]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/blame - fs/dcache.c
dcache: Translating dentry into pathname without taking rename_lock
[mirror_ubuntu-focal-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>
07f3f05c 40#include "internal.h"
b2dba1af 41#include "mount.h"
1da177e4 42
789680d1
NP
43/*
44 * Usage:
873feea0
NP
45 * dcache->d_inode->i_lock protects:
46 * - i_dentry, d_alias, d_inode of aliases
ceb5bdc2
NP
47 * dcache_hash_bucket lock protects:
48 * - the dcache hash table
49 * s_anon bl list spinlock protects:
50 * - the s_anon list (see __d_drop)
23044507
NP
51 * dcache_lru_lock protects:
52 * - the dcache lru lists and counters
53 * d_lock protects:
54 * - d_flags
55 * - d_name
56 * - d_lru
b7ab39f6 57 * - d_count
da502956 58 * - d_unhashed()
2fd6b7f5
NP
59 * - d_parent and d_subdirs
60 * - childrens' d_child and d_parent
b23fb0a6 61 * - d_alias, d_inode
789680d1
NP
62 *
63 * Ordering:
873feea0 64 * dentry->d_inode->i_lock
b5c84bf6
NP
65 * dentry->d_lock
66 * dcache_lru_lock
ceb5bdc2
NP
67 * dcache_hash_bucket lock
68 * s_anon lock
789680d1 69 *
da502956
NP
70 * If there is an ancestor relationship:
71 * dentry->d_parent->...->d_parent->d_lock
72 * ...
73 * dentry->d_parent->d_lock
74 * dentry->d_lock
75 *
76 * If no ancestor relationship:
789680d1
NP
77 * if (dentry1 < dentry2)
78 * dentry1->d_lock
79 * dentry2->d_lock
80 */
fa3536cc 81int sysctl_vfs_cache_pressure __read_mostly = 100;
1da177e4
LT
82EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
83
23044507 84static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
74c3cbe3 85__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
1da177e4 86
949854d0 87EXPORT_SYMBOL(rename_lock);
1da177e4 88
e18b890b 89static struct kmem_cache *dentry_cache __read_mostly;
1da177e4 90
232d2d60
WL
91/**
92 * read_seqbegin_or_lock - begin a sequence number check or locking block
93 * lock: sequence lock
94 * seq : sequence number to be checked
95 *
96 * First try it once optimistically without taking the lock. If that fails,
97 * take the lock. The sequence number is also used as a marker for deciding
98 * whether to be a reader (even) or writer (odd).
99 * N.B. seq must be initialized to an even number to begin with.
100 */
101static inline void read_seqbegin_or_lock(seqlock_t *lock, int *seq)
102{
103 if (!(*seq & 1)) { /* Even */
104 *seq = read_seqbegin(lock);
105 rcu_read_lock();
106 } else /* Odd */
107 write_seqlock(lock);
108}
109
110/**
111 * read_seqretry_or_unlock - end a seqretry or lock block & return retry status
112 * lock : sequence lock
113 * seq : sequence number
114 * Return: 1 to retry operation again, 0 to continue
115 */
116static inline int read_seqretry_or_unlock(seqlock_t *lock, int *seq)
117{
118 if (!(*seq & 1)) { /* Even */
119 rcu_read_unlock();
120 if (read_seqretry(lock, *seq)) {
121 (*seq)++; /* Take writer lock */
122 return 1;
123 }
124 } else /* Odd */
125 write_sequnlock(lock);
126 return 0;
127}
128
1da177e4
LT
129/*
130 * This is the single most critical data structure when it comes
131 * to the dcache: the hashtable for lookups. Somebody should try
132 * to make this good - I've just made it work.
133 *
134 * This hash-function tries to avoid losing too many bits of hash
135 * information, yet avoid using a prime hash-size or similar.
136 */
137#define D_HASHBITS d_hash_shift
138#define D_HASHMASK d_hash_mask
139
fa3536cc
ED
140static unsigned int d_hash_mask __read_mostly;
141static unsigned int d_hash_shift __read_mostly;
ceb5bdc2 142
b07ad996 143static struct hlist_bl_head *dentry_hashtable __read_mostly;
ceb5bdc2 144
8966be90 145static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
6d7d1a0d 146 unsigned int hash)
ceb5bdc2 147{
6d7d1a0d
LT
148 hash += (unsigned long) parent / L1_CACHE_BYTES;
149 hash = hash + (hash >> D_HASHBITS);
ceb5bdc2
NP
150 return dentry_hashtable + (hash & D_HASHMASK);
151}
152
1da177e4
LT
153/* Statistics gathering. */
154struct dentry_stat_t dentry_stat = {
155 .age_limit = 45,
156};
157
3e880fb5 158static DEFINE_PER_CPU(unsigned int, nr_dentry);
312d3ca8
CH
159
160#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
3e880fb5
NP
161static int get_nr_dentry(void)
162{
163 int i;
164 int sum = 0;
165 for_each_possible_cpu(i)
166 sum += per_cpu(nr_dentry, i);
167 return sum < 0 ? 0 : sum;
168}
169
312d3ca8
CH
170int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
171 size_t *lenp, loff_t *ppos)
172{
3e880fb5 173 dentry_stat.nr_dentry = get_nr_dentry();
312d3ca8
CH
174 return proc_dointvec(table, write, buffer, lenp, ppos);
175}
176#endif
177
5483f18e
LT
178/*
179 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
180 * The strings are both count bytes long, and count is non-zero.
181 */
e419b4cc
LT
182#ifdef CONFIG_DCACHE_WORD_ACCESS
183
184#include <asm/word-at-a-time.h>
185/*
186 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
187 * aligned allocation for this particular component. We don't
188 * strictly need the load_unaligned_zeropad() safety, but it
189 * doesn't hurt either.
190 *
191 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
192 * need the careful unaligned handling.
193 */
94753db5 194static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
5483f18e 195{
bfcfaa77 196 unsigned long a,b,mask;
bfcfaa77
LT
197
198 for (;;) {
12f8ad4b 199 a = *(unsigned long *)cs;
e419b4cc 200 b = load_unaligned_zeropad(ct);
bfcfaa77
LT
201 if (tcount < sizeof(unsigned long))
202 break;
203 if (unlikely(a != b))
204 return 1;
205 cs += sizeof(unsigned long);
206 ct += sizeof(unsigned long);
207 tcount -= sizeof(unsigned long);
208 if (!tcount)
209 return 0;
210 }
211 mask = ~(~0ul << tcount*8);
212 return unlikely(!!((a ^ b) & mask));
e419b4cc
LT
213}
214
bfcfaa77 215#else
e419b4cc 216
94753db5 217static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
e419b4cc 218{
5483f18e
LT
219 do {
220 if (*cs != *ct)
221 return 1;
222 cs++;
223 ct++;
224 tcount--;
225 } while (tcount);
226 return 0;
227}
228
e419b4cc
LT
229#endif
230
94753db5
LT
231static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
232{
6326c71f 233 const unsigned char *cs;
94753db5
LT
234 /*
235 * Be careful about RCU walk racing with rename:
236 * use ACCESS_ONCE to fetch the name pointer.
237 *
238 * NOTE! Even if a rename will mean that the length
239 * was not loaded atomically, we don't care. The
240 * RCU walk will check the sequence count eventually,
241 * and catch it. And we won't overrun the buffer,
242 * because we're reading the name pointer atomically,
243 * and a dentry name is guaranteed to be properly
244 * terminated with a NUL byte.
245 *
246 * End result: even if 'len' is wrong, we'll exit
247 * early because the data cannot match (there can
248 * be no NUL in the ct/tcount data)
249 */
6326c71f
LT
250 cs = ACCESS_ONCE(dentry->d_name.name);
251 smp_read_barrier_depends();
252 return dentry_string_cmp(cs, ct, tcount);
94753db5
LT
253}
254
9c82ab9c 255static void __d_free(struct rcu_head *head)
1da177e4 256{
9c82ab9c
CH
257 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
258
b3d9b7a3 259 WARN_ON(!hlist_unhashed(&dentry->d_alias));
1da177e4
LT
260 if (dname_external(dentry))
261 kfree(dentry->d_name.name);
262 kmem_cache_free(dentry_cache, dentry);
263}
264
265/*
b5c84bf6 266 * no locks, please.
1da177e4
LT
267 */
268static void d_free(struct dentry *dentry)
269{
98474236 270 BUG_ON(dentry->d_lockref.count);
3e880fb5 271 this_cpu_dec(nr_dentry);
1da177e4
LT
272 if (dentry->d_op && dentry->d_op->d_release)
273 dentry->d_op->d_release(dentry);
312d3ca8 274
dea3667b
LT
275 /* if dentry was never visible to RCU, immediate free is OK */
276 if (!(dentry->d_flags & DCACHE_RCUACCESS))
9c82ab9c 277 __d_free(&dentry->d_u.d_rcu);
b3423415 278 else
9c82ab9c 279 call_rcu(&dentry->d_u.d_rcu, __d_free);
1da177e4
LT
280}
281
31e6b01f
NP
282/**
283 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
ff5fdb61 284 * @dentry: the target dentry
31e6b01f
NP
285 * After this call, in-progress rcu-walk path lookup will fail. This
286 * should be called after unhashing, and after changing d_inode (if
287 * the dentry has not already been unhashed).
288 */
289static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
290{
291 assert_spin_locked(&dentry->d_lock);
292 /* Go through a barrier */
293 write_seqcount_barrier(&dentry->d_seq);
294}
295
1da177e4
LT
296/*
297 * Release the dentry's inode, using the filesystem
31e6b01f
NP
298 * d_iput() operation if defined. Dentry has no refcount
299 * and is unhashed.
1da177e4 300 */
858119e1 301static void dentry_iput(struct dentry * dentry)
31f3e0b3 302 __releases(dentry->d_lock)
873feea0 303 __releases(dentry->d_inode->i_lock)
1da177e4
LT
304{
305 struct inode *inode = dentry->d_inode;
306 if (inode) {
307 dentry->d_inode = NULL;
b3d9b7a3 308 hlist_del_init(&dentry->d_alias);
1da177e4 309 spin_unlock(&dentry->d_lock);
873feea0 310 spin_unlock(&inode->i_lock);
f805fbda
LT
311 if (!inode->i_nlink)
312 fsnotify_inoderemove(inode);
1da177e4
LT
313 if (dentry->d_op && dentry->d_op->d_iput)
314 dentry->d_op->d_iput(dentry, inode);
315 else
316 iput(inode);
317 } else {
318 spin_unlock(&dentry->d_lock);
1da177e4
LT
319 }
320}
321
31e6b01f
NP
322/*
323 * Release the dentry's inode, using the filesystem
324 * d_iput() operation if defined. dentry remains in-use.
325 */
326static void dentry_unlink_inode(struct dentry * dentry)
327 __releases(dentry->d_lock)
873feea0 328 __releases(dentry->d_inode->i_lock)
31e6b01f
NP
329{
330 struct inode *inode = dentry->d_inode;
331 dentry->d_inode = NULL;
b3d9b7a3 332 hlist_del_init(&dentry->d_alias);
31e6b01f
NP
333 dentry_rcuwalk_barrier(dentry);
334 spin_unlock(&dentry->d_lock);
873feea0 335 spin_unlock(&inode->i_lock);
31e6b01f
NP
336 if (!inode->i_nlink)
337 fsnotify_inoderemove(inode);
338 if (dentry->d_op && dentry->d_op->d_iput)
339 dentry->d_op->d_iput(dentry, inode);
340 else
341 iput(inode);
342}
343
da3bbdd4 344/*
f0023bc6 345 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
da3bbdd4
KM
346 */
347static void dentry_lru_add(struct dentry *dentry)
348{
a4633357 349 if (list_empty(&dentry->d_lru)) {
23044507 350 spin_lock(&dcache_lru_lock);
a4633357
CH
351 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
352 dentry->d_sb->s_nr_dentry_unused++;
86c8749e 353 dentry_stat.nr_unused++;
23044507 354 spin_unlock(&dcache_lru_lock);
a4633357 355 }
da3bbdd4
KM
356}
357
23044507
NP
358static void __dentry_lru_del(struct dentry *dentry)
359{
360 list_del_init(&dentry->d_lru);
eaf5f907 361 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
23044507
NP
362 dentry->d_sb->s_nr_dentry_unused--;
363 dentry_stat.nr_unused--;
364}
365
f0023bc6
SW
366/*
367 * Remove a dentry with references from the LRU.
368 */
da3bbdd4
KM
369static void dentry_lru_del(struct dentry *dentry)
370{
371 if (!list_empty(&dentry->d_lru)) {
23044507
NP
372 spin_lock(&dcache_lru_lock);
373 __dentry_lru_del(dentry);
374 spin_unlock(&dcache_lru_lock);
da3bbdd4
KM
375 }
376}
377
b48f03b3 378static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
da3bbdd4 379{
23044507 380 spin_lock(&dcache_lru_lock);
a4633357 381 if (list_empty(&dentry->d_lru)) {
b48f03b3 382 list_add_tail(&dentry->d_lru, list);
a4633357 383 dentry->d_sb->s_nr_dentry_unused++;
86c8749e 384 dentry_stat.nr_unused++;
a4633357 385 } else {
b48f03b3 386 list_move_tail(&dentry->d_lru, list);
da3bbdd4 387 }
23044507 388 spin_unlock(&dcache_lru_lock);
da3bbdd4
KM
389}
390
d52b9086
MS
391/**
392 * d_kill - kill dentry and return parent
393 * @dentry: dentry to kill
ff5fdb61 394 * @parent: parent dentry
d52b9086 395 *
31f3e0b3 396 * The dentry must already be unhashed and removed from the LRU.
d52b9086
MS
397 *
398 * If this is the root of the dentry tree, return NULL.
23044507 399 *
b5c84bf6
NP
400 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
401 * d_kill.
d52b9086 402 */
2fd6b7f5 403static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
31f3e0b3 404 __releases(dentry->d_lock)
2fd6b7f5 405 __releases(parent->d_lock)
873feea0 406 __releases(dentry->d_inode->i_lock)
d52b9086 407{
d52b9086 408 list_del(&dentry->d_u.d_child);
c83ce989
TM
409 /*
410 * Inform try_to_ascend() that we are no longer attached to the
411 * dentry tree
412 */
b161dfa6 413 dentry->d_flags |= DCACHE_DENTRY_KILLED;
2fd6b7f5
NP
414 if (parent)
415 spin_unlock(&parent->d_lock);
d52b9086 416 dentry_iput(dentry);
b7ab39f6
NP
417 /*
418 * dentry_iput drops the locks, at which point nobody (except
419 * transient RCU lookups) can reach this dentry.
420 */
d52b9086 421 d_free(dentry);
871c0067 422 return parent;
d52b9086
MS
423}
424
c6627c60
DH
425/*
426 * Unhash a dentry without inserting an RCU walk barrier or checking that
427 * dentry->d_lock is locked. The caller must take care of that, if
428 * appropriate.
429 */
430static void __d_shrink(struct dentry *dentry)
431{
432 if (!d_unhashed(dentry)) {
433 struct hlist_bl_head *b;
434 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
435 b = &dentry->d_sb->s_anon;
436 else
437 b = d_hash(dentry->d_parent, dentry->d_name.hash);
438
439 hlist_bl_lock(b);
440 __hlist_bl_del(&dentry->d_hash);
441 dentry->d_hash.pprev = NULL;
442 hlist_bl_unlock(b);
443 }
444}
445
789680d1
NP
446/**
447 * d_drop - drop a dentry
448 * @dentry: dentry to drop
449 *
450 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
451 * be found through a VFS lookup any more. Note that this is different from
452 * deleting the dentry - d_delete will try to mark the dentry negative if
453 * possible, giving a successful _negative_ lookup, while d_drop will
454 * just make the cache lookup fail.
455 *
456 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
457 * reason (NFS timeouts or autofs deletes).
458 *
459 * __d_drop requires dentry->d_lock.
460 */
461void __d_drop(struct dentry *dentry)
462{
dea3667b 463 if (!d_unhashed(dentry)) {
c6627c60 464 __d_shrink(dentry);
dea3667b 465 dentry_rcuwalk_barrier(dentry);
789680d1
NP
466 }
467}
468EXPORT_SYMBOL(__d_drop);
469
470void d_drop(struct dentry *dentry)
471{
789680d1
NP
472 spin_lock(&dentry->d_lock);
473 __d_drop(dentry);
474 spin_unlock(&dentry->d_lock);
789680d1
NP
475}
476EXPORT_SYMBOL(d_drop);
477
77812a1e
NP
478/*
479 * Finish off a dentry we've decided to kill.
480 * dentry->d_lock must be held, returns with it unlocked.
481 * If ref is non-zero, then decrement the refcount too.
482 * Returns dentry requiring refcount drop, or NULL if we're done.
483 */
484static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
485 __releases(dentry->d_lock)
486{
873feea0 487 struct inode *inode;
77812a1e
NP
488 struct dentry *parent;
489
873feea0
NP
490 inode = dentry->d_inode;
491 if (inode && !spin_trylock(&inode->i_lock)) {
77812a1e
NP
492relock:
493 spin_unlock(&dentry->d_lock);
494 cpu_relax();
495 return dentry; /* try again with same dentry */
496 }
497 if (IS_ROOT(dentry))
498 parent = NULL;
499 else
500 parent = dentry->d_parent;
501 if (parent && !spin_trylock(&parent->d_lock)) {
873feea0
NP
502 if (inode)
503 spin_unlock(&inode->i_lock);
77812a1e
NP
504 goto relock;
505 }
31e6b01f 506
77812a1e 507 if (ref)
98474236 508 dentry->d_lockref.count--;
f0023bc6 509 /*
f0023bc6
SW
510 * inform the fs via d_prune that this dentry is about to be
511 * unhashed and destroyed.
512 */
590fb51f 513 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
61572bb1
YZ
514 dentry->d_op->d_prune(dentry);
515
516 dentry_lru_del(dentry);
77812a1e
NP
517 /* if it was on the hash then remove it */
518 __d_drop(dentry);
519 return d_kill(dentry, parent);
520}
521
1da177e4
LT
522/*
523 * This is dput
524 *
525 * This is complicated by the fact that we do not want to put
526 * dentries that are no longer on any hash chain on the unused
527 * list: we'd much rather just get rid of them immediately.
528 *
529 * However, that implies that we have to traverse the dentry
530 * tree upwards to the parents which might _also_ now be
531 * scheduled for deletion (it may have been only waiting for
532 * its last child to go away).
533 *
534 * This tail recursion is done by hand as we don't want to depend
535 * on the compiler to always get this right (gcc generally doesn't).
536 * Real recursion would eat up our stack space.
537 */
538
539/*
540 * dput - release a dentry
541 * @dentry: dentry to release
542 *
543 * Release a dentry. This will drop the usage count and if appropriate
544 * call the dentry unlink method as well as removing it from the queues and
545 * releasing its resources. If the parent dentries were scheduled for release
546 * they too may now get deleted.
1da177e4 547 */
1da177e4
LT
548void dput(struct dentry *dentry)
549{
550 if (!dentry)
551 return;
552
553repeat:
98474236 554 if (dentry->d_lockref.count == 1)
1da177e4 555 might_sleep();
98474236 556 if (lockref_put_or_lock(&dentry->d_lockref))
1da177e4 557 return;
1da177e4 558
fb045adb 559 if (dentry->d_flags & DCACHE_OP_DELETE) {
1da177e4 560 if (dentry->d_op->d_delete(dentry))
61f3dee4 561 goto kill_it;
1da177e4 562 }
265ac902 563
1da177e4
LT
564 /* Unreachable? Get rid of it */
565 if (d_unhashed(dentry))
566 goto kill_it;
265ac902 567
39e3c955 568 dentry->d_flags |= DCACHE_REFERENCED;
a4633357 569 dentry_lru_add(dentry);
265ac902 570
98474236 571 dentry->d_lockref.count--;
61f3dee4 572 spin_unlock(&dentry->d_lock);
1da177e4
LT
573 return;
574
d52b9086 575kill_it:
77812a1e 576 dentry = dentry_kill(dentry, 1);
d52b9086
MS
577 if (dentry)
578 goto repeat;
1da177e4 579}
ec4f8605 580EXPORT_SYMBOL(dput);
1da177e4
LT
581
582/**
583 * d_invalidate - invalidate a dentry
584 * @dentry: dentry to invalidate
585 *
586 * Try to invalidate the dentry if it turns out to be
587 * possible. If there are other dentries that can be
588 * reached through this one we can't delete it and we
589 * return -EBUSY. On success we return 0.
590 *
591 * no dcache lock.
592 */
593
594int d_invalidate(struct dentry * dentry)
595{
596 /*
597 * If it's already been dropped, return OK.
598 */
da502956 599 spin_lock(&dentry->d_lock);
1da177e4 600 if (d_unhashed(dentry)) {
da502956 601 spin_unlock(&dentry->d_lock);
1da177e4
LT
602 return 0;
603 }
604 /*
605 * Check whether to do a partial shrink_dcache
606 * to get rid of unused child entries.
607 */
608 if (!list_empty(&dentry->d_subdirs)) {
da502956 609 spin_unlock(&dentry->d_lock);
1da177e4 610 shrink_dcache_parent(dentry);
da502956 611 spin_lock(&dentry->d_lock);
1da177e4
LT
612 }
613
614 /*
615 * Somebody else still using it?
616 *
617 * If it's a directory, we can't drop it
618 * for fear of somebody re-populating it
619 * with children (even though dropping it
620 * would make it unreachable from the root,
621 * we might still populate it if it was a
622 * working directory or similar).
50e69630
AV
623 * We also need to leave mountpoints alone,
624 * directory or not.
1da177e4 625 */
98474236 626 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
50e69630 627 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
1da177e4 628 spin_unlock(&dentry->d_lock);
1da177e4
LT
629 return -EBUSY;
630 }
631 }
632
633 __d_drop(dentry);
634 spin_unlock(&dentry->d_lock);
1da177e4
LT
635 return 0;
636}
ec4f8605 637EXPORT_SYMBOL(d_invalidate);
1da177e4 638
b5c84bf6 639/* This must be called with d_lock held */
dc0474be 640static inline void __dget_dlock(struct dentry *dentry)
23044507 641{
98474236 642 dentry->d_lockref.count++;
23044507
NP
643}
644
dc0474be 645static inline void __dget(struct dentry *dentry)
1da177e4 646{
98474236 647 lockref_get(&dentry->d_lockref);
1da177e4
LT
648}
649
b7ab39f6
NP
650struct dentry *dget_parent(struct dentry *dentry)
651{
df3d0bbc 652 int gotref;
b7ab39f6
NP
653 struct dentry *ret;
654
df3d0bbc
WL
655 /*
656 * Do optimistic parent lookup without any
657 * locking.
658 */
659 rcu_read_lock();
660 ret = ACCESS_ONCE(dentry->d_parent);
661 gotref = lockref_get_not_zero(&ret->d_lockref);
662 rcu_read_unlock();
663 if (likely(gotref)) {
664 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
665 return ret;
666 dput(ret);
667 }
668
b7ab39f6 669repeat:
a734eb45
NP
670 /*
671 * Don't need rcu_dereference because we re-check it was correct under
672 * the lock.
673 */
674 rcu_read_lock();
b7ab39f6 675 ret = dentry->d_parent;
a734eb45
NP
676 spin_lock(&ret->d_lock);
677 if (unlikely(ret != dentry->d_parent)) {
678 spin_unlock(&ret->d_lock);
679 rcu_read_unlock();
b7ab39f6
NP
680 goto repeat;
681 }
a734eb45 682 rcu_read_unlock();
98474236
WL
683 BUG_ON(!ret->d_lockref.count);
684 ret->d_lockref.count++;
b7ab39f6 685 spin_unlock(&ret->d_lock);
b7ab39f6
NP
686 return ret;
687}
688EXPORT_SYMBOL(dget_parent);
689
1da177e4
LT
690/**
691 * d_find_alias - grab a hashed alias of inode
692 * @inode: inode in question
32ba9c3f
LT
693 * @want_discon: flag, used by d_splice_alias, to request
694 * that only a DISCONNECTED alias be returned.
1da177e4
LT
695 *
696 * If inode has a hashed alias, or is a directory and has any alias,
697 * acquire the reference to alias and return it. Otherwise return NULL.
698 * Notice that if inode is a directory there can be only one alias and
699 * it can be unhashed only if it has no children, or if it is the root
700 * of a filesystem.
701 *
21c0d8fd 702 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
32ba9c3f
LT
703 * any other hashed alias over that one unless @want_discon is set,
704 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
1da177e4 705 */
32ba9c3f 706static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
1da177e4 707{
da502956 708 struct dentry *alias, *discon_alias;
1da177e4 709
da502956
NP
710again:
711 discon_alias = NULL;
b67bfe0d 712 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
da502956 713 spin_lock(&alias->d_lock);
1da177e4 714 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
21c0d8fd 715 if (IS_ROOT(alias) &&
da502956 716 (alias->d_flags & DCACHE_DISCONNECTED)) {
1da177e4 717 discon_alias = alias;
32ba9c3f 718 } else if (!want_discon) {
dc0474be 719 __dget_dlock(alias);
da502956
NP
720 spin_unlock(&alias->d_lock);
721 return alias;
722 }
723 }
724 spin_unlock(&alias->d_lock);
725 }
726 if (discon_alias) {
727 alias = discon_alias;
728 spin_lock(&alias->d_lock);
729 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
730 if (IS_ROOT(alias) &&
731 (alias->d_flags & DCACHE_DISCONNECTED)) {
dc0474be 732 __dget_dlock(alias);
da502956 733 spin_unlock(&alias->d_lock);
1da177e4
LT
734 return alias;
735 }
736 }
da502956
NP
737 spin_unlock(&alias->d_lock);
738 goto again;
1da177e4 739 }
da502956 740 return NULL;
1da177e4
LT
741}
742
da502956 743struct dentry *d_find_alias(struct inode *inode)
1da177e4 744{
214fda1f
DH
745 struct dentry *de = NULL;
746
b3d9b7a3 747 if (!hlist_empty(&inode->i_dentry)) {
873feea0 748 spin_lock(&inode->i_lock);
32ba9c3f 749 de = __d_find_alias(inode, 0);
873feea0 750 spin_unlock(&inode->i_lock);
214fda1f 751 }
1da177e4
LT
752 return de;
753}
ec4f8605 754EXPORT_SYMBOL(d_find_alias);
1da177e4
LT
755
756/*
757 * Try to kill dentries associated with this inode.
758 * WARNING: you must own a reference to inode.
759 */
760void d_prune_aliases(struct inode *inode)
761{
0cdca3f9 762 struct dentry *dentry;
1da177e4 763restart:
873feea0 764 spin_lock(&inode->i_lock);
b67bfe0d 765 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
1da177e4 766 spin_lock(&dentry->d_lock);
98474236 767 if (!dentry->d_lockref.count) {
590fb51f
YZ
768 /*
769 * inform the fs via d_prune that this dentry
770 * is about to be unhashed and destroyed.
771 */
772 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
773 !d_unhashed(dentry))
774 dentry->d_op->d_prune(dentry);
775
dc0474be 776 __dget_dlock(dentry);
1da177e4
LT
777 __d_drop(dentry);
778 spin_unlock(&dentry->d_lock);
873feea0 779 spin_unlock(&inode->i_lock);
1da177e4
LT
780 dput(dentry);
781 goto restart;
782 }
783 spin_unlock(&dentry->d_lock);
784 }
873feea0 785 spin_unlock(&inode->i_lock);
1da177e4 786}
ec4f8605 787EXPORT_SYMBOL(d_prune_aliases);
1da177e4
LT
788
789/*
77812a1e
NP
790 * Try to throw away a dentry - free the inode, dput the parent.
791 * Requires dentry->d_lock is held, and dentry->d_count == 0.
792 * Releases dentry->d_lock.
d702ccb3 793 *
77812a1e 794 * This may fail if locks cannot be acquired no problem, just try again.
1da177e4 795 */
77812a1e 796static void try_prune_one_dentry(struct dentry *dentry)
31f3e0b3 797 __releases(dentry->d_lock)
1da177e4 798{
77812a1e 799 struct dentry *parent;
d52b9086 800
77812a1e 801 parent = dentry_kill(dentry, 0);
d52b9086 802 /*
77812a1e
NP
803 * If dentry_kill returns NULL, we have nothing more to do.
804 * if it returns the same dentry, trylocks failed. In either
805 * case, just loop again.
806 *
807 * Otherwise, we need to prune ancestors too. This is necessary
808 * to prevent quadratic behavior of shrink_dcache_parent(), but
809 * is also expected to be beneficial in reducing dentry cache
810 * fragmentation.
d52b9086 811 */
77812a1e
NP
812 if (!parent)
813 return;
814 if (parent == dentry)
815 return;
816
817 /* Prune ancestors. */
818 dentry = parent;
d52b9086 819 while (dentry) {
98474236 820 if (lockref_put_or_lock(&dentry->d_lockref))
89e60548 821 return;
77812a1e 822 dentry = dentry_kill(dentry, 1);
d52b9086 823 }
1da177e4
LT
824}
825
3049cfe2 826static void shrink_dentry_list(struct list_head *list)
1da177e4 827{
da3bbdd4 828 struct dentry *dentry;
da3bbdd4 829
ec33679d
NP
830 rcu_read_lock();
831 for (;;) {
ec33679d
NP
832 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
833 if (&dentry->d_lru == list)
834 break; /* empty */
835 spin_lock(&dentry->d_lock);
836 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
837 spin_unlock(&dentry->d_lock);
23044507
NP
838 continue;
839 }
840
1da177e4
LT
841 /*
842 * We found an inuse dentry which was not removed from
da3bbdd4
KM
843 * the LRU because of laziness during lookup. Do not free
844 * it - just keep it off the LRU list.
1da177e4 845 */
98474236 846 if (dentry->d_lockref.count) {
ec33679d 847 dentry_lru_del(dentry);
da3bbdd4 848 spin_unlock(&dentry->d_lock);
1da177e4
LT
849 continue;
850 }
ec33679d 851
ec33679d 852 rcu_read_unlock();
77812a1e
NP
853
854 try_prune_one_dentry(dentry);
855
ec33679d 856 rcu_read_lock();
da3bbdd4 857 }
ec33679d 858 rcu_read_unlock();
3049cfe2
CH
859}
860
861/**
b48f03b3
DC
862 * prune_dcache_sb - shrink the dcache
863 * @sb: superblock
864 * @count: number of entries to try to free
865 *
866 * Attempt to shrink the superblock dcache LRU by @count entries. This is
867 * done when we need more memory an called from the superblock shrinker
868 * function.
3049cfe2 869 *
b48f03b3
DC
870 * This function may fail to free any resources if all the dentries are in
871 * use.
3049cfe2 872 */
b48f03b3 873void prune_dcache_sb(struct super_block *sb, int count)
3049cfe2 874{
3049cfe2
CH
875 struct dentry *dentry;
876 LIST_HEAD(referenced);
877 LIST_HEAD(tmp);
3049cfe2 878
23044507
NP
879relock:
880 spin_lock(&dcache_lru_lock);
3049cfe2
CH
881 while (!list_empty(&sb->s_dentry_lru)) {
882 dentry = list_entry(sb->s_dentry_lru.prev,
883 struct dentry, d_lru);
884 BUG_ON(dentry->d_sb != sb);
885
23044507
NP
886 if (!spin_trylock(&dentry->d_lock)) {
887 spin_unlock(&dcache_lru_lock);
888 cpu_relax();
889 goto relock;
890 }
891
b48f03b3 892 if (dentry->d_flags & DCACHE_REFERENCED) {
23044507
NP
893 dentry->d_flags &= ~DCACHE_REFERENCED;
894 list_move(&dentry->d_lru, &referenced);
3049cfe2 895 spin_unlock(&dentry->d_lock);
23044507
NP
896 } else {
897 list_move_tail(&dentry->d_lru, &tmp);
eaf5f907 898 dentry->d_flags |= DCACHE_SHRINK_LIST;
23044507 899 spin_unlock(&dentry->d_lock);
b0d40c92 900 if (!--count)
23044507 901 break;
3049cfe2 902 }
ec33679d 903 cond_resched_lock(&dcache_lru_lock);
3049cfe2 904 }
da3bbdd4
KM
905 if (!list_empty(&referenced))
906 list_splice(&referenced, &sb->s_dentry_lru);
23044507 907 spin_unlock(&dcache_lru_lock);
ec33679d
NP
908
909 shrink_dentry_list(&tmp);
da3bbdd4
KM
910}
911
1da177e4
LT
912/**
913 * shrink_dcache_sb - shrink dcache for a superblock
914 * @sb: superblock
915 *
3049cfe2
CH
916 * Shrink the dcache for the specified super block. This is used to free
917 * the dcache before unmounting a file system.
1da177e4 918 */
3049cfe2 919void shrink_dcache_sb(struct super_block *sb)
1da177e4 920{
3049cfe2
CH
921 LIST_HEAD(tmp);
922
23044507 923 spin_lock(&dcache_lru_lock);
3049cfe2
CH
924 while (!list_empty(&sb->s_dentry_lru)) {
925 list_splice_init(&sb->s_dentry_lru, &tmp);
ec33679d 926 spin_unlock(&dcache_lru_lock);
3049cfe2 927 shrink_dentry_list(&tmp);
ec33679d 928 spin_lock(&dcache_lru_lock);
3049cfe2 929 }
23044507 930 spin_unlock(&dcache_lru_lock);
1da177e4 931}
ec4f8605 932EXPORT_SYMBOL(shrink_dcache_sb);
1da177e4 933
c636ebdb
DH
934/*
935 * destroy a single subtree of dentries for unmount
936 * - see the comments on shrink_dcache_for_umount() for a description of the
937 * locking
938 */
939static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
940{
941 struct dentry *parent;
942
943 BUG_ON(!IS_ROOT(dentry));
944
c636ebdb
DH
945 for (;;) {
946 /* descend to the first leaf in the current subtree */
43c1c9cd 947 while (!list_empty(&dentry->d_subdirs))
c636ebdb
DH
948 dentry = list_entry(dentry->d_subdirs.next,
949 struct dentry, d_u.d_child);
c636ebdb
DH
950
951 /* consume the dentries from this leaf up through its parents
952 * until we find one with children or run out altogether */
953 do {
954 struct inode *inode;
955
f0023bc6 956 /*
61572bb1 957 * inform the fs that this dentry is about to be
f0023bc6
SW
958 * unhashed and destroyed.
959 */
590fb51f
YZ
960 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
961 !d_unhashed(dentry))
61572bb1
YZ
962 dentry->d_op->d_prune(dentry);
963
964 dentry_lru_del(dentry);
43c1c9cd
DH
965 __d_shrink(dentry);
966
98474236 967 if (dentry->d_lockref.count != 0) {
c636ebdb
DH
968 printk(KERN_ERR
969 "BUG: Dentry %p{i=%lx,n=%s}"
970 " still in use (%d)"
971 " [unmount of %s %s]\n",
972 dentry,
973 dentry->d_inode ?
974 dentry->d_inode->i_ino : 0UL,
975 dentry->d_name.name,
98474236 976 dentry->d_lockref.count,
c636ebdb
DH
977 dentry->d_sb->s_type->name,
978 dentry->d_sb->s_id);
979 BUG();
980 }
981
2fd6b7f5 982 if (IS_ROOT(dentry)) {
c636ebdb 983 parent = NULL;
2fd6b7f5
NP
984 list_del(&dentry->d_u.d_child);
985 } else {
871c0067 986 parent = dentry->d_parent;
98474236 987 parent->d_lockref.count--;
2fd6b7f5 988 list_del(&dentry->d_u.d_child);
871c0067 989 }
c636ebdb 990
c636ebdb
DH
991 inode = dentry->d_inode;
992 if (inode) {
993 dentry->d_inode = NULL;
b3d9b7a3 994 hlist_del_init(&dentry->d_alias);
c636ebdb
DH
995 if (dentry->d_op && dentry->d_op->d_iput)
996 dentry->d_op->d_iput(dentry, inode);
997 else
998 iput(inode);
999 }
1000
1001 d_free(dentry);
1002
1003 /* finished when we fall off the top of the tree,
1004 * otherwise we ascend to the parent and move to the
1005 * next sibling if there is one */
1006 if (!parent)
312d3ca8 1007 return;
c636ebdb 1008 dentry = parent;
c636ebdb
DH
1009 } while (list_empty(&dentry->d_subdirs));
1010
1011 dentry = list_entry(dentry->d_subdirs.next,
1012 struct dentry, d_u.d_child);
1013 }
1014}
1015
1016/*
1017 * destroy the dentries attached to a superblock on unmounting
b5c84bf6 1018 * - we don't need to use dentry->d_lock because:
c636ebdb
DH
1019 * - the superblock is detached from all mountings and open files, so the
1020 * dentry trees will not be rearranged by the VFS
1021 * - s_umount is write-locked, so the memory pressure shrinker will ignore
1022 * any dentries belonging to this superblock that it comes across
1023 * - the filesystem itself is no longer permitted to rearrange the dentries
1024 * in this superblock
1025 */
1026void shrink_dcache_for_umount(struct super_block *sb)
1027{
1028 struct dentry *dentry;
1029
1030 if (down_read_trylock(&sb->s_umount))
1031 BUG();
1032
1033 dentry = sb->s_root;
1034 sb->s_root = NULL;
98474236 1035 dentry->d_lockref.count--;
c636ebdb
DH
1036 shrink_dcache_for_umount_subtree(dentry);
1037
ceb5bdc2
NP
1038 while (!hlist_bl_empty(&sb->s_anon)) {
1039 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
c636ebdb
DH
1040 shrink_dcache_for_umount_subtree(dentry);
1041 }
1042}
1043
c826cb7d
LT
1044/*
1045 * This tries to ascend one level of parenthood, but
1046 * we can race with renaming, so we need to re-check
1047 * the parenthood after dropping the lock and check
1048 * that the sequence number still matches.
1049 */
1050static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1051{
1052 struct dentry *new = old->d_parent;
1053
1054 rcu_read_lock();
1055 spin_unlock(&old->d_lock);
1056 spin_lock(&new->d_lock);
1057
1058 /*
1059 * might go back up the wrong parent if we have had a rename
1060 * or deletion
1061 */
1062 if (new != old->d_parent ||
b161dfa6 1063 (old->d_flags & DCACHE_DENTRY_KILLED) ||
c826cb7d
LT
1064 (!locked && read_seqretry(&rename_lock, seq))) {
1065 spin_unlock(&new->d_lock);
1066 new = NULL;
1067 }
1068 rcu_read_unlock();
1069 return new;
1070}
1071
db14fc3a
MS
1072/**
1073 * enum d_walk_ret - action to talke during tree walk
1074 * @D_WALK_CONTINUE: contrinue walk
1075 * @D_WALK_QUIT: quit walk
1076 * @D_WALK_NORETRY: quit when retry is needed
1077 * @D_WALK_SKIP: skip this dentry and its children
1078 */
1079enum d_walk_ret {
1080 D_WALK_CONTINUE,
1081 D_WALK_QUIT,
1082 D_WALK_NORETRY,
1083 D_WALK_SKIP,
1084};
c826cb7d 1085
1da177e4 1086/**
db14fc3a
MS
1087 * d_walk - walk the dentry tree
1088 * @parent: start of walk
1089 * @data: data passed to @enter() and @finish()
1090 * @enter: callback when first entering the dentry
1091 * @finish: callback when successfully finished the walk
1da177e4 1092 *
db14fc3a 1093 * The @enter() and @finish() callbacks are called with d_lock held.
1da177e4 1094 */
db14fc3a
MS
1095static void d_walk(struct dentry *parent, void *data,
1096 enum d_walk_ret (*enter)(void *, struct dentry *),
1097 void (*finish)(void *))
1da177e4 1098{
949854d0 1099 struct dentry *this_parent;
1da177e4 1100 struct list_head *next;
949854d0 1101 unsigned seq;
58db63d0 1102 int locked = 0;
db14fc3a
MS
1103 enum d_walk_ret ret;
1104 bool retry = true;
949854d0 1105
949854d0 1106 seq = read_seqbegin(&rename_lock);
58db63d0
NP
1107again:
1108 this_parent = parent;
2fd6b7f5 1109 spin_lock(&this_parent->d_lock);
db14fc3a
MS
1110
1111 ret = enter(data, this_parent);
1112 switch (ret) {
1113 case D_WALK_CONTINUE:
1114 break;
1115 case D_WALK_QUIT:
1116 case D_WALK_SKIP:
1117 goto out_unlock;
1118 case D_WALK_NORETRY:
1119 retry = false;
1120 break;
1121 }
1da177e4
LT
1122repeat:
1123 next = this_parent->d_subdirs.next;
1124resume:
1125 while (next != &this_parent->d_subdirs) {
1126 struct list_head *tmp = next;
5160ee6f 1127 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1da177e4 1128 next = tmp->next;
2fd6b7f5
NP
1129
1130 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
db14fc3a
MS
1131
1132 ret = enter(data, dentry);
1133 switch (ret) {
1134 case D_WALK_CONTINUE:
1135 break;
1136 case D_WALK_QUIT:
2fd6b7f5 1137 spin_unlock(&dentry->d_lock);
db14fc3a
MS
1138 goto out_unlock;
1139 case D_WALK_NORETRY:
1140 retry = false;
1141 break;
1142 case D_WALK_SKIP:
1143 spin_unlock(&dentry->d_lock);
1144 continue;
2fd6b7f5 1145 }
db14fc3a 1146
1da177e4 1147 if (!list_empty(&dentry->d_subdirs)) {
2fd6b7f5
NP
1148 spin_unlock(&this_parent->d_lock);
1149 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1da177e4 1150 this_parent = dentry;
2fd6b7f5 1151 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1da177e4
LT
1152 goto repeat;
1153 }
2fd6b7f5 1154 spin_unlock(&dentry->d_lock);
1da177e4
LT
1155 }
1156 /*
1157 * All done at this level ... ascend and resume the search.
1158 */
1159 if (this_parent != parent) {
c826cb7d
LT
1160 struct dentry *child = this_parent;
1161 this_parent = try_to_ascend(this_parent, locked, seq);
1162 if (!this_parent)
949854d0 1163 goto rename_retry;
949854d0 1164 next = child->d_u.d_child.next;
1da177e4
LT
1165 goto resume;
1166 }
db14fc3a
MS
1167 if (!locked && read_seqretry(&rename_lock, seq)) {
1168 spin_unlock(&this_parent->d_lock);
949854d0 1169 goto rename_retry;
db14fc3a
MS
1170 }
1171 if (finish)
1172 finish(data);
1173
1174out_unlock:
1175 spin_unlock(&this_parent->d_lock);
58db63d0
NP
1176 if (locked)
1177 write_sequnlock(&rename_lock);
db14fc3a 1178 return;
58db63d0
NP
1179
1180rename_retry:
db14fc3a
MS
1181 if (!retry)
1182 return;
8110e16d
MS
1183 if (locked)
1184 goto again;
58db63d0
NP
1185 locked = 1;
1186 write_seqlock(&rename_lock);
1187 goto again;
1da177e4 1188}
db14fc3a
MS
1189
1190/*
1191 * Search for at least 1 mount point in the dentry's subdirs.
1192 * We descend to the next level whenever the d_subdirs
1193 * list is non-empty and continue searching.
1194 */
1195
1196/**
1197 * have_submounts - check for mounts over a dentry
1198 * @parent: dentry to check.
1199 *
1200 * Return true if the parent or its subdirectories contain
1201 * a mount point
1202 */
1203
1204static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1205{
1206 int *ret = data;
1207 if (d_mountpoint(dentry)) {
1208 *ret = 1;
1209 return D_WALK_QUIT;
1210 }
1211 return D_WALK_CONTINUE;
1212}
1213
1214int have_submounts(struct dentry *parent)
1215{
1216 int ret = 0;
1217
1218 d_walk(parent, &ret, check_mount, NULL);
1219
1220 return ret;
1221}
ec4f8605 1222EXPORT_SYMBOL(have_submounts);
1da177e4 1223
eed81007
MS
1224/*
1225 * Called by mount code to set a mountpoint and check if the mountpoint is
1226 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1227 * subtree can become unreachable).
1228 *
1229 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1230 * this reason take rename_lock and d_lock on dentry and ancestors.
1231 */
1232int d_set_mounted(struct dentry *dentry)
1233{
1234 struct dentry *p;
1235 int ret = -ENOENT;
1236 write_seqlock(&rename_lock);
1237 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1238 /* Need exclusion wrt. check_submounts_and_drop() */
1239 spin_lock(&p->d_lock);
1240 if (unlikely(d_unhashed(p))) {
1241 spin_unlock(&p->d_lock);
1242 goto out;
1243 }
1244 spin_unlock(&p->d_lock);
1245 }
1246 spin_lock(&dentry->d_lock);
1247 if (!d_unlinked(dentry)) {
1248 dentry->d_flags |= DCACHE_MOUNTED;
1249 ret = 0;
1250 }
1251 spin_unlock(&dentry->d_lock);
1252out:
1253 write_sequnlock(&rename_lock);
1254 return ret;
1255}
1256
1da177e4 1257/*
fd517909 1258 * Search the dentry child list of the specified parent,
1da177e4
LT
1259 * and move any unused dentries to the end of the unused
1260 * list for prune_dcache(). We descend to the next level
1261 * whenever the d_subdirs list is non-empty and continue
1262 * searching.
1263 *
1264 * It returns zero iff there are no unused children,
1265 * otherwise it returns the number of children moved to
1266 * the end of the unused list. This may not be the total
1267 * number of unused children, because select_parent can
1268 * drop the lock and return early due to latency
1269 * constraints.
1270 */
1da177e4 1271
db14fc3a
MS
1272struct select_data {
1273 struct dentry *start;
1274 struct list_head dispose;
1275 int found;
1276};
23044507 1277
db14fc3a
MS
1278static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1279{
1280 struct select_data *data = _data;
1281 enum d_walk_ret ret = D_WALK_CONTINUE;
1da177e4 1282
db14fc3a
MS
1283 if (data->start == dentry)
1284 goto out;
2fd6b7f5 1285
1da177e4 1286 /*
db14fc3a
MS
1287 * move only zero ref count dentries to the dispose list.
1288 *
1289 * Those which are presently on the shrink list, being processed
1290 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1291 * loop in shrink_dcache_parent() might not make any progress
1292 * and loop forever.
1da177e4 1293 */
db14fc3a
MS
1294 if (dentry->d_lockref.count) {
1295 dentry_lru_del(dentry);
1296 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1297 dentry_lru_move_list(dentry, &data->dispose);
1298 dentry->d_flags |= DCACHE_SHRINK_LIST;
1299 data->found++;
1300 ret = D_WALK_NORETRY;
1da177e4 1301 }
db14fc3a
MS
1302 /*
1303 * We can return to the caller if we have found some (this
1304 * ensures forward progress). We'll be coming back to find
1305 * the rest.
1306 */
1307 if (data->found && need_resched())
1308 ret = D_WALK_QUIT;
1da177e4 1309out:
db14fc3a 1310 return ret;
1da177e4
LT
1311}
1312
1313/**
1314 * shrink_dcache_parent - prune dcache
1315 * @parent: parent of entries to prune
1316 *
1317 * Prune the dcache to remove unused children of the parent dentry.
1318 */
db14fc3a 1319void shrink_dcache_parent(struct dentry *parent)
1da177e4 1320{
db14fc3a
MS
1321 for (;;) {
1322 struct select_data data;
1da177e4 1323
db14fc3a
MS
1324 INIT_LIST_HEAD(&data.dispose);
1325 data.start = parent;
1326 data.found = 0;
1327
1328 d_walk(parent, &data, select_collect, NULL);
1329 if (!data.found)
1330 break;
1331
1332 shrink_dentry_list(&data.dispose);
421348f1
GT
1333 cond_resched();
1334 }
1da177e4 1335}
ec4f8605 1336EXPORT_SYMBOL(shrink_dcache_parent);
1da177e4 1337
848ac114
MS
1338static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1339{
1340 struct select_data *data = _data;
1341
1342 if (d_mountpoint(dentry)) {
1343 data->found = -EBUSY;
1344 return D_WALK_QUIT;
1345 }
1346
1347 return select_collect(_data, dentry);
1348}
1349
1350static void check_and_drop(void *_data)
1351{
1352 struct select_data *data = _data;
1353
1354 if (d_mountpoint(data->start))
1355 data->found = -EBUSY;
1356 if (!data->found)
1357 __d_drop(data->start);
1358}
1359
1360/**
1361 * check_submounts_and_drop - prune dcache, check for submounts and drop
1362 *
1363 * All done as a single atomic operation relative to has_unlinked_ancestor().
1364 * Returns 0 if successfully unhashed @parent. If there were submounts then
1365 * return -EBUSY.
1366 *
1367 * @dentry: dentry to prune and drop
1368 */
1369int check_submounts_and_drop(struct dentry *dentry)
1370{
1371 int ret = 0;
1372
1373 /* Negative dentries can be dropped without further checks */
1374 if (!dentry->d_inode) {
1375 d_drop(dentry);
1376 goto out;
1377 }
1378
1379 for (;;) {
1380 struct select_data data;
1381
1382 INIT_LIST_HEAD(&data.dispose);
1383 data.start = dentry;
1384 data.found = 0;
1385
1386 d_walk(dentry, &data, check_and_collect, check_and_drop);
1387 ret = data.found;
1388
1389 if (!list_empty(&data.dispose))
1390 shrink_dentry_list(&data.dispose);
1391
1392 if (ret <= 0)
1393 break;
1394
1395 cond_resched();
1396 }
1397
1398out:
1399 return ret;
1400}
1401EXPORT_SYMBOL(check_submounts_and_drop);
1402
1da177e4 1403/**
a4464dbc
AV
1404 * __d_alloc - allocate a dcache entry
1405 * @sb: filesystem it will belong to
1da177e4
LT
1406 * @name: qstr of the name
1407 *
1408 * Allocates a dentry. It returns %NULL if there is insufficient memory
1409 * available. On a success the dentry is returned. The name passed in is
1410 * copied and the copy passed in may be reused after this call.
1411 */
1412
a4464dbc 1413struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1da177e4
LT
1414{
1415 struct dentry *dentry;
1416 char *dname;
1417
e12ba74d 1418 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1da177e4
LT
1419 if (!dentry)
1420 return NULL;
1421
6326c71f
LT
1422 /*
1423 * We guarantee that the inline name is always NUL-terminated.
1424 * This way the memcpy() done by the name switching in rename
1425 * will still always have a NUL at the end, even if we might
1426 * be overwriting an internal NUL character
1427 */
1428 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1da177e4
LT
1429 if (name->len > DNAME_INLINE_LEN-1) {
1430 dname = kmalloc(name->len + 1, GFP_KERNEL);
1431 if (!dname) {
1432 kmem_cache_free(dentry_cache, dentry);
1433 return NULL;
1434 }
1435 } else {
1436 dname = dentry->d_iname;
1437 }
1da177e4
LT
1438
1439 dentry->d_name.len = name->len;
1440 dentry->d_name.hash = name->hash;
1441 memcpy(dname, name->name, name->len);
1442 dname[name->len] = 0;
1443
6326c71f
LT
1444 /* Make sure we always see the terminating NUL character */
1445 smp_wmb();
1446 dentry->d_name.name = dname;
1447
98474236 1448 dentry->d_lockref.count = 1;
dea3667b 1449 dentry->d_flags = 0;
1da177e4 1450 spin_lock_init(&dentry->d_lock);
31e6b01f 1451 seqcount_init(&dentry->d_seq);
1da177e4 1452 dentry->d_inode = NULL;
a4464dbc
AV
1453 dentry->d_parent = dentry;
1454 dentry->d_sb = sb;
1da177e4
LT
1455 dentry->d_op = NULL;
1456 dentry->d_fsdata = NULL;
ceb5bdc2 1457 INIT_HLIST_BL_NODE(&dentry->d_hash);
1da177e4
LT
1458 INIT_LIST_HEAD(&dentry->d_lru);
1459 INIT_LIST_HEAD(&dentry->d_subdirs);
b3d9b7a3 1460 INIT_HLIST_NODE(&dentry->d_alias);
2fd6b7f5 1461 INIT_LIST_HEAD(&dentry->d_u.d_child);
a4464dbc 1462 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1da177e4 1463
3e880fb5 1464 this_cpu_inc(nr_dentry);
312d3ca8 1465
1da177e4
LT
1466 return dentry;
1467}
a4464dbc
AV
1468
1469/**
1470 * d_alloc - allocate a dcache entry
1471 * @parent: parent of entry to allocate
1472 * @name: qstr of the name
1473 *
1474 * Allocates a dentry. It returns %NULL if there is insufficient memory
1475 * available. On a success the dentry is returned. The name passed in is
1476 * copied and the copy passed in may be reused after this call.
1477 */
1478struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1479{
1480 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1481 if (!dentry)
1482 return NULL;
1483
1484 spin_lock(&parent->d_lock);
1485 /*
1486 * don't need child lock because it is not subject
1487 * to concurrency here
1488 */
1489 __dget_dlock(parent);
1490 dentry->d_parent = parent;
1491 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1492 spin_unlock(&parent->d_lock);
1493
1494 return dentry;
1495}
ec4f8605 1496EXPORT_SYMBOL(d_alloc);
1da177e4 1497
4b936885
NP
1498struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1499{
a4464dbc
AV
1500 struct dentry *dentry = __d_alloc(sb, name);
1501 if (dentry)
4b936885 1502 dentry->d_flags |= DCACHE_DISCONNECTED;
4b936885
NP
1503 return dentry;
1504}
1505EXPORT_SYMBOL(d_alloc_pseudo);
1506
1da177e4
LT
1507struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1508{
1509 struct qstr q;
1510
1511 q.name = name;
1512 q.len = strlen(name);
1513 q.hash = full_name_hash(q.name, q.len);
1514 return d_alloc(parent, &q);
1515}
ef26ca97 1516EXPORT_SYMBOL(d_alloc_name);
1da177e4 1517
fb045adb
NP
1518void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1519{
6f7f7caa
LT
1520 WARN_ON_ONCE(dentry->d_op);
1521 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
fb045adb
NP
1522 DCACHE_OP_COMPARE |
1523 DCACHE_OP_REVALIDATE |
ecf3d1f1 1524 DCACHE_OP_WEAK_REVALIDATE |
fb045adb
NP
1525 DCACHE_OP_DELETE ));
1526 dentry->d_op = op;
1527 if (!op)
1528 return;
1529 if (op->d_hash)
1530 dentry->d_flags |= DCACHE_OP_HASH;
1531 if (op->d_compare)
1532 dentry->d_flags |= DCACHE_OP_COMPARE;
1533 if (op->d_revalidate)
1534 dentry->d_flags |= DCACHE_OP_REVALIDATE;
ecf3d1f1
JL
1535 if (op->d_weak_revalidate)
1536 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
fb045adb
NP
1537 if (op->d_delete)
1538 dentry->d_flags |= DCACHE_OP_DELETE;
f0023bc6
SW
1539 if (op->d_prune)
1540 dentry->d_flags |= DCACHE_OP_PRUNE;
fb045adb
NP
1541
1542}
1543EXPORT_SYMBOL(d_set_d_op);
1544
360da900
OH
1545static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1546{
b23fb0a6 1547 spin_lock(&dentry->d_lock);
9875cf80
DH
1548 if (inode) {
1549 if (unlikely(IS_AUTOMOUNT(inode)))
1550 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
b3d9b7a3 1551 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
9875cf80 1552 }
360da900 1553 dentry->d_inode = inode;
31e6b01f 1554 dentry_rcuwalk_barrier(dentry);
b23fb0a6 1555 spin_unlock(&dentry->d_lock);
360da900
OH
1556 fsnotify_d_instantiate(dentry, inode);
1557}
1558
1da177e4
LT
1559/**
1560 * d_instantiate - fill in inode information for a dentry
1561 * @entry: dentry to complete
1562 * @inode: inode to attach to this dentry
1563 *
1564 * Fill in inode information in the entry.
1565 *
1566 * This turns negative dentries into productive full members
1567 * of society.
1568 *
1569 * NOTE! This assumes that the inode count has been incremented
1570 * (or otherwise set) by the caller to indicate that it is now
1571 * in use by the dcache.
1572 */
1573
1574void d_instantiate(struct dentry *entry, struct inode * inode)
1575{
b3d9b7a3 1576 BUG_ON(!hlist_unhashed(&entry->d_alias));
873feea0
NP
1577 if (inode)
1578 spin_lock(&inode->i_lock);
360da900 1579 __d_instantiate(entry, inode);
873feea0
NP
1580 if (inode)
1581 spin_unlock(&inode->i_lock);
1da177e4
LT
1582 security_d_instantiate(entry, inode);
1583}
ec4f8605 1584EXPORT_SYMBOL(d_instantiate);
1da177e4
LT
1585
1586/**
1587 * d_instantiate_unique - instantiate a non-aliased dentry
1588 * @entry: dentry to instantiate
1589 * @inode: inode to attach to this dentry
1590 *
1591 * Fill in inode information in the entry. On success, it returns NULL.
1592 * If an unhashed alias of "entry" already exists, then we return the
e866cfa9 1593 * aliased dentry instead and drop one reference to inode.
1da177e4
LT
1594 *
1595 * Note that in order to avoid conflicts with rename() etc, the caller
1596 * had better be holding the parent directory semaphore.
e866cfa9
OD
1597 *
1598 * This also assumes that the inode count has been incremented
1599 * (or otherwise set) by the caller to indicate that it is now
1600 * in use by the dcache.
1da177e4 1601 */
770bfad8
DH
1602static struct dentry *__d_instantiate_unique(struct dentry *entry,
1603 struct inode *inode)
1da177e4
LT
1604{
1605 struct dentry *alias;
1606 int len = entry->d_name.len;
1607 const char *name = entry->d_name.name;
1608 unsigned int hash = entry->d_name.hash;
1609
770bfad8 1610 if (!inode) {
360da900 1611 __d_instantiate(entry, NULL);
770bfad8
DH
1612 return NULL;
1613 }
1614
b67bfe0d 1615 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
9abca360
NP
1616 /*
1617 * Don't need alias->d_lock here, because aliases with
1618 * d_parent == entry->d_parent are not subject to name or
1619 * parent changes, because the parent inode i_mutex is held.
1620 */
12f8ad4b 1621 if (alias->d_name.hash != hash)
1da177e4
LT
1622 continue;
1623 if (alias->d_parent != entry->d_parent)
1624 continue;
ee983e89
LT
1625 if (alias->d_name.len != len)
1626 continue;
12f8ad4b 1627 if (dentry_cmp(alias, name, len))
1da177e4 1628 continue;
dc0474be 1629 __dget(alias);
1da177e4
LT
1630 return alias;
1631 }
770bfad8 1632
360da900 1633 __d_instantiate(entry, inode);
1da177e4
LT
1634 return NULL;
1635}
770bfad8
DH
1636
1637struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1638{
1639 struct dentry *result;
1640
b3d9b7a3 1641 BUG_ON(!hlist_unhashed(&entry->d_alias));
770bfad8 1642
873feea0
NP
1643 if (inode)
1644 spin_lock(&inode->i_lock);
770bfad8 1645 result = __d_instantiate_unique(entry, inode);
873feea0
NP
1646 if (inode)
1647 spin_unlock(&inode->i_lock);
770bfad8
DH
1648
1649 if (!result) {
1650 security_d_instantiate(entry, inode);
1651 return NULL;
1652 }
1653
1654 BUG_ON(!d_unhashed(result));
1655 iput(inode);
1656 return result;
1657}
1658
1da177e4
LT
1659EXPORT_SYMBOL(d_instantiate_unique);
1660
adc0e91a
AV
1661struct dentry *d_make_root(struct inode *root_inode)
1662{
1663 struct dentry *res = NULL;
1664
1665 if (root_inode) {
26fe5750 1666 static const struct qstr name = QSTR_INIT("/", 1);
adc0e91a
AV
1667
1668 res = __d_alloc(root_inode->i_sb, &name);
1669 if (res)
1670 d_instantiate(res, root_inode);
1671 else
1672 iput(root_inode);
1673 }
1674 return res;
1675}
1676EXPORT_SYMBOL(d_make_root);
1677
d891eedb
BF
1678static struct dentry * __d_find_any_alias(struct inode *inode)
1679{
1680 struct dentry *alias;
1681
b3d9b7a3 1682 if (hlist_empty(&inode->i_dentry))
d891eedb 1683 return NULL;
b3d9b7a3 1684 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
d891eedb
BF
1685 __dget(alias);
1686 return alias;
1687}
1688
46f72b34
SW
1689/**
1690 * d_find_any_alias - find any alias for a given inode
1691 * @inode: inode to find an alias for
1692 *
1693 * If any aliases exist for the given inode, take and return a
1694 * reference for one of them. If no aliases exist, return %NULL.
1695 */
1696struct dentry *d_find_any_alias(struct inode *inode)
d891eedb
BF
1697{
1698 struct dentry *de;
1699
1700 spin_lock(&inode->i_lock);
1701 de = __d_find_any_alias(inode);
1702 spin_unlock(&inode->i_lock);
1703 return de;
1704}
46f72b34 1705EXPORT_SYMBOL(d_find_any_alias);
d891eedb 1706
4ea3ada2
CH
1707/**
1708 * d_obtain_alias - find or allocate a dentry for a given inode
1709 * @inode: inode to allocate the dentry for
1710 *
1711 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1712 * similar open by handle operations. The returned dentry may be anonymous,
1713 * or may have a full name (if the inode was already in the cache).
1714 *
1715 * When called on a directory inode, we must ensure that the inode only ever
1716 * has one dentry. If a dentry is found, that is returned instead of
1717 * allocating a new one.
1718 *
1719 * On successful return, the reference to the inode has been transferred
44003728
CH
1720 * to the dentry. In case of an error the reference on the inode is released.
1721 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1722 * be passed in and will be the error will be propagate to the return value,
1723 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
4ea3ada2
CH
1724 */
1725struct dentry *d_obtain_alias(struct inode *inode)
1726{
b911a6bd 1727 static const struct qstr anonstring = QSTR_INIT("/", 1);
9308a612
CH
1728 struct dentry *tmp;
1729 struct dentry *res;
4ea3ada2
CH
1730
1731 if (!inode)
44003728 1732 return ERR_PTR(-ESTALE);
4ea3ada2
CH
1733 if (IS_ERR(inode))
1734 return ERR_CAST(inode);
1735
d891eedb 1736 res = d_find_any_alias(inode);
9308a612
CH
1737 if (res)
1738 goto out_iput;
1739
a4464dbc 1740 tmp = __d_alloc(inode->i_sb, &anonstring);
9308a612
CH
1741 if (!tmp) {
1742 res = ERR_PTR(-ENOMEM);
1743 goto out_iput;
4ea3ada2 1744 }
b5c84bf6 1745
873feea0 1746 spin_lock(&inode->i_lock);
d891eedb 1747 res = __d_find_any_alias(inode);
9308a612 1748 if (res) {
873feea0 1749 spin_unlock(&inode->i_lock);
9308a612
CH
1750 dput(tmp);
1751 goto out_iput;
1752 }
1753
1754 /* attach a disconnected dentry */
1755 spin_lock(&tmp->d_lock);
9308a612
CH
1756 tmp->d_inode = inode;
1757 tmp->d_flags |= DCACHE_DISCONNECTED;
b3d9b7a3 1758 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1879fd6a 1759 hlist_bl_lock(&tmp->d_sb->s_anon);
ceb5bdc2 1760 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1879fd6a 1761 hlist_bl_unlock(&tmp->d_sb->s_anon);
9308a612 1762 spin_unlock(&tmp->d_lock);
873feea0 1763 spin_unlock(&inode->i_lock);
24ff6663 1764 security_d_instantiate(tmp, inode);
9308a612 1765
9308a612
CH
1766 return tmp;
1767
1768 out_iput:
24ff6663
JB
1769 if (res && !IS_ERR(res))
1770 security_d_instantiate(res, inode);
9308a612
CH
1771 iput(inode);
1772 return res;
4ea3ada2 1773}
adc48720 1774EXPORT_SYMBOL(d_obtain_alias);
1da177e4
LT
1775
1776/**
1777 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1778 * @inode: the inode which may have a disconnected dentry
1779 * @dentry: a negative dentry which we want to point to the inode.
1780 *
1781 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1782 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1783 * and return it, else simply d_add the inode to the dentry and return NULL.
1784 *
1785 * This is needed in the lookup routine of any filesystem that is exportable
1786 * (via knfsd) so that we can build dcache paths to directories effectively.
1787 *
1788 * If a dentry was found and moved, then it is returned. Otherwise NULL
1789 * is returned. This matches the expected return value of ->lookup.
1790 *
6d4ade98
SW
1791 * Cluster filesystems may call this function with a negative, hashed dentry.
1792 * In that case, we know that the inode will be a regular file, and also this
1793 * will only occur during atomic_open. So we need to check for the dentry
1794 * being already hashed only in the final case.
1da177e4
LT
1795 */
1796struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1797{
1798 struct dentry *new = NULL;
1799
a9049376
AV
1800 if (IS_ERR(inode))
1801 return ERR_CAST(inode);
1802
21c0d8fd 1803 if (inode && S_ISDIR(inode->i_mode)) {
873feea0 1804 spin_lock(&inode->i_lock);
32ba9c3f 1805 new = __d_find_alias(inode, 1);
1da177e4 1806 if (new) {
32ba9c3f 1807 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
873feea0 1808 spin_unlock(&inode->i_lock);
1da177e4 1809 security_d_instantiate(new, inode);
1da177e4
LT
1810 d_move(new, dentry);
1811 iput(inode);
1812 } else {
873feea0 1813 /* already taking inode->i_lock, so d_add() by hand */
360da900 1814 __d_instantiate(dentry, inode);
873feea0 1815 spin_unlock(&inode->i_lock);
1da177e4
LT
1816 security_d_instantiate(dentry, inode);
1817 d_rehash(dentry);
1818 }
6d4ade98
SW
1819 } else {
1820 d_instantiate(dentry, inode);
1821 if (d_unhashed(dentry))
1822 d_rehash(dentry);
1823 }
1da177e4
LT
1824 return new;
1825}
ec4f8605 1826EXPORT_SYMBOL(d_splice_alias);
1da177e4 1827
9403540c
BN
1828/**
1829 * d_add_ci - lookup or allocate new dentry with case-exact name
1830 * @inode: the inode case-insensitive lookup has found
1831 * @dentry: the negative dentry that was passed to the parent's lookup func
1832 * @name: the case-exact name to be associated with the returned dentry
1833 *
1834 * This is to avoid filling the dcache with case-insensitive names to the
1835 * same inode, only the actual correct case is stored in the dcache for
1836 * case-insensitive filesystems.
1837 *
1838 * For a case-insensitive lookup match and if the the case-exact dentry
1839 * already exists in in the dcache, use it and return it.
1840 *
1841 * If no entry exists with the exact case name, allocate new dentry with
1842 * the exact case, and return the spliced entry.
1843 */
e45b590b 1844struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
9403540c
BN
1845 struct qstr *name)
1846{
9403540c
BN
1847 struct dentry *found;
1848 struct dentry *new;
1849
b6520c81
CH
1850 /*
1851 * First check if a dentry matching the name already exists,
1852 * if not go ahead and create it now.
1853 */
9403540c 1854 found = d_hash_and_lookup(dentry->d_parent, name);
4f522a24
AV
1855 if (unlikely(IS_ERR(found)))
1856 goto err_out;
9403540c
BN
1857 if (!found) {
1858 new = d_alloc(dentry->d_parent, name);
1859 if (!new) {
4f522a24 1860 found = ERR_PTR(-ENOMEM);
9403540c
BN
1861 goto err_out;
1862 }
b6520c81 1863
9403540c
BN
1864 found = d_splice_alias(inode, new);
1865 if (found) {
1866 dput(new);
1867 return found;
1868 }
1869 return new;
1870 }
b6520c81
CH
1871
1872 /*
1873 * If a matching dentry exists, and it's not negative use it.
1874 *
1875 * Decrement the reference count to balance the iget() done
1876 * earlier on.
1877 */
9403540c
BN
1878 if (found->d_inode) {
1879 if (unlikely(found->d_inode != inode)) {
1880 /* This can't happen because bad inodes are unhashed. */
1881 BUG_ON(!is_bad_inode(inode));
1882 BUG_ON(!is_bad_inode(found->d_inode));
1883 }
9403540c
BN
1884 iput(inode);
1885 return found;
1886 }
b6520c81 1887
9403540c 1888 /*
9403540c 1889 * Negative dentry: instantiate it unless the inode is a directory and
b6520c81 1890 * already has a dentry.
9403540c 1891 */
4513d899
AV
1892 new = d_splice_alias(inode, found);
1893 if (new) {
1894 dput(found);
1895 found = new;
9403540c 1896 }
4513d899 1897 return found;
9403540c
BN
1898
1899err_out:
1900 iput(inode);
4f522a24 1901 return found;
9403540c 1902}
ec4f8605 1903EXPORT_SYMBOL(d_add_ci);
1da177e4 1904
12f8ad4b
LT
1905/*
1906 * Do the slow-case of the dentry name compare.
1907 *
1908 * Unlike the dentry_cmp() function, we need to atomically
da53be12 1909 * load the name and length information, so that the
12f8ad4b
LT
1910 * filesystem can rely on them, and can use the 'name' and
1911 * 'len' information without worrying about walking off the
1912 * end of memory etc.
1913 *
1914 * Thus the read_seqcount_retry() and the "duplicate" info
1915 * in arguments (the low-level filesystem should not look
1916 * at the dentry inode or name contents directly, since
1917 * rename can change them while we're in RCU mode).
1918 */
1919enum slow_d_compare {
1920 D_COMP_OK,
1921 D_COMP_NOMATCH,
1922 D_COMP_SEQRETRY,
1923};
1924
1925static noinline enum slow_d_compare slow_dentry_cmp(
1926 const struct dentry *parent,
12f8ad4b
LT
1927 struct dentry *dentry,
1928 unsigned int seq,
1929 const struct qstr *name)
1930{
1931 int tlen = dentry->d_name.len;
1932 const char *tname = dentry->d_name.name;
12f8ad4b
LT
1933
1934 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1935 cpu_relax();
1936 return D_COMP_SEQRETRY;
1937 }
da53be12 1938 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
12f8ad4b
LT
1939 return D_COMP_NOMATCH;
1940 return D_COMP_OK;
1941}
1942
31e6b01f
NP
1943/**
1944 * __d_lookup_rcu - search for a dentry (racy, store-free)
1945 * @parent: parent dentry
1946 * @name: qstr of name we wish to find
1f1e6e52 1947 * @seqp: returns d_seq value at the point where the dentry was found
31e6b01f
NP
1948 * Returns: dentry, or NULL
1949 *
1950 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1951 * resolution (store-free path walking) design described in
1952 * Documentation/filesystems/path-lookup.txt.
1953 *
1954 * This is not to be used outside core vfs.
1955 *
1956 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1957 * held, and rcu_read_lock held. The returned dentry must not be stored into
1958 * without taking d_lock and checking d_seq sequence count against @seq
1959 * returned here.
1960 *
15570086 1961 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
31e6b01f
NP
1962 * function.
1963 *
1964 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1965 * the returned dentry, so long as its parent's seqlock is checked after the
1966 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1967 * is formed, giving integrity down the path walk.
12f8ad4b
LT
1968 *
1969 * NOTE! The caller *has* to check the resulting dentry against the sequence
1970 * number we've returned before using any of the resulting dentry state!
31e6b01f 1971 */
8966be90
LT
1972struct dentry *__d_lookup_rcu(const struct dentry *parent,
1973 const struct qstr *name,
da53be12 1974 unsigned *seqp)
31e6b01f 1975{
26fe5750 1976 u64 hashlen = name->hash_len;
31e6b01f 1977 const unsigned char *str = name->name;
26fe5750 1978 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
ceb5bdc2 1979 struct hlist_bl_node *node;
31e6b01f
NP
1980 struct dentry *dentry;
1981
1982 /*
1983 * Note: There is significant duplication with __d_lookup_rcu which is
1984 * required to prevent single threaded performance regressions
1985 * especially on architectures where smp_rmb (in seqcounts) are costly.
1986 * Keep the two functions in sync.
1987 */
1988
1989 /*
1990 * The hash list is protected using RCU.
1991 *
1992 * Carefully use d_seq when comparing a candidate dentry, to avoid
1993 * races with d_move().
1994 *
1995 * It is possible that concurrent renames can mess up our list
1996 * walk here and result in missing our dentry, resulting in the
1997 * false-negative result. d_lookup() protects against concurrent
1998 * renames using rename_lock seqlock.
1999 *
b0a4bb83 2000 * See Documentation/filesystems/path-lookup.txt for more details.
31e6b01f 2001 */
b07ad996 2002 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
8966be90 2003 unsigned seq;
31e6b01f 2004
31e6b01f 2005seqretry:
12f8ad4b
LT
2006 /*
2007 * The dentry sequence count protects us from concurrent
da53be12 2008 * renames, and thus protects parent and name fields.
12f8ad4b
LT
2009 *
2010 * The caller must perform a seqcount check in order
da53be12 2011 * to do anything useful with the returned dentry.
12f8ad4b
LT
2012 *
2013 * NOTE! We do a "raw" seqcount_begin here. That means that
2014 * we don't wait for the sequence count to stabilize if it
2015 * is in the middle of a sequence change. If we do the slow
2016 * dentry compare, we will do seqretries until it is stable,
2017 * and if we end up with a successful lookup, we actually
2018 * want to exit RCU lookup anyway.
2019 */
2020 seq = raw_seqcount_begin(&dentry->d_seq);
31e6b01f
NP
2021 if (dentry->d_parent != parent)
2022 continue;
2e321806
LT
2023 if (d_unhashed(dentry))
2024 continue;
12f8ad4b 2025
830c0f0e 2026 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
26fe5750
LT
2027 if (dentry->d_name.hash != hashlen_hash(hashlen))
2028 continue;
da53be12
LT
2029 *seqp = seq;
2030 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
12f8ad4b
LT
2031 case D_COMP_OK:
2032 return dentry;
2033 case D_COMP_NOMATCH:
31e6b01f 2034 continue;
12f8ad4b
LT
2035 default:
2036 goto seqretry;
2037 }
31e6b01f 2038 }
12f8ad4b 2039
26fe5750 2040 if (dentry->d_name.hash_len != hashlen)
ee983e89 2041 continue;
da53be12 2042 *seqp = seq;
26fe5750 2043 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
12f8ad4b 2044 return dentry;
31e6b01f
NP
2045 }
2046 return NULL;
2047}
2048
1da177e4
LT
2049/**
2050 * d_lookup - search for a dentry
2051 * @parent: parent dentry
2052 * @name: qstr of name we wish to find
b04f784e 2053 * Returns: dentry, or NULL
1da177e4 2054 *
b04f784e
NP
2055 * d_lookup searches the children of the parent dentry for the name in
2056 * question. If the dentry is found its reference count is incremented and the
2057 * dentry is returned. The caller must use dput to free the entry when it has
2058 * finished using it. %NULL is returned if the dentry does not exist.
1da177e4 2059 */
da2d8455 2060struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
1da177e4 2061{
31e6b01f 2062 struct dentry *dentry;
949854d0 2063 unsigned seq;
1da177e4
LT
2064
2065 do {
2066 seq = read_seqbegin(&rename_lock);
2067 dentry = __d_lookup(parent, name);
2068 if (dentry)
2069 break;
2070 } while (read_seqretry(&rename_lock, seq));
2071 return dentry;
2072}
ec4f8605 2073EXPORT_SYMBOL(d_lookup);
1da177e4 2074
31e6b01f 2075/**
b04f784e
NP
2076 * __d_lookup - search for a dentry (racy)
2077 * @parent: parent dentry
2078 * @name: qstr of name we wish to find
2079 * Returns: dentry, or NULL
2080 *
2081 * __d_lookup is like d_lookup, however it may (rarely) return a
2082 * false-negative result due to unrelated rename activity.
2083 *
2084 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2085 * however it must be used carefully, eg. with a following d_lookup in
2086 * the case of failure.
2087 *
2088 * __d_lookup callers must be commented.
2089 */
a713ca2a 2090struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
1da177e4
LT
2091{
2092 unsigned int len = name->len;
2093 unsigned int hash = name->hash;
2094 const unsigned char *str = name->name;
b07ad996 2095 struct hlist_bl_head *b = d_hash(parent, hash);
ceb5bdc2 2096 struct hlist_bl_node *node;
31e6b01f 2097 struct dentry *found = NULL;
665a7583 2098 struct dentry *dentry;
1da177e4 2099
31e6b01f
NP
2100 /*
2101 * Note: There is significant duplication with __d_lookup_rcu which is
2102 * required to prevent single threaded performance regressions
2103 * especially on architectures where smp_rmb (in seqcounts) are costly.
2104 * Keep the two functions in sync.
2105 */
2106
b04f784e
NP
2107 /*
2108 * The hash list is protected using RCU.
2109 *
2110 * Take d_lock when comparing a candidate dentry, to avoid races
2111 * with d_move().
2112 *
2113 * It is possible that concurrent renames can mess up our list
2114 * walk here and result in missing our dentry, resulting in the
2115 * false-negative result. d_lookup() protects against concurrent
2116 * renames using rename_lock seqlock.
2117 *
b0a4bb83 2118 * See Documentation/filesystems/path-lookup.txt for more details.
b04f784e 2119 */
1da177e4
LT
2120 rcu_read_lock();
2121
b07ad996 2122 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1da177e4 2123
1da177e4
LT
2124 if (dentry->d_name.hash != hash)
2125 continue;
1da177e4
LT
2126
2127 spin_lock(&dentry->d_lock);
1da177e4
LT
2128 if (dentry->d_parent != parent)
2129 goto next;
d0185c08
LT
2130 if (d_unhashed(dentry))
2131 goto next;
2132
1da177e4
LT
2133 /*
2134 * It is safe to compare names since d_move() cannot
2135 * change the qstr (protected by d_lock).
2136 */
fb045adb 2137 if (parent->d_flags & DCACHE_OP_COMPARE) {
12f8ad4b
LT
2138 int tlen = dentry->d_name.len;
2139 const char *tname = dentry->d_name.name;
da53be12 2140 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1da177e4
LT
2141 goto next;
2142 } else {
ee983e89
LT
2143 if (dentry->d_name.len != len)
2144 goto next;
12f8ad4b 2145 if (dentry_cmp(dentry, str, len))
1da177e4
LT
2146 goto next;
2147 }
2148
98474236 2149 dentry->d_lockref.count++;
d0185c08 2150 found = dentry;
1da177e4
LT
2151 spin_unlock(&dentry->d_lock);
2152 break;
2153next:
2154 spin_unlock(&dentry->d_lock);
2155 }
2156 rcu_read_unlock();
2157
2158 return found;
2159}
2160
3e7e241f
EB
2161/**
2162 * d_hash_and_lookup - hash the qstr then search for a dentry
2163 * @dir: Directory to search in
2164 * @name: qstr of name we wish to find
2165 *
4f522a24 2166 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
3e7e241f
EB
2167 */
2168struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2169{
3e7e241f
EB
2170 /*
2171 * Check for a fs-specific hash function. Note that we must
2172 * calculate the standard hash first, as the d_op->d_hash()
2173 * routine may choose to leave the hash value unchanged.
2174 */
2175 name->hash = full_name_hash(name->name, name->len);
fb045adb 2176 if (dir->d_flags & DCACHE_OP_HASH) {
da53be12 2177 int err = dir->d_op->d_hash(dir, name);
4f522a24
AV
2178 if (unlikely(err < 0))
2179 return ERR_PTR(err);
3e7e241f 2180 }
4f522a24 2181 return d_lookup(dir, name);
3e7e241f 2182}
4f522a24 2183EXPORT_SYMBOL(d_hash_and_lookup);
3e7e241f 2184
1da177e4 2185/**
786a5e15 2186 * d_validate - verify dentry provided from insecure source (deprecated)
1da177e4 2187 * @dentry: The dentry alleged to be valid child of @dparent
ff5fdb61 2188 * @dparent: The parent dentry (known to be valid)
1da177e4
LT
2189 *
2190 * An insecure source has sent us a dentry, here we verify it and dget() it.
2191 * This is used by ncpfs in its readdir implementation.
2192 * Zero is returned in the dentry is invalid.
786a5e15
NP
2193 *
2194 * This function is slow for big directories, and deprecated, do not use it.
1da177e4 2195 */
d3a23e16 2196int d_validate(struct dentry *dentry, struct dentry *dparent)
1da177e4 2197{
786a5e15 2198 struct dentry *child;
d3a23e16 2199
2fd6b7f5 2200 spin_lock(&dparent->d_lock);
786a5e15
NP
2201 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2202 if (dentry == child) {
2fd6b7f5 2203 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
dc0474be 2204 __dget_dlock(dentry);
2fd6b7f5
NP
2205 spin_unlock(&dentry->d_lock);
2206 spin_unlock(&dparent->d_lock);
1da177e4
LT
2207 return 1;
2208 }
2209 }
2fd6b7f5 2210 spin_unlock(&dparent->d_lock);
786a5e15 2211
1da177e4
LT
2212 return 0;
2213}
ec4f8605 2214EXPORT_SYMBOL(d_validate);
1da177e4
LT
2215
2216/*
2217 * When a file is deleted, we have two options:
2218 * - turn this dentry into a negative dentry
2219 * - unhash this dentry and free it.
2220 *
2221 * Usually, we want to just turn this into
2222 * a negative dentry, but if anybody else is
2223 * currently using the dentry or the inode
2224 * we can't do that and we fall back on removing
2225 * it from the hash queues and waiting for
2226 * it to be deleted later when it has no users
2227 */
2228
2229/**
2230 * d_delete - delete a dentry
2231 * @dentry: The dentry to delete
2232 *
2233 * Turn the dentry into a negative dentry if possible, otherwise
2234 * remove it from the hash queues so it can be deleted later
2235 */
2236
2237void d_delete(struct dentry * dentry)
2238{
873feea0 2239 struct inode *inode;
7a91bf7f 2240 int isdir = 0;
1da177e4
LT
2241 /*
2242 * Are we the only user?
2243 */
357f8e65 2244again:
1da177e4 2245 spin_lock(&dentry->d_lock);
873feea0
NP
2246 inode = dentry->d_inode;
2247 isdir = S_ISDIR(inode->i_mode);
98474236 2248 if (dentry->d_lockref.count == 1) {
1fe0c023 2249 if (!spin_trylock(&inode->i_lock)) {
357f8e65
NP
2250 spin_unlock(&dentry->d_lock);
2251 cpu_relax();
2252 goto again;
2253 }
13e3c5e5 2254 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
31e6b01f 2255 dentry_unlink_inode(dentry);
7a91bf7f 2256 fsnotify_nameremove(dentry, isdir);
1da177e4
LT
2257 return;
2258 }
2259
2260 if (!d_unhashed(dentry))
2261 __d_drop(dentry);
2262
2263 spin_unlock(&dentry->d_lock);
7a91bf7f
JM
2264
2265 fsnotify_nameremove(dentry, isdir);
1da177e4 2266}
ec4f8605 2267EXPORT_SYMBOL(d_delete);
1da177e4 2268
b07ad996 2269static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
1da177e4 2270{
ceb5bdc2 2271 BUG_ON(!d_unhashed(entry));
1879fd6a 2272 hlist_bl_lock(b);
dea3667b 2273 entry->d_flags |= DCACHE_RCUACCESS;
b07ad996 2274 hlist_bl_add_head_rcu(&entry->d_hash, b);
1879fd6a 2275 hlist_bl_unlock(b);
1da177e4
LT
2276}
2277
770bfad8
DH
2278static void _d_rehash(struct dentry * entry)
2279{
2280 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2281}
2282
1da177e4
LT
2283/**
2284 * d_rehash - add an entry back to the hash
2285 * @entry: dentry to add to the hash
2286 *
2287 * Adds a dentry to the hash according to its name.
2288 */
2289
2290void d_rehash(struct dentry * entry)
2291{
1da177e4 2292 spin_lock(&entry->d_lock);
770bfad8 2293 _d_rehash(entry);
1da177e4 2294 spin_unlock(&entry->d_lock);
1da177e4 2295}
ec4f8605 2296EXPORT_SYMBOL(d_rehash);
1da177e4 2297
fb2d5b86
NP
2298/**
2299 * dentry_update_name_case - update case insensitive dentry with a new name
2300 * @dentry: dentry to be updated
2301 * @name: new name
2302 *
2303 * Update a case insensitive dentry with new case of name.
2304 *
2305 * dentry must have been returned by d_lookup with name @name. Old and new
2306 * name lengths must match (ie. no d_compare which allows mismatched name
2307 * lengths).
2308 *
2309 * Parent inode i_mutex must be held over d_lookup and into this call (to
2310 * keep renames and concurrent inserts, and readdir(2) away).
2311 */
2312void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2313{
7ebfa57f 2314 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
fb2d5b86
NP
2315 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2316
fb2d5b86 2317 spin_lock(&dentry->d_lock);
31e6b01f 2318 write_seqcount_begin(&dentry->d_seq);
fb2d5b86 2319 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
31e6b01f 2320 write_seqcount_end(&dentry->d_seq);
fb2d5b86 2321 spin_unlock(&dentry->d_lock);
fb2d5b86
NP
2322}
2323EXPORT_SYMBOL(dentry_update_name_case);
2324
1da177e4
LT
2325static void switch_names(struct dentry *dentry, struct dentry *target)
2326{
2327 if (dname_external(target)) {
2328 if (dname_external(dentry)) {
2329 /*
2330 * Both external: swap the pointers
2331 */
9a8d5bb4 2332 swap(target->d_name.name, dentry->d_name.name);
1da177e4
LT
2333 } else {
2334 /*
2335 * dentry:internal, target:external. Steal target's
2336 * storage and make target internal.
2337 */
321bcf92
BF
2338 memcpy(target->d_iname, dentry->d_name.name,
2339 dentry->d_name.len + 1);
1da177e4
LT
2340 dentry->d_name.name = target->d_name.name;
2341 target->d_name.name = target->d_iname;
2342 }
2343 } else {
2344 if (dname_external(dentry)) {
2345 /*
2346 * dentry:external, target:internal. Give dentry's
2347 * storage to target and make dentry internal
2348 */
2349 memcpy(dentry->d_iname, target->d_name.name,
2350 target->d_name.len + 1);
2351 target->d_name.name = dentry->d_name.name;
2352 dentry->d_name.name = dentry->d_iname;
2353 } else {
2354 /*
2355 * Both are internal. Just copy target to dentry
2356 */
2357 memcpy(dentry->d_iname, target->d_name.name,
2358 target->d_name.len + 1);
dc711ca3
AV
2359 dentry->d_name.len = target->d_name.len;
2360 return;
1da177e4
LT
2361 }
2362 }
9a8d5bb4 2363 swap(dentry->d_name.len, target->d_name.len);
1da177e4
LT
2364}
2365
2fd6b7f5
NP
2366static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2367{
2368 /*
2369 * XXXX: do we really need to take target->d_lock?
2370 */
2371 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2372 spin_lock(&target->d_parent->d_lock);
2373 else {
2374 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2375 spin_lock(&dentry->d_parent->d_lock);
2376 spin_lock_nested(&target->d_parent->d_lock,
2377 DENTRY_D_LOCK_NESTED);
2378 } else {
2379 spin_lock(&target->d_parent->d_lock);
2380 spin_lock_nested(&dentry->d_parent->d_lock,
2381 DENTRY_D_LOCK_NESTED);
2382 }
2383 }
2384 if (target < dentry) {
2385 spin_lock_nested(&target->d_lock, 2);
2386 spin_lock_nested(&dentry->d_lock, 3);
2387 } else {
2388 spin_lock_nested(&dentry->d_lock, 2);
2389 spin_lock_nested(&target->d_lock, 3);
2390 }
2391}
2392
2393static void dentry_unlock_parents_for_move(struct dentry *dentry,
2394 struct dentry *target)
2395{
2396 if (target->d_parent != dentry->d_parent)
2397 spin_unlock(&dentry->d_parent->d_lock);
2398 if (target->d_parent != target)
2399 spin_unlock(&target->d_parent->d_lock);
2400}
2401
1da177e4 2402/*
2fd6b7f5
NP
2403 * When switching names, the actual string doesn't strictly have to
2404 * be preserved in the target - because we're dropping the target
2405 * anyway. As such, we can just do a simple memcpy() to copy over
2406 * the new name before we switch.
2407 *
2408 * Note that we have to be a lot more careful about getting the hash
2409 * switched - we have to switch the hash value properly even if it
2410 * then no longer matches the actual (corrupted) string of the target.
2411 * The hash value has to match the hash queue that the dentry is on..
1da177e4 2412 */
9eaef27b 2413/*
18367501 2414 * __d_move - move a dentry
1da177e4
LT
2415 * @dentry: entry to move
2416 * @target: new dentry
2417 *
2418 * Update the dcache to reflect the move of a file name. Negative
c46c8877
JL
2419 * dcache entries should not be moved in this way. Caller must hold
2420 * rename_lock, the i_mutex of the source and target directories,
2421 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
1da177e4 2422 */
18367501 2423static void __d_move(struct dentry * dentry, struct dentry * target)
1da177e4 2424{
1da177e4
LT
2425 if (!dentry->d_inode)
2426 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2427
2fd6b7f5
NP
2428 BUG_ON(d_ancestor(dentry, target));
2429 BUG_ON(d_ancestor(target, dentry));
2430
2fd6b7f5 2431 dentry_lock_for_move(dentry, target);
1da177e4 2432
31e6b01f
NP
2433 write_seqcount_begin(&dentry->d_seq);
2434 write_seqcount_begin(&target->d_seq);
2435
ceb5bdc2
NP
2436 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2437
2438 /*
2439 * Move the dentry to the target hash queue. Don't bother checking
2440 * for the same hash queue because of how unlikely it is.
2441 */
2442 __d_drop(dentry);
789680d1 2443 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
1da177e4
LT
2444
2445 /* Unhash the target: dput() will then get rid of it */
2446 __d_drop(target);
2447
5160ee6f
ED
2448 list_del(&dentry->d_u.d_child);
2449 list_del(&target->d_u.d_child);
1da177e4
LT
2450
2451 /* Switch the names.. */
2452 switch_names(dentry, target);
9a8d5bb4 2453 swap(dentry->d_name.hash, target->d_name.hash);
1da177e4
LT
2454
2455 /* ... and switch the parents */
2456 if (IS_ROOT(dentry)) {
2457 dentry->d_parent = target->d_parent;
2458 target->d_parent = target;
5160ee6f 2459 INIT_LIST_HEAD(&target->d_u.d_child);
1da177e4 2460 } else {
9a8d5bb4 2461 swap(dentry->d_parent, target->d_parent);
1da177e4
LT
2462
2463 /* And add them back to the (new) parent lists */
5160ee6f 2464 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
1da177e4
LT
2465 }
2466
5160ee6f 2467 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2fd6b7f5 2468
31e6b01f
NP
2469 write_seqcount_end(&target->d_seq);
2470 write_seqcount_end(&dentry->d_seq);
2471
2fd6b7f5 2472 dentry_unlock_parents_for_move(dentry, target);
1da177e4 2473 spin_unlock(&target->d_lock);
c32ccd87 2474 fsnotify_d_move(dentry);
1da177e4 2475 spin_unlock(&dentry->d_lock);
18367501
AV
2476}
2477
2478/*
2479 * d_move - move a dentry
2480 * @dentry: entry to move
2481 * @target: new dentry
2482 *
2483 * Update the dcache to reflect the move of a file name. Negative
c46c8877
JL
2484 * dcache entries should not be moved in this way. See the locking
2485 * requirements for __d_move.
18367501
AV
2486 */
2487void d_move(struct dentry *dentry, struct dentry *target)
2488{
2489 write_seqlock(&rename_lock);
2490 __d_move(dentry, target);
1da177e4 2491 write_sequnlock(&rename_lock);
9eaef27b 2492}
ec4f8605 2493EXPORT_SYMBOL(d_move);
1da177e4 2494
e2761a11
OH
2495/**
2496 * d_ancestor - search for an ancestor
2497 * @p1: ancestor dentry
2498 * @p2: child dentry
2499 *
2500 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2501 * an ancestor of p2, else NULL.
9eaef27b 2502 */
e2761a11 2503struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
9eaef27b
TM
2504{
2505 struct dentry *p;
2506
871c0067 2507 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
9eaef27b 2508 if (p->d_parent == p1)
e2761a11 2509 return p;
9eaef27b 2510 }
e2761a11 2511 return NULL;
9eaef27b
TM
2512}
2513
2514/*
2515 * This helper attempts to cope with remotely renamed directories
2516 *
2517 * It assumes that the caller is already holding
18367501 2518 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
9eaef27b
TM
2519 *
2520 * Note: If ever the locking in lock_rename() changes, then please
2521 * remember to update this too...
9eaef27b 2522 */
873feea0
NP
2523static struct dentry *__d_unalias(struct inode *inode,
2524 struct dentry *dentry, struct dentry *alias)
9eaef27b
TM
2525{
2526 struct mutex *m1 = NULL, *m2 = NULL;
ee3efa91 2527 struct dentry *ret = ERR_PTR(-EBUSY);
9eaef27b
TM
2528
2529 /* If alias and dentry share a parent, then no extra locks required */
2530 if (alias->d_parent == dentry->d_parent)
2531 goto out_unalias;
2532
9eaef27b 2533 /* See lock_rename() */
9eaef27b
TM
2534 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2535 goto out_err;
2536 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2537 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2538 goto out_err;
2539 m2 = &alias->d_parent->d_inode->i_mutex;
2540out_unalias:
ee3efa91
AV
2541 if (likely(!d_mountpoint(alias))) {
2542 __d_move(alias, dentry);
2543 ret = alias;
2544 }
9eaef27b 2545out_err:
873feea0 2546 spin_unlock(&inode->i_lock);
9eaef27b
TM
2547 if (m2)
2548 mutex_unlock(m2);
2549 if (m1)
2550 mutex_unlock(m1);
2551 return ret;
2552}
2553
770bfad8
DH
2554/*
2555 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2556 * named dentry in place of the dentry to be replaced.
2fd6b7f5 2557 * returns with anon->d_lock held!
770bfad8
DH
2558 */
2559static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2560{
740da42e 2561 struct dentry *dparent;
770bfad8 2562
2fd6b7f5 2563 dentry_lock_for_move(anon, dentry);
770bfad8 2564
31e6b01f
NP
2565 write_seqcount_begin(&dentry->d_seq);
2566 write_seqcount_begin(&anon->d_seq);
2567
770bfad8 2568 dparent = dentry->d_parent;
770bfad8 2569
2fd6b7f5
NP
2570 switch_names(dentry, anon);
2571 swap(dentry->d_name.hash, anon->d_name.hash);
2572
740da42e
AV
2573 dentry->d_parent = dentry;
2574 list_del_init(&dentry->d_u.d_child);
2575 anon->d_parent = dparent;
9ed53b12 2576 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
770bfad8 2577
31e6b01f
NP
2578 write_seqcount_end(&dentry->d_seq);
2579 write_seqcount_end(&anon->d_seq);
2580
2fd6b7f5
NP
2581 dentry_unlock_parents_for_move(anon, dentry);
2582 spin_unlock(&dentry->d_lock);
2583
2584 /* anon->d_lock still locked, returns locked */
770bfad8
DH
2585 anon->d_flags &= ~DCACHE_DISCONNECTED;
2586}
2587
2588/**
2589 * d_materialise_unique - introduce an inode into the tree
2590 * @dentry: candidate dentry
2591 * @inode: inode to bind to the dentry, to which aliases may be attached
2592 *
2593 * Introduces an dentry into the tree, substituting an extant disconnected
c46c8877
JL
2594 * root directory alias in its place if there is one. Caller must hold the
2595 * i_mutex of the parent directory.
770bfad8
DH
2596 */
2597struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2598{
9eaef27b 2599 struct dentry *actual;
770bfad8
DH
2600
2601 BUG_ON(!d_unhashed(dentry));
2602
770bfad8
DH
2603 if (!inode) {
2604 actual = dentry;
360da900 2605 __d_instantiate(dentry, NULL);
357f8e65
NP
2606 d_rehash(actual);
2607 goto out_nolock;
770bfad8
DH
2608 }
2609
873feea0 2610 spin_lock(&inode->i_lock);
357f8e65 2611
9eaef27b
TM
2612 if (S_ISDIR(inode->i_mode)) {
2613 struct dentry *alias;
2614
2615 /* Does an aliased dentry already exist? */
32ba9c3f 2616 alias = __d_find_alias(inode, 0);
9eaef27b
TM
2617 if (alias) {
2618 actual = alias;
18367501
AV
2619 write_seqlock(&rename_lock);
2620
2621 if (d_ancestor(alias, dentry)) {
2622 /* Check for loops */
2623 actual = ERR_PTR(-ELOOP);
b18dafc8 2624 spin_unlock(&inode->i_lock);
18367501
AV
2625 } else if (IS_ROOT(alias)) {
2626 /* Is this an anonymous mountpoint that we
2627 * could splice into our tree? */
9eaef27b 2628 __d_materialise_dentry(dentry, alias);
18367501 2629 write_sequnlock(&rename_lock);
9eaef27b
TM
2630 __d_drop(alias);
2631 goto found;
18367501
AV
2632 } else {
2633 /* Nope, but we must(!) avoid directory
b18dafc8 2634 * aliasing. This drops inode->i_lock */
18367501 2635 actual = __d_unalias(inode, dentry, alias);
9eaef27b 2636 }
18367501 2637 write_sequnlock(&rename_lock);
dd179946
DH
2638 if (IS_ERR(actual)) {
2639 if (PTR_ERR(actual) == -ELOOP)
2640 pr_warn_ratelimited(
2641 "VFS: Lookup of '%s' in %s %s"
2642 " would have caused loop\n",
2643 dentry->d_name.name,
2644 inode->i_sb->s_type->name,
2645 inode->i_sb->s_id);
9eaef27b 2646 dput(alias);
dd179946 2647 }
9eaef27b
TM
2648 goto out_nolock;
2649 }
770bfad8
DH
2650 }
2651
2652 /* Add a unique reference */
2653 actual = __d_instantiate_unique(dentry, inode);
2654 if (!actual)
2655 actual = dentry;
357f8e65
NP
2656 else
2657 BUG_ON(!d_unhashed(actual));
770bfad8 2658
770bfad8
DH
2659 spin_lock(&actual->d_lock);
2660found:
2661 _d_rehash(actual);
2662 spin_unlock(&actual->d_lock);
873feea0 2663 spin_unlock(&inode->i_lock);
9eaef27b 2664out_nolock:
770bfad8
DH
2665 if (actual == dentry) {
2666 security_d_instantiate(dentry, inode);
2667 return NULL;
2668 }
2669
2670 iput(inode);
2671 return actual;
770bfad8 2672}
ec4f8605 2673EXPORT_SYMBOL_GPL(d_materialise_unique);
770bfad8 2674
cdd16d02 2675static int prepend(char **buffer, int *buflen, const char *str, int namelen)
6092d048
RP
2676{
2677 *buflen -= namelen;
2678 if (*buflen < 0)
2679 return -ENAMETOOLONG;
2680 *buffer -= namelen;
2681 memcpy(*buffer, str, namelen);
2682 return 0;
2683}
2684
232d2d60
WL
2685/**
2686 * prepend_name - prepend a pathname in front of current buffer pointer
2687 * buffer: buffer pointer
2688 * buflen: allocated length of the buffer
2689 * name: name string and length qstr structure
2690 *
2691 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2692 * make sure that either the old or the new name pointer and length are
2693 * fetched. However, there may be mismatch between length and pointer.
2694 * The length cannot be trusted, we need to copy it byte-by-byte until
2695 * the length is reached or a null byte is found. It also prepends "/" at
2696 * the beginning of the name. The sequence number check at the caller will
2697 * retry it again when a d_move() does happen. So any garbage in the buffer
2698 * due to mismatched pointer and length will be discarded.
2699 */
cdd16d02
MS
2700static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2701{
232d2d60
WL
2702 const char *dname = ACCESS_ONCE(name->name);
2703 u32 dlen = ACCESS_ONCE(name->len);
2704 char *p;
2705
2706 if (*buflen < dlen + 1)
2707 return -ENAMETOOLONG;
2708 *buflen -= dlen + 1;
2709 p = *buffer -= dlen + 1;
2710 *p++ = '/';
2711 while (dlen--) {
2712 char c = *dname++;
2713 if (!c)
2714 break;
2715 *p++ = c;
2716 }
2717 return 0;
cdd16d02
MS
2718}
2719
1da177e4 2720/**
208898c1 2721 * prepend_path - Prepend path string to a buffer
9d1bc601 2722 * @path: the dentry/vfsmount to report
02125a82 2723 * @root: root vfsmnt/dentry
f2eb6575
MS
2724 * @buffer: pointer to the end of the buffer
2725 * @buflen: pointer to buffer length
552ce544 2726 *
232d2d60
WL
2727 * The function tries to write out the pathname without taking any lock other
2728 * than the RCU read lock to make sure that dentries won't go away. It only
2729 * checks the sequence number of the global rename_lock as any change in the
2730 * dentry's d_seq will be preceded by changes in the rename_lock sequence
2731 * number. If the sequence number had been change, it will restart the whole
2732 * pathname back-tracing sequence again. It performs a total of 3 trials of
2733 * lockless back-tracing sequences before falling back to take the
2734 * rename_lock.
1da177e4 2735 */
02125a82
AV
2736static int prepend_path(const struct path *path,
2737 const struct path *root,
f2eb6575 2738 char **buffer, int *buflen)
1da177e4 2739{
9d1bc601
MS
2740 struct dentry *dentry = path->dentry;
2741 struct vfsmount *vfsmnt = path->mnt;
0714a533 2742 struct mount *mnt = real_mount(vfsmnt);
f2eb6575 2743 int error = 0;
232d2d60
WL
2744 unsigned seq = 0;
2745 char *bptr;
2746 int blen;
6092d048 2747
232d2d60
WL
2748restart:
2749 bptr = *buffer;
2750 blen = *buflen;
2751 read_seqbegin_or_lock(&rename_lock, &seq);
f2eb6575 2752 while (dentry != root->dentry || vfsmnt != root->mnt) {
1da177e4
LT
2753 struct dentry * parent;
2754
1da177e4 2755 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
552ce544 2756 /* Global root? */
232d2d60
WL
2757 if (mnt_has_parent(mnt)) {
2758 dentry = mnt->mnt_mountpoint;
2759 mnt = mnt->mnt_parent;
2760 vfsmnt = &mnt->mnt;
2761 continue;
2762 }
2763 /*
2764 * Filesystems needing to implement special "root names"
2765 * should do so with ->d_dname()
2766 */
2767 if (IS_ROOT(dentry) &&
2768 (dentry->d_name.len != 1 ||
2769 dentry->d_name.name[0] != '/')) {
2770 WARN(1, "Root dentry has weird name <%.*s>\n",
2771 (int) dentry->d_name.len,
2772 dentry->d_name.name);
2773 }
2774 if (!error)
2775 error = is_mounted(vfsmnt) ? 1 : 2;
2776 break;
1da177e4
LT
2777 }
2778 parent = dentry->d_parent;
2779 prefetch(parent);
232d2d60 2780 error = prepend_name(&bptr, &blen, &dentry->d_name);
f2eb6575
MS
2781 if (error)
2782 break;
2783
1da177e4
LT
2784 dentry = parent;
2785 }
232d2d60
WL
2786 if (read_seqretry_or_unlock(&rename_lock, &seq))
2787 goto restart;
1da177e4 2788
232d2d60
WL
2789 if (error >= 0 && bptr == *buffer) {
2790 if (--blen < 0)
2791 error = -ENAMETOOLONG;
2792 else
2793 *--bptr = '/';
2794 }
2795 *buffer = bptr;
2796 *buflen = blen;
7ea600b5 2797 return error;
f2eb6575 2798}
be285c71 2799
f2eb6575
MS
2800/**
2801 * __d_path - return the path of a dentry
2802 * @path: the dentry/vfsmount to report
02125a82 2803 * @root: root vfsmnt/dentry
cd956a1c 2804 * @buf: buffer to return value in
f2eb6575
MS
2805 * @buflen: buffer length
2806 *
ffd1f4ed 2807 * Convert a dentry into an ASCII path name.
f2eb6575
MS
2808 *
2809 * Returns a pointer into the buffer or an error code if the
2810 * path was too long.
2811 *
be148247 2812 * "buflen" should be positive.
f2eb6575 2813 *
02125a82 2814 * If the path is not reachable from the supplied root, return %NULL.
f2eb6575 2815 */
02125a82
AV
2816char *__d_path(const struct path *path,
2817 const struct path *root,
f2eb6575
MS
2818 char *buf, int buflen)
2819{
2820 char *res = buf + buflen;
2821 int error;
2822
2823 prepend(&res, &buflen, "\0", 1);
7ea600b5 2824 br_read_lock(&vfsmount_lock);
f2eb6575 2825 error = prepend_path(path, root, &res, &buflen);
7ea600b5 2826 br_read_unlock(&vfsmount_lock);
be148247 2827
02125a82
AV
2828 if (error < 0)
2829 return ERR_PTR(error);
2830 if (error > 0)
2831 return NULL;
2832 return res;
2833}
2834
2835char *d_absolute_path(const struct path *path,
2836 char *buf, int buflen)
2837{
2838 struct path root = {};
2839 char *res = buf + buflen;
2840 int error;
2841
2842 prepend(&res, &buflen, "\0", 1);
7ea600b5 2843 br_read_lock(&vfsmount_lock);
02125a82 2844 error = prepend_path(path, &root, &res, &buflen);
7ea600b5 2845 br_read_unlock(&vfsmount_lock);
02125a82
AV
2846
2847 if (error > 1)
2848 error = -EINVAL;
2849 if (error < 0)
f2eb6575 2850 return ERR_PTR(error);
f2eb6575 2851 return res;
1da177e4
LT
2852}
2853
ffd1f4ed
MS
2854/*
2855 * same as __d_path but appends "(deleted)" for unlinked files.
2856 */
02125a82
AV
2857static int path_with_deleted(const struct path *path,
2858 const struct path *root,
2859 char **buf, int *buflen)
ffd1f4ed
MS
2860{
2861 prepend(buf, buflen, "\0", 1);
2862 if (d_unlinked(path->dentry)) {
2863 int error = prepend(buf, buflen, " (deleted)", 10);
2864 if (error)
2865 return error;
2866 }
2867
2868 return prepend_path(path, root, buf, buflen);
2869}
2870
8df9d1a4
MS
2871static int prepend_unreachable(char **buffer, int *buflen)
2872{
2873 return prepend(buffer, buflen, "(unreachable)", 13);
2874}
2875
a03a8a70
JB
2876/**
2877 * d_path - return the path of a dentry
cf28b486 2878 * @path: path to report
a03a8a70
JB
2879 * @buf: buffer to return value in
2880 * @buflen: buffer length
2881 *
2882 * Convert a dentry into an ASCII path name. If the entry has been deleted
2883 * the string " (deleted)" is appended. Note that this is ambiguous.
2884 *
52afeefb
AV
2885 * Returns a pointer into the buffer or an error code if the path was
2886 * too long. Note: Callers should use the returned pointer, not the passed
2887 * in buffer, to use the name! The implementation often starts at an offset
2888 * into the buffer, and may leave 0 bytes at the start.
a03a8a70 2889 *
31f3e0b3 2890 * "buflen" should be positive.
a03a8a70 2891 */
20d4fdc1 2892char *d_path(const struct path *path, char *buf, int buflen)
1da177e4 2893{
ffd1f4ed 2894 char *res = buf + buflen;
6ac08c39 2895 struct path root;
ffd1f4ed 2896 int error;
1da177e4 2897
c23fbb6b
ED
2898 /*
2899 * We have various synthetic filesystems that never get mounted. On
2900 * these filesystems dentries are never used for lookup purposes, and
2901 * thus don't need to be hashed. They also don't need a name until a
2902 * user wants to identify the object in /proc/pid/fd/. The little hack
2903 * below allows us to generate a name for these objects on demand:
2904 */
cf28b486
JB
2905 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2906 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
c23fbb6b 2907
f7ad3c6b 2908 get_fs_root(current->fs, &root);
7ea600b5 2909 br_read_lock(&vfsmount_lock);
02125a82 2910 error = path_with_deleted(path, &root, &res, &buflen);
7ea600b5 2911 br_read_unlock(&vfsmount_lock);
02125a82 2912 if (error < 0)
ffd1f4ed 2913 res = ERR_PTR(error);
6ac08c39 2914 path_put(&root);
1da177e4
LT
2915 return res;
2916}
ec4f8605 2917EXPORT_SYMBOL(d_path);
1da177e4 2918
c23fbb6b
ED
2919/*
2920 * Helper function for dentry_operations.d_dname() members
2921 */
2922char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2923 const char *fmt, ...)
2924{
2925 va_list args;
2926 char temp[64];
2927 int sz;
2928
2929 va_start(args, fmt);
2930 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2931 va_end(args);
2932
2933 if (sz > sizeof(temp) || sz > buflen)
2934 return ERR_PTR(-ENAMETOOLONG);
2935
2936 buffer += buflen - sz;
2937 return memcpy(buffer, temp, sz);
2938}
2939
118b2302
AV
2940char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
2941{
2942 char *end = buffer + buflen;
2943 /* these dentries are never renamed, so d_lock is not needed */
2944 if (prepend(&end, &buflen, " (deleted)", 11) ||
232d2d60 2945 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
118b2302
AV
2946 prepend(&end, &buflen, "/", 1))
2947 end = ERR_PTR(-ENAMETOOLONG);
232d2d60 2948 return end;
118b2302
AV
2949}
2950
6092d048
RP
2951/*
2952 * Write full pathname from the root of the filesystem into the buffer.
2953 */
ec2447c2 2954static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
6092d048 2955{
232d2d60
WL
2956 char *end, *retval;
2957 int len, seq = 0;
2958 int error = 0;
6092d048 2959
232d2d60
WL
2960restart:
2961 end = buf + buflen;
2962 len = buflen;
2963 prepend(&end, &len, "\0", 1);
6092d048
RP
2964 if (buflen < 1)
2965 goto Elong;
2966 /* Get '/' right */
2967 retval = end-1;
2968 *retval = '/';
232d2d60 2969 read_seqbegin_or_lock(&rename_lock, &seq);
cdd16d02
MS
2970 while (!IS_ROOT(dentry)) {
2971 struct dentry *parent = dentry->d_parent;
9abca360 2972 int error;
6092d048 2973
6092d048 2974 prefetch(parent);
232d2d60
WL
2975 error = prepend_name(&end, &len, &dentry->d_name);
2976 if (error)
2977 break;
6092d048
RP
2978
2979 retval = end;
2980 dentry = parent;
2981 }
232d2d60
WL
2982 if (read_seqretry_or_unlock(&rename_lock, &seq))
2983 goto restart;
2984 if (error)
2985 goto Elong;
c103135c
AV
2986 return retval;
2987Elong:
2988 return ERR_PTR(-ENAMETOOLONG);
2989}
ec2447c2
NP
2990
2991char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2992{
232d2d60 2993 return __dentry_path(dentry, buf, buflen);
ec2447c2
NP
2994}
2995EXPORT_SYMBOL(dentry_path_raw);
c103135c
AV
2996
2997char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2998{
2999 char *p = NULL;
3000 char *retval;
3001
c103135c
AV
3002 if (d_unlinked(dentry)) {
3003 p = buf + buflen;
3004 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3005 goto Elong;
3006 buflen++;
3007 }
3008 retval = __dentry_path(dentry, buf, buflen);
c103135c
AV
3009 if (!IS_ERR(retval) && p)
3010 *p = '/'; /* restore '/' overriden with '\0' */
6092d048
RP
3011 return retval;
3012Elong:
6092d048
RP
3013 return ERR_PTR(-ENAMETOOLONG);
3014}
3015
1da177e4
LT
3016/*
3017 * NOTE! The user-level library version returns a
3018 * character pointer. The kernel system call just
3019 * returns the length of the buffer filled (which
3020 * includes the ending '\0' character), or a negative
3021 * error value. So libc would do something like
3022 *
3023 * char *getcwd(char * buf, size_t size)
3024 * {
3025 * int retval;
3026 *
3027 * retval = sys_getcwd(buf, size);
3028 * if (retval >= 0)
3029 * return buf;
3030 * errno = -retval;
3031 * return NULL;
3032 * }
3033 */
3cdad428 3034SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
1da177e4 3035{
552ce544 3036 int error;
6ac08c39 3037 struct path pwd, root;
552ce544 3038 char *page = (char *) __get_free_page(GFP_USER);
1da177e4
LT
3039
3040 if (!page)
3041 return -ENOMEM;
3042
f7ad3c6b 3043 get_fs_root_and_pwd(current->fs, &root, &pwd);
1da177e4 3044
552ce544 3045 error = -ENOENT;
7ea600b5 3046 br_read_lock(&vfsmount_lock);
f3da392e 3047 if (!d_unlinked(pwd.dentry)) {
552ce544 3048 unsigned long len;
8df9d1a4
MS
3049 char *cwd = page + PAGE_SIZE;
3050 int buflen = PAGE_SIZE;
1da177e4 3051
8df9d1a4 3052 prepend(&cwd, &buflen, "\0", 1);
02125a82 3053 error = prepend_path(&pwd, &root, &cwd, &buflen);
7ea600b5 3054 br_read_unlock(&vfsmount_lock);
552ce544 3055
02125a82 3056 if (error < 0)
552ce544
LT
3057 goto out;
3058
8df9d1a4 3059 /* Unreachable from current root */
02125a82 3060 if (error > 0) {
8df9d1a4
MS
3061 error = prepend_unreachable(&cwd, &buflen);
3062 if (error)
3063 goto out;
3064 }
3065
552ce544
LT
3066 error = -ERANGE;
3067 len = PAGE_SIZE + page - cwd;
3068 if (len <= size) {
3069 error = len;
3070 if (copy_to_user(buf, cwd, len))
3071 error = -EFAULT;
3072 }
949854d0 3073 } else {
7ea600b5 3074 br_read_unlock(&vfsmount_lock);
949854d0 3075 }
1da177e4
LT
3076
3077out:
6ac08c39
JB
3078 path_put(&pwd);
3079 path_put(&root);
1da177e4
LT
3080 free_page((unsigned long) page);
3081 return error;
3082}
3083
3084/*
3085 * Test whether new_dentry is a subdirectory of old_dentry.
3086 *
3087 * Trivially implemented using the dcache structure
3088 */
3089
3090/**
3091 * is_subdir - is new dentry a subdirectory of old_dentry
3092 * @new_dentry: new dentry
3093 * @old_dentry: old dentry
3094 *
3095 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3096 * Returns 0 otherwise.
3097 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3098 */
3099
e2761a11 3100int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
1da177e4
LT
3101{
3102 int result;
949854d0 3103 unsigned seq;
1da177e4 3104
e2761a11
OH
3105 if (new_dentry == old_dentry)
3106 return 1;
3107
e2761a11 3108 do {
1da177e4 3109 /* for restarting inner loop in case of seq retry */
1da177e4 3110 seq = read_seqbegin(&rename_lock);
949854d0
NP
3111 /*
3112 * Need rcu_readlock to protect against the d_parent trashing
3113 * due to d_move
3114 */
3115 rcu_read_lock();
e2761a11 3116 if (d_ancestor(old_dentry, new_dentry))
1da177e4 3117 result = 1;
e2761a11
OH
3118 else
3119 result = 0;
949854d0 3120 rcu_read_unlock();
1da177e4 3121 } while (read_seqretry(&rename_lock, seq));
1da177e4
LT
3122
3123 return result;
3124}
3125
db14fc3a 3126static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
1da177e4 3127{
db14fc3a
MS
3128 struct dentry *root = data;
3129 if (dentry != root) {
3130 if (d_unhashed(dentry) || !dentry->d_inode)
3131 return D_WALK_SKIP;
1da177e4 3132
01ddc4ed
MS
3133 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3134 dentry->d_flags |= DCACHE_GENOCIDE;
3135 dentry->d_lockref.count--;
3136 }
1da177e4 3137 }
db14fc3a
MS
3138 return D_WALK_CONTINUE;
3139}
58db63d0 3140
db14fc3a
MS
3141void d_genocide(struct dentry *parent)
3142{
3143 d_walk(parent, parent, d_genocide_kill, NULL);
1da177e4
LT
3144}
3145
60545d0d 3146void d_tmpfile(struct dentry *dentry, struct inode *inode)
1da177e4 3147{
60545d0d
AV
3148 inode_dec_link_count(inode);
3149 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3150 !hlist_unhashed(&dentry->d_alias) ||
3151 !d_unlinked(dentry));
3152 spin_lock(&dentry->d_parent->d_lock);
3153 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3154 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3155 (unsigned long long)inode->i_ino);
3156 spin_unlock(&dentry->d_lock);
3157 spin_unlock(&dentry->d_parent->d_lock);
3158 d_instantiate(dentry, inode);
1da177e4 3159}
60545d0d 3160EXPORT_SYMBOL(d_tmpfile);
1da177e4
LT
3161
3162static __initdata unsigned long dhash_entries;
3163static int __init set_dhash_entries(char *str)
3164{
3165 if (!str)
3166 return 0;
3167 dhash_entries = simple_strtoul(str, &str, 0);
3168 return 1;
3169}
3170__setup("dhash_entries=", set_dhash_entries);
3171
3172static void __init dcache_init_early(void)
3173{
074b8517 3174 unsigned int loop;
1da177e4
LT
3175
3176 /* If hashes are distributed across NUMA nodes, defer
3177 * hash allocation until vmalloc space is available.
3178 */
3179 if (hashdist)
3180 return;
3181
3182 dentry_hashtable =
3183 alloc_large_system_hash("Dentry cache",
b07ad996 3184 sizeof(struct hlist_bl_head),
1da177e4
LT
3185 dhash_entries,
3186 13,
3187 HASH_EARLY,
3188 &d_hash_shift,
3189 &d_hash_mask,
31fe62b9 3190 0,
1da177e4
LT
3191 0);
3192
074b8517 3193 for (loop = 0; loop < (1U << d_hash_shift); loop++)
b07ad996 3194 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
1da177e4
LT
3195}
3196
74bf17cf 3197static void __init dcache_init(void)
1da177e4 3198{
074b8517 3199 unsigned int loop;
1da177e4
LT
3200
3201 /*
3202 * A constructor could be added for stable state like the lists,
3203 * but it is probably not worth it because of the cache nature
3204 * of the dcache.
3205 */
0a31bd5f
CL
3206 dentry_cache = KMEM_CACHE(dentry,
3207 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
1da177e4
LT
3208
3209 /* Hash may have been set up in dcache_init_early */
3210 if (!hashdist)
3211 return;
3212
3213 dentry_hashtable =
3214 alloc_large_system_hash("Dentry cache",
b07ad996 3215 sizeof(struct hlist_bl_head),
1da177e4
LT
3216 dhash_entries,
3217 13,
3218 0,
3219 &d_hash_shift,
3220 &d_hash_mask,
31fe62b9 3221 0,
1da177e4
LT
3222 0);
3223
074b8517 3224 for (loop = 0; loop < (1U << d_hash_shift); loop++)
b07ad996 3225 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
1da177e4
LT
3226}
3227
3228/* SLAB cache for __getname() consumers */
e18b890b 3229struct kmem_cache *names_cachep __read_mostly;
ec4f8605 3230EXPORT_SYMBOL(names_cachep);
1da177e4 3231
1da177e4
LT
3232EXPORT_SYMBOL(d_genocide);
3233
1da177e4
LT
3234void __init vfs_caches_init_early(void)
3235{
3236 dcache_init_early();
3237 inode_init_early();
3238}
3239
3240void __init vfs_caches_init(unsigned long mempages)
3241{
3242 unsigned long reserve;
3243
3244 /* Base hash sizes on available memory, with a reserve equal to
3245 150% of current kernel size */
3246
3247 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3248 mempages -= reserve;
3249
3250 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
20c2df83 3251 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1da177e4 3252
74bf17cf
DC
3253 dcache_init();
3254 inode_init();
1da177e4 3255 files_init(mempages);
74bf17cf 3256 mnt_init();
1da177e4
LT
3257 bdev_cache_init();
3258 chrdev_init();
3259}