blob: 58e26bee7ef463fe36f3f7ecbc84b7dfd1a34134 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/dcache.c
3 *
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
8
9/*
10 * Notes on the allocation strategy:
11 *
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040021#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/hash.h>
25#include <linux/cache.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/file.h>
29#include <asm/uaccess.h>
30#include <linux/security.h>
31#include <linux/seqlock.h>
32#include <linux/swap.h>
33#include <linux/bootmem.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Frederic Weisbecker613afbf2009-07-16 15:44:29 +020035#include <linux/hardirq.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110036#include <linux/bit_spinlock.h>
37#include <linux/rculist_bl.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070038#include <linux/prefetch.h>
David Howellsdd179942011-08-16 15:31:30 +010039#include <linux/ratelimit.h>
Dave Chinnerf6041562013-08-28 10:18:00 +100040#include <linux/list_lru.h>
David Howells07f3f052006-09-30 20:52:18 +020041#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050042#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Nick Piggin789680d2011-01-07 17:49:30 +110044/*
45 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110046 * dcache->d_inode->i_lock protects:
47 * - i_dentry, d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110048 * dcache_hash_bucket lock protects:
49 * - the dcache hash table
50 * s_anon bl list spinlock protects:
51 * - the s_anon list (see __d_drop)
Dave Chinner19156842013-08-28 10:17:55 +100052 * dentry->d_sb->s_dentry_lru_lock protects:
Nick Piggin23044502011-01-07 17:49:31 +110053 * - the dcache lru lists and counters
54 * d_lock protects:
55 * - d_flags
56 * - d_name
57 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110058 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110059 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110060 * - d_parent and d_subdirs
61 * - childrens' d_child and d_parent
Nick Pigginb23fb0a2011-01-07 17:49:35 +110062 * - d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110063 *
64 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110065 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110066 * dentry->d_lock
Dave Chinner19156842013-08-28 10:17:55 +100067 * dentry->d_sb->s_dentry_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110068 * dcache_hash_bucket lock
69 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110070 *
Nick Pigginda502952011-01-07 17:49:33 +110071 * If there is an ancestor relationship:
72 * dentry->d_parent->...->d_parent->d_lock
73 * ...
74 * dentry->d_parent->d_lock
75 * dentry->d_lock
76 *
77 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110078 * if (dentry1 < dentry2)
79 * dentry1->d_lock
80 * dentry2->d_lock
81 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080082int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
84
Al Viro74c3cbe2007-07-22 08:04:18 -040085__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Nick Piggin949854d2011-01-07 17:49:37 +110087EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Christoph Lametere18b8902006-12-06 20:33:20 -080089static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/*
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
95 *
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
98 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800100static unsigned int d_hash_mask __read_mostly;
101static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100102
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700103static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvalds8966be92012-03-02 14:23:30 -0800105static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700106 unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100107{
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700108 hash += (unsigned long) parent / L1_CACHE_BYTES;
Al Viro482db902013-10-25 16:41:01 -0400109 hash = hash + (hash >> d_hash_shift);
110 return dentry_hashtable + (hash & d_hash_mask);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100111}
112
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113/* Statistics gathering. */
114struct dentry_stat_t dentry_stat = {
115 .age_limit = 45,
116};
117
Glauber Costa3942c072013-08-28 10:17:53 +1000118static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000119static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400120
121#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000122
123/*
124 * Here we resort to our own counters instead of using generic per-cpu counters
125 * for consistency with what the vfs inode code does. We are expected to harvest
126 * better code and performance by having our own specialized counters.
127 *
128 * Please note that the loop is done over all possible CPUs, not over all online
129 * CPUs. The reason for this is that we don't want to play games with CPUs going
130 * on and off. If one of them goes off, we will just keep their counters.
131 *
132 * glommer: See cffbc8a for details, and if you ever intend to change this,
133 * please update all vfs counters to match.
134 */
Glauber Costa3942c072013-08-28 10:17:53 +1000135static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100136{
137 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000138 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100139 for_each_possible_cpu(i)
140 sum += per_cpu(nr_dentry, i);
141 return sum < 0 ? 0 : sum;
142}
143
Dave Chinner62d36c72013-08-28 10:17:54 +1000144static long get_nr_dentry_unused(void)
145{
146 int i;
147 long sum = 0;
148 for_each_possible_cpu(i)
149 sum += per_cpu(nr_dentry_unused, i);
150 return sum < 0 ? 0 : sum;
151}
152
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400153int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
154 size_t *lenp, loff_t *ppos)
155{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100156 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000157 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000158 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400159}
160#endif
161
Linus Torvalds5483f182012-03-04 15:51:42 -0800162/*
163 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
164 * The strings are both count bytes long, and count is non-zero.
165 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700166#ifdef CONFIG_DCACHE_WORD_ACCESS
167
168#include <asm/word-at-a-time.h>
169/*
170 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
171 * aligned allocation for this particular component. We don't
172 * strictly need the load_unaligned_zeropad() safety, but it
173 * doesn't hurt either.
174 *
175 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
176 * need the careful unaligned handling.
177 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700178static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800179{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800180 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800181
182 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700183 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700184 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800185 if (tcount < sizeof(unsigned long))
186 break;
187 if (unlikely(a != b))
188 return 1;
189 cs += sizeof(unsigned long);
190 ct += sizeof(unsigned long);
191 tcount -= sizeof(unsigned long);
192 if (!tcount)
193 return 0;
194 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000195 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800196 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700197}
198
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800199#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700200
Linus Torvalds94753db52012-05-10 12:19:19 -0700201static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700202{
Linus Torvalds5483f182012-03-04 15:51:42 -0800203 do {
204 if (*cs != *ct)
205 return 1;
206 cs++;
207 ct++;
208 tcount--;
209 } while (tcount);
210 return 0;
211}
212
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700213#endif
214
Linus Torvalds94753db52012-05-10 12:19:19 -0700215static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
216{
Linus Torvalds6326c712012-05-21 16:14:04 -0700217 const unsigned char *cs;
Linus Torvalds94753db52012-05-10 12:19:19 -0700218 /*
219 * Be careful about RCU walk racing with rename:
220 * use ACCESS_ONCE to fetch the name pointer.
221 *
222 * NOTE! Even if a rename will mean that the length
223 * was not loaded atomically, we don't care. The
224 * RCU walk will check the sequence count eventually,
225 * and catch it. And we won't overrun the buffer,
226 * because we're reading the name pointer atomically,
227 * and a dentry name is guaranteed to be properly
228 * terminated with a NUL byte.
229 *
230 * End result: even if 'len' is wrong, we'll exit
231 * early because the data cannot match (there can
232 * be no NUL in the ct/tcount data)
233 */
Linus Torvalds6326c712012-05-21 16:14:04 -0700234 cs = ACCESS_ONCE(dentry->d_name.name);
235 smp_read_barrier_depends();
236 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700237}
238
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400239static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400241 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
242
Al Virob3d9b7a2012-06-09 13:51:19 -0400243 WARN_ON(!hlist_unhashed(&dentry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 if (dname_external(dentry))
245 kfree(dentry->d_name.name);
246 kmem_cache_free(dentry_cache, dentry);
247}
248
Al Virob4f03542014-04-29 23:40:14 -0400249static void dentry_free(struct dentry *dentry)
250{
251 /* if dentry was never visible to RCU, immediate free is OK */
252 if (!(dentry->d_flags & DCACHE_RCUACCESS))
253 __d_free(&dentry->d_u.d_rcu);
254 else
255 call_rcu(&dentry->d_u.d_rcu, __d_free);
256}
257
Nick Piggin31e6b012011-01-07 17:49:52 +1100258/**
259 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800260 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100261 * After this call, in-progress rcu-walk path lookup will fail. This
262 * should be called after unhashing, and after changing d_inode (if
263 * the dentry has not already been unhashed).
264 */
265static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
266{
267 assert_spin_locked(&dentry->d_lock);
268 /* Go through a barrier */
269 write_seqcount_barrier(&dentry->d_seq);
270}
271
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272/*
273 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100274 * d_iput() operation if defined. Dentry has no refcount
275 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800277static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200278 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100279 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280{
281 struct inode *inode = dentry->d_inode;
282 if (inode) {
283 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400284 hlist_del_init(&dentry->d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100286 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700287 if (!inode->i_nlink)
288 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 if (dentry->d_op && dentry->d_op->d_iput)
290 dentry->d_op->d_iput(dentry, inode);
291 else
292 iput(inode);
293 } else {
294 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 }
296}
297
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700298/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100299 * Release the dentry's inode, using the filesystem
300 * d_iput() operation if defined. dentry remains in-use.
301 */
302static void dentry_unlink_inode(struct dentry * dentry)
303 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100304 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100305{
306 struct inode *inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +0100307 __d_clear_type(dentry);
Nick Piggin31e6b012011-01-07 17:49:52 +1100308 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400309 hlist_del_init(&dentry->d_alias);
Nick Piggin31e6b012011-01-07 17:49:52 +1100310 dentry_rcuwalk_barrier(dentry);
311 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100312 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100313 if (!inode->i_nlink)
314 fsnotify_inoderemove(inode);
315 if (dentry->d_op && dentry->d_op->d_iput)
316 dentry->d_op->d_iput(dentry, inode);
317 else
318 iput(inode);
319}
320
321/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400322 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
323 * is in use - which includes both the "real" per-superblock
324 * LRU list _and_ the DCACHE_SHRINK_LIST use.
325 *
326 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
327 * on the shrink list (ie not on the superblock LRU list).
328 *
329 * The per-cpu "nr_dentry_unused" counters are updated with
330 * the DCACHE_LRU_LIST bit.
331 *
332 * These helper functions make sure we always follow the
333 * rules. d_lock must be held by the caller.
334 */
335#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
336static void d_lru_add(struct dentry *dentry)
337{
338 D_FLAG_VERIFY(dentry, 0);
339 dentry->d_flags |= DCACHE_LRU_LIST;
340 this_cpu_inc(nr_dentry_unused);
341 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
342}
343
344static void d_lru_del(struct dentry *dentry)
345{
346 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
347 dentry->d_flags &= ~DCACHE_LRU_LIST;
348 this_cpu_dec(nr_dentry_unused);
349 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
350}
351
352static void d_shrink_del(struct dentry *dentry)
353{
354 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
355 list_del_init(&dentry->d_lru);
356 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
357 this_cpu_dec(nr_dentry_unused);
358}
359
360static void d_shrink_add(struct dentry *dentry, struct list_head *list)
361{
362 D_FLAG_VERIFY(dentry, 0);
363 list_add(&dentry->d_lru, list);
364 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
365 this_cpu_inc(nr_dentry_unused);
366}
367
368/*
369 * These can only be called under the global LRU lock, ie during the
370 * callback for freeing the LRU list. "isolate" removes it from the
371 * LRU lists entirely, while shrink_move moves it to the indicated
372 * private list.
373 */
374static void d_lru_isolate(struct dentry *dentry)
375{
376 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
377 dentry->d_flags &= ~DCACHE_LRU_LIST;
378 this_cpu_dec(nr_dentry_unused);
379 list_del_init(&dentry->d_lru);
380}
381
382static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
383{
384 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
385 dentry->d_flags |= DCACHE_SHRINK_LIST;
386 list_move_tail(&dentry->d_lru, list);
387}
388
389/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000390 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700391 */
392static void dentry_lru_add(struct dentry *dentry)
393{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400394 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
395 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700396}
397
Sage Weilf0023bc2011-10-28 10:02:42 -0700398/*
399 * Remove a dentry with references from the LRU.
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000400 *
401 * If we are on the shrink list, then we can get to try_prune_one_dentry() and
402 * lose our last reference through the parent walk. In this case, we need to
403 * remove ourselves from the shrink list, not the LRU.
Sage Weilf0023bc2011-10-28 10:02:42 -0700404 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700405static void dentry_lru_del(struct dentry *dentry)
406{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400407 if (dentry->d_flags & DCACHE_LRU_LIST) {
408 if (dentry->d_flags & DCACHE_SHRINK_LIST)
409 return d_shrink_del(dentry);
410 d_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700411 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700412}
413
Miklos Szeredid52b9082007-05-08 00:23:46 -0700414/**
Nick Piggin789680d2011-01-07 17:49:30 +1100415 * d_drop - drop a dentry
416 * @dentry: dentry to drop
417 *
418 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
419 * be found through a VFS lookup any more. Note that this is different from
420 * deleting the dentry - d_delete will try to mark the dentry negative if
421 * possible, giving a successful _negative_ lookup, while d_drop will
422 * just make the cache lookup fail.
423 *
424 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
425 * reason (NFS timeouts or autofs deletes).
426 *
427 * __d_drop requires dentry->d_lock.
428 */
429void __d_drop(struct dentry *dentry)
430{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700431 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400432 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400433 /*
434 * Hashed dentries are normally on the dentry hashtable,
435 * with the exception of those newly allocated by
436 * d_obtain_alias, which are always IS_ROOT:
437 */
438 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400439 b = &dentry->d_sb->s_anon;
440 else
441 b = d_hash(dentry->d_parent, dentry->d_name.hash);
442
443 hlist_bl_lock(b);
444 __hlist_bl_del(&dentry->d_hash);
445 dentry->d_hash.pprev = NULL;
446 hlist_bl_unlock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700447 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100448 }
449}
450EXPORT_SYMBOL(__d_drop);
451
452void d_drop(struct dentry *dentry)
453{
Nick Piggin789680d2011-01-07 17:49:30 +1100454 spin_lock(&dentry->d_lock);
455 __d_drop(dentry);
456 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100457}
458EXPORT_SYMBOL(d_drop);
459
Nick Piggin77812a12011-01-07 17:49:48 +1100460/*
461 * Finish off a dentry we've decided to kill.
462 * dentry->d_lock must be held, returns with it unlocked.
463 * If ref is non-zero, then decrement the refcount too.
464 * Returns dentry requiring refcount drop, or NULL if we're done.
465 */
Linus Torvalds358eec12013-10-31 15:43:02 -0700466static struct dentry *
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000467dentry_kill(struct dentry *dentry, int unlock_on_failure)
Nick Piggin77812a12011-01-07 17:49:48 +1100468 __releases(dentry->d_lock)
469{
Nick Piggin873feea2011-01-07 17:50:06 +1100470 struct inode *inode;
Al Viro41edf272014-05-01 10:30:00 -0400471 struct dentry *parent = NULL;
472 bool can_free = true;
473
474 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
475 can_free = dentry->d_flags & DCACHE_MAY_FREE;
476 spin_unlock(&dentry->d_lock);
477 goto out;
478 }
Nick Piggin77812a12011-01-07 17:49:48 +1100479
Nick Piggin873feea2011-01-07 17:50:06 +1100480 inode = dentry->d_inode;
481 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin77812a12011-01-07 17:49:48 +1100482relock:
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000483 if (unlock_on_failure) {
484 spin_unlock(&dentry->d_lock);
485 cpu_relax();
486 }
Nick Piggin77812a12011-01-07 17:49:48 +1100487 return dentry; /* try again with same dentry */
488 }
Al Viro41edf272014-05-01 10:30:00 -0400489 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100490 parent = dentry->d_parent;
491 if (parent && !spin_trylock(&parent->d_lock)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100492 if (inode)
493 spin_unlock(&inode->i_lock);
Nick Piggin77812a12011-01-07 17:49:48 +1100494 goto relock;
495 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100496
Linus Torvalds0d984392013-09-08 13:46:52 -0700497 /*
498 * The dentry is now unrecoverably dead to the world.
499 */
500 lockref_mark_dead(&dentry->d_lockref);
501
Sage Weilf0023bc2011-10-28 10:02:42 -0700502 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700503 * inform the fs via d_prune that this dentry is about to be
504 * unhashed and destroyed.
505 */
Yan, Zheng590fb512013-08-13 15:42:02 +0800506 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
Yan, Zheng61572bb2013-04-15 14:13:21 +0800507 dentry->d_op->d_prune(dentry);
508
Al Viro01b60352014-04-29 23:42:52 -0400509 if (dentry->d_flags & DCACHE_LRU_LIST) {
510 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
511 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400512 }
Nick Piggin77812a12011-01-07 17:49:48 +1100513 /* if it was on the hash then remove it */
514 __d_drop(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400515 list_del(&dentry->d_u.d_child);
516 /*
517 * Inform d_walk() that we are no longer attached to the
518 * dentry tree
519 */
520 dentry->d_flags |= DCACHE_DENTRY_KILLED;
521 if (parent)
522 spin_unlock(&parent->d_lock);
523 dentry_iput(dentry);
524 /*
525 * dentry_iput drops the locks, at which point nobody (except
526 * transient RCU lookups) can reach this dentry.
527 */
528 BUG_ON((int)dentry->d_lockref.count > 0);
529 this_cpu_dec(nr_dentry);
530 if (dentry->d_op && dentry->d_op->d_release)
531 dentry->d_op->d_release(dentry);
532
Al Viro41edf272014-05-01 10:30:00 -0400533 spin_lock(&dentry->d_lock);
534 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
535 dentry->d_flags |= DCACHE_MAY_FREE;
536 can_free = false;
537 }
538 spin_unlock(&dentry->d_lock);
539out:
540 if (likely(can_free))
541 dentry_free(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400542 return parent;
Nick Piggin77812a12011-01-07 17:49:48 +1100543}
544
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545/*
546 * This is dput
547 *
548 * This is complicated by the fact that we do not want to put
549 * dentries that are no longer on any hash chain on the unused
550 * list: we'd much rather just get rid of them immediately.
551 *
552 * However, that implies that we have to traverse the dentry
553 * tree upwards to the parents which might _also_ now be
554 * scheduled for deletion (it may have been only waiting for
555 * its last child to go away).
556 *
557 * This tail recursion is done by hand as we don't want to depend
558 * on the compiler to always get this right (gcc generally doesn't).
559 * Real recursion would eat up our stack space.
560 */
561
562/*
563 * dput - release a dentry
564 * @dentry: dentry to release
565 *
566 * Release a dentry. This will drop the usage count and if appropriate
567 * call the dentry unlink method as well as removing it from the queues and
568 * releasing its resources. If the parent dentries were scheduled for release
569 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571void dput(struct dentry *dentry)
572{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700573 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 return;
575
576repeat:
Waiman Long98474232013-08-28 18:24:59 -0700577 if (lockref_put_or_lock(&dentry->d_lockref))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700580 /* Unreachable? Get rid of it */
581 if (unlikely(d_unhashed(dentry)))
582 goto kill_it;
583
584 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100586 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 }
Nick Piggin265ac902010-10-10 05:36:24 -0400588
Linus Torvalds358eec12013-10-31 15:43:02 -0700589 if (!(dentry->d_flags & DCACHE_REFERENCED))
590 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400591 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400592
Waiman Long98474232013-08-28 18:24:59 -0700593 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100594 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 return;
596
Miklos Szeredid52b9082007-05-08 00:23:46 -0700597kill_it:
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000598 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700599 if (dentry)
600 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700602EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603
604/**
605 * d_invalidate - invalidate a dentry
606 * @dentry: dentry to invalidate
607 *
608 * Try to invalidate the dentry if it turns out to be
609 * possible. If there are other dentries that can be
610 * reached through this one we can't delete it and we
611 * return -EBUSY. On success we return 0.
612 *
613 * no dcache lock.
614 */
615
616int d_invalidate(struct dentry * dentry)
617{
618 /*
619 * If it's already been dropped, return OK.
620 */
Nick Pigginda502952011-01-07 17:49:33 +1100621 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100623 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 return 0;
625 }
626 /*
627 * Check whether to do a partial shrink_dcache
628 * to get rid of unused child entries.
629 */
630 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100631 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100633 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 }
635
636 /*
637 * Somebody else still using it?
638 *
639 * If it's a directory, we can't drop it
640 * for fear of somebody re-populating it
641 * with children (even though dropping it
642 * would make it unreachable from the root,
643 * we might still populate it if it was a
644 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000645 * We also need to leave mountpoints alone,
646 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 */
Waiman Long98474232013-08-28 18:24:59 -0700648 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
Al Viro50e69632011-11-07 16:39:57 +0000649 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 return -EBUSY;
652 }
653 }
654
655 __d_drop(dentry);
656 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 return 0;
658}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700659EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100661/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100662static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663{
Waiman Long98474232013-08-28 18:24:59 -0700664 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
666
Nick Piggindc0474b2011-01-07 17:49:43 +1100667static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100668{
Waiman Long98474232013-08-28 18:24:59 -0700669 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100670}
671
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100672struct dentry *dget_parent(struct dentry *dentry)
673{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700674 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100675 struct dentry *ret;
676
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700677 /*
678 * Do optimistic parent lookup without any
679 * locking.
680 */
681 rcu_read_lock();
682 ret = ACCESS_ONCE(dentry->d_parent);
683 gotref = lockref_get_not_zero(&ret->d_lockref);
684 rcu_read_unlock();
685 if (likely(gotref)) {
686 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
687 return ret;
688 dput(ret);
689 }
690
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100691repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100692 /*
693 * Don't need rcu_dereference because we re-check it was correct under
694 * the lock.
695 */
696 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100697 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100698 spin_lock(&ret->d_lock);
699 if (unlikely(ret != dentry->d_parent)) {
700 spin_unlock(&ret->d_lock);
701 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100702 goto repeat;
703 }
Nick Piggina734eb42011-01-07 17:49:44 +1100704 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700705 BUG_ON(!ret->d_lockref.count);
706 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100707 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100708 return ret;
709}
710EXPORT_SYMBOL(dget_parent);
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712/**
713 * d_find_alias - grab a hashed alias of inode
714 * @inode: inode in question
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700715 * @want_discon: flag, used by d_splice_alias, to request
716 * that only a DISCONNECTED alias be returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 *
718 * If inode has a hashed alias, or is a directory and has any alias,
719 * acquire the reference to alias and return it. Otherwise return NULL.
720 * Notice that if inode is a directory there can be only one alias and
721 * it can be unhashed only if it has no children, or if it is the root
722 * of a filesystem.
723 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700724 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700725 * any other hashed alias over that one unless @want_discon is set,
726 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 */
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700728static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729{
Nick Pigginda502952011-01-07 17:49:33 +1100730 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
Nick Pigginda502952011-01-07 17:49:33 +1100732again:
733 discon_alias = NULL;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800734 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100735 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700737 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100738 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 discon_alias = alias;
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700740 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100741 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100742 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 return alias;
744 }
745 }
Nick Pigginda502952011-01-07 17:49:33 +1100746 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 }
Nick Pigginda502952011-01-07 17:49:33 +1100748 if (discon_alias) {
749 alias = discon_alias;
750 spin_lock(&alias->d_lock);
751 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
752 if (IS_ROOT(alias) &&
753 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100754 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100755 spin_unlock(&alias->d_lock);
756 return alias;
757 }
758 }
759 spin_unlock(&alias->d_lock);
760 goto again;
761 }
762 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763}
764
Nick Pigginda502952011-01-07 17:49:33 +1100765struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766{
David Howells214fda12006-03-25 03:06:36 -0800767 struct dentry *de = NULL;
768
Al Virob3d9b7a2012-06-09 13:51:19 -0400769 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100770 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700771 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100772 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800773 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 return de;
775}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700776EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778/*
779 * Try to kill dentries associated with this inode.
780 * WARNING: you must own a reference to inode.
781 */
782void d_prune_aliases(struct inode *inode)
783{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700784 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100786 spin_lock(&inode->i_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800787 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700789 if (!dentry->d_lockref.count) {
Yan, Zheng590fb512013-08-13 15:42:02 +0800790 /*
791 * inform the fs via d_prune that this dentry
792 * is about to be unhashed and destroyed.
793 */
794 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
795 !d_unhashed(dentry))
796 dentry->d_op->d_prune(dentry);
797
Nick Piggindc0474b2011-01-07 17:49:43 +1100798 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 __d_drop(dentry);
800 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100801 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 dput(dentry);
803 goto restart;
804 }
805 spin_unlock(&dentry->d_lock);
806 }
Nick Piggin873feea2011-01-07 17:50:06 +1100807 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700809EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400811static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812{
Al Viro5c47e6d2014-04-29 16:13:18 -0400813 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700814
Nick Pigginec336792011-01-07 17:49:47 +1100815 rcu_read_lock();
816 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100817 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
818 if (&dentry->d_lru == list)
819 break; /* empty */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400820
821 /*
822 * Get the dentry lock, and re-verify that the dentry is
823 * this on the shrinking list. If it is, we know that
824 * DCACHE_SHRINK_LIST and DCACHE_LRU_LIST are set.
825 */
Nick Pigginec336792011-01-07 17:49:47 +1100826 spin_lock(&dentry->d_lock);
827 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
828 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100829 continue;
830 }
831
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000833 * The dispose list is isolated and dentries are not accounted
834 * to the LRU here, so we can simply remove it from the list
835 * here regardless of whether it is referenced or not.
836 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400837 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000838
839 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000841 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 */
Al Viro41edf272014-05-01 10:30:00 -0400843 if ((int)dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700844 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 continue;
846 }
Nick Pigginec336792011-01-07 17:49:47 +1100847 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100848
Al Viro5c47e6d2014-04-29 16:13:18 -0400849 parent = dentry_kill(dentry, 0);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400850 /*
Al Viro5c47e6d2014-04-29 16:13:18 -0400851 * If dentry_kill returns NULL, we have nothing more to do.
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400852 */
Al Viro5c47e6d2014-04-29 16:13:18 -0400853 if (!parent) {
854 rcu_read_lock();
855 continue;
856 }
857 if (unlikely(parent == dentry)) {
858 /*
859 * trylocks have failed and d_lock has been held the
860 * whole time, so it could not have been added to any
861 * other lists. Just add it back to the shrink list.
862 */
863 rcu_read_lock();
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400864 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000865 spin_unlock(&dentry->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400866 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000867 }
Al Viro5c47e6d2014-04-29 16:13:18 -0400868 /*
869 * We need to prune ancestors too. This is necessary to prevent
870 * quadratic behavior of shrink_dcache_parent(), but is also
871 * expected to be beneficial in reducing dentry cache
872 * fragmentation.
873 */
874 dentry = parent;
875 while (dentry && !lockref_put_or_lock(&dentry->d_lockref))
876 dentry = dentry_kill(dentry, 1);
877 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 }
Nick Pigginec336792011-01-07 17:49:47 +1100879 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400880}
881
Dave Chinnerf6041562013-08-28 10:18:00 +1000882static enum lru_status
883dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
884{
885 struct list_head *freeable = arg;
886 struct dentry *dentry = container_of(item, struct dentry, d_lru);
887
888
889 /*
890 * we are inverting the lru lock/dentry->d_lock here,
891 * so use a trylock. If we fail to get the lock, just skip
892 * it
893 */
894 if (!spin_trylock(&dentry->d_lock))
895 return LRU_SKIP;
896
897 /*
898 * Referenced dentries are still in use. If they have active
899 * counts, just remove them from the LRU. Otherwise give them
900 * another pass through the LRU.
901 */
902 if (dentry->d_lockref.count) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400903 d_lru_isolate(dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +1000904 spin_unlock(&dentry->d_lock);
905 return LRU_REMOVED;
906 }
907
908 if (dentry->d_flags & DCACHE_REFERENCED) {
909 dentry->d_flags &= ~DCACHE_REFERENCED;
910 spin_unlock(&dentry->d_lock);
911
912 /*
913 * The list move itself will be made by the common LRU code. At
914 * this point, we've dropped the dentry->d_lock but keep the
915 * lru lock. This is safe to do, since every list movement is
916 * protected by the lru lock even if both locks are held.
917 *
918 * This is guaranteed by the fact that all LRU management
919 * functions are intermediated by the LRU API calls like
920 * list_lru_add and list_lru_del. List movement in this file
921 * only ever occur through this functions or through callbacks
922 * like this one, that are called from the LRU API.
923 *
924 * The only exceptions to this are functions like
925 * shrink_dentry_list, and code that first checks for the
926 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
927 * operating only with stack provided lists after they are
928 * properly isolated from the main list. It is thus, always a
929 * local access.
930 */
931 return LRU_ROTATE;
932 }
933
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400934 d_lru_shrink_move(dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +1000935 spin_unlock(&dentry->d_lock);
936
937 return LRU_REMOVED;
938}
939
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400940/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000941 * prune_dcache_sb - shrink the dcache
942 * @sb: superblock
Dave Chinnerf6041562013-08-28 10:18:00 +1000943 * @nr_to_scan : number of entries to try to free
Dave Chinner9b17c622013-08-28 10:18:05 +1000944 * @nid: which node to scan for freeable entities
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400945 *
Dave Chinnerf6041562013-08-28 10:18:00 +1000946 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000947 * done when we need more memory an called from the superblock shrinker
948 * function.
949 *
950 * This function may fail to free any resources if all the dentries are in
951 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400952 */
Dave Chinner9b17c622013-08-28 10:18:05 +1000953long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
954 int nid)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400955{
Dave Chinnerf6041562013-08-28 10:18:00 +1000956 LIST_HEAD(dispose);
957 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400958
Dave Chinner9b17c622013-08-28 10:18:05 +1000959 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
960 &dispose, &nr_to_scan);
Dave Chinnerf6041562013-08-28 10:18:00 +1000961 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +1000962 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963}
964
Glauber Costa4e717f52013-08-28 10:18:03 +1000965static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
966 spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000967{
Glauber Costa4e717f52013-08-28 10:18:03 +1000968 struct list_head *freeable = arg;
969 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000970
Glauber Costa4e717f52013-08-28 10:18:03 +1000971 /*
972 * we are inverting the lru lock/dentry->d_lock here,
973 * so use a trylock. If we fail to get the lock, just skip
974 * it
975 */
976 if (!spin_trylock(&dentry->d_lock))
977 return LRU_SKIP;
978
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400979 d_lru_shrink_move(dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +1000980 spin_unlock(&dentry->d_lock);
981
982 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000983}
984
Glauber Costa4e717f52013-08-28 10:18:03 +1000985
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700986/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 * shrink_dcache_sb - shrink dcache for a superblock
988 * @sb: superblock
989 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400990 * Shrink the dcache for the specified super block. This is used to free
991 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400993void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
Glauber Costa4e717f52013-08-28 10:18:03 +1000995 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400996
Glauber Costa4e717f52013-08-28 10:18:03 +1000997 do {
998 LIST_HEAD(dispose);
999
1000 freed = list_lru_walk(&sb->s_dentry_lru,
1001 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1002
1003 this_cpu_sub(nr_dentry_unused, freed);
1004 shrink_dentry_list(&dispose);
1005 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001007EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001010 * enum d_walk_ret - action to talke during tree walk
1011 * @D_WALK_CONTINUE: contrinue walk
1012 * @D_WALK_QUIT: quit walk
1013 * @D_WALK_NORETRY: quit when retry is needed
1014 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001016enum d_walk_ret {
1017 D_WALK_CONTINUE,
1018 D_WALK_QUIT,
1019 D_WALK_NORETRY,
1020 D_WALK_SKIP,
1021};
1022
1023/**
1024 * d_walk - walk the dentry tree
1025 * @parent: start of walk
1026 * @data: data passed to @enter() and @finish()
1027 * @enter: callback when first entering the dentry
1028 * @finish: callback when successfully finished the walk
1029 *
1030 * The @enter() and @finish() callbacks are called with d_lock held.
1031 */
1032static void d_walk(struct dentry *parent, void *data,
1033 enum d_walk_ret (*enter)(void *, struct dentry *),
1034 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035{
Nick Piggin949854d2011-01-07 17:49:37 +11001036 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001038 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001039 enum d_walk_ret ret;
1040 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001041
Nick Piggin58db63d2011-01-07 17:49:39 +11001042again:
Al Viro48f5ec22013-09-09 15:22:25 -04001043 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001044 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001045 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001046
1047 ret = enter(data, this_parent);
1048 switch (ret) {
1049 case D_WALK_CONTINUE:
1050 break;
1051 case D_WALK_QUIT:
1052 case D_WALK_SKIP:
1053 goto out_unlock;
1054 case D_WALK_NORETRY:
1055 retry = false;
1056 break;
1057 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058repeat:
1059 next = this_parent->d_subdirs.next;
1060resume:
1061 while (next != &this_parent->d_subdirs) {
1062 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001063 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001065
1066 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001067
1068 ret = enter(data, dentry);
1069 switch (ret) {
1070 case D_WALK_CONTINUE:
1071 break;
1072 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001073 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001074 goto out_unlock;
1075 case D_WALK_NORETRY:
1076 retry = false;
1077 break;
1078 case D_WALK_SKIP:
1079 spin_unlock(&dentry->d_lock);
1080 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001081 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001082
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001084 spin_unlock(&this_parent->d_lock);
1085 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001087 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 goto repeat;
1089 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001090 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 }
1092 /*
1093 * All done at this level ... ascend and resume the search.
1094 */
1095 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001096 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001097 this_parent = child->d_parent;
1098
1099 rcu_read_lock();
1100 spin_unlock(&child->d_lock);
1101 spin_lock(&this_parent->d_lock);
1102
1103 /*
1104 * might go back up the wrong parent if we have had a rename
1105 * or deletion
1106 */
1107 if (this_parent != child->d_parent ||
1108 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1109 need_seqretry(&rename_lock, seq)) {
1110 spin_unlock(&this_parent->d_lock);
1111 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11001112 goto rename_retry;
Al Viro31dec132013-10-25 17:04:27 -04001113 }
1114 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11001115 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 goto resume;
1117 }
Al Viro48f5ec22013-09-09 15:22:25 -04001118 if (need_seqretry(&rename_lock, seq)) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001119 spin_unlock(&this_parent->d_lock);
1120 goto rename_retry;
1121 }
1122 if (finish)
1123 finish(data);
1124
1125out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001126 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001127 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001128 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001129
1130rename_retry:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001131 if (!retry)
1132 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001133 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001134 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001136
1137/*
1138 * Search for at least 1 mount point in the dentry's subdirs.
1139 * We descend to the next level whenever the d_subdirs
1140 * list is non-empty and continue searching.
1141 */
1142
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001143static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1144{
1145 int *ret = data;
1146 if (d_mountpoint(dentry)) {
1147 *ret = 1;
1148 return D_WALK_QUIT;
1149 }
1150 return D_WALK_CONTINUE;
1151}
1152
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001153/**
1154 * have_submounts - check for mounts over a dentry
1155 * @parent: dentry to check.
1156 *
1157 * Return true if the parent or its subdirectories contain
1158 * a mount point
1159 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001160int have_submounts(struct dentry *parent)
1161{
1162 int ret = 0;
1163
1164 d_walk(parent, &ret, check_mount, NULL);
1165
1166 return ret;
1167}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001168EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
1170/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001171 * Called by mount code to set a mountpoint and check if the mountpoint is
1172 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1173 * subtree can become unreachable).
1174 *
1175 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1176 * this reason take rename_lock and d_lock on dentry and ancestors.
1177 */
1178int d_set_mounted(struct dentry *dentry)
1179{
1180 struct dentry *p;
1181 int ret = -ENOENT;
1182 write_seqlock(&rename_lock);
1183 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1184 /* Need exclusion wrt. check_submounts_and_drop() */
1185 spin_lock(&p->d_lock);
1186 if (unlikely(d_unhashed(p))) {
1187 spin_unlock(&p->d_lock);
1188 goto out;
1189 }
1190 spin_unlock(&p->d_lock);
1191 }
1192 spin_lock(&dentry->d_lock);
1193 if (!d_unlinked(dentry)) {
1194 dentry->d_flags |= DCACHE_MOUNTED;
1195 ret = 0;
1196 }
1197 spin_unlock(&dentry->d_lock);
1198out:
1199 write_sequnlock(&rename_lock);
1200 return ret;
1201}
1202
1203/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001204 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 * and move any unused dentries to the end of the unused
1206 * list for prune_dcache(). We descend to the next level
1207 * whenever the d_subdirs list is non-empty and continue
1208 * searching.
1209 *
1210 * It returns zero iff there are no unused children,
1211 * otherwise it returns the number of children moved to
1212 * the end of the unused list. This may not be the total
1213 * number of unused children, because select_parent can
1214 * drop the lock and return early due to latency
1215 * constraints.
1216 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001217
1218struct select_data {
1219 struct dentry *start;
1220 struct list_head dispose;
1221 int found;
1222};
1223
1224static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001226 struct select_data *data = _data;
1227 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001229 if (data->start == dentry)
1230 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001232 /*
1233 * move only zero ref count dentries to the dispose list.
1234 *
1235 * Those which are presently on the shrink list, being processed
1236 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1237 * loop in shrink_dcache_parent() might not make any progress
1238 * and loop forever.
1239 */
1240 if (dentry->d_lockref.count) {
1241 dentry_lru_del(dentry);
1242 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001243 /*
1244 * We can't use d_lru_shrink_move() because we
1245 * need to get the global LRU lock and do the
Linus Torvalds05a82522013-09-15 07:11:01 -04001246 * LRU accounting.
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001247 */
1248 d_lru_del(dentry);
1249 d_shrink_add(dentry, &data->dispose);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001250 data->found++;
1251 ret = D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 }
1253 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001254 * We can return to the caller if we have found some (this
1255 * ensures forward progress). We'll be coming back to find
1256 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001258 if (data->found && need_resched())
1259 ret = D_WALK_QUIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001261 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262}
1263
1264/**
1265 * shrink_dcache_parent - prune dcache
1266 * @parent: parent of entries to prune
1267 *
1268 * Prune the dcache to remove unused children of the parent dentry.
1269 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001270void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001272 for (;;) {
1273 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001275 INIT_LIST_HEAD(&data.dispose);
1276 data.start = parent;
1277 data.found = 0;
1278
1279 d_walk(parent, &data, select_collect, NULL);
1280 if (!data.found)
1281 break;
1282
1283 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001284 cond_resched();
1285 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001287EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288
Al Viro42c32602013-11-08 12:31:16 -05001289static enum d_walk_ret umount_collect(void *_data, struct dentry *dentry)
1290{
1291 struct select_data *data = _data;
1292 enum d_walk_ret ret = D_WALK_CONTINUE;
1293
1294 if (dentry->d_lockref.count) {
1295 dentry_lru_del(dentry);
1296 if (likely(!list_empty(&dentry->d_subdirs)))
1297 goto out;
1298 if (dentry == data->start && dentry->d_lockref.count == 1)
1299 goto out;
1300 printk(KERN_ERR
1301 "BUG: Dentry %p{i=%lx,n=%s}"
1302 " still in use (%d)"
1303 " [unmount of %s %s]\n",
1304 dentry,
1305 dentry->d_inode ?
1306 dentry->d_inode->i_ino : 0UL,
1307 dentry->d_name.name,
1308 dentry->d_lockref.count,
1309 dentry->d_sb->s_type->name,
1310 dentry->d_sb->s_id);
1311 BUG();
1312 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1313 /*
1314 * We can't use d_lru_shrink_move() because we
1315 * need to get the global LRU lock and do the
1316 * LRU accounting.
1317 */
1318 if (dentry->d_flags & DCACHE_LRU_LIST)
1319 d_lru_del(dentry);
1320 d_shrink_add(dentry, &data->dispose);
1321 data->found++;
1322 ret = D_WALK_NORETRY;
1323 }
1324out:
1325 if (data->found && need_resched())
1326 ret = D_WALK_QUIT;
1327 return ret;
1328}
1329
1330/*
1331 * destroy the dentries attached to a superblock on unmounting
1332 */
1333void shrink_dcache_for_umount(struct super_block *sb)
1334{
1335 struct dentry *dentry;
1336
1337 if (down_read_trylock(&sb->s_umount))
1338 BUG();
1339
1340 dentry = sb->s_root;
1341 sb->s_root = NULL;
1342 for (;;) {
1343 struct select_data data;
1344
1345 INIT_LIST_HEAD(&data.dispose);
1346 data.start = dentry;
1347 data.found = 0;
1348
1349 d_walk(dentry, &data, umount_collect, NULL);
1350 if (!data.found)
1351 break;
1352
1353 shrink_dentry_list(&data.dispose);
1354 cond_resched();
1355 }
1356 d_drop(dentry);
1357 dput(dentry);
1358
1359 while (!hlist_bl_empty(&sb->s_anon)) {
1360 struct select_data data;
1361 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1362
1363 INIT_LIST_HEAD(&data.dispose);
1364 data.start = NULL;
1365 data.found = 0;
1366
1367 d_walk(dentry, &data, umount_collect, NULL);
1368 if (data.found)
1369 shrink_dentry_list(&data.dispose);
1370 cond_resched();
1371 }
1372}
1373
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001374static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1375{
1376 struct select_data *data = _data;
1377
1378 if (d_mountpoint(dentry)) {
1379 data->found = -EBUSY;
1380 return D_WALK_QUIT;
1381 }
1382
1383 return select_collect(_data, dentry);
1384}
1385
1386static void check_and_drop(void *_data)
1387{
1388 struct select_data *data = _data;
1389
1390 if (d_mountpoint(data->start))
1391 data->found = -EBUSY;
1392 if (!data->found)
1393 __d_drop(data->start);
1394}
1395
1396/**
1397 * check_submounts_and_drop - prune dcache, check for submounts and drop
1398 *
1399 * All done as a single atomic operation relative to has_unlinked_ancestor().
1400 * Returns 0 if successfully unhashed @parent. If there were submounts then
1401 * return -EBUSY.
1402 *
1403 * @dentry: dentry to prune and drop
1404 */
1405int check_submounts_and_drop(struct dentry *dentry)
1406{
1407 int ret = 0;
1408
1409 /* Negative dentries can be dropped without further checks */
1410 if (!dentry->d_inode) {
1411 d_drop(dentry);
1412 goto out;
1413 }
1414
1415 for (;;) {
1416 struct select_data data;
1417
1418 INIT_LIST_HEAD(&data.dispose);
1419 data.start = dentry;
1420 data.found = 0;
1421
1422 d_walk(dentry, &data, check_and_collect, check_and_drop);
1423 ret = data.found;
1424
1425 if (!list_empty(&data.dispose))
1426 shrink_dentry_list(&data.dispose);
1427
1428 if (ret <= 0)
1429 break;
1430
1431 cond_resched();
1432 }
1433
1434out:
1435 return ret;
1436}
1437EXPORT_SYMBOL(check_submounts_and_drop);
1438
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439/**
Al Viroa4464db2011-07-07 15:03:58 -04001440 * __d_alloc - allocate a dcache entry
1441 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 * @name: qstr of the name
1443 *
1444 * Allocates a dentry. It returns %NULL if there is insufficient memory
1445 * available. On a success the dentry is returned. The name passed in is
1446 * copied and the copy passed in may be reused after this call.
1447 */
1448
Al Viroa4464db2011-07-07 15:03:58 -04001449struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450{
1451 struct dentry *dentry;
1452 char *dname;
1453
Mel Gormane12ba742007-10-16 01:25:52 -07001454 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 if (!dentry)
1456 return NULL;
1457
Linus Torvalds6326c712012-05-21 16:14:04 -07001458 /*
1459 * We guarantee that the inline name is always NUL-terminated.
1460 * This way the memcpy() done by the name switching in rename
1461 * will still always have a NUL at the end, even if we might
1462 * be overwriting an internal NUL character
1463 */
1464 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 if (name->len > DNAME_INLINE_LEN-1) {
1466 dname = kmalloc(name->len + 1, GFP_KERNEL);
1467 if (!dname) {
1468 kmem_cache_free(dentry_cache, dentry);
1469 return NULL;
1470 }
1471 } else {
1472 dname = dentry->d_iname;
1473 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
1475 dentry->d_name.len = name->len;
1476 dentry->d_name.hash = name->hash;
1477 memcpy(dname, name->name, name->len);
1478 dname[name->len] = 0;
1479
Linus Torvalds6326c712012-05-21 16:14:04 -07001480 /* Make sure we always see the terminating NUL character */
1481 smp_wmb();
1482 dentry->d_name.name = dname;
1483
Waiman Long98474232013-08-28 18:24:59 -07001484 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001485 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001487 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001489 dentry->d_parent = dentry;
1490 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 dentry->d_op = NULL;
1492 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001493 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 INIT_LIST_HEAD(&dentry->d_lru);
1495 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001496 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001497 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001498 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
Nick Piggin3e880fb2011-01-07 17:49:19 +11001500 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 return dentry;
1503}
Al Viroa4464db2011-07-07 15:03:58 -04001504
1505/**
1506 * d_alloc - allocate a dcache entry
1507 * @parent: parent of entry to allocate
1508 * @name: qstr of the name
1509 *
1510 * Allocates a dentry. It returns %NULL if there is insufficient memory
1511 * available. On a success the dentry is returned. The name passed in is
1512 * copied and the copy passed in may be reused after this call.
1513 */
1514struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1515{
1516 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1517 if (!dentry)
1518 return NULL;
1519
1520 spin_lock(&parent->d_lock);
1521 /*
1522 * don't need child lock because it is not subject
1523 * to concurrency here
1524 */
1525 __dget_dlock(parent);
1526 dentry->d_parent = parent;
1527 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1528 spin_unlock(&parent->d_lock);
1529
1530 return dentry;
1531}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001532EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001534/**
1535 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1536 * @sb: the superblock
1537 * @name: qstr of the name
1538 *
1539 * For a filesystem that just pins its dentries in memory and never
1540 * performs lookups at all, return an unhashed IS_ROOT dentry.
1541 */
Nick Piggin4b936882011-01-07 17:50:07 +11001542struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1543{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001544 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001545}
1546EXPORT_SYMBOL(d_alloc_pseudo);
1547
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1549{
1550 struct qstr q;
1551
1552 q.name = name;
1553 q.len = strlen(name);
1554 q.hash = full_name_hash(q.name, q.len);
1555 return d_alloc(parent, &q);
1556}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001557EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558
Nick Pigginfb045ad2011-01-07 17:49:55 +11001559void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1560{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001561 WARN_ON_ONCE(dentry->d_op);
1562 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001563 DCACHE_OP_COMPARE |
1564 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001565 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001566 DCACHE_OP_DELETE ));
1567 dentry->d_op = op;
1568 if (!op)
1569 return;
1570 if (op->d_hash)
1571 dentry->d_flags |= DCACHE_OP_HASH;
1572 if (op->d_compare)
1573 dentry->d_flags |= DCACHE_OP_COMPARE;
1574 if (op->d_revalidate)
1575 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001576 if (op->d_weak_revalidate)
1577 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001578 if (op->d_delete)
1579 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001580 if (op->d_prune)
1581 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001582
1583}
1584EXPORT_SYMBOL(d_set_d_op);
1585
David Howellsb18825a2013-09-12 19:22:53 +01001586static unsigned d_flags_for_inode(struct inode *inode)
1587{
1588 unsigned add_flags = DCACHE_FILE_TYPE;
1589
1590 if (!inode)
1591 return DCACHE_MISS_TYPE;
1592
1593 if (S_ISDIR(inode->i_mode)) {
1594 add_flags = DCACHE_DIRECTORY_TYPE;
1595 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1596 if (unlikely(!inode->i_op->lookup))
1597 add_flags = DCACHE_AUTODIR_TYPE;
1598 else
1599 inode->i_opflags |= IOP_LOOKUP;
1600 }
1601 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1602 if (unlikely(inode->i_op->follow_link))
1603 add_flags = DCACHE_SYMLINK_TYPE;
1604 else
1605 inode->i_opflags |= IOP_NOFOLLOW;
1606 }
1607
1608 if (unlikely(IS_AUTOMOUNT(inode)))
1609 add_flags |= DCACHE_NEED_AUTOMOUNT;
1610 return add_flags;
1611}
1612
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001613static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1614{
David Howellsb18825a2013-09-12 19:22:53 +01001615 unsigned add_flags = d_flags_for_inode(inode);
1616
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001617 spin_lock(&dentry->d_lock);
Al Viro22213312014-04-19 12:30:58 -04001618 __d_set_type(dentry, add_flags);
David Howellsb18825a2013-09-12 19:22:53 +01001619 if (inode)
Al Virob3d9b7a2012-06-09 13:51:19 -04001620 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001621 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001622 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001623 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001624 fsnotify_d_instantiate(dentry, inode);
1625}
1626
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627/**
1628 * d_instantiate - fill in inode information for a dentry
1629 * @entry: dentry to complete
1630 * @inode: inode to attach to this dentry
1631 *
1632 * Fill in inode information in the entry.
1633 *
1634 * This turns negative dentries into productive full members
1635 * of society.
1636 *
1637 * NOTE! This assumes that the inode count has been incremented
1638 * (or otherwise set) by the caller to indicate that it is now
1639 * in use by the dcache.
1640 */
1641
1642void d_instantiate(struct dentry *entry, struct inode * inode)
1643{
Al Virob3d9b7a2012-06-09 13:51:19 -04001644 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001645 if (inode)
1646 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001647 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001648 if (inode)
1649 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 security_d_instantiate(entry, inode);
1651}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001652EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
1654/**
1655 * d_instantiate_unique - instantiate a non-aliased dentry
1656 * @entry: dentry to instantiate
1657 * @inode: inode to attach to this dentry
1658 *
1659 * Fill in inode information in the entry. On success, it returns NULL.
1660 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001661 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 *
1663 * Note that in order to avoid conflicts with rename() etc, the caller
1664 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001665 *
1666 * This also assumes that the inode count has been incremented
1667 * (or otherwise set) by the caller to indicate that it is now
1668 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 */
David Howells770bfad2006-08-22 20:06:07 -04001670static struct dentry *__d_instantiate_unique(struct dentry *entry,
1671 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672{
1673 struct dentry *alias;
1674 int len = entry->d_name.len;
1675 const char *name = entry->d_name.name;
1676 unsigned int hash = entry->d_name.hash;
1677
David Howells770bfad2006-08-22 20:06:07 -04001678 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001679 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001680 return NULL;
1681 }
1682
Sasha Levinb67bfe02013-02-27 17:06:00 -08001683 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001684 /*
1685 * Don't need alias->d_lock here, because aliases with
1686 * d_parent == entry->d_parent are not subject to name or
1687 * parent changes, because the parent inode i_mutex is held.
1688 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001689 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 continue;
1691 if (alias->d_parent != entry->d_parent)
1692 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001693 if (alias->d_name.len != len)
1694 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001695 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001697 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 return alias;
1699 }
David Howells770bfad2006-08-22 20:06:07 -04001700
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001701 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 return NULL;
1703}
David Howells770bfad2006-08-22 20:06:07 -04001704
1705struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1706{
1707 struct dentry *result;
1708
Al Virob3d9b7a2012-06-09 13:51:19 -04001709 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001710
Nick Piggin873feea2011-01-07 17:50:06 +11001711 if (inode)
1712 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001713 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001714 if (inode)
1715 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001716
1717 if (!result) {
1718 security_d_instantiate(entry, inode);
1719 return NULL;
1720 }
1721
1722 BUG_ON(!d_unhashed(result));
1723 iput(inode);
1724 return result;
1725}
1726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727EXPORT_SYMBOL(d_instantiate_unique);
1728
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001729/**
1730 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1731 * @entry: dentry to complete
1732 * @inode: inode to attach to this dentry
1733 *
1734 * Fill in inode information in the entry. If a directory alias is found, then
1735 * return an error (and drop inode). Together with d_materialise_unique() this
1736 * guarantees that a directory inode may never have more than one alias.
1737 */
1738int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1739{
1740 BUG_ON(!hlist_unhashed(&entry->d_alias));
1741
1742 spin_lock(&inode->i_lock);
1743 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1744 spin_unlock(&inode->i_lock);
1745 iput(inode);
1746 return -EBUSY;
1747 }
1748 __d_instantiate(entry, inode);
1749 spin_unlock(&inode->i_lock);
1750 security_d_instantiate(entry, inode);
1751
1752 return 0;
1753}
1754EXPORT_SYMBOL(d_instantiate_no_diralias);
1755
Al Viroadc0e912012-01-08 16:49:21 -05001756struct dentry *d_make_root(struct inode *root_inode)
1757{
1758 struct dentry *res = NULL;
1759
1760 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001761 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001762
1763 res = __d_alloc(root_inode->i_sb, &name);
1764 if (res)
1765 d_instantiate(res, root_inode);
1766 else
1767 iput(root_inode);
1768 }
1769 return res;
1770}
1771EXPORT_SYMBOL(d_make_root);
1772
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001773static struct dentry * __d_find_any_alias(struct inode *inode)
1774{
1775 struct dentry *alias;
1776
Al Virob3d9b7a2012-06-09 13:51:19 -04001777 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001778 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001779 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001780 __dget(alias);
1781 return alias;
1782}
1783
Sage Weil46f72b32012-01-10 09:04:37 -08001784/**
1785 * d_find_any_alias - find any alias for a given inode
1786 * @inode: inode to find an alias for
1787 *
1788 * If any aliases exist for the given inode, take and return a
1789 * reference for one of them. If no aliases exist, return %NULL.
1790 */
1791struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001792{
1793 struct dentry *de;
1794
1795 spin_lock(&inode->i_lock);
1796 de = __d_find_any_alias(inode);
1797 spin_unlock(&inode->i_lock);
1798 return de;
1799}
Sage Weil46f72b32012-01-10 09:04:37 -08001800EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001801
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001803 * d_obtain_alias - find or allocate a dentry for a given inode
1804 * @inode: inode to allocate the dentry for
1805 *
1806 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1807 * similar open by handle operations. The returned dentry may be anonymous,
1808 * or may have a full name (if the inode was already in the cache).
1809 *
1810 * When called on a directory inode, we must ensure that the inode only ever
1811 * has one dentry. If a dentry is found, that is returned instead of
1812 * allocating a new one.
1813 *
1814 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001815 * to the dentry. In case of an error the reference on the inode is released.
1816 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1817 * be passed in and will be the error will be propagate to the return value,
1818 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001819 */
1820struct dentry *d_obtain_alias(struct inode *inode)
1821{
NeilBrownb911a6b2012-11-08 16:09:37 -08001822 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001823 struct dentry *tmp;
1824 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001825 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001826
1827 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001828 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001829 if (IS_ERR(inode))
1830 return ERR_CAST(inode);
1831
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001832 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001833 if (res)
1834 goto out_iput;
1835
Al Viroa4464db2011-07-07 15:03:58 -04001836 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001837 if (!tmp) {
1838 res = ERR_PTR(-ENOMEM);
1839 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001840 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001841
Nick Piggin873feea2011-01-07 17:50:06 +11001842 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001843 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001844 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001845 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001846 dput(tmp);
1847 goto out_iput;
1848 }
1849
1850 /* attach a disconnected dentry */
David Howellsb18825a2013-09-12 19:22:53 +01001851 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1852
Christoph Hellwig9308a612008-08-11 15:49:12 +02001853 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001854 tmp->d_inode = inode;
David Howellsb18825a2013-09-12 19:22:53 +01001855 tmp->d_flags |= add_flags;
Al Virob3d9b7a2012-06-09 13:51:19 -04001856 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001857 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001858 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001859 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001860 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001861 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001862 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001863
Christoph Hellwig9308a612008-08-11 15:49:12 +02001864 return tmp;
1865
1866 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001867 if (res && !IS_ERR(res))
1868 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001869 iput(inode);
1870 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001871}
Benny Halevyadc48722009-02-27 14:02:59 -08001872EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
1874/**
1875 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1876 * @inode: the inode which may have a disconnected dentry
1877 * @dentry: a negative dentry which we want to point to the inode.
1878 *
1879 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1880 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1881 * and return it, else simply d_add the inode to the dentry and return NULL.
1882 *
1883 * This is needed in the lookup routine of any filesystem that is exportable
1884 * (via knfsd) so that we can build dcache paths to directories effectively.
1885 *
1886 * If a dentry was found and moved, then it is returned. Otherwise NULL
1887 * is returned. This matches the expected return value of ->lookup.
1888 *
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001889 * Cluster filesystems may call this function with a negative, hashed dentry.
1890 * In that case, we know that the inode will be a regular file, and also this
1891 * will only occur during atomic_open. So we need to check for the dentry
1892 * being already hashed only in the final case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 */
1894struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1895{
1896 struct dentry *new = NULL;
1897
Al Viroa9049372011-07-08 21:20:11 -04001898 if (IS_ERR(inode))
1899 return ERR_CAST(inode);
1900
NeilBrown21c0d8f2006-10-04 02:16:16 -07001901 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001902 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001903 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001905 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001906 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 d_move(new, dentry);
1909 iput(inode);
1910 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001911 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001912 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001913 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 security_d_instantiate(dentry, inode);
1915 d_rehash(dentry);
1916 }
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001917 } else {
1918 d_instantiate(dentry, inode);
1919 if (d_unhashed(dentry))
1920 d_rehash(dentry);
1921 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 return new;
1923}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001924EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925
Barry Naujok94035402008-05-21 16:50:46 +10001926/**
1927 * d_add_ci - lookup or allocate new dentry with case-exact name
1928 * @inode: the inode case-insensitive lookup has found
1929 * @dentry: the negative dentry that was passed to the parent's lookup func
1930 * @name: the case-exact name to be associated with the returned dentry
1931 *
1932 * This is to avoid filling the dcache with case-insensitive names to the
1933 * same inode, only the actual correct case is stored in the dcache for
1934 * case-insensitive filesystems.
1935 *
1936 * For a case-insensitive lookup match and if the the case-exact dentry
1937 * already exists in in the dcache, use it and return it.
1938 *
1939 * If no entry exists with the exact case name, allocate new dentry with
1940 * the exact case, and return the spliced entry.
1941 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001942struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001943 struct qstr *name)
1944{
Barry Naujok94035402008-05-21 16:50:46 +10001945 struct dentry *found;
1946 struct dentry *new;
1947
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001948 /*
1949 * First check if a dentry matching the name already exists,
1950 * if not go ahead and create it now.
1951 */
Barry Naujok94035402008-05-21 16:50:46 +10001952 found = d_hash_and_lookup(dentry->d_parent, name);
Al Viro4f522a22013-02-11 23:20:37 -05001953 if (unlikely(IS_ERR(found)))
1954 goto err_out;
Barry Naujok94035402008-05-21 16:50:46 +10001955 if (!found) {
1956 new = d_alloc(dentry->d_parent, name);
1957 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05001958 found = ERR_PTR(-ENOMEM);
Barry Naujok94035402008-05-21 16:50:46 +10001959 goto err_out;
1960 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001961
Barry Naujok94035402008-05-21 16:50:46 +10001962 found = d_splice_alias(inode, new);
1963 if (found) {
1964 dput(new);
1965 return found;
1966 }
1967 return new;
1968 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001969
1970 /*
1971 * If a matching dentry exists, and it's not negative use it.
1972 *
1973 * Decrement the reference count to balance the iget() done
1974 * earlier on.
1975 */
Barry Naujok94035402008-05-21 16:50:46 +10001976 if (found->d_inode) {
1977 if (unlikely(found->d_inode != inode)) {
1978 /* This can't happen because bad inodes are unhashed. */
1979 BUG_ON(!is_bad_inode(inode));
1980 BUG_ON(!is_bad_inode(found->d_inode));
1981 }
Barry Naujok94035402008-05-21 16:50:46 +10001982 iput(inode);
1983 return found;
1984 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001985
Barry Naujok94035402008-05-21 16:50:46 +10001986 /*
1987 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001988 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001989 */
Al Viro4513d892011-07-17 10:52:14 -04001990 new = d_splice_alias(inode, found);
1991 if (new) {
1992 dput(found);
1993 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10001994 }
Al Viro4513d892011-07-17 10:52:14 -04001995 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001996
1997err_out:
1998 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05001999 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002000}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002001EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002003/*
2004 * Do the slow-case of the dentry name compare.
2005 *
2006 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07002007 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002008 * filesystem can rely on them, and can use the 'name' and
2009 * 'len' information without worrying about walking off the
2010 * end of memory etc.
2011 *
2012 * Thus the read_seqcount_retry() and the "duplicate" info
2013 * in arguments (the low-level filesystem should not look
2014 * at the dentry inode or name contents directly, since
2015 * rename can change them while we're in RCU mode).
2016 */
2017enum slow_d_compare {
2018 D_COMP_OK,
2019 D_COMP_NOMATCH,
2020 D_COMP_SEQRETRY,
2021};
2022
2023static noinline enum slow_d_compare slow_dentry_cmp(
2024 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002025 struct dentry *dentry,
2026 unsigned int seq,
2027 const struct qstr *name)
2028{
2029 int tlen = dentry->d_name.len;
2030 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002031
2032 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2033 cpu_relax();
2034 return D_COMP_SEQRETRY;
2035 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002036 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002037 return D_COMP_NOMATCH;
2038 return D_COMP_OK;
2039}
2040
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002042 * __d_lookup_rcu - search for a dentry (racy, store-free)
2043 * @parent: parent dentry
2044 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002045 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002046 * Returns: dentry, or NULL
2047 *
2048 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2049 * resolution (store-free path walking) design described in
2050 * Documentation/filesystems/path-lookup.txt.
2051 *
2052 * This is not to be used outside core vfs.
2053 *
2054 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2055 * held, and rcu_read_lock held. The returned dentry must not be stored into
2056 * without taking d_lock and checking d_seq sequence count against @seq
2057 * returned here.
2058 *
Linus Torvalds15570082013-09-02 11:38:06 -07002059 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002060 * function.
2061 *
2062 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2063 * the returned dentry, so long as its parent's seqlock is checked after the
2064 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2065 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002066 *
2067 * NOTE! The caller *has* to check the resulting dentry against the sequence
2068 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002069 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002070struct dentry *__d_lookup_rcu(const struct dentry *parent,
2071 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002072 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002073{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002074 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002075 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002076 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002077 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002078 struct dentry *dentry;
2079
2080 /*
2081 * Note: There is significant duplication with __d_lookup_rcu which is
2082 * required to prevent single threaded performance regressions
2083 * especially on architectures where smp_rmb (in seqcounts) are costly.
2084 * Keep the two functions in sync.
2085 */
2086
2087 /*
2088 * The hash list is protected using RCU.
2089 *
2090 * Carefully use d_seq when comparing a candidate dentry, to avoid
2091 * races with d_move().
2092 *
2093 * It is possible that concurrent renames can mess up our list
2094 * walk here and result in missing our dentry, resulting in the
2095 * false-negative result. d_lookup() protects against concurrent
2096 * renames using rename_lock seqlock.
2097 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002098 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002099 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002100 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002101 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002102
Nick Piggin31e6b012011-01-07 17:49:52 +11002103seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002104 /*
2105 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002106 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002107 *
2108 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002109 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002110 *
2111 * NOTE! We do a "raw" seqcount_begin here. That means that
2112 * we don't wait for the sequence count to stabilize if it
2113 * is in the middle of a sequence change. If we do the slow
2114 * dentry compare, we will do seqretries until it is stable,
2115 * and if we end up with a successful lookup, we actually
2116 * want to exit RCU lookup anyway.
2117 */
2118 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002119 if (dentry->d_parent != parent)
2120 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002121 if (d_unhashed(dentry))
2122 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002123
2124 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002125 if (dentry->d_name.hash != hashlen_hash(hashlen))
2126 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002127 *seqp = seq;
2128 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002129 case D_COMP_OK:
2130 return dentry;
2131 case D_COMP_NOMATCH:
2132 continue;
2133 default:
2134 goto seqretry;
2135 }
2136 }
2137
Linus Torvalds26fe5752012-05-10 13:14:12 -07002138 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002139 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002140 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002141 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002142 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002143 }
2144 return NULL;
2145}
2146
2147/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 * d_lookup - search for a dentry
2149 * @parent: parent dentry
2150 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002151 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002153 * d_lookup searches the children of the parent dentry for the name in
2154 * question. If the dentry is found its reference count is incremented and the
2155 * dentry is returned. The caller must use dput to free the entry when it has
2156 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 */
Al Viroda2d8452013-01-24 18:29:34 -05002158struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159{
Nick Piggin31e6b012011-01-07 17:49:52 +11002160 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002161 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162
2163 do {
2164 seq = read_seqbegin(&rename_lock);
2165 dentry = __d_lookup(parent, name);
2166 if (dentry)
2167 break;
2168 } while (read_seqretry(&rename_lock, seq));
2169 return dentry;
2170}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002171EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172
Nick Piggin31e6b012011-01-07 17:49:52 +11002173/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002174 * __d_lookup - search for a dentry (racy)
2175 * @parent: parent dentry
2176 * @name: qstr of name we wish to find
2177 * Returns: dentry, or NULL
2178 *
2179 * __d_lookup is like d_lookup, however it may (rarely) return a
2180 * false-negative result due to unrelated rename activity.
2181 *
2182 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2183 * however it must be used carefully, eg. with a following d_lookup in
2184 * the case of failure.
2185 *
2186 * __d_lookup callers must be commented.
2187 */
Al Viroa713ca22013-01-24 18:27:00 -05002188struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189{
2190 unsigned int len = name->len;
2191 unsigned int hash = name->hash;
2192 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002193 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002194 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002195 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002196 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197
Nick Pigginb04f7842010-08-18 04:37:34 +10002198 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002199 * Note: There is significant duplication with __d_lookup_rcu which is
2200 * required to prevent single threaded performance regressions
2201 * especially on architectures where smp_rmb (in seqcounts) are costly.
2202 * Keep the two functions in sync.
2203 */
2204
2205 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002206 * The hash list is protected using RCU.
2207 *
2208 * Take d_lock when comparing a candidate dentry, to avoid races
2209 * with d_move().
2210 *
2211 * It is possible that concurrent renames can mess up our list
2212 * walk here and result in missing our dentry, resulting in the
2213 * false-negative result. d_lookup() protects against concurrent
2214 * renames using rename_lock seqlock.
2215 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002216 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002217 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 rcu_read_lock();
2219
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002220 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 if (dentry->d_name.hash != hash)
2223 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224
2225 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 if (dentry->d_parent != parent)
2227 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002228 if (d_unhashed(dentry))
2229 goto next;
2230
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 /*
2232 * It is safe to compare names since d_move() cannot
2233 * change the qstr (protected by d_lock).
2234 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002235 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002236 int tlen = dentry->d_name.len;
2237 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002238 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 goto next;
2240 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002241 if (dentry->d_name.len != len)
2242 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002243 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 goto next;
2245 }
2246
Waiman Long98474232013-08-28 18:24:59 -07002247 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002248 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 spin_unlock(&dentry->d_lock);
2250 break;
2251next:
2252 spin_unlock(&dentry->d_lock);
2253 }
2254 rcu_read_unlock();
2255
2256 return found;
2257}
2258
2259/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002260 * d_hash_and_lookup - hash the qstr then search for a dentry
2261 * @dir: Directory to search in
2262 * @name: qstr of name we wish to find
2263 *
Al Viro4f522a22013-02-11 23:20:37 -05002264 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002265 */
2266struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2267{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002268 /*
2269 * Check for a fs-specific hash function. Note that we must
2270 * calculate the standard hash first, as the d_op->d_hash()
2271 * routine may choose to leave the hash value unchanged.
2272 */
2273 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002274 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002275 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002276 if (unlikely(err < 0))
2277 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002278 }
Al Viro4f522a22013-02-11 23:20:37 -05002279 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002280}
Al Viro4f522a22013-02-11 23:20:37 -05002281EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002282
2283/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002284 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002286 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 *
2288 * An insecure source has sent us a dentry, here we verify it and dget() it.
2289 * This is used by ncpfs in its readdir implementation.
2290 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002291 *
2292 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002294int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295{
Nick Piggin786a5e12011-01-07 17:49:16 +11002296 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002297
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002298 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002299 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2300 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002301 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002302 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002303 spin_unlock(&dentry->d_lock);
2304 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 return 1;
2306 }
2307 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002308 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002309
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 return 0;
2311}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002312EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313
2314/*
2315 * When a file is deleted, we have two options:
2316 * - turn this dentry into a negative dentry
2317 * - unhash this dentry and free it.
2318 *
2319 * Usually, we want to just turn this into
2320 * a negative dentry, but if anybody else is
2321 * currently using the dentry or the inode
2322 * we can't do that and we fall back on removing
2323 * it from the hash queues and waiting for
2324 * it to be deleted later when it has no users
2325 */
2326
2327/**
2328 * d_delete - delete a dentry
2329 * @dentry: The dentry to delete
2330 *
2331 * Turn the dentry into a negative dentry if possible, otherwise
2332 * remove it from the hash queues so it can be deleted later
2333 */
2334
2335void d_delete(struct dentry * dentry)
2336{
Nick Piggin873feea2011-01-07 17:50:06 +11002337 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002338 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 /*
2340 * Are we the only user?
2341 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002342again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002344 inode = dentry->d_inode;
2345 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002346 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002347 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002348 spin_unlock(&dentry->d_lock);
2349 cpu_relax();
2350 goto again;
2351 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002352 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002353 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002354 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 return;
2356 }
2357
2358 if (!d_unhashed(dentry))
2359 __d_drop(dentry);
2360
2361 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002362
2363 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002365EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002367static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002369 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002370 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002371 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002372 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002373 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374}
2375
David Howells770bfad2006-08-22 20:06:07 -04002376static void _d_rehash(struct dentry * entry)
2377{
2378 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2379}
2380
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381/**
2382 * d_rehash - add an entry back to the hash
2383 * @entry: dentry to add to the hash
2384 *
2385 * Adds a dentry to the hash according to its name.
2386 */
2387
2388void d_rehash(struct dentry * entry)
2389{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002391 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002394EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002396/**
2397 * dentry_update_name_case - update case insensitive dentry with a new name
2398 * @dentry: dentry to be updated
2399 * @name: new name
2400 *
2401 * Update a case insensitive dentry with new case of name.
2402 *
2403 * dentry must have been returned by d_lookup with name @name. Old and new
2404 * name lengths must match (ie. no d_compare which allows mismatched name
2405 * lengths).
2406 *
2407 * Parent inode i_mutex must be held over d_lookup and into this call (to
2408 * keep renames and concurrent inserts, and readdir(2) away).
2409 */
2410void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2411{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002412 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002413 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2414
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002415 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002416 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002417 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002418 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002419 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002420}
2421EXPORT_SYMBOL(dentry_update_name_case);
2422
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423static void switch_names(struct dentry *dentry, struct dentry *target)
2424{
2425 if (dname_external(target)) {
2426 if (dname_external(dentry)) {
2427 /*
2428 * Both external: swap the pointers
2429 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002430 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 } else {
2432 /*
2433 * dentry:internal, target:external. Steal target's
2434 * storage and make target internal.
2435 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002436 memcpy(target->d_iname, dentry->d_name.name,
2437 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 dentry->d_name.name = target->d_name.name;
2439 target->d_name.name = target->d_iname;
2440 }
2441 } else {
2442 if (dname_external(dentry)) {
2443 /*
2444 * dentry:external, target:internal. Give dentry's
2445 * storage to target and make dentry internal
2446 */
2447 memcpy(dentry->d_iname, target->d_name.name,
2448 target->d_name.len + 1);
2449 target->d_name.name = dentry->d_name.name;
2450 dentry->d_name.name = dentry->d_iname;
2451 } else {
2452 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002453 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002455 unsigned int i;
2456 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2457 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2458 swap(((long *) &dentry->d_iname)[i],
2459 ((long *) &target->d_iname)[i]);
2460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 }
2462 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002463 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464}
2465
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002466static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2467{
2468 /*
2469 * XXXX: do we really need to take target->d_lock?
2470 */
2471 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2472 spin_lock(&target->d_parent->d_lock);
2473 else {
2474 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2475 spin_lock(&dentry->d_parent->d_lock);
2476 spin_lock_nested(&target->d_parent->d_lock,
2477 DENTRY_D_LOCK_NESTED);
2478 } else {
2479 spin_lock(&target->d_parent->d_lock);
2480 spin_lock_nested(&dentry->d_parent->d_lock,
2481 DENTRY_D_LOCK_NESTED);
2482 }
2483 }
2484 if (target < dentry) {
2485 spin_lock_nested(&target->d_lock, 2);
2486 spin_lock_nested(&dentry->d_lock, 3);
2487 } else {
2488 spin_lock_nested(&dentry->d_lock, 2);
2489 spin_lock_nested(&target->d_lock, 3);
2490 }
2491}
2492
2493static void dentry_unlock_parents_for_move(struct dentry *dentry,
2494 struct dentry *target)
2495{
2496 if (target->d_parent != dentry->d_parent)
2497 spin_unlock(&dentry->d_parent->d_lock);
2498 if (target->d_parent != target)
2499 spin_unlock(&target->d_parent->d_lock);
2500}
2501
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002503 * When switching names, the actual string doesn't strictly have to
2504 * be preserved in the target - because we're dropping the target
2505 * anyway. As such, we can just do a simple memcpy() to copy over
2506 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002508 * Note that we have to be a lot more careful about getting the hash
2509 * switched - we have to switch the hash value properly even if it
2510 * then no longer matches the actual (corrupted) string of the target.
2511 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002513/*
Al Viro18367502011-07-12 21:42:24 -04002514 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 * @dentry: entry to move
2516 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002517 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 *
2519 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002520 * dcache entries should not be moved in this way. Caller must hold
2521 * rename_lock, the i_mutex of the source and target directories,
2522 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002524static void __d_move(struct dentry *dentry, struct dentry *target,
2525 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 if (!dentry->d_inode)
2528 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2529
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002530 BUG_ON(d_ancestor(dentry, target));
2531 BUG_ON(d_ancestor(target, dentry));
2532
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002533 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534
Nick Piggin31e6b012011-01-07 17:49:52 +11002535 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002536 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002537
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002538 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2539
2540 /*
2541 * Move the dentry to the target hash queue. Don't bother checking
2542 * for the same hash queue because of how unlikely it is.
2543 */
2544 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002545 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Miklos Szeredida1ce062014-04-01 17:08:43 +02002547 /*
2548 * Unhash the target (d_delete() is not usable here). If exchanging
2549 * the two dentries, then rehash onto the other's hash queue.
2550 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002552 if (exchange) {
2553 __d_rehash(target,
2554 d_hash(dentry->d_parent, dentry->d_name.hash));
2555 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556
Eric Dumazet5160ee62006-01-08 01:03:32 -08002557 list_del(&dentry->d_u.d_child);
2558 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559
2560 /* Switch the names.. */
2561 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002562 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563
2564 /* ... and switch the parents */
2565 if (IS_ROOT(dentry)) {
2566 dentry->d_parent = target->d_parent;
2567 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002568 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002570 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571
2572 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002573 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 }
2575
Eric Dumazet5160ee62006-01-08 01:03:32 -08002576 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002577
Nick Piggin31e6b012011-01-07 17:49:52 +11002578 write_seqcount_end(&target->d_seq);
2579 write_seqcount_end(&dentry->d_seq);
2580
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002581 dentry_unlock_parents_for_move(dentry, target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002582 if (exchange)
2583 fsnotify_d_move(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002585 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002587}
2588
2589/*
2590 * d_move - move a dentry
2591 * @dentry: entry to move
2592 * @target: new dentry
2593 *
2594 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002595 * dcache entries should not be moved in this way. See the locking
2596 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002597 */
2598void d_move(struct dentry *dentry, struct dentry *target)
2599{
2600 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002601 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002603}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002604EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
Miklos Szeredida1ce062014-04-01 17:08:43 +02002606/*
2607 * d_exchange - exchange two dentries
2608 * @dentry1: first dentry
2609 * @dentry2: second dentry
2610 */
2611void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2612{
2613 write_seqlock(&rename_lock);
2614
2615 WARN_ON(!dentry1->d_inode);
2616 WARN_ON(!dentry2->d_inode);
2617 WARN_ON(IS_ROOT(dentry1));
2618 WARN_ON(IS_ROOT(dentry2));
2619
2620 __d_move(dentry1, dentry2, true);
2621
2622 write_sequnlock(&rename_lock);
2623}
2624
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002625/**
2626 * d_ancestor - search for an ancestor
2627 * @p1: ancestor dentry
2628 * @p2: child dentry
2629 *
2630 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2631 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002632 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002633struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002634{
2635 struct dentry *p;
2636
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002637 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002638 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002639 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002640 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002641 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002642}
2643
2644/*
2645 * This helper attempts to cope with remotely renamed directories
2646 *
2647 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002648 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002649 *
2650 * Note: If ever the locking in lock_rename() changes, then please
2651 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002652 */
Nick Piggin873feea2011-01-07 17:50:06 +11002653static struct dentry *__d_unalias(struct inode *inode,
2654 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002655{
2656 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002657 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002658
2659 /* If alias and dentry share a parent, then no extra locks required */
2660 if (alias->d_parent == dentry->d_parent)
2661 goto out_unalias;
2662
Trond Myklebust9eaef272006-10-21 10:24:20 -07002663 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002664 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2665 goto out_err;
2666 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2667 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2668 goto out_err;
2669 m2 = &alias->d_parent->d_inode->i_mutex;
2670out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002671 if (likely(!d_mountpoint(alias))) {
Miklos Szeredida1ce062014-04-01 17:08:43 +02002672 __d_move(alias, dentry, false);
Al Viroee3efa92012-06-08 15:59:33 -04002673 ret = alias;
2674 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002675out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002676 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002677 if (m2)
2678 mutex_unlock(m2);
2679 if (m1)
2680 mutex_unlock(m1);
2681 return ret;
2682}
2683
2684/*
David Howells770bfad2006-08-22 20:06:07 -04002685 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2686 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002687 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002688 */
2689static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2690{
Al Viro740da422013-01-30 10:13:38 -05002691 struct dentry *dparent;
David Howells770bfad2006-08-22 20:06:07 -04002692
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002693 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002694
Nick Piggin31e6b012011-01-07 17:49:52 +11002695 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002696 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002697
David Howells770bfad2006-08-22 20:06:07 -04002698 dparent = dentry->d_parent;
David Howells770bfad2006-08-22 20:06:07 -04002699
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002700 switch_names(dentry, anon);
2701 swap(dentry->d_name.hash, anon->d_name.hash);
2702
Al Viro740da422013-01-30 10:13:38 -05002703 dentry->d_parent = dentry;
2704 list_del_init(&dentry->d_u.d_child);
2705 anon->d_parent = dparent;
Wei Yongjun9ed53b12013-03-12 00:10:50 +08002706 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
David Howells770bfad2006-08-22 20:06:07 -04002707
Nick Piggin31e6b012011-01-07 17:49:52 +11002708 write_seqcount_end(&dentry->d_seq);
2709 write_seqcount_end(&anon->d_seq);
2710
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002711 dentry_unlock_parents_for_move(anon, dentry);
2712 spin_unlock(&dentry->d_lock);
2713
2714 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002715}
2716
2717/**
2718 * d_materialise_unique - introduce an inode into the tree
2719 * @dentry: candidate dentry
2720 * @inode: inode to bind to the dentry, to which aliases may be attached
2721 *
2722 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002723 * root directory alias in its place if there is one. Caller must hold the
2724 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002725 */
2726struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2727{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002728 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002729
2730 BUG_ON(!d_unhashed(dentry));
2731
David Howells770bfad2006-08-22 20:06:07 -04002732 if (!inode) {
2733 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002734 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002735 d_rehash(actual);
2736 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002737 }
2738
Nick Piggin873feea2011-01-07 17:50:06 +11002739 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002740
Trond Myklebust9eaef272006-10-21 10:24:20 -07002741 if (S_ISDIR(inode->i_mode)) {
2742 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002743
Trond Myklebust9eaef272006-10-21 10:24:20 -07002744 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002745 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002746 if (alias) {
2747 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002748 write_seqlock(&rename_lock);
2749
2750 if (d_ancestor(alias, dentry)) {
2751 /* Check for loops */
2752 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002753 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002754 } else if (IS_ROOT(alias)) {
2755 /* Is this an anonymous mountpoint that we
2756 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002757 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002758 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002759 __d_drop(alias);
2760 goto found;
Al Viro18367502011-07-12 21:42:24 -04002761 } else {
2762 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002763 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002764 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002765 }
Al Viro18367502011-07-12 21:42:24 -04002766 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002767 if (IS_ERR(actual)) {
2768 if (PTR_ERR(actual) == -ELOOP)
2769 pr_warn_ratelimited(
2770 "VFS: Lookup of '%s' in %s %s"
2771 " would have caused loop\n",
2772 dentry->d_name.name,
2773 inode->i_sb->s_type->name,
2774 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002775 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002776 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002777 goto out_nolock;
2778 }
David Howells770bfad2006-08-22 20:06:07 -04002779 }
2780
2781 /* Add a unique reference */
2782 actual = __d_instantiate_unique(dentry, inode);
2783 if (!actual)
2784 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002785 else
2786 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002787
David Howells770bfad2006-08-22 20:06:07 -04002788 spin_lock(&actual->d_lock);
2789found:
2790 _d_rehash(actual);
2791 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002792 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002793out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002794 if (actual == dentry) {
2795 security_d_instantiate(dentry, inode);
2796 return NULL;
2797 }
2798
2799 iput(inode);
2800 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002801}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002802EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002803
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002804static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002805{
2806 *buflen -= namelen;
2807 if (*buflen < 0)
2808 return -ENAMETOOLONG;
2809 *buffer -= namelen;
2810 memcpy(*buffer, str, namelen);
2811 return 0;
2812}
2813
Waiman Long232d2d62013-09-09 12:18:13 -04002814/**
2815 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002816 * @buffer: buffer pointer
2817 * @buflen: allocated length of the buffer
2818 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002819 *
2820 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2821 * make sure that either the old or the new name pointer and length are
2822 * fetched. However, there may be mismatch between length and pointer.
2823 * The length cannot be trusted, we need to copy it byte-by-byte until
2824 * the length is reached or a null byte is found. It also prepends "/" at
2825 * the beginning of the name. The sequence number check at the caller will
2826 * retry it again when a d_move() does happen. So any garbage in the buffer
2827 * due to mismatched pointer and length will be discarded.
2828 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002829static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2830{
Waiman Long232d2d62013-09-09 12:18:13 -04002831 const char *dname = ACCESS_ONCE(name->name);
2832 u32 dlen = ACCESS_ONCE(name->len);
2833 char *p;
2834
Waiman Long232d2d62013-09-09 12:18:13 -04002835 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04002836 if (*buflen < 0)
2837 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04002838 p = *buffer -= dlen + 1;
2839 *p++ = '/';
2840 while (dlen--) {
2841 char c = *dname++;
2842 if (!c)
2843 break;
2844 *p++ = c;
2845 }
2846 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002847}
2848
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002850 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002851 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002852 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002853 * @buffer: pointer to the end of the buffer
2854 * @buflen: pointer to buffer length
2855 *
Waiman Long18129972013-09-12 10:55:35 -04002856 * The function will first try to write out the pathname without taking any
2857 * lock other than the RCU read lock to make sure that dentries won't go away.
2858 * It only checks the sequence number of the global rename_lock as any change
2859 * in the dentry's d_seq will be preceded by changes in the rename_lock
2860 * sequence number. If the sequence number had been changed, it will restart
2861 * the whole pathname back-tracing sequence again by taking the rename_lock.
2862 * In this case, there is no need to take the RCU read lock as the recursive
2863 * parent pointer references will keep the dentry chain alive as long as no
2864 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002865 */
Al Viro02125a82011-12-05 08:43:34 -05002866static int prepend_path(const struct path *path,
2867 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002868 char **buffer, int *buflen)
2869{
Al Viroede4ceb2013-11-13 07:45:40 -05002870 struct dentry *dentry;
2871 struct vfsmount *vfsmnt;
2872 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002873 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002874 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002875 char *bptr;
2876 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002877
Al Viro48f5ec22013-09-09 15:22:25 -04002878 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002879restart_mnt:
2880 read_seqbegin_or_lock(&mount_lock, &m_seq);
2881 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002882 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002883restart:
2884 bptr = *buffer;
2885 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002886 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002887 dentry = path->dentry;
2888 vfsmnt = path->mnt;
2889 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002890 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002891 while (dentry != root->dentry || vfsmnt != root->mnt) {
2892 struct dentry * parent;
2893
2894 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002895 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002896 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002897 if (mnt != parent) {
2898 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2899 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002900 vfsmnt = &mnt->mnt;
2901 continue;
2902 }
2903 /*
2904 * Filesystems needing to implement special "root names"
2905 * should do so with ->d_dname()
2906 */
2907 if (IS_ROOT(dentry) &&
2908 (dentry->d_name.len != 1 ||
2909 dentry->d_name.name[0] != '/')) {
2910 WARN(1, "Root dentry has weird name <%.*s>\n",
2911 (int) dentry->d_name.len,
2912 dentry->d_name.name);
2913 }
2914 if (!error)
2915 error = is_mounted(vfsmnt) ? 1 : 2;
2916 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002917 }
2918 parent = dentry->d_parent;
2919 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002920 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002921 if (error)
2922 break;
2923
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002924 dentry = parent;
2925 }
Al Viro48f5ec22013-09-09 15:22:25 -04002926 if (!(seq & 1))
2927 rcu_read_unlock();
2928 if (need_seqretry(&rename_lock, seq)) {
2929 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002930 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002931 }
Al Viro48f5ec22013-09-09 15:22:25 -04002932 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002933
2934 if (!(m_seq & 1))
2935 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002936 if (need_seqretry(&mount_lock, m_seq)) {
2937 m_seq = 1;
2938 goto restart_mnt;
2939 }
2940 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002941
Waiman Long232d2d62013-09-09 12:18:13 -04002942 if (error >= 0 && bptr == *buffer) {
2943 if (--blen < 0)
2944 error = -ENAMETOOLONG;
2945 else
2946 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002947 }
Waiman Long232d2d62013-09-09 12:18:13 -04002948 *buffer = bptr;
2949 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002950 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002951}
2952
2953/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002954 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002955 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002956 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002957 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958 * @buflen: buffer length
2959 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002960 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002962 * Returns a pointer into the buffer or an error code if the
2963 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002964 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002965 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002966 *
Al Viro02125a82011-12-05 08:43:34 -05002967 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968 */
Al Viro02125a82011-12-05 08:43:34 -05002969char *__d_path(const struct path *path,
2970 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002971 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002973 char *res = buf + buflen;
2974 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002976 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002977 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002978
Al Viro02125a82011-12-05 08:43:34 -05002979 if (error < 0)
2980 return ERR_PTR(error);
2981 if (error > 0)
2982 return NULL;
2983 return res;
2984}
2985
2986char *d_absolute_path(const struct path *path,
2987 char *buf, int buflen)
2988{
2989 struct path root = {};
2990 char *res = buf + buflen;
2991 int error;
2992
2993 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002994 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05002995
2996 if (error > 1)
2997 error = -EINVAL;
2998 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002999 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003000 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001}
3002
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003003/*
3004 * same as __d_path but appends "(deleted)" for unlinked files.
3005 */
Al Viro02125a82011-12-05 08:43:34 -05003006static int path_with_deleted(const struct path *path,
3007 const struct path *root,
3008 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003009{
3010 prepend(buf, buflen, "\0", 1);
3011 if (d_unlinked(path->dentry)) {
3012 int error = prepend(buf, buflen, " (deleted)", 10);
3013 if (error)
3014 return error;
3015 }
3016
3017 return prepend_path(path, root, buf, buflen);
3018}
3019
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003020static int prepend_unreachable(char **buffer, int *buflen)
3021{
3022 return prepend(buffer, buflen, "(unreachable)", 13);
3023}
3024
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003025static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3026{
3027 unsigned seq;
3028
3029 do {
3030 seq = read_seqcount_begin(&fs->seq);
3031 *root = fs->root;
3032 } while (read_seqcount_retry(&fs->seq, seq));
3033}
3034
Jan Bluncka03a8a702008-02-14 19:38:32 -08003035/**
3036 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003037 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003038 * @buf: buffer to return value in
3039 * @buflen: buffer length
3040 *
3041 * Convert a dentry into an ASCII path name. If the entry has been deleted
3042 * the string " (deleted)" is appended. Note that this is ambiguous.
3043 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003044 * Returns a pointer into the buffer or an error code if the path was
3045 * too long. Note: Callers should use the returned pointer, not the passed
3046 * in buffer, to use the name! The implementation often starts at an offset
3047 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003048 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003049 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003050 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003051char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003053 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003054 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003055 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003057 /*
3058 * We have various synthetic filesystems that never get mounted. On
3059 * these filesystems dentries are never used for lookup purposes, and
3060 * thus don't need to be hashed. They also don't need a name until a
3061 * user wants to identify the object in /proc/pid/fd/. The little hack
3062 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003063 *
3064 * Some pseudo inodes are mountable. When they are mounted
3065 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3066 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003067 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003068 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3069 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003070 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003071
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003072 rcu_read_lock();
3073 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003074 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003075 rcu_read_unlock();
3076
Al Viro02125a82011-12-05 08:43:34 -05003077 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003078 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 return res;
3080}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003081EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082
3083/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003084 * Helper function for dentry_operations.d_dname() members
3085 */
3086char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3087 const char *fmt, ...)
3088{
3089 va_list args;
3090 char temp[64];
3091 int sz;
3092
3093 va_start(args, fmt);
3094 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3095 va_end(args);
3096
3097 if (sz > sizeof(temp) || sz > buflen)
3098 return ERR_PTR(-ENAMETOOLONG);
3099
3100 buffer += buflen - sz;
3101 return memcpy(buffer, temp, sz);
3102}
3103
Al Viro118b2302013-08-24 12:08:17 -04003104char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3105{
3106 char *end = buffer + buflen;
3107 /* these dentries are never renamed, so d_lock is not needed */
3108 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003109 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003110 prepend(&end, &buflen, "/", 1))
3111 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003112 return end;
Al Viro118b2302013-08-24 12:08:17 -04003113}
David Herrmann31bbe162014-01-03 14:09:47 +01003114EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003115
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003116/*
Ram Pai6092d042008-03-27 13:06:20 +01003117 * Write full pathname from the root of the filesystem into the buffer.
3118 */
Al Virof6500802014-01-26 12:37:55 -05003119static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003120{
Al Virof6500802014-01-26 12:37:55 -05003121 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003122 char *end, *retval;
3123 int len, seq = 0;
3124 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003125
Al Virof6500802014-01-26 12:37:55 -05003126 if (buflen < 2)
3127 goto Elong;
3128
Al Viro48f5ec22013-09-09 15:22:25 -04003129 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003130restart:
Al Virof6500802014-01-26 12:37:55 -05003131 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003132 end = buf + buflen;
3133 len = buflen;
3134 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003135 /* Get '/' right */
3136 retval = end-1;
3137 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003138 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003139 while (!IS_ROOT(dentry)) {
3140 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003141
Ram Pai6092d042008-03-27 13:06:20 +01003142 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003143 error = prepend_name(&end, &len, &dentry->d_name);
3144 if (error)
3145 break;
Ram Pai6092d042008-03-27 13:06:20 +01003146
3147 retval = end;
3148 dentry = parent;
3149 }
Al Viro48f5ec22013-09-09 15:22:25 -04003150 if (!(seq & 1))
3151 rcu_read_unlock();
3152 if (need_seqretry(&rename_lock, seq)) {
3153 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003154 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003155 }
3156 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003157 if (error)
3158 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003159 return retval;
3160Elong:
3161 return ERR_PTR(-ENAMETOOLONG);
3162}
Nick Pigginec2447c2011-01-07 17:49:29 +11003163
3164char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3165{
Waiman Long232d2d62013-09-09 12:18:13 -04003166 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003167}
3168EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003169
3170char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3171{
3172 char *p = NULL;
3173 char *retval;
3174
Al Viroc1031352010-06-06 22:31:14 -04003175 if (d_unlinked(dentry)) {
3176 p = buf + buflen;
3177 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3178 goto Elong;
3179 buflen++;
3180 }
3181 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003182 if (!IS_ERR(retval) && p)
3183 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003184 return retval;
3185Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003186 return ERR_PTR(-ENAMETOOLONG);
3187}
3188
Linus Torvalds8b19e342013-09-12 10:35:47 -07003189static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3190 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003191{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003192 unsigned seq;
3193
3194 do {
3195 seq = read_seqcount_begin(&fs->seq);
3196 *root = fs->root;
3197 *pwd = fs->pwd;
3198 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003199}
3200
Ram Pai6092d042008-03-27 13:06:20 +01003201/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202 * NOTE! The user-level library version returns a
3203 * character pointer. The kernel system call just
3204 * returns the length of the buffer filled (which
3205 * includes the ending '\0' character), or a negative
3206 * error value. So libc would do something like
3207 *
3208 * char *getcwd(char * buf, size_t size)
3209 * {
3210 * int retval;
3211 *
3212 * retval = sys_getcwd(buf, size);
3213 * if (retval >= 0)
3214 * return buf;
3215 * errno = -retval;
3216 * return NULL;
3217 * }
3218 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003219SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220{
Linus Torvalds552ce542007-02-13 12:08:18 -08003221 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003222 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003223 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224
3225 if (!page)
3226 return -ENOMEM;
3227
Linus Torvalds8b19e342013-09-12 10:35:47 -07003228 rcu_read_lock();
3229 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230
Linus Torvalds552ce542007-02-13 12:08:18 -08003231 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003232 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003233 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003234 char *cwd = page + PATH_MAX;
3235 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003237 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003238 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003239 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003240
Al Viro02125a82011-12-05 08:43:34 -05003241 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003242 goto out;
3243
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003244 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003245 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003246 error = prepend_unreachable(&cwd, &buflen);
3247 if (error)
3248 goto out;
3249 }
3250
Linus Torvalds552ce542007-02-13 12:08:18 -08003251 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003252 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003253 if (len <= size) {
3254 error = len;
3255 if (copy_to_user(buf, cwd, len))
3256 error = -EFAULT;
3257 }
Nick Piggin949854d2011-01-07 17:49:37 +11003258 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003259 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261
3262out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003263 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264 return error;
3265}
3266
3267/*
3268 * Test whether new_dentry is a subdirectory of old_dentry.
3269 *
3270 * Trivially implemented using the dcache structure
3271 */
3272
3273/**
3274 * is_subdir - is new dentry a subdirectory of old_dentry
3275 * @new_dentry: new dentry
3276 * @old_dentry: old dentry
3277 *
3278 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3279 * Returns 0 otherwise.
3280 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3281 */
3282
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003283int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284{
3285 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11003286 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003288 if (new_dentry == old_dentry)
3289 return 1;
3290
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003291 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003294 /*
3295 * Need rcu_readlock to protect against the d_parent trashing
3296 * due to d_move
3297 */
3298 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003299 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003301 else
3302 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11003303 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003304 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305
3306 return result;
3307}
3308
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003309static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003311 struct dentry *root = data;
3312 if (dentry != root) {
3313 if (d_unhashed(dentry) || !dentry->d_inode)
3314 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003316 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3317 dentry->d_flags |= DCACHE_GENOCIDE;
3318 dentry->d_lockref.count--;
3319 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003321 return D_WALK_CONTINUE;
3322}
Nick Piggin58db63d2011-01-07 17:49:39 +11003323
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003324void d_genocide(struct dentry *parent)
3325{
3326 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003327}
3328
Al Viro60545d02013-06-07 01:20:27 -04003329void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330{
Al Viro60545d02013-06-07 01:20:27 -04003331 inode_dec_link_count(inode);
3332 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3333 !hlist_unhashed(&dentry->d_alias) ||
3334 !d_unlinked(dentry));
3335 spin_lock(&dentry->d_parent->d_lock);
3336 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3337 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3338 (unsigned long long)inode->i_ino);
3339 spin_unlock(&dentry->d_lock);
3340 spin_unlock(&dentry->d_parent->d_lock);
3341 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342}
Al Viro60545d02013-06-07 01:20:27 -04003343EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344
3345static __initdata unsigned long dhash_entries;
3346static int __init set_dhash_entries(char *str)
3347{
3348 if (!str)
3349 return 0;
3350 dhash_entries = simple_strtoul(str, &str, 0);
3351 return 1;
3352}
3353__setup("dhash_entries=", set_dhash_entries);
3354
3355static void __init dcache_init_early(void)
3356{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003357 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358
3359 /* If hashes are distributed across NUMA nodes, defer
3360 * hash allocation until vmalloc space is available.
3361 */
3362 if (hashdist)
3363 return;
3364
3365 dentry_hashtable =
3366 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003367 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368 dhash_entries,
3369 13,
3370 HASH_EARLY,
3371 &d_hash_shift,
3372 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003373 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374 0);
3375
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003376 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003377 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378}
3379
Denis Cheng74bf17c2007-10-16 23:26:30 -07003380static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003382 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383
3384 /*
3385 * A constructor could be added for stable state like the lists,
3386 * but it is probably not worth it because of the cache nature
3387 * of the dcache.
3388 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003389 dentry_cache = KMEM_CACHE(dentry,
3390 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391
3392 /* Hash may have been set up in dcache_init_early */
3393 if (!hashdist)
3394 return;
3395
3396 dentry_hashtable =
3397 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003398 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399 dhash_entries,
3400 13,
3401 0,
3402 &d_hash_shift,
3403 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003404 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405 0);
3406
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003407 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003408 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409}
3410
3411/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003412struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003413EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415EXPORT_SYMBOL(d_genocide);
3416
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417void __init vfs_caches_init_early(void)
3418{
3419 dcache_init_early();
3420 inode_init_early();
3421}
3422
3423void __init vfs_caches_init(unsigned long mempages)
3424{
3425 unsigned long reserve;
3426
3427 /* Base hash sizes on available memory, with a reserve equal to
3428 150% of current kernel size */
3429
3430 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3431 mempages -= reserve;
3432
3433 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003434 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003435
Denis Cheng74bf17c2007-10-16 23:26:30 -07003436 dcache_init();
3437 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003439 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440 bdev_cache_init();
3441 chrdev_init();
3442}