Merge tag 'mm-hotfixes-stable-2025-07-11-16-16' of git://git.kernel.org/pub/scm/linux...
[linux-block.git] / fs / dcache.c
... / ...
CommitLineData
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * fs/dcache.c
4 *
5 * Complete reimplementation
6 * (C) 1997 Thomas Schoebel-Theuer,
7 * with heavy changes by Linus Torvalds
8 */
9
10/*
11 * Notes on the allocation strategy:
12 *
13 * The dcache is a master of the icache - whenever a dcache entry
14 * exists, the inode will always exist. "iput()" is done either when
15 * the dcache entry is deleted or garbage collected.
16 */
17
18#include <linux/ratelimit.h>
19#include <linux/string.h>
20#include <linux/mm.h>
21#include <linux/fs.h>
22#include <linux/fscrypt.h>
23#include <linux/fsnotify.h>
24#include <linux/slab.h>
25#include <linux/init.h>
26#include <linux/hash.h>
27#include <linux/cache.h>
28#include <linux/export.h>
29#include <linux/security.h>
30#include <linux/seqlock.h>
31#include <linux/memblock.h>
32#include <linux/bit_spinlock.h>
33#include <linux/rculist_bl.h>
34#include <linux/list_lru.h>
35#include "internal.h"
36#include "mount.h"
37
38#include <asm/runtime-const.h>
39
40/*
41 * Usage:
42 * dcache->d_inode->i_lock protects:
43 * - i_dentry, d_u.d_alias, d_inode of aliases
44 * dcache_hash_bucket lock protects:
45 * - the dcache hash table
46 * s_roots bl list spinlock protects:
47 * - the s_roots list (see __d_drop)
48 * dentry->d_sb->s_dentry_lru_lock protects:
49 * - the dcache lru lists and counters
50 * d_lock protects:
51 * - d_flags
52 * - d_name
53 * - d_lru
54 * - d_count
55 * - d_unhashed()
56 * - d_parent and d_chilren
57 * - childrens' d_sib and d_parent
58 * - d_u.d_alias, d_inode
59 *
60 * Ordering:
61 * dentry->d_inode->i_lock
62 * dentry->d_lock
63 * dentry->d_sb->s_dentry_lru_lock
64 * dcache_hash_bucket lock
65 * s_roots lock
66 *
67 * If there is an ancestor relationship:
68 * dentry->d_parent->...->d_parent->d_lock
69 * ...
70 * dentry->d_parent->d_lock
71 * dentry->d_lock
72 *
73 * If no ancestor relationship:
74 * arbitrary, since it's serialized on rename_lock
75 */
76static int sysctl_vfs_cache_pressure __read_mostly = 100;
77static int sysctl_vfs_cache_pressure_denom __read_mostly = 100;
78
79unsigned long vfs_pressure_ratio(unsigned long val)
80{
81 return mult_frac(val, sysctl_vfs_cache_pressure, sysctl_vfs_cache_pressure_denom);
82}
83EXPORT_SYMBOL_GPL(vfs_pressure_ratio);
84
85__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
86
87EXPORT_SYMBOL(rename_lock);
88
89static struct kmem_cache *dentry_cache __ro_after_init;
90
91const struct qstr empty_name = QSTR_INIT("", 0);
92EXPORT_SYMBOL(empty_name);
93const struct qstr slash_name = QSTR_INIT("/", 1);
94EXPORT_SYMBOL(slash_name);
95const struct qstr dotdot_name = QSTR_INIT("..", 2);
96EXPORT_SYMBOL(dotdot_name);
97
98/*
99 * This is the single most critical data structure when it comes
100 * to the dcache: the hashtable for lookups. Somebody should try
101 * to make this good - I've just made it work.
102 *
103 * This hash-function tries to avoid losing too many bits of hash
104 * information, yet avoid using a prime hash-size or similar.
105 *
106 * Marking the variables "used" ensures that the compiler doesn't
107 * optimize them away completely on architectures with runtime
108 * constant infrastructure, this allows debuggers to see their
109 * values. But updating these values has no effect on those arches.
110 */
111
112static unsigned int d_hash_shift __ro_after_init __used;
113
114static struct hlist_bl_head *dentry_hashtable __ro_after_init __used;
115
116static inline struct hlist_bl_head *d_hash(unsigned long hashlen)
117{
118 return runtime_const_ptr(dentry_hashtable) +
119 runtime_const_shift_right_32(hashlen, d_hash_shift);
120}
121
122#define IN_LOOKUP_SHIFT 10
123static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
124
125static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
126 unsigned int hash)
127{
128 hash += (unsigned long) parent / L1_CACHE_BYTES;
129 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
130}
131
132struct dentry_stat_t {
133 long nr_dentry;
134 long nr_unused;
135 long age_limit; /* age in seconds */
136 long want_pages; /* pages requested by system */
137 long nr_negative; /* # of unused negative dentries */
138 long dummy; /* Reserved for future use */
139};
140
141static DEFINE_PER_CPU(long, nr_dentry);
142static DEFINE_PER_CPU(long, nr_dentry_unused);
143static DEFINE_PER_CPU(long, nr_dentry_negative);
144static int dentry_negative_policy;
145
146#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
147/* Statistics gathering. */
148static struct dentry_stat_t dentry_stat = {
149 .age_limit = 45,
150};
151
152/*
153 * Here we resort to our own counters instead of using generic per-cpu counters
154 * for consistency with what the vfs inode code does. We are expected to harvest
155 * better code and performance by having our own specialized counters.
156 *
157 * Please note that the loop is done over all possible CPUs, not over all online
158 * CPUs. The reason for this is that we don't want to play games with CPUs going
159 * on and off. If one of them goes off, we will just keep their counters.
160 *
161 * glommer: See cffbc8a for details, and if you ever intend to change this,
162 * please update all vfs counters to match.
163 */
164static long get_nr_dentry(void)
165{
166 int i;
167 long sum = 0;
168 for_each_possible_cpu(i)
169 sum += per_cpu(nr_dentry, i);
170 return sum < 0 ? 0 : sum;
171}
172
173static long get_nr_dentry_unused(void)
174{
175 int i;
176 long sum = 0;
177 for_each_possible_cpu(i)
178 sum += per_cpu(nr_dentry_unused, i);
179 return sum < 0 ? 0 : sum;
180}
181
182static long get_nr_dentry_negative(void)
183{
184 int i;
185 long sum = 0;
186
187 for_each_possible_cpu(i)
188 sum += per_cpu(nr_dentry_negative, i);
189 return sum < 0 ? 0 : sum;
190}
191
192static int proc_nr_dentry(const struct ctl_table *table, int write, void *buffer,
193 size_t *lenp, loff_t *ppos)
194{
195 dentry_stat.nr_dentry = get_nr_dentry();
196 dentry_stat.nr_unused = get_nr_dentry_unused();
197 dentry_stat.nr_negative = get_nr_dentry_negative();
198 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
199}
200
201static const struct ctl_table fs_dcache_sysctls[] = {
202 {
203 .procname = "dentry-state",
204 .data = &dentry_stat,
205 .maxlen = 6*sizeof(long),
206 .mode = 0444,
207 .proc_handler = proc_nr_dentry,
208 },
209 {
210 .procname = "dentry-negative",
211 .data = &dentry_negative_policy,
212 .maxlen = sizeof(dentry_negative_policy),
213 .mode = 0644,
214 .proc_handler = proc_dointvec_minmax,
215 .extra1 = SYSCTL_ZERO,
216 .extra2 = SYSCTL_ONE,
217 },
218};
219
220static const struct ctl_table vm_dcache_sysctls[] = {
221 {
222 .procname = "vfs_cache_pressure",
223 .data = &sysctl_vfs_cache_pressure,
224 .maxlen = sizeof(sysctl_vfs_cache_pressure),
225 .mode = 0644,
226 .proc_handler = proc_dointvec_minmax,
227 .extra1 = SYSCTL_ZERO,
228 },
229 {
230 .procname = "vfs_cache_pressure_denom",
231 .data = &sysctl_vfs_cache_pressure_denom,
232 .maxlen = sizeof(sysctl_vfs_cache_pressure_denom),
233 .mode = 0644,
234 .proc_handler = proc_dointvec_minmax,
235 .extra1 = SYSCTL_ONE_HUNDRED,
236 },
237};
238
239static int __init init_fs_dcache_sysctls(void)
240{
241 register_sysctl_init("vm", vm_dcache_sysctls);
242 register_sysctl_init("fs", fs_dcache_sysctls);
243 return 0;
244}
245fs_initcall(init_fs_dcache_sysctls);
246#endif
247
248/*
249 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
250 * The strings are both count bytes long, and count is non-zero.
251 */
252#ifdef CONFIG_DCACHE_WORD_ACCESS
253
254#include <asm/word-at-a-time.h>
255/*
256 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
257 * aligned allocation for this particular component. We don't
258 * strictly need the load_unaligned_zeropad() safety, but it
259 * doesn't hurt either.
260 *
261 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
262 * need the careful unaligned handling.
263 */
264static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
265{
266 unsigned long a,b,mask;
267
268 for (;;) {
269 a = read_word_at_a_time(cs);
270 b = load_unaligned_zeropad(ct);
271 if (tcount < sizeof(unsigned long))
272 break;
273 if (unlikely(a != b))
274 return 1;
275 cs += sizeof(unsigned long);
276 ct += sizeof(unsigned long);
277 tcount -= sizeof(unsigned long);
278 if (!tcount)
279 return 0;
280 }
281 mask = bytemask_from_count(tcount);
282 return unlikely(!!((a ^ b) & mask));
283}
284
285#else
286
287static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
288{
289 do {
290 if (*cs != *ct)
291 return 1;
292 cs++;
293 ct++;
294 tcount--;
295 } while (tcount);
296 return 0;
297}
298
299#endif
300
301static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
302{
303 /*
304 * Be careful about RCU walk racing with rename:
305 * use 'READ_ONCE' to fetch the name pointer.
306 *
307 * NOTE! Even if a rename will mean that the length
308 * was not loaded atomically, we don't care. The
309 * RCU walk will check the sequence count eventually,
310 * and catch it. And we won't overrun the buffer,
311 * because we're reading the name pointer atomically,
312 * and a dentry name is guaranteed to be properly
313 * terminated with a NUL byte.
314 *
315 * End result: even if 'len' is wrong, we'll exit
316 * early because the data cannot match (there can
317 * be no NUL in the ct/tcount data)
318 */
319 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
320
321 return dentry_string_cmp(cs, ct, tcount);
322}
323
324/*
325 * long names are allocated separately from dentry and never modified.
326 * Refcounted, freeing is RCU-delayed. See take_dentry_name_snapshot()
327 * for the reason why ->count and ->head can't be combined into a union.
328 * dentry_string_cmp() relies upon ->name[] being word-aligned.
329 */
330struct external_name {
331 atomic_t count;
332 struct rcu_head head;
333 unsigned char name[] __aligned(sizeof(unsigned long));
334};
335
336static inline struct external_name *external_name(struct dentry *dentry)
337{
338 return container_of(dentry->d_name.name, struct external_name, name[0]);
339}
340
341static void __d_free(struct rcu_head *head)
342{
343 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
344
345 kmem_cache_free(dentry_cache, dentry);
346}
347
348static void __d_free_external(struct rcu_head *head)
349{
350 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
351 kfree(external_name(dentry));
352 kmem_cache_free(dentry_cache, dentry);
353}
354
355static inline int dname_external(const struct dentry *dentry)
356{
357 return dentry->d_name.name != dentry->d_shortname.string;
358}
359
360void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
361{
362 unsigned seq;
363 const unsigned char *s;
364
365 rcu_read_lock();
366retry:
367 seq = read_seqcount_begin(&dentry->d_seq);
368 s = READ_ONCE(dentry->d_name.name);
369 name->name.hash_len = dentry->d_name.hash_len;
370 name->name.name = name->inline_name.string;
371 if (likely(s == dentry->d_shortname.string)) {
372 name->inline_name = dentry->d_shortname;
373 } else {
374 struct external_name *p;
375 p = container_of(s, struct external_name, name[0]);
376 // get a valid reference
377 if (unlikely(!atomic_inc_not_zero(&p->count)))
378 goto retry;
379 name->name.name = s;
380 }
381 if (read_seqcount_retry(&dentry->d_seq, seq)) {
382 release_dentry_name_snapshot(name);
383 goto retry;
384 }
385 rcu_read_unlock();
386}
387EXPORT_SYMBOL(take_dentry_name_snapshot);
388
389void release_dentry_name_snapshot(struct name_snapshot *name)
390{
391 if (unlikely(name->name.name != name->inline_name.string)) {
392 struct external_name *p;
393 p = container_of(name->name.name, struct external_name, name[0]);
394 if (unlikely(atomic_dec_and_test(&p->count)))
395 kfree_rcu(p, head);
396 }
397}
398EXPORT_SYMBOL(release_dentry_name_snapshot);
399
400static inline void __d_set_inode_and_type(struct dentry *dentry,
401 struct inode *inode,
402 unsigned type_flags)
403{
404 unsigned flags;
405
406 dentry->d_inode = inode;
407 flags = READ_ONCE(dentry->d_flags);
408 flags &= ~DCACHE_ENTRY_TYPE;
409 flags |= type_flags;
410 smp_store_release(&dentry->d_flags, flags);
411}
412
413static inline void __d_clear_type_and_inode(struct dentry *dentry)
414{
415 unsigned flags = READ_ONCE(dentry->d_flags);
416
417 flags &= ~DCACHE_ENTRY_TYPE;
418 WRITE_ONCE(dentry->d_flags, flags);
419 dentry->d_inode = NULL;
420 /*
421 * The negative counter only tracks dentries on the LRU. Don't inc if
422 * d_lru is on another list.
423 */
424 if ((flags & (DCACHE_LRU_LIST|DCACHE_SHRINK_LIST)) == DCACHE_LRU_LIST)
425 this_cpu_inc(nr_dentry_negative);
426}
427
428static void dentry_free(struct dentry *dentry)
429{
430 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
431 if (unlikely(dname_external(dentry))) {
432 struct external_name *p = external_name(dentry);
433 if (likely(atomic_dec_and_test(&p->count))) {
434 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
435 return;
436 }
437 }
438 /* if dentry was never visible to RCU, immediate free is OK */
439 if (dentry->d_flags & DCACHE_NORCU)
440 __d_free(&dentry->d_u.d_rcu);
441 else
442 call_rcu(&dentry->d_u.d_rcu, __d_free);
443}
444
445/*
446 * Release the dentry's inode, using the filesystem
447 * d_iput() operation if defined.
448 */
449static void dentry_unlink_inode(struct dentry * dentry)
450 __releases(dentry->d_lock)
451 __releases(dentry->d_inode->i_lock)
452{
453 struct inode *inode = dentry->d_inode;
454
455 raw_write_seqcount_begin(&dentry->d_seq);
456 __d_clear_type_and_inode(dentry);
457 hlist_del_init(&dentry->d_u.d_alias);
458 raw_write_seqcount_end(&dentry->d_seq);
459 spin_unlock(&dentry->d_lock);
460 spin_unlock(&inode->i_lock);
461 if (!inode->i_nlink)
462 fsnotify_inoderemove(inode);
463 if (dentry->d_op && dentry->d_op->d_iput)
464 dentry->d_op->d_iput(dentry, inode);
465 else
466 iput(inode);
467}
468
469/*
470 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
471 * is in use - which includes both the "real" per-superblock
472 * LRU list _and_ the DCACHE_SHRINK_LIST use.
473 *
474 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
475 * on the shrink list (ie not on the superblock LRU list).
476 *
477 * The per-cpu "nr_dentry_unused" counters are updated with
478 * the DCACHE_LRU_LIST bit.
479 *
480 * The per-cpu "nr_dentry_negative" counters are only updated
481 * when deleted from or added to the per-superblock LRU list, not
482 * from/to the shrink list. That is to avoid an unneeded dec/inc
483 * pair when moving from LRU to shrink list in select_collect().
484 *
485 * These helper functions make sure we always follow the
486 * rules. d_lock must be held by the caller.
487 */
488#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
489static void d_lru_add(struct dentry *dentry)
490{
491 D_FLAG_VERIFY(dentry, 0);
492 dentry->d_flags |= DCACHE_LRU_LIST;
493 this_cpu_inc(nr_dentry_unused);
494 if (d_is_negative(dentry))
495 this_cpu_inc(nr_dentry_negative);
496 WARN_ON_ONCE(!list_lru_add_obj(
497 &dentry->d_sb->s_dentry_lru, &dentry->d_lru));
498}
499
500static void d_lru_del(struct dentry *dentry)
501{
502 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
503 dentry->d_flags &= ~DCACHE_LRU_LIST;
504 this_cpu_dec(nr_dentry_unused);
505 if (d_is_negative(dentry))
506 this_cpu_dec(nr_dentry_negative);
507 WARN_ON_ONCE(!list_lru_del_obj(
508 &dentry->d_sb->s_dentry_lru, &dentry->d_lru));
509}
510
511static void d_shrink_del(struct dentry *dentry)
512{
513 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
514 list_del_init(&dentry->d_lru);
515 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
516 this_cpu_dec(nr_dentry_unused);
517}
518
519static void d_shrink_add(struct dentry *dentry, struct list_head *list)
520{
521 D_FLAG_VERIFY(dentry, 0);
522 list_add(&dentry->d_lru, list);
523 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
524 this_cpu_inc(nr_dentry_unused);
525}
526
527/*
528 * These can only be called under the global LRU lock, ie during the
529 * callback for freeing the LRU list. "isolate" removes it from the
530 * LRU lists entirely, while shrink_move moves it to the indicated
531 * private list.
532 */
533static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
534{
535 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
536 dentry->d_flags &= ~DCACHE_LRU_LIST;
537 this_cpu_dec(nr_dentry_unused);
538 if (d_is_negative(dentry))
539 this_cpu_dec(nr_dentry_negative);
540 list_lru_isolate(lru, &dentry->d_lru);
541}
542
543static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
544 struct list_head *list)
545{
546 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
547 dentry->d_flags |= DCACHE_SHRINK_LIST;
548 if (d_is_negative(dentry))
549 this_cpu_dec(nr_dentry_negative);
550 list_lru_isolate_move(lru, &dentry->d_lru, list);
551}
552
553static void ___d_drop(struct dentry *dentry)
554{
555 struct hlist_bl_head *b;
556 /*
557 * Hashed dentries are normally on the dentry hashtable,
558 * with the exception of those newly allocated by
559 * d_obtain_root, which are always IS_ROOT:
560 */
561 if (unlikely(IS_ROOT(dentry)))
562 b = &dentry->d_sb->s_roots;
563 else
564 b = d_hash(dentry->d_name.hash);
565
566 hlist_bl_lock(b);
567 __hlist_bl_del(&dentry->d_hash);
568 hlist_bl_unlock(b);
569}
570
571void __d_drop(struct dentry *dentry)
572{
573 if (!d_unhashed(dentry)) {
574 ___d_drop(dentry);
575 dentry->d_hash.pprev = NULL;
576 write_seqcount_invalidate(&dentry->d_seq);
577 }
578}
579EXPORT_SYMBOL(__d_drop);
580
581/**
582 * d_drop - drop a dentry
583 * @dentry: dentry to drop
584 *
585 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
586 * be found through a VFS lookup any more. Note that this is different from
587 * deleting the dentry - d_delete will try to mark the dentry negative if
588 * possible, giving a successful _negative_ lookup, while d_drop will
589 * just make the cache lookup fail.
590 *
591 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
592 * reason (NFS timeouts or autofs deletes).
593 *
594 * __d_drop requires dentry->d_lock
595 *
596 * ___d_drop doesn't mark dentry as "unhashed"
597 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
598 */
599void d_drop(struct dentry *dentry)
600{
601 spin_lock(&dentry->d_lock);
602 __d_drop(dentry);
603 spin_unlock(&dentry->d_lock);
604}
605EXPORT_SYMBOL(d_drop);
606
607static inline void dentry_unlist(struct dentry *dentry)
608{
609 struct dentry *next;
610 /*
611 * Inform d_walk() and shrink_dentry_list() that we are no longer
612 * attached to the dentry tree
613 */
614 dentry->d_flags |= DCACHE_DENTRY_KILLED;
615 if (unlikely(hlist_unhashed(&dentry->d_sib)))
616 return;
617 __hlist_del(&dentry->d_sib);
618 /*
619 * Cursors can move around the list of children. While we'd been
620 * a normal list member, it didn't matter - ->d_sib.next would've
621 * been updated. However, from now on it won't be and for the
622 * things like d_walk() it might end up with a nasty surprise.
623 * Normally d_walk() doesn't care about cursors moving around -
624 * ->d_lock on parent prevents that and since a cursor has no children
625 * of its own, we get through it without ever unlocking the parent.
626 * There is one exception, though - if we ascend from a child that
627 * gets killed as soon as we unlock it, the next sibling is found
628 * using the value left in its ->d_sib.next. And if _that_
629 * pointed to a cursor, and cursor got moved (e.g. by lseek())
630 * before d_walk() regains parent->d_lock, we'll end up skipping
631 * everything the cursor had been moved past.
632 *
633 * Solution: make sure that the pointer left behind in ->d_sib.next
634 * points to something that won't be moving around. I.e. skip the
635 * cursors.
636 */
637 while (dentry->d_sib.next) {
638 next = hlist_entry(dentry->d_sib.next, struct dentry, d_sib);
639 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
640 break;
641 dentry->d_sib.next = next->d_sib.next;
642 }
643}
644
645static struct dentry *__dentry_kill(struct dentry *dentry)
646{
647 struct dentry *parent = NULL;
648 bool can_free = true;
649
650 /*
651 * The dentry is now unrecoverably dead to the world.
652 */
653 lockref_mark_dead(&dentry->d_lockref);
654
655 /*
656 * inform the fs via d_prune that this dentry is about to be
657 * unhashed and destroyed.
658 */
659 if (dentry->d_flags & DCACHE_OP_PRUNE)
660 dentry->d_op->d_prune(dentry);
661
662 if (dentry->d_flags & DCACHE_LRU_LIST) {
663 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
664 d_lru_del(dentry);
665 }
666 /* if it was on the hash then remove it */
667 __d_drop(dentry);
668 if (dentry->d_inode)
669 dentry_unlink_inode(dentry);
670 else
671 spin_unlock(&dentry->d_lock);
672 this_cpu_dec(nr_dentry);
673 if (dentry->d_op && dentry->d_op->d_release)
674 dentry->d_op->d_release(dentry);
675
676 cond_resched();
677 /* now that it's negative, ->d_parent is stable */
678 if (!IS_ROOT(dentry)) {
679 parent = dentry->d_parent;
680 spin_lock(&parent->d_lock);
681 }
682 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
683 dentry_unlist(dentry);
684 if (dentry->d_flags & DCACHE_SHRINK_LIST)
685 can_free = false;
686 spin_unlock(&dentry->d_lock);
687 if (likely(can_free))
688 dentry_free(dentry);
689 if (parent && --parent->d_lockref.count) {
690 spin_unlock(&parent->d_lock);
691 return NULL;
692 }
693 return parent;
694}
695
696/*
697 * Lock a dentry for feeding it to __dentry_kill().
698 * Called under rcu_read_lock() and dentry->d_lock; the former
699 * guarantees that nothing we access will be freed under us.
700 * Note that dentry is *not* protected from concurrent dentry_kill(),
701 * d_delete(), etc.
702 *
703 * Return false if dentry is busy. Otherwise, return true and have
704 * that dentry's inode locked.
705 */
706
707static bool lock_for_kill(struct dentry *dentry)
708{
709 struct inode *inode = dentry->d_inode;
710
711 if (unlikely(dentry->d_lockref.count))
712 return false;
713
714 if (!inode || likely(spin_trylock(&inode->i_lock)))
715 return true;
716
717 do {
718 spin_unlock(&dentry->d_lock);
719 spin_lock(&inode->i_lock);
720 spin_lock(&dentry->d_lock);
721 if (likely(inode == dentry->d_inode))
722 break;
723 spin_unlock(&inode->i_lock);
724 inode = dentry->d_inode;
725 } while (inode);
726 if (likely(!dentry->d_lockref.count))
727 return true;
728 if (inode)
729 spin_unlock(&inode->i_lock);
730 return false;
731}
732
733/*
734 * Decide if dentry is worth retaining. Usually this is called with dentry
735 * locked; if not locked, we are more limited and might not be able to tell
736 * without a lock. False in this case means "punt to locked path and recheck".
737 *
738 * In case we aren't locked, these predicates are not "stable". However, it is
739 * sufficient that at some point after we dropped the reference the dentry was
740 * hashed and the flags had the proper value. Other dentry users may have
741 * re-gotten a reference to the dentry and change that, but our work is done -
742 * we can leave the dentry around with a zero refcount.
743 */
744static inline bool retain_dentry(struct dentry *dentry, bool locked)
745{
746 unsigned int d_flags;
747
748 smp_rmb();
749 d_flags = READ_ONCE(dentry->d_flags);
750
751 // Unreachable? Nobody would be able to look it up, no point retaining
752 if (unlikely(d_unhashed(dentry)))
753 return false;
754
755 // Same if it's disconnected
756 if (unlikely(d_flags & DCACHE_DISCONNECTED))
757 return false;
758
759 // ->d_delete() might tell us not to bother, but that requires
760 // ->d_lock; can't decide without it
761 if (unlikely(d_flags & DCACHE_OP_DELETE)) {
762 if (!locked || dentry->d_op->d_delete(dentry))
763 return false;
764 }
765
766 // Explicitly told not to bother
767 if (unlikely(d_flags & DCACHE_DONTCACHE))
768 return false;
769
770 // At this point it looks like we ought to keep it. We also might
771 // need to do something - put it on LRU if it wasn't there already
772 // and mark it referenced if it was on LRU, but not marked yet.
773 // Unfortunately, both actions require ->d_lock, so in lockless
774 // case we'd have to punt rather than doing those.
775 if (unlikely(!(d_flags & DCACHE_LRU_LIST))) {
776 if (!locked)
777 return false;
778 d_lru_add(dentry);
779 } else if (unlikely(!(d_flags & DCACHE_REFERENCED))) {
780 if (!locked)
781 return false;
782 dentry->d_flags |= DCACHE_REFERENCED;
783 }
784 return true;
785}
786
787void d_mark_dontcache(struct inode *inode)
788{
789 struct dentry *de;
790
791 spin_lock(&inode->i_lock);
792 hlist_for_each_entry(de, &inode->i_dentry, d_u.d_alias) {
793 spin_lock(&de->d_lock);
794 de->d_flags |= DCACHE_DONTCACHE;
795 spin_unlock(&de->d_lock);
796 }
797 inode->i_state |= I_DONTCACHE;
798 spin_unlock(&inode->i_lock);
799}
800EXPORT_SYMBOL(d_mark_dontcache);
801
802/*
803 * Try to do a lockless dput(), and return whether that was successful.
804 *
805 * If unsuccessful, we return false, having already taken the dentry lock.
806 * In that case refcount is guaranteed to be zero and we have already
807 * decided that it's not worth keeping around.
808 *
809 * The caller needs to hold the RCU read lock, so that the dentry is
810 * guaranteed to stay around even if the refcount goes down to zero!
811 */
812static inline bool fast_dput(struct dentry *dentry)
813{
814 int ret;
815
816 /*
817 * try to decrement the lockref optimistically.
818 */
819 ret = lockref_put_return(&dentry->d_lockref);
820
821 /*
822 * If the lockref_put_return() failed due to the lock being held
823 * by somebody else, the fast path has failed. We will need to
824 * get the lock, and then check the count again.
825 */
826 if (unlikely(ret < 0)) {
827 spin_lock(&dentry->d_lock);
828 if (WARN_ON_ONCE(dentry->d_lockref.count <= 0)) {
829 spin_unlock(&dentry->d_lock);
830 return true;
831 }
832 dentry->d_lockref.count--;
833 goto locked;
834 }
835
836 /*
837 * If we weren't the last ref, we're done.
838 */
839 if (ret)
840 return true;
841
842 /*
843 * Can we decide that decrement of refcount is all we needed without
844 * taking the lock? There's a very common case when it's all we need -
845 * dentry looks like it ought to be retained and there's nothing else
846 * to do.
847 */
848 if (retain_dentry(dentry, false))
849 return true;
850
851 /*
852 * Either not worth retaining or we can't tell without the lock.
853 * Get the lock, then. We've already decremented the refcount to 0,
854 * but we'll need to re-check the situation after getting the lock.
855 */
856 spin_lock(&dentry->d_lock);
857
858 /*
859 * Did somebody else grab a reference to it in the meantime, and
860 * we're no longer the last user after all? Alternatively, somebody
861 * else could have killed it and marked it dead. Either way, we
862 * don't need to do anything else.
863 */
864locked:
865 if (dentry->d_lockref.count || retain_dentry(dentry, true)) {
866 spin_unlock(&dentry->d_lock);
867 return true;
868 }
869 return false;
870}
871
872
873/*
874 * This is dput
875 *
876 * This is complicated by the fact that we do not want to put
877 * dentries that are no longer on any hash chain on the unused
878 * list: we'd much rather just get rid of them immediately.
879 *
880 * However, that implies that we have to traverse the dentry
881 * tree upwards to the parents which might _also_ now be
882 * scheduled for deletion (it may have been only waiting for
883 * its last child to go away).
884 *
885 * This tail recursion is done by hand as we don't want to depend
886 * on the compiler to always get this right (gcc generally doesn't).
887 * Real recursion would eat up our stack space.
888 */
889
890/*
891 * dput - release a dentry
892 * @dentry: dentry to release
893 *
894 * Release a dentry. This will drop the usage count and if appropriate
895 * call the dentry unlink method as well as removing it from the queues and
896 * releasing its resources. If the parent dentries were scheduled for release
897 * they too may now get deleted.
898 */
899void dput(struct dentry *dentry)
900{
901 if (!dentry)
902 return;
903 might_sleep();
904 rcu_read_lock();
905 if (likely(fast_dput(dentry))) {
906 rcu_read_unlock();
907 return;
908 }
909 while (lock_for_kill(dentry)) {
910 rcu_read_unlock();
911 dentry = __dentry_kill(dentry);
912 if (!dentry)
913 return;
914 if (retain_dentry(dentry, true)) {
915 spin_unlock(&dentry->d_lock);
916 return;
917 }
918 rcu_read_lock();
919 }
920 rcu_read_unlock();
921 spin_unlock(&dentry->d_lock);
922}
923EXPORT_SYMBOL(dput);
924
925static void to_shrink_list(struct dentry *dentry, struct list_head *list)
926__must_hold(&dentry->d_lock)
927{
928 if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
929 if (dentry->d_flags & DCACHE_LRU_LIST)
930 d_lru_del(dentry);
931 d_shrink_add(dentry, list);
932 }
933}
934
935void dput_to_list(struct dentry *dentry, struct list_head *list)
936{
937 rcu_read_lock();
938 if (likely(fast_dput(dentry))) {
939 rcu_read_unlock();
940 return;
941 }
942 rcu_read_unlock();
943 to_shrink_list(dentry, list);
944 spin_unlock(&dentry->d_lock);
945}
946
947struct dentry *dget_parent(struct dentry *dentry)
948{
949 int gotref;
950 struct dentry *ret;
951 unsigned seq;
952
953 /*
954 * Do optimistic parent lookup without any
955 * locking.
956 */
957 rcu_read_lock();
958 seq = raw_seqcount_begin(&dentry->d_seq);
959 ret = READ_ONCE(dentry->d_parent);
960 gotref = lockref_get_not_zero(&ret->d_lockref);
961 rcu_read_unlock();
962 if (likely(gotref)) {
963 if (!read_seqcount_retry(&dentry->d_seq, seq))
964 return ret;
965 dput(ret);
966 }
967
968repeat:
969 /*
970 * Don't need rcu_dereference because we re-check it was correct under
971 * the lock.
972 */
973 rcu_read_lock();
974 ret = dentry->d_parent;
975 spin_lock(&ret->d_lock);
976 if (unlikely(ret != dentry->d_parent)) {
977 spin_unlock(&ret->d_lock);
978 rcu_read_unlock();
979 goto repeat;
980 }
981 rcu_read_unlock();
982 BUG_ON(!ret->d_lockref.count);
983 ret->d_lockref.count++;
984 spin_unlock(&ret->d_lock);
985 return ret;
986}
987EXPORT_SYMBOL(dget_parent);
988
989static struct dentry * __d_find_any_alias(struct inode *inode)
990{
991 struct dentry *alias;
992
993 if (hlist_empty(&inode->i_dentry))
994 return NULL;
995 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
996 lockref_get(&alias->d_lockref);
997 return alias;
998}
999
1000/**
1001 * d_find_any_alias - find any alias for a given inode
1002 * @inode: inode to find an alias for
1003 *
1004 * If any aliases exist for the given inode, take and return a
1005 * reference for one of them. If no aliases exist, return %NULL.
1006 */
1007struct dentry *d_find_any_alias(struct inode *inode)
1008{
1009 struct dentry *de;
1010
1011 spin_lock(&inode->i_lock);
1012 de = __d_find_any_alias(inode);
1013 spin_unlock(&inode->i_lock);
1014 return de;
1015}
1016EXPORT_SYMBOL(d_find_any_alias);
1017
1018static struct dentry *__d_find_alias(struct inode *inode)
1019{
1020 struct dentry *alias;
1021
1022 if (S_ISDIR(inode->i_mode))
1023 return __d_find_any_alias(inode);
1024
1025 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1026 spin_lock(&alias->d_lock);
1027 if (!d_unhashed(alias)) {
1028 dget_dlock(alias);
1029 spin_unlock(&alias->d_lock);
1030 return alias;
1031 }
1032 spin_unlock(&alias->d_lock);
1033 }
1034 return NULL;
1035}
1036
1037/**
1038 * d_find_alias - grab a hashed alias of inode
1039 * @inode: inode in question
1040 *
1041 * If inode has a hashed alias, or is a directory and has any alias,
1042 * acquire the reference to alias and return it. Otherwise return NULL.
1043 * Notice that if inode is a directory there can be only one alias and
1044 * it can be unhashed only if it has no children, or if it is the root
1045 * of a filesystem, or if the directory was renamed and d_revalidate
1046 * was the first vfs operation to notice.
1047 *
1048 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
1049 * any other hashed alias over that one.
1050 */
1051struct dentry *d_find_alias(struct inode *inode)
1052{
1053 struct dentry *de = NULL;
1054
1055 if (!hlist_empty(&inode->i_dentry)) {
1056 spin_lock(&inode->i_lock);
1057 de = __d_find_alias(inode);
1058 spin_unlock(&inode->i_lock);
1059 }
1060 return de;
1061}
1062EXPORT_SYMBOL(d_find_alias);
1063
1064/*
1065 * Caller MUST be holding rcu_read_lock() and be guaranteed
1066 * that inode won't get freed until rcu_read_unlock().
1067 */
1068struct dentry *d_find_alias_rcu(struct inode *inode)
1069{
1070 struct hlist_head *l = &inode->i_dentry;
1071 struct dentry *de = NULL;
1072
1073 spin_lock(&inode->i_lock);
1074 // ->i_dentry and ->i_rcu are colocated, but the latter won't be
1075 // used without having I_FREEING set, which means no aliases left
1076 if (likely(!(inode->i_state & I_FREEING) && !hlist_empty(l))) {
1077 if (S_ISDIR(inode->i_mode)) {
1078 de = hlist_entry(l->first, struct dentry, d_u.d_alias);
1079 } else {
1080 hlist_for_each_entry(de, l, d_u.d_alias)
1081 if (!d_unhashed(de))
1082 break;
1083 }
1084 }
1085 spin_unlock(&inode->i_lock);
1086 return de;
1087}
1088
1089/*
1090 * Try to kill dentries associated with this inode.
1091 * WARNING: you must own a reference to inode.
1092 */
1093void d_prune_aliases(struct inode *inode)
1094{
1095 LIST_HEAD(dispose);
1096 struct dentry *dentry;
1097
1098 spin_lock(&inode->i_lock);
1099 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
1100 spin_lock(&dentry->d_lock);
1101 if (!dentry->d_lockref.count)
1102 to_shrink_list(dentry, &dispose);
1103 spin_unlock(&dentry->d_lock);
1104 }
1105 spin_unlock(&inode->i_lock);
1106 shrink_dentry_list(&dispose);
1107}
1108EXPORT_SYMBOL(d_prune_aliases);
1109
1110static inline void shrink_kill(struct dentry *victim)
1111{
1112 do {
1113 rcu_read_unlock();
1114 victim = __dentry_kill(victim);
1115 rcu_read_lock();
1116 } while (victim && lock_for_kill(victim));
1117 rcu_read_unlock();
1118 if (victim)
1119 spin_unlock(&victim->d_lock);
1120}
1121
1122void shrink_dentry_list(struct list_head *list)
1123{
1124 while (!list_empty(list)) {
1125 struct dentry *dentry;
1126
1127 dentry = list_entry(list->prev, struct dentry, d_lru);
1128 spin_lock(&dentry->d_lock);
1129 rcu_read_lock();
1130 if (!lock_for_kill(dentry)) {
1131 bool can_free;
1132 rcu_read_unlock();
1133 d_shrink_del(dentry);
1134 can_free = dentry->d_flags & DCACHE_DENTRY_KILLED;
1135 spin_unlock(&dentry->d_lock);
1136 if (can_free)
1137 dentry_free(dentry);
1138 continue;
1139 }
1140 d_shrink_del(dentry);
1141 shrink_kill(dentry);
1142 }
1143}
1144
1145static enum lru_status dentry_lru_isolate(struct list_head *item,
1146 struct list_lru_one *lru, void *arg)
1147{
1148 struct list_head *freeable = arg;
1149 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1150
1151
1152 /*
1153 * we are inverting the lru lock/dentry->d_lock here,
1154 * so use a trylock. If we fail to get the lock, just skip
1155 * it
1156 */
1157 if (!spin_trylock(&dentry->d_lock))
1158 return LRU_SKIP;
1159
1160 /*
1161 * Referenced dentries are still in use. If they have active
1162 * counts, just remove them from the LRU. Otherwise give them
1163 * another pass through the LRU.
1164 */
1165 if (dentry->d_lockref.count) {
1166 d_lru_isolate(lru, dentry);
1167 spin_unlock(&dentry->d_lock);
1168 return LRU_REMOVED;
1169 }
1170
1171 if (dentry->d_flags & DCACHE_REFERENCED) {
1172 dentry->d_flags &= ~DCACHE_REFERENCED;
1173 spin_unlock(&dentry->d_lock);
1174
1175 /*
1176 * The list move itself will be made by the common LRU code. At
1177 * this point, we've dropped the dentry->d_lock but keep the
1178 * lru lock. This is safe to do, since every list movement is
1179 * protected by the lru lock even if both locks are held.
1180 *
1181 * This is guaranteed by the fact that all LRU management
1182 * functions are intermediated by the LRU API calls like
1183 * list_lru_add_obj and list_lru_del_obj. List movement in this file
1184 * only ever occur through this functions or through callbacks
1185 * like this one, that are called from the LRU API.
1186 *
1187 * The only exceptions to this are functions like
1188 * shrink_dentry_list, and code that first checks for the
1189 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1190 * operating only with stack provided lists after they are
1191 * properly isolated from the main list. It is thus, always a
1192 * local access.
1193 */
1194 return LRU_ROTATE;
1195 }
1196
1197 d_lru_shrink_move(lru, dentry, freeable);
1198 spin_unlock(&dentry->d_lock);
1199
1200 return LRU_REMOVED;
1201}
1202
1203/**
1204 * prune_dcache_sb - shrink the dcache
1205 * @sb: superblock
1206 * @sc: shrink control, passed to list_lru_shrink_walk()
1207 *
1208 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1209 * is done when we need more memory and called from the superblock shrinker
1210 * function.
1211 *
1212 * This function may fail to free any resources if all the dentries are in
1213 * use.
1214 */
1215long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1216{
1217 LIST_HEAD(dispose);
1218 long freed;
1219
1220 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1221 dentry_lru_isolate, &dispose);
1222 shrink_dentry_list(&dispose);
1223 return freed;
1224}
1225
1226static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1227 struct list_lru_one *lru, void *arg)
1228{
1229 struct list_head *freeable = arg;
1230 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1231
1232 /*
1233 * we are inverting the lru lock/dentry->d_lock here,
1234 * so use a trylock. If we fail to get the lock, just skip
1235 * it
1236 */
1237 if (!spin_trylock(&dentry->d_lock))
1238 return LRU_SKIP;
1239
1240 d_lru_shrink_move(lru, dentry, freeable);
1241 spin_unlock(&dentry->d_lock);
1242
1243 return LRU_REMOVED;
1244}
1245
1246
1247/**
1248 * shrink_dcache_sb - shrink dcache for a superblock
1249 * @sb: superblock
1250 *
1251 * Shrink the dcache for the specified super block. This is used to free
1252 * the dcache before unmounting a file system.
1253 */
1254void shrink_dcache_sb(struct super_block *sb)
1255{
1256 do {
1257 LIST_HEAD(dispose);
1258
1259 list_lru_walk(&sb->s_dentry_lru,
1260 dentry_lru_isolate_shrink, &dispose, 1024);
1261 shrink_dentry_list(&dispose);
1262 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1263}
1264EXPORT_SYMBOL(shrink_dcache_sb);
1265
1266/**
1267 * enum d_walk_ret - action to talke during tree walk
1268 * @D_WALK_CONTINUE: contrinue walk
1269 * @D_WALK_QUIT: quit walk
1270 * @D_WALK_NORETRY: quit when retry is needed
1271 * @D_WALK_SKIP: skip this dentry and its children
1272 */
1273enum d_walk_ret {
1274 D_WALK_CONTINUE,
1275 D_WALK_QUIT,
1276 D_WALK_NORETRY,
1277 D_WALK_SKIP,
1278};
1279
1280/**
1281 * d_walk - walk the dentry tree
1282 * @parent: start of walk
1283 * @data: data passed to @enter() and @finish()
1284 * @enter: callback when first entering the dentry
1285 *
1286 * The @enter() callbacks are called with d_lock held.
1287 */
1288static void d_walk(struct dentry *parent, void *data,
1289 enum d_walk_ret (*enter)(void *, struct dentry *))
1290{
1291 struct dentry *this_parent, *dentry;
1292 unsigned seq = 0;
1293 enum d_walk_ret ret;
1294 bool retry = true;
1295
1296again:
1297 read_seqbegin_or_lock(&rename_lock, &seq);
1298 this_parent = parent;
1299 spin_lock(&this_parent->d_lock);
1300
1301 ret = enter(data, this_parent);
1302 switch (ret) {
1303 case D_WALK_CONTINUE:
1304 break;
1305 case D_WALK_QUIT:
1306 case D_WALK_SKIP:
1307 goto out_unlock;
1308 case D_WALK_NORETRY:
1309 retry = false;
1310 break;
1311 }
1312repeat:
1313 dentry = d_first_child(this_parent);
1314resume:
1315 hlist_for_each_entry_from(dentry, d_sib) {
1316 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1317 continue;
1318
1319 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1320
1321 ret = enter(data, dentry);
1322 switch (ret) {
1323 case D_WALK_CONTINUE:
1324 break;
1325 case D_WALK_QUIT:
1326 spin_unlock(&dentry->d_lock);
1327 goto out_unlock;
1328 case D_WALK_NORETRY:
1329 retry = false;
1330 break;
1331 case D_WALK_SKIP:
1332 spin_unlock(&dentry->d_lock);
1333 continue;
1334 }
1335
1336 if (!hlist_empty(&dentry->d_children)) {
1337 spin_unlock(&this_parent->d_lock);
1338 spin_release(&dentry->d_lock.dep_map, _RET_IP_);
1339 this_parent = dentry;
1340 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1341 goto repeat;
1342 }
1343 spin_unlock(&dentry->d_lock);
1344 }
1345 /*
1346 * All done at this level ... ascend and resume the search.
1347 */
1348 rcu_read_lock();
1349ascend:
1350 if (this_parent != parent) {
1351 dentry = this_parent;
1352 this_parent = dentry->d_parent;
1353
1354 spin_unlock(&dentry->d_lock);
1355 spin_lock(&this_parent->d_lock);
1356
1357 /* might go back up the wrong parent if we have had a rename. */
1358 if (need_seqretry(&rename_lock, seq))
1359 goto rename_retry;
1360 /* go into the first sibling still alive */
1361 hlist_for_each_entry_continue(dentry, d_sib) {
1362 if (likely(!(dentry->d_flags & DCACHE_DENTRY_KILLED))) {
1363 rcu_read_unlock();
1364 goto resume;
1365 }
1366 }
1367 goto ascend;
1368 }
1369 if (need_seqretry(&rename_lock, seq))
1370 goto rename_retry;
1371 rcu_read_unlock();
1372
1373out_unlock:
1374 spin_unlock(&this_parent->d_lock);
1375 done_seqretry(&rename_lock, seq);
1376 return;
1377
1378rename_retry:
1379 spin_unlock(&this_parent->d_lock);
1380 rcu_read_unlock();
1381 BUG_ON(seq & 1);
1382 if (!retry)
1383 return;
1384 seq = 1;
1385 goto again;
1386}
1387
1388struct check_mount {
1389 struct vfsmount *mnt;
1390 unsigned int mounted;
1391};
1392
1393static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1394{
1395 struct check_mount *info = data;
1396 struct path path = { .mnt = info->mnt, .dentry = dentry };
1397
1398 if (likely(!d_mountpoint(dentry)))
1399 return D_WALK_CONTINUE;
1400 if (__path_is_mountpoint(&path)) {
1401 info->mounted = 1;
1402 return D_WALK_QUIT;
1403 }
1404 return D_WALK_CONTINUE;
1405}
1406
1407/**
1408 * path_has_submounts - check for mounts over a dentry in the
1409 * current namespace.
1410 * @parent: path to check.
1411 *
1412 * Return true if the parent or its subdirectories contain
1413 * a mount point in the current namespace.
1414 */
1415int path_has_submounts(const struct path *parent)
1416{
1417 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1418
1419 read_seqlock_excl(&mount_lock);
1420 d_walk(parent->dentry, &data, path_check_mount);
1421 read_sequnlock_excl(&mount_lock);
1422
1423 return data.mounted;
1424}
1425EXPORT_SYMBOL(path_has_submounts);
1426
1427/*
1428 * Called by mount code to set a mountpoint and check if the mountpoint is
1429 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1430 * subtree can become unreachable).
1431 *
1432 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1433 * this reason take rename_lock and d_lock on dentry and ancestors.
1434 */
1435int d_set_mounted(struct dentry *dentry)
1436{
1437 struct dentry *p;
1438 int ret = -ENOENT;
1439 write_seqlock(&rename_lock);
1440 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1441 /* Need exclusion wrt. d_invalidate() */
1442 spin_lock(&p->d_lock);
1443 if (unlikely(d_unhashed(p))) {
1444 spin_unlock(&p->d_lock);
1445 goto out;
1446 }
1447 spin_unlock(&p->d_lock);
1448 }
1449 spin_lock(&dentry->d_lock);
1450 if (!d_unlinked(dentry)) {
1451 ret = -EBUSY;
1452 if (!d_mountpoint(dentry)) {
1453 dentry->d_flags |= DCACHE_MOUNTED;
1454 ret = 0;
1455 }
1456 }
1457 spin_unlock(&dentry->d_lock);
1458out:
1459 write_sequnlock(&rename_lock);
1460 return ret;
1461}
1462
1463/*
1464 * Search the dentry child list of the specified parent,
1465 * and move any unused dentries to the end of the unused
1466 * list for prune_dcache(). We descend to the next level
1467 * whenever the d_children list is non-empty and continue
1468 * searching.
1469 *
1470 * It returns zero iff there are no unused children,
1471 * otherwise it returns the number of children moved to
1472 * the end of the unused list. This may not be the total
1473 * number of unused children, because select_parent can
1474 * drop the lock and return early due to latency
1475 * constraints.
1476 */
1477
1478struct select_data {
1479 struct dentry *start;
1480 union {
1481 long found;
1482 struct dentry *victim;
1483 };
1484 struct list_head dispose;
1485};
1486
1487static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1488{
1489 struct select_data *data = _data;
1490 enum d_walk_ret ret = D_WALK_CONTINUE;
1491
1492 if (data->start == dentry)
1493 goto out;
1494
1495 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1496 data->found++;
1497 } else if (!dentry->d_lockref.count) {
1498 to_shrink_list(dentry, &data->dispose);
1499 data->found++;
1500 } else if (dentry->d_lockref.count < 0) {
1501 data->found++;
1502 }
1503 /*
1504 * We can return to the caller if we have found some (this
1505 * ensures forward progress). We'll be coming back to find
1506 * the rest.
1507 */
1508 if (!list_empty(&data->dispose))
1509 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1510out:
1511 return ret;
1512}
1513
1514static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry)
1515{
1516 struct select_data *data = _data;
1517 enum d_walk_ret ret = D_WALK_CONTINUE;
1518
1519 if (data->start == dentry)
1520 goto out;
1521
1522 if (!dentry->d_lockref.count) {
1523 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1524 rcu_read_lock();
1525 data->victim = dentry;
1526 return D_WALK_QUIT;
1527 }
1528 to_shrink_list(dentry, &data->dispose);
1529 }
1530 /*
1531 * We can return to the caller if we have found some (this
1532 * ensures forward progress). We'll be coming back to find
1533 * the rest.
1534 */
1535 if (!list_empty(&data->dispose))
1536 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1537out:
1538 return ret;
1539}
1540
1541/**
1542 * shrink_dcache_parent - prune dcache
1543 * @parent: parent of entries to prune
1544 *
1545 * Prune the dcache to remove unused children of the parent dentry.
1546 */
1547void shrink_dcache_parent(struct dentry *parent)
1548{
1549 for (;;) {
1550 struct select_data data = {.start = parent};
1551
1552 INIT_LIST_HEAD(&data.dispose);
1553 d_walk(parent, &data, select_collect);
1554
1555 if (!list_empty(&data.dispose)) {
1556 shrink_dentry_list(&data.dispose);
1557 continue;
1558 }
1559
1560 cond_resched();
1561 if (!data.found)
1562 break;
1563 data.victim = NULL;
1564 d_walk(parent, &data, select_collect2);
1565 if (data.victim) {
1566 spin_lock(&data.victim->d_lock);
1567 if (!lock_for_kill(data.victim)) {
1568 spin_unlock(&data.victim->d_lock);
1569 rcu_read_unlock();
1570 } else {
1571 shrink_kill(data.victim);
1572 }
1573 }
1574 if (!list_empty(&data.dispose))
1575 shrink_dentry_list(&data.dispose);
1576 }
1577}
1578EXPORT_SYMBOL(shrink_dcache_parent);
1579
1580static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1581{
1582 /* it has busy descendents; complain about those instead */
1583 if (!hlist_empty(&dentry->d_children))
1584 return D_WALK_CONTINUE;
1585
1586 /* root with refcount 1 is fine */
1587 if (dentry == _data && dentry->d_lockref.count == 1)
1588 return D_WALK_CONTINUE;
1589
1590 WARN(1, "BUG: Dentry %p{i=%lx,n=%pd} "
1591 " still in use (%d) [unmount of %s %s]\n",
1592 dentry,
1593 dentry->d_inode ?
1594 dentry->d_inode->i_ino : 0UL,
1595 dentry,
1596 dentry->d_lockref.count,
1597 dentry->d_sb->s_type->name,
1598 dentry->d_sb->s_id);
1599 return D_WALK_CONTINUE;
1600}
1601
1602static void do_one_tree(struct dentry *dentry)
1603{
1604 shrink_dcache_parent(dentry);
1605 d_walk(dentry, dentry, umount_check);
1606 d_drop(dentry);
1607 dput(dentry);
1608}
1609
1610/*
1611 * destroy the dentries attached to a superblock on unmounting
1612 */
1613void shrink_dcache_for_umount(struct super_block *sb)
1614{
1615 struct dentry *dentry;
1616
1617 rwsem_assert_held_write(&sb->s_umount);
1618
1619 dentry = sb->s_root;
1620 sb->s_root = NULL;
1621 do_one_tree(dentry);
1622
1623 while (!hlist_bl_empty(&sb->s_roots)) {
1624 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1625 do_one_tree(dentry);
1626 }
1627}
1628
1629static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
1630{
1631 struct dentry **victim = _data;
1632 if (d_mountpoint(dentry)) {
1633 *victim = dget_dlock(dentry);
1634 return D_WALK_QUIT;
1635 }
1636 return D_WALK_CONTINUE;
1637}
1638
1639/**
1640 * d_invalidate - detach submounts, prune dcache, and drop
1641 * @dentry: dentry to invalidate (aka detach, prune and drop)
1642 */
1643void d_invalidate(struct dentry *dentry)
1644{
1645 bool had_submounts = false;
1646 spin_lock(&dentry->d_lock);
1647 if (d_unhashed(dentry)) {
1648 spin_unlock(&dentry->d_lock);
1649 return;
1650 }
1651 __d_drop(dentry);
1652 spin_unlock(&dentry->d_lock);
1653
1654 /* Negative dentries can be dropped without further checks */
1655 if (!dentry->d_inode)
1656 return;
1657
1658 shrink_dcache_parent(dentry);
1659 for (;;) {
1660 struct dentry *victim = NULL;
1661 d_walk(dentry, &victim, find_submount);
1662 if (!victim) {
1663 if (had_submounts)
1664 shrink_dcache_parent(dentry);
1665 return;
1666 }
1667 had_submounts = true;
1668 detach_mounts(victim);
1669 dput(victim);
1670 }
1671}
1672EXPORT_SYMBOL(d_invalidate);
1673
1674/**
1675 * __d_alloc - allocate a dcache entry
1676 * @sb: filesystem it will belong to
1677 * @name: qstr of the name
1678 *
1679 * Allocates a dentry. It returns %NULL if there is insufficient memory
1680 * available. On a success the dentry is returned. The name passed in is
1681 * copied and the copy passed in may be reused after this call.
1682 */
1683
1684static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1685{
1686 struct dentry *dentry;
1687 char *dname;
1688 int err;
1689
1690 dentry = kmem_cache_alloc_lru(dentry_cache, &sb->s_dentry_lru,
1691 GFP_KERNEL);
1692 if (!dentry)
1693 return NULL;
1694
1695 /*
1696 * We guarantee that the inline name is always NUL-terminated.
1697 * This way the memcpy() done by the name switching in rename
1698 * will still always have a NUL at the end, even if we might
1699 * be overwriting an internal NUL character
1700 */
1701 dentry->d_shortname.string[DNAME_INLINE_LEN-1] = 0;
1702 if (unlikely(!name)) {
1703 name = &slash_name;
1704 dname = dentry->d_shortname.string;
1705 } else if (name->len > DNAME_INLINE_LEN-1) {
1706 size_t size = offsetof(struct external_name, name[1]);
1707 struct external_name *p = kmalloc(size + name->len,
1708 GFP_KERNEL_ACCOUNT |
1709 __GFP_RECLAIMABLE);
1710 if (!p) {
1711 kmem_cache_free(dentry_cache, dentry);
1712 return NULL;
1713 }
1714 atomic_set(&p->count, 1);
1715 dname = p->name;
1716 } else {
1717 dname = dentry->d_shortname.string;
1718 }
1719
1720 dentry->d_name.len = name->len;
1721 dentry->d_name.hash = name->hash;
1722 memcpy(dname, name->name, name->len);
1723 dname[name->len] = 0;
1724
1725 /* Make sure we always see the terminating NUL character */
1726 smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1727
1728 dentry->d_flags = 0;
1729 lockref_init(&dentry->d_lockref);
1730 seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock);
1731 dentry->d_inode = NULL;
1732 dentry->d_parent = dentry;
1733 dentry->d_sb = sb;
1734 dentry->d_op = NULL;
1735 dentry->d_fsdata = NULL;
1736 INIT_HLIST_BL_NODE(&dentry->d_hash);
1737 INIT_LIST_HEAD(&dentry->d_lru);
1738 INIT_HLIST_HEAD(&dentry->d_children);
1739 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1740 INIT_HLIST_NODE(&dentry->d_sib);
1741 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1742
1743 if (dentry->d_op && dentry->d_op->d_init) {
1744 err = dentry->d_op->d_init(dentry);
1745 if (err) {
1746 if (dname_external(dentry))
1747 kfree(external_name(dentry));
1748 kmem_cache_free(dentry_cache, dentry);
1749 return NULL;
1750 }
1751 }
1752
1753 this_cpu_inc(nr_dentry);
1754
1755 return dentry;
1756}
1757
1758/**
1759 * d_alloc - allocate a dcache entry
1760 * @parent: parent of entry to allocate
1761 * @name: qstr of the name
1762 *
1763 * Allocates a dentry. It returns %NULL if there is insufficient memory
1764 * available. On a success the dentry is returned. The name passed in is
1765 * copied and the copy passed in may be reused after this call.
1766 */
1767struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1768{
1769 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1770 if (!dentry)
1771 return NULL;
1772 spin_lock(&parent->d_lock);
1773 /*
1774 * don't need child lock because it is not subject
1775 * to concurrency here
1776 */
1777 dentry->d_parent = dget_dlock(parent);
1778 hlist_add_head(&dentry->d_sib, &parent->d_children);
1779 spin_unlock(&parent->d_lock);
1780
1781 return dentry;
1782}
1783EXPORT_SYMBOL(d_alloc);
1784
1785struct dentry *d_alloc_anon(struct super_block *sb)
1786{
1787 return __d_alloc(sb, NULL);
1788}
1789EXPORT_SYMBOL(d_alloc_anon);
1790
1791struct dentry *d_alloc_cursor(struct dentry * parent)
1792{
1793 struct dentry *dentry = d_alloc_anon(parent->d_sb);
1794 if (dentry) {
1795 dentry->d_flags |= DCACHE_DENTRY_CURSOR;
1796 dentry->d_parent = dget(parent);
1797 }
1798 return dentry;
1799}
1800
1801/**
1802 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1803 * @sb: the superblock
1804 * @name: qstr of the name
1805 *
1806 * For a filesystem that just pins its dentries in memory and never
1807 * performs lookups at all, return an unhashed IS_ROOT dentry.
1808 * This is used for pipes, sockets et.al. - the stuff that should
1809 * never be anyone's children or parents. Unlike all other
1810 * dentries, these will not have RCU delay between dropping the
1811 * last reference and freeing them.
1812 *
1813 * The only user is alloc_file_pseudo() and that's what should
1814 * be considered a public interface. Don't use directly.
1815 */
1816struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1817{
1818 static const struct dentry_operations anon_ops = {
1819 .d_dname = simple_dname
1820 };
1821 struct dentry *dentry = __d_alloc(sb, name);
1822 if (likely(dentry)) {
1823 dentry->d_flags |= DCACHE_NORCU;
1824 if (!sb->s_d_op)
1825 d_set_d_op(dentry, &anon_ops);
1826 }
1827 return dentry;
1828}
1829
1830struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1831{
1832 struct qstr q;
1833
1834 q.name = name;
1835 q.hash_len = hashlen_string(parent, name);
1836 return d_alloc(parent, &q);
1837}
1838EXPORT_SYMBOL(d_alloc_name);
1839
1840void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1841{
1842 WARN_ON_ONCE(dentry->d_op);
1843 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1844 DCACHE_OP_COMPARE |
1845 DCACHE_OP_REVALIDATE |
1846 DCACHE_OP_WEAK_REVALIDATE |
1847 DCACHE_OP_DELETE |
1848 DCACHE_OP_REAL));
1849 dentry->d_op = op;
1850 if (!op)
1851 return;
1852 if (op->d_hash)
1853 dentry->d_flags |= DCACHE_OP_HASH;
1854 if (op->d_compare)
1855 dentry->d_flags |= DCACHE_OP_COMPARE;
1856 if (op->d_revalidate)
1857 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1858 if (op->d_weak_revalidate)
1859 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1860 if (op->d_delete)
1861 dentry->d_flags |= DCACHE_OP_DELETE;
1862 if (op->d_prune)
1863 dentry->d_flags |= DCACHE_OP_PRUNE;
1864 if (op->d_real)
1865 dentry->d_flags |= DCACHE_OP_REAL;
1866
1867}
1868EXPORT_SYMBOL(d_set_d_op);
1869
1870static unsigned d_flags_for_inode(struct inode *inode)
1871{
1872 unsigned add_flags = DCACHE_REGULAR_TYPE;
1873
1874 if (!inode)
1875 return DCACHE_MISS_TYPE;
1876
1877 if (S_ISDIR(inode->i_mode)) {
1878 add_flags = DCACHE_DIRECTORY_TYPE;
1879 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1880 if (unlikely(!inode->i_op->lookup))
1881 add_flags = DCACHE_AUTODIR_TYPE;
1882 else
1883 inode->i_opflags |= IOP_LOOKUP;
1884 }
1885 goto type_determined;
1886 }
1887
1888 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1889 if (unlikely(inode->i_op->get_link)) {
1890 add_flags = DCACHE_SYMLINK_TYPE;
1891 goto type_determined;
1892 }
1893 inode->i_opflags |= IOP_NOFOLLOW;
1894 }
1895
1896 if (unlikely(!S_ISREG(inode->i_mode)))
1897 add_flags = DCACHE_SPECIAL_TYPE;
1898
1899type_determined:
1900 if (unlikely(IS_AUTOMOUNT(inode)))
1901 add_flags |= DCACHE_NEED_AUTOMOUNT;
1902 return add_flags;
1903}
1904
1905static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1906{
1907 unsigned add_flags = d_flags_for_inode(inode);
1908 WARN_ON(d_in_lookup(dentry));
1909
1910 spin_lock(&dentry->d_lock);
1911 /*
1912 * The negative counter only tracks dentries on the LRU. Don't dec if
1913 * d_lru is on another list.
1914 */
1915 if ((dentry->d_flags &
1916 (DCACHE_LRU_LIST|DCACHE_SHRINK_LIST)) == DCACHE_LRU_LIST)
1917 this_cpu_dec(nr_dentry_negative);
1918 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1919 raw_write_seqcount_begin(&dentry->d_seq);
1920 __d_set_inode_and_type(dentry, inode, add_flags);
1921 raw_write_seqcount_end(&dentry->d_seq);
1922 fsnotify_update_flags(dentry);
1923 spin_unlock(&dentry->d_lock);
1924}
1925
1926/**
1927 * d_instantiate - fill in inode information for a dentry
1928 * @entry: dentry to complete
1929 * @inode: inode to attach to this dentry
1930 *
1931 * Fill in inode information in the entry.
1932 *
1933 * This turns negative dentries into productive full members
1934 * of society.
1935 *
1936 * NOTE! This assumes that the inode count has been incremented
1937 * (or otherwise set) by the caller to indicate that it is now
1938 * in use by the dcache.
1939 */
1940
1941void d_instantiate(struct dentry *entry, struct inode * inode)
1942{
1943 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1944 if (inode) {
1945 security_d_instantiate(entry, inode);
1946 spin_lock(&inode->i_lock);
1947 __d_instantiate(entry, inode);
1948 spin_unlock(&inode->i_lock);
1949 }
1950}
1951EXPORT_SYMBOL(d_instantiate);
1952
1953/*
1954 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1955 * with lockdep-related part of unlock_new_inode() done before
1956 * anything else. Use that instead of open-coding d_instantiate()/
1957 * unlock_new_inode() combinations.
1958 */
1959void d_instantiate_new(struct dentry *entry, struct inode *inode)
1960{
1961 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1962 BUG_ON(!inode);
1963 lockdep_annotate_inode_mutex_key(inode);
1964 security_d_instantiate(entry, inode);
1965 spin_lock(&inode->i_lock);
1966 __d_instantiate(entry, inode);
1967 WARN_ON(!(inode->i_state & I_NEW));
1968 inode->i_state &= ~I_NEW & ~I_CREATING;
1969 /*
1970 * Pairs with the barrier in prepare_to_wait_event() to make sure
1971 * ___wait_var_event() either sees the bit cleared or
1972 * waitqueue_active() check in wake_up_var() sees the waiter.
1973 */
1974 smp_mb();
1975 inode_wake_up_bit(inode, __I_NEW);
1976 spin_unlock(&inode->i_lock);
1977}
1978EXPORT_SYMBOL(d_instantiate_new);
1979
1980struct dentry *d_make_root(struct inode *root_inode)
1981{
1982 struct dentry *res = NULL;
1983
1984 if (root_inode) {
1985 res = d_alloc_anon(root_inode->i_sb);
1986 if (res)
1987 d_instantiate(res, root_inode);
1988 else
1989 iput(root_inode);
1990 }
1991 return res;
1992}
1993EXPORT_SYMBOL(d_make_root);
1994
1995static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
1996{
1997 struct super_block *sb;
1998 struct dentry *new, *res;
1999
2000 if (!inode)
2001 return ERR_PTR(-ESTALE);
2002 if (IS_ERR(inode))
2003 return ERR_CAST(inode);
2004
2005 sb = inode->i_sb;
2006
2007 res = d_find_any_alias(inode); /* existing alias? */
2008 if (res)
2009 goto out;
2010
2011 new = d_alloc_anon(sb);
2012 if (!new) {
2013 res = ERR_PTR(-ENOMEM);
2014 goto out;
2015 }
2016
2017 security_d_instantiate(new, inode);
2018 spin_lock(&inode->i_lock);
2019 res = __d_find_any_alias(inode); /* recheck under lock */
2020 if (likely(!res)) { /* still no alias, attach a disconnected dentry */
2021 unsigned add_flags = d_flags_for_inode(inode);
2022
2023 if (disconnected)
2024 add_flags |= DCACHE_DISCONNECTED;
2025
2026 spin_lock(&new->d_lock);
2027 __d_set_inode_and_type(new, inode, add_flags);
2028 hlist_add_head(&new->d_u.d_alias, &inode->i_dentry);
2029 if (!disconnected) {
2030 hlist_bl_lock(&sb->s_roots);
2031 hlist_bl_add_head(&new->d_hash, &sb->s_roots);
2032 hlist_bl_unlock(&sb->s_roots);
2033 }
2034 spin_unlock(&new->d_lock);
2035 spin_unlock(&inode->i_lock);
2036 inode = NULL; /* consumed by new->d_inode */
2037 res = new;
2038 } else {
2039 spin_unlock(&inode->i_lock);
2040 dput(new);
2041 }
2042
2043 out:
2044 iput(inode);
2045 return res;
2046}
2047
2048/**
2049 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2050 * @inode: inode to allocate the dentry for
2051 *
2052 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2053 * similar open by handle operations. The returned dentry may be anonymous,
2054 * or may have a full name (if the inode was already in the cache).
2055 *
2056 * When called on a directory inode, we must ensure that the inode only ever
2057 * has one dentry. If a dentry is found, that is returned instead of
2058 * allocating a new one.
2059 *
2060 * On successful return, the reference to the inode has been transferred
2061 * to the dentry. In case of an error the reference on the inode is released.
2062 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2063 * be passed in and the error will be propagated to the return value,
2064 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2065 */
2066struct dentry *d_obtain_alias(struct inode *inode)
2067{
2068 return __d_obtain_alias(inode, true);
2069}
2070EXPORT_SYMBOL(d_obtain_alias);
2071
2072/**
2073 * d_obtain_root - find or allocate a dentry for a given inode
2074 * @inode: inode to allocate the dentry for
2075 *
2076 * Obtain an IS_ROOT dentry for the root of a filesystem.
2077 *
2078 * We must ensure that directory inodes only ever have one dentry. If a
2079 * dentry is found, that is returned instead of allocating a new one.
2080 *
2081 * On successful return, the reference to the inode has been transferred
2082 * to the dentry. In case of an error the reference on the inode is
2083 * released. A %NULL or IS_ERR inode may be passed in and will be the
2084 * error will be propagate to the return value, with a %NULL @inode
2085 * replaced by ERR_PTR(-ESTALE).
2086 */
2087struct dentry *d_obtain_root(struct inode *inode)
2088{
2089 return __d_obtain_alias(inode, false);
2090}
2091EXPORT_SYMBOL(d_obtain_root);
2092
2093/**
2094 * d_add_ci - lookup or allocate new dentry with case-exact name
2095 * @dentry: the negative dentry that was passed to the parent's lookup func
2096 * @inode: the inode case-insensitive lookup has found
2097 * @name: the case-exact name to be associated with the returned dentry
2098 *
2099 * This is to avoid filling the dcache with case-insensitive names to the
2100 * same inode, only the actual correct case is stored in the dcache for
2101 * case-insensitive filesystems.
2102 *
2103 * For a case-insensitive lookup match and if the case-exact dentry
2104 * already exists in the dcache, use it and return it.
2105 *
2106 * If no entry exists with the exact case name, allocate new dentry with
2107 * the exact case, and return the spliced entry.
2108 */
2109struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2110 struct qstr *name)
2111{
2112 struct dentry *found, *res;
2113
2114 /*
2115 * First check if a dentry matching the name already exists,
2116 * if not go ahead and create it now.
2117 */
2118 found = d_hash_and_lookup(dentry->d_parent, name);
2119 if (found) {
2120 iput(inode);
2121 return found;
2122 }
2123 if (d_in_lookup(dentry)) {
2124 found = d_alloc_parallel(dentry->d_parent, name,
2125 dentry->d_wait);
2126 if (IS_ERR(found) || !d_in_lookup(found)) {
2127 iput(inode);
2128 return found;
2129 }
2130 } else {
2131 found = d_alloc(dentry->d_parent, name);
2132 if (!found) {
2133 iput(inode);
2134 return ERR_PTR(-ENOMEM);
2135 }
2136 }
2137 res = d_splice_alias(inode, found);
2138 if (res) {
2139 d_lookup_done(found);
2140 dput(found);
2141 return res;
2142 }
2143 return found;
2144}
2145EXPORT_SYMBOL(d_add_ci);
2146
2147/**
2148 * d_same_name - compare dentry name with case-exact name
2149 * @dentry: the negative dentry that was passed to the parent's lookup func
2150 * @parent: parent dentry
2151 * @name: the case-exact name to be associated with the returned dentry
2152 *
2153 * Return: true if names are same, or false
2154 */
2155bool d_same_name(const struct dentry *dentry, const struct dentry *parent,
2156 const struct qstr *name)
2157{
2158 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2159 if (dentry->d_name.len != name->len)
2160 return false;
2161 return dentry_cmp(dentry, name->name, name->len) == 0;
2162 }
2163 return parent->d_op->d_compare(dentry,
2164 dentry->d_name.len, dentry->d_name.name,
2165 name) == 0;
2166}
2167EXPORT_SYMBOL_GPL(d_same_name);
2168
2169/*
2170 * This is __d_lookup_rcu() when the parent dentry has
2171 * DCACHE_OP_COMPARE, which makes things much nastier.
2172 */
2173static noinline struct dentry *__d_lookup_rcu_op_compare(
2174 const struct dentry *parent,
2175 const struct qstr *name,
2176 unsigned *seqp)
2177{
2178 u64 hashlen = name->hash_len;
2179 struct hlist_bl_head *b = d_hash(hashlen);
2180 struct hlist_bl_node *node;
2181 struct dentry *dentry;
2182
2183 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2184 int tlen;
2185 const char *tname;
2186 unsigned seq;
2187
2188seqretry:
2189 seq = raw_seqcount_begin(&dentry->d_seq);
2190 if (dentry->d_parent != parent)
2191 continue;
2192 if (d_unhashed(dentry))
2193 continue;
2194 if (dentry->d_name.hash != hashlen_hash(hashlen))
2195 continue;
2196 tlen = dentry->d_name.len;
2197 tname = dentry->d_name.name;
2198 /* we want a consistent (name,len) pair */
2199 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2200 cpu_relax();
2201 goto seqretry;
2202 }
2203 if (parent->d_op->d_compare(dentry, tlen, tname, name) != 0)
2204 continue;
2205 *seqp = seq;
2206 return dentry;
2207 }
2208 return NULL;
2209}
2210
2211/**
2212 * __d_lookup_rcu - search for a dentry (racy, store-free)
2213 * @parent: parent dentry
2214 * @name: qstr of name we wish to find
2215 * @seqp: returns d_seq value at the point where the dentry was found
2216 * Returns: dentry, or NULL
2217 *
2218 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2219 * resolution (store-free path walking) design described in
2220 * Documentation/filesystems/path-lookup.txt.
2221 *
2222 * This is not to be used outside core vfs.
2223 *
2224 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2225 * held, and rcu_read_lock held. The returned dentry must not be stored into
2226 * without taking d_lock and checking d_seq sequence count against @seq
2227 * returned here.
2228 *
2229 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2230 * the returned dentry, so long as its parent's seqlock is checked after the
2231 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2232 * is formed, giving integrity down the path walk.
2233 *
2234 * NOTE! The caller *has* to check the resulting dentry against the sequence
2235 * number we've returned before using any of the resulting dentry state!
2236 */
2237struct dentry *__d_lookup_rcu(const struct dentry *parent,
2238 const struct qstr *name,
2239 unsigned *seqp)
2240{
2241 u64 hashlen = name->hash_len;
2242 const unsigned char *str = name->name;
2243 struct hlist_bl_head *b = d_hash(hashlen);
2244 struct hlist_bl_node *node;
2245 struct dentry *dentry;
2246
2247 /*
2248 * Note: There is significant duplication with __d_lookup_rcu which is
2249 * required to prevent single threaded performance regressions
2250 * especially on architectures where smp_rmb (in seqcounts) are costly.
2251 * Keep the two functions in sync.
2252 */
2253
2254 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE))
2255 return __d_lookup_rcu_op_compare(parent, name, seqp);
2256
2257 /*
2258 * The hash list is protected using RCU.
2259 *
2260 * Carefully use d_seq when comparing a candidate dentry, to avoid
2261 * races with d_move().
2262 *
2263 * It is possible that concurrent renames can mess up our list
2264 * walk here and result in missing our dentry, resulting in the
2265 * false-negative result. d_lookup() protects against concurrent
2266 * renames using rename_lock seqlock.
2267 *
2268 * See Documentation/filesystems/path-lookup.txt for more details.
2269 */
2270 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2271 unsigned seq;
2272
2273 /*
2274 * The dentry sequence count protects us from concurrent
2275 * renames, and thus protects parent and name fields.
2276 *
2277 * The caller must perform a seqcount check in order
2278 * to do anything useful with the returned dentry.
2279 *
2280 * NOTE! We do a "raw" seqcount_begin here. That means that
2281 * we don't wait for the sequence count to stabilize if it
2282 * is in the middle of a sequence change. If we do the slow
2283 * dentry compare, we will do seqretries until it is stable,
2284 * and if we end up with a successful lookup, we actually
2285 * want to exit RCU lookup anyway.
2286 *
2287 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2288 * we are still guaranteed NUL-termination of ->d_name.name.
2289 */
2290 seq = raw_seqcount_begin(&dentry->d_seq);
2291 if (dentry->d_parent != parent)
2292 continue;
2293 if (d_unhashed(dentry))
2294 continue;
2295 if (dentry->d_name.hash_len != hashlen)
2296 continue;
2297 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2298 continue;
2299 *seqp = seq;
2300 return dentry;
2301 }
2302 return NULL;
2303}
2304
2305/**
2306 * d_lookup - search for a dentry
2307 * @parent: parent dentry
2308 * @name: qstr of name we wish to find
2309 * Returns: dentry, or NULL
2310 *
2311 * d_lookup searches the children of the parent dentry for the name in
2312 * question. If the dentry is found its reference count is incremented and the
2313 * dentry is returned. The caller must use dput to free the entry when it has
2314 * finished using it. %NULL is returned if the dentry does not exist.
2315 */
2316struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2317{
2318 struct dentry *dentry;
2319 unsigned seq;
2320
2321 do {
2322 seq = read_seqbegin(&rename_lock);
2323 dentry = __d_lookup(parent, name);
2324 if (dentry)
2325 break;
2326 } while (read_seqretry(&rename_lock, seq));
2327 return dentry;
2328}
2329EXPORT_SYMBOL(d_lookup);
2330
2331/**
2332 * __d_lookup - search for a dentry (racy)
2333 * @parent: parent dentry
2334 * @name: qstr of name we wish to find
2335 * Returns: dentry, or NULL
2336 *
2337 * __d_lookup is like d_lookup, however it may (rarely) return a
2338 * false-negative result due to unrelated rename activity.
2339 *
2340 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2341 * however it must be used carefully, eg. with a following d_lookup in
2342 * the case of failure.
2343 *
2344 * __d_lookup callers must be commented.
2345 */
2346struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2347{
2348 unsigned int hash = name->hash;
2349 struct hlist_bl_head *b = d_hash(hash);
2350 struct hlist_bl_node *node;
2351 struct dentry *found = NULL;
2352 struct dentry *dentry;
2353
2354 /*
2355 * Note: There is significant duplication with __d_lookup_rcu which is
2356 * required to prevent single threaded performance regressions
2357 * especially on architectures where smp_rmb (in seqcounts) are costly.
2358 * Keep the two functions in sync.
2359 */
2360
2361 /*
2362 * The hash list is protected using RCU.
2363 *
2364 * Take d_lock when comparing a candidate dentry, to avoid races
2365 * with d_move().
2366 *
2367 * It is possible that concurrent renames can mess up our list
2368 * walk here and result in missing our dentry, resulting in the
2369 * false-negative result. d_lookup() protects against concurrent
2370 * renames using rename_lock seqlock.
2371 *
2372 * See Documentation/filesystems/path-lookup.txt for more details.
2373 */
2374 rcu_read_lock();
2375
2376 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2377
2378 if (dentry->d_name.hash != hash)
2379 continue;
2380
2381 spin_lock(&dentry->d_lock);
2382 if (dentry->d_parent != parent)
2383 goto next;
2384 if (d_unhashed(dentry))
2385 goto next;
2386
2387 if (!d_same_name(dentry, parent, name))
2388 goto next;
2389
2390 dentry->d_lockref.count++;
2391 found = dentry;
2392 spin_unlock(&dentry->d_lock);
2393 break;
2394next:
2395 spin_unlock(&dentry->d_lock);
2396 }
2397 rcu_read_unlock();
2398
2399 return found;
2400}
2401
2402/**
2403 * d_hash_and_lookup - hash the qstr then search for a dentry
2404 * @dir: Directory to search in
2405 * @name: qstr of name we wish to find
2406 *
2407 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2408 */
2409struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2410{
2411 /*
2412 * Check for a fs-specific hash function. Note that we must
2413 * calculate the standard hash first, as the d_op->d_hash()
2414 * routine may choose to leave the hash value unchanged.
2415 */
2416 name->hash = full_name_hash(dir, name->name, name->len);
2417 if (dir->d_flags & DCACHE_OP_HASH) {
2418 int err = dir->d_op->d_hash(dir, name);
2419 if (unlikely(err < 0))
2420 return ERR_PTR(err);
2421 }
2422 return d_lookup(dir, name);
2423}
2424
2425/*
2426 * When a file is deleted, we have two options:
2427 * - turn this dentry into a negative dentry
2428 * - unhash this dentry and free it.
2429 *
2430 * Usually, we want to just turn this into
2431 * a negative dentry, but if anybody else is
2432 * currently using the dentry or the inode
2433 * we can't do that and we fall back on removing
2434 * it from the hash queues and waiting for
2435 * it to be deleted later when it has no users
2436 */
2437
2438/**
2439 * d_delete - delete a dentry
2440 * @dentry: The dentry to delete
2441 *
2442 * Turn the dentry into a negative dentry if possible, otherwise
2443 * remove it from the hash queues so it can be deleted later
2444 */
2445
2446void d_delete(struct dentry * dentry)
2447{
2448 struct inode *inode = dentry->d_inode;
2449
2450 spin_lock(&inode->i_lock);
2451 spin_lock(&dentry->d_lock);
2452 /*
2453 * Are we the only user?
2454 */
2455 if (dentry->d_lockref.count == 1) {
2456 if (dentry_negative_policy)
2457 __d_drop(dentry);
2458 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2459 dentry_unlink_inode(dentry);
2460 } else {
2461 __d_drop(dentry);
2462 spin_unlock(&dentry->d_lock);
2463 spin_unlock(&inode->i_lock);
2464 }
2465}
2466EXPORT_SYMBOL(d_delete);
2467
2468static void __d_rehash(struct dentry *entry)
2469{
2470 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2471
2472 hlist_bl_lock(b);
2473 hlist_bl_add_head_rcu(&entry->d_hash, b);
2474 hlist_bl_unlock(b);
2475}
2476
2477/**
2478 * d_rehash - add an entry back to the hash
2479 * @entry: dentry to add to the hash
2480 *
2481 * Adds a dentry to the hash according to its name.
2482 */
2483
2484void d_rehash(struct dentry * entry)
2485{
2486 spin_lock(&entry->d_lock);
2487 __d_rehash(entry);
2488 spin_unlock(&entry->d_lock);
2489}
2490EXPORT_SYMBOL(d_rehash);
2491
2492static inline unsigned start_dir_add(struct inode *dir)
2493{
2494 preempt_disable_nested();
2495 for (;;) {
2496 unsigned n = dir->i_dir_seq;
2497 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2498 return n;
2499 cpu_relax();
2500 }
2501}
2502
2503static inline void end_dir_add(struct inode *dir, unsigned int n,
2504 wait_queue_head_t *d_wait)
2505{
2506 smp_store_release(&dir->i_dir_seq, n + 2);
2507 preempt_enable_nested();
2508 if (wq_has_sleeper(d_wait))
2509 wake_up_all(d_wait);
2510}
2511
2512static void d_wait_lookup(struct dentry *dentry)
2513{
2514 if (d_in_lookup(dentry)) {
2515 DECLARE_WAITQUEUE(wait, current);
2516 add_wait_queue(dentry->d_wait, &wait);
2517 do {
2518 set_current_state(TASK_UNINTERRUPTIBLE);
2519 spin_unlock(&dentry->d_lock);
2520 schedule();
2521 spin_lock(&dentry->d_lock);
2522 } while (d_in_lookup(dentry));
2523 }
2524}
2525
2526struct dentry *d_alloc_parallel(struct dentry *parent,
2527 const struct qstr *name,
2528 wait_queue_head_t *wq)
2529{
2530 unsigned int hash = name->hash;
2531 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2532 struct hlist_bl_node *node;
2533 struct dentry *new = d_alloc(parent, name);
2534 struct dentry *dentry;
2535 unsigned seq, r_seq, d_seq;
2536
2537 if (unlikely(!new))
2538 return ERR_PTR(-ENOMEM);
2539
2540retry:
2541 rcu_read_lock();
2542 seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2543 r_seq = read_seqbegin(&rename_lock);
2544 dentry = __d_lookup_rcu(parent, name, &d_seq);
2545 if (unlikely(dentry)) {
2546 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2547 rcu_read_unlock();
2548 goto retry;
2549 }
2550 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2551 rcu_read_unlock();
2552 dput(dentry);
2553 goto retry;
2554 }
2555 rcu_read_unlock();
2556 dput(new);
2557 return dentry;
2558 }
2559 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2560 rcu_read_unlock();
2561 goto retry;
2562 }
2563
2564 if (unlikely(seq & 1)) {
2565 rcu_read_unlock();
2566 goto retry;
2567 }
2568
2569 hlist_bl_lock(b);
2570 if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2571 hlist_bl_unlock(b);
2572 rcu_read_unlock();
2573 goto retry;
2574 }
2575 /*
2576 * No changes for the parent since the beginning of d_lookup().
2577 * Since all removals from the chain happen with hlist_bl_lock(),
2578 * any potential in-lookup matches are going to stay here until
2579 * we unlock the chain. All fields are stable in everything
2580 * we encounter.
2581 */
2582 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2583 if (dentry->d_name.hash != hash)
2584 continue;
2585 if (dentry->d_parent != parent)
2586 continue;
2587 if (!d_same_name(dentry, parent, name))
2588 continue;
2589 hlist_bl_unlock(b);
2590 /* now we can try to grab a reference */
2591 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2592 rcu_read_unlock();
2593 goto retry;
2594 }
2595
2596 rcu_read_unlock();
2597 /*
2598 * somebody is likely to be still doing lookup for it;
2599 * wait for them to finish
2600 */
2601 spin_lock(&dentry->d_lock);
2602 d_wait_lookup(dentry);
2603 /*
2604 * it's not in-lookup anymore; in principle we should repeat
2605 * everything from dcache lookup, but it's likely to be what
2606 * d_lookup() would've found anyway. If it is, just return it;
2607 * otherwise we really have to repeat the whole thing.
2608 */
2609 if (unlikely(dentry->d_name.hash != hash))
2610 goto mismatch;
2611 if (unlikely(dentry->d_parent != parent))
2612 goto mismatch;
2613 if (unlikely(d_unhashed(dentry)))
2614 goto mismatch;
2615 if (unlikely(!d_same_name(dentry, parent, name)))
2616 goto mismatch;
2617 /* OK, it *is* a hashed match; return it */
2618 spin_unlock(&dentry->d_lock);
2619 dput(new);
2620 return dentry;
2621 }
2622 rcu_read_unlock();
2623 /* we can't take ->d_lock here; it's OK, though. */
2624 new->d_flags |= DCACHE_PAR_LOOKUP;
2625 new->d_wait = wq;
2626 hlist_bl_add_head(&new->d_u.d_in_lookup_hash, b);
2627 hlist_bl_unlock(b);
2628 return new;
2629mismatch:
2630 spin_unlock(&dentry->d_lock);
2631 dput(dentry);
2632 goto retry;
2633}
2634EXPORT_SYMBOL(d_alloc_parallel);
2635
2636/*
2637 * - Unhash the dentry
2638 * - Retrieve and clear the waitqueue head in dentry
2639 * - Return the waitqueue head
2640 */
2641static wait_queue_head_t *__d_lookup_unhash(struct dentry *dentry)
2642{
2643 wait_queue_head_t *d_wait;
2644 struct hlist_bl_head *b;
2645
2646 lockdep_assert_held(&dentry->d_lock);
2647
2648 b = in_lookup_hash(dentry->d_parent, dentry->d_name.hash);
2649 hlist_bl_lock(b);
2650 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2651 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2652 d_wait = dentry->d_wait;
2653 dentry->d_wait = NULL;
2654 hlist_bl_unlock(b);
2655 INIT_HLIST_NODE(&dentry->d_u.d_alias);
2656 INIT_LIST_HEAD(&dentry->d_lru);
2657 return d_wait;
2658}
2659
2660void __d_lookup_unhash_wake(struct dentry *dentry)
2661{
2662 spin_lock(&dentry->d_lock);
2663 wake_up_all(__d_lookup_unhash(dentry));
2664 spin_unlock(&dentry->d_lock);
2665}
2666EXPORT_SYMBOL(__d_lookup_unhash_wake);
2667
2668/* inode->i_lock held if inode is non-NULL */
2669
2670static inline void __d_add(struct dentry *dentry, struct inode *inode)
2671{
2672 wait_queue_head_t *d_wait;
2673 struct inode *dir = NULL;
2674 unsigned n;
2675 spin_lock(&dentry->d_lock);
2676 if (unlikely(d_in_lookup(dentry))) {
2677 dir = dentry->d_parent->d_inode;
2678 n = start_dir_add(dir);
2679 d_wait = __d_lookup_unhash(dentry);
2680 }
2681 if (inode) {
2682 unsigned add_flags = d_flags_for_inode(inode);
2683 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2684 raw_write_seqcount_begin(&dentry->d_seq);
2685 __d_set_inode_and_type(dentry, inode, add_flags);
2686 raw_write_seqcount_end(&dentry->d_seq);
2687 fsnotify_update_flags(dentry);
2688 }
2689 __d_rehash(dentry);
2690 if (dir)
2691 end_dir_add(dir, n, d_wait);
2692 spin_unlock(&dentry->d_lock);
2693 if (inode)
2694 spin_unlock(&inode->i_lock);
2695}
2696
2697/**
2698 * d_add - add dentry to hash queues
2699 * @entry: dentry to add
2700 * @inode: The inode to attach to this dentry
2701 *
2702 * This adds the entry to the hash queues and initializes @inode.
2703 * The entry was actually filled in earlier during d_alloc().
2704 */
2705
2706void d_add(struct dentry *entry, struct inode *inode)
2707{
2708 if (inode) {
2709 security_d_instantiate(entry, inode);
2710 spin_lock(&inode->i_lock);
2711 }
2712 __d_add(entry, inode);
2713}
2714EXPORT_SYMBOL(d_add);
2715
2716static void swap_names(struct dentry *dentry, struct dentry *target)
2717{
2718 if (unlikely(dname_external(target))) {
2719 if (unlikely(dname_external(dentry))) {
2720 /*
2721 * Both external: swap the pointers
2722 */
2723 swap(target->d_name.name, dentry->d_name.name);
2724 } else {
2725 /*
2726 * dentry:internal, target:external. Steal target's
2727 * storage and make target internal.
2728 */
2729 dentry->d_name.name = target->d_name.name;
2730 target->d_shortname = dentry->d_shortname;
2731 target->d_name.name = target->d_shortname.string;
2732 }
2733 } else {
2734 if (unlikely(dname_external(dentry))) {
2735 /*
2736 * dentry:external, target:internal. Give dentry's
2737 * storage to target and make dentry internal
2738 */
2739 target->d_name.name = dentry->d_name.name;
2740 dentry->d_shortname = target->d_shortname;
2741 dentry->d_name.name = dentry->d_shortname.string;
2742 } else {
2743 /*
2744 * Both are internal.
2745 */
2746 for (int i = 0; i < DNAME_INLINE_WORDS; i++)
2747 swap(dentry->d_shortname.words[i],
2748 target->d_shortname.words[i]);
2749 }
2750 }
2751 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2752}
2753
2754static void copy_name(struct dentry *dentry, struct dentry *target)
2755{
2756 struct external_name *old_name = NULL;
2757 if (unlikely(dname_external(dentry)))
2758 old_name = external_name(dentry);
2759 if (unlikely(dname_external(target))) {
2760 atomic_inc(&external_name(target)->count);
2761 dentry->d_name = target->d_name;
2762 } else {
2763 dentry->d_shortname = target->d_shortname;
2764 dentry->d_name.name = dentry->d_shortname.string;
2765 dentry->d_name.hash_len = target->d_name.hash_len;
2766 }
2767 if (old_name && likely(atomic_dec_and_test(&old_name->count)))
2768 kfree_rcu(old_name, head);
2769}
2770
2771/*
2772 * __d_move - move a dentry
2773 * @dentry: entry to move
2774 * @target: new dentry
2775 * @exchange: exchange the two dentries
2776 *
2777 * Update the dcache to reflect the move of a file name. Negative
2778 * dcache entries should not be moved in this way. Caller must hold
2779 * rename_lock, the i_mutex of the source and target directories,
2780 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2781 */
2782static void __d_move(struct dentry *dentry, struct dentry *target,
2783 bool exchange)
2784{
2785 struct dentry *old_parent, *p;
2786 wait_queue_head_t *d_wait;
2787 struct inode *dir = NULL;
2788 unsigned n;
2789
2790 WARN_ON(!dentry->d_inode);
2791 if (WARN_ON(dentry == target))
2792 return;
2793
2794 BUG_ON(d_ancestor(target, dentry));
2795 old_parent = dentry->d_parent;
2796 p = d_ancestor(old_parent, target);
2797 if (IS_ROOT(dentry)) {
2798 BUG_ON(p);
2799 spin_lock(&target->d_parent->d_lock);
2800 } else if (!p) {
2801 /* target is not a descendent of dentry->d_parent */
2802 spin_lock(&target->d_parent->d_lock);
2803 spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
2804 } else {
2805 BUG_ON(p == dentry);
2806 spin_lock(&old_parent->d_lock);
2807 if (p != target)
2808 spin_lock_nested(&target->d_parent->d_lock,
2809 DENTRY_D_LOCK_NESTED);
2810 }
2811 spin_lock_nested(&dentry->d_lock, 2);
2812 spin_lock_nested(&target->d_lock, 3);
2813
2814 if (unlikely(d_in_lookup(target))) {
2815 dir = target->d_parent->d_inode;
2816 n = start_dir_add(dir);
2817 d_wait = __d_lookup_unhash(target);
2818 }
2819
2820 write_seqcount_begin(&dentry->d_seq);
2821 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2822
2823 /* unhash both */
2824 if (!d_unhashed(dentry))
2825 ___d_drop(dentry);
2826 if (!d_unhashed(target))
2827 ___d_drop(target);
2828
2829 /* ... and switch them in the tree */
2830 dentry->d_parent = target->d_parent;
2831 if (!exchange) {
2832 copy_name(dentry, target);
2833 target->d_hash.pprev = NULL;
2834 dentry->d_parent->d_lockref.count++;
2835 if (dentry != old_parent) /* wasn't IS_ROOT */
2836 WARN_ON(!--old_parent->d_lockref.count);
2837 } else {
2838 target->d_parent = old_parent;
2839 swap_names(dentry, target);
2840 if (!hlist_unhashed(&target->d_sib))
2841 __hlist_del(&target->d_sib);
2842 hlist_add_head(&target->d_sib, &target->d_parent->d_children);
2843 __d_rehash(target);
2844 fsnotify_update_flags(target);
2845 }
2846 if (!hlist_unhashed(&dentry->d_sib))
2847 __hlist_del(&dentry->d_sib);
2848 hlist_add_head(&dentry->d_sib, &dentry->d_parent->d_children);
2849 __d_rehash(dentry);
2850 fsnotify_update_flags(dentry);
2851 fscrypt_handle_d_move(dentry);
2852
2853 write_seqcount_end(&target->d_seq);
2854 write_seqcount_end(&dentry->d_seq);
2855
2856 if (dir)
2857 end_dir_add(dir, n, d_wait);
2858
2859 if (dentry->d_parent != old_parent)
2860 spin_unlock(&dentry->d_parent->d_lock);
2861 if (dentry != old_parent)
2862 spin_unlock(&old_parent->d_lock);
2863 spin_unlock(&target->d_lock);
2864 spin_unlock(&dentry->d_lock);
2865}
2866
2867/*
2868 * d_move - move a dentry
2869 * @dentry: entry to move
2870 * @target: new dentry
2871 *
2872 * Update the dcache to reflect the move of a file name. Negative
2873 * dcache entries should not be moved in this way. See the locking
2874 * requirements for __d_move.
2875 */
2876void d_move(struct dentry *dentry, struct dentry *target)
2877{
2878 write_seqlock(&rename_lock);
2879 __d_move(dentry, target, false);
2880 write_sequnlock(&rename_lock);
2881}
2882EXPORT_SYMBOL(d_move);
2883
2884/*
2885 * d_exchange - exchange two dentries
2886 * @dentry1: first dentry
2887 * @dentry2: second dentry
2888 */
2889void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2890{
2891 write_seqlock(&rename_lock);
2892
2893 WARN_ON(!dentry1->d_inode);
2894 WARN_ON(!dentry2->d_inode);
2895 WARN_ON(IS_ROOT(dentry1));
2896 WARN_ON(IS_ROOT(dentry2));
2897
2898 __d_move(dentry1, dentry2, true);
2899
2900 write_sequnlock(&rename_lock);
2901}
2902
2903/**
2904 * d_ancestor - search for an ancestor
2905 * @p1: ancestor dentry
2906 * @p2: child dentry
2907 *
2908 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2909 * an ancestor of p2, else NULL.
2910 */
2911struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2912{
2913 struct dentry *p;
2914
2915 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2916 if (p->d_parent == p1)
2917 return p;
2918 }
2919 return NULL;
2920}
2921
2922/*
2923 * This helper attempts to cope with remotely renamed directories
2924 *
2925 * It assumes that the caller is already holding
2926 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2927 *
2928 * Note: If ever the locking in lock_rename() changes, then please
2929 * remember to update this too...
2930 */
2931static int __d_unalias(struct dentry *dentry, struct dentry *alias)
2932{
2933 struct mutex *m1 = NULL;
2934 struct rw_semaphore *m2 = NULL;
2935 int ret = -ESTALE;
2936
2937 /* If alias and dentry share a parent, then no extra locks required */
2938 if (alias->d_parent == dentry->d_parent)
2939 goto out_unalias;
2940
2941 /* See lock_rename() */
2942 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2943 goto out_err;
2944 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2945 if (!inode_trylock_shared(alias->d_parent->d_inode))
2946 goto out_err;
2947 m2 = &alias->d_parent->d_inode->i_rwsem;
2948out_unalias:
2949 if (alias->d_op && alias->d_op->d_unalias_trylock &&
2950 !alias->d_op->d_unalias_trylock(alias))
2951 goto out_err;
2952 __d_move(alias, dentry, false);
2953 if (alias->d_op && alias->d_op->d_unalias_unlock)
2954 alias->d_op->d_unalias_unlock(alias);
2955 ret = 0;
2956out_err:
2957 if (m2)
2958 up_read(m2);
2959 if (m1)
2960 mutex_unlock(m1);
2961 return ret;
2962}
2963
2964/**
2965 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2966 * @inode: the inode which may have a disconnected dentry
2967 * @dentry: a negative dentry which we want to point to the inode.
2968 *
2969 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2970 * place of the given dentry and return it, else simply d_add the inode
2971 * to the dentry and return NULL.
2972 *
2973 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2974 * we should error out: directories can't have multiple aliases.
2975 *
2976 * This is needed in the lookup routine of any filesystem that is exportable
2977 * (via knfsd) so that we can build dcache paths to directories effectively.
2978 *
2979 * If a dentry was found and moved, then it is returned. Otherwise NULL
2980 * is returned. This matches the expected return value of ->lookup.
2981 *
2982 * Cluster filesystems may call this function with a negative, hashed dentry.
2983 * In that case, we know that the inode will be a regular file, and also this
2984 * will only occur during atomic_open. So we need to check for the dentry
2985 * being already hashed only in the final case.
2986 */
2987struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2988{
2989 if (IS_ERR(inode))
2990 return ERR_CAST(inode);
2991
2992 BUG_ON(!d_unhashed(dentry));
2993
2994 if (!inode)
2995 goto out;
2996
2997 security_d_instantiate(dentry, inode);
2998 spin_lock(&inode->i_lock);
2999 if (S_ISDIR(inode->i_mode)) {
3000 struct dentry *new = __d_find_any_alias(inode);
3001 if (unlikely(new)) {
3002 /* The reference to new ensures it remains an alias */
3003 spin_unlock(&inode->i_lock);
3004 write_seqlock(&rename_lock);
3005 if (unlikely(d_ancestor(new, dentry))) {
3006 write_sequnlock(&rename_lock);
3007 dput(new);
3008 new = ERR_PTR(-ELOOP);
3009 pr_warn_ratelimited(
3010 "VFS: Lookup of '%s' in %s %s"
3011 " would have caused loop\n",
3012 dentry->d_name.name,
3013 inode->i_sb->s_type->name,
3014 inode->i_sb->s_id);
3015 } else if (!IS_ROOT(new)) {
3016 struct dentry *old_parent = dget(new->d_parent);
3017 int err = __d_unalias(dentry, new);
3018 write_sequnlock(&rename_lock);
3019 if (err) {
3020 dput(new);
3021 new = ERR_PTR(err);
3022 }
3023 dput(old_parent);
3024 } else {
3025 __d_move(new, dentry, false);
3026 write_sequnlock(&rename_lock);
3027 }
3028 iput(inode);
3029 return new;
3030 }
3031 }
3032out:
3033 __d_add(dentry, inode);
3034 return NULL;
3035}
3036EXPORT_SYMBOL(d_splice_alias);
3037
3038/*
3039 * Test whether new_dentry is a subdirectory of old_dentry.
3040 *
3041 * Trivially implemented using the dcache structure
3042 */
3043
3044/**
3045 * is_subdir - is new dentry a subdirectory of old_dentry
3046 * @new_dentry: new dentry
3047 * @old_dentry: old dentry
3048 *
3049 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3050 * Returns false otherwise.
3051 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3052 */
3053
3054bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3055{
3056 bool subdir;
3057 unsigned seq;
3058
3059 if (new_dentry == old_dentry)
3060 return true;
3061
3062 /* Access d_parent under rcu as d_move() may change it. */
3063 rcu_read_lock();
3064 seq = read_seqbegin(&rename_lock);
3065 subdir = d_ancestor(old_dentry, new_dentry);
3066 /* Try lockless once... */
3067 if (read_seqretry(&rename_lock, seq)) {
3068 /* ...else acquire lock for progress even on deep chains. */
3069 read_seqlock_excl(&rename_lock);
3070 subdir = d_ancestor(old_dentry, new_dentry);
3071 read_sequnlock_excl(&rename_lock);
3072 }
3073 rcu_read_unlock();
3074 return subdir;
3075}
3076EXPORT_SYMBOL(is_subdir);
3077
3078static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3079{
3080 struct dentry *root = data;
3081 if (dentry != root) {
3082 if (d_unhashed(dentry) || !dentry->d_inode)
3083 return D_WALK_SKIP;
3084
3085 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3086 dentry->d_flags |= DCACHE_GENOCIDE;
3087 dentry->d_lockref.count--;
3088 }
3089 }
3090 return D_WALK_CONTINUE;
3091}
3092
3093void d_genocide(struct dentry *parent)
3094{
3095 d_walk(parent, parent, d_genocide_kill);
3096}
3097
3098void d_mark_tmpfile(struct file *file, struct inode *inode)
3099{
3100 struct dentry *dentry = file->f_path.dentry;
3101
3102 BUG_ON(dname_external(dentry) ||
3103 !hlist_unhashed(&dentry->d_u.d_alias) ||
3104 !d_unlinked(dentry));
3105 spin_lock(&dentry->d_parent->d_lock);
3106 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3107 dentry->d_name.len = sprintf(dentry->d_shortname.string, "#%llu",
3108 (unsigned long long)inode->i_ino);
3109 spin_unlock(&dentry->d_lock);
3110 spin_unlock(&dentry->d_parent->d_lock);
3111}
3112EXPORT_SYMBOL(d_mark_tmpfile);
3113
3114void d_tmpfile(struct file *file, struct inode *inode)
3115{
3116 struct dentry *dentry = file->f_path.dentry;
3117
3118 inode_dec_link_count(inode);
3119 d_mark_tmpfile(file, inode);
3120 d_instantiate(dentry, inode);
3121}
3122EXPORT_SYMBOL(d_tmpfile);
3123
3124/*
3125 * Obtain inode number of the parent dentry.
3126 */
3127ino_t d_parent_ino(struct dentry *dentry)
3128{
3129 struct dentry *parent;
3130 struct inode *iparent;
3131 unsigned seq;
3132 ino_t ret;
3133
3134 scoped_guard(rcu) {
3135 seq = raw_seqcount_begin(&dentry->d_seq);
3136 parent = READ_ONCE(dentry->d_parent);
3137 iparent = d_inode_rcu(parent);
3138 if (likely(iparent)) {
3139 ret = iparent->i_ino;
3140 if (!read_seqcount_retry(&dentry->d_seq, seq))
3141 return ret;
3142 }
3143 }
3144
3145 spin_lock(&dentry->d_lock);
3146 ret = dentry->d_parent->d_inode->i_ino;
3147 spin_unlock(&dentry->d_lock);
3148 return ret;
3149}
3150EXPORT_SYMBOL(d_parent_ino);
3151
3152static __initdata unsigned long dhash_entries;
3153static int __init set_dhash_entries(char *str)
3154{
3155 if (!str)
3156 return 0;
3157 dhash_entries = simple_strtoul(str, &str, 0);
3158 return 1;
3159}
3160__setup("dhash_entries=", set_dhash_entries);
3161
3162static void __init dcache_init_early(void)
3163{
3164 /* If hashes are distributed across NUMA nodes, defer
3165 * hash allocation until vmalloc space is available.
3166 */
3167 if (hashdist)
3168 return;
3169
3170 dentry_hashtable =
3171 alloc_large_system_hash("Dentry cache",
3172 sizeof(struct hlist_bl_head),
3173 dhash_entries,
3174 13,
3175 HASH_EARLY | HASH_ZERO,
3176 &d_hash_shift,
3177 NULL,
3178 0,
3179 0);
3180 d_hash_shift = 32 - d_hash_shift;
3181
3182 runtime_const_init(shift, d_hash_shift);
3183 runtime_const_init(ptr, dentry_hashtable);
3184}
3185
3186static void __init dcache_init(void)
3187{
3188 /*
3189 * A constructor could be added for stable state like the lists,
3190 * but it is probably not worth it because of the cache nature
3191 * of the dcache.
3192 */
3193 dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3194 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_ACCOUNT,
3195 d_shortname.string);
3196
3197 /* Hash may have been set up in dcache_init_early */
3198 if (!hashdist)
3199 return;
3200
3201 dentry_hashtable =
3202 alloc_large_system_hash("Dentry cache",
3203 sizeof(struct hlist_bl_head),
3204 dhash_entries,
3205 13,
3206 HASH_ZERO,
3207 &d_hash_shift,
3208 NULL,
3209 0,
3210 0);
3211 d_hash_shift = 32 - d_hash_shift;
3212
3213 runtime_const_init(shift, d_hash_shift);
3214 runtime_const_init(ptr, dentry_hashtable);
3215}
3216
3217/* SLAB cache for __getname() consumers */
3218struct kmem_cache *names_cachep __ro_after_init;
3219EXPORT_SYMBOL(names_cachep);
3220
3221void __init vfs_caches_init_early(void)
3222{
3223 int i;
3224
3225 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3226 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3227
3228 dcache_init_early();
3229 inode_init_early();
3230}
3231
3232void __init vfs_caches_init(void)
3233{
3234 names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3235 SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
3236
3237 dcache_init();
3238 inode_init();
3239 files_init();
3240 files_maxfiles_init();
3241 mnt_init();
3242 bdev_cache_init();
3243 chrdev_init();
3244}