Version:  2.0.40 2.2.26 2.4.37 3.3 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

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

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