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

TOMOYO Linux Cross Reference
Linux/fs/dcache.c

Version: ~ [ linux-5.15-rc1 ] ~ [ linux-5.14.5 ] ~ [ linux-5.13.18 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.66 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.147 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.206 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.246 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.282 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.283 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-only
  2 /*
  3  * fs/dcache.c
  4  *
  5  * Complete reimplementation
  6  * (C) 1997 Thomas Schoebel-Theuer,
  7  * with heavy changes by Linus Torvalds
  8  */
  9 
 10 /*
 11  * Notes on the allocation strategy:
 12  *
 13  * The dcache is a master of the icache - whenever a dcache entry
 14  * exists, the inode will always exist. "iput()" is done either when
 15  * the dcache entry is deleted or garbage collected.
 16  */
 17 
 18 #include <linux/ratelimit.h>
 19 #include <linux/string.h>
 20 #include <linux/mm.h>
 21 #include <linux/fs.h>
 22 #include <linux/fscrypt.h>
 23 #include <linux/fsnotify.h>
 24 #include <linux/slab.h>
 25 #include <linux/init.h>
 26 #include <linux/hash.h>
 27 #include <linux/cache.h>
 28 #include <linux/export.h>
 29 #include <linux/security.h>
 30 #include <linux/seqlock.h>
 31 #include <linux/memblock.h>
 32 #include <linux/bit_spinlock.h>
 33 #include <linux/rculist_bl.h>
 34 #include <linux/list_lru.h>
 35 #include "internal.h"
 36 #include "mount.h"
 37 
 38 /*
 39  * Usage:
 40  * dcache->d_inode->i_lock protects:
 41  *   - i_dentry, d_u.d_alias, d_inode of aliases
 42  * dcache_hash_bucket lock protects:
 43  *   - the dcache hash table
 44  * s_roots bl list spinlock protects:
 45  *   - the s_roots list (see __d_drop)
 46  * dentry->d_sb->s_dentry_lru_lock protects:
 47  *   - the dcache lru lists and counters
 48  * d_lock protects:
 49  *   - d_flags
 50  *   - d_name
 51  *   - d_lru
 52  *   - d_count
 53  *   - d_unhashed()
 54  *   - d_parent and d_subdirs
 55  *   - childrens' d_child and d_parent
 56  *   - d_u.d_alias, d_inode
 57  *
 58  * Ordering:
 59  * dentry->d_inode->i_lock
 60  *   dentry->d_lock
 61  *     dentry->d_sb->s_dentry_lru_lock
 62  *     dcache_hash_bucket lock
 63  *     s_roots lock
 64  *
 65  * If there is an ancestor relationship:
 66  * dentry->d_parent->...->d_parent->d_lock
 67  *   ...
 68  *     dentry->d_parent->d_lock
 69  *       dentry->d_lock
 70  *
 71  * If no ancestor relationship:
 72  * arbitrary, since it's serialized on rename_lock
 73  */
 74 int sysctl_vfs_cache_pressure __read_mostly = 100;
 75 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
 76 
 77 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
 78 
 79 EXPORT_SYMBOL(rename_lock);
 80 
 81 static struct kmem_cache *dentry_cache __read_mostly;
 82 
 83 const struct qstr empty_name = QSTR_INIT("", 0);
 84 EXPORT_SYMBOL(empty_name);
 85 const struct qstr slash_name = QSTR_INIT("/", 1);
 86 EXPORT_SYMBOL(slash_name);
 87 
 88 /*
 89  * This is the single most critical data structure when it comes
 90  * to the dcache: the hashtable for lookups. Somebody should try
 91  * to make this good - I've just made it work.
 92  *
 93  * This hash-function tries to avoid losing too many bits of hash
 94  * information, yet avoid using a prime hash-size or similar.
 95  */
 96 
 97 static unsigned int d_hash_shift __read_mostly;
 98 
 99 static struct hlist_bl_head *dentry_hashtable __read_mostly;
100 
101 static inline struct hlist_bl_head *d_hash(unsigned int hash)
102 {
103         return dentry_hashtable + (hash >> d_hash_shift);
104 }
105 
106 #define IN_LOOKUP_SHIFT 10
107 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
108 
109 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
110                                         unsigned int hash)
111 {
112         hash += (unsigned long) parent / L1_CACHE_BYTES;
113         return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
114 }
115 
116 
117 /* Statistics gathering. */
118 struct dentry_stat_t dentry_stat = {
119         .age_limit = 45,
120 };
121 
122 static DEFINE_PER_CPU(long, nr_dentry);
123 static DEFINE_PER_CPU(long, nr_dentry_unused);
124 static DEFINE_PER_CPU(long, nr_dentry_negative);
125 
126 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
127 
128 /*
129  * Here we resort to our own counters instead of using generic per-cpu counters
130  * for consistency with what the vfs inode code does. We are expected to harvest
131  * better code and performance by having our own specialized counters.
132  *
133  * Please note that the loop is done over all possible CPUs, not over all online
134  * CPUs. The reason for this is that we don't want to play games with CPUs going
135  * on and off. If one of them goes off, we will just keep their counters.
136  *
137  * glommer: See cffbc8a for details, and if you ever intend to change this,
138  * please update all vfs counters to match.
139  */
140 static long get_nr_dentry(void)
141 {
142         int i;
143         long sum = 0;
144         for_each_possible_cpu(i)
145                 sum += per_cpu(nr_dentry, i);
146         return sum < 0 ? 0 : sum;
147 }
148 
149 static long get_nr_dentry_unused(void)
150 {
151         int i;
152         long sum = 0;
153         for_each_possible_cpu(i)
154                 sum += per_cpu(nr_dentry_unused, i);
155         return sum < 0 ? 0 : sum;
156 }
157 
158 static long get_nr_dentry_negative(void)
159 {
160         int i;
161         long sum = 0;
162 
163         for_each_possible_cpu(i)
164                 sum += per_cpu(nr_dentry_negative, i);
165         return sum < 0 ? 0 : sum;
166 }
167 
168 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
169                    size_t *lenp, loff_t *ppos)
170 {
171         dentry_stat.nr_dentry = get_nr_dentry();
172         dentry_stat.nr_unused = get_nr_dentry_unused();
173         dentry_stat.nr_negative = get_nr_dentry_negative();
174         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
175 }
176 #endif
177 
178 /*
179  * Compare 2 name strings, return 0 if they match, otherwise non-zero.
180  * The strings are both count bytes long, and count is non-zero.
181  */
182 #ifdef CONFIG_DCACHE_WORD_ACCESS
183 
184 #include <asm/word-at-a-time.h>
185 /*
186  * NOTE! 'cs' and 'scount' come from a dentry, so it has a
187  * aligned allocation for this particular component. We don't
188  * strictly need the load_unaligned_zeropad() safety, but it
189  * doesn't hurt either.
190  *
191  * In contrast, 'ct' and 'tcount' can be from a pathname, and do
192  * need the careful unaligned handling.
193  */
194 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
195 {
196         unsigned long a,b,mask;
197 
198         for (;;) {
199                 a = read_word_at_a_time(cs);
200                 b = load_unaligned_zeropad(ct);
201                 if (tcount < sizeof(unsigned long))
202                         break;
203                 if (unlikely(a != b))
204                         return 1;
205                 cs += sizeof(unsigned long);
206                 ct += sizeof(unsigned long);
207                 tcount -= sizeof(unsigned long);
208                 if (!tcount)
209                         return 0;
210         }
211         mask = bytemask_from_count(tcount);
212         return unlikely(!!((a ^ b) & mask));
213 }
214 
215 #else
216 
217 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
218 {
219         do {
220                 if (*cs != *ct)
221                         return 1;
222                 cs++;
223                 ct++;
224                 tcount--;
225         } while (tcount);
226         return 0;
227 }
228 
229 #endif
230 
231 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
232 {
233         /*
234          * Be careful about RCU walk racing with rename:
235          * use 'READ_ONCE' to fetch the name pointer.
236          *
237          * NOTE! Even if a rename will mean that the length
238          * was not loaded atomically, we don't care. The
239          * RCU walk will check the sequence count eventually,
240          * and catch it. And we won't overrun the buffer,
241          * because we're reading the name pointer atomically,
242          * and a dentry name is guaranteed to be properly
243          * terminated with a NUL byte.
244          *
245          * End result: even if 'len' is wrong, we'll exit
246          * early because the data cannot match (there can
247          * be no NUL in the ct/tcount data)
248          */
249         const unsigned char *cs = READ_ONCE(dentry->d_name.name);
250 
251         return dentry_string_cmp(cs, ct, tcount);
252 }
253 
254 struct external_name {
255         union {
256                 atomic_t count;
257                 struct rcu_head head;
258         } u;
259         unsigned char name[];
260 };
261 
262 static inline struct external_name *external_name(struct dentry *dentry)
263 {
264         return container_of(dentry->d_name.name, struct external_name, name[0]);
265 }
266 
267 static void __d_free(struct rcu_head *head)
268 {
269         struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
270 
271         kmem_cache_free(dentry_cache, dentry); 
272 }
273 
274 static void __d_free_external(struct rcu_head *head)
275 {
276         struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
277         kfree(external_name(dentry));
278         kmem_cache_free(dentry_cache, dentry);
279 }
280 
281 static inline int dname_external(const struct dentry *dentry)
282 {
283         return dentry->d_name.name != dentry->d_iname;
284 }
285 
286 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
287 {
288         spin_lock(&dentry->d_lock);
289         name->name = dentry->d_name;
290         if (unlikely(dname_external(dentry))) {
291                 atomic_inc(&external_name(dentry)->u.count);
292         } else {
293                 memcpy(name->inline_name, dentry->d_iname,
294                        dentry->d_name.len + 1);
295                 name->name.name = name->inline_name;
296         }
297         spin_unlock(&dentry->d_lock);
298 }
299 EXPORT_SYMBOL(take_dentry_name_snapshot);
300 
301 void release_dentry_name_snapshot(struct name_snapshot *name)
302 {
303         if (unlikely(name->name.name != name->inline_name)) {
304                 struct external_name *p;
305                 p = container_of(name->name.name, struct external_name, name[0]);
306                 if (unlikely(atomic_dec_and_test(&p->u.count)))
307                         kfree_rcu(p, u.head);
308         }
309 }
310 EXPORT_SYMBOL(release_dentry_name_snapshot);
311 
312 static inline void __d_set_inode_and_type(struct dentry *dentry,
313                                           struct inode *inode,
314                                           unsigned type_flags)
315 {
316         unsigned flags;
317 
318         dentry->d_inode = inode;
319         flags = READ_ONCE(dentry->d_flags);
320         flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
321         flags |= type_flags;
322         WRITE_ONCE(dentry->d_flags, flags);
323 }
324 
325 static inline void __d_clear_type_and_inode(struct dentry *dentry)
326 {
327         unsigned flags = READ_ONCE(dentry->d_flags);
328 
329         flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
330         WRITE_ONCE(dentry->d_flags, flags);
331         dentry->d_inode = NULL;
332         if (dentry->d_flags & DCACHE_LRU_LIST)
333                 this_cpu_inc(nr_dentry_negative);
334 }
335 
336 static void dentry_free(struct dentry *dentry)
337 {
338         WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
339         if (unlikely(dname_external(dentry))) {
340                 struct external_name *p = external_name(dentry);
341                 if (likely(atomic_dec_and_test(&p->u.count))) {
342                         call_rcu(&dentry->d_u.d_rcu, __d_free_external);
343                         return;
344                 }
345         }
346         /* if dentry was never visible to RCU, immediate free is OK */
347         if (dentry->d_flags & DCACHE_NORCU)
348                 __d_free(&dentry->d_u.d_rcu);
349         else
350                 call_rcu(&dentry->d_u.d_rcu, __d_free);
351 }
352 
353 /*
354  * Release the dentry's inode, using the filesystem
355  * d_iput() operation if defined.
356  */
357 static void dentry_unlink_inode(struct dentry * dentry)
358         __releases(dentry->d_lock)
359         __releases(dentry->d_inode->i_lock)
360 {
361         struct inode *inode = dentry->d_inode;
362 
363         raw_write_seqcount_begin(&dentry->d_seq);
364         __d_clear_type_and_inode(dentry);
365         hlist_del_init(&dentry->d_u.d_alias);
366         raw_write_seqcount_end(&dentry->d_seq);
367         spin_unlock(&dentry->d_lock);
368         spin_unlock(&inode->i_lock);
369         if (!inode->i_nlink)
370                 fsnotify_inoderemove(inode);
371         if (dentry->d_op && dentry->d_op->d_iput)
372                 dentry->d_op->d_iput(dentry, inode);
373         else
374                 iput(inode);
375 }
376 
377 /*
378  * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
379  * is in use - which includes both the "real" per-superblock
380  * LRU list _and_ the DCACHE_SHRINK_LIST use.
381  *
382  * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
383  * on the shrink list (ie not on the superblock LRU list).
384  *
385  * The per-cpu "nr_dentry_unused" counters are updated with
386  * the DCACHE_LRU_LIST bit.
387  *
388  * The per-cpu "nr_dentry_negative" counters are only updated
389  * when deleted from or added to the per-superblock LRU list, not
390  * from/to the shrink list. That is to avoid an unneeded dec/inc
391  * pair when moving from LRU to shrink list in select_collect().
392  *
393  * These helper functions make sure we always follow the
394  * rules. d_lock must be held by the caller.
395  */
396 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
397 static void d_lru_add(struct dentry *dentry)
398 {
399         D_FLAG_VERIFY(dentry, 0);
400         dentry->d_flags |= DCACHE_LRU_LIST;
401         this_cpu_inc(nr_dentry_unused);
402         if (d_is_negative(dentry))
403                 this_cpu_inc(nr_dentry_negative);
404         WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
405 }
406 
407 static void d_lru_del(struct dentry *dentry)
408 {
409         D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
410         dentry->d_flags &= ~DCACHE_LRU_LIST;
411         this_cpu_dec(nr_dentry_unused);
412         if (d_is_negative(dentry))
413                 this_cpu_dec(nr_dentry_negative);
414         WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
415 }
416 
417 static void d_shrink_del(struct dentry *dentry)
418 {
419         D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
420         list_del_init(&dentry->d_lru);
421         dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
422         this_cpu_dec(nr_dentry_unused);
423 }
424 
425 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
426 {
427         D_FLAG_VERIFY(dentry, 0);
428         list_add(&dentry->d_lru, list);
429         dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
430         this_cpu_inc(nr_dentry_unused);
431 }
432 
433 /*
434  * These can only be called under the global LRU lock, ie during the
435  * callback for freeing the LRU list. "isolate" removes it from the
436  * LRU lists entirely, while shrink_move moves it to the indicated
437  * private list.
438  */
439 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
440 {
441         D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
442         dentry->d_flags &= ~DCACHE_LRU_LIST;
443         this_cpu_dec(nr_dentry_unused);
444         if (d_is_negative(dentry))
445                 this_cpu_dec(nr_dentry_negative);
446         list_lru_isolate(lru, &dentry->d_lru);
447 }
448 
449 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
450                               struct list_head *list)
451 {
452         D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
453         dentry->d_flags |= DCACHE_SHRINK_LIST;
454         if (d_is_negative(dentry))
455                 this_cpu_dec(nr_dentry_negative);
456         list_lru_isolate_move(lru, &dentry->d_lru, list);
457 }
458 
459 /**
460  * d_drop - drop a dentry
461  * @dentry: dentry to drop
462  *
463  * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
464  * be found through a VFS lookup any more. Note that this is different from
465  * deleting the dentry - d_delete will try to mark the dentry negative if
466  * possible, giving a successful _negative_ lookup, while d_drop will
467  * just make the cache lookup fail.
468  *
469  * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
470  * reason (NFS timeouts or autofs deletes).
471  *
472  * __d_drop requires dentry->d_lock
473  * ___d_drop doesn't mark dentry as "unhashed"
474  *   (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
475  */
476 static void ___d_drop(struct dentry *dentry)
477 {
478         struct hlist_bl_head *b;
479         /*
480          * Hashed dentries are normally on the dentry hashtable,
481          * with the exception of those newly allocated by
482          * d_obtain_root, which are always IS_ROOT:
483          */
484         if (unlikely(IS_ROOT(dentry)))
485                 b = &dentry->d_sb->s_roots;
486         else
487                 b = d_hash(dentry->d_name.hash);
488 
489         hlist_bl_lock(b);
490         __hlist_bl_del(&dentry->d_hash);
491         hlist_bl_unlock(b);
492 }
493 
494 void __d_drop(struct dentry *dentry)
495 {
496         if (!d_unhashed(dentry)) {
497                 ___d_drop(dentry);
498                 dentry->d_hash.pprev = NULL;
499                 write_seqcount_invalidate(&dentry->d_seq);
500         }
501 }
502 EXPORT_SYMBOL(__d_drop);
503 
504 void d_drop(struct dentry *dentry)
505 {
506         spin_lock(&dentry->d_lock);
507         __d_drop(dentry);
508         spin_unlock(&dentry->d_lock);
509 }
510 EXPORT_SYMBOL(d_drop);
511 
512 static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
513 {
514         struct dentry *next;
515         /*
516          * Inform d_walk() and shrink_dentry_list() that we are no longer
517          * attached to the dentry tree
518          */
519         dentry->d_flags |= DCACHE_DENTRY_KILLED;
520         if (unlikely(list_empty(&dentry->d_child)))
521                 return;
522         __list_del_entry(&dentry->d_child);
523         /*
524          * Cursors can move around the list of children.  While we'd been
525          * a normal list member, it didn't matter - ->d_child.next would've
526          * been updated.  However, from now on it won't be and for the
527          * things like d_walk() it might end up with a nasty surprise.
528          * Normally d_walk() doesn't care about cursors moving around -
529          * ->d_lock on parent prevents that and since a cursor has no children
530          * of its own, we get through it without ever unlocking the parent.
531          * There is one exception, though - if we ascend from a child that
532          * gets killed as soon as we unlock it, the next sibling is found
533          * using the value left in its ->d_child.next.  And if _that_
534          * pointed to a cursor, and cursor got moved (e.g. by lseek())
535          * before d_walk() regains parent->d_lock, we'll end up skipping
536          * everything the cursor had been moved past.
537          *
538          * Solution: make sure that the pointer left behind in ->d_child.next
539          * points to something that won't be moving around.  I.e. skip the
540          * cursors.
541          */
542         while (dentry->d_child.next != &parent->d_subdirs) {
543                 next = list_entry(dentry->d_child.next, struct dentry, d_child);
544                 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
545                         break;
546                 dentry->d_child.next = next->d_child.next;
547         }
548 }
549 
550 static void __dentry_kill(struct dentry *dentry)
551 {
552         struct dentry *parent = NULL;
553         bool can_free = true;
554         if (!IS_ROOT(dentry))
555                 parent = dentry->d_parent;
556 
557         /*
558          * The dentry is now unrecoverably dead to the world.
559          */
560         lockref_mark_dead(&dentry->d_lockref);
561 
562         /*
563          * inform the fs via d_prune that this dentry is about to be
564          * unhashed and destroyed.
565          */
566         if (dentry->d_flags & DCACHE_OP_PRUNE)
567                 dentry->d_op->d_prune(dentry);
568 
569         if (dentry->d_flags & DCACHE_LRU_LIST) {
570                 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
571                         d_lru_del(dentry);
572         }
573         /* if it was on the hash then remove it */
574         __d_drop(dentry);
575         dentry_unlist(dentry, parent);
576         if (parent)
577                 spin_unlock(&parent->d_lock);
578         if (dentry->d_inode)
579                 dentry_unlink_inode(dentry);
580         else
581                 spin_unlock(&dentry->d_lock);
582         this_cpu_dec(nr_dentry);
583         if (dentry->d_op && dentry->d_op->d_release)
584                 dentry->d_op->d_release(dentry);
585 
586         spin_lock(&dentry->d_lock);
587         if (dentry->d_flags & DCACHE_SHRINK_LIST) {
588                 dentry->d_flags |= DCACHE_MAY_FREE;
589                 can_free = false;
590         }
591         spin_unlock(&dentry->d_lock);
592         if (likely(can_free))
593                 dentry_free(dentry);
594         cond_resched();
595 }
596 
597 static struct dentry *__lock_parent(struct dentry *dentry)
598 {
599         struct dentry *parent;
600         rcu_read_lock();
601         spin_unlock(&dentry->d_lock);
602 again:
603         parent = READ_ONCE(dentry->d_parent);
604         spin_lock(&parent->d_lock);
605         /*
606          * We can't blindly lock dentry until we are sure
607          * that we won't violate the locking order.
608          * Any changes of dentry->d_parent must have
609          * been done with parent->d_lock held, so
610          * spin_lock() above is enough of a barrier
611          * for checking if it's still our child.
612          */
613         if (unlikely(parent != dentry->d_parent)) {
614                 spin_unlock(&parent->d_lock);
615                 goto again;
616         }
617         rcu_read_unlock();
618         if (parent != dentry)
619                 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
620         else
621                 parent = NULL;
622         return parent;
623 }
624 
625 static inline struct dentry *lock_parent(struct dentry *dentry)
626 {
627         struct dentry *parent = dentry->d_parent;
628         if (IS_ROOT(dentry))
629                 return NULL;
630         if (likely(spin_trylock(&parent->d_lock)))
631                 return parent;
632         return __lock_parent(dentry);
633 }
634 
635 static inline bool retain_dentry(struct dentry *dentry)
636 {
637         WARN_ON(d_in_lookup(dentry));
638 
639         /* Unreachable? Get rid of it */
640         if (unlikely(d_unhashed(dentry)))
641                 return false;
642 
643         if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
644                 return false;
645 
646         if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
647                 if (dentry->d_op->d_delete(dentry))
648                         return false;
649         }
650         /* retain; LRU fodder */
651         dentry->d_lockref.count--;
652         if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
653                 d_lru_add(dentry);
654         else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
655                 dentry->d_flags |= DCACHE_REFERENCED;
656         return true;
657 }
658 
659 /*
660  * Finish off a dentry we've decided to kill.
661  * dentry->d_lock must be held, returns with it unlocked.
662  * Returns dentry requiring refcount drop, or NULL if we're done.
663  */
664 static struct dentry *dentry_kill(struct dentry *dentry)
665         __releases(dentry->d_lock)
666 {
667         struct inode *inode = dentry->d_inode;
668         struct dentry *parent = NULL;
669 
670         if (inode && unlikely(!spin_trylock(&inode->i_lock)))
671                 goto slow_positive;
672 
673         if (!IS_ROOT(dentry)) {
674                 parent = dentry->d_parent;
675                 if (unlikely(!spin_trylock(&parent->d_lock))) {
676                         parent = __lock_parent(dentry);
677                         if (likely(inode || !dentry->d_inode))
678                                 goto got_locks;
679                         /* negative that became positive */
680                         if (parent)
681                                 spin_unlock(&parent->d_lock);
682                         inode = dentry->d_inode;
683                         goto slow_positive;
684                 }
685         }
686         __dentry_kill(dentry);
687         return parent;
688 
689 slow_positive:
690         spin_unlock(&dentry->d_lock);
691         spin_lock(&inode->i_lock);
692         spin_lock(&dentry->d_lock);
693         parent = lock_parent(dentry);
694 got_locks:
695         if (unlikely(dentry->d_lockref.count != 1)) {
696                 dentry->d_lockref.count--;
697         } else if (likely(!retain_dentry(dentry))) {
698                 __dentry_kill(dentry);
699                 return parent;
700         }
701         /* we are keeping it, after all */
702         if (inode)
703                 spin_unlock(&inode->i_lock);
704         if (parent)
705                 spin_unlock(&parent->d_lock);
706         spin_unlock(&dentry->d_lock);
707         return NULL;
708 }
709 
710 /*
711  * Try to do a lockless dput(), and return whether that was successful.
712  *
713  * If unsuccessful, we return false, having already taken the dentry lock.
714  *
715  * The caller needs to hold the RCU read lock, so that the dentry is
716  * guaranteed to stay around even if the refcount goes down to zero!
717  */
718 static inline bool fast_dput(struct dentry *dentry)
719 {
720         int ret;
721         unsigned int d_flags;
722 
723         /*
724          * If we have a d_op->d_delete() operation, we sould not
725          * let the dentry count go to zero, so use "put_or_lock".
726          */
727         if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
728                 return lockref_put_or_lock(&dentry->d_lockref);
729 
730         /*
731          * .. otherwise, we can try to just decrement the
732          * lockref optimistically.
733          */
734         ret = lockref_put_return(&dentry->d_lockref);
735 
736         /*
737          * If the lockref_put_return() failed due to the lock being held
738          * by somebody else, the fast path has failed. We will need to
739          * get the lock, and then check the count again.
740          */
741         if (unlikely(ret < 0)) {
742                 spin_lock(&dentry->d_lock);
743                 if (dentry->d_lockref.count > 1) {
744                         dentry->d_lockref.count--;
745                         spin_unlock(&dentry->d_lock);
746                         return true;
747                 }
748                 return false;
749         }
750 
751         /*
752          * If we weren't the last ref, we're done.
753          */
754         if (ret)
755                 return true;
756 
757         /*
758          * Careful, careful. The reference count went down
759          * to zero, but we don't hold the dentry lock, so
760          * somebody else could get it again, and do another
761          * dput(), and we need to not race with that.
762          *
763          * However, there is a very special and common case
764          * where we don't care, because there is nothing to
765          * do: the dentry is still hashed, it does not have
766          * a 'delete' op, and it's referenced and already on
767          * the LRU list.
768          *
769          * NOTE! Since we aren't locked, these values are
770          * not "stable". However, it is sufficient that at
771          * some point after we dropped the reference the
772          * dentry was hashed and the flags had the proper
773          * value. Other dentry users may have re-gotten
774          * a reference to the dentry and change that, but
775          * our work is done - we can leave the dentry
776          * around with a zero refcount.
777          */
778         smp_rmb();
779         d_flags = READ_ONCE(dentry->d_flags);
780         d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
781 
782         /* Nothing to do? Dropping the reference was all we needed? */
783         if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
784                 return true;
785 
786         /*
787          * Not the fast normal case? Get the lock. We've already decremented
788          * the refcount, but we'll need to re-check the situation after
789          * getting the lock.
790          */
791         spin_lock(&dentry->d_lock);
792 
793         /*
794          * Did somebody else grab a reference to it in the meantime, and
795          * we're no longer the last user after all? Alternatively, somebody
796          * else could have killed it and marked it dead. Either way, we
797          * don't need to do anything else.
798          */
799         if (dentry->d_lockref.count) {
800                 spin_unlock(&dentry->d_lock);
801                 return true;
802         }
803 
804         /*
805          * Re-get the reference we optimistically dropped. We hold the
806          * lock, and we just tested that it was zero, so we can just
807          * set it to 1.
808          */
809         dentry->d_lockref.count = 1;
810         return false;
811 }
812 
813 
814 /* 
815  * This is dput
816  *
817  * This is complicated by the fact that we do not want to put
818  * dentries that are no longer on any hash chain on the unused
819  * list: we'd much rather just get rid of them immediately.
820  *
821  * However, that implies that we have to traverse the dentry
822  * tree upwards to the parents which might _also_ now be
823  * scheduled for deletion (it may have been only waiting for
824  * its last child to go away).
825  *
826  * This tail recursion is done by hand as we don't want to depend
827  * on the compiler to always get this right (gcc generally doesn't).
828  * Real recursion would eat up our stack space.
829  */
830 
831 /*
832  * dput - release a dentry
833  * @dentry: dentry to release 
834  *
835  * Release a dentry. This will drop the usage count and if appropriate
836  * call the dentry unlink method as well as removing it from the queues and
837  * releasing its resources. If the parent dentries were scheduled for release
838  * they too may now get deleted.
839  */
840 void dput(struct dentry *dentry)
841 {
842         while (dentry) {
843                 might_sleep();
844 
845                 rcu_read_lock();
846                 if (likely(fast_dput(dentry))) {
847                         rcu_read_unlock();
848                         return;
849                 }
850 
851                 /* Slow case: now with the dentry lock held */
852                 rcu_read_unlock();
853 
854                 if (likely(retain_dentry(dentry))) {
855                         spin_unlock(&dentry->d_lock);
856                         return;
857                 }
858 
859                 dentry = dentry_kill(dentry);
860         }
861 }
862 EXPORT_SYMBOL(dput);
863 
864 static void __dput_to_list(struct dentry *dentry, struct list_head *list)
865 __must_hold(&dentry->d_lock)
866 {
867         if (dentry->d_flags & DCACHE_SHRINK_LIST) {
868                 /* let the owner of the list it's on deal with it */
869                 --dentry->d_lockref.count;
870         } else {
871                 if (dentry->d_flags & DCACHE_LRU_LIST)
872                         d_lru_del(dentry);
873                 if (!--dentry->d_lockref.count)
874                         d_shrink_add(dentry, list);
875         }
876 }
877 
878 void dput_to_list(struct dentry *dentry, struct list_head *list)
879 {
880         rcu_read_lock();
881         if (likely(fast_dput(dentry))) {
882                 rcu_read_unlock();
883                 return;
884         }
885         rcu_read_unlock();
886         if (!retain_dentry(dentry))
887                 __dput_to_list(dentry, list);
888         spin_unlock(&dentry->d_lock);
889 }
890 
891 /* This must be called with d_lock held */
892 static inline void __dget_dlock(struct dentry *dentry)
893 {
894         dentry->d_lockref.count++;
895 }
896 
897 static inline void __dget(struct dentry *dentry)
898 {
899         lockref_get(&dentry->d_lockref);
900 }
901 
902 struct dentry *dget_parent(struct dentry *dentry)
903 {
904         int gotref;
905         struct dentry *ret;
906 
907         /*
908          * Do optimistic parent lookup without any
909          * locking.
910          */
911         rcu_read_lock();
912         ret = READ_ONCE(dentry->d_parent);
913         gotref = lockref_get_not_zero(&ret->d_lockref);
914         rcu_read_unlock();
915         if (likely(gotref)) {
916                 if (likely(ret == READ_ONCE(dentry->d_parent)))
917                         return ret;
918                 dput(ret);
919         }
920 
921 repeat:
922         /*
923          * Don't need rcu_dereference because we re-check it was correct under
924          * the lock.
925          */
926         rcu_read_lock();
927         ret = dentry->d_parent;
928         spin_lock(&ret->d_lock);
929         if (unlikely(ret != dentry->d_parent)) {
930                 spin_unlock(&ret->d_lock);
931                 rcu_read_unlock();
932                 goto repeat;
933         }
934         rcu_read_unlock();
935         BUG_ON(!ret->d_lockref.count);
936         ret->d_lockref.count++;
937         spin_unlock(&ret->d_lock);
938         return ret;
939 }
940 EXPORT_SYMBOL(dget_parent);
941 
942 static struct dentry * __d_find_any_alias(struct inode *inode)
943 {
944         struct dentry *alias;
945 
946         if (hlist_empty(&inode->i_dentry))
947                 return NULL;
948         alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
949         __dget(alias);
950         return alias;
951 }
952 
953 /**
954  * d_find_any_alias - find any alias for a given inode
955  * @inode: inode to find an alias for
956  *
957  * If any aliases exist for the given inode, take and return a
958  * reference for one of them.  If no aliases exist, return %NULL.
959  */
960 struct dentry *d_find_any_alias(struct inode *inode)
961 {
962         struct dentry *de;
963 
964         spin_lock(&inode->i_lock);
965         de = __d_find_any_alias(inode);
966         spin_unlock(&inode->i_lock);
967         return de;
968 }
969 EXPORT_SYMBOL(d_find_any_alias);
970 
971 /**
972  * d_find_alias - grab a hashed alias of inode
973  * @inode: inode in question
974  *
975  * If inode has a hashed alias, or is a directory and has any alias,
976  * acquire the reference to alias and return it. Otherwise return NULL.
977  * Notice that if inode is a directory there can be only one alias and
978  * it can be unhashed only if it has no children, or if it is the root
979  * of a filesystem, or if the directory was renamed and d_revalidate
980  * was the first vfs operation to notice.
981  *
982  * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
983  * any other hashed alias over that one.
984  */
985 static struct dentry *__d_find_alias(struct inode *inode)
986 {
987         struct dentry *alias;
988 
989         if (S_ISDIR(inode->i_mode))
990                 return __d_find_any_alias(inode);
991 
992         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
993                 spin_lock(&alias->d_lock);
994                 if (!d_unhashed(alias)) {
995                         __dget_dlock(alias);
996                         spin_unlock(&alias->d_lock);
997                         return alias;
998                 }
999                 spin_unlock(&alias->d_lock);
1000         }
1001         return NULL;
1002 }
1003 
1004 struct dentry *d_find_alias(struct inode *inode)
1005 {
1006         struct dentry *de = NULL;
1007 
1008         if (!hlist_empty(&inode->i_dentry)) {
1009                 spin_lock(&inode->i_lock);
1010                 de = __d_find_alias(inode);
1011                 spin_unlock(&inode->i_lock);
1012         }
1013         return de;
1014 }
1015 EXPORT_SYMBOL(d_find_alias);
1016 
1017 /*
1018  *      Try to kill dentries associated with this inode.
1019  * WARNING: you must own a reference to inode.
1020  */
1021 void d_prune_aliases(struct inode *inode)
1022 {
1023         struct dentry *dentry;
1024 restart:
1025         spin_lock(&inode->i_lock);
1026         hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
1027                 spin_lock(&dentry->d_lock);
1028                 if (!dentry->d_lockref.count) {
1029                         struct dentry *parent = lock_parent(dentry);
1030                         if (likely(!dentry->d_lockref.count)) {
1031                                 __dentry_kill(dentry);
1032                                 dput(parent);
1033                                 goto restart;
1034                         }
1035                         if (parent)
1036                                 spin_unlock(&parent->d_lock);
1037                 }
1038                 spin_unlock(&dentry->d_lock);
1039         }
1040         spin_unlock(&inode->i_lock);
1041 }
1042 EXPORT_SYMBOL(d_prune_aliases);
1043 
1044 /*
1045  * Lock a dentry from shrink list.
1046  * Called under rcu_read_lock() and dentry->d_lock; the former
1047  * guarantees that nothing we access will be freed under us.
1048  * Note that dentry is *not* protected from concurrent dentry_kill(),
1049  * d_delete(), etc.
1050  *
1051  * Return false if dentry has been disrupted or grabbed, leaving
1052  * the caller to kick it off-list.  Otherwise, return true and have
1053  * that dentry's inode and parent both locked.
1054  */
1055 static bool shrink_lock_dentry(struct dentry *dentry)
1056 {
1057         struct inode *inode;
1058         struct dentry *parent;
1059 
1060         if (dentry->d_lockref.count)
1061                 return false;
1062 
1063         inode = dentry->d_inode;
1064         if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1065                 spin_unlock(&dentry->d_lock);
1066                 spin_lock(&inode->i_lock);
1067                 spin_lock(&dentry->d_lock);
1068                 if (unlikely(dentry->d_lockref.count))
1069                         goto out;
1070                 /* changed inode means that somebody had grabbed it */
1071                 if (unlikely(inode != dentry->d_inode))
1072                         goto out;
1073         }
1074 
1075         parent = dentry->d_parent;
1076         if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock)))
1077                 return true;
1078 
1079         spin_unlock(&dentry->d_lock);
1080         spin_lock(&parent->d_lock);
1081         if (unlikely(parent != dentry->d_parent)) {
1082                 spin_unlock(&parent->d_lock);
1083                 spin_lock(&dentry->d_lock);
1084                 goto out;
1085         }
1086         spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1087         if (likely(!dentry->d_lockref.count))
1088                 return true;
1089         spin_unlock(&parent->d_lock);
1090 out:
1091         if (inode)
1092                 spin_unlock(&inode->i_lock);
1093         return false;
1094 }
1095 
1096 void shrink_dentry_list(struct list_head *list)
1097 {
1098         while (!list_empty(list)) {
1099                 struct dentry *dentry, *parent;
1100 
1101                 dentry = list_entry(list->prev, struct dentry, d_lru);
1102                 spin_lock(&dentry->d_lock);
1103                 rcu_read_lock();
1104                 if (!shrink_lock_dentry(dentry)) {
1105                         bool can_free = false;
1106                         rcu_read_unlock();
1107                         d_shrink_del(dentry);
1108                         if (dentry->d_lockref.count < 0)
1109                                 can_free = dentry->d_flags & DCACHE_MAY_FREE;
1110                         spin_unlock(&dentry->d_lock);
1111                         if (can_free)
1112                                 dentry_free(dentry);
1113                         continue;
1114                 }
1115                 rcu_read_unlock();
1116                 d_shrink_del(dentry);
1117                 parent = dentry->d_parent;
1118                 if (parent != dentry)
1119                         __dput_to_list(parent, list);
1120                 __dentry_kill(dentry);
1121         }
1122 }
1123 
1124 static enum lru_status dentry_lru_isolate(struct list_head *item,
1125                 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1126 {
1127         struct list_head *freeable = arg;
1128         struct dentry   *dentry = container_of(item, struct dentry, d_lru);
1129 
1130 
1131         /*
1132          * we are inverting the lru lock/dentry->d_lock here,
1133          * so use a trylock. If we fail to get the lock, just skip
1134          * it
1135          */
1136         if (!spin_trylock(&dentry->d_lock))
1137                 return LRU_SKIP;
1138 
1139         /*
1140          * Referenced dentries are still in use. If they have active
1141          * counts, just remove them from the LRU. Otherwise give them
1142          * another pass through the LRU.
1143          */
1144         if (dentry->d_lockref.count) {
1145                 d_lru_isolate(lru, dentry);
1146                 spin_unlock(&dentry->d_lock);
1147                 return LRU_REMOVED;
1148         }
1149 
1150         if (dentry->d_flags & DCACHE_REFERENCED) {
1151                 dentry->d_flags &= ~DCACHE_REFERENCED;
1152                 spin_unlock(&dentry->d_lock);
1153 
1154                 /*
1155                  * The list move itself will be made by the common LRU code. At
1156                  * this point, we've dropped the dentry->d_lock but keep the
1157                  * lru lock. This is safe to do, since every list movement is
1158                  * protected by the lru lock even if both locks are held.
1159                  *
1160                  * This is guaranteed by the fact that all LRU management
1161                  * functions are intermediated by the LRU API calls like
1162                  * list_lru_add and list_lru_del. List movement in this file
1163                  * only ever occur through this functions or through callbacks
1164                  * like this one, that are called from the LRU API.
1165                  *
1166                  * The only exceptions to this are functions like
1167                  * shrink_dentry_list, and code that first checks for the
1168                  * DCACHE_SHRINK_LIST flag.  Those are guaranteed to be
1169                  * operating only with stack provided lists after they are
1170                  * properly isolated from the main list.  It is thus, always a
1171                  * local access.
1172                  */
1173                 return LRU_ROTATE;
1174         }
1175 
1176         d_lru_shrink_move(lru, dentry, freeable);
1177         spin_unlock(&dentry->d_lock);
1178 
1179         return LRU_REMOVED;
1180 }
1181 
1182 /**
1183  * prune_dcache_sb - shrink the dcache
1184  * @sb: superblock
1185  * @sc: shrink control, passed to list_lru_shrink_walk()
1186  *
1187  * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1188  * is done when we need more memory and called from the superblock shrinker
1189  * function.
1190  *
1191  * This function may fail to free any resources if all the dentries are in
1192  * use.
1193  */
1194 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1195 {
1196         LIST_HEAD(dispose);
1197         long freed;
1198 
1199         freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1200                                      dentry_lru_isolate, &dispose);
1201         shrink_dentry_list(&dispose);
1202         return freed;
1203 }
1204 
1205 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1206                 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1207 {
1208         struct list_head *freeable = arg;
1209         struct dentry   *dentry = container_of(item, struct dentry, d_lru);
1210 
1211         /*
1212          * we are inverting the lru lock/dentry->d_lock here,
1213          * so use a trylock. If we fail to get the lock, just skip
1214          * it
1215          */
1216         if (!spin_trylock(&dentry->d_lock))
1217                 return LRU_SKIP;
1218 
1219         d_lru_shrink_move(lru, dentry, freeable);
1220         spin_unlock(&dentry->d_lock);
1221 
1222         return LRU_REMOVED;
1223 }
1224 
1225 
1226 /**
1227  * shrink_dcache_sb - shrink dcache for a superblock
1228  * @sb: superblock
1229  *
1230  * Shrink the dcache for the specified super block. This is used to free
1231  * the dcache before unmounting a file system.
1232  */
1233 void shrink_dcache_sb(struct super_block *sb)
1234 {
1235         do {
1236                 LIST_HEAD(dispose);
1237 
1238                 list_lru_walk(&sb->s_dentry_lru,
1239                         dentry_lru_isolate_shrink, &dispose, 1024);
1240                 shrink_dentry_list(&dispose);
1241         } while (list_lru_count(&sb->s_dentry_lru) > 0);
1242 }
1243 EXPORT_SYMBOL(shrink_dcache_sb);
1244 
1245 /**
1246  * enum d_walk_ret - action to talke during tree walk
1247  * @D_WALK_CONTINUE:    contrinue walk
1248  * @D_WALK_QUIT:        quit walk
1249  * @D_WALK_NORETRY:     quit when retry is needed
1250  * @D_WALK_SKIP:        skip this dentry and its children
1251  */
1252 enum d_walk_ret {
1253         D_WALK_CONTINUE,
1254         D_WALK_QUIT,
1255         D_WALK_NORETRY,
1256         D_WALK_SKIP,
1257 };
1258 
1259 /**
1260  * d_walk - walk the dentry tree
1261  * @parent:     start of walk
1262  * @data:       data passed to @enter() and @finish()
1263  * @enter:      callback when first entering the dentry
1264  *
1265  * The @enter() callbacks are called with d_lock held.
1266  */
1267 static void d_walk(struct dentry *parent, void *data,
1268                    enum d_walk_ret (*enter)(void *, struct dentry *))
1269 {
1270         struct dentry *this_parent;
1271         struct list_head *next;
1272         unsigned seq = 0;
1273         enum d_walk_ret ret;
1274         bool retry = true;
1275 
1276 again:
1277         read_seqbegin_or_lock(&rename_lock, &seq);
1278         this_parent = parent;
1279         spin_lock(&this_parent->d_lock);
1280 
1281         ret = enter(data, this_parent);
1282         switch (ret) {
1283         case D_WALK_CONTINUE:
1284                 break;
1285         case D_WALK_QUIT:
1286         case D_WALK_SKIP:
1287                 goto out_unlock;
1288         case D_WALK_NORETRY:
1289                 retry = false;
1290                 break;
1291         }
1292 repeat:
1293         next = this_parent->d_subdirs.next;
1294 resume:
1295         while (next != &this_parent->d_subdirs) {
1296                 struct list_head *tmp = next;
1297                 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1298                 next = tmp->next;
1299 
1300                 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1301                         continue;
1302 
1303                 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1304 
1305                 ret = enter(data, dentry);
1306                 switch (ret) {
1307                 case D_WALK_CONTINUE:
1308                         break;
1309                 case D_WALK_QUIT:
1310                         spin_unlock(&dentry->d_lock);
1311                         goto out_unlock;
1312                 case D_WALK_NORETRY:
1313                         retry = false;
1314                         break;
1315                 case D_WALK_SKIP:
1316                         spin_unlock(&dentry->d_lock);
1317                         continue;
1318                 }
1319 
1320                 if (!list_empty(&dentry->d_subdirs)) {
1321                         spin_unlock(&this_parent->d_lock);
1322                         spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1323                         this_parent = dentry;
1324                         spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1325                         goto repeat;
1326                 }
1327                 spin_unlock(&dentry->d_lock);
1328         }
1329         /*
1330          * All done at this level ... ascend and resume the search.
1331          */
1332         rcu_read_lock();
1333 ascend:
1334         if (this_parent != parent) {
1335                 struct dentry *child = this_parent;
1336                 this_parent = child->d_parent;
1337 
1338                 spin_unlock(&child->d_lock);
1339                 spin_lock(&this_parent->d_lock);
1340 
1341                 /* might go back up the wrong parent if we have had a rename. */
1342                 if (need_seqretry(&rename_lock, seq))
1343                         goto rename_retry;
1344                 /* go into the first sibling still alive */
1345                 do {
1346                         next = child->d_child.next;
1347                         if (next == &this_parent->d_subdirs)
1348                                 goto ascend;
1349                         child = list_entry(next, struct dentry, d_child);
1350                 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1351                 rcu_read_unlock();
1352                 goto resume;
1353         }
1354         if (need_seqretry(&rename_lock, seq))
1355                 goto rename_retry;
1356         rcu_read_unlock();
1357 
1358 out_unlock:
1359         spin_unlock(&this_parent->d_lock);
1360         done_seqretry(&rename_lock, seq);
1361         return;
1362 
1363 rename_retry:
1364         spin_unlock(&this_parent->d_lock);
1365         rcu_read_unlock();
1366         BUG_ON(seq & 1);
1367         if (!retry)
1368                 return;
1369         seq = 1;
1370         goto again;
1371 }
1372 
1373 struct check_mount {
1374         struct vfsmount *mnt;
1375         unsigned int mounted;
1376 };
1377 
1378 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1379 {
1380         struct check_mount *info = data;
1381         struct path path = { .mnt = info->mnt, .dentry = dentry };
1382 
1383         if (likely(!d_mountpoint(dentry)))
1384                 return D_WALK_CONTINUE;
1385         if (__path_is_mountpoint(&path)) {
1386                 info->mounted = 1;
1387                 return D_WALK_QUIT;
1388         }
1389         return D_WALK_CONTINUE;
1390 }
1391 
1392 /**
1393  * path_has_submounts - check for mounts over a dentry in the
1394  *                      current namespace.
1395  * @parent: path to check.
1396  *
1397  * Return true if the parent or its subdirectories contain
1398  * a mount point in the current namespace.
1399  */
1400 int path_has_submounts(const struct path *parent)
1401 {
1402         struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1403 
1404         read_seqlock_excl(&mount_lock);
1405         d_walk(parent->dentry, &data, path_check_mount);
1406         read_sequnlock_excl(&mount_lock);
1407 
1408         return data.mounted;
1409 }
1410 EXPORT_SYMBOL(path_has_submounts);
1411 
1412 /*
1413  * Called by mount code to set a mountpoint and check if the mountpoint is
1414  * reachable (e.g. NFS can unhash a directory dentry and then the complete
1415  * subtree can become unreachable).
1416  *
1417  * Only one of d_invalidate() and d_set_mounted() must succeed.  For
1418  * this reason take rename_lock and d_lock on dentry and ancestors.
1419  */
1420 int d_set_mounted(struct dentry *dentry)
1421 {
1422         struct dentry *p;
1423         int ret = -ENOENT;
1424         write_seqlock(&rename_lock);
1425         for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1426                 /* Need exclusion wrt. d_invalidate() */
1427                 spin_lock(&p->d_lock);
1428                 if (unlikely(d_unhashed(p))) {
1429                         spin_unlock(&p->d_lock);
1430                         goto out;
1431                 }
1432                 spin_unlock(&p->d_lock);
1433         }
1434         spin_lock(&dentry->d_lock);
1435         if (!d_unlinked(dentry)) {
1436                 ret = -EBUSY;
1437                 if (!d_mountpoint(dentry)) {
1438                         dentry->d_flags |= DCACHE_MOUNTED;
1439                         ret = 0;
1440                 }
1441         }
1442         spin_unlock(&dentry->d_lock);
1443 out:
1444         write_sequnlock(&rename_lock);
1445         return ret;
1446 }
1447 
1448 /*
1449  * Search the dentry child list of the specified parent,
1450  * and move any unused dentries to the end of the unused
1451  * list for prune_dcache(). We descend to the next level
1452  * whenever the d_subdirs list is non-empty and continue
1453  * searching.
1454  *
1455  * It returns zero iff there are no unused children,
1456  * otherwise  it returns the number of children moved to
1457  * the end of the unused list. This may not be the total
1458  * number of unused children, because select_parent can
1459  * drop the lock and return early due to latency
1460  * constraints.
1461  */
1462 
1463 struct select_data {
1464         struct dentry *start;
1465         union {
1466                 long found;
1467                 struct dentry *victim;
1468         };
1469         struct list_head dispose;
1470 };
1471 
1472 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1473 {
1474         struct select_data *data = _data;
1475         enum d_walk_ret ret = D_WALK_CONTINUE;
1476 
1477         if (data->start == dentry)
1478                 goto out;
1479 
1480         if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1481                 data->found++;
1482         } else {
1483                 if (dentry->d_flags & DCACHE_LRU_LIST)
1484                         d_lru_del(dentry);
1485                 if (!dentry->d_lockref.count) {
1486                         d_shrink_add(dentry, &data->dispose);
1487                         data->found++;
1488                 }
1489         }
1490         /*
1491          * We can return to the caller if we have found some (this
1492          * ensures forward progress). We'll be coming back to find
1493          * the rest.
1494          */
1495         if (!list_empty(&data->dispose))
1496                 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1497 out:
1498         return ret;
1499 }
1500 
1501 static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry)
1502 {
1503         struct select_data *data = _data;
1504         enum d_walk_ret ret = D_WALK_CONTINUE;
1505 
1506         if (data->start == dentry)
1507                 goto out;
1508 
1509         if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1510                 if (!dentry->d_lockref.count) {
1511                         rcu_read_lock();
1512                         data->victim = dentry;
1513                         return D_WALK_QUIT;
1514                 }
1515         } else {
1516                 if (dentry->d_flags & DCACHE_LRU_LIST)
1517                         d_lru_del(dentry);
1518                 if (!dentry->d_lockref.count)
1519                         d_shrink_add(dentry, &data->dispose);
1520         }
1521         /*
1522          * We can return to the caller if we have found some (this
1523          * ensures forward progress). We'll be coming back to find
1524          * the rest.
1525          */
1526         if (!list_empty(&data->dispose))
1527                 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1528 out:
1529         return ret;
1530 }
1531 
1532 /**
1533  * shrink_dcache_parent - prune dcache
1534  * @parent: parent of entries to prune
1535  *
1536  * Prune the dcache to remove unused children of the parent dentry.
1537  */
1538 void shrink_dcache_parent(struct dentry *parent)
1539 {
1540         for (;;) {
1541                 struct select_data data = {.start = parent};
1542 
1543                 INIT_LIST_HEAD(&data.dispose);
1544                 d_walk(parent, &data, select_collect);
1545 
1546                 if (!list_empty(&data.dispose)) {
1547                         shrink_dentry_list(&data.dispose);
1548                         continue;
1549                 }
1550 
1551                 cond_resched();
1552                 if (!data.found)
1553                         break;
1554                 data.victim = NULL;
1555                 d_walk(parent, &data, select_collect2);
1556                 if (data.victim) {
1557                         struct dentry *parent;
1558                         spin_lock(&data.victim->d_lock);
1559                         if (!shrink_lock_dentry(data.victim)) {
1560                                 spin_unlock(&data.victim->d_lock);
1561                                 rcu_read_unlock();
1562                         } else {
1563                                 rcu_read_unlock();
1564                                 parent = data.victim->d_parent;
1565                                 if (parent != data.victim)
1566                                         __dput_to_list(parent, &data.dispose);
1567                                 __dentry_kill(data.victim);
1568                         }
1569                 }
1570                 if (!list_empty(&data.dispose))
1571                         shrink_dentry_list(&data.dispose);
1572         }
1573 }
1574 EXPORT_SYMBOL(shrink_dcache_parent);
1575 
1576 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1577 {
1578         /* it has busy descendents; complain about those instead */
1579         if (!list_empty(&dentry->d_subdirs))
1580                 return D_WALK_CONTINUE;
1581 
1582         /* root with refcount 1 is fine */
1583         if (dentry == _data && dentry->d_lockref.count == 1)
1584                 return D_WALK_CONTINUE;
1585 
1586         printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1587                         " still in use (%d) [unmount of %s %s]\n",
1588                        dentry,
1589                        dentry->d_inode ?
1590                        dentry->d_inode->i_ino : 0UL,
1591                        dentry,
1592                        dentry->d_lockref.count,
1593                        dentry->d_sb->s_type->name,
1594                        dentry->d_sb->s_id);
1595         WARN_ON(1);
1596         return D_WALK_CONTINUE;
1597 }
1598 
1599 static void do_one_tree(struct dentry *dentry)
1600 {
1601         shrink_dcache_parent(dentry);
1602         d_walk(dentry, dentry, umount_check);
1603         d_drop(dentry);
1604         dput(dentry);
1605 }
1606 
1607 /*
1608  * destroy the dentries attached to a superblock on unmounting
1609  */
1610 void shrink_dcache_for_umount(struct super_block *sb)
1611 {
1612         struct dentry *dentry;
1613 
1614         WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1615 
1616         dentry = sb->s_root;
1617         sb->s_root = NULL;
1618         do_one_tree(dentry);
1619 
1620         while (!hlist_bl_empty(&sb->s_roots)) {
1621                 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1622                 do_one_tree(dentry);
1623         }
1624 }
1625 
1626 static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
1627 {
1628         struct dentry **victim = _data;
1629         if (d_mountpoint(dentry)) {
1630                 __dget_dlock(dentry);
1631                 *victim = dentry;
1632                 return D_WALK_QUIT;
1633         }
1634         return D_WALK_CONTINUE;
1635 }
1636 
1637 /**
1638  * d_invalidate - detach submounts, prune dcache, and drop
1639  * @dentry: dentry to invalidate (aka detach, prune and drop)
1640  */
1641 void d_invalidate(struct dentry *dentry)
1642 {
1643         bool had_submounts = false;
1644         spin_lock(&dentry->d_lock);
1645         if (d_unhashed(dentry)) {
1646                 spin_unlock(&dentry->d_lock);
1647                 return;
1648         }
1649         __d_drop(dentry);
1650         spin_unlock(&dentry->d_lock);
1651 
1652         /* Negative dentries can be dropped without further checks */
1653         if (!dentry->d_inode)
1654                 return;
1655 
1656         shrink_dcache_parent(dentry);
1657         for (;;) {
1658                 struct dentry *victim = NULL;
1659                 d_walk(dentry, &victim, find_submount);
1660                 if (!victim) {
1661                         if (had_submounts)
1662                                 shrink_dcache_parent(dentry);
1663                         return;
1664                 }
1665                 had_submounts = true;
1666                 detach_mounts(victim);
1667                 dput(victim);
1668         }
1669 }
1670 EXPORT_SYMBOL(d_invalidate);
1671 
1672 /**
1673  * __d_alloc    -       allocate a dcache entry
1674  * @sb: filesystem it will belong to
1675  * @name: qstr of the name
1676  *
1677  * Allocates a dentry. It returns %NULL if there is insufficient memory
1678  * available. On a success the dentry is returned. The name passed in is
1679  * copied and the copy passed in may be reused after this call.
1680  */
1681  
1682 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1683 {
1684         struct dentry *dentry;
1685         char *dname;
1686         int err;
1687 
1688         dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1689         if (!dentry)
1690                 return NULL;
1691 
1692         /*
1693          * We guarantee that the inline name is always NUL-terminated.
1694          * This way the memcpy() done by the name switching in rename
1695          * will still always have a NUL at the end, even if we might
1696          * be overwriting an internal NUL character
1697          */
1698         dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1699         if (unlikely(!name)) {
1700                 name = &slash_name;
1701                 dname = dentry->d_iname;
1702         } else if (name->len > DNAME_INLINE_LEN-1) {
1703                 size_t size = offsetof(struct external_name, name[1]);
1704                 struct external_name *p = kmalloc(size + name->len,
1705                                                   GFP_KERNEL_ACCOUNT |
1706                                                   __GFP_RECLAIMABLE);
1707                 if (!p) {
1708                         kmem_cache_free(dentry_cache, dentry); 
1709                         return NULL;
1710                 }
1711                 atomic_set(&p->u.count, 1);
1712                 dname = p->name;
1713         } else  {
1714                 dname = dentry->d_iname;
1715         }       
1716 
1717         dentry->d_name.len = name->len;
1718         dentry->d_name.hash = name->hash;
1719         memcpy(dname, name->name, name->len);
1720         dname[name->len] = 0;
1721 
1722         /* Make sure we always see the terminating NUL character */
1723         smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1724 
1725         dentry->d_lockref.count = 1;
1726         dentry->d_flags = 0;
1727         spin_lock_init(&dentry->d_lock);
1728         seqcount_init(&dentry->d_seq);
1729         dentry->d_inode = NULL;
1730         dentry->d_parent = dentry;
1731         dentry->d_sb = sb;
1732         dentry->d_op = NULL;
1733         dentry->d_fsdata = NULL;
1734         INIT_HLIST_BL_NODE(&dentry->d_hash);
1735         INIT_LIST_HEAD(&dentry->d_lru);
1736         INIT_LIST_HEAD(&dentry->d_subdirs);
1737         INIT_HLIST_NODE(&dentry->d_u.d_alias);
1738         INIT_LIST_HEAD(&dentry->d_child);
1739         d_set_d_op(dentry, dentry->d_sb->s_d_op);
1740 
1741         if (dentry->d_op && dentry->d_op->d_init) {
1742                 err = dentry->d_op->d_init(dentry);
1743                 if (err) {
1744                         if (dname_external(dentry))
1745                                 kfree(external_name(dentry));
1746                         kmem_cache_free(dentry_cache, dentry);
1747                         return NULL;
1748                 }
1749         }
1750 
1751         this_cpu_inc(nr_dentry);
1752 
1753         return dentry;
1754 }
1755 
1756 /**
1757  * d_alloc      -       allocate a dcache entry
1758  * @parent: parent of entry to allocate
1759  * @name: qstr of the name
1760  *
1761  * Allocates a dentry. It returns %NULL if there is insufficient memory
1762  * available. On a success the dentry is returned. The name passed in is
1763  * copied and the copy passed in may be reused after this call.
1764  */
1765 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1766 {
1767         struct dentry *dentry = __d_alloc(parent->d_sb, name);
1768         if (!dentry)
1769                 return NULL;
1770         spin_lock(&parent->d_lock);
1771         /*
1772          * don't need child lock because it is not subject
1773          * to concurrency here
1774          */
1775         __dget_dlock(parent);
1776         dentry->d_parent = parent;
1777         list_add(&dentry->d_child, &parent->d_subdirs);
1778         spin_unlock(&parent->d_lock);
1779 
1780         return dentry;
1781 }
1782 EXPORT_SYMBOL(d_alloc);
1783 
1784 struct dentry *d_alloc_anon(struct super_block *sb)
1785 {
1786         return __d_alloc(sb, NULL);
1787 }
1788 EXPORT_SYMBOL(d_alloc_anon);
1789 
1790 struct dentry *d_alloc_cursor(struct dentry * parent)
1791 {
1792         struct dentry *dentry = d_alloc_anon(parent->d_sb);
1793         if (dentry) {
1794                 dentry->d_flags |= DCACHE_DENTRY_CURSOR;
1795                 dentry->d_parent = dget(parent);
1796         }
1797         return dentry;
1798 }
1799 
1800 /**
1801  * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1802  * @sb: the superblock
1803  * @name: qstr of the name
1804  *
1805  * For a filesystem that just pins its dentries in memory and never
1806  * performs lookups at all, return an unhashed IS_ROOT dentry.
1807  * This is used for pipes, sockets et.al. - the stuff that should
1808  * never be anyone's children or parents.  Unlike all other
1809  * dentries, these will not have RCU delay between dropping the
1810  * last reference and freeing them.
1811  *
1812  * The only user is alloc_file_pseudo() and that's what should
1813  * be considered a public interface.  Don't use directly.
1814  */
1815 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1816 {
1817         struct dentry *dentry = __d_alloc(sb, name);
1818         if (likely(dentry))
1819                 dentry->d_flags |= DCACHE_NORCU;
1820         return dentry;
1821 }
1822 
1823 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1824 {
1825         struct qstr q;
1826 
1827         q.name = name;
1828         q.hash_len = hashlen_string(parent, name);
1829         return d_alloc(parent, &q);
1830 }
1831 EXPORT_SYMBOL(d_alloc_name);
1832 
1833 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1834 {
1835         WARN_ON_ONCE(dentry->d_op);
1836         WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH  |
1837                                 DCACHE_OP_COMPARE       |
1838                                 DCACHE_OP_REVALIDATE    |
1839                                 DCACHE_OP_WEAK_REVALIDATE       |
1840                                 DCACHE_OP_DELETE        |
1841                                 DCACHE_OP_REAL));
1842         dentry->d_op = op;
1843         if (!op)
1844                 return;
1845         if (op->d_hash)
1846                 dentry->d_flags |= DCACHE_OP_HASH;
1847         if (op->d_compare)
1848                 dentry->d_flags |= DCACHE_OP_COMPARE;
1849         if (op->d_revalidate)
1850                 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1851         if (op->d_weak_revalidate)
1852                 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1853         if (op->d_delete)
1854                 dentry->d_flags |= DCACHE_OP_DELETE;
1855         if (op->d_prune)
1856                 dentry->d_flags |= DCACHE_OP_PRUNE;
1857         if (op->d_real)
1858                 dentry->d_flags |= DCACHE_OP_REAL;
1859 
1860 }
1861 EXPORT_SYMBOL(d_set_d_op);
1862 
1863 
1864 /*
1865  * d_set_fallthru - Mark a dentry as falling through to a lower layer
1866  * @dentry - The dentry to mark
1867  *
1868  * Mark a dentry as falling through to the lower layer (as set with
1869  * d_pin_lower()).  This flag may be recorded on the medium.
1870  */
1871 void d_set_fallthru(struct dentry *dentry)
1872 {
1873         spin_lock(&dentry->d_lock);
1874         dentry->d_flags |= DCACHE_FALLTHRU;
1875         spin_unlock(&dentry->d_lock);
1876 }
1877 EXPORT_SYMBOL(d_set_fallthru);
1878 
1879 static unsigned d_flags_for_inode(struct inode *inode)
1880 {
1881         unsigned add_flags = DCACHE_REGULAR_TYPE;
1882 
1883         if (!inode)
1884                 return DCACHE_MISS_TYPE;
1885 
1886         if (S_ISDIR(inode->i_mode)) {
1887                 add_flags = DCACHE_DIRECTORY_TYPE;
1888                 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1889                         if (unlikely(!inode->i_op->lookup))
1890                                 add_flags = DCACHE_AUTODIR_TYPE;
1891                         else
1892                                 inode->i_opflags |= IOP_LOOKUP;
1893                 }
1894                 goto type_determined;
1895         }
1896 
1897         if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1898                 if (unlikely(inode->i_op->get_link)) {
1899                         add_flags = DCACHE_SYMLINK_TYPE;
1900                         goto type_determined;
1901                 }
1902                 inode->i_opflags |= IOP_NOFOLLOW;
1903         }
1904 
1905         if (unlikely(!S_ISREG(inode->i_mode)))
1906                 add_flags = DCACHE_SPECIAL_TYPE;
1907 
1908 type_determined:
1909         if (unlikely(IS_AUTOMOUNT(inode)))
1910                 add_flags |= DCACHE_NEED_AUTOMOUNT;
1911         return add_flags;
1912 }
1913 
1914 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1915 {
1916         unsigned add_flags = d_flags_for_inode(inode);
1917         WARN_ON(d_in_lookup(dentry));
1918 
1919         spin_lock(&dentry->d_lock);
1920         /*
1921          * Decrement negative dentry count if it was in the LRU list.
1922          */
1923         if (dentry->d_flags & DCACHE_LRU_LIST)
1924                 this_cpu_dec(nr_dentry_negative);
1925         hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1926         raw_write_seqcount_begin(&dentry->d_seq);
1927         __d_set_inode_and_type(dentry, inode, add_flags);
1928         raw_write_seqcount_end(&dentry->d_seq);
1929         fsnotify_update_flags(dentry);
1930         spin_unlock(&dentry->d_lock);
1931 }
1932 
1933 /**
1934  * d_instantiate - fill in inode information for a dentry
1935  * @entry: dentry to complete
1936  * @inode: inode to attach to this dentry
1937  *
1938  * Fill in inode information in the entry.
1939  *
1940  * This turns negative dentries into productive full members
1941  * of society.
1942  *
1943  * NOTE! This assumes that the inode count has been incremented
1944  * (or otherwise set) by the caller to indicate that it is now
1945  * in use by the dcache.
1946  */
1947  
1948 void d_instantiate(struct dentry *entry, struct inode * inode)
1949 {
1950         BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1951         if (inode) {
1952                 security_d_instantiate(entry, inode);
1953                 spin_lock(&inode->i_lock);
1954                 __d_instantiate(entry, inode);
1955                 spin_unlock(&inode->i_lock);
1956         }
1957 }
1958 EXPORT_SYMBOL(d_instantiate);
1959 
1960 /*
1961  * This should be equivalent to d_instantiate() + unlock_new_inode(),
1962  * with lockdep-related part of unlock_new_inode() done before
1963  * anything else.  Use that instead of open-coding d_instantiate()/
1964  * unlock_new_inode() combinations.
1965  */
1966 void d_instantiate_new(struct dentry *entry, struct inode *inode)
1967 {
1968         BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1969         BUG_ON(!inode);
1970         lockdep_annotate_inode_mutex_key(inode);
1971         security_d_instantiate(entry, inode);
1972         spin_lock(&inode->i_lock);
1973         __d_instantiate(entry, inode);
1974         WARN_ON(!(inode->i_state & I_NEW));
1975         inode->i_state &= ~I_NEW & ~I_CREATING;
1976         smp_mb();
1977         wake_up_bit(&inode->i_state, __I_NEW);
1978         spin_unlock(&inode->i_lock);
1979 }
1980 EXPORT_SYMBOL(d_instantiate_new);
1981 
1982 struct dentry *d_make_root(struct inode *root_inode)
1983 {
1984         struct dentry *res = NULL;
1985 
1986         if (root_inode) {
1987                 res = d_alloc_anon(root_inode->i_sb);
1988                 if (res)
1989                         d_instantiate(res, root_inode);
1990                 else
1991                         iput(root_inode);
1992         }
1993         return res;
1994 }
1995 EXPORT_SYMBOL(d_make_root);
1996 
1997 static struct dentry *__d_instantiate_anon(struct dentry *dentry,
1998                                            struct inode *inode,
1999                                            bool disconnected)
2000 {
2001         struct dentry *res;
2002         unsigned add_flags;
2003 
2004         security_d_instantiate(dentry, inode);
2005         spin_lock(&inode->i_lock);
2006         res = __d_find_any_alias(inode);
2007         if (res) {
2008                 spin_unlock(&inode->i_lock);
2009                 dput(dentry);
2010                 goto out_iput;
2011         }
2012 
2013         /* attach a disconnected dentry */
2014         add_flags = d_flags_for_inode(inode);
2015 
2016         if (disconnected)
2017                 add_flags |= DCACHE_DISCONNECTED;
2018 
2019         spin_lock(&dentry->d_lock);
2020         __d_set_inode_and_type(dentry, inode, add_flags);
2021         hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2022         if (!disconnected) {
2023                 hlist_bl_lock(&dentry->d_sb->s_roots);
2024                 hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots);
2025                 hlist_bl_unlock(&dentry->d_sb->s_roots);
2026         }
2027         spin_unlock(&dentry->d_lock);
2028         spin_unlock(&inode->i_lock);
2029 
2030         return dentry;
2031 
2032  out_iput:
2033         iput(inode);
2034         return res;
2035 }
2036 
2037 struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode)
2038 {
2039         return __d_instantiate_anon(dentry, inode, true);
2040 }
2041 EXPORT_SYMBOL(d_instantiate_anon);
2042 
2043 static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
2044 {
2045         struct dentry *tmp;
2046         struct dentry *res;
2047 
2048         if (!inode)
2049                 return ERR_PTR(-ESTALE);
2050         if (IS_ERR(inode))
2051                 return ERR_CAST(inode);
2052 
2053         res = d_find_any_alias(inode);
2054         if (res)
2055                 goto out_iput;
2056 
2057         tmp = d_alloc_anon(inode->i_sb);
2058         if (!tmp) {
2059                 res = ERR_PTR(-ENOMEM);
2060                 goto out_iput;
2061         }
2062 
2063         return __d_instantiate_anon(tmp, inode, disconnected);
2064 
2065 out_iput:
2066         iput(inode);
2067         return res;
2068 }
2069 
2070 /**
2071  * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2072  * @inode: inode to allocate the dentry for
2073  *
2074  * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2075  * similar open by handle operations.  The returned dentry may be anonymous,
2076  * or may have a full name (if the inode was already in the cache).
2077  *
2078  * When called on a directory inode, we must ensure that the inode only ever
2079  * has one dentry.  If a dentry is found, that is returned instead of
2080  * allocating a new one.
2081  *
2082  * On successful return, the reference to the inode has been transferred
2083  * to the dentry.  In case of an error the reference on the inode is released.
2084  * To make it easier to use in export operations a %NULL or IS_ERR inode may
2085  * be passed in and the error will be propagated to the return value,
2086  * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2087  */
2088 struct dentry *d_obtain_alias(struct inode *inode)
2089 {
2090         return __d_obtain_alias(inode, true);
2091 }
2092 EXPORT_SYMBOL(d_obtain_alias);
2093 
2094 /**
2095  * d_obtain_root - find or allocate a dentry for a given inode
2096  * @inode: inode to allocate the dentry for
2097  *
2098  * Obtain an IS_ROOT dentry for the root of a filesystem.
2099  *
2100  * We must ensure that directory inodes only ever have one dentry.  If a
2101  * dentry is found, that is returned instead of allocating a new one.
2102  *
2103  * On successful return, the reference to the inode has been transferred
2104  * to the dentry.  In case of an error the reference on the inode is
2105  * released.  A %NULL or IS_ERR inode may be passed in and will be the
2106  * error will be propagate to the return value, with a %NULL @inode
2107  * replaced by ERR_PTR(-ESTALE).
2108  */
2109 struct dentry *d_obtain_root(struct inode *inode)
2110 {
2111         return __d_obtain_alias(inode, false);
2112 }
2113 EXPORT_SYMBOL(d_obtain_root);
2114 
2115 /**
2116  * d_add_ci - lookup or allocate new dentry with case-exact name
2117  * @inode:  the inode case-insensitive lookup has found
2118  * @dentry: the negative dentry that was passed to the parent's lookup func
2119  * @name:   the case-exact name to be associated with the returned dentry
2120  *
2121  * This is to avoid filling the dcache with case-insensitive names to the
2122  * same inode, only the actual correct case is stored in the dcache for
2123  * case-insensitive filesystems.
2124  *
2125  * For a case-insensitive lookup match and if the the case-exact dentry
2126  * already exists in in the dcache, use it and return it.
2127  *
2128  * If no entry exists with the exact case name, allocate new dentry with
2129  * the exact case, and return the spliced entry.
2130  */
2131 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2132                         struct qstr *name)
2133 {
2134         struct dentry *found, *res;
2135 
2136         /*
2137          * First check if a dentry matching the name already exists,
2138          * if not go ahead and create it now.
2139          */
2140         found = d_hash_and_lookup(dentry->d_parent, name);
2141         if (found) {
2142                 iput(inode);
2143                 return found;
2144         }
2145         if (d_in_lookup(dentry)) {
2146                 found = d_alloc_parallel(dentry->d_parent, name,
2147                                         dentry->d_wait);
2148                 if (IS_ERR(found) || !d_in_lookup(found)) {
2149                         iput(inode);
2150                         return found;
2151                 }
2152         } else {
2153                 found = d_alloc(dentry->d_parent, name);
2154                 if (!found) {
2155                         iput(inode);
2156                         return ERR_PTR(-ENOMEM);
2157                 } 
2158         }
2159         res = d_splice_alias(inode, found);
2160         if (res) {
2161                 dput(found);
2162                 return res;
2163         }
2164         return found;
2165 }
2166 EXPORT_SYMBOL(d_add_ci);
2167 
2168 
2169 static inline bool d_same_name(const struct dentry *dentry,
2170                                 const struct dentry *parent,
2171                                 const struct qstr *name)
2172 {
2173         if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2174                 if (dentry->d_name.len != name->len)
2175                         return false;
2176                 return dentry_cmp(dentry, name->name, name->len) == 0;
2177         }
2178         return parent->d_op->d_compare(dentry,
2179                                        dentry->d_name.len, dentry->d_name.name,
2180                                        name) == 0;
2181 }
2182 
2183 /**
2184  * __d_lookup_rcu - search for a dentry (racy, store-free)
2185  * @parent: parent dentry
2186  * @name: qstr of name we wish to find
2187  * @seqp: returns d_seq value at the point where the dentry was found
2188  * Returns: dentry, or NULL
2189  *
2190  * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2191  * resolution (store-free path walking) design described in
2192  * Documentation/filesystems/path-lookup.txt.
2193  *
2194  * This is not to be used outside core vfs.
2195  *
2196  * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2197  * held, and rcu_read_lock held. The returned dentry must not be stored into
2198  * without taking d_lock and checking d_seq sequence count against @seq
2199  * returned here.
2200  *
2201  * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2202  * function.
2203  *
2204  * Alternatively, __d_lookup_rcu may be called again to look up the child of
2205  * the returned dentry, so long as its parent's seqlock is checked after the
2206  * child is looked up. Thus, an interlocking stepping of sequence lock checks
2207  * is formed, giving integrity down the path walk.
2208  *
2209  * NOTE! The caller *has* to check the resulting dentry against the sequence
2210  * number we've returned before using any of the resulting dentry state!
2211  */
2212 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2213                                 const struct qstr *name,
2214                                 unsigned *seqp)
2215 {
2216         u64 hashlen = name->hash_len;
2217         const unsigned char *str = name->name;
2218         struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2219         struct hlist_bl_node *node;
2220         struct dentry *dentry;
2221 
2222         /*
2223          * Note: There is significant duplication with __d_lookup_rcu which is
2224          * required to prevent single threaded performance regressions
2225          * especially on architectures where smp_rmb (in seqcounts) are costly.
2226          * Keep the two functions in sync.
2227          */
2228 
2229         /*
2230          * The hash list is protected using RCU.
2231          *
2232          * Carefully use d_seq when comparing a candidate dentry, to avoid
2233          * races with d_move().
2234          *
2235          * It is possible that concurrent renames can mess up our list
2236          * walk here and result in missing our dentry, resulting in the
2237          * false-negative result. d_lookup() protects against concurrent
2238          * renames using rename_lock seqlock.
2239          *
2240          * See Documentation/filesystems/path-lookup.txt for more details.
2241          */
2242         hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2243                 unsigned seq;
2244 
2245 seqretry:
2246                 /*
2247                  * The dentry sequence count protects us from concurrent
2248                  * renames, and thus protects parent and name fields.
2249                  *
2250                  * The caller must perform a seqcount check in order
2251                  * to do anything useful with the returned dentry.
2252                  *
2253                  * NOTE! We do a "raw" seqcount_begin here. That means that
2254                  * we don't wait for the sequence count to stabilize if it
2255                  * is in the middle of a sequence change. If we do the slow
2256                  * dentry compare, we will do seqretries until it is stable,
2257                  * and if we end up with a successful lookup, we actually
2258                  * want to exit RCU lookup anyway.
2259                  *
2260                  * Note that raw_seqcount_begin still *does* smp_rmb(), so
2261                  * we are still guaranteed NUL-termination of ->d_name.name.
2262                  */
2263                 seq = raw_seqcount_begin(&dentry->d_seq);
2264                 if (dentry->d_parent != parent)
2265                         continue;
2266                 if (d_unhashed(dentry))
2267                         continue;
2268 
2269                 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2270                         int tlen;
2271                         const char *tname;
2272                         if (dentry->d_name.hash != hashlen_hash(hashlen))
2273                                 continue;
2274                         tlen = dentry->d_name.len;
2275                         tname = dentry->d_name.name;
2276                         /* we want a consistent (name,len) pair */
2277                         if (read_seqcount_retry(&dentry->d_seq, seq)) {
2278                                 cpu_relax();
2279                                 goto seqretry;
2280                         }
2281                         if (parent->d_op->d_compare(dentry,
2282                                                     tlen, tname, name) != 0)
2283                                 continue;
2284                 } else {
2285                         if (dentry->d_name.hash_len != hashlen)
2286                                 continue;
2287                         if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2288                                 continue;
2289                 }
2290                 *seqp = seq;
2291                 return dentry;
2292         }
2293         return NULL;
2294 }
2295 
2296 /**
2297  * d_lookup - search for a dentry
2298  * @parent: parent dentry
2299  * @name: qstr of name we wish to find
2300  * Returns: dentry, or NULL
2301  *
2302  * d_lookup searches the children of the parent dentry for the name in
2303  * question. If the dentry is found its reference count is incremented and the
2304  * dentry is returned. The caller must use dput to free the entry when it has
2305  * finished using it. %NULL is returned if the dentry does not exist.
2306  */
2307 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2308 {
2309         struct dentry *dentry;
2310         unsigned seq;
2311 
2312         do {
2313                 seq = read_seqbegin(&rename_lock);
2314                 dentry = __d_lookup(parent, name);
2315                 if (dentry)
2316                         break;
2317         } while (read_seqretry(&rename_lock, seq));
2318         return dentry;
2319 }
2320 EXPORT_SYMBOL(d_lookup);
2321 
2322 /**
2323  * __d_lookup - search for a dentry (racy)
2324  * @parent: parent dentry
2325  * @name: qstr of name we wish to find
2326  * Returns: dentry, or NULL
2327  *
2328  * __d_lookup is like d_lookup, however it may (rarely) return a
2329  * false-negative result due to unrelated rename activity.
2330  *
2331  * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2332  * however it must be used carefully, eg. with a following d_lookup in
2333  * the case of failure.
2334  *
2335  * __d_lookup callers must be commented.
2336  */
2337 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2338 {
2339         unsigned int hash = name->hash;
2340         struct hlist_bl_head *b = d_hash(hash);
2341         struct hlist_bl_node *node;
2342         struct dentry *found = NULL;
2343         struct dentry *dentry;
2344 
2345         /*
2346          * Note: There is significant duplication with __d_lookup_rcu which is
2347          * required to prevent single threaded performance regressions
2348          * especially on architectures where smp_rmb (in seqcounts) are costly.
2349          * Keep the two functions in sync.
2350          */
2351 
2352         /*
2353          * The hash list is protected using RCU.
2354          *
2355          * Take d_lock when comparing a candidate dentry, to avoid races
2356          * with d_move().
2357          *
2358          * It is possible that concurrent renames can mess up our list
2359          * walk here and result in missing our dentry, resulting in the
2360          * false-negative result. d_lookup() protects against concurrent
2361          * renames using rename_lock seqlock.
2362          *
2363          * See Documentation/filesystems/path-lookup.txt for more details.
2364          */
2365         rcu_read_lock();
2366         
2367         hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2368 
2369                 if (dentry->d_name.hash != hash)
2370                         continue;
2371 
2372                 spin_lock(&dentry->d_lock);
2373                 if (dentry->d_parent != parent)
2374                         goto next;
2375                 if (d_unhashed(dentry))
2376                         goto next;
2377 
2378                 if (!d_same_name(dentry, parent, name))
2379                         goto next;
2380 
2381                 dentry->d_lockref.count++;
2382                 found = dentry;
2383                 spin_unlock(&dentry->d_lock);
2384                 break;
2385 next:
2386                 spin_unlock(&dentry->d_lock);
2387         }
2388         rcu_read_unlock();
2389 
2390         return found;
2391 }
2392 
2393 /**
2394  * d_hash_and_lookup - hash the qstr then search for a dentry
2395  * @dir: Directory to search in
2396  * @name: qstr of name we wish to find
2397  *
2398  * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2399  */
2400 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2401 {
2402         /*
2403          * Check for a fs-specific hash function. Note that we must
2404          * calculate the standard hash first, as the d_op->d_hash()
2405          * routine may choose to leave the hash value unchanged.
2406          */
2407         name->hash = full_name_hash(dir, name->name, name->len);
2408         if (dir->d_flags & DCACHE_OP_HASH) {
2409                 int err = dir->d_op->d_hash(dir, name);
2410                 if (unlikely(err < 0))
2411                         return ERR_PTR(err);
2412         }
2413         return d_lookup(dir, name);
2414 }
2415 EXPORT_SYMBOL(d_hash_and_lookup);
2416 
2417 /*
2418  * When a file is deleted, we have two options:
2419  * - turn this dentry into a negative dentry
2420  * - unhash this dentry and free it.
2421  *
2422  * Usually, we want to just turn this into
2423  * a negative dentry, but if anybody else is
2424  * currently using the dentry or the inode
2425  * we can't do that and we fall back on removing
2426  * it from the hash queues and waiting for
2427  * it to be deleted later when it has no users
2428  */
2429  
2430 /**
2431  * d_delete - delete a dentry
2432  * @dentry: The dentry to delete
2433  *
2434  * Turn the dentry into a negative dentry if possible, otherwise
2435  * remove it from the hash queues so it can be deleted later
2436  */
2437  
2438 void d_delete(struct dentry * dentry)
2439 {
2440         struct inode *inode = dentry->d_inode;
2441 
2442         spin_lock(&inode->i_lock);
2443         spin_lock(&dentry->d_lock);
2444         /*
2445          * Are we the only user?
2446          */
2447         if (dentry->d_lockref.count == 1) {
2448                 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2449                 dentry_unlink_inode(dentry);
2450         } else {
2451                 __d_drop(dentry);
2452                 spin_unlock(&dentry->d_lock);
2453                 spin_unlock(&inode->i_lock);
2454         }
2455 }
2456 EXPORT_SYMBOL(d_delete);
2457 
2458 static void __d_rehash(struct dentry *entry)
2459 {
2460         struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2461 
2462         hlist_bl_lock(b);
2463         hlist_bl_add_head_rcu(&entry->d_hash, b);
2464         hlist_bl_unlock(b);
2465 }
2466 
2467 /**
2468  * d_rehash     - add an entry back to the hash
2469  * @entry: dentry to add to the hash
2470  *
2471  * Adds a dentry to the hash according to its name.
2472  */
2473  
2474 void d_rehash(struct dentry * entry)
2475 {
2476         spin_lock(&entry->d_lock);
2477         __d_rehash(entry);
2478         spin_unlock(&entry->d_lock);
2479 }
2480 EXPORT_SYMBOL(d_rehash);
2481 
2482 static inline unsigned start_dir_add(struct inode *dir)
2483 {
2484 
2485         for (;;) {
2486                 unsigned n = dir->i_dir_seq;
2487                 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2488                         return n;
2489                 cpu_relax();
2490         }
2491 }
2492 
2493 static inline void end_dir_add(struct inode *dir, unsigned n)
2494 {
2495         smp_store_release(&dir->i_dir_seq, n + 2);
2496 }
2497 
2498 static void d_wait_lookup(struct dentry *dentry)
2499 {
2500         if (d_in_lookup(dentry)) {
2501                 DECLARE_WAITQUEUE(wait, current);
2502                 add_wait_queue(dentry->d_wait, &wait);
2503                 do {
2504                         set_current_state(TASK_UNINTERRUPTIBLE);
2505                         spin_unlock(&dentry->d_lock);
2506                         schedule();
2507                         spin_lock(&dentry->d_lock);
2508                 } while (d_in_lookup(dentry));
2509         }
2510 }
2511 
2512 struct dentry *d_alloc_parallel(struct dentry *parent,
2513                                 const struct qstr *name,
2514                                 wait_queue_head_t *wq)
2515 {
2516         unsigned int hash = name->hash;
2517         struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2518         struct hlist_bl_node *node;
2519         struct dentry *new = d_alloc(parent, name);
2520         struct dentry *dentry;
2521         unsigned seq, r_seq, d_seq;
2522 
2523         if (unlikely(!new))
2524                 return ERR_PTR(-ENOMEM);
2525 
2526 retry:
2527         rcu_read_lock();
2528         seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2529         r_seq = read_seqbegin(&rename_lock);
2530         dentry = __d_lookup_rcu(parent, name, &d_seq);
2531         if (unlikely(dentry)) {
2532                 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2533                         rcu_read_unlock();
2534                         goto retry;
2535                 }
2536                 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2537                         rcu_read_unlock();
2538                         dput(dentry);
2539                         goto retry;
2540                 }
2541                 rcu_read_unlock();
2542                 dput(new);
2543                 return dentry;
2544         }
2545         if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2546                 rcu_read_unlock();
2547                 goto retry;
2548         }
2549 
2550         if (unlikely(seq & 1)) {
2551                 rcu_read_unlock();
2552                 goto retry;
2553         }
2554 
2555         hlist_bl_lock(b);
2556         if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2557                 hlist_bl_unlock(b);
2558                 rcu_read_unlock();
2559                 goto retry;
2560         }
2561         /*
2562          * No changes for the parent since the beginning of d_lookup().
2563          * Since all removals from the chain happen with hlist_bl_lock(),
2564          * any potential in-lookup matches are going to stay here until
2565          * we unlock the chain.  All fields are stable in everything
2566          * we encounter.
2567          */
2568         hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2569                 if (dentry->d_name.hash != hash)
2570                         continue;
2571                 if (dentry->d_parent != parent)
2572                         continue;
2573                 if (!d_same_name(dentry, parent, name))
2574                         continue;
2575                 hlist_bl_unlock(b);
2576                 /* now we can try to grab a reference */
2577                 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2578                         rcu_read_unlock();
2579                         goto retry;
2580                 }
2581 
2582                 rcu_read_unlock();
2583                 /*
2584                  * somebody is likely to be still doing lookup for it;
2585                  * wait for them to finish
2586                  */
2587                 spin_lock(&dentry->d_lock);
2588                 d_wait_lookup(dentry);
2589                 /*
2590                  * it's not in-lookup anymore; in principle we should repeat
2591                  * everything from dcache lookup, but it's likely to be what
2592                  * d_lookup() would've found anyway.  If it is, just return it;
2593                  * otherwise we really have to repeat the whole thing.
2594                  */
2595                 if (unlikely(dentry->d_name.hash != hash))
2596                         goto mismatch;
2597                 if (unlikely(dentry->d_parent != parent))
2598                         goto mismatch;
2599                 if (unlikely(d_unhashed(dentry)))
2600                         goto mismatch;
2601                 if (unlikely(!d_same_name(dentry, parent, name)))
2602                         goto mismatch;
2603                 /* OK, it *is* a hashed match; return it */
2604                 spin_unlock(&dentry->d_lock);
2605                 dput(new);
2606                 return dentry;
2607         }
2608         rcu_read_unlock();
2609         /* we can't take ->d_lock here; it's OK, though. */
2610         new->d_flags |= DCACHE_PAR_LOOKUP;
2611         new->d_wait = wq;
2612         hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2613         hlist_bl_unlock(b);
2614         return new;
2615 mismatch:
2616         spin_unlock(&dentry->d_lock);
2617         dput(dentry);
2618         goto retry;
2619 }
2620 EXPORT_SYMBOL(d_alloc_parallel);
2621 
2622 void __d_lookup_done(struct dentry *dentry)
2623 {
2624         struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2625                                                  dentry->d_name.hash);
2626         hlist_bl_lock(b);
2627         dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2628         __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2629         wake_up_all(dentry->d_wait);
2630         dentry->d_wait = NULL;
2631         hlist_bl_unlock(b);
2632         INIT_HLIST_NODE(&dentry->d_u.d_alias);
2633         INIT_LIST_HEAD(&dentry->d_lru);
2634 }
2635 EXPORT_SYMBOL(__d_lookup_done);
2636 
2637 /* inode->i_lock held if inode is non-NULL */
2638 
2639 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2640 {
2641         struct inode *dir = NULL;
2642         unsigned n;
2643         spin_lock(&dentry->d_lock);
2644         if (unlikely(d_in_lookup(dentry))) {
2645                 dir = dentry->d_parent->d_inode;
2646                 n = start_dir_add(dir);
2647                 __d_lookup_done(dentry);
2648         }
2649         if (inode) {
2650                 unsigned add_flags = d_flags_for_inode(inode);
2651                 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2652                 raw_write_seqcount_begin(&dentry->d_seq);
2653                 __d_set_inode_and_type(dentry, inode, add_flags);
2654                 raw_write_seqcount_end(&dentry->d_seq);
2655                 fsnotify_update_flags(dentry);
2656         }
2657         __d_rehash(dentry);
2658         if (dir)
2659                 end_dir_add(dir, n);
2660         spin_unlock(&dentry->d_lock);
2661         if (inode)
2662                 spin_unlock(&inode->i_lock);
2663 }
2664 
2665 /**
2666  * d_add - add dentry to hash queues
2667  * @entry: dentry to add
2668  * @inode: The inode to attach to this dentry
2669  *
2670  * This adds the entry to the hash queues and initializes @inode.
2671  * The entry was actually filled in earlier during d_alloc().
2672  */
2673 
2674 void d_add(struct dentry *entry, struct inode *inode)
2675 {
2676         if (inode) {
2677                 security_d_instantiate(entry, inode);
2678                 spin_lock(&inode->i_lock);
2679         }
2680         __d_add(entry, inode);
2681 }
2682 EXPORT_SYMBOL(d_add);
2683 
2684 /**
2685  * d_exact_alias - find and hash an exact unhashed alias
2686  * @entry: dentry to add
2687  * @inode: The inode to go with this dentry
2688  *
2689  * If an unhashed dentry with the same name/parent and desired
2690  * inode already exists, hash and return it.  Otherwise, return
2691  * NULL.
2692  *
2693  * Parent directory should be locked.
2694  */
2695 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2696 {
2697         struct dentry *alias;
2698         unsigned int hash = entry->d_name.hash;
2699 
2700         spin_lock(&inode->i_lock);
2701         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2702                 /*
2703                  * Don't need alias->d_lock here, because aliases with
2704                  * d_parent == entry->d_parent are not subject to name or
2705                  * parent changes, because the parent inode i_mutex is held.
2706                  */
2707                 if (alias->d_name.hash != hash)
2708                         continue;
2709                 if (alias->d_parent != entry->d_parent)
2710                         continue;
2711                 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2712                         continue;
2713                 spin_lock(&alias->d_lock);
2714                 if (!d_unhashed(alias)) {
2715                         spin_unlock(&alias->d_lock);
2716                         alias = NULL;
2717                 } else {
2718                         __dget_dlock(alias);
2719                         __d_rehash(alias);
2720                         spin_unlock(&alias->d_lock);
2721                 }
2722                 spin_unlock(&inode->i_lock);
2723                 return alias;
2724         }
2725         spin_unlock(&inode->i_lock);
2726         return NULL;
2727 }
2728 EXPORT_SYMBOL(d_exact_alias);
2729 
2730 static void swap_names(struct dentry *dentry, struct dentry *target)
2731 {
2732         if (unlikely(dname_external(target))) {
2733                 if (unlikely(dname_external(dentry))) {
2734                         /*
2735                          * Both external: swap the pointers
2736                          */
2737                         swap(target->d_name.name, dentry->d_name.name);
2738                 } else {
2739                         /*
2740                          * dentry:internal, target:external.  Steal target's
2741                          * storage and make target internal.
2742                          */
2743                         memcpy(target->d_iname, dentry->d_name.name,
2744                                         dentry->d_name.len + 1);
2745                         dentry->d_name.name = target->d_name.name;
2746                         target->d_name.name = target->d_iname;
2747                 }
2748         } else {
2749                 if (unlikely(dname_external(dentry))) {
2750                         /*
2751                          * dentry:external, target:internal.  Give dentry's
2752                          * storage to target and make dentry internal
2753                          */
2754                         memcpy(dentry->d_iname, target->d_name.name,
2755                                         target->d_name.len + 1);
2756                         target->d_name.name = dentry->d_name.name;
2757                         dentry->d_name.name = dentry->d_iname;
2758                 } else {
2759                         /*
2760                          * Both are internal.
2761                          */
2762                         unsigned int i;
2763                         BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2764                         for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2765                                 swap(((long *) &dentry->d_iname)[i],
2766                                      ((long *) &target->d_iname)[i]);
2767                         }
2768                 }
2769         }
2770         swap(dentry->d_name.hash_len, target->d_name.hash_len);
2771 }
2772 
2773 static void copy_name(struct dentry *dentry, struct dentry *target)
2774 {
2775         struct external_name *old_name = NULL;
2776         if (unlikely(dname_external(dentry)))
2777                 old_name = external_name(dentry);
2778         if (unlikely(dname_external(target))) {
2779                 atomic_inc(&external_name(target)->u.count);
2780                 dentry->d_name = target->d_name;
2781         } else {
2782                 memcpy(dentry->d_iname, target->d_name.name,
2783                                 target->d_name.len + 1);
2784                 dentry->d_name.name = dentry->d_iname;
2785                 dentry->d_name.hash_len = target->d_name.hash_len;
2786         }
2787         if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2788                 kfree_rcu(old_name, u.head);
2789 }
2790 
2791 /*
2792  * __d_move - move a dentry
2793  * @dentry: entry to move
2794  * @target: new dentry
2795  * @exchange: exchange the two dentries
2796  *
2797  * Update the dcache to reflect the move of a file name. Negative
2798  * dcache entries should not be moved in this way. Caller must hold
2799  * rename_lock, the i_mutex of the source and target directories,
2800  * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2801  */
2802 static void __d_move(struct dentry *dentry, struct dentry *target,
2803                      bool exchange)
2804 {
2805         struct dentry *old_parent, *p;
2806         struct inode *dir = NULL;
2807         unsigned n;
2808 
2809         WARN_ON(!dentry->d_inode);
2810         if (WARN_ON(dentry == target))
2811                 return;
2812 
2813         BUG_ON(d_ancestor(target, dentry));
2814         old_parent = dentry->d_parent;
2815         p = d_ancestor(old_parent, target);
2816         if (IS_ROOT(dentry)) {
2817                 BUG_ON(p);
2818                 spin_lock(&target->d_parent->d_lock);
2819         } else if (!p) {
2820                 /* target is not a descendent of dentry->d_parent */
2821                 spin_lock(&target->d_parent->d_lock);
2822                 spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
2823         } else {
2824                 BUG_ON(p == dentry);
2825                 spin_lock(&old_parent->d_lock);
2826                 if (p != target)
2827                         spin_lock_nested(&target->d_parent->d_lock,
2828                                         DENTRY_D_LOCK_NESTED);
2829         }
2830         spin_lock_nested(&dentry->d_lock, 2);
2831         spin_lock_nested(&target->d_lock, 3);
2832 
2833         if (unlikely(d_in_lookup(target))) {
2834                 dir = target->d_parent->d_inode;
2835                 n = start_dir_add(dir);
2836                 __d_lookup_done(target);
2837         }
2838 
2839         write_seqcount_begin(&dentry->d_seq);
2840         write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2841 
2842         /* unhash both */
2843         if (!d_unhashed(dentry))
2844                 ___d_drop(dentry);
2845         if (!d_unhashed(target))
2846                 ___d_drop(target);
2847 
2848         /* ... and switch them in the tree */
2849         dentry->d_parent = target->d_parent;
2850         if (!exchange) {
2851                 copy_name(dentry, target);
2852                 target->d_hash.pprev = NULL;
2853                 dentry->d_parent->d_lockref.count++;
2854                 if (dentry != old_parent) /* wasn't IS_ROOT */
2855                         WARN_ON(!--old_parent->d_lockref.count);
2856         } else {
2857                 target->d_parent = old_parent;
2858                 swap_names(dentry, target);
2859                 list_move(&target->d_child, &target->d_parent->d_subdirs);
2860                 __d_rehash(target);
2861                 fsnotify_update_flags(target);
2862         }
2863         list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2864         __d_rehash(dentry);
2865         fsnotify_update_flags(dentry);
2866         fscrypt_handle_d_move(dentry);
2867 
2868         write_seqcount_end(&target->d_seq);
2869         write_seqcount_end(&dentry->d_seq);
2870 
2871         if (dir)
2872                 end_dir_add(dir, n);
2873 
2874         if (dentry->d_parent != old_parent)
2875                 spin_unlock(&dentry->d_parent->d_lock);
2876         if (dentry != old_parent)
2877                 spin_unlock(&old_parent->d_lock);
2878         spin_unlock(&target->d_lock);
2879         spin_unlock(&dentry->d_lock);
2880 }
2881 
2882 /*
2883  * d_move - move a dentry
2884  * @dentry: entry to move
2885  * @target: new dentry
2886  *
2887  * Update the dcache to reflect the move of a file name. Negative
2888  * dcache entries should not be moved in this way. See the locking
2889  * requirements for __d_move.
2890  */
2891 void d_move(struct dentry *dentry, struct dentry *target)
2892 {
2893         write_seqlock(&rename_lock);
2894         __d_move(dentry, target, false);
2895         write_sequnlock(&rename_lock);
2896 }
2897 EXPORT_SYMBOL(d_move);
2898 
2899 /*
2900  * d_exchange - exchange two dentries
2901  * @dentry1: first dentry
2902  * @dentry2: second dentry
2903  */
2904 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2905 {
2906         write_seqlock(&rename_lock);
2907 
2908         WARN_ON(!dentry1->d_inode);
2909         WARN_ON(!dentry2->d_inode);
2910         WARN_ON(IS_ROOT(dentry1));
2911         WARN_ON(IS_ROOT(dentry2));
2912 
2913         __d_move(dentry1, dentry2, true);
2914 
2915         write_sequnlock(&rename_lock);
2916 }
2917 
2918 /**
2919  * d_ancestor - search for an ancestor
2920  * @p1: ancestor dentry
2921  * @p2: child dentry
2922  *
2923  * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2924  * an ancestor of p2, else NULL.
2925  */
2926 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2927 {
2928         struct dentry *p;
2929 
2930         for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2931                 if (p->d_parent == p1)
2932                         return p;
2933         }
2934         return NULL;
2935 }
2936 
2937 /*
2938  * This helper attempts to cope with remotely renamed directories
2939  *
2940  * It assumes that the caller is already holding
2941  * dentry->d_parent->d_inode->i_mutex, and rename_lock
2942  *
2943  * Note: If ever the locking in lock_rename() changes, then please
2944  * remember to update this too...
2945  */
2946 static int __d_unalias(struct inode *inode,
2947                 struct dentry *dentry, struct dentry *alias)
2948 {
2949         struct mutex *m1 = NULL;
2950         struct rw_semaphore *m2 = NULL;
2951         int ret = -ESTALE;
2952 
2953         /* If alias and dentry share a parent, then no extra locks required */
2954         if (alias->d_parent == dentry->d_parent)
2955                 goto out_unalias;
2956 
2957         /* See lock_rename() */
2958         if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2959                 goto out_err;
2960         m1 = &dentry->d_sb->s_vfs_rename_mutex;
2961         if (!inode_trylock_shared(alias->d_parent->d_inode))
2962                 goto out_err;
2963         m2 = &alias->d_parent->d_inode->i_rwsem;
2964 out_unalias:
2965         __d_move(alias, dentry, false);
2966         ret = 0;
2967 out_err:
2968         if (m2)
2969                 up_read(m2);
2970         if (m1)
2971                 mutex_unlock(m1);
2972         return ret;
2973 }
2974 
2975 /**
2976  * d_splice_alias - splice a disconnected dentry into the tree if one exists
2977  * @inode:  the inode which may have a disconnected dentry
2978  * @dentry: a negative dentry which we want to point to the inode.
2979  *
2980  * If inode is a directory and has an IS_ROOT alias, then d_move that in
2981  * place of the given dentry and return it, else simply d_add the inode
2982  * to the dentry and return NULL.
2983  *
2984  * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2985  * we should error out: directories can't have multiple aliases.
2986  *
2987  * This is needed in the lookup routine of any filesystem that is exportable
2988  * (via knfsd) so that we can build dcache paths to directories effectively.
2989  *
2990  * If a dentry was found and moved, then it is returned.  Otherwise NULL
2991  * is returned.  This matches the expected return value of ->lookup.
2992  *
2993  * Cluster filesystems may call this function with a negative, hashed dentry.
2994  * In that case, we know that the inode will be a regular file, and also this
2995  * will only occur during atomic_open. So we need to check for the dentry
2996  * being already hashed only in the final case.
2997  */
2998 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2999 {
3000         if (IS_ERR(inode))
3001                 return ERR_CAST(inode);
3002 
3003         BUG_ON(!d_unhashed(dentry));
3004 
3005         if (!inode)
3006                 goto out;
3007 
3008         security_d_instantiate(dentry, inode);
3009         spin_lock(&inode->i_lock);
3010         if (S_ISDIR(inode->i_mode)) {
3011                 struct dentry *new = __d_find_any_alias(inode);
3012                 if (unlikely(new)) {
3013                         /* The reference to new ensures it remains an alias */
3014                         spin_unlock(&inode->i_lock);
3015                         write_seqlock(&rename_lock);
3016                         if (unlikely(d_ancestor(new, dentry))) {
3017                                 write_sequnlock(&rename_lock);
3018                                 dput(new);
3019                                 new = ERR_PTR(-ELOOP);
3020                                 pr_warn_ratelimited(
3021                                         "VFS: Lookup of '%s' in %s %s"
3022                                         " would have caused loop\n",
3023                                         dentry->d_name.name,
3024                                         inode->i_sb->s_type->name,
3025                                         inode->i_sb->s_id);
3026                         } else if (!IS_ROOT(new)) {
3027                                 struct dentry *old_parent = dget(new->d_parent);
3028                                 int err = __d_unalias(inode, dentry, new);
3029                                 write_sequnlock(&rename_lock);
3030                                 if (err) {
3031                                         dput(new);
3032                                         new = ERR_PTR(err);
3033                                 }
3034                                 dput(old_parent);
3035                         } else {
3036                                 __d_move(new, dentry, false);
3037                                 write_sequnlock(&rename_lock);
3038                         }
3039                         iput(inode);
3040                         return new;
3041                 }
3042         }
3043 out:
3044         __d_add(dentry, inode);
3045         return NULL;
3046 }
3047 EXPORT_SYMBOL(d_splice_alias);
3048 
3049 /*
3050  * Test whether new_dentry is a subdirectory of old_dentry.
3051  *
3052  * Trivially implemented using the dcache structure
3053  */
3054 
3055 /**
3056  * is_subdir - is new dentry a subdirectory of old_dentry
3057  * @new_dentry: new dentry
3058  * @old_dentry: old dentry
3059  *
3060  * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3061  * Returns false otherwise.
3062  * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3063  */
3064   
3065 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3066 {
3067         bool result;
3068         unsigned seq;
3069 
3070         if (new_dentry == old_dentry)
3071                 return true;
3072 
3073         do {
3074                 /* for restarting inner loop in case of seq retry */
3075                 seq = read_seqbegin(&rename_lock);
3076                 /*
3077                  * Need rcu_readlock to protect against the d_parent trashing
3078                  * due to d_move
3079                  */
3080                 rcu_read_lock();
3081                 if (d_ancestor(old_dentry, new_dentry))
3082                         result = true;
3083                 else
3084                         result = false;
3085                 rcu_read_unlock();
3086         } while (read_seqretry(&rename_lock, seq));
3087 
3088         return result;
3089 }
3090 EXPORT_SYMBOL(is_subdir);
3091 
3092 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3093 {
3094         struct dentry *root = data;
3095         if (dentry != root) {
3096                 if (d_unhashed(dentry) || !dentry->d_inode)
3097                         return D_WALK_SKIP;
3098 
3099                 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3100                         dentry->d_flags |= DCACHE_GENOCIDE;
3101                         dentry->d_lockref.count--;
3102                 }
3103         }
3104         return D_WALK_CONTINUE;
3105 }
3106 
3107 void d_genocide(struct dentry *parent)
3108 {
3109         d_walk(parent, parent, d_genocide_kill);
3110 }
3111 
3112 EXPORT_SYMBOL(d_genocide);
3113 
3114 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3115 {
3116         inode_dec_link_count(inode);
3117         BUG_ON(dentry->d_name.name != dentry->d_iname ||
3118                 !hlist_unhashed(&dentry->d_u.d_alias) ||
3119                 !d_unlinked(dentry));
3120         spin_lock(&dentry->d_parent->d_lock);
3121         spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3122         dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3123                                 (unsigned long long)inode->i_ino);
3124         spin_unlock(&dentry->d_lock);
3125         spin_unlock(&dentry->d_parent->d_lock);
3126         d_instantiate(dentry, inode);
3127 }
3128 EXPORT_SYMBOL(d_tmpfile);
3129 
3130 static __initdata unsigned long dhash_entries;
3131 static int __init set_dhash_entries(char *str)
3132 {
3133         if (!str)
3134                 return 0;
3135         dhash_entries = simple_strtoul(str, &str, 0);
3136         return 1;
3137 }
3138 __setup("dhash_entries=", set_dhash_entries);
3139 
3140 static void __init dcache_init_early(void)
3141 {
3142         /* If hashes are distributed across NUMA nodes, defer
3143          * hash allocation until vmalloc space is available.
3144          */
3145         if (hashdist)
3146                 return;
3147 
3148         dentry_hashtable =
3149                 alloc_large_system_hash("Dentry cache",
3150                                         sizeof(struct hlist_bl_head),
3151                                         dhash_entries,
3152                                         13,
3153                                         HASH_EARLY | HASH_ZERO,
3154                                         &d_hash_shift,
3155                                         NULL,
3156                                         0,
3157                                         0);
3158         d_hash_shift = 32 - d_hash_shift;
3159 }
3160 
3161 static void __init dcache_init(void)
3162 {
3163         /*
3164          * A constructor could be added for stable state like the lists,
3165          * but it is probably not worth it because of the cache nature
3166          * of the dcache.
3167          */
3168         dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3169                 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT,
3170                 d_iname);
3171 
3172         /* Hash may have been set up in dcache_init_early */
3173         if (!hashdist)
3174                 return;
3175 
3176         dentry_hashtable =
3177                 alloc_large_system_hash("Dentry cache",
3178                                         sizeof(struct hlist_bl_head),
3179                                         dhash_entries,
3180                                         13,
3181                                         HASH_ZERO,
3182                                         &d_hash_shift,
3183                                         NULL,
3184                                         0,
3185                                         0);
3186         d_hash_shift = 32 - d_hash_shift;
3187 }
3188 
3189 /* SLAB cache for __getname() consumers */
3190 struct kmem_cache *names_cachep __read_mostly;
3191 EXPORT_SYMBOL(names_cachep);
3192 
3193 void __init vfs_caches_init_early(void)
3194 {
3195         int i;
3196 
3197         for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3198                 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3199 
3200         dcache_init_early();
3201         inode_init_early();
3202 }
3203 
3204 void __init vfs_caches_init(void)
3205 {
3206         names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3207                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
3208 
3209         dcache_init();
3210         inode_init();
3211         files_init();
3212         files_maxfiles_init();
3213         mnt_init();
3214         bdev_cache_init();
3215         chrdev_init();
3216 }
3217 

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

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

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

osdn.jp