blob: 67957f5b325c8ac85fd72fe730c2854f70f06988 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/dcache.c
3 *
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
8
9/*
10 * Notes on the allocation strategy:
11 *
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040021#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/hash.h>
25#include <linux/cache.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/file.h>
29#include <asm/uaccess.h>
30#include <linux/security.h>
31#include <linux/seqlock.h>
32#include <linux/swap.h>
33#include <linux/bootmem.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Frederic Weisbecker613afbf2009-07-16 15:44:29 +020035#include <linux/hardirq.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110036#include <linux/bit_spinlock.h>
37#include <linux/rculist_bl.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070038#include <linux/prefetch.h>
David Howellsdd179942011-08-16 15:31:30 +010039#include <linux/ratelimit.h>
Dave Chinnerf6041562013-08-28 10:18:00 +100040#include <linux/list_lru.h>
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -080041#include <linux/kasan.h>
42
David Howells07f3f052006-09-30 20:52:18 +020043#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050044#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Nick Piggin789680d2011-01-07 17:49:30 +110046/*
47 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110048 * dcache->d_inode->i_lock protects:
Al Viro946e51f2014-10-26 19:19:16 -040049 * - i_dentry, d_u.d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110050 * dcache_hash_bucket lock protects:
51 * - the dcache hash table
52 * s_anon bl list spinlock protects:
53 * - the s_anon list (see __d_drop)
Dave Chinner19156842013-08-28 10:17:55 +100054 * dentry->d_sb->s_dentry_lru_lock protects:
Nick Piggin23044502011-01-07 17:49:31 +110055 * - the dcache lru lists and counters
56 * d_lock protects:
57 * - d_flags
58 * - d_name
59 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110060 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110061 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110062 * - d_parent and d_subdirs
63 * - childrens' d_child and d_parent
Al Viro946e51f2014-10-26 19:19:16 -040064 * - d_u.d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110065 *
66 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110067 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110068 * dentry->d_lock
Dave Chinner19156842013-08-28 10:17:55 +100069 * dentry->d_sb->s_dentry_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110070 * dcache_hash_bucket lock
71 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110072 *
Nick Pigginda502952011-01-07 17:49:33 +110073 * If there is an ancestor relationship:
74 * dentry->d_parent->...->d_parent->d_lock
75 * ...
76 * dentry->d_parent->d_lock
77 * dentry->d_lock
78 *
79 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110080 * if (dentry1 < dentry2)
81 * dentry1->d_lock
82 * dentry2->d_lock
83 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080084int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
86
Al Viro74c3cbe2007-07-22 08:04:18 -040087__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Nick Piggin949854d2011-01-07 17:49:37 +110089EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Christoph Lametere18b8902006-12-06 20:33:20 -080091static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Linus Torvalds1da177e2005-04-16 15:20:36 -070093/*
94 * This is the single most critical data structure when it comes
95 * to the dcache: the hashtable for lookups. Somebody should try
96 * to make this good - I've just made it work.
97 *
98 * This hash-function tries to avoid losing too many bits of hash
99 * information, yet avoid using a prime hash-size or similar.
100 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800102static unsigned int d_hash_mask __read_mostly;
103static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700105static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100106
Linus Torvalds8387ff22016-06-10 07:51:30 -0700107static inline struct hlist_bl_head *d_hash(unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100108{
George Spelvin703b5fa2016-06-10 00:22:12 -0400109 return dentry_hashtable + (hash >> (32 - d_hash_shift));
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100110}
111
Al Viro94bdd652016-04-15 02:42:04 -0400112#define IN_LOOKUP_SHIFT 10
113static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
114
115static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
116 unsigned int hash)
117{
118 hash += (unsigned long) parent / L1_CACHE_BYTES;
119 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
120}
121
122
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123/* Statistics gathering. */
124struct dentry_stat_t dentry_stat = {
125 .age_limit = 45,
126};
127
Glauber Costa3942c072013-08-28 10:17:53 +1000128static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000129static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400130
131#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000132
133/*
134 * Here we resort to our own counters instead of using generic per-cpu counters
135 * for consistency with what the vfs inode code does. We are expected to harvest
136 * better code and performance by having our own specialized counters.
137 *
138 * Please note that the loop is done over all possible CPUs, not over all online
139 * CPUs. The reason for this is that we don't want to play games with CPUs going
140 * on and off. If one of them goes off, we will just keep their counters.
141 *
142 * glommer: See cffbc8a for details, and if you ever intend to change this,
143 * please update all vfs counters to match.
144 */
Glauber Costa3942c072013-08-28 10:17:53 +1000145static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100146{
147 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000148 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100149 for_each_possible_cpu(i)
150 sum += per_cpu(nr_dentry, i);
151 return sum < 0 ? 0 : sum;
152}
153
Dave Chinner62d36c72013-08-28 10:17:54 +1000154static long get_nr_dentry_unused(void)
155{
156 int i;
157 long sum = 0;
158 for_each_possible_cpu(i)
159 sum += per_cpu(nr_dentry_unused, i);
160 return sum < 0 ? 0 : sum;
161}
162
Joe Perches1f7e0612014-06-06 14:38:05 -0700163int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400164 size_t *lenp, loff_t *ppos)
165{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100166 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000167 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000168 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400169}
170#endif
171
Linus Torvalds5483f182012-03-04 15:51:42 -0800172/*
173 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
174 * The strings are both count bytes long, and count is non-zero.
175 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700176#ifdef CONFIG_DCACHE_WORD_ACCESS
177
178#include <asm/word-at-a-time.h>
179/*
180 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
181 * aligned allocation for this particular component. We don't
182 * strictly need the load_unaligned_zeropad() safety, but it
183 * doesn't hurt either.
184 *
185 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
186 * need the careful unaligned handling.
187 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700188static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800189{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800190 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800191
192 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700193 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700194 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800195 if (tcount < sizeof(unsigned long))
196 break;
197 if (unlikely(a != b))
198 return 1;
199 cs += sizeof(unsigned long);
200 ct += sizeof(unsigned long);
201 tcount -= sizeof(unsigned long);
202 if (!tcount)
203 return 0;
204 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000205 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800206 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700207}
208
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800209#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700210
Linus Torvalds94753db52012-05-10 12:19:19 -0700211static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700212{
Linus Torvalds5483f182012-03-04 15:51:42 -0800213 do {
214 if (*cs != *ct)
215 return 1;
216 cs++;
217 ct++;
218 tcount--;
219 } while (tcount);
220 return 0;
221}
222
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700223#endif
224
Linus Torvalds94753db52012-05-10 12:19:19 -0700225static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
226{
Linus Torvalds94753db52012-05-10 12:19:19 -0700227 /*
228 * Be careful about RCU walk racing with rename:
He Kuangae0a8432016-03-26 09:12:10 +0000229 * use 'lockless_dereference' to fetch the name pointer.
Linus Torvalds94753db52012-05-10 12:19:19 -0700230 *
231 * NOTE! Even if a rename will mean that the length
232 * was not loaded atomically, we don't care. The
233 * RCU walk will check the sequence count eventually,
234 * and catch it. And we won't overrun the buffer,
235 * because we're reading the name pointer atomically,
236 * and a dentry name is guaranteed to be properly
237 * terminated with a NUL byte.
238 *
239 * End result: even if 'len' is wrong, we'll exit
240 * early because the data cannot match (there can
241 * be no NUL in the ct/tcount data)
242 */
He Kuangae0a8432016-03-26 09:12:10 +0000243 const unsigned char *cs = lockless_dereference(dentry->d_name.name);
244
Linus Torvalds6326c712012-05-21 16:14:04 -0700245 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700246}
247
Al Viro8d85b482014-09-29 14:54:27 -0400248struct external_name {
249 union {
250 atomic_t count;
251 struct rcu_head head;
252 } u;
253 unsigned char name[];
254};
255
256static inline struct external_name *external_name(struct dentry *dentry)
257{
258 return container_of(dentry->d_name.name, struct external_name, name[0]);
259}
260
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400261static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400263 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
264
Al Viro8d85b482014-09-29 14:54:27 -0400265 kmem_cache_free(dentry_cache, dentry);
266}
267
268static void __d_free_external(struct rcu_head *head)
269{
270 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
Al Viro8d85b482014-09-29 14:54:27 -0400271 kfree(external_name(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 kmem_cache_free(dentry_cache, dentry);
273}
274
Al Viro810bb172014-10-12 12:45:37 -0400275static inline int dname_external(const struct dentry *dentry)
276{
277 return dentry->d_name.name != dentry->d_iname;
278}
279
Al Viroad25f112017-07-07 14:51:19 -0400280void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
281{
282 spin_lock(&dentry->d_lock);
283 if (unlikely(dname_external(dentry))) {
284 struct external_name *p = external_name(dentry);
285 atomic_inc(&p->u.count);
286 spin_unlock(&dentry->d_lock);
287 name->name = p->name;
288 } else {
289 memcpy(name->inline_name, dentry->d_iname, DNAME_INLINE_LEN);
290 spin_unlock(&dentry->d_lock);
291 name->name = name->inline_name;
292 }
293}
294EXPORT_SYMBOL(take_dentry_name_snapshot);
295
296void release_dentry_name_snapshot(struct name_snapshot *name)
297{
298 if (unlikely(name->name != name->inline_name)) {
299 struct external_name *p;
300 p = container_of(name->name, struct external_name, name[0]);
301 if (unlikely(atomic_dec_and_test(&p->u.count)))
302 kfree_rcu(p, u.head);
303 }
304}
305EXPORT_SYMBOL(release_dentry_name_snapshot);
306
David Howells4bf46a22015-03-05 14:09:22 +0000307static inline void __d_set_inode_and_type(struct dentry *dentry,
308 struct inode *inode,
309 unsigned type_flags)
310{
311 unsigned flags;
312
313 dentry->d_inode = inode;
David Howells4bf46a22015-03-05 14:09:22 +0000314 flags = READ_ONCE(dentry->d_flags);
315 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
316 flags |= type_flags;
317 WRITE_ONCE(dentry->d_flags, flags);
318}
319
David Howells4bf46a22015-03-05 14:09:22 +0000320static inline void __d_clear_type_and_inode(struct dentry *dentry)
321{
322 unsigned flags = READ_ONCE(dentry->d_flags);
323
324 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
325 WRITE_ONCE(dentry->d_flags, flags);
David Howells4bf46a22015-03-05 14:09:22 +0000326 dentry->d_inode = NULL;
327}
328
Al Virob4f03542014-04-29 23:40:14 -0400329static void dentry_free(struct dentry *dentry)
330{
Al Viro946e51f2014-10-26 19:19:16 -0400331 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400332 if (unlikely(dname_external(dentry))) {
333 struct external_name *p = external_name(dentry);
334 if (likely(atomic_dec_and_test(&p->u.count))) {
335 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
336 return;
337 }
338 }
Al Virob4f03542014-04-29 23:40:14 -0400339 /* if dentry was never visible to RCU, immediate free is OK */
340 if (!(dentry->d_flags & DCACHE_RCUACCESS))
341 __d_free(&dentry->d_u.d_rcu);
342 else
343 call_rcu(&dentry->d_u.d_rcu, __d_free);
344}
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346/*
347 * Release the dentry's inode, using the filesystem
Al Viro550dce02016-05-29 20:13:30 -0400348 * d_iput() operation if defined.
Nick Piggin31e6b012011-01-07 17:49:52 +1100349 */
350static void dentry_unlink_inode(struct dentry * dentry)
351 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100352 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100353{
354 struct inode *inode = dentry->d_inode;
Al Viro550dce02016-05-29 20:13:30 -0400355 bool hashed = !d_unhashed(dentry);
Al Viroa528aca2016-02-29 12:12:46 -0500356
Al Viro550dce02016-05-29 20:13:30 -0400357 if (hashed)
358 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +0000359 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400360 hlist_del_init(&dentry->d_u.d_alias);
Al Viro550dce02016-05-29 20:13:30 -0400361 if (hashed)
362 raw_write_seqcount_end(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100363 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100364 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100365 if (!inode->i_nlink)
366 fsnotify_inoderemove(inode);
367 if (dentry->d_op && dentry->d_op->d_iput)
368 dentry->d_op->d_iput(dentry, inode);
369 else
370 iput(inode);
371}
372
373/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400374 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
375 * is in use - which includes both the "real" per-superblock
376 * LRU list _and_ the DCACHE_SHRINK_LIST use.
377 *
378 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
379 * on the shrink list (ie not on the superblock LRU list).
380 *
381 * The per-cpu "nr_dentry_unused" counters are updated with
382 * the DCACHE_LRU_LIST bit.
383 *
384 * These helper functions make sure we always follow the
385 * rules. d_lock must be held by the caller.
386 */
387#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
388static void d_lru_add(struct dentry *dentry)
389{
390 D_FLAG_VERIFY(dentry, 0);
391 dentry->d_flags |= DCACHE_LRU_LIST;
392 this_cpu_inc(nr_dentry_unused);
393 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
394}
395
396static void d_lru_del(struct dentry *dentry)
397{
398 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
399 dentry->d_flags &= ~DCACHE_LRU_LIST;
400 this_cpu_dec(nr_dentry_unused);
401 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
402}
403
404static void d_shrink_del(struct dentry *dentry)
405{
406 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
407 list_del_init(&dentry->d_lru);
408 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
409 this_cpu_dec(nr_dentry_unused);
410}
411
412static void d_shrink_add(struct dentry *dentry, struct list_head *list)
413{
414 D_FLAG_VERIFY(dentry, 0);
415 list_add(&dentry->d_lru, list);
416 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
417 this_cpu_inc(nr_dentry_unused);
418}
419
420/*
421 * These can only be called under the global LRU lock, ie during the
422 * callback for freeing the LRU list. "isolate" removes it from the
423 * LRU lists entirely, while shrink_move moves it to the indicated
424 * private list.
425 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800426static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400427{
428 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
429 dentry->d_flags &= ~DCACHE_LRU_LIST;
430 this_cpu_dec(nr_dentry_unused);
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800431 list_lru_isolate(lru, &dentry->d_lru);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400432}
433
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800434static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
435 struct list_head *list)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400436{
437 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
438 dentry->d_flags |= DCACHE_SHRINK_LIST;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800439 list_lru_isolate_move(lru, &dentry->d_lru, list);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400440}
441
442/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000443 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700444 */
445static void dentry_lru_add(struct dentry *dentry)
446{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400447 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
448 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700449}
450
Miklos Szeredid52b9082007-05-08 00:23:46 -0700451/**
Nick Piggin789680d2011-01-07 17:49:30 +1100452 * d_drop - drop a dentry
453 * @dentry: dentry to drop
454 *
455 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
456 * be found through a VFS lookup any more. Note that this is different from
457 * deleting the dentry - d_delete will try to mark the dentry negative if
458 * possible, giving a successful _negative_ lookup, while d_drop will
459 * just make the cache lookup fail.
460 *
461 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
462 * reason (NFS timeouts or autofs deletes).
463 *
464 * __d_drop requires dentry->d_lock.
465 */
466void __d_drop(struct dentry *dentry)
467{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700468 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400469 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400470 /*
471 * Hashed dentries are normally on the dentry hashtable,
472 * with the exception of those newly allocated by
473 * d_obtain_alias, which are always IS_ROOT:
474 */
475 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400476 b = &dentry->d_sb->s_anon;
477 else
Linus Torvalds8387ff22016-06-10 07:51:30 -0700478 b = d_hash(dentry->d_name.hash);
Al Virob61625d2013-10-04 11:09:01 -0400479
480 hlist_bl_lock(b);
481 __hlist_bl_del(&dentry->d_hash);
482 dentry->d_hash.pprev = NULL;
483 hlist_bl_unlock(b);
Al Virod6141462016-07-28 13:05:50 -0400484 /* After this call, in-progress rcu-walk path lookup will fail. */
485 write_seqcount_invalidate(&dentry->d_seq);
Nick Piggin789680d2011-01-07 17:49:30 +1100486 }
487}
488EXPORT_SYMBOL(__d_drop);
489
490void d_drop(struct dentry *dentry)
491{
Nick Piggin789680d2011-01-07 17:49:30 +1100492 spin_lock(&dentry->d_lock);
493 __d_drop(dentry);
494 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100495}
496EXPORT_SYMBOL(d_drop);
497
Al Viroba65dc52016-06-10 11:32:47 -0400498static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
499{
500 struct dentry *next;
501 /*
502 * Inform d_walk() and shrink_dentry_list() that we are no longer
503 * attached to the dentry tree
504 */
505 dentry->d_flags |= DCACHE_DENTRY_KILLED;
506 if (unlikely(list_empty(&dentry->d_child)))
507 return;
508 __list_del_entry(&dentry->d_child);
509 /*
510 * Cursors can move around the list of children. While we'd been
511 * a normal list member, it didn't matter - ->d_child.next would've
512 * been updated. However, from now on it won't be and for the
513 * things like d_walk() it might end up with a nasty surprise.
514 * Normally d_walk() doesn't care about cursors moving around -
515 * ->d_lock on parent prevents that and since a cursor has no children
516 * of its own, we get through it without ever unlocking the parent.
517 * There is one exception, though - if we ascend from a child that
518 * gets killed as soon as we unlock it, the next sibling is found
519 * using the value left in its ->d_child.next. And if _that_
520 * pointed to a cursor, and cursor got moved (e.g. by lseek())
521 * before d_walk() regains parent->d_lock, we'll end up skipping
522 * everything the cursor had been moved past.
523 *
524 * Solution: make sure that the pointer left behind in ->d_child.next
525 * points to something that won't be moving around. I.e. skip the
526 * cursors.
527 */
528 while (dentry->d_child.next != &parent->d_subdirs) {
529 next = list_entry(dentry->d_child.next, struct dentry, d_child);
530 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
531 break;
532 dentry->d_child.next = next->d_child.next;
533 }
534}
535
Al Viroe55fd012014-05-28 13:51:12 -0400536static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100537{
Al Viro41edf272014-05-01 10:30:00 -0400538 struct dentry *parent = NULL;
539 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400540 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100541 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100542
Linus Torvalds0d984392013-09-08 13:46:52 -0700543 /*
544 * The dentry is now unrecoverably dead to the world.
545 */
546 lockref_mark_dead(&dentry->d_lockref);
547
Sage Weilf0023bc2011-10-28 10:02:42 -0700548 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700549 * inform the fs via d_prune that this dentry is about to be
550 * unhashed and destroyed.
551 */
Al Viro29266202014-05-30 11:39:02 -0400552 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800553 dentry->d_op->d_prune(dentry);
554
Al Viro01b60352014-04-29 23:42:52 -0400555 if (dentry->d_flags & DCACHE_LRU_LIST) {
556 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
557 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400558 }
Nick Piggin77812a12011-01-07 17:49:48 +1100559 /* if it was on the hash then remove it */
560 __d_drop(dentry);
Al Viroba65dc52016-06-10 11:32:47 -0400561 dentry_unlist(dentry, parent);
Al Viro03b3b882014-04-29 15:45:28 -0400562 if (parent)
563 spin_unlock(&parent->d_lock);
Al Viro550dce02016-05-29 20:13:30 -0400564 if (dentry->d_inode)
565 dentry_unlink_inode(dentry);
566 else
567 spin_unlock(&dentry->d_lock);
Al Viro03b3b882014-04-29 15:45:28 -0400568 this_cpu_dec(nr_dentry);
569 if (dentry->d_op && dentry->d_op->d_release)
570 dentry->d_op->d_release(dentry);
571
Al Viro41edf272014-05-01 10:30:00 -0400572 spin_lock(&dentry->d_lock);
573 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
574 dentry->d_flags |= DCACHE_MAY_FREE;
575 can_free = false;
576 }
577 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400578 if (likely(can_free))
579 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400580}
581
582/*
583 * Finish off a dentry we've decided to kill.
584 * dentry->d_lock must be held, returns with it unlocked.
585 * If ref is non-zero, then decrement the refcount too.
586 * Returns dentry requiring refcount drop, or NULL if we're done.
587 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400588static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400589 __releases(dentry->d_lock)
590{
591 struct inode *inode = dentry->d_inode;
592 struct dentry *parent = NULL;
593
594 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
595 goto failed;
596
597 if (!IS_ROOT(dentry)) {
598 parent = dentry->d_parent;
599 if (unlikely(!spin_trylock(&parent->d_lock))) {
600 if (inode)
601 spin_unlock(&inode->i_lock);
602 goto failed;
603 }
604 }
605
606 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400607 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400608
609failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400610 spin_unlock(&dentry->d_lock);
Al Viroe55fd012014-05-28 13:51:12 -0400611 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100612}
613
Al Viro046b9612014-05-29 08:54:52 -0400614static inline struct dentry *lock_parent(struct dentry *dentry)
615{
616 struct dentry *parent = dentry->d_parent;
617 if (IS_ROOT(dentry))
618 return NULL;
Linus Torvalds360f5472015-01-09 15:19:03 -0800619 if (unlikely(dentry->d_lockref.count < 0))
Al Viroc2338f22014-06-12 00:29:13 -0400620 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400621 if (likely(spin_trylock(&parent->d_lock)))
622 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400623 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400624 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400625again:
626 parent = ACCESS_ONCE(dentry->d_parent);
627 spin_lock(&parent->d_lock);
628 /*
629 * We can't blindly lock dentry until we are sure
630 * that we won't violate the locking order.
631 * Any changes of dentry->d_parent must have
632 * been done with parent->d_lock held, so
633 * spin_lock() above is enough of a barrier
634 * for checking if it's still our child.
635 */
636 if (unlikely(parent != dentry->d_parent)) {
637 spin_unlock(&parent->d_lock);
638 goto again;
639 }
640 rcu_read_unlock();
641 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700642 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400643 else
644 parent = NULL;
645 return parent;
646}
647
Linus Torvalds360f5472015-01-09 15:19:03 -0800648/*
649 * Try to do a lockless dput(), and return whether that was successful.
650 *
651 * If unsuccessful, we return false, having already taken the dentry lock.
652 *
653 * The caller needs to hold the RCU read lock, so that the dentry is
654 * guaranteed to stay around even if the refcount goes down to zero!
655 */
656static inline bool fast_dput(struct dentry *dentry)
657{
658 int ret;
659 unsigned int d_flags;
660
661 /*
662 * If we have a d_op->d_delete() operation, we sould not
Al Viro75a6f822015-07-08 02:42:38 +0100663 * let the dentry count go to zero, so use "put_or_lock".
Linus Torvalds360f5472015-01-09 15:19:03 -0800664 */
665 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
666 return lockref_put_or_lock(&dentry->d_lockref);
667
668 /*
669 * .. otherwise, we can try to just decrement the
670 * lockref optimistically.
671 */
672 ret = lockref_put_return(&dentry->d_lockref);
673
674 /*
675 * If the lockref_put_return() failed due to the lock being held
676 * by somebody else, the fast path has failed. We will need to
677 * get the lock, and then check the count again.
678 */
679 if (unlikely(ret < 0)) {
680 spin_lock(&dentry->d_lock);
681 if (dentry->d_lockref.count > 1) {
682 dentry->d_lockref.count--;
683 spin_unlock(&dentry->d_lock);
684 return 1;
685 }
686 return 0;
687 }
688
689 /*
690 * If we weren't the last ref, we're done.
691 */
692 if (ret)
693 return 1;
694
695 /*
696 * Careful, careful. The reference count went down
697 * to zero, but we don't hold the dentry lock, so
698 * somebody else could get it again, and do another
699 * dput(), and we need to not race with that.
700 *
701 * However, there is a very special and common case
702 * where we don't care, because there is nothing to
703 * do: the dentry is still hashed, it does not have
704 * a 'delete' op, and it's referenced and already on
705 * the LRU list.
706 *
707 * NOTE! Since we aren't locked, these values are
708 * not "stable". However, it is sufficient that at
709 * some point after we dropped the reference the
710 * dentry was hashed and the flags had the proper
711 * value. Other dentry users may have re-gotten
712 * a reference to the dentry and change that, but
713 * our work is done - we can leave the dentry
714 * around with a zero refcount.
715 */
716 smp_rmb();
717 d_flags = ACCESS_ONCE(dentry->d_flags);
Al Viro75a6f822015-07-08 02:42:38 +0100718 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
Linus Torvalds360f5472015-01-09 15:19:03 -0800719
720 /* Nothing to do? Dropping the reference was all we needed? */
721 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
722 return 1;
723
724 /*
725 * Not the fast normal case? Get the lock. We've already decremented
726 * the refcount, but we'll need to re-check the situation after
727 * getting the lock.
728 */
729 spin_lock(&dentry->d_lock);
730
731 /*
732 * Did somebody else grab a reference to it in the meantime, and
733 * we're no longer the last user after all? Alternatively, somebody
734 * else could have killed it and marked it dead. Either way, we
735 * don't need to do anything else.
736 */
737 if (dentry->d_lockref.count) {
738 spin_unlock(&dentry->d_lock);
739 return 1;
740 }
741
742 /*
743 * Re-get the reference we optimistically dropped. We hold the
744 * lock, and we just tested that it was zero, so we can just
745 * set it to 1.
746 */
747 dentry->d_lockref.count = 1;
748 return 0;
749}
750
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752/*
753 * This is dput
754 *
755 * This is complicated by the fact that we do not want to put
756 * dentries that are no longer on any hash chain on the unused
757 * list: we'd much rather just get rid of them immediately.
758 *
759 * However, that implies that we have to traverse the dentry
760 * tree upwards to the parents which might _also_ now be
761 * scheduled for deletion (it may have been only waiting for
762 * its last child to go away).
763 *
764 * This tail recursion is done by hand as we don't want to depend
765 * on the compiler to always get this right (gcc generally doesn't).
766 * Real recursion would eat up our stack space.
767 */
768
769/*
770 * dput - release a dentry
771 * @dentry: dentry to release
772 *
773 * Release a dentry. This will drop the usage count and if appropriate
774 * call the dentry unlink method as well as removing it from the queues and
775 * releasing its resources. If the parent dentries were scheduled for release
776 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778void dput(struct dentry *dentry)
779{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700780 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 return;
782
783repeat:
Wei Fang47be6182016-07-06 11:32:20 +0800784 might_sleep();
785
Linus Torvalds360f5472015-01-09 15:19:03 -0800786 rcu_read_lock();
787 if (likely(fast_dput(dentry))) {
788 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 return;
Linus Torvalds360f5472015-01-09 15:19:03 -0800790 }
791
792 /* Slow case: now with the dentry lock held */
793 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
Al Viro85c7f812016-04-14 19:52:13 -0400795 WARN_ON(d_in_lookup(dentry));
796
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700797 /* Unreachable? Get rid of it */
798 if (unlikely(d_unhashed(dentry)))
799 goto kill_it;
800
Al Viro75a6f822015-07-08 02:42:38 +0100801 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
802 goto kill_it;
803
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700804 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100806 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 }
Nick Piggin265ac902010-10-10 05:36:24 -0400808
Linus Torvalds358eec12013-10-31 15:43:02 -0700809 if (!(dentry->d_flags & DCACHE_REFERENCED))
810 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400811 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400812
Waiman Long98474232013-08-28 18:24:59 -0700813 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100814 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 return;
816
Miklos Szeredid52b9082007-05-08 00:23:46 -0700817kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400818 dentry = dentry_kill(dentry);
Wei Fang47be6182016-07-06 11:32:20 +0800819 if (dentry) {
820 cond_resched();
Miklos Szeredid52b9082007-05-08 00:23:46 -0700821 goto repeat;
Wei Fang47be6182016-07-06 11:32:20 +0800822 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700824EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100827/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100828static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829{
Waiman Long98474232013-08-28 18:24:59 -0700830 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831}
832
Nick Piggindc0474b2011-01-07 17:49:43 +1100833static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100834{
Waiman Long98474232013-08-28 18:24:59 -0700835 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100836}
837
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100838struct dentry *dget_parent(struct dentry *dentry)
839{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700840 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100841 struct dentry *ret;
842
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700843 /*
844 * Do optimistic parent lookup without any
845 * locking.
846 */
847 rcu_read_lock();
848 ret = ACCESS_ONCE(dentry->d_parent);
849 gotref = lockref_get_not_zero(&ret->d_lockref);
850 rcu_read_unlock();
851 if (likely(gotref)) {
852 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
853 return ret;
854 dput(ret);
855 }
856
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100857repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100858 /*
859 * Don't need rcu_dereference because we re-check it was correct under
860 * the lock.
861 */
862 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100863 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100864 spin_lock(&ret->d_lock);
865 if (unlikely(ret != dentry->d_parent)) {
866 spin_unlock(&ret->d_lock);
867 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100868 goto repeat;
869 }
Nick Piggina734eb42011-01-07 17:49:44 +1100870 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700871 BUG_ON(!ret->d_lockref.count);
872 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100873 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100874 return ret;
875}
876EXPORT_SYMBOL(dget_parent);
877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878/**
879 * d_find_alias - grab a hashed alias of inode
880 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 *
882 * If inode has a hashed alias, or is a directory and has any alias,
883 * acquire the reference to alias and return it. Otherwise return NULL.
884 * Notice that if inode is a directory there can be only one alias and
885 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800886 * of a filesystem, or if the directory was renamed and d_revalidate
887 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700889 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500890 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500892static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893{
Nick Pigginda502952011-01-07 17:49:33 +1100894 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895
Nick Pigginda502952011-01-07 17:49:33 +1100896again:
897 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400898 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100899 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700901 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100902 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500904 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100905 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100906 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 return alias;
908 }
909 }
Nick Pigginda502952011-01-07 17:49:33 +1100910 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 }
Nick Pigginda502952011-01-07 17:49:33 +1100912 if (discon_alias) {
913 alias = discon_alias;
914 spin_lock(&alias->d_lock);
915 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500916 __dget_dlock(alias);
917 spin_unlock(&alias->d_lock);
918 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100919 }
920 spin_unlock(&alias->d_lock);
921 goto again;
922 }
923 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924}
925
Nick Pigginda502952011-01-07 17:49:33 +1100926struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927{
David Howells214fda12006-03-25 03:06:36 -0800928 struct dentry *de = NULL;
929
Al Virob3d9b7a2012-06-09 13:51:19 -0400930 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100931 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500932 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100933 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800934 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 return de;
936}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700937EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
939/*
940 * Try to kill dentries associated with this inode.
941 * WARNING: you must own a reference to inode.
942 */
943void d_prune_aliases(struct inode *inode)
944{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700945 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100947 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400948 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700950 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400951 struct dentry *parent = lock_parent(dentry);
952 if (likely(!dentry->d_lockref.count)) {
953 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800954 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400955 goto restart;
956 }
957 if (parent)
958 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 }
960 spin_unlock(&dentry->d_lock);
961 }
Nick Piggin873feea2011-01-07 17:50:06 +1100962 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700964EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400966static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967{
Al Viro5c47e6d2014-04-29 16:13:18 -0400968 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700969
Miklos Szeredi60942f22014-05-02 15:38:39 -0400970 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400971 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400972 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100973 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400974 parent = lock_parent(dentry);
975
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000977 * The dispose list is isolated and dentries are not accounted
978 * to the LRU here, so we can simply remove it from the list
979 * here regardless of whether it is referenced or not.
980 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400981 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000982
983 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000985 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800987 if (dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700988 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400989 if (parent)
990 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 continue;
992 }
Nick Piggin77812a12011-01-07 17:49:48 +1100993
Al Viro64fd72e2014-05-28 09:48:44 -0400994
995 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
996 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
997 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400998 if (parent)
999 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -04001000 if (can_free)
1001 dentry_free(dentry);
1002 continue;
1003 }
1004
Al Viroff2fde92014-05-28 13:59:13 -04001005 inode = dentry->d_inode;
1006 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001007 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001008 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -04001009 if (parent)
1010 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -04001011 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001012 }
Al Viroff2fde92014-05-28 13:59:13 -04001013
Al Viroff2fde92014-05-28 13:59:13 -04001014 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -04001015
Al Viro5c47e6d2014-04-29 16:13:18 -04001016 /*
1017 * We need to prune ancestors too. This is necessary to prevent
1018 * quadratic behavior of shrink_dcache_parent(), but is also
1019 * expected to be beneficial in reducing dentry cache
1020 * fragmentation.
1021 */
1022 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -04001023 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1024 parent = lock_parent(dentry);
1025 if (dentry->d_lockref.count != 1) {
1026 dentry->d_lockref.count--;
1027 spin_unlock(&dentry->d_lock);
1028 if (parent)
1029 spin_unlock(&parent->d_lock);
1030 break;
1031 }
1032 inode = dentry->d_inode; /* can't be NULL */
1033 if (unlikely(!spin_trylock(&inode->i_lock))) {
1034 spin_unlock(&dentry->d_lock);
1035 if (parent)
1036 spin_unlock(&parent->d_lock);
1037 cpu_relax();
1038 continue;
1039 }
1040 __dentry_kill(dentry);
1041 dentry = parent;
1042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001044}
1045
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001046static enum lru_status dentry_lru_isolate(struct list_head *item,
1047 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +10001048{
1049 struct list_head *freeable = arg;
1050 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1051
1052
1053 /*
1054 * we are inverting the lru lock/dentry->d_lock here,
1055 * so use a trylock. If we fail to get the lock, just skip
1056 * it
1057 */
1058 if (!spin_trylock(&dentry->d_lock))
1059 return LRU_SKIP;
1060
1061 /*
1062 * Referenced dentries are still in use. If they have active
1063 * counts, just remove them from the LRU. Otherwise give them
1064 * another pass through the LRU.
1065 */
1066 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001067 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001068 spin_unlock(&dentry->d_lock);
1069 return LRU_REMOVED;
1070 }
1071
1072 if (dentry->d_flags & DCACHE_REFERENCED) {
1073 dentry->d_flags &= ~DCACHE_REFERENCED;
1074 spin_unlock(&dentry->d_lock);
1075
1076 /*
1077 * The list move itself will be made by the common LRU code. At
1078 * this point, we've dropped the dentry->d_lock but keep the
1079 * lru lock. This is safe to do, since every list movement is
1080 * protected by the lru lock even if both locks are held.
1081 *
1082 * This is guaranteed by the fact that all LRU management
1083 * functions are intermediated by the LRU API calls like
1084 * list_lru_add and list_lru_del. List movement in this file
1085 * only ever occur through this functions or through callbacks
1086 * like this one, that are called from the LRU API.
1087 *
1088 * The only exceptions to this are functions like
1089 * shrink_dentry_list, and code that first checks for the
1090 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1091 * operating only with stack provided lists after they are
1092 * properly isolated from the main list. It is thus, always a
1093 * local access.
1094 */
1095 return LRU_ROTATE;
1096 }
1097
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001098 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001099 spin_unlock(&dentry->d_lock);
1100
1101 return LRU_REMOVED;
1102}
1103
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001104/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001105 * prune_dcache_sb - shrink the dcache
1106 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001107 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001108 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001109 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1110 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001111 * function.
1112 *
1113 * This function may fail to free any resources if all the dentries are in
1114 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001115 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001116long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001117{
Dave Chinnerf6041562013-08-28 10:18:00 +10001118 LIST_HEAD(dispose);
1119 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001120
Vladimir Davydov503c3582015-02-12 14:58:47 -08001121 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1122 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001123 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001124 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125}
1126
Glauber Costa4e717f52013-08-28 10:18:03 +10001127static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001128 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001129{
Glauber Costa4e717f52013-08-28 10:18:03 +10001130 struct list_head *freeable = arg;
1131 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001132
Glauber Costa4e717f52013-08-28 10:18:03 +10001133 /*
1134 * we are inverting the lru lock/dentry->d_lock here,
1135 * so use a trylock. If we fail to get the lock, just skip
1136 * it
1137 */
1138 if (!spin_trylock(&dentry->d_lock))
1139 return LRU_SKIP;
1140
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001141 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001142 spin_unlock(&dentry->d_lock);
1143
1144 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001145}
1146
Glauber Costa4e717f52013-08-28 10:18:03 +10001147
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001148/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 * shrink_dcache_sb - shrink dcache for a superblock
1150 * @sb: superblock
1151 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001152 * Shrink the dcache for the specified super block. This is used to free
1153 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001155void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156{
Glauber Costa4e717f52013-08-28 10:18:03 +10001157 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001158
Glauber Costa4e717f52013-08-28 10:18:03 +10001159 do {
1160 LIST_HEAD(dispose);
1161
1162 freed = list_lru_walk(&sb->s_dentry_lru,
Sahitya Tummalaa9aa6522a2017-07-10 15:50:00 -07001163 dentry_lru_isolate_shrink, &dispose, 1024);
Glauber Costa4e717f52013-08-28 10:18:03 +10001164
1165 this_cpu_sub(nr_dentry_unused, freed);
1166 shrink_dentry_list(&dispose);
Sahitya Tummalaa9aa6522a2017-07-10 15:50:00 -07001167 cond_resched();
1168 } while (list_lru_count(&sb->s_dentry_lru) > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001170EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001173 * enum d_walk_ret - action to talke during tree walk
1174 * @D_WALK_CONTINUE: contrinue walk
1175 * @D_WALK_QUIT: quit walk
1176 * @D_WALK_NORETRY: quit when retry is needed
1177 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001179enum d_walk_ret {
1180 D_WALK_CONTINUE,
1181 D_WALK_QUIT,
1182 D_WALK_NORETRY,
1183 D_WALK_SKIP,
1184};
1185
1186/**
1187 * d_walk - walk the dentry tree
1188 * @parent: start of walk
1189 * @data: data passed to @enter() and @finish()
1190 * @enter: callback when first entering the dentry
1191 * @finish: callback when successfully finished the walk
1192 *
1193 * The @enter() and @finish() callbacks are called with d_lock held.
1194 */
1195static void d_walk(struct dentry *parent, void *data,
1196 enum d_walk_ret (*enter)(void *, struct dentry *),
1197 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198{
Nick Piggin949854d2011-01-07 17:49:37 +11001199 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001201 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001202 enum d_walk_ret ret;
1203 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001204
Nick Piggin58db63d2011-01-07 17:49:39 +11001205again:
Al Viro48f5ec22013-09-09 15:22:25 -04001206 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001207 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001208 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001209
1210 ret = enter(data, this_parent);
1211 switch (ret) {
1212 case D_WALK_CONTINUE:
1213 break;
1214 case D_WALK_QUIT:
1215 case D_WALK_SKIP:
1216 goto out_unlock;
1217 case D_WALK_NORETRY:
1218 retry = false;
1219 break;
1220 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221repeat:
1222 next = this_parent->d_subdirs.next;
1223resume:
1224 while (next != &this_parent->d_subdirs) {
1225 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001226 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001228
Al Viroba65dc52016-06-10 11:32:47 -04001229 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1230 continue;
1231
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001232 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001233
1234 ret = enter(data, dentry);
1235 switch (ret) {
1236 case D_WALK_CONTINUE:
1237 break;
1238 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001239 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001240 goto out_unlock;
1241 case D_WALK_NORETRY:
1242 retry = false;
1243 break;
1244 case D_WALK_SKIP:
1245 spin_unlock(&dentry->d_lock);
1246 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001247 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001248
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001250 spin_unlock(&this_parent->d_lock);
1251 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001253 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 goto repeat;
1255 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001256 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 }
1258 /*
1259 * All done at this level ... ascend and resume the search.
1260 */
Al Viroca5358e2014-10-26 19:31:10 -04001261 rcu_read_lock();
1262ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001264 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001265 this_parent = child->d_parent;
1266
Al Viro31dec132013-10-25 17:04:27 -04001267 spin_unlock(&child->d_lock);
1268 spin_lock(&this_parent->d_lock);
1269
Al Viroca5358e2014-10-26 19:31:10 -04001270 /* might go back up the wrong parent if we have had a rename. */
1271 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001272 goto rename_retry;
Al Viro21591842015-05-28 23:09:19 -04001273 /* go into the first sibling still alive */
1274 do {
1275 next = child->d_child.next;
Al Viroca5358e2014-10-26 19:31:10 -04001276 if (next == &this_parent->d_subdirs)
1277 goto ascend;
1278 child = list_entry(next, struct dentry, d_child);
Al Viro21591842015-05-28 23:09:19 -04001279 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
Al Viro31dec132013-10-25 17:04:27 -04001280 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 goto resume;
1282 }
Al Viroca5358e2014-10-26 19:31:10 -04001283 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001284 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001285 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001286 if (finish)
1287 finish(data);
1288
1289out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001290 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001291 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001292 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001293
1294rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001295 spin_unlock(&this_parent->d_lock);
1296 rcu_read_unlock();
1297 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001298 if (!retry)
1299 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001300 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001301 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001303
1304/*
1305 * Search for at least 1 mount point in the dentry's subdirs.
1306 * We descend to the next level whenever the d_subdirs
1307 * list is non-empty and continue searching.
1308 */
1309
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001310static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1311{
1312 int *ret = data;
1313 if (d_mountpoint(dentry)) {
1314 *ret = 1;
1315 return D_WALK_QUIT;
1316 }
1317 return D_WALK_CONTINUE;
1318}
1319
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001320/**
1321 * have_submounts - check for mounts over a dentry
1322 * @parent: dentry to check.
1323 *
1324 * Return true if the parent or its subdirectories contain
1325 * a mount point
1326 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001327int have_submounts(struct dentry *parent)
1328{
1329 int ret = 0;
1330
1331 d_walk(parent, &ret, check_mount, NULL);
1332
1333 return ret;
1334}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001335EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
1337/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001338 * Called by mount code to set a mountpoint and check if the mountpoint is
1339 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1340 * subtree can become unreachable).
1341 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001342 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001343 * this reason take rename_lock and d_lock on dentry and ancestors.
1344 */
1345int d_set_mounted(struct dentry *dentry)
1346{
1347 struct dentry *p;
1348 int ret = -ENOENT;
1349 write_seqlock(&rename_lock);
1350 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001351 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001352 spin_lock(&p->d_lock);
1353 if (unlikely(d_unhashed(p))) {
1354 spin_unlock(&p->d_lock);
1355 goto out;
1356 }
1357 spin_unlock(&p->d_lock);
1358 }
1359 spin_lock(&dentry->d_lock);
1360 if (!d_unlinked(dentry)) {
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +13001361 ret = -EBUSY;
1362 if (!d_mountpoint(dentry)) {
1363 dentry->d_flags |= DCACHE_MOUNTED;
1364 ret = 0;
1365 }
Miklos Szeredieed81002013-09-05 14:39:11 +02001366 }
1367 spin_unlock(&dentry->d_lock);
1368out:
1369 write_sequnlock(&rename_lock);
1370 return ret;
1371}
1372
1373/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001374 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 * and move any unused dentries to the end of the unused
1376 * list for prune_dcache(). We descend to the next level
1377 * whenever the d_subdirs list is non-empty and continue
1378 * searching.
1379 *
1380 * It returns zero iff there are no unused children,
1381 * otherwise it returns the number of children moved to
1382 * the end of the unused list. This may not be the total
1383 * number of unused children, because select_parent can
1384 * drop the lock and return early due to latency
1385 * constraints.
1386 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001387
1388struct select_data {
1389 struct dentry *start;
1390 struct list_head dispose;
1391 int found;
1392};
1393
1394static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001396 struct select_data *data = _data;
1397 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001399 if (data->start == dentry)
1400 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
Al Virofe915222014-05-03 00:02:25 -04001402 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001403 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001404 } else {
1405 if (dentry->d_flags & DCACHE_LRU_LIST)
1406 d_lru_del(dentry);
1407 if (!dentry->d_lockref.count) {
1408 d_shrink_add(dentry, &data->dispose);
1409 data->found++;
1410 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 }
1412 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001413 * We can return to the caller if we have found some (this
1414 * ensures forward progress). We'll be coming back to find
1415 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 */
Al Virofe915222014-05-03 00:02:25 -04001417 if (!list_empty(&data->dispose))
1418 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001420 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421}
1422
1423/**
1424 * shrink_dcache_parent - prune dcache
1425 * @parent: parent of entries to prune
1426 *
1427 * Prune the dcache to remove unused children of the parent dentry.
1428 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001429void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001431 for (;;) {
1432 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001434 INIT_LIST_HEAD(&data.dispose);
1435 data.start = parent;
1436 data.found = 0;
1437
1438 d_walk(parent, &data, select_collect, NULL);
1439 if (!data.found)
1440 break;
1441
1442 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001443 cond_resched();
1444 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001446EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Al Viro9c8c10e2014-05-02 20:36:10 -04001448static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001449{
Al Viro9c8c10e2014-05-02 20:36:10 -04001450 /* it has busy descendents; complain about those instead */
1451 if (!list_empty(&dentry->d_subdirs))
1452 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001453
Al Viro9c8c10e2014-05-02 20:36:10 -04001454 /* root with refcount 1 is fine */
1455 if (dentry == _data && dentry->d_lockref.count == 1)
1456 return D_WALK_CONTINUE;
1457
1458 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1459 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001460 dentry,
1461 dentry->d_inode ?
1462 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001463 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001464 dentry->d_lockref.count,
1465 dentry->d_sb->s_type->name,
1466 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001467 WARN_ON(1);
1468 return D_WALK_CONTINUE;
1469}
1470
1471static void do_one_tree(struct dentry *dentry)
1472{
1473 shrink_dcache_parent(dentry);
1474 d_walk(dentry, dentry, umount_check, NULL);
1475 d_drop(dentry);
1476 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001477}
1478
1479/*
1480 * destroy the dentries attached to a superblock on unmounting
1481 */
1482void shrink_dcache_for_umount(struct super_block *sb)
1483{
1484 struct dentry *dentry;
1485
Al Viro9c8c10e2014-05-02 20:36:10 -04001486 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001487
1488 dentry = sb->s_root;
1489 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001490 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001491
1492 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001493 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1494 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001495 }
1496}
1497
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001498struct detach_data {
1499 struct select_data select;
1500 struct dentry *mountpoint;
1501};
1502static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001503{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001504 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001505
1506 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001507 __dget_dlock(dentry);
1508 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001509 return D_WALK_QUIT;
1510 }
1511
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001512 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001513}
1514
1515static void check_and_drop(void *_data)
1516{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001517 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001518
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001519 if (!data->mountpoint && !data->select.found)
1520 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001521}
1522
1523/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001524 * d_invalidate - detach submounts, prune dcache, and drop
1525 * @dentry: dentry to invalidate (aka detach, prune and drop)
1526 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001527 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001528 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001529 * The final d_drop is done as an atomic operation relative to
1530 * rename_lock ensuring there are no races with d_set_mounted. This
1531 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001532 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001533void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001534{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001535 /*
1536 * If it's already been dropped, return OK.
1537 */
1538 spin_lock(&dentry->d_lock);
1539 if (d_unhashed(dentry)) {
1540 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001541 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001542 }
1543 spin_unlock(&dentry->d_lock);
1544
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001545 /* Negative dentries can be dropped without further checks */
1546 if (!dentry->d_inode) {
1547 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001548 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001549 }
1550
1551 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001552 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001553
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001554 data.mountpoint = NULL;
1555 INIT_LIST_HEAD(&data.select.dispose);
1556 data.select.start = dentry;
1557 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001558
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001559 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001560
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001561 if (data.select.found)
1562 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001563
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001564 if (data.mountpoint) {
1565 detach_mounts(data.mountpoint);
1566 dput(data.mountpoint);
1567 }
1568
1569 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001570 break;
1571
1572 cond_resched();
1573 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001574}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001575EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001576
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577/**
Al Viroa4464db2011-07-07 15:03:58 -04001578 * __d_alloc - allocate a dcache entry
1579 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 * @name: qstr of the name
1581 *
1582 * Allocates a dentry. It returns %NULL if there is insufficient memory
1583 * available. On a success the dentry is returned. The name passed in is
1584 * copied and the copy passed in may be reused after this call.
1585 */
1586
Al Viroa4464db2011-07-07 15:03:58 -04001587struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588{
1589 struct dentry *dentry;
1590 char *dname;
Miklos Szeredi285b1022016-06-28 11:47:32 +02001591 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592
Mel Gormane12ba742007-10-16 01:25:52 -07001593 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 if (!dentry)
1595 return NULL;
1596
Linus Torvalds6326c712012-05-21 16:14:04 -07001597 /*
1598 * We guarantee that the inline name is always NUL-terminated.
1599 * This way the memcpy() done by the name switching in rename
1600 * will still always have a NUL at the end, even if we might
1601 * be overwriting an internal NUL character
1602 */
1603 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Al Viro798434b2016-03-24 20:38:43 -04001604 if (unlikely(!name)) {
1605 static const struct qstr anon = QSTR_INIT("/", 1);
1606 name = &anon;
1607 dname = dentry->d_iname;
1608 } else if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001609 size_t size = offsetof(struct external_name, name[1]);
Vladimir Davydov5d097052016-01-14 15:18:21 -08001610 struct external_name *p = kmalloc(size + name->len,
1611 GFP_KERNEL_ACCOUNT);
Al Viro8d85b482014-09-29 14:54:27 -04001612 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 kmem_cache_free(dentry_cache, dentry);
1614 return NULL;
1615 }
Al Viro8d85b482014-09-29 14:54:27 -04001616 atomic_set(&p->u.count, 1);
1617 dname = p->name;
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -08001618 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1619 kasan_unpoison_shadow(dname,
1620 round_up(name->len + 1, sizeof(unsigned long)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 } else {
1622 dname = dentry->d_iname;
1623 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624
1625 dentry->d_name.len = name->len;
1626 dentry->d_name.hash = name->hash;
1627 memcpy(dname, name->name, name->len);
1628 dname[name->len] = 0;
1629
Linus Torvalds6326c712012-05-21 16:14:04 -07001630 /* Make sure we always see the terminating NUL character */
1631 smp_wmb();
1632 dentry->d_name.name = dname;
1633
Waiman Long98474232013-08-28 18:24:59 -07001634 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001635 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001637 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001639 dentry->d_parent = dentry;
1640 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 dentry->d_op = NULL;
1642 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001643 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 INIT_LIST_HEAD(&dentry->d_lru);
1645 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001646 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1647 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001648 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649
Miklos Szeredi285b1022016-06-28 11:47:32 +02001650 if (dentry->d_op && dentry->d_op->d_init) {
1651 err = dentry->d_op->d_init(dentry);
1652 if (err) {
1653 if (dname_external(dentry))
1654 kfree(external_name(dentry));
1655 kmem_cache_free(dentry_cache, dentry);
1656 return NULL;
1657 }
1658 }
1659
Nick Piggin3e880fb2011-01-07 17:49:19 +11001660 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001661
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 return dentry;
1663}
Al Viroa4464db2011-07-07 15:03:58 -04001664
1665/**
1666 * d_alloc - allocate a dcache entry
1667 * @parent: parent of entry to allocate
1668 * @name: qstr of the name
1669 *
1670 * Allocates a dentry. It returns %NULL if there is insufficient memory
1671 * available. On a success the dentry is returned. The name passed in is
1672 * copied and the copy passed in may be reused after this call.
1673 */
1674struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1675{
1676 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1677 if (!dentry)
1678 return NULL;
Al Viro3d56c252016-06-07 21:26:55 -04001679 dentry->d_flags |= DCACHE_RCUACCESS;
Al Viroa4464db2011-07-07 15:03:58 -04001680 spin_lock(&parent->d_lock);
1681 /*
1682 * don't need child lock because it is not subject
1683 * to concurrency here
1684 */
1685 __dget_dlock(parent);
1686 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001687 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001688 spin_unlock(&parent->d_lock);
1689
1690 return dentry;
1691}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001692EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693
Al Viroba65dc52016-06-10 11:32:47 -04001694struct dentry *d_alloc_cursor(struct dentry * parent)
1695{
1696 struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
1697 if (dentry) {
1698 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1699 dentry->d_parent = dget(parent);
1700 }
1701 return dentry;
1702}
1703
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001704/**
1705 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1706 * @sb: the superblock
1707 * @name: qstr of the name
1708 *
1709 * For a filesystem that just pins its dentries in memory and never
1710 * performs lookups at all, return an unhashed IS_ROOT dentry.
1711 */
Nick Piggin4b936882011-01-07 17:50:07 +11001712struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1713{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001714 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001715}
1716EXPORT_SYMBOL(d_alloc_pseudo);
1717
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1719{
1720 struct qstr q;
1721
1722 q.name = name;
Linus Torvalds8387ff22016-06-10 07:51:30 -07001723 q.hash_len = hashlen_string(parent, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 return d_alloc(parent, &q);
1725}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001726EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727
Nick Pigginfb045ad2011-01-07 17:49:55 +11001728void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1729{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001730 WARN_ON_ONCE(dentry->d_op);
1731 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001732 DCACHE_OP_COMPARE |
1733 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001734 DCACHE_OP_WEAK_REVALIDATE |
David Howells4bacc9c2015-06-18 14:32:31 +01001735 DCACHE_OP_DELETE |
Miklos Szeredid101a122016-03-26 16:14:37 -04001736 DCACHE_OP_REAL));
Nick Pigginfb045ad2011-01-07 17:49:55 +11001737 dentry->d_op = op;
1738 if (!op)
1739 return;
1740 if (op->d_hash)
1741 dentry->d_flags |= DCACHE_OP_HASH;
1742 if (op->d_compare)
1743 dentry->d_flags |= DCACHE_OP_COMPARE;
1744 if (op->d_revalidate)
1745 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001746 if (op->d_weak_revalidate)
1747 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001748 if (op->d_delete)
1749 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001750 if (op->d_prune)
1751 dentry->d_flags |= DCACHE_OP_PRUNE;
Miklos Szeredid101a122016-03-26 16:14:37 -04001752 if (op->d_real)
1753 dentry->d_flags |= DCACHE_OP_REAL;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001754
1755}
1756EXPORT_SYMBOL(d_set_d_op);
1757
David Howellsdf1a0852015-01-29 12:02:28 +00001758
1759/*
1760 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1761 * @dentry - The dentry to mark
1762 *
1763 * Mark a dentry as falling through to the lower layer (as set with
1764 * d_pin_lower()). This flag may be recorded on the medium.
1765 */
1766void d_set_fallthru(struct dentry *dentry)
1767{
1768 spin_lock(&dentry->d_lock);
1769 dentry->d_flags |= DCACHE_FALLTHRU;
1770 spin_unlock(&dentry->d_lock);
1771}
1772EXPORT_SYMBOL(d_set_fallthru);
1773
David Howellsb18825a2013-09-12 19:22:53 +01001774static unsigned d_flags_for_inode(struct inode *inode)
1775{
David Howells44bdb5e2015-01-29 12:02:29 +00001776 unsigned add_flags = DCACHE_REGULAR_TYPE;
David Howellsb18825a2013-09-12 19:22:53 +01001777
1778 if (!inode)
1779 return DCACHE_MISS_TYPE;
1780
1781 if (S_ISDIR(inode->i_mode)) {
1782 add_flags = DCACHE_DIRECTORY_TYPE;
1783 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1784 if (unlikely(!inode->i_op->lookup))
1785 add_flags = DCACHE_AUTODIR_TYPE;
1786 else
1787 inode->i_opflags |= IOP_LOOKUP;
1788 }
David Howells44bdb5e2015-01-29 12:02:29 +00001789 goto type_determined;
David Howellsb18825a2013-09-12 19:22:53 +01001790 }
1791
David Howells44bdb5e2015-01-29 12:02:29 +00001792 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
Al Viro6b255392015-11-17 10:20:54 -05001793 if (unlikely(inode->i_op->get_link)) {
David Howells44bdb5e2015-01-29 12:02:29 +00001794 add_flags = DCACHE_SYMLINK_TYPE;
1795 goto type_determined;
1796 }
1797 inode->i_opflags |= IOP_NOFOLLOW;
1798 }
1799
1800 if (unlikely(!S_ISREG(inode->i_mode)))
1801 add_flags = DCACHE_SPECIAL_TYPE;
1802
1803type_determined:
David Howellsb18825a2013-09-12 19:22:53 +01001804 if (unlikely(IS_AUTOMOUNT(inode)))
1805 add_flags |= DCACHE_NEED_AUTOMOUNT;
1806 return add_flags;
1807}
1808
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001809static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1810{
David Howellsb18825a2013-09-12 19:22:53 +01001811 unsigned add_flags = d_flags_for_inode(inode);
Al Viro85c7f812016-04-14 19:52:13 -04001812 WARN_ON(d_in_lookup(dentry));
David Howellsb18825a2013-09-12 19:22:53 +01001813
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001814 spin_lock(&dentry->d_lock);
Al Virode689f52016-03-09 18:05:42 -05001815 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
Al Viroa528aca2016-02-29 12:12:46 -05001816 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +00001817 __d_set_inode_and_type(dentry, inode, add_flags);
Al Viroa528aca2016-02-29 12:12:46 -05001818 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04001819 fsnotify_update_flags(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001820 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001821}
1822
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823/**
1824 * d_instantiate - fill in inode information for a dentry
1825 * @entry: dentry to complete
1826 * @inode: inode to attach to this dentry
1827 *
1828 * Fill in inode information in the entry.
1829 *
1830 * This turns negative dentries into productive full members
1831 * of society.
1832 *
1833 * NOTE! This assumes that the inode count has been incremented
1834 * (or otherwise set) by the caller to indicate that it is now
1835 * in use by the dcache.
1836 */
1837
1838void d_instantiate(struct dentry *entry, struct inode * inode)
1839{
Al Viro946e51f2014-10-26 19:19:16 -04001840 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Al Virode689f52016-03-09 18:05:42 -05001841 if (inode) {
Al Virob9680912016-04-11 00:53:26 -04001842 security_d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001843 spin_lock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001844 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001845 spin_unlock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001846 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001848EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849
1850/**
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001851 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1852 * @entry: dentry to complete
1853 * @inode: inode to attach to this dentry
1854 *
1855 * Fill in inode information in the entry. If a directory alias is found, then
1856 * return an error (and drop inode). Together with d_materialise_unique() this
1857 * guarantees that a directory inode may never have more than one alias.
1858 */
1859int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1860{
Al Viro946e51f2014-10-26 19:19:16 -04001861 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001862
Al Virob9680912016-04-11 00:53:26 -04001863 security_d_instantiate(entry, inode);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001864 spin_lock(&inode->i_lock);
1865 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1866 spin_unlock(&inode->i_lock);
1867 iput(inode);
1868 return -EBUSY;
1869 }
1870 __d_instantiate(entry, inode);
1871 spin_unlock(&inode->i_lock);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001872
1873 return 0;
1874}
1875EXPORT_SYMBOL(d_instantiate_no_diralias);
1876
Al Viroadc0e912012-01-08 16:49:21 -05001877struct dentry *d_make_root(struct inode *root_inode)
1878{
1879 struct dentry *res = NULL;
1880
1881 if (root_inode) {
Al Viro798434b2016-03-24 20:38:43 -04001882 res = __d_alloc(root_inode->i_sb, NULL);
Al Viroadc0e912012-01-08 16:49:21 -05001883 if (res)
1884 d_instantiate(res, root_inode);
1885 else
1886 iput(root_inode);
1887 }
1888 return res;
1889}
1890EXPORT_SYMBOL(d_make_root);
1891
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001892static struct dentry * __d_find_any_alias(struct inode *inode)
1893{
1894 struct dentry *alias;
1895
Al Virob3d9b7a2012-06-09 13:51:19 -04001896 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001897 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001898 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001899 __dget(alias);
1900 return alias;
1901}
1902
Sage Weil46f72b32012-01-10 09:04:37 -08001903/**
1904 * d_find_any_alias - find any alias for a given inode
1905 * @inode: inode to find an alias for
1906 *
1907 * If any aliases exist for the given inode, take and return a
1908 * reference for one of them. If no aliases exist, return %NULL.
1909 */
1910struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001911{
1912 struct dentry *de;
1913
1914 spin_lock(&inode->i_lock);
1915 de = __d_find_any_alias(inode);
1916 spin_unlock(&inode->i_lock);
1917 return de;
1918}
Sage Weil46f72b32012-01-10 09:04:37 -08001919EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001920
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001921static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001922{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001923 struct dentry *tmp;
1924 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001925 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001926
1927 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001928 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001929 if (IS_ERR(inode))
1930 return ERR_CAST(inode);
1931
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001932 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001933 if (res)
1934 goto out_iput;
1935
Al Viro798434b2016-03-24 20:38:43 -04001936 tmp = __d_alloc(inode->i_sb, NULL);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001937 if (!tmp) {
1938 res = ERR_PTR(-ENOMEM);
1939 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001940 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001941
Al Virob9680912016-04-11 00:53:26 -04001942 security_d_instantiate(tmp, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001943 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001944 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001945 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001946 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001947 dput(tmp);
1948 goto out_iput;
1949 }
1950
1951 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001952 add_flags = d_flags_for_inode(inode);
1953
1954 if (disconnected)
1955 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001956
Christoph Hellwig9308a612008-08-11 15:49:12 +02001957 spin_lock(&tmp->d_lock);
David Howells4bf46a22015-03-05 14:09:22 +00001958 __d_set_inode_and_type(tmp, inode, add_flags);
Al Viro946e51f2014-10-26 19:19:16 -04001959 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001960 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001961 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001962 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001963 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001964 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001965
Christoph Hellwig9308a612008-08-11 15:49:12 +02001966 return tmp;
1967
1968 out_iput:
1969 iput(inode);
1970 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001971}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001972
1973/**
1974 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1975 * @inode: inode to allocate the dentry for
1976 *
1977 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1978 * similar open by handle operations. The returned dentry may be anonymous,
1979 * or may have a full name (if the inode was already in the cache).
1980 *
1981 * When called on a directory inode, we must ensure that the inode only ever
1982 * has one dentry. If a dentry is found, that is returned instead of
1983 * allocating a new one.
1984 *
1985 * On successful return, the reference to the inode has been transferred
1986 * to the dentry. In case of an error the reference on the inode is released.
1987 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1988 * be passed in and the error will be propagated to the return value,
1989 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1990 */
1991struct dentry *d_obtain_alias(struct inode *inode)
1992{
1993 return __d_obtain_alias(inode, 1);
1994}
Benny Halevyadc48722009-02-27 14:02:59 -08001995EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996
1997/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001998 * d_obtain_root - find or allocate a dentry for a given inode
1999 * @inode: inode to allocate the dentry for
2000 *
2001 * Obtain an IS_ROOT dentry for the root of a filesystem.
2002 *
2003 * We must ensure that directory inodes only ever have one dentry. If a
2004 * dentry is found, that is returned instead of allocating a new one.
2005 *
2006 * On successful return, the reference to the inode has been transferred
2007 * to the dentry. In case of an error the reference on the inode is
2008 * released. A %NULL or IS_ERR inode may be passed in and will be the
2009 * error will be propagate to the return value, with a %NULL @inode
2010 * replaced by ERR_PTR(-ESTALE).
2011 */
2012struct dentry *d_obtain_root(struct inode *inode)
2013{
2014 return __d_obtain_alias(inode, 0);
2015}
2016EXPORT_SYMBOL(d_obtain_root);
2017
2018/**
Barry Naujok94035402008-05-21 16:50:46 +10002019 * d_add_ci - lookup or allocate new dentry with case-exact name
2020 * @inode: the inode case-insensitive lookup has found
2021 * @dentry: the negative dentry that was passed to the parent's lookup func
2022 * @name: the case-exact name to be associated with the returned dentry
2023 *
2024 * This is to avoid filling the dcache with case-insensitive names to the
2025 * same inode, only the actual correct case is stored in the dcache for
2026 * case-insensitive filesystems.
2027 *
2028 * For a case-insensitive lookup match and if the the case-exact dentry
2029 * already exists in in the dcache, use it and return it.
2030 *
2031 * If no entry exists with the exact case name, allocate new dentry with
2032 * the exact case, and return the spliced entry.
2033 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02002034struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10002035 struct qstr *name)
2036{
Al Virod9171b92016-04-15 03:33:13 -04002037 struct dentry *found, *res;
Barry Naujok94035402008-05-21 16:50:46 +10002038
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002039 /*
2040 * First check if a dentry matching the name already exists,
2041 * if not go ahead and create it now.
2042 */
Barry Naujok94035402008-05-21 16:50:46 +10002043 found = d_hash_and_lookup(dentry->d_parent, name);
Al Virod9171b92016-04-15 03:33:13 -04002044 if (found) {
2045 iput(inode);
2046 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002047 }
Al Virod9171b92016-04-15 03:33:13 -04002048 if (d_in_lookup(dentry)) {
2049 found = d_alloc_parallel(dentry->d_parent, name,
2050 dentry->d_wait);
2051 if (IS_ERR(found) || !d_in_lookup(found)) {
2052 iput(inode);
2053 return found;
2054 }
2055 } else {
2056 found = d_alloc(dentry->d_parent, name);
2057 if (!found) {
2058 iput(inode);
2059 return ERR_PTR(-ENOMEM);
2060 }
2061 }
2062 res = d_splice_alias(inode, found);
2063 if (res) {
2064 dput(found);
2065 return res;
2066 }
Al Viro4f522a22013-02-11 23:20:37 -05002067 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002068}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002069EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002071
Al Virod4c91a82016-06-25 23:33:49 -04002072static inline bool d_same_name(const struct dentry *dentry,
2073 const struct dentry *parent,
2074 const struct qstr *name)
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002075{
Al Virod4c91a82016-06-25 23:33:49 -04002076 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2077 if (dentry->d_name.len != name->len)
2078 return false;
2079 return dentry_cmp(dentry, name->name, name->len) == 0;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002080 }
Al Viro6fa67e72016-07-31 16:37:25 -04002081 return parent->d_op->d_compare(dentry,
Al Virod4c91a82016-06-25 23:33:49 -04002082 dentry->d_name.len, dentry->d_name.name,
2083 name) == 0;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002084}
2085
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002087 * __d_lookup_rcu - search for a dentry (racy, store-free)
2088 * @parent: parent dentry
2089 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002090 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002091 * Returns: dentry, or NULL
2092 *
2093 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2094 * resolution (store-free path walking) design described in
2095 * Documentation/filesystems/path-lookup.txt.
2096 *
2097 * This is not to be used outside core vfs.
2098 *
2099 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2100 * held, and rcu_read_lock held. The returned dentry must not be stored into
2101 * without taking d_lock and checking d_seq sequence count against @seq
2102 * returned here.
2103 *
Linus Torvalds15570082013-09-02 11:38:06 -07002104 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002105 * function.
2106 *
2107 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2108 * the returned dentry, so long as its parent's seqlock is checked after the
2109 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2110 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002111 *
2112 * NOTE! The caller *has* to check the resulting dentry against the sequence
2113 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002114 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002115struct dentry *__d_lookup_rcu(const struct dentry *parent,
2116 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002117 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002118{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002119 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002120 const unsigned char *str = name->name;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002121 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002122 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002123 struct dentry *dentry;
2124
2125 /*
2126 * Note: There is significant duplication with __d_lookup_rcu which is
2127 * required to prevent single threaded performance regressions
2128 * especially on architectures where smp_rmb (in seqcounts) are costly.
2129 * Keep the two functions in sync.
2130 */
2131
2132 /*
2133 * The hash list is protected using RCU.
2134 *
2135 * Carefully use d_seq when comparing a candidate dentry, to avoid
2136 * races with d_move().
2137 *
2138 * It is possible that concurrent renames can mess up our list
2139 * walk here and result in missing our dentry, resulting in the
2140 * false-negative result. d_lookup() protects against concurrent
2141 * renames using rename_lock seqlock.
2142 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002143 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002144 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002145 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002146 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002147
Nick Piggin31e6b012011-01-07 17:49:52 +11002148seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002149 /*
2150 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002151 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002152 *
2153 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002154 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002155 *
2156 * NOTE! We do a "raw" seqcount_begin here. That means that
2157 * we don't wait for the sequence count to stabilize if it
2158 * is in the middle of a sequence change. If we do the slow
2159 * dentry compare, we will do seqretries until it is stable,
2160 * and if we end up with a successful lookup, we actually
2161 * want to exit RCU lookup anyway.
Al Virod4c91a82016-06-25 23:33:49 -04002162 *
2163 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2164 * we are still guaranteed NUL-termination of ->d_name.name.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002165 */
2166 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002167 if (dentry->d_parent != parent)
2168 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002169 if (d_unhashed(dentry))
2170 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002171
2172 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Al Virod4c91a82016-06-25 23:33:49 -04002173 int tlen;
2174 const char *tname;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002175 if (dentry->d_name.hash != hashlen_hash(hashlen))
2176 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002177 tlen = dentry->d_name.len;
2178 tname = dentry->d_name.name;
2179 /* we want a consistent (name,len) pair */
2180 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2181 cpu_relax();
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002182 goto seqretry;
2183 }
Al Viro6fa67e72016-07-31 16:37:25 -04002184 if (parent->d_op->d_compare(dentry,
Al Virod4c91a82016-06-25 23:33:49 -04002185 tlen, tname, name) != 0)
2186 continue;
2187 } else {
2188 if (dentry->d_name.hash_len != hashlen)
2189 continue;
2190 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2191 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002192 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002193 *seqp = seq;
Al Virod4c91a82016-06-25 23:33:49 -04002194 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002195 }
2196 return NULL;
2197}
2198
2199/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 * d_lookup - search for a dentry
2201 * @parent: parent dentry
2202 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002203 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002205 * d_lookup searches the children of the parent dentry for the name in
2206 * question. If the dentry is found its reference count is incremented and the
2207 * dentry is returned. The caller must use dput to free the entry when it has
2208 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 */
Al Viroda2d8452013-01-24 18:29:34 -05002210struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211{
Nick Piggin31e6b012011-01-07 17:49:52 +11002212 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002213 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214
Daeseok Younb8314f92014-08-11 11:46:53 +09002215 do {
2216 seq = read_seqbegin(&rename_lock);
2217 dentry = __d_lookup(parent, name);
2218 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 break;
2220 } while (read_seqretry(&rename_lock, seq));
2221 return dentry;
2222}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002223EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224
Nick Piggin31e6b012011-01-07 17:49:52 +11002225/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002226 * __d_lookup - search for a dentry (racy)
2227 * @parent: parent dentry
2228 * @name: qstr of name we wish to find
2229 * Returns: dentry, or NULL
2230 *
2231 * __d_lookup is like d_lookup, however it may (rarely) return a
2232 * false-negative result due to unrelated rename activity.
2233 *
2234 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2235 * however it must be used carefully, eg. with a following d_lookup in
2236 * the case of failure.
2237 *
2238 * __d_lookup callers must be commented.
2239 */
Al Viroa713ca22013-01-24 18:27:00 -05002240struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 unsigned int hash = name->hash;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002243 struct hlist_bl_head *b = d_hash(hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002244 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002245 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002246 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
Nick Pigginb04f7842010-08-18 04:37:34 +10002248 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002249 * Note: There is significant duplication with __d_lookup_rcu which is
2250 * required to prevent single threaded performance regressions
2251 * especially on architectures where smp_rmb (in seqcounts) are costly.
2252 * Keep the two functions in sync.
2253 */
2254
2255 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002256 * The hash list is protected using RCU.
2257 *
2258 * Take d_lock when comparing a candidate dentry, to avoid races
2259 * with d_move().
2260 *
2261 * It is possible that concurrent renames can mess up our list
2262 * walk here and result in missing our dentry, resulting in the
2263 * false-negative result. d_lookup() protects against concurrent
2264 * renames using rename_lock seqlock.
2265 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002266 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002267 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 rcu_read_lock();
2269
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002270 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 if (dentry->d_name.hash != hash)
2273 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274
2275 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 if (dentry->d_parent != parent)
2277 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002278 if (d_unhashed(dentry))
2279 goto next;
2280
Al Virod4c91a82016-06-25 23:33:49 -04002281 if (!d_same_name(dentry, parent, name))
2282 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283
Waiman Long98474232013-08-28 18:24:59 -07002284 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002285 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 spin_unlock(&dentry->d_lock);
2287 break;
2288next:
2289 spin_unlock(&dentry->d_lock);
2290 }
2291 rcu_read_unlock();
2292
2293 return found;
2294}
2295
2296/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002297 * d_hash_and_lookup - hash the qstr then search for a dentry
2298 * @dir: Directory to search in
2299 * @name: qstr of name we wish to find
2300 *
Al Viro4f522a22013-02-11 23:20:37 -05002301 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002302 */
2303struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2304{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002305 /*
2306 * Check for a fs-specific hash function. Note that we must
2307 * calculate the standard hash first, as the d_op->d_hash()
2308 * routine may choose to leave the hash value unchanged.
2309 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002310 name->hash = full_name_hash(dir, name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002311 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002312 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002313 if (unlikely(err < 0))
2314 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002315 }
Al Viro4f522a22013-02-11 23:20:37 -05002316 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002317}
Al Viro4f522a22013-02-11 23:20:37 -05002318EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002319
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320/*
2321 * When a file is deleted, we have two options:
2322 * - turn this dentry into a negative dentry
2323 * - unhash this dentry and free it.
2324 *
2325 * Usually, we want to just turn this into
2326 * a negative dentry, but if anybody else is
2327 * currently using the dentry or the inode
2328 * we can't do that and we fall back on removing
2329 * it from the hash queues and waiting for
2330 * it to be deleted later when it has no users
2331 */
2332
2333/**
2334 * d_delete - delete a dentry
2335 * @dentry: The dentry to delete
2336 *
2337 * Turn the dentry into a negative dentry if possible, otherwise
2338 * remove it from the hash queues so it can be deleted later
2339 */
2340
2341void d_delete(struct dentry * dentry)
2342{
Nick Piggin873feea2011-01-07 17:50:06 +11002343 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002344 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 /*
2346 * Are we the only user?
2347 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002348again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002350 inode = dentry->d_inode;
2351 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002352 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002353 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002354 spin_unlock(&dentry->d_lock);
2355 cpu_relax();
2356 goto again;
2357 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002358 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002359 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002360 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 return;
2362 }
2363
2364 if (!d_unhashed(dentry))
2365 __d_drop(dentry);
2366
2367 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002368
2369 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002371EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372
Al Viro15d3c582016-07-29 17:45:21 -04002373static void __d_rehash(struct dentry *entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374{
Al Viro15d3c582016-07-29 17:45:21 -04002375 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002376 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002377 hlist_bl_lock(b);
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002378 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002379 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380}
2381
2382/**
2383 * d_rehash - add an entry back to the hash
2384 * @entry: dentry to add to the hash
2385 *
2386 * Adds a dentry to the hash according to its name.
2387 */
2388
2389void d_rehash(struct dentry * entry)
2390{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 spin_lock(&entry->d_lock);
Al Viro15d3c582016-07-29 17:45:21 -04002392 __d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002395EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396
Al Viro84e710d2016-04-15 00:58:55 -04002397static inline unsigned start_dir_add(struct inode *dir)
2398{
2399
2400 for (;;) {
2401 unsigned n = dir->i_dir_seq;
2402 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2403 return n;
2404 cpu_relax();
2405 }
2406}
2407
2408static inline void end_dir_add(struct inode *dir, unsigned n)
2409{
2410 smp_store_release(&dir->i_dir_seq, n + 2);
2411}
2412
Al Virod9171b92016-04-15 03:33:13 -04002413static void d_wait_lookup(struct dentry *dentry)
2414{
2415 if (d_in_lookup(dentry)) {
2416 DECLARE_WAITQUEUE(wait, current);
2417 add_wait_queue(dentry->d_wait, &wait);
2418 do {
2419 set_current_state(TASK_UNINTERRUPTIBLE);
2420 spin_unlock(&dentry->d_lock);
2421 schedule();
2422 spin_lock(&dentry->d_lock);
2423 } while (d_in_lookup(dentry));
2424 }
2425}
2426
Al Viro94bdd652016-04-15 02:42:04 -04002427struct dentry *d_alloc_parallel(struct dentry *parent,
Al Virod9171b92016-04-15 03:33:13 -04002428 const struct qstr *name,
2429 wait_queue_head_t *wq)
Al Viro94bdd652016-04-15 02:42:04 -04002430{
Al Viro94bdd652016-04-15 02:42:04 -04002431 unsigned int hash = name->hash;
Al Viro94bdd652016-04-15 02:42:04 -04002432 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2433 struct hlist_bl_node *node;
2434 struct dentry *new = d_alloc(parent, name);
2435 struct dentry *dentry;
2436 unsigned seq, r_seq, d_seq;
2437
2438 if (unlikely(!new))
2439 return ERR_PTR(-ENOMEM);
2440
2441retry:
2442 rcu_read_lock();
2443 seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1;
2444 r_seq = read_seqbegin(&rename_lock);
2445 dentry = __d_lookup_rcu(parent, name, &d_seq);
2446 if (unlikely(dentry)) {
2447 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2448 rcu_read_unlock();
2449 goto retry;
2450 }
2451 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2452 rcu_read_unlock();
2453 dput(dentry);
2454 goto retry;
2455 }
2456 rcu_read_unlock();
2457 dput(new);
2458 return dentry;
2459 }
2460 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2461 rcu_read_unlock();
2462 goto retry;
2463 }
2464 hlist_bl_lock(b);
2465 if (unlikely(parent->d_inode->i_dir_seq != seq)) {
2466 hlist_bl_unlock(b);
2467 rcu_read_unlock();
2468 goto retry;
2469 }
Al Viro94bdd652016-04-15 02:42:04 -04002470 /*
2471 * No changes for the parent since the beginning of d_lookup().
2472 * Since all removals from the chain happen with hlist_bl_lock(),
2473 * any potential in-lookup matches are going to stay here until
2474 * we unlock the chain. All fields are stable in everything
2475 * we encounter.
2476 */
2477 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2478 if (dentry->d_name.hash != hash)
2479 continue;
2480 if (dentry->d_parent != parent)
2481 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002482 if (!d_same_name(dentry, parent, name))
2483 continue;
Al Viro94bdd652016-04-15 02:42:04 -04002484 hlist_bl_unlock(b);
Al Viroe7d6ef92016-06-20 01:35:59 -04002485 /* now we can try to grab a reference */
2486 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2487 rcu_read_unlock();
2488 goto retry;
2489 }
2490
2491 rcu_read_unlock();
2492 /*
2493 * somebody is likely to be still doing lookup for it;
2494 * wait for them to finish
2495 */
Al Virod9171b92016-04-15 03:33:13 -04002496 spin_lock(&dentry->d_lock);
2497 d_wait_lookup(dentry);
2498 /*
2499 * it's not in-lookup anymore; in principle we should repeat
2500 * everything from dcache lookup, but it's likely to be what
2501 * d_lookup() would've found anyway. If it is, just return it;
2502 * otherwise we really have to repeat the whole thing.
2503 */
2504 if (unlikely(dentry->d_name.hash != hash))
2505 goto mismatch;
2506 if (unlikely(dentry->d_parent != parent))
2507 goto mismatch;
2508 if (unlikely(d_unhashed(dentry)))
2509 goto mismatch;
Al Virod4c91a82016-06-25 23:33:49 -04002510 if (unlikely(!d_same_name(dentry, parent, name)))
2511 goto mismatch;
Al Virod9171b92016-04-15 03:33:13 -04002512 /* OK, it *is* a hashed match; return it */
2513 spin_unlock(&dentry->d_lock);
Al Viro94bdd652016-04-15 02:42:04 -04002514 dput(new);
2515 return dentry;
2516 }
Al Viroe7d6ef92016-06-20 01:35:59 -04002517 rcu_read_unlock();
Al Viro94bdd652016-04-15 02:42:04 -04002518 /* we can't take ->d_lock here; it's OK, though. */
2519 new->d_flags |= DCACHE_PAR_LOOKUP;
Al Virod9171b92016-04-15 03:33:13 -04002520 new->d_wait = wq;
Al Viro94bdd652016-04-15 02:42:04 -04002521 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2522 hlist_bl_unlock(b);
2523 return new;
Al Virod9171b92016-04-15 03:33:13 -04002524mismatch:
2525 spin_unlock(&dentry->d_lock);
2526 dput(dentry);
2527 goto retry;
Al Viro94bdd652016-04-15 02:42:04 -04002528}
2529EXPORT_SYMBOL(d_alloc_parallel);
2530
Al Viro85c7f812016-04-14 19:52:13 -04002531void __d_lookup_done(struct dentry *dentry)
2532{
Al Viro94bdd652016-04-15 02:42:04 -04002533 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2534 dentry->d_name.hash);
2535 hlist_bl_lock(b);
Al Viro85c7f812016-04-14 19:52:13 -04002536 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
Al Viro94bdd652016-04-15 02:42:04 -04002537 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
Al Virod9171b92016-04-15 03:33:13 -04002538 wake_up_all(dentry->d_wait);
2539 dentry->d_wait = NULL;
Al Viro94bdd652016-04-15 02:42:04 -04002540 hlist_bl_unlock(b);
2541 INIT_HLIST_NODE(&dentry->d_u.d_alias);
Al Virod9171b92016-04-15 03:33:13 -04002542 INIT_LIST_HEAD(&dentry->d_lru);
Al Viro85c7f812016-04-14 19:52:13 -04002543}
2544EXPORT_SYMBOL(__d_lookup_done);
Al Viroed782b52016-03-09 19:52:39 -05002545
2546/* inode->i_lock held if inode is non-NULL */
2547
2548static inline void __d_add(struct dentry *dentry, struct inode *inode)
2549{
Al Viro84e710d2016-04-15 00:58:55 -04002550 struct inode *dir = NULL;
2551 unsigned n;
Al Viro0568d702016-04-14 19:40:56 -04002552 spin_lock(&dentry->d_lock);
Al Viro84e710d2016-04-15 00:58:55 -04002553 if (unlikely(d_in_lookup(dentry))) {
2554 dir = dentry->d_parent->d_inode;
2555 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002556 __d_lookup_done(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002557 }
Al Viroed782b52016-03-09 19:52:39 -05002558 if (inode) {
Al Viro0568d702016-04-14 19:40:56 -04002559 unsigned add_flags = d_flags_for_inode(inode);
2560 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2561 raw_write_seqcount_begin(&dentry->d_seq);
2562 __d_set_inode_and_type(dentry, inode, add_flags);
2563 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04002564 fsnotify_update_flags(dentry);
Al Viroed782b52016-03-09 19:52:39 -05002565 }
Al Viro15d3c582016-07-29 17:45:21 -04002566 __d_rehash(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002567 if (dir)
2568 end_dir_add(dir, n);
Al Viro0568d702016-04-14 19:40:56 -04002569 spin_unlock(&dentry->d_lock);
2570 if (inode)
2571 spin_unlock(&inode->i_lock);
Al Viroed782b52016-03-09 19:52:39 -05002572}
2573
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002574/**
Al Viro34d0d192016-03-08 21:01:03 -05002575 * d_add - add dentry to hash queues
2576 * @entry: dentry to add
2577 * @inode: The inode to attach to this dentry
2578 *
2579 * This adds the entry to the hash queues and initializes @inode.
2580 * The entry was actually filled in earlier during d_alloc().
2581 */
2582
2583void d_add(struct dentry *entry, struct inode *inode)
2584{
Al Virob9680912016-04-11 00:53:26 -04002585 if (inode) {
2586 security_d_instantiate(entry, inode);
Al Viroed782b52016-03-09 19:52:39 -05002587 spin_lock(&inode->i_lock);
Al Virob9680912016-04-11 00:53:26 -04002588 }
Al Viroed782b52016-03-09 19:52:39 -05002589 __d_add(entry, inode);
Al Viro34d0d192016-03-08 21:01:03 -05002590}
2591EXPORT_SYMBOL(d_add);
2592
2593/**
Al Viro668d0cd2016-03-08 12:44:17 -05002594 * d_exact_alias - find and hash an exact unhashed alias
2595 * @entry: dentry to add
2596 * @inode: The inode to go with this dentry
2597 *
2598 * If an unhashed dentry with the same name/parent and desired
2599 * inode already exists, hash and return it. Otherwise, return
2600 * NULL.
2601 *
2602 * Parent directory should be locked.
2603 */
2604struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2605{
2606 struct dentry *alias;
Al Viro668d0cd2016-03-08 12:44:17 -05002607 unsigned int hash = entry->d_name.hash;
2608
2609 spin_lock(&inode->i_lock);
2610 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2611 /*
2612 * Don't need alias->d_lock here, because aliases with
2613 * d_parent == entry->d_parent are not subject to name or
2614 * parent changes, because the parent inode i_mutex is held.
2615 */
2616 if (alias->d_name.hash != hash)
2617 continue;
2618 if (alias->d_parent != entry->d_parent)
2619 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002620 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
Al Viro668d0cd2016-03-08 12:44:17 -05002621 continue;
2622 spin_lock(&alias->d_lock);
2623 if (!d_unhashed(alias)) {
2624 spin_unlock(&alias->d_lock);
2625 alias = NULL;
2626 } else {
2627 __dget_dlock(alias);
Al Viro15d3c582016-07-29 17:45:21 -04002628 __d_rehash(alias);
Al Viro668d0cd2016-03-08 12:44:17 -05002629 spin_unlock(&alias->d_lock);
2630 }
2631 spin_unlock(&inode->i_lock);
2632 return alias;
2633 }
2634 spin_unlock(&inode->i_lock);
2635 return NULL;
2636}
2637EXPORT_SYMBOL(d_exact_alias);
2638
2639/**
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002640 * dentry_update_name_case - update case insensitive dentry with a new name
2641 * @dentry: dentry to be updated
2642 * @name: new name
2643 *
2644 * Update a case insensitive dentry with new case of name.
2645 *
2646 * dentry must have been returned by d_lookup with name @name. Old and new
2647 * name lengths must match (ie. no d_compare which allows mismatched name
2648 * lengths).
2649 *
2650 * Parent inode i_mutex must be held over d_lookup and into this call (to
2651 * keep renames and concurrent inserts, and readdir(2) away).
2652 */
Al Viro9aba36d2016-07-20 22:28:45 -04002653void dentry_update_name_case(struct dentry *dentry, const struct qstr *name)
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002654{
Al Viro59551022016-01-22 15:40:57 -05002655 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002656 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2657
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002658 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002659 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002660 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002661 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002662 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002663}
2664EXPORT_SYMBOL(dentry_update_name_case);
2665
Al Viro8d85b482014-09-29 14:54:27 -04002666static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667{
Al Viro8d85b482014-09-29 14:54:27 -04002668 if (unlikely(dname_external(target))) {
2669 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 /*
2671 * Both external: swap the pointers
2672 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002673 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 } else {
2675 /*
2676 * dentry:internal, target:external. Steal target's
2677 * storage and make target internal.
2678 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002679 memcpy(target->d_iname, dentry->d_name.name,
2680 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 dentry->d_name.name = target->d_name.name;
2682 target->d_name.name = target->d_iname;
2683 }
2684 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002685 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 /*
2687 * dentry:external, target:internal. Give dentry's
2688 * storage to target and make dentry internal
2689 */
2690 memcpy(dentry->d_iname, target->d_name.name,
2691 target->d_name.len + 1);
2692 target->d_name.name = dentry->d_name.name;
2693 dentry->d_name.name = dentry->d_iname;
2694 } else {
2695 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002696 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002698 unsigned int i;
2699 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002700 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2701 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002702 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2703 swap(((long *) &dentry->d_iname)[i],
2704 ((long *) &target->d_iname)[i]);
2705 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 }
2707 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002708 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709}
2710
Al Viro8d85b482014-09-29 14:54:27 -04002711static void copy_name(struct dentry *dentry, struct dentry *target)
2712{
2713 struct external_name *old_name = NULL;
2714 if (unlikely(dname_external(dentry)))
2715 old_name = external_name(dentry);
2716 if (unlikely(dname_external(target))) {
2717 atomic_inc(&external_name(target)->u.count);
2718 dentry->d_name = target->d_name;
2719 } else {
2720 memcpy(dentry->d_iname, target->d_name.name,
2721 target->d_name.len + 1);
2722 dentry->d_name.name = dentry->d_iname;
2723 dentry->d_name.hash_len = target->d_name.hash_len;
2724 }
2725 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2726 kfree_rcu(old_name, u.head);
2727}
2728
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002729static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2730{
2731 /*
2732 * XXXX: do we really need to take target->d_lock?
2733 */
2734 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2735 spin_lock(&target->d_parent->d_lock);
2736 else {
2737 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2738 spin_lock(&dentry->d_parent->d_lock);
2739 spin_lock_nested(&target->d_parent->d_lock,
2740 DENTRY_D_LOCK_NESTED);
2741 } else {
2742 spin_lock(&target->d_parent->d_lock);
2743 spin_lock_nested(&dentry->d_parent->d_lock,
2744 DENTRY_D_LOCK_NESTED);
2745 }
2746 }
2747 if (target < dentry) {
2748 spin_lock_nested(&target->d_lock, 2);
2749 spin_lock_nested(&dentry->d_lock, 3);
2750 } else {
2751 spin_lock_nested(&dentry->d_lock, 2);
2752 spin_lock_nested(&target->d_lock, 3);
2753 }
2754}
2755
Al Viro986c0192014-09-26 23:11:15 -04002756static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002757{
2758 if (target->d_parent != dentry->d_parent)
2759 spin_unlock(&dentry->d_parent->d_lock);
2760 if (target->d_parent != target)
2761 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002762 spin_unlock(&target->d_lock);
2763 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002764}
2765
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002767 * When switching names, the actual string doesn't strictly have to
2768 * be preserved in the target - because we're dropping the target
2769 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002770 * the new name before we switch, unless we are going to rehash
2771 * it. Note that if we *do* unhash the target, we are not allowed
2772 * to rehash it without giving it a new name/hash key - whether
2773 * we swap or overwrite the names here, resulting name won't match
2774 * the reality in filesystem; it's only there for d_path() purposes.
2775 * Note that all of this is happening under rename_lock, so the
2776 * any hash lookup seeing it in the middle of manipulations will
2777 * be discarded anyway. So we do not care what happens to the hash
2778 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002780/*
Al Viro18367502011-07-12 21:42:24 -04002781 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 * @dentry: entry to move
2783 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002784 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 *
2786 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002787 * dcache entries should not be moved in this way. Caller must hold
2788 * rename_lock, the i_mutex of the source and target directories,
2789 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002791static void __d_move(struct dentry *dentry, struct dentry *target,
2792 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793{
Al Viro84e710d2016-04-15 00:58:55 -04002794 struct inode *dir = NULL;
2795 unsigned n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 if (!dentry->d_inode)
2797 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2798
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002799 BUG_ON(d_ancestor(dentry, target));
2800 BUG_ON(d_ancestor(target, dentry));
2801
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002802 dentry_lock_for_move(dentry, target);
Al Viro84e710d2016-04-15 00:58:55 -04002803 if (unlikely(d_in_lookup(target))) {
2804 dir = target->d_parent->d_inode;
2805 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002806 __d_lookup_done(target);
Al Viro84e710d2016-04-15 00:58:55 -04002807 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808
Nick Piggin31e6b012011-01-07 17:49:52 +11002809 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002810 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002811
Al Viro15d3c582016-07-29 17:45:21 -04002812 /* unhash both */
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002813 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002814 __d_drop(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 __d_drop(target);
2816
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002818 if (exchange)
2819 swap_names(dentry, target);
2820 else
2821 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822
Al Viro15d3c582016-07-29 17:45:21 -04002823 /* rehash in new place(s) */
2824 __d_rehash(dentry);
2825 if (exchange)
2826 __d_rehash(target);
2827
Al Viro63cf4272014-09-26 23:06:14 -04002828 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002830 /* splicing a tree */
Al Viro3d56c252016-06-07 21:26:55 -04002831 dentry->d_flags |= DCACHE_RCUACCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 dentry->d_parent = target->d_parent;
2833 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002834 list_del_init(&target->d_child);
2835 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002837 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002838 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002839 list_move(&target->d_child, &target->d_parent->d_subdirs);
2840 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002841 if (exchange)
Al Viroaffda482016-05-29 18:35:12 -04002842 fsnotify_update_flags(target);
2843 fsnotify_update_flags(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 }
2845
Nick Piggin31e6b012011-01-07 17:49:52 +11002846 write_seqcount_end(&target->d_seq);
2847 write_seqcount_end(&dentry->d_seq);
2848
Al Viro84e710d2016-04-15 00:58:55 -04002849 if (dir)
2850 end_dir_add(dir, n);
Al Viro986c0192014-09-26 23:11:15 -04002851 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002852}
2853
2854/*
2855 * d_move - move a dentry
2856 * @dentry: entry to move
2857 * @target: new dentry
2858 *
2859 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002860 * dcache entries should not be moved in this way. See the locking
2861 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002862 */
2863void d_move(struct dentry *dentry, struct dentry *target)
2864{
2865 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002866 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002868}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002869EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870
Miklos Szeredida1ce062014-04-01 17:08:43 +02002871/*
2872 * d_exchange - exchange two dentries
2873 * @dentry1: first dentry
2874 * @dentry2: second dentry
2875 */
2876void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2877{
2878 write_seqlock(&rename_lock);
2879
2880 WARN_ON(!dentry1->d_inode);
2881 WARN_ON(!dentry2->d_inode);
2882 WARN_ON(IS_ROOT(dentry1));
2883 WARN_ON(IS_ROOT(dentry2));
2884
2885 __d_move(dentry1, dentry2, true);
2886
2887 write_sequnlock(&rename_lock);
2888}
2889
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002890/**
2891 * d_ancestor - search for an ancestor
2892 * @p1: ancestor dentry
2893 * @p2: child dentry
2894 *
2895 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2896 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002897 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002898struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002899{
2900 struct dentry *p;
2901
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002902 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002903 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002904 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002905 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002906 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002907}
2908
2909/*
2910 * This helper attempts to cope with remotely renamed directories
2911 *
2912 * It assumes that the caller is already holding
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002913 * dentry->d_parent->d_inode->i_mutex, and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002914 *
2915 * Note: If ever the locking in lock_rename() changes, then please
2916 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002917 */
Al Virob5ae6b12014-10-12 22:16:02 -04002918static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002919 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002920{
Al Viro9902af72016-04-15 15:08:36 -04002921 struct mutex *m1 = NULL;
2922 struct rw_semaphore *m2 = NULL;
J. Bruce Fields3d330dc2015-02-10 10:55:53 -05002923 int ret = -ESTALE;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002924
2925 /* If alias and dentry share a parent, then no extra locks required */
2926 if (alias->d_parent == dentry->d_parent)
2927 goto out_unalias;
2928
Trond Myklebust9eaef272006-10-21 10:24:20 -07002929 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002930 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2931 goto out_err;
2932 m1 = &dentry->d_sb->s_vfs_rename_mutex;
Al Viro9902af72016-04-15 15:08:36 -04002933 if (!inode_trylock_shared(alias->d_parent->d_inode))
Trond Myklebust9eaef272006-10-21 10:24:20 -07002934 goto out_err;
Al Viro9902af72016-04-15 15:08:36 -04002935 m2 = &alias->d_parent->d_inode->i_rwsem;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002936out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002937 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04002938 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002939out_err:
Trond Myklebust9eaef272006-10-21 10:24:20 -07002940 if (m2)
Al Viro9902af72016-04-15 15:08:36 -04002941 up_read(m2);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002942 if (m1)
2943 mutex_unlock(m1);
2944 return ret;
2945}
2946
David Howells770bfad2006-08-22 20:06:07 -04002947/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002948 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2949 * @inode: the inode which may have a disconnected dentry
2950 * @dentry: a negative dentry which we want to point to the inode.
2951 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002952 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2953 * place of the given dentry and return it, else simply d_add the inode
2954 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002955 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002956 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2957 * we should error out: directories can't have multiple aliases.
2958 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002959 * This is needed in the lookup routine of any filesystem that is exportable
2960 * (via knfsd) so that we can build dcache paths to directories effectively.
2961 *
2962 * If a dentry was found and moved, then it is returned. Otherwise NULL
2963 * is returned. This matches the expected return value of ->lookup.
2964 *
2965 * Cluster filesystems may call this function with a negative, hashed dentry.
2966 * In that case, we know that the inode will be a regular file, and also this
2967 * will only occur during atomic_open. So we need to check for the dentry
2968 * being already hashed only in the final case.
2969 */
2970struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2971{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002972 if (IS_ERR(inode))
2973 return ERR_CAST(inode);
2974
David Howells770bfad2006-08-22 20:06:07 -04002975 BUG_ON(!d_unhashed(dentry));
2976
Al Virode689f52016-03-09 18:05:42 -05002977 if (!inode)
Al Virob5ae6b12014-10-12 22:16:02 -04002978 goto out;
Al Virode689f52016-03-09 18:05:42 -05002979
Al Virob9680912016-04-11 00:53:26 -04002980 security_d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11002981 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002982 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04002983 struct dentry *new = __d_find_any_alias(inode);
2984 if (unlikely(new)) {
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002985 /* The reference to new ensures it remains an alias */
2986 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002987 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002988 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04002989 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002990 dput(new);
2991 new = ERR_PTR(-ELOOP);
2992 pr_warn_ratelimited(
2993 "VFS: Lookup of '%s' in %s %s"
2994 " would have caused loop\n",
2995 dentry->d_name.name,
2996 inode->i_sb->s_type->name,
2997 inode->i_sb->s_id);
2998 } else if (!IS_ROOT(new)) {
2999 int err = __d_unalias(inode, dentry, new);
3000 write_sequnlock(&rename_lock);
3001 if (err) {
3002 dput(new);
3003 new = ERR_PTR(err);
3004 }
Al Viro18367502011-07-12 21:42:24 -04003005 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04003006 __d_move(new, dentry, false);
3007 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003008 }
Al Virob5ae6b12014-10-12 22:16:02 -04003009 iput(inode);
3010 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003011 }
David Howells770bfad2006-08-22 20:06:07 -04003012 }
Al Virob5ae6b12014-10-12 22:16:02 -04003013out:
Al Viroed782b52016-03-09 19:52:39 -05003014 __d_add(dentry, inode);
Al Virob5ae6b12014-10-12 22:16:02 -04003015 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04003016}
Al Virob5ae6b12014-10-12 22:16:02 -04003017EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04003018
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003019static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01003020{
3021 *buflen -= namelen;
3022 if (*buflen < 0)
3023 return -ENAMETOOLONG;
3024 *buffer -= namelen;
3025 memcpy(*buffer, str, namelen);
3026 return 0;
3027}
3028
Waiman Long232d2d62013-09-09 12:18:13 -04003029/**
3030 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04003031 * @buffer: buffer pointer
3032 * @buflen: allocated length of the buffer
3033 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04003034 *
3035 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3036 * make sure that either the old or the new name pointer and length are
3037 * fetched. However, there may be mismatch between length and pointer.
3038 * The length cannot be trusted, we need to copy it byte-by-byte until
3039 * the length is reached or a null byte is found. It also prepends "/" at
3040 * the beginning of the name. The sequence number check at the caller will
3041 * retry it again when a d_move() does happen. So any garbage in the buffer
3042 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04003043 *
3044 * Data dependency barrier is needed to make sure that we see that terminating
3045 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04003046 */
Al Viro9aba36d2016-07-20 22:28:45 -04003047static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003048{
Waiman Long232d2d62013-09-09 12:18:13 -04003049 const char *dname = ACCESS_ONCE(name->name);
3050 u32 dlen = ACCESS_ONCE(name->len);
3051 char *p;
3052
Al Viro6d13f692014-09-29 14:46:30 -04003053 smp_read_barrier_depends();
3054
Waiman Long232d2d62013-09-09 12:18:13 -04003055 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04003056 if (*buflen < 0)
3057 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04003058 p = *buffer -= dlen + 1;
3059 *p++ = '/';
3060 while (dlen--) {
3061 char c = *dname++;
3062 if (!c)
3063 break;
3064 *p++ = c;
3065 }
3066 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003067}
3068
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069/**
Randy Dunlap208898c2010-11-18 15:02:49 -08003070 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003071 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003072 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003073 * @buffer: pointer to the end of the buffer
3074 * @buflen: pointer to buffer length
3075 *
Waiman Long18129972013-09-12 10:55:35 -04003076 * The function will first try to write out the pathname without taking any
3077 * lock other than the RCU read lock to make sure that dentries won't go away.
3078 * It only checks the sequence number of the global rename_lock as any change
3079 * in the dentry's d_seq will be preceded by changes in the rename_lock
3080 * sequence number. If the sequence number had been changed, it will restart
3081 * the whole pathname back-tracing sequence again by taking the rename_lock.
3082 * In this case, there is no need to take the RCU read lock as the recursive
3083 * parent pointer references will keep the dentry chain alive as long as no
3084 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003085 */
Al Viro02125a82011-12-05 08:43:34 -05003086static int prepend_path(const struct path *path,
3087 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003088 char **buffer, int *buflen)
3089{
Al Viroede4ceb2013-11-13 07:45:40 -05003090 struct dentry *dentry;
3091 struct vfsmount *vfsmnt;
3092 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003093 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04003094 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04003095 char *bptr;
3096 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003097
Al Viro48f5ec22013-09-09 15:22:25 -04003098 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04003099restart_mnt:
3100 read_seqbegin_or_lock(&mount_lock, &m_seq);
3101 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003102 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003103restart:
3104 bptr = *buffer;
3105 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04003106 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05003107 dentry = path->dentry;
3108 vfsmnt = path->mnt;
3109 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04003110 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003111 while (dentry != root->dentry || vfsmnt != root->mnt) {
3112 struct dentry * parent;
3113
3114 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04003115 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Eric W. Biedermancde93be2015-08-15 13:36:12 -05003116 /* Escaped? */
3117 if (dentry != vfsmnt->mnt_root) {
3118 bptr = *buffer;
3119 blen = *buflen;
3120 error = 3;
3121 break;
3122 }
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003123 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04003124 if (mnt != parent) {
3125 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
3126 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04003127 vfsmnt = &mnt->mnt;
3128 continue;
3129 }
Waiman Long232d2d62013-09-09 12:18:13 -04003130 if (!error)
3131 error = is_mounted(vfsmnt) ? 1 : 2;
3132 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003133 }
3134 parent = dentry->d_parent;
3135 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003136 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003137 if (error)
3138 break;
3139
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003140 dentry = parent;
3141 }
Al Viro48f5ec22013-09-09 15:22:25 -04003142 if (!(seq & 1))
3143 rcu_read_unlock();
3144 if (need_seqretry(&rename_lock, seq)) {
3145 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003146 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003147 }
Al Viro48f5ec22013-09-09 15:22:25 -04003148 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003149
3150 if (!(m_seq & 1))
3151 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04003152 if (need_seqretry(&mount_lock, m_seq)) {
3153 m_seq = 1;
3154 goto restart_mnt;
3155 }
3156 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003157
Waiman Long232d2d62013-09-09 12:18:13 -04003158 if (error >= 0 && bptr == *buffer) {
3159 if (--blen < 0)
3160 error = -ENAMETOOLONG;
3161 else
3162 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003163 }
Waiman Long232d2d62013-09-09 12:18:13 -04003164 *buffer = bptr;
3165 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04003166 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003167}
3168
3169/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003170 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003171 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003172 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07003173 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07003174 * @buflen: buffer length
3175 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003176 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003177 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003178 * Returns a pointer into the buffer or an error code if the
3179 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08003180 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04003181 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003182 *
Al Viro02125a82011-12-05 08:43:34 -05003183 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 */
Al Viro02125a82011-12-05 08:43:34 -05003185char *__d_path(const struct path *path,
3186 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003187 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003189 char *res = buf + buflen;
3190 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003192 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003193 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04003194
Al Viro02125a82011-12-05 08:43:34 -05003195 if (error < 0)
3196 return ERR_PTR(error);
3197 if (error > 0)
3198 return NULL;
3199 return res;
3200}
3201
3202char *d_absolute_path(const struct path *path,
3203 char *buf, int buflen)
3204{
3205 struct path root = {};
3206 char *res = buf + buflen;
3207 int error;
3208
3209 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003210 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05003211
3212 if (error > 1)
3213 error = -EINVAL;
3214 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003215 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003216 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003217}
3218
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003219/*
3220 * same as __d_path but appends "(deleted)" for unlinked files.
3221 */
Al Viro02125a82011-12-05 08:43:34 -05003222static int path_with_deleted(const struct path *path,
3223 const struct path *root,
3224 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003225{
3226 prepend(buf, buflen, "\0", 1);
3227 if (d_unlinked(path->dentry)) {
3228 int error = prepend(buf, buflen, " (deleted)", 10);
3229 if (error)
3230 return error;
3231 }
3232
3233 return prepend_path(path, root, buf, buflen);
3234}
3235
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003236static int prepend_unreachable(char **buffer, int *buflen)
3237{
3238 return prepend(buffer, buflen, "(unreachable)", 13);
3239}
3240
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003241static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3242{
3243 unsigned seq;
3244
3245 do {
3246 seq = read_seqcount_begin(&fs->seq);
3247 *root = fs->root;
3248 } while (read_seqcount_retry(&fs->seq, seq));
3249}
3250
Jan Bluncka03a8a702008-02-14 19:38:32 -08003251/**
3252 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003253 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003254 * @buf: buffer to return value in
3255 * @buflen: buffer length
3256 *
3257 * Convert a dentry into an ASCII path name. If the entry has been deleted
3258 * the string " (deleted)" is appended. Note that this is ambiguous.
3259 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003260 * Returns a pointer into the buffer or an error code if the path was
3261 * too long. Note: Callers should use the returned pointer, not the passed
3262 * in buffer, to use the name! The implementation often starts at an offset
3263 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003264 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003265 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003266 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003267char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003269 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003270 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003271 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003272
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003273 /*
3274 * We have various synthetic filesystems that never get mounted. On
3275 * these filesystems dentries are never used for lookup purposes, and
3276 * thus don't need to be hashed. They also don't need a name until a
3277 * user wants to identify the object in /proc/pid/fd/. The little hack
3278 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003279 *
3280 * Some pseudo inodes are mountable. When they are mounted
3281 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3282 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003283 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003284 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3285 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003286 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003287
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003288 rcu_read_lock();
3289 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003290 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003291 rcu_read_unlock();
3292
Al Viro02125a82011-12-05 08:43:34 -05003293 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003294 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 return res;
3296}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003297EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298
3299/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003300 * Helper function for dentry_operations.d_dname() members
3301 */
3302char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3303 const char *fmt, ...)
3304{
3305 va_list args;
3306 char temp[64];
3307 int sz;
3308
3309 va_start(args, fmt);
3310 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3311 va_end(args);
3312
3313 if (sz > sizeof(temp) || sz > buflen)
3314 return ERR_PTR(-ENAMETOOLONG);
3315
3316 buffer += buflen - sz;
3317 return memcpy(buffer, temp, sz);
3318}
3319
Al Viro118b2302013-08-24 12:08:17 -04003320char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3321{
3322 char *end = buffer + buflen;
3323 /* these dentries are never renamed, so d_lock is not needed */
3324 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003325 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003326 prepend(&end, &buflen, "/", 1))
3327 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003328 return end;
Al Viro118b2302013-08-24 12:08:17 -04003329}
David Herrmann31bbe162014-01-03 14:09:47 +01003330EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003331
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003332/*
Ram Pai6092d042008-03-27 13:06:20 +01003333 * Write full pathname from the root of the filesystem into the buffer.
3334 */
Al Virof6500802014-01-26 12:37:55 -05003335static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003336{
Al Virof6500802014-01-26 12:37:55 -05003337 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003338 char *end, *retval;
3339 int len, seq = 0;
3340 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003341
Al Virof6500802014-01-26 12:37:55 -05003342 if (buflen < 2)
3343 goto Elong;
3344
Al Viro48f5ec22013-09-09 15:22:25 -04003345 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003346restart:
Al Virof6500802014-01-26 12:37:55 -05003347 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003348 end = buf + buflen;
3349 len = buflen;
3350 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003351 /* Get '/' right */
3352 retval = end-1;
3353 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003354 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003355 while (!IS_ROOT(dentry)) {
3356 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003357
Ram Pai6092d042008-03-27 13:06:20 +01003358 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003359 error = prepend_name(&end, &len, &dentry->d_name);
3360 if (error)
3361 break;
Ram Pai6092d042008-03-27 13:06:20 +01003362
3363 retval = end;
3364 dentry = parent;
3365 }
Al Viro48f5ec22013-09-09 15:22:25 -04003366 if (!(seq & 1))
3367 rcu_read_unlock();
3368 if (need_seqretry(&rename_lock, seq)) {
3369 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003370 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003371 }
3372 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003373 if (error)
3374 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003375 return retval;
3376Elong:
3377 return ERR_PTR(-ENAMETOOLONG);
3378}
Nick Pigginec2447c2011-01-07 17:49:29 +11003379
3380char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3381{
Waiman Long232d2d62013-09-09 12:18:13 -04003382 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003383}
3384EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003385
3386char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3387{
3388 char *p = NULL;
3389 char *retval;
3390
Al Viroc1031352010-06-06 22:31:14 -04003391 if (d_unlinked(dentry)) {
3392 p = buf + buflen;
3393 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3394 goto Elong;
3395 buflen++;
3396 }
3397 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003398 if (!IS_ERR(retval) && p)
3399 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003400 return retval;
3401Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003402 return ERR_PTR(-ENAMETOOLONG);
3403}
3404
Linus Torvalds8b19e342013-09-12 10:35:47 -07003405static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3406 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003407{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003408 unsigned seq;
3409
3410 do {
3411 seq = read_seqcount_begin(&fs->seq);
3412 *root = fs->root;
3413 *pwd = fs->pwd;
3414 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003415}
3416
Ram Pai6092d042008-03-27 13:06:20 +01003417/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418 * NOTE! The user-level library version returns a
3419 * character pointer. The kernel system call just
3420 * returns the length of the buffer filled (which
3421 * includes the ending '\0' character), or a negative
3422 * error value. So libc would do something like
3423 *
3424 * char *getcwd(char * buf, size_t size)
3425 * {
3426 * int retval;
3427 *
3428 * retval = sys_getcwd(buf, size);
3429 * if (retval >= 0)
3430 * return buf;
3431 * errno = -retval;
3432 * return NULL;
3433 * }
3434 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003435SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436{
Linus Torvalds552ce542007-02-13 12:08:18 -08003437 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003438 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003439 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440
3441 if (!page)
3442 return -ENOMEM;
3443
Linus Torvalds8b19e342013-09-12 10:35:47 -07003444 rcu_read_lock();
3445 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446
Linus Torvalds552ce542007-02-13 12:08:18 -08003447 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003448 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003449 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003450 char *cwd = page + PATH_MAX;
3451 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003452
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003453 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003454 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003455 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003456
Al Viro02125a82011-12-05 08:43:34 -05003457 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003458 goto out;
3459
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003460 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003461 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003462 error = prepend_unreachable(&cwd, &buflen);
3463 if (error)
3464 goto out;
3465 }
3466
Linus Torvalds552ce542007-02-13 12:08:18 -08003467 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003468 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003469 if (len <= size) {
3470 error = len;
3471 if (copy_to_user(buf, cwd, len))
3472 error = -EFAULT;
3473 }
Nick Piggin949854d2011-01-07 17:49:37 +11003474 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003475 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003476 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477
3478out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003479 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480 return error;
3481}
3482
3483/*
3484 * Test whether new_dentry is a subdirectory of old_dentry.
3485 *
3486 * Trivially implemented using the dcache structure
3487 */
3488
3489/**
3490 * is_subdir - is new dentry a subdirectory of old_dentry
3491 * @new_dentry: new dentry
3492 * @old_dentry: old dentry
3493 *
Yaowei Baia6e57872015-11-17 14:40:11 +08003494 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3495 * Returns false otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3497 */
3498
Yaowei Baia6e57872015-11-17 14:40:11 +08003499bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003500{
Yaowei Baia6e57872015-11-17 14:40:11 +08003501 bool result;
Nick Piggin949854d2011-01-07 17:49:37 +11003502 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003504 if (new_dentry == old_dentry)
Yaowei Baia6e57872015-11-17 14:40:11 +08003505 return true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003506
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003507 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003509 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003510 /*
3511 * Need rcu_readlock to protect against the d_parent trashing
3512 * due to d_move
3513 */
3514 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003515 if (d_ancestor(old_dentry, new_dentry))
Yaowei Baia6e57872015-11-17 14:40:11 +08003516 result = true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003517 else
Yaowei Baia6e57872015-11-17 14:40:11 +08003518 result = false;
Nick Piggin949854d2011-01-07 17:49:37 +11003519 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003521
3522 return result;
3523}
3524
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003525static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003526{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003527 struct dentry *root = data;
3528 if (dentry != root) {
3529 if (d_unhashed(dentry) || !dentry->d_inode)
3530 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003532 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3533 dentry->d_flags |= DCACHE_GENOCIDE;
3534 dentry->d_lockref.count--;
3535 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003537 return D_WALK_CONTINUE;
3538}
Nick Piggin58db63d2011-01-07 17:49:39 +11003539
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003540void d_genocide(struct dentry *parent)
3541{
3542 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543}
3544
Al Viro60545d02013-06-07 01:20:27 -04003545void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003546{
Al Viro60545d02013-06-07 01:20:27 -04003547 inode_dec_link_count(inode);
3548 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003549 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003550 !d_unlinked(dentry));
3551 spin_lock(&dentry->d_parent->d_lock);
3552 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3553 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3554 (unsigned long long)inode->i_ino);
3555 spin_unlock(&dentry->d_lock);
3556 spin_unlock(&dentry->d_parent->d_lock);
3557 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558}
Al Viro60545d02013-06-07 01:20:27 -04003559EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560
3561static __initdata unsigned long dhash_entries;
3562static int __init set_dhash_entries(char *str)
3563{
3564 if (!str)
3565 return 0;
3566 dhash_entries = simple_strtoul(str, &str, 0);
3567 return 1;
3568}
3569__setup("dhash_entries=", set_dhash_entries);
3570
3571static void __init dcache_init_early(void)
3572{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003573 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574
3575 /* If hashes are distributed across NUMA nodes, defer
3576 * hash allocation until vmalloc space is available.
3577 */
3578 if (hashdist)
3579 return;
3580
3581 dentry_hashtable =
3582 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003583 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584 dhash_entries,
3585 13,
3586 HASH_EARLY,
3587 &d_hash_shift,
3588 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003589 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590 0);
3591
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003592 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003593 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594}
3595
Denis Cheng74bf17c2007-10-16 23:26:30 -07003596static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003598 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003599
3600 /*
3601 * A constructor could be added for stable state like the lists,
3602 * but it is probably not worth it because of the cache nature
3603 * of the dcache.
3604 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003605 dentry_cache = KMEM_CACHE(dentry,
Vladimir Davydov5d097052016-01-14 15:18:21 -08003606 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003607
3608 /* Hash may have been set up in dcache_init_early */
3609 if (!hashdist)
3610 return;
3611
3612 dentry_hashtable =
3613 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003614 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003615 dhash_entries,
3616 13,
3617 0,
3618 &d_hash_shift,
3619 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003620 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621 0);
3622
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003623 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003624 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625}
3626
3627/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003628struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003629EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003630
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631EXPORT_SYMBOL(d_genocide);
3632
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633void __init vfs_caches_init_early(void)
3634{
3635 dcache_init_early();
3636 inode_init_early();
3637}
3638
Mel Gorman4248b0d2015-08-06 15:46:20 -07003639void __init vfs_caches_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003640{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003642 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643
Denis Cheng74bf17c2007-10-16 23:26:30 -07003644 dcache_init();
3645 inode_init();
Mel Gorman4248b0d2015-08-06 15:46:20 -07003646 files_init();
3647 files_maxfiles_init();
Denis Cheng74bf17c2007-10-16 23:26:30 -07003648 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649 bdev_cache_init();
3650 chrdev_init();
3651}