blob: 34c852af215c0f1ff2d73800ed0bc02c9a2f0794 [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>
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080029#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#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 Howells07f3f05c2006-09-30 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
David Howellscdf01222017-07-04 17:25:22 +010093const struct qstr empty_name = QSTR_INIT("", 0);
94EXPORT_SYMBOL(empty_name);
95const struct qstr slash_name = QSTR_INIT("/", 1);
96EXPORT_SYMBOL(slash_name);
97
Linus Torvalds1da177e2005-04-16 15:20:36 -070098/*
99 * This is the single most critical data structure when it comes
100 * to the dcache: the hashtable for lookups. Somebody should try
101 * to make this good - I've just made it work.
102 *
103 * This hash-function tries to avoid losing too many bits of hash
104 * information, yet avoid using a prime hash-size or similar.
105 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800107static unsigned int d_hash_mask __read_mostly;
108static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100109
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700110static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100111
Linus Torvalds8387ff22016-06-10 07:51:30 -0700112static inline struct hlist_bl_head *d_hash(unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100113{
George Spelvin703b5fa2016-06-10 00:22:12 -0400114 return dentry_hashtable + (hash >> (32 - d_hash_shift));
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100115}
116
Al Viro94bdd652016-04-15 02:42:04 -0400117#define IN_LOOKUP_SHIFT 10
118static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
119
120static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
121 unsigned int hash)
122{
123 hash += (unsigned long) parent / L1_CACHE_BYTES;
124 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
125}
126
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128/* Statistics gathering. */
129struct dentry_stat_t dentry_stat = {
130 .age_limit = 45,
131};
132
Glauber Costa3942c072013-08-28 10:17:53 +1000133static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000134static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400135
136#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000137
138/*
139 * Here we resort to our own counters instead of using generic per-cpu counters
140 * for consistency with what the vfs inode code does. We are expected to harvest
141 * better code and performance by having our own specialized counters.
142 *
143 * Please note that the loop is done over all possible CPUs, not over all online
144 * CPUs. The reason for this is that we don't want to play games with CPUs going
145 * on and off. If one of them goes off, we will just keep their counters.
146 *
147 * glommer: See cffbc8a for details, and if you ever intend to change this,
148 * please update all vfs counters to match.
149 */
Glauber Costa3942c072013-08-28 10:17:53 +1000150static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100151{
152 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000153 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100154 for_each_possible_cpu(i)
155 sum += per_cpu(nr_dentry, i);
156 return sum < 0 ? 0 : sum;
157}
158
Dave Chinner62d36c72013-08-28 10:17:54 +1000159static long get_nr_dentry_unused(void)
160{
161 int i;
162 long sum = 0;
163 for_each_possible_cpu(i)
164 sum += per_cpu(nr_dentry_unused, i);
165 return sum < 0 ? 0 : sum;
166}
167
Joe Perches1f7e0612014-06-06 14:38:05 -0700168int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400169 size_t *lenp, loff_t *ppos)
170{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100171 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000172 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000173 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400174}
175#endif
176
Linus Torvalds5483f182012-03-04 15:51:42 -0800177/*
178 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
179 * The strings are both count bytes long, and count is non-zero.
180 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700181#ifdef CONFIG_DCACHE_WORD_ACCESS
182
183#include <asm/word-at-a-time.h>
184/*
185 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
186 * aligned allocation for this particular component. We don't
187 * strictly need the load_unaligned_zeropad() safety, but it
188 * doesn't hurt either.
189 *
190 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
191 * need the careful unaligned handling.
192 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700193static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800194{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800195 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800196
197 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700198 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700199 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800200 if (tcount < sizeof(unsigned long))
201 break;
202 if (unlikely(a != b))
203 return 1;
204 cs += sizeof(unsigned long);
205 ct += sizeof(unsigned long);
206 tcount -= sizeof(unsigned long);
207 if (!tcount)
208 return 0;
209 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000210 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800211 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700212}
213
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800214#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700215
Linus Torvalds94753db52012-05-10 12:19:19 -0700216static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700217{
Linus Torvalds5483f182012-03-04 15:51:42 -0800218 do {
219 if (*cs != *ct)
220 return 1;
221 cs++;
222 ct++;
223 tcount--;
224 } while (tcount);
225 return 0;
226}
227
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700228#endif
229
Linus Torvalds94753db52012-05-10 12:19:19 -0700230static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
231{
Linus Torvalds94753db52012-05-10 12:19:19 -0700232 /*
233 * Be careful about RCU walk racing with rename:
Will Deacon33822902017-10-24 11:22:48 +0100234 * use 'READ_ONCE' to fetch the name pointer.
Linus Torvalds94753db52012-05-10 12:19:19 -0700235 *
236 * NOTE! Even if a rename will mean that the length
237 * was not loaded atomically, we don't care. The
238 * RCU walk will check the sequence count eventually,
239 * and catch it. And we won't overrun the buffer,
240 * because we're reading the name pointer atomically,
241 * and a dentry name is guaranteed to be properly
242 * terminated with a NUL byte.
243 *
244 * End result: even if 'len' is wrong, we'll exit
245 * early because the data cannot match (there can
246 * be no NUL in the ct/tcount data)
247 */
Will Deacon33822902017-10-24 11:22:48 +0100248 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
He Kuangae0a8432016-03-26 09:12:10 +0000249
Linus Torvalds6326c712012-05-21 16:14:04 -0700250 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700251}
252
Al Viro8d85b482014-09-29 14:54:27 -0400253struct external_name {
254 union {
255 atomic_t count;
256 struct rcu_head head;
257 } u;
258 unsigned char name[];
259};
260
261static inline struct external_name *external_name(struct dentry *dentry)
262{
263 return container_of(dentry->d_name.name, struct external_name, name[0]);
264}
265
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400266static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400268 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
269
Al Viro8d85b482014-09-29 14:54:27 -0400270 kmem_cache_free(dentry_cache, dentry);
271}
272
273static void __d_free_external(struct rcu_head *head)
274{
275 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
Al Viro8d85b482014-09-29 14:54:27 -0400276 kfree(external_name(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 kmem_cache_free(dentry_cache, dentry);
278}
279
Al Viro810bb172014-10-12 12:45:37 -0400280static inline int dname_external(const struct dentry *dentry)
281{
282 return dentry->d_name.name != dentry->d_iname;
283}
284
Al Viro49d31c22017-07-07 14:51:19 -0400285void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
286{
287 spin_lock(&dentry->d_lock);
288 if (unlikely(dname_external(dentry))) {
289 struct external_name *p = external_name(dentry);
290 atomic_inc(&p->u.count);
291 spin_unlock(&dentry->d_lock);
292 name->name = p->name;
293 } else {
294 memcpy(name->inline_name, dentry->d_iname, DNAME_INLINE_LEN);
295 spin_unlock(&dentry->d_lock);
296 name->name = name->inline_name;
297 }
298}
299EXPORT_SYMBOL(take_dentry_name_snapshot);
300
301void release_dentry_name_snapshot(struct name_snapshot *name)
302{
303 if (unlikely(name->name != name->inline_name)) {
304 struct external_name *p;
305 p = container_of(name->name, struct external_name, name[0]);
306 if (unlikely(atomic_dec_and_test(&p->u.count)))
307 kfree_rcu(p, u.head);
308 }
309}
310EXPORT_SYMBOL(release_dentry_name_snapshot);
311
David Howells4bf46a22015-03-05 14:09:22 +0000312static inline void __d_set_inode_and_type(struct dentry *dentry,
313 struct inode *inode,
314 unsigned type_flags)
315{
316 unsigned flags;
317
318 dentry->d_inode = inode;
David Howells4bf46a22015-03-05 14:09:22 +0000319 flags = READ_ONCE(dentry->d_flags);
320 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
321 flags |= type_flags;
322 WRITE_ONCE(dentry->d_flags, flags);
323}
324
David Howells4bf46a22015-03-05 14:09:22 +0000325static inline void __d_clear_type_and_inode(struct dentry *dentry)
326{
327 unsigned flags = READ_ONCE(dentry->d_flags);
328
329 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
330 WRITE_ONCE(dentry->d_flags, flags);
David Howells4bf46a22015-03-05 14:09:22 +0000331 dentry->d_inode = NULL;
332}
333
Al Virob4f03542014-04-29 23:40:14 -0400334static void dentry_free(struct dentry *dentry)
335{
Al Viro946e51f2014-10-26 19:19:16 -0400336 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400337 if (unlikely(dname_external(dentry))) {
338 struct external_name *p = external_name(dentry);
339 if (likely(atomic_dec_and_test(&p->u.count))) {
340 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
341 return;
342 }
343 }
Al Virob4f03542014-04-29 23:40:14 -0400344 /* if dentry was never visible to RCU, immediate free is OK */
345 if (!(dentry->d_flags & DCACHE_RCUACCESS))
346 __d_free(&dentry->d_u.d_rcu);
347 else
348 call_rcu(&dentry->d_u.d_rcu, __d_free);
349}
350
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351/*
352 * Release the dentry's inode, using the filesystem
Al Viro550dce02016-05-29 20:13:30 -0400353 * d_iput() operation if defined.
Nick Piggin31e6b012011-01-07 17:49:52 +1100354 */
355static void dentry_unlink_inode(struct dentry * dentry)
356 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100357 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100358{
359 struct inode *inode = dentry->d_inode;
Al Viro550dce02016-05-29 20:13:30 -0400360 bool hashed = !d_unhashed(dentry);
Al Viroa528aca2016-02-29 12:12:46 -0500361
Al Viro550dce02016-05-29 20:13:30 -0400362 if (hashed)
363 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +0000364 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400365 hlist_del_init(&dentry->d_u.d_alias);
Al Viro550dce02016-05-29 20:13:30 -0400366 if (hashed)
367 raw_write_seqcount_end(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100368 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100369 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100370 if (!inode->i_nlink)
371 fsnotify_inoderemove(inode);
372 if (dentry->d_op && dentry->d_op->d_iput)
373 dentry->d_op->d_iput(dentry, inode);
374 else
375 iput(inode);
376}
377
378/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400379 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
380 * is in use - which includes both the "real" per-superblock
381 * LRU list _and_ the DCACHE_SHRINK_LIST use.
382 *
383 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
384 * on the shrink list (ie not on the superblock LRU list).
385 *
386 * The per-cpu "nr_dentry_unused" counters are updated with
387 * the DCACHE_LRU_LIST bit.
388 *
389 * These helper functions make sure we always follow the
390 * rules. d_lock must be held by the caller.
391 */
392#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
393static void d_lru_add(struct dentry *dentry)
394{
395 D_FLAG_VERIFY(dentry, 0);
396 dentry->d_flags |= DCACHE_LRU_LIST;
397 this_cpu_inc(nr_dentry_unused);
398 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
399}
400
401static void d_lru_del(struct dentry *dentry)
402{
403 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
404 dentry->d_flags &= ~DCACHE_LRU_LIST;
405 this_cpu_dec(nr_dentry_unused);
406 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
407}
408
409static void d_shrink_del(struct dentry *dentry)
410{
411 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
412 list_del_init(&dentry->d_lru);
413 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
414 this_cpu_dec(nr_dentry_unused);
415}
416
417static void d_shrink_add(struct dentry *dentry, struct list_head *list)
418{
419 D_FLAG_VERIFY(dentry, 0);
420 list_add(&dentry->d_lru, list);
421 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
422 this_cpu_inc(nr_dentry_unused);
423}
424
425/*
426 * These can only be called under the global LRU lock, ie during the
427 * callback for freeing the LRU list. "isolate" removes it from the
428 * LRU lists entirely, while shrink_move moves it to the indicated
429 * private list.
430 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800431static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400432{
433 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
434 dentry->d_flags &= ~DCACHE_LRU_LIST;
435 this_cpu_dec(nr_dentry_unused);
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800436 list_lru_isolate(lru, &dentry->d_lru);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400437}
438
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800439static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
440 struct list_head *list)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400441{
442 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
443 dentry->d_flags |= DCACHE_SHRINK_LIST;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800444 list_lru_isolate_move(lru, &dentry->d_lru, list);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400445}
446
447/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000448 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700449 */
450static void dentry_lru_add(struct dentry *dentry)
451{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400452 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
453 d_lru_add(dentry);
Josef Bacik563f4002017-04-18 16:04:17 -0400454 else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
455 dentry->d_flags |= DCACHE_REFERENCED;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700456}
457
Miklos Szeredid52b9082007-05-08 00:23:46 -0700458/**
Nick Piggin789680d2011-01-07 17:49:30 +1100459 * d_drop - drop a dentry
460 * @dentry: dentry to drop
461 *
462 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
463 * be found through a VFS lookup any more. Note that this is different from
464 * deleting the dentry - d_delete will try to mark the dentry negative if
465 * possible, giving a successful _negative_ lookup, while d_drop will
466 * just make the cache lookup fail.
467 *
468 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
469 * reason (NFS timeouts or autofs deletes).
470 *
471 * __d_drop requires dentry->d_lock.
472 */
473void __d_drop(struct dentry *dentry)
474{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700475 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400476 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400477 /*
478 * Hashed dentries are normally on the dentry hashtable,
479 * with the exception of those newly allocated by
480 * d_obtain_alias, which are always IS_ROOT:
481 */
482 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400483 b = &dentry->d_sb->s_anon;
484 else
Linus Torvalds8387ff22016-06-10 07:51:30 -0700485 b = d_hash(dentry->d_name.hash);
Al Virob61625d2013-10-04 11:09:01 -0400486
487 hlist_bl_lock(b);
488 __hlist_bl_del(&dentry->d_hash);
489 dentry->d_hash.pprev = NULL;
490 hlist_bl_unlock(b);
Al Virod6141462016-07-28 13:05:50 -0400491 /* After this call, in-progress rcu-walk path lookup will fail. */
492 write_seqcount_invalidate(&dentry->d_seq);
Nick Piggin789680d2011-01-07 17:49:30 +1100493 }
494}
495EXPORT_SYMBOL(__d_drop);
496
497void d_drop(struct dentry *dentry)
498{
Nick Piggin789680d2011-01-07 17:49:30 +1100499 spin_lock(&dentry->d_lock);
500 __d_drop(dentry);
501 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100502}
503EXPORT_SYMBOL(d_drop);
504
Al Viroba65dc52016-06-10 11:32:47 -0400505static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
506{
507 struct dentry *next;
508 /*
509 * Inform d_walk() and shrink_dentry_list() that we are no longer
510 * attached to the dentry tree
511 */
512 dentry->d_flags |= DCACHE_DENTRY_KILLED;
513 if (unlikely(list_empty(&dentry->d_child)))
514 return;
515 __list_del_entry(&dentry->d_child);
516 /*
517 * Cursors can move around the list of children. While we'd been
518 * a normal list member, it didn't matter - ->d_child.next would've
519 * been updated. However, from now on it won't be and for the
520 * things like d_walk() it might end up with a nasty surprise.
521 * Normally d_walk() doesn't care about cursors moving around -
522 * ->d_lock on parent prevents that and since a cursor has no children
523 * of its own, we get through it without ever unlocking the parent.
524 * There is one exception, though - if we ascend from a child that
525 * gets killed as soon as we unlock it, the next sibling is found
526 * using the value left in its ->d_child.next. And if _that_
527 * pointed to a cursor, and cursor got moved (e.g. by lseek())
528 * before d_walk() regains parent->d_lock, we'll end up skipping
529 * everything the cursor had been moved past.
530 *
531 * Solution: make sure that the pointer left behind in ->d_child.next
532 * points to something that won't be moving around. I.e. skip the
533 * cursors.
534 */
535 while (dentry->d_child.next != &parent->d_subdirs) {
536 next = list_entry(dentry->d_child.next, struct dentry, d_child);
537 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
538 break;
539 dentry->d_child.next = next->d_child.next;
540 }
541}
542
Al Viroe55fd012014-05-28 13:51:12 -0400543static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100544{
Al Viro41edf272014-05-01 10:30:00 -0400545 struct dentry *parent = NULL;
546 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400547 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100548 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100549
Linus Torvalds0d984392013-09-08 13:46:52 -0700550 /*
551 * The dentry is now unrecoverably dead to the world.
552 */
553 lockref_mark_dead(&dentry->d_lockref);
554
Sage Weilf0023bc2011-10-28 10:02:42 -0700555 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700556 * inform the fs via d_prune that this dentry is about to be
557 * unhashed and destroyed.
558 */
Al Viro29266202014-05-30 11:39:02 -0400559 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800560 dentry->d_op->d_prune(dentry);
561
Al Viro01b60352014-04-29 23:42:52 -0400562 if (dentry->d_flags & DCACHE_LRU_LIST) {
563 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
564 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400565 }
Nick Piggin77812a12011-01-07 17:49:48 +1100566 /* if it was on the hash then remove it */
567 __d_drop(dentry);
Al Viroba65dc52016-06-10 11:32:47 -0400568 dentry_unlist(dentry, parent);
Al Viro03b3b882014-04-29 15:45:28 -0400569 if (parent)
570 spin_unlock(&parent->d_lock);
Al Viro550dce02016-05-29 20:13:30 -0400571 if (dentry->d_inode)
572 dentry_unlink_inode(dentry);
573 else
574 spin_unlock(&dentry->d_lock);
Al Viro03b3b882014-04-29 15:45:28 -0400575 this_cpu_dec(nr_dentry);
576 if (dentry->d_op && dentry->d_op->d_release)
577 dentry->d_op->d_release(dentry);
578
Al Viro41edf272014-05-01 10:30:00 -0400579 spin_lock(&dentry->d_lock);
580 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
581 dentry->d_flags |= DCACHE_MAY_FREE;
582 can_free = false;
583 }
584 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400585 if (likely(can_free))
586 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400587}
588
589/*
590 * Finish off a dentry we've decided to kill.
591 * dentry->d_lock must be held, returns with it unlocked.
592 * If ref is non-zero, then decrement the refcount too.
593 * Returns dentry requiring refcount drop, or NULL if we're done.
594 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400595static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400596 __releases(dentry->d_lock)
597{
598 struct inode *inode = dentry->d_inode;
599 struct dentry *parent = NULL;
600
601 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
602 goto failed;
603
604 if (!IS_ROOT(dentry)) {
605 parent = dentry->d_parent;
606 if (unlikely(!spin_trylock(&parent->d_lock))) {
607 if (inode)
608 spin_unlock(&inode->i_lock);
609 goto failed;
610 }
611 }
612
613 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400614 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400615
616failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400617 spin_unlock(&dentry->d_lock);
Al Viroe55fd012014-05-28 13:51:12 -0400618 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100619}
620
Al Viro046b9612014-05-29 08:54:52 -0400621static inline struct dentry *lock_parent(struct dentry *dentry)
622{
623 struct dentry *parent = dentry->d_parent;
624 if (IS_ROOT(dentry))
625 return NULL;
Linus Torvalds360f5472015-01-09 15:19:03 -0800626 if (unlikely(dentry->d_lockref.count < 0))
Al Viroc2338f22014-06-12 00:29:13 -0400627 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400628 if (likely(spin_trylock(&parent->d_lock)))
629 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400630 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400631 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400632again:
633 parent = ACCESS_ONCE(dentry->d_parent);
634 spin_lock(&parent->d_lock);
635 /*
636 * We can't blindly lock dentry until we are sure
637 * that we won't violate the locking order.
638 * Any changes of dentry->d_parent must have
639 * been done with parent->d_lock held, so
640 * spin_lock() above is enough of a barrier
641 * for checking if it's still our child.
642 */
643 if (unlikely(parent != dentry->d_parent)) {
644 spin_unlock(&parent->d_lock);
645 goto again;
646 }
647 rcu_read_unlock();
648 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700649 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400650 else
651 parent = NULL;
652 return parent;
653}
654
Linus Torvalds360f5472015-01-09 15:19:03 -0800655/*
656 * Try to do a lockless dput(), and return whether that was successful.
657 *
658 * If unsuccessful, we return false, having already taken the dentry lock.
659 *
660 * The caller needs to hold the RCU read lock, so that the dentry is
661 * guaranteed to stay around even if the refcount goes down to zero!
662 */
663static inline bool fast_dput(struct dentry *dentry)
664{
665 int ret;
666 unsigned int d_flags;
667
668 /*
669 * If we have a d_op->d_delete() operation, we sould not
Al Viro75a6f822015-07-08 02:42:38 +0100670 * let the dentry count go to zero, so use "put_or_lock".
Linus Torvalds360f5472015-01-09 15:19:03 -0800671 */
672 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
673 return lockref_put_or_lock(&dentry->d_lockref);
674
675 /*
676 * .. otherwise, we can try to just decrement the
677 * lockref optimistically.
678 */
679 ret = lockref_put_return(&dentry->d_lockref);
680
681 /*
682 * If the lockref_put_return() failed due to the lock being held
683 * by somebody else, the fast path has failed. We will need to
684 * get the lock, and then check the count again.
685 */
686 if (unlikely(ret < 0)) {
687 spin_lock(&dentry->d_lock);
688 if (dentry->d_lockref.count > 1) {
689 dentry->d_lockref.count--;
690 spin_unlock(&dentry->d_lock);
691 return 1;
692 }
693 return 0;
694 }
695
696 /*
697 * If we weren't the last ref, we're done.
698 */
699 if (ret)
700 return 1;
701
702 /*
703 * Careful, careful. The reference count went down
704 * to zero, but we don't hold the dentry lock, so
705 * somebody else could get it again, and do another
706 * dput(), and we need to not race with that.
707 *
708 * However, there is a very special and common case
709 * where we don't care, because there is nothing to
710 * do: the dentry is still hashed, it does not have
711 * a 'delete' op, and it's referenced and already on
712 * the LRU list.
713 *
714 * NOTE! Since we aren't locked, these values are
715 * not "stable". However, it is sufficient that at
716 * some point after we dropped the reference the
717 * dentry was hashed and the flags had the proper
718 * value. Other dentry users may have re-gotten
719 * a reference to the dentry and change that, but
720 * our work is done - we can leave the dentry
721 * around with a zero refcount.
722 */
723 smp_rmb();
724 d_flags = ACCESS_ONCE(dentry->d_flags);
Al Viro75a6f822015-07-08 02:42:38 +0100725 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
Linus Torvalds360f5472015-01-09 15:19:03 -0800726
727 /* Nothing to do? Dropping the reference was all we needed? */
728 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
729 return 1;
730
731 /*
732 * Not the fast normal case? Get the lock. We've already decremented
733 * the refcount, but we'll need to re-check the situation after
734 * getting the lock.
735 */
736 spin_lock(&dentry->d_lock);
737
738 /*
739 * Did somebody else grab a reference to it in the meantime, and
740 * we're no longer the last user after all? Alternatively, somebody
741 * else could have killed it and marked it dead. Either way, we
742 * don't need to do anything else.
743 */
744 if (dentry->d_lockref.count) {
745 spin_unlock(&dentry->d_lock);
746 return 1;
747 }
748
749 /*
750 * Re-get the reference we optimistically dropped. We hold the
751 * lock, and we just tested that it was zero, so we can just
752 * set it to 1.
753 */
754 dentry->d_lockref.count = 1;
755 return 0;
756}
757
758
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759/*
760 * This is dput
761 *
762 * This is complicated by the fact that we do not want to put
763 * dentries that are no longer on any hash chain on the unused
764 * list: we'd much rather just get rid of them immediately.
765 *
766 * However, that implies that we have to traverse the dentry
767 * tree upwards to the parents which might _also_ now be
768 * scheduled for deletion (it may have been only waiting for
769 * its last child to go away).
770 *
771 * This tail recursion is done by hand as we don't want to depend
772 * on the compiler to always get this right (gcc generally doesn't).
773 * Real recursion would eat up our stack space.
774 */
775
776/*
777 * dput - release a dentry
778 * @dentry: dentry to release
779 *
780 * Release a dentry. This will drop the usage count and if appropriate
781 * call the dentry unlink method as well as removing it from the queues and
782 * releasing its resources. If the parent dentries were scheduled for release
783 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785void dput(struct dentry *dentry)
786{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700787 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 return;
789
790repeat:
Wei Fang47be6182016-07-06 11:32:20 +0800791 might_sleep();
792
Linus Torvalds360f5472015-01-09 15:19:03 -0800793 rcu_read_lock();
794 if (likely(fast_dput(dentry))) {
795 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 return;
Linus Torvalds360f5472015-01-09 15:19:03 -0800797 }
798
799 /* Slow case: now with the dentry lock held */
800 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
Al Viro85c7f812016-04-14 19:52:13 -0400802 WARN_ON(d_in_lookup(dentry));
803
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700804 /* Unreachable? Get rid of it */
805 if (unlikely(d_unhashed(dentry)))
806 goto kill_it;
807
Al Viro75a6f822015-07-08 02:42:38 +0100808 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
809 goto kill_it;
810
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700811 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100813 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 }
Nick Piggin265ac902010-10-10 05:36:24 -0400815
Christoph Hellwiga4633352010-10-10 05:36:26 -0400816 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400817
Waiman Long98474232013-08-28 18:24:59 -0700818 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100819 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 return;
821
Miklos Szeredid52b9082007-05-08 00:23:46 -0700822kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400823 dentry = dentry_kill(dentry);
Wei Fang47be6182016-07-06 11:32:20 +0800824 if (dentry) {
825 cond_resched();
Miklos Szeredid52b9082007-05-08 00:23:46 -0700826 goto repeat;
Wei Fang47be6182016-07-06 11:32:20 +0800827 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700829EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100832/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100833static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834{
Waiman Long98474232013-08-28 18:24:59 -0700835 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836}
837
Nick Piggindc0474b2011-01-07 17:49:43 +1100838static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100839{
Waiman Long98474232013-08-28 18:24:59 -0700840 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100841}
842
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100843struct dentry *dget_parent(struct dentry *dentry)
844{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700845 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100846 struct dentry *ret;
847
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700848 /*
849 * Do optimistic parent lookup without any
850 * locking.
851 */
852 rcu_read_lock();
853 ret = ACCESS_ONCE(dentry->d_parent);
854 gotref = lockref_get_not_zero(&ret->d_lockref);
855 rcu_read_unlock();
856 if (likely(gotref)) {
857 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
858 return ret;
859 dput(ret);
860 }
861
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100862repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100863 /*
864 * Don't need rcu_dereference because we re-check it was correct under
865 * the lock.
866 */
867 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100868 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100869 spin_lock(&ret->d_lock);
870 if (unlikely(ret != dentry->d_parent)) {
871 spin_unlock(&ret->d_lock);
872 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100873 goto repeat;
874 }
Nick Piggina734eb42011-01-07 17:49:44 +1100875 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700876 BUG_ON(!ret->d_lockref.count);
877 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100878 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100879 return ret;
880}
881EXPORT_SYMBOL(dget_parent);
882
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883/**
884 * d_find_alias - grab a hashed alias of inode
885 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 *
887 * If inode has a hashed alias, or is a directory and has any alias,
888 * acquire the reference to alias and return it. Otherwise return NULL.
889 * Notice that if inode is a directory there can be only one alias and
890 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800891 * of a filesystem, or if the directory was renamed and d_revalidate
892 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700894 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500895 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500897static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898{
Nick Pigginda502952011-01-07 17:49:33 +1100899 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
Nick Pigginda502952011-01-07 17:49:33 +1100901again:
902 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400903 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100904 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700906 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100907 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500909 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100910 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100911 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 return alias;
913 }
914 }
Nick Pigginda502952011-01-07 17:49:33 +1100915 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 }
Nick Pigginda502952011-01-07 17:49:33 +1100917 if (discon_alias) {
918 alias = discon_alias;
919 spin_lock(&alias->d_lock);
920 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500921 __dget_dlock(alias);
922 spin_unlock(&alias->d_lock);
923 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100924 }
925 spin_unlock(&alias->d_lock);
926 goto again;
927 }
928 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929}
930
Nick Pigginda502952011-01-07 17:49:33 +1100931struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932{
David Howells214fda12006-03-25 03:06:36 -0800933 struct dentry *de = NULL;
934
Al Virob3d9b7a2012-06-09 13:51:19 -0400935 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100936 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500937 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100938 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800939 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 return de;
941}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700942EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944/*
945 * Try to kill dentries associated with this inode.
946 * WARNING: you must own a reference to inode.
947 */
948void d_prune_aliases(struct inode *inode)
949{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700950 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100952 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400953 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700955 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400956 struct dentry *parent = lock_parent(dentry);
957 if (likely(!dentry->d_lockref.count)) {
958 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800959 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400960 goto restart;
961 }
962 if (parent)
963 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 }
965 spin_unlock(&dentry->d_lock);
966 }
Nick Piggin873feea2011-01-07 17:50:06 +1100967 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700969EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400971static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972{
Al Viro5c47e6d2014-04-29 16:13:18 -0400973 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700974
Miklos Szeredi60942f22014-05-02 15:38:39 -0400975 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400976 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400977 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100978 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400979 parent = lock_parent(dentry);
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000982 * The dispose list is isolated and dentries are not accounted
983 * to the LRU here, so we can simply remove it from the list
984 * here regardless of whether it is referenced or not.
985 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400986 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000987
988 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000990 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800992 if (dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700993 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400994 if (parent)
995 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 continue;
997 }
Nick Piggin77812a12011-01-07 17:49:48 +1100998
Al Viro64fd72e2014-05-28 09:48:44 -0400999
1000 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
1001 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
1002 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -04001003 if (parent)
1004 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -04001005 if (can_free)
1006 dentry_free(dentry);
1007 continue;
1008 }
1009
Al Viroff2fde92014-05-28 13:59:13 -04001010 inode = dentry->d_inode;
1011 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001012 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001013 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -04001014 if (parent)
1015 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -04001016 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001017 }
Al Viroff2fde92014-05-28 13:59:13 -04001018
Al Viroff2fde92014-05-28 13:59:13 -04001019 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -04001020
Al Viro5c47e6d2014-04-29 16:13:18 -04001021 /*
1022 * We need to prune ancestors too. This is necessary to prevent
1023 * quadratic behavior of shrink_dcache_parent(), but is also
1024 * expected to be beneficial in reducing dentry cache
1025 * fragmentation.
1026 */
1027 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -04001028 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1029 parent = lock_parent(dentry);
1030 if (dentry->d_lockref.count != 1) {
1031 dentry->d_lockref.count--;
1032 spin_unlock(&dentry->d_lock);
1033 if (parent)
1034 spin_unlock(&parent->d_lock);
1035 break;
1036 }
1037 inode = dentry->d_inode; /* can't be NULL */
1038 if (unlikely(!spin_trylock(&inode->i_lock))) {
1039 spin_unlock(&dentry->d_lock);
1040 if (parent)
1041 spin_unlock(&parent->d_lock);
1042 cpu_relax();
1043 continue;
1044 }
1045 __dentry_kill(dentry);
1046 dentry = parent;
1047 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001049}
1050
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001051static enum lru_status dentry_lru_isolate(struct list_head *item,
1052 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +10001053{
1054 struct list_head *freeable = arg;
1055 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1056
1057
1058 /*
1059 * we are inverting the lru lock/dentry->d_lock here,
1060 * so use a trylock. If we fail to get the lock, just skip
1061 * it
1062 */
1063 if (!spin_trylock(&dentry->d_lock))
1064 return LRU_SKIP;
1065
1066 /*
1067 * Referenced dentries are still in use. If they have active
1068 * counts, just remove them from the LRU. Otherwise give them
1069 * another pass through the LRU.
1070 */
1071 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001072 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001073 spin_unlock(&dentry->d_lock);
1074 return LRU_REMOVED;
1075 }
1076
1077 if (dentry->d_flags & DCACHE_REFERENCED) {
1078 dentry->d_flags &= ~DCACHE_REFERENCED;
1079 spin_unlock(&dentry->d_lock);
1080
1081 /*
1082 * The list move itself will be made by the common LRU code. At
1083 * this point, we've dropped the dentry->d_lock but keep the
1084 * lru lock. This is safe to do, since every list movement is
1085 * protected by the lru lock even if both locks are held.
1086 *
1087 * This is guaranteed by the fact that all LRU management
1088 * functions are intermediated by the LRU API calls like
1089 * list_lru_add and list_lru_del. List movement in this file
1090 * only ever occur through this functions or through callbacks
1091 * like this one, that are called from the LRU API.
1092 *
1093 * The only exceptions to this are functions like
1094 * shrink_dentry_list, and code that first checks for the
1095 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1096 * operating only with stack provided lists after they are
1097 * properly isolated from the main list. It is thus, always a
1098 * local access.
1099 */
1100 return LRU_ROTATE;
1101 }
1102
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001103 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001104 spin_unlock(&dentry->d_lock);
1105
1106 return LRU_REMOVED;
1107}
1108
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001109/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001110 * prune_dcache_sb - shrink the dcache
1111 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001112 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001113 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001114 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1115 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001116 * function.
1117 *
1118 * This function may fail to free any resources if all the dentries are in
1119 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001120 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001121long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001122{
Dave Chinnerf6041562013-08-28 10:18:00 +10001123 LIST_HEAD(dispose);
1124 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001125
Vladimir Davydov503c3582015-02-12 14:58:47 -08001126 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1127 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001128 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001129 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130}
1131
Glauber Costa4e717f52013-08-28 10:18:03 +10001132static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001133 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001134{
Glauber Costa4e717f52013-08-28 10:18:03 +10001135 struct list_head *freeable = arg;
1136 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001137
Glauber Costa4e717f52013-08-28 10:18:03 +10001138 /*
1139 * we are inverting the lru lock/dentry->d_lock here,
1140 * so use a trylock. If we fail to get the lock, just skip
1141 * it
1142 */
1143 if (!spin_trylock(&dentry->d_lock))
1144 return LRU_SKIP;
1145
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001146 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001147 spin_unlock(&dentry->d_lock);
1148
1149 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001150}
1151
Glauber Costa4e717f52013-08-28 10:18:03 +10001152
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001153/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 * shrink_dcache_sb - shrink dcache for a superblock
1155 * @sb: superblock
1156 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001157 * Shrink the dcache for the specified super block. This is used to free
1158 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001160void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161{
Glauber Costa4e717f52013-08-28 10:18:03 +10001162 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001163
Glauber Costa4e717f52013-08-28 10:18:03 +10001164 do {
1165 LIST_HEAD(dispose);
1166
1167 freed = list_lru_walk(&sb->s_dentry_lru,
Sahitya Tummalab17c0702017-07-10 15:50:00 -07001168 dentry_lru_isolate_shrink, &dispose, 1024);
Glauber Costa4e717f52013-08-28 10:18:03 +10001169
1170 this_cpu_sub(nr_dentry_unused, freed);
1171 shrink_dentry_list(&dispose);
Sahitya Tummalab17c0702017-07-10 15:50:00 -07001172 cond_resched();
1173 } while (list_lru_count(&sb->s_dentry_lru) > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001175EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001178 * enum d_walk_ret - action to talke during tree walk
1179 * @D_WALK_CONTINUE: contrinue walk
1180 * @D_WALK_QUIT: quit walk
1181 * @D_WALK_NORETRY: quit when retry is needed
1182 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001184enum d_walk_ret {
1185 D_WALK_CONTINUE,
1186 D_WALK_QUIT,
1187 D_WALK_NORETRY,
1188 D_WALK_SKIP,
1189};
1190
1191/**
1192 * d_walk - walk the dentry tree
1193 * @parent: start of walk
1194 * @data: data passed to @enter() and @finish()
1195 * @enter: callback when first entering the dentry
1196 * @finish: callback when successfully finished the walk
1197 *
1198 * The @enter() and @finish() callbacks are called with d_lock held.
1199 */
1200static void d_walk(struct dentry *parent, void *data,
1201 enum d_walk_ret (*enter)(void *, struct dentry *),
1202 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203{
Nick Piggin949854d2011-01-07 17:49:37 +11001204 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001206 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001207 enum d_walk_ret ret;
1208 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001209
Nick Piggin58db63d2011-01-07 17:49:39 +11001210again:
Al Viro48f5ec22013-09-09 15:22:25 -04001211 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001212 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001213 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001214
1215 ret = enter(data, this_parent);
1216 switch (ret) {
1217 case D_WALK_CONTINUE:
1218 break;
1219 case D_WALK_QUIT:
1220 case D_WALK_SKIP:
1221 goto out_unlock;
1222 case D_WALK_NORETRY:
1223 retry = false;
1224 break;
1225 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226repeat:
1227 next = this_parent->d_subdirs.next;
1228resume:
1229 while (next != &this_parent->d_subdirs) {
1230 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001231 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001233
Al Viroba65dc52016-06-10 11:32:47 -04001234 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1235 continue;
1236
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001237 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001238
1239 ret = enter(data, dentry);
1240 switch (ret) {
1241 case D_WALK_CONTINUE:
1242 break;
1243 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001244 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001245 goto out_unlock;
1246 case D_WALK_NORETRY:
1247 retry = false;
1248 break;
1249 case D_WALK_SKIP:
1250 spin_unlock(&dentry->d_lock);
1251 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001252 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001253
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001255 spin_unlock(&this_parent->d_lock);
1256 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001258 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 goto repeat;
1260 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001261 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 }
1263 /*
1264 * All done at this level ... ascend and resume the search.
1265 */
Al Viroca5358e2014-10-26 19:31:10 -04001266 rcu_read_lock();
1267ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001269 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001270 this_parent = child->d_parent;
1271
Al Viro31dec132013-10-25 17:04:27 -04001272 spin_unlock(&child->d_lock);
1273 spin_lock(&this_parent->d_lock);
1274
Al Viroca5358e2014-10-26 19:31:10 -04001275 /* might go back up the wrong parent if we have had a rename. */
1276 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001277 goto rename_retry;
Al Viro21591842015-05-28 23:09:19 -04001278 /* go into the first sibling still alive */
1279 do {
1280 next = child->d_child.next;
Al Viroca5358e2014-10-26 19:31:10 -04001281 if (next == &this_parent->d_subdirs)
1282 goto ascend;
1283 child = list_entry(next, struct dentry, d_child);
Al Viro21591842015-05-28 23:09:19 -04001284 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
Al Viro31dec132013-10-25 17:04:27 -04001285 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 goto resume;
1287 }
Al Viroca5358e2014-10-26 19:31:10 -04001288 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001289 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001290 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001291 if (finish)
1292 finish(data);
1293
1294out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001295 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001296 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001297 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001298
1299rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001300 spin_unlock(&this_parent->d_lock);
1301 rcu_read_unlock();
1302 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001303 if (!retry)
1304 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001305 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001306 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001308
Ian Kent01619492016-11-24 08:03:41 +11001309struct check_mount {
1310 struct vfsmount *mnt;
1311 unsigned int mounted;
1312};
1313
1314static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1315{
1316 struct check_mount *info = data;
1317 struct path path = { .mnt = info->mnt, .dentry = dentry };
1318
1319 if (likely(!d_mountpoint(dentry)))
1320 return D_WALK_CONTINUE;
1321 if (__path_is_mountpoint(&path)) {
1322 info->mounted = 1;
1323 return D_WALK_QUIT;
1324 }
1325 return D_WALK_CONTINUE;
1326}
1327
1328/**
1329 * path_has_submounts - check for mounts over a dentry in the
1330 * current namespace.
1331 * @parent: path to check.
1332 *
1333 * Return true if the parent or its subdirectories contain
1334 * a mount point in the current namespace.
1335 */
1336int path_has_submounts(const struct path *parent)
1337{
1338 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1339
1340 read_seqlock_excl(&mount_lock);
1341 d_walk(parent->dentry, &data, path_check_mount, NULL);
1342 read_sequnlock_excl(&mount_lock);
1343
1344 return data.mounted;
1345}
1346EXPORT_SYMBOL(path_has_submounts);
1347
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001349 * Called by mount code to set a mountpoint and check if the mountpoint is
1350 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1351 * subtree can become unreachable).
1352 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001353 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001354 * this reason take rename_lock and d_lock on dentry and ancestors.
1355 */
1356int d_set_mounted(struct dentry *dentry)
1357{
1358 struct dentry *p;
1359 int ret = -ENOENT;
1360 write_seqlock(&rename_lock);
1361 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001362 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001363 spin_lock(&p->d_lock);
1364 if (unlikely(d_unhashed(p))) {
1365 spin_unlock(&p->d_lock);
1366 goto out;
1367 }
1368 spin_unlock(&p->d_lock);
1369 }
1370 spin_lock(&dentry->d_lock);
1371 if (!d_unlinked(dentry)) {
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13001372 ret = -EBUSY;
1373 if (!d_mountpoint(dentry)) {
1374 dentry->d_flags |= DCACHE_MOUNTED;
1375 ret = 0;
1376 }
Miklos Szeredieed81002013-09-05 14:39:11 +02001377 }
1378 spin_unlock(&dentry->d_lock);
1379out:
1380 write_sequnlock(&rename_lock);
1381 return ret;
1382}
1383
1384/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001385 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 * and move any unused dentries to the end of the unused
1387 * list for prune_dcache(). We descend to the next level
1388 * whenever the d_subdirs list is non-empty and continue
1389 * searching.
1390 *
1391 * It returns zero iff there are no unused children,
1392 * otherwise it returns the number of children moved to
1393 * the end of the unused list. This may not be the total
1394 * number of unused children, because select_parent can
1395 * drop the lock and return early due to latency
1396 * constraints.
1397 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001398
1399struct select_data {
1400 struct dentry *start;
1401 struct list_head dispose;
1402 int found;
1403};
1404
1405static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001407 struct select_data *data = _data;
1408 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001410 if (data->start == dentry)
1411 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
Al Virofe915222014-05-03 00:02:25 -04001413 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001414 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001415 } else {
1416 if (dentry->d_flags & DCACHE_LRU_LIST)
1417 d_lru_del(dentry);
1418 if (!dentry->d_lockref.count) {
1419 d_shrink_add(dentry, &data->dispose);
1420 data->found++;
1421 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 }
1423 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001424 * We can return to the caller if we have found some (this
1425 * ensures forward progress). We'll be coming back to find
1426 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 */
Al Virofe915222014-05-03 00:02:25 -04001428 if (!list_empty(&data->dispose))
1429 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001431 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432}
1433
1434/**
1435 * shrink_dcache_parent - prune dcache
1436 * @parent: parent of entries to prune
1437 *
1438 * Prune the dcache to remove unused children of the parent dentry.
1439 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001440void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001442 for (;;) {
1443 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001445 INIT_LIST_HEAD(&data.dispose);
1446 data.start = parent;
1447 data.found = 0;
1448
1449 d_walk(parent, &data, select_collect, NULL);
1450 if (!data.found)
1451 break;
1452
1453 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001454 cond_resched();
1455 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001457EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
Al Viro9c8c10e2014-05-02 20:36:10 -04001459static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001460{
Al Viro9c8c10e2014-05-02 20:36:10 -04001461 /* it has busy descendents; complain about those instead */
1462 if (!list_empty(&dentry->d_subdirs))
1463 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001464
Al Viro9c8c10e2014-05-02 20:36:10 -04001465 /* root with refcount 1 is fine */
1466 if (dentry == _data && dentry->d_lockref.count == 1)
1467 return D_WALK_CONTINUE;
1468
1469 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1470 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001471 dentry,
1472 dentry->d_inode ?
1473 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001474 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001475 dentry->d_lockref.count,
1476 dentry->d_sb->s_type->name,
1477 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001478 WARN_ON(1);
1479 return D_WALK_CONTINUE;
1480}
1481
1482static void do_one_tree(struct dentry *dentry)
1483{
1484 shrink_dcache_parent(dentry);
1485 d_walk(dentry, dentry, umount_check, NULL);
1486 d_drop(dentry);
1487 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001488}
1489
1490/*
1491 * destroy the dentries attached to a superblock on unmounting
1492 */
1493void shrink_dcache_for_umount(struct super_block *sb)
1494{
1495 struct dentry *dentry;
1496
Al Viro9c8c10e2014-05-02 20:36:10 -04001497 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001498
1499 dentry = sb->s_root;
1500 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001501 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001502
1503 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001504 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1505 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001506 }
1507}
1508
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001509struct detach_data {
1510 struct select_data select;
1511 struct dentry *mountpoint;
1512};
1513static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001514{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001515 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001516
1517 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001518 __dget_dlock(dentry);
1519 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001520 return D_WALK_QUIT;
1521 }
1522
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001523 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001524}
1525
1526static void check_and_drop(void *_data)
1527{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001528 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001529
Al Viro81be24d2017-06-03 07:20:09 +01001530 if (!data->mountpoint && list_empty(&data->select.dispose))
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001531 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001532}
1533
1534/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001535 * d_invalidate - detach submounts, prune dcache, and drop
1536 * @dentry: dentry to invalidate (aka detach, prune and drop)
1537 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001538 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001539 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001540 * The final d_drop is done as an atomic operation relative to
1541 * rename_lock ensuring there are no races with d_set_mounted. This
1542 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001543 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001544void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001545{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001546 /*
1547 * If it's already been dropped, return OK.
1548 */
1549 spin_lock(&dentry->d_lock);
1550 if (d_unhashed(dentry)) {
1551 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001552 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001553 }
1554 spin_unlock(&dentry->d_lock);
1555
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001556 /* Negative dentries can be dropped without further checks */
1557 if (!dentry->d_inode) {
1558 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001559 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001560 }
1561
1562 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001563 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001564
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001565 data.mountpoint = NULL;
1566 INIT_LIST_HEAD(&data.select.dispose);
1567 data.select.start = dentry;
1568 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001569
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001570 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001571
Al Viro81be24d2017-06-03 07:20:09 +01001572 if (!list_empty(&data.select.dispose))
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001573 shrink_dentry_list(&data.select.dispose);
Al Viro81be24d2017-06-03 07:20:09 +01001574 else if (!data.mountpoint)
1575 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001576
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001577 if (data.mountpoint) {
1578 detach_mounts(data.mountpoint);
1579 dput(data.mountpoint);
1580 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001581 cond_resched();
1582 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001583}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001584EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001585
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586/**
Al Viroa4464db2011-07-07 15:03:58 -04001587 * __d_alloc - allocate a dcache entry
1588 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 * @name: qstr of the name
1590 *
1591 * Allocates a dentry. It returns %NULL if there is insufficient memory
1592 * available. On a success the dentry is returned. The name passed in is
1593 * copied and the copy passed in may be reused after this call.
1594 */
1595
Al Viroa4464db2011-07-07 15:03:58 -04001596struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597{
1598 struct dentry *dentry;
1599 char *dname;
Miklos Szeredi285b1022016-06-28 11:47:32 +02001600 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
Mel Gormane12ba742007-10-16 01:25:52 -07001602 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 if (!dentry)
1604 return NULL;
1605
Linus Torvalds6326c712012-05-21 16:14:04 -07001606 /*
1607 * We guarantee that the inline name is always NUL-terminated.
1608 * This way the memcpy() done by the name switching in rename
1609 * will still always have a NUL at the end, even if we might
1610 * be overwriting an internal NUL character
1611 */
1612 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Al Viro798434b2016-03-24 20:38:43 -04001613 if (unlikely(!name)) {
David Howellscdf01222017-07-04 17:25:22 +01001614 name = &slash_name;
Al Viro798434b2016-03-24 20:38:43 -04001615 dname = dentry->d_iname;
1616 } else if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001617 size_t size = offsetof(struct external_name, name[1]);
Vladimir Davydov5d097052016-01-14 15:18:21 -08001618 struct external_name *p = kmalloc(size + name->len,
1619 GFP_KERNEL_ACCOUNT);
Al Viro8d85b482014-09-29 14:54:27 -04001620 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 kmem_cache_free(dentry_cache, dentry);
1622 return NULL;
1623 }
Al Viro8d85b482014-09-29 14:54:27 -04001624 atomic_set(&p->u.count, 1);
1625 dname = p->name;
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -08001626 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1627 kasan_unpoison_shadow(dname,
1628 round_up(name->len + 1, sizeof(unsigned long)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 } else {
1630 dname = dentry->d_iname;
1631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
1633 dentry->d_name.len = name->len;
1634 dentry->d_name.hash = name->hash;
1635 memcpy(dname, name->name, name->len);
1636 dname[name->len] = 0;
1637
Linus Torvalds6326c712012-05-21 16:14:04 -07001638 /* Make sure we always see the terminating NUL character */
1639 smp_wmb();
1640 dentry->d_name.name = dname;
1641
Waiman Long98474232013-08-28 18:24:59 -07001642 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001643 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001645 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001647 dentry->d_parent = dentry;
1648 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 dentry->d_op = NULL;
1650 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001651 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 INIT_LIST_HEAD(&dentry->d_lru);
1653 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001654 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1655 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001656 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657
Miklos Szeredi285b1022016-06-28 11:47:32 +02001658 if (dentry->d_op && dentry->d_op->d_init) {
1659 err = dentry->d_op->d_init(dentry);
1660 if (err) {
1661 if (dname_external(dentry))
1662 kfree(external_name(dentry));
1663 kmem_cache_free(dentry_cache, dentry);
1664 return NULL;
1665 }
1666 }
1667
Nick Piggin3e880fb2011-01-07 17:49:19 +11001668 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001669
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 return dentry;
1671}
Al Viroa4464db2011-07-07 15:03:58 -04001672
1673/**
1674 * d_alloc - allocate a dcache entry
1675 * @parent: parent of entry to allocate
1676 * @name: qstr of the name
1677 *
1678 * Allocates a dentry. It returns %NULL if there is insufficient memory
1679 * available. On a success the dentry is returned. The name passed in is
1680 * copied and the copy passed in may be reused after this call.
1681 */
1682struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1683{
1684 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1685 if (!dentry)
1686 return NULL;
Al Viro3d56c252016-06-07 21:26:55 -04001687 dentry->d_flags |= DCACHE_RCUACCESS;
Al Viroa4464db2011-07-07 15:03:58 -04001688 spin_lock(&parent->d_lock);
1689 /*
1690 * don't need child lock because it is not subject
1691 * to concurrency here
1692 */
1693 __dget_dlock(parent);
1694 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001695 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001696 spin_unlock(&parent->d_lock);
1697
1698 return dentry;
1699}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001700EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
Al Viroba65dc52016-06-10 11:32:47 -04001702struct dentry *d_alloc_cursor(struct dentry * parent)
1703{
1704 struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
1705 if (dentry) {
1706 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1707 dentry->d_parent = dget(parent);
1708 }
1709 return dentry;
1710}
1711
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001712/**
1713 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1714 * @sb: the superblock
1715 * @name: qstr of the name
1716 *
1717 * For a filesystem that just pins its dentries in memory and never
1718 * performs lookups at all, return an unhashed IS_ROOT dentry.
1719 */
Nick Piggin4b936882011-01-07 17:50:07 +11001720struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1721{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001722 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001723}
1724EXPORT_SYMBOL(d_alloc_pseudo);
1725
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1727{
1728 struct qstr q;
1729
1730 q.name = name;
Linus Torvalds8387ff22016-06-10 07:51:30 -07001731 q.hash_len = hashlen_string(parent, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 return d_alloc(parent, &q);
1733}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001734EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735
Nick Pigginfb045ad2011-01-07 17:49:55 +11001736void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1737{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001738 WARN_ON_ONCE(dentry->d_op);
1739 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001740 DCACHE_OP_COMPARE |
1741 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001742 DCACHE_OP_WEAK_REVALIDATE |
David Howells4bacc9c2015-06-18 14:32:31 +01001743 DCACHE_OP_DELETE |
Miklos Szeredid101a122016-03-26 16:14:37 -04001744 DCACHE_OP_REAL));
Nick Pigginfb045ad2011-01-07 17:49:55 +11001745 dentry->d_op = op;
1746 if (!op)
1747 return;
1748 if (op->d_hash)
1749 dentry->d_flags |= DCACHE_OP_HASH;
1750 if (op->d_compare)
1751 dentry->d_flags |= DCACHE_OP_COMPARE;
1752 if (op->d_revalidate)
1753 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001754 if (op->d_weak_revalidate)
1755 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001756 if (op->d_delete)
1757 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001758 if (op->d_prune)
1759 dentry->d_flags |= DCACHE_OP_PRUNE;
Miklos Szeredid101a122016-03-26 16:14:37 -04001760 if (op->d_real)
1761 dentry->d_flags |= DCACHE_OP_REAL;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001762
1763}
1764EXPORT_SYMBOL(d_set_d_op);
1765
David Howellsdf1a0852015-01-29 12:02:28 +00001766
1767/*
1768 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1769 * @dentry - The dentry to mark
1770 *
1771 * Mark a dentry as falling through to the lower layer (as set with
1772 * d_pin_lower()). This flag may be recorded on the medium.
1773 */
1774void d_set_fallthru(struct dentry *dentry)
1775{
1776 spin_lock(&dentry->d_lock);
1777 dentry->d_flags |= DCACHE_FALLTHRU;
1778 spin_unlock(&dentry->d_lock);
1779}
1780EXPORT_SYMBOL(d_set_fallthru);
1781
David Howellsb18825a2013-09-12 19:22:53 +01001782static unsigned d_flags_for_inode(struct inode *inode)
1783{
David Howells44bdb5e2015-01-29 12:02:29 +00001784 unsigned add_flags = DCACHE_REGULAR_TYPE;
David Howellsb18825a2013-09-12 19:22:53 +01001785
1786 if (!inode)
1787 return DCACHE_MISS_TYPE;
1788
1789 if (S_ISDIR(inode->i_mode)) {
1790 add_flags = DCACHE_DIRECTORY_TYPE;
1791 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1792 if (unlikely(!inode->i_op->lookup))
1793 add_flags = DCACHE_AUTODIR_TYPE;
1794 else
1795 inode->i_opflags |= IOP_LOOKUP;
1796 }
David Howells44bdb5e2015-01-29 12:02:29 +00001797 goto type_determined;
David Howellsb18825a2013-09-12 19:22:53 +01001798 }
1799
David Howells44bdb5e2015-01-29 12:02:29 +00001800 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
Al Viro6b255392015-11-17 10:20:54 -05001801 if (unlikely(inode->i_op->get_link)) {
David Howells44bdb5e2015-01-29 12:02:29 +00001802 add_flags = DCACHE_SYMLINK_TYPE;
1803 goto type_determined;
1804 }
1805 inode->i_opflags |= IOP_NOFOLLOW;
1806 }
1807
1808 if (unlikely(!S_ISREG(inode->i_mode)))
1809 add_flags = DCACHE_SPECIAL_TYPE;
1810
1811type_determined:
David Howellsb18825a2013-09-12 19:22:53 +01001812 if (unlikely(IS_AUTOMOUNT(inode)))
1813 add_flags |= DCACHE_NEED_AUTOMOUNT;
1814 return add_flags;
1815}
1816
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001817static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1818{
David Howellsb18825a2013-09-12 19:22:53 +01001819 unsigned add_flags = d_flags_for_inode(inode);
Al Viro85c7f812016-04-14 19:52:13 -04001820 WARN_ON(d_in_lookup(dentry));
David Howellsb18825a2013-09-12 19:22:53 +01001821
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001822 spin_lock(&dentry->d_lock);
Al Virode689f52016-03-09 18:05:42 -05001823 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
Al Viroa528aca2016-02-29 12:12:46 -05001824 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +00001825 __d_set_inode_and_type(dentry, inode, add_flags);
Al Viroa528aca2016-02-29 12:12:46 -05001826 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04001827 fsnotify_update_flags(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001828 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001829}
1830
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831/**
1832 * d_instantiate - fill in inode information for a dentry
1833 * @entry: dentry to complete
1834 * @inode: inode to attach to this dentry
1835 *
1836 * Fill in inode information in the entry.
1837 *
1838 * This turns negative dentries into productive full members
1839 * of society.
1840 *
1841 * NOTE! This assumes that the inode count has been incremented
1842 * (or otherwise set) by the caller to indicate that it is now
1843 * in use by the dcache.
1844 */
1845
1846void d_instantiate(struct dentry *entry, struct inode * inode)
1847{
Al Viro946e51f2014-10-26 19:19:16 -04001848 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Al Virode689f52016-03-09 18:05:42 -05001849 if (inode) {
Al Virob9680912016-04-11 00:53:26 -04001850 security_d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001851 spin_lock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001852 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001853 spin_unlock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001856EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857
1858/**
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001859 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1860 * @entry: dentry to complete
1861 * @inode: inode to attach to this dentry
1862 *
1863 * Fill in inode information in the entry. If a directory alias is found, then
1864 * return an error (and drop inode). Together with d_materialise_unique() this
1865 * guarantees that a directory inode may never have more than one alias.
1866 */
1867int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1868{
Al Viro946e51f2014-10-26 19:19:16 -04001869 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001870
Al Virob9680912016-04-11 00:53:26 -04001871 security_d_instantiate(entry, inode);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001872 spin_lock(&inode->i_lock);
1873 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1874 spin_unlock(&inode->i_lock);
1875 iput(inode);
1876 return -EBUSY;
1877 }
1878 __d_instantiate(entry, inode);
1879 spin_unlock(&inode->i_lock);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001880
1881 return 0;
1882}
1883EXPORT_SYMBOL(d_instantiate_no_diralias);
1884
Al Viroadc0e912012-01-08 16:49:21 -05001885struct dentry *d_make_root(struct inode *root_inode)
1886{
1887 struct dentry *res = NULL;
1888
1889 if (root_inode) {
Al Viro798434b2016-03-24 20:38:43 -04001890 res = __d_alloc(root_inode->i_sb, NULL);
Al Viroadc0e912012-01-08 16:49:21 -05001891 if (res)
1892 d_instantiate(res, root_inode);
1893 else
1894 iput(root_inode);
1895 }
1896 return res;
1897}
1898EXPORT_SYMBOL(d_make_root);
1899
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001900static struct dentry * __d_find_any_alias(struct inode *inode)
1901{
1902 struct dentry *alias;
1903
Al Virob3d9b7a2012-06-09 13:51:19 -04001904 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001905 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001906 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001907 __dget(alias);
1908 return alias;
1909}
1910
Sage Weil46f72b32012-01-10 09:04:37 -08001911/**
1912 * d_find_any_alias - find any alias for a given inode
1913 * @inode: inode to find an alias for
1914 *
1915 * If any aliases exist for the given inode, take and return a
1916 * reference for one of them. If no aliases exist, return %NULL.
1917 */
1918struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001919{
1920 struct dentry *de;
1921
1922 spin_lock(&inode->i_lock);
1923 de = __d_find_any_alias(inode);
1924 spin_unlock(&inode->i_lock);
1925 return de;
1926}
Sage Weil46f72b32012-01-10 09:04:37 -08001927EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001928
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001929static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001930{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001931 struct dentry *tmp;
1932 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001933 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001934
1935 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001936 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001937 if (IS_ERR(inode))
1938 return ERR_CAST(inode);
1939
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001940 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001941 if (res)
1942 goto out_iput;
1943
Al Viro798434b2016-03-24 20:38:43 -04001944 tmp = __d_alloc(inode->i_sb, NULL);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001945 if (!tmp) {
1946 res = ERR_PTR(-ENOMEM);
1947 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001948 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001949
Al Virob9680912016-04-11 00:53:26 -04001950 security_d_instantiate(tmp, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001951 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001952 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001953 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001954 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001955 dput(tmp);
1956 goto out_iput;
1957 }
1958
1959 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001960 add_flags = d_flags_for_inode(inode);
1961
1962 if (disconnected)
1963 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001964
Christoph Hellwig9308a612008-08-11 15:49:12 +02001965 spin_lock(&tmp->d_lock);
David Howells4bf46a22015-03-05 14:09:22 +00001966 __d_set_inode_and_type(tmp, inode, add_flags);
Al Viro946e51f2014-10-26 19:19:16 -04001967 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001968 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001969 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001970 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001971 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001972 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001973
Christoph Hellwig9308a612008-08-11 15:49:12 +02001974 return tmp;
1975
1976 out_iput:
1977 iput(inode);
1978 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001979}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001980
1981/**
1982 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1983 * @inode: inode to allocate the dentry for
1984 *
1985 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1986 * similar open by handle operations. The returned dentry may be anonymous,
1987 * or may have a full name (if the inode was already in the cache).
1988 *
1989 * When called on a directory inode, we must ensure that the inode only ever
1990 * has one dentry. If a dentry is found, that is returned instead of
1991 * allocating a new one.
1992 *
1993 * On successful return, the reference to the inode has been transferred
1994 * to the dentry. In case of an error the reference on the inode is released.
1995 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1996 * be passed in and the error will be propagated to the return value,
1997 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1998 */
1999struct dentry *d_obtain_alias(struct inode *inode)
2000{
2001 return __d_obtain_alias(inode, 1);
2002}
Benny Halevyadc48722009-02-27 14:02:59 -08002003EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004
2005/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002006 * d_obtain_root - find or allocate a dentry for a given inode
2007 * @inode: inode to allocate the dentry for
2008 *
2009 * Obtain an IS_ROOT dentry for the root of a filesystem.
2010 *
2011 * We must ensure that directory inodes only ever have one dentry. If a
2012 * dentry is found, that is returned instead of allocating a new one.
2013 *
2014 * On successful return, the reference to the inode has been transferred
2015 * to the dentry. In case of an error the reference on the inode is
2016 * released. A %NULL or IS_ERR inode may be passed in and will be the
2017 * error will be propagate to the return value, with a %NULL @inode
2018 * replaced by ERR_PTR(-ESTALE).
2019 */
2020struct dentry *d_obtain_root(struct inode *inode)
2021{
2022 return __d_obtain_alias(inode, 0);
2023}
2024EXPORT_SYMBOL(d_obtain_root);
2025
2026/**
Barry Naujok94035402008-05-21 16:50:46 +10002027 * d_add_ci - lookup or allocate new dentry with case-exact name
2028 * @inode: the inode case-insensitive lookup has found
2029 * @dentry: the negative dentry that was passed to the parent's lookup func
2030 * @name: the case-exact name to be associated with the returned dentry
2031 *
2032 * This is to avoid filling the dcache with case-insensitive names to the
2033 * same inode, only the actual correct case is stored in the dcache for
2034 * case-insensitive filesystems.
2035 *
2036 * For a case-insensitive lookup match and if the the case-exact dentry
2037 * already exists in in the dcache, use it and return it.
2038 *
2039 * If no entry exists with the exact case name, allocate new dentry with
2040 * the exact case, and return the spliced entry.
2041 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02002042struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10002043 struct qstr *name)
2044{
Al Virod9171b92016-04-15 03:33:13 -04002045 struct dentry *found, *res;
Barry Naujok94035402008-05-21 16:50:46 +10002046
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002047 /*
2048 * First check if a dentry matching the name already exists,
2049 * if not go ahead and create it now.
2050 */
Barry Naujok94035402008-05-21 16:50:46 +10002051 found = d_hash_and_lookup(dentry->d_parent, name);
Al Virod9171b92016-04-15 03:33:13 -04002052 if (found) {
2053 iput(inode);
2054 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002055 }
Al Virod9171b92016-04-15 03:33:13 -04002056 if (d_in_lookup(dentry)) {
2057 found = d_alloc_parallel(dentry->d_parent, name,
2058 dentry->d_wait);
2059 if (IS_ERR(found) || !d_in_lookup(found)) {
2060 iput(inode);
2061 return found;
2062 }
2063 } else {
2064 found = d_alloc(dentry->d_parent, name);
2065 if (!found) {
2066 iput(inode);
2067 return ERR_PTR(-ENOMEM);
2068 }
2069 }
2070 res = d_splice_alias(inode, found);
2071 if (res) {
2072 dput(found);
2073 return res;
2074 }
Al Viro4f522a22013-02-11 23:20:37 -05002075 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002076}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002077EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002079
Al Virod4c91a82016-06-25 23:33:49 -04002080static inline bool d_same_name(const struct dentry *dentry,
2081 const struct dentry *parent,
2082 const struct qstr *name)
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002083{
Al Virod4c91a82016-06-25 23:33:49 -04002084 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2085 if (dentry->d_name.len != name->len)
2086 return false;
2087 return dentry_cmp(dentry, name->name, name->len) == 0;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002088 }
Al Viro6fa67e72016-07-31 16:37:25 -04002089 return parent->d_op->d_compare(dentry,
Al Virod4c91a82016-06-25 23:33:49 -04002090 dentry->d_name.len, dentry->d_name.name,
2091 name) == 0;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002092}
2093
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002095 * __d_lookup_rcu - search for a dentry (racy, store-free)
2096 * @parent: parent dentry
2097 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002098 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002099 * Returns: dentry, or NULL
2100 *
2101 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2102 * resolution (store-free path walking) design described in
2103 * Documentation/filesystems/path-lookup.txt.
2104 *
2105 * This is not to be used outside core vfs.
2106 *
2107 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2108 * held, and rcu_read_lock held. The returned dentry must not be stored into
2109 * without taking d_lock and checking d_seq sequence count against @seq
2110 * returned here.
2111 *
Linus Torvalds15570082013-09-02 11:38:06 -07002112 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002113 * function.
2114 *
2115 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2116 * the returned dentry, so long as its parent's seqlock is checked after the
2117 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2118 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002119 *
2120 * NOTE! The caller *has* to check the resulting dentry against the sequence
2121 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002122 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002123struct dentry *__d_lookup_rcu(const struct dentry *parent,
2124 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002125 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002126{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002127 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002128 const unsigned char *str = name->name;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002129 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002130 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002131 struct dentry *dentry;
2132
2133 /*
2134 * Note: There is significant duplication with __d_lookup_rcu which is
2135 * required to prevent single threaded performance regressions
2136 * especially on architectures where smp_rmb (in seqcounts) are costly.
2137 * Keep the two functions in sync.
2138 */
2139
2140 /*
2141 * The hash list is protected using RCU.
2142 *
2143 * Carefully use d_seq when comparing a candidate dentry, to avoid
2144 * races with d_move().
2145 *
2146 * It is possible that concurrent renames can mess up our list
2147 * walk here and result in missing our dentry, resulting in the
2148 * false-negative result. d_lookup() protects against concurrent
2149 * renames using rename_lock seqlock.
2150 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002151 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002152 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002153 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002154 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002155
Nick Piggin31e6b012011-01-07 17:49:52 +11002156seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002157 /*
2158 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002159 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002160 *
2161 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002162 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002163 *
2164 * NOTE! We do a "raw" seqcount_begin here. That means that
2165 * we don't wait for the sequence count to stabilize if it
2166 * is in the middle of a sequence change. If we do the slow
2167 * dentry compare, we will do seqretries until it is stable,
2168 * and if we end up with a successful lookup, we actually
2169 * want to exit RCU lookup anyway.
Al Virod4c91a82016-06-25 23:33:49 -04002170 *
2171 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2172 * we are still guaranteed NUL-termination of ->d_name.name.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002173 */
2174 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002175 if (dentry->d_parent != parent)
2176 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002177 if (d_unhashed(dentry))
2178 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002179
2180 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Al Virod4c91a82016-06-25 23:33:49 -04002181 int tlen;
2182 const char *tname;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002183 if (dentry->d_name.hash != hashlen_hash(hashlen))
2184 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002185 tlen = dentry->d_name.len;
2186 tname = dentry->d_name.name;
2187 /* we want a consistent (name,len) pair */
2188 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2189 cpu_relax();
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002190 goto seqretry;
2191 }
Al Viro6fa67e72016-07-31 16:37:25 -04002192 if (parent->d_op->d_compare(dentry,
Al Virod4c91a82016-06-25 23:33:49 -04002193 tlen, tname, name) != 0)
2194 continue;
2195 } else {
2196 if (dentry->d_name.hash_len != hashlen)
2197 continue;
2198 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2199 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002200 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002201 *seqp = seq;
Al Virod4c91a82016-06-25 23:33:49 -04002202 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002203 }
2204 return NULL;
2205}
2206
2207/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 * d_lookup - search for a dentry
2209 * @parent: parent dentry
2210 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002211 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002213 * d_lookup searches the children of the parent dentry for the name in
2214 * question. If the dentry is found its reference count is incremented and the
2215 * dentry is returned. The caller must use dput to free the entry when it has
2216 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 */
Al Viroda2d8452013-01-24 18:29:34 -05002218struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219{
Nick Piggin31e6b012011-01-07 17:49:52 +11002220 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002221 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222
Daeseok Younb8314f92014-08-11 11:46:53 +09002223 do {
2224 seq = read_seqbegin(&rename_lock);
2225 dentry = __d_lookup(parent, name);
2226 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 break;
2228 } while (read_seqretry(&rename_lock, seq));
2229 return dentry;
2230}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002231EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232
Nick Piggin31e6b012011-01-07 17:49:52 +11002233/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002234 * __d_lookup - search for a dentry (racy)
2235 * @parent: parent dentry
2236 * @name: qstr of name we wish to find
2237 * Returns: dentry, or NULL
2238 *
2239 * __d_lookup is like d_lookup, however it may (rarely) return a
2240 * false-negative result due to unrelated rename activity.
2241 *
2242 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2243 * however it must be used carefully, eg. with a following d_lookup in
2244 * the case of failure.
2245 *
2246 * __d_lookup callers must be commented.
2247 */
Al Viroa713ca22013-01-24 18:27:00 -05002248struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 unsigned int hash = name->hash;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002251 struct hlist_bl_head *b = d_hash(hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002252 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002253 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002254 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255
Nick Pigginb04f7842010-08-18 04:37:34 +10002256 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002257 * Note: There is significant duplication with __d_lookup_rcu which is
2258 * required to prevent single threaded performance regressions
2259 * especially on architectures where smp_rmb (in seqcounts) are costly.
2260 * Keep the two functions in sync.
2261 */
2262
2263 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002264 * The hash list is protected using RCU.
2265 *
2266 * Take d_lock when comparing a candidate dentry, to avoid races
2267 * with d_move().
2268 *
2269 * It is possible that concurrent renames can mess up our list
2270 * walk here and result in missing our dentry, resulting in the
2271 * false-negative result. d_lookup() protects against concurrent
2272 * renames using rename_lock seqlock.
2273 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002274 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002275 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 rcu_read_lock();
2277
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002278 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 if (dentry->d_name.hash != hash)
2281 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282
2283 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 if (dentry->d_parent != parent)
2285 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002286 if (d_unhashed(dentry))
2287 goto next;
2288
Al Virod4c91a82016-06-25 23:33:49 -04002289 if (!d_same_name(dentry, parent, name))
2290 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
Waiman Long98474232013-08-28 18:24:59 -07002292 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002293 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 spin_unlock(&dentry->d_lock);
2295 break;
2296next:
2297 spin_unlock(&dentry->d_lock);
2298 }
2299 rcu_read_unlock();
2300
2301 return found;
2302}
2303
2304/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002305 * d_hash_and_lookup - hash the qstr then search for a dentry
2306 * @dir: Directory to search in
2307 * @name: qstr of name we wish to find
2308 *
Al Viro4f522a22013-02-11 23:20:37 -05002309 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002310 */
2311struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2312{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002313 /*
2314 * Check for a fs-specific hash function. Note that we must
2315 * calculate the standard hash first, as the d_op->d_hash()
2316 * routine may choose to leave the hash value unchanged.
2317 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002318 name->hash = full_name_hash(dir, name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002319 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002320 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002321 if (unlikely(err < 0))
2322 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002323 }
Al Viro4f522a22013-02-11 23:20:37 -05002324 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002325}
Al Viro4f522a22013-02-11 23:20:37 -05002326EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002327
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328/*
2329 * When a file is deleted, we have two options:
2330 * - turn this dentry into a negative dentry
2331 * - unhash this dentry and free it.
2332 *
2333 * Usually, we want to just turn this into
2334 * a negative dentry, but if anybody else is
2335 * currently using the dentry or the inode
2336 * we can't do that and we fall back on removing
2337 * it from the hash queues and waiting for
2338 * it to be deleted later when it has no users
2339 */
2340
2341/**
2342 * d_delete - delete a dentry
2343 * @dentry: The dentry to delete
2344 *
2345 * Turn the dentry into a negative dentry if possible, otherwise
2346 * remove it from the hash queues so it can be deleted later
2347 */
2348
2349void d_delete(struct dentry * dentry)
2350{
Nick Piggin873feea2011-01-07 17:50:06 +11002351 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002352 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 /*
2354 * Are we the only user?
2355 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002356again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002358 inode = dentry->d_inode;
2359 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002360 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002361 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002362 spin_unlock(&dentry->d_lock);
2363 cpu_relax();
2364 goto again;
2365 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002366 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002367 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002368 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 return;
2370 }
2371
2372 if (!d_unhashed(dentry))
2373 __d_drop(dentry);
2374
2375 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002376
2377 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002379EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380
Al Viro15d3c582016-07-29 17:45:21 -04002381static void __d_rehash(struct dentry *entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382{
Al Viro15d3c582016-07-29 17:45:21 -04002383 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002384 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002385 hlist_bl_lock(b);
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002386 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002387 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388}
2389
2390/**
2391 * d_rehash - add an entry back to the hash
2392 * @entry: dentry to add to the hash
2393 *
2394 * Adds a dentry to the hash according to its name.
2395 */
2396
2397void d_rehash(struct dentry * entry)
2398{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 spin_lock(&entry->d_lock);
Al Viro15d3c582016-07-29 17:45:21 -04002400 __d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002403EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404
Al Viro84e710d2016-04-15 00:58:55 -04002405static inline unsigned start_dir_add(struct inode *dir)
2406{
2407
2408 for (;;) {
2409 unsigned n = dir->i_dir_seq;
2410 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2411 return n;
2412 cpu_relax();
2413 }
2414}
2415
2416static inline void end_dir_add(struct inode *dir, unsigned n)
2417{
2418 smp_store_release(&dir->i_dir_seq, n + 2);
2419}
2420
Al Virod9171b92016-04-15 03:33:13 -04002421static void d_wait_lookup(struct dentry *dentry)
2422{
2423 if (d_in_lookup(dentry)) {
2424 DECLARE_WAITQUEUE(wait, current);
2425 add_wait_queue(dentry->d_wait, &wait);
2426 do {
2427 set_current_state(TASK_UNINTERRUPTIBLE);
2428 spin_unlock(&dentry->d_lock);
2429 schedule();
2430 spin_lock(&dentry->d_lock);
2431 } while (d_in_lookup(dentry));
2432 }
2433}
2434
Al Viro94bdd652016-04-15 02:42:04 -04002435struct dentry *d_alloc_parallel(struct dentry *parent,
Al Virod9171b92016-04-15 03:33:13 -04002436 const struct qstr *name,
2437 wait_queue_head_t *wq)
Al Viro94bdd652016-04-15 02:42:04 -04002438{
Al Viro94bdd652016-04-15 02:42:04 -04002439 unsigned int hash = name->hash;
Al Viro94bdd652016-04-15 02:42:04 -04002440 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2441 struct hlist_bl_node *node;
2442 struct dentry *new = d_alloc(parent, name);
2443 struct dentry *dentry;
2444 unsigned seq, r_seq, d_seq;
2445
2446 if (unlikely(!new))
2447 return ERR_PTR(-ENOMEM);
2448
2449retry:
2450 rcu_read_lock();
2451 seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1;
2452 r_seq = read_seqbegin(&rename_lock);
2453 dentry = __d_lookup_rcu(parent, name, &d_seq);
2454 if (unlikely(dentry)) {
2455 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2456 rcu_read_unlock();
2457 goto retry;
2458 }
2459 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2460 rcu_read_unlock();
2461 dput(dentry);
2462 goto retry;
2463 }
2464 rcu_read_unlock();
2465 dput(new);
2466 return dentry;
2467 }
2468 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2469 rcu_read_unlock();
2470 goto retry;
2471 }
2472 hlist_bl_lock(b);
2473 if (unlikely(parent->d_inode->i_dir_seq != seq)) {
2474 hlist_bl_unlock(b);
2475 rcu_read_unlock();
2476 goto retry;
2477 }
Al Viro94bdd652016-04-15 02:42:04 -04002478 /*
2479 * No changes for the parent since the beginning of d_lookup().
2480 * Since all removals from the chain happen with hlist_bl_lock(),
2481 * any potential in-lookup matches are going to stay here until
2482 * we unlock the chain. All fields are stable in everything
2483 * we encounter.
2484 */
2485 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2486 if (dentry->d_name.hash != hash)
2487 continue;
2488 if (dentry->d_parent != parent)
2489 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002490 if (!d_same_name(dentry, parent, name))
2491 continue;
Al Viro94bdd652016-04-15 02:42:04 -04002492 hlist_bl_unlock(b);
Al Viroe7d6ef92016-06-20 01:35:59 -04002493 /* now we can try to grab a reference */
2494 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2495 rcu_read_unlock();
2496 goto retry;
2497 }
2498
2499 rcu_read_unlock();
2500 /*
2501 * somebody is likely to be still doing lookup for it;
2502 * wait for them to finish
2503 */
Al Virod9171b92016-04-15 03:33:13 -04002504 spin_lock(&dentry->d_lock);
2505 d_wait_lookup(dentry);
2506 /*
2507 * it's not in-lookup anymore; in principle we should repeat
2508 * everything from dcache lookup, but it's likely to be what
2509 * d_lookup() would've found anyway. If it is, just return it;
2510 * otherwise we really have to repeat the whole thing.
2511 */
2512 if (unlikely(dentry->d_name.hash != hash))
2513 goto mismatch;
2514 if (unlikely(dentry->d_parent != parent))
2515 goto mismatch;
2516 if (unlikely(d_unhashed(dentry)))
2517 goto mismatch;
Al Virod4c91a82016-06-25 23:33:49 -04002518 if (unlikely(!d_same_name(dentry, parent, name)))
2519 goto mismatch;
Al Virod9171b92016-04-15 03:33:13 -04002520 /* OK, it *is* a hashed match; return it */
2521 spin_unlock(&dentry->d_lock);
Al Viro94bdd652016-04-15 02:42:04 -04002522 dput(new);
2523 return dentry;
2524 }
Al Viroe7d6ef92016-06-20 01:35:59 -04002525 rcu_read_unlock();
Al Viro94bdd652016-04-15 02:42:04 -04002526 /* we can't take ->d_lock here; it's OK, though. */
2527 new->d_flags |= DCACHE_PAR_LOOKUP;
Al Virod9171b92016-04-15 03:33:13 -04002528 new->d_wait = wq;
Al Viro94bdd652016-04-15 02:42:04 -04002529 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2530 hlist_bl_unlock(b);
2531 return new;
Al Virod9171b92016-04-15 03:33:13 -04002532mismatch:
2533 spin_unlock(&dentry->d_lock);
2534 dput(dentry);
2535 goto retry;
Al Viro94bdd652016-04-15 02:42:04 -04002536}
2537EXPORT_SYMBOL(d_alloc_parallel);
2538
Al Viro85c7f812016-04-14 19:52:13 -04002539void __d_lookup_done(struct dentry *dentry)
2540{
Al Viro94bdd652016-04-15 02:42:04 -04002541 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2542 dentry->d_name.hash);
2543 hlist_bl_lock(b);
Al Viro85c7f812016-04-14 19:52:13 -04002544 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
Al Viro94bdd652016-04-15 02:42:04 -04002545 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
Al Virod9171b92016-04-15 03:33:13 -04002546 wake_up_all(dentry->d_wait);
2547 dentry->d_wait = NULL;
Al Viro94bdd652016-04-15 02:42:04 -04002548 hlist_bl_unlock(b);
2549 INIT_HLIST_NODE(&dentry->d_u.d_alias);
Al Virod9171b92016-04-15 03:33:13 -04002550 INIT_LIST_HEAD(&dentry->d_lru);
Al Viro85c7f812016-04-14 19:52:13 -04002551}
2552EXPORT_SYMBOL(__d_lookup_done);
Al Viroed782b52016-03-09 19:52:39 -05002553
2554/* inode->i_lock held if inode is non-NULL */
2555
2556static inline void __d_add(struct dentry *dentry, struct inode *inode)
2557{
Al Viro84e710d2016-04-15 00:58:55 -04002558 struct inode *dir = NULL;
2559 unsigned n;
Al Viro0568d702016-04-14 19:40:56 -04002560 spin_lock(&dentry->d_lock);
Al Viro84e710d2016-04-15 00:58:55 -04002561 if (unlikely(d_in_lookup(dentry))) {
2562 dir = dentry->d_parent->d_inode;
2563 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002564 __d_lookup_done(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002565 }
Al Viroed782b52016-03-09 19:52:39 -05002566 if (inode) {
Al Viro0568d702016-04-14 19:40:56 -04002567 unsigned add_flags = d_flags_for_inode(inode);
2568 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2569 raw_write_seqcount_begin(&dentry->d_seq);
2570 __d_set_inode_and_type(dentry, inode, add_flags);
2571 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04002572 fsnotify_update_flags(dentry);
Al Viroed782b52016-03-09 19:52:39 -05002573 }
Al Viro15d3c582016-07-29 17:45:21 -04002574 __d_rehash(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002575 if (dir)
2576 end_dir_add(dir, n);
Al Viro0568d702016-04-14 19:40:56 -04002577 spin_unlock(&dentry->d_lock);
2578 if (inode)
2579 spin_unlock(&inode->i_lock);
Al Viroed782b52016-03-09 19:52:39 -05002580}
2581
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002582/**
Al Viro34d0d192016-03-08 21:01:03 -05002583 * d_add - add dentry to hash queues
2584 * @entry: dentry to add
2585 * @inode: The inode to attach to this dentry
2586 *
2587 * This adds the entry to the hash queues and initializes @inode.
2588 * The entry was actually filled in earlier during d_alloc().
2589 */
2590
2591void d_add(struct dentry *entry, struct inode *inode)
2592{
Al Virob9680912016-04-11 00:53:26 -04002593 if (inode) {
2594 security_d_instantiate(entry, inode);
Al Viroed782b52016-03-09 19:52:39 -05002595 spin_lock(&inode->i_lock);
Al Virob9680912016-04-11 00:53:26 -04002596 }
Al Viroed782b52016-03-09 19:52:39 -05002597 __d_add(entry, inode);
Al Viro34d0d192016-03-08 21:01:03 -05002598}
2599EXPORT_SYMBOL(d_add);
2600
2601/**
Al Viro668d0cd2016-03-08 12:44:17 -05002602 * d_exact_alias - find and hash an exact unhashed alias
2603 * @entry: dentry to add
2604 * @inode: The inode to go with this dentry
2605 *
2606 * If an unhashed dentry with the same name/parent and desired
2607 * inode already exists, hash and return it. Otherwise, return
2608 * NULL.
2609 *
2610 * Parent directory should be locked.
2611 */
2612struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2613{
2614 struct dentry *alias;
Al Viro668d0cd2016-03-08 12:44:17 -05002615 unsigned int hash = entry->d_name.hash;
2616
2617 spin_lock(&inode->i_lock);
2618 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2619 /*
2620 * Don't need alias->d_lock here, because aliases with
2621 * d_parent == entry->d_parent are not subject to name or
2622 * parent changes, because the parent inode i_mutex is held.
2623 */
2624 if (alias->d_name.hash != hash)
2625 continue;
2626 if (alias->d_parent != entry->d_parent)
2627 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002628 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
Al Viro668d0cd2016-03-08 12:44:17 -05002629 continue;
2630 spin_lock(&alias->d_lock);
2631 if (!d_unhashed(alias)) {
2632 spin_unlock(&alias->d_lock);
2633 alias = NULL;
2634 } else {
2635 __dget_dlock(alias);
Al Viro15d3c582016-07-29 17:45:21 -04002636 __d_rehash(alias);
Al Viro668d0cd2016-03-08 12:44:17 -05002637 spin_unlock(&alias->d_lock);
2638 }
2639 spin_unlock(&inode->i_lock);
2640 return alias;
2641 }
2642 spin_unlock(&inode->i_lock);
2643 return NULL;
2644}
2645EXPORT_SYMBOL(d_exact_alias);
2646
2647/**
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002648 * dentry_update_name_case - update case insensitive dentry with a new name
2649 * @dentry: dentry to be updated
2650 * @name: new name
2651 *
2652 * Update a case insensitive dentry with new case of name.
2653 *
2654 * dentry must have been returned by d_lookup with name @name. Old and new
2655 * name lengths must match (ie. no d_compare which allows mismatched name
2656 * lengths).
2657 *
2658 * Parent inode i_mutex must be held over d_lookup and into this call (to
2659 * keep renames and concurrent inserts, and readdir(2) away).
2660 */
Al Viro9aba36d2016-07-20 22:28:45 -04002661void dentry_update_name_case(struct dentry *dentry, const struct qstr *name)
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002662{
Al Viro59551022016-01-22 15:40:57 -05002663 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002664 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2665
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002666 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002667 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002668 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002669 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002670 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002671}
2672EXPORT_SYMBOL(dentry_update_name_case);
2673
Al Viro8d85b482014-09-29 14:54:27 -04002674static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675{
Al Viro8d85b482014-09-29 14:54:27 -04002676 if (unlikely(dname_external(target))) {
2677 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 /*
2679 * Both external: swap the pointers
2680 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002681 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 } else {
2683 /*
2684 * dentry:internal, target:external. Steal target's
2685 * storage and make target internal.
2686 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002687 memcpy(target->d_iname, dentry->d_name.name,
2688 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 dentry->d_name.name = target->d_name.name;
2690 target->d_name.name = target->d_iname;
2691 }
2692 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002693 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 /*
2695 * dentry:external, target:internal. Give dentry's
2696 * storage to target and make dentry internal
2697 */
2698 memcpy(dentry->d_iname, target->d_name.name,
2699 target->d_name.len + 1);
2700 target->d_name.name = dentry->d_name.name;
2701 dentry->d_name.name = dentry->d_iname;
2702 } else {
2703 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002704 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002706 unsigned int i;
2707 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002708 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2709 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002710 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2711 swap(((long *) &dentry->d_iname)[i],
2712 ((long *) &target->d_iname)[i]);
2713 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714 }
2715 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002716 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717}
2718
Al Viro8d85b482014-09-29 14:54:27 -04002719static void copy_name(struct dentry *dentry, struct dentry *target)
2720{
2721 struct external_name *old_name = NULL;
2722 if (unlikely(dname_external(dentry)))
2723 old_name = external_name(dentry);
2724 if (unlikely(dname_external(target))) {
2725 atomic_inc(&external_name(target)->u.count);
2726 dentry->d_name = target->d_name;
2727 } else {
2728 memcpy(dentry->d_iname, target->d_name.name,
2729 target->d_name.len + 1);
2730 dentry->d_name.name = dentry->d_iname;
2731 dentry->d_name.hash_len = target->d_name.hash_len;
2732 }
2733 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2734 kfree_rcu(old_name, u.head);
2735}
2736
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002737static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2738{
2739 /*
2740 * XXXX: do we really need to take target->d_lock?
2741 */
2742 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2743 spin_lock(&target->d_parent->d_lock);
2744 else {
2745 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2746 spin_lock(&dentry->d_parent->d_lock);
2747 spin_lock_nested(&target->d_parent->d_lock,
2748 DENTRY_D_LOCK_NESTED);
2749 } else {
2750 spin_lock(&target->d_parent->d_lock);
2751 spin_lock_nested(&dentry->d_parent->d_lock,
2752 DENTRY_D_LOCK_NESTED);
2753 }
2754 }
2755 if (target < dentry) {
2756 spin_lock_nested(&target->d_lock, 2);
2757 spin_lock_nested(&dentry->d_lock, 3);
2758 } else {
2759 spin_lock_nested(&dentry->d_lock, 2);
2760 spin_lock_nested(&target->d_lock, 3);
2761 }
2762}
2763
Al Viro986c0192014-09-26 23:11:15 -04002764static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002765{
2766 if (target->d_parent != dentry->d_parent)
2767 spin_unlock(&dentry->d_parent->d_lock);
2768 if (target->d_parent != target)
2769 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002770 spin_unlock(&target->d_lock);
2771 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002772}
2773
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002775 * When switching names, the actual string doesn't strictly have to
2776 * be preserved in the target - because we're dropping the target
2777 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002778 * the new name before we switch, unless we are going to rehash
2779 * it. Note that if we *do* unhash the target, we are not allowed
2780 * to rehash it without giving it a new name/hash key - whether
2781 * we swap or overwrite the names here, resulting name won't match
2782 * the reality in filesystem; it's only there for d_path() purposes.
2783 * Note that all of this is happening under rename_lock, so the
2784 * any hash lookup seeing it in the middle of manipulations will
2785 * be discarded anyway. So we do not care what happens to the hash
2786 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002788/*
Al Viro18367502011-07-12 21:42:24 -04002789 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 * @dentry: entry to move
2791 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002792 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 *
2794 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002795 * dcache entries should not be moved in this way. Caller must hold
2796 * rename_lock, the i_mutex of the source and target directories,
2797 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002799static void __d_move(struct dentry *dentry, struct dentry *target,
2800 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801{
Al Viro84e710d2016-04-15 00:58:55 -04002802 struct inode *dir = NULL;
2803 unsigned n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 if (!dentry->d_inode)
2805 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2806
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002807 BUG_ON(d_ancestor(dentry, target));
2808 BUG_ON(d_ancestor(target, dentry));
2809
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002810 dentry_lock_for_move(dentry, target);
Al Viro84e710d2016-04-15 00:58:55 -04002811 if (unlikely(d_in_lookup(target))) {
2812 dir = target->d_parent->d_inode;
2813 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002814 __d_lookup_done(target);
Al Viro84e710d2016-04-15 00:58:55 -04002815 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816
Nick Piggin31e6b012011-01-07 17:49:52 +11002817 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002818 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002819
Al Viro15d3c582016-07-29 17:45:21 -04002820 /* unhash both */
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002821 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002822 __d_drop(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 __d_drop(target);
2824
Linus Torvalds1da177e2005-04-16 15:20:36 -07002825 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002826 if (exchange)
2827 swap_names(dentry, target);
2828 else
2829 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830
Al Viro15d3c582016-07-29 17:45:21 -04002831 /* rehash in new place(s) */
2832 __d_rehash(dentry);
2833 if (exchange)
2834 __d_rehash(target);
2835
Al Viro63cf4272014-09-26 23:06:14 -04002836 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002838 /* splicing a tree */
Al Viro3d56c252016-06-07 21:26:55 -04002839 dentry->d_flags |= DCACHE_RCUACCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 dentry->d_parent = target->d_parent;
2841 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002842 list_del_init(&target->d_child);
2843 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002845 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002846 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002847 list_move(&target->d_child, &target->d_parent->d_subdirs);
2848 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002849 if (exchange)
Al Viroaffda482016-05-29 18:35:12 -04002850 fsnotify_update_flags(target);
2851 fsnotify_update_flags(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 }
2853
Nick Piggin31e6b012011-01-07 17:49:52 +11002854 write_seqcount_end(&target->d_seq);
2855 write_seqcount_end(&dentry->d_seq);
2856
Al Viro84e710d2016-04-15 00:58:55 -04002857 if (dir)
2858 end_dir_add(dir, n);
Al Viro986c0192014-09-26 23:11:15 -04002859 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002860}
2861
2862/*
2863 * d_move - move a dentry
2864 * @dentry: entry to move
2865 * @target: new dentry
2866 *
2867 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002868 * dcache entries should not be moved in this way. See the locking
2869 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002870 */
2871void d_move(struct dentry *dentry, struct dentry *target)
2872{
2873 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002874 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002876}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002877EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878
Miklos Szeredida1ce062014-04-01 17:08:43 +02002879/*
2880 * d_exchange - exchange two dentries
2881 * @dentry1: first dentry
2882 * @dentry2: second dentry
2883 */
2884void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2885{
2886 write_seqlock(&rename_lock);
2887
2888 WARN_ON(!dentry1->d_inode);
2889 WARN_ON(!dentry2->d_inode);
2890 WARN_ON(IS_ROOT(dentry1));
2891 WARN_ON(IS_ROOT(dentry2));
2892
2893 __d_move(dentry1, dentry2, true);
2894
2895 write_sequnlock(&rename_lock);
2896}
2897
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002898/**
2899 * d_ancestor - search for an ancestor
2900 * @p1: ancestor dentry
2901 * @p2: child dentry
2902 *
2903 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2904 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002905 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002906struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002907{
2908 struct dentry *p;
2909
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002910 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002911 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002912 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002913 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002914 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002915}
2916
2917/*
2918 * This helper attempts to cope with remotely renamed directories
2919 *
2920 * It assumes that the caller is already holding
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002921 * dentry->d_parent->d_inode->i_mutex, and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002922 *
2923 * Note: If ever the locking in lock_rename() changes, then please
2924 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002925 */
Al Virob5ae6b12014-10-12 22:16:02 -04002926static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002927 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002928{
Al Viro9902af72016-04-15 15:08:36 -04002929 struct mutex *m1 = NULL;
2930 struct rw_semaphore *m2 = NULL;
J. Bruce Fields3d330dc2015-02-10 10:55:53 -05002931 int ret = -ESTALE;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002932
2933 /* If alias and dentry share a parent, then no extra locks required */
2934 if (alias->d_parent == dentry->d_parent)
2935 goto out_unalias;
2936
Trond Myklebust9eaef272006-10-21 10:24:20 -07002937 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002938 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2939 goto out_err;
2940 m1 = &dentry->d_sb->s_vfs_rename_mutex;
Al Viro9902af72016-04-15 15:08:36 -04002941 if (!inode_trylock_shared(alias->d_parent->d_inode))
Trond Myklebust9eaef272006-10-21 10:24:20 -07002942 goto out_err;
Al Viro9902af72016-04-15 15:08:36 -04002943 m2 = &alias->d_parent->d_inode->i_rwsem;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002944out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002945 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04002946 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002947out_err:
Trond Myklebust9eaef272006-10-21 10:24:20 -07002948 if (m2)
Al Viro9902af72016-04-15 15:08:36 -04002949 up_read(m2);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002950 if (m1)
2951 mutex_unlock(m1);
2952 return ret;
2953}
2954
David Howells770bfad2006-08-22 20:06:07 -04002955/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002956 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2957 * @inode: the inode which may have a disconnected dentry
2958 * @dentry: a negative dentry which we want to point to the inode.
2959 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002960 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2961 * place of the given dentry and return it, else simply d_add the inode
2962 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002963 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002964 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2965 * we should error out: directories can't have multiple aliases.
2966 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002967 * This is needed in the lookup routine of any filesystem that is exportable
2968 * (via knfsd) so that we can build dcache paths to directories effectively.
2969 *
2970 * If a dentry was found and moved, then it is returned. Otherwise NULL
2971 * is returned. This matches the expected return value of ->lookup.
2972 *
2973 * Cluster filesystems may call this function with a negative, hashed dentry.
2974 * In that case, we know that the inode will be a regular file, and also this
2975 * will only occur during atomic_open. So we need to check for the dentry
2976 * being already hashed only in the final case.
2977 */
2978struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2979{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002980 if (IS_ERR(inode))
2981 return ERR_CAST(inode);
2982
David Howells770bfad2006-08-22 20:06:07 -04002983 BUG_ON(!d_unhashed(dentry));
2984
Al Virode689f52016-03-09 18:05:42 -05002985 if (!inode)
Al Virob5ae6b12014-10-12 22:16:02 -04002986 goto out;
Al Virode689f52016-03-09 18:05:42 -05002987
Al Virob9680912016-04-11 00:53:26 -04002988 security_d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11002989 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002990 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04002991 struct dentry *new = __d_find_any_alias(inode);
2992 if (unlikely(new)) {
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002993 /* The reference to new ensures it remains an alias */
2994 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002995 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002996 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04002997 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002998 dput(new);
2999 new = ERR_PTR(-ELOOP);
3000 pr_warn_ratelimited(
3001 "VFS: Lookup of '%s' in %s %s"
3002 " would have caused loop\n",
3003 dentry->d_name.name,
3004 inode->i_sb->s_type->name,
3005 inode->i_sb->s_id);
3006 } else if (!IS_ROOT(new)) {
3007 int err = __d_unalias(inode, dentry, new);
3008 write_sequnlock(&rename_lock);
3009 if (err) {
3010 dput(new);
3011 new = ERR_PTR(err);
3012 }
Al Viro18367502011-07-12 21:42:24 -04003013 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04003014 __d_move(new, dentry, false);
3015 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003016 }
Al Virob5ae6b12014-10-12 22:16:02 -04003017 iput(inode);
3018 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003019 }
David Howells770bfad2006-08-22 20:06:07 -04003020 }
Al Virob5ae6b12014-10-12 22:16:02 -04003021out:
Al Viroed782b52016-03-09 19:52:39 -05003022 __d_add(dentry, inode);
Al Virob5ae6b12014-10-12 22:16:02 -04003023 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04003024}
Al Virob5ae6b12014-10-12 22:16:02 -04003025EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04003026
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003027static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01003028{
3029 *buflen -= namelen;
3030 if (*buflen < 0)
3031 return -ENAMETOOLONG;
3032 *buffer -= namelen;
3033 memcpy(*buffer, str, namelen);
3034 return 0;
3035}
3036
Waiman Long232d2d62013-09-09 12:18:13 -04003037/**
3038 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04003039 * @buffer: buffer pointer
3040 * @buflen: allocated length of the buffer
3041 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04003042 *
3043 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3044 * make sure that either the old or the new name pointer and length are
3045 * fetched. However, there may be mismatch between length and pointer.
3046 * The length cannot be trusted, we need to copy it byte-by-byte until
3047 * the length is reached or a null byte is found. It also prepends "/" at
3048 * the beginning of the name. The sequence number check at the caller will
3049 * retry it again when a d_move() does happen. So any garbage in the buffer
3050 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04003051 *
3052 * Data dependency barrier is needed to make sure that we see that terminating
3053 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04003054 */
Al Viro9aba36d2016-07-20 22:28:45 -04003055static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003056{
Waiman Long232d2d62013-09-09 12:18:13 -04003057 const char *dname = ACCESS_ONCE(name->name);
3058 u32 dlen = ACCESS_ONCE(name->len);
3059 char *p;
3060
Al Viro6d13f692014-09-29 14:46:30 -04003061 smp_read_barrier_depends();
3062
Waiman Long232d2d62013-09-09 12:18:13 -04003063 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04003064 if (*buflen < 0)
3065 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04003066 p = *buffer -= dlen + 1;
3067 *p++ = '/';
3068 while (dlen--) {
3069 char c = *dname++;
3070 if (!c)
3071 break;
3072 *p++ = c;
3073 }
3074 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003075}
3076
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077/**
Randy Dunlap208898c2010-11-18 15:02:49 -08003078 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003079 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003080 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003081 * @buffer: pointer to the end of the buffer
3082 * @buflen: pointer to buffer length
3083 *
Waiman Long18129972013-09-12 10:55:35 -04003084 * The function will first try to write out the pathname without taking any
3085 * lock other than the RCU read lock to make sure that dentries won't go away.
3086 * It only checks the sequence number of the global rename_lock as any change
3087 * in the dentry's d_seq will be preceded by changes in the rename_lock
3088 * sequence number. If the sequence number had been changed, it will restart
3089 * the whole pathname back-tracing sequence again by taking the rename_lock.
3090 * In this case, there is no need to take the RCU read lock as the recursive
3091 * parent pointer references will keep the dentry chain alive as long as no
3092 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003093 */
Al Viro02125a82011-12-05 08:43:34 -05003094static int prepend_path(const struct path *path,
3095 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003096 char **buffer, int *buflen)
3097{
Al Viroede4ceb2013-11-13 07:45:40 -05003098 struct dentry *dentry;
3099 struct vfsmount *vfsmnt;
3100 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003101 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04003102 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04003103 char *bptr;
3104 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003105
Al Viro48f5ec22013-09-09 15:22:25 -04003106 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04003107restart_mnt:
3108 read_seqbegin_or_lock(&mount_lock, &m_seq);
3109 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003110 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003111restart:
3112 bptr = *buffer;
3113 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04003114 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05003115 dentry = path->dentry;
3116 vfsmnt = path->mnt;
3117 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04003118 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003119 while (dentry != root->dentry || vfsmnt != root->mnt) {
3120 struct dentry * parent;
3121
3122 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04003123 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Eric W. Biedermancde93be2015-08-15 13:36:12 -05003124 /* Escaped? */
3125 if (dentry != vfsmnt->mnt_root) {
3126 bptr = *buffer;
3127 blen = *buflen;
3128 error = 3;
3129 break;
3130 }
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003131 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04003132 if (mnt != parent) {
3133 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
3134 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04003135 vfsmnt = &mnt->mnt;
3136 continue;
3137 }
Waiman Long232d2d62013-09-09 12:18:13 -04003138 if (!error)
3139 error = is_mounted(vfsmnt) ? 1 : 2;
3140 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003141 }
3142 parent = dentry->d_parent;
3143 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003144 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003145 if (error)
3146 break;
3147
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003148 dentry = parent;
3149 }
Al Viro48f5ec22013-09-09 15:22:25 -04003150 if (!(seq & 1))
3151 rcu_read_unlock();
3152 if (need_seqretry(&rename_lock, seq)) {
3153 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003154 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003155 }
Al Viro48f5ec22013-09-09 15:22:25 -04003156 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003157
3158 if (!(m_seq & 1))
3159 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04003160 if (need_seqretry(&mount_lock, m_seq)) {
3161 m_seq = 1;
3162 goto restart_mnt;
3163 }
3164 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003165
Waiman Long232d2d62013-09-09 12:18:13 -04003166 if (error >= 0 && bptr == *buffer) {
3167 if (--blen < 0)
3168 error = -ENAMETOOLONG;
3169 else
3170 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003171 }
Waiman Long232d2d62013-09-09 12:18:13 -04003172 *buffer = bptr;
3173 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04003174 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003175}
3176
3177/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003178 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003179 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003180 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07003181 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182 * @buflen: buffer length
3183 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003184 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003186 * Returns a pointer into the buffer or an error code if the
3187 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08003188 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04003189 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003190 *
Al Viro02125a82011-12-05 08:43:34 -05003191 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 */
Al Viro02125a82011-12-05 08:43:34 -05003193char *__d_path(const struct path *path,
3194 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003195 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003197 char *res = buf + buflen;
3198 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003200 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003201 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04003202
Al Viro02125a82011-12-05 08:43:34 -05003203 if (error < 0)
3204 return ERR_PTR(error);
3205 if (error > 0)
3206 return NULL;
3207 return res;
3208}
3209
3210char *d_absolute_path(const struct path *path,
3211 char *buf, int buflen)
3212{
3213 struct path root = {};
3214 char *res = buf + buflen;
3215 int error;
3216
3217 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003218 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05003219
3220 if (error > 1)
3221 error = -EINVAL;
3222 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003223 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003224 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003225}
3226
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003227/*
3228 * same as __d_path but appends "(deleted)" for unlinked files.
3229 */
Al Viro02125a82011-12-05 08:43:34 -05003230static int path_with_deleted(const struct path *path,
3231 const struct path *root,
3232 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003233{
3234 prepend(buf, buflen, "\0", 1);
3235 if (d_unlinked(path->dentry)) {
3236 int error = prepend(buf, buflen, " (deleted)", 10);
3237 if (error)
3238 return error;
3239 }
3240
3241 return prepend_path(path, root, buf, buflen);
3242}
3243
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003244static int prepend_unreachable(char **buffer, int *buflen)
3245{
3246 return prepend(buffer, buflen, "(unreachable)", 13);
3247}
3248
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003249static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3250{
3251 unsigned seq;
3252
3253 do {
3254 seq = read_seqcount_begin(&fs->seq);
3255 *root = fs->root;
3256 } while (read_seqcount_retry(&fs->seq, seq));
3257}
3258
Jan Bluncka03a8a702008-02-14 19:38:32 -08003259/**
3260 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003261 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003262 * @buf: buffer to return value in
3263 * @buflen: buffer length
3264 *
3265 * Convert a dentry into an ASCII path name. If the entry has been deleted
3266 * the string " (deleted)" is appended. Note that this is ambiguous.
3267 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003268 * Returns a pointer into the buffer or an error code if the path was
3269 * too long. Note: Callers should use the returned pointer, not the passed
3270 * in buffer, to use the name! The implementation often starts at an offset
3271 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003272 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003273 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003274 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003275char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003277 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003278 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003279 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003281 /*
3282 * We have various synthetic filesystems that never get mounted. On
3283 * these filesystems dentries are never used for lookup purposes, and
3284 * thus don't need to be hashed. They also don't need a name until a
3285 * user wants to identify the object in /proc/pid/fd/. The little hack
3286 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003287 *
3288 * Some pseudo inodes are mountable. When they are mounted
3289 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3290 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003291 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003292 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3293 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003294 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003295
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003296 rcu_read_lock();
3297 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003298 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003299 rcu_read_unlock();
3300
Al Viro02125a82011-12-05 08:43:34 -05003301 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003302 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303 return res;
3304}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003305EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306
3307/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003308 * Helper function for dentry_operations.d_dname() members
3309 */
3310char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3311 const char *fmt, ...)
3312{
3313 va_list args;
3314 char temp[64];
3315 int sz;
3316
3317 va_start(args, fmt);
3318 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3319 va_end(args);
3320
3321 if (sz > sizeof(temp) || sz > buflen)
3322 return ERR_PTR(-ENAMETOOLONG);
3323
3324 buffer += buflen - sz;
3325 return memcpy(buffer, temp, sz);
3326}
3327
Al Viro118b2302013-08-24 12:08:17 -04003328char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3329{
3330 char *end = buffer + buflen;
3331 /* these dentries are never renamed, so d_lock is not needed */
3332 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003333 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003334 prepend(&end, &buflen, "/", 1))
3335 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003336 return end;
Al Viro118b2302013-08-24 12:08:17 -04003337}
David Herrmann31bbe162014-01-03 14:09:47 +01003338EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003339
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003340/*
Ram Pai6092d042008-03-27 13:06:20 +01003341 * Write full pathname from the root of the filesystem into the buffer.
3342 */
Al Virof6500802014-01-26 12:37:55 -05003343static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003344{
Al Virof6500802014-01-26 12:37:55 -05003345 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003346 char *end, *retval;
3347 int len, seq = 0;
3348 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003349
Al Virof6500802014-01-26 12:37:55 -05003350 if (buflen < 2)
3351 goto Elong;
3352
Al Viro48f5ec22013-09-09 15:22:25 -04003353 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003354restart:
Al Virof6500802014-01-26 12:37:55 -05003355 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003356 end = buf + buflen;
3357 len = buflen;
3358 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003359 /* Get '/' right */
3360 retval = end-1;
3361 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003362 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003363 while (!IS_ROOT(dentry)) {
3364 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003365
Ram Pai6092d042008-03-27 13:06:20 +01003366 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003367 error = prepend_name(&end, &len, &dentry->d_name);
3368 if (error)
3369 break;
Ram Pai6092d042008-03-27 13:06:20 +01003370
3371 retval = end;
3372 dentry = parent;
3373 }
Al Viro48f5ec22013-09-09 15:22:25 -04003374 if (!(seq & 1))
3375 rcu_read_unlock();
3376 if (need_seqretry(&rename_lock, seq)) {
3377 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003378 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003379 }
3380 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003381 if (error)
3382 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003383 return retval;
3384Elong:
3385 return ERR_PTR(-ENAMETOOLONG);
3386}
Nick Pigginec2447c2011-01-07 17:49:29 +11003387
3388char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3389{
Waiman Long232d2d62013-09-09 12:18:13 -04003390 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003391}
3392EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003393
3394char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3395{
3396 char *p = NULL;
3397 char *retval;
3398
Al Viroc1031352010-06-06 22:31:14 -04003399 if (d_unlinked(dentry)) {
3400 p = buf + buflen;
3401 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3402 goto Elong;
3403 buflen++;
3404 }
3405 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003406 if (!IS_ERR(retval) && p)
3407 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003408 return retval;
3409Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003410 return ERR_PTR(-ENAMETOOLONG);
3411}
3412
Linus Torvalds8b19e342013-09-12 10:35:47 -07003413static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3414 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003415{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003416 unsigned seq;
3417
3418 do {
3419 seq = read_seqcount_begin(&fs->seq);
3420 *root = fs->root;
3421 *pwd = fs->pwd;
3422 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003423}
3424
Ram Pai6092d042008-03-27 13:06:20 +01003425/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 * NOTE! The user-level library version returns a
3427 * character pointer. The kernel system call just
3428 * returns the length of the buffer filled (which
3429 * includes the ending '\0' character), or a negative
3430 * error value. So libc would do something like
3431 *
3432 * char *getcwd(char * buf, size_t size)
3433 * {
3434 * int retval;
3435 *
3436 * retval = sys_getcwd(buf, size);
3437 * if (retval >= 0)
3438 * return buf;
3439 * errno = -retval;
3440 * return NULL;
3441 * }
3442 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003443SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444{
Linus Torvalds552ce542007-02-13 12:08:18 -08003445 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003446 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003447 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448
3449 if (!page)
3450 return -ENOMEM;
3451
Linus Torvalds8b19e342013-09-12 10:35:47 -07003452 rcu_read_lock();
3453 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454
Linus Torvalds552ce542007-02-13 12:08:18 -08003455 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003456 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003457 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003458 char *cwd = page + PATH_MAX;
3459 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003461 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003462 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003463 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003464
Al Viro02125a82011-12-05 08:43:34 -05003465 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003466 goto out;
3467
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003468 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003469 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003470 error = prepend_unreachable(&cwd, &buflen);
3471 if (error)
3472 goto out;
3473 }
3474
Linus Torvalds552ce542007-02-13 12:08:18 -08003475 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003476 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003477 if (len <= size) {
3478 error = len;
3479 if (copy_to_user(buf, cwd, len))
3480 error = -EFAULT;
3481 }
Nick Piggin949854d2011-01-07 17:49:37 +11003482 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003483 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003484 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485
3486out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003487 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488 return error;
3489}
3490
3491/*
3492 * Test whether new_dentry is a subdirectory of old_dentry.
3493 *
3494 * Trivially implemented using the dcache structure
3495 */
3496
3497/**
3498 * is_subdir - is new dentry a subdirectory of old_dentry
3499 * @new_dentry: new dentry
3500 * @old_dentry: old dentry
3501 *
Yaowei Baia6e57872015-11-17 14:40:11 +08003502 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3503 * Returns false otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003504 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3505 */
3506
Yaowei Baia6e57872015-11-17 14:40:11 +08003507bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508{
Yaowei Baia6e57872015-11-17 14:40:11 +08003509 bool result;
Nick Piggin949854d2011-01-07 17:49:37 +11003510 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003512 if (new_dentry == old_dentry)
Yaowei Baia6e57872015-11-17 14:40:11 +08003513 return true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003514
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003515 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003518 /*
3519 * Need rcu_readlock to protect against the d_parent trashing
3520 * due to d_move
3521 */
3522 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003523 if (d_ancestor(old_dentry, new_dentry))
Yaowei Baia6e57872015-11-17 14:40:11 +08003524 result = true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003525 else
Yaowei Baia6e57872015-11-17 14:40:11 +08003526 result = false;
Nick Piggin949854d2011-01-07 17:49:37 +11003527 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003528 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003529
3530 return result;
3531}
3532
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003533static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003534{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003535 struct dentry *root = data;
3536 if (dentry != root) {
3537 if (d_unhashed(dentry) || !dentry->d_inode)
3538 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003539
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003540 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3541 dentry->d_flags |= DCACHE_GENOCIDE;
3542 dentry->d_lockref.count--;
3543 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003544 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003545 return D_WALK_CONTINUE;
3546}
Nick Piggin58db63d2011-01-07 17:49:39 +11003547
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003548void d_genocide(struct dentry *parent)
3549{
3550 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551}
3552
Al Viro60545d02013-06-07 01:20:27 -04003553void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003554{
Al Viro60545d02013-06-07 01:20:27 -04003555 inode_dec_link_count(inode);
3556 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003557 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003558 !d_unlinked(dentry));
3559 spin_lock(&dentry->d_parent->d_lock);
3560 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3561 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3562 (unsigned long long)inode->i_ino);
3563 spin_unlock(&dentry->d_lock);
3564 spin_unlock(&dentry->d_parent->d_lock);
3565 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003566}
Al Viro60545d02013-06-07 01:20:27 -04003567EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003568
3569static __initdata unsigned long dhash_entries;
3570static int __init set_dhash_entries(char *str)
3571{
3572 if (!str)
3573 return 0;
3574 dhash_entries = simple_strtoul(str, &str, 0);
3575 return 1;
3576}
3577__setup("dhash_entries=", set_dhash_entries);
3578
3579static void __init dcache_init_early(void)
3580{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581 /* If hashes are distributed across NUMA nodes, defer
3582 * hash allocation until vmalloc space is available.
3583 */
3584 if (hashdist)
3585 return;
3586
3587 dentry_hashtable =
3588 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003589 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590 dhash_entries,
3591 13,
Pavel Tatashin3d375d72017-07-06 15:39:11 -07003592 HASH_EARLY | HASH_ZERO,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593 &d_hash_shift,
3594 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003595 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003596 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597}
3598
Denis Cheng74bf17c2007-10-16 23:26:30 -07003599static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003600{
Pavel Tatashin3d375d72017-07-06 15:39:11 -07003601 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602 * A constructor could be added for stable state like the lists,
3603 * but it is probably not worth it because of the cache nature
Pavel Tatashin3d375d72017-07-06 15:39:11 -07003604 * of the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003605 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003606 dentry_cache = KMEM_CACHE(dentry,
Vladimir Davydov5d097052016-01-14 15:18:21 -08003607 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608
3609 /* Hash may have been set up in dcache_init_early */
3610 if (!hashdist)
3611 return;
3612
3613 dentry_hashtable =
3614 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003615 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616 dhash_entries,
3617 13,
Pavel Tatashin3d375d72017-07-06 15:39:11 -07003618 HASH_ZERO,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003619 &d_hash_shift,
3620 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003621 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003622 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623}
3624
3625/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003626struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003627EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003628
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629EXPORT_SYMBOL(d_genocide);
3630
Linus Torvalds1da177e2005-04-16 15:20:36 -07003631void __init vfs_caches_init_early(void)
3632{
Sebastian Andrzej Siewior69163632017-06-27 18:19:11 +02003633 int i;
3634
3635 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3636 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3637
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638 dcache_init_early();
3639 inode_init_early();
3640}
3641
Mel Gorman4248b0d2015-08-06 15:46:20 -07003642void __init vfs_caches_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003645 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646
Denis Cheng74bf17c2007-10-16 23:26:30 -07003647 dcache_init();
3648 inode_init();
Mel Gorman4248b0d2015-08-06 15:46:20 -07003649 files_init();
3650 files_maxfiles_init();
Denis Cheng74bf17c2007-10-16 23:26:30 -07003651 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652 bdev_cache_init();
3653 chrdev_init();
3654}