]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/dcache.c
parallel lookups machinery, part 3
[mirror_ubuntu-bionic-kernel.git] / fs / dcache.c
CommitLineData
1da177e4
LT
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
1da177e4
LT
17#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
7a91bf7f 21#include <linux/fsnotify.h>
1da177e4
LT
22#include <linux/slab.h>
23#include <linux/init.h>
1da177e4
LT
24#include <linux/hash.h>
25#include <linux/cache.h>
630d9c47 26#include <linux/export.h>
1da177e4
LT
27#include <linux/mount.h>
28#include <linux/file.h>
29#include <asm/uaccess.h>
30#include <linux/security.h>
31#include <linux/seqlock.h>
32#include <linux/swap.h>
33#include <linux/bootmem.h>
5ad4e53b 34#include <linux/fs_struct.h>
613afbf8 35#include <linux/hardirq.h>
ceb5bdc2
NP
36#include <linux/bit_spinlock.h>
37#include <linux/rculist_bl.h>
268bb0ce 38#include <linux/prefetch.h>
dd179946 39#include <linux/ratelimit.h>
f6041567 40#include <linux/list_lru.h>
df4c0e36
AR
41#include <linux/kasan.h>
42
07f3f05c 43#include "internal.h"
b2dba1af 44#include "mount.h"
1da177e4 45
789680d1
NP
46/*
47 * Usage:
873feea0 48 * dcache->d_inode->i_lock protects:
946e51f2 49 * - i_dentry, d_u.d_alias, d_inode of aliases
ceb5bdc2
NP
50 * dcache_hash_bucket lock protects:
51 * - the dcache hash table
52 * s_anon bl list spinlock protects:
53 * - the s_anon list (see __d_drop)
19156840 54 * dentry->d_sb->s_dentry_lru_lock protects:
23044507
NP
55 * - the dcache lru lists and counters
56 * d_lock protects:
57 * - d_flags
58 * - d_name
59 * - d_lru
b7ab39f6 60 * - d_count
da502956 61 * - d_unhashed()
2fd6b7f5
NP
62 * - d_parent and d_subdirs
63 * - childrens' d_child and d_parent
946e51f2 64 * - d_u.d_alias, d_inode
789680d1
NP
65 *
66 * Ordering:
873feea0 67 * dentry->d_inode->i_lock
b5c84bf6 68 * dentry->d_lock
19156840 69 * dentry->d_sb->s_dentry_lru_lock
ceb5bdc2
NP
70 * dcache_hash_bucket lock
71 * s_anon lock
789680d1 72 *
da502956
NP
73 * If there is an ancestor relationship:
74 * dentry->d_parent->...->d_parent->d_lock
75 * ...
76 * dentry->d_parent->d_lock
77 * dentry->d_lock
78 *
79 * If no ancestor relationship:
789680d1
NP
80 * if (dentry1 < dentry2)
81 * dentry1->d_lock
82 * dentry2->d_lock
83 */
fa3536cc 84int sysctl_vfs_cache_pressure __read_mostly = 100;
1da177e4
LT
85EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
86
74c3cbe3 87__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
1da177e4 88
949854d0 89EXPORT_SYMBOL(rename_lock);
1da177e4 90
e18b890b 91static struct kmem_cache *dentry_cache __read_mostly;
1da177e4 92
1da177e4
LT
93/*
94 * This is the single most critical data structure when it comes
95 * to the dcache: the hashtable for lookups. Somebody should try
96 * to make this good - I've just made it work.
97 *
98 * This hash-function tries to avoid losing too many bits of hash
99 * information, yet avoid using a prime hash-size or similar.
100 */
1da177e4 101
fa3536cc
ED
102static unsigned int d_hash_mask __read_mostly;
103static unsigned int d_hash_shift __read_mostly;
ceb5bdc2 104
b07ad996 105static struct hlist_bl_head *dentry_hashtable __read_mostly;
ceb5bdc2 106
8966be90 107static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
6d7d1a0d 108 unsigned int hash)
ceb5bdc2 109{
6d7d1a0d 110 hash += (unsigned long) parent / L1_CACHE_BYTES;
99d263d4 111 return dentry_hashtable + hash_32(hash, d_hash_shift);
ceb5bdc2
NP
112}
113
94bdd655
AV
114#define IN_LOOKUP_SHIFT 10
115static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
116
117static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
118 unsigned int hash)
119{
120 hash += (unsigned long) parent / L1_CACHE_BYTES;
121 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
122}
123
124
1da177e4
LT
125/* Statistics gathering. */
126struct dentry_stat_t dentry_stat = {
127 .age_limit = 45,
128};
129
3942c07c 130static DEFINE_PER_CPU(long, nr_dentry);
62d36c77 131static DEFINE_PER_CPU(long, nr_dentry_unused);
312d3ca8
CH
132
133#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
62d36c77
DC
134
135/*
136 * Here we resort to our own counters instead of using generic per-cpu counters
137 * for consistency with what the vfs inode code does. We are expected to harvest
138 * better code and performance by having our own specialized counters.
139 *
140 * Please note that the loop is done over all possible CPUs, not over all online
141 * CPUs. The reason for this is that we don't want to play games with CPUs going
142 * on and off. If one of them goes off, we will just keep their counters.
143 *
144 * glommer: See cffbc8a for details, and if you ever intend to change this,
145 * please update all vfs counters to match.
146 */
3942c07c 147static long get_nr_dentry(void)
3e880fb5
NP
148{
149 int i;
3942c07c 150 long sum = 0;
3e880fb5
NP
151 for_each_possible_cpu(i)
152 sum += per_cpu(nr_dentry, i);
153 return sum < 0 ? 0 : sum;
154}
155
62d36c77
DC
156static long get_nr_dentry_unused(void)
157{
158 int i;
159 long sum = 0;
160 for_each_possible_cpu(i)
161 sum += per_cpu(nr_dentry_unused, i);
162 return sum < 0 ? 0 : sum;
163}
164
1f7e0616 165int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
312d3ca8
CH
166 size_t *lenp, loff_t *ppos)
167{
3e880fb5 168 dentry_stat.nr_dentry = get_nr_dentry();
62d36c77 169 dentry_stat.nr_unused = get_nr_dentry_unused();
3942c07c 170 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
312d3ca8
CH
171}
172#endif
173
5483f18e
LT
174/*
175 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
176 * The strings are both count bytes long, and count is non-zero.
177 */
e419b4cc
LT
178#ifdef CONFIG_DCACHE_WORD_ACCESS
179
180#include <asm/word-at-a-time.h>
181/*
182 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
183 * aligned allocation for this particular component. We don't
184 * strictly need the load_unaligned_zeropad() safety, but it
185 * doesn't hurt either.
186 *
187 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
188 * need the careful unaligned handling.
189 */
94753db5 190static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
5483f18e 191{
bfcfaa77 192 unsigned long a,b,mask;
bfcfaa77
LT
193
194 for (;;) {
12f8ad4b 195 a = *(unsigned long *)cs;
e419b4cc 196 b = load_unaligned_zeropad(ct);
bfcfaa77
LT
197 if (tcount < sizeof(unsigned long))
198 break;
199 if (unlikely(a != b))
200 return 1;
201 cs += sizeof(unsigned long);
202 ct += sizeof(unsigned long);
203 tcount -= sizeof(unsigned long);
204 if (!tcount)
205 return 0;
206 }
a5c21dce 207 mask = bytemask_from_count(tcount);
bfcfaa77 208 return unlikely(!!((a ^ b) & mask));
e419b4cc
LT
209}
210
bfcfaa77 211#else
e419b4cc 212
94753db5 213static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
e419b4cc 214{
5483f18e
LT
215 do {
216 if (*cs != *ct)
217 return 1;
218 cs++;
219 ct++;
220 tcount--;
221 } while (tcount);
222 return 0;
223}
224
e419b4cc
LT
225#endif
226
94753db5
LT
227static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
228{
6326c71f 229 const unsigned char *cs;
94753db5
LT
230 /*
231 * Be careful about RCU walk racing with rename:
232 * use ACCESS_ONCE to fetch the name pointer.
233 *
234 * NOTE! Even if a rename will mean that the length
235 * was not loaded atomically, we don't care. The
236 * RCU walk will check the sequence count eventually,
237 * and catch it. And we won't overrun the buffer,
238 * because we're reading the name pointer atomically,
239 * and a dentry name is guaranteed to be properly
240 * terminated with a NUL byte.
241 *
242 * End result: even if 'len' is wrong, we'll exit
243 * early because the data cannot match (there can
244 * be no NUL in the ct/tcount data)
245 */
6326c71f
LT
246 cs = ACCESS_ONCE(dentry->d_name.name);
247 smp_read_barrier_depends();
248 return dentry_string_cmp(cs, ct, tcount);
94753db5
LT
249}
250
8d85b484
AV
251struct external_name {
252 union {
253 atomic_t count;
254 struct rcu_head head;
255 } u;
256 unsigned char name[];
257};
258
259static inline struct external_name *external_name(struct dentry *dentry)
260{
261 return container_of(dentry->d_name.name, struct external_name, name[0]);
262}
263
9c82ab9c 264static void __d_free(struct rcu_head *head)
1da177e4 265{
9c82ab9c
CH
266 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
267
8d85b484
AV
268 kmem_cache_free(dentry_cache, dentry);
269}
270
271static void __d_free_external(struct rcu_head *head)
272{
273 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
8d85b484 274 kfree(external_name(dentry));
1da177e4
LT
275 kmem_cache_free(dentry_cache, dentry);
276}
277
810bb172
AV
278static inline int dname_external(const struct dentry *dentry)
279{
280 return dentry->d_name.name != dentry->d_iname;
281}
282
4bf46a27
DH
283static inline void __d_set_inode_and_type(struct dentry *dentry,
284 struct inode *inode,
285 unsigned type_flags)
286{
287 unsigned flags;
288
289 dentry->d_inode = inode;
4bf46a27
DH
290 flags = READ_ONCE(dentry->d_flags);
291 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
292 flags |= type_flags;
293 WRITE_ONCE(dentry->d_flags, flags);
294}
295
4bf46a27
DH
296static inline void __d_clear_type_and_inode(struct dentry *dentry)
297{
298 unsigned flags = READ_ONCE(dentry->d_flags);
299
300 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
301 WRITE_ONCE(dentry->d_flags, flags);
4bf46a27
DH
302 dentry->d_inode = NULL;
303}
304
b4f0354e
AV
305static void dentry_free(struct dentry *dentry)
306{
946e51f2 307 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
8d85b484
AV
308 if (unlikely(dname_external(dentry))) {
309 struct external_name *p = external_name(dentry);
310 if (likely(atomic_dec_and_test(&p->u.count))) {
311 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
312 return;
313 }
314 }
b4f0354e
AV
315 /* if dentry was never visible to RCU, immediate free is OK */
316 if (!(dentry->d_flags & DCACHE_RCUACCESS))
317 __d_free(&dentry->d_u.d_rcu);
318 else
319 call_rcu(&dentry->d_u.d_rcu, __d_free);
320}
321
31e6b01f 322/**
a7c6f571 323 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
ff5fdb61 324 * @dentry: the target dentry
31e6b01f
NP
325 * After this call, in-progress rcu-walk path lookup will fail. This
326 * should be called after unhashing, and after changing d_inode (if
327 * the dentry has not already been unhashed).
328 */
a7c6f571 329static inline void dentry_rcuwalk_invalidate(struct dentry *dentry)
31e6b01f 330{
a7c6f571
PZ
331 lockdep_assert_held(&dentry->d_lock);
332 /* Go through am invalidation barrier */
333 write_seqcount_invalidate(&dentry->d_seq);
31e6b01f
NP
334}
335
1da177e4
LT
336/*
337 * Release the dentry's inode, using the filesystem
31e6b01f
NP
338 * d_iput() operation if defined. Dentry has no refcount
339 * and is unhashed.
1da177e4 340 */
858119e1 341static void dentry_iput(struct dentry * dentry)
31f3e0b3 342 __releases(dentry->d_lock)
873feea0 343 __releases(dentry->d_inode->i_lock)
1da177e4
LT
344{
345 struct inode *inode = dentry->d_inode;
346 if (inode) {
4bf46a27 347 __d_clear_type_and_inode(dentry);
946e51f2 348 hlist_del_init(&dentry->d_u.d_alias);
1da177e4 349 spin_unlock(&dentry->d_lock);
873feea0 350 spin_unlock(&inode->i_lock);
f805fbda
LT
351 if (!inode->i_nlink)
352 fsnotify_inoderemove(inode);
1da177e4
LT
353 if (dentry->d_op && dentry->d_op->d_iput)
354 dentry->d_op->d_iput(dentry, inode);
355 else
356 iput(inode);
357 } else {
358 spin_unlock(&dentry->d_lock);
1da177e4
LT
359 }
360}
361
31e6b01f
NP
362/*
363 * Release the dentry's inode, using the filesystem
364 * d_iput() operation if defined. dentry remains in-use.
365 */
366static void dentry_unlink_inode(struct dentry * dentry)
367 __releases(dentry->d_lock)
873feea0 368 __releases(dentry->d_inode->i_lock)
31e6b01f
NP
369{
370 struct inode *inode = dentry->d_inode;
a528aca7
AV
371
372 raw_write_seqcount_begin(&dentry->d_seq);
4bf46a27 373 __d_clear_type_and_inode(dentry);
946e51f2 374 hlist_del_init(&dentry->d_u.d_alias);
a528aca7 375 raw_write_seqcount_end(&dentry->d_seq);
31e6b01f 376 spin_unlock(&dentry->d_lock);
873feea0 377 spin_unlock(&inode->i_lock);
31e6b01f
NP
378 if (!inode->i_nlink)
379 fsnotify_inoderemove(inode);
380 if (dentry->d_op && dentry->d_op->d_iput)
381 dentry->d_op->d_iput(dentry, inode);
382 else
383 iput(inode);
384}
385
89dc77bc
LT
386/*
387 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
388 * is in use - which includes both the "real" per-superblock
389 * LRU list _and_ the DCACHE_SHRINK_LIST use.
390 *
391 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
392 * on the shrink list (ie not on the superblock LRU list).
393 *
394 * The per-cpu "nr_dentry_unused" counters are updated with
395 * the DCACHE_LRU_LIST bit.
396 *
397 * These helper functions make sure we always follow the
398 * rules. d_lock must be held by the caller.
399 */
400#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
401static void d_lru_add(struct dentry *dentry)
402{
403 D_FLAG_VERIFY(dentry, 0);
404 dentry->d_flags |= DCACHE_LRU_LIST;
405 this_cpu_inc(nr_dentry_unused);
406 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
407}
408
409static void d_lru_del(struct dentry *dentry)
410{
411 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
412 dentry->d_flags &= ~DCACHE_LRU_LIST;
413 this_cpu_dec(nr_dentry_unused);
414 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
415}
416
417static void d_shrink_del(struct dentry *dentry)
418{
419 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
420 list_del_init(&dentry->d_lru);
421 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
422 this_cpu_dec(nr_dentry_unused);
423}
424
425static void d_shrink_add(struct dentry *dentry, struct list_head *list)
426{
427 D_FLAG_VERIFY(dentry, 0);
428 list_add(&dentry->d_lru, list);
429 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
430 this_cpu_inc(nr_dentry_unused);
431}
432
433/*
434 * These can only be called under the global LRU lock, ie during the
435 * callback for freeing the LRU list. "isolate" removes it from the
436 * LRU lists entirely, while shrink_move moves it to the indicated
437 * private list.
438 */
3f97b163 439static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
89dc77bc
LT
440{
441 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
442 dentry->d_flags &= ~DCACHE_LRU_LIST;
443 this_cpu_dec(nr_dentry_unused);
3f97b163 444 list_lru_isolate(lru, &dentry->d_lru);
89dc77bc
LT
445}
446
3f97b163
VD
447static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
448 struct list_head *list)
89dc77bc
LT
449{
450 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
451 dentry->d_flags |= DCACHE_SHRINK_LIST;
3f97b163 452 list_lru_isolate_move(lru, &dentry->d_lru, list);
89dc77bc
LT
453}
454
da3bbdd4 455/*
f6041567 456 * dentry_lru_(add|del)_list) must be called with d_lock held.
da3bbdd4
KM
457 */
458static void dentry_lru_add(struct dentry *dentry)
459{
89dc77bc
LT
460 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
461 d_lru_add(dentry);
da3bbdd4
KM
462}
463
789680d1
NP
464/**
465 * d_drop - drop a dentry
466 * @dentry: dentry to drop
467 *
468 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
469 * be found through a VFS lookup any more. Note that this is different from
470 * deleting the dentry - d_delete will try to mark the dentry negative if
471 * possible, giving a successful _negative_ lookup, while d_drop will
472 * just make the cache lookup fail.
473 *
474 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
475 * reason (NFS timeouts or autofs deletes).
476 *
477 * __d_drop requires dentry->d_lock.
478 */
479void __d_drop(struct dentry *dentry)
480{
dea3667b 481 if (!d_unhashed(dentry)) {
b61625d2 482 struct hlist_bl_head *b;
7632e465
BF
483 /*
484 * Hashed dentries are normally on the dentry hashtable,
485 * with the exception of those newly allocated by
486 * d_obtain_alias, which are always IS_ROOT:
487 */
488 if (unlikely(IS_ROOT(dentry)))
b61625d2
AV
489 b = &dentry->d_sb->s_anon;
490 else
491 b = d_hash(dentry->d_parent, dentry->d_name.hash);
492
493 hlist_bl_lock(b);
494 __hlist_bl_del(&dentry->d_hash);
495 dentry->d_hash.pprev = NULL;
496 hlist_bl_unlock(b);
a7c6f571 497 dentry_rcuwalk_invalidate(dentry);
789680d1
NP
498 }
499}
500EXPORT_SYMBOL(__d_drop);
501
502void d_drop(struct dentry *dentry)
503{
789680d1
NP
504 spin_lock(&dentry->d_lock);
505 __d_drop(dentry);
506 spin_unlock(&dentry->d_lock);
789680d1
NP
507}
508EXPORT_SYMBOL(d_drop);
509
e55fd011 510static void __dentry_kill(struct dentry *dentry)
77812a1e 511{
41edf278
AV
512 struct dentry *parent = NULL;
513 bool can_free = true;
41edf278 514 if (!IS_ROOT(dentry))
77812a1e 515 parent = dentry->d_parent;
31e6b01f 516
0d98439e
LT
517 /*
518 * The dentry is now unrecoverably dead to the world.
519 */
520 lockref_mark_dead(&dentry->d_lockref);
521
f0023bc6 522 /*
f0023bc6
SW
523 * inform the fs via d_prune that this dentry is about to be
524 * unhashed and destroyed.
525 */
29266201 526 if (dentry->d_flags & DCACHE_OP_PRUNE)
61572bb1
YZ
527 dentry->d_op->d_prune(dentry);
528
01b60351
AV
529 if (dentry->d_flags & DCACHE_LRU_LIST) {
530 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
531 d_lru_del(dentry);
01b60351 532 }
77812a1e
NP
533 /* if it was on the hash then remove it */
534 __d_drop(dentry);
ca5358ef 535 __list_del_entry(&dentry->d_child);
03b3b889
AV
536 /*
537 * Inform d_walk() that we are no longer attached to the
538 * dentry tree
539 */
540 dentry->d_flags |= DCACHE_DENTRY_KILLED;
541 if (parent)
542 spin_unlock(&parent->d_lock);
543 dentry_iput(dentry);
544 /*
545 * dentry_iput drops the locks, at which point nobody (except
546 * transient RCU lookups) can reach this dentry.
547 */
360f5479 548 BUG_ON(dentry->d_lockref.count > 0);
03b3b889
AV
549 this_cpu_dec(nr_dentry);
550 if (dentry->d_op && dentry->d_op->d_release)
551 dentry->d_op->d_release(dentry);
552
41edf278
AV
553 spin_lock(&dentry->d_lock);
554 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
555 dentry->d_flags |= DCACHE_MAY_FREE;
556 can_free = false;
557 }
558 spin_unlock(&dentry->d_lock);
41edf278
AV
559 if (likely(can_free))
560 dentry_free(dentry);
e55fd011
AV
561}
562
563/*
564 * Finish off a dentry we've decided to kill.
565 * dentry->d_lock must be held, returns with it unlocked.
566 * If ref is non-zero, then decrement the refcount too.
567 * Returns dentry requiring refcount drop, or NULL if we're done.
568 */
8cbf74da 569static struct dentry *dentry_kill(struct dentry *dentry)
e55fd011
AV
570 __releases(dentry->d_lock)
571{
572 struct inode *inode = dentry->d_inode;
573 struct dentry *parent = NULL;
574
575 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
576 goto failed;
577
578 if (!IS_ROOT(dentry)) {
579 parent = dentry->d_parent;
580 if (unlikely(!spin_trylock(&parent->d_lock))) {
581 if (inode)
582 spin_unlock(&inode->i_lock);
583 goto failed;
584 }
585 }
586
587 __dentry_kill(dentry);
03b3b889 588 return parent;
e55fd011
AV
589
590failed:
8cbf74da
AV
591 spin_unlock(&dentry->d_lock);
592 cpu_relax();
e55fd011 593 return dentry; /* try again with same dentry */
77812a1e
NP
594}
595
046b961b
AV
596static inline struct dentry *lock_parent(struct dentry *dentry)
597{
598 struct dentry *parent = dentry->d_parent;
599 if (IS_ROOT(dentry))
600 return NULL;
360f5479 601 if (unlikely(dentry->d_lockref.count < 0))
c2338f2d 602 return NULL;
046b961b
AV
603 if (likely(spin_trylock(&parent->d_lock)))
604 return parent;
046b961b 605 rcu_read_lock();
c2338f2d 606 spin_unlock(&dentry->d_lock);
046b961b
AV
607again:
608 parent = ACCESS_ONCE(dentry->d_parent);
609 spin_lock(&parent->d_lock);
610 /*
611 * We can't blindly lock dentry until we are sure
612 * that we won't violate the locking order.
613 * Any changes of dentry->d_parent must have
614 * been done with parent->d_lock held, so
615 * spin_lock() above is enough of a barrier
616 * for checking if it's still our child.
617 */
618 if (unlikely(parent != dentry->d_parent)) {
619 spin_unlock(&parent->d_lock);
620 goto again;
621 }
622 rcu_read_unlock();
623 if (parent != dentry)
9f12600f 624 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
046b961b
AV
625 else
626 parent = NULL;
627 return parent;
628}
629
360f5479
LT
630/*
631 * Try to do a lockless dput(), and return whether that was successful.
632 *
633 * If unsuccessful, we return false, having already taken the dentry lock.
634 *
635 * The caller needs to hold the RCU read lock, so that the dentry is
636 * guaranteed to stay around even if the refcount goes down to zero!
637 */
638static inline bool fast_dput(struct dentry *dentry)
639{
640 int ret;
641 unsigned int d_flags;
642
643 /*
644 * If we have a d_op->d_delete() operation, we sould not
75a6f82a 645 * let the dentry count go to zero, so use "put_or_lock".
360f5479
LT
646 */
647 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
648 return lockref_put_or_lock(&dentry->d_lockref);
649
650 /*
651 * .. otherwise, we can try to just decrement the
652 * lockref optimistically.
653 */
654 ret = lockref_put_return(&dentry->d_lockref);
655
656 /*
657 * If the lockref_put_return() failed due to the lock being held
658 * by somebody else, the fast path has failed. We will need to
659 * get the lock, and then check the count again.
660 */
661 if (unlikely(ret < 0)) {
662 spin_lock(&dentry->d_lock);
663 if (dentry->d_lockref.count > 1) {
664 dentry->d_lockref.count--;
665 spin_unlock(&dentry->d_lock);
666 return 1;
667 }
668 return 0;
669 }
670
671 /*
672 * If we weren't the last ref, we're done.
673 */
674 if (ret)
675 return 1;
676
677 /*
678 * Careful, careful. The reference count went down
679 * to zero, but we don't hold the dentry lock, so
680 * somebody else could get it again, and do another
681 * dput(), and we need to not race with that.
682 *
683 * However, there is a very special and common case
684 * where we don't care, because there is nothing to
685 * do: the dentry is still hashed, it does not have
686 * a 'delete' op, and it's referenced and already on
687 * the LRU list.
688 *
689 * NOTE! Since we aren't locked, these values are
690 * not "stable". However, it is sufficient that at
691 * some point after we dropped the reference the
692 * dentry was hashed and the flags had the proper
693 * value. Other dentry users may have re-gotten
694 * a reference to the dentry and change that, but
695 * our work is done - we can leave the dentry
696 * around with a zero refcount.
697 */
698 smp_rmb();
699 d_flags = ACCESS_ONCE(dentry->d_flags);
75a6f82a 700 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
360f5479
LT
701
702 /* Nothing to do? Dropping the reference was all we needed? */
703 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
704 return 1;
705
706 /*
707 * Not the fast normal case? Get the lock. We've already decremented
708 * the refcount, but we'll need to re-check the situation after
709 * getting the lock.
710 */
711 spin_lock(&dentry->d_lock);
712
713 /*
714 * Did somebody else grab a reference to it in the meantime, and
715 * we're no longer the last user after all? Alternatively, somebody
716 * else could have killed it and marked it dead. Either way, we
717 * don't need to do anything else.
718 */
719 if (dentry->d_lockref.count) {
720 spin_unlock(&dentry->d_lock);
721 return 1;
722 }
723
724 /*
725 * Re-get the reference we optimistically dropped. We hold the
726 * lock, and we just tested that it was zero, so we can just
727 * set it to 1.
728 */
729 dentry->d_lockref.count = 1;
730 return 0;
731}
732
733
1da177e4
LT
734/*
735 * This is dput
736 *
737 * This is complicated by the fact that we do not want to put
738 * dentries that are no longer on any hash chain on the unused
739 * list: we'd much rather just get rid of them immediately.
740 *
741 * However, that implies that we have to traverse the dentry
742 * tree upwards to the parents which might _also_ now be
743 * scheduled for deletion (it may have been only waiting for
744 * its last child to go away).
745 *
746 * This tail recursion is done by hand as we don't want to depend
747 * on the compiler to always get this right (gcc generally doesn't).
748 * Real recursion would eat up our stack space.
749 */
750
751/*
752 * dput - release a dentry
753 * @dentry: dentry to release
754 *
755 * Release a dentry. This will drop the usage count and if appropriate
756 * call the dentry unlink method as well as removing it from the queues and
757 * releasing its resources. If the parent dentries were scheduled for release
758 * they too may now get deleted.
1da177e4 759 */
1da177e4
LT
760void dput(struct dentry *dentry)
761{
8aab6a27 762 if (unlikely(!dentry))
1da177e4
LT
763 return;
764
765repeat:
360f5479
LT
766 rcu_read_lock();
767 if (likely(fast_dput(dentry))) {
768 rcu_read_unlock();
1da177e4 769 return;
360f5479
LT
770 }
771
772 /* Slow case: now with the dentry lock held */
773 rcu_read_unlock();
1da177e4 774
85c7f810
AV
775 WARN_ON(d_in_lookup(dentry));
776
8aab6a27
LT
777 /* Unreachable? Get rid of it */
778 if (unlikely(d_unhashed(dentry)))
779 goto kill_it;
780
75a6f82a
AV
781 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
782 goto kill_it;
783
8aab6a27 784 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
1da177e4 785 if (dentry->d_op->d_delete(dentry))
61f3dee4 786 goto kill_it;
1da177e4 787 }
265ac902 788
358eec18
LT
789 if (!(dentry->d_flags & DCACHE_REFERENCED))
790 dentry->d_flags |= DCACHE_REFERENCED;
a4633357 791 dentry_lru_add(dentry);
265ac902 792
98474236 793 dentry->d_lockref.count--;
61f3dee4 794 spin_unlock(&dentry->d_lock);
1da177e4
LT
795 return;
796
d52b9086 797kill_it:
8cbf74da 798 dentry = dentry_kill(dentry);
d52b9086
MS
799 if (dentry)
800 goto repeat;
1da177e4 801}
ec4f8605 802EXPORT_SYMBOL(dput);
1da177e4 803
1da177e4 804
b5c84bf6 805/* This must be called with d_lock held */
dc0474be 806static inline void __dget_dlock(struct dentry *dentry)
23044507 807{
98474236 808 dentry->d_lockref.count++;
23044507
NP
809}
810
dc0474be 811static inline void __dget(struct dentry *dentry)
1da177e4 812{
98474236 813 lockref_get(&dentry->d_lockref);
1da177e4
LT
814}
815
b7ab39f6
NP
816struct dentry *dget_parent(struct dentry *dentry)
817{
df3d0bbc 818 int gotref;
b7ab39f6
NP
819 struct dentry *ret;
820
df3d0bbc
WL
821 /*
822 * Do optimistic parent lookup without any
823 * locking.
824 */
825 rcu_read_lock();
826 ret = ACCESS_ONCE(dentry->d_parent);
827 gotref = lockref_get_not_zero(&ret->d_lockref);
828 rcu_read_unlock();
829 if (likely(gotref)) {
830 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
831 return ret;
832 dput(ret);
833 }
834
b7ab39f6 835repeat:
a734eb45
NP
836 /*
837 * Don't need rcu_dereference because we re-check it was correct under
838 * the lock.
839 */
840 rcu_read_lock();
b7ab39f6 841 ret = dentry->d_parent;
a734eb45
NP
842 spin_lock(&ret->d_lock);
843 if (unlikely(ret != dentry->d_parent)) {
844 spin_unlock(&ret->d_lock);
845 rcu_read_unlock();
b7ab39f6
NP
846 goto repeat;
847 }
a734eb45 848 rcu_read_unlock();
98474236
WL
849 BUG_ON(!ret->d_lockref.count);
850 ret->d_lockref.count++;
b7ab39f6 851 spin_unlock(&ret->d_lock);
b7ab39f6
NP
852 return ret;
853}
854EXPORT_SYMBOL(dget_parent);
855
1da177e4
LT
856/**
857 * d_find_alias - grab a hashed alias of inode
858 * @inode: inode in question
1da177e4
LT
859 *
860 * If inode has a hashed alias, or is a directory and has any alias,
861 * acquire the reference to alias and return it. Otherwise return NULL.
862 * Notice that if inode is a directory there can be only one alias and
863 * it can be unhashed only if it has no children, or if it is the root
3ccb354d
EB
864 * of a filesystem, or if the directory was renamed and d_revalidate
865 * was the first vfs operation to notice.
1da177e4 866 *
21c0d8fd 867 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
52ed46f0 868 * any other hashed alias over that one.
1da177e4 869 */
52ed46f0 870static struct dentry *__d_find_alias(struct inode *inode)
1da177e4 871{
da502956 872 struct dentry *alias, *discon_alias;
1da177e4 873
da502956
NP
874again:
875 discon_alias = NULL;
946e51f2 876 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
da502956 877 spin_lock(&alias->d_lock);
1da177e4 878 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
21c0d8fd 879 if (IS_ROOT(alias) &&
da502956 880 (alias->d_flags & DCACHE_DISCONNECTED)) {
1da177e4 881 discon_alias = alias;
52ed46f0 882 } else {
dc0474be 883 __dget_dlock(alias);
da502956
NP
884 spin_unlock(&alias->d_lock);
885 return alias;
886 }
887 }
888 spin_unlock(&alias->d_lock);
889 }
890 if (discon_alias) {
891 alias = discon_alias;
892 spin_lock(&alias->d_lock);
893 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
8d80d7da
BF
894 __dget_dlock(alias);
895 spin_unlock(&alias->d_lock);
896 return alias;
1da177e4 897 }
da502956
NP
898 spin_unlock(&alias->d_lock);
899 goto again;
1da177e4 900 }
da502956 901 return NULL;
1da177e4
LT
902}
903
da502956 904struct dentry *d_find_alias(struct inode *inode)
1da177e4 905{
214fda1f
DH
906 struct dentry *de = NULL;
907
b3d9b7a3 908 if (!hlist_empty(&inode->i_dentry)) {
873feea0 909 spin_lock(&inode->i_lock);
52ed46f0 910 de = __d_find_alias(inode);
873feea0 911 spin_unlock(&inode->i_lock);
214fda1f 912 }
1da177e4
LT
913 return de;
914}
ec4f8605 915EXPORT_SYMBOL(d_find_alias);
1da177e4
LT
916
917/*
918 * Try to kill dentries associated with this inode.
919 * WARNING: you must own a reference to inode.
920 */
921void d_prune_aliases(struct inode *inode)
922{
0cdca3f9 923 struct dentry *dentry;
1da177e4 924restart:
873feea0 925 spin_lock(&inode->i_lock);
946e51f2 926 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
1da177e4 927 spin_lock(&dentry->d_lock);
98474236 928 if (!dentry->d_lockref.count) {
29355c39
AV
929 struct dentry *parent = lock_parent(dentry);
930 if (likely(!dentry->d_lockref.count)) {
931 __dentry_kill(dentry);
4a7795d3 932 dput(parent);
29355c39
AV
933 goto restart;
934 }
935 if (parent)
936 spin_unlock(&parent->d_lock);
1da177e4
LT
937 }
938 spin_unlock(&dentry->d_lock);
939 }
873feea0 940 spin_unlock(&inode->i_lock);
1da177e4 941}
ec4f8605 942EXPORT_SYMBOL(d_prune_aliases);
1da177e4 943
3049cfe2 944static void shrink_dentry_list(struct list_head *list)
1da177e4 945{
5c47e6d0 946 struct dentry *dentry, *parent;
da3bbdd4 947
60942f2f 948 while (!list_empty(list)) {
ff2fde99 949 struct inode *inode;
60942f2f 950 dentry = list_entry(list->prev, struct dentry, d_lru);
ec33679d 951 spin_lock(&dentry->d_lock);
046b961b
AV
952 parent = lock_parent(dentry);
953
dd1f6b2e
DC
954 /*
955 * The dispose list is isolated and dentries are not accounted
956 * to the LRU here, so we can simply remove it from the list
957 * here regardless of whether it is referenced or not.
958 */
89dc77bc 959 d_shrink_del(dentry);
dd1f6b2e 960
1da177e4
LT
961 /*
962 * We found an inuse dentry which was not removed from
dd1f6b2e 963 * the LRU because of laziness during lookup. Do not free it.
1da177e4 964 */
360f5479 965 if (dentry->d_lockref.count > 0) {
da3bbdd4 966 spin_unlock(&dentry->d_lock);
046b961b
AV
967 if (parent)
968 spin_unlock(&parent->d_lock);
1da177e4
LT
969 continue;
970 }
77812a1e 971
64fd72e0
AV
972
973 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
974 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
975 spin_unlock(&dentry->d_lock);
046b961b
AV
976 if (parent)
977 spin_unlock(&parent->d_lock);
64fd72e0
AV
978 if (can_free)
979 dentry_free(dentry);
980 continue;
981 }
982
ff2fde99
AV
983 inode = dentry->d_inode;
984 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
89dc77bc 985 d_shrink_add(dentry, list);
dd1f6b2e 986 spin_unlock(&dentry->d_lock);
046b961b
AV
987 if (parent)
988 spin_unlock(&parent->d_lock);
5c47e6d0 989 continue;
dd1f6b2e 990 }
ff2fde99 991
ff2fde99 992 __dentry_kill(dentry);
046b961b 993
5c47e6d0
AV
994 /*
995 * We need to prune ancestors too. This is necessary to prevent
996 * quadratic behavior of shrink_dcache_parent(), but is also
997 * expected to be beneficial in reducing dentry cache
998 * fragmentation.
999 */
1000 dentry = parent;
b2b80195
AV
1001 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1002 parent = lock_parent(dentry);
1003 if (dentry->d_lockref.count != 1) {
1004 dentry->d_lockref.count--;
1005 spin_unlock(&dentry->d_lock);
1006 if (parent)
1007 spin_unlock(&parent->d_lock);
1008 break;
1009 }
1010 inode = dentry->d_inode; /* can't be NULL */
1011 if (unlikely(!spin_trylock(&inode->i_lock))) {
1012 spin_unlock(&dentry->d_lock);
1013 if (parent)
1014 spin_unlock(&parent->d_lock);
1015 cpu_relax();
1016 continue;
1017 }
1018 __dentry_kill(dentry);
1019 dentry = parent;
1020 }
da3bbdd4 1021 }
3049cfe2
CH
1022}
1023
3f97b163
VD
1024static enum lru_status dentry_lru_isolate(struct list_head *item,
1025 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
f6041567
DC
1026{
1027 struct list_head *freeable = arg;
1028 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1029
1030
1031 /*
1032 * we are inverting the lru lock/dentry->d_lock here,
1033 * so use a trylock. If we fail to get the lock, just skip
1034 * it
1035 */
1036 if (!spin_trylock(&dentry->d_lock))
1037 return LRU_SKIP;
1038
1039 /*
1040 * Referenced dentries are still in use. If they have active
1041 * counts, just remove them from the LRU. Otherwise give them
1042 * another pass through the LRU.
1043 */
1044 if (dentry->d_lockref.count) {
3f97b163 1045 d_lru_isolate(lru, dentry);
f6041567
DC
1046 spin_unlock(&dentry->d_lock);
1047 return LRU_REMOVED;
1048 }
1049
1050 if (dentry->d_flags & DCACHE_REFERENCED) {
1051 dentry->d_flags &= ~DCACHE_REFERENCED;
1052 spin_unlock(&dentry->d_lock);
1053
1054 /*
1055 * The list move itself will be made by the common LRU code. At
1056 * this point, we've dropped the dentry->d_lock but keep the
1057 * lru lock. This is safe to do, since every list movement is
1058 * protected by the lru lock even if both locks are held.
1059 *
1060 * This is guaranteed by the fact that all LRU management
1061 * functions are intermediated by the LRU API calls like
1062 * list_lru_add and list_lru_del. List movement in this file
1063 * only ever occur through this functions or through callbacks
1064 * like this one, that are called from the LRU API.
1065 *
1066 * The only exceptions to this are functions like
1067 * shrink_dentry_list, and code that first checks for the
1068 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1069 * operating only with stack provided lists after they are
1070 * properly isolated from the main list. It is thus, always a
1071 * local access.
1072 */
1073 return LRU_ROTATE;
1074 }
1075
3f97b163 1076 d_lru_shrink_move(lru, dentry, freeable);
f6041567
DC
1077 spin_unlock(&dentry->d_lock);
1078
1079 return LRU_REMOVED;
1080}
1081
3049cfe2 1082/**
b48f03b3
DC
1083 * prune_dcache_sb - shrink the dcache
1084 * @sb: superblock
503c358c 1085 * @sc: shrink control, passed to list_lru_shrink_walk()
b48f03b3 1086 *
503c358c
VD
1087 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1088 * is done when we need more memory and called from the superblock shrinker
b48f03b3 1089 * function.
3049cfe2 1090 *
b48f03b3
DC
1091 * This function may fail to free any resources if all the dentries are in
1092 * use.
3049cfe2 1093 */
503c358c 1094long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
3049cfe2 1095{
f6041567
DC
1096 LIST_HEAD(dispose);
1097 long freed;
3049cfe2 1098
503c358c
VD
1099 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1100 dentry_lru_isolate, &dispose);
f6041567 1101 shrink_dentry_list(&dispose);
0a234c6d 1102 return freed;
da3bbdd4 1103}
23044507 1104
4e717f5c 1105static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
3f97b163 1106 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
dd1f6b2e 1107{
4e717f5c
GC
1108 struct list_head *freeable = arg;
1109 struct dentry *dentry = container_of(item, struct dentry, d_lru);
dd1f6b2e 1110
4e717f5c
GC
1111 /*
1112 * we are inverting the lru lock/dentry->d_lock here,
1113 * so use a trylock. If we fail to get the lock, just skip
1114 * it
1115 */
1116 if (!spin_trylock(&dentry->d_lock))
1117 return LRU_SKIP;
1118
3f97b163 1119 d_lru_shrink_move(lru, dentry, freeable);
4e717f5c 1120 spin_unlock(&dentry->d_lock);
ec33679d 1121
4e717f5c 1122 return LRU_REMOVED;
da3bbdd4
KM
1123}
1124
4e717f5c 1125
1da177e4
LT
1126/**
1127 * shrink_dcache_sb - shrink dcache for a superblock
1128 * @sb: superblock
1129 *
3049cfe2
CH
1130 * Shrink the dcache for the specified super block. This is used to free
1131 * the dcache before unmounting a file system.
1da177e4 1132 */
3049cfe2 1133void shrink_dcache_sb(struct super_block *sb)
1da177e4 1134{
4e717f5c
GC
1135 long freed;
1136
1137 do {
1138 LIST_HEAD(dispose);
1139
1140 freed = list_lru_walk(&sb->s_dentry_lru,
1141 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
3049cfe2 1142
4e717f5c
GC
1143 this_cpu_sub(nr_dentry_unused, freed);
1144 shrink_dentry_list(&dispose);
1145 } while (freed > 0);
1da177e4 1146}
ec4f8605 1147EXPORT_SYMBOL(shrink_dcache_sb);
1da177e4 1148
db14fc3a
MS
1149/**
1150 * enum d_walk_ret - action to talke during tree walk
1151 * @D_WALK_CONTINUE: contrinue walk
1152 * @D_WALK_QUIT: quit walk
1153 * @D_WALK_NORETRY: quit when retry is needed
1154 * @D_WALK_SKIP: skip this dentry and its children
1155 */
1156enum d_walk_ret {
1157 D_WALK_CONTINUE,
1158 D_WALK_QUIT,
1159 D_WALK_NORETRY,
1160 D_WALK_SKIP,
1161};
c826cb7d 1162
1da177e4 1163/**
db14fc3a
MS
1164 * d_walk - walk the dentry tree
1165 * @parent: start of walk
1166 * @data: data passed to @enter() and @finish()
1167 * @enter: callback when first entering the dentry
1168 * @finish: callback when successfully finished the walk
1da177e4 1169 *
db14fc3a 1170 * The @enter() and @finish() callbacks are called with d_lock held.
1da177e4 1171 */
db14fc3a
MS
1172static void d_walk(struct dentry *parent, void *data,
1173 enum d_walk_ret (*enter)(void *, struct dentry *),
1174 void (*finish)(void *))
1da177e4 1175{
949854d0 1176 struct dentry *this_parent;
1da177e4 1177 struct list_head *next;
48f5ec21 1178 unsigned seq = 0;
db14fc3a
MS
1179 enum d_walk_ret ret;
1180 bool retry = true;
949854d0 1181
58db63d0 1182again:
48f5ec21 1183 read_seqbegin_or_lock(&rename_lock, &seq);
58db63d0 1184 this_parent = parent;
2fd6b7f5 1185 spin_lock(&this_parent->d_lock);
db14fc3a
MS
1186
1187 ret = enter(data, this_parent);
1188 switch (ret) {
1189 case D_WALK_CONTINUE:
1190 break;
1191 case D_WALK_QUIT:
1192 case D_WALK_SKIP:
1193 goto out_unlock;
1194 case D_WALK_NORETRY:
1195 retry = false;
1196 break;
1197 }
1da177e4
LT
1198repeat:
1199 next = this_parent->d_subdirs.next;
1200resume:
1201 while (next != &this_parent->d_subdirs) {
1202 struct list_head *tmp = next;
946e51f2 1203 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1da177e4 1204 next = tmp->next;
2fd6b7f5
NP
1205
1206 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
db14fc3a
MS
1207
1208 ret = enter(data, dentry);
1209 switch (ret) {
1210 case D_WALK_CONTINUE:
1211 break;
1212 case D_WALK_QUIT:
2fd6b7f5 1213 spin_unlock(&dentry->d_lock);
db14fc3a
MS
1214 goto out_unlock;
1215 case D_WALK_NORETRY:
1216 retry = false;
1217 break;
1218 case D_WALK_SKIP:
1219 spin_unlock(&dentry->d_lock);
1220 continue;
2fd6b7f5 1221 }
db14fc3a 1222
1da177e4 1223 if (!list_empty(&dentry->d_subdirs)) {
2fd6b7f5
NP
1224 spin_unlock(&this_parent->d_lock);
1225 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1da177e4 1226 this_parent = dentry;
2fd6b7f5 1227 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1da177e4
LT
1228 goto repeat;
1229 }
2fd6b7f5 1230 spin_unlock(&dentry->d_lock);
1da177e4
LT
1231 }
1232 /*
1233 * All done at this level ... ascend and resume the search.
1234 */
ca5358ef
AV
1235 rcu_read_lock();
1236ascend:
1da177e4 1237 if (this_parent != parent) {
c826cb7d 1238 struct dentry *child = this_parent;
31dec132
AV
1239 this_parent = child->d_parent;
1240
31dec132
AV
1241 spin_unlock(&child->d_lock);
1242 spin_lock(&this_parent->d_lock);
1243
ca5358ef
AV
1244 /* might go back up the wrong parent if we have had a rename. */
1245 if (need_seqretry(&rename_lock, seq))
949854d0 1246 goto rename_retry;
2159184e
AV
1247 /* go into the first sibling still alive */
1248 do {
1249 next = child->d_child.next;
ca5358ef
AV
1250 if (next == &this_parent->d_subdirs)
1251 goto ascend;
1252 child = list_entry(next, struct dentry, d_child);
2159184e 1253 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
31dec132 1254 rcu_read_unlock();
1da177e4
LT
1255 goto resume;
1256 }
ca5358ef 1257 if (need_seqretry(&rename_lock, seq))
949854d0 1258 goto rename_retry;
ca5358ef 1259 rcu_read_unlock();
db14fc3a
MS
1260 if (finish)
1261 finish(data);
1262
1263out_unlock:
1264 spin_unlock(&this_parent->d_lock);
48f5ec21 1265 done_seqretry(&rename_lock, seq);
db14fc3a 1266 return;
58db63d0
NP
1267
1268rename_retry:
ca5358ef
AV
1269 spin_unlock(&this_parent->d_lock);
1270 rcu_read_unlock();
1271 BUG_ON(seq & 1);
db14fc3a
MS
1272 if (!retry)
1273 return;
48f5ec21 1274 seq = 1;
58db63d0 1275 goto again;
1da177e4 1276}
db14fc3a
MS
1277
1278/*
1279 * Search for at least 1 mount point in the dentry's subdirs.
1280 * We descend to the next level whenever the d_subdirs
1281 * list is non-empty and continue searching.
1282 */
1283
db14fc3a
MS
1284static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1285{
1286 int *ret = data;
1287 if (d_mountpoint(dentry)) {
1288 *ret = 1;
1289 return D_WALK_QUIT;
1290 }
1291 return D_WALK_CONTINUE;
1292}
1293
69c88dc7
RD
1294/**
1295 * have_submounts - check for mounts over a dentry
1296 * @parent: dentry to check.
1297 *
1298 * Return true if the parent or its subdirectories contain
1299 * a mount point
1300 */
db14fc3a
MS
1301int have_submounts(struct dentry *parent)
1302{
1303 int ret = 0;
1304
1305 d_walk(parent, &ret, check_mount, NULL);
1306
1307 return ret;
1308}
ec4f8605 1309EXPORT_SYMBOL(have_submounts);
1da177e4 1310
eed81007
MS
1311/*
1312 * Called by mount code to set a mountpoint and check if the mountpoint is
1313 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1314 * subtree can become unreachable).
1315 *
1ffe46d1 1316 * Only one of d_invalidate() and d_set_mounted() must succeed. For
eed81007
MS
1317 * this reason take rename_lock and d_lock on dentry and ancestors.
1318 */
1319int d_set_mounted(struct dentry *dentry)
1320{
1321 struct dentry *p;
1322 int ret = -ENOENT;
1323 write_seqlock(&rename_lock);
1324 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1ffe46d1 1325 /* Need exclusion wrt. d_invalidate() */
eed81007
MS
1326 spin_lock(&p->d_lock);
1327 if (unlikely(d_unhashed(p))) {
1328 spin_unlock(&p->d_lock);
1329 goto out;
1330 }
1331 spin_unlock(&p->d_lock);
1332 }
1333 spin_lock(&dentry->d_lock);
1334 if (!d_unlinked(dentry)) {
1335 dentry->d_flags |= DCACHE_MOUNTED;
1336 ret = 0;
1337 }
1338 spin_unlock(&dentry->d_lock);
1339out:
1340 write_sequnlock(&rename_lock);
1341 return ret;
1342}
1343
1da177e4 1344/*
fd517909 1345 * Search the dentry child list of the specified parent,
1da177e4
LT
1346 * and move any unused dentries to the end of the unused
1347 * list for prune_dcache(). We descend to the next level
1348 * whenever the d_subdirs list is non-empty and continue
1349 * searching.
1350 *
1351 * It returns zero iff there are no unused children,
1352 * otherwise it returns the number of children moved to
1353 * the end of the unused list. This may not be the total
1354 * number of unused children, because select_parent can
1355 * drop the lock and return early due to latency
1356 * constraints.
1357 */
1da177e4 1358
db14fc3a
MS
1359struct select_data {
1360 struct dentry *start;
1361 struct list_head dispose;
1362 int found;
1363};
23044507 1364
db14fc3a
MS
1365static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1366{
1367 struct select_data *data = _data;
1368 enum d_walk_ret ret = D_WALK_CONTINUE;
1da177e4 1369
db14fc3a
MS
1370 if (data->start == dentry)
1371 goto out;
2fd6b7f5 1372
fe91522a 1373 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
db14fc3a 1374 data->found++;
fe91522a
AV
1375 } else {
1376 if (dentry->d_flags & DCACHE_LRU_LIST)
1377 d_lru_del(dentry);
1378 if (!dentry->d_lockref.count) {
1379 d_shrink_add(dentry, &data->dispose);
1380 data->found++;
1381 }
1da177e4 1382 }
db14fc3a
MS
1383 /*
1384 * We can return to the caller if we have found some (this
1385 * ensures forward progress). We'll be coming back to find
1386 * the rest.
1387 */
fe91522a
AV
1388 if (!list_empty(&data->dispose))
1389 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1da177e4 1390out:
db14fc3a 1391 return ret;
1da177e4
LT
1392}
1393
1394/**
1395 * shrink_dcache_parent - prune dcache
1396 * @parent: parent of entries to prune
1397 *
1398 * Prune the dcache to remove unused children of the parent dentry.
1399 */
db14fc3a 1400void shrink_dcache_parent(struct dentry *parent)
1da177e4 1401{
db14fc3a
MS
1402 for (;;) {
1403 struct select_data data;
1da177e4 1404
db14fc3a
MS
1405 INIT_LIST_HEAD(&data.dispose);
1406 data.start = parent;
1407 data.found = 0;
1408
1409 d_walk(parent, &data, select_collect, NULL);
1410 if (!data.found)
1411 break;
1412
1413 shrink_dentry_list(&data.dispose);
421348f1
GT
1414 cond_resched();
1415 }
1da177e4 1416}
ec4f8605 1417EXPORT_SYMBOL(shrink_dcache_parent);
1da177e4 1418
9c8c10e2 1419static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
42c32608 1420{
9c8c10e2
AV
1421 /* it has busy descendents; complain about those instead */
1422 if (!list_empty(&dentry->d_subdirs))
1423 return D_WALK_CONTINUE;
42c32608 1424
9c8c10e2
AV
1425 /* root with refcount 1 is fine */
1426 if (dentry == _data && dentry->d_lockref.count == 1)
1427 return D_WALK_CONTINUE;
1428
1429 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1430 " still in use (%d) [unmount of %s %s]\n",
42c32608
AV
1431 dentry,
1432 dentry->d_inode ?
1433 dentry->d_inode->i_ino : 0UL,
9c8c10e2 1434 dentry,
42c32608
AV
1435 dentry->d_lockref.count,
1436 dentry->d_sb->s_type->name,
1437 dentry->d_sb->s_id);
9c8c10e2
AV
1438 WARN_ON(1);
1439 return D_WALK_CONTINUE;
1440}
1441
1442static void do_one_tree(struct dentry *dentry)
1443{
1444 shrink_dcache_parent(dentry);
1445 d_walk(dentry, dentry, umount_check, NULL);
1446 d_drop(dentry);
1447 dput(dentry);
42c32608
AV
1448}
1449
1450/*
1451 * destroy the dentries attached to a superblock on unmounting
1452 */
1453void shrink_dcache_for_umount(struct super_block *sb)
1454{
1455 struct dentry *dentry;
1456
9c8c10e2 1457 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
42c32608
AV
1458
1459 dentry = sb->s_root;
1460 sb->s_root = NULL;
9c8c10e2 1461 do_one_tree(dentry);
42c32608
AV
1462
1463 while (!hlist_bl_empty(&sb->s_anon)) {
9c8c10e2
AV
1464 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1465 do_one_tree(dentry);
42c32608
AV
1466 }
1467}
1468
8ed936b5
EB
1469struct detach_data {
1470 struct select_data select;
1471 struct dentry *mountpoint;
1472};
1473static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
848ac114 1474{
8ed936b5 1475 struct detach_data *data = _data;
848ac114
MS
1476
1477 if (d_mountpoint(dentry)) {
8ed936b5
EB
1478 __dget_dlock(dentry);
1479 data->mountpoint = dentry;
848ac114
MS
1480 return D_WALK_QUIT;
1481 }
1482
8ed936b5 1483 return select_collect(&data->select, dentry);
848ac114
MS
1484}
1485
1486static void check_and_drop(void *_data)
1487{
8ed936b5 1488 struct detach_data *data = _data;
848ac114 1489
8ed936b5
EB
1490 if (!data->mountpoint && !data->select.found)
1491 __d_drop(data->select.start);
848ac114
MS
1492}
1493
1494/**
1ffe46d1
EB
1495 * d_invalidate - detach submounts, prune dcache, and drop
1496 * @dentry: dentry to invalidate (aka detach, prune and drop)
1497 *
1ffe46d1 1498 * no dcache lock.
848ac114 1499 *
8ed936b5
EB
1500 * The final d_drop is done as an atomic operation relative to
1501 * rename_lock ensuring there are no races with d_set_mounted. This
1502 * ensures there are no unhashed dentries on the path to a mountpoint.
848ac114 1503 */
5542aa2f 1504void d_invalidate(struct dentry *dentry)
848ac114 1505{
1ffe46d1
EB
1506 /*
1507 * If it's already been dropped, return OK.
1508 */
1509 spin_lock(&dentry->d_lock);
1510 if (d_unhashed(dentry)) {
1511 spin_unlock(&dentry->d_lock);
5542aa2f 1512 return;
1ffe46d1
EB
1513 }
1514 spin_unlock(&dentry->d_lock);
1515
848ac114
MS
1516 /* Negative dentries can be dropped without further checks */
1517 if (!dentry->d_inode) {
1518 d_drop(dentry);
5542aa2f 1519 return;
848ac114
MS
1520 }
1521
1522 for (;;) {
8ed936b5 1523 struct detach_data data;
848ac114 1524
8ed936b5
EB
1525 data.mountpoint = NULL;
1526 INIT_LIST_HEAD(&data.select.dispose);
1527 data.select.start = dentry;
1528 data.select.found = 0;
1529
1530 d_walk(dentry, &data, detach_and_collect, check_and_drop);
848ac114 1531
8ed936b5
EB
1532 if (data.select.found)
1533 shrink_dentry_list(&data.select.dispose);
848ac114 1534
8ed936b5
EB
1535 if (data.mountpoint) {
1536 detach_mounts(data.mountpoint);
1537 dput(data.mountpoint);
1538 }
848ac114 1539
8ed936b5 1540 if (!data.mountpoint && !data.select.found)
848ac114
MS
1541 break;
1542
1543 cond_resched();
1544 }
848ac114 1545}
1ffe46d1 1546EXPORT_SYMBOL(d_invalidate);
848ac114 1547
1da177e4 1548/**
a4464dbc
AV
1549 * __d_alloc - allocate a dcache entry
1550 * @sb: filesystem it will belong to
1da177e4
LT
1551 * @name: qstr of the name
1552 *
1553 * Allocates a dentry. It returns %NULL if there is insufficient memory
1554 * available. On a success the dentry is returned. The name passed in is
1555 * copied and the copy passed in may be reused after this call.
1556 */
1557
a4464dbc 1558struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1da177e4
LT
1559{
1560 struct dentry *dentry;
1561 char *dname;
1562
e12ba74d 1563 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1da177e4
LT
1564 if (!dentry)
1565 return NULL;
1566
6326c71f
LT
1567 /*
1568 * We guarantee that the inline name is always NUL-terminated.
1569 * This way the memcpy() done by the name switching in rename
1570 * will still always have a NUL at the end, even if we might
1571 * be overwriting an internal NUL character
1572 */
1573 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1da177e4 1574 if (name->len > DNAME_INLINE_LEN-1) {
8d85b484 1575 size_t size = offsetof(struct external_name, name[1]);
5d097056
VD
1576 struct external_name *p = kmalloc(size + name->len,
1577 GFP_KERNEL_ACCOUNT);
8d85b484 1578 if (!p) {
1da177e4
LT
1579 kmem_cache_free(dentry_cache, dentry);
1580 return NULL;
1581 }
8d85b484
AV
1582 atomic_set(&p->u.count, 1);
1583 dname = p->name;
df4c0e36
AR
1584 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1585 kasan_unpoison_shadow(dname,
1586 round_up(name->len + 1, sizeof(unsigned long)));
1da177e4
LT
1587 } else {
1588 dname = dentry->d_iname;
1589 }
1da177e4
LT
1590
1591 dentry->d_name.len = name->len;
1592 dentry->d_name.hash = name->hash;
1593 memcpy(dname, name->name, name->len);
1594 dname[name->len] = 0;
1595
6326c71f
LT
1596 /* Make sure we always see the terminating NUL character */
1597 smp_wmb();
1598 dentry->d_name.name = dname;
1599
98474236 1600 dentry->d_lockref.count = 1;
dea3667b 1601 dentry->d_flags = 0;
1da177e4 1602 spin_lock_init(&dentry->d_lock);
31e6b01f 1603 seqcount_init(&dentry->d_seq);
1da177e4 1604 dentry->d_inode = NULL;
a4464dbc
AV
1605 dentry->d_parent = dentry;
1606 dentry->d_sb = sb;
1da177e4
LT
1607 dentry->d_op = NULL;
1608 dentry->d_fsdata = NULL;
ceb5bdc2 1609 INIT_HLIST_BL_NODE(&dentry->d_hash);
1da177e4
LT
1610 INIT_LIST_HEAD(&dentry->d_lru);
1611 INIT_LIST_HEAD(&dentry->d_subdirs);
946e51f2
AV
1612 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1613 INIT_LIST_HEAD(&dentry->d_child);
a4464dbc 1614 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1da177e4 1615
3e880fb5 1616 this_cpu_inc(nr_dentry);
312d3ca8 1617
1da177e4
LT
1618 return dentry;
1619}
a4464dbc
AV
1620
1621/**
1622 * d_alloc - allocate a dcache entry
1623 * @parent: parent of entry to allocate
1624 * @name: qstr of the name
1625 *
1626 * Allocates a dentry. It returns %NULL if there is insufficient memory
1627 * available. On a success the dentry is returned. The name passed in is
1628 * copied and the copy passed in may be reused after this call.
1629 */
1630struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1631{
1632 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1633 if (!dentry)
1634 return NULL;
1635
1636 spin_lock(&parent->d_lock);
1637 /*
1638 * don't need child lock because it is not subject
1639 * to concurrency here
1640 */
1641 __dget_dlock(parent);
1642 dentry->d_parent = parent;
946e51f2 1643 list_add(&dentry->d_child, &parent->d_subdirs);
a4464dbc
AV
1644 spin_unlock(&parent->d_lock);
1645
1646 return dentry;
1647}
ec4f8605 1648EXPORT_SYMBOL(d_alloc);
1da177e4 1649
e1a24bb0
BF
1650/**
1651 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1652 * @sb: the superblock
1653 * @name: qstr of the name
1654 *
1655 * For a filesystem that just pins its dentries in memory and never
1656 * performs lookups at all, return an unhashed IS_ROOT dentry.
1657 */
4b936885
NP
1658struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1659{
e1a24bb0 1660 return __d_alloc(sb, name);
4b936885
NP
1661}
1662EXPORT_SYMBOL(d_alloc_pseudo);
1663
1da177e4
LT
1664struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1665{
1666 struct qstr q;
1667
1668 q.name = name;
1669 q.len = strlen(name);
1670 q.hash = full_name_hash(q.name, q.len);
1671 return d_alloc(parent, &q);
1672}
ef26ca97 1673EXPORT_SYMBOL(d_alloc_name);
1da177e4 1674
fb045adb
NP
1675void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1676{
6f7f7caa
LT
1677 WARN_ON_ONCE(dentry->d_op);
1678 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
fb045adb
NP
1679 DCACHE_OP_COMPARE |
1680 DCACHE_OP_REVALIDATE |
ecf3d1f1 1681 DCACHE_OP_WEAK_REVALIDATE |
4bacc9c9 1682 DCACHE_OP_DELETE |
d101a125
MS
1683 DCACHE_OP_SELECT_INODE |
1684 DCACHE_OP_REAL));
fb045adb
NP
1685 dentry->d_op = op;
1686 if (!op)
1687 return;
1688 if (op->d_hash)
1689 dentry->d_flags |= DCACHE_OP_HASH;
1690 if (op->d_compare)
1691 dentry->d_flags |= DCACHE_OP_COMPARE;
1692 if (op->d_revalidate)
1693 dentry->d_flags |= DCACHE_OP_REVALIDATE;
ecf3d1f1
JL
1694 if (op->d_weak_revalidate)
1695 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
fb045adb
NP
1696 if (op->d_delete)
1697 dentry->d_flags |= DCACHE_OP_DELETE;
f0023bc6
SW
1698 if (op->d_prune)
1699 dentry->d_flags |= DCACHE_OP_PRUNE;
4bacc9c9
DH
1700 if (op->d_select_inode)
1701 dentry->d_flags |= DCACHE_OP_SELECT_INODE;
d101a125
MS
1702 if (op->d_real)
1703 dentry->d_flags |= DCACHE_OP_REAL;
fb045adb
NP
1704
1705}
1706EXPORT_SYMBOL(d_set_d_op);
1707
df1a085a
DH
1708
1709/*
1710 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1711 * @dentry - The dentry to mark
1712 *
1713 * Mark a dentry as falling through to the lower layer (as set with
1714 * d_pin_lower()). This flag may be recorded on the medium.
1715 */
1716void d_set_fallthru(struct dentry *dentry)
1717{
1718 spin_lock(&dentry->d_lock);
1719 dentry->d_flags |= DCACHE_FALLTHRU;
1720 spin_unlock(&dentry->d_lock);
1721}
1722EXPORT_SYMBOL(d_set_fallthru);
1723
b18825a7
DH
1724static unsigned d_flags_for_inode(struct inode *inode)
1725{
44bdb5e5 1726 unsigned add_flags = DCACHE_REGULAR_TYPE;
b18825a7
DH
1727
1728 if (!inode)
1729 return DCACHE_MISS_TYPE;
1730
1731 if (S_ISDIR(inode->i_mode)) {
1732 add_flags = DCACHE_DIRECTORY_TYPE;
1733 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1734 if (unlikely(!inode->i_op->lookup))
1735 add_flags = DCACHE_AUTODIR_TYPE;
1736 else
1737 inode->i_opflags |= IOP_LOOKUP;
1738 }
44bdb5e5
DH
1739 goto type_determined;
1740 }
1741
1742 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
6b255391 1743 if (unlikely(inode->i_op->get_link)) {
b18825a7 1744 add_flags = DCACHE_SYMLINK_TYPE;
44bdb5e5
DH
1745 goto type_determined;
1746 }
1747 inode->i_opflags |= IOP_NOFOLLOW;
b18825a7
DH
1748 }
1749
44bdb5e5
DH
1750 if (unlikely(!S_ISREG(inode->i_mode)))
1751 add_flags = DCACHE_SPECIAL_TYPE;
1752
1753type_determined:
b18825a7
DH
1754 if (unlikely(IS_AUTOMOUNT(inode)))
1755 add_flags |= DCACHE_NEED_AUTOMOUNT;
1756 return add_flags;
1757}
1758
360da900
OH
1759static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1760{
b18825a7 1761 unsigned add_flags = d_flags_for_inode(inode);
85c7f810 1762 WARN_ON(d_in_lookup(dentry));
b18825a7 1763
b23fb0a6 1764 spin_lock(&dentry->d_lock);
de689f5e 1765 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
a528aca7 1766 raw_write_seqcount_begin(&dentry->d_seq);
4bf46a27 1767 __d_set_inode_and_type(dentry, inode, add_flags);
a528aca7 1768 raw_write_seqcount_end(&dentry->d_seq);
de689f5e 1769 __fsnotify_d_instantiate(dentry);
b23fb0a6 1770 spin_unlock(&dentry->d_lock);
360da900
OH
1771}
1772
1da177e4
LT
1773/**
1774 * d_instantiate - fill in inode information for a dentry
1775 * @entry: dentry to complete
1776 * @inode: inode to attach to this dentry
1777 *
1778 * Fill in inode information in the entry.
1779 *
1780 * This turns negative dentries into productive full members
1781 * of society.
1782 *
1783 * NOTE! This assumes that the inode count has been incremented
1784 * (or otherwise set) by the caller to indicate that it is now
1785 * in use by the dcache.
1786 */
1787
1788void d_instantiate(struct dentry *entry, struct inode * inode)
1789{
946e51f2 1790 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
de689f5e 1791 if (inode) {
b9680917 1792 security_d_instantiate(entry, inode);
873feea0 1793 spin_lock(&inode->i_lock);
de689f5e 1794 __d_instantiate(entry, inode);
873feea0 1795 spin_unlock(&inode->i_lock);
de689f5e 1796 }
1da177e4 1797}
ec4f8605 1798EXPORT_SYMBOL(d_instantiate);
1da177e4 1799
b70a80e7
MS
1800/**
1801 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1802 * @entry: dentry to complete
1803 * @inode: inode to attach to this dentry
1804 *
1805 * Fill in inode information in the entry. If a directory alias is found, then
1806 * return an error (and drop inode). Together with d_materialise_unique() this
1807 * guarantees that a directory inode may never have more than one alias.
1808 */
1809int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1810{
946e51f2 1811 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
b70a80e7 1812
b9680917 1813 security_d_instantiate(entry, inode);
b70a80e7
MS
1814 spin_lock(&inode->i_lock);
1815 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1816 spin_unlock(&inode->i_lock);
1817 iput(inode);
1818 return -EBUSY;
1819 }
1820 __d_instantiate(entry, inode);
1821 spin_unlock(&inode->i_lock);
b70a80e7
MS
1822
1823 return 0;
1824}
1825EXPORT_SYMBOL(d_instantiate_no_diralias);
1826
adc0e91a
AV
1827struct dentry *d_make_root(struct inode *root_inode)
1828{
1829 struct dentry *res = NULL;
1830
1831 if (root_inode) {
26fe5750 1832 static const struct qstr name = QSTR_INIT("/", 1);
adc0e91a
AV
1833
1834 res = __d_alloc(root_inode->i_sb, &name);
1835 if (res)
1836 d_instantiate(res, root_inode);
1837 else
1838 iput(root_inode);
1839 }
1840 return res;
1841}
1842EXPORT_SYMBOL(d_make_root);
1843
d891eedb
BF
1844static struct dentry * __d_find_any_alias(struct inode *inode)
1845{
1846 struct dentry *alias;
1847
b3d9b7a3 1848 if (hlist_empty(&inode->i_dentry))
d891eedb 1849 return NULL;
946e51f2 1850 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
d891eedb
BF
1851 __dget(alias);
1852 return alias;
1853}
1854
46f72b34
SW
1855/**
1856 * d_find_any_alias - find any alias for a given inode
1857 * @inode: inode to find an alias for
1858 *
1859 * If any aliases exist for the given inode, take and return a
1860 * reference for one of them. If no aliases exist, return %NULL.
1861 */
1862struct dentry *d_find_any_alias(struct inode *inode)
d891eedb
BF
1863{
1864 struct dentry *de;
1865
1866 spin_lock(&inode->i_lock);
1867 de = __d_find_any_alias(inode);
1868 spin_unlock(&inode->i_lock);
1869 return de;
1870}
46f72b34 1871EXPORT_SYMBOL(d_find_any_alias);
d891eedb 1872
49c7dd28 1873static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
4ea3ada2 1874{
b911a6bd 1875 static const struct qstr anonstring = QSTR_INIT("/", 1);
9308a612
CH
1876 struct dentry *tmp;
1877 struct dentry *res;
b18825a7 1878 unsigned add_flags;
4ea3ada2
CH
1879
1880 if (!inode)
44003728 1881 return ERR_PTR(-ESTALE);
4ea3ada2
CH
1882 if (IS_ERR(inode))
1883 return ERR_CAST(inode);
1884
d891eedb 1885 res = d_find_any_alias(inode);
9308a612
CH
1886 if (res)
1887 goto out_iput;
1888
a4464dbc 1889 tmp = __d_alloc(inode->i_sb, &anonstring);
9308a612
CH
1890 if (!tmp) {
1891 res = ERR_PTR(-ENOMEM);
1892 goto out_iput;
4ea3ada2 1893 }
b5c84bf6 1894
b9680917 1895 security_d_instantiate(tmp, inode);
873feea0 1896 spin_lock(&inode->i_lock);
d891eedb 1897 res = __d_find_any_alias(inode);
9308a612 1898 if (res) {
873feea0 1899 spin_unlock(&inode->i_lock);
9308a612
CH
1900 dput(tmp);
1901 goto out_iput;
1902 }
1903
1904 /* attach a disconnected dentry */
1a0a397e
BF
1905 add_flags = d_flags_for_inode(inode);
1906
1907 if (disconnected)
1908 add_flags |= DCACHE_DISCONNECTED;
b18825a7 1909
9308a612 1910 spin_lock(&tmp->d_lock);
4bf46a27 1911 __d_set_inode_and_type(tmp, inode, add_flags);
946e51f2 1912 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1879fd6a 1913 hlist_bl_lock(&tmp->d_sb->s_anon);
ceb5bdc2 1914 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1879fd6a 1915 hlist_bl_unlock(&tmp->d_sb->s_anon);
9308a612 1916 spin_unlock(&tmp->d_lock);
873feea0 1917 spin_unlock(&inode->i_lock);
9308a612 1918
9308a612
CH
1919 return tmp;
1920
1921 out_iput:
1922 iput(inode);
1923 return res;
4ea3ada2 1924}
1a0a397e
BF
1925
1926/**
1927 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1928 * @inode: inode to allocate the dentry for
1929 *
1930 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1931 * similar open by handle operations. The returned dentry may be anonymous,
1932 * or may have a full name (if the inode was already in the cache).
1933 *
1934 * When called on a directory inode, we must ensure that the inode only ever
1935 * has one dentry. If a dentry is found, that is returned instead of
1936 * allocating a new one.
1937 *
1938 * On successful return, the reference to the inode has been transferred
1939 * to the dentry. In case of an error the reference on the inode is released.
1940 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1941 * be passed in and the error will be propagated to the return value,
1942 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1943 */
1944struct dentry *d_obtain_alias(struct inode *inode)
1945{
1946 return __d_obtain_alias(inode, 1);
1947}
adc48720 1948EXPORT_SYMBOL(d_obtain_alias);
1da177e4 1949
1a0a397e
BF
1950/**
1951 * d_obtain_root - find or allocate a dentry for a given inode
1952 * @inode: inode to allocate the dentry for
1953 *
1954 * Obtain an IS_ROOT dentry for the root of a filesystem.
1955 *
1956 * We must ensure that directory inodes only ever have one dentry. If a
1957 * dentry is found, that is returned instead of allocating a new one.
1958 *
1959 * On successful return, the reference to the inode has been transferred
1960 * to the dentry. In case of an error the reference on the inode is
1961 * released. A %NULL or IS_ERR inode may be passed in and will be the
1962 * error will be propagate to the return value, with a %NULL @inode
1963 * replaced by ERR_PTR(-ESTALE).
1964 */
1965struct dentry *d_obtain_root(struct inode *inode)
1966{
1967 return __d_obtain_alias(inode, 0);
1968}
1969EXPORT_SYMBOL(d_obtain_root);
1970
9403540c
BN
1971/**
1972 * d_add_ci - lookup or allocate new dentry with case-exact name
1973 * @inode: the inode case-insensitive lookup has found
1974 * @dentry: the negative dentry that was passed to the parent's lookup func
1975 * @name: the case-exact name to be associated with the returned dentry
1976 *
1977 * This is to avoid filling the dcache with case-insensitive names to the
1978 * same inode, only the actual correct case is stored in the dcache for
1979 * case-insensitive filesystems.
1980 *
1981 * For a case-insensitive lookup match and if the the case-exact dentry
1982 * already exists in in the dcache, use it and return it.
1983 *
1984 * If no entry exists with the exact case name, allocate new dentry with
1985 * the exact case, and return the spliced entry.
1986 */
e45b590b 1987struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
9403540c
BN
1988 struct qstr *name)
1989{
9403540c
BN
1990 struct dentry *found;
1991 struct dentry *new;
1992
b6520c81
CH
1993 /*
1994 * First check if a dentry matching the name already exists,
1995 * if not go ahead and create it now.
1996 */
9403540c 1997 found = d_hash_and_lookup(dentry->d_parent, name);
9403540c
BN
1998 if (!found) {
1999 new = d_alloc(dentry->d_parent, name);
2000 if (!new) {
4f522a24 2001 found = ERR_PTR(-ENOMEM);
427c77d4
AV
2002 } else {
2003 found = d_splice_alias(inode, new);
2004 if (found) {
2005 dput(new);
2006 return found;
2007 }
2008 return new;
9403540c 2009 }
9403540c 2010 }
9403540c 2011 iput(inode);
4f522a24 2012 return found;
9403540c 2013}
ec4f8605 2014EXPORT_SYMBOL(d_add_ci);
1da177e4 2015
12f8ad4b
LT
2016/*
2017 * Do the slow-case of the dentry name compare.
2018 *
2019 * Unlike the dentry_cmp() function, we need to atomically
da53be12 2020 * load the name and length information, so that the
12f8ad4b
LT
2021 * filesystem can rely on them, and can use the 'name' and
2022 * 'len' information without worrying about walking off the
2023 * end of memory etc.
2024 *
2025 * Thus the read_seqcount_retry() and the "duplicate" info
2026 * in arguments (the low-level filesystem should not look
2027 * at the dentry inode or name contents directly, since
2028 * rename can change them while we're in RCU mode).
2029 */
2030enum slow_d_compare {
2031 D_COMP_OK,
2032 D_COMP_NOMATCH,
2033 D_COMP_SEQRETRY,
2034};
2035
2036static noinline enum slow_d_compare slow_dentry_cmp(
2037 const struct dentry *parent,
12f8ad4b
LT
2038 struct dentry *dentry,
2039 unsigned int seq,
2040 const struct qstr *name)
2041{
2042 int tlen = dentry->d_name.len;
2043 const char *tname = dentry->d_name.name;
12f8ad4b
LT
2044
2045 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2046 cpu_relax();
2047 return D_COMP_SEQRETRY;
2048 }
da53be12 2049 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
12f8ad4b
LT
2050 return D_COMP_NOMATCH;
2051 return D_COMP_OK;
2052}
2053
31e6b01f
NP
2054/**
2055 * __d_lookup_rcu - search for a dentry (racy, store-free)
2056 * @parent: parent dentry
2057 * @name: qstr of name we wish to find
1f1e6e52 2058 * @seqp: returns d_seq value at the point where the dentry was found
31e6b01f
NP
2059 * Returns: dentry, or NULL
2060 *
2061 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2062 * resolution (store-free path walking) design described in
2063 * Documentation/filesystems/path-lookup.txt.
2064 *
2065 * This is not to be used outside core vfs.
2066 *
2067 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2068 * held, and rcu_read_lock held. The returned dentry must not be stored into
2069 * without taking d_lock and checking d_seq sequence count against @seq
2070 * returned here.
2071 *
15570086 2072 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
31e6b01f
NP
2073 * function.
2074 *
2075 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2076 * the returned dentry, so long as its parent's seqlock is checked after the
2077 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2078 * is formed, giving integrity down the path walk.
12f8ad4b
LT
2079 *
2080 * NOTE! The caller *has* to check the resulting dentry against the sequence
2081 * number we've returned before using any of the resulting dentry state!
31e6b01f 2082 */
8966be90
LT
2083struct dentry *__d_lookup_rcu(const struct dentry *parent,
2084 const struct qstr *name,
da53be12 2085 unsigned *seqp)
31e6b01f 2086{
26fe5750 2087 u64 hashlen = name->hash_len;
31e6b01f 2088 const unsigned char *str = name->name;
26fe5750 2089 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
ceb5bdc2 2090 struct hlist_bl_node *node;
31e6b01f
NP
2091 struct dentry *dentry;
2092
2093 /*
2094 * Note: There is significant duplication with __d_lookup_rcu which is
2095 * required to prevent single threaded performance regressions
2096 * especially on architectures where smp_rmb (in seqcounts) are costly.
2097 * Keep the two functions in sync.
2098 */
2099
2100 /*
2101 * The hash list is protected using RCU.
2102 *
2103 * Carefully use d_seq when comparing a candidate dentry, to avoid
2104 * races with d_move().
2105 *
2106 * It is possible that concurrent renames can mess up our list
2107 * walk here and result in missing our dentry, resulting in the
2108 * false-negative result. d_lookup() protects against concurrent
2109 * renames using rename_lock seqlock.
2110 *
b0a4bb83 2111 * See Documentation/filesystems/path-lookup.txt for more details.
31e6b01f 2112 */
b07ad996 2113 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
8966be90 2114 unsigned seq;
31e6b01f 2115
31e6b01f 2116seqretry:
12f8ad4b
LT
2117 /*
2118 * The dentry sequence count protects us from concurrent
da53be12 2119 * renames, and thus protects parent and name fields.
12f8ad4b
LT
2120 *
2121 * The caller must perform a seqcount check in order
da53be12 2122 * to do anything useful with the returned dentry.
12f8ad4b
LT
2123 *
2124 * NOTE! We do a "raw" seqcount_begin here. That means that
2125 * we don't wait for the sequence count to stabilize if it
2126 * is in the middle of a sequence change. If we do the slow
2127 * dentry compare, we will do seqretries until it is stable,
2128 * and if we end up with a successful lookup, we actually
2129 * want to exit RCU lookup anyway.
2130 */
2131 seq = raw_seqcount_begin(&dentry->d_seq);
31e6b01f
NP
2132 if (dentry->d_parent != parent)
2133 continue;
2e321806
LT
2134 if (d_unhashed(dentry))
2135 continue;
12f8ad4b 2136
830c0f0e 2137 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
26fe5750
LT
2138 if (dentry->d_name.hash != hashlen_hash(hashlen))
2139 continue;
da53be12
LT
2140 *seqp = seq;
2141 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
12f8ad4b
LT
2142 case D_COMP_OK:
2143 return dentry;
2144 case D_COMP_NOMATCH:
31e6b01f 2145 continue;
12f8ad4b
LT
2146 default:
2147 goto seqretry;
2148 }
31e6b01f 2149 }
12f8ad4b 2150
26fe5750 2151 if (dentry->d_name.hash_len != hashlen)
ee983e89 2152 continue;
da53be12 2153 *seqp = seq;
26fe5750 2154 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
12f8ad4b 2155 return dentry;
31e6b01f
NP
2156 }
2157 return NULL;
2158}
2159
1da177e4
LT
2160/**
2161 * d_lookup - search for a dentry
2162 * @parent: parent dentry
2163 * @name: qstr of name we wish to find
b04f784e 2164 * Returns: dentry, or NULL
1da177e4 2165 *
b04f784e
NP
2166 * d_lookup searches the children of the parent dentry for the name in
2167 * question. If the dentry is found its reference count is incremented and the
2168 * dentry is returned. The caller must use dput to free the entry when it has
2169 * finished using it. %NULL is returned if the dentry does not exist.
1da177e4 2170 */
da2d8455 2171struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
1da177e4 2172{
31e6b01f 2173 struct dentry *dentry;
949854d0 2174 unsigned seq;
1da177e4 2175
b8314f93
DY
2176 do {
2177 seq = read_seqbegin(&rename_lock);
2178 dentry = __d_lookup(parent, name);
2179 if (dentry)
1da177e4
LT
2180 break;
2181 } while (read_seqretry(&rename_lock, seq));
2182 return dentry;
2183}
ec4f8605 2184EXPORT_SYMBOL(d_lookup);
1da177e4 2185
31e6b01f 2186/**
b04f784e
NP
2187 * __d_lookup - search for a dentry (racy)
2188 * @parent: parent dentry
2189 * @name: qstr of name we wish to find
2190 * Returns: dentry, or NULL
2191 *
2192 * __d_lookup is like d_lookup, however it may (rarely) return a
2193 * false-negative result due to unrelated rename activity.
2194 *
2195 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2196 * however it must be used carefully, eg. with a following d_lookup in
2197 * the case of failure.
2198 *
2199 * __d_lookup callers must be commented.
2200 */
a713ca2a 2201struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
1da177e4
LT
2202{
2203 unsigned int len = name->len;
2204 unsigned int hash = name->hash;
2205 const unsigned char *str = name->name;
b07ad996 2206 struct hlist_bl_head *b = d_hash(parent, hash);
ceb5bdc2 2207 struct hlist_bl_node *node;
31e6b01f 2208 struct dentry *found = NULL;
665a7583 2209 struct dentry *dentry;
1da177e4 2210
31e6b01f
NP
2211 /*
2212 * Note: There is significant duplication with __d_lookup_rcu which is
2213 * required to prevent single threaded performance regressions
2214 * especially on architectures where smp_rmb (in seqcounts) are costly.
2215 * Keep the two functions in sync.
2216 */
2217
b04f784e
NP
2218 /*
2219 * The hash list is protected using RCU.
2220 *
2221 * Take d_lock when comparing a candidate dentry, to avoid races
2222 * with d_move().
2223 *
2224 * It is possible that concurrent renames can mess up our list
2225 * walk here and result in missing our dentry, resulting in the
2226 * false-negative result. d_lookup() protects against concurrent
2227 * renames using rename_lock seqlock.
2228 *
b0a4bb83 2229 * See Documentation/filesystems/path-lookup.txt for more details.
b04f784e 2230 */
1da177e4
LT
2231 rcu_read_lock();
2232
b07ad996 2233 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1da177e4 2234
1da177e4
LT
2235 if (dentry->d_name.hash != hash)
2236 continue;
1da177e4
LT
2237
2238 spin_lock(&dentry->d_lock);
1da177e4
LT
2239 if (dentry->d_parent != parent)
2240 goto next;
d0185c08
LT
2241 if (d_unhashed(dentry))
2242 goto next;
2243
1da177e4
LT
2244 /*
2245 * It is safe to compare names since d_move() cannot
2246 * change the qstr (protected by d_lock).
2247 */
fb045adb 2248 if (parent->d_flags & DCACHE_OP_COMPARE) {
12f8ad4b
LT
2249 int tlen = dentry->d_name.len;
2250 const char *tname = dentry->d_name.name;
da53be12 2251 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1da177e4
LT
2252 goto next;
2253 } else {
ee983e89
LT
2254 if (dentry->d_name.len != len)
2255 goto next;
12f8ad4b 2256 if (dentry_cmp(dentry, str, len))
1da177e4
LT
2257 goto next;
2258 }
2259
98474236 2260 dentry->d_lockref.count++;
d0185c08 2261 found = dentry;
1da177e4
LT
2262 spin_unlock(&dentry->d_lock);
2263 break;
2264next:
2265 spin_unlock(&dentry->d_lock);
2266 }
2267 rcu_read_unlock();
2268
2269 return found;
2270}
2271
3e7e241f
EB
2272/**
2273 * d_hash_and_lookup - hash the qstr then search for a dentry
2274 * @dir: Directory to search in
2275 * @name: qstr of name we wish to find
2276 *
4f522a24 2277 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
3e7e241f
EB
2278 */
2279struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2280{
3e7e241f
EB
2281 /*
2282 * Check for a fs-specific hash function. Note that we must
2283 * calculate the standard hash first, as the d_op->d_hash()
2284 * routine may choose to leave the hash value unchanged.
2285 */
2286 name->hash = full_name_hash(name->name, name->len);
fb045adb 2287 if (dir->d_flags & DCACHE_OP_HASH) {
da53be12 2288 int err = dir->d_op->d_hash(dir, name);
4f522a24
AV
2289 if (unlikely(err < 0))
2290 return ERR_PTR(err);
3e7e241f 2291 }
4f522a24 2292 return d_lookup(dir, name);
3e7e241f 2293}
4f522a24 2294EXPORT_SYMBOL(d_hash_and_lookup);
3e7e241f 2295
1da177e4
LT
2296/*
2297 * When a file is deleted, we have two options:
2298 * - turn this dentry into a negative dentry
2299 * - unhash this dentry and free it.
2300 *
2301 * Usually, we want to just turn this into
2302 * a negative dentry, but if anybody else is
2303 * currently using the dentry or the inode
2304 * we can't do that and we fall back on removing
2305 * it from the hash queues and waiting for
2306 * it to be deleted later when it has no users
2307 */
2308
2309/**
2310 * d_delete - delete a dentry
2311 * @dentry: The dentry to delete
2312 *
2313 * Turn the dentry into a negative dentry if possible, otherwise
2314 * remove it from the hash queues so it can be deleted later
2315 */
2316
2317void d_delete(struct dentry * dentry)
2318{
873feea0 2319 struct inode *inode;
7a91bf7f 2320 int isdir = 0;
1da177e4
LT
2321 /*
2322 * Are we the only user?
2323 */
357f8e65 2324again:
1da177e4 2325 spin_lock(&dentry->d_lock);
873feea0
NP
2326 inode = dentry->d_inode;
2327 isdir = S_ISDIR(inode->i_mode);
98474236 2328 if (dentry->d_lockref.count == 1) {
1fe0c023 2329 if (!spin_trylock(&inode->i_lock)) {
357f8e65
NP
2330 spin_unlock(&dentry->d_lock);
2331 cpu_relax();
2332 goto again;
2333 }
13e3c5e5 2334 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
31e6b01f 2335 dentry_unlink_inode(dentry);
7a91bf7f 2336 fsnotify_nameremove(dentry, isdir);
1da177e4
LT
2337 return;
2338 }
2339
2340 if (!d_unhashed(dentry))
2341 __d_drop(dentry);
2342
2343 spin_unlock(&dentry->d_lock);
7a91bf7f
JM
2344
2345 fsnotify_nameremove(dentry, isdir);
1da177e4 2346}
ec4f8605 2347EXPORT_SYMBOL(d_delete);
1da177e4 2348
b07ad996 2349static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
1da177e4 2350{
ceb5bdc2 2351 BUG_ON(!d_unhashed(entry));
1879fd6a 2352 hlist_bl_lock(b);
dea3667b 2353 entry->d_flags |= DCACHE_RCUACCESS;
b07ad996 2354 hlist_bl_add_head_rcu(&entry->d_hash, b);
1879fd6a 2355 hlist_bl_unlock(b);
1da177e4
LT
2356}
2357
770bfad8
DH
2358static void _d_rehash(struct dentry * entry)
2359{
2360 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2361}
2362
1da177e4
LT
2363/**
2364 * d_rehash - add an entry back to the hash
2365 * @entry: dentry to add to the hash
2366 *
2367 * Adds a dentry to the hash according to its name.
2368 */
2369
2370void d_rehash(struct dentry * entry)
2371{
1da177e4 2372 spin_lock(&entry->d_lock);
770bfad8 2373 _d_rehash(entry);
1da177e4 2374 spin_unlock(&entry->d_lock);
1da177e4 2375}
ec4f8605 2376EXPORT_SYMBOL(d_rehash);
1da177e4 2377
84e710da
AV
2378static inline unsigned start_dir_add(struct inode *dir)
2379{
2380
2381 for (;;) {
2382 unsigned n = dir->i_dir_seq;
2383 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2384 return n;
2385 cpu_relax();
2386 }
2387}
2388
2389static inline void end_dir_add(struct inode *dir, unsigned n)
2390{
2391 smp_store_release(&dir->i_dir_seq, n + 2);
2392}
2393
94bdd655
AV
2394struct dentry *d_alloc_parallel(struct dentry *parent,
2395 const struct qstr *name)
2396{
2397 unsigned int len = name->len;
2398 unsigned int hash = name->hash;
2399 const unsigned char *str = name->name;
2400 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2401 struct hlist_bl_node *node;
2402 struct dentry *new = d_alloc(parent, name);
2403 struct dentry *dentry;
2404 unsigned seq, r_seq, d_seq;
2405
2406 if (unlikely(!new))
2407 return ERR_PTR(-ENOMEM);
2408
2409retry:
2410 rcu_read_lock();
2411 seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1;
2412 r_seq = read_seqbegin(&rename_lock);
2413 dentry = __d_lookup_rcu(parent, name, &d_seq);
2414 if (unlikely(dentry)) {
2415 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2416 rcu_read_unlock();
2417 goto retry;
2418 }
2419 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2420 rcu_read_unlock();
2421 dput(dentry);
2422 goto retry;
2423 }
2424 rcu_read_unlock();
2425 dput(new);
2426 return dentry;
2427 }
2428 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2429 rcu_read_unlock();
2430 goto retry;
2431 }
2432 hlist_bl_lock(b);
2433 if (unlikely(parent->d_inode->i_dir_seq != seq)) {
2434 hlist_bl_unlock(b);
2435 rcu_read_unlock();
2436 goto retry;
2437 }
2438 rcu_read_unlock();
2439 /*
2440 * No changes for the parent since the beginning of d_lookup().
2441 * Since all removals from the chain happen with hlist_bl_lock(),
2442 * any potential in-lookup matches are going to stay here until
2443 * we unlock the chain. All fields are stable in everything
2444 * we encounter.
2445 */
2446 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2447 if (dentry->d_name.hash != hash)
2448 continue;
2449 if (dentry->d_parent != parent)
2450 continue;
2451 if (d_unhashed(dentry))
2452 continue;
2453 if (parent->d_flags & DCACHE_OP_COMPARE) {
2454 int tlen = dentry->d_name.len;
2455 const char *tname = dentry->d_name.name;
2456 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2457 continue;
2458 } else {
2459 if (dentry->d_name.len != len)
2460 continue;
2461 if (dentry_cmp(dentry, str, len))
2462 continue;
2463 }
2464 dget(dentry);
2465 hlist_bl_unlock(b);
2466 /* impossible until we actually enable parallel lookups */
2467 BUG();
2468 /* and this will be "wait for it to stop being in-lookup" */
2469 /* this one will be handled in the next commit */
2470 dput(new);
2471 return dentry;
2472 }
2473 /* we can't take ->d_lock here; it's OK, though. */
2474 new->d_flags |= DCACHE_PAR_LOOKUP;
2475 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2476 hlist_bl_unlock(b);
2477 return new;
2478}
2479EXPORT_SYMBOL(d_alloc_parallel);
2480
85c7f810
AV
2481void __d_lookup_done(struct dentry *dentry)
2482{
94bdd655
AV
2483 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2484 dentry->d_name.hash);
2485 hlist_bl_lock(b);
85c7f810 2486 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
94bdd655
AV
2487 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2488 hlist_bl_unlock(b);
2489 INIT_HLIST_NODE(&dentry->d_u.d_alias);
85c7f810
AV
2490 /* more stuff will land here */
2491}
2492EXPORT_SYMBOL(__d_lookup_done);
ed782b5a
AV
2493
2494/* inode->i_lock held if inode is non-NULL */
2495
2496static inline void __d_add(struct dentry *dentry, struct inode *inode)
2497{
84e710da
AV
2498 struct inode *dir = NULL;
2499 unsigned n;
0568d705 2500 spin_lock(&dentry->d_lock);
84e710da
AV
2501 if (unlikely(d_in_lookup(dentry))) {
2502 dir = dentry->d_parent->d_inode;
2503 n = start_dir_add(dir);
85c7f810 2504 __d_lookup_done(dentry);
84e710da 2505 }
ed782b5a 2506 if (inode) {
0568d705
AV
2507 unsigned add_flags = d_flags_for_inode(inode);
2508 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2509 raw_write_seqcount_begin(&dentry->d_seq);
2510 __d_set_inode_and_type(dentry, inode, add_flags);
2511 raw_write_seqcount_end(&dentry->d_seq);
2512 __fsnotify_d_instantiate(dentry);
ed782b5a 2513 }
0568d705 2514 _d_rehash(dentry);
84e710da
AV
2515 if (dir)
2516 end_dir_add(dir, n);
0568d705
AV
2517 spin_unlock(&dentry->d_lock);
2518 if (inode)
2519 spin_unlock(&inode->i_lock);
ed782b5a
AV
2520}
2521
34d0d19d
AV
2522/**
2523 * d_add - add dentry to hash queues
2524 * @entry: dentry to add
2525 * @inode: The inode to attach to this dentry
2526 *
2527 * This adds the entry to the hash queues and initializes @inode.
2528 * The entry was actually filled in earlier during d_alloc().
2529 */
2530
2531void d_add(struct dentry *entry, struct inode *inode)
2532{
b9680917
AV
2533 if (inode) {
2534 security_d_instantiate(entry, inode);
ed782b5a 2535 spin_lock(&inode->i_lock);
b9680917 2536 }
ed782b5a 2537 __d_add(entry, inode);
34d0d19d
AV
2538}
2539EXPORT_SYMBOL(d_add);
2540
668d0cd5
AV
2541/**
2542 * d_exact_alias - find and hash an exact unhashed alias
2543 * @entry: dentry to add
2544 * @inode: The inode to go with this dentry
2545 *
2546 * If an unhashed dentry with the same name/parent and desired
2547 * inode already exists, hash and return it. Otherwise, return
2548 * NULL.
2549 *
2550 * Parent directory should be locked.
2551 */
2552struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2553{
2554 struct dentry *alias;
2555 int len = entry->d_name.len;
2556 const char *name = entry->d_name.name;
2557 unsigned int hash = entry->d_name.hash;
2558
2559 spin_lock(&inode->i_lock);
2560 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2561 /*
2562 * Don't need alias->d_lock here, because aliases with
2563 * d_parent == entry->d_parent are not subject to name or
2564 * parent changes, because the parent inode i_mutex is held.
2565 */
2566 if (alias->d_name.hash != hash)
2567 continue;
2568 if (alias->d_parent != entry->d_parent)
2569 continue;
2570 if (alias->d_name.len != len)
2571 continue;
2572 if (dentry_cmp(alias, name, len))
2573 continue;
2574 spin_lock(&alias->d_lock);
2575 if (!d_unhashed(alias)) {
2576 spin_unlock(&alias->d_lock);
2577 alias = NULL;
2578 } else {
2579 __dget_dlock(alias);
2580 _d_rehash(alias);
2581 spin_unlock(&alias->d_lock);
2582 }
2583 spin_unlock(&inode->i_lock);
2584 return alias;
2585 }
2586 spin_unlock(&inode->i_lock);
2587 return NULL;
2588}
2589EXPORT_SYMBOL(d_exact_alias);
2590
fb2d5b86
NP
2591/**
2592 * dentry_update_name_case - update case insensitive dentry with a new name
2593 * @dentry: dentry to be updated
2594 * @name: new name
2595 *
2596 * Update a case insensitive dentry with new case of name.
2597 *
2598 * dentry must have been returned by d_lookup with name @name. Old and new
2599 * name lengths must match (ie. no d_compare which allows mismatched name
2600 * lengths).
2601 *
2602 * Parent inode i_mutex must be held over d_lookup and into this call (to
2603 * keep renames and concurrent inserts, and readdir(2) away).
2604 */
2605void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2606{
5955102c 2607 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
fb2d5b86
NP
2608 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2609
fb2d5b86 2610 spin_lock(&dentry->d_lock);
31e6b01f 2611 write_seqcount_begin(&dentry->d_seq);
fb2d5b86 2612 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
31e6b01f 2613 write_seqcount_end(&dentry->d_seq);
fb2d5b86 2614 spin_unlock(&dentry->d_lock);
fb2d5b86
NP
2615}
2616EXPORT_SYMBOL(dentry_update_name_case);
2617
8d85b484 2618static void swap_names(struct dentry *dentry, struct dentry *target)
1da177e4 2619{
8d85b484
AV
2620 if (unlikely(dname_external(target))) {
2621 if (unlikely(dname_external(dentry))) {
1da177e4
LT
2622 /*
2623 * Both external: swap the pointers
2624 */
9a8d5bb4 2625 swap(target->d_name.name, dentry->d_name.name);
1da177e4
LT
2626 } else {
2627 /*
2628 * dentry:internal, target:external. Steal target's
2629 * storage and make target internal.
2630 */
321bcf92
BF
2631 memcpy(target->d_iname, dentry->d_name.name,
2632 dentry->d_name.len + 1);
1da177e4
LT
2633 dentry->d_name.name = target->d_name.name;
2634 target->d_name.name = target->d_iname;
2635 }
2636 } else {
8d85b484 2637 if (unlikely(dname_external(dentry))) {
1da177e4
LT
2638 /*
2639 * dentry:external, target:internal. Give dentry's
2640 * storage to target and make dentry internal
2641 */
2642 memcpy(dentry->d_iname, target->d_name.name,
2643 target->d_name.len + 1);
2644 target->d_name.name = dentry->d_name.name;
2645 dentry->d_name.name = dentry->d_iname;
2646 } else {
2647 /*
da1ce067 2648 * Both are internal.
1da177e4 2649 */
da1ce067
MS
2650 unsigned int i;
2651 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
08d4f772
MP
2652 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2653 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
da1ce067
MS
2654 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2655 swap(((long *) &dentry->d_iname)[i],
2656 ((long *) &target->d_iname)[i]);
2657 }
1da177e4
LT
2658 }
2659 }
a28ddb87 2660 swap(dentry->d_name.hash_len, target->d_name.hash_len);
1da177e4
LT
2661}
2662
8d85b484
AV
2663static void copy_name(struct dentry *dentry, struct dentry *target)
2664{
2665 struct external_name *old_name = NULL;
2666 if (unlikely(dname_external(dentry)))
2667 old_name = external_name(dentry);
2668 if (unlikely(dname_external(target))) {
2669 atomic_inc(&external_name(target)->u.count);
2670 dentry->d_name = target->d_name;
2671 } else {
2672 memcpy(dentry->d_iname, target->d_name.name,
2673 target->d_name.len + 1);
2674 dentry->d_name.name = dentry->d_iname;
2675 dentry->d_name.hash_len = target->d_name.hash_len;
2676 }
2677 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2678 kfree_rcu(old_name, u.head);
2679}
2680
2fd6b7f5
NP
2681static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2682{
2683 /*
2684 * XXXX: do we really need to take target->d_lock?
2685 */
2686 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2687 spin_lock(&target->d_parent->d_lock);
2688 else {
2689 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2690 spin_lock(&dentry->d_parent->d_lock);
2691 spin_lock_nested(&target->d_parent->d_lock,
2692 DENTRY_D_LOCK_NESTED);
2693 } else {
2694 spin_lock(&target->d_parent->d_lock);
2695 spin_lock_nested(&dentry->d_parent->d_lock,
2696 DENTRY_D_LOCK_NESTED);
2697 }
2698 }
2699 if (target < dentry) {
2700 spin_lock_nested(&target->d_lock, 2);
2701 spin_lock_nested(&dentry->d_lock, 3);
2702 } else {
2703 spin_lock_nested(&dentry->d_lock, 2);
2704 spin_lock_nested(&target->d_lock, 3);
2705 }
2706}
2707
986c0194 2708static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2fd6b7f5
NP
2709{
2710 if (target->d_parent != dentry->d_parent)
2711 spin_unlock(&dentry->d_parent->d_lock);
2712 if (target->d_parent != target)
2713 spin_unlock(&target->d_parent->d_lock);
986c0194
AV
2714 spin_unlock(&target->d_lock);
2715 spin_unlock(&dentry->d_lock);
2fd6b7f5
NP
2716}
2717
1da177e4 2718/*
2fd6b7f5
NP
2719 * When switching names, the actual string doesn't strictly have to
2720 * be preserved in the target - because we're dropping the target
2721 * anyway. As such, we can just do a simple memcpy() to copy over
d2fa4a84
ME
2722 * the new name before we switch, unless we are going to rehash
2723 * it. Note that if we *do* unhash the target, we are not allowed
2724 * to rehash it without giving it a new name/hash key - whether
2725 * we swap or overwrite the names here, resulting name won't match
2726 * the reality in filesystem; it's only there for d_path() purposes.
2727 * Note that all of this is happening under rename_lock, so the
2728 * any hash lookup seeing it in the middle of manipulations will
2729 * be discarded anyway. So we do not care what happens to the hash
2730 * key in that case.
1da177e4 2731 */
9eaef27b 2732/*
18367501 2733 * __d_move - move a dentry
1da177e4
LT
2734 * @dentry: entry to move
2735 * @target: new dentry
da1ce067 2736 * @exchange: exchange the two dentries
1da177e4
LT
2737 *
2738 * Update the dcache to reflect the move of a file name. Negative
c46c8877
JL
2739 * dcache entries should not be moved in this way. Caller must hold
2740 * rename_lock, the i_mutex of the source and target directories,
2741 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
1da177e4 2742 */
da1ce067
MS
2743static void __d_move(struct dentry *dentry, struct dentry *target,
2744 bool exchange)
1da177e4 2745{
84e710da
AV
2746 struct inode *dir = NULL;
2747 unsigned n;
1da177e4
LT
2748 if (!dentry->d_inode)
2749 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2750
2fd6b7f5
NP
2751 BUG_ON(d_ancestor(dentry, target));
2752 BUG_ON(d_ancestor(target, dentry));
2753
2fd6b7f5 2754 dentry_lock_for_move(dentry, target);
84e710da
AV
2755 if (unlikely(d_in_lookup(target))) {
2756 dir = target->d_parent->d_inode;
2757 n = start_dir_add(dir);
85c7f810 2758 __d_lookup_done(target);
84e710da 2759 }
1da177e4 2760
31e6b01f 2761 write_seqcount_begin(&dentry->d_seq);
1ca7d67c 2762 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
31e6b01f 2763
ceb5bdc2
NP
2764 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2765
2766 /*
2767 * Move the dentry to the target hash queue. Don't bother checking
2768 * for the same hash queue because of how unlikely it is.
2769 */
2770 __d_drop(dentry);
789680d1 2771 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
1da177e4 2772
da1ce067
MS
2773 /*
2774 * Unhash the target (d_delete() is not usable here). If exchanging
2775 * the two dentries, then rehash onto the other's hash queue.
2776 */
1da177e4 2777 __d_drop(target);
da1ce067
MS
2778 if (exchange) {
2779 __d_rehash(target,
2780 d_hash(dentry->d_parent, dentry->d_name.hash));
2781 }
1da177e4 2782
1da177e4 2783 /* Switch the names.. */
8d85b484
AV
2784 if (exchange)
2785 swap_names(dentry, target);
2786 else
2787 copy_name(dentry, target);
1da177e4 2788
63cf427a 2789 /* ... and switch them in the tree */
1da177e4 2790 if (IS_ROOT(dentry)) {
63cf427a 2791 /* splicing a tree */
1da177e4
LT
2792 dentry->d_parent = target->d_parent;
2793 target->d_parent = target;
946e51f2
AV
2794 list_del_init(&target->d_child);
2795 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
1da177e4 2796 } else {
63cf427a 2797 /* swapping two dentries */
9a8d5bb4 2798 swap(dentry->d_parent, target->d_parent);
946e51f2
AV
2799 list_move(&target->d_child, &target->d_parent->d_subdirs);
2800 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
63cf427a
AV
2801 if (exchange)
2802 fsnotify_d_move(target);
2803 fsnotify_d_move(dentry);
1da177e4
LT
2804 }
2805
31e6b01f
NP
2806 write_seqcount_end(&target->d_seq);
2807 write_seqcount_end(&dentry->d_seq);
2808
84e710da
AV
2809 if (dir)
2810 end_dir_add(dir, n);
986c0194 2811 dentry_unlock_for_move(dentry, target);
18367501
AV
2812}
2813
2814/*
2815 * d_move - move a dentry
2816 * @dentry: entry to move
2817 * @target: new dentry
2818 *
2819 * Update the dcache to reflect the move of a file name. Negative
c46c8877
JL
2820 * dcache entries should not be moved in this way. See the locking
2821 * requirements for __d_move.
18367501
AV
2822 */
2823void d_move(struct dentry *dentry, struct dentry *target)
2824{
2825 write_seqlock(&rename_lock);
da1ce067 2826 __d_move(dentry, target, false);
1da177e4 2827 write_sequnlock(&rename_lock);
9eaef27b 2828}
ec4f8605 2829EXPORT_SYMBOL(d_move);
1da177e4 2830
da1ce067
MS
2831/*
2832 * d_exchange - exchange two dentries
2833 * @dentry1: first dentry
2834 * @dentry2: second dentry
2835 */
2836void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2837{
2838 write_seqlock(&rename_lock);
2839
2840 WARN_ON(!dentry1->d_inode);
2841 WARN_ON(!dentry2->d_inode);
2842 WARN_ON(IS_ROOT(dentry1));
2843 WARN_ON(IS_ROOT(dentry2));
2844
2845 __d_move(dentry1, dentry2, true);
2846
2847 write_sequnlock(&rename_lock);
2848}
2849
e2761a11
OH
2850/**
2851 * d_ancestor - search for an ancestor
2852 * @p1: ancestor dentry
2853 * @p2: child dentry
2854 *
2855 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2856 * an ancestor of p2, else NULL.
9eaef27b 2857 */
e2761a11 2858struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
9eaef27b
TM
2859{
2860 struct dentry *p;
2861
871c0067 2862 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
9eaef27b 2863 if (p->d_parent == p1)
e2761a11 2864 return p;
9eaef27b 2865 }
e2761a11 2866 return NULL;
9eaef27b
TM
2867}
2868
2869/*
2870 * This helper attempts to cope with remotely renamed directories
2871 *
2872 * It assumes that the caller is already holding
a03e283b 2873 * dentry->d_parent->d_inode->i_mutex, and rename_lock
9eaef27b
TM
2874 *
2875 * Note: If ever the locking in lock_rename() changes, then please
2876 * remember to update this too...
9eaef27b 2877 */
b5ae6b15 2878static int __d_unalias(struct inode *inode,
873feea0 2879 struct dentry *dentry, struct dentry *alias)
9eaef27b
TM
2880{
2881 struct mutex *m1 = NULL, *m2 = NULL;
3d330dc1 2882 int ret = -ESTALE;
9eaef27b
TM
2883
2884 /* If alias and dentry share a parent, then no extra locks required */
2885 if (alias->d_parent == dentry->d_parent)
2886 goto out_unalias;
2887
9eaef27b 2888 /* See lock_rename() */
9eaef27b
TM
2889 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2890 goto out_err;
2891 m1 = &dentry->d_sb->s_vfs_rename_mutex;
5955102c 2892 if (!inode_trylock(alias->d_parent->d_inode))
9eaef27b
TM
2893 goto out_err;
2894 m2 = &alias->d_parent->d_inode->i_mutex;
2895out_unalias:
8ed936b5 2896 __d_move(alias, dentry, false);
b5ae6b15 2897 ret = 0;
9eaef27b 2898out_err:
9eaef27b
TM
2899 if (m2)
2900 mutex_unlock(m2);
2901 if (m1)
2902 mutex_unlock(m1);
2903 return ret;
2904}
2905
3f70bd51
BF
2906/**
2907 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2908 * @inode: the inode which may have a disconnected dentry
2909 * @dentry: a negative dentry which we want to point to the inode.
2910 *
da093a9b
BF
2911 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2912 * place of the given dentry and return it, else simply d_add the inode
2913 * to the dentry and return NULL.
3f70bd51 2914 *
908790fa
BF
2915 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2916 * we should error out: directories can't have multiple aliases.
2917 *
3f70bd51
BF
2918 * This is needed in the lookup routine of any filesystem that is exportable
2919 * (via knfsd) so that we can build dcache paths to directories effectively.
2920 *
2921 * If a dentry was found and moved, then it is returned. Otherwise NULL
2922 * is returned. This matches the expected return value of ->lookup.
2923 *
2924 * Cluster filesystems may call this function with a negative, hashed dentry.
2925 * In that case, we know that the inode will be a regular file, and also this
2926 * will only occur during atomic_open. So we need to check for the dentry
2927 * being already hashed only in the final case.
2928 */
2929struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2930{
3f70bd51
BF
2931 if (IS_ERR(inode))
2932 return ERR_CAST(inode);
2933
770bfad8
DH
2934 BUG_ON(!d_unhashed(dentry));
2935
de689f5e 2936 if (!inode)
b5ae6b15 2937 goto out;
de689f5e 2938
b9680917 2939 security_d_instantiate(dentry, inode);
873feea0 2940 spin_lock(&inode->i_lock);
9eaef27b 2941 if (S_ISDIR(inode->i_mode)) {
b5ae6b15
AV
2942 struct dentry *new = __d_find_any_alias(inode);
2943 if (unlikely(new)) {
a03e283b
EB
2944 /* The reference to new ensures it remains an alias */
2945 spin_unlock(&inode->i_lock);
18367501 2946 write_seqlock(&rename_lock);
b5ae6b15
AV
2947 if (unlikely(d_ancestor(new, dentry))) {
2948 write_sequnlock(&rename_lock);
b5ae6b15
AV
2949 dput(new);
2950 new = ERR_PTR(-ELOOP);
2951 pr_warn_ratelimited(
2952 "VFS: Lookup of '%s' in %s %s"
2953 " would have caused loop\n",
2954 dentry->d_name.name,
2955 inode->i_sb->s_type->name,
2956 inode->i_sb->s_id);
2957 } else if (!IS_ROOT(new)) {
2958 int err = __d_unalias(inode, dentry, new);
18367501 2959 write_sequnlock(&rename_lock);
b5ae6b15
AV
2960 if (err) {
2961 dput(new);
2962 new = ERR_PTR(err);
2963 }
18367501 2964 } else {
b5ae6b15
AV
2965 __d_move(new, dentry, false);
2966 write_sequnlock(&rename_lock);
dd179946 2967 }
b5ae6b15
AV
2968 iput(inode);
2969 return new;
9eaef27b 2970 }
770bfad8 2971 }
b5ae6b15 2972out:
ed782b5a 2973 __d_add(dentry, inode);
b5ae6b15 2974 return NULL;
770bfad8 2975}
b5ae6b15 2976EXPORT_SYMBOL(d_splice_alias);
770bfad8 2977
cdd16d02 2978static int prepend(char **buffer, int *buflen, const char *str, int namelen)
6092d048
RP
2979{
2980 *buflen -= namelen;
2981 if (*buflen < 0)
2982 return -ENAMETOOLONG;
2983 *buffer -= namelen;
2984 memcpy(*buffer, str, namelen);
2985 return 0;
2986}
2987
232d2d60
WL
2988/**
2989 * prepend_name - prepend a pathname in front of current buffer pointer
18129977
WL
2990 * @buffer: buffer pointer
2991 * @buflen: allocated length of the buffer
2992 * @name: name string and length qstr structure
232d2d60
WL
2993 *
2994 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2995 * make sure that either the old or the new name pointer and length are
2996 * fetched. However, there may be mismatch between length and pointer.
2997 * The length cannot be trusted, we need to copy it byte-by-byte until
2998 * the length is reached or a null byte is found. It also prepends "/" at
2999 * the beginning of the name. The sequence number check at the caller will
3000 * retry it again when a d_move() does happen. So any garbage in the buffer
3001 * due to mismatched pointer and length will be discarded.
6d13f694
AV
3002 *
3003 * Data dependency barrier is needed to make sure that we see that terminating
3004 * NUL. Alpha strikes again, film at 11...
232d2d60 3005 */
cdd16d02
MS
3006static int prepend_name(char **buffer, int *buflen, struct qstr *name)
3007{
232d2d60
WL
3008 const char *dname = ACCESS_ONCE(name->name);
3009 u32 dlen = ACCESS_ONCE(name->len);
3010 char *p;
3011
6d13f694
AV
3012 smp_read_barrier_depends();
3013
232d2d60 3014 *buflen -= dlen + 1;
e825196d
AV
3015 if (*buflen < 0)
3016 return -ENAMETOOLONG;
232d2d60
WL
3017 p = *buffer -= dlen + 1;
3018 *p++ = '/';
3019 while (dlen--) {
3020 char c = *dname++;
3021 if (!c)
3022 break;
3023 *p++ = c;
3024 }
3025 return 0;
cdd16d02
MS
3026}
3027
1da177e4 3028/**
208898c1 3029 * prepend_path - Prepend path string to a buffer
9d1bc601 3030 * @path: the dentry/vfsmount to report
02125a82 3031 * @root: root vfsmnt/dentry
f2eb6575
MS
3032 * @buffer: pointer to the end of the buffer
3033 * @buflen: pointer to buffer length
552ce544 3034 *
18129977
WL
3035 * The function will first try to write out the pathname without taking any
3036 * lock other than the RCU read lock to make sure that dentries won't go away.
3037 * It only checks the sequence number of the global rename_lock as any change
3038 * in the dentry's d_seq will be preceded by changes in the rename_lock
3039 * sequence number. If the sequence number had been changed, it will restart
3040 * the whole pathname back-tracing sequence again by taking the rename_lock.
3041 * In this case, there is no need to take the RCU read lock as the recursive
3042 * parent pointer references will keep the dentry chain alive as long as no
3043 * rename operation is performed.
1da177e4 3044 */
02125a82
AV
3045static int prepend_path(const struct path *path,
3046 const struct path *root,
f2eb6575 3047 char **buffer, int *buflen)
1da177e4 3048{
ede4cebc
AV
3049 struct dentry *dentry;
3050 struct vfsmount *vfsmnt;
3051 struct mount *mnt;
f2eb6575 3052 int error = 0;
48a066e7 3053 unsigned seq, m_seq = 0;
232d2d60
WL
3054 char *bptr;
3055 int blen;
6092d048 3056
48f5ec21 3057 rcu_read_lock();
48a066e7
AV
3058restart_mnt:
3059 read_seqbegin_or_lock(&mount_lock, &m_seq);
3060 seq = 0;
4ec6c2ae 3061 rcu_read_lock();
232d2d60
WL
3062restart:
3063 bptr = *buffer;
3064 blen = *buflen;
48a066e7 3065 error = 0;
ede4cebc
AV
3066 dentry = path->dentry;
3067 vfsmnt = path->mnt;
3068 mnt = real_mount(vfsmnt);
232d2d60 3069 read_seqbegin_or_lock(&rename_lock, &seq);
f2eb6575 3070 while (dentry != root->dentry || vfsmnt != root->mnt) {
1da177e4
LT
3071 struct dentry * parent;
3072
1da177e4 3073 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
48a066e7 3074 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
cde93be4
EB
3075 /* Escaped? */
3076 if (dentry != vfsmnt->mnt_root) {
3077 bptr = *buffer;
3078 blen = *buflen;
3079 error = 3;
3080 break;
3081 }
552ce544 3082 /* Global root? */
48a066e7
AV
3083 if (mnt != parent) {
3084 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
3085 mnt = parent;
232d2d60
WL
3086 vfsmnt = &mnt->mnt;
3087 continue;
3088 }
232d2d60
WL
3089 if (!error)
3090 error = is_mounted(vfsmnt) ? 1 : 2;
3091 break;
1da177e4
LT
3092 }
3093 parent = dentry->d_parent;
3094 prefetch(parent);
232d2d60 3095 error = prepend_name(&bptr, &blen, &dentry->d_name);
f2eb6575
MS
3096 if (error)
3097 break;
3098
1da177e4
LT
3099 dentry = parent;
3100 }
48f5ec21
AV
3101 if (!(seq & 1))
3102 rcu_read_unlock();
3103 if (need_seqretry(&rename_lock, seq)) {
3104 seq = 1;
232d2d60 3105 goto restart;
48f5ec21
AV
3106 }
3107 done_seqretry(&rename_lock, seq);
4ec6c2ae
LZ
3108
3109 if (!(m_seq & 1))
3110 rcu_read_unlock();
48a066e7
AV
3111 if (need_seqretry(&mount_lock, m_seq)) {
3112 m_seq = 1;
3113 goto restart_mnt;
3114 }
3115 done_seqretry(&mount_lock, m_seq);
1da177e4 3116
232d2d60
WL
3117 if (error >= 0 && bptr == *buffer) {
3118 if (--blen < 0)
3119 error = -ENAMETOOLONG;
3120 else
3121 *--bptr = '/';
3122 }
3123 *buffer = bptr;
3124 *buflen = blen;
7ea600b5 3125 return error;
f2eb6575 3126}
be285c71 3127
f2eb6575
MS
3128/**
3129 * __d_path - return the path of a dentry
3130 * @path: the dentry/vfsmount to report
02125a82 3131 * @root: root vfsmnt/dentry
cd956a1c 3132 * @buf: buffer to return value in
f2eb6575
MS
3133 * @buflen: buffer length
3134 *
ffd1f4ed 3135 * Convert a dentry into an ASCII path name.
f2eb6575
MS
3136 *
3137 * Returns a pointer into the buffer or an error code if the
3138 * path was too long.
3139 *
be148247 3140 * "buflen" should be positive.
f2eb6575 3141 *
02125a82 3142 * If the path is not reachable from the supplied root, return %NULL.
f2eb6575 3143 */
02125a82
AV
3144char *__d_path(const struct path *path,
3145 const struct path *root,
f2eb6575
MS
3146 char *buf, int buflen)
3147{
3148 char *res = buf + buflen;
3149 int error;
3150
3151 prepend(&res, &buflen, "\0", 1);
f2eb6575 3152 error = prepend_path(path, root, &res, &buflen);
be148247 3153
02125a82
AV
3154 if (error < 0)
3155 return ERR_PTR(error);
3156 if (error > 0)
3157 return NULL;
3158 return res;
3159}
3160
3161char *d_absolute_path(const struct path *path,
3162 char *buf, int buflen)
3163{
3164 struct path root = {};
3165 char *res = buf + buflen;
3166 int error;
3167
3168 prepend(&res, &buflen, "\0", 1);
02125a82 3169 error = prepend_path(path, &root, &res, &buflen);
02125a82
AV
3170
3171 if (error > 1)
3172 error = -EINVAL;
3173 if (error < 0)
f2eb6575 3174 return ERR_PTR(error);
f2eb6575 3175 return res;
1da177e4
LT
3176}
3177
ffd1f4ed
MS
3178/*
3179 * same as __d_path but appends "(deleted)" for unlinked files.
3180 */
02125a82
AV
3181static int path_with_deleted(const struct path *path,
3182 const struct path *root,
3183 char **buf, int *buflen)
ffd1f4ed
MS
3184{
3185 prepend(buf, buflen, "\0", 1);
3186 if (d_unlinked(path->dentry)) {
3187 int error = prepend(buf, buflen, " (deleted)", 10);
3188 if (error)
3189 return error;
3190 }
3191
3192 return prepend_path(path, root, buf, buflen);
3193}
3194
8df9d1a4
MS
3195static int prepend_unreachable(char **buffer, int *buflen)
3196{
3197 return prepend(buffer, buflen, "(unreachable)", 13);
3198}
3199
68f0d9d9
LT
3200static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3201{
3202 unsigned seq;
3203
3204 do {
3205 seq = read_seqcount_begin(&fs->seq);
3206 *root = fs->root;
3207 } while (read_seqcount_retry(&fs->seq, seq));
3208}
3209
a03a8a70
JB
3210/**
3211 * d_path - return the path of a dentry
cf28b486 3212 * @path: path to report
a03a8a70
JB
3213 * @buf: buffer to return value in
3214 * @buflen: buffer length
3215 *
3216 * Convert a dentry into an ASCII path name. If the entry has been deleted
3217 * the string " (deleted)" is appended. Note that this is ambiguous.
3218 *
52afeefb
AV
3219 * Returns a pointer into the buffer or an error code if the path was
3220 * too long. Note: Callers should use the returned pointer, not the passed
3221 * in buffer, to use the name! The implementation often starts at an offset
3222 * into the buffer, and may leave 0 bytes at the start.
a03a8a70 3223 *
31f3e0b3 3224 * "buflen" should be positive.
a03a8a70 3225 */
20d4fdc1 3226char *d_path(const struct path *path, char *buf, int buflen)
1da177e4 3227{
ffd1f4ed 3228 char *res = buf + buflen;
6ac08c39 3229 struct path root;
ffd1f4ed 3230 int error;
1da177e4 3231
c23fbb6b
ED
3232 /*
3233 * We have various synthetic filesystems that never get mounted. On
3234 * these filesystems dentries are never used for lookup purposes, and
3235 * thus don't need to be hashed. They also don't need a name until a
3236 * user wants to identify the object in /proc/pid/fd/. The little hack
3237 * below allows us to generate a name for these objects on demand:
f48cfddc
EB
3238 *
3239 * Some pseudo inodes are mountable. When they are mounted
3240 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3241 * and instead have d_path return the mounted path.
c23fbb6b 3242 */
f48cfddc
EB
3243 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3244 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
cf28b486 3245 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
c23fbb6b 3246
68f0d9d9
LT
3247 rcu_read_lock();
3248 get_fs_root_rcu(current->fs, &root);
02125a82 3249 error = path_with_deleted(path, &root, &res, &buflen);
68f0d9d9
LT
3250 rcu_read_unlock();
3251
02125a82 3252 if (error < 0)
ffd1f4ed 3253 res = ERR_PTR(error);
1da177e4
LT
3254 return res;
3255}
ec4f8605 3256EXPORT_SYMBOL(d_path);
1da177e4 3257
c23fbb6b
ED
3258/*
3259 * Helper function for dentry_operations.d_dname() members
3260 */
3261char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3262 const char *fmt, ...)
3263{
3264 va_list args;
3265 char temp[64];
3266 int sz;
3267
3268 va_start(args, fmt);
3269 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3270 va_end(args);
3271
3272 if (sz > sizeof(temp) || sz > buflen)
3273 return ERR_PTR(-ENAMETOOLONG);
3274
3275 buffer += buflen - sz;
3276 return memcpy(buffer, temp, sz);
3277}
3278
118b2302
AV
3279char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3280{
3281 char *end = buffer + buflen;
3282 /* these dentries are never renamed, so d_lock is not needed */
3283 if (prepend(&end, &buflen, " (deleted)", 11) ||
232d2d60 3284 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
118b2302
AV
3285 prepend(&end, &buflen, "/", 1))
3286 end = ERR_PTR(-ENAMETOOLONG);
232d2d60 3287 return end;
118b2302 3288}
31bbe16f 3289EXPORT_SYMBOL(simple_dname);
118b2302 3290
6092d048
RP
3291/*
3292 * Write full pathname from the root of the filesystem into the buffer.
3293 */
f6500801 3294static char *__dentry_path(struct dentry *d, char *buf, int buflen)
6092d048 3295{
f6500801 3296 struct dentry *dentry;
232d2d60
WL
3297 char *end, *retval;
3298 int len, seq = 0;
3299 int error = 0;
6092d048 3300
f6500801
AV
3301 if (buflen < 2)
3302 goto Elong;
3303
48f5ec21 3304 rcu_read_lock();
232d2d60 3305restart:
f6500801 3306 dentry = d;
232d2d60
WL
3307 end = buf + buflen;
3308 len = buflen;
3309 prepend(&end, &len, "\0", 1);
6092d048
RP
3310 /* Get '/' right */
3311 retval = end-1;
3312 *retval = '/';
232d2d60 3313 read_seqbegin_or_lock(&rename_lock, &seq);
cdd16d02
MS
3314 while (!IS_ROOT(dentry)) {
3315 struct dentry *parent = dentry->d_parent;
6092d048 3316
6092d048 3317 prefetch(parent);
232d2d60
WL
3318 error = prepend_name(&end, &len, &dentry->d_name);
3319 if (error)
3320 break;
6092d048
RP
3321
3322 retval = end;
3323 dentry = parent;
3324 }
48f5ec21
AV
3325 if (!(seq & 1))
3326 rcu_read_unlock();
3327 if (need_seqretry(&rename_lock, seq)) {
3328 seq = 1;
232d2d60 3329 goto restart;
48f5ec21
AV
3330 }
3331 done_seqretry(&rename_lock, seq);
232d2d60
WL
3332 if (error)
3333 goto Elong;
c103135c
AV
3334 return retval;
3335Elong:
3336 return ERR_PTR(-ENAMETOOLONG);
3337}
ec2447c2
NP
3338
3339char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3340{
232d2d60 3341 return __dentry_path(dentry, buf, buflen);
ec2447c2
NP
3342}
3343EXPORT_SYMBOL(dentry_path_raw);
c103135c
AV
3344
3345char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3346{
3347 char *p = NULL;
3348 char *retval;
3349
c103135c
AV
3350 if (d_unlinked(dentry)) {
3351 p = buf + buflen;
3352 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3353 goto Elong;
3354 buflen++;
3355 }
3356 retval = __dentry_path(dentry, buf, buflen);
c103135c
AV
3357 if (!IS_ERR(retval) && p)
3358 *p = '/'; /* restore '/' overriden with '\0' */
6092d048
RP
3359 return retval;
3360Elong:
6092d048
RP
3361 return ERR_PTR(-ENAMETOOLONG);
3362}
3363
8b19e341
LT
3364static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3365 struct path *pwd)
5762482f 3366{
8b19e341
LT
3367 unsigned seq;
3368
3369 do {
3370 seq = read_seqcount_begin(&fs->seq);
3371 *root = fs->root;
3372 *pwd = fs->pwd;
3373 } while (read_seqcount_retry(&fs->seq, seq));
5762482f
LT
3374}
3375
1da177e4
LT
3376/*
3377 * NOTE! The user-level library version returns a
3378 * character pointer. The kernel system call just
3379 * returns the length of the buffer filled (which
3380 * includes the ending '\0' character), or a negative
3381 * error value. So libc would do something like
3382 *
3383 * char *getcwd(char * buf, size_t size)
3384 * {
3385 * int retval;
3386 *
3387 * retval = sys_getcwd(buf, size);
3388 * if (retval >= 0)
3389 * return buf;
3390 * errno = -retval;
3391 * return NULL;
3392 * }
3393 */
3cdad428 3394SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
1da177e4 3395{
552ce544 3396 int error;
6ac08c39 3397 struct path pwd, root;
3272c544 3398 char *page = __getname();
1da177e4
LT
3399
3400 if (!page)
3401 return -ENOMEM;
3402
8b19e341
LT
3403 rcu_read_lock();
3404 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
1da177e4 3405
552ce544 3406 error = -ENOENT;
f3da392e 3407 if (!d_unlinked(pwd.dentry)) {
552ce544 3408 unsigned long len;
3272c544
LT
3409 char *cwd = page + PATH_MAX;
3410 int buflen = PATH_MAX;
1da177e4 3411
8df9d1a4 3412 prepend(&cwd, &buflen, "\0", 1);
02125a82 3413 error = prepend_path(&pwd, &root, &cwd, &buflen);
ff812d72 3414 rcu_read_unlock();
552ce544 3415
02125a82 3416 if (error < 0)
552ce544
LT
3417 goto out;
3418
8df9d1a4 3419 /* Unreachable from current root */
02125a82 3420 if (error > 0) {
8df9d1a4
MS
3421 error = prepend_unreachable(&cwd, &buflen);
3422 if (error)
3423 goto out;
3424 }
3425
552ce544 3426 error = -ERANGE;
3272c544 3427 len = PATH_MAX + page - cwd;
552ce544
LT
3428 if (len <= size) {
3429 error = len;
3430 if (copy_to_user(buf, cwd, len))
3431 error = -EFAULT;
3432 }
949854d0 3433 } else {
ff812d72 3434 rcu_read_unlock();
949854d0 3435 }
1da177e4
LT
3436
3437out:
3272c544 3438 __putname(page);
1da177e4
LT
3439 return error;
3440}
3441
3442/*
3443 * Test whether new_dentry is a subdirectory of old_dentry.
3444 *
3445 * Trivially implemented using the dcache structure
3446 */
3447
3448/**
3449 * is_subdir - is new dentry a subdirectory of old_dentry
3450 * @new_dentry: new dentry
3451 * @old_dentry: old dentry
3452 *
a6e5787f
YB
3453 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3454 * Returns false otherwise.
1da177e4
LT
3455 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3456 */
3457
a6e5787f 3458bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
1da177e4 3459{
a6e5787f 3460 bool result;
949854d0 3461 unsigned seq;
1da177e4 3462
e2761a11 3463 if (new_dentry == old_dentry)
a6e5787f 3464 return true;
e2761a11 3465
e2761a11 3466 do {
1da177e4 3467 /* for restarting inner loop in case of seq retry */
1da177e4 3468 seq = read_seqbegin(&rename_lock);
949854d0
NP
3469 /*
3470 * Need rcu_readlock to protect against the d_parent trashing
3471 * due to d_move
3472 */
3473 rcu_read_lock();
e2761a11 3474 if (d_ancestor(old_dentry, new_dentry))
a6e5787f 3475 result = true;
e2761a11 3476 else
a6e5787f 3477 result = false;
949854d0 3478 rcu_read_unlock();
1da177e4 3479 } while (read_seqretry(&rename_lock, seq));
1da177e4
LT
3480
3481 return result;
3482}
3483
db14fc3a 3484static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
1da177e4 3485{
db14fc3a
MS
3486 struct dentry *root = data;
3487 if (dentry != root) {
3488 if (d_unhashed(dentry) || !dentry->d_inode)
3489 return D_WALK_SKIP;
1da177e4 3490
01ddc4ed
MS
3491 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3492 dentry->d_flags |= DCACHE_GENOCIDE;
3493 dentry->d_lockref.count--;
3494 }
1da177e4 3495 }
db14fc3a
MS
3496 return D_WALK_CONTINUE;
3497}
58db63d0 3498
db14fc3a
MS
3499void d_genocide(struct dentry *parent)
3500{
3501 d_walk(parent, parent, d_genocide_kill, NULL);
1da177e4
LT
3502}
3503
60545d0d 3504void d_tmpfile(struct dentry *dentry, struct inode *inode)
1da177e4 3505{
60545d0d
AV
3506 inode_dec_link_count(inode);
3507 BUG_ON(dentry->d_name.name != dentry->d_iname ||
946e51f2 3508 !hlist_unhashed(&dentry->d_u.d_alias) ||
60545d0d
AV
3509 !d_unlinked(dentry));
3510 spin_lock(&dentry->d_parent->d_lock);
3511 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3512 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3513 (unsigned long long)inode->i_ino);
3514 spin_unlock(&dentry->d_lock);
3515 spin_unlock(&dentry->d_parent->d_lock);
3516 d_instantiate(dentry, inode);
1da177e4 3517}
60545d0d 3518EXPORT_SYMBOL(d_tmpfile);
1da177e4
LT
3519
3520static __initdata unsigned long dhash_entries;
3521static int __init set_dhash_entries(char *str)
3522{
3523 if (!str)
3524 return 0;
3525 dhash_entries = simple_strtoul(str, &str, 0);
3526 return 1;
3527}
3528__setup("dhash_entries=", set_dhash_entries);
3529
3530static void __init dcache_init_early(void)
3531{
074b8517 3532 unsigned int loop;
1da177e4
LT
3533
3534 /* If hashes are distributed across NUMA nodes, defer
3535 * hash allocation until vmalloc space is available.
3536 */
3537 if (hashdist)
3538 return;
3539
3540 dentry_hashtable =
3541 alloc_large_system_hash("Dentry cache",
b07ad996 3542 sizeof(struct hlist_bl_head),
1da177e4
LT
3543 dhash_entries,
3544 13,
3545 HASH_EARLY,
3546 &d_hash_shift,
3547 &d_hash_mask,
31fe62b9 3548 0,
1da177e4
LT
3549 0);
3550
074b8517 3551 for (loop = 0; loop < (1U << d_hash_shift); loop++)
b07ad996 3552 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
1da177e4
LT
3553}
3554
74bf17cf 3555static void __init dcache_init(void)
1da177e4 3556{
074b8517 3557 unsigned int loop;
1da177e4
LT
3558
3559 /*
3560 * A constructor could be added for stable state like the lists,
3561 * but it is probably not worth it because of the cache nature
3562 * of the dcache.
3563 */
0a31bd5f 3564 dentry_cache = KMEM_CACHE(dentry,
5d097056 3565 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
1da177e4
LT
3566
3567 /* Hash may have been set up in dcache_init_early */
3568 if (!hashdist)
3569 return;
3570
3571 dentry_hashtable =
3572 alloc_large_system_hash("Dentry cache",
b07ad996 3573 sizeof(struct hlist_bl_head),
1da177e4
LT
3574 dhash_entries,
3575 13,
3576 0,
3577 &d_hash_shift,
3578 &d_hash_mask,
31fe62b9 3579 0,
1da177e4
LT
3580 0);
3581
074b8517 3582 for (loop = 0; loop < (1U << d_hash_shift); loop++)
b07ad996 3583 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
1da177e4
LT
3584}
3585
3586/* SLAB cache for __getname() consumers */
e18b890b 3587struct kmem_cache *names_cachep __read_mostly;
ec4f8605 3588EXPORT_SYMBOL(names_cachep);
1da177e4 3589
1da177e4
LT
3590EXPORT_SYMBOL(d_genocide);
3591
1da177e4
LT
3592void __init vfs_caches_init_early(void)
3593{
3594 dcache_init_early();
3595 inode_init_early();
3596}
3597
4248b0da 3598void __init vfs_caches_init(void)
1da177e4 3599{
1da177e4 3600 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
20c2df83 3601 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1da177e4 3602
74bf17cf
DC
3603 dcache_init();
3604 inode_init();
4248b0da
MG
3605 files_init();
3606 files_maxfiles_init();
74bf17cf 3607 mnt_init();
1da177e4
LT
3608 bdev_cache_init();
3609 chrdev_init();
3610}