Version:  2.0.40 2.2.26 2.4.37 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0

Linux/fs/dcache.c

  1 /*
  2  * fs/dcache.c
  3  *
  4  * Complete reimplementation
  5  * (C) 1997 Thomas Schoebel-Theuer,
  6  * with heavy changes by Linus Torvalds
  7  */
  8 
  9 /*
 10  * Notes on the allocation strategy:
 11  *
 12  * The dcache is a master of the icache - whenever a dcache entry
 13  * exists, the inode will always exist. "iput()" is done either when
 14  * the dcache entry is deleted or garbage collected.
 15  */
 16 
 17 #include <linux/syscalls.h>
 18 #include <linux/string.h>
 19 #include <linux/mm.h>
 20 #include <linux/fs.h>
 21 #include <linux/fsnotify.h>
 22 #include <linux/slab.h>
 23 #include <linux/init.h>
 24 #include <linux/hash.h>
 25 #include <linux/cache.h>
 26 #include <linux/export.h>
 27 #include <linux/mount.h>
 28 #include <linux/file.h>
 29 #include <asm/uaccess.h>
 30 #include <linux/security.h>
 31 #include <linux/seqlock.h>
 32 #include <linux/swap.h>
 33 #include <linux/bootmem.h>
 34 #include <linux/fs_struct.h>
 35 #include <linux/hardirq.h>
 36 #include <linux/bit_spinlock.h>
 37 #include <linux/rculist_bl.h>
 38 #include <linux/prefetch.h>
 39 #include <linux/ratelimit.h>
 40 #include <linux/list_lru.h>
 41 #include <linux/kasan.h>
 42 
 43 #include "internal.h"
 44 #include "mount.h"
 45 
 46 /*
 47  * Usage:
 48  * dcache->d_inode->i_lock protects:
 49  *   - i_dentry, d_u.d_alias, d_inode of aliases
 50  * dcache_hash_bucket lock protects:
 51  *   - the dcache hash table
 52  * s_anon bl list spinlock protects:
 53  *   - the s_anon list (see __d_drop)
 54  * dentry->d_sb->s_dentry_lru_lock protects:
 55  *   - the dcache lru lists and counters
 56  * d_lock protects:
 57  *   - d_flags
 58  *   - d_name
 59  *   - d_lru
 60  *   - d_count
 61  *   - d_unhashed()
 62  *   - d_parent and d_subdirs
 63  *   - childrens' d_child and d_parent
 64  *   - d_u.d_alias, d_inode
 65  *
 66  * Ordering:
 67  * dentry->d_inode->i_lock
 68  *   dentry->d_lock
 69  *     dentry->d_sb->s_dentry_lru_lock
 70  *     dcache_hash_bucket lock
 71  *     s_anon lock
 72  *
 73  * If there is an ancestor relationship:
 74  * dentry->d_parent->...->d_parent->d_lock
 75  *   ...
 76  *     dentry->d_parent->d_lock
 77  *       dentry->d_lock
 78  *
 79  * If no ancestor relationship:
 80  * if (dentry1 < dentry2)
 81  *   dentry1->d_lock
 82  *     dentry2->d_lock
 83  */
 84 int sysctl_vfs_cache_pressure __read_mostly = 100;
 85 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
 86 
 87 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
 88 
 89 EXPORT_SYMBOL(rename_lock);
 90 
 91 static struct kmem_cache *dentry_cache __read_mostly;
 92 
 93 /*
 94  * This is the single most critical data structure when it comes
 95  * to the dcache: the hashtable for lookups. Somebody should try
 96  * to make this good - I've just made it work.
 97  *
 98  * This hash-function tries to avoid losing too many bits of hash
 99  * information, yet avoid using a prime hash-size or similar.
100  */
101 
102 static unsigned int d_hash_mask __read_mostly;
103 static unsigned int d_hash_shift __read_mostly;
104 
105 static struct hlist_bl_head *dentry_hashtable __read_mostly;
106 
107 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
108                                         unsigned int hash)
109 {
110         hash += (unsigned long) parent / L1_CACHE_BYTES;
111         return dentry_hashtable + hash_32(hash, d_hash_shift);
112 }
113 
114 /* Statistics gathering. */
115 struct dentry_stat_t dentry_stat = {
116         .age_limit = 45,
117 };
118 
119 static DEFINE_PER_CPU(long, nr_dentry);
120 static DEFINE_PER_CPU(long, nr_dentry_unused);
121 
122 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
123 
124 /*
125  * Here we resort to our own counters instead of using generic per-cpu counters
126  * for consistency with what the vfs inode code does. We are expected to harvest
127  * better code and performance by having our own specialized counters.
128  *
129  * Please note that the loop is done over all possible CPUs, not over all online
130  * CPUs. The reason for this is that we don't want to play games with CPUs going
131  * on and off. If one of them goes off, we will just keep their counters.
132  *
133  * glommer: See cffbc8a for details, and if you ever intend to change this,
134  * please update all vfs counters to match.
135  */
136 static long get_nr_dentry(void)
137 {
138         int i;
139         long sum = 0;
140         for_each_possible_cpu(i)
141                 sum += per_cpu(nr_dentry, i);
142         return sum < 0 ? 0 : sum;
143 }
144 
145 static long get_nr_dentry_unused(void)
146 {
147         int i;
148         long sum = 0;
149         for_each_possible_cpu(i)
150                 sum += per_cpu(nr_dentry_unused, i);
151         return sum < 0 ? 0 : sum;
152 }
153 
154 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
155                    size_t *lenp, loff_t *ppos)
156 {
157         dentry_stat.nr_dentry = get_nr_dentry();
158         dentry_stat.nr_unused = get_nr_dentry_unused();
159         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
160 }
161 #endif
162 
163 /*
164  * Compare 2 name strings, return 0 if they match, otherwise non-zero.
165  * The strings are both count bytes long, and count is non-zero.
166  */
167 #ifdef CONFIG_DCACHE_WORD_ACCESS
168 
169 #include <asm/word-at-a-time.h>
170 /*
171  * NOTE! 'cs' and 'scount' come from a dentry, so it has a
172  * aligned allocation for this particular component. We don't
173  * strictly need the load_unaligned_zeropad() safety, but it
174  * doesn't hurt either.
175  *
176  * In contrast, 'ct' and 'tcount' can be from a pathname, and do
177  * need the careful unaligned handling.
178  */
179 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
180 {
181         unsigned long a,b,mask;
182 
183         for (;;) {
184                 a = *(unsigned long *)cs;
185                 b = load_unaligned_zeropad(ct);
186                 if (tcount < sizeof(unsigned long))
187                         break;
188                 if (unlikely(a != b))
189                         return 1;
190                 cs += sizeof(unsigned long);
191                 ct += sizeof(unsigned long);
192                 tcount -= sizeof(unsigned long);
193                 if (!tcount)
194                         return 0;
195         }
196         mask = bytemask_from_count(tcount);
197         return unlikely(!!((a ^ b) & mask));
198 }
199 
200 #else
201 
202 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
203 {
204         do {
205                 if (*cs != *ct)
206                         return 1;
207                 cs++;
208                 ct++;
209                 tcount--;
210         } while (tcount);
211         return 0;
212 }
213 
214 #endif
215 
216 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
217 {
218         const unsigned char *cs;
219         /*
220          * Be careful about RCU walk racing with rename:
221          * use ACCESS_ONCE to fetch the name pointer.
222          *
223          * NOTE! Even if a rename will mean that the length
224          * was not loaded atomically, we don't care. The
225          * RCU walk will check the sequence count eventually,
226          * and catch it. And we won't overrun the buffer,
227          * because we're reading the name pointer atomically,
228          * and a dentry name is guaranteed to be properly
229          * terminated with a NUL byte.
230          *
231          * End result: even if 'len' is wrong, we'll exit
232          * early because the data cannot match (there can
233          * be no NUL in the ct/tcount data)
234          */
235         cs = ACCESS_ONCE(dentry->d_name.name);
236         smp_read_barrier_depends();
237         return dentry_string_cmp(cs, ct, tcount);
238 }
239 
240 struct external_name {
241         union {
242                 atomic_t count;
243                 struct rcu_head head;
244         } u;
245         unsigned char name[];
246 };
247 
248 static inline struct external_name *external_name(struct dentry *dentry)
249 {
250         return container_of(dentry->d_name.name, struct external_name, name[0]);
251 }
252 
253 static void __d_free(struct rcu_head *head)
254 {
255         struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
256 
257         kmem_cache_free(dentry_cache, dentry); 
258 }
259 
260 static void __d_free_external(struct rcu_head *head)
261 {
262         struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
263         kfree(external_name(dentry));
264         kmem_cache_free(dentry_cache, dentry); 
265 }
266 
267 static inline int dname_external(const struct dentry *dentry)
268 {
269         return dentry->d_name.name != dentry->d_iname;
270 }
271 
272 static void dentry_free(struct dentry *dentry)
273 {
274         WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
275         if (unlikely(dname_external(dentry))) {
276                 struct external_name *p = external_name(dentry);
277                 if (likely(atomic_dec_and_test(&p->u.count))) {
278                         call_rcu(&dentry->d_u.d_rcu, __d_free_external);
279                         return;
280                 }
281         }
282         /* if dentry was never visible to RCU, immediate free is OK */
283         if (!(dentry->d_flags & DCACHE_RCUACCESS))
284                 __d_free(&dentry->d_u.d_rcu);
285         else
286                 call_rcu(&dentry->d_u.d_rcu, __d_free);
287 }
288 
289 /**
290  * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
291  * @dentry: the target dentry
292  * After this call, in-progress rcu-walk path lookup will fail. This
293  * should be called after unhashing, and after changing d_inode (if
294  * the dentry has not already been unhashed).
295  */
296 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
297 {
298         assert_spin_locked(&dentry->d_lock);
299         /* Go through a barrier */
300         write_seqcount_barrier(&dentry->d_seq);
301 }
302 
303 /*
304  * Release the dentry's inode, using the filesystem
305  * d_iput() operation if defined. Dentry has no refcount
306  * and is unhashed.
307  */
308 static void dentry_iput(struct dentry * dentry)
309         __releases(dentry->d_lock)
310         __releases(dentry->d_inode->i_lock)
311 {
312         struct inode *inode = dentry->d_inode;
313         if (inode) {
314                 dentry->d_inode = NULL;
315                 hlist_del_init(&dentry->d_u.d_alias);
316                 spin_unlock(&dentry->d_lock);
317                 spin_unlock(&inode->i_lock);
318                 if (!inode->i_nlink)
319                         fsnotify_inoderemove(inode);
320                 if (dentry->d_op && dentry->d_op->d_iput)
321                         dentry->d_op->d_iput(dentry, inode);
322                 else
323                         iput(inode);
324         } else {
325                 spin_unlock(&dentry->d_lock);
326         }
327 }
328 
329 /*
330  * Release the dentry's inode, using the filesystem
331  * d_iput() operation if defined. dentry remains in-use.
332  */
333 static void dentry_unlink_inode(struct dentry * dentry)
334         __releases(dentry->d_lock)
335         __releases(dentry->d_inode->i_lock)
336 {
337         struct inode *inode = dentry->d_inode;
338         __d_clear_type(dentry);
339         dentry->d_inode = NULL;
340         hlist_del_init(&dentry->d_u.d_alias);
341         dentry_rcuwalk_barrier(dentry);
342         spin_unlock(&dentry->d_lock);
343         spin_unlock(&inode->i_lock);
344         if (!inode->i_nlink)
345                 fsnotify_inoderemove(inode);
346         if (dentry->d_op && dentry->d_op->d_iput)
347                 dentry->d_op->d_iput(dentry, inode);
348         else
349                 iput(inode);
350 }
351 
352 /*
353  * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
354  * is in use - which includes both the "real" per-superblock
355  * LRU list _and_ the DCACHE_SHRINK_LIST use.
356  *
357  * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
358  * on the shrink list (ie not on the superblock LRU list).
359  *
360  * The per-cpu "nr_dentry_unused" counters are updated with
361  * the DCACHE_LRU_LIST bit.
362  *
363  * These helper functions make sure we always follow the
364  * rules. d_lock must be held by the caller.
365  */
366 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
367 static void d_lru_add(struct dentry *dentry)
368 {
369         D_FLAG_VERIFY(dentry, 0);
370         dentry->d_flags |= DCACHE_LRU_LIST;
371         this_cpu_inc(nr_dentry_unused);
372         WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
373 }
374 
375 static void d_lru_del(struct dentry *dentry)
376 {
377         D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
378         dentry->d_flags &= ~DCACHE_LRU_LIST;
379         this_cpu_dec(nr_dentry_unused);
380         WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
381 }
382 
383 static void d_shrink_del(struct dentry *dentry)
384 {
385         D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
386         list_del_init(&dentry->d_lru);
387         dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
388         this_cpu_dec(nr_dentry_unused);
389 }
390 
391 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
392 {
393         D_FLAG_VERIFY(dentry, 0);
394         list_add(&dentry->d_lru, list);
395         dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
396         this_cpu_inc(nr_dentry_unused);
397 }
398 
399 /*
400  * These can only be called under the global LRU lock, ie during the
401  * callback for freeing the LRU list. "isolate" removes it from the
402  * LRU lists entirely, while shrink_move moves it to the indicated
403  * private list.
404  */
405 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
406 {
407         D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
408         dentry->d_flags &= ~DCACHE_LRU_LIST;
409         this_cpu_dec(nr_dentry_unused);
410         list_lru_isolate(lru, &dentry->d_lru);
411 }
412 
413 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
414                               struct list_head *list)
415 {
416         D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
417         dentry->d_flags |= DCACHE_SHRINK_LIST;
418         list_lru_isolate_move(lru, &dentry->d_lru, list);
419 }
420 
421 /*
422  * dentry_lru_(add|del)_list) must be called with d_lock held.
423  */
424 static void dentry_lru_add(struct dentry *dentry)
425 {
426         if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
427                 d_lru_add(dentry);
428 }
429 
430 /**
431  * d_drop - drop a dentry
432  * @dentry: dentry to drop
433  *
434  * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
435  * be found through a VFS lookup any more. Note that this is different from
436  * deleting the dentry - d_delete will try to mark the dentry negative if
437  * possible, giving a successful _negative_ lookup, while d_drop will
438  * just make the cache lookup fail.
439  *
440  * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
441  * reason (NFS timeouts or autofs deletes).
442  *
443  * __d_drop requires dentry->d_lock.
444  */
445 void __d_drop(struct dentry *dentry)
446 {
447         if (!d_unhashed(dentry)) {
448                 struct hlist_bl_head *b;
449                 /*
450                  * Hashed dentries are normally on the dentry hashtable,
451                  * with the exception of those newly allocated by
452                  * d_obtain_alias, which are always IS_ROOT:
453                  */
454                 if (unlikely(IS_ROOT(dentry)))
455                         b = &dentry->d_sb->s_anon;
456                 else
457                         b = d_hash(dentry->d_parent, dentry->d_name.hash);
458 
459                 hlist_bl_lock(b);
460                 __hlist_bl_del(&dentry->d_hash);
461                 dentry->d_hash.pprev = NULL;
462                 hlist_bl_unlock(b);
463                 dentry_rcuwalk_barrier(dentry);
464         }
465 }
466 EXPORT_SYMBOL(__d_drop);
467 
468 void d_drop(struct dentry *dentry)
469 {
470         spin_lock(&dentry->d_lock);
471         __d_drop(dentry);
472         spin_unlock(&dentry->d_lock);
473 }
474 EXPORT_SYMBOL(d_drop);
475 
476 static void __dentry_kill(struct dentry *dentry)
477 {
478         struct dentry *parent = NULL;
479         bool can_free = true;
480         if (!IS_ROOT(dentry))
481                 parent = dentry->d_parent;
482 
483         /*
484          * The dentry is now unrecoverably dead to the world.
485          */
486         lockref_mark_dead(&dentry->d_lockref);
487 
488         /*
489          * inform the fs via d_prune that this dentry is about to be
490          * unhashed and destroyed.
491          */
492         if (dentry->d_flags & DCACHE_OP_PRUNE)
493                 dentry->d_op->d_prune(dentry);
494 
495         if (dentry->d_flags & DCACHE_LRU_LIST) {
496                 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
497                         d_lru_del(dentry);
498         }
499         /* if it was on the hash then remove it */
500         __d_drop(dentry);
501         __list_del_entry(&dentry->d_child);
502         /*
503          * Inform d_walk() that we are no longer attached to the
504          * dentry tree
505          */
506         dentry->d_flags |= DCACHE_DENTRY_KILLED;
507         if (parent)
508                 spin_unlock(&parent->d_lock);
509         dentry_iput(dentry);
510         /*
511          * dentry_iput drops the locks, at which point nobody (except
512          * transient RCU lookups) can reach this dentry.
513          */
514         BUG_ON(dentry->d_lockref.count > 0);
515         this_cpu_dec(nr_dentry);
516         if (dentry->d_op && dentry->d_op->d_release)
517                 dentry->d_op->d_release(dentry);
518 
519         spin_lock(&dentry->d_lock);
520         if (dentry->d_flags & DCACHE_SHRINK_LIST) {
521                 dentry->d_flags |= DCACHE_MAY_FREE;
522                 can_free = false;
523         }
524         spin_unlock(&dentry->d_lock);
525         if (likely(can_free))
526                 dentry_free(dentry);
527 }
528 
529 /*
530  * Finish off a dentry we've decided to kill.
531  * dentry->d_lock must be held, returns with it unlocked.
532  * If ref is non-zero, then decrement the refcount too.
533  * Returns dentry requiring refcount drop, or NULL if we're done.
534  */
535 static struct dentry *dentry_kill(struct dentry *dentry)
536         __releases(dentry->d_lock)
537 {
538         struct inode *inode = dentry->d_inode;
539         struct dentry *parent = NULL;
540 
541         if (inode && unlikely(!spin_trylock(&inode->i_lock)))
542                 goto failed;
543 
544         if (!IS_ROOT(dentry)) {
545                 parent = dentry->d_parent;
546                 if (unlikely(!spin_trylock(&parent->d_lock))) {
547                         if (inode)
548                                 spin_unlock(&inode->i_lock);
549                         goto failed;
550                 }
551         }
552 
553         __dentry_kill(dentry);
554         return parent;
555 
556 failed:
557         spin_unlock(&dentry->d_lock);
558         cpu_relax();
559         return dentry; /* try again with same dentry */
560 }
561 
562 static inline struct dentry *lock_parent(struct dentry *dentry)
563 {
564         struct dentry *parent = dentry->d_parent;
565         if (IS_ROOT(dentry))
566                 return NULL;
567         if (unlikely(dentry->d_lockref.count < 0))
568                 return NULL;
569         if (likely(spin_trylock(&parent->d_lock)))
570                 return parent;
571         rcu_read_lock();
572         spin_unlock(&dentry->d_lock);
573 again:
574         parent = ACCESS_ONCE(dentry->d_parent);
575         spin_lock(&parent->d_lock);
576         /*
577          * We can't blindly lock dentry until we are sure
578          * that we won't violate the locking order.
579          * Any changes of dentry->d_parent must have
580          * been done with parent->d_lock held, so
581          * spin_lock() above is enough of a barrier
582          * for checking if it's still our child.
583          */
584         if (unlikely(parent != dentry->d_parent)) {
585                 spin_unlock(&parent->d_lock);
586                 goto again;
587         }
588         rcu_read_unlock();
589         if (parent != dentry)
590                 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
591         else
592                 parent = NULL;
593         return parent;
594 }
595 
596 /*
597  * Try to do a lockless dput(), and return whether that was successful.
598  *
599  * If unsuccessful, we return false, having already taken the dentry lock.
600  *
601  * The caller needs to hold the RCU read lock, so that the dentry is
602  * guaranteed to stay around even if the refcount goes down to zero!
603  */
604 static inline bool fast_dput(struct dentry *dentry)
605 {
606         int ret;
607         unsigned int d_flags;
608 
609         /*
610          * If we have a d_op->d_delete() operation, we sould not
611          * let the dentry count go to zero, so use "put__or_lock".
612          */
613         if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
614                 return lockref_put_or_lock(&dentry->d_lockref);
615 
616         /*
617          * .. otherwise, we can try to just decrement the
618          * lockref optimistically.
619          */
620         ret = lockref_put_return(&dentry->d_lockref);
621 
622         /*
623          * If the lockref_put_return() failed due to the lock being held
624          * by somebody else, the fast path has failed. We will need to
625          * get the lock, and then check the count again.
626          */
627         if (unlikely(ret < 0)) {
628                 spin_lock(&dentry->d_lock);
629                 if (dentry->d_lockref.count > 1) {
630                         dentry->d_lockref.count--;
631                         spin_unlock(&dentry->d_lock);
632                         return 1;
633                 }
634                 return 0;
635         }
636 
637         /*
638          * If we weren't the last ref, we're done.
639          */
640         if (ret)
641                 return 1;
642 
643         /*
644          * Careful, careful. The reference count went down
645          * to zero, but we don't hold the dentry lock, so
646          * somebody else could get it again, and do another
647          * dput(), and we need to not race with that.
648          *
649          * However, there is a very special and common case
650          * where we don't care, because there is nothing to
651          * do: the dentry is still hashed, it does not have
652          * a 'delete' op, and it's referenced and already on
653          * the LRU list.
654          *
655          * NOTE! Since we aren't locked, these values are
656          * not "stable". However, it is sufficient that at
657          * some point after we dropped the reference the
658          * dentry was hashed and the flags had the proper
659          * value. Other dentry users may have re-gotten
660          * a reference to the dentry and change that, but
661          * our work is done - we can leave the dentry
662          * around with a zero refcount.
663          */
664         smp_rmb();
665         d_flags = ACCESS_ONCE(dentry->d_flags);
666         d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST;
667 
668         /* Nothing to do? Dropping the reference was all we needed? */
669         if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
670                 return 1;
671 
672         /*
673          * Not the fast normal case? Get the lock. We've already decremented
674          * the refcount, but we'll need to re-check the situation after
675          * getting the lock.
676          */
677         spin_lock(&dentry->d_lock);
678 
679         /*
680          * Did somebody else grab a reference to it in the meantime, and
681          * we're no longer the last user after all? Alternatively, somebody
682          * else could have killed it and marked it dead. Either way, we
683          * don't need to do anything else.
684          */
685         if (dentry->d_lockref.count) {
686                 spin_unlock(&dentry->d_lock);
687                 return 1;
688         }
689 
690         /*
691          * Re-get the reference we optimistically dropped. We hold the
692          * lock, and we just tested that it was zero, so we can just
693          * set it to 1.
694          */
695         dentry->d_lockref.count = 1;
696         return 0;
697 }
698 
699 
700 /* 
701  * This is dput
702  *
703  * This is complicated by the fact that we do not want to put
704  * dentries that are no longer on any hash chain on the unused
705  * list: we'd much rather just get rid of them immediately.
706  *
707  * However, that implies that we have to traverse the dentry
708  * tree upwards to the parents which might _also_ now be
709  * scheduled for deletion (it may have been only waiting for
710  * its last child to go away).
711  *
712  * This tail recursion is done by hand as we don't want to depend
713  * on the compiler to always get this right (gcc generally doesn't).
714  * Real recursion would eat up our stack space.
715  */
716 
717 /*
718  * dput - release a dentry
719  * @dentry: dentry to release 
720  *
721  * Release a dentry. This will drop the usage count and if appropriate
722  * call the dentry unlink method as well as removing it from the queues and
723  * releasing its resources. If the parent dentries were scheduled for release
724  * they too may now get deleted.
725  */
726 void dput(struct dentry *dentry)
727 {
728         if (unlikely(!dentry))
729                 return;
730 
731 repeat:
732         rcu_read_lock();
733         if (likely(fast_dput(dentry))) {
734                 rcu_read_unlock();
735                 return;
736         }
737 
738         /* Slow case: now with the dentry lock held */
739         rcu_read_unlock();
740 
741         /* Unreachable? Get rid of it */
742         if (unlikely(d_unhashed(dentry)))
743                 goto kill_it;
744 
745         if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
746                 if (dentry->d_op->d_delete(dentry))
747                         goto kill_it;
748         }
749 
750         if (!(dentry->d_flags & DCACHE_REFERENCED))
751                 dentry->d_flags |= DCACHE_REFERENCED;
752         dentry_lru_add(dentry);
753 
754         dentry->d_lockref.count--;
755         spin_unlock(&dentry->d_lock);
756         return;
757 
758 kill_it:
759         dentry = dentry_kill(dentry);
760         if (dentry)
761                 goto repeat;
762 }
763 EXPORT_SYMBOL(dput);
764 
765 
766 /* This must be called with d_lock held */
767 static inline void __dget_dlock(struct dentry *dentry)
768 {
769         dentry->d_lockref.count++;
770 }
771 
772 static inline void __dget(struct dentry *dentry)
773 {
774         lockref_get(&dentry->d_lockref);
775 }
776 
777 struct dentry *dget_parent(struct dentry *dentry)
778 {
779         int gotref;
780         struct dentry *ret;
781 
782         /*
783          * Do optimistic parent lookup without any
784          * locking.
785          */
786         rcu_read_lock();
787         ret = ACCESS_ONCE(dentry->d_parent);
788         gotref = lockref_get_not_zero(&ret->d_lockref);
789         rcu_read_unlock();
790         if (likely(gotref)) {
791                 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
792                         return ret;
793                 dput(ret);
794         }
795 
796 repeat:
797         /*
798          * Don't need rcu_dereference because we re-check it was correct under
799          * the lock.
800          */
801         rcu_read_lock();
802         ret = dentry->d_parent;
803         spin_lock(&ret->d_lock);
804         if (unlikely(ret != dentry->d_parent)) {
805                 spin_unlock(&ret->d_lock);
806                 rcu_read_unlock();
807                 goto repeat;
808         }
809         rcu_read_unlock();
810         BUG_ON(!ret->d_lockref.count);
811         ret->d_lockref.count++;
812         spin_unlock(&ret->d_lock);
813         return ret;
814 }
815 EXPORT_SYMBOL(dget_parent);
816 
817 /**
818  * d_find_alias - grab a hashed alias of inode
819  * @inode: inode in question
820  *
821  * If inode has a hashed alias, or is a directory and has any alias,
822  * acquire the reference to alias and return it. Otherwise return NULL.
823  * Notice that if inode is a directory there can be only one alias and
824  * it can be unhashed only if it has no children, or if it is the root
825  * of a filesystem, or if the directory was renamed and d_revalidate
826  * was the first vfs operation to notice.
827  *
828  * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
829  * any other hashed alias over that one.
830  */
831 static struct dentry *__d_find_alias(struct inode *inode)
832 {
833         struct dentry *alias, *discon_alias;
834 
835 again:
836         discon_alias = NULL;
837         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
838                 spin_lock(&alias->d_lock);
839                 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
840                         if (IS_ROOT(alias) &&
841                             (alias->d_flags & DCACHE_DISCONNECTED)) {
842                                 discon_alias = alias;
843                         } else {
844                                 __dget_dlock(alias);
845                                 spin_unlock(&alias->d_lock);
846                                 return alias;
847                         }
848                 }
849                 spin_unlock(&alias->d_lock);
850         }
851         if (discon_alias) {
852                 alias = discon_alias;
853                 spin_lock(&alias->d_lock);
854                 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
855                         __dget_dlock(alias);
856                         spin_unlock(&alias->d_lock);
857                         return alias;
858                 }
859                 spin_unlock(&alias->d_lock);
860                 goto again;
861         }
862         return NULL;
863 }
864 
865 struct dentry *d_find_alias(struct inode *inode)
866 {
867         struct dentry *de = NULL;
868 
869         if (!hlist_empty(&inode->i_dentry)) {
870                 spin_lock(&inode->i_lock);
871                 de = __d_find_alias(inode);
872                 spin_unlock(&inode->i_lock);
873         }
874         return de;
875 }
876 EXPORT_SYMBOL(d_find_alias);
877 
878 /*
879  *      Try to kill dentries associated with this inode.
880  * WARNING: you must own a reference to inode.
881  */
882 void d_prune_aliases(struct inode *inode)
883 {
884         struct dentry *dentry;
885 restart:
886         spin_lock(&inode->i_lock);
887         hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
888                 spin_lock(&dentry->d_lock);
889                 if (!dentry->d_lockref.count) {
890                         struct dentry *parent = lock_parent(dentry);
891                         if (likely(!dentry->d_lockref.count)) {
892                                 __dentry_kill(dentry);
893                                 dput(parent);
894                                 goto restart;
895                         }
896                         if (parent)
897                                 spin_unlock(&parent->d_lock);
898                 }
899                 spin_unlock(&dentry->d_lock);
900         }
901         spin_unlock(&inode->i_lock);
902 }
903 EXPORT_SYMBOL(d_prune_aliases);
904 
905 static void shrink_dentry_list(struct list_head *list)
906 {
907         struct dentry *dentry, *parent;
908 
909         while (!list_empty(list)) {
910                 struct inode *inode;
911                 dentry = list_entry(list->prev, struct dentry, d_lru);
912                 spin_lock(&dentry->d_lock);
913                 parent = lock_parent(dentry);
914 
915                 /*
916                  * The dispose list is isolated and dentries are not accounted
917                  * to the LRU here, so we can simply remove it from the list
918                  * here regardless of whether it is referenced or not.
919                  */
920                 d_shrink_del(dentry);
921 
922                 /*
923                  * We found an inuse dentry which was not removed from
924                  * the LRU because of laziness during lookup. Do not free it.
925                  */
926                 if (dentry->d_lockref.count > 0) {
927                         spin_unlock(&dentry->d_lock);
928                         if (parent)
929                                 spin_unlock(&parent->d_lock);
930                         continue;
931                 }
932 
933 
934                 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
935                         bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
936                         spin_unlock(&dentry->d_lock);
937                         if (parent)
938                                 spin_unlock(&parent->d_lock);
939                         if (can_free)
940                                 dentry_free(dentry);
941                         continue;
942                 }
943 
944                 inode = dentry->d_inode;
945                 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
946                         d_shrink_add(dentry, list);
947                         spin_unlock(&dentry->d_lock);
948                         if (parent)
949                                 spin_unlock(&parent->d_lock);
950                         continue;
951                 }
952 
953                 __dentry_kill(dentry);
954 
955                 /*
956                  * We need to prune ancestors too. This is necessary to prevent
957                  * quadratic behavior of shrink_dcache_parent(), but is also
958                  * expected to be beneficial in reducing dentry cache
959                  * fragmentation.
960                  */
961                 dentry = parent;
962                 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
963                         parent = lock_parent(dentry);
964                         if (dentry->d_lockref.count != 1) {
965                                 dentry->d_lockref.count--;
966                                 spin_unlock(&dentry->d_lock);
967                                 if (parent)
968                                         spin_unlock(&parent->d_lock);
969                                 break;
970                         }
971                         inode = dentry->d_inode;        /* can't be NULL */
972                         if (unlikely(!spin_trylock(&inode->i_lock))) {
973                                 spin_unlock(&dentry->d_lock);
974                                 if (parent)
975                                         spin_unlock(&parent->d_lock);
976                                 cpu_relax();
977                                 continue;
978                         }
979                         __dentry_kill(dentry);
980                         dentry = parent;
981                 }
982         }
983 }
984 
985 static enum lru_status dentry_lru_isolate(struct list_head *item,
986                 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
987 {
988         struct list_head *freeable = arg;
989         struct dentry   *dentry = container_of(item, struct dentry, d_lru);
990 
991 
992         /*
993          * we are inverting the lru lock/dentry->d_lock here,
994          * so use a trylock. If we fail to get the lock, just skip
995          * it
996          */
997         if (!spin_trylock(&dentry->d_lock))
998                 return LRU_SKIP;
999 
1000         /*
1001          * Referenced dentries are still in use. If they have active
1002          * counts, just remove them from the LRU. Otherwise give them
1003          * another pass through the LRU.
1004          */
1005         if (dentry->d_lockref.count) {
1006                 d_lru_isolate(lru, dentry);
1007                 spin_unlock(&dentry->d_lock);
1008                 return LRU_REMOVED;
1009         }
1010 
1011         if (dentry->d_flags & DCACHE_REFERENCED) {
1012                 dentry->d_flags &= ~DCACHE_REFERENCED;
1013                 spin_unlock(&dentry->d_lock);
1014 
1015                 /*
1016                  * The list move itself will be made by the common LRU code. At
1017                  * this point, we've dropped the dentry->d_lock but keep the
1018                  * lru lock. This is safe to do, since every list movement is
1019                  * protected by the lru lock even if both locks are held.
1020                  *
1021                  * This is guaranteed by the fact that all LRU management
1022                  * functions are intermediated by the LRU API calls like
1023                  * list_lru_add and list_lru_del. List movement in this file
1024                  * only ever occur through this functions or through callbacks
1025                  * like this one, that are called from the LRU API.
1026                  *
1027                  * The only exceptions to this are functions like
1028                  * shrink_dentry_list, and code that first checks for the
1029                  * DCACHE_SHRINK_LIST flag.  Those are guaranteed to be
1030                  * operating only with stack provided lists after they are
1031                  * properly isolated from the main list.  It is thus, always a
1032                  * local access.
1033                  */
1034                 return LRU_ROTATE;
1035         }
1036 
1037         d_lru_shrink_move(lru, dentry, freeable);
1038         spin_unlock(&dentry->d_lock);
1039 
1040         return LRU_REMOVED;
1041 }
1042 
1043 /**
1044  * prune_dcache_sb - shrink the dcache
1045  * @sb: superblock
1046  * @sc: shrink control, passed to list_lru_shrink_walk()
1047  *
1048  * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1049  * is done when we need more memory and called from the superblock shrinker
1050  * function.
1051  *
1052  * This function may fail to free any resources if all the dentries are in
1053  * use.
1054  */
1055 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1056 {
1057         LIST_HEAD(dispose);
1058         long freed;
1059 
1060         freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1061                                      dentry_lru_isolate, &dispose);
1062         shrink_dentry_list(&dispose);
1063         return freed;
1064 }
1065 
1066 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1067                 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1068 {
1069         struct list_head *freeable = arg;
1070         struct dentry   *dentry = container_of(item, struct dentry, d_lru);
1071 
1072         /*
1073          * we are inverting the lru lock/dentry->d_lock here,
1074          * so use a trylock. If we fail to get the lock, just skip
1075          * it
1076          */
1077         if (!spin_trylock(&dentry->d_lock))
1078                 return LRU_SKIP;
1079 
1080         d_lru_shrink_move(lru, dentry, freeable);
1081         spin_unlock(&dentry->d_lock);
1082 
1083         return LRU_REMOVED;
1084 }
1085 
1086 
1087 /**
1088  * shrink_dcache_sb - shrink dcache for a superblock
1089  * @sb: superblock
1090  *
1091  * Shrink the dcache for the specified super block. This is used to free
1092  * the dcache before unmounting a file system.
1093  */
1094 void shrink_dcache_sb(struct super_block *sb)
1095 {
1096         long freed;
1097 
1098         do {
1099                 LIST_HEAD(dispose);
1100 
1101                 freed = list_lru_walk(&sb->s_dentry_lru,
1102                         dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1103 
1104                 this_cpu_sub(nr_dentry_unused, freed);
1105                 shrink_dentry_list(&dispose);
1106         } while (freed > 0);
1107 }
1108 EXPORT_SYMBOL(shrink_dcache_sb);
1109 
1110 /**
1111  * enum d_walk_ret - action to talke during tree walk
1112  * @D_WALK_CONTINUE:    contrinue walk
1113  * @D_WALK_QUIT:        quit walk
1114  * @D_WALK_NORETRY:     quit when retry is needed
1115  * @D_WALK_SKIP:        skip this dentry and its children
1116  */
1117 enum d_walk_ret {
1118         D_WALK_CONTINUE,
1119         D_WALK_QUIT,
1120         D_WALK_NORETRY,
1121         D_WALK_SKIP,
1122 };
1123 
1124 /**
1125  * d_walk - walk the dentry tree
1126  * @parent:     start of walk
1127  * @data:       data passed to @enter() and @finish()
1128  * @enter:      callback when first entering the dentry
1129  * @finish:     callback when successfully finished the walk
1130  *
1131  * The @enter() and @finish() callbacks are called with d_lock held.
1132  */
1133 static void d_walk(struct dentry *parent, void *data,
1134                    enum d_walk_ret (*enter)(void *, struct dentry *),
1135                    void (*finish)(void *))
1136 {
1137         struct dentry *this_parent;
1138         struct list_head *next;
1139         unsigned seq = 0;
1140         enum d_walk_ret ret;
1141         bool retry = true;
1142 
1143 again:
1144         read_seqbegin_or_lock(&rename_lock, &seq);
1145         this_parent = parent;
1146         spin_lock(&this_parent->d_lock);
1147 
1148         ret = enter(data, this_parent);
1149         switch (ret) {
1150         case D_WALK_CONTINUE:
1151                 break;
1152         case D_WALK_QUIT:
1153         case D_WALK_SKIP:
1154                 goto out_unlock;
1155         case D_WALK_NORETRY:
1156                 retry = false;
1157                 break;
1158         }
1159 repeat:
1160         next = this_parent->d_subdirs.next;
1161 resume:
1162         while (next != &this_parent->d_subdirs) {
1163                 struct list_head *tmp = next;
1164                 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1165                 next = tmp->next;
1166 
1167                 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1168 
1169                 ret = enter(data, dentry);
1170                 switch (ret) {
1171                 case D_WALK_CONTINUE:
1172                         break;
1173                 case D_WALK_QUIT:
1174                         spin_unlock(&dentry->d_lock);
1175                         goto out_unlock;
1176                 case D_WALK_NORETRY:
1177                         retry = false;
1178                         break;
1179                 case D_WALK_SKIP:
1180                         spin_unlock(&dentry->d_lock);
1181                         continue;
1182                 }
1183 
1184                 if (!list_empty(&dentry->d_subdirs)) {
1185                         spin_unlock(&this_parent->d_lock);
1186                         spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1187                         this_parent = dentry;
1188                         spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1189                         goto repeat;
1190                 }
1191                 spin_unlock(&dentry->d_lock);
1192         }
1193         /*
1194          * All done at this level ... ascend and resume the search.
1195          */
1196         rcu_read_lock();
1197 ascend:
1198         if (this_parent != parent) {
1199                 struct dentry *child = this_parent;
1200                 this_parent = child->d_parent;
1201 
1202                 spin_unlock(&child->d_lock);
1203                 spin_lock(&this_parent->d_lock);
1204 
1205                 /* might go back up the wrong parent if we have had a rename. */
1206                 if (need_seqretry(&rename_lock, seq))
1207                         goto rename_retry;
1208                 next = child->d_child.next;
1209                 while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
1210                         if (next == &this_parent->d_subdirs)
1211                                 goto ascend;
1212                         child = list_entry(next, struct dentry, d_child);
1213                         next = next->next;
1214                 }
1215                 rcu_read_unlock();
1216                 goto resume;
1217         }
1218         if (need_seqretry(&rename_lock, seq))
1219                 goto rename_retry;
1220         rcu_read_unlock();
1221         if (finish)
1222                 finish(data);
1223 
1224 out_unlock:
1225         spin_unlock(&this_parent->d_lock);
1226         done_seqretry(&rename_lock, seq);
1227         return;
1228 
1229 rename_retry:
1230         spin_unlock(&this_parent->d_lock);
1231         rcu_read_unlock();
1232         BUG_ON(seq & 1);
1233         if (!retry)
1234                 return;
1235         seq = 1;
1236         goto again;
1237 }
1238 
1239 /*
1240  * Search for at least 1 mount point in the dentry's subdirs.
1241  * We descend to the next level whenever the d_subdirs
1242  * list is non-empty and continue searching.
1243  */
1244 
1245 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1246 {
1247         int *ret = data;
1248         if (d_mountpoint(dentry)) {
1249                 *ret = 1;
1250                 return D_WALK_QUIT;
1251         }
1252         return D_WALK_CONTINUE;
1253 }
1254 
1255 /**
1256  * have_submounts - check for mounts over a dentry
1257  * @parent: dentry to check.
1258  *
1259  * Return true if the parent or its subdirectories contain
1260  * a mount point
1261  */
1262 int have_submounts(struct dentry *parent)
1263 {
1264         int ret = 0;
1265 
1266         d_walk(parent, &ret, check_mount, NULL);
1267 
1268         return ret;
1269 }
1270 EXPORT_SYMBOL(have_submounts);
1271 
1272 /*
1273  * Called by mount code to set a mountpoint and check if the mountpoint is
1274  * reachable (e.g. NFS can unhash a directory dentry and then the complete
1275  * subtree can become unreachable).
1276  *
1277  * Only one of d_invalidate() and d_set_mounted() must succeed.  For
1278  * this reason take rename_lock and d_lock on dentry and ancestors.
1279  */
1280 int d_set_mounted(struct dentry *dentry)
1281 {
1282         struct dentry *p;
1283         int ret = -ENOENT;
1284         write_seqlock(&rename_lock);
1285         for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1286                 /* Need exclusion wrt. d_invalidate() */
1287                 spin_lock(&p->d_lock);
1288                 if (unlikely(d_unhashed(p))) {
1289                         spin_unlock(&p->d_lock);
1290                         goto out;
1291                 }
1292                 spin_unlock(&p->d_lock);
1293         }
1294         spin_lock(&dentry->d_lock);
1295         if (!d_unlinked(dentry)) {
1296                 dentry->d_flags |= DCACHE_MOUNTED;
1297                 ret = 0;
1298         }
1299         spin_unlock(&dentry->d_lock);
1300 out:
1301         write_sequnlock(&rename_lock);
1302         return ret;
1303 }
1304 
1305 /*
1306  * Search the dentry child list of the specified parent,
1307  * and move any unused dentries to the end of the unused
1308  * list for prune_dcache(). We descend to the next level
1309  * whenever the d_subdirs list is non-empty and continue
1310  * searching.
1311  *
1312  * It returns zero iff there are no unused children,
1313  * otherwise  it returns the number of children moved to
1314  * the end of the unused list. This may not be the total
1315  * number of unused children, because select_parent can
1316  * drop the lock and return early due to latency
1317  * constraints.
1318  */
1319 
1320 struct select_data {
1321         struct dentry *start;
1322         struct list_head dispose;
1323         int found;
1324 };
1325 
1326 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1327 {
1328         struct select_data *data = _data;
1329         enum d_walk_ret ret = D_WALK_CONTINUE;
1330 
1331         if (data->start == dentry)
1332                 goto out;
1333 
1334         if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1335                 data->found++;
1336         } else {
1337                 if (dentry->d_flags & DCACHE_LRU_LIST)
1338                         d_lru_del(dentry);
1339                 if (!dentry->d_lockref.count) {
1340                         d_shrink_add(dentry, &data->dispose);
1341                         data->found++;
1342                 }
1343         }
1344         /*
1345          * We can return to the caller if we have found some (this
1346          * ensures forward progress). We'll be coming back to find
1347          * the rest.
1348          */
1349         if (!list_empty(&data->dispose))
1350                 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1351 out:
1352         return ret;
1353 }
1354 
1355 /**
1356  * shrink_dcache_parent - prune dcache
1357  * @parent: parent of entries to prune
1358  *
1359  * Prune the dcache to remove unused children of the parent dentry.
1360  */
1361 void shrink_dcache_parent(struct dentry *parent)
1362 {
1363         for (;;) {
1364                 struct select_data data;
1365 
1366                 INIT_LIST_HEAD(&data.dispose);
1367                 data.start = parent;
1368                 data.found = 0;
1369 
1370                 d_walk(parent, &data, select_collect, NULL);
1371                 if (!data.found)
1372                         break;
1373 
1374                 shrink_dentry_list(&data.dispose);
1375                 cond_resched();
1376         }
1377 }
1378 EXPORT_SYMBOL(shrink_dcache_parent);
1379 
1380 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1381 {
1382         /* it has busy descendents; complain about those instead */
1383         if (!list_empty(&dentry->d_subdirs))
1384                 return D_WALK_CONTINUE;
1385 
1386         /* root with refcount 1 is fine */
1387         if (dentry == _data && dentry->d_lockref.count == 1)
1388                 return D_WALK_CONTINUE;
1389 
1390         printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1391                         " still in use (%d) [unmount of %s %s]\n",
1392                        dentry,
1393                        dentry->d_inode ?
1394                        dentry->d_inode->i_ino : 0UL,
1395                        dentry,
1396                        dentry->d_lockref.count,
1397                        dentry->d_sb->s_type->name,
1398                        dentry->d_sb->s_id);
1399         WARN_ON(1);
1400         return D_WALK_CONTINUE;
1401 }
1402 
1403 static void do_one_tree(struct dentry *dentry)
1404 {
1405         shrink_dcache_parent(dentry);
1406         d_walk(dentry, dentry, umount_check, NULL);
1407         d_drop(dentry);
1408         dput(dentry);
1409 }
1410 
1411 /*
1412  * destroy the dentries attached to a superblock on unmounting
1413  */
1414 void shrink_dcache_for_umount(struct super_block *sb)
1415 {
1416         struct dentry *dentry;
1417 
1418         WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1419 
1420         dentry = sb->s_root;
1421         sb->s_root = NULL;
1422         do_one_tree(dentry);
1423 
1424         while (!hlist_bl_empty(&sb->s_anon)) {
1425                 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1426                 do_one_tree(dentry);
1427         }
1428 }
1429 
1430 struct detach_data {
1431         struct select_data select;
1432         struct dentry *mountpoint;
1433 };
1434 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1435 {
1436         struct detach_data *data = _data;
1437 
1438         if (d_mountpoint(dentry)) {
1439                 __dget_dlock(dentry);
1440                 data->mountpoint = dentry;
1441                 return D_WALK_QUIT;
1442         }
1443 
1444         return select_collect(&data->select, dentry);
1445 }
1446 
1447 static void check_and_drop(void *_data)
1448 {
1449         struct detach_data *data = _data;
1450 
1451         if (!data->mountpoint && !data->select.found)
1452                 __d_drop(data->select.start);
1453 }
1454 
1455 /**
1456  * d_invalidate - detach submounts, prune dcache, and drop
1457  * @dentry: dentry to invalidate (aka detach, prune and drop)
1458  *
1459  * no dcache lock.
1460  *
1461  * The final d_drop is done as an atomic operation relative to
1462  * rename_lock ensuring there are no races with d_set_mounted.  This
1463  * ensures there are no unhashed dentries on the path to a mountpoint.
1464  */
1465 void d_invalidate(struct dentry *dentry)
1466 {
1467         /*
1468          * If it's already been dropped, return OK.
1469          */
1470         spin_lock(&dentry->d_lock);
1471         if (d_unhashed(dentry)) {
1472                 spin_unlock(&dentry->d_lock);
1473                 return;
1474         }
1475         spin_unlock(&dentry->d_lock);
1476 
1477         /* Negative dentries can be dropped without further checks */
1478         if (!dentry->d_inode) {
1479                 d_drop(dentry);
1480                 return;
1481         }
1482 
1483         for (;;) {
1484                 struct detach_data data;
1485 
1486                 data.mountpoint = NULL;
1487                 INIT_LIST_HEAD(&data.select.dispose);
1488                 data.select.start = dentry;
1489                 data.select.found = 0;
1490 
1491                 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1492 
1493                 if (data.select.found)
1494                         shrink_dentry_list(&data.select.dispose);
1495 
1496                 if (data.mountpoint) {
1497                         detach_mounts(data.mountpoint);
1498                         dput(data.mountpoint);
1499                 }
1500 
1501                 if (!data.mountpoint && !data.select.found)
1502                         break;
1503 
1504                 cond_resched();
1505         }
1506 }
1507 EXPORT_SYMBOL(d_invalidate);
1508 
1509 /**
1510  * __d_alloc    -       allocate a dcache entry
1511  * @sb: filesystem it will belong to
1512  * @name: qstr of the name
1513  *
1514  * Allocates a dentry. It returns %NULL if there is insufficient memory
1515  * available. On a success the dentry is returned. The name passed in is
1516  * copied and the copy passed in may be reused after this call.
1517  */
1518  
1519 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1520 {
1521         struct dentry *dentry;
1522         char *dname;
1523 
1524         dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1525         if (!dentry)
1526                 return NULL;
1527 
1528         /*
1529          * We guarantee that the inline name is always NUL-terminated.
1530          * This way the memcpy() done by the name switching in rename
1531          * will still always have a NUL at the end, even if we might
1532          * be overwriting an internal NUL character
1533          */
1534         dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1535         if (name->len > DNAME_INLINE_LEN-1) {
1536                 size_t size = offsetof(struct external_name, name[1]);
1537                 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1538                 if (!p) {
1539                         kmem_cache_free(dentry_cache, dentry); 
1540                         return NULL;
1541                 }
1542                 atomic_set(&p->u.count, 1);
1543                 dname = p->name;
1544                 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1545                         kasan_unpoison_shadow(dname,
1546                                 round_up(name->len + 1, sizeof(unsigned long)));
1547         } else  {
1548                 dname = dentry->d_iname;
1549         }       
1550 
1551         dentry->d_name.len = name->len;
1552         dentry->d_name.hash = name->hash;
1553         memcpy(dname, name->name, name->len);
1554         dname[name->len] = 0;
1555 
1556         /* Make sure we always see the terminating NUL character */
1557         smp_wmb();
1558         dentry->d_name.name = dname;
1559 
1560         dentry->d_lockref.count = 1;
1561         dentry->d_flags = 0;
1562         spin_lock_init(&dentry->d_lock);
1563         seqcount_init(&dentry->d_seq);
1564         dentry->d_inode = NULL;
1565         dentry->d_parent = dentry;
1566         dentry->d_sb = sb;
1567         dentry->d_op = NULL;
1568         dentry->d_fsdata = NULL;
1569         INIT_HLIST_BL_NODE(&dentry->d_hash);
1570         INIT_LIST_HEAD(&dentry->d_lru);
1571         INIT_LIST_HEAD(&dentry->d_subdirs);
1572         INIT_HLIST_NODE(&dentry->d_u.d_alias);
1573         INIT_LIST_HEAD(&dentry->d_child);
1574         d_set_d_op(dentry, dentry->d_sb->s_d_op);
1575 
1576         this_cpu_inc(nr_dentry);
1577 
1578         return dentry;
1579 }
1580 
1581 /**
1582  * d_alloc      -       allocate a dcache entry
1583  * @parent: parent of entry to allocate
1584  * @name: qstr of the name
1585  *
1586  * Allocates a dentry. It returns %NULL if there is insufficient memory
1587  * available. On a success the dentry is returned. The name passed in is
1588  * copied and the copy passed in may be reused after this call.
1589  */
1590 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1591 {
1592         struct dentry *dentry = __d_alloc(parent->d_sb, name);
1593         if (!dentry)
1594                 return NULL;
1595 
1596         spin_lock(&parent->d_lock);
1597         /*
1598          * don't need child lock because it is not subject
1599          * to concurrency here
1600          */
1601         __dget_dlock(parent);
1602         dentry->d_parent = parent;
1603         list_add(&dentry->d_child, &parent->d_subdirs);
1604         spin_unlock(&parent->d_lock);
1605 
1606         return dentry;
1607 }
1608 EXPORT_SYMBOL(d_alloc);
1609 
1610 /**
1611  * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1612  * @sb: the superblock
1613  * @name: qstr of the name
1614  *
1615  * For a filesystem that just pins its dentries in memory and never
1616  * performs lookups at all, return an unhashed IS_ROOT dentry.
1617  */
1618 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1619 {
1620         return __d_alloc(sb, name);
1621 }
1622 EXPORT_SYMBOL(d_alloc_pseudo);
1623 
1624 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1625 {
1626         struct qstr q;
1627 
1628         q.name = name;
1629         q.len = strlen(name);
1630         q.hash = full_name_hash(q.name, q.len);
1631         return d_alloc(parent, &q);
1632 }
1633 EXPORT_SYMBOL(d_alloc_name);
1634 
1635 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1636 {
1637         WARN_ON_ONCE(dentry->d_op);
1638         WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH  |
1639                                 DCACHE_OP_COMPARE       |
1640                                 DCACHE_OP_REVALIDATE    |
1641                                 DCACHE_OP_WEAK_REVALIDATE       |
1642                                 DCACHE_OP_DELETE ));
1643         dentry->d_op = op;
1644         if (!op)
1645                 return;
1646         if (op->d_hash)
1647                 dentry->d_flags |= DCACHE_OP_HASH;
1648         if (op->d_compare)
1649                 dentry->d_flags |= DCACHE_OP_COMPARE;
1650         if (op->d_revalidate)
1651                 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1652         if (op->d_weak_revalidate)
1653                 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1654         if (op->d_delete)
1655                 dentry->d_flags |= DCACHE_OP_DELETE;
1656         if (op->d_prune)
1657                 dentry->d_flags |= DCACHE_OP_PRUNE;
1658 
1659 }
1660 EXPORT_SYMBOL(d_set_d_op);
1661 
1662 
1663 /*
1664  * d_set_fallthru - Mark a dentry as falling through to a lower layer
1665  * @dentry - The dentry to mark
1666  *
1667  * Mark a dentry as falling through to the lower layer (as set with
1668  * d_pin_lower()).  This flag may be recorded on the medium.
1669  */
1670 void d_set_fallthru(struct dentry *dentry)
1671 {
1672         spin_lock(&dentry->d_lock);
1673         dentry->d_flags |= DCACHE_FALLTHRU;
1674         spin_unlock(&dentry->d_lock);
1675 }
1676 EXPORT_SYMBOL(d_set_fallthru);
1677 
1678 static unsigned d_flags_for_inode(struct inode *inode)
1679 {
1680         unsigned add_flags = DCACHE_REGULAR_TYPE;
1681 
1682         if (!inode)
1683                 return DCACHE_MISS_TYPE;
1684 
1685         if (S_ISDIR(inode->i_mode)) {
1686                 add_flags = DCACHE_DIRECTORY_TYPE;
1687                 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1688                         if (unlikely(!inode->i_op->lookup))
1689                                 add_flags = DCACHE_AUTODIR_TYPE;
1690                         else
1691                                 inode->i_opflags |= IOP_LOOKUP;
1692                 }
1693                 goto type_determined;
1694         }
1695 
1696         if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1697                 if (unlikely(inode->i_op->follow_link)) {
1698                         add_flags = DCACHE_SYMLINK_TYPE;
1699                         goto type_determined;
1700                 }
1701                 inode->i_opflags |= IOP_NOFOLLOW;
1702         }
1703 
1704         if (unlikely(!S_ISREG(inode->i_mode)))
1705                 add_flags = DCACHE_SPECIAL_TYPE;
1706 
1707 type_determined:
1708         if (unlikely(IS_AUTOMOUNT(inode)))
1709                 add_flags |= DCACHE_NEED_AUTOMOUNT;
1710         return add_flags;
1711 }
1712 
1713 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1714 {
1715         unsigned add_flags = d_flags_for_inode(inode);
1716 
1717         spin_lock(&dentry->d_lock);
1718         dentry->d_flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
1719         dentry->d_flags |= add_flags;
1720         if (inode)
1721                 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1722         dentry->d_inode = inode;
1723         dentry_rcuwalk_barrier(dentry);
1724         spin_unlock(&dentry->d_lock);
1725         fsnotify_d_instantiate(dentry, inode);
1726 }
1727 
1728 /**
1729  * d_instantiate - fill in inode information for a dentry
1730  * @entry: dentry to complete
1731  * @inode: inode to attach to this dentry
1732  *
1733  * Fill in inode information in the entry.
1734  *
1735  * This turns negative dentries into productive full members
1736  * of society.
1737  *
1738  * NOTE! This assumes that the inode count has been incremented
1739  * (or otherwise set) by the caller to indicate that it is now
1740  * in use by the dcache.
1741  */
1742  
1743 void d_instantiate(struct dentry *entry, struct inode * inode)
1744 {
1745         BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1746         if (inode)
1747                 spin_lock(&inode->i_lock);
1748         __d_instantiate(entry, inode);
1749         if (inode)
1750                 spin_unlock(&inode->i_lock);
1751         security_d_instantiate(entry, inode);
1752 }
1753 EXPORT_SYMBOL(d_instantiate);
1754 
1755 /**
1756  * d_instantiate_unique - instantiate a non-aliased dentry
1757  * @entry: dentry to instantiate
1758  * @inode: inode to attach to this dentry
1759  *
1760  * Fill in inode information in the entry. On success, it returns NULL.
1761  * If an unhashed alias of "entry" already exists, then we return the
1762  * aliased dentry instead and drop one reference to inode.
1763  *
1764  * Note that in order to avoid conflicts with rename() etc, the caller
1765  * had better be holding the parent directory semaphore.
1766  *
1767  * This also assumes that the inode count has been incremented
1768  * (or otherwise set) by the caller to indicate that it is now
1769  * in use by the dcache.
1770  */
1771 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1772                                              struct inode *inode)
1773 {
1774         struct dentry *alias;
1775         int len = entry->d_name.len;
1776         const char *name = entry->d_name.name;
1777         unsigned int hash = entry->d_name.hash;
1778 
1779         if (!inode) {
1780                 __d_instantiate(entry, NULL);
1781                 return NULL;
1782         }
1783 
1784         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1785                 /*
1786                  * Don't need alias->d_lock here, because aliases with
1787                  * d_parent == entry->d_parent are not subject to name or
1788                  * parent changes, because the parent inode i_mutex is held.
1789                  */
1790                 if (alias->d_name.hash != hash)
1791                         continue;
1792                 if (alias->d_parent != entry->d_parent)
1793                         continue;
1794                 if (alias->d_name.len != len)
1795                         continue;
1796                 if (dentry_cmp(alias, name, len))
1797                         continue;
1798                 __dget(alias);
1799                 return alias;
1800         }
1801 
1802         __d_instantiate(entry, inode);
1803         return NULL;
1804 }
1805 
1806 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1807 {
1808         struct dentry *result;
1809 
1810         BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1811 
1812         if (inode)
1813                 spin_lock(&inode->i_lock);
1814         result = __d_instantiate_unique(entry, inode);
1815         if (inode)
1816                 spin_unlock(&inode->i_lock);
1817 
1818         if (!result) {
1819                 security_d_instantiate(entry, inode);
1820                 return NULL;
1821         }
1822 
1823         BUG_ON(!d_unhashed(result));
1824         iput(inode);
1825         return result;
1826 }
1827 
1828 EXPORT_SYMBOL(d_instantiate_unique);
1829 
1830 /**
1831  * d_instantiate_no_diralias - instantiate a non-aliased dentry
1832  * @entry: dentry to complete
1833  * @inode: inode to attach to this dentry
1834  *
1835  * Fill in inode information in the entry.  If a directory alias is found, then
1836  * return an error (and drop inode).  Together with d_materialise_unique() this
1837  * guarantees that a directory inode may never have more than one alias.
1838  */
1839 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1840 {
1841         BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1842 
1843         spin_lock(&inode->i_lock);
1844         if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1845                 spin_unlock(&inode->i_lock);
1846                 iput(inode);
1847                 return -EBUSY;
1848         }
1849         __d_instantiate(entry, inode);
1850         spin_unlock(&inode->i_lock);
1851         security_d_instantiate(entry, inode);
1852 
1853         return 0;
1854 }
1855 EXPORT_SYMBOL(d_instantiate_no_diralias);
1856 
1857 struct dentry *d_make_root(struct inode *root_inode)
1858 {
1859         struct dentry *res = NULL;
1860 
1861         if (root_inode) {
1862                 static const struct qstr name = QSTR_INIT("/", 1);
1863 
1864                 res = __d_alloc(root_inode->i_sb, &name);
1865                 if (res)
1866                         d_instantiate(res, root_inode);
1867                 else
1868                         iput(root_inode);
1869         }
1870         return res;
1871 }
1872 EXPORT_SYMBOL(d_make_root);
1873 
1874 static struct dentry * __d_find_any_alias(struct inode *inode)
1875 {
1876         struct dentry *alias;
1877 
1878         if (hlist_empty(&inode->i_dentry))
1879                 return NULL;
1880         alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1881         __dget(alias);
1882         return alias;
1883 }
1884 
1885 /**
1886  * d_find_any_alias - find any alias for a given inode
1887  * @inode: inode to find an alias for
1888  *
1889  * If any aliases exist for the given inode, take and return a
1890  * reference for one of them.  If no aliases exist, return %NULL.
1891  */
1892 struct dentry *d_find_any_alias(struct inode *inode)
1893 {
1894         struct dentry *de;
1895 
1896         spin_lock(&inode->i_lock);
1897         de = __d_find_any_alias(inode);
1898         spin_unlock(&inode->i_lock);
1899         return de;
1900 }
1901 EXPORT_SYMBOL(d_find_any_alias);
1902 
1903 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1904 {
1905         static const struct qstr anonstring = QSTR_INIT("/", 1);
1906         struct dentry *tmp;
1907         struct dentry *res;
1908         unsigned add_flags;
1909 
1910         if (!inode)
1911                 return ERR_PTR(-ESTALE);
1912         if (IS_ERR(inode))
1913                 return ERR_CAST(inode);
1914 
1915         res = d_find_any_alias(inode);
1916         if (res)
1917                 goto out_iput;
1918 
1919         tmp = __d_alloc(inode->i_sb, &anonstring);
1920         if (!tmp) {
1921                 res = ERR_PTR(-ENOMEM);
1922                 goto out_iput;
1923         }
1924 
1925         spin_lock(&inode->i_lock);
1926         res = __d_find_any_alias(inode);
1927         if (res) {
1928                 spin_unlock(&inode->i_lock);
1929                 dput(tmp);
1930                 goto out_iput;
1931         }
1932 
1933         /* attach a disconnected dentry */
1934         add_flags = d_flags_for_inode(inode);
1935 
1936         if (disconnected)
1937                 add_flags |= DCACHE_DISCONNECTED;
1938 
1939         spin_lock(&tmp->d_lock);
1940         tmp->d_inode = inode;
1941         tmp->d_flags |= add_flags;
1942         hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1943         hlist_bl_lock(&tmp->d_sb->s_anon);
1944         hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1945         hlist_bl_unlock(&tmp->d_sb->s_anon);
1946         spin_unlock(&tmp->d_lock);
1947         spin_unlock(&inode->i_lock);
1948         security_d_instantiate(tmp, inode);
1949 
1950         return tmp;
1951 
1952  out_iput:
1953         if (res && !IS_ERR(res))
1954                 security_d_instantiate(res, inode);
1955         iput(inode);
1956         return res;
1957 }
1958 
1959 /**
1960  * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1961  * @inode: inode to allocate the dentry for
1962  *
1963  * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1964  * similar open by handle operations.  The returned dentry may be anonymous,
1965  * or may have a full name (if the inode was already in the cache).
1966  *
1967  * When called on a directory inode, we must ensure that the inode only ever
1968  * has one dentry.  If a dentry is found, that is returned instead of
1969  * allocating a new one.
1970  *
1971  * On successful return, the reference to the inode has been transferred
1972  * to the dentry.  In case of an error the reference on the inode is released.
1973  * To make it easier to use in export operations a %NULL or IS_ERR inode may
1974  * be passed in and the error will be propagated to the return value,
1975  * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1976  */
1977 struct dentry *d_obtain_alias(struct inode *inode)
1978 {
1979         return __d_obtain_alias(inode, 1);
1980 }
1981 EXPORT_SYMBOL(d_obtain_alias);
1982 
1983 /**
1984  * d_obtain_root - find or allocate a dentry for a given inode
1985  * @inode: inode to allocate the dentry for
1986  *
1987  * Obtain an IS_ROOT dentry for the root of a filesystem.
1988  *
1989  * We must ensure that directory inodes only ever have one dentry.  If a
1990  * dentry is found, that is returned instead of allocating a new one.
1991  *
1992  * On successful return, the reference to the inode has been transferred
1993  * to the dentry.  In case of an error the reference on the inode is
1994  * released.  A %NULL or IS_ERR inode may be passed in and will be the
1995  * error will be propagate to the return value, with a %NULL @inode
1996  * replaced by ERR_PTR(-ESTALE).
1997  */
1998 struct dentry *d_obtain_root(struct inode *inode)
1999 {
2000         return __d_obtain_alias(inode, 0);
2001 }
2002 EXPORT_SYMBOL(d_obtain_root);
2003 
2004 /**
2005  * d_add_ci - lookup or allocate new dentry with case-exact name
2006  * @inode:  the inode case-insensitive lookup has found
2007  * @dentry: the negative dentry that was passed to the parent's lookup func
2008  * @name:   the case-exact name to be associated with the returned dentry
2009  *
2010  * This is to avoid filling the dcache with case-insensitive names to the
2011  * same inode, only the actual correct case is stored in the dcache for
2012  * case-insensitive filesystems.
2013  *
2014  * For a case-insensitive lookup match and if the the case-exact dentry
2015  * already exists in in the dcache, use it and return it.
2016  *
2017  * If no entry exists with the exact case name, allocate new dentry with
2018  * the exact case, and return the spliced entry.
2019  */
2020 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2021                         struct qstr *name)
2022 {
2023         struct dentry *found;
2024         struct dentry *new;
2025 
2026         /*
2027          * First check if a dentry matching the name already exists,
2028          * if not go ahead and create it now.
2029          */
2030         found = d_hash_and_lookup(dentry->d_parent, name);
2031         if (!found) {
2032                 new = d_alloc(dentry->d_parent, name);
2033                 if (!new) {
2034                         found = ERR_PTR(-ENOMEM);
2035                 } else {
2036                         found = d_splice_alias(inode, new);
2037                         if (found) {
2038                                 dput(new);
2039                                 return found;
2040                         }
2041                         return new;
2042                 }
2043         }
2044         iput(inode);
2045         return found;
2046 }
2047 EXPORT_SYMBOL(d_add_ci);
2048 
2049 /*
2050  * Do the slow-case of the dentry name compare.
2051  *
2052  * Unlike the dentry_cmp() function, we need to atomically
2053  * load the name and length information, so that the
2054  * filesystem can rely on them, and can use the 'name' and
2055  * 'len' information without worrying about walking off the
2056  * end of memory etc.
2057  *
2058  * Thus the read_seqcount_retry() and the "duplicate" info
2059  * in arguments (the low-level filesystem should not look
2060  * at the dentry inode or name contents directly, since
2061  * rename can change them while we're in RCU mode).
2062  */
2063 enum slow_d_compare {
2064         D_COMP_OK,
2065         D_COMP_NOMATCH,
2066         D_COMP_SEQRETRY,
2067 };
2068 
2069 static noinline enum slow_d_compare slow_dentry_cmp(
2070                 const struct dentry *parent,
2071                 struct dentry *dentry,
2072                 unsigned int seq,
2073                 const struct qstr *name)
2074 {
2075         int tlen = dentry->d_name.len;
2076         const char *tname = dentry->d_name.name;
2077 
2078         if (read_seqcount_retry(&dentry->d_seq, seq)) {
2079                 cpu_relax();
2080                 return D_COMP_SEQRETRY;
2081         }
2082         if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2083                 return D_COMP_NOMATCH;
2084         return D_COMP_OK;
2085 }
2086 
2087 /**
2088  * __d_lookup_rcu - search for a dentry (racy, store-free)
2089  * @parent: parent dentry
2090  * @name: qstr of name we wish to find
2091  * @seqp: returns d_seq value at the point where the dentry was found
2092  * Returns: dentry, or NULL
2093  *
2094  * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2095  * resolution (store-free path walking) design described in
2096  * Documentation/filesystems/path-lookup.txt.
2097  *
2098  * This is not to be used outside core vfs.
2099  *
2100  * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2101  * held, and rcu_read_lock held. The returned dentry must not be stored into
2102  * without taking d_lock and checking d_seq sequence count against @seq
2103  * returned here.
2104  *
2105  * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2106  * function.
2107  *
2108  * Alternatively, __d_lookup_rcu may be called again to look up the child of
2109  * the returned dentry, so long as its parent's seqlock is checked after the
2110  * child is looked up. Thus, an interlocking stepping of sequence lock checks
2111  * is formed, giving integrity down the path walk.
2112  *
2113  * NOTE! The caller *has* to check the resulting dentry against the sequence
2114  * number we've returned before using any of the resulting dentry state!
2115  */
2116 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2117                                 const struct qstr *name,
2118                                 unsigned *seqp)
2119 {
2120         u64 hashlen = name->hash_len;
2121         const unsigned char *str = name->name;
2122         struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2123         struct hlist_bl_node *node;
2124         struct dentry *dentry;
2125 
2126         /*
2127          * Note: There is significant duplication with __d_lookup_rcu which is
2128          * required to prevent single threaded performance regressions
2129          * especially on architectures where smp_rmb (in seqcounts) are costly.
2130          * Keep the two functions in sync.
2131          */
2132 
2133         /*
2134          * The hash list is protected using RCU.
2135          *
2136          * Carefully use d_seq when comparing a candidate dentry, to avoid
2137          * races with d_move().
2138          *
2139          * It is possible that concurrent renames can mess up our list
2140          * walk here and result in missing our dentry, resulting in the
2141          * false-negative result. d_lookup() protects against concurrent
2142          * renames using rename_lock seqlock.
2143          *
2144          * See Documentation/filesystems/path-lookup.txt for more details.
2145          */
2146         hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2147                 unsigned seq;
2148 
2149 seqretry:
2150                 /*
2151                  * The dentry sequence count protects us from concurrent
2152                  * renames, and thus protects parent and name fields.
2153                  *
2154                  * The caller must perform a seqcount check in order
2155                  * to do anything useful with the returned dentry.
2156                  *
2157                  * NOTE! We do a "raw" seqcount_begin here. That means that
2158                  * we don't wait for the sequence count to stabilize if it
2159                  * is in the middle of a sequence change. If we do the slow
2160                  * dentry compare, we will do seqretries until it is stable,
2161                  * and if we end up with a successful lookup, we actually
2162                  * want to exit RCU lookup anyway.
2163                  */
2164                 seq = raw_seqcount_begin(&dentry->d_seq);
2165                 if (dentry->d_parent != parent)
2166                         continue;
2167                 if (d_unhashed(dentry))
2168                         continue;
2169 
2170                 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2171                         if (dentry->d_name.hash != hashlen_hash(hashlen))
2172                                 continue;
2173                         *seqp = seq;
2174                         switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2175                         case D_COMP_OK:
2176                                 return dentry;
2177                         case D_COMP_NOMATCH:
2178                                 continue;
2179                         default:
2180                                 goto seqretry;
2181                         }
2182                 }
2183 
2184                 if (dentry->d_name.hash_len != hashlen)
2185                         continue;
2186                 *seqp = seq;
2187                 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2188                         return dentry;
2189         }
2190         return NULL;
2191 }
2192 
2193 /**
2194  * d_lookup - search for a dentry
2195  * @parent: parent dentry
2196  * @name: qstr of name we wish to find
2197  * Returns: dentry, or NULL
2198  *
2199  * d_lookup searches the children of the parent dentry for the name in
2200  * question. If the dentry is found its reference count is incremented and the
2201  * dentry is returned. The caller must use dput to free the entry when it has
2202  * finished using it. %NULL is returned if the dentry does not exist.
2203  */
2204 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2205 {
2206         struct dentry *dentry;
2207         unsigned seq;
2208 
2209         do {
2210                 seq = read_seqbegin(&rename_lock);
2211                 dentry = __d_lookup(parent, name);
2212                 if (dentry)
2213                         break;
2214         } while (read_seqretry(&rename_lock, seq));
2215         return dentry;
2216 }
2217 EXPORT_SYMBOL(d_lookup);
2218 
2219 /**
2220  * __d_lookup - search for a dentry (racy)
2221  * @parent: parent dentry
2222  * @name: qstr of name we wish to find
2223  * Returns: dentry, or NULL
2224  *
2225  * __d_lookup is like d_lookup, however it may (rarely) return a
2226  * false-negative result due to unrelated rename activity.
2227  *
2228  * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2229  * however it must be used carefully, eg. with a following d_lookup in
2230  * the case of failure.
2231  *
2232  * __d_lookup callers must be commented.
2233  */
2234 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2235 {
2236         unsigned int len = name->len;
2237         unsigned int hash = name->hash;
2238         const unsigned char *str = name->name;
2239         struct hlist_bl_head *b = d_hash(parent, hash);
2240         struct hlist_bl_node *node;
2241         struct dentry *found = NULL;
2242         struct dentry *dentry;
2243 
2244         /*
2245          * Note: There is significant duplication with __d_lookup_rcu which is
2246          * required to prevent single threaded performance regressions
2247          * especially on architectures where smp_rmb (in seqcounts) are costly.
2248          * Keep the two functions in sync.
2249          */
2250 
2251         /*
2252          * The hash list is protected using RCU.
2253          *
2254          * Take d_lock when comparing a candidate dentry, to avoid races
2255          * with d_move().
2256          *
2257          * It is possible that concurrent renames can mess up our list
2258          * walk here and result in missing our dentry, resulting in the
2259          * false-negative result. d_lookup() protects against concurrent
2260          * renames using rename_lock seqlock.
2261          *
2262          * See Documentation/filesystems/path-lookup.txt for more details.
2263          */
2264         rcu_read_lock();
2265         
2266         hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2267 
2268                 if (dentry->d_name.hash != hash)
2269                         continue;
2270 
2271                 spin_lock(&dentry->d_lock);
2272                 if (dentry->d_parent != parent)
2273                         goto next;
2274                 if (d_unhashed(dentry))
2275                         goto next;
2276 
2277                 /*
2278                  * It is safe to compare names since d_move() cannot
2279                  * change the qstr (protected by d_lock).
2280                  */
2281                 if (parent->d_flags & DCACHE_OP_COMPARE) {
2282                         int tlen = dentry->d_name.len;
2283                         const char *tname = dentry->d_name.name;
2284                         if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2285                                 goto next;
2286                 } else {
2287                         if (dentry->d_name.len != len)
2288                                 goto next;
2289                         if (dentry_cmp(dentry, str, len))
2290                                 goto next;
2291                 }
2292 
2293                 dentry->d_lockref.count++;
2294                 found = dentry;
2295                 spin_unlock(&dentry->d_lock);
2296                 break;
2297 next:
2298                 spin_unlock(&dentry->d_lock);
2299         }
2300         rcu_read_unlock();
2301 
2302         return found;
2303 }
2304 
2305 /**
2306  * d_hash_and_lookup - hash the qstr then search for a dentry
2307  * @dir: Directory to search in
2308  * @name: qstr of name we wish to find
2309  *
2310  * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2311  */
2312 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2313 {
2314         /*
2315          * Check for a fs-specific hash function. Note that we must
2316          * calculate the standard hash first, as the d_op->d_hash()
2317          * routine may choose to leave the hash value unchanged.
2318          */
2319         name->hash = full_name_hash(name->name, name->len);
2320         if (dir->d_flags & DCACHE_OP_HASH) {
2321                 int err = dir->d_op->d_hash(dir, name);
2322                 if (unlikely(err < 0))
2323                         return ERR_PTR(err);
2324         }
2325         return d_lookup(dir, name);
2326 }
2327 EXPORT_SYMBOL(d_hash_and_lookup);
2328 
2329 /*
2330  * When a file is deleted, we have two options:
2331  * - turn this dentry into a negative dentry
2332  * - unhash this dentry and free it.
2333  *
2334  * Usually, we want to just turn this into
2335  * a negative dentry, but if anybody else is
2336  * currently using the dentry or the inode
2337  * we can't do that and we fall back on removing
2338  * it from the hash queues and waiting for
2339  * it to be deleted later when it has no users
2340  */
2341  
2342 /**
2343  * d_delete - delete a dentry
2344  * @dentry: The dentry to delete
2345  *
2346  * Turn the dentry into a negative dentry if possible, otherwise
2347  * remove it from the hash queues so it can be deleted later
2348  */
2349  
2350 void d_delete(struct dentry * dentry)
2351 {
2352         struct inode *inode;
2353         int isdir = 0;
2354         /*
2355          * Are we the only user?
2356          */
2357 again:
2358         spin_lock(&dentry->d_lock);
2359         inode = dentry->d_inode;
2360         isdir = S_ISDIR(inode->i_mode);
2361         if (dentry->d_lockref.count == 1) {
2362                 if (!spin_trylock(&inode->i_lock)) {
2363                         spin_unlock(&dentry->d_lock);
2364                         cpu_relax();
2365                         goto again;
2366                 }
2367                 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2368                 dentry_unlink_inode(dentry);
2369                 fsnotify_nameremove(dentry, isdir);
2370                 return;
2371         }
2372 
2373         if (!d_unhashed(dentry))
2374                 __d_drop(dentry);
2375 
2376         spin_unlock(&dentry->d_lock);
2377 
2378         fsnotify_nameremove(dentry, isdir);
2379 }
2380 EXPORT_SYMBOL(d_delete);
2381 
2382 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2383 {
2384         BUG_ON(!d_unhashed(entry));
2385         hlist_bl_lock(b);
2386         entry->d_flags |= DCACHE_RCUACCESS;
2387         hlist_bl_add_head_rcu(&entry->d_hash, b);
2388         hlist_bl_unlock(b);
2389 }
2390 
2391 static void _d_rehash(struct dentry * entry)
2392 {
2393         __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2394 }
2395 
2396 /**
2397  * d_rehash     - add an entry back to the hash
2398  * @entry: dentry to add to the hash
2399  *
2400  * Adds a dentry to the hash according to its name.
2401  */
2402  
2403 void d_rehash(struct dentry * entry)
2404 {
2405         spin_lock(&entry->d_lock);
2406         _d_rehash(entry);
2407         spin_unlock(&entry->d_lock);
2408 }
2409 EXPORT_SYMBOL(d_rehash);
2410 
2411 /**
2412  * dentry_update_name_case - update case insensitive dentry with a new name
2413  * @dentry: dentry to be updated
2414  * @name: new name
2415  *
2416  * Update a case insensitive dentry with new case of name.
2417  *
2418  * dentry must have been returned by d_lookup with name @name. Old and new
2419  * name lengths must match (ie. no d_compare which allows mismatched name
2420  * lengths).
2421  *
2422  * Parent inode i_mutex must be held over d_lookup and into this call (to
2423  * keep renames and concurrent inserts, and readdir(2) away).
2424  */
2425 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2426 {
2427         BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2428         BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2429 
2430         spin_lock(&dentry->d_lock);
2431         write_seqcount_begin(&dentry->d_seq);
2432         memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2433         write_seqcount_end(&dentry->d_seq);
2434         spin_unlock(&dentry->d_lock);
2435 }
2436 EXPORT_SYMBOL(dentry_update_name_case);
2437 
2438 static void swap_names(struct dentry *dentry, struct dentry *target)
2439 {
2440         if (unlikely(dname_external(target))) {
2441                 if (unlikely(dname_external(dentry))) {
2442                         /*
2443                          * Both external: swap the pointers
2444                          */
2445                         swap(target->d_name.name, dentry->d_name.name);
2446                 } else {
2447                         /*
2448                          * dentry:internal, target:external.  Steal target's
2449                          * storage and make target internal.
2450                          */
2451                         memcpy(target->d_iname, dentry->d_name.name,
2452                                         dentry->d_name.len + 1);
2453                         dentry->d_name.name = target->d_name.name;
2454                         target->d_name.name = target->d_iname;
2455                 }
2456         } else {
2457                 if (unlikely(dname_external(dentry))) {
2458                         /*
2459                          * dentry:external, target:internal.  Give dentry's
2460                          * storage to target and make dentry internal
2461                          */
2462                         memcpy(dentry->d_iname, target->d_name.name,
2463                                         target->d_name.len + 1);
2464                         target->d_name.name = dentry->d_name.name;
2465                         dentry->d_name.name = dentry->d_iname;
2466                 } else {
2467                         /*
2468                          * Both are internal.
2469                          */
2470                         unsigned int i;
2471                         BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2472                         kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2473                         kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
2474                         for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2475                                 swap(((long *) &dentry->d_iname)[i],
2476                                      ((long *) &target->d_iname)[i]);
2477                         }
2478                 }
2479         }
2480         swap(dentry->d_name.hash_len, target->d_name.hash_len);
2481 }
2482 
2483 static void copy_name(struct dentry *dentry, struct dentry *target)
2484 {
2485         struct external_name *old_name = NULL;
2486         if (unlikely(dname_external(dentry)))
2487                 old_name = external_name(dentry);
2488         if (unlikely(dname_external(target))) {
2489                 atomic_inc(&external_name(target)->u.count);
2490                 dentry->d_name = target->d_name;
2491         } else {
2492                 memcpy(dentry->d_iname, target->d_name.name,
2493                                 target->d_name.len + 1);
2494                 dentry->d_name.name = dentry->d_iname;
2495                 dentry->d_name.hash_len = target->d_name.hash_len;
2496         }
2497         if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2498                 kfree_rcu(old_name, u.head);
2499 }
2500 
2501 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2502 {
2503         /*
2504          * XXXX: do we really need to take target->d_lock?
2505          */
2506         if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2507                 spin_lock(&target->d_parent->d_lock);
2508         else {
2509                 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2510                         spin_lock(&dentry->d_parent->d_lock);
2511                         spin_lock_nested(&target->d_parent->d_lock,
2512                                                 DENTRY_D_LOCK_NESTED);
2513                 } else {
2514                         spin_lock(&target->d_parent->d_lock);
2515                         spin_lock_nested(&dentry->d_parent->d_lock,
2516                                                 DENTRY_D_LOCK_NESTED);
2517                 }
2518         }
2519         if (target < dentry) {
2520                 spin_lock_nested(&target->d_lock, 2);
2521                 spin_lock_nested(&dentry->d_lock, 3);
2522         } else {
2523                 spin_lock_nested(&dentry->d_lock, 2);
2524                 spin_lock_nested(&target->d_lock, 3);
2525         }
2526 }
2527 
2528 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2529 {
2530         if (target->d_parent != dentry->d_parent)
2531                 spin_unlock(&dentry->d_parent->d_lock);
2532         if (target->d_parent != target)
2533                 spin_unlock(&target->d_parent->d_lock);
2534         spin_unlock(&target->d_lock);
2535         spin_unlock(&dentry->d_lock);
2536 }
2537 
2538 /*
2539  * When switching names, the actual string doesn't strictly have to
2540  * be preserved in the target - because we're dropping the target
2541  * anyway. As such, we can just do a simple memcpy() to copy over
2542  * the new name before we switch, unless we are going to rehash
2543  * it.  Note that if we *do* unhash the target, we are not allowed
2544  * to rehash it without giving it a new name/hash key - whether
2545  * we swap or overwrite the names here, resulting name won't match
2546  * the reality in filesystem; it's only there for d_path() purposes.
2547  * Note that all of this is happening under rename_lock, so the
2548  * any hash lookup seeing it in the middle of manipulations will
2549  * be discarded anyway.  So we do not care what happens to the hash
2550  * key in that case.
2551  */
2552 /*
2553  * __d_move - move a dentry
2554  * @dentry: entry to move
2555  * @target: new dentry
2556  * @exchange: exchange the two dentries
2557  *
2558  * Update the dcache to reflect the move of a file name. Negative
2559  * dcache entries should not be moved in this way. Caller must hold
2560  * rename_lock, the i_mutex of the source and target directories,
2561  * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2562  */
2563 static void __d_move(struct dentry *dentry, struct dentry *target,
2564                      bool exchange)
2565 {
2566         if (!dentry->d_inode)
2567                 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2568 
2569         BUG_ON(d_ancestor(dentry, target));
2570         BUG_ON(d_ancestor(target, dentry));
2571 
2572         dentry_lock_for_move(dentry, target);
2573 
2574         write_seqcount_begin(&dentry->d_seq);
2575         write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2576 
2577         /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2578 
2579         /*
2580          * Move the dentry to the target hash queue. Don't bother checking
2581          * for the same hash queue because of how unlikely it is.
2582          */
2583         __d_drop(dentry);
2584         __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2585 
2586         /*
2587          * Unhash the target (d_delete() is not usable here).  If exchanging
2588          * the two dentries, then rehash onto the other's hash queue.
2589          */
2590         __d_drop(target);
2591         if (exchange) {
2592                 __d_rehash(target,
2593                            d_hash(dentry->d_parent, dentry->d_name.hash));
2594         }
2595 
2596         /* Switch the names.. */
2597         if (exchange)
2598                 swap_names(dentry, target);
2599         else
2600                 copy_name(dentry, target);
2601 
2602         /* ... and switch them in the tree */
2603         if (IS_ROOT(dentry)) {
2604                 /* splicing a tree */
2605                 dentry->d_parent = target->d_parent;
2606                 target->d_parent = target;
2607                 list_del_init(&target->d_child);
2608                 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2609         } else {
2610                 /* swapping two dentries */
2611                 swap(dentry->d_parent, target->d_parent);
2612                 list_move(&target->d_child, &target->d_parent->d_subdirs);
2613                 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2614                 if (exchange)
2615                         fsnotify_d_move(target);
2616                 fsnotify_d_move(dentry);
2617         }
2618 
2619         write_seqcount_end(&target->d_seq);
2620         write_seqcount_end(&dentry->d_seq);
2621 
2622         dentry_unlock_for_move(dentry, target);
2623 }
2624 
2625 /*
2626  * d_move - move a dentry
2627  * @dentry: entry to move
2628  * @target: new dentry
2629  *
2630  * Update the dcache to reflect the move of a file name. Negative
2631  * dcache entries should not be moved in this way. See the locking
2632  * requirements for __d_move.
2633  */
2634 void d_move(struct dentry *dentry, struct dentry *target)
2635 {
2636         write_seqlock(&rename_lock);
2637         __d_move(dentry, target, false);
2638         write_sequnlock(&rename_lock);
2639 }
2640 EXPORT_SYMBOL(d_move);
2641 
2642 /*
2643  * d_exchange - exchange two dentries
2644  * @dentry1: first dentry
2645  * @dentry2: second dentry
2646  */
2647 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2648 {
2649         write_seqlock(&rename_lock);
2650 
2651         WARN_ON(!dentry1->d_inode);
2652         WARN_ON(!dentry2->d_inode);
2653         WARN_ON(IS_ROOT(dentry1));
2654         WARN_ON(IS_ROOT(dentry2));
2655 
2656         __d_move(dentry1, dentry2, true);
2657 
2658         write_sequnlock(&rename_lock);
2659 }
2660 
2661 /**
2662  * d_ancestor - search for an ancestor
2663  * @p1: ancestor dentry
2664  * @p2: child dentry
2665  *
2666  * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2667  * an ancestor of p2, else NULL.
2668  */
2669 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2670 {
2671         struct dentry *p;
2672 
2673         for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2674                 if (p->d_parent == p1)
2675                         return p;
2676         }
2677         return NULL;
2678 }
2679 
2680 /*
2681  * This helper attempts to cope with remotely renamed directories
2682  *
2683  * It assumes that the caller is already holding
2684  * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2685  *
2686  * Note: If ever the locking in lock_rename() changes, then please
2687  * remember to update this too...
2688  */
2689 static int __d_unalias(struct inode *inode,
2690                 struct dentry *dentry, struct dentry *alias)
2691 {
2692         struct mutex *m1 = NULL, *m2 = NULL;
2693         int ret = -EBUSY;
2694 
2695         /* If alias and dentry share a parent, then no extra locks required */
2696         if (alias->d_parent == dentry->d_parent)
2697                 goto out_unalias;
2698 
2699         /* See lock_rename() */
2700         if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2701                 goto out_err;
2702         m1 = &dentry->d_sb->s_vfs_rename_mutex;
2703         if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2704                 goto out_err;
2705         m2 = &alias->d_parent->d_inode->i_mutex;
2706 out_unalias:
2707         __d_move(alias, dentry, false);
2708         ret = 0;
2709 out_err:
2710         spin_unlock(&inode->i_lock);
2711         if (m2)
2712                 mutex_unlock(m2);
2713         if (m1)
2714                 mutex_unlock(m1);
2715         return ret;
2716 }
2717 
2718 /**
2719  * d_splice_alias - splice a disconnected dentry into the tree if one exists
2720  * @inode:  the inode which may have a disconnected dentry
2721  * @dentry: a negative dentry which we want to point to the inode.
2722  *
2723  * If inode is a directory and has an IS_ROOT alias, then d_move that in
2724  * place of the given dentry and return it, else simply d_add the inode
2725  * to the dentry and return NULL.
2726  *
2727  * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2728  * we should error out: directories can't have multiple aliases.
2729  *
2730  * This is needed in the lookup routine of any filesystem that is exportable
2731  * (via knfsd) so that we can build dcache paths to directories effectively.
2732  *
2733  * If a dentry was found and moved, then it is returned.  Otherwise NULL
2734  * is returned.  This matches the expected return value of ->lookup.
2735  *
2736  * Cluster filesystems may call this function with a negative, hashed dentry.
2737  * In that case, we know that the inode will be a regular file, and also this
2738  * will only occur during atomic_open. So we need to check for the dentry
2739  * being already hashed only in the final case.
2740  */
2741 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2742 {
2743         if (IS_ERR(inode))
2744                 return ERR_CAST(inode);
2745 
2746         BUG_ON(!d_unhashed(dentry));
2747 
2748         if (!inode) {
2749                 __d_instantiate(dentry, NULL);
2750                 goto out;
2751         }
2752         spin_lock(&inode->i_lock);
2753         if (S_ISDIR(inode->i_mode)) {
2754                 struct dentry *new = __d_find_any_alias(inode);
2755                 if (unlikely(new)) {
2756                         write_seqlock(&rename_lock);
2757                         if (unlikely(d_ancestor(new, dentry))) {
2758                                 write_sequnlock(&rename_lock);
2759                                 spin_unlock(&inode->i_lock);
2760                                 dput(new);
2761                                 new = ERR_PTR(-ELOOP);
2762                                 pr_warn_ratelimited(
2763                                         "VFS: Lookup of '%s' in %s %s"
2764                                         " would have caused loop\n",
2765                                         dentry->d_name.name,
2766                                         inode->i_sb->s_type->name,
2767                                         inode->i_sb->s_id);
2768                         } else if (!IS_ROOT(new)) {
2769                                 int err = __d_unalias(inode, dentry, new);
2770                                 write_sequnlock(&rename_lock);
2771                                 if (err) {
2772                                         dput(new);
2773                                         new = ERR_PTR(err);
2774                                 }
2775                         } else {
2776                                 __d_move(new, dentry, false);
2777                                 write_sequnlock(&rename_lock);
2778                                 spin_unlock(&inode->i_lock);
2779                                 security_d_instantiate(new, inode);
2780                         }
2781                         iput(inode);
2782                         return new;
2783                 }
2784         }
2785         /* already taking inode->i_lock, so d_add() by hand */
2786         __d_instantiate(dentry, inode);
2787         spin_unlock(&inode->i_lock);
2788 out:
2789         security_d_instantiate(dentry, inode);
2790         d_rehash(dentry);
2791         return NULL;
2792 }
2793 EXPORT_SYMBOL(d_splice_alias);
2794 
2795 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2796 {
2797         *buflen -= namelen;
2798         if (*buflen < 0)
2799                 return -ENAMETOOLONG;
2800         *buffer -= namelen;
2801         memcpy(*buffer, str, namelen);
2802         return 0;
2803 }
2804 
2805 /**
2806  * prepend_name - prepend a pathname in front of current buffer pointer
2807  * @buffer: buffer pointer
2808  * @buflen: allocated length of the buffer
2809  * @name:   name string and length qstr structure
2810  *
2811  * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2812  * make sure that either the old or the new name pointer and length are
2813  * fetched. However, there may be mismatch between length and pointer.
2814  * The length cannot be trusted, we need to copy it byte-by-byte until
2815  * the length is reached or a null byte is found. It also prepends "/" at
2816  * the beginning of the name. The sequence number check at the caller will
2817  * retry it again when a d_move() does happen. So any garbage in the buffer
2818  * due to mismatched pointer and length will be discarded.
2819  *
2820  * Data dependency barrier is needed to make sure that we see that terminating
2821  * NUL.  Alpha strikes again, film at 11...
2822  */
2823 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2824 {
2825         const char *dname = ACCESS_ONCE(name->name);
2826         u32 dlen = ACCESS_ONCE(name->len);
2827         char *p;
2828 
2829         smp_read_barrier_depends();
2830 
2831         *buflen -= dlen + 1;
2832         if (*buflen < 0)
2833                 return -ENAMETOOLONG;
2834         p = *buffer -= dlen + 1;
2835         *p++ = '/';
2836         while (dlen--) {
2837                 char c = *dname++;
2838                 if (!c)
2839                         break;
2840                 *p++ = c;
2841         }
2842         return 0;
2843 }
2844 
2845 /**
2846  * prepend_path - Prepend path string to a buffer
2847  * @path: the dentry/vfsmount to report
2848  * @root: root vfsmnt/dentry
2849  * @buffer: pointer to the end of the buffer
2850  * @buflen: pointer to buffer length
2851  *
2852  * The function will first try to write out the pathname without taking any
2853  * lock other than the RCU read lock to make sure that dentries won't go away.
2854  * It only checks the sequence number of the global rename_lock as any change
2855  * in the dentry's d_seq will be preceded by changes in the rename_lock
2856  * sequence number. If the sequence number had been changed, it will restart
2857  * the whole pathname back-tracing sequence again by taking the rename_lock.
2858  * In this case, there is no need to take the RCU read lock as the recursive
2859  * parent pointer references will keep the dentry chain alive as long as no
2860  * rename operation is performed.
2861  */
2862 static int prepend_path(const struct path *path,
2863                         const struct path *root,
2864                         char **buffer, int *buflen)
2865 {
2866         struct dentry *dentry;
2867         struct vfsmount *vfsmnt;
2868         struct mount *mnt;
2869         int error = 0;
2870         unsigned seq, m_seq = 0;
2871         char *bptr;
2872         int blen;
2873 
2874         rcu_read_lock();
2875 restart_mnt:
2876         read_seqbegin_or_lock(&mount_lock, &m_seq);
2877         seq = 0;
2878         rcu_read_lock();
2879 restart:
2880         bptr = *buffer;
2881         blen = *buflen;
2882         error = 0;
2883         dentry = path->dentry;
2884         vfsmnt = path->mnt;
2885         mnt = real_mount(vfsmnt);
2886         read_seqbegin_or_lock(&rename_lock, &seq);
2887         while (dentry != root->dentry || vfsmnt != root->mnt) {
2888                 struct dentry * parent;
2889 
2890                 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2891                         struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2892                         /* Global root? */
2893                         if (mnt != parent) {
2894                                 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2895                                 mnt = parent;
2896                                 vfsmnt = &mnt->mnt;
2897                                 continue;
2898                         }
2899                         /*
2900                          * Filesystems needing to implement special "root names"
2901                          * should do so with ->d_dname()
2902                          */
2903                         if (IS_ROOT(dentry) &&
2904                            (dentry->d_name.len != 1 ||
2905                             dentry->d_name.name[0] != '/')) {
2906                                 WARN(1, "Root dentry has weird name <%.*s>\n",
2907                                      (int) dentry->d_name.len,
2908                                      dentry->d_name.name);
2909                         }
2910                         if (!error)
2911                                 error = is_mounted(vfsmnt) ? 1 : 2;
2912                         break;
2913                 }
2914                 parent = dentry->d_parent;
2915                 prefetch(parent);
2916                 error = prepend_name(&bptr, &blen, &dentry->d_name);
2917                 if (error)
2918                         break;
2919 
2920                 dentry = parent;
2921         }
2922         if (!(seq & 1))
2923                 rcu_read_unlock();
2924         if (need_seqretry(&rename_lock, seq)) {
2925                 seq = 1;
2926                 goto restart;
2927         }
2928         done_seqretry(&rename_lock, seq);
2929 
2930         if (!(m_seq & 1))
2931                 rcu_read_unlock();
2932         if (need_seqretry(&mount_lock, m_seq)) {
2933                 m_seq = 1;
2934                 goto restart_mnt;
2935         }
2936         done_seqretry(&mount_lock, m_seq);
2937 
2938         if (error >= 0 && bptr == *buffer) {
2939                 if (--blen < 0)
2940                         error = -ENAMETOOLONG;
2941                 else
2942                         *--bptr = '/';
2943         }
2944         *buffer = bptr;
2945         *buflen = blen;
2946         return error;
2947 }
2948 
2949 /**
2950  * __d_path - return the path of a dentry
2951  * @path: the dentry/vfsmount to report
2952  * @root: root vfsmnt/dentry
2953  * @buf: buffer to return value in
2954  * @buflen: buffer length
2955  *
2956  * Convert a dentry into an ASCII path name.
2957  *
2958  * Returns a pointer into the buffer or an error code if the
2959  * path was too long.
2960  *
2961  * "buflen" should be positive.
2962  *
2963  * If the path is not reachable from the supplied root, return %NULL.
2964  */
2965 char *__d_path(const struct path *path,
2966                const struct path *root,
2967                char *buf, int buflen)
2968 {
2969         char *res = buf + buflen;
2970         int error;
2971 
2972         prepend(&res, &buflen, "\0", 1);
2973         error = prepend_path(path, root, &res, &buflen);
2974 
2975         if (error < 0)
2976                 return ERR_PTR(error);
2977         if (error > 0)
2978                 return NULL;
2979         return res;
2980 }
2981 
2982 char *d_absolute_path(const struct path *path,
2983                char *buf, int buflen)
2984 {
2985         struct path root = {};
2986         char *res = buf + buflen;
2987         int error;
2988 
2989         prepend(&res, &buflen, "\0", 1);
2990         error = prepend_path(path, &root, &res, &buflen);
2991 
2992         if (error > 1)
2993                 error = -EINVAL;
2994         if (error < 0)
2995                 return ERR_PTR(error);
2996         return res;
2997 }
2998 
2999 /*
3000  * same as __d_path but appends "(deleted)" for unlinked files.
3001  */
3002 static int path_with_deleted(const struct path *path,
3003                              const struct path *root,
3004                              char **buf, int *buflen)
3005 {
3006         prepend(buf, buflen, "\0", 1);
3007         if (d_unlinked(path->dentry)) {
3008                 int error = prepend(buf, buflen, " (deleted)", 10);
3009                 if (error)
3010                         return error;
3011         }
3012 
3013         return prepend_path(path, root, buf, buflen);
3014 }
3015 
3016 static int prepend_unreachable(char **buffer, int *buflen)
3017 {
3018         return prepend(buffer, buflen, "(unreachable)", 13);
3019 }
3020 
3021 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3022 {
3023         unsigned seq;
3024 
3025         do {
3026                 seq = read_seqcount_begin(&fs->seq);
3027                 *root = fs->root;
3028         } while (read_seqcount_retry(&fs->seq, seq));
3029 }
3030 
3031 /**
3032  * d_path - return the path of a dentry
3033  * @path: path to report
3034  * @buf: buffer to return value in
3035  * @buflen: buffer length
3036  *
3037  * Convert a dentry into an ASCII path name. If the entry has been deleted
3038  * the string " (deleted)" is appended. Note that this is ambiguous.
3039  *
3040  * Returns a pointer into the buffer or an error code if the path was
3041  * too long. Note: Callers should use the returned pointer, not the passed
3042  * in buffer, to use the name! The implementation often starts at an offset
3043  * into the buffer, and may leave 0 bytes at the start.
3044  *
3045  * "buflen" should be positive.
3046  */
3047 char *d_path(const struct path *path, char *buf, int buflen)
3048 {
3049         char *res = buf + buflen;
3050         struct path root;
3051         int error;
3052 
3053         /*
3054          * We have various synthetic filesystems that never get mounted.  On
3055          * these filesystems dentries are never used for lookup purposes, and
3056          * thus don't need to be hashed.  They also don't need a name until a
3057          * user wants to identify the object in /proc/pid/fd/.  The little hack
3058          * below allows us to generate a name for these objects on demand:
3059          *
3060          * Some pseudo inodes are mountable.  When they are mounted
3061          * path->dentry == path->mnt->mnt_root.  In that case don't call d_dname
3062          * and instead have d_path return the mounted path.
3063          */
3064         if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3065             (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3066                 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3067 
3068         rcu_read_lock();
3069         get_fs_root_rcu(current->fs, &root);
3070         error = path_with_deleted(path, &root, &res, &buflen);
3071         rcu_read_unlock();
3072 
3073         if (error < 0)
3074                 res = ERR_PTR(error);
3075         return res;
3076 }
3077 EXPORT_SYMBOL(d_path);
3078 
3079 /*
3080  * Helper function for dentry_operations.d_dname() members
3081  */
3082 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3083                         const char *fmt, ...)
3084 {
3085         va_list args;
3086         char temp[64];
3087         int sz;
3088 
3089         va_start(args, fmt);
3090         sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3091         va_end(args);
3092 
3093         if (sz > sizeof(temp) || sz > buflen)
3094                 return ERR_PTR(-ENAMETOOLONG);
3095 
3096         buffer += buflen - sz;
3097         return memcpy(buffer, temp, sz);
3098 }
3099 
3100 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3101 {
3102         char *end = buffer + buflen;
3103         /* these dentries are never renamed, so d_lock is not needed */
3104         if (prepend(&end, &buflen, " (deleted)", 11) ||
3105             prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3106             prepend(&end, &buflen, "/", 1))  
3107                 end = ERR_PTR(-ENAMETOOLONG);
3108         return end;
3109 }
3110 EXPORT_SYMBOL(simple_dname);
3111 
3112 /*
3113  * Write full pathname from the root of the filesystem into the buffer.
3114  */
3115 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3116 {
3117         struct dentry *dentry;
3118         char *end, *retval;
3119         int len, seq = 0;
3120         int error = 0;
3121 
3122         if (buflen < 2)
3123                 goto Elong;
3124 
3125         rcu_read_lock();
3126 restart:
3127         dentry = d;
3128         end = buf + buflen;
3129         len = buflen;
3130         prepend(&end, &len, "\0", 1);
3131         /* Get '/' right */
3132         retval = end-1;
3133         *retval = '/';
3134         read_seqbegin_or_lock(&rename_lock, &seq);
3135         while (!IS_ROOT(dentry)) {
3136                 struct dentry *parent = dentry->d_parent;
3137 
3138                 prefetch(parent);
3139                 error = prepend_name(&end, &len, &dentry->d_name);
3140                 if (error)
3141                         break;
3142 
3143                 retval = end;
3144                 dentry = parent;
3145         }
3146         if (!(seq & 1))
3147                 rcu_read_unlock();
3148         if (need_seqretry(&rename_lock, seq)) {
3149                 seq = 1;
3150                 goto restart;
3151         }
3152         done_seqretry(&rename_lock, seq);
3153         if (error)
3154                 goto Elong;
3155         return retval;
3156 Elong:
3157         return ERR_PTR(-ENAMETOOLONG);
3158 }
3159 
3160 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3161 {
3162         return __dentry_path(dentry, buf, buflen);
3163 }
3164 EXPORT_SYMBOL(dentry_path_raw);
3165 
3166 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3167 {
3168         char *p = NULL;
3169         char *retval;
3170 
3171         if (d_unlinked(dentry)) {
3172                 p = buf + buflen;
3173                 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3174                         goto Elong;
3175                 buflen++;
3176         }
3177         retval = __dentry_path(dentry, buf, buflen);
3178         if (!IS_ERR(retval) && p)
3179                 *p = '/';       /* restore '/' overriden with '\0' */
3180         return retval;
3181 Elong:
3182         return ERR_PTR(-ENAMETOOLONG);
3183 }
3184 
3185 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3186                                     struct path *pwd)
3187 {
3188         unsigned seq;
3189 
3190         do {
3191                 seq = read_seqcount_begin(&fs->seq);
3192                 *root = fs->root;
3193                 *pwd = fs->pwd;
3194         } while (read_seqcount_retry(&fs->seq, seq));
3195 }
3196 
3197 /*
3198  * NOTE! The user-level library version returns a
3199  * character pointer. The kernel system call just
3200  * returns the length of the buffer filled (which
3201  * includes the ending '\0' character), or a negative
3202  * error value. So libc would do something like
3203  *
3204  *      char *getcwd(char * buf, size_t size)
3205  *      {
3206  *              int retval;
3207  *
3208  *              retval = sys_getcwd(buf, size);
3209  *              if (retval >= 0)
3210  *                      return buf;
3211  *              errno = -retval;
3212  *              return NULL;
3213  *      }
3214  */
3215 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3216 {
3217         int error;
3218         struct path pwd, root;
3219         char *page = __getname();
3220 
3221         if (!page)
3222                 return -ENOMEM;
3223 
3224         rcu_read_lock();
3225         get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3226 
3227         error = -ENOENT;
3228         if (!d_unlinked(pwd.dentry)) {
3229                 unsigned long len;
3230                 char *cwd = page + PATH_MAX;
3231                 int buflen = PATH_MAX;
3232 
3233                 prepend(&cwd, &buflen, "\0", 1);
3234                 error = prepend_path(&pwd, &root, &cwd, &buflen);
3235                 rcu_read_unlock();
3236 
3237                 if (error < 0)
3238                         goto out;
3239 
3240                 /* Unreachable from current root */
3241                 if (error > 0) {
3242                         error = prepend_unreachable(&cwd, &buflen);
3243                         if (error)
3244                                 goto out;
3245                 }
3246 
3247                 error = -ERANGE;
3248                 len = PATH_MAX + page - cwd;
3249                 if (len <= size) {
3250                         error = len;
3251                         if (copy_to_user(buf, cwd, len))
3252                                 error = -EFAULT;
3253                 }
3254         } else {
3255                 rcu_read_unlock();
3256         }
3257 
3258 out:
3259         __putname(page);
3260         return error;
3261 }
3262 
3263 /*
3264  * Test whether new_dentry is a subdirectory of old_dentry.
3265  *
3266  * Trivially implemented using the dcache structure
3267  */
3268 
3269 /**
3270  * is_subdir - is new dentry a subdirectory of old_dentry
3271  * @new_dentry: new dentry
3272  * @old_dentry: old dentry
3273  *
3274  * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3275  * Returns 0 otherwise.
3276  * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3277  */
3278   
3279 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3280 {
3281         int result;
3282         unsigned seq;
3283 
3284         if (new_dentry == old_dentry)
3285                 return 1;
3286 
3287         do {
3288                 /* for restarting inner loop in case of seq retry */
3289                 seq = read_seqbegin(&rename_lock);
3290                 /*
3291                  * Need rcu_readlock to protect against the d_parent trashing
3292                  * due to d_move
3293                  */
3294                 rcu_read_lock();
3295                 if (d_ancestor(old_dentry, new_dentry))
3296                         result = 1;
3297                 else
3298                         result = 0;
3299                 rcu_read_unlock();
3300         } while (read_seqretry(&rename_lock, seq));
3301 
3302         return result;
3303 }
3304 
3305 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3306 {
3307         struct dentry *root = data;
3308         if (dentry != root) {
3309                 if (d_unhashed(dentry) || !dentry->d_inode)
3310                         return D_WALK_SKIP;
3311 
3312                 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3313                         dentry->d_flags |= DCACHE_GENOCIDE;
3314                         dentry->d_lockref.count--;
3315                 }
3316         }
3317         return D_WALK_CONTINUE;
3318 }
3319 
3320 void d_genocide(struct dentry *parent)
3321 {
3322         d_walk(parent, parent, d_genocide_kill, NULL);
3323 }
3324 
3325 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3326 {
3327         inode_dec_link_count(inode);
3328         BUG_ON(dentry->d_name.name != dentry->d_iname ||
3329                 !hlist_unhashed(&dentry->d_u.d_alias) ||
3330                 !d_unlinked(dentry));
3331         spin_lock(&dentry->d_parent->d_lock);
3332         spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3333         dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3334                                 (unsigned long long)inode->i_ino);
3335         spin_unlock(&dentry->d_lock);
3336         spin_unlock(&dentry->d_parent->d_lock);
3337         d_instantiate(dentry, inode);
3338 }
3339 EXPORT_SYMBOL(d_tmpfile);
3340 
3341 static __initdata unsigned long dhash_entries;
3342 static int __init set_dhash_entries(char *str)
3343 {
3344         if (!str)
3345                 return 0;
3346         dhash_entries = simple_strtoul(str, &str, 0);
3347         return 1;
3348 }
3349 __setup("dhash_entries=", set_dhash_entries);
3350 
3351 static void __init dcache_init_early(void)
3352 {
3353         unsigned int loop;
3354 
3355         /* If hashes are distributed across NUMA nodes, defer
3356          * hash allocation until vmalloc space is available.
3357          */
3358         if (hashdist)
3359                 return;
3360 
3361         dentry_hashtable =
3362                 alloc_large_system_hash("Dentry cache",
3363                                         sizeof(struct hlist_bl_head),
3364                                         dhash_entries,
3365                                         13,
3366                                         HASH_EARLY,
3367                                         &d_hash_shift,
3368                                         &d_hash_mask,
3369                                         0,
3370                                         0);
3371 
3372         for (loop = 0; loop < (1U << d_hash_shift); loop++)
3373                 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3374 }
3375 
3376 static void __init dcache_init(void)
3377 {
3378         unsigned int loop;
3379 
3380         /* 
3381          * A constructor could be added for stable state like the lists,
3382          * but it is probably not worth it because of the cache nature
3383          * of the dcache. 
3384          */
3385         dentry_cache = KMEM_CACHE(dentry,
3386                 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3387 
3388         /* Hash may have been set up in dcache_init_early */
3389         if (!hashdist)
3390                 return;
3391 
3392         dentry_hashtable =
3393                 alloc_large_system_hash("Dentry cache",
3394                                         sizeof(struct hlist_bl_head),
3395                                         dhash_entries,
3396                                         13,
3397                                         0,
3398                                         &d_hash_shift,
3399                                         &d_hash_mask,
3400                                         0,
3401                                         0);
3402 
3403         for (loop = 0; loop < (1U << d_hash_shift); loop++)
3404                 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3405 }
3406 
3407 /* SLAB cache for __getname() consumers */
3408 struct kmem_cache *names_cachep __read_mostly;
3409 EXPORT_SYMBOL(names_cachep);
3410 
3411 EXPORT_SYMBOL(d_genocide);
3412 
3413 void __init vfs_caches_init_early(void)
3414 {
3415         dcache_init_early();
3416         inode_init_early();
3417 }
3418 
3419 void __init vfs_caches_init(unsigned long mempages)
3420 {
3421         unsigned long reserve;
3422 
3423         /* Base hash sizes on available memory, with a reserve equal to
3424            150% of current kernel size */
3425 
3426         reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3427         mempages -= reserve;
3428 
3429         names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3430                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3431 
3432         dcache_init();
3433         inode_init();
3434         files_init(mempages);
3435         mnt_init();
3436         bdev_cache_init();
3437         chrdev_init();
3438 }
3439 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us