blob: d54a99baf4f314d43bf1d79abf814da0ff3bef01 [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
Miklos Szeredid52b9082007-05-08 00:23:46 -0700398/**
Nick Piggin789680d2011-01-07 17:49:30 +1100399 * d_drop - drop a dentry
400 * @dentry: dentry to drop
401 *
402 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
403 * be found through a VFS lookup any more. Note that this is different from
404 * deleting the dentry - d_delete will try to mark the dentry negative if
405 * possible, giving a successful _negative_ lookup, while d_drop will
406 * just make the cache lookup fail.
407 *
408 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
409 * reason (NFS timeouts or autofs deletes).
410 *
411 * __d_drop requires dentry->d_lock.
412 */
413void __d_drop(struct dentry *dentry)
414{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700415 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400416 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400417 /*
418 * Hashed dentries are normally on the dentry hashtable,
419 * with the exception of those newly allocated by
420 * d_obtain_alias, which are always IS_ROOT:
421 */
422 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400423 b = &dentry->d_sb->s_anon;
424 else
425 b = d_hash(dentry->d_parent, dentry->d_name.hash);
426
427 hlist_bl_lock(b);
428 __hlist_bl_del(&dentry->d_hash);
429 dentry->d_hash.pprev = NULL;
430 hlist_bl_unlock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700431 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100432 }
433}
434EXPORT_SYMBOL(__d_drop);
435
436void d_drop(struct dentry *dentry)
437{
Nick Piggin789680d2011-01-07 17:49:30 +1100438 spin_lock(&dentry->d_lock);
439 __d_drop(dentry);
440 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100441}
442EXPORT_SYMBOL(d_drop);
443
Al Viroe55fd012014-05-28 13:51:12 -0400444static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100445{
Al Viro41edf272014-05-01 10:30:00 -0400446 struct dentry *parent = NULL;
447 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400448 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100449 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100450
Linus Torvalds0d984392013-09-08 13:46:52 -0700451 /*
452 * The dentry is now unrecoverably dead to the world.
453 */
454 lockref_mark_dead(&dentry->d_lockref);
455
Sage Weilf0023bc2011-10-28 10:02:42 -0700456 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700457 * inform the fs via d_prune that this dentry is about to be
458 * unhashed and destroyed.
459 */
Yan, Zheng590fb512013-08-13 15:42:02 +0800460 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
Yan, Zheng61572bb2013-04-15 14:13:21 +0800461 dentry->d_op->d_prune(dentry);
462
Al Viro01b60352014-04-29 23:42:52 -0400463 if (dentry->d_flags & DCACHE_LRU_LIST) {
464 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
465 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400466 }
Nick Piggin77812a12011-01-07 17:49:48 +1100467 /* if it was on the hash then remove it */
468 __d_drop(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400469 list_del(&dentry->d_u.d_child);
470 /*
471 * Inform d_walk() that we are no longer attached to the
472 * dentry tree
473 */
474 dentry->d_flags |= DCACHE_DENTRY_KILLED;
475 if (parent)
476 spin_unlock(&parent->d_lock);
477 dentry_iput(dentry);
478 /*
479 * dentry_iput drops the locks, at which point nobody (except
480 * transient RCU lookups) can reach this dentry.
481 */
482 BUG_ON((int)dentry->d_lockref.count > 0);
483 this_cpu_dec(nr_dentry);
484 if (dentry->d_op && dentry->d_op->d_release)
485 dentry->d_op->d_release(dentry);
486
Al Viro41edf272014-05-01 10:30:00 -0400487 spin_lock(&dentry->d_lock);
488 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
489 dentry->d_flags |= DCACHE_MAY_FREE;
490 can_free = false;
491 }
492 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400493 if (likely(can_free))
494 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400495}
496
497/*
498 * Finish off a dentry we've decided to kill.
499 * dentry->d_lock must be held, returns with it unlocked.
500 * If ref is non-zero, then decrement the refcount too.
501 * Returns dentry requiring refcount drop, or NULL if we're done.
502 */
503static struct dentry *
504dentry_kill(struct dentry *dentry, int unlock_on_failure)
505 __releases(dentry->d_lock)
506{
507 struct inode *inode = dentry->d_inode;
508 struct dentry *parent = NULL;
509
510 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
511 goto failed;
512
513 if (!IS_ROOT(dentry)) {
514 parent = dentry->d_parent;
515 if (unlikely(!spin_trylock(&parent->d_lock))) {
516 if (inode)
517 spin_unlock(&inode->i_lock);
518 goto failed;
519 }
520 }
521
522 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400523 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400524
525failed:
526 if (unlock_on_failure) {
527 spin_unlock(&dentry->d_lock);
528 cpu_relax();
529 }
530 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100531}
532
Al Viro046b9612014-05-29 08:54:52 -0400533static inline struct dentry *lock_parent(struct dentry *dentry)
534{
535 struct dentry *parent = dentry->d_parent;
536 if (IS_ROOT(dentry))
537 return NULL;
538 if (likely(spin_trylock(&parent->d_lock)))
539 return parent;
540 spin_unlock(&dentry->d_lock);
541 rcu_read_lock();
542again:
543 parent = ACCESS_ONCE(dentry->d_parent);
544 spin_lock(&parent->d_lock);
545 /*
546 * We can't blindly lock dentry until we are sure
547 * that we won't violate the locking order.
548 * Any changes of dentry->d_parent must have
549 * been done with parent->d_lock held, so
550 * spin_lock() above is enough of a barrier
551 * for checking if it's still our child.
552 */
553 if (unlikely(parent != dentry->d_parent)) {
554 spin_unlock(&parent->d_lock);
555 goto again;
556 }
557 rcu_read_unlock();
558 if (parent != dentry)
559 spin_lock(&dentry->d_lock);
560 else
561 parent = NULL;
562 return parent;
563}
564
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565/*
566 * This is dput
567 *
568 * This is complicated by the fact that we do not want to put
569 * dentries that are no longer on any hash chain on the unused
570 * list: we'd much rather just get rid of them immediately.
571 *
572 * However, that implies that we have to traverse the dentry
573 * tree upwards to the parents which might _also_ now be
574 * scheduled for deletion (it may have been only waiting for
575 * its last child to go away).
576 *
577 * This tail recursion is done by hand as we don't want to depend
578 * on the compiler to always get this right (gcc generally doesn't).
579 * Real recursion would eat up our stack space.
580 */
581
582/*
583 * dput - release a dentry
584 * @dentry: dentry to release
585 *
586 * Release a dentry. This will drop the usage count and if appropriate
587 * call the dentry unlink method as well as removing it from the queues and
588 * releasing its resources. If the parent dentries were scheduled for release
589 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591void dput(struct dentry *dentry)
592{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700593 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 return;
595
596repeat:
Waiman Long98474232013-08-28 18:24:59 -0700597 if (lockref_put_or_lock(&dentry->d_lockref))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700600 /* Unreachable? Get rid of it */
601 if (unlikely(d_unhashed(dentry)))
602 goto kill_it;
603
604 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100606 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 }
Nick Piggin265ac902010-10-10 05:36:24 -0400608
Linus Torvalds358eec12013-10-31 15:43:02 -0700609 if (!(dentry->d_flags & DCACHE_REFERENCED))
610 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400611 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400612
Waiman Long98474232013-08-28 18:24:59 -0700613 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100614 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 return;
616
Miklos Szeredid52b9082007-05-08 00:23:46 -0700617kill_it:
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000618 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700619 if (dentry)
620 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700622EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
624/**
625 * d_invalidate - invalidate a dentry
626 * @dentry: dentry to invalidate
627 *
628 * Try to invalidate the dentry if it turns out to be
629 * possible. If there are other dentries that can be
630 * reached through this one we can't delete it and we
631 * return -EBUSY. On success we return 0.
632 *
633 * no dcache lock.
634 */
635
636int d_invalidate(struct dentry * dentry)
637{
638 /*
639 * If it's already been dropped, return OK.
640 */
Nick Pigginda502952011-01-07 17:49:33 +1100641 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100643 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 return 0;
645 }
646 /*
647 * Check whether to do a partial shrink_dcache
648 * to get rid of unused child entries.
649 */
650 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100651 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100653 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 }
655
656 /*
657 * Somebody else still using it?
658 *
659 * If it's a directory, we can't drop it
660 * for fear of somebody re-populating it
661 * with children (even though dropping it
662 * would make it unreachable from the root,
663 * we might still populate it if it was a
664 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000665 * We also need to leave mountpoints alone,
666 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 */
Waiman Long98474232013-08-28 18:24:59 -0700668 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
Al Viro50e69632011-11-07 16:39:57 +0000669 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 return -EBUSY;
672 }
673 }
674
675 __d_drop(dentry);
676 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 return 0;
678}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700679EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100681/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100682static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
Waiman Long98474232013-08-28 18:24:59 -0700684 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685}
686
Nick Piggindc0474b2011-01-07 17:49:43 +1100687static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100688{
Waiman Long98474232013-08-28 18:24:59 -0700689 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100690}
691
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100692struct dentry *dget_parent(struct dentry *dentry)
693{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700694 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100695 struct dentry *ret;
696
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700697 /*
698 * Do optimistic parent lookup without any
699 * locking.
700 */
701 rcu_read_lock();
702 ret = ACCESS_ONCE(dentry->d_parent);
703 gotref = lockref_get_not_zero(&ret->d_lockref);
704 rcu_read_unlock();
705 if (likely(gotref)) {
706 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
707 return ret;
708 dput(ret);
709 }
710
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100711repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100712 /*
713 * Don't need rcu_dereference because we re-check it was correct under
714 * the lock.
715 */
716 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100717 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100718 spin_lock(&ret->d_lock);
719 if (unlikely(ret != dentry->d_parent)) {
720 spin_unlock(&ret->d_lock);
721 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100722 goto repeat;
723 }
Nick Piggina734eb42011-01-07 17:49:44 +1100724 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700725 BUG_ON(!ret->d_lockref.count);
726 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100727 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100728 return ret;
729}
730EXPORT_SYMBOL(dget_parent);
731
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732/**
733 * d_find_alias - grab a hashed alias of inode
734 * @inode: inode in question
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700735 * @want_discon: flag, used by d_splice_alias, to request
736 * that only a DISCONNECTED alias be returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 *
738 * If inode has a hashed alias, or is a directory and has any alias,
739 * acquire the reference to alias and return it. Otherwise return NULL.
740 * Notice that if inode is a directory there can be only one alias and
741 * it can be unhashed only if it has no children, or if it is the root
742 * of a filesystem.
743 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700744 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700745 * any other hashed alias over that one unless @want_discon is set,
746 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 */
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700748static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
Nick Pigginda502952011-01-07 17:49:33 +1100750 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
Nick Pigginda502952011-01-07 17:49:33 +1100752again:
753 discon_alias = NULL;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800754 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100755 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700757 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100758 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 discon_alias = alias;
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700760 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100761 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100762 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 return alias;
764 }
765 }
Nick Pigginda502952011-01-07 17:49:33 +1100766 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 }
Nick Pigginda502952011-01-07 17:49:33 +1100768 if (discon_alias) {
769 alias = discon_alias;
770 spin_lock(&alias->d_lock);
771 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
772 if (IS_ROOT(alias) &&
773 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100774 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100775 spin_unlock(&alias->d_lock);
776 return alias;
777 }
778 }
779 spin_unlock(&alias->d_lock);
780 goto again;
781 }
782 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783}
784
Nick Pigginda502952011-01-07 17:49:33 +1100785struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786{
David Howells214fda12006-03-25 03:06:36 -0800787 struct dentry *de = NULL;
788
Al Virob3d9b7a2012-06-09 13:51:19 -0400789 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100790 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700791 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100792 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800793 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 return de;
795}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700796EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
798/*
799 * Try to kill dentries associated with this inode.
800 * WARNING: you must own a reference to inode.
801 */
802void d_prune_aliases(struct inode *inode)
803{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700804 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100806 spin_lock(&inode->i_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800807 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700809 if (!dentry->d_lockref.count) {
Yan, Zheng590fb512013-08-13 15:42:02 +0800810 /*
811 * inform the fs via d_prune that this dentry
812 * is about to be unhashed and destroyed.
813 */
814 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
815 !d_unhashed(dentry))
816 dentry->d_op->d_prune(dentry);
817
Nick Piggindc0474b2011-01-07 17:49:43 +1100818 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 __d_drop(dentry);
820 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100821 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 dput(dentry);
823 goto restart;
824 }
825 spin_unlock(&dentry->d_lock);
826 }
Nick Piggin873feea2011-01-07 17:50:06 +1100827 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700829EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400831static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
Al Viro5c47e6d2014-04-29 16:13:18 -0400833 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700834
Miklos Szeredi60942f22014-05-02 15:38:39 -0400835 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400836 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400837 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100838 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400839 parent = lock_parent(dentry);
840
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000842 * The dispose list is isolated and dentries are not accounted
843 * to the LRU here, so we can simply remove it from the list
844 * here regardless of whether it is referenced or not.
845 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400846 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000847
848 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000850 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 */
Al Viro41edf272014-05-01 10:30:00 -0400852 if ((int)dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700853 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400854 if (parent)
855 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 continue;
857 }
Nick Piggin77812a12011-01-07 17:49:48 +1100858
Al Viro64fd72e2014-05-28 09:48:44 -0400859
860 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
861 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
862 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400863 if (parent)
864 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -0400865 if (can_free)
866 dentry_free(dentry);
867 continue;
868 }
869
Al Viroff2fde92014-05-28 13:59:13 -0400870 inode = dentry->d_inode;
871 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400872 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000873 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400874 if (parent)
875 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400876 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000877 }
Al Viroff2fde92014-05-28 13:59:13 -0400878
Al Viroff2fde92014-05-28 13:59:13 -0400879 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -0400880
Al Viro5c47e6d2014-04-29 16:13:18 -0400881 /*
882 * We need to prune ancestors too. This is necessary to prevent
883 * quadratic behavior of shrink_dcache_parent(), but is also
884 * expected to be beneficial in reducing dentry cache
885 * fragmentation.
886 */
887 dentry = parent;
888 while (dentry && !lockref_put_or_lock(&dentry->d_lockref))
889 dentry = dentry_kill(dentry, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400891}
892
Dave Chinnerf6041562013-08-28 10:18:00 +1000893static enum lru_status
894dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
895{
896 struct list_head *freeable = arg;
897 struct dentry *dentry = container_of(item, struct dentry, d_lru);
898
899
900 /*
901 * we are inverting the lru lock/dentry->d_lock here,
902 * so use a trylock. If we fail to get the lock, just skip
903 * it
904 */
905 if (!spin_trylock(&dentry->d_lock))
906 return LRU_SKIP;
907
908 /*
909 * Referenced dentries are still in use. If they have active
910 * counts, just remove them from the LRU. Otherwise give them
911 * another pass through the LRU.
912 */
913 if (dentry->d_lockref.count) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400914 d_lru_isolate(dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +1000915 spin_unlock(&dentry->d_lock);
916 return LRU_REMOVED;
917 }
918
919 if (dentry->d_flags & DCACHE_REFERENCED) {
920 dentry->d_flags &= ~DCACHE_REFERENCED;
921 spin_unlock(&dentry->d_lock);
922
923 /*
924 * The list move itself will be made by the common LRU code. At
925 * this point, we've dropped the dentry->d_lock but keep the
926 * lru lock. This is safe to do, since every list movement is
927 * protected by the lru lock even if both locks are held.
928 *
929 * This is guaranteed by the fact that all LRU management
930 * functions are intermediated by the LRU API calls like
931 * list_lru_add and list_lru_del. List movement in this file
932 * only ever occur through this functions or through callbacks
933 * like this one, that are called from the LRU API.
934 *
935 * The only exceptions to this are functions like
936 * shrink_dentry_list, and code that first checks for the
937 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
938 * operating only with stack provided lists after they are
939 * properly isolated from the main list. It is thus, always a
940 * local access.
941 */
942 return LRU_ROTATE;
943 }
944
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400945 d_lru_shrink_move(dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +1000946 spin_unlock(&dentry->d_lock);
947
948 return LRU_REMOVED;
949}
950
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400951/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000952 * prune_dcache_sb - shrink the dcache
953 * @sb: superblock
Dave Chinnerf6041562013-08-28 10:18:00 +1000954 * @nr_to_scan : number of entries to try to free
Dave Chinner9b17c622013-08-28 10:18:05 +1000955 * @nid: which node to scan for freeable entities
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400956 *
Dave Chinnerf6041562013-08-28 10:18:00 +1000957 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000958 * done when we need more memory an called from the superblock shrinker
959 * function.
960 *
961 * This function may fail to free any resources if all the dentries are in
962 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400963 */
Dave Chinner9b17c622013-08-28 10:18:05 +1000964long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
965 int nid)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400966{
Dave Chinnerf6041562013-08-28 10:18:00 +1000967 LIST_HEAD(dispose);
968 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400969
Dave Chinner9b17c622013-08-28 10:18:05 +1000970 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
971 &dispose, &nr_to_scan);
Dave Chinnerf6041562013-08-28 10:18:00 +1000972 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +1000973 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974}
975
Glauber Costa4e717f52013-08-28 10:18:03 +1000976static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
977 spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000978{
Glauber Costa4e717f52013-08-28 10:18:03 +1000979 struct list_head *freeable = arg;
980 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000981
Glauber Costa4e717f52013-08-28 10:18:03 +1000982 /*
983 * we are inverting the lru lock/dentry->d_lock here,
984 * so use a trylock. If we fail to get the lock, just skip
985 * it
986 */
987 if (!spin_trylock(&dentry->d_lock))
988 return LRU_SKIP;
989
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400990 d_lru_shrink_move(dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +1000991 spin_unlock(&dentry->d_lock);
992
993 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000994}
995
Glauber Costa4e717f52013-08-28 10:18:03 +1000996
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700997/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 * shrink_dcache_sb - shrink dcache for a superblock
999 * @sb: superblock
1000 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001001 * Shrink the dcache for the specified super block. This is used to free
1002 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001004void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005{
Glauber Costa4e717f52013-08-28 10:18:03 +10001006 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001007
Glauber Costa4e717f52013-08-28 10:18:03 +10001008 do {
1009 LIST_HEAD(dispose);
1010
1011 freed = list_lru_walk(&sb->s_dentry_lru,
1012 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1013
1014 this_cpu_sub(nr_dentry_unused, freed);
1015 shrink_dentry_list(&dispose);
1016 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001018EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001021 * enum d_walk_ret - action to talke during tree walk
1022 * @D_WALK_CONTINUE: contrinue walk
1023 * @D_WALK_QUIT: quit walk
1024 * @D_WALK_NORETRY: quit when retry is needed
1025 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001027enum d_walk_ret {
1028 D_WALK_CONTINUE,
1029 D_WALK_QUIT,
1030 D_WALK_NORETRY,
1031 D_WALK_SKIP,
1032};
1033
1034/**
1035 * d_walk - walk the dentry tree
1036 * @parent: start of walk
1037 * @data: data passed to @enter() and @finish()
1038 * @enter: callback when first entering the dentry
1039 * @finish: callback when successfully finished the walk
1040 *
1041 * The @enter() and @finish() callbacks are called with d_lock held.
1042 */
1043static void d_walk(struct dentry *parent, void *data,
1044 enum d_walk_ret (*enter)(void *, struct dentry *),
1045 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046{
Nick Piggin949854d2011-01-07 17:49:37 +11001047 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001049 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001050 enum d_walk_ret ret;
1051 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001052
Nick Piggin58db63d2011-01-07 17:49:39 +11001053again:
Al Viro48f5ec22013-09-09 15:22:25 -04001054 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001055 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001056 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001057
1058 ret = enter(data, this_parent);
1059 switch (ret) {
1060 case D_WALK_CONTINUE:
1061 break;
1062 case D_WALK_QUIT:
1063 case D_WALK_SKIP:
1064 goto out_unlock;
1065 case D_WALK_NORETRY:
1066 retry = false;
1067 break;
1068 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069repeat:
1070 next = this_parent->d_subdirs.next;
1071resume:
1072 while (next != &this_parent->d_subdirs) {
1073 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001074 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001076
1077 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001078
1079 ret = enter(data, dentry);
1080 switch (ret) {
1081 case D_WALK_CONTINUE:
1082 break;
1083 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001084 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001085 goto out_unlock;
1086 case D_WALK_NORETRY:
1087 retry = false;
1088 break;
1089 case D_WALK_SKIP:
1090 spin_unlock(&dentry->d_lock);
1091 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001092 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001093
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001095 spin_unlock(&this_parent->d_lock);
1096 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001098 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 goto repeat;
1100 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001101 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 }
1103 /*
1104 * All done at this level ... ascend and resume the search.
1105 */
1106 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001107 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001108 this_parent = child->d_parent;
1109
1110 rcu_read_lock();
1111 spin_unlock(&child->d_lock);
1112 spin_lock(&this_parent->d_lock);
1113
1114 /*
1115 * might go back up the wrong parent if we have had a rename
1116 * or deletion
1117 */
1118 if (this_parent != child->d_parent ||
1119 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1120 need_seqretry(&rename_lock, seq)) {
1121 spin_unlock(&this_parent->d_lock);
1122 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11001123 goto rename_retry;
Al Viro31dec132013-10-25 17:04:27 -04001124 }
1125 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11001126 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 goto resume;
1128 }
Al Viro48f5ec22013-09-09 15:22:25 -04001129 if (need_seqretry(&rename_lock, seq)) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001130 spin_unlock(&this_parent->d_lock);
1131 goto rename_retry;
1132 }
1133 if (finish)
1134 finish(data);
1135
1136out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001137 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001138 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001139 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001140
1141rename_retry:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001142 if (!retry)
1143 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001144 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001145 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001147
1148/*
1149 * Search for at least 1 mount point in the dentry's subdirs.
1150 * We descend to the next level whenever the d_subdirs
1151 * list is non-empty and continue searching.
1152 */
1153
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001154static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1155{
1156 int *ret = data;
1157 if (d_mountpoint(dentry)) {
1158 *ret = 1;
1159 return D_WALK_QUIT;
1160 }
1161 return D_WALK_CONTINUE;
1162}
1163
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001164/**
1165 * have_submounts - check for mounts over a dentry
1166 * @parent: dentry to check.
1167 *
1168 * Return true if the parent or its subdirectories contain
1169 * a mount point
1170 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001171int have_submounts(struct dentry *parent)
1172{
1173 int ret = 0;
1174
1175 d_walk(parent, &ret, check_mount, NULL);
1176
1177 return ret;
1178}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001179EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
1181/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001182 * Called by mount code to set a mountpoint and check if the mountpoint is
1183 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1184 * subtree can become unreachable).
1185 *
1186 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1187 * this reason take rename_lock and d_lock on dentry and ancestors.
1188 */
1189int d_set_mounted(struct dentry *dentry)
1190{
1191 struct dentry *p;
1192 int ret = -ENOENT;
1193 write_seqlock(&rename_lock);
1194 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1195 /* Need exclusion wrt. check_submounts_and_drop() */
1196 spin_lock(&p->d_lock);
1197 if (unlikely(d_unhashed(p))) {
1198 spin_unlock(&p->d_lock);
1199 goto out;
1200 }
1201 spin_unlock(&p->d_lock);
1202 }
1203 spin_lock(&dentry->d_lock);
1204 if (!d_unlinked(dentry)) {
1205 dentry->d_flags |= DCACHE_MOUNTED;
1206 ret = 0;
1207 }
1208 spin_unlock(&dentry->d_lock);
1209out:
1210 write_sequnlock(&rename_lock);
1211 return ret;
1212}
1213
1214/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001215 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 * and move any unused dentries to the end of the unused
1217 * list for prune_dcache(). We descend to the next level
1218 * whenever the d_subdirs list is non-empty and continue
1219 * searching.
1220 *
1221 * It returns zero iff there are no unused children,
1222 * otherwise it returns the number of children moved to
1223 * the end of the unused list. This may not be the total
1224 * number of unused children, because select_parent can
1225 * drop the lock and return early due to latency
1226 * constraints.
1227 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001228
1229struct select_data {
1230 struct dentry *start;
1231 struct list_head dispose;
1232 int found;
1233};
1234
1235static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001237 struct select_data *data = _data;
1238 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001240 if (data->start == dentry)
1241 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242
Al Virofe915222014-05-03 00:02:25 -04001243 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001244 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001245 } else {
1246 if (dentry->d_flags & DCACHE_LRU_LIST)
1247 d_lru_del(dentry);
1248 if (!dentry->d_lockref.count) {
1249 d_shrink_add(dentry, &data->dispose);
1250 data->found++;
1251 }
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 */
Al Virofe915222014-05-03 00:02:25 -04001258 if (!list_empty(&data->dispose))
1259 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
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 Viro9c8c10e2014-05-02 20:36:10 -04001289static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001290{
Al Viro9c8c10e2014-05-02 20:36:10 -04001291 /* it has busy descendents; complain about those instead */
1292 if (!list_empty(&dentry->d_subdirs))
1293 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001294
Al Viro9c8c10e2014-05-02 20:36:10 -04001295 /* root with refcount 1 is fine */
1296 if (dentry == _data && dentry->d_lockref.count == 1)
1297 return D_WALK_CONTINUE;
1298
1299 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1300 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001301 dentry,
1302 dentry->d_inode ?
1303 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001304 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001305 dentry->d_lockref.count,
1306 dentry->d_sb->s_type->name,
1307 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001308 WARN_ON(1);
1309 return D_WALK_CONTINUE;
1310}
1311
1312static void do_one_tree(struct dentry *dentry)
1313{
1314 shrink_dcache_parent(dentry);
1315 d_walk(dentry, dentry, umount_check, NULL);
1316 d_drop(dentry);
1317 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001318}
1319
1320/*
1321 * destroy the dentries attached to a superblock on unmounting
1322 */
1323void shrink_dcache_for_umount(struct super_block *sb)
1324{
1325 struct dentry *dentry;
1326
Al Viro9c8c10e2014-05-02 20:36:10 -04001327 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001328
1329 dentry = sb->s_root;
1330 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001331 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001332
1333 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001334 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1335 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001336 }
1337}
1338
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001339static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1340{
1341 struct select_data *data = _data;
1342
1343 if (d_mountpoint(dentry)) {
1344 data->found = -EBUSY;
1345 return D_WALK_QUIT;
1346 }
1347
1348 return select_collect(_data, dentry);
1349}
1350
1351static void check_and_drop(void *_data)
1352{
1353 struct select_data *data = _data;
1354
1355 if (d_mountpoint(data->start))
1356 data->found = -EBUSY;
1357 if (!data->found)
1358 __d_drop(data->start);
1359}
1360
1361/**
1362 * check_submounts_and_drop - prune dcache, check for submounts and drop
1363 *
1364 * All done as a single atomic operation relative to has_unlinked_ancestor().
1365 * Returns 0 if successfully unhashed @parent. If there were submounts then
1366 * return -EBUSY.
1367 *
1368 * @dentry: dentry to prune and drop
1369 */
1370int check_submounts_and_drop(struct dentry *dentry)
1371{
1372 int ret = 0;
1373
1374 /* Negative dentries can be dropped without further checks */
1375 if (!dentry->d_inode) {
1376 d_drop(dentry);
1377 goto out;
1378 }
1379
1380 for (;;) {
1381 struct select_data data;
1382
1383 INIT_LIST_HEAD(&data.dispose);
1384 data.start = dentry;
1385 data.found = 0;
1386
1387 d_walk(dentry, &data, check_and_collect, check_and_drop);
1388 ret = data.found;
1389
1390 if (!list_empty(&data.dispose))
1391 shrink_dentry_list(&data.dispose);
1392
1393 if (ret <= 0)
1394 break;
1395
1396 cond_resched();
1397 }
1398
1399out:
1400 return ret;
1401}
1402EXPORT_SYMBOL(check_submounts_and_drop);
1403
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404/**
Al Viroa4464db2011-07-07 15:03:58 -04001405 * __d_alloc - allocate a dcache entry
1406 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 * @name: qstr of the name
1408 *
1409 * Allocates a dentry. It returns %NULL if there is insufficient memory
1410 * available. On a success the dentry is returned. The name passed in is
1411 * copied and the copy passed in may be reused after this call.
1412 */
1413
Al Viroa4464db2011-07-07 15:03:58 -04001414struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415{
1416 struct dentry *dentry;
1417 char *dname;
1418
Mel Gormane12ba742007-10-16 01:25:52 -07001419 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 if (!dentry)
1421 return NULL;
1422
Linus Torvalds6326c712012-05-21 16:14:04 -07001423 /*
1424 * We guarantee that the inline name is always NUL-terminated.
1425 * This way the memcpy() done by the name switching in rename
1426 * will still always have a NUL at the end, even if we might
1427 * be overwriting an internal NUL character
1428 */
1429 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 if (name->len > DNAME_INLINE_LEN-1) {
1431 dname = kmalloc(name->len + 1, GFP_KERNEL);
1432 if (!dname) {
1433 kmem_cache_free(dentry_cache, dentry);
1434 return NULL;
1435 }
1436 } else {
1437 dname = dentry->d_iname;
1438 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
1440 dentry->d_name.len = name->len;
1441 dentry->d_name.hash = name->hash;
1442 memcpy(dname, name->name, name->len);
1443 dname[name->len] = 0;
1444
Linus Torvalds6326c712012-05-21 16:14:04 -07001445 /* Make sure we always see the terminating NUL character */
1446 smp_wmb();
1447 dentry->d_name.name = dname;
1448
Waiman Long98474232013-08-28 18:24:59 -07001449 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001450 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001452 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001454 dentry->d_parent = dentry;
1455 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 dentry->d_op = NULL;
1457 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001458 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 INIT_LIST_HEAD(&dentry->d_lru);
1460 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001461 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001462 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001463 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
Nick Piggin3e880fb2011-01-07 17:49:19 +11001465 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001466
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 return dentry;
1468}
Al Viroa4464db2011-07-07 15:03:58 -04001469
1470/**
1471 * d_alloc - allocate a dcache entry
1472 * @parent: parent of entry to allocate
1473 * @name: qstr of the name
1474 *
1475 * Allocates a dentry. It returns %NULL if there is insufficient memory
1476 * available. On a success the dentry is returned. The name passed in is
1477 * copied and the copy passed in may be reused after this call.
1478 */
1479struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1480{
1481 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1482 if (!dentry)
1483 return NULL;
1484
1485 spin_lock(&parent->d_lock);
1486 /*
1487 * don't need child lock because it is not subject
1488 * to concurrency here
1489 */
1490 __dget_dlock(parent);
1491 dentry->d_parent = parent;
1492 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1493 spin_unlock(&parent->d_lock);
1494
1495 return dentry;
1496}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001497EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001499/**
1500 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1501 * @sb: the superblock
1502 * @name: qstr of the name
1503 *
1504 * For a filesystem that just pins its dentries in memory and never
1505 * performs lookups at all, return an unhashed IS_ROOT dentry.
1506 */
Nick Piggin4b936882011-01-07 17:50:07 +11001507struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1508{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001509 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001510}
1511EXPORT_SYMBOL(d_alloc_pseudo);
1512
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1514{
1515 struct qstr q;
1516
1517 q.name = name;
1518 q.len = strlen(name);
1519 q.hash = full_name_hash(q.name, q.len);
1520 return d_alloc(parent, &q);
1521}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001522EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
Nick Pigginfb045ad2011-01-07 17:49:55 +11001524void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1525{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001526 WARN_ON_ONCE(dentry->d_op);
1527 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001528 DCACHE_OP_COMPARE |
1529 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001530 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001531 DCACHE_OP_DELETE ));
1532 dentry->d_op = op;
1533 if (!op)
1534 return;
1535 if (op->d_hash)
1536 dentry->d_flags |= DCACHE_OP_HASH;
1537 if (op->d_compare)
1538 dentry->d_flags |= DCACHE_OP_COMPARE;
1539 if (op->d_revalidate)
1540 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001541 if (op->d_weak_revalidate)
1542 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001543 if (op->d_delete)
1544 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001545 if (op->d_prune)
1546 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001547
1548}
1549EXPORT_SYMBOL(d_set_d_op);
1550
David Howellsb18825a2013-09-12 19:22:53 +01001551static unsigned d_flags_for_inode(struct inode *inode)
1552{
1553 unsigned add_flags = DCACHE_FILE_TYPE;
1554
1555 if (!inode)
1556 return DCACHE_MISS_TYPE;
1557
1558 if (S_ISDIR(inode->i_mode)) {
1559 add_flags = DCACHE_DIRECTORY_TYPE;
1560 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1561 if (unlikely(!inode->i_op->lookup))
1562 add_flags = DCACHE_AUTODIR_TYPE;
1563 else
1564 inode->i_opflags |= IOP_LOOKUP;
1565 }
1566 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1567 if (unlikely(inode->i_op->follow_link))
1568 add_flags = DCACHE_SYMLINK_TYPE;
1569 else
1570 inode->i_opflags |= IOP_NOFOLLOW;
1571 }
1572
1573 if (unlikely(IS_AUTOMOUNT(inode)))
1574 add_flags |= DCACHE_NEED_AUTOMOUNT;
1575 return add_flags;
1576}
1577
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001578static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1579{
David Howellsb18825a2013-09-12 19:22:53 +01001580 unsigned add_flags = d_flags_for_inode(inode);
1581
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001582 spin_lock(&dentry->d_lock);
Al Viro22213312014-04-19 12:30:58 -04001583 __d_set_type(dentry, add_flags);
David Howellsb18825a2013-09-12 19:22:53 +01001584 if (inode)
Al Virob3d9b7a2012-06-09 13:51:19 -04001585 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001586 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001587 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001588 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001589 fsnotify_d_instantiate(dentry, inode);
1590}
1591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592/**
1593 * d_instantiate - fill in inode information for a dentry
1594 * @entry: dentry to complete
1595 * @inode: inode to attach to this dentry
1596 *
1597 * Fill in inode information in the entry.
1598 *
1599 * This turns negative dentries into productive full members
1600 * of society.
1601 *
1602 * NOTE! This assumes that the inode count has been incremented
1603 * (or otherwise set) by the caller to indicate that it is now
1604 * in use by the dcache.
1605 */
1606
1607void d_instantiate(struct dentry *entry, struct inode * inode)
1608{
Al Virob3d9b7a2012-06-09 13:51:19 -04001609 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001610 if (inode)
1611 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001612 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001613 if (inode)
1614 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 security_d_instantiate(entry, inode);
1616}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001617EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
1619/**
1620 * d_instantiate_unique - instantiate a non-aliased dentry
1621 * @entry: dentry to instantiate
1622 * @inode: inode to attach to this dentry
1623 *
1624 * Fill in inode information in the entry. On success, it returns NULL.
1625 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001626 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 *
1628 * Note that in order to avoid conflicts with rename() etc, the caller
1629 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001630 *
1631 * This also assumes that the inode count has been incremented
1632 * (or otherwise set) by the caller to indicate that it is now
1633 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 */
David Howells770bfad2006-08-22 20:06:07 -04001635static struct dentry *__d_instantiate_unique(struct dentry *entry,
1636 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637{
1638 struct dentry *alias;
1639 int len = entry->d_name.len;
1640 const char *name = entry->d_name.name;
1641 unsigned int hash = entry->d_name.hash;
1642
David Howells770bfad2006-08-22 20:06:07 -04001643 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001644 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001645 return NULL;
1646 }
1647
Sasha Levinb67bfe02013-02-27 17:06:00 -08001648 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001649 /*
1650 * Don't need alias->d_lock here, because aliases with
1651 * d_parent == entry->d_parent are not subject to name or
1652 * parent changes, because the parent inode i_mutex is held.
1653 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001654 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 continue;
1656 if (alias->d_parent != entry->d_parent)
1657 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001658 if (alias->d_name.len != len)
1659 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001660 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001662 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 return alias;
1664 }
David Howells770bfad2006-08-22 20:06:07 -04001665
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001666 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 return NULL;
1668}
David Howells770bfad2006-08-22 20:06:07 -04001669
1670struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1671{
1672 struct dentry *result;
1673
Al Virob3d9b7a2012-06-09 13:51:19 -04001674 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001675
Nick Piggin873feea2011-01-07 17:50:06 +11001676 if (inode)
1677 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001678 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001679 if (inode)
1680 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001681
1682 if (!result) {
1683 security_d_instantiate(entry, inode);
1684 return NULL;
1685 }
1686
1687 BUG_ON(!d_unhashed(result));
1688 iput(inode);
1689 return result;
1690}
1691
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692EXPORT_SYMBOL(d_instantiate_unique);
1693
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001694/**
1695 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1696 * @entry: dentry to complete
1697 * @inode: inode to attach to this dentry
1698 *
1699 * Fill in inode information in the entry. If a directory alias is found, then
1700 * return an error (and drop inode). Together with d_materialise_unique() this
1701 * guarantees that a directory inode may never have more than one alias.
1702 */
1703int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1704{
1705 BUG_ON(!hlist_unhashed(&entry->d_alias));
1706
1707 spin_lock(&inode->i_lock);
1708 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1709 spin_unlock(&inode->i_lock);
1710 iput(inode);
1711 return -EBUSY;
1712 }
1713 __d_instantiate(entry, inode);
1714 spin_unlock(&inode->i_lock);
1715 security_d_instantiate(entry, inode);
1716
1717 return 0;
1718}
1719EXPORT_SYMBOL(d_instantiate_no_diralias);
1720
Al Viroadc0e912012-01-08 16:49:21 -05001721struct dentry *d_make_root(struct inode *root_inode)
1722{
1723 struct dentry *res = NULL;
1724
1725 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001726 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001727
1728 res = __d_alloc(root_inode->i_sb, &name);
1729 if (res)
1730 d_instantiate(res, root_inode);
1731 else
1732 iput(root_inode);
1733 }
1734 return res;
1735}
1736EXPORT_SYMBOL(d_make_root);
1737
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001738static struct dentry * __d_find_any_alias(struct inode *inode)
1739{
1740 struct dentry *alias;
1741
Al Virob3d9b7a2012-06-09 13:51:19 -04001742 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001743 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001744 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001745 __dget(alias);
1746 return alias;
1747}
1748
Sage Weil46f72b32012-01-10 09:04:37 -08001749/**
1750 * d_find_any_alias - find any alias for a given inode
1751 * @inode: inode to find an alias for
1752 *
1753 * If any aliases exist for the given inode, take and return a
1754 * reference for one of them. If no aliases exist, return %NULL.
1755 */
1756struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001757{
1758 struct dentry *de;
1759
1760 spin_lock(&inode->i_lock);
1761 de = __d_find_any_alias(inode);
1762 spin_unlock(&inode->i_lock);
1763 return de;
1764}
Sage Weil46f72b32012-01-10 09:04:37 -08001765EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001766
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001768 * d_obtain_alias - find or allocate a dentry for a given inode
1769 * @inode: inode to allocate the dentry for
1770 *
1771 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1772 * similar open by handle operations. The returned dentry may be anonymous,
1773 * or may have a full name (if the inode was already in the cache).
1774 *
1775 * When called on a directory inode, we must ensure that the inode only ever
1776 * has one dentry. If a dentry is found, that is returned instead of
1777 * allocating a new one.
1778 *
1779 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001780 * to the dentry. In case of an error the reference on the inode is released.
1781 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1782 * be passed in and will be the error will be propagate to the return value,
1783 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001784 */
1785struct dentry *d_obtain_alias(struct inode *inode)
1786{
NeilBrownb911a6b2012-11-08 16:09:37 -08001787 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001788 struct dentry *tmp;
1789 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001790 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001791
1792 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001793 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001794 if (IS_ERR(inode))
1795 return ERR_CAST(inode);
1796
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001797 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001798 if (res)
1799 goto out_iput;
1800
Al Viroa4464db2011-07-07 15:03:58 -04001801 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001802 if (!tmp) {
1803 res = ERR_PTR(-ENOMEM);
1804 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001805 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001806
Nick Piggin873feea2011-01-07 17:50:06 +11001807 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001808 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001809 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001810 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001811 dput(tmp);
1812 goto out_iput;
1813 }
1814
1815 /* attach a disconnected dentry */
David Howellsb18825a2013-09-12 19:22:53 +01001816 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1817
Christoph Hellwig9308a612008-08-11 15:49:12 +02001818 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001819 tmp->d_inode = inode;
David Howellsb18825a2013-09-12 19:22:53 +01001820 tmp->d_flags |= add_flags;
Al Virob3d9b7a2012-06-09 13:51:19 -04001821 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001822 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001823 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001824 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001825 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001826 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001827 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001828
Christoph Hellwig9308a612008-08-11 15:49:12 +02001829 return tmp;
1830
1831 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001832 if (res && !IS_ERR(res))
1833 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001834 iput(inode);
1835 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001836}
Benny Halevyadc48722009-02-27 14:02:59 -08001837EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838
1839/**
1840 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1841 * @inode: the inode which may have a disconnected dentry
1842 * @dentry: a negative dentry which we want to point to the inode.
1843 *
1844 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1845 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1846 * and return it, else simply d_add the inode to the dentry and return NULL.
1847 *
1848 * This is needed in the lookup routine of any filesystem that is exportable
1849 * (via knfsd) so that we can build dcache paths to directories effectively.
1850 *
1851 * If a dentry was found and moved, then it is returned. Otherwise NULL
1852 * is returned. This matches the expected return value of ->lookup.
1853 *
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001854 * Cluster filesystems may call this function with a negative, hashed dentry.
1855 * In that case, we know that the inode will be a regular file, and also this
1856 * will only occur during atomic_open. So we need to check for the dentry
1857 * being already hashed only in the final case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 */
1859struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1860{
1861 struct dentry *new = NULL;
1862
Al Viroa9049372011-07-08 21:20:11 -04001863 if (IS_ERR(inode))
1864 return ERR_CAST(inode);
1865
NeilBrown21c0d8f2006-10-04 02:16:16 -07001866 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001867 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001868 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001870 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001871 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 d_move(new, dentry);
1874 iput(inode);
1875 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001876 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001877 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001878 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 security_d_instantiate(dentry, inode);
1880 d_rehash(dentry);
1881 }
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001882 } else {
1883 d_instantiate(dentry, inode);
1884 if (d_unhashed(dentry))
1885 d_rehash(dentry);
1886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 return new;
1888}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001889EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
Barry Naujok94035402008-05-21 16:50:46 +10001891/**
1892 * d_add_ci - lookup or allocate new dentry with case-exact name
1893 * @inode: the inode case-insensitive lookup has found
1894 * @dentry: the negative dentry that was passed to the parent's lookup func
1895 * @name: the case-exact name to be associated with the returned dentry
1896 *
1897 * This is to avoid filling the dcache with case-insensitive names to the
1898 * same inode, only the actual correct case is stored in the dcache for
1899 * case-insensitive filesystems.
1900 *
1901 * For a case-insensitive lookup match and if the the case-exact dentry
1902 * already exists in in the dcache, use it and return it.
1903 *
1904 * If no entry exists with the exact case name, allocate new dentry with
1905 * the exact case, and return the spliced entry.
1906 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001907struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001908 struct qstr *name)
1909{
Barry Naujok94035402008-05-21 16:50:46 +10001910 struct dentry *found;
1911 struct dentry *new;
1912
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001913 /*
1914 * First check if a dentry matching the name already exists,
1915 * if not go ahead and create it now.
1916 */
Barry Naujok94035402008-05-21 16:50:46 +10001917 found = d_hash_and_lookup(dentry->d_parent, name);
Al Viro4f522a22013-02-11 23:20:37 -05001918 if (unlikely(IS_ERR(found)))
1919 goto err_out;
Barry Naujok94035402008-05-21 16:50:46 +10001920 if (!found) {
1921 new = d_alloc(dentry->d_parent, name);
1922 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05001923 found = ERR_PTR(-ENOMEM);
Barry Naujok94035402008-05-21 16:50:46 +10001924 goto err_out;
1925 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001926
Barry Naujok94035402008-05-21 16:50:46 +10001927 found = d_splice_alias(inode, new);
1928 if (found) {
1929 dput(new);
1930 return found;
1931 }
1932 return new;
1933 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001934
1935 /*
1936 * If a matching dentry exists, and it's not negative use it.
1937 *
1938 * Decrement the reference count to balance the iget() done
1939 * earlier on.
1940 */
Barry Naujok94035402008-05-21 16:50:46 +10001941 if (found->d_inode) {
1942 if (unlikely(found->d_inode != inode)) {
1943 /* This can't happen because bad inodes are unhashed. */
1944 BUG_ON(!is_bad_inode(inode));
1945 BUG_ON(!is_bad_inode(found->d_inode));
1946 }
Barry Naujok94035402008-05-21 16:50:46 +10001947 iput(inode);
1948 return found;
1949 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001950
Barry Naujok94035402008-05-21 16:50:46 +10001951 /*
1952 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001953 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001954 */
Al Viro4513d892011-07-17 10:52:14 -04001955 new = d_splice_alias(inode, found);
1956 if (new) {
1957 dput(found);
1958 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10001959 }
Al Viro4513d892011-07-17 10:52:14 -04001960 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001961
1962err_out:
1963 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05001964 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001965}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001966EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001968/*
1969 * Do the slow-case of the dentry name compare.
1970 *
1971 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07001972 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001973 * filesystem can rely on them, and can use the 'name' and
1974 * 'len' information without worrying about walking off the
1975 * end of memory etc.
1976 *
1977 * Thus the read_seqcount_retry() and the "duplicate" info
1978 * in arguments (the low-level filesystem should not look
1979 * at the dentry inode or name contents directly, since
1980 * rename can change them while we're in RCU mode).
1981 */
1982enum slow_d_compare {
1983 D_COMP_OK,
1984 D_COMP_NOMATCH,
1985 D_COMP_SEQRETRY,
1986};
1987
1988static noinline enum slow_d_compare slow_dentry_cmp(
1989 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001990 struct dentry *dentry,
1991 unsigned int seq,
1992 const struct qstr *name)
1993{
1994 int tlen = dentry->d_name.len;
1995 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001996
1997 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1998 cpu_relax();
1999 return D_COMP_SEQRETRY;
2000 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002001 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002002 return D_COMP_NOMATCH;
2003 return D_COMP_OK;
2004}
2005
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002007 * __d_lookup_rcu - search for a dentry (racy, store-free)
2008 * @parent: parent dentry
2009 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002010 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002011 * Returns: dentry, or NULL
2012 *
2013 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2014 * resolution (store-free path walking) design described in
2015 * Documentation/filesystems/path-lookup.txt.
2016 *
2017 * This is not to be used outside core vfs.
2018 *
2019 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2020 * held, and rcu_read_lock held. The returned dentry must not be stored into
2021 * without taking d_lock and checking d_seq sequence count against @seq
2022 * returned here.
2023 *
Linus Torvalds15570082013-09-02 11:38:06 -07002024 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002025 * function.
2026 *
2027 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2028 * the returned dentry, so long as its parent's seqlock is checked after the
2029 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2030 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002031 *
2032 * NOTE! The caller *has* to check the resulting dentry against the sequence
2033 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002034 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002035struct dentry *__d_lookup_rcu(const struct dentry *parent,
2036 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002037 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002038{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002039 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002040 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002041 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002042 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002043 struct dentry *dentry;
2044
2045 /*
2046 * Note: There is significant duplication with __d_lookup_rcu which is
2047 * required to prevent single threaded performance regressions
2048 * especially on architectures where smp_rmb (in seqcounts) are costly.
2049 * Keep the two functions in sync.
2050 */
2051
2052 /*
2053 * The hash list is protected using RCU.
2054 *
2055 * Carefully use d_seq when comparing a candidate dentry, to avoid
2056 * races with d_move().
2057 *
2058 * It is possible that concurrent renames can mess up our list
2059 * walk here and result in missing our dentry, resulting in the
2060 * false-negative result. d_lookup() protects against concurrent
2061 * renames using rename_lock seqlock.
2062 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002063 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002064 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002065 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002066 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002067
Nick Piggin31e6b012011-01-07 17:49:52 +11002068seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002069 /*
2070 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002071 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002072 *
2073 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002074 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002075 *
2076 * NOTE! We do a "raw" seqcount_begin here. That means that
2077 * we don't wait for the sequence count to stabilize if it
2078 * is in the middle of a sequence change. If we do the slow
2079 * dentry compare, we will do seqretries until it is stable,
2080 * and if we end up with a successful lookup, we actually
2081 * want to exit RCU lookup anyway.
2082 */
2083 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002084 if (dentry->d_parent != parent)
2085 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002086 if (d_unhashed(dentry))
2087 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002088
2089 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002090 if (dentry->d_name.hash != hashlen_hash(hashlen))
2091 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002092 *seqp = seq;
2093 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002094 case D_COMP_OK:
2095 return dentry;
2096 case D_COMP_NOMATCH:
2097 continue;
2098 default:
2099 goto seqretry;
2100 }
2101 }
2102
Linus Torvalds26fe5752012-05-10 13:14:12 -07002103 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002104 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002105 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002106 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002107 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002108 }
2109 return NULL;
2110}
2111
2112/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 * d_lookup - search for a dentry
2114 * @parent: parent dentry
2115 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002116 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002118 * d_lookup searches the children of the parent dentry for the name in
2119 * question. If the dentry is found its reference count is incremented and the
2120 * dentry is returned. The caller must use dput to free the entry when it has
2121 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 */
Al Viroda2d8452013-01-24 18:29:34 -05002123struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124{
Nick Piggin31e6b012011-01-07 17:49:52 +11002125 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002126 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127
2128 do {
2129 seq = read_seqbegin(&rename_lock);
2130 dentry = __d_lookup(parent, name);
2131 if (dentry)
2132 break;
2133 } while (read_seqretry(&rename_lock, seq));
2134 return dentry;
2135}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002136EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137
Nick Piggin31e6b012011-01-07 17:49:52 +11002138/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002139 * __d_lookup - search for a dentry (racy)
2140 * @parent: parent dentry
2141 * @name: qstr of name we wish to find
2142 * Returns: dentry, or NULL
2143 *
2144 * __d_lookup is like d_lookup, however it may (rarely) return a
2145 * false-negative result due to unrelated rename activity.
2146 *
2147 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2148 * however it must be used carefully, eg. with a following d_lookup in
2149 * the case of failure.
2150 *
2151 * __d_lookup callers must be commented.
2152 */
Al Viroa713ca22013-01-24 18:27:00 -05002153struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154{
2155 unsigned int len = name->len;
2156 unsigned int hash = name->hash;
2157 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002158 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002159 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002160 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002161 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162
Nick Pigginb04f7842010-08-18 04:37:34 +10002163 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002164 * Note: There is significant duplication with __d_lookup_rcu which is
2165 * required to prevent single threaded performance regressions
2166 * especially on architectures where smp_rmb (in seqcounts) are costly.
2167 * Keep the two functions in sync.
2168 */
2169
2170 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002171 * The hash list is protected using RCU.
2172 *
2173 * Take d_lock when comparing a candidate dentry, to avoid races
2174 * with d_move().
2175 *
2176 * It is possible that concurrent renames can mess up our list
2177 * walk here and result in missing our dentry, resulting in the
2178 * false-negative result. d_lookup() protects against concurrent
2179 * renames using rename_lock seqlock.
2180 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002181 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002182 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 rcu_read_lock();
2184
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002185 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 if (dentry->d_name.hash != hash)
2188 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
2190 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 if (dentry->d_parent != parent)
2192 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002193 if (d_unhashed(dentry))
2194 goto next;
2195
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 /*
2197 * It is safe to compare names since d_move() cannot
2198 * change the qstr (protected by d_lock).
2199 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002200 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002201 int tlen = dentry->d_name.len;
2202 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002203 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 goto next;
2205 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002206 if (dentry->d_name.len != len)
2207 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002208 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 goto next;
2210 }
2211
Waiman Long98474232013-08-28 18:24:59 -07002212 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002213 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 spin_unlock(&dentry->d_lock);
2215 break;
2216next:
2217 spin_unlock(&dentry->d_lock);
2218 }
2219 rcu_read_unlock();
2220
2221 return found;
2222}
2223
2224/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002225 * d_hash_and_lookup - hash the qstr then search for a dentry
2226 * @dir: Directory to search in
2227 * @name: qstr of name we wish to find
2228 *
Al Viro4f522a22013-02-11 23:20:37 -05002229 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002230 */
2231struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2232{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002233 /*
2234 * Check for a fs-specific hash function. Note that we must
2235 * calculate the standard hash first, as the d_op->d_hash()
2236 * routine may choose to leave the hash value unchanged.
2237 */
2238 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002239 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002240 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002241 if (unlikely(err < 0))
2242 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002243 }
Al Viro4f522a22013-02-11 23:20:37 -05002244 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002245}
Al Viro4f522a22013-02-11 23:20:37 -05002246EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002247
2248/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002249 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002251 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 *
2253 * An insecure source has sent us a dentry, here we verify it and dget() it.
2254 * This is used by ncpfs in its readdir implementation.
2255 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002256 *
2257 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002259int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260{
Nick Piggin786a5e12011-01-07 17:49:16 +11002261 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002262
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002263 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002264 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2265 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002266 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002267 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002268 spin_unlock(&dentry->d_lock);
2269 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 return 1;
2271 }
2272 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002273 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002274
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 return 0;
2276}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002277EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278
2279/*
2280 * When a file is deleted, we have two options:
2281 * - turn this dentry into a negative dentry
2282 * - unhash this dentry and free it.
2283 *
2284 * Usually, we want to just turn this into
2285 * a negative dentry, but if anybody else is
2286 * currently using the dentry or the inode
2287 * we can't do that and we fall back on removing
2288 * it from the hash queues and waiting for
2289 * it to be deleted later when it has no users
2290 */
2291
2292/**
2293 * d_delete - delete a dentry
2294 * @dentry: The dentry to delete
2295 *
2296 * Turn the dentry into a negative dentry if possible, otherwise
2297 * remove it from the hash queues so it can be deleted later
2298 */
2299
2300void d_delete(struct dentry * dentry)
2301{
Nick Piggin873feea2011-01-07 17:50:06 +11002302 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002303 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 /*
2305 * Are we the only user?
2306 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002307again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002309 inode = dentry->d_inode;
2310 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002311 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002312 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002313 spin_unlock(&dentry->d_lock);
2314 cpu_relax();
2315 goto again;
2316 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002317 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002318 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002319 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 return;
2321 }
2322
2323 if (!d_unhashed(dentry))
2324 __d_drop(dentry);
2325
2326 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002327
2328 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002330EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002332static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002334 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002335 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002336 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002337 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002338 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339}
2340
David Howells770bfad2006-08-22 20:06:07 -04002341static void _d_rehash(struct dentry * entry)
2342{
2343 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2344}
2345
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346/**
2347 * d_rehash - add an entry back to the hash
2348 * @entry: dentry to add to the hash
2349 *
2350 * Adds a dentry to the hash according to its name.
2351 */
2352
2353void d_rehash(struct dentry * entry)
2354{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002356 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002359EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002361/**
2362 * dentry_update_name_case - update case insensitive dentry with a new name
2363 * @dentry: dentry to be updated
2364 * @name: new name
2365 *
2366 * Update a case insensitive dentry with new case of name.
2367 *
2368 * dentry must have been returned by d_lookup with name @name. Old and new
2369 * name lengths must match (ie. no d_compare which allows mismatched name
2370 * lengths).
2371 *
2372 * Parent inode i_mutex must be held over d_lookup and into this call (to
2373 * keep renames and concurrent inserts, and readdir(2) away).
2374 */
2375void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2376{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002377 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002378 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2379
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002380 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002381 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002382 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002383 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002384 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002385}
2386EXPORT_SYMBOL(dentry_update_name_case);
2387
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388static void switch_names(struct dentry *dentry, struct dentry *target)
2389{
2390 if (dname_external(target)) {
2391 if (dname_external(dentry)) {
2392 /*
2393 * Both external: swap the pointers
2394 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002395 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 } else {
2397 /*
2398 * dentry:internal, target:external. Steal target's
2399 * storage and make target internal.
2400 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002401 memcpy(target->d_iname, dentry->d_name.name,
2402 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 dentry->d_name.name = target->d_name.name;
2404 target->d_name.name = target->d_iname;
2405 }
2406 } else {
2407 if (dname_external(dentry)) {
2408 /*
2409 * dentry:external, target:internal. Give dentry's
2410 * storage to target and make dentry internal
2411 */
2412 memcpy(dentry->d_iname, target->d_name.name,
2413 target->d_name.len + 1);
2414 target->d_name.name = dentry->d_name.name;
2415 dentry->d_name.name = dentry->d_iname;
2416 } else {
2417 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002418 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002420 unsigned int i;
2421 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2422 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2423 swap(((long *) &dentry->d_iname)[i],
2424 ((long *) &target->d_iname)[i]);
2425 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 }
2427 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002428 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429}
2430
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002431static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2432{
2433 /*
2434 * XXXX: do we really need to take target->d_lock?
2435 */
2436 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2437 spin_lock(&target->d_parent->d_lock);
2438 else {
2439 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2440 spin_lock(&dentry->d_parent->d_lock);
2441 spin_lock_nested(&target->d_parent->d_lock,
2442 DENTRY_D_LOCK_NESTED);
2443 } else {
2444 spin_lock(&target->d_parent->d_lock);
2445 spin_lock_nested(&dentry->d_parent->d_lock,
2446 DENTRY_D_LOCK_NESTED);
2447 }
2448 }
2449 if (target < dentry) {
2450 spin_lock_nested(&target->d_lock, 2);
2451 spin_lock_nested(&dentry->d_lock, 3);
2452 } else {
2453 spin_lock_nested(&dentry->d_lock, 2);
2454 spin_lock_nested(&target->d_lock, 3);
2455 }
2456}
2457
2458static void dentry_unlock_parents_for_move(struct dentry *dentry,
2459 struct dentry *target)
2460{
2461 if (target->d_parent != dentry->d_parent)
2462 spin_unlock(&dentry->d_parent->d_lock);
2463 if (target->d_parent != target)
2464 spin_unlock(&target->d_parent->d_lock);
2465}
2466
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002468 * When switching names, the actual string doesn't strictly have to
2469 * be preserved in the target - because we're dropping the target
2470 * anyway. As such, we can just do a simple memcpy() to copy over
2471 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002473 * Note that we have to be a lot more careful about getting the hash
2474 * switched - we have to switch the hash value properly even if it
2475 * then no longer matches the actual (corrupted) string of the target.
2476 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002478/*
Al Viro18367502011-07-12 21:42:24 -04002479 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 * @dentry: entry to move
2481 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002482 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 *
2484 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002485 * dcache entries should not be moved in this way. Caller must hold
2486 * rename_lock, the i_mutex of the source and target directories,
2487 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002489static void __d_move(struct dentry *dentry, struct dentry *target,
2490 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 if (!dentry->d_inode)
2493 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2494
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002495 BUG_ON(d_ancestor(dentry, target));
2496 BUG_ON(d_ancestor(target, dentry));
2497
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002498 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499
Nick Piggin31e6b012011-01-07 17:49:52 +11002500 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002501 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002502
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002503 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2504
2505 /*
2506 * Move the dentry to the target hash queue. Don't bother checking
2507 * for the same hash queue because of how unlikely it is.
2508 */
2509 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002510 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511
Miklos Szeredida1ce062014-04-01 17:08:43 +02002512 /*
2513 * Unhash the target (d_delete() is not usable here). If exchanging
2514 * the two dentries, then rehash onto the other's hash queue.
2515 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002517 if (exchange) {
2518 __d_rehash(target,
2519 d_hash(dentry->d_parent, dentry->d_name.hash));
2520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521
Eric Dumazet5160ee62006-01-08 01:03:32 -08002522 list_del(&dentry->d_u.d_child);
2523 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524
2525 /* Switch the names.. */
2526 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002527 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528
2529 /* ... and switch the parents */
2530 if (IS_ROOT(dentry)) {
2531 dentry->d_parent = target->d_parent;
2532 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002533 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002535 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536
2537 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002538 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 }
2540
Eric Dumazet5160ee62006-01-08 01:03:32 -08002541 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002542
Nick Piggin31e6b012011-01-07 17:49:52 +11002543 write_seqcount_end(&target->d_seq);
2544 write_seqcount_end(&dentry->d_seq);
2545
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002546 dentry_unlock_parents_for_move(dentry, target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002547 if (exchange)
2548 fsnotify_d_move(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002550 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002552}
2553
2554/*
2555 * d_move - move a dentry
2556 * @dentry: entry to move
2557 * @target: new dentry
2558 *
2559 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002560 * dcache entries should not be moved in this way. See the locking
2561 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002562 */
2563void d_move(struct dentry *dentry, struct dentry *target)
2564{
2565 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002566 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002568}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002569EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
Miklos Szeredida1ce062014-04-01 17:08:43 +02002571/*
2572 * d_exchange - exchange two dentries
2573 * @dentry1: first dentry
2574 * @dentry2: second dentry
2575 */
2576void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2577{
2578 write_seqlock(&rename_lock);
2579
2580 WARN_ON(!dentry1->d_inode);
2581 WARN_ON(!dentry2->d_inode);
2582 WARN_ON(IS_ROOT(dentry1));
2583 WARN_ON(IS_ROOT(dentry2));
2584
2585 __d_move(dentry1, dentry2, true);
2586
2587 write_sequnlock(&rename_lock);
2588}
2589
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002590/**
2591 * d_ancestor - search for an ancestor
2592 * @p1: ancestor dentry
2593 * @p2: child dentry
2594 *
2595 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2596 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002597 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002598struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002599{
2600 struct dentry *p;
2601
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002602 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002603 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002604 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002605 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002606 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002607}
2608
2609/*
2610 * This helper attempts to cope with remotely renamed directories
2611 *
2612 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002613 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002614 *
2615 * Note: If ever the locking in lock_rename() changes, then please
2616 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002617 */
Nick Piggin873feea2011-01-07 17:50:06 +11002618static struct dentry *__d_unalias(struct inode *inode,
2619 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002620{
2621 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002622 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002623
2624 /* If alias and dentry share a parent, then no extra locks required */
2625 if (alias->d_parent == dentry->d_parent)
2626 goto out_unalias;
2627
Trond Myklebust9eaef272006-10-21 10:24:20 -07002628 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002629 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2630 goto out_err;
2631 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2632 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2633 goto out_err;
2634 m2 = &alias->d_parent->d_inode->i_mutex;
2635out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002636 if (likely(!d_mountpoint(alias))) {
Miklos Szeredida1ce062014-04-01 17:08:43 +02002637 __d_move(alias, dentry, false);
Al Viroee3efa92012-06-08 15:59:33 -04002638 ret = alias;
2639 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002640out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002641 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002642 if (m2)
2643 mutex_unlock(m2);
2644 if (m1)
2645 mutex_unlock(m1);
2646 return ret;
2647}
2648
2649/*
David Howells770bfad2006-08-22 20:06:07 -04002650 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2651 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002652 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002653 */
2654static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2655{
Al Viro740da422013-01-30 10:13:38 -05002656 struct dentry *dparent;
David Howells770bfad2006-08-22 20:06:07 -04002657
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002658 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002659
Nick Piggin31e6b012011-01-07 17:49:52 +11002660 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002661 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002662
David Howells770bfad2006-08-22 20:06:07 -04002663 dparent = dentry->d_parent;
David Howells770bfad2006-08-22 20:06:07 -04002664
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002665 switch_names(dentry, anon);
2666 swap(dentry->d_name.hash, anon->d_name.hash);
2667
Al Viro740da422013-01-30 10:13:38 -05002668 dentry->d_parent = dentry;
2669 list_del_init(&dentry->d_u.d_child);
2670 anon->d_parent = dparent;
Wei Yongjun9ed53b12013-03-12 00:10:50 +08002671 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
David Howells770bfad2006-08-22 20:06:07 -04002672
Nick Piggin31e6b012011-01-07 17:49:52 +11002673 write_seqcount_end(&dentry->d_seq);
2674 write_seqcount_end(&anon->d_seq);
2675
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002676 dentry_unlock_parents_for_move(anon, dentry);
2677 spin_unlock(&dentry->d_lock);
2678
2679 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002680}
2681
2682/**
2683 * d_materialise_unique - introduce an inode into the tree
2684 * @dentry: candidate dentry
2685 * @inode: inode to bind to the dentry, to which aliases may be attached
2686 *
2687 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002688 * root directory alias in its place if there is one. Caller must hold the
2689 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002690 */
2691struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2692{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002693 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002694
2695 BUG_ON(!d_unhashed(dentry));
2696
David Howells770bfad2006-08-22 20:06:07 -04002697 if (!inode) {
2698 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002699 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002700 d_rehash(actual);
2701 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002702 }
2703
Nick Piggin873feea2011-01-07 17:50:06 +11002704 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002705
Trond Myklebust9eaef272006-10-21 10:24:20 -07002706 if (S_ISDIR(inode->i_mode)) {
2707 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002708
Trond Myklebust9eaef272006-10-21 10:24:20 -07002709 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002710 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002711 if (alias) {
2712 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002713 write_seqlock(&rename_lock);
2714
2715 if (d_ancestor(alias, dentry)) {
2716 /* Check for loops */
2717 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002718 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002719 } else if (IS_ROOT(alias)) {
2720 /* Is this an anonymous mountpoint that we
2721 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002722 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002723 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002724 __d_drop(alias);
2725 goto found;
Al Viro18367502011-07-12 21:42:24 -04002726 } else {
2727 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002728 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002729 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002730 }
Al Viro18367502011-07-12 21:42:24 -04002731 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002732 if (IS_ERR(actual)) {
2733 if (PTR_ERR(actual) == -ELOOP)
2734 pr_warn_ratelimited(
2735 "VFS: Lookup of '%s' in %s %s"
2736 " would have caused loop\n",
2737 dentry->d_name.name,
2738 inode->i_sb->s_type->name,
2739 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002740 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002741 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002742 goto out_nolock;
2743 }
David Howells770bfad2006-08-22 20:06:07 -04002744 }
2745
2746 /* Add a unique reference */
2747 actual = __d_instantiate_unique(dentry, inode);
2748 if (!actual)
2749 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002750 else
2751 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002752
David Howells770bfad2006-08-22 20:06:07 -04002753 spin_lock(&actual->d_lock);
2754found:
2755 _d_rehash(actual);
2756 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002757 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002758out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002759 if (actual == dentry) {
2760 security_d_instantiate(dentry, inode);
2761 return NULL;
2762 }
2763
2764 iput(inode);
2765 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002766}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002767EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002768
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002769static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002770{
2771 *buflen -= namelen;
2772 if (*buflen < 0)
2773 return -ENAMETOOLONG;
2774 *buffer -= namelen;
2775 memcpy(*buffer, str, namelen);
2776 return 0;
2777}
2778
Waiman Long232d2d62013-09-09 12:18:13 -04002779/**
2780 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002781 * @buffer: buffer pointer
2782 * @buflen: allocated length of the buffer
2783 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002784 *
2785 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2786 * make sure that either the old or the new name pointer and length are
2787 * fetched. However, there may be mismatch between length and pointer.
2788 * The length cannot be trusted, we need to copy it byte-by-byte until
2789 * the length is reached or a null byte is found. It also prepends "/" at
2790 * the beginning of the name. The sequence number check at the caller will
2791 * retry it again when a d_move() does happen. So any garbage in the buffer
2792 * due to mismatched pointer and length will be discarded.
2793 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002794static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2795{
Waiman Long232d2d62013-09-09 12:18:13 -04002796 const char *dname = ACCESS_ONCE(name->name);
2797 u32 dlen = ACCESS_ONCE(name->len);
2798 char *p;
2799
Waiman Long232d2d62013-09-09 12:18:13 -04002800 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04002801 if (*buflen < 0)
2802 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04002803 p = *buffer -= dlen + 1;
2804 *p++ = '/';
2805 while (dlen--) {
2806 char c = *dname++;
2807 if (!c)
2808 break;
2809 *p++ = c;
2810 }
2811 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002812}
2813
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002815 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002816 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002817 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002818 * @buffer: pointer to the end of the buffer
2819 * @buflen: pointer to buffer length
2820 *
Waiman Long18129972013-09-12 10:55:35 -04002821 * The function will first try to write out the pathname without taking any
2822 * lock other than the RCU read lock to make sure that dentries won't go away.
2823 * It only checks the sequence number of the global rename_lock as any change
2824 * in the dentry's d_seq will be preceded by changes in the rename_lock
2825 * sequence number. If the sequence number had been changed, it will restart
2826 * the whole pathname back-tracing sequence again by taking the rename_lock.
2827 * In this case, there is no need to take the RCU read lock as the recursive
2828 * parent pointer references will keep the dentry chain alive as long as no
2829 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002830 */
Al Viro02125a82011-12-05 08:43:34 -05002831static int prepend_path(const struct path *path,
2832 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002833 char **buffer, int *buflen)
2834{
Al Viroede4ceb2013-11-13 07:45:40 -05002835 struct dentry *dentry;
2836 struct vfsmount *vfsmnt;
2837 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002838 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002839 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002840 char *bptr;
2841 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002842
Al Viro48f5ec22013-09-09 15:22:25 -04002843 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002844restart_mnt:
2845 read_seqbegin_or_lock(&mount_lock, &m_seq);
2846 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002847 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002848restart:
2849 bptr = *buffer;
2850 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002851 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002852 dentry = path->dentry;
2853 vfsmnt = path->mnt;
2854 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002855 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002856 while (dentry != root->dentry || vfsmnt != root->mnt) {
2857 struct dentry * parent;
2858
2859 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002860 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002861 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002862 if (mnt != parent) {
2863 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2864 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002865 vfsmnt = &mnt->mnt;
2866 continue;
2867 }
2868 /*
2869 * Filesystems needing to implement special "root names"
2870 * should do so with ->d_dname()
2871 */
2872 if (IS_ROOT(dentry) &&
2873 (dentry->d_name.len != 1 ||
2874 dentry->d_name.name[0] != '/')) {
2875 WARN(1, "Root dentry has weird name <%.*s>\n",
2876 (int) dentry->d_name.len,
2877 dentry->d_name.name);
2878 }
2879 if (!error)
2880 error = is_mounted(vfsmnt) ? 1 : 2;
2881 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002882 }
2883 parent = dentry->d_parent;
2884 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002885 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002886 if (error)
2887 break;
2888
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002889 dentry = parent;
2890 }
Al Viro48f5ec22013-09-09 15:22:25 -04002891 if (!(seq & 1))
2892 rcu_read_unlock();
2893 if (need_seqretry(&rename_lock, seq)) {
2894 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002895 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002896 }
Al Viro48f5ec22013-09-09 15:22:25 -04002897 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002898
2899 if (!(m_seq & 1))
2900 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002901 if (need_seqretry(&mount_lock, m_seq)) {
2902 m_seq = 1;
2903 goto restart_mnt;
2904 }
2905 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002906
Waiman Long232d2d62013-09-09 12:18:13 -04002907 if (error >= 0 && bptr == *buffer) {
2908 if (--blen < 0)
2909 error = -ENAMETOOLONG;
2910 else
2911 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002912 }
Waiman Long232d2d62013-09-09 12:18:13 -04002913 *buffer = bptr;
2914 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002915 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002916}
2917
2918/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002919 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002920 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002921 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002922 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 * @buflen: buffer length
2924 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002925 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002927 * Returns a pointer into the buffer or an error code if the
2928 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002929 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002930 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002931 *
Al Viro02125a82011-12-05 08:43:34 -05002932 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933 */
Al Viro02125a82011-12-05 08:43:34 -05002934char *__d_path(const struct path *path,
2935 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002936 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002938 char *res = buf + buflen;
2939 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002941 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002942 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002943
Al Viro02125a82011-12-05 08:43:34 -05002944 if (error < 0)
2945 return ERR_PTR(error);
2946 if (error > 0)
2947 return NULL;
2948 return res;
2949}
2950
2951char *d_absolute_path(const struct path *path,
2952 char *buf, int buflen)
2953{
2954 struct path root = {};
2955 char *res = buf + buflen;
2956 int error;
2957
2958 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002959 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05002960
2961 if (error > 1)
2962 error = -EINVAL;
2963 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002964 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002965 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966}
2967
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002968/*
2969 * same as __d_path but appends "(deleted)" for unlinked files.
2970 */
Al Viro02125a82011-12-05 08:43:34 -05002971static int path_with_deleted(const struct path *path,
2972 const struct path *root,
2973 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002974{
2975 prepend(buf, buflen, "\0", 1);
2976 if (d_unlinked(path->dentry)) {
2977 int error = prepend(buf, buflen, " (deleted)", 10);
2978 if (error)
2979 return error;
2980 }
2981
2982 return prepend_path(path, root, buf, buflen);
2983}
2984
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002985static int prepend_unreachable(char **buffer, int *buflen)
2986{
2987 return prepend(buffer, buflen, "(unreachable)", 13);
2988}
2989
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07002990static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
2991{
2992 unsigned seq;
2993
2994 do {
2995 seq = read_seqcount_begin(&fs->seq);
2996 *root = fs->root;
2997 } while (read_seqcount_retry(&fs->seq, seq));
2998}
2999
Jan Bluncka03a8a702008-02-14 19:38:32 -08003000/**
3001 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003002 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003003 * @buf: buffer to return value in
3004 * @buflen: buffer length
3005 *
3006 * Convert a dentry into an ASCII path name. If the entry has been deleted
3007 * the string " (deleted)" is appended. Note that this is ambiguous.
3008 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003009 * Returns a pointer into the buffer or an error code if the path was
3010 * too long. Note: Callers should use the returned pointer, not the passed
3011 * in buffer, to use the name! The implementation often starts at an offset
3012 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003013 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003014 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003015 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003016char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003018 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003019 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003020 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003022 /*
3023 * We have various synthetic filesystems that never get mounted. On
3024 * these filesystems dentries are never used for lookup purposes, and
3025 * thus don't need to be hashed. They also don't need a name until a
3026 * user wants to identify the object in /proc/pid/fd/. The little hack
3027 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003028 *
3029 * Some pseudo inodes are mountable. When they are mounted
3030 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3031 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003032 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003033 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3034 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003035 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003036
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003037 rcu_read_lock();
3038 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003039 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003040 rcu_read_unlock();
3041
Al Viro02125a82011-12-05 08:43:34 -05003042 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003043 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 return res;
3045}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003046EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047
3048/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003049 * Helper function for dentry_operations.d_dname() members
3050 */
3051char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3052 const char *fmt, ...)
3053{
3054 va_list args;
3055 char temp[64];
3056 int sz;
3057
3058 va_start(args, fmt);
3059 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3060 va_end(args);
3061
3062 if (sz > sizeof(temp) || sz > buflen)
3063 return ERR_PTR(-ENAMETOOLONG);
3064
3065 buffer += buflen - sz;
3066 return memcpy(buffer, temp, sz);
3067}
3068
Al Viro118b2302013-08-24 12:08:17 -04003069char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3070{
3071 char *end = buffer + buflen;
3072 /* these dentries are never renamed, so d_lock is not needed */
3073 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003074 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003075 prepend(&end, &buflen, "/", 1))
3076 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003077 return end;
Al Viro118b2302013-08-24 12:08:17 -04003078}
David Herrmann31bbe162014-01-03 14:09:47 +01003079EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003080
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003081/*
Ram Pai6092d042008-03-27 13:06:20 +01003082 * Write full pathname from the root of the filesystem into the buffer.
3083 */
Al Virof6500802014-01-26 12:37:55 -05003084static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003085{
Al Virof6500802014-01-26 12:37:55 -05003086 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003087 char *end, *retval;
3088 int len, seq = 0;
3089 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003090
Al Virof6500802014-01-26 12:37:55 -05003091 if (buflen < 2)
3092 goto Elong;
3093
Al Viro48f5ec22013-09-09 15:22:25 -04003094 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003095restart:
Al Virof6500802014-01-26 12:37:55 -05003096 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003097 end = buf + buflen;
3098 len = buflen;
3099 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003100 /* Get '/' right */
3101 retval = end-1;
3102 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003103 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003104 while (!IS_ROOT(dentry)) {
3105 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003106
Ram Pai6092d042008-03-27 13:06:20 +01003107 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003108 error = prepend_name(&end, &len, &dentry->d_name);
3109 if (error)
3110 break;
Ram Pai6092d042008-03-27 13:06:20 +01003111
3112 retval = end;
3113 dentry = parent;
3114 }
Al Viro48f5ec22013-09-09 15:22:25 -04003115 if (!(seq & 1))
3116 rcu_read_unlock();
3117 if (need_seqretry(&rename_lock, seq)) {
3118 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003119 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003120 }
3121 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003122 if (error)
3123 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003124 return retval;
3125Elong:
3126 return ERR_PTR(-ENAMETOOLONG);
3127}
Nick Pigginec2447c2011-01-07 17:49:29 +11003128
3129char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3130{
Waiman Long232d2d62013-09-09 12:18:13 -04003131 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003132}
3133EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003134
3135char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3136{
3137 char *p = NULL;
3138 char *retval;
3139
Al Viroc1031352010-06-06 22:31:14 -04003140 if (d_unlinked(dentry)) {
3141 p = buf + buflen;
3142 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3143 goto Elong;
3144 buflen++;
3145 }
3146 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003147 if (!IS_ERR(retval) && p)
3148 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003149 return retval;
3150Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003151 return ERR_PTR(-ENAMETOOLONG);
3152}
3153
Linus Torvalds8b19e342013-09-12 10:35:47 -07003154static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3155 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003156{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003157 unsigned seq;
3158
3159 do {
3160 seq = read_seqcount_begin(&fs->seq);
3161 *root = fs->root;
3162 *pwd = fs->pwd;
3163 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003164}
3165
Ram Pai6092d042008-03-27 13:06:20 +01003166/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 * NOTE! The user-level library version returns a
3168 * character pointer. The kernel system call just
3169 * returns the length of the buffer filled (which
3170 * includes the ending '\0' character), or a negative
3171 * error value. So libc would do something like
3172 *
3173 * char *getcwd(char * buf, size_t size)
3174 * {
3175 * int retval;
3176 *
3177 * retval = sys_getcwd(buf, size);
3178 * if (retval >= 0)
3179 * return buf;
3180 * errno = -retval;
3181 * return NULL;
3182 * }
3183 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003184SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185{
Linus Torvalds552ce542007-02-13 12:08:18 -08003186 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003187 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003188 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189
3190 if (!page)
3191 return -ENOMEM;
3192
Linus Torvalds8b19e342013-09-12 10:35:47 -07003193 rcu_read_lock();
3194 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195
Linus Torvalds552ce542007-02-13 12:08:18 -08003196 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003197 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003198 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003199 char *cwd = page + PATH_MAX;
3200 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003202 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003203 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003204 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003205
Al Viro02125a82011-12-05 08:43:34 -05003206 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003207 goto out;
3208
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003209 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003210 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003211 error = prepend_unreachable(&cwd, &buflen);
3212 if (error)
3213 goto out;
3214 }
3215
Linus Torvalds552ce542007-02-13 12:08:18 -08003216 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003217 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003218 if (len <= size) {
3219 error = len;
3220 if (copy_to_user(buf, cwd, len))
3221 error = -EFAULT;
3222 }
Nick Piggin949854d2011-01-07 17:49:37 +11003223 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003224 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003225 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226
3227out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003228 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229 return error;
3230}
3231
3232/*
3233 * Test whether new_dentry is a subdirectory of old_dentry.
3234 *
3235 * Trivially implemented using the dcache structure
3236 */
3237
3238/**
3239 * is_subdir - is new dentry a subdirectory of old_dentry
3240 * @new_dentry: new dentry
3241 * @old_dentry: old dentry
3242 *
3243 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3244 * Returns 0 otherwise.
3245 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3246 */
3247
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003248int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249{
3250 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11003251 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003253 if (new_dentry == old_dentry)
3254 return 1;
3255
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003256 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003259 /*
3260 * Need rcu_readlock to protect against the d_parent trashing
3261 * due to d_move
3262 */
3263 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003264 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003266 else
3267 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11003268 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003270
3271 return result;
3272}
3273
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003274static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003276 struct dentry *root = data;
3277 if (dentry != root) {
3278 if (d_unhashed(dentry) || !dentry->d_inode)
3279 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003281 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3282 dentry->d_flags |= DCACHE_GENOCIDE;
3283 dentry->d_lockref.count--;
3284 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003286 return D_WALK_CONTINUE;
3287}
Nick Piggin58db63d2011-01-07 17:49:39 +11003288
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003289void d_genocide(struct dentry *parent)
3290{
3291 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292}
3293
Al Viro60545d02013-06-07 01:20:27 -04003294void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295{
Al Viro60545d02013-06-07 01:20:27 -04003296 inode_dec_link_count(inode);
3297 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3298 !hlist_unhashed(&dentry->d_alias) ||
3299 !d_unlinked(dentry));
3300 spin_lock(&dentry->d_parent->d_lock);
3301 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3302 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3303 (unsigned long long)inode->i_ino);
3304 spin_unlock(&dentry->d_lock);
3305 spin_unlock(&dentry->d_parent->d_lock);
3306 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307}
Al Viro60545d02013-06-07 01:20:27 -04003308EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309
3310static __initdata unsigned long dhash_entries;
3311static int __init set_dhash_entries(char *str)
3312{
3313 if (!str)
3314 return 0;
3315 dhash_entries = simple_strtoul(str, &str, 0);
3316 return 1;
3317}
3318__setup("dhash_entries=", set_dhash_entries);
3319
3320static void __init dcache_init_early(void)
3321{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003322 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323
3324 /* If hashes are distributed across NUMA nodes, defer
3325 * hash allocation until vmalloc space is available.
3326 */
3327 if (hashdist)
3328 return;
3329
3330 dentry_hashtable =
3331 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003332 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333 dhash_entries,
3334 13,
3335 HASH_EARLY,
3336 &d_hash_shift,
3337 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003338 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339 0);
3340
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003341 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003342 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343}
3344
Denis Cheng74bf17c2007-10-16 23:26:30 -07003345static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003347 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348
3349 /*
3350 * A constructor could be added for stable state like the lists,
3351 * but it is probably not worth it because of the cache nature
3352 * of the dcache.
3353 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003354 dentry_cache = KMEM_CACHE(dentry,
3355 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356
3357 /* Hash may have been set up in dcache_init_early */
3358 if (!hashdist)
3359 return;
3360
3361 dentry_hashtable =
3362 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003363 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 dhash_entries,
3365 13,
3366 0,
3367 &d_hash_shift,
3368 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003369 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370 0);
3371
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003372 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003373 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003374}
3375
3376/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003377struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003378EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380EXPORT_SYMBOL(d_genocide);
3381
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382void __init vfs_caches_init_early(void)
3383{
3384 dcache_init_early();
3385 inode_init_early();
3386}
3387
3388void __init vfs_caches_init(unsigned long mempages)
3389{
3390 unsigned long reserve;
3391
3392 /* Base hash sizes on available memory, with a reserve equal to
3393 150% of current kernel size */
3394
3395 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3396 mempages -= reserve;
3397
3398 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003399 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400
Denis Cheng74bf17c2007-10-16 23:26:30 -07003401 dcache_init();
3402 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003404 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405 bdev_cache_init();
3406 chrdev_init();
3407}