blob: 494a9def5dced9af43b4cf6c1bfa50d6d010322a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/dcache.c
3 *
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
8
9/*
10 * Notes on the allocation strategy:
11 *
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040021#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/hash.h>
25#include <linux/cache.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/file.h>
29#include <asm/uaccess.h>
30#include <linux/security.h>
31#include <linux/seqlock.h>
32#include <linux/swap.h>
33#include <linux/bootmem.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Frederic Weisbecker613afbf2009-07-16 15:44:29 +020035#include <linux/hardirq.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110036#include <linux/bit_spinlock.h>
37#include <linux/rculist_bl.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070038#include <linux/prefetch.h>
David Howellsdd179942011-08-16 15:31:30 +010039#include <linux/ratelimit.h>
Dave Chinnerf6041562013-08-28 10:18:00 +100040#include <linux/list_lru.h>
David Howells07f3f052006-09-30 20:52:18 +020041#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050042#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Nick Piggin789680d2011-01-07 17:49:30 +110044/*
45 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110046 * dcache->d_inode->i_lock protects:
47 * - i_dentry, d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110048 * dcache_hash_bucket lock protects:
49 * - the dcache hash table
50 * s_anon bl list spinlock protects:
51 * - the s_anon list (see __d_drop)
Dave Chinner19156842013-08-28 10:17:55 +100052 * dentry->d_sb->s_dentry_lru_lock protects:
Nick Piggin23044502011-01-07 17:49:31 +110053 * - the dcache lru lists and counters
54 * d_lock protects:
55 * - d_flags
56 * - d_name
57 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110058 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110059 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110060 * - d_parent and d_subdirs
61 * - childrens' d_child and d_parent
Nick Pigginb23fb0a2011-01-07 17:49:35 +110062 * - d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110063 *
64 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110065 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110066 * dentry->d_lock
Dave Chinner19156842013-08-28 10:17:55 +100067 * dentry->d_sb->s_dentry_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110068 * dcache_hash_bucket lock
69 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110070 *
Nick Pigginda502952011-01-07 17:49:33 +110071 * If there is an ancestor relationship:
72 * dentry->d_parent->...->d_parent->d_lock
73 * ...
74 * dentry->d_parent->d_lock
75 * dentry->d_lock
76 *
77 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110078 * if (dentry1 < dentry2)
79 * dentry1->d_lock
80 * dentry2->d_lock
81 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080082int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
84
Al Viro74c3cbe2007-07-22 08:04:18 -040085__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Nick Piggin949854d2011-01-07 17:49:37 +110087EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Christoph Lametere18b8902006-12-06 20:33:20 -080089static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/*
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
95 *
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
98 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800100static unsigned int d_hash_mask __read_mostly;
101static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100102
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700103static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvalds8966be92012-03-02 14:23:30 -0800105static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700106 unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100107{
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700108 hash += (unsigned long) parent / L1_CACHE_BYTES;
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);
Al Viro22213312014-04-19 12:30:58 -04001650 __d_set_type(dentry, add_flags);
David Howellsb18825a2013-09-12 19:22:53 +01001651 if (inode)
Al Virob3d9b7a2012-06-09 13:51:19 -04001652 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001653 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001654 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001655 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001656 fsnotify_d_instantiate(dentry, inode);
1657}
1658
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659/**
1660 * d_instantiate - fill in inode information for a dentry
1661 * @entry: dentry to complete
1662 * @inode: inode to attach to this dentry
1663 *
1664 * Fill in inode information in the entry.
1665 *
1666 * This turns negative dentries into productive full members
1667 * of society.
1668 *
1669 * NOTE! This assumes that the inode count has been incremented
1670 * (or otherwise set) by the caller to indicate that it is now
1671 * in use by the dcache.
1672 */
1673
1674void d_instantiate(struct dentry *entry, struct inode * inode)
1675{
Al Virob3d9b7a2012-06-09 13:51:19 -04001676 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001677 if (inode)
1678 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001679 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001680 if (inode)
1681 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 security_d_instantiate(entry, inode);
1683}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001684EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
1686/**
1687 * d_instantiate_unique - instantiate a non-aliased dentry
1688 * @entry: dentry to instantiate
1689 * @inode: inode to attach to this dentry
1690 *
1691 * Fill in inode information in the entry. On success, it returns NULL.
1692 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001693 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 *
1695 * Note that in order to avoid conflicts with rename() etc, the caller
1696 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001697 *
1698 * This also assumes that the inode count has been incremented
1699 * (or otherwise set) by the caller to indicate that it is now
1700 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 */
David Howells770bfad2006-08-22 20:06:07 -04001702static struct dentry *__d_instantiate_unique(struct dentry *entry,
1703 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704{
1705 struct dentry *alias;
1706 int len = entry->d_name.len;
1707 const char *name = entry->d_name.name;
1708 unsigned int hash = entry->d_name.hash;
1709
David Howells770bfad2006-08-22 20:06:07 -04001710 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001711 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001712 return NULL;
1713 }
1714
Sasha Levinb67bfe02013-02-27 17:06:00 -08001715 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001716 /*
1717 * Don't need alias->d_lock here, because aliases with
1718 * d_parent == entry->d_parent are not subject to name or
1719 * parent changes, because the parent inode i_mutex is held.
1720 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001721 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 continue;
1723 if (alias->d_parent != entry->d_parent)
1724 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001725 if (alias->d_name.len != len)
1726 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001727 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001729 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 return alias;
1731 }
David Howells770bfad2006-08-22 20:06:07 -04001732
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001733 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 return NULL;
1735}
David Howells770bfad2006-08-22 20:06:07 -04001736
1737struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1738{
1739 struct dentry *result;
1740
Al Virob3d9b7a2012-06-09 13:51:19 -04001741 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001742
Nick Piggin873feea2011-01-07 17:50:06 +11001743 if (inode)
1744 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001745 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001746 if (inode)
1747 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001748
1749 if (!result) {
1750 security_d_instantiate(entry, inode);
1751 return NULL;
1752 }
1753
1754 BUG_ON(!d_unhashed(result));
1755 iput(inode);
1756 return result;
1757}
1758
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759EXPORT_SYMBOL(d_instantiate_unique);
1760
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001761/**
1762 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1763 * @entry: dentry to complete
1764 * @inode: inode to attach to this dentry
1765 *
1766 * Fill in inode information in the entry. If a directory alias is found, then
1767 * return an error (and drop inode). Together with d_materialise_unique() this
1768 * guarantees that a directory inode may never have more than one alias.
1769 */
1770int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1771{
1772 BUG_ON(!hlist_unhashed(&entry->d_alias));
1773
1774 spin_lock(&inode->i_lock);
1775 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1776 spin_unlock(&inode->i_lock);
1777 iput(inode);
1778 return -EBUSY;
1779 }
1780 __d_instantiate(entry, inode);
1781 spin_unlock(&inode->i_lock);
1782 security_d_instantiate(entry, inode);
1783
1784 return 0;
1785}
1786EXPORT_SYMBOL(d_instantiate_no_diralias);
1787
Al Viroadc0e912012-01-08 16:49:21 -05001788struct dentry *d_make_root(struct inode *root_inode)
1789{
1790 struct dentry *res = NULL;
1791
1792 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001793 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001794
1795 res = __d_alloc(root_inode->i_sb, &name);
1796 if (res)
1797 d_instantiate(res, root_inode);
1798 else
1799 iput(root_inode);
1800 }
1801 return res;
1802}
1803EXPORT_SYMBOL(d_make_root);
1804
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001805static struct dentry * __d_find_any_alias(struct inode *inode)
1806{
1807 struct dentry *alias;
1808
Al Virob3d9b7a2012-06-09 13:51:19 -04001809 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001810 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001811 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001812 __dget(alias);
1813 return alias;
1814}
1815
Sage Weil46f72b32012-01-10 09:04:37 -08001816/**
1817 * d_find_any_alias - find any alias for a given inode
1818 * @inode: inode to find an alias for
1819 *
1820 * If any aliases exist for the given inode, take and return a
1821 * reference for one of them. If no aliases exist, return %NULL.
1822 */
1823struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001824{
1825 struct dentry *de;
1826
1827 spin_lock(&inode->i_lock);
1828 de = __d_find_any_alias(inode);
1829 spin_unlock(&inode->i_lock);
1830 return de;
1831}
Sage Weil46f72b32012-01-10 09:04:37 -08001832EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001833
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001835 * d_obtain_alias - find or allocate a dentry for a given inode
1836 * @inode: inode to allocate the dentry for
1837 *
1838 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1839 * similar open by handle operations. The returned dentry may be anonymous,
1840 * or may have a full name (if the inode was already in the cache).
1841 *
1842 * When called on a directory inode, we must ensure that the inode only ever
1843 * has one dentry. If a dentry is found, that is returned instead of
1844 * allocating a new one.
1845 *
1846 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001847 * to the dentry. In case of an error the reference on the inode is released.
1848 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1849 * be passed in and will be the error will be propagate to the return value,
1850 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001851 */
1852struct dentry *d_obtain_alias(struct inode *inode)
1853{
NeilBrownb911a6b2012-11-08 16:09:37 -08001854 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001855 struct dentry *tmp;
1856 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001857 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001858
1859 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001860 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001861 if (IS_ERR(inode))
1862 return ERR_CAST(inode);
1863
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001864 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001865 if (res)
1866 goto out_iput;
1867
Al Viroa4464db2011-07-07 15:03:58 -04001868 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001869 if (!tmp) {
1870 res = ERR_PTR(-ENOMEM);
1871 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001872 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001873
Nick Piggin873feea2011-01-07 17:50:06 +11001874 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001875 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001876 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001877 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001878 dput(tmp);
1879 goto out_iput;
1880 }
1881
1882 /* attach a disconnected dentry */
David Howellsb18825a2013-09-12 19:22:53 +01001883 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1884
Christoph Hellwig9308a612008-08-11 15:49:12 +02001885 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001886 tmp->d_inode = inode;
David Howellsb18825a2013-09-12 19:22:53 +01001887 tmp->d_flags |= add_flags;
Al Virob3d9b7a2012-06-09 13:51:19 -04001888 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001889 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001890 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001891 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001892 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001893 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001894 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001895
Christoph Hellwig9308a612008-08-11 15:49:12 +02001896 return tmp;
1897
1898 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001899 if (res && !IS_ERR(res))
1900 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001901 iput(inode);
1902 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001903}
Benny Halevyadc48722009-02-27 14:02:59 -08001904EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
1906/**
1907 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1908 * @inode: the inode which may have a disconnected dentry
1909 * @dentry: a negative dentry which we want to point to the inode.
1910 *
1911 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1912 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1913 * and return it, else simply d_add the inode to the dentry and return NULL.
1914 *
1915 * This is needed in the lookup routine of any filesystem that is exportable
1916 * (via knfsd) so that we can build dcache paths to directories effectively.
1917 *
1918 * If a dentry was found and moved, then it is returned. Otherwise NULL
1919 * is returned. This matches the expected return value of ->lookup.
1920 *
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001921 * Cluster filesystems may call this function with a negative, hashed dentry.
1922 * In that case, we know that the inode will be a regular file, and also this
1923 * will only occur during atomic_open. So we need to check for the dentry
1924 * being already hashed only in the final case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 */
1926struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1927{
1928 struct dentry *new = NULL;
1929
Al Viroa9049372011-07-08 21:20:11 -04001930 if (IS_ERR(inode))
1931 return ERR_CAST(inode);
1932
NeilBrown21c0d8f2006-10-04 02:16:16 -07001933 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001934 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001935 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001937 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001938 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 d_move(new, dentry);
1941 iput(inode);
1942 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001943 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001944 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001945 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 security_d_instantiate(dentry, inode);
1947 d_rehash(dentry);
1948 }
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001949 } else {
1950 d_instantiate(dentry, inode);
1951 if (d_unhashed(dentry))
1952 d_rehash(dentry);
1953 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return new;
1955}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001956EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
Barry Naujok94035402008-05-21 16:50:46 +10001958/**
1959 * d_add_ci - lookup or allocate new dentry with case-exact name
1960 * @inode: the inode case-insensitive lookup has found
1961 * @dentry: the negative dentry that was passed to the parent's lookup func
1962 * @name: the case-exact name to be associated with the returned dentry
1963 *
1964 * This is to avoid filling the dcache with case-insensitive names to the
1965 * same inode, only the actual correct case is stored in the dcache for
1966 * case-insensitive filesystems.
1967 *
1968 * For a case-insensitive lookup match and if the the case-exact dentry
1969 * already exists in in the dcache, use it and return it.
1970 *
1971 * If no entry exists with the exact case name, allocate new dentry with
1972 * the exact case, and return the spliced entry.
1973 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001974struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001975 struct qstr *name)
1976{
Barry Naujok94035402008-05-21 16:50:46 +10001977 struct dentry *found;
1978 struct dentry *new;
1979
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001980 /*
1981 * First check if a dentry matching the name already exists,
1982 * if not go ahead and create it now.
1983 */
Barry Naujok94035402008-05-21 16:50:46 +10001984 found = d_hash_and_lookup(dentry->d_parent, name);
Al Viro4f522a22013-02-11 23:20:37 -05001985 if (unlikely(IS_ERR(found)))
1986 goto err_out;
Barry Naujok94035402008-05-21 16:50:46 +10001987 if (!found) {
1988 new = d_alloc(dentry->d_parent, name);
1989 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05001990 found = ERR_PTR(-ENOMEM);
Barry Naujok94035402008-05-21 16:50:46 +10001991 goto err_out;
1992 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001993
Barry Naujok94035402008-05-21 16:50:46 +10001994 found = d_splice_alias(inode, new);
1995 if (found) {
1996 dput(new);
1997 return found;
1998 }
1999 return new;
2000 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002001
2002 /*
2003 * If a matching dentry exists, and it's not negative use it.
2004 *
2005 * Decrement the reference count to balance the iget() done
2006 * earlier on.
2007 */
Barry Naujok94035402008-05-21 16:50:46 +10002008 if (found->d_inode) {
2009 if (unlikely(found->d_inode != inode)) {
2010 /* This can't happen because bad inodes are unhashed. */
2011 BUG_ON(!is_bad_inode(inode));
2012 BUG_ON(!is_bad_inode(found->d_inode));
2013 }
Barry Naujok94035402008-05-21 16:50:46 +10002014 iput(inode);
2015 return found;
2016 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002017
Barry Naujok94035402008-05-21 16:50:46 +10002018 /*
2019 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002020 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10002021 */
Al Viro4513d892011-07-17 10:52:14 -04002022 new = d_splice_alias(inode, found);
2023 if (new) {
2024 dput(found);
2025 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10002026 }
Al Viro4513d892011-07-17 10:52:14 -04002027 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002028
2029err_out:
2030 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05002031 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002032}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002033EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002035/*
2036 * Do the slow-case of the dentry name compare.
2037 *
2038 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07002039 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002040 * filesystem can rely on them, and can use the 'name' and
2041 * 'len' information without worrying about walking off the
2042 * end of memory etc.
2043 *
2044 * Thus the read_seqcount_retry() and the "duplicate" info
2045 * in arguments (the low-level filesystem should not look
2046 * at the dentry inode or name contents directly, since
2047 * rename can change them while we're in RCU mode).
2048 */
2049enum slow_d_compare {
2050 D_COMP_OK,
2051 D_COMP_NOMATCH,
2052 D_COMP_SEQRETRY,
2053};
2054
2055static noinline enum slow_d_compare slow_dentry_cmp(
2056 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002057 struct dentry *dentry,
2058 unsigned int seq,
2059 const struct qstr *name)
2060{
2061 int tlen = dentry->d_name.len;
2062 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002063
2064 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2065 cpu_relax();
2066 return D_COMP_SEQRETRY;
2067 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002068 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002069 return D_COMP_NOMATCH;
2070 return D_COMP_OK;
2071}
2072
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002074 * __d_lookup_rcu - search for a dentry (racy, store-free)
2075 * @parent: parent dentry
2076 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002077 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002078 * Returns: dentry, or NULL
2079 *
2080 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2081 * resolution (store-free path walking) design described in
2082 * Documentation/filesystems/path-lookup.txt.
2083 *
2084 * This is not to be used outside core vfs.
2085 *
2086 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2087 * held, and rcu_read_lock held. The returned dentry must not be stored into
2088 * without taking d_lock and checking d_seq sequence count against @seq
2089 * returned here.
2090 *
Linus Torvalds15570082013-09-02 11:38:06 -07002091 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002092 * function.
2093 *
2094 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2095 * the returned dentry, so long as its parent's seqlock is checked after the
2096 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2097 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002098 *
2099 * NOTE! The caller *has* to check the resulting dentry against the sequence
2100 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002101 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002102struct dentry *__d_lookup_rcu(const struct dentry *parent,
2103 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002104 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002105{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002106 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002107 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002108 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002109 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002110 struct dentry *dentry;
2111
2112 /*
2113 * Note: There is significant duplication with __d_lookup_rcu which is
2114 * required to prevent single threaded performance regressions
2115 * especially on architectures where smp_rmb (in seqcounts) are costly.
2116 * Keep the two functions in sync.
2117 */
2118
2119 /*
2120 * The hash list is protected using RCU.
2121 *
2122 * Carefully use d_seq when comparing a candidate dentry, to avoid
2123 * races with d_move().
2124 *
2125 * It is possible that concurrent renames can mess up our list
2126 * walk here and result in missing our dentry, resulting in the
2127 * false-negative result. d_lookup() protects against concurrent
2128 * renames using rename_lock seqlock.
2129 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002130 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002131 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002132 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002133 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002134
Nick Piggin31e6b012011-01-07 17:49:52 +11002135seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002136 /*
2137 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002138 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002139 *
2140 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002141 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002142 *
2143 * NOTE! We do a "raw" seqcount_begin here. That means that
2144 * we don't wait for the sequence count to stabilize if it
2145 * is in the middle of a sequence change. If we do the slow
2146 * dentry compare, we will do seqretries until it is stable,
2147 * and if we end up with a successful lookup, we actually
2148 * want to exit RCU lookup anyway.
2149 */
2150 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002151 if (dentry->d_parent != parent)
2152 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002153 if (d_unhashed(dentry))
2154 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002155
2156 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002157 if (dentry->d_name.hash != hashlen_hash(hashlen))
2158 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002159 *seqp = seq;
2160 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002161 case D_COMP_OK:
2162 return dentry;
2163 case D_COMP_NOMATCH:
2164 continue;
2165 default:
2166 goto seqretry;
2167 }
2168 }
2169
Linus Torvalds26fe5752012-05-10 13:14:12 -07002170 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002171 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002172 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002173 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002174 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002175 }
2176 return NULL;
2177}
2178
2179/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 * d_lookup - search for a dentry
2181 * @parent: parent dentry
2182 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002183 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002185 * d_lookup searches the children of the parent dentry for the name in
2186 * question. If the dentry is found its reference count is incremented and the
2187 * dentry is returned. The caller must use dput to free the entry when it has
2188 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 */
Al Viroda2d8452013-01-24 18:29:34 -05002190struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191{
Nick Piggin31e6b012011-01-07 17:49:52 +11002192 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002193 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194
2195 do {
2196 seq = read_seqbegin(&rename_lock);
2197 dentry = __d_lookup(parent, name);
2198 if (dentry)
2199 break;
2200 } while (read_seqretry(&rename_lock, seq));
2201 return dentry;
2202}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002203EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204
Nick Piggin31e6b012011-01-07 17:49:52 +11002205/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002206 * __d_lookup - search for a dentry (racy)
2207 * @parent: parent dentry
2208 * @name: qstr of name we wish to find
2209 * Returns: dentry, or NULL
2210 *
2211 * __d_lookup is like d_lookup, however it may (rarely) return a
2212 * false-negative result due to unrelated rename activity.
2213 *
2214 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2215 * however it must be used carefully, eg. with a following d_lookup in
2216 * the case of failure.
2217 *
2218 * __d_lookup callers must be commented.
2219 */
Al Viroa713ca22013-01-24 18:27:00 -05002220struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221{
2222 unsigned int len = name->len;
2223 unsigned int hash = name->hash;
2224 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002225 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002226 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002227 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002228 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
Nick Pigginb04f7842010-08-18 04:37:34 +10002230 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002231 * Note: There is significant duplication with __d_lookup_rcu which is
2232 * required to prevent single threaded performance regressions
2233 * especially on architectures where smp_rmb (in seqcounts) are costly.
2234 * Keep the two functions in sync.
2235 */
2236
2237 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002238 * The hash list is protected using RCU.
2239 *
2240 * Take d_lock when comparing a candidate dentry, to avoid races
2241 * with d_move().
2242 *
2243 * It is possible that concurrent renames can mess up our list
2244 * walk here and result in missing our dentry, resulting in the
2245 * false-negative result. d_lookup() protects against concurrent
2246 * renames using rename_lock seqlock.
2247 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002248 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002249 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 rcu_read_lock();
2251
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002252 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254 if (dentry->d_name.hash != hash)
2255 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256
2257 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 if (dentry->d_parent != parent)
2259 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002260 if (d_unhashed(dentry))
2261 goto next;
2262
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 /*
2264 * It is safe to compare names since d_move() cannot
2265 * change the qstr (protected by d_lock).
2266 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002267 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002268 int tlen = dentry->d_name.len;
2269 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002270 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 goto next;
2272 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002273 if (dentry->d_name.len != len)
2274 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002275 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 goto next;
2277 }
2278
Waiman Long98474232013-08-28 18:24:59 -07002279 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002280 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 spin_unlock(&dentry->d_lock);
2282 break;
2283next:
2284 spin_unlock(&dentry->d_lock);
2285 }
2286 rcu_read_unlock();
2287
2288 return found;
2289}
2290
2291/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002292 * d_hash_and_lookup - hash the qstr then search for a dentry
2293 * @dir: Directory to search in
2294 * @name: qstr of name we wish to find
2295 *
Al Viro4f522a22013-02-11 23:20:37 -05002296 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002297 */
2298struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2299{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002300 /*
2301 * Check for a fs-specific hash function. Note that we must
2302 * calculate the standard hash first, as the d_op->d_hash()
2303 * routine may choose to leave the hash value unchanged.
2304 */
2305 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002306 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002307 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002308 if (unlikely(err < 0))
2309 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002310 }
Al Viro4f522a22013-02-11 23:20:37 -05002311 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002312}
Al Viro4f522a22013-02-11 23:20:37 -05002313EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002314
2315/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002316 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002318 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 *
2320 * An insecure source has sent us a dentry, here we verify it and dget() it.
2321 * This is used by ncpfs in its readdir implementation.
2322 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002323 *
2324 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002326int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327{
Nick Piggin786a5e12011-01-07 17:49:16 +11002328 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002329
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002330 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002331 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2332 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002333 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002334 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002335 spin_unlock(&dentry->d_lock);
2336 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 return 1;
2338 }
2339 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002340 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002341
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 return 0;
2343}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002344EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345
2346/*
2347 * When a file is deleted, we have two options:
2348 * - turn this dentry into a negative dentry
2349 * - unhash this dentry and free it.
2350 *
2351 * Usually, we want to just turn this into
2352 * a negative dentry, but if anybody else is
2353 * currently using the dentry or the inode
2354 * we can't do that and we fall back on removing
2355 * it from the hash queues and waiting for
2356 * it to be deleted later when it has no users
2357 */
2358
2359/**
2360 * d_delete - delete a dentry
2361 * @dentry: The dentry to delete
2362 *
2363 * Turn the dentry into a negative dentry if possible, otherwise
2364 * remove it from the hash queues so it can be deleted later
2365 */
2366
2367void d_delete(struct dentry * dentry)
2368{
Nick Piggin873feea2011-01-07 17:50:06 +11002369 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002370 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 /*
2372 * Are we the only user?
2373 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002374again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002376 inode = dentry->d_inode;
2377 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002378 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002379 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002380 spin_unlock(&dentry->d_lock);
2381 cpu_relax();
2382 goto again;
2383 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002384 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002385 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002386 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 return;
2388 }
2389
2390 if (!d_unhashed(dentry))
2391 __d_drop(dentry);
2392
2393 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002394
2395 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002397EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002399static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002401 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002402 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002403 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002404 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002405 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406}
2407
David Howells770bfad2006-08-22 20:06:07 -04002408static void _d_rehash(struct dentry * entry)
2409{
2410 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2411}
2412
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413/**
2414 * d_rehash - add an entry back to the hash
2415 * @entry: dentry to add to the hash
2416 *
2417 * Adds a dentry to the hash according to its name.
2418 */
2419
2420void d_rehash(struct dentry * entry)
2421{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002423 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002426EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002428/**
2429 * dentry_update_name_case - update case insensitive dentry with a new name
2430 * @dentry: dentry to be updated
2431 * @name: new name
2432 *
2433 * Update a case insensitive dentry with new case of name.
2434 *
2435 * dentry must have been returned by d_lookup with name @name. Old and new
2436 * name lengths must match (ie. no d_compare which allows mismatched name
2437 * lengths).
2438 *
2439 * Parent inode i_mutex must be held over d_lookup and into this call (to
2440 * keep renames and concurrent inserts, and readdir(2) away).
2441 */
2442void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2443{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002444 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002445 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2446
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002447 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002448 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002449 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002450 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002451 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002452}
2453EXPORT_SYMBOL(dentry_update_name_case);
2454
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455static void switch_names(struct dentry *dentry, struct dentry *target)
2456{
2457 if (dname_external(target)) {
2458 if (dname_external(dentry)) {
2459 /*
2460 * Both external: swap the pointers
2461 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002462 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 } else {
2464 /*
2465 * dentry:internal, target:external. Steal target's
2466 * storage and make target internal.
2467 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002468 memcpy(target->d_iname, dentry->d_name.name,
2469 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 dentry->d_name.name = target->d_name.name;
2471 target->d_name.name = target->d_iname;
2472 }
2473 } else {
2474 if (dname_external(dentry)) {
2475 /*
2476 * dentry:external, target:internal. Give dentry's
2477 * storage to target and make dentry internal
2478 */
2479 memcpy(dentry->d_iname, target->d_name.name,
2480 target->d_name.len + 1);
2481 target->d_name.name = dentry->d_name.name;
2482 dentry->d_name.name = dentry->d_iname;
2483 } else {
2484 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002485 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002487 unsigned int i;
2488 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2489 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2490 swap(((long *) &dentry->d_iname)[i],
2491 ((long *) &target->d_iname)[i]);
2492 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 }
2494 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002495 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496}
2497
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002498static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2499{
2500 /*
2501 * XXXX: do we really need to take target->d_lock?
2502 */
2503 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2504 spin_lock(&target->d_parent->d_lock);
2505 else {
2506 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2507 spin_lock(&dentry->d_parent->d_lock);
2508 spin_lock_nested(&target->d_parent->d_lock,
2509 DENTRY_D_LOCK_NESTED);
2510 } else {
2511 spin_lock(&target->d_parent->d_lock);
2512 spin_lock_nested(&dentry->d_parent->d_lock,
2513 DENTRY_D_LOCK_NESTED);
2514 }
2515 }
2516 if (target < dentry) {
2517 spin_lock_nested(&target->d_lock, 2);
2518 spin_lock_nested(&dentry->d_lock, 3);
2519 } else {
2520 spin_lock_nested(&dentry->d_lock, 2);
2521 spin_lock_nested(&target->d_lock, 3);
2522 }
2523}
2524
2525static void dentry_unlock_parents_for_move(struct dentry *dentry,
2526 struct dentry *target)
2527{
2528 if (target->d_parent != dentry->d_parent)
2529 spin_unlock(&dentry->d_parent->d_lock);
2530 if (target->d_parent != target)
2531 spin_unlock(&target->d_parent->d_lock);
2532}
2533
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002535 * When switching names, the actual string doesn't strictly have to
2536 * be preserved in the target - because we're dropping the target
2537 * anyway. As such, we can just do a simple memcpy() to copy over
2538 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002540 * Note that we have to be a lot more careful about getting the hash
2541 * switched - we have to switch the hash value properly even if it
2542 * then no longer matches the actual (corrupted) string of the target.
2543 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002545/*
Al Viro18367502011-07-12 21:42:24 -04002546 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 * @dentry: entry to move
2548 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002549 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 *
2551 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002552 * dcache entries should not be moved in this way. Caller must hold
2553 * rename_lock, the i_mutex of the source and target directories,
2554 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002556static void __d_move(struct dentry *dentry, struct dentry *target,
2557 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 if (!dentry->d_inode)
2560 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2561
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002562 BUG_ON(d_ancestor(dentry, target));
2563 BUG_ON(d_ancestor(target, dentry));
2564
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002565 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566
Nick Piggin31e6b012011-01-07 17:49:52 +11002567 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002568 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002569
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002570 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2571
2572 /*
2573 * Move the dentry to the target hash queue. Don't bother checking
2574 * for the same hash queue because of how unlikely it is.
2575 */
2576 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002577 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578
Miklos Szeredida1ce062014-04-01 17:08:43 +02002579 /*
2580 * Unhash the target (d_delete() is not usable here). If exchanging
2581 * the two dentries, then rehash onto the other's hash queue.
2582 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002584 if (exchange) {
2585 __d_rehash(target,
2586 d_hash(dentry->d_parent, dentry->d_name.hash));
2587 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588
Eric Dumazet5160ee62006-01-08 01:03:32 -08002589 list_del(&dentry->d_u.d_child);
2590 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591
2592 /* Switch the names.. */
2593 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002594 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595
2596 /* ... and switch the parents */
2597 if (IS_ROOT(dentry)) {
2598 dentry->d_parent = target->d_parent;
2599 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002600 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002602 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603
2604 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002605 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 }
2607
Eric Dumazet5160ee62006-01-08 01:03:32 -08002608 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002609
Nick Piggin31e6b012011-01-07 17:49:52 +11002610 write_seqcount_end(&target->d_seq);
2611 write_seqcount_end(&dentry->d_seq);
2612
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002613 dentry_unlock_parents_for_move(dentry, target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002614 if (exchange)
2615 fsnotify_d_move(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002617 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002619}
2620
2621/*
2622 * d_move - move a dentry
2623 * @dentry: entry to move
2624 * @target: new dentry
2625 *
2626 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002627 * dcache entries should not be moved in this way. See the locking
2628 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002629 */
2630void d_move(struct dentry *dentry, struct dentry *target)
2631{
2632 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002633 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002635}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002636EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637
Miklos Szeredida1ce062014-04-01 17:08:43 +02002638/*
2639 * d_exchange - exchange two dentries
2640 * @dentry1: first dentry
2641 * @dentry2: second dentry
2642 */
2643void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2644{
2645 write_seqlock(&rename_lock);
2646
2647 WARN_ON(!dentry1->d_inode);
2648 WARN_ON(!dentry2->d_inode);
2649 WARN_ON(IS_ROOT(dentry1));
2650 WARN_ON(IS_ROOT(dentry2));
2651
2652 __d_move(dentry1, dentry2, true);
2653
2654 write_sequnlock(&rename_lock);
2655}
2656
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002657/**
2658 * d_ancestor - search for an ancestor
2659 * @p1: ancestor dentry
2660 * @p2: child dentry
2661 *
2662 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2663 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002664 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002665struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002666{
2667 struct dentry *p;
2668
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002669 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002670 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002671 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002672 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002673 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002674}
2675
2676/*
2677 * This helper attempts to cope with remotely renamed directories
2678 *
2679 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002680 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002681 *
2682 * Note: If ever the locking in lock_rename() changes, then please
2683 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002684 */
Nick Piggin873feea2011-01-07 17:50:06 +11002685static struct dentry *__d_unalias(struct inode *inode,
2686 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002687{
2688 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002689 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002690
2691 /* If alias and dentry share a parent, then no extra locks required */
2692 if (alias->d_parent == dentry->d_parent)
2693 goto out_unalias;
2694
Trond Myklebust9eaef272006-10-21 10:24:20 -07002695 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002696 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2697 goto out_err;
2698 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2699 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2700 goto out_err;
2701 m2 = &alias->d_parent->d_inode->i_mutex;
2702out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002703 if (likely(!d_mountpoint(alias))) {
Miklos Szeredida1ce062014-04-01 17:08:43 +02002704 __d_move(alias, dentry, false);
Al Viroee3efa92012-06-08 15:59:33 -04002705 ret = alias;
2706 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002707out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002708 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002709 if (m2)
2710 mutex_unlock(m2);
2711 if (m1)
2712 mutex_unlock(m1);
2713 return ret;
2714}
2715
2716/*
David Howells770bfad2006-08-22 20:06:07 -04002717 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2718 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002719 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002720 */
2721static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2722{
Al Viro740da422013-01-30 10:13:38 -05002723 struct dentry *dparent;
David Howells770bfad2006-08-22 20:06:07 -04002724
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002725 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002726
Nick Piggin31e6b012011-01-07 17:49:52 +11002727 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002728 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002729
David Howells770bfad2006-08-22 20:06:07 -04002730 dparent = dentry->d_parent;
David Howells770bfad2006-08-22 20:06:07 -04002731
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002732 switch_names(dentry, anon);
2733 swap(dentry->d_name.hash, anon->d_name.hash);
2734
Al Viro740da422013-01-30 10:13:38 -05002735 dentry->d_parent = dentry;
2736 list_del_init(&dentry->d_u.d_child);
2737 anon->d_parent = dparent;
Wei Yongjun9ed53b12013-03-12 00:10:50 +08002738 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
David Howells770bfad2006-08-22 20:06:07 -04002739
Nick Piggin31e6b012011-01-07 17:49:52 +11002740 write_seqcount_end(&dentry->d_seq);
2741 write_seqcount_end(&anon->d_seq);
2742
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002743 dentry_unlock_parents_for_move(anon, dentry);
2744 spin_unlock(&dentry->d_lock);
2745
2746 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002747}
2748
2749/**
2750 * d_materialise_unique - introduce an inode into the tree
2751 * @dentry: candidate dentry
2752 * @inode: inode to bind to the dentry, to which aliases may be attached
2753 *
2754 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002755 * root directory alias in its place if there is one. Caller must hold the
2756 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002757 */
2758struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2759{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002760 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002761
2762 BUG_ON(!d_unhashed(dentry));
2763
David Howells770bfad2006-08-22 20:06:07 -04002764 if (!inode) {
2765 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002766 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002767 d_rehash(actual);
2768 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002769 }
2770
Nick Piggin873feea2011-01-07 17:50:06 +11002771 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002772
Trond Myklebust9eaef272006-10-21 10:24:20 -07002773 if (S_ISDIR(inode->i_mode)) {
2774 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002775
Trond Myklebust9eaef272006-10-21 10:24:20 -07002776 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002777 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002778 if (alias) {
2779 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002780 write_seqlock(&rename_lock);
2781
2782 if (d_ancestor(alias, dentry)) {
2783 /* Check for loops */
2784 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002785 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002786 } else if (IS_ROOT(alias)) {
2787 /* Is this an anonymous mountpoint that we
2788 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002789 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002790 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002791 __d_drop(alias);
2792 goto found;
Al Viro18367502011-07-12 21:42:24 -04002793 } else {
2794 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002795 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002796 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002797 }
Al Viro18367502011-07-12 21:42:24 -04002798 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002799 if (IS_ERR(actual)) {
2800 if (PTR_ERR(actual) == -ELOOP)
2801 pr_warn_ratelimited(
2802 "VFS: Lookup of '%s' in %s %s"
2803 " would have caused loop\n",
2804 dentry->d_name.name,
2805 inode->i_sb->s_type->name,
2806 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002807 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002808 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002809 goto out_nolock;
2810 }
David Howells770bfad2006-08-22 20:06:07 -04002811 }
2812
2813 /* Add a unique reference */
2814 actual = __d_instantiate_unique(dentry, inode);
2815 if (!actual)
2816 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002817 else
2818 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002819
David Howells770bfad2006-08-22 20:06:07 -04002820 spin_lock(&actual->d_lock);
2821found:
2822 _d_rehash(actual);
2823 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002824 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002825out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002826 if (actual == dentry) {
2827 security_d_instantiate(dentry, inode);
2828 return NULL;
2829 }
2830
2831 iput(inode);
2832 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002833}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002834EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002835
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002836static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002837{
2838 *buflen -= namelen;
2839 if (*buflen < 0)
2840 return -ENAMETOOLONG;
2841 *buffer -= namelen;
2842 memcpy(*buffer, str, namelen);
2843 return 0;
2844}
2845
Waiman Long232d2d62013-09-09 12:18:13 -04002846/**
2847 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002848 * @buffer: buffer pointer
2849 * @buflen: allocated length of the buffer
2850 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002851 *
2852 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2853 * make sure that either the old or the new name pointer and length are
2854 * fetched. However, there may be mismatch between length and pointer.
2855 * The length cannot be trusted, we need to copy it byte-by-byte until
2856 * the length is reached or a null byte is found. It also prepends "/" at
2857 * the beginning of the name. The sequence number check at the caller will
2858 * retry it again when a d_move() does happen. So any garbage in the buffer
2859 * due to mismatched pointer and length will be discarded.
2860 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002861static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2862{
Waiman Long232d2d62013-09-09 12:18:13 -04002863 const char *dname = ACCESS_ONCE(name->name);
2864 u32 dlen = ACCESS_ONCE(name->len);
2865 char *p;
2866
Waiman Long232d2d62013-09-09 12:18:13 -04002867 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04002868 if (*buflen < 0)
2869 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04002870 p = *buffer -= dlen + 1;
2871 *p++ = '/';
2872 while (dlen--) {
2873 char c = *dname++;
2874 if (!c)
2875 break;
2876 *p++ = c;
2877 }
2878 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002879}
2880
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002882 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002883 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002884 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002885 * @buffer: pointer to the end of the buffer
2886 * @buflen: pointer to buffer length
2887 *
Waiman Long18129972013-09-12 10:55:35 -04002888 * The function will first try to write out the pathname without taking any
2889 * lock other than the RCU read lock to make sure that dentries won't go away.
2890 * It only checks the sequence number of the global rename_lock as any change
2891 * in the dentry's d_seq will be preceded by changes in the rename_lock
2892 * sequence number. If the sequence number had been changed, it will restart
2893 * the whole pathname back-tracing sequence again by taking the rename_lock.
2894 * In this case, there is no need to take the RCU read lock as the recursive
2895 * parent pointer references will keep the dentry chain alive as long as no
2896 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002897 */
Al Viro02125a82011-12-05 08:43:34 -05002898static int prepend_path(const struct path *path,
2899 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002900 char **buffer, int *buflen)
2901{
Al Viroede4ceb2013-11-13 07:45:40 -05002902 struct dentry *dentry;
2903 struct vfsmount *vfsmnt;
2904 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002905 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002906 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002907 char *bptr;
2908 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002909
Al Viro48f5ec22013-09-09 15:22:25 -04002910 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002911restart_mnt:
2912 read_seqbegin_or_lock(&mount_lock, &m_seq);
2913 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002914 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002915restart:
2916 bptr = *buffer;
2917 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002918 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002919 dentry = path->dentry;
2920 vfsmnt = path->mnt;
2921 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002922 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002923 while (dentry != root->dentry || vfsmnt != root->mnt) {
2924 struct dentry * parent;
2925
2926 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002927 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002928 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002929 if (mnt != parent) {
2930 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2931 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002932 vfsmnt = &mnt->mnt;
2933 continue;
2934 }
2935 /*
2936 * Filesystems needing to implement special "root names"
2937 * should do so with ->d_dname()
2938 */
2939 if (IS_ROOT(dentry) &&
2940 (dentry->d_name.len != 1 ||
2941 dentry->d_name.name[0] != '/')) {
2942 WARN(1, "Root dentry has weird name <%.*s>\n",
2943 (int) dentry->d_name.len,
2944 dentry->d_name.name);
2945 }
2946 if (!error)
2947 error = is_mounted(vfsmnt) ? 1 : 2;
2948 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002949 }
2950 parent = dentry->d_parent;
2951 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002952 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002953 if (error)
2954 break;
2955
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002956 dentry = parent;
2957 }
Al Viro48f5ec22013-09-09 15:22:25 -04002958 if (!(seq & 1))
2959 rcu_read_unlock();
2960 if (need_seqretry(&rename_lock, seq)) {
2961 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002962 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002963 }
Al Viro48f5ec22013-09-09 15:22:25 -04002964 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002965
2966 if (!(m_seq & 1))
2967 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002968 if (need_seqretry(&mount_lock, m_seq)) {
2969 m_seq = 1;
2970 goto restart_mnt;
2971 }
2972 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002973
Waiman Long232d2d62013-09-09 12:18:13 -04002974 if (error >= 0 && bptr == *buffer) {
2975 if (--blen < 0)
2976 error = -ENAMETOOLONG;
2977 else
2978 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002979 }
Waiman Long232d2d62013-09-09 12:18:13 -04002980 *buffer = bptr;
2981 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002982 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002983}
2984
2985/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002986 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002987 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002988 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002989 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 * @buflen: buffer length
2991 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002992 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002994 * Returns a pointer into the buffer or an error code if the
2995 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002996 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002997 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002998 *
Al Viro02125a82011-12-05 08:43:34 -05002999 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 */
Al Viro02125a82011-12-05 08:43:34 -05003001char *__d_path(const struct path *path,
3002 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003003 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003005 char *res = buf + buflen;
3006 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003008 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003009 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04003010
Al Viro02125a82011-12-05 08:43:34 -05003011 if (error < 0)
3012 return ERR_PTR(error);
3013 if (error > 0)
3014 return NULL;
3015 return res;
3016}
3017
3018char *d_absolute_path(const struct path *path,
3019 char *buf, int buflen)
3020{
3021 struct path root = {};
3022 char *res = buf + buflen;
3023 int error;
3024
3025 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003026 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05003027
3028 if (error > 1)
3029 error = -EINVAL;
3030 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003031 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003032 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033}
3034
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003035/*
3036 * same as __d_path but appends "(deleted)" for unlinked files.
3037 */
Al Viro02125a82011-12-05 08:43:34 -05003038static int path_with_deleted(const struct path *path,
3039 const struct path *root,
3040 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003041{
3042 prepend(buf, buflen, "\0", 1);
3043 if (d_unlinked(path->dentry)) {
3044 int error = prepend(buf, buflen, " (deleted)", 10);
3045 if (error)
3046 return error;
3047 }
3048
3049 return prepend_path(path, root, buf, buflen);
3050}
3051
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003052static int prepend_unreachable(char **buffer, int *buflen)
3053{
3054 return prepend(buffer, buflen, "(unreachable)", 13);
3055}
3056
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003057static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3058{
3059 unsigned seq;
3060
3061 do {
3062 seq = read_seqcount_begin(&fs->seq);
3063 *root = fs->root;
3064 } while (read_seqcount_retry(&fs->seq, seq));
3065}
3066
Jan Bluncka03a8a702008-02-14 19:38:32 -08003067/**
3068 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003069 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003070 * @buf: buffer to return value in
3071 * @buflen: buffer length
3072 *
3073 * Convert a dentry into an ASCII path name. If the entry has been deleted
3074 * the string " (deleted)" is appended. Note that this is ambiguous.
3075 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003076 * Returns a pointer into the buffer or an error code if the path was
3077 * too long. Note: Callers should use the returned pointer, not the passed
3078 * in buffer, to use the name! The implementation often starts at an offset
3079 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003080 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003081 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003082 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003083char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003085 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003086 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003087 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003089 /*
3090 * We have various synthetic filesystems that never get mounted. On
3091 * these filesystems dentries are never used for lookup purposes, and
3092 * thus don't need to be hashed. They also don't need a name until a
3093 * user wants to identify the object in /proc/pid/fd/. The little hack
3094 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003095 *
3096 * Some pseudo inodes are mountable. When they are mounted
3097 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3098 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003099 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003100 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3101 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003102 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003103
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003104 rcu_read_lock();
3105 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003106 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003107 rcu_read_unlock();
3108
Al Viro02125a82011-12-05 08:43:34 -05003109 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003110 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 return res;
3112}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003113EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114
3115/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003116 * Helper function for dentry_operations.d_dname() members
3117 */
3118char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3119 const char *fmt, ...)
3120{
3121 va_list args;
3122 char temp[64];
3123 int sz;
3124
3125 va_start(args, fmt);
3126 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3127 va_end(args);
3128
3129 if (sz > sizeof(temp) || sz > buflen)
3130 return ERR_PTR(-ENAMETOOLONG);
3131
3132 buffer += buflen - sz;
3133 return memcpy(buffer, temp, sz);
3134}
3135
Al Viro118b2302013-08-24 12:08:17 -04003136char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3137{
3138 char *end = buffer + buflen;
3139 /* these dentries are never renamed, so d_lock is not needed */
3140 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003141 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003142 prepend(&end, &buflen, "/", 1))
3143 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003144 return end;
Al Viro118b2302013-08-24 12:08:17 -04003145}
David Herrmann31bbe162014-01-03 14:09:47 +01003146EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003147
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003148/*
Ram Pai6092d042008-03-27 13:06:20 +01003149 * Write full pathname from the root of the filesystem into the buffer.
3150 */
Al Virof6500802014-01-26 12:37:55 -05003151static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003152{
Al Virof6500802014-01-26 12:37:55 -05003153 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003154 char *end, *retval;
3155 int len, seq = 0;
3156 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003157
Al Virof6500802014-01-26 12:37:55 -05003158 if (buflen < 2)
3159 goto Elong;
3160
Al Viro48f5ec22013-09-09 15:22:25 -04003161 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003162restart:
Al Virof6500802014-01-26 12:37:55 -05003163 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003164 end = buf + buflen;
3165 len = buflen;
3166 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003167 /* Get '/' right */
3168 retval = end-1;
3169 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003170 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003171 while (!IS_ROOT(dentry)) {
3172 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003173
Ram Pai6092d042008-03-27 13:06:20 +01003174 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003175 error = prepend_name(&end, &len, &dentry->d_name);
3176 if (error)
3177 break;
Ram Pai6092d042008-03-27 13:06:20 +01003178
3179 retval = end;
3180 dentry = parent;
3181 }
Al Viro48f5ec22013-09-09 15:22:25 -04003182 if (!(seq & 1))
3183 rcu_read_unlock();
3184 if (need_seqretry(&rename_lock, seq)) {
3185 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003186 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003187 }
3188 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003189 if (error)
3190 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003191 return retval;
3192Elong:
3193 return ERR_PTR(-ENAMETOOLONG);
3194}
Nick Pigginec2447c2011-01-07 17:49:29 +11003195
3196char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3197{
Waiman Long232d2d62013-09-09 12:18:13 -04003198 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003199}
3200EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003201
3202char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3203{
3204 char *p = NULL;
3205 char *retval;
3206
Al Viroc1031352010-06-06 22:31:14 -04003207 if (d_unlinked(dentry)) {
3208 p = buf + buflen;
3209 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3210 goto Elong;
3211 buflen++;
3212 }
3213 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003214 if (!IS_ERR(retval) && p)
3215 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003216 return retval;
3217Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003218 return ERR_PTR(-ENAMETOOLONG);
3219}
3220
Linus Torvalds8b19e342013-09-12 10:35:47 -07003221static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3222 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003223{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003224 unsigned seq;
3225
3226 do {
3227 seq = read_seqcount_begin(&fs->seq);
3228 *root = fs->root;
3229 *pwd = fs->pwd;
3230 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003231}
3232
Ram Pai6092d042008-03-27 13:06:20 +01003233/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234 * NOTE! The user-level library version returns a
3235 * character pointer. The kernel system call just
3236 * returns the length of the buffer filled (which
3237 * includes the ending '\0' character), or a negative
3238 * error value. So libc would do something like
3239 *
3240 * char *getcwd(char * buf, size_t size)
3241 * {
3242 * int retval;
3243 *
3244 * retval = sys_getcwd(buf, size);
3245 * if (retval >= 0)
3246 * return buf;
3247 * errno = -retval;
3248 * return NULL;
3249 * }
3250 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003251SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252{
Linus Torvalds552ce542007-02-13 12:08:18 -08003253 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003254 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003255 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256
3257 if (!page)
3258 return -ENOMEM;
3259
Linus Torvalds8b19e342013-09-12 10:35:47 -07003260 rcu_read_lock();
3261 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262
Linus Torvalds552ce542007-02-13 12:08:18 -08003263 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003264 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003265 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003266 char *cwd = page + PATH_MAX;
3267 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003269 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003270 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003271 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003272
Al Viro02125a82011-12-05 08:43:34 -05003273 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003274 goto out;
3275
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003276 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003277 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003278 error = prepend_unreachable(&cwd, &buflen);
3279 if (error)
3280 goto out;
3281 }
3282
Linus Torvalds552ce542007-02-13 12:08:18 -08003283 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003284 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003285 if (len <= size) {
3286 error = len;
3287 if (copy_to_user(buf, cwd, len))
3288 error = -EFAULT;
3289 }
Nick Piggin949854d2011-01-07 17:49:37 +11003290 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003291 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003292 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293
3294out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003295 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003296 return error;
3297}
3298
3299/*
3300 * Test whether new_dentry is a subdirectory of old_dentry.
3301 *
3302 * Trivially implemented using the dcache structure
3303 */
3304
3305/**
3306 * is_subdir - is new dentry a subdirectory of old_dentry
3307 * @new_dentry: new dentry
3308 * @old_dentry: old dentry
3309 *
3310 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3311 * Returns 0 otherwise.
3312 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3313 */
3314
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003315int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316{
3317 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11003318 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003320 if (new_dentry == old_dentry)
3321 return 1;
3322
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003323 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003326 /*
3327 * Need rcu_readlock to protect against the d_parent trashing
3328 * due to d_move
3329 */
3330 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003331 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003333 else
3334 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11003335 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003336 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337
3338 return result;
3339}
3340
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003341static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003343 struct dentry *root = data;
3344 if (dentry != root) {
3345 if (d_unhashed(dentry) || !dentry->d_inode)
3346 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003348 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3349 dentry->d_flags |= DCACHE_GENOCIDE;
3350 dentry->d_lockref.count--;
3351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003353 return D_WALK_CONTINUE;
3354}
Nick Piggin58db63d2011-01-07 17:49:39 +11003355
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003356void d_genocide(struct dentry *parent)
3357{
3358 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359}
3360
Al Viro60545d02013-06-07 01:20:27 -04003361void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362{
Al Viro60545d02013-06-07 01:20:27 -04003363 inode_dec_link_count(inode);
3364 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3365 !hlist_unhashed(&dentry->d_alias) ||
3366 !d_unlinked(dentry));
3367 spin_lock(&dentry->d_parent->d_lock);
3368 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3369 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3370 (unsigned long long)inode->i_ino);
3371 spin_unlock(&dentry->d_lock);
3372 spin_unlock(&dentry->d_parent->d_lock);
3373 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374}
Al Viro60545d02013-06-07 01:20:27 -04003375EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376
3377static __initdata unsigned long dhash_entries;
3378static int __init set_dhash_entries(char *str)
3379{
3380 if (!str)
3381 return 0;
3382 dhash_entries = simple_strtoul(str, &str, 0);
3383 return 1;
3384}
3385__setup("dhash_entries=", set_dhash_entries);
3386
3387static void __init dcache_init_early(void)
3388{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003389 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003390
3391 /* If hashes are distributed across NUMA nodes, defer
3392 * hash allocation until vmalloc space is available.
3393 */
3394 if (hashdist)
3395 return;
3396
3397 dentry_hashtable =
3398 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003399 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 dhash_entries,
3401 13,
3402 HASH_EARLY,
3403 &d_hash_shift,
3404 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003405 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406 0);
3407
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003408 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003409 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410}
3411
Denis Cheng74bf17c2007-10-16 23:26:30 -07003412static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003414 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415
3416 /*
3417 * A constructor could be added for stable state like the lists,
3418 * but it is probably not worth it because of the cache nature
3419 * of the dcache.
3420 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003421 dentry_cache = KMEM_CACHE(dentry,
3422 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003423
3424 /* Hash may have been set up in dcache_init_early */
3425 if (!hashdist)
3426 return;
3427
3428 dentry_hashtable =
3429 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003430 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431 dhash_entries,
3432 13,
3433 0,
3434 &d_hash_shift,
3435 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003436 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003437 0);
3438
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003439 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003440 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441}
3442
3443/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003444struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003445EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447EXPORT_SYMBOL(d_genocide);
3448
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449void __init vfs_caches_init_early(void)
3450{
3451 dcache_init_early();
3452 inode_init_early();
3453}
3454
3455void __init vfs_caches_init(unsigned long mempages)
3456{
3457 unsigned long reserve;
3458
3459 /* Base hash sizes on available memory, with a reserve equal to
3460 150% of current kernel size */
3461
3462 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3463 mempages -= reserve;
3464
3465 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003466 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003467
Denis Cheng74bf17c2007-10-16 23:26:30 -07003468 dcache_init();
3469 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003470 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003471 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472 bdev_cache_init();
3473 chrdev_init();
3474}