blob: cb4a10690868263cb58e28012ee8f9abc4880da5 [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 Howells07f3f05c2006-09-30 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;
Al Viro482db902013-10-25 16:41:01 -0400109 hash = hash + (hash >> d_hash_shift);
110 return dentry_hashtable + (hash & d_hash_mask);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100111}
112
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113/* Statistics gathering. */
114struct dentry_stat_t dentry_stat = {
115 .age_limit = 45,
116};
117
Glauber Costa3942c072013-08-28 10:17:53 +1000118static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000119static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400120
121#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000122
123/*
124 * Here we resort to our own counters instead of using generic per-cpu counters
125 * for consistency with what the vfs inode code does. We are expected to harvest
126 * better code and performance by having our own specialized counters.
127 *
128 * Please note that the loop is done over all possible CPUs, not over all online
129 * CPUs. The reason for this is that we don't want to play games with CPUs going
130 * on and off. If one of them goes off, we will just keep their counters.
131 *
132 * glommer: See cffbc8a for details, and if you ever intend to change this,
133 * please update all vfs counters to match.
134 */
Glauber Costa3942c072013-08-28 10:17:53 +1000135static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100136{
137 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000138 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100139 for_each_possible_cpu(i)
140 sum += per_cpu(nr_dentry, i);
141 return sum < 0 ? 0 : sum;
142}
143
Dave Chinner62d36c72013-08-28 10:17:54 +1000144static long get_nr_dentry_unused(void)
145{
146 int i;
147 long sum = 0;
148 for_each_possible_cpu(i)
149 sum += per_cpu(nr_dentry_unused, i);
150 return sum < 0 ? 0 : sum;
151}
152
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400153int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
154 size_t *lenp, loff_t *ppos)
155{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100156 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000157 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000158 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400159}
160#endif
161
Linus Torvalds5483f182012-03-04 15:51:42 -0800162/*
163 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
164 * The strings are both count bytes long, and count is non-zero.
165 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700166#ifdef CONFIG_DCACHE_WORD_ACCESS
167
168#include <asm/word-at-a-time.h>
169/*
170 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
171 * aligned allocation for this particular component. We don't
172 * strictly need the load_unaligned_zeropad() safety, but it
173 * doesn't hurt either.
174 *
175 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
176 * need the careful unaligned handling.
177 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700178static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800179{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800180 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800181
182 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700183 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700184 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800185 if (tcount < sizeof(unsigned long))
186 break;
187 if (unlikely(a != b))
188 return 1;
189 cs += sizeof(unsigned long);
190 ct += sizeof(unsigned long);
191 tcount -= sizeof(unsigned long);
192 if (!tcount)
193 return 0;
194 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000195 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800196 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700197}
198
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800199#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700200
Linus Torvalds94753db52012-05-10 12:19:19 -0700201static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700202{
Linus Torvalds5483f182012-03-04 15:51:42 -0800203 do {
204 if (*cs != *ct)
205 return 1;
206 cs++;
207 ct++;
208 tcount--;
209 } while (tcount);
210 return 0;
211}
212
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700213#endif
214
Linus Torvalds94753db52012-05-10 12:19:19 -0700215static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
216{
Linus Torvalds6326c712012-05-21 16:14:04 -0700217 const unsigned char *cs;
Linus Torvalds94753db52012-05-10 12:19:19 -0700218 /*
219 * Be careful about RCU walk racing with rename:
220 * use ACCESS_ONCE to fetch the name pointer.
221 *
222 * NOTE! Even if a rename will mean that the length
223 * was not loaded atomically, we don't care. The
224 * RCU walk will check the sequence count eventually,
225 * and catch it. And we won't overrun the buffer,
226 * because we're reading the name pointer atomically,
227 * and a dentry name is guaranteed to be properly
228 * terminated with a NUL byte.
229 *
230 * End result: even if 'len' is wrong, we'll exit
231 * early because the data cannot match (there can
232 * be no NUL in the ct/tcount data)
233 */
Linus Torvalds6326c712012-05-21 16:14:04 -0700234 cs = ACCESS_ONCE(dentry->d_name.name);
235 smp_read_barrier_depends();
236 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700237}
238
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400239static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400241 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
242
Al Virob3d9b7a2012-06-09 13:51:19 -0400243 WARN_ON(!hlist_unhashed(&dentry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 if (dname_external(dentry))
245 kfree(dentry->d_name.name);
246 kmem_cache_free(dentry_cache, dentry);
247}
248
249/*
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100250 * no locks, please.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 */
252static void d_free(struct dentry *dentry)
253{
Linus Torvalds0d984392013-09-08 13:46:52 -0700254 BUG_ON((int)dentry->d_lockref.count > 0);
Nick Piggin3e880fb2011-01-07 17:49:19 +1100255 this_cpu_dec(nr_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 if (dentry->d_op && dentry->d_op->d_release)
257 dentry->d_op->d_release(dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400258
Linus Torvaldsdea36672011-04-24 07:58:46 -0700259 /* if dentry was never visible to RCU, immediate free is OK */
260 if (!(dentry->d_flags & DCACHE_RCUACCESS))
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400261 __d_free(&dentry->d_u.d_rcu);
Eric Dumazetb3423412006-12-06 20:38:48 -0800262 else
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400263 call_rcu(&dentry->d_u.d_rcu, __d_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
265
Nick Piggin31e6b012011-01-07 17:49:52 +1100266/**
267 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800268 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100269 * After this call, in-progress rcu-walk path lookup will fail. This
270 * should be called after unhashing, and after changing d_inode (if
271 * the dentry has not already been unhashed).
272 */
273static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
274{
275 assert_spin_locked(&dentry->d_lock);
276 /* Go through a barrier */
277 write_seqcount_barrier(&dentry->d_seq);
278}
279
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280/*
281 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100282 * d_iput() operation if defined. Dentry has no refcount
283 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800285static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200286 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100287 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288{
289 struct inode *inode = dentry->d_inode;
290 if (inode) {
291 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400292 hlist_del_init(&dentry->d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100294 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700295 if (!inode->i_nlink)
296 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 if (dentry->d_op && dentry->d_op->d_iput)
298 dentry->d_op->d_iput(dentry, inode);
299 else
300 iput(inode);
301 } else {
302 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 }
304}
305
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700306/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100307 * Release the dentry's inode, using the filesystem
308 * d_iput() operation if defined. dentry remains in-use.
309 */
310static void dentry_unlink_inode(struct dentry * dentry)
311 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100312 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100313{
314 struct inode *inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +0100315 __d_clear_type(dentry);
Nick Piggin31e6b012011-01-07 17:49:52 +1100316 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400317 hlist_del_init(&dentry->d_alias);
Nick Piggin31e6b012011-01-07 17:49:52 +1100318 dentry_rcuwalk_barrier(dentry);
319 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100320 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100321 if (!inode->i_nlink)
322 fsnotify_inoderemove(inode);
323 if (dentry->d_op && dentry->d_op->d_iput)
324 dentry->d_op->d_iput(dentry, inode);
325 else
326 iput(inode);
327}
328
329/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400330 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
331 * is in use - which includes both the "real" per-superblock
332 * LRU list _and_ the DCACHE_SHRINK_LIST use.
333 *
334 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
335 * on the shrink list (ie not on the superblock LRU list).
336 *
337 * The per-cpu "nr_dentry_unused" counters are updated with
338 * the DCACHE_LRU_LIST bit.
339 *
340 * These helper functions make sure we always follow the
341 * rules. d_lock must be held by the caller.
342 */
343#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
344static void d_lru_add(struct dentry *dentry)
345{
346 D_FLAG_VERIFY(dentry, 0);
347 dentry->d_flags |= DCACHE_LRU_LIST;
348 this_cpu_inc(nr_dentry_unused);
349 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
350}
351
352static void d_lru_del(struct dentry *dentry)
353{
354 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
355 dentry->d_flags &= ~DCACHE_LRU_LIST;
356 this_cpu_dec(nr_dentry_unused);
357 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
358}
359
360static void d_shrink_del(struct dentry *dentry)
361{
362 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
363 list_del_init(&dentry->d_lru);
364 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
365 this_cpu_dec(nr_dentry_unused);
366}
367
368static void d_shrink_add(struct dentry *dentry, struct list_head *list)
369{
370 D_FLAG_VERIFY(dentry, 0);
371 list_add(&dentry->d_lru, list);
372 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
373 this_cpu_inc(nr_dentry_unused);
374}
375
376/*
377 * These can only be called under the global LRU lock, ie during the
378 * callback for freeing the LRU list. "isolate" removes it from the
379 * LRU lists entirely, while shrink_move moves it to the indicated
380 * private list.
381 */
382static void d_lru_isolate(struct dentry *dentry)
383{
384 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
385 dentry->d_flags &= ~DCACHE_LRU_LIST;
386 this_cpu_dec(nr_dentry_unused);
387 list_del_init(&dentry->d_lru);
388}
389
390static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
391{
392 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
393 dentry->d_flags |= DCACHE_SHRINK_LIST;
394 list_move_tail(&dentry->d_lru, list);
395}
396
397/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000398 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700399 */
400static void dentry_lru_add(struct dentry *dentry)
401{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400402 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
403 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700404}
405
Sage Weilf0023bc2011-10-28 10:02:42 -0700406/*
407 * Remove a dentry with references from the LRU.
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000408 *
409 * If we are on the shrink list, then we can get to try_prune_one_dentry() and
410 * lose our last reference through the parent walk. In this case, we need to
411 * remove ourselves from the shrink list, not the LRU.
Sage Weilf0023bc2011-10-28 10:02:42 -0700412 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700413static void dentry_lru_del(struct dentry *dentry)
414{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400415 if (dentry->d_flags & DCACHE_LRU_LIST) {
416 if (dentry->d_flags & DCACHE_SHRINK_LIST)
417 return d_shrink_del(dentry);
418 d_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700419 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700420}
421
Miklos Szeredid52b9082007-05-08 00:23:46 -0700422/**
423 * d_kill - kill dentry and return parent
424 * @dentry: dentry to kill
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800425 * @parent: parent dentry
Miklos Szeredid52b9082007-05-08 00:23:46 -0700426 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200427 * The dentry must already be unhashed and removed from the LRU.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700428 *
429 * If this is the root of the dentry tree, return NULL.
Nick Piggin23044502011-01-07 17:49:31 +1100430 *
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100431 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
432 * d_kill.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700433 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100434static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200435 __releases(dentry->d_lock)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100436 __releases(parent->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100437 __releases(dentry->d_inode->i_lock)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700438{
Miklos Szeredid52b9082007-05-08 00:23:46 -0700439 list_del(&dentry->d_u.d_child);
Trond Myklebustc83ce982011-03-15 13:36:43 -0400440 /*
Al Viro31dec132013-10-25 17:04:27 -0400441 * Inform d_walk() that we are no longer attached to the
Trond Myklebustc83ce982011-03-15 13:36:43 -0400442 * dentry tree
443 */
Miklos Szeredib161dfa62012-09-17 22:31:38 +0200444 dentry->d_flags |= DCACHE_DENTRY_KILLED;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100445 if (parent)
446 spin_unlock(&parent->d_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700447 dentry_iput(dentry);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100448 /*
449 * dentry_iput drops the locks, at which point nobody (except
450 * transient RCU lookups) can reach this dentry.
451 */
Miklos Szeredid52b9082007-05-08 00:23:46 -0700452 d_free(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900453 return parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700454}
455
Nick Piggin789680d2011-01-07 17:49:30 +1100456/**
457 * d_drop - drop a dentry
458 * @dentry: dentry to drop
459 *
460 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
461 * be found through a VFS lookup any more. Note that this is different from
462 * deleting the dentry - d_delete will try to mark the dentry negative if
463 * possible, giving a successful _negative_ lookup, while d_drop will
464 * just make the cache lookup fail.
465 *
466 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
467 * reason (NFS timeouts or autofs deletes).
468 *
469 * __d_drop requires dentry->d_lock.
470 */
471void __d_drop(struct dentry *dentry)
472{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700473 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400474 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400475 /*
476 * Hashed dentries are normally on the dentry hashtable,
477 * with the exception of those newly allocated by
478 * d_obtain_alias, which are always IS_ROOT:
479 */
480 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400481 b = &dentry->d_sb->s_anon;
482 else
483 b = d_hash(dentry->d_parent, dentry->d_name.hash);
484
485 hlist_bl_lock(b);
486 __hlist_bl_del(&dentry->d_hash);
487 dentry->d_hash.pprev = NULL;
488 hlist_bl_unlock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700489 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100490 }
491}
492EXPORT_SYMBOL(__d_drop);
493
494void d_drop(struct dentry *dentry)
495{
Nick Piggin789680d2011-01-07 17:49:30 +1100496 spin_lock(&dentry->d_lock);
497 __d_drop(dentry);
498 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100499}
500EXPORT_SYMBOL(d_drop);
501
Nick Piggin77812a12011-01-07 17:49:48 +1100502/*
503 * Finish off a dentry we've decided to kill.
504 * dentry->d_lock must be held, returns with it unlocked.
505 * If ref is non-zero, then decrement the refcount too.
506 * Returns dentry requiring refcount drop, or NULL if we're done.
507 */
Linus Torvalds358eec12013-10-31 15:43:02 -0700508static struct dentry *
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000509dentry_kill(struct dentry *dentry, int unlock_on_failure)
Nick Piggin77812a12011-01-07 17:49:48 +1100510 __releases(dentry->d_lock)
511{
Nick Piggin873feea2011-01-07 17:50:06 +1100512 struct inode *inode;
Nick Piggin77812a12011-01-07 17:49:48 +1100513 struct dentry *parent;
514
Nick Piggin873feea2011-01-07 17:50:06 +1100515 inode = dentry->d_inode;
516 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin77812a12011-01-07 17:49:48 +1100517relock:
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000518 if (unlock_on_failure) {
519 spin_unlock(&dentry->d_lock);
520 cpu_relax();
521 }
Nick Piggin77812a12011-01-07 17:49:48 +1100522 return dentry; /* try again with same dentry */
523 }
524 if (IS_ROOT(dentry))
525 parent = NULL;
526 else
527 parent = dentry->d_parent;
528 if (parent && !spin_trylock(&parent->d_lock)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100529 if (inode)
530 spin_unlock(&inode->i_lock);
Nick Piggin77812a12011-01-07 17:49:48 +1100531 goto relock;
532 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100533
Linus Torvalds0d984392013-09-08 13:46:52 -0700534 /*
535 * The dentry is now unrecoverably dead to the world.
536 */
537 lockref_mark_dead(&dentry->d_lockref);
538
Sage Weilf0023bc2011-10-28 10:02:42 -0700539 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700540 * inform the fs via d_prune that this dentry is about to be
541 * unhashed and destroyed.
542 */
Yan, Zheng590fb512013-08-13 15:42:02 +0800543 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
Yan, Zheng61572bb2013-04-15 14:13:21 +0800544 dentry->d_op->d_prune(dentry);
545
546 dentry_lru_del(dentry);
Nick Piggin77812a12011-01-07 17:49:48 +1100547 /* if it was on the hash then remove it */
548 __d_drop(dentry);
549 return d_kill(dentry, parent);
550}
551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552/*
553 * This is dput
554 *
555 * This is complicated by the fact that we do not want to put
556 * dentries that are no longer on any hash chain on the unused
557 * list: we'd much rather just get rid of them immediately.
558 *
559 * However, that implies that we have to traverse the dentry
560 * tree upwards to the parents which might _also_ now be
561 * scheduled for deletion (it may have been only waiting for
562 * its last child to go away).
563 *
564 * This tail recursion is done by hand as we don't want to depend
565 * on the compiler to always get this right (gcc generally doesn't).
566 * Real recursion would eat up our stack space.
567 */
568
569/*
570 * dput - release a dentry
571 * @dentry: dentry to release
572 *
573 * Release a dentry. This will drop the usage count and if appropriate
574 * call the dentry unlink method as well as removing it from the queues and
575 * releasing its resources. If the parent dentries were scheduled for release
576 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578void dput(struct dentry *dentry)
579{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700580 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 return;
582
583repeat:
Waiman Long98474232013-08-28 18:24:59 -0700584 if (lockref_put_or_lock(&dentry->d_lockref))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700587 /* Unreachable? Get rid of it */
588 if (unlikely(d_unhashed(dentry)))
589 goto kill_it;
590
591 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100593 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
Nick Piggin265ac902010-10-10 05:36:24 -0400595
Linus Torvalds358eec12013-10-31 15:43:02 -0700596 if (!(dentry->d_flags & DCACHE_REFERENCED))
597 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400598 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400599
Waiman Long98474232013-08-28 18:24:59 -0700600 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100601 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 return;
603
Miklos Szeredid52b9082007-05-08 00:23:46 -0700604kill_it:
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000605 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700606 if (dentry)
607 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700609EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
611/**
612 * d_invalidate - invalidate a dentry
613 * @dentry: dentry to invalidate
614 *
615 * Try to invalidate the dentry if it turns out to be
616 * possible. If there are other dentries that can be
617 * reached through this one we can't delete it and we
618 * return -EBUSY. On success we return 0.
619 *
620 * no dcache lock.
621 */
622
623int d_invalidate(struct dentry * dentry)
624{
625 /*
626 * If it's already been dropped, return OK.
627 */
Nick Pigginda502952011-01-07 17:49:33 +1100628 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100630 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 return 0;
632 }
633 /*
634 * Check whether to do a partial shrink_dcache
635 * to get rid of unused child entries.
636 */
637 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100638 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100640 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 }
642
643 /*
644 * Somebody else still using it?
645 *
646 * If it's a directory, we can't drop it
647 * for fear of somebody re-populating it
648 * with children (even though dropping it
649 * would make it unreachable from the root,
650 * we might still populate it if it was a
651 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000652 * We also need to leave mountpoints alone,
653 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 */
Waiman Long98474232013-08-28 18:24:59 -0700655 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
Al Viro50e69632011-11-07 16:39:57 +0000656 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 return -EBUSY;
659 }
660 }
661
662 __d_drop(dentry);
663 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 return 0;
665}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700666EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100668/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100669static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670{
Waiman Long98474232013-08-28 18:24:59 -0700671 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
673
Nick Piggindc0474b2011-01-07 17:49:43 +1100674static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100675{
Waiman Long98474232013-08-28 18:24:59 -0700676 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100677}
678
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100679struct dentry *dget_parent(struct dentry *dentry)
680{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700681 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100682 struct dentry *ret;
683
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700684 /*
685 * Do optimistic parent lookup without any
686 * locking.
687 */
688 rcu_read_lock();
689 ret = ACCESS_ONCE(dentry->d_parent);
690 gotref = lockref_get_not_zero(&ret->d_lockref);
691 rcu_read_unlock();
692 if (likely(gotref)) {
693 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
694 return ret;
695 dput(ret);
696 }
697
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100698repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100699 /*
700 * Don't need rcu_dereference because we re-check it was correct under
701 * the lock.
702 */
703 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100704 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100705 spin_lock(&ret->d_lock);
706 if (unlikely(ret != dentry->d_parent)) {
707 spin_unlock(&ret->d_lock);
708 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100709 goto repeat;
710 }
Nick Piggina734eb42011-01-07 17:49:44 +1100711 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700712 BUG_ON(!ret->d_lockref.count);
713 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100714 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100715 return ret;
716}
717EXPORT_SYMBOL(dget_parent);
718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719/**
720 * d_find_alias - grab a hashed alias of inode
721 * @inode: inode in question
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700722 * @want_discon: flag, used by d_splice_alias, to request
723 * that only a DISCONNECTED alias be returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 *
725 * If inode has a hashed alias, or is a directory and has any alias,
726 * acquire the reference to alias and return it. Otherwise return NULL.
727 * Notice that if inode is a directory there can be only one alias and
728 * it can be unhashed only if it has no children, or if it is the root
729 * of a filesystem.
730 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700731 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700732 * any other hashed alias over that one unless @want_discon is set,
733 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 */
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700735static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736{
Nick Pigginda502952011-01-07 17:49:33 +1100737 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
Nick Pigginda502952011-01-07 17:49:33 +1100739again:
740 discon_alias = NULL;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800741 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100742 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700744 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100745 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 discon_alias = alias;
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700747 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100748 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100749 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 return alias;
751 }
752 }
Nick Pigginda502952011-01-07 17:49:33 +1100753 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 }
Nick Pigginda502952011-01-07 17:49:33 +1100755 if (discon_alias) {
756 alias = discon_alias;
757 spin_lock(&alias->d_lock);
758 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
759 if (IS_ROOT(alias) &&
760 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100761 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100762 spin_unlock(&alias->d_lock);
763 return alias;
764 }
765 }
766 spin_unlock(&alias->d_lock);
767 goto again;
768 }
769 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770}
771
Nick Pigginda502952011-01-07 17:49:33 +1100772struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773{
David Howells214fda12006-03-25 03:06:36 -0800774 struct dentry *de = NULL;
775
Al Virob3d9b7a2012-06-09 13:51:19 -0400776 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100777 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700778 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100779 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800780 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 return de;
782}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700783EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
785/*
786 * Try to kill dentries associated with this inode.
787 * WARNING: you must own a reference to inode.
788 */
789void d_prune_aliases(struct inode *inode)
790{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700791 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100793 spin_lock(&inode->i_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800794 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700796 if (!dentry->d_lockref.count) {
Yan, Zheng590fb512013-08-13 15:42:02 +0800797 /*
798 * inform the fs via d_prune that this dentry
799 * is about to be unhashed and destroyed.
800 */
801 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
802 !d_unhashed(dentry))
803 dentry->d_op->d_prune(dentry);
804
Nick Piggindc0474b2011-01-07 17:49:43 +1100805 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 __d_drop(dentry);
807 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100808 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 dput(dentry);
810 goto restart;
811 }
812 spin_unlock(&dentry->d_lock);
813 }
Nick Piggin873feea2011-01-07 17:50:06 +1100814 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700816EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
818/*
Nick Piggin77812a12011-01-07 17:49:48 +1100819 * Try to throw away a dentry - free the inode, dput the parent.
820 * Requires dentry->d_lock is held, and dentry->d_count == 0.
821 * Releases dentry->d_lock.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700822 *
Nick Piggin77812a12011-01-07 17:49:48 +1100823 * This may fail if locks cannot be acquired no problem, just try again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 */
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000825static struct dentry * try_prune_one_dentry(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200826 __releases(dentry->d_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827{
Nick Piggin77812a12011-01-07 17:49:48 +1100828 struct dentry *parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700829
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000830 parent = dentry_kill(dentry, 0);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700831 /*
Nick Piggin77812a12011-01-07 17:49:48 +1100832 * If dentry_kill returns NULL, we have nothing more to do.
833 * if it returns the same dentry, trylocks failed. In either
834 * case, just loop again.
835 *
836 * Otherwise, we need to prune ancestors too. This is necessary
837 * to prevent quadratic behavior of shrink_dcache_parent(), but
838 * is also expected to be beneficial in reducing dentry cache
839 * fragmentation.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700840 */
Nick Piggin77812a12011-01-07 17:49:48 +1100841 if (!parent)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000842 return NULL;
Nick Piggin77812a12011-01-07 17:49:48 +1100843 if (parent == dentry)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000844 return dentry;
Nick Piggin77812a12011-01-07 17:49:48 +1100845
846 /* Prune ancestors. */
847 dentry = parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700848 while (dentry) {
Waiman Long98474232013-08-28 18:24:59 -0700849 if (lockref_put_or_lock(&dentry->d_lockref))
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000850 return NULL;
851 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700852 }
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000853 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854}
855
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400856static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700858 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700859
Nick Pigginec336792011-01-07 17:49:47 +1100860 rcu_read_lock();
861 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100862 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
863 if (&dentry->d_lru == list)
864 break; /* empty */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400865
866 /*
867 * Get the dentry lock, and re-verify that the dentry is
868 * this on the shrinking list. If it is, we know that
869 * DCACHE_SHRINK_LIST and DCACHE_LRU_LIST are set.
870 */
Nick Pigginec336792011-01-07 17:49:47 +1100871 spin_lock(&dentry->d_lock);
872 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
873 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100874 continue;
875 }
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000878 * The dispose list is isolated and dentries are not accounted
879 * to the LRU here, so we can simply remove it from the list
880 * here regardless of whether it is referenced or not.
881 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400882 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000883
884 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000886 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 */
Waiman Long98474232013-08-28 18:24:59 -0700888 if (dentry->d_lockref.count) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700889 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 continue;
891 }
Nick Pigginec336792011-01-07 17:49:47 +1100892 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100893
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400894 /*
895 * If 'try_to_prune()' returns a dentry, it will
896 * be the same one we passed in, and d_lock will
897 * have been held the whole time, so it will not
898 * have been added to any other lists. We failed
899 * to get the inode lock.
900 *
901 * We just add it back to the shrink list.
902 */
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000903 dentry = try_prune_one_dentry(dentry);
Nick Piggin77812a12011-01-07 17:49:48 +1100904
Nick Pigginec336792011-01-07 17:49:47 +1100905 rcu_read_lock();
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000906 if (dentry) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400907 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000908 spin_unlock(&dentry->d_lock);
909 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 }
Nick Pigginec336792011-01-07 17:49:47 +1100911 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400912}
913
Dave Chinnerf6041562013-08-28 10:18:00 +1000914static enum lru_status
915dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
916{
917 struct list_head *freeable = arg;
918 struct dentry *dentry = container_of(item, struct dentry, d_lru);
919
920
921 /*
922 * we are inverting the lru lock/dentry->d_lock here,
923 * so use a trylock. If we fail to get the lock, just skip
924 * it
925 */
926 if (!spin_trylock(&dentry->d_lock))
927 return LRU_SKIP;
928
929 /*
930 * Referenced dentries are still in use. If they have active
931 * counts, just remove them from the LRU. Otherwise give them
932 * another pass through the LRU.
933 */
934 if (dentry->d_lockref.count) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400935 d_lru_isolate(dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +1000936 spin_unlock(&dentry->d_lock);
937 return LRU_REMOVED;
938 }
939
940 if (dentry->d_flags & DCACHE_REFERENCED) {
941 dentry->d_flags &= ~DCACHE_REFERENCED;
942 spin_unlock(&dentry->d_lock);
943
944 /*
945 * The list move itself will be made by the common LRU code. At
946 * this point, we've dropped the dentry->d_lock but keep the
947 * lru lock. This is safe to do, since every list movement is
948 * protected by the lru lock even if both locks are held.
949 *
950 * This is guaranteed by the fact that all LRU management
951 * functions are intermediated by the LRU API calls like
952 * list_lru_add and list_lru_del. List movement in this file
953 * only ever occur through this functions or through callbacks
954 * like this one, that are called from the LRU API.
955 *
956 * The only exceptions to this are functions like
957 * shrink_dentry_list, and code that first checks for the
958 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
959 * operating only with stack provided lists after they are
960 * properly isolated from the main list. It is thus, always a
961 * local access.
962 */
963 return LRU_ROTATE;
964 }
965
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400966 d_lru_shrink_move(dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +1000967 spin_unlock(&dentry->d_lock);
968
969 return LRU_REMOVED;
970}
971
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400972/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000973 * prune_dcache_sb - shrink the dcache
974 * @sb: superblock
Dave Chinnerf6041562013-08-28 10:18:00 +1000975 * @nr_to_scan : number of entries to try to free
Dave Chinner9b17c622013-08-28 10:18:05 +1000976 * @nid: which node to scan for freeable entities
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400977 *
Dave Chinnerf6041562013-08-28 10:18:00 +1000978 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000979 * done when we need more memory an called from the superblock shrinker
980 * function.
981 *
982 * This function may fail to free any resources if all the dentries are in
983 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400984 */
Dave Chinner9b17c622013-08-28 10:18:05 +1000985long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
986 int nid)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400987{
Dave Chinnerf6041562013-08-28 10:18:00 +1000988 LIST_HEAD(dispose);
989 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400990
Dave Chinner9b17c622013-08-28 10:18:05 +1000991 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
992 &dispose, &nr_to_scan);
Dave Chinnerf6041562013-08-28 10:18:00 +1000993 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +1000994 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995}
996
Glauber Costa4e717f52013-08-28 10:18:03 +1000997static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
998 spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000999{
Glauber Costa4e717f52013-08-28 10:18:03 +10001000 struct list_head *freeable = arg;
1001 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001002
Glauber Costa4e717f52013-08-28 10:18:03 +10001003 /*
1004 * we are inverting the lru lock/dentry->d_lock here,
1005 * so use a trylock. If we fail to get the lock, just skip
1006 * it
1007 */
1008 if (!spin_trylock(&dentry->d_lock))
1009 return LRU_SKIP;
1010
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001011 d_lru_shrink_move(dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001012 spin_unlock(&dentry->d_lock);
1013
1014 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001015}
1016
Glauber Costa4e717f52013-08-28 10:18:03 +10001017
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001018/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 * shrink_dcache_sb - shrink dcache for a superblock
1020 * @sb: superblock
1021 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001022 * Shrink the dcache for the specified super block. This is used to free
1023 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001025void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026{
Glauber Costa4e717f52013-08-28 10:18:03 +10001027 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001028
Glauber Costa4e717f52013-08-28 10:18:03 +10001029 do {
1030 LIST_HEAD(dispose);
1031
1032 freed = list_lru_walk(&sb->s_dentry_lru,
1033 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1034
1035 this_cpu_sub(nr_dentry_unused, freed);
1036 shrink_dentry_list(&dispose);
1037 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001039EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001042 * enum d_walk_ret - action to talke during tree walk
1043 * @D_WALK_CONTINUE: contrinue walk
1044 * @D_WALK_QUIT: quit walk
1045 * @D_WALK_NORETRY: quit when retry is needed
1046 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001048enum d_walk_ret {
1049 D_WALK_CONTINUE,
1050 D_WALK_QUIT,
1051 D_WALK_NORETRY,
1052 D_WALK_SKIP,
1053};
1054
1055/**
1056 * d_walk - walk the dentry tree
1057 * @parent: start of walk
1058 * @data: data passed to @enter() and @finish()
1059 * @enter: callback when first entering the dentry
1060 * @finish: callback when successfully finished the walk
1061 *
1062 * The @enter() and @finish() callbacks are called with d_lock held.
1063 */
1064static void d_walk(struct dentry *parent, void *data,
1065 enum d_walk_ret (*enter)(void *, struct dentry *),
1066 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067{
Nick Piggin949854d2011-01-07 17:49:37 +11001068 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001070 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001071 enum d_walk_ret ret;
1072 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001073
Nick Piggin58db63d2011-01-07 17:49:39 +11001074again:
Al Viro48f5ec22013-09-09 15:22:25 -04001075 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001076 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001077 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001078
1079 ret = enter(data, this_parent);
1080 switch (ret) {
1081 case D_WALK_CONTINUE:
1082 break;
1083 case D_WALK_QUIT:
1084 case D_WALK_SKIP:
1085 goto out_unlock;
1086 case D_WALK_NORETRY:
1087 retry = false;
1088 break;
1089 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090repeat:
1091 next = this_parent->d_subdirs.next;
1092resume:
1093 while (next != &this_parent->d_subdirs) {
1094 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001095 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001097
1098 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001099
1100 ret = enter(data, dentry);
1101 switch (ret) {
1102 case D_WALK_CONTINUE:
1103 break;
1104 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001105 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001106 goto out_unlock;
1107 case D_WALK_NORETRY:
1108 retry = false;
1109 break;
1110 case D_WALK_SKIP:
1111 spin_unlock(&dentry->d_lock);
1112 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001113 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001114
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001116 spin_unlock(&this_parent->d_lock);
1117 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001119 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 goto repeat;
1121 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001122 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 }
1124 /*
1125 * All done at this level ... ascend and resume the search.
1126 */
1127 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001128 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001129 this_parent = child->d_parent;
1130
1131 rcu_read_lock();
1132 spin_unlock(&child->d_lock);
1133 spin_lock(&this_parent->d_lock);
1134
1135 /*
1136 * might go back up the wrong parent if we have had a rename
1137 * or deletion
1138 */
1139 if (this_parent != child->d_parent ||
1140 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1141 need_seqretry(&rename_lock, seq)) {
1142 spin_unlock(&this_parent->d_lock);
1143 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11001144 goto rename_retry;
Al Viro31dec132013-10-25 17:04:27 -04001145 }
1146 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11001147 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 goto resume;
1149 }
Al Viro48f5ec22013-09-09 15:22:25 -04001150 if (need_seqretry(&rename_lock, seq)) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001151 spin_unlock(&this_parent->d_lock);
1152 goto rename_retry;
1153 }
1154 if (finish)
1155 finish(data);
1156
1157out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001158 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001159 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001160 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001161
1162rename_retry:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001163 if (!retry)
1164 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001165 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001166 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001168
1169/*
1170 * Search for at least 1 mount point in the dentry's subdirs.
1171 * We descend to the next level whenever the d_subdirs
1172 * list is non-empty and continue searching.
1173 */
1174
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001175static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1176{
1177 int *ret = data;
1178 if (d_mountpoint(dentry)) {
1179 *ret = 1;
1180 return D_WALK_QUIT;
1181 }
1182 return D_WALK_CONTINUE;
1183}
1184
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001185/**
1186 * have_submounts - check for mounts over a dentry
1187 * @parent: dentry to check.
1188 *
1189 * Return true if the parent or its subdirectories contain
1190 * a mount point
1191 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001192int have_submounts(struct dentry *parent)
1193{
1194 int ret = 0;
1195
1196 d_walk(parent, &ret, check_mount, NULL);
1197
1198 return ret;
1199}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001200EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
1202/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001203 * Called by mount code to set a mountpoint and check if the mountpoint is
1204 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1205 * subtree can become unreachable).
1206 *
1207 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1208 * this reason take rename_lock and d_lock on dentry and ancestors.
1209 */
1210int d_set_mounted(struct dentry *dentry)
1211{
1212 struct dentry *p;
1213 int ret = -ENOENT;
1214 write_seqlock(&rename_lock);
1215 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1216 /* Need exclusion wrt. check_submounts_and_drop() */
1217 spin_lock(&p->d_lock);
1218 if (unlikely(d_unhashed(p))) {
1219 spin_unlock(&p->d_lock);
1220 goto out;
1221 }
1222 spin_unlock(&p->d_lock);
1223 }
1224 spin_lock(&dentry->d_lock);
1225 if (!d_unlinked(dentry)) {
1226 dentry->d_flags |= DCACHE_MOUNTED;
1227 ret = 0;
1228 }
1229 spin_unlock(&dentry->d_lock);
1230out:
1231 write_sequnlock(&rename_lock);
1232 return ret;
1233}
1234
1235/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001236 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 * and move any unused dentries to the end of the unused
1238 * list for prune_dcache(). We descend to the next level
1239 * whenever the d_subdirs list is non-empty and continue
1240 * searching.
1241 *
1242 * It returns zero iff there are no unused children,
1243 * otherwise it returns the number of children moved to
1244 * the end of the unused list. This may not be the total
1245 * number of unused children, because select_parent can
1246 * drop the lock and return early due to latency
1247 * constraints.
1248 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001249
1250struct select_data {
1251 struct dentry *start;
1252 struct list_head dispose;
1253 int found;
1254};
1255
1256static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001258 struct select_data *data = _data;
1259 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001261 if (data->start == dentry)
1262 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001264 /*
1265 * move only zero ref count dentries to the dispose list.
1266 *
1267 * Those which are presently on the shrink list, being processed
1268 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1269 * loop in shrink_dcache_parent() might not make any progress
1270 * and loop forever.
1271 */
1272 if (dentry->d_lockref.count) {
1273 dentry_lru_del(dentry);
1274 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001275 /*
1276 * We can't use d_lru_shrink_move() because we
1277 * need to get the global LRU lock and do the
Linus Torvalds05a82522013-09-15 07:11:01 -04001278 * LRU accounting.
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001279 */
1280 d_lru_del(dentry);
1281 d_shrink_add(dentry, &data->dispose);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001282 data->found++;
1283 ret = D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 }
1285 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001286 * We can return to the caller if we have found some (this
1287 * ensures forward progress). We'll be coming back to find
1288 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001290 if (data->found && need_resched())
1291 ret = D_WALK_QUIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001293 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294}
1295
1296/**
1297 * shrink_dcache_parent - prune dcache
1298 * @parent: parent of entries to prune
1299 *
1300 * Prune the dcache to remove unused children of the parent dentry.
1301 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001302void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001304 for (;;) {
1305 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001307 INIT_LIST_HEAD(&data.dispose);
1308 data.start = parent;
1309 data.found = 0;
1310
1311 d_walk(parent, &data, select_collect, NULL);
1312 if (!data.found)
1313 break;
1314
1315 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001316 cond_resched();
1317 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001319EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
Al Viro42c32602013-11-08 12:31:16 -05001321static enum d_walk_ret umount_collect(void *_data, struct dentry *dentry)
1322{
1323 struct select_data *data = _data;
1324 enum d_walk_ret ret = D_WALK_CONTINUE;
1325
1326 if (dentry->d_lockref.count) {
1327 dentry_lru_del(dentry);
1328 if (likely(!list_empty(&dentry->d_subdirs)))
1329 goto out;
1330 if (dentry == data->start && dentry->d_lockref.count == 1)
1331 goto out;
1332 printk(KERN_ERR
1333 "BUG: Dentry %p{i=%lx,n=%s}"
1334 " still in use (%d)"
1335 " [unmount of %s %s]\n",
1336 dentry,
1337 dentry->d_inode ?
1338 dentry->d_inode->i_ino : 0UL,
1339 dentry->d_name.name,
1340 dentry->d_lockref.count,
1341 dentry->d_sb->s_type->name,
1342 dentry->d_sb->s_id);
1343 BUG();
1344 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1345 /*
1346 * We can't use d_lru_shrink_move() because we
1347 * need to get the global LRU lock and do the
1348 * LRU accounting.
1349 */
1350 if (dentry->d_flags & DCACHE_LRU_LIST)
1351 d_lru_del(dentry);
1352 d_shrink_add(dentry, &data->dispose);
1353 data->found++;
1354 ret = D_WALK_NORETRY;
1355 }
1356out:
1357 if (data->found && need_resched())
1358 ret = D_WALK_QUIT;
1359 return ret;
1360}
1361
1362/*
1363 * destroy the dentries attached to a superblock on unmounting
1364 */
1365void shrink_dcache_for_umount(struct super_block *sb)
1366{
1367 struct dentry *dentry;
1368
1369 if (down_read_trylock(&sb->s_umount))
1370 BUG();
1371
1372 dentry = sb->s_root;
1373 sb->s_root = NULL;
1374 for (;;) {
1375 struct select_data data;
1376
1377 INIT_LIST_HEAD(&data.dispose);
1378 data.start = dentry;
1379 data.found = 0;
1380
1381 d_walk(dentry, &data, umount_collect, NULL);
1382 if (!data.found)
1383 break;
1384
1385 shrink_dentry_list(&data.dispose);
1386 cond_resched();
1387 }
1388 d_drop(dentry);
1389 dput(dentry);
1390
1391 while (!hlist_bl_empty(&sb->s_anon)) {
1392 struct select_data data;
1393 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1394
1395 INIT_LIST_HEAD(&data.dispose);
1396 data.start = NULL;
1397 data.found = 0;
1398
1399 d_walk(dentry, &data, umount_collect, NULL);
1400 if (data.found)
1401 shrink_dentry_list(&data.dispose);
1402 cond_resched();
1403 }
1404}
1405
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001406static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1407{
1408 struct select_data *data = _data;
1409
1410 if (d_mountpoint(dentry)) {
1411 data->found = -EBUSY;
1412 return D_WALK_QUIT;
1413 }
1414
1415 return select_collect(_data, dentry);
1416}
1417
1418static void check_and_drop(void *_data)
1419{
1420 struct select_data *data = _data;
1421
1422 if (d_mountpoint(data->start))
1423 data->found = -EBUSY;
1424 if (!data->found)
1425 __d_drop(data->start);
1426}
1427
1428/**
1429 * check_submounts_and_drop - prune dcache, check for submounts and drop
1430 *
1431 * All done as a single atomic operation relative to has_unlinked_ancestor().
1432 * Returns 0 if successfully unhashed @parent. If there were submounts then
1433 * return -EBUSY.
1434 *
1435 * @dentry: dentry to prune and drop
1436 */
1437int check_submounts_and_drop(struct dentry *dentry)
1438{
1439 int ret = 0;
1440
1441 /* Negative dentries can be dropped without further checks */
1442 if (!dentry->d_inode) {
1443 d_drop(dentry);
1444 goto out;
1445 }
1446
1447 for (;;) {
1448 struct select_data data;
1449
1450 INIT_LIST_HEAD(&data.dispose);
1451 data.start = dentry;
1452 data.found = 0;
1453
1454 d_walk(dentry, &data, check_and_collect, check_and_drop);
1455 ret = data.found;
1456
1457 if (!list_empty(&data.dispose))
1458 shrink_dentry_list(&data.dispose);
1459
1460 if (ret <= 0)
1461 break;
1462
1463 cond_resched();
1464 }
1465
1466out:
1467 return ret;
1468}
1469EXPORT_SYMBOL(check_submounts_and_drop);
1470
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471/**
Al Viroa4464db2011-07-07 15:03:58 -04001472 * __d_alloc - allocate a dcache entry
1473 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 * @name: qstr of the name
1475 *
1476 * Allocates a dentry. It returns %NULL if there is insufficient memory
1477 * available. On a success the dentry is returned. The name passed in is
1478 * copied and the copy passed in may be reused after this call.
1479 */
1480
Al Viroa4464db2011-07-07 15:03:58 -04001481struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482{
1483 struct dentry *dentry;
1484 char *dname;
1485
Mel Gormane12ba742007-10-16 01:25:52 -07001486 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 if (!dentry)
1488 return NULL;
1489
Linus Torvalds6326c712012-05-21 16:14:04 -07001490 /*
1491 * We guarantee that the inline name is always NUL-terminated.
1492 * This way the memcpy() done by the name switching in rename
1493 * will still always have a NUL at the end, even if we might
1494 * be overwriting an internal NUL character
1495 */
1496 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 if (name->len > DNAME_INLINE_LEN-1) {
1498 dname = kmalloc(name->len + 1, GFP_KERNEL);
1499 if (!dname) {
1500 kmem_cache_free(dentry_cache, dentry);
1501 return NULL;
1502 }
1503 } else {
1504 dname = dentry->d_iname;
1505 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
1507 dentry->d_name.len = name->len;
1508 dentry->d_name.hash = name->hash;
1509 memcpy(dname, name->name, name->len);
1510 dname[name->len] = 0;
1511
Linus Torvalds6326c712012-05-21 16:14:04 -07001512 /* Make sure we always see the terminating NUL character */
1513 smp_wmb();
1514 dentry->d_name.name = dname;
1515
Waiman Long98474232013-08-28 18:24:59 -07001516 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001517 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001519 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001521 dentry->d_parent = dentry;
1522 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 dentry->d_op = NULL;
1524 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001525 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 INIT_LIST_HEAD(&dentry->d_lru);
1527 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001528 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001529 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001530 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
Nick Piggin3e880fb2011-01-07 17:49:19 +11001532 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001533
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 return dentry;
1535}
Al Viroa4464db2011-07-07 15:03:58 -04001536
1537/**
1538 * d_alloc - allocate a dcache entry
1539 * @parent: parent of entry to allocate
1540 * @name: qstr of the name
1541 *
1542 * Allocates a dentry. It returns %NULL if there is insufficient memory
1543 * available. On a success the dentry is returned. The name passed in is
1544 * copied and the copy passed in may be reused after this call.
1545 */
1546struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1547{
1548 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1549 if (!dentry)
1550 return NULL;
1551
1552 spin_lock(&parent->d_lock);
1553 /*
1554 * don't need child lock because it is not subject
1555 * to concurrency here
1556 */
1557 __dget_dlock(parent);
1558 dentry->d_parent = parent;
1559 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1560 spin_unlock(&parent->d_lock);
1561
1562 return dentry;
1563}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001564EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001566/**
1567 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1568 * @sb: the superblock
1569 * @name: qstr of the name
1570 *
1571 * For a filesystem that just pins its dentries in memory and never
1572 * performs lookups at all, return an unhashed IS_ROOT dentry.
1573 */
Nick Piggin4b936882011-01-07 17:50:07 +11001574struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1575{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001576 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001577}
1578EXPORT_SYMBOL(d_alloc_pseudo);
1579
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1581{
1582 struct qstr q;
1583
1584 q.name = name;
1585 q.len = strlen(name);
1586 q.hash = full_name_hash(q.name, q.len);
1587 return d_alloc(parent, &q);
1588}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001589EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
Nick Pigginfb045ad2011-01-07 17:49:55 +11001591void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1592{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001593 WARN_ON_ONCE(dentry->d_op);
1594 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001595 DCACHE_OP_COMPARE |
1596 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001597 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001598 DCACHE_OP_DELETE ));
1599 dentry->d_op = op;
1600 if (!op)
1601 return;
1602 if (op->d_hash)
1603 dentry->d_flags |= DCACHE_OP_HASH;
1604 if (op->d_compare)
1605 dentry->d_flags |= DCACHE_OP_COMPARE;
1606 if (op->d_revalidate)
1607 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001608 if (op->d_weak_revalidate)
1609 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001610 if (op->d_delete)
1611 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001612 if (op->d_prune)
1613 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001614
1615}
1616EXPORT_SYMBOL(d_set_d_op);
1617
David Howellsb18825a2013-09-12 19:22:53 +01001618static unsigned d_flags_for_inode(struct inode *inode)
1619{
1620 unsigned add_flags = DCACHE_FILE_TYPE;
1621
1622 if (!inode)
1623 return DCACHE_MISS_TYPE;
1624
1625 if (S_ISDIR(inode->i_mode)) {
1626 add_flags = DCACHE_DIRECTORY_TYPE;
1627 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1628 if (unlikely(!inode->i_op->lookup))
1629 add_flags = DCACHE_AUTODIR_TYPE;
1630 else
1631 inode->i_opflags |= IOP_LOOKUP;
1632 }
1633 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1634 if (unlikely(inode->i_op->follow_link))
1635 add_flags = DCACHE_SYMLINK_TYPE;
1636 else
1637 inode->i_opflags |= IOP_NOFOLLOW;
1638 }
1639
1640 if (unlikely(IS_AUTOMOUNT(inode)))
1641 add_flags |= DCACHE_NEED_AUTOMOUNT;
1642 return add_flags;
1643}
1644
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001645static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1646{
David Howellsb18825a2013-09-12 19:22:53 +01001647 unsigned add_flags = d_flags_for_inode(inode);
1648
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001649 spin_lock(&dentry->d_lock);
David Howellsb18825a2013-09-12 19:22:53 +01001650 dentry->d_flags &= ~DCACHE_ENTRY_TYPE;
1651 dentry->d_flags |= add_flags;
1652 if (inode)
Al Virob3d9b7a2012-06-09 13:51:19 -04001653 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001654 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001655 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001656 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001657 fsnotify_d_instantiate(dentry, inode);
1658}
1659
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660/**
1661 * d_instantiate - fill in inode information for a dentry
1662 * @entry: dentry to complete
1663 * @inode: inode to attach to this dentry
1664 *
1665 * Fill in inode information in the entry.
1666 *
1667 * This turns negative dentries into productive full members
1668 * of society.
1669 *
1670 * NOTE! This assumes that the inode count has been incremented
1671 * (or otherwise set) by the caller to indicate that it is now
1672 * in use by the dcache.
1673 */
1674
1675void d_instantiate(struct dentry *entry, struct inode * inode)
1676{
Al Virob3d9b7a2012-06-09 13:51:19 -04001677 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001678 if (inode)
1679 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001680 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001681 if (inode)
1682 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 security_d_instantiate(entry, inode);
1684}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001685EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686
1687/**
1688 * d_instantiate_unique - instantiate a non-aliased dentry
1689 * @entry: dentry to instantiate
1690 * @inode: inode to attach to this dentry
1691 *
1692 * Fill in inode information in the entry. On success, it returns NULL.
1693 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001694 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 *
1696 * Note that in order to avoid conflicts with rename() etc, the caller
1697 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001698 *
1699 * This also assumes that the inode count has been incremented
1700 * (or otherwise set) by the caller to indicate that it is now
1701 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 */
David Howells770bfad2006-08-22 20:06:07 -04001703static struct dentry *__d_instantiate_unique(struct dentry *entry,
1704 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705{
1706 struct dentry *alias;
1707 int len = entry->d_name.len;
1708 const char *name = entry->d_name.name;
1709 unsigned int hash = entry->d_name.hash;
1710
David Howells770bfad2006-08-22 20:06:07 -04001711 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001712 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001713 return NULL;
1714 }
1715
Sasha Levinb67bfe02013-02-27 17:06:00 -08001716 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001717 /*
1718 * Don't need alias->d_lock here, because aliases with
1719 * d_parent == entry->d_parent are not subject to name or
1720 * parent changes, because the parent inode i_mutex is held.
1721 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001722 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 continue;
1724 if (alias->d_parent != entry->d_parent)
1725 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001726 if (alias->d_name.len != len)
1727 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001728 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001730 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 return alias;
1732 }
David Howells770bfad2006-08-22 20:06:07 -04001733
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001734 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 return NULL;
1736}
David Howells770bfad2006-08-22 20:06:07 -04001737
1738struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1739{
1740 struct dentry *result;
1741
Al Virob3d9b7a2012-06-09 13:51:19 -04001742 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001743
Nick Piggin873feea2011-01-07 17:50:06 +11001744 if (inode)
1745 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001746 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001747 if (inode)
1748 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001749
1750 if (!result) {
1751 security_d_instantiate(entry, inode);
1752 return NULL;
1753 }
1754
1755 BUG_ON(!d_unhashed(result));
1756 iput(inode);
1757 return result;
1758}
1759
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760EXPORT_SYMBOL(d_instantiate_unique);
1761
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001762/**
1763 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1764 * @entry: dentry to complete
1765 * @inode: inode to attach to this dentry
1766 *
1767 * Fill in inode information in the entry. If a directory alias is found, then
1768 * return an error (and drop inode). Together with d_materialise_unique() this
1769 * guarantees that a directory inode may never have more than one alias.
1770 */
1771int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1772{
1773 BUG_ON(!hlist_unhashed(&entry->d_alias));
1774
1775 spin_lock(&inode->i_lock);
1776 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1777 spin_unlock(&inode->i_lock);
1778 iput(inode);
1779 return -EBUSY;
1780 }
1781 __d_instantiate(entry, inode);
1782 spin_unlock(&inode->i_lock);
1783 security_d_instantiate(entry, inode);
1784
1785 return 0;
1786}
1787EXPORT_SYMBOL(d_instantiate_no_diralias);
1788
Al Viroadc0e912012-01-08 16:49:21 -05001789struct dentry *d_make_root(struct inode *root_inode)
1790{
1791 struct dentry *res = NULL;
1792
1793 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001794 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001795
1796 res = __d_alloc(root_inode->i_sb, &name);
1797 if (res)
1798 d_instantiate(res, root_inode);
1799 else
1800 iput(root_inode);
1801 }
1802 return res;
1803}
1804EXPORT_SYMBOL(d_make_root);
1805
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001806static struct dentry * __d_find_any_alias(struct inode *inode)
1807{
1808 struct dentry *alias;
1809
Al Virob3d9b7a2012-06-09 13:51:19 -04001810 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001811 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001812 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001813 __dget(alias);
1814 return alias;
1815}
1816
Sage Weil46f72b32012-01-10 09:04:37 -08001817/**
1818 * d_find_any_alias - find any alias for a given inode
1819 * @inode: inode to find an alias for
1820 *
1821 * If any aliases exist for the given inode, take and return a
1822 * reference for one of them. If no aliases exist, return %NULL.
1823 */
1824struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001825{
1826 struct dentry *de;
1827
1828 spin_lock(&inode->i_lock);
1829 de = __d_find_any_alias(inode);
1830 spin_unlock(&inode->i_lock);
1831 return de;
1832}
Sage Weil46f72b32012-01-10 09:04:37 -08001833EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001836 * d_obtain_alias - find or allocate a dentry for a given inode
1837 * @inode: inode to allocate the dentry for
1838 *
1839 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1840 * similar open by handle operations. The returned dentry may be anonymous,
1841 * or may have a full name (if the inode was already in the cache).
1842 *
1843 * When called on a directory inode, we must ensure that the inode only ever
1844 * has one dentry. If a dentry is found, that is returned instead of
1845 * allocating a new one.
1846 *
1847 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001848 * to the dentry. In case of an error the reference on the inode is released.
1849 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1850 * be passed in and will be the error will be propagate to the return value,
1851 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001852 */
1853struct dentry *d_obtain_alias(struct inode *inode)
1854{
NeilBrownb911a6b2012-11-08 16:09:37 -08001855 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001856 struct dentry *tmp;
1857 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001858 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001859
1860 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001861 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001862 if (IS_ERR(inode))
1863 return ERR_CAST(inode);
1864
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001865 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001866 if (res)
1867 goto out_iput;
1868
Al Viroa4464db2011-07-07 15:03:58 -04001869 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001870 if (!tmp) {
1871 res = ERR_PTR(-ENOMEM);
1872 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001873 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001874
Nick Piggin873feea2011-01-07 17:50:06 +11001875 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001876 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001877 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001878 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001879 dput(tmp);
1880 goto out_iput;
1881 }
1882
1883 /* attach a disconnected dentry */
David Howellsb18825a2013-09-12 19:22:53 +01001884 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1885
Christoph Hellwig9308a612008-08-11 15:49:12 +02001886 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001887 tmp->d_inode = inode;
David Howellsb18825a2013-09-12 19:22:53 +01001888 tmp->d_flags |= add_flags;
Al Virob3d9b7a2012-06-09 13:51:19 -04001889 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001890 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001891 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001892 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001893 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001894 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001895 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001896
Christoph Hellwig9308a612008-08-11 15:49:12 +02001897 return tmp;
1898
1899 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001900 if (res && !IS_ERR(res))
1901 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001902 iput(inode);
1903 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001904}
Benny Halevyadc48722009-02-27 14:02:59 -08001905EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906
1907/**
1908 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1909 * @inode: the inode which may have a disconnected dentry
1910 * @dentry: a negative dentry which we want to point to the inode.
1911 *
1912 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1913 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1914 * and return it, else simply d_add the inode to the dentry and return NULL.
1915 *
1916 * This is needed in the lookup routine of any filesystem that is exportable
1917 * (via knfsd) so that we can build dcache paths to directories effectively.
1918 *
1919 * If a dentry was found and moved, then it is returned. Otherwise NULL
1920 * is returned. This matches the expected return value of ->lookup.
1921 *
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001922 * Cluster filesystems may call this function with a negative, hashed dentry.
1923 * In that case, we know that the inode will be a regular file, and also this
1924 * will only occur during atomic_open. So we need to check for the dentry
1925 * being already hashed only in the final case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 */
1927struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1928{
1929 struct dentry *new = NULL;
1930
Al Viroa9049372011-07-08 21:20:11 -04001931 if (IS_ERR(inode))
1932 return ERR_CAST(inode);
1933
NeilBrown21c0d8f2006-10-04 02:16:16 -07001934 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001935 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001936 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001938 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001939 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 d_move(new, dentry);
1942 iput(inode);
1943 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001944 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001945 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001946 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 security_d_instantiate(dentry, inode);
1948 d_rehash(dentry);
1949 }
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001950 } else {
1951 d_instantiate(dentry, inode);
1952 if (d_unhashed(dentry))
1953 d_rehash(dentry);
1954 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 return new;
1956}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001957EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958
Barry Naujok94035402008-05-21 16:50:46 +10001959/**
1960 * d_add_ci - lookup or allocate new dentry with case-exact name
1961 * @inode: the inode case-insensitive lookup has found
1962 * @dentry: the negative dentry that was passed to the parent's lookup func
1963 * @name: the case-exact name to be associated with the returned dentry
1964 *
1965 * This is to avoid filling the dcache with case-insensitive names to the
1966 * same inode, only the actual correct case is stored in the dcache for
1967 * case-insensitive filesystems.
1968 *
1969 * For a case-insensitive lookup match and if the the case-exact dentry
1970 * already exists in in the dcache, use it and return it.
1971 *
1972 * If no entry exists with the exact case name, allocate new dentry with
1973 * the exact case, and return the spliced entry.
1974 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001975struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001976 struct qstr *name)
1977{
Barry Naujok94035402008-05-21 16:50:46 +10001978 struct dentry *found;
1979 struct dentry *new;
1980
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001981 /*
1982 * First check if a dentry matching the name already exists,
1983 * if not go ahead and create it now.
1984 */
Barry Naujok94035402008-05-21 16:50:46 +10001985 found = d_hash_and_lookup(dentry->d_parent, name);
Al Viro4f522a22013-02-11 23:20:37 -05001986 if (unlikely(IS_ERR(found)))
1987 goto err_out;
Barry Naujok94035402008-05-21 16:50:46 +10001988 if (!found) {
1989 new = d_alloc(dentry->d_parent, name);
1990 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05001991 found = ERR_PTR(-ENOMEM);
Barry Naujok94035402008-05-21 16:50:46 +10001992 goto err_out;
1993 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001994
Barry Naujok94035402008-05-21 16:50:46 +10001995 found = d_splice_alias(inode, new);
1996 if (found) {
1997 dput(new);
1998 return found;
1999 }
2000 return new;
2001 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002002
2003 /*
2004 * If a matching dentry exists, and it's not negative use it.
2005 *
2006 * Decrement the reference count to balance the iget() done
2007 * earlier on.
2008 */
Barry Naujok94035402008-05-21 16:50:46 +10002009 if (found->d_inode) {
2010 if (unlikely(found->d_inode != inode)) {
2011 /* This can't happen because bad inodes are unhashed. */
2012 BUG_ON(!is_bad_inode(inode));
2013 BUG_ON(!is_bad_inode(found->d_inode));
2014 }
Barry Naujok94035402008-05-21 16:50:46 +10002015 iput(inode);
2016 return found;
2017 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002018
Barry Naujok94035402008-05-21 16:50:46 +10002019 /*
2020 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002021 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10002022 */
Al Viro4513d892011-07-17 10:52:14 -04002023 new = d_splice_alias(inode, found);
2024 if (new) {
2025 dput(found);
2026 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10002027 }
Al Viro4513d892011-07-17 10:52:14 -04002028 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002029
2030err_out:
2031 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05002032 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002033}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002034EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002036/*
2037 * Do the slow-case of the dentry name compare.
2038 *
2039 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07002040 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002041 * filesystem can rely on them, and can use the 'name' and
2042 * 'len' information without worrying about walking off the
2043 * end of memory etc.
2044 *
2045 * Thus the read_seqcount_retry() and the "duplicate" info
2046 * in arguments (the low-level filesystem should not look
2047 * at the dentry inode or name contents directly, since
2048 * rename can change them while we're in RCU mode).
2049 */
2050enum slow_d_compare {
2051 D_COMP_OK,
2052 D_COMP_NOMATCH,
2053 D_COMP_SEQRETRY,
2054};
2055
2056static noinline enum slow_d_compare slow_dentry_cmp(
2057 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002058 struct dentry *dentry,
2059 unsigned int seq,
2060 const struct qstr *name)
2061{
2062 int tlen = dentry->d_name.len;
2063 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002064
2065 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2066 cpu_relax();
2067 return D_COMP_SEQRETRY;
2068 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002069 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002070 return D_COMP_NOMATCH;
2071 return D_COMP_OK;
2072}
2073
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002075 * __d_lookup_rcu - search for a dentry (racy, store-free)
2076 * @parent: parent dentry
2077 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002078 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002079 * Returns: dentry, or NULL
2080 *
2081 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2082 * resolution (store-free path walking) design described in
2083 * Documentation/filesystems/path-lookup.txt.
2084 *
2085 * This is not to be used outside core vfs.
2086 *
2087 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2088 * held, and rcu_read_lock held. The returned dentry must not be stored into
2089 * without taking d_lock and checking d_seq sequence count against @seq
2090 * returned here.
2091 *
Linus Torvalds15570082013-09-02 11:38:06 -07002092 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002093 * function.
2094 *
2095 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2096 * the returned dentry, so long as its parent's seqlock is checked after the
2097 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2098 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002099 *
2100 * NOTE! The caller *has* to check the resulting dentry against the sequence
2101 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002102 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002103struct dentry *__d_lookup_rcu(const struct dentry *parent,
2104 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002105 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002106{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002107 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002108 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002109 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002110 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002111 struct dentry *dentry;
2112
2113 /*
2114 * Note: There is significant duplication with __d_lookup_rcu which is
2115 * required to prevent single threaded performance regressions
2116 * especially on architectures where smp_rmb (in seqcounts) are costly.
2117 * Keep the two functions in sync.
2118 */
2119
2120 /*
2121 * The hash list is protected using RCU.
2122 *
2123 * Carefully use d_seq when comparing a candidate dentry, to avoid
2124 * races with d_move().
2125 *
2126 * It is possible that concurrent renames can mess up our list
2127 * walk here and result in missing our dentry, resulting in the
2128 * false-negative result. d_lookup() protects against concurrent
2129 * renames using rename_lock seqlock.
2130 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002131 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002132 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002133 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002134 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002135
Nick Piggin31e6b012011-01-07 17:49:52 +11002136seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002137 /*
2138 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002139 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002140 *
2141 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002142 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002143 *
2144 * NOTE! We do a "raw" seqcount_begin here. That means that
2145 * we don't wait for the sequence count to stabilize if it
2146 * is in the middle of a sequence change. If we do the slow
2147 * dentry compare, we will do seqretries until it is stable,
2148 * and if we end up with a successful lookup, we actually
2149 * want to exit RCU lookup anyway.
2150 */
2151 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002152 if (dentry->d_parent != parent)
2153 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002154 if (d_unhashed(dentry))
2155 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002156
2157 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002158 if (dentry->d_name.hash != hashlen_hash(hashlen))
2159 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002160 *seqp = seq;
2161 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002162 case D_COMP_OK:
2163 return dentry;
2164 case D_COMP_NOMATCH:
2165 continue;
2166 default:
2167 goto seqretry;
2168 }
2169 }
2170
Linus Torvalds26fe5752012-05-10 13:14:12 -07002171 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002172 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002173 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002174 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002175 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002176 }
2177 return NULL;
2178}
2179
2180/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 * d_lookup - search for a dentry
2182 * @parent: parent dentry
2183 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002184 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002186 * d_lookup searches the children of the parent dentry for the name in
2187 * question. If the dentry is found its reference count is incremented and the
2188 * dentry is returned. The caller must use dput to free the entry when it has
2189 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 */
Al Viroda2d8452013-01-24 18:29:34 -05002191struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192{
Nick Piggin31e6b012011-01-07 17:49:52 +11002193 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002194 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195
2196 do {
2197 seq = read_seqbegin(&rename_lock);
2198 dentry = __d_lookup(parent, name);
2199 if (dentry)
2200 break;
2201 } while (read_seqretry(&rename_lock, seq));
2202 return dentry;
2203}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002204EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205
Nick Piggin31e6b012011-01-07 17:49:52 +11002206/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002207 * __d_lookup - search for a dentry (racy)
2208 * @parent: parent dentry
2209 * @name: qstr of name we wish to find
2210 * Returns: dentry, or NULL
2211 *
2212 * __d_lookup is like d_lookup, however it may (rarely) return a
2213 * false-negative result due to unrelated rename activity.
2214 *
2215 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2216 * however it must be used carefully, eg. with a following d_lookup in
2217 * the case of failure.
2218 *
2219 * __d_lookup callers must be commented.
2220 */
Al Viroa713ca22013-01-24 18:27:00 -05002221struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222{
2223 unsigned int len = name->len;
2224 unsigned int hash = name->hash;
2225 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002226 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002227 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002228 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002229 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230
Nick Pigginb04f7842010-08-18 04:37:34 +10002231 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002232 * Note: There is significant duplication with __d_lookup_rcu which is
2233 * required to prevent single threaded performance regressions
2234 * especially on architectures where smp_rmb (in seqcounts) are costly.
2235 * Keep the two functions in sync.
2236 */
2237
2238 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002239 * The hash list is protected using RCU.
2240 *
2241 * Take d_lock when comparing a candidate dentry, to avoid races
2242 * with d_move().
2243 *
2244 * It is possible that concurrent renames can mess up our list
2245 * walk here and result in missing our dentry, resulting in the
2246 * false-negative result. d_lookup() protects against concurrent
2247 * renames using rename_lock seqlock.
2248 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002249 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002250 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 rcu_read_lock();
2252
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002253 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 if (dentry->d_name.hash != hash)
2256 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257
2258 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 if (dentry->d_parent != parent)
2260 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002261 if (d_unhashed(dentry))
2262 goto next;
2263
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 /*
2265 * It is safe to compare names since d_move() cannot
2266 * change the qstr (protected by d_lock).
2267 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002268 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002269 int tlen = dentry->d_name.len;
2270 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002271 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 goto next;
2273 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002274 if (dentry->d_name.len != len)
2275 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002276 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 goto next;
2278 }
2279
Waiman Long98474232013-08-28 18:24:59 -07002280 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002281 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 spin_unlock(&dentry->d_lock);
2283 break;
2284next:
2285 spin_unlock(&dentry->d_lock);
2286 }
2287 rcu_read_unlock();
2288
2289 return found;
2290}
2291
2292/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002293 * d_hash_and_lookup - hash the qstr then search for a dentry
2294 * @dir: Directory to search in
2295 * @name: qstr of name we wish to find
2296 *
Al Viro4f522a22013-02-11 23:20:37 -05002297 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002298 */
2299struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2300{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002301 /*
2302 * Check for a fs-specific hash function. Note that we must
2303 * calculate the standard hash first, as the d_op->d_hash()
2304 * routine may choose to leave the hash value unchanged.
2305 */
2306 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002307 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002308 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002309 if (unlikely(err < 0))
2310 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002311 }
Al Viro4f522a22013-02-11 23:20:37 -05002312 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002313}
Al Viro4f522a22013-02-11 23:20:37 -05002314EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002315
2316/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002317 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002319 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 *
2321 * An insecure source has sent us a dentry, here we verify it and dget() it.
2322 * This is used by ncpfs in its readdir implementation.
2323 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002324 *
2325 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002327int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328{
Nick Piggin786a5e12011-01-07 17:49:16 +11002329 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002330
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002331 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002332 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2333 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002334 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002335 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002336 spin_unlock(&dentry->d_lock);
2337 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 return 1;
2339 }
2340 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002341 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002342
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 return 0;
2344}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002345EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346
2347/*
2348 * When a file is deleted, we have two options:
2349 * - turn this dentry into a negative dentry
2350 * - unhash this dentry and free it.
2351 *
2352 * Usually, we want to just turn this into
2353 * a negative dentry, but if anybody else is
2354 * currently using the dentry or the inode
2355 * we can't do that and we fall back on removing
2356 * it from the hash queues and waiting for
2357 * it to be deleted later when it has no users
2358 */
2359
2360/**
2361 * d_delete - delete a dentry
2362 * @dentry: The dentry to delete
2363 *
2364 * Turn the dentry into a negative dentry if possible, otherwise
2365 * remove it from the hash queues so it can be deleted later
2366 */
2367
2368void d_delete(struct dentry * dentry)
2369{
Nick Piggin873feea2011-01-07 17:50:06 +11002370 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002371 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 /*
2373 * Are we the only user?
2374 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002375again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002377 inode = dentry->d_inode;
2378 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002379 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002380 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002381 spin_unlock(&dentry->d_lock);
2382 cpu_relax();
2383 goto again;
2384 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002385 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002386 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002387 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 return;
2389 }
2390
2391 if (!d_unhashed(dentry))
2392 __d_drop(dentry);
2393
2394 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002395
2396 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002398EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002400static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002402 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002403 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002404 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002405 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002406 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407}
2408
David Howells770bfad2006-08-22 20:06:07 -04002409static void _d_rehash(struct dentry * entry)
2410{
2411 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2412}
2413
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414/**
2415 * d_rehash - add an entry back to the hash
2416 * @entry: dentry to add to the hash
2417 *
2418 * Adds a dentry to the hash according to its name.
2419 */
2420
2421void d_rehash(struct dentry * entry)
2422{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002424 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002427EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002429/**
2430 * dentry_update_name_case - update case insensitive dentry with a new name
2431 * @dentry: dentry to be updated
2432 * @name: new name
2433 *
2434 * Update a case insensitive dentry with new case of name.
2435 *
2436 * dentry must have been returned by d_lookup with name @name. Old and new
2437 * name lengths must match (ie. no d_compare which allows mismatched name
2438 * lengths).
2439 *
2440 * Parent inode i_mutex must be held over d_lookup and into this call (to
2441 * keep renames and concurrent inserts, and readdir(2) away).
2442 */
2443void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2444{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002445 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002446 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2447
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002448 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002449 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002450 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002451 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002452 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002453}
2454EXPORT_SYMBOL(dentry_update_name_case);
2455
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456static void switch_names(struct dentry *dentry, struct dentry *target)
2457{
2458 if (dname_external(target)) {
2459 if (dname_external(dentry)) {
2460 /*
2461 * Both external: swap the pointers
2462 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002463 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 } else {
2465 /*
2466 * dentry:internal, target:external. Steal target's
2467 * storage and make target internal.
2468 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002469 memcpy(target->d_iname, dentry->d_name.name,
2470 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 dentry->d_name.name = target->d_name.name;
2472 target->d_name.name = target->d_iname;
2473 }
2474 } else {
2475 if (dname_external(dentry)) {
2476 /*
2477 * dentry:external, target:internal. Give dentry's
2478 * storage to target and make dentry internal
2479 */
2480 memcpy(dentry->d_iname, target->d_name.name,
2481 target->d_name.len + 1);
2482 target->d_name.name = dentry->d_name.name;
2483 dentry->d_name.name = dentry->d_iname;
2484 } else {
2485 /*
2486 * Both are internal. Just copy target to dentry
2487 */
2488 memcpy(dentry->d_iname, target->d_name.name,
2489 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05002490 dentry->d_name.len = target->d_name.len;
2491 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 }
2493 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002494 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495}
2496
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002497static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2498{
2499 /*
2500 * XXXX: do we really need to take target->d_lock?
2501 */
2502 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2503 spin_lock(&target->d_parent->d_lock);
2504 else {
2505 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2506 spin_lock(&dentry->d_parent->d_lock);
2507 spin_lock_nested(&target->d_parent->d_lock,
2508 DENTRY_D_LOCK_NESTED);
2509 } else {
2510 spin_lock(&target->d_parent->d_lock);
2511 spin_lock_nested(&dentry->d_parent->d_lock,
2512 DENTRY_D_LOCK_NESTED);
2513 }
2514 }
2515 if (target < dentry) {
2516 spin_lock_nested(&target->d_lock, 2);
2517 spin_lock_nested(&dentry->d_lock, 3);
2518 } else {
2519 spin_lock_nested(&dentry->d_lock, 2);
2520 spin_lock_nested(&target->d_lock, 3);
2521 }
2522}
2523
2524static void dentry_unlock_parents_for_move(struct dentry *dentry,
2525 struct dentry *target)
2526{
2527 if (target->d_parent != dentry->d_parent)
2528 spin_unlock(&dentry->d_parent->d_lock);
2529 if (target->d_parent != target)
2530 spin_unlock(&target->d_parent->d_lock);
2531}
2532
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002534 * When switching names, the actual string doesn't strictly have to
2535 * be preserved in the target - because we're dropping the target
2536 * anyway. As such, we can just do a simple memcpy() to copy over
2537 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002539 * Note that we have to be a lot more careful about getting the hash
2540 * switched - we have to switch the hash value properly even if it
2541 * then no longer matches the actual (corrupted) string of the target.
2542 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002544/*
Al Viro18367502011-07-12 21:42:24 -04002545 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 * @dentry: entry to move
2547 * @target: new dentry
2548 *
2549 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002550 * dcache entries should not be moved in this way. Caller must hold
2551 * rename_lock, the i_mutex of the source and target directories,
2552 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 */
Al Viro18367502011-07-12 21:42:24 -04002554static void __d_move(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 if (!dentry->d_inode)
2557 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2558
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002559 BUG_ON(d_ancestor(dentry, target));
2560 BUG_ON(d_ancestor(target, dentry));
2561
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002562 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563
Nick Piggin31e6b012011-01-07 17:49:52 +11002564 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002565 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002566
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002567 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2568
2569 /*
2570 * Move the dentry to the target hash queue. Don't bother checking
2571 * for the same hash queue because of how unlikely it is.
2572 */
2573 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002574 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575
2576 /* Unhash the target: dput() will then get rid of it */
2577 __d_drop(target);
2578
Eric Dumazet5160ee62006-01-08 01:03:32 -08002579 list_del(&dentry->d_u.d_child);
2580 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581
2582 /* Switch the names.. */
2583 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002584 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585
2586 /* ... and switch the parents */
2587 if (IS_ROOT(dentry)) {
2588 dentry->d_parent = target->d_parent;
2589 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002590 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002592 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593
2594 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002595 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 }
2597
Eric Dumazet5160ee62006-01-08 01:03:32 -08002598 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002599
Nick Piggin31e6b012011-01-07 17:49:52 +11002600 write_seqcount_end(&target->d_seq);
2601 write_seqcount_end(&dentry->d_seq);
2602
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002603 dentry_unlock_parents_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002605 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002607}
2608
2609/*
2610 * d_move - move a dentry
2611 * @dentry: entry to move
2612 * @target: new dentry
2613 *
2614 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002615 * dcache entries should not be moved in this way. See the locking
2616 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002617 */
2618void d_move(struct dentry *dentry, struct dentry *target)
2619{
2620 write_seqlock(&rename_lock);
2621 __d_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002623}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002624EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002626/**
2627 * d_ancestor - search for an ancestor
2628 * @p1: ancestor dentry
2629 * @p2: child dentry
2630 *
2631 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2632 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002633 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002634struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002635{
2636 struct dentry *p;
2637
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002638 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002639 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002640 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002641 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002642 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002643}
2644
2645/*
2646 * This helper attempts to cope with remotely renamed directories
2647 *
2648 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002649 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002650 *
2651 * Note: If ever the locking in lock_rename() changes, then please
2652 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002653 */
Nick Piggin873feea2011-01-07 17:50:06 +11002654static struct dentry *__d_unalias(struct inode *inode,
2655 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002656{
2657 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002658 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002659
2660 /* If alias and dentry share a parent, then no extra locks required */
2661 if (alias->d_parent == dentry->d_parent)
2662 goto out_unalias;
2663
Trond Myklebust9eaef272006-10-21 10:24:20 -07002664 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002665 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2666 goto out_err;
2667 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2668 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2669 goto out_err;
2670 m2 = &alias->d_parent->d_inode->i_mutex;
2671out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002672 if (likely(!d_mountpoint(alias))) {
2673 __d_move(alias, dentry);
2674 ret = alias;
2675 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002676out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002677 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002678 if (m2)
2679 mutex_unlock(m2);
2680 if (m1)
2681 mutex_unlock(m1);
2682 return ret;
2683}
2684
2685/*
David Howells770bfad2006-08-22 20:06:07 -04002686 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2687 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002688 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002689 */
2690static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2691{
Al Viro740da422013-01-30 10:13:38 -05002692 struct dentry *dparent;
David Howells770bfad2006-08-22 20:06:07 -04002693
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002694 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002695
Nick Piggin31e6b012011-01-07 17:49:52 +11002696 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002697 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002698
David Howells770bfad2006-08-22 20:06:07 -04002699 dparent = dentry->d_parent;
David Howells770bfad2006-08-22 20:06:07 -04002700
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002701 switch_names(dentry, anon);
2702 swap(dentry->d_name.hash, anon->d_name.hash);
2703
Al Viro740da422013-01-30 10:13:38 -05002704 dentry->d_parent = dentry;
2705 list_del_init(&dentry->d_u.d_child);
2706 anon->d_parent = dparent;
Wei Yongjun9ed53b12013-03-12 00:10:50 +08002707 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
David Howells770bfad2006-08-22 20:06:07 -04002708
Nick Piggin31e6b012011-01-07 17:49:52 +11002709 write_seqcount_end(&dentry->d_seq);
2710 write_seqcount_end(&anon->d_seq);
2711
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002712 dentry_unlock_parents_for_move(anon, dentry);
2713 spin_unlock(&dentry->d_lock);
2714
2715 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002716}
2717
2718/**
2719 * d_materialise_unique - introduce an inode into the tree
2720 * @dentry: candidate dentry
2721 * @inode: inode to bind to the dentry, to which aliases may be attached
2722 *
2723 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002724 * root directory alias in its place if there is one. Caller must hold the
2725 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002726 */
2727struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2728{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002729 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002730
2731 BUG_ON(!d_unhashed(dentry));
2732
David Howells770bfad2006-08-22 20:06:07 -04002733 if (!inode) {
2734 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002735 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002736 d_rehash(actual);
2737 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002738 }
2739
Nick Piggin873feea2011-01-07 17:50:06 +11002740 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002741
Trond Myklebust9eaef272006-10-21 10:24:20 -07002742 if (S_ISDIR(inode->i_mode)) {
2743 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002744
Trond Myklebust9eaef272006-10-21 10:24:20 -07002745 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002746 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002747 if (alias) {
2748 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002749 write_seqlock(&rename_lock);
2750
2751 if (d_ancestor(alias, dentry)) {
2752 /* Check for loops */
2753 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002754 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002755 } else if (IS_ROOT(alias)) {
2756 /* Is this an anonymous mountpoint that we
2757 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002758 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002759 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002760 __d_drop(alias);
2761 goto found;
Al Viro18367502011-07-12 21:42:24 -04002762 } else {
2763 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002764 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002765 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002766 }
Al Viro18367502011-07-12 21:42:24 -04002767 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002768 if (IS_ERR(actual)) {
2769 if (PTR_ERR(actual) == -ELOOP)
2770 pr_warn_ratelimited(
2771 "VFS: Lookup of '%s' in %s %s"
2772 " would have caused loop\n",
2773 dentry->d_name.name,
2774 inode->i_sb->s_type->name,
2775 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002776 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002777 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002778 goto out_nolock;
2779 }
David Howells770bfad2006-08-22 20:06:07 -04002780 }
2781
2782 /* Add a unique reference */
2783 actual = __d_instantiate_unique(dentry, inode);
2784 if (!actual)
2785 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002786 else
2787 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002788
David Howells770bfad2006-08-22 20:06:07 -04002789 spin_lock(&actual->d_lock);
2790found:
2791 _d_rehash(actual);
2792 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002793 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002794out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002795 if (actual == dentry) {
2796 security_d_instantiate(dentry, inode);
2797 return NULL;
2798 }
2799
2800 iput(inode);
2801 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002802}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002803EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002804
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002805static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002806{
2807 *buflen -= namelen;
2808 if (*buflen < 0)
2809 return -ENAMETOOLONG;
2810 *buffer -= namelen;
2811 memcpy(*buffer, str, namelen);
2812 return 0;
2813}
2814
Waiman Long232d2d62013-09-09 12:18:13 -04002815/**
2816 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002817 * @buffer: buffer pointer
2818 * @buflen: allocated length of the buffer
2819 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002820 *
2821 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2822 * make sure that either the old or the new name pointer and length are
2823 * fetched. However, there may be mismatch between length and pointer.
2824 * The length cannot be trusted, we need to copy it byte-by-byte until
2825 * the length is reached or a null byte is found. It also prepends "/" at
2826 * the beginning of the name. The sequence number check at the caller will
2827 * retry it again when a d_move() does happen. So any garbage in the buffer
2828 * due to mismatched pointer and length will be discarded.
2829 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002830static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2831{
Waiman Long232d2d62013-09-09 12:18:13 -04002832 const char *dname = ACCESS_ONCE(name->name);
2833 u32 dlen = ACCESS_ONCE(name->len);
2834 char *p;
2835
2836 if (*buflen < dlen + 1)
2837 return -ENAMETOOLONG;
2838 *buflen -= dlen + 1;
2839 p = *buffer -= dlen + 1;
2840 *p++ = '/';
2841 while (dlen--) {
2842 char c = *dname++;
2843 if (!c)
2844 break;
2845 *p++ = c;
2846 }
2847 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002848}
2849
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002851 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002852 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002853 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002854 * @buffer: pointer to the end of the buffer
2855 * @buflen: pointer to buffer length
2856 *
Waiman Long18129972013-09-12 10:55:35 -04002857 * The function will first try to write out the pathname without taking any
2858 * lock other than the RCU read lock to make sure that dentries won't go away.
2859 * It only checks the sequence number of the global rename_lock as any change
2860 * in the dentry's d_seq will be preceded by changes in the rename_lock
2861 * sequence number. If the sequence number had been changed, it will restart
2862 * the whole pathname back-tracing sequence again by taking the rename_lock.
2863 * In this case, there is no need to take the RCU read lock as the recursive
2864 * parent pointer references will keep the dentry chain alive as long as no
2865 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002866 */
Al Viro02125a82011-12-05 08:43:34 -05002867static int prepend_path(const struct path *path,
2868 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002869 char **buffer, int *buflen)
2870{
Al Viroede4ceb2013-11-13 07:45:40 -05002871 struct dentry *dentry;
2872 struct vfsmount *vfsmnt;
2873 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002874 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002875 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002876 char *bptr;
2877 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002878
Al Viro48f5ec22013-09-09 15:22:25 -04002879 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002880restart_mnt:
2881 read_seqbegin_or_lock(&mount_lock, &m_seq);
2882 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002883 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002884restart:
2885 bptr = *buffer;
2886 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002887 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002888 dentry = path->dentry;
2889 vfsmnt = path->mnt;
2890 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002891 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002892 while (dentry != root->dentry || vfsmnt != root->mnt) {
2893 struct dentry * parent;
2894
2895 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002896 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002897 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002898 if (mnt != parent) {
2899 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2900 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002901 vfsmnt = &mnt->mnt;
2902 continue;
2903 }
2904 /*
2905 * Filesystems needing to implement special "root names"
2906 * should do so with ->d_dname()
2907 */
2908 if (IS_ROOT(dentry) &&
2909 (dentry->d_name.len != 1 ||
2910 dentry->d_name.name[0] != '/')) {
2911 WARN(1, "Root dentry has weird name <%.*s>\n",
2912 (int) dentry->d_name.len,
2913 dentry->d_name.name);
2914 }
2915 if (!error)
2916 error = is_mounted(vfsmnt) ? 1 : 2;
2917 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002918 }
2919 parent = dentry->d_parent;
2920 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002921 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002922 if (error)
2923 break;
2924
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002925 dentry = parent;
2926 }
Al Viro48f5ec22013-09-09 15:22:25 -04002927 if (!(seq & 1))
2928 rcu_read_unlock();
2929 if (need_seqretry(&rename_lock, seq)) {
2930 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002931 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002932 }
Al Viro48f5ec22013-09-09 15:22:25 -04002933 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002934
2935 if (!(m_seq & 1))
2936 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002937 if (need_seqretry(&mount_lock, m_seq)) {
2938 m_seq = 1;
2939 goto restart_mnt;
2940 }
2941 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002942
Waiman Long232d2d62013-09-09 12:18:13 -04002943 if (error >= 0 && bptr == *buffer) {
2944 if (--blen < 0)
2945 error = -ENAMETOOLONG;
2946 else
2947 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002948 }
Waiman Long232d2d62013-09-09 12:18:13 -04002949 *buffer = bptr;
2950 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002951 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002952}
2953
2954/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002955 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002956 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002957 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002958 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959 * @buflen: buffer length
2960 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002961 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002963 * Returns a pointer into the buffer or an error code if the
2964 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002965 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002966 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002967 *
Al Viro02125a82011-12-05 08:43:34 -05002968 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 */
Al Viro02125a82011-12-05 08:43:34 -05002970char *__d_path(const struct path *path,
2971 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002972 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002974 char *res = buf + buflen;
2975 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002977 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002978 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002979
Al Viro02125a82011-12-05 08:43:34 -05002980 if (error < 0)
2981 return ERR_PTR(error);
2982 if (error > 0)
2983 return NULL;
2984 return res;
2985}
2986
2987char *d_absolute_path(const struct path *path,
2988 char *buf, int buflen)
2989{
2990 struct path root = {};
2991 char *res = buf + buflen;
2992 int error;
2993
2994 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002995 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05002996
2997 if (error > 1)
2998 error = -EINVAL;
2999 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003000 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003001 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002}
3003
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003004/*
3005 * same as __d_path but appends "(deleted)" for unlinked files.
3006 */
Al Viro02125a82011-12-05 08:43:34 -05003007static int path_with_deleted(const struct path *path,
3008 const struct path *root,
3009 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003010{
3011 prepend(buf, buflen, "\0", 1);
3012 if (d_unlinked(path->dentry)) {
3013 int error = prepend(buf, buflen, " (deleted)", 10);
3014 if (error)
3015 return error;
3016 }
3017
3018 return prepend_path(path, root, buf, buflen);
3019}
3020
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003021static int prepend_unreachable(char **buffer, int *buflen)
3022{
3023 return prepend(buffer, buflen, "(unreachable)", 13);
3024}
3025
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003026static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3027{
3028 unsigned seq;
3029
3030 do {
3031 seq = read_seqcount_begin(&fs->seq);
3032 *root = fs->root;
3033 } while (read_seqcount_retry(&fs->seq, seq));
3034}
3035
Jan Bluncka03a8a702008-02-14 19:38:32 -08003036/**
3037 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003038 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003039 * @buf: buffer to return value in
3040 * @buflen: buffer length
3041 *
3042 * Convert a dentry into an ASCII path name. If the entry has been deleted
3043 * the string " (deleted)" is appended. Note that this is ambiguous.
3044 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003045 * Returns a pointer into the buffer or an error code if the path was
3046 * too long. Note: Callers should use the returned pointer, not the passed
3047 * in buffer, to use the name! The implementation often starts at an offset
3048 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003049 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003050 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003051 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003052char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003054 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003055 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003056 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003058 /*
3059 * We have various synthetic filesystems that never get mounted. On
3060 * these filesystems dentries are never used for lookup purposes, and
3061 * thus don't need to be hashed. They also don't need a name until a
3062 * user wants to identify the object in /proc/pid/fd/. The little hack
3063 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003064 *
3065 * Some pseudo inodes are mountable. When they are mounted
3066 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3067 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003068 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003069 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3070 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003071 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003072
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003073 rcu_read_lock();
3074 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003075 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003076 rcu_read_unlock();
3077
Al Viro02125a82011-12-05 08:43:34 -05003078 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003079 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080 return res;
3081}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003082EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083
3084/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003085 * Helper function for dentry_operations.d_dname() members
3086 */
3087char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3088 const char *fmt, ...)
3089{
3090 va_list args;
3091 char temp[64];
3092 int sz;
3093
3094 va_start(args, fmt);
3095 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3096 va_end(args);
3097
3098 if (sz > sizeof(temp) || sz > buflen)
3099 return ERR_PTR(-ENAMETOOLONG);
3100
3101 buffer += buflen - sz;
3102 return memcpy(buffer, temp, sz);
3103}
3104
Al Viro118b2302013-08-24 12:08:17 -04003105char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3106{
3107 char *end = buffer + buflen;
3108 /* these dentries are never renamed, so d_lock is not needed */
3109 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003110 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003111 prepend(&end, &buflen, "/", 1))
3112 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003113 return end;
Al Viro118b2302013-08-24 12:08:17 -04003114}
3115
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003116/*
Ram Pai6092d042008-03-27 13:06:20 +01003117 * Write full pathname from the root of the filesystem into the buffer.
3118 */
Nick Pigginec2447c2011-01-07 17:49:29 +11003119static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003120{
Waiman Long232d2d62013-09-09 12:18:13 -04003121 char *end, *retval;
3122 int len, seq = 0;
3123 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003124
Al Viro48f5ec22013-09-09 15:22:25 -04003125 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003126restart:
3127 end = buf + buflen;
3128 len = buflen;
3129 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003130 if (buflen < 1)
3131 goto Elong;
3132 /* Get '/' right */
3133 retval = end-1;
3134 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003135 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003136 while (!IS_ROOT(dentry)) {
3137 struct dentry *parent = dentry->d_parent;
Nick Piggin9abca362011-01-07 17:49:36 +11003138 int error;
Ram Pai6092d042008-03-27 13:06:20 +01003139
Ram Pai6092d042008-03-27 13:06:20 +01003140 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003141 error = prepend_name(&end, &len, &dentry->d_name);
3142 if (error)
3143 break;
Ram Pai6092d042008-03-27 13:06:20 +01003144
3145 retval = end;
3146 dentry = parent;
3147 }
Al Viro48f5ec22013-09-09 15:22:25 -04003148 if (!(seq & 1))
3149 rcu_read_unlock();
3150 if (need_seqretry(&rename_lock, seq)) {
3151 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003152 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003153 }
3154 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003155 if (error)
3156 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003157 return retval;
3158Elong:
3159 return ERR_PTR(-ENAMETOOLONG);
3160}
Nick Pigginec2447c2011-01-07 17:49:29 +11003161
3162char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3163{
Waiman Long232d2d62013-09-09 12:18:13 -04003164 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003165}
3166EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003167
3168char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3169{
3170 char *p = NULL;
3171 char *retval;
3172
Al Viroc1031352010-06-06 22:31:14 -04003173 if (d_unlinked(dentry)) {
3174 p = buf + buflen;
3175 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3176 goto Elong;
3177 buflen++;
3178 }
3179 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003180 if (!IS_ERR(retval) && p)
3181 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003182 return retval;
3183Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003184 return ERR_PTR(-ENAMETOOLONG);
3185}
3186
Linus Torvalds8b19e342013-09-12 10:35:47 -07003187static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3188 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003189{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003190 unsigned seq;
3191
3192 do {
3193 seq = read_seqcount_begin(&fs->seq);
3194 *root = fs->root;
3195 *pwd = fs->pwd;
3196 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003197}
3198
Ram Pai6092d042008-03-27 13:06:20 +01003199/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200 * NOTE! The user-level library version returns a
3201 * character pointer. The kernel system call just
3202 * returns the length of the buffer filled (which
3203 * includes the ending '\0' character), or a negative
3204 * error value. So libc would do something like
3205 *
3206 * char *getcwd(char * buf, size_t size)
3207 * {
3208 * int retval;
3209 *
3210 * retval = sys_getcwd(buf, size);
3211 * if (retval >= 0)
3212 * return buf;
3213 * errno = -retval;
3214 * return NULL;
3215 * }
3216 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003217SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218{
Linus Torvalds552ce542007-02-13 12:08:18 -08003219 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003220 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003221 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222
3223 if (!page)
3224 return -ENOMEM;
3225
Linus Torvalds8b19e342013-09-12 10:35:47 -07003226 rcu_read_lock();
3227 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003228
Linus Torvalds552ce542007-02-13 12:08:18 -08003229 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003230 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003231 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003232 char *cwd = page + PATH_MAX;
3233 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003235 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003236 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003237 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003238
Al Viro02125a82011-12-05 08:43:34 -05003239 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003240 goto out;
3241
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003242 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003243 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003244 error = prepend_unreachable(&cwd, &buflen);
3245 if (error)
3246 goto out;
3247 }
3248
Linus Torvalds552ce542007-02-13 12:08:18 -08003249 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003250 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003251 if (len <= size) {
3252 error = len;
3253 if (copy_to_user(buf, cwd, len))
3254 error = -EFAULT;
3255 }
Nick Piggin949854d2011-01-07 17:49:37 +11003256 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003257 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003258 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259
3260out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003261 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262 return error;
3263}
3264
3265/*
3266 * Test whether new_dentry is a subdirectory of old_dentry.
3267 *
3268 * Trivially implemented using the dcache structure
3269 */
3270
3271/**
3272 * is_subdir - is new dentry a subdirectory of old_dentry
3273 * @new_dentry: new dentry
3274 * @old_dentry: old dentry
3275 *
3276 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3277 * Returns 0 otherwise.
3278 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3279 */
3280
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003281int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282{
3283 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11003284 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003286 if (new_dentry == old_dentry)
3287 return 1;
3288
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003289 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003290 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003292 /*
3293 * Need rcu_readlock to protect against the d_parent trashing
3294 * due to d_move
3295 */
3296 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003297 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003299 else
3300 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11003301 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303
3304 return result;
3305}
3306
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003307static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003309 struct dentry *root = data;
3310 if (dentry != root) {
3311 if (d_unhashed(dentry) || !dentry->d_inode)
3312 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003314 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3315 dentry->d_flags |= DCACHE_GENOCIDE;
3316 dentry->d_lockref.count--;
3317 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003319 return D_WALK_CONTINUE;
3320}
Nick Piggin58db63d2011-01-07 17:49:39 +11003321
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003322void d_genocide(struct dentry *parent)
3323{
3324 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325}
3326
Al Viro60545d02013-06-07 01:20:27 -04003327void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328{
Al Viro60545d02013-06-07 01:20:27 -04003329 inode_dec_link_count(inode);
3330 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3331 !hlist_unhashed(&dentry->d_alias) ||
3332 !d_unlinked(dentry));
3333 spin_lock(&dentry->d_parent->d_lock);
3334 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3335 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3336 (unsigned long long)inode->i_ino);
3337 spin_unlock(&dentry->d_lock);
3338 spin_unlock(&dentry->d_parent->d_lock);
3339 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340}
Al Viro60545d02013-06-07 01:20:27 -04003341EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342
3343static __initdata unsigned long dhash_entries;
3344static int __init set_dhash_entries(char *str)
3345{
3346 if (!str)
3347 return 0;
3348 dhash_entries = simple_strtoul(str, &str, 0);
3349 return 1;
3350}
3351__setup("dhash_entries=", set_dhash_entries);
3352
3353static void __init dcache_init_early(void)
3354{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003355 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356
3357 /* If hashes are distributed across NUMA nodes, defer
3358 * hash allocation until vmalloc space is available.
3359 */
3360 if (hashdist)
3361 return;
3362
3363 dentry_hashtable =
3364 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003365 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366 dhash_entries,
3367 13,
3368 HASH_EARLY,
3369 &d_hash_shift,
3370 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003371 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 0);
3373
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003374 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003375 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376}
3377
Denis Cheng74bf17c2007-10-16 23:26:30 -07003378static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003380 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381
3382 /*
3383 * A constructor could be added for stable state like the lists,
3384 * but it is probably not worth it because of the cache nature
3385 * of the dcache.
3386 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003387 dentry_cache = KMEM_CACHE(dentry,
3388 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389
3390 /* Hash may have been set up in dcache_init_early */
3391 if (!hashdist)
3392 return;
3393
3394 dentry_hashtable =
3395 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003396 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397 dhash_entries,
3398 13,
3399 0,
3400 &d_hash_shift,
3401 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003402 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403 0);
3404
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003405 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003406 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407}
3408
3409/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003410struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003411EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413EXPORT_SYMBOL(d_genocide);
3414
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415void __init vfs_caches_init_early(void)
3416{
3417 dcache_init_early();
3418 inode_init_early();
3419}
3420
3421void __init vfs_caches_init(unsigned long mempages)
3422{
3423 unsigned long reserve;
3424
3425 /* Base hash sizes on available memory, with a reserve equal to
3426 150% of current kernel size */
3427
3428 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3429 mempages -= reserve;
3430
3431 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003432 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433
Denis Cheng74bf17c2007-10-16 23:26:30 -07003434 dcache_init();
3435 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003437 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438 bdev_cache_init();
3439 chrdev_init();
3440}