blob: e368d4f412f97ed11039e3329c77358b3fc1fea0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/dcache.c
3 *
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
8
9/*
10 * Notes on the allocation strategy:
11 *
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040021#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/hash.h>
25#include <linux/cache.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/file.h>
29#include <asm/uaccess.h>
30#include <linux/security.h>
31#include <linux/seqlock.h>
32#include <linux/swap.h>
33#include <linux/bootmem.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Frederic Weisbecker613afbf2009-07-16 15:44:29 +020035#include <linux/hardirq.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110036#include <linux/bit_spinlock.h>
37#include <linux/rculist_bl.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070038#include <linux/prefetch.h>
David Howellsdd179942011-08-16 15:31:30 +010039#include <linux/ratelimit.h>
Dave Chinnerf6041562013-08-28 10:18:00 +100040#include <linux/list_lru.h>
David Howells07f3f05c2006-09-30 20:52:18 +020041#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050042#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Nick Piggin789680d2011-01-07 17:49:30 +110044/*
45 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110046 * dcache->d_inode->i_lock protects:
Al Viro946e51f2014-10-26 19:19:16 -040047 * - i_dentry, d_u.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
Al Viro946e51f2014-10-26 19:19:16 -040062 * - d_u.d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110063 *
64 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110065 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110066 * dentry->d_lock
Dave Chinner19156842013-08-28 10:17:55 +100067 * dentry->d_sb->s_dentry_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110068 * dcache_hash_bucket lock
69 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110070 *
Nick Pigginda502952011-01-07 17:49:33 +110071 * If there is an ancestor relationship:
72 * dentry->d_parent->...->d_parent->d_lock
73 * ...
74 * dentry->d_parent->d_lock
75 * dentry->d_lock
76 *
77 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110078 * if (dentry1 < dentry2)
79 * dentry1->d_lock
80 * dentry2->d_lock
81 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080082int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
84
Al Viro74c3cbe2007-07-22 08:04:18 -040085__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Nick Piggin949854d2011-01-07 17:49:37 +110087EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Christoph Lametere18b8902006-12-06 20:33:20 -080089static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/*
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
95 *
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
98 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800100static unsigned int d_hash_mask __read_mostly;
101static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100102
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700103static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvalds8966be92012-03-02 14:23:30 -0800105static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700106 unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100107{
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700108 hash += (unsigned long) parent / L1_CACHE_BYTES;
Linus Torvalds99d263d2014-09-13 11:30:10 -0700109 return dentry_hashtable + hash_32(hash, d_hash_shift);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100110}
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112/* Statistics gathering. */
113struct dentry_stat_t dentry_stat = {
114 .age_limit = 45,
115};
116
Glauber Costa3942c072013-08-28 10:17:53 +1000117static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000118static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400119
120#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000121
122/*
123 * Here we resort to our own counters instead of using generic per-cpu counters
124 * for consistency with what the vfs inode code does. We are expected to harvest
125 * better code and performance by having our own specialized counters.
126 *
127 * Please note that the loop is done over all possible CPUs, not over all online
128 * CPUs. The reason for this is that we don't want to play games with CPUs going
129 * on and off. If one of them goes off, we will just keep their counters.
130 *
131 * glommer: See cffbc8a for details, and if you ever intend to change this,
132 * please update all vfs counters to match.
133 */
Glauber Costa3942c072013-08-28 10:17:53 +1000134static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100135{
136 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000137 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100138 for_each_possible_cpu(i)
139 sum += per_cpu(nr_dentry, i);
140 return sum < 0 ? 0 : sum;
141}
142
Dave Chinner62d36c72013-08-28 10:17:54 +1000143static long get_nr_dentry_unused(void)
144{
145 int i;
146 long sum = 0;
147 for_each_possible_cpu(i)
148 sum += per_cpu(nr_dentry_unused, i);
149 return sum < 0 ? 0 : sum;
150}
151
Joe Perches1f7e0612014-06-06 14:38:05 -0700152int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400153 size_t *lenp, loff_t *ppos)
154{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100155 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000156 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000157 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400158}
159#endif
160
Linus Torvalds5483f182012-03-04 15:51:42 -0800161/*
162 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
163 * The strings are both count bytes long, and count is non-zero.
164 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700165#ifdef CONFIG_DCACHE_WORD_ACCESS
166
167#include <asm/word-at-a-time.h>
168/*
169 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
170 * aligned allocation for this particular component. We don't
171 * strictly need the load_unaligned_zeropad() safety, but it
172 * doesn't hurt either.
173 *
174 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
175 * need the careful unaligned handling.
176 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700177static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800178{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800179 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800180
181 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700182 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700183 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800184 if (tcount < sizeof(unsigned long))
185 break;
186 if (unlikely(a != b))
187 return 1;
188 cs += sizeof(unsigned long);
189 ct += sizeof(unsigned long);
190 tcount -= sizeof(unsigned long);
191 if (!tcount)
192 return 0;
193 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000194 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800195 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700196}
197
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800198#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700199
Linus Torvalds94753db52012-05-10 12:19:19 -0700200static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700201{
Linus Torvalds5483f182012-03-04 15:51:42 -0800202 do {
203 if (*cs != *ct)
204 return 1;
205 cs++;
206 ct++;
207 tcount--;
208 } while (tcount);
209 return 0;
210}
211
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700212#endif
213
Linus Torvalds94753db52012-05-10 12:19:19 -0700214static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
215{
Linus Torvalds6326c712012-05-21 16:14:04 -0700216 const unsigned char *cs;
Linus Torvalds94753db52012-05-10 12:19:19 -0700217 /*
218 * Be careful about RCU walk racing with rename:
219 * use ACCESS_ONCE to fetch the name pointer.
220 *
221 * NOTE! Even if a rename will mean that the length
222 * was not loaded atomically, we don't care. The
223 * RCU walk will check the sequence count eventually,
224 * and catch it. And we won't overrun the buffer,
225 * because we're reading the name pointer atomically,
226 * and a dentry name is guaranteed to be properly
227 * terminated with a NUL byte.
228 *
229 * End result: even if 'len' is wrong, we'll exit
230 * early because the data cannot match (there can
231 * be no NUL in the ct/tcount data)
232 */
Linus Torvalds6326c712012-05-21 16:14:04 -0700233 cs = ACCESS_ONCE(dentry->d_name.name);
234 smp_read_barrier_depends();
235 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700236}
237
Al Viro8d85b482014-09-29 14:54:27 -0400238struct external_name {
239 union {
240 atomic_t count;
241 struct rcu_head head;
242 } u;
243 unsigned char name[];
244};
245
246static inline struct external_name *external_name(struct dentry *dentry)
247{
248 return container_of(dentry->d_name.name, struct external_name, name[0]);
249}
250
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400251static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400253 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
254
Al Viro8d85b482014-09-29 14:54:27 -0400255 kmem_cache_free(dentry_cache, dentry);
256}
257
258static void __d_free_external(struct rcu_head *head)
259{
260 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
Al Viro8d85b482014-09-29 14:54:27 -0400261 kfree(external_name(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 kmem_cache_free(dentry_cache, dentry);
263}
264
Al Viro810bb172014-10-12 12:45:37 -0400265static inline int dname_external(const struct dentry *dentry)
266{
267 return dentry->d_name.name != dentry->d_iname;
268}
269
Al Virob4f03542014-04-29 23:40:14 -0400270static void dentry_free(struct dentry *dentry)
271{
Al Viro946e51f2014-10-26 19:19:16 -0400272 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400273 if (unlikely(dname_external(dentry))) {
274 struct external_name *p = external_name(dentry);
275 if (likely(atomic_dec_and_test(&p->u.count))) {
276 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
277 return;
278 }
279 }
Al Virob4f03542014-04-29 23:40:14 -0400280 /* if dentry was never visible to RCU, immediate free is OK */
281 if (!(dentry->d_flags & DCACHE_RCUACCESS))
282 __d_free(&dentry->d_u.d_rcu);
283 else
284 call_rcu(&dentry->d_u.d_rcu, __d_free);
285}
286
Nick Piggin31e6b012011-01-07 17:49:52 +1100287/**
288 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800289 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100290 * After this call, in-progress rcu-walk path lookup will fail. This
291 * should be called after unhashing, and after changing d_inode (if
292 * the dentry has not already been unhashed).
293 */
294static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
295{
296 assert_spin_locked(&dentry->d_lock);
297 /* Go through a barrier */
298 write_seqcount_barrier(&dentry->d_seq);
299}
300
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301/*
302 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100303 * d_iput() operation if defined. Dentry has no refcount
304 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800306static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200307 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100308 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309{
310 struct inode *inode = dentry->d_inode;
311 if (inode) {
312 dentry->d_inode = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400313 hlist_del_init(&dentry->d_u.d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100315 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700316 if (!inode->i_nlink)
317 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 if (dentry->d_op && dentry->d_op->d_iput)
319 dentry->d_op->d_iput(dentry, inode);
320 else
321 iput(inode);
322 } else {
323 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 }
325}
326
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700327/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100328 * Release the dentry's inode, using the filesystem
329 * d_iput() operation if defined. dentry remains in-use.
330 */
331static void dentry_unlink_inode(struct dentry * dentry)
332 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100333 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100334{
335 struct inode *inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +0100336 __d_clear_type(dentry);
Nick Piggin31e6b012011-01-07 17:49:52 +1100337 dentry->d_inode = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400338 hlist_del_init(&dentry->d_u.d_alias);
Nick Piggin31e6b012011-01-07 17:49:52 +1100339 dentry_rcuwalk_barrier(dentry);
340 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100341 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100342 if (!inode->i_nlink)
343 fsnotify_inoderemove(inode);
344 if (dentry->d_op && dentry->d_op->d_iput)
345 dentry->d_op->d_iput(dentry, inode);
346 else
347 iput(inode);
348}
349
350/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400351 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
352 * is in use - which includes both the "real" per-superblock
353 * LRU list _and_ the DCACHE_SHRINK_LIST use.
354 *
355 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
356 * on the shrink list (ie not on the superblock LRU list).
357 *
358 * The per-cpu "nr_dentry_unused" counters are updated with
359 * the DCACHE_LRU_LIST bit.
360 *
361 * These helper functions make sure we always follow the
362 * rules. d_lock must be held by the caller.
363 */
364#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
365static void d_lru_add(struct dentry *dentry)
366{
367 D_FLAG_VERIFY(dentry, 0);
368 dentry->d_flags |= DCACHE_LRU_LIST;
369 this_cpu_inc(nr_dentry_unused);
370 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
371}
372
373static void d_lru_del(struct dentry *dentry)
374{
375 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
376 dentry->d_flags &= ~DCACHE_LRU_LIST;
377 this_cpu_dec(nr_dentry_unused);
378 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
379}
380
381static void d_shrink_del(struct dentry *dentry)
382{
383 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
384 list_del_init(&dentry->d_lru);
385 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
386 this_cpu_dec(nr_dentry_unused);
387}
388
389static void d_shrink_add(struct dentry *dentry, struct list_head *list)
390{
391 D_FLAG_VERIFY(dentry, 0);
392 list_add(&dentry->d_lru, list);
393 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
394 this_cpu_inc(nr_dentry_unused);
395}
396
397/*
398 * These can only be called under the global LRU lock, ie during the
399 * callback for freeing the LRU list. "isolate" removes it from the
400 * LRU lists entirely, while shrink_move moves it to the indicated
401 * private list.
402 */
403static void d_lru_isolate(struct dentry *dentry)
404{
405 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
406 dentry->d_flags &= ~DCACHE_LRU_LIST;
407 this_cpu_dec(nr_dentry_unused);
408 list_del_init(&dentry->d_lru);
409}
410
411static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
412{
413 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
414 dentry->d_flags |= DCACHE_SHRINK_LIST;
415 list_move_tail(&dentry->d_lru, list);
416}
417
418/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000419 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700420 */
421static void dentry_lru_add(struct dentry *dentry)
422{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400423 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
424 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700425}
426
Miklos Szeredid52b9082007-05-08 00:23:46 -0700427/**
Nick Piggin789680d2011-01-07 17:49:30 +1100428 * d_drop - drop a dentry
429 * @dentry: dentry to drop
430 *
431 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
432 * be found through a VFS lookup any more. Note that this is different from
433 * deleting the dentry - d_delete will try to mark the dentry negative if
434 * possible, giving a successful _negative_ lookup, while d_drop will
435 * just make the cache lookup fail.
436 *
437 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
438 * reason (NFS timeouts or autofs deletes).
439 *
440 * __d_drop requires dentry->d_lock.
441 */
442void __d_drop(struct dentry *dentry)
443{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700444 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400445 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400446 /*
447 * Hashed dentries are normally on the dentry hashtable,
448 * with the exception of those newly allocated by
449 * d_obtain_alias, which are always IS_ROOT:
450 */
451 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400452 b = &dentry->d_sb->s_anon;
453 else
454 b = d_hash(dentry->d_parent, dentry->d_name.hash);
455
456 hlist_bl_lock(b);
457 __hlist_bl_del(&dentry->d_hash);
458 dentry->d_hash.pprev = NULL;
459 hlist_bl_unlock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700460 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100461 }
462}
463EXPORT_SYMBOL(__d_drop);
464
465void d_drop(struct dentry *dentry)
466{
Nick Piggin789680d2011-01-07 17:49:30 +1100467 spin_lock(&dentry->d_lock);
468 __d_drop(dentry);
469 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100470}
471EXPORT_SYMBOL(d_drop);
472
Al Viroe55fd012014-05-28 13:51:12 -0400473static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100474{
Al Viro41edf272014-05-01 10:30:00 -0400475 struct dentry *parent = NULL;
476 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400477 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100478 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100479
Linus Torvalds0d984392013-09-08 13:46:52 -0700480 /*
481 * The dentry is now unrecoverably dead to the world.
482 */
483 lockref_mark_dead(&dentry->d_lockref);
484
Sage Weilf0023bc2011-10-28 10:02:42 -0700485 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700486 * inform the fs via d_prune that this dentry is about to be
487 * unhashed and destroyed.
488 */
Al Viro29266202014-05-30 11:39:02 -0400489 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800490 dentry->d_op->d_prune(dentry);
491
Al Viro01b60352014-04-29 23:42:52 -0400492 if (dentry->d_flags & DCACHE_LRU_LIST) {
493 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
494 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400495 }
Nick Piggin77812a12011-01-07 17:49:48 +1100496 /* if it was on the hash then remove it */
497 __d_drop(dentry);
Al Viroca5358e2014-10-26 19:31:10 -0400498 __list_del_entry(&dentry->d_child);
Al Viro03b3b882014-04-29 15:45:28 -0400499 /*
500 * Inform d_walk() that we are no longer attached to the
501 * dentry tree
502 */
503 dentry->d_flags |= DCACHE_DENTRY_KILLED;
504 if (parent)
505 spin_unlock(&parent->d_lock);
506 dentry_iput(dentry);
507 /*
508 * dentry_iput drops the locks, at which point nobody (except
509 * transient RCU lookups) can reach this dentry.
510 */
511 BUG_ON((int)dentry->d_lockref.count > 0);
512 this_cpu_dec(nr_dentry);
513 if (dentry->d_op && dentry->d_op->d_release)
514 dentry->d_op->d_release(dentry);
515
Al Viro41edf272014-05-01 10:30:00 -0400516 spin_lock(&dentry->d_lock);
517 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
518 dentry->d_flags |= DCACHE_MAY_FREE;
519 can_free = false;
520 }
521 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400522 if (likely(can_free))
523 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400524}
525
526/*
527 * Finish off a dentry we've decided to kill.
528 * dentry->d_lock must be held, returns with it unlocked.
529 * If ref is non-zero, then decrement the refcount too.
530 * Returns dentry requiring refcount drop, or NULL if we're done.
531 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400532static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400533 __releases(dentry->d_lock)
534{
535 struct inode *inode = dentry->d_inode;
536 struct dentry *parent = NULL;
537
538 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
539 goto failed;
540
541 if (!IS_ROOT(dentry)) {
542 parent = dentry->d_parent;
543 if (unlikely(!spin_trylock(&parent->d_lock))) {
544 if (inode)
545 spin_unlock(&inode->i_lock);
546 goto failed;
547 }
548 }
549
550 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400551 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400552
553failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400554 spin_unlock(&dentry->d_lock);
555 cpu_relax();
Al Viroe55fd012014-05-28 13:51:12 -0400556 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100557}
558
Al Viro046b9612014-05-29 08:54:52 -0400559static inline struct dentry *lock_parent(struct dentry *dentry)
560{
561 struct dentry *parent = dentry->d_parent;
562 if (IS_ROOT(dentry))
563 return NULL;
Al Viroc2338f22014-06-12 00:29:13 -0400564 if (unlikely((int)dentry->d_lockref.count < 0))
565 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400566 if (likely(spin_trylock(&parent->d_lock)))
567 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400568 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400569 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400570again:
571 parent = ACCESS_ONCE(dentry->d_parent);
572 spin_lock(&parent->d_lock);
573 /*
574 * We can't blindly lock dentry until we are sure
575 * that we won't violate the locking order.
576 * Any changes of dentry->d_parent must have
577 * been done with parent->d_lock held, so
578 * spin_lock() above is enough of a barrier
579 * for checking if it's still our child.
580 */
581 if (unlikely(parent != dentry->d_parent)) {
582 spin_unlock(&parent->d_lock);
583 goto again;
584 }
585 rcu_read_unlock();
586 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700587 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400588 else
589 parent = NULL;
590 return parent;
591}
592
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593/*
594 * This is dput
595 *
596 * This is complicated by the fact that we do not want to put
597 * dentries that are no longer on any hash chain on the unused
598 * list: we'd much rather just get rid of them immediately.
599 *
600 * However, that implies that we have to traverse the dentry
601 * tree upwards to the parents which might _also_ now be
602 * scheduled for deletion (it may have been only waiting for
603 * its last child to go away).
604 *
605 * This tail recursion is done by hand as we don't want to depend
606 * on the compiler to always get this right (gcc generally doesn't).
607 * Real recursion would eat up our stack space.
608 */
609
610/*
611 * dput - release a dentry
612 * @dentry: dentry to release
613 *
614 * Release a dentry. This will drop the usage count and if appropriate
615 * call the dentry unlink method as well as removing it from the queues and
616 * releasing its resources. If the parent dentries were scheduled for release
617 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619void dput(struct dentry *dentry)
620{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700621 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 return;
623
624repeat:
Waiman Long98474232013-08-28 18:24:59 -0700625 if (lockref_put_or_lock(&dentry->d_lockref))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700628 /* Unreachable? Get rid of it */
629 if (unlikely(d_unhashed(dentry)))
630 goto kill_it;
631
632 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100634 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 }
Nick Piggin265ac902010-10-10 05:36:24 -0400636
Linus Torvalds358eec12013-10-31 15:43:02 -0700637 if (!(dentry->d_flags & DCACHE_REFERENCED))
638 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400639 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400640
Waiman Long98474232013-08-28 18:24:59 -0700641 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100642 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 return;
644
Miklos Szeredid52b9082007-05-08 00:23:46 -0700645kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400646 dentry = dentry_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700647 if (dentry)
648 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700650EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100653/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100654static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655{
Waiman Long98474232013-08-28 18:24:59 -0700656 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657}
658
Nick Piggindc0474b2011-01-07 17:49:43 +1100659static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100660{
Waiman Long98474232013-08-28 18:24:59 -0700661 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100662}
663
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100664struct dentry *dget_parent(struct dentry *dentry)
665{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700666 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100667 struct dentry *ret;
668
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700669 /*
670 * Do optimistic parent lookup without any
671 * locking.
672 */
673 rcu_read_lock();
674 ret = ACCESS_ONCE(dentry->d_parent);
675 gotref = lockref_get_not_zero(&ret->d_lockref);
676 rcu_read_unlock();
677 if (likely(gotref)) {
678 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
679 return ret;
680 dput(ret);
681 }
682
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100683repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100684 /*
685 * Don't need rcu_dereference because we re-check it was correct under
686 * the lock.
687 */
688 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100689 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100690 spin_lock(&ret->d_lock);
691 if (unlikely(ret != dentry->d_parent)) {
692 spin_unlock(&ret->d_lock);
693 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100694 goto repeat;
695 }
Nick Piggina734eb42011-01-07 17:49:44 +1100696 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700697 BUG_ON(!ret->d_lockref.count);
698 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100699 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100700 return ret;
701}
702EXPORT_SYMBOL(dget_parent);
703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704/**
705 * d_find_alias - grab a hashed alias of inode
706 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 *
708 * If inode has a hashed alias, or is a directory and has any alias,
709 * acquire the reference to alias and return it. Otherwise return NULL.
710 * Notice that if inode is a directory there can be only one alias and
711 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800712 * of a filesystem, or if the directory was renamed and d_revalidate
713 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700715 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500716 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500718static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719{
Nick Pigginda502952011-01-07 17:49:33 +1100720 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
Nick Pigginda502952011-01-07 17:49:33 +1100722again:
723 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400724 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100725 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700727 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100728 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500730 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100731 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100732 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 return alias;
734 }
735 }
Nick Pigginda502952011-01-07 17:49:33 +1100736 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 }
Nick Pigginda502952011-01-07 17:49:33 +1100738 if (discon_alias) {
739 alias = discon_alias;
740 spin_lock(&alias->d_lock);
741 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500742 __dget_dlock(alias);
743 spin_unlock(&alias->d_lock);
744 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100745 }
746 spin_unlock(&alias->d_lock);
747 goto again;
748 }
749 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750}
751
Nick Pigginda502952011-01-07 17:49:33 +1100752struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753{
David Howells214fda12006-03-25 03:06:36 -0800754 struct dentry *de = NULL;
755
Al Virob3d9b7a2012-06-09 13:51:19 -0400756 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100757 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500758 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100759 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800760 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 return de;
762}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700763EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
765/*
766 * Try to kill dentries associated with this inode.
767 * WARNING: you must own a reference to inode.
768 */
769void d_prune_aliases(struct inode *inode)
770{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700771 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100773 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400774 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700776 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400777 struct dentry *parent = lock_parent(dentry);
778 if (likely(!dentry->d_lockref.count)) {
779 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800780 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400781 goto restart;
782 }
783 if (parent)
784 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 }
786 spin_unlock(&dentry->d_lock);
787 }
Nick Piggin873feea2011-01-07 17:50:06 +1100788 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700790EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400792static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
Al Viro5c47e6d2014-04-29 16:13:18 -0400794 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700795
Miklos Szeredi60942f22014-05-02 15:38:39 -0400796 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400797 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400798 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100799 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400800 parent = lock_parent(dentry);
801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000803 * The dispose list is isolated and dentries are not accounted
804 * to the LRU here, so we can simply remove it from the list
805 * here regardless of whether it is referenced or not.
806 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400807 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000808
809 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000811 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 */
Al Viro41edf272014-05-01 10:30:00 -0400813 if ((int)dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700814 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400815 if (parent)
816 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 continue;
818 }
Nick Piggin77812a12011-01-07 17:49:48 +1100819
Al Viro64fd72e2014-05-28 09:48:44 -0400820
821 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
822 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
823 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400824 if (parent)
825 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -0400826 if (can_free)
827 dentry_free(dentry);
828 continue;
829 }
830
Al Viroff2fde92014-05-28 13:59:13 -0400831 inode = dentry->d_inode;
832 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400833 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000834 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400835 if (parent)
836 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400837 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000838 }
Al Viroff2fde92014-05-28 13:59:13 -0400839
Al Viroff2fde92014-05-28 13:59:13 -0400840 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -0400841
Al Viro5c47e6d2014-04-29 16:13:18 -0400842 /*
843 * We need to prune ancestors too. This is necessary to prevent
844 * quadratic behavior of shrink_dcache_parent(), but is also
845 * expected to be beneficial in reducing dentry cache
846 * fragmentation.
847 */
848 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -0400849 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
850 parent = lock_parent(dentry);
851 if (dentry->d_lockref.count != 1) {
852 dentry->d_lockref.count--;
853 spin_unlock(&dentry->d_lock);
854 if (parent)
855 spin_unlock(&parent->d_lock);
856 break;
857 }
858 inode = dentry->d_inode; /* can't be NULL */
859 if (unlikely(!spin_trylock(&inode->i_lock))) {
860 spin_unlock(&dentry->d_lock);
861 if (parent)
862 spin_unlock(&parent->d_lock);
863 cpu_relax();
864 continue;
865 }
866 __dentry_kill(dentry);
867 dentry = parent;
868 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400870}
871
Dave Chinnerf6041562013-08-28 10:18:00 +1000872static enum lru_status
873dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
874{
875 struct list_head *freeable = arg;
876 struct dentry *dentry = container_of(item, struct dentry, d_lru);
877
878
879 /*
880 * we are inverting the lru lock/dentry->d_lock here,
881 * so use a trylock. If we fail to get the lock, just skip
882 * it
883 */
884 if (!spin_trylock(&dentry->d_lock))
885 return LRU_SKIP;
886
887 /*
888 * Referenced dentries are still in use. If they have active
889 * counts, just remove them from the LRU. Otherwise give them
890 * another pass through the LRU.
891 */
892 if (dentry->d_lockref.count) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400893 d_lru_isolate(dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +1000894 spin_unlock(&dentry->d_lock);
895 return LRU_REMOVED;
896 }
897
898 if (dentry->d_flags & DCACHE_REFERENCED) {
899 dentry->d_flags &= ~DCACHE_REFERENCED;
900 spin_unlock(&dentry->d_lock);
901
902 /*
903 * The list move itself will be made by the common LRU code. At
904 * this point, we've dropped the dentry->d_lock but keep the
905 * lru lock. This is safe to do, since every list movement is
906 * protected by the lru lock even if both locks are held.
907 *
908 * This is guaranteed by the fact that all LRU management
909 * functions are intermediated by the LRU API calls like
910 * list_lru_add and list_lru_del. List movement in this file
911 * only ever occur through this functions or through callbacks
912 * like this one, that are called from the LRU API.
913 *
914 * The only exceptions to this are functions like
915 * shrink_dentry_list, and code that first checks for the
916 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
917 * operating only with stack provided lists after they are
918 * properly isolated from the main list. It is thus, always a
919 * local access.
920 */
921 return LRU_ROTATE;
922 }
923
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400924 d_lru_shrink_move(dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +1000925 spin_unlock(&dentry->d_lock);
926
927 return LRU_REMOVED;
928}
929
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400930/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000931 * prune_dcache_sb - shrink the dcache
932 * @sb: superblock
Dave Chinnerf6041562013-08-28 10:18:00 +1000933 * @nr_to_scan : number of entries to try to free
Dave Chinner9b17c622013-08-28 10:18:05 +1000934 * @nid: which node to scan for freeable entities
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400935 *
Dave Chinnerf6041562013-08-28 10:18:00 +1000936 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000937 * done when we need more memory an called from the superblock shrinker
938 * function.
939 *
940 * This function may fail to free any resources if all the dentries are in
941 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400942 */
Dave Chinner9b17c622013-08-28 10:18:05 +1000943long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
944 int nid)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400945{
Dave Chinnerf6041562013-08-28 10:18:00 +1000946 LIST_HEAD(dispose);
947 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400948
Dave Chinner9b17c622013-08-28 10:18:05 +1000949 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
950 &dispose, &nr_to_scan);
Dave Chinnerf6041562013-08-28 10:18:00 +1000951 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +1000952 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953}
954
Glauber Costa4e717f52013-08-28 10:18:03 +1000955static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
956 spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000957{
Glauber Costa4e717f52013-08-28 10:18:03 +1000958 struct list_head *freeable = arg;
959 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000960
Glauber Costa4e717f52013-08-28 10:18:03 +1000961 /*
962 * we are inverting the lru lock/dentry->d_lock here,
963 * so use a trylock. If we fail to get the lock, just skip
964 * it
965 */
966 if (!spin_trylock(&dentry->d_lock))
967 return LRU_SKIP;
968
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400969 d_lru_shrink_move(dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +1000970 spin_unlock(&dentry->d_lock);
971
972 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000973}
974
Glauber Costa4e717f52013-08-28 10:18:03 +1000975
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700976/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 * shrink_dcache_sb - shrink dcache for a superblock
978 * @sb: superblock
979 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400980 * Shrink the dcache for the specified super block. This is used to free
981 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400983void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984{
Glauber Costa4e717f52013-08-28 10:18:03 +1000985 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400986
Glauber Costa4e717f52013-08-28 10:18:03 +1000987 do {
988 LIST_HEAD(dispose);
989
990 freed = list_lru_walk(&sb->s_dentry_lru,
991 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
992
993 this_cpu_sub(nr_dentry_unused, freed);
994 shrink_dentry_list(&dispose);
995 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700997EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001000 * enum d_walk_ret - action to talke during tree walk
1001 * @D_WALK_CONTINUE: contrinue walk
1002 * @D_WALK_QUIT: quit walk
1003 * @D_WALK_NORETRY: quit when retry is needed
1004 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001006enum d_walk_ret {
1007 D_WALK_CONTINUE,
1008 D_WALK_QUIT,
1009 D_WALK_NORETRY,
1010 D_WALK_SKIP,
1011};
1012
1013/**
1014 * d_walk - walk the dentry tree
1015 * @parent: start of walk
1016 * @data: data passed to @enter() and @finish()
1017 * @enter: callback when first entering the dentry
1018 * @finish: callback when successfully finished the walk
1019 *
1020 * The @enter() and @finish() callbacks are called with d_lock held.
1021 */
1022static void d_walk(struct dentry *parent, void *data,
1023 enum d_walk_ret (*enter)(void *, struct dentry *),
1024 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
Nick Piggin949854d2011-01-07 17:49:37 +11001026 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001028 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001029 enum d_walk_ret ret;
1030 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001031
Nick Piggin58db63d2011-01-07 17:49:39 +11001032again:
Al Viro48f5ec22013-09-09 15:22:25 -04001033 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001034 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001035 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001036
1037 ret = enter(data, this_parent);
1038 switch (ret) {
1039 case D_WALK_CONTINUE:
1040 break;
1041 case D_WALK_QUIT:
1042 case D_WALK_SKIP:
1043 goto out_unlock;
1044 case D_WALK_NORETRY:
1045 retry = false;
1046 break;
1047 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048repeat:
1049 next = this_parent->d_subdirs.next;
1050resume:
1051 while (next != &this_parent->d_subdirs) {
1052 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001053 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001055
1056 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001057
1058 ret = enter(data, dentry);
1059 switch (ret) {
1060 case D_WALK_CONTINUE:
1061 break;
1062 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001063 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001064 goto out_unlock;
1065 case D_WALK_NORETRY:
1066 retry = false;
1067 break;
1068 case D_WALK_SKIP:
1069 spin_unlock(&dentry->d_lock);
1070 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001071 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001072
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001074 spin_unlock(&this_parent->d_lock);
1075 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001077 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 goto repeat;
1079 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001080 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 }
1082 /*
1083 * All done at this level ... ascend and resume the search.
1084 */
Al Viroca5358e2014-10-26 19:31:10 -04001085 rcu_read_lock();
1086ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001088 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001089 this_parent = child->d_parent;
1090
Al Viro31dec132013-10-25 17:04:27 -04001091 spin_unlock(&child->d_lock);
1092 spin_lock(&this_parent->d_lock);
1093
Al Viroca5358e2014-10-26 19:31:10 -04001094 /* might go back up the wrong parent if we have had a rename. */
1095 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001096 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001097 next = child->d_child.next;
1098 while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
1099 if (next == &this_parent->d_subdirs)
1100 goto ascend;
1101 child = list_entry(next, struct dentry, d_child);
1102 next = next->next;
Al Viro31dec132013-10-25 17:04:27 -04001103 }
1104 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 goto resume;
1106 }
Al Viroca5358e2014-10-26 19:31:10 -04001107 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001108 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001109 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001110 if (finish)
1111 finish(data);
1112
1113out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001114 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001115 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001116 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001117
1118rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001119 spin_unlock(&this_parent->d_lock);
1120 rcu_read_unlock();
1121 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001122 if (!retry)
1123 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001124 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001125 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001127
1128/*
1129 * Search for at least 1 mount point in the dentry's subdirs.
1130 * We descend to the next level whenever the d_subdirs
1131 * list is non-empty and continue searching.
1132 */
1133
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001134static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1135{
1136 int *ret = data;
1137 if (d_mountpoint(dentry)) {
1138 *ret = 1;
1139 return D_WALK_QUIT;
1140 }
1141 return D_WALK_CONTINUE;
1142}
1143
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001144/**
1145 * have_submounts - check for mounts over a dentry
1146 * @parent: dentry to check.
1147 *
1148 * Return true if the parent or its subdirectories contain
1149 * a mount point
1150 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001151int have_submounts(struct dentry *parent)
1152{
1153 int ret = 0;
1154
1155 d_walk(parent, &ret, check_mount, NULL);
1156
1157 return ret;
1158}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001159EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160
1161/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001162 * Called by mount code to set a mountpoint and check if the mountpoint is
1163 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1164 * subtree can become unreachable).
1165 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001166 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001167 * this reason take rename_lock and d_lock on dentry and ancestors.
1168 */
1169int d_set_mounted(struct dentry *dentry)
1170{
1171 struct dentry *p;
1172 int ret = -ENOENT;
1173 write_seqlock(&rename_lock);
1174 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001175 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001176 spin_lock(&p->d_lock);
1177 if (unlikely(d_unhashed(p))) {
1178 spin_unlock(&p->d_lock);
1179 goto out;
1180 }
1181 spin_unlock(&p->d_lock);
1182 }
1183 spin_lock(&dentry->d_lock);
1184 if (!d_unlinked(dentry)) {
1185 dentry->d_flags |= DCACHE_MOUNTED;
1186 ret = 0;
1187 }
1188 spin_unlock(&dentry->d_lock);
1189out:
1190 write_sequnlock(&rename_lock);
1191 return ret;
1192}
1193
1194/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001195 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 * and move any unused dentries to the end of the unused
1197 * list for prune_dcache(). We descend to the next level
1198 * whenever the d_subdirs list is non-empty and continue
1199 * searching.
1200 *
1201 * It returns zero iff there are no unused children,
1202 * otherwise it returns the number of children moved to
1203 * the end of the unused list. This may not be the total
1204 * number of unused children, because select_parent can
1205 * drop the lock and return early due to latency
1206 * constraints.
1207 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001208
1209struct select_data {
1210 struct dentry *start;
1211 struct list_head dispose;
1212 int found;
1213};
1214
1215static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001217 struct select_data *data = _data;
1218 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001220 if (data->start == dentry)
1221 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
Al Virofe915222014-05-03 00:02:25 -04001223 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001224 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001225 } else {
1226 if (dentry->d_flags & DCACHE_LRU_LIST)
1227 d_lru_del(dentry);
1228 if (!dentry->d_lockref.count) {
1229 d_shrink_add(dentry, &data->dispose);
1230 data->found++;
1231 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 }
1233 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001234 * We can return to the caller if we have found some (this
1235 * ensures forward progress). We'll be coming back to find
1236 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 */
Al Virofe915222014-05-03 00:02:25 -04001238 if (!list_empty(&data->dispose))
1239 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001241 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242}
1243
1244/**
1245 * shrink_dcache_parent - prune dcache
1246 * @parent: parent of entries to prune
1247 *
1248 * Prune the dcache to remove unused children of the parent dentry.
1249 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001250void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001252 for (;;) {
1253 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001255 INIT_LIST_HEAD(&data.dispose);
1256 data.start = parent;
1257 data.found = 0;
1258
1259 d_walk(parent, &data, select_collect, NULL);
1260 if (!data.found)
1261 break;
1262
1263 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001264 cond_resched();
1265 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001267EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268
Al Viro9c8c10e2014-05-02 20:36:10 -04001269static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001270{
Al Viro9c8c10e2014-05-02 20:36:10 -04001271 /* it has busy descendents; complain about those instead */
1272 if (!list_empty(&dentry->d_subdirs))
1273 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001274
Al Viro9c8c10e2014-05-02 20:36:10 -04001275 /* root with refcount 1 is fine */
1276 if (dentry == _data && dentry->d_lockref.count == 1)
1277 return D_WALK_CONTINUE;
1278
1279 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1280 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001281 dentry,
1282 dentry->d_inode ?
1283 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001284 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001285 dentry->d_lockref.count,
1286 dentry->d_sb->s_type->name,
1287 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001288 WARN_ON(1);
1289 return D_WALK_CONTINUE;
1290}
1291
1292static void do_one_tree(struct dentry *dentry)
1293{
1294 shrink_dcache_parent(dentry);
1295 d_walk(dentry, dentry, umount_check, NULL);
1296 d_drop(dentry);
1297 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001298}
1299
1300/*
1301 * destroy the dentries attached to a superblock on unmounting
1302 */
1303void shrink_dcache_for_umount(struct super_block *sb)
1304{
1305 struct dentry *dentry;
1306
Al Viro9c8c10e2014-05-02 20:36:10 -04001307 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001308
1309 dentry = sb->s_root;
1310 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001311 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001312
1313 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001314 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1315 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001316 }
1317}
1318
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001319struct detach_data {
1320 struct select_data select;
1321 struct dentry *mountpoint;
1322};
1323static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001324{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001325 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001326
1327 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001328 __dget_dlock(dentry);
1329 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001330 return D_WALK_QUIT;
1331 }
1332
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001333 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001334}
1335
1336static void check_and_drop(void *_data)
1337{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001338 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001339
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001340 if (!data->mountpoint && !data->select.found)
1341 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001342}
1343
1344/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001345 * d_invalidate - detach submounts, prune dcache, and drop
1346 * @dentry: dentry to invalidate (aka detach, prune and drop)
1347 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001348 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001349 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001350 * The final d_drop is done as an atomic operation relative to
1351 * rename_lock ensuring there are no races with d_set_mounted. This
1352 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001353 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001354void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001355{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001356 /*
1357 * If it's already been dropped, return OK.
1358 */
1359 spin_lock(&dentry->d_lock);
1360 if (d_unhashed(dentry)) {
1361 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001362 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001363 }
1364 spin_unlock(&dentry->d_lock);
1365
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001366 /* Negative dentries can be dropped without further checks */
1367 if (!dentry->d_inode) {
1368 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001369 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001370 }
1371
1372 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001373 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001374
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001375 data.mountpoint = NULL;
1376 INIT_LIST_HEAD(&data.select.dispose);
1377 data.select.start = dentry;
1378 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001379
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001380 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001381
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001382 if (data.select.found)
1383 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001384
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001385 if (data.mountpoint) {
1386 detach_mounts(data.mountpoint);
1387 dput(data.mountpoint);
1388 }
1389
1390 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001391 break;
1392
1393 cond_resched();
1394 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001395}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001396EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001397
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398/**
Al Viroa4464db2011-07-07 15:03:58 -04001399 * __d_alloc - allocate a dcache entry
1400 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 * @name: qstr of the name
1402 *
1403 * Allocates a dentry. It returns %NULL if there is insufficient memory
1404 * available. On a success the dentry is returned. The name passed in is
1405 * copied and the copy passed in may be reused after this call.
1406 */
1407
Al Viroa4464db2011-07-07 15:03:58 -04001408struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409{
1410 struct dentry *dentry;
1411 char *dname;
1412
Mel Gormane12ba742007-10-16 01:25:52 -07001413 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 if (!dentry)
1415 return NULL;
1416
Linus Torvalds6326c712012-05-21 16:14:04 -07001417 /*
1418 * We guarantee that the inline name is always NUL-terminated.
1419 * This way the memcpy() done by the name switching in rename
1420 * will still always have a NUL at the end, even if we might
1421 * be overwriting an internal NUL character
1422 */
1423 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001425 size_t size = offsetof(struct external_name, name[1]);
1426 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1427 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 kmem_cache_free(dentry_cache, dentry);
1429 return NULL;
1430 }
Al Viro8d85b482014-09-29 14:54:27 -04001431 atomic_set(&p->u.count, 1);
1432 dname = p->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 } else {
1434 dname = dentry->d_iname;
1435 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
1437 dentry->d_name.len = name->len;
1438 dentry->d_name.hash = name->hash;
1439 memcpy(dname, name->name, name->len);
1440 dname[name->len] = 0;
1441
Linus Torvalds6326c712012-05-21 16:14:04 -07001442 /* Make sure we always see the terminating NUL character */
1443 smp_wmb();
1444 dentry->d_name.name = dname;
1445
Waiman Long98474232013-08-28 18:24:59 -07001446 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001447 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001449 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001451 dentry->d_parent = dentry;
1452 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 dentry->d_op = NULL;
1454 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001455 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 INIT_LIST_HEAD(&dentry->d_lru);
1457 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001458 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1459 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001460 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461
Nick Piggin3e880fb2011-01-07 17:49:19 +11001462 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001463
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 return dentry;
1465}
Al Viroa4464db2011-07-07 15:03:58 -04001466
1467/**
1468 * d_alloc - allocate a dcache entry
1469 * @parent: parent of entry to allocate
1470 * @name: qstr of the name
1471 *
1472 * Allocates a dentry. It returns %NULL if there is insufficient memory
1473 * available. On a success the dentry is returned. The name passed in is
1474 * copied and the copy passed in may be reused after this call.
1475 */
1476struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1477{
1478 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1479 if (!dentry)
1480 return NULL;
1481
1482 spin_lock(&parent->d_lock);
1483 /*
1484 * don't need child lock because it is not subject
1485 * to concurrency here
1486 */
1487 __dget_dlock(parent);
1488 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001489 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001490 spin_unlock(&parent->d_lock);
1491
1492 return dentry;
1493}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001494EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001496/**
1497 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1498 * @sb: the superblock
1499 * @name: qstr of the name
1500 *
1501 * For a filesystem that just pins its dentries in memory and never
1502 * performs lookups at all, return an unhashed IS_ROOT dentry.
1503 */
Nick Piggin4b936882011-01-07 17:50:07 +11001504struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1505{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001506 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001507}
1508EXPORT_SYMBOL(d_alloc_pseudo);
1509
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1511{
1512 struct qstr q;
1513
1514 q.name = name;
1515 q.len = strlen(name);
1516 q.hash = full_name_hash(q.name, q.len);
1517 return d_alloc(parent, &q);
1518}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001519EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520
Nick Pigginfb045ad2011-01-07 17:49:55 +11001521void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1522{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001523 WARN_ON_ONCE(dentry->d_op);
1524 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001525 DCACHE_OP_COMPARE |
1526 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001527 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001528 DCACHE_OP_DELETE ));
1529 dentry->d_op = op;
1530 if (!op)
1531 return;
1532 if (op->d_hash)
1533 dentry->d_flags |= DCACHE_OP_HASH;
1534 if (op->d_compare)
1535 dentry->d_flags |= DCACHE_OP_COMPARE;
1536 if (op->d_revalidate)
1537 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001538 if (op->d_weak_revalidate)
1539 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001540 if (op->d_delete)
1541 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001542 if (op->d_prune)
1543 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001544
1545}
1546EXPORT_SYMBOL(d_set_d_op);
1547
David Howellsb18825a2013-09-12 19:22:53 +01001548static unsigned d_flags_for_inode(struct inode *inode)
1549{
1550 unsigned add_flags = DCACHE_FILE_TYPE;
1551
1552 if (!inode)
1553 return DCACHE_MISS_TYPE;
1554
1555 if (S_ISDIR(inode->i_mode)) {
1556 add_flags = DCACHE_DIRECTORY_TYPE;
1557 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1558 if (unlikely(!inode->i_op->lookup))
1559 add_flags = DCACHE_AUTODIR_TYPE;
1560 else
1561 inode->i_opflags |= IOP_LOOKUP;
1562 }
1563 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1564 if (unlikely(inode->i_op->follow_link))
1565 add_flags = DCACHE_SYMLINK_TYPE;
1566 else
1567 inode->i_opflags |= IOP_NOFOLLOW;
1568 }
1569
1570 if (unlikely(IS_AUTOMOUNT(inode)))
1571 add_flags |= DCACHE_NEED_AUTOMOUNT;
1572 return add_flags;
1573}
1574
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001575static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1576{
David Howellsb18825a2013-09-12 19:22:53 +01001577 unsigned add_flags = d_flags_for_inode(inode);
1578
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001579 spin_lock(&dentry->d_lock);
Al Viro22213312014-04-19 12:30:58 -04001580 __d_set_type(dentry, add_flags);
David Howellsb18825a2013-09-12 19:22:53 +01001581 if (inode)
Al Viro946e51f2014-10-26 19:19:16 -04001582 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001583 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001584 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001585 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001586 fsnotify_d_instantiate(dentry, inode);
1587}
1588
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589/**
1590 * d_instantiate - fill in inode information for a dentry
1591 * @entry: dentry to complete
1592 * @inode: inode to attach to this dentry
1593 *
1594 * Fill in inode information in the entry.
1595 *
1596 * This turns negative dentries into productive full members
1597 * of society.
1598 *
1599 * NOTE! This assumes that the inode count has been incremented
1600 * (or otherwise set) by the caller to indicate that it is now
1601 * in use by the dcache.
1602 */
1603
1604void d_instantiate(struct dentry *entry, struct inode * inode)
1605{
Al Viro946e51f2014-10-26 19:19:16 -04001606 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001607 if (inode)
1608 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001609 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001610 if (inode)
1611 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 security_d_instantiate(entry, inode);
1613}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001614EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615
1616/**
1617 * d_instantiate_unique - instantiate a non-aliased dentry
1618 * @entry: dentry to instantiate
1619 * @inode: inode to attach to this dentry
1620 *
1621 * Fill in inode information in the entry. On success, it returns NULL.
1622 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001623 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 *
1625 * Note that in order to avoid conflicts with rename() etc, the caller
1626 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001627 *
1628 * This also assumes that the inode count has been incremented
1629 * (or otherwise set) by the caller to indicate that it is now
1630 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 */
David Howells770bfad2006-08-22 20:06:07 -04001632static struct dentry *__d_instantiate_unique(struct dentry *entry,
1633 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634{
1635 struct dentry *alias;
1636 int len = entry->d_name.len;
1637 const char *name = entry->d_name.name;
1638 unsigned int hash = entry->d_name.hash;
1639
David Howells770bfad2006-08-22 20:06:07 -04001640 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001641 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001642 return NULL;
1643 }
1644
Al Viro946e51f2014-10-26 19:19:16 -04001645 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001646 /*
1647 * Don't need alias->d_lock here, because aliases with
1648 * d_parent == entry->d_parent are not subject to name or
1649 * parent changes, because the parent inode i_mutex is held.
1650 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001651 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 continue;
1653 if (alias->d_parent != entry->d_parent)
1654 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001655 if (alias->d_name.len != len)
1656 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001657 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001659 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 return alias;
1661 }
David Howells770bfad2006-08-22 20:06:07 -04001662
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001663 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 return NULL;
1665}
David Howells770bfad2006-08-22 20:06:07 -04001666
1667struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1668{
1669 struct dentry *result;
1670
Al Viro946e51f2014-10-26 19:19:16 -04001671 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001672
Nick Piggin873feea2011-01-07 17:50:06 +11001673 if (inode)
1674 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001675 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001676 if (inode)
1677 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001678
1679 if (!result) {
1680 security_d_instantiate(entry, inode);
1681 return NULL;
1682 }
1683
1684 BUG_ON(!d_unhashed(result));
1685 iput(inode);
1686 return result;
1687}
1688
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689EXPORT_SYMBOL(d_instantiate_unique);
1690
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001691/**
1692 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1693 * @entry: dentry to complete
1694 * @inode: inode to attach to this dentry
1695 *
1696 * Fill in inode information in the entry. If a directory alias is found, then
1697 * return an error (and drop inode). Together with d_materialise_unique() this
1698 * guarantees that a directory inode may never have more than one alias.
1699 */
1700int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1701{
Al Viro946e51f2014-10-26 19:19:16 -04001702 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001703
1704 spin_lock(&inode->i_lock);
1705 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1706 spin_unlock(&inode->i_lock);
1707 iput(inode);
1708 return -EBUSY;
1709 }
1710 __d_instantiate(entry, inode);
1711 spin_unlock(&inode->i_lock);
1712 security_d_instantiate(entry, inode);
1713
1714 return 0;
1715}
1716EXPORT_SYMBOL(d_instantiate_no_diralias);
1717
Al Viroadc0e912012-01-08 16:49:21 -05001718struct dentry *d_make_root(struct inode *root_inode)
1719{
1720 struct dentry *res = NULL;
1721
1722 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001723 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001724
1725 res = __d_alloc(root_inode->i_sb, &name);
1726 if (res)
1727 d_instantiate(res, root_inode);
1728 else
1729 iput(root_inode);
1730 }
1731 return res;
1732}
1733EXPORT_SYMBOL(d_make_root);
1734
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001735static struct dentry * __d_find_any_alias(struct inode *inode)
1736{
1737 struct dentry *alias;
1738
Al Virob3d9b7a2012-06-09 13:51:19 -04001739 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001740 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001741 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001742 __dget(alias);
1743 return alias;
1744}
1745
Sage Weil46f72b32012-01-10 09:04:37 -08001746/**
1747 * d_find_any_alias - find any alias for a given inode
1748 * @inode: inode to find an alias for
1749 *
1750 * If any aliases exist for the given inode, take and return a
1751 * reference for one of them. If no aliases exist, return %NULL.
1752 */
1753struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001754{
1755 struct dentry *de;
1756
1757 spin_lock(&inode->i_lock);
1758 de = __d_find_any_alias(inode);
1759 spin_unlock(&inode->i_lock);
1760 return de;
1761}
Sage Weil46f72b32012-01-10 09:04:37 -08001762EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001763
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001764static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001765{
NeilBrownb911a6b2012-11-08 16:09:37 -08001766 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001767 struct dentry *tmp;
1768 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001769 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001770
1771 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001772 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001773 if (IS_ERR(inode))
1774 return ERR_CAST(inode);
1775
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001776 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001777 if (res)
1778 goto out_iput;
1779
Al Viroa4464db2011-07-07 15:03:58 -04001780 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001781 if (!tmp) {
1782 res = ERR_PTR(-ENOMEM);
1783 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001784 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001785
Nick Piggin873feea2011-01-07 17:50:06 +11001786 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001787 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001788 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001789 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001790 dput(tmp);
1791 goto out_iput;
1792 }
1793
1794 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001795 add_flags = d_flags_for_inode(inode);
1796
1797 if (disconnected)
1798 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001799
Christoph Hellwig9308a612008-08-11 15:49:12 +02001800 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001801 tmp->d_inode = inode;
David Howellsb18825a2013-09-12 19:22:53 +01001802 tmp->d_flags |= add_flags;
Al Viro946e51f2014-10-26 19:19:16 -04001803 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001804 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001805 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001806 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001807 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001808 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001809 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001810
Christoph Hellwig9308a612008-08-11 15:49:12 +02001811 return tmp;
1812
1813 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001814 if (res && !IS_ERR(res))
1815 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001816 iput(inode);
1817 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001818}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001819
1820/**
1821 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1822 * @inode: inode to allocate the dentry for
1823 *
1824 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1825 * similar open by handle operations. The returned dentry may be anonymous,
1826 * or may have a full name (if the inode was already in the cache).
1827 *
1828 * When called on a directory inode, we must ensure that the inode only ever
1829 * has one dentry. If a dentry is found, that is returned instead of
1830 * allocating a new one.
1831 *
1832 * On successful return, the reference to the inode has been transferred
1833 * to the dentry. In case of an error the reference on the inode is released.
1834 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1835 * be passed in and the error will be propagated to the return value,
1836 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1837 */
1838struct dentry *d_obtain_alias(struct inode *inode)
1839{
1840 return __d_obtain_alias(inode, 1);
1841}
Benny Halevyadc48722009-02-27 14:02:59 -08001842EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843
1844/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001845 * d_obtain_root - find or allocate a dentry for a given inode
1846 * @inode: inode to allocate the dentry for
1847 *
1848 * Obtain an IS_ROOT dentry for the root of a filesystem.
1849 *
1850 * We must ensure that directory inodes only ever have one dentry. If a
1851 * dentry is found, that is returned instead of allocating a new one.
1852 *
1853 * On successful return, the reference to the inode has been transferred
1854 * to the dentry. In case of an error the reference on the inode is
1855 * released. A %NULL or IS_ERR inode may be passed in and will be the
1856 * error will be propagate to the return value, with a %NULL @inode
1857 * replaced by ERR_PTR(-ESTALE).
1858 */
1859struct dentry *d_obtain_root(struct inode *inode)
1860{
1861 return __d_obtain_alias(inode, 0);
1862}
1863EXPORT_SYMBOL(d_obtain_root);
1864
1865/**
Barry Naujok94035402008-05-21 16:50:46 +10001866 * d_add_ci - lookup or allocate new dentry with case-exact name
1867 * @inode: the inode case-insensitive lookup has found
1868 * @dentry: the negative dentry that was passed to the parent's lookup func
1869 * @name: the case-exact name to be associated with the returned dentry
1870 *
1871 * This is to avoid filling the dcache with case-insensitive names to the
1872 * same inode, only the actual correct case is stored in the dcache for
1873 * case-insensitive filesystems.
1874 *
1875 * For a case-insensitive lookup match and if the the case-exact dentry
1876 * already exists in in the dcache, use it and return it.
1877 *
1878 * If no entry exists with the exact case name, allocate new dentry with
1879 * the exact case, and return the spliced entry.
1880 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001881struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001882 struct qstr *name)
1883{
Barry Naujok94035402008-05-21 16:50:46 +10001884 struct dentry *found;
1885 struct dentry *new;
1886
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001887 /*
1888 * First check if a dentry matching the name already exists,
1889 * if not go ahead and create it now.
1890 */
Barry Naujok94035402008-05-21 16:50:46 +10001891 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10001892 if (!found) {
1893 new = d_alloc(dentry->d_parent, name);
1894 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05001895 found = ERR_PTR(-ENOMEM);
Al Viro427c77d2014-10-12 18:46:26 -04001896 } else {
1897 found = d_splice_alias(inode, new);
1898 if (found) {
1899 dput(new);
1900 return found;
1901 }
1902 return new;
Barry Naujok94035402008-05-21 16:50:46 +10001903 }
Barry Naujok94035402008-05-21 16:50:46 +10001904 }
Barry Naujok94035402008-05-21 16:50:46 +10001905 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05001906 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001907}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001908EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001910/*
1911 * Do the slow-case of the dentry name compare.
1912 *
1913 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07001914 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001915 * filesystem can rely on them, and can use the 'name' and
1916 * 'len' information without worrying about walking off the
1917 * end of memory etc.
1918 *
1919 * Thus the read_seqcount_retry() and the "duplicate" info
1920 * in arguments (the low-level filesystem should not look
1921 * at the dentry inode or name contents directly, since
1922 * rename can change them while we're in RCU mode).
1923 */
1924enum slow_d_compare {
1925 D_COMP_OK,
1926 D_COMP_NOMATCH,
1927 D_COMP_SEQRETRY,
1928};
1929
1930static noinline enum slow_d_compare slow_dentry_cmp(
1931 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001932 struct dentry *dentry,
1933 unsigned int seq,
1934 const struct qstr *name)
1935{
1936 int tlen = dentry->d_name.len;
1937 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001938
1939 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1940 cpu_relax();
1941 return D_COMP_SEQRETRY;
1942 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07001943 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001944 return D_COMP_NOMATCH;
1945 return D_COMP_OK;
1946}
1947
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948/**
Nick Piggin31e6b012011-01-07 17:49:52 +11001949 * __d_lookup_rcu - search for a dentry (racy, store-free)
1950 * @parent: parent dentry
1951 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07001952 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11001953 * Returns: dentry, or NULL
1954 *
1955 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1956 * resolution (store-free path walking) design described in
1957 * Documentation/filesystems/path-lookup.txt.
1958 *
1959 * This is not to be used outside core vfs.
1960 *
1961 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1962 * held, and rcu_read_lock held. The returned dentry must not be stored into
1963 * without taking d_lock and checking d_seq sequence count against @seq
1964 * returned here.
1965 *
Linus Torvalds15570082013-09-02 11:38:06 -07001966 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11001967 * function.
1968 *
1969 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1970 * the returned dentry, so long as its parent's seqlock is checked after the
1971 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1972 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001973 *
1974 * NOTE! The caller *has* to check the resulting dentry against the sequence
1975 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11001976 */
Linus Torvalds8966be92012-03-02 14:23:30 -08001977struct dentry *__d_lookup_rcu(const struct dentry *parent,
1978 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07001979 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11001980{
Linus Torvalds26fe5752012-05-10 13:14:12 -07001981 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11001982 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001983 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001984 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001985 struct dentry *dentry;
1986
1987 /*
1988 * Note: There is significant duplication with __d_lookup_rcu which is
1989 * required to prevent single threaded performance regressions
1990 * especially on architectures where smp_rmb (in seqcounts) are costly.
1991 * Keep the two functions in sync.
1992 */
1993
1994 /*
1995 * The hash list is protected using RCU.
1996 *
1997 * Carefully use d_seq when comparing a candidate dentry, to avoid
1998 * races with d_move().
1999 *
2000 * It is possible that concurrent renames can mess up our list
2001 * walk here and result in missing our dentry, resulting in the
2002 * false-negative result. d_lookup() protects against concurrent
2003 * renames using rename_lock seqlock.
2004 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002005 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002006 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002007 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002008 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002009
Nick Piggin31e6b012011-01-07 17:49:52 +11002010seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002011 /*
2012 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002013 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002014 *
2015 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002016 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002017 *
2018 * NOTE! We do a "raw" seqcount_begin here. That means that
2019 * we don't wait for the sequence count to stabilize if it
2020 * is in the middle of a sequence change. If we do the slow
2021 * dentry compare, we will do seqretries until it is stable,
2022 * and if we end up with a successful lookup, we actually
2023 * want to exit RCU lookup anyway.
2024 */
2025 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002026 if (dentry->d_parent != parent)
2027 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002028 if (d_unhashed(dentry))
2029 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002030
2031 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002032 if (dentry->d_name.hash != hashlen_hash(hashlen))
2033 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002034 *seqp = seq;
2035 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002036 case D_COMP_OK:
2037 return dentry;
2038 case D_COMP_NOMATCH:
2039 continue;
2040 default:
2041 goto seqretry;
2042 }
2043 }
2044
Linus Torvalds26fe5752012-05-10 13:14:12 -07002045 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002046 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002047 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002048 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002049 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002050 }
2051 return NULL;
2052}
2053
2054/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 * d_lookup - search for a dentry
2056 * @parent: parent dentry
2057 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002058 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002060 * d_lookup searches the children of the parent dentry for the name in
2061 * question. If the dentry is found its reference count is incremented and the
2062 * dentry is returned. The caller must use dput to free the entry when it has
2063 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 */
Al Viroda2d8452013-01-24 18:29:34 -05002065struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066{
Nick Piggin31e6b012011-01-07 17:49:52 +11002067 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002068 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
Daeseok Younb8314f92014-08-11 11:46:53 +09002070 do {
2071 seq = read_seqbegin(&rename_lock);
2072 dentry = __d_lookup(parent, name);
2073 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 break;
2075 } while (read_seqretry(&rename_lock, seq));
2076 return dentry;
2077}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002078EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
Nick Piggin31e6b012011-01-07 17:49:52 +11002080/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002081 * __d_lookup - search for a dentry (racy)
2082 * @parent: parent dentry
2083 * @name: qstr of name we wish to find
2084 * Returns: dentry, or NULL
2085 *
2086 * __d_lookup is like d_lookup, however it may (rarely) return a
2087 * false-negative result due to unrelated rename activity.
2088 *
2089 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2090 * however it must be used carefully, eg. with a following d_lookup in
2091 * the case of failure.
2092 *
2093 * __d_lookup callers must be commented.
2094 */
Al Viroa713ca22013-01-24 18:27:00 -05002095struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096{
2097 unsigned int len = name->len;
2098 unsigned int hash = name->hash;
2099 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002100 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002101 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002102 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002103 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Nick Pigginb04f7842010-08-18 04:37:34 +10002105 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002106 * Note: There is significant duplication with __d_lookup_rcu which is
2107 * required to prevent single threaded performance regressions
2108 * especially on architectures where smp_rmb (in seqcounts) are costly.
2109 * Keep the two functions in sync.
2110 */
2111
2112 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002113 * The hash list is protected using RCU.
2114 *
2115 * Take d_lock when comparing a candidate dentry, to avoid races
2116 * with d_move().
2117 *
2118 * It is possible that concurrent renames can mess up our list
2119 * walk here and result in missing our dentry, resulting in the
2120 * false-negative result. d_lookup() protects against concurrent
2121 * renames using rename_lock seqlock.
2122 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002123 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002124 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 rcu_read_lock();
2126
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002127 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 if (dentry->d_name.hash != hash)
2130 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
2132 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 if (dentry->d_parent != parent)
2134 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002135 if (d_unhashed(dentry))
2136 goto next;
2137
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 /*
2139 * It is safe to compare names since d_move() cannot
2140 * change the qstr (protected by d_lock).
2141 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002142 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002143 int tlen = dentry->d_name.len;
2144 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002145 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 goto next;
2147 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002148 if (dentry->d_name.len != len)
2149 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002150 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 goto next;
2152 }
2153
Waiman Long98474232013-08-28 18:24:59 -07002154 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002155 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 spin_unlock(&dentry->d_lock);
2157 break;
2158next:
2159 spin_unlock(&dentry->d_lock);
2160 }
2161 rcu_read_unlock();
2162
2163 return found;
2164}
2165
2166/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002167 * d_hash_and_lookup - hash the qstr then search for a dentry
2168 * @dir: Directory to search in
2169 * @name: qstr of name we wish to find
2170 *
Al Viro4f522a22013-02-11 23:20:37 -05002171 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002172 */
2173struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2174{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002175 /*
2176 * Check for a fs-specific hash function. Note that we must
2177 * calculate the standard hash first, as the d_op->d_hash()
2178 * routine may choose to leave the hash value unchanged.
2179 */
2180 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002181 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002182 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002183 if (unlikely(err < 0))
2184 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002185 }
Al Viro4f522a22013-02-11 23:20:37 -05002186 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002187}
Al Viro4f522a22013-02-11 23:20:37 -05002188EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002189
2190/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002191 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002193 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 *
2195 * An insecure source has sent us a dentry, here we verify it and dget() it.
2196 * This is used by ncpfs in its readdir implementation.
2197 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002198 *
2199 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002201int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202{
Nick Piggin786a5e12011-01-07 17:49:16 +11002203 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002204
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002205 spin_lock(&dparent->d_lock);
Al Viro946e51f2014-10-26 19:19:16 -04002206 list_for_each_entry(child, &dparent->d_subdirs, d_child) {
Nick Piggin786a5e12011-01-07 17:49:16 +11002207 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002208 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002209 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002210 spin_unlock(&dentry->d_lock);
2211 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 return 1;
2213 }
2214 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002215 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002216
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 return 0;
2218}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002219EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220
2221/*
2222 * When a file is deleted, we have two options:
2223 * - turn this dentry into a negative dentry
2224 * - unhash this dentry and free it.
2225 *
2226 * Usually, we want to just turn this into
2227 * a negative dentry, but if anybody else is
2228 * currently using the dentry or the inode
2229 * we can't do that and we fall back on removing
2230 * it from the hash queues and waiting for
2231 * it to be deleted later when it has no users
2232 */
2233
2234/**
2235 * d_delete - delete a dentry
2236 * @dentry: The dentry to delete
2237 *
2238 * Turn the dentry into a negative dentry if possible, otherwise
2239 * remove it from the hash queues so it can be deleted later
2240 */
2241
2242void d_delete(struct dentry * dentry)
2243{
Nick Piggin873feea2011-01-07 17:50:06 +11002244 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002245 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 /*
2247 * Are we the only user?
2248 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002249again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002251 inode = dentry->d_inode;
2252 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002253 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002254 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002255 spin_unlock(&dentry->d_lock);
2256 cpu_relax();
2257 goto again;
2258 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002259 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002260 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002261 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 return;
2263 }
2264
2265 if (!d_unhashed(dentry))
2266 __d_drop(dentry);
2267
2268 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002269
2270 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002272EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002274static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002276 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002277 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002278 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002279 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002280 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281}
2282
David Howells770bfad2006-08-22 20:06:07 -04002283static void _d_rehash(struct dentry * entry)
2284{
2285 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2286}
2287
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288/**
2289 * d_rehash - add an entry back to the hash
2290 * @entry: dentry to add to the hash
2291 *
2292 * Adds a dentry to the hash according to its name.
2293 */
2294
2295void d_rehash(struct dentry * entry)
2296{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002298 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002301EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002303/**
2304 * dentry_update_name_case - update case insensitive dentry with a new name
2305 * @dentry: dentry to be updated
2306 * @name: new name
2307 *
2308 * Update a case insensitive dentry with new case of name.
2309 *
2310 * dentry must have been returned by d_lookup with name @name. Old and new
2311 * name lengths must match (ie. no d_compare which allows mismatched name
2312 * lengths).
2313 *
2314 * Parent inode i_mutex must be held over d_lookup and into this call (to
2315 * keep renames and concurrent inserts, and readdir(2) away).
2316 */
2317void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2318{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002319 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002320 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2321
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002322 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002323 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002324 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002325 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002326 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002327}
2328EXPORT_SYMBOL(dentry_update_name_case);
2329
Al Viro8d85b482014-09-29 14:54:27 -04002330static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331{
Al Viro8d85b482014-09-29 14:54:27 -04002332 if (unlikely(dname_external(target))) {
2333 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 /*
2335 * Both external: swap the pointers
2336 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002337 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 } else {
2339 /*
2340 * dentry:internal, target:external. Steal target's
2341 * storage and make target internal.
2342 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002343 memcpy(target->d_iname, dentry->d_name.name,
2344 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 dentry->d_name.name = target->d_name.name;
2346 target->d_name.name = target->d_iname;
2347 }
2348 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002349 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 /*
2351 * dentry:external, target:internal. Give dentry's
2352 * storage to target and make dentry internal
2353 */
2354 memcpy(dentry->d_iname, target->d_name.name,
2355 target->d_name.len + 1);
2356 target->d_name.name = dentry->d_name.name;
2357 dentry->d_name.name = dentry->d_iname;
2358 } else {
2359 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002360 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002362 unsigned int i;
2363 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002364 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2365 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002366 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2367 swap(((long *) &dentry->d_iname)[i],
2368 ((long *) &target->d_iname)[i]);
2369 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 }
2371 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002372 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373}
2374
Al Viro8d85b482014-09-29 14:54:27 -04002375static void copy_name(struct dentry *dentry, struct dentry *target)
2376{
2377 struct external_name *old_name = NULL;
2378 if (unlikely(dname_external(dentry)))
2379 old_name = external_name(dentry);
2380 if (unlikely(dname_external(target))) {
2381 atomic_inc(&external_name(target)->u.count);
2382 dentry->d_name = target->d_name;
2383 } else {
2384 memcpy(dentry->d_iname, target->d_name.name,
2385 target->d_name.len + 1);
2386 dentry->d_name.name = dentry->d_iname;
2387 dentry->d_name.hash_len = target->d_name.hash_len;
2388 }
2389 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2390 kfree_rcu(old_name, u.head);
2391}
2392
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002393static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2394{
2395 /*
2396 * XXXX: do we really need to take target->d_lock?
2397 */
2398 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2399 spin_lock(&target->d_parent->d_lock);
2400 else {
2401 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2402 spin_lock(&dentry->d_parent->d_lock);
2403 spin_lock_nested(&target->d_parent->d_lock,
2404 DENTRY_D_LOCK_NESTED);
2405 } else {
2406 spin_lock(&target->d_parent->d_lock);
2407 spin_lock_nested(&dentry->d_parent->d_lock,
2408 DENTRY_D_LOCK_NESTED);
2409 }
2410 }
2411 if (target < dentry) {
2412 spin_lock_nested(&target->d_lock, 2);
2413 spin_lock_nested(&dentry->d_lock, 3);
2414 } else {
2415 spin_lock_nested(&dentry->d_lock, 2);
2416 spin_lock_nested(&target->d_lock, 3);
2417 }
2418}
2419
Al Viro986c0192014-09-26 23:11:15 -04002420static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002421{
2422 if (target->d_parent != dentry->d_parent)
2423 spin_unlock(&dentry->d_parent->d_lock);
2424 if (target->d_parent != target)
2425 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002426 spin_unlock(&target->d_lock);
2427 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002428}
2429
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002431 * When switching names, the actual string doesn't strictly have to
2432 * be preserved in the target - because we're dropping the target
2433 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002434 * the new name before we switch, unless we are going to rehash
2435 * it. Note that if we *do* unhash the target, we are not allowed
2436 * to rehash it without giving it a new name/hash key - whether
2437 * we swap or overwrite the names here, resulting name won't match
2438 * the reality in filesystem; it's only there for d_path() purposes.
2439 * Note that all of this is happening under rename_lock, so the
2440 * any hash lookup seeing it in the middle of manipulations will
2441 * be discarded anyway. So we do not care what happens to the hash
2442 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002444/*
Al Viro18367502011-07-12 21:42:24 -04002445 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 * @dentry: entry to move
2447 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002448 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 *
2450 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002451 * dcache entries should not be moved in this way. Caller must hold
2452 * rename_lock, the i_mutex of the source and target directories,
2453 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002455static void __d_move(struct dentry *dentry, struct dentry *target,
2456 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 if (!dentry->d_inode)
2459 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2460
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002461 BUG_ON(d_ancestor(dentry, target));
2462 BUG_ON(d_ancestor(target, dentry));
2463
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002464 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
Nick Piggin31e6b012011-01-07 17:49:52 +11002466 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002467 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002468
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002469 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2470
2471 /*
2472 * Move the dentry to the target hash queue. Don't bother checking
2473 * for the same hash queue because of how unlikely it is.
2474 */
2475 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002476 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477
Miklos Szeredida1ce062014-04-01 17:08:43 +02002478 /*
2479 * Unhash the target (d_delete() is not usable here). If exchanging
2480 * the two dentries, then rehash onto the other's hash queue.
2481 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002483 if (exchange) {
2484 __d_rehash(target,
2485 d_hash(dentry->d_parent, dentry->d_name.hash));
2486 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002489 if (exchange)
2490 swap_names(dentry, target);
2491 else
2492 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493
Al Viro63cf4272014-09-26 23:06:14 -04002494 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002496 /* splicing a tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 dentry->d_parent = target->d_parent;
2498 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002499 list_del_init(&target->d_child);
2500 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002502 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002503 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002504 list_move(&target->d_child, &target->d_parent->d_subdirs);
2505 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002506 if (exchange)
2507 fsnotify_d_move(target);
2508 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 }
2510
Nick Piggin31e6b012011-01-07 17:49:52 +11002511 write_seqcount_end(&target->d_seq);
2512 write_seqcount_end(&dentry->d_seq);
2513
Al Viro986c0192014-09-26 23:11:15 -04002514 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002515}
2516
2517/*
2518 * d_move - move a dentry
2519 * @dentry: entry to move
2520 * @target: new dentry
2521 *
2522 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002523 * dcache entries should not be moved in this way. See the locking
2524 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002525 */
2526void d_move(struct dentry *dentry, struct dentry *target)
2527{
2528 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002529 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002531}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002532EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533
Miklos Szeredida1ce062014-04-01 17:08:43 +02002534/*
2535 * d_exchange - exchange two dentries
2536 * @dentry1: first dentry
2537 * @dentry2: second dentry
2538 */
2539void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2540{
2541 write_seqlock(&rename_lock);
2542
2543 WARN_ON(!dentry1->d_inode);
2544 WARN_ON(!dentry2->d_inode);
2545 WARN_ON(IS_ROOT(dentry1));
2546 WARN_ON(IS_ROOT(dentry2));
2547
2548 __d_move(dentry1, dentry2, true);
2549
2550 write_sequnlock(&rename_lock);
2551}
2552
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002553/**
2554 * d_ancestor - search for an ancestor
2555 * @p1: ancestor dentry
2556 * @p2: child dentry
2557 *
2558 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2559 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002560 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002561struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002562{
2563 struct dentry *p;
2564
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002565 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002566 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002567 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002568 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002569 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002570}
2571
2572/*
2573 * This helper attempts to cope with remotely renamed directories
2574 *
2575 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002576 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002577 *
2578 * Note: If ever the locking in lock_rename() changes, then please
2579 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002580 */
Al Virob5ae6b12014-10-12 22:16:02 -04002581static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002582 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002583{
2584 struct mutex *m1 = NULL, *m2 = NULL;
Al Virob5ae6b12014-10-12 22:16:02 -04002585 int ret = -EBUSY;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002586
2587 /* If alias and dentry share a parent, then no extra locks required */
2588 if (alias->d_parent == dentry->d_parent)
2589 goto out_unalias;
2590
Trond Myklebust9eaef272006-10-21 10:24:20 -07002591 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002592 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2593 goto out_err;
2594 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2595 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2596 goto out_err;
2597 m2 = &alias->d_parent->d_inode->i_mutex;
2598out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002599 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04002600 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002601out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002602 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002603 if (m2)
2604 mutex_unlock(m2);
2605 if (m1)
2606 mutex_unlock(m1);
2607 return ret;
2608}
2609
David Howells770bfad2006-08-22 20:06:07 -04002610/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002611 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2612 * @inode: the inode which may have a disconnected dentry
2613 * @dentry: a negative dentry which we want to point to the inode.
2614 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002615 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2616 * place of the given dentry and return it, else simply d_add the inode
2617 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002618 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002619 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2620 * we should error out: directories can't have multiple aliases.
2621 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002622 * This is needed in the lookup routine of any filesystem that is exportable
2623 * (via knfsd) so that we can build dcache paths to directories effectively.
2624 *
2625 * If a dentry was found and moved, then it is returned. Otherwise NULL
2626 * is returned. This matches the expected return value of ->lookup.
2627 *
2628 * Cluster filesystems may call this function with a negative, hashed dentry.
2629 * In that case, we know that the inode will be a regular file, and also this
2630 * will only occur during atomic_open. So we need to check for the dentry
2631 * being already hashed only in the final case.
2632 */
2633struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2634{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002635 if (IS_ERR(inode))
2636 return ERR_CAST(inode);
2637
David Howells770bfad2006-08-22 20:06:07 -04002638 BUG_ON(!d_unhashed(dentry));
2639
David Howells770bfad2006-08-22 20:06:07 -04002640 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002641 __d_instantiate(dentry, NULL);
Al Virob5ae6b12014-10-12 22:16:02 -04002642 goto out;
David Howells770bfad2006-08-22 20:06:07 -04002643 }
Nick Piggin873feea2011-01-07 17:50:06 +11002644 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002645 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04002646 struct dentry *new = __d_find_any_alias(inode);
2647 if (unlikely(new)) {
Al Viro18367502011-07-12 21:42:24 -04002648 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002649 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04002650 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002651 spin_unlock(&inode->i_lock);
2652 dput(new);
2653 new = ERR_PTR(-ELOOP);
2654 pr_warn_ratelimited(
2655 "VFS: Lookup of '%s' in %s %s"
2656 " would have caused loop\n",
2657 dentry->d_name.name,
2658 inode->i_sb->s_type->name,
2659 inode->i_sb->s_id);
2660 } else if (!IS_ROOT(new)) {
2661 int err = __d_unalias(inode, dentry, new);
2662 write_sequnlock(&rename_lock);
2663 if (err) {
2664 dput(new);
2665 new = ERR_PTR(err);
2666 }
Al Viro18367502011-07-12 21:42:24 -04002667 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04002668 __d_move(new, dentry, false);
2669 write_sequnlock(&rename_lock);
2670 spin_unlock(&inode->i_lock);
2671 security_d_instantiate(new, inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002672 }
Al Virob5ae6b12014-10-12 22:16:02 -04002673 iput(inode);
2674 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002675 }
David Howells770bfad2006-08-22 20:06:07 -04002676 }
Al Virob5ae6b12014-10-12 22:16:02 -04002677 /* already taking inode->i_lock, so d_add() by hand */
2678 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11002679 spin_unlock(&inode->i_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002680out:
2681 security_d_instantiate(dentry, inode);
2682 d_rehash(dentry);
2683 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04002684}
Al Virob5ae6b12014-10-12 22:16:02 -04002685EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04002686
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002687static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002688{
2689 *buflen -= namelen;
2690 if (*buflen < 0)
2691 return -ENAMETOOLONG;
2692 *buffer -= namelen;
2693 memcpy(*buffer, str, namelen);
2694 return 0;
2695}
2696
Waiman Long232d2d62013-09-09 12:18:13 -04002697/**
2698 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002699 * @buffer: buffer pointer
2700 * @buflen: allocated length of the buffer
2701 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002702 *
2703 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2704 * make sure that either the old or the new name pointer and length are
2705 * fetched. However, there may be mismatch between length and pointer.
2706 * The length cannot be trusted, we need to copy it byte-by-byte until
2707 * the length is reached or a null byte is found. It also prepends "/" at
2708 * the beginning of the name. The sequence number check at the caller will
2709 * retry it again when a d_move() does happen. So any garbage in the buffer
2710 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04002711 *
2712 * Data dependency barrier is needed to make sure that we see that terminating
2713 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04002714 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002715static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2716{
Waiman Long232d2d62013-09-09 12:18:13 -04002717 const char *dname = ACCESS_ONCE(name->name);
2718 u32 dlen = ACCESS_ONCE(name->len);
2719 char *p;
2720
Al Viro6d13f692014-09-29 14:46:30 -04002721 smp_read_barrier_depends();
2722
Waiman Long232d2d62013-09-09 12:18:13 -04002723 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04002724 if (*buflen < 0)
2725 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04002726 p = *buffer -= dlen + 1;
2727 *p++ = '/';
2728 while (dlen--) {
2729 char c = *dname++;
2730 if (!c)
2731 break;
2732 *p++ = c;
2733 }
2734 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002735}
2736
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002738 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002739 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002740 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002741 * @buffer: pointer to the end of the buffer
2742 * @buflen: pointer to buffer length
2743 *
Waiman Long18129972013-09-12 10:55:35 -04002744 * The function will first try to write out the pathname without taking any
2745 * lock other than the RCU read lock to make sure that dentries won't go away.
2746 * It only checks the sequence number of the global rename_lock as any change
2747 * in the dentry's d_seq will be preceded by changes in the rename_lock
2748 * sequence number. If the sequence number had been changed, it will restart
2749 * the whole pathname back-tracing sequence again by taking the rename_lock.
2750 * In this case, there is no need to take the RCU read lock as the recursive
2751 * parent pointer references will keep the dentry chain alive as long as no
2752 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002753 */
Al Viro02125a82011-12-05 08:43:34 -05002754static int prepend_path(const struct path *path,
2755 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002756 char **buffer, int *buflen)
2757{
Al Viroede4ceb2013-11-13 07:45:40 -05002758 struct dentry *dentry;
2759 struct vfsmount *vfsmnt;
2760 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002761 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002762 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002763 char *bptr;
2764 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002765
Al Viro48f5ec22013-09-09 15:22:25 -04002766 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002767restart_mnt:
2768 read_seqbegin_or_lock(&mount_lock, &m_seq);
2769 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002770 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002771restart:
2772 bptr = *buffer;
2773 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002774 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002775 dentry = path->dentry;
2776 vfsmnt = path->mnt;
2777 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002778 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002779 while (dentry != root->dentry || vfsmnt != root->mnt) {
2780 struct dentry * parent;
2781
2782 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002783 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002784 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002785 if (mnt != parent) {
2786 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2787 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002788 vfsmnt = &mnt->mnt;
2789 continue;
2790 }
2791 /*
2792 * Filesystems needing to implement special "root names"
2793 * should do so with ->d_dname()
2794 */
2795 if (IS_ROOT(dentry) &&
2796 (dentry->d_name.len != 1 ||
2797 dentry->d_name.name[0] != '/')) {
2798 WARN(1, "Root dentry has weird name <%.*s>\n",
2799 (int) dentry->d_name.len,
2800 dentry->d_name.name);
2801 }
2802 if (!error)
2803 error = is_mounted(vfsmnt) ? 1 : 2;
2804 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002805 }
2806 parent = dentry->d_parent;
2807 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002808 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002809 if (error)
2810 break;
2811
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002812 dentry = parent;
2813 }
Al Viro48f5ec22013-09-09 15:22:25 -04002814 if (!(seq & 1))
2815 rcu_read_unlock();
2816 if (need_seqretry(&rename_lock, seq)) {
2817 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002818 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002819 }
Al Viro48f5ec22013-09-09 15:22:25 -04002820 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002821
2822 if (!(m_seq & 1))
2823 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002824 if (need_seqretry(&mount_lock, m_seq)) {
2825 m_seq = 1;
2826 goto restart_mnt;
2827 }
2828 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002829
Waiman Long232d2d62013-09-09 12:18:13 -04002830 if (error >= 0 && bptr == *buffer) {
2831 if (--blen < 0)
2832 error = -ENAMETOOLONG;
2833 else
2834 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002835 }
Waiman Long232d2d62013-09-09 12:18:13 -04002836 *buffer = bptr;
2837 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002838 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002839}
2840
2841/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002842 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002843 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002844 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002845 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 * @buflen: buffer length
2847 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002848 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002850 * Returns a pointer into the buffer or an error code if the
2851 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002852 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002853 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002854 *
Al Viro02125a82011-12-05 08:43:34 -05002855 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 */
Al Viro02125a82011-12-05 08:43:34 -05002857char *__d_path(const struct path *path,
2858 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002859 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002861 char *res = buf + buflen;
2862 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002864 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002865 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002866
Al Viro02125a82011-12-05 08:43:34 -05002867 if (error < 0)
2868 return ERR_PTR(error);
2869 if (error > 0)
2870 return NULL;
2871 return res;
2872}
2873
2874char *d_absolute_path(const struct path *path,
2875 char *buf, int buflen)
2876{
2877 struct path root = {};
2878 char *res = buf + buflen;
2879 int error;
2880
2881 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002882 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05002883
2884 if (error > 1)
2885 error = -EINVAL;
2886 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002887 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002888 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889}
2890
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002891/*
2892 * same as __d_path but appends "(deleted)" for unlinked files.
2893 */
Al Viro02125a82011-12-05 08:43:34 -05002894static int path_with_deleted(const struct path *path,
2895 const struct path *root,
2896 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002897{
2898 prepend(buf, buflen, "\0", 1);
2899 if (d_unlinked(path->dentry)) {
2900 int error = prepend(buf, buflen, " (deleted)", 10);
2901 if (error)
2902 return error;
2903 }
2904
2905 return prepend_path(path, root, buf, buflen);
2906}
2907
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002908static int prepend_unreachable(char **buffer, int *buflen)
2909{
2910 return prepend(buffer, buflen, "(unreachable)", 13);
2911}
2912
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07002913static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
2914{
2915 unsigned seq;
2916
2917 do {
2918 seq = read_seqcount_begin(&fs->seq);
2919 *root = fs->root;
2920 } while (read_seqcount_retry(&fs->seq, seq));
2921}
2922
Jan Bluncka03a8a702008-02-14 19:38:32 -08002923/**
2924 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08002925 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08002926 * @buf: buffer to return value in
2927 * @buflen: buffer length
2928 *
2929 * Convert a dentry into an ASCII path name. If the entry has been deleted
2930 * the string " (deleted)" is appended. Note that this is ambiguous.
2931 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002932 * Returns a pointer into the buffer or an error code if the path was
2933 * too long. Note: Callers should use the returned pointer, not the passed
2934 * in buffer, to use the name! The implementation often starts at an offset
2935 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002936 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002937 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002938 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07002939char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002941 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002942 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002943 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002945 /*
2946 * We have various synthetic filesystems that never get mounted. On
2947 * these filesystems dentries are never used for lookup purposes, and
2948 * thus don't need to be hashed. They also don't need a name until a
2949 * user wants to identify the object in /proc/pid/fd/. The little hack
2950 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08002951 *
2952 * Some pseudo inodes are mountable. When they are mounted
2953 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
2954 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002955 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08002956 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
2957 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08002958 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002959
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07002960 rcu_read_lock();
2961 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05002962 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07002963 rcu_read_unlock();
2964
Al Viro02125a82011-12-05 08:43:34 -05002965 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002966 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 return res;
2968}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002969EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970
2971/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002972 * Helper function for dentry_operations.d_dname() members
2973 */
2974char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2975 const char *fmt, ...)
2976{
2977 va_list args;
2978 char temp[64];
2979 int sz;
2980
2981 va_start(args, fmt);
2982 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2983 va_end(args);
2984
2985 if (sz > sizeof(temp) || sz > buflen)
2986 return ERR_PTR(-ENAMETOOLONG);
2987
2988 buffer += buflen - sz;
2989 return memcpy(buffer, temp, sz);
2990}
2991
Al Viro118b2302013-08-24 12:08:17 -04002992char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
2993{
2994 char *end = buffer + buflen;
2995 /* these dentries are never renamed, so d_lock is not needed */
2996 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04002997 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04002998 prepend(&end, &buflen, "/", 1))
2999 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003000 return end;
Al Viro118b2302013-08-24 12:08:17 -04003001}
David Herrmann31bbe162014-01-03 14:09:47 +01003002EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003003
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003004/*
Ram Pai6092d042008-03-27 13:06:20 +01003005 * Write full pathname from the root of the filesystem into the buffer.
3006 */
Al Virof6500802014-01-26 12:37:55 -05003007static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003008{
Al Virof6500802014-01-26 12:37:55 -05003009 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003010 char *end, *retval;
3011 int len, seq = 0;
3012 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003013
Al Virof6500802014-01-26 12:37:55 -05003014 if (buflen < 2)
3015 goto Elong;
3016
Al Viro48f5ec22013-09-09 15:22:25 -04003017 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003018restart:
Al Virof6500802014-01-26 12:37:55 -05003019 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003020 end = buf + buflen;
3021 len = buflen;
3022 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003023 /* Get '/' right */
3024 retval = end-1;
3025 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003026 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003027 while (!IS_ROOT(dentry)) {
3028 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003029
Ram Pai6092d042008-03-27 13:06:20 +01003030 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003031 error = prepend_name(&end, &len, &dentry->d_name);
3032 if (error)
3033 break;
Ram Pai6092d042008-03-27 13:06:20 +01003034
3035 retval = end;
3036 dentry = parent;
3037 }
Al Viro48f5ec22013-09-09 15:22:25 -04003038 if (!(seq & 1))
3039 rcu_read_unlock();
3040 if (need_seqretry(&rename_lock, seq)) {
3041 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003042 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003043 }
3044 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003045 if (error)
3046 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003047 return retval;
3048Elong:
3049 return ERR_PTR(-ENAMETOOLONG);
3050}
Nick Pigginec2447c2011-01-07 17:49:29 +11003051
3052char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3053{
Waiman Long232d2d62013-09-09 12:18:13 -04003054 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003055}
3056EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003057
3058char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3059{
3060 char *p = NULL;
3061 char *retval;
3062
Al Viroc1031352010-06-06 22:31:14 -04003063 if (d_unlinked(dentry)) {
3064 p = buf + buflen;
3065 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3066 goto Elong;
3067 buflen++;
3068 }
3069 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003070 if (!IS_ERR(retval) && p)
3071 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003072 return retval;
3073Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003074 return ERR_PTR(-ENAMETOOLONG);
3075}
3076
Linus Torvalds8b19e342013-09-12 10:35:47 -07003077static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3078 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003079{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003080 unsigned seq;
3081
3082 do {
3083 seq = read_seqcount_begin(&fs->seq);
3084 *root = fs->root;
3085 *pwd = fs->pwd;
3086 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003087}
3088
Ram Pai6092d042008-03-27 13:06:20 +01003089/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 * NOTE! The user-level library version returns a
3091 * character pointer. The kernel system call just
3092 * returns the length of the buffer filled (which
3093 * includes the ending '\0' character), or a negative
3094 * error value. So libc would do something like
3095 *
3096 * char *getcwd(char * buf, size_t size)
3097 * {
3098 * int retval;
3099 *
3100 * retval = sys_getcwd(buf, size);
3101 * if (retval >= 0)
3102 * return buf;
3103 * errno = -retval;
3104 * return NULL;
3105 * }
3106 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003107SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108{
Linus Torvalds552ce542007-02-13 12:08:18 -08003109 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003110 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003111 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003112
3113 if (!page)
3114 return -ENOMEM;
3115
Linus Torvalds8b19e342013-09-12 10:35:47 -07003116 rcu_read_lock();
3117 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118
Linus Torvalds552ce542007-02-13 12:08:18 -08003119 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003120 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003121 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003122 char *cwd = page + PATH_MAX;
3123 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003125 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003126 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003127 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003128
Al Viro02125a82011-12-05 08:43:34 -05003129 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003130 goto out;
3131
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003132 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003133 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003134 error = prepend_unreachable(&cwd, &buflen);
3135 if (error)
3136 goto out;
3137 }
3138
Linus Torvalds552ce542007-02-13 12:08:18 -08003139 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003140 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003141 if (len <= size) {
3142 error = len;
3143 if (copy_to_user(buf, cwd, len))
3144 error = -EFAULT;
3145 }
Nick Piggin949854d2011-01-07 17:49:37 +11003146 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003147 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003148 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149
3150out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003151 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152 return error;
3153}
3154
3155/*
3156 * Test whether new_dentry is a subdirectory of old_dentry.
3157 *
3158 * Trivially implemented using the dcache structure
3159 */
3160
3161/**
3162 * is_subdir - is new dentry a subdirectory of old_dentry
3163 * @new_dentry: new dentry
3164 * @old_dentry: old dentry
3165 *
3166 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3167 * Returns 0 otherwise.
3168 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3169 */
3170
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003171int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172{
3173 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11003174 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003175
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003176 if (new_dentry == old_dentry)
3177 return 1;
3178
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003179 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003180 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003182 /*
3183 * Need rcu_readlock to protect against the d_parent trashing
3184 * due to d_move
3185 */
3186 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003187 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003189 else
3190 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11003191 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193
3194 return result;
3195}
3196
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003197static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003199 struct dentry *root = data;
3200 if (dentry != root) {
3201 if (d_unhashed(dentry) || !dentry->d_inode)
3202 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003204 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3205 dentry->d_flags |= DCACHE_GENOCIDE;
3206 dentry->d_lockref.count--;
3207 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003209 return D_WALK_CONTINUE;
3210}
Nick Piggin58db63d2011-01-07 17:49:39 +11003211
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003212void d_genocide(struct dentry *parent)
3213{
3214 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215}
3216
Al Viro60545d02013-06-07 01:20:27 -04003217void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218{
Al Viro60545d02013-06-07 01:20:27 -04003219 inode_dec_link_count(inode);
3220 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003221 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003222 !d_unlinked(dentry));
3223 spin_lock(&dentry->d_parent->d_lock);
3224 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3225 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3226 (unsigned long long)inode->i_ino);
3227 spin_unlock(&dentry->d_lock);
3228 spin_unlock(&dentry->d_parent->d_lock);
3229 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230}
Al Viro60545d02013-06-07 01:20:27 -04003231EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232
3233static __initdata unsigned long dhash_entries;
3234static int __init set_dhash_entries(char *str)
3235{
3236 if (!str)
3237 return 0;
3238 dhash_entries = simple_strtoul(str, &str, 0);
3239 return 1;
3240}
3241__setup("dhash_entries=", set_dhash_entries);
3242
3243static void __init dcache_init_early(void)
3244{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003245 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246
3247 /* If hashes are distributed across NUMA nodes, defer
3248 * hash allocation until vmalloc space is available.
3249 */
3250 if (hashdist)
3251 return;
3252
3253 dentry_hashtable =
3254 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003255 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256 dhash_entries,
3257 13,
3258 HASH_EARLY,
3259 &d_hash_shift,
3260 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003261 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262 0);
3263
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003264 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003265 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266}
3267
Denis Cheng74bf17c2007-10-16 23:26:30 -07003268static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003270 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271
3272 /*
3273 * A constructor could be added for stable state like the lists,
3274 * but it is probably not worth it because of the cache nature
3275 * of the dcache.
3276 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003277 dentry_cache = KMEM_CACHE(dentry,
3278 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279
3280 /* Hash may have been set up in dcache_init_early */
3281 if (!hashdist)
3282 return;
3283
3284 dentry_hashtable =
3285 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003286 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287 dhash_entries,
3288 13,
3289 0,
3290 &d_hash_shift,
3291 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003292 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293 0);
3294
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003295 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003296 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297}
3298
3299/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003300struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003301EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303EXPORT_SYMBOL(d_genocide);
3304
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305void __init vfs_caches_init_early(void)
3306{
3307 dcache_init_early();
3308 inode_init_early();
3309}
3310
3311void __init vfs_caches_init(unsigned long mempages)
3312{
3313 unsigned long reserve;
3314
3315 /* Base hash sizes on available memory, with a reserve equal to
3316 150% of current kernel size */
3317
3318 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3319 mempages -= reserve;
3320
3321 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003322 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323
Denis Cheng74bf17c2007-10-16 23:26:30 -07003324 dcache_init();
3325 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003327 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328 bdev_cache_init();
3329 chrdev_init();
3330}