blob: 362396ae5650ea81c62e68bfd1bce61a55df7c87 [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>
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -080041#include <linux/kasan.h>
42
David Howells07f3f052006-09-30 20:52:18 +020043#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050044#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Nick Piggin789680d2011-01-07 17:49:30 +110046/*
47 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110048 * dcache->d_inode->i_lock protects:
Al Viro946e51f2014-10-26 19:19:16 -040049 * - i_dentry, d_u.d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110050 * dcache_hash_bucket lock protects:
51 * - the dcache hash table
52 * s_anon bl list spinlock protects:
53 * - the s_anon list (see __d_drop)
Dave Chinner19156842013-08-28 10:17:55 +100054 * dentry->d_sb->s_dentry_lru_lock protects:
Nick Piggin23044502011-01-07 17:49:31 +110055 * - the dcache lru lists and counters
56 * d_lock protects:
57 * - d_flags
58 * - d_name
59 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110060 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110061 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110062 * - d_parent and d_subdirs
63 * - childrens' d_child and d_parent
Al Viro946e51f2014-10-26 19:19:16 -040064 * - d_u.d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110065 *
66 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110067 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110068 * dentry->d_lock
Dave Chinner19156842013-08-28 10:17:55 +100069 * dentry->d_sb->s_dentry_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110070 * dcache_hash_bucket lock
71 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110072 *
Nick Pigginda502952011-01-07 17:49:33 +110073 * If there is an ancestor relationship:
74 * dentry->d_parent->...->d_parent->d_lock
75 * ...
76 * dentry->d_parent->d_lock
77 * dentry->d_lock
78 *
79 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110080 * if (dentry1 < dentry2)
81 * dentry1->d_lock
82 * dentry2->d_lock
83 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080084int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
86
Al Viro74c3cbe2007-07-22 08:04:18 -040087__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Nick Piggin949854d2011-01-07 17:49:37 +110089EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Christoph Lametere18b8902006-12-06 20:33:20 -080091static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Linus Torvalds1da177e2005-04-16 15:20:36 -070093/*
94 * This is the single most critical data structure when it comes
95 * to the dcache: the hashtable for lookups. Somebody should try
96 * to make this good - I've just made it work.
97 *
98 * This hash-function tries to avoid losing too many bits of hash
99 * information, yet avoid using a prime hash-size or similar.
100 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800102static unsigned int d_hash_mask __read_mostly;
103static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700105static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100106
Linus Torvalds8387ff22016-06-10 07:51:30 -0700107static inline struct hlist_bl_head *d_hash(unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100108{
George Spelvin703b5fa2016-06-10 00:22:12 -0400109 return dentry_hashtable + (hash >> (32 - d_hash_shift));
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100110}
111
Al Viro94bdd652016-04-15 02:42:04 -0400112#define IN_LOOKUP_SHIFT 10
113static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
114
115static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
116 unsigned int hash)
117{
118 hash += (unsigned long) parent / L1_CACHE_BYTES;
119 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
120}
121
122
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123/* Statistics gathering. */
124struct dentry_stat_t dentry_stat = {
125 .age_limit = 45,
126};
127
Glauber Costa3942c072013-08-28 10:17:53 +1000128static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000129static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400130
131#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000132
133/*
134 * Here we resort to our own counters instead of using generic per-cpu counters
135 * for consistency with what the vfs inode code does. We are expected to harvest
136 * better code and performance by having our own specialized counters.
137 *
138 * Please note that the loop is done over all possible CPUs, not over all online
139 * CPUs. The reason for this is that we don't want to play games with CPUs going
140 * on and off. If one of them goes off, we will just keep their counters.
141 *
142 * glommer: See cffbc8a for details, and if you ever intend to change this,
143 * please update all vfs counters to match.
144 */
Glauber Costa3942c072013-08-28 10:17:53 +1000145static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100146{
147 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000148 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100149 for_each_possible_cpu(i)
150 sum += per_cpu(nr_dentry, i);
151 return sum < 0 ? 0 : sum;
152}
153
Dave Chinner62d36c72013-08-28 10:17:54 +1000154static long get_nr_dentry_unused(void)
155{
156 int i;
157 long sum = 0;
158 for_each_possible_cpu(i)
159 sum += per_cpu(nr_dentry_unused, i);
160 return sum < 0 ? 0 : sum;
161}
162
Joe Perches1f7e0612014-06-06 14:38:05 -0700163int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400164 size_t *lenp, loff_t *ppos)
165{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100166 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000167 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000168 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400169}
170#endif
171
Linus Torvalds5483f182012-03-04 15:51:42 -0800172/*
173 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
174 * The strings are both count bytes long, and count is non-zero.
175 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700176#ifdef CONFIG_DCACHE_WORD_ACCESS
177
178#include <asm/word-at-a-time.h>
179/*
180 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
181 * aligned allocation for this particular component. We don't
182 * strictly need the load_unaligned_zeropad() safety, but it
183 * doesn't hurt either.
184 *
185 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
186 * need the careful unaligned handling.
187 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700188static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800189{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800190 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800191
192 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700193 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700194 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800195 if (tcount < sizeof(unsigned long))
196 break;
197 if (unlikely(a != b))
198 return 1;
199 cs += sizeof(unsigned long);
200 ct += sizeof(unsigned long);
201 tcount -= sizeof(unsigned long);
202 if (!tcount)
203 return 0;
204 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000205 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800206 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700207}
208
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800209#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700210
Linus Torvalds94753db52012-05-10 12:19:19 -0700211static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700212{
Linus Torvalds5483f182012-03-04 15:51:42 -0800213 do {
214 if (*cs != *ct)
215 return 1;
216 cs++;
217 ct++;
218 tcount--;
219 } while (tcount);
220 return 0;
221}
222
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700223#endif
224
Linus Torvalds94753db52012-05-10 12:19:19 -0700225static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
226{
Linus Torvalds94753db52012-05-10 12:19:19 -0700227 /*
228 * Be careful about RCU walk racing with rename:
He Kuangae0a8432016-03-26 09:12:10 +0000229 * use 'lockless_dereference' to fetch the name pointer.
Linus Torvalds94753db52012-05-10 12:19:19 -0700230 *
231 * NOTE! Even if a rename will mean that the length
232 * was not loaded atomically, we don't care. The
233 * RCU walk will check the sequence count eventually,
234 * and catch it. And we won't overrun the buffer,
235 * because we're reading the name pointer atomically,
236 * and a dentry name is guaranteed to be properly
237 * terminated with a NUL byte.
238 *
239 * End result: even if 'len' is wrong, we'll exit
240 * early because the data cannot match (there can
241 * be no NUL in the ct/tcount data)
242 */
He Kuangae0a8432016-03-26 09:12:10 +0000243 const unsigned char *cs = lockless_dereference(dentry->d_name.name);
244
Linus Torvalds6326c712012-05-21 16:14:04 -0700245 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700246}
247
Al Viro8d85b482014-09-29 14:54:27 -0400248struct external_name {
249 union {
250 atomic_t count;
251 struct rcu_head head;
252 } u;
253 unsigned char name[];
254};
255
256static inline struct external_name *external_name(struct dentry *dentry)
257{
258 return container_of(dentry->d_name.name, struct external_name, name[0]);
259}
260
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400261static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400263 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
264
Al Viro8d85b482014-09-29 14:54:27 -0400265 kmem_cache_free(dentry_cache, dentry);
266}
267
268static void __d_free_external(struct rcu_head *head)
269{
270 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
Al Viro8d85b482014-09-29 14:54:27 -0400271 kfree(external_name(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 kmem_cache_free(dentry_cache, dentry);
273}
274
Al Viro810bb172014-10-12 12:45:37 -0400275static inline int dname_external(const struct dentry *dentry)
276{
277 return dentry->d_name.name != dentry->d_iname;
278}
279
David Howells4bf46a22015-03-05 14:09:22 +0000280static inline void __d_set_inode_and_type(struct dentry *dentry,
281 struct inode *inode,
282 unsigned type_flags)
283{
284 unsigned flags;
285
286 dentry->d_inode = inode;
David Howells4bf46a22015-03-05 14:09:22 +0000287 flags = READ_ONCE(dentry->d_flags);
288 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
289 flags |= type_flags;
290 WRITE_ONCE(dentry->d_flags, flags);
291}
292
David Howells4bf46a22015-03-05 14:09:22 +0000293static inline void __d_clear_type_and_inode(struct dentry *dentry)
294{
295 unsigned flags = READ_ONCE(dentry->d_flags);
296
297 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
298 WRITE_ONCE(dentry->d_flags, flags);
David Howells4bf46a22015-03-05 14:09:22 +0000299 dentry->d_inode = NULL;
300}
301
Al Virob4f03542014-04-29 23:40:14 -0400302static void dentry_free(struct dentry *dentry)
303{
Al Viro946e51f2014-10-26 19:19:16 -0400304 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400305 if (unlikely(dname_external(dentry))) {
306 struct external_name *p = external_name(dentry);
307 if (likely(atomic_dec_and_test(&p->u.count))) {
308 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
309 return;
310 }
311 }
Al Virob4f03542014-04-29 23:40:14 -0400312 /* if dentry was never visible to RCU, immediate free is OK */
313 if (!(dentry->d_flags & DCACHE_RCUACCESS))
314 __d_free(&dentry->d_u.d_rcu);
315 else
316 call_rcu(&dentry->d_u.d_rcu, __d_free);
317}
318
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319/*
320 * Release the dentry's inode, using the filesystem
Al Viro550dce02016-05-29 20:13:30 -0400321 * d_iput() operation if defined.
Nick Piggin31e6b012011-01-07 17:49:52 +1100322 */
323static void dentry_unlink_inode(struct dentry * dentry)
324 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100325 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100326{
327 struct inode *inode = dentry->d_inode;
Al Viro550dce02016-05-29 20:13:30 -0400328 bool hashed = !d_unhashed(dentry);
Al Viroa528aca2016-02-29 12:12:46 -0500329
Al Viro550dce02016-05-29 20:13:30 -0400330 if (hashed)
331 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +0000332 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400333 hlist_del_init(&dentry->d_u.d_alias);
Al Viro550dce02016-05-29 20:13:30 -0400334 if (hashed)
335 raw_write_seqcount_end(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100336 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100337 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100338 if (!inode->i_nlink)
339 fsnotify_inoderemove(inode);
340 if (dentry->d_op && dentry->d_op->d_iput)
341 dentry->d_op->d_iput(dentry, inode);
342 else
343 iput(inode);
344}
345
346/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400347 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
348 * is in use - which includes both the "real" per-superblock
349 * LRU list _and_ the DCACHE_SHRINK_LIST use.
350 *
351 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
352 * on the shrink list (ie not on the superblock LRU list).
353 *
354 * The per-cpu "nr_dentry_unused" counters are updated with
355 * the DCACHE_LRU_LIST bit.
356 *
357 * These helper functions make sure we always follow the
358 * rules. d_lock must be held by the caller.
359 */
360#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
361static void d_lru_add(struct dentry *dentry)
362{
363 D_FLAG_VERIFY(dentry, 0);
364 dentry->d_flags |= DCACHE_LRU_LIST;
365 this_cpu_inc(nr_dentry_unused);
366 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
367}
368
369static void d_lru_del(struct dentry *dentry)
370{
371 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
372 dentry->d_flags &= ~DCACHE_LRU_LIST;
373 this_cpu_dec(nr_dentry_unused);
374 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
375}
376
377static void d_shrink_del(struct dentry *dentry)
378{
379 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
380 list_del_init(&dentry->d_lru);
381 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
382 this_cpu_dec(nr_dentry_unused);
383}
384
385static void d_shrink_add(struct dentry *dentry, struct list_head *list)
386{
387 D_FLAG_VERIFY(dentry, 0);
388 list_add(&dentry->d_lru, list);
389 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
390 this_cpu_inc(nr_dentry_unused);
391}
392
393/*
394 * These can only be called under the global LRU lock, ie during the
395 * callback for freeing the LRU list. "isolate" removes it from the
396 * LRU lists entirely, while shrink_move moves it to the indicated
397 * private list.
398 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800399static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400400{
401 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
402 dentry->d_flags &= ~DCACHE_LRU_LIST;
403 this_cpu_dec(nr_dentry_unused);
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800404 list_lru_isolate(lru, &dentry->d_lru);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400405}
406
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800407static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
408 struct list_head *list)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400409{
410 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
411 dentry->d_flags |= DCACHE_SHRINK_LIST;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800412 list_lru_isolate_move(lru, &dentry->d_lru, list);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400413}
414
415/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000416 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700417 */
418static void dentry_lru_add(struct dentry *dentry)
419{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400420 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
421 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700422}
423
Miklos Szeredid52b9082007-05-08 00:23:46 -0700424/**
Nick Piggin789680d2011-01-07 17:49:30 +1100425 * d_drop - drop a dentry
426 * @dentry: dentry to drop
427 *
428 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
429 * be found through a VFS lookup any more. Note that this is different from
430 * deleting the dentry - d_delete will try to mark the dentry negative if
431 * possible, giving a successful _negative_ lookup, while d_drop will
432 * just make the cache lookup fail.
433 *
434 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
435 * reason (NFS timeouts or autofs deletes).
436 *
437 * __d_drop requires dentry->d_lock.
438 */
439void __d_drop(struct dentry *dentry)
440{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700441 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400442 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400443 /*
444 * Hashed dentries are normally on the dentry hashtable,
445 * with the exception of those newly allocated by
446 * d_obtain_alias, which are always IS_ROOT:
447 */
448 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400449 b = &dentry->d_sb->s_anon;
450 else
Linus Torvalds8387ff22016-06-10 07:51:30 -0700451 b = d_hash(dentry->d_name.hash);
Al Virob61625d2013-10-04 11:09:01 -0400452
453 hlist_bl_lock(b);
454 __hlist_bl_del(&dentry->d_hash);
455 dentry->d_hash.pprev = NULL;
456 hlist_bl_unlock(b);
Al Virod6141462016-07-28 13:05:50 -0400457 /* After this call, in-progress rcu-walk path lookup will fail. */
458 write_seqcount_invalidate(&dentry->d_seq);
Nick Piggin789680d2011-01-07 17:49:30 +1100459 }
460}
461EXPORT_SYMBOL(__d_drop);
462
463void d_drop(struct dentry *dentry)
464{
Nick Piggin789680d2011-01-07 17:49:30 +1100465 spin_lock(&dentry->d_lock);
466 __d_drop(dentry);
467 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100468}
469EXPORT_SYMBOL(d_drop);
470
Al Viroba65dc52016-06-10 11:32:47 -0400471static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
472{
473 struct dentry *next;
474 /*
475 * Inform d_walk() and shrink_dentry_list() that we are no longer
476 * attached to the dentry tree
477 */
478 dentry->d_flags |= DCACHE_DENTRY_KILLED;
479 if (unlikely(list_empty(&dentry->d_child)))
480 return;
481 __list_del_entry(&dentry->d_child);
482 /*
483 * Cursors can move around the list of children. While we'd been
484 * a normal list member, it didn't matter - ->d_child.next would've
485 * been updated. However, from now on it won't be and for the
486 * things like d_walk() it might end up with a nasty surprise.
487 * Normally d_walk() doesn't care about cursors moving around -
488 * ->d_lock on parent prevents that and since a cursor has no children
489 * of its own, we get through it without ever unlocking the parent.
490 * There is one exception, though - if we ascend from a child that
491 * gets killed as soon as we unlock it, the next sibling is found
492 * using the value left in its ->d_child.next. And if _that_
493 * pointed to a cursor, and cursor got moved (e.g. by lseek())
494 * before d_walk() regains parent->d_lock, we'll end up skipping
495 * everything the cursor had been moved past.
496 *
497 * Solution: make sure that the pointer left behind in ->d_child.next
498 * points to something that won't be moving around. I.e. skip the
499 * cursors.
500 */
501 while (dentry->d_child.next != &parent->d_subdirs) {
502 next = list_entry(dentry->d_child.next, struct dentry, d_child);
503 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
504 break;
505 dentry->d_child.next = next->d_child.next;
506 }
507}
508
Al Viroe55fd012014-05-28 13:51:12 -0400509static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100510{
Al Viro41edf272014-05-01 10:30:00 -0400511 struct dentry *parent = NULL;
512 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400513 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100514 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100515
Linus Torvalds0d984392013-09-08 13:46:52 -0700516 /*
517 * The dentry is now unrecoverably dead to the world.
518 */
519 lockref_mark_dead(&dentry->d_lockref);
520
Sage Weilf0023bc2011-10-28 10:02:42 -0700521 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700522 * inform the fs via d_prune that this dentry is about to be
523 * unhashed and destroyed.
524 */
Al Viro29266202014-05-30 11:39:02 -0400525 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800526 dentry->d_op->d_prune(dentry);
527
Al Viro01b60352014-04-29 23:42:52 -0400528 if (dentry->d_flags & DCACHE_LRU_LIST) {
529 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
530 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400531 }
Nick Piggin77812a12011-01-07 17:49:48 +1100532 /* if it was on the hash then remove it */
533 __d_drop(dentry);
Al Viroba65dc52016-06-10 11:32:47 -0400534 dentry_unlist(dentry, parent);
Al Viro03b3b882014-04-29 15:45:28 -0400535 if (parent)
536 spin_unlock(&parent->d_lock);
Al Viro550dce02016-05-29 20:13:30 -0400537 if (dentry->d_inode)
538 dentry_unlink_inode(dentry);
539 else
540 spin_unlock(&dentry->d_lock);
Al Viro03b3b882014-04-29 15:45:28 -0400541 this_cpu_dec(nr_dentry);
542 if (dentry->d_op && dentry->d_op->d_release)
543 dentry->d_op->d_release(dentry);
544
Al Viro41edf272014-05-01 10:30:00 -0400545 spin_lock(&dentry->d_lock);
546 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
547 dentry->d_flags |= DCACHE_MAY_FREE;
548 can_free = false;
549 }
550 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400551 if (likely(can_free))
552 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400553}
554
555/*
556 * Finish off a dentry we've decided to kill.
557 * dentry->d_lock must be held, returns with it unlocked.
558 * If ref is non-zero, then decrement the refcount too.
559 * Returns dentry requiring refcount drop, or NULL if we're done.
560 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400561static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400562 __releases(dentry->d_lock)
563{
564 struct inode *inode = dentry->d_inode;
565 struct dentry *parent = NULL;
566
567 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
568 goto failed;
569
570 if (!IS_ROOT(dentry)) {
571 parent = dentry->d_parent;
572 if (unlikely(!spin_trylock(&parent->d_lock))) {
573 if (inode)
574 spin_unlock(&inode->i_lock);
575 goto failed;
576 }
577 }
578
579 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400580 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400581
582failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400583 spin_unlock(&dentry->d_lock);
Al Viroe55fd012014-05-28 13:51:12 -0400584 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100585}
586
Al Viro046b9612014-05-29 08:54:52 -0400587static inline struct dentry *lock_parent(struct dentry *dentry)
588{
589 struct dentry *parent = dentry->d_parent;
590 if (IS_ROOT(dentry))
591 return NULL;
Linus Torvalds360f5472015-01-09 15:19:03 -0800592 if (unlikely(dentry->d_lockref.count < 0))
Al Viroc2338f22014-06-12 00:29:13 -0400593 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400594 if (likely(spin_trylock(&parent->d_lock)))
595 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400596 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400597 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400598again:
599 parent = ACCESS_ONCE(dentry->d_parent);
600 spin_lock(&parent->d_lock);
601 /*
602 * We can't blindly lock dentry until we are sure
603 * that we won't violate the locking order.
604 * Any changes of dentry->d_parent must have
605 * been done with parent->d_lock held, so
606 * spin_lock() above is enough of a barrier
607 * for checking if it's still our child.
608 */
609 if (unlikely(parent != dentry->d_parent)) {
610 spin_unlock(&parent->d_lock);
611 goto again;
612 }
613 rcu_read_unlock();
614 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700615 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400616 else
617 parent = NULL;
618 return parent;
619}
620
Linus Torvalds360f5472015-01-09 15:19:03 -0800621/*
622 * Try to do a lockless dput(), and return whether that was successful.
623 *
624 * If unsuccessful, we return false, having already taken the dentry lock.
625 *
626 * The caller needs to hold the RCU read lock, so that the dentry is
627 * guaranteed to stay around even if the refcount goes down to zero!
628 */
629static inline bool fast_dput(struct dentry *dentry)
630{
631 int ret;
632 unsigned int d_flags;
633
634 /*
635 * If we have a d_op->d_delete() operation, we sould not
Al Viro75a6f822015-07-08 02:42:38 +0100636 * let the dentry count go to zero, so use "put_or_lock".
Linus Torvalds360f5472015-01-09 15:19:03 -0800637 */
638 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
639 return lockref_put_or_lock(&dentry->d_lockref);
640
641 /*
642 * .. otherwise, we can try to just decrement the
643 * lockref optimistically.
644 */
645 ret = lockref_put_return(&dentry->d_lockref);
646
647 /*
648 * If the lockref_put_return() failed due to the lock being held
649 * by somebody else, the fast path has failed. We will need to
650 * get the lock, and then check the count again.
651 */
652 if (unlikely(ret < 0)) {
653 spin_lock(&dentry->d_lock);
654 if (dentry->d_lockref.count > 1) {
655 dentry->d_lockref.count--;
656 spin_unlock(&dentry->d_lock);
657 return 1;
658 }
659 return 0;
660 }
661
662 /*
663 * If we weren't the last ref, we're done.
664 */
665 if (ret)
666 return 1;
667
668 /*
669 * Careful, careful. The reference count went down
670 * to zero, but we don't hold the dentry lock, so
671 * somebody else could get it again, and do another
672 * dput(), and we need to not race with that.
673 *
674 * However, there is a very special and common case
675 * where we don't care, because there is nothing to
676 * do: the dentry is still hashed, it does not have
677 * a 'delete' op, and it's referenced and already on
678 * the LRU list.
679 *
680 * NOTE! Since we aren't locked, these values are
681 * not "stable". However, it is sufficient that at
682 * some point after we dropped the reference the
683 * dentry was hashed and the flags had the proper
684 * value. Other dentry users may have re-gotten
685 * a reference to the dentry and change that, but
686 * our work is done - we can leave the dentry
687 * around with a zero refcount.
688 */
689 smp_rmb();
690 d_flags = ACCESS_ONCE(dentry->d_flags);
Al Viro75a6f822015-07-08 02:42:38 +0100691 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
Linus Torvalds360f5472015-01-09 15:19:03 -0800692
693 /* Nothing to do? Dropping the reference was all we needed? */
694 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
695 return 1;
696
697 /*
698 * Not the fast normal case? Get the lock. We've already decremented
699 * the refcount, but we'll need to re-check the situation after
700 * getting the lock.
701 */
702 spin_lock(&dentry->d_lock);
703
704 /*
705 * Did somebody else grab a reference to it in the meantime, and
706 * we're no longer the last user after all? Alternatively, somebody
707 * else could have killed it and marked it dead. Either way, we
708 * don't need to do anything else.
709 */
710 if (dentry->d_lockref.count) {
711 spin_unlock(&dentry->d_lock);
712 return 1;
713 }
714
715 /*
716 * Re-get the reference we optimistically dropped. We hold the
717 * lock, and we just tested that it was zero, so we can just
718 * set it to 1.
719 */
720 dentry->d_lockref.count = 1;
721 return 0;
722}
723
724
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725/*
726 * This is dput
727 *
728 * This is complicated by the fact that we do not want to put
729 * dentries that are no longer on any hash chain on the unused
730 * list: we'd much rather just get rid of them immediately.
731 *
732 * However, that implies that we have to traverse the dentry
733 * tree upwards to the parents which might _also_ now be
734 * scheduled for deletion (it may have been only waiting for
735 * its last child to go away).
736 *
737 * This tail recursion is done by hand as we don't want to depend
738 * on the compiler to always get this right (gcc generally doesn't).
739 * Real recursion would eat up our stack space.
740 */
741
742/*
743 * dput - release a dentry
744 * @dentry: dentry to release
745 *
746 * Release a dentry. This will drop the usage count and if appropriate
747 * call the dentry unlink method as well as removing it from the queues and
748 * releasing its resources. If the parent dentries were scheduled for release
749 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751void dput(struct dentry *dentry)
752{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700753 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 return;
755
756repeat:
Wei Fang47be6182016-07-06 11:32:20 +0800757 might_sleep();
758
Linus Torvalds360f5472015-01-09 15:19:03 -0800759 rcu_read_lock();
760 if (likely(fast_dput(dentry))) {
761 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 return;
Linus Torvalds360f5472015-01-09 15:19:03 -0800763 }
764
765 /* Slow case: now with the dentry lock held */
766 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Al Viro85c7f812016-04-14 19:52:13 -0400768 WARN_ON(d_in_lookup(dentry));
769
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700770 /* Unreachable? Get rid of it */
771 if (unlikely(d_unhashed(dentry)))
772 goto kill_it;
773
Al Viro75a6f822015-07-08 02:42:38 +0100774 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
775 goto kill_it;
776
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700777 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100779 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 }
Nick Piggin265ac902010-10-10 05:36:24 -0400781
Linus Torvalds358eec12013-10-31 15:43:02 -0700782 if (!(dentry->d_flags & DCACHE_REFERENCED))
783 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400784 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400785
Waiman Long98474232013-08-28 18:24:59 -0700786 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100787 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 return;
789
Miklos Szeredid52b9082007-05-08 00:23:46 -0700790kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400791 dentry = dentry_kill(dentry);
Wei Fang47be6182016-07-06 11:32:20 +0800792 if (dentry) {
793 cond_resched();
Miklos Szeredid52b9082007-05-08 00:23:46 -0700794 goto repeat;
Wei Fang47be6182016-07-06 11:32:20 +0800795 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700797EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100800/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100801static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
Waiman Long98474232013-08-28 18:24:59 -0700803 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805
Nick Piggindc0474b2011-01-07 17:49:43 +1100806static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100807{
Waiman Long98474232013-08-28 18:24:59 -0700808 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100809}
810
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100811struct dentry *dget_parent(struct dentry *dentry)
812{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700813 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100814 struct dentry *ret;
815
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700816 /*
817 * Do optimistic parent lookup without any
818 * locking.
819 */
820 rcu_read_lock();
821 ret = ACCESS_ONCE(dentry->d_parent);
822 gotref = lockref_get_not_zero(&ret->d_lockref);
823 rcu_read_unlock();
824 if (likely(gotref)) {
825 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
826 return ret;
827 dput(ret);
828 }
829
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100830repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100831 /*
832 * Don't need rcu_dereference because we re-check it was correct under
833 * the lock.
834 */
835 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100836 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100837 spin_lock(&ret->d_lock);
838 if (unlikely(ret != dentry->d_parent)) {
839 spin_unlock(&ret->d_lock);
840 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100841 goto repeat;
842 }
Nick Piggina734eb42011-01-07 17:49:44 +1100843 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700844 BUG_ON(!ret->d_lockref.count);
845 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100846 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100847 return ret;
848}
849EXPORT_SYMBOL(dget_parent);
850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851/**
852 * d_find_alias - grab a hashed alias of inode
853 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 *
855 * If inode has a hashed alias, or is a directory and has any alias,
856 * acquire the reference to alias and return it. Otherwise return NULL.
857 * Notice that if inode is a directory there can be only one alias and
858 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800859 * of a filesystem, or if the directory was renamed and d_revalidate
860 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700862 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500863 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500865static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866{
Nick Pigginda502952011-01-07 17:49:33 +1100867 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Nick Pigginda502952011-01-07 17:49:33 +1100869again:
870 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400871 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100872 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700874 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100875 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500877 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100878 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100879 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 return alias;
881 }
882 }
Nick Pigginda502952011-01-07 17:49:33 +1100883 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 }
Nick Pigginda502952011-01-07 17:49:33 +1100885 if (discon_alias) {
886 alias = discon_alias;
887 spin_lock(&alias->d_lock);
888 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500889 __dget_dlock(alias);
890 spin_unlock(&alias->d_lock);
891 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100892 }
893 spin_unlock(&alias->d_lock);
894 goto again;
895 }
896 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
Nick Pigginda502952011-01-07 17:49:33 +1100899struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900{
David Howells214fda12006-03-25 03:06:36 -0800901 struct dentry *de = NULL;
902
Al Virob3d9b7a2012-06-09 13:51:19 -0400903 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100904 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500905 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100906 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800907 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 return de;
909}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700910EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
912/*
913 * Try to kill dentries associated with this inode.
914 * WARNING: you must own a reference to inode.
915 */
916void d_prune_aliases(struct inode *inode)
917{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700918 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100920 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400921 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700923 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400924 struct dentry *parent = lock_parent(dentry);
925 if (likely(!dentry->d_lockref.count)) {
926 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800927 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400928 goto restart;
929 }
930 if (parent)
931 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 }
933 spin_unlock(&dentry->d_lock);
934 }
Nick Piggin873feea2011-01-07 17:50:06 +1100935 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700937EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400939static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
Al Viro5c47e6d2014-04-29 16:13:18 -0400941 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700942
Miklos Szeredi60942f22014-05-02 15:38:39 -0400943 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400944 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400945 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100946 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400947 parent = lock_parent(dentry);
948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000950 * The dispose list is isolated and dentries are not accounted
951 * to the LRU here, so we can simply remove it from the list
952 * here regardless of whether it is referenced or not.
953 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400954 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000955
956 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000958 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800960 if (dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700961 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400962 if (parent)
963 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 continue;
965 }
Nick Piggin77812a12011-01-07 17:49:48 +1100966
Al Viro64fd72e2014-05-28 09:48:44 -0400967
968 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
969 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
970 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400971 if (parent)
972 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -0400973 if (can_free)
974 dentry_free(dentry);
975 continue;
976 }
977
Al Viroff2fde92014-05-28 13:59:13 -0400978 inode = dentry->d_inode;
979 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400980 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000981 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400982 if (parent)
983 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400984 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000985 }
Al Viroff2fde92014-05-28 13:59:13 -0400986
Al Viroff2fde92014-05-28 13:59:13 -0400987 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -0400988
Al Viro5c47e6d2014-04-29 16:13:18 -0400989 /*
990 * We need to prune ancestors too. This is necessary to prevent
991 * quadratic behavior of shrink_dcache_parent(), but is also
992 * expected to be beneficial in reducing dentry cache
993 * fragmentation.
994 */
995 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -0400996 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
997 parent = lock_parent(dentry);
998 if (dentry->d_lockref.count != 1) {
999 dentry->d_lockref.count--;
1000 spin_unlock(&dentry->d_lock);
1001 if (parent)
1002 spin_unlock(&parent->d_lock);
1003 break;
1004 }
1005 inode = dentry->d_inode; /* can't be NULL */
1006 if (unlikely(!spin_trylock(&inode->i_lock))) {
1007 spin_unlock(&dentry->d_lock);
1008 if (parent)
1009 spin_unlock(&parent->d_lock);
1010 cpu_relax();
1011 continue;
1012 }
1013 __dentry_kill(dentry);
1014 dentry = parent;
1015 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001017}
1018
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001019static enum lru_status dentry_lru_isolate(struct list_head *item,
1020 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +10001021{
1022 struct list_head *freeable = arg;
1023 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1024
1025
1026 /*
1027 * we are inverting the lru lock/dentry->d_lock here,
1028 * so use a trylock. If we fail to get the lock, just skip
1029 * it
1030 */
1031 if (!spin_trylock(&dentry->d_lock))
1032 return LRU_SKIP;
1033
1034 /*
1035 * Referenced dentries are still in use. If they have active
1036 * counts, just remove them from the LRU. Otherwise give them
1037 * another pass through the LRU.
1038 */
1039 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001040 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001041 spin_unlock(&dentry->d_lock);
1042 return LRU_REMOVED;
1043 }
1044
1045 if (dentry->d_flags & DCACHE_REFERENCED) {
1046 dentry->d_flags &= ~DCACHE_REFERENCED;
1047 spin_unlock(&dentry->d_lock);
1048
1049 /*
1050 * The list move itself will be made by the common LRU code. At
1051 * this point, we've dropped the dentry->d_lock but keep the
1052 * lru lock. This is safe to do, since every list movement is
1053 * protected by the lru lock even if both locks are held.
1054 *
1055 * This is guaranteed by the fact that all LRU management
1056 * functions are intermediated by the LRU API calls like
1057 * list_lru_add and list_lru_del. List movement in this file
1058 * only ever occur through this functions or through callbacks
1059 * like this one, that are called from the LRU API.
1060 *
1061 * The only exceptions to this are functions like
1062 * shrink_dentry_list, and code that first checks for the
1063 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1064 * operating only with stack provided lists after they are
1065 * properly isolated from the main list. It is thus, always a
1066 * local access.
1067 */
1068 return LRU_ROTATE;
1069 }
1070
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001071 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001072 spin_unlock(&dentry->d_lock);
1073
1074 return LRU_REMOVED;
1075}
1076
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001077/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001078 * prune_dcache_sb - shrink the dcache
1079 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001080 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001081 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001082 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1083 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001084 * function.
1085 *
1086 * This function may fail to free any resources if all the dentries are in
1087 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001088 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001089long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001090{
Dave Chinnerf6041562013-08-28 10:18:00 +10001091 LIST_HEAD(dispose);
1092 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001093
Vladimir Davydov503c3582015-02-12 14:58:47 -08001094 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1095 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001096 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001097 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098}
1099
Glauber Costa4e717f52013-08-28 10:18:03 +10001100static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001101 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001102{
Glauber Costa4e717f52013-08-28 10:18:03 +10001103 struct list_head *freeable = arg;
1104 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001105
Glauber Costa4e717f52013-08-28 10:18:03 +10001106 /*
1107 * we are inverting the lru lock/dentry->d_lock here,
1108 * so use a trylock. If we fail to get the lock, just skip
1109 * it
1110 */
1111 if (!spin_trylock(&dentry->d_lock))
1112 return LRU_SKIP;
1113
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001114 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001115 spin_unlock(&dentry->d_lock);
1116
1117 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001118}
1119
Glauber Costa4e717f52013-08-28 10:18:03 +10001120
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001121/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 * shrink_dcache_sb - shrink dcache for a superblock
1123 * @sb: superblock
1124 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001125 * Shrink the dcache for the specified super block. This is used to free
1126 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001128void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129{
Glauber Costa4e717f52013-08-28 10:18:03 +10001130 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001131
Glauber Costa4e717f52013-08-28 10:18:03 +10001132 do {
1133 LIST_HEAD(dispose);
1134
1135 freed = list_lru_walk(&sb->s_dentry_lru,
1136 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1137
1138 this_cpu_sub(nr_dentry_unused, freed);
1139 shrink_dentry_list(&dispose);
1140 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001142EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001145 * enum d_walk_ret - action to talke during tree walk
1146 * @D_WALK_CONTINUE: contrinue walk
1147 * @D_WALK_QUIT: quit walk
1148 * @D_WALK_NORETRY: quit when retry is needed
1149 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001151enum d_walk_ret {
1152 D_WALK_CONTINUE,
1153 D_WALK_QUIT,
1154 D_WALK_NORETRY,
1155 D_WALK_SKIP,
1156};
1157
1158/**
1159 * d_walk - walk the dentry tree
1160 * @parent: start of walk
1161 * @data: data passed to @enter() and @finish()
1162 * @enter: callback when first entering the dentry
1163 * @finish: callback when successfully finished the walk
1164 *
1165 * The @enter() and @finish() callbacks are called with d_lock held.
1166 */
1167static void d_walk(struct dentry *parent, void *data,
1168 enum d_walk_ret (*enter)(void *, struct dentry *),
1169 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170{
Nick Piggin949854d2011-01-07 17:49:37 +11001171 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001173 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001174 enum d_walk_ret ret;
1175 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001176
Nick Piggin58db63d2011-01-07 17:49:39 +11001177again:
Al Viro48f5ec22013-09-09 15:22:25 -04001178 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001179 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001180 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001181
1182 ret = enter(data, this_parent);
1183 switch (ret) {
1184 case D_WALK_CONTINUE:
1185 break;
1186 case D_WALK_QUIT:
1187 case D_WALK_SKIP:
1188 goto out_unlock;
1189 case D_WALK_NORETRY:
1190 retry = false;
1191 break;
1192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193repeat:
1194 next = this_parent->d_subdirs.next;
1195resume:
1196 while (next != &this_parent->d_subdirs) {
1197 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001198 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001200
Al Viroba65dc52016-06-10 11:32:47 -04001201 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1202 continue;
1203
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001204 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001205
1206 ret = enter(data, dentry);
1207 switch (ret) {
1208 case D_WALK_CONTINUE:
1209 break;
1210 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001211 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001212 goto out_unlock;
1213 case D_WALK_NORETRY:
1214 retry = false;
1215 break;
1216 case D_WALK_SKIP:
1217 spin_unlock(&dentry->d_lock);
1218 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001219 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001220
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001222 spin_unlock(&this_parent->d_lock);
1223 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001225 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 goto repeat;
1227 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001228 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 }
1230 /*
1231 * All done at this level ... ascend and resume the search.
1232 */
Al Viroca5358e2014-10-26 19:31:10 -04001233 rcu_read_lock();
1234ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001236 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001237 this_parent = child->d_parent;
1238
Al Viro31dec132013-10-25 17:04:27 -04001239 spin_unlock(&child->d_lock);
1240 spin_lock(&this_parent->d_lock);
1241
Al Viroca5358e2014-10-26 19:31:10 -04001242 /* might go back up the wrong parent if we have had a rename. */
1243 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001244 goto rename_retry;
Al Viro21591842015-05-28 23:09:19 -04001245 /* go into the first sibling still alive */
1246 do {
1247 next = child->d_child.next;
Al Viroca5358e2014-10-26 19:31:10 -04001248 if (next == &this_parent->d_subdirs)
1249 goto ascend;
1250 child = list_entry(next, struct dentry, d_child);
Al Viro21591842015-05-28 23:09:19 -04001251 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
Al Viro31dec132013-10-25 17:04:27 -04001252 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 goto resume;
1254 }
Al Viroca5358e2014-10-26 19:31:10 -04001255 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001256 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001257 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001258 if (finish)
1259 finish(data);
1260
1261out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001262 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001263 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001264 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001265
1266rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001267 spin_unlock(&this_parent->d_lock);
1268 rcu_read_unlock();
1269 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001270 if (!retry)
1271 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001272 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001273 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001275
1276/*
1277 * Search for at least 1 mount point in the dentry's subdirs.
1278 * We descend to the next level whenever the d_subdirs
1279 * list is non-empty and continue searching.
1280 */
1281
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001282static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1283{
1284 int *ret = data;
1285 if (d_mountpoint(dentry)) {
1286 *ret = 1;
1287 return D_WALK_QUIT;
1288 }
1289 return D_WALK_CONTINUE;
1290}
1291
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001292/**
1293 * have_submounts - check for mounts over a dentry
1294 * @parent: dentry to check.
1295 *
1296 * Return true if the parent or its subdirectories contain
1297 * a mount point
1298 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001299int have_submounts(struct dentry *parent)
1300{
1301 int ret = 0;
1302
1303 d_walk(parent, &ret, check_mount, NULL);
1304
1305 return ret;
1306}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001307EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
1309/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001310 * Called by mount code to set a mountpoint and check if the mountpoint is
1311 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1312 * subtree can become unreachable).
1313 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001314 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001315 * this reason take rename_lock and d_lock on dentry and ancestors.
1316 */
1317int d_set_mounted(struct dentry *dentry)
1318{
1319 struct dentry *p;
1320 int ret = -ENOENT;
1321 write_seqlock(&rename_lock);
1322 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001323 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001324 spin_lock(&p->d_lock);
1325 if (unlikely(d_unhashed(p))) {
1326 spin_unlock(&p->d_lock);
1327 goto out;
1328 }
1329 spin_unlock(&p->d_lock);
1330 }
1331 spin_lock(&dentry->d_lock);
1332 if (!d_unlinked(dentry)) {
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +13001333 ret = -EBUSY;
1334 if (!d_mountpoint(dentry)) {
1335 dentry->d_flags |= DCACHE_MOUNTED;
1336 ret = 0;
1337 }
Miklos Szeredieed81002013-09-05 14:39:11 +02001338 }
1339 spin_unlock(&dentry->d_lock);
1340out:
1341 write_sequnlock(&rename_lock);
1342 return ret;
1343}
1344
1345/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001346 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 * and move any unused dentries to the end of the unused
1348 * list for prune_dcache(). We descend to the next level
1349 * whenever the d_subdirs list is non-empty and continue
1350 * searching.
1351 *
1352 * It returns zero iff there are no unused children,
1353 * otherwise it returns the number of children moved to
1354 * the end of the unused list. This may not be the total
1355 * number of unused children, because select_parent can
1356 * drop the lock and return early due to latency
1357 * constraints.
1358 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001359
1360struct select_data {
1361 struct dentry *start;
1362 struct list_head dispose;
1363 int found;
1364};
1365
1366static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001368 struct select_data *data = _data;
1369 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001371 if (data->start == dentry)
1372 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
Al Virofe915222014-05-03 00:02:25 -04001374 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001375 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001376 } else {
1377 if (dentry->d_flags & DCACHE_LRU_LIST)
1378 d_lru_del(dentry);
1379 if (!dentry->d_lockref.count) {
1380 d_shrink_add(dentry, &data->dispose);
1381 data->found++;
1382 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 }
1384 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001385 * We can return to the caller if we have found some (this
1386 * ensures forward progress). We'll be coming back to find
1387 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 */
Al Virofe915222014-05-03 00:02:25 -04001389 if (!list_empty(&data->dispose))
1390 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001392 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393}
1394
1395/**
1396 * shrink_dcache_parent - prune dcache
1397 * @parent: parent of entries to prune
1398 *
1399 * Prune the dcache to remove unused children of the parent dentry.
1400 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001401void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001403 for (;;) {
1404 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001406 INIT_LIST_HEAD(&data.dispose);
1407 data.start = parent;
1408 data.found = 0;
1409
1410 d_walk(parent, &data, select_collect, NULL);
1411 if (!data.found)
1412 break;
1413
1414 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001415 cond_resched();
1416 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001418EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
Al Viro9c8c10e2014-05-02 20:36:10 -04001420static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001421{
Al Viro9c8c10e2014-05-02 20:36:10 -04001422 /* it has busy descendents; complain about those instead */
1423 if (!list_empty(&dentry->d_subdirs))
1424 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001425
Al Viro9c8c10e2014-05-02 20:36:10 -04001426 /* root with refcount 1 is fine */
1427 if (dentry == _data && dentry->d_lockref.count == 1)
1428 return D_WALK_CONTINUE;
1429
1430 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1431 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001432 dentry,
1433 dentry->d_inode ?
1434 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001435 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001436 dentry->d_lockref.count,
1437 dentry->d_sb->s_type->name,
1438 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001439 WARN_ON(1);
1440 return D_WALK_CONTINUE;
1441}
1442
1443static void do_one_tree(struct dentry *dentry)
1444{
1445 shrink_dcache_parent(dentry);
1446 d_walk(dentry, dentry, umount_check, NULL);
1447 d_drop(dentry);
1448 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001449}
1450
1451/*
1452 * destroy the dentries attached to a superblock on unmounting
1453 */
1454void shrink_dcache_for_umount(struct super_block *sb)
1455{
1456 struct dentry *dentry;
1457
Al Viro9c8c10e2014-05-02 20:36:10 -04001458 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001459
1460 dentry = sb->s_root;
1461 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001462 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001463
1464 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001465 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1466 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001467 }
1468}
1469
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001470struct detach_data {
1471 struct select_data select;
1472 struct dentry *mountpoint;
1473};
1474static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001475{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001476 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001477
1478 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001479 __dget_dlock(dentry);
1480 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001481 return D_WALK_QUIT;
1482 }
1483
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001484 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001485}
1486
1487static void check_and_drop(void *_data)
1488{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001489 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001490
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001491 if (!data->mountpoint && !data->select.found)
1492 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001493}
1494
1495/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001496 * d_invalidate - detach submounts, prune dcache, and drop
1497 * @dentry: dentry to invalidate (aka detach, prune and drop)
1498 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001499 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001500 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001501 * The final d_drop is done as an atomic operation relative to
1502 * rename_lock ensuring there are no races with d_set_mounted. This
1503 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001504 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001505void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001506{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001507 /*
1508 * If it's already been dropped, return OK.
1509 */
1510 spin_lock(&dentry->d_lock);
1511 if (d_unhashed(dentry)) {
1512 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001513 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001514 }
1515 spin_unlock(&dentry->d_lock);
1516
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001517 /* Negative dentries can be dropped without further checks */
1518 if (!dentry->d_inode) {
1519 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001520 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001521 }
1522
1523 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001524 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001525
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001526 data.mountpoint = NULL;
1527 INIT_LIST_HEAD(&data.select.dispose);
1528 data.select.start = dentry;
1529 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001530
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001531 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001532
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001533 if (data.select.found)
1534 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001535
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001536 if (data.mountpoint) {
1537 detach_mounts(data.mountpoint);
1538 dput(data.mountpoint);
1539 }
1540
1541 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001542 break;
1543
1544 cond_resched();
1545 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001546}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001547EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549/**
Al Viroa4464db2011-07-07 15:03:58 -04001550 * __d_alloc - allocate a dcache entry
1551 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 * @name: qstr of the name
1553 *
1554 * Allocates a dentry. It returns %NULL if there is insufficient memory
1555 * available. On a success the dentry is returned. The name passed in is
1556 * copied and the copy passed in may be reused after this call.
1557 */
1558
Al Viroa4464db2011-07-07 15:03:58 -04001559struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560{
1561 struct dentry *dentry;
1562 char *dname;
Miklos Szeredi285b1022016-06-28 11:47:32 +02001563 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Mel Gormane12ba742007-10-16 01:25:52 -07001565 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 if (!dentry)
1567 return NULL;
1568
Linus Torvalds6326c712012-05-21 16:14:04 -07001569 /*
1570 * We guarantee that the inline name is always NUL-terminated.
1571 * This way the memcpy() done by the name switching in rename
1572 * will still always have a NUL at the end, even if we might
1573 * be overwriting an internal NUL character
1574 */
1575 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Al Viro798434b2016-03-24 20:38:43 -04001576 if (unlikely(!name)) {
1577 static const struct qstr anon = QSTR_INIT("/", 1);
1578 name = &anon;
1579 dname = dentry->d_iname;
1580 } else if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001581 size_t size = offsetof(struct external_name, name[1]);
Vladimir Davydov5d097052016-01-14 15:18:21 -08001582 struct external_name *p = kmalloc(size + name->len,
1583 GFP_KERNEL_ACCOUNT);
Al Viro8d85b482014-09-29 14:54:27 -04001584 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 kmem_cache_free(dentry_cache, dentry);
1586 return NULL;
1587 }
Al Viro8d85b482014-09-29 14:54:27 -04001588 atomic_set(&p->u.count, 1);
1589 dname = p->name;
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -08001590 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1591 kasan_unpoison_shadow(dname,
1592 round_up(name->len + 1, sizeof(unsigned long)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 } else {
1594 dname = dentry->d_iname;
1595 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
1597 dentry->d_name.len = name->len;
1598 dentry->d_name.hash = name->hash;
1599 memcpy(dname, name->name, name->len);
1600 dname[name->len] = 0;
1601
Linus Torvalds6326c712012-05-21 16:14:04 -07001602 /* Make sure we always see the terminating NUL character */
1603 smp_wmb();
1604 dentry->d_name.name = dname;
1605
Waiman Long98474232013-08-28 18:24:59 -07001606 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001607 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001609 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001611 dentry->d_parent = dentry;
1612 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 dentry->d_op = NULL;
1614 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001615 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 INIT_LIST_HEAD(&dentry->d_lru);
1617 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001618 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1619 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001620 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
Miklos Szeredi285b1022016-06-28 11:47:32 +02001622 if (dentry->d_op && dentry->d_op->d_init) {
1623 err = dentry->d_op->d_init(dentry);
1624 if (err) {
1625 if (dname_external(dentry))
1626 kfree(external_name(dentry));
1627 kmem_cache_free(dentry_cache, dentry);
1628 return NULL;
1629 }
1630 }
1631
Nick Piggin3e880fb2011-01-07 17:49:19 +11001632 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001633
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 return dentry;
1635}
Al Viroa4464db2011-07-07 15:03:58 -04001636
1637/**
1638 * d_alloc - allocate a dcache entry
1639 * @parent: parent of entry to allocate
1640 * @name: qstr of the name
1641 *
1642 * Allocates a dentry. It returns %NULL if there is insufficient memory
1643 * available. On a success the dentry is returned. The name passed in is
1644 * copied and the copy passed in may be reused after this call.
1645 */
1646struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1647{
1648 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1649 if (!dentry)
1650 return NULL;
Al Viro3d56c252016-06-07 21:26:55 -04001651 dentry->d_flags |= DCACHE_RCUACCESS;
Al Viroa4464db2011-07-07 15:03:58 -04001652 spin_lock(&parent->d_lock);
1653 /*
1654 * don't need child lock because it is not subject
1655 * to concurrency here
1656 */
1657 __dget_dlock(parent);
1658 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001659 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001660 spin_unlock(&parent->d_lock);
1661
1662 return dentry;
1663}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001664EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
Al Viroba65dc52016-06-10 11:32:47 -04001666struct dentry *d_alloc_cursor(struct dentry * parent)
1667{
1668 struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
1669 if (dentry) {
1670 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1671 dentry->d_parent = dget(parent);
1672 }
1673 return dentry;
1674}
1675
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001676/**
1677 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1678 * @sb: the superblock
1679 * @name: qstr of the name
1680 *
1681 * For a filesystem that just pins its dentries in memory and never
1682 * performs lookups at all, return an unhashed IS_ROOT dentry.
1683 */
Nick Piggin4b936882011-01-07 17:50:07 +11001684struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1685{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001686 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001687}
1688EXPORT_SYMBOL(d_alloc_pseudo);
1689
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1691{
1692 struct qstr q;
1693
1694 q.name = name;
Linus Torvalds8387ff22016-06-10 07:51:30 -07001695 q.hash_len = hashlen_string(parent, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 return d_alloc(parent, &q);
1697}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001698EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699
Nick Pigginfb045ad2011-01-07 17:49:55 +11001700void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1701{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001702 WARN_ON_ONCE(dentry->d_op);
1703 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001704 DCACHE_OP_COMPARE |
1705 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001706 DCACHE_OP_WEAK_REVALIDATE |
David Howells4bacc9c2015-06-18 14:32:31 +01001707 DCACHE_OP_DELETE |
Miklos Szeredid101a122016-03-26 16:14:37 -04001708 DCACHE_OP_REAL));
Nick Pigginfb045ad2011-01-07 17:49:55 +11001709 dentry->d_op = op;
1710 if (!op)
1711 return;
1712 if (op->d_hash)
1713 dentry->d_flags |= DCACHE_OP_HASH;
1714 if (op->d_compare)
1715 dentry->d_flags |= DCACHE_OP_COMPARE;
1716 if (op->d_revalidate)
1717 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001718 if (op->d_weak_revalidate)
1719 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001720 if (op->d_delete)
1721 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001722 if (op->d_prune)
1723 dentry->d_flags |= DCACHE_OP_PRUNE;
Miklos Szeredid101a122016-03-26 16:14:37 -04001724 if (op->d_real)
1725 dentry->d_flags |= DCACHE_OP_REAL;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001726
1727}
1728EXPORT_SYMBOL(d_set_d_op);
1729
David Howellsdf1a0852015-01-29 12:02:28 +00001730
1731/*
1732 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1733 * @dentry - The dentry to mark
1734 *
1735 * Mark a dentry as falling through to the lower layer (as set with
1736 * d_pin_lower()). This flag may be recorded on the medium.
1737 */
1738void d_set_fallthru(struct dentry *dentry)
1739{
1740 spin_lock(&dentry->d_lock);
1741 dentry->d_flags |= DCACHE_FALLTHRU;
1742 spin_unlock(&dentry->d_lock);
1743}
1744EXPORT_SYMBOL(d_set_fallthru);
1745
David Howellsb18825a2013-09-12 19:22:53 +01001746static unsigned d_flags_for_inode(struct inode *inode)
1747{
David Howells44bdb5e2015-01-29 12:02:29 +00001748 unsigned add_flags = DCACHE_REGULAR_TYPE;
David Howellsb18825a2013-09-12 19:22:53 +01001749
1750 if (!inode)
1751 return DCACHE_MISS_TYPE;
1752
1753 if (S_ISDIR(inode->i_mode)) {
1754 add_flags = DCACHE_DIRECTORY_TYPE;
1755 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1756 if (unlikely(!inode->i_op->lookup))
1757 add_flags = DCACHE_AUTODIR_TYPE;
1758 else
1759 inode->i_opflags |= IOP_LOOKUP;
1760 }
David Howells44bdb5e2015-01-29 12:02:29 +00001761 goto type_determined;
David Howellsb18825a2013-09-12 19:22:53 +01001762 }
1763
David Howells44bdb5e2015-01-29 12:02:29 +00001764 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
Al Viro6b255392015-11-17 10:20:54 -05001765 if (unlikely(inode->i_op->get_link)) {
David Howells44bdb5e2015-01-29 12:02:29 +00001766 add_flags = DCACHE_SYMLINK_TYPE;
1767 goto type_determined;
1768 }
1769 inode->i_opflags |= IOP_NOFOLLOW;
1770 }
1771
1772 if (unlikely(!S_ISREG(inode->i_mode)))
1773 add_flags = DCACHE_SPECIAL_TYPE;
1774
1775type_determined:
David Howellsb18825a2013-09-12 19:22:53 +01001776 if (unlikely(IS_AUTOMOUNT(inode)))
1777 add_flags |= DCACHE_NEED_AUTOMOUNT;
1778 return add_flags;
1779}
1780
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001781static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1782{
David Howellsb18825a2013-09-12 19:22:53 +01001783 unsigned add_flags = d_flags_for_inode(inode);
Al Viro85c7f812016-04-14 19:52:13 -04001784 WARN_ON(d_in_lookup(dentry));
David Howellsb18825a2013-09-12 19:22:53 +01001785
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001786 spin_lock(&dentry->d_lock);
Al Virode689f52016-03-09 18:05:42 -05001787 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
Al Viroa528aca2016-02-29 12:12:46 -05001788 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +00001789 __d_set_inode_and_type(dentry, inode, add_flags);
Al Viroa528aca2016-02-29 12:12:46 -05001790 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04001791 fsnotify_update_flags(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001792 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001793}
1794
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795/**
1796 * d_instantiate - fill in inode information for a dentry
1797 * @entry: dentry to complete
1798 * @inode: inode to attach to this dentry
1799 *
1800 * Fill in inode information in the entry.
1801 *
1802 * This turns negative dentries into productive full members
1803 * of society.
1804 *
1805 * NOTE! This assumes that the inode count has been incremented
1806 * (or otherwise set) by the caller to indicate that it is now
1807 * in use by the dcache.
1808 */
1809
1810void d_instantiate(struct dentry *entry, struct inode * inode)
1811{
Al Viro946e51f2014-10-26 19:19:16 -04001812 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Al Virode689f52016-03-09 18:05:42 -05001813 if (inode) {
Al Virob9680912016-04-11 00:53:26 -04001814 security_d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001815 spin_lock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001816 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001817 spin_unlock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001818 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001820EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821
1822/**
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001823 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1824 * @entry: dentry to complete
1825 * @inode: inode to attach to this dentry
1826 *
1827 * Fill in inode information in the entry. If a directory alias is found, then
1828 * return an error (and drop inode). Together with d_materialise_unique() this
1829 * guarantees that a directory inode may never have more than one alias.
1830 */
1831int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1832{
Al Viro946e51f2014-10-26 19:19:16 -04001833 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001834
Al Virob9680912016-04-11 00:53:26 -04001835 security_d_instantiate(entry, inode);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001836 spin_lock(&inode->i_lock);
1837 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1838 spin_unlock(&inode->i_lock);
1839 iput(inode);
1840 return -EBUSY;
1841 }
1842 __d_instantiate(entry, inode);
1843 spin_unlock(&inode->i_lock);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001844
1845 return 0;
1846}
1847EXPORT_SYMBOL(d_instantiate_no_diralias);
1848
Al Viroadc0e912012-01-08 16:49:21 -05001849struct dentry *d_make_root(struct inode *root_inode)
1850{
1851 struct dentry *res = NULL;
1852
1853 if (root_inode) {
Al Viro798434b2016-03-24 20:38:43 -04001854 res = __d_alloc(root_inode->i_sb, NULL);
Al Viroadc0e912012-01-08 16:49:21 -05001855 if (res)
1856 d_instantiate(res, root_inode);
1857 else
1858 iput(root_inode);
1859 }
1860 return res;
1861}
1862EXPORT_SYMBOL(d_make_root);
1863
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001864static struct dentry * __d_find_any_alias(struct inode *inode)
1865{
1866 struct dentry *alias;
1867
Al Virob3d9b7a2012-06-09 13:51:19 -04001868 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001869 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001870 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001871 __dget(alias);
1872 return alias;
1873}
1874
Sage Weil46f72b32012-01-10 09:04:37 -08001875/**
1876 * d_find_any_alias - find any alias for a given inode
1877 * @inode: inode to find an alias for
1878 *
1879 * If any aliases exist for the given inode, take and return a
1880 * reference for one of them. If no aliases exist, return %NULL.
1881 */
1882struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001883{
1884 struct dentry *de;
1885
1886 spin_lock(&inode->i_lock);
1887 de = __d_find_any_alias(inode);
1888 spin_unlock(&inode->i_lock);
1889 return de;
1890}
Sage Weil46f72b32012-01-10 09:04:37 -08001891EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001892
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001893static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001894{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001895 struct dentry *tmp;
1896 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001897 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001898
1899 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001900 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001901 if (IS_ERR(inode))
1902 return ERR_CAST(inode);
1903
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001904 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001905 if (res)
1906 goto out_iput;
1907
Al Viro798434b2016-03-24 20:38:43 -04001908 tmp = __d_alloc(inode->i_sb, NULL);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001909 if (!tmp) {
1910 res = ERR_PTR(-ENOMEM);
1911 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001912 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001913
Al Virob9680912016-04-11 00:53:26 -04001914 security_d_instantiate(tmp, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001915 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001916 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001917 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001918 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001919 dput(tmp);
1920 goto out_iput;
1921 }
1922
1923 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001924 add_flags = d_flags_for_inode(inode);
1925
1926 if (disconnected)
1927 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001928
Christoph Hellwig9308a612008-08-11 15:49:12 +02001929 spin_lock(&tmp->d_lock);
David Howells4bf46a22015-03-05 14:09:22 +00001930 __d_set_inode_and_type(tmp, inode, add_flags);
Al Viro946e51f2014-10-26 19:19:16 -04001931 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001932 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001933 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001934 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001935 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001936 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001937
Christoph Hellwig9308a612008-08-11 15:49:12 +02001938 return tmp;
1939
1940 out_iput:
1941 iput(inode);
1942 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001943}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001944
1945/**
1946 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1947 * @inode: inode to allocate the dentry for
1948 *
1949 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1950 * similar open by handle operations. The returned dentry may be anonymous,
1951 * or may have a full name (if the inode was already in the cache).
1952 *
1953 * When called on a directory inode, we must ensure that the inode only ever
1954 * has one dentry. If a dentry is found, that is returned instead of
1955 * allocating a new one.
1956 *
1957 * On successful return, the reference to the inode has been transferred
1958 * to the dentry. In case of an error the reference on the inode is released.
1959 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1960 * be passed in and the error will be propagated to the return value,
1961 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1962 */
1963struct dentry *d_obtain_alias(struct inode *inode)
1964{
1965 return __d_obtain_alias(inode, 1);
1966}
Benny Halevyadc48722009-02-27 14:02:59 -08001967EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968
1969/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001970 * d_obtain_root - find or allocate a dentry for a given inode
1971 * @inode: inode to allocate the dentry for
1972 *
1973 * Obtain an IS_ROOT dentry for the root of a filesystem.
1974 *
1975 * We must ensure that directory inodes only ever have one dentry. If a
1976 * dentry is found, that is returned instead of allocating a new one.
1977 *
1978 * On successful return, the reference to the inode has been transferred
1979 * to the dentry. In case of an error the reference on the inode is
1980 * released. A %NULL or IS_ERR inode may be passed in and will be the
1981 * error will be propagate to the return value, with a %NULL @inode
1982 * replaced by ERR_PTR(-ESTALE).
1983 */
1984struct dentry *d_obtain_root(struct inode *inode)
1985{
1986 return __d_obtain_alias(inode, 0);
1987}
1988EXPORT_SYMBOL(d_obtain_root);
1989
1990/**
Barry Naujok94035402008-05-21 16:50:46 +10001991 * d_add_ci - lookup or allocate new dentry with case-exact name
1992 * @inode: the inode case-insensitive lookup has found
1993 * @dentry: the negative dentry that was passed to the parent's lookup func
1994 * @name: the case-exact name to be associated with the returned dentry
1995 *
1996 * This is to avoid filling the dcache with case-insensitive names to the
1997 * same inode, only the actual correct case is stored in the dcache for
1998 * case-insensitive filesystems.
1999 *
2000 * For a case-insensitive lookup match and if the the case-exact dentry
2001 * already exists in in the dcache, use it and return it.
2002 *
2003 * If no entry exists with the exact case name, allocate new dentry with
2004 * the exact case, and return the spliced entry.
2005 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02002006struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10002007 struct qstr *name)
2008{
Al Virod9171b92016-04-15 03:33:13 -04002009 struct dentry *found, *res;
Barry Naujok94035402008-05-21 16:50:46 +10002010
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002011 /*
2012 * First check if a dentry matching the name already exists,
2013 * if not go ahead and create it now.
2014 */
Barry Naujok94035402008-05-21 16:50:46 +10002015 found = d_hash_and_lookup(dentry->d_parent, name);
Al Virod9171b92016-04-15 03:33:13 -04002016 if (found) {
2017 iput(inode);
2018 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002019 }
Al Virod9171b92016-04-15 03:33:13 -04002020 if (d_in_lookup(dentry)) {
2021 found = d_alloc_parallel(dentry->d_parent, name,
2022 dentry->d_wait);
2023 if (IS_ERR(found) || !d_in_lookup(found)) {
2024 iput(inode);
2025 return found;
2026 }
2027 } else {
2028 found = d_alloc(dentry->d_parent, name);
2029 if (!found) {
2030 iput(inode);
2031 return ERR_PTR(-ENOMEM);
2032 }
2033 }
2034 res = d_splice_alias(inode, found);
2035 if (res) {
2036 dput(found);
2037 return res;
2038 }
Al Viro4f522a22013-02-11 23:20:37 -05002039 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002040}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002041EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002043
Al Virod4c91a82016-06-25 23:33:49 -04002044static inline bool d_same_name(const struct dentry *dentry,
2045 const struct dentry *parent,
2046 const struct qstr *name)
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002047{
Al Virod4c91a82016-06-25 23:33:49 -04002048 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2049 if (dentry->d_name.len != name->len)
2050 return false;
2051 return dentry_cmp(dentry, name->name, name->len) == 0;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002052 }
Al Viro6fa67e72016-07-31 16:37:25 -04002053 return parent->d_op->d_compare(dentry,
Al Virod4c91a82016-06-25 23:33:49 -04002054 dentry->d_name.len, dentry->d_name.name,
2055 name) == 0;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002056}
2057
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002059 * __d_lookup_rcu - search for a dentry (racy, store-free)
2060 * @parent: parent dentry
2061 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002062 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002063 * Returns: dentry, or NULL
2064 *
2065 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2066 * resolution (store-free path walking) design described in
2067 * Documentation/filesystems/path-lookup.txt.
2068 *
2069 * This is not to be used outside core vfs.
2070 *
2071 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2072 * held, and rcu_read_lock held. The returned dentry must not be stored into
2073 * without taking d_lock and checking d_seq sequence count against @seq
2074 * returned here.
2075 *
Linus Torvalds15570082013-09-02 11:38:06 -07002076 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002077 * function.
2078 *
2079 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2080 * the returned dentry, so long as its parent's seqlock is checked after the
2081 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2082 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002083 *
2084 * NOTE! The caller *has* to check the resulting dentry against the sequence
2085 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002086 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002087struct dentry *__d_lookup_rcu(const struct dentry *parent,
2088 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002089 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002090{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002091 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002092 const unsigned char *str = name->name;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002093 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002094 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002095 struct dentry *dentry;
2096
2097 /*
2098 * Note: There is significant duplication with __d_lookup_rcu which is
2099 * required to prevent single threaded performance regressions
2100 * especially on architectures where smp_rmb (in seqcounts) are costly.
2101 * Keep the two functions in sync.
2102 */
2103
2104 /*
2105 * The hash list is protected using RCU.
2106 *
2107 * Carefully use d_seq when comparing a candidate dentry, to avoid
2108 * races with d_move().
2109 *
2110 * It is possible that concurrent renames can mess up our list
2111 * walk here and result in missing our dentry, resulting in the
2112 * false-negative result. d_lookup() protects against concurrent
2113 * renames using rename_lock seqlock.
2114 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002115 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002116 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002117 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002118 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002119
Nick Piggin31e6b012011-01-07 17:49:52 +11002120seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002121 /*
2122 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002123 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002124 *
2125 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002126 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002127 *
2128 * NOTE! We do a "raw" seqcount_begin here. That means that
2129 * we don't wait for the sequence count to stabilize if it
2130 * is in the middle of a sequence change. If we do the slow
2131 * dentry compare, we will do seqretries until it is stable,
2132 * and if we end up with a successful lookup, we actually
2133 * want to exit RCU lookup anyway.
Al Virod4c91a82016-06-25 23:33:49 -04002134 *
2135 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2136 * we are still guaranteed NUL-termination of ->d_name.name.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002137 */
2138 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002139 if (dentry->d_parent != parent)
2140 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002141 if (d_unhashed(dentry))
2142 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002143
2144 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Al Virod4c91a82016-06-25 23:33:49 -04002145 int tlen;
2146 const char *tname;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002147 if (dentry->d_name.hash != hashlen_hash(hashlen))
2148 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002149 tlen = dentry->d_name.len;
2150 tname = dentry->d_name.name;
2151 /* we want a consistent (name,len) pair */
2152 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2153 cpu_relax();
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002154 goto seqretry;
2155 }
Al Viro6fa67e72016-07-31 16:37:25 -04002156 if (parent->d_op->d_compare(dentry,
Al Virod4c91a82016-06-25 23:33:49 -04002157 tlen, tname, name) != 0)
2158 continue;
2159 } else {
2160 if (dentry->d_name.hash_len != hashlen)
2161 continue;
2162 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2163 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002164 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002165 *seqp = seq;
Al Virod4c91a82016-06-25 23:33:49 -04002166 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002167 }
2168 return NULL;
2169}
2170
2171/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 * d_lookup - search for a dentry
2173 * @parent: parent dentry
2174 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002175 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002177 * d_lookup searches the children of the parent dentry for the name in
2178 * question. If the dentry is found its reference count is incremented and the
2179 * dentry is returned. The caller must use dput to free the entry when it has
2180 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 */
Al Viroda2d8452013-01-24 18:29:34 -05002182struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183{
Nick Piggin31e6b012011-01-07 17:49:52 +11002184 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002185 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186
Daeseok Younb8314f92014-08-11 11:46:53 +09002187 do {
2188 seq = read_seqbegin(&rename_lock);
2189 dentry = __d_lookup(parent, name);
2190 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 break;
2192 } while (read_seqretry(&rename_lock, seq));
2193 return dentry;
2194}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002195EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196
Nick Piggin31e6b012011-01-07 17:49:52 +11002197/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002198 * __d_lookup - search for a dentry (racy)
2199 * @parent: parent dentry
2200 * @name: qstr of name we wish to find
2201 * Returns: dentry, or NULL
2202 *
2203 * __d_lookup is like d_lookup, however it may (rarely) return a
2204 * false-negative result due to unrelated rename activity.
2205 *
2206 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2207 * however it must be used carefully, eg. with a following d_lookup in
2208 * the case of failure.
2209 *
2210 * __d_lookup callers must be commented.
2211 */
Al Viroa713ca22013-01-24 18:27:00 -05002212struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 unsigned int hash = name->hash;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002215 struct hlist_bl_head *b = d_hash(hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002216 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002217 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002218 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219
Nick Pigginb04f7842010-08-18 04:37:34 +10002220 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002221 * Note: There is significant duplication with __d_lookup_rcu which is
2222 * required to prevent single threaded performance regressions
2223 * especially on architectures where smp_rmb (in seqcounts) are costly.
2224 * Keep the two functions in sync.
2225 */
2226
2227 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002228 * The hash list is protected using RCU.
2229 *
2230 * Take d_lock when comparing a candidate dentry, to avoid races
2231 * with d_move().
2232 *
2233 * It is possible that concurrent renames can mess up our list
2234 * walk here and result in missing our dentry, resulting in the
2235 * false-negative result. d_lookup() protects against concurrent
2236 * renames using rename_lock seqlock.
2237 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002238 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002239 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 rcu_read_lock();
2241
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002242 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 if (dentry->d_name.hash != hash)
2245 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246
2247 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 if (dentry->d_parent != parent)
2249 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002250 if (d_unhashed(dentry))
2251 goto next;
2252
Al Virod4c91a82016-06-25 23:33:49 -04002253 if (!d_same_name(dentry, parent, name))
2254 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255
Waiman Long98474232013-08-28 18:24:59 -07002256 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002257 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 spin_unlock(&dentry->d_lock);
2259 break;
2260next:
2261 spin_unlock(&dentry->d_lock);
2262 }
2263 rcu_read_unlock();
2264
2265 return found;
2266}
2267
2268/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002269 * d_hash_and_lookup - hash the qstr then search for a dentry
2270 * @dir: Directory to search in
2271 * @name: qstr of name we wish to find
2272 *
Al Viro4f522a22013-02-11 23:20:37 -05002273 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002274 */
2275struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2276{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002277 /*
2278 * Check for a fs-specific hash function. Note that we must
2279 * calculate the standard hash first, as the d_op->d_hash()
2280 * routine may choose to leave the hash value unchanged.
2281 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002282 name->hash = full_name_hash(dir, name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002283 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002284 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002285 if (unlikely(err < 0))
2286 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002287 }
Al Viro4f522a22013-02-11 23:20:37 -05002288 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002289}
Al Viro4f522a22013-02-11 23:20:37 -05002290EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002291
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292/*
2293 * When a file is deleted, we have two options:
2294 * - turn this dentry into a negative dentry
2295 * - unhash this dentry and free it.
2296 *
2297 * Usually, we want to just turn this into
2298 * a negative dentry, but if anybody else is
2299 * currently using the dentry or the inode
2300 * we can't do that and we fall back on removing
2301 * it from the hash queues and waiting for
2302 * it to be deleted later when it has no users
2303 */
2304
2305/**
2306 * d_delete - delete a dentry
2307 * @dentry: The dentry to delete
2308 *
2309 * Turn the dentry into a negative dentry if possible, otherwise
2310 * remove it from the hash queues so it can be deleted later
2311 */
2312
2313void d_delete(struct dentry * dentry)
2314{
Nick Piggin873feea2011-01-07 17:50:06 +11002315 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002316 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 /*
2318 * Are we the only user?
2319 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002320again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002322 inode = dentry->d_inode;
2323 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002324 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002325 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002326 spin_unlock(&dentry->d_lock);
2327 cpu_relax();
2328 goto again;
2329 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002330 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002331 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002332 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 return;
2334 }
2335
2336 if (!d_unhashed(dentry))
2337 __d_drop(dentry);
2338
2339 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002340
2341 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002343EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344
Al Viro15d3c582016-07-29 17:45:21 -04002345static void __d_rehash(struct dentry *entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346{
Al Viro15d3c582016-07-29 17:45:21 -04002347 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002348 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002349 hlist_bl_lock(b);
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002350 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002351 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352}
2353
2354/**
2355 * d_rehash - add an entry back to the hash
2356 * @entry: dentry to add to the hash
2357 *
2358 * Adds a dentry to the hash according to its name.
2359 */
2360
2361void d_rehash(struct dentry * entry)
2362{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 spin_lock(&entry->d_lock);
Al Viro15d3c582016-07-29 17:45:21 -04002364 __d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002367EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368
Al Viro84e710d2016-04-15 00:58:55 -04002369static inline unsigned start_dir_add(struct inode *dir)
2370{
2371
2372 for (;;) {
2373 unsigned n = dir->i_dir_seq;
2374 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2375 return n;
2376 cpu_relax();
2377 }
2378}
2379
2380static inline void end_dir_add(struct inode *dir, unsigned n)
2381{
2382 smp_store_release(&dir->i_dir_seq, n + 2);
2383}
2384
Al Virod9171b92016-04-15 03:33:13 -04002385static void d_wait_lookup(struct dentry *dentry)
2386{
2387 if (d_in_lookup(dentry)) {
2388 DECLARE_WAITQUEUE(wait, current);
2389 add_wait_queue(dentry->d_wait, &wait);
2390 do {
2391 set_current_state(TASK_UNINTERRUPTIBLE);
2392 spin_unlock(&dentry->d_lock);
2393 schedule();
2394 spin_lock(&dentry->d_lock);
2395 } while (d_in_lookup(dentry));
2396 }
2397}
2398
Al Viro94bdd652016-04-15 02:42:04 -04002399struct dentry *d_alloc_parallel(struct dentry *parent,
Al Virod9171b92016-04-15 03:33:13 -04002400 const struct qstr *name,
2401 wait_queue_head_t *wq)
Al Viro94bdd652016-04-15 02:42:04 -04002402{
Al Viro94bdd652016-04-15 02:42:04 -04002403 unsigned int hash = name->hash;
Al Viro94bdd652016-04-15 02:42:04 -04002404 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2405 struct hlist_bl_node *node;
2406 struct dentry *new = d_alloc(parent, name);
2407 struct dentry *dentry;
2408 unsigned seq, r_seq, d_seq;
2409
2410 if (unlikely(!new))
2411 return ERR_PTR(-ENOMEM);
2412
2413retry:
2414 rcu_read_lock();
2415 seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1;
2416 r_seq = read_seqbegin(&rename_lock);
2417 dentry = __d_lookup_rcu(parent, name, &d_seq);
2418 if (unlikely(dentry)) {
2419 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2420 rcu_read_unlock();
2421 goto retry;
2422 }
2423 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2424 rcu_read_unlock();
2425 dput(dentry);
2426 goto retry;
2427 }
2428 rcu_read_unlock();
2429 dput(new);
2430 return dentry;
2431 }
2432 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2433 rcu_read_unlock();
2434 goto retry;
2435 }
2436 hlist_bl_lock(b);
2437 if (unlikely(parent->d_inode->i_dir_seq != seq)) {
2438 hlist_bl_unlock(b);
2439 rcu_read_unlock();
2440 goto retry;
2441 }
Al Viro94bdd652016-04-15 02:42:04 -04002442 /*
2443 * No changes for the parent since the beginning of d_lookup().
2444 * Since all removals from the chain happen with hlist_bl_lock(),
2445 * any potential in-lookup matches are going to stay here until
2446 * we unlock the chain. All fields are stable in everything
2447 * we encounter.
2448 */
2449 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2450 if (dentry->d_name.hash != hash)
2451 continue;
2452 if (dentry->d_parent != parent)
2453 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002454 if (!d_same_name(dentry, parent, name))
2455 continue;
Al Viro94bdd652016-04-15 02:42:04 -04002456 hlist_bl_unlock(b);
Al Viroe7d6ef92016-06-20 01:35:59 -04002457 /* now we can try to grab a reference */
2458 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2459 rcu_read_unlock();
2460 goto retry;
2461 }
2462
2463 rcu_read_unlock();
2464 /*
2465 * somebody is likely to be still doing lookup for it;
2466 * wait for them to finish
2467 */
Al Virod9171b92016-04-15 03:33:13 -04002468 spin_lock(&dentry->d_lock);
2469 d_wait_lookup(dentry);
2470 /*
2471 * it's not in-lookup anymore; in principle we should repeat
2472 * everything from dcache lookup, but it's likely to be what
2473 * d_lookup() would've found anyway. If it is, just return it;
2474 * otherwise we really have to repeat the whole thing.
2475 */
2476 if (unlikely(dentry->d_name.hash != hash))
2477 goto mismatch;
2478 if (unlikely(dentry->d_parent != parent))
2479 goto mismatch;
2480 if (unlikely(d_unhashed(dentry)))
2481 goto mismatch;
Al Virod4c91a82016-06-25 23:33:49 -04002482 if (unlikely(!d_same_name(dentry, parent, name)))
2483 goto mismatch;
Al Virod9171b92016-04-15 03:33:13 -04002484 /* OK, it *is* a hashed match; return it */
2485 spin_unlock(&dentry->d_lock);
Al Viro94bdd652016-04-15 02:42:04 -04002486 dput(new);
2487 return dentry;
2488 }
Al Viroe7d6ef92016-06-20 01:35:59 -04002489 rcu_read_unlock();
Al Viro94bdd652016-04-15 02:42:04 -04002490 /* we can't take ->d_lock here; it's OK, though. */
2491 new->d_flags |= DCACHE_PAR_LOOKUP;
Al Virod9171b92016-04-15 03:33:13 -04002492 new->d_wait = wq;
Al Viro94bdd652016-04-15 02:42:04 -04002493 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2494 hlist_bl_unlock(b);
2495 return new;
Al Virod9171b92016-04-15 03:33:13 -04002496mismatch:
2497 spin_unlock(&dentry->d_lock);
2498 dput(dentry);
2499 goto retry;
Al Viro94bdd652016-04-15 02:42:04 -04002500}
2501EXPORT_SYMBOL(d_alloc_parallel);
2502
Al Viro85c7f812016-04-14 19:52:13 -04002503void __d_lookup_done(struct dentry *dentry)
2504{
Al Viro94bdd652016-04-15 02:42:04 -04002505 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2506 dentry->d_name.hash);
2507 hlist_bl_lock(b);
Al Viro85c7f812016-04-14 19:52:13 -04002508 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
Al Viro94bdd652016-04-15 02:42:04 -04002509 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
Al Virod9171b92016-04-15 03:33:13 -04002510 wake_up_all(dentry->d_wait);
2511 dentry->d_wait = NULL;
Al Viro94bdd652016-04-15 02:42:04 -04002512 hlist_bl_unlock(b);
2513 INIT_HLIST_NODE(&dentry->d_u.d_alias);
Al Virod9171b92016-04-15 03:33:13 -04002514 INIT_LIST_HEAD(&dentry->d_lru);
Al Viro85c7f812016-04-14 19:52:13 -04002515}
2516EXPORT_SYMBOL(__d_lookup_done);
Al Viroed782b52016-03-09 19:52:39 -05002517
2518/* inode->i_lock held if inode is non-NULL */
2519
2520static inline void __d_add(struct dentry *dentry, struct inode *inode)
2521{
Al Viro84e710d2016-04-15 00:58:55 -04002522 struct inode *dir = NULL;
2523 unsigned n;
Al Viro0568d702016-04-14 19:40:56 -04002524 spin_lock(&dentry->d_lock);
Al Viro84e710d2016-04-15 00:58:55 -04002525 if (unlikely(d_in_lookup(dentry))) {
2526 dir = dentry->d_parent->d_inode;
2527 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002528 __d_lookup_done(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002529 }
Al Viroed782b52016-03-09 19:52:39 -05002530 if (inode) {
Al Viro0568d702016-04-14 19:40:56 -04002531 unsigned add_flags = d_flags_for_inode(inode);
2532 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2533 raw_write_seqcount_begin(&dentry->d_seq);
2534 __d_set_inode_and_type(dentry, inode, add_flags);
2535 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04002536 fsnotify_update_flags(dentry);
Al Viroed782b52016-03-09 19:52:39 -05002537 }
Al Viro15d3c582016-07-29 17:45:21 -04002538 __d_rehash(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002539 if (dir)
2540 end_dir_add(dir, n);
Al Viro0568d702016-04-14 19:40:56 -04002541 spin_unlock(&dentry->d_lock);
2542 if (inode)
2543 spin_unlock(&inode->i_lock);
Al Viroed782b52016-03-09 19:52:39 -05002544}
2545
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002546/**
Al Viro34d0d192016-03-08 21:01:03 -05002547 * d_add - add dentry to hash queues
2548 * @entry: dentry to add
2549 * @inode: The inode to attach to this dentry
2550 *
2551 * This adds the entry to the hash queues and initializes @inode.
2552 * The entry was actually filled in earlier during d_alloc().
2553 */
2554
2555void d_add(struct dentry *entry, struct inode *inode)
2556{
Al Virob9680912016-04-11 00:53:26 -04002557 if (inode) {
2558 security_d_instantiate(entry, inode);
Al Viroed782b52016-03-09 19:52:39 -05002559 spin_lock(&inode->i_lock);
Al Virob9680912016-04-11 00:53:26 -04002560 }
Al Viroed782b52016-03-09 19:52:39 -05002561 __d_add(entry, inode);
Al Viro34d0d192016-03-08 21:01:03 -05002562}
2563EXPORT_SYMBOL(d_add);
2564
2565/**
Al Viro668d0cd2016-03-08 12:44:17 -05002566 * d_exact_alias - find and hash an exact unhashed alias
2567 * @entry: dentry to add
2568 * @inode: The inode to go with this dentry
2569 *
2570 * If an unhashed dentry with the same name/parent and desired
2571 * inode already exists, hash and return it. Otherwise, return
2572 * NULL.
2573 *
2574 * Parent directory should be locked.
2575 */
2576struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2577{
2578 struct dentry *alias;
Al Viro668d0cd2016-03-08 12:44:17 -05002579 unsigned int hash = entry->d_name.hash;
2580
2581 spin_lock(&inode->i_lock);
2582 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2583 /*
2584 * Don't need alias->d_lock here, because aliases with
2585 * d_parent == entry->d_parent are not subject to name or
2586 * parent changes, because the parent inode i_mutex is held.
2587 */
2588 if (alias->d_name.hash != hash)
2589 continue;
2590 if (alias->d_parent != entry->d_parent)
2591 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002592 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
Al Viro668d0cd2016-03-08 12:44:17 -05002593 continue;
2594 spin_lock(&alias->d_lock);
2595 if (!d_unhashed(alias)) {
2596 spin_unlock(&alias->d_lock);
2597 alias = NULL;
2598 } else {
2599 __dget_dlock(alias);
Al Viro15d3c582016-07-29 17:45:21 -04002600 __d_rehash(alias);
Al Viro668d0cd2016-03-08 12:44:17 -05002601 spin_unlock(&alias->d_lock);
2602 }
2603 spin_unlock(&inode->i_lock);
2604 return alias;
2605 }
2606 spin_unlock(&inode->i_lock);
2607 return NULL;
2608}
2609EXPORT_SYMBOL(d_exact_alias);
2610
2611/**
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002612 * dentry_update_name_case - update case insensitive dentry with a new name
2613 * @dentry: dentry to be updated
2614 * @name: new name
2615 *
2616 * Update a case insensitive dentry with new case of name.
2617 *
2618 * dentry must have been returned by d_lookup with name @name. Old and new
2619 * name lengths must match (ie. no d_compare which allows mismatched name
2620 * lengths).
2621 *
2622 * Parent inode i_mutex must be held over d_lookup and into this call (to
2623 * keep renames and concurrent inserts, and readdir(2) away).
2624 */
Al Viro9aba36d2016-07-20 22:28:45 -04002625void dentry_update_name_case(struct dentry *dentry, const struct qstr *name)
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002626{
Al Viro59551022016-01-22 15:40:57 -05002627 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002628 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2629
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002630 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002631 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002632 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002633 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002634 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002635}
2636EXPORT_SYMBOL(dentry_update_name_case);
2637
Al Viro8d85b482014-09-29 14:54:27 -04002638static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639{
Al Viro8d85b482014-09-29 14:54:27 -04002640 if (unlikely(dname_external(target))) {
2641 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 /*
2643 * Both external: swap the pointers
2644 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002645 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 } else {
2647 /*
2648 * dentry:internal, target:external. Steal target's
2649 * storage and make target internal.
2650 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002651 memcpy(target->d_iname, dentry->d_name.name,
2652 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 dentry->d_name.name = target->d_name.name;
2654 target->d_name.name = target->d_iname;
2655 }
2656 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002657 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 /*
2659 * dentry:external, target:internal. Give dentry's
2660 * storage to target and make dentry internal
2661 */
2662 memcpy(dentry->d_iname, target->d_name.name,
2663 target->d_name.len + 1);
2664 target->d_name.name = dentry->d_name.name;
2665 dentry->d_name.name = dentry->d_iname;
2666 } else {
2667 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002668 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002670 unsigned int i;
2671 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002672 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2673 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002674 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2675 swap(((long *) &dentry->d_iname)[i],
2676 ((long *) &target->d_iname)[i]);
2677 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 }
2679 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002680 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681}
2682
Al Viro8d85b482014-09-29 14:54:27 -04002683static void copy_name(struct dentry *dentry, struct dentry *target)
2684{
2685 struct external_name *old_name = NULL;
2686 if (unlikely(dname_external(dentry)))
2687 old_name = external_name(dentry);
2688 if (unlikely(dname_external(target))) {
2689 atomic_inc(&external_name(target)->u.count);
2690 dentry->d_name = target->d_name;
2691 } else {
2692 memcpy(dentry->d_iname, target->d_name.name,
2693 target->d_name.len + 1);
2694 dentry->d_name.name = dentry->d_iname;
2695 dentry->d_name.hash_len = target->d_name.hash_len;
2696 }
2697 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2698 kfree_rcu(old_name, u.head);
2699}
2700
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002701static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2702{
2703 /*
2704 * XXXX: do we really need to take target->d_lock?
2705 */
2706 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2707 spin_lock(&target->d_parent->d_lock);
2708 else {
2709 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2710 spin_lock(&dentry->d_parent->d_lock);
2711 spin_lock_nested(&target->d_parent->d_lock,
2712 DENTRY_D_LOCK_NESTED);
2713 } else {
2714 spin_lock(&target->d_parent->d_lock);
2715 spin_lock_nested(&dentry->d_parent->d_lock,
2716 DENTRY_D_LOCK_NESTED);
2717 }
2718 }
2719 if (target < dentry) {
2720 spin_lock_nested(&target->d_lock, 2);
2721 spin_lock_nested(&dentry->d_lock, 3);
2722 } else {
2723 spin_lock_nested(&dentry->d_lock, 2);
2724 spin_lock_nested(&target->d_lock, 3);
2725 }
2726}
2727
Al Viro986c0192014-09-26 23:11:15 -04002728static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002729{
2730 if (target->d_parent != dentry->d_parent)
2731 spin_unlock(&dentry->d_parent->d_lock);
2732 if (target->d_parent != target)
2733 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002734 spin_unlock(&target->d_lock);
2735 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002736}
2737
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002739 * When switching names, the actual string doesn't strictly have to
2740 * be preserved in the target - because we're dropping the target
2741 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002742 * the new name before we switch, unless we are going to rehash
2743 * it. Note that if we *do* unhash the target, we are not allowed
2744 * to rehash it without giving it a new name/hash key - whether
2745 * we swap or overwrite the names here, resulting name won't match
2746 * the reality in filesystem; it's only there for d_path() purposes.
2747 * Note that all of this is happening under rename_lock, so the
2748 * any hash lookup seeing it in the middle of manipulations will
2749 * be discarded anyway. So we do not care what happens to the hash
2750 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002752/*
Al Viro18367502011-07-12 21:42:24 -04002753 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 * @dentry: entry to move
2755 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002756 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757 *
2758 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002759 * dcache entries should not be moved in this way. Caller must hold
2760 * rename_lock, the i_mutex of the source and target directories,
2761 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002763static void __d_move(struct dentry *dentry, struct dentry *target,
2764 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765{
Al Viro84e710d2016-04-15 00:58:55 -04002766 struct inode *dir = NULL;
2767 unsigned n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 if (!dentry->d_inode)
2769 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2770
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002771 BUG_ON(d_ancestor(dentry, target));
2772 BUG_ON(d_ancestor(target, dentry));
2773
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002774 dentry_lock_for_move(dentry, target);
Al Viro84e710d2016-04-15 00:58:55 -04002775 if (unlikely(d_in_lookup(target))) {
2776 dir = target->d_parent->d_inode;
2777 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002778 __d_lookup_done(target);
Al Viro84e710d2016-04-15 00:58:55 -04002779 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780
Nick Piggin31e6b012011-01-07 17:49:52 +11002781 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002782 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002783
Al Viro15d3c582016-07-29 17:45:21 -04002784 /* unhash both */
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002785 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002786 __d_drop(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 __d_drop(target);
2788
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002790 if (exchange)
2791 swap_names(dentry, target);
2792 else
2793 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794
Al Viro15d3c582016-07-29 17:45:21 -04002795 /* rehash in new place(s) */
2796 __d_rehash(dentry);
2797 if (exchange)
2798 __d_rehash(target);
2799
Al Viro63cf4272014-09-26 23:06:14 -04002800 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002802 /* splicing a tree */
Al Viro3d56c252016-06-07 21:26:55 -04002803 dentry->d_flags |= DCACHE_RCUACCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 dentry->d_parent = target->d_parent;
2805 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002806 list_del_init(&target->d_child);
2807 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002809 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002810 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002811 list_move(&target->d_child, &target->d_parent->d_subdirs);
2812 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002813 if (exchange)
Al Viroaffda482016-05-29 18:35:12 -04002814 fsnotify_update_flags(target);
2815 fsnotify_update_flags(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 }
2817
Nick Piggin31e6b012011-01-07 17:49:52 +11002818 write_seqcount_end(&target->d_seq);
2819 write_seqcount_end(&dentry->d_seq);
2820
Al Viro84e710d2016-04-15 00:58:55 -04002821 if (dir)
2822 end_dir_add(dir, n);
Al Viro986c0192014-09-26 23:11:15 -04002823 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002824}
2825
2826/*
2827 * d_move - move a dentry
2828 * @dentry: entry to move
2829 * @target: new dentry
2830 *
2831 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002832 * dcache entries should not be moved in this way. See the locking
2833 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002834 */
2835void d_move(struct dentry *dentry, struct dentry *target)
2836{
2837 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002838 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002840}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002841EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842
Miklos Szeredida1ce062014-04-01 17:08:43 +02002843/*
2844 * d_exchange - exchange two dentries
2845 * @dentry1: first dentry
2846 * @dentry2: second dentry
2847 */
2848void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2849{
2850 write_seqlock(&rename_lock);
2851
2852 WARN_ON(!dentry1->d_inode);
2853 WARN_ON(!dentry2->d_inode);
2854 WARN_ON(IS_ROOT(dentry1));
2855 WARN_ON(IS_ROOT(dentry2));
2856
2857 __d_move(dentry1, dentry2, true);
2858
2859 write_sequnlock(&rename_lock);
2860}
2861
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002862/**
2863 * d_ancestor - search for an ancestor
2864 * @p1: ancestor dentry
2865 * @p2: child dentry
2866 *
2867 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2868 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002869 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002870struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002871{
2872 struct dentry *p;
2873
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002874 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002875 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002876 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002877 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002878 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002879}
2880
2881/*
2882 * This helper attempts to cope with remotely renamed directories
2883 *
2884 * It assumes that the caller is already holding
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002885 * dentry->d_parent->d_inode->i_mutex, and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002886 *
2887 * Note: If ever the locking in lock_rename() changes, then please
2888 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002889 */
Al Virob5ae6b12014-10-12 22:16:02 -04002890static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002891 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002892{
Al Viro9902af72016-04-15 15:08:36 -04002893 struct mutex *m1 = NULL;
2894 struct rw_semaphore *m2 = NULL;
J. Bruce Fields3d330dc2015-02-10 10:55:53 -05002895 int ret = -ESTALE;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002896
2897 /* If alias and dentry share a parent, then no extra locks required */
2898 if (alias->d_parent == dentry->d_parent)
2899 goto out_unalias;
2900
Trond Myklebust9eaef272006-10-21 10:24:20 -07002901 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002902 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2903 goto out_err;
2904 m1 = &dentry->d_sb->s_vfs_rename_mutex;
Al Viro9902af72016-04-15 15:08:36 -04002905 if (!inode_trylock_shared(alias->d_parent->d_inode))
Trond Myklebust9eaef272006-10-21 10:24:20 -07002906 goto out_err;
Al Viro9902af72016-04-15 15:08:36 -04002907 m2 = &alias->d_parent->d_inode->i_rwsem;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002908out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002909 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04002910 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002911out_err:
Trond Myklebust9eaef272006-10-21 10:24:20 -07002912 if (m2)
Al Viro9902af72016-04-15 15:08:36 -04002913 up_read(m2);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002914 if (m1)
2915 mutex_unlock(m1);
2916 return ret;
2917}
2918
David Howells770bfad2006-08-22 20:06:07 -04002919/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002920 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2921 * @inode: the inode which may have a disconnected dentry
2922 * @dentry: a negative dentry which we want to point to the inode.
2923 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002924 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2925 * place of the given dentry and return it, else simply d_add the inode
2926 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002927 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002928 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2929 * we should error out: directories can't have multiple aliases.
2930 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002931 * This is needed in the lookup routine of any filesystem that is exportable
2932 * (via knfsd) so that we can build dcache paths to directories effectively.
2933 *
2934 * If a dentry was found and moved, then it is returned. Otherwise NULL
2935 * is returned. This matches the expected return value of ->lookup.
2936 *
2937 * Cluster filesystems may call this function with a negative, hashed dentry.
2938 * In that case, we know that the inode will be a regular file, and also this
2939 * will only occur during atomic_open. So we need to check for the dentry
2940 * being already hashed only in the final case.
2941 */
2942struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2943{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002944 if (IS_ERR(inode))
2945 return ERR_CAST(inode);
2946
David Howells770bfad2006-08-22 20:06:07 -04002947 BUG_ON(!d_unhashed(dentry));
2948
Al Virode689f52016-03-09 18:05:42 -05002949 if (!inode)
Al Virob5ae6b12014-10-12 22:16:02 -04002950 goto out;
Al Virode689f52016-03-09 18:05:42 -05002951
Al Virob9680912016-04-11 00:53:26 -04002952 security_d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11002953 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002954 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04002955 struct dentry *new = __d_find_any_alias(inode);
2956 if (unlikely(new)) {
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002957 /* The reference to new ensures it remains an alias */
2958 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002959 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002960 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04002961 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002962 dput(new);
2963 new = ERR_PTR(-ELOOP);
2964 pr_warn_ratelimited(
2965 "VFS: Lookup of '%s' in %s %s"
2966 " would have caused loop\n",
2967 dentry->d_name.name,
2968 inode->i_sb->s_type->name,
2969 inode->i_sb->s_id);
2970 } else if (!IS_ROOT(new)) {
2971 int err = __d_unalias(inode, dentry, new);
2972 write_sequnlock(&rename_lock);
2973 if (err) {
2974 dput(new);
2975 new = ERR_PTR(err);
2976 }
Al Viro18367502011-07-12 21:42:24 -04002977 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04002978 __d_move(new, dentry, false);
2979 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002980 }
Al Virob5ae6b12014-10-12 22:16:02 -04002981 iput(inode);
2982 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002983 }
David Howells770bfad2006-08-22 20:06:07 -04002984 }
Al Virob5ae6b12014-10-12 22:16:02 -04002985out:
Al Viroed782b52016-03-09 19:52:39 -05002986 __d_add(dentry, inode);
Al Virob5ae6b12014-10-12 22:16:02 -04002987 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04002988}
Al Virob5ae6b12014-10-12 22:16:02 -04002989EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04002990
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002991static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002992{
2993 *buflen -= namelen;
2994 if (*buflen < 0)
2995 return -ENAMETOOLONG;
2996 *buffer -= namelen;
2997 memcpy(*buffer, str, namelen);
2998 return 0;
2999}
3000
Waiman Long232d2d62013-09-09 12:18:13 -04003001/**
3002 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04003003 * @buffer: buffer pointer
3004 * @buflen: allocated length of the buffer
3005 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04003006 *
3007 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3008 * make sure that either the old or the new name pointer and length are
3009 * fetched. However, there may be mismatch between length and pointer.
3010 * The length cannot be trusted, we need to copy it byte-by-byte until
3011 * the length is reached or a null byte is found. It also prepends "/" at
3012 * the beginning of the name. The sequence number check at the caller will
3013 * retry it again when a d_move() does happen. So any garbage in the buffer
3014 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04003015 *
3016 * Data dependency barrier is needed to make sure that we see that terminating
3017 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04003018 */
Al Viro9aba36d2016-07-20 22:28:45 -04003019static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003020{
Waiman Long232d2d62013-09-09 12:18:13 -04003021 const char *dname = ACCESS_ONCE(name->name);
3022 u32 dlen = ACCESS_ONCE(name->len);
3023 char *p;
3024
Al Viro6d13f692014-09-29 14:46:30 -04003025 smp_read_barrier_depends();
3026
Waiman Long232d2d62013-09-09 12:18:13 -04003027 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04003028 if (*buflen < 0)
3029 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04003030 p = *buffer -= dlen + 1;
3031 *p++ = '/';
3032 while (dlen--) {
3033 char c = *dname++;
3034 if (!c)
3035 break;
3036 *p++ = c;
3037 }
3038 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003039}
3040
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041/**
Randy Dunlap208898c2010-11-18 15:02:49 -08003042 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003043 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003044 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003045 * @buffer: pointer to the end of the buffer
3046 * @buflen: pointer to buffer length
3047 *
Waiman Long18129972013-09-12 10:55:35 -04003048 * The function will first try to write out the pathname without taking any
3049 * lock other than the RCU read lock to make sure that dentries won't go away.
3050 * It only checks the sequence number of the global rename_lock as any change
3051 * in the dentry's d_seq will be preceded by changes in the rename_lock
3052 * sequence number. If the sequence number had been changed, it will restart
3053 * the whole pathname back-tracing sequence again by taking the rename_lock.
3054 * In this case, there is no need to take the RCU read lock as the recursive
3055 * parent pointer references will keep the dentry chain alive as long as no
3056 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003057 */
Al Viro02125a82011-12-05 08:43:34 -05003058static int prepend_path(const struct path *path,
3059 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003060 char **buffer, int *buflen)
3061{
Al Viroede4ceb2013-11-13 07:45:40 -05003062 struct dentry *dentry;
3063 struct vfsmount *vfsmnt;
3064 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003065 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04003066 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04003067 char *bptr;
3068 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003069
Al Viro48f5ec22013-09-09 15:22:25 -04003070 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04003071restart_mnt:
3072 read_seqbegin_or_lock(&mount_lock, &m_seq);
3073 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003074 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003075restart:
3076 bptr = *buffer;
3077 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04003078 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05003079 dentry = path->dentry;
3080 vfsmnt = path->mnt;
3081 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04003082 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003083 while (dentry != root->dentry || vfsmnt != root->mnt) {
3084 struct dentry * parent;
3085
3086 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04003087 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Eric W. Biedermancde93be2015-08-15 13:36:12 -05003088 /* Escaped? */
3089 if (dentry != vfsmnt->mnt_root) {
3090 bptr = *buffer;
3091 blen = *buflen;
3092 error = 3;
3093 break;
3094 }
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003095 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04003096 if (mnt != parent) {
3097 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
3098 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04003099 vfsmnt = &mnt->mnt;
3100 continue;
3101 }
Waiman Long232d2d62013-09-09 12:18:13 -04003102 if (!error)
3103 error = is_mounted(vfsmnt) ? 1 : 2;
3104 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003105 }
3106 parent = dentry->d_parent;
3107 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003108 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003109 if (error)
3110 break;
3111
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003112 dentry = parent;
3113 }
Al Viro48f5ec22013-09-09 15:22:25 -04003114 if (!(seq & 1))
3115 rcu_read_unlock();
3116 if (need_seqretry(&rename_lock, seq)) {
3117 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003118 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003119 }
Al Viro48f5ec22013-09-09 15:22:25 -04003120 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003121
3122 if (!(m_seq & 1))
3123 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04003124 if (need_seqretry(&mount_lock, m_seq)) {
3125 m_seq = 1;
3126 goto restart_mnt;
3127 }
3128 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003129
Waiman Long232d2d62013-09-09 12:18:13 -04003130 if (error >= 0 && bptr == *buffer) {
3131 if (--blen < 0)
3132 error = -ENAMETOOLONG;
3133 else
3134 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003135 }
Waiman Long232d2d62013-09-09 12:18:13 -04003136 *buffer = bptr;
3137 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04003138 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003139}
3140
3141/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003142 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003143 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003144 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07003145 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146 * @buflen: buffer length
3147 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003148 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003149 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003150 * Returns a pointer into the buffer or an error code if the
3151 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08003152 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04003153 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003154 *
Al Viro02125a82011-12-05 08:43:34 -05003155 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003156 */
Al Viro02125a82011-12-05 08:43:34 -05003157char *__d_path(const struct path *path,
3158 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003159 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003160{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003161 char *res = buf + buflen;
3162 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003164 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003165 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04003166
Al Viro02125a82011-12-05 08:43:34 -05003167 if (error < 0)
3168 return ERR_PTR(error);
3169 if (error > 0)
3170 return NULL;
3171 return res;
3172}
3173
3174char *d_absolute_path(const struct path *path,
3175 char *buf, int buflen)
3176{
3177 struct path root = {};
3178 char *res = buf + buflen;
3179 int error;
3180
3181 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003182 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05003183
3184 if (error > 1)
3185 error = -EINVAL;
3186 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003187 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003188 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003189}
Guenter Roeck671662b2016-03-24 10:32:35 -07003190EXPORT_SYMBOL(d_absolute_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003192/*
3193 * same as __d_path but appends "(deleted)" for unlinked files.
3194 */
Al Viro02125a82011-12-05 08:43:34 -05003195static int path_with_deleted(const struct path *path,
3196 const struct path *root,
3197 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003198{
3199 prepend(buf, buflen, "\0", 1);
3200 if (d_unlinked(path->dentry)) {
3201 int error = prepend(buf, buflen, " (deleted)", 10);
3202 if (error)
3203 return error;
3204 }
3205
3206 return prepend_path(path, root, buf, buflen);
3207}
3208
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003209static int prepend_unreachable(char **buffer, int *buflen)
3210{
3211 return prepend(buffer, buflen, "(unreachable)", 13);
3212}
3213
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003214static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3215{
3216 unsigned seq;
3217
3218 do {
3219 seq = read_seqcount_begin(&fs->seq);
3220 *root = fs->root;
3221 } while (read_seqcount_retry(&fs->seq, seq));
3222}
3223
Jan Bluncka03a8a702008-02-14 19:38:32 -08003224/**
3225 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003226 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003227 * @buf: buffer to return value in
3228 * @buflen: buffer length
3229 *
3230 * Convert a dentry into an ASCII path name. If the entry has been deleted
3231 * the string " (deleted)" is appended. Note that this is ambiguous.
3232 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003233 * Returns a pointer into the buffer or an error code if the path was
3234 * too long. Note: Callers should use the returned pointer, not the passed
3235 * in buffer, to use the name! The implementation often starts at an offset
3236 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003237 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003238 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003239 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003240char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003242 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003243 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003244 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003245
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003246 /*
3247 * We have various synthetic filesystems that never get mounted. On
3248 * these filesystems dentries are never used for lookup purposes, and
3249 * thus don't need to be hashed. They also don't need a name until a
3250 * user wants to identify the object in /proc/pid/fd/. The little hack
3251 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003252 *
3253 * Some pseudo inodes are mountable. When they are mounted
3254 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3255 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003256 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003257 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3258 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003259 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003260
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003261 rcu_read_lock();
3262 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003263 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003264 rcu_read_unlock();
3265
Al Viro02125a82011-12-05 08:43:34 -05003266 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003267 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268 return res;
3269}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003270EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271
3272/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003273 * Helper function for dentry_operations.d_dname() members
3274 */
3275char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3276 const char *fmt, ...)
3277{
3278 va_list args;
3279 char temp[64];
3280 int sz;
3281
3282 va_start(args, fmt);
3283 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3284 va_end(args);
3285
3286 if (sz > sizeof(temp) || sz > buflen)
3287 return ERR_PTR(-ENAMETOOLONG);
3288
3289 buffer += buflen - sz;
3290 return memcpy(buffer, temp, sz);
3291}
3292
Al Viro118b2302013-08-24 12:08:17 -04003293char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3294{
3295 char *end = buffer + buflen;
3296 /* these dentries are never renamed, so d_lock is not needed */
3297 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003298 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003299 prepend(&end, &buflen, "/", 1))
3300 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003301 return end;
Al Viro118b2302013-08-24 12:08:17 -04003302}
David Herrmann31bbe162014-01-03 14:09:47 +01003303EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003304
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003305/*
Ram Pai6092d042008-03-27 13:06:20 +01003306 * Write full pathname from the root of the filesystem into the buffer.
3307 */
Al Virof6500802014-01-26 12:37:55 -05003308static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003309{
Al Virof6500802014-01-26 12:37:55 -05003310 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003311 char *end, *retval;
3312 int len, seq = 0;
3313 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003314
Al Virof6500802014-01-26 12:37:55 -05003315 if (buflen < 2)
3316 goto Elong;
3317
Al Viro48f5ec22013-09-09 15:22:25 -04003318 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003319restart:
Al Virof6500802014-01-26 12:37:55 -05003320 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003321 end = buf + buflen;
3322 len = buflen;
3323 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003324 /* Get '/' right */
3325 retval = end-1;
3326 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003327 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003328 while (!IS_ROOT(dentry)) {
3329 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003330
Ram Pai6092d042008-03-27 13:06:20 +01003331 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003332 error = prepend_name(&end, &len, &dentry->d_name);
3333 if (error)
3334 break;
Ram Pai6092d042008-03-27 13:06:20 +01003335
3336 retval = end;
3337 dentry = parent;
3338 }
Al Viro48f5ec22013-09-09 15:22:25 -04003339 if (!(seq & 1))
3340 rcu_read_unlock();
3341 if (need_seqretry(&rename_lock, seq)) {
3342 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003343 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003344 }
3345 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003346 if (error)
3347 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003348 return retval;
3349Elong:
3350 return ERR_PTR(-ENAMETOOLONG);
3351}
Nick Pigginec2447c2011-01-07 17:49:29 +11003352
3353char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3354{
Waiman Long232d2d62013-09-09 12:18:13 -04003355 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003356}
3357EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003358
3359char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3360{
3361 char *p = NULL;
3362 char *retval;
3363
Al Viroc1031352010-06-06 22:31:14 -04003364 if (d_unlinked(dentry)) {
3365 p = buf + buflen;
3366 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3367 goto Elong;
3368 buflen++;
3369 }
3370 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003371 if (!IS_ERR(retval) && p)
3372 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003373 return retval;
3374Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003375 return ERR_PTR(-ENAMETOOLONG);
3376}
3377
Linus Torvalds8b19e342013-09-12 10:35:47 -07003378static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3379 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003380{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003381 unsigned seq;
3382
3383 do {
3384 seq = read_seqcount_begin(&fs->seq);
3385 *root = fs->root;
3386 *pwd = fs->pwd;
3387 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003388}
3389
Ram Pai6092d042008-03-27 13:06:20 +01003390/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 * NOTE! The user-level library version returns a
3392 * character pointer. The kernel system call just
3393 * returns the length of the buffer filled (which
3394 * includes the ending '\0' character), or a negative
3395 * error value. So libc would do something like
3396 *
3397 * char *getcwd(char * buf, size_t size)
3398 * {
3399 * int retval;
3400 *
3401 * retval = sys_getcwd(buf, size);
3402 * if (retval >= 0)
3403 * return buf;
3404 * errno = -retval;
3405 * return NULL;
3406 * }
3407 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003408SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409{
Linus Torvalds552ce542007-02-13 12:08:18 -08003410 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003411 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003412 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413
3414 if (!page)
3415 return -ENOMEM;
3416
Linus Torvalds8b19e342013-09-12 10:35:47 -07003417 rcu_read_lock();
3418 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419
Linus Torvalds552ce542007-02-13 12:08:18 -08003420 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003421 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003422 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003423 char *cwd = page + PATH_MAX;
3424 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003425
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003426 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003427 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003428 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003429
Al Viro02125a82011-12-05 08:43:34 -05003430 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003431 goto out;
3432
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003433 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003434 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003435 error = prepend_unreachable(&cwd, &buflen);
3436 if (error)
3437 goto out;
3438 }
3439
Linus Torvalds552ce542007-02-13 12:08:18 -08003440 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003441 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003442 if (len <= size) {
3443 error = len;
3444 if (copy_to_user(buf, cwd, len))
3445 error = -EFAULT;
3446 }
Nick Piggin949854d2011-01-07 17:49:37 +11003447 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003448 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003449 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450
3451out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003452 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 return error;
3454}
3455
3456/*
3457 * Test whether new_dentry is a subdirectory of old_dentry.
3458 *
3459 * Trivially implemented using the dcache structure
3460 */
3461
3462/**
3463 * is_subdir - is new dentry a subdirectory of old_dentry
3464 * @new_dentry: new dentry
3465 * @old_dentry: old dentry
3466 *
Yaowei Baia6e57872015-11-17 14:40:11 +08003467 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3468 * Returns false otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3470 */
3471
Yaowei Baia6e57872015-11-17 14:40:11 +08003472bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003473{
Yaowei Baia6e57872015-11-17 14:40:11 +08003474 bool result;
Nick Piggin949854d2011-01-07 17:49:37 +11003475 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003476
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003477 if (new_dentry == old_dentry)
Yaowei Baia6e57872015-11-17 14:40:11 +08003478 return true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003479
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003480 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003482 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003483 /*
3484 * Need rcu_readlock to protect against the d_parent trashing
3485 * due to d_move
3486 */
3487 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003488 if (d_ancestor(old_dentry, new_dentry))
Yaowei Baia6e57872015-11-17 14:40:11 +08003489 result = true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003490 else
Yaowei Baia6e57872015-11-17 14:40:11 +08003491 result = false;
Nick Piggin949854d2011-01-07 17:49:37 +11003492 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494
3495 return result;
3496}
3497
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003498static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003500 struct dentry *root = data;
3501 if (dentry != root) {
3502 if (d_unhashed(dentry) || !dentry->d_inode)
3503 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003504
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003505 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3506 dentry->d_flags |= DCACHE_GENOCIDE;
3507 dentry->d_lockref.count--;
3508 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003509 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003510 return D_WALK_CONTINUE;
3511}
Nick Piggin58db63d2011-01-07 17:49:39 +11003512
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003513void d_genocide(struct dentry *parent)
3514{
3515 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516}
3517
Al Viro60545d02013-06-07 01:20:27 -04003518void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519{
Al Viro60545d02013-06-07 01:20:27 -04003520 inode_dec_link_count(inode);
3521 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003522 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003523 !d_unlinked(dentry));
3524 spin_lock(&dentry->d_parent->d_lock);
3525 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3526 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3527 (unsigned long long)inode->i_ino);
3528 spin_unlock(&dentry->d_lock);
3529 spin_unlock(&dentry->d_parent->d_lock);
3530 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531}
Al Viro60545d02013-06-07 01:20:27 -04003532EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003533
3534static __initdata unsigned long dhash_entries;
3535static int __init set_dhash_entries(char *str)
3536{
3537 if (!str)
3538 return 0;
3539 dhash_entries = simple_strtoul(str, &str, 0);
3540 return 1;
3541}
3542__setup("dhash_entries=", set_dhash_entries);
3543
3544static void __init dcache_init_early(void)
3545{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003546 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547
3548 /* If hashes are distributed across NUMA nodes, defer
3549 * hash allocation until vmalloc space is available.
3550 */
3551 if (hashdist)
3552 return;
3553
3554 dentry_hashtable =
3555 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003556 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557 dhash_entries,
3558 13,
3559 HASH_EARLY,
3560 &d_hash_shift,
3561 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003562 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003563 0);
3564
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003565 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003566 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567}
3568
Denis Cheng74bf17c2007-10-16 23:26:30 -07003569static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003571 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003572
3573 /*
3574 * A constructor could be added for stable state like the lists,
3575 * but it is probably not worth it because of the cache nature
3576 * of the dcache.
3577 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003578 dentry_cache = KMEM_CACHE(dentry,
Vladimir Davydov5d097052016-01-14 15:18:21 -08003579 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580
3581 /* Hash may have been set up in dcache_init_early */
3582 if (!hashdist)
3583 return;
3584
3585 dentry_hashtable =
3586 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003587 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003588 dhash_entries,
3589 13,
3590 0,
3591 &d_hash_shift,
3592 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003593 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594 0);
3595
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003596 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003597 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598}
3599
3600/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003601struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003602EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003603
Linus Torvalds1da177e2005-04-16 15:20:36 -07003604EXPORT_SYMBOL(d_genocide);
3605
Linus Torvalds1da177e2005-04-16 15:20:36 -07003606void __init vfs_caches_init_early(void)
3607{
3608 dcache_init_early();
3609 inode_init_early();
3610}
3611
Mel Gorman4248b0d2015-08-06 15:46:20 -07003612void __init vfs_caches_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003613{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003615 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003616
Denis Cheng74bf17c2007-10-16 23:26:30 -07003617 dcache_init();
3618 inode_init();
Mel Gorman4248b0d2015-08-06 15:46:20 -07003619 files_init();
3620 files_maxfiles_init();
Denis Cheng74bf17c2007-10-16 23:26:30 -07003621 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003622 bdev_cache_init();
3623 chrdev_init();
3624}