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_alias, d_inode of aliases
44 * dcache_hash_bucket lock protects:
45 * - the dcache hash table
46 * s_roots_lock protects:
47 * - the s_roots list (see __d_move()/dentry_unlist()/d_obtain_root())
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_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 */
76 static int sysctl_vfs_cache_pressure __read_mostly = 100;
77 static int sysctl_vfs_cache_pressure_denom __read_mostly = 100;
78
vfs_pressure_ratio(unsigned long val)79 unsigned long vfs_pressure_ratio(unsigned long val)
80 {
81 return mult_frac(val, sysctl_vfs_cache_pressure, sysctl_vfs_cache_pressure_denom);
82 }
83 EXPORT_SYMBOL_GPL(vfs_pressure_ratio);
84
85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
86
87 EXPORT_SYMBOL(rename_lock);
88
89 static struct kmem_cache *__dentry_cache __ro_after_init;
90 #define dentry_cache runtime_const_ptr(__dentry_cache)
91
92 const struct qstr empty_name = QSTR_INIT("", 0);
93 EXPORT_SYMBOL(empty_name);
94 const struct qstr slash_name = QSTR_INIT("/", 1);
95 EXPORT_SYMBOL(slash_name);
96 const struct qstr dotdot_name = QSTR_INIT("..", 2);
97 EXPORT_SYMBOL(dotdot_name);
98
99 /*
100 * This is the single most critical data structure when it comes
101 * to the dcache: the hashtable for lookups. Somebody should try
102 * to make this good - I've just made it work.
103 *
104 * This hash-function tries to avoid losing too many bits of hash
105 * information, yet avoid using a prime hash-size or similar.
106 *
107 * Marking the variables "used" ensures that the compiler doesn't
108 * optimize them away completely on architectures with runtime
109 * constant infrastructure, this allows debuggers to see their
110 * values. But updating these values has no effect on those arches.
111 */
112
113 static unsigned int d_hash_shift __ro_after_init __used;
114
115 static struct hlist_bl_head *dentry_hashtable __ro_after_init __used;
116
d_hash(unsigned long hashlen)117 static inline struct hlist_bl_head *d_hash(unsigned long hashlen)
118 {
119 return runtime_const_ptr(dentry_hashtable) +
120 runtime_const_shift_right_32(hashlen, d_hash_shift);
121 }
122
123 #define IN_LOOKUP_SHIFT 10
124 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
125
in_lookup_hash(const struct dentry * parent,unsigned int hash)126 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
127 unsigned int hash)
128 {
129 hash += (unsigned long) parent / L1_CACHE_BYTES;
130 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
131 }
132
133 struct dentry_stat_t {
134 long nr_dentry;
135 long nr_unused;
136 long age_limit; /* age in seconds */
137 long want_pages; /* pages requested by system */
138 long nr_negative; /* # of unused negative dentries */
139 long dummy; /* Reserved for future use */
140 };
141
142 static DEFINE_PER_CPU(long, nr_dentry);
143 static DEFINE_PER_CPU(long, nr_dentry_unused);
144 static DEFINE_PER_CPU(long, nr_dentry_negative);
145 static int dentry_negative_policy;
146
147 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
148 /* Statistics gathering. */
149 static struct dentry_stat_t dentry_stat = {
150 .age_limit = 45,
151 };
152
153 /*
154 * Here we resort to our own counters instead of using generic per-cpu counters
155 * for consistency with what the vfs inode code does. We are expected to harvest
156 * better code and performance by having our own specialized counters.
157 *
158 * Please note that the loop is done over all possible CPUs, not over all online
159 * CPUs. The reason for this is that we don't want to play games with CPUs going
160 * on and off. If one of them goes off, we will just keep their counters.
161 *
162 * glommer: See cffbc8a for details, and if you ever intend to change this,
163 * please update all vfs counters to match.
164 */
get_nr_dentry(void)165 static long get_nr_dentry(void)
166 {
167 int i;
168 long sum = 0;
169 for_each_possible_cpu(i)
170 sum += per_cpu(nr_dentry, i);
171 return sum < 0 ? 0 : sum;
172 }
173
get_nr_dentry_unused(void)174 static long get_nr_dentry_unused(void)
175 {
176 int i;
177 long sum = 0;
178 for_each_possible_cpu(i)
179 sum += per_cpu(nr_dentry_unused, i);
180 return sum < 0 ? 0 : sum;
181 }
182
get_nr_dentry_negative(void)183 static long get_nr_dentry_negative(void)
184 {
185 int i;
186 long sum = 0;
187
188 for_each_possible_cpu(i)
189 sum += per_cpu(nr_dentry_negative, i);
190 return sum < 0 ? 0 : sum;
191 }
192
proc_nr_dentry(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)193 static int proc_nr_dentry(const struct ctl_table *table, int write, void *buffer,
194 size_t *lenp, loff_t *ppos)
195 {
196 dentry_stat.nr_dentry = get_nr_dentry();
197 dentry_stat.nr_unused = get_nr_dentry_unused();
198 dentry_stat.nr_negative = get_nr_dentry_negative();
199 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
200 }
201
202 static const struct ctl_table fs_dcache_sysctls[] = {
203 {
204 .procname = "dentry-state",
205 .data = &dentry_stat,
206 .maxlen = 6*sizeof(long),
207 .mode = 0444,
208 .proc_handler = proc_nr_dentry,
209 },
210 {
211 .procname = "dentry-negative",
212 .data = &dentry_negative_policy,
213 .maxlen = sizeof(dentry_negative_policy),
214 .mode = 0644,
215 .proc_handler = proc_dointvec_minmax,
216 .extra1 = SYSCTL_ZERO,
217 .extra2 = SYSCTL_ONE,
218 },
219 };
220
221 static const struct ctl_table vm_dcache_sysctls[] = {
222 {
223 .procname = "vfs_cache_pressure",
224 .data = &sysctl_vfs_cache_pressure,
225 .maxlen = sizeof(sysctl_vfs_cache_pressure),
226 .mode = 0644,
227 .proc_handler = proc_dointvec_minmax,
228 .extra1 = SYSCTL_ZERO,
229 },
230 {
231 .procname = "vfs_cache_pressure_denom",
232 .data = &sysctl_vfs_cache_pressure_denom,
233 .maxlen = sizeof(sysctl_vfs_cache_pressure_denom),
234 .mode = 0644,
235 .proc_handler = proc_dointvec_minmax,
236 .extra1 = SYSCTL_ONE_HUNDRED,
237 },
238 };
239
init_fs_dcache_sysctls(void)240 static int __init init_fs_dcache_sysctls(void)
241 {
242 register_sysctl_init("vm", vm_dcache_sysctls);
243 register_sysctl_init("fs", fs_dcache_sysctls);
244 return 0;
245 }
246 fs_initcall(init_fs_dcache_sysctls);
247 #endif
248
249 /*
250 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
251 * The strings are both count bytes long, and count is non-zero.
252 */
253 #ifdef CONFIG_DCACHE_WORD_ACCESS
254
255 #include <asm/word-at-a-time.h>
256 /*
257 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
258 * aligned allocation for this particular component. We don't
259 * strictly need the load_unaligned_zeropad() safety, but it
260 * doesn't hurt either.
261 *
262 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
263 * need the careful unaligned handling.
264 */
dentry_string_cmp(const unsigned char * cs,const unsigned char * ct,unsigned tcount)265 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
266 {
267 unsigned long a,b,mask;
268
269 for (;;) {
270 a = read_word_at_a_time(cs);
271 b = load_unaligned_zeropad(ct);
272 if (tcount < sizeof(unsigned long))
273 break;
274 if (unlikely(a != b))
275 return 1;
276 cs += sizeof(unsigned long);
277 ct += sizeof(unsigned long);
278 tcount -= sizeof(unsigned long);
279 if (!tcount)
280 return 0;
281 }
282 mask = bytemask_from_count(tcount);
283 return unlikely(!!((a ^ b) & mask));
284 }
285
286 #else
287
dentry_string_cmp(const unsigned char * cs,const unsigned char * ct,unsigned tcount)288 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
289 {
290 do {
291 if (*cs != *ct)
292 return 1;
293 cs++;
294 ct++;
295 tcount--;
296 } while (tcount);
297 return 0;
298 }
299
300 #endif
301
dentry_cmp(const struct dentry * dentry,const unsigned char * ct,unsigned tcount)302 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
303 {
304 /*
305 * Be careful about RCU walk racing with rename:
306 * use 'READ_ONCE' to fetch the name pointer.
307 *
308 * NOTE! Even if a rename will mean that the length
309 * was not loaded atomically, we don't care. The
310 * RCU walk will check the sequence count eventually,
311 * and catch it. And we won't overrun the buffer,
312 * because we're reading the name pointer atomically,
313 * and a dentry name is guaranteed to be properly
314 * terminated with a NUL byte.
315 *
316 * End result: even if 'len' is wrong, we'll exit
317 * early because the data cannot match (there can
318 * be no NUL in the ct/tcount data)
319 */
320 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
321
322 return dentry_string_cmp(cs, ct, tcount);
323 }
324
325 /*
326 * long names are allocated separately from dentry and never modified.
327 * Refcounted, freeing is RCU-delayed. See take_dentry_name_snapshot()
328 * for the reason why ->count and ->head can't be combined into a union.
329 * dentry_string_cmp() relies upon ->name[] being word-aligned.
330 */
331 struct external_name {
332 atomic_t count;
333 struct rcu_head head;
334 unsigned char name[] __aligned(sizeof(unsigned long));
335 };
336
external_name(struct dentry * dentry)337 static inline struct external_name *external_name(struct dentry *dentry)
338 {
339 return container_of(dentry->d_name.name, struct external_name, name[0]);
340 }
341
__d_free(struct rcu_head * head)342 static void __d_free(struct rcu_head *head)
343 {
344 struct dentry *dentry = container_of(head, struct dentry, d_rcu);
345
346 kmem_cache_free(dentry_cache, dentry);
347 }
348
__d_free_external(struct rcu_head * head)349 static void __d_free_external(struct rcu_head *head)
350 {
351 struct dentry *dentry = container_of(head, struct dentry, d_rcu);
352 kfree(external_name(dentry));
353 kmem_cache_free(dentry_cache, dentry);
354 }
355
dname_external(const struct dentry * dentry)356 static inline int dname_external(const struct dentry *dentry)
357 {
358 return dentry->d_name.name != dentry->d_shortname.string;
359 }
360
take_dentry_name_snapshot(struct name_snapshot * name,struct dentry * dentry)361 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
362 {
363 unsigned seq;
364 const unsigned char *s;
365
366 rcu_read_lock();
367 retry:
368 seq = read_seqcount_begin(&dentry->d_seq);
369 s = READ_ONCE(dentry->d_name.name);
370 name->name.hash_len = dentry->d_name.hash_len;
371 name->name.name = name->inline_name.string;
372 if (likely(s == dentry->d_shortname.string)) {
373 name->inline_name = dentry->d_shortname;
374 } else {
375 struct external_name *p;
376 p = container_of(s, struct external_name, name[0]);
377 // get a valid reference
378 if (unlikely(!atomic_inc_not_zero(&p->count)))
379 goto retry;
380 name->name.name = s;
381 }
382 if (read_seqcount_retry(&dentry->d_seq, seq)) {
383 release_dentry_name_snapshot(name);
384 goto retry;
385 }
386 rcu_read_unlock();
387 }
388 EXPORT_SYMBOL(take_dentry_name_snapshot);
389
release_dentry_name_snapshot(struct name_snapshot * name)390 void release_dentry_name_snapshot(struct name_snapshot *name)
391 {
392 if (unlikely(name->name.name != name->inline_name.string)) {
393 struct external_name *p;
394 p = container_of(name->name.name, struct external_name, name[0]);
395 if (unlikely(atomic_dec_and_test(&p->count)))
396 kfree_rcu(p, head);
397 }
398 }
399 EXPORT_SYMBOL(release_dentry_name_snapshot);
400
__d_set_inode_and_type(struct dentry * dentry,struct inode * inode,unsigned type_flags)401 static inline void __d_set_inode_and_type(struct dentry *dentry,
402 struct inode *inode,
403 unsigned type_flags)
404 {
405 unsigned flags;
406
407 dentry->d_inode = inode;
408 flags = READ_ONCE(dentry->d_flags);
409 flags &= ~DCACHE_ENTRY_TYPE;
410 flags |= type_flags;
411 smp_store_release(&dentry->d_flags, flags);
412 }
413
__d_clear_type_and_inode(struct dentry * dentry)414 static inline void __d_clear_type_and_inode(struct dentry *dentry)
415 {
416 unsigned flags = READ_ONCE(dentry->d_flags);
417
418 flags &= ~DCACHE_ENTRY_TYPE;
419 WRITE_ONCE(dentry->d_flags, flags);
420 dentry->d_inode = NULL;
421 /*
422 * The negative counter only tracks dentries on the LRU. Don't inc if
423 * d_lru is on another list.
424 */
425 if ((flags & (DCACHE_LRU_LIST|DCACHE_SHRINK_LIST)) == DCACHE_LRU_LIST)
426 this_cpu_inc(nr_dentry_negative);
427 }
428
429 #define DENTRY_WARN_ONCE(condition, dentry) \
430 WARN_ONCE((condition), "dentry=%p d_flags=0x%x\n", (dentry), (dentry)->d_flags)
431 #define D_FLAG_VERIFY(dentry, x) \
432 DENTRY_WARN_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x), (dentry))
433
dentry_free(struct dentry * dentry)434 static void dentry_free(struct dentry *dentry)
435 {
436 DENTRY_WARN_ONCE(d_really_is_positive(dentry), dentry);
437 DENTRY_WARN_ONCE(!lockref_is_dead(&dentry->d_lockref), dentry);
438 D_FLAG_VERIFY(dentry, 0);
439 if (unlikely(dname_external(dentry))) {
440 struct external_name *p = external_name(dentry);
441 if (likely(atomic_dec_and_test(&p->count))) {
442 call_rcu(&dentry->d_rcu, __d_free_external);
443 return;
444 }
445 }
446 /* if dentry was never visible to RCU, immediate free is OK */
447 if (dentry->d_flags & DCACHE_NORCU)
448 __d_free(&dentry->d_rcu);
449 else
450 call_rcu(&dentry->d_rcu, __d_free);
451 }
452
453 /*
454 * Release the dentry's inode, using the filesystem
455 * d_iput() operation if defined.
456 */
dentry_unlink_inode(struct dentry * dentry)457 static void dentry_unlink_inode(struct dentry * dentry)
458 __releases(dentry->d_lock)
459 __releases(dentry->d_inode->i_lock)
460 {
461 struct inode *inode = dentry->d_inode;
462
463 raw_write_seqcount_begin(&dentry->d_seq);
464 __d_clear_type_and_inode(dentry);
465 __hlist_del(&dentry->d_alias);
466 /*
467 * dentry becomes negative, so the space occupied by ->d_alias
468 * belongs to ->waiters now.
469 */
470 dentry->waiters = NULL;
471 raw_write_seqcount_end(&dentry->d_seq);
472 spin_unlock(&dentry->d_lock);
473 spin_unlock(&inode->i_lock);
474 if (!inode->i_nlink)
475 fsnotify_inoderemove(inode);
476 if (dentry->d_op && dentry->d_op->d_iput)
477 dentry->d_op->d_iput(dentry, inode);
478 else
479 iput(inode);
480 }
481
482 /*
483 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
484 * is in use - which includes both the "real" per-superblock
485 * LRU list _and_ the DCACHE_SHRINK_LIST use.
486 *
487 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
488 * on the shrink list (ie not on the superblock LRU list).
489 *
490 * The per-cpu "nr_dentry_unused" counters are updated with
491 * the DCACHE_LRU_LIST bit.
492 *
493 * The per-cpu "nr_dentry_negative" counters are only updated
494 * when deleted from or added to the per-superblock LRU list, not
495 * from/to the shrink list. That is to avoid an unneeded dec/inc
496 * pair when moving from LRU to shrink list in select_collect().
497 *
498 * These helper functions make sure we always follow the
499 * rules. d_lock must be held by the caller.
500 */
d_lru_add(struct dentry * dentry)501 static void d_lru_add(struct dentry *dentry)
502 {
503 D_FLAG_VERIFY(dentry, 0);
504 dentry->d_flags |= DCACHE_LRU_LIST;
505 this_cpu_inc(nr_dentry_unused);
506 if (d_is_negative(dentry))
507 this_cpu_inc(nr_dentry_negative);
508 WARN_ON_ONCE(!list_lru_add_obj(
509 &dentry->d_sb->s_dentry_lru, &dentry->d_lru));
510 }
511
d_lru_del(struct dentry * dentry)512 static void d_lru_del(struct dentry *dentry)
513 {
514 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
515 dentry->d_flags &= ~DCACHE_LRU_LIST;
516 this_cpu_dec(nr_dentry_unused);
517 if (d_is_negative(dentry))
518 this_cpu_dec(nr_dentry_negative);
519 WARN_ON_ONCE(!list_lru_del_obj(
520 &dentry->d_sb->s_dentry_lru, &dentry->d_lru));
521 }
522
d_shrink_del(struct dentry * dentry)523 static void d_shrink_del(struct dentry *dentry)
524 {
525 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
526 list_del_init(&dentry->d_lru);
527 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
528 this_cpu_dec(nr_dentry_unused);
529 }
530
d_shrink_add(struct dentry * dentry,struct list_head * list)531 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
532 {
533 D_FLAG_VERIFY(dentry, 0);
534 list_add(&dentry->d_lru, list);
535 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
536 this_cpu_inc(nr_dentry_unused);
537 }
538
539 /*
540 * These can only be called under the global LRU lock, ie during the
541 * callback for freeing the LRU list. "isolate" removes it from the
542 * LRU lists entirely, while shrink_move moves it to the indicated
543 * private list.
544 */
d_lru_isolate(struct list_lru_one * lru,struct dentry * dentry)545 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
546 {
547 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
548 dentry->d_flags &= ~DCACHE_LRU_LIST;
549 this_cpu_dec(nr_dentry_unused);
550 if (d_is_negative(dentry))
551 this_cpu_dec(nr_dentry_negative);
552 list_lru_isolate(lru, &dentry->d_lru);
553 }
554
d_lru_shrink_move(struct list_lru_one * lru,struct dentry * dentry,struct list_head * list)555 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
556 struct list_head *list)
557 {
558 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
559 dentry->d_flags |= DCACHE_SHRINK_LIST;
560 if (d_is_negative(dentry))
561 this_cpu_dec(nr_dentry_negative);
562 list_lru_isolate_move(lru, &dentry->d_lru, list);
563 }
564
___d_drop(struct dentry * dentry)565 static void ___d_drop(struct dentry *dentry)
566 {
567 struct hlist_bl_head *b = d_hash(dentry->d_name.hash);
568
569 hlist_bl_lock(b);
570 __hlist_bl_del(&dentry->d_hash);
571 hlist_bl_unlock(b);
572 }
573
__d_drop(struct dentry * dentry)574 void __d_drop(struct dentry *dentry)
575 {
576 if (!d_unhashed(dentry)) {
577 ___d_drop(dentry);
578 dentry->d_hash.pprev = NULL;
579 write_seqcount_invalidate(&dentry->d_seq);
580 }
581 }
582 EXPORT_SYMBOL(__d_drop);
583
584 /**
585 * d_drop - drop a dentry
586 * @dentry: dentry to drop
587 *
588 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
589 * be found through a VFS lookup any more. Note that this is different from
590 * deleting the dentry - d_delete will try to mark the dentry negative if
591 * possible, giving a successful _negative_ lookup, while d_drop will
592 * just make the cache lookup fail.
593 *
594 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
595 * reason (NFS timeouts or autofs deletes).
596 *
597 * __d_drop requires dentry->d_lock
598 *
599 * ___d_drop doesn't mark dentry as "unhashed"
600 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
601 */
d_drop(struct dentry * dentry)602 void d_drop(struct dentry *dentry)
603 {
604 spin_lock(&dentry->d_lock);
605 __d_drop(dentry);
606 spin_unlock(&dentry->d_lock);
607 }
608 EXPORT_SYMBOL(d_drop);
609
610 struct completion_list {
611 struct completion_list *next;
612 struct completion completion;
613 };
614
615 /*
616 * shrink_dcache_tree() needs to be notified when dentry in process of
617 * being evicted finally gets unlisted. Such dentries are
618 * already with negative ->d_count
619 * already negative
620 * already not in in-lookup hash
621 * reachable only via ->d_sib.
622 *
623 * Use ->waiters for a single-linked list of struct completion_list of
624 * waiters.
625 */
d_add_waiter(struct dentry * dentry,struct completion_list * p)626 static inline bool d_add_waiter(struct dentry *dentry, struct completion_list *p)
627 {
628 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED))
629 return false;
630 init_completion(&p->completion);
631 p->next = dentry->waiters;
632 dentry->waiters = p;
633 return true;
634 }
635
d_complete_waiters(struct dentry * dentry)636 static inline void d_complete_waiters(struct dentry *dentry)
637 {
638 struct completion_list *v = dentry->waiters;
639 if (unlikely(v)) {
640 /* some shrink_dcache_tree() instances are waiting */
641 dentry->waiters = NULL;
642 while (v) {
643 struct completion *r = &v->completion;
644 v = v->next;
645 complete(r);
646 }
647 }
648 }
649
unlink_secondary_root(struct dentry * dentry)650 static void unlink_secondary_root(struct dentry *dentry)
651 {
652 spin_lock(&dentry->d_sb->s_roots_lock);
653 hlist_del_init(&dentry->d_sib);
654 spin_unlock(&dentry->d_sb->s_roots_lock);
655 }
656
dentry_unlist(struct dentry * dentry)657 static inline void dentry_unlist(struct dentry *dentry)
658 {
659 struct dentry *next;
660 /*
661 * Inform d_walk() and shrink_dentry_list() that we are no longer
662 * attached to the dentry tree
663 */
664 dentry->d_flags |= DCACHE_DENTRY_KILLED;
665 d_complete_waiters(dentry);
666 if (unlikely(hlist_unhashed(&dentry->d_sib)))
667 return;
668 if (unlikely(IS_ROOT(dentry))) {
669 unlink_secondary_root(dentry); // secondary root goes away
670 return;
671 }
672 __hlist_del(&dentry->d_sib);
673 /*
674 * Cursors can move around the list of children. While we'd been
675 * a normal list member, it didn't matter - ->d_sib.next would've
676 * been updated. However, from now on it won't be and for the
677 * things like d_walk() it might end up with a nasty surprise.
678 * Normally d_walk() doesn't care about cursors moving around -
679 * ->d_lock on parent prevents that and since a cursor has no children
680 * of its own, we get through it without ever unlocking the parent.
681 * There is one exception, though - if we ascend from a child that
682 * gets killed as soon as we unlock it, the next sibling is found
683 * using the value left in its ->d_sib.next. And if _that_
684 * pointed to a cursor, and cursor got moved (e.g. by lseek())
685 * before d_walk() regains parent->d_lock, we'll end up skipping
686 * everything the cursor had been moved past.
687 *
688 * Solution: make sure that the pointer left behind in ->d_sib.next
689 * points to something that won't be moving around. I.e. skip the
690 * cursors.
691 */
692 while (dentry->d_sib.next) {
693 next = hlist_entry(dentry->d_sib.next, struct dentry, d_sib);
694 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
695 break;
696 dentry->d_sib.next = next->d_sib.next;
697 }
698 }
699
700 /*
701 * Prepare locking environment for killing a dentry.
702 * Called under dentry->d_lock. To proceed with eviction of a positive dentry
703 * we need to get ->i_lock of the inode of that dentry as well.
704 * However, ->i_lock nests outside of ->d_lock, so if trylock fails we might
705 * have to drop and regain the latter. Dentry state can change while its
706 * ->d_lock is not held - it might end up getting killed, becoming busy,
707 * negative, etc., so we need to be careful.
708 *
709 * For NORCU dentries memory safety relies upon having only one call of
710 * lock_for_kill() in the entire lifetime of dentry and dentry_free() being
711 * called only by the caller of lock_for_kill(). That this is NORCU-specific;
712 * the crucial part is that refcounts of NORCU dentries never grow once having
713 * dropped to zero.
714 *
715 * For normal dentries we can not assume that there won't be concurrent calls
716 * of dentry_free() - dentry might end up being evicted by another thread
717 * while we are dropping/retaking locks on the slow path. Memory safety is
718 * provided by keeping the RCU read-side critical area contiguous with
719 * an explicit rcu_read_lock() scope bridging over the break in spinlock scopes.
720 *
721 * If dentry is busy (or busy dying, or already dead), unlock dentry
722 * and return false. Otherwise, return true and have that dentry's
723 * inode (if any) locked in addition to dentry itself.
724 */
lock_for_kill(struct dentry * dentry)725 static bool lock_for_kill(struct dentry *dentry)
726 {
727 struct inode *inode = dentry->d_inode;
728
729 if (unlikely(dentry->d_lockref.count)) {
730 spin_unlock(&dentry->d_lock);
731 return false;
732 }
733
734 if (!inode || likely(spin_trylock(&inode->i_lock)))
735 return true;
736
737 // Too bad - we need to drop ->d_lock and take locks in correct order.
738 // To avoid breaking RCU read-side critical area when we drop ->d_lock,
739 // take an explicit rcu_read_lock() while we are switching locks.
740 rcu_read_lock();
741 do {
742 spin_unlock(&dentry->d_lock);
743 spin_lock(&inode->i_lock);
744 spin_lock(&dentry->d_lock);
745 // make sure we'd locked the right inode - ->d_inode might've
746 // changed while we were not holding ->d_lock
747 if (likely(inode == dentry->d_inode))
748 break;
749 spin_unlock(&inode->i_lock);
750 inode = dentry->d_inode;
751 } while (inode);
752 rcu_read_unlock();
753 if (likely(!dentry->d_lockref.count))
754 return true;
755 if (inode)
756 spin_unlock(&inode->i_lock);
757 spin_unlock(&dentry->d_lock);
758 return false;
759 }
760
761 /**
762 * dentry_kill - evict a dentry
763 * @dentry: dentry to be evicted
764 *
765 * All dentry evictions are done by this function. The reference we are
766 * passed does not contribute to the refcount; the caller had either
767 * already decremented the refcount or it had never held one in the
768 * first place. @dentry->d_lock is held by the caller and dropped
769 * by dentry_kill(@dentry).
770 *
771 * We are guaranteed that nobody had called dentry_free(@dentry)
772 * prior to the beginning of RCU read-side critical area we are in.
773 *
774 * Caller must not access @dentry after the call.
775 *
776 * If eviction of @dentry drops the last reference to its parent,
777 * the reference to parent is returned to caller. In that case
778 * it is guaranteed to satisfy the requirements for dentry_kill()
779 * argument - its ->d_lock is held and we are guaranteed that nobody
780 * had passed it to dentry_free() prior to acquisition of its ->d_lock.
781 * Otherwise %NULL is returned.
782 *
783 * If @dentry is idle and remains such after we assemble the full
784 * locking environment for eviction (see lock_for_kill() for details)
785 * we mark it doomed (see lockref_mark_dead()) and proceed to detaching
786 * it from any filesystem objects. Otherwise we drop ->d_lock and
787 * return %NULL.
788 *
789 * Once @dentry is detached from the filesystem objects, we complete
790 * detaching it from dentry tree. The parent, if any, gets locked
791 * and its refcount is decremented; dentry is carefully removed from
792 * the tree (see dentry_unlist() for details) and marked killed
793 * (%DCACHE_DENTRY_KILLED set in ->d_flags). At that point it's just
794 * an inert chunk of memory, accessible only via RCU references
795 * and possibly via a shrink list. If it is not on any shrink lists,
796 * we call dentry_free(), which schedules actual freeing of memory.
797 * Othewise freeing is left to the owner of the shrink list in question.
798 */
dentry_kill(struct dentry * dentry)799 static struct dentry *dentry_kill(struct dentry *dentry)
800 {
801 struct dentry *parent = NULL;
802 bool can_free = true;
803
804 if (unlikely(!lock_for_kill(dentry)))
805 return NULL;
806
807 /*
808 * The dentry is now unrecoverably dead to the world.
809 */
810 lockref_mark_dead(&dentry->d_lockref);
811
812 /*
813 * inform the fs via d_prune that this dentry is about to be
814 * unhashed and destroyed.
815 */
816 if (dentry->d_flags & DCACHE_OP_PRUNE)
817 dentry->d_op->d_prune(dentry);
818
819 if (dentry->d_flags & DCACHE_LRU_LIST) {
820 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
821 d_lru_del(dentry);
822 }
823 /* if it was on the hash then remove it */
824 __d_drop(dentry);
825 if (dentry->d_inode)
826 dentry_unlink_inode(dentry);
827 else
828 spin_unlock(&dentry->d_lock);
829 this_cpu_dec(nr_dentry);
830 if (dentry->d_op && dentry->d_op->d_release)
831 dentry->d_op->d_release(dentry);
832
833 cond_resched();
834 /* now that it's negative, ->d_parent is stable */
835 if (!IS_ROOT(dentry)) {
836 parent = dentry->d_parent;
837 spin_lock(&parent->d_lock);
838 }
839 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
840 dentry_unlist(dentry);
841 if (dentry->d_flags & DCACHE_SHRINK_LIST)
842 can_free = false;
843 spin_unlock(&dentry->d_lock);
844 if (likely(can_free))
845 dentry_free(dentry);
846 if (parent && --parent->d_lockref.count) {
847 spin_unlock(&parent->d_lock);
848 return NULL;
849 }
850 return parent;
851 }
852
853 /*
854 * Decide if dentry is worth retaining. Usually this is called with dentry
855 * locked; if not locked, we are more limited and might not be able to tell
856 * without a lock. False in this case means "punt to locked path and recheck".
857 *
858 * In case we aren't locked, these predicates are not "stable". However, it is
859 * sufficient that at some point after we dropped the reference the dentry was
860 * hashed and the flags had the proper value. Other dentry users may have
861 * re-gotten a reference to the dentry and change that, but our work is done -
862 * we can leave the dentry around with a zero refcount.
863 */
retain_dentry(struct dentry * dentry,bool locked)864 static inline bool retain_dentry(struct dentry *dentry, bool locked)
865 {
866 unsigned int d_flags;
867
868 smp_rmb();
869 d_flags = READ_ONCE(dentry->d_flags);
870
871 // Unreachable? Nobody would be able to look it up, no point retaining
872 if (unlikely(d_unhashed(dentry)))
873 return false;
874
875 // Same if it's disconnected
876 if (unlikely(d_flags & DCACHE_DISCONNECTED))
877 return false;
878
879 // ->d_delete() might tell us not to bother, but that requires
880 // ->d_lock; can't decide without it
881 if (unlikely(d_flags & DCACHE_OP_DELETE)) {
882 if (!locked || dentry->d_op->d_delete(dentry))
883 return false;
884 }
885
886 // Explicitly told not to bother
887 if (unlikely(d_flags & DCACHE_DONTCACHE))
888 return false;
889
890 // At this point it looks like we ought to keep it. We also might
891 // need to do something - put it on LRU if it wasn't there already
892 // and mark it referenced if it was on LRU, but not marked yet.
893 // Unfortunately, both actions require ->d_lock, so in lockless
894 // case we'd have to punt rather than doing those.
895 if (unlikely(!(d_flags & DCACHE_LRU_LIST))) {
896 if (!locked)
897 return false;
898 d_lru_add(dentry);
899 } else if (unlikely(!(d_flags & DCACHE_REFERENCED))) {
900 if (!locked)
901 return false;
902 dentry->d_flags |= DCACHE_REFERENCED;
903 }
904 return true;
905 }
906
d_mark_dontcache(struct inode * inode)907 void d_mark_dontcache(struct inode *inode)
908 {
909 struct dentry *de;
910
911 spin_lock(&inode->i_lock);
912 for_each_alias(de, inode) {
913 spin_lock(&de->d_lock);
914 de->d_flags |= DCACHE_DONTCACHE;
915 spin_unlock(&de->d_lock);
916 }
917 inode_state_set(inode, I_DONTCACHE);
918 spin_unlock(&inode->i_lock);
919 }
920 EXPORT_SYMBOL(d_mark_dontcache);
921
922 /*
923 * Try to do a lockless dput(), and return whether that was successful.
924 *
925 * If unsuccessful, we return false, having already taken the dentry lock.
926 * In that case refcount is guaranteed to be zero and we have already
927 * decided that it's not worth keeping around.
928 */
fast_dput(struct dentry * dentry)929 static inline bool fast_dput(struct dentry *dentry)
930 {
931 int ret;
932
933 /*
934 * Try to decrement the lockref optimistically.
935 * RCU read lock held so that dentry is guaranteed to stay around
936 * even if the refcount goes down to zero.
937 */
938 rcu_read_lock();
939 ret = lockref_put_return(&dentry->d_lockref);
940
941 /*
942 * If the lockref_put_return() failed due to the lock being held
943 * by somebody else, the fast path has failed. We will need to
944 * get the lock, and then check the count again.
945 */
946 if (unlikely(ret < 0)) {
947 spin_lock(&dentry->d_lock);
948 rcu_read_unlock();
949 if (WARN_ON_ONCE(lockref_is_dead_or_zero(&dentry->d_lockref))) {
950 spin_unlock(&dentry->d_lock);
951 return true;
952 }
953 dentry->d_lockref.count--;
954 goto locked;
955 }
956
957 /*
958 * If we weren't the last ref, we're done.
959 */
960 if (ret) {
961 rcu_read_unlock();
962 return true;
963 }
964
965 /*
966 * Can we decide that decrement of refcount is all we needed without
967 * taking the lock? There's a very common case when it's all we need -
968 * dentry looks like it ought to be retained and there's nothing else
969 * to do.
970 */
971 if (retain_dentry(dentry, false)) {
972 rcu_read_unlock();
973 return true;
974 }
975
976 /*
977 * Either not worth retaining or we can't tell without the lock.
978 * Get the lock, then. We've already decremented the refcount to 0,
979 * but we'll need to re-check the situation after getting the lock.
980 */
981 spin_lock(&dentry->d_lock);
982 rcu_read_unlock();
983
984 /*
985 * Did somebody else grab a reference to it in the meantime, and
986 * we're no longer the last user after all? Alternatively, somebody
987 * else could have killed it and marked it dead. Either way, we
988 * don't need to do anything else.
989 */
990 locked:
991 if (dentry->d_lockref.count || retain_dentry(dentry, true)) {
992 spin_unlock(&dentry->d_lock);
993 return true;
994 }
995 return false;
996 }
997
finish_dput(struct dentry * dentry)998 static void finish_dput(struct dentry *dentry)
999 __releases(dentry->d_lock)
1000 {
1001 while ((dentry = dentry_kill(dentry)) != NULL) {
1002 if (retain_dentry(dentry, true)) {
1003 spin_unlock(&dentry->d_lock);
1004 return;
1005 }
1006 }
1007 }
1008
1009 /*
1010 * This is dput
1011 *
1012 * This is complicated by the fact that we do not want to put
1013 * dentries that are no longer on any hash chain on the unused
1014 * list: we'd much rather just get rid of them immediately.
1015 *
1016 * However, that implies that we have to traverse the dentry
1017 * tree upwards to the parents which might _also_ now be
1018 * scheduled for deletion (it may have been only waiting for
1019 * its last child to go away).
1020 *
1021 * This tail recursion is done by hand as we don't want to depend
1022 * on the compiler to always get this right (gcc generally doesn't).
1023 * Real recursion would eat up our stack space.
1024 */
1025
1026 /*
1027 * dput - release a dentry
1028 * @dentry: dentry to release
1029 *
1030 * Release a dentry. This will drop the usage count and if appropriate
1031 * call the dentry unlink method as well as removing it from the queues and
1032 * releasing its resources. If the parent dentries were scheduled for release
1033 * they too may now get deleted.
1034 */
dput(struct dentry * dentry)1035 void dput(struct dentry *dentry)
1036 {
1037 if (!dentry)
1038 return;
1039 might_sleep();
1040 if (likely(fast_dput(dentry)))
1041 return;
1042 finish_dput(dentry);
1043 }
1044 EXPORT_SYMBOL(dput);
1045
d_make_discardable(struct dentry * dentry)1046 void d_make_discardable(struct dentry *dentry)
1047 {
1048 spin_lock(&dentry->d_lock);
1049 WARN_ON(!(dentry->d_flags & DCACHE_PERSISTENT));
1050 dentry->d_flags &= ~DCACHE_PERSISTENT;
1051 dentry->d_lockref.count--;
1052 finish_dput(dentry);
1053 }
1054 EXPORT_SYMBOL(d_make_discardable);
1055
1056 /**
1057 * __move_to_shrink_list - try to place a dentry into a shrink list
1058 * @dentry: dentry to try putting into shrink list
1059 * @list: the list to put @dentry into.
1060 * Returns: true @dentry had been placed into @list, false otherwise
1061 *
1062 * If @dentry is idle and not already include into a shrink list, move
1063 * it into @list and return %true; otherwise do nothing and return %false.
1064 *
1065 * Caller must be holding @dentry->d_lock. There must have been no calls of
1066 * dentry_free(@dentry) prior to the beginning of the RCU read-side critical
1067 * area in which __move_to_shrink_list(@dentry, @list) is called.
1068 *
1069 * @list should be thread-private and eventually emptied by passing it to
1070 * shrink_dentry_list().
1071 */
1072
__move_to_shrink_list(struct dentry * dentry,struct list_head * list)1073 bool __move_to_shrink_list(struct dentry *dentry, struct list_head *list)
1074 __must_hold(&dentry->d_lock)
1075 {
1076 if (likely(!dentry->d_lockref.count &&
1077 !(dentry->d_flags & DCACHE_SHRINK_LIST))) {
1078 if (dentry->d_flags & DCACHE_LRU_LIST)
1079 d_lru_del(dentry);
1080 d_shrink_add(dentry, list);
1081 return true;
1082 }
1083 return false;
1084 }
1085 EXPORT_SYMBOL(__move_to_shrink_list);
1086
dput_to_list(struct dentry * dentry,struct list_head * list)1087 void dput_to_list(struct dentry *dentry, struct list_head *list)
1088 {
1089 if (likely(fast_dput(dentry)))
1090 return;
1091 __move_to_shrink_list(dentry, list);
1092 spin_unlock(&dentry->d_lock);
1093 }
1094
dget_parent(struct dentry * dentry)1095 struct dentry *dget_parent(struct dentry *dentry)
1096 {
1097 int gotref;
1098 struct dentry *ret;
1099 unsigned seq;
1100
1101 /*
1102 * Do optimistic parent lookup without any
1103 * locking.
1104 */
1105 rcu_read_lock();
1106 seq = raw_seqcount_begin(&dentry->d_seq);
1107 ret = READ_ONCE(dentry->d_parent);
1108 gotref = lockref_get_not_zero(&ret->d_lockref);
1109 rcu_read_unlock();
1110 if (likely(gotref)) {
1111 if (!read_seqcount_retry(&dentry->d_seq, seq))
1112 return ret;
1113 dput(ret);
1114 }
1115
1116 repeat:
1117 /*
1118 * Don't need rcu_dereference because we re-check it was correct under
1119 * the lock.
1120 */
1121 rcu_read_lock();
1122 ret = dentry->d_parent;
1123 spin_lock(&ret->d_lock);
1124 if (unlikely(ret != dentry->d_parent)) {
1125 spin_unlock(&ret->d_lock);
1126 rcu_read_unlock();
1127 goto repeat;
1128 }
1129 rcu_read_unlock();
1130 BUG_ON(!ret->d_lockref.count);
1131 ret->d_lockref.count++;
1132 spin_unlock(&ret->d_lock);
1133 return ret;
1134 }
1135 EXPORT_SYMBOL(dget_parent);
1136
1137 /*
1138 * inode is a directory, inode->i_lock is held by the caller
1139 */
__d_find_dir_alias(struct inode * inode)1140 static struct dentry * __d_find_dir_alias(struct inode *inode)
1141 {
1142 struct dentry *alias;
1143
1144 if (hlist_empty(&inode->i_dentry))
1145 return NULL;
1146 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1147 lockref_get(&alias->d_lockref);
1148 return alias;
1149 }
1150
__d_find_any_alias(struct inode * inode)1151 static struct dentry * __d_find_any_alias(struct inode *inode)
1152 {
1153 struct dentry *alias;
1154
1155 if (hlist_empty(&inode->i_dentry))
1156 return NULL;
1157 for_each_alias(alias, inode)
1158 if (dget_alias_ilocked(alias))
1159 return alias;
1160 return NULL;
1161 }
1162
1163 /**
1164 * d_find_any_alias - find any alias for a given inode
1165 * @inode: inode to find an alias for
1166 *
1167 * If any aliases exist for the given inode, take and return a
1168 * reference for one of them. If no aliases exist, return %NULL.
1169 */
d_find_any_alias(struct inode * inode)1170 struct dentry *d_find_any_alias(struct inode *inode)
1171 {
1172 struct dentry *de;
1173
1174 spin_lock(&inode->i_lock);
1175 de = __d_find_any_alias(inode);
1176 spin_unlock(&inode->i_lock);
1177 return de;
1178 }
1179 EXPORT_SYMBOL(d_find_any_alias);
1180
__d_find_alias(struct inode * inode)1181 static struct dentry *__d_find_alias(struct inode *inode)
1182 {
1183 struct dentry *alias;
1184
1185 if (S_ISDIR(inode->i_mode))
1186 return __d_find_dir_alias(inode);
1187
1188 for_each_alias(alias, inode) {
1189 spin_lock(&alias->d_lock);
1190 if (!d_unhashed(alias)) {
1191 dget_dlock(alias);
1192 spin_unlock(&alias->d_lock);
1193 return alias;
1194 }
1195 spin_unlock(&alias->d_lock);
1196 }
1197 return NULL;
1198 }
1199
1200 /**
1201 * d_find_alias - grab a hashed alias of inode
1202 * @inode: inode in question
1203 *
1204 * If inode has a hashed alias, or is a directory and has any alias,
1205 * acquire the reference to alias and return it. Otherwise return NULL.
1206 * Notice that if inode is a directory there can be only one alias and
1207 * it can be unhashed only if it has no children, or if it is the root
1208 * of a filesystem, or if the directory was renamed and d_revalidate
1209 * was the first vfs operation to notice.
1210 *
1211 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
1212 * any other hashed alias over that one.
1213 */
d_find_alias(struct inode * inode)1214 struct dentry *d_find_alias(struct inode *inode)
1215 {
1216 struct dentry *de = NULL;
1217
1218 if (!hlist_empty(&inode->i_dentry)) {
1219 spin_lock(&inode->i_lock);
1220 de = __d_find_alias(inode);
1221 spin_unlock(&inode->i_lock);
1222 }
1223 return de;
1224 }
1225 EXPORT_SYMBOL(d_find_alias);
1226
1227 /*
1228 * Caller MUST be holding rcu_read_lock() and be guaranteed
1229 * that inode won't get freed until rcu_read_unlock().
1230 */
d_find_alias_rcu(struct inode * inode)1231 struct dentry *d_find_alias_rcu(struct inode *inode)
1232 {
1233 struct hlist_head *l = &inode->i_dentry;
1234 struct dentry *de = NULL;
1235
1236 spin_lock(&inode->i_lock);
1237 // ->i_dentry and ->i_rcu are colocated, but the latter won't be
1238 // used without having I_FREEING set, which means no aliases left
1239 if (likely(!(inode_state_read(inode) & I_FREEING) && !hlist_empty(l))) {
1240 if (S_ISDIR(inode->i_mode)) {
1241 de = hlist_entry(l->first, struct dentry, d_alias);
1242 } else {
1243 hlist_for_each_entry(de, l, d_alias)
1244 if (!d_unhashed(de))
1245 break;
1246 }
1247 }
1248 spin_unlock(&inode->i_lock);
1249 return de;
1250 }
1251
1252 /*
1253 * Try to kill dentries associated with this inode.
1254 * WARNING: you must own a reference to inode.
1255 */
d_prune_aliases(struct inode * inode)1256 void d_prune_aliases(struct inode *inode)
1257 {
1258 LIST_HEAD(dispose);
1259 struct dentry *dentry;
1260
1261 spin_lock(&inode->i_lock);
1262 for_each_alias(dentry, inode) {
1263 spin_lock(&dentry->d_lock);
1264 if (likely(!(dentry->d_flags & DCACHE_NORCU)))
1265 __move_to_shrink_list(dentry, &dispose);
1266 spin_unlock(&dentry->d_lock);
1267 }
1268 spin_unlock(&inode->i_lock);
1269 shrink_dentry_list(&dispose);
1270 }
1271 EXPORT_SYMBOL(d_prune_aliases);
1272
shrink_kill(struct dentry * victim)1273 static inline void shrink_kill(struct dentry *victim)
1274 {
1275 while ((victim = dentry_kill(victim)) != NULL)
1276 ;
1277 }
1278
shrink_dentry_list(struct list_head * list)1279 void shrink_dentry_list(struct list_head *list)
1280 {
1281 while (!list_empty(list)) {
1282 struct dentry *dentry;
1283
1284 dentry = list_entry(list->prev, struct dentry, d_lru);
1285 spin_lock(&dentry->d_lock);
1286 d_shrink_del(dentry);
1287 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
1288 spin_unlock(&dentry->d_lock);
1289 dentry_free(dentry);
1290 continue;
1291 }
1292 shrink_kill(dentry);
1293 }
1294 }
1295 EXPORT_SYMBOL(shrink_dentry_list);
1296
dentry_lru_isolate(struct list_head * item,struct list_lru_one * lru,void * arg)1297 static enum lru_status dentry_lru_isolate(struct list_head *item,
1298 struct list_lru_one *lru, void *arg)
1299 {
1300 struct list_head *freeable = arg;
1301 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1302
1303
1304 /*
1305 * we are inverting the lru lock/dentry->d_lock here,
1306 * so use a trylock. If we fail to get the lock, just skip
1307 * it
1308 */
1309 if (!spin_trylock(&dentry->d_lock))
1310 return LRU_SKIP;
1311
1312 /*
1313 * Referenced dentries are still in use. If they have active
1314 * counts, just remove them from the LRU. Otherwise give them
1315 * another pass through the LRU.
1316 */
1317 if (dentry->d_lockref.count) {
1318 d_lru_isolate(lru, dentry);
1319 spin_unlock(&dentry->d_lock);
1320 return LRU_REMOVED;
1321 }
1322
1323 if (dentry->d_flags & DCACHE_REFERENCED) {
1324 dentry->d_flags &= ~DCACHE_REFERENCED;
1325 spin_unlock(&dentry->d_lock);
1326
1327 /*
1328 * The list move itself will be made by the common LRU code. At
1329 * this point, we've dropped the dentry->d_lock but keep the
1330 * lru lock. This is safe to do, since every list movement is
1331 * protected by the lru lock even if both locks are held.
1332 *
1333 * This is guaranteed by the fact that all LRU management
1334 * functions are intermediated by the LRU API calls like
1335 * list_lru_add_obj and list_lru_del_obj. List movement in this file
1336 * only ever occur through this functions or through callbacks
1337 * like this one, that are called from the LRU API.
1338 *
1339 * The only exceptions to this are functions like
1340 * shrink_dentry_list, and code that first checks for the
1341 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1342 * operating only with stack provided lists after they are
1343 * properly isolated from the main list. It is thus, always a
1344 * local access.
1345 */
1346 return LRU_ROTATE;
1347 }
1348
1349 d_lru_shrink_move(lru, dentry, freeable);
1350 spin_unlock(&dentry->d_lock);
1351
1352 return LRU_REMOVED;
1353 }
1354
1355 /**
1356 * prune_dcache_sb - shrink the dcache
1357 * @sb: superblock
1358 * @sc: shrink control, passed to list_lru_shrink_walk()
1359 *
1360 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1361 * is done when we need more memory and called from the superblock shrinker
1362 * function.
1363 *
1364 * This function may fail to free any resources if all the dentries are in
1365 * use.
1366 */
prune_dcache_sb(struct super_block * sb,struct shrink_control * sc)1367 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1368 {
1369 LIST_HEAD(dispose);
1370 long freed;
1371
1372 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1373 dentry_lru_isolate, &dispose);
1374 shrink_dentry_list(&dispose);
1375 return freed;
1376 }
1377
dentry_lru_isolate_shrink(struct list_head * item,struct list_lru_one * lru,void * arg)1378 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1379 struct list_lru_one *lru, void *arg)
1380 {
1381 struct list_head *freeable = arg;
1382 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1383
1384 /*
1385 * we are inverting the lru lock/dentry->d_lock here,
1386 * so use a trylock. If we fail to get the lock, just skip
1387 * it
1388 */
1389 if (!spin_trylock(&dentry->d_lock))
1390 return LRU_SKIP;
1391
1392 d_lru_shrink_move(lru, dentry, freeable);
1393 spin_unlock(&dentry->d_lock);
1394
1395 return LRU_REMOVED;
1396 }
1397
1398
1399 /**
1400 * shrink_dcache_sb - shrink dcache for a superblock
1401 * @sb: superblock
1402 *
1403 * Shrink the dcache for the specified super block. This is used to free
1404 * the dcache before unmounting a file system.
1405 */
shrink_dcache_sb(struct super_block * sb)1406 void shrink_dcache_sb(struct super_block *sb)
1407 {
1408 do {
1409 LIST_HEAD(dispose);
1410
1411 list_lru_walk(&sb->s_dentry_lru,
1412 dentry_lru_isolate_shrink, &dispose, 1024);
1413 shrink_dentry_list(&dispose);
1414 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1415 }
1416 EXPORT_SYMBOL(shrink_dcache_sb);
1417
1418 /**
1419 * enum d_walk_ret - action to take during tree walk
1420 * @D_WALK_CONTINUE: continue walk
1421 * @D_WALK_QUIT: quit walk
1422 * @D_WALK_NORETRY: quit when retry is needed
1423 * @D_WALK_SKIP: skip this dentry and its children
1424 */
1425 enum d_walk_ret {
1426 D_WALK_CONTINUE,
1427 D_WALK_QUIT,
1428 D_WALK_NORETRY,
1429 D_WALK_SKIP,
1430 };
1431
1432 /**
1433 * d_walk - walk the dentry tree
1434 * @parent: start of walk
1435 * @data: data passed to @enter() and @finish()
1436 * @enter: callback when first entering the dentry
1437 *
1438 * The @enter() callbacks are called with d_lock held.
1439 */
d_walk(struct dentry * parent,void * data,enum d_walk_ret (* enter)(void *,struct dentry *))1440 static void d_walk(struct dentry *parent, void *data,
1441 enum d_walk_ret (*enter)(void *, struct dentry *))
1442 {
1443 struct dentry *this_parent, *dentry;
1444 unsigned seq = 0;
1445 enum d_walk_ret ret;
1446 bool retry = true;
1447
1448 again:
1449 read_seqbegin_or_lock(&rename_lock, &seq);
1450 this_parent = parent;
1451 spin_lock(&this_parent->d_lock);
1452 if (unlikely(this_parent->d_flags & DCACHE_DENTRY_CURSOR))
1453 goto out_unlock;
1454
1455 ret = enter(data, this_parent);
1456 switch (ret) {
1457 case D_WALK_CONTINUE:
1458 break;
1459 case D_WALK_QUIT:
1460 case D_WALK_SKIP:
1461 goto out_unlock;
1462 case D_WALK_NORETRY:
1463 retry = false;
1464 break;
1465 }
1466 repeat:
1467 dentry = d_first_child(this_parent);
1468 resume:
1469 hlist_for_each_entry_from(dentry, d_sib) {
1470 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1471 continue;
1472
1473 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1474
1475 ret = enter(data, dentry);
1476 switch (ret) {
1477 case D_WALK_CONTINUE:
1478 break;
1479 case D_WALK_QUIT:
1480 spin_unlock(&dentry->d_lock);
1481 goto out_unlock;
1482 case D_WALK_NORETRY:
1483 retry = false;
1484 break;
1485 case D_WALK_SKIP:
1486 spin_unlock(&dentry->d_lock);
1487 continue;
1488 }
1489
1490 if (!hlist_empty(&dentry->d_children)) {
1491 spin_unlock(&this_parent->d_lock);
1492 spin_release(&dentry->d_lock.dep_map, _RET_IP_);
1493 this_parent = dentry;
1494 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1495 goto repeat;
1496 }
1497 spin_unlock(&dentry->d_lock);
1498 }
1499 /*
1500 * All done at this level ... ascend and resume the search.
1501 */
1502 ascend:
1503 if (this_parent != parent) {
1504 dentry = this_parent;
1505 this_parent = dentry->d_parent;
1506
1507 rcu_read_lock();
1508 spin_unlock(&dentry->d_lock);
1509 spin_lock(&this_parent->d_lock);
1510 rcu_read_unlock();
1511
1512 /* might go back up the wrong parent if we have had a rename. */
1513 if (need_seqretry(&rename_lock, seq))
1514 goto rename_retry;
1515 /* go into the first sibling still alive */
1516 hlist_for_each_entry_continue(dentry, d_sib) {
1517 if (likely(!(dentry->d_flags & DCACHE_DENTRY_KILLED))) {
1518 goto resume;
1519 }
1520 }
1521 goto ascend;
1522 }
1523 if (need_seqretry(&rename_lock, seq))
1524 goto rename_retry;
1525
1526 out_unlock:
1527 spin_unlock(&this_parent->d_lock);
1528 done_seqretry(&rename_lock, seq);
1529 return;
1530
1531 rename_retry:
1532 spin_unlock(&this_parent->d_lock);
1533 BUG_ON(seq & 1);
1534 if (!retry)
1535 return;
1536 seq = 1;
1537 goto again;
1538 }
1539
1540 struct check_mount {
1541 struct vfsmount *mnt;
1542 unsigned int mounted;
1543 };
1544
1545 /* locks: mount_locked_reader && dentry->d_lock */
path_check_mount(void * data,struct dentry * dentry)1546 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1547 {
1548 struct check_mount *info = data;
1549 struct path path = { .mnt = info->mnt, .dentry = dentry };
1550
1551 if (likely(!d_mountpoint(dentry)))
1552 return D_WALK_CONTINUE;
1553 if (__path_is_mountpoint(&path)) {
1554 info->mounted = 1;
1555 return D_WALK_QUIT;
1556 }
1557 return D_WALK_CONTINUE;
1558 }
1559
1560 /**
1561 * path_has_submounts - check for mounts over a dentry in the
1562 * current namespace.
1563 * @parent: path to check.
1564 *
1565 * Return true if the parent or its subdirectories contain
1566 * a mount point in the current namespace.
1567 */
path_has_submounts(const struct path * parent)1568 int path_has_submounts(const struct path *parent)
1569 {
1570 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1571
1572 guard(mount_locked_reader)();
1573 d_walk(parent->dentry, &data, path_check_mount);
1574
1575 return data.mounted;
1576 }
1577 EXPORT_SYMBOL(path_has_submounts);
1578
1579 /*
1580 * Called by mount code to set a mountpoint and check if the mountpoint is
1581 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1582 * subtree can become unreachable).
1583 *
1584 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1585 * this reason take rename_lock and d_lock on dentry and ancestors.
1586 */
d_set_mounted(struct dentry * dentry)1587 int d_set_mounted(struct dentry *dentry)
1588 {
1589 struct dentry *p;
1590 int ret = -ENOENT;
1591 read_seqlock_excl(&rename_lock);
1592 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1593 /* Need exclusion wrt. d_invalidate() */
1594 spin_lock(&p->d_lock);
1595 if (unlikely(d_unhashed(p))) {
1596 spin_unlock(&p->d_lock);
1597 goto out;
1598 }
1599 spin_unlock(&p->d_lock);
1600 }
1601 spin_lock(&dentry->d_lock);
1602 if (!d_unlinked(dentry)) {
1603 ret = -EBUSY;
1604 if (!d_mountpoint(dentry)) {
1605 dentry->d_flags |= DCACHE_MOUNTED;
1606 ret = 0;
1607 }
1608 }
1609 spin_unlock(&dentry->d_lock);
1610 out:
1611 read_sequnlock_excl(&rename_lock);
1612 return ret;
1613 }
1614
1615 /*
1616 * Search the dentry child list of the specified parent,
1617 * and move any unused dentries to the end of the unused
1618 * list for prune_dcache(). We descend to the next level
1619 * whenever the d_children list is non-empty and continue
1620 * searching.
1621 *
1622 * It returns zero iff there are no unused children,
1623 * otherwise it returns the number of children moved to
1624 * the end of the unused list. This may not be the total
1625 * number of unused children, because select_parent can
1626 * drop the lock and return early due to latency
1627 * constraints.
1628 */
1629
1630 struct select_data {
1631 struct dentry *start;
1632 union {
1633 long found;
1634 struct dentry *victim;
1635 };
1636 struct list_head dispose;
1637 };
1638
select_collect(void * _data,struct dentry * dentry)1639 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1640 {
1641 struct select_data *data = _data;
1642 enum d_walk_ret ret = D_WALK_CONTINUE;
1643
1644 if (data->start == dentry)
1645 goto out;
1646
1647 if (lockref_is_dead_or_zero(&dentry->d_lockref)) {
1648 __move_to_shrink_list(dentry, &data->dispose);
1649 data->found++;
1650 }
1651 /*
1652 * We can return to the caller if we have found some (this
1653 * ensures forward progress). We'll be coming back to find
1654 * the rest.
1655 */
1656 if (!list_empty(&data->dispose))
1657 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1658 out:
1659 return ret;
1660 }
1661
select_collect_umount(void * _data,struct dentry * dentry)1662 static enum d_walk_ret select_collect_umount(void *_data, struct dentry *dentry)
1663 {
1664 if (dentry->d_flags & DCACHE_PERSISTENT) {
1665 dentry->d_flags &= ~DCACHE_PERSISTENT;
1666 dentry->d_lockref.count--;
1667 }
1668 return select_collect(_data, dentry);
1669 }
1670
select_collect2(void * _data,struct dentry * dentry)1671 static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry)
1672 {
1673 struct select_data *data = _data;
1674 enum d_walk_ret ret = D_WALK_CONTINUE;
1675
1676 if (data->start == dentry)
1677 goto out;
1678
1679 if (lockref_is_dead_or_zero(&dentry->d_lockref)) {
1680 if (!__move_to_shrink_list(dentry, &data->dispose)) {
1681 /*
1682 * We need an enter RCU read-side critical area that
1683 * would extend past the return from d_walk() and
1684 * we are in the scope of ->d_lock that will terminate
1685 * before that, so we use rcu_read_lock() to bridge
1686 * over to the scope of ->d_lock in d_walk() caller.
1687 * The scope of rcu_read_lock() spans from here to
1688 * paired rcu_read_unlock() in shrink_dcache_tree().
1689 */
1690 rcu_read_lock();
1691 data->victim = dentry;
1692 return D_WALK_QUIT;
1693 }
1694 }
1695 /*
1696 * We can return to the caller if we have found some (this
1697 * ensures forward progress). We'll be coming back to find
1698 * the rest.
1699 */
1700 if (!list_empty(&data->dispose))
1701 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1702 out:
1703 return ret;
1704 }
1705
1706 /**
1707 * shrink_dcache_tree - prune dcache
1708 * @parent: parent of entries to prune
1709 * @for_umount: true if we want to unpin the persistent ones
1710 *
1711 * Prune the dcache to remove unused children of the parent dentry.
1712 */
shrink_dcache_tree(struct dentry * parent,bool for_umount)1713 static void shrink_dcache_tree(struct dentry *parent, bool for_umount)
1714 {
1715 for (;;) {
1716 struct completion_list wait;
1717 bool need_wait = false;
1718 struct select_data data = { .start = parent };
1719
1720 INIT_LIST_HEAD(&data.dispose);
1721 d_walk(parent, &data,
1722 for_umount ? select_collect_umount : select_collect);
1723
1724 if (!list_empty(&data.dispose)) {
1725 shrink_dentry_list(&data.dispose);
1726 continue;
1727 }
1728
1729 cond_resched();
1730 if (!data.found)
1731 break;
1732 data.victim = NULL;
1733 d_walk(parent, &data, select_collect2);
1734 if (data.victim) {
1735 struct dentry *v = data.victim;
1736 /*
1737 * select_collect2() has picked a dentry that was
1738 * either dying or on a shrink list and arranged
1739 * for it to be returned to us. We are still in
1740 * the RCU read-side critical area started there
1741 * (rcu_read_lock() scope opened in select_collect2()),
1742 * so dentry couldn't have been freed yet, but its
1743 * state might've changed since we dropped ->d_lock
1744 * on the way out. Switch over to ->d_lock scope
1745 * and recheck the dentry state.
1746 */
1747 spin_lock(&v->d_lock);
1748 rcu_read_unlock();
1749
1750 if (unlikely(lockref_is_dead(&v->d_lockref))) {
1751 // It's doomed; if it isn't dead yet, notify us
1752 // once it becomes invisible to d_walk().
1753 need_wait = d_add_waiter(v, &wait);
1754 spin_unlock(&v->d_lock);
1755 } else {
1756 shrink_kill(v);
1757 }
1758 }
1759 shrink_dentry_list(&data.dispose);
1760 if (unlikely(need_wait))
1761 wait_for_completion(&wait.completion);
1762 }
1763 }
1764
shrink_dcache_parent(struct dentry * parent)1765 void shrink_dcache_parent(struct dentry *parent)
1766 {
1767 shrink_dcache_tree(parent, false);
1768 }
1769 EXPORT_SYMBOL(shrink_dcache_parent);
1770
umount_check(void * _data,struct dentry * dentry)1771 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1772 {
1773 /* it has busy descendents; complain about those instead */
1774 if (!hlist_empty(&dentry->d_children))
1775 return D_WALK_CONTINUE;
1776
1777 /* root with refcount 1 is fine */
1778 if (dentry == _data && dentry->d_lockref.count == 1)
1779 return D_WALK_CONTINUE;
1780
1781 WARN(1, "BUG: Dentry %p{i=%llx,n=%pd} "
1782 " still in use (%d) [unmount of %s %s]\n",
1783 dentry,
1784 dentry->d_inode ?
1785 dentry->d_inode->i_ino : (u64)0,
1786 dentry,
1787 dentry->d_lockref.count,
1788 dentry->d_sb->s_type->name,
1789 dentry->d_sb->s_id);
1790 return D_WALK_CONTINUE;
1791 }
1792
do_one_tree(struct dentry * dentry)1793 static void do_one_tree(struct dentry *dentry)
1794 {
1795 shrink_dcache_tree(dentry, true);
1796 d_walk(dentry, dentry, umount_check);
1797 spin_lock(&dentry->d_lock);
1798 __d_drop(dentry);
1799 /* A busy root survives the dput() below so don't leave it on ->s_roots. */
1800 if (unlikely(!hlist_unhashed(&dentry->d_sib)))
1801 unlink_secondary_root(dentry);
1802 spin_unlock(&dentry->d_lock);
1803 dput(dentry);
1804 }
1805
1806 /*
1807 * destroy the dentries attached to a superblock on unmounting
1808 */
shrink_dcache_for_umount(struct super_block * sb)1809 void shrink_dcache_for_umount(struct super_block *sb)
1810 {
1811 struct dentry *dentry;
1812
1813 rwsem_assert_held_write(&sb->s_umount);
1814
1815 dentry = sb->s_root;
1816 sb->s_root = NULL;
1817 do_one_tree(dentry);
1818
1819 for (;;) {
1820 spin_lock(&sb->s_roots_lock);
1821 dentry = hlist_entry_safe(sb->s_roots.first,
1822 struct dentry, d_sib);
1823 if (!dentry) {
1824 spin_unlock(&sb->s_roots_lock);
1825 break;
1826 }
1827 rcu_read_lock();
1828 spin_unlock(&sb->s_roots_lock);
1829 spin_lock(&dentry->d_lock);
1830 rcu_read_unlock();
1831 if (unlikely(lockref_is_dead(&dentry->d_lockref))) {
1832 struct completion_list wait;
1833 bool need_wait = d_add_waiter(dentry, &wait);
1834
1835 spin_unlock(&dentry->d_lock);
1836 if (need_wait)
1837 wait_for_completion(&wait.completion);
1838 } else {
1839 dget_dlock(dentry);
1840 spin_unlock(&dentry->d_lock);
1841 do_one_tree(dentry);
1842 }
1843 }
1844 }
1845
find_submount(void * _data,struct dentry * dentry)1846 static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
1847 {
1848 struct dentry **victim = _data;
1849 if (d_mountpoint(dentry)) {
1850 *victim = dget_dlock(dentry);
1851 return D_WALK_QUIT;
1852 }
1853 return D_WALK_CONTINUE;
1854 }
1855
1856 /**
1857 * d_invalidate - detach submounts, prune dcache, and drop
1858 * @dentry: dentry to invalidate (aka detach, prune and drop)
1859 */
d_invalidate(struct dentry * dentry)1860 void d_invalidate(struct dentry *dentry)
1861 {
1862 bool had_submounts = false;
1863 spin_lock(&dentry->d_lock);
1864 if (d_unhashed(dentry)) {
1865 spin_unlock(&dentry->d_lock);
1866 return;
1867 }
1868 __d_drop(dentry);
1869 spin_unlock(&dentry->d_lock);
1870
1871 /* Negative dentries can be dropped without further checks */
1872 if (!dentry->d_inode)
1873 return;
1874
1875 shrink_dcache_parent(dentry);
1876 for (;;) {
1877 struct dentry *victim = NULL;
1878 d_walk(dentry, &victim, find_submount);
1879 if (!victim) {
1880 if (had_submounts)
1881 shrink_dcache_parent(dentry);
1882 return;
1883 }
1884 had_submounts = true;
1885 detach_mounts(victim);
1886 dput(victim);
1887 }
1888 }
1889 EXPORT_SYMBOL(d_invalidate);
1890
1891 /**
1892 * __d_alloc - allocate a dcache entry
1893 * @sb: filesystem it will belong to
1894 * @name: qstr of the name
1895 *
1896 * Allocates a dentry. It returns %NULL if there is insufficient memory
1897 * available. On a success the dentry is returned. The name passed in is
1898 * copied and the copy passed in may be reused after this call.
1899 */
1900
__d_alloc(struct super_block * sb,const struct qstr * name)1901 static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1902 {
1903 struct dentry *dentry;
1904 char *dname;
1905 int err;
1906
1907 dentry = kmem_cache_alloc_lru(dentry_cache, &sb->s_dentry_lru,
1908 GFP_KERNEL);
1909 if (!dentry)
1910 return NULL;
1911
1912 /*
1913 * We guarantee that the inline name is always NUL-terminated.
1914 * This way the memcpy() done by the name switching in rename
1915 * will still always have a NUL at the end, even if we might
1916 * be overwriting an internal NUL character
1917 */
1918 dentry->d_shortname.string[DNAME_INLINE_LEN-1] = 0;
1919
1920 /* Racy __d_lookup_rcu() walk may read past the NUL; harmless */
1921 kmsan_unpoison_memory(dentry->d_shortname.string, DNAME_INLINE_LEN);
1922
1923 if (unlikely(!name)) {
1924 name = &slash_name;
1925 dname = dentry->d_shortname.string;
1926 } else if (name->len > DNAME_INLINE_LEN-1) {
1927 struct external_name *p;
1928
1929 p = kmalloc_flex(*p, name, name->len + 1,
1930 GFP_KERNEL_ACCOUNT | __GFP_RECLAIMABLE);
1931 if (!p) {
1932 kmem_cache_free(dentry_cache, dentry);
1933 return NULL;
1934 }
1935 atomic_set(&p->count, 1);
1936 dname = p->name;
1937 } else {
1938 dname = dentry->d_shortname.string;
1939 }
1940
1941 dentry->__d_name.len = name->len;
1942 dentry->__d_name.hash = name->hash;
1943 memcpy(dname, name->name, name->len);
1944 dname[name->len] = 0;
1945
1946 /* Make sure we always see the terminating NUL character */
1947 smp_store_release(&dentry->__d_name.name, dname); /* ^^^ */
1948
1949 dentry->d_flags = 0;
1950 lockref_init(&dentry->d_lockref);
1951 seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock);
1952 dentry->d_inode = NULL;
1953 dentry->d_parent = dentry;
1954 dentry->d_sb = sb;
1955 dentry->d_op = sb->__s_d_op;
1956 dentry->d_flags = sb->s_d_flags;
1957 dentry->d_fsdata = NULL;
1958 INIT_HLIST_BL_NODE(&dentry->d_hash);
1959 INIT_LIST_HEAD(&dentry->d_lru);
1960 INIT_HLIST_HEAD(&dentry->d_children);
1961 dentry->waiters = NULL;
1962 INIT_HLIST_NODE(&dentry->d_sib);
1963
1964 if (dentry->d_op && dentry->d_op->d_init) {
1965 err = dentry->d_op->d_init(dentry);
1966 if (err) {
1967 if (dname_external(dentry))
1968 kfree(external_name(dentry));
1969 kmem_cache_free(dentry_cache, dentry);
1970 return NULL;
1971 }
1972 }
1973
1974 this_cpu_inc(nr_dentry);
1975
1976 return dentry;
1977 }
1978
1979 /**
1980 * d_alloc - allocate a dcache entry
1981 * @parent: parent of entry to allocate
1982 * @name: qstr of the name
1983 *
1984 * Allocates a dentry. It returns %NULL if there is insufficient memory
1985 * available. On a success the dentry is returned. The name passed in is
1986 * copied and the copy passed in may be reused after this call.
1987 */
d_alloc(struct dentry * parent,const struct qstr * name)1988 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1989 {
1990 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1991 if (!dentry)
1992 return NULL;
1993 spin_lock(&parent->d_lock);
1994 /*
1995 * don't need child lock because it is not subject
1996 * to concurrency here
1997 */
1998 dentry->d_parent = dget_dlock(parent);
1999 hlist_add_head(&dentry->d_sib, &parent->d_children);
2000 spin_unlock(&parent->d_lock);
2001
2002 return dentry;
2003 }
2004 EXPORT_SYMBOL(d_alloc);
2005
d_alloc_anon(struct super_block * sb)2006 struct dentry *d_alloc_anon(struct super_block *sb)
2007 {
2008 return __d_alloc(sb, NULL);
2009 }
2010 EXPORT_SYMBOL(d_alloc_anon);
2011
d_alloc_cursor(struct dentry * parent)2012 struct dentry *d_alloc_cursor(struct dentry * parent)
2013 {
2014 struct dentry *dentry = d_alloc_anon(parent->d_sb);
2015 if (dentry) {
2016 dentry->d_flags |= DCACHE_DENTRY_CURSOR | DCACHE_NORCU;
2017 dentry->d_parent = dget(parent);
2018 }
2019 return dentry;
2020 }
2021
2022 /**
2023 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
2024 * @sb: the superblock
2025 * @name: qstr of the name
2026 *
2027 * For a filesystem that just pins its dentries in memory and never
2028 * performs lookups at all, return an unhashed IS_ROOT dentry.
2029 * This is used for pipes, sockets et.al. - the stuff that should
2030 * never be anyone's children or parents. Unlike all other
2031 * dentries, these will not have RCU delay between dropping the
2032 * last reference and freeing them.
2033 *
2034 * The only user is alloc_file_pseudo() and that's what should
2035 * be considered a public interface. Don't use directly.
2036 */
d_alloc_pseudo(struct super_block * sb,const struct qstr * name)2037 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
2038 {
2039 static const struct dentry_operations anon_ops = {
2040 .d_dname = simple_dname
2041 };
2042 struct dentry *dentry = __d_alloc(sb, name);
2043 if (likely(dentry)) {
2044 dentry->d_flags |= DCACHE_NORCU;
2045 /* d_op_flags(&anon_ops) is 0 */
2046 if (!dentry->d_op)
2047 dentry->d_op = &anon_ops;
2048 }
2049 return dentry;
2050 }
2051
d_alloc_name(struct dentry * parent,const char * name)2052 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
2053 {
2054 struct qstr q;
2055
2056 q.name = name;
2057 q.hash_len = hashlen_string(parent, name);
2058 return d_alloc(parent, &q);
2059 }
2060 EXPORT_SYMBOL(d_alloc_name);
2061
2062 #define DCACHE_OP_FLAGS \
2063 (DCACHE_OP_HASH | DCACHE_OP_COMPARE | DCACHE_OP_REVALIDATE | \
2064 DCACHE_OP_WEAK_REVALIDATE | DCACHE_OP_DELETE | DCACHE_OP_PRUNE | \
2065 DCACHE_OP_REAL)
2066
d_op_flags(const struct dentry_operations * op)2067 static unsigned int d_op_flags(const struct dentry_operations *op)
2068 {
2069 unsigned int flags = 0;
2070 if (op) {
2071 if (op->d_hash)
2072 flags |= DCACHE_OP_HASH;
2073 if (op->d_compare)
2074 flags |= DCACHE_OP_COMPARE;
2075 if (op->d_revalidate)
2076 flags |= DCACHE_OP_REVALIDATE;
2077 if (op->d_weak_revalidate)
2078 flags |= DCACHE_OP_WEAK_REVALIDATE;
2079 if (op->d_delete)
2080 flags |= DCACHE_OP_DELETE;
2081 if (op->d_prune)
2082 flags |= DCACHE_OP_PRUNE;
2083 if (op->d_real)
2084 flags |= DCACHE_OP_REAL;
2085 }
2086 return flags;
2087 }
2088
d_set_d_op(struct dentry * dentry,const struct dentry_operations * op)2089 static void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
2090 {
2091 unsigned int flags = d_op_flags(op);
2092 WARN_ON_ONCE(dentry->d_op);
2093 WARN_ON_ONCE(dentry->d_flags & DCACHE_OP_FLAGS);
2094 dentry->d_op = op;
2095 if (flags)
2096 dentry->d_flags |= flags;
2097 }
2098
set_default_d_op(struct super_block * s,const struct dentry_operations * ops)2099 void set_default_d_op(struct super_block *s, const struct dentry_operations *ops)
2100 {
2101 unsigned int flags = d_op_flags(ops);
2102 s->__s_d_op = ops;
2103 s->s_d_flags = (s->s_d_flags & ~DCACHE_OP_FLAGS) | flags;
2104 }
2105 EXPORT_SYMBOL(set_default_d_op);
2106
d_flags_for_inode(struct inode * inode)2107 static unsigned d_flags_for_inode(struct inode *inode)
2108 {
2109 unsigned add_flags = DCACHE_REGULAR_TYPE;
2110
2111 if (!inode)
2112 return DCACHE_MISS_TYPE;
2113
2114 if (S_ISDIR(inode->i_mode)) {
2115 add_flags = DCACHE_DIRECTORY_TYPE;
2116 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
2117 if (unlikely(!inode->i_op->lookup))
2118 add_flags = DCACHE_AUTODIR_TYPE;
2119 else
2120 inode->i_opflags |= IOP_LOOKUP;
2121 }
2122 goto type_determined;
2123 }
2124
2125 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
2126 if (unlikely(inode->i_op->get_link)) {
2127 add_flags = DCACHE_SYMLINK_TYPE;
2128 goto type_determined;
2129 }
2130 inode->i_opflags |= IOP_NOFOLLOW;
2131 }
2132
2133 if (unlikely(!S_ISREG(inode->i_mode)))
2134 add_flags = DCACHE_SPECIAL_TYPE;
2135
2136 type_determined:
2137 if (unlikely(IS_AUTOMOUNT(inode)))
2138 add_flags |= DCACHE_NEED_AUTOMOUNT;
2139 return add_flags;
2140 }
2141
__d_instantiate(struct dentry * dentry,struct inode * inode)2142 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
2143 {
2144 unsigned add_flags = d_flags_for_inode(inode);
2145 WARN_ON(d_in_lookup(dentry));
2146
2147 /*
2148 * The negative counter only tracks dentries on the LRU. Don't dec if
2149 * d_lru is on another list.
2150 */
2151 if ((dentry->d_flags &
2152 (DCACHE_LRU_LIST|DCACHE_SHRINK_LIST)) == DCACHE_LRU_LIST)
2153 this_cpu_dec(nr_dentry_negative);
2154 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
2155 raw_write_seqcount_begin(&dentry->d_seq);
2156 __d_set_inode_and_type(dentry, inode, add_flags);
2157 raw_write_seqcount_end(&dentry->d_seq);
2158 fsnotify_update_flags(dentry);
2159 }
2160
2161 /**
2162 * d_instantiate - fill in inode information for a dentry
2163 * @entry: dentry to complete
2164 * @inode: inode to attach to this dentry
2165 *
2166 * Fill in inode information in the entry.
2167 *
2168 * This turns negative dentries into productive full members
2169 * of society.
2170 *
2171 * NOTE! This assumes that the inode count has been incremented
2172 * (or otherwise set) by the caller to indicate that it is now
2173 * in use by the dcache.
2174 */
2175
d_instantiate(struct dentry * entry,struct inode * inode)2176 void d_instantiate(struct dentry *entry, struct inode * inode)
2177 {
2178 BUG_ON(d_really_is_positive(entry));
2179 if (inode) {
2180 security_d_instantiate(entry, inode);
2181 spin_lock(&inode->i_lock);
2182 spin_lock(&entry->d_lock);
2183 __d_instantiate(entry, inode);
2184 spin_unlock(&entry->d_lock);
2185 spin_unlock(&inode->i_lock);
2186 }
2187 }
2188 EXPORT_SYMBOL(d_instantiate);
2189
2190 /*
2191 * This should be equivalent to d_instantiate() + unlock_new_inode(),
2192 * with lockdep-related part of unlock_new_inode() done before
2193 * anything else. Use that instead of open-coding d_instantiate()/
2194 * unlock_new_inode() combinations.
2195 */
d_instantiate_new(struct dentry * entry,struct inode * inode)2196 void d_instantiate_new(struct dentry *entry, struct inode *inode)
2197 {
2198 BUG_ON(d_really_is_positive(entry));
2199 BUG_ON(!inode);
2200 lockdep_annotate_inode_mutex_key(inode);
2201 security_d_instantiate(entry, inode);
2202 spin_lock(&inode->i_lock);
2203 spin_lock(&entry->d_lock);
2204 __d_instantiate(entry, inode);
2205 spin_unlock(&entry->d_lock);
2206 WARN_ON(!(inode_state_read(inode) & I_NEW));
2207 /*
2208 * Paired with igrab_from_hash()
2209 */
2210 smp_wmb();
2211 inode_state_clear(inode, I_NEW | I_CREATING);
2212 inode_wake_up_bit(inode, __I_NEW);
2213 spin_unlock(&inode->i_lock);
2214 }
2215 EXPORT_SYMBOL(d_instantiate_new);
2216
d_make_root(struct inode * root_inode)2217 struct dentry *d_make_root(struct inode *root_inode)
2218 {
2219 struct dentry *res = NULL;
2220
2221 if (root_inode) {
2222 res = d_alloc_anon(root_inode->i_sb);
2223 if (res)
2224 d_instantiate(res, root_inode);
2225 else
2226 iput(root_inode);
2227 }
2228 return res;
2229 }
2230 EXPORT_SYMBOL(d_make_root);
2231
__d_obtain_alias(struct inode * inode,bool disconnected)2232 static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
2233 {
2234 struct super_block *sb;
2235 struct dentry *new, *res;
2236
2237 if (!inode)
2238 return ERR_PTR(-ESTALE);
2239 if (IS_ERR(inode))
2240 return ERR_CAST(inode);
2241
2242 sb = inode->i_sb;
2243
2244 res = d_find_any_alias(inode); /* existing alias? */
2245 if (res)
2246 goto out;
2247
2248 new = d_alloc_anon(sb);
2249 if (!new) {
2250 res = ERR_PTR(-ENOMEM);
2251 goto out;
2252 }
2253
2254 security_d_instantiate(new, inode);
2255 spin_lock(&inode->i_lock);
2256 res = __d_find_any_alias(inode); /* recheck under lock */
2257 if (likely(!res)) { /* still no alias, attach a disconnected dentry */
2258 unsigned add_flags = d_flags_for_inode(inode);
2259
2260 if (disconnected)
2261 add_flags |= DCACHE_DISCONNECTED;
2262
2263 spin_lock(&new->d_lock);
2264 __d_set_inode_and_type(new, inode, add_flags);
2265 hlist_add_head(&new->d_alias, &inode->i_dentry);
2266 if (!disconnected) {
2267 spin_lock(&sb->s_roots_lock);
2268 hlist_add_head(&new->d_sib, &sb->s_roots);
2269 spin_unlock(&sb->s_roots_lock);
2270 }
2271 spin_unlock(&new->d_lock);
2272 spin_unlock(&inode->i_lock);
2273 inode = NULL; /* consumed by new->d_inode */
2274 res = new;
2275 } else {
2276 spin_unlock(&inode->i_lock);
2277 dput(new);
2278 }
2279
2280 out:
2281 iput(inode);
2282 return res;
2283 }
2284
2285 /**
2286 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2287 * @inode: inode to allocate the dentry for
2288 *
2289 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2290 * similar open by handle operations. The returned dentry may be anonymous,
2291 * or may have a full name (if the inode was already in the cache).
2292 *
2293 * When called on a directory inode, we must ensure that the inode only ever
2294 * has one dentry. If a dentry is found, that is returned instead of
2295 * allocating a new one.
2296 *
2297 * On successful return, the reference to the inode has been transferred
2298 * to the dentry. In case of an error the reference on the inode is released.
2299 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2300 * be passed in and the error will be propagated to the return value,
2301 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2302 */
d_obtain_alias(struct inode * inode)2303 struct dentry *d_obtain_alias(struct inode *inode)
2304 {
2305 return __d_obtain_alias(inode, true);
2306 }
2307 EXPORT_SYMBOL(d_obtain_alias);
2308
2309 /**
2310 * d_obtain_root - find or allocate a dentry for a given inode
2311 * @inode: inode to allocate the dentry for
2312 *
2313 * Obtain an IS_ROOT dentry for the root of a filesystem.
2314 *
2315 * We must ensure that directory inodes only ever have one dentry. If a
2316 * dentry is found, that is returned instead of allocating a new one.
2317 *
2318 * On successful return, the reference to the inode has been transferred
2319 * to the dentry. In case of an error the reference on the inode is
2320 * released. A %NULL or IS_ERR inode may be passed in and will be the
2321 * error will be propagate to the return value, with a %NULL @inode
2322 * replaced by ERR_PTR(-ESTALE).
2323 */
d_obtain_root(struct inode * inode)2324 struct dentry *d_obtain_root(struct inode *inode)
2325 {
2326 return __d_obtain_alias(inode, false);
2327 }
2328 EXPORT_SYMBOL(d_obtain_root);
2329
2330 /**
2331 * d_add_ci - lookup or allocate new dentry with case-exact name
2332 * @dentry: the negative dentry that was passed to the parent's lookup func
2333 * @inode: the inode case-insensitive lookup has found
2334 * @name: the case-exact name to be associated with the returned dentry
2335 *
2336 * This is to avoid filling the dcache with case-insensitive names to the
2337 * same inode, only the actual correct case is stored in the dcache for
2338 * case-insensitive filesystems.
2339 *
2340 * For a case-insensitive lookup match and if the case-exact dentry
2341 * already exists in the dcache, use it and return it.
2342 *
2343 * If no entry exists with the exact case name, allocate new dentry with
2344 * the exact case, and return the spliced entry.
2345 */
d_add_ci(struct dentry * dentry,struct inode * inode,struct qstr * name)2346 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2347 struct qstr *name)
2348 {
2349 struct dentry *found, *res;
2350
2351 /*
2352 * First check if a dentry matching the name already exists,
2353 * if not go ahead and create it now.
2354 */
2355 found = d_hash_and_lookup(dentry->d_parent, name);
2356 if (found) {
2357 iput(inode);
2358 return found;
2359 }
2360 if (d_in_lookup(dentry)) {
2361 found = d_alloc_parallel(dentry->d_parent, name);
2362 if (IS_ERR(found) || !d_in_lookup(found)) {
2363 iput(inode);
2364 return found;
2365 }
2366 } else {
2367 found = d_alloc(dentry->d_parent, name);
2368 if (!found) {
2369 iput(inode);
2370 return ERR_PTR(-ENOMEM);
2371 }
2372 }
2373 res = d_splice_alias(inode, found);
2374 if (res) {
2375 d_lookup_done(found);
2376 dput(found);
2377 return res;
2378 }
2379 return found;
2380 }
2381 EXPORT_SYMBOL(d_add_ci);
2382
2383 /**
2384 * d_same_name - compare dentry name with case-exact name
2385 * @dentry: the negative dentry that was passed to the parent's lookup func
2386 * @parent: parent dentry
2387 * @name: the case-exact name to be associated with the returned dentry
2388 *
2389 * Return: true if names are same, or false
2390 */
d_same_name(const struct dentry * dentry,const struct dentry * parent,const struct qstr * name)2391 bool d_same_name(const struct dentry *dentry, const struct dentry *parent,
2392 const struct qstr *name)
2393 {
2394 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2395 if (dentry->d_name.len != name->len)
2396 return false;
2397 return dentry_cmp(dentry, name->name, name->len) == 0;
2398 }
2399 return parent->d_op->d_compare(dentry,
2400 dentry->d_name.len, dentry->d_name.name,
2401 name) == 0;
2402 }
2403 EXPORT_SYMBOL_GPL(d_same_name);
2404
2405 /*
2406 * This is __d_lookup_rcu() when the parent dentry has
2407 * DCACHE_OP_COMPARE, which makes things much nastier.
2408 */
__d_lookup_rcu_op_compare(const struct dentry * parent,const struct qstr * name,unsigned * seqp)2409 static noinline struct dentry *__d_lookup_rcu_op_compare(
2410 const struct dentry *parent,
2411 const struct qstr *name,
2412 unsigned *seqp)
2413 {
2414 u64 hashlen = name->hash_len;
2415 struct hlist_bl_head *b = d_hash(hashlen);
2416 struct hlist_bl_node *node;
2417 struct dentry *dentry;
2418
2419 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2420 int tlen;
2421 const char *tname;
2422 unsigned seq;
2423
2424 seqretry:
2425 seq = raw_seqcount_begin(&dentry->d_seq);
2426 if (dentry->d_parent != parent)
2427 continue;
2428 if (d_unhashed(dentry))
2429 continue;
2430 if (dentry->d_name.hash != hashlen_hash(hashlen))
2431 continue;
2432 tlen = dentry->d_name.len;
2433 tname = dentry->d_name.name;
2434 /* we want a consistent (name,len) pair */
2435 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2436 cpu_relax();
2437 goto seqretry;
2438 }
2439 if (parent->d_op->d_compare(dentry, tlen, tname, name) != 0)
2440 continue;
2441 *seqp = seq;
2442 return dentry;
2443 }
2444 return NULL;
2445 }
2446
2447 /**
2448 * __d_lookup_rcu - search for a dentry (racy, store-free)
2449 * @parent: parent dentry
2450 * @name: qstr of name we wish to find
2451 * @seqp: returns d_seq value at the point where the dentry was found
2452 * Returns: dentry, or NULL
2453 *
2454 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2455 * resolution (store-free path walking) design described in
2456 * Documentation/filesystems/path-lookup.txt.
2457 *
2458 * This is not to be used outside core vfs.
2459 *
2460 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2461 * held, and rcu_read_lock held. The returned dentry must not be stored into
2462 * without taking d_lock and checking d_seq sequence count against @seq
2463 * returned here.
2464 *
2465 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2466 * the returned dentry, so long as its parent's seqlock is checked after the
2467 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2468 * is formed, giving integrity down the path walk.
2469 *
2470 * NOTE! The caller *has* to check the resulting dentry against the sequence
2471 * number we've returned before using any of the resulting dentry state!
2472 */
__d_lookup_rcu(const struct dentry * parent,const struct qstr * name,unsigned * seqp)2473 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2474 const struct qstr *name,
2475 unsigned *seqp)
2476 {
2477 u64 hashlen = name->hash_len;
2478 const unsigned char *str = name->name;
2479 struct hlist_bl_head *b = d_hash(hashlen);
2480 struct hlist_bl_node *node;
2481 struct dentry *dentry;
2482
2483 /*
2484 * Note: There is significant duplication with __d_lookup_rcu which is
2485 * required to prevent single threaded performance regressions
2486 * especially on architectures where smp_rmb (in seqcounts) are costly.
2487 * Keep the two functions in sync.
2488 */
2489
2490 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE))
2491 return __d_lookup_rcu_op_compare(parent, name, seqp);
2492
2493 /*
2494 * The hash list is protected using RCU.
2495 *
2496 * Carefully use d_seq when comparing a candidate dentry, to avoid
2497 * races with d_move().
2498 *
2499 * It is possible that concurrent renames can mess up our list
2500 * walk here and result in missing our dentry, resulting in the
2501 * false-negative result. d_lookup() protects against concurrent
2502 * renames using rename_lock seqlock.
2503 *
2504 * See Documentation/filesystems/path-lookup.txt for more details.
2505 */
2506 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2507 unsigned seq;
2508
2509 /*
2510 * The dentry sequence count protects us from concurrent
2511 * renames, and thus protects parent and name fields.
2512 *
2513 * The caller must perform a seqcount check in order
2514 * to do anything useful with the returned dentry.
2515 *
2516 * NOTE! We do a "raw" seqcount_begin here. That means that
2517 * we don't wait for the sequence count to stabilize if it
2518 * is in the middle of a sequence change. If we do the slow
2519 * dentry compare, we will do seqretries until it is stable,
2520 * and if we end up with a successful lookup, we actually
2521 * want to exit RCU lookup anyway.
2522 *
2523 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2524 * we are still guaranteed NUL-termination of ->d_name.name.
2525 */
2526 seq = raw_seqcount_begin(&dentry->d_seq);
2527 if (dentry->d_parent != parent)
2528 continue;
2529 if (dentry->d_name.hash_len != hashlen)
2530 continue;
2531 if (unlikely(dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0))
2532 continue;
2533 /*
2534 * Check for the dentry being unhashed.
2535 *
2536 * As tempting as it is, we *can't* skip it because of a race window
2537 * between us finding the dentry before it gets unhashed and loading
2538 * the sequence counter after unhashing is finished.
2539 *
2540 * We can at least predict on it.
2541 */
2542 if (unlikely(d_unhashed(dentry)))
2543 continue;
2544 *seqp = seq;
2545 return dentry;
2546 }
2547 return NULL;
2548 }
2549
2550 /**
2551 * d_lookup - search for a dentry
2552 * @parent: parent dentry
2553 * @name: qstr of name we wish to find
2554 * Returns: dentry, or NULL
2555 *
2556 * d_lookup searches the children of the parent dentry for the name in
2557 * question. If the dentry is found its reference count is incremented and the
2558 * dentry is returned. The caller must use dput to free the entry when it has
2559 * finished using it. %NULL is returned if the dentry does not exist.
2560 */
d_lookup(const struct dentry * parent,const struct qstr * name)2561 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2562 {
2563 struct dentry *dentry;
2564 unsigned seq;
2565
2566 do {
2567 seq = read_seqbegin(&rename_lock);
2568 dentry = __d_lookup(parent, name);
2569 if (dentry)
2570 break;
2571 } while (read_seqretry(&rename_lock, seq));
2572 return dentry;
2573 }
2574 EXPORT_SYMBOL(d_lookup);
2575
2576 /**
2577 * __d_lookup - search for a dentry (racy)
2578 * @parent: parent dentry
2579 * @name: qstr of name we wish to find
2580 * Returns: dentry, or NULL
2581 *
2582 * __d_lookup is like d_lookup, however it may (rarely) return a
2583 * false-negative result due to unrelated rename activity.
2584 *
2585 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2586 * however it must be used carefully, eg. with a following d_lookup in
2587 * the case of failure.
2588 *
2589 * __d_lookup callers must be commented.
2590 */
__d_lookup(const struct dentry * parent,const struct qstr * name)2591 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2592 {
2593 unsigned int hash = name->hash;
2594 struct hlist_bl_head *b = d_hash(hash);
2595 struct hlist_bl_node *node;
2596 struct dentry *found = NULL;
2597 struct dentry *dentry;
2598
2599 /*
2600 * Note: There is significant duplication with __d_lookup_rcu which is
2601 * required to prevent single threaded performance regressions
2602 * especially on architectures where smp_rmb (in seqcounts) are costly.
2603 * Keep the two functions in sync.
2604 */
2605
2606 /*
2607 * The hash list is protected using RCU.
2608 *
2609 * Take d_lock when comparing a candidate dentry, to avoid races
2610 * with d_move().
2611 *
2612 * It is possible that concurrent renames can mess up our list
2613 * walk here and result in missing our dentry, resulting in the
2614 * false-negative result. d_lookup() protects against concurrent
2615 * renames using rename_lock seqlock.
2616 *
2617 * See Documentation/filesystems/path-lookup.txt for more details.
2618 */
2619 rcu_read_lock();
2620
2621 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2622
2623 if (dentry->d_name.hash != hash)
2624 continue;
2625
2626 spin_lock(&dentry->d_lock);
2627 if (dentry->d_parent != parent)
2628 goto next;
2629 if (d_unhashed(dentry))
2630 goto next;
2631
2632 if (!d_same_name(dentry, parent, name))
2633 goto next;
2634
2635 dentry->d_lockref.count++;
2636 found = dentry;
2637 spin_unlock(&dentry->d_lock);
2638 break;
2639 next:
2640 spin_unlock(&dentry->d_lock);
2641 }
2642 rcu_read_unlock();
2643
2644 return found;
2645 }
2646
2647 /**
2648 * d_hash_and_lookup - hash the qstr then search for a dentry
2649 * @dir: Directory to search in
2650 * @name: qstr of name we wish to find
2651 *
2652 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2653 */
d_hash_and_lookup(struct dentry * dir,struct qstr * name)2654 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2655 {
2656 /*
2657 * Check for a fs-specific hash function. Note that we must
2658 * calculate the standard hash first, as the d_op->d_hash()
2659 * routine may choose to leave the hash value unchanged.
2660 */
2661 name->hash = full_name_hash(dir, name->name, name->len);
2662 if (dir->d_flags & DCACHE_OP_HASH) {
2663 int err = dir->d_op->d_hash(dir, name);
2664 if (unlikely(err < 0))
2665 return ERR_PTR(err);
2666 }
2667 return d_lookup(dir, name);
2668 }
2669
2670 /*
2671 * When a file is deleted, we have two options:
2672 * - turn this dentry into a negative dentry
2673 * - unhash this dentry and free it.
2674 *
2675 * Usually, we want to just turn this into
2676 * a negative dentry, but if anybody else is
2677 * currently using the dentry or the inode
2678 * we can't do that and we fall back on removing
2679 * it from the hash queues and waiting for
2680 * it to be deleted later when it has no users
2681 */
2682
2683 /**
2684 * d_delete - delete a dentry
2685 * @dentry: The dentry to delete
2686 *
2687 * Turn the dentry into a negative dentry if possible, otherwise
2688 * remove it from the hash queues so it can be deleted later
2689 */
2690
d_delete(struct dentry * dentry)2691 void d_delete(struct dentry * dentry)
2692 {
2693 struct inode *inode = dentry->d_inode;
2694
2695 spin_lock(&inode->i_lock);
2696 spin_lock(&dentry->d_lock);
2697 /*
2698 * Are we the only user?
2699 */
2700 if (dentry->d_lockref.count == 1) {
2701 if (dentry_negative_policy)
2702 __d_drop(dentry);
2703 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2704 dentry_unlink_inode(dentry);
2705 } else {
2706 __d_drop(dentry);
2707 spin_unlock(&dentry->d_lock);
2708 spin_unlock(&inode->i_lock);
2709 }
2710 }
2711 EXPORT_SYMBOL(d_delete);
2712
__d_rehash(struct dentry * entry)2713 static void __d_rehash(struct dentry *entry)
2714 {
2715 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2716
2717 hlist_bl_lock(b);
2718 hlist_bl_add_head_rcu(&entry->d_hash, b);
2719 hlist_bl_unlock(b);
2720 }
2721
2722 /**
2723 * d_rehash - add an entry back to the hash
2724 * @entry: dentry to add to the hash
2725 *
2726 * Adds a dentry to the hash according to its name.
2727 */
2728
d_rehash(struct dentry * entry)2729 void d_rehash(struct dentry * entry)
2730 {
2731 spin_lock(&entry->d_lock);
2732 __d_rehash(entry);
2733 spin_unlock(&entry->d_lock);
2734 }
2735 EXPORT_SYMBOL(d_rehash);
2736
start_dir_add(struct inode * dir)2737 static inline unsigned start_dir_add(struct inode *dir)
2738 {
2739 preempt_disable_nested();
2740 for (;;) {
2741 unsigned n = READ_ONCE(dir->i_dir_seq);
2742 if (!(n & 1) && try_cmpxchg(&dir->i_dir_seq, &n, n + 1))
2743 return n;
2744 cpu_relax();
2745 }
2746 }
2747
end_dir_add(struct inode * dir,unsigned int n)2748 static inline void end_dir_add(struct inode *dir, unsigned int n)
2749 {
2750 smp_store_release(&dir->i_dir_seq, n + 2);
2751 preempt_enable_nested();
2752 }
2753
d_wait_lookup(struct dentry * dentry)2754 static void d_wait_lookup(struct dentry *dentry)
2755 {
2756 if (likely(d_in_lookup(dentry))) {
2757 dentry->d_flags |= DCACHE_LOOKUP_WAITERS;
2758 wait_var_event_spinlock(&dentry->d_flags,
2759 !d_in_lookup(dentry),
2760 &dentry->d_lock);
2761 }
2762 }
2763
d_alloc_parallel(struct dentry * parent,const struct qstr * name)2764 struct dentry *d_alloc_parallel(struct dentry *parent,
2765 const struct qstr *name)
2766 {
2767 unsigned int hash = name->hash;
2768 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2769 struct hlist_bl_node *node;
2770 struct dentry *new = __d_alloc(parent->d_sb, name);
2771 struct dentry *dentry;
2772 unsigned seq, r_seq, d_seq;
2773
2774 if (unlikely(!new))
2775 return ERR_PTR(-ENOMEM);
2776
2777 new->d_flags |= DCACHE_PAR_LOOKUP;
2778 spin_lock(&parent->d_lock);
2779 new->d_parent = dget_dlock(parent);
2780 hlist_add_head(&new->d_sib, &parent->d_children);
2781 if (parent->d_flags & DCACHE_DISCONNECTED)
2782 new->d_flags |= DCACHE_DISCONNECTED;
2783 spin_unlock(&parent->d_lock);
2784
2785 retry:
2786 seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2787 r_seq = read_seqbegin(&rename_lock);
2788 rcu_read_lock();
2789 dentry = __d_lookup_rcu(parent, name, &d_seq);
2790 if (unlikely(dentry)) {
2791 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2792 rcu_read_unlock();
2793 goto retry;
2794 }
2795 rcu_read_unlock();
2796 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2797 dput(dentry);
2798 goto retry;
2799 }
2800 dput(new);
2801 return dentry;
2802 }
2803 rcu_read_unlock();
2804 if (unlikely(read_seqretry(&rename_lock, r_seq)))
2805 goto retry;
2806
2807 if (unlikely(seq & 1))
2808 goto retry;
2809
2810 hlist_bl_lock(b);
2811 if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2812 hlist_bl_unlock(b);
2813 goto retry;
2814 }
2815 /*
2816 * No changes for the parent since the beginning of d_lookup().
2817 * Since all removals from the chain happen with hlist_bl_lock(),
2818 * any potential in-lookup matches are going to stay here until
2819 * we unlock the chain. All fields are stable in everything
2820 * we encounter.
2821 */
2822 hlist_bl_for_each_entry(dentry, node, b, d_in_lookup_hash) {
2823 if (dentry->d_name.hash != hash)
2824 continue;
2825 if (dentry->d_parent != parent)
2826 continue;
2827 if (!d_same_name(dentry, parent, name))
2828 continue;
2829 rcu_read_lock();
2830 hlist_bl_unlock(b);
2831 spin_lock(&dentry->d_lock);
2832 rcu_read_unlock();
2833 /* now we can try to grab a reference */
2834 if (unlikely(lockref_is_dead(&dentry->d_lockref))) {
2835 spin_unlock(&dentry->d_lock);
2836 goto retry;
2837 }
2838 /*
2839 * somebody is likely to be still doing lookup for it;
2840 * pin it and wait for them to finish
2841 */
2842 dget_dlock(dentry);
2843 d_wait_lookup(dentry);
2844 /*
2845 * it's not in-lookup anymore; in principle we should repeat
2846 * everything from dcache lookup, but it's likely to be what
2847 * d_lookup() would've found anyway. If it is, just return it;
2848 * otherwise we really have to repeat the whole thing.
2849 */
2850 if (unlikely(dentry->d_name.hash != hash))
2851 goto mismatch;
2852 if (unlikely(dentry->d_parent != parent))
2853 goto mismatch;
2854 if (unlikely(d_unhashed(dentry)))
2855 goto mismatch;
2856 if (unlikely(!d_same_name(dentry, parent, name)))
2857 goto mismatch;
2858 /* OK, it *is* a hashed match; return it */
2859 spin_unlock(&dentry->d_lock);
2860 dput(new);
2861 return dentry;
2862 }
2863 hlist_bl_add_head(&new->d_in_lookup_hash, b);
2864 hlist_bl_unlock(b);
2865 return new;
2866 mismatch:
2867 spin_unlock(&dentry->d_lock);
2868 dput(dentry);
2869 goto retry;
2870 }
2871 EXPORT_SYMBOL(d_alloc_parallel);
2872
2873 /*
2874 * Move dentry from in-lookup state to busy-negative one.
2875 *
2876 * From now on d_in_lookup(dentry) will return false and dentry is gone from
2877 * in-lookup hash.
2878 *
2879 * Anyone who had been waiting on it in d_alloc_parallel() is free to
2880 * proceed after that. Note that waking such waiters up is left to
2881 * the callers; PREEMPT_RT kernels can't have that wakeup done while
2882 * in write-side critical area for ->i_dir_seq, so it's done by calling
2883 * __d_wake_in_lookup_waiters() once it's safe to do so.
2884 *
2885 * Both __d_lookup_unhash() and __d_wake_in_lookup_waiters() should
2886 * be called within the same ->d_lock scope. PAR_LOOKUP is cleared
2887 * here, while LOOKUP_WAITERS (set by somebody finding dentry in
2888 * the in-lookup hash and setting down to wait) is checked and cleared
2889 * in __d_wake_in_lookup_waiters(). Both are gone by the end of
2890 * ->d_lock scope.
2891 */
__d_lookup_unhash(struct dentry * dentry)2892 static void __d_lookup_unhash(struct dentry *dentry)
2893 {
2894 struct hlist_bl_head *b;
2895
2896 lockdep_assert_held(&dentry->d_lock);
2897
2898 b = in_lookup_hash(dentry->d_parent, dentry->d_name.hash);
2899 hlist_bl_lock(b);
2900 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2901 __hlist_bl_del(&dentry->d_in_lookup_hash);
2902 hlist_bl_unlock(b);
2903 dentry->waiters = NULL;
2904 }
2905
__d_wake_in_lookup_waiters(struct dentry * dentry)2906 static inline void __d_wake_in_lookup_waiters(struct dentry *dentry)
2907 {
2908 if (dentry->d_flags & DCACHE_LOOKUP_WAITERS) {
2909 wake_up_var_locked(&dentry->d_flags, &dentry->d_lock);
2910 dentry->d_flags &= ~DCACHE_LOOKUP_WAITERS;
2911 }
2912 }
2913
__d_lookup_unhash_wake(struct dentry * dentry)2914 void __d_lookup_unhash_wake(struct dentry *dentry)
2915 {
2916 spin_lock(&dentry->d_lock);
2917 __d_lookup_unhash(dentry);
2918 __d_wake_in_lookup_waiters(dentry);
2919 spin_unlock(&dentry->d_lock);
2920 }
2921 EXPORT_SYMBOL(__d_lookup_unhash_wake);
2922
2923 /* inode->i_lock held if inode is non-NULL */
2924
__d_add(struct dentry * dentry,struct inode * inode,const struct dentry_operations * ops)2925 static inline void __d_add(struct dentry *dentry, struct inode *inode,
2926 const struct dentry_operations *ops)
2927 {
2928 struct inode *dir = NULL;
2929 unsigned n;
2930 spin_lock(&dentry->d_lock);
2931 if (unlikely(d_in_lookup(dentry))) {
2932 dir = dentry->d_parent->d_inode;
2933 n = start_dir_add(dir);
2934 __d_lookup_unhash(dentry);
2935 }
2936 if (unlikely(ops))
2937 d_set_d_op(dentry, ops);
2938 if (inode) {
2939 unsigned add_flags = d_flags_for_inode(inode);
2940 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
2941 raw_write_seqcount_begin(&dentry->d_seq);
2942 __d_set_inode_and_type(dentry, inode, add_flags);
2943 raw_write_seqcount_end(&dentry->d_seq);
2944 fsnotify_update_flags(dentry);
2945 }
2946 __d_rehash(dentry);
2947 if (dir) {
2948 end_dir_add(dir, n);
2949 __d_wake_in_lookup_waiters(dentry);
2950 }
2951 spin_unlock(&dentry->d_lock);
2952 if (inode)
2953 spin_unlock(&inode->i_lock);
2954 }
2955
2956 /**
2957 * d_add - add dentry to hash queues
2958 * @entry: dentry to add
2959 * @inode: The inode to attach to this dentry
2960 *
2961 * This adds the entry to the hash queues and initializes @inode.
2962 * The entry was actually filled in earlier during d_alloc().
2963 */
2964
d_add(struct dentry * entry,struct inode * inode)2965 void d_add(struct dentry *entry, struct inode *inode)
2966 {
2967 if (inode) {
2968 security_d_instantiate(entry, inode);
2969 spin_lock(&inode->i_lock);
2970 }
2971 __d_add(entry, inode, NULL);
2972 }
2973 EXPORT_SYMBOL(d_add);
2974
d_make_persistent(struct dentry * dentry,struct inode * inode)2975 struct dentry *d_make_persistent(struct dentry *dentry, struct inode *inode)
2976 {
2977 WARN_ON(d_really_is_positive(dentry));
2978 WARN_ON(!inode);
2979 security_d_instantiate(dentry, inode);
2980 spin_lock(&inode->i_lock);
2981 spin_lock(&dentry->d_lock);
2982 __d_instantiate(dentry, inode);
2983 dentry->d_flags |= DCACHE_PERSISTENT;
2984 dget_dlock(dentry);
2985 if (d_unhashed(dentry))
2986 __d_rehash(dentry);
2987 spin_unlock(&dentry->d_lock);
2988 spin_unlock(&inode->i_lock);
2989 return dentry;
2990 }
2991 EXPORT_SYMBOL(d_make_persistent);
2992
swap_names(struct dentry * dentry,struct dentry * target)2993 static void swap_names(struct dentry *dentry, struct dentry *target)
2994 {
2995 if (unlikely(dname_external(target))) {
2996 if (unlikely(dname_external(dentry))) {
2997 /*
2998 * Both external: swap the pointers
2999 */
3000 swap(target->__d_name.name, dentry->__d_name.name);
3001 } else {
3002 /*
3003 * dentry:internal, target:external. Steal target's
3004 * storage and make target internal.
3005 */
3006 dentry->__d_name.name = target->__d_name.name;
3007 target->d_shortname = dentry->d_shortname;
3008 target->__d_name.name = target->d_shortname.string;
3009 }
3010 } else {
3011 if (unlikely(dname_external(dentry))) {
3012 /*
3013 * dentry:external, target:internal. Give dentry's
3014 * storage to target and make dentry internal
3015 */
3016 target->__d_name.name = dentry->__d_name.name;
3017 dentry->d_shortname = target->d_shortname;
3018 dentry->__d_name.name = dentry->d_shortname.string;
3019 } else {
3020 /*
3021 * Both are internal.
3022 */
3023 for (int i = 0; i < DNAME_INLINE_WORDS; i++)
3024 swap(dentry->d_shortname.words[i],
3025 target->d_shortname.words[i]);
3026 }
3027 }
3028 swap(dentry->__d_name.hash_len, target->__d_name.hash_len);
3029 }
3030
copy_name(struct dentry * dentry,struct dentry * target)3031 static void copy_name(struct dentry *dentry, struct dentry *target)
3032 {
3033 struct external_name *old_name = NULL;
3034 if (unlikely(dname_external(dentry)))
3035 old_name = external_name(dentry);
3036 if (unlikely(dname_external(target))) {
3037 atomic_inc(&external_name(target)->count);
3038 dentry->__d_name = target->__d_name;
3039 } else {
3040 dentry->d_shortname = target->d_shortname;
3041 dentry->__d_name.name = dentry->d_shortname.string;
3042 dentry->__d_name.hash_len = target->__d_name.hash_len;
3043 }
3044 if (old_name && likely(atomic_dec_and_test(&old_name->count)))
3045 kfree_rcu(old_name, head);
3046 }
3047
3048 /*
3049 * __d_move - move a dentry
3050 * @dentry: entry to move
3051 * @target: new dentry
3052 * @exchange: exchange the two dentries
3053 *
3054 * Update the dcache to reflect the move of a file name. Negative dcache
3055 * entries should not be moved in this way. Caller must hold rename_lock, the
3056 * i_rwsem of the source and target directories (exclusively), and the sb->
3057 * s_vfs_rename_mutex if they differ. See lock_rename().
3058 */
__d_move(struct dentry * dentry,struct dentry * target,bool exchange)3059 static void __d_move(struct dentry *dentry, struct dentry *target,
3060 bool exchange)
3061 {
3062 struct dentry *old_parent, *p;
3063 struct inode *dir = NULL;
3064 unsigned n;
3065
3066 WARN_ON(!dentry->d_inode);
3067 if (WARN_ON(dentry == target))
3068 return;
3069
3070 BUG_ON(d_ancestor(target, dentry));
3071 old_parent = dentry->d_parent;
3072 p = d_ancestor(old_parent, target);
3073 if (IS_ROOT(dentry)) {
3074 BUG_ON(p);
3075 spin_lock(&target->d_parent->d_lock);
3076 } else if (!p) {
3077 /* target is not a descendent of dentry->d_parent */
3078 spin_lock(&target->d_parent->d_lock);
3079 spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
3080 } else {
3081 BUG_ON(p == dentry);
3082 spin_lock(&old_parent->d_lock);
3083 if (p != target)
3084 spin_lock_nested(&target->d_parent->d_lock,
3085 DENTRY_D_LOCK_NESTED);
3086 }
3087 spin_lock_nested(&dentry->d_lock, 2);
3088 spin_lock_nested(&target->d_lock, 3);
3089
3090 if (unlikely(d_in_lookup(target))) {
3091 dir = target->d_parent->d_inode;
3092 n = start_dir_add(dir);
3093 __d_lookup_unhash(target);
3094 }
3095
3096 write_seqcount_begin(&dentry->d_seq);
3097 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
3098
3099 /* unhash both */
3100 if (!d_unhashed(dentry))
3101 ___d_drop(dentry);
3102 if (!d_unhashed(target))
3103 ___d_drop(target);
3104
3105 /* ... and switch them in the tree */
3106 dentry->d_parent = target->d_parent;
3107 if (!exchange) {
3108 copy_name(dentry, target);
3109 target->d_hash.pprev = NULL;
3110 dentry->d_parent->d_lockref.count++;
3111 if (dentry != old_parent) /* wasn't IS_ROOT */
3112 WARN_ON(!--old_parent->d_lockref.count);
3113 } else {
3114 target->d_parent = old_parent;
3115 swap_names(dentry, target);
3116 if (!hlist_unhashed(&target->d_sib))
3117 __hlist_del(&target->d_sib);
3118 hlist_add_head(&target->d_sib, &target->d_parent->d_children);
3119 __d_rehash(target);
3120 fsnotify_update_flags(target);
3121 }
3122 if (!hlist_unhashed(&dentry->d_sib))
3123 __hlist_del(&dentry->d_sib);
3124 hlist_add_head(&dentry->d_sib, &dentry->d_parent->d_children);
3125 __d_rehash(dentry);
3126 fsnotify_update_flags(dentry);
3127 fscrypt_handle_d_move(dentry);
3128
3129 write_seqcount_end(&target->d_seq);
3130 write_seqcount_end(&dentry->d_seq);
3131
3132 if (dir) {
3133 end_dir_add(dir, n);
3134 __d_wake_in_lookup_waiters(target);
3135 }
3136 if (dentry->d_parent != old_parent)
3137 spin_unlock(&dentry->d_parent->d_lock);
3138 if (dentry != old_parent)
3139 spin_unlock(&old_parent->d_lock);
3140 spin_unlock(&target->d_lock);
3141 spin_unlock(&dentry->d_lock);
3142 }
3143
3144 /*
3145 * d_move - move a dentry
3146 * @dentry: entry to move
3147 * @target: new dentry
3148 *
3149 * Update the dcache to reflect the move of a file name. Negative
3150 * dcache entries should not be moved in this way. See the locking
3151 * requirements for __d_move.
3152 */
d_move(struct dentry * dentry,struct dentry * target)3153 void d_move(struct dentry *dentry, struct dentry *target)
3154 {
3155 write_seqlock(&rename_lock);
3156 __d_move(dentry, target, false);
3157 write_sequnlock(&rename_lock);
3158 }
3159 EXPORT_SYMBOL(d_move);
3160
3161 /*
3162 * d_exchange - exchange two dentries
3163 * @dentry1: first dentry
3164 * @dentry2: second dentry
3165 */
d_exchange(struct dentry * dentry1,struct dentry * dentry2)3166 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
3167 {
3168 write_seqlock(&rename_lock);
3169
3170 WARN_ON(!dentry1->d_inode);
3171 WARN_ON(!dentry2->d_inode);
3172 WARN_ON(IS_ROOT(dentry1));
3173 WARN_ON(IS_ROOT(dentry2));
3174
3175 __d_move(dentry1, dentry2, true);
3176
3177 write_sequnlock(&rename_lock);
3178 }
3179 EXPORT_SYMBOL(d_exchange);
3180
3181 /**
3182 * d_ancestor - search for an ancestor
3183 * @p1: ancestor dentry
3184 * @p2: child dentry
3185 *
3186 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
3187 * an ancestor of p2, else NULL.
3188 */
d_ancestor(struct dentry * p1,struct dentry * p2)3189 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
3190 {
3191 struct dentry *p;
3192
3193 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
3194 if (p->d_parent == p1)
3195 return p;
3196 }
3197 return NULL;
3198 }
3199
3200 /*
3201 * This helper attempts to cope with remotely renamed directories
3202 *
3203 * It assumes that the caller is already holding
3204 * dentry->d_parent->d_inode->i_rwsem, and rename_lock
3205 *
3206 * Note: If ever the locking in lock_rename() changes, then please
3207 * remember to update this too...
3208 */
__d_unalias(struct dentry * dentry,struct dentry * alias)3209 static int __d_unalias(struct dentry *dentry, struct dentry *alias)
3210 {
3211 struct mutex *m1 = NULL;
3212 struct rw_semaphore *m2 = NULL;
3213 int ret = -ESTALE;
3214
3215 /* If alias and dentry share a parent, then no extra locks required */
3216 if (alias->d_parent == dentry->d_parent)
3217 goto out_unalias;
3218
3219 /* See lock_rename() */
3220 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
3221 goto out_err;
3222 m1 = &dentry->d_sb->s_vfs_rename_mutex;
3223 if (!inode_trylock_shared(alias->d_parent->d_inode))
3224 goto out_err;
3225 m2 = &alias->d_parent->d_inode->i_rwsem;
3226 out_unalias:
3227 if (alias->d_op && alias->d_op->d_unalias_trylock &&
3228 !alias->d_op->d_unalias_trylock(alias))
3229 goto out_err;
3230 __d_move(alias, dentry, false);
3231 if (alias->d_op && alias->d_op->d_unalias_unlock)
3232 alias->d_op->d_unalias_unlock(alias);
3233 ret = 0;
3234 out_err:
3235 if (m2)
3236 up_read(m2);
3237 if (m1)
3238 mutex_unlock(m1);
3239 return ret;
3240 }
3241
d_splice_alias_ops(struct inode * inode,struct dentry * dentry,const struct dentry_operations * ops)3242 struct dentry *d_splice_alias_ops(struct inode *inode, struct dentry *dentry,
3243 const struct dentry_operations *ops)
3244 {
3245 if (IS_ERR(inode))
3246 return ERR_CAST(inode);
3247
3248 BUG_ON(!d_unhashed(dentry));
3249
3250 if (!inode)
3251 goto out;
3252
3253 security_d_instantiate(dentry, inode);
3254 spin_lock(&inode->i_lock);
3255 if (S_ISDIR(inode->i_mode)) {
3256 struct dentry *new = __d_find_dir_alias(inode);
3257 if (unlikely(new)) {
3258 /* The reference to new ensures it remains an alias */
3259 spin_unlock(&inode->i_lock);
3260 write_seqlock(&rename_lock);
3261 if (unlikely(d_ancestor(new, dentry))) {
3262 write_sequnlock(&rename_lock);
3263 dput(new);
3264 new = ERR_PTR(-ELOOP);
3265 pr_warn_ratelimited(
3266 "VFS: Lookup of '%s' in %s %s"
3267 " would have caused loop\n",
3268 dentry->d_name.name,
3269 inode->i_sb->s_type->name,
3270 inode->i_sb->s_id);
3271 } else if (!IS_ROOT(new)) {
3272 struct dentry *old_parent = dget(new->d_parent);
3273 int err = __d_unalias(dentry, new);
3274 write_sequnlock(&rename_lock);
3275 if (err) {
3276 dput(new);
3277 new = ERR_PTR(err);
3278 }
3279 dput(old_parent);
3280 } else {
3281 if (unlikely(!hlist_unhashed(&new->d_sib))) {
3282 // secondary root getting spliced
3283 spin_lock(&new->d_lock);
3284 unlink_secondary_root(new);
3285 spin_unlock(&new->d_lock);
3286 }
3287 __d_move(new, dentry, false);
3288 write_sequnlock(&rename_lock);
3289 }
3290 iput(inode);
3291 return new;
3292 }
3293 }
3294 out:
3295 __d_add(dentry, inode, ops);
3296 return NULL;
3297 }
3298
3299 /**
3300 * d_splice_alias - splice a disconnected dentry into the tree if one exists
3301 * @inode: the inode which may have a disconnected dentry
3302 * @dentry: a negative dentry which we want to point to the inode.
3303 *
3304 * If inode is a directory and has an IS_ROOT alias, then d_move that in
3305 * place of the given dentry and return it, else simply d_add the inode
3306 * to the dentry and return NULL.
3307 *
3308 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3309 * we should error out: directories can't have multiple aliases.
3310 *
3311 * This is needed in the lookup routine of any filesystem that is exportable
3312 * (via knfsd) so that we can build dcache paths to directories effectively.
3313 *
3314 * If a dentry was found and moved, then it is returned. Otherwise NULL
3315 * is returned. This matches the expected return value of ->lookup.
3316 *
3317 * Cluster filesystems may call this function with a negative, hashed dentry.
3318 * In that case, we know that the inode will be a regular file, and also this
3319 * will only occur during atomic_open. So we need to check for the dentry
3320 * being already hashed only in the final case.
3321 */
d_splice_alias(struct inode * inode,struct dentry * dentry)3322 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3323 {
3324 return d_splice_alias_ops(inode, dentry, NULL);
3325 }
3326 EXPORT_SYMBOL(d_splice_alias);
3327
3328 /*
3329 * Test whether new_dentry is a subdirectory of old_dentry.
3330 *
3331 * Trivially implemented using the dcache structure
3332 */
3333
3334 /**
3335 * is_subdir - is new dentry a subdirectory of old_dentry
3336 * @new_dentry: new dentry
3337 * @old_dentry: old dentry
3338 *
3339 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3340 * Returns false otherwise.
3341 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3342 */
3343
is_subdir(struct dentry * new_dentry,struct dentry * old_dentry)3344 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3345 {
3346 bool subdir;
3347 unsigned seq;
3348
3349 if (new_dentry == old_dentry)
3350 return true;
3351
3352 /* Access d_parent under rcu as d_move() may change it. */
3353 rcu_read_lock();
3354 seq = read_seqbegin(&rename_lock);
3355 subdir = d_ancestor(old_dentry, new_dentry);
3356 /* Try lockless once... */
3357 if (read_seqretry(&rename_lock, seq)) {
3358 /* ...else acquire lock for progress even on deep chains. */
3359 read_seqlock_excl(&rename_lock);
3360 subdir = d_ancestor(old_dentry, new_dentry);
3361 read_sequnlock_excl(&rename_lock);
3362 }
3363 rcu_read_unlock();
3364 return subdir;
3365 }
3366 EXPORT_SYMBOL(is_subdir);
3367
d_mark_tmpfile(struct file * file,struct inode * inode)3368 void d_mark_tmpfile(struct file *file, struct inode *inode)
3369 {
3370 struct dentry *dentry = file->f_path.dentry;
3371
3372 BUG_ON(dname_external(dentry) ||
3373 d_really_is_positive(dentry) ||
3374 !d_unlinked(dentry));
3375 spin_lock(&dentry->d_parent->d_lock);
3376 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3377 dentry->__d_name.len = sprintf(dentry->d_shortname.string, "#%llu",
3378 (unsigned long long)inode->i_ino);
3379 spin_unlock(&dentry->d_lock);
3380 spin_unlock(&dentry->d_parent->d_lock);
3381 }
3382 EXPORT_SYMBOL(d_mark_tmpfile);
3383
d_mark_tmpfile_name(struct file * file,const struct qstr * name)3384 int d_mark_tmpfile_name(struct file *file, const struct qstr *name)
3385 {
3386 struct dentry *dentry = file->f_path.dentry;
3387 char *dname = dentry->d_shortname.string;
3388
3389 if (unlikely(dname_external(dentry) ||
3390 d_really_is_positive(dentry) ||
3391 !d_unlinked(dentry)))
3392 return -EINVAL;
3393 if (unlikely(name->len > DNAME_INLINE_LEN - 1))
3394 return -ENAMETOOLONG;
3395
3396 spin_lock(&dentry->d_parent->d_lock);
3397 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3398 dentry->__d_name.len = name->len;
3399 memcpy(dname, name->name, name->len);
3400 dname[name->len] = '\0';
3401 spin_unlock(&dentry->d_lock);
3402 spin_unlock(&dentry->d_parent->d_lock);
3403 return 0;
3404 }
3405 EXPORT_SYMBOL(d_mark_tmpfile_name);
3406
d_tmpfile(struct file * file,struct inode * inode)3407 void d_tmpfile(struct file *file, struct inode *inode)
3408 {
3409 struct dentry *dentry = file->f_path.dentry;
3410
3411 inode_dec_link_count(inode);
3412 d_mark_tmpfile(file, inode);
3413 d_instantiate(dentry, inode);
3414 }
3415 EXPORT_SYMBOL(d_tmpfile);
3416
3417 /*
3418 * Obtain inode number of the parent dentry.
3419 */
d_parent_ino(struct dentry * dentry)3420 ino_t d_parent_ino(struct dentry *dentry)
3421 {
3422 struct dentry *parent;
3423 struct inode *iparent;
3424 unsigned seq;
3425 ino_t ret;
3426
3427 scoped_guard(rcu) {
3428 seq = raw_seqcount_begin(&dentry->d_seq);
3429 parent = READ_ONCE(dentry->d_parent);
3430 iparent = d_inode_rcu(parent);
3431 if (likely(iparent)) {
3432 ret = iparent->i_ino;
3433 if (!read_seqcount_retry(&dentry->d_seq, seq))
3434 return ret;
3435 }
3436 }
3437
3438 spin_lock(&dentry->d_lock);
3439 ret = dentry->d_parent->d_inode->i_ino;
3440 spin_unlock(&dentry->d_lock);
3441 return ret;
3442 }
3443 EXPORT_SYMBOL(d_parent_ino);
3444
3445 static __initdata unsigned long dhash_entries;
set_dhash_entries(char * str)3446 static int __init set_dhash_entries(char *str)
3447 {
3448 return kstrtoul(str, 0, &dhash_entries) == 0;
3449 }
3450 __setup("dhash_entries=", set_dhash_entries);
3451
dcache_init_early(void)3452 static void __init dcache_init_early(void)
3453 {
3454 /* If hashes are distributed across NUMA nodes, defer
3455 * hash allocation until vmalloc space is available.
3456 */
3457 if (hashdist)
3458 return;
3459
3460 dentry_hashtable =
3461 alloc_large_system_hash("Dentry cache",
3462 sizeof(struct hlist_bl_head),
3463 dhash_entries,
3464 13,
3465 HASH_EARLY | HASH_ZERO,
3466 &d_hash_shift,
3467 NULL,
3468 2,
3469 0);
3470 d_hash_shift = 32 - d_hash_shift;
3471
3472 runtime_const_init(shift, d_hash_shift);
3473 runtime_const_init(ptr, dentry_hashtable);
3474 }
3475
dcache_init(void)3476 static void __init dcache_init(void)
3477 {
3478 /*
3479 * A constructor could be added for stable state like the lists,
3480 * but it is probably not worth it because of the cache nature
3481 * of the dcache.
3482 */
3483 __dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3484 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_ACCOUNT,
3485 d_shortname.string);
3486 runtime_const_init(ptr, __dentry_cache);
3487
3488 /* Hash may have been set up in dcache_init_early */
3489 if (!hashdist)
3490 return;
3491
3492 dentry_hashtable =
3493 alloc_large_system_hash("Dentry cache",
3494 sizeof(struct hlist_bl_head),
3495 dhash_entries,
3496 13,
3497 HASH_ZERO,
3498 &d_hash_shift,
3499 NULL,
3500 2,
3501 0);
3502 d_hash_shift = 32 - d_hash_shift;
3503
3504 runtime_const_init(shift, d_hash_shift);
3505 runtime_const_init(ptr, dentry_hashtable);
3506 }
3507
vfs_caches_init_early(void)3508 void __init vfs_caches_init_early(void)
3509 {
3510 int i;
3511
3512 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3513 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3514
3515 dcache_init_early();
3516 inode_init_early();
3517 }
3518
vfs_caches_init(void)3519 void __init vfs_caches_init(void)
3520 {
3521 filename_init();
3522 dcache_init();
3523 inode_init();
3524 files_init();
3525 files_maxfiles_init();
3526 mnt_init();
3527 bdev_cache_init();
3528 chrdev_init();
3529 }
3530