blob: dc400fd29f4d1c3c8e2265b4275aaabe4250e1fb [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
Al Virob4f03542014-04-29 23:40:14 -0400272static void dentry_free(struct dentry *dentry)
273{
Al Viro946e51f2014-10-26 19:19:16 -0400274 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400275 if (unlikely(dname_external(dentry))) {
276 struct external_name *p = external_name(dentry);
277 if (likely(atomic_dec_and_test(&p->u.count))) {
278 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
279 return;
280 }
281 }
Al Virob4f03542014-04-29 23:40:14 -0400282 /* if dentry was never visible to RCU, immediate free is OK */
283 if (!(dentry->d_flags & DCACHE_RCUACCESS))
284 __d_free(&dentry->d_u.d_rcu);
285 else
286 call_rcu(&dentry->d_u.d_rcu, __d_free);
287}
288
Nick Piggin31e6b012011-01-07 17:49:52 +1100289/**
290 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800291 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100292 * After this call, in-progress rcu-walk path lookup will fail. This
293 * should be called after unhashing, and after changing d_inode (if
294 * the dentry has not already been unhashed).
295 */
296static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
297{
298 assert_spin_locked(&dentry->d_lock);
299 /* Go through a barrier */
300 write_seqcount_barrier(&dentry->d_seq);
301}
302
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303/*
304 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100305 * d_iput() operation if defined. Dentry has no refcount
306 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800308static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200309 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100310 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311{
312 struct inode *inode = dentry->d_inode;
313 if (inode) {
314 dentry->d_inode = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400315 hlist_del_init(&dentry->d_u.d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100317 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700318 if (!inode->i_nlink)
319 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 if (dentry->d_op && dentry->d_op->d_iput)
321 dentry->d_op->d_iput(dentry, inode);
322 else
323 iput(inode);
324 } else {
325 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 }
327}
328
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700329/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100330 * Release the dentry's inode, using the filesystem
331 * d_iput() operation if defined. dentry remains in-use.
332 */
333static void dentry_unlink_inode(struct dentry * dentry)
334 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100335 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100336{
337 struct inode *inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +0100338 __d_clear_type(dentry);
Nick Piggin31e6b012011-01-07 17:49:52 +1100339 dentry->d_inode = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400340 hlist_del_init(&dentry->d_u.d_alias);
Nick Piggin31e6b012011-01-07 17:49:52 +1100341 dentry_rcuwalk_barrier(dentry);
342 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100343 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100344 if (!inode->i_nlink)
345 fsnotify_inoderemove(inode);
346 if (dentry->d_op && dentry->d_op->d_iput)
347 dentry->d_op->d_iput(dentry, inode);
348 else
349 iput(inode);
350}
351
352/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400353 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
354 * is in use - which includes both the "real" per-superblock
355 * LRU list _and_ the DCACHE_SHRINK_LIST use.
356 *
357 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
358 * on the shrink list (ie not on the superblock LRU list).
359 *
360 * The per-cpu "nr_dentry_unused" counters are updated with
361 * the DCACHE_LRU_LIST bit.
362 *
363 * These helper functions make sure we always follow the
364 * rules. d_lock must be held by the caller.
365 */
366#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
367static void d_lru_add(struct dentry *dentry)
368{
369 D_FLAG_VERIFY(dentry, 0);
370 dentry->d_flags |= DCACHE_LRU_LIST;
371 this_cpu_inc(nr_dentry_unused);
372 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
373}
374
375static void d_lru_del(struct dentry *dentry)
376{
377 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
378 dentry->d_flags &= ~DCACHE_LRU_LIST;
379 this_cpu_dec(nr_dentry_unused);
380 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
381}
382
383static void d_shrink_del(struct dentry *dentry)
384{
385 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
386 list_del_init(&dentry->d_lru);
387 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
388 this_cpu_dec(nr_dentry_unused);
389}
390
391static void d_shrink_add(struct dentry *dentry, struct list_head *list)
392{
393 D_FLAG_VERIFY(dentry, 0);
394 list_add(&dentry->d_lru, list);
395 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
396 this_cpu_inc(nr_dentry_unused);
397}
398
399/*
400 * These can only be called under the global LRU lock, ie during the
401 * callback for freeing the LRU list. "isolate" removes it from the
402 * LRU lists entirely, while shrink_move moves it to the indicated
403 * private list.
404 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800405static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400406{
407 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
408 dentry->d_flags &= ~DCACHE_LRU_LIST;
409 this_cpu_dec(nr_dentry_unused);
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800410 list_lru_isolate(lru, &dentry->d_lru);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400411}
412
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800413static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
414 struct list_head *list)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400415{
416 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
417 dentry->d_flags |= DCACHE_SHRINK_LIST;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800418 list_lru_isolate_move(lru, &dentry->d_lru, list);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400419}
420
421/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000422 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700423 */
424static void dentry_lru_add(struct dentry *dentry)
425{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400426 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
427 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700428}
429
Miklos Szeredid52b9082007-05-08 00:23:46 -0700430/**
Nick Piggin789680d2011-01-07 17:49:30 +1100431 * d_drop - drop a dentry
432 * @dentry: dentry to drop
433 *
434 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
435 * be found through a VFS lookup any more. Note that this is different from
436 * deleting the dentry - d_delete will try to mark the dentry negative if
437 * possible, giving a successful _negative_ lookup, while d_drop will
438 * just make the cache lookup fail.
439 *
440 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
441 * reason (NFS timeouts or autofs deletes).
442 *
443 * __d_drop requires dentry->d_lock.
444 */
445void __d_drop(struct dentry *dentry)
446{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700447 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400448 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400449 /*
450 * Hashed dentries are normally on the dentry hashtable,
451 * with the exception of those newly allocated by
452 * d_obtain_alias, which are always IS_ROOT:
453 */
454 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400455 b = &dentry->d_sb->s_anon;
456 else
457 b = d_hash(dentry->d_parent, dentry->d_name.hash);
458
459 hlist_bl_lock(b);
460 __hlist_bl_del(&dentry->d_hash);
461 dentry->d_hash.pprev = NULL;
462 hlist_bl_unlock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700463 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100464 }
465}
466EXPORT_SYMBOL(__d_drop);
467
468void d_drop(struct dentry *dentry)
469{
Nick Piggin789680d2011-01-07 17:49:30 +1100470 spin_lock(&dentry->d_lock);
471 __d_drop(dentry);
472 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100473}
474EXPORT_SYMBOL(d_drop);
475
Al Viroe55fd012014-05-28 13:51:12 -0400476static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100477{
Al Viro41edf272014-05-01 10:30:00 -0400478 struct dentry *parent = NULL;
479 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400480 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100481 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100482
Linus Torvalds0d984392013-09-08 13:46:52 -0700483 /*
484 * The dentry is now unrecoverably dead to the world.
485 */
486 lockref_mark_dead(&dentry->d_lockref);
487
Sage Weilf0023bc2011-10-28 10:02:42 -0700488 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700489 * inform the fs via d_prune that this dentry is about to be
490 * unhashed and destroyed.
491 */
Al Viro29266202014-05-30 11:39:02 -0400492 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800493 dentry->d_op->d_prune(dentry);
494
Al Viro01b60352014-04-29 23:42:52 -0400495 if (dentry->d_flags & DCACHE_LRU_LIST) {
496 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
497 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400498 }
Nick Piggin77812a12011-01-07 17:49:48 +1100499 /* if it was on the hash then remove it */
500 __d_drop(dentry);
Al Viroca5358e2014-10-26 19:31:10 -0400501 __list_del_entry(&dentry->d_child);
Al Viro03b3b882014-04-29 15:45:28 -0400502 /*
503 * Inform d_walk() that we are no longer attached to the
504 * dentry tree
505 */
506 dentry->d_flags |= DCACHE_DENTRY_KILLED;
507 if (parent)
508 spin_unlock(&parent->d_lock);
509 dentry_iput(dentry);
510 /*
511 * dentry_iput drops the locks, at which point nobody (except
512 * transient RCU lookups) can reach this dentry.
513 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800514 BUG_ON(dentry->d_lockref.count > 0);
Al Viro03b3b882014-04-29 15:45:28 -0400515 this_cpu_dec(nr_dentry);
516 if (dentry->d_op && dentry->d_op->d_release)
517 dentry->d_op->d_release(dentry);
518
Al Viro41edf272014-05-01 10:30:00 -0400519 spin_lock(&dentry->d_lock);
520 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
521 dentry->d_flags |= DCACHE_MAY_FREE;
522 can_free = false;
523 }
524 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400525 if (likely(can_free))
526 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400527}
528
529/*
530 * Finish off a dentry we've decided to kill.
531 * dentry->d_lock must be held, returns with it unlocked.
532 * If ref is non-zero, then decrement the refcount too.
533 * Returns dentry requiring refcount drop, or NULL if we're done.
534 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400535static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400536 __releases(dentry->d_lock)
537{
538 struct inode *inode = dentry->d_inode;
539 struct dentry *parent = NULL;
540
541 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
542 goto failed;
543
544 if (!IS_ROOT(dentry)) {
545 parent = dentry->d_parent;
546 if (unlikely(!spin_trylock(&parent->d_lock))) {
547 if (inode)
548 spin_unlock(&inode->i_lock);
549 goto failed;
550 }
551 }
552
553 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400554 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400555
556failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400557 spin_unlock(&dentry->d_lock);
558 cpu_relax();
Al Viroe55fd012014-05-28 13:51:12 -0400559 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100560}
561
Al Viro046b9612014-05-29 08:54:52 -0400562static inline struct dentry *lock_parent(struct dentry *dentry)
563{
564 struct dentry *parent = dentry->d_parent;
565 if (IS_ROOT(dentry))
566 return NULL;
Linus Torvalds360f5472015-01-09 15:19:03 -0800567 if (unlikely(dentry->d_lockref.count < 0))
Al Viroc2338f22014-06-12 00:29:13 -0400568 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400569 if (likely(spin_trylock(&parent->d_lock)))
570 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400571 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400572 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400573again:
574 parent = ACCESS_ONCE(dentry->d_parent);
575 spin_lock(&parent->d_lock);
576 /*
577 * We can't blindly lock dentry until we are sure
578 * that we won't violate the locking order.
579 * Any changes of dentry->d_parent must have
580 * been done with parent->d_lock held, so
581 * spin_lock() above is enough of a barrier
582 * for checking if it's still our child.
583 */
584 if (unlikely(parent != dentry->d_parent)) {
585 spin_unlock(&parent->d_lock);
586 goto again;
587 }
588 rcu_read_unlock();
589 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700590 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400591 else
592 parent = NULL;
593 return parent;
594}
595
Linus Torvalds360f5472015-01-09 15:19:03 -0800596/*
597 * Try to do a lockless dput(), and return whether that was successful.
598 *
599 * If unsuccessful, we return false, having already taken the dentry lock.
600 *
601 * The caller needs to hold the RCU read lock, so that the dentry is
602 * guaranteed to stay around even if the refcount goes down to zero!
603 */
604static inline bool fast_dput(struct dentry *dentry)
605{
606 int ret;
607 unsigned int d_flags;
608
609 /*
610 * If we have a d_op->d_delete() operation, we sould not
611 * let the dentry count go to zero, so use "put__or_lock".
612 */
613 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
614 return lockref_put_or_lock(&dentry->d_lockref);
615
616 /*
617 * .. otherwise, we can try to just decrement the
618 * lockref optimistically.
619 */
620 ret = lockref_put_return(&dentry->d_lockref);
621
622 /*
623 * If the lockref_put_return() failed due to the lock being held
624 * by somebody else, the fast path has failed. We will need to
625 * get the lock, and then check the count again.
626 */
627 if (unlikely(ret < 0)) {
628 spin_lock(&dentry->d_lock);
629 if (dentry->d_lockref.count > 1) {
630 dentry->d_lockref.count--;
631 spin_unlock(&dentry->d_lock);
632 return 1;
633 }
634 return 0;
635 }
636
637 /*
638 * If we weren't the last ref, we're done.
639 */
640 if (ret)
641 return 1;
642
643 /*
644 * Careful, careful. The reference count went down
645 * to zero, but we don't hold the dentry lock, so
646 * somebody else could get it again, and do another
647 * dput(), and we need to not race with that.
648 *
649 * However, there is a very special and common case
650 * where we don't care, because there is nothing to
651 * do: the dentry is still hashed, it does not have
652 * a 'delete' op, and it's referenced and already on
653 * the LRU list.
654 *
655 * NOTE! Since we aren't locked, these values are
656 * not "stable". However, it is sufficient that at
657 * some point after we dropped the reference the
658 * dentry was hashed and the flags had the proper
659 * value. Other dentry users may have re-gotten
660 * a reference to the dentry and change that, but
661 * our work is done - we can leave the dentry
662 * around with a zero refcount.
663 */
664 smp_rmb();
665 d_flags = ACCESS_ONCE(dentry->d_flags);
666 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST;
667
668 /* Nothing to do? Dropping the reference was all we needed? */
669 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
670 return 1;
671
672 /*
673 * Not the fast normal case? Get the lock. We've already decremented
674 * the refcount, but we'll need to re-check the situation after
675 * getting the lock.
676 */
677 spin_lock(&dentry->d_lock);
678
679 /*
680 * Did somebody else grab a reference to it in the meantime, and
681 * we're no longer the last user after all? Alternatively, somebody
682 * else could have killed it and marked it dead. Either way, we
683 * don't need to do anything else.
684 */
685 if (dentry->d_lockref.count) {
686 spin_unlock(&dentry->d_lock);
687 return 1;
688 }
689
690 /*
691 * Re-get the reference we optimistically dropped. We hold the
692 * lock, and we just tested that it was zero, so we can just
693 * set it to 1.
694 */
695 dentry->d_lockref.count = 1;
696 return 0;
697}
698
699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700/*
701 * This is dput
702 *
703 * This is complicated by the fact that we do not want to put
704 * dentries that are no longer on any hash chain on the unused
705 * list: we'd much rather just get rid of them immediately.
706 *
707 * However, that implies that we have to traverse the dentry
708 * tree upwards to the parents which might _also_ now be
709 * scheduled for deletion (it may have been only waiting for
710 * its last child to go away).
711 *
712 * This tail recursion is done by hand as we don't want to depend
713 * on the compiler to always get this right (gcc generally doesn't).
714 * Real recursion would eat up our stack space.
715 */
716
717/*
718 * dput - release a dentry
719 * @dentry: dentry to release
720 *
721 * Release a dentry. This will drop the usage count and if appropriate
722 * call the dentry unlink method as well as removing it from the queues and
723 * releasing its resources. If the parent dentries were scheduled for release
724 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726void dput(struct dentry *dentry)
727{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700728 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 return;
730
731repeat:
Linus Torvalds360f5472015-01-09 15:19:03 -0800732 rcu_read_lock();
733 if (likely(fast_dput(dentry))) {
734 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 return;
Linus Torvalds360f5472015-01-09 15:19:03 -0800736 }
737
738 /* Slow case: now with the dentry lock held */
739 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700741 /* Unreachable? Get rid of it */
742 if (unlikely(d_unhashed(dentry)))
743 goto kill_it;
744
745 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100747 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 }
Nick Piggin265ac902010-10-10 05:36:24 -0400749
Linus Torvalds358eec12013-10-31 15:43:02 -0700750 if (!(dentry->d_flags & DCACHE_REFERENCED))
751 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400752 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400753
Waiman Long98474232013-08-28 18:24:59 -0700754 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100755 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 return;
757
Miklos Szeredid52b9082007-05-08 00:23:46 -0700758kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400759 dentry = dentry_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700760 if (dentry)
761 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700763EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100766/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100767static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768{
Waiman Long98474232013-08-28 18:24:59 -0700769 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770}
771
Nick Piggindc0474b2011-01-07 17:49:43 +1100772static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100773{
Waiman Long98474232013-08-28 18:24:59 -0700774 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100775}
776
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100777struct dentry *dget_parent(struct dentry *dentry)
778{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700779 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100780 struct dentry *ret;
781
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700782 /*
783 * Do optimistic parent lookup without any
784 * locking.
785 */
786 rcu_read_lock();
787 ret = ACCESS_ONCE(dentry->d_parent);
788 gotref = lockref_get_not_zero(&ret->d_lockref);
789 rcu_read_unlock();
790 if (likely(gotref)) {
791 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
792 return ret;
793 dput(ret);
794 }
795
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100796repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100797 /*
798 * Don't need rcu_dereference because we re-check it was correct under
799 * the lock.
800 */
801 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100802 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100803 spin_lock(&ret->d_lock);
804 if (unlikely(ret != dentry->d_parent)) {
805 spin_unlock(&ret->d_lock);
806 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100807 goto repeat;
808 }
Nick Piggina734eb42011-01-07 17:49:44 +1100809 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700810 BUG_ON(!ret->d_lockref.count);
811 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100812 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100813 return ret;
814}
815EXPORT_SYMBOL(dget_parent);
816
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817/**
818 * d_find_alias - grab a hashed alias of inode
819 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 *
821 * If inode has a hashed alias, or is a directory and has any alias,
822 * acquire the reference to alias and return it. Otherwise return NULL.
823 * Notice that if inode is a directory there can be only one alias and
824 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800825 * of a filesystem, or if the directory was renamed and d_revalidate
826 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700828 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500829 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500831static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
Nick Pigginda502952011-01-07 17:49:33 +1100833 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
Nick Pigginda502952011-01-07 17:49:33 +1100835again:
836 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400837 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100838 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700840 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100841 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500843 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100844 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100845 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 return alias;
847 }
848 }
Nick Pigginda502952011-01-07 17:49:33 +1100849 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 }
Nick Pigginda502952011-01-07 17:49:33 +1100851 if (discon_alias) {
852 alias = discon_alias;
853 spin_lock(&alias->d_lock);
854 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500855 __dget_dlock(alias);
856 spin_unlock(&alias->d_lock);
857 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100858 }
859 spin_unlock(&alias->d_lock);
860 goto again;
861 }
862 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863}
864
Nick Pigginda502952011-01-07 17:49:33 +1100865struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866{
David Howells214fda12006-03-25 03:06:36 -0800867 struct dentry *de = NULL;
868
Al Virob3d9b7a2012-06-09 13:51:19 -0400869 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100870 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500871 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100872 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800873 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 return de;
875}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700876EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
878/*
879 * Try to kill dentries associated with this inode.
880 * WARNING: you must own a reference to inode.
881 */
882void d_prune_aliases(struct inode *inode)
883{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700884 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100886 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400887 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700889 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400890 struct dentry *parent = lock_parent(dentry);
891 if (likely(!dentry->d_lockref.count)) {
892 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800893 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400894 goto restart;
895 }
896 if (parent)
897 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 }
899 spin_unlock(&dentry->d_lock);
900 }
Nick Piggin873feea2011-01-07 17:50:06 +1100901 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700903EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400905static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906{
Al Viro5c47e6d2014-04-29 16:13:18 -0400907 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700908
Miklos Szeredi60942f22014-05-02 15:38:39 -0400909 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400910 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400911 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100912 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400913 parent = lock_parent(dentry);
914
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000916 * The dispose list is isolated and dentries are not accounted
917 * to the LRU here, so we can simply remove it from the list
918 * here regardless of whether it is referenced or not.
919 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400920 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000921
922 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000924 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800926 if (dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700927 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400928 if (parent)
929 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 continue;
931 }
Nick Piggin77812a12011-01-07 17:49:48 +1100932
Al Viro64fd72e2014-05-28 09:48:44 -0400933
934 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
935 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
936 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400937 if (parent)
938 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -0400939 if (can_free)
940 dentry_free(dentry);
941 continue;
942 }
943
Al Viroff2fde92014-05-28 13:59:13 -0400944 inode = dentry->d_inode;
945 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400946 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000947 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400948 if (parent)
949 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400950 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000951 }
Al Viroff2fde92014-05-28 13:59:13 -0400952
Al Viroff2fde92014-05-28 13:59:13 -0400953 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -0400954
Al Viro5c47e6d2014-04-29 16:13:18 -0400955 /*
956 * We need to prune ancestors too. This is necessary to prevent
957 * quadratic behavior of shrink_dcache_parent(), but is also
958 * expected to be beneficial in reducing dentry cache
959 * fragmentation.
960 */
961 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -0400962 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
963 parent = lock_parent(dentry);
964 if (dentry->d_lockref.count != 1) {
965 dentry->d_lockref.count--;
966 spin_unlock(&dentry->d_lock);
967 if (parent)
968 spin_unlock(&parent->d_lock);
969 break;
970 }
971 inode = dentry->d_inode; /* can't be NULL */
972 if (unlikely(!spin_trylock(&inode->i_lock))) {
973 spin_unlock(&dentry->d_lock);
974 if (parent)
975 spin_unlock(&parent->d_lock);
976 cpu_relax();
977 continue;
978 }
979 __dentry_kill(dentry);
980 dentry = parent;
981 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400983}
984
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800985static enum lru_status dentry_lru_isolate(struct list_head *item,
986 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +1000987{
988 struct list_head *freeable = arg;
989 struct dentry *dentry = container_of(item, struct dentry, d_lru);
990
991
992 /*
993 * we are inverting the lru lock/dentry->d_lock here,
994 * so use a trylock. If we fail to get the lock, just skip
995 * it
996 */
997 if (!spin_trylock(&dentry->d_lock))
998 return LRU_SKIP;
999
1000 /*
1001 * Referenced dentries are still in use. If they have active
1002 * counts, just remove them from the LRU. Otherwise give them
1003 * another pass through the LRU.
1004 */
1005 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001006 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001007 spin_unlock(&dentry->d_lock);
1008 return LRU_REMOVED;
1009 }
1010
1011 if (dentry->d_flags & DCACHE_REFERENCED) {
1012 dentry->d_flags &= ~DCACHE_REFERENCED;
1013 spin_unlock(&dentry->d_lock);
1014
1015 /*
1016 * The list move itself will be made by the common LRU code. At
1017 * this point, we've dropped the dentry->d_lock but keep the
1018 * lru lock. This is safe to do, since every list movement is
1019 * protected by the lru lock even if both locks are held.
1020 *
1021 * This is guaranteed by the fact that all LRU management
1022 * functions are intermediated by the LRU API calls like
1023 * list_lru_add and list_lru_del. List movement in this file
1024 * only ever occur through this functions or through callbacks
1025 * like this one, that are called from the LRU API.
1026 *
1027 * The only exceptions to this are functions like
1028 * shrink_dentry_list, and code that first checks for the
1029 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1030 * operating only with stack provided lists after they are
1031 * properly isolated from the main list. It is thus, always a
1032 * local access.
1033 */
1034 return LRU_ROTATE;
1035 }
1036
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001037 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001038 spin_unlock(&dentry->d_lock);
1039
1040 return LRU_REMOVED;
1041}
1042
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001043/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001044 * prune_dcache_sb - shrink the dcache
1045 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001046 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001047 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001048 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1049 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001050 * function.
1051 *
1052 * This function may fail to free any resources if all the dentries are in
1053 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001054 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001055long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001056{
Dave Chinnerf6041562013-08-28 10:18:00 +10001057 LIST_HEAD(dispose);
1058 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001059
Vladimir Davydov503c3582015-02-12 14:58:47 -08001060 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1061 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001062 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001063 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064}
1065
Glauber Costa4e717f52013-08-28 10:18:03 +10001066static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001067 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001068{
Glauber Costa4e717f52013-08-28 10:18:03 +10001069 struct list_head *freeable = arg;
1070 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001071
Glauber Costa4e717f52013-08-28 10:18:03 +10001072 /*
1073 * we are inverting the lru lock/dentry->d_lock here,
1074 * so use a trylock. If we fail to get the lock, just skip
1075 * it
1076 */
1077 if (!spin_trylock(&dentry->d_lock))
1078 return LRU_SKIP;
1079
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001080 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001081 spin_unlock(&dentry->d_lock);
1082
1083 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001084}
1085
Glauber Costa4e717f52013-08-28 10:18:03 +10001086
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001087/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 * shrink_dcache_sb - shrink dcache for a superblock
1089 * @sb: superblock
1090 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001091 * Shrink the dcache for the specified super block. This is used to free
1092 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001094void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095{
Glauber Costa4e717f52013-08-28 10:18:03 +10001096 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001097
Glauber Costa4e717f52013-08-28 10:18:03 +10001098 do {
1099 LIST_HEAD(dispose);
1100
1101 freed = list_lru_walk(&sb->s_dentry_lru,
1102 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1103
1104 this_cpu_sub(nr_dentry_unused, freed);
1105 shrink_dentry_list(&dispose);
1106 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001108EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001111 * enum d_walk_ret - action to talke during tree walk
1112 * @D_WALK_CONTINUE: contrinue walk
1113 * @D_WALK_QUIT: quit walk
1114 * @D_WALK_NORETRY: quit when retry is needed
1115 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001117enum d_walk_ret {
1118 D_WALK_CONTINUE,
1119 D_WALK_QUIT,
1120 D_WALK_NORETRY,
1121 D_WALK_SKIP,
1122};
1123
1124/**
1125 * d_walk - walk the dentry tree
1126 * @parent: start of walk
1127 * @data: data passed to @enter() and @finish()
1128 * @enter: callback when first entering the dentry
1129 * @finish: callback when successfully finished the walk
1130 *
1131 * The @enter() and @finish() callbacks are called with d_lock held.
1132 */
1133static void d_walk(struct dentry *parent, void *data,
1134 enum d_walk_ret (*enter)(void *, struct dentry *),
1135 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136{
Nick Piggin949854d2011-01-07 17:49:37 +11001137 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001139 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001140 enum d_walk_ret ret;
1141 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001142
Nick Piggin58db63d2011-01-07 17:49:39 +11001143again:
Al Viro48f5ec22013-09-09 15:22:25 -04001144 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001145 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001146 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001147
1148 ret = enter(data, this_parent);
1149 switch (ret) {
1150 case D_WALK_CONTINUE:
1151 break;
1152 case D_WALK_QUIT:
1153 case D_WALK_SKIP:
1154 goto out_unlock;
1155 case D_WALK_NORETRY:
1156 retry = false;
1157 break;
1158 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159repeat:
1160 next = this_parent->d_subdirs.next;
1161resume:
1162 while (next != &this_parent->d_subdirs) {
1163 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001164 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001166
1167 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001168
1169 ret = enter(data, dentry);
1170 switch (ret) {
1171 case D_WALK_CONTINUE:
1172 break;
1173 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001174 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001175 goto out_unlock;
1176 case D_WALK_NORETRY:
1177 retry = false;
1178 break;
1179 case D_WALK_SKIP:
1180 spin_unlock(&dentry->d_lock);
1181 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001182 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001183
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001185 spin_unlock(&this_parent->d_lock);
1186 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001188 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 goto repeat;
1190 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001191 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 }
1193 /*
1194 * All done at this level ... ascend and resume the search.
1195 */
Al Viroca5358e2014-10-26 19:31:10 -04001196 rcu_read_lock();
1197ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001199 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001200 this_parent = child->d_parent;
1201
Al Viro31dec132013-10-25 17:04:27 -04001202 spin_unlock(&child->d_lock);
1203 spin_lock(&this_parent->d_lock);
1204
Al Viroca5358e2014-10-26 19:31:10 -04001205 /* might go back up the wrong parent if we have had a rename. */
1206 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001207 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001208 next = child->d_child.next;
1209 while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
1210 if (next == &this_parent->d_subdirs)
1211 goto ascend;
1212 child = list_entry(next, struct dentry, d_child);
1213 next = next->next;
Al Viro31dec132013-10-25 17:04:27 -04001214 }
1215 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 goto resume;
1217 }
Al Viroca5358e2014-10-26 19:31:10 -04001218 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001219 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001220 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001221 if (finish)
1222 finish(data);
1223
1224out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001225 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001226 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001227 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001228
1229rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001230 spin_unlock(&this_parent->d_lock);
1231 rcu_read_unlock();
1232 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001233 if (!retry)
1234 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001235 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001236 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001238
1239/*
1240 * Search for at least 1 mount point in the dentry's subdirs.
1241 * We descend to the next level whenever the d_subdirs
1242 * list is non-empty and continue searching.
1243 */
1244
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001245static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1246{
1247 int *ret = data;
1248 if (d_mountpoint(dentry)) {
1249 *ret = 1;
1250 return D_WALK_QUIT;
1251 }
1252 return D_WALK_CONTINUE;
1253}
1254
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001255/**
1256 * have_submounts - check for mounts over a dentry
1257 * @parent: dentry to check.
1258 *
1259 * Return true if the parent or its subdirectories contain
1260 * a mount point
1261 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001262int have_submounts(struct dentry *parent)
1263{
1264 int ret = 0;
1265
1266 d_walk(parent, &ret, check_mount, NULL);
1267
1268 return ret;
1269}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001270EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
1272/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001273 * Called by mount code to set a mountpoint and check if the mountpoint is
1274 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1275 * subtree can become unreachable).
1276 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001277 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001278 * this reason take rename_lock and d_lock on dentry and ancestors.
1279 */
1280int d_set_mounted(struct dentry *dentry)
1281{
1282 struct dentry *p;
1283 int ret = -ENOENT;
1284 write_seqlock(&rename_lock);
1285 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001286 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001287 spin_lock(&p->d_lock);
1288 if (unlikely(d_unhashed(p))) {
1289 spin_unlock(&p->d_lock);
1290 goto out;
1291 }
1292 spin_unlock(&p->d_lock);
1293 }
1294 spin_lock(&dentry->d_lock);
1295 if (!d_unlinked(dentry)) {
1296 dentry->d_flags |= DCACHE_MOUNTED;
1297 ret = 0;
1298 }
1299 spin_unlock(&dentry->d_lock);
1300out:
1301 write_sequnlock(&rename_lock);
1302 return ret;
1303}
1304
1305/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001306 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 * and move any unused dentries to the end of the unused
1308 * list for prune_dcache(). We descend to the next level
1309 * whenever the d_subdirs list is non-empty and continue
1310 * searching.
1311 *
1312 * It returns zero iff there are no unused children,
1313 * otherwise it returns the number of children moved to
1314 * the end of the unused list. This may not be the total
1315 * number of unused children, because select_parent can
1316 * drop the lock and return early due to latency
1317 * constraints.
1318 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001319
1320struct select_data {
1321 struct dentry *start;
1322 struct list_head dispose;
1323 int found;
1324};
1325
1326static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001328 struct select_data *data = _data;
1329 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001331 if (data->start == dentry)
1332 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Al Virofe915222014-05-03 00:02:25 -04001334 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001335 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001336 } else {
1337 if (dentry->d_flags & DCACHE_LRU_LIST)
1338 d_lru_del(dentry);
1339 if (!dentry->d_lockref.count) {
1340 d_shrink_add(dentry, &data->dispose);
1341 data->found++;
1342 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 }
1344 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001345 * We can return to the caller if we have found some (this
1346 * ensures forward progress). We'll be coming back to find
1347 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 */
Al Virofe915222014-05-03 00:02:25 -04001349 if (!list_empty(&data->dispose))
1350 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001352 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353}
1354
1355/**
1356 * shrink_dcache_parent - prune dcache
1357 * @parent: parent of entries to prune
1358 *
1359 * Prune the dcache to remove unused children of the parent dentry.
1360 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001361void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001363 for (;;) {
1364 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001366 INIT_LIST_HEAD(&data.dispose);
1367 data.start = parent;
1368 data.found = 0;
1369
1370 d_walk(parent, &data, select_collect, NULL);
1371 if (!data.found)
1372 break;
1373
1374 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001375 cond_resched();
1376 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001378EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379
Al Viro9c8c10e2014-05-02 20:36:10 -04001380static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001381{
Al Viro9c8c10e2014-05-02 20:36:10 -04001382 /* it has busy descendents; complain about those instead */
1383 if (!list_empty(&dentry->d_subdirs))
1384 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001385
Al Viro9c8c10e2014-05-02 20:36:10 -04001386 /* root with refcount 1 is fine */
1387 if (dentry == _data && dentry->d_lockref.count == 1)
1388 return D_WALK_CONTINUE;
1389
1390 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1391 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001392 dentry,
1393 dentry->d_inode ?
1394 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001395 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001396 dentry->d_lockref.count,
1397 dentry->d_sb->s_type->name,
1398 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001399 WARN_ON(1);
1400 return D_WALK_CONTINUE;
1401}
1402
1403static void do_one_tree(struct dentry *dentry)
1404{
1405 shrink_dcache_parent(dentry);
1406 d_walk(dentry, dentry, umount_check, NULL);
1407 d_drop(dentry);
1408 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001409}
1410
1411/*
1412 * destroy the dentries attached to a superblock on unmounting
1413 */
1414void shrink_dcache_for_umount(struct super_block *sb)
1415{
1416 struct dentry *dentry;
1417
Al Viro9c8c10e2014-05-02 20:36:10 -04001418 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001419
1420 dentry = sb->s_root;
1421 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001422 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001423
1424 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001425 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1426 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001427 }
1428}
1429
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001430struct detach_data {
1431 struct select_data select;
1432 struct dentry *mountpoint;
1433};
1434static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001435{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001436 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001437
1438 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001439 __dget_dlock(dentry);
1440 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001441 return D_WALK_QUIT;
1442 }
1443
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001444 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001445}
1446
1447static void check_and_drop(void *_data)
1448{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001449 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001450
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001451 if (!data->mountpoint && !data->select.found)
1452 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001453}
1454
1455/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001456 * d_invalidate - detach submounts, prune dcache, and drop
1457 * @dentry: dentry to invalidate (aka detach, prune and drop)
1458 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001459 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001460 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001461 * The final d_drop is done as an atomic operation relative to
1462 * rename_lock ensuring there are no races with d_set_mounted. This
1463 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001464 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001465void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001466{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001467 /*
1468 * If it's already been dropped, return OK.
1469 */
1470 spin_lock(&dentry->d_lock);
1471 if (d_unhashed(dentry)) {
1472 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001473 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001474 }
1475 spin_unlock(&dentry->d_lock);
1476
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001477 /* Negative dentries can be dropped without further checks */
1478 if (!dentry->d_inode) {
1479 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001480 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001481 }
1482
1483 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001484 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001485
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001486 data.mountpoint = NULL;
1487 INIT_LIST_HEAD(&data.select.dispose);
1488 data.select.start = dentry;
1489 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001490
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001491 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001492
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001493 if (data.select.found)
1494 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001495
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001496 if (data.mountpoint) {
1497 detach_mounts(data.mountpoint);
1498 dput(data.mountpoint);
1499 }
1500
1501 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001502 break;
1503
1504 cond_resched();
1505 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001506}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001507EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001508
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509/**
Al Viroa4464db2011-07-07 15:03:58 -04001510 * __d_alloc - allocate a dcache entry
1511 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 * @name: qstr of the name
1513 *
1514 * Allocates a dentry. It returns %NULL if there is insufficient memory
1515 * available. On a success the dentry is returned. The name passed in is
1516 * copied and the copy passed in may be reused after this call.
1517 */
1518
Al Viroa4464db2011-07-07 15:03:58 -04001519struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520{
1521 struct dentry *dentry;
1522 char *dname;
1523
Mel Gormane12ba742007-10-16 01:25:52 -07001524 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 if (!dentry)
1526 return NULL;
1527
Linus Torvalds6326c712012-05-21 16:14:04 -07001528 /*
1529 * We guarantee that the inline name is always NUL-terminated.
1530 * This way the memcpy() done by the name switching in rename
1531 * will still always have a NUL at the end, even if we might
1532 * be overwriting an internal NUL character
1533 */
1534 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001536 size_t size = offsetof(struct external_name, name[1]);
1537 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1538 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 kmem_cache_free(dentry_cache, dentry);
1540 return NULL;
1541 }
Al Viro8d85b482014-09-29 14:54:27 -04001542 atomic_set(&p->u.count, 1);
1543 dname = p->name;
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -08001544 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1545 kasan_unpoison_shadow(dname,
1546 round_up(name->len + 1, sizeof(unsigned long)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 } else {
1548 dname = dentry->d_iname;
1549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
1551 dentry->d_name.len = name->len;
1552 dentry->d_name.hash = name->hash;
1553 memcpy(dname, name->name, name->len);
1554 dname[name->len] = 0;
1555
Linus Torvalds6326c712012-05-21 16:14:04 -07001556 /* Make sure we always see the terminating NUL character */
1557 smp_wmb();
1558 dentry->d_name.name = dname;
1559
Waiman Long98474232013-08-28 18:24:59 -07001560 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001561 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001563 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001565 dentry->d_parent = dentry;
1566 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 dentry->d_op = NULL;
1568 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001569 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 INIT_LIST_HEAD(&dentry->d_lru);
1571 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001572 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1573 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001574 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
Nick Piggin3e880fb2011-01-07 17:49:19 +11001576 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001577
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 return dentry;
1579}
Al Viroa4464db2011-07-07 15:03:58 -04001580
1581/**
1582 * d_alloc - allocate a dcache entry
1583 * @parent: parent of entry to allocate
1584 * @name: qstr of the name
1585 *
1586 * Allocates a dentry. It returns %NULL if there is insufficient memory
1587 * available. On a success the dentry is returned. The name passed in is
1588 * copied and the copy passed in may be reused after this call.
1589 */
1590struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1591{
1592 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1593 if (!dentry)
1594 return NULL;
1595
1596 spin_lock(&parent->d_lock);
1597 /*
1598 * don't need child lock because it is not subject
1599 * to concurrency here
1600 */
1601 __dget_dlock(parent);
1602 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001603 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001604 spin_unlock(&parent->d_lock);
1605
1606 return dentry;
1607}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001608EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001610/**
1611 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1612 * @sb: the superblock
1613 * @name: qstr of the name
1614 *
1615 * For a filesystem that just pins its dentries in memory and never
1616 * performs lookups at all, return an unhashed IS_ROOT dentry.
1617 */
Nick Piggin4b936882011-01-07 17:50:07 +11001618struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1619{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001620 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001621}
1622EXPORT_SYMBOL(d_alloc_pseudo);
1623
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1625{
1626 struct qstr q;
1627
1628 q.name = name;
1629 q.len = strlen(name);
1630 q.hash = full_name_hash(q.name, q.len);
1631 return d_alloc(parent, &q);
1632}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001633EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634
Nick Pigginfb045ad2011-01-07 17:49:55 +11001635void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1636{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001637 WARN_ON_ONCE(dentry->d_op);
1638 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001639 DCACHE_OP_COMPARE |
1640 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001641 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001642 DCACHE_OP_DELETE ));
1643 dentry->d_op = op;
1644 if (!op)
1645 return;
1646 if (op->d_hash)
1647 dentry->d_flags |= DCACHE_OP_HASH;
1648 if (op->d_compare)
1649 dentry->d_flags |= DCACHE_OP_COMPARE;
1650 if (op->d_revalidate)
1651 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001652 if (op->d_weak_revalidate)
1653 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001654 if (op->d_delete)
1655 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001656 if (op->d_prune)
1657 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001658
1659}
1660EXPORT_SYMBOL(d_set_d_op);
1661
David Howellsb18825a2013-09-12 19:22:53 +01001662static unsigned d_flags_for_inode(struct inode *inode)
1663{
1664 unsigned add_flags = DCACHE_FILE_TYPE;
1665
1666 if (!inode)
1667 return DCACHE_MISS_TYPE;
1668
1669 if (S_ISDIR(inode->i_mode)) {
1670 add_flags = DCACHE_DIRECTORY_TYPE;
1671 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1672 if (unlikely(!inode->i_op->lookup))
1673 add_flags = DCACHE_AUTODIR_TYPE;
1674 else
1675 inode->i_opflags |= IOP_LOOKUP;
1676 }
1677 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1678 if (unlikely(inode->i_op->follow_link))
1679 add_flags = DCACHE_SYMLINK_TYPE;
1680 else
1681 inode->i_opflags |= IOP_NOFOLLOW;
1682 }
1683
1684 if (unlikely(IS_AUTOMOUNT(inode)))
1685 add_flags |= DCACHE_NEED_AUTOMOUNT;
1686 return add_flags;
1687}
1688
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001689static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1690{
David Howellsb18825a2013-09-12 19:22:53 +01001691 unsigned add_flags = d_flags_for_inode(inode);
1692
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001693 spin_lock(&dentry->d_lock);
Al Viro22213312014-04-19 12:30:58 -04001694 __d_set_type(dentry, add_flags);
David Howellsb18825a2013-09-12 19:22:53 +01001695 if (inode)
Al Viro946e51f2014-10-26 19:19:16 -04001696 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001697 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001698 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001699 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001700 fsnotify_d_instantiate(dentry, inode);
1701}
1702
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703/**
1704 * d_instantiate - fill in inode information for a dentry
1705 * @entry: dentry to complete
1706 * @inode: inode to attach to this dentry
1707 *
1708 * Fill in inode information in the entry.
1709 *
1710 * This turns negative dentries into productive full members
1711 * of society.
1712 *
1713 * NOTE! This assumes that the inode count has been incremented
1714 * (or otherwise set) by the caller to indicate that it is now
1715 * in use by the dcache.
1716 */
1717
1718void d_instantiate(struct dentry *entry, struct inode * inode)
1719{
Al Viro946e51f2014-10-26 19:19:16 -04001720 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001721 if (inode)
1722 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001723 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001724 if (inode)
1725 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 security_d_instantiate(entry, inode);
1727}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001728EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729
1730/**
1731 * d_instantiate_unique - instantiate a non-aliased dentry
1732 * @entry: dentry to instantiate
1733 * @inode: inode to attach to this dentry
1734 *
1735 * Fill in inode information in the entry. On success, it returns NULL.
1736 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001737 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 *
1739 * Note that in order to avoid conflicts with rename() etc, the caller
1740 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001741 *
1742 * This also assumes that the inode count has been incremented
1743 * (or otherwise set) by the caller to indicate that it is now
1744 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 */
David Howells770bfad2006-08-22 20:06:07 -04001746static struct dentry *__d_instantiate_unique(struct dentry *entry,
1747 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748{
1749 struct dentry *alias;
1750 int len = entry->d_name.len;
1751 const char *name = entry->d_name.name;
1752 unsigned int hash = entry->d_name.hash;
1753
David Howells770bfad2006-08-22 20:06:07 -04001754 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001755 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001756 return NULL;
1757 }
1758
Al Viro946e51f2014-10-26 19:19:16 -04001759 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001760 /*
1761 * Don't need alias->d_lock here, because aliases with
1762 * d_parent == entry->d_parent are not subject to name or
1763 * parent changes, because the parent inode i_mutex is held.
1764 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001765 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 continue;
1767 if (alias->d_parent != entry->d_parent)
1768 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001769 if (alias->d_name.len != len)
1770 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001771 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001773 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 return alias;
1775 }
David Howells770bfad2006-08-22 20:06:07 -04001776
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001777 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 return NULL;
1779}
David Howells770bfad2006-08-22 20:06:07 -04001780
1781struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1782{
1783 struct dentry *result;
1784
Al Viro946e51f2014-10-26 19:19:16 -04001785 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001786
Nick Piggin873feea2011-01-07 17:50:06 +11001787 if (inode)
1788 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001789 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001790 if (inode)
1791 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001792
1793 if (!result) {
1794 security_d_instantiate(entry, inode);
1795 return NULL;
1796 }
1797
1798 BUG_ON(!d_unhashed(result));
1799 iput(inode);
1800 return result;
1801}
1802
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803EXPORT_SYMBOL(d_instantiate_unique);
1804
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001805/**
1806 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1807 * @entry: dentry to complete
1808 * @inode: inode to attach to this dentry
1809 *
1810 * Fill in inode information in the entry. If a directory alias is found, then
1811 * return an error (and drop inode). Together with d_materialise_unique() this
1812 * guarantees that a directory inode may never have more than one alias.
1813 */
1814int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1815{
Al Viro946e51f2014-10-26 19:19:16 -04001816 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001817
1818 spin_lock(&inode->i_lock);
1819 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1820 spin_unlock(&inode->i_lock);
1821 iput(inode);
1822 return -EBUSY;
1823 }
1824 __d_instantiate(entry, inode);
1825 spin_unlock(&inode->i_lock);
1826 security_d_instantiate(entry, inode);
1827
1828 return 0;
1829}
1830EXPORT_SYMBOL(d_instantiate_no_diralias);
1831
Al Viroadc0e912012-01-08 16:49:21 -05001832struct dentry *d_make_root(struct inode *root_inode)
1833{
1834 struct dentry *res = NULL;
1835
1836 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001837 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001838
1839 res = __d_alloc(root_inode->i_sb, &name);
1840 if (res)
1841 d_instantiate(res, root_inode);
1842 else
1843 iput(root_inode);
1844 }
1845 return res;
1846}
1847EXPORT_SYMBOL(d_make_root);
1848
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001849static struct dentry * __d_find_any_alias(struct inode *inode)
1850{
1851 struct dentry *alias;
1852
Al Virob3d9b7a2012-06-09 13:51:19 -04001853 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001854 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001855 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001856 __dget(alias);
1857 return alias;
1858}
1859
Sage Weil46f72b32012-01-10 09:04:37 -08001860/**
1861 * d_find_any_alias - find any alias for a given inode
1862 * @inode: inode to find an alias for
1863 *
1864 * If any aliases exist for the given inode, take and return a
1865 * reference for one of them. If no aliases exist, return %NULL.
1866 */
1867struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001868{
1869 struct dentry *de;
1870
1871 spin_lock(&inode->i_lock);
1872 de = __d_find_any_alias(inode);
1873 spin_unlock(&inode->i_lock);
1874 return de;
1875}
Sage Weil46f72b32012-01-10 09:04:37 -08001876EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001877
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001878static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001879{
NeilBrownb911a6b2012-11-08 16:09:37 -08001880 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001881 struct dentry *tmp;
1882 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001883 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001884
1885 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001886 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001887 if (IS_ERR(inode))
1888 return ERR_CAST(inode);
1889
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001890 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001891 if (res)
1892 goto out_iput;
1893
Al Viroa4464db2011-07-07 15:03:58 -04001894 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001895 if (!tmp) {
1896 res = ERR_PTR(-ENOMEM);
1897 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001898 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001899
Nick Piggin873feea2011-01-07 17:50:06 +11001900 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001901 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001902 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001903 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001904 dput(tmp);
1905 goto out_iput;
1906 }
1907
1908 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001909 add_flags = d_flags_for_inode(inode);
1910
1911 if (disconnected)
1912 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001913
Christoph Hellwig9308a612008-08-11 15:49:12 +02001914 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001915 tmp->d_inode = inode;
David Howellsb18825a2013-09-12 19:22:53 +01001916 tmp->d_flags |= add_flags;
Al Viro946e51f2014-10-26 19:19:16 -04001917 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001918 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001919 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001920 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001921 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001922 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001923 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001924
Christoph Hellwig9308a612008-08-11 15:49:12 +02001925 return tmp;
1926
1927 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001928 if (res && !IS_ERR(res))
1929 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001930 iput(inode);
1931 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001932}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001933
1934/**
1935 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1936 * @inode: inode to allocate the dentry for
1937 *
1938 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1939 * similar open by handle operations. The returned dentry may be anonymous,
1940 * or may have a full name (if the inode was already in the cache).
1941 *
1942 * When called on a directory inode, we must ensure that the inode only ever
1943 * has one dentry. If a dentry is found, that is returned instead of
1944 * allocating a new one.
1945 *
1946 * On successful return, the reference to the inode has been transferred
1947 * to the dentry. In case of an error the reference on the inode is released.
1948 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1949 * be passed in and the error will be propagated to the return value,
1950 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1951 */
1952struct dentry *d_obtain_alias(struct inode *inode)
1953{
1954 return __d_obtain_alias(inode, 1);
1955}
Benny Halevyadc48722009-02-27 14:02:59 -08001956EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
1958/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001959 * d_obtain_root - find or allocate a dentry for a given inode
1960 * @inode: inode to allocate the dentry for
1961 *
1962 * Obtain an IS_ROOT dentry for the root of a filesystem.
1963 *
1964 * We must ensure that directory inodes only ever have one dentry. If a
1965 * dentry is found, that is returned instead of allocating a new one.
1966 *
1967 * On successful return, the reference to the inode has been transferred
1968 * to the dentry. In case of an error the reference on the inode is
1969 * released. A %NULL or IS_ERR inode may be passed in and will be the
1970 * error will be propagate to the return value, with a %NULL @inode
1971 * replaced by ERR_PTR(-ESTALE).
1972 */
1973struct dentry *d_obtain_root(struct inode *inode)
1974{
1975 return __d_obtain_alias(inode, 0);
1976}
1977EXPORT_SYMBOL(d_obtain_root);
1978
1979/**
Barry Naujok94035402008-05-21 16:50:46 +10001980 * d_add_ci - lookup or allocate new dentry with case-exact name
1981 * @inode: the inode case-insensitive lookup has found
1982 * @dentry: the negative dentry that was passed to the parent's lookup func
1983 * @name: the case-exact name to be associated with the returned dentry
1984 *
1985 * This is to avoid filling the dcache with case-insensitive names to the
1986 * same inode, only the actual correct case is stored in the dcache for
1987 * case-insensitive filesystems.
1988 *
1989 * For a case-insensitive lookup match and if the the case-exact dentry
1990 * already exists in in the dcache, use it and return it.
1991 *
1992 * If no entry exists with the exact case name, allocate new dentry with
1993 * the exact case, and return the spliced entry.
1994 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001995struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001996 struct qstr *name)
1997{
Barry Naujok94035402008-05-21 16:50:46 +10001998 struct dentry *found;
1999 struct dentry *new;
2000
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002001 /*
2002 * First check if a dentry matching the name already exists,
2003 * if not go ahead and create it now.
2004 */
Barry Naujok94035402008-05-21 16:50:46 +10002005 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10002006 if (!found) {
2007 new = d_alloc(dentry->d_parent, name);
2008 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05002009 found = ERR_PTR(-ENOMEM);
Al Viro427c77d2014-10-12 18:46:26 -04002010 } else {
2011 found = d_splice_alias(inode, new);
2012 if (found) {
2013 dput(new);
2014 return found;
2015 }
2016 return new;
Barry Naujok94035402008-05-21 16:50:46 +10002017 }
Barry Naujok94035402008-05-21 16:50:46 +10002018 }
Barry Naujok94035402008-05-21 16:50:46 +10002019 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05002020 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002021}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002022EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002024/*
2025 * Do the slow-case of the dentry name compare.
2026 *
2027 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07002028 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002029 * filesystem can rely on them, and can use the 'name' and
2030 * 'len' information without worrying about walking off the
2031 * end of memory etc.
2032 *
2033 * Thus the read_seqcount_retry() and the "duplicate" info
2034 * in arguments (the low-level filesystem should not look
2035 * at the dentry inode or name contents directly, since
2036 * rename can change them while we're in RCU mode).
2037 */
2038enum slow_d_compare {
2039 D_COMP_OK,
2040 D_COMP_NOMATCH,
2041 D_COMP_SEQRETRY,
2042};
2043
2044static noinline enum slow_d_compare slow_dentry_cmp(
2045 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002046 struct dentry *dentry,
2047 unsigned int seq,
2048 const struct qstr *name)
2049{
2050 int tlen = dentry->d_name.len;
2051 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002052
2053 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2054 cpu_relax();
2055 return D_COMP_SEQRETRY;
2056 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002057 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002058 return D_COMP_NOMATCH;
2059 return D_COMP_OK;
2060}
2061
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002063 * __d_lookup_rcu - search for a dentry (racy, store-free)
2064 * @parent: parent dentry
2065 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002066 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002067 * Returns: dentry, or NULL
2068 *
2069 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2070 * resolution (store-free path walking) design described in
2071 * Documentation/filesystems/path-lookup.txt.
2072 *
2073 * This is not to be used outside core vfs.
2074 *
2075 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2076 * held, and rcu_read_lock held. The returned dentry must not be stored into
2077 * without taking d_lock and checking d_seq sequence count against @seq
2078 * returned here.
2079 *
Linus Torvalds15570082013-09-02 11:38:06 -07002080 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002081 * function.
2082 *
2083 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2084 * the returned dentry, so long as its parent's seqlock is checked after the
2085 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2086 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002087 *
2088 * NOTE! The caller *has* to check the resulting dentry against the sequence
2089 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002090 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002091struct dentry *__d_lookup_rcu(const struct dentry *parent,
2092 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002093 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002094{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002095 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002096 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002097 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002098 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002099 struct dentry *dentry;
2100
2101 /*
2102 * Note: There is significant duplication with __d_lookup_rcu which is
2103 * required to prevent single threaded performance regressions
2104 * especially on architectures where smp_rmb (in seqcounts) are costly.
2105 * Keep the two functions in sync.
2106 */
2107
2108 /*
2109 * The hash list is protected using RCU.
2110 *
2111 * Carefully use d_seq when comparing a candidate dentry, to avoid
2112 * races with d_move().
2113 *
2114 * It is possible that concurrent renames can mess up our list
2115 * walk here and result in missing our dentry, resulting in the
2116 * false-negative result. d_lookup() protects against concurrent
2117 * renames using rename_lock seqlock.
2118 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002119 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002120 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002121 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002122 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002123
Nick Piggin31e6b012011-01-07 17:49:52 +11002124seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002125 /*
2126 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002127 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002128 *
2129 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002130 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002131 *
2132 * NOTE! We do a "raw" seqcount_begin here. That means that
2133 * we don't wait for the sequence count to stabilize if it
2134 * is in the middle of a sequence change. If we do the slow
2135 * dentry compare, we will do seqretries until it is stable,
2136 * and if we end up with a successful lookup, we actually
2137 * want to exit RCU lookup anyway.
2138 */
2139 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002140 if (dentry->d_parent != parent)
2141 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002142 if (d_unhashed(dentry))
2143 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002144
2145 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002146 if (dentry->d_name.hash != hashlen_hash(hashlen))
2147 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002148 *seqp = seq;
2149 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002150 case D_COMP_OK:
2151 return dentry;
2152 case D_COMP_NOMATCH:
2153 continue;
2154 default:
2155 goto seqretry;
2156 }
2157 }
2158
Linus Torvalds26fe5752012-05-10 13:14:12 -07002159 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002160 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002161 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002162 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002163 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002164 }
2165 return NULL;
2166}
2167
2168/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 * d_lookup - search for a dentry
2170 * @parent: parent dentry
2171 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002172 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002174 * d_lookup searches the children of the parent dentry for the name in
2175 * question. If the dentry is found its reference count is incremented and the
2176 * dentry is returned. The caller must use dput to free the entry when it has
2177 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 */
Al Viroda2d8452013-01-24 18:29:34 -05002179struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180{
Nick Piggin31e6b012011-01-07 17:49:52 +11002181 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002182 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183
Daeseok Younb8314f92014-08-11 11:46:53 +09002184 do {
2185 seq = read_seqbegin(&rename_lock);
2186 dentry = __d_lookup(parent, name);
2187 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 break;
2189 } while (read_seqretry(&rename_lock, seq));
2190 return dentry;
2191}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002192EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193
Nick Piggin31e6b012011-01-07 17:49:52 +11002194/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002195 * __d_lookup - search for a dentry (racy)
2196 * @parent: parent dentry
2197 * @name: qstr of name we wish to find
2198 * Returns: dentry, or NULL
2199 *
2200 * __d_lookup is like d_lookup, however it may (rarely) return a
2201 * false-negative result due to unrelated rename activity.
2202 *
2203 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2204 * however it must be used carefully, eg. with a following d_lookup in
2205 * the case of failure.
2206 *
2207 * __d_lookup callers must be commented.
2208 */
Al Viroa713ca22013-01-24 18:27:00 -05002209struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210{
2211 unsigned int len = name->len;
2212 unsigned int hash = name->hash;
2213 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002214 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002215 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002216 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002217 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218
Nick Pigginb04f7842010-08-18 04:37:34 +10002219 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002220 * Note: There is significant duplication with __d_lookup_rcu which is
2221 * required to prevent single threaded performance regressions
2222 * especially on architectures where smp_rmb (in seqcounts) are costly.
2223 * Keep the two functions in sync.
2224 */
2225
2226 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002227 * The hash list is protected using RCU.
2228 *
2229 * Take d_lock when comparing a candidate dentry, to avoid races
2230 * with d_move().
2231 *
2232 * It is possible that concurrent renames can mess up our list
2233 * walk here and result in missing our dentry, resulting in the
2234 * false-negative result. d_lookup() protects against concurrent
2235 * renames using rename_lock seqlock.
2236 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002237 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002238 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 rcu_read_lock();
2240
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002241 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 if (dentry->d_name.hash != hash)
2244 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
2246 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 if (dentry->d_parent != parent)
2248 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002249 if (d_unhashed(dentry))
2250 goto next;
2251
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 /*
2253 * It is safe to compare names since d_move() cannot
2254 * change the qstr (protected by d_lock).
2255 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002256 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002257 int tlen = dentry->d_name.len;
2258 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002259 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 goto next;
2261 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002262 if (dentry->d_name.len != len)
2263 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002264 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 goto next;
2266 }
2267
Waiman Long98474232013-08-28 18:24:59 -07002268 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002269 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 spin_unlock(&dentry->d_lock);
2271 break;
2272next:
2273 spin_unlock(&dentry->d_lock);
2274 }
2275 rcu_read_unlock();
2276
2277 return found;
2278}
2279
2280/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002281 * d_hash_and_lookup - hash the qstr then search for a dentry
2282 * @dir: Directory to search in
2283 * @name: qstr of name we wish to find
2284 *
Al Viro4f522a22013-02-11 23:20:37 -05002285 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002286 */
2287struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2288{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002289 /*
2290 * Check for a fs-specific hash function. Note that we must
2291 * calculate the standard hash first, as the d_op->d_hash()
2292 * routine may choose to leave the hash value unchanged.
2293 */
2294 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002295 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002296 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002297 if (unlikely(err < 0))
2298 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002299 }
Al Viro4f522a22013-02-11 23:20:37 -05002300 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002301}
Al Viro4f522a22013-02-11 23:20:37 -05002302EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002303
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304/*
2305 * When a file is deleted, we have two options:
2306 * - turn this dentry into a negative dentry
2307 * - unhash this dentry and free it.
2308 *
2309 * Usually, we want to just turn this into
2310 * a negative dentry, but if anybody else is
2311 * currently using the dentry or the inode
2312 * we can't do that and we fall back on removing
2313 * it from the hash queues and waiting for
2314 * it to be deleted later when it has no users
2315 */
2316
2317/**
2318 * d_delete - delete a dentry
2319 * @dentry: The dentry to delete
2320 *
2321 * Turn the dentry into a negative dentry if possible, otherwise
2322 * remove it from the hash queues so it can be deleted later
2323 */
2324
2325void d_delete(struct dentry * dentry)
2326{
Nick Piggin873feea2011-01-07 17:50:06 +11002327 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002328 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 /*
2330 * Are we the only user?
2331 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002332again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002334 inode = dentry->d_inode;
2335 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002336 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002337 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002338 spin_unlock(&dentry->d_lock);
2339 cpu_relax();
2340 goto again;
2341 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002342 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002343 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002344 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 return;
2346 }
2347
2348 if (!d_unhashed(dentry))
2349 __d_drop(dentry);
2350
2351 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002352
2353 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002355EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002357static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002359 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002360 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002361 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002362 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002363 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364}
2365
David Howells770bfad2006-08-22 20:06:07 -04002366static void _d_rehash(struct dentry * entry)
2367{
2368 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2369}
2370
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371/**
2372 * d_rehash - add an entry back to the hash
2373 * @entry: dentry to add to the hash
2374 *
2375 * Adds a dentry to the hash according to its name.
2376 */
2377
2378void d_rehash(struct dentry * entry)
2379{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002381 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002384EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002386/**
2387 * dentry_update_name_case - update case insensitive dentry with a new name
2388 * @dentry: dentry to be updated
2389 * @name: new name
2390 *
2391 * Update a case insensitive dentry with new case of name.
2392 *
2393 * dentry must have been returned by d_lookup with name @name. Old and new
2394 * name lengths must match (ie. no d_compare which allows mismatched name
2395 * lengths).
2396 *
2397 * Parent inode i_mutex must be held over d_lookup and into this call (to
2398 * keep renames and concurrent inserts, and readdir(2) away).
2399 */
2400void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2401{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002402 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002403 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2404
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002405 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002406 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002407 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002408 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002409 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002410}
2411EXPORT_SYMBOL(dentry_update_name_case);
2412
Al Viro8d85b482014-09-29 14:54:27 -04002413static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414{
Al Viro8d85b482014-09-29 14:54:27 -04002415 if (unlikely(dname_external(target))) {
2416 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 /*
2418 * Both external: swap the pointers
2419 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002420 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 } else {
2422 /*
2423 * dentry:internal, target:external. Steal target's
2424 * storage and make target internal.
2425 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002426 memcpy(target->d_iname, dentry->d_name.name,
2427 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 dentry->d_name.name = target->d_name.name;
2429 target->d_name.name = target->d_iname;
2430 }
2431 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002432 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 /*
2434 * dentry:external, target:internal. Give dentry's
2435 * storage to target and make dentry internal
2436 */
2437 memcpy(dentry->d_iname, target->d_name.name,
2438 target->d_name.len + 1);
2439 target->d_name.name = dentry->d_name.name;
2440 dentry->d_name.name = dentry->d_iname;
2441 } else {
2442 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002443 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002445 unsigned int i;
2446 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002447 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2448 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002449 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2450 swap(((long *) &dentry->d_iname)[i],
2451 ((long *) &target->d_iname)[i]);
2452 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453 }
2454 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002455 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456}
2457
Al Viro8d85b482014-09-29 14:54:27 -04002458static void copy_name(struct dentry *dentry, struct dentry *target)
2459{
2460 struct external_name *old_name = NULL;
2461 if (unlikely(dname_external(dentry)))
2462 old_name = external_name(dentry);
2463 if (unlikely(dname_external(target))) {
2464 atomic_inc(&external_name(target)->u.count);
2465 dentry->d_name = target->d_name;
2466 } else {
2467 memcpy(dentry->d_iname, target->d_name.name,
2468 target->d_name.len + 1);
2469 dentry->d_name.name = dentry->d_iname;
2470 dentry->d_name.hash_len = target->d_name.hash_len;
2471 }
2472 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2473 kfree_rcu(old_name, u.head);
2474}
2475
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002476static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2477{
2478 /*
2479 * XXXX: do we really need to take target->d_lock?
2480 */
2481 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2482 spin_lock(&target->d_parent->d_lock);
2483 else {
2484 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2485 spin_lock(&dentry->d_parent->d_lock);
2486 spin_lock_nested(&target->d_parent->d_lock,
2487 DENTRY_D_LOCK_NESTED);
2488 } else {
2489 spin_lock(&target->d_parent->d_lock);
2490 spin_lock_nested(&dentry->d_parent->d_lock,
2491 DENTRY_D_LOCK_NESTED);
2492 }
2493 }
2494 if (target < dentry) {
2495 spin_lock_nested(&target->d_lock, 2);
2496 spin_lock_nested(&dentry->d_lock, 3);
2497 } else {
2498 spin_lock_nested(&dentry->d_lock, 2);
2499 spin_lock_nested(&target->d_lock, 3);
2500 }
2501}
2502
Al Viro986c0192014-09-26 23:11:15 -04002503static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002504{
2505 if (target->d_parent != dentry->d_parent)
2506 spin_unlock(&dentry->d_parent->d_lock);
2507 if (target->d_parent != target)
2508 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002509 spin_unlock(&target->d_lock);
2510 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002511}
2512
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002514 * When switching names, the actual string doesn't strictly have to
2515 * be preserved in the target - because we're dropping the target
2516 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002517 * the new name before we switch, unless we are going to rehash
2518 * it. Note that if we *do* unhash the target, we are not allowed
2519 * to rehash it without giving it a new name/hash key - whether
2520 * we swap or overwrite the names here, resulting name won't match
2521 * the reality in filesystem; it's only there for d_path() purposes.
2522 * Note that all of this is happening under rename_lock, so the
2523 * any hash lookup seeing it in the middle of manipulations will
2524 * be discarded anyway. So we do not care what happens to the hash
2525 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002527/*
Al Viro18367502011-07-12 21:42:24 -04002528 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 * @dentry: entry to move
2530 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002531 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 *
2533 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002534 * dcache entries should not be moved in this way. Caller must hold
2535 * rename_lock, the i_mutex of the source and target directories,
2536 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002538static void __d_move(struct dentry *dentry, struct dentry *target,
2539 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 if (!dentry->d_inode)
2542 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2543
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002544 BUG_ON(d_ancestor(dentry, target));
2545 BUG_ON(d_ancestor(target, dentry));
2546
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002547 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548
Nick Piggin31e6b012011-01-07 17:49:52 +11002549 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002550 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002551
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002552 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2553
2554 /*
2555 * Move the dentry to the target hash queue. Don't bother checking
2556 * for the same hash queue because of how unlikely it is.
2557 */
2558 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002559 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
Miklos Szeredida1ce062014-04-01 17:08:43 +02002561 /*
2562 * Unhash the target (d_delete() is not usable here). If exchanging
2563 * the two dentries, then rehash onto the other's hash queue.
2564 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002566 if (exchange) {
2567 __d_rehash(target,
2568 d_hash(dentry->d_parent, dentry->d_name.hash));
2569 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002572 if (exchange)
2573 swap_names(dentry, target);
2574 else
2575 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576
Al Viro63cf4272014-09-26 23:06:14 -04002577 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002579 /* splicing a tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002580 dentry->d_parent = target->d_parent;
2581 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002582 list_del_init(&target->d_child);
2583 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002585 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002586 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002587 list_move(&target->d_child, &target->d_parent->d_subdirs);
2588 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002589 if (exchange)
2590 fsnotify_d_move(target);
2591 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 }
2593
Nick Piggin31e6b012011-01-07 17:49:52 +11002594 write_seqcount_end(&target->d_seq);
2595 write_seqcount_end(&dentry->d_seq);
2596
Al Viro986c0192014-09-26 23:11:15 -04002597 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002598}
2599
2600/*
2601 * d_move - move a dentry
2602 * @dentry: entry to move
2603 * @target: new dentry
2604 *
2605 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002606 * dcache entries should not be moved in this way. See the locking
2607 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002608 */
2609void d_move(struct dentry *dentry, struct dentry *target)
2610{
2611 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002612 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002614}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002615EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
Miklos Szeredida1ce062014-04-01 17:08:43 +02002617/*
2618 * d_exchange - exchange two dentries
2619 * @dentry1: first dentry
2620 * @dentry2: second dentry
2621 */
2622void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2623{
2624 write_seqlock(&rename_lock);
2625
2626 WARN_ON(!dentry1->d_inode);
2627 WARN_ON(!dentry2->d_inode);
2628 WARN_ON(IS_ROOT(dentry1));
2629 WARN_ON(IS_ROOT(dentry2));
2630
2631 __d_move(dentry1, dentry2, true);
2632
2633 write_sequnlock(&rename_lock);
2634}
2635
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002636/**
2637 * d_ancestor - search for an ancestor
2638 * @p1: ancestor dentry
2639 * @p2: child dentry
2640 *
2641 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2642 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002643 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002644struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002645{
2646 struct dentry *p;
2647
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002648 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002649 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002650 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002651 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002652 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002653}
2654
2655/*
2656 * This helper attempts to cope with remotely renamed directories
2657 *
2658 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002659 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002660 *
2661 * Note: If ever the locking in lock_rename() changes, then please
2662 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002663 */
Al Virob5ae6b12014-10-12 22:16:02 -04002664static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002665 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002666{
2667 struct mutex *m1 = NULL, *m2 = NULL;
Al Virob5ae6b12014-10-12 22:16:02 -04002668 int ret = -EBUSY;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002669
2670 /* If alias and dentry share a parent, then no extra locks required */
2671 if (alias->d_parent == dentry->d_parent)
2672 goto out_unalias;
2673
Trond Myklebust9eaef272006-10-21 10:24:20 -07002674 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002675 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2676 goto out_err;
2677 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2678 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2679 goto out_err;
2680 m2 = &alias->d_parent->d_inode->i_mutex;
2681out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002682 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04002683 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002684out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002685 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002686 if (m2)
2687 mutex_unlock(m2);
2688 if (m1)
2689 mutex_unlock(m1);
2690 return ret;
2691}
2692
David Howells770bfad2006-08-22 20:06:07 -04002693/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002694 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2695 * @inode: the inode which may have a disconnected dentry
2696 * @dentry: a negative dentry which we want to point to the inode.
2697 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002698 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2699 * place of the given dentry and return it, else simply d_add the inode
2700 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002701 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002702 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2703 * we should error out: directories can't have multiple aliases.
2704 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002705 * This is needed in the lookup routine of any filesystem that is exportable
2706 * (via knfsd) so that we can build dcache paths to directories effectively.
2707 *
2708 * If a dentry was found and moved, then it is returned. Otherwise NULL
2709 * is returned. This matches the expected return value of ->lookup.
2710 *
2711 * Cluster filesystems may call this function with a negative, hashed dentry.
2712 * In that case, we know that the inode will be a regular file, and also this
2713 * will only occur during atomic_open. So we need to check for the dentry
2714 * being already hashed only in the final case.
2715 */
2716struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2717{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002718 if (IS_ERR(inode))
2719 return ERR_CAST(inode);
2720
David Howells770bfad2006-08-22 20:06:07 -04002721 BUG_ON(!d_unhashed(dentry));
2722
David Howells770bfad2006-08-22 20:06:07 -04002723 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002724 __d_instantiate(dentry, NULL);
Al Virob5ae6b12014-10-12 22:16:02 -04002725 goto out;
David Howells770bfad2006-08-22 20:06:07 -04002726 }
Nick Piggin873feea2011-01-07 17:50:06 +11002727 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002728 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04002729 struct dentry *new = __d_find_any_alias(inode);
2730 if (unlikely(new)) {
Al Viro18367502011-07-12 21:42:24 -04002731 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002732 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04002733 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002734 spin_unlock(&inode->i_lock);
2735 dput(new);
2736 new = ERR_PTR(-ELOOP);
2737 pr_warn_ratelimited(
2738 "VFS: Lookup of '%s' in %s %s"
2739 " would have caused loop\n",
2740 dentry->d_name.name,
2741 inode->i_sb->s_type->name,
2742 inode->i_sb->s_id);
2743 } else if (!IS_ROOT(new)) {
2744 int err = __d_unalias(inode, dentry, new);
2745 write_sequnlock(&rename_lock);
2746 if (err) {
2747 dput(new);
2748 new = ERR_PTR(err);
2749 }
Al Viro18367502011-07-12 21:42:24 -04002750 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04002751 __d_move(new, dentry, false);
2752 write_sequnlock(&rename_lock);
2753 spin_unlock(&inode->i_lock);
2754 security_d_instantiate(new, inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002755 }
Al Virob5ae6b12014-10-12 22:16:02 -04002756 iput(inode);
2757 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002758 }
David Howells770bfad2006-08-22 20:06:07 -04002759 }
Al Virob5ae6b12014-10-12 22:16:02 -04002760 /* already taking inode->i_lock, so d_add() by hand */
2761 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11002762 spin_unlock(&inode->i_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002763out:
2764 security_d_instantiate(dentry, inode);
2765 d_rehash(dentry);
2766 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04002767}
Al Virob5ae6b12014-10-12 22:16:02 -04002768EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04002769
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002770static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002771{
2772 *buflen -= namelen;
2773 if (*buflen < 0)
2774 return -ENAMETOOLONG;
2775 *buffer -= namelen;
2776 memcpy(*buffer, str, namelen);
2777 return 0;
2778}
2779
Waiman Long232d2d62013-09-09 12:18:13 -04002780/**
2781 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002782 * @buffer: buffer pointer
2783 * @buflen: allocated length of the buffer
2784 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002785 *
2786 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2787 * make sure that either the old or the new name pointer and length are
2788 * fetched. However, there may be mismatch between length and pointer.
2789 * The length cannot be trusted, we need to copy it byte-by-byte until
2790 * the length is reached or a null byte is found. It also prepends "/" at
2791 * the beginning of the name. The sequence number check at the caller will
2792 * retry it again when a d_move() does happen. So any garbage in the buffer
2793 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04002794 *
2795 * Data dependency barrier is needed to make sure that we see that terminating
2796 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04002797 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002798static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2799{
Waiman Long232d2d62013-09-09 12:18:13 -04002800 const char *dname = ACCESS_ONCE(name->name);
2801 u32 dlen = ACCESS_ONCE(name->len);
2802 char *p;
2803
Al Viro6d13f692014-09-29 14:46:30 -04002804 smp_read_barrier_depends();
2805
Waiman Long232d2d62013-09-09 12:18:13 -04002806 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04002807 if (*buflen < 0)
2808 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04002809 p = *buffer -= dlen + 1;
2810 *p++ = '/';
2811 while (dlen--) {
2812 char c = *dname++;
2813 if (!c)
2814 break;
2815 *p++ = c;
2816 }
2817 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002818}
2819
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002821 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002822 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002823 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002824 * @buffer: pointer to the end of the buffer
2825 * @buflen: pointer to buffer length
2826 *
Waiman Long18129972013-09-12 10:55:35 -04002827 * The function will first try to write out the pathname without taking any
2828 * lock other than the RCU read lock to make sure that dentries won't go away.
2829 * It only checks the sequence number of the global rename_lock as any change
2830 * in the dentry's d_seq will be preceded by changes in the rename_lock
2831 * sequence number. If the sequence number had been changed, it will restart
2832 * the whole pathname back-tracing sequence again by taking the rename_lock.
2833 * In this case, there is no need to take the RCU read lock as the recursive
2834 * parent pointer references will keep the dentry chain alive as long as no
2835 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002836 */
Al Viro02125a82011-12-05 08:43:34 -05002837static int prepend_path(const struct path *path,
2838 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002839 char **buffer, int *buflen)
2840{
Al Viroede4ceb2013-11-13 07:45:40 -05002841 struct dentry *dentry;
2842 struct vfsmount *vfsmnt;
2843 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002844 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002845 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002846 char *bptr;
2847 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002848
Al Viro48f5ec22013-09-09 15:22:25 -04002849 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002850restart_mnt:
2851 read_seqbegin_or_lock(&mount_lock, &m_seq);
2852 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002853 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002854restart:
2855 bptr = *buffer;
2856 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002857 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002858 dentry = path->dentry;
2859 vfsmnt = path->mnt;
2860 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002861 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002862 while (dentry != root->dentry || vfsmnt != root->mnt) {
2863 struct dentry * parent;
2864
2865 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002866 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002867 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002868 if (mnt != parent) {
2869 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2870 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002871 vfsmnt = &mnt->mnt;
2872 continue;
2873 }
2874 /*
2875 * Filesystems needing to implement special "root names"
2876 * should do so with ->d_dname()
2877 */
2878 if (IS_ROOT(dentry) &&
2879 (dentry->d_name.len != 1 ||
2880 dentry->d_name.name[0] != '/')) {
2881 WARN(1, "Root dentry has weird name <%.*s>\n",
2882 (int) dentry->d_name.len,
2883 dentry->d_name.name);
2884 }
2885 if (!error)
2886 error = is_mounted(vfsmnt) ? 1 : 2;
2887 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002888 }
2889 parent = dentry->d_parent;
2890 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002891 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002892 if (error)
2893 break;
2894
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002895 dentry = parent;
2896 }
Al Viro48f5ec22013-09-09 15:22:25 -04002897 if (!(seq & 1))
2898 rcu_read_unlock();
2899 if (need_seqretry(&rename_lock, seq)) {
2900 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002901 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002902 }
Al Viro48f5ec22013-09-09 15:22:25 -04002903 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002904
2905 if (!(m_seq & 1))
2906 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002907 if (need_seqretry(&mount_lock, m_seq)) {
2908 m_seq = 1;
2909 goto restart_mnt;
2910 }
2911 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002912
Waiman Long232d2d62013-09-09 12:18:13 -04002913 if (error >= 0 && bptr == *buffer) {
2914 if (--blen < 0)
2915 error = -ENAMETOOLONG;
2916 else
2917 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002918 }
Waiman Long232d2d62013-09-09 12:18:13 -04002919 *buffer = bptr;
2920 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002921 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002922}
2923
2924/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002925 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002926 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002927 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002928 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929 * @buflen: buffer length
2930 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002931 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002933 * Returns a pointer into the buffer or an error code if the
2934 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002935 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002936 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002937 *
Al Viro02125a82011-12-05 08:43:34 -05002938 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 */
Al Viro02125a82011-12-05 08:43:34 -05002940char *__d_path(const struct path *path,
2941 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002942 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002944 char *res = buf + buflen;
2945 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002947 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002948 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002949
Al Viro02125a82011-12-05 08:43:34 -05002950 if (error < 0)
2951 return ERR_PTR(error);
2952 if (error > 0)
2953 return NULL;
2954 return res;
2955}
2956
2957char *d_absolute_path(const struct path *path,
2958 char *buf, int buflen)
2959{
2960 struct path root = {};
2961 char *res = buf + buflen;
2962 int error;
2963
2964 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002965 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05002966
2967 if (error > 1)
2968 error = -EINVAL;
2969 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002970 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002971 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972}
2973
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002974/*
2975 * same as __d_path but appends "(deleted)" for unlinked files.
2976 */
Al Viro02125a82011-12-05 08:43:34 -05002977static int path_with_deleted(const struct path *path,
2978 const struct path *root,
2979 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002980{
2981 prepend(buf, buflen, "\0", 1);
2982 if (d_unlinked(path->dentry)) {
2983 int error = prepend(buf, buflen, " (deleted)", 10);
2984 if (error)
2985 return error;
2986 }
2987
2988 return prepend_path(path, root, buf, buflen);
2989}
2990
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002991static int prepend_unreachable(char **buffer, int *buflen)
2992{
2993 return prepend(buffer, buflen, "(unreachable)", 13);
2994}
2995
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07002996static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
2997{
2998 unsigned seq;
2999
3000 do {
3001 seq = read_seqcount_begin(&fs->seq);
3002 *root = fs->root;
3003 } while (read_seqcount_retry(&fs->seq, seq));
3004}
3005
Jan Bluncka03a8a702008-02-14 19:38:32 -08003006/**
3007 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003008 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003009 * @buf: buffer to return value in
3010 * @buflen: buffer length
3011 *
3012 * Convert a dentry into an ASCII path name. If the entry has been deleted
3013 * the string " (deleted)" is appended. Note that this is ambiguous.
3014 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003015 * Returns a pointer into the buffer or an error code if the path was
3016 * too long. Note: Callers should use the returned pointer, not the passed
3017 * in buffer, to use the name! The implementation often starts at an offset
3018 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003019 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003020 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003021 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003022char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003024 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003025 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003026 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003028 /*
3029 * We have various synthetic filesystems that never get mounted. On
3030 * these filesystems dentries are never used for lookup purposes, and
3031 * thus don't need to be hashed. They also don't need a name until a
3032 * user wants to identify the object in /proc/pid/fd/. The little hack
3033 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003034 *
3035 * Some pseudo inodes are mountable. When they are mounted
3036 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3037 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003038 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003039 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3040 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003041 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003042
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003043 rcu_read_lock();
3044 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003045 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003046 rcu_read_unlock();
3047
Al Viro02125a82011-12-05 08:43:34 -05003048 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003049 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050 return res;
3051}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003052EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053
3054/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003055 * Helper function for dentry_operations.d_dname() members
3056 */
3057char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3058 const char *fmt, ...)
3059{
3060 va_list args;
3061 char temp[64];
3062 int sz;
3063
3064 va_start(args, fmt);
3065 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3066 va_end(args);
3067
3068 if (sz > sizeof(temp) || sz > buflen)
3069 return ERR_PTR(-ENAMETOOLONG);
3070
3071 buffer += buflen - sz;
3072 return memcpy(buffer, temp, sz);
3073}
3074
Al Viro118b2302013-08-24 12:08:17 -04003075char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3076{
3077 char *end = buffer + buflen;
3078 /* these dentries are never renamed, so d_lock is not needed */
3079 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003080 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003081 prepend(&end, &buflen, "/", 1))
3082 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003083 return end;
Al Viro118b2302013-08-24 12:08:17 -04003084}
David Herrmann31bbe162014-01-03 14:09:47 +01003085EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003086
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003087/*
Ram Pai6092d042008-03-27 13:06:20 +01003088 * Write full pathname from the root of the filesystem into the buffer.
3089 */
Al Virof6500802014-01-26 12:37:55 -05003090static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003091{
Al Virof6500802014-01-26 12:37:55 -05003092 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003093 char *end, *retval;
3094 int len, seq = 0;
3095 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003096
Al Virof6500802014-01-26 12:37:55 -05003097 if (buflen < 2)
3098 goto Elong;
3099
Al Viro48f5ec22013-09-09 15:22:25 -04003100 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003101restart:
Al Virof6500802014-01-26 12:37:55 -05003102 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003103 end = buf + buflen;
3104 len = buflen;
3105 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003106 /* Get '/' right */
3107 retval = end-1;
3108 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003109 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003110 while (!IS_ROOT(dentry)) {
3111 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003112
Ram Pai6092d042008-03-27 13:06:20 +01003113 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003114 error = prepend_name(&end, &len, &dentry->d_name);
3115 if (error)
3116 break;
Ram Pai6092d042008-03-27 13:06:20 +01003117
3118 retval = end;
3119 dentry = parent;
3120 }
Al Viro48f5ec22013-09-09 15:22:25 -04003121 if (!(seq & 1))
3122 rcu_read_unlock();
3123 if (need_seqretry(&rename_lock, seq)) {
3124 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003125 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003126 }
3127 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003128 if (error)
3129 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003130 return retval;
3131Elong:
3132 return ERR_PTR(-ENAMETOOLONG);
3133}
Nick Pigginec2447c2011-01-07 17:49:29 +11003134
3135char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3136{
Waiman Long232d2d62013-09-09 12:18:13 -04003137 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003138}
3139EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003140
3141char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3142{
3143 char *p = NULL;
3144 char *retval;
3145
Al Viroc1031352010-06-06 22:31:14 -04003146 if (d_unlinked(dentry)) {
3147 p = buf + buflen;
3148 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3149 goto Elong;
3150 buflen++;
3151 }
3152 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003153 if (!IS_ERR(retval) && p)
3154 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003155 return retval;
3156Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003157 return ERR_PTR(-ENAMETOOLONG);
3158}
3159
Linus Torvalds8b19e342013-09-12 10:35:47 -07003160static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3161 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003162{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003163 unsigned seq;
3164
3165 do {
3166 seq = read_seqcount_begin(&fs->seq);
3167 *root = fs->root;
3168 *pwd = fs->pwd;
3169 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003170}
3171
Ram Pai6092d042008-03-27 13:06:20 +01003172/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173 * NOTE! The user-level library version returns a
3174 * character pointer. The kernel system call just
3175 * returns the length of the buffer filled (which
3176 * includes the ending '\0' character), or a negative
3177 * error value. So libc would do something like
3178 *
3179 * char *getcwd(char * buf, size_t size)
3180 * {
3181 * int retval;
3182 *
3183 * retval = sys_getcwd(buf, size);
3184 * if (retval >= 0)
3185 * return buf;
3186 * errno = -retval;
3187 * return NULL;
3188 * }
3189 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003190SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191{
Linus Torvalds552ce542007-02-13 12:08:18 -08003192 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003193 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003194 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195
3196 if (!page)
3197 return -ENOMEM;
3198
Linus Torvalds8b19e342013-09-12 10:35:47 -07003199 rcu_read_lock();
3200 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201
Linus Torvalds552ce542007-02-13 12:08:18 -08003202 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003203 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003204 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003205 char *cwd = page + PATH_MAX;
3206 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003208 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003209 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003210 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003211
Al Viro02125a82011-12-05 08:43:34 -05003212 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003213 goto out;
3214
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003215 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003216 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003217 error = prepend_unreachable(&cwd, &buflen);
3218 if (error)
3219 goto out;
3220 }
3221
Linus Torvalds552ce542007-02-13 12:08:18 -08003222 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003223 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003224 if (len <= size) {
3225 error = len;
3226 if (copy_to_user(buf, cwd, len))
3227 error = -EFAULT;
3228 }
Nick Piggin949854d2011-01-07 17:49:37 +11003229 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003230 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003231 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232
3233out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003234 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003235 return error;
3236}
3237
3238/*
3239 * Test whether new_dentry is a subdirectory of old_dentry.
3240 *
3241 * Trivially implemented using the dcache structure
3242 */
3243
3244/**
3245 * is_subdir - is new dentry a subdirectory of old_dentry
3246 * @new_dentry: new dentry
3247 * @old_dentry: old dentry
3248 *
3249 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3250 * Returns 0 otherwise.
3251 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3252 */
3253
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003254int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255{
3256 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11003257 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003259 if (new_dentry == old_dentry)
3260 return 1;
3261
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003262 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003265 /*
3266 * Need rcu_readlock to protect against the d_parent trashing
3267 * due to d_move
3268 */
3269 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003270 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003272 else
3273 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11003274 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276
3277 return result;
3278}
3279
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003280static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003282 struct dentry *root = data;
3283 if (dentry != root) {
3284 if (d_unhashed(dentry) || !dentry->d_inode)
3285 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003287 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3288 dentry->d_flags |= DCACHE_GENOCIDE;
3289 dentry->d_lockref.count--;
3290 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003292 return D_WALK_CONTINUE;
3293}
Nick Piggin58db63d2011-01-07 17:49:39 +11003294
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003295void d_genocide(struct dentry *parent)
3296{
3297 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298}
3299
Al Viro60545d02013-06-07 01:20:27 -04003300void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301{
Al Viro60545d02013-06-07 01:20:27 -04003302 inode_dec_link_count(inode);
3303 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003304 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003305 !d_unlinked(dentry));
3306 spin_lock(&dentry->d_parent->d_lock);
3307 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3308 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3309 (unsigned long long)inode->i_ino);
3310 spin_unlock(&dentry->d_lock);
3311 spin_unlock(&dentry->d_parent->d_lock);
3312 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313}
Al Viro60545d02013-06-07 01:20:27 -04003314EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315
3316static __initdata unsigned long dhash_entries;
3317static int __init set_dhash_entries(char *str)
3318{
3319 if (!str)
3320 return 0;
3321 dhash_entries = simple_strtoul(str, &str, 0);
3322 return 1;
3323}
3324__setup("dhash_entries=", set_dhash_entries);
3325
3326static void __init dcache_init_early(void)
3327{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003328 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329
3330 /* If hashes are distributed across NUMA nodes, defer
3331 * hash allocation until vmalloc space is available.
3332 */
3333 if (hashdist)
3334 return;
3335
3336 dentry_hashtable =
3337 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003338 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339 dhash_entries,
3340 13,
3341 HASH_EARLY,
3342 &d_hash_shift,
3343 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003344 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345 0);
3346
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003347 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003348 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349}
3350
Denis Cheng74bf17c2007-10-16 23:26:30 -07003351static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003353 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354
3355 /*
3356 * A constructor could be added for stable state like the lists,
3357 * but it is probably not worth it because of the cache nature
3358 * of the dcache.
3359 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003360 dentry_cache = KMEM_CACHE(dentry,
3361 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362
3363 /* Hash may have been set up in dcache_init_early */
3364 if (!hashdist)
3365 return;
3366
3367 dentry_hashtable =
3368 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003369 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370 dhash_entries,
3371 13,
3372 0,
3373 &d_hash_shift,
3374 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003375 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376 0);
3377
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003378 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003379 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380}
3381
3382/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003383struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003384EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386EXPORT_SYMBOL(d_genocide);
3387
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388void __init vfs_caches_init_early(void)
3389{
3390 dcache_init_early();
3391 inode_init_early();
3392}
3393
3394void __init vfs_caches_init(unsigned long mempages)
3395{
3396 unsigned long reserve;
3397
3398 /* Base hash sizes on available memory, with a reserve equal to
3399 150% of current kernel size */
3400
3401 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3402 mempages -= reserve;
3403
3404 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003405 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406
Denis Cheng74bf17c2007-10-16 23:26:30 -07003407 dcache_init();
3408 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003410 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411 bdev_cache_init();
3412 chrdev_init();
3413}