blob: e7484f9c73b4de22602f4e5b41d56cd966d4172f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/dcache.c
3 *
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
8
9/*
10 * Notes on the allocation strategy:
11 *
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040021#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/hash.h>
25#include <linux/cache.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/file.h>
29#include <asm/uaccess.h>
30#include <linux/security.h>
31#include <linux/seqlock.h>
32#include <linux/swap.h>
33#include <linux/bootmem.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Frederic Weisbecker613afbf2009-07-16 15:44:29 +020035#include <linux/hardirq.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110036#include <linux/bit_spinlock.h>
37#include <linux/rculist_bl.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070038#include <linux/prefetch.h>
David Howellsdd179942011-08-16 15:31:30 +010039#include <linux/ratelimit.h>
Dave Chinnerf6041562013-08-28 10:18:00 +100040#include <linux/list_lru.h>
David Howells07f3f052006-09-30 20:52:18 +020041#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050042#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Nick Piggin789680d2011-01-07 17:49:30 +110044/*
45 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110046 * dcache->d_inode->i_lock protects:
47 * - i_dentry, d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110048 * dcache_hash_bucket lock protects:
49 * - the dcache hash table
50 * s_anon bl list spinlock protects:
51 * - the s_anon list (see __d_drop)
Dave Chinner19156842013-08-28 10:17:55 +100052 * dentry->d_sb->s_dentry_lru_lock protects:
Nick Piggin23044502011-01-07 17:49:31 +110053 * - the dcache lru lists and counters
54 * d_lock protects:
55 * - d_flags
56 * - d_name
57 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110058 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110059 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110060 * - d_parent and d_subdirs
61 * - childrens' d_child and d_parent
Nick Pigginb23fb0a2011-01-07 17:49:35 +110062 * - d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110063 *
64 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110065 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110066 * dentry->d_lock
Dave Chinner19156842013-08-28 10:17:55 +100067 * dentry->d_sb->s_dentry_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110068 * dcache_hash_bucket lock
69 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110070 *
Nick Pigginda502952011-01-07 17:49:33 +110071 * If there is an ancestor relationship:
72 * dentry->d_parent->...->d_parent->d_lock
73 * ...
74 * dentry->d_parent->d_lock
75 * dentry->d_lock
76 *
77 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110078 * if (dentry1 < dentry2)
79 * dentry1->d_lock
80 * dentry2->d_lock
81 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080082int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
84
Al Viro74c3cbe2007-07-22 08:04:18 -040085__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Nick Piggin949854d2011-01-07 17:49:37 +110087EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Christoph Lametere18b8902006-12-06 20:33:20 -080089static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/*
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
95 *
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
98 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800100static unsigned int d_hash_mask __read_mostly;
101static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100102
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700103static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvalds8966be92012-03-02 14:23:30 -0800105static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700106 unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100107{
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700108 hash += (unsigned long) parent / L1_CACHE_BYTES;
Linus Torvalds99d263d2014-09-13 11:30:10 -0700109 return dentry_hashtable + hash_32(hash, d_hash_shift);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100110}
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112/* Statistics gathering. */
113struct dentry_stat_t dentry_stat = {
114 .age_limit = 45,
115};
116
Glauber Costa3942c072013-08-28 10:17:53 +1000117static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000118static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400119
120#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000121
122/*
123 * Here we resort to our own counters instead of using generic per-cpu counters
124 * for consistency with what the vfs inode code does. We are expected to harvest
125 * better code and performance by having our own specialized counters.
126 *
127 * Please note that the loop is done over all possible CPUs, not over all online
128 * CPUs. The reason for this is that we don't want to play games with CPUs going
129 * on and off. If one of them goes off, we will just keep their counters.
130 *
131 * glommer: See cffbc8a for details, and if you ever intend to change this,
132 * please update all vfs counters to match.
133 */
Glauber Costa3942c072013-08-28 10:17:53 +1000134static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100135{
136 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000137 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100138 for_each_possible_cpu(i)
139 sum += per_cpu(nr_dentry, i);
140 return sum < 0 ? 0 : sum;
141}
142
Dave Chinner62d36c72013-08-28 10:17:54 +1000143static long get_nr_dentry_unused(void)
144{
145 int i;
146 long sum = 0;
147 for_each_possible_cpu(i)
148 sum += per_cpu(nr_dentry_unused, i);
149 return sum < 0 ? 0 : sum;
150}
151
Joe Perches1f7e0612014-06-06 14:38:05 -0700152int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400153 size_t *lenp, loff_t *ppos)
154{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100155 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000156 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000157 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400158}
159#endif
160
Linus Torvalds5483f182012-03-04 15:51:42 -0800161/*
162 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
163 * The strings are both count bytes long, and count is non-zero.
164 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700165#ifdef CONFIG_DCACHE_WORD_ACCESS
166
167#include <asm/word-at-a-time.h>
168/*
169 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
170 * aligned allocation for this particular component. We don't
171 * strictly need the load_unaligned_zeropad() safety, but it
172 * doesn't hurt either.
173 *
174 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
175 * need the careful unaligned handling.
176 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700177static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800178{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800179 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800180
181 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700182 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700183 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800184 if (tcount < sizeof(unsigned long))
185 break;
186 if (unlikely(a != b))
187 return 1;
188 cs += sizeof(unsigned long);
189 ct += sizeof(unsigned long);
190 tcount -= sizeof(unsigned long);
191 if (!tcount)
192 return 0;
193 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000194 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800195 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700196}
197
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800198#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700199
Linus Torvalds94753db52012-05-10 12:19:19 -0700200static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700201{
Linus Torvalds5483f182012-03-04 15:51:42 -0800202 do {
203 if (*cs != *ct)
204 return 1;
205 cs++;
206 ct++;
207 tcount--;
208 } while (tcount);
209 return 0;
210}
211
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700212#endif
213
Linus Torvalds94753db52012-05-10 12:19:19 -0700214static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
215{
Linus Torvalds6326c712012-05-21 16:14:04 -0700216 const unsigned char *cs;
Linus Torvalds94753db52012-05-10 12:19:19 -0700217 /*
218 * Be careful about RCU walk racing with rename:
219 * use ACCESS_ONCE to fetch the name pointer.
220 *
221 * NOTE! Even if a rename will mean that the length
222 * was not loaded atomically, we don't care. The
223 * RCU walk will check the sequence count eventually,
224 * and catch it. And we won't overrun the buffer,
225 * because we're reading the name pointer atomically,
226 * and a dentry name is guaranteed to be properly
227 * terminated with a NUL byte.
228 *
229 * End result: even if 'len' is wrong, we'll exit
230 * early because the data cannot match (there can
231 * be no NUL in the ct/tcount data)
232 */
Linus Torvalds6326c712012-05-21 16:14:04 -0700233 cs = ACCESS_ONCE(dentry->d_name.name);
234 smp_read_barrier_depends();
235 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700236}
237
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400238static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400240 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
241
Al Virob3d9b7a2012-06-09 13:51:19 -0400242 WARN_ON(!hlist_unhashed(&dentry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 if (dname_external(dentry))
244 kfree(dentry->d_name.name);
245 kmem_cache_free(dentry_cache, dentry);
246}
247
Al Virob4f03542014-04-29 23:40:14 -0400248static void dentry_free(struct dentry *dentry)
249{
250 /* if dentry was never visible to RCU, immediate free is OK */
251 if (!(dentry->d_flags & DCACHE_RCUACCESS))
252 __d_free(&dentry->d_u.d_rcu);
253 else
254 call_rcu(&dentry->d_u.d_rcu, __d_free);
255}
256
Nick Piggin31e6b012011-01-07 17:49:52 +1100257/**
258 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800259 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100260 * After this call, in-progress rcu-walk path lookup will fail. This
261 * should be called after unhashing, and after changing d_inode (if
262 * the dentry has not already been unhashed).
263 */
264static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
265{
266 assert_spin_locked(&dentry->d_lock);
267 /* Go through a barrier */
268 write_seqcount_barrier(&dentry->d_seq);
269}
270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271/*
272 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100273 * d_iput() operation if defined. Dentry has no refcount
274 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800276static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200277 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100278 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279{
280 struct inode *inode = dentry->d_inode;
281 if (inode) {
282 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400283 hlist_del_init(&dentry->d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100285 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700286 if (!inode->i_nlink)
287 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 if (dentry->d_op && dentry->d_op->d_iput)
289 dentry->d_op->d_iput(dentry, inode);
290 else
291 iput(inode);
292 } else {
293 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 }
295}
296
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700297/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100298 * Release the dentry's inode, using the filesystem
299 * d_iput() operation if defined. dentry remains in-use.
300 */
301static void dentry_unlink_inode(struct dentry * dentry)
302 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100303 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100304{
305 struct inode *inode = dentry->d_inode;
David Howellsb18825a2013-09-12 19:22:53 +0100306 __d_clear_type(dentry);
Nick Piggin31e6b012011-01-07 17:49:52 +1100307 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400308 hlist_del_init(&dentry->d_alias);
Nick Piggin31e6b012011-01-07 17:49:52 +1100309 dentry_rcuwalk_barrier(dentry);
310 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100311 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100312 if (!inode->i_nlink)
313 fsnotify_inoderemove(inode);
314 if (dentry->d_op && dentry->d_op->d_iput)
315 dentry->d_op->d_iput(dentry, inode);
316 else
317 iput(inode);
318}
319
320/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400321 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
322 * is in use - which includes both the "real" per-superblock
323 * LRU list _and_ the DCACHE_SHRINK_LIST use.
324 *
325 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
326 * on the shrink list (ie not on the superblock LRU list).
327 *
328 * The per-cpu "nr_dentry_unused" counters are updated with
329 * the DCACHE_LRU_LIST bit.
330 *
331 * These helper functions make sure we always follow the
332 * rules. d_lock must be held by the caller.
333 */
334#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
335static void d_lru_add(struct dentry *dentry)
336{
337 D_FLAG_VERIFY(dentry, 0);
338 dentry->d_flags |= DCACHE_LRU_LIST;
339 this_cpu_inc(nr_dentry_unused);
340 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
341}
342
343static void d_lru_del(struct dentry *dentry)
344{
345 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
346 dentry->d_flags &= ~DCACHE_LRU_LIST;
347 this_cpu_dec(nr_dentry_unused);
348 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
349}
350
351static void d_shrink_del(struct dentry *dentry)
352{
353 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
354 list_del_init(&dentry->d_lru);
355 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
356 this_cpu_dec(nr_dentry_unused);
357}
358
359static void d_shrink_add(struct dentry *dentry, struct list_head *list)
360{
361 D_FLAG_VERIFY(dentry, 0);
362 list_add(&dentry->d_lru, list);
363 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
364 this_cpu_inc(nr_dentry_unused);
365}
366
367/*
368 * These can only be called under the global LRU lock, ie during the
369 * callback for freeing the LRU list. "isolate" removes it from the
370 * LRU lists entirely, while shrink_move moves it to the indicated
371 * private list.
372 */
373static void d_lru_isolate(struct dentry *dentry)
374{
375 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
376 dentry->d_flags &= ~DCACHE_LRU_LIST;
377 this_cpu_dec(nr_dentry_unused);
378 list_del_init(&dentry->d_lru);
379}
380
381static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
382{
383 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
384 dentry->d_flags |= DCACHE_SHRINK_LIST;
385 list_move_tail(&dentry->d_lru, list);
386}
387
388/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000389 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700390 */
391static void dentry_lru_add(struct dentry *dentry)
392{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400393 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
394 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700395}
396
Miklos Szeredid52b9082007-05-08 00:23:46 -0700397/**
Nick Piggin789680d2011-01-07 17:49:30 +1100398 * d_drop - drop a dentry
399 * @dentry: dentry to drop
400 *
401 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
402 * be found through a VFS lookup any more. Note that this is different from
403 * deleting the dentry - d_delete will try to mark the dentry negative if
404 * possible, giving a successful _negative_ lookup, while d_drop will
405 * just make the cache lookup fail.
406 *
407 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
408 * reason (NFS timeouts or autofs deletes).
409 *
410 * __d_drop requires dentry->d_lock.
411 */
412void __d_drop(struct dentry *dentry)
413{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700414 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400415 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400416 /*
417 * Hashed dentries are normally on the dentry hashtable,
418 * with the exception of those newly allocated by
419 * d_obtain_alias, which are always IS_ROOT:
420 */
421 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400422 b = &dentry->d_sb->s_anon;
423 else
424 b = d_hash(dentry->d_parent, dentry->d_name.hash);
425
426 hlist_bl_lock(b);
427 __hlist_bl_del(&dentry->d_hash);
428 dentry->d_hash.pprev = NULL;
429 hlist_bl_unlock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700430 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100431 }
432}
433EXPORT_SYMBOL(__d_drop);
434
435void d_drop(struct dentry *dentry)
436{
Nick Piggin789680d2011-01-07 17:49:30 +1100437 spin_lock(&dentry->d_lock);
438 __d_drop(dentry);
439 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100440}
441EXPORT_SYMBOL(d_drop);
442
Al Viroe55fd012014-05-28 13:51:12 -0400443static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100444{
Al Viro41edf272014-05-01 10:30:00 -0400445 struct dentry *parent = NULL;
446 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400447 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100448 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100449
Linus Torvalds0d984392013-09-08 13:46:52 -0700450 /*
451 * The dentry is now unrecoverably dead to the world.
452 */
453 lockref_mark_dead(&dentry->d_lockref);
454
Sage Weilf0023bc2011-10-28 10:02:42 -0700455 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700456 * inform the fs via d_prune that this dentry is about to be
457 * unhashed and destroyed.
458 */
Yan, Zheng590fb512013-08-13 15:42:02 +0800459 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
Yan, Zheng61572bb2013-04-15 14:13:21 +0800460 dentry->d_op->d_prune(dentry);
461
Al Viro01b60352014-04-29 23:42:52 -0400462 if (dentry->d_flags & DCACHE_LRU_LIST) {
463 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
464 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400465 }
Nick Piggin77812a12011-01-07 17:49:48 +1100466 /* if it was on the hash then remove it */
467 __d_drop(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400468 list_del(&dentry->d_u.d_child);
469 /*
470 * Inform d_walk() that we are no longer attached to the
471 * dentry tree
472 */
473 dentry->d_flags |= DCACHE_DENTRY_KILLED;
474 if (parent)
475 spin_unlock(&parent->d_lock);
476 dentry_iput(dentry);
477 /*
478 * dentry_iput drops the locks, at which point nobody (except
479 * transient RCU lookups) can reach this dentry.
480 */
481 BUG_ON((int)dentry->d_lockref.count > 0);
482 this_cpu_dec(nr_dentry);
483 if (dentry->d_op && dentry->d_op->d_release)
484 dentry->d_op->d_release(dentry);
485
Al Viro41edf272014-05-01 10:30:00 -0400486 spin_lock(&dentry->d_lock);
487 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
488 dentry->d_flags |= DCACHE_MAY_FREE;
489 can_free = false;
490 }
491 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400492 if (likely(can_free))
493 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400494}
495
496/*
497 * Finish off a dentry we've decided to kill.
498 * dentry->d_lock must be held, returns with it unlocked.
499 * If ref is non-zero, then decrement the refcount too.
500 * Returns dentry requiring refcount drop, or NULL if we're done.
501 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400502static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400503 __releases(dentry->d_lock)
504{
505 struct inode *inode = dentry->d_inode;
506 struct dentry *parent = NULL;
507
508 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
509 goto failed;
510
511 if (!IS_ROOT(dentry)) {
512 parent = dentry->d_parent;
513 if (unlikely(!spin_trylock(&parent->d_lock))) {
514 if (inode)
515 spin_unlock(&inode->i_lock);
516 goto failed;
517 }
518 }
519
520 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400521 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400522
523failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400524 spin_unlock(&dentry->d_lock);
525 cpu_relax();
Al Viroe55fd012014-05-28 13:51:12 -0400526 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100527}
528
Al Viro046b9612014-05-29 08:54:52 -0400529static inline struct dentry *lock_parent(struct dentry *dentry)
530{
531 struct dentry *parent = dentry->d_parent;
532 if (IS_ROOT(dentry))
533 return NULL;
Al Viroc2338f22014-06-12 00:29:13 -0400534 if (unlikely((int)dentry->d_lockref.count < 0))
535 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400536 if (likely(spin_trylock(&parent->d_lock)))
537 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400538 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400539 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400540again:
541 parent = ACCESS_ONCE(dentry->d_parent);
542 spin_lock(&parent->d_lock);
543 /*
544 * We can't blindly lock dentry until we are sure
545 * that we won't violate the locking order.
546 * Any changes of dentry->d_parent must have
547 * been done with parent->d_lock held, so
548 * spin_lock() above is enough of a barrier
549 * for checking if it's still our child.
550 */
551 if (unlikely(parent != dentry->d_parent)) {
552 spin_unlock(&parent->d_lock);
553 goto again;
554 }
555 rcu_read_unlock();
556 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700557 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400558 else
559 parent = NULL;
560 return parent;
561}
562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563/*
564 * This is dput
565 *
566 * This is complicated by the fact that we do not want to put
567 * dentries that are no longer on any hash chain on the unused
568 * list: we'd much rather just get rid of them immediately.
569 *
570 * However, that implies that we have to traverse the dentry
571 * tree upwards to the parents which might _also_ now be
572 * scheduled for deletion (it may have been only waiting for
573 * its last child to go away).
574 *
575 * This tail recursion is done by hand as we don't want to depend
576 * on the compiler to always get this right (gcc generally doesn't).
577 * Real recursion would eat up our stack space.
578 */
579
580/*
581 * dput - release a dentry
582 * @dentry: dentry to release
583 *
584 * Release a dentry. This will drop the usage count and if appropriate
585 * call the dentry unlink method as well as removing it from the queues and
586 * releasing its resources. If the parent dentries were scheduled for release
587 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589void dput(struct dentry *dentry)
590{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700591 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 return;
593
594repeat:
Waiman Long98474232013-08-28 18:24:59 -0700595 if (lockref_put_or_lock(&dentry->d_lockref))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700598 /* Unreachable? Get rid of it */
599 if (unlikely(d_unhashed(dentry)))
600 goto kill_it;
601
602 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100604 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 }
Nick Piggin265ac902010-10-10 05:36:24 -0400606
Linus Torvalds358eec12013-10-31 15:43:02 -0700607 if (!(dentry->d_flags & DCACHE_REFERENCED))
608 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400609 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400610
Waiman Long98474232013-08-28 18:24:59 -0700611 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100612 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 return;
614
Miklos Szeredid52b9082007-05-08 00:23:46 -0700615kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400616 dentry = dentry_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700617 if (dentry)
618 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700620EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
622/**
623 * d_invalidate - invalidate a dentry
624 * @dentry: dentry to invalidate
625 *
626 * Try to invalidate the dentry if it turns out to be
627 * possible. If there are other dentries that can be
628 * reached through this one we can't delete it and we
629 * return -EBUSY. On success we return 0.
630 *
631 * no dcache lock.
632 */
633
634int d_invalidate(struct dentry * dentry)
635{
636 /*
637 * If it's already been dropped, return OK.
638 */
Nick Pigginda502952011-01-07 17:49:33 +1100639 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100641 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 return 0;
643 }
644 /*
645 * Check whether to do a partial shrink_dcache
646 * to get rid of unused child entries.
647 */
648 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100649 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100651 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 }
653
654 /*
655 * Somebody else still using it?
656 *
657 * If it's a directory, we can't drop it
658 * for fear of somebody re-populating it
659 * with children (even though dropping it
660 * would make it unreachable from the root,
661 * we might still populate it if it was a
662 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000663 * We also need to leave mountpoints alone,
664 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 */
Waiman Long98474232013-08-28 18:24:59 -0700666 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
Al Viro50e69632011-11-07 16:39:57 +0000667 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 return -EBUSY;
670 }
671 }
672
673 __d_drop(dentry);
674 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 return 0;
676}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700677EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100679/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100680static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
Waiman Long98474232013-08-28 18:24:59 -0700682 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683}
684
Nick Piggindc0474b2011-01-07 17:49:43 +1100685static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100686{
Waiman Long98474232013-08-28 18:24:59 -0700687 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100688}
689
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100690struct dentry *dget_parent(struct dentry *dentry)
691{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700692 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100693 struct dentry *ret;
694
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700695 /*
696 * Do optimistic parent lookup without any
697 * locking.
698 */
699 rcu_read_lock();
700 ret = ACCESS_ONCE(dentry->d_parent);
701 gotref = lockref_get_not_zero(&ret->d_lockref);
702 rcu_read_unlock();
703 if (likely(gotref)) {
704 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
705 return ret;
706 dput(ret);
707 }
708
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100709repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100710 /*
711 * Don't need rcu_dereference because we re-check it was correct under
712 * the lock.
713 */
714 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100715 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100716 spin_lock(&ret->d_lock);
717 if (unlikely(ret != dentry->d_parent)) {
718 spin_unlock(&ret->d_lock);
719 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100720 goto repeat;
721 }
Nick Piggina734eb42011-01-07 17:49:44 +1100722 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700723 BUG_ON(!ret->d_lockref.count);
724 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100725 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100726 return ret;
727}
728EXPORT_SYMBOL(dget_parent);
729
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730/**
731 * d_find_alias - grab a hashed alias of inode
732 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 *
734 * If inode has a hashed alias, or is a directory and has any alias,
735 * acquire the reference to alias and return it. Otherwise return NULL.
736 * Notice that if inode is a directory there can be only one alias and
737 * it can be unhashed only if it has no children, or if it is the root
738 * of a filesystem.
739 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700740 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500741 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500743static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
Nick Pigginda502952011-01-07 17:49:33 +1100745 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746
Nick Pigginda502952011-01-07 17:49:33 +1100747again:
748 discon_alias = NULL;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800749 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100750 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700752 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100753 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500755 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100756 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100757 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 return alias;
759 }
760 }
Nick Pigginda502952011-01-07 17:49:33 +1100761 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
Nick Pigginda502952011-01-07 17:49:33 +1100763 if (discon_alias) {
764 alias = discon_alias;
765 spin_lock(&alias->d_lock);
766 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500767 __dget_dlock(alias);
768 spin_unlock(&alias->d_lock);
769 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100770 }
771 spin_unlock(&alias->d_lock);
772 goto again;
773 }
774 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775}
776
Nick Pigginda502952011-01-07 17:49:33 +1100777struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778{
David Howells214fda12006-03-25 03:06:36 -0800779 struct dentry *de = NULL;
780
Al Virob3d9b7a2012-06-09 13:51:19 -0400781 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100782 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500783 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100784 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800785 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 return de;
787}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700788EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
790/*
791 * Try to kill dentries associated with this inode.
792 * WARNING: you must own a reference to inode.
793 */
794void d_prune_aliases(struct inode *inode)
795{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700796 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100798 spin_lock(&inode->i_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800799 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700801 if (!dentry->d_lockref.count) {
Yan, Zheng590fb512013-08-13 15:42:02 +0800802 /*
803 * inform the fs via d_prune that this dentry
804 * is about to be unhashed and destroyed.
805 */
806 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
807 !d_unhashed(dentry))
808 dentry->d_op->d_prune(dentry);
809
Nick Piggindc0474b2011-01-07 17:49:43 +1100810 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 __d_drop(dentry);
812 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100813 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 dput(dentry);
815 goto restart;
816 }
817 spin_unlock(&dentry->d_lock);
818 }
Nick Piggin873feea2011-01-07 17:50:06 +1100819 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700821EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400823static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824{
Al Viro5c47e6d2014-04-29 16:13:18 -0400825 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700826
Miklos Szeredi60942f22014-05-02 15:38:39 -0400827 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400828 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400829 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100830 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400831 parent = lock_parent(dentry);
832
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000834 * The dispose list is isolated and dentries are not accounted
835 * to the LRU here, so we can simply remove it from the list
836 * here regardless of whether it is referenced or not.
837 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400838 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000839
840 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000842 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 */
Al Viro41edf272014-05-01 10:30:00 -0400844 if ((int)dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700845 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400846 if (parent)
847 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 continue;
849 }
Nick Piggin77812a12011-01-07 17:49:48 +1100850
Al Viro64fd72e2014-05-28 09:48:44 -0400851
852 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
853 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
854 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400855 if (parent)
856 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -0400857 if (can_free)
858 dentry_free(dentry);
859 continue;
860 }
861
Al Viroff2fde92014-05-28 13:59:13 -0400862 inode = dentry->d_inode;
863 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400864 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000865 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400866 if (parent)
867 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400868 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000869 }
Al Viroff2fde92014-05-28 13:59:13 -0400870
Al Viroff2fde92014-05-28 13:59:13 -0400871 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -0400872
Al Viro5c47e6d2014-04-29 16:13:18 -0400873 /*
874 * We need to prune ancestors too. This is necessary to prevent
875 * quadratic behavior of shrink_dcache_parent(), but is also
876 * expected to be beneficial in reducing dentry cache
877 * fragmentation.
878 */
879 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -0400880 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
881 parent = lock_parent(dentry);
882 if (dentry->d_lockref.count != 1) {
883 dentry->d_lockref.count--;
884 spin_unlock(&dentry->d_lock);
885 if (parent)
886 spin_unlock(&parent->d_lock);
887 break;
888 }
889 inode = dentry->d_inode; /* can't be NULL */
890 if (unlikely(!spin_trylock(&inode->i_lock))) {
891 spin_unlock(&dentry->d_lock);
892 if (parent)
893 spin_unlock(&parent->d_lock);
894 cpu_relax();
895 continue;
896 }
897 __dentry_kill(dentry);
898 dentry = parent;
899 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400901}
902
Dave Chinnerf6041562013-08-28 10:18:00 +1000903static enum lru_status
904dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
905{
906 struct list_head *freeable = arg;
907 struct dentry *dentry = container_of(item, struct dentry, d_lru);
908
909
910 /*
911 * we are inverting the lru lock/dentry->d_lock here,
912 * so use a trylock. If we fail to get the lock, just skip
913 * it
914 */
915 if (!spin_trylock(&dentry->d_lock))
916 return LRU_SKIP;
917
918 /*
919 * Referenced dentries are still in use. If they have active
920 * counts, just remove them from the LRU. Otherwise give them
921 * another pass through the LRU.
922 */
923 if (dentry->d_lockref.count) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400924 d_lru_isolate(dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +1000925 spin_unlock(&dentry->d_lock);
926 return LRU_REMOVED;
927 }
928
929 if (dentry->d_flags & DCACHE_REFERENCED) {
930 dentry->d_flags &= ~DCACHE_REFERENCED;
931 spin_unlock(&dentry->d_lock);
932
933 /*
934 * The list move itself will be made by the common LRU code. At
935 * this point, we've dropped the dentry->d_lock but keep the
936 * lru lock. This is safe to do, since every list movement is
937 * protected by the lru lock even if both locks are held.
938 *
939 * This is guaranteed by the fact that all LRU management
940 * functions are intermediated by the LRU API calls like
941 * list_lru_add and list_lru_del. List movement in this file
942 * only ever occur through this functions or through callbacks
943 * like this one, that are called from the LRU API.
944 *
945 * The only exceptions to this are functions like
946 * shrink_dentry_list, and code that first checks for the
947 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
948 * operating only with stack provided lists after they are
949 * properly isolated from the main list. It is thus, always a
950 * local access.
951 */
952 return LRU_ROTATE;
953 }
954
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400955 d_lru_shrink_move(dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +1000956 spin_unlock(&dentry->d_lock);
957
958 return LRU_REMOVED;
959}
960
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400961/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000962 * prune_dcache_sb - shrink the dcache
963 * @sb: superblock
Dave Chinnerf6041562013-08-28 10:18:00 +1000964 * @nr_to_scan : number of entries to try to free
Dave Chinner9b17c622013-08-28 10:18:05 +1000965 * @nid: which node to scan for freeable entities
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400966 *
Dave Chinnerf6041562013-08-28 10:18:00 +1000967 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000968 * done when we need more memory an called from the superblock shrinker
969 * function.
970 *
971 * This function may fail to free any resources if all the dentries are in
972 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400973 */
Dave Chinner9b17c622013-08-28 10:18:05 +1000974long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
975 int nid)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400976{
Dave Chinnerf6041562013-08-28 10:18:00 +1000977 LIST_HEAD(dispose);
978 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400979
Dave Chinner9b17c622013-08-28 10:18:05 +1000980 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
981 &dispose, &nr_to_scan);
Dave Chinnerf6041562013-08-28 10:18:00 +1000982 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +1000983 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984}
985
Glauber Costa4e717f52013-08-28 10:18:03 +1000986static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
987 spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000988{
Glauber Costa4e717f52013-08-28 10:18:03 +1000989 struct list_head *freeable = arg;
990 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000991
Glauber Costa4e717f52013-08-28 10:18:03 +1000992 /*
993 * we are inverting the lru lock/dentry->d_lock here,
994 * so use a trylock. If we fail to get the lock, just skip
995 * it
996 */
997 if (!spin_trylock(&dentry->d_lock))
998 return LRU_SKIP;
999
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001000 d_lru_shrink_move(dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001001 spin_unlock(&dentry->d_lock);
1002
1003 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001004}
1005
Glauber Costa4e717f52013-08-28 10:18:03 +10001006
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001007/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 * shrink_dcache_sb - shrink dcache for a superblock
1009 * @sb: superblock
1010 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001011 * Shrink the dcache for the specified super block. This is used to free
1012 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001014void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015{
Glauber Costa4e717f52013-08-28 10:18:03 +10001016 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001017
Glauber Costa4e717f52013-08-28 10:18:03 +10001018 do {
1019 LIST_HEAD(dispose);
1020
1021 freed = list_lru_walk(&sb->s_dentry_lru,
1022 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1023
1024 this_cpu_sub(nr_dentry_unused, freed);
1025 shrink_dentry_list(&dispose);
1026 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001028EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001031 * enum d_walk_ret - action to talke during tree walk
1032 * @D_WALK_CONTINUE: contrinue walk
1033 * @D_WALK_QUIT: quit walk
1034 * @D_WALK_NORETRY: quit when retry is needed
1035 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001037enum d_walk_ret {
1038 D_WALK_CONTINUE,
1039 D_WALK_QUIT,
1040 D_WALK_NORETRY,
1041 D_WALK_SKIP,
1042};
1043
1044/**
1045 * d_walk - walk the dentry tree
1046 * @parent: start of walk
1047 * @data: data passed to @enter() and @finish()
1048 * @enter: callback when first entering the dentry
1049 * @finish: callback when successfully finished the walk
1050 *
1051 * The @enter() and @finish() callbacks are called with d_lock held.
1052 */
1053static void d_walk(struct dentry *parent, void *data,
1054 enum d_walk_ret (*enter)(void *, struct dentry *),
1055 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056{
Nick Piggin949854d2011-01-07 17:49:37 +11001057 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001059 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001060 enum d_walk_ret ret;
1061 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001062
Nick Piggin58db63d2011-01-07 17:49:39 +11001063again:
Al Viro48f5ec22013-09-09 15:22:25 -04001064 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001065 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001066 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001067
1068 ret = enter(data, this_parent);
1069 switch (ret) {
1070 case D_WALK_CONTINUE:
1071 break;
1072 case D_WALK_QUIT:
1073 case D_WALK_SKIP:
1074 goto out_unlock;
1075 case D_WALK_NORETRY:
1076 retry = false;
1077 break;
1078 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079repeat:
1080 next = this_parent->d_subdirs.next;
1081resume:
1082 while (next != &this_parent->d_subdirs) {
1083 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001084 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001086
1087 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001088
1089 ret = enter(data, dentry);
1090 switch (ret) {
1091 case D_WALK_CONTINUE:
1092 break;
1093 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001094 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001095 goto out_unlock;
1096 case D_WALK_NORETRY:
1097 retry = false;
1098 break;
1099 case D_WALK_SKIP:
1100 spin_unlock(&dentry->d_lock);
1101 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001102 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001103
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001105 spin_unlock(&this_parent->d_lock);
1106 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001108 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 goto repeat;
1110 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001111 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 }
1113 /*
1114 * All done at this level ... ascend and resume the search.
1115 */
1116 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001117 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001118 this_parent = child->d_parent;
1119
1120 rcu_read_lock();
1121 spin_unlock(&child->d_lock);
1122 spin_lock(&this_parent->d_lock);
1123
1124 /*
1125 * might go back up the wrong parent if we have had a rename
1126 * or deletion
1127 */
1128 if (this_parent != child->d_parent ||
1129 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1130 need_seqretry(&rename_lock, seq)) {
1131 spin_unlock(&this_parent->d_lock);
1132 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11001133 goto rename_retry;
Al Viro31dec132013-10-25 17:04:27 -04001134 }
1135 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11001136 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 goto resume;
1138 }
Al Viro48f5ec22013-09-09 15:22:25 -04001139 if (need_seqretry(&rename_lock, seq)) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001140 spin_unlock(&this_parent->d_lock);
1141 goto rename_retry;
1142 }
1143 if (finish)
1144 finish(data);
1145
1146out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001147 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001148 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001149 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001150
1151rename_retry:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001152 if (!retry)
1153 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001154 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001155 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001157
1158/*
1159 * Search for at least 1 mount point in the dentry's subdirs.
1160 * We descend to the next level whenever the d_subdirs
1161 * list is non-empty and continue searching.
1162 */
1163
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001164static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1165{
1166 int *ret = data;
1167 if (d_mountpoint(dentry)) {
1168 *ret = 1;
1169 return D_WALK_QUIT;
1170 }
1171 return D_WALK_CONTINUE;
1172}
1173
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001174/**
1175 * have_submounts - check for mounts over a dentry
1176 * @parent: dentry to check.
1177 *
1178 * Return true if the parent or its subdirectories contain
1179 * a mount point
1180 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001181int have_submounts(struct dentry *parent)
1182{
1183 int ret = 0;
1184
1185 d_walk(parent, &ret, check_mount, NULL);
1186
1187 return ret;
1188}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001189EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
1191/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001192 * Called by mount code to set a mountpoint and check if the mountpoint is
1193 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1194 * subtree can become unreachable).
1195 *
1196 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1197 * this reason take rename_lock and d_lock on dentry and ancestors.
1198 */
1199int d_set_mounted(struct dentry *dentry)
1200{
1201 struct dentry *p;
1202 int ret = -ENOENT;
1203 write_seqlock(&rename_lock);
1204 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1205 /* Need exclusion wrt. check_submounts_and_drop() */
1206 spin_lock(&p->d_lock);
1207 if (unlikely(d_unhashed(p))) {
1208 spin_unlock(&p->d_lock);
1209 goto out;
1210 }
1211 spin_unlock(&p->d_lock);
1212 }
1213 spin_lock(&dentry->d_lock);
1214 if (!d_unlinked(dentry)) {
1215 dentry->d_flags |= DCACHE_MOUNTED;
1216 ret = 0;
1217 }
1218 spin_unlock(&dentry->d_lock);
1219out:
1220 write_sequnlock(&rename_lock);
1221 return ret;
1222}
1223
1224/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001225 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 * and move any unused dentries to the end of the unused
1227 * list for prune_dcache(). We descend to the next level
1228 * whenever the d_subdirs list is non-empty and continue
1229 * searching.
1230 *
1231 * It returns zero iff there are no unused children,
1232 * otherwise it returns the number of children moved to
1233 * the end of the unused list. This may not be the total
1234 * number of unused children, because select_parent can
1235 * drop the lock and return early due to latency
1236 * constraints.
1237 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001238
1239struct select_data {
1240 struct dentry *start;
1241 struct list_head dispose;
1242 int found;
1243};
1244
1245static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001247 struct select_data *data = _data;
1248 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001250 if (data->start == dentry)
1251 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
Al Virofe915222014-05-03 00:02:25 -04001253 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001254 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001255 } else {
1256 if (dentry->d_flags & DCACHE_LRU_LIST)
1257 d_lru_del(dentry);
1258 if (!dentry->d_lockref.count) {
1259 d_shrink_add(dentry, &data->dispose);
1260 data->found++;
1261 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 }
1263 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001264 * We can return to the caller if we have found some (this
1265 * ensures forward progress). We'll be coming back to find
1266 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 */
Al Virofe915222014-05-03 00:02:25 -04001268 if (!list_empty(&data->dispose))
1269 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001271 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272}
1273
1274/**
1275 * shrink_dcache_parent - prune dcache
1276 * @parent: parent of entries to prune
1277 *
1278 * Prune the dcache to remove unused children of the parent dentry.
1279 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001280void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001282 for (;;) {
1283 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001285 INIT_LIST_HEAD(&data.dispose);
1286 data.start = parent;
1287 data.found = 0;
1288
1289 d_walk(parent, &data, select_collect, NULL);
1290 if (!data.found)
1291 break;
1292
1293 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001294 cond_resched();
1295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001297EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
Al Viro9c8c10e2014-05-02 20:36:10 -04001299static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001300{
Al Viro9c8c10e2014-05-02 20:36:10 -04001301 /* it has busy descendents; complain about those instead */
1302 if (!list_empty(&dentry->d_subdirs))
1303 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001304
Al Viro9c8c10e2014-05-02 20:36:10 -04001305 /* root with refcount 1 is fine */
1306 if (dentry == _data && dentry->d_lockref.count == 1)
1307 return D_WALK_CONTINUE;
1308
1309 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1310 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001311 dentry,
1312 dentry->d_inode ?
1313 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001314 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001315 dentry->d_lockref.count,
1316 dentry->d_sb->s_type->name,
1317 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001318 WARN_ON(1);
1319 return D_WALK_CONTINUE;
1320}
1321
1322static void do_one_tree(struct dentry *dentry)
1323{
1324 shrink_dcache_parent(dentry);
1325 d_walk(dentry, dentry, umount_check, NULL);
1326 d_drop(dentry);
1327 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001328}
1329
1330/*
1331 * destroy the dentries attached to a superblock on unmounting
1332 */
1333void shrink_dcache_for_umount(struct super_block *sb)
1334{
1335 struct dentry *dentry;
1336
Al Viro9c8c10e2014-05-02 20:36:10 -04001337 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001338
1339 dentry = sb->s_root;
1340 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001341 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001342
1343 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001344 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1345 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001346 }
1347}
1348
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001349static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1350{
1351 struct select_data *data = _data;
1352
1353 if (d_mountpoint(dentry)) {
1354 data->found = -EBUSY;
1355 return D_WALK_QUIT;
1356 }
1357
1358 return select_collect(_data, dentry);
1359}
1360
1361static void check_and_drop(void *_data)
1362{
1363 struct select_data *data = _data;
1364
1365 if (d_mountpoint(data->start))
1366 data->found = -EBUSY;
1367 if (!data->found)
1368 __d_drop(data->start);
1369}
1370
1371/**
1372 * check_submounts_and_drop - prune dcache, check for submounts and drop
1373 *
1374 * All done as a single atomic operation relative to has_unlinked_ancestor().
1375 * Returns 0 if successfully unhashed @parent. If there were submounts then
1376 * return -EBUSY.
1377 *
1378 * @dentry: dentry to prune and drop
1379 */
1380int check_submounts_and_drop(struct dentry *dentry)
1381{
1382 int ret = 0;
1383
1384 /* Negative dentries can be dropped without further checks */
1385 if (!dentry->d_inode) {
1386 d_drop(dentry);
1387 goto out;
1388 }
1389
1390 for (;;) {
1391 struct select_data data;
1392
1393 INIT_LIST_HEAD(&data.dispose);
1394 data.start = dentry;
1395 data.found = 0;
1396
1397 d_walk(dentry, &data, check_and_collect, check_and_drop);
1398 ret = data.found;
1399
1400 if (!list_empty(&data.dispose))
1401 shrink_dentry_list(&data.dispose);
1402
1403 if (ret <= 0)
1404 break;
1405
1406 cond_resched();
1407 }
1408
1409out:
1410 return ret;
1411}
1412EXPORT_SYMBOL(check_submounts_and_drop);
1413
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414/**
Al Viroa4464db2011-07-07 15:03:58 -04001415 * __d_alloc - allocate a dcache entry
1416 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 * @name: qstr of the name
1418 *
1419 * Allocates a dentry. It returns %NULL if there is insufficient memory
1420 * available. On a success the dentry is returned. The name passed in is
1421 * copied and the copy passed in may be reused after this call.
1422 */
1423
Al Viroa4464db2011-07-07 15:03:58 -04001424struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425{
1426 struct dentry *dentry;
1427 char *dname;
1428
Mel Gormane12ba742007-10-16 01:25:52 -07001429 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 if (!dentry)
1431 return NULL;
1432
Linus Torvalds6326c712012-05-21 16:14:04 -07001433 /*
1434 * We guarantee that the inline name is always NUL-terminated.
1435 * This way the memcpy() done by the name switching in rename
1436 * will still always have a NUL at the end, even if we might
1437 * be overwriting an internal NUL character
1438 */
1439 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 if (name->len > DNAME_INLINE_LEN-1) {
1441 dname = kmalloc(name->len + 1, GFP_KERNEL);
1442 if (!dname) {
1443 kmem_cache_free(dentry_cache, dentry);
1444 return NULL;
1445 }
1446 } else {
1447 dname = dentry->d_iname;
1448 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
1450 dentry->d_name.len = name->len;
1451 dentry->d_name.hash = name->hash;
1452 memcpy(dname, name->name, name->len);
1453 dname[name->len] = 0;
1454
Linus Torvalds6326c712012-05-21 16:14:04 -07001455 /* Make sure we always see the terminating NUL character */
1456 smp_wmb();
1457 dentry->d_name.name = dname;
1458
Waiman Long98474232013-08-28 18:24:59 -07001459 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001460 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001462 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001464 dentry->d_parent = dentry;
1465 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 dentry->d_op = NULL;
1467 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001468 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 INIT_LIST_HEAD(&dentry->d_lru);
1470 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001471 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001472 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001473 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
Nick Piggin3e880fb2011-01-07 17:49:19 +11001475 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001476
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 return dentry;
1478}
Al Viroa4464db2011-07-07 15:03:58 -04001479
1480/**
1481 * d_alloc - allocate a dcache entry
1482 * @parent: parent of entry to allocate
1483 * @name: qstr of the name
1484 *
1485 * Allocates a dentry. It returns %NULL if there is insufficient memory
1486 * available. On a success the dentry is returned. The name passed in is
1487 * copied and the copy passed in may be reused after this call.
1488 */
1489struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1490{
1491 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1492 if (!dentry)
1493 return NULL;
1494
1495 spin_lock(&parent->d_lock);
1496 /*
1497 * don't need child lock because it is not subject
1498 * to concurrency here
1499 */
1500 __dget_dlock(parent);
1501 dentry->d_parent = parent;
1502 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1503 spin_unlock(&parent->d_lock);
1504
1505 return dentry;
1506}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001507EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001509/**
1510 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1511 * @sb: the superblock
1512 * @name: qstr of the name
1513 *
1514 * For a filesystem that just pins its dentries in memory and never
1515 * performs lookups at all, return an unhashed IS_ROOT dentry.
1516 */
Nick Piggin4b936882011-01-07 17:50:07 +11001517struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1518{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001519 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001520}
1521EXPORT_SYMBOL(d_alloc_pseudo);
1522
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1524{
1525 struct qstr q;
1526
1527 q.name = name;
1528 q.len = strlen(name);
1529 q.hash = full_name_hash(q.name, q.len);
1530 return d_alloc(parent, &q);
1531}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001532EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
Nick Pigginfb045ad2011-01-07 17:49:55 +11001534void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1535{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001536 WARN_ON_ONCE(dentry->d_op);
1537 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001538 DCACHE_OP_COMPARE |
1539 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001540 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001541 DCACHE_OP_DELETE ));
1542 dentry->d_op = op;
1543 if (!op)
1544 return;
1545 if (op->d_hash)
1546 dentry->d_flags |= DCACHE_OP_HASH;
1547 if (op->d_compare)
1548 dentry->d_flags |= DCACHE_OP_COMPARE;
1549 if (op->d_revalidate)
1550 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001551 if (op->d_weak_revalidate)
1552 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001553 if (op->d_delete)
1554 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001555 if (op->d_prune)
1556 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001557
1558}
1559EXPORT_SYMBOL(d_set_d_op);
1560
David Howellsb18825a2013-09-12 19:22:53 +01001561static unsigned d_flags_for_inode(struct inode *inode)
1562{
1563 unsigned add_flags = DCACHE_FILE_TYPE;
1564
1565 if (!inode)
1566 return DCACHE_MISS_TYPE;
1567
1568 if (S_ISDIR(inode->i_mode)) {
1569 add_flags = DCACHE_DIRECTORY_TYPE;
1570 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1571 if (unlikely(!inode->i_op->lookup))
1572 add_flags = DCACHE_AUTODIR_TYPE;
1573 else
1574 inode->i_opflags |= IOP_LOOKUP;
1575 }
1576 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1577 if (unlikely(inode->i_op->follow_link))
1578 add_flags = DCACHE_SYMLINK_TYPE;
1579 else
1580 inode->i_opflags |= IOP_NOFOLLOW;
1581 }
1582
1583 if (unlikely(IS_AUTOMOUNT(inode)))
1584 add_flags |= DCACHE_NEED_AUTOMOUNT;
1585 return add_flags;
1586}
1587
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001588static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1589{
David Howellsb18825a2013-09-12 19:22:53 +01001590 unsigned add_flags = d_flags_for_inode(inode);
1591
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001592 spin_lock(&dentry->d_lock);
Al Viro22213312014-04-19 12:30:58 -04001593 __d_set_type(dentry, add_flags);
David Howellsb18825a2013-09-12 19:22:53 +01001594 if (inode)
Al Virob3d9b7a2012-06-09 13:51:19 -04001595 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001596 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001597 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001598 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001599 fsnotify_d_instantiate(dentry, inode);
1600}
1601
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602/**
1603 * d_instantiate - fill in inode information for a dentry
1604 * @entry: dentry to complete
1605 * @inode: inode to attach to this dentry
1606 *
1607 * Fill in inode information in the entry.
1608 *
1609 * This turns negative dentries into productive full members
1610 * of society.
1611 *
1612 * NOTE! This assumes that the inode count has been incremented
1613 * (or otherwise set) by the caller to indicate that it is now
1614 * in use by the dcache.
1615 */
1616
1617void d_instantiate(struct dentry *entry, struct inode * inode)
1618{
Al Virob3d9b7a2012-06-09 13:51:19 -04001619 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001620 if (inode)
1621 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001622 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001623 if (inode)
1624 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 security_d_instantiate(entry, inode);
1626}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001627EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629/**
1630 * d_instantiate_unique - instantiate a non-aliased dentry
1631 * @entry: dentry to instantiate
1632 * @inode: inode to attach to this dentry
1633 *
1634 * Fill in inode information in the entry. On success, it returns NULL.
1635 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001636 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 *
1638 * Note that in order to avoid conflicts with rename() etc, the caller
1639 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001640 *
1641 * This also assumes that the inode count has been incremented
1642 * (or otherwise set) by the caller to indicate that it is now
1643 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 */
David Howells770bfad2006-08-22 20:06:07 -04001645static struct dentry *__d_instantiate_unique(struct dentry *entry,
1646 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647{
1648 struct dentry *alias;
1649 int len = entry->d_name.len;
1650 const char *name = entry->d_name.name;
1651 unsigned int hash = entry->d_name.hash;
1652
David Howells770bfad2006-08-22 20:06:07 -04001653 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001654 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001655 return NULL;
1656 }
1657
Sasha Levinb67bfe02013-02-27 17:06:00 -08001658 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001659 /*
1660 * Don't need alias->d_lock here, because aliases with
1661 * d_parent == entry->d_parent are not subject to name or
1662 * parent changes, because the parent inode i_mutex is held.
1663 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001664 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 continue;
1666 if (alias->d_parent != entry->d_parent)
1667 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001668 if (alias->d_name.len != len)
1669 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001670 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001672 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 return alias;
1674 }
David Howells770bfad2006-08-22 20:06:07 -04001675
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001676 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 return NULL;
1678}
David Howells770bfad2006-08-22 20:06:07 -04001679
1680struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1681{
1682 struct dentry *result;
1683
Al Virob3d9b7a2012-06-09 13:51:19 -04001684 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001685
Nick Piggin873feea2011-01-07 17:50:06 +11001686 if (inode)
1687 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001688 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001689 if (inode)
1690 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001691
1692 if (!result) {
1693 security_d_instantiate(entry, inode);
1694 return NULL;
1695 }
1696
1697 BUG_ON(!d_unhashed(result));
1698 iput(inode);
1699 return result;
1700}
1701
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702EXPORT_SYMBOL(d_instantiate_unique);
1703
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001704/**
1705 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1706 * @entry: dentry to complete
1707 * @inode: inode to attach to this dentry
1708 *
1709 * Fill in inode information in the entry. If a directory alias is found, then
1710 * return an error (and drop inode). Together with d_materialise_unique() this
1711 * guarantees that a directory inode may never have more than one alias.
1712 */
1713int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1714{
1715 BUG_ON(!hlist_unhashed(&entry->d_alias));
1716
1717 spin_lock(&inode->i_lock);
1718 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1719 spin_unlock(&inode->i_lock);
1720 iput(inode);
1721 return -EBUSY;
1722 }
1723 __d_instantiate(entry, inode);
1724 spin_unlock(&inode->i_lock);
1725 security_d_instantiate(entry, inode);
1726
1727 return 0;
1728}
1729EXPORT_SYMBOL(d_instantiate_no_diralias);
1730
Al Viroadc0e912012-01-08 16:49:21 -05001731struct dentry *d_make_root(struct inode *root_inode)
1732{
1733 struct dentry *res = NULL;
1734
1735 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001736 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001737
1738 res = __d_alloc(root_inode->i_sb, &name);
1739 if (res)
1740 d_instantiate(res, root_inode);
1741 else
1742 iput(root_inode);
1743 }
1744 return res;
1745}
1746EXPORT_SYMBOL(d_make_root);
1747
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001748static struct dentry * __d_find_any_alias(struct inode *inode)
1749{
1750 struct dentry *alias;
1751
Al Virob3d9b7a2012-06-09 13:51:19 -04001752 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001753 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001754 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001755 __dget(alias);
1756 return alias;
1757}
1758
Sage Weil46f72b32012-01-10 09:04:37 -08001759/**
1760 * d_find_any_alias - find any alias for a given inode
1761 * @inode: inode to find an alias for
1762 *
1763 * If any aliases exist for the given inode, take and return a
1764 * reference for one of them. If no aliases exist, return %NULL.
1765 */
1766struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001767{
1768 struct dentry *de;
1769
1770 spin_lock(&inode->i_lock);
1771 de = __d_find_any_alias(inode);
1772 spin_unlock(&inode->i_lock);
1773 return de;
1774}
Sage Weil46f72b32012-01-10 09:04:37 -08001775EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001776
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001777static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001778{
NeilBrownb911a6b2012-11-08 16:09:37 -08001779 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001780 struct dentry *tmp;
1781 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001782 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001783
1784 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001785 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001786 if (IS_ERR(inode))
1787 return ERR_CAST(inode);
1788
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001789 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001790 if (res)
1791 goto out_iput;
1792
Al Viroa4464db2011-07-07 15:03:58 -04001793 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001794 if (!tmp) {
1795 res = ERR_PTR(-ENOMEM);
1796 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001797 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001798
Nick Piggin873feea2011-01-07 17:50:06 +11001799 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001800 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001801 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001802 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001803 dput(tmp);
1804 goto out_iput;
1805 }
1806
1807 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001808 add_flags = d_flags_for_inode(inode);
1809
1810 if (disconnected)
1811 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001812
Christoph Hellwig9308a612008-08-11 15:49:12 +02001813 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001814 tmp->d_inode = inode;
David Howellsb18825a2013-09-12 19:22:53 +01001815 tmp->d_flags |= add_flags;
Al Virob3d9b7a2012-06-09 13:51:19 -04001816 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001817 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001818 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001819 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001820 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001821 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001822 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001823
Christoph Hellwig9308a612008-08-11 15:49:12 +02001824 return tmp;
1825
1826 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001827 if (res && !IS_ERR(res))
1828 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001829 iput(inode);
1830 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001831}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001832
1833/**
1834 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1835 * @inode: inode to allocate the dentry for
1836 *
1837 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1838 * similar open by handle operations. The returned dentry may be anonymous,
1839 * or may have a full name (if the inode was already in the cache).
1840 *
1841 * When called on a directory inode, we must ensure that the inode only ever
1842 * has one dentry. If a dentry is found, that is returned instead of
1843 * allocating a new one.
1844 *
1845 * On successful return, the reference to the inode has been transferred
1846 * to the dentry. In case of an error the reference on the inode is released.
1847 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1848 * be passed in and the error will be propagated to the return value,
1849 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1850 */
1851struct dentry *d_obtain_alias(struct inode *inode)
1852{
1853 return __d_obtain_alias(inode, 1);
1854}
Benny Halevyadc48722009-02-27 14:02:59 -08001855EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856
1857/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001858 * d_obtain_root - find or allocate a dentry for a given inode
1859 * @inode: inode to allocate the dentry for
1860 *
1861 * Obtain an IS_ROOT dentry for the root of a filesystem.
1862 *
1863 * We must ensure that directory inodes only ever have one dentry. If a
1864 * dentry is found, that is returned instead of allocating a new one.
1865 *
1866 * On successful return, the reference to the inode has been transferred
1867 * to the dentry. In case of an error the reference on the inode is
1868 * released. A %NULL or IS_ERR inode may be passed in and will be the
1869 * error will be propagate to the return value, with a %NULL @inode
1870 * replaced by ERR_PTR(-ESTALE).
1871 */
1872struct dentry *d_obtain_root(struct inode *inode)
1873{
1874 return __d_obtain_alias(inode, 0);
1875}
1876EXPORT_SYMBOL(d_obtain_root);
1877
1878/**
Barry Naujok94035402008-05-21 16:50:46 +10001879 * d_add_ci - lookup or allocate new dentry with case-exact name
1880 * @inode: the inode case-insensitive lookup has found
1881 * @dentry: the negative dentry that was passed to the parent's lookup func
1882 * @name: the case-exact name to be associated with the returned dentry
1883 *
1884 * This is to avoid filling the dcache with case-insensitive names to the
1885 * same inode, only the actual correct case is stored in the dcache for
1886 * case-insensitive filesystems.
1887 *
1888 * For a case-insensitive lookup match and if the the case-exact dentry
1889 * already exists in in the dcache, use it and return it.
1890 *
1891 * If no entry exists with the exact case name, allocate new dentry with
1892 * the exact case, and return the spliced entry.
1893 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001894struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001895 struct qstr *name)
1896{
Barry Naujok94035402008-05-21 16:50:46 +10001897 struct dentry *found;
1898 struct dentry *new;
1899
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001900 /*
1901 * First check if a dentry matching the name already exists,
1902 * if not go ahead and create it now.
1903 */
Barry Naujok94035402008-05-21 16:50:46 +10001904 found = d_hash_and_lookup(dentry->d_parent, name);
Al Viro4f522a22013-02-11 23:20:37 -05001905 if (unlikely(IS_ERR(found)))
1906 goto err_out;
Barry Naujok94035402008-05-21 16:50:46 +10001907 if (!found) {
1908 new = d_alloc(dentry->d_parent, name);
1909 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05001910 found = ERR_PTR(-ENOMEM);
Barry Naujok94035402008-05-21 16:50:46 +10001911 goto err_out;
1912 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001913
Barry Naujok94035402008-05-21 16:50:46 +10001914 found = d_splice_alias(inode, new);
1915 if (found) {
1916 dput(new);
1917 return found;
1918 }
1919 return new;
1920 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001921
1922 /*
1923 * If a matching dentry exists, and it's not negative use it.
1924 *
1925 * Decrement the reference count to balance the iget() done
1926 * earlier on.
1927 */
Barry Naujok94035402008-05-21 16:50:46 +10001928 if (found->d_inode) {
1929 if (unlikely(found->d_inode != inode)) {
1930 /* This can't happen because bad inodes are unhashed. */
1931 BUG_ON(!is_bad_inode(inode));
1932 BUG_ON(!is_bad_inode(found->d_inode));
1933 }
Barry Naujok94035402008-05-21 16:50:46 +10001934 iput(inode);
1935 return found;
1936 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001937
Barry Naujok94035402008-05-21 16:50:46 +10001938 /*
1939 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001940 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001941 */
Al Viro4513d892011-07-17 10:52:14 -04001942 new = d_splice_alias(inode, found);
1943 if (new) {
1944 dput(found);
1945 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10001946 }
Al Viro4513d892011-07-17 10:52:14 -04001947 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001948
1949err_out:
1950 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05001951 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001952}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001953EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001955/*
1956 * Do the slow-case of the dentry name compare.
1957 *
1958 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07001959 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001960 * filesystem can rely on them, and can use the 'name' and
1961 * 'len' information without worrying about walking off the
1962 * end of memory etc.
1963 *
1964 * Thus the read_seqcount_retry() and the "duplicate" info
1965 * in arguments (the low-level filesystem should not look
1966 * at the dentry inode or name contents directly, since
1967 * rename can change them while we're in RCU mode).
1968 */
1969enum slow_d_compare {
1970 D_COMP_OK,
1971 D_COMP_NOMATCH,
1972 D_COMP_SEQRETRY,
1973};
1974
1975static noinline enum slow_d_compare slow_dentry_cmp(
1976 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001977 struct dentry *dentry,
1978 unsigned int seq,
1979 const struct qstr *name)
1980{
1981 int tlen = dentry->d_name.len;
1982 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001983
1984 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1985 cpu_relax();
1986 return D_COMP_SEQRETRY;
1987 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07001988 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001989 return D_COMP_NOMATCH;
1990 return D_COMP_OK;
1991}
1992
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993/**
Nick Piggin31e6b012011-01-07 17:49:52 +11001994 * __d_lookup_rcu - search for a dentry (racy, store-free)
1995 * @parent: parent dentry
1996 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07001997 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11001998 * Returns: dentry, or NULL
1999 *
2000 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2001 * resolution (store-free path walking) design described in
2002 * Documentation/filesystems/path-lookup.txt.
2003 *
2004 * This is not to be used outside core vfs.
2005 *
2006 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2007 * held, and rcu_read_lock held. The returned dentry must not be stored into
2008 * without taking d_lock and checking d_seq sequence count against @seq
2009 * returned here.
2010 *
Linus Torvalds15570082013-09-02 11:38:06 -07002011 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002012 * function.
2013 *
2014 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2015 * the returned dentry, so long as its parent's seqlock is checked after the
2016 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2017 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002018 *
2019 * NOTE! The caller *has* to check the resulting dentry against the sequence
2020 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002021 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002022struct dentry *__d_lookup_rcu(const struct dentry *parent,
2023 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002024 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002025{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002026 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002027 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002028 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002029 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002030 struct dentry *dentry;
2031
2032 /*
2033 * Note: There is significant duplication with __d_lookup_rcu which is
2034 * required to prevent single threaded performance regressions
2035 * especially on architectures where smp_rmb (in seqcounts) are costly.
2036 * Keep the two functions in sync.
2037 */
2038
2039 /*
2040 * The hash list is protected using RCU.
2041 *
2042 * Carefully use d_seq when comparing a candidate dentry, to avoid
2043 * races with d_move().
2044 *
2045 * It is possible that concurrent renames can mess up our list
2046 * walk here and result in missing our dentry, resulting in the
2047 * false-negative result. d_lookup() protects against concurrent
2048 * renames using rename_lock seqlock.
2049 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002050 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002051 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002052 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002053 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002054
Nick Piggin31e6b012011-01-07 17:49:52 +11002055seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002056 /*
2057 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002058 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002059 *
2060 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002061 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002062 *
2063 * NOTE! We do a "raw" seqcount_begin here. That means that
2064 * we don't wait for the sequence count to stabilize if it
2065 * is in the middle of a sequence change. If we do the slow
2066 * dentry compare, we will do seqretries until it is stable,
2067 * and if we end up with a successful lookup, we actually
2068 * want to exit RCU lookup anyway.
2069 */
2070 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002071 if (dentry->d_parent != parent)
2072 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002073 if (d_unhashed(dentry))
2074 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002075
2076 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002077 if (dentry->d_name.hash != hashlen_hash(hashlen))
2078 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002079 *seqp = seq;
2080 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002081 case D_COMP_OK:
2082 return dentry;
2083 case D_COMP_NOMATCH:
2084 continue;
2085 default:
2086 goto seqretry;
2087 }
2088 }
2089
Linus Torvalds26fe5752012-05-10 13:14:12 -07002090 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002091 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002092 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002093 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002094 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002095 }
2096 return NULL;
2097}
2098
2099/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 * d_lookup - search for a dentry
2101 * @parent: parent dentry
2102 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002103 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002105 * d_lookup searches the children of the parent dentry for the name in
2106 * question. If the dentry is found its reference count is incremented and the
2107 * dentry is returned. The caller must use dput to free the entry when it has
2108 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 */
Al Viroda2d8452013-01-24 18:29:34 -05002110struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111{
Nick Piggin31e6b012011-01-07 17:49:52 +11002112 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002113 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114
2115 do {
2116 seq = read_seqbegin(&rename_lock);
2117 dentry = __d_lookup(parent, name);
2118 if (dentry)
2119 break;
2120 } while (read_seqretry(&rename_lock, seq));
2121 return dentry;
2122}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002123EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
Nick Piggin31e6b012011-01-07 17:49:52 +11002125/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002126 * __d_lookup - search for a dentry (racy)
2127 * @parent: parent dentry
2128 * @name: qstr of name we wish to find
2129 * Returns: dentry, or NULL
2130 *
2131 * __d_lookup is like d_lookup, however it may (rarely) return a
2132 * false-negative result due to unrelated rename activity.
2133 *
2134 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2135 * however it must be used carefully, eg. with a following d_lookup in
2136 * the case of failure.
2137 *
2138 * __d_lookup callers must be commented.
2139 */
Al Viroa713ca22013-01-24 18:27:00 -05002140struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141{
2142 unsigned int len = name->len;
2143 unsigned int hash = name->hash;
2144 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002145 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002146 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002147 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002148 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149
Nick Pigginb04f7842010-08-18 04:37:34 +10002150 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002151 * Note: There is significant duplication with __d_lookup_rcu which is
2152 * required to prevent single threaded performance regressions
2153 * especially on architectures where smp_rmb (in seqcounts) are costly.
2154 * Keep the two functions in sync.
2155 */
2156
2157 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002158 * The hash list is protected using RCU.
2159 *
2160 * Take d_lock when comparing a candidate dentry, to avoid races
2161 * with d_move().
2162 *
2163 * It is possible that concurrent renames can mess up our list
2164 * walk here and result in missing our dentry, resulting in the
2165 * false-negative result. d_lookup() protects against concurrent
2166 * renames using rename_lock seqlock.
2167 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002168 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002169 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 rcu_read_lock();
2171
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002172 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 if (dentry->d_name.hash != hash)
2175 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176
2177 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 if (dentry->d_parent != parent)
2179 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002180 if (d_unhashed(dentry))
2181 goto next;
2182
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 /*
2184 * It is safe to compare names since d_move() cannot
2185 * change the qstr (protected by d_lock).
2186 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002187 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002188 int tlen = dentry->d_name.len;
2189 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002190 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 goto next;
2192 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002193 if (dentry->d_name.len != len)
2194 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002195 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 goto next;
2197 }
2198
Waiman Long98474232013-08-28 18:24:59 -07002199 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002200 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 spin_unlock(&dentry->d_lock);
2202 break;
2203next:
2204 spin_unlock(&dentry->d_lock);
2205 }
2206 rcu_read_unlock();
2207
2208 return found;
2209}
2210
2211/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002212 * d_hash_and_lookup - hash the qstr then search for a dentry
2213 * @dir: Directory to search in
2214 * @name: qstr of name we wish to find
2215 *
Al Viro4f522a22013-02-11 23:20:37 -05002216 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002217 */
2218struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2219{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002220 /*
2221 * Check for a fs-specific hash function. Note that we must
2222 * calculate the standard hash first, as the d_op->d_hash()
2223 * routine may choose to leave the hash value unchanged.
2224 */
2225 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002226 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002227 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002228 if (unlikely(err < 0))
2229 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002230 }
Al Viro4f522a22013-02-11 23:20:37 -05002231 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002232}
Al Viro4f522a22013-02-11 23:20:37 -05002233EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002234
2235/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002236 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002238 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 *
2240 * An insecure source has sent us a dentry, here we verify it and dget() it.
2241 * This is used by ncpfs in its readdir implementation.
2242 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002243 *
2244 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002246int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247{
Nick Piggin786a5e12011-01-07 17:49:16 +11002248 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002249
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002250 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002251 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2252 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002253 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002254 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002255 spin_unlock(&dentry->d_lock);
2256 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 return 1;
2258 }
2259 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002260 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002261
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262 return 0;
2263}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002264EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
2266/*
2267 * When a file is deleted, we have two options:
2268 * - turn this dentry into a negative dentry
2269 * - unhash this dentry and free it.
2270 *
2271 * Usually, we want to just turn this into
2272 * a negative dentry, but if anybody else is
2273 * currently using the dentry or the inode
2274 * we can't do that and we fall back on removing
2275 * it from the hash queues and waiting for
2276 * it to be deleted later when it has no users
2277 */
2278
2279/**
2280 * d_delete - delete a dentry
2281 * @dentry: The dentry to delete
2282 *
2283 * Turn the dentry into a negative dentry if possible, otherwise
2284 * remove it from the hash queues so it can be deleted later
2285 */
2286
2287void d_delete(struct dentry * dentry)
2288{
Nick Piggin873feea2011-01-07 17:50:06 +11002289 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002290 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 /*
2292 * Are we the only user?
2293 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002294again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002296 inode = dentry->d_inode;
2297 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002298 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002299 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002300 spin_unlock(&dentry->d_lock);
2301 cpu_relax();
2302 goto again;
2303 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002304 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002305 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002306 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 return;
2308 }
2309
2310 if (!d_unhashed(dentry))
2311 __d_drop(dentry);
2312
2313 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002314
2315 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002317EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002319static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002321 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002322 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002323 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002324 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002325 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326}
2327
David Howells770bfad2006-08-22 20:06:07 -04002328static void _d_rehash(struct dentry * entry)
2329{
2330 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2331}
2332
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333/**
2334 * d_rehash - add an entry back to the hash
2335 * @entry: dentry to add to the hash
2336 *
2337 * Adds a dentry to the hash according to its name.
2338 */
2339
2340void d_rehash(struct dentry * entry)
2341{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002343 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002346EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002348/**
2349 * dentry_update_name_case - update case insensitive dentry with a new name
2350 * @dentry: dentry to be updated
2351 * @name: new name
2352 *
2353 * Update a case insensitive dentry with new case of name.
2354 *
2355 * dentry must have been returned by d_lookup with name @name. Old and new
2356 * name lengths must match (ie. no d_compare which allows mismatched name
2357 * lengths).
2358 *
2359 * Parent inode i_mutex must be held over d_lookup and into this call (to
2360 * keep renames and concurrent inserts, and readdir(2) away).
2361 */
2362void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2363{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002364 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002365 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2366
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002367 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002368 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002369 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002370 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002371 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002372}
2373EXPORT_SYMBOL(dentry_update_name_case);
2374
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002375static void switch_names(struct dentry *dentry, struct dentry *target,
2376 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377{
2378 if (dname_external(target)) {
2379 if (dname_external(dentry)) {
2380 /*
2381 * Both external: swap the pointers
2382 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002383 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 } else {
2385 /*
2386 * dentry:internal, target:external. Steal target's
2387 * storage and make target internal.
2388 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002389 memcpy(target->d_iname, dentry->d_name.name,
2390 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 dentry->d_name.name = target->d_name.name;
2392 target->d_name.name = target->d_iname;
2393 }
2394 } else {
2395 if (dname_external(dentry)) {
2396 /*
2397 * dentry:external, target:internal. Give dentry's
2398 * storage to target and make dentry internal
2399 */
2400 memcpy(dentry->d_iname, target->d_name.name,
2401 target->d_name.len + 1);
2402 target->d_name.name = dentry->d_name.name;
2403 dentry->d_name.name = dentry->d_iname;
2404 } else {
2405 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002406 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002408 unsigned int i;
2409 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002410 if (!exchange) {
2411 memcpy(dentry->d_iname, target->d_name.name,
2412 target->d_name.len + 1);
2413 dentry->d_name.hash_len = target->d_name.hash_len;
2414 return;
2415 }
Miklos Szeredida1ce062014-04-01 17:08:43 +02002416 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2417 swap(((long *) &dentry->d_iname)[i],
2418 ((long *) &target->d_iname)[i]);
2419 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 }
2421 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002422 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423}
2424
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002425static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2426{
2427 /*
2428 * XXXX: do we really need to take target->d_lock?
2429 */
2430 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2431 spin_lock(&target->d_parent->d_lock);
2432 else {
2433 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2434 spin_lock(&dentry->d_parent->d_lock);
2435 spin_lock_nested(&target->d_parent->d_lock,
2436 DENTRY_D_LOCK_NESTED);
2437 } else {
2438 spin_lock(&target->d_parent->d_lock);
2439 spin_lock_nested(&dentry->d_parent->d_lock,
2440 DENTRY_D_LOCK_NESTED);
2441 }
2442 }
2443 if (target < dentry) {
2444 spin_lock_nested(&target->d_lock, 2);
2445 spin_lock_nested(&dentry->d_lock, 3);
2446 } else {
2447 spin_lock_nested(&dentry->d_lock, 2);
2448 spin_lock_nested(&target->d_lock, 3);
2449 }
2450}
2451
Al Viro986c0192014-09-26 23:11:15 -04002452static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002453{
2454 if (target->d_parent != dentry->d_parent)
2455 spin_unlock(&dentry->d_parent->d_lock);
2456 if (target->d_parent != target)
2457 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002458 spin_unlock(&target->d_lock);
2459 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002460}
2461
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002463 * When switching names, the actual string doesn't strictly have to
2464 * be preserved in the target - because we're dropping the target
2465 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002466 * the new name before we switch, unless we are going to rehash
2467 * it. Note that if we *do* unhash the target, we are not allowed
2468 * to rehash it without giving it a new name/hash key - whether
2469 * we swap or overwrite the names here, resulting name won't match
2470 * the reality in filesystem; it's only there for d_path() purposes.
2471 * Note that all of this is happening under rename_lock, so the
2472 * any hash lookup seeing it in the middle of manipulations will
2473 * be discarded anyway. So we do not care what happens to the hash
2474 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002476/*
Al Viro18367502011-07-12 21:42:24 -04002477 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 * @dentry: entry to move
2479 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002480 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 *
2482 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002483 * dcache entries should not be moved in this way. Caller must hold
2484 * rename_lock, the i_mutex of the source and target directories,
2485 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002487static void __d_move(struct dentry *dentry, struct dentry *target,
2488 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 if (!dentry->d_inode)
2491 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2492
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002493 BUG_ON(d_ancestor(dentry, target));
2494 BUG_ON(d_ancestor(target, dentry));
2495
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002496 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497
Nick Piggin31e6b012011-01-07 17:49:52 +11002498 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002499 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002500
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002501 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2502
2503 /*
2504 * Move the dentry to the target hash queue. Don't bother checking
2505 * for the same hash queue because of how unlikely it is.
2506 */
2507 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002508 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509
Miklos Szeredida1ce062014-04-01 17:08:43 +02002510 /*
2511 * Unhash the target (d_delete() is not usable here). If exchanging
2512 * the two dentries, then rehash onto the other's hash queue.
2513 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002515 if (exchange) {
2516 __d_rehash(target,
2517 d_hash(dentry->d_parent, dentry->d_name.hash));
2518 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 /* Switch the names.. */
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002521 switch_names(dentry, target, exchange);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522
Al Viro63cf4272014-09-26 23:06:14 -04002523 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002525 /* splicing a tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 dentry->d_parent = target->d_parent;
2527 target->d_parent = target;
Al Viro9d8cd302014-09-26 21:34:01 -04002528 list_del_init(&target->d_u.d_child);
Al Viro63cf4272014-09-26 23:06:14 -04002529 list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002531 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002532 swap(dentry->d_parent, target->d_parent);
Al Viro9d8cd302014-09-26 21:34:01 -04002533 list_move(&target->d_u.d_child, &target->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002534 list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2535 if (exchange)
2536 fsnotify_d_move(target);
2537 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 }
2539
Nick Piggin31e6b012011-01-07 17:49:52 +11002540 write_seqcount_end(&target->d_seq);
2541 write_seqcount_end(&dentry->d_seq);
2542
Al Viro986c0192014-09-26 23:11:15 -04002543 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002544}
2545
2546/*
2547 * d_move - move a dentry
2548 * @dentry: entry to move
2549 * @target: new dentry
2550 *
2551 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002552 * dcache entries should not be moved in this way. See the locking
2553 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002554 */
2555void d_move(struct dentry *dentry, struct dentry *target)
2556{
2557 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002558 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002560}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002561EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562
Miklos Szeredida1ce062014-04-01 17:08:43 +02002563/*
2564 * d_exchange - exchange two dentries
2565 * @dentry1: first dentry
2566 * @dentry2: second dentry
2567 */
2568void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2569{
2570 write_seqlock(&rename_lock);
2571
2572 WARN_ON(!dentry1->d_inode);
2573 WARN_ON(!dentry2->d_inode);
2574 WARN_ON(IS_ROOT(dentry1));
2575 WARN_ON(IS_ROOT(dentry2));
2576
2577 __d_move(dentry1, dentry2, true);
2578
2579 write_sequnlock(&rename_lock);
2580}
2581
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002582/**
2583 * d_ancestor - search for an ancestor
2584 * @p1: ancestor dentry
2585 * @p2: child dentry
2586 *
2587 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2588 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002589 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002590struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002591{
2592 struct dentry *p;
2593
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002594 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002595 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002596 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002597 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002598 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002599}
2600
2601/*
2602 * This helper attempts to cope with remotely renamed directories
2603 *
2604 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002605 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002606 *
2607 * Note: If ever the locking in lock_rename() changes, then please
2608 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002609 */
Nick Piggin873feea2011-01-07 17:50:06 +11002610static struct dentry *__d_unalias(struct inode *inode,
2611 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002612{
2613 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002614 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002615
2616 /* If alias and dentry share a parent, then no extra locks required */
2617 if (alias->d_parent == dentry->d_parent)
2618 goto out_unalias;
2619
Trond Myklebust9eaef272006-10-21 10:24:20 -07002620 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002621 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2622 goto out_err;
2623 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2624 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2625 goto out_err;
2626 m2 = &alias->d_parent->d_inode->i_mutex;
2627out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002628 if (likely(!d_mountpoint(alias))) {
Miklos Szeredida1ce062014-04-01 17:08:43 +02002629 __d_move(alias, dentry, false);
Al Viroee3efa92012-06-08 15:59:33 -04002630 ret = alias;
2631 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002632out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002633 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002634 if (m2)
2635 mutex_unlock(m2);
2636 if (m1)
2637 mutex_unlock(m1);
2638 return ret;
2639}
2640
David Howells770bfad2006-08-22 20:06:07 -04002641/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002642 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2643 * @inode: the inode which may have a disconnected dentry
2644 * @dentry: a negative dentry which we want to point to the inode.
2645 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002646 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2647 * place of the given dentry and return it, else simply d_add the inode
2648 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002649 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002650 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2651 * we should error out: directories can't have multiple aliases.
2652 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002653 * This is needed in the lookup routine of any filesystem that is exportable
2654 * (via knfsd) so that we can build dcache paths to directories effectively.
2655 *
2656 * If a dentry was found and moved, then it is returned. Otherwise NULL
2657 * is returned. This matches the expected return value of ->lookup.
2658 *
2659 * Cluster filesystems may call this function with a negative, hashed dentry.
2660 * In that case, we know that the inode will be a regular file, and also this
2661 * will only occur during atomic_open. So we need to check for the dentry
2662 * being already hashed only in the final case.
2663 */
2664struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2665{
2666 struct dentry *new = NULL;
2667
2668 if (IS_ERR(inode))
2669 return ERR_CAST(inode);
2670
2671 if (inode && S_ISDIR(inode->i_mode)) {
2672 spin_lock(&inode->i_lock);
J. Bruce Fields908790f2014-02-17 17:58:42 -05002673 new = __d_find_any_alias(inode);
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002674 if (new) {
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002675 if (!IS_ROOT(new)) {
J. Bruce Fields908790f2014-02-17 17:58:42 -05002676 spin_unlock(&inode->i_lock);
2677 dput(new);
2678 return ERR_PTR(-EIO);
2679 }
J. Bruce Fields95ad5c22014-03-12 12:19:23 -04002680 if (d_ancestor(new, dentry)) {
2681 spin_unlock(&inode->i_lock);
2682 dput(new);
2683 return ERR_PTR(-EIO);
2684 }
J. Bruce Fields75a23522014-02-17 17:45:56 -05002685 write_seqlock(&rename_lock);
Al Viro63cf4272014-09-26 23:06:14 -04002686 __d_move(new, dentry, false);
J. Bruce Fields75a23522014-02-17 17:45:56 -05002687 write_sequnlock(&rename_lock);
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002688 spin_unlock(&inode->i_lock);
2689 security_d_instantiate(new, inode);
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002690 iput(inode);
2691 } else {
2692 /* already taking inode->i_lock, so d_add() by hand */
2693 __d_instantiate(dentry, inode);
2694 spin_unlock(&inode->i_lock);
2695 security_d_instantiate(dentry, inode);
2696 d_rehash(dentry);
2697 }
2698 } else {
2699 d_instantiate(dentry, inode);
2700 if (d_unhashed(dentry))
2701 d_rehash(dentry);
2702 }
2703 return new;
2704}
2705EXPORT_SYMBOL(d_splice_alias);
2706
2707/**
David Howells770bfad2006-08-22 20:06:07 -04002708 * d_materialise_unique - introduce an inode into the tree
2709 * @dentry: candidate dentry
2710 * @inode: inode to bind to the dentry, to which aliases may be attached
2711 *
2712 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002713 * root directory alias in its place if there is one. Caller must hold the
2714 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002715 */
2716struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2717{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002718 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002719
2720 BUG_ON(!d_unhashed(dentry));
2721
David Howells770bfad2006-08-22 20:06:07 -04002722 if (!inode) {
2723 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002724 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002725 d_rehash(actual);
2726 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002727 }
2728
Nick Piggin873feea2011-01-07 17:50:06 +11002729 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002730
Trond Myklebust9eaef272006-10-21 10:24:20 -07002731 if (S_ISDIR(inode->i_mode)) {
2732 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002733
Trond Myklebust9eaef272006-10-21 10:24:20 -07002734 /* Does an aliased dentry already exist? */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -05002735 alias = __d_find_alias(inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002736 if (alias) {
2737 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002738 write_seqlock(&rename_lock);
2739
2740 if (d_ancestor(alias, dentry)) {
2741 /* Check for loops */
2742 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002743 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002744 } else if (IS_ROOT(alias)) {
2745 /* Is this an anonymous mountpoint that we
2746 * could splice into our tree? */
Al Viro63cf4272014-09-26 23:06:14 -04002747 __d_move(alias, dentry, false);
Al Viro18367502011-07-12 21:42:24 -04002748 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002749 goto found;
Al Viro18367502011-07-12 21:42:24 -04002750 } else {
2751 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002752 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002753 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002754 }
Al Viro18367502011-07-12 21:42:24 -04002755 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002756 if (IS_ERR(actual)) {
2757 if (PTR_ERR(actual) == -ELOOP)
2758 pr_warn_ratelimited(
2759 "VFS: Lookup of '%s' in %s %s"
2760 " would have caused loop\n",
2761 dentry->d_name.name,
2762 inode->i_sb->s_type->name,
2763 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002764 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002765 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002766 goto out_nolock;
2767 }
David Howells770bfad2006-08-22 20:06:07 -04002768 }
2769
2770 /* Add a unique reference */
2771 actual = __d_instantiate_unique(dentry, inode);
2772 if (!actual)
2773 actual = dentry;
David Howells770bfad2006-08-22 20:06:07 -04002774
Al Viro8527dd72014-09-26 21:26:50 -04002775 d_rehash(actual);
Al Viro5cc38212014-09-26 21:20:39 -04002776found:
Nick Piggin873feea2011-01-07 17:50:06 +11002777 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002778out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002779 if (actual == dentry) {
2780 security_d_instantiate(dentry, inode);
2781 return NULL;
2782 }
2783
2784 iput(inode);
2785 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002786}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002787EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002788
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002789static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002790{
2791 *buflen -= namelen;
2792 if (*buflen < 0)
2793 return -ENAMETOOLONG;
2794 *buffer -= namelen;
2795 memcpy(*buffer, str, namelen);
2796 return 0;
2797}
2798
Waiman Long232d2d62013-09-09 12:18:13 -04002799/**
2800 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002801 * @buffer: buffer pointer
2802 * @buflen: allocated length of the buffer
2803 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002804 *
2805 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2806 * make sure that either the old or the new name pointer and length are
2807 * fetched. However, there may be mismatch between length and pointer.
2808 * The length cannot be trusted, we need to copy it byte-by-byte until
2809 * the length is reached or a null byte is found. It also prepends "/" at
2810 * the beginning of the name. The sequence number check at the caller will
2811 * retry it again when a d_move() does happen. So any garbage in the buffer
2812 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04002813 *
2814 * Data dependency barrier is needed to make sure that we see that terminating
2815 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04002816 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002817static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2818{
Waiman Long232d2d62013-09-09 12:18:13 -04002819 const char *dname = ACCESS_ONCE(name->name);
2820 u32 dlen = ACCESS_ONCE(name->len);
2821 char *p;
2822
Al Viro6d13f692014-09-29 14:46:30 -04002823 smp_read_barrier_depends();
2824
Waiman Long232d2d62013-09-09 12:18:13 -04002825 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04002826 if (*buflen < 0)
2827 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04002828 p = *buffer -= dlen + 1;
2829 *p++ = '/';
2830 while (dlen--) {
2831 char c = *dname++;
2832 if (!c)
2833 break;
2834 *p++ = c;
2835 }
2836 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002837}
2838
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002840 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002841 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002842 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002843 * @buffer: pointer to the end of the buffer
2844 * @buflen: pointer to buffer length
2845 *
Waiman Long18129972013-09-12 10:55:35 -04002846 * The function will first try to write out the pathname without taking any
2847 * lock other than the RCU read lock to make sure that dentries won't go away.
2848 * It only checks the sequence number of the global rename_lock as any change
2849 * in the dentry's d_seq will be preceded by changes in the rename_lock
2850 * sequence number. If the sequence number had been changed, it will restart
2851 * the whole pathname back-tracing sequence again by taking the rename_lock.
2852 * In this case, there is no need to take the RCU read lock as the recursive
2853 * parent pointer references will keep the dentry chain alive as long as no
2854 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002855 */
Al Viro02125a82011-12-05 08:43:34 -05002856static int prepend_path(const struct path *path,
2857 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002858 char **buffer, int *buflen)
2859{
Al Viroede4ceb2013-11-13 07:45:40 -05002860 struct dentry *dentry;
2861 struct vfsmount *vfsmnt;
2862 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002863 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002864 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002865 char *bptr;
2866 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002867
Al Viro48f5ec22013-09-09 15:22:25 -04002868 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002869restart_mnt:
2870 read_seqbegin_or_lock(&mount_lock, &m_seq);
2871 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002872 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002873restart:
2874 bptr = *buffer;
2875 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002876 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002877 dentry = path->dentry;
2878 vfsmnt = path->mnt;
2879 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002880 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002881 while (dentry != root->dentry || vfsmnt != root->mnt) {
2882 struct dentry * parent;
2883
2884 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002885 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002886 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002887 if (mnt != parent) {
2888 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2889 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002890 vfsmnt = &mnt->mnt;
2891 continue;
2892 }
2893 /*
2894 * Filesystems needing to implement special "root names"
2895 * should do so with ->d_dname()
2896 */
2897 if (IS_ROOT(dentry) &&
2898 (dentry->d_name.len != 1 ||
2899 dentry->d_name.name[0] != '/')) {
2900 WARN(1, "Root dentry has weird name <%.*s>\n",
2901 (int) dentry->d_name.len,
2902 dentry->d_name.name);
2903 }
2904 if (!error)
2905 error = is_mounted(vfsmnt) ? 1 : 2;
2906 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002907 }
2908 parent = dentry->d_parent;
2909 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002910 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002911 if (error)
2912 break;
2913
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002914 dentry = parent;
2915 }
Al Viro48f5ec22013-09-09 15:22:25 -04002916 if (!(seq & 1))
2917 rcu_read_unlock();
2918 if (need_seqretry(&rename_lock, seq)) {
2919 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002920 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002921 }
Al Viro48f5ec22013-09-09 15:22:25 -04002922 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002923
2924 if (!(m_seq & 1))
2925 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002926 if (need_seqretry(&mount_lock, m_seq)) {
2927 m_seq = 1;
2928 goto restart_mnt;
2929 }
2930 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002931
Waiman Long232d2d62013-09-09 12:18:13 -04002932 if (error >= 0 && bptr == *buffer) {
2933 if (--blen < 0)
2934 error = -ENAMETOOLONG;
2935 else
2936 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002937 }
Waiman Long232d2d62013-09-09 12:18:13 -04002938 *buffer = bptr;
2939 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002940 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002941}
2942
2943/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002944 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002945 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002946 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002947 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948 * @buflen: buffer length
2949 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002950 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002952 * Returns a pointer into the buffer or an error code if the
2953 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002954 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002955 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002956 *
Al Viro02125a82011-12-05 08:43:34 -05002957 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958 */
Al Viro02125a82011-12-05 08:43:34 -05002959char *__d_path(const struct path *path,
2960 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002961 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002963 char *res = buf + buflen;
2964 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002966 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002967 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002968
Al Viro02125a82011-12-05 08:43:34 -05002969 if (error < 0)
2970 return ERR_PTR(error);
2971 if (error > 0)
2972 return NULL;
2973 return res;
2974}
2975
2976char *d_absolute_path(const struct path *path,
2977 char *buf, int buflen)
2978{
2979 struct path root = {};
2980 char *res = buf + buflen;
2981 int error;
2982
2983 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002984 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05002985
2986 if (error > 1)
2987 error = -EINVAL;
2988 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002989 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002990 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991}
2992
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002993/*
2994 * same as __d_path but appends "(deleted)" for unlinked files.
2995 */
Al Viro02125a82011-12-05 08:43:34 -05002996static int path_with_deleted(const struct path *path,
2997 const struct path *root,
2998 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002999{
3000 prepend(buf, buflen, "\0", 1);
3001 if (d_unlinked(path->dentry)) {
3002 int error = prepend(buf, buflen, " (deleted)", 10);
3003 if (error)
3004 return error;
3005 }
3006
3007 return prepend_path(path, root, buf, buflen);
3008}
3009
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003010static int prepend_unreachable(char **buffer, int *buflen)
3011{
3012 return prepend(buffer, buflen, "(unreachable)", 13);
3013}
3014
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003015static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3016{
3017 unsigned seq;
3018
3019 do {
3020 seq = read_seqcount_begin(&fs->seq);
3021 *root = fs->root;
3022 } while (read_seqcount_retry(&fs->seq, seq));
3023}
3024
Jan Bluncka03a8a702008-02-14 19:38:32 -08003025/**
3026 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003027 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003028 * @buf: buffer to return value in
3029 * @buflen: buffer length
3030 *
3031 * Convert a dentry into an ASCII path name. If the entry has been deleted
3032 * the string " (deleted)" is appended. Note that this is ambiguous.
3033 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003034 * Returns a pointer into the buffer or an error code if the path was
3035 * too long. Note: Callers should use the returned pointer, not the passed
3036 * in buffer, to use the name! The implementation often starts at an offset
3037 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003038 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003039 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003040 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003041char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003043 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003044 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003045 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003047 /*
3048 * We have various synthetic filesystems that never get mounted. On
3049 * these filesystems dentries are never used for lookup purposes, and
3050 * thus don't need to be hashed. They also don't need a name until a
3051 * user wants to identify the object in /proc/pid/fd/. The little hack
3052 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003053 *
3054 * Some pseudo inodes are mountable. When they are mounted
3055 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3056 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003057 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003058 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3059 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003060 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003061
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003062 rcu_read_lock();
3063 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003064 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003065 rcu_read_unlock();
3066
Al Viro02125a82011-12-05 08:43:34 -05003067 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003068 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 return res;
3070}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003071EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072
3073/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003074 * Helper function for dentry_operations.d_dname() members
3075 */
3076char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3077 const char *fmt, ...)
3078{
3079 va_list args;
3080 char temp[64];
3081 int sz;
3082
3083 va_start(args, fmt);
3084 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3085 va_end(args);
3086
3087 if (sz > sizeof(temp) || sz > buflen)
3088 return ERR_PTR(-ENAMETOOLONG);
3089
3090 buffer += buflen - sz;
3091 return memcpy(buffer, temp, sz);
3092}
3093
Al Viro118b2302013-08-24 12:08:17 -04003094char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3095{
3096 char *end = buffer + buflen;
3097 /* these dentries are never renamed, so d_lock is not needed */
3098 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003099 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003100 prepend(&end, &buflen, "/", 1))
3101 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003102 return end;
Al Viro118b2302013-08-24 12:08:17 -04003103}
David Herrmann31bbe162014-01-03 14:09:47 +01003104EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003105
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003106/*
Ram Pai6092d042008-03-27 13:06:20 +01003107 * Write full pathname from the root of the filesystem into the buffer.
3108 */
Al Virof6500802014-01-26 12:37:55 -05003109static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003110{
Al Virof6500802014-01-26 12:37:55 -05003111 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003112 char *end, *retval;
3113 int len, seq = 0;
3114 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003115
Al Virof6500802014-01-26 12:37:55 -05003116 if (buflen < 2)
3117 goto Elong;
3118
Al Viro48f5ec22013-09-09 15:22:25 -04003119 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003120restart:
Al Virof6500802014-01-26 12:37:55 -05003121 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003122 end = buf + buflen;
3123 len = buflen;
3124 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003125 /* Get '/' right */
3126 retval = end-1;
3127 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003128 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003129 while (!IS_ROOT(dentry)) {
3130 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003131
Ram Pai6092d042008-03-27 13:06:20 +01003132 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003133 error = prepend_name(&end, &len, &dentry->d_name);
3134 if (error)
3135 break;
Ram Pai6092d042008-03-27 13:06:20 +01003136
3137 retval = end;
3138 dentry = parent;
3139 }
Al Viro48f5ec22013-09-09 15:22:25 -04003140 if (!(seq & 1))
3141 rcu_read_unlock();
3142 if (need_seqretry(&rename_lock, seq)) {
3143 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003144 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003145 }
3146 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003147 if (error)
3148 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003149 return retval;
3150Elong:
3151 return ERR_PTR(-ENAMETOOLONG);
3152}
Nick Pigginec2447c2011-01-07 17:49:29 +11003153
3154char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3155{
Waiman Long232d2d62013-09-09 12:18:13 -04003156 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003157}
3158EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003159
3160char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3161{
3162 char *p = NULL;
3163 char *retval;
3164
Al Viroc1031352010-06-06 22:31:14 -04003165 if (d_unlinked(dentry)) {
3166 p = buf + buflen;
3167 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3168 goto Elong;
3169 buflen++;
3170 }
3171 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003172 if (!IS_ERR(retval) && p)
3173 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003174 return retval;
3175Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003176 return ERR_PTR(-ENAMETOOLONG);
3177}
3178
Linus Torvalds8b19e342013-09-12 10:35:47 -07003179static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3180 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003181{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003182 unsigned seq;
3183
3184 do {
3185 seq = read_seqcount_begin(&fs->seq);
3186 *root = fs->root;
3187 *pwd = fs->pwd;
3188 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003189}
3190
Ram Pai6092d042008-03-27 13:06:20 +01003191/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 * NOTE! The user-level library version returns a
3193 * character pointer. The kernel system call just
3194 * returns the length of the buffer filled (which
3195 * includes the ending '\0' character), or a negative
3196 * error value. So libc would do something like
3197 *
3198 * char *getcwd(char * buf, size_t size)
3199 * {
3200 * int retval;
3201 *
3202 * retval = sys_getcwd(buf, size);
3203 * if (retval >= 0)
3204 * return buf;
3205 * errno = -retval;
3206 * return NULL;
3207 * }
3208 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003209SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210{
Linus Torvalds552ce542007-02-13 12:08:18 -08003211 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003212 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003213 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214
3215 if (!page)
3216 return -ENOMEM;
3217
Linus Torvalds8b19e342013-09-12 10:35:47 -07003218 rcu_read_lock();
3219 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220
Linus Torvalds552ce542007-02-13 12:08:18 -08003221 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003222 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003223 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003224 char *cwd = page + PATH_MAX;
3225 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003226
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003227 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003228 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003229 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003230
Al Viro02125a82011-12-05 08:43:34 -05003231 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003232 goto out;
3233
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003234 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003235 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003236 error = prepend_unreachable(&cwd, &buflen);
3237 if (error)
3238 goto out;
3239 }
3240
Linus Torvalds552ce542007-02-13 12:08:18 -08003241 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003242 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003243 if (len <= size) {
3244 error = len;
3245 if (copy_to_user(buf, cwd, len))
3246 error = -EFAULT;
3247 }
Nick Piggin949854d2011-01-07 17:49:37 +11003248 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003249 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251
3252out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003253 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254 return error;
3255}
3256
3257/*
3258 * Test whether new_dentry is a subdirectory of old_dentry.
3259 *
3260 * Trivially implemented using the dcache structure
3261 */
3262
3263/**
3264 * is_subdir - is new dentry a subdirectory of old_dentry
3265 * @new_dentry: new dentry
3266 * @old_dentry: old dentry
3267 *
3268 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3269 * Returns 0 otherwise.
3270 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3271 */
3272
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003273int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274{
3275 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11003276 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003277
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003278 if (new_dentry == old_dentry)
3279 return 1;
3280
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003281 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003284 /*
3285 * Need rcu_readlock to protect against the d_parent trashing
3286 * due to d_move
3287 */
3288 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003289 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003290 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003291 else
3292 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11003293 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295
3296 return result;
3297}
3298
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003299static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003301 struct dentry *root = data;
3302 if (dentry != root) {
3303 if (d_unhashed(dentry) || !dentry->d_inode)
3304 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003306 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3307 dentry->d_flags |= DCACHE_GENOCIDE;
3308 dentry->d_lockref.count--;
3309 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003311 return D_WALK_CONTINUE;
3312}
Nick Piggin58db63d2011-01-07 17:49:39 +11003313
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003314void d_genocide(struct dentry *parent)
3315{
3316 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317}
3318
Al Viro60545d02013-06-07 01:20:27 -04003319void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320{
Al Viro60545d02013-06-07 01:20:27 -04003321 inode_dec_link_count(inode);
3322 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3323 !hlist_unhashed(&dentry->d_alias) ||
3324 !d_unlinked(dentry));
3325 spin_lock(&dentry->d_parent->d_lock);
3326 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3327 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3328 (unsigned long long)inode->i_ino);
3329 spin_unlock(&dentry->d_lock);
3330 spin_unlock(&dentry->d_parent->d_lock);
3331 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332}
Al Viro60545d02013-06-07 01:20:27 -04003333EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334
3335static __initdata unsigned long dhash_entries;
3336static int __init set_dhash_entries(char *str)
3337{
3338 if (!str)
3339 return 0;
3340 dhash_entries = simple_strtoul(str, &str, 0);
3341 return 1;
3342}
3343__setup("dhash_entries=", set_dhash_entries);
3344
3345static void __init dcache_init_early(void)
3346{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003347 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348
3349 /* If hashes are distributed across NUMA nodes, defer
3350 * hash allocation until vmalloc space is available.
3351 */
3352 if (hashdist)
3353 return;
3354
3355 dentry_hashtable =
3356 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003357 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358 dhash_entries,
3359 13,
3360 HASH_EARLY,
3361 &d_hash_shift,
3362 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003363 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 0);
3365
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003366 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003367 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368}
3369
Denis Cheng74bf17c2007-10-16 23:26:30 -07003370static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003372 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003373
3374 /*
3375 * A constructor could be added for stable state like the lists,
3376 * but it is probably not worth it because of the cache nature
3377 * of the dcache.
3378 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003379 dentry_cache = KMEM_CACHE(dentry,
3380 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381
3382 /* Hash may have been set up in dcache_init_early */
3383 if (!hashdist)
3384 return;
3385
3386 dentry_hashtable =
3387 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003388 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389 dhash_entries,
3390 13,
3391 0,
3392 &d_hash_shift,
3393 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003394 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395 0);
3396
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003397 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003398 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399}
3400
3401/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003402struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003403EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003404
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405EXPORT_SYMBOL(d_genocide);
3406
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407void __init vfs_caches_init_early(void)
3408{
3409 dcache_init_early();
3410 inode_init_early();
3411}
3412
3413void __init vfs_caches_init(unsigned long mempages)
3414{
3415 unsigned long reserve;
3416
3417 /* Base hash sizes on available memory, with a reserve equal to
3418 150% of current kernel size */
3419
3420 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3421 mempages -= reserve;
3422
3423 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003424 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003425
Denis Cheng74bf17c2007-10-16 23:26:30 -07003426 dcache_init();
3427 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003429 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430 bdev_cache_init();
3431 chrdev_init();
3432}