blob: 06f65857a855725247c1190d243c0e19cccd8570 [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
Joe Perches1f7e0612014-06-06 14:38:05 -0700153int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400154 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 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400503static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400504 __releases(dentry->d_lock)
505{
506 struct inode *inode = dentry->d_inode;
507 struct dentry *parent = NULL;
508
509 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
510 goto failed;
511
512 if (!IS_ROOT(dentry)) {
513 parent = dentry->d_parent;
514 if (unlikely(!spin_trylock(&parent->d_lock))) {
515 if (inode)
516 spin_unlock(&inode->i_lock);
517 goto failed;
518 }
519 }
520
521 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400522 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400523
524failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400525 spin_unlock(&dentry->d_lock);
526 cpu_relax();
Al Viroe55fd012014-05-28 13:51:12 -0400527 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100528}
529
Al Viro046b9612014-05-29 08:54:52 -0400530static inline struct dentry *lock_parent(struct dentry *dentry)
531{
532 struct dentry *parent = dentry->d_parent;
533 if (IS_ROOT(dentry))
534 return NULL;
Al Viroc2338f22014-06-12 00:29:13 -0400535 if (unlikely((int)dentry->d_lockref.count < 0))
536 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400537 if (likely(spin_trylock(&parent->d_lock)))
538 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400539 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400540 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400541again:
542 parent = ACCESS_ONCE(dentry->d_parent);
543 spin_lock(&parent->d_lock);
544 /*
545 * We can't blindly lock dentry until we are sure
546 * that we won't violate the locking order.
547 * Any changes of dentry->d_parent must have
548 * been done with parent->d_lock held, so
549 * spin_lock() above is enough of a barrier
550 * for checking if it's still our child.
551 */
552 if (unlikely(parent != dentry->d_parent)) {
553 spin_unlock(&parent->d_lock);
554 goto again;
555 }
556 rcu_read_unlock();
557 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700558 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400559 else
560 parent = NULL;
561 return parent;
562}
563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564/*
565 * This is dput
566 *
567 * This is complicated by the fact that we do not want to put
568 * dentries that are no longer on any hash chain on the unused
569 * list: we'd much rather just get rid of them immediately.
570 *
571 * However, that implies that we have to traverse the dentry
572 * tree upwards to the parents which might _also_ now be
573 * scheduled for deletion (it may have been only waiting for
574 * its last child to go away).
575 *
576 * This tail recursion is done by hand as we don't want to depend
577 * on the compiler to always get this right (gcc generally doesn't).
578 * Real recursion would eat up our stack space.
579 */
580
581/*
582 * dput - release a dentry
583 * @dentry: dentry to release
584 *
585 * Release a dentry. This will drop the usage count and if appropriate
586 * call the dentry unlink method as well as removing it from the queues and
587 * releasing its resources. If the parent dentries were scheduled for release
588 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590void dput(struct dentry *dentry)
591{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700592 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 return;
594
595repeat:
Waiman Long98474232013-08-28 18:24:59 -0700596 if (lockref_put_or_lock(&dentry->d_lockref))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700599 /* Unreachable? Get rid of it */
600 if (unlikely(d_unhashed(dentry)))
601 goto kill_it;
602
603 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100605 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 }
Nick Piggin265ac902010-10-10 05:36:24 -0400607
Linus Torvalds358eec12013-10-31 15:43:02 -0700608 if (!(dentry->d_flags & DCACHE_REFERENCED))
609 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400610 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400611
Waiman Long98474232013-08-28 18:24:59 -0700612 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100613 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 return;
615
Miklos Szeredid52b9082007-05-08 00:23:46 -0700616kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400617 dentry = dentry_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700618 if (dentry)
619 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700621EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622
623/**
624 * d_invalidate - invalidate a dentry
625 * @dentry: dentry to invalidate
626 *
627 * Try to invalidate the dentry if it turns out to be
628 * possible. If there are other dentries that can be
629 * reached through this one we can't delete it and we
630 * return -EBUSY. On success we return 0.
631 *
632 * no dcache lock.
633 */
634
635int d_invalidate(struct dentry * dentry)
636{
637 /*
638 * If it's already been dropped, return OK.
639 */
Nick Pigginda502952011-01-07 17:49:33 +1100640 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100642 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 return 0;
644 }
645 /*
646 * Check whether to do a partial shrink_dcache
647 * to get rid of unused child entries.
648 */
649 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100650 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100652 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 }
654
655 /*
656 * Somebody else still using it?
657 *
658 * If it's a directory, we can't drop it
659 * for fear of somebody re-populating it
660 * with children (even though dropping it
661 * would make it unreachable from the root,
662 * we might still populate it if it was a
663 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000664 * We also need to leave mountpoints alone,
665 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 */
Waiman Long98474232013-08-28 18:24:59 -0700667 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
Al Viro50e69632011-11-07 16:39:57 +0000668 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 return -EBUSY;
671 }
672 }
673
674 __d_drop(dentry);
675 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 return 0;
677}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700678EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100680/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100681static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
Waiman Long98474232013-08-28 18:24:59 -0700683 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684}
685
Nick Piggindc0474b2011-01-07 17:49:43 +1100686static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100687{
Waiman Long98474232013-08-28 18:24:59 -0700688 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100689}
690
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100691struct dentry *dget_parent(struct dentry *dentry)
692{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700693 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100694 struct dentry *ret;
695
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700696 /*
697 * Do optimistic parent lookup without any
698 * locking.
699 */
700 rcu_read_lock();
701 ret = ACCESS_ONCE(dentry->d_parent);
702 gotref = lockref_get_not_zero(&ret->d_lockref);
703 rcu_read_unlock();
704 if (likely(gotref)) {
705 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
706 return ret;
707 dput(ret);
708 }
709
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100710repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100711 /*
712 * Don't need rcu_dereference because we re-check it was correct under
713 * the lock.
714 */
715 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100716 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100717 spin_lock(&ret->d_lock);
718 if (unlikely(ret != dentry->d_parent)) {
719 spin_unlock(&ret->d_lock);
720 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100721 goto repeat;
722 }
Nick Piggina734eb42011-01-07 17:49:44 +1100723 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700724 BUG_ON(!ret->d_lockref.count);
725 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100726 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100727 return ret;
728}
729EXPORT_SYMBOL(dget_parent);
730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731/**
732 * d_find_alias - grab a hashed alias of inode
733 * @inode: inode in question
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700734 * @want_discon: flag, used by d_splice_alias, to request
735 * that only a DISCONNECTED alias be returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 *
737 * If inode has a hashed alias, or is a directory and has any alias,
738 * acquire the reference to alias and return it. Otherwise return NULL.
739 * Notice that if inode is a directory there can be only one alias and
740 * it can be unhashed only if it has no children, or if it is the root
741 * of a filesystem.
742 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700743 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700744 * any other hashed alias over that one unless @want_discon is set,
745 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 */
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700747static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748{
Nick Pigginda502952011-01-07 17:49:33 +1100749 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Nick Pigginda502952011-01-07 17:49:33 +1100751again:
752 discon_alias = NULL;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800753 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100754 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700756 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100757 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 discon_alias = alias;
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700759 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100760 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100761 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 return alias;
763 }
764 }
Nick Pigginda502952011-01-07 17:49:33 +1100765 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 }
Nick Pigginda502952011-01-07 17:49:33 +1100767 if (discon_alias) {
768 alias = discon_alias;
769 spin_lock(&alias->d_lock);
770 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
771 if (IS_ROOT(alias) &&
772 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100773 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100774 spin_unlock(&alias->d_lock);
775 return alias;
776 }
777 }
778 spin_unlock(&alias->d_lock);
779 goto again;
780 }
781 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782}
783
Nick Pigginda502952011-01-07 17:49:33 +1100784struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
David Howells214fda12006-03-25 03:06:36 -0800786 struct dentry *de = NULL;
787
Al Virob3d9b7a2012-06-09 13:51:19 -0400788 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100789 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700790 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100791 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 return de;
794}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700795EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
797/*
798 * Try to kill dentries associated with this inode.
799 * WARNING: you must own a reference to inode.
800 */
801void d_prune_aliases(struct inode *inode)
802{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700803 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100805 spin_lock(&inode->i_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800806 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700808 if (!dentry->d_lockref.count) {
Yan, Zheng590fb512013-08-13 15:42:02 +0800809 /*
810 * inform the fs via d_prune that this dentry
811 * is about to be unhashed and destroyed.
812 */
813 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
814 !d_unhashed(dentry))
815 dentry->d_op->d_prune(dentry);
816
Nick Piggindc0474b2011-01-07 17:49:43 +1100817 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 __d_drop(dentry);
819 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100820 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 dput(dentry);
822 goto restart;
823 }
824 spin_unlock(&dentry->d_lock);
825 }
Nick Piggin873feea2011-01-07 17:50:06 +1100826 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700828EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400830static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831{
Al Viro5c47e6d2014-04-29 16:13:18 -0400832 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700833
Miklos Szeredi60942f22014-05-02 15:38:39 -0400834 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400835 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400836 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100837 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400838 parent = lock_parent(dentry);
839
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000841 * The dispose list is isolated and dentries are not accounted
842 * to the LRU here, so we can simply remove it from the list
843 * here regardless of whether it is referenced or not.
844 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400845 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000846
847 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000849 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 */
Al Viro41edf272014-05-01 10:30:00 -0400851 if ((int)dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700852 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400853 if (parent)
854 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 continue;
856 }
Nick Piggin77812a12011-01-07 17:49:48 +1100857
Al Viro64fd72e2014-05-28 09:48:44 -0400858
859 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
860 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
861 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400862 if (parent)
863 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -0400864 if (can_free)
865 dentry_free(dentry);
866 continue;
867 }
868
Al Viroff2fde92014-05-28 13:59:13 -0400869 inode = dentry->d_inode;
870 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400871 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000872 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400873 if (parent)
874 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400875 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000876 }
Al Viroff2fde92014-05-28 13:59:13 -0400877
Al Viroff2fde92014-05-28 13:59:13 -0400878 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -0400879
Al Viro5c47e6d2014-04-29 16:13:18 -0400880 /*
881 * We need to prune ancestors too. This is necessary to prevent
882 * quadratic behavior of shrink_dcache_parent(), but is also
883 * expected to be beneficial in reducing dentry cache
884 * fragmentation.
885 */
886 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -0400887 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
888 parent = lock_parent(dentry);
889 if (dentry->d_lockref.count != 1) {
890 dentry->d_lockref.count--;
891 spin_unlock(&dentry->d_lock);
892 if (parent)
893 spin_unlock(&parent->d_lock);
894 break;
895 }
896 inode = dentry->d_inode; /* can't be NULL */
897 if (unlikely(!spin_trylock(&inode->i_lock))) {
898 spin_unlock(&dentry->d_lock);
899 if (parent)
900 spin_unlock(&parent->d_lock);
901 cpu_relax();
902 continue;
903 }
904 __dentry_kill(dentry);
905 dentry = parent;
906 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400908}
909
Dave Chinnerf6041562013-08-28 10:18:00 +1000910static enum lru_status
911dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
912{
913 struct list_head *freeable = arg;
914 struct dentry *dentry = container_of(item, struct dentry, d_lru);
915
916
917 /*
918 * we are inverting the lru lock/dentry->d_lock here,
919 * so use a trylock. If we fail to get the lock, just skip
920 * it
921 */
922 if (!spin_trylock(&dentry->d_lock))
923 return LRU_SKIP;
924
925 /*
926 * Referenced dentries are still in use. If they have active
927 * counts, just remove them from the LRU. Otherwise give them
928 * another pass through the LRU.
929 */
930 if (dentry->d_lockref.count) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400931 d_lru_isolate(dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +1000932 spin_unlock(&dentry->d_lock);
933 return LRU_REMOVED;
934 }
935
936 if (dentry->d_flags & DCACHE_REFERENCED) {
937 dentry->d_flags &= ~DCACHE_REFERENCED;
938 spin_unlock(&dentry->d_lock);
939
940 /*
941 * The list move itself will be made by the common LRU code. At
942 * this point, we've dropped the dentry->d_lock but keep the
943 * lru lock. This is safe to do, since every list movement is
944 * protected by the lru lock even if both locks are held.
945 *
946 * This is guaranteed by the fact that all LRU management
947 * functions are intermediated by the LRU API calls like
948 * list_lru_add and list_lru_del. List movement in this file
949 * only ever occur through this functions or through callbacks
950 * like this one, that are called from the LRU API.
951 *
952 * The only exceptions to this are functions like
953 * shrink_dentry_list, and code that first checks for the
954 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
955 * operating only with stack provided lists after they are
956 * properly isolated from the main list. It is thus, always a
957 * local access.
958 */
959 return LRU_ROTATE;
960 }
961
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400962 d_lru_shrink_move(dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +1000963 spin_unlock(&dentry->d_lock);
964
965 return LRU_REMOVED;
966}
967
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400968/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000969 * prune_dcache_sb - shrink the dcache
970 * @sb: superblock
Dave Chinnerf6041562013-08-28 10:18:00 +1000971 * @nr_to_scan : number of entries to try to free
Dave Chinner9b17c622013-08-28 10:18:05 +1000972 * @nid: which node to scan for freeable entities
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400973 *
Dave Chinnerf6041562013-08-28 10:18:00 +1000974 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000975 * done when we need more memory an called from the superblock shrinker
976 * function.
977 *
978 * This function may fail to free any resources if all the dentries are in
979 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400980 */
Dave Chinner9b17c622013-08-28 10:18:05 +1000981long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
982 int nid)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400983{
Dave Chinnerf6041562013-08-28 10:18:00 +1000984 LIST_HEAD(dispose);
985 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400986
Dave Chinner9b17c622013-08-28 10:18:05 +1000987 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
988 &dispose, &nr_to_scan);
Dave Chinnerf6041562013-08-28 10:18:00 +1000989 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +1000990 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991}
992
Glauber Costa4e717f52013-08-28 10:18:03 +1000993static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
994 spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000995{
Glauber Costa4e717f52013-08-28 10:18:03 +1000996 struct list_head *freeable = arg;
997 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000998
Glauber Costa4e717f52013-08-28 10:18:03 +1000999 /*
1000 * we are inverting the lru lock/dentry->d_lock here,
1001 * so use a trylock. If we fail to get the lock, just skip
1002 * it
1003 */
1004 if (!spin_trylock(&dentry->d_lock))
1005 return LRU_SKIP;
1006
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001007 d_lru_shrink_move(dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001008 spin_unlock(&dentry->d_lock);
1009
1010 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001011}
1012
Glauber Costa4e717f52013-08-28 10:18:03 +10001013
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001014/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 * shrink_dcache_sb - shrink dcache for a superblock
1016 * @sb: superblock
1017 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001018 * Shrink the dcache for the specified super block. This is used to free
1019 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001021void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022{
Glauber Costa4e717f52013-08-28 10:18:03 +10001023 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001024
Glauber Costa4e717f52013-08-28 10:18:03 +10001025 do {
1026 LIST_HEAD(dispose);
1027
1028 freed = list_lru_walk(&sb->s_dentry_lru,
1029 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1030
1031 this_cpu_sub(nr_dentry_unused, freed);
1032 shrink_dentry_list(&dispose);
1033 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001035EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001038 * enum d_walk_ret - action to talke during tree walk
1039 * @D_WALK_CONTINUE: contrinue walk
1040 * @D_WALK_QUIT: quit walk
1041 * @D_WALK_NORETRY: quit when retry is needed
1042 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001044enum d_walk_ret {
1045 D_WALK_CONTINUE,
1046 D_WALK_QUIT,
1047 D_WALK_NORETRY,
1048 D_WALK_SKIP,
1049};
1050
1051/**
1052 * d_walk - walk the dentry tree
1053 * @parent: start of walk
1054 * @data: data passed to @enter() and @finish()
1055 * @enter: callback when first entering the dentry
1056 * @finish: callback when successfully finished the walk
1057 *
1058 * The @enter() and @finish() callbacks are called with d_lock held.
1059 */
1060static void d_walk(struct dentry *parent, void *data,
1061 enum d_walk_ret (*enter)(void *, struct dentry *),
1062 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063{
Nick Piggin949854d2011-01-07 17:49:37 +11001064 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001066 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001067 enum d_walk_ret ret;
1068 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001069
Nick Piggin58db63d2011-01-07 17:49:39 +11001070again:
Al Viro48f5ec22013-09-09 15:22:25 -04001071 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001072 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001073 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001074
1075 ret = enter(data, this_parent);
1076 switch (ret) {
1077 case D_WALK_CONTINUE:
1078 break;
1079 case D_WALK_QUIT:
1080 case D_WALK_SKIP:
1081 goto out_unlock;
1082 case D_WALK_NORETRY:
1083 retry = false;
1084 break;
1085 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086repeat:
1087 next = this_parent->d_subdirs.next;
1088resume:
1089 while (next != &this_parent->d_subdirs) {
1090 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001091 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001093
1094 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001095
1096 ret = enter(data, dentry);
1097 switch (ret) {
1098 case D_WALK_CONTINUE:
1099 break;
1100 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001101 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001102 goto out_unlock;
1103 case D_WALK_NORETRY:
1104 retry = false;
1105 break;
1106 case D_WALK_SKIP:
1107 spin_unlock(&dentry->d_lock);
1108 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001109 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001110
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001112 spin_unlock(&this_parent->d_lock);
1113 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001115 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 goto repeat;
1117 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001118 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 }
1120 /*
1121 * All done at this level ... ascend and resume the search.
1122 */
1123 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001124 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001125 this_parent = child->d_parent;
1126
1127 rcu_read_lock();
1128 spin_unlock(&child->d_lock);
1129 spin_lock(&this_parent->d_lock);
1130
1131 /*
1132 * might go back up the wrong parent if we have had a rename
1133 * or deletion
1134 */
1135 if (this_parent != child->d_parent ||
1136 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1137 need_seqretry(&rename_lock, seq)) {
1138 spin_unlock(&this_parent->d_lock);
1139 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11001140 goto rename_retry;
Al Viro31dec132013-10-25 17:04:27 -04001141 }
1142 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11001143 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 goto resume;
1145 }
Al Viro48f5ec22013-09-09 15:22:25 -04001146 if (need_seqretry(&rename_lock, seq)) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001147 spin_unlock(&this_parent->d_lock);
1148 goto rename_retry;
1149 }
1150 if (finish)
1151 finish(data);
1152
1153out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001154 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001155 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001156 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001157
1158rename_retry:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001159 if (!retry)
1160 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001161 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001162 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001164
1165/*
1166 * Search for at least 1 mount point in the dentry's subdirs.
1167 * We descend to the next level whenever the d_subdirs
1168 * list is non-empty and continue searching.
1169 */
1170
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001171static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1172{
1173 int *ret = data;
1174 if (d_mountpoint(dentry)) {
1175 *ret = 1;
1176 return D_WALK_QUIT;
1177 }
1178 return D_WALK_CONTINUE;
1179}
1180
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001181/**
1182 * have_submounts - check for mounts over a dentry
1183 * @parent: dentry to check.
1184 *
1185 * Return true if the parent or its subdirectories contain
1186 * a mount point
1187 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001188int have_submounts(struct dentry *parent)
1189{
1190 int ret = 0;
1191
1192 d_walk(parent, &ret, check_mount, NULL);
1193
1194 return ret;
1195}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001196EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
1198/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001199 * Called by mount code to set a mountpoint and check if the mountpoint is
1200 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1201 * subtree can become unreachable).
1202 *
1203 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1204 * this reason take rename_lock and d_lock on dentry and ancestors.
1205 */
1206int d_set_mounted(struct dentry *dentry)
1207{
1208 struct dentry *p;
1209 int ret = -ENOENT;
1210 write_seqlock(&rename_lock);
1211 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1212 /* Need exclusion wrt. check_submounts_and_drop() */
1213 spin_lock(&p->d_lock);
1214 if (unlikely(d_unhashed(p))) {
1215 spin_unlock(&p->d_lock);
1216 goto out;
1217 }
1218 spin_unlock(&p->d_lock);
1219 }
1220 spin_lock(&dentry->d_lock);
1221 if (!d_unlinked(dentry)) {
1222 dentry->d_flags |= DCACHE_MOUNTED;
1223 ret = 0;
1224 }
1225 spin_unlock(&dentry->d_lock);
1226out:
1227 write_sequnlock(&rename_lock);
1228 return ret;
1229}
1230
1231/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001232 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 * and move any unused dentries to the end of the unused
1234 * list for prune_dcache(). We descend to the next level
1235 * whenever the d_subdirs list is non-empty and continue
1236 * searching.
1237 *
1238 * It returns zero iff there are no unused children,
1239 * otherwise it returns the number of children moved to
1240 * the end of the unused list. This may not be the total
1241 * number of unused children, because select_parent can
1242 * drop the lock and return early due to latency
1243 * constraints.
1244 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001245
1246struct select_data {
1247 struct dentry *start;
1248 struct list_head dispose;
1249 int found;
1250};
1251
1252static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001254 struct select_data *data = _data;
1255 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001257 if (data->start == dentry)
1258 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
Al Virofe915222014-05-03 00:02:25 -04001260 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001261 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001262 } else {
1263 if (dentry->d_flags & DCACHE_LRU_LIST)
1264 d_lru_del(dentry);
1265 if (!dentry->d_lockref.count) {
1266 d_shrink_add(dentry, &data->dispose);
1267 data->found++;
1268 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 }
1270 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001271 * We can return to the caller if we have found some (this
1272 * ensures forward progress). We'll be coming back to find
1273 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 */
Al Virofe915222014-05-03 00:02:25 -04001275 if (!list_empty(&data->dispose))
1276 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001278 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279}
1280
1281/**
1282 * shrink_dcache_parent - prune dcache
1283 * @parent: parent of entries to prune
1284 *
1285 * Prune the dcache to remove unused children of the parent dentry.
1286 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001287void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001289 for (;;) {
1290 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001292 INIT_LIST_HEAD(&data.dispose);
1293 data.start = parent;
1294 data.found = 0;
1295
1296 d_walk(parent, &data, select_collect, NULL);
1297 if (!data.found)
1298 break;
1299
1300 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001301 cond_resched();
1302 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001304EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
Al Viro9c8c10e2014-05-02 20:36:10 -04001306static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001307{
Al Viro9c8c10e2014-05-02 20:36:10 -04001308 /* it has busy descendents; complain about those instead */
1309 if (!list_empty(&dentry->d_subdirs))
1310 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001311
Al Viro9c8c10e2014-05-02 20:36:10 -04001312 /* root with refcount 1 is fine */
1313 if (dentry == _data && dentry->d_lockref.count == 1)
1314 return D_WALK_CONTINUE;
1315
1316 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1317 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001318 dentry,
1319 dentry->d_inode ?
1320 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001321 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001322 dentry->d_lockref.count,
1323 dentry->d_sb->s_type->name,
1324 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001325 WARN_ON(1);
1326 return D_WALK_CONTINUE;
1327}
1328
1329static void do_one_tree(struct dentry *dentry)
1330{
1331 shrink_dcache_parent(dentry);
1332 d_walk(dentry, dentry, umount_check, NULL);
1333 d_drop(dentry);
1334 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001335}
1336
1337/*
1338 * destroy the dentries attached to a superblock on unmounting
1339 */
1340void shrink_dcache_for_umount(struct super_block *sb)
1341{
1342 struct dentry *dentry;
1343
Al Viro9c8c10e2014-05-02 20:36:10 -04001344 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001345
1346 dentry = sb->s_root;
1347 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001348 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001349
1350 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001351 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1352 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001353 }
1354}
1355
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001356static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1357{
1358 struct select_data *data = _data;
1359
1360 if (d_mountpoint(dentry)) {
1361 data->found = -EBUSY;
1362 return D_WALK_QUIT;
1363 }
1364
1365 return select_collect(_data, dentry);
1366}
1367
1368static void check_and_drop(void *_data)
1369{
1370 struct select_data *data = _data;
1371
1372 if (d_mountpoint(data->start))
1373 data->found = -EBUSY;
1374 if (!data->found)
1375 __d_drop(data->start);
1376}
1377
1378/**
1379 * check_submounts_and_drop - prune dcache, check for submounts and drop
1380 *
1381 * All done as a single atomic operation relative to has_unlinked_ancestor().
1382 * Returns 0 if successfully unhashed @parent. If there were submounts then
1383 * return -EBUSY.
1384 *
1385 * @dentry: dentry to prune and drop
1386 */
1387int check_submounts_and_drop(struct dentry *dentry)
1388{
1389 int ret = 0;
1390
1391 /* Negative dentries can be dropped without further checks */
1392 if (!dentry->d_inode) {
1393 d_drop(dentry);
1394 goto out;
1395 }
1396
1397 for (;;) {
1398 struct select_data data;
1399
1400 INIT_LIST_HEAD(&data.dispose);
1401 data.start = dentry;
1402 data.found = 0;
1403
1404 d_walk(dentry, &data, check_and_collect, check_and_drop);
1405 ret = data.found;
1406
1407 if (!list_empty(&data.dispose))
1408 shrink_dentry_list(&data.dispose);
1409
1410 if (ret <= 0)
1411 break;
1412
1413 cond_resched();
1414 }
1415
1416out:
1417 return ret;
1418}
1419EXPORT_SYMBOL(check_submounts_and_drop);
1420
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421/**
Al Viroa4464db2011-07-07 15:03:58 -04001422 * __d_alloc - allocate a dcache entry
1423 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 * @name: qstr of the name
1425 *
1426 * Allocates a dentry. It returns %NULL if there is insufficient memory
1427 * available. On a success the dentry is returned. The name passed in is
1428 * copied and the copy passed in may be reused after this call.
1429 */
1430
Al Viroa4464db2011-07-07 15:03:58 -04001431struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432{
1433 struct dentry *dentry;
1434 char *dname;
1435
Mel Gormane12ba742007-10-16 01:25:52 -07001436 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 if (!dentry)
1438 return NULL;
1439
Linus Torvalds6326c712012-05-21 16:14:04 -07001440 /*
1441 * We guarantee that the inline name is always NUL-terminated.
1442 * This way the memcpy() done by the name switching in rename
1443 * will still always have a NUL at the end, even if we might
1444 * be overwriting an internal NUL character
1445 */
1446 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 if (name->len > DNAME_INLINE_LEN-1) {
1448 dname = kmalloc(name->len + 1, GFP_KERNEL);
1449 if (!dname) {
1450 kmem_cache_free(dentry_cache, dentry);
1451 return NULL;
1452 }
1453 } else {
1454 dname = dentry->d_iname;
1455 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
1457 dentry->d_name.len = name->len;
1458 dentry->d_name.hash = name->hash;
1459 memcpy(dname, name->name, name->len);
1460 dname[name->len] = 0;
1461
Linus Torvalds6326c712012-05-21 16:14:04 -07001462 /* Make sure we always see the terminating NUL character */
1463 smp_wmb();
1464 dentry->d_name.name = dname;
1465
Waiman Long98474232013-08-28 18:24:59 -07001466 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001467 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001469 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001471 dentry->d_parent = dentry;
1472 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 dentry->d_op = NULL;
1474 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001475 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 INIT_LIST_HEAD(&dentry->d_lru);
1477 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001478 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001479 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001480 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481
Nick Piggin3e880fb2011-01-07 17:49:19 +11001482 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001483
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 return dentry;
1485}
Al Viroa4464db2011-07-07 15:03:58 -04001486
1487/**
1488 * d_alloc - allocate a dcache entry
1489 * @parent: parent of entry to allocate
1490 * @name: qstr of the name
1491 *
1492 * Allocates a dentry. It returns %NULL if there is insufficient memory
1493 * available. On a success the dentry is returned. The name passed in is
1494 * copied and the copy passed in may be reused after this call.
1495 */
1496struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1497{
1498 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1499 if (!dentry)
1500 return NULL;
1501
1502 spin_lock(&parent->d_lock);
1503 /*
1504 * don't need child lock because it is not subject
1505 * to concurrency here
1506 */
1507 __dget_dlock(parent);
1508 dentry->d_parent = parent;
1509 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1510 spin_unlock(&parent->d_lock);
1511
1512 return dentry;
1513}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001514EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001516/**
1517 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1518 * @sb: the superblock
1519 * @name: qstr of the name
1520 *
1521 * For a filesystem that just pins its dentries in memory and never
1522 * performs lookups at all, return an unhashed IS_ROOT dentry.
1523 */
Nick Piggin4b936882011-01-07 17:50:07 +11001524struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1525{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001526 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001527}
1528EXPORT_SYMBOL(d_alloc_pseudo);
1529
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1531{
1532 struct qstr q;
1533
1534 q.name = name;
1535 q.len = strlen(name);
1536 q.hash = full_name_hash(q.name, q.len);
1537 return d_alloc(parent, &q);
1538}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001539EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
Nick Pigginfb045ad2011-01-07 17:49:55 +11001541void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1542{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001543 WARN_ON_ONCE(dentry->d_op);
1544 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001545 DCACHE_OP_COMPARE |
1546 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001547 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001548 DCACHE_OP_DELETE ));
1549 dentry->d_op = op;
1550 if (!op)
1551 return;
1552 if (op->d_hash)
1553 dentry->d_flags |= DCACHE_OP_HASH;
1554 if (op->d_compare)
1555 dentry->d_flags |= DCACHE_OP_COMPARE;
1556 if (op->d_revalidate)
1557 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001558 if (op->d_weak_revalidate)
1559 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001560 if (op->d_delete)
1561 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001562 if (op->d_prune)
1563 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001564
1565}
1566EXPORT_SYMBOL(d_set_d_op);
1567
David Howellsb18825a2013-09-12 19:22:53 +01001568static unsigned d_flags_for_inode(struct inode *inode)
1569{
1570 unsigned add_flags = DCACHE_FILE_TYPE;
1571
1572 if (!inode)
1573 return DCACHE_MISS_TYPE;
1574
1575 if (S_ISDIR(inode->i_mode)) {
1576 add_flags = DCACHE_DIRECTORY_TYPE;
1577 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1578 if (unlikely(!inode->i_op->lookup))
1579 add_flags = DCACHE_AUTODIR_TYPE;
1580 else
1581 inode->i_opflags |= IOP_LOOKUP;
1582 }
1583 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1584 if (unlikely(inode->i_op->follow_link))
1585 add_flags = DCACHE_SYMLINK_TYPE;
1586 else
1587 inode->i_opflags |= IOP_NOFOLLOW;
1588 }
1589
1590 if (unlikely(IS_AUTOMOUNT(inode)))
1591 add_flags |= DCACHE_NEED_AUTOMOUNT;
1592 return add_flags;
1593}
1594
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001595static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1596{
David Howellsb18825a2013-09-12 19:22:53 +01001597 unsigned add_flags = d_flags_for_inode(inode);
1598
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001599 spin_lock(&dentry->d_lock);
Al Viro22213312014-04-19 12:30:58 -04001600 __d_set_type(dentry, add_flags);
David Howellsb18825a2013-09-12 19:22:53 +01001601 if (inode)
Al Virob3d9b7a2012-06-09 13:51:19 -04001602 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001603 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001604 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001605 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001606 fsnotify_d_instantiate(dentry, inode);
1607}
1608
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609/**
1610 * d_instantiate - fill in inode information for a dentry
1611 * @entry: dentry to complete
1612 * @inode: inode to attach to this dentry
1613 *
1614 * Fill in inode information in the entry.
1615 *
1616 * This turns negative dentries into productive full members
1617 * of society.
1618 *
1619 * NOTE! This assumes that the inode count has been incremented
1620 * (or otherwise set) by the caller to indicate that it is now
1621 * in use by the dcache.
1622 */
1623
1624void d_instantiate(struct dentry *entry, struct inode * inode)
1625{
Al Virob3d9b7a2012-06-09 13:51:19 -04001626 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001627 if (inode)
1628 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001629 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001630 if (inode)
1631 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 security_d_instantiate(entry, inode);
1633}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001634EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
1636/**
1637 * d_instantiate_unique - instantiate a non-aliased dentry
1638 * @entry: dentry to instantiate
1639 * @inode: inode to attach to this dentry
1640 *
1641 * Fill in inode information in the entry. On success, it returns NULL.
1642 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001643 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 *
1645 * Note that in order to avoid conflicts with rename() etc, the caller
1646 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001647 *
1648 * This also assumes that the inode count has been incremented
1649 * (or otherwise set) by the caller to indicate that it is now
1650 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 */
David Howells770bfad2006-08-22 20:06:07 -04001652static struct dentry *__d_instantiate_unique(struct dentry *entry,
1653 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
1655 struct dentry *alias;
1656 int len = entry->d_name.len;
1657 const char *name = entry->d_name.name;
1658 unsigned int hash = entry->d_name.hash;
1659
David Howells770bfad2006-08-22 20:06:07 -04001660 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001661 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001662 return NULL;
1663 }
1664
Sasha Levinb67bfe02013-02-27 17:06:00 -08001665 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001666 /*
1667 * Don't need alias->d_lock here, because aliases with
1668 * d_parent == entry->d_parent are not subject to name or
1669 * parent changes, because the parent inode i_mutex is held.
1670 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001671 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 continue;
1673 if (alias->d_parent != entry->d_parent)
1674 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001675 if (alias->d_name.len != len)
1676 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001677 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001679 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 return alias;
1681 }
David Howells770bfad2006-08-22 20:06:07 -04001682
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001683 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 return NULL;
1685}
David Howells770bfad2006-08-22 20:06:07 -04001686
1687struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1688{
1689 struct dentry *result;
1690
Al Virob3d9b7a2012-06-09 13:51:19 -04001691 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001692
Nick Piggin873feea2011-01-07 17:50:06 +11001693 if (inode)
1694 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001695 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001696 if (inode)
1697 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001698
1699 if (!result) {
1700 security_d_instantiate(entry, inode);
1701 return NULL;
1702 }
1703
1704 BUG_ON(!d_unhashed(result));
1705 iput(inode);
1706 return result;
1707}
1708
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709EXPORT_SYMBOL(d_instantiate_unique);
1710
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001711/**
1712 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1713 * @entry: dentry to complete
1714 * @inode: inode to attach to this dentry
1715 *
1716 * Fill in inode information in the entry. If a directory alias is found, then
1717 * return an error (and drop inode). Together with d_materialise_unique() this
1718 * guarantees that a directory inode may never have more than one alias.
1719 */
1720int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1721{
1722 BUG_ON(!hlist_unhashed(&entry->d_alias));
1723
1724 spin_lock(&inode->i_lock);
1725 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1726 spin_unlock(&inode->i_lock);
1727 iput(inode);
1728 return -EBUSY;
1729 }
1730 __d_instantiate(entry, inode);
1731 spin_unlock(&inode->i_lock);
1732 security_d_instantiate(entry, inode);
1733
1734 return 0;
1735}
1736EXPORT_SYMBOL(d_instantiate_no_diralias);
1737
Al Viroadc0e912012-01-08 16:49:21 -05001738struct dentry *d_make_root(struct inode *root_inode)
1739{
1740 struct dentry *res = NULL;
1741
1742 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001743 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001744
1745 res = __d_alloc(root_inode->i_sb, &name);
1746 if (res)
1747 d_instantiate(res, root_inode);
1748 else
1749 iput(root_inode);
1750 }
1751 return res;
1752}
1753EXPORT_SYMBOL(d_make_root);
1754
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001755static struct dentry * __d_find_any_alias(struct inode *inode)
1756{
1757 struct dentry *alias;
1758
Al Virob3d9b7a2012-06-09 13:51:19 -04001759 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001760 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001761 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001762 __dget(alias);
1763 return alias;
1764}
1765
Sage Weil46f72b32012-01-10 09:04:37 -08001766/**
1767 * d_find_any_alias - find any alias for a given inode
1768 * @inode: inode to find an alias for
1769 *
1770 * If any aliases exist for the given inode, take and return a
1771 * reference for one of them. If no aliases exist, return %NULL.
1772 */
1773struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001774{
1775 struct dentry *de;
1776
1777 spin_lock(&inode->i_lock);
1778 de = __d_find_any_alias(inode);
1779 spin_unlock(&inode->i_lock);
1780 return de;
1781}
Sage Weil46f72b32012-01-10 09:04:37 -08001782EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001783
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001785 * d_obtain_alias - find or allocate a dentry for a given inode
1786 * @inode: inode to allocate the dentry for
1787 *
1788 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1789 * similar open by handle operations. The returned dentry may be anonymous,
1790 * or may have a full name (if the inode was already in the cache).
1791 *
1792 * When called on a directory inode, we must ensure that the inode only ever
1793 * has one dentry. If a dentry is found, that is returned instead of
1794 * allocating a new one.
1795 *
1796 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001797 * to the dentry. In case of an error the reference on the inode is released.
1798 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1799 * be passed in and will be the error will be propagate to the return value,
1800 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001801 */
1802struct dentry *d_obtain_alias(struct inode *inode)
1803{
NeilBrownb911a6b2012-11-08 16:09:37 -08001804 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001805 struct dentry *tmp;
1806 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001807 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001808
1809 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001810 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001811 if (IS_ERR(inode))
1812 return ERR_CAST(inode);
1813
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001814 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001815 if (res)
1816 goto out_iput;
1817
Al Viroa4464db2011-07-07 15:03:58 -04001818 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001819 if (!tmp) {
1820 res = ERR_PTR(-ENOMEM);
1821 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001822 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001823
Nick Piggin873feea2011-01-07 17:50:06 +11001824 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001825 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001826 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001827 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001828 dput(tmp);
1829 goto out_iput;
1830 }
1831
1832 /* attach a disconnected dentry */
David Howellsb18825a2013-09-12 19:22:53 +01001833 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1834
Christoph Hellwig9308a612008-08-11 15:49:12 +02001835 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001836 tmp->d_inode = inode;
David Howellsb18825a2013-09-12 19:22:53 +01001837 tmp->d_flags |= add_flags;
Al Virob3d9b7a2012-06-09 13:51:19 -04001838 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001839 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001840 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001841 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001842 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001843 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001844 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001845
Christoph Hellwig9308a612008-08-11 15:49:12 +02001846 return tmp;
1847
1848 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001849 if (res && !IS_ERR(res))
1850 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001851 iput(inode);
1852 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001853}
Benny Halevyadc48722009-02-27 14:02:59 -08001854EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
1856/**
1857 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1858 * @inode: the inode which may have a disconnected dentry
1859 * @dentry: a negative dentry which we want to point to the inode.
1860 *
1861 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1862 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1863 * and return it, else simply d_add the inode to the dentry and return NULL.
1864 *
1865 * This is needed in the lookup routine of any filesystem that is exportable
1866 * (via knfsd) so that we can build dcache paths to directories effectively.
1867 *
1868 * If a dentry was found and moved, then it is returned. Otherwise NULL
1869 * is returned. This matches the expected return value of ->lookup.
1870 *
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001871 * Cluster filesystems may call this function with a negative, hashed dentry.
1872 * In that case, we know that the inode will be a regular file, and also this
1873 * will only occur during atomic_open. So we need to check for the dentry
1874 * being already hashed only in the final case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 */
1876struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1877{
1878 struct dentry *new = NULL;
1879
Al Viroa9049372011-07-08 21:20:11 -04001880 if (IS_ERR(inode))
1881 return ERR_CAST(inode);
1882
NeilBrown21c0d8f2006-10-04 02:16:16 -07001883 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001884 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001885 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001887 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001888 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 d_move(new, dentry);
1891 iput(inode);
1892 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001893 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001894 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001895 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 security_d_instantiate(dentry, inode);
1897 d_rehash(dentry);
1898 }
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001899 } else {
1900 d_instantiate(dentry, inode);
1901 if (d_unhashed(dentry))
1902 d_rehash(dentry);
1903 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 return new;
1905}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001906EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
Barry Naujok94035402008-05-21 16:50:46 +10001908/**
1909 * d_add_ci - lookup or allocate new dentry with case-exact name
1910 * @inode: the inode case-insensitive lookup has found
1911 * @dentry: the negative dentry that was passed to the parent's lookup func
1912 * @name: the case-exact name to be associated with the returned dentry
1913 *
1914 * This is to avoid filling the dcache with case-insensitive names to the
1915 * same inode, only the actual correct case is stored in the dcache for
1916 * case-insensitive filesystems.
1917 *
1918 * For a case-insensitive lookup match and if the the case-exact dentry
1919 * already exists in in the dcache, use it and return it.
1920 *
1921 * If no entry exists with the exact case name, allocate new dentry with
1922 * the exact case, and return the spliced entry.
1923 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001924struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001925 struct qstr *name)
1926{
Barry Naujok94035402008-05-21 16:50:46 +10001927 struct dentry *found;
1928 struct dentry *new;
1929
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001930 /*
1931 * First check if a dentry matching the name already exists,
1932 * if not go ahead and create it now.
1933 */
Barry Naujok94035402008-05-21 16:50:46 +10001934 found = d_hash_and_lookup(dentry->d_parent, name);
Al Viro4f522a22013-02-11 23:20:37 -05001935 if (unlikely(IS_ERR(found)))
1936 goto err_out;
Barry Naujok94035402008-05-21 16:50:46 +10001937 if (!found) {
1938 new = d_alloc(dentry->d_parent, name);
1939 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05001940 found = ERR_PTR(-ENOMEM);
Barry Naujok94035402008-05-21 16:50:46 +10001941 goto err_out;
1942 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001943
Barry Naujok94035402008-05-21 16:50:46 +10001944 found = d_splice_alias(inode, new);
1945 if (found) {
1946 dput(new);
1947 return found;
1948 }
1949 return new;
1950 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001951
1952 /*
1953 * If a matching dentry exists, and it's not negative use it.
1954 *
1955 * Decrement the reference count to balance the iget() done
1956 * earlier on.
1957 */
Barry Naujok94035402008-05-21 16:50:46 +10001958 if (found->d_inode) {
1959 if (unlikely(found->d_inode != inode)) {
1960 /* This can't happen because bad inodes are unhashed. */
1961 BUG_ON(!is_bad_inode(inode));
1962 BUG_ON(!is_bad_inode(found->d_inode));
1963 }
Barry Naujok94035402008-05-21 16:50:46 +10001964 iput(inode);
1965 return found;
1966 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001967
Barry Naujok94035402008-05-21 16:50:46 +10001968 /*
1969 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001970 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001971 */
Al Viro4513d892011-07-17 10:52:14 -04001972 new = d_splice_alias(inode, found);
1973 if (new) {
1974 dput(found);
1975 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10001976 }
Al Viro4513d892011-07-17 10:52:14 -04001977 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001978
1979err_out:
1980 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05001981 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001982}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001983EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001985/*
1986 * Do the slow-case of the dentry name compare.
1987 *
1988 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07001989 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001990 * filesystem can rely on them, and can use the 'name' and
1991 * 'len' information without worrying about walking off the
1992 * end of memory etc.
1993 *
1994 * Thus the read_seqcount_retry() and the "duplicate" info
1995 * in arguments (the low-level filesystem should not look
1996 * at the dentry inode or name contents directly, since
1997 * rename can change them while we're in RCU mode).
1998 */
1999enum slow_d_compare {
2000 D_COMP_OK,
2001 D_COMP_NOMATCH,
2002 D_COMP_SEQRETRY,
2003};
2004
2005static noinline enum slow_d_compare slow_dentry_cmp(
2006 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002007 struct dentry *dentry,
2008 unsigned int seq,
2009 const struct qstr *name)
2010{
2011 int tlen = dentry->d_name.len;
2012 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002013
2014 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2015 cpu_relax();
2016 return D_COMP_SEQRETRY;
2017 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002018 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002019 return D_COMP_NOMATCH;
2020 return D_COMP_OK;
2021}
2022
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002024 * __d_lookup_rcu - search for a dentry (racy, store-free)
2025 * @parent: parent dentry
2026 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002027 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002028 * Returns: dentry, or NULL
2029 *
2030 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2031 * resolution (store-free path walking) design described in
2032 * Documentation/filesystems/path-lookup.txt.
2033 *
2034 * This is not to be used outside core vfs.
2035 *
2036 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2037 * held, and rcu_read_lock held. The returned dentry must not be stored into
2038 * without taking d_lock and checking d_seq sequence count against @seq
2039 * returned here.
2040 *
Linus Torvalds15570082013-09-02 11:38:06 -07002041 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002042 * function.
2043 *
2044 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2045 * the returned dentry, so long as its parent's seqlock is checked after the
2046 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2047 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002048 *
2049 * NOTE! The caller *has* to check the resulting dentry against the sequence
2050 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002051 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002052struct dentry *__d_lookup_rcu(const struct dentry *parent,
2053 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002054 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002055{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002056 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002057 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002058 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002059 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002060 struct dentry *dentry;
2061
2062 /*
2063 * Note: There is significant duplication with __d_lookup_rcu which is
2064 * required to prevent single threaded performance regressions
2065 * especially on architectures where smp_rmb (in seqcounts) are costly.
2066 * Keep the two functions in sync.
2067 */
2068
2069 /*
2070 * The hash list is protected using RCU.
2071 *
2072 * Carefully use d_seq when comparing a candidate dentry, to avoid
2073 * races with d_move().
2074 *
2075 * It is possible that concurrent renames can mess up our list
2076 * walk here and result in missing our dentry, resulting in the
2077 * false-negative result. d_lookup() protects against concurrent
2078 * renames using rename_lock seqlock.
2079 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002080 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002081 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002082 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002083 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002084
Nick Piggin31e6b012011-01-07 17:49:52 +11002085seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002086 /*
2087 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002088 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002089 *
2090 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002091 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002092 *
2093 * NOTE! We do a "raw" seqcount_begin here. That means that
2094 * we don't wait for the sequence count to stabilize if it
2095 * is in the middle of a sequence change. If we do the slow
2096 * dentry compare, we will do seqretries until it is stable,
2097 * and if we end up with a successful lookup, we actually
2098 * want to exit RCU lookup anyway.
2099 */
2100 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002101 if (dentry->d_parent != parent)
2102 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002103 if (d_unhashed(dentry))
2104 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002105
2106 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002107 if (dentry->d_name.hash != hashlen_hash(hashlen))
2108 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002109 *seqp = seq;
2110 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002111 case D_COMP_OK:
2112 return dentry;
2113 case D_COMP_NOMATCH:
2114 continue;
2115 default:
2116 goto seqretry;
2117 }
2118 }
2119
Linus Torvalds26fe5752012-05-10 13:14:12 -07002120 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002121 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002122 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002123 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002124 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002125 }
2126 return NULL;
2127}
2128
2129/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 * d_lookup - search for a dentry
2131 * @parent: parent dentry
2132 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002133 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002135 * d_lookup searches the children of the parent dentry for the name in
2136 * question. If the dentry is found its reference count is incremented and the
2137 * dentry is returned. The caller must use dput to free the entry when it has
2138 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 */
Al Viroda2d8452013-01-24 18:29:34 -05002140struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141{
Nick Piggin31e6b012011-01-07 17:49:52 +11002142 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002143 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144
2145 do {
2146 seq = read_seqbegin(&rename_lock);
2147 dentry = __d_lookup(parent, name);
2148 if (dentry)
2149 break;
2150 } while (read_seqretry(&rename_lock, seq));
2151 return dentry;
2152}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002153EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154
Nick Piggin31e6b012011-01-07 17:49:52 +11002155/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002156 * __d_lookup - search for a dentry (racy)
2157 * @parent: parent dentry
2158 * @name: qstr of name we wish to find
2159 * Returns: dentry, or NULL
2160 *
2161 * __d_lookup is like d_lookup, however it may (rarely) return a
2162 * false-negative result due to unrelated rename activity.
2163 *
2164 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2165 * however it must be used carefully, eg. with a following d_lookup in
2166 * the case of failure.
2167 *
2168 * __d_lookup callers must be commented.
2169 */
Al Viroa713ca22013-01-24 18:27:00 -05002170struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171{
2172 unsigned int len = name->len;
2173 unsigned int hash = name->hash;
2174 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002175 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002176 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002177 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002178 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179
Nick Pigginb04f7842010-08-18 04:37:34 +10002180 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002181 * Note: There is significant duplication with __d_lookup_rcu which is
2182 * required to prevent single threaded performance regressions
2183 * especially on architectures where smp_rmb (in seqcounts) are costly.
2184 * Keep the two functions in sync.
2185 */
2186
2187 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002188 * The hash list is protected using RCU.
2189 *
2190 * Take d_lock when comparing a candidate dentry, to avoid races
2191 * with d_move().
2192 *
2193 * It is possible that concurrent renames can mess up our list
2194 * walk here and result in missing our dentry, resulting in the
2195 * false-negative result. d_lookup() protects against concurrent
2196 * renames using rename_lock seqlock.
2197 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002198 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002199 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 rcu_read_lock();
2201
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002202 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 if (dentry->d_name.hash != hash)
2205 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206
2207 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 if (dentry->d_parent != parent)
2209 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002210 if (d_unhashed(dentry))
2211 goto next;
2212
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 /*
2214 * It is safe to compare names since d_move() cannot
2215 * change the qstr (protected by d_lock).
2216 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002217 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002218 int tlen = dentry->d_name.len;
2219 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002220 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 goto next;
2222 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002223 if (dentry->d_name.len != len)
2224 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002225 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 goto next;
2227 }
2228
Waiman Long98474232013-08-28 18:24:59 -07002229 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002230 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 spin_unlock(&dentry->d_lock);
2232 break;
2233next:
2234 spin_unlock(&dentry->d_lock);
2235 }
2236 rcu_read_unlock();
2237
2238 return found;
2239}
2240
2241/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002242 * d_hash_and_lookup - hash the qstr then search for a dentry
2243 * @dir: Directory to search in
2244 * @name: qstr of name we wish to find
2245 *
Al Viro4f522a22013-02-11 23:20:37 -05002246 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002247 */
2248struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2249{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002250 /*
2251 * Check for a fs-specific hash function. Note that we must
2252 * calculate the standard hash first, as the d_op->d_hash()
2253 * routine may choose to leave the hash value unchanged.
2254 */
2255 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002256 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002257 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002258 if (unlikely(err < 0))
2259 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002260 }
Al Viro4f522a22013-02-11 23:20:37 -05002261 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002262}
Al Viro4f522a22013-02-11 23:20:37 -05002263EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002264
2265/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002266 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002268 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 *
2270 * An insecure source has sent us a dentry, here we verify it and dget() it.
2271 * This is used by ncpfs in its readdir implementation.
2272 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002273 *
2274 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002276int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277{
Nick Piggin786a5e12011-01-07 17:49:16 +11002278 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002279
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002280 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002281 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2282 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002283 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002284 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002285 spin_unlock(&dentry->d_lock);
2286 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 return 1;
2288 }
2289 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002290 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002291
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 return 0;
2293}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002294EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295
2296/*
2297 * When a file is deleted, we have two options:
2298 * - turn this dentry into a negative dentry
2299 * - unhash this dentry and free it.
2300 *
2301 * Usually, we want to just turn this into
2302 * a negative dentry, but if anybody else is
2303 * currently using the dentry or the inode
2304 * we can't do that and we fall back on removing
2305 * it from the hash queues and waiting for
2306 * it to be deleted later when it has no users
2307 */
2308
2309/**
2310 * d_delete - delete a dentry
2311 * @dentry: The dentry to delete
2312 *
2313 * Turn the dentry into a negative dentry if possible, otherwise
2314 * remove it from the hash queues so it can be deleted later
2315 */
2316
2317void d_delete(struct dentry * dentry)
2318{
Nick Piggin873feea2011-01-07 17:50:06 +11002319 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002320 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 /*
2322 * Are we the only user?
2323 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002324again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002326 inode = dentry->d_inode;
2327 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002328 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002329 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002330 spin_unlock(&dentry->d_lock);
2331 cpu_relax();
2332 goto again;
2333 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002334 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002335 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002336 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 return;
2338 }
2339
2340 if (!d_unhashed(dentry))
2341 __d_drop(dentry);
2342
2343 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002344
2345 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002347EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002349static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002351 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002352 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002353 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002354 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002355 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356}
2357
David Howells770bfad2006-08-22 20:06:07 -04002358static void _d_rehash(struct dentry * entry)
2359{
2360 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2361}
2362
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363/**
2364 * d_rehash - add an entry back to the hash
2365 * @entry: dentry to add to the hash
2366 *
2367 * Adds a dentry to the hash according to its name.
2368 */
2369
2370void d_rehash(struct dentry * entry)
2371{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002373 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002376EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002378/**
2379 * dentry_update_name_case - update case insensitive dentry with a new name
2380 * @dentry: dentry to be updated
2381 * @name: new name
2382 *
2383 * Update a case insensitive dentry with new case of name.
2384 *
2385 * dentry must have been returned by d_lookup with name @name. Old and new
2386 * name lengths must match (ie. no d_compare which allows mismatched name
2387 * lengths).
2388 *
2389 * Parent inode i_mutex must be held over d_lookup and into this call (to
2390 * keep renames and concurrent inserts, and readdir(2) away).
2391 */
2392void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2393{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002394 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002395 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2396
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002397 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002398 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002399 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002400 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002401 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002402}
2403EXPORT_SYMBOL(dentry_update_name_case);
2404
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405static void switch_names(struct dentry *dentry, struct dentry *target)
2406{
2407 if (dname_external(target)) {
2408 if (dname_external(dentry)) {
2409 /*
2410 * Both external: swap the pointers
2411 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002412 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 } else {
2414 /*
2415 * dentry:internal, target:external. Steal target's
2416 * storage and make target internal.
2417 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002418 memcpy(target->d_iname, dentry->d_name.name,
2419 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 dentry->d_name.name = target->d_name.name;
2421 target->d_name.name = target->d_iname;
2422 }
2423 } else {
2424 if (dname_external(dentry)) {
2425 /*
2426 * dentry:external, target:internal. Give dentry's
2427 * storage to target and make dentry internal
2428 */
2429 memcpy(dentry->d_iname, target->d_name.name,
2430 target->d_name.len + 1);
2431 target->d_name.name = dentry->d_name.name;
2432 dentry->d_name.name = dentry->d_iname;
2433 } else {
2434 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002435 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002437 unsigned int i;
2438 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2439 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2440 swap(((long *) &dentry->d_iname)[i],
2441 ((long *) &target->d_iname)[i]);
2442 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 }
2444 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002445 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446}
2447
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002448static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2449{
2450 /*
2451 * XXXX: do we really need to take target->d_lock?
2452 */
2453 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2454 spin_lock(&target->d_parent->d_lock);
2455 else {
2456 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2457 spin_lock(&dentry->d_parent->d_lock);
2458 spin_lock_nested(&target->d_parent->d_lock,
2459 DENTRY_D_LOCK_NESTED);
2460 } else {
2461 spin_lock(&target->d_parent->d_lock);
2462 spin_lock_nested(&dentry->d_parent->d_lock,
2463 DENTRY_D_LOCK_NESTED);
2464 }
2465 }
2466 if (target < dentry) {
2467 spin_lock_nested(&target->d_lock, 2);
2468 spin_lock_nested(&dentry->d_lock, 3);
2469 } else {
2470 spin_lock_nested(&dentry->d_lock, 2);
2471 spin_lock_nested(&target->d_lock, 3);
2472 }
2473}
2474
2475static void dentry_unlock_parents_for_move(struct dentry *dentry,
2476 struct dentry *target)
2477{
2478 if (target->d_parent != dentry->d_parent)
2479 spin_unlock(&dentry->d_parent->d_lock);
2480 if (target->d_parent != target)
2481 spin_unlock(&target->d_parent->d_lock);
2482}
2483
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002485 * When switching names, the actual string doesn't strictly have to
2486 * be preserved in the target - because we're dropping the target
2487 * anyway. As such, we can just do a simple memcpy() to copy over
2488 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002490 * Note that we have to be a lot more careful about getting the hash
2491 * switched - we have to switch the hash value properly even if it
2492 * then no longer matches the actual (corrupted) string of the target.
2493 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002495/*
Al Viro18367502011-07-12 21:42:24 -04002496 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 * @dentry: entry to move
2498 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002499 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 *
2501 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002502 * dcache entries should not be moved in this way. Caller must hold
2503 * rename_lock, the i_mutex of the source and target directories,
2504 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002506static void __d_move(struct dentry *dentry, struct dentry *target,
2507 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 if (!dentry->d_inode)
2510 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2511
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002512 BUG_ON(d_ancestor(dentry, target));
2513 BUG_ON(d_ancestor(target, dentry));
2514
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002515 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516
Nick Piggin31e6b012011-01-07 17:49:52 +11002517 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002518 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002519
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002520 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2521
2522 /*
2523 * Move the dentry to the target hash queue. Don't bother checking
2524 * for the same hash queue because of how unlikely it is.
2525 */
2526 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002527 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528
Miklos Szeredida1ce062014-04-01 17:08:43 +02002529 /*
2530 * Unhash the target (d_delete() is not usable here). If exchanging
2531 * the two dentries, then rehash onto the other's hash queue.
2532 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002534 if (exchange) {
2535 __d_rehash(target,
2536 d_hash(dentry->d_parent, dentry->d_name.hash));
2537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
Eric Dumazet5160ee62006-01-08 01:03:32 -08002539 list_del(&dentry->d_u.d_child);
2540 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541
2542 /* Switch the names.. */
2543 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002544 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
2546 /* ... and switch the parents */
2547 if (IS_ROOT(dentry)) {
2548 dentry->d_parent = target->d_parent;
2549 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002550 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002552 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553
2554 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002555 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 }
2557
Eric Dumazet5160ee62006-01-08 01:03:32 -08002558 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002559
Nick Piggin31e6b012011-01-07 17:49:52 +11002560 write_seqcount_end(&target->d_seq);
2561 write_seqcount_end(&dentry->d_seq);
2562
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002563 dentry_unlock_parents_for_move(dentry, target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002564 if (exchange)
2565 fsnotify_d_move(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002567 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002569}
2570
2571/*
2572 * d_move - move a dentry
2573 * @dentry: entry to move
2574 * @target: new dentry
2575 *
2576 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002577 * dcache entries should not be moved in this way. See the locking
2578 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002579 */
2580void d_move(struct dentry *dentry, struct dentry *target)
2581{
2582 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002583 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002585}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002586EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587
Miklos Szeredida1ce062014-04-01 17:08:43 +02002588/*
2589 * d_exchange - exchange two dentries
2590 * @dentry1: first dentry
2591 * @dentry2: second dentry
2592 */
2593void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2594{
2595 write_seqlock(&rename_lock);
2596
2597 WARN_ON(!dentry1->d_inode);
2598 WARN_ON(!dentry2->d_inode);
2599 WARN_ON(IS_ROOT(dentry1));
2600 WARN_ON(IS_ROOT(dentry2));
2601
2602 __d_move(dentry1, dentry2, true);
2603
2604 write_sequnlock(&rename_lock);
2605}
2606
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002607/**
2608 * d_ancestor - search for an ancestor
2609 * @p1: ancestor dentry
2610 * @p2: child dentry
2611 *
2612 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2613 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002614 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002615struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002616{
2617 struct dentry *p;
2618
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002619 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002620 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002621 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002622 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002623 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002624}
2625
2626/*
2627 * This helper attempts to cope with remotely renamed directories
2628 *
2629 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002630 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002631 *
2632 * Note: If ever the locking in lock_rename() changes, then please
2633 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002634 */
Nick Piggin873feea2011-01-07 17:50:06 +11002635static struct dentry *__d_unalias(struct inode *inode,
2636 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002637{
2638 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002639 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002640
2641 /* If alias and dentry share a parent, then no extra locks required */
2642 if (alias->d_parent == dentry->d_parent)
2643 goto out_unalias;
2644
Trond Myklebust9eaef272006-10-21 10:24:20 -07002645 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002646 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2647 goto out_err;
2648 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2649 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2650 goto out_err;
2651 m2 = &alias->d_parent->d_inode->i_mutex;
2652out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002653 if (likely(!d_mountpoint(alias))) {
Miklos Szeredida1ce062014-04-01 17:08:43 +02002654 __d_move(alias, dentry, false);
Al Viroee3efa92012-06-08 15:59:33 -04002655 ret = alias;
2656 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002657out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002658 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002659 if (m2)
2660 mutex_unlock(m2);
2661 if (m1)
2662 mutex_unlock(m1);
2663 return ret;
2664}
2665
2666/*
David Howells770bfad2006-08-22 20:06:07 -04002667 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2668 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002669 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002670 */
2671static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2672{
Al Viro740da422013-01-30 10:13:38 -05002673 struct dentry *dparent;
David Howells770bfad2006-08-22 20:06:07 -04002674
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002675 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002676
Nick Piggin31e6b012011-01-07 17:49:52 +11002677 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002678 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002679
David Howells770bfad2006-08-22 20:06:07 -04002680 dparent = dentry->d_parent;
David Howells770bfad2006-08-22 20:06:07 -04002681
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002682 switch_names(dentry, anon);
2683 swap(dentry->d_name.hash, anon->d_name.hash);
2684
Al Viro740da422013-01-30 10:13:38 -05002685 dentry->d_parent = dentry;
2686 list_del_init(&dentry->d_u.d_child);
2687 anon->d_parent = dparent;
Wei Yongjun9ed53b12013-03-12 00:10:50 +08002688 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
David Howells770bfad2006-08-22 20:06:07 -04002689
Nick Piggin31e6b012011-01-07 17:49:52 +11002690 write_seqcount_end(&dentry->d_seq);
2691 write_seqcount_end(&anon->d_seq);
2692
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002693 dentry_unlock_parents_for_move(anon, dentry);
2694 spin_unlock(&dentry->d_lock);
2695
2696 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002697}
2698
2699/**
2700 * d_materialise_unique - introduce an inode into the tree
2701 * @dentry: candidate dentry
2702 * @inode: inode to bind to the dentry, to which aliases may be attached
2703 *
2704 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002705 * root directory alias in its place if there is one. Caller must hold the
2706 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002707 */
2708struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2709{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002710 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002711
2712 BUG_ON(!d_unhashed(dentry));
2713
David Howells770bfad2006-08-22 20:06:07 -04002714 if (!inode) {
2715 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002716 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002717 d_rehash(actual);
2718 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002719 }
2720
Nick Piggin873feea2011-01-07 17:50:06 +11002721 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002722
Trond Myklebust9eaef272006-10-21 10:24:20 -07002723 if (S_ISDIR(inode->i_mode)) {
2724 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002725
Trond Myklebust9eaef272006-10-21 10:24:20 -07002726 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002727 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002728 if (alias) {
2729 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002730 write_seqlock(&rename_lock);
2731
2732 if (d_ancestor(alias, dentry)) {
2733 /* Check for loops */
2734 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002735 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002736 } else if (IS_ROOT(alias)) {
2737 /* Is this an anonymous mountpoint that we
2738 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002739 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002740 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002741 __d_drop(alias);
2742 goto found;
Al Viro18367502011-07-12 21:42:24 -04002743 } else {
2744 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002745 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002746 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002747 }
Al Viro18367502011-07-12 21:42:24 -04002748 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002749 if (IS_ERR(actual)) {
2750 if (PTR_ERR(actual) == -ELOOP)
2751 pr_warn_ratelimited(
2752 "VFS: Lookup of '%s' in %s %s"
2753 " would have caused loop\n",
2754 dentry->d_name.name,
2755 inode->i_sb->s_type->name,
2756 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002757 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002758 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002759 goto out_nolock;
2760 }
David Howells770bfad2006-08-22 20:06:07 -04002761 }
2762
2763 /* Add a unique reference */
2764 actual = __d_instantiate_unique(dentry, inode);
2765 if (!actual)
2766 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002767 else
2768 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002769
David Howells770bfad2006-08-22 20:06:07 -04002770 spin_lock(&actual->d_lock);
2771found:
2772 _d_rehash(actual);
2773 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002774 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002775out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002776 if (actual == dentry) {
2777 security_d_instantiate(dentry, inode);
2778 return NULL;
2779 }
2780
2781 iput(inode);
2782 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002783}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002784EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002785
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002786static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002787{
2788 *buflen -= namelen;
2789 if (*buflen < 0)
2790 return -ENAMETOOLONG;
2791 *buffer -= namelen;
2792 memcpy(*buffer, str, namelen);
2793 return 0;
2794}
2795
Waiman Long232d2d62013-09-09 12:18:13 -04002796/**
2797 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002798 * @buffer: buffer pointer
2799 * @buflen: allocated length of the buffer
2800 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002801 *
2802 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2803 * make sure that either the old or the new name pointer and length are
2804 * fetched. However, there may be mismatch between length and pointer.
2805 * The length cannot be trusted, we need to copy it byte-by-byte until
2806 * the length is reached or a null byte is found. It also prepends "/" at
2807 * the beginning of the name. The sequence number check at the caller will
2808 * retry it again when a d_move() does happen. So any garbage in the buffer
2809 * due to mismatched pointer and length will be discarded.
2810 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002811static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2812{
Waiman Long232d2d62013-09-09 12:18:13 -04002813 const char *dname = ACCESS_ONCE(name->name);
2814 u32 dlen = ACCESS_ONCE(name->len);
2815 char *p;
2816
Waiman Long232d2d62013-09-09 12:18:13 -04002817 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04002818 if (*buflen < 0)
2819 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04002820 p = *buffer -= dlen + 1;
2821 *p++ = '/';
2822 while (dlen--) {
2823 char c = *dname++;
2824 if (!c)
2825 break;
2826 *p++ = c;
2827 }
2828 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002829}
2830
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002832 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002833 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002834 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002835 * @buffer: pointer to the end of the buffer
2836 * @buflen: pointer to buffer length
2837 *
Waiman Long18129972013-09-12 10:55:35 -04002838 * The function will first try to write out the pathname without taking any
2839 * lock other than the RCU read lock to make sure that dentries won't go away.
2840 * It only checks the sequence number of the global rename_lock as any change
2841 * in the dentry's d_seq will be preceded by changes in the rename_lock
2842 * sequence number. If the sequence number had been changed, it will restart
2843 * the whole pathname back-tracing sequence again by taking the rename_lock.
2844 * In this case, there is no need to take the RCU read lock as the recursive
2845 * parent pointer references will keep the dentry chain alive as long as no
2846 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002847 */
Al Viro02125a82011-12-05 08:43:34 -05002848static int prepend_path(const struct path *path,
2849 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002850 char **buffer, int *buflen)
2851{
Al Viroede4ceb2013-11-13 07:45:40 -05002852 struct dentry *dentry;
2853 struct vfsmount *vfsmnt;
2854 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002855 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002856 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002857 char *bptr;
2858 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002859
Al Viro48f5ec22013-09-09 15:22:25 -04002860 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002861restart_mnt:
2862 read_seqbegin_or_lock(&mount_lock, &m_seq);
2863 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002864 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002865restart:
2866 bptr = *buffer;
2867 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002868 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002869 dentry = path->dentry;
2870 vfsmnt = path->mnt;
2871 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002872 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002873 while (dentry != root->dentry || vfsmnt != root->mnt) {
2874 struct dentry * parent;
2875
2876 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002877 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002878 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002879 if (mnt != parent) {
2880 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2881 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002882 vfsmnt = &mnt->mnt;
2883 continue;
2884 }
2885 /*
2886 * Filesystems needing to implement special "root names"
2887 * should do so with ->d_dname()
2888 */
2889 if (IS_ROOT(dentry) &&
2890 (dentry->d_name.len != 1 ||
2891 dentry->d_name.name[0] != '/')) {
2892 WARN(1, "Root dentry has weird name <%.*s>\n",
2893 (int) dentry->d_name.len,
2894 dentry->d_name.name);
2895 }
2896 if (!error)
2897 error = is_mounted(vfsmnt) ? 1 : 2;
2898 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002899 }
2900 parent = dentry->d_parent;
2901 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002902 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002903 if (error)
2904 break;
2905
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002906 dentry = parent;
2907 }
Al Viro48f5ec22013-09-09 15:22:25 -04002908 if (!(seq & 1))
2909 rcu_read_unlock();
2910 if (need_seqretry(&rename_lock, seq)) {
2911 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002912 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002913 }
Al Viro48f5ec22013-09-09 15:22:25 -04002914 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002915
2916 if (!(m_seq & 1))
2917 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002918 if (need_seqretry(&mount_lock, m_seq)) {
2919 m_seq = 1;
2920 goto restart_mnt;
2921 }
2922 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002923
Waiman Long232d2d62013-09-09 12:18:13 -04002924 if (error >= 0 && bptr == *buffer) {
2925 if (--blen < 0)
2926 error = -ENAMETOOLONG;
2927 else
2928 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002929 }
Waiman Long232d2d62013-09-09 12:18:13 -04002930 *buffer = bptr;
2931 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002932 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002933}
2934
2935/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002936 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002937 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002938 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002939 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940 * @buflen: buffer length
2941 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002942 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002944 * Returns a pointer into the buffer or an error code if the
2945 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002946 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002947 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002948 *
Al Viro02125a82011-12-05 08:43:34 -05002949 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 */
Al Viro02125a82011-12-05 08:43:34 -05002951char *__d_path(const struct path *path,
2952 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002953 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002955 char *res = buf + buflen;
2956 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002957
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002958 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002959 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002960
Al Viro02125a82011-12-05 08:43:34 -05002961 if (error < 0)
2962 return ERR_PTR(error);
2963 if (error > 0)
2964 return NULL;
2965 return res;
2966}
2967
2968char *d_absolute_path(const struct path *path,
2969 char *buf, int buflen)
2970{
2971 struct path root = {};
2972 char *res = buf + buflen;
2973 int error;
2974
2975 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002976 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05002977
2978 if (error > 1)
2979 error = -EINVAL;
2980 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002981 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002982 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983}
2984
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002985/*
2986 * same as __d_path but appends "(deleted)" for unlinked files.
2987 */
Al Viro02125a82011-12-05 08:43:34 -05002988static int path_with_deleted(const struct path *path,
2989 const struct path *root,
2990 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002991{
2992 prepend(buf, buflen, "\0", 1);
2993 if (d_unlinked(path->dentry)) {
2994 int error = prepend(buf, buflen, " (deleted)", 10);
2995 if (error)
2996 return error;
2997 }
2998
2999 return prepend_path(path, root, buf, buflen);
3000}
3001
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003002static int prepend_unreachable(char **buffer, int *buflen)
3003{
3004 return prepend(buffer, buflen, "(unreachable)", 13);
3005}
3006
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003007static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3008{
3009 unsigned seq;
3010
3011 do {
3012 seq = read_seqcount_begin(&fs->seq);
3013 *root = fs->root;
3014 } while (read_seqcount_retry(&fs->seq, seq));
3015}
3016
Jan Bluncka03a8a702008-02-14 19:38:32 -08003017/**
3018 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003019 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003020 * @buf: buffer to return value in
3021 * @buflen: buffer length
3022 *
3023 * Convert a dentry into an ASCII path name. If the entry has been deleted
3024 * the string " (deleted)" is appended. Note that this is ambiguous.
3025 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003026 * Returns a pointer into the buffer or an error code if the path was
3027 * too long. Note: Callers should use the returned pointer, not the passed
3028 * in buffer, to use the name! The implementation often starts at an offset
3029 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003030 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003031 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003032 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003033char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003035 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003036 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003037 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003039 /*
3040 * We have various synthetic filesystems that never get mounted. On
3041 * these filesystems dentries are never used for lookup purposes, and
3042 * thus don't need to be hashed. They also don't need a name until a
3043 * user wants to identify the object in /proc/pid/fd/. The little hack
3044 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003045 *
3046 * Some pseudo inodes are mountable. When they are mounted
3047 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3048 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003049 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003050 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3051 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003052 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003053
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003054 rcu_read_lock();
3055 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003056 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003057 rcu_read_unlock();
3058
Al Viro02125a82011-12-05 08:43:34 -05003059 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003060 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 return res;
3062}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003063EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064
3065/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003066 * Helper function for dentry_operations.d_dname() members
3067 */
3068char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3069 const char *fmt, ...)
3070{
3071 va_list args;
3072 char temp[64];
3073 int sz;
3074
3075 va_start(args, fmt);
3076 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3077 va_end(args);
3078
3079 if (sz > sizeof(temp) || sz > buflen)
3080 return ERR_PTR(-ENAMETOOLONG);
3081
3082 buffer += buflen - sz;
3083 return memcpy(buffer, temp, sz);
3084}
3085
Al Viro118b2302013-08-24 12:08:17 -04003086char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3087{
3088 char *end = buffer + buflen;
3089 /* these dentries are never renamed, so d_lock is not needed */
3090 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003091 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003092 prepend(&end, &buflen, "/", 1))
3093 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003094 return end;
Al Viro118b2302013-08-24 12:08:17 -04003095}
David Herrmann31bbe162014-01-03 14:09:47 +01003096EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003097
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003098/*
Ram Pai6092d042008-03-27 13:06:20 +01003099 * Write full pathname from the root of the filesystem into the buffer.
3100 */
Al Virof6500802014-01-26 12:37:55 -05003101static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003102{
Al Virof6500802014-01-26 12:37:55 -05003103 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003104 char *end, *retval;
3105 int len, seq = 0;
3106 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003107
Al Virof6500802014-01-26 12:37:55 -05003108 if (buflen < 2)
3109 goto Elong;
3110
Al Viro48f5ec22013-09-09 15:22:25 -04003111 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003112restart:
Al Virof6500802014-01-26 12:37:55 -05003113 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003114 end = buf + buflen;
3115 len = buflen;
3116 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003117 /* Get '/' right */
3118 retval = end-1;
3119 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003120 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003121 while (!IS_ROOT(dentry)) {
3122 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003123
Ram Pai6092d042008-03-27 13:06:20 +01003124 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003125 error = prepend_name(&end, &len, &dentry->d_name);
3126 if (error)
3127 break;
Ram Pai6092d042008-03-27 13:06:20 +01003128
3129 retval = end;
3130 dentry = parent;
3131 }
Al Viro48f5ec22013-09-09 15:22:25 -04003132 if (!(seq & 1))
3133 rcu_read_unlock();
3134 if (need_seqretry(&rename_lock, seq)) {
3135 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003136 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003137 }
3138 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003139 if (error)
3140 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003141 return retval;
3142Elong:
3143 return ERR_PTR(-ENAMETOOLONG);
3144}
Nick Pigginec2447c2011-01-07 17:49:29 +11003145
3146char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3147{
Waiman Long232d2d62013-09-09 12:18:13 -04003148 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003149}
3150EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003151
3152char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3153{
3154 char *p = NULL;
3155 char *retval;
3156
Al Viroc1031352010-06-06 22:31:14 -04003157 if (d_unlinked(dentry)) {
3158 p = buf + buflen;
3159 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3160 goto Elong;
3161 buflen++;
3162 }
3163 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003164 if (!IS_ERR(retval) && p)
3165 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003166 return retval;
3167Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003168 return ERR_PTR(-ENAMETOOLONG);
3169}
3170
Linus Torvalds8b19e342013-09-12 10:35:47 -07003171static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3172 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003173{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003174 unsigned seq;
3175
3176 do {
3177 seq = read_seqcount_begin(&fs->seq);
3178 *root = fs->root;
3179 *pwd = fs->pwd;
3180 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003181}
3182
Ram Pai6092d042008-03-27 13:06:20 +01003183/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 * NOTE! The user-level library version returns a
3185 * character pointer. The kernel system call just
3186 * returns the length of the buffer filled (which
3187 * includes the ending '\0' character), or a negative
3188 * error value. So libc would do something like
3189 *
3190 * char *getcwd(char * buf, size_t size)
3191 * {
3192 * int retval;
3193 *
3194 * retval = sys_getcwd(buf, size);
3195 * if (retval >= 0)
3196 * return buf;
3197 * errno = -retval;
3198 * return NULL;
3199 * }
3200 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003201SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202{
Linus Torvalds552ce542007-02-13 12:08:18 -08003203 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003204 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003205 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206
3207 if (!page)
3208 return -ENOMEM;
3209
Linus Torvalds8b19e342013-09-12 10:35:47 -07003210 rcu_read_lock();
3211 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212
Linus Torvalds552ce542007-02-13 12:08:18 -08003213 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003214 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003215 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003216 char *cwd = page + PATH_MAX;
3217 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003219 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003220 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003221 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003222
Al Viro02125a82011-12-05 08:43:34 -05003223 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003224 goto out;
3225
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003226 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003227 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003228 error = prepend_unreachable(&cwd, &buflen);
3229 if (error)
3230 goto out;
3231 }
3232
Linus Torvalds552ce542007-02-13 12:08:18 -08003233 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003234 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003235 if (len <= size) {
3236 error = len;
3237 if (copy_to_user(buf, cwd, len))
3238 error = -EFAULT;
3239 }
Nick Piggin949854d2011-01-07 17:49:37 +11003240 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003241 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003242 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003243
3244out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003245 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246 return error;
3247}
3248
3249/*
3250 * Test whether new_dentry is a subdirectory of old_dentry.
3251 *
3252 * Trivially implemented using the dcache structure
3253 */
3254
3255/**
3256 * is_subdir - is new dentry a subdirectory of old_dentry
3257 * @new_dentry: new dentry
3258 * @old_dentry: old dentry
3259 *
3260 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3261 * Returns 0 otherwise.
3262 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3263 */
3264
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003265int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266{
3267 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11003268 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003270 if (new_dentry == old_dentry)
3271 return 1;
3272
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003273 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003276 /*
3277 * Need rcu_readlock to protect against the d_parent trashing
3278 * due to d_move
3279 */
3280 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003281 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003283 else
3284 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11003285 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003287
3288 return result;
3289}
3290
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003291static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003293 struct dentry *root = data;
3294 if (dentry != root) {
3295 if (d_unhashed(dentry) || !dentry->d_inode)
3296 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003298 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3299 dentry->d_flags |= DCACHE_GENOCIDE;
3300 dentry->d_lockref.count--;
3301 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003303 return D_WALK_CONTINUE;
3304}
Nick Piggin58db63d2011-01-07 17:49:39 +11003305
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003306void d_genocide(struct dentry *parent)
3307{
3308 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003309}
3310
Al Viro60545d02013-06-07 01:20:27 -04003311void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312{
Al Viro60545d02013-06-07 01:20:27 -04003313 inode_dec_link_count(inode);
3314 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3315 !hlist_unhashed(&dentry->d_alias) ||
3316 !d_unlinked(dentry));
3317 spin_lock(&dentry->d_parent->d_lock);
3318 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3319 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3320 (unsigned long long)inode->i_ino);
3321 spin_unlock(&dentry->d_lock);
3322 spin_unlock(&dentry->d_parent->d_lock);
3323 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324}
Al Viro60545d02013-06-07 01:20:27 -04003325EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003326
3327static __initdata unsigned long dhash_entries;
3328static int __init set_dhash_entries(char *str)
3329{
3330 if (!str)
3331 return 0;
3332 dhash_entries = simple_strtoul(str, &str, 0);
3333 return 1;
3334}
3335__setup("dhash_entries=", set_dhash_entries);
3336
3337static void __init dcache_init_early(void)
3338{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003339 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340
3341 /* If hashes are distributed across NUMA nodes, defer
3342 * hash allocation until vmalloc space is available.
3343 */
3344 if (hashdist)
3345 return;
3346
3347 dentry_hashtable =
3348 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003349 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 dhash_entries,
3351 13,
3352 HASH_EARLY,
3353 &d_hash_shift,
3354 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003355 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356 0);
3357
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003358 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003359 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360}
3361
Denis Cheng74bf17c2007-10-16 23:26:30 -07003362static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003364 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365
3366 /*
3367 * A constructor could be added for stable state like the lists,
3368 * but it is probably not worth it because of the cache nature
3369 * of the dcache.
3370 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003371 dentry_cache = KMEM_CACHE(dentry,
3372 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373
3374 /* Hash may have been set up in dcache_init_early */
3375 if (!hashdist)
3376 return;
3377
3378 dentry_hashtable =
3379 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003380 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381 dhash_entries,
3382 13,
3383 0,
3384 &d_hash_shift,
3385 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003386 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387 0);
3388
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003389 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003390 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391}
3392
3393/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003394struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003395EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397EXPORT_SYMBOL(d_genocide);
3398
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399void __init vfs_caches_init_early(void)
3400{
3401 dcache_init_early();
3402 inode_init_early();
3403}
3404
3405void __init vfs_caches_init(unsigned long mempages)
3406{
3407 unsigned long reserve;
3408
3409 /* Base hash sizes on available memory, with a reserve equal to
3410 150% of current kernel size */
3411
3412 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3413 mempages -= reserve;
3414
3415 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003416 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417
Denis Cheng74bf17c2007-10-16 23:26:30 -07003418 dcache_init();
3419 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003421 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422 bdev_cache_init();
3423 chrdev_init();
3424}