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

TOMOYO Linux Cross Reference
Linux/fs/dcache.c

Version: ~ [ linux-5.1-rc2 ] ~ [ linux-5.0.4 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.31 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.108 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.165 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.177 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.137 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.63 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * fs/dcache.c
  3  *
  4  * Complete reimplementation
  5  * (C) 1997 Thomas Schoebel-Theuer,
  6  * with heavy changes by Linus Torvalds
  7  */
  8 
  9 /*
 10  * Notes on the allocation strategy:
 11  *
 12  * The dcache is a master of the icache - whenever a dcache entry
 13  * exists, the inode will always exist. "iput()" is done either when
 14  * the dcache entry is deleted or garbage collected.
 15  */
 16 
 17 #include <linux/syscalls.h>
 18 #include <linux/string.h>
 19 #include <linux/mm.h>
 20 #include <linux/fs.h>
 21 #include <linux/fsnotify.h>
 22 #include <linux/slab.h>
 23 #include <linux/init.h>
 24 #include <linux/hash.h>
 25 #include <linux/cache.h>
 26 #include <linux/export.h>
 27 #include <linux/mount.h>
 28 #include <linux/file.h>
 29 #include <linux/uaccess.h>
 30 #include <linux/security.h>
 31 #include <linux/seqlock.h>
 32 #include <linux/swap.h>
 33 #include <linux/bootmem.h>
 34 #include <linux/fs_struct.h>
 35 #include <linux/bit_spinlock.h>
 36 #include <linux/rculist_bl.h>
 37 #include <linux/prefetch.h>
 38 #include <linux/ratelimit.h>
 39 #include <linux/list_lru.h>
 40 #include "internal.h"
 41 #include "mount.h"
 42 
 43 /*
 44  * Usage:
 45  * dcache->d_inode->i_lock protects:
 46  *   - i_dentry, d_u.d_alias, d_inode of aliases
 47  * dcache_hash_bucket lock protects:
 48  *   - the dcache hash table
 49  * s_roots bl list spinlock protects:
 50  *   - the s_roots list (see __d_drop)
 51  * dentry->d_sb->s_dentry_lru_lock protects:
 52  *   - the dcache lru lists and counters
 53  * d_lock protects:
 54  *   - d_flags
 55  *   - d_name
 56  *   - d_lru
 57  *   - d_count
 58  *   - d_unhashed()
 59  *   - d_parent and d_subdirs
 60  *   - childrens' d_child and d_parent
 61  *   - d_u.d_alias, d_inode
 62  *
 63  * Ordering:
 64  * dentry->d_inode->i_lock
 65  *   dentry->d_lock
 66  *     dentry->d_sb->s_dentry_lru_lock
 67  *     dcache_hash_bucket lock
 68  *     s_roots lock
 69  *
 70  * If there is an ancestor relationship:
 71  * dentry->d_parent->...->d_parent->d_lock
 72  *   ...
 73  *     dentry->d_parent->d_lock
 74  *       dentry->d_lock
 75  *
 76  * If no ancestor relationship:
 77  * if (dentry1 < dentry2)
 78  *   dentry1->d_lock
 79  *     dentry2->d_lock
 80  */
 81 int sysctl_vfs_cache_pressure __read_mostly = 100;
 82 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
 83 
 84 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
 85 
 86 EXPORT_SYMBOL(rename_lock);
 87 
 88 static struct kmem_cache *dentry_cache __read_mostly;
 89 
 90 const struct qstr empty_name = QSTR_INIT("", 0);
 91 EXPORT_SYMBOL(empty_name);
 92 const struct qstr slash_name = QSTR_INIT("/", 1);
 93 EXPORT_SYMBOL(slash_name);
 94 
 95 /*
 96  * This is the single most critical data structure when it comes
 97  * to the dcache: the hashtable for lookups. Somebody should try
 98  * to make this good - I've just made it work.
 99  *
100  * This hash-function tries to avoid losing too many bits of hash
101  * information, yet avoid using a prime hash-size or similar.
102  */
103 
104 static unsigned int d_hash_shift __read_mostly;
105 
106 static struct hlist_bl_head *dentry_hashtable __read_mostly;
107 
108 static inline struct hlist_bl_head *d_hash(unsigned int hash)
109 {
110         return dentry_hashtable + (hash >> d_hash_shift);
111 }
112 
113 #define IN_LOOKUP_SHIFT 10
114 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
115 
116 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
117                                         unsigned int hash)
118 {
119         hash += (unsigned long) parent / L1_CACHE_BYTES;
120         return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
121 }
122 
123 
124 /* Statistics gathering. */
125 struct dentry_stat_t dentry_stat = {
126         .age_limit = 45,
127 };
128 
129 static DEFINE_PER_CPU(long, nr_dentry);
130 static DEFINE_PER_CPU(long, nr_dentry_unused);
131 
132 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
133 
134 /*
135  * Here we resort to our own counters instead of using generic per-cpu counters
136  * for consistency with what the vfs inode code does. We are expected to harvest
137  * better code and performance by having our own specialized counters.
138  *
139  * Please note that the loop is done over all possible CPUs, not over all online
140  * CPUs. The reason for this is that we don't want to play games with CPUs going
141  * on and off. If one of them goes off, we will just keep their counters.
142  *
143  * glommer: See cffbc8a for details, and if you ever intend to change this,
144  * please update all vfs counters to match.
145  */
146 static long get_nr_dentry(void)
147 {
148         int i;
149         long sum = 0;
150         for_each_possible_cpu(i)
151                 sum += per_cpu(nr_dentry, i);
152         return sum < 0 ? 0 : sum;
153 }
154 
155 static long get_nr_dentry_unused(void)
156 {
157         int i;
158         long sum = 0;
159         for_each_possible_cpu(i)
160                 sum += per_cpu(nr_dentry_unused, i);
161         return sum < 0 ? 0 : sum;
162 }
163 
164 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
165                    size_t *lenp, loff_t *ppos)
166 {
167         dentry_stat.nr_dentry = get_nr_dentry();
168         dentry_stat.nr_unused = get_nr_dentry_unused();
169         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
170 }
171 #endif
172 
173 /*
174  * Compare 2 name strings, return 0 if they match, otherwise non-zero.
175  * The strings are both count bytes long, and count is non-zero.
176  */
177 #ifdef CONFIG_DCACHE_WORD_ACCESS
178 
179 #include <asm/word-at-a-time.h>
180 /*
181  * NOTE! 'cs' and 'scount' come from a dentry, so it has a
182  * aligned allocation for this particular component. We don't
183  * strictly need the load_unaligned_zeropad() safety, but it
184  * doesn't hurt either.
185  *
186  * In contrast, 'ct' and 'tcount' can be from a pathname, and do
187  * need the careful unaligned handling.
188  */
189 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
190 {
191         unsigned long a,b,mask;
192 
193         for (;;) {
194                 a = read_word_at_a_time(cs);
195                 b = load_unaligned_zeropad(ct);
196                 if (tcount < sizeof(unsigned long))
197                         break;
198                 if (unlikely(a != b))
199                         return 1;
200                 cs += sizeof(unsigned long);
201                 ct += sizeof(unsigned long);
202                 tcount -= sizeof(unsigned long);
203                 if (!tcount)
204                         return 0;
205         }
206         mask = bytemask_from_count(tcount);
207         return unlikely(!!((a ^ b) & mask));
208 }
209 
210 #else
211 
212 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
213 {
214         do {
215                 if (*cs != *ct)
216                         return 1;
217                 cs++;
218                 ct++;
219                 tcount--;
220         } while (tcount);
221         return 0;
222 }
223 
224 #endif
225 
226 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
227 {
228         /*
229          * Be careful about RCU walk racing with rename:
230          * use 'READ_ONCE' to fetch the name pointer.
231          *
232          * NOTE! Even if a rename will mean that the length
233          * was not loaded atomically, we don't care. The
234          * RCU walk will check the sequence count eventually,
235          * and catch it. And we won't overrun the buffer,
236          * because we're reading the name pointer atomically,
237          * and a dentry name is guaranteed to be properly
238          * terminated with a NUL byte.
239          *
240          * End result: even if 'len' is wrong, we'll exit
241          * early because the data cannot match (there can
242          * be no NUL in the ct/tcount data)
243          */
244         const unsigned char *cs = READ_ONCE(dentry->d_name.name);
245 
246         return dentry_string_cmp(cs, ct, tcount);
247 }
248 
249 struct external_name {
250         union {
251                 atomic_t count;
252                 struct rcu_head head;
253         } u;
254         unsigned char name[];
255 };
256 
257 static inline struct external_name *external_name(struct dentry *dentry)
258 {
259         return container_of(dentry->d_name.name, struct external_name, name[0]);
260 }
261 
262 static void __d_free(struct rcu_head *head)
263 {
264         struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
265 
266         kmem_cache_free(dentry_cache, dentry); 
267 }
268 
269 static void __d_free_external(struct rcu_head *head)
270 {
271         struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
272         kfree(external_name(dentry));
273         kmem_cache_free(dentry_cache, dentry); 
274 }
275 
276 static inline int dname_external(const struct dentry *dentry)
277 {
278         return dentry->d_name.name != dentry->d_iname;
279 }
280 
281 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
282 {
283         spin_lock(&dentry->d_lock);
284         if (unlikely(dname_external(dentry))) {
285                 struct external_name *p = external_name(dentry);
286                 atomic_inc(&p->u.count);
287                 spin_unlock(&dentry->d_lock);
288                 name->name = p->name;
289         } else {
290                 memcpy(name->inline_name, dentry->d_iname, DNAME_INLINE_LEN);
291                 spin_unlock(&dentry->d_lock);
292                 name->name = name->inline_name;
293         }
294 }
295 EXPORT_SYMBOL(take_dentry_name_snapshot);
296 
297 void release_dentry_name_snapshot(struct name_snapshot *name)
298 {
299         if (unlikely(name->name != name->inline_name)) {
300                 struct external_name *p;
301                 p = container_of(name->name, struct external_name, name[0]);
302                 if (unlikely(atomic_dec_and_test(&p->u.count)))
303                         kfree_rcu(p, u.head);
304         }
305 }
306 EXPORT_SYMBOL(release_dentry_name_snapshot);
307 
308 static inline void __d_set_inode_and_type(struct dentry *dentry,
309                                           struct inode *inode,
310                                           unsigned type_flags)
311 {
312         unsigned flags;
313 
314         dentry->d_inode = inode;
315         flags = READ_ONCE(dentry->d_flags);
316         flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
317         flags |= type_flags;
318         WRITE_ONCE(dentry->d_flags, flags);
319 }
320 
321 static inline void __d_clear_type_and_inode(struct dentry *dentry)
322 {
323         unsigned flags = READ_ONCE(dentry->d_flags);
324 
325         flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
326         WRITE_ONCE(dentry->d_flags, flags);
327         dentry->d_inode = NULL;
328 }
329 
330 static void dentry_free(struct dentry *dentry)
331 {
332         WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
333         if (unlikely(dname_external(dentry))) {
334                 struct external_name *p = external_name(dentry);
335                 if (likely(atomic_dec_and_test(&p->u.count))) {
336                         call_rcu(&dentry->d_u.d_rcu, __d_free_external);
337                         return;
338                 }
339         }
340         /* if dentry was never visible to RCU, immediate free is OK */
341         if (!(dentry->d_flags & DCACHE_RCUACCESS))
342                 __d_free(&dentry->d_u.d_rcu);
343         else
344                 call_rcu(&dentry->d_u.d_rcu, __d_free);
345 }
346 
347 /*
348  * Release the dentry's inode, using the filesystem
349  * d_iput() operation if defined.
350  */
351 static void dentry_unlink_inode(struct dentry * dentry)
352         __releases(dentry->d_lock)
353         __releases(dentry->d_inode->i_lock)
354 {
355         struct inode *inode = dentry->d_inode;
356         bool hashed = !d_unhashed(dentry);
357 
358         if (hashed)
359                 raw_write_seqcount_begin(&dentry->d_seq);
360         __d_clear_type_and_inode(dentry);
361         hlist_del_init(&dentry->d_u.d_alias);
362         if (hashed)
363                 raw_write_seqcount_end(&dentry->d_seq);
364         spin_unlock(&dentry->d_lock);
365         spin_unlock(&inode->i_lock);
366         if (!inode->i_nlink)
367                 fsnotify_inoderemove(inode);
368         if (dentry->d_op && dentry->d_op->d_iput)
369                 dentry->d_op->d_iput(dentry, inode);
370         else
371                 iput(inode);
372 }
373 
374 /*
375  * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
376  * is in use - which includes both the "real" per-superblock
377  * LRU list _and_ the DCACHE_SHRINK_LIST use.
378  *
379  * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
380  * on the shrink list (ie not on the superblock LRU list).
381  *
382  * The per-cpu "nr_dentry_unused" counters are updated with
383  * the DCACHE_LRU_LIST bit.
384  *
385  * These helper functions make sure we always follow the
386  * rules. d_lock must be held by the caller.
387  */
388 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
389 static void d_lru_add(struct dentry *dentry)
390 {
391         D_FLAG_VERIFY(dentry, 0);
392         dentry->d_flags |= DCACHE_LRU_LIST;
393         this_cpu_inc(nr_dentry_unused);
394         WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
395 }
396 
397 static void d_lru_del(struct dentry *dentry)
398 {
399         D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
400         dentry->d_flags &= ~DCACHE_LRU_LIST;
401         this_cpu_dec(nr_dentry_unused);
402         WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
403 }
404 
405 static void d_shrink_del(struct dentry *dentry)
406 {
407         D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
408         list_del_init(&dentry->d_lru);
409         dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
410         this_cpu_dec(nr_dentry_unused);
411 }
412 
413 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
414 {
415         D_FLAG_VERIFY(dentry, 0);
416         list_add(&dentry->d_lru, list);
417         dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
418         this_cpu_inc(nr_dentry_unused);
419 }
420 
421 /*
422  * These can only be called under the global LRU lock, ie during the
423  * callback for freeing the LRU list. "isolate" removes it from the
424  * LRU lists entirely, while shrink_move moves it to the indicated
425  * private list.
426  */
427 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
428 {
429         D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
430         dentry->d_flags &= ~DCACHE_LRU_LIST;
431         this_cpu_dec(nr_dentry_unused);
432         list_lru_isolate(lru, &dentry->d_lru);
433 }
434 
435 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
436                               struct list_head *list)
437 {
438         D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
439         dentry->d_flags |= DCACHE_SHRINK_LIST;
440         list_lru_isolate_move(lru, &dentry->d_lru, list);
441 }
442 
443 /*
444  * dentry_lru_(add|del)_list) must be called with d_lock held.
445  */
446 static void dentry_lru_add(struct dentry *dentry)
447 {
448         if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
449                 d_lru_add(dentry);
450         else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
451                 dentry->d_flags |= DCACHE_REFERENCED;
452 }
453 
454 /**
455  * d_drop - drop a dentry
456  * @dentry: dentry to drop
457  *
458  * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
459  * be found through a VFS lookup any more. Note that this is different from
460  * deleting the dentry - d_delete will try to mark the dentry negative if
461  * possible, giving a successful _negative_ lookup, while d_drop will
462  * just make the cache lookup fail.
463  *
464  * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
465  * reason (NFS timeouts or autofs deletes).
466  *
467  * __d_drop requires dentry->d_lock
468  * ___d_drop doesn't mark dentry as "unhashed"
469  *   (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
470  */
471 static void ___d_drop(struct dentry *dentry)
472 {
473         if (!d_unhashed(dentry)) {
474                 struct hlist_bl_head *b;
475                 /*
476                  * Hashed dentries are normally on the dentry hashtable,
477                  * with the exception of those newly allocated by
478                  * d_obtain_root, which are always IS_ROOT:
479                  */
480                 if (unlikely(IS_ROOT(dentry)))
481                         b = &dentry->d_sb->s_roots;
482                 else
483                         b = d_hash(dentry->d_name.hash);
484 
485                 hlist_bl_lock(b);
486                 __hlist_bl_del(&dentry->d_hash);
487                 hlist_bl_unlock(b);
488                 /* After this call, in-progress rcu-walk path lookup will fail. */
489                 write_seqcount_invalidate(&dentry->d_seq);
490         }
491 }
492 
493 void __d_drop(struct dentry *dentry)
494 {
495         ___d_drop(dentry);
496         dentry->d_hash.pprev = NULL;
497 }
498 EXPORT_SYMBOL(__d_drop);
499 
500 void d_drop(struct dentry *dentry)
501 {
502         spin_lock(&dentry->d_lock);
503         __d_drop(dentry);
504         spin_unlock(&dentry->d_lock);
505 }
506 EXPORT_SYMBOL(d_drop);
507 
508 static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
509 {
510         struct dentry *next;
511         /*
512          * Inform d_walk() and shrink_dentry_list() that we are no longer
513          * attached to the dentry tree
514          */
515         dentry->d_flags |= DCACHE_DENTRY_KILLED;
516         if (unlikely(list_empty(&dentry->d_child)))
517                 return;
518         __list_del_entry(&dentry->d_child);
519         /*
520          * Cursors can move around the list of children.  While we'd been
521          * a normal list member, it didn't matter - ->d_child.next would've
522          * been updated.  However, from now on it won't be and for the
523          * things like d_walk() it might end up with a nasty surprise.
524          * Normally d_walk() doesn't care about cursors moving around -
525          * ->d_lock on parent prevents that and since a cursor has no children
526          * of its own, we get through it without ever unlocking the parent.
527          * There is one exception, though - if we ascend from a child that
528          * gets killed as soon as we unlock it, the next sibling is found
529          * using the value left in its ->d_child.next.  And if _that_
530          * pointed to a cursor, and cursor got moved (e.g. by lseek())
531          * before d_walk() regains parent->d_lock, we'll end up skipping
532          * everything the cursor had been moved past.
533          *
534          * Solution: make sure that the pointer left behind in ->d_child.next
535          * points to something that won't be moving around.  I.e. skip the
536          * cursors.
537          */
538         while (dentry->d_child.next != &parent->d_subdirs) {
539                 next = list_entry(dentry->d_child.next, struct dentry, d_child);
540                 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
541                         break;
542                 dentry->d_child.next = next->d_child.next;
543         }
544 }
545 
546 static void __dentry_kill(struct dentry *dentry)
547 {
548         struct dentry *parent = NULL;
549         bool can_free = true;
550         if (!IS_ROOT(dentry))
551                 parent = dentry->d_parent;
552 
553         /*
554          * The dentry is now unrecoverably dead to the world.
555          */
556         lockref_mark_dead(&dentry->d_lockref);
557 
558         /*
559          * inform the fs via d_prune that this dentry is about to be
560          * unhashed and destroyed.
561          */
562         if (dentry->d_flags & DCACHE_OP_PRUNE)
563                 dentry->d_op->d_prune(dentry);
564 
565         if (dentry->d_flags & DCACHE_LRU_LIST) {
566                 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
567                         d_lru_del(dentry);
568         }
569         /* if it was on the hash then remove it */
570         __d_drop(dentry);
571         dentry_unlist(dentry, parent);
572         if (parent)
573                 spin_unlock(&parent->d_lock);
574         if (dentry->d_inode)
575                 dentry_unlink_inode(dentry);
576         else
577                 spin_unlock(&dentry->d_lock);
578         this_cpu_dec(nr_dentry);
579         if (dentry->d_op && dentry->d_op->d_release)
580                 dentry->d_op->d_release(dentry);
581 
582         spin_lock(&dentry->d_lock);
583         if (dentry->d_flags & DCACHE_SHRINK_LIST) {
584                 dentry->d_flags |= DCACHE_MAY_FREE;
585                 can_free = false;
586         }
587         spin_unlock(&dentry->d_lock);
588         if (likely(can_free))
589                 dentry_free(dentry);
590 }
591 
592 /*
593  * Finish off a dentry we've decided to kill.
594  * dentry->d_lock must be held, returns with it unlocked.
595  * If ref is non-zero, then decrement the refcount too.
596  * Returns dentry requiring refcount drop, or NULL if we're done.
597  */
598 static struct dentry *dentry_kill(struct dentry *dentry)
599         __releases(dentry->d_lock)
600 {
601         struct inode *inode = dentry->d_inode;
602         struct dentry *parent = NULL;
603 
604         if (inode && unlikely(!spin_trylock(&inode->i_lock)))
605                 goto failed;
606 
607         if (!IS_ROOT(dentry)) {
608                 parent = dentry->d_parent;
609                 if (unlikely(!spin_trylock(&parent->d_lock))) {
610                         if (inode)
611                                 spin_unlock(&inode->i_lock);
612                         goto failed;
613                 }
614         }
615 
616         __dentry_kill(dentry);
617         return parent;
618 
619 failed:
620         spin_unlock(&dentry->d_lock);
621         return dentry; /* try again with same dentry */
622 }
623 
624 static inline struct dentry *lock_parent(struct dentry *dentry)
625 {
626         struct dentry *parent = dentry->d_parent;
627         if (IS_ROOT(dentry))
628                 return NULL;
629         if (unlikely(dentry->d_lockref.count < 0))
630                 return NULL;
631         if (likely(spin_trylock(&parent->d_lock)))
632                 return parent;
633         rcu_read_lock();
634         spin_unlock(&dentry->d_lock);
635 again:
636         parent = READ_ONCE(dentry->d_parent);
637         spin_lock(&parent->d_lock);
638         /*
639          * We can't blindly lock dentry until we are sure
640          * that we won't violate the locking order.
641          * Any changes of dentry->d_parent must have
642          * been done with parent->d_lock held, so
643          * spin_lock() above is enough of a barrier
644          * for checking if it's still our child.
645          */
646         if (unlikely(parent != dentry->d_parent)) {
647                 spin_unlock(&parent->d_lock);
648                 goto again;
649         }
650         if (parent != dentry) {
651                 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
652                 if (unlikely(dentry->d_lockref.count < 0)) {
653                         spin_unlock(&parent->d_lock);
654                         parent = NULL;
655                 }
656         } else {
657                 parent = NULL;
658         }
659         rcu_read_unlock();
660         return parent;
661 }
662 
663 /*
664  * Try to do a lockless dput(), and return whether that was successful.
665  *
666  * If unsuccessful, we return false, having already taken the dentry lock.
667  *
668  * The caller needs to hold the RCU read lock, so that the dentry is
669  * guaranteed to stay around even if the refcount goes down to zero!
670  */
671 static inline bool fast_dput(struct dentry *dentry)
672 {
673         int ret;
674         unsigned int d_flags;
675 
676         /*
677          * If we have a d_op->d_delete() operation, we sould not
678          * let the dentry count go to zero, so use "put_or_lock".
679          */
680         if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
681                 return lockref_put_or_lock(&dentry->d_lockref);
682 
683         /*
684          * .. otherwise, we can try to just decrement the
685          * lockref optimistically.
686          */
687         ret = lockref_put_return(&dentry->d_lockref);
688 
689         /*
690          * If the lockref_put_return() failed due to the lock being held
691          * by somebody else, the fast path has failed. We will need to
692          * get the lock, and then check the count again.
693          */
694         if (unlikely(ret < 0)) {
695                 spin_lock(&dentry->d_lock);
696                 if (dentry->d_lockref.count > 1) {
697                         dentry->d_lockref.count--;
698                         spin_unlock(&dentry->d_lock);
699                         return 1;
700                 }
701                 return 0;
702         }
703 
704         /*
705          * If we weren't the last ref, we're done.
706          */
707         if (ret)
708                 return 1;
709 
710         /*
711          * Careful, careful. The reference count went down
712          * to zero, but we don't hold the dentry lock, so
713          * somebody else could get it again, and do another
714          * dput(), and we need to not race with that.
715          *
716          * However, there is a very special and common case
717          * where we don't care, because there is nothing to
718          * do: the dentry is still hashed, it does not have
719          * a 'delete' op, and it's referenced and already on
720          * the LRU list.
721          *
722          * NOTE! Since we aren't locked, these values are
723          * not "stable". However, it is sufficient that at
724          * some point after we dropped the reference the
725          * dentry was hashed and the flags had the proper
726          * value. Other dentry users may have re-gotten
727          * a reference to the dentry and change that, but
728          * our work is done - we can leave the dentry
729          * around with a zero refcount.
730          */
731         smp_rmb();
732         d_flags = READ_ONCE(dentry->d_flags);
733         d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
734 
735         /* Nothing to do? Dropping the reference was all we needed? */
736         if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
737                 return 1;
738 
739         /*
740          * Not the fast normal case? Get the lock. We've already decremented
741          * the refcount, but we'll need to re-check the situation after
742          * getting the lock.
743          */
744         spin_lock(&dentry->d_lock);
745 
746         /*
747          * Did somebody else grab a reference to it in the meantime, and
748          * we're no longer the last user after all? Alternatively, somebody
749          * else could have killed it and marked it dead. Either way, we
750          * don't need to do anything else.
751          */
752         if (dentry->d_lockref.count) {
753                 spin_unlock(&dentry->d_lock);
754                 return 1;
755         }
756 
757         /*
758          * Re-get the reference we optimistically dropped. We hold the
759          * lock, and we just tested that it was zero, so we can just
760          * set it to 1.
761          */
762         dentry->d_lockref.count = 1;
763         return 0;
764 }
765 
766 
767 /* 
768  * This is dput
769  *
770  * This is complicated by the fact that we do not want to put
771  * dentries that are no longer on any hash chain on the unused
772  * list: we'd much rather just get rid of them immediately.
773  *
774  * However, that implies that we have to traverse the dentry
775  * tree upwards to the parents which might _also_ now be
776  * scheduled for deletion (it may have been only waiting for
777  * its last child to go away).
778  *
779  * This tail recursion is done by hand as we don't want to depend
780  * on the compiler to always get this right (gcc generally doesn't).
781  * Real recursion would eat up our stack space.
782  */
783 
784 /*
785  * dput - release a dentry
786  * @dentry: dentry to release 
787  *
788  * Release a dentry. This will drop the usage count and if appropriate
789  * call the dentry unlink method as well as removing it from the queues and
790  * releasing its resources. If the parent dentries were scheduled for release
791  * they too may now get deleted.
792  */
793 void dput(struct dentry *dentry)
794 {
795         if (unlikely(!dentry))
796                 return;
797 
798 repeat:
799         might_sleep();
800 
801         rcu_read_lock();
802         if (likely(fast_dput(dentry))) {
803                 rcu_read_unlock();
804                 return;
805         }
806 
807         /* Slow case: now with the dentry lock held */
808         rcu_read_unlock();
809 
810         WARN_ON(d_in_lookup(dentry));
811 
812         /* Unreachable? Get rid of it */
813         if (unlikely(d_unhashed(dentry)))
814                 goto kill_it;
815 
816         if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
817                 goto kill_it;
818 
819         if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
820                 if (dentry->d_op->d_delete(dentry))
821                         goto kill_it;
822         }
823 
824         dentry_lru_add(dentry);
825 
826         dentry->d_lockref.count--;
827         spin_unlock(&dentry->d_lock);
828         return;
829 
830 kill_it:
831         dentry = dentry_kill(dentry);
832         if (dentry) {
833                 cond_resched();
834                 goto repeat;
835         }
836 }
837 EXPORT_SYMBOL(dput);
838 
839 
840 /* This must be called with d_lock held */
841 static inline void __dget_dlock(struct dentry *dentry)
842 {
843         dentry->d_lockref.count++;
844 }
845 
846 static inline void __dget(struct dentry *dentry)
847 {
848         lockref_get(&dentry->d_lockref);
849 }
850 
851 struct dentry *dget_parent(struct dentry *dentry)
852 {
853         int gotref;
854         struct dentry *ret;
855 
856         /*
857          * Do optimistic parent lookup without any
858          * locking.
859          */
860         rcu_read_lock();
861         ret = READ_ONCE(dentry->d_parent);
862         gotref = lockref_get_not_zero(&ret->d_lockref);
863         rcu_read_unlock();
864         if (likely(gotref)) {
865                 if (likely(ret == READ_ONCE(dentry->d_parent)))
866                         return ret;
867                 dput(ret);
868         }
869 
870 repeat:
871         /*
872          * Don't need rcu_dereference because we re-check it was correct under
873          * the lock.
874          */
875         rcu_read_lock();
876         ret = dentry->d_parent;
877         spin_lock(&ret->d_lock);
878         if (unlikely(ret != dentry->d_parent)) {
879                 spin_unlock(&ret->d_lock);
880                 rcu_read_unlock();
881                 goto repeat;
882         }
883         rcu_read_unlock();
884         BUG_ON(!ret->d_lockref.count);
885         ret->d_lockref.count++;
886         spin_unlock(&ret->d_lock);
887         return ret;
888 }
889 EXPORT_SYMBOL(dget_parent);
890 
891 /**
892  * d_find_alias - grab a hashed alias of inode
893  * @inode: inode in question
894  *
895  * If inode has a hashed alias, or is a directory and has any alias,
896  * acquire the reference to alias and return it. Otherwise return NULL.
897  * Notice that if inode is a directory there can be only one alias and
898  * it can be unhashed only if it has no children, or if it is the root
899  * of a filesystem, or if the directory was renamed and d_revalidate
900  * was the first vfs operation to notice.
901  *
902  * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
903  * any other hashed alias over that one.
904  */
905 static struct dentry *__d_find_alias(struct inode *inode)
906 {
907         struct dentry *alias, *discon_alias;
908 
909 again:
910         discon_alias = NULL;
911         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
912                 spin_lock(&alias->d_lock);
913                 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
914                         if (IS_ROOT(alias) &&
915                             (alias->d_flags & DCACHE_DISCONNECTED)) {
916                                 discon_alias = alias;
917                         } else {
918                                 __dget_dlock(alias);
919                                 spin_unlock(&alias->d_lock);
920                                 return alias;
921                         }
922                 }
923                 spin_unlock(&alias->d_lock);
924         }
925         if (discon_alias) {
926                 alias = discon_alias;
927                 spin_lock(&alias->d_lock);
928                 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
929                         __dget_dlock(alias);
930                         spin_unlock(&alias->d_lock);
931                         return alias;
932                 }
933                 spin_unlock(&alias->d_lock);
934                 goto again;
935         }
936         return NULL;
937 }
938 
939 struct dentry *d_find_alias(struct inode *inode)
940 {
941         struct dentry *de = NULL;
942 
943         if (!hlist_empty(&inode->i_dentry)) {
944                 spin_lock(&inode->i_lock);
945                 de = __d_find_alias(inode);
946                 spin_unlock(&inode->i_lock);
947         }
948         return de;
949 }
950 EXPORT_SYMBOL(d_find_alias);
951 
952 /*
953  *      Try to kill dentries associated with this inode.
954  * WARNING: you must own a reference to inode.
955  */
956 void d_prune_aliases(struct inode *inode)
957 {
958         struct dentry *dentry;
959 restart:
960         spin_lock(&inode->i_lock);
961         hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
962                 spin_lock(&dentry->d_lock);
963                 if (!dentry->d_lockref.count) {
964                         struct dentry *parent = lock_parent(dentry);
965                         if (likely(!dentry->d_lockref.count)) {
966                                 __dentry_kill(dentry);
967                                 dput(parent);
968                                 goto restart;
969                         }
970                         if (parent)
971                                 spin_unlock(&parent->d_lock);
972                 }
973                 spin_unlock(&dentry->d_lock);
974         }
975         spin_unlock(&inode->i_lock);
976 }
977 EXPORT_SYMBOL(d_prune_aliases);
978 
979 static void shrink_dentry_list(struct list_head *list)
980 {
981         struct dentry *dentry, *parent;
982 
983         while (!list_empty(list)) {
984                 struct inode *inode;
985                 dentry = list_entry(list->prev, struct dentry, d_lru);
986                 spin_lock(&dentry->d_lock);
987                 parent = lock_parent(dentry);
988 
989                 /*
990                  * The dispose list is isolated and dentries are not accounted
991                  * to the LRU here, so we can simply remove it from the list
992                  * here regardless of whether it is referenced or not.
993                  */
994                 d_shrink_del(dentry);
995 
996                 /*
997                  * We found an inuse dentry which was not removed from
998                  * the LRU because of laziness during lookup. Do not free it.
999                  */
1000                 if (dentry->d_lockref.count > 0) {
1001                         spin_unlock(&dentry->d_lock);
1002                         if (parent)
1003                                 spin_unlock(&parent->d_lock);
1004                         continue;
1005                 }
1006 
1007 
1008                 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
1009                         bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
1010                         spin_unlock(&dentry->d_lock);
1011                         if (parent)
1012                                 spin_unlock(&parent->d_lock);
1013                         if (can_free)
1014                                 dentry_free(dentry);
1015                         continue;
1016                 }
1017 
1018                 inode = dentry->d_inode;
1019                 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1020                         d_shrink_add(dentry, list);
1021                         spin_unlock(&dentry->d_lock);
1022                         if (parent)
1023                                 spin_unlock(&parent->d_lock);
1024                         continue;
1025                 }
1026 
1027                 __dentry_kill(dentry);
1028 
1029                 /*
1030                  * We need to prune ancestors too. This is necessary to prevent
1031                  * quadratic behavior of shrink_dcache_parent(), but is also
1032                  * expected to be beneficial in reducing dentry cache
1033                  * fragmentation.
1034                  */
1035                 dentry = parent;
1036                 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1037                         parent = lock_parent(dentry);
1038                         if (dentry->d_lockref.count != 1) {
1039                                 dentry->d_lockref.count--;
1040                                 spin_unlock(&dentry->d_lock);
1041                                 if (parent)
1042                                         spin_unlock(&parent->d_lock);
1043                                 break;
1044                         }
1045                         inode = dentry->d_inode;        /* can't be NULL */
1046                         if (unlikely(!spin_trylock(&inode->i_lock))) {
1047                                 spin_unlock(&dentry->d_lock);
1048                                 if (parent)
1049                                         spin_unlock(&parent->d_lock);
1050                                 cpu_relax();
1051                                 continue;
1052                         }
1053                         __dentry_kill(dentry);
1054                         dentry = parent;
1055                 }
1056         }
1057 }
1058 
1059 static enum lru_status dentry_lru_isolate(struct list_head *item,
1060                 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1061 {
1062         struct list_head *freeable = arg;
1063         struct dentry   *dentry = container_of(item, struct dentry, d_lru);
1064 
1065 
1066         /*
1067          * we are inverting the lru lock/dentry->d_lock here,
1068          * so use a trylock. If we fail to get the lock, just skip
1069          * it
1070          */
1071         if (!spin_trylock(&dentry->d_lock))
1072                 return LRU_SKIP;
1073 
1074         /*
1075          * Referenced dentries are still in use. If they have active
1076          * counts, just remove them from the LRU. Otherwise give them
1077          * another pass through the LRU.
1078          */
1079         if (dentry->d_lockref.count) {
1080                 d_lru_isolate(lru, dentry);
1081                 spin_unlock(&dentry->d_lock);
1082                 return LRU_REMOVED;
1083         }
1084 
1085         if (dentry->d_flags & DCACHE_REFERENCED) {
1086                 dentry->d_flags &= ~DCACHE_REFERENCED;
1087                 spin_unlock(&dentry->d_lock);
1088 
1089                 /*
1090                  * The list move itself will be made by the common LRU code. At
1091                  * this point, we've dropped the dentry->d_lock but keep the
1092                  * lru lock. This is safe to do, since every list movement is
1093                  * protected by the lru lock even if both locks are held.
1094                  *
1095                  * This is guaranteed by the fact that all LRU management
1096                  * functions are intermediated by the LRU API calls like
1097                  * list_lru_add and list_lru_del. List movement in this file
1098                  * only ever occur through this functions or through callbacks
1099                  * like this one, that are called from the LRU API.
1100                  *
1101                  * The only exceptions to this are functions like
1102                  * shrink_dentry_list, and code that first checks for the
1103                  * DCACHE_SHRINK_LIST flag.  Those are guaranteed to be
1104                  * operating only with stack provided lists after they are
1105                  * properly isolated from the main list.  It is thus, always a
1106                  * local access.
1107                  */
1108                 return LRU_ROTATE;
1109         }
1110 
1111         d_lru_shrink_move(lru, dentry, freeable);
1112         spin_unlock(&dentry->d_lock);
1113 
1114         return LRU_REMOVED;
1115 }
1116 
1117 /**
1118  * prune_dcache_sb - shrink the dcache
1119  * @sb: superblock
1120  * @sc: shrink control, passed to list_lru_shrink_walk()
1121  *
1122  * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1123  * is done when we need more memory and called from the superblock shrinker
1124  * function.
1125  *
1126  * This function may fail to free any resources if all the dentries are in
1127  * use.
1128  */
1129 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1130 {
1131         LIST_HEAD(dispose);
1132         long freed;
1133 
1134         freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1135                                      dentry_lru_isolate, &dispose);
1136         shrink_dentry_list(&dispose);
1137         return freed;
1138 }
1139 
1140 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1141                 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1142 {
1143         struct list_head *freeable = arg;
1144         struct dentry   *dentry = container_of(item, struct dentry, d_lru);
1145 
1146         /*
1147          * we are inverting the lru lock/dentry->d_lock here,
1148          * so use a trylock. If we fail to get the lock, just skip
1149          * it
1150          */
1151         if (!spin_trylock(&dentry->d_lock))
1152                 return LRU_SKIP;
1153 
1154         d_lru_shrink_move(lru, dentry, freeable);
1155         spin_unlock(&dentry->d_lock);
1156 
1157         return LRU_REMOVED;
1158 }
1159 
1160 
1161 /**
1162  * shrink_dcache_sb - shrink dcache for a superblock
1163  * @sb: superblock
1164  *
1165  * Shrink the dcache for the specified super block. This is used to free
1166  * the dcache before unmounting a file system.
1167  */
1168 void shrink_dcache_sb(struct super_block *sb)
1169 {
1170         long freed;
1171 
1172         do {
1173                 LIST_HEAD(dispose);
1174 
1175                 freed = list_lru_walk(&sb->s_dentry_lru,
1176                         dentry_lru_isolate_shrink, &dispose, 1024);
1177 
1178                 this_cpu_sub(nr_dentry_unused, freed);
1179                 shrink_dentry_list(&dispose);
1180                 cond_resched();
1181         } while (list_lru_count(&sb->s_dentry_lru) > 0);
1182 }
1183 EXPORT_SYMBOL(shrink_dcache_sb);
1184 
1185 /**
1186  * enum d_walk_ret - action to talke during tree walk
1187  * @D_WALK_CONTINUE:    contrinue walk
1188  * @D_WALK_QUIT:        quit walk
1189  * @D_WALK_NORETRY:     quit when retry is needed
1190  * @D_WALK_SKIP:        skip this dentry and its children
1191  */
1192 enum d_walk_ret {
1193         D_WALK_CONTINUE,
1194         D_WALK_QUIT,
1195         D_WALK_NORETRY,
1196         D_WALK_SKIP,
1197 };
1198 
1199 /**
1200  * d_walk - walk the dentry tree
1201  * @parent:     start of walk
1202  * @data:       data passed to @enter() and @finish()
1203  * @enter:      callback when first entering the dentry
1204  * @finish:     callback when successfully finished the walk
1205  *
1206  * The @enter() and @finish() callbacks are called with d_lock held.
1207  */
1208 static void d_walk(struct dentry *parent, void *data,
1209                    enum d_walk_ret (*enter)(void *, struct dentry *),
1210                    void (*finish)(void *))
1211 {
1212         struct dentry *this_parent;
1213         struct list_head *next;
1214         unsigned seq = 0;
1215         enum d_walk_ret ret;
1216         bool retry = true;
1217 
1218 again:
1219         read_seqbegin_or_lock(&rename_lock, &seq);
1220         this_parent = parent;
1221         spin_lock(&this_parent->d_lock);
1222 
1223         ret = enter(data, this_parent);
1224         switch (ret) {
1225         case D_WALK_CONTINUE:
1226                 break;
1227         case D_WALK_QUIT:
1228         case D_WALK_SKIP:
1229                 goto out_unlock;
1230         case D_WALK_NORETRY:
1231                 retry = false;
1232                 break;
1233         }
1234 repeat:
1235         next = this_parent->d_subdirs.next;
1236 resume:
1237         while (next != &this_parent->d_subdirs) {
1238                 struct list_head *tmp = next;
1239                 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1240                 next = tmp->next;
1241 
1242                 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1243                         continue;
1244 
1245                 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1246 
1247                 ret = enter(data, dentry);
1248                 switch (ret) {
1249                 case D_WALK_CONTINUE:
1250                         break;
1251                 case D_WALK_QUIT:
1252                         spin_unlock(&dentry->d_lock);
1253                         goto out_unlock;
1254                 case D_WALK_NORETRY:
1255                         retry = false;
1256                         break;
1257                 case D_WALK_SKIP:
1258                         spin_unlock(&dentry->d_lock);
1259                         continue;
1260                 }
1261 
1262                 if (!list_empty(&dentry->d_subdirs)) {
1263                         spin_unlock(&this_parent->d_lock);
1264                         spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1265                         this_parent = dentry;
1266                         spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1267                         goto repeat;
1268                 }
1269                 spin_unlock(&dentry->d_lock);
1270         }
1271         /*
1272          * All done at this level ... ascend and resume the search.
1273          */
1274         rcu_read_lock();
1275 ascend:
1276         if (this_parent != parent) {
1277                 struct dentry *child = this_parent;
1278                 this_parent = child->d_parent;
1279 
1280                 spin_unlock(&child->d_lock);
1281                 spin_lock(&this_parent->d_lock);
1282 
1283                 /* might go back up the wrong parent if we have had a rename. */
1284                 if (need_seqretry(&rename_lock, seq))
1285                         goto rename_retry;
1286                 /* go into the first sibling still alive */
1287                 do {
1288                         next = child->d_child.next;
1289                         if (next == &this_parent->d_subdirs)
1290                                 goto ascend;
1291                         child = list_entry(next, struct dentry, d_child);
1292                 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1293                 rcu_read_unlock();
1294                 goto resume;
1295         }
1296         if (need_seqretry(&rename_lock, seq))
1297                 goto rename_retry;
1298         rcu_read_unlock();
1299         if (finish)
1300                 finish(data);
1301 
1302 out_unlock:
1303         spin_unlock(&this_parent->d_lock);
1304         done_seqretry(&rename_lock, seq);
1305         return;
1306 
1307 rename_retry:
1308         spin_unlock(&this_parent->d_lock);
1309         rcu_read_unlock();
1310         BUG_ON(seq & 1);
1311         if (!retry)
1312                 return;
1313         seq = 1;
1314         goto again;
1315 }
1316 
1317 struct check_mount {
1318         struct vfsmount *mnt;
1319         unsigned int mounted;
1320 };
1321 
1322 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1323 {
1324         struct check_mount *info = data;
1325         struct path path = { .mnt = info->mnt, .dentry = dentry };
1326 
1327         if (likely(!d_mountpoint(dentry)))
1328                 return D_WALK_CONTINUE;
1329         if (__path_is_mountpoint(&path)) {
1330                 info->mounted = 1;
1331                 return D_WALK_QUIT;
1332         }
1333         return D_WALK_CONTINUE;
1334 }
1335 
1336 /**
1337  * path_has_submounts - check for mounts over a dentry in the
1338  *                      current namespace.
1339  * @parent: path to check.
1340  *
1341  * Return true if the parent or its subdirectories contain
1342  * a mount point in the current namespace.
1343  */
1344 int path_has_submounts(const struct path *parent)
1345 {
1346         struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1347 
1348         read_seqlock_excl(&mount_lock);
1349         d_walk(parent->dentry, &data, path_check_mount, NULL);
1350         read_sequnlock_excl(&mount_lock);
1351 
1352         return data.mounted;
1353 }
1354 EXPORT_SYMBOL(path_has_submounts);
1355 
1356 /*
1357  * Called by mount code to set a mountpoint and check if the mountpoint is
1358  * reachable (e.g. NFS can unhash a directory dentry and then the complete
1359  * subtree can become unreachable).
1360  *
1361  * Only one of d_invalidate() and d_set_mounted() must succeed.  For
1362  * this reason take rename_lock and d_lock on dentry and ancestors.
1363  */
1364 int d_set_mounted(struct dentry *dentry)
1365 {
1366         struct dentry *p;
1367         int ret = -ENOENT;
1368         write_seqlock(&rename_lock);
1369         for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1370                 /* Need exclusion wrt. d_invalidate() */
1371                 spin_lock(&p->d_lock);
1372                 if (unlikely(d_unhashed(p))) {
1373                         spin_unlock(&p->d_lock);
1374                         goto out;
1375                 }
1376                 spin_unlock(&p->d_lock);
1377         }
1378         spin_lock(&dentry->d_lock);
1379         if (!d_unlinked(dentry)) {
1380                 ret = -EBUSY;
1381                 if (!d_mountpoint(dentry)) {
1382                         dentry->d_flags |= DCACHE_MOUNTED;
1383                         ret = 0;
1384                 }
1385         }
1386         spin_unlock(&dentry->d_lock);
1387 out:
1388         write_sequnlock(&rename_lock);
1389         return ret;
1390 }
1391 
1392 /*
1393  * Search the dentry child list of the specified parent,
1394  * and move any unused dentries to the end of the unused
1395  * list for prune_dcache(). We descend to the next level
1396  * whenever the d_subdirs list is non-empty and continue
1397  * searching.
1398  *
1399  * It returns zero iff there are no unused children,
1400  * otherwise  it returns the number of children moved to
1401  * the end of the unused list. This may not be the total
1402  * number of unused children, because select_parent can
1403  * drop the lock and return early due to latency
1404  * constraints.
1405  */
1406 
1407 struct select_data {
1408         struct dentry *start;
1409         struct list_head dispose;
1410         int found;
1411 };
1412 
1413 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1414 {
1415         struct select_data *data = _data;
1416         enum d_walk_ret ret = D_WALK_CONTINUE;
1417 
1418         if (data->start == dentry)
1419                 goto out;
1420 
1421         if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1422                 data->found++;
1423         } else {
1424                 if (dentry->d_flags & DCACHE_LRU_LIST)
1425                         d_lru_del(dentry);
1426                 if (!dentry->d_lockref.count) {
1427                         d_shrink_add(dentry, &data->dispose);
1428                         data->found++;
1429                 }
1430         }
1431         /*
1432          * We can return to the caller if we have found some (this
1433          * ensures forward progress). We'll be coming back to find
1434          * the rest.
1435          */
1436         if (!list_empty(&data->dispose))
1437                 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1438 out:
1439         return ret;
1440 }
1441 
1442 /**
1443  * shrink_dcache_parent - prune dcache
1444  * @parent: parent of entries to prune
1445  *
1446  * Prune the dcache to remove unused children of the parent dentry.
1447  */
1448 void shrink_dcache_parent(struct dentry *parent)
1449 {
1450         for (;;) {
1451                 struct select_data data;
1452 
1453                 INIT_LIST_HEAD(&data.dispose);
1454                 data.start = parent;
1455                 data.found = 0;
1456 
1457                 d_walk(parent, &data, select_collect, NULL);
1458                 if (!data.found)
1459                         break;
1460 
1461                 shrink_dentry_list(&data.dispose);
1462                 cond_resched();
1463         }
1464 }
1465 EXPORT_SYMBOL(shrink_dcache_parent);
1466 
1467 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1468 {
1469         /* it has busy descendents; complain about those instead */
1470         if (!list_empty(&dentry->d_subdirs))
1471                 return D_WALK_CONTINUE;
1472 
1473         /* root with refcount 1 is fine */
1474         if (dentry == _data && dentry->d_lockref.count == 1)
1475                 return D_WALK_CONTINUE;
1476 
1477         printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1478                         " still in use (%d) [unmount of %s %s]\n",
1479                        dentry,
1480                        dentry->d_inode ?
1481                        dentry->d_inode->i_ino : 0UL,
1482                        dentry,
1483                        dentry->d_lockref.count,
1484                        dentry->d_sb->s_type->name,
1485                        dentry->d_sb->s_id);
1486         WARN_ON(1);
1487         return D_WALK_CONTINUE;
1488 }
1489 
1490 static void do_one_tree(struct dentry *dentry)
1491 {
1492         shrink_dcache_parent(dentry);
1493         d_walk(dentry, dentry, umount_check, NULL);
1494         d_drop(dentry);
1495         dput(dentry);
1496 }
1497 
1498 /*
1499  * destroy the dentries attached to a superblock on unmounting
1500  */
1501 void shrink_dcache_for_umount(struct super_block *sb)
1502 {
1503         struct dentry *dentry;
1504 
1505         WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1506 
1507         dentry = sb->s_root;
1508         sb->s_root = NULL;
1509         do_one_tree(dentry);
1510 
1511         while (!hlist_bl_empty(&sb->s_roots)) {
1512                 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1513                 do_one_tree(dentry);
1514         }
1515 }
1516 
1517 struct detach_data {
1518         struct select_data select;
1519         struct dentry *mountpoint;
1520 };
1521 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1522 {
1523         struct detach_data *data = _data;
1524 
1525         if (d_mountpoint(dentry)) {
1526                 __dget_dlock(dentry);
1527                 data->mountpoint = dentry;
1528                 return D_WALK_QUIT;
1529         }
1530 
1531         return select_collect(&data->select, dentry);
1532 }
1533 
1534 static void check_and_drop(void *_data)
1535 {
1536         struct detach_data *data = _data;
1537 
1538         if (!data->mountpoint && list_empty(&data->select.dispose))
1539                 __d_drop(data->select.start);
1540 }
1541 
1542 /**
1543  * d_invalidate - detach submounts, prune dcache, and drop
1544  * @dentry: dentry to invalidate (aka detach, prune and drop)
1545  *
1546  * no dcache lock.
1547  *
1548  * The final d_drop is done as an atomic operation relative to
1549  * rename_lock ensuring there are no races with d_set_mounted.  This
1550  * ensures there are no unhashed dentries on the path to a mountpoint.
1551  */
1552 void d_invalidate(struct dentry *dentry)
1553 {
1554         /*
1555          * If it's already been dropped, return OK.
1556          */
1557         spin_lock(&dentry->d_lock);
1558         if (d_unhashed(dentry)) {
1559                 spin_unlock(&dentry->d_lock);
1560                 return;
1561         }
1562         spin_unlock(&dentry->d_lock);
1563 
1564         /* Negative dentries can be dropped without further checks */
1565         if (!dentry->d_inode) {
1566                 d_drop(dentry);
1567                 return;
1568         }
1569 
1570         for (;;) {
1571                 struct detach_data data;
1572 
1573                 data.mountpoint = NULL;
1574                 INIT_LIST_HEAD(&data.select.dispose);
1575                 data.select.start = dentry;
1576                 data.select.found = 0;
1577 
1578                 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1579 
1580                 if (!list_empty(&data.select.dispose))
1581                         shrink_dentry_list(&data.select.dispose);
1582                 else if (!data.mountpoint)
1583                         return;
1584 
1585                 if (data.mountpoint) {
1586                         detach_mounts(data.mountpoint);
1587                         dput(data.mountpoint);
1588                 }
1589                 cond_resched();
1590         }
1591 }
1592 EXPORT_SYMBOL(d_invalidate);
1593 
1594 /**
1595  * __d_alloc    -       allocate a dcache entry
1596  * @sb: filesystem it will belong to
1597  * @name: qstr of the name
1598  *
1599  * Allocates a dentry. It returns %NULL if there is insufficient memory
1600  * available. On a success the dentry is returned. The name passed in is
1601  * copied and the copy passed in may be reused after this call.
1602  */
1603  
1604 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1605 {
1606         struct dentry *dentry;
1607         char *dname;
1608         int err;
1609 
1610         dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1611         if (!dentry)
1612                 return NULL;
1613 
1614         /*
1615          * We guarantee that the inline name is always NUL-terminated.
1616          * This way the memcpy() done by the name switching in rename
1617          * will still always have a NUL at the end, even if we might
1618          * be overwriting an internal NUL character
1619          */
1620         dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1621         if (unlikely(!name)) {
1622                 name = &slash_name;
1623                 dname = dentry->d_iname;
1624         } else if (name->len > DNAME_INLINE_LEN-1) {
1625                 size_t size = offsetof(struct external_name, name[1]);
1626                 struct external_name *p = kmalloc(size + name->len,
1627                                                   GFP_KERNEL_ACCOUNT);
1628                 if (!p) {
1629                         kmem_cache_free(dentry_cache, dentry); 
1630                         return NULL;
1631                 }
1632                 atomic_set(&p->u.count, 1);
1633                 dname = p->name;
1634         } else  {
1635                 dname = dentry->d_iname;
1636         }       
1637 
1638         dentry->d_name.len = name->len;
1639         dentry->d_name.hash = name->hash;
1640         memcpy(dname, name->name, name->len);
1641         dname[name->len] = 0;
1642 
1643         /* Make sure we always see the terminating NUL character */
1644         smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1645 
1646         dentry->d_lockref.count = 1;
1647         dentry->d_flags = 0;
1648         spin_lock_init(&dentry->d_lock);
1649         seqcount_init(&dentry->d_seq);
1650         dentry->d_inode = NULL;
1651         dentry->d_parent = dentry;
1652         dentry->d_sb = sb;
1653         dentry->d_op = NULL;
1654         dentry->d_fsdata = NULL;
1655         INIT_HLIST_BL_NODE(&dentry->d_hash);
1656         INIT_LIST_HEAD(&dentry->d_lru);
1657         INIT_LIST_HEAD(&dentry->d_subdirs);
1658         INIT_HLIST_NODE(&dentry->d_u.d_alias);
1659         INIT_LIST_HEAD(&dentry->d_child);
1660         d_set_d_op(dentry, dentry->d_sb->s_d_op);
1661 
1662         if (dentry->d_op && dentry->d_op->d_init) {
1663                 err = dentry->d_op->d_init(dentry);
1664                 if (err) {
1665                         if (dname_external(dentry))
1666                                 kfree(external_name(dentry));
1667                         kmem_cache_free(dentry_cache, dentry);
1668                         return NULL;
1669                 }
1670         }
1671 
1672         this_cpu_inc(nr_dentry);
1673 
1674         return dentry;
1675 }
1676 
1677 /**
1678  * d_alloc      -       allocate a dcache entry
1679  * @parent: parent of entry to allocate
1680  * @name: qstr of the name
1681  *
1682  * Allocates a dentry. It returns %NULL if there is insufficient memory
1683  * available. On a success the dentry is returned. The name passed in is
1684  * copied and the copy passed in may be reused after this call.
1685  */
1686 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1687 {
1688         struct dentry *dentry = __d_alloc(parent->d_sb, name);
1689         if (!dentry)
1690                 return NULL;
1691         dentry->d_flags |= DCACHE_RCUACCESS;
1692         spin_lock(&parent->d_lock);
1693         /*
1694          * don't need child lock because it is not subject
1695          * to concurrency here
1696          */
1697         __dget_dlock(parent);
1698         dentry->d_parent = parent;
1699         list_add(&dentry->d_child, &parent->d_subdirs);
1700         spin_unlock(&parent->d_lock);
1701 
1702         return dentry;
1703 }
1704 EXPORT_SYMBOL(d_alloc);
1705 
1706 struct dentry *d_alloc_anon(struct super_block *sb)
1707 {
1708         return __d_alloc(sb, NULL);
1709 }
1710 EXPORT_SYMBOL(d_alloc_anon);
1711 
1712 struct dentry *d_alloc_cursor(struct dentry * parent)
1713 {
1714         struct dentry *dentry = d_alloc_anon(parent->d_sb);
1715         if (dentry) {
1716                 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1717                 dentry->d_parent = dget(parent);
1718         }
1719         return dentry;
1720 }
1721 
1722 /**
1723  * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1724  * @sb: the superblock
1725  * @name: qstr of the name
1726  *
1727  * For a filesystem that just pins its dentries in memory and never
1728  * performs lookups at all, return an unhashed IS_ROOT dentry.
1729  */
1730 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1731 {
1732         return __d_alloc(sb, name);
1733 }
1734 EXPORT_SYMBOL(d_alloc_pseudo);
1735 
1736 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1737 {
1738         struct qstr q;
1739 
1740         q.name = name;
1741         q.hash_len = hashlen_string(parent, name);
1742         return d_alloc(parent, &q);
1743 }
1744 EXPORT_SYMBOL(d_alloc_name);
1745 
1746 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1747 {
1748         WARN_ON_ONCE(dentry->d_op);
1749         WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH  |
1750                                 DCACHE_OP_COMPARE       |
1751                                 DCACHE_OP_REVALIDATE    |
1752                                 DCACHE_OP_WEAK_REVALIDATE       |
1753                                 DCACHE_OP_DELETE        |
1754                                 DCACHE_OP_REAL));
1755         dentry->d_op = op;
1756         if (!op)
1757                 return;
1758         if (op->d_hash)
1759                 dentry->d_flags |= DCACHE_OP_HASH;
1760         if (op->d_compare)
1761                 dentry->d_flags |= DCACHE_OP_COMPARE;
1762         if (op->d_revalidate)
1763                 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1764         if (op->d_weak_revalidate)
1765                 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1766         if (op->d_delete)
1767                 dentry->d_flags |= DCACHE_OP_DELETE;
1768         if (op->d_prune)
1769                 dentry->d_flags |= DCACHE_OP_PRUNE;
1770         if (op->d_real)
1771                 dentry->d_flags |= DCACHE_OP_REAL;
1772 
1773 }
1774 EXPORT_SYMBOL(d_set_d_op);
1775 
1776 
1777 /*
1778  * d_set_fallthru - Mark a dentry as falling through to a lower layer
1779  * @dentry - The dentry to mark
1780  *
1781  * Mark a dentry as falling through to the lower layer (as set with
1782  * d_pin_lower()).  This flag may be recorded on the medium.
1783  */
1784 void d_set_fallthru(struct dentry *dentry)
1785 {
1786         spin_lock(&dentry->d_lock);
1787         dentry->d_flags |= DCACHE_FALLTHRU;
1788         spin_unlock(&dentry->d_lock);
1789 }
1790 EXPORT_SYMBOL(d_set_fallthru);
1791 
1792 static unsigned d_flags_for_inode(struct inode *inode)
1793 {
1794         unsigned add_flags = DCACHE_REGULAR_TYPE;
1795 
1796         if (!inode)
1797                 return DCACHE_MISS_TYPE;
1798 
1799         if (S_ISDIR(inode->i_mode)) {
1800                 add_flags = DCACHE_DIRECTORY_TYPE;
1801                 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1802                         if (unlikely(!inode->i_op->lookup))
1803                                 add_flags = DCACHE_AUTODIR_TYPE;
1804                         else
1805                                 inode->i_opflags |= IOP_LOOKUP;
1806                 }
1807                 goto type_determined;
1808         }
1809 
1810         if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1811                 if (unlikely(inode->i_op->get_link)) {
1812                         add_flags = DCACHE_SYMLINK_TYPE;
1813                         goto type_determined;
1814                 }
1815                 inode->i_opflags |= IOP_NOFOLLOW;
1816         }
1817 
1818         if (unlikely(!S_ISREG(inode->i_mode)))
1819                 add_flags = DCACHE_SPECIAL_TYPE;
1820 
1821 type_determined:
1822         if (unlikely(IS_AUTOMOUNT(inode)))
1823                 add_flags |= DCACHE_NEED_AUTOMOUNT;
1824         return add_flags;
1825 }
1826 
1827 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1828 {
1829         unsigned add_flags = d_flags_for_inode(inode);
1830         WARN_ON(d_in_lookup(dentry));
1831 
1832         spin_lock(&dentry->d_lock);
1833         hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1834         raw_write_seqcount_begin(&dentry->d_seq);
1835         __d_set_inode_and_type(dentry, inode, add_flags);
1836         raw_write_seqcount_end(&dentry->d_seq);
1837         fsnotify_update_flags(dentry);
1838         spin_unlock(&dentry->d_lock);
1839 }
1840 
1841 /**
1842  * d_instantiate - fill in inode information for a dentry
1843  * @entry: dentry to complete
1844  * @inode: inode to attach to this dentry
1845  *
1846  * Fill in inode information in the entry.
1847  *
1848  * This turns negative dentries into productive full members
1849  * of society.
1850  *
1851  * NOTE! This assumes that the inode count has been incremented
1852  * (or otherwise set) by the caller to indicate that it is now
1853  * in use by the dcache.
1854  */
1855  
1856 void d_instantiate(struct dentry *entry, struct inode * inode)
1857 {
1858         BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1859         if (inode) {
1860                 security_d_instantiate(entry, inode);
1861                 spin_lock(&inode->i_lock);
1862                 __d_instantiate(entry, inode);
1863                 spin_unlock(&inode->i_lock);
1864         }
1865 }
1866 EXPORT_SYMBOL(d_instantiate);
1867 
1868 /*
1869  * This should be equivalent to d_instantiate() + unlock_new_inode(),
1870  * with lockdep-related part of unlock_new_inode() done before
1871  * anything else.  Use that instead of open-coding d_instantiate()/
1872  * unlock_new_inode() combinations.
1873  */
1874 void d_instantiate_new(struct dentry *entry, struct inode *inode)
1875 {
1876         BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1877         BUG_ON(!inode);
1878         lockdep_annotate_inode_mutex_key(inode);
1879         security_d_instantiate(entry, inode);
1880         spin_lock(&inode->i_lock);
1881         __d_instantiate(entry, inode);
1882         WARN_ON(!(inode->i_state & I_NEW));
1883         inode->i_state &= ~I_NEW;
1884         smp_mb();
1885         wake_up_bit(&inode->i_state, __I_NEW);
1886         spin_unlock(&inode->i_lock);
1887 }
1888 EXPORT_SYMBOL(d_instantiate_new);
1889 
1890 /**
1891  * d_instantiate_no_diralias - instantiate a non-aliased dentry
1892  * @entry: dentry to complete
1893  * @inode: inode to attach to this dentry
1894  *
1895  * Fill in inode information in the entry.  If a directory alias is found, then
1896  * return an error (and drop inode).  Together with d_materialise_unique() this
1897  * guarantees that a directory inode may never have more than one alias.
1898  */
1899 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1900 {
1901         BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1902 
1903         security_d_instantiate(entry, inode);
1904         spin_lock(&inode->i_lock);
1905         if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1906                 spin_unlock(&inode->i_lock);
1907                 iput(inode);
1908                 return -EBUSY;
1909         }
1910         __d_instantiate(entry, inode);
1911         spin_unlock(&inode->i_lock);
1912 
1913         return 0;
1914 }
1915 EXPORT_SYMBOL(d_instantiate_no_diralias);
1916 
1917 struct dentry *d_make_root(struct inode *root_inode)
1918 {
1919         struct dentry *res = NULL;
1920 
1921         if (root_inode) {
1922                 res = d_alloc_anon(root_inode->i_sb);
1923                 if (res)
1924                         d_instantiate(res, root_inode);
1925                 else
1926                         iput(root_inode);
1927         }
1928         return res;
1929 }
1930 EXPORT_SYMBOL(d_make_root);
1931 
1932 static struct dentry * __d_find_any_alias(struct inode *inode)
1933 {
1934         struct dentry *alias;
1935 
1936         if (hlist_empty(&inode->i_dentry))
1937                 return NULL;
1938         alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1939         __dget(alias);
1940         return alias;
1941 }
1942 
1943 /**
1944  * d_find_any_alias - find any alias for a given inode
1945  * @inode: inode to find an alias for
1946  *
1947  * If any aliases exist for the given inode, take and return a
1948  * reference for one of them.  If no aliases exist, return %NULL.
1949  */
1950 struct dentry *d_find_any_alias(struct inode *inode)
1951 {
1952         struct dentry *de;
1953 
1954         spin_lock(&inode->i_lock);
1955         de = __d_find_any_alias(inode);
1956         spin_unlock(&inode->i_lock);
1957         return de;
1958 }
1959 EXPORT_SYMBOL(d_find_any_alias);
1960 
1961 static struct dentry *__d_instantiate_anon(struct dentry *dentry,
1962                                            struct inode *inode,
1963                                            bool disconnected)
1964 {
1965         struct dentry *res;
1966         unsigned add_flags;
1967 
1968         security_d_instantiate(dentry, inode);
1969         spin_lock(&inode->i_lock);
1970         res = __d_find_any_alias(inode);
1971         if (res) {
1972                 spin_unlock(&inode->i_lock);
1973                 dput(dentry);
1974                 goto out_iput;
1975         }
1976 
1977         /* attach a disconnected dentry */
1978         add_flags = d_flags_for_inode(inode);
1979 
1980         if (disconnected)
1981                 add_flags |= DCACHE_DISCONNECTED;
1982 
1983         spin_lock(&dentry->d_lock);
1984         __d_set_inode_and_type(dentry, inode, add_flags);
1985         hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1986         if (!disconnected) {
1987                 hlist_bl_lock(&dentry->d_sb->s_roots);
1988                 hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots);
1989                 hlist_bl_unlock(&dentry->d_sb->s_roots);
1990         }
1991         spin_unlock(&dentry->d_lock);
1992         spin_unlock(&inode->i_lock);
1993 
1994         return dentry;
1995 
1996  out_iput:
1997         iput(inode);
1998         return res;
1999 }
2000 
2001 struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode)
2002 {
2003         return __d_instantiate_anon(dentry, inode, true);
2004 }
2005 EXPORT_SYMBOL(d_instantiate_anon);
2006 
2007 static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
2008 {
2009         struct dentry *tmp;
2010         struct dentry *res;
2011 
2012         if (!inode)
2013                 return ERR_PTR(-ESTALE);
2014         if (IS_ERR(inode))
2015                 return ERR_CAST(inode);
2016 
2017         res = d_find_any_alias(inode);
2018         if (res)
2019                 goto out_iput;
2020 
2021         tmp = d_alloc_anon(inode->i_sb);
2022         if (!tmp) {
2023                 res = ERR_PTR(-ENOMEM);
2024                 goto out_iput;
2025         }
2026 
2027         return __d_instantiate_anon(tmp, inode, disconnected);
2028 
2029 out_iput:
2030         iput(inode);
2031         return res;
2032 }
2033 
2034 /**
2035  * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2036  * @inode: inode to allocate the dentry for
2037  *
2038  * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2039  * similar open by handle operations.  The returned dentry may be anonymous,
2040  * or may have a full name (if the inode was already in the cache).
2041  *
2042  * When called on a directory inode, we must ensure that the inode only ever
2043  * has one dentry.  If a dentry is found, that is returned instead of
2044  * allocating a new one.
2045  *
2046  * On successful return, the reference to the inode has been transferred
2047  * to the dentry.  In case of an error the reference on the inode is released.
2048  * To make it easier to use in export operations a %NULL or IS_ERR inode may
2049  * be passed in and the error will be propagated to the return value,
2050  * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2051  */
2052 struct dentry *d_obtain_alias(struct inode *inode)
2053 {
2054         return __d_obtain_alias(inode, true);
2055 }
2056 EXPORT_SYMBOL(d_obtain_alias);
2057 
2058 /**
2059  * d_obtain_root - find or allocate a dentry for a given inode
2060  * @inode: inode to allocate the dentry for
2061  *
2062  * Obtain an IS_ROOT dentry for the root of a filesystem.
2063  *
2064  * We must ensure that directory inodes only ever have one dentry.  If a
2065  * dentry is found, that is returned instead of allocating a new one.
2066  *
2067  * On successful return, the reference to the inode has been transferred
2068  * to the dentry.  In case of an error the reference on the inode is
2069  * released.  A %NULL or IS_ERR inode may be passed in and will be the
2070  * error will be propagate to the return value, with a %NULL @inode
2071  * replaced by ERR_PTR(-ESTALE).
2072  */
2073 struct dentry *d_obtain_root(struct inode *inode)
2074 {
2075         return __d_obtain_alias(inode, false);
2076 }
2077 EXPORT_SYMBOL(d_obtain_root);
2078 
2079 /**
2080  * d_add_ci - lookup or allocate new dentry with case-exact name
2081  * @inode:  the inode case-insensitive lookup has found
2082  * @dentry: the negative dentry that was passed to the parent's lookup func
2083  * @name:   the case-exact name to be associated with the returned dentry
2084  *
2085  * This is to avoid filling the dcache with case-insensitive names to the
2086  * same inode, only the actual correct case is stored in the dcache for
2087  * case-insensitive filesystems.
2088  *
2089  * For a case-insensitive lookup match and if the the case-exact dentry
2090  * already exists in in the dcache, use it and return it.
2091  *
2092  * If no entry exists with the exact case name, allocate new dentry with
2093  * the exact case, and return the spliced entry.
2094  */
2095 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2096                         struct qstr *name)
2097 {
2098         struct dentry *found, *res;
2099 
2100         /*
2101          * First check if a dentry matching the name already exists,
2102          * if not go ahead and create it now.
2103          */
2104         found = d_hash_and_lookup(dentry->d_parent, name);
2105         if (found) {
2106                 iput(inode);
2107                 return found;
2108         }
2109         if (d_in_lookup(dentry)) {
2110                 found = d_alloc_parallel(dentry->d_parent, name,
2111                                         dentry->d_wait);
2112                 if (IS_ERR(found) || !d_in_lookup(found)) {
2113                         iput(inode);
2114                         return found;
2115                 }
2116         } else {
2117                 found = d_alloc(dentry->d_parent, name);
2118                 if (!found) {
2119                         iput(inode);
2120                         return ERR_PTR(-ENOMEM);
2121                 } 
2122         }
2123         res = d_splice_alias(inode, found);
2124         if (res) {
2125                 dput(found);
2126                 return res;
2127         }
2128         return found;
2129 }
2130 EXPORT_SYMBOL(d_add_ci);
2131 
2132 
2133 static inline bool d_same_name(const struct dentry *dentry,
2134                                 const struct dentry *parent,
2135                                 const struct qstr *name)
2136 {
2137         if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2138                 if (dentry->d_name.len != name->len)
2139                         return false;
2140                 return dentry_cmp(dentry, name->name, name->len) == 0;
2141         }
2142         return parent->d_op->d_compare(dentry,
2143                                        dentry->d_name.len, dentry->d_name.name,
2144                                        name) == 0;
2145 }
2146 
2147 /**
2148  * __d_lookup_rcu - search for a dentry (racy, store-free)
2149  * @parent: parent dentry
2150  * @name: qstr of name we wish to find
2151  * @seqp: returns d_seq value at the point where the dentry was found
2152  * Returns: dentry, or NULL
2153  *
2154  * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2155  * resolution (store-free path walking) design described in
2156  * Documentation/filesystems/path-lookup.txt.
2157  *
2158  * This is not to be used outside core vfs.
2159  *
2160  * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2161  * held, and rcu_read_lock held. The returned dentry must not be stored into
2162  * without taking d_lock and checking d_seq sequence count against @seq
2163  * returned here.
2164  *
2165  * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2166  * function.
2167  *
2168  * Alternatively, __d_lookup_rcu may be called again to look up the child of
2169  * the returned dentry, so long as its parent's seqlock is checked after the
2170  * child is looked up. Thus, an interlocking stepping of sequence lock checks
2171  * is formed, giving integrity down the path walk.
2172  *
2173  * NOTE! The caller *has* to check the resulting dentry against the sequence
2174  * number we've returned before using any of the resulting dentry state!
2175  */
2176 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2177                                 const struct qstr *name,
2178                                 unsigned *seqp)
2179 {
2180         u64 hashlen = name->hash_len;
2181         const unsigned char *str = name->name;
2182         struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2183         struct hlist_bl_node *node;
2184         struct dentry *dentry;
2185 
2186         /*
2187          * Note: There is significant duplication with __d_lookup_rcu which is
2188          * required to prevent single threaded performance regressions
2189          * especially on architectures where smp_rmb (in seqcounts) are costly.
2190          * Keep the two functions in sync.
2191          */
2192 
2193         /*
2194          * The hash list is protected using RCU.
2195          *
2196          * Carefully use d_seq when comparing a candidate dentry, to avoid
2197          * races with d_move().
2198          *
2199          * It is possible that concurrent renames can mess up our list
2200          * walk here and result in missing our dentry, resulting in the
2201          * false-negative result. d_lookup() protects against concurrent
2202          * renames using rename_lock seqlock.
2203          *
2204          * See Documentation/filesystems/path-lookup.txt for more details.
2205          */
2206         hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2207                 unsigned seq;
2208 
2209 seqretry:
2210                 /*
2211                  * The dentry sequence count protects us from concurrent
2212                  * renames, and thus protects parent and name fields.
2213                  *
2214                  * The caller must perform a seqcount check in order
2215                  * to do anything useful with the returned dentry.
2216                  *
2217                  * NOTE! We do a "raw" seqcount_begin here. That means that
2218                  * we don't wait for the sequence count to stabilize if it
2219                  * is in the middle of a sequence change. If we do the slow
2220                  * dentry compare, we will do seqretries until it is stable,
2221                  * and if we end up with a successful lookup, we actually
2222                  * want to exit RCU lookup anyway.
2223                  *
2224                  * Note that raw_seqcount_begin still *does* smp_rmb(), so
2225                  * we are still guaranteed NUL-termination of ->d_name.name.
2226                  */
2227                 seq = raw_seqcount_begin(&dentry->d_seq);
2228                 if (dentry->d_parent != parent)
2229                         continue;
2230                 if (d_unhashed(dentry))
2231                         continue;
2232 
2233                 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2234                         int tlen;
2235                         const char *tname;
2236                         if (dentry->d_name.hash != hashlen_hash(hashlen))
2237                                 continue;
2238                         tlen = dentry->d_name.len;
2239                         tname = dentry->d_name.name;
2240                         /* we want a consistent (name,len) pair */
2241                         if (read_seqcount_retry(&dentry->d_seq, seq)) {
2242                                 cpu_relax();
2243                                 goto seqretry;
2244                         }
2245                         if (parent->d_op->d_compare(dentry,
2246                                                     tlen, tname, name) != 0)
2247                                 continue;
2248                 } else {
2249                         if (dentry->d_name.hash_len != hashlen)
2250                                 continue;
2251                         if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2252                                 continue;
2253                 }
2254                 *seqp = seq;
2255                 return dentry;
2256         }
2257         return NULL;
2258 }
2259 
2260 /**
2261  * d_lookup - search for a dentry
2262  * @parent: parent dentry
2263  * @name: qstr of name we wish to find
2264  * Returns: dentry, or NULL
2265  *
2266  * d_lookup searches the children of the parent dentry for the name in
2267  * question. If the dentry is found its reference count is incremented and the
2268  * dentry is returned. The caller must use dput to free the entry when it has
2269  * finished using it. %NULL is returned if the dentry does not exist.
2270  */
2271 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2272 {
2273         struct dentry *dentry;
2274         unsigned seq;
2275 
2276         do {
2277                 seq = read_seqbegin(&rename_lock);
2278                 dentry = __d_lookup(parent, name);
2279                 if (dentry)
2280                         break;
2281         } while (read_seqretry(&rename_lock, seq));
2282         return dentry;
2283 }
2284 EXPORT_SYMBOL(d_lookup);
2285 
2286 /**
2287  * __d_lookup - search for a dentry (racy)
2288  * @parent: parent dentry
2289  * @name: qstr of name we wish to find
2290  * Returns: dentry, or NULL
2291  *
2292  * __d_lookup is like d_lookup, however it may (rarely) return a
2293  * false-negative result due to unrelated rename activity.
2294  *
2295  * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2296  * however it must be used carefully, eg. with a following d_lookup in
2297  * the case of failure.
2298  *
2299  * __d_lookup callers must be commented.
2300  */
2301 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2302 {
2303         unsigned int hash = name->hash;
2304         struct hlist_bl_head *b = d_hash(hash);
2305         struct hlist_bl_node *node;
2306         struct dentry *found = NULL;
2307         struct dentry *dentry;
2308 
2309         /*
2310          * Note: There is significant duplication with __d_lookup_rcu which is
2311          * required to prevent single threaded performance regressions
2312          * especially on architectures where smp_rmb (in seqcounts) are costly.
2313          * Keep the two functions in sync.
2314          */
2315 
2316         /*
2317          * The hash list is protected using RCU.
2318          *
2319          * Take d_lock when comparing a candidate dentry, to avoid races
2320          * with d_move().
2321          *
2322          * It is possible that concurrent renames can mess up our list
2323          * walk here and result in missing our dentry, resulting in the
2324          * false-negative result. d_lookup() protects against concurrent
2325          * renames using rename_lock seqlock.
2326          *
2327          * See Documentation/filesystems/path-lookup.txt for more details.
2328          */
2329         rcu_read_lock();
2330         
2331         hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2332 
2333                 if (dentry->d_name.hash != hash)
2334                         continue;
2335 
2336                 spin_lock(&dentry->d_lock);
2337                 if (dentry->d_parent != parent)
2338                         goto next;
2339                 if (d_unhashed(dentry))
2340                         goto next;
2341 
2342                 if (!d_same_name(dentry, parent, name))
2343                         goto next;
2344 
2345                 dentry->d_lockref.count++;
2346                 found = dentry;
2347                 spin_unlock(&dentry->d_lock);
2348                 break;
2349 next:
2350                 spin_unlock(&dentry->d_lock);
2351         }
2352         rcu_read_unlock();
2353 
2354         return found;
2355 }
2356 
2357 /**
2358  * d_hash_and_lookup - hash the qstr then search for a dentry
2359  * @dir: Directory to search in
2360  * @name: qstr of name we wish to find
2361  *
2362  * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2363  */
2364 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2365 {
2366         /*
2367          * Check for a fs-specific hash function. Note that we must
2368          * calculate the standard hash first, as the d_op->d_hash()
2369          * routine may choose to leave the hash value unchanged.
2370          */
2371         name->hash = full_name_hash(dir, name->name, name->len);
2372         if (dir->d_flags & DCACHE_OP_HASH) {
2373                 int err = dir->d_op->d_hash(dir, name);
2374                 if (unlikely(err < 0))
2375                         return ERR_PTR(err);
2376         }
2377         return d_lookup(dir, name);
2378 }
2379 EXPORT_SYMBOL(d_hash_and_lookup);
2380 
2381 /*
2382  * When a file is deleted, we have two options:
2383  * - turn this dentry into a negative dentry
2384  * - unhash this dentry and free it.
2385  *
2386  * Usually, we want to just turn this into
2387  * a negative dentry, but if anybody else is
2388  * currently using the dentry or the inode
2389  * we can't do that and we fall back on removing
2390  * it from the hash queues and waiting for
2391  * it to be deleted later when it has no users
2392  */
2393  
2394 /**
2395  * d_delete - delete a dentry
2396  * @dentry: The dentry to delete
2397  *
2398  * Turn the dentry into a negative dentry if possible, otherwise
2399  * remove it from the hash queues so it can be deleted later
2400  */
2401  
2402 void d_delete(struct dentry * dentry)
2403 {
2404         struct inode *inode;
2405         int isdir = 0;
2406         /*
2407          * Are we the only user?
2408          */
2409 again:
2410         spin_lock(&dentry->d_lock);
2411         inode = dentry->d_inode;
2412         isdir = S_ISDIR(inode->i_mode);
2413         if (dentry->d_lockref.count == 1) {
2414                 if (!spin_trylock(&inode->i_lock)) {
2415                         spin_unlock(&dentry->d_lock);
2416                         cpu_relax();
2417                         goto again;
2418                 }
2419                 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2420                 dentry_unlink_inode(dentry);
2421                 fsnotify_nameremove(dentry, isdir);
2422                 return;
2423         }
2424 
2425         if (!d_unhashed(dentry))
2426                 __d_drop(dentry);
2427 
2428         spin_unlock(&dentry->d_lock);
2429 
2430         fsnotify_nameremove(dentry, isdir);
2431 }
2432 EXPORT_SYMBOL(d_delete);
2433 
2434 static void __d_rehash(struct dentry *entry)
2435 {
2436         struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2437 
2438         hlist_bl_lock(b);
2439         hlist_bl_add_head_rcu(&entry->d_hash, b);
2440         hlist_bl_unlock(b);
2441 }
2442 
2443 /**
2444  * d_rehash     - add an entry back to the hash
2445  * @entry: dentry to add to the hash
2446  *
2447  * Adds a dentry to the hash according to its name.
2448  */
2449  
2450 void d_rehash(struct dentry * entry)
2451 {
2452         spin_lock(&entry->d_lock);
2453         __d_rehash(entry);
2454         spin_unlock(&entry->d_lock);
2455 }
2456 EXPORT_SYMBOL(d_rehash);
2457 
2458 static inline unsigned start_dir_add(struct inode *dir)
2459 {
2460 
2461         for (;;) {
2462                 unsigned n = dir->i_dir_seq;
2463                 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2464                         return n;
2465                 cpu_relax();
2466         }
2467 }
2468 
2469 static inline void end_dir_add(struct inode *dir, unsigned n)
2470 {
2471         smp_store_release(&dir->i_dir_seq, n + 2);
2472 }
2473 
2474 static void d_wait_lookup(struct dentry *dentry)
2475 {
2476         if (d_in_lookup(dentry)) {
2477                 DECLARE_WAITQUEUE(wait, current);
2478                 add_wait_queue(dentry->d_wait, &wait);
2479                 do {
2480                         set_current_state(TASK_UNINTERRUPTIBLE);
2481                         spin_unlock(&dentry->d_lock);
2482                         schedule();
2483                         spin_lock(&dentry->d_lock);
2484                 } while (d_in_lookup(dentry));
2485         }
2486 }
2487 
2488 struct dentry *d_alloc_parallel(struct dentry *parent,
2489                                 const struct qstr *name,
2490                                 wait_queue_head_t *wq)
2491 {
2492         unsigned int hash = name->hash;
2493         struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2494         struct hlist_bl_node *node;
2495         struct dentry *new = d_alloc(parent, name);
2496         struct dentry *dentry;
2497         unsigned seq, r_seq, d_seq;
2498 
2499         if (unlikely(!new))
2500                 return ERR_PTR(-ENOMEM);
2501 
2502 retry:
2503         rcu_read_lock();
2504         seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2505         r_seq = read_seqbegin(&rename_lock);
2506         dentry = __d_lookup_rcu(parent, name, &d_seq);
2507         if (unlikely(dentry)) {
2508                 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2509                         rcu_read_unlock();
2510                         goto retry;
2511                 }
2512                 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2513                         rcu_read_unlock();
2514                         dput(dentry);
2515                         goto retry;
2516                 }
2517                 rcu_read_unlock();
2518                 dput(new);
2519                 return dentry;
2520         }
2521         if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2522                 rcu_read_unlock();
2523                 goto retry;
2524         }
2525 
2526         if (unlikely(seq & 1)) {
2527                 rcu_read_unlock();
2528                 goto retry;
2529         }
2530 
2531         hlist_bl_lock(b);
2532         if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2533                 hlist_bl_unlock(b);
2534                 rcu_read_unlock();
2535                 goto retry;
2536         }
2537         /*
2538          * No changes for the parent since the beginning of d_lookup().
2539          * Since all removals from the chain happen with hlist_bl_lock(),
2540          * any potential in-lookup matches are going to stay here until
2541          * we unlock the chain.  All fields are stable in everything
2542          * we encounter.
2543          */
2544         hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2545                 if (dentry->d_name.hash != hash)
2546                         continue;
2547                 if (dentry->d_parent != parent)
2548                         continue;
2549                 if (!d_same_name(dentry, parent, name))
2550                         continue;
2551                 hlist_bl_unlock(b);
2552                 /* now we can try to grab a reference */
2553                 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2554                         rcu_read_unlock();
2555                         goto retry;
2556                 }
2557 
2558                 rcu_read_unlock();
2559                 /*
2560                  * somebody is likely to be still doing lookup for it;
2561                  * wait for them to finish
2562                  */
2563                 spin_lock(&dentry->d_lock);
2564                 d_wait_lookup(dentry);
2565                 /*
2566                  * it's not in-lookup anymore; in principle we should repeat
2567                  * everything from dcache lookup, but it's likely to be what
2568                  * d_lookup() would've found anyway.  If it is, just return it;
2569                  * otherwise we really have to repeat the whole thing.
2570                  */
2571                 if (unlikely(dentry->d_name.hash != hash))
2572                         goto mismatch;
2573                 if (unlikely(dentry->d_parent != parent))
2574                         goto mismatch;
2575                 if (unlikely(d_unhashed(dentry)))
2576                         goto mismatch;
2577                 if (unlikely(!d_same_name(dentry, parent, name)))
2578                         goto mismatch;
2579                 /* OK, it *is* a hashed match; return it */
2580                 spin_unlock(&dentry->d_lock);
2581                 dput(new);
2582                 return dentry;
2583         }
2584         rcu_read_unlock();
2585         /* we can't take ->d_lock here; it's OK, though. */
2586         new->d_flags |= DCACHE_PAR_LOOKUP;
2587         new->d_wait = wq;
2588         hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2589         hlist_bl_unlock(b);
2590         return new;
2591 mismatch:
2592         spin_unlock(&dentry->d_lock);
2593         dput(dentry);
2594         goto retry;
2595 }
2596 EXPORT_SYMBOL(d_alloc_parallel);
2597 
2598 void __d_lookup_done(struct dentry *dentry)
2599 {
2600         struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2601                                                  dentry->d_name.hash);
2602         hlist_bl_lock(b);
2603         dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2604         __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2605         wake_up_all(dentry->d_wait);
2606         dentry->d_wait = NULL;
2607         hlist_bl_unlock(b);
2608         INIT_HLIST_NODE(&dentry->d_u.d_alias);
2609         INIT_LIST_HEAD(&dentry->d_lru);
2610 }
2611 EXPORT_SYMBOL(__d_lookup_done);
2612 
2613 /* inode->i_lock held if inode is non-NULL */
2614 
2615 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2616 {
2617         struct inode *dir = NULL;
2618         unsigned n;
2619         spin_lock(&dentry->d_lock);
2620         if (unlikely(d_in_lookup(dentry))) {
2621                 dir = dentry->d_parent->d_inode;
2622                 n = start_dir_add(dir);
2623                 __d_lookup_done(dentry);
2624         }
2625         if (inode) {
2626                 unsigned add_flags = d_flags_for_inode(inode);
2627                 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2628                 raw_write_seqcount_begin(&dentry->d_seq);
2629                 __d_set_inode_and_type(dentry, inode, add_flags);
2630                 raw_write_seqcount_end(&dentry->d_seq);
2631                 fsnotify_update_flags(dentry);
2632         }
2633         __d_rehash(dentry);
2634         if (dir)
2635                 end_dir_add(dir, n);
2636         spin_unlock(&dentry->d_lock);
2637         if (inode)
2638                 spin_unlock(&inode->i_lock);
2639 }
2640 
2641 /**
2642  * d_add - add dentry to hash queues
2643  * @entry: dentry to add
2644  * @inode: The inode to attach to this dentry
2645  *
2646  * This adds the entry to the hash queues and initializes @inode.
2647  * The entry was actually filled in earlier during d_alloc().
2648  */
2649 
2650 void d_add(struct dentry *entry, struct inode *inode)
2651 {
2652         if (inode) {
2653                 security_d_instantiate(entry, inode);
2654                 spin_lock(&inode->i_lock);
2655         }
2656         __d_add(entry, inode);
2657 }
2658 EXPORT_SYMBOL(d_add);
2659 
2660 /**
2661  * d_exact_alias - find and hash an exact unhashed alias
2662  * @entry: dentry to add
2663  * @inode: The inode to go with this dentry
2664  *
2665  * If an unhashed dentry with the same name/parent and desired
2666  * inode already exists, hash and return it.  Otherwise, return
2667  * NULL.
2668  *
2669  * Parent directory should be locked.
2670  */
2671 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2672 {
2673         struct dentry *alias;
2674         unsigned int hash = entry->d_name.hash;
2675 
2676         spin_lock(&inode->i_lock);
2677         hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2678                 /*
2679                  * Don't need alias->d_lock here, because aliases with
2680                  * d_parent == entry->d_parent are not subject to name or
2681                  * parent changes, because the parent inode i_mutex is held.
2682                  */
2683                 if (alias->d_name.hash != hash)
2684                         continue;
2685                 if (alias->d_parent != entry->d_parent)
2686                         continue;
2687                 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2688                         continue;
2689                 spin_lock(&alias->d_lock);
2690                 if (!d_unhashed(alias)) {
2691                         spin_unlock(&alias->d_lock);
2692                         alias = NULL;
2693                 } else {
2694                         __dget_dlock(alias);
2695                         __d_rehash(alias);
2696                         spin_unlock(&alias->d_lock);
2697                 }
2698                 spin_unlock(&inode->i_lock);
2699                 return alias;
2700         }
2701         spin_unlock(&inode->i_lock);
2702         return NULL;
2703 }
2704 EXPORT_SYMBOL(d_exact_alias);
2705 
2706 /**
2707  * dentry_update_name_case - update case insensitive dentry with a new name
2708  * @dentry: dentry to be updated
2709  * @name: new name
2710  *
2711  * Update a case insensitive dentry with new case of name.
2712  *
2713  * dentry must have been returned by d_lookup with name @name. Old and new
2714  * name lengths must match (ie. no d_compare which allows mismatched name
2715  * lengths).
2716  *
2717  * Parent inode i_mutex must be held over d_lookup and into this call (to
2718  * keep renames and concurrent inserts, and readdir(2) away).
2719  */
2720 void dentry_update_name_case(struct dentry *dentry, const struct qstr *name)
2721 {
2722         BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
2723         BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2724 
2725         spin_lock(&dentry->d_lock);
2726         write_seqcount_begin(&dentry->d_seq);
2727         memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2728         write_seqcount_end(&dentry->d_seq);
2729         spin_unlock(&dentry->d_lock);
2730 }
2731 EXPORT_SYMBOL(dentry_update_name_case);
2732 
2733 static void swap_names(struct dentry *dentry, struct dentry *target)
2734 {
2735         if (unlikely(dname_external(target))) {
2736                 if (unlikely(dname_external(dentry))) {
2737                         /*
2738                          * Both external: swap the pointers
2739                          */
2740                         swap(target->d_name.name, dentry->d_name.name);
2741                 } else {
2742                         /*
2743                          * dentry:internal, target:external.  Steal target's
2744                          * storage and make target internal.
2745                          */
2746                         memcpy(target->d_iname, dentry->d_name.name,
2747                                         dentry->d_name.len + 1);
2748                         dentry->d_name.name = target->d_name.name;
2749                         target->d_name.name = target->d_iname;
2750                 }
2751         } else {
2752                 if (unlikely(dname_external(dentry))) {
2753                         /*
2754                          * dentry:external, target:internal.  Give dentry's
2755                          * storage to target and make dentry internal
2756                          */
2757                         memcpy(dentry->d_iname, target->d_name.name,
2758                                         target->d_name.len + 1);
2759                         target->d_name.name = dentry->d_name.name;
2760                         dentry->d_name.name = dentry->d_iname;
2761                 } else {
2762                         /*
2763                          * Both are internal.
2764                          */
2765                         unsigned int i;
2766                         BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2767                         for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2768                                 swap(((long *) &dentry->d_iname)[i],
2769                                      ((long *) &target->d_iname)[i]);
2770                         }
2771                 }
2772         }
2773         swap(dentry->d_name.hash_len, target->d_name.hash_len);
2774 }
2775 
2776 static void copy_name(struct dentry *dentry, struct dentry *target)
2777 {
2778         struct external_name *old_name = NULL;
2779         if (unlikely(dname_external(dentry)))
2780                 old_name = external_name(dentry);
2781         if (unlikely(dname_external(target))) {
2782                 atomic_inc(&external_name(target)->u.count);
2783                 dentry->d_name = target->d_name;
2784         } else {
2785                 memcpy(dentry->d_iname, target->d_name.name,
2786                                 target->d_name.len + 1);
2787                 dentry->d_name.name = dentry->d_iname;
2788                 dentry->d_name.hash_len = target->d_name.hash_len;
2789         }
2790         if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2791                 kfree_rcu(old_name, u.head);
2792 }
2793 
2794 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2795 {
2796         /*
2797          * XXXX: do we really need to take target->d_lock?
2798          */
2799         if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2800                 spin_lock(&target->d_parent->d_lock);
2801         else {
2802                 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2803                         spin_lock(&dentry->d_parent->d_lock);
2804                         spin_lock_nested(&target->d_parent->d_lock,
2805                                                 DENTRY_D_LOCK_NESTED);
2806                 } else {
2807                         spin_lock(&target->d_parent->d_lock);
2808                         spin_lock_nested(&dentry->d_parent->d_lock,
2809                                                 DENTRY_D_LOCK_NESTED);
2810                 }
2811         }
2812         if (target < dentry) {
2813                 spin_lock_nested(&target->d_lock, 2);
2814                 spin_lock_nested(&dentry->d_lock, 3);
2815         } else {
2816                 spin_lock_nested(&dentry->d_lock, 2);
2817                 spin_lock_nested(&target->d_lock, 3);
2818         }
2819 }
2820 
2821 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2822 {
2823         if (target->d_parent != dentry->d_parent)
2824                 spin_unlock(&dentry->d_parent->d_lock);
2825         if (target->d_parent != target)
2826                 spin_unlock(&target->d_parent->d_lock);
2827         spin_unlock(&target->d_lock);
2828         spin_unlock(&dentry->d_lock);
2829 }
2830 
2831 /*
2832  * When switching names, the actual string doesn't strictly have to
2833  * be preserved in the target - because we're dropping the target
2834  * anyway. As such, we can just do a simple memcpy() to copy over
2835  * the new name before we switch, unless we are going to rehash
2836  * it.  Note that if we *do* unhash the target, we are not allowed
2837  * to rehash it without giving it a new name/hash key - whether
2838  * we swap or overwrite the names here, resulting name won't match
2839  * the reality in filesystem; it's only there for d_path() purposes.
2840  * Note that all of this is happening under rename_lock, so the
2841  * any hash lookup seeing it in the middle of manipulations will
2842  * be discarded anyway.  So we do not care what happens to the hash
2843  * key in that case.
2844  */
2845 /*
2846  * __d_move - move a dentry
2847  * @dentry: entry to move
2848  * @target: new dentry
2849  * @exchange: exchange the two dentries
2850  *
2851  * Update the dcache to reflect the move of a file name. Negative
2852  * dcache entries should not be moved in this way. Caller must hold
2853  * rename_lock, the i_mutex of the source and target directories,
2854  * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2855  */
2856 static void __d_move(struct dentry *dentry, struct dentry *target,
2857                      bool exchange)
2858 {
2859         struct inode *dir = NULL;
2860         unsigned n;
2861         if (!dentry->d_inode)
2862                 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2863 
2864         BUG_ON(d_ancestor(dentry, target));
2865         BUG_ON(d_ancestor(target, dentry));
2866 
2867         dentry_lock_for_move(dentry, target);
2868         if (unlikely(d_in_lookup(target))) {
2869                 dir = target->d_parent->d_inode;
2870                 n = start_dir_add(dir);
2871                 __d_lookup_done(target);
2872         }
2873 
2874         write_seqcount_begin(&dentry->d_seq);
2875         write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2876 
2877         /* unhash both */
2878         /* ___d_drop does write_seqcount_barrier, but they're OK to nest. */
2879         ___d_drop(dentry);
2880         ___d_drop(target);
2881 
2882         /* Switch the names.. */
2883         if (exchange)
2884                 swap_names(dentry, target);
2885         else
2886                 copy_name(dentry, target);
2887 
2888         /* rehash in new place(s) */
2889         __d_rehash(dentry);
2890         if (exchange)
2891                 __d_rehash(target);
2892         else
2893                 target->d_hash.pprev = NULL;
2894 
2895         /* ... and switch them in the tree */
2896         if (IS_ROOT(dentry)) {
2897                 /* splicing a tree */
2898                 dentry->d_flags |= DCACHE_RCUACCESS;
2899                 dentry->d_parent = target->d_parent;
2900                 target->d_parent = target;
2901                 list_del_init(&target->d_child);
2902                 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2903         } else {
2904                 /* swapping two dentries */
2905                 swap(dentry->d_parent, target->d_parent);
2906                 list_move(&target->d_child, &target->d_parent->d_subdirs);
2907                 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2908                 if (exchange)
2909                         fsnotify_update_flags(target);
2910                 fsnotify_update_flags(dentry);
2911         }
2912 
2913         write_seqcount_end(&target->d_seq);
2914         write_seqcount_end(&dentry->d_seq);
2915 
2916         if (dir)
2917                 end_dir_add(dir, n);
2918         dentry_unlock_for_move(dentry, target);
2919 }
2920 
2921 /*
2922  * d_move - move a dentry
2923  * @dentry: entry to move
2924  * @target: new dentry
2925  *
2926  * Update the dcache to reflect the move of a file name. Negative
2927  * dcache entries should not be moved in this way. See the locking
2928  * requirements for __d_move.
2929  */
2930 void d_move(struct dentry *dentry, struct dentry *target)
2931 {
2932         write_seqlock(&rename_lock);
2933         __d_move(dentry, target, false);
2934         write_sequnlock(&rename_lock);
2935 }
2936 EXPORT_SYMBOL(d_move);
2937 
2938 /*
2939  * d_exchange - exchange two dentries
2940  * @dentry1: first dentry
2941  * @dentry2: second dentry
2942  */
2943 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2944 {
2945         write_seqlock(&rename_lock);
2946 
2947         WARN_ON(!dentry1->d_inode);
2948         WARN_ON(!dentry2->d_inode);
2949         WARN_ON(IS_ROOT(dentry1));
2950         WARN_ON(IS_ROOT(dentry2));
2951 
2952         __d_move(dentry1, dentry2, true);
2953 
2954         write_sequnlock(&rename_lock);
2955 }
2956 
2957 /**
2958  * d_ancestor - search for an ancestor
2959  * @p1: ancestor dentry
2960  * @p2: child dentry
2961  *
2962  * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2963  * an ancestor of p2, else NULL.
2964  */
2965 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2966 {
2967         struct dentry *p;
2968 
2969         for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2970                 if (p->d_parent == p1)
2971                         return p;
2972         }
2973         return NULL;
2974 }
2975 
2976 /*
2977  * This helper attempts to cope with remotely renamed directories
2978  *
2979  * It assumes that the caller is already holding
2980  * dentry->d_parent->d_inode->i_mutex, and rename_lock
2981  *
2982  * Note: If ever the locking in lock_rename() changes, then please
2983  * remember to update this too...
2984  */
2985 static int __d_unalias(struct inode *inode,
2986                 struct dentry *dentry, struct dentry *alias)
2987 {
2988         struct mutex *m1 = NULL;
2989         struct rw_semaphore *m2 = NULL;
2990         int ret = -ESTALE;
2991 
2992         /* If alias and dentry share a parent, then no extra locks required */
2993         if (alias->d_parent == dentry->d_parent)
2994                 goto out_unalias;
2995 
2996         /* See lock_rename() */
2997         if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2998                 goto out_err;
2999         m1 = &dentry->d_sb->s_vfs_rename_mutex;
3000         if (!inode_trylock_shared(alias->d_parent->d_inode))
3001                 goto out_err;
3002         m2 = &alias->d_parent->d_inode->i_rwsem;
3003 out_unalias:
3004         __d_move(alias, dentry, false);
3005         ret = 0;
3006 out_err:
3007         if (m2)
3008                 up_read(m2);
3009         if (m1)
3010                 mutex_unlock(m1);
3011         return ret;
3012 }
3013 
3014 /**
3015  * d_splice_alias - splice a disconnected dentry into the tree if one exists
3016  * @inode:  the inode which may have a disconnected dentry
3017  * @dentry: a negative dentry which we want to point to the inode.
3018  *
3019  * If inode is a directory and has an IS_ROOT alias, then d_move that in
3020  * place of the given dentry and return it, else simply d_add the inode
3021  * to the dentry and return NULL.
3022  *
3023  * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3024  * we should error out: directories can't have multiple aliases.
3025  *
3026  * This is needed in the lookup routine of any filesystem that is exportable
3027  * (via knfsd) so that we can build dcache paths to directories effectively.
3028  *
3029  * If a dentry was found and moved, then it is returned.  Otherwise NULL
3030  * is returned.  This matches the expected return value of ->lookup.
3031  *
3032  * Cluster filesystems may call this function with a negative, hashed dentry.
3033  * In that case, we know that the inode will be a regular file, and also this
3034  * will only occur during atomic_open. So we need to check for the dentry
3035  * being already hashed only in the final case.
3036  */
3037 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3038 {
3039         if (IS_ERR(inode))
3040                 return ERR_CAST(inode);
3041 
3042         BUG_ON(!d_unhashed(dentry));
3043 
3044         if (!inode)
3045                 goto out;
3046 
3047         security_d_instantiate(dentry, inode);
3048         spin_lock(&inode->i_lock);
3049         if (S_ISDIR(inode->i_mode)) {
3050                 struct dentry *new = __d_find_any_alias(inode);
3051                 if (unlikely(new)) {
3052                         /* The reference to new ensures it remains an alias */
3053                         spin_unlock(&inode->i_lock);
3054                         write_seqlock(&rename_lock);
3055                         if (unlikely(d_ancestor(new, dentry))) {
3056                                 write_sequnlock(&rename_lock);
3057                                 dput(new);
3058                                 new = ERR_PTR(-ELOOP);
3059                                 pr_warn_ratelimited(
3060                                         "VFS: Lookup of '%s' in %s %s"
3061                                         " would have caused loop\n",
3062                                         dentry->d_name.name,
3063                                         inode->i_sb->s_type->name,
3064                                         inode->i_sb->s_id);
3065                         } else if (!IS_ROOT(new)) {
3066                                 int err = __d_unalias(inode, dentry, new);
3067                                 write_sequnlock(&rename_lock);
3068                                 if (err) {
3069                                         dput(new);
3070                                         new = ERR_PTR(err);
3071                                 }
3072                         } else {
3073                                 __d_move(new, dentry, false);
3074                                 write_sequnlock(&rename_lock);
3075                         }
3076                         iput(inode);
3077                         return new;
3078                 }
3079         }
3080 out:
3081         __d_add(dentry, inode);
3082         return NULL;
3083 }
3084 EXPORT_SYMBOL(d_splice_alias);
3085 
3086 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
3087 {
3088         *buflen -= namelen;
3089         if (*buflen < 0)
3090                 return -ENAMETOOLONG;
3091         *buffer -= namelen;
3092         memcpy(*buffer, str, namelen);
3093         return 0;
3094 }
3095 
3096 /**
3097  * prepend_name - prepend a pathname in front of current buffer pointer
3098  * @buffer: buffer pointer
3099  * @buflen: allocated length of the buffer
3100  * @name:   name string and length qstr structure
3101  *
3102  * With RCU path tracing, it may race with d_move(). Use READ_ONCE() to
3103  * make sure that either the old or the new name pointer and length are
3104  * fetched. However, there may be mismatch between length and pointer.
3105  * The length cannot be trusted, we need to copy it byte-by-byte until
3106  * the length is reached or a null byte is found. It also prepends "/" at
3107  * the beginning of the name. The sequence number check at the caller will
3108  * retry it again when a d_move() does happen. So any garbage in the buffer
3109  * due to mismatched pointer and length will be discarded.
3110  *
3111  * Load acquire is needed to make sure that we see that terminating NUL.
3112  */
3113 static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
3114 {
3115         const char *dname = smp_load_acquire(&name->name); /* ^^^ */
3116         u32 dlen = READ_ONCE(name->len);
3117         char *p;
3118 
3119         *buflen -= dlen + 1;
3120         if (*buflen < 0)
3121                 return -ENAMETOOLONG;
3122         p = *buffer -= dlen + 1;
3123         *p++ = '/';
3124         while (dlen--) {
3125                 char c = *dname++;
3126                 if (!c)
3127                         break;
3128                 *p++ = c;
3129         }
3130         return 0;
3131 }
3132 
3133 /**
3134  * prepend_path - Prepend path string to a buffer
3135  * @path: the dentry/vfsmount to report
3136  * @root: root vfsmnt/dentry
3137  * @buffer: pointer to the end of the buffer
3138  * @buflen: pointer to buffer length
3139  *
3140  * The function will first try to write out the pathname without taking any
3141  * lock other than the RCU read lock to make sure that dentries won't go away.
3142  * It only checks the sequence number of the global rename_lock as any change
3143  * in the dentry's d_seq will be preceded by changes in the rename_lock
3144  * sequence number. If the sequence number had been changed, it will restart
3145  * the whole pathname back-tracing sequence again by taking the rename_lock.
3146  * In this case, there is no need to take the RCU read lock as the recursive
3147  * parent pointer references will keep the dentry chain alive as long as no
3148  * rename operation is performed.
3149  */
3150 static int prepend_path(const struct path *path,
3151                         const struct path *root,
3152                         char **buffer, int *buflen)
3153 {
3154         struct dentry *dentry;
3155         struct vfsmount *vfsmnt;
3156         struct mount *mnt;
3157         int error = 0;
3158         unsigned seq, m_seq = 0;
3159         char *bptr;
3160         int blen;
3161 
3162         rcu_read_lock();
3163 restart_mnt:
3164         read_seqbegin_or_lock(&mount_lock, &m_seq);
3165         seq = 0;
3166         rcu_read_lock();
3167 restart:
3168         bptr = *buffer;
3169         blen = *buflen;
3170         error = 0;
3171         dentry = path->dentry;
3172         vfsmnt = path->mnt;
3173         mnt = real_mount(vfsmnt);
3174         read_seqbegin_or_lock(&rename_lock, &seq);
3175         while (dentry != root->dentry || vfsmnt != root->mnt) {
3176                 struct dentry * parent;
3177 
3178                 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
3179                         struct mount *parent = READ_ONCE(mnt->mnt_parent);
3180                         /* Escaped? */
3181                         if (dentry != vfsmnt->mnt_root) {
3182                                 bptr = *buffer;
3183                                 blen = *buflen;
3184                                 error = 3;
3185                                 break;
3186                         }
3187                         /* Global root? */
3188                         if (mnt != parent) {
3189                                 dentry = READ_ONCE(mnt->mnt_mountpoint);
3190                                 mnt = parent;
3191                                 vfsmnt = &mnt->mnt;
3192                                 continue;
3193                         }
3194                         if (!error)
3195                                 error = is_mounted(vfsmnt) ? 1 : 2;
3196                         break;
3197                 }
3198                 parent = dentry->d_parent;
3199                 prefetch(parent);
3200                 error = prepend_name(&bptr, &blen, &dentry->d_name);
3201                 if (error)
3202                         break;
3203 
3204                 dentry = parent;
3205         }
3206         if (!(seq & 1))
3207                 rcu_read_unlock();
3208         if (need_seqretry(&rename_lock, seq)) {
3209                 seq = 1;
3210                 goto restart;
3211         }
3212         done_seqretry(&rename_lock, seq);
3213 
3214         if (!(m_seq & 1))
3215                 rcu_read_unlock();
3216         if (need_seqretry(&mount_lock, m_seq)) {
3217                 m_seq = 1;
3218                 goto restart_mnt;
3219         }
3220         done_seqretry(&mount_lock, m_seq);
3221 
3222         if (error >= 0 && bptr == *buffer) {
3223                 if (--blen < 0)
3224                         error = -ENAMETOOLONG;
3225                 else
3226                         *--bptr = '/';
3227         }
3228         *buffer = bptr;
3229         *buflen = blen;
3230         return error;
3231 }
3232 
3233 /**
3234  * __d_path - return the path of a dentry
3235  * @path: the dentry/vfsmount to report
3236  * @root: root vfsmnt/dentry
3237  * @buf: buffer to return value in
3238  * @buflen: buffer length
3239  *
3240  * Convert a dentry into an ASCII path name.
3241  *
3242  * Returns a pointer into the buffer or an error code if the
3243  * path was too long.
3244  *
3245  * "buflen" should be positive.
3246  *
3247  * If the path is not reachable from the supplied root, return %NULL.
3248  */
3249 char *__d_path(const struct path *path,
3250                const struct path *root,
3251                char *buf, int buflen)
3252 {
3253         char *res = buf + buflen;
3254         int error;
3255 
3256         prepend(&res, &buflen, "\0", 1);
3257         error = prepend_path(path, root, &res, &buflen);
3258 
3259         if (error < 0)
3260                 return ERR_PTR(error);
3261         if (error > 0)
3262                 return NULL;
3263         return res;
3264 }
3265 
3266 char *d_absolute_path(const struct path *path,
3267                char *buf, int buflen)
3268 {
3269         struct path root = {};
3270         char *res = buf + buflen;
3271         int error;
3272 
3273         prepend(&res, &buflen, "\0", 1);
3274         error = prepend_path(path, &root, &res, &buflen);
3275 
3276         if (error > 1)
3277                 error = -EINVAL;
3278         if (error < 0)
3279                 return ERR_PTR(error);
3280         return res;
3281 }
3282 
3283 /*
3284  * same as __d_path but appends "(deleted)" for unlinked files.
3285  */
3286 static int path_with_deleted(const struct path *path,
3287                              const struct path *root,
3288                              char **buf, int *buflen)
3289 {
3290         prepend(buf, buflen, "\0", 1);
3291         if (d_unlinked(path->dentry)) {
3292                 int error = prepend(buf, buflen, " (deleted)", 10);
3293                 if (error)
3294                         return error;
3295         }
3296 
3297         return prepend_path(path, root, buf, buflen);
3298 }
3299 
3300 static int prepend_unreachable(char **buffer, int *buflen)
3301 {
3302         return prepend(buffer, buflen, "(unreachable)", 13);
3303 }
3304 
3305 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3306 {
3307         unsigned seq;
3308 
3309         do {
3310                 seq = read_seqcount_begin(&fs->seq);
3311                 *root = fs->root;
3312         } while (read_seqcount_retry(&fs->seq, seq));
3313 }
3314 
3315 /**
3316  * d_path - return the path of a dentry
3317  * @path: path to report
3318  * @buf: buffer to return value in
3319  * @buflen: buffer length
3320  *
3321  * Convert a dentry into an ASCII path name. If the entry has been deleted
3322  * the string " (deleted)" is appended. Note that this is ambiguous.
3323  *
3324  * Returns a pointer into the buffer or an error code if the path was
3325  * too long. Note: Callers should use the returned pointer, not the passed
3326  * in buffer, to use the name! The implementation often starts at an offset
3327  * into the buffer, and may leave 0 bytes at the start.
3328  *
3329  * "buflen" should be positive.
3330  */
3331 char *d_path(const struct path *path, char *buf, int buflen)
3332 {
3333         char *res = buf + buflen;
3334         struct path root;
3335         int error;
3336 
3337         /*
3338          * We have various synthetic filesystems that never get mounted.  On
3339          * these filesystems dentries are never used for lookup purposes, and
3340          * thus don't need to be hashed.  They also don't need a name until a
3341          * user wants to identify the object in /proc/pid/fd/.  The little hack
3342          * below allows us to generate a name for these objects on demand:
3343          *
3344          * Some pseudo inodes are mountable.  When they are mounted
3345          * path->dentry == path->mnt->mnt_root.  In that case don't call d_dname
3346          * and instead have d_path return the mounted path.
3347          */
3348         if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3349             (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3350                 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3351 
3352         rcu_read_lock();
3353         get_fs_root_rcu(current->fs, &root);
3354         error = path_with_deleted(path, &root, &res, &buflen);
3355         rcu_read_unlock();
3356 
3357         if (error < 0)
3358                 res = ERR_PTR(error);
3359         return res;
3360 }
3361 EXPORT_SYMBOL(d_path);
3362 
3363 /*
3364  * Helper function for dentry_operations.d_dname() members
3365  */
3366 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3367                         const char *fmt, ...)
3368 {
3369         va_list args;
3370         char temp[64];
3371         int sz;
3372 
3373         va_start(args, fmt);
3374         sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3375         va_end(args);
3376 
3377         if (sz > sizeof(temp) || sz > buflen)
3378                 return ERR_PTR(-ENAMETOOLONG);
3379 
3380         buffer += buflen - sz;
3381         return memcpy(buffer, temp, sz);
3382 }
3383 
3384 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3385 {
3386         char *end = buffer + buflen;
3387         /* these dentries are never renamed, so d_lock is not needed */
3388         if (prepend(&end, &buflen, " (deleted)", 11) ||
3389             prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3390             prepend(&end, &buflen, "/", 1))  
3391                 end = ERR_PTR(-ENAMETOOLONG);
3392         return end;
3393 }
3394 EXPORT_SYMBOL(simple_dname);
3395 
3396 /*
3397  * Write full pathname from the root of the filesystem into the buffer.
3398  */
3399 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3400 {
3401         struct dentry *dentry;
3402         char *end, *retval;
3403         int len, seq = 0;
3404         int error = 0;
3405 
3406         if (buflen < 2)
3407                 goto Elong;
3408 
3409         rcu_read_lock();
3410 restart:
3411         dentry = d;
3412         end = buf + buflen;
3413         len = buflen;
3414         prepend(&end, &len, "\0", 1);
3415         /* Get '/' right */
3416         retval = end-1;
3417         *retval = '/';
3418         read_seqbegin_or_lock(&rename_lock, &seq);
3419         while (!IS_ROOT(dentry)) {
3420                 struct dentry *parent = dentry->d_parent;
3421 
3422                 prefetch(parent);
3423                 error = prepend_name(&end, &len, &dentry->d_name);
3424                 if (error)
3425                         break;
3426 
3427                 retval = end;
3428                 dentry = parent;
3429         }
3430         if (!(seq & 1))
3431                 rcu_read_unlock();
3432         if (need_seqretry(&rename_lock, seq)) {
3433                 seq = 1;
3434                 goto restart;
3435         }
3436         done_seqretry(&rename_lock, seq);
3437         if (error)
3438                 goto Elong;
3439         return retval;
3440 Elong:
3441         return ERR_PTR(-ENAMETOOLONG);
3442 }
3443 
3444 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3445 {
3446         return __dentry_path(dentry, buf, buflen);
3447 }
3448 EXPORT_SYMBOL(dentry_path_raw);
3449 
3450 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3451 {
3452         char *p = NULL;
3453         char *retval;
3454 
3455         if (d_unlinked(dentry)) {
3456                 p = buf + buflen;
3457                 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3458                         goto Elong;
3459                 buflen++;
3460         }
3461         retval = __dentry_path(dentry, buf, buflen);
3462         if (!IS_ERR(retval) && p)
3463                 *p = '/';       /* restore '/' overriden with '\0' */
3464         return retval;
3465 Elong:
3466         return ERR_PTR(-ENAMETOOLONG);
3467 }
3468 
3469 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3470                                     struct path *pwd)
3471 {
3472         unsigned seq;
3473 
3474         do {
3475                 seq = read_seqcount_begin(&fs->seq);
3476                 *root = fs->root;
3477                 *pwd = fs->pwd;
3478         } while (read_seqcount_retry(&fs->seq, seq));
3479 }
3480 
3481 /*
3482  * NOTE! The user-level library version returns a
3483  * character pointer. The kernel system call just
3484  * returns the length of the buffer filled (which
3485  * includes the ending '\0' character), or a negative
3486  * error value. So libc would do something like
3487  *
3488  *      char *getcwd(char * buf, size_t size)
3489  *      {
3490  *              int retval;
3491  *
3492  *              retval = sys_getcwd(buf, size);
3493  *              if (retval >= 0)
3494  *                      return buf;
3495  *              errno = -retval;
3496  *              return NULL;
3497  *      }
3498  */
3499 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3500 {
3501         int error;
3502         struct path pwd, root;
3503         char *page = __getname();
3504 
3505         if (!page)
3506                 return -ENOMEM;
3507 
3508         rcu_read_lock();
3509         get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3510 
3511         error = -ENOENT;
3512         if (!d_unlinked(pwd.dentry)) {
3513                 unsigned long len;
3514                 char *cwd = page + PATH_MAX;
3515                 int buflen = PATH_MAX;
3516 
3517                 prepend(&cwd, &buflen, "\0", 1);
3518                 error = prepend_path(&pwd, &root, &cwd, &buflen);
3519                 rcu_read_unlock();
3520 
3521                 if (error < 0)
3522                         goto out;
3523 
3524                 /* Unreachable from current root */
3525                 if (error > 0) {
3526                         error = prepend_unreachable(&cwd, &buflen);
3527                         if (error)
3528                                 goto out;
3529                 }
3530 
3531                 error = -ERANGE;
3532                 len = PATH_MAX + page - cwd;
3533                 if (len <= size) {
3534                         error = len;
3535                         if (copy_to_user(buf, cwd, len))
3536                                 error = -EFAULT;
3537                 }
3538         } else {
3539                 rcu_read_unlock();
3540         }
3541 
3542 out:
3543         __putname(page);
3544         return error;
3545 }
3546 
3547 /*
3548  * Test whether new_dentry is a subdirectory of old_dentry.
3549  *
3550  * Trivially implemented using the dcache structure
3551  */
3552 
3553 /**
3554  * is_subdir - is new dentry a subdirectory of old_dentry
3555  * @new_dentry: new dentry
3556  * @old_dentry: old dentry
3557  *
3558  * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3559  * Returns false otherwise.
3560  * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3561  */
3562   
3563 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3564 {
3565         bool result;
3566         unsigned seq;
3567 
3568         if (new_dentry == old_dentry)
3569                 return true;
3570 
3571         do {
3572                 /* for restarting inner loop in case of seq retry */
3573                 seq = read_seqbegin(&rename_lock);
3574                 /*
3575                  * Need rcu_readlock to protect against the d_parent trashing
3576                  * due to d_move
3577                  */
3578                 rcu_read_lock();
3579                 if (d_ancestor(old_dentry, new_dentry))
3580                         result = true;
3581                 else
3582                         result = false;
3583                 rcu_read_unlock();
3584         } while (read_seqretry(&rename_lock, seq));
3585 
3586         return result;
3587 }
3588 EXPORT_SYMBOL(is_subdir);
3589 
3590 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3591 {
3592         struct dentry *root = data;
3593         if (dentry != root) {
3594                 if (d_unhashed(dentry) || !dentry->d_inode)
3595                         return D_WALK_SKIP;
3596 
3597                 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3598                         dentry->d_flags |= DCACHE_GENOCIDE;
3599                         dentry->d_lockref.count--;
3600                 }
3601         }
3602         return D_WALK_CONTINUE;
3603 }
3604 
3605 void d_genocide(struct dentry *parent)
3606 {
3607         d_walk(parent, parent, d_genocide_kill, NULL);
3608 }
3609 
3610 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3611 {
3612         inode_dec_link_count(inode);
3613         BUG_ON(dentry->d_name.name != dentry->d_iname ||
3614                 !hlist_unhashed(&dentry->d_u.d_alias) ||
3615                 !d_unlinked(dentry));
3616         spin_lock(&dentry->d_parent->d_lock);
3617         spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3618         dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3619                                 (unsigned long long)inode->i_ino);
3620         spin_unlock(&dentry->d_lock);
3621         spin_unlock(&dentry->d_parent->d_lock);
3622         d_instantiate(dentry, inode);
3623 }
3624 EXPORT_SYMBOL(d_tmpfile);
3625 
3626 static __initdata unsigned long dhash_entries;
3627 static int __init set_dhash_entries(char *str)
3628 {
3629         if (!str)
3630                 return 0;
3631         dhash_entries = simple_strtoul(str, &str, 0);
3632         return 1;
3633 }
3634 __setup("dhash_entries=", set_dhash_entries);
3635 
3636 static void __init dcache_init_early(void)
3637 {
3638         /* If hashes are distributed across NUMA nodes, defer
3639          * hash allocation until vmalloc space is available.
3640          */
3641         if (hashdist)
3642                 return;
3643 
3644         dentry_hashtable =
3645                 alloc_large_system_hash("Dentry cache",
3646                                         sizeof(struct hlist_bl_head),
3647                                         dhash_entries,
3648                                         13,
3649                                         HASH_EARLY | HASH_ZERO,
3650                                         &d_hash_shift,
3651                                         NULL,
3652                                         0,
3653                                         0);
3654         d_hash_shift = 32 - d_hash_shift;
3655 }
3656 
3657 static void __init dcache_init(void)
3658 {
3659         /*
3660          * A constructor could be added for stable state like the lists,
3661          * but it is probably not worth it because of the cache nature
3662          * of the dcache.
3663          */
3664         dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3665                 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT,
3666                 d_iname);
3667 
3668         /* Hash may have been set up in dcache_init_early */
3669         if (!hashdist)
3670                 return;
3671 
3672         dentry_hashtable =
3673                 alloc_large_system_hash("Dentry cache",
3674                                         sizeof(struct hlist_bl_head),
3675                                         dhash_entries,
3676                                         13,
3677                                         HASH_ZERO,
3678                                         &d_hash_shift,
3679                                         NULL,
3680                                         0,
3681                                         0);
3682         d_hash_shift = 32 - d_hash_shift;
3683 }
3684 
3685 /* SLAB cache for __getname() consumers */
3686 struct kmem_cache *names_cachep __read_mostly;
3687 EXPORT_SYMBOL(names_cachep);
3688 
3689 EXPORT_SYMBOL(d_genocide);
3690 
3691 void __init vfs_caches_init_early(void)
3692 {
3693         int i;
3694 
3695         for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3696                 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3697 
3698         dcache_init_early();
3699         inode_init_early();
3700 }
3701 
3702 void __init vfs_caches_init(void)
3703 {
3704         names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3705                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
3706 
3707         dcache_init();
3708         inode_init();
3709         files_init();
3710         files_maxfiles_init();
3711         mnt_init();
3712         bdev_cache_init();
3713         chrdev_init();
3714 }
3715 

~ [ 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