blob: f9c63c108881daaee1f8a1bbb3d384009cb977de [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/dcache.c
3 *
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
8
9/*
10 * Notes on the allocation strategy:
11 *
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040021#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/hash.h>
25#include <linux/cache.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/file.h>
29#include <asm/uaccess.h>
30#include <linux/security.h>
31#include <linux/seqlock.h>
32#include <linux/swap.h>
33#include <linux/bootmem.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Frederic Weisbecker613afbf2009-07-16 15:44:29 +020035#include <linux/hardirq.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110036#include <linux/bit_spinlock.h>
37#include <linux/rculist_bl.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070038#include <linux/prefetch.h>
David Howellsdd179942011-08-16 15:31:30 +010039#include <linux/ratelimit.h>
Dave Chinnerf6041562013-08-28 10:18:00 +100040#include <linux/list_lru.h>
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -080041#include <linux/kasan.h>
42
David Howells07f3f052006-09-30 20:52:18 +020043#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050044#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Nick Piggin789680d2011-01-07 17:49:30 +110046/*
47 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110048 * dcache->d_inode->i_lock protects:
Al Viro946e51f2014-10-26 19:19:16 -040049 * - i_dentry, d_u.d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110050 * dcache_hash_bucket lock protects:
51 * - the dcache hash table
52 * s_anon bl list spinlock protects:
53 * - the s_anon list (see __d_drop)
Dave Chinner19156842013-08-28 10:17:55 +100054 * dentry->d_sb->s_dentry_lru_lock protects:
Nick Piggin23044502011-01-07 17:49:31 +110055 * - the dcache lru lists and counters
56 * d_lock protects:
57 * - d_flags
58 * - d_name
59 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110060 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110061 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110062 * - d_parent and d_subdirs
63 * - childrens' d_child and d_parent
Al Viro946e51f2014-10-26 19:19:16 -040064 * - d_u.d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110065 *
66 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110067 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110068 * dentry->d_lock
Dave Chinner19156842013-08-28 10:17:55 +100069 * dentry->d_sb->s_dentry_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110070 * dcache_hash_bucket lock
71 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110072 *
Nick Pigginda502952011-01-07 17:49:33 +110073 * If there is an ancestor relationship:
74 * dentry->d_parent->...->d_parent->d_lock
75 * ...
76 * dentry->d_parent->d_lock
77 * dentry->d_lock
78 *
79 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110080 * if (dentry1 < dentry2)
81 * dentry1->d_lock
82 * dentry2->d_lock
83 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080084int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
86
Al Viro74c3cbe2007-07-22 08:04:18 -040087__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Nick Piggin949854d2011-01-07 17:49:37 +110089EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Christoph Lametere18b8902006-12-06 20:33:20 -080091static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Linus Torvalds1da177e2005-04-16 15:20:36 -070093/*
94 * This is the single most critical data structure when it comes
95 * to the dcache: the hashtable for lookups. Somebody should try
96 * to make this good - I've just made it work.
97 *
98 * This hash-function tries to avoid losing too many bits of hash
99 * information, yet avoid using a prime hash-size or similar.
100 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800102static unsigned int d_hash_mask __read_mostly;
103static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700105static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100106
Linus Torvalds8966be92012-03-02 14:23:30 -0800107static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700108 unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100109{
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700110 hash += (unsigned long) parent / L1_CACHE_BYTES;
Linus Torvalds99d263d2014-09-13 11:30:10 -0700111 return dentry_hashtable + hash_32(hash, d_hash_shift);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100112}
113
Al Viro94bdd652016-04-15 02:42:04 -0400114#define IN_LOOKUP_SHIFT 10
115static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
116
117static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
118 unsigned int hash)
119{
120 hash += (unsigned long) parent / L1_CACHE_BYTES;
121 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
122}
123
124
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125/* Statistics gathering. */
126struct dentry_stat_t dentry_stat = {
127 .age_limit = 45,
128};
129
Glauber Costa3942c072013-08-28 10:17:53 +1000130static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000131static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400132
133#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000134
135/*
136 * Here we resort to our own counters instead of using generic per-cpu counters
137 * for consistency with what the vfs inode code does. We are expected to harvest
138 * better code and performance by having our own specialized counters.
139 *
140 * Please note that the loop is done over all possible CPUs, not over all online
141 * CPUs. The reason for this is that we don't want to play games with CPUs going
142 * on and off. If one of them goes off, we will just keep their counters.
143 *
144 * glommer: See cffbc8a for details, and if you ever intend to change this,
145 * please update all vfs counters to match.
146 */
Glauber Costa3942c072013-08-28 10:17:53 +1000147static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100148{
149 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000150 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100151 for_each_possible_cpu(i)
152 sum += per_cpu(nr_dentry, i);
153 return sum < 0 ? 0 : sum;
154}
155
Dave Chinner62d36c72013-08-28 10:17:54 +1000156static long get_nr_dentry_unused(void)
157{
158 int i;
159 long sum = 0;
160 for_each_possible_cpu(i)
161 sum += per_cpu(nr_dentry_unused, i);
162 return sum < 0 ? 0 : sum;
163}
164
Joe Perches1f7e0612014-06-06 14:38:05 -0700165int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400166 size_t *lenp, loff_t *ppos)
167{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100168 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000169 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000170 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400171}
172#endif
173
Linus Torvalds5483f182012-03-04 15:51:42 -0800174/*
175 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
176 * The strings are both count bytes long, and count is non-zero.
177 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700178#ifdef CONFIG_DCACHE_WORD_ACCESS
179
180#include <asm/word-at-a-time.h>
181/*
182 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
183 * aligned allocation for this particular component. We don't
184 * strictly need the load_unaligned_zeropad() safety, but it
185 * doesn't hurt either.
186 *
187 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
188 * need the careful unaligned handling.
189 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700190static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800191{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800192 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800193
194 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700195 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700196 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800197 if (tcount < sizeof(unsigned long))
198 break;
199 if (unlikely(a != b))
200 return 1;
201 cs += sizeof(unsigned long);
202 ct += sizeof(unsigned long);
203 tcount -= sizeof(unsigned long);
204 if (!tcount)
205 return 0;
206 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000207 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800208 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700209}
210
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800211#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700212
Linus Torvalds94753db52012-05-10 12:19:19 -0700213static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700214{
Linus Torvalds5483f182012-03-04 15:51:42 -0800215 do {
216 if (*cs != *ct)
217 return 1;
218 cs++;
219 ct++;
220 tcount--;
221 } while (tcount);
222 return 0;
223}
224
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700225#endif
226
Linus Torvalds94753db52012-05-10 12:19:19 -0700227static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
228{
Linus Torvalds6326c712012-05-21 16:14:04 -0700229 const unsigned char *cs;
Linus Torvalds94753db52012-05-10 12:19:19 -0700230 /*
231 * Be careful about RCU walk racing with rename:
232 * use ACCESS_ONCE to fetch the name pointer.
233 *
234 * NOTE! Even if a rename will mean that the length
235 * was not loaded atomically, we don't care. The
236 * RCU walk will check the sequence count eventually,
237 * and catch it. And we won't overrun the buffer,
238 * because we're reading the name pointer atomically,
239 * and a dentry name is guaranteed to be properly
240 * terminated with a NUL byte.
241 *
242 * End result: even if 'len' is wrong, we'll exit
243 * early because the data cannot match (there can
244 * be no NUL in the ct/tcount data)
245 */
Linus Torvalds6326c712012-05-21 16:14:04 -0700246 cs = ACCESS_ONCE(dentry->d_name.name);
247 smp_read_barrier_depends();
248 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700249}
250
Al Viro8d85b482014-09-29 14:54:27 -0400251struct external_name {
252 union {
253 atomic_t count;
254 struct rcu_head head;
255 } u;
256 unsigned char name[];
257};
258
259static inline struct external_name *external_name(struct dentry *dentry)
260{
261 return container_of(dentry->d_name.name, struct external_name, name[0]);
262}
263
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400264static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400266 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
267
Al Viro8d85b482014-09-29 14:54:27 -0400268 kmem_cache_free(dentry_cache, dentry);
269}
270
271static void __d_free_external(struct rcu_head *head)
272{
273 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
Al Viro8d85b482014-09-29 14:54:27 -0400274 kfree(external_name(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 kmem_cache_free(dentry_cache, dentry);
276}
277
Al Viro810bb172014-10-12 12:45:37 -0400278static inline int dname_external(const struct dentry *dentry)
279{
280 return dentry->d_name.name != dentry->d_iname;
281}
282
David Howells4bf46a22015-03-05 14:09:22 +0000283static inline void __d_set_inode_and_type(struct dentry *dentry,
284 struct inode *inode,
285 unsigned type_flags)
286{
287 unsigned flags;
288
289 dentry->d_inode = inode;
David Howells4bf46a22015-03-05 14:09:22 +0000290 flags = READ_ONCE(dentry->d_flags);
291 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
292 flags |= type_flags;
293 WRITE_ONCE(dentry->d_flags, flags);
294}
295
David Howells4bf46a22015-03-05 14:09:22 +0000296static inline void __d_clear_type_and_inode(struct dentry *dentry)
297{
298 unsigned flags = READ_ONCE(dentry->d_flags);
299
300 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
301 WRITE_ONCE(dentry->d_flags, flags);
David Howells4bf46a22015-03-05 14:09:22 +0000302 dentry->d_inode = NULL;
303}
304
Al Virob4f03542014-04-29 23:40:14 -0400305static void dentry_free(struct dentry *dentry)
306{
Al Viro946e51f2014-10-26 19:19:16 -0400307 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400308 if (unlikely(dname_external(dentry))) {
309 struct external_name *p = external_name(dentry);
310 if (likely(atomic_dec_and_test(&p->u.count))) {
311 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
312 return;
313 }
314 }
Al Virob4f03542014-04-29 23:40:14 -0400315 /* if dentry was never visible to RCU, immediate free is OK */
316 if (!(dentry->d_flags & DCACHE_RCUACCESS))
317 __d_free(&dentry->d_u.d_rcu);
318 else
319 call_rcu(&dentry->d_u.d_rcu, __d_free);
320}
321
Nick Piggin31e6b012011-01-07 17:49:52 +1100322/**
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200323 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800324 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100325 * After this call, in-progress rcu-walk path lookup will fail. This
326 * should be called after unhashing, and after changing d_inode (if
327 * the dentry has not already been unhashed).
328 */
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200329static inline void dentry_rcuwalk_invalidate(struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100330{
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200331 lockdep_assert_held(&dentry->d_lock);
332 /* Go through am invalidation barrier */
333 write_seqcount_invalidate(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100334}
335
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336/*
337 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100338 * d_iput() operation if defined. Dentry has no refcount
339 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800341static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200342 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100343 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344{
345 struct inode *inode = dentry->d_inode;
346 if (inode) {
David Howells4bf46a22015-03-05 14:09:22 +0000347 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400348 hlist_del_init(&dentry->d_u.d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100350 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700351 if (!inode->i_nlink)
352 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 if (dentry->d_op && dentry->d_op->d_iput)
354 dentry->d_op->d_iput(dentry, inode);
355 else
356 iput(inode);
357 } else {
358 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 }
360}
361
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700362/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100363 * Release the dentry's inode, using the filesystem
364 * d_iput() operation if defined. dentry remains in-use.
365 */
366static void dentry_unlink_inode(struct dentry * dentry)
367 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100368 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100369{
370 struct inode *inode = dentry->d_inode;
Al Viroa528aca2016-02-29 12:12:46 -0500371
372 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +0000373 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400374 hlist_del_init(&dentry->d_u.d_alias);
Al Viroa528aca2016-02-29 12:12:46 -0500375 raw_write_seqcount_end(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100376 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100377 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100378 if (!inode->i_nlink)
379 fsnotify_inoderemove(inode);
380 if (dentry->d_op && dentry->d_op->d_iput)
381 dentry->d_op->d_iput(dentry, inode);
382 else
383 iput(inode);
384}
385
386/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400387 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
388 * is in use - which includes both the "real" per-superblock
389 * LRU list _and_ the DCACHE_SHRINK_LIST use.
390 *
391 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
392 * on the shrink list (ie not on the superblock LRU list).
393 *
394 * The per-cpu "nr_dentry_unused" counters are updated with
395 * the DCACHE_LRU_LIST bit.
396 *
397 * These helper functions make sure we always follow the
398 * rules. d_lock must be held by the caller.
399 */
400#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
401static void d_lru_add(struct dentry *dentry)
402{
403 D_FLAG_VERIFY(dentry, 0);
404 dentry->d_flags |= DCACHE_LRU_LIST;
405 this_cpu_inc(nr_dentry_unused);
406 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
407}
408
409static void d_lru_del(struct dentry *dentry)
410{
411 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
412 dentry->d_flags &= ~DCACHE_LRU_LIST;
413 this_cpu_dec(nr_dentry_unused);
414 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
415}
416
417static void d_shrink_del(struct dentry *dentry)
418{
419 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
420 list_del_init(&dentry->d_lru);
421 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
422 this_cpu_dec(nr_dentry_unused);
423}
424
425static void d_shrink_add(struct dentry *dentry, struct list_head *list)
426{
427 D_FLAG_VERIFY(dentry, 0);
428 list_add(&dentry->d_lru, list);
429 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
430 this_cpu_inc(nr_dentry_unused);
431}
432
433/*
434 * These can only be called under the global LRU lock, ie during the
435 * callback for freeing the LRU list. "isolate" removes it from the
436 * LRU lists entirely, while shrink_move moves it to the indicated
437 * private list.
438 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800439static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400440{
441 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
442 dentry->d_flags &= ~DCACHE_LRU_LIST;
443 this_cpu_dec(nr_dentry_unused);
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800444 list_lru_isolate(lru, &dentry->d_lru);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400445}
446
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800447static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
448 struct list_head *list)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400449{
450 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
451 dentry->d_flags |= DCACHE_SHRINK_LIST;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800452 list_lru_isolate_move(lru, &dentry->d_lru, list);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400453}
454
455/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000456 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700457 */
458static void dentry_lru_add(struct dentry *dentry)
459{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400460 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
461 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700462}
463
Miklos Szeredid52b9082007-05-08 00:23:46 -0700464/**
Nick Piggin789680d2011-01-07 17:49:30 +1100465 * d_drop - drop a dentry
466 * @dentry: dentry to drop
467 *
468 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
469 * be found through a VFS lookup any more. Note that this is different from
470 * deleting the dentry - d_delete will try to mark the dentry negative if
471 * possible, giving a successful _negative_ lookup, while d_drop will
472 * just make the cache lookup fail.
473 *
474 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
475 * reason (NFS timeouts or autofs deletes).
476 *
477 * __d_drop requires dentry->d_lock.
478 */
479void __d_drop(struct dentry *dentry)
480{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700481 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400482 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400483 /*
484 * Hashed dentries are normally on the dentry hashtable,
485 * with the exception of those newly allocated by
486 * d_obtain_alias, which are always IS_ROOT:
487 */
488 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400489 b = &dentry->d_sb->s_anon;
490 else
491 b = d_hash(dentry->d_parent, dentry->d_name.hash);
492
493 hlist_bl_lock(b);
494 __hlist_bl_del(&dentry->d_hash);
495 dentry->d_hash.pprev = NULL;
496 hlist_bl_unlock(b);
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200497 dentry_rcuwalk_invalidate(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100498 }
499}
500EXPORT_SYMBOL(__d_drop);
501
502void d_drop(struct dentry *dentry)
503{
Nick Piggin789680d2011-01-07 17:49:30 +1100504 spin_lock(&dentry->d_lock);
505 __d_drop(dentry);
506 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100507}
508EXPORT_SYMBOL(d_drop);
509
Al Viroe55fd012014-05-28 13:51:12 -0400510static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100511{
Al Viro41edf272014-05-01 10:30:00 -0400512 struct dentry *parent = NULL;
513 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400514 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100515 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100516
Linus Torvalds0d984392013-09-08 13:46:52 -0700517 /*
518 * The dentry is now unrecoverably dead to the world.
519 */
520 lockref_mark_dead(&dentry->d_lockref);
521
Sage Weilf0023bc2011-10-28 10:02:42 -0700522 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700523 * inform the fs via d_prune that this dentry is about to be
524 * unhashed and destroyed.
525 */
Al Viro29266202014-05-30 11:39:02 -0400526 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800527 dentry->d_op->d_prune(dentry);
528
Al Viro01b60352014-04-29 23:42:52 -0400529 if (dentry->d_flags & DCACHE_LRU_LIST) {
530 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
531 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400532 }
Nick Piggin77812a12011-01-07 17:49:48 +1100533 /* if it was on the hash then remove it */
534 __d_drop(dentry);
Al Viroca5358e2014-10-26 19:31:10 -0400535 __list_del_entry(&dentry->d_child);
Al Viro03b3b882014-04-29 15:45:28 -0400536 /*
537 * Inform d_walk() that we are no longer attached to the
538 * dentry tree
539 */
540 dentry->d_flags |= DCACHE_DENTRY_KILLED;
541 if (parent)
542 spin_unlock(&parent->d_lock);
543 dentry_iput(dentry);
544 /*
545 * dentry_iput drops the locks, at which point nobody (except
546 * transient RCU lookups) can reach this dentry.
547 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800548 BUG_ON(dentry->d_lockref.count > 0);
Al Viro03b3b882014-04-29 15:45:28 -0400549 this_cpu_dec(nr_dentry);
550 if (dentry->d_op && dentry->d_op->d_release)
551 dentry->d_op->d_release(dentry);
552
Al Viro41edf272014-05-01 10:30:00 -0400553 spin_lock(&dentry->d_lock);
554 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
555 dentry->d_flags |= DCACHE_MAY_FREE;
556 can_free = false;
557 }
558 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400559 if (likely(can_free))
560 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400561}
562
563/*
564 * Finish off a dentry we've decided to kill.
565 * dentry->d_lock must be held, returns with it unlocked.
566 * If ref is non-zero, then decrement the refcount too.
567 * Returns dentry requiring refcount drop, or NULL if we're done.
568 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400569static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400570 __releases(dentry->d_lock)
571{
572 struct inode *inode = dentry->d_inode;
573 struct dentry *parent = NULL;
574
575 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
576 goto failed;
577
578 if (!IS_ROOT(dentry)) {
579 parent = dentry->d_parent;
580 if (unlikely(!spin_trylock(&parent->d_lock))) {
581 if (inode)
582 spin_unlock(&inode->i_lock);
583 goto failed;
584 }
585 }
586
587 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400588 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400589
590failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400591 spin_unlock(&dentry->d_lock);
592 cpu_relax();
Al Viroe55fd012014-05-28 13:51:12 -0400593 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100594}
595
Al Viro046b9612014-05-29 08:54:52 -0400596static inline struct dentry *lock_parent(struct dentry *dentry)
597{
598 struct dentry *parent = dentry->d_parent;
599 if (IS_ROOT(dentry))
600 return NULL;
Linus Torvalds360f5472015-01-09 15:19:03 -0800601 if (unlikely(dentry->d_lockref.count < 0))
Al Viroc2338f22014-06-12 00:29:13 -0400602 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400603 if (likely(spin_trylock(&parent->d_lock)))
604 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400605 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400606 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400607again:
608 parent = ACCESS_ONCE(dentry->d_parent);
609 spin_lock(&parent->d_lock);
610 /*
611 * We can't blindly lock dentry until we are sure
612 * that we won't violate the locking order.
613 * Any changes of dentry->d_parent must have
614 * been done with parent->d_lock held, so
615 * spin_lock() above is enough of a barrier
616 * for checking if it's still our child.
617 */
618 if (unlikely(parent != dentry->d_parent)) {
619 spin_unlock(&parent->d_lock);
620 goto again;
621 }
622 rcu_read_unlock();
623 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700624 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400625 else
626 parent = NULL;
627 return parent;
628}
629
Linus Torvalds360f5472015-01-09 15:19:03 -0800630/*
631 * Try to do a lockless dput(), and return whether that was successful.
632 *
633 * If unsuccessful, we return false, having already taken the dentry lock.
634 *
635 * The caller needs to hold the RCU read lock, so that the dentry is
636 * guaranteed to stay around even if the refcount goes down to zero!
637 */
638static inline bool fast_dput(struct dentry *dentry)
639{
640 int ret;
641 unsigned int d_flags;
642
643 /*
644 * If we have a d_op->d_delete() operation, we sould not
Al Viro75a6f822015-07-08 02:42:38 +0100645 * let the dentry count go to zero, so use "put_or_lock".
Linus Torvalds360f5472015-01-09 15:19:03 -0800646 */
647 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
648 return lockref_put_or_lock(&dentry->d_lockref);
649
650 /*
651 * .. otherwise, we can try to just decrement the
652 * lockref optimistically.
653 */
654 ret = lockref_put_return(&dentry->d_lockref);
655
656 /*
657 * If the lockref_put_return() failed due to the lock being held
658 * by somebody else, the fast path has failed. We will need to
659 * get the lock, and then check the count again.
660 */
661 if (unlikely(ret < 0)) {
662 spin_lock(&dentry->d_lock);
663 if (dentry->d_lockref.count > 1) {
664 dentry->d_lockref.count--;
665 spin_unlock(&dentry->d_lock);
666 return 1;
667 }
668 return 0;
669 }
670
671 /*
672 * If we weren't the last ref, we're done.
673 */
674 if (ret)
675 return 1;
676
677 /*
678 * Careful, careful. The reference count went down
679 * to zero, but we don't hold the dentry lock, so
680 * somebody else could get it again, and do another
681 * dput(), and we need to not race with that.
682 *
683 * However, there is a very special and common case
684 * where we don't care, because there is nothing to
685 * do: the dentry is still hashed, it does not have
686 * a 'delete' op, and it's referenced and already on
687 * the LRU list.
688 *
689 * NOTE! Since we aren't locked, these values are
690 * not "stable". However, it is sufficient that at
691 * some point after we dropped the reference the
692 * dentry was hashed and the flags had the proper
693 * value. Other dentry users may have re-gotten
694 * a reference to the dentry and change that, but
695 * our work is done - we can leave the dentry
696 * around with a zero refcount.
697 */
698 smp_rmb();
699 d_flags = ACCESS_ONCE(dentry->d_flags);
Al Viro75a6f822015-07-08 02:42:38 +0100700 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
Linus Torvalds360f5472015-01-09 15:19:03 -0800701
702 /* Nothing to do? Dropping the reference was all we needed? */
703 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
704 return 1;
705
706 /*
707 * Not the fast normal case? Get the lock. We've already decremented
708 * the refcount, but we'll need to re-check the situation after
709 * getting the lock.
710 */
711 spin_lock(&dentry->d_lock);
712
713 /*
714 * Did somebody else grab a reference to it in the meantime, and
715 * we're no longer the last user after all? Alternatively, somebody
716 * else could have killed it and marked it dead. Either way, we
717 * don't need to do anything else.
718 */
719 if (dentry->d_lockref.count) {
720 spin_unlock(&dentry->d_lock);
721 return 1;
722 }
723
724 /*
725 * Re-get the reference we optimistically dropped. We hold the
726 * lock, and we just tested that it was zero, so we can just
727 * set it to 1.
728 */
729 dentry->d_lockref.count = 1;
730 return 0;
731}
732
733
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734/*
735 * This is dput
736 *
737 * This is complicated by the fact that we do not want to put
738 * dentries that are no longer on any hash chain on the unused
739 * list: we'd much rather just get rid of them immediately.
740 *
741 * However, that implies that we have to traverse the dentry
742 * tree upwards to the parents which might _also_ now be
743 * scheduled for deletion (it may have been only waiting for
744 * its last child to go away).
745 *
746 * This tail recursion is done by hand as we don't want to depend
747 * on the compiler to always get this right (gcc generally doesn't).
748 * Real recursion would eat up our stack space.
749 */
750
751/*
752 * dput - release a dentry
753 * @dentry: dentry to release
754 *
755 * Release a dentry. This will drop the usage count and if appropriate
756 * call the dentry unlink method as well as removing it from the queues and
757 * releasing its resources. If the parent dentries were scheduled for release
758 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760void dput(struct dentry *dentry)
761{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700762 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 return;
764
765repeat:
Linus Torvalds360f5472015-01-09 15:19:03 -0800766 rcu_read_lock();
767 if (likely(fast_dput(dentry))) {
768 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 return;
Linus Torvalds360f5472015-01-09 15:19:03 -0800770 }
771
772 /* Slow case: now with the dentry lock held */
773 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Al Viro85c7f812016-04-14 19:52:13 -0400775 WARN_ON(d_in_lookup(dentry));
776
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700777 /* Unreachable? Get rid of it */
778 if (unlikely(d_unhashed(dentry)))
779 goto kill_it;
780
Al Viro75a6f822015-07-08 02:42:38 +0100781 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
782 goto kill_it;
783
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700784 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100786 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 }
Nick Piggin265ac902010-10-10 05:36:24 -0400788
Linus Torvalds358eec12013-10-31 15:43:02 -0700789 if (!(dentry->d_flags & DCACHE_REFERENCED))
790 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400791 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400792
Waiman Long98474232013-08-28 18:24:59 -0700793 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100794 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 return;
796
Miklos Szeredid52b9082007-05-08 00:23:46 -0700797kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400798 dentry = dentry_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700799 if (dentry)
800 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700802EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100805/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100806static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807{
Waiman Long98474232013-08-28 18:24:59 -0700808 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809}
810
Nick Piggindc0474b2011-01-07 17:49:43 +1100811static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100812{
Waiman Long98474232013-08-28 18:24:59 -0700813 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100814}
815
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100816struct dentry *dget_parent(struct dentry *dentry)
817{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700818 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100819 struct dentry *ret;
820
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700821 /*
822 * Do optimistic parent lookup without any
823 * locking.
824 */
825 rcu_read_lock();
826 ret = ACCESS_ONCE(dentry->d_parent);
827 gotref = lockref_get_not_zero(&ret->d_lockref);
828 rcu_read_unlock();
829 if (likely(gotref)) {
830 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
831 return ret;
832 dput(ret);
833 }
834
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100835repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100836 /*
837 * Don't need rcu_dereference because we re-check it was correct under
838 * the lock.
839 */
840 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100841 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100842 spin_lock(&ret->d_lock);
843 if (unlikely(ret != dentry->d_parent)) {
844 spin_unlock(&ret->d_lock);
845 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100846 goto repeat;
847 }
Nick Piggina734eb42011-01-07 17:49:44 +1100848 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700849 BUG_ON(!ret->d_lockref.count);
850 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100851 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100852 return ret;
853}
854EXPORT_SYMBOL(dget_parent);
855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856/**
857 * d_find_alias - grab a hashed alias of inode
858 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 *
860 * If inode has a hashed alias, or is a directory and has any alias,
861 * acquire the reference to alias and return it. Otherwise return NULL.
862 * Notice that if inode is a directory there can be only one alias and
863 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800864 * of a filesystem, or if the directory was renamed and d_revalidate
865 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700867 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500868 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500870static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871{
Nick Pigginda502952011-01-07 17:49:33 +1100872 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
Nick Pigginda502952011-01-07 17:49:33 +1100874again:
875 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400876 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100877 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700879 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100880 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500882 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100883 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100884 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 return alias;
886 }
887 }
Nick Pigginda502952011-01-07 17:49:33 +1100888 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
Nick Pigginda502952011-01-07 17:49:33 +1100890 if (discon_alias) {
891 alias = discon_alias;
892 spin_lock(&alias->d_lock);
893 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500894 __dget_dlock(alias);
895 spin_unlock(&alias->d_lock);
896 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100897 }
898 spin_unlock(&alias->d_lock);
899 goto again;
900 }
901 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902}
903
Nick Pigginda502952011-01-07 17:49:33 +1100904struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905{
David Howells214fda12006-03-25 03:06:36 -0800906 struct dentry *de = NULL;
907
Al Virob3d9b7a2012-06-09 13:51:19 -0400908 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100909 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500910 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100911 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800912 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 return de;
914}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700915EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
917/*
918 * Try to kill dentries associated with this inode.
919 * WARNING: you must own a reference to inode.
920 */
921void d_prune_aliases(struct inode *inode)
922{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700923 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100925 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400926 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700928 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400929 struct dentry *parent = lock_parent(dentry);
930 if (likely(!dentry->d_lockref.count)) {
931 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800932 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400933 goto restart;
934 }
935 if (parent)
936 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 }
938 spin_unlock(&dentry->d_lock);
939 }
Nick Piggin873feea2011-01-07 17:50:06 +1100940 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700942EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400944static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945{
Al Viro5c47e6d2014-04-29 16:13:18 -0400946 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700947
Miklos Szeredi60942f22014-05-02 15:38:39 -0400948 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400949 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400950 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100951 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400952 parent = lock_parent(dentry);
953
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000955 * The dispose list is isolated and dentries are not accounted
956 * to the LRU here, so we can simply remove it from the list
957 * here regardless of whether it is referenced or not.
958 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400959 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000960
961 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000963 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800965 if (dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700966 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400967 if (parent)
968 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 continue;
970 }
Nick Piggin77812a12011-01-07 17:49:48 +1100971
Al Viro64fd72e2014-05-28 09:48:44 -0400972
973 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
974 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
975 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400976 if (parent)
977 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -0400978 if (can_free)
979 dentry_free(dentry);
980 continue;
981 }
982
Al Viroff2fde92014-05-28 13:59:13 -0400983 inode = dentry->d_inode;
984 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400985 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000986 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400987 if (parent)
988 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400989 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000990 }
Al Viroff2fde92014-05-28 13:59:13 -0400991
Al Viroff2fde92014-05-28 13:59:13 -0400992 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -0400993
Al Viro5c47e6d2014-04-29 16:13:18 -0400994 /*
995 * We need to prune ancestors too. This is necessary to prevent
996 * quadratic behavior of shrink_dcache_parent(), but is also
997 * expected to be beneficial in reducing dentry cache
998 * fragmentation.
999 */
1000 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -04001001 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1002 parent = lock_parent(dentry);
1003 if (dentry->d_lockref.count != 1) {
1004 dentry->d_lockref.count--;
1005 spin_unlock(&dentry->d_lock);
1006 if (parent)
1007 spin_unlock(&parent->d_lock);
1008 break;
1009 }
1010 inode = dentry->d_inode; /* can't be NULL */
1011 if (unlikely(!spin_trylock(&inode->i_lock))) {
1012 spin_unlock(&dentry->d_lock);
1013 if (parent)
1014 spin_unlock(&parent->d_lock);
1015 cpu_relax();
1016 continue;
1017 }
1018 __dentry_kill(dentry);
1019 dentry = parent;
1020 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001022}
1023
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001024static enum lru_status dentry_lru_isolate(struct list_head *item,
1025 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +10001026{
1027 struct list_head *freeable = arg;
1028 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1029
1030
1031 /*
1032 * we are inverting the lru lock/dentry->d_lock here,
1033 * so use a trylock. If we fail to get the lock, just skip
1034 * it
1035 */
1036 if (!spin_trylock(&dentry->d_lock))
1037 return LRU_SKIP;
1038
1039 /*
1040 * Referenced dentries are still in use. If they have active
1041 * counts, just remove them from the LRU. Otherwise give them
1042 * another pass through the LRU.
1043 */
1044 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001045 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001046 spin_unlock(&dentry->d_lock);
1047 return LRU_REMOVED;
1048 }
1049
1050 if (dentry->d_flags & DCACHE_REFERENCED) {
1051 dentry->d_flags &= ~DCACHE_REFERENCED;
1052 spin_unlock(&dentry->d_lock);
1053
1054 /*
1055 * The list move itself will be made by the common LRU code. At
1056 * this point, we've dropped the dentry->d_lock but keep the
1057 * lru lock. This is safe to do, since every list movement is
1058 * protected by the lru lock even if both locks are held.
1059 *
1060 * This is guaranteed by the fact that all LRU management
1061 * functions are intermediated by the LRU API calls like
1062 * list_lru_add and list_lru_del. List movement in this file
1063 * only ever occur through this functions or through callbacks
1064 * like this one, that are called from the LRU API.
1065 *
1066 * The only exceptions to this are functions like
1067 * shrink_dentry_list, and code that first checks for the
1068 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1069 * operating only with stack provided lists after they are
1070 * properly isolated from the main list. It is thus, always a
1071 * local access.
1072 */
1073 return LRU_ROTATE;
1074 }
1075
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001076 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001077 spin_unlock(&dentry->d_lock);
1078
1079 return LRU_REMOVED;
1080}
1081
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001082/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001083 * prune_dcache_sb - shrink the dcache
1084 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001085 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001086 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001087 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1088 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001089 * function.
1090 *
1091 * This function may fail to free any resources if all the dentries are in
1092 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001093 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001094long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001095{
Dave Chinnerf6041562013-08-28 10:18:00 +10001096 LIST_HEAD(dispose);
1097 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001098
Vladimir Davydov503c3582015-02-12 14:58:47 -08001099 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1100 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001101 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001102 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103}
1104
Glauber Costa4e717f52013-08-28 10:18:03 +10001105static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001106 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001107{
Glauber Costa4e717f52013-08-28 10:18:03 +10001108 struct list_head *freeable = arg;
1109 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001110
Glauber Costa4e717f52013-08-28 10:18:03 +10001111 /*
1112 * we are inverting the lru lock/dentry->d_lock here,
1113 * so use a trylock. If we fail to get the lock, just skip
1114 * it
1115 */
1116 if (!spin_trylock(&dentry->d_lock))
1117 return LRU_SKIP;
1118
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001119 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001120 spin_unlock(&dentry->d_lock);
1121
1122 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001123}
1124
Glauber Costa4e717f52013-08-28 10:18:03 +10001125
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001126/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 * shrink_dcache_sb - shrink dcache for a superblock
1128 * @sb: superblock
1129 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001130 * Shrink the dcache for the specified super block. This is used to free
1131 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001133void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134{
Glauber Costa4e717f52013-08-28 10:18:03 +10001135 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001136
Glauber Costa4e717f52013-08-28 10:18:03 +10001137 do {
1138 LIST_HEAD(dispose);
1139
1140 freed = list_lru_walk(&sb->s_dentry_lru,
1141 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1142
1143 this_cpu_sub(nr_dentry_unused, freed);
1144 shrink_dentry_list(&dispose);
1145 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001147EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001150 * enum d_walk_ret - action to talke during tree walk
1151 * @D_WALK_CONTINUE: contrinue walk
1152 * @D_WALK_QUIT: quit walk
1153 * @D_WALK_NORETRY: quit when retry is needed
1154 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001156enum d_walk_ret {
1157 D_WALK_CONTINUE,
1158 D_WALK_QUIT,
1159 D_WALK_NORETRY,
1160 D_WALK_SKIP,
1161};
1162
1163/**
1164 * d_walk - walk the dentry tree
1165 * @parent: start of walk
1166 * @data: data passed to @enter() and @finish()
1167 * @enter: callback when first entering the dentry
1168 * @finish: callback when successfully finished the walk
1169 *
1170 * The @enter() and @finish() callbacks are called with d_lock held.
1171 */
1172static void d_walk(struct dentry *parent, void *data,
1173 enum d_walk_ret (*enter)(void *, struct dentry *),
1174 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175{
Nick Piggin949854d2011-01-07 17:49:37 +11001176 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001178 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001179 enum d_walk_ret ret;
1180 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001181
Nick Piggin58db63d2011-01-07 17:49:39 +11001182again:
Al Viro48f5ec22013-09-09 15:22:25 -04001183 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001184 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001185 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001186
1187 ret = enter(data, this_parent);
1188 switch (ret) {
1189 case D_WALK_CONTINUE:
1190 break;
1191 case D_WALK_QUIT:
1192 case D_WALK_SKIP:
1193 goto out_unlock;
1194 case D_WALK_NORETRY:
1195 retry = false;
1196 break;
1197 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198repeat:
1199 next = this_parent->d_subdirs.next;
1200resume:
1201 while (next != &this_parent->d_subdirs) {
1202 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001203 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001205
1206 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001207
1208 ret = enter(data, dentry);
1209 switch (ret) {
1210 case D_WALK_CONTINUE:
1211 break;
1212 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001213 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001214 goto out_unlock;
1215 case D_WALK_NORETRY:
1216 retry = false;
1217 break;
1218 case D_WALK_SKIP:
1219 spin_unlock(&dentry->d_lock);
1220 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001221 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001222
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001224 spin_unlock(&this_parent->d_lock);
1225 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001227 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 goto repeat;
1229 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001230 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 }
1232 /*
1233 * All done at this level ... ascend and resume the search.
1234 */
Al Viroca5358e2014-10-26 19:31:10 -04001235 rcu_read_lock();
1236ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001238 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001239 this_parent = child->d_parent;
1240
Al Viro31dec132013-10-25 17:04:27 -04001241 spin_unlock(&child->d_lock);
1242 spin_lock(&this_parent->d_lock);
1243
Al Viroca5358e2014-10-26 19:31:10 -04001244 /* might go back up the wrong parent if we have had a rename. */
1245 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001246 goto rename_retry;
Al Viro21591842015-05-28 23:09:19 -04001247 /* go into the first sibling still alive */
1248 do {
1249 next = child->d_child.next;
Al Viroca5358e2014-10-26 19:31:10 -04001250 if (next == &this_parent->d_subdirs)
1251 goto ascend;
1252 child = list_entry(next, struct dentry, d_child);
Al Viro21591842015-05-28 23:09:19 -04001253 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
Al Viro31dec132013-10-25 17:04:27 -04001254 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 goto resume;
1256 }
Al Viroca5358e2014-10-26 19:31:10 -04001257 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001258 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001259 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001260 if (finish)
1261 finish(data);
1262
1263out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001264 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001265 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001266 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001267
1268rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001269 spin_unlock(&this_parent->d_lock);
1270 rcu_read_unlock();
1271 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001272 if (!retry)
1273 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001274 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001275 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001277
1278/*
1279 * Search for at least 1 mount point in the dentry's subdirs.
1280 * We descend to the next level whenever the d_subdirs
1281 * list is non-empty and continue searching.
1282 */
1283
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001284static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1285{
1286 int *ret = data;
1287 if (d_mountpoint(dentry)) {
1288 *ret = 1;
1289 return D_WALK_QUIT;
1290 }
1291 return D_WALK_CONTINUE;
1292}
1293
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001294/**
1295 * have_submounts - check for mounts over a dentry
1296 * @parent: dentry to check.
1297 *
1298 * Return true if the parent or its subdirectories contain
1299 * a mount point
1300 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001301int have_submounts(struct dentry *parent)
1302{
1303 int ret = 0;
1304
1305 d_walk(parent, &ret, check_mount, NULL);
1306
1307 return ret;
1308}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001309EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
1311/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001312 * Called by mount code to set a mountpoint and check if the mountpoint is
1313 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1314 * subtree can become unreachable).
1315 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001316 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001317 * this reason take rename_lock and d_lock on dentry and ancestors.
1318 */
1319int d_set_mounted(struct dentry *dentry)
1320{
1321 struct dentry *p;
1322 int ret = -ENOENT;
1323 write_seqlock(&rename_lock);
1324 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001325 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001326 spin_lock(&p->d_lock);
1327 if (unlikely(d_unhashed(p))) {
1328 spin_unlock(&p->d_lock);
1329 goto out;
1330 }
1331 spin_unlock(&p->d_lock);
1332 }
1333 spin_lock(&dentry->d_lock);
1334 if (!d_unlinked(dentry)) {
1335 dentry->d_flags |= DCACHE_MOUNTED;
1336 ret = 0;
1337 }
1338 spin_unlock(&dentry->d_lock);
1339out:
1340 write_sequnlock(&rename_lock);
1341 return ret;
1342}
1343
1344/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001345 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 * and move any unused dentries to the end of the unused
1347 * list for prune_dcache(). We descend to the next level
1348 * whenever the d_subdirs list is non-empty and continue
1349 * searching.
1350 *
1351 * It returns zero iff there are no unused children,
1352 * otherwise it returns the number of children moved to
1353 * the end of the unused list. This may not be the total
1354 * number of unused children, because select_parent can
1355 * drop the lock and return early due to latency
1356 * constraints.
1357 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001358
1359struct select_data {
1360 struct dentry *start;
1361 struct list_head dispose;
1362 int found;
1363};
1364
1365static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001367 struct select_data *data = _data;
1368 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001370 if (data->start == dentry)
1371 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
Al Virofe915222014-05-03 00:02:25 -04001373 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001374 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001375 } else {
1376 if (dentry->d_flags & DCACHE_LRU_LIST)
1377 d_lru_del(dentry);
1378 if (!dentry->d_lockref.count) {
1379 d_shrink_add(dentry, &data->dispose);
1380 data->found++;
1381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 }
1383 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001384 * We can return to the caller if we have found some (this
1385 * ensures forward progress). We'll be coming back to find
1386 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 */
Al Virofe915222014-05-03 00:02:25 -04001388 if (!list_empty(&data->dispose))
1389 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001391 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392}
1393
1394/**
1395 * shrink_dcache_parent - prune dcache
1396 * @parent: parent of entries to prune
1397 *
1398 * Prune the dcache to remove unused children of the parent dentry.
1399 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001400void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001402 for (;;) {
1403 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001405 INIT_LIST_HEAD(&data.dispose);
1406 data.start = parent;
1407 data.found = 0;
1408
1409 d_walk(parent, &data, select_collect, NULL);
1410 if (!data.found)
1411 break;
1412
1413 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001414 cond_resched();
1415 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001417EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Al Viro9c8c10e2014-05-02 20:36:10 -04001419static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001420{
Al Viro9c8c10e2014-05-02 20:36:10 -04001421 /* it has busy descendents; complain about those instead */
1422 if (!list_empty(&dentry->d_subdirs))
1423 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001424
Al Viro9c8c10e2014-05-02 20:36:10 -04001425 /* root with refcount 1 is fine */
1426 if (dentry == _data && dentry->d_lockref.count == 1)
1427 return D_WALK_CONTINUE;
1428
1429 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1430 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001431 dentry,
1432 dentry->d_inode ?
1433 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001434 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001435 dentry->d_lockref.count,
1436 dentry->d_sb->s_type->name,
1437 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001438 WARN_ON(1);
1439 return D_WALK_CONTINUE;
1440}
1441
1442static void do_one_tree(struct dentry *dentry)
1443{
1444 shrink_dcache_parent(dentry);
1445 d_walk(dentry, dentry, umount_check, NULL);
1446 d_drop(dentry);
1447 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001448}
1449
1450/*
1451 * destroy the dentries attached to a superblock on unmounting
1452 */
1453void shrink_dcache_for_umount(struct super_block *sb)
1454{
1455 struct dentry *dentry;
1456
Al Viro9c8c10e2014-05-02 20:36:10 -04001457 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001458
1459 dentry = sb->s_root;
1460 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001461 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001462
1463 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001464 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1465 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001466 }
1467}
1468
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001469struct detach_data {
1470 struct select_data select;
1471 struct dentry *mountpoint;
1472};
1473static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001474{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001475 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001476
1477 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001478 __dget_dlock(dentry);
1479 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001480 return D_WALK_QUIT;
1481 }
1482
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001483 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001484}
1485
1486static void check_and_drop(void *_data)
1487{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001488 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001489
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001490 if (!data->mountpoint && !data->select.found)
1491 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001492}
1493
1494/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001495 * d_invalidate - detach submounts, prune dcache, and drop
1496 * @dentry: dentry to invalidate (aka detach, prune and drop)
1497 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001498 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001499 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001500 * The final d_drop is done as an atomic operation relative to
1501 * rename_lock ensuring there are no races with d_set_mounted. This
1502 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001503 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001504void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001505{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001506 /*
1507 * If it's already been dropped, return OK.
1508 */
1509 spin_lock(&dentry->d_lock);
1510 if (d_unhashed(dentry)) {
1511 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001512 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001513 }
1514 spin_unlock(&dentry->d_lock);
1515
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001516 /* Negative dentries can be dropped without further checks */
1517 if (!dentry->d_inode) {
1518 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001519 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001520 }
1521
1522 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001523 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001524
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001525 data.mountpoint = NULL;
1526 INIT_LIST_HEAD(&data.select.dispose);
1527 data.select.start = dentry;
1528 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001529
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001530 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001531
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001532 if (data.select.found)
1533 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001534
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001535 if (data.mountpoint) {
1536 detach_mounts(data.mountpoint);
1537 dput(data.mountpoint);
1538 }
1539
1540 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001541 break;
1542
1543 cond_resched();
1544 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001545}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001546EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001547
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548/**
Al Viroa4464db2011-07-07 15:03:58 -04001549 * __d_alloc - allocate a dcache entry
1550 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 * @name: qstr of the name
1552 *
1553 * Allocates a dentry. It returns %NULL if there is insufficient memory
1554 * available. On a success the dentry is returned. The name passed in is
1555 * copied and the copy passed in may be reused after this call.
1556 */
1557
Al Viroa4464db2011-07-07 15:03:58 -04001558struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559{
1560 struct dentry *dentry;
1561 char *dname;
1562
Mel Gormane12ba742007-10-16 01:25:52 -07001563 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 if (!dentry)
1565 return NULL;
1566
Linus Torvalds6326c712012-05-21 16:14:04 -07001567 /*
1568 * We guarantee that the inline name is always NUL-terminated.
1569 * This way the memcpy() done by the name switching in rename
1570 * will still always have a NUL at the end, even if we might
1571 * be overwriting an internal NUL character
1572 */
1573 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Al Viro798434b2016-03-24 20:38:43 -04001574 if (unlikely(!name)) {
1575 static const struct qstr anon = QSTR_INIT("/", 1);
1576 name = &anon;
1577 dname = dentry->d_iname;
1578 } else if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001579 size_t size = offsetof(struct external_name, name[1]);
Vladimir Davydov5d097052016-01-14 15:18:21 -08001580 struct external_name *p = kmalloc(size + name->len,
1581 GFP_KERNEL_ACCOUNT);
Al Viro8d85b482014-09-29 14:54:27 -04001582 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 kmem_cache_free(dentry_cache, dentry);
1584 return NULL;
1585 }
Al Viro8d85b482014-09-29 14:54:27 -04001586 atomic_set(&p->u.count, 1);
1587 dname = p->name;
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -08001588 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1589 kasan_unpoison_shadow(dname,
1590 round_up(name->len + 1, sizeof(unsigned long)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 } else {
1592 dname = dentry->d_iname;
1593 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
1595 dentry->d_name.len = name->len;
1596 dentry->d_name.hash = name->hash;
1597 memcpy(dname, name->name, name->len);
1598 dname[name->len] = 0;
1599
Linus Torvalds6326c712012-05-21 16:14:04 -07001600 /* Make sure we always see the terminating NUL character */
1601 smp_wmb();
1602 dentry->d_name.name = dname;
1603
Waiman Long98474232013-08-28 18:24:59 -07001604 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001605 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001607 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001609 dentry->d_parent = dentry;
1610 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 dentry->d_op = NULL;
1612 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001613 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 INIT_LIST_HEAD(&dentry->d_lru);
1615 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001616 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1617 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001618 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
Nick Piggin3e880fb2011-01-07 17:49:19 +11001620 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001621
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 return dentry;
1623}
Al Viroa4464db2011-07-07 15:03:58 -04001624
1625/**
1626 * d_alloc - allocate a dcache entry
1627 * @parent: parent of entry to allocate
1628 * @name: qstr of the name
1629 *
1630 * Allocates a dentry. It returns %NULL if there is insufficient memory
1631 * available. On a success the dentry is returned. The name passed in is
1632 * copied and the copy passed in may be reused after this call.
1633 */
1634struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1635{
1636 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1637 if (!dentry)
1638 return NULL;
1639
1640 spin_lock(&parent->d_lock);
1641 /*
1642 * don't need child lock because it is not subject
1643 * to concurrency here
1644 */
1645 __dget_dlock(parent);
1646 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001647 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001648 spin_unlock(&parent->d_lock);
1649
1650 return dentry;
1651}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001652EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001654/**
1655 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1656 * @sb: the superblock
1657 * @name: qstr of the name
1658 *
1659 * For a filesystem that just pins its dentries in memory and never
1660 * performs lookups at all, return an unhashed IS_ROOT dentry.
1661 */
Nick Piggin4b936882011-01-07 17:50:07 +11001662struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1663{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001664 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001665}
1666EXPORT_SYMBOL(d_alloc_pseudo);
1667
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1669{
1670 struct qstr q;
1671
1672 q.name = name;
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001673 q.hash_len = hashlen_string(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 return d_alloc(parent, &q);
1675}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001676EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677
Nick Pigginfb045ad2011-01-07 17:49:55 +11001678void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1679{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001680 WARN_ON_ONCE(dentry->d_op);
1681 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001682 DCACHE_OP_COMPARE |
1683 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001684 DCACHE_OP_WEAK_REVALIDATE |
David Howells4bacc9c2015-06-18 14:32:31 +01001685 DCACHE_OP_DELETE |
Miklos Szeredid101a122016-03-26 16:14:37 -04001686 DCACHE_OP_SELECT_INODE |
1687 DCACHE_OP_REAL));
Nick Pigginfb045ad2011-01-07 17:49:55 +11001688 dentry->d_op = op;
1689 if (!op)
1690 return;
1691 if (op->d_hash)
1692 dentry->d_flags |= DCACHE_OP_HASH;
1693 if (op->d_compare)
1694 dentry->d_flags |= DCACHE_OP_COMPARE;
1695 if (op->d_revalidate)
1696 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001697 if (op->d_weak_revalidate)
1698 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001699 if (op->d_delete)
1700 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001701 if (op->d_prune)
1702 dentry->d_flags |= DCACHE_OP_PRUNE;
David Howells4bacc9c2015-06-18 14:32:31 +01001703 if (op->d_select_inode)
1704 dentry->d_flags |= DCACHE_OP_SELECT_INODE;
Miklos Szeredid101a122016-03-26 16:14:37 -04001705 if (op->d_real)
1706 dentry->d_flags |= DCACHE_OP_REAL;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001707
1708}
1709EXPORT_SYMBOL(d_set_d_op);
1710
David Howellsdf1a0852015-01-29 12:02:28 +00001711
1712/*
1713 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1714 * @dentry - The dentry to mark
1715 *
1716 * Mark a dentry as falling through to the lower layer (as set with
1717 * d_pin_lower()). This flag may be recorded on the medium.
1718 */
1719void d_set_fallthru(struct dentry *dentry)
1720{
1721 spin_lock(&dentry->d_lock);
1722 dentry->d_flags |= DCACHE_FALLTHRU;
1723 spin_unlock(&dentry->d_lock);
1724}
1725EXPORT_SYMBOL(d_set_fallthru);
1726
David Howellsb18825a2013-09-12 19:22:53 +01001727static unsigned d_flags_for_inode(struct inode *inode)
1728{
David Howells44bdb5e2015-01-29 12:02:29 +00001729 unsigned add_flags = DCACHE_REGULAR_TYPE;
David Howellsb18825a2013-09-12 19:22:53 +01001730
1731 if (!inode)
1732 return DCACHE_MISS_TYPE;
1733
1734 if (S_ISDIR(inode->i_mode)) {
1735 add_flags = DCACHE_DIRECTORY_TYPE;
1736 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1737 if (unlikely(!inode->i_op->lookup))
1738 add_flags = DCACHE_AUTODIR_TYPE;
1739 else
1740 inode->i_opflags |= IOP_LOOKUP;
1741 }
David Howells44bdb5e2015-01-29 12:02:29 +00001742 goto type_determined;
David Howellsb18825a2013-09-12 19:22:53 +01001743 }
1744
David Howells44bdb5e2015-01-29 12:02:29 +00001745 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
Al Viro6b255392015-11-17 10:20:54 -05001746 if (unlikely(inode->i_op->get_link)) {
David Howells44bdb5e2015-01-29 12:02:29 +00001747 add_flags = DCACHE_SYMLINK_TYPE;
1748 goto type_determined;
1749 }
1750 inode->i_opflags |= IOP_NOFOLLOW;
1751 }
1752
1753 if (unlikely(!S_ISREG(inode->i_mode)))
1754 add_flags = DCACHE_SPECIAL_TYPE;
1755
1756type_determined:
David Howellsb18825a2013-09-12 19:22:53 +01001757 if (unlikely(IS_AUTOMOUNT(inode)))
1758 add_flags |= DCACHE_NEED_AUTOMOUNT;
1759 return add_flags;
1760}
1761
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001762static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1763{
David Howellsb18825a2013-09-12 19:22:53 +01001764 unsigned add_flags = d_flags_for_inode(inode);
Al Viro85c7f812016-04-14 19:52:13 -04001765 WARN_ON(d_in_lookup(dentry));
David Howellsb18825a2013-09-12 19:22:53 +01001766
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001767 spin_lock(&dentry->d_lock);
Al Virode689f52016-03-09 18:05:42 -05001768 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
Al Viroa528aca2016-02-29 12:12:46 -05001769 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +00001770 __d_set_inode_and_type(dentry, inode, add_flags);
Al Viroa528aca2016-02-29 12:12:46 -05001771 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04001772 fsnotify_update_flags(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001773 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001774}
1775
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776/**
1777 * d_instantiate - fill in inode information for a dentry
1778 * @entry: dentry to complete
1779 * @inode: inode to attach to this dentry
1780 *
1781 * Fill in inode information in the entry.
1782 *
1783 * This turns negative dentries into productive full members
1784 * of society.
1785 *
1786 * NOTE! This assumes that the inode count has been incremented
1787 * (or otherwise set) by the caller to indicate that it is now
1788 * in use by the dcache.
1789 */
1790
1791void d_instantiate(struct dentry *entry, struct inode * inode)
1792{
Al Viro946e51f2014-10-26 19:19:16 -04001793 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Al Virode689f52016-03-09 18:05:42 -05001794 if (inode) {
Al Virob9680912016-04-11 00:53:26 -04001795 security_d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001796 spin_lock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001797 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001798 spin_unlock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001799 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001801EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
1803/**
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001804 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1805 * @entry: dentry to complete
1806 * @inode: inode to attach to this dentry
1807 *
1808 * Fill in inode information in the entry. If a directory alias is found, then
1809 * return an error (and drop inode). Together with d_materialise_unique() this
1810 * guarantees that a directory inode may never have more than one alias.
1811 */
1812int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1813{
Al Viro946e51f2014-10-26 19:19:16 -04001814 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001815
Al Virob9680912016-04-11 00:53:26 -04001816 security_d_instantiate(entry, inode);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001817 spin_lock(&inode->i_lock);
1818 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1819 spin_unlock(&inode->i_lock);
1820 iput(inode);
1821 return -EBUSY;
1822 }
1823 __d_instantiate(entry, inode);
1824 spin_unlock(&inode->i_lock);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001825
1826 return 0;
1827}
1828EXPORT_SYMBOL(d_instantiate_no_diralias);
1829
Al Viroadc0e912012-01-08 16:49:21 -05001830struct dentry *d_make_root(struct inode *root_inode)
1831{
1832 struct dentry *res = NULL;
1833
1834 if (root_inode) {
Al Viro798434b2016-03-24 20:38:43 -04001835 res = __d_alloc(root_inode->i_sb, NULL);
Al Viroadc0e912012-01-08 16:49:21 -05001836 if (res)
1837 d_instantiate(res, root_inode);
1838 else
1839 iput(root_inode);
1840 }
1841 return res;
1842}
1843EXPORT_SYMBOL(d_make_root);
1844
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001845static struct dentry * __d_find_any_alias(struct inode *inode)
1846{
1847 struct dentry *alias;
1848
Al Virob3d9b7a2012-06-09 13:51:19 -04001849 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001850 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001851 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001852 __dget(alias);
1853 return alias;
1854}
1855
Sage Weil46f72b32012-01-10 09:04:37 -08001856/**
1857 * d_find_any_alias - find any alias for a given inode
1858 * @inode: inode to find an alias for
1859 *
1860 * If any aliases exist for the given inode, take and return a
1861 * reference for one of them. If no aliases exist, return %NULL.
1862 */
1863struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001864{
1865 struct dentry *de;
1866
1867 spin_lock(&inode->i_lock);
1868 de = __d_find_any_alias(inode);
1869 spin_unlock(&inode->i_lock);
1870 return de;
1871}
Sage Weil46f72b32012-01-10 09:04:37 -08001872EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001873
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001874static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001875{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001876 struct dentry *tmp;
1877 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001878 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001879
1880 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001881 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001882 if (IS_ERR(inode))
1883 return ERR_CAST(inode);
1884
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001885 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001886 if (res)
1887 goto out_iput;
1888
Al Viro798434b2016-03-24 20:38:43 -04001889 tmp = __d_alloc(inode->i_sb, NULL);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001890 if (!tmp) {
1891 res = ERR_PTR(-ENOMEM);
1892 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001893 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001894
Al Virob9680912016-04-11 00:53:26 -04001895 security_d_instantiate(tmp, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001896 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001897 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001898 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001899 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001900 dput(tmp);
1901 goto out_iput;
1902 }
1903
1904 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001905 add_flags = d_flags_for_inode(inode);
1906
1907 if (disconnected)
1908 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001909
Christoph Hellwig9308a612008-08-11 15:49:12 +02001910 spin_lock(&tmp->d_lock);
David Howells4bf46a22015-03-05 14:09:22 +00001911 __d_set_inode_and_type(tmp, inode, add_flags);
Al Viro946e51f2014-10-26 19:19:16 -04001912 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001913 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001914 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001915 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001916 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001917 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001918
Christoph Hellwig9308a612008-08-11 15:49:12 +02001919 return tmp;
1920
1921 out_iput:
1922 iput(inode);
1923 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001924}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001925
1926/**
1927 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1928 * @inode: inode to allocate the dentry for
1929 *
1930 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1931 * similar open by handle operations. The returned dentry may be anonymous,
1932 * or may have a full name (if the inode was already in the cache).
1933 *
1934 * When called on a directory inode, we must ensure that the inode only ever
1935 * has one dentry. If a dentry is found, that is returned instead of
1936 * allocating a new one.
1937 *
1938 * On successful return, the reference to the inode has been transferred
1939 * to the dentry. In case of an error the reference on the inode is released.
1940 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1941 * be passed in and the error will be propagated to the return value,
1942 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1943 */
1944struct dentry *d_obtain_alias(struct inode *inode)
1945{
1946 return __d_obtain_alias(inode, 1);
1947}
Benny Halevyadc48722009-02-27 14:02:59 -08001948EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
1950/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001951 * d_obtain_root - find or allocate a dentry for a given inode
1952 * @inode: inode to allocate the dentry for
1953 *
1954 * Obtain an IS_ROOT dentry for the root of a filesystem.
1955 *
1956 * We must ensure that directory inodes only ever have one dentry. If a
1957 * dentry is found, that is returned instead of allocating a new one.
1958 *
1959 * On successful return, the reference to the inode has been transferred
1960 * to the dentry. In case of an error the reference on the inode is
1961 * released. A %NULL or IS_ERR inode may be passed in and will be the
1962 * error will be propagate to the return value, with a %NULL @inode
1963 * replaced by ERR_PTR(-ESTALE).
1964 */
1965struct dentry *d_obtain_root(struct inode *inode)
1966{
1967 return __d_obtain_alias(inode, 0);
1968}
1969EXPORT_SYMBOL(d_obtain_root);
1970
1971/**
Barry Naujok94035402008-05-21 16:50:46 +10001972 * d_add_ci - lookup or allocate new dentry with case-exact name
1973 * @inode: the inode case-insensitive lookup has found
1974 * @dentry: the negative dentry that was passed to the parent's lookup func
1975 * @name: the case-exact name to be associated with the returned dentry
1976 *
1977 * This is to avoid filling the dcache with case-insensitive names to the
1978 * same inode, only the actual correct case is stored in the dcache for
1979 * case-insensitive filesystems.
1980 *
1981 * For a case-insensitive lookup match and if the the case-exact dentry
1982 * already exists in in the dcache, use it and return it.
1983 *
1984 * If no entry exists with the exact case name, allocate new dentry with
1985 * the exact case, and return the spliced entry.
1986 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001987struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001988 struct qstr *name)
1989{
Al Virod9171b92016-04-15 03:33:13 -04001990 struct dentry *found, *res;
Barry Naujok94035402008-05-21 16:50:46 +10001991
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001992 /*
1993 * First check if a dentry matching the name already exists,
1994 * if not go ahead and create it now.
1995 */
Barry Naujok94035402008-05-21 16:50:46 +10001996 found = d_hash_and_lookup(dentry->d_parent, name);
Al Virod9171b92016-04-15 03:33:13 -04001997 if (found) {
1998 iput(inode);
1999 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002000 }
Al Virod9171b92016-04-15 03:33:13 -04002001 if (d_in_lookup(dentry)) {
2002 found = d_alloc_parallel(dentry->d_parent, name,
2003 dentry->d_wait);
2004 if (IS_ERR(found) || !d_in_lookup(found)) {
2005 iput(inode);
2006 return found;
2007 }
2008 } else {
2009 found = d_alloc(dentry->d_parent, name);
2010 if (!found) {
2011 iput(inode);
2012 return ERR_PTR(-ENOMEM);
2013 }
2014 }
2015 res = d_splice_alias(inode, found);
2016 if (res) {
2017 dput(found);
2018 return res;
2019 }
Al Viro4f522a22013-02-11 23:20:37 -05002020 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002021}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002022EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002024/*
2025 * Do the slow-case of the dentry name compare.
2026 *
2027 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07002028 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002029 * filesystem can rely on them, and can use the 'name' and
2030 * 'len' information without worrying about walking off the
2031 * end of memory etc.
2032 *
2033 * Thus the read_seqcount_retry() and the "duplicate" info
2034 * in arguments (the low-level filesystem should not look
2035 * at the dentry inode or name contents directly, since
2036 * rename can change them while we're in RCU mode).
2037 */
2038enum slow_d_compare {
2039 D_COMP_OK,
2040 D_COMP_NOMATCH,
2041 D_COMP_SEQRETRY,
2042};
2043
2044static noinline enum slow_d_compare slow_dentry_cmp(
2045 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002046 struct dentry *dentry,
2047 unsigned int seq,
2048 const struct qstr *name)
2049{
2050 int tlen = dentry->d_name.len;
2051 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002052
2053 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2054 cpu_relax();
2055 return D_COMP_SEQRETRY;
2056 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002057 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002058 return D_COMP_NOMATCH;
2059 return D_COMP_OK;
2060}
2061
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002063 * __d_lookup_rcu - search for a dentry (racy, store-free)
2064 * @parent: parent dentry
2065 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002066 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002067 * Returns: dentry, or NULL
2068 *
2069 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2070 * resolution (store-free path walking) design described in
2071 * Documentation/filesystems/path-lookup.txt.
2072 *
2073 * This is not to be used outside core vfs.
2074 *
2075 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2076 * held, and rcu_read_lock held. The returned dentry must not be stored into
2077 * without taking d_lock and checking d_seq sequence count against @seq
2078 * returned here.
2079 *
Linus Torvalds15570082013-09-02 11:38:06 -07002080 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002081 * function.
2082 *
2083 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2084 * the returned dentry, so long as its parent's seqlock is checked after the
2085 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2086 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002087 *
2088 * NOTE! The caller *has* to check the resulting dentry against the sequence
2089 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002090 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002091struct dentry *__d_lookup_rcu(const struct dentry *parent,
2092 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002093 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002094{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002095 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002096 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002097 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002098 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002099 struct dentry *dentry;
2100
2101 /*
2102 * Note: There is significant duplication with __d_lookup_rcu which is
2103 * required to prevent single threaded performance regressions
2104 * especially on architectures where smp_rmb (in seqcounts) are costly.
2105 * Keep the two functions in sync.
2106 */
2107
2108 /*
2109 * The hash list is protected using RCU.
2110 *
2111 * Carefully use d_seq when comparing a candidate dentry, to avoid
2112 * races with d_move().
2113 *
2114 * It is possible that concurrent renames can mess up our list
2115 * walk here and result in missing our dentry, resulting in the
2116 * false-negative result. d_lookup() protects against concurrent
2117 * renames using rename_lock seqlock.
2118 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002119 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002120 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002121 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002122 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002123
Nick Piggin31e6b012011-01-07 17:49:52 +11002124seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002125 /*
2126 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002127 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002128 *
2129 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002130 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002131 *
2132 * NOTE! We do a "raw" seqcount_begin here. That means that
2133 * we don't wait for the sequence count to stabilize if it
2134 * is in the middle of a sequence change. If we do the slow
2135 * dentry compare, we will do seqretries until it is stable,
2136 * and if we end up with a successful lookup, we actually
2137 * want to exit RCU lookup anyway.
2138 */
2139 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002140 if (dentry->d_parent != parent)
2141 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002142 if (d_unhashed(dentry))
2143 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002144
2145 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002146 if (dentry->d_name.hash != hashlen_hash(hashlen))
2147 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002148 *seqp = seq;
2149 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002150 case D_COMP_OK:
2151 return dentry;
2152 case D_COMP_NOMATCH:
2153 continue;
2154 default:
2155 goto seqretry;
2156 }
2157 }
2158
Linus Torvalds26fe5752012-05-10 13:14:12 -07002159 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002160 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002161 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002162 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002163 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002164 }
2165 return NULL;
2166}
2167
2168/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 * d_lookup - search for a dentry
2170 * @parent: parent dentry
2171 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002172 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002174 * d_lookup searches the children of the parent dentry for the name in
2175 * question. If the dentry is found its reference count is incremented and the
2176 * dentry is returned. The caller must use dput to free the entry when it has
2177 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 */
Al Viroda2d8452013-01-24 18:29:34 -05002179struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180{
Nick Piggin31e6b012011-01-07 17:49:52 +11002181 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002182 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183
Daeseok Younb8314f92014-08-11 11:46:53 +09002184 do {
2185 seq = read_seqbegin(&rename_lock);
2186 dentry = __d_lookup(parent, name);
2187 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 break;
2189 } while (read_seqretry(&rename_lock, seq));
2190 return dentry;
2191}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002192EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193
Nick Piggin31e6b012011-01-07 17:49:52 +11002194/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002195 * __d_lookup - search for a dentry (racy)
2196 * @parent: parent dentry
2197 * @name: qstr of name we wish to find
2198 * Returns: dentry, or NULL
2199 *
2200 * __d_lookup is like d_lookup, however it may (rarely) return a
2201 * false-negative result due to unrelated rename activity.
2202 *
2203 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2204 * however it must be used carefully, eg. with a following d_lookup in
2205 * the case of failure.
2206 *
2207 * __d_lookup callers must be commented.
2208 */
Al Viroa713ca22013-01-24 18:27:00 -05002209struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210{
2211 unsigned int len = name->len;
2212 unsigned int hash = name->hash;
2213 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002214 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002215 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002216 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002217 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218
Nick Pigginb04f7842010-08-18 04:37:34 +10002219 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002220 * Note: There is significant duplication with __d_lookup_rcu which is
2221 * required to prevent single threaded performance regressions
2222 * especially on architectures where smp_rmb (in seqcounts) are costly.
2223 * Keep the two functions in sync.
2224 */
2225
2226 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002227 * The hash list is protected using RCU.
2228 *
2229 * Take d_lock when comparing a candidate dentry, to avoid races
2230 * with d_move().
2231 *
2232 * It is possible that concurrent renames can mess up our list
2233 * walk here and result in missing our dentry, resulting in the
2234 * false-negative result. d_lookup() protects against concurrent
2235 * renames using rename_lock seqlock.
2236 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002237 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002238 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 rcu_read_lock();
2240
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002241 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 if (dentry->d_name.hash != hash)
2244 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
2246 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 if (dentry->d_parent != parent)
2248 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002249 if (d_unhashed(dentry))
2250 goto next;
2251
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 /*
2253 * It is safe to compare names since d_move() cannot
2254 * change the qstr (protected by d_lock).
2255 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002256 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002257 int tlen = dentry->d_name.len;
2258 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002259 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 goto next;
2261 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002262 if (dentry->d_name.len != len)
2263 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002264 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 goto next;
2266 }
2267
Waiman Long98474232013-08-28 18:24:59 -07002268 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002269 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 spin_unlock(&dentry->d_lock);
2271 break;
2272next:
2273 spin_unlock(&dentry->d_lock);
2274 }
2275 rcu_read_unlock();
2276
2277 return found;
2278}
2279
2280/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002281 * d_hash_and_lookup - hash the qstr then search for a dentry
2282 * @dir: Directory to search in
2283 * @name: qstr of name we wish to find
2284 *
Al Viro4f522a22013-02-11 23:20:37 -05002285 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002286 */
2287struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2288{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002289 /*
2290 * Check for a fs-specific hash function. Note that we must
2291 * calculate the standard hash first, as the d_op->d_hash()
2292 * routine may choose to leave the hash value unchanged.
2293 */
2294 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002295 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002296 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002297 if (unlikely(err < 0))
2298 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002299 }
Al Viro4f522a22013-02-11 23:20:37 -05002300 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002301}
Al Viro4f522a22013-02-11 23:20:37 -05002302EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002303
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304/*
2305 * When a file is deleted, we have two options:
2306 * - turn this dentry into a negative dentry
2307 * - unhash this dentry and free it.
2308 *
2309 * Usually, we want to just turn this into
2310 * a negative dentry, but if anybody else is
2311 * currently using the dentry or the inode
2312 * we can't do that and we fall back on removing
2313 * it from the hash queues and waiting for
2314 * it to be deleted later when it has no users
2315 */
2316
2317/**
2318 * d_delete - delete a dentry
2319 * @dentry: The dentry to delete
2320 *
2321 * Turn the dentry into a negative dentry if possible, otherwise
2322 * remove it from the hash queues so it can be deleted later
2323 */
2324
2325void d_delete(struct dentry * dentry)
2326{
Nick Piggin873feea2011-01-07 17:50:06 +11002327 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002328 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 /*
2330 * Are we the only user?
2331 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002332again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002334 inode = dentry->d_inode;
2335 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002336 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002337 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002338 spin_unlock(&dentry->d_lock);
2339 cpu_relax();
2340 goto again;
2341 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002342 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002343 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002344 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 return;
2346 }
2347
2348 if (!d_unhashed(dentry))
2349 __d_drop(dentry);
2350
2351 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002352
2353 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002355EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002357static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002359 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002360 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002361 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002362 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002363 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364}
2365
David Howells770bfad2006-08-22 20:06:07 -04002366static void _d_rehash(struct dentry * entry)
2367{
2368 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2369}
2370
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371/**
2372 * d_rehash - add an entry back to the hash
2373 * @entry: dentry to add to the hash
2374 *
2375 * Adds a dentry to the hash according to its name.
2376 */
2377
2378void d_rehash(struct dentry * entry)
2379{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002381 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002384EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385
Al Viro84e710d2016-04-15 00:58:55 -04002386static inline unsigned start_dir_add(struct inode *dir)
2387{
2388
2389 for (;;) {
2390 unsigned n = dir->i_dir_seq;
2391 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2392 return n;
2393 cpu_relax();
2394 }
2395}
2396
2397static inline void end_dir_add(struct inode *dir, unsigned n)
2398{
2399 smp_store_release(&dir->i_dir_seq, n + 2);
2400}
2401
Al Virod9171b92016-04-15 03:33:13 -04002402static void d_wait_lookup(struct dentry *dentry)
2403{
2404 if (d_in_lookup(dentry)) {
2405 DECLARE_WAITQUEUE(wait, current);
2406 add_wait_queue(dentry->d_wait, &wait);
2407 do {
2408 set_current_state(TASK_UNINTERRUPTIBLE);
2409 spin_unlock(&dentry->d_lock);
2410 schedule();
2411 spin_lock(&dentry->d_lock);
2412 } while (d_in_lookup(dentry));
2413 }
2414}
2415
Al Viro94bdd652016-04-15 02:42:04 -04002416struct dentry *d_alloc_parallel(struct dentry *parent,
Al Virod9171b92016-04-15 03:33:13 -04002417 const struct qstr *name,
2418 wait_queue_head_t *wq)
Al Viro94bdd652016-04-15 02:42:04 -04002419{
2420 unsigned int len = name->len;
2421 unsigned int hash = name->hash;
2422 const unsigned char *str = name->name;
2423 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2424 struct hlist_bl_node *node;
2425 struct dentry *new = d_alloc(parent, name);
2426 struct dentry *dentry;
2427 unsigned seq, r_seq, d_seq;
2428
2429 if (unlikely(!new))
2430 return ERR_PTR(-ENOMEM);
2431
2432retry:
2433 rcu_read_lock();
2434 seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1;
2435 r_seq = read_seqbegin(&rename_lock);
2436 dentry = __d_lookup_rcu(parent, name, &d_seq);
2437 if (unlikely(dentry)) {
2438 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2439 rcu_read_unlock();
2440 goto retry;
2441 }
2442 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2443 rcu_read_unlock();
2444 dput(dentry);
2445 goto retry;
2446 }
2447 rcu_read_unlock();
2448 dput(new);
2449 return dentry;
2450 }
2451 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2452 rcu_read_unlock();
2453 goto retry;
2454 }
2455 hlist_bl_lock(b);
2456 if (unlikely(parent->d_inode->i_dir_seq != seq)) {
2457 hlist_bl_unlock(b);
2458 rcu_read_unlock();
2459 goto retry;
2460 }
2461 rcu_read_unlock();
2462 /*
2463 * No changes for the parent since the beginning of d_lookup().
2464 * Since all removals from the chain happen with hlist_bl_lock(),
2465 * any potential in-lookup matches are going to stay here until
2466 * we unlock the chain. All fields are stable in everything
2467 * we encounter.
2468 */
2469 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2470 if (dentry->d_name.hash != hash)
2471 continue;
2472 if (dentry->d_parent != parent)
2473 continue;
2474 if (d_unhashed(dentry))
2475 continue;
2476 if (parent->d_flags & DCACHE_OP_COMPARE) {
2477 int tlen = dentry->d_name.len;
2478 const char *tname = dentry->d_name.name;
2479 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2480 continue;
2481 } else {
2482 if (dentry->d_name.len != len)
2483 continue;
2484 if (dentry_cmp(dentry, str, len))
2485 continue;
2486 }
2487 dget(dentry);
2488 hlist_bl_unlock(b);
Al Virod9171b92016-04-15 03:33:13 -04002489 /* somebody is doing lookup for it right now; wait for it */
2490 spin_lock(&dentry->d_lock);
2491 d_wait_lookup(dentry);
2492 /*
2493 * it's not in-lookup anymore; in principle we should repeat
2494 * everything from dcache lookup, but it's likely to be what
2495 * d_lookup() would've found anyway. If it is, just return it;
2496 * otherwise we really have to repeat the whole thing.
2497 */
2498 if (unlikely(dentry->d_name.hash != hash))
2499 goto mismatch;
2500 if (unlikely(dentry->d_parent != parent))
2501 goto mismatch;
2502 if (unlikely(d_unhashed(dentry)))
2503 goto mismatch;
2504 if (parent->d_flags & DCACHE_OP_COMPARE) {
2505 int tlen = dentry->d_name.len;
2506 const char *tname = dentry->d_name.name;
2507 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2508 goto mismatch;
2509 } else {
2510 if (unlikely(dentry->d_name.len != len))
2511 goto mismatch;
2512 if (unlikely(dentry_cmp(dentry, str, len)))
2513 goto mismatch;
2514 }
2515 /* OK, it *is* a hashed match; return it */
2516 spin_unlock(&dentry->d_lock);
Al Viro94bdd652016-04-15 02:42:04 -04002517 dput(new);
2518 return dentry;
2519 }
2520 /* we can't take ->d_lock here; it's OK, though. */
2521 new->d_flags |= DCACHE_PAR_LOOKUP;
Al Virod9171b92016-04-15 03:33:13 -04002522 new->d_wait = wq;
Al Viro94bdd652016-04-15 02:42:04 -04002523 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2524 hlist_bl_unlock(b);
2525 return new;
Al Virod9171b92016-04-15 03:33:13 -04002526mismatch:
2527 spin_unlock(&dentry->d_lock);
2528 dput(dentry);
2529 goto retry;
Al Viro94bdd652016-04-15 02:42:04 -04002530}
2531EXPORT_SYMBOL(d_alloc_parallel);
2532
Al Viro85c7f812016-04-14 19:52:13 -04002533void __d_lookup_done(struct dentry *dentry)
2534{
Al Viro94bdd652016-04-15 02:42:04 -04002535 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2536 dentry->d_name.hash);
2537 hlist_bl_lock(b);
Al Viro85c7f812016-04-14 19:52:13 -04002538 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
Al Viro94bdd652016-04-15 02:42:04 -04002539 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
Al Virod9171b92016-04-15 03:33:13 -04002540 wake_up_all(dentry->d_wait);
2541 dentry->d_wait = NULL;
Al Viro94bdd652016-04-15 02:42:04 -04002542 hlist_bl_unlock(b);
2543 INIT_HLIST_NODE(&dentry->d_u.d_alias);
Al Virod9171b92016-04-15 03:33:13 -04002544 INIT_LIST_HEAD(&dentry->d_lru);
Al Viro85c7f812016-04-14 19:52:13 -04002545}
2546EXPORT_SYMBOL(__d_lookup_done);
Al Viroed782b52016-03-09 19:52:39 -05002547
2548/* inode->i_lock held if inode is non-NULL */
2549
2550static inline void __d_add(struct dentry *dentry, struct inode *inode)
2551{
Al Viro84e710d2016-04-15 00:58:55 -04002552 struct inode *dir = NULL;
2553 unsigned n;
Al Viro0568d702016-04-14 19:40:56 -04002554 spin_lock(&dentry->d_lock);
Al Viro84e710d2016-04-15 00:58:55 -04002555 if (unlikely(d_in_lookup(dentry))) {
2556 dir = dentry->d_parent->d_inode;
2557 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002558 __d_lookup_done(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002559 }
Al Viroed782b52016-03-09 19:52:39 -05002560 if (inode) {
Al Viro0568d702016-04-14 19:40:56 -04002561 unsigned add_flags = d_flags_for_inode(inode);
2562 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2563 raw_write_seqcount_begin(&dentry->d_seq);
2564 __d_set_inode_and_type(dentry, inode, add_flags);
2565 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04002566 fsnotify_update_flags(dentry);
Al Viroed782b52016-03-09 19:52:39 -05002567 }
Al Viro0568d702016-04-14 19:40:56 -04002568 _d_rehash(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002569 if (dir)
2570 end_dir_add(dir, n);
Al Viro0568d702016-04-14 19:40:56 -04002571 spin_unlock(&dentry->d_lock);
2572 if (inode)
2573 spin_unlock(&inode->i_lock);
Al Viroed782b52016-03-09 19:52:39 -05002574}
2575
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002576/**
Al Viro34d0d192016-03-08 21:01:03 -05002577 * d_add - add dentry to hash queues
2578 * @entry: dentry to add
2579 * @inode: The inode to attach to this dentry
2580 *
2581 * This adds the entry to the hash queues and initializes @inode.
2582 * The entry was actually filled in earlier during d_alloc().
2583 */
2584
2585void d_add(struct dentry *entry, struct inode *inode)
2586{
Al Virob9680912016-04-11 00:53:26 -04002587 if (inode) {
2588 security_d_instantiate(entry, inode);
Al Viroed782b52016-03-09 19:52:39 -05002589 spin_lock(&inode->i_lock);
Al Virob9680912016-04-11 00:53:26 -04002590 }
Al Viroed782b52016-03-09 19:52:39 -05002591 __d_add(entry, inode);
Al Viro34d0d192016-03-08 21:01:03 -05002592}
2593EXPORT_SYMBOL(d_add);
2594
2595/**
Al Viro668d0cd2016-03-08 12:44:17 -05002596 * d_exact_alias - find and hash an exact unhashed alias
2597 * @entry: dentry to add
2598 * @inode: The inode to go with this dentry
2599 *
2600 * If an unhashed dentry with the same name/parent and desired
2601 * inode already exists, hash and return it. Otherwise, return
2602 * NULL.
2603 *
2604 * Parent directory should be locked.
2605 */
2606struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2607{
2608 struct dentry *alias;
2609 int len = entry->d_name.len;
2610 const char *name = entry->d_name.name;
2611 unsigned int hash = entry->d_name.hash;
2612
2613 spin_lock(&inode->i_lock);
2614 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2615 /*
2616 * Don't need alias->d_lock here, because aliases with
2617 * d_parent == entry->d_parent are not subject to name or
2618 * parent changes, because the parent inode i_mutex is held.
2619 */
2620 if (alias->d_name.hash != hash)
2621 continue;
2622 if (alias->d_parent != entry->d_parent)
2623 continue;
2624 if (alias->d_name.len != len)
2625 continue;
2626 if (dentry_cmp(alias, name, len))
2627 continue;
2628 spin_lock(&alias->d_lock);
2629 if (!d_unhashed(alias)) {
2630 spin_unlock(&alias->d_lock);
2631 alias = NULL;
2632 } else {
2633 __dget_dlock(alias);
2634 _d_rehash(alias);
2635 spin_unlock(&alias->d_lock);
2636 }
2637 spin_unlock(&inode->i_lock);
2638 return alias;
2639 }
2640 spin_unlock(&inode->i_lock);
2641 return NULL;
2642}
2643EXPORT_SYMBOL(d_exact_alias);
2644
2645/**
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002646 * dentry_update_name_case - update case insensitive dentry with a new name
2647 * @dentry: dentry to be updated
2648 * @name: new name
2649 *
2650 * Update a case insensitive dentry with new case of name.
2651 *
2652 * dentry must have been returned by d_lookup with name @name. Old and new
2653 * name lengths must match (ie. no d_compare which allows mismatched name
2654 * lengths).
2655 *
2656 * Parent inode i_mutex must be held over d_lookup and into this call (to
2657 * keep renames and concurrent inserts, and readdir(2) away).
2658 */
2659void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2660{
Al Viro59551022016-01-22 15:40:57 -05002661 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002662 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2663
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002664 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002665 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002666 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002667 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002668 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002669}
2670EXPORT_SYMBOL(dentry_update_name_case);
2671
Al Viro8d85b482014-09-29 14:54:27 -04002672static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673{
Al Viro8d85b482014-09-29 14:54:27 -04002674 if (unlikely(dname_external(target))) {
2675 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 /*
2677 * Both external: swap the pointers
2678 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002679 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 } else {
2681 /*
2682 * dentry:internal, target:external. Steal target's
2683 * storage and make target internal.
2684 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002685 memcpy(target->d_iname, dentry->d_name.name,
2686 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 dentry->d_name.name = target->d_name.name;
2688 target->d_name.name = target->d_iname;
2689 }
2690 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002691 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 /*
2693 * dentry:external, target:internal. Give dentry's
2694 * storage to target and make dentry internal
2695 */
2696 memcpy(dentry->d_iname, target->d_name.name,
2697 target->d_name.len + 1);
2698 target->d_name.name = dentry->d_name.name;
2699 dentry->d_name.name = dentry->d_iname;
2700 } else {
2701 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002702 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002704 unsigned int i;
2705 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002706 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2707 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002708 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2709 swap(((long *) &dentry->d_iname)[i],
2710 ((long *) &target->d_iname)[i]);
2711 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 }
2713 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002714 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715}
2716
Al Viro8d85b482014-09-29 14:54:27 -04002717static void copy_name(struct dentry *dentry, struct dentry *target)
2718{
2719 struct external_name *old_name = NULL;
2720 if (unlikely(dname_external(dentry)))
2721 old_name = external_name(dentry);
2722 if (unlikely(dname_external(target))) {
2723 atomic_inc(&external_name(target)->u.count);
2724 dentry->d_name = target->d_name;
2725 } else {
2726 memcpy(dentry->d_iname, target->d_name.name,
2727 target->d_name.len + 1);
2728 dentry->d_name.name = dentry->d_iname;
2729 dentry->d_name.hash_len = target->d_name.hash_len;
2730 }
2731 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2732 kfree_rcu(old_name, u.head);
2733}
2734
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002735static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2736{
2737 /*
2738 * XXXX: do we really need to take target->d_lock?
2739 */
2740 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2741 spin_lock(&target->d_parent->d_lock);
2742 else {
2743 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2744 spin_lock(&dentry->d_parent->d_lock);
2745 spin_lock_nested(&target->d_parent->d_lock,
2746 DENTRY_D_LOCK_NESTED);
2747 } else {
2748 spin_lock(&target->d_parent->d_lock);
2749 spin_lock_nested(&dentry->d_parent->d_lock,
2750 DENTRY_D_LOCK_NESTED);
2751 }
2752 }
2753 if (target < dentry) {
2754 spin_lock_nested(&target->d_lock, 2);
2755 spin_lock_nested(&dentry->d_lock, 3);
2756 } else {
2757 spin_lock_nested(&dentry->d_lock, 2);
2758 spin_lock_nested(&target->d_lock, 3);
2759 }
2760}
2761
Al Viro986c0192014-09-26 23:11:15 -04002762static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002763{
2764 if (target->d_parent != dentry->d_parent)
2765 spin_unlock(&dentry->d_parent->d_lock);
2766 if (target->d_parent != target)
2767 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002768 spin_unlock(&target->d_lock);
2769 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002770}
2771
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002773 * When switching names, the actual string doesn't strictly have to
2774 * be preserved in the target - because we're dropping the target
2775 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002776 * the new name before we switch, unless we are going to rehash
2777 * it. Note that if we *do* unhash the target, we are not allowed
2778 * to rehash it without giving it a new name/hash key - whether
2779 * we swap or overwrite the names here, resulting name won't match
2780 * the reality in filesystem; it's only there for d_path() purposes.
2781 * Note that all of this is happening under rename_lock, so the
2782 * any hash lookup seeing it in the middle of manipulations will
2783 * be discarded anyway. So we do not care what happens to the hash
2784 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002786/*
Al Viro18367502011-07-12 21:42:24 -04002787 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 * @dentry: entry to move
2789 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002790 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 *
2792 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002793 * dcache entries should not be moved in this way. Caller must hold
2794 * rename_lock, the i_mutex of the source and target directories,
2795 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002797static void __d_move(struct dentry *dentry, struct dentry *target,
2798 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799{
Al Viro84e710d2016-04-15 00:58:55 -04002800 struct inode *dir = NULL;
2801 unsigned n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 if (!dentry->d_inode)
2803 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2804
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002805 BUG_ON(d_ancestor(dentry, target));
2806 BUG_ON(d_ancestor(target, dentry));
2807
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002808 dentry_lock_for_move(dentry, target);
Al Viro84e710d2016-04-15 00:58:55 -04002809 if (unlikely(d_in_lookup(target))) {
2810 dir = target->d_parent->d_inode;
2811 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002812 __d_lookup_done(target);
Al Viro84e710d2016-04-15 00:58:55 -04002813 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814
Nick Piggin31e6b012011-01-07 17:49:52 +11002815 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002816 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002817
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002818 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2819
2820 /*
2821 * Move the dentry to the target hash queue. Don't bother checking
2822 * for the same hash queue because of how unlikely it is.
2823 */
2824 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002825 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826
Miklos Szeredida1ce062014-04-01 17:08:43 +02002827 /*
2828 * Unhash the target (d_delete() is not usable here). If exchanging
2829 * the two dentries, then rehash onto the other's hash queue.
2830 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002832 if (exchange) {
2833 __d_rehash(target,
2834 d_hash(dentry->d_parent, dentry->d_name.hash));
2835 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002838 if (exchange)
2839 swap_names(dentry, target);
2840 else
2841 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842
Al Viro63cf4272014-09-26 23:06:14 -04002843 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002845 /* splicing a tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 dentry->d_parent = target->d_parent;
2847 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002848 list_del_init(&target->d_child);
2849 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002851 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002852 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002853 list_move(&target->d_child, &target->d_parent->d_subdirs);
2854 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002855 if (exchange)
Al Viroaffda482016-05-29 18:35:12 -04002856 fsnotify_update_flags(target);
2857 fsnotify_update_flags(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 }
2859
Nick Piggin31e6b012011-01-07 17:49:52 +11002860 write_seqcount_end(&target->d_seq);
2861 write_seqcount_end(&dentry->d_seq);
2862
Al Viro84e710d2016-04-15 00:58:55 -04002863 if (dir)
2864 end_dir_add(dir, n);
Al Viro986c0192014-09-26 23:11:15 -04002865 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002866}
2867
2868/*
2869 * d_move - move a dentry
2870 * @dentry: entry to move
2871 * @target: new dentry
2872 *
2873 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002874 * dcache entries should not be moved in this way. See the locking
2875 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002876 */
2877void d_move(struct dentry *dentry, struct dentry *target)
2878{
2879 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002880 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002882}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002883EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884
Miklos Szeredida1ce062014-04-01 17:08:43 +02002885/*
2886 * d_exchange - exchange two dentries
2887 * @dentry1: first dentry
2888 * @dentry2: second dentry
2889 */
2890void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2891{
2892 write_seqlock(&rename_lock);
2893
2894 WARN_ON(!dentry1->d_inode);
2895 WARN_ON(!dentry2->d_inode);
2896 WARN_ON(IS_ROOT(dentry1));
2897 WARN_ON(IS_ROOT(dentry2));
2898
2899 __d_move(dentry1, dentry2, true);
2900
2901 write_sequnlock(&rename_lock);
2902}
2903
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002904/**
2905 * d_ancestor - search for an ancestor
2906 * @p1: ancestor dentry
2907 * @p2: child dentry
2908 *
2909 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2910 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002911 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002912struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002913{
2914 struct dentry *p;
2915
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002916 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002917 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002918 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002919 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002920 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002921}
2922
2923/*
2924 * This helper attempts to cope with remotely renamed directories
2925 *
2926 * It assumes that the caller is already holding
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002927 * dentry->d_parent->d_inode->i_mutex, and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002928 *
2929 * Note: If ever the locking in lock_rename() changes, then please
2930 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002931 */
Al Virob5ae6b12014-10-12 22:16:02 -04002932static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002933 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002934{
Al Viro9902af72016-04-15 15:08:36 -04002935 struct mutex *m1 = NULL;
2936 struct rw_semaphore *m2 = NULL;
J. Bruce Fields3d330dc2015-02-10 10:55:53 -05002937 int ret = -ESTALE;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002938
2939 /* If alias and dentry share a parent, then no extra locks required */
2940 if (alias->d_parent == dentry->d_parent)
2941 goto out_unalias;
2942
Trond Myklebust9eaef272006-10-21 10:24:20 -07002943 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002944 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2945 goto out_err;
2946 m1 = &dentry->d_sb->s_vfs_rename_mutex;
Al Viro9902af72016-04-15 15:08:36 -04002947 if (!inode_trylock_shared(alias->d_parent->d_inode))
Trond Myklebust9eaef272006-10-21 10:24:20 -07002948 goto out_err;
Al Viro9902af72016-04-15 15:08:36 -04002949 m2 = &alias->d_parent->d_inode->i_rwsem;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002950out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002951 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04002952 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002953out_err:
Trond Myklebust9eaef272006-10-21 10:24:20 -07002954 if (m2)
Al Viro9902af72016-04-15 15:08:36 -04002955 up_read(m2);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002956 if (m1)
2957 mutex_unlock(m1);
2958 return ret;
2959}
2960
David Howells770bfad2006-08-22 20:06:07 -04002961/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002962 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2963 * @inode: the inode which may have a disconnected dentry
2964 * @dentry: a negative dentry which we want to point to the inode.
2965 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002966 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2967 * place of the given dentry and return it, else simply d_add the inode
2968 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002969 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002970 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2971 * we should error out: directories can't have multiple aliases.
2972 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002973 * This is needed in the lookup routine of any filesystem that is exportable
2974 * (via knfsd) so that we can build dcache paths to directories effectively.
2975 *
2976 * If a dentry was found and moved, then it is returned. Otherwise NULL
2977 * is returned. This matches the expected return value of ->lookup.
2978 *
2979 * Cluster filesystems may call this function with a negative, hashed dentry.
2980 * In that case, we know that the inode will be a regular file, and also this
2981 * will only occur during atomic_open. So we need to check for the dentry
2982 * being already hashed only in the final case.
2983 */
2984struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2985{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002986 if (IS_ERR(inode))
2987 return ERR_CAST(inode);
2988
David Howells770bfad2006-08-22 20:06:07 -04002989 BUG_ON(!d_unhashed(dentry));
2990
Al Virode689f52016-03-09 18:05:42 -05002991 if (!inode)
Al Virob5ae6b12014-10-12 22:16:02 -04002992 goto out;
Al Virode689f52016-03-09 18:05:42 -05002993
Al Virob9680912016-04-11 00:53:26 -04002994 security_d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11002995 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002996 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04002997 struct dentry *new = __d_find_any_alias(inode);
2998 if (unlikely(new)) {
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002999 /* The reference to new ensures it remains an alias */
3000 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04003001 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04003002 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04003003 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04003004 dput(new);
3005 new = ERR_PTR(-ELOOP);
3006 pr_warn_ratelimited(
3007 "VFS: Lookup of '%s' in %s %s"
3008 " would have caused loop\n",
3009 dentry->d_name.name,
3010 inode->i_sb->s_type->name,
3011 inode->i_sb->s_id);
3012 } else if (!IS_ROOT(new)) {
3013 int err = __d_unalias(inode, dentry, new);
3014 write_sequnlock(&rename_lock);
3015 if (err) {
3016 dput(new);
3017 new = ERR_PTR(err);
3018 }
Al Viro18367502011-07-12 21:42:24 -04003019 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04003020 __d_move(new, dentry, false);
3021 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003022 }
Al Virob5ae6b12014-10-12 22:16:02 -04003023 iput(inode);
3024 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003025 }
David Howells770bfad2006-08-22 20:06:07 -04003026 }
Al Virob5ae6b12014-10-12 22:16:02 -04003027out:
Al Viroed782b52016-03-09 19:52:39 -05003028 __d_add(dentry, inode);
Al Virob5ae6b12014-10-12 22:16:02 -04003029 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04003030}
Al Virob5ae6b12014-10-12 22:16:02 -04003031EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04003032
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003033static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01003034{
3035 *buflen -= namelen;
3036 if (*buflen < 0)
3037 return -ENAMETOOLONG;
3038 *buffer -= namelen;
3039 memcpy(*buffer, str, namelen);
3040 return 0;
3041}
3042
Waiman Long232d2d62013-09-09 12:18:13 -04003043/**
3044 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04003045 * @buffer: buffer pointer
3046 * @buflen: allocated length of the buffer
3047 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04003048 *
3049 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3050 * make sure that either the old or the new name pointer and length are
3051 * fetched. However, there may be mismatch between length and pointer.
3052 * The length cannot be trusted, we need to copy it byte-by-byte until
3053 * the length is reached or a null byte is found. It also prepends "/" at
3054 * the beginning of the name. The sequence number check at the caller will
3055 * retry it again when a d_move() does happen. So any garbage in the buffer
3056 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04003057 *
3058 * Data dependency barrier is needed to make sure that we see that terminating
3059 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04003060 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003061static int prepend_name(char **buffer, int *buflen, struct qstr *name)
3062{
Waiman Long232d2d62013-09-09 12:18:13 -04003063 const char *dname = ACCESS_ONCE(name->name);
3064 u32 dlen = ACCESS_ONCE(name->len);
3065 char *p;
3066
Al Viro6d13f692014-09-29 14:46:30 -04003067 smp_read_barrier_depends();
3068
Waiman Long232d2d62013-09-09 12:18:13 -04003069 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04003070 if (*buflen < 0)
3071 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04003072 p = *buffer -= dlen + 1;
3073 *p++ = '/';
3074 while (dlen--) {
3075 char c = *dname++;
3076 if (!c)
3077 break;
3078 *p++ = c;
3079 }
3080 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003081}
3082
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083/**
Randy Dunlap208898c2010-11-18 15:02:49 -08003084 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003085 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003086 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003087 * @buffer: pointer to the end of the buffer
3088 * @buflen: pointer to buffer length
3089 *
Waiman Long18129972013-09-12 10:55:35 -04003090 * The function will first try to write out the pathname without taking any
3091 * lock other than the RCU read lock to make sure that dentries won't go away.
3092 * It only checks the sequence number of the global rename_lock as any change
3093 * in the dentry's d_seq will be preceded by changes in the rename_lock
3094 * sequence number. If the sequence number had been changed, it will restart
3095 * the whole pathname back-tracing sequence again by taking the rename_lock.
3096 * In this case, there is no need to take the RCU read lock as the recursive
3097 * parent pointer references will keep the dentry chain alive as long as no
3098 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003099 */
Al Viro02125a82011-12-05 08:43:34 -05003100static int prepend_path(const struct path *path,
3101 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003102 char **buffer, int *buflen)
3103{
Al Viroede4ceb2013-11-13 07:45:40 -05003104 struct dentry *dentry;
3105 struct vfsmount *vfsmnt;
3106 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003107 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04003108 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04003109 char *bptr;
3110 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003111
Al Viro48f5ec22013-09-09 15:22:25 -04003112 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04003113restart_mnt:
3114 read_seqbegin_or_lock(&mount_lock, &m_seq);
3115 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003116 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003117restart:
3118 bptr = *buffer;
3119 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04003120 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05003121 dentry = path->dentry;
3122 vfsmnt = path->mnt;
3123 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04003124 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003125 while (dentry != root->dentry || vfsmnt != root->mnt) {
3126 struct dentry * parent;
3127
3128 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04003129 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Eric W. Biedermancde93be2015-08-15 13:36:12 -05003130 /* Escaped? */
3131 if (dentry != vfsmnt->mnt_root) {
3132 bptr = *buffer;
3133 blen = *buflen;
3134 error = 3;
3135 break;
3136 }
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003137 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04003138 if (mnt != parent) {
3139 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
3140 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04003141 vfsmnt = &mnt->mnt;
3142 continue;
3143 }
Waiman Long232d2d62013-09-09 12:18:13 -04003144 if (!error)
3145 error = is_mounted(vfsmnt) ? 1 : 2;
3146 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003147 }
3148 parent = dentry->d_parent;
3149 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003150 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003151 if (error)
3152 break;
3153
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003154 dentry = parent;
3155 }
Al Viro48f5ec22013-09-09 15:22:25 -04003156 if (!(seq & 1))
3157 rcu_read_unlock();
3158 if (need_seqretry(&rename_lock, seq)) {
3159 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003160 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003161 }
Al Viro48f5ec22013-09-09 15:22:25 -04003162 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003163
3164 if (!(m_seq & 1))
3165 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04003166 if (need_seqretry(&mount_lock, m_seq)) {
3167 m_seq = 1;
3168 goto restart_mnt;
3169 }
3170 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003171
Waiman Long232d2d62013-09-09 12:18:13 -04003172 if (error >= 0 && bptr == *buffer) {
3173 if (--blen < 0)
3174 error = -ENAMETOOLONG;
3175 else
3176 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003177 }
Waiman Long232d2d62013-09-09 12:18:13 -04003178 *buffer = bptr;
3179 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04003180 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003181}
3182
3183/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003184 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003185 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003186 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07003187 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 * @buflen: buffer length
3189 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003190 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003192 * Returns a pointer into the buffer or an error code if the
3193 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08003194 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04003195 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003196 *
Al Viro02125a82011-12-05 08:43:34 -05003197 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198 */
Al Viro02125a82011-12-05 08:43:34 -05003199char *__d_path(const struct path *path,
3200 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003201 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003203 char *res = buf + buflen;
3204 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003205
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003206 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003207 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04003208
Al Viro02125a82011-12-05 08:43:34 -05003209 if (error < 0)
3210 return ERR_PTR(error);
3211 if (error > 0)
3212 return NULL;
3213 return res;
3214}
3215
3216char *d_absolute_path(const struct path *path,
3217 char *buf, int buflen)
3218{
3219 struct path root = {};
3220 char *res = buf + buflen;
3221 int error;
3222
3223 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003224 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05003225
3226 if (error > 1)
3227 error = -EINVAL;
3228 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003229 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003230 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003231}
3232
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003233/*
3234 * same as __d_path but appends "(deleted)" for unlinked files.
3235 */
Al Viro02125a82011-12-05 08:43:34 -05003236static int path_with_deleted(const struct path *path,
3237 const struct path *root,
3238 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003239{
3240 prepend(buf, buflen, "\0", 1);
3241 if (d_unlinked(path->dentry)) {
3242 int error = prepend(buf, buflen, " (deleted)", 10);
3243 if (error)
3244 return error;
3245 }
3246
3247 return prepend_path(path, root, buf, buflen);
3248}
3249
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003250static int prepend_unreachable(char **buffer, int *buflen)
3251{
3252 return prepend(buffer, buflen, "(unreachable)", 13);
3253}
3254
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003255static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3256{
3257 unsigned seq;
3258
3259 do {
3260 seq = read_seqcount_begin(&fs->seq);
3261 *root = fs->root;
3262 } while (read_seqcount_retry(&fs->seq, seq));
3263}
3264
Jan Bluncka03a8a702008-02-14 19:38:32 -08003265/**
3266 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003267 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003268 * @buf: buffer to return value in
3269 * @buflen: buffer length
3270 *
3271 * Convert a dentry into an ASCII path name. If the entry has been deleted
3272 * the string " (deleted)" is appended. Note that this is ambiguous.
3273 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003274 * Returns a pointer into the buffer or an error code if the path was
3275 * too long. Note: Callers should use the returned pointer, not the passed
3276 * in buffer, to use the name! The implementation often starts at an offset
3277 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003278 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003279 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003280 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003281char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003283 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003284 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003285 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003287 /*
3288 * We have various synthetic filesystems that never get mounted. On
3289 * these filesystems dentries are never used for lookup purposes, and
3290 * thus don't need to be hashed. They also don't need a name until a
3291 * user wants to identify the object in /proc/pid/fd/. The little hack
3292 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003293 *
3294 * Some pseudo inodes are mountable. When they are mounted
3295 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3296 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003297 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003298 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3299 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003300 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003301
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003302 rcu_read_lock();
3303 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003304 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003305 rcu_read_unlock();
3306
Al Viro02125a82011-12-05 08:43:34 -05003307 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003308 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309 return res;
3310}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003311EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312
3313/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003314 * Helper function for dentry_operations.d_dname() members
3315 */
3316char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3317 const char *fmt, ...)
3318{
3319 va_list args;
3320 char temp[64];
3321 int sz;
3322
3323 va_start(args, fmt);
3324 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3325 va_end(args);
3326
3327 if (sz > sizeof(temp) || sz > buflen)
3328 return ERR_PTR(-ENAMETOOLONG);
3329
3330 buffer += buflen - sz;
3331 return memcpy(buffer, temp, sz);
3332}
3333
Al Viro118b2302013-08-24 12:08:17 -04003334char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3335{
3336 char *end = buffer + buflen;
3337 /* these dentries are never renamed, so d_lock is not needed */
3338 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003339 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003340 prepend(&end, &buflen, "/", 1))
3341 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003342 return end;
Al Viro118b2302013-08-24 12:08:17 -04003343}
David Herrmann31bbe162014-01-03 14:09:47 +01003344EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003345
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003346/*
Ram Pai6092d042008-03-27 13:06:20 +01003347 * Write full pathname from the root of the filesystem into the buffer.
3348 */
Al Virof6500802014-01-26 12:37:55 -05003349static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003350{
Al Virof6500802014-01-26 12:37:55 -05003351 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003352 char *end, *retval;
3353 int len, seq = 0;
3354 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003355
Al Virof6500802014-01-26 12:37:55 -05003356 if (buflen < 2)
3357 goto Elong;
3358
Al Viro48f5ec22013-09-09 15:22:25 -04003359 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003360restart:
Al Virof6500802014-01-26 12:37:55 -05003361 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003362 end = buf + buflen;
3363 len = buflen;
3364 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003365 /* Get '/' right */
3366 retval = end-1;
3367 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003368 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003369 while (!IS_ROOT(dentry)) {
3370 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003371
Ram Pai6092d042008-03-27 13:06:20 +01003372 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003373 error = prepend_name(&end, &len, &dentry->d_name);
3374 if (error)
3375 break;
Ram Pai6092d042008-03-27 13:06:20 +01003376
3377 retval = end;
3378 dentry = parent;
3379 }
Al Viro48f5ec22013-09-09 15:22:25 -04003380 if (!(seq & 1))
3381 rcu_read_unlock();
3382 if (need_seqretry(&rename_lock, seq)) {
3383 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003384 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003385 }
3386 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003387 if (error)
3388 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003389 return retval;
3390Elong:
3391 return ERR_PTR(-ENAMETOOLONG);
3392}
Nick Pigginec2447c2011-01-07 17:49:29 +11003393
3394char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3395{
Waiman Long232d2d62013-09-09 12:18:13 -04003396 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003397}
3398EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003399
3400char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3401{
3402 char *p = NULL;
3403 char *retval;
3404
Al Viroc1031352010-06-06 22:31:14 -04003405 if (d_unlinked(dentry)) {
3406 p = buf + buflen;
3407 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3408 goto Elong;
3409 buflen++;
3410 }
3411 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003412 if (!IS_ERR(retval) && p)
3413 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003414 return retval;
3415Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003416 return ERR_PTR(-ENAMETOOLONG);
3417}
3418
Linus Torvalds8b19e342013-09-12 10:35:47 -07003419static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3420 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003421{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003422 unsigned seq;
3423
3424 do {
3425 seq = read_seqcount_begin(&fs->seq);
3426 *root = fs->root;
3427 *pwd = fs->pwd;
3428 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003429}
3430
Ram Pai6092d042008-03-27 13:06:20 +01003431/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432 * NOTE! The user-level library version returns a
3433 * character pointer. The kernel system call just
3434 * returns the length of the buffer filled (which
3435 * includes the ending '\0' character), or a negative
3436 * error value. So libc would do something like
3437 *
3438 * char *getcwd(char * buf, size_t size)
3439 * {
3440 * int retval;
3441 *
3442 * retval = sys_getcwd(buf, size);
3443 * if (retval >= 0)
3444 * return buf;
3445 * errno = -retval;
3446 * return NULL;
3447 * }
3448 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003449SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450{
Linus Torvalds552ce542007-02-13 12:08:18 -08003451 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003452 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003453 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454
3455 if (!page)
3456 return -ENOMEM;
3457
Linus Torvalds8b19e342013-09-12 10:35:47 -07003458 rcu_read_lock();
3459 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460
Linus Torvalds552ce542007-02-13 12:08:18 -08003461 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003462 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003463 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003464 char *cwd = page + PATH_MAX;
3465 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003467 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003468 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003469 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003470
Al Viro02125a82011-12-05 08:43:34 -05003471 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003472 goto out;
3473
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003474 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003475 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003476 error = prepend_unreachable(&cwd, &buflen);
3477 if (error)
3478 goto out;
3479 }
3480
Linus Torvalds552ce542007-02-13 12:08:18 -08003481 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003482 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003483 if (len <= size) {
3484 error = len;
3485 if (copy_to_user(buf, cwd, len))
3486 error = -EFAULT;
3487 }
Nick Piggin949854d2011-01-07 17:49:37 +11003488 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003489 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003490 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491
3492out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003493 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494 return error;
3495}
3496
3497/*
3498 * Test whether new_dentry is a subdirectory of old_dentry.
3499 *
3500 * Trivially implemented using the dcache structure
3501 */
3502
3503/**
3504 * is_subdir - is new dentry a subdirectory of old_dentry
3505 * @new_dentry: new dentry
3506 * @old_dentry: old dentry
3507 *
Yaowei Baia6e57872015-11-17 14:40:11 +08003508 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3509 * Returns false otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3511 */
3512
Yaowei Baia6e57872015-11-17 14:40:11 +08003513bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003514{
Yaowei Baia6e57872015-11-17 14:40:11 +08003515 bool result;
Nick Piggin949854d2011-01-07 17:49:37 +11003516 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003517
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003518 if (new_dentry == old_dentry)
Yaowei Baia6e57872015-11-17 14:40:11 +08003519 return true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003520
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003521 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003524 /*
3525 * Need rcu_readlock to protect against the d_parent trashing
3526 * due to d_move
3527 */
3528 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003529 if (d_ancestor(old_dentry, new_dentry))
Yaowei Baia6e57872015-11-17 14:40:11 +08003530 result = true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003531 else
Yaowei Baia6e57872015-11-17 14:40:11 +08003532 result = false;
Nick Piggin949854d2011-01-07 17:49:37 +11003533 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003534 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535
3536 return result;
3537}
3538
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003539static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003541 struct dentry *root = data;
3542 if (dentry != root) {
3543 if (d_unhashed(dentry) || !dentry->d_inode)
3544 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003546 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3547 dentry->d_flags |= DCACHE_GENOCIDE;
3548 dentry->d_lockref.count--;
3549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003551 return D_WALK_CONTINUE;
3552}
Nick Piggin58db63d2011-01-07 17:49:39 +11003553
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003554void d_genocide(struct dentry *parent)
3555{
3556 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557}
3558
Al Viro60545d02013-06-07 01:20:27 -04003559void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560{
Al Viro60545d02013-06-07 01:20:27 -04003561 inode_dec_link_count(inode);
3562 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003563 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003564 !d_unlinked(dentry));
3565 spin_lock(&dentry->d_parent->d_lock);
3566 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3567 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3568 (unsigned long long)inode->i_ino);
3569 spin_unlock(&dentry->d_lock);
3570 spin_unlock(&dentry->d_parent->d_lock);
3571 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572}
Al Viro60545d02013-06-07 01:20:27 -04003573EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003574
3575static __initdata unsigned long dhash_entries;
3576static int __init set_dhash_entries(char *str)
3577{
3578 if (!str)
3579 return 0;
3580 dhash_entries = simple_strtoul(str, &str, 0);
3581 return 1;
3582}
3583__setup("dhash_entries=", set_dhash_entries);
3584
3585static void __init dcache_init_early(void)
3586{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003587 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003588
3589 /* If hashes are distributed across NUMA nodes, defer
3590 * hash allocation until vmalloc space is available.
3591 */
3592 if (hashdist)
3593 return;
3594
3595 dentry_hashtable =
3596 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003597 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598 dhash_entries,
3599 13,
3600 HASH_EARLY,
3601 &d_hash_shift,
3602 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003603 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604 0);
3605
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003606 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003607 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608}
3609
Denis Cheng74bf17c2007-10-16 23:26:30 -07003610static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003611{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003612 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003613
3614 /*
3615 * A constructor could be added for stable state like the lists,
3616 * but it is probably not worth it because of the cache nature
3617 * of the dcache.
3618 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003619 dentry_cache = KMEM_CACHE(dentry,
Vladimir Davydov5d097052016-01-14 15:18:21 -08003620 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003621
3622 /* Hash may have been set up in dcache_init_early */
3623 if (!hashdist)
3624 return;
3625
3626 dentry_hashtable =
3627 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003628 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 dhash_entries,
3630 13,
3631 0,
3632 &d_hash_shift,
3633 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003634 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003635 0);
3636
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003637 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003638 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003639}
3640
3641/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003642struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003643EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644
Linus Torvalds1da177e2005-04-16 15:20:36 -07003645EXPORT_SYMBOL(d_genocide);
3646
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647void __init vfs_caches_init_early(void)
3648{
3649 dcache_init_early();
3650 inode_init_early();
3651}
3652
Mel Gorman4248b0d2015-08-06 15:46:20 -07003653void __init vfs_caches_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003655 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003656 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657
Denis Cheng74bf17c2007-10-16 23:26:30 -07003658 dcache_init();
3659 inode_init();
Mel Gorman4248b0d2015-08-06 15:46:20 -07003660 files_init();
3661 files_maxfiles_init();
Denis Cheng74bf17c2007-10-16 23:26:30 -07003662 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003663 bdev_cache_init();
3664 chrdev_init();
3665}