blob: 3ffef7f4e5cdd9d00ca454130070f4619a8707d6 [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>
David Howells07f3f052006-09-30 20:52:18 +020041#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050042#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Nick Piggin789680d2011-01-07 17:49:30 +110044/*
45 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110046 * dcache->d_inode->i_lock protects:
47 * - i_dentry, d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110048 * dcache_hash_bucket lock protects:
49 * - the dcache hash table
50 * s_anon bl list spinlock protects:
51 * - the s_anon list (see __d_drop)
Dave Chinner19156842013-08-28 10:17:55 +100052 * dentry->d_sb->s_dentry_lru_lock protects:
Nick Piggin23044502011-01-07 17:49:31 +110053 * - the dcache lru lists and counters
54 * d_lock protects:
55 * - d_flags
56 * - d_name
57 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110058 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110059 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110060 * - d_parent and d_subdirs
61 * - childrens' d_child and d_parent
Nick Pigginb23fb0a2011-01-07 17:49:35 +110062 * - d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110063 *
64 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110065 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110066 * dentry->d_lock
Dave Chinner19156842013-08-28 10:17:55 +100067 * dentry->d_sb->s_dentry_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110068 * dcache_hash_bucket lock
69 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110070 *
Nick Pigginda502952011-01-07 17:49:33 +110071 * If there is an ancestor relationship:
72 * dentry->d_parent->...->d_parent->d_lock
73 * ...
74 * dentry->d_parent->d_lock
75 * dentry->d_lock
76 *
77 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110078 * if (dentry1 < dentry2)
79 * dentry1->d_lock
80 * dentry2->d_lock
81 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080082int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
84
Al Viro74c3cbe2007-07-22 08:04:18 -040085__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Nick Piggin949854d2011-01-07 17:49:37 +110087EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Christoph Lametere18b8902006-12-06 20:33:20 -080089static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/*
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
95 *
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
98 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800100static unsigned int d_hash_mask __read_mostly;
101static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100102
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700103static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvalds8966be92012-03-02 14:23:30 -0800105static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700106 unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100107{
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700108 hash += (unsigned long) parent / L1_CACHE_BYTES;
Linus Torvalds99d263d2014-09-13 11:30:10 -0700109 return dentry_hashtable + hash_32(hash, d_hash_shift);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100110}
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112/* Statistics gathering. */
113struct dentry_stat_t dentry_stat = {
114 .age_limit = 45,
115};
116
Glauber Costa3942c072013-08-28 10:17:53 +1000117static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000118static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400119
120#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000121
122/*
123 * Here we resort to our own counters instead of using generic per-cpu counters
124 * for consistency with what the vfs inode code does. We are expected to harvest
125 * better code and performance by having our own specialized counters.
126 *
127 * Please note that the loop is done over all possible CPUs, not over all online
128 * CPUs. The reason for this is that we don't want to play games with CPUs going
129 * on and off. If one of them goes off, we will just keep their counters.
130 *
131 * glommer: See cffbc8a for details, and if you ever intend to change this,
132 * please update all vfs counters to match.
133 */
Glauber Costa3942c072013-08-28 10:17:53 +1000134static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100135{
136 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000137 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100138 for_each_possible_cpu(i)
139 sum += per_cpu(nr_dentry, i);
140 return sum < 0 ? 0 : sum;
141}
142
Dave Chinner62d36c72013-08-28 10:17:54 +1000143static long get_nr_dentry_unused(void)
144{
145 int i;
146 long sum = 0;
147 for_each_possible_cpu(i)
148 sum += per_cpu(nr_dentry_unused, i);
149 return sum < 0 ? 0 : sum;
150}
151
Joe Perches1f7e0612014-06-06 14:38:05 -0700152int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400153 size_t *lenp, loff_t *ppos)
154{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100155 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000156 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000157 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400158}
159#endif
160
Linus Torvalds5483f182012-03-04 15:51:42 -0800161/*
162 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
163 * The strings are both count bytes long, and count is non-zero.
164 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700165#ifdef CONFIG_DCACHE_WORD_ACCESS
166
167#include <asm/word-at-a-time.h>
168/*
169 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
170 * aligned allocation for this particular component. We don't
171 * strictly need the load_unaligned_zeropad() safety, but it
172 * doesn't hurt either.
173 *
174 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
175 * need the careful unaligned handling.
176 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700177static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800178{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800179 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800180
181 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700182 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700183 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800184 if (tcount < sizeof(unsigned long))
185 break;
186 if (unlikely(a != b))
187 return 1;
188 cs += sizeof(unsigned long);
189 ct += sizeof(unsigned long);
190 tcount -= sizeof(unsigned long);
191 if (!tcount)
192 return 0;
193 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000194 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800195 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700196}
197
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800198#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700199
Linus Torvalds94753db52012-05-10 12:19:19 -0700200static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700201{
Linus Torvalds5483f182012-03-04 15:51:42 -0800202 do {
203 if (*cs != *ct)
204 return 1;
205 cs++;
206 ct++;
207 tcount--;
208 } while (tcount);
209 return 0;
210}
211
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700212#endif
213
Linus Torvalds94753db52012-05-10 12:19:19 -0700214static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
215{
Linus Torvalds6326c712012-05-21 16:14:04 -0700216 const unsigned char *cs;
Linus Torvalds94753db52012-05-10 12:19:19 -0700217 /*
218 * Be careful about RCU walk racing with rename:
219 * use ACCESS_ONCE to fetch the name pointer.
220 *
221 * NOTE! Even if a rename will mean that the length
222 * was not loaded atomically, we don't care. The
223 * RCU walk will check the sequence count eventually,
224 * and catch it. And we won't overrun the buffer,
225 * because we're reading the name pointer atomically,
226 * and a dentry name is guaranteed to be properly
227 * terminated with a NUL byte.
228 *
229 * End result: even if 'len' is wrong, we'll exit
230 * early because the data cannot match (there can
231 * be no NUL in the ct/tcount data)
232 */
Linus Torvalds6326c712012-05-21 16:14:04 -0700233 cs = ACCESS_ONCE(dentry->d_name.name);
234 smp_read_barrier_depends();
235 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700236}
237
Al Viro8d85b482014-09-29 14:54:27 -0400238struct external_name {
239 union {
240 atomic_t count;
241 struct rcu_head head;
242 } u;
243 unsigned char name[];
244};
245
246static inline struct external_name *external_name(struct dentry *dentry)
247{
248 return container_of(dentry->d_name.name, struct external_name, name[0]);
249}
250
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400251static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400253 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
254
Al Virob3d9b7a2012-06-09 13:51:19 -0400255 WARN_ON(!hlist_unhashed(&dentry->d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400256 kmem_cache_free(dentry_cache, dentry);
257}
258
259static void __d_free_external(struct rcu_head *head)
260{
261 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
262 WARN_ON(!hlist_unhashed(&dentry->d_alias));
263 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 Viro8d85b482014-09-29 14:54:27 -0400274 if (unlikely(dname_external(dentry))) {
275 struct external_name *p = external_name(dentry);
276 if (likely(atomic_dec_and_test(&p->u.count))) {
277 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
278 return;
279 }
280 }
Al Virob4f03542014-04-29 23:40:14 -0400281 /* if dentry was never visible to RCU, immediate free is OK */
282 if (!(dentry->d_flags & DCACHE_RCUACCESS))
283 __d_free(&dentry->d_u.d_rcu);
284 else
285 call_rcu(&dentry->d_u.d_rcu, __d_free);
286}
287
Nick Piggin31e6b012011-01-07 17:49:52 +1100288/**
289 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800290 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100291 * After this call, in-progress rcu-walk path lookup will fail. This
292 * should be called after unhashing, and after changing d_inode (if
293 * the dentry has not already been unhashed).
294 */
295static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
296{
297 assert_spin_locked(&dentry->d_lock);
298 /* Go through a barrier */
299 write_seqcount_barrier(&dentry->d_seq);
300}
301
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302/*
303 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100304 * d_iput() operation if defined. Dentry has no refcount
305 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800307static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200308 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100309 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310{
311 struct inode *inode = dentry->d_inode;
312 if (inode) {
313 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400314 hlist_del_init(&dentry->d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100316 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700317 if (!inode->i_nlink)
318 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 if (dentry->d_op && dentry->d_op->d_iput)
320 dentry->d_op->d_iput(dentry, inode);
321 else
322 iput(inode);
323 } else {
324 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 }
326}
327
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700328/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100329 * Release the dentry's inode, using the filesystem
330 * d_iput() operation if defined. dentry remains in-use.
331 */
332static void dentry_unlink_inode(struct dentry * dentry)
333 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100334 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100335{
336 struct inode *inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +0100337 __d_clear_type(dentry);
Nick Piggin31e6b012011-01-07 17:49:52 +1100338 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400339 hlist_del_init(&dentry->d_alias);
Nick Piggin31e6b012011-01-07 17:49:52 +1100340 dentry_rcuwalk_barrier(dentry);
341 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100342 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100343 if (!inode->i_nlink)
344 fsnotify_inoderemove(inode);
345 if (dentry->d_op && dentry->d_op->d_iput)
346 dentry->d_op->d_iput(dentry, inode);
347 else
348 iput(inode);
349}
350
351/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400352 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
353 * is in use - which includes both the "real" per-superblock
354 * LRU list _and_ the DCACHE_SHRINK_LIST use.
355 *
356 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
357 * on the shrink list (ie not on the superblock LRU list).
358 *
359 * The per-cpu "nr_dentry_unused" counters are updated with
360 * the DCACHE_LRU_LIST bit.
361 *
362 * These helper functions make sure we always follow the
363 * rules. d_lock must be held by the caller.
364 */
365#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
366static void d_lru_add(struct dentry *dentry)
367{
368 D_FLAG_VERIFY(dentry, 0);
369 dentry->d_flags |= DCACHE_LRU_LIST;
370 this_cpu_inc(nr_dentry_unused);
371 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
372}
373
374static void d_lru_del(struct dentry *dentry)
375{
376 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
377 dentry->d_flags &= ~DCACHE_LRU_LIST;
378 this_cpu_dec(nr_dentry_unused);
379 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
380}
381
382static void d_shrink_del(struct dentry *dentry)
383{
384 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
385 list_del_init(&dentry->d_lru);
386 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
387 this_cpu_dec(nr_dentry_unused);
388}
389
390static void d_shrink_add(struct dentry *dentry, struct list_head *list)
391{
392 D_FLAG_VERIFY(dentry, 0);
393 list_add(&dentry->d_lru, list);
394 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
395 this_cpu_inc(nr_dentry_unused);
396}
397
398/*
399 * These can only be called under the global LRU lock, ie during the
400 * callback for freeing the LRU list. "isolate" removes it from the
401 * LRU lists entirely, while shrink_move moves it to the indicated
402 * private list.
403 */
404static void d_lru_isolate(struct dentry *dentry)
405{
406 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
407 dentry->d_flags &= ~DCACHE_LRU_LIST;
408 this_cpu_dec(nr_dentry_unused);
409 list_del_init(&dentry->d_lru);
410}
411
412static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
413{
414 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
415 dentry->d_flags |= DCACHE_SHRINK_LIST;
416 list_move_tail(&dentry->d_lru, list);
417}
418
419/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000420 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700421 */
422static void dentry_lru_add(struct dentry *dentry)
423{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400424 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
425 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700426}
427
Miklos Szeredid52b9082007-05-08 00:23:46 -0700428/**
Nick Piggin789680d2011-01-07 17:49:30 +1100429 * d_drop - drop a dentry
430 * @dentry: dentry to drop
431 *
432 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
433 * be found through a VFS lookup any more. Note that this is different from
434 * deleting the dentry - d_delete will try to mark the dentry negative if
435 * possible, giving a successful _negative_ lookup, while d_drop will
436 * just make the cache lookup fail.
437 *
438 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
439 * reason (NFS timeouts or autofs deletes).
440 *
441 * __d_drop requires dentry->d_lock.
442 */
443void __d_drop(struct dentry *dentry)
444{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700445 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400446 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400447 /*
448 * Hashed dentries are normally on the dentry hashtable,
449 * with the exception of those newly allocated by
450 * d_obtain_alias, which are always IS_ROOT:
451 */
452 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400453 b = &dentry->d_sb->s_anon;
454 else
455 b = d_hash(dentry->d_parent, dentry->d_name.hash);
456
457 hlist_bl_lock(b);
458 __hlist_bl_del(&dentry->d_hash);
459 dentry->d_hash.pprev = NULL;
460 hlist_bl_unlock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700461 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100462 }
463}
464EXPORT_SYMBOL(__d_drop);
465
466void d_drop(struct dentry *dentry)
467{
Nick Piggin789680d2011-01-07 17:49:30 +1100468 spin_lock(&dentry->d_lock);
469 __d_drop(dentry);
470 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100471}
472EXPORT_SYMBOL(d_drop);
473
Al Viroe55fd012014-05-28 13:51:12 -0400474static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100475{
Al Viro41edf272014-05-01 10:30:00 -0400476 struct dentry *parent = NULL;
477 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400478 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100479 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100480
Linus Torvalds0d984392013-09-08 13:46:52 -0700481 /*
482 * The dentry is now unrecoverably dead to the world.
483 */
484 lockref_mark_dead(&dentry->d_lockref);
485
Sage Weilf0023bc2011-10-28 10:02:42 -0700486 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700487 * inform the fs via d_prune that this dentry is about to be
488 * unhashed and destroyed.
489 */
Al Viro29266202014-05-30 11:39:02 -0400490 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800491 dentry->d_op->d_prune(dentry);
492
Al Viro01b60352014-04-29 23:42:52 -0400493 if (dentry->d_flags & DCACHE_LRU_LIST) {
494 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
495 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400496 }
Nick Piggin77812a12011-01-07 17:49:48 +1100497 /* if it was on the hash then remove it */
498 __d_drop(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400499 list_del(&dentry->d_u.d_child);
500 /*
501 * Inform d_walk() that we are no longer attached to the
502 * dentry tree
503 */
504 dentry->d_flags |= DCACHE_DENTRY_KILLED;
505 if (parent)
506 spin_unlock(&parent->d_lock);
507 dentry_iput(dentry);
508 /*
509 * dentry_iput drops the locks, at which point nobody (except
510 * transient RCU lookups) can reach this dentry.
511 */
512 BUG_ON((int)dentry->d_lockref.count > 0);
513 this_cpu_dec(nr_dentry);
514 if (dentry->d_op && dentry->d_op->d_release)
515 dentry->d_op->d_release(dentry);
516
Al Viro41edf272014-05-01 10:30:00 -0400517 spin_lock(&dentry->d_lock);
518 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
519 dentry->d_flags |= DCACHE_MAY_FREE;
520 can_free = false;
521 }
522 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400523 if (likely(can_free))
524 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400525}
526
527/*
528 * Finish off a dentry we've decided to kill.
529 * dentry->d_lock must be held, returns with it unlocked.
530 * If ref is non-zero, then decrement the refcount too.
531 * Returns dentry requiring refcount drop, or NULL if we're done.
532 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400533static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400534 __releases(dentry->d_lock)
535{
536 struct inode *inode = dentry->d_inode;
537 struct dentry *parent = NULL;
538
539 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
540 goto failed;
541
542 if (!IS_ROOT(dentry)) {
543 parent = dentry->d_parent;
544 if (unlikely(!spin_trylock(&parent->d_lock))) {
545 if (inode)
546 spin_unlock(&inode->i_lock);
547 goto failed;
548 }
549 }
550
551 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400552 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400553
554failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400555 spin_unlock(&dentry->d_lock);
556 cpu_relax();
Al Viroe55fd012014-05-28 13:51:12 -0400557 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100558}
559
Al Viro046b9612014-05-29 08:54:52 -0400560static inline struct dentry *lock_parent(struct dentry *dentry)
561{
562 struct dentry *parent = dentry->d_parent;
563 if (IS_ROOT(dentry))
564 return NULL;
Al Viroc2338f22014-06-12 00:29:13 -0400565 if (unlikely((int)dentry->d_lockref.count < 0))
566 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400567 if (likely(spin_trylock(&parent->d_lock)))
568 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400569 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400570 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400571again:
572 parent = ACCESS_ONCE(dentry->d_parent);
573 spin_lock(&parent->d_lock);
574 /*
575 * We can't blindly lock dentry until we are sure
576 * that we won't violate the locking order.
577 * Any changes of dentry->d_parent must have
578 * been done with parent->d_lock held, so
579 * spin_lock() above is enough of a barrier
580 * for checking if it's still our child.
581 */
582 if (unlikely(parent != dentry->d_parent)) {
583 spin_unlock(&parent->d_lock);
584 goto again;
585 }
586 rcu_read_unlock();
587 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700588 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400589 else
590 parent = NULL;
591 return parent;
592}
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594/*
595 * This is dput
596 *
597 * This is complicated by the fact that we do not want to put
598 * dentries that are no longer on any hash chain on the unused
599 * list: we'd much rather just get rid of them immediately.
600 *
601 * However, that implies that we have to traverse the dentry
602 * tree upwards to the parents which might _also_ now be
603 * scheduled for deletion (it may have been only waiting for
604 * its last child to go away).
605 *
606 * This tail recursion is done by hand as we don't want to depend
607 * on the compiler to always get this right (gcc generally doesn't).
608 * Real recursion would eat up our stack space.
609 */
610
611/*
612 * dput - release a dentry
613 * @dentry: dentry to release
614 *
615 * Release a dentry. This will drop the usage count and if appropriate
616 * call the dentry unlink method as well as removing it from the queues and
617 * releasing its resources. If the parent dentries were scheduled for release
618 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620void dput(struct dentry *dentry)
621{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700622 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 return;
624
625repeat:
Waiman Long98474232013-08-28 18:24:59 -0700626 if (lockref_put_or_lock(&dentry->d_lockref))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700629 /* Unreachable? Get rid of it */
630 if (unlikely(d_unhashed(dentry)))
631 goto kill_it;
632
633 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100635 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 }
Nick Piggin265ac902010-10-10 05:36:24 -0400637
Linus Torvalds358eec12013-10-31 15:43:02 -0700638 if (!(dentry->d_flags & DCACHE_REFERENCED))
639 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400640 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400641
Waiman Long98474232013-08-28 18:24:59 -0700642 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100643 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 return;
645
Miklos Szeredid52b9082007-05-08 00:23:46 -0700646kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400647 dentry = dentry_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700648 if (dentry)
649 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700651EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100654/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100655static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656{
Waiman Long98474232013-08-28 18:24:59 -0700657 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658}
659
Nick Piggindc0474b2011-01-07 17:49:43 +1100660static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100661{
Waiman Long98474232013-08-28 18:24:59 -0700662 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100663}
664
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100665struct dentry *dget_parent(struct dentry *dentry)
666{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700667 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100668 struct dentry *ret;
669
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700670 /*
671 * Do optimistic parent lookup without any
672 * locking.
673 */
674 rcu_read_lock();
675 ret = ACCESS_ONCE(dentry->d_parent);
676 gotref = lockref_get_not_zero(&ret->d_lockref);
677 rcu_read_unlock();
678 if (likely(gotref)) {
679 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
680 return ret;
681 dput(ret);
682 }
683
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100684repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100685 /*
686 * Don't need rcu_dereference because we re-check it was correct under
687 * the lock.
688 */
689 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100690 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100691 spin_lock(&ret->d_lock);
692 if (unlikely(ret != dentry->d_parent)) {
693 spin_unlock(&ret->d_lock);
694 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100695 goto repeat;
696 }
Nick Piggina734eb42011-01-07 17:49:44 +1100697 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700698 BUG_ON(!ret->d_lockref.count);
699 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100700 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100701 return ret;
702}
703EXPORT_SYMBOL(dget_parent);
704
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705/**
706 * d_find_alias - grab a hashed alias of inode
707 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 *
709 * If inode has a hashed alias, or is a directory and has any alias,
710 * acquire the reference to alias and return it. Otherwise return NULL.
711 * Notice that if inode is a directory there can be only one alias and
712 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800713 * of a filesystem, or if the directory was renamed and d_revalidate
714 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700716 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500717 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500719static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720{
Nick Pigginda502952011-01-07 17:49:33 +1100721 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Nick Pigginda502952011-01-07 17:49:33 +1100723again:
724 discon_alias = NULL;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800725 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100726 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700728 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100729 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500731 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100732 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100733 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 return alias;
735 }
736 }
Nick Pigginda502952011-01-07 17:49:33 +1100737 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 }
Nick Pigginda502952011-01-07 17:49:33 +1100739 if (discon_alias) {
740 alias = discon_alias;
741 spin_lock(&alias->d_lock);
742 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500743 __dget_dlock(alias);
744 spin_unlock(&alias->d_lock);
745 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100746 }
747 spin_unlock(&alias->d_lock);
748 goto again;
749 }
750 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751}
752
Nick Pigginda502952011-01-07 17:49:33 +1100753struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754{
David Howells214fda12006-03-25 03:06:36 -0800755 struct dentry *de = NULL;
756
Al Virob3d9b7a2012-06-09 13:51:19 -0400757 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100758 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500759 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100760 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800761 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 return de;
763}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700764EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
766/*
767 * Try to kill dentries associated with this inode.
768 * WARNING: you must own a reference to inode.
769 */
770void d_prune_aliases(struct inode *inode)
771{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700772 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100774 spin_lock(&inode->i_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800775 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700777 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400778 struct dentry *parent = lock_parent(dentry);
779 if (likely(!dentry->d_lockref.count)) {
780 __dentry_kill(dentry);
781 goto restart;
782 }
783 if (parent)
784 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 }
786 spin_unlock(&dentry->d_lock);
787 }
Nick Piggin873feea2011-01-07 17:50:06 +1100788 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700790EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400792static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
Al Viro5c47e6d2014-04-29 16:13:18 -0400794 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700795
Miklos Szeredi60942f22014-05-02 15:38:39 -0400796 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400797 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400798 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100799 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400800 parent = lock_parent(dentry);
801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000803 * The dispose list is isolated and dentries are not accounted
804 * to the LRU here, so we can simply remove it from the list
805 * here regardless of whether it is referenced or not.
806 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400807 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000808
809 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000811 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 */
Al Viro41edf272014-05-01 10:30:00 -0400813 if ((int)dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700814 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400815 if (parent)
816 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 continue;
818 }
Nick Piggin77812a12011-01-07 17:49:48 +1100819
Al Viro64fd72e2014-05-28 09:48:44 -0400820
821 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
822 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
823 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400824 if (parent)
825 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -0400826 if (can_free)
827 dentry_free(dentry);
828 continue;
829 }
830
Al Viroff2fde92014-05-28 13:59:13 -0400831 inode = dentry->d_inode;
832 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400833 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000834 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400835 if (parent)
836 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400837 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000838 }
Al Viroff2fde92014-05-28 13:59:13 -0400839
Al Viroff2fde92014-05-28 13:59:13 -0400840 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -0400841
Al Viro5c47e6d2014-04-29 16:13:18 -0400842 /*
843 * We need to prune ancestors too. This is necessary to prevent
844 * quadratic behavior of shrink_dcache_parent(), but is also
845 * expected to be beneficial in reducing dentry cache
846 * fragmentation.
847 */
848 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -0400849 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
850 parent = lock_parent(dentry);
851 if (dentry->d_lockref.count != 1) {
852 dentry->d_lockref.count--;
853 spin_unlock(&dentry->d_lock);
854 if (parent)
855 spin_unlock(&parent->d_lock);
856 break;
857 }
858 inode = dentry->d_inode; /* can't be NULL */
859 if (unlikely(!spin_trylock(&inode->i_lock))) {
860 spin_unlock(&dentry->d_lock);
861 if (parent)
862 spin_unlock(&parent->d_lock);
863 cpu_relax();
864 continue;
865 }
866 __dentry_kill(dentry);
867 dentry = parent;
868 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400870}
871
Dave Chinnerf6041562013-08-28 10:18:00 +1000872static enum lru_status
873dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
874{
875 struct list_head *freeable = arg;
876 struct dentry *dentry = container_of(item, struct dentry, d_lru);
877
878
879 /*
880 * we are inverting the lru lock/dentry->d_lock here,
881 * so use a trylock. If we fail to get the lock, just skip
882 * it
883 */
884 if (!spin_trylock(&dentry->d_lock))
885 return LRU_SKIP;
886
887 /*
888 * Referenced dentries are still in use. If they have active
889 * counts, just remove them from the LRU. Otherwise give them
890 * another pass through the LRU.
891 */
892 if (dentry->d_lockref.count) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400893 d_lru_isolate(dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +1000894 spin_unlock(&dentry->d_lock);
895 return LRU_REMOVED;
896 }
897
898 if (dentry->d_flags & DCACHE_REFERENCED) {
899 dentry->d_flags &= ~DCACHE_REFERENCED;
900 spin_unlock(&dentry->d_lock);
901
902 /*
903 * The list move itself will be made by the common LRU code. At
904 * this point, we've dropped the dentry->d_lock but keep the
905 * lru lock. This is safe to do, since every list movement is
906 * protected by the lru lock even if both locks are held.
907 *
908 * This is guaranteed by the fact that all LRU management
909 * functions are intermediated by the LRU API calls like
910 * list_lru_add and list_lru_del. List movement in this file
911 * only ever occur through this functions or through callbacks
912 * like this one, that are called from the LRU API.
913 *
914 * The only exceptions to this are functions like
915 * shrink_dentry_list, and code that first checks for the
916 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
917 * operating only with stack provided lists after they are
918 * properly isolated from the main list. It is thus, always a
919 * local access.
920 */
921 return LRU_ROTATE;
922 }
923
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400924 d_lru_shrink_move(dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +1000925 spin_unlock(&dentry->d_lock);
926
927 return LRU_REMOVED;
928}
929
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400930/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000931 * prune_dcache_sb - shrink the dcache
932 * @sb: superblock
Dave Chinnerf6041562013-08-28 10:18:00 +1000933 * @nr_to_scan : number of entries to try to free
Dave Chinner9b17c622013-08-28 10:18:05 +1000934 * @nid: which node to scan for freeable entities
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400935 *
Dave Chinnerf6041562013-08-28 10:18:00 +1000936 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000937 * done when we need more memory an called from the superblock shrinker
938 * function.
939 *
940 * This function may fail to free any resources if all the dentries are in
941 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400942 */
Dave Chinner9b17c622013-08-28 10:18:05 +1000943long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
944 int nid)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400945{
Dave Chinnerf6041562013-08-28 10:18:00 +1000946 LIST_HEAD(dispose);
947 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400948
Dave Chinner9b17c622013-08-28 10:18:05 +1000949 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
950 &dispose, &nr_to_scan);
Dave Chinnerf6041562013-08-28 10:18:00 +1000951 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +1000952 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953}
954
Glauber Costa4e717f52013-08-28 10:18:03 +1000955static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
956 spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000957{
Glauber Costa4e717f52013-08-28 10:18:03 +1000958 struct list_head *freeable = arg;
959 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000960
Glauber Costa4e717f52013-08-28 10:18:03 +1000961 /*
962 * we are inverting the lru lock/dentry->d_lock here,
963 * so use a trylock. If we fail to get the lock, just skip
964 * it
965 */
966 if (!spin_trylock(&dentry->d_lock))
967 return LRU_SKIP;
968
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400969 d_lru_shrink_move(dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +1000970 spin_unlock(&dentry->d_lock);
971
972 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000973}
974
Glauber Costa4e717f52013-08-28 10:18:03 +1000975
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700976/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 * shrink_dcache_sb - shrink dcache for a superblock
978 * @sb: superblock
979 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400980 * Shrink the dcache for the specified super block. This is used to free
981 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400983void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984{
Glauber Costa4e717f52013-08-28 10:18:03 +1000985 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400986
Glauber Costa4e717f52013-08-28 10:18:03 +1000987 do {
988 LIST_HEAD(dispose);
989
990 freed = list_lru_walk(&sb->s_dentry_lru,
991 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
992
993 this_cpu_sub(nr_dentry_unused, freed);
994 shrink_dentry_list(&dispose);
995 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700997EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001000 * enum d_walk_ret - action to talke during tree walk
1001 * @D_WALK_CONTINUE: contrinue walk
1002 * @D_WALK_QUIT: quit walk
1003 * @D_WALK_NORETRY: quit when retry is needed
1004 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001006enum d_walk_ret {
1007 D_WALK_CONTINUE,
1008 D_WALK_QUIT,
1009 D_WALK_NORETRY,
1010 D_WALK_SKIP,
1011};
1012
1013/**
1014 * d_walk - walk the dentry tree
1015 * @parent: start of walk
1016 * @data: data passed to @enter() and @finish()
1017 * @enter: callback when first entering the dentry
1018 * @finish: callback when successfully finished the walk
1019 *
1020 * The @enter() and @finish() callbacks are called with d_lock held.
1021 */
1022static void d_walk(struct dentry *parent, void *data,
1023 enum d_walk_ret (*enter)(void *, struct dentry *),
1024 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
Nick Piggin949854d2011-01-07 17:49:37 +11001026 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001028 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001029 enum d_walk_ret ret;
1030 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001031
Nick Piggin58db63d2011-01-07 17:49:39 +11001032again:
Al Viro48f5ec22013-09-09 15:22:25 -04001033 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001034 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001035 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001036
1037 ret = enter(data, this_parent);
1038 switch (ret) {
1039 case D_WALK_CONTINUE:
1040 break;
1041 case D_WALK_QUIT:
1042 case D_WALK_SKIP:
1043 goto out_unlock;
1044 case D_WALK_NORETRY:
1045 retry = false;
1046 break;
1047 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048repeat:
1049 next = this_parent->d_subdirs.next;
1050resume:
1051 while (next != &this_parent->d_subdirs) {
1052 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001053 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001055
1056 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001057
1058 ret = enter(data, dentry);
1059 switch (ret) {
1060 case D_WALK_CONTINUE:
1061 break;
1062 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001063 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001064 goto out_unlock;
1065 case D_WALK_NORETRY:
1066 retry = false;
1067 break;
1068 case D_WALK_SKIP:
1069 spin_unlock(&dentry->d_lock);
1070 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001071 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001072
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001074 spin_unlock(&this_parent->d_lock);
1075 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001077 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 goto repeat;
1079 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001080 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 }
1082 /*
1083 * All done at this level ... ascend and resume the search.
1084 */
1085 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001086 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001087 this_parent = child->d_parent;
1088
1089 rcu_read_lock();
1090 spin_unlock(&child->d_lock);
1091 spin_lock(&this_parent->d_lock);
1092
1093 /*
1094 * might go back up the wrong parent if we have had a rename
1095 * or deletion
1096 */
1097 if (this_parent != child->d_parent ||
1098 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1099 need_seqretry(&rename_lock, seq)) {
1100 spin_unlock(&this_parent->d_lock);
1101 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11001102 goto rename_retry;
Al Viro31dec132013-10-25 17:04:27 -04001103 }
1104 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11001105 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 goto resume;
1107 }
Al Viro48f5ec22013-09-09 15:22:25 -04001108 if (need_seqretry(&rename_lock, seq)) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001109 spin_unlock(&this_parent->d_lock);
1110 goto rename_retry;
1111 }
1112 if (finish)
1113 finish(data);
1114
1115out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001116 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001117 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001118 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001119
1120rename_retry:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001121 if (!retry)
1122 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001123 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001124 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001126
1127/*
1128 * Search for at least 1 mount point in the dentry's subdirs.
1129 * We descend to the next level whenever the d_subdirs
1130 * list is non-empty and continue searching.
1131 */
1132
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001133static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1134{
1135 int *ret = data;
1136 if (d_mountpoint(dentry)) {
1137 *ret = 1;
1138 return D_WALK_QUIT;
1139 }
1140 return D_WALK_CONTINUE;
1141}
1142
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001143/**
1144 * have_submounts - check for mounts over a dentry
1145 * @parent: dentry to check.
1146 *
1147 * Return true if the parent or its subdirectories contain
1148 * a mount point
1149 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001150int have_submounts(struct dentry *parent)
1151{
1152 int ret = 0;
1153
1154 d_walk(parent, &ret, check_mount, NULL);
1155
1156 return ret;
1157}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001158EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
1160/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001161 * Called by mount code to set a mountpoint and check if the mountpoint is
1162 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1163 * subtree can become unreachable).
1164 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001165 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001166 * this reason take rename_lock and d_lock on dentry and ancestors.
1167 */
1168int d_set_mounted(struct dentry *dentry)
1169{
1170 struct dentry *p;
1171 int ret = -ENOENT;
1172 write_seqlock(&rename_lock);
1173 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001174 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001175 spin_lock(&p->d_lock);
1176 if (unlikely(d_unhashed(p))) {
1177 spin_unlock(&p->d_lock);
1178 goto out;
1179 }
1180 spin_unlock(&p->d_lock);
1181 }
1182 spin_lock(&dentry->d_lock);
1183 if (!d_unlinked(dentry)) {
1184 dentry->d_flags |= DCACHE_MOUNTED;
1185 ret = 0;
1186 }
1187 spin_unlock(&dentry->d_lock);
1188out:
1189 write_sequnlock(&rename_lock);
1190 return ret;
1191}
1192
1193/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001194 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 * and move any unused dentries to the end of the unused
1196 * list for prune_dcache(). We descend to the next level
1197 * whenever the d_subdirs list is non-empty and continue
1198 * searching.
1199 *
1200 * It returns zero iff there are no unused children,
1201 * otherwise it returns the number of children moved to
1202 * the end of the unused list. This may not be the total
1203 * number of unused children, because select_parent can
1204 * drop the lock and return early due to latency
1205 * constraints.
1206 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001207
1208struct select_data {
1209 struct dentry *start;
1210 struct list_head dispose;
1211 int found;
1212};
1213
1214static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001216 struct select_data *data = _data;
1217 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001219 if (data->start == dentry)
1220 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
Al Virofe915222014-05-03 00:02:25 -04001222 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001223 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001224 } else {
1225 if (dentry->d_flags & DCACHE_LRU_LIST)
1226 d_lru_del(dentry);
1227 if (!dentry->d_lockref.count) {
1228 d_shrink_add(dentry, &data->dispose);
1229 data->found++;
1230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 }
1232 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001233 * We can return to the caller if we have found some (this
1234 * ensures forward progress). We'll be coming back to find
1235 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 */
Al Virofe915222014-05-03 00:02:25 -04001237 if (!list_empty(&data->dispose))
1238 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001240 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241}
1242
1243/**
1244 * shrink_dcache_parent - prune dcache
1245 * @parent: parent of entries to prune
1246 *
1247 * Prune the dcache to remove unused children of the parent dentry.
1248 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001249void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001251 for (;;) {
1252 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001254 INIT_LIST_HEAD(&data.dispose);
1255 data.start = parent;
1256 data.found = 0;
1257
1258 d_walk(parent, &data, select_collect, NULL);
1259 if (!data.found)
1260 break;
1261
1262 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001263 cond_resched();
1264 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001266EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Al Viro9c8c10e2014-05-02 20:36:10 -04001268static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001269{
Al Viro9c8c10e2014-05-02 20:36:10 -04001270 /* it has busy descendents; complain about those instead */
1271 if (!list_empty(&dentry->d_subdirs))
1272 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001273
Al Viro9c8c10e2014-05-02 20:36:10 -04001274 /* root with refcount 1 is fine */
1275 if (dentry == _data && dentry->d_lockref.count == 1)
1276 return D_WALK_CONTINUE;
1277
1278 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1279 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001280 dentry,
1281 dentry->d_inode ?
1282 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001283 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001284 dentry->d_lockref.count,
1285 dentry->d_sb->s_type->name,
1286 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001287 WARN_ON(1);
1288 return D_WALK_CONTINUE;
1289}
1290
1291static void do_one_tree(struct dentry *dentry)
1292{
1293 shrink_dcache_parent(dentry);
1294 d_walk(dentry, dentry, umount_check, NULL);
1295 d_drop(dentry);
1296 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001297}
1298
1299/*
1300 * destroy the dentries attached to a superblock on unmounting
1301 */
1302void shrink_dcache_for_umount(struct super_block *sb)
1303{
1304 struct dentry *dentry;
1305
Al Viro9c8c10e2014-05-02 20:36:10 -04001306 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001307
1308 dentry = sb->s_root;
1309 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001310 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001311
1312 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001313 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1314 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001315 }
1316}
1317
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001318struct detach_data {
1319 struct select_data select;
1320 struct dentry *mountpoint;
1321};
1322static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001323{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001324 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001325
1326 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001327 __dget_dlock(dentry);
1328 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001329 return D_WALK_QUIT;
1330 }
1331
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001332 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001333}
1334
1335static void check_and_drop(void *_data)
1336{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001337 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001338
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001339 if (!data->mountpoint && !data->select.found)
1340 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001341}
1342
1343/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001344 * d_invalidate - detach submounts, prune dcache, and drop
1345 * @dentry: dentry to invalidate (aka detach, prune and drop)
1346 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001347 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001348 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001349 * The final d_drop is done as an atomic operation relative to
1350 * rename_lock ensuring there are no races with d_set_mounted. This
1351 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001352 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001353void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001354{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001355 /*
1356 * If it's already been dropped, return OK.
1357 */
1358 spin_lock(&dentry->d_lock);
1359 if (d_unhashed(dentry)) {
1360 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001361 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001362 }
1363 spin_unlock(&dentry->d_lock);
1364
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001365 /* Negative dentries can be dropped without further checks */
1366 if (!dentry->d_inode) {
1367 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001368 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001369 }
1370
1371 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001372 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001373
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001374 data.mountpoint = NULL;
1375 INIT_LIST_HEAD(&data.select.dispose);
1376 data.select.start = dentry;
1377 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001378
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001379 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001380
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001381 if (data.select.found)
1382 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001383
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001384 if (data.mountpoint) {
1385 detach_mounts(data.mountpoint);
1386 dput(data.mountpoint);
1387 }
1388
1389 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001390 break;
1391
1392 cond_resched();
1393 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001394}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001395EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397/**
Al Viroa4464db2011-07-07 15:03:58 -04001398 * __d_alloc - allocate a dcache entry
1399 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 * @name: qstr of the name
1401 *
1402 * Allocates a dentry. It returns %NULL if there is insufficient memory
1403 * available. On a success the dentry is returned. The name passed in is
1404 * copied and the copy passed in may be reused after this call.
1405 */
1406
Al Viroa4464db2011-07-07 15:03:58 -04001407struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408{
1409 struct dentry *dentry;
1410 char *dname;
1411
Mel Gormane12ba742007-10-16 01:25:52 -07001412 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 if (!dentry)
1414 return NULL;
1415
Linus Torvalds6326c712012-05-21 16:14:04 -07001416 /*
1417 * We guarantee that the inline name is always NUL-terminated.
1418 * This way the memcpy() done by the name switching in rename
1419 * will still always have a NUL at the end, even if we might
1420 * be overwriting an internal NUL character
1421 */
1422 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001424 size_t size = offsetof(struct external_name, name[1]);
1425 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1426 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 kmem_cache_free(dentry_cache, dentry);
1428 return NULL;
1429 }
Al Viro8d85b482014-09-29 14:54:27 -04001430 atomic_set(&p->u.count, 1);
1431 dname = p->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 } else {
1433 dname = dentry->d_iname;
1434 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436 dentry->d_name.len = name->len;
1437 dentry->d_name.hash = name->hash;
1438 memcpy(dname, name->name, name->len);
1439 dname[name->len] = 0;
1440
Linus Torvalds6326c712012-05-21 16:14:04 -07001441 /* Make sure we always see the terminating NUL character */
1442 smp_wmb();
1443 dentry->d_name.name = dname;
1444
Waiman Long98474232013-08-28 18:24:59 -07001445 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001446 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001448 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001450 dentry->d_parent = dentry;
1451 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 dentry->d_op = NULL;
1453 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001454 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 INIT_LIST_HEAD(&dentry->d_lru);
1456 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001457 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001458 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001459 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
Nick Piggin3e880fb2011-01-07 17:49:19 +11001461 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 return dentry;
1464}
Al Viroa4464db2011-07-07 15:03:58 -04001465
1466/**
1467 * d_alloc - allocate a dcache entry
1468 * @parent: parent of entry to allocate
1469 * @name: qstr of the name
1470 *
1471 * Allocates a dentry. It returns %NULL if there is insufficient memory
1472 * available. On a success the dentry is returned. The name passed in is
1473 * copied and the copy passed in may be reused after this call.
1474 */
1475struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1476{
1477 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1478 if (!dentry)
1479 return NULL;
1480
1481 spin_lock(&parent->d_lock);
1482 /*
1483 * don't need child lock because it is not subject
1484 * to concurrency here
1485 */
1486 __dget_dlock(parent);
1487 dentry->d_parent = parent;
1488 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1489 spin_unlock(&parent->d_lock);
1490
1491 return dentry;
1492}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001493EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001495/**
1496 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1497 * @sb: the superblock
1498 * @name: qstr of the name
1499 *
1500 * For a filesystem that just pins its dentries in memory and never
1501 * performs lookups at all, return an unhashed IS_ROOT dentry.
1502 */
Nick Piggin4b936882011-01-07 17:50:07 +11001503struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1504{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001505 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001506}
1507EXPORT_SYMBOL(d_alloc_pseudo);
1508
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1510{
1511 struct qstr q;
1512
1513 q.name = name;
1514 q.len = strlen(name);
1515 q.hash = full_name_hash(q.name, q.len);
1516 return d_alloc(parent, &q);
1517}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001518EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519
Nick Pigginfb045ad2011-01-07 17:49:55 +11001520void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1521{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001522 WARN_ON_ONCE(dentry->d_op);
1523 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001524 DCACHE_OP_COMPARE |
1525 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001526 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001527 DCACHE_OP_DELETE ));
1528 dentry->d_op = op;
1529 if (!op)
1530 return;
1531 if (op->d_hash)
1532 dentry->d_flags |= DCACHE_OP_HASH;
1533 if (op->d_compare)
1534 dentry->d_flags |= DCACHE_OP_COMPARE;
1535 if (op->d_revalidate)
1536 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001537 if (op->d_weak_revalidate)
1538 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001539 if (op->d_delete)
1540 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001541 if (op->d_prune)
1542 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001543
1544}
1545EXPORT_SYMBOL(d_set_d_op);
1546
David Howellsb18825a2013-09-12 19:22:53 +01001547static unsigned d_flags_for_inode(struct inode *inode)
1548{
1549 unsigned add_flags = DCACHE_FILE_TYPE;
1550
1551 if (!inode)
1552 return DCACHE_MISS_TYPE;
1553
1554 if (S_ISDIR(inode->i_mode)) {
1555 add_flags = DCACHE_DIRECTORY_TYPE;
1556 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1557 if (unlikely(!inode->i_op->lookup))
1558 add_flags = DCACHE_AUTODIR_TYPE;
1559 else
1560 inode->i_opflags |= IOP_LOOKUP;
1561 }
1562 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1563 if (unlikely(inode->i_op->follow_link))
1564 add_flags = DCACHE_SYMLINK_TYPE;
1565 else
1566 inode->i_opflags |= IOP_NOFOLLOW;
1567 }
1568
1569 if (unlikely(IS_AUTOMOUNT(inode)))
1570 add_flags |= DCACHE_NEED_AUTOMOUNT;
1571 return add_flags;
1572}
1573
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001574static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1575{
David Howellsb18825a2013-09-12 19:22:53 +01001576 unsigned add_flags = d_flags_for_inode(inode);
1577
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001578 spin_lock(&dentry->d_lock);
Al Viro22213312014-04-19 12:30:58 -04001579 __d_set_type(dentry, add_flags);
David Howellsb18825a2013-09-12 19:22:53 +01001580 if (inode)
Al Virob3d9b7a2012-06-09 13:51:19 -04001581 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001582 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001583 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001584 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001585 fsnotify_d_instantiate(dentry, inode);
1586}
1587
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588/**
1589 * d_instantiate - fill in inode information for a dentry
1590 * @entry: dentry to complete
1591 * @inode: inode to attach to this dentry
1592 *
1593 * Fill in inode information in the entry.
1594 *
1595 * This turns negative dentries into productive full members
1596 * of society.
1597 *
1598 * NOTE! This assumes that the inode count has been incremented
1599 * (or otherwise set) by the caller to indicate that it is now
1600 * in use by the dcache.
1601 */
1602
1603void d_instantiate(struct dentry *entry, struct inode * inode)
1604{
Al Virob3d9b7a2012-06-09 13:51:19 -04001605 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001606 if (inode)
1607 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001608 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001609 if (inode)
1610 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 security_d_instantiate(entry, inode);
1612}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001613EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614
1615/**
1616 * d_instantiate_unique - instantiate a non-aliased dentry
1617 * @entry: dentry to instantiate
1618 * @inode: inode to attach to this dentry
1619 *
1620 * Fill in inode information in the entry. On success, it returns NULL.
1621 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001622 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 *
1624 * Note that in order to avoid conflicts with rename() etc, the caller
1625 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001626 *
1627 * This also assumes that the inode count has been incremented
1628 * (or otherwise set) by the caller to indicate that it is now
1629 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 */
David Howells770bfad2006-08-22 20:06:07 -04001631static struct dentry *__d_instantiate_unique(struct dentry *entry,
1632 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633{
1634 struct dentry *alias;
1635 int len = entry->d_name.len;
1636 const char *name = entry->d_name.name;
1637 unsigned int hash = entry->d_name.hash;
1638
David Howells770bfad2006-08-22 20:06:07 -04001639 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001640 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001641 return NULL;
1642 }
1643
Sasha Levinb67bfe02013-02-27 17:06:00 -08001644 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001645 /*
1646 * Don't need alias->d_lock here, because aliases with
1647 * d_parent == entry->d_parent are not subject to name or
1648 * parent changes, because the parent inode i_mutex is held.
1649 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001650 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 continue;
1652 if (alias->d_parent != entry->d_parent)
1653 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001654 if (alias->d_name.len != len)
1655 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001656 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001658 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 return alias;
1660 }
David Howells770bfad2006-08-22 20:06:07 -04001661
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001662 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 return NULL;
1664}
David Howells770bfad2006-08-22 20:06:07 -04001665
1666struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1667{
1668 struct dentry *result;
1669
Al Virob3d9b7a2012-06-09 13:51:19 -04001670 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001671
Nick Piggin873feea2011-01-07 17:50:06 +11001672 if (inode)
1673 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001674 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001675 if (inode)
1676 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001677
1678 if (!result) {
1679 security_d_instantiate(entry, inode);
1680 return NULL;
1681 }
1682
1683 BUG_ON(!d_unhashed(result));
1684 iput(inode);
1685 return result;
1686}
1687
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688EXPORT_SYMBOL(d_instantiate_unique);
1689
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001690/**
1691 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1692 * @entry: dentry to complete
1693 * @inode: inode to attach to this dentry
1694 *
1695 * Fill in inode information in the entry. If a directory alias is found, then
1696 * return an error (and drop inode). Together with d_materialise_unique() this
1697 * guarantees that a directory inode may never have more than one alias.
1698 */
1699int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1700{
1701 BUG_ON(!hlist_unhashed(&entry->d_alias));
1702
1703 spin_lock(&inode->i_lock);
1704 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1705 spin_unlock(&inode->i_lock);
1706 iput(inode);
1707 return -EBUSY;
1708 }
1709 __d_instantiate(entry, inode);
1710 spin_unlock(&inode->i_lock);
1711 security_d_instantiate(entry, inode);
1712
1713 return 0;
1714}
1715EXPORT_SYMBOL(d_instantiate_no_diralias);
1716
Al Viroadc0e912012-01-08 16:49:21 -05001717struct dentry *d_make_root(struct inode *root_inode)
1718{
1719 struct dentry *res = NULL;
1720
1721 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001722 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001723
1724 res = __d_alloc(root_inode->i_sb, &name);
1725 if (res)
1726 d_instantiate(res, root_inode);
1727 else
1728 iput(root_inode);
1729 }
1730 return res;
1731}
1732EXPORT_SYMBOL(d_make_root);
1733
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001734static struct dentry * __d_find_any_alias(struct inode *inode)
1735{
1736 struct dentry *alias;
1737
Al Virob3d9b7a2012-06-09 13:51:19 -04001738 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001739 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001740 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001741 __dget(alias);
1742 return alias;
1743}
1744
Sage Weil46f72b32012-01-10 09:04:37 -08001745/**
1746 * d_find_any_alias - find any alias for a given inode
1747 * @inode: inode to find an alias for
1748 *
1749 * If any aliases exist for the given inode, take and return a
1750 * reference for one of them. If no aliases exist, return %NULL.
1751 */
1752struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001753{
1754 struct dentry *de;
1755
1756 spin_lock(&inode->i_lock);
1757 de = __d_find_any_alias(inode);
1758 spin_unlock(&inode->i_lock);
1759 return de;
1760}
Sage Weil46f72b32012-01-10 09:04:37 -08001761EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001762
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001763static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001764{
NeilBrownb911a6b2012-11-08 16:09:37 -08001765 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001766 struct dentry *tmp;
1767 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001768 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001769
1770 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001771 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001772 if (IS_ERR(inode))
1773 return ERR_CAST(inode);
1774
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001775 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001776 if (res)
1777 goto out_iput;
1778
Al Viroa4464db2011-07-07 15:03:58 -04001779 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001780 if (!tmp) {
1781 res = ERR_PTR(-ENOMEM);
1782 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001783 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001784
Nick Piggin873feea2011-01-07 17:50:06 +11001785 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001786 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001787 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001788 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001789 dput(tmp);
1790 goto out_iput;
1791 }
1792
1793 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001794 add_flags = d_flags_for_inode(inode);
1795
1796 if (disconnected)
1797 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001798
Christoph Hellwig9308a612008-08-11 15:49:12 +02001799 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001800 tmp->d_inode = inode;
David Howellsb18825a2013-09-12 19:22:53 +01001801 tmp->d_flags |= add_flags;
Al Virob3d9b7a2012-06-09 13:51:19 -04001802 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001803 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001804 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001805 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001806 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001807 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001808 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001809
Christoph Hellwig9308a612008-08-11 15:49:12 +02001810 return tmp;
1811
1812 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001813 if (res && !IS_ERR(res))
1814 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001815 iput(inode);
1816 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001817}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001818
1819/**
1820 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1821 * @inode: inode to allocate the dentry for
1822 *
1823 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1824 * similar open by handle operations. The returned dentry may be anonymous,
1825 * or may have a full name (if the inode was already in the cache).
1826 *
1827 * When called on a directory inode, we must ensure that the inode only ever
1828 * has one dentry. If a dentry is found, that is returned instead of
1829 * allocating a new one.
1830 *
1831 * On successful return, the reference to the inode has been transferred
1832 * to the dentry. In case of an error the reference on the inode is released.
1833 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1834 * be passed in and the error will be propagated to the return value,
1835 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1836 */
1837struct dentry *d_obtain_alias(struct inode *inode)
1838{
1839 return __d_obtain_alias(inode, 1);
1840}
Benny Halevyadc48722009-02-27 14:02:59 -08001841EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842
1843/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001844 * d_obtain_root - find or allocate a dentry for a given inode
1845 * @inode: inode to allocate the dentry for
1846 *
1847 * Obtain an IS_ROOT dentry for the root of a filesystem.
1848 *
1849 * We must ensure that directory inodes only ever have one dentry. If a
1850 * dentry is found, that is returned instead of allocating a new one.
1851 *
1852 * On successful return, the reference to the inode has been transferred
1853 * to the dentry. In case of an error the reference on the inode is
1854 * released. A %NULL or IS_ERR inode may be passed in and will be the
1855 * error will be propagate to the return value, with a %NULL @inode
1856 * replaced by ERR_PTR(-ESTALE).
1857 */
1858struct dentry *d_obtain_root(struct inode *inode)
1859{
1860 return __d_obtain_alias(inode, 0);
1861}
1862EXPORT_SYMBOL(d_obtain_root);
1863
1864/**
Barry Naujok94035402008-05-21 16:50:46 +10001865 * d_add_ci - lookup or allocate new dentry with case-exact name
1866 * @inode: the inode case-insensitive lookup has found
1867 * @dentry: the negative dentry that was passed to the parent's lookup func
1868 * @name: the case-exact name to be associated with the returned dentry
1869 *
1870 * This is to avoid filling the dcache with case-insensitive names to the
1871 * same inode, only the actual correct case is stored in the dcache for
1872 * case-insensitive filesystems.
1873 *
1874 * For a case-insensitive lookup match and if the the case-exact dentry
1875 * already exists in in the dcache, use it and return it.
1876 *
1877 * If no entry exists with the exact case name, allocate new dentry with
1878 * the exact case, and return the spliced entry.
1879 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001880struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001881 struct qstr *name)
1882{
Barry Naujok94035402008-05-21 16:50:46 +10001883 struct dentry *found;
1884 struct dentry *new;
1885
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001886 /*
1887 * First check if a dentry matching the name already exists,
1888 * if not go ahead and create it now.
1889 */
Barry Naujok94035402008-05-21 16:50:46 +10001890 found = d_hash_and_lookup(dentry->d_parent, name);
Al Viro4f522a22013-02-11 23:20:37 -05001891 if (unlikely(IS_ERR(found)))
1892 goto err_out;
Barry Naujok94035402008-05-21 16:50:46 +10001893 if (!found) {
1894 new = d_alloc(dentry->d_parent, name);
1895 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05001896 found = ERR_PTR(-ENOMEM);
Barry Naujok94035402008-05-21 16:50:46 +10001897 goto err_out;
1898 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001899
Barry Naujok94035402008-05-21 16:50:46 +10001900 found = d_splice_alias(inode, new);
1901 if (found) {
1902 dput(new);
1903 return found;
1904 }
1905 return new;
1906 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001907
1908 /*
1909 * If a matching dentry exists, and it's not negative use it.
1910 *
1911 * Decrement the reference count to balance the iget() done
1912 * earlier on.
1913 */
Barry Naujok94035402008-05-21 16:50:46 +10001914 if (found->d_inode) {
1915 if (unlikely(found->d_inode != inode)) {
1916 /* This can't happen because bad inodes are unhashed. */
1917 BUG_ON(!is_bad_inode(inode));
1918 BUG_ON(!is_bad_inode(found->d_inode));
1919 }
Barry Naujok94035402008-05-21 16:50:46 +10001920 iput(inode);
1921 return found;
1922 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001923
Barry Naujok94035402008-05-21 16:50:46 +10001924 /*
1925 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001926 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001927 */
Al Viro4513d892011-07-17 10:52:14 -04001928 new = d_splice_alias(inode, found);
1929 if (new) {
1930 dput(found);
1931 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10001932 }
Al Viro4513d892011-07-17 10:52:14 -04001933 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001934
1935err_out:
1936 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05001937 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001938}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001939EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001941/*
1942 * Do the slow-case of the dentry name compare.
1943 *
1944 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07001945 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001946 * filesystem can rely on them, and can use the 'name' and
1947 * 'len' information without worrying about walking off the
1948 * end of memory etc.
1949 *
1950 * Thus the read_seqcount_retry() and the "duplicate" info
1951 * in arguments (the low-level filesystem should not look
1952 * at the dentry inode or name contents directly, since
1953 * rename can change them while we're in RCU mode).
1954 */
1955enum slow_d_compare {
1956 D_COMP_OK,
1957 D_COMP_NOMATCH,
1958 D_COMP_SEQRETRY,
1959};
1960
1961static noinline enum slow_d_compare slow_dentry_cmp(
1962 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001963 struct dentry *dentry,
1964 unsigned int seq,
1965 const struct qstr *name)
1966{
1967 int tlen = dentry->d_name.len;
1968 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001969
1970 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1971 cpu_relax();
1972 return D_COMP_SEQRETRY;
1973 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07001974 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001975 return D_COMP_NOMATCH;
1976 return D_COMP_OK;
1977}
1978
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979/**
Nick Piggin31e6b012011-01-07 17:49:52 +11001980 * __d_lookup_rcu - search for a dentry (racy, store-free)
1981 * @parent: parent dentry
1982 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07001983 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11001984 * Returns: dentry, or NULL
1985 *
1986 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1987 * resolution (store-free path walking) design described in
1988 * Documentation/filesystems/path-lookup.txt.
1989 *
1990 * This is not to be used outside core vfs.
1991 *
1992 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1993 * held, and rcu_read_lock held. The returned dentry must not be stored into
1994 * without taking d_lock and checking d_seq sequence count against @seq
1995 * returned here.
1996 *
Linus Torvalds15570082013-09-02 11:38:06 -07001997 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11001998 * function.
1999 *
2000 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2001 * the returned dentry, so long as its parent's seqlock is checked after the
2002 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2003 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002004 *
2005 * NOTE! The caller *has* to check the resulting dentry against the sequence
2006 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002007 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002008struct dentry *__d_lookup_rcu(const struct dentry *parent,
2009 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002010 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002011{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002012 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002013 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002014 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002015 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002016 struct dentry *dentry;
2017
2018 /*
2019 * Note: There is significant duplication with __d_lookup_rcu which is
2020 * required to prevent single threaded performance regressions
2021 * especially on architectures where smp_rmb (in seqcounts) are costly.
2022 * Keep the two functions in sync.
2023 */
2024
2025 /*
2026 * The hash list is protected using RCU.
2027 *
2028 * Carefully use d_seq when comparing a candidate dentry, to avoid
2029 * races with d_move().
2030 *
2031 * It is possible that concurrent renames can mess up our list
2032 * walk here and result in missing our dentry, resulting in the
2033 * false-negative result. d_lookup() protects against concurrent
2034 * renames using rename_lock seqlock.
2035 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002036 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002037 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002038 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002039 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002040
Nick Piggin31e6b012011-01-07 17:49:52 +11002041seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002042 /*
2043 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002044 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002045 *
2046 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002047 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002048 *
2049 * NOTE! We do a "raw" seqcount_begin here. That means that
2050 * we don't wait for the sequence count to stabilize if it
2051 * is in the middle of a sequence change. If we do the slow
2052 * dentry compare, we will do seqretries until it is stable,
2053 * and if we end up with a successful lookup, we actually
2054 * want to exit RCU lookup anyway.
2055 */
2056 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002057 if (dentry->d_parent != parent)
2058 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002059 if (d_unhashed(dentry))
2060 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002061
2062 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002063 if (dentry->d_name.hash != hashlen_hash(hashlen))
2064 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002065 *seqp = seq;
2066 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002067 case D_COMP_OK:
2068 return dentry;
2069 case D_COMP_NOMATCH:
2070 continue;
2071 default:
2072 goto seqretry;
2073 }
2074 }
2075
Linus Torvalds26fe5752012-05-10 13:14:12 -07002076 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002077 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002078 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002079 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002080 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002081 }
2082 return NULL;
2083}
2084
2085/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 * d_lookup - search for a dentry
2087 * @parent: parent dentry
2088 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002089 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002091 * d_lookup searches the children of the parent dentry for the name in
2092 * question. If the dentry is found its reference count is incremented and the
2093 * dentry is returned. The caller must use dput to free the entry when it has
2094 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 */
Al Viroda2d8452013-01-24 18:29:34 -05002096struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097{
Nick Piggin31e6b012011-01-07 17:49:52 +11002098 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002099 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100
Daeseok Younb8314f92014-08-11 11:46:53 +09002101 do {
2102 seq = read_seqbegin(&rename_lock);
2103 dentry = __d_lookup(parent, name);
2104 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 break;
2106 } while (read_seqretry(&rename_lock, seq));
2107 return dentry;
2108}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002109EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
Nick Piggin31e6b012011-01-07 17:49:52 +11002111/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002112 * __d_lookup - search for a dentry (racy)
2113 * @parent: parent dentry
2114 * @name: qstr of name we wish to find
2115 * Returns: dentry, or NULL
2116 *
2117 * __d_lookup is like d_lookup, however it may (rarely) return a
2118 * false-negative result due to unrelated rename activity.
2119 *
2120 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2121 * however it must be used carefully, eg. with a following d_lookup in
2122 * the case of failure.
2123 *
2124 * __d_lookup callers must be commented.
2125 */
Al Viroa713ca22013-01-24 18:27:00 -05002126struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127{
2128 unsigned int len = name->len;
2129 unsigned int hash = name->hash;
2130 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002131 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002132 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002133 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002134 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135
Nick Pigginb04f7842010-08-18 04:37:34 +10002136 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002137 * Note: There is significant duplication with __d_lookup_rcu which is
2138 * required to prevent single threaded performance regressions
2139 * especially on architectures where smp_rmb (in seqcounts) are costly.
2140 * Keep the two functions in sync.
2141 */
2142
2143 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002144 * The hash list is protected using RCU.
2145 *
2146 * Take d_lock when comparing a candidate dentry, to avoid races
2147 * with d_move().
2148 *
2149 * It is possible that concurrent renames can mess up our list
2150 * walk here and result in missing our dentry, resulting in the
2151 * false-negative result. d_lookup() protects against concurrent
2152 * renames using rename_lock seqlock.
2153 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002154 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002155 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 rcu_read_lock();
2157
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002158 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 if (dentry->d_name.hash != hash)
2161 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162
2163 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 if (dentry->d_parent != parent)
2165 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002166 if (d_unhashed(dentry))
2167 goto next;
2168
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 /*
2170 * It is safe to compare names since d_move() cannot
2171 * change the qstr (protected by d_lock).
2172 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002173 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002174 int tlen = dentry->d_name.len;
2175 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002176 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 goto next;
2178 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002179 if (dentry->d_name.len != len)
2180 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002181 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 goto next;
2183 }
2184
Waiman Long98474232013-08-28 18:24:59 -07002185 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002186 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 spin_unlock(&dentry->d_lock);
2188 break;
2189next:
2190 spin_unlock(&dentry->d_lock);
2191 }
2192 rcu_read_unlock();
2193
2194 return found;
2195}
2196
2197/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002198 * d_hash_and_lookup - hash the qstr then search for a dentry
2199 * @dir: Directory to search in
2200 * @name: qstr of name we wish to find
2201 *
Al Viro4f522a22013-02-11 23:20:37 -05002202 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002203 */
2204struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2205{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002206 /*
2207 * Check for a fs-specific hash function. Note that we must
2208 * calculate the standard hash first, as the d_op->d_hash()
2209 * routine may choose to leave the hash value unchanged.
2210 */
2211 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002212 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002213 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002214 if (unlikely(err < 0))
2215 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002216 }
Al Viro4f522a22013-02-11 23:20:37 -05002217 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002218}
Al Viro4f522a22013-02-11 23:20:37 -05002219EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002220
2221/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002222 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002224 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 *
2226 * An insecure source has sent us a dentry, here we verify it and dget() it.
2227 * This is used by ncpfs in its readdir implementation.
2228 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002229 *
2230 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002232int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233{
Nick Piggin786a5e12011-01-07 17:49:16 +11002234 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002235
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002236 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002237 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2238 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002239 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002240 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002241 spin_unlock(&dentry->d_lock);
2242 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 return 1;
2244 }
2245 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002246 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002247
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 return 0;
2249}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002250EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251
2252/*
2253 * When a file is deleted, we have two options:
2254 * - turn this dentry into a negative dentry
2255 * - unhash this dentry and free it.
2256 *
2257 * Usually, we want to just turn this into
2258 * a negative dentry, but if anybody else is
2259 * currently using the dentry or the inode
2260 * we can't do that and we fall back on removing
2261 * it from the hash queues and waiting for
2262 * it to be deleted later when it has no users
2263 */
2264
2265/**
2266 * d_delete - delete a dentry
2267 * @dentry: The dentry to delete
2268 *
2269 * Turn the dentry into a negative dentry if possible, otherwise
2270 * remove it from the hash queues so it can be deleted later
2271 */
2272
2273void d_delete(struct dentry * dentry)
2274{
Nick Piggin873feea2011-01-07 17:50:06 +11002275 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002276 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 /*
2278 * Are we the only user?
2279 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002280again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002282 inode = dentry->d_inode;
2283 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002284 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002285 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002286 spin_unlock(&dentry->d_lock);
2287 cpu_relax();
2288 goto again;
2289 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002290 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002291 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002292 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 return;
2294 }
2295
2296 if (!d_unhashed(dentry))
2297 __d_drop(dentry);
2298
2299 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002300
2301 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002303EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002305static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002307 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002308 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002309 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002310 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002311 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312}
2313
David Howells770bfad2006-08-22 20:06:07 -04002314static void _d_rehash(struct dentry * entry)
2315{
2316 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2317}
2318
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319/**
2320 * d_rehash - add an entry back to the hash
2321 * @entry: dentry to add to the hash
2322 *
2323 * Adds a dentry to the hash according to its name.
2324 */
2325
2326void d_rehash(struct dentry * entry)
2327{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002329 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002332EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002334/**
2335 * dentry_update_name_case - update case insensitive dentry with a new name
2336 * @dentry: dentry to be updated
2337 * @name: new name
2338 *
2339 * Update a case insensitive dentry with new case of name.
2340 *
2341 * dentry must have been returned by d_lookup with name @name. Old and new
2342 * name lengths must match (ie. no d_compare which allows mismatched name
2343 * lengths).
2344 *
2345 * Parent inode i_mutex must be held over d_lookup and into this call (to
2346 * keep renames and concurrent inserts, and readdir(2) away).
2347 */
2348void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2349{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002350 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002351 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2352
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002353 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002354 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002355 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002356 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002357 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002358}
2359EXPORT_SYMBOL(dentry_update_name_case);
2360
Al Viro8d85b482014-09-29 14:54:27 -04002361static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362{
Al Viro8d85b482014-09-29 14:54:27 -04002363 if (unlikely(dname_external(target))) {
2364 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 /*
2366 * Both external: swap the pointers
2367 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002368 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 } else {
2370 /*
2371 * dentry:internal, target:external. Steal target's
2372 * storage and make target internal.
2373 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002374 memcpy(target->d_iname, dentry->d_name.name,
2375 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 dentry->d_name.name = target->d_name.name;
2377 target->d_name.name = target->d_iname;
2378 }
2379 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002380 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 /*
2382 * dentry:external, target:internal. Give dentry's
2383 * storage to target and make dentry internal
2384 */
2385 memcpy(dentry->d_iname, target->d_name.name,
2386 target->d_name.len + 1);
2387 target->d_name.name = dentry->d_name.name;
2388 dentry->d_name.name = dentry->d_iname;
2389 } else {
2390 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002391 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002393 unsigned int i;
2394 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2395 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2396 swap(((long *) &dentry->d_iname)[i],
2397 ((long *) &target->d_iname)[i]);
2398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 }
2400 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002401 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402}
2403
Al Viro8d85b482014-09-29 14:54:27 -04002404static void copy_name(struct dentry *dentry, struct dentry *target)
2405{
2406 struct external_name *old_name = NULL;
2407 if (unlikely(dname_external(dentry)))
2408 old_name = external_name(dentry);
2409 if (unlikely(dname_external(target))) {
2410 atomic_inc(&external_name(target)->u.count);
2411 dentry->d_name = target->d_name;
2412 } else {
2413 memcpy(dentry->d_iname, target->d_name.name,
2414 target->d_name.len + 1);
2415 dentry->d_name.name = dentry->d_iname;
2416 dentry->d_name.hash_len = target->d_name.hash_len;
2417 }
2418 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2419 kfree_rcu(old_name, u.head);
2420}
2421
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002422static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2423{
2424 /*
2425 * XXXX: do we really need to take target->d_lock?
2426 */
2427 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2428 spin_lock(&target->d_parent->d_lock);
2429 else {
2430 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2431 spin_lock(&dentry->d_parent->d_lock);
2432 spin_lock_nested(&target->d_parent->d_lock,
2433 DENTRY_D_LOCK_NESTED);
2434 } else {
2435 spin_lock(&target->d_parent->d_lock);
2436 spin_lock_nested(&dentry->d_parent->d_lock,
2437 DENTRY_D_LOCK_NESTED);
2438 }
2439 }
2440 if (target < dentry) {
2441 spin_lock_nested(&target->d_lock, 2);
2442 spin_lock_nested(&dentry->d_lock, 3);
2443 } else {
2444 spin_lock_nested(&dentry->d_lock, 2);
2445 spin_lock_nested(&target->d_lock, 3);
2446 }
2447}
2448
Al Viro986c0192014-09-26 23:11:15 -04002449static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002450{
2451 if (target->d_parent != dentry->d_parent)
2452 spin_unlock(&dentry->d_parent->d_lock);
2453 if (target->d_parent != target)
2454 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002455 spin_unlock(&target->d_lock);
2456 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002457}
2458
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002460 * When switching names, the actual string doesn't strictly have to
2461 * be preserved in the target - because we're dropping the target
2462 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002463 * the new name before we switch, unless we are going to rehash
2464 * it. Note that if we *do* unhash the target, we are not allowed
2465 * to rehash it without giving it a new name/hash key - whether
2466 * we swap or overwrite the names here, resulting name won't match
2467 * the reality in filesystem; it's only there for d_path() purposes.
2468 * Note that all of this is happening under rename_lock, so the
2469 * any hash lookup seeing it in the middle of manipulations will
2470 * be discarded anyway. So we do not care what happens to the hash
2471 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002473/*
Al Viro18367502011-07-12 21:42:24 -04002474 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 * @dentry: entry to move
2476 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002477 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 *
2479 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002480 * dcache entries should not be moved in this way. Caller must hold
2481 * rename_lock, the i_mutex of the source and target directories,
2482 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002484static void __d_move(struct dentry *dentry, struct dentry *target,
2485 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 if (!dentry->d_inode)
2488 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2489
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002490 BUG_ON(d_ancestor(dentry, target));
2491 BUG_ON(d_ancestor(target, dentry));
2492
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002493 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494
Nick Piggin31e6b012011-01-07 17:49:52 +11002495 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002496 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002497
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002498 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2499
2500 /*
2501 * Move the dentry to the target hash queue. Don't bother checking
2502 * for the same hash queue because of how unlikely it is.
2503 */
2504 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002505 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506
Miklos Szeredida1ce062014-04-01 17:08:43 +02002507 /*
2508 * Unhash the target (d_delete() is not usable here). If exchanging
2509 * the two dentries, then rehash onto the other's hash queue.
2510 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002512 if (exchange) {
2513 __d_rehash(target,
2514 d_hash(dentry->d_parent, dentry->d_name.hash));
2515 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002518 if (exchange)
2519 swap_names(dentry, target);
2520 else
2521 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522
Al Viro63cf4272014-09-26 23:06:14 -04002523 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002525 /* splicing a tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 dentry->d_parent = target->d_parent;
2527 target->d_parent = target;
Al Viro9d8cd302014-09-26 21:34:01 -04002528 list_del_init(&target->d_u.d_child);
Al Viro63cf4272014-09-26 23:06:14 -04002529 list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002531 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002532 swap(dentry->d_parent, target->d_parent);
Al Viro9d8cd302014-09-26 21:34:01 -04002533 list_move(&target->d_u.d_child, &target->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002534 list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2535 if (exchange)
2536 fsnotify_d_move(target);
2537 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 }
2539
Nick Piggin31e6b012011-01-07 17:49:52 +11002540 write_seqcount_end(&target->d_seq);
2541 write_seqcount_end(&dentry->d_seq);
2542
Al Viro986c0192014-09-26 23:11:15 -04002543 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002544}
2545
2546/*
2547 * d_move - move a dentry
2548 * @dentry: entry to move
2549 * @target: new dentry
2550 *
2551 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002552 * dcache entries should not be moved in this way. See the locking
2553 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002554 */
2555void d_move(struct dentry *dentry, struct dentry *target)
2556{
2557 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002558 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002560}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002561EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562
Miklos Szeredida1ce062014-04-01 17:08:43 +02002563/*
2564 * d_exchange - exchange two dentries
2565 * @dentry1: first dentry
2566 * @dentry2: second dentry
2567 */
2568void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2569{
2570 write_seqlock(&rename_lock);
2571
2572 WARN_ON(!dentry1->d_inode);
2573 WARN_ON(!dentry2->d_inode);
2574 WARN_ON(IS_ROOT(dentry1));
2575 WARN_ON(IS_ROOT(dentry2));
2576
2577 __d_move(dentry1, dentry2, true);
2578
2579 write_sequnlock(&rename_lock);
2580}
2581
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002582/**
2583 * d_ancestor - search for an ancestor
2584 * @p1: ancestor dentry
2585 * @p2: child dentry
2586 *
2587 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2588 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002589 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002590struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002591{
2592 struct dentry *p;
2593
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002594 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002595 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002596 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002597 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002598 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002599}
2600
2601/*
2602 * This helper attempts to cope with remotely renamed directories
2603 *
2604 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002605 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002606 *
2607 * Note: If ever the locking in lock_rename() changes, then please
2608 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002609 */
Nick Piggin873feea2011-01-07 17:50:06 +11002610static struct dentry *__d_unalias(struct inode *inode,
2611 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002612{
2613 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002614 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002615
2616 /* If alias and dentry share a parent, then no extra locks required */
2617 if (alias->d_parent == dentry->d_parent)
2618 goto out_unalias;
2619
Trond Myklebust9eaef272006-10-21 10:24:20 -07002620 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002621 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2622 goto out_err;
2623 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2624 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2625 goto out_err;
2626 m2 = &alias->d_parent->d_inode->i_mutex;
2627out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002628 __d_move(alias, dentry, false);
2629 ret = alias;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002630out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002631 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002632 if (m2)
2633 mutex_unlock(m2);
2634 if (m1)
2635 mutex_unlock(m1);
2636 return ret;
2637}
2638
David Howells770bfad2006-08-22 20:06:07 -04002639/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002640 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2641 * @inode: the inode which may have a disconnected dentry
2642 * @dentry: a negative dentry which we want to point to the inode.
2643 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002644 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2645 * place of the given dentry and return it, else simply d_add the inode
2646 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002647 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002648 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2649 * we should error out: directories can't have multiple aliases.
2650 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002651 * This is needed in the lookup routine of any filesystem that is exportable
2652 * (via knfsd) so that we can build dcache paths to directories effectively.
2653 *
2654 * If a dentry was found and moved, then it is returned. Otherwise NULL
2655 * is returned. This matches the expected return value of ->lookup.
2656 *
2657 * Cluster filesystems may call this function with a negative, hashed dentry.
2658 * In that case, we know that the inode will be a regular file, and also this
2659 * will only occur during atomic_open. So we need to check for the dentry
2660 * being already hashed only in the final case.
2661 */
2662struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2663{
2664 struct dentry *new = NULL;
2665
2666 if (IS_ERR(inode))
2667 return ERR_CAST(inode);
2668
2669 if (inode && S_ISDIR(inode->i_mode)) {
2670 spin_lock(&inode->i_lock);
J. Bruce Fields908790f2014-02-17 17:58:42 -05002671 new = __d_find_any_alias(inode);
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002672 if (new) {
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002673 if (!IS_ROOT(new)) {
J. Bruce Fields908790f2014-02-17 17:58:42 -05002674 spin_unlock(&inode->i_lock);
2675 dput(new);
Al Viro51486b92014-10-23 13:26:21 -04002676 iput(inode);
J. Bruce Fields908790f2014-02-17 17:58:42 -05002677 return ERR_PTR(-EIO);
2678 }
J. Bruce Fields95ad5c22014-03-12 12:19:23 -04002679 if (d_ancestor(new, dentry)) {
2680 spin_unlock(&inode->i_lock);
2681 dput(new);
Al Viro51486b92014-10-23 13:26:21 -04002682 iput(inode);
J. Bruce Fields95ad5c22014-03-12 12:19:23 -04002683 return ERR_PTR(-EIO);
2684 }
J. Bruce Fields75a23522014-02-17 17:45:56 -05002685 write_seqlock(&rename_lock);
Al Viro63cf4272014-09-26 23:06:14 -04002686 __d_move(new, dentry, false);
J. Bruce Fields75a23522014-02-17 17:45:56 -05002687 write_sequnlock(&rename_lock);
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002688 spin_unlock(&inode->i_lock);
2689 security_d_instantiate(new, inode);
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002690 iput(inode);
2691 } else {
2692 /* already taking inode->i_lock, so d_add() by hand */
2693 __d_instantiate(dentry, inode);
2694 spin_unlock(&inode->i_lock);
2695 security_d_instantiate(dentry, inode);
2696 d_rehash(dentry);
2697 }
2698 } else {
2699 d_instantiate(dentry, inode);
2700 if (d_unhashed(dentry))
2701 d_rehash(dentry);
2702 }
2703 return new;
2704}
2705EXPORT_SYMBOL(d_splice_alias);
2706
2707/**
David Howells770bfad2006-08-22 20:06:07 -04002708 * d_materialise_unique - introduce an inode into the tree
2709 * @dentry: candidate dentry
2710 * @inode: inode to bind to the dentry, to which aliases may be attached
2711 *
2712 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002713 * root directory alias in its place if there is one. Caller must hold the
2714 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002715 */
2716struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2717{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002718 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002719
2720 BUG_ON(!d_unhashed(dentry));
2721
David Howells770bfad2006-08-22 20:06:07 -04002722 if (!inode) {
2723 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002724 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002725 d_rehash(actual);
2726 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002727 }
2728
Nick Piggin873feea2011-01-07 17:50:06 +11002729 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002730
Trond Myklebust9eaef272006-10-21 10:24:20 -07002731 if (S_ISDIR(inode->i_mode)) {
2732 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002733
Trond Myklebust9eaef272006-10-21 10:24:20 -07002734 /* Does an aliased dentry already exist? */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -05002735 alias = __d_find_alias(inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002736 if (alias) {
2737 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002738 write_seqlock(&rename_lock);
2739
2740 if (d_ancestor(alias, dentry)) {
2741 /* Check for loops */
2742 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002743 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002744 } else if (IS_ROOT(alias)) {
2745 /* Is this an anonymous mountpoint that we
2746 * could splice into our tree? */
Al Viro63cf4272014-09-26 23:06:14 -04002747 __d_move(alias, dentry, false);
Al Viro18367502011-07-12 21:42:24 -04002748 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002749 goto found;
Al Viro18367502011-07-12 21:42:24 -04002750 } else {
2751 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002752 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002753 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002754 }
Al Viro18367502011-07-12 21:42:24 -04002755 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002756 if (IS_ERR(actual)) {
2757 if (PTR_ERR(actual) == -ELOOP)
2758 pr_warn_ratelimited(
2759 "VFS: Lookup of '%s' in %s %s"
2760 " would have caused loop\n",
2761 dentry->d_name.name,
2762 inode->i_sb->s_type->name,
2763 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002764 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002765 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002766 goto out_nolock;
2767 }
David Howells770bfad2006-08-22 20:06:07 -04002768 }
2769
2770 /* Add a unique reference */
2771 actual = __d_instantiate_unique(dentry, inode);
2772 if (!actual)
2773 actual = dentry;
David Howells770bfad2006-08-22 20:06:07 -04002774
Al Viro8527dd72014-09-26 21:26:50 -04002775 d_rehash(actual);
Al Viro5cc38212014-09-26 21:20:39 -04002776found:
Nick Piggin873feea2011-01-07 17:50:06 +11002777 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002778out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002779 if (actual == dentry) {
2780 security_d_instantiate(dentry, inode);
2781 return NULL;
2782 }
2783
2784 iput(inode);
2785 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002786}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002787EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002788
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002789static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002790{
2791 *buflen -= namelen;
2792 if (*buflen < 0)
2793 return -ENAMETOOLONG;
2794 *buffer -= namelen;
2795 memcpy(*buffer, str, namelen);
2796 return 0;
2797}
2798
Waiman Long232d2d62013-09-09 12:18:13 -04002799/**
2800 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002801 * @buffer: buffer pointer
2802 * @buflen: allocated length of the buffer
2803 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002804 *
2805 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2806 * make sure that either the old or the new name pointer and length are
2807 * fetched. However, there may be mismatch between length and pointer.
2808 * The length cannot be trusted, we need to copy it byte-by-byte until
2809 * the length is reached or a null byte is found. It also prepends "/" at
2810 * the beginning of the name. The sequence number check at the caller will
2811 * retry it again when a d_move() does happen. So any garbage in the buffer
2812 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04002813 *
2814 * Data dependency barrier is needed to make sure that we see that terminating
2815 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04002816 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002817static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2818{
Waiman Long232d2d62013-09-09 12:18:13 -04002819 const char *dname = ACCESS_ONCE(name->name);
2820 u32 dlen = ACCESS_ONCE(name->len);
2821 char *p;
2822
Al Viro6d13f692014-09-29 14:46:30 -04002823 smp_read_barrier_depends();
2824
Waiman Long232d2d62013-09-09 12:18:13 -04002825 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04002826 if (*buflen < 0)
2827 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04002828 p = *buffer -= dlen + 1;
2829 *p++ = '/';
2830 while (dlen--) {
2831 char c = *dname++;
2832 if (!c)
2833 break;
2834 *p++ = c;
2835 }
2836 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002837}
2838
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002840 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002841 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002842 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002843 * @buffer: pointer to the end of the buffer
2844 * @buflen: pointer to buffer length
2845 *
Waiman Long18129972013-09-12 10:55:35 -04002846 * The function will first try to write out the pathname without taking any
2847 * lock other than the RCU read lock to make sure that dentries won't go away.
2848 * It only checks the sequence number of the global rename_lock as any change
2849 * in the dentry's d_seq will be preceded by changes in the rename_lock
2850 * sequence number. If the sequence number had been changed, it will restart
2851 * the whole pathname back-tracing sequence again by taking the rename_lock.
2852 * In this case, there is no need to take the RCU read lock as the recursive
2853 * parent pointer references will keep the dentry chain alive as long as no
2854 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002855 */
Al Viro02125a82011-12-05 08:43:34 -05002856static int prepend_path(const struct path *path,
2857 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002858 char **buffer, int *buflen)
2859{
Al Viroede4ceb2013-11-13 07:45:40 -05002860 struct dentry *dentry;
2861 struct vfsmount *vfsmnt;
2862 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002863 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002864 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002865 char *bptr;
2866 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002867
Al Viro48f5ec22013-09-09 15:22:25 -04002868 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002869restart_mnt:
2870 read_seqbegin_or_lock(&mount_lock, &m_seq);
2871 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002872 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002873restart:
2874 bptr = *buffer;
2875 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002876 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002877 dentry = path->dentry;
2878 vfsmnt = path->mnt;
2879 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002880 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002881 while (dentry != root->dentry || vfsmnt != root->mnt) {
2882 struct dentry * parent;
2883
2884 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002885 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002886 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002887 if (mnt != parent) {
2888 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2889 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002890 vfsmnt = &mnt->mnt;
2891 continue;
2892 }
2893 /*
2894 * Filesystems needing to implement special "root names"
2895 * should do so with ->d_dname()
2896 */
2897 if (IS_ROOT(dentry) &&
2898 (dentry->d_name.len != 1 ||
2899 dentry->d_name.name[0] != '/')) {
2900 WARN(1, "Root dentry has weird name <%.*s>\n",
2901 (int) dentry->d_name.len,
2902 dentry->d_name.name);
2903 }
2904 if (!error)
2905 error = is_mounted(vfsmnt) ? 1 : 2;
2906 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002907 }
2908 parent = dentry->d_parent;
2909 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002910 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002911 if (error)
2912 break;
2913
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002914 dentry = parent;
2915 }
Al Viro48f5ec22013-09-09 15:22:25 -04002916 if (!(seq & 1))
2917 rcu_read_unlock();
2918 if (need_seqretry(&rename_lock, seq)) {
2919 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002920 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002921 }
Al Viro48f5ec22013-09-09 15:22:25 -04002922 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002923
2924 if (!(m_seq & 1))
2925 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002926 if (need_seqretry(&mount_lock, m_seq)) {
2927 m_seq = 1;
2928 goto restart_mnt;
2929 }
2930 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002931
Waiman Long232d2d62013-09-09 12:18:13 -04002932 if (error >= 0 && bptr == *buffer) {
2933 if (--blen < 0)
2934 error = -ENAMETOOLONG;
2935 else
2936 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002937 }
Waiman Long232d2d62013-09-09 12:18:13 -04002938 *buffer = bptr;
2939 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002940 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002941}
2942
2943/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002944 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002945 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002946 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002947 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948 * @buflen: buffer length
2949 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002950 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002952 * Returns a pointer into the buffer or an error code if the
2953 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002954 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002955 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002956 *
Al Viro02125a82011-12-05 08:43:34 -05002957 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958 */
Al Viro02125a82011-12-05 08:43:34 -05002959char *__d_path(const struct path *path,
2960 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002961 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002963 char *res = buf + buflen;
2964 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002966 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002967 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002968
Al Viro02125a82011-12-05 08:43:34 -05002969 if (error < 0)
2970 return ERR_PTR(error);
2971 if (error > 0)
2972 return NULL;
2973 return res;
2974}
2975
2976char *d_absolute_path(const struct path *path,
2977 char *buf, int buflen)
2978{
2979 struct path root = {};
2980 char *res = buf + buflen;
2981 int error;
2982
2983 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002984 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05002985
2986 if (error > 1)
2987 error = -EINVAL;
2988 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002989 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002990 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991}
2992
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002993/*
2994 * same as __d_path but appends "(deleted)" for unlinked files.
2995 */
Al Viro02125a82011-12-05 08:43:34 -05002996static int path_with_deleted(const struct path *path,
2997 const struct path *root,
2998 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002999{
3000 prepend(buf, buflen, "\0", 1);
3001 if (d_unlinked(path->dentry)) {
3002 int error = prepend(buf, buflen, " (deleted)", 10);
3003 if (error)
3004 return error;
3005 }
3006
3007 return prepend_path(path, root, buf, buflen);
3008}
3009
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003010static int prepend_unreachable(char **buffer, int *buflen)
3011{
3012 return prepend(buffer, buflen, "(unreachable)", 13);
3013}
3014
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003015static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3016{
3017 unsigned seq;
3018
3019 do {
3020 seq = read_seqcount_begin(&fs->seq);
3021 *root = fs->root;
3022 } while (read_seqcount_retry(&fs->seq, seq));
3023}
3024
Jan Bluncka03a8a702008-02-14 19:38:32 -08003025/**
3026 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003027 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003028 * @buf: buffer to return value in
3029 * @buflen: buffer length
3030 *
3031 * Convert a dentry into an ASCII path name. If the entry has been deleted
3032 * the string " (deleted)" is appended. Note that this is ambiguous.
3033 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003034 * Returns a pointer into the buffer or an error code if the path was
3035 * too long. Note: Callers should use the returned pointer, not the passed
3036 * in buffer, to use the name! The implementation often starts at an offset
3037 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003038 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003039 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003040 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003041char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003043 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003044 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003045 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003047 /*
3048 * We have various synthetic filesystems that never get mounted. On
3049 * these filesystems dentries are never used for lookup purposes, and
3050 * thus don't need to be hashed. They also don't need a name until a
3051 * user wants to identify the object in /proc/pid/fd/. The little hack
3052 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003053 *
3054 * Some pseudo inodes are mountable. When they are mounted
3055 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3056 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003057 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003058 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3059 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003060 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003061
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003062 rcu_read_lock();
3063 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003064 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003065 rcu_read_unlock();
3066
Al Viro02125a82011-12-05 08:43:34 -05003067 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003068 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 return res;
3070}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003071EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072
3073/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003074 * Helper function for dentry_operations.d_dname() members
3075 */
3076char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3077 const char *fmt, ...)
3078{
3079 va_list args;
3080 char temp[64];
3081 int sz;
3082
3083 va_start(args, fmt);
3084 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3085 va_end(args);
3086
3087 if (sz > sizeof(temp) || sz > buflen)
3088 return ERR_PTR(-ENAMETOOLONG);
3089
3090 buffer += buflen - sz;
3091 return memcpy(buffer, temp, sz);
3092}
3093
Al Viro118b2302013-08-24 12:08:17 -04003094char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3095{
3096 char *end = buffer + buflen;
3097 /* these dentries are never renamed, so d_lock is not needed */
3098 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003099 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003100 prepend(&end, &buflen, "/", 1))
3101 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003102 return end;
Al Viro118b2302013-08-24 12:08:17 -04003103}
David Herrmann31bbe162014-01-03 14:09:47 +01003104EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003105
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003106/*
Ram Pai6092d042008-03-27 13:06:20 +01003107 * Write full pathname from the root of the filesystem into the buffer.
3108 */
Al Virof6500802014-01-26 12:37:55 -05003109static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003110{
Al Virof6500802014-01-26 12:37:55 -05003111 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003112 char *end, *retval;
3113 int len, seq = 0;
3114 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003115
Al Virof6500802014-01-26 12:37:55 -05003116 if (buflen < 2)
3117 goto Elong;
3118
Al Viro48f5ec22013-09-09 15:22:25 -04003119 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003120restart:
Al Virof6500802014-01-26 12:37:55 -05003121 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003122 end = buf + buflen;
3123 len = buflen;
3124 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003125 /* Get '/' right */
3126 retval = end-1;
3127 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003128 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003129 while (!IS_ROOT(dentry)) {
3130 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003131
Ram Pai6092d042008-03-27 13:06:20 +01003132 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003133 error = prepend_name(&end, &len, &dentry->d_name);
3134 if (error)
3135 break;
Ram Pai6092d042008-03-27 13:06:20 +01003136
3137 retval = end;
3138 dentry = parent;
3139 }
Al Viro48f5ec22013-09-09 15:22:25 -04003140 if (!(seq & 1))
3141 rcu_read_unlock();
3142 if (need_seqretry(&rename_lock, seq)) {
3143 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003144 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003145 }
3146 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003147 if (error)
3148 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003149 return retval;
3150Elong:
3151 return ERR_PTR(-ENAMETOOLONG);
3152}
Nick Pigginec2447c2011-01-07 17:49:29 +11003153
3154char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3155{
Waiman Long232d2d62013-09-09 12:18:13 -04003156 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003157}
3158EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003159
3160char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3161{
3162 char *p = NULL;
3163 char *retval;
3164
Al Viroc1031352010-06-06 22:31:14 -04003165 if (d_unlinked(dentry)) {
3166 p = buf + buflen;
3167 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3168 goto Elong;
3169 buflen++;
3170 }
3171 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003172 if (!IS_ERR(retval) && p)
3173 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003174 return retval;
3175Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003176 return ERR_PTR(-ENAMETOOLONG);
3177}
3178
Linus Torvalds8b19e342013-09-12 10:35:47 -07003179static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3180 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003181{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003182 unsigned seq;
3183
3184 do {
3185 seq = read_seqcount_begin(&fs->seq);
3186 *root = fs->root;
3187 *pwd = fs->pwd;
3188 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003189}
3190
Ram Pai6092d042008-03-27 13:06:20 +01003191/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 * NOTE! The user-level library version returns a
3193 * character pointer. The kernel system call just
3194 * returns the length of the buffer filled (which
3195 * includes the ending '\0' character), or a negative
3196 * error value. So libc would do something like
3197 *
3198 * char *getcwd(char * buf, size_t size)
3199 * {
3200 * int retval;
3201 *
3202 * retval = sys_getcwd(buf, size);
3203 * if (retval >= 0)
3204 * return buf;
3205 * errno = -retval;
3206 * return NULL;
3207 * }
3208 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003209SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210{
Linus Torvalds552ce542007-02-13 12:08:18 -08003211 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003212 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003213 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214
3215 if (!page)
3216 return -ENOMEM;
3217
Linus Torvalds8b19e342013-09-12 10:35:47 -07003218 rcu_read_lock();
3219 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220
Linus Torvalds552ce542007-02-13 12:08:18 -08003221 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003222 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003223 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003224 char *cwd = page + PATH_MAX;
3225 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003227 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003228 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003229 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003230
Al Viro02125a82011-12-05 08:43:34 -05003231 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003232 goto out;
3233
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003234 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003235 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003236 error = prepend_unreachable(&cwd, &buflen);
3237 if (error)
3238 goto out;
3239 }
3240
Linus Torvalds552ce542007-02-13 12:08:18 -08003241 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003242 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003243 if (len <= size) {
3244 error = len;
3245 if (copy_to_user(buf, cwd, len))
3246 error = -EFAULT;
3247 }
Nick Piggin949854d2011-01-07 17:49:37 +11003248 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003249 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251
3252out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003253 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254 return error;
3255}
3256
3257/*
3258 * Test whether new_dentry is a subdirectory of old_dentry.
3259 *
3260 * Trivially implemented using the dcache structure
3261 */
3262
3263/**
3264 * is_subdir - is new dentry a subdirectory of old_dentry
3265 * @new_dentry: new dentry
3266 * @old_dentry: old dentry
3267 *
3268 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3269 * Returns 0 otherwise.
3270 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3271 */
3272
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003273int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274{
3275 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11003276 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003278 if (new_dentry == old_dentry)
3279 return 1;
3280
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003281 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003284 /*
3285 * Need rcu_readlock to protect against the d_parent trashing
3286 * due to d_move
3287 */
3288 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003289 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003290 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003291 else
3292 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11003293 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295
3296 return result;
3297}
3298
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003299static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003301 struct dentry *root = data;
3302 if (dentry != root) {
3303 if (d_unhashed(dentry) || !dentry->d_inode)
3304 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003306 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3307 dentry->d_flags |= DCACHE_GENOCIDE;
3308 dentry->d_lockref.count--;
3309 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003311 return D_WALK_CONTINUE;
3312}
Nick Piggin58db63d2011-01-07 17:49:39 +11003313
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003314void d_genocide(struct dentry *parent)
3315{
3316 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317}
3318
Al Viro60545d02013-06-07 01:20:27 -04003319void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320{
Al Viro60545d02013-06-07 01:20:27 -04003321 inode_dec_link_count(inode);
3322 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3323 !hlist_unhashed(&dentry->d_alias) ||
3324 !d_unlinked(dentry));
3325 spin_lock(&dentry->d_parent->d_lock);
3326 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3327 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3328 (unsigned long long)inode->i_ino);
3329 spin_unlock(&dentry->d_lock);
3330 spin_unlock(&dentry->d_parent->d_lock);
3331 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332}
Al Viro60545d02013-06-07 01:20:27 -04003333EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334
3335static __initdata unsigned long dhash_entries;
3336static int __init set_dhash_entries(char *str)
3337{
3338 if (!str)
3339 return 0;
3340 dhash_entries = simple_strtoul(str, &str, 0);
3341 return 1;
3342}
3343__setup("dhash_entries=", set_dhash_entries);
3344
3345static void __init dcache_init_early(void)
3346{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003347 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348
3349 /* If hashes are distributed across NUMA nodes, defer
3350 * hash allocation until vmalloc space is available.
3351 */
3352 if (hashdist)
3353 return;
3354
3355 dentry_hashtable =
3356 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003357 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358 dhash_entries,
3359 13,
3360 HASH_EARLY,
3361 &d_hash_shift,
3362 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003363 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 0);
3365
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003366 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003367 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368}
3369
Denis Cheng74bf17c2007-10-16 23:26:30 -07003370static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003372 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373
3374 /*
3375 * A constructor could be added for stable state like the lists,
3376 * but it is probably not worth it because of the cache nature
3377 * of the dcache.
3378 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003379 dentry_cache = KMEM_CACHE(dentry,
3380 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381
3382 /* Hash may have been set up in dcache_init_early */
3383 if (!hashdist)
3384 return;
3385
3386 dentry_hashtable =
3387 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003388 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389 dhash_entries,
3390 13,
3391 0,
3392 &d_hash_shift,
3393 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003394 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395 0);
3396
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003397 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003398 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399}
3400
3401/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003402struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003403EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003404
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405EXPORT_SYMBOL(d_genocide);
3406
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407void __init vfs_caches_init_early(void)
3408{
3409 dcache_init_early();
3410 inode_init_early();
3411}
3412
3413void __init vfs_caches_init(unsigned long mempages)
3414{
3415 unsigned long reserve;
3416
3417 /* Base hash sizes on available memory, with a reserve equal to
3418 150% of current kernel size */
3419
3420 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3421 mempages -= reserve;
3422
3423 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003424 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003425
Denis Cheng74bf17c2007-10-16 23:26:30 -07003426 dcache_init();
3427 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003429 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430 bdev_cache_init();
3431 chrdev_init();
3432}