~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/kernfs/dir.c

Version: ~ [ linux-5.1-rc1 ] ~ [ linux-5.0.2 ] ~ [ linux-4.20.16 ] ~ [ linux-4.19.29 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.106 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.163 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.176 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.136 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.63 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp