blob: d6847d7b123d4ca0fe4a9100b751b673d8801cd9 [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 Viroba65dc52016-06-10 11:32:47 -0400510static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
511{
512 struct dentry *next;
513 /*
514 * Inform d_walk() and shrink_dentry_list() that we are no longer
515 * attached to the dentry tree
516 */
517 dentry->d_flags |= DCACHE_DENTRY_KILLED;
518 if (unlikely(list_empty(&dentry->d_child)))
519 return;
520 __list_del_entry(&dentry->d_child);
521 /*
522 * Cursors can move around the list of children. While we'd been
523 * a normal list member, it didn't matter - ->d_child.next would've
524 * been updated. However, from now on it won't be and for the
525 * things like d_walk() it might end up with a nasty surprise.
526 * Normally d_walk() doesn't care about cursors moving around -
527 * ->d_lock on parent prevents that and since a cursor has no children
528 * of its own, we get through it without ever unlocking the parent.
529 * There is one exception, though - if we ascend from a child that
530 * gets killed as soon as we unlock it, the next sibling is found
531 * using the value left in its ->d_child.next. And if _that_
532 * pointed to a cursor, and cursor got moved (e.g. by lseek())
533 * before d_walk() regains parent->d_lock, we'll end up skipping
534 * everything the cursor had been moved past.
535 *
536 * Solution: make sure that the pointer left behind in ->d_child.next
537 * points to something that won't be moving around. I.e. skip the
538 * cursors.
539 */
540 while (dentry->d_child.next != &parent->d_subdirs) {
541 next = list_entry(dentry->d_child.next, struct dentry, d_child);
542 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
543 break;
544 dentry->d_child.next = next->d_child.next;
545 }
546}
547
Al Viroe55fd012014-05-28 13:51:12 -0400548static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100549{
Al Viro41edf272014-05-01 10:30:00 -0400550 struct dentry *parent = NULL;
551 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400552 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100553 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100554
Linus Torvalds0d984392013-09-08 13:46:52 -0700555 /*
556 * The dentry is now unrecoverably dead to the world.
557 */
558 lockref_mark_dead(&dentry->d_lockref);
559
Sage Weilf0023bc2011-10-28 10:02:42 -0700560 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700561 * inform the fs via d_prune that this dentry is about to be
562 * unhashed and destroyed.
563 */
Al Viro29266202014-05-30 11:39:02 -0400564 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800565 dentry->d_op->d_prune(dentry);
566
Al Viro01b60352014-04-29 23:42:52 -0400567 if (dentry->d_flags & DCACHE_LRU_LIST) {
568 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
569 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400570 }
Nick Piggin77812a12011-01-07 17:49:48 +1100571 /* if it was on the hash then remove it */
572 __d_drop(dentry);
Al Viroba65dc52016-06-10 11:32:47 -0400573 dentry_unlist(dentry, parent);
Al Viro03b3b882014-04-29 15:45:28 -0400574 if (parent)
575 spin_unlock(&parent->d_lock);
576 dentry_iput(dentry);
577 /*
578 * dentry_iput drops the locks, at which point nobody (except
579 * transient RCU lookups) can reach this dentry.
580 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800581 BUG_ON(dentry->d_lockref.count > 0);
Al Viro03b3b882014-04-29 15:45:28 -0400582 this_cpu_dec(nr_dentry);
583 if (dentry->d_op && dentry->d_op->d_release)
584 dentry->d_op->d_release(dentry);
585
Al Viro41edf272014-05-01 10:30:00 -0400586 spin_lock(&dentry->d_lock);
587 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
588 dentry->d_flags |= DCACHE_MAY_FREE;
589 can_free = false;
590 }
591 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400592 if (likely(can_free))
593 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400594}
595
596/*
597 * Finish off a dentry we've decided to kill.
598 * dentry->d_lock must be held, returns with it unlocked.
599 * If ref is non-zero, then decrement the refcount too.
600 * Returns dentry requiring refcount drop, or NULL if we're done.
601 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400602static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400603 __releases(dentry->d_lock)
604{
605 struct inode *inode = dentry->d_inode;
606 struct dentry *parent = NULL;
607
608 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
609 goto failed;
610
611 if (!IS_ROOT(dentry)) {
612 parent = dentry->d_parent;
613 if (unlikely(!spin_trylock(&parent->d_lock))) {
614 if (inode)
615 spin_unlock(&inode->i_lock);
616 goto failed;
617 }
618 }
619
620 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400621 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400622
623failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400624 spin_unlock(&dentry->d_lock);
625 cpu_relax();
Al Viroe55fd012014-05-28 13:51:12 -0400626 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100627}
628
Al Viro046b9612014-05-29 08:54:52 -0400629static inline struct dentry *lock_parent(struct dentry *dentry)
630{
631 struct dentry *parent = dentry->d_parent;
632 if (IS_ROOT(dentry))
633 return NULL;
Linus Torvalds360f5472015-01-09 15:19:03 -0800634 if (unlikely(dentry->d_lockref.count < 0))
Al Viroc2338f22014-06-12 00:29:13 -0400635 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400636 if (likely(spin_trylock(&parent->d_lock)))
637 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400638 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400639 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400640again:
641 parent = ACCESS_ONCE(dentry->d_parent);
642 spin_lock(&parent->d_lock);
643 /*
644 * We can't blindly lock dentry until we are sure
645 * that we won't violate the locking order.
646 * Any changes of dentry->d_parent must have
647 * been done with parent->d_lock held, so
648 * spin_lock() above is enough of a barrier
649 * for checking if it's still our child.
650 */
651 if (unlikely(parent != dentry->d_parent)) {
652 spin_unlock(&parent->d_lock);
653 goto again;
654 }
655 rcu_read_unlock();
656 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700657 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400658 else
659 parent = NULL;
660 return parent;
661}
662
Linus Torvalds360f5472015-01-09 15:19:03 -0800663/*
664 * Try to do a lockless dput(), and return whether that was successful.
665 *
666 * If unsuccessful, we return false, having already taken the dentry lock.
667 *
668 * The caller needs to hold the RCU read lock, so that the dentry is
669 * guaranteed to stay around even if the refcount goes down to zero!
670 */
671static inline bool fast_dput(struct dentry *dentry)
672{
673 int ret;
674 unsigned int d_flags;
675
676 /*
677 * If we have a d_op->d_delete() operation, we sould not
Al Viro75a6f822015-07-08 02:42:38 +0100678 * let the dentry count go to zero, so use "put_or_lock".
Linus Torvalds360f5472015-01-09 15:19:03 -0800679 */
680 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
681 return lockref_put_or_lock(&dentry->d_lockref);
682
683 /*
684 * .. otherwise, we can try to just decrement the
685 * lockref optimistically.
686 */
687 ret = lockref_put_return(&dentry->d_lockref);
688
689 /*
690 * If the lockref_put_return() failed due to the lock being held
691 * by somebody else, the fast path has failed. We will need to
692 * get the lock, and then check the count again.
693 */
694 if (unlikely(ret < 0)) {
695 spin_lock(&dentry->d_lock);
696 if (dentry->d_lockref.count > 1) {
697 dentry->d_lockref.count--;
698 spin_unlock(&dentry->d_lock);
699 return 1;
700 }
701 return 0;
702 }
703
704 /*
705 * If we weren't the last ref, we're done.
706 */
707 if (ret)
708 return 1;
709
710 /*
711 * Careful, careful. The reference count went down
712 * to zero, but we don't hold the dentry lock, so
713 * somebody else could get it again, and do another
714 * dput(), and we need to not race with that.
715 *
716 * However, there is a very special and common case
717 * where we don't care, because there is nothing to
718 * do: the dentry is still hashed, it does not have
719 * a 'delete' op, and it's referenced and already on
720 * the LRU list.
721 *
722 * NOTE! Since we aren't locked, these values are
723 * not "stable". However, it is sufficient that at
724 * some point after we dropped the reference the
725 * dentry was hashed and the flags had the proper
726 * value. Other dentry users may have re-gotten
727 * a reference to the dentry and change that, but
728 * our work is done - we can leave the dentry
729 * around with a zero refcount.
730 */
731 smp_rmb();
732 d_flags = ACCESS_ONCE(dentry->d_flags);
Al Viro75a6f822015-07-08 02:42:38 +0100733 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
Linus Torvalds360f5472015-01-09 15:19:03 -0800734
735 /* Nothing to do? Dropping the reference was all we needed? */
736 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
737 return 1;
738
739 /*
740 * Not the fast normal case? Get the lock. We've already decremented
741 * the refcount, but we'll need to re-check the situation after
742 * getting the lock.
743 */
744 spin_lock(&dentry->d_lock);
745
746 /*
747 * Did somebody else grab a reference to it in the meantime, and
748 * we're no longer the last user after all? Alternatively, somebody
749 * else could have killed it and marked it dead. Either way, we
750 * don't need to do anything else.
751 */
752 if (dentry->d_lockref.count) {
753 spin_unlock(&dentry->d_lock);
754 return 1;
755 }
756
757 /*
758 * Re-get the reference we optimistically dropped. We hold the
759 * lock, and we just tested that it was zero, so we can just
760 * set it to 1.
761 */
762 dentry->d_lockref.count = 1;
763 return 0;
764}
765
766
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767/*
768 * This is dput
769 *
770 * This is complicated by the fact that we do not want to put
771 * dentries that are no longer on any hash chain on the unused
772 * list: we'd much rather just get rid of them immediately.
773 *
774 * However, that implies that we have to traverse the dentry
775 * tree upwards to the parents which might _also_ now be
776 * scheduled for deletion (it may have been only waiting for
777 * its last child to go away).
778 *
779 * This tail recursion is done by hand as we don't want to depend
780 * on the compiler to always get this right (gcc generally doesn't).
781 * Real recursion would eat up our stack space.
782 */
783
784/*
785 * dput - release a dentry
786 * @dentry: dentry to release
787 *
788 * Release a dentry. This will drop the usage count and if appropriate
789 * call the dentry unlink method as well as removing it from the queues and
790 * releasing its resources. If the parent dentries were scheduled for release
791 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793void dput(struct dentry *dentry)
794{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700795 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 return;
797
798repeat:
Linus Torvalds360f5472015-01-09 15:19:03 -0800799 rcu_read_lock();
800 if (likely(fast_dput(dentry))) {
801 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 return;
Linus Torvalds360f5472015-01-09 15:19:03 -0800803 }
804
805 /* Slow case: now with the dentry lock held */
806 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
Al Viro85c7f812016-04-14 19:52:13 -0400808 WARN_ON(d_in_lookup(dentry));
809
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700810 /* Unreachable? Get rid of it */
811 if (unlikely(d_unhashed(dentry)))
812 goto kill_it;
813
Al Viro75a6f822015-07-08 02:42:38 +0100814 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
815 goto kill_it;
816
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700817 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100819 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 }
Nick Piggin265ac902010-10-10 05:36:24 -0400821
Linus Torvalds358eec12013-10-31 15:43:02 -0700822 if (!(dentry->d_flags & DCACHE_REFERENCED))
823 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400824 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400825
Waiman Long98474232013-08-28 18:24:59 -0700826 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100827 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 return;
829
Miklos Szeredid52b9082007-05-08 00:23:46 -0700830kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400831 dentry = dentry_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700832 if (dentry)
833 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700835EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100838/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100839static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840{
Waiman Long98474232013-08-28 18:24:59 -0700841 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842}
843
Nick Piggindc0474b2011-01-07 17:49:43 +1100844static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100845{
Waiman Long98474232013-08-28 18:24:59 -0700846 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100847}
848
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100849struct dentry *dget_parent(struct dentry *dentry)
850{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700851 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100852 struct dentry *ret;
853
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700854 /*
855 * Do optimistic parent lookup without any
856 * locking.
857 */
858 rcu_read_lock();
859 ret = ACCESS_ONCE(dentry->d_parent);
860 gotref = lockref_get_not_zero(&ret->d_lockref);
861 rcu_read_unlock();
862 if (likely(gotref)) {
863 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
864 return ret;
865 dput(ret);
866 }
867
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100868repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100869 /*
870 * Don't need rcu_dereference because we re-check it was correct under
871 * the lock.
872 */
873 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100874 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100875 spin_lock(&ret->d_lock);
876 if (unlikely(ret != dentry->d_parent)) {
877 spin_unlock(&ret->d_lock);
878 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100879 goto repeat;
880 }
Nick Piggina734eb42011-01-07 17:49:44 +1100881 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700882 BUG_ON(!ret->d_lockref.count);
883 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100884 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100885 return ret;
886}
887EXPORT_SYMBOL(dget_parent);
888
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889/**
890 * d_find_alias - grab a hashed alias of inode
891 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 *
893 * If inode has a hashed alias, or is a directory and has any alias,
894 * acquire the reference to alias and return it. Otherwise return NULL.
895 * Notice that if inode is a directory there can be only one alias and
896 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800897 * of a filesystem, or if the directory was renamed and d_revalidate
898 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700900 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500901 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500903static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904{
Nick Pigginda502952011-01-07 17:49:33 +1100905 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
Nick Pigginda502952011-01-07 17:49:33 +1100907again:
908 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400909 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100910 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700912 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100913 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500915 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100916 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100917 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 return alias;
919 }
920 }
Nick Pigginda502952011-01-07 17:49:33 +1100921 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 }
Nick Pigginda502952011-01-07 17:49:33 +1100923 if (discon_alias) {
924 alias = discon_alias;
925 spin_lock(&alias->d_lock);
926 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500927 __dget_dlock(alias);
928 spin_unlock(&alias->d_lock);
929 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100930 }
931 spin_unlock(&alias->d_lock);
932 goto again;
933 }
934 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935}
936
Nick Pigginda502952011-01-07 17:49:33 +1100937struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938{
David Howells214fda12006-03-25 03:06:36 -0800939 struct dentry *de = NULL;
940
Al Virob3d9b7a2012-06-09 13:51:19 -0400941 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100942 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500943 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100944 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800945 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 return de;
947}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700948EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
950/*
951 * Try to kill dentries associated with this inode.
952 * WARNING: you must own a reference to inode.
953 */
954void d_prune_aliases(struct inode *inode)
955{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700956 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100958 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400959 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700961 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400962 struct dentry *parent = lock_parent(dentry);
963 if (likely(!dentry->d_lockref.count)) {
964 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800965 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400966 goto restart;
967 }
968 if (parent)
969 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 }
971 spin_unlock(&dentry->d_lock);
972 }
Nick Piggin873feea2011-01-07 17:50:06 +1100973 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700975EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400977static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978{
Al Viro5c47e6d2014-04-29 16:13:18 -0400979 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700980
Miklos Szeredi60942f22014-05-02 15:38:39 -0400981 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400982 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400983 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100984 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400985 parent = lock_parent(dentry);
986
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000988 * The dispose list is isolated and dentries are not accounted
989 * to the LRU here, so we can simply remove it from the list
990 * here regardless of whether it is referenced or not.
991 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400992 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000993
994 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000996 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800998 if (dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700999 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -04001000 if (parent)
1001 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 continue;
1003 }
Nick Piggin77812a12011-01-07 17:49:48 +11001004
Al Viro64fd72e2014-05-28 09:48:44 -04001005
1006 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
1007 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
1008 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -04001009 if (parent)
1010 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -04001011 if (can_free)
1012 dentry_free(dentry);
1013 continue;
1014 }
1015
Al Viroff2fde92014-05-28 13:59:13 -04001016 inode = dentry->d_inode;
1017 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001018 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001019 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -04001020 if (parent)
1021 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -04001022 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001023 }
Al Viroff2fde92014-05-28 13:59:13 -04001024
Al Viroff2fde92014-05-28 13:59:13 -04001025 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -04001026
Al Viro5c47e6d2014-04-29 16:13:18 -04001027 /*
1028 * We need to prune ancestors too. This is necessary to prevent
1029 * quadratic behavior of shrink_dcache_parent(), but is also
1030 * expected to be beneficial in reducing dentry cache
1031 * fragmentation.
1032 */
1033 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -04001034 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1035 parent = lock_parent(dentry);
1036 if (dentry->d_lockref.count != 1) {
1037 dentry->d_lockref.count--;
1038 spin_unlock(&dentry->d_lock);
1039 if (parent)
1040 spin_unlock(&parent->d_lock);
1041 break;
1042 }
1043 inode = dentry->d_inode; /* can't be NULL */
1044 if (unlikely(!spin_trylock(&inode->i_lock))) {
1045 spin_unlock(&dentry->d_lock);
1046 if (parent)
1047 spin_unlock(&parent->d_lock);
1048 cpu_relax();
1049 continue;
1050 }
1051 __dentry_kill(dentry);
1052 dentry = parent;
1053 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001055}
1056
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001057static enum lru_status dentry_lru_isolate(struct list_head *item,
1058 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +10001059{
1060 struct list_head *freeable = arg;
1061 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1062
1063
1064 /*
1065 * we are inverting the lru lock/dentry->d_lock here,
1066 * so use a trylock. If we fail to get the lock, just skip
1067 * it
1068 */
1069 if (!spin_trylock(&dentry->d_lock))
1070 return LRU_SKIP;
1071
1072 /*
1073 * Referenced dentries are still in use. If they have active
1074 * counts, just remove them from the LRU. Otherwise give them
1075 * another pass through the LRU.
1076 */
1077 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001078 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001079 spin_unlock(&dentry->d_lock);
1080 return LRU_REMOVED;
1081 }
1082
1083 if (dentry->d_flags & DCACHE_REFERENCED) {
1084 dentry->d_flags &= ~DCACHE_REFERENCED;
1085 spin_unlock(&dentry->d_lock);
1086
1087 /*
1088 * The list move itself will be made by the common LRU code. At
1089 * this point, we've dropped the dentry->d_lock but keep the
1090 * lru lock. This is safe to do, since every list movement is
1091 * protected by the lru lock even if both locks are held.
1092 *
1093 * This is guaranteed by the fact that all LRU management
1094 * functions are intermediated by the LRU API calls like
1095 * list_lru_add and list_lru_del. List movement in this file
1096 * only ever occur through this functions or through callbacks
1097 * like this one, that are called from the LRU API.
1098 *
1099 * The only exceptions to this are functions like
1100 * shrink_dentry_list, and code that first checks for the
1101 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1102 * operating only with stack provided lists after they are
1103 * properly isolated from the main list. It is thus, always a
1104 * local access.
1105 */
1106 return LRU_ROTATE;
1107 }
1108
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001109 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001110 spin_unlock(&dentry->d_lock);
1111
1112 return LRU_REMOVED;
1113}
1114
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001115/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001116 * prune_dcache_sb - shrink the dcache
1117 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001118 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001119 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001120 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1121 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001122 * function.
1123 *
1124 * This function may fail to free any resources if all the dentries are in
1125 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001126 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001127long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001128{
Dave Chinnerf6041562013-08-28 10:18:00 +10001129 LIST_HEAD(dispose);
1130 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001131
Vladimir Davydov503c3582015-02-12 14:58:47 -08001132 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1133 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001134 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001135 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136}
1137
Glauber Costa4e717f52013-08-28 10:18:03 +10001138static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001139 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001140{
Glauber Costa4e717f52013-08-28 10:18:03 +10001141 struct list_head *freeable = arg;
1142 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001143
Glauber Costa4e717f52013-08-28 10:18:03 +10001144 /*
1145 * we are inverting the lru lock/dentry->d_lock here,
1146 * so use a trylock. If we fail to get the lock, just skip
1147 * it
1148 */
1149 if (!spin_trylock(&dentry->d_lock))
1150 return LRU_SKIP;
1151
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001152 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001153 spin_unlock(&dentry->d_lock);
1154
1155 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001156}
1157
Glauber Costa4e717f52013-08-28 10:18:03 +10001158
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001159/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 * shrink_dcache_sb - shrink dcache for a superblock
1161 * @sb: superblock
1162 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001163 * Shrink the dcache for the specified super block. This is used to free
1164 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001166void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167{
Glauber Costa4e717f52013-08-28 10:18:03 +10001168 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001169
Glauber Costa4e717f52013-08-28 10:18:03 +10001170 do {
1171 LIST_HEAD(dispose);
1172
1173 freed = list_lru_walk(&sb->s_dentry_lru,
1174 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1175
1176 this_cpu_sub(nr_dentry_unused, freed);
1177 shrink_dentry_list(&dispose);
1178 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001180EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001183 * enum d_walk_ret - action to talke during tree walk
1184 * @D_WALK_CONTINUE: contrinue walk
1185 * @D_WALK_QUIT: quit walk
1186 * @D_WALK_NORETRY: quit when retry is needed
1187 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001189enum d_walk_ret {
1190 D_WALK_CONTINUE,
1191 D_WALK_QUIT,
1192 D_WALK_NORETRY,
1193 D_WALK_SKIP,
1194};
1195
1196/**
1197 * d_walk - walk the dentry tree
1198 * @parent: start of walk
1199 * @data: data passed to @enter() and @finish()
1200 * @enter: callback when first entering the dentry
1201 * @finish: callback when successfully finished the walk
1202 *
1203 * The @enter() and @finish() callbacks are called with d_lock held.
1204 */
1205static void d_walk(struct dentry *parent, void *data,
1206 enum d_walk_ret (*enter)(void *, struct dentry *),
1207 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208{
Nick Piggin949854d2011-01-07 17:49:37 +11001209 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001211 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001212 enum d_walk_ret ret;
1213 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001214
Nick Piggin58db63d2011-01-07 17:49:39 +11001215again:
Al Viro48f5ec22013-09-09 15:22:25 -04001216 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001217 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001218 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001219
1220 ret = enter(data, this_parent);
1221 switch (ret) {
1222 case D_WALK_CONTINUE:
1223 break;
1224 case D_WALK_QUIT:
1225 case D_WALK_SKIP:
1226 goto out_unlock;
1227 case D_WALK_NORETRY:
1228 retry = false;
1229 break;
1230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231repeat:
1232 next = this_parent->d_subdirs.next;
1233resume:
1234 while (next != &this_parent->d_subdirs) {
1235 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001236 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001238
Al Viroba65dc52016-06-10 11:32:47 -04001239 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1240 continue;
1241
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001242 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001243
1244 ret = enter(data, dentry);
1245 switch (ret) {
1246 case D_WALK_CONTINUE:
1247 break;
1248 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001249 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001250 goto out_unlock;
1251 case D_WALK_NORETRY:
1252 retry = false;
1253 break;
1254 case D_WALK_SKIP:
1255 spin_unlock(&dentry->d_lock);
1256 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001257 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001260 spin_unlock(&this_parent->d_lock);
1261 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001263 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 goto repeat;
1265 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001266 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 }
1268 /*
1269 * All done at this level ... ascend and resume the search.
1270 */
Al Viroca5358e2014-10-26 19:31:10 -04001271 rcu_read_lock();
1272ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001274 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001275 this_parent = child->d_parent;
1276
Al Viro31dec132013-10-25 17:04:27 -04001277 spin_unlock(&child->d_lock);
1278 spin_lock(&this_parent->d_lock);
1279
Al Viroca5358e2014-10-26 19:31:10 -04001280 /* might go back up the wrong parent if we have had a rename. */
1281 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001282 goto rename_retry;
Al Viro21591842015-05-28 23:09:19 -04001283 /* go into the first sibling still alive */
1284 do {
1285 next = child->d_child.next;
Al Viroca5358e2014-10-26 19:31:10 -04001286 if (next == &this_parent->d_subdirs)
1287 goto ascend;
1288 child = list_entry(next, struct dentry, d_child);
Al Viro21591842015-05-28 23:09:19 -04001289 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
Al Viro31dec132013-10-25 17:04:27 -04001290 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 goto resume;
1292 }
Al Viroca5358e2014-10-26 19:31:10 -04001293 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001294 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001295 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001296 if (finish)
1297 finish(data);
1298
1299out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001300 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001301 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001302 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001303
1304rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001305 spin_unlock(&this_parent->d_lock);
1306 rcu_read_unlock();
1307 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001308 if (!retry)
1309 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001310 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001311 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001313
1314/*
1315 * Search for at least 1 mount point in the dentry's subdirs.
1316 * We descend to the next level whenever the d_subdirs
1317 * list is non-empty and continue searching.
1318 */
1319
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001320static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1321{
1322 int *ret = data;
1323 if (d_mountpoint(dentry)) {
1324 *ret = 1;
1325 return D_WALK_QUIT;
1326 }
1327 return D_WALK_CONTINUE;
1328}
1329
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001330/**
1331 * have_submounts - check for mounts over a dentry
1332 * @parent: dentry to check.
1333 *
1334 * Return true if the parent or its subdirectories contain
1335 * a mount point
1336 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001337int have_submounts(struct dentry *parent)
1338{
1339 int ret = 0;
1340
1341 d_walk(parent, &ret, check_mount, NULL);
1342
1343 return ret;
1344}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001345EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
1347/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001348 * Called by mount code to set a mountpoint and check if the mountpoint is
1349 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1350 * subtree can become unreachable).
1351 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001352 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001353 * this reason take rename_lock and d_lock on dentry and ancestors.
1354 */
1355int d_set_mounted(struct dentry *dentry)
1356{
1357 struct dentry *p;
1358 int ret = -ENOENT;
1359 write_seqlock(&rename_lock);
1360 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001361 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001362 spin_lock(&p->d_lock);
1363 if (unlikely(d_unhashed(p))) {
1364 spin_unlock(&p->d_lock);
1365 goto out;
1366 }
1367 spin_unlock(&p->d_lock);
1368 }
1369 spin_lock(&dentry->d_lock);
1370 if (!d_unlinked(dentry)) {
1371 dentry->d_flags |= DCACHE_MOUNTED;
1372 ret = 0;
1373 }
1374 spin_unlock(&dentry->d_lock);
1375out:
1376 write_sequnlock(&rename_lock);
1377 return ret;
1378}
1379
1380/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001381 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 * and move any unused dentries to the end of the unused
1383 * list for prune_dcache(). We descend to the next level
1384 * whenever the d_subdirs list is non-empty and continue
1385 * searching.
1386 *
1387 * It returns zero iff there are no unused children,
1388 * otherwise it returns the number of children moved to
1389 * the end of the unused list. This may not be the total
1390 * number of unused children, because select_parent can
1391 * drop the lock and return early due to latency
1392 * constraints.
1393 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001394
1395struct select_data {
1396 struct dentry *start;
1397 struct list_head dispose;
1398 int found;
1399};
1400
1401static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001403 struct select_data *data = _data;
1404 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001406 if (data->start == dentry)
1407 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
Al Virofe915222014-05-03 00:02:25 -04001409 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001410 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001411 } else {
1412 if (dentry->d_flags & DCACHE_LRU_LIST)
1413 d_lru_del(dentry);
1414 if (!dentry->d_lockref.count) {
1415 d_shrink_add(dentry, &data->dispose);
1416 data->found++;
1417 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 }
1419 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001420 * We can return to the caller if we have found some (this
1421 * ensures forward progress). We'll be coming back to find
1422 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 */
Al Virofe915222014-05-03 00:02:25 -04001424 if (!list_empty(&data->dispose))
1425 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001427 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428}
1429
1430/**
1431 * shrink_dcache_parent - prune dcache
1432 * @parent: parent of entries to prune
1433 *
1434 * Prune the dcache to remove unused children of the parent dentry.
1435 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001436void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001438 for (;;) {
1439 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001441 INIT_LIST_HEAD(&data.dispose);
1442 data.start = parent;
1443 data.found = 0;
1444
1445 d_walk(parent, &data, select_collect, NULL);
1446 if (!data.found)
1447 break;
1448
1449 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001450 cond_resched();
1451 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001453EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
Al Viro9c8c10e2014-05-02 20:36:10 -04001455static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001456{
Al Viro9c8c10e2014-05-02 20:36:10 -04001457 /* it has busy descendents; complain about those instead */
1458 if (!list_empty(&dentry->d_subdirs))
1459 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001460
Al Viro9c8c10e2014-05-02 20:36:10 -04001461 /* root with refcount 1 is fine */
1462 if (dentry == _data && dentry->d_lockref.count == 1)
1463 return D_WALK_CONTINUE;
1464
1465 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1466 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001467 dentry,
1468 dentry->d_inode ?
1469 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001470 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001471 dentry->d_lockref.count,
1472 dentry->d_sb->s_type->name,
1473 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001474 WARN_ON(1);
1475 return D_WALK_CONTINUE;
1476}
1477
1478static void do_one_tree(struct dentry *dentry)
1479{
1480 shrink_dcache_parent(dentry);
1481 d_walk(dentry, dentry, umount_check, NULL);
1482 d_drop(dentry);
1483 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001484}
1485
1486/*
1487 * destroy the dentries attached to a superblock on unmounting
1488 */
1489void shrink_dcache_for_umount(struct super_block *sb)
1490{
1491 struct dentry *dentry;
1492
Al Viro9c8c10e2014-05-02 20:36:10 -04001493 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001494
1495 dentry = sb->s_root;
1496 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001497 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001498
1499 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001500 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1501 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001502 }
1503}
1504
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001505struct detach_data {
1506 struct select_data select;
1507 struct dentry *mountpoint;
1508};
1509static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001510{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001511 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001512
1513 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001514 __dget_dlock(dentry);
1515 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001516 return D_WALK_QUIT;
1517 }
1518
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001519 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001520}
1521
1522static void check_and_drop(void *_data)
1523{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001524 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001525
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001526 if (!data->mountpoint && !data->select.found)
1527 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001528}
1529
1530/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001531 * d_invalidate - detach submounts, prune dcache, and drop
1532 * @dentry: dentry to invalidate (aka detach, prune and drop)
1533 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001534 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001535 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001536 * The final d_drop is done as an atomic operation relative to
1537 * rename_lock ensuring there are no races with d_set_mounted. This
1538 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001539 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001540void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001541{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001542 /*
1543 * If it's already been dropped, return OK.
1544 */
1545 spin_lock(&dentry->d_lock);
1546 if (d_unhashed(dentry)) {
1547 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001548 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001549 }
1550 spin_unlock(&dentry->d_lock);
1551
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001552 /* Negative dentries can be dropped without further checks */
1553 if (!dentry->d_inode) {
1554 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001555 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001556 }
1557
1558 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001559 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001560
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001561 data.mountpoint = NULL;
1562 INIT_LIST_HEAD(&data.select.dispose);
1563 data.select.start = dentry;
1564 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001565
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001566 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001567
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001568 if (data.select.found)
1569 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001570
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001571 if (data.mountpoint) {
1572 detach_mounts(data.mountpoint);
1573 dput(data.mountpoint);
1574 }
1575
1576 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001577 break;
1578
1579 cond_resched();
1580 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001581}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001582EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001583
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584/**
Al Viroa4464db2011-07-07 15:03:58 -04001585 * __d_alloc - allocate a dcache entry
1586 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 * @name: qstr of the name
1588 *
1589 * Allocates a dentry. It returns %NULL if there is insufficient memory
1590 * available. On a success the dentry is returned. The name passed in is
1591 * copied and the copy passed in may be reused after this call.
1592 */
1593
Al Viroa4464db2011-07-07 15:03:58 -04001594struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595{
1596 struct dentry *dentry;
1597 char *dname;
1598
Mel Gormane12ba742007-10-16 01:25:52 -07001599 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 if (!dentry)
1601 return NULL;
1602
Linus Torvalds6326c712012-05-21 16:14:04 -07001603 /*
1604 * We guarantee that the inline name is always NUL-terminated.
1605 * This way the memcpy() done by the name switching in rename
1606 * will still always have a NUL at the end, even if we might
1607 * be overwriting an internal NUL character
1608 */
1609 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Al Viro798434b2016-03-24 20:38:43 -04001610 if (unlikely(!name)) {
1611 static const struct qstr anon = QSTR_INIT("/", 1);
1612 name = &anon;
1613 dname = dentry->d_iname;
1614 } else if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001615 size_t size = offsetof(struct external_name, name[1]);
Vladimir Davydov5d097052016-01-14 15:18:21 -08001616 struct external_name *p = kmalloc(size + name->len,
1617 GFP_KERNEL_ACCOUNT);
Al Viro8d85b482014-09-29 14:54:27 -04001618 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 kmem_cache_free(dentry_cache, dentry);
1620 return NULL;
1621 }
Al Viro8d85b482014-09-29 14:54:27 -04001622 atomic_set(&p->u.count, 1);
1623 dname = p->name;
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -08001624 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1625 kasan_unpoison_shadow(dname,
1626 round_up(name->len + 1, sizeof(unsigned long)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 } else {
1628 dname = dentry->d_iname;
1629 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
1631 dentry->d_name.len = name->len;
1632 dentry->d_name.hash = name->hash;
1633 memcpy(dname, name->name, name->len);
1634 dname[name->len] = 0;
1635
Linus Torvalds6326c712012-05-21 16:14:04 -07001636 /* Make sure we always see the terminating NUL character */
1637 smp_wmb();
1638 dentry->d_name.name = dname;
1639
Waiman Long98474232013-08-28 18:24:59 -07001640 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001641 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001643 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001645 dentry->d_parent = dentry;
1646 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 dentry->d_op = NULL;
1648 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001649 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 INIT_LIST_HEAD(&dentry->d_lru);
1651 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001652 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1653 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001654 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Nick Piggin3e880fb2011-01-07 17:49:19 +11001656 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001657
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 return dentry;
1659}
Al Viroa4464db2011-07-07 15:03:58 -04001660
1661/**
1662 * d_alloc - allocate a dcache entry
1663 * @parent: parent of entry to allocate
1664 * @name: qstr of the name
1665 *
1666 * Allocates a dentry. It returns %NULL if there is insufficient memory
1667 * available. On a success the dentry is returned. The name passed in is
1668 * copied and the copy passed in may be reused after this call.
1669 */
1670struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1671{
1672 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1673 if (!dentry)
1674 return NULL;
Al Viro3d56c252016-06-07 21:26:55 -04001675 dentry->d_flags |= DCACHE_RCUACCESS;
Al Viroa4464db2011-07-07 15:03:58 -04001676 spin_lock(&parent->d_lock);
1677 /*
1678 * don't need child lock because it is not subject
1679 * to concurrency here
1680 */
1681 __dget_dlock(parent);
1682 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001683 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001684 spin_unlock(&parent->d_lock);
1685
1686 return dentry;
1687}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001688EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689
Al Viroba65dc52016-06-10 11:32:47 -04001690struct dentry *d_alloc_cursor(struct dentry * parent)
1691{
1692 struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
1693 if (dentry) {
1694 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1695 dentry->d_parent = dget(parent);
1696 }
1697 return dentry;
1698}
1699
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001700/**
1701 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1702 * @sb: the superblock
1703 * @name: qstr of the name
1704 *
1705 * For a filesystem that just pins its dentries in memory and never
1706 * performs lookups at all, return an unhashed IS_ROOT dentry.
1707 */
Nick Piggin4b936882011-01-07 17:50:07 +11001708struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1709{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001710 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001711}
1712EXPORT_SYMBOL(d_alloc_pseudo);
1713
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1715{
1716 struct qstr q;
1717
1718 q.name = name;
George Spelvinfcfd2fb2016-05-20 08:41:37 -04001719 q.hash_len = hashlen_string(name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 return d_alloc(parent, &q);
1721}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001722EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
Nick Pigginfb045ad2011-01-07 17:49:55 +11001724void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1725{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001726 WARN_ON_ONCE(dentry->d_op);
1727 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001728 DCACHE_OP_COMPARE |
1729 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001730 DCACHE_OP_WEAK_REVALIDATE |
David Howells4bacc9c2015-06-18 14:32:31 +01001731 DCACHE_OP_DELETE |
Miklos Szeredid101a122016-03-26 16:14:37 -04001732 DCACHE_OP_SELECT_INODE |
1733 DCACHE_OP_REAL));
Nick Pigginfb045ad2011-01-07 17:49:55 +11001734 dentry->d_op = op;
1735 if (!op)
1736 return;
1737 if (op->d_hash)
1738 dentry->d_flags |= DCACHE_OP_HASH;
1739 if (op->d_compare)
1740 dentry->d_flags |= DCACHE_OP_COMPARE;
1741 if (op->d_revalidate)
1742 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001743 if (op->d_weak_revalidate)
1744 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001745 if (op->d_delete)
1746 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001747 if (op->d_prune)
1748 dentry->d_flags |= DCACHE_OP_PRUNE;
David Howells4bacc9c2015-06-18 14:32:31 +01001749 if (op->d_select_inode)
1750 dentry->d_flags |= DCACHE_OP_SELECT_INODE;
Miklos Szeredid101a122016-03-26 16:14:37 -04001751 if (op->d_real)
1752 dentry->d_flags |= DCACHE_OP_REAL;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001753
1754}
1755EXPORT_SYMBOL(d_set_d_op);
1756
David Howellsdf1a0852015-01-29 12:02:28 +00001757
1758/*
1759 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1760 * @dentry - The dentry to mark
1761 *
1762 * Mark a dentry as falling through to the lower layer (as set with
1763 * d_pin_lower()). This flag may be recorded on the medium.
1764 */
1765void d_set_fallthru(struct dentry *dentry)
1766{
1767 spin_lock(&dentry->d_lock);
1768 dentry->d_flags |= DCACHE_FALLTHRU;
1769 spin_unlock(&dentry->d_lock);
1770}
1771EXPORT_SYMBOL(d_set_fallthru);
1772
David Howellsb18825a2013-09-12 19:22:53 +01001773static unsigned d_flags_for_inode(struct inode *inode)
1774{
David Howells44bdb5e2015-01-29 12:02:29 +00001775 unsigned add_flags = DCACHE_REGULAR_TYPE;
David Howellsb18825a2013-09-12 19:22:53 +01001776
1777 if (!inode)
1778 return DCACHE_MISS_TYPE;
1779
1780 if (S_ISDIR(inode->i_mode)) {
1781 add_flags = DCACHE_DIRECTORY_TYPE;
1782 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1783 if (unlikely(!inode->i_op->lookup))
1784 add_flags = DCACHE_AUTODIR_TYPE;
1785 else
1786 inode->i_opflags |= IOP_LOOKUP;
1787 }
David Howells44bdb5e2015-01-29 12:02:29 +00001788 goto type_determined;
David Howellsb18825a2013-09-12 19:22:53 +01001789 }
1790
David Howells44bdb5e2015-01-29 12:02:29 +00001791 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
Al Viro6b255392015-11-17 10:20:54 -05001792 if (unlikely(inode->i_op->get_link)) {
David Howells44bdb5e2015-01-29 12:02:29 +00001793 add_flags = DCACHE_SYMLINK_TYPE;
1794 goto type_determined;
1795 }
1796 inode->i_opflags |= IOP_NOFOLLOW;
1797 }
1798
1799 if (unlikely(!S_ISREG(inode->i_mode)))
1800 add_flags = DCACHE_SPECIAL_TYPE;
1801
1802type_determined:
David Howellsb18825a2013-09-12 19:22:53 +01001803 if (unlikely(IS_AUTOMOUNT(inode)))
1804 add_flags |= DCACHE_NEED_AUTOMOUNT;
1805 return add_flags;
1806}
1807
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001808static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1809{
David Howellsb18825a2013-09-12 19:22:53 +01001810 unsigned add_flags = d_flags_for_inode(inode);
Al Viro85c7f812016-04-14 19:52:13 -04001811 WARN_ON(d_in_lookup(dentry));
David Howellsb18825a2013-09-12 19:22:53 +01001812
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001813 spin_lock(&dentry->d_lock);
Al Virode689f52016-03-09 18:05:42 -05001814 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
Al Viroa528aca2016-02-29 12:12:46 -05001815 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +00001816 __d_set_inode_and_type(dentry, inode, add_flags);
Al Viroa528aca2016-02-29 12:12:46 -05001817 raw_write_seqcount_end(&dentry->d_seq);
Al Virode689f52016-03-09 18:05:42 -05001818 __fsnotify_d_instantiate(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001819 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001820}
1821
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822/**
1823 * d_instantiate - fill in inode information for a dentry
1824 * @entry: dentry to complete
1825 * @inode: inode to attach to this dentry
1826 *
1827 * Fill in inode information in the entry.
1828 *
1829 * This turns negative dentries into productive full members
1830 * of society.
1831 *
1832 * NOTE! This assumes that the inode count has been incremented
1833 * (or otherwise set) by the caller to indicate that it is now
1834 * in use by the dcache.
1835 */
1836
1837void d_instantiate(struct dentry *entry, struct inode * inode)
1838{
Al Viro946e51f2014-10-26 19:19:16 -04001839 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Al Virode689f52016-03-09 18:05:42 -05001840 if (inode) {
Al Virob9680912016-04-11 00:53:26 -04001841 security_d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001842 spin_lock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001843 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001844 spin_unlock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001845 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001847EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848
1849/**
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001850 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1851 * @entry: dentry to complete
1852 * @inode: inode to attach to this dentry
1853 *
1854 * Fill in inode information in the entry. If a directory alias is found, then
1855 * return an error (and drop inode). Together with d_materialise_unique() this
1856 * guarantees that a directory inode may never have more than one alias.
1857 */
1858int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1859{
Al Viro946e51f2014-10-26 19:19:16 -04001860 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001861
Al Virob9680912016-04-11 00:53:26 -04001862 security_d_instantiate(entry, inode);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001863 spin_lock(&inode->i_lock);
1864 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1865 spin_unlock(&inode->i_lock);
1866 iput(inode);
1867 return -EBUSY;
1868 }
1869 __d_instantiate(entry, inode);
1870 spin_unlock(&inode->i_lock);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001871
1872 return 0;
1873}
1874EXPORT_SYMBOL(d_instantiate_no_diralias);
1875
Al Viroadc0e912012-01-08 16:49:21 -05001876struct dentry *d_make_root(struct inode *root_inode)
1877{
1878 struct dentry *res = NULL;
1879
1880 if (root_inode) {
Al Viro798434b2016-03-24 20:38:43 -04001881 res = __d_alloc(root_inode->i_sb, NULL);
Al Viroadc0e912012-01-08 16:49:21 -05001882 if (res)
1883 d_instantiate(res, root_inode);
1884 else
1885 iput(root_inode);
1886 }
1887 return res;
1888}
1889EXPORT_SYMBOL(d_make_root);
1890
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001891static struct dentry * __d_find_any_alias(struct inode *inode)
1892{
1893 struct dentry *alias;
1894
Al Virob3d9b7a2012-06-09 13:51:19 -04001895 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001896 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001897 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001898 __dget(alias);
1899 return alias;
1900}
1901
Sage Weil46f72b32012-01-10 09:04:37 -08001902/**
1903 * d_find_any_alias - find any alias for a given inode
1904 * @inode: inode to find an alias for
1905 *
1906 * If any aliases exist for the given inode, take and return a
1907 * reference for one of them. If no aliases exist, return %NULL.
1908 */
1909struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001910{
1911 struct dentry *de;
1912
1913 spin_lock(&inode->i_lock);
1914 de = __d_find_any_alias(inode);
1915 spin_unlock(&inode->i_lock);
1916 return de;
1917}
Sage Weil46f72b32012-01-10 09:04:37 -08001918EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001919
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001920static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001921{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001922 struct dentry *tmp;
1923 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001924 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001925
1926 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001927 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001928 if (IS_ERR(inode))
1929 return ERR_CAST(inode);
1930
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001931 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001932 if (res)
1933 goto out_iput;
1934
Al Viro798434b2016-03-24 20:38:43 -04001935 tmp = __d_alloc(inode->i_sb, NULL);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001936 if (!tmp) {
1937 res = ERR_PTR(-ENOMEM);
1938 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001939 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001940
Al Virob9680912016-04-11 00:53:26 -04001941 security_d_instantiate(tmp, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001942 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001943 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001944 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001945 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001946 dput(tmp);
1947 goto out_iput;
1948 }
1949
1950 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001951 add_flags = d_flags_for_inode(inode);
1952
1953 if (disconnected)
1954 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001955
Christoph Hellwig9308a612008-08-11 15:49:12 +02001956 spin_lock(&tmp->d_lock);
David Howells4bf46a22015-03-05 14:09:22 +00001957 __d_set_inode_and_type(tmp, inode, add_flags);
Al Viro946e51f2014-10-26 19:19:16 -04001958 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001959 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001960 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001961 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001962 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001963 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001964
Christoph Hellwig9308a612008-08-11 15:49:12 +02001965 return tmp;
1966
1967 out_iput:
1968 iput(inode);
1969 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001970}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001971
1972/**
1973 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1974 * @inode: inode to allocate the dentry for
1975 *
1976 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1977 * similar open by handle operations. The returned dentry may be anonymous,
1978 * or may have a full name (if the inode was already in the cache).
1979 *
1980 * When called on a directory inode, we must ensure that the inode only ever
1981 * has one dentry. If a dentry is found, that is returned instead of
1982 * allocating a new one.
1983 *
1984 * On successful return, the reference to the inode has been transferred
1985 * to the dentry. In case of an error the reference on the inode is released.
1986 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1987 * be passed in and the error will be propagated to the return value,
1988 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1989 */
1990struct dentry *d_obtain_alias(struct inode *inode)
1991{
1992 return __d_obtain_alias(inode, 1);
1993}
Benny Halevyadc48722009-02-27 14:02:59 -08001994EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995
1996/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001997 * d_obtain_root - find or allocate a dentry for a given inode
1998 * @inode: inode to allocate the dentry for
1999 *
2000 * Obtain an IS_ROOT dentry for the root of a filesystem.
2001 *
2002 * We must ensure that directory inodes only ever have one dentry. If a
2003 * dentry is found, that is returned instead of allocating a new one.
2004 *
2005 * On successful return, the reference to the inode has been transferred
2006 * to the dentry. In case of an error the reference on the inode is
2007 * released. A %NULL or IS_ERR inode may be passed in and will be the
2008 * error will be propagate to the return value, with a %NULL @inode
2009 * replaced by ERR_PTR(-ESTALE).
2010 */
2011struct dentry *d_obtain_root(struct inode *inode)
2012{
2013 return __d_obtain_alias(inode, 0);
2014}
2015EXPORT_SYMBOL(d_obtain_root);
2016
2017/**
Barry Naujok94035402008-05-21 16:50:46 +10002018 * d_add_ci - lookup or allocate new dentry with case-exact name
2019 * @inode: the inode case-insensitive lookup has found
2020 * @dentry: the negative dentry that was passed to the parent's lookup func
2021 * @name: the case-exact name to be associated with the returned dentry
2022 *
2023 * This is to avoid filling the dcache with case-insensitive names to the
2024 * same inode, only the actual correct case is stored in the dcache for
2025 * case-insensitive filesystems.
2026 *
2027 * For a case-insensitive lookup match and if the the case-exact dentry
2028 * already exists in in the dcache, use it and return it.
2029 *
2030 * If no entry exists with the exact case name, allocate new dentry with
2031 * the exact case, and return the spliced entry.
2032 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02002033struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10002034 struct qstr *name)
2035{
Al Virod9171b92016-04-15 03:33:13 -04002036 struct dentry *found, *res;
Barry Naujok94035402008-05-21 16:50:46 +10002037
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002038 /*
2039 * First check if a dentry matching the name already exists,
2040 * if not go ahead and create it now.
2041 */
Barry Naujok94035402008-05-21 16:50:46 +10002042 found = d_hash_and_lookup(dentry->d_parent, name);
Al Virod9171b92016-04-15 03:33:13 -04002043 if (found) {
2044 iput(inode);
2045 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002046 }
Al Virod9171b92016-04-15 03:33:13 -04002047 if (d_in_lookup(dentry)) {
2048 found = d_alloc_parallel(dentry->d_parent, name,
2049 dentry->d_wait);
2050 if (IS_ERR(found) || !d_in_lookup(found)) {
2051 iput(inode);
2052 return found;
2053 }
2054 } else {
2055 found = d_alloc(dentry->d_parent, name);
2056 if (!found) {
2057 iput(inode);
2058 return ERR_PTR(-ENOMEM);
2059 }
2060 }
2061 res = d_splice_alias(inode, found);
2062 if (res) {
2063 dput(found);
2064 return res;
2065 }
Al Viro4f522a22013-02-11 23:20:37 -05002066 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002067}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002068EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002070/*
2071 * Do the slow-case of the dentry name compare.
2072 *
2073 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07002074 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002075 * filesystem can rely on them, and can use the 'name' and
2076 * 'len' information without worrying about walking off the
2077 * end of memory etc.
2078 *
2079 * Thus the read_seqcount_retry() and the "duplicate" info
2080 * in arguments (the low-level filesystem should not look
2081 * at the dentry inode or name contents directly, since
2082 * rename can change them while we're in RCU mode).
2083 */
2084enum slow_d_compare {
2085 D_COMP_OK,
2086 D_COMP_NOMATCH,
2087 D_COMP_SEQRETRY,
2088};
2089
2090static noinline enum slow_d_compare slow_dentry_cmp(
2091 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002092 struct dentry *dentry,
2093 unsigned int seq,
2094 const struct qstr *name)
2095{
2096 int tlen = dentry->d_name.len;
2097 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002098
2099 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2100 cpu_relax();
2101 return D_COMP_SEQRETRY;
2102 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002103 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002104 return D_COMP_NOMATCH;
2105 return D_COMP_OK;
2106}
2107
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002109 * __d_lookup_rcu - search for a dentry (racy, store-free)
2110 * @parent: parent dentry
2111 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002112 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002113 * Returns: dentry, or NULL
2114 *
2115 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2116 * resolution (store-free path walking) design described in
2117 * Documentation/filesystems/path-lookup.txt.
2118 *
2119 * This is not to be used outside core vfs.
2120 *
2121 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2122 * held, and rcu_read_lock held. The returned dentry must not be stored into
2123 * without taking d_lock and checking d_seq sequence count against @seq
2124 * returned here.
2125 *
Linus Torvalds15570082013-09-02 11:38:06 -07002126 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002127 * function.
2128 *
2129 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2130 * the returned dentry, so long as its parent's seqlock is checked after the
2131 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2132 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002133 *
2134 * NOTE! The caller *has* to check the resulting dentry against the sequence
2135 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002136 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002137struct dentry *__d_lookup_rcu(const struct dentry *parent,
2138 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002139 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002140{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002141 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002142 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002143 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002144 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002145 struct dentry *dentry;
2146
2147 /*
2148 * Note: There is significant duplication with __d_lookup_rcu which is
2149 * required to prevent single threaded performance regressions
2150 * especially on architectures where smp_rmb (in seqcounts) are costly.
2151 * Keep the two functions in sync.
2152 */
2153
2154 /*
2155 * The hash list is protected using RCU.
2156 *
2157 * Carefully use d_seq when comparing a candidate dentry, to avoid
2158 * races with d_move().
2159 *
2160 * It is possible that concurrent renames can mess up our list
2161 * walk here and result in missing our dentry, resulting in the
2162 * false-negative result. d_lookup() protects against concurrent
2163 * renames using rename_lock seqlock.
2164 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002165 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002166 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002167 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002168 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002169
Nick Piggin31e6b012011-01-07 17:49:52 +11002170seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002171 /*
2172 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002173 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002174 *
2175 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002176 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002177 *
2178 * NOTE! We do a "raw" seqcount_begin here. That means that
2179 * we don't wait for the sequence count to stabilize if it
2180 * is in the middle of a sequence change. If we do the slow
2181 * dentry compare, we will do seqretries until it is stable,
2182 * and if we end up with a successful lookup, we actually
2183 * want to exit RCU lookup anyway.
2184 */
2185 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002186 if (dentry->d_parent != parent)
2187 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002188 if (d_unhashed(dentry))
2189 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002190
2191 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002192 if (dentry->d_name.hash != hashlen_hash(hashlen))
2193 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002194 *seqp = seq;
2195 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002196 case D_COMP_OK:
2197 return dentry;
2198 case D_COMP_NOMATCH:
2199 continue;
2200 default:
2201 goto seqretry;
2202 }
2203 }
2204
Linus Torvalds26fe5752012-05-10 13:14:12 -07002205 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002206 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002207 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002208 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002209 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002210 }
2211 return NULL;
2212}
2213
2214/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 * d_lookup - search for a dentry
2216 * @parent: parent dentry
2217 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002218 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002220 * d_lookup searches the children of the parent dentry for the name in
2221 * question. If the dentry is found its reference count is incremented and the
2222 * dentry is returned. The caller must use dput to free the entry when it has
2223 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 */
Al Viroda2d8452013-01-24 18:29:34 -05002225struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226{
Nick Piggin31e6b012011-01-07 17:49:52 +11002227 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002228 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
Daeseok Younb8314f92014-08-11 11:46:53 +09002230 do {
2231 seq = read_seqbegin(&rename_lock);
2232 dentry = __d_lookup(parent, name);
2233 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 break;
2235 } while (read_seqretry(&rename_lock, seq));
2236 return dentry;
2237}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002238EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239
Nick Piggin31e6b012011-01-07 17:49:52 +11002240/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002241 * __d_lookup - search for a dentry (racy)
2242 * @parent: parent dentry
2243 * @name: qstr of name we wish to find
2244 * Returns: dentry, or NULL
2245 *
2246 * __d_lookup is like d_lookup, however it may (rarely) return a
2247 * false-negative result due to unrelated rename activity.
2248 *
2249 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2250 * however it must be used carefully, eg. with a following d_lookup in
2251 * the case of failure.
2252 *
2253 * __d_lookup callers must be commented.
2254 */
Al Viroa713ca22013-01-24 18:27:00 -05002255struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256{
2257 unsigned int len = name->len;
2258 unsigned int hash = name->hash;
2259 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002260 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002261 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002262 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002263 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
Nick Pigginb04f7842010-08-18 04:37:34 +10002265 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002266 * Note: There is significant duplication with __d_lookup_rcu which is
2267 * required to prevent single threaded performance regressions
2268 * especially on architectures where smp_rmb (in seqcounts) are costly.
2269 * Keep the two functions in sync.
2270 */
2271
2272 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002273 * The hash list is protected using RCU.
2274 *
2275 * Take d_lock when comparing a candidate dentry, to avoid races
2276 * with d_move().
2277 *
2278 * It is possible that concurrent renames can mess up our list
2279 * walk here and result in missing our dentry, resulting in the
2280 * false-negative result. d_lookup() protects against concurrent
2281 * renames using rename_lock seqlock.
2282 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002283 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002284 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 rcu_read_lock();
2286
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002287 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 if (dentry->d_name.hash != hash)
2290 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
2292 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 if (dentry->d_parent != parent)
2294 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002295 if (d_unhashed(dentry))
2296 goto next;
2297
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 /*
2299 * It is safe to compare names since d_move() cannot
2300 * change the qstr (protected by d_lock).
2301 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002302 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002303 int tlen = dentry->d_name.len;
2304 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002305 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 goto next;
2307 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002308 if (dentry->d_name.len != len)
2309 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002310 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 goto next;
2312 }
2313
Waiman Long98474232013-08-28 18:24:59 -07002314 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002315 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 spin_unlock(&dentry->d_lock);
2317 break;
2318next:
2319 spin_unlock(&dentry->d_lock);
2320 }
2321 rcu_read_unlock();
2322
2323 return found;
2324}
2325
2326/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002327 * d_hash_and_lookup - hash the qstr then search for a dentry
2328 * @dir: Directory to search in
2329 * @name: qstr of name we wish to find
2330 *
Al Viro4f522a22013-02-11 23:20:37 -05002331 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002332 */
2333struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2334{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002335 /*
2336 * Check for a fs-specific hash function. Note that we must
2337 * calculate the standard hash first, as the d_op->d_hash()
2338 * routine may choose to leave the hash value unchanged.
2339 */
2340 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002341 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002342 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002343 if (unlikely(err < 0))
2344 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002345 }
Al Viro4f522a22013-02-11 23:20:37 -05002346 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002347}
Al Viro4f522a22013-02-11 23:20:37 -05002348EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002349
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350/*
2351 * When a file is deleted, we have two options:
2352 * - turn this dentry into a negative dentry
2353 * - unhash this dentry and free it.
2354 *
2355 * Usually, we want to just turn this into
2356 * a negative dentry, but if anybody else is
2357 * currently using the dentry or the inode
2358 * we can't do that and we fall back on removing
2359 * it from the hash queues and waiting for
2360 * it to be deleted later when it has no users
2361 */
2362
2363/**
2364 * d_delete - delete a dentry
2365 * @dentry: The dentry to delete
2366 *
2367 * Turn the dentry into a negative dentry if possible, otherwise
2368 * remove it from the hash queues so it can be deleted later
2369 */
2370
2371void d_delete(struct dentry * dentry)
2372{
Nick Piggin873feea2011-01-07 17:50:06 +11002373 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002374 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 /*
2376 * Are we the only user?
2377 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002378again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002380 inode = dentry->d_inode;
2381 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002382 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002383 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002384 spin_unlock(&dentry->d_lock);
2385 cpu_relax();
2386 goto again;
2387 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002388 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002389 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002390 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 return;
2392 }
2393
2394 if (!d_unhashed(dentry))
2395 __d_drop(dentry);
2396
2397 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002398
2399 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002401EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002403static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002405 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002406 hlist_bl_lock(b);
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002407 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002408 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409}
2410
David Howells770bfad2006-08-22 20:06:07 -04002411static void _d_rehash(struct dentry * entry)
2412{
2413 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2414}
2415
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416/**
2417 * d_rehash - add an entry back to the hash
2418 * @entry: dentry to add to the hash
2419 *
2420 * Adds a dentry to the hash according to its name.
2421 */
2422
2423void d_rehash(struct dentry * entry)
2424{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002426 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002429EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
Al Viro84e710d2016-04-15 00:58:55 -04002431static inline unsigned start_dir_add(struct inode *dir)
2432{
2433
2434 for (;;) {
2435 unsigned n = dir->i_dir_seq;
2436 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2437 return n;
2438 cpu_relax();
2439 }
2440}
2441
2442static inline void end_dir_add(struct inode *dir, unsigned n)
2443{
2444 smp_store_release(&dir->i_dir_seq, n + 2);
2445}
2446
Al Virod9171b92016-04-15 03:33:13 -04002447static void d_wait_lookup(struct dentry *dentry)
2448{
2449 if (d_in_lookup(dentry)) {
2450 DECLARE_WAITQUEUE(wait, current);
2451 add_wait_queue(dentry->d_wait, &wait);
2452 do {
2453 set_current_state(TASK_UNINTERRUPTIBLE);
2454 spin_unlock(&dentry->d_lock);
2455 schedule();
2456 spin_lock(&dentry->d_lock);
2457 } while (d_in_lookup(dentry));
2458 }
2459}
2460
Al Viro94bdd652016-04-15 02:42:04 -04002461struct dentry *d_alloc_parallel(struct dentry *parent,
Al Virod9171b92016-04-15 03:33:13 -04002462 const struct qstr *name,
2463 wait_queue_head_t *wq)
Al Viro94bdd652016-04-15 02:42:04 -04002464{
2465 unsigned int len = name->len;
2466 unsigned int hash = name->hash;
2467 const unsigned char *str = name->name;
2468 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2469 struct hlist_bl_node *node;
2470 struct dentry *new = d_alloc(parent, name);
2471 struct dentry *dentry;
2472 unsigned seq, r_seq, d_seq;
2473
2474 if (unlikely(!new))
2475 return ERR_PTR(-ENOMEM);
2476
2477retry:
2478 rcu_read_lock();
2479 seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1;
2480 r_seq = read_seqbegin(&rename_lock);
2481 dentry = __d_lookup_rcu(parent, name, &d_seq);
2482 if (unlikely(dentry)) {
2483 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2484 rcu_read_unlock();
2485 goto retry;
2486 }
2487 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2488 rcu_read_unlock();
2489 dput(dentry);
2490 goto retry;
2491 }
2492 rcu_read_unlock();
2493 dput(new);
2494 return dentry;
2495 }
2496 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2497 rcu_read_unlock();
2498 goto retry;
2499 }
2500 hlist_bl_lock(b);
2501 if (unlikely(parent->d_inode->i_dir_seq != seq)) {
2502 hlist_bl_unlock(b);
2503 rcu_read_unlock();
2504 goto retry;
2505 }
Al Viro94bdd652016-04-15 02:42:04 -04002506 /*
2507 * No changes for the parent since the beginning of d_lookup().
2508 * Since all removals from the chain happen with hlist_bl_lock(),
2509 * any potential in-lookup matches are going to stay here until
2510 * we unlock the chain. All fields are stable in everything
2511 * we encounter.
2512 */
2513 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2514 if (dentry->d_name.hash != hash)
2515 continue;
2516 if (dentry->d_parent != parent)
2517 continue;
Al Viro94bdd652016-04-15 02:42:04 -04002518 if (parent->d_flags & DCACHE_OP_COMPARE) {
2519 int tlen = dentry->d_name.len;
2520 const char *tname = dentry->d_name.name;
2521 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2522 continue;
2523 } else {
2524 if (dentry->d_name.len != len)
2525 continue;
2526 if (dentry_cmp(dentry, str, len))
2527 continue;
2528 }
Al Viro94bdd652016-04-15 02:42:04 -04002529 hlist_bl_unlock(b);
Al Viroe7d6ef92016-06-20 01:35:59 -04002530 /* now we can try to grab a reference */
2531 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2532 rcu_read_unlock();
2533 goto retry;
2534 }
2535
2536 rcu_read_unlock();
2537 /*
2538 * somebody is likely to be still doing lookup for it;
2539 * wait for them to finish
2540 */
Al Virod9171b92016-04-15 03:33:13 -04002541 spin_lock(&dentry->d_lock);
2542 d_wait_lookup(dentry);
2543 /*
2544 * it's not in-lookup anymore; in principle we should repeat
2545 * everything from dcache lookup, but it's likely to be what
2546 * d_lookup() would've found anyway. If it is, just return it;
2547 * otherwise we really have to repeat the whole thing.
2548 */
2549 if (unlikely(dentry->d_name.hash != hash))
2550 goto mismatch;
2551 if (unlikely(dentry->d_parent != parent))
2552 goto mismatch;
2553 if (unlikely(d_unhashed(dentry)))
2554 goto mismatch;
2555 if (parent->d_flags & DCACHE_OP_COMPARE) {
2556 int tlen = dentry->d_name.len;
2557 const char *tname = dentry->d_name.name;
2558 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2559 goto mismatch;
2560 } else {
2561 if (unlikely(dentry->d_name.len != len))
2562 goto mismatch;
2563 if (unlikely(dentry_cmp(dentry, str, len)))
2564 goto mismatch;
2565 }
2566 /* OK, it *is* a hashed match; return it */
2567 spin_unlock(&dentry->d_lock);
Al Viro94bdd652016-04-15 02:42:04 -04002568 dput(new);
2569 return dentry;
2570 }
Al Viroe7d6ef92016-06-20 01:35:59 -04002571 rcu_read_unlock();
Al Viro94bdd652016-04-15 02:42:04 -04002572 /* we can't take ->d_lock here; it's OK, though. */
2573 new->d_flags |= DCACHE_PAR_LOOKUP;
Al Virod9171b92016-04-15 03:33:13 -04002574 new->d_wait = wq;
Al Viro94bdd652016-04-15 02:42:04 -04002575 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2576 hlist_bl_unlock(b);
2577 return new;
Al Virod9171b92016-04-15 03:33:13 -04002578mismatch:
2579 spin_unlock(&dentry->d_lock);
2580 dput(dentry);
2581 goto retry;
Al Viro94bdd652016-04-15 02:42:04 -04002582}
2583EXPORT_SYMBOL(d_alloc_parallel);
2584
Al Viro85c7f812016-04-14 19:52:13 -04002585void __d_lookup_done(struct dentry *dentry)
2586{
Al Viro94bdd652016-04-15 02:42:04 -04002587 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2588 dentry->d_name.hash);
2589 hlist_bl_lock(b);
Al Viro85c7f812016-04-14 19:52:13 -04002590 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
Al Viro94bdd652016-04-15 02:42:04 -04002591 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
Al Virod9171b92016-04-15 03:33:13 -04002592 wake_up_all(dentry->d_wait);
2593 dentry->d_wait = NULL;
Al Viro94bdd652016-04-15 02:42:04 -04002594 hlist_bl_unlock(b);
2595 INIT_HLIST_NODE(&dentry->d_u.d_alias);
Al Virod9171b92016-04-15 03:33:13 -04002596 INIT_LIST_HEAD(&dentry->d_lru);
Al Viro85c7f812016-04-14 19:52:13 -04002597}
2598EXPORT_SYMBOL(__d_lookup_done);
Al Viroed782b52016-03-09 19:52:39 -05002599
2600/* inode->i_lock held if inode is non-NULL */
2601
2602static inline void __d_add(struct dentry *dentry, struct inode *inode)
2603{
Al Viro84e710d2016-04-15 00:58:55 -04002604 struct inode *dir = NULL;
2605 unsigned n;
Al Viro0568d702016-04-14 19:40:56 -04002606 spin_lock(&dentry->d_lock);
Al Viro84e710d2016-04-15 00:58:55 -04002607 if (unlikely(d_in_lookup(dentry))) {
2608 dir = dentry->d_parent->d_inode;
2609 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002610 __d_lookup_done(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002611 }
Al Viroed782b52016-03-09 19:52:39 -05002612 if (inode) {
Al Viro0568d702016-04-14 19:40:56 -04002613 unsigned add_flags = d_flags_for_inode(inode);
2614 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2615 raw_write_seqcount_begin(&dentry->d_seq);
2616 __d_set_inode_and_type(dentry, inode, add_flags);
2617 raw_write_seqcount_end(&dentry->d_seq);
2618 __fsnotify_d_instantiate(dentry);
Al Viroed782b52016-03-09 19:52:39 -05002619 }
Al Viro0568d702016-04-14 19:40:56 -04002620 _d_rehash(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002621 if (dir)
2622 end_dir_add(dir, n);
Al Viro0568d702016-04-14 19:40:56 -04002623 spin_unlock(&dentry->d_lock);
2624 if (inode)
2625 spin_unlock(&inode->i_lock);
Al Viroed782b52016-03-09 19:52:39 -05002626}
2627
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002628/**
Al Viro34d0d192016-03-08 21:01:03 -05002629 * d_add - add dentry to hash queues
2630 * @entry: dentry to add
2631 * @inode: The inode to attach to this dentry
2632 *
2633 * This adds the entry to the hash queues and initializes @inode.
2634 * The entry was actually filled in earlier during d_alloc().
2635 */
2636
2637void d_add(struct dentry *entry, struct inode *inode)
2638{
Al Virob9680912016-04-11 00:53:26 -04002639 if (inode) {
2640 security_d_instantiate(entry, inode);
Al Viroed782b52016-03-09 19:52:39 -05002641 spin_lock(&inode->i_lock);
Al Virob9680912016-04-11 00:53:26 -04002642 }
Al Viroed782b52016-03-09 19:52:39 -05002643 __d_add(entry, inode);
Al Viro34d0d192016-03-08 21:01:03 -05002644}
2645EXPORT_SYMBOL(d_add);
2646
2647/**
Al Viro668d0cd2016-03-08 12:44:17 -05002648 * d_exact_alias - find and hash an exact unhashed alias
2649 * @entry: dentry to add
2650 * @inode: The inode to go with this dentry
2651 *
2652 * If an unhashed dentry with the same name/parent and desired
2653 * inode already exists, hash and return it. Otherwise, return
2654 * NULL.
2655 *
2656 * Parent directory should be locked.
2657 */
2658struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2659{
2660 struct dentry *alias;
2661 int len = entry->d_name.len;
2662 const char *name = entry->d_name.name;
2663 unsigned int hash = entry->d_name.hash;
2664
2665 spin_lock(&inode->i_lock);
2666 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2667 /*
2668 * Don't need alias->d_lock here, because aliases with
2669 * d_parent == entry->d_parent are not subject to name or
2670 * parent changes, because the parent inode i_mutex is held.
2671 */
2672 if (alias->d_name.hash != hash)
2673 continue;
2674 if (alias->d_parent != entry->d_parent)
2675 continue;
2676 if (alias->d_name.len != len)
2677 continue;
2678 if (dentry_cmp(alias, name, len))
2679 continue;
2680 spin_lock(&alias->d_lock);
2681 if (!d_unhashed(alias)) {
2682 spin_unlock(&alias->d_lock);
2683 alias = NULL;
2684 } else {
2685 __dget_dlock(alias);
2686 _d_rehash(alias);
2687 spin_unlock(&alias->d_lock);
2688 }
2689 spin_unlock(&inode->i_lock);
2690 return alias;
2691 }
2692 spin_unlock(&inode->i_lock);
2693 return NULL;
2694}
2695EXPORT_SYMBOL(d_exact_alias);
2696
2697/**
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002698 * dentry_update_name_case - update case insensitive dentry with a new name
2699 * @dentry: dentry to be updated
2700 * @name: new name
2701 *
2702 * Update a case insensitive dentry with new case of name.
2703 *
2704 * dentry must have been returned by d_lookup with name @name. Old and new
2705 * name lengths must match (ie. no d_compare which allows mismatched name
2706 * lengths).
2707 *
2708 * Parent inode i_mutex must be held over d_lookup and into this call (to
2709 * keep renames and concurrent inserts, and readdir(2) away).
2710 */
2711void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2712{
Al Viro59551022016-01-22 15:40:57 -05002713 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002714 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2715
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002716 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002717 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002718 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002719 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002720 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002721}
2722EXPORT_SYMBOL(dentry_update_name_case);
2723
Al Viro8d85b482014-09-29 14:54:27 -04002724static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725{
Al Viro8d85b482014-09-29 14:54:27 -04002726 if (unlikely(dname_external(target))) {
2727 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 /*
2729 * Both external: swap the pointers
2730 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002731 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 } else {
2733 /*
2734 * dentry:internal, target:external. Steal target's
2735 * storage and make target internal.
2736 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002737 memcpy(target->d_iname, dentry->d_name.name,
2738 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 dentry->d_name.name = target->d_name.name;
2740 target->d_name.name = target->d_iname;
2741 }
2742 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002743 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 /*
2745 * dentry:external, target:internal. Give dentry's
2746 * storage to target and make dentry internal
2747 */
2748 memcpy(dentry->d_iname, target->d_name.name,
2749 target->d_name.len + 1);
2750 target->d_name.name = dentry->d_name.name;
2751 dentry->d_name.name = dentry->d_iname;
2752 } else {
2753 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002754 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002756 unsigned int i;
2757 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002758 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2759 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002760 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2761 swap(((long *) &dentry->d_iname)[i],
2762 ((long *) &target->d_iname)[i]);
2763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764 }
2765 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002766 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767}
2768
Al Viro8d85b482014-09-29 14:54:27 -04002769static void copy_name(struct dentry *dentry, struct dentry *target)
2770{
2771 struct external_name *old_name = NULL;
2772 if (unlikely(dname_external(dentry)))
2773 old_name = external_name(dentry);
2774 if (unlikely(dname_external(target))) {
2775 atomic_inc(&external_name(target)->u.count);
2776 dentry->d_name = target->d_name;
2777 } else {
2778 memcpy(dentry->d_iname, target->d_name.name,
2779 target->d_name.len + 1);
2780 dentry->d_name.name = dentry->d_iname;
2781 dentry->d_name.hash_len = target->d_name.hash_len;
2782 }
2783 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2784 kfree_rcu(old_name, u.head);
2785}
2786
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002787static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2788{
2789 /*
2790 * XXXX: do we really need to take target->d_lock?
2791 */
2792 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2793 spin_lock(&target->d_parent->d_lock);
2794 else {
2795 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2796 spin_lock(&dentry->d_parent->d_lock);
2797 spin_lock_nested(&target->d_parent->d_lock,
2798 DENTRY_D_LOCK_NESTED);
2799 } else {
2800 spin_lock(&target->d_parent->d_lock);
2801 spin_lock_nested(&dentry->d_parent->d_lock,
2802 DENTRY_D_LOCK_NESTED);
2803 }
2804 }
2805 if (target < dentry) {
2806 spin_lock_nested(&target->d_lock, 2);
2807 spin_lock_nested(&dentry->d_lock, 3);
2808 } else {
2809 spin_lock_nested(&dentry->d_lock, 2);
2810 spin_lock_nested(&target->d_lock, 3);
2811 }
2812}
2813
Al Viro986c0192014-09-26 23:11:15 -04002814static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002815{
2816 if (target->d_parent != dentry->d_parent)
2817 spin_unlock(&dentry->d_parent->d_lock);
2818 if (target->d_parent != target)
2819 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002820 spin_unlock(&target->d_lock);
2821 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002822}
2823
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002825 * When switching names, the actual string doesn't strictly have to
2826 * be preserved in the target - because we're dropping the target
2827 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002828 * the new name before we switch, unless we are going to rehash
2829 * it. Note that if we *do* unhash the target, we are not allowed
2830 * to rehash it without giving it a new name/hash key - whether
2831 * we swap or overwrite the names here, resulting name won't match
2832 * the reality in filesystem; it's only there for d_path() purposes.
2833 * Note that all of this is happening under rename_lock, so the
2834 * any hash lookup seeing it in the middle of manipulations will
2835 * be discarded anyway. So we do not care what happens to the hash
2836 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002838/*
Al Viro18367502011-07-12 21:42:24 -04002839 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 * @dentry: entry to move
2841 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002842 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 *
2844 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002845 * dcache entries should not be moved in this way. Caller must hold
2846 * rename_lock, the i_mutex of the source and target directories,
2847 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002849static void __d_move(struct dentry *dentry, struct dentry *target,
2850 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851{
Al Viro84e710d2016-04-15 00:58:55 -04002852 struct inode *dir = NULL;
2853 unsigned n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 if (!dentry->d_inode)
2855 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2856
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002857 BUG_ON(d_ancestor(dentry, target));
2858 BUG_ON(d_ancestor(target, dentry));
2859
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002860 dentry_lock_for_move(dentry, target);
Al Viro84e710d2016-04-15 00:58:55 -04002861 if (unlikely(d_in_lookup(target))) {
2862 dir = target->d_parent->d_inode;
2863 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002864 __d_lookup_done(target);
Al Viro84e710d2016-04-15 00:58:55 -04002865 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866
Nick Piggin31e6b012011-01-07 17:49:52 +11002867 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002868 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002869
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002870 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2871
2872 /*
2873 * Move the dentry to the target hash queue. Don't bother checking
2874 * for the same hash queue because of how unlikely it is.
2875 */
2876 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002877 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878
Miklos Szeredida1ce062014-04-01 17:08:43 +02002879 /*
2880 * Unhash the target (d_delete() is not usable here). If exchanging
2881 * the two dentries, then rehash onto the other's hash queue.
2882 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002884 if (exchange) {
2885 __d_rehash(target,
2886 d_hash(dentry->d_parent, dentry->d_name.hash));
2887 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002890 if (exchange)
2891 swap_names(dentry, target);
2892 else
2893 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894
Al Viro63cf4272014-09-26 23:06:14 -04002895 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002897 /* splicing a tree */
Al Viro3d56c252016-06-07 21:26:55 -04002898 dentry->d_flags |= DCACHE_RCUACCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 dentry->d_parent = target->d_parent;
2900 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002901 list_del_init(&target->d_child);
2902 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002904 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002905 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002906 list_move(&target->d_child, &target->d_parent->d_subdirs);
2907 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002908 if (exchange)
2909 fsnotify_d_move(target);
2910 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 }
2912
Nick Piggin31e6b012011-01-07 17:49:52 +11002913 write_seqcount_end(&target->d_seq);
2914 write_seqcount_end(&dentry->d_seq);
2915
Al Viro84e710d2016-04-15 00:58:55 -04002916 if (dir)
2917 end_dir_add(dir, n);
Al Viro986c0192014-09-26 23:11:15 -04002918 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002919}
2920
2921/*
2922 * d_move - move a dentry
2923 * @dentry: entry to move
2924 * @target: new dentry
2925 *
2926 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002927 * dcache entries should not be moved in this way. See the locking
2928 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002929 */
2930void d_move(struct dentry *dentry, struct dentry *target)
2931{
2932 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002933 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002935}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002936EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937
Miklos Szeredida1ce062014-04-01 17:08:43 +02002938/*
2939 * d_exchange - exchange two dentries
2940 * @dentry1: first dentry
2941 * @dentry2: second dentry
2942 */
2943void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2944{
2945 write_seqlock(&rename_lock);
2946
2947 WARN_ON(!dentry1->d_inode);
2948 WARN_ON(!dentry2->d_inode);
2949 WARN_ON(IS_ROOT(dentry1));
2950 WARN_ON(IS_ROOT(dentry2));
2951
2952 __d_move(dentry1, dentry2, true);
2953
2954 write_sequnlock(&rename_lock);
2955}
2956
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002957/**
2958 * d_ancestor - search for an ancestor
2959 * @p1: ancestor dentry
2960 * @p2: child dentry
2961 *
2962 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2963 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002964 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002965struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002966{
2967 struct dentry *p;
2968
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002969 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002970 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002971 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002972 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002973 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002974}
2975
2976/*
2977 * This helper attempts to cope with remotely renamed directories
2978 *
2979 * It assumes that the caller is already holding
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002980 * dentry->d_parent->d_inode->i_mutex, and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002981 *
2982 * Note: If ever the locking in lock_rename() changes, then please
2983 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002984 */
Al Virob5ae6b12014-10-12 22:16:02 -04002985static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002986 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002987{
Al Viro9902af72016-04-15 15:08:36 -04002988 struct mutex *m1 = NULL;
2989 struct rw_semaphore *m2 = NULL;
J. Bruce Fields3d330dc2015-02-10 10:55:53 -05002990 int ret = -ESTALE;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002991
2992 /* If alias and dentry share a parent, then no extra locks required */
2993 if (alias->d_parent == dentry->d_parent)
2994 goto out_unalias;
2995
Trond Myklebust9eaef272006-10-21 10:24:20 -07002996 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002997 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2998 goto out_err;
2999 m1 = &dentry->d_sb->s_vfs_rename_mutex;
Al Viro9902af72016-04-15 15:08:36 -04003000 if (!inode_trylock_shared(alias->d_parent->d_inode))
Trond Myklebust9eaef272006-10-21 10:24:20 -07003001 goto out_err;
Al Viro9902af72016-04-15 15:08:36 -04003002 m2 = &alias->d_parent->d_inode->i_rwsem;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003003out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003004 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04003005 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003006out_err:
Trond Myklebust9eaef272006-10-21 10:24:20 -07003007 if (m2)
Al Viro9902af72016-04-15 15:08:36 -04003008 up_read(m2);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003009 if (m1)
3010 mutex_unlock(m1);
3011 return ret;
3012}
3013
David Howells770bfad2006-08-22 20:06:07 -04003014/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05003015 * d_splice_alias - splice a disconnected dentry into the tree if one exists
3016 * @inode: the inode which may have a disconnected dentry
3017 * @dentry: a negative dentry which we want to point to the inode.
3018 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05003019 * If inode is a directory and has an IS_ROOT alias, then d_move that in
3020 * place of the given dentry and return it, else simply d_add the inode
3021 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05003022 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05003023 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3024 * we should error out: directories can't have multiple aliases.
3025 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05003026 * This is needed in the lookup routine of any filesystem that is exportable
3027 * (via knfsd) so that we can build dcache paths to directories effectively.
3028 *
3029 * If a dentry was found and moved, then it is returned. Otherwise NULL
3030 * is returned. This matches the expected return value of ->lookup.
3031 *
3032 * Cluster filesystems may call this function with a negative, hashed dentry.
3033 * In that case, we know that the inode will be a regular file, and also this
3034 * will only occur during atomic_open. So we need to check for the dentry
3035 * being already hashed only in the final case.
3036 */
3037struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3038{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05003039 if (IS_ERR(inode))
3040 return ERR_CAST(inode);
3041
David Howells770bfad2006-08-22 20:06:07 -04003042 BUG_ON(!d_unhashed(dentry));
3043
Al Virode689f52016-03-09 18:05:42 -05003044 if (!inode)
Al Virob5ae6b12014-10-12 22:16:02 -04003045 goto out;
Al Virode689f52016-03-09 18:05:42 -05003046
Al Virob9680912016-04-11 00:53:26 -04003047 security_d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11003048 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003049 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04003050 struct dentry *new = __d_find_any_alias(inode);
3051 if (unlikely(new)) {
Eric W. Biedermana03e2832015-08-15 13:36:41 -05003052 /* The reference to new ensures it remains an alias */
3053 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04003054 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04003055 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04003056 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04003057 dput(new);
3058 new = ERR_PTR(-ELOOP);
3059 pr_warn_ratelimited(
3060 "VFS: Lookup of '%s' in %s %s"
3061 " would have caused loop\n",
3062 dentry->d_name.name,
3063 inode->i_sb->s_type->name,
3064 inode->i_sb->s_id);
3065 } else if (!IS_ROOT(new)) {
3066 int err = __d_unalias(inode, dentry, new);
3067 write_sequnlock(&rename_lock);
3068 if (err) {
3069 dput(new);
3070 new = ERR_PTR(err);
3071 }
Al Viro18367502011-07-12 21:42:24 -04003072 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04003073 __d_move(new, dentry, false);
3074 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003075 }
Al Virob5ae6b12014-10-12 22:16:02 -04003076 iput(inode);
3077 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003078 }
David Howells770bfad2006-08-22 20:06:07 -04003079 }
Al Virob5ae6b12014-10-12 22:16:02 -04003080out:
Al Viroed782b52016-03-09 19:52:39 -05003081 __d_add(dentry, inode);
Al Virob5ae6b12014-10-12 22:16:02 -04003082 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04003083}
Al Virob5ae6b12014-10-12 22:16:02 -04003084EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04003085
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003086static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01003087{
3088 *buflen -= namelen;
3089 if (*buflen < 0)
3090 return -ENAMETOOLONG;
3091 *buffer -= namelen;
3092 memcpy(*buffer, str, namelen);
3093 return 0;
3094}
3095
Waiman Long232d2d62013-09-09 12:18:13 -04003096/**
3097 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04003098 * @buffer: buffer pointer
3099 * @buflen: allocated length of the buffer
3100 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04003101 *
3102 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3103 * make sure that either the old or the new name pointer and length are
3104 * fetched. However, there may be mismatch between length and pointer.
3105 * The length cannot be trusted, we need to copy it byte-by-byte until
3106 * the length is reached or a null byte is found. It also prepends "/" at
3107 * the beginning of the name. The sequence number check at the caller will
3108 * retry it again when a d_move() does happen. So any garbage in the buffer
3109 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04003110 *
3111 * Data dependency barrier is needed to make sure that we see that terminating
3112 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04003113 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003114static int prepend_name(char **buffer, int *buflen, struct qstr *name)
3115{
Waiman Long232d2d62013-09-09 12:18:13 -04003116 const char *dname = ACCESS_ONCE(name->name);
3117 u32 dlen = ACCESS_ONCE(name->len);
3118 char *p;
3119
Al Viro6d13f692014-09-29 14:46:30 -04003120 smp_read_barrier_depends();
3121
Waiman Long232d2d62013-09-09 12:18:13 -04003122 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04003123 if (*buflen < 0)
3124 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04003125 p = *buffer -= dlen + 1;
3126 *p++ = '/';
3127 while (dlen--) {
3128 char c = *dname++;
3129 if (!c)
3130 break;
3131 *p++ = c;
3132 }
3133 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003134}
3135
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136/**
Randy Dunlap208898c2010-11-18 15:02:49 -08003137 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003138 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003139 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003140 * @buffer: pointer to the end of the buffer
3141 * @buflen: pointer to buffer length
3142 *
Waiman Long18129972013-09-12 10:55:35 -04003143 * The function will first try to write out the pathname without taking any
3144 * lock other than the RCU read lock to make sure that dentries won't go away.
3145 * It only checks the sequence number of the global rename_lock as any change
3146 * in the dentry's d_seq will be preceded by changes in the rename_lock
3147 * sequence number. If the sequence number had been changed, it will restart
3148 * the whole pathname back-tracing sequence again by taking the rename_lock.
3149 * In this case, there is no need to take the RCU read lock as the recursive
3150 * parent pointer references will keep the dentry chain alive as long as no
3151 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003152 */
Al Viro02125a82011-12-05 08:43:34 -05003153static int prepend_path(const struct path *path,
3154 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003155 char **buffer, int *buflen)
3156{
Al Viroede4ceb2013-11-13 07:45:40 -05003157 struct dentry *dentry;
3158 struct vfsmount *vfsmnt;
3159 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003160 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04003161 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04003162 char *bptr;
3163 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003164
Al Viro48f5ec22013-09-09 15:22:25 -04003165 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04003166restart_mnt:
3167 read_seqbegin_or_lock(&mount_lock, &m_seq);
3168 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003169 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003170restart:
3171 bptr = *buffer;
3172 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04003173 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05003174 dentry = path->dentry;
3175 vfsmnt = path->mnt;
3176 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04003177 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003178 while (dentry != root->dentry || vfsmnt != root->mnt) {
3179 struct dentry * parent;
3180
3181 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04003182 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Eric W. Biedermancde93be2015-08-15 13:36:12 -05003183 /* Escaped? */
3184 if (dentry != vfsmnt->mnt_root) {
3185 bptr = *buffer;
3186 blen = *buflen;
3187 error = 3;
3188 break;
3189 }
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003190 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04003191 if (mnt != parent) {
3192 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
3193 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04003194 vfsmnt = &mnt->mnt;
3195 continue;
3196 }
Waiman Long232d2d62013-09-09 12:18:13 -04003197 if (!error)
3198 error = is_mounted(vfsmnt) ? 1 : 2;
3199 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003200 }
3201 parent = dentry->d_parent;
3202 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003203 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003204 if (error)
3205 break;
3206
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003207 dentry = parent;
3208 }
Al Viro48f5ec22013-09-09 15:22:25 -04003209 if (!(seq & 1))
3210 rcu_read_unlock();
3211 if (need_seqretry(&rename_lock, seq)) {
3212 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003213 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003214 }
Al Viro48f5ec22013-09-09 15:22:25 -04003215 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003216
3217 if (!(m_seq & 1))
3218 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04003219 if (need_seqretry(&mount_lock, m_seq)) {
3220 m_seq = 1;
3221 goto restart_mnt;
3222 }
3223 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003224
Waiman Long232d2d62013-09-09 12:18:13 -04003225 if (error >= 0 && bptr == *buffer) {
3226 if (--blen < 0)
3227 error = -ENAMETOOLONG;
3228 else
3229 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003230 }
Waiman Long232d2d62013-09-09 12:18:13 -04003231 *buffer = bptr;
3232 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04003233 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003234}
3235
3236/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003237 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003238 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003239 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07003240 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241 * @buflen: buffer length
3242 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003243 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003245 * Returns a pointer into the buffer or an error code if the
3246 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08003247 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04003248 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003249 *
Al Viro02125a82011-12-05 08:43:34 -05003250 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251 */
Al Viro02125a82011-12-05 08:43:34 -05003252char *__d_path(const struct path *path,
3253 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003254 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003256 char *res = buf + buflen;
3257 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003259 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003260 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04003261
Al Viro02125a82011-12-05 08:43:34 -05003262 if (error < 0)
3263 return ERR_PTR(error);
3264 if (error > 0)
3265 return NULL;
3266 return res;
3267}
3268
3269char *d_absolute_path(const struct path *path,
3270 char *buf, int buflen)
3271{
3272 struct path root = {};
3273 char *res = buf + buflen;
3274 int error;
3275
3276 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003277 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05003278
3279 if (error > 1)
3280 error = -EINVAL;
3281 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003282 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003283 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284}
3285
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003286/*
3287 * same as __d_path but appends "(deleted)" for unlinked files.
3288 */
Al Viro02125a82011-12-05 08:43:34 -05003289static int path_with_deleted(const struct path *path,
3290 const struct path *root,
3291 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003292{
3293 prepend(buf, buflen, "\0", 1);
3294 if (d_unlinked(path->dentry)) {
3295 int error = prepend(buf, buflen, " (deleted)", 10);
3296 if (error)
3297 return error;
3298 }
3299
3300 return prepend_path(path, root, buf, buflen);
3301}
3302
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003303static int prepend_unreachable(char **buffer, int *buflen)
3304{
3305 return prepend(buffer, buflen, "(unreachable)", 13);
3306}
3307
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003308static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3309{
3310 unsigned seq;
3311
3312 do {
3313 seq = read_seqcount_begin(&fs->seq);
3314 *root = fs->root;
3315 } while (read_seqcount_retry(&fs->seq, seq));
3316}
3317
Jan Bluncka03a8a702008-02-14 19:38:32 -08003318/**
3319 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003320 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003321 * @buf: buffer to return value in
3322 * @buflen: buffer length
3323 *
3324 * Convert a dentry into an ASCII path name. If the entry has been deleted
3325 * the string " (deleted)" is appended. Note that this is ambiguous.
3326 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003327 * Returns a pointer into the buffer or an error code if the path was
3328 * too long. Note: Callers should use the returned pointer, not the passed
3329 * in buffer, to use the name! The implementation often starts at an offset
3330 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003331 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003332 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003333 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003334char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003336 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003337 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003338 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003340 /*
3341 * We have various synthetic filesystems that never get mounted. On
3342 * these filesystems dentries are never used for lookup purposes, and
3343 * thus don't need to be hashed. They also don't need a name until a
3344 * user wants to identify the object in /proc/pid/fd/. The little hack
3345 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003346 *
3347 * Some pseudo inodes are mountable. When they are mounted
3348 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3349 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003350 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003351 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3352 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003353 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003354
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003355 rcu_read_lock();
3356 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003357 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003358 rcu_read_unlock();
3359
Al Viro02125a82011-12-05 08:43:34 -05003360 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003361 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362 return res;
3363}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003364EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365
3366/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003367 * Helper function for dentry_operations.d_dname() members
3368 */
3369char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3370 const char *fmt, ...)
3371{
3372 va_list args;
3373 char temp[64];
3374 int sz;
3375
3376 va_start(args, fmt);
3377 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3378 va_end(args);
3379
3380 if (sz > sizeof(temp) || sz > buflen)
3381 return ERR_PTR(-ENAMETOOLONG);
3382
3383 buffer += buflen - sz;
3384 return memcpy(buffer, temp, sz);
3385}
3386
Al Viro118b2302013-08-24 12:08:17 -04003387char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3388{
3389 char *end = buffer + buflen;
3390 /* these dentries are never renamed, so d_lock is not needed */
3391 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003392 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003393 prepend(&end, &buflen, "/", 1))
3394 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003395 return end;
Al Viro118b2302013-08-24 12:08:17 -04003396}
David Herrmann31bbe162014-01-03 14:09:47 +01003397EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003398
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003399/*
Ram Pai6092d042008-03-27 13:06:20 +01003400 * Write full pathname from the root of the filesystem into the buffer.
3401 */
Al Virof6500802014-01-26 12:37:55 -05003402static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003403{
Al Virof6500802014-01-26 12:37:55 -05003404 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003405 char *end, *retval;
3406 int len, seq = 0;
3407 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003408
Al Virof6500802014-01-26 12:37:55 -05003409 if (buflen < 2)
3410 goto Elong;
3411
Al Viro48f5ec22013-09-09 15:22:25 -04003412 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003413restart:
Al Virof6500802014-01-26 12:37:55 -05003414 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003415 end = buf + buflen;
3416 len = buflen;
3417 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003418 /* Get '/' right */
3419 retval = end-1;
3420 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003421 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003422 while (!IS_ROOT(dentry)) {
3423 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003424
Ram Pai6092d042008-03-27 13:06:20 +01003425 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003426 error = prepend_name(&end, &len, &dentry->d_name);
3427 if (error)
3428 break;
Ram Pai6092d042008-03-27 13:06:20 +01003429
3430 retval = end;
3431 dentry = parent;
3432 }
Al Viro48f5ec22013-09-09 15:22:25 -04003433 if (!(seq & 1))
3434 rcu_read_unlock();
3435 if (need_seqretry(&rename_lock, seq)) {
3436 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003437 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003438 }
3439 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003440 if (error)
3441 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003442 return retval;
3443Elong:
3444 return ERR_PTR(-ENAMETOOLONG);
3445}
Nick Pigginec2447c2011-01-07 17:49:29 +11003446
3447char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3448{
Waiman Long232d2d62013-09-09 12:18:13 -04003449 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003450}
3451EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003452
3453char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3454{
3455 char *p = NULL;
3456 char *retval;
3457
Al Viroc1031352010-06-06 22:31:14 -04003458 if (d_unlinked(dentry)) {
3459 p = buf + buflen;
3460 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3461 goto Elong;
3462 buflen++;
3463 }
3464 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003465 if (!IS_ERR(retval) && p)
3466 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003467 return retval;
3468Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003469 return ERR_PTR(-ENAMETOOLONG);
3470}
3471
Linus Torvalds8b19e342013-09-12 10:35:47 -07003472static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3473 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003474{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003475 unsigned seq;
3476
3477 do {
3478 seq = read_seqcount_begin(&fs->seq);
3479 *root = fs->root;
3480 *pwd = fs->pwd;
3481 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003482}
3483
Ram Pai6092d042008-03-27 13:06:20 +01003484/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485 * NOTE! The user-level library version returns a
3486 * character pointer. The kernel system call just
3487 * returns the length of the buffer filled (which
3488 * includes the ending '\0' character), or a negative
3489 * error value. So libc would do something like
3490 *
3491 * char *getcwd(char * buf, size_t size)
3492 * {
3493 * int retval;
3494 *
3495 * retval = sys_getcwd(buf, size);
3496 * if (retval >= 0)
3497 * return buf;
3498 * errno = -retval;
3499 * return NULL;
3500 * }
3501 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003502SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503{
Linus Torvalds552ce542007-02-13 12:08:18 -08003504 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003505 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003506 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003507
3508 if (!page)
3509 return -ENOMEM;
3510
Linus Torvalds8b19e342013-09-12 10:35:47 -07003511 rcu_read_lock();
3512 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003513
Linus Torvalds552ce542007-02-13 12:08:18 -08003514 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003515 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003516 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003517 char *cwd = page + PATH_MAX;
3518 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003520 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003521 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003522 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003523
Al Viro02125a82011-12-05 08:43:34 -05003524 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003525 goto out;
3526
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003527 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003528 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003529 error = prepend_unreachable(&cwd, &buflen);
3530 if (error)
3531 goto out;
3532 }
3533
Linus Torvalds552ce542007-02-13 12:08:18 -08003534 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003535 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003536 if (len <= size) {
3537 error = len;
3538 if (copy_to_user(buf, cwd, len))
3539 error = -EFAULT;
3540 }
Nick Piggin949854d2011-01-07 17:49:37 +11003541 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003542 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003543 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003544
3545out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003546 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547 return error;
3548}
3549
3550/*
3551 * Test whether new_dentry is a subdirectory of old_dentry.
3552 *
3553 * Trivially implemented using the dcache structure
3554 */
3555
3556/**
3557 * is_subdir - is new dentry a subdirectory of old_dentry
3558 * @new_dentry: new dentry
3559 * @old_dentry: old dentry
3560 *
Yaowei Baia6e57872015-11-17 14:40:11 +08003561 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3562 * Returns false otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3564 */
3565
Yaowei Baia6e57872015-11-17 14:40:11 +08003566bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567{
Yaowei Baia6e57872015-11-17 14:40:11 +08003568 bool result;
Nick Piggin949854d2011-01-07 17:49:37 +11003569 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003571 if (new_dentry == old_dentry)
Yaowei Baia6e57872015-11-17 14:40:11 +08003572 return true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003573
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003574 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003577 /*
3578 * Need rcu_readlock to protect against the d_parent trashing
3579 * due to d_move
3580 */
3581 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003582 if (d_ancestor(old_dentry, new_dentry))
Yaowei Baia6e57872015-11-17 14:40:11 +08003583 result = true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003584 else
Yaowei Baia6e57872015-11-17 14:40:11 +08003585 result = false;
Nick Piggin949854d2011-01-07 17:49:37 +11003586 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003588
3589 return result;
3590}
3591
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003592static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003594 struct dentry *root = data;
3595 if (dentry != root) {
3596 if (d_unhashed(dentry) || !dentry->d_inode)
3597 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003599 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3600 dentry->d_flags |= DCACHE_GENOCIDE;
3601 dentry->d_lockref.count--;
3602 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003603 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003604 return D_WALK_CONTINUE;
3605}
Nick Piggin58db63d2011-01-07 17:49:39 +11003606
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003607void d_genocide(struct dentry *parent)
3608{
3609 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610}
3611
Al Viro60545d02013-06-07 01:20:27 -04003612void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003613{
Al Viro60545d02013-06-07 01:20:27 -04003614 inode_dec_link_count(inode);
3615 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003616 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003617 !d_unlinked(dentry));
3618 spin_lock(&dentry->d_parent->d_lock);
3619 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3620 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3621 (unsigned long long)inode->i_ino);
3622 spin_unlock(&dentry->d_lock);
3623 spin_unlock(&dentry->d_parent->d_lock);
3624 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003625}
Al Viro60545d02013-06-07 01:20:27 -04003626EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627
3628static __initdata unsigned long dhash_entries;
3629static int __init set_dhash_entries(char *str)
3630{
3631 if (!str)
3632 return 0;
3633 dhash_entries = simple_strtoul(str, &str, 0);
3634 return 1;
3635}
3636__setup("dhash_entries=", set_dhash_entries);
3637
3638static void __init dcache_init_early(void)
3639{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003640 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003641
3642 /* If hashes are distributed across NUMA nodes, defer
3643 * hash allocation until vmalloc space is available.
3644 */
3645 if (hashdist)
3646 return;
3647
3648 dentry_hashtable =
3649 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003650 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003651 dhash_entries,
3652 13,
3653 HASH_EARLY,
3654 &d_hash_shift,
3655 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003656 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003657 0);
3658
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003659 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003660 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661}
3662
Denis Cheng74bf17c2007-10-16 23:26:30 -07003663static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003665 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003666
3667 /*
3668 * A constructor could be added for stable state like the lists,
3669 * but it is probably not worth it because of the cache nature
3670 * of the dcache.
3671 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003672 dentry_cache = KMEM_CACHE(dentry,
Vladimir Davydov5d097052016-01-14 15:18:21 -08003673 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674
3675 /* Hash may have been set up in dcache_init_early */
3676 if (!hashdist)
3677 return;
3678
3679 dentry_hashtable =
3680 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003681 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682 dhash_entries,
3683 13,
3684 0,
3685 &d_hash_shift,
3686 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003687 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688 0);
3689
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003690 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003691 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692}
3693
3694/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003695struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003696EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003697
Linus Torvalds1da177e2005-04-16 15:20:36 -07003698EXPORT_SYMBOL(d_genocide);
3699
Linus Torvalds1da177e2005-04-16 15:20:36 -07003700void __init vfs_caches_init_early(void)
3701{
3702 dcache_init_early();
3703 inode_init_early();
3704}
3705
Mel Gorman4248b0d2015-08-06 15:46:20 -07003706void __init vfs_caches_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003708 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003709 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003710
Denis Cheng74bf17c2007-10-16 23:26:30 -07003711 dcache_init();
3712 inode_init();
Mel Gorman4248b0d2015-08-06 15:46:20 -07003713 files_init();
3714 files_maxfiles_init();
Denis Cheng74bf17c2007-10-16 23:26:30 -07003715 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716 bdev_cache_init();
3717 chrdev_init();
3718}