]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/kernfs/dir.c
x86 / CPU: Always show current CPU frequency in /proc/cpuinfo
[mirror_ubuntu-artful-kernel.git] / fs / kernfs / dir.c
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
21 DEFINE_MUTEX(kernfs_mutex);
22 static DEFINE_SPINLOCK(kernfs_rename_lock); /* kn->parent and ->name */
23 static 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
27 static bool kernfs_active(struct kernfs_node *kn)
28 {
29 lockdep_assert_held(&kernfs_mutex);
30 return atomic_read(&kn->active) >= 0;
31 }
32
33 static 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
42 static 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 */
51 static 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
62 static 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 */
122 static 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 common = kernfs_common_ancestor(kn_from, kn_to);
141 if (WARN_ON(!common))
142 return -EINVAL;
143
144 depth_to = kernfs_depth(common, kn_to);
145 depth_from = kernfs_depth(common, kn_from);
146
147 if (buf)
148 buf[0] = '\0';
149
150 for (i = 0; i < depth_from; i++)
151 len += strlcpy(buf + len, parent_str,
152 len < buflen ? buflen - len : 0);
153
154 /* Calculate how many bytes we need for the rest */
155 for (i = depth_to - 1; i >= 0; i--) {
156 for (kn = kn_to, j = 0; j < i; j++)
157 kn = kn->parent;
158 len += strlcpy(buf + len, "/",
159 len < buflen ? buflen - len : 0);
160 len += strlcpy(buf + len, kn->name,
161 len < buflen ? buflen - len : 0);
162 }
163
164 return len;
165 }
166
167 /**
168 * kernfs_name - obtain the name of a given node
169 * @kn: kernfs_node of interest
170 * @buf: buffer to copy @kn's name into
171 * @buflen: size of @buf
172 *
173 * Copies the name of @kn into @buf of @buflen bytes. The behavior is
174 * similar to strlcpy(). It returns the length of @kn's name and if @buf
175 * isn't long enough, it's filled upto @buflen-1 and nul terminated.
176 *
177 * Fills buffer with "(null)" if @kn is NULL.
178 *
179 * This function can be called from any context.
180 */
181 int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
182 {
183 unsigned long flags;
184 int ret;
185
186 spin_lock_irqsave(&kernfs_rename_lock, flags);
187 ret = kernfs_name_locked(kn, buf, buflen);
188 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
189 return ret;
190 }
191
192 /**
193 * kernfs_path_from_node - build path of node @to relative to @from.
194 * @from: parent kernfs_node relative to which we need to build the path
195 * @to: kernfs_node of interest
196 * @buf: buffer to copy @to's path into
197 * @buflen: size of @buf
198 *
199 * Builds @to's path relative to @from in @buf. @from and @to must
200 * be on the same kernfs-root. If @from is not parent of @to, then a relative
201 * path (which includes '..'s) as needed to reach from @from to @to is
202 * returned.
203 *
204 * Returns the length of the full path. If the full length is equal to or
205 * greater than @buflen, @buf contains the truncated path with the trailing
206 * '\0'. On error, -errno is returned.
207 */
208 int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
209 char *buf, size_t buflen)
210 {
211 unsigned long flags;
212 int ret;
213
214 spin_lock_irqsave(&kernfs_rename_lock, flags);
215 ret = kernfs_path_from_node_locked(to, from, buf, buflen);
216 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
217 return ret;
218 }
219 EXPORT_SYMBOL_GPL(kernfs_path_from_node);
220
221 /**
222 * pr_cont_kernfs_name - pr_cont name of a kernfs_node
223 * @kn: kernfs_node of interest
224 *
225 * This function can be called from any context.
226 */
227 void pr_cont_kernfs_name(struct kernfs_node *kn)
228 {
229 unsigned long flags;
230
231 spin_lock_irqsave(&kernfs_rename_lock, flags);
232
233 kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf));
234 pr_cont("%s", kernfs_pr_cont_buf);
235
236 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
237 }
238
239 /**
240 * pr_cont_kernfs_path - pr_cont path of a kernfs_node
241 * @kn: kernfs_node of interest
242 *
243 * This function can be called from any context.
244 */
245 void pr_cont_kernfs_path(struct kernfs_node *kn)
246 {
247 unsigned long flags;
248 int sz;
249
250 spin_lock_irqsave(&kernfs_rename_lock, flags);
251
252 sz = kernfs_path_from_node_locked(kn, NULL, kernfs_pr_cont_buf,
253 sizeof(kernfs_pr_cont_buf));
254 if (sz < 0) {
255 pr_cont("(error)");
256 goto out;
257 }
258
259 if (sz >= sizeof(kernfs_pr_cont_buf)) {
260 pr_cont("(name too long)");
261 goto out;
262 }
263
264 pr_cont("%s", kernfs_pr_cont_buf);
265
266 out:
267 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
268 }
269
270 /**
271 * kernfs_get_parent - determine the parent node and pin it
272 * @kn: kernfs_node of interest
273 *
274 * Determines @kn's parent, pins and returns it. This function can be
275 * called from any context.
276 */
277 struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn)
278 {
279 struct kernfs_node *parent;
280 unsigned long flags;
281
282 spin_lock_irqsave(&kernfs_rename_lock, flags);
283 parent = kn->parent;
284 kernfs_get(parent);
285 spin_unlock_irqrestore(&kernfs_rename_lock, flags);
286
287 return parent;
288 }
289
290 /**
291 * kernfs_name_hash
292 * @name: Null terminated string to hash
293 * @ns: Namespace tag to hash
294 *
295 * Returns 31 bit hash of ns + name (so it fits in an off_t )
296 */
297 static unsigned int kernfs_name_hash(const char *name, const void *ns)
298 {
299 unsigned long hash = init_name_hash(ns);
300 unsigned int len = strlen(name);
301 while (len--)
302 hash = partial_name_hash(*name++, hash);
303 hash = end_name_hash(hash);
304 hash &= 0x7fffffffU;
305 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
306 if (hash < 2)
307 hash += 2;
308 if (hash >= INT_MAX)
309 hash = INT_MAX - 1;
310 return hash;
311 }
312
313 static int kernfs_name_compare(unsigned int hash, const char *name,
314 const void *ns, const struct kernfs_node *kn)
315 {
316 if (hash < kn->hash)
317 return -1;
318 if (hash > kn->hash)
319 return 1;
320 if (ns < kn->ns)
321 return -1;
322 if (ns > kn->ns)
323 return 1;
324 return strcmp(name, kn->name);
325 }
326
327 static int kernfs_sd_compare(const struct kernfs_node *left,
328 const struct kernfs_node *right)
329 {
330 return kernfs_name_compare(left->hash, left->name, left->ns, right);
331 }
332
333 /**
334 * kernfs_link_sibling - link kernfs_node into sibling rbtree
335 * @kn: kernfs_node of interest
336 *
337 * Link @kn into its sibling rbtree which starts from
338 * @kn->parent->dir.children.
339 *
340 * Locking:
341 * mutex_lock(kernfs_mutex)
342 *
343 * RETURNS:
344 * 0 on susccess -EEXIST on failure.
345 */
346 static int kernfs_link_sibling(struct kernfs_node *kn)
347 {
348 struct rb_node **node = &kn->parent->dir.children.rb_node;
349 struct rb_node *parent = NULL;
350
351 while (*node) {
352 struct kernfs_node *pos;
353 int result;
354
355 pos = rb_to_kn(*node);
356 parent = *node;
357 result = kernfs_sd_compare(kn, pos);
358 if (result < 0)
359 node = &pos->rb.rb_left;
360 else if (result > 0)
361 node = &pos->rb.rb_right;
362 else
363 return -EEXIST;
364 }
365
366 /* add new node and rebalance the tree */
367 rb_link_node(&kn->rb, parent, node);
368 rb_insert_color(&kn->rb, &kn->parent->dir.children);
369
370 /* successfully added, account subdir number */
371 if (kernfs_type(kn) == KERNFS_DIR)
372 kn->parent->dir.subdirs++;
373
374 return 0;
375 }
376
377 /**
378 * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree
379 * @kn: kernfs_node of interest
380 *
381 * Try to unlink @kn from its sibling rbtree which starts from
382 * kn->parent->dir.children. Returns %true if @kn was actually
383 * removed, %false if @kn wasn't on the rbtree.
384 *
385 * Locking:
386 * mutex_lock(kernfs_mutex)
387 */
388 static bool kernfs_unlink_sibling(struct kernfs_node *kn)
389 {
390 if (RB_EMPTY_NODE(&kn->rb))
391 return false;
392
393 if (kernfs_type(kn) == KERNFS_DIR)
394 kn->parent->dir.subdirs--;
395
396 rb_erase(&kn->rb, &kn->parent->dir.children);
397 RB_CLEAR_NODE(&kn->rb);
398 return true;
399 }
400
401 /**
402 * kernfs_get_active - get an active reference to kernfs_node
403 * @kn: kernfs_node to get an active reference to
404 *
405 * Get an active reference of @kn. This function is noop if @kn
406 * is NULL.
407 *
408 * RETURNS:
409 * Pointer to @kn on success, NULL on failure.
410 */
411 struct kernfs_node *kernfs_get_active(struct kernfs_node *kn)
412 {
413 if (unlikely(!kn))
414 return NULL;
415
416 if (!atomic_inc_unless_negative(&kn->active))
417 return NULL;
418
419 if (kernfs_lockdep(kn))
420 rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_);
421 return kn;
422 }
423
424 /**
425 * kernfs_put_active - put an active reference to kernfs_node
426 * @kn: kernfs_node to put an active reference to
427 *
428 * Put an active reference to @kn. This function is noop if @kn
429 * is NULL.
430 */
431 void kernfs_put_active(struct kernfs_node *kn)
432 {
433 struct kernfs_root *root = kernfs_root(kn);
434 int v;
435
436 if (unlikely(!kn))
437 return;
438
439 if (kernfs_lockdep(kn))
440 rwsem_release(&kn->dep_map, 1, _RET_IP_);
441 v = atomic_dec_return(&kn->active);
442 if (likely(v != KN_DEACTIVATED_BIAS))
443 return;
444
445 wake_up_all(&root->deactivate_waitq);
446 }
447
448 /**
449 * kernfs_drain - drain kernfs_node
450 * @kn: kernfs_node to drain
451 *
452 * Drain existing usages and nuke all existing mmaps of @kn. Mutiple
453 * removers may invoke this function concurrently on @kn and all will
454 * return after draining is complete.
455 */
456 static void kernfs_drain(struct kernfs_node *kn)
457 __releases(&kernfs_mutex) __acquires(&kernfs_mutex)
458 {
459 struct kernfs_root *root = kernfs_root(kn);
460
461 lockdep_assert_held(&kernfs_mutex);
462 WARN_ON_ONCE(kernfs_active(kn));
463
464 mutex_unlock(&kernfs_mutex);
465
466 if (kernfs_lockdep(kn)) {
467 rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_);
468 if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS)
469 lock_contended(&kn->dep_map, _RET_IP_);
470 }
471
472 /* but everyone should wait for draining */
473 wait_event(root->deactivate_waitq,
474 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
475
476 if (kernfs_lockdep(kn)) {
477 lock_acquired(&kn->dep_map, _RET_IP_);
478 rwsem_release(&kn->dep_map, 1, _RET_IP_);
479 }
480
481 kernfs_drain_open_files(kn);
482
483 mutex_lock(&kernfs_mutex);
484 }
485
486 /**
487 * kernfs_get - get a reference count on a kernfs_node
488 * @kn: the target kernfs_node
489 */
490 void kernfs_get(struct kernfs_node *kn)
491 {
492 if (kn) {
493 WARN_ON(!atomic_read(&kn->count));
494 atomic_inc(&kn->count);
495 }
496 }
497 EXPORT_SYMBOL_GPL(kernfs_get);
498
499 /**
500 * kernfs_put - put a reference count on a kernfs_node
501 * @kn: the target kernfs_node
502 *
503 * Put a reference count of @kn and destroy it if it reached zero.
504 */
505 void kernfs_put(struct kernfs_node *kn)
506 {
507 struct kernfs_node *parent;
508 struct kernfs_root *root;
509
510 if (!kn || !atomic_dec_and_test(&kn->count))
511 return;
512 root = kernfs_root(kn);
513 repeat:
514 /*
515 * Moving/renaming is always done while holding reference.
516 * kn->parent won't change beneath us.
517 */
518 parent = kn->parent;
519
520 WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS,
521 "kernfs_put: %s/%s: released with incorrect active_ref %d\n",
522 parent ? parent->name : "", kn->name, atomic_read(&kn->active));
523
524 if (kernfs_type(kn) == KERNFS_LINK)
525 kernfs_put(kn->symlink.target_kn);
526
527 kfree_const(kn->name);
528
529 if (kn->iattr) {
530 if (kn->iattr->ia_secdata)
531 security_release_secctx(kn->iattr->ia_secdata,
532 kn->iattr->ia_secdata_len);
533 simple_xattrs_free(&kn->iattr->xattrs);
534 }
535 kfree(kn->iattr);
536 ida_simple_remove(&root->ino_ida, kn->ino);
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 ida_destroy(&root->ino_ida);
546 kfree(root);
547 }
548 }
549 EXPORT_SYMBOL_GPL(kernfs_put);
550
551 static 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 = dentry->d_fsdata;
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 (dentry->d_parent->d_fsdata != 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;
584 out_bad:
585 mutex_unlock(&kernfs_mutex);
586 out_bad_unlocked:
587 return 0;
588 }
589
590 static void kernfs_dop_release(struct dentry *dentry)
591 {
592 kernfs_put(dentry->d_fsdata);
593 }
594
595 const struct dentry_operations kernfs_dops = {
596 .d_revalidate = kernfs_dop_revalidate,
597 .d_release = kernfs_dop_release,
598 };
599
600 /**
601 * kernfs_node_from_dentry - determine kernfs_node associated with a dentry
602 * @dentry: the dentry in question
603 *
604 * Return the kernfs_node associated with @dentry. If @dentry is not a
605 * kernfs one, %NULL is returned.
606 *
607 * While the returned kernfs_node will stay accessible as long as @dentry
608 * is accessible, the returned node can be in any state and the caller is
609 * fully responsible for determining what's accessible.
610 */
611 struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry)
612 {
613 if (dentry->d_sb->s_op == &kernfs_sops)
614 return dentry->d_fsdata;
615 return NULL;
616 }
617
618 static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root,
619 const char *name, umode_t mode,
620 unsigned flags)
621 {
622 struct kernfs_node *kn;
623 int ret;
624
625 name = kstrdup_const(name, GFP_KERNEL);
626 if (!name)
627 return NULL;
628
629 kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL);
630 if (!kn)
631 goto err_out1;
632
633 ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL);
634 if (ret < 0)
635 goto err_out2;
636 kn->ino = ret;
637
638 atomic_set(&kn->count, 1);
639 atomic_set(&kn->active, KN_DEACTIVATED_BIAS);
640 RB_CLEAR_NODE(&kn->rb);
641
642 kn->name = name;
643 kn->mode = mode;
644 kn->flags = flags;
645
646 return kn;
647
648 err_out2:
649 kmem_cache_free(kernfs_node_cache, kn);
650 err_out1:
651 kfree_const(name);
652 return NULL;
653 }
654
655 struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
656 const char *name, umode_t mode,
657 unsigned flags)
658 {
659 struct kernfs_node *kn;
660
661 kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags);
662 if (kn) {
663 kernfs_get(parent);
664 kn->parent = parent;
665 }
666 return kn;
667 }
668
669 /**
670 * kernfs_add_one - add kernfs_node to parent without warning
671 * @kn: kernfs_node to be added
672 *
673 * The caller must already have initialized @kn->parent. This
674 * function increments nlink of the parent's inode if @kn is a
675 * directory and link into the children list of the parent.
676 *
677 * RETURNS:
678 * 0 on success, -EEXIST if entry with the given name already
679 * exists.
680 */
681 int kernfs_add_one(struct kernfs_node *kn)
682 {
683 struct kernfs_node *parent = kn->parent;
684 struct kernfs_iattrs *ps_iattr;
685 bool has_ns;
686 int ret;
687
688 mutex_lock(&kernfs_mutex);
689
690 ret = -EINVAL;
691 has_ns = kernfs_ns_enabled(parent);
692 if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
693 has_ns ? "required" : "invalid", parent->name, kn->name))
694 goto out_unlock;
695
696 if (kernfs_type(parent) != KERNFS_DIR)
697 goto out_unlock;
698
699 ret = -ENOENT;
700 if (parent->flags & KERNFS_EMPTY_DIR)
701 goto out_unlock;
702
703 if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent))
704 goto out_unlock;
705
706 kn->hash = kernfs_name_hash(kn->name, kn->ns);
707
708 ret = kernfs_link_sibling(kn);
709 if (ret)
710 goto out_unlock;
711
712 /* Update timestamps on the parent */
713 ps_iattr = parent->iattr;
714 if (ps_iattr) {
715 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
716 ktime_get_real_ts(&ps_iattrs->ia_ctime);
717 ps_iattrs->ia_mtime = ps_iattrs->ia_ctime;
718 }
719
720 mutex_unlock(&kernfs_mutex);
721
722 /*
723 * Activate the new node unless CREATE_DEACTIVATED is requested.
724 * If not activated here, the kernfs user is responsible for
725 * activating the node with kernfs_activate(). A node which hasn't
726 * been activated is not visible to userland and its removal won't
727 * trigger deactivation.
728 */
729 if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
730 kernfs_activate(kn);
731 return 0;
732
733 out_unlock:
734 mutex_unlock(&kernfs_mutex);
735 return ret;
736 }
737
738 /**
739 * kernfs_find_ns - find kernfs_node with the given name
740 * @parent: kernfs_node to search under
741 * @name: name to look for
742 * @ns: the namespace tag to use
743 *
744 * Look for kernfs_node with name @name under @parent. Returns pointer to
745 * the found kernfs_node on success, %NULL on failure.
746 */
747 static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent,
748 const unsigned char *name,
749 const void *ns)
750 {
751 struct rb_node *node = parent->dir.children.rb_node;
752 bool has_ns = kernfs_ns_enabled(parent);
753 unsigned int hash;
754
755 lockdep_assert_held(&kernfs_mutex);
756
757 if (has_ns != (bool)ns) {
758 WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n",
759 has_ns ? "required" : "invalid", parent->name, name);
760 return NULL;
761 }
762
763 hash = kernfs_name_hash(name, ns);
764 while (node) {
765 struct kernfs_node *kn;
766 int result;
767
768 kn = rb_to_kn(node);
769 result = kernfs_name_compare(hash, name, ns, kn);
770 if (result < 0)
771 node = node->rb_left;
772 else if (result > 0)
773 node = node->rb_right;
774 else
775 return kn;
776 }
777 return NULL;
778 }
779
780 static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent,
781 const unsigned char *path,
782 const void *ns)
783 {
784 size_t len;
785 char *p, *name;
786
787 lockdep_assert_held(&kernfs_mutex);
788
789 /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */
790 spin_lock_irq(&kernfs_rename_lock);
791
792 len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf));
793
794 if (len >= sizeof(kernfs_pr_cont_buf)) {
795 spin_unlock_irq(&kernfs_rename_lock);
796 return NULL;
797 }
798
799 p = kernfs_pr_cont_buf;
800
801 while ((name = strsep(&p, "/")) && parent) {
802 if (*name == '\0')
803 continue;
804 parent = kernfs_find_ns(parent, name, ns);
805 }
806
807 spin_unlock_irq(&kernfs_rename_lock);
808
809 return parent;
810 }
811
812 /**
813 * kernfs_find_and_get_ns - find and get kernfs_node with the given name
814 * @parent: kernfs_node to search under
815 * @name: name to look for
816 * @ns: the namespace tag to use
817 *
818 * Look for kernfs_node with name @name under @parent and get a reference
819 * if found. This function may sleep and returns pointer to the found
820 * kernfs_node on success, %NULL on failure.
821 */
822 struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent,
823 const char *name, const void *ns)
824 {
825 struct kernfs_node *kn;
826
827 mutex_lock(&kernfs_mutex);
828 kn = kernfs_find_ns(parent, name, ns);
829 kernfs_get(kn);
830 mutex_unlock(&kernfs_mutex);
831
832 return kn;
833 }
834 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
835
836 /**
837 * kernfs_walk_and_get_ns - find and get kernfs_node with the given path
838 * @parent: kernfs_node to search under
839 * @path: path to look for
840 * @ns: the namespace tag to use
841 *
842 * Look for kernfs_node with path @path under @parent and get a reference
843 * if found. This function may sleep and returns pointer to the found
844 * kernfs_node on success, %NULL on failure.
845 */
846 struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent,
847 const char *path, const void *ns)
848 {
849 struct kernfs_node *kn;
850
851 mutex_lock(&kernfs_mutex);
852 kn = kernfs_walk_ns(parent, path, ns);
853 kernfs_get(kn);
854 mutex_unlock(&kernfs_mutex);
855
856 return kn;
857 }
858
859 /**
860 * kernfs_create_root - create a new kernfs hierarchy
861 * @scops: optional syscall operations for the hierarchy
862 * @flags: KERNFS_ROOT_* flags
863 * @priv: opaque data associated with the new directory
864 *
865 * Returns the root of the new hierarchy on success, ERR_PTR() value on
866 * failure.
867 */
868 struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops,
869 unsigned int flags, void *priv)
870 {
871 struct kernfs_root *root;
872 struct kernfs_node *kn;
873
874 root = kzalloc(sizeof(*root), GFP_KERNEL);
875 if (!root)
876 return ERR_PTR(-ENOMEM);
877
878 ida_init(&root->ino_ida);
879 INIT_LIST_HEAD(&root->supers);
880
881 kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO,
882 KERNFS_DIR);
883 if (!kn) {
884 ida_destroy(&root->ino_ida);
885 kfree(root);
886 return ERR_PTR(-ENOMEM);
887 }
888
889 kn->priv = priv;
890 kn->dir.root = root;
891
892 root->syscall_ops = scops;
893 root->flags = flags;
894 root->kn = kn;
895 init_waitqueue_head(&root->deactivate_waitq);
896
897 if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED))
898 kernfs_activate(kn);
899
900 return root;
901 }
902
903 /**
904 * kernfs_destroy_root - destroy a kernfs hierarchy
905 * @root: root of the hierarchy to destroy
906 *
907 * Destroy the hierarchy anchored at @root by removing all existing
908 * directories and destroying @root.
909 */
910 void kernfs_destroy_root(struct kernfs_root *root)
911 {
912 kernfs_remove(root->kn); /* will also free @root */
913 }
914
915 /**
916 * kernfs_create_dir_ns - create a directory
917 * @parent: parent in which to create a new directory
918 * @name: name of the new directory
919 * @mode: mode of the new directory
920 * @priv: opaque data associated with the new directory
921 * @ns: optional namespace tag of the directory
922 *
923 * Returns the created node on success, ERR_PTR() value on failure.
924 */
925 struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent,
926 const char *name, umode_t mode,
927 void *priv, const void *ns)
928 {
929 struct kernfs_node *kn;
930 int rc;
931
932 /* allocate */
933 kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR);
934 if (!kn)
935 return ERR_PTR(-ENOMEM);
936
937 kn->dir.root = parent->dir.root;
938 kn->ns = ns;
939 kn->priv = priv;
940
941 /* link in */
942 rc = kernfs_add_one(kn);
943 if (!rc)
944 return kn;
945
946 kernfs_put(kn);
947 return ERR_PTR(rc);
948 }
949
950 /**
951 * kernfs_create_empty_dir - create an always empty directory
952 * @parent: parent in which to create a new directory
953 * @name: name of the new directory
954 *
955 * Returns the created node on success, ERR_PTR() value on failure.
956 */
957 struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent,
958 const char *name)
959 {
960 struct kernfs_node *kn;
961 int rc;
962
963 /* allocate */
964 kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR, KERNFS_DIR);
965 if (!kn)
966 return ERR_PTR(-ENOMEM);
967
968 kn->flags |= KERNFS_EMPTY_DIR;
969 kn->dir.root = parent->dir.root;
970 kn->ns = NULL;
971 kn->priv = NULL;
972
973 /* link in */
974 rc = kernfs_add_one(kn);
975 if (!rc)
976 return kn;
977
978 kernfs_put(kn);
979 return ERR_PTR(rc);
980 }
981
982 static struct dentry *kernfs_iop_lookup(struct inode *dir,
983 struct dentry *dentry,
984 unsigned int flags)
985 {
986 struct dentry *ret;
987 struct kernfs_node *parent = dentry->d_parent->d_fsdata;
988 struct kernfs_node *kn;
989 struct inode *inode;
990 const void *ns = NULL;
991
992 mutex_lock(&kernfs_mutex);
993
994 if (kernfs_ns_enabled(parent))
995 ns = kernfs_info(dir->i_sb)->ns;
996
997 kn = kernfs_find_ns(parent, dentry->d_name.name, ns);
998
999 /* no such entry */
1000 if (!kn || !kernfs_active(kn)) {
1001 ret = NULL;
1002 goto out_unlock;
1003 }
1004 kernfs_get(kn);
1005 dentry->d_fsdata = kn;
1006
1007 /* attach dentry and inode */
1008 inode = kernfs_get_inode(dir->i_sb, kn);
1009 if (!inode) {
1010 ret = ERR_PTR(-ENOMEM);
1011 goto out_unlock;
1012 }
1013
1014 /* instantiate and hash dentry */
1015 ret = d_splice_alias(inode, dentry);
1016 out_unlock:
1017 mutex_unlock(&kernfs_mutex);
1018 return ret;
1019 }
1020
1021 static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry,
1022 umode_t mode)
1023 {
1024 struct kernfs_node *parent = dir->i_private;
1025 struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops;
1026 int ret;
1027
1028 if (!scops || !scops->mkdir)
1029 return -EPERM;
1030
1031 if (!kernfs_get_active(parent))
1032 return -ENODEV;
1033
1034 ret = scops->mkdir(parent, dentry->d_name.name, mode);
1035
1036 kernfs_put_active(parent);
1037 return ret;
1038 }
1039
1040 static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry)
1041 {
1042 struct kernfs_node *kn = dentry->d_fsdata;
1043 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1044 int ret;
1045
1046 if (!scops || !scops->rmdir)
1047 return -EPERM;
1048
1049 if (!kernfs_get_active(kn))
1050 return -ENODEV;
1051
1052 ret = scops->rmdir(kn);
1053
1054 kernfs_put_active(kn);
1055 return ret;
1056 }
1057
1058 static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry,
1059 struct inode *new_dir, struct dentry *new_dentry,
1060 unsigned int flags)
1061 {
1062 struct kernfs_node *kn = old_dentry->d_fsdata;
1063 struct kernfs_node *new_parent = new_dir->i_private;
1064 struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops;
1065 int ret;
1066
1067 if (flags)
1068 return -EINVAL;
1069
1070 if (!scops || !scops->rename)
1071 return -EPERM;
1072
1073 if (!kernfs_get_active(kn))
1074 return -ENODEV;
1075
1076 if (!kernfs_get_active(new_parent)) {
1077 kernfs_put_active(kn);
1078 return -ENODEV;
1079 }
1080
1081 ret = scops->rename(kn, new_parent, new_dentry->d_name.name);
1082
1083 kernfs_put_active(new_parent);
1084 kernfs_put_active(kn);
1085 return ret;
1086 }
1087
1088 const struct inode_operations kernfs_dir_iops = {
1089 .lookup = kernfs_iop_lookup,
1090 .permission = kernfs_iop_permission,
1091 .setattr = kernfs_iop_setattr,
1092 .getattr = kernfs_iop_getattr,
1093 .listxattr = kernfs_iop_listxattr,
1094
1095 .mkdir = kernfs_iop_mkdir,
1096 .rmdir = kernfs_iop_rmdir,
1097 .rename = kernfs_iop_rename,
1098 };
1099
1100 static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos)
1101 {
1102 struct kernfs_node *last;
1103
1104 while (true) {
1105 struct rb_node *rbn;
1106
1107 last = pos;
1108
1109 if (kernfs_type(pos) != KERNFS_DIR)
1110 break;
1111
1112 rbn = rb_first(&pos->dir.children);
1113 if (!rbn)
1114 break;
1115
1116 pos = rb_to_kn(rbn);
1117 }
1118
1119 return last;
1120 }
1121
1122 /**
1123 * kernfs_next_descendant_post - find the next descendant for post-order walk
1124 * @pos: the current position (%NULL to initiate traversal)
1125 * @root: kernfs_node whose descendants to walk
1126 *
1127 * Find the next descendant to visit for post-order traversal of @root's
1128 * descendants. @root is included in the iteration and the last node to be
1129 * visited.
1130 */
1131 static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos,
1132 struct kernfs_node *root)
1133 {
1134 struct rb_node *rbn;
1135
1136 lockdep_assert_held(&kernfs_mutex);
1137
1138 /* if first iteration, visit leftmost descendant which may be root */
1139 if (!pos)
1140 return kernfs_leftmost_descendant(root);
1141
1142 /* if we visited @root, we're done */
1143 if (pos == root)
1144 return NULL;
1145
1146 /* if there's an unvisited sibling, visit its leftmost descendant */
1147 rbn = rb_next(&pos->rb);
1148 if (rbn)
1149 return kernfs_leftmost_descendant(rb_to_kn(rbn));
1150
1151 /* no sibling left, visit parent */
1152 return pos->parent;
1153 }
1154
1155 /**
1156 * kernfs_activate - activate a node which started deactivated
1157 * @kn: kernfs_node whose subtree is to be activated
1158 *
1159 * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node
1160 * needs to be explicitly activated. A node which hasn't been activated
1161 * isn't visible to userland and deactivation is skipped during its
1162 * removal. This is useful to construct atomic init sequences where
1163 * creation of multiple nodes should either succeed or fail atomically.
1164 *
1165 * The caller is responsible for ensuring that this function is not called
1166 * after kernfs_remove*() is invoked on @kn.
1167 */
1168 void kernfs_activate(struct kernfs_node *kn)
1169 {
1170 struct kernfs_node *pos;
1171
1172 mutex_lock(&kernfs_mutex);
1173
1174 pos = NULL;
1175 while ((pos = kernfs_next_descendant_post(pos, kn))) {
1176 if (!pos || (pos->flags & KERNFS_ACTIVATED))
1177 continue;
1178
1179 WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb));
1180 WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS);
1181
1182 atomic_sub(KN_DEACTIVATED_BIAS, &pos->active);
1183 pos->flags |= KERNFS_ACTIVATED;
1184 }
1185
1186 mutex_unlock(&kernfs_mutex);
1187 }
1188
1189 static void __kernfs_remove(struct kernfs_node *kn)
1190 {
1191 struct kernfs_node *pos;
1192
1193 lockdep_assert_held(&kernfs_mutex);
1194
1195 /*
1196 * Short-circuit if non-root @kn has already finished removal.
1197 * This is for kernfs_remove_self() which plays with active ref
1198 * after removal.
1199 */
1200 if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb)))
1201 return;
1202
1203 pr_debug("kernfs %s: removing\n", kn->name);
1204
1205 /* prevent any new usage under @kn by deactivating all nodes */
1206 pos = NULL;
1207 while ((pos = kernfs_next_descendant_post(pos, kn)))
1208 if (kernfs_active(pos))
1209 atomic_add(KN_DEACTIVATED_BIAS, &pos->active);
1210
1211 /* deactivate and unlink the subtree node-by-node */
1212 do {
1213 pos = kernfs_leftmost_descendant(kn);
1214
1215 /*
1216 * kernfs_drain() drops kernfs_mutex temporarily and @pos's
1217 * base ref could have been put by someone else by the time
1218 * the function returns. Make sure it doesn't go away
1219 * underneath us.
1220 */
1221 kernfs_get(pos);
1222
1223 /*
1224 * Drain iff @kn was activated. This avoids draining and
1225 * its lockdep annotations for nodes which have never been
1226 * activated and allows embedding kernfs_remove() in create
1227 * error paths without worrying about draining.
1228 */
1229 if (kn->flags & KERNFS_ACTIVATED)
1230 kernfs_drain(pos);
1231 else
1232 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
1233
1234 /*
1235 * kernfs_unlink_sibling() succeeds once per node. Use it
1236 * to decide who's responsible for cleanups.
1237 */
1238 if (!pos->parent || kernfs_unlink_sibling(pos)) {
1239 struct kernfs_iattrs *ps_iattr =
1240 pos->parent ? pos->parent->iattr : NULL;
1241
1242 /* update timestamps on the parent */
1243 if (ps_iattr) {
1244 ktime_get_real_ts(&ps_iattr->ia_iattr.ia_ctime);
1245 ps_iattr->ia_iattr.ia_mtime =
1246 ps_iattr->ia_iattr.ia_ctime;
1247 }
1248
1249 kernfs_put(pos);
1250 }
1251
1252 kernfs_put(pos);
1253 } while (pos != kn);
1254 }
1255
1256 /**
1257 * kernfs_remove - remove a kernfs_node recursively
1258 * @kn: the kernfs_node to remove
1259 *
1260 * Remove @kn along with all its subdirectories and files.
1261 */
1262 void kernfs_remove(struct kernfs_node *kn)
1263 {
1264 mutex_lock(&kernfs_mutex);
1265 __kernfs_remove(kn);
1266 mutex_unlock(&kernfs_mutex);
1267 }
1268
1269 /**
1270 * kernfs_break_active_protection - break out of active protection
1271 * @kn: the self kernfs_node
1272 *
1273 * The caller must be running off of a kernfs operation which is invoked
1274 * with an active reference - e.g. one of kernfs_ops. Each invocation of
1275 * this function must also be matched with an invocation of
1276 * kernfs_unbreak_active_protection().
1277 *
1278 * This function releases the active reference of @kn the caller is
1279 * holding. Once this function is called, @kn may be removed at any point
1280 * and the caller is solely responsible for ensuring that the objects it
1281 * dereferences are accessible.
1282 */
1283 void kernfs_break_active_protection(struct kernfs_node *kn)
1284 {
1285 /*
1286 * Take out ourself out of the active ref dependency chain. If
1287 * we're called without an active ref, lockdep will complain.
1288 */
1289 kernfs_put_active(kn);
1290 }
1291
1292 /**
1293 * kernfs_unbreak_active_protection - undo kernfs_break_active_protection()
1294 * @kn: the self kernfs_node
1295 *
1296 * If kernfs_break_active_protection() was called, this function must be
1297 * invoked before finishing the kernfs operation. Note that while this
1298 * function restores the active reference, it doesn't and can't actually
1299 * restore the active protection - @kn may already or be in the process of
1300 * being removed. Once kernfs_break_active_protection() is invoked, that
1301 * protection is irreversibly gone for the kernfs operation instance.
1302 *
1303 * While this function may be called at any point after
1304 * kernfs_break_active_protection() is invoked, its most useful location
1305 * would be right before the enclosing kernfs operation returns.
1306 */
1307 void kernfs_unbreak_active_protection(struct kernfs_node *kn)
1308 {
1309 /*
1310 * @kn->active could be in any state; however, the increment we do
1311 * here will be undone as soon as the enclosing kernfs operation
1312 * finishes and this temporary bump can't break anything. If @kn
1313 * is alive, nothing changes. If @kn is being deactivated, the
1314 * soon-to-follow put will either finish deactivation or restore
1315 * deactivated state. If @kn is already removed, the temporary
1316 * bump is guaranteed to be gone before @kn is released.
1317 */
1318 atomic_inc(&kn->active);
1319 if (kernfs_lockdep(kn))
1320 rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_);
1321 }
1322
1323 /**
1324 * kernfs_remove_self - remove a kernfs_node from its own method
1325 * @kn: the self kernfs_node to remove
1326 *
1327 * The caller must be running off of a kernfs operation which is invoked
1328 * with an active reference - e.g. one of kernfs_ops. This can be used to
1329 * implement a file operation which deletes itself.
1330 *
1331 * For example, the "delete" file for a sysfs device directory can be
1332 * implemented by invoking kernfs_remove_self() on the "delete" file
1333 * itself. This function breaks the circular dependency of trying to
1334 * deactivate self while holding an active ref itself. It isn't necessary
1335 * to modify the usual removal path to use kernfs_remove_self(). The
1336 * "delete" implementation can simply invoke kernfs_remove_self() on self
1337 * before proceeding with the usual removal path. kernfs will ignore later
1338 * kernfs_remove() on self.
1339 *
1340 * kernfs_remove_self() can be called multiple times concurrently on the
1341 * same kernfs_node. Only the first one actually performs removal and
1342 * returns %true. All others will wait until the kernfs operation which
1343 * won self-removal finishes and return %false. Note that the losers wait
1344 * for the completion of not only the winning kernfs_remove_self() but also
1345 * the whole kernfs_ops which won the arbitration. This can be used to
1346 * guarantee, for example, all concurrent writes to a "delete" file to
1347 * finish only after the whole operation is complete.
1348 */
1349 bool kernfs_remove_self(struct kernfs_node *kn)
1350 {
1351 bool ret;
1352
1353 mutex_lock(&kernfs_mutex);
1354 kernfs_break_active_protection(kn);
1355
1356 /*
1357 * SUICIDAL is used to arbitrate among competing invocations. Only
1358 * the first one will actually perform removal. When the removal
1359 * is complete, SUICIDED is set and the active ref is restored
1360 * while holding kernfs_mutex. The ones which lost arbitration
1361 * waits for SUICDED && drained which can happen only after the
1362 * enclosing kernfs operation which executed the winning instance
1363 * of kernfs_remove_self() finished.
1364 */
1365 if (!(kn->flags & KERNFS_SUICIDAL)) {
1366 kn->flags |= KERNFS_SUICIDAL;
1367 __kernfs_remove(kn);
1368 kn->flags |= KERNFS_SUICIDED;
1369 ret = true;
1370 } else {
1371 wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq;
1372 DEFINE_WAIT(wait);
1373
1374 while (true) {
1375 prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE);
1376
1377 if ((kn->flags & KERNFS_SUICIDED) &&
1378 atomic_read(&kn->active) == KN_DEACTIVATED_BIAS)
1379 break;
1380
1381 mutex_unlock(&kernfs_mutex);
1382 schedule();
1383 mutex_lock(&kernfs_mutex);
1384 }
1385 finish_wait(waitq, &wait);
1386 WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb));
1387 ret = false;
1388 }
1389
1390 /*
1391 * This must be done while holding kernfs_mutex; otherwise, waiting
1392 * for SUICIDED && deactivated could finish prematurely.
1393 */
1394 kernfs_unbreak_active_protection(kn);
1395
1396 mutex_unlock(&kernfs_mutex);
1397 return ret;
1398 }
1399
1400 /**
1401 * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it
1402 * @parent: parent of the target
1403 * @name: name of the kernfs_node to remove
1404 * @ns: namespace tag of the kernfs_node to remove
1405 *
1406 * Look for the kernfs_node with @name and @ns under @parent and remove it.
1407 * Returns 0 on success, -ENOENT if such entry doesn't exist.
1408 */
1409 int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
1410 const void *ns)
1411 {
1412 struct kernfs_node *kn;
1413
1414 if (!parent) {
1415 WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n",
1416 name);
1417 return -ENOENT;
1418 }
1419
1420 mutex_lock(&kernfs_mutex);
1421
1422 kn = kernfs_find_ns(parent, name, ns);
1423 if (kn)
1424 __kernfs_remove(kn);
1425
1426 mutex_unlock(&kernfs_mutex);
1427
1428 if (kn)
1429 return 0;
1430 else
1431 return -ENOENT;
1432 }
1433
1434 /**
1435 * kernfs_rename_ns - move and rename a kernfs_node
1436 * @kn: target node
1437 * @new_parent: new parent to put @sd under
1438 * @new_name: new name
1439 * @new_ns: new namespace tag
1440 */
1441 int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent,
1442 const char *new_name, const void *new_ns)
1443 {
1444 struct kernfs_node *old_parent;
1445 const char *old_name = NULL;
1446 int error;
1447
1448 /* can't move or rename root */
1449 if (!kn->parent)
1450 return -EINVAL;
1451
1452 mutex_lock(&kernfs_mutex);
1453
1454 error = -ENOENT;
1455 if (!kernfs_active(kn) || !kernfs_active(new_parent) ||
1456 (new_parent->flags & KERNFS_EMPTY_DIR))
1457 goto out;
1458
1459 error = 0;
1460 if ((kn->parent == new_parent) && (kn->ns == new_ns) &&
1461 (strcmp(kn->name, new_name) == 0))
1462 goto out; /* nothing to rename */
1463
1464 error = -EEXIST;
1465 if (kernfs_find_ns(new_parent, new_name, new_ns))
1466 goto out;
1467
1468 /* rename kernfs_node */
1469 if (strcmp(kn->name, new_name) != 0) {
1470 error = -ENOMEM;
1471 new_name = kstrdup_const(new_name, GFP_KERNEL);
1472 if (!new_name)
1473 goto out;
1474 } else {
1475 new_name = NULL;
1476 }
1477
1478 /*
1479 * Move to the appropriate place in the appropriate directories rbtree.
1480 */
1481 kernfs_unlink_sibling(kn);
1482 kernfs_get(new_parent);
1483
1484 /* rename_lock protects ->parent and ->name accessors */
1485 spin_lock_irq(&kernfs_rename_lock);
1486
1487 old_parent = kn->parent;
1488 kn->parent = new_parent;
1489
1490 kn->ns = new_ns;
1491 if (new_name) {
1492 old_name = kn->name;
1493 kn->name = new_name;
1494 }
1495
1496 spin_unlock_irq(&kernfs_rename_lock);
1497
1498 kn->hash = kernfs_name_hash(kn->name, kn->ns);
1499 kernfs_link_sibling(kn);
1500
1501 kernfs_put(old_parent);
1502 kfree_const(old_name);
1503
1504 error = 0;
1505 out:
1506 mutex_unlock(&kernfs_mutex);
1507 return error;
1508 }
1509
1510 /* Relationship between s_mode and the DT_xxx types */
1511 static inline unsigned char dt_type(struct kernfs_node *kn)
1512 {
1513 return (kn->mode >> 12) & 15;
1514 }
1515
1516 static int kernfs_dir_fop_release(struct inode *inode, struct file *filp)
1517 {
1518 kernfs_put(filp->private_data);
1519 return 0;
1520 }
1521
1522 static struct kernfs_node *kernfs_dir_pos(const void *ns,
1523 struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos)
1524 {
1525 if (pos) {
1526 int valid = kernfs_active(pos) &&
1527 pos->parent == parent && hash == pos->hash;
1528 kernfs_put(pos);
1529 if (!valid)
1530 pos = NULL;
1531 }
1532 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1533 struct rb_node *node = parent->dir.children.rb_node;
1534 while (node) {
1535 pos = rb_to_kn(node);
1536
1537 if (hash < pos->hash)
1538 node = node->rb_left;
1539 else if (hash > pos->hash)
1540 node = node->rb_right;
1541 else
1542 break;
1543 }
1544 }
1545 /* Skip over entries which are dying/dead or in the wrong namespace */
1546 while (pos && (!kernfs_active(pos) || pos->ns != ns)) {
1547 struct rb_node *node = rb_next(&pos->rb);
1548 if (!node)
1549 pos = NULL;
1550 else
1551 pos = rb_to_kn(node);
1552 }
1553 return pos;
1554 }
1555
1556 static struct kernfs_node *kernfs_dir_next_pos(const void *ns,
1557 struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos)
1558 {
1559 pos = kernfs_dir_pos(ns, parent, ino, pos);
1560 if (pos) {
1561 do {
1562 struct rb_node *node = rb_next(&pos->rb);
1563 if (!node)
1564 pos = NULL;
1565 else
1566 pos = rb_to_kn(node);
1567 } while (pos && (!kernfs_active(pos) || pos->ns != ns));
1568 }
1569 return pos;
1570 }
1571
1572 static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx)
1573 {
1574 struct dentry *dentry = file->f_path.dentry;
1575 struct kernfs_node *parent = dentry->d_fsdata;
1576 struct kernfs_node *pos = file->private_data;
1577 const void *ns = NULL;
1578
1579 if (!dir_emit_dots(file, ctx))
1580 return 0;
1581 mutex_lock(&kernfs_mutex);
1582
1583 if (kernfs_ns_enabled(parent))
1584 ns = kernfs_info(dentry->d_sb)->ns;
1585
1586 for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos);
1587 pos;
1588 pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) {
1589 const char *name = pos->name;
1590 unsigned int type = dt_type(pos);
1591 int len = strlen(name);
1592 ino_t ino = pos->ino;
1593
1594 ctx->pos = pos->hash;
1595 file->private_data = pos;
1596 kernfs_get(pos);
1597
1598 mutex_unlock(&kernfs_mutex);
1599 if (!dir_emit(ctx, name, len, ino, type))
1600 return 0;
1601 mutex_lock(&kernfs_mutex);
1602 }
1603 mutex_unlock(&kernfs_mutex);
1604 file->private_data = NULL;
1605 ctx->pos = INT_MAX;
1606 return 0;
1607 }
1608
1609 const struct file_operations kernfs_dir_fops = {
1610 .read = generic_read_dir,
1611 .iterate_shared = kernfs_fop_readdir,
1612 .release = kernfs_dir_fop_release,
1613 .llseek = generic_file_llseek,
1614 };