blob: 4d18d8d755e8e09e9a3fcc15ddc0d2445c1ecde4 [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
Al Viroad25f112017-07-07 14:51:19 -0400280void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
281{
282 spin_lock(&dentry->d_lock);
283 if (unlikely(dname_external(dentry))) {
284 struct external_name *p = external_name(dentry);
285 atomic_inc(&p->u.count);
286 spin_unlock(&dentry->d_lock);
287 name->name = p->name;
288 } else {
Tetsuo Handafc2afa82018-08-17 15:44:34 -0700289 memcpy(name->inline_name, dentry->d_iname,
290 dentry->d_name.len + 1);
Al Viroad25f112017-07-07 14:51:19 -0400291 spin_unlock(&dentry->d_lock);
292 name->name = name->inline_name;
293 }
294}
295EXPORT_SYMBOL(take_dentry_name_snapshot);
296
297void release_dentry_name_snapshot(struct name_snapshot *name)
298{
299 if (unlikely(name->name != name->inline_name)) {
300 struct external_name *p;
301 p = container_of(name->name, struct external_name, name[0]);
302 if (unlikely(atomic_dec_and_test(&p->u.count)))
303 kfree_rcu(p, u.head);
304 }
305}
306EXPORT_SYMBOL(release_dentry_name_snapshot);
307
David Howells4bf46a22015-03-05 14:09:22 +0000308static inline void __d_set_inode_and_type(struct dentry *dentry,
309 struct inode *inode,
310 unsigned type_flags)
311{
312 unsigned flags;
313
314 dentry->d_inode = inode;
David Howells4bf46a22015-03-05 14:09:22 +0000315 flags = READ_ONCE(dentry->d_flags);
316 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
317 flags |= type_flags;
318 WRITE_ONCE(dentry->d_flags, flags);
319}
320
David Howells4bf46a22015-03-05 14:09:22 +0000321static inline void __d_clear_type_and_inode(struct dentry *dentry)
322{
323 unsigned flags = READ_ONCE(dentry->d_flags);
324
325 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
326 WRITE_ONCE(dentry->d_flags, flags);
David Howells4bf46a22015-03-05 14:09:22 +0000327 dentry->d_inode = NULL;
328}
329
Al Virob4f03542014-04-29 23:40:14 -0400330static void dentry_free(struct dentry *dentry)
331{
Al Viro946e51f2014-10-26 19:19:16 -0400332 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400333 if (unlikely(dname_external(dentry))) {
334 struct external_name *p = external_name(dentry);
335 if (likely(atomic_dec_and_test(&p->u.count))) {
336 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
337 return;
338 }
339 }
Al Virob4f03542014-04-29 23:40:14 -0400340 /* if dentry was never visible to RCU, immediate free is OK */
341 if (!(dentry->d_flags & DCACHE_RCUACCESS))
342 __d_free(&dentry->d_u.d_rcu);
343 else
344 call_rcu(&dentry->d_u.d_rcu, __d_free);
345}
346
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347/*
348 * Release the dentry's inode, using the filesystem
Al Viro550dce02016-05-29 20:13:30 -0400349 * d_iput() operation if defined.
Nick Piggin31e6b012011-01-07 17:49:52 +1100350 */
351static void dentry_unlink_inode(struct dentry * dentry)
352 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100353 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100354{
355 struct inode *inode = dentry->d_inode;
Al Viroa528aca2016-02-29 12:12:46 -0500356
Al Viro59199c02018-08-09 10:15:54 -0400357 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +0000358 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400359 hlist_del_init(&dentry->d_u.d_alias);
Al Viro59199c02018-08-09 10:15:54 -0400360 raw_write_seqcount_end(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100361 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100362 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100363 if (!inode->i_nlink)
364 fsnotify_inoderemove(inode);
365 if (dentry->d_op && dentry->d_op->d_iput)
366 dentry->d_op->d_iput(dentry, inode);
367 else
368 iput(inode);
369}
370
371/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400372 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
373 * is in use - which includes both the "real" per-superblock
374 * LRU list _and_ the DCACHE_SHRINK_LIST use.
375 *
376 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
377 * on the shrink list (ie not on the superblock LRU list).
378 *
379 * The per-cpu "nr_dentry_unused" counters are updated with
380 * the DCACHE_LRU_LIST bit.
381 *
382 * These helper functions make sure we always follow the
383 * rules. d_lock must be held by the caller.
384 */
385#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
386static void d_lru_add(struct dentry *dentry)
387{
388 D_FLAG_VERIFY(dentry, 0);
389 dentry->d_flags |= DCACHE_LRU_LIST;
390 this_cpu_inc(nr_dentry_unused);
391 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
392}
393
394static void d_lru_del(struct dentry *dentry)
395{
396 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
397 dentry->d_flags &= ~DCACHE_LRU_LIST;
398 this_cpu_dec(nr_dentry_unused);
399 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
400}
401
402static void d_shrink_del(struct dentry *dentry)
403{
404 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
405 list_del_init(&dentry->d_lru);
406 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
407 this_cpu_dec(nr_dentry_unused);
408}
409
410static void d_shrink_add(struct dentry *dentry, struct list_head *list)
411{
412 D_FLAG_VERIFY(dentry, 0);
413 list_add(&dentry->d_lru, list);
414 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
415 this_cpu_inc(nr_dentry_unused);
416}
417
418/*
419 * These can only be called under the global LRU lock, ie during the
420 * callback for freeing the LRU list. "isolate" removes it from the
421 * LRU lists entirely, while shrink_move moves it to the indicated
422 * private list.
423 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800424static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400425{
426 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
427 dentry->d_flags &= ~DCACHE_LRU_LIST;
428 this_cpu_dec(nr_dentry_unused);
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800429 list_lru_isolate(lru, &dentry->d_lru);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400430}
431
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800432static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
433 struct list_head *list)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400434{
435 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
436 dentry->d_flags |= DCACHE_SHRINK_LIST;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800437 list_lru_isolate_move(lru, &dentry->d_lru, list);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400438}
439
440/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000441 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700442 */
443static void dentry_lru_add(struct dentry *dentry)
444{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400445 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
446 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700447}
448
Miklos Szeredid52b9082007-05-08 00:23:46 -0700449/**
Nick Piggin789680d2011-01-07 17:49:30 +1100450 * d_drop - drop a dentry
451 * @dentry: dentry to drop
452 *
453 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
454 * be found through a VFS lookup any more. Note that this is different from
455 * deleting the dentry - d_delete will try to mark the dentry negative if
456 * possible, giving a successful _negative_ lookup, while d_drop will
457 * just make the cache lookup fail.
458 *
459 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
460 * reason (NFS timeouts or autofs deletes).
461 *
NeilBrown3aa66ba2017-11-10 15:45:41 +1100462 * __d_drop requires dentry->d_lock
463 * ___d_drop doesn't mark dentry as "unhashed"
464 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
Nick Piggin789680d2011-01-07 17:49:30 +1100465 */
NeilBrown3aa66ba2017-11-10 15:45:41 +1100466static void ___d_drop(struct dentry *dentry)
Nick Piggin789680d2011-01-07 17:49:30 +1100467{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700468 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400469 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400470 /*
471 * Hashed dentries are normally on the dentry hashtable,
472 * with the exception of those newly allocated by
473 * d_obtain_alias, which are always IS_ROOT:
474 */
475 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400476 b = &dentry->d_sb->s_anon;
477 else
Linus Torvalds8387ff22016-06-10 07:51:30 -0700478 b = d_hash(dentry->d_name.hash);
Al Virob61625d2013-10-04 11:09:01 -0400479
480 hlist_bl_lock(b);
481 __hlist_bl_del(&dentry->d_hash);
Al Virob61625d2013-10-04 11:09:01 -0400482 hlist_bl_unlock(b);
Al Virod6141462016-07-28 13:05:50 -0400483 /* After this call, in-progress rcu-walk path lookup will fail. */
484 write_seqcount_invalidate(&dentry->d_seq);
Nick Piggin789680d2011-01-07 17:49:30 +1100485 }
486}
NeilBrown3aa66ba2017-11-10 15:45:41 +1100487
488void __d_drop(struct dentry *dentry)
489{
490 ___d_drop(dentry);
491 dentry->d_hash.pprev = NULL;
492}
Nick Piggin789680d2011-01-07 17:49:30 +1100493EXPORT_SYMBOL(__d_drop);
494
495void d_drop(struct dentry *dentry)
496{
Nick Piggin789680d2011-01-07 17:49:30 +1100497 spin_lock(&dentry->d_lock);
498 __d_drop(dentry);
499 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100500}
501EXPORT_SYMBOL(d_drop);
502
Al Viroba65dc52016-06-10 11:32:47 -0400503static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
504{
505 struct dentry *next;
506 /*
507 * Inform d_walk() and shrink_dentry_list() that we are no longer
508 * attached to the dentry tree
509 */
510 dentry->d_flags |= DCACHE_DENTRY_KILLED;
511 if (unlikely(list_empty(&dentry->d_child)))
512 return;
513 __list_del_entry(&dentry->d_child);
514 /*
515 * Cursors can move around the list of children. While we'd been
516 * a normal list member, it didn't matter - ->d_child.next would've
517 * been updated. However, from now on it won't be and for the
518 * things like d_walk() it might end up with a nasty surprise.
519 * Normally d_walk() doesn't care about cursors moving around -
520 * ->d_lock on parent prevents that and since a cursor has no children
521 * of its own, we get through it without ever unlocking the parent.
522 * There is one exception, though - if we ascend from a child that
523 * gets killed as soon as we unlock it, the next sibling is found
524 * using the value left in its ->d_child.next. And if _that_
525 * pointed to a cursor, and cursor got moved (e.g. by lseek())
526 * before d_walk() regains parent->d_lock, we'll end up skipping
527 * everything the cursor had been moved past.
528 *
529 * Solution: make sure that the pointer left behind in ->d_child.next
530 * points to something that won't be moving around. I.e. skip the
531 * cursors.
532 */
533 while (dentry->d_child.next != &parent->d_subdirs) {
534 next = list_entry(dentry->d_child.next, struct dentry, d_child);
535 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
536 break;
537 dentry->d_child.next = next->d_child.next;
538 }
539}
540
Al Viroe55fd012014-05-28 13:51:12 -0400541static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100542{
Al Viro41edf272014-05-01 10:30:00 -0400543 struct dentry *parent = NULL;
544 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400545 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100546 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100547
Linus Torvalds0d984392013-09-08 13:46:52 -0700548 /*
549 * The dentry is now unrecoverably dead to the world.
550 */
551 lockref_mark_dead(&dentry->d_lockref);
552
Sage Weilf0023bc2011-10-28 10:02:42 -0700553 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700554 * inform the fs via d_prune that this dentry is about to be
555 * unhashed and destroyed.
556 */
Al Viro29266202014-05-30 11:39:02 -0400557 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800558 dentry->d_op->d_prune(dentry);
559
Al Viro01b60352014-04-29 23:42:52 -0400560 if (dentry->d_flags & DCACHE_LRU_LIST) {
561 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
562 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400563 }
Nick Piggin77812a12011-01-07 17:49:48 +1100564 /* if it was on the hash then remove it */
565 __d_drop(dentry);
Al Viroba65dc52016-06-10 11:32:47 -0400566 dentry_unlist(dentry, parent);
Al Viro03b3b882014-04-29 15:45:28 -0400567 if (parent)
568 spin_unlock(&parent->d_lock);
Al Viro550dce02016-05-29 20:13:30 -0400569 if (dentry->d_inode)
570 dentry_unlink_inode(dentry);
571 else
572 spin_unlock(&dentry->d_lock);
Al Viro03b3b882014-04-29 15:45:28 -0400573 this_cpu_dec(nr_dentry);
574 if (dentry->d_op && dentry->d_op->d_release)
575 dentry->d_op->d_release(dentry);
576
Al Viro41edf272014-05-01 10:30:00 -0400577 spin_lock(&dentry->d_lock);
578 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
579 dentry->d_flags |= DCACHE_MAY_FREE;
580 can_free = false;
581 }
582 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400583 if (likely(can_free))
584 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400585}
586
587/*
588 * Finish off a dentry we've decided to kill.
589 * dentry->d_lock must be held, returns with it unlocked.
590 * If ref is non-zero, then decrement the refcount too.
591 * Returns dentry requiring refcount drop, or NULL if we're done.
592 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400593static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400594 __releases(dentry->d_lock)
595{
596 struct inode *inode = dentry->d_inode;
597 struct dentry *parent = NULL;
598
599 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
600 goto failed;
601
602 if (!IS_ROOT(dentry)) {
603 parent = dentry->d_parent;
604 if (unlikely(!spin_trylock(&parent->d_lock))) {
605 if (inode)
606 spin_unlock(&inode->i_lock);
607 goto failed;
608 }
609 }
610
611 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400612 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400613
614failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400615 spin_unlock(&dentry->d_lock);
Al Viroe55fd012014-05-28 13:51:12 -0400616 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100617}
618
Al Viro046b9612014-05-29 08:54:52 -0400619static inline struct dentry *lock_parent(struct dentry *dentry)
620{
621 struct dentry *parent = dentry->d_parent;
622 if (IS_ROOT(dentry))
623 return NULL;
Linus Torvalds360f5472015-01-09 15:19:03 -0800624 if (unlikely(dentry->d_lockref.count < 0))
Al Viroc2338f22014-06-12 00:29:13 -0400625 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400626 if (likely(spin_trylock(&parent->d_lock)))
627 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400628 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400629 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400630again:
631 parent = ACCESS_ONCE(dentry->d_parent);
632 spin_lock(&parent->d_lock);
633 /*
634 * We can't blindly lock dentry until we are sure
635 * that we won't violate the locking order.
636 * Any changes of dentry->d_parent must have
637 * been done with parent->d_lock held, so
638 * spin_lock() above is enough of a barrier
639 * for checking if it's still our child.
640 */
641 if (unlikely(parent != dentry->d_parent)) {
642 spin_unlock(&parent->d_lock);
643 goto again;
644 }
Al Viro05f16fe2018-02-23 20:47:17 -0500645 if (parent != dentry) {
Linus Torvalds9f126002014-05-31 09:13:21 -0700646 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro05f16fe2018-02-23 20:47:17 -0500647 if (unlikely(dentry->d_lockref.count < 0)) {
648 spin_unlock(&parent->d_lock);
649 parent = NULL;
650 }
651 } else {
Al Viro046b9612014-05-29 08:54:52 -0400652 parent = NULL;
Al Viro05f16fe2018-02-23 20:47:17 -0500653 }
654 rcu_read_unlock();
Al Viro046b9612014-05-29 08:54:52 -0400655 return parent;
656}
657
Linus Torvalds360f5472015-01-09 15:19:03 -0800658/*
659 * Try to do a lockless dput(), and return whether that was successful.
660 *
661 * If unsuccessful, we return false, having already taken the dentry lock.
662 *
663 * The caller needs to hold the RCU read lock, so that the dentry is
664 * guaranteed to stay around even if the refcount goes down to zero!
665 */
666static inline bool fast_dput(struct dentry *dentry)
667{
668 int ret;
669 unsigned int d_flags;
670
671 /*
672 * If we have a d_op->d_delete() operation, we sould not
Al Viro75a6f822015-07-08 02:42:38 +0100673 * let the dentry count go to zero, so use "put_or_lock".
Linus Torvalds360f5472015-01-09 15:19:03 -0800674 */
675 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
676 return lockref_put_or_lock(&dentry->d_lockref);
677
678 /*
679 * .. otherwise, we can try to just decrement the
680 * lockref optimistically.
681 */
682 ret = lockref_put_return(&dentry->d_lockref);
683
684 /*
685 * If the lockref_put_return() failed due to the lock being held
686 * by somebody else, the fast path has failed. We will need to
687 * get the lock, and then check the count again.
688 */
689 if (unlikely(ret < 0)) {
690 spin_lock(&dentry->d_lock);
691 if (dentry->d_lockref.count > 1) {
692 dentry->d_lockref.count--;
693 spin_unlock(&dentry->d_lock);
694 return 1;
695 }
696 return 0;
697 }
698
699 /*
700 * If we weren't the last ref, we're done.
701 */
702 if (ret)
703 return 1;
704
705 /*
706 * Careful, careful. The reference count went down
707 * to zero, but we don't hold the dentry lock, so
708 * somebody else could get it again, and do another
709 * dput(), and we need to not race with that.
710 *
711 * However, there is a very special and common case
712 * where we don't care, because there is nothing to
713 * do: the dentry is still hashed, it does not have
714 * a 'delete' op, and it's referenced and already on
715 * the LRU list.
716 *
717 * NOTE! Since we aren't locked, these values are
718 * not "stable". However, it is sufficient that at
719 * some point after we dropped the reference the
720 * dentry was hashed and the flags had the proper
721 * value. Other dentry users may have re-gotten
722 * a reference to the dentry and change that, but
723 * our work is done - we can leave the dentry
724 * around with a zero refcount.
725 */
726 smp_rmb();
727 d_flags = ACCESS_ONCE(dentry->d_flags);
Al Viro75a6f822015-07-08 02:42:38 +0100728 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
Linus Torvalds360f5472015-01-09 15:19:03 -0800729
730 /* Nothing to do? Dropping the reference was all we needed? */
731 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
732 return 1;
733
734 /*
735 * Not the fast normal case? Get the lock. We've already decremented
736 * the refcount, but we'll need to re-check the situation after
737 * getting the lock.
738 */
739 spin_lock(&dentry->d_lock);
740
741 /*
742 * Did somebody else grab a reference to it in the meantime, and
743 * we're no longer the last user after all? Alternatively, somebody
744 * else could have killed it and marked it dead. Either way, we
745 * don't need to do anything else.
746 */
747 if (dentry->d_lockref.count) {
748 spin_unlock(&dentry->d_lock);
749 return 1;
750 }
751
752 /*
753 * Re-get the reference we optimistically dropped. We hold the
754 * lock, and we just tested that it was zero, so we can just
755 * set it to 1.
756 */
757 dentry->d_lockref.count = 1;
758 return 0;
759}
760
761
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762/*
763 * This is dput
764 *
765 * This is complicated by the fact that we do not want to put
766 * dentries that are no longer on any hash chain on the unused
767 * list: we'd much rather just get rid of them immediately.
768 *
769 * However, that implies that we have to traverse the dentry
770 * tree upwards to the parents which might _also_ now be
771 * scheduled for deletion (it may have been only waiting for
772 * its last child to go away).
773 *
774 * This tail recursion is done by hand as we don't want to depend
775 * on the compiler to always get this right (gcc generally doesn't).
776 * Real recursion would eat up our stack space.
777 */
778
779/*
780 * dput - release a dentry
781 * @dentry: dentry to release
782 *
783 * Release a dentry. This will drop the usage count and if appropriate
784 * call the dentry unlink method as well as removing it from the queues and
785 * releasing its resources. If the parent dentries were scheduled for release
786 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788void dput(struct dentry *dentry)
789{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700790 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 return;
792
793repeat:
Wei Fang47be6182016-07-06 11:32:20 +0800794 might_sleep();
795
Linus Torvalds360f5472015-01-09 15:19:03 -0800796 rcu_read_lock();
797 if (likely(fast_dput(dentry))) {
798 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 return;
Linus Torvalds360f5472015-01-09 15:19:03 -0800800 }
801
802 /* Slow case: now with the dentry lock held */
803 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
Al Viro85c7f812016-04-14 19:52:13 -0400805 WARN_ON(d_in_lookup(dentry));
806
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700807 /* Unreachable? Get rid of it */
808 if (unlikely(d_unhashed(dentry)))
809 goto kill_it;
810
Al Viro75a6f822015-07-08 02:42:38 +0100811 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
812 goto kill_it;
813
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700814 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100816 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 }
Nick Piggin265ac902010-10-10 05:36:24 -0400818
Linus Torvalds358eec12013-10-31 15:43:02 -0700819 if (!(dentry->d_flags & DCACHE_REFERENCED))
820 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400821 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400822
Waiman Long98474232013-08-28 18:24:59 -0700823 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100824 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 return;
826
Miklos Szeredid52b9082007-05-08 00:23:46 -0700827kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400828 dentry = dentry_kill(dentry);
Wei Fang47be6182016-07-06 11:32:20 +0800829 if (dentry) {
830 cond_resched();
Miklos Szeredid52b9082007-05-08 00:23:46 -0700831 goto repeat;
Wei Fang47be6182016-07-06 11:32:20 +0800832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700834EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100837/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100838static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839{
Waiman Long98474232013-08-28 18:24:59 -0700840 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841}
842
Nick Piggindc0474b2011-01-07 17:49:43 +1100843static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100844{
Waiman Long98474232013-08-28 18:24:59 -0700845 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100846}
847
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100848struct dentry *dget_parent(struct dentry *dentry)
849{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700850 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100851 struct dentry *ret;
852
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700853 /*
854 * Do optimistic parent lookup without any
855 * locking.
856 */
857 rcu_read_lock();
858 ret = ACCESS_ONCE(dentry->d_parent);
859 gotref = lockref_get_not_zero(&ret->d_lockref);
860 rcu_read_unlock();
861 if (likely(gotref)) {
862 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
863 return ret;
864 dput(ret);
865 }
866
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100867repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100868 /*
869 * Don't need rcu_dereference because we re-check it was correct under
870 * the lock.
871 */
872 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100873 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100874 spin_lock(&ret->d_lock);
875 if (unlikely(ret != dentry->d_parent)) {
876 spin_unlock(&ret->d_lock);
877 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100878 goto repeat;
879 }
Nick Piggina734eb42011-01-07 17:49:44 +1100880 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700881 BUG_ON(!ret->d_lockref.count);
882 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100883 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100884 return ret;
885}
886EXPORT_SYMBOL(dget_parent);
887
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888/**
889 * d_find_alias - grab a hashed alias of inode
890 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 *
892 * If inode has a hashed alias, or is a directory and has any alias,
893 * acquire the reference to alias and return it. Otherwise return NULL.
894 * Notice that if inode is a directory there can be only one alias and
895 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800896 * of a filesystem, or if the directory was renamed and d_revalidate
897 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700899 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500900 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500902static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903{
Nick Pigginda502952011-01-07 17:49:33 +1100904 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
Nick Pigginda502952011-01-07 17:49:33 +1100906again:
907 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400908 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100909 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700911 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100912 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500914 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100915 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100916 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 return alias;
918 }
919 }
Nick Pigginda502952011-01-07 17:49:33 +1100920 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 }
Nick Pigginda502952011-01-07 17:49:33 +1100922 if (discon_alias) {
923 alias = discon_alias;
924 spin_lock(&alias->d_lock);
925 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500926 __dget_dlock(alias);
927 spin_unlock(&alias->d_lock);
928 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100929 }
930 spin_unlock(&alias->d_lock);
931 goto again;
932 }
933 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934}
935
Nick Pigginda502952011-01-07 17:49:33 +1100936struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
David Howells214fda12006-03-25 03:06:36 -0800938 struct dentry *de = NULL;
939
Al Virob3d9b7a2012-06-09 13:51:19 -0400940 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100941 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500942 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100943 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800944 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 return de;
946}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700947EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
949/*
950 * Try to kill dentries associated with this inode.
951 * WARNING: you must own a reference to inode.
952 */
953void d_prune_aliases(struct inode *inode)
954{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700955 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100957 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400958 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700960 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400961 struct dentry *parent = lock_parent(dentry);
962 if (likely(!dentry->d_lockref.count)) {
963 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800964 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400965 goto restart;
966 }
967 if (parent)
968 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 }
970 spin_unlock(&dentry->d_lock);
971 }
Nick Piggin873feea2011-01-07 17:50:06 +1100972 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700974EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400976static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977{
Al Viro5c47e6d2014-04-29 16:13:18 -0400978 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700979
Miklos Szeredi60942f22014-05-02 15:38:39 -0400980 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400981 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400982 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100983 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400984 parent = lock_parent(dentry);
985
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000987 * The dispose list is isolated and dentries are not accounted
988 * to the LRU here, so we can simply remove it from the list
989 * here regardless of whether it is referenced or not.
990 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400991 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000992
993 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000995 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800997 if (dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700998 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400999 if (parent)
1000 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 continue;
1002 }
Nick Piggin77812a12011-01-07 17:49:48 +11001003
Al Viro64fd72e2014-05-28 09:48:44 -04001004
1005 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
1006 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
1007 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -04001008 if (parent)
1009 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -04001010 if (can_free)
1011 dentry_free(dentry);
1012 continue;
1013 }
1014
Al Viroff2fde92014-05-28 13:59:13 -04001015 inode = dentry->d_inode;
1016 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001017 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001018 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -04001019 if (parent)
1020 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -04001021 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001022 }
Al Viroff2fde92014-05-28 13:59:13 -04001023
Al Viroff2fde92014-05-28 13:59:13 -04001024 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -04001025
Al Viro5c47e6d2014-04-29 16:13:18 -04001026 /*
1027 * We need to prune ancestors too. This is necessary to prevent
1028 * quadratic behavior of shrink_dcache_parent(), but is also
1029 * expected to be beneficial in reducing dentry cache
1030 * fragmentation.
1031 */
1032 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -04001033 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1034 parent = lock_parent(dentry);
1035 if (dentry->d_lockref.count != 1) {
1036 dentry->d_lockref.count--;
1037 spin_unlock(&dentry->d_lock);
1038 if (parent)
1039 spin_unlock(&parent->d_lock);
1040 break;
1041 }
1042 inode = dentry->d_inode; /* can't be NULL */
1043 if (unlikely(!spin_trylock(&inode->i_lock))) {
1044 spin_unlock(&dentry->d_lock);
1045 if (parent)
1046 spin_unlock(&parent->d_lock);
1047 cpu_relax();
1048 continue;
1049 }
1050 __dentry_kill(dentry);
1051 dentry = parent;
1052 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001054}
1055
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001056static enum lru_status dentry_lru_isolate(struct list_head *item,
1057 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +10001058{
1059 struct list_head *freeable = arg;
1060 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1061
1062
1063 /*
1064 * we are inverting the lru lock/dentry->d_lock here,
1065 * so use a trylock. If we fail to get the lock, just skip
1066 * it
1067 */
1068 if (!spin_trylock(&dentry->d_lock))
1069 return LRU_SKIP;
1070
1071 /*
1072 * Referenced dentries are still in use. If they have active
1073 * counts, just remove them from the LRU. Otherwise give them
1074 * another pass through the LRU.
1075 */
1076 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001077 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001078 spin_unlock(&dentry->d_lock);
1079 return LRU_REMOVED;
1080 }
1081
1082 if (dentry->d_flags & DCACHE_REFERENCED) {
1083 dentry->d_flags &= ~DCACHE_REFERENCED;
1084 spin_unlock(&dentry->d_lock);
1085
1086 /*
1087 * The list move itself will be made by the common LRU code. At
1088 * this point, we've dropped the dentry->d_lock but keep the
1089 * lru lock. This is safe to do, since every list movement is
1090 * protected by the lru lock even if both locks are held.
1091 *
1092 * This is guaranteed by the fact that all LRU management
1093 * functions are intermediated by the LRU API calls like
1094 * list_lru_add and list_lru_del. List movement in this file
1095 * only ever occur through this functions or through callbacks
1096 * like this one, that are called from the LRU API.
1097 *
1098 * The only exceptions to this are functions like
1099 * shrink_dentry_list, and code that first checks for the
1100 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1101 * operating only with stack provided lists after they are
1102 * properly isolated from the main list. It is thus, always a
1103 * local access.
1104 */
1105 return LRU_ROTATE;
1106 }
1107
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001108 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001109 spin_unlock(&dentry->d_lock);
1110
1111 return LRU_REMOVED;
1112}
1113
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001114/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001115 * prune_dcache_sb - shrink the dcache
1116 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001117 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001118 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001119 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1120 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001121 * function.
1122 *
1123 * This function may fail to free any resources if all the dentries are in
1124 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001125 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001126long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001127{
Dave Chinnerf6041562013-08-28 10:18:00 +10001128 LIST_HEAD(dispose);
1129 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001130
Vladimir Davydov503c3582015-02-12 14:58:47 -08001131 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1132 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001133 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001134 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135}
1136
Glauber Costa4e717f52013-08-28 10:18:03 +10001137static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001138 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001139{
Glauber Costa4e717f52013-08-28 10:18:03 +10001140 struct list_head *freeable = arg;
1141 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001142
Glauber Costa4e717f52013-08-28 10:18:03 +10001143 /*
1144 * we are inverting the lru lock/dentry->d_lock here,
1145 * so use a trylock. If we fail to get the lock, just skip
1146 * it
1147 */
1148 if (!spin_trylock(&dentry->d_lock))
1149 return LRU_SKIP;
1150
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001151 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001152 spin_unlock(&dentry->d_lock);
1153
1154 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001155}
1156
Glauber Costa4e717f52013-08-28 10:18:03 +10001157
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001158/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 * shrink_dcache_sb - shrink dcache for a superblock
1160 * @sb: superblock
1161 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001162 * Shrink the dcache for the specified super block. This is used to free
1163 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001165void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166{
Glauber Costa4e717f52013-08-28 10:18:03 +10001167 do {
1168 LIST_HEAD(dispose);
1169
Waiman Long92744322019-01-30 13:52:36 -05001170 list_lru_walk(&sb->s_dentry_lru,
Sahitya Tummalaa9aa6522a2017-07-10 15:50:00 -07001171 dentry_lru_isolate_shrink, &dispose, 1024);
Glauber Costa4e717f52013-08-28 10:18:03 +10001172 shrink_dentry_list(&dispose);
Sahitya Tummalaa9aa6522a2017-07-10 15:50:00 -07001173 cond_resched();
1174 } while (list_lru_count(&sb->s_dentry_lru) > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001176EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001179 * enum d_walk_ret - action to talke during tree walk
1180 * @D_WALK_CONTINUE: contrinue walk
1181 * @D_WALK_QUIT: quit walk
1182 * @D_WALK_NORETRY: quit when retry is needed
1183 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001185enum d_walk_ret {
1186 D_WALK_CONTINUE,
1187 D_WALK_QUIT,
1188 D_WALK_NORETRY,
1189 D_WALK_SKIP,
1190};
1191
1192/**
1193 * d_walk - walk the dentry tree
1194 * @parent: start of walk
1195 * @data: data passed to @enter() and @finish()
1196 * @enter: callback when first entering the dentry
1197 * @finish: callback when successfully finished the walk
1198 *
1199 * The @enter() and @finish() callbacks are called with d_lock held.
1200 */
1201static void d_walk(struct dentry *parent, void *data,
1202 enum d_walk_ret (*enter)(void *, struct dentry *),
1203 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204{
Nick Piggin949854d2011-01-07 17:49:37 +11001205 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001207 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001208 enum d_walk_ret ret;
1209 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001210
Nick Piggin58db63d2011-01-07 17:49:39 +11001211again:
Al Viro48f5ec22013-09-09 15:22:25 -04001212 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001213 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001214 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001215
1216 ret = enter(data, this_parent);
1217 switch (ret) {
1218 case D_WALK_CONTINUE:
1219 break;
1220 case D_WALK_QUIT:
1221 case D_WALK_SKIP:
1222 goto out_unlock;
1223 case D_WALK_NORETRY:
1224 retry = false;
1225 break;
1226 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227repeat:
1228 next = this_parent->d_subdirs.next;
1229resume:
1230 while (next != &this_parent->d_subdirs) {
1231 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001232 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001234
Al Viroba65dc52016-06-10 11:32:47 -04001235 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1236 continue;
1237
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001238 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001239
1240 ret = enter(data, dentry);
1241 switch (ret) {
1242 case D_WALK_CONTINUE:
1243 break;
1244 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001245 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001246 goto out_unlock;
1247 case D_WALK_NORETRY:
1248 retry = false;
1249 break;
1250 case D_WALK_SKIP:
1251 spin_unlock(&dentry->d_lock);
1252 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001253 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001254
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001256 spin_unlock(&this_parent->d_lock);
1257 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001259 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 goto repeat;
1261 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001262 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 }
1264 /*
1265 * All done at this level ... ascend and resume the search.
1266 */
Al Viroca5358e2014-10-26 19:31:10 -04001267 rcu_read_lock();
1268ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001270 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001271 this_parent = child->d_parent;
1272
Al Viro31dec132013-10-25 17:04:27 -04001273 spin_unlock(&child->d_lock);
1274 spin_lock(&this_parent->d_lock);
1275
Al Viroca5358e2014-10-26 19:31:10 -04001276 /* might go back up the wrong parent if we have had a rename. */
1277 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001278 goto rename_retry;
Al Viro21591842015-05-28 23:09:19 -04001279 /* go into the first sibling still alive */
1280 do {
1281 next = child->d_child.next;
Al Viroca5358e2014-10-26 19:31:10 -04001282 if (next == &this_parent->d_subdirs)
1283 goto ascend;
1284 child = list_entry(next, struct dentry, d_child);
Al Viro21591842015-05-28 23:09:19 -04001285 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
Al Viro31dec132013-10-25 17:04:27 -04001286 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 goto resume;
1288 }
Al Viroca5358e2014-10-26 19:31:10 -04001289 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001290 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001291 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001292 if (finish)
1293 finish(data);
1294
1295out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001296 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001297 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001298 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001299
1300rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001301 spin_unlock(&this_parent->d_lock);
1302 rcu_read_unlock();
1303 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001304 if (!retry)
1305 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001306 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001307 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001309
1310/*
1311 * Search for at least 1 mount point in the dentry's subdirs.
1312 * We descend to the next level whenever the d_subdirs
1313 * list is non-empty and continue searching.
1314 */
1315
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001316static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1317{
1318 int *ret = data;
1319 if (d_mountpoint(dentry)) {
1320 *ret = 1;
1321 return D_WALK_QUIT;
1322 }
1323 return D_WALK_CONTINUE;
1324}
1325
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001326/**
1327 * have_submounts - check for mounts over a dentry
1328 * @parent: dentry to check.
1329 *
1330 * Return true if the parent or its subdirectories contain
1331 * a mount point
1332 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001333int have_submounts(struct dentry *parent)
1334{
1335 int ret = 0;
1336
1337 d_walk(parent, &ret, check_mount, NULL);
1338
1339 return ret;
1340}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001341EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
1343/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001344 * Called by mount code to set a mountpoint and check if the mountpoint is
1345 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1346 * subtree can become unreachable).
1347 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001348 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001349 * this reason take rename_lock and d_lock on dentry and ancestors.
1350 */
1351int d_set_mounted(struct dentry *dentry)
1352{
1353 struct dentry *p;
1354 int ret = -ENOENT;
1355 write_seqlock(&rename_lock);
1356 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001357 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001358 spin_lock(&p->d_lock);
1359 if (unlikely(d_unhashed(p))) {
1360 spin_unlock(&p->d_lock);
1361 goto out;
1362 }
1363 spin_unlock(&p->d_lock);
1364 }
1365 spin_lock(&dentry->d_lock);
1366 if (!d_unlinked(dentry)) {
Eric W. Biederman1a62a0f2017-01-03 14:18:43 +13001367 ret = -EBUSY;
1368 if (!d_mountpoint(dentry)) {
1369 dentry->d_flags |= DCACHE_MOUNTED;
1370 ret = 0;
1371 }
Miklos Szeredieed81002013-09-05 14:39:11 +02001372 }
1373 spin_unlock(&dentry->d_lock);
1374out:
1375 write_sequnlock(&rename_lock);
1376 return ret;
1377}
1378
1379/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001380 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 * and move any unused dentries to the end of the unused
1382 * list for prune_dcache(). We descend to the next level
1383 * whenever the d_subdirs list is non-empty and continue
1384 * searching.
1385 *
1386 * It returns zero iff there are no unused children,
1387 * otherwise it returns the number of children moved to
1388 * the end of the unused list. This may not be the total
1389 * number of unused children, because select_parent can
1390 * drop the lock and return early due to latency
1391 * constraints.
1392 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001393
1394struct select_data {
1395 struct dentry *start;
1396 struct list_head dispose;
1397 int found;
1398};
1399
1400static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001402 struct select_data *data = _data;
1403 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001405 if (data->start == dentry)
1406 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
Al Virofe915222014-05-03 00:02:25 -04001408 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001409 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001410 } else {
1411 if (dentry->d_flags & DCACHE_LRU_LIST)
1412 d_lru_del(dentry);
1413 if (!dentry->d_lockref.count) {
1414 d_shrink_add(dentry, &data->dispose);
1415 data->found++;
1416 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 }
1418 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001419 * We can return to the caller if we have found some (this
1420 * ensures forward progress). We'll be coming back to find
1421 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 */
Al Virofe915222014-05-03 00:02:25 -04001423 if (!list_empty(&data->dispose))
1424 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001426 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427}
1428
1429/**
1430 * shrink_dcache_parent - prune dcache
1431 * @parent: parent of entries to prune
1432 *
1433 * Prune the dcache to remove unused children of the parent dentry.
1434 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001435void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001437 for (;;) {
1438 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001440 INIT_LIST_HEAD(&data.dispose);
1441 data.start = parent;
1442 data.found = 0;
1443
1444 d_walk(parent, &data, select_collect, NULL);
1445 if (!data.found)
1446 break;
1447
1448 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001449 cond_resched();
1450 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001452EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453
Al Viro9c8c10e2014-05-02 20:36:10 -04001454static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001455{
Al Viro9c8c10e2014-05-02 20:36:10 -04001456 /* it has busy descendents; complain about those instead */
1457 if (!list_empty(&dentry->d_subdirs))
1458 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001459
Al Viro9c8c10e2014-05-02 20:36:10 -04001460 /* root with refcount 1 is fine */
1461 if (dentry == _data && dentry->d_lockref.count == 1)
1462 return D_WALK_CONTINUE;
1463
1464 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1465 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001466 dentry,
1467 dentry->d_inode ?
1468 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001469 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001470 dentry->d_lockref.count,
1471 dentry->d_sb->s_type->name,
1472 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001473 WARN_ON(1);
1474 return D_WALK_CONTINUE;
1475}
1476
1477static void do_one_tree(struct dentry *dentry)
1478{
1479 shrink_dcache_parent(dentry);
1480 d_walk(dentry, dentry, umount_check, NULL);
1481 d_drop(dentry);
1482 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001483}
1484
1485/*
1486 * destroy the dentries attached to a superblock on unmounting
1487 */
1488void shrink_dcache_for_umount(struct super_block *sb)
1489{
1490 struct dentry *dentry;
1491
Al Viro9c8c10e2014-05-02 20:36:10 -04001492 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001493
1494 dentry = sb->s_root;
1495 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001496 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001497
1498 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001499 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1500 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001501 }
1502}
1503
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001504struct detach_data {
1505 struct select_data select;
1506 struct dentry *mountpoint;
1507};
1508static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001509{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001510 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001511
1512 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001513 __dget_dlock(dentry);
1514 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001515 return D_WALK_QUIT;
1516 }
1517
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001518 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001519}
1520
1521static void check_and_drop(void *_data)
1522{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001523 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001524
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001525 if (!data->mountpoint && !data->select.found)
1526 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001527}
1528
1529/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001530 * d_invalidate - detach submounts, prune dcache, and drop
1531 * @dentry: dentry to invalidate (aka detach, prune and drop)
1532 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001533 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001534 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001535 * The final d_drop is done as an atomic operation relative to
1536 * rename_lock ensuring there are no races with d_set_mounted. This
1537 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001538 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001539void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001540{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001541 /*
1542 * If it's already been dropped, return OK.
1543 */
1544 spin_lock(&dentry->d_lock);
1545 if (d_unhashed(dentry)) {
1546 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001547 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001548 }
1549 spin_unlock(&dentry->d_lock);
1550
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001551 /* Negative dentries can be dropped without further checks */
1552 if (!dentry->d_inode) {
1553 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001554 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001555 }
1556
1557 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001558 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001559
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001560 data.mountpoint = NULL;
1561 INIT_LIST_HEAD(&data.select.dispose);
1562 data.select.start = dentry;
1563 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001564
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001565 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001566
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001567 if (data.select.found)
1568 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001569
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001570 if (data.mountpoint) {
1571 detach_mounts(data.mountpoint);
1572 dput(data.mountpoint);
1573 }
1574
1575 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001576 break;
1577
1578 cond_resched();
1579 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001580}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001581EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001582
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583/**
Al Viroa4464db2011-07-07 15:03:58 -04001584 * __d_alloc - allocate a dcache entry
1585 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 * @name: qstr of the name
1587 *
1588 * Allocates a dentry. It returns %NULL if there is insufficient memory
1589 * available. On a success the dentry is returned. The name passed in is
1590 * copied and the copy passed in may be reused after this call.
1591 */
1592
Al Viroa4464db2011-07-07 15:03:58 -04001593struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594{
1595 struct dentry *dentry;
1596 char *dname;
Miklos Szeredi285b1022016-06-28 11:47:32 +02001597 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598
Mel Gormane12ba742007-10-16 01:25:52 -07001599 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 if (!dentry)
1601 return NULL;
1602
Linus Torvalds6326c712012-05-21 16:14:04 -07001603 /*
1604 * We guarantee that the inline name is always NUL-terminated.
1605 * This way the memcpy() done by the name switching in rename
1606 * will still always have a NUL at the end, even if we might
1607 * be overwriting an internal NUL character
1608 */
1609 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Al Viro798434b2016-03-24 20:38:43 -04001610 if (unlikely(!name)) {
1611 static const struct qstr anon = QSTR_INIT("/", 1);
1612 name = &anon;
1613 dname = dentry->d_iname;
1614 } else if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001615 size_t size = offsetof(struct external_name, name[1]);
Vladimir Davydov5d097052016-01-14 15:18:21 -08001616 struct external_name *p = kmalloc(size + name->len,
1617 GFP_KERNEL_ACCOUNT);
Al Viro8d85b482014-09-29 14:54:27 -04001618 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 kmem_cache_free(dentry_cache, dentry);
1620 return NULL;
1621 }
Al Viro8d85b482014-09-29 14:54:27 -04001622 atomic_set(&p->u.count, 1);
1623 dname = p->name;
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -08001624 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1625 kasan_unpoison_shadow(dname,
1626 round_up(name->len + 1, sizeof(unsigned long)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 } else {
1628 dname = dentry->d_iname;
1629 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
1631 dentry->d_name.len = name->len;
1632 dentry->d_name.hash = name->hash;
1633 memcpy(dname, name->name, name->len);
1634 dname[name->len] = 0;
1635
Linus Torvalds6326c712012-05-21 16:14:04 -07001636 /* Make sure we always see the terminating NUL character */
1637 smp_wmb();
1638 dentry->d_name.name = dname;
1639
Waiman Long98474232013-08-28 18:24:59 -07001640 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001641 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001643 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001645 dentry->d_parent = dentry;
1646 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 dentry->d_op = NULL;
1648 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001649 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 INIT_LIST_HEAD(&dentry->d_lru);
1651 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001652 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1653 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001654 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Miklos Szeredi285b1022016-06-28 11:47:32 +02001656 if (dentry->d_op && dentry->d_op->d_init) {
1657 err = dentry->d_op->d_init(dentry);
1658 if (err) {
1659 if (dname_external(dentry))
1660 kfree(external_name(dentry));
1661 kmem_cache_free(dentry_cache, dentry);
1662 return NULL;
1663 }
1664 }
1665
Nick Piggin3e880fb2011-01-07 17:49:19 +11001666 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001667
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 return dentry;
1669}
Al Viroa4464db2011-07-07 15:03:58 -04001670
1671/**
1672 * d_alloc - allocate a dcache entry
1673 * @parent: parent of entry to allocate
1674 * @name: qstr of the name
1675 *
1676 * Allocates a dentry. It returns %NULL if there is insufficient memory
1677 * available. On a success the dentry is returned. The name passed in is
1678 * copied and the copy passed in may be reused after this call.
1679 */
1680struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1681{
1682 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1683 if (!dentry)
1684 return NULL;
Al Viro3d56c252016-06-07 21:26:55 -04001685 dentry->d_flags |= DCACHE_RCUACCESS;
Al Viroa4464db2011-07-07 15:03:58 -04001686 spin_lock(&parent->d_lock);
1687 /*
1688 * don't need child lock because it is not subject
1689 * to concurrency here
1690 */
1691 __dget_dlock(parent);
1692 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001693 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001694 spin_unlock(&parent->d_lock);
1695
1696 return dentry;
1697}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001698EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699
Al Viroba65dc52016-06-10 11:32:47 -04001700struct dentry *d_alloc_cursor(struct dentry * parent)
1701{
1702 struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
1703 if (dentry) {
1704 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1705 dentry->d_parent = dget(parent);
1706 }
1707 return dentry;
1708}
1709
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001710/**
1711 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1712 * @sb: the superblock
1713 * @name: qstr of the name
1714 *
1715 * For a filesystem that just pins its dentries in memory and never
1716 * performs lookups at all, return an unhashed IS_ROOT dentry.
1717 */
Nick Piggin4b936882011-01-07 17:50:07 +11001718struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1719{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001720 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001721}
1722EXPORT_SYMBOL(d_alloc_pseudo);
1723
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1725{
1726 struct qstr q;
1727
1728 q.name = name;
Linus Torvalds8387ff22016-06-10 07:51:30 -07001729 q.hash_len = hashlen_string(parent, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 return d_alloc(parent, &q);
1731}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001732EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
Nick Pigginfb045ad2011-01-07 17:49:55 +11001734void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1735{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001736 WARN_ON_ONCE(dentry->d_op);
1737 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001738 DCACHE_OP_COMPARE |
1739 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001740 DCACHE_OP_WEAK_REVALIDATE |
David Howells4bacc9c2015-06-18 14:32:31 +01001741 DCACHE_OP_DELETE |
Miklos Szeredid101a122016-03-26 16:14:37 -04001742 DCACHE_OP_REAL));
Nick Pigginfb045ad2011-01-07 17:49:55 +11001743 dentry->d_op = op;
1744 if (!op)
1745 return;
1746 if (op->d_hash)
1747 dentry->d_flags |= DCACHE_OP_HASH;
1748 if (op->d_compare)
1749 dentry->d_flags |= DCACHE_OP_COMPARE;
1750 if (op->d_revalidate)
1751 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001752 if (op->d_weak_revalidate)
1753 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001754 if (op->d_delete)
1755 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001756 if (op->d_prune)
1757 dentry->d_flags |= DCACHE_OP_PRUNE;
Miklos Szeredid101a122016-03-26 16:14:37 -04001758 if (op->d_real)
1759 dentry->d_flags |= DCACHE_OP_REAL;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001760
1761}
1762EXPORT_SYMBOL(d_set_d_op);
1763
David Howellsdf1a0852015-01-29 12:02:28 +00001764
1765/*
1766 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1767 * @dentry - The dentry to mark
1768 *
1769 * Mark a dentry as falling through to the lower layer (as set with
1770 * d_pin_lower()). This flag may be recorded on the medium.
1771 */
1772void d_set_fallthru(struct dentry *dentry)
1773{
1774 spin_lock(&dentry->d_lock);
1775 dentry->d_flags |= DCACHE_FALLTHRU;
1776 spin_unlock(&dentry->d_lock);
1777}
1778EXPORT_SYMBOL(d_set_fallthru);
1779
David Howellsb18825a2013-09-12 19:22:53 +01001780static unsigned d_flags_for_inode(struct inode *inode)
1781{
David Howells44bdb5e2015-01-29 12:02:29 +00001782 unsigned add_flags = DCACHE_REGULAR_TYPE;
David Howellsb18825a2013-09-12 19:22:53 +01001783
1784 if (!inode)
1785 return DCACHE_MISS_TYPE;
1786
1787 if (S_ISDIR(inode->i_mode)) {
1788 add_flags = DCACHE_DIRECTORY_TYPE;
1789 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1790 if (unlikely(!inode->i_op->lookup))
1791 add_flags = DCACHE_AUTODIR_TYPE;
1792 else
1793 inode->i_opflags |= IOP_LOOKUP;
1794 }
David Howells44bdb5e2015-01-29 12:02:29 +00001795 goto type_determined;
David Howellsb18825a2013-09-12 19:22:53 +01001796 }
1797
David Howells44bdb5e2015-01-29 12:02:29 +00001798 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
Al Viro6b255392015-11-17 10:20:54 -05001799 if (unlikely(inode->i_op->get_link)) {
David Howells44bdb5e2015-01-29 12:02:29 +00001800 add_flags = DCACHE_SYMLINK_TYPE;
1801 goto type_determined;
1802 }
1803 inode->i_opflags |= IOP_NOFOLLOW;
1804 }
1805
1806 if (unlikely(!S_ISREG(inode->i_mode)))
1807 add_flags = DCACHE_SPECIAL_TYPE;
1808
1809type_determined:
David Howellsb18825a2013-09-12 19:22:53 +01001810 if (unlikely(IS_AUTOMOUNT(inode)))
1811 add_flags |= DCACHE_NEED_AUTOMOUNT;
1812 return add_flags;
1813}
1814
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001815static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1816{
David Howellsb18825a2013-09-12 19:22:53 +01001817 unsigned add_flags = d_flags_for_inode(inode);
Al Viro85c7f812016-04-14 19:52:13 -04001818 WARN_ON(d_in_lookup(dentry));
David Howellsb18825a2013-09-12 19:22:53 +01001819
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001820 spin_lock(&dentry->d_lock);
Al Virode689f52016-03-09 18:05:42 -05001821 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
Al Viroa528aca2016-02-29 12:12:46 -05001822 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +00001823 __d_set_inode_and_type(dentry, inode, add_flags);
Al Viroa528aca2016-02-29 12:12:46 -05001824 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04001825 fsnotify_update_flags(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001826 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001827}
1828
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829/**
1830 * d_instantiate - fill in inode information for a dentry
1831 * @entry: dentry to complete
1832 * @inode: inode to attach to this dentry
1833 *
1834 * Fill in inode information in the entry.
1835 *
1836 * This turns negative dentries into productive full members
1837 * of society.
1838 *
1839 * NOTE! This assumes that the inode count has been incremented
1840 * (or otherwise set) by the caller to indicate that it is now
1841 * in use by the dcache.
1842 */
1843
1844void d_instantiate(struct dentry *entry, struct inode * inode)
1845{
Al Viro946e51f2014-10-26 19:19:16 -04001846 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Al Virode689f52016-03-09 18:05:42 -05001847 if (inode) {
Al Virob9680912016-04-11 00:53:26 -04001848 security_d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001849 spin_lock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001850 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001851 spin_unlock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05001852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001854EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
Al Viro2d2d3f12018-05-04 08:23:01 -04001856/*
1857 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1858 * with lockdep-related part of unlock_new_inode() done before
1859 * anything else. Use that instead of open-coding d_instantiate()/
1860 * unlock_new_inode() combinations.
1861 */
1862void d_instantiate_new(struct dentry *entry, struct inode *inode)
1863{
1864 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1865 BUG_ON(!inode);
1866 lockdep_annotate_inode_mutex_key(inode);
1867 security_d_instantiate(entry, inode);
1868 spin_lock(&inode->i_lock);
1869 __d_instantiate(entry, inode);
1870 WARN_ON(!(inode->i_state & I_NEW));
1871 inode->i_state &= ~I_NEW;
1872 smp_mb();
1873 wake_up_bit(&inode->i_state, __I_NEW);
1874 spin_unlock(&inode->i_lock);
1875}
1876EXPORT_SYMBOL(d_instantiate_new);
1877
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878/**
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001879 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1880 * @entry: dentry to complete
1881 * @inode: inode to attach to this dentry
1882 *
1883 * Fill in inode information in the entry. If a directory alias is found, then
1884 * return an error (and drop inode). Together with d_materialise_unique() this
1885 * guarantees that a directory inode may never have more than one alias.
1886 */
1887int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1888{
Al Viro946e51f2014-10-26 19:19:16 -04001889 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001890
Al Virob9680912016-04-11 00:53:26 -04001891 security_d_instantiate(entry, inode);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001892 spin_lock(&inode->i_lock);
1893 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1894 spin_unlock(&inode->i_lock);
1895 iput(inode);
1896 return -EBUSY;
1897 }
1898 __d_instantiate(entry, inode);
1899 spin_unlock(&inode->i_lock);
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001900
1901 return 0;
1902}
1903EXPORT_SYMBOL(d_instantiate_no_diralias);
1904
Al Viroadc0e912012-01-08 16:49:21 -05001905struct dentry *d_make_root(struct inode *root_inode)
1906{
1907 struct dentry *res = NULL;
1908
1909 if (root_inode) {
Al Viro798434b2016-03-24 20:38:43 -04001910 res = __d_alloc(root_inode->i_sb, NULL);
Al Virocfac7df2018-08-06 09:03:58 -04001911 if (res) {
1912 res->d_flags |= DCACHE_RCUACCESS;
Al Viroadc0e912012-01-08 16:49:21 -05001913 d_instantiate(res, root_inode);
Al Virocfac7df2018-08-06 09:03:58 -04001914 } else {
Al Viroadc0e912012-01-08 16:49:21 -05001915 iput(root_inode);
Al Virocfac7df2018-08-06 09:03:58 -04001916 }
Al Viroadc0e912012-01-08 16:49:21 -05001917 }
1918 return res;
1919}
1920EXPORT_SYMBOL(d_make_root);
1921
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001922static struct dentry * __d_find_any_alias(struct inode *inode)
1923{
1924 struct dentry *alias;
1925
Al Virob3d9b7a2012-06-09 13:51:19 -04001926 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001927 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001928 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001929 __dget(alias);
1930 return alias;
1931}
1932
Sage Weil46f72b32012-01-10 09:04:37 -08001933/**
1934 * d_find_any_alias - find any alias for a given inode
1935 * @inode: inode to find an alias for
1936 *
1937 * If any aliases exist for the given inode, take and return a
1938 * reference for one of them. If no aliases exist, return %NULL.
1939 */
1940struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001941{
1942 struct dentry *de;
1943
1944 spin_lock(&inode->i_lock);
1945 de = __d_find_any_alias(inode);
1946 spin_unlock(&inode->i_lock);
1947 return de;
1948}
Sage Weil46f72b32012-01-10 09:04:37 -08001949EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001950
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001951static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001952{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001953 struct dentry *tmp;
1954 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001955 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001956
1957 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001958 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001959 if (IS_ERR(inode))
1960 return ERR_CAST(inode);
1961
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001962 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001963 if (res)
1964 goto out_iput;
1965
Al Viro798434b2016-03-24 20:38:43 -04001966 tmp = __d_alloc(inode->i_sb, NULL);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001967 if (!tmp) {
1968 res = ERR_PTR(-ENOMEM);
1969 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001970 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001971
Al Virob9680912016-04-11 00:53:26 -04001972 security_d_instantiate(tmp, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001973 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001974 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001975 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001976 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001977 dput(tmp);
1978 goto out_iput;
1979 }
1980
1981 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001982 add_flags = d_flags_for_inode(inode);
1983
1984 if (disconnected)
1985 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001986
Christoph Hellwig9308a612008-08-11 15:49:12 +02001987 spin_lock(&tmp->d_lock);
David Howells4bf46a22015-03-05 14:09:22 +00001988 __d_set_inode_and_type(tmp, inode, add_flags);
Al Viro946e51f2014-10-26 19:19:16 -04001989 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001990 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001991 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001992 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001993 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001994 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001995
Christoph Hellwig9308a612008-08-11 15:49:12 +02001996 return tmp;
1997
1998 out_iput:
1999 iput(inode);
2000 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02002001}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002002
2003/**
2004 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2005 * @inode: inode to allocate the dentry for
2006 *
2007 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2008 * similar open by handle operations. The returned dentry may be anonymous,
2009 * or may have a full name (if the inode was already in the cache).
2010 *
2011 * When called on a directory inode, we must ensure that the inode only ever
2012 * has one dentry. If a dentry is found, that is returned instead of
2013 * allocating a new one.
2014 *
2015 * On successful return, the reference to the inode has been transferred
2016 * to the dentry. In case of an error the reference on the inode is released.
2017 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2018 * be passed in and the error will be propagated to the return value,
2019 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2020 */
2021struct dentry *d_obtain_alias(struct inode *inode)
2022{
2023 return __d_obtain_alias(inode, 1);
2024}
Benny Halevyadc48722009-02-27 14:02:59 -08002025EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
2027/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002028 * d_obtain_root - find or allocate a dentry for a given inode
2029 * @inode: inode to allocate the dentry for
2030 *
2031 * Obtain an IS_ROOT dentry for the root of a filesystem.
2032 *
2033 * We must ensure that directory inodes only ever have one dentry. If a
2034 * dentry is found, that is returned instead of allocating a new one.
2035 *
2036 * On successful return, the reference to the inode has been transferred
2037 * to the dentry. In case of an error the reference on the inode is
2038 * released. A %NULL or IS_ERR inode may be passed in and will be the
2039 * error will be propagate to the return value, with a %NULL @inode
2040 * replaced by ERR_PTR(-ESTALE).
2041 */
2042struct dentry *d_obtain_root(struct inode *inode)
2043{
2044 return __d_obtain_alias(inode, 0);
2045}
2046EXPORT_SYMBOL(d_obtain_root);
2047
2048/**
Barry Naujok94035402008-05-21 16:50:46 +10002049 * d_add_ci - lookup or allocate new dentry with case-exact name
2050 * @inode: the inode case-insensitive lookup has found
2051 * @dentry: the negative dentry that was passed to the parent's lookup func
2052 * @name: the case-exact name to be associated with the returned dentry
2053 *
2054 * This is to avoid filling the dcache with case-insensitive names to the
2055 * same inode, only the actual correct case is stored in the dcache for
2056 * case-insensitive filesystems.
2057 *
2058 * For a case-insensitive lookup match and if the the case-exact dentry
2059 * already exists in in the dcache, use it and return it.
2060 *
2061 * If no entry exists with the exact case name, allocate new dentry with
2062 * the exact case, and return the spliced entry.
2063 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02002064struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10002065 struct qstr *name)
2066{
Al Virod9171b92016-04-15 03:33:13 -04002067 struct dentry *found, *res;
Barry Naujok94035402008-05-21 16:50:46 +10002068
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002069 /*
2070 * First check if a dentry matching the name already exists,
2071 * if not go ahead and create it now.
2072 */
Barry Naujok94035402008-05-21 16:50:46 +10002073 found = d_hash_and_lookup(dentry->d_parent, name);
Al Virod9171b92016-04-15 03:33:13 -04002074 if (found) {
2075 iput(inode);
2076 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002077 }
Al Virod9171b92016-04-15 03:33:13 -04002078 if (d_in_lookup(dentry)) {
2079 found = d_alloc_parallel(dentry->d_parent, name,
2080 dentry->d_wait);
2081 if (IS_ERR(found) || !d_in_lookup(found)) {
2082 iput(inode);
2083 return found;
2084 }
2085 } else {
2086 found = d_alloc(dentry->d_parent, name);
2087 if (!found) {
2088 iput(inode);
2089 return ERR_PTR(-ENOMEM);
2090 }
2091 }
2092 res = d_splice_alias(inode, found);
2093 if (res) {
2094 dput(found);
2095 return res;
2096 }
Al Viro4f522a22013-02-11 23:20:37 -05002097 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002098}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002099EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002101
Al Virod4c91a82016-06-25 23:33:49 -04002102static inline bool d_same_name(const struct dentry *dentry,
2103 const struct dentry *parent,
2104 const struct qstr *name)
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002105{
Al Virod4c91a82016-06-25 23:33:49 -04002106 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2107 if (dentry->d_name.len != name->len)
2108 return false;
2109 return dentry_cmp(dentry, name->name, name->len) == 0;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002110 }
Al Viro6fa67e72016-07-31 16:37:25 -04002111 return parent->d_op->d_compare(dentry,
Al Virod4c91a82016-06-25 23:33:49 -04002112 dentry->d_name.len, dentry->d_name.name,
2113 name) == 0;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002114}
2115
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002117 * __d_lookup_rcu - search for a dentry (racy, store-free)
2118 * @parent: parent dentry
2119 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002120 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002121 * Returns: dentry, or NULL
2122 *
2123 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2124 * resolution (store-free path walking) design described in
2125 * Documentation/filesystems/path-lookup.txt.
2126 *
2127 * This is not to be used outside core vfs.
2128 *
2129 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2130 * held, and rcu_read_lock held. The returned dentry must not be stored into
2131 * without taking d_lock and checking d_seq sequence count against @seq
2132 * returned here.
2133 *
Linus Torvalds15570082013-09-02 11:38:06 -07002134 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002135 * function.
2136 *
2137 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2138 * the returned dentry, so long as its parent's seqlock is checked after the
2139 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2140 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002141 *
2142 * NOTE! The caller *has* to check the resulting dentry against the sequence
2143 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002144 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002145struct dentry *__d_lookup_rcu(const struct dentry *parent,
2146 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002147 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002148{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002149 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002150 const unsigned char *str = name->name;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002151 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002152 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002153 struct dentry *dentry;
2154
2155 /*
2156 * Note: There is significant duplication with __d_lookup_rcu which is
2157 * required to prevent single threaded performance regressions
2158 * especially on architectures where smp_rmb (in seqcounts) are costly.
2159 * Keep the two functions in sync.
2160 */
2161
2162 /*
2163 * The hash list is protected using RCU.
2164 *
2165 * Carefully use d_seq when comparing a candidate dentry, to avoid
2166 * races with d_move().
2167 *
2168 * It is possible that concurrent renames can mess up our list
2169 * walk here and result in missing our dentry, resulting in the
2170 * false-negative result. d_lookup() protects against concurrent
2171 * renames using rename_lock seqlock.
2172 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002173 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002174 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002175 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002176 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002177
Nick Piggin31e6b012011-01-07 17:49:52 +11002178seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002179 /*
2180 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002181 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002182 *
2183 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002184 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002185 *
2186 * NOTE! We do a "raw" seqcount_begin here. That means that
2187 * we don't wait for the sequence count to stabilize if it
2188 * is in the middle of a sequence change. If we do the slow
2189 * dentry compare, we will do seqretries until it is stable,
2190 * and if we end up with a successful lookup, we actually
2191 * want to exit RCU lookup anyway.
Al Virod4c91a82016-06-25 23:33:49 -04002192 *
2193 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2194 * we are still guaranteed NUL-termination of ->d_name.name.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002195 */
2196 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002197 if (dentry->d_parent != parent)
2198 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002199 if (d_unhashed(dentry))
2200 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002201
2202 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Al Virod4c91a82016-06-25 23:33:49 -04002203 int tlen;
2204 const char *tname;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002205 if (dentry->d_name.hash != hashlen_hash(hashlen))
2206 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002207 tlen = dentry->d_name.len;
2208 tname = dentry->d_name.name;
2209 /* we want a consistent (name,len) pair */
2210 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2211 cpu_relax();
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002212 goto seqretry;
2213 }
Al Viro6fa67e72016-07-31 16:37:25 -04002214 if (parent->d_op->d_compare(dentry,
Al Virod4c91a82016-06-25 23:33:49 -04002215 tlen, tname, name) != 0)
2216 continue;
2217 } else {
2218 if (dentry->d_name.hash_len != hashlen)
2219 continue;
2220 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2221 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002222 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002223 *seqp = seq;
Al Virod4c91a82016-06-25 23:33:49 -04002224 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002225 }
2226 return NULL;
2227}
2228
2229/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 * d_lookup - search for a dentry
2231 * @parent: parent dentry
2232 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002233 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002235 * d_lookup searches the children of the parent dentry for the name in
2236 * question. If the dentry is found its reference count is incremented and the
2237 * dentry is returned. The caller must use dput to free the entry when it has
2238 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 */
Al Viroda2d8452013-01-24 18:29:34 -05002240struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241{
Nick Piggin31e6b012011-01-07 17:49:52 +11002242 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002243 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
Daeseok Younb8314f92014-08-11 11:46:53 +09002245 do {
2246 seq = read_seqbegin(&rename_lock);
2247 dentry = __d_lookup(parent, name);
2248 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 break;
2250 } while (read_seqretry(&rename_lock, seq));
2251 return dentry;
2252}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002253EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002254
Nick Piggin31e6b012011-01-07 17:49:52 +11002255/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002256 * __d_lookup - search for a dentry (racy)
2257 * @parent: parent dentry
2258 * @name: qstr of name we wish to find
2259 * Returns: dentry, or NULL
2260 *
2261 * __d_lookup is like d_lookup, however it may (rarely) return a
2262 * false-negative result due to unrelated rename activity.
2263 *
2264 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2265 * however it must be used carefully, eg. with a following d_lookup in
2266 * the case of failure.
2267 *
2268 * __d_lookup callers must be commented.
2269 */
Al Viroa713ca22013-01-24 18:27:00 -05002270struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 unsigned int hash = name->hash;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002273 struct hlist_bl_head *b = d_hash(hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002274 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002275 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002276 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
Nick Pigginb04f7842010-08-18 04:37:34 +10002278 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002279 * Note: There is significant duplication with __d_lookup_rcu which is
2280 * required to prevent single threaded performance regressions
2281 * especially on architectures where smp_rmb (in seqcounts) are costly.
2282 * Keep the two functions in sync.
2283 */
2284
2285 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002286 * The hash list is protected using RCU.
2287 *
2288 * Take d_lock when comparing a candidate dentry, to avoid races
2289 * with d_move().
2290 *
2291 * It is possible that concurrent renames can mess up our list
2292 * walk here and result in missing our dentry, resulting in the
2293 * false-negative result. d_lookup() protects against concurrent
2294 * renames using rename_lock seqlock.
2295 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002296 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002297 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 rcu_read_lock();
2299
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002300 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 if (dentry->d_name.hash != hash)
2303 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
2305 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 if (dentry->d_parent != parent)
2307 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002308 if (d_unhashed(dentry))
2309 goto next;
2310
Al Virod4c91a82016-06-25 23:33:49 -04002311 if (!d_same_name(dentry, parent, name))
2312 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313
Waiman Long98474232013-08-28 18:24:59 -07002314 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002315 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 spin_unlock(&dentry->d_lock);
2317 break;
2318next:
2319 spin_unlock(&dentry->d_lock);
2320 }
2321 rcu_read_unlock();
2322
2323 return found;
2324}
2325
2326/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002327 * d_hash_and_lookup - hash the qstr then search for a dentry
2328 * @dir: Directory to search in
2329 * @name: qstr of name we wish to find
2330 *
Al Viro4f522a22013-02-11 23:20:37 -05002331 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002332 */
2333struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2334{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002335 /*
2336 * Check for a fs-specific hash function. Note that we must
2337 * calculate the standard hash first, as the d_op->d_hash()
2338 * routine may choose to leave the hash value unchanged.
2339 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002340 name->hash = full_name_hash(dir, name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002341 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002342 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002343 if (unlikely(err < 0))
2344 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002345 }
Al Viro4f522a22013-02-11 23:20:37 -05002346 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002347}
Al Viro4f522a22013-02-11 23:20:37 -05002348EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002349
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350/*
2351 * When a file is deleted, we have two options:
2352 * - turn this dentry into a negative dentry
2353 * - unhash this dentry and free it.
2354 *
2355 * Usually, we want to just turn this into
2356 * a negative dentry, but if anybody else is
2357 * currently using the dentry or the inode
2358 * we can't do that and we fall back on removing
2359 * it from the hash queues and waiting for
2360 * it to be deleted later when it has no users
2361 */
2362
2363/**
2364 * d_delete - delete a dentry
2365 * @dentry: The dentry to delete
2366 *
2367 * Turn the dentry into a negative dentry if possible, otherwise
2368 * remove it from the hash queues so it can be deleted later
2369 */
2370
2371void d_delete(struct dentry * dentry)
2372{
Nick Piggin873feea2011-01-07 17:50:06 +11002373 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002374 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 /*
2376 * Are we the only user?
2377 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002378again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002380 inode = dentry->d_inode;
2381 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002382 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002383 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002384 spin_unlock(&dentry->d_lock);
2385 cpu_relax();
2386 goto again;
2387 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002388 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002389 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002390 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 return;
2392 }
2393
2394 if (!d_unhashed(dentry))
2395 __d_drop(dentry);
2396
2397 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002398
2399 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002401EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
Al Viro15d3c582016-07-29 17:45:21 -04002403static void __d_rehash(struct dentry *entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404{
Al Viro15d3c582016-07-29 17:45:21 -04002405 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
NeilBrown3aa66ba2017-11-10 15:45:41 +11002406
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002407 hlist_bl_lock(b);
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002408 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002409 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410}
2411
2412/**
2413 * d_rehash - add an entry back to the hash
2414 * @entry: dentry to add to the hash
2415 *
2416 * Adds a dentry to the hash according to its name.
2417 */
2418
2419void d_rehash(struct dentry * entry)
2420{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 spin_lock(&entry->d_lock);
Al Viro15d3c582016-07-29 17:45:21 -04002422 __d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002425EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426
Al Viro84e710d2016-04-15 00:58:55 -04002427static inline unsigned start_dir_add(struct inode *dir)
2428{
2429
2430 for (;;) {
2431 unsigned n = dir->i_dir_seq;
2432 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2433 return n;
2434 cpu_relax();
2435 }
2436}
2437
2438static inline void end_dir_add(struct inode *dir, unsigned n)
2439{
2440 smp_store_release(&dir->i_dir_seq, n + 2);
2441}
2442
Al Virod9171b92016-04-15 03:33:13 -04002443static void d_wait_lookup(struct dentry *dentry)
2444{
2445 if (d_in_lookup(dentry)) {
2446 DECLARE_WAITQUEUE(wait, current);
2447 add_wait_queue(dentry->d_wait, &wait);
2448 do {
2449 set_current_state(TASK_UNINTERRUPTIBLE);
2450 spin_unlock(&dentry->d_lock);
2451 schedule();
2452 spin_lock(&dentry->d_lock);
2453 } while (d_in_lookup(dentry));
2454 }
2455}
2456
Al Viro94bdd652016-04-15 02:42:04 -04002457struct dentry *d_alloc_parallel(struct dentry *parent,
Al Virod9171b92016-04-15 03:33:13 -04002458 const struct qstr *name,
2459 wait_queue_head_t *wq)
Al Viro94bdd652016-04-15 02:42:04 -04002460{
Al Viro94bdd652016-04-15 02:42:04 -04002461 unsigned int hash = name->hash;
Al Viro94bdd652016-04-15 02:42:04 -04002462 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2463 struct hlist_bl_node *node;
2464 struct dentry *new = d_alloc(parent, name);
2465 struct dentry *dentry;
2466 unsigned seq, r_seq, d_seq;
2467
2468 if (unlikely(!new))
2469 return ERR_PTR(-ENOMEM);
2470
2471retry:
2472 rcu_read_lock();
Will Deaconbcefedb2018-02-19 14:55:54 +00002473 seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
Al Viro94bdd652016-04-15 02:42:04 -04002474 r_seq = read_seqbegin(&rename_lock);
2475 dentry = __d_lookup_rcu(parent, name, &d_seq);
2476 if (unlikely(dentry)) {
2477 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2478 rcu_read_unlock();
2479 goto retry;
2480 }
2481 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2482 rcu_read_unlock();
2483 dput(dentry);
2484 goto retry;
2485 }
2486 rcu_read_unlock();
2487 dput(new);
2488 return dentry;
2489 }
2490 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2491 rcu_read_unlock();
2492 goto retry;
2493 }
Will Deaconbcefedb2018-02-19 14:55:54 +00002494
2495 if (unlikely(seq & 1)) {
2496 rcu_read_unlock();
2497 goto retry;
2498 }
2499
Al Viro94bdd652016-04-15 02:42:04 -04002500 hlist_bl_lock(b);
Will Deacon527ed412018-02-19 14:55:55 +00002501 if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
Al Viro94bdd652016-04-15 02:42:04 -04002502 hlist_bl_unlock(b);
2503 rcu_read_unlock();
2504 goto retry;
2505 }
Al Viro94bdd652016-04-15 02:42:04 -04002506 /*
2507 * No changes for the parent since the beginning of d_lookup().
2508 * Since all removals from the chain happen with hlist_bl_lock(),
2509 * any potential in-lookup matches are going to stay here until
2510 * we unlock the chain. All fields are stable in everything
2511 * we encounter.
2512 */
2513 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2514 if (dentry->d_name.hash != hash)
2515 continue;
2516 if (dentry->d_parent != parent)
2517 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002518 if (!d_same_name(dentry, parent, name))
2519 continue;
Al Viro94bdd652016-04-15 02:42:04 -04002520 hlist_bl_unlock(b);
Al Viroe7d6ef92016-06-20 01:35:59 -04002521 /* now we can try to grab a reference */
2522 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2523 rcu_read_unlock();
2524 goto retry;
2525 }
2526
2527 rcu_read_unlock();
2528 /*
2529 * somebody is likely to be still doing lookup for it;
2530 * wait for them to finish
2531 */
Al Virod9171b92016-04-15 03:33:13 -04002532 spin_lock(&dentry->d_lock);
2533 d_wait_lookup(dentry);
2534 /*
2535 * it's not in-lookup anymore; in principle we should repeat
2536 * everything from dcache lookup, but it's likely to be what
2537 * d_lookup() would've found anyway. If it is, just return it;
2538 * otherwise we really have to repeat the whole thing.
2539 */
2540 if (unlikely(dentry->d_name.hash != hash))
2541 goto mismatch;
2542 if (unlikely(dentry->d_parent != parent))
2543 goto mismatch;
2544 if (unlikely(d_unhashed(dentry)))
2545 goto mismatch;
Al Virod4c91a82016-06-25 23:33:49 -04002546 if (unlikely(!d_same_name(dentry, parent, name)))
2547 goto mismatch;
Al Virod9171b92016-04-15 03:33:13 -04002548 /* OK, it *is* a hashed match; return it */
2549 spin_unlock(&dentry->d_lock);
Al Viro94bdd652016-04-15 02:42:04 -04002550 dput(new);
2551 return dentry;
2552 }
Al Viroe7d6ef92016-06-20 01:35:59 -04002553 rcu_read_unlock();
Al Viro94bdd652016-04-15 02:42:04 -04002554 /* we can't take ->d_lock here; it's OK, though. */
2555 new->d_flags |= DCACHE_PAR_LOOKUP;
Al Virod9171b92016-04-15 03:33:13 -04002556 new->d_wait = wq;
Al Viro94bdd652016-04-15 02:42:04 -04002557 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2558 hlist_bl_unlock(b);
2559 return new;
Al Virod9171b92016-04-15 03:33:13 -04002560mismatch:
2561 spin_unlock(&dentry->d_lock);
2562 dput(dentry);
2563 goto retry;
Al Viro94bdd652016-04-15 02:42:04 -04002564}
2565EXPORT_SYMBOL(d_alloc_parallel);
2566
Al Viro85c7f812016-04-14 19:52:13 -04002567void __d_lookup_done(struct dentry *dentry)
2568{
Al Viro94bdd652016-04-15 02:42:04 -04002569 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2570 dentry->d_name.hash);
2571 hlist_bl_lock(b);
Al Viro85c7f812016-04-14 19:52:13 -04002572 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
Al Viro94bdd652016-04-15 02:42:04 -04002573 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
Al Virod9171b92016-04-15 03:33:13 -04002574 wake_up_all(dentry->d_wait);
2575 dentry->d_wait = NULL;
Al Viro94bdd652016-04-15 02:42:04 -04002576 hlist_bl_unlock(b);
2577 INIT_HLIST_NODE(&dentry->d_u.d_alias);
Al Virod9171b92016-04-15 03:33:13 -04002578 INIT_LIST_HEAD(&dentry->d_lru);
Al Viro85c7f812016-04-14 19:52:13 -04002579}
2580EXPORT_SYMBOL(__d_lookup_done);
Al Viroed782b52016-03-09 19:52:39 -05002581
2582/* inode->i_lock held if inode is non-NULL */
2583
2584static inline void __d_add(struct dentry *dentry, struct inode *inode)
2585{
Al Viro84e710d2016-04-15 00:58:55 -04002586 struct inode *dir = NULL;
2587 unsigned n;
Al Viro0568d702016-04-14 19:40:56 -04002588 spin_lock(&dentry->d_lock);
Al Viro84e710d2016-04-15 00:58:55 -04002589 if (unlikely(d_in_lookup(dentry))) {
2590 dir = dentry->d_parent->d_inode;
2591 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002592 __d_lookup_done(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002593 }
Al Viroed782b52016-03-09 19:52:39 -05002594 if (inode) {
Al Viro0568d702016-04-14 19:40:56 -04002595 unsigned add_flags = d_flags_for_inode(inode);
2596 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2597 raw_write_seqcount_begin(&dentry->d_seq);
2598 __d_set_inode_and_type(dentry, inode, add_flags);
2599 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04002600 fsnotify_update_flags(dentry);
Al Viroed782b52016-03-09 19:52:39 -05002601 }
Al Viro15d3c582016-07-29 17:45:21 -04002602 __d_rehash(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002603 if (dir)
2604 end_dir_add(dir, n);
Al Viro0568d702016-04-14 19:40:56 -04002605 spin_unlock(&dentry->d_lock);
2606 if (inode)
2607 spin_unlock(&inode->i_lock);
Al Viroed782b52016-03-09 19:52:39 -05002608}
2609
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002610/**
Al Viro34d0d192016-03-08 21:01:03 -05002611 * d_add - add dentry to hash queues
2612 * @entry: dentry to add
2613 * @inode: The inode to attach to this dentry
2614 *
2615 * This adds the entry to the hash queues and initializes @inode.
2616 * The entry was actually filled in earlier during d_alloc().
2617 */
2618
2619void d_add(struct dentry *entry, struct inode *inode)
2620{
Al Virob9680912016-04-11 00:53:26 -04002621 if (inode) {
2622 security_d_instantiate(entry, inode);
Al Viroed782b52016-03-09 19:52:39 -05002623 spin_lock(&inode->i_lock);
Al Virob9680912016-04-11 00:53:26 -04002624 }
Al Viroed782b52016-03-09 19:52:39 -05002625 __d_add(entry, inode);
Al Viro34d0d192016-03-08 21:01:03 -05002626}
2627EXPORT_SYMBOL(d_add);
2628
2629/**
Al Viro668d0cd2016-03-08 12:44:17 -05002630 * d_exact_alias - find and hash an exact unhashed alias
2631 * @entry: dentry to add
2632 * @inode: The inode to go with this dentry
2633 *
2634 * If an unhashed dentry with the same name/parent and desired
2635 * inode already exists, hash and return it. Otherwise, return
2636 * NULL.
2637 *
2638 * Parent directory should be locked.
2639 */
2640struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2641{
2642 struct dentry *alias;
Al Viro668d0cd2016-03-08 12:44:17 -05002643 unsigned int hash = entry->d_name.hash;
2644
2645 spin_lock(&inode->i_lock);
2646 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2647 /*
2648 * Don't need alias->d_lock here, because aliases with
2649 * d_parent == entry->d_parent are not subject to name or
2650 * parent changes, because the parent inode i_mutex is held.
2651 */
2652 if (alias->d_name.hash != hash)
2653 continue;
2654 if (alias->d_parent != entry->d_parent)
2655 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002656 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
Al Viro668d0cd2016-03-08 12:44:17 -05002657 continue;
2658 spin_lock(&alias->d_lock);
2659 if (!d_unhashed(alias)) {
2660 spin_unlock(&alias->d_lock);
2661 alias = NULL;
2662 } else {
2663 __dget_dlock(alias);
Al Viro15d3c582016-07-29 17:45:21 -04002664 __d_rehash(alias);
Al Viro668d0cd2016-03-08 12:44:17 -05002665 spin_unlock(&alias->d_lock);
2666 }
2667 spin_unlock(&inode->i_lock);
2668 return alias;
2669 }
2670 spin_unlock(&inode->i_lock);
2671 return NULL;
2672}
2673EXPORT_SYMBOL(d_exact_alias);
2674
2675/**
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002676 * dentry_update_name_case - update case insensitive dentry with a new name
2677 * @dentry: dentry to be updated
2678 * @name: new name
2679 *
2680 * Update a case insensitive dentry with new case of name.
2681 *
2682 * dentry must have been returned by d_lookup with name @name. Old and new
2683 * name lengths must match (ie. no d_compare which allows mismatched name
2684 * lengths).
2685 *
2686 * Parent inode i_mutex must be held over d_lookup and into this call (to
2687 * keep renames and concurrent inserts, and readdir(2) away).
2688 */
Al Viro9aba36d2016-07-20 22:28:45 -04002689void dentry_update_name_case(struct dentry *dentry, const struct qstr *name)
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002690{
Al Viro59551022016-01-22 15:40:57 -05002691 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002692 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2693
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002694 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002695 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002696 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002697 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002698 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002699}
2700EXPORT_SYMBOL(dentry_update_name_case);
2701
Al Viro8d85b482014-09-29 14:54:27 -04002702static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703{
Al Viro8d85b482014-09-29 14:54:27 -04002704 if (unlikely(dname_external(target))) {
2705 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 /*
2707 * Both external: swap the pointers
2708 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002709 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 } else {
2711 /*
2712 * dentry:internal, target:external. Steal target's
2713 * storage and make target internal.
2714 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002715 memcpy(target->d_iname, dentry->d_name.name,
2716 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 dentry->d_name.name = target->d_name.name;
2718 target->d_name.name = target->d_iname;
2719 }
2720 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002721 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 /*
2723 * dentry:external, target:internal. Give dentry's
2724 * storage to target and make dentry internal
2725 */
2726 memcpy(dentry->d_iname, target->d_name.name,
2727 target->d_name.len + 1);
2728 target->d_name.name = dentry->d_name.name;
2729 dentry->d_name.name = dentry->d_iname;
2730 } else {
2731 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002732 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002734 unsigned int i;
2735 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002736 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2737 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002738 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2739 swap(((long *) &dentry->d_iname)[i],
2740 ((long *) &target->d_iname)[i]);
2741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 }
2743 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002744 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745}
2746
Al Viro8d85b482014-09-29 14:54:27 -04002747static void copy_name(struct dentry *dentry, struct dentry *target)
2748{
2749 struct external_name *old_name = NULL;
2750 if (unlikely(dname_external(dentry)))
2751 old_name = external_name(dentry);
2752 if (unlikely(dname_external(target))) {
2753 atomic_inc(&external_name(target)->u.count);
2754 dentry->d_name = target->d_name;
2755 } else {
2756 memcpy(dentry->d_iname, target->d_name.name,
2757 target->d_name.len + 1);
2758 dentry->d_name.name = dentry->d_iname;
2759 dentry->d_name.hash_len = target->d_name.hash_len;
2760 }
2761 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2762 kfree_rcu(old_name, u.head);
2763}
2764
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002765static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2766{
2767 /*
2768 * XXXX: do we really need to take target->d_lock?
2769 */
2770 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2771 spin_lock(&target->d_parent->d_lock);
2772 else {
2773 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2774 spin_lock(&dentry->d_parent->d_lock);
2775 spin_lock_nested(&target->d_parent->d_lock,
2776 DENTRY_D_LOCK_NESTED);
2777 } else {
2778 spin_lock(&target->d_parent->d_lock);
2779 spin_lock_nested(&dentry->d_parent->d_lock,
2780 DENTRY_D_LOCK_NESTED);
2781 }
2782 }
2783 if (target < dentry) {
2784 spin_lock_nested(&target->d_lock, 2);
2785 spin_lock_nested(&dentry->d_lock, 3);
2786 } else {
2787 spin_lock_nested(&dentry->d_lock, 2);
2788 spin_lock_nested(&target->d_lock, 3);
2789 }
2790}
2791
Al Viro986c0192014-09-26 23:11:15 -04002792static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002793{
2794 if (target->d_parent != dentry->d_parent)
2795 spin_unlock(&dentry->d_parent->d_lock);
2796 if (target->d_parent != target)
2797 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002798 spin_unlock(&target->d_lock);
2799 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002800}
2801
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002803 * When switching names, the actual string doesn't strictly have to
2804 * be preserved in the target - because we're dropping the target
2805 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002806 * the new name before we switch, unless we are going to rehash
2807 * it. Note that if we *do* unhash the target, we are not allowed
2808 * to rehash it without giving it a new name/hash key - whether
2809 * we swap or overwrite the names here, resulting name won't match
2810 * the reality in filesystem; it's only there for d_path() purposes.
2811 * Note that all of this is happening under rename_lock, so the
2812 * any hash lookup seeing it in the middle of manipulations will
2813 * be discarded anyway. So we do not care what happens to the hash
2814 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002816/*
Al Viro18367502011-07-12 21:42:24 -04002817 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818 * @dentry: entry to move
2819 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002820 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 *
2822 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002823 * dcache entries should not be moved in this way. Caller must hold
2824 * rename_lock, the i_mutex of the source and target directories,
2825 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002827static void __d_move(struct dentry *dentry, struct dentry *target,
2828 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829{
Al Viro84e710d2016-04-15 00:58:55 -04002830 struct inode *dir = NULL;
2831 unsigned n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 if (!dentry->d_inode)
2833 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2834
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002835 BUG_ON(d_ancestor(dentry, target));
2836 BUG_ON(d_ancestor(target, dentry));
2837
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002838 dentry_lock_for_move(dentry, target);
Al Viro84e710d2016-04-15 00:58:55 -04002839 if (unlikely(d_in_lookup(target))) {
2840 dir = target->d_parent->d_inode;
2841 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002842 __d_lookup_done(target);
Al Viro84e710d2016-04-15 00:58:55 -04002843 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844
Nick Piggin31e6b012011-01-07 17:49:52 +11002845 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002846 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002847
Al Viro15d3c582016-07-29 17:45:21 -04002848 /* unhash both */
NeilBrown3aa66ba2017-11-10 15:45:41 +11002849 /* ___d_drop does write_seqcount_barrier, but they're OK to nest. */
2850 ___d_drop(dentry);
2851 ___d_drop(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002854 if (exchange)
2855 swap_names(dentry, target);
2856 else
2857 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858
Al Viro15d3c582016-07-29 17:45:21 -04002859 /* rehash in new place(s) */
2860 __d_rehash(dentry);
2861 if (exchange)
2862 __d_rehash(target);
NeilBrown3aa66ba2017-11-10 15:45:41 +11002863 else
2864 target->d_hash.pprev = NULL;
Al Viro15d3c582016-07-29 17:45:21 -04002865
Al Viro63cf4272014-09-26 23:06:14 -04002866 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002868 /* splicing a tree */
Al Viro3d56c252016-06-07 21:26:55 -04002869 dentry->d_flags |= DCACHE_RCUACCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 dentry->d_parent = target->d_parent;
2871 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002872 list_del_init(&target->d_child);
2873 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002875 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002876 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002877 list_move(&target->d_child, &target->d_parent->d_subdirs);
2878 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002879 if (exchange)
Al Viroaffda482016-05-29 18:35:12 -04002880 fsnotify_update_flags(target);
2881 fsnotify_update_flags(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 }
2883
Nick Piggin31e6b012011-01-07 17:49:52 +11002884 write_seqcount_end(&target->d_seq);
2885 write_seqcount_end(&dentry->d_seq);
2886
Al Viro84e710d2016-04-15 00:58:55 -04002887 if (dir)
2888 end_dir_add(dir, n);
Al Viro986c0192014-09-26 23:11:15 -04002889 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002890}
2891
2892/*
2893 * d_move - move a dentry
2894 * @dentry: entry to move
2895 * @target: new dentry
2896 *
2897 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002898 * dcache entries should not be moved in this way. See the locking
2899 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002900 */
2901void d_move(struct dentry *dentry, struct dentry *target)
2902{
2903 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002904 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002906}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002907EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908
Miklos Szeredida1ce062014-04-01 17:08:43 +02002909/*
2910 * d_exchange - exchange two dentries
2911 * @dentry1: first dentry
2912 * @dentry2: second dentry
2913 */
2914void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2915{
2916 write_seqlock(&rename_lock);
2917
2918 WARN_ON(!dentry1->d_inode);
2919 WARN_ON(!dentry2->d_inode);
2920 WARN_ON(IS_ROOT(dentry1));
2921 WARN_ON(IS_ROOT(dentry2));
2922
2923 __d_move(dentry1, dentry2, true);
2924
2925 write_sequnlock(&rename_lock);
2926}
2927
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002928/**
2929 * d_ancestor - search for an ancestor
2930 * @p1: ancestor dentry
2931 * @p2: child dentry
2932 *
2933 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2934 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002935 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002936struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002937{
2938 struct dentry *p;
2939
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002940 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002941 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002942 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002943 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002944 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002945}
2946
2947/*
2948 * This helper attempts to cope with remotely renamed directories
2949 *
2950 * It assumes that the caller is already holding
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002951 * dentry->d_parent->d_inode->i_mutex, and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002952 *
2953 * Note: If ever the locking in lock_rename() changes, then please
2954 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002955 */
Al Virob5ae6b12014-10-12 22:16:02 -04002956static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002957 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002958{
Al Viro9902af72016-04-15 15:08:36 -04002959 struct mutex *m1 = NULL;
2960 struct rw_semaphore *m2 = NULL;
J. Bruce Fields3d330dc2015-02-10 10:55:53 -05002961 int ret = -ESTALE;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002962
2963 /* If alias and dentry share a parent, then no extra locks required */
2964 if (alias->d_parent == dentry->d_parent)
2965 goto out_unalias;
2966
Trond Myklebust9eaef272006-10-21 10:24:20 -07002967 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002968 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2969 goto out_err;
2970 m1 = &dentry->d_sb->s_vfs_rename_mutex;
Al Viro9902af72016-04-15 15:08:36 -04002971 if (!inode_trylock_shared(alias->d_parent->d_inode))
Trond Myklebust9eaef272006-10-21 10:24:20 -07002972 goto out_err;
Al Viro9902af72016-04-15 15:08:36 -04002973 m2 = &alias->d_parent->d_inode->i_rwsem;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002974out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002975 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04002976 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002977out_err:
Trond Myklebust9eaef272006-10-21 10:24:20 -07002978 if (m2)
Al Viro9902af72016-04-15 15:08:36 -04002979 up_read(m2);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002980 if (m1)
2981 mutex_unlock(m1);
2982 return ret;
2983}
2984
David Howells770bfad2006-08-22 20:06:07 -04002985/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002986 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2987 * @inode: the inode which may have a disconnected dentry
2988 * @dentry: a negative dentry which we want to point to the inode.
2989 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002990 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2991 * place of the given dentry and return it, else simply d_add the inode
2992 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002993 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002994 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2995 * we should error out: directories can't have multiple aliases.
2996 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002997 * This is needed in the lookup routine of any filesystem that is exportable
2998 * (via knfsd) so that we can build dcache paths to directories effectively.
2999 *
3000 * If a dentry was found and moved, then it is returned. Otherwise NULL
3001 * is returned. This matches the expected return value of ->lookup.
3002 *
3003 * Cluster filesystems may call this function with a negative, hashed dentry.
3004 * In that case, we know that the inode will be a regular file, and also this
3005 * will only occur during atomic_open. So we need to check for the dentry
3006 * being already hashed only in the final case.
3007 */
3008struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3009{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05003010 if (IS_ERR(inode))
3011 return ERR_CAST(inode);
3012
David Howells770bfad2006-08-22 20:06:07 -04003013 BUG_ON(!d_unhashed(dentry));
3014
Al Virode689f52016-03-09 18:05:42 -05003015 if (!inode)
Al Virob5ae6b12014-10-12 22:16:02 -04003016 goto out;
Al Virode689f52016-03-09 18:05:42 -05003017
Al Virob9680912016-04-11 00:53:26 -04003018 security_d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11003019 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003020 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04003021 struct dentry *new = __d_find_any_alias(inode);
3022 if (unlikely(new)) {
Eric W. Biedermana03e2832015-08-15 13:36:41 -05003023 /* The reference to new ensures it remains an alias */
3024 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04003025 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04003026 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04003027 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04003028 dput(new);
3029 new = ERR_PTR(-ELOOP);
3030 pr_warn_ratelimited(
3031 "VFS: Lookup of '%s' in %s %s"
3032 " would have caused loop\n",
3033 dentry->d_name.name,
3034 inode->i_sb->s_type->name,
3035 inode->i_sb->s_id);
3036 } else if (!IS_ROOT(new)) {
3037 int err = __d_unalias(inode, dentry, new);
3038 write_sequnlock(&rename_lock);
3039 if (err) {
3040 dput(new);
3041 new = ERR_PTR(err);
3042 }
Al Viro18367502011-07-12 21:42:24 -04003043 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04003044 __d_move(new, dentry, false);
3045 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003046 }
Al Virob5ae6b12014-10-12 22:16:02 -04003047 iput(inode);
3048 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003049 }
David Howells770bfad2006-08-22 20:06:07 -04003050 }
Al Virob5ae6b12014-10-12 22:16:02 -04003051out:
Al Viroed782b52016-03-09 19:52:39 -05003052 __d_add(dentry, inode);
Al Virob5ae6b12014-10-12 22:16:02 -04003053 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04003054}
Al Virob5ae6b12014-10-12 22:16:02 -04003055EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04003056
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003057static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01003058{
3059 *buflen -= namelen;
3060 if (*buflen < 0)
3061 return -ENAMETOOLONG;
3062 *buffer -= namelen;
3063 memcpy(*buffer, str, namelen);
3064 return 0;
3065}
3066
Waiman Long232d2d62013-09-09 12:18:13 -04003067/**
3068 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04003069 * @buffer: buffer pointer
3070 * @buflen: allocated length of the buffer
3071 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04003072 *
3073 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3074 * make sure that either the old or the new name pointer and length are
3075 * fetched. However, there may be mismatch between length and pointer.
3076 * The length cannot be trusted, we need to copy it byte-by-byte until
3077 * the length is reached or a null byte is found. It also prepends "/" at
3078 * the beginning of the name. The sequence number check at the caller will
3079 * retry it again when a d_move() does happen. So any garbage in the buffer
3080 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04003081 *
3082 * Data dependency barrier is needed to make sure that we see that terminating
3083 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04003084 */
Al Viro9aba36d2016-07-20 22:28:45 -04003085static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003086{
Waiman Long232d2d62013-09-09 12:18:13 -04003087 const char *dname = ACCESS_ONCE(name->name);
3088 u32 dlen = ACCESS_ONCE(name->len);
3089 char *p;
3090
Al Viro6d13f692014-09-29 14:46:30 -04003091 smp_read_barrier_depends();
3092
Waiman Long232d2d62013-09-09 12:18:13 -04003093 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04003094 if (*buflen < 0)
3095 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04003096 p = *buffer -= dlen + 1;
3097 *p++ = '/';
3098 while (dlen--) {
3099 char c = *dname++;
3100 if (!c)
3101 break;
3102 *p++ = c;
3103 }
3104 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003105}
3106
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107/**
Randy Dunlap208898c2010-11-18 15:02:49 -08003108 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003109 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003110 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003111 * @buffer: pointer to the end of the buffer
3112 * @buflen: pointer to buffer length
3113 *
Waiman Long18129972013-09-12 10:55:35 -04003114 * The function will first try to write out the pathname without taking any
3115 * lock other than the RCU read lock to make sure that dentries won't go away.
3116 * It only checks the sequence number of the global rename_lock as any change
3117 * in the dentry's d_seq will be preceded by changes in the rename_lock
3118 * sequence number. If the sequence number had been changed, it will restart
3119 * the whole pathname back-tracing sequence again by taking the rename_lock.
3120 * In this case, there is no need to take the RCU read lock as the recursive
3121 * parent pointer references will keep the dentry chain alive as long as no
3122 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003123 */
Al Viro02125a82011-12-05 08:43:34 -05003124static int prepend_path(const struct path *path,
3125 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003126 char **buffer, int *buflen)
3127{
Al Viroede4ceb2013-11-13 07:45:40 -05003128 struct dentry *dentry;
3129 struct vfsmount *vfsmnt;
3130 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003131 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04003132 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04003133 char *bptr;
3134 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003135
Al Viro48f5ec22013-09-09 15:22:25 -04003136 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04003137restart_mnt:
3138 read_seqbegin_or_lock(&mount_lock, &m_seq);
3139 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003140 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003141restart:
3142 bptr = *buffer;
3143 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04003144 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05003145 dentry = path->dentry;
3146 vfsmnt = path->mnt;
3147 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04003148 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003149 while (dentry != root->dentry || vfsmnt != root->mnt) {
3150 struct dentry * parent;
3151
3152 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04003153 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Eric W. Biedermancde93be2015-08-15 13:36:12 -05003154 /* Escaped? */
3155 if (dentry != vfsmnt->mnt_root) {
3156 bptr = *buffer;
3157 blen = *buflen;
3158 error = 3;
3159 break;
3160 }
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003161 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04003162 if (mnt != parent) {
3163 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
3164 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04003165 vfsmnt = &mnt->mnt;
3166 continue;
3167 }
Waiman Long232d2d62013-09-09 12:18:13 -04003168 if (!error)
3169 error = is_mounted(vfsmnt) ? 1 : 2;
3170 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003171 }
3172 parent = dentry->d_parent;
3173 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003174 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003175 if (error)
3176 break;
3177
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003178 dentry = parent;
3179 }
Al Viro48f5ec22013-09-09 15:22:25 -04003180 if (!(seq & 1))
3181 rcu_read_unlock();
3182 if (need_seqretry(&rename_lock, seq)) {
3183 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003184 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003185 }
Al Viro48f5ec22013-09-09 15:22:25 -04003186 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08003187
3188 if (!(m_seq & 1))
3189 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04003190 if (need_seqretry(&mount_lock, m_seq)) {
3191 m_seq = 1;
3192 goto restart_mnt;
3193 }
3194 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003195
Waiman Long232d2d62013-09-09 12:18:13 -04003196 if (error >= 0 && bptr == *buffer) {
3197 if (--blen < 0)
3198 error = -ENAMETOOLONG;
3199 else
3200 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003201 }
Waiman Long232d2d62013-09-09 12:18:13 -04003202 *buffer = bptr;
3203 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04003204 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003205}
3206
3207/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003208 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003209 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003210 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07003211 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212 * @buflen: buffer length
3213 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003214 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003216 * Returns a pointer into the buffer or an error code if the
3217 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08003218 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04003219 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003220 *
Al Viro02125a82011-12-05 08:43:34 -05003221 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222 */
Al Viro02125a82011-12-05 08:43:34 -05003223char *__d_path(const struct path *path,
3224 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003225 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003227 char *res = buf + buflen;
3228 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003230 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003231 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04003232
Al Viro02125a82011-12-05 08:43:34 -05003233 if (error < 0)
3234 return ERR_PTR(error);
3235 if (error > 0)
3236 return NULL;
3237 return res;
3238}
3239
3240char *d_absolute_path(const struct path *path,
3241 char *buf, int buflen)
3242{
3243 struct path root = {};
3244 char *res = buf + buflen;
3245 int error;
3246
3247 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003248 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05003249
3250 if (error > 1)
3251 error = -EINVAL;
3252 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003253 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003254 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255}
Guenter Roeck671662b2016-03-24 10:32:35 -07003256EXPORT_SYMBOL(d_absolute_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003258/*
3259 * same as __d_path but appends "(deleted)" for unlinked files.
3260 */
Al Viro02125a82011-12-05 08:43:34 -05003261static int path_with_deleted(const struct path *path,
3262 const struct path *root,
3263 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003264{
3265 prepend(buf, buflen, "\0", 1);
3266 if (d_unlinked(path->dentry)) {
3267 int error = prepend(buf, buflen, " (deleted)", 10);
3268 if (error)
3269 return error;
3270 }
3271
3272 return prepend_path(path, root, buf, buflen);
3273}
3274
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003275static int prepend_unreachable(char **buffer, int *buflen)
3276{
3277 return prepend(buffer, buflen, "(unreachable)", 13);
3278}
3279
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003280static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3281{
3282 unsigned seq;
3283
3284 do {
3285 seq = read_seqcount_begin(&fs->seq);
3286 *root = fs->root;
3287 } while (read_seqcount_retry(&fs->seq, seq));
3288}
3289
Jan Bluncka03a8a702008-02-14 19:38:32 -08003290/**
3291 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003292 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003293 * @buf: buffer to return value in
3294 * @buflen: buffer length
3295 *
3296 * Convert a dentry into an ASCII path name. If the entry has been deleted
3297 * the string " (deleted)" is appended. Note that this is ambiguous.
3298 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003299 * Returns a pointer into the buffer or an error code if the path was
3300 * too long. Note: Callers should use the returned pointer, not the passed
3301 * in buffer, to use the name! The implementation often starts at an offset
3302 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003303 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003304 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003305 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003306char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003308 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003309 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003310 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003312 /*
3313 * We have various synthetic filesystems that never get mounted. On
3314 * these filesystems dentries are never used for lookup purposes, and
3315 * thus don't need to be hashed. They also don't need a name until a
3316 * user wants to identify the object in /proc/pid/fd/. The little hack
3317 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003318 *
3319 * Some pseudo inodes are mountable. When they are mounted
3320 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3321 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003322 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003323 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3324 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003325 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003326
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003327 rcu_read_lock();
3328 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003329 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003330 rcu_read_unlock();
3331
Al Viro02125a82011-12-05 08:43:34 -05003332 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003333 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334 return res;
3335}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003336EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337
3338/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003339 * Helper function for dentry_operations.d_dname() members
3340 */
3341char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3342 const char *fmt, ...)
3343{
3344 va_list args;
3345 char temp[64];
3346 int sz;
3347
3348 va_start(args, fmt);
3349 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3350 va_end(args);
3351
3352 if (sz > sizeof(temp) || sz > buflen)
3353 return ERR_PTR(-ENAMETOOLONG);
3354
3355 buffer += buflen - sz;
3356 return memcpy(buffer, temp, sz);
3357}
3358
Al Viro118b2302013-08-24 12:08:17 -04003359char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3360{
3361 char *end = buffer + buflen;
3362 /* these dentries are never renamed, so d_lock is not needed */
3363 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003364 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003365 prepend(&end, &buflen, "/", 1))
3366 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003367 return end;
Al Viro118b2302013-08-24 12:08:17 -04003368}
David Herrmann31bbe162014-01-03 14:09:47 +01003369EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003370
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003371/*
Ram Pai6092d042008-03-27 13:06:20 +01003372 * Write full pathname from the root of the filesystem into the buffer.
3373 */
Al Virof6500802014-01-26 12:37:55 -05003374static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003375{
Al Virof6500802014-01-26 12:37:55 -05003376 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003377 char *end, *retval;
3378 int len, seq = 0;
3379 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003380
Al Virof6500802014-01-26 12:37:55 -05003381 if (buflen < 2)
3382 goto Elong;
3383
Al Viro48f5ec22013-09-09 15:22:25 -04003384 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003385restart:
Al Virof6500802014-01-26 12:37:55 -05003386 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003387 end = buf + buflen;
3388 len = buflen;
3389 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003390 /* Get '/' right */
3391 retval = end-1;
3392 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003393 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003394 while (!IS_ROOT(dentry)) {
3395 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003396
Ram Pai6092d042008-03-27 13:06:20 +01003397 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003398 error = prepend_name(&end, &len, &dentry->d_name);
3399 if (error)
3400 break;
Ram Pai6092d042008-03-27 13:06:20 +01003401
3402 retval = end;
3403 dentry = parent;
3404 }
Al Viro48f5ec22013-09-09 15:22:25 -04003405 if (!(seq & 1))
3406 rcu_read_unlock();
3407 if (need_seqretry(&rename_lock, seq)) {
3408 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003409 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003410 }
3411 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003412 if (error)
3413 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003414 return retval;
3415Elong:
3416 return ERR_PTR(-ENAMETOOLONG);
3417}
Nick Pigginec2447c2011-01-07 17:49:29 +11003418
3419char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3420{
Waiman Long232d2d62013-09-09 12:18:13 -04003421 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003422}
3423EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003424
3425char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3426{
3427 char *p = NULL;
3428 char *retval;
3429
Al Viroc1031352010-06-06 22:31:14 -04003430 if (d_unlinked(dentry)) {
3431 p = buf + buflen;
3432 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3433 goto Elong;
3434 buflen++;
3435 }
3436 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003437 if (!IS_ERR(retval) && p)
3438 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003439 return retval;
3440Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003441 return ERR_PTR(-ENAMETOOLONG);
3442}
3443
Linus Torvalds8b19e342013-09-12 10:35:47 -07003444static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3445 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003446{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003447 unsigned seq;
3448
3449 do {
3450 seq = read_seqcount_begin(&fs->seq);
3451 *root = fs->root;
3452 *pwd = fs->pwd;
3453 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003454}
3455
Ram Pai6092d042008-03-27 13:06:20 +01003456/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 * NOTE! The user-level library version returns a
3458 * character pointer. The kernel system call just
3459 * returns the length of the buffer filled (which
3460 * includes the ending '\0' character), or a negative
3461 * error value. So libc would do something like
3462 *
3463 * char *getcwd(char * buf, size_t size)
3464 * {
3465 * int retval;
3466 *
3467 * retval = sys_getcwd(buf, size);
3468 * if (retval >= 0)
3469 * return buf;
3470 * errno = -retval;
3471 * return NULL;
3472 * }
3473 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003474SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475{
Linus Torvalds552ce542007-02-13 12:08:18 -08003476 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003477 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003478 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003479
3480 if (!page)
3481 return -ENOMEM;
3482
Linus Torvalds8b19e342013-09-12 10:35:47 -07003483 rcu_read_lock();
3484 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485
Linus Torvalds552ce542007-02-13 12:08:18 -08003486 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003487 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003488 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003489 char *cwd = page + PATH_MAX;
3490 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003491
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003492 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003493 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003494 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003495
Al Viro02125a82011-12-05 08:43:34 -05003496 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003497 goto out;
3498
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003499 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003500 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003501 error = prepend_unreachable(&cwd, &buflen);
3502 if (error)
3503 goto out;
3504 }
3505
Linus Torvalds552ce542007-02-13 12:08:18 -08003506 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003507 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003508 if (len <= size) {
3509 error = len;
3510 if (copy_to_user(buf, cwd, len))
3511 error = -EFAULT;
3512 }
Nick Piggin949854d2011-01-07 17:49:37 +11003513 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003514 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003515 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516
3517out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003518 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003519 return error;
3520}
3521
3522/*
3523 * Test whether new_dentry is a subdirectory of old_dentry.
3524 *
3525 * Trivially implemented using the dcache structure
3526 */
3527
3528/**
3529 * is_subdir - is new dentry a subdirectory of old_dentry
3530 * @new_dentry: new dentry
3531 * @old_dentry: old dentry
3532 *
Yaowei Baia6e57872015-11-17 14:40:11 +08003533 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3534 * Returns false otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3536 */
3537
Yaowei Baia6e57872015-11-17 14:40:11 +08003538bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003539{
Yaowei Baia6e57872015-11-17 14:40:11 +08003540 bool result;
Nick Piggin949854d2011-01-07 17:49:37 +11003541 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003543 if (new_dentry == old_dentry)
Yaowei Baia6e57872015-11-17 14:40:11 +08003544 return true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003545
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003546 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003548 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003549 /*
3550 * Need rcu_readlock to protect against the d_parent trashing
3551 * due to d_move
3552 */
3553 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003554 if (d_ancestor(old_dentry, new_dentry))
Yaowei Baia6e57872015-11-17 14:40:11 +08003555 result = true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003556 else
Yaowei Baia6e57872015-11-17 14:40:11 +08003557 result = false;
Nick Piggin949854d2011-01-07 17:49:37 +11003558 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003559 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560
3561 return result;
3562}
3563
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003564static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003566 struct dentry *root = data;
3567 if (dentry != root) {
3568 if (d_unhashed(dentry) || !dentry->d_inode)
3569 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003571 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3572 dentry->d_flags |= DCACHE_GENOCIDE;
3573 dentry->d_lockref.count--;
3574 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003575 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003576 return D_WALK_CONTINUE;
3577}
Nick Piggin58db63d2011-01-07 17:49:39 +11003578
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003579void d_genocide(struct dentry *parent)
3580{
3581 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003582}
3583
Al Viro60545d02013-06-07 01:20:27 -04003584void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003585{
Al Viro60545d02013-06-07 01:20:27 -04003586 inode_dec_link_count(inode);
3587 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003588 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003589 !d_unlinked(dentry));
3590 spin_lock(&dentry->d_parent->d_lock);
3591 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3592 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3593 (unsigned long long)inode->i_ino);
3594 spin_unlock(&dentry->d_lock);
3595 spin_unlock(&dentry->d_parent->d_lock);
3596 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597}
Al Viro60545d02013-06-07 01:20:27 -04003598EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003599
3600static __initdata unsigned long dhash_entries;
3601static int __init set_dhash_entries(char *str)
3602{
3603 if (!str)
3604 return 0;
3605 dhash_entries = simple_strtoul(str, &str, 0);
3606 return 1;
3607}
3608__setup("dhash_entries=", set_dhash_entries);
3609
3610static void __init dcache_init_early(void)
3611{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003612 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003613
3614 /* If hashes are distributed across NUMA nodes, defer
3615 * hash allocation until vmalloc space is available.
3616 */
3617 if (hashdist)
3618 return;
3619
3620 dentry_hashtable =
3621 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003622 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623 dhash_entries,
3624 13,
3625 HASH_EARLY,
3626 &d_hash_shift,
3627 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003628 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 0);
3630
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003631 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003632 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633}
3634
Denis Cheng74bf17c2007-10-16 23:26:30 -07003635static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003637 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638
3639 /*
3640 * A constructor could be added for stable state like the lists,
3641 * but it is probably not worth it because of the cache nature
3642 * of the dcache.
3643 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003644 dentry_cache = KMEM_CACHE(dentry,
Vladimir Davydov5d097052016-01-14 15:18:21 -08003645 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646
3647 /* Hash may have been set up in dcache_init_early */
3648 if (!hashdist)
3649 return;
3650
3651 dentry_hashtable =
3652 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003653 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003654 dhash_entries,
3655 13,
3656 0,
3657 &d_hash_shift,
3658 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003659 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660 0);
3661
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003662 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003663 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664}
3665
3666/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003667struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003668EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003669
Linus Torvalds1da177e2005-04-16 15:20:36 -07003670EXPORT_SYMBOL(d_genocide);
3671
Linus Torvalds1da177e2005-04-16 15:20:36 -07003672void __init vfs_caches_init_early(void)
3673{
3674 dcache_init_early();
3675 inode_init_early();
3676}
3677
Mel Gorman4248b0d2015-08-06 15:46:20 -07003678void __init vfs_caches_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003681 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003682
Denis Cheng74bf17c2007-10-16 23:26:30 -07003683 dcache_init();
3684 inode_init();
Mel Gorman4248b0d2015-08-06 15:46:20 -07003685 files_init();
3686 files_maxfiles_init();
Denis Cheng74bf17c2007-10-16 23:26:30 -07003687 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003688 bdev_cache_init();
3689 chrdev_init();
3690}