blob: 656ce522a218f29850e2415b0a038c8208a5dd52 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114/* Statistics gathering. */
115struct dentry_stat_t dentry_stat = {
116 .age_limit = 45,
117};
118
Glauber Costa3942c072013-08-28 10:17:53 +1000119static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000120static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400121
122#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000123
124/*
125 * Here we resort to our own counters instead of using generic per-cpu counters
126 * for consistency with what the vfs inode code does. We are expected to harvest
127 * better code and performance by having our own specialized counters.
128 *
129 * Please note that the loop is done over all possible CPUs, not over all online
130 * CPUs. The reason for this is that we don't want to play games with CPUs going
131 * on and off. If one of them goes off, we will just keep their counters.
132 *
133 * glommer: See cffbc8a for details, and if you ever intend to change this,
134 * please update all vfs counters to match.
135 */
Glauber Costa3942c072013-08-28 10:17:53 +1000136static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100137{
138 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000139 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100140 for_each_possible_cpu(i)
141 sum += per_cpu(nr_dentry, i);
142 return sum < 0 ? 0 : sum;
143}
144
Dave Chinner62d36c72013-08-28 10:17:54 +1000145static long get_nr_dentry_unused(void)
146{
147 int i;
148 long sum = 0;
149 for_each_possible_cpu(i)
150 sum += per_cpu(nr_dentry_unused, i);
151 return sum < 0 ? 0 : sum;
152}
153
Joe Perches1f7e0612014-06-06 14:38:05 -0700154int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400155 size_t *lenp, loff_t *ppos)
156{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100157 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000158 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000159 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400160}
161#endif
162
Linus Torvalds5483f182012-03-04 15:51:42 -0800163/*
164 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
165 * The strings are both count bytes long, and count is non-zero.
166 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700167#ifdef CONFIG_DCACHE_WORD_ACCESS
168
169#include <asm/word-at-a-time.h>
170/*
171 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
172 * aligned allocation for this particular component. We don't
173 * strictly need the load_unaligned_zeropad() safety, but it
174 * doesn't hurt either.
175 *
176 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
177 * need the careful unaligned handling.
178 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700179static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800180{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800181 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800182
183 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700184 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700185 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800186 if (tcount < sizeof(unsigned long))
187 break;
188 if (unlikely(a != b))
189 return 1;
190 cs += sizeof(unsigned long);
191 ct += sizeof(unsigned long);
192 tcount -= sizeof(unsigned long);
193 if (!tcount)
194 return 0;
195 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000196 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800197 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700198}
199
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800200#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700201
Linus Torvalds94753db52012-05-10 12:19:19 -0700202static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700203{
Linus Torvalds5483f182012-03-04 15:51:42 -0800204 do {
205 if (*cs != *ct)
206 return 1;
207 cs++;
208 ct++;
209 tcount--;
210 } while (tcount);
211 return 0;
212}
213
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700214#endif
215
Linus Torvalds94753db52012-05-10 12:19:19 -0700216static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
217{
Linus Torvalds6326c712012-05-21 16:14:04 -0700218 const unsigned char *cs;
Linus Torvalds94753db52012-05-10 12:19:19 -0700219 /*
220 * Be careful about RCU walk racing with rename:
221 * use ACCESS_ONCE to fetch the name pointer.
222 *
223 * NOTE! Even if a rename will mean that the length
224 * was not loaded atomically, we don't care. The
225 * RCU walk will check the sequence count eventually,
226 * and catch it. And we won't overrun the buffer,
227 * because we're reading the name pointer atomically,
228 * and a dentry name is guaranteed to be properly
229 * terminated with a NUL byte.
230 *
231 * End result: even if 'len' is wrong, we'll exit
232 * early because the data cannot match (there can
233 * be no NUL in the ct/tcount data)
234 */
Linus Torvalds6326c712012-05-21 16:14:04 -0700235 cs = ACCESS_ONCE(dentry->d_name.name);
236 smp_read_barrier_depends();
237 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700238}
239
Al Viro8d85b482014-09-29 14:54:27 -0400240struct external_name {
241 union {
242 atomic_t count;
243 struct rcu_head head;
244 } u;
245 unsigned char name[];
246};
247
248static inline struct external_name *external_name(struct dentry *dentry)
249{
250 return container_of(dentry->d_name.name, struct external_name, name[0]);
251}
252
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400253static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400255 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
256
Al Viro8d85b482014-09-29 14:54:27 -0400257 kmem_cache_free(dentry_cache, dentry);
258}
259
260static void __d_free_external(struct rcu_head *head)
261{
262 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
Al Viro8d85b482014-09-29 14:54:27 -0400263 kfree(external_name(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 kmem_cache_free(dentry_cache, dentry);
265}
266
Al Viro810bb172014-10-12 12:45:37 -0400267static inline int dname_external(const struct dentry *dentry)
268{
269 return dentry->d_name.name != dentry->d_iname;
270}
271
David Howells4bf46a22015-03-05 14:09:22 +0000272/*
273 * Make sure other CPUs see the inode attached before the type is set.
274 */
275static inline void __d_set_inode_and_type(struct dentry *dentry,
276 struct inode *inode,
277 unsigned type_flags)
278{
279 unsigned flags;
280
281 dentry->d_inode = inode;
282 smp_wmb();
283 flags = READ_ONCE(dentry->d_flags);
284 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
285 flags |= type_flags;
286 WRITE_ONCE(dentry->d_flags, flags);
287}
288
289/*
290 * Ideally, we want to make sure that other CPUs see the flags cleared before
291 * the inode is detached, but this is really a violation of RCU principles
292 * since the ordering suggests we should always set inode before flags.
293 *
294 * We should instead replace or discard the entire dentry - but that sucks
295 * performancewise on mass deletion/rename.
296 */
297static inline void __d_clear_type_and_inode(struct dentry *dentry)
298{
299 unsigned flags = READ_ONCE(dentry->d_flags);
300
301 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
302 WRITE_ONCE(dentry->d_flags, flags);
303 smp_wmb();
304 dentry->d_inode = NULL;
305}
306
Al Virob4f03542014-04-29 23:40:14 -0400307static void dentry_free(struct dentry *dentry)
308{
Al Viro946e51f2014-10-26 19:19:16 -0400309 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400310 if (unlikely(dname_external(dentry))) {
311 struct external_name *p = external_name(dentry);
312 if (likely(atomic_dec_and_test(&p->u.count))) {
313 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
314 return;
315 }
316 }
Al Virob4f03542014-04-29 23:40:14 -0400317 /* if dentry was never visible to RCU, immediate free is OK */
318 if (!(dentry->d_flags & DCACHE_RCUACCESS))
319 __d_free(&dentry->d_u.d_rcu);
320 else
321 call_rcu(&dentry->d_u.d_rcu, __d_free);
322}
323
Nick Piggin31e6b012011-01-07 17:49:52 +1100324/**
325 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800326 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100327 * After this call, in-progress rcu-walk path lookup will fail. This
328 * should be called after unhashing, and after changing d_inode (if
329 * the dentry has not already been unhashed).
330 */
331static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
332{
333 assert_spin_locked(&dentry->d_lock);
334 /* Go through a barrier */
335 write_seqcount_barrier(&dentry->d_seq);
336}
337
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338/*
339 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100340 * d_iput() operation if defined. Dentry has no refcount
341 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800343static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200344 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100345 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346{
347 struct inode *inode = dentry->d_inode;
348 if (inode) {
David Howells4bf46a22015-03-05 14:09:22 +0000349 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400350 hlist_del_init(&dentry->d_u.d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100352 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700353 if (!inode->i_nlink)
354 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 if (dentry->d_op && dentry->d_op->d_iput)
356 dentry->d_op->d_iput(dentry, inode);
357 else
358 iput(inode);
359 } else {
360 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 }
362}
363
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700364/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100365 * Release the dentry's inode, using the filesystem
366 * d_iput() operation if defined. dentry remains in-use.
367 */
368static void dentry_unlink_inode(struct dentry * dentry)
369 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100370 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100371{
372 struct inode *inode = dentry->d_inode;
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);
Nick Piggin31e6b012011-01-07 17:49:52 +1100375 dentry_rcuwalk_barrier(dentry);
376 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);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700497 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100498 }
499}
500EXPORT_SYMBOL(__d_drop);
501
502void d_drop(struct dentry *dentry)
503{
Nick Piggin789680d2011-01-07 17:49:30 +1100504 spin_lock(&dentry->d_lock);
505 __d_drop(dentry);
506 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100507}
508EXPORT_SYMBOL(d_drop);
509
Al Viroe55fd012014-05-28 13:51:12 -0400510static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100511{
Al Viro41edf272014-05-01 10:30:00 -0400512 struct dentry *parent = NULL;
513 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400514 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100515 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100516
Linus Torvalds0d984392013-09-08 13:46:52 -0700517 /*
518 * The dentry is now unrecoverably dead to the world.
519 */
520 lockref_mark_dead(&dentry->d_lockref);
521
Sage Weilf0023bc2011-10-28 10:02:42 -0700522 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700523 * inform the fs via d_prune that this dentry is about to be
524 * unhashed and destroyed.
525 */
Al Viro29266202014-05-30 11:39:02 -0400526 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800527 dentry->d_op->d_prune(dentry);
528
Al Viro01b60352014-04-29 23:42:52 -0400529 if (dentry->d_flags & DCACHE_LRU_LIST) {
530 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
531 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400532 }
Nick Piggin77812a12011-01-07 17:49:48 +1100533 /* if it was on the hash then remove it */
534 __d_drop(dentry);
Al Viroca5358e2014-10-26 19:31:10 -0400535 __list_del_entry(&dentry->d_child);
Al Viro03b3b882014-04-29 15:45:28 -0400536 /*
537 * Inform d_walk() that we are no longer attached to the
538 * dentry tree
539 */
540 dentry->d_flags |= DCACHE_DENTRY_KILLED;
541 if (parent)
542 spin_unlock(&parent->d_lock);
543 dentry_iput(dentry);
544 /*
545 * dentry_iput drops the locks, at which point nobody (except
546 * transient RCU lookups) can reach this dentry.
547 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800548 BUG_ON(dentry->d_lockref.count > 0);
Al Viro03b3b882014-04-29 15:45:28 -0400549 this_cpu_dec(nr_dentry);
550 if (dentry->d_op && dentry->d_op->d_release)
551 dentry->d_op->d_release(dentry);
552
Al Viro41edf272014-05-01 10:30:00 -0400553 spin_lock(&dentry->d_lock);
554 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
555 dentry->d_flags |= DCACHE_MAY_FREE;
556 can_free = false;
557 }
558 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400559 if (likely(can_free))
560 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400561}
562
563/*
564 * Finish off a dentry we've decided to kill.
565 * dentry->d_lock must be held, returns with it unlocked.
566 * If ref is non-zero, then decrement the refcount too.
567 * Returns dentry requiring refcount drop, or NULL if we're done.
568 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400569static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400570 __releases(dentry->d_lock)
571{
572 struct inode *inode = dentry->d_inode;
573 struct dentry *parent = NULL;
574
575 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
576 goto failed;
577
578 if (!IS_ROOT(dentry)) {
579 parent = dentry->d_parent;
580 if (unlikely(!spin_trylock(&parent->d_lock))) {
581 if (inode)
582 spin_unlock(&inode->i_lock);
583 goto failed;
584 }
585 }
586
587 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400588 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400589
590failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400591 spin_unlock(&dentry->d_lock);
592 cpu_relax();
Al Viroe55fd012014-05-28 13:51:12 -0400593 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100594}
595
Al Viro046b9612014-05-29 08:54:52 -0400596static inline struct dentry *lock_parent(struct dentry *dentry)
597{
598 struct dentry *parent = dentry->d_parent;
599 if (IS_ROOT(dentry))
600 return NULL;
Linus Torvalds360f5472015-01-09 15:19:03 -0800601 if (unlikely(dentry->d_lockref.count < 0))
Al Viroc2338f22014-06-12 00:29:13 -0400602 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400603 if (likely(spin_trylock(&parent->d_lock)))
604 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400605 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400606 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400607again:
608 parent = ACCESS_ONCE(dentry->d_parent);
609 spin_lock(&parent->d_lock);
610 /*
611 * We can't blindly lock dentry until we are sure
612 * that we won't violate the locking order.
613 * Any changes of dentry->d_parent must have
614 * been done with parent->d_lock held, so
615 * spin_lock() above is enough of a barrier
616 * for checking if it's still our child.
617 */
618 if (unlikely(parent != dentry->d_parent)) {
619 spin_unlock(&parent->d_lock);
620 goto again;
621 }
622 rcu_read_unlock();
623 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700624 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400625 else
626 parent = NULL;
627 return parent;
628}
629
Linus Torvalds360f5472015-01-09 15:19:03 -0800630/*
631 * Try to do a lockless dput(), and return whether that was successful.
632 *
633 * If unsuccessful, we return false, having already taken the dentry lock.
634 *
635 * The caller needs to hold the RCU read lock, so that the dentry is
636 * guaranteed to stay around even if the refcount goes down to zero!
637 */
638static inline bool fast_dput(struct dentry *dentry)
639{
640 int ret;
641 unsigned int d_flags;
642
643 /*
644 * If we have a d_op->d_delete() operation, we sould not
645 * let the dentry count go to zero, so use "put__or_lock".
646 */
647 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
648 return lockref_put_or_lock(&dentry->d_lockref);
649
650 /*
651 * .. otherwise, we can try to just decrement the
652 * lockref optimistically.
653 */
654 ret = lockref_put_return(&dentry->d_lockref);
655
656 /*
657 * If the lockref_put_return() failed due to the lock being held
658 * by somebody else, the fast path has failed. We will need to
659 * get the lock, and then check the count again.
660 */
661 if (unlikely(ret < 0)) {
662 spin_lock(&dentry->d_lock);
663 if (dentry->d_lockref.count > 1) {
664 dentry->d_lockref.count--;
665 spin_unlock(&dentry->d_lock);
666 return 1;
667 }
668 return 0;
669 }
670
671 /*
672 * If we weren't the last ref, we're done.
673 */
674 if (ret)
675 return 1;
676
677 /*
678 * Careful, careful. The reference count went down
679 * to zero, but we don't hold the dentry lock, so
680 * somebody else could get it again, and do another
681 * dput(), and we need to not race with that.
682 *
683 * However, there is a very special and common case
684 * where we don't care, because there is nothing to
685 * do: the dentry is still hashed, it does not have
686 * a 'delete' op, and it's referenced and already on
687 * the LRU list.
688 *
689 * NOTE! Since we aren't locked, these values are
690 * not "stable". However, it is sufficient that at
691 * some point after we dropped the reference the
692 * dentry was hashed and the flags had the proper
693 * value. Other dentry users may have re-gotten
694 * a reference to the dentry and change that, but
695 * our work is done - we can leave the dentry
696 * around with a zero refcount.
697 */
698 smp_rmb();
699 d_flags = ACCESS_ONCE(dentry->d_flags);
700 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST;
701
702 /* Nothing to do? Dropping the reference was all we needed? */
703 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
704 return 1;
705
706 /*
707 * Not the fast normal case? Get the lock. We've already decremented
708 * the refcount, but we'll need to re-check the situation after
709 * getting the lock.
710 */
711 spin_lock(&dentry->d_lock);
712
713 /*
714 * Did somebody else grab a reference to it in the meantime, and
715 * we're no longer the last user after all? Alternatively, somebody
716 * else could have killed it and marked it dead. Either way, we
717 * don't need to do anything else.
718 */
719 if (dentry->d_lockref.count) {
720 spin_unlock(&dentry->d_lock);
721 return 1;
722 }
723
724 /*
725 * Re-get the reference we optimistically dropped. We hold the
726 * lock, and we just tested that it was zero, so we can just
727 * set it to 1.
728 */
729 dentry->d_lockref.count = 1;
730 return 0;
731}
732
733
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734/*
735 * This is dput
736 *
737 * This is complicated by the fact that we do not want to put
738 * dentries that are no longer on any hash chain on the unused
739 * list: we'd much rather just get rid of them immediately.
740 *
741 * However, that implies that we have to traverse the dentry
742 * tree upwards to the parents which might _also_ now be
743 * scheduled for deletion (it may have been only waiting for
744 * its last child to go away).
745 *
746 * This tail recursion is done by hand as we don't want to depend
747 * on the compiler to always get this right (gcc generally doesn't).
748 * Real recursion would eat up our stack space.
749 */
750
751/*
752 * dput - release a dentry
753 * @dentry: dentry to release
754 *
755 * Release a dentry. This will drop the usage count and if appropriate
756 * call the dentry unlink method as well as removing it from the queues and
757 * releasing its resources. If the parent dentries were scheduled for release
758 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760void dput(struct dentry *dentry)
761{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700762 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 return;
764
765repeat:
Linus Torvalds360f5472015-01-09 15:19:03 -0800766 rcu_read_lock();
767 if (likely(fast_dput(dentry))) {
768 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 return;
Linus Torvalds360f5472015-01-09 15:19:03 -0800770 }
771
772 /* Slow case: now with the dentry lock held */
773 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700775 /* Unreachable? Get rid of it */
776 if (unlikely(d_unhashed(dentry)))
777 goto kill_it;
778
779 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100781 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 }
Nick Piggin265ac902010-10-10 05:36:24 -0400783
Linus Torvalds358eec12013-10-31 15:43:02 -0700784 if (!(dentry->d_flags & DCACHE_REFERENCED))
785 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400786 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400787
Waiman Long98474232013-08-28 18:24:59 -0700788 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100789 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 return;
791
Miklos Szeredid52b9082007-05-08 00:23:46 -0700792kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400793 dentry = dentry_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700794 if (dentry)
795 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700797EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100800/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100801static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
Waiman Long98474232013-08-28 18:24:59 -0700803 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805
Nick Piggindc0474b2011-01-07 17:49:43 +1100806static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100807{
Waiman Long98474232013-08-28 18:24:59 -0700808 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100809}
810
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100811struct dentry *dget_parent(struct dentry *dentry)
812{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700813 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100814 struct dentry *ret;
815
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700816 /*
817 * Do optimistic parent lookup without any
818 * locking.
819 */
820 rcu_read_lock();
821 ret = ACCESS_ONCE(dentry->d_parent);
822 gotref = lockref_get_not_zero(&ret->d_lockref);
823 rcu_read_unlock();
824 if (likely(gotref)) {
825 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
826 return ret;
827 dput(ret);
828 }
829
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100830repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100831 /*
832 * Don't need rcu_dereference because we re-check it was correct under
833 * the lock.
834 */
835 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100836 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100837 spin_lock(&ret->d_lock);
838 if (unlikely(ret != dentry->d_parent)) {
839 spin_unlock(&ret->d_lock);
840 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100841 goto repeat;
842 }
Nick Piggina734eb42011-01-07 17:49:44 +1100843 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700844 BUG_ON(!ret->d_lockref.count);
845 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100846 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100847 return ret;
848}
849EXPORT_SYMBOL(dget_parent);
850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851/**
852 * d_find_alias - grab a hashed alias of inode
853 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 *
855 * If inode has a hashed alias, or is a directory and has any alias,
856 * acquire the reference to alias and return it. Otherwise return NULL.
857 * Notice that if inode is a directory there can be only one alias and
858 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800859 * of a filesystem, or if the directory was renamed and d_revalidate
860 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700862 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500863 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500865static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866{
Nick Pigginda502952011-01-07 17:49:33 +1100867 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Nick Pigginda502952011-01-07 17:49:33 +1100869again:
870 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400871 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100872 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700874 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100875 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500877 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100878 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100879 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 return alias;
881 }
882 }
Nick Pigginda502952011-01-07 17:49:33 +1100883 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 }
Nick Pigginda502952011-01-07 17:49:33 +1100885 if (discon_alias) {
886 alias = discon_alias;
887 spin_lock(&alias->d_lock);
888 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500889 __dget_dlock(alias);
890 spin_unlock(&alias->d_lock);
891 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100892 }
893 spin_unlock(&alias->d_lock);
894 goto again;
895 }
896 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
Nick Pigginda502952011-01-07 17:49:33 +1100899struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900{
David Howells214fda12006-03-25 03:06:36 -0800901 struct dentry *de = NULL;
902
Al Virob3d9b7a2012-06-09 13:51:19 -0400903 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100904 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500905 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100906 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800907 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 return de;
909}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700910EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
912/*
913 * Try to kill dentries associated with this inode.
914 * WARNING: you must own a reference to inode.
915 */
916void d_prune_aliases(struct inode *inode)
917{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700918 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100920 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400921 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700923 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400924 struct dentry *parent = lock_parent(dentry);
925 if (likely(!dentry->d_lockref.count)) {
926 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800927 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400928 goto restart;
929 }
930 if (parent)
931 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 }
933 spin_unlock(&dentry->d_lock);
934 }
Nick Piggin873feea2011-01-07 17:50:06 +1100935 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700937EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400939static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
Al Viro5c47e6d2014-04-29 16:13:18 -0400941 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700942
Miklos Szeredi60942f22014-05-02 15:38:39 -0400943 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400944 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400945 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100946 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400947 parent = lock_parent(dentry);
948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000950 * The dispose list is isolated and dentries are not accounted
951 * to the LRU here, so we can simply remove it from the list
952 * here regardless of whether it is referenced or not.
953 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400954 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000955
956 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000958 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800960 if (dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700961 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400962 if (parent)
963 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 continue;
965 }
Nick Piggin77812a12011-01-07 17:49:48 +1100966
Al Viro64fd72e2014-05-28 09:48:44 -0400967
968 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
969 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
970 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400971 if (parent)
972 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -0400973 if (can_free)
974 dentry_free(dentry);
975 continue;
976 }
977
Al Viroff2fde92014-05-28 13:59:13 -0400978 inode = dentry->d_inode;
979 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400980 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000981 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400982 if (parent)
983 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400984 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000985 }
Al Viroff2fde92014-05-28 13:59:13 -0400986
Al Viroff2fde92014-05-28 13:59:13 -0400987 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -0400988
Al Viro5c47e6d2014-04-29 16:13:18 -0400989 /*
990 * We need to prune ancestors too. This is necessary to prevent
991 * quadratic behavior of shrink_dcache_parent(), but is also
992 * expected to be beneficial in reducing dentry cache
993 * fragmentation.
994 */
995 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -0400996 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
997 parent = lock_parent(dentry);
998 if (dentry->d_lockref.count != 1) {
999 dentry->d_lockref.count--;
1000 spin_unlock(&dentry->d_lock);
1001 if (parent)
1002 spin_unlock(&parent->d_lock);
1003 break;
1004 }
1005 inode = dentry->d_inode; /* can't be NULL */
1006 if (unlikely(!spin_trylock(&inode->i_lock))) {
1007 spin_unlock(&dentry->d_lock);
1008 if (parent)
1009 spin_unlock(&parent->d_lock);
1010 cpu_relax();
1011 continue;
1012 }
1013 __dentry_kill(dentry);
1014 dentry = parent;
1015 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001017}
1018
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001019static enum lru_status dentry_lru_isolate(struct list_head *item,
1020 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +10001021{
1022 struct list_head *freeable = arg;
1023 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1024
1025
1026 /*
1027 * we are inverting the lru lock/dentry->d_lock here,
1028 * so use a trylock. If we fail to get the lock, just skip
1029 * it
1030 */
1031 if (!spin_trylock(&dentry->d_lock))
1032 return LRU_SKIP;
1033
1034 /*
1035 * Referenced dentries are still in use. If they have active
1036 * counts, just remove them from the LRU. Otherwise give them
1037 * another pass through the LRU.
1038 */
1039 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001040 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001041 spin_unlock(&dentry->d_lock);
1042 return LRU_REMOVED;
1043 }
1044
1045 if (dentry->d_flags & DCACHE_REFERENCED) {
1046 dentry->d_flags &= ~DCACHE_REFERENCED;
1047 spin_unlock(&dentry->d_lock);
1048
1049 /*
1050 * The list move itself will be made by the common LRU code. At
1051 * this point, we've dropped the dentry->d_lock but keep the
1052 * lru lock. This is safe to do, since every list movement is
1053 * protected by the lru lock even if both locks are held.
1054 *
1055 * This is guaranteed by the fact that all LRU management
1056 * functions are intermediated by the LRU API calls like
1057 * list_lru_add and list_lru_del. List movement in this file
1058 * only ever occur through this functions or through callbacks
1059 * like this one, that are called from the LRU API.
1060 *
1061 * The only exceptions to this are functions like
1062 * shrink_dentry_list, and code that first checks for the
1063 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1064 * operating only with stack provided lists after they are
1065 * properly isolated from the main list. It is thus, always a
1066 * local access.
1067 */
1068 return LRU_ROTATE;
1069 }
1070
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001071 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001072 spin_unlock(&dentry->d_lock);
1073
1074 return LRU_REMOVED;
1075}
1076
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001077/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001078 * prune_dcache_sb - shrink the dcache
1079 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001080 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001081 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001082 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1083 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001084 * function.
1085 *
1086 * This function may fail to free any resources if all the dentries are in
1087 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001088 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001089long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001090{
Dave Chinnerf6041562013-08-28 10:18:00 +10001091 LIST_HEAD(dispose);
1092 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001093
Vladimir Davydov503c3582015-02-12 14:58:47 -08001094 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1095 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001096 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001097 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098}
1099
Glauber Costa4e717f52013-08-28 10:18:03 +10001100static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001101 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001102{
Glauber Costa4e717f52013-08-28 10:18:03 +10001103 struct list_head *freeable = arg;
1104 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001105
Glauber Costa4e717f52013-08-28 10:18:03 +10001106 /*
1107 * we are inverting the lru lock/dentry->d_lock here,
1108 * so use a trylock. If we fail to get the lock, just skip
1109 * it
1110 */
1111 if (!spin_trylock(&dentry->d_lock))
1112 return LRU_SKIP;
1113
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001114 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001115 spin_unlock(&dentry->d_lock);
1116
1117 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001118}
1119
Glauber Costa4e717f52013-08-28 10:18:03 +10001120
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001121/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 * shrink_dcache_sb - shrink dcache for a superblock
1123 * @sb: superblock
1124 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001125 * Shrink the dcache for the specified super block. This is used to free
1126 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001128void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129{
Glauber Costa4e717f52013-08-28 10:18:03 +10001130 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001131
Glauber Costa4e717f52013-08-28 10:18:03 +10001132 do {
1133 LIST_HEAD(dispose);
1134
1135 freed = list_lru_walk(&sb->s_dentry_lru,
1136 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1137
1138 this_cpu_sub(nr_dentry_unused, freed);
1139 shrink_dentry_list(&dispose);
1140 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001142EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001145 * enum d_walk_ret - action to talke during tree walk
1146 * @D_WALK_CONTINUE: contrinue walk
1147 * @D_WALK_QUIT: quit walk
1148 * @D_WALK_NORETRY: quit when retry is needed
1149 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001151enum d_walk_ret {
1152 D_WALK_CONTINUE,
1153 D_WALK_QUIT,
1154 D_WALK_NORETRY,
1155 D_WALK_SKIP,
1156};
1157
1158/**
1159 * d_walk - walk the dentry tree
1160 * @parent: start of walk
1161 * @data: data passed to @enter() and @finish()
1162 * @enter: callback when first entering the dentry
1163 * @finish: callback when successfully finished the walk
1164 *
1165 * The @enter() and @finish() callbacks are called with d_lock held.
1166 */
1167static void d_walk(struct dentry *parent, void *data,
1168 enum d_walk_ret (*enter)(void *, struct dentry *),
1169 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170{
Nick Piggin949854d2011-01-07 17:49:37 +11001171 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001173 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001174 enum d_walk_ret ret;
1175 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001176
Nick Piggin58db63d2011-01-07 17:49:39 +11001177again:
Al Viro48f5ec22013-09-09 15:22:25 -04001178 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001179 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001180 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001181
1182 ret = enter(data, this_parent);
1183 switch (ret) {
1184 case D_WALK_CONTINUE:
1185 break;
1186 case D_WALK_QUIT:
1187 case D_WALK_SKIP:
1188 goto out_unlock;
1189 case D_WALK_NORETRY:
1190 retry = false;
1191 break;
1192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193repeat:
1194 next = this_parent->d_subdirs.next;
1195resume:
1196 while (next != &this_parent->d_subdirs) {
1197 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001198 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001200
1201 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001202
1203 ret = enter(data, dentry);
1204 switch (ret) {
1205 case D_WALK_CONTINUE:
1206 break;
1207 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001208 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001209 goto out_unlock;
1210 case D_WALK_NORETRY:
1211 retry = false;
1212 break;
1213 case D_WALK_SKIP:
1214 spin_unlock(&dentry->d_lock);
1215 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001216 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001217
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001219 spin_unlock(&this_parent->d_lock);
1220 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001222 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 goto repeat;
1224 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001225 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 }
1227 /*
1228 * All done at this level ... ascend and resume the search.
1229 */
Al Viroca5358e2014-10-26 19:31:10 -04001230 rcu_read_lock();
1231ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001233 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001234 this_parent = child->d_parent;
1235
Al Viro31dec132013-10-25 17:04:27 -04001236 spin_unlock(&child->d_lock);
1237 spin_lock(&this_parent->d_lock);
1238
Al Viroca5358e2014-10-26 19:31:10 -04001239 /* might go back up the wrong parent if we have had a rename. */
1240 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001241 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001242 next = child->d_child.next;
1243 while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
1244 if (next == &this_parent->d_subdirs)
1245 goto ascend;
1246 child = list_entry(next, struct dentry, d_child);
1247 next = next->next;
Al Viro31dec132013-10-25 17:04:27 -04001248 }
1249 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 goto resume;
1251 }
Al Viroca5358e2014-10-26 19:31:10 -04001252 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001253 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001254 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001255 if (finish)
1256 finish(data);
1257
1258out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001259 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001260 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001261 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001262
1263rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001264 spin_unlock(&this_parent->d_lock);
1265 rcu_read_unlock();
1266 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001267 if (!retry)
1268 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001269 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001270 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001272
1273/*
1274 * Search for at least 1 mount point in the dentry's subdirs.
1275 * We descend to the next level whenever the d_subdirs
1276 * list is non-empty and continue searching.
1277 */
1278
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001279static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1280{
1281 int *ret = data;
1282 if (d_mountpoint(dentry)) {
1283 *ret = 1;
1284 return D_WALK_QUIT;
1285 }
1286 return D_WALK_CONTINUE;
1287}
1288
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001289/**
1290 * have_submounts - check for mounts over a dentry
1291 * @parent: dentry to check.
1292 *
1293 * Return true if the parent or its subdirectories contain
1294 * a mount point
1295 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001296int have_submounts(struct dentry *parent)
1297{
1298 int ret = 0;
1299
1300 d_walk(parent, &ret, check_mount, NULL);
1301
1302 return ret;
1303}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001304EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
1306/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001307 * Called by mount code to set a mountpoint and check if the mountpoint is
1308 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1309 * subtree can become unreachable).
1310 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001311 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001312 * this reason take rename_lock and d_lock on dentry and ancestors.
1313 */
1314int d_set_mounted(struct dentry *dentry)
1315{
1316 struct dentry *p;
1317 int ret = -ENOENT;
1318 write_seqlock(&rename_lock);
1319 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001320 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001321 spin_lock(&p->d_lock);
1322 if (unlikely(d_unhashed(p))) {
1323 spin_unlock(&p->d_lock);
1324 goto out;
1325 }
1326 spin_unlock(&p->d_lock);
1327 }
1328 spin_lock(&dentry->d_lock);
1329 if (!d_unlinked(dentry)) {
1330 dentry->d_flags |= DCACHE_MOUNTED;
1331 ret = 0;
1332 }
1333 spin_unlock(&dentry->d_lock);
1334out:
1335 write_sequnlock(&rename_lock);
1336 return ret;
1337}
1338
1339/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001340 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 * and move any unused dentries to the end of the unused
1342 * list for prune_dcache(). We descend to the next level
1343 * whenever the d_subdirs list is non-empty and continue
1344 * searching.
1345 *
1346 * It returns zero iff there are no unused children,
1347 * otherwise it returns the number of children moved to
1348 * the end of the unused list. This may not be the total
1349 * number of unused children, because select_parent can
1350 * drop the lock and return early due to latency
1351 * constraints.
1352 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001353
1354struct select_data {
1355 struct dentry *start;
1356 struct list_head dispose;
1357 int found;
1358};
1359
1360static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001362 struct select_data *data = _data;
1363 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001365 if (data->start == dentry)
1366 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Al Virofe915222014-05-03 00:02:25 -04001368 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001369 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001370 } else {
1371 if (dentry->d_flags & DCACHE_LRU_LIST)
1372 d_lru_del(dentry);
1373 if (!dentry->d_lockref.count) {
1374 d_shrink_add(dentry, &data->dispose);
1375 data->found++;
1376 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 }
1378 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001379 * We can return to the caller if we have found some (this
1380 * ensures forward progress). We'll be coming back to find
1381 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 */
Al Virofe915222014-05-03 00:02:25 -04001383 if (!list_empty(&data->dispose))
1384 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001386 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387}
1388
1389/**
1390 * shrink_dcache_parent - prune dcache
1391 * @parent: parent of entries to prune
1392 *
1393 * Prune the dcache to remove unused children of the parent dentry.
1394 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001395void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001397 for (;;) {
1398 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001400 INIT_LIST_HEAD(&data.dispose);
1401 data.start = parent;
1402 data.found = 0;
1403
1404 d_walk(parent, &data, select_collect, NULL);
1405 if (!data.found)
1406 break;
1407
1408 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001409 cond_resched();
1410 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001412EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
Al Viro9c8c10e2014-05-02 20:36:10 -04001414static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001415{
Al Viro9c8c10e2014-05-02 20:36:10 -04001416 /* it has busy descendents; complain about those instead */
1417 if (!list_empty(&dentry->d_subdirs))
1418 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001419
Al Viro9c8c10e2014-05-02 20:36:10 -04001420 /* root with refcount 1 is fine */
1421 if (dentry == _data && dentry->d_lockref.count == 1)
1422 return D_WALK_CONTINUE;
1423
1424 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1425 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001426 dentry,
1427 dentry->d_inode ?
1428 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001429 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001430 dentry->d_lockref.count,
1431 dentry->d_sb->s_type->name,
1432 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001433 WARN_ON(1);
1434 return D_WALK_CONTINUE;
1435}
1436
1437static void do_one_tree(struct dentry *dentry)
1438{
1439 shrink_dcache_parent(dentry);
1440 d_walk(dentry, dentry, umount_check, NULL);
1441 d_drop(dentry);
1442 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001443}
1444
1445/*
1446 * destroy the dentries attached to a superblock on unmounting
1447 */
1448void shrink_dcache_for_umount(struct super_block *sb)
1449{
1450 struct dentry *dentry;
1451
Al Viro9c8c10e2014-05-02 20:36:10 -04001452 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001453
1454 dentry = sb->s_root;
1455 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001456 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001457
1458 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001459 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1460 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001461 }
1462}
1463
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001464struct detach_data {
1465 struct select_data select;
1466 struct dentry *mountpoint;
1467};
1468static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001469{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001470 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001471
1472 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001473 __dget_dlock(dentry);
1474 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001475 return D_WALK_QUIT;
1476 }
1477
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001478 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001479}
1480
1481static void check_and_drop(void *_data)
1482{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001483 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001484
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001485 if (!data->mountpoint && !data->select.found)
1486 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001487}
1488
1489/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001490 * d_invalidate - detach submounts, prune dcache, and drop
1491 * @dentry: dentry to invalidate (aka detach, prune and drop)
1492 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001493 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001494 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001495 * The final d_drop is done as an atomic operation relative to
1496 * rename_lock ensuring there are no races with d_set_mounted. This
1497 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001498 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001499void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001500{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001501 /*
1502 * If it's already been dropped, return OK.
1503 */
1504 spin_lock(&dentry->d_lock);
1505 if (d_unhashed(dentry)) {
1506 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001507 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001508 }
1509 spin_unlock(&dentry->d_lock);
1510
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001511 /* Negative dentries can be dropped without further checks */
1512 if (!dentry->d_inode) {
1513 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001514 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001515 }
1516
1517 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001518 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001519
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001520 data.mountpoint = NULL;
1521 INIT_LIST_HEAD(&data.select.dispose);
1522 data.select.start = dentry;
1523 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001524
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001525 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001526
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001527 if (data.select.found)
1528 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001529
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001530 if (data.mountpoint) {
1531 detach_mounts(data.mountpoint);
1532 dput(data.mountpoint);
1533 }
1534
1535 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001536 break;
1537
1538 cond_resched();
1539 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001540}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001541EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001542
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543/**
Al Viroa4464db2011-07-07 15:03:58 -04001544 * __d_alloc - allocate a dcache entry
1545 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 * @name: qstr of the name
1547 *
1548 * Allocates a dentry. It returns %NULL if there is insufficient memory
1549 * available. On a success the dentry is returned. The name passed in is
1550 * copied and the copy passed in may be reused after this call.
1551 */
1552
Al Viroa4464db2011-07-07 15:03:58 -04001553struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554{
1555 struct dentry *dentry;
1556 char *dname;
1557
Mel Gormane12ba742007-10-16 01:25:52 -07001558 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 if (!dentry)
1560 return NULL;
1561
Linus Torvalds6326c712012-05-21 16:14:04 -07001562 /*
1563 * We guarantee that the inline name is always NUL-terminated.
1564 * This way the memcpy() done by the name switching in rename
1565 * will still always have a NUL at the end, even if we might
1566 * be overwriting an internal NUL character
1567 */
1568 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001570 size_t size = offsetof(struct external_name, name[1]);
1571 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1572 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 kmem_cache_free(dentry_cache, dentry);
1574 return NULL;
1575 }
Al Viro8d85b482014-09-29 14:54:27 -04001576 atomic_set(&p->u.count, 1);
1577 dname = p->name;
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -08001578 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1579 kasan_unpoison_shadow(dname,
1580 round_up(name->len + 1, sizeof(unsigned long)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 } else {
1582 dname = dentry->d_iname;
1583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
1585 dentry->d_name.len = name->len;
1586 dentry->d_name.hash = name->hash;
1587 memcpy(dname, name->name, name->len);
1588 dname[name->len] = 0;
1589
Linus Torvalds6326c712012-05-21 16:14:04 -07001590 /* Make sure we always see the terminating NUL character */
1591 smp_wmb();
1592 dentry->d_name.name = dname;
1593
Waiman Long98474232013-08-28 18:24:59 -07001594 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001595 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001597 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001599 dentry->d_parent = dentry;
1600 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 dentry->d_op = NULL;
1602 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001603 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 INIT_LIST_HEAD(&dentry->d_lru);
1605 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001606 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1607 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001608 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
Nick Piggin3e880fb2011-01-07 17:49:19 +11001610 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001611
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 return dentry;
1613}
Al Viroa4464db2011-07-07 15:03:58 -04001614
1615/**
1616 * d_alloc - allocate a dcache entry
1617 * @parent: parent of entry to allocate
1618 * @name: qstr of the name
1619 *
1620 * Allocates a dentry. It returns %NULL if there is insufficient memory
1621 * available. On a success the dentry is returned. The name passed in is
1622 * copied and the copy passed in may be reused after this call.
1623 */
1624struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1625{
1626 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1627 if (!dentry)
1628 return NULL;
1629
1630 spin_lock(&parent->d_lock);
1631 /*
1632 * don't need child lock because it is not subject
1633 * to concurrency here
1634 */
1635 __dget_dlock(parent);
1636 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001637 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001638 spin_unlock(&parent->d_lock);
1639
1640 return dentry;
1641}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001642EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001644/**
1645 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1646 * @sb: the superblock
1647 * @name: qstr of the name
1648 *
1649 * For a filesystem that just pins its dentries in memory and never
1650 * performs lookups at all, return an unhashed IS_ROOT dentry.
1651 */
Nick Piggin4b936882011-01-07 17:50:07 +11001652struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1653{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001654 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001655}
1656EXPORT_SYMBOL(d_alloc_pseudo);
1657
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1659{
1660 struct qstr q;
1661
1662 q.name = name;
1663 q.len = strlen(name);
1664 q.hash = full_name_hash(q.name, q.len);
1665 return d_alloc(parent, &q);
1666}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001667EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
Nick Pigginfb045ad2011-01-07 17:49:55 +11001669void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1670{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001671 WARN_ON_ONCE(dentry->d_op);
1672 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001673 DCACHE_OP_COMPARE |
1674 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001675 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001676 DCACHE_OP_DELETE ));
1677 dentry->d_op = op;
1678 if (!op)
1679 return;
1680 if (op->d_hash)
1681 dentry->d_flags |= DCACHE_OP_HASH;
1682 if (op->d_compare)
1683 dentry->d_flags |= DCACHE_OP_COMPARE;
1684 if (op->d_revalidate)
1685 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001686 if (op->d_weak_revalidate)
1687 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001688 if (op->d_delete)
1689 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001690 if (op->d_prune)
1691 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001692
1693}
1694EXPORT_SYMBOL(d_set_d_op);
1695
David Howellsdf1a0852015-01-29 12:02:28 +00001696
1697/*
1698 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1699 * @dentry - The dentry to mark
1700 *
1701 * Mark a dentry as falling through to the lower layer (as set with
1702 * d_pin_lower()). This flag may be recorded on the medium.
1703 */
1704void d_set_fallthru(struct dentry *dentry)
1705{
1706 spin_lock(&dentry->d_lock);
1707 dentry->d_flags |= DCACHE_FALLTHRU;
1708 spin_unlock(&dentry->d_lock);
1709}
1710EXPORT_SYMBOL(d_set_fallthru);
1711
David Howellsb18825a2013-09-12 19:22:53 +01001712static unsigned d_flags_for_inode(struct inode *inode)
1713{
David Howells44bdb5e2015-01-29 12:02:29 +00001714 unsigned add_flags = DCACHE_REGULAR_TYPE;
David Howellsb18825a2013-09-12 19:22:53 +01001715
1716 if (!inode)
1717 return DCACHE_MISS_TYPE;
1718
1719 if (S_ISDIR(inode->i_mode)) {
1720 add_flags = DCACHE_DIRECTORY_TYPE;
1721 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1722 if (unlikely(!inode->i_op->lookup))
1723 add_flags = DCACHE_AUTODIR_TYPE;
1724 else
1725 inode->i_opflags |= IOP_LOOKUP;
1726 }
David Howells44bdb5e2015-01-29 12:02:29 +00001727 goto type_determined;
David Howellsb18825a2013-09-12 19:22:53 +01001728 }
1729
David Howells44bdb5e2015-01-29 12:02:29 +00001730 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1731 if (unlikely(inode->i_op->follow_link)) {
1732 add_flags = DCACHE_SYMLINK_TYPE;
1733 goto type_determined;
1734 }
1735 inode->i_opflags |= IOP_NOFOLLOW;
1736 }
1737
1738 if (unlikely(!S_ISREG(inode->i_mode)))
1739 add_flags = DCACHE_SPECIAL_TYPE;
1740
1741type_determined:
David Howellsb18825a2013-09-12 19:22:53 +01001742 if (unlikely(IS_AUTOMOUNT(inode)))
1743 add_flags |= DCACHE_NEED_AUTOMOUNT;
1744 return add_flags;
1745}
1746
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001747static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1748{
David Howellsb18825a2013-09-12 19:22:53 +01001749 unsigned add_flags = d_flags_for_inode(inode);
1750
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001751 spin_lock(&dentry->d_lock);
David Howellsb18825a2013-09-12 19:22:53 +01001752 if (inode)
Al Viro946e51f2014-10-26 19:19:16 -04001753 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
David Howells4bf46a22015-03-05 14:09:22 +00001754 __d_set_inode_and_type(dentry, inode, add_flags);
Nick Piggin31e6b012011-01-07 17:49:52 +11001755 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001756 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001757 fsnotify_d_instantiate(dentry, inode);
1758}
1759
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760/**
1761 * d_instantiate - fill in inode information for a dentry
1762 * @entry: dentry to complete
1763 * @inode: inode to attach to this dentry
1764 *
1765 * Fill in inode information in the entry.
1766 *
1767 * This turns negative dentries into productive full members
1768 * of society.
1769 *
1770 * NOTE! This assumes that the inode count has been incremented
1771 * (or otherwise set) by the caller to indicate that it is now
1772 * in use by the dcache.
1773 */
1774
1775void d_instantiate(struct dentry *entry, struct inode * inode)
1776{
Al Viro946e51f2014-10-26 19:19:16 -04001777 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001778 if (inode)
1779 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001780 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001781 if (inode)
1782 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 security_d_instantiate(entry, inode);
1784}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001785EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
1787/**
1788 * d_instantiate_unique - instantiate a non-aliased dentry
1789 * @entry: dentry to instantiate
1790 * @inode: inode to attach to this dentry
1791 *
1792 * Fill in inode information in the entry. On success, it returns NULL.
1793 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001794 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 *
1796 * Note that in order to avoid conflicts with rename() etc, the caller
1797 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001798 *
1799 * This also assumes that the inode count has been incremented
1800 * (or otherwise set) by the caller to indicate that it is now
1801 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 */
David Howells770bfad2006-08-22 20:06:07 -04001803static struct dentry *__d_instantiate_unique(struct dentry *entry,
1804 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805{
1806 struct dentry *alias;
1807 int len = entry->d_name.len;
1808 const char *name = entry->d_name.name;
1809 unsigned int hash = entry->d_name.hash;
1810
David Howells770bfad2006-08-22 20:06:07 -04001811 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001812 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001813 return NULL;
1814 }
1815
Al Viro946e51f2014-10-26 19:19:16 -04001816 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001817 /*
1818 * Don't need alias->d_lock here, because aliases with
1819 * d_parent == entry->d_parent are not subject to name or
1820 * parent changes, because the parent inode i_mutex is held.
1821 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001822 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 continue;
1824 if (alias->d_parent != entry->d_parent)
1825 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001826 if (alias->d_name.len != len)
1827 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001828 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001830 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 return alias;
1832 }
David Howells770bfad2006-08-22 20:06:07 -04001833
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001834 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 return NULL;
1836}
David Howells770bfad2006-08-22 20:06:07 -04001837
1838struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1839{
1840 struct dentry *result;
1841
Al Viro946e51f2014-10-26 19:19:16 -04001842 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001843
Nick Piggin873feea2011-01-07 17:50:06 +11001844 if (inode)
1845 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001846 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001847 if (inode)
1848 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001849
1850 if (!result) {
1851 security_d_instantiate(entry, inode);
1852 return NULL;
1853 }
1854
1855 BUG_ON(!d_unhashed(result));
1856 iput(inode);
1857 return result;
1858}
1859
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860EXPORT_SYMBOL(d_instantiate_unique);
1861
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001862/**
1863 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1864 * @entry: dentry to complete
1865 * @inode: inode to attach to this dentry
1866 *
1867 * Fill in inode information in the entry. If a directory alias is found, then
1868 * return an error (and drop inode). Together with d_materialise_unique() this
1869 * guarantees that a directory inode may never have more than one alias.
1870 */
1871int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1872{
Al Viro946e51f2014-10-26 19:19:16 -04001873 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001874
1875 spin_lock(&inode->i_lock);
1876 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1877 spin_unlock(&inode->i_lock);
1878 iput(inode);
1879 return -EBUSY;
1880 }
1881 __d_instantiate(entry, inode);
1882 spin_unlock(&inode->i_lock);
1883 security_d_instantiate(entry, inode);
1884
1885 return 0;
1886}
1887EXPORT_SYMBOL(d_instantiate_no_diralias);
1888
Al Viroadc0e912012-01-08 16:49:21 -05001889struct dentry *d_make_root(struct inode *root_inode)
1890{
1891 struct dentry *res = NULL;
1892
1893 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001894 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001895
1896 res = __d_alloc(root_inode->i_sb, &name);
1897 if (res)
1898 d_instantiate(res, root_inode);
1899 else
1900 iput(root_inode);
1901 }
1902 return res;
1903}
1904EXPORT_SYMBOL(d_make_root);
1905
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001906static struct dentry * __d_find_any_alias(struct inode *inode)
1907{
1908 struct dentry *alias;
1909
Al Virob3d9b7a2012-06-09 13:51:19 -04001910 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001911 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001912 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001913 __dget(alias);
1914 return alias;
1915}
1916
Sage Weil46f72b32012-01-10 09:04:37 -08001917/**
1918 * d_find_any_alias - find any alias for a given inode
1919 * @inode: inode to find an alias for
1920 *
1921 * If any aliases exist for the given inode, take and return a
1922 * reference for one of them. If no aliases exist, return %NULL.
1923 */
1924struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001925{
1926 struct dentry *de;
1927
1928 spin_lock(&inode->i_lock);
1929 de = __d_find_any_alias(inode);
1930 spin_unlock(&inode->i_lock);
1931 return de;
1932}
Sage Weil46f72b32012-01-10 09:04:37 -08001933EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001934
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001935static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001936{
NeilBrownb911a6b2012-11-08 16:09:37 -08001937 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001938 struct dentry *tmp;
1939 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001940 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001941
1942 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001943 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001944 if (IS_ERR(inode))
1945 return ERR_CAST(inode);
1946
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001947 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001948 if (res)
1949 goto out_iput;
1950
Al Viroa4464db2011-07-07 15:03:58 -04001951 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001952 if (!tmp) {
1953 res = ERR_PTR(-ENOMEM);
1954 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001955 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001956
Nick Piggin873feea2011-01-07 17:50:06 +11001957 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001958 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001959 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001960 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001961 dput(tmp);
1962 goto out_iput;
1963 }
1964
1965 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001966 add_flags = d_flags_for_inode(inode);
1967
1968 if (disconnected)
1969 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001970
Christoph Hellwig9308a612008-08-11 15:49:12 +02001971 spin_lock(&tmp->d_lock);
David Howells4bf46a22015-03-05 14:09:22 +00001972 __d_set_inode_and_type(tmp, inode, add_flags);
Al Viro946e51f2014-10-26 19:19:16 -04001973 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001974 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001975 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001976 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001977 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001978 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001979 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001980
Christoph Hellwig9308a612008-08-11 15:49:12 +02001981 return tmp;
1982
1983 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001984 if (res && !IS_ERR(res))
1985 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001986 iput(inode);
1987 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001988}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001989
1990/**
1991 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1992 * @inode: inode to allocate the dentry for
1993 *
1994 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1995 * similar open by handle operations. The returned dentry may be anonymous,
1996 * or may have a full name (if the inode was already in the cache).
1997 *
1998 * When called on a directory inode, we must ensure that the inode only ever
1999 * has one dentry. If a dentry is found, that is returned instead of
2000 * allocating a new one.
2001 *
2002 * On successful return, the reference to the inode has been transferred
2003 * to the dentry. In case of an error the reference on the inode is released.
2004 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2005 * be passed in and the error will be propagated to the return value,
2006 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2007 */
2008struct dentry *d_obtain_alias(struct inode *inode)
2009{
2010 return __d_obtain_alias(inode, 1);
2011}
Benny Halevyadc48722009-02-27 14:02:59 -08002012EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013
2014/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002015 * d_obtain_root - find or allocate a dentry for a given inode
2016 * @inode: inode to allocate the dentry for
2017 *
2018 * Obtain an IS_ROOT dentry for the root of a filesystem.
2019 *
2020 * We must ensure that directory inodes only ever have one dentry. If a
2021 * dentry is found, that is returned instead of allocating a new one.
2022 *
2023 * On successful return, the reference to the inode has been transferred
2024 * to the dentry. In case of an error the reference on the inode is
2025 * released. A %NULL or IS_ERR inode may be passed in and will be the
2026 * error will be propagate to the return value, with a %NULL @inode
2027 * replaced by ERR_PTR(-ESTALE).
2028 */
2029struct dentry *d_obtain_root(struct inode *inode)
2030{
2031 return __d_obtain_alias(inode, 0);
2032}
2033EXPORT_SYMBOL(d_obtain_root);
2034
2035/**
Barry Naujok94035402008-05-21 16:50:46 +10002036 * d_add_ci - lookup or allocate new dentry with case-exact name
2037 * @inode: the inode case-insensitive lookup has found
2038 * @dentry: the negative dentry that was passed to the parent's lookup func
2039 * @name: the case-exact name to be associated with the returned dentry
2040 *
2041 * This is to avoid filling the dcache with case-insensitive names to the
2042 * same inode, only the actual correct case is stored in the dcache for
2043 * case-insensitive filesystems.
2044 *
2045 * For a case-insensitive lookup match and if the the case-exact dentry
2046 * already exists in in the dcache, use it and return it.
2047 *
2048 * If no entry exists with the exact case name, allocate new dentry with
2049 * the exact case, and return the spliced entry.
2050 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02002051struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10002052 struct qstr *name)
2053{
Barry Naujok94035402008-05-21 16:50:46 +10002054 struct dentry *found;
2055 struct dentry *new;
2056
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002057 /*
2058 * First check if a dentry matching the name already exists,
2059 * if not go ahead and create it now.
2060 */
Barry Naujok94035402008-05-21 16:50:46 +10002061 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10002062 if (!found) {
2063 new = d_alloc(dentry->d_parent, name);
2064 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05002065 found = ERR_PTR(-ENOMEM);
Al Viro427c77d2014-10-12 18:46:26 -04002066 } else {
2067 found = d_splice_alias(inode, new);
2068 if (found) {
2069 dput(new);
2070 return found;
2071 }
2072 return new;
Barry Naujok94035402008-05-21 16:50:46 +10002073 }
Barry Naujok94035402008-05-21 16:50:46 +10002074 }
Barry Naujok94035402008-05-21 16:50:46 +10002075 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05002076 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002077}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002078EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002080/*
2081 * Do the slow-case of the dentry name compare.
2082 *
2083 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07002084 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002085 * filesystem can rely on them, and can use the 'name' and
2086 * 'len' information without worrying about walking off the
2087 * end of memory etc.
2088 *
2089 * Thus the read_seqcount_retry() and the "duplicate" info
2090 * in arguments (the low-level filesystem should not look
2091 * at the dentry inode or name contents directly, since
2092 * rename can change them while we're in RCU mode).
2093 */
2094enum slow_d_compare {
2095 D_COMP_OK,
2096 D_COMP_NOMATCH,
2097 D_COMP_SEQRETRY,
2098};
2099
2100static noinline enum slow_d_compare slow_dentry_cmp(
2101 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002102 struct dentry *dentry,
2103 unsigned int seq,
2104 const struct qstr *name)
2105{
2106 int tlen = dentry->d_name.len;
2107 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002108
2109 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2110 cpu_relax();
2111 return D_COMP_SEQRETRY;
2112 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002113 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002114 return D_COMP_NOMATCH;
2115 return D_COMP_OK;
2116}
2117
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002119 * __d_lookup_rcu - search for a dentry (racy, store-free)
2120 * @parent: parent dentry
2121 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002122 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002123 * Returns: dentry, or NULL
2124 *
2125 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2126 * resolution (store-free path walking) design described in
2127 * Documentation/filesystems/path-lookup.txt.
2128 *
2129 * This is not to be used outside core vfs.
2130 *
2131 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2132 * held, and rcu_read_lock held. The returned dentry must not be stored into
2133 * without taking d_lock and checking d_seq sequence count against @seq
2134 * returned here.
2135 *
Linus Torvalds15570082013-09-02 11:38:06 -07002136 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002137 * function.
2138 *
2139 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2140 * the returned dentry, so long as its parent's seqlock is checked after the
2141 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2142 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002143 *
2144 * NOTE! The caller *has* to check the resulting dentry against the sequence
2145 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002146 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002147struct dentry *__d_lookup_rcu(const struct dentry *parent,
2148 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002149 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002150{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002151 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002152 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002153 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002154 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002155 struct dentry *dentry;
2156
2157 /*
2158 * Note: There is significant duplication with __d_lookup_rcu which is
2159 * required to prevent single threaded performance regressions
2160 * especially on architectures where smp_rmb (in seqcounts) are costly.
2161 * Keep the two functions in sync.
2162 */
2163
2164 /*
2165 * The hash list is protected using RCU.
2166 *
2167 * Carefully use d_seq when comparing a candidate dentry, to avoid
2168 * races with d_move().
2169 *
2170 * It is possible that concurrent renames can mess up our list
2171 * walk here and result in missing our dentry, resulting in the
2172 * false-negative result. d_lookup() protects against concurrent
2173 * renames using rename_lock seqlock.
2174 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002175 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002176 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002177 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002178 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002179
Nick Piggin31e6b012011-01-07 17:49:52 +11002180seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002181 /*
2182 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002183 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002184 *
2185 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002186 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002187 *
2188 * NOTE! We do a "raw" seqcount_begin here. That means that
2189 * we don't wait for the sequence count to stabilize if it
2190 * is in the middle of a sequence change. If we do the slow
2191 * dentry compare, we will do seqretries until it is stable,
2192 * and if we end up with a successful lookup, we actually
2193 * want to exit RCU lookup anyway.
2194 */
2195 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002196 if (dentry->d_parent != parent)
2197 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002198 if (d_unhashed(dentry))
2199 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002200
2201 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002202 if (dentry->d_name.hash != hashlen_hash(hashlen))
2203 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002204 *seqp = seq;
2205 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002206 case D_COMP_OK:
2207 return dentry;
2208 case D_COMP_NOMATCH:
2209 continue;
2210 default:
2211 goto seqretry;
2212 }
2213 }
2214
Linus Torvalds26fe5752012-05-10 13:14:12 -07002215 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002216 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002217 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002218 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002219 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002220 }
2221 return NULL;
2222}
2223
2224/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 * d_lookup - search for a dentry
2226 * @parent: parent dentry
2227 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002228 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002230 * d_lookup searches the children of the parent dentry for the name in
2231 * question. If the dentry is found its reference count is incremented and the
2232 * dentry is returned. The caller must use dput to free the entry when it has
2233 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 */
Al Viroda2d8452013-01-24 18:29:34 -05002235struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236{
Nick Piggin31e6b012011-01-07 17:49:52 +11002237 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002238 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239
Daeseok Younb8314f92014-08-11 11:46:53 +09002240 do {
2241 seq = read_seqbegin(&rename_lock);
2242 dentry = __d_lookup(parent, name);
2243 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 break;
2245 } while (read_seqretry(&rename_lock, seq));
2246 return dentry;
2247}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002248EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249
Nick Piggin31e6b012011-01-07 17:49:52 +11002250/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002251 * __d_lookup - search for a dentry (racy)
2252 * @parent: parent dentry
2253 * @name: qstr of name we wish to find
2254 * Returns: dentry, or NULL
2255 *
2256 * __d_lookup is like d_lookup, however it may (rarely) return a
2257 * false-negative result due to unrelated rename activity.
2258 *
2259 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2260 * however it must be used carefully, eg. with a following d_lookup in
2261 * the case of failure.
2262 *
2263 * __d_lookup callers must be commented.
2264 */
Al Viroa713ca22013-01-24 18:27:00 -05002265struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266{
2267 unsigned int len = name->len;
2268 unsigned int hash = name->hash;
2269 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002270 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002271 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002272 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002273 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274
Nick Pigginb04f7842010-08-18 04:37:34 +10002275 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002276 * Note: There is significant duplication with __d_lookup_rcu which is
2277 * required to prevent single threaded performance regressions
2278 * especially on architectures where smp_rmb (in seqcounts) are costly.
2279 * Keep the two functions in sync.
2280 */
2281
2282 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002283 * The hash list is protected using RCU.
2284 *
2285 * Take d_lock when comparing a candidate dentry, to avoid races
2286 * with d_move().
2287 *
2288 * It is possible that concurrent renames can mess up our list
2289 * walk here and result in missing our dentry, resulting in the
2290 * false-negative result. d_lookup() protects against concurrent
2291 * renames using rename_lock seqlock.
2292 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002293 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002294 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 rcu_read_lock();
2296
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002297 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 if (dentry->d_name.hash != hash)
2300 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
2302 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 if (dentry->d_parent != parent)
2304 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002305 if (d_unhashed(dentry))
2306 goto next;
2307
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 /*
2309 * It is safe to compare names since d_move() cannot
2310 * change the qstr (protected by d_lock).
2311 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002312 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002313 int tlen = dentry->d_name.len;
2314 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002315 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 goto next;
2317 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002318 if (dentry->d_name.len != len)
2319 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002320 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 goto next;
2322 }
2323
Waiman Long98474232013-08-28 18:24:59 -07002324 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002325 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 spin_unlock(&dentry->d_lock);
2327 break;
2328next:
2329 spin_unlock(&dentry->d_lock);
2330 }
2331 rcu_read_unlock();
2332
2333 return found;
2334}
2335
2336/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002337 * d_hash_and_lookup - hash the qstr then search for a dentry
2338 * @dir: Directory to search in
2339 * @name: qstr of name we wish to find
2340 *
Al Viro4f522a22013-02-11 23:20:37 -05002341 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002342 */
2343struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2344{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002345 /*
2346 * Check for a fs-specific hash function. Note that we must
2347 * calculate the standard hash first, as the d_op->d_hash()
2348 * routine may choose to leave the hash value unchanged.
2349 */
2350 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002351 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002352 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002353 if (unlikely(err < 0))
2354 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002355 }
Al Viro4f522a22013-02-11 23:20:37 -05002356 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002357}
Al Viro4f522a22013-02-11 23:20:37 -05002358EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002359
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360/*
2361 * When a file is deleted, we have two options:
2362 * - turn this dentry into a negative dentry
2363 * - unhash this dentry and free it.
2364 *
2365 * Usually, we want to just turn this into
2366 * a negative dentry, but if anybody else is
2367 * currently using the dentry or the inode
2368 * we can't do that and we fall back on removing
2369 * it from the hash queues and waiting for
2370 * it to be deleted later when it has no users
2371 */
2372
2373/**
2374 * d_delete - delete a dentry
2375 * @dentry: The dentry to delete
2376 *
2377 * Turn the dentry into a negative dentry if possible, otherwise
2378 * remove it from the hash queues so it can be deleted later
2379 */
2380
2381void d_delete(struct dentry * dentry)
2382{
Nick Piggin873feea2011-01-07 17:50:06 +11002383 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002384 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 /*
2386 * Are we the only user?
2387 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002388again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002390 inode = dentry->d_inode;
2391 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002392 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002393 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002394 spin_unlock(&dentry->d_lock);
2395 cpu_relax();
2396 goto again;
2397 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002398 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002399 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002400 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 return;
2402 }
2403
2404 if (!d_unhashed(dentry))
2405 __d_drop(dentry);
2406
2407 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002408
2409 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002411EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002413static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002415 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002416 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002417 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002418 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002419 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420}
2421
David Howells770bfad2006-08-22 20:06:07 -04002422static void _d_rehash(struct dentry * entry)
2423{
2424 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2425}
2426
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427/**
2428 * d_rehash - add an entry back to the hash
2429 * @entry: dentry to add to the hash
2430 *
2431 * Adds a dentry to the hash according to its name.
2432 */
2433
2434void d_rehash(struct dentry * entry)
2435{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002437 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002440EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002442/**
2443 * dentry_update_name_case - update case insensitive dentry with a new name
2444 * @dentry: dentry to be updated
2445 * @name: new name
2446 *
2447 * Update a case insensitive dentry with new case of name.
2448 *
2449 * dentry must have been returned by d_lookup with name @name. Old and new
2450 * name lengths must match (ie. no d_compare which allows mismatched name
2451 * lengths).
2452 *
2453 * Parent inode i_mutex must be held over d_lookup and into this call (to
2454 * keep renames and concurrent inserts, and readdir(2) away).
2455 */
2456void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2457{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002458 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002459 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2460
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002461 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002462 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002463 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002464 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002465 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002466}
2467EXPORT_SYMBOL(dentry_update_name_case);
2468
Al Viro8d85b482014-09-29 14:54:27 -04002469static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470{
Al Viro8d85b482014-09-29 14:54:27 -04002471 if (unlikely(dname_external(target))) {
2472 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 /*
2474 * Both external: swap the pointers
2475 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002476 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 } else {
2478 /*
2479 * dentry:internal, target:external. Steal target's
2480 * storage and make target internal.
2481 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002482 memcpy(target->d_iname, dentry->d_name.name,
2483 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 dentry->d_name.name = target->d_name.name;
2485 target->d_name.name = target->d_iname;
2486 }
2487 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002488 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 /*
2490 * dentry:external, target:internal. Give dentry's
2491 * storage to target and make dentry internal
2492 */
2493 memcpy(dentry->d_iname, target->d_name.name,
2494 target->d_name.len + 1);
2495 target->d_name.name = dentry->d_name.name;
2496 dentry->d_name.name = dentry->d_iname;
2497 } else {
2498 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002499 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002501 unsigned int i;
2502 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002503 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2504 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002505 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2506 swap(((long *) &dentry->d_iname)[i],
2507 ((long *) &target->d_iname)[i]);
2508 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 }
2510 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002511 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512}
2513
Al Viro8d85b482014-09-29 14:54:27 -04002514static void copy_name(struct dentry *dentry, struct dentry *target)
2515{
2516 struct external_name *old_name = NULL;
2517 if (unlikely(dname_external(dentry)))
2518 old_name = external_name(dentry);
2519 if (unlikely(dname_external(target))) {
2520 atomic_inc(&external_name(target)->u.count);
2521 dentry->d_name = target->d_name;
2522 } else {
2523 memcpy(dentry->d_iname, target->d_name.name,
2524 target->d_name.len + 1);
2525 dentry->d_name.name = dentry->d_iname;
2526 dentry->d_name.hash_len = target->d_name.hash_len;
2527 }
2528 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2529 kfree_rcu(old_name, u.head);
2530}
2531
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002532static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2533{
2534 /*
2535 * XXXX: do we really need to take target->d_lock?
2536 */
2537 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2538 spin_lock(&target->d_parent->d_lock);
2539 else {
2540 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2541 spin_lock(&dentry->d_parent->d_lock);
2542 spin_lock_nested(&target->d_parent->d_lock,
2543 DENTRY_D_LOCK_NESTED);
2544 } else {
2545 spin_lock(&target->d_parent->d_lock);
2546 spin_lock_nested(&dentry->d_parent->d_lock,
2547 DENTRY_D_LOCK_NESTED);
2548 }
2549 }
2550 if (target < dentry) {
2551 spin_lock_nested(&target->d_lock, 2);
2552 spin_lock_nested(&dentry->d_lock, 3);
2553 } else {
2554 spin_lock_nested(&dentry->d_lock, 2);
2555 spin_lock_nested(&target->d_lock, 3);
2556 }
2557}
2558
Al Viro986c0192014-09-26 23:11:15 -04002559static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002560{
2561 if (target->d_parent != dentry->d_parent)
2562 spin_unlock(&dentry->d_parent->d_lock);
2563 if (target->d_parent != target)
2564 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002565 spin_unlock(&target->d_lock);
2566 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002567}
2568
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002570 * When switching names, the actual string doesn't strictly have to
2571 * be preserved in the target - because we're dropping the target
2572 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002573 * the new name before we switch, unless we are going to rehash
2574 * it. Note that if we *do* unhash the target, we are not allowed
2575 * to rehash it without giving it a new name/hash key - whether
2576 * we swap or overwrite the names here, resulting name won't match
2577 * the reality in filesystem; it's only there for d_path() purposes.
2578 * Note that all of this is happening under rename_lock, so the
2579 * any hash lookup seeing it in the middle of manipulations will
2580 * be discarded anyway. So we do not care what happens to the hash
2581 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002583/*
Al Viro18367502011-07-12 21:42:24 -04002584 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 * @dentry: entry to move
2586 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002587 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 *
2589 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002590 * dcache entries should not be moved in this way. Caller must hold
2591 * rename_lock, the i_mutex of the source and target directories,
2592 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002594static void __d_move(struct dentry *dentry, struct dentry *target,
2595 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 if (!dentry->d_inode)
2598 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2599
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002600 BUG_ON(d_ancestor(dentry, target));
2601 BUG_ON(d_ancestor(target, dentry));
2602
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002603 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604
Nick Piggin31e6b012011-01-07 17:49:52 +11002605 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002606 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002607
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002608 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2609
2610 /*
2611 * Move the dentry to the target hash queue. Don't bother checking
2612 * for the same hash queue because of how unlikely it is.
2613 */
2614 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002615 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
Miklos Szeredida1ce062014-04-01 17:08:43 +02002617 /*
2618 * Unhash the target (d_delete() is not usable here). If exchanging
2619 * the two dentries, then rehash onto the other's hash queue.
2620 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002622 if (exchange) {
2623 __d_rehash(target,
2624 d_hash(dentry->d_parent, dentry->d_name.hash));
2625 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002628 if (exchange)
2629 swap_names(dentry, target);
2630 else
2631 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632
Al Viro63cf4272014-09-26 23:06:14 -04002633 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002635 /* splicing a tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 dentry->d_parent = target->d_parent;
2637 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002638 list_del_init(&target->d_child);
2639 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002641 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002642 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002643 list_move(&target->d_child, &target->d_parent->d_subdirs);
2644 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002645 if (exchange)
2646 fsnotify_d_move(target);
2647 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 }
2649
Nick Piggin31e6b012011-01-07 17:49:52 +11002650 write_seqcount_end(&target->d_seq);
2651 write_seqcount_end(&dentry->d_seq);
2652
Al Viro986c0192014-09-26 23:11:15 -04002653 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002654}
2655
2656/*
2657 * d_move - move a dentry
2658 * @dentry: entry to move
2659 * @target: new dentry
2660 *
2661 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002662 * dcache entries should not be moved in this way. See the locking
2663 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002664 */
2665void d_move(struct dentry *dentry, struct dentry *target)
2666{
2667 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002668 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002670}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002671EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
Miklos Szeredida1ce062014-04-01 17:08:43 +02002673/*
2674 * d_exchange - exchange two dentries
2675 * @dentry1: first dentry
2676 * @dentry2: second dentry
2677 */
2678void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2679{
2680 write_seqlock(&rename_lock);
2681
2682 WARN_ON(!dentry1->d_inode);
2683 WARN_ON(!dentry2->d_inode);
2684 WARN_ON(IS_ROOT(dentry1));
2685 WARN_ON(IS_ROOT(dentry2));
2686
2687 __d_move(dentry1, dentry2, true);
2688
2689 write_sequnlock(&rename_lock);
2690}
2691
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002692/**
2693 * d_ancestor - search for an ancestor
2694 * @p1: ancestor dentry
2695 * @p2: child dentry
2696 *
2697 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2698 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002699 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002700struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002701{
2702 struct dentry *p;
2703
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002704 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002705 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002706 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002707 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002708 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002709}
2710
2711/*
2712 * This helper attempts to cope with remotely renamed directories
2713 *
2714 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002715 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002716 *
2717 * Note: If ever the locking in lock_rename() changes, then please
2718 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002719 */
Al Virob5ae6b12014-10-12 22:16:02 -04002720static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002721 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002722{
2723 struct mutex *m1 = NULL, *m2 = NULL;
J. Bruce Fields3d330dc2015-02-10 10:55:53 -05002724 int ret = -ESTALE;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002725
2726 /* If alias and dentry share a parent, then no extra locks required */
2727 if (alias->d_parent == dentry->d_parent)
2728 goto out_unalias;
2729
Trond Myklebust9eaef272006-10-21 10:24:20 -07002730 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002731 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2732 goto out_err;
2733 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2734 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2735 goto out_err;
2736 m2 = &alias->d_parent->d_inode->i_mutex;
2737out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002738 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04002739 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002740out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002741 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002742 if (m2)
2743 mutex_unlock(m2);
2744 if (m1)
2745 mutex_unlock(m1);
2746 return ret;
2747}
2748
David Howells770bfad2006-08-22 20:06:07 -04002749/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002750 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2751 * @inode: the inode which may have a disconnected dentry
2752 * @dentry: a negative dentry which we want to point to the inode.
2753 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002754 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2755 * place of the given dentry and return it, else simply d_add the inode
2756 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002757 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002758 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2759 * we should error out: directories can't have multiple aliases.
2760 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002761 * This is needed in the lookup routine of any filesystem that is exportable
2762 * (via knfsd) so that we can build dcache paths to directories effectively.
2763 *
2764 * If a dentry was found and moved, then it is returned. Otherwise NULL
2765 * is returned. This matches the expected return value of ->lookup.
2766 *
2767 * Cluster filesystems may call this function with a negative, hashed dentry.
2768 * In that case, we know that the inode will be a regular file, and also this
2769 * will only occur during atomic_open. So we need to check for the dentry
2770 * being already hashed only in the final case.
2771 */
2772struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2773{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002774 if (IS_ERR(inode))
2775 return ERR_CAST(inode);
2776
David Howells770bfad2006-08-22 20:06:07 -04002777 BUG_ON(!d_unhashed(dentry));
2778
David Howells770bfad2006-08-22 20:06:07 -04002779 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002780 __d_instantiate(dentry, NULL);
Al Virob5ae6b12014-10-12 22:16:02 -04002781 goto out;
David Howells770bfad2006-08-22 20:06:07 -04002782 }
Nick Piggin873feea2011-01-07 17:50:06 +11002783 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002784 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04002785 struct dentry *new = __d_find_any_alias(inode);
2786 if (unlikely(new)) {
Al Viro18367502011-07-12 21:42:24 -04002787 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002788 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04002789 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002790 spin_unlock(&inode->i_lock);
2791 dput(new);
2792 new = ERR_PTR(-ELOOP);
2793 pr_warn_ratelimited(
2794 "VFS: Lookup of '%s' in %s %s"
2795 " would have caused loop\n",
2796 dentry->d_name.name,
2797 inode->i_sb->s_type->name,
2798 inode->i_sb->s_id);
2799 } else if (!IS_ROOT(new)) {
2800 int err = __d_unalias(inode, dentry, new);
2801 write_sequnlock(&rename_lock);
2802 if (err) {
2803 dput(new);
2804 new = ERR_PTR(err);
2805 }
Al Viro18367502011-07-12 21:42:24 -04002806 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04002807 __d_move(new, dentry, false);
2808 write_sequnlock(&rename_lock);
2809 spin_unlock(&inode->i_lock);
2810 security_d_instantiate(new, inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002811 }
Al Virob5ae6b12014-10-12 22:16:02 -04002812 iput(inode);
2813 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002814 }
David Howells770bfad2006-08-22 20:06:07 -04002815 }
Al Virob5ae6b12014-10-12 22:16:02 -04002816 /* already taking inode->i_lock, so d_add() by hand */
2817 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11002818 spin_unlock(&inode->i_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002819out:
2820 security_d_instantiate(dentry, inode);
2821 d_rehash(dentry);
2822 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04002823}
Al Virob5ae6b12014-10-12 22:16:02 -04002824EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04002825
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002826static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002827{
2828 *buflen -= namelen;
2829 if (*buflen < 0)
2830 return -ENAMETOOLONG;
2831 *buffer -= namelen;
2832 memcpy(*buffer, str, namelen);
2833 return 0;
2834}
2835
Waiman Long232d2d62013-09-09 12:18:13 -04002836/**
2837 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002838 * @buffer: buffer pointer
2839 * @buflen: allocated length of the buffer
2840 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002841 *
2842 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2843 * make sure that either the old or the new name pointer and length are
2844 * fetched. However, there may be mismatch between length and pointer.
2845 * The length cannot be trusted, we need to copy it byte-by-byte until
2846 * the length is reached or a null byte is found. It also prepends "/" at
2847 * the beginning of the name. The sequence number check at the caller will
2848 * retry it again when a d_move() does happen. So any garbage in the buffer
2849 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04002850 *
2851 * Data dependency barrier is needed to make sure that we see that terminating
2852 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04002853 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002854static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2855{
Waiman Long232d2d62013-09-09 12:18:13 -04002856 const char *dname = ACCESS_ONCE(name->name);
2857 u32 dlen = ACCESS_ONCE(name->len);
2858 char *p;
2859
Al Viro6d13f692014-09-29 14:46:30 -04002860 smp_read_barrier_depends();
2861
Waiman Long232d2d62013-09-09 12:18:13 -04002862 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04002863 if (*buflen < 0)
2864 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04002865 p = *buffer -= dlen + 1;
2866 *p++ = '/';
2867 while (dlen--) {
2868 char c = *dname++;
2869 if (!c)
2870 break;
2871 *p++ = c;
2872 }
2873 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002874}
2875
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002877 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002878 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002879 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002880 * @buffer: pointer to the end of the buffer
2881 * @buflen: pointer to buffer length
2882 *
Waiman Long18129972013-09-12 10:55:35 -04002883 * The function will first try to write out the pathname without taking any
2884 * lock other than the RCU read lock to make sure that dentries won't go away.
2885 * It only checks the sequence number of the global rename_lock as any change
2886 * in the dentry's d_seq will be preceded by changes in the rename_lock
2887 * sequence number. If the sequence number had been changed, it will restart
2888 * the whole pathname back-tracing sequence again by taking the rename_lock.
2889 * In this case, there is no need to take the RCU read lock as the recursive
2890 * parent pointer references will keep the dentry chain alive as long as no
2891 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002892 */
Al Viro02125a82011-12-05 08:43:34 -05002893static int prepend_path(const struct path *path,
2894 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002895 char **buffer, int *buflen)
2896{
Al Viroede4ceb2013-11-13 07:45:40 -05002897 struct dentry *dentry;
2898 struct vfsmount *vfsmnt;
2899 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002900 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002901 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002902 char *bptr;
2903 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002904
Al Viro48f5ec22013-09-09 15:22:25 -04002905 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002906restart_mnt:
2907 read_seqbegin_or_lock(&mount_lock, &m_seq);
2908 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002909 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002910restart:
2911 bptr = *buffer;
2912 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002913 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002914 dentry = path->dentry;
2915 vfsmnt = path->mnt;
2916 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002917 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002918 while (dentry != root->dentry || vfsmnt != root->mnt) {
2919 struct dentry * parent;
2920
2921 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002922 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002923 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002924 if (mnt != parent) {
2925 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2926 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002927 vfsmnt = &mnt->mnt;
2928 continue;
2929 }
2930 /*
2931 * Filesystems needing to implement special "root names"
2932 * should do so with ->d_dname()
2933 */
2934 if (IS_ROOT(dentry) &&
2935 (dentry->d_name.len != 1 ||
2936 dentry->d_name.name[0] != '/')) {
2937 WARN(1, "Root dentry has weird name <%.*s>\n",
2938 (int) dentry->d_name.len,
2939 dentry->d_name.name);
2940 }
2941 if (!error)
2942 error = is_mounted(vfsmnt) ? 1 : 2;
2943 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002944 }
2945 parent = dentry->d_parent;
2946 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002947 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002948 if (error)
2949 break;
2950
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002951 dentry = parent;
2952 }
Al Viro48f5ec22013-09-09 15:22:25 -04002953 if (!(seq & 1))
2954 rcu_read_unlock();
2955 if (need_seqretry(&rename_lock, seq)) {
2956 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002957 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002958 }
Al Viro48f5ec22013-09-09 15:22:25 -04002959 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002960
2961 if (!(m_seq & 1))
2962 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002963 if (need_seqretry(&mount_lock, m_seq)) {
2964 m_seq = 1;
2965 goto restart_mnt;
2966 }
2967 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002968
Waiman Long232d2d62013-09-09 12:18:13 -04002969 if (error >= 0 && bptr == *buffer) {
2970 if (--blen < 0)
2971 error = -ENAMETOOLONG;
2972 else
2973 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002974 }
Waiman Long232d2d62013-09-09 12:18:13 -04002975 *buffer = bptr;
2976 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002977 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002978}
2979
2980/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002981 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002982 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002983 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002984 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 * @buflen: buffer length
2986 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002987 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002989 * Returns a pointer into the buffer or an error code if the
2990 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002991 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002992 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002993 *
Al Viro02125a82011-12-05 08:43:34 -05002994 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 */
Al Viro02125a82011-12-05 08:43:34 -05002996char *__d_path(const struct path *path,
2997 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002998 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003000 char *res = buf + buflen;
3001 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003003 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003004 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04003005
Al Viro02125a82011-12-05 08:43:34 -05003006 if (error < 0)
3007 return ERR_PTR(error);
3008 if (error > 0)
3009 return NULL;
3010 return res;
3011}
3012
3013char *d_absolute_path(const struct path *path,
3014 char *buf, int buflen)
3015{
3016 struct path root = {};
3017 char *res = buf + buflen;
3018 int error;
3019
3020 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003021 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05003022
3023 if (error > 1)
3024 error = -EINVAL;
3025 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003026 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003027 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028}
3029
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003030/*
3031 * same as __d_path but appends "(deleted)" for unlinked files.
3032 */
Al Viro02125a82011-12-05 08:43:34 -05003033static int path_with_deleted(const struct path *path,
3034 const struct path *root,
3035 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003036{
3037 prepend(buf, buflen, "\0", 1);
3038 if (d_unlinked(path->dentry)) {
3039 int error = prepend(buf, buflen, " (deleted)", 10);
3040 if (error)
3041 return error;
3042 }
3043
3044 return prepend_path(path, root, buf, buflen);
3045}
3046
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003047static int prepend_unreachable(char **buffer, int *buflen)
3048{
3049 return prepend(buffer, buflen, "(unreachable)", 13);
3050}
3051
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003052static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3053{
3054 unsigned seq;
3055
3056 do {
3057 seq = read_seqcount_begin(&fs->seq);
3058 *root = fs->root;
3059 } while (read_seqcount_retry(&fs->seq, seq));
3060}
3061
Jan Bluncka03a8a702008-02-14 19:38:32 -08003062/**
3063 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003064 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003065 * @buf: buffer to return value in
3066 * @buflen: buffer length
3067 *
3068 * Convert a dentry into an ASCII path name. If the entry has been deleted
3069 * the string " (deleted)" is appended. Note that this is ambiguous.
3070 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003071 * Returns a pointer into the buffer or an error code if the path was
3072 * too long. Note: Callers should use the returned pointer, not the passed
3073 * in buffer, to use the name! The implementation often starts at an offset
3074 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003075 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003076 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003077 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003078char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003080 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003081 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003082 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003084 /*
3085 * We have various synthetic filesystems that never get mounted. On
3086 * these filesystems dentries are never used for lookup purposes, and
3087 * thus don't need to be hashed. They also don't need a name until a
3088 * user wants to identify the object in /proc/pid/fd/. The little hack
3089 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003090 *
3091 * Some pseudo inodes are mountable. When they are mounted
3092 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3093 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003094 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003095 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3096 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003097 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003098
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003099 rcu_read_lock();
3100 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003101 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003102 rcu_read_unlock();
3103
Al Viro02125a82011-12-05 08:43:34 -05003104 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003105 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106 return res;
3107}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003108EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109
3110/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003111 * Helper function for dentry_operations.d_dname() members
3112 */
3113char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3114 const char *fmt, ...)
3115{
3116 va_list args;
3117 char temp[64];
3118 int sz;
3119
3120 va_start(args, fmt);
3121 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3122 va_end(args);
3123
3124 if (sz > sizeof(temp) || sz > buflen)
3125 return ERR_PTR(-ENAMETOOLONG);
3126
3127 buffer += buflen - sz;
3128 return memcpy(buffer, temp, sz);
3129}
3130
Al Viro118b2302013-08-24 12:08:17 -04003131char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3132{
3133 char *end = buffer + buflen;
3134 /* these dentries are never renamed, so d_lock is not needed */
3135 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003136 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003137 prepend(&end, &buflen, "/", 1))
3138 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003139 return end;
Al Viro118b2302013-08-24 12:08:17 -04003140}
David Herrmann31bbe162014-01-03 14:09:47 +01003141EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003142
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003143/*
Ram Pai6092d042008-03-27 13:06:20 +01003144 * Write full pathname from the root of the filesystem into the buffer.
3145 */
Al Virof6500802014-01-26 12:37:55 -05003146static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003147{
Al Virof6500802014-01-26 12:37:55 -05003148 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003149 char *end, *retval;
3150 int len, seq = 0;
3151 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003152
Al Virof6500802014-01-26 12:37:55 -05003153 if (buflen < 2)
3154 goto Elong;
3155
Al Viro48f5ec22013-09-09 15:22:25 -04003156 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003157restart:
Al Virof6500802014-01-26 12:37:55 -05003158 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003159 end = buf + buflen;
3160 len = buflen;
3161 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003162 /* Get '/' right */
3163 retval = end-1;
3164 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003165 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003166 while (!IS_ROOT(dentry)) {
3167 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003168
Ram Pai6092d042008-03-27 13:06:20 +01003169 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003170 error = prepend_name(&end, &len, &dentry->d_name);
3171 if (error)
3172 break;
Ram Pai6092d042008-03-27 13:06:20 +01003173
3174 retval = end;
3175 dentry = parent;
3176 }
Al Viro48f5ec22013-09-09 15:22:25 -04003177 if (!(seq & 1))
3178 rcu_read_unlock();
3179 if (need_seqretry(&rename_lock, seq)) {
3180 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003181 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003182 }
3183 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003184 if (error)
3185 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003186 return retval;
3187Elong:
3188 return ERR_PTR(-ENAMETOOLONG);
3189}
Nick Pigginec2447c2011-01-07 17:49:29 +11003190
3191char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3192{
Waiman Long232d2d62013-09-09 12:18:13 -04003193 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003194}
3195EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003196
3197char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3198{
3199 char *p = NULL;
3200 char *retval;
3201
Al Viroc1031352010-06-06 22:31:14 -04003202 if (d_unlinked(dentry)) {
3203 p = buf + buflen;
3204 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3205 goto Elong;
3206 buflen++;
3207 }
3208 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003209 if (!IS_ERR(retval) && p)
3210 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003211 return retval;
3212Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003213 return ERR_PTR(-ENAMETOOLONG);
3214}
3215
Linus Torvalds8b19e342013-09-12 10:35:47 -07003216static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3217 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003218{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003219 unsigned seq;
3220
3221 do {
3222 seq = read_seqcount_begin(&fs->seq);
3223 *root = fs->root;
3224 *pwd = fs->pwd;
3225 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003226}
3227
Ram Pai6092d042008-03-27 13:06:20 +01003228/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229 * NOTE! The user-level library version returns a
3230 * character pointer. The kernel system call just
3231 * returns the length of the buffer filled (which
3232 * includes the ending '\0' character), or a negative
3233 * error value. So libc would do something like
3234 *
3235 * char *getcwd(char * buf, size_t size)
3236 * {
3237 * int retval;
3238 *
3239 * retval = sys_getcwd(buf, size);
3240 * if (retval >= 0)
3241 * return buf;
3242 * errno = -retval;
3243 * return NULL;
3244 * }
3245 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003246SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247{
Linus Torvalds552ce542007-02-13 12:08:18 -08003248 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003249 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003250 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251
3252 if (!page)
3253 return -ENOMEM;
3254
Linus Torvalds8b19e342013-09-12 10:35:47 -07003255 rcu_read_lock();
3256 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257
Linus Torvalds552ce542007-02-13 12:08:18 -08003258 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003259 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003260 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003261 char *cwd = page + PATH_MAX;
3262 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003264 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003265 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003266 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003267
Al Viro02125a82011-12-05 08:43:34 -05003268 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003269 goto out;
3270
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003271 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003272 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003273 error = prepend_unreachable(&cwd, &buflen);
3274 if (error)
3275 goto out;
3276 }
3277
Linus Torvalds552ce542007-02-13 12:08:18 -08003278 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003279 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003280 if (len <= size) {
3281 error = len;
3282 if (copy_to_user(buf, cwd, len))
3283 error = -EFAULT;
3284 }
Nick Piggin949854d2011-01-07 17:49:37 +11003285 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003286 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003287 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288
3289out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003290 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291 return error;
3292}
3293
3294/*
3295 * Test whether new_dentry is a subdirectory of old_dentry.
3296 *
3297 * Trivially implemented using the dcache structure
3298 */
3299
3300/**
3301 * is_subdir - is new dentry a subdirectory of old_dentry
3302 * @new_dentry: new dentry
3303 * @old_dentry: old dentry
3304 *
3305 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3306 * Returns 0 otherwise.
3307 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3308 */
3309
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003310int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311{
3312 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11003313 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003315 if (new_dentry == old_dentry)
3316 return 1;
3317
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003318 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003321 /*
3322 * Need rcu_readlock to protect against the d_parent trashing
3323 * due to d_move
3324 */
3325 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003326 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003328 else
3329 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11003330 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003331 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332
3333 return result;
3334}
3335
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003336static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003338 struct dentry *root = data;
3339 if (dentry != root) {
3340 if (d_unhashed(dentry) || !dentry->d_inode)
3341 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003343 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3344 dentry->d_flags |= DCACHE_GENOCIDE;
3345 dentry->d_lockref.count--;
3346 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003348 return D_WALK_CONTINUE;
3349}
Nick Piggin58db63d2011-01-07 17:49:39 +11003350
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003351void d_genocide(struct dentry *parent)
3352{
3353 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354}
3355
Al Viro60545d02013-06-07 01:20:27 -04003356void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357{
Al Viro60545d02013-06-07 01:20:27 -04003358 inode_dec_link_count(inode);
3359 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003360 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003361 !d_unlinked(dentry));
3362 spin_lock(&dentry->d_parent->d_lock);
3363 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3364 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3365 (unsigned long long)inode->i_ino);
3366 spin_unlock(&dentry->d_lock);
3367 spin_unlock(&dentry->d_parent->d_lock);
3368 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369}
Al Viro60545d02013-06-07 01:20:27 -04003370EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371
3372static __initdata unsigned long dhash_entries;
3373static int __init set_dhash_entries(char *str)
3374{
3375 if (!str)
3376 return 0;
3377 dhash_entries = simple_strtoul(str, &str, 0);
3378 return 1;
3379}
3380__setup("dhash_entries=", set_dhash_entries);
3381
3382static void __init dcache_init_early(void)
3383{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003384 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385
3386 /* If hashes are distributed across NUMA nodes, defer
3387 * hash allocation until vmalloc space is available.
3388 */
3389 if (hashdist)
3390 return;
3391
3392 dentry_hashtable =
3393 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003394 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395 dhash_entries,
3396 13,
3397 HASH_EARLY,
3398 &d_hash_shift,
3399 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003400 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003401 0);
3402
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003403 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003404 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405}
3406
Denis Cheng74bf17c2007-10-16 23:26:30 -07003407static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003409 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410
3411 /*
3412 * A constructor could be added for stable state like the lists,
3413 * but it is probably not worth it because of the cache nature
3414 * of the dcache.
3415 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003416 dentry_cache = KMEM_CACHE(dentry,
3417 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418
3419 /* Hash may have been set up in dcache_init_early */
3420 if (!hashdist)
3421 return;
3422
3423 dentry_hashtable =
3424 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003425 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 dhash_entries,
3427 13,
3428 0,
3429 &d_hash_shift,
3430 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003431 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432 0);
3433
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003434 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003435 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436}
3437
3438/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003439struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003440EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442EXPORT_SYMBOL(d_genocide);
3443
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444void __init vfs_caches_init_early(void)
3445{
3446 dcache_init_early();
3447 inode_init_early();
3448}
3449
3450void __init vfs_caches_init(unsigned long mempages)
3451{
3452 unsigned long reserve;
3453
3454 /* Base hash sizes on available memory, with a reserve equal to
3455 150% of current kernel size */
3456
3457 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3458 mempages -= reserve;
3459
3460 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003461 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003462
Denis Cheng74bf17c2007-10-16 23:26:30 -07003463 dcache_init();
3464 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003465 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003466 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003467 bdev_cache_init();
3468 chrdev_init();
3469}