]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | ||
1da177e4 LT |
17 | #include <linux/syscalls.h> |
18 | #include <linux/string.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/fs.h> | |
7a91bf7f | 21 | #include <linux/fsnotify.h> |
1da177e4 LT |
22 | #include <linux/slab.h> |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/hash.h> |
25 | #include <linux/cache.h> | |
630d9c47 | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #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> | |
5ad4e53b | 34 | #include <linux/fs_struct.h> |
613afbf8 | 35 | #include <linux/hardirq.h> |
ceb5bdc2 NP |
36 | #include <linux/bit_spinlock.h> |
37 | #include <linux/rculist_bl.h> | |
268bb0ce | 38 | #include <linux/prefetch.h> |
dd179946 | 39 | #include <linux/ratelimit.h> |
f6041567 | 40 | #include <linux/list_lru.h> |
07f3f05c | 41 | #include "internal.h" |
b2dba1af | 42 | #include "mount.h" |
1da177e4 | 43 | |
789680d1 NP |
44 | /* |
45 | * Usage: | |
873feea0 NP |
46 | * dcache->d_inode->i_lock protects: |
47 | * - i_dentry, d_alias, d_inode of aliases | |
ceb5bdc2 NP |
48 | * 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) | |
19156840 | 52 | * dentry->d_sb->s_dentry_lru_lock protects: |
23044507 NP |
53 | * - the dcache lru lists and counters |
54 | * d_lock protects: | |
55 | * - d_flags | |
56 | * - d_name | |
57 | * - d_lru | |
b7ab39f6 | 58 | * - d_count |
da502956 | 59 | * - d_unhashed() |
2fd6b7f5 NP |
60 | * - d_parent and d_subdirs |
61 | * - childrens' d_child and d_parent | |
b23fb0a6 | 62 | * - d_alias, d_inode |
789680d1 NP |
63 | * |
64 | * Ordering: | |
873feea0 | 65 | * dentry->d_inode->i_lock |
b5c84bf6 | 66 | * dentry->d_lock |
19156840 | 67 | * dentry->d_sb->s_dentry_lru_lock |
ceb5bdc2 NP |
68 | * dcache_hash_bucket lock |
69 | * s_anon lock | |
789680d1 | 70 | * |
da502956 NP |
71 | * 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: | |
789680d1 NP |
78 | * if (dentry1 < dentry2) |
79 | * dentry1->d_lock | |
80 | * dentry2->d_lock | |
81 | */ | |
fa3536cc | 82 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
83 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
84 | ||
74c3cbe3 | 85 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 86 | |
949854d0 | 87 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 88 | |
e18b890b | 89 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 90 | |
1da177e4 LT |
91 | /* |
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 | */ | |
1da177e4 | 99 | |
fa3536cc ED |
100 | static unsigned int d_hash_mask __read_mostly; |
101 | static unsigned int d_hash_shift __read_mostly; | |
ceb5bdc2 | 102 | |
b07ad996 | 103 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 104 | |
8966be90 | 105 | static inline struct hlist_bl_head *d_hash(const struct dentry *parent, |
6d7d1a0d | 106 | unsigned int hash) |
ceb5bdc2 | 107 | { |
6d7d1a0d | 108 | hash += (unsigned long) parent / L1_CACHE_BYTES; |
99d263d4 | 109 | return dentry_hashtable + hash_32(hash, d_hash_shift); |
ceb5bdc2 NP |
110 | } |
111 | ||
1da177e4 LT |
112 | /* Statistics gathering. */ |
113 | struct dentry_stat_t dentry_stat = { | |
114 | .age_limit = 45, | |
115 | }; | |
116 | ||
3942c07c | 117 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 118 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
312d3ca8 CH |
119 | |
120 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
62d36c77 DC |
121 | |
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 | */ | |
3942c07c | 134 | static long get_nr_dentry(void) |
3e880fb5 NP |
135 | { |
136 | int i; | |
3942c07c | 137 | long sum = 0; |
3e880fb5 NP |
138 | for_each_possible_cpu(i) |
139 | sum += per_cpu(nr_dentry, i); | |
140 | return sum < 0 ? 0 : sum; | |
141 | } | |
142 | ||
62d36c77 DC |
143 | static 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 | ||
1f7e0616 | 152 | int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer, |
312d3ca8 CH |
153 | size_t *lenp, loff_t *ppos) |
154 | { | |
3e880fb5 | 155 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 156 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
3942c07c | 157 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 CH |
158 | } |
159 | #endif | |
160 | ||
5483f18e LT |
161 | /* |
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 | */ | |
e419b4cc LT |
165 | #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 | */ | |
94753db5 | 177 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 178 | { |
bfcfaa77 | 179 | unsigned long a,b,mask; |
bfcfaa77 LT |
180 | |
181 | for (;;) { | |
12f8ad4b | 182 | a = *(unsigned long *)cs; |
e419b4cc | 183 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
184 | 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 | } | |
a5c21dce | 194 | mask = bytemask_from_count(tcount); |
bfcfaa77 | 195 | return unlikely(!!((a ^ b) & mask)); |
e419b4cc LT |
196 | } |
197 | ||
bfcfaa77 | 198 | #else |
e419b4cc | 199 | |
94753db5 | 200 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 201 | { |
5483f18e LT |
202 | do { |
203 | if (*cs != *ct) | |
204 | return 1; | |
205 | cs++; | |
206 | ct++; | |
207 | tcount--; | |
208 | } while (tcount); | |
209 | return 0; | |
210 | } | |
211 | ||
e419b4cc LT |
212 | #endif |
213 | ||
94753db5 LT |
214 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
215 | { | |
6326c71f | 216 | const unsigned char *cs; |
94753db5 LT |
217 | /* |
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 | */ | |
6326c71f LT |
233 | cs = ACCESS_ONCE(dentry->d_name.name); |
234 | smp_read_barrier_depends(); | |
235 | return dentry_string_cmp(cs, ct, tcount); | |
94753db5 LT |
236 | } |
237 | ||
8d85b484 AV |
238 | struct external_name { |
239 | union { | |
240 | atomic_t count; | |
241 | struct rcu_head head; | |
242 | } u; | |
243 | unsigned char name[]; | |
244 | }; | |
245 | ||
246 | static inline struct external_name *external_name(struct dentry *dentry) | |
247 | { | |
248 | return container_of(dentry->d_name.name, struct external_name, name[0]); | |
249 | } | |
250 | ||
9c82ab9c | 251 | static void __d_free(struct rcu_head *head) |
1da177e4 | 252 | { |
9c82ab9c CH |
253 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
254 | ||
b3d9b7a3 | 255 | WARN_ON(!hlist_unhashed(&dentry->d_alias)); |
8d85b484 AV |
256 | kmem_cache_free(dentry_cache, dentry); |
257 | } | |
258 | ||
259 | static void __d_free_external(struct rcu_head *head) | |
260 | { | |
261 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
262 | WARN_ON(!hlist_unhashed(&dentry->d_alias)); | |
263 | kfree(external_name(dentry)); | |
1da177e4 LT |
264 | kmem_cache_free(dentry_cache, dentry); |
265 | } | |
266 | ||
b4f0354e AV |
267 | static void dentry_free(struct dentry *dentry) |
268 | { | |
8d85b484 AV |
269 | if (unlikely(dname_external(dentry))) { |
270 | struct external_name *p = external_name(dentry); | |
271 | if (likely(atomic_dec_and_test(&p->u.count))) { | |
272 | call_rcu(&dentry->d_u.d_rcu, __d_free_external); | |
273 | return; | |
274 | } | |
275 | } | |
b4f0354e AV |
276 | /* if dentry was never visible to RCU, immediate free is OK */ |
277 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
278 | __d_free(&dentry->d_u.d_rcu); | |
279 | else | |
280 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
281 | } | |
282 | ||
31e6b01f NP |
283 | /** |
284 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
ff5fdb61 | 285 | * @dentry: the target dentry |
31e6b01f NP |
286 | * After this call, in-progress rcu-walk path lookup will fail. This |
287 | * should be called after unhashing, and after changing d_inode (if | |
288 | * the dentry has not already been unhashed). | |
289 | */ | |
290 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
291 | { | |
292 | assert_spin_locked(&dentry->d_lock); | |
293 | /* Go through a barrier */ | |
294 | write_seqcount_barrier(&dentry->d_seq); | |
295 | } | |
296 | ||
1da177e4 LT |
297 | /* |
298 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
299 | * d_iput() operation if defined. Dentry has no refcount |
300 | * and is unhashed. | |
1da177e4 | 301 | */ |
858119e1 | 302 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 303 | __releases(dentry->d_lock) |
873feea0 | 304 | __releases(dentry->d_inode->i_lock) |
1da177e4 LT |
305 | { |
306 | struct inode *inode = dentry->d_inode; | |
307 | if (inode) { | |
308 | dentry->d_inode = NULL; | |
b3d9b7a3 | 309 | hlist_del_init(&dentry->d_alias); |
1da177e4 | 310 | spin_unlock(&dentry->d_lock); |
873feea0 | 311 | spin_unlock(&inode->i_lock); |
f805fbda LT |
312 | if (!inode->i_nlink) |
313 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
314 | if (dentry->d_op && dentry->d_op->d_iput) |
315 | dentry->d_op->d_iput(dentry, inode); | |
316 | else | |
317 | iput(inode); | |
318 | } else { | |
319 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
320 | } |
321 | } | |
322 | ||
31e6b01f NP |
323 | /* |
324 | * Release the dentry's inode, using the filesystem | |
325 | * d_iput() operation if defined. dentry remains in-use. | |
326 | */ | |
327 | static void dentry_unlink_inode(struct dentry * dentry) | |
328 | __releases(dentry->d_lock) | |
873feea0 | 329 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
330 | { |
331 | struct inode *inode = dentry->d_inode; | |
b18825a7 | 332 | __d_clear_type(dentry); |
31e6b01f | 333 | dentry->d_inode = NULL; |
b3d9b7a3 | 334 | hlist_del_init(&dentry->d_alias); |
31e6b01f NP |
335 | dentry_rcuwalk_barrier(dentry); |
336 | spin_unlock(&dentry->d_lock); | |
873feea0 | 337 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
338 | if (!inode->i_nlink) |
339 | fsnotify_inoderemove(inode); | |
340 | if (dentry->d_op && dentry->d_op->d_iput) | |
341 | dentry->d_op->d_iput(dentry, inode); | |
342 | else | |
343 | iput(inode); | |
344 | } | |
345 | ||
89dc77bc LT |
346 | /* |
347 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
348 | * is in use - which includes both the "real" per-superblock | |
349 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
350 | * | |
351 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
352 | * on the shrink list (ie not on the superblock LRU list). | |
353 | * | |
354 | * The per-cpu "nr_dentry_unused" counters are updated with | |
355 | * the DCACHE_LRU_LIST bit. | |
356 | * | |
357 | * These helper functions make sure we always follow the | |
358 | * rules. d_lock must be held by the caller. | |
359 | */ | |
360 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
361 | static void d_lru_add(struct dentry *dentry) | |
362 | { | |
363 | D_FLAG_VERIFY(dentry, 0); | |
364 | dentry->d_flags |= DCACHE_LRU_LIST; | |
365 | this_cpu_inc(nr_dentry_unused); | |
366 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
367 | } | |
368 | ||
369 | static void d_lru_del(struct dentry *dentry) | |
370 | { | |
371 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
372 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
373 | this_cpu_dec(nr_dentry_unused); | |
374 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
375 | } | |
376 | ||
377 | static void d_shrink_del(struct dentry *dentry) | |
378 | { | |
379 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
380 | list_del_init(&dentry->d_lru); | |
381 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
382 | this_cpu_dec(nr_dentry_unused); | |
383 | } | |
384 | ||
385 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
386 | { | |
387 | D_FLAG_VERIFY(dentry, 0); | |
388 | list_add(&dentry->d_lru, list); | |
389 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
390 | this_cpu_inc(nr_dentry_unused); | |
391 | } | |
392 | ||
393 | /* | |
394 | * These can only be called under the global LRU lock, ie during the | |
395 | * callback for freeing the LRU list. "isolate" removes it from the | |
396 | * LRU lists entirely, while shrink_move moves it to the indicated | |
397 | * private list. | |
398 | */ | |
399 | static void d_lru_isolate(struct dentry *dentry) | |
400 | { | |
401 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
402 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
403 | this_cpu_dec(nr_dentry_unused); | |
404 | list_del_init(&dentry->d_lru); | |
405 | } | |
406 | ||
407 | static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list) | |
408 | { | |
409 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
410 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
411 | list_move_tail(&dentry->d_lru, list); | |
412 | } | |
413 | ||
da3bbdd4 | 414 | /* |
f6041567 | 415 | * dentry_lru_(add|del)_list) must be called with d_lock held. |
da3bbdd4 KM |
416 | */ |
417 | static void dentry_lru_add(struct dentry *dentry) | |
418 | { | |
89dc77bc LT |
419 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) |
420 | d_lru_add(dentry); | |
da3bbdd4 KM |
421 | } |
422 | ||
789680d1 NP |
423 | /** |
424 | * d_drop - drop a dentry | |
425 | * @dentry: dentry to drop | |
426 | * | |
427 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
428 | * be found through a VFS lookup any more. Note that this is different from | |
429 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
430 | * possible, giving a successful _negative_ lookup, while d_drop will | |
431 | * just make the cache lookup fail. | |
432 | * | |
433 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
434 | * reason (NFS timeouts or autofs deletes). | |
435 | * | |
436 | * __d_drop requires dentry->d_lock. | |
437 | */ | |
438 | void __d_drop(struct dentry *dentry) | |
439 | { | |
dea3667b | 440 | if (!d_unhashed(dentry)) { |
b61625d2 | 441 | struct hlist_bl_head *b; |
7632e465 BF |
442 | /* |
443 | * Hashed dentries are normally on the dentry hashtable, | |
444 | * with the exception of those newly allocated by | |
445 | * d_obtain_alias, which are always IS_ROOT: | |
446 | */ | |
447 | if (unlikely(IS_ROOT(dentry))) | |
b61625d2 AV |
448 | b = &dentry->d_sb->s_anon; |
449 | else | |
450 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
451 | ||
452 | hlist_bl_lock(b); | |
453 | __hlist_bl_del(&dentry->d_hash); | |
454 | dentry->d_hash.pprev = NULL; | |
455 | hlist_bl_unlock(b); | |
dea3667b | 456 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
457 | } |
458 | } | |
459 | EXPORT_SYMBOL(__d_drop); | |
460 | ||
461 | void d_drop(struct dentry *dentry) | |
462 | { | |
789680d1 NP |
463 | spin_lock(&dentry->d_lock); |
464 | __d_drop(dentry); | |
465 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
466 | } |
467 | EXPORT_SYMBOL(d_drop); | |
468 | ||
e55fd011 | 469 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 470 | { |
41edf278 AV |
471 | struct dentry *parent = NULL; |
472 | bool can_free = true; | |
41edf278 | 473 | if (!IS_ROOT(dentry)) |
77812a1e | 474 | parent = dentry->d_parent; |
31e6b01f | 475 | |
0d98439e LT |
476 | /* |
477 | * The dentry is now unrecoverably dead to the world. | |
478 | */ | |
479 | lockref_mark_dead(&dentry->d_lockref); | |
480 | ||
f0023bc6 | 481 | /* |
f0023bc6 SW |
482 | * inform the fs via d_prune that this dentry is about to be |
483 | * unhashed and destroyed. | |
484 | */ | |
590fb51f | 485 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry)) |
61572bb1 YZ |
486 | dentry->d_op->d_prune(dentry); |
487 | ||
01b60351 AV |
488 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
489 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
490 | d_lru_del(dentry); | |
01b60351 | 491 | } |
77812a1e NP |
492 | /* if it was on the hash then remove it */ |
493 | __d_drop(dentry); | |
03b3b889 AV |
494 | list_del(&dentry->d_u.d_child); |
495 | /* | |
496 | * Inform d_walk() that we are no longer attached to the | |
497 | * dentry tree | |
498 | */ | |
499 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
500 | if (parent) | |
501 | spin_unlock(&parent->d_lock); | |
502 | dentry_iput(dentry); | |
503 | /* | |
504 | * dentry_iput drops the locks, at which point nobody (except | |
505 | * transient RCU lookups) can reach this dentry. | |
506 | */ | |
507 | BUG_ON((int)dentry->d_lockref.count > 0); | |
508 | this_cpu_dec(nr_dentry); | |
509 | if (dentry->d_op && dentry->d_op->d_release) | |
510 | dentry->d_op->d_release(dentry); | |
511 | ||
41edf278 AV |
512 | spin_lock(&dentry->d_lock); |
513 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
514 | dentry->d_flags |= DCACHE_MAY_FREE; | |
515 | can_free = false; | |
516 | } | |
517 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
518 | if (likely(can_free)) |
519 | dentry_free(dentry); | |
e55fd011 AV |
520 | } |
521 | ||
522 | /* | |
523 | * Finish off a dentry we've decided to kill. | |
524 | * dentry->d_lock must be held, returns with it unlocked. | |
525 | * If ref is non-zero, then decrement the refcount too. | |
526 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
527 | */ | |
8cbf74da | 528 | static struct dentry *dentry_kill(struct dentry *dentry) |
e55fd011 AV |
529 | __releases(dentry->d_lock) |
530 | { | |
531 | struct inode *inode = dentry->d_inode; | |
532 | struct dentry *parent = NULL; | |
533 | ||
534 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
535 | goto failed; | |
536 | ||
537 | if (!IS_ROOT(dentry)) { | |
538 | parent = dentry->d_parent; | |
539 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
540 | if (inode) | |
541 | spin_unlock(&inode->i_lock); | |
542 | goto failed; | |
543 | } | |
544 | } | |
545 | ||
546 | __dentry_kill(dentry); | |
03b3b889 | 547 | return parent; |
e55fd011 AV |
548 | |
549 | failed: | |
8cbf74da AV |
550 | spin_unlock(&dentry->d_lock); |
551 | cpu_relax(); | |
e55fd011 | 552 | return dentry; /* try again with same dentry */ |
77812a1e NP |
553 | } |
554 | ||
046b961b AV |
555 | static inline struct dentry *lock_parent(struct dentry *dentry) |
556 | { | |
557 | struct dentry *parent = dentry->d_parent; | |
558 | if (IS_ROOT(dentry)) | |
559 | return NULL; | |
c2338f2d AV |
560 | if (unlikely((int)dentry->d_lockref.count < 0)) |
561 | return NULL; | |
046b961b AV |
562 | if (likely(spin_trylock(&parent->d_lock))) |
563 | return parent; | |
046b961b | 564 | rcu_read_lock(); |
c2338f2d | 565 | spin_unlock(&dentry->d_lock); |
046b961b AV |
566 | again: |
567 | parent = ACCESS_ONCE(dentry->d_parent); | |
568 | spin_lock(&parent->d_lock); | |
569 | /* | |
570 | * We can't blindly lock dentry until we are sure | |
571 | * that we won't violate the locking order. | |
572 | * Any changes of dentry->d_parent must have | |
573 | * been done with parent->d_lock held, so | |
574 | * spin_lock() above is enough of a barrier | |
575 | * for checking if it's still our child. | |
576 | */ | |
577 | if (unlikely(parent != dentry->d_parent)) { | |
578 | spin_unlock(&parent->d_lock); | |
579 | goto again; | |
580 | } | |
581 | rcu_read_unlock(); | |
582 | if (parent != dentry) | |
9f12600f | 583 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
046b961b AV |
584 | else |
585 | parent = NULL; | |
586 | return parent; | |
587 | } | |
588 | ||
1da177e4 LT |
589 | /* |
590 | * This is dput | |
591 | * | |
592 | * This is complicated by the fact that we do not want to put | |
593 | * dentries that are no longer on any hash chain on the unused | |
594 | * list: we'd much rather just get rid of them immediately. | |
595 | * | |
596 | * However, that implies that we have to traverse the dentry | |
597 | * tree upwards to the parents which might _also_ now be | |
598 | * scheduled for deletion (it may have been only waiting for | |
599 | * its last child to go away). | |
600 | * | |
601 | * This tail recursion is done by hand as we don't want to depend | |
602 | * on the compiler to always get this right (gcc generally doesn't). | |
603 | * Real recursion would eat up our stack space. | |
604 | */ | |
605 | ||
606 | /* | |
607 | * dput - release a dentry | |
608 | * @dentry: dentry to release | |
609 | * | |
610 | * Release a dentry. This will drop the usage count and if appropriate | |
611 | * call the dentry unlink method as well as removing it from the queues and | |
612 | * releasing its resources. If the parent dentries were scheduled for release | |
613 | * they too may now get deleted. | |
1da177e4 | 614 | */ |
1da177e4 LT |
615 | void dput(struct dentry *dentry) |
616 | { | |
8aab6a27 | 617 | if (unlikely(!dentry)) |
1da177e4 LT |
618 | return; |
619 | ||
620 | repeat: | |
98474236 | 621 | if (lockref_put_or_lock(&dentry->d_lockref)) |
1da177e4 | 622 | return; |
1da177e4 | 623 | |
8aab6a27 LT |
624 | /* Unreachable? Get rid of it */ |
625 | if (unlikely(d_unhashed(dentry))) | |
626 | goto kill_it; | |
627 | ||
628 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
1da177e4 | 629 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 630 | goto kill_it; |
1da177e4 | 631 | } |
265ac902 | 632 | |
358eec18 LT |
633 | if (!(dentry->d_flags & DCACHE_REFERENCED)) |
634 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 635 | dentry_lru_add(dentry); |
265ac902 | 636 | |
98474236 | 637 | dentry->d_lockref.count--; |
61f3dee4 | 638 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
639 | return; |
640 | ||
d52b9086 | 641 | kill_it: |
8cbf74da | 642 | dentry = dentry_kill(dentry); |
d52b9086 MS |
643 | if (dentry) |
644 | goto repeat; | |
1da177e4 | 645 | } |
ec4f8605 | 646 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
647 | |
648 | /** | |
649 | * d_invalidate - invalidate a dentry | |
650 | * @dentry: dentry to invalidate | |
651 | * | |
652 | * Try to invalidate the dentry if it turns out to be | |
653 | * possible. If there are other dentries that can be | |
654 | * reached through this one we can't delete it and we | |
655 | * return -EBUSY. On success we return 0. | |
656 | * | |
657 | * no dcache lock. | |
658 | */ | |
659 | ||
660 | int d_invalidate(struct dentry * dentry) | |
661 | { | |
662 | /* | |
663 | * If it's already been dropped, return OK. | |
664 | */ | |
da502956 | 665 | spin_lock(&dentry->d_lock); |
1da177e4 | 666 | if (d_unhashed(dentry)) { |
da502956 | 667 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
668 | return 0; |
669 | } | |
670 | /* | |
671 | * Check whether to do a partial shrink_dcache | |
672 | * to get rid of unused child entries. | |
673 | */ | |
674 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 675 | spin_unlock(&dentry->d_lock); |
1da177e4 | 676 | shrink_dcache_parent(dentry); |
da502956 | 677 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
678 | } |
679 | ||
680 | /* | |
681 | * Somebody else still using it? | |
682 | * | |
683 | * If it's a directory, we can't drop it | |
684 | * for fear of somebody re-populating it | |
685 | * with children (even though dropping it | |
686 | * would make it unreachable from the root, | |
687 | * we might still populate it if it was a | |
688 | * working directory or similar). | |
50e69630 AV |
689 | * We also need to leave mountpoints alone, |
690 | * directory or not. | |
1da177e4 | 691 | */ |
98474236 | 692 | if (dentry->d_lockref.count > 1 && dentry->d_inode) { |
50e69630 | 693 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { |
1da177e4 | 694 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
695 | return -EBUSY; |
696 | } | |
697 | } | |
698 | ||
699 | __d_drop(dentry); | |
700 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
701 | return 0; |
702 | } | |
ec4f8605 | 703 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 704 | |
b5c84bf6 | 705 | /* This must be called with d_lock held */ |
dc0474be | 706 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 707 | { |
98474236 | 708 | dentry->d_lockref.count++; |
23044507 NP |
709 | } |
710 | ||
dc0474be | 711 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 712 | { |
98474236 | 713 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
714 | } |
715 | ||
b7ab39f6 NP |
716 | struct dentry *dget_parent(struct dentry *dentry) |
717 | { | |
df3d0bbc | 718 | int gotref; |
b7ab39f6 NP |
719 | struct dentry *ret; |
720 | ||
df3d0bbc WL |
721 | /* |
722 | * Do optimistic parent lookup without any | |
723 | * locking. | |
724 | */ | |
725 | rcu_read_lock(); | |
726 | ret = ACCESS_ONCE(dentry->d_parent); | |
727 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
728 | rcu_read_unlock(); | |
729 | if (likely(gotref)) { | |
730 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
731 | return ret; | |
732 | dput(ret); | |
733 | } | |
734 | ||
b7ab39f6 | 735 | repeat: |
a734eb45 NP |
736 | /* |
737 | * Don't need rcu_dereference because we re-check it was correct under | |
738 | * the lock. | |
739 | */ | |
740 | rcu_read_lock(); | |
b7ab39f6 | 741 | ret = dentry->d_parent; |
a734eb45 NP |
742 | spin_lock(&ret->d_lock); |
743 | if (unlikely(ret != dentry->d_parent)) { | |
744 | spin_unlock(&ret->d_lock); | |
745 | rcu_read_unlock(); | |
b7ab39f6 NP |
746 | goto repeat; |
747 | } | |
a734eb45 | 748 | rcu_read_unlock(); |
98474236 WL |
749 | BUG_ON(!ret->d_lockref.count); |
750 | ret->d_lockref.count++; | |
b7ab39f6 | 751 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
752 | return ret; |
753 | } | |
754 | EXPORT_SYMBOL(dget_parent); | |
755 | ||
1da177e4 LT |
756 | /** |
757 | * d_find_alias - grab a hashed alias of inode | |
758 | * @inode: inode in question | |
1da177e4 LT |
759 | * |
760 | * If inode has a hashed alias, or is a directory and has any alias, | |
761 | * acquire the reference to alias and return it. Otherwise return NULL. | |
762 | * Notice that if inode is a directory there can be only one alias and | |
763 | * it can be unhashed only if it has no children, or if it is the root | |
764 | * of a filesystem. | |
765 | * | |
21c0d8fd | 766 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
52ed46f0 | 767 | * any other hashed alias over that one. |
1da177e4 | 768 | */ |
52ed46f0 | 769 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 770 | { |
da502956 | 771 | struct dentry *alias, *discon_alias; |
1da177e4 | 772 | |
da502956 NP |
773 | again: |
774 | discon_alias = NULL; | |
b67bfe0d | 775 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
da502956 | 776 | spin_lock(&alias->d_lock); |
1da177e4 | 777 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 778 | if (IS_ROOT(alias) && |
da502956 | 779 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 780 | discon_alias = alias; |
52ed46f0 | 781 | } else { |
dc0474be | 782 | __dget_dlock(alias); |
da502956 NP |
783 | spin_unlock(&alias->d_lock); |
784 | return alias; | |
785 | } | |
786 | } | |
787 | spin_unlock(&alias->d_lock); | |
788 | } | |
789 | if (discon_alias) { | |
790 | alias = discon_alias; | |
791 | spin_lock(&alias->d_lock); | |
792 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
8d80d7da BF |
793 | __dget_dlock(alias); |
794 | spin_unlock(&alias->d_lock); | |
795 | return alias; | |
1da177e4 | 796 | } |
da502956 NP |
797 | spin_unlock(&alias->d_lock); |
798 | goto again; | |
1da177e4 | 799 | } |
da502956 | 800 | return NULL; |
1da177e4 LT |
801 | } |
802 | ||
da502956 | 803 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 804 | { |
214fda1f DH |
805 | struct dentry *de = NULL; |
806 | ||
b3d9b7a3 | 807 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 808 | spin_lock(&inode->i_lock); |
52ed46f0 | 809 | de = __d_find_alias(inode); |
873feea0 | 810 | spin_unlock(&inode->i_lock); |
214fda1f | 811 | } |
1da177e4 LT |
812 | return de; |
813 | } | |
ec4f8605 | 814 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
815 | |
816 | /* | |
817 | * Try to kill dentries associated with this inode. | |
818 | * WARNING: you must own a reference to inode. | |
819 | */ | |
820 | void d_prune_aliases(struct inode *inode) | |
821 | { | |
0cdca3f9 | 822 | struct dentry *dentry; |
1da177e4 | 823 | restart: |
873feea0 | 824 | spin_lock(&inode->i_lock); |
b67bfe0d | 825 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 826 | spin_lock(&dentry->d_lock); |
98474236 | 827 | if (!dentry->d_lockref.count) { |
590fb51f YZ |
828 | /* |
829 | * inform the fs via d_prune that this dentry | |
830 | * is about to be unhashed and destroyed. | |
831 | */ | |
832 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && | |
833 | !d_unhashed(dentry)) | |
834 | dentry->d_op->d_prune(dentry); | |
835 | ||
dc0474be | 836 | __dget_dlock(dentry); |
1da177e4 LT |
837 | __d_drop(dentry); |
838 | spin_unlock(&dentry->d_lock); | |
873feea0 | 839 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
840 | dput(dentry); |
841 | goto restart; | |
842 | } | |
843 | spin_unlock(&dentry->d_lock); | |
844 | } | |
873feea0 | 845 | spin_unlock(&inode->i_lock); |
1da177e4 | 846 | } |
ec4f8605 | 847 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 848 | |
3049cfe2 | 849 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 850 | { |
5c47e6d0 | 851 | struct dentry *dentry, *parent; |
da3bbdd4 | 852 | |
60942f2f | 853 | while (!list_empty(list)) { |
ff2fde99 | 854 | struct inode *inode; |
60942f2f | 855 | dentry = list_entry(list->prev, struct dentry, d_lru); |
ec33679d | 856 | spin_lock(&dentry->d_lock); |
046b961b AV |
857 | parent = lock_parent(dentry); |
858 | ||
dd1f6b2e DC |
859 | /* |
860 | * The dispose list is isolated and dentries are not accounted | |
861 | * to the LRU here, so we can simply remove it from the list | |
862 | * here regardless of whether it is referenced or not. | |
863 | */ | |
89dc77bc | 864 | d_shrink_del(dentry); |
dd1f6b2e | 865 | |
1da177e4 LT |
866 | /* |
867 | * We found an inuse dentry which was not removed from | |
dd1f6b2e | 868 | * the LRU because of laziness during lookup. Do not free it. |
1da177e4 | 869 | */ |
41edf278 | 870 | if ((int)dentry->d_lockref.count > 0) { |
da3bbdd4 | 871 | spin_unlock(&dentry->d_lock); |
046b961b AV |
872 | if (parent) |
873 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
874 | continue; |
875 | } | |
77812a1e | 876 | |
64fd72e0 AV |
877 | |
878 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) { | |
879 | bool can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
880 | spin_unlock(&dentry->d_lock); | |
046b961b AV |
881 | if (parent) |
882 | spin_unlock(&parent->d_lock); | |
64fd72e0 AV |
883 | if (can_free) |
884 | dentry_free(dentry); | |
885 | continue; | |
886 | } | |
887 | ||
ff2fde99 AV |
888 | inode = dentry->d_inode; |
889 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
89dc77bc | 890 | d_shrink_add(dentry, list); |
dd1f6b2e | 891 | spin_unlock(&dentry->d_lock); |
046b961b AV |
892 | if (parent) |
893 | spin_unlock(&parent->d_lock); | |
5c47e6d0 | 894 | continue; |
dd1f6b2e | 895 | } |
ff2fde99 | 896 | |
ff2fde99 | 897 | __dentry_kill(dentry); |
046b961b | 898 | |
5c47e6d0 AV |
899 | /* |
900 | * We need to prune ancestors too. This is necessary to prevent | |
901 | * quadratic behavior of shrink_dcache_parent(), but is also | |
902 | * expected to be beneficial in reducing dentry cache | |
903 | * fragmentation. | |
904 | */ | |
905 | dentry = parent; | |
b2b80195 AV |
906 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) { |
907 | parent = lock_parent(dentry); | |
908 | if (dentry->d_lockref.count != 1) { | |
909 | dentry->d_lockref.count--; | |
910 | spin_unlock(&dentry->d_lock); | |
911 | if (parent) | |
912 | spin_unlock(&parent->d_lock); | |
913 | break; | |
914 | } | |
915 | inode = dentry->d_inode; /* can't be NULL */ | |
916 | if (unlikely(!spin_trylock(&inode->i_lock))) { | |
917 | spin_unlock(&dentry->d_lock); | |
918 | if (parent) | |
919 | spin_unlock(&parent->d_lock); | |
920 | cpu_relax(); | |
921 | continue; | |
922 | } | |
923 | __dentry_kill(dentry); | |
924 | dentry = parent; | |
925 | } | |
da3bbdd4 | 926 | } |
3049cfe2 CH |
927 | } |
928 | ||
f6041567 DC |
929 | static enum lru_status |
930 | dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg) | |
931 | { | |
932 | struct list_head *freeable = arg; | |
933 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
934 | ||
935 | ||
936 | /* | |
937 | * we are inverting the lru lock/dentry->d_lock here, | |
938 | * so use a trylock. If we fail to get the lock, just skip | |
939 | * it | |
940 | */ | |
941 | if (!spin_trylock(&dentry->d_lock)) | |
942 | return LRU_SKIP; | |
943 | ||
944 | /* | |
945 | * Referenced dentries are still in use. If they have active | |
946 | * counts, just remove them from the LRU. Otherwise give them | |
947 | * another pass through the LRU. | |
948 | */ | |
949 | if (dentry->d_lockref.count) { | |
89dc77bc | 950 | d_lru_isolate(dentry); |
f6041567 DC |
951 | spin_unlock(&dentry->d_lock); |
952 | return LRU_REMOVED; | |
953 | } | |
954 | ||
955 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
956 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
957 | spin_unlock(&dentry->d_lock); | |
958 | ||
959 | /* | |
960 | * The list move itself will be made by the common LRU code. At | |
961 | * this point, we've dropped the dentry->d_lock but keep the | |
962 | * lru lock. This is safe to do, since every list movement is | |
963 | * protected by the lru lock even if both locks are held. | |
964 | * | |
965 | * This is guaranteed by the fact that all LRU management | |
966 | * functions are intermediated by the LRU API calls like | |
967 | * list_lru_add and list_lru_del. List movement in this file | |
968 | * only ever occur through this functions or through callbacks | |
969 | * like this one, that are called from the LRU API. | |
970 | * | |
971 | * The only exceptions to this are functions like | |
972 | * shrink_dentry_list, and code that first checks for the | |
973 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
974 | * operating only with stack provided lists after they are | |
975 | * properly isolated from the main list. It is thus, always a | |
976 | * local access. | |
977 | */ | |
978 | return LRU_ROTATE; | |
979 | } | |
980 | ||
89dc77bc | 981 | d_lru_shrink_move(dentry, freeable); |
f6041567 DC |
982 | spin_unlock(&dentry->d_lock); |
983 | ||
984 | return LRU_REMOVED; | |
985 | } | |
986 | ||
3049cfe2 | 987 | /** |
b48f03b3 DC |
988 | * prune_dcache_sb - shrink the dcache |
989 | * @sb: superblock | |
f6041567 | 990 | * @nr_to_scan : number of entries to try to free |
9b17c623 | 991 | * @nid: which node to scan for freeable entities |
b48f03b3 | 992 | * |
f6041567 | 993 | * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is |
b48f03b3 DC |
994 | * done when we need more memory an called from the superblock shrinker |
995 | * function. | |
3049cfe2 | 996 | * |
b48f03b3 DC |
997 | * This function may fail to free any resources if all the dentries are in |
998 | * use. | |
3049cfe2 | 999 | */ |
9b17c623 DC |
1000 | long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan, |
1001 | int nid) | |
3049cfe2 | 1002 | { |
f6041567 DC |
1003 | LIST_HEAD(dispose); |
1004 | long freed; | |
3049cfe2 | 1005 | |
9b17c623 DC |
1006 | freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate, |
1007 | &dispose, &nr_to_scan); | |
f6041567 | 1008 | shrink_dentry_list(&dispose); |
0a234c6d | 1009 | return freed; |
da3bbdd4 | 1010 | } |
23044507 | 1011 | |
4e717f5c GC |
1012 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
1013 | spinlock_t *lru_lock, void *arg) | |
dd1f6b2e | 1014 | { |
4e717f5c GC |
1015 | struct list_head *freeable = arg; |
1016 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 1017 | |
4e717f5c GC |
1018 | /* |
1019 | * we are inverting the lru lock/dentry->d_lock here, | |
1020 | * so use a trylock. If we fail to get the lock, just skip | |
1021 | * it | |
1022 | */ | |
1023 | if (!spin_trylock(&dentry->d_lock)) | |
1024 | return LRU_SKIP; | |
1025 | ||
89dc77bc | 1026 | d_lru_shrink_move(dentry, freeable); |
4e717f5c | 1027 | spin_unlock(&dentry->d_lock); |
ec33679d | 1028 | |
4e717f5c | 1029 | return LRU_REMOVED; |
da3bbdd4 KM |
1030 | } |
1031 | ||
4e717f5c | 1032 | |
1da177e4 LT |
1033 | /** |
1034 | * shrink_dcache_sb - shrink dcache for a superblock | |
1035 | * @sb: superblock | |
1036 | * | |
3049cfe2 CH |
1037 | * Shrink the dcache for the specified super block. This is used to free |
1038 | * the dcache before unmounting a file system. | |
1da177e4 | 1039 | */ |
3049cfe2 | 1040 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1041 | { |
4e717f5c GC |
1042 | long freed; |
1043 | ||
1044 | do { | |
1045 | LIST_HEAD(dispose); | |
1046 | ||
1047 | freed = list_lru_walk(&sb->s_dentry_lru, | |
1048 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); | |
3049cfe2 | 1049 | |
4e717f5c GC |
1050 | this_cpu_sub(nr_dentry_unused, freed); |
1051 | shrink_dentry_list(&dispose); | |
1052 | } while (freed > 0); | |
1da177e4 | 1053 | } |
ec4f8605 | 1054 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1055 | |
db14fc3a MS |
1056 | /** |
1057 | * enum d_walk_ret - action to talke during tree walk | |
1058 | * @D_WALK_CONTINUE: contrinue walk | |
1059 | * @D_WALK_QUIT: quit walk | |
1060 | * @D_WALK_NORETRY: quit when retry is needed | |
1061 | * @D_WALK_SKIP: skip this dentry and its children | |
1062 | */ | |
1063 | enum d_walk_ret { | |
1064 | D_WALK_CONTINUE, | |
1065 | D_WALK_QUIT, | |
1066 | D_WALK_NORETRY, | |
1067 | D_WALK_SKIP, | |
1068 | }; | |
c826cb7d | 1069 | |
1da177e4 | 1070 | /** |
db14fc3a MS |
1071 | * d_walk - walk the dentry tree |
1072 | * @parent: start of walk | |
1073 | * @data: data passed to @enter() and @finish() | |
1074 | * @enter: callback when first entering the dentry | |
1075 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1076 | * |
db14fc3a | 1077 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1078 | */ |
db14fc3a MS |
1079 | static void d_walk(struct dentry *parent, void *data, |
1080 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1081 | void (*finish)(void *)) | |
1da177e4 | 1082 | { |
949854d0 | 1083 | struct dentry *this_parent; |
1da177e4 | 1084 | struct list_head *next; |
48f5ec21 | 1085 | unsigned seq = 0; |
db14fc3a MS |
1086 | enum d_walk_ret ret; |
1087 | bool retry = true; | |
949854d0 | 1088 | |
58db63d0 | 1089 | again: |
48f5ec21 | 1090 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1091 | this_parent = parent; |
2fd6b7f5 | 1092 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1093 | |
1094 | ret = enter(data, this_parent); | |
1095 | switch (ret) { | |
1096 | case D_WALK_CONTINUE: | |
1097 | break; | |
1098 | case D_WALK_QUIT: | |
1099 | case D_WALK_SKIP: | |
1100 | goto out_unlock; | |
1101 | case D_WALK_NORETRY: | |
1102 | retry = false; | |
1103 | break; | |
1104 | } | |
1da177e4 LT |
1105 | repeat: |
1106 | next = this_parent->d_subdirs.next; | |
1107 | resume: | |
1108 | while (next != &this_parent->d_subdirs) { | |
1109 | struct list_head *tmp = next; | |
5160ee6f | 1110 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1111 | next = tmp->next; |
2fd6b7f5 NP |
1112 | |
1113 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
db14fc3a MS |
1114 | |
1115 | ret = enter(data, dentry); | |
1116 | switch (ret) { | |
1117 | case D_WALK_CONTINUE: | |
1118 | break; | |
1119 | case D_WALK_QUIT: | |
2fd6b7f5 | 1120 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1121 | goto out_unlock; |
1122 | case D_WALK_NORETRY: | |
1123 | retry = false; | |
1124 | break; | |
1125 | case D_WALK_SKIP: | |
1126 | spin_unlock(&dentry->d_lock); | |
1127 | continue; | |
2fd6b7f5 | 1128 | } |
db14fc3a | 1129 | |
1da177e4 | 1130 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1131 | spin_unlock(&this_parent->d_lock); |
1132 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1133 | this_parent = dentry; |
2fd6b7f5 | 1134 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1135 | goto repeat; |
1136 | } | |
2fd6b7f5 | 1137 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1138 | } |
1139 | /* | |
1140 | * All done at this level ... ascend and resume the search. | |
1141 | */ | |
1142 | if (this_parent != parent) { | |
c826cb7d | 1143 | struct dentry *child = this_parent; |
31dec132 AV |
1144 | this_parent = child->d_parent; |
1145 | ||
1146 | rcu_read_lock(); | |
1147 | spin_unlock(&child->d_lock); | |
1148 | spin_lock(&this_parent->d_lock); | |
1149 | ||
1150 | /* | |
1151 | * might go back up the wrong parent if we have had a rename | |
1152 | * or deletion | |
1153 | */ | |
1154 | if (this_parent != child->d_parent || | |
1155 | (child->d_flags & DCACHE_DENTRY_KILLED) || | |
1156 | need_seqretry(&rename_lock, seq)) { | |
1157 | spin_unlock(&this_parent->d_lock); | |
1158 | rcu_read_unlock(); | |
949854d0 | 1159 | goto rename_retry; |
31dec132 AV |
1160 | } |
1161 | rcu_read_unlock(); | |
949854d0 | 1162 | next = child->d_u.d_child.next; |
1da177e4 LT |
1163 | goto resume; |
1164 | } | |
48f5ec21 | 1165 | if (need_seqretry(&rename_lock, seq)) { |
db14fc3a | 1166 | spin_unlock(&this_parent->d_lock); |
949854d0 | 1167 | goto rename_retry; |
db14fc3a MS |
1168 | } |
1169 | if (finish) | |
1170 | finish(data); | |
1171 | ||
1172 | out_unlock: | |
1173 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1174 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1175 | return; |
58db63d0 NP |
1176 | |
1177 | rename_retry: | |
db14fc3a MS |
1178 | if (!retry) |
1179 | return; | |
48f5ec21 | 1180 | seq = 1; |
58db63d0 | 1181 | goto again; |
1da177e4 | 1182 | } |
db14fc3a MS |
1183 | |
1184 | /* | |
1185 | * Search for at least 1 mount point in the dentry's subdirs. | |
1186 | * We descend to the next level whenever the d_subdirs | |
1187 | * list is non-empty and continue searching. | |
1188 | */ | |
1189 | ||
db14fc3a MS |
1190 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) |
1191 | { | |
1192 | int *ret = data; | |
1193 | if (d_mountpoint(dentry)) { | |
1194 | *ret = 1; | |
1195 | return D_WALK_QUIT; | |
1196 | } | |
1197 | return D_WALK_CONTINUE; | |
1198 | } | |
1199 | ||
69c88dc7 RD |
1200 | /** |
1201 | * have_submounts - check for mounts over a dentry | |
1202 | * @parent: dentry to check. | |
1203 | * | |
1204 | * Return true if the parent or its subdirectories contain | |
1205 | * a mount point | |
1206 | */ | |
db14fc3a MS |
1207 | int have_submounts(struct dentry *parent) |
1208 | { | |
1209 | int ret = 0; | |
1210 | ||
1211 | d_walk(parent, &ret, check_mount, NULL); | |
1212 | ||
1213 | return ret; | |
1214 | } | |
ec4f8605 | 1215 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1216 | |
eed81007 MS |
1217 | /* |
1218 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1219 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1220 | * subtree can become unreachable). | |
1221 | * | |
1222 | * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For | |
1223 | * this reason take rename_lock and d_lock on dentry and ancestors. | |
1224 | */ | |
1225 | int d_set_mounted(struct dentry *dentry) | |
1226 | { | |
1227 | struct dentry *p; | |
1228 | int ret = -ENOENT; | |
1229 | write_seqlock(&rename_lock); | |
1230 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1231 | /* Need exclusion wrt. check_submounts_and_drop() */ | |
1232 | spin_lock(&p->d_lock); | |
1233 | if (unlikely(d_unhashed(p))) { | |
1234 | spin_unlock(&p->d_lock); | |
1235 | goto out; | |
1236 | } | |
1237 | spin_unlock(&p->d_lock); | |
1238 | } | |
1239 | spin_lock(&dentry->d_lock); | |
1240 | if (!d_unlinked(dentry)) { | |
1241 | dentry->d_flags |= DCACHE_MOUNTED; | |
1242 | ret = 0; | |
1243 | } | |
1244 | spin_unlock(&dentry->d_lock); | |
1245 | out: | |
1246 | write_sequnlock(&rename_lock); | |
1247 | return ret; | |
1248 | } | |
1249 | ||
1da177e4 | 1250 | /* |
fd517909 | 1251 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1252 | * and move any unused dentries to the end of the unused |
1253 | * list for prune_dcache(). We descend to the next level | |
1254 | * whenever the d_subdirs list is non-empty and continue | |
1255 | * searching. | |
1256 | * | |
1257 | * It returns zero iff there are no unused children, | |
1258 | * otherwise it returns the number of children moved to | |
1259 | * the end of the unused list. This may not be the total | |
1260 | * number of unused children, because select_parent can | |
1261 | * drop the lock and return early due to latency | |
1262 | * constraints. | |
1263 | */ | |
1da177e4 | 1264 | |
db14fc3a MS |
1265 | struct select_data { |
1266 | struct dentry *start; | |
1267 | struct list_head dispose; | |
1268 | int found; | |
1269 | }; | |
23044507 | 1270 | |
db14fc3a MS |
1271 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1272 | { | |
1273 | struct select_data *data = _data; | |
1274 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1275 | |
db14fc3a MS |
1276 | if (data->start == dentry) |
1277 | goto out; | |
2fd6b7f5 | 1278 | |
fe91522a | 1279 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1280 | data->found++; |
fe91522a AV |
1281 | } else { |
1282 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1283 | d_lru_del(dentry); | |
1284 | if (!dentry->d_lockref.count) { | |
1285 | d_shrink_add(dentry, &data->dispose); | |
1286 | data->found++; | |
1287 | } | |
1da177e4 | 1288 | } |
db14fc3a MS |
1289 | /* |
1290 | * We can return to the caller if we have found some (this | |
1291 | * ensures forward progress). We'll be coming back to find | |
1292 | * the rest. | |
1293 | */ | |
fe91522a AV |
1294 | if (!list_empty(&data->dispose)) |
1295 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1296 | out: |
db14fc3a | 1297 | return ret; |
1da177e4 LT |
1298 | } |
1299 | ||
1300 | /** | |
1301 | * shrink_dcache_parent - prune dcache | |
1302 | * @parent: parent of entries to prune | |
1303 | * | |
1304 | * Prune the dcache to remove unused children of the parent dentry. | |
1305 | */ | |
db14fc3a | 1306 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1307 | { |
db14fc3a MS |
1308 | for (;;) { |
1309 | struct select_data data; | |
1da177e4 | 1310 | |
db14fc3a MS |
1311 | INIT_LIST_HEAD(&data.dispose); |
1312 | data.start = parent; | |
1313 | data.found = 0; | |
1314 | ||
1315 | d_walk(parent, &data, select_collect, NULL); | |
1316 | if (!data.found) | |
1317 | break; | |
1318 | ||
1319 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1320 | cond_resched(); |
1321 | } | |
1da177e4 | 1322 | } |
ec4f8605 | 1323 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1324 | |
9c8c10e2 | 1325 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1326 | { |
9c8c10e2 AV |
1327 | /* it has busy descendents; complain about those instead */ |
1328 | if (!list_empty(&dentry->d_subdirs)) | |
1329 | return D_WALK_CONTINUE; | |
42c32608 | 1330 | |
9c8c10e2 AV |
1331 | /* root with refcount 1 is fine */ |
1332 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1333 | return D_WALK_CONTINUE; | |
1334 | ||
1335 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1336 | " still in use (%d) [unmount of %s %s]\n", | |
42c32608 AV |
1337 | dentry, |
1338 | dentry->d_inode ? | |
1339 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1340 | dentry, |
42c32608 AV |
1341 | dentry->d_lockref.count, |
1342 | dentry->d_sb->s_type->name, | |
1343 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1344 | WARN_ON(1); |
1345 | return D_WALK_CONTINUE; | |
1346 | } | |
1347 | ||
1348 | static void do_one_tree(struct dentry *dentry) | |
1349 | { | |
1350 | shrink_dcache_parent(dentry); | |
1351 | d_walk(dentry, dentry, umount_check, NULL); | |
1352 | d_drop(dentry); | |
1353 | dput(dentry); | |
42c32608 AV |
1354 | } |
1355 | ||
1356 | /* | |
1357 | * destroy the dentries attached to a superblock on unmounting | |
1358 | */ | |
1359 | void shrink_dcache_for_umount(struct super_block *sb) | |
1360 | { | |
1361 | struct dentry *dentry; | |
1362 | ||
9c8c10e2 | 1363 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1364 | |
1365 | dentry = sb->s_root; | |
1366 | sb->s_root = NULL; | |
9c8c10e2 | 1367 | do_one_tree(dentry); |
42c32608 AV |
1368 | |
1369 | while (!hlist_bl_empty(&sb->s_anon)) { | |
9c8c10e2 AV |
1370 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash)); |
1371 | do_one_tree(dentry); | |
42c32608 AV |
1372 | } |
1373 | } | |
1374 | ||
848ac114 MS |
1375 | static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry) |
1376 | { | |
1377 | struct select_data *data = _data; | |
1378 | ||
1379 | if (d_mountpoint(dentry)) { | |
1380 | data->found = -EBUSY; | |
1381 | return D_WALK_QUIT; | |
1382 | } | |
1383 | ||
1384 | return select_collect(_data, dentry); | |
1385 | } | |
1386 | ||
1387 | static void check_and_drop(void *_data) | |
1388 | { | |
1389 | struct select_data *data = _data; | |
1390 | ||
1391 | if (d_mountpoint(data->start)) | |
1392 | data->found = -EBUSY; | |
1393 | if (!data->found) | |
1394 | __d_drop(data->start); | |
1395 | } | |
1396 | ||
1397 | /** | |
1398 | * check_submounts_and_drop - prune dcache, check for submounts and drop | |
1399 | * | |
1400 | * All done as a single atomic operation relative to has_unlinked_ancestor(). | |
1401 | * Returns 0 if successfully unhashed @parent. If there were submounts then | |
1402 | * return -EBUSY. | |
1403 | * | |
1404 | * @dentry: dentry to prune and drop | |
1405 | */ | |
1406 | int check_submounts_and_drop(struct dentry *dentry) | |
1407 | { | |
1408 | int ret = 0; | |
1409 | ||
1410 | /* Negative dentries can be dropped without further checks */ | |
1411 | if (!dentry->d_inode) { | |
1412 | d_drop(dentry); | |
1413 | goto out; | |
1414 | } | |
1415 | ||
1416 | for (;;) { | |
1417 | struct select_data data; | |
1418 | ||
1419 | INIT_LIST_HEAD(&data.dispose); | |
1420 | data.start = dentry; | |
1421 | data.found = 0; | |
1422 | ||
1423 | d_walk(dentry, &data, check_and_collect, check_and_drop); | |
1424 | ret = data.found; | |
1425 | ||
1426 | if (!list_empty(&data.dispose)) | |
1427 | shrink_dentry_list(&data.dispose); | |
1428 | ||
1429 | if (ret <= 0) | |
1430 | break; | |
1431 | ||
1432 | cond_resched(); | |
1433 | } | |
1434 | ||
1435 | out: | |
1436 | return ret; | |
1437 | } | |
1438 | EXPORT_SYMBOL(check_submounts_and_drop); | |
1439 | ||
1da177e4 | 1440 | /** |
a4464dbc AV |
1441 | * __d_alloc - allocate a dcache entry |
1442 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1443 | * @name: qstr of the name |
1444 | * | |
1445 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1446 | * available. On a success the dentry is returned. The name passed in is | |
1447 | * copied and the copy passed in may be reused after this call. | |
1448 | */ | |
1449 | ||
a4464dbc | 1450 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1451 | { |
1452 | struct dentry *dentry; | |
1453 | char *dname; | |
1454 | ||
e12ba74d | 1455 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1456 | if (!dentry) |
1457 | return NULL; | |
1458 | ||
6326c71f LT |
1459 | /* |
1460 | * We guarantee that the inline name is always NUL-terminated. | |
1461 | * This way the memcpy() done by the name switching in rename | |
1462 | * will still always have a NUL at the end, even if we might | |
1463 | * be overwriting an internal NUL character | |
1464 | */ | |
1465 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 | 1466 | if (name->len > DNAME_INLINE_LEN-1) { |
8d85b484 AV |
1467 | size_t size = offsetof(struct external_name, name[1]); |
1468 | struct external_name *p = kmalloc(size + name->len, GFP_KERNEL); | |
1469 | if (!p) { | |
1da177e4 LT |
1470 | kmem_cache_free(dentry_cache, dentry); |
1471 | return NULL; | |
1472 | } | |
8d85b484 AV |
1473 | atomic_set(&p->u.count, 1); |
1474 | dname = p->name; | |
1da177e4 LT |
1475 | } else { |
1476 | dname = dentry->d_iname; | |
1477 | } | |
1da177e4 LT |
1478 | |
1479 | dentry->d_name.len = name->len; | |
1480 | dentry->d_name.hash = name->hash; | |
1481 | memcpy(dname, name->name, name->len); | |
1482 | dname[name->len] = 0; | |
1483 | ||
6326c71f LT |
1484 | /* Make sure we always see the terminating NUL character */ |
1485 | smp_wmb(); | |
1486 | dentry->d_name.name = dname; | |
1487 | ||
98474236 | 1488 | dentry->d_lockref.count = 1; |
dea3667b | 1489 | dentry->d_flags = 0; |
1da177e4 | 1490 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1491 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1492 | dentry->d_inode = NULL; |
a4464dbc AV |
1493 | dentry->d_parent = dentry; |
1494 | dentry->d_sb = sb; | |
1da177e4 LT |
1495 | dentry->d_op = NULL; |
1496 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1497 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1498 | INIT_LIST_HEAD(&dentry->d_lru); |
1499 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
b3d9b7a3 | 1500 | INIT_HLIST_NODE(&dentry->d_alias); |
2fd6b7f5 | 1501 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1502 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1503 | |
3e880fb5 | 1504 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1505 | |
1da177e4 LT |
1506 | return dentry; |
1507 | } | |
a4464dbc AV |
1508 | |
1509 | /** | |
1510 | * d_alloc - allocate a dcache entry | |
1511 | * @parent: parent of entry to allocate | |
1512 | * @name: qstr of the name | |
1513 | * | |
1514 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1515 | * available. On a success the dentry is returned. The name passed in is | |
1516 | * copied and the copy passed in may be reused after this call. | |
1517 | */ | |
1518 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1519 | { | |
1520 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1521 | if (!dentry) | |
1522 | return NULL; | |
1523 | ||
1524 | spin_lock(&parent->d_lock); | |
1525 | /* | |
1526 | * don't need child lock because it is not subject | |
1527 | * to concurrency here | |
1528 | */ | |
1529 | __dget_dlock(parent); | |
1530 | dentry->d_parent = parent; | |
1531 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1532 | spin_unlock(&parent->d_lock); | |
1533 | ||
1534 | return dentry; | |
1535 | } | |
ec4f8605 | 1536 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1537 | |
e1a24bb0 BF |
1538 | /** |
1539 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1540 | * @sb: the superblock | |
1541 | * @name: qstr of the name | |
1542 | * | |
1543 | * For a filesystem that just pins its dentries in memory and never | |
1544 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1545 | */ | |
4b936885 NP |
1546 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1547 | { | |
e1a24bb0 | 1548 | return __d_alloc(sb, name); |
4b936885 NP |
1549 | } |
1550 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1551 | ||
1da177e4 LT |
1552 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1553 | { | |
1554 | struct qstr q; | |
1555 | ||
1556 | q.name = name; | |
1557 | q.len = strlen(name); | |
1558 | q.hash = full_name_hash(q.name, q.len); | |
1559 | return d_alloc(parent, &q); | |
1560 | } | |
ef26ca97 | 1561 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1562 | |
fb045adb NP |
1563 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1564 | { | |
6f7f7caa LT |
1565 | WARN_ON_ONCE(dentry->d_op); |
1566 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1567 | DCACHE_OP_COMPARE | |
1568 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1569 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1570 | DCACHE_OP_DELETE )); |
1571 | dentry->d_op = op; | |
1572 | if (!op) | |
1573 | return; | |
1574 | if (op->d_hash) | |
1575 | dentry->d_flags |= DCACHE_OP_HASH; | |
1576 | if (op->d_compare) | |
1577 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1578 | if (op->d_revalidate) | |
1579 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1580 | if (op->d_weak_revalidate) |
1581 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1582 | if (op->d_delete) |
1583 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1584 | if (op->d_prune) |
1585 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1586 | |
1587 | } | |
1588 | EXPORT_SYMBOL(d_set_d_op); | |
1589 | ||
b18825a7 DH |
1590 | static unsigned d_flags_for_inode(struct inode *inode) |
1591 | { | |
1592 | unsigned add_flags = DCACHE_FILE_TYPE; | |
1593 | ||
1594 | if (!inode) | |
1595 | return DCACHE_MISS_TYPE; | |
1596 | ||
1597 | if (S_ISDIR(inode->i_mode)) { | |
1598 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1599 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1600 | if (unlikely(!inode->i_op->lookup)) | |
1601 | add_flags = DCACHE_AUTODIR_TYPE; | |
1602 | else | |
1603 | inode->i_opflags |= IOP_LOOKUP; | |
1604 | } | |
1605 | } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1606 | if (unlikely(inode->i_op->follow_link)) | |
1607 | add_flags = DCACHE_SYMLINK_TYPE; | |
1608 | else | |
1609 | inode->i_opflags |= IOP_NOFOLLOW; | |
1610 | } | |
1611 | ||
1612 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1613 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1614 | return add_flags; | |
1615 | } | |
1616 | ||
360da900 OH |
1617 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1618 | { | |
b18825a7 DH |
1619 | unsigned add_flags = d_flags_for_inode(inode); |
1620 | ||
b23fb0a6 | 1621 | spin_lock(&dentry->d_lock); |
22213318 | 1622 | __d_set_type(dentry, add_flags); |
b18825a7 | 1623 | if (inode) |
b3d9b7a3 | 1624 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); |
360da900 | 1625 | dentry->d_inode = inode; |
31e6b01f | 1626 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1627 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1628 | fsnotify_d_instantiate(dentry, inode); |
1629 | } | |
1630 | ||
1da177e4 LT |
1631 | /** |
1632 | * d_instantiate - fill in inode information for a dentry | |
1633 | * @entry: dentry to complete | |
1634 | * @inode: inode to attach to this dentry | |
1635 | * | |
1636 | * Fill in inode information in the entry. | |
1637 | * | |
1638 | * This turns negative dentries into productive full members | |
1639 | * of society. | |
1640 | * | |
1641 | * NOTE! This 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. | |
1644 | */ | |
1645 | ||
1646 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1647 | { | |
b3d9b7a3 | 1648 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
873feea0 NP |
1649 | if (inode) |
1650 | spin_lock(&inode->i_lock); | |
360da900 | 1651 | __d_instantiate(entry, inode); |
873feea0 NP |
1652 | if (inode) |
1653 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1654 | security_d_instantiate(entry, inode); |
1655 | } | |
ec4f8605 | 1656 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1657 | |
1658 | /** | |
1659 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1660 | * @entry: dentry to instantiate | |
1661 | * @inode: inode to attach to this dentry | |
1662 | * | |
1663 | * Fill in inode information in the entry. On success, it returns NULL. | |
1664 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1665 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1666 | * |
1667 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1668 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1669 | * |
1670 | * This also assumes that the inode count has been incremented | |
1671 | * (or otherwise set) by the caller to indicate that it is now | |
1672 | * in use by the dcache. | |
1da177e4 | 1673 | */ |
770bfad8 DH |
1674 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1675 | struct inode *inode) | |
1da177e4 LT |
1676 | { |
1677 | struct dentry *alias; | |
1678 | int len = entry->d_name.len; | |
1679 | const char *name = entry->d_name.name; | |
1680 | unsigned int hash = entry->d_name.hash; | |
1681 | ||
770bfad8 | 1682 | if (!inode) { |
360da900 | 1683 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1684 | return NULL; |
1685 | } | |
1686 | ||
b67bfe0d | 1687 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
9abca360 NP |
1688 | /* |
1689 | * Don't need alias->d_lock here, because aliases with | |
1690 | * d_parent == entry->d_parent are not subject to name or | |
1691 | * parent changes, because the parent inode i_mutex is held. | |
1692 | */ | |
12f8ad4b | 1693 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1694 | continue; |
1695 | if (alias->d_parent != entry->d_parent) | |
1696 | continue; | |
ee983e89 LT |
1697 | if (alias->d_name.len != len) |
1698 | continue; | |
12f8ad4b | 1699 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1700 | continue; |
dc0474be | 1701 | __dget(alias); |
1da177e4 LT |
1702 | return alias; |
1703 | } | |
770bfad8 | 1704 | |
360da900 | 1705 | __d_instantiate(entry, inode); |
1da177e4 LT |
1706 | return NULL; |
1707 | } | |
770bfad8 DH |
1708 | |
1709 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1710 | { | |
1711 | struct dentry *result; | |
1712 | ||
b3d9b7a3 | 1713 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
770bfad8 | 1714 | |
873feea0 NP |
1715 | if (inode) |
1716 | spin_lock(&inode->i_lock); | |
770bfad8 | 1717 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1718 | if (inode) |
1719 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1720 | |
1721 | if (!result) { | |
1722 | security_d_instantiate(entry, inode); | |
1723 | return NULL; | |
1724 | } | |
1725 | ||
1726 | BUG_ON(!d_unhashed(result)); | |
1727 | iput(inode); | |
1728 | return result; | |
1729 | } | |
1730 | ||
1da177e4 LT |
1731 | EXPORT_SYMBOL(d_instantiate_unique); |
1732 | ||
b70a80e7 MS |
1733 | /** |
1734 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1735 | * @entry: dentry to complete | |
1736 | * @inode: inode to attach to this dentry | |
1737 | * | |
1738 | * Fill in inode information in the entry. If a directory alias is found, then | |
1739 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1740 | * guarantees that a directory inode may never have more than one alias. | |
1741 | */ | |
1742 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1743 | { | |
1744 | BUG_ON(!hlist_unhashed(&entry->d_alias)); | |
1745 | ||
1746 | spin_lock(&inode->i_lock); | |
1747 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1748 | spin_unlock(&inode->i_lock); | |
1749 | iput(inode); | |
1750 | return -EBUSY; | |
1751 | } | |
1752 | __d_instantiate(entry, inode); | |
1753 | spin_unlock(&inode->i_lock); | |
1754 | security_d_instantiate(entry, inode); | |
1755 | ||
1756 | return 0; | |
1757 | } | |
1758 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1759 | ||
adc0e91a AV |
1760 | struct dentry *d_make_root(struct inode *root_inode) |
1761 | { | |
1762 | struct dentry *res = NULL; | |
1763 | ||
1764 | if (root_inode) { | |
26fe5750 | 1765 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1766 | |
1767 | res = __d_alloc(root_inode->i_sb, &name); | |
1768 | if (res) | |
1769 | d_instantiate(res, root_inode); | |
1770 | else | |
1771 | iput(root_inode); | |
1772 | } | |
1773 | return res; | |
1774 | } | |
1775 | EXPORT_SYMBOL(d_make_root); | |
1776 | ||
d891eedb BF |
1777 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1778 | { | |
1779 | struct dentry *alias; | |
1780 | ||
b3d9b7a3 | 1781 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1782 | return NULL; |
b3d9b7a3 | 1783 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); |
d891eedb BF |
1784 | __dget(alias); |
1785 | return alias; | |
1786 | } | |
1787 | ||
46f72b34 SW |
1788 | /** |
1789 | * d_find_any_alias - find any alias for a given inode | |
1790 | * @inode: inode to find an alias for | |
1791 | * | |
1792 | * If any aliases exist for the given inode, take and return a | |
1793 | * reference for one of them. If no aliases exist, return %NULL. | |
1794 | */ | |
1795 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1796 | { |
1797 | struct dentry *de; | |
1798 | ||
1799 | spin_lock(&inode->i_lock); | |
1800 | de = __d_find_any_alias(inode); | |
1801 | spin_unlock(&inode->i_lock); | |
1802 | return de; | |
1803 | } | |
46f72b34 | 1804 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1805 | |
49c7dd28 | 1806 | static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected) |
4ea3ada2 | 1807 | { |
b911a6bd | 1808 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1809 | struct dentry *tmp; |
1810 | struct dentry *res; | |
b18825a7 | 1811 | unsigned add_flags; |
4ea3ada2 CH |
1812 | |
1813 | if (!inode) | |
44003728 | 1814 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1815 | if (IS_ERR(inode)) |
1816 | return ERR_CAST(inode); | |
1817 | ||
d891eedb | 1818 | res = d_find_any_alias(inode); |
9308a612 CH |
1819 | if (res) |
1820 | goto out_iput; | |
1821 | ||
a4464dbc | 1822 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1823 | if (!tmp) { |
1824 | res = ERR_PTR(-ENOMEM); | |
1825 | goto out_iput; | |
4ea3ada2 | 1826 | } |
b5c84bf6 | 1827 | |
873feea0 | 1828 | spin_lock(&inode->i_lock); |
d891eedb | 1829 | res = __d_find_any_alias(inode); |
9308a612 | 1830 | if (res) { |
873feea0 | 1831 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1832 | dput(tmp); |
1833 | goto out_iput; | |
1834 | } | |
1835 | ||
1836 | /* attach a disconnected dentry */ | |
1a0a397e BF |
1837 | add_flags = d_flags_for_inode(inode); |
1838 | ||
1839 | if (disconnected) | |
1840 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 1841 | |
9308a612 | 1842 | spin_lock(&tmp->d_lock); |
9308a612 | 1843 | tmp->d_inode = inode; |
b18825a7 | 1844 | tmp->d_flags |= add_flags; |
b3d9b7a3 | 1845 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1846 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1847 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1848 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1849 | spin_unlock(&tmp->d_lock); |
873feea0 | 1850 | spin_unlock(&inode->i_lock); |
24ff6663 | 1851 | security_d_instantiate(tmp, inode); |
9308a612 | 1852 | |
9308a612 CH |
1853 | return tmp; |
1854 | ||
1855 | out_iput: | |
24ff6663 JB |
1856 | if (res && !IS_ERR(res)) |
1857 | security_d_instantiate(res, inode); | |
9308a612 CH |
1858 | iput(inode); |
1859 | return res; | |
4ea3ada2 | 1860 | } |
1a0a397e BF |
1861 | |
1862 | /** | |
1863 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
1864 | * @inode: inode to allocate the dentry for | |
1865 | * | |
1866 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1867 | * similar open by handle operations. The returned dentry may be anonymous, | |
1868 | * or may have a full name (if the inode was already in the cache). | |
1869 | * | |
1870 | * When called on a directory inode, we must ensure that the inode only ever | |
1871 | * has one dentry. If a dentry is found, that is returned instead of | |
1872 | * allocating a new one. | |
1873 | * | |
1874 | * On successful return, the reference to the inode has been transferred | |
1875 | * to the dentry. In case of an error the reference on the inode is released. | |
1876 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1877 | * be passed in and the error will be propagated to the return value, | |
1878 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
1879 | */ | |
1880 | struct dentry *d_obtain_alias(struct inode *inode) | |
1881 | { | |
1882 | return __d_obtain_alias(inode, 1); | |
1883 | } | |
adc48720 | 1884 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 1885 | |
1a0a397e BF |
1886 | /** |
1887 | * d_obtain_root - find or allocate a dentry for a given inode | |
1888 | * @inode: inode to allocate the dentry for | |
1889 | * | |
1890 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
1891 | * | |
1892 | * We must ensure that directory inodes only ever have one dentry. If a | |
1893 | * dentry is found, that is returned instead of allocating a new one. | |
1894 | * | |
1895 | * On successful return, the reference to the inode has been transferred | |
1896 | * to the dentry. In case of an error the reference on the inode is | |
1897 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
1898 | * error will be propagate to the return value, with a %NULL @inode | |
1899 | * replaced by ERR_PTR(-ESTALE). | |
1900 | */ | |
1901 | struct dentry *d_obtain_root(struct inode *inode) | |
1902 | { | |
1903 | return __d_obtain_alias(inode, 0); | |
1904 | } | |
1905 | EXPORT_SYMBOL(d_obtain_root); | |
1906 | ||
9403540c BN |
1907 | /** |
1908 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1909 | * @inode: the inode case-insensitive lookup has found | |
1910 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1911 | * @name: the case-exact name to be associated with the returned dentry | |
1912 | * | |
1913 | * This is to avoid filling the dcache with case-insensitive names to the | |
1914 | * same inode, only the actual correct case is stored in the dcache for | |
1915 | * case-insensitive filesystems. | |
1916 | * | |
1917 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1918 | * already exists in in the dcache, use it and return it. | |
1919 | * | |
1920 | * If no entry exists with the exact case name, allocate new dentry with | |
1921 | * the exact case, and return the spliced entry. | |
1922 | */ | |
e45b590b | 1923 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1924 | struct qstr *name) |
1925 | { | |
9403540c BN |
1926 | struct dentry *found; |
1927 | struct dentry *new; | |
1928 | ||
b6520c81 CH |
1929 | /* |
1930 | * First check if a dentry matching the name already exists, | |
1931 | * if not go ahead and create it now. | |
1932 | */ | |
9403540c | 1933 | found = d_hash_and_lookup(dentry->d_parent, name); |
4f522a24 AV |
1934 | if (unlikely(IS_ERR(found))) |
1935 | goto err_out; | |
9403540c BN |
1936 | if (!found) { |
1937 | new = d_alloc(dentry->d_parent, name); | |
1938 | if (!new) { | |
4f522a24 | 1939 | found = ERR_PTR(-ENOMEM); |
9403540c BN |
1940 | goto err_out; |
1941 | } | |
b6520c81 | 1942 | |
9403540c BN |
1943 | found = d_splice_alias(inode, new); |
1944 | if (found) { | |
1945 | dput(new); | |
1946 | return found; | |
1947 | } | |
1948 | return new; | |
1949 | } | |
b6520c81 CH |
1950 | |
1951 | /* | |
1952 | * If a matching dentry exists, and it's not negative use it. | |
1953 | * | |
1954 | * Decrement the reference count to balance the iget() done | |
1955 | * earlier on. | |
1956 | */ | |
9403540c BN |
1957 | if (found->d_inode) { |
1958 | if (unlikely(found->d_inode != inode)) { | |
1959 | /* This can't happen because bad inodes are unhashed. */ | |
1960 | BUG_ON(!is_bad_inode(inode)); | |
1961 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1962 | } | |
9403540c BN |
1963 | iput(inode); |
1964 | return found; | |
1965 | } | |
b6520c81 | 1966 | |
9403540c | 1967 | /* |
9403540c | 1968 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1969 | * already has a dentry. |
9403540c | 1970 | */ |
4513d899 AV |
1971 | new = d_splice_alias(inode, found); |
1972 | if (new) { | |
1973 | dput(found); | |
1974 | found = new; | |
9403540c | 1975 | } |
4513d899 | 1976 | return found; |
9403540c BN |
1977 | |
1978 | err_out: | |
1979 | iput(inode); | |
4f522a24 | 1980 | return found; |
9403540c | 1981 | } |
ec4f8605 | 1982 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1983 | |
12f8ad4b LT |
1984 | /* |
1985 | * Do the slow-case of the dentry name compare. | |
1986 | * | |
1987 | * Unlike the dentry_cmp() function, we need to atomically | |
da53be12 | 1988 | * load the name and length information, so that the |
12f8ad4b LT |
1989 | * filesystem can rely on them, and can use the 'name' and |
1990 | * 'len' information without worrying about walking off the | |
1991 | * end of memory etc. | |
1992 | * | |
1993 | * Thus the read_seqcount_retry() and the "duplicate" info | |
1994 | * in arguments (the low-level filesystem should not look | |
1995 | * at the dentry inode or name contents directly, since | |
1996 | * rename can change them while we're in RCU mode). | |
1997 | */ | |
1998 | enum slow_d_compare { | |
1999 | D_COMP_OK, | |
2000 | D_COMP_NOMATCH, | |
2001 | D_COMP_SEQRETRY, | |
2002 | }; | |
2003 | ||
2004 | static noinline enum slow_d_compare slow_dentry_cmp( | |
2005 | const struct dentry *parent, | |
12f8ad4b LT |
2006 | struct dentry *dentry, |
2007 | unsigned int seq, | |
2008 | const struct qstr *name) | |
2009 | { | |
2010 | int tlen = dentry->d_name.len; | |
2011 | const char *tname = dentry->d_name.name; | |
12f8ad4b LT |
2012 | |
2013 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2014 | cpu_relax(); | |
2015 | return D_COMP_SEQRETRY; | |
2016 | } | |
da53be12 | 2017 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
12f8ad4b LT |
2018 | return D_COMP_NOMATCH; |
2019 | return D_COMP_OK; | |
2020 | } | |
2021 | ||
31e6b01f NP |
2022 | /** |
2023 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2024 | * @parent: parent dentry | |
2025 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2026 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2027 | * Returns: dentry, or NULL |
2028 | * | |
2029 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2030 | * resolution (store-free path walking) design described in | |
2031 | * Documentation/filesystems/path-lookup.txt. | |
2032 | * | |
2033 | * This is not to be used outside core vfs. | |
2034 | * | |
2035 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2036 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2037 | * without taking d_lock and checking d_seq sequence count against @seq | |
2038 | * returned here. | |
2039 | * | |
15570086 | 2040 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2041 | * function. |
2042 | * | |
2043 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2044 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2045 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2046 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2047 | * |
2048 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2049 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2050 | */ |
8966be90 LT |
2051 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2052 | const struct qstr *name, | |
da53be12 | 2053 | unsigned *seqp) |
31e6b01f | 2054 | { |
26fe5750 | 2055 | u64 hashlen = name->hash_len; |
31e6b01f | 2056 | const unsigned char *str = name->name; |
26fe5750 | 2057 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 2058 | struct hlist_bl_node *node; |
31e6b01f NP |
2059 | struct dentry *dentry; |
2060 | ||
2061 | /* | |
2062 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2063 | * required to prevent single threaded performance regressions | |
2064 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2065 | * Keep the two functions in sync. | |
2066 | */ | |
2067 | ||
2068 | /* | |
2069 | * The hash list is protected using RCU. | |
2070 | * | |
2071 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2072 | * races with d_move(). | |
2073 | * | |
2074 | * It is possible that concurrent renames can mess up our list | |
2075 | * walk here and result in missing our dentry, resulting in the | |
2076 | * false-negative result. d_lookup() protects against concurrent | |
2077 | * renames using rename_lock seqlock. | |
2078 | * | |
b0a4bb83 | 2079 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2080 | */ |
b07ad996 | 2081 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2082 | unsigned seq; |
31e6b01f | 2083 | |
31e6b01f | 2084 | seqretry: |
12f8ad4b LT |
2085 | /* |
2086 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2087 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2088 | * |
2089 | * The caller must perform a seqcount check in order | |
da53be12 | 2090 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2091 | * |
2092 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2093 | * we don't wait for the sequence count to stabilize if it | |
2094 | * is in the middle of a sequence change. If we do the slow | |
2095 | * dentry compare, we will do seqretries until it is stable, | |
2096 | * and if we end up with a successful lookup, we actually | |
2097 | * want to exit RCU lookup anyway. | |
2098 | */ | |
2099 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2100 | if (dentry->d_parent != parent) |
2101 | continue; | |
2e321806 LT |
2102 | if (d_unhashed(dentry)) |
2103 | continue; | |
12f8ad4b | 2104 | |
830c0f0e | 2105 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
2106 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2107 | continue; | |
da53be12 LT |
2108 | *seqp = seq; |
2109 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
12f8ad4b LT |
2110 | case D_COMP_OK: |
2111 | return dentry; | |
2112 | case D_COMP_NOMATCH: | |
31e6b01f | 2113 | continue; |
12f8ad4b LT |
2114 | default: |
2115 | goto seqretry; | |
2116 | } | |
31e6b01f | 2117 | } |
12f8ad4b | 2118 | |
26fe5750 | 2119 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 2120 | continue; |
da53be12 | 2121 | *seqp = seq; |
26fe5750 | 2122 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 2123 | return dentry; |
31e6b01f NP |
2124 | } |
2125 | return NULL; | |
2126 | } | |
2127 | ||
1da177e4 LT |
2128 | /** |
2129 | * d_lookup - search for a dentry | |
2130 | * @parent: parent dentry | |
2131 | * @name: qstr of name we wish to find | |
b04f784e | 2132 | * Returns: dentry, or NULL |
1da177e4 | 2133 | * |
b04f784e NP |
2134 | * d_lookup searches the children of the parent dentry for the name in |
2135 | * question. If the dentry is found its reference count is incremented and the | |
2136 | * dentry is returned. The caller must use dput to free the entry when it has | |
2137 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2138 | */ |
da2d8455 | 2139 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2140 | { |
31e6b01f | 2141 | struct dentry *dentry; |
949854d0 | 2142 | unsigned seq; |
1da177e4 LT |
2143 | |
2144 | do { | |
2145 | seq = read_seqbegin(&rename_lock); | |
2146 | dentry = __d_lookup(parent, name); | |
2147 | if (dentry) | |
2148 | break; | |
2149 | } while (read_seqretry(&rename_lock, seq)); | |
2150 | return dentry; | |
2151 | } | |
ec4f8605 | 2152 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2153 | |
31e6b01f | 2154 | /** |
b04f784e NP |
2155 | * __d_lookup - search for a dentry (racy) |
2156 | * @parent: parent dentry | |
2157 | * @name: qstr of name we wish to find | |
2158 | * Returns: dentry, or NULL | |
2159 | * | |
2160 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2161 | * false-negative result due to unrelated rename activity. | |
2162 | * | |
2163 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2164 | * however it must be used carefully, eg. with a following d_lookup in | |
2165 | * the case of failure. | |
2166 | * | |
2167 | * __d_lookup callers must be commented. | |
2168 | */ | |
a713ca2a | 2169 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
2170 | { |
2171 | unsigned int len = name->len; | |
2172 | unsigned int hash = name->hash; | |
2173 | const unsigned char *str = name->name; | |
b07ad996 | 2174 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 2175 | struct hlist_bl_node *node; |
31e6b01f | 2176 | struct dentry *found = NULL; |
665a7583 | 2177 | struct dentry *dentry; |
1da177e4 | 2178 | |
31e6b01f NP |
2179 | /* |
2180 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2181 | * required to prevent single threaded performance regressions | |
2182 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2183 | * Keep the two functions in sync. | |
2184 | */ | |
2185 | ||
b04f784e NP |
2186 | /* |
2187 | * The hash list is protected using RCU. | |
2188 | * | |
2189 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2190 | * with d_move(). | |
2191 | * | |
2192 | * It is possible that concurrent renames can mess up our list | |
2193 | * walk here and result in missing our dentry, resulting in the | |
2194 | * false-negative result. d_lookup() protects against concurrent | |
2195 | * renames using rename_lock seqlock. | |
2196 | * | |
b0a4bb83 | 2197 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2198 | */ |
1da177e4 LT |
2199 | rcu_read_lock(); |
2200 | ||
b07ad996 | 2201 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2202 | |
1da177e4 LT |
2203 | if (dentry->d_name.hash != hash) |
2204 | continue; | |
1da177e4 LT |
2205 | |
2206 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2207 | if (dentry->d_parent != parent) |
2208 | goto next; | |
d0185c08 LT |
2209 | if (d_unhashed(dentry)) |
2210 | goto next; | |
2211 | ||
1da177e4 LT |
2212 | /* |
2213 | * It is safe to compare names since d_move() cannot | |
2214 | * change the qstr (protected by d_lock). | |
2215 | */ | |
fb045adb | 2216 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
2217 | int tlen = dentry->d_name.len; |
2218 | const char *tname = dentry->d_name.name; | |
da53be12 | 2219 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1da177e4 LT |
2220 | goto next; |
2221 | } else { | |
ee983e89 LT |
2222 | if (dentry->d_name.len != len) |
2223 | goto next; | |
12f8ad4b | 2224 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
2225 | goto next; |
2226 | } | |
2227 | ||
98474236 | 2228 | dentry->d_lockref.count++; |
d0185c08 | 2229 | found = dentry; |
1da177e4 LT |
2230 | spin_unlock(&dentry->d_lock); |
2231 | break; | |
2232 | next: | |
2233 | spin_unlock(&dentry->d_lock); | |
2234 | } | |
2235 | rcu_read_unlock(); | |
2236 | ||
2237 | return found; | |
2238 | } | |
2239 | ||
3e7e241f EB |
2240 | /** |
2241 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2242 | * @dir: Directory to search in | |
2243 | * @name: qstr of name we wish to find | |
2244 | * | |
4f522a24 | 2245 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2246 | */ |
2247 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2248 | { | |
3e7e241f EB |
2249 | /* |
2250 | * Check for a fs-specific hash function. Note that we must | |
2251 | * calculate the standard hash first, as the d_op->d_hash() | |
2252 | * routine may choose to leave the hash value unchanged. | |
2253 | */ | |
2254 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2255 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2256 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2257 | if (unlikely(err < 0)) |
2258 | return ERR_PTR(err); | |
3e7e241f | 2259 | } |
4f522a24 | 2260 | return d_lookup(dir, name); |
3e7e241f | 2261 | } |
4f522a24 | 2262 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2263 | |
1da177e4 | 2264 | /** |
786a5e15 | 2265 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2266 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2267 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2268 | * |
2269 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2270 | * This is used by ncpfs in its readdir implementation. | |
2271 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2272 | * |
2273 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2274 | */ |
d3a23e16 | 2275 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2276 | { |
786a5e15 | 2277 | struct dentry *child; |
d3a23e16 | 2278 | |
2fd6b7f5 | 2279 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2280 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2281 | if (dentry == child) { | |
2fd6b7f5 | 2282 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2283 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2284 | spin_unlock(&dentry->d_lock); |
2285 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2286 | return 1; |
2287 | } | |
2288 | } | |
2fd6b7f5 | 2289 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2290 | |
1da177e4 LT |
2291 | return 0; |
2292 | } | |
ec4f8605 | 2293 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2294 | |
2295 | /* | |
2296 | * When a file is deleted, we have two options: | |
2297 | * - turn this dentry into a negative dentry | |
2298 | * - unhash this dentry and free it. | |
2299 | * | |
2300 | * Usually, we want to just turn this into | |
2301 | * a negative dentry, but if anybody else is | |
2302 | * currently using the dentry or the inode | |
2303 | * we can't do that and we fall back on removing | |
2304 | * it from the hash queues and waiting for | |
2305 | * it to be deleted later when it has no users | |
2306 | */ | |
2307 | ||
2308 | /** | |
2309 | * d_delete - delete a dentry | |
2310 | * @dentry: The dentry to delete | |
2311 | * | |
2312 | * Turn the dentry into a negative dentry if possible, otherwise | |
2313 | * remove it from the hash queues so it can be deleted later | |
2314 | */ | |
2315 | ||
2316 | void d_delete(struct dentry * dentry) | |
2317 | { | |
873feea0 | 2318 | struct inode *inode; |
7a91bf7f | 2319 | int isdir = 0; |
1da177e4 LT |
2320 | /* |
2321 | * Are we the only user? | |
2322 | */ | |
357f8e65 | 2323 | again: |
1da177e4 | 2324 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2325 | inode = dentry->d_inode; |
2326 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2327 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2328 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2329 | spin_unlock(&dentry->d_lock); |
2330 | cpu_relax(); | |
2331 | goto again; | |
2332 | } | |
13e3c5e5 | 2333 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2334 | dentry_unlink_inode(dentry); |
7a91bf7f | 2335 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2336 | return; |
2337 | } | |
2338 | ||
2339 | if (!d_unhashed(dentry)) | |
2340 | __d_drop(dentry); | |
2341 | ||
2342 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2343 | |
2344 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2345 | } |
ec4f8605 | 2346 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2347 | |
b07ad996 | 2348 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2349 | { |
ceb5bdc2 | 2350 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2351 | hlist_bl_lock(b); |
dea3667b | 2352 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2353 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2354 | hlist_bl_unlock(b); |
1da177e4 LT |
2355 | } |
2356 | ||
770bfad8 DH |
2357 | static void _d_rehash(struct dentry * entry) |
2358 | { | |
2359 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2360 | } | |
2361 | ||
1da177e4 LT |
2362 | /** |
2363 | * d_rehash - add an entry back to the hash | |
2364 | * @entry: dentry to add to the hash | |
2365 | * | |
2366 | * Adds a dentry to the hash according to its name. | |
2367 | */ | |
2368 | ||
2369 | void d_rehash(struct dentry * entry) | |
2370 | { | |
1da177e4 | 2371 | spin_lock(&entry->d_lock); |
770bfad8 | 2372 | _d_rehash(entry); |
1da177e4 | 2373 | spin_unlock(&entry->d_lock); |
1da177e4 | 2374 | } |
ec4f8605 | 2375 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2376 | |
fb2d5b86 NP |
2377 | /** |
2378 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2379 | * @dentry: dentry to be updated | |
2380 | * @name: new name | |
2381 | * | |
2382 | * Update a case insensitive dentry with new case of name. | |
2383 | * | |
2384 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2385 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2386 | * lengths). | |
2387 | * | |
2388 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2389 | * keep renames and concurrent inserts, and readdir(2) away). | |
2390 | */ | |
2391 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2392 | { | |
7ebfa57f | 2393 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2394 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2395 | ||
fb2d5b86 | 2396 | spin_lock(&dentry->d_lock); |
31e6b01f | 2397 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2398 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2399 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2400 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2401 | } |
2402 | EXPORT_SYMBOL(dentry_update_name_case); | |
2403 | ||
8d85b484 | 2404 | static void swap_names(struct dentry *dentry, struct dentry *target) |
1da177e4 | 2405 | { |
8d85b484 AV |
2406 | if (unlikely(dname_external(target))) { |
2407 | if (unlikely(dname_external(dentry))) { | |
1da177e4 LT |
2408 | /* |
2409 | * Both external: swap the pointers | |
2410 | */ | |
9a8d5bb4 | 2411 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2412 | } else { |
2413 | /* | |
2414 | * dentry:internal, target:external. Steal target's | |
2415 | * storage and make target internal. | |
2416 | */ | |
321bcf92 BF |
2417 | memcpy(target->d_iname, dentry->d_name.name, |
2418 | dentry->d_name.len + 1); | |
1da177e4 LT |
2419 | dentry->d_name.name = target->d_name.name; |
2420 | target->d_name.name = target->d_iname; | |
2421 | } | |
2422 | } else { | |
8d85b484 | 2423 | if (unlikely(dname_external(dentry))) { |
1da177e4 LT |
2424 | /* |
2425 | * dentry:external, target:internal. Give dentry's | |
2426 | * storage to target and make dentry internal | |
2427 | */ | |
2428 | memcpy(dentry->d_iname, target->d_name.name, | |
2429 | target->d_name.len + 1); | |
2430 | target->d_name.name = dentry->d_name.name; | |
2431 | dentry->d_name.name = dentry->d_iname; | |
2432 | } else { | |
2433 | /* | |
da1ce067 | 2434 | * Both are internal. |
1da177e4 | 2435 | */ |
da1ce067 MS |
2436 | unsigned int i; |
2437 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
2438 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { | |
2439 | swap(((long *) &dentry->d_iname)[i], | |
2440 | ((long *) &target->d_iname)[i]); | |
2441 | } | |
1da177e4 LT |
2442 | } |
2443 | } | |
a28ddb87 | 2444 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
1da177e4 LT |
2445 | } |
2446 | ||
8d85b484 AV |
2447 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2448 | { | |
2449 | struct external_name *old_name = NULL; | |
2450 | if (unlikely(dname_external(dentry))) | |
2451 | old_name = external_name(dentry); | |
2452 | if (unlikely(dname_external(target))) { | |
2453 | atomic_inc(&external_name(target)->u.count); | |
2454 | dentry->d_name = target->d_name; | |
2455 | } else { | |
2456 | memcpy(dentry->d_iname, target->d_name.name, | |
2457 | target->d_name.len + 1); | |
2458 | dentry->d_name.name = dentry->d_iname; | |
2459 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2460 | } | |
2461 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
2462 | kfree_rcu(old_name, u.head); | |
2463 | } | |
2464 | ||
2fd6b7f5 NP |
2465 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2466 | { | |
2467 | /* | |
2468 | * XXXX: do we really need to take target->d_lock? | |
2469 | */ | |
2470 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2471 | spin_lock(&target->d_parent->d_lock); | |
2472 | else { | |
2473 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2474 | spin_lock(&dentry->d_parent->d_lock); | |
2475 | spin_lock_nested(&target->d_parent->d_lock, | |
2476 | DENTRY_D_LOCK_NESTED); | |
2477 | } else { | |
2478 | spin_lock(&target->d_parent->d_lock); | |
2479 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2480 | DENTRY_D_LOCK_NESTED); | |
2481 | } | |
2482 | } | |
2483 | if (target < dentry) { | |
2484 | spin_lock_nested(&target->d_lock, 2); | |
2485 | spin_lock_nested(&dentry->d_lock, 3); | |
2486 | } else { | |
2487 | spin_lock_nested(&dentry->d_lock, 2); | |
2488 | spin_lock_nested(&target->d_lock, 3); | |
2489 | } | |
2490 | } | |
2491 | ||
986c0194 | 2492 | static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target) |
2fd6b7f5 NP |
2493 | { |
2494 | if (target->d_parent != dentry->d_parent) | |
2495 | spin_unlock(&dentry->d_parent->d_lock); | |
2496 | if (target->d_parent != target) | |
2497 | spin_unlock(&target->d_parent->d_lock); | |
986c0194 AV |
2498 | spin_unlock(&target->d_lock); |
2499 | spin_unlock(&dentry->d_lock); | |
2fd6b7f5 NP |
2500 | } |
2501 | ||
1da177e4 | 2502 | /* |
2fd6b7f5 NP |
2503 | * When switching names, the actual string doesn't strictly have to |
2504 | * be preserved in the target - because we're dropping the target | |
2505 | * anyway. As such, we can just do a simple memcpy() to copy over | |
d2fa4a84 ME |
2506 | * the new name before we switch, unless we are going to rehash |
2507 | * it. Note that if we *do* unhash the target, we are not allowed | |
2508 | * to rehash it without giving it a new name/hash key - whether | |
2509 | * we swap or overwrite the names here, resulting name won't match | |
2510 | * the reality in filesystem; it's only there for d_path() purposes. | |
2511 | * Note that all of this is happening under rename_lock, so the | |
2512 | * any hash lookup seeing it in the middle of manipulations will | |
2513 | * be discarded anyway. So we do not care what happens to the hash | |
2514 | * key in that case. | |
1da177e4 | 2515 | */ |
9eaef27b | 2516 | /* |
18367501 | 2517 | * __d_move - move a dentry |
1da177e4 LT |
2518 | * @dentry: entry to move |
2519 | * @target: new dentry | |
da1ce067 | 2520 | * @exchange: exchange the two dentries |
1da177e4 LT |
2521 | * |
2522 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2523 | * dcache entries should not be moved in this way. Caller must hold |
2524 | * rename_lock, the i_mutex of the source and target directories, | |
2525 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2526 | */ |
da1ce067 MS |
2527 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2528 | bool exchange) | |
1da177e4 | 2529 | { |
1da177e4 LT |
2530 | if (!dentry->d_inode) |
2531 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2532 | ||
2fd6b7f5 NP |
2533 | BUG_ON(d_ancestor(dentry, target)); |
2534 | BUG_ON(d_ancestor(target, dentry)); | |
2535 | ||
2fd6b7f5 | 2536 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2537 | |
31e6b01f | 2538 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2539 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2540 | |
ceb5bdc2 NP |
2541 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2542 | ||
2543 | /* | |
2544 | * Move the dentry to the target hash queue. Don't bother checking | |
2545 | * for the same hash queue because of how unlikely it is. | |
2546 | */ | |
2547 | __d_drop(dentry); | |
789680d1 | 2548 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 | 2549 | |
da1ce067 MS |
2550 | /* |
2551 | * Unhash the target (d_delete() is not usable here). If exchanging | |
2552 | * the two dentries, then rehash onto the other's hash queue. | |
2553 | */ | |
1da177e4 | 2554 | __d_drop(target); |
da1ce067 MS |
2555 | if (exchange) { |
2556 | __d_rehash(target, | |
2557 | d_hash(dentry->d_parent, dentry->d_name.hash)); | |
2558 | } | |
1da177e4 | 2559 | |
1da177e4 | 2560 | /* Switch the names.. */ |
8d85b484 AV |
2561 | if (exchange) |
2562 | swap_names(dentry, target); | |
2563 | else | |
2564 | copy_name(dentry, target); | |
1da177e4 | 2565 | |
63cf427a | 2566 | /* ... and switch them in the tree */ |
1da177e4 | 2567 | if (IS_ROOT(dentry)) { |
63cf427a | 2568 | /* splicing a tree */ |
1da177e4 LT |
2569 | dentry->d_parent = target->d_parent; |
2570 | target->d_parent = target; | |
9d8cd306 | 2571 | list_del_init(&target->d_u.d_child); |
63cf427a | 2572 | list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
1da177e4 | 2573 | } else { |
63cf427a | 2574 | /* swapping two dentries */ |
9a8d5bb4 | 2575 | swap(dentry->d_parent, target->d_parent); |
9d8cd306 | 2576 | list_move(&target->d_u.d_child, &target->d_parent->d_subdirs); |
63cf427a AV |
2577 | list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2578 | if (exchange) | |
2579 | fsnotify_d_move(target); | |
2580 | fsnotify_d_move(dentry); | |
1da177e4 LT |
2581 | } |
2582 | ||
31e6b01f NP |
2583 | write_seqcount_end(&target->d_seq); |
2584 | write_seqcount_end(&dentry->d_seq); | |
2585 | ||
986c0194 | 2586 | dentry_unlock_for_move(dentry, target); |
18367501 AV |
2587 | } |
2588 | ||
2589 | /* | |
2590 | * d_move - move a dentry | |
2591 | * @dentry: entry to move | |
2592 | * @target: new dentry | |
2593 | * | |
2594 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2595 | * dcache entries should not be moved in this way. See the locking |
2596 | * requirements for __d_move. | |
18367501 AV |
2597 | */ |
2598 | void d_move(struct dentry *dentry, struct dentry *target) | |
2599 | { | |
2600 | write_seqlock(&rename_lock); | |
da1ce067 | 2601 | __d_move(dentry, target, false); |
1da177e4 | 2602 | write_sequnlock(&rename_lock); |
9eaef27b | 2603 | } |
ec4f8605 | 2604 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2605 | |
da1ce067 MS |
2606 | /* |
2607 | * d_exchange - exchange two dentries | |
2608 | * @dentry1: first dentry | |
2609 | * @dentry2: second dentry | |
2610 | */ | |
2611 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2612 | { | |
2613 | write_seqlock(&rename_lock); | |
2614 | ||
2615 | WARN_ON(!dentry1->d_inode); | |
2616 | WARN_ON(!dentry2->d_inode); | |
2617 | WARN_ON(IS_ROOT(dentry1)); | |
2618 | WARN_ON(IS_ROOT(dentry2)); | |
2619 | ||
2620 | __d_move(dentry1, dentry2, true); | |
2621 | ||
2622 | write_sequnlock(&rename_lock); | |
2623 | } | |
2624 | ||
e2761a11 OH |
2625 | /** |
2626 | * d_ancestor - search for an ancestor | |
2627 | * @p1: ancestor dentry | |
2628 | * @p2: child dentry | |
2629 | * | |
2630 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2631 | * an ancestor of p2, else NULL. | |
9eaef27b | 2632 | */ |
e2761a11 | 2633 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2634 | { |
2635 | struct dentry *p; | |
2636 | ||
871c0067 | 2637 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2638 | if (p->d_parent == p1) |
e2761a11 | 2639 | return p; |
9eaef27b | 2640 | } |
e2761a11 | 2641 | return NULL; |
9eaef27b TM |
2642 | } |
2643 | ||
2644 | /* | |
2645 | * This helper attempts to cope with remotely renamed directories | |
2646 | * | |
2647 | * It assumes that the caller is already holding | |
18367501 | 2648 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2649 | * |
2650 | * Note: If ever the locking in lock_rename() changes, then please | |
2651 | * remember to update this too... | |
9eaef27b | 2652 | */ |
873feea0 NP |
2653 | static struct dentry *__d_unalias(struct inode *inode, |
2654 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2655 | { |
2656 | struct mutex *m1 = NULL, *m2 = NULL; | |
ee3efa91 | 2657 | struct dentry *ret = ERR_PTR(-EBUSY); |
9eaef27b TM |
2658 | |
2659 | /* If alias and dentry share a parent, then no extra locks required */ | |
2660 | if (alias->d_parent == dentry->d_parent) | |
2661 | goto out_unalias; | |
2662 | ||
9eaef27b | 2663 | /* See lock_rename() */ |
9eaef27b TM |
2664 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2665 | goto out_err; | |
2666 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2667 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2668 | goto out_err; | |
2669 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2670 | out_unalias: | |
ee3efa91 | 2671 | if (likely(!d_mountpoint(alias))) { |
da1ce067 | 2672 | __d_move(alias, dentry, false); |
ee3efa91 AV |
2673 | ret = alias; |
2674 | } | |
9eaef27b | 2675 | out_err: |
873feea0 | 2676 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2677 | if (m2) |
2678 | mutex_unlock(m2); | |
2679 | if (m1) | |
2680 | mutex_unlock(m1); | |
2681 | return ret; | |
2682 | } | |
2683 | ||
3f70bd51 BF |
2684 | /** |
2685 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
2686 | * @inode: the inode which may have a disconnected dentry | |
2687 | * @dentry: a negative dentry which we want to point to the inode. | |
2688 | * | |
da093a9b BF |
2689 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
2690 | * place of the given dentry and return it, else simply d_add the inode | |
2691 | * to the dentry and return NULL. | |
3f70bd51 | 2692 | * |
908790fa BF |
2693 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
2694 | * we should error out: directories can't have multiple aliases. | |
2695 | * | |
3f70bd51 BF |
2696 | * This is needed in the lookup routine of any filesystem that is exportable |
2697 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
2698 | * | |
2699 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
2700 | * is returned. This matches the expected return value of ->lookup. | |
2701 | * | |
2702 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
2703 | * In that case, we know that the inode will be a regular file, and also this | |
2704 | * will only occur during atomic_open. So we need to check for the dentry | |
2705 | * being already hashed only in the final case. | |
2706 | */ | |
2707 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
2708 | { | |
2709 | struct dentry *new = NULL; | |
2710 | ||
2711 | if (IS_ERR(inode)) | |
2712 | return ERR_CAST(inode); | |
2713 | ||
2714 | if (inode && S_ISDIR(inode->i_mode)) { | |
2715 | spin_lock(&inode->i_lock); | |
908790fa | 2716 | new = __d_find_any_alias(inode); |
3f70bd51 | 2717 | if (new) { |
da093a9b | 2718 | if (!IS_ROOT(new)) { |
908790fa BF |
2719 | spin_unlock(&inode->i_lock); |
2720 | dput(new); | |
2721 | return ERR_PTR(-EIO); | |
2722 | } | |
95ad5c29 BF |
2723 | if (d_ancestor(new, dentry)) { |
2724 | spin_unlock(&inode->i_lock); | |
2725 | dput(new); | |
2726 | return ERR_PTR(-EIO); | |
2727 | } | |
75a2352d | 2728 | write_seqlock(&rename_lock); |
63cf427a | 2729 | __d_move(new, dentry, false); |
75a2352d | 2730 | write_sequnlock(&rename_lock); |
3f70bd51 BF |
2731 | spin_unlock(&inode->i_lock); |
2732 | security_d_instantiate(new, inode); | |
3f70bd51 BF |
2733 | iput(inode); |
2734 | } else { | |
2735 | /* already taking inode->i_lock, so d_add() by hand */ | |
2736 | __d_instantiate(dentry, inode); | |
2737 | spin_unlock(&inode->i_lock); | |
2738 | security_d_instantiate(dentry, inode); | |
2739 | d_rehash(dentry); | |
2740 | } | |
2741 | } else { | |
2742 | d_instantiate(dentry, inode); | |
2743 | if (d_unhashed(dentry)) | |
2744 | d_rehash(dentry); | |
2745 | } | |
2746 | return new; | |
2747 | } | |
2748 | EXPORT_SYMBOL(d_splice_alias); | |
2749 | ||
770bfad8 DH |
2750 | /** |
2751 | * d_materialise_unique - introduce an inode into the tree | |
2752 | * @dentry: candidate dentry | |
2753 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2754 | * | |
2755 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2756 | * root directory alias in its place if there is one. Caller must hold the |
2757 | * i_mutex of the parent directory. | |
770bfad8 DH |
2758 | */ |
2759 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2760 | { | |
9eaef27b | 2761 | struct dentry *actual; |
770bfad8 DH |
2762 | |
2763 | BUG_ON(!d_unhashed(dentry)); | |
2764 | ||
770bfad8 DH |
2765 | if (!inode) { |
2766 | actual = dentry; | |
360da900 | 2767 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2768 | d_rehash(actual); |
2769 | goto out_nolock; | |
770bfad8 DH |
2770 | } |
2771 | ||
873feea0 | 2772 | spin_lock(&inode->i_lock); |
357f8e65 | 2773 | |
9eaef27b TM |
2774 | if (S_ISDIR(inode->i_mode)) { |
2775 | struct dentry *alias; | |
2776 | ||
2777 | /* Does an aliased dentry already exist? */ | |
52ed46f0 | 2778 | alias = __d_find_alias(inode); |
9eaef27b TM |
2779 | if (alias) { |
2780 | actual = alias; | |
18367501 AV |
2781 | write_seqlock(&rename_lock); |
2782 | ||
2783 | if (d_ancestor(alias, dentry)) { | |
2784 | /* Check for loops */ | |
2785 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2786 | spin_unlock(&inode->i_lock); |
18367501 AV |
2787 | } else if (IS_ROOT(alias)) { |
2788 | /* Is this an anonymous mountpoint that we | |
2789 | * could splice into our tree? */ | |
63cf427a | 2790 | __d_move(alias, dentry, false); |
18367501 | 2791 | write_sequnlock(&rename_lock); |
9eaef27b | 2792 | goto found; |
18367501 AV |
2793 | } else { |
2794 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2795 | * aliasing. This drops inode->i_lock */ |
18367501 | 2796 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2797 | } |
18367501 | 2798 | write_sequnlock(&rename_lock); |
dd179946 DH |
2799 | if (IS_ERR(actual)) { |
2800 | if (PTR_ERR(actual) == -ELOOP) | |
2801 | pr_warn_ratelimited( | |
2802 | "VFS: Lookup of '%s' in %s %s" | |
2803 | " would have caused loop\n", | |
2804 | dentry->d_name.name, | |
2805 | inode->i_sb->s_type->name, | |
2806 | inode->i_sb->s_id); | |
9eaef27b | 2807 | dput(alias); |
dd179946 | 2808 | } |
9eaef27b TM |
2809 | goto out_nolock; |
2810 | } | |
770bfad8 DH |
2811 | } |
2812 | ||
2813 | /* Add a unique reference */ | |
2814 | actual = __d_instantiate_unique(dentry, inode); | |
2815 | if (!actual) | |
2816 | actual = dentry; | |
770bfad8 | 2817 | |
8527dd71 | 2818 | d_rehash(actual); |
5cc3821b | 2819 | found: |
873feea0 | 2820 | spin_unlock(&inode->i_lock); |
9eaef27b | 2821 | out_nolock: |
770bfad8 DH |
2822 | if (actual == dentry) { |
2823 | security_d_instantiate(dentry, inode); | |
2824 | return NULL; | |
2825 | } | |
2826 | ||
2827 | iput(inode); | |
2828 | return actual; | |
770bfad8 | 2829 | } |
ec4f8605 | 2830 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2831 | |
cdd16d02 | 2832 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2833 | { |
2834 | *buflen -= namelen; | |
2835 | if (*buflen < 0) | |
2836 | return -ENAMETOOLONG; | |
2837 | *buffer -= namelen; | |
2838 | memcpy(*buffer, str, namelen); | |
2839 | return 0; | |
2840 | } | |
2841 | ||
232d2d60 WL |
2842 | /** |
2843 | * prepend_name - prepend a pathname in front of current buffer pointer | |
18129977 WL |
2844 | * @buffer: buffer pointer |
2845 | * @buflen: allocated length of the buffer | |
2846 | * @name: name string and length qstr structure | |
232d2d60 WL |
2847 | * |
2848 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
2849 | * make sure that either the old or the new name pointer and length are | |
2850 | * fetched. However, there may be mismatch between length and pointer. | |
2851 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
2852 | * the length is reached or a null byte is found. It also prepends "/" at | |
2853 | * the beginning of the name. The sequence number check at the caller will | |
2854 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
2855 | * due to mismatched pointer and length will be discarded. | |
6d13f694 AV |
2856 | * |
2857 | * Data dependency barrier is needed to make sure that we see that terminating | |
2858 | * NUL. Alpha strikes again, film at 11... | |
232d2d60 | 2859 | */ |
cdd16d02 MS |
2860 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2861 | { | |
232d2d60 WL |
2862 | const char *dname = ACCESS_ONCE(name->name); |
2863 | u32 dlen = ACCESS_ONCE(name->len); | |
2864 | char *p; | |
2865 | ||
6d13f694 AV |
2866 | smp_read_barrier_depends(); |
2867 | ||
232d2d60 | 2868 | *buflen -= dlen + 1; |
e825196d AV |
2869 | if (*buflen < 0) |
2870 | return -ENAMETOOLONG; | |
232d2d60 WL |
2871 | p = *buffer -= dlen + 1; |
2872 | *p++ = '/'; | |
2873 | while (dlen--) { | |
2874 | char c = *dname++; | |
2875 | if (!c) | |
2876 | break; | |
2877 | *p++ = c; | |
2878 | } | |
2879 | return 0; | |
cdd16d02 MS |
2880 | } |
2881 | ||
1da177e4 | 2882 | /** |
208898c1 | 2883 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2884 | * @path: the dentry/vfsmount to report |
02125a82 | 2885 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2886 | * @buffer: pointer to the end of the buffer |
2887 | * @buflen: pointer to buffer length | |
552ce544 | 2888 | * |
18129977 WL |
2889 | * The function will first try to write out the pathname without taking any |
2890 | * lock other than the RCU read lock to make sure that dentries won't go away. | |
2891 | * It only checks the sequence number of the global rename_lock as any change | |
2892 | * in the dentry's d_seq will be preceded by changes in the rename_lock | |
2893 | * sequence number. If the sequence number had been changed, it will restart | |
2894 | * the whole pathname back-tracing sequence again by taking the rename_lock. | |
2895 | * In this case, there is no need to take the RCU read lock as the recursive | |
2896 | * parent pointer references will keep the dentry chain alive as long as no | |
2897 | * rename operation is performed. | |
1da177e4 | 2898 | */ |
02125a82 AV |
2899 | static int prepend_path(const struct path *path, |
2900 | const struct path *root, | |
f2eb6575 | 2901 | char **buffer, int *buflen) |
1da177e4 | 2902 | { |
ede4cebc AV |
2903 | struct dentry *dentry; |
2904 | struct vfsmount *vfsmnt; | |
2905 | struct mount *mnt; | |
f2eb6575 | 2906 | int error = 0; |
48a066e7 | 2907 | unsigned seq, m_seq = 0; |
232d2d60 WL |
2908 | char *bptr; |
2909 | int blen; | |
6092d048 | 2910 | |
48f5ec21 | 2911 | rcu_read_lock(); |
48a066e7 AV |
2912 | restart_mnt: |
2913 | read_seqbegin_or_lock(&mount_lock, &m_seq); | |
2914 | seq = 0; | |
4ec6c2ae | 2915 | rcu_read_lock(); |
232d2d60 WL |
2916 | restart: |
2917 | bptr = *buffer; | |
2918 | blen = *buflen; | |
48a066e7 | 2919 | error = 0; |
ede4cebc AV |
2920 | dentry = path->dentry; |
2921 | vfsmnt = path->mnt; | |
2922 | mnt = real_mount(vfsmnt); | |
232d2d60 | 2923 | read_seqbegin_or_lock(&rename_lock, &seq); |
f2eb6575 | 2924 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2925 | struct dentry * parent; |
2926 | ||
1da177e4 | 2927 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
48a066e7 | 2928 | struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); |
552ce544 | 2929 | /* Global root? */ |
48a066e7 AV |
2930 | if (mnt != parent) { |
2931 | dentry = ACCESS_ONCE(mnt->mnt_mountpoint); | |
2932 | mnt = parent; | |
232d2d60 WL |
2933 | vfsmnt = &mnt->mnt; |
2934 | continue; | |
2935 | } | |
2936 | /* | |
2937 | * Filesystems needing to implement special "root names" | |
2938 | * should do so with ->d_dname() | |
2939 | */ | |
2940 | if (IS_ROOT(dentry) && | |
2941 | (dentry->d_name.len != 1 || | |
2942 | dentry->d_name.name[0] != '/')) { | |
2943 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2944 | (int) dentry->d_name.len, | |
2945 | dentry->d_name.name); | |
2946 | } | |
2947 | if (!error) | |
2948 | error = is_mounted(vfsmnt) ? 1 : 2; | |
2949 | break; | |
1da177e4 LT |
2950 | } |
2951 | parent = dentry->d_parent; | |
2952 | prefetch(parent); | |
232d2d60 | 2953 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
2954 | if (error) |
2955 | break; | |
2956 | ||
1da177e4 LT |
2957 | dentry = parent; |
2958 | } | |
48f5ec21 AV |
2959 | if (!(seq & 1)) |
2960 | rcu_read_unlock(); | |
2961 | if (need_seqretry(&rename_lock, seq)) { | |
2962 | seq = 1; | |
232d2d60 | 2963 | goto restart; |
48f5ec21 AV |
2964 | } |
2965 | done_seqretry(&rename_lock, seq); | |
4ec6c2ae LZ |
2966 | |
2967 | if (!(m_seq & 1)) | |
2968 | rcu_read_unlock(); | |
48a066e7 AV |
2969 | if (need_seqretry(&mount_lock, m_seq)) { |
2970 | m_seq = 1; | |
2971 | goto restart_mnt; | |
2972 | } | |
2973 | done_seqretry(&mount_lock, m_seq); | |
1da177e4 | 2974 | |
232d2d60 WL |
2975 | if (error >= 0 && bptr == *buffer) { |
2976 | if (--blen < 0) | |
2977 | error = -ENAMETOOLONG; | |
2978 | else | |
2979 | *--bptr = '/'; | |
2980 | } | |
2981 | *buffer = bptr; | |
2982 | *buflen = blen; | |
7ea600b5 | 2983 | return error; |
f2eb6575 | 2984 | } |
be285c71 | 2985 | |
f2eb6575 MS |
2986 | /** |
2987 | * __d_path - return the path of a dentry | |
2988 | * @path: the dentry/vfsmount to report | |
02125a82 | 2989 | * @root: root vfsmnt/dentry |
cd956a1c | 2990 | * @buf: buffer to return value in |
f2eb6575 MS |
2991 | * @buflen: buffer length |
2992 | * | |
ffd1f4ed | 2993 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2994 | * |
2995 | * Returns a pointer into the buffer or an error code if the | |
2996 | * path was too long. | |
2997 | * | |
be148247 | 2998 | * "buflen" should be positive. |
f2eb6575 | 2999 | * |
02125a82 | 3000 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 3001 | */ |
02125a82 AV |
3002 | char *__d_path(const struct path *path, |
3003 | const struct path *root, | |
f2eb6575 MS |
3004 | char *buf, int buflen) |
3005 | { | |
3006 | char *res = buf + buflen; | |
3007 | int error; | |
3008 | ||
3009 | prepend(&res, &buflen, "\0", 1); | |
f2eb6575 | 3010 | error = prepend_path(path, root, &res, &buflen); |
be148247 | 3011 | |
02125a82 AV |
3012 | if (error < 0) |
3013 | return ERR_PTR(error); | |
3014 | if (error > 0) | |
3015 | return NULL; | |
3016 | return res; | |
3017 | } | |
3018 | ||
3019 | char *d_absolute_path(const struct path *path, | |
3020 | char *buf, int buflen) | |
3021 | { | |
3022 | struct path root = {}; | |
3023 | char *res = buf + buflen; | |
3024 | int error; | |
3025 | ||
3026 | prepend(&res, &buflen, "\0", 1); | |
02125a82 | 3027 | error = prepend_path(path, &root, &res, &buflen); |
02125a82 AV |
3028 | |
3029 | if (error > 1) | |
3030 | error = -EINVAL; | |
3031 | if (error < 0) | |
f2eb6575 | 3032 | return ERR_PTR(error); |
f2eb6575 | 3033 | return res; |
1da177e4 LT |
3034 | } |
3035 | ||
ffd1f4ed MS |
3036 | /* |
3037 | * same as __d_path but appends "(deleted)" for unlinked files. | |
3038 | */ | |
02125a82 AV |
3039 | static int path_with_deleted(const struct path *path, |
3040 | const struct path *root, | |
3041 | char **buf, int *buflen) | |
ffd1f4ed MS |
3042 | { |
3043 | prepend(buf, buflen, "\0", 1); | |
3044 | if (d_unlinked(path->dentry)) { | |
3045 | int error = prepend(buf, buflen, " (deleted)", 10); | |
3046 | if (error) | |
3047 | return error; | |
3048 | } | |
3049 | ||
3050 | return prepend_path(path, root, buf, buflen); | |
3051 | } | |
3052 | ||
8df9d1a4 MS |
3053 | static int prepend_unreachable(char **buffer, int *buflen) |
3054 | { | |
3055 | return prepend(buffer, buflen, "(unreachable)", 13); | |
3056 | } | |
3057 | ||
68f0d9d9 LT |
3058 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) |
3059 | { | |
3060 | unsigned seq; | |
3061 | ||
3062 | do { | |
3063 | seq = read_seqcount_begin(&fs->seq); | |
3064 | *root = fs->root; | |
3065 | } while (read_seqcount_retry(&fs->seq, seq)); | |
3066 | } | |
3067 | ||
a03a8a70 JB |
3068 | /** |
3069 | * d_path - return the path of a dentry | |
cf28b486 | 3070 | * @path: path to report |
a03a8a70 JB |
3071 | * @buf: buffer to return value in |
3072 | * @buflen: buffer length | |
3073 | * | |
3074 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
3075 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
3076 | * | |
52afeefb AV |
3077 | * Returns a pointer into the buffer or an error code if the path was |
3078 | * too long. Note: Callers should use the returned pointer, not the passed | |
3079 | * in buffer, to use the name! The implementation often starts at an offset | |
3080 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 3081 | * |
31f3e0b3 | 3082 | * "buflen" should be positive. |
a03a8a70 | 3083 | */ |
20d4fdc1 | 3084 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 3085 | { |
ffd1f4ed | 3086 | char *res = buf + buflen; |
6ac08c39 | 3087 | struct path root; |
ffd1f4ed | 3088 | int error; |
1da177e4 | 3089 | |
c23fbb6b ED |
3090 | /* |
3091 | * We have various synthetic filesystems that never get mounted. On | |
3092 | * these filesystems dentries are never used for lookup purposes, and | |
3093 | * thus don't need to be hashed. They also don't need a name until a | |
3094 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
3095 | * below allows us to generate a name for these objects on demand: | |
f48cfddc EB |
3096 | * |
3097 | * Some pseudo inodes are mountable. When they are mounted | |
3098 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname | |
3099 | * and instead have d_path return the mounted path. | |
c23fbb6b | 3100 | */ |
f48cfddc EB |
3101 | if (path->dentry->d_op && path->dentry->d_op->d_dname && |
3102 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) | |
cf28b486 | 3103 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
c23fbb6b | 3104 | |
68f0d9d9 LT |
3105 | rcu_read_lock(); |
3106 | get_fs_root_rcu(current->fs, &root); | |
02125a82 | 3107 | error = path_with_deleted(path, &root, &res, &buflen); |
68f0d9d9 LT |
3108 | rcu_read_unlock(); |
3109 | ||
02125a82 | 3110 | if (error < 0) |
ffd1f4ed | 3111 | res = ERR_PTR(error); |
1da177e4 LT |
3112 | return res; |
3113 | } | |
ec4f8605 | 3114 | EXPORT_SYMBOL(d_path); |
1da177e4 | 3115 | |
c23fbb6b ED |
3116 | /* |
3117 | * Helper function for dentry_operations.d_dname() members | |
3118 | */ | |
3119 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
3120 | const char *fmt, ...) | |
3121 | { | |
3122 | va_list args; | |
3123 | char temp[64]; | |
3124 | int sz; | |
3125 | ||
3126 | va_start(args, fmt); | |
3127 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
3128 | va_end(args); | |
3129 | ||
3130 | if (sz > sizeof(temp) || sz > buflen) | |
3131 | return ERR_PTR(-ENAMETOOLONG); | |
3132 | ||
3133 | buffer += buflen - sz; | |
3134 | return memcpy(buffer, temp, sz); | |
3135 | } | |
3136 | ||
118b2302 AV |
3137 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
3138 | { | |
3139 | char *end = buffer + buflen; | |
3140 | /* these dentries are never renamed, so d_lock is not needed */ | |
3141 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
232d2d60 | 3142 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
118b2302 AV |
3143 | prepend(&end, &buflen, "/", 1)) |
3144 | end = ERR_PTR(-ENAMETOOLONG); | |
232d2d60 | 3145 | return end; |
118b2302 | 3146 | } |
31bbe16f | 3147 | EXPORT_SYMBOL(simple_dname); |
118b2302 | 3148 | |
6092d048 RP |
3149 | /* |
3150 | * Write full pathname from the root of the filesystem into the buffer. | |
3151 | */ | |
f6500801 | 3152 | static char *__dentry_path(struct dentry *d, char *buf, int buflen) |
6092d048 | 3153 | { |
f6500801 | 3154 | struct dentry *dentry; |
232d2d60 WL |
3155 | char *end, *retval; |
3156 | int len, seq = 0; | |
3157 | int error = 0; | |
6092d048 | 3158 | |
f6500801 AV |
3159 | if (buflen < 2) |
3160 | goto Elong; | |
3161 | ||
48f5ec21 | 3162 | rcu_read_lock(); |
232d2d60 | 3163 | restart: |
f6500801 | 3164 | dentry = d; |
232d2d60 WL |
3165 | end = buf + buflen; |
3166 | len = buflen; | |
3167 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
3168 | /* Get '/' right */ |
3169 | retval = end-1; | |
3170 | *retval = '/'; | |
232d2d60 | 3171 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
3172 | while (!IS_ROOT(dentry)) { |
3173 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 3174 | |
6092d048 | 3175 | prefetch(parent); |
232d2d60 WL |
3176 | error = prepend_name(&end, &len, &dentry->d_name); |
3177 | if (error) | |
3178 | break; | |
6092d048 RP |
3179 | |
3180 | retval = end; | |
3181 | dentry = parent; | |
3182 | } | |
48f5ec21 AV |
3183 | if (!(seq & 1)) |
3184 | rcu_read_unlock(); | |
3185 | if (need_seqretry(&rename_lock, seq)) { | |
3186 | seq = 1; | |
232d2d60 | 3187 | goto restart; |
48f5ec21 AV |
3188 | } |
3189 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3190 | if (error) |
3191 | goto Elong; | |
c103135c AV |
3192 | return retval; |
3193 | Elong: | |
3194 | return ERR_PTR(-ENAMETOOLONG); | |
3195 | } | |
ec2447c2 NP |
3196 | |
3197 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3198 | { | |
232d2d60 | 3199 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3200 | } |
3201 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3202 | |
3203 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3204 | { | |
3205 | char *p = NULL; | |
3206 | char *retval; | |
3207 | ||
c103135c AV |
3208 | if (d_unlinked(dentry)) { |
3209 | p = buf + buflen; | |
3210 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3211 | goto Elong; | |
3212 | buflen++; | |
3213 | } | |
3214 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3215 | if (!IS_ERR(retval) && p) |
3216 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3217 | return retval; |
3218 | Elong: | |
6092d048 RP |
3219 | return ERR_PTR(-ENAMETOOLONG); |
3220 | } | |
3221 | ||
8b19e341 LT |
3222 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, |
3223 | struct path *pwd) | |
5762482f | 3224 | { |
8b19e341 LT |
3225 | unsigned seq; |
3226 | ||
3227 | do { | |
3228 | seq = read_seqcount_begin(&fs->seq); | |
3229 | *root = fs->root; | |
3230 | *pwd = fs->pwd; | |
3231 | } while (read_seqcount_retry(&fs->seq, seq)); | |
5762482f LT |
3232 | } |
3233 | ||
1da177e4 LT |
3234 | /* |
3235 | * NOTE! The user-level library version returns a | |
3236 | * character pointer. The kernel system call just | |
3237 | * returns the length of the buffer filled (which | |
3238 | * includes the ending '\0' character), or a negative | |
3239 | * error value. So libc would do something like | |
3240 | * | |
3241 | * char *getcwd(char * buf, size_t size) | |
3242 | * { | |
3243 | * int retval; | |
3244 | * | |
3245 | * retval = sys_getcwd(buf, size); | |
3246 | * if (retval >= 0) | |
3247 | * return buf; | |
3248 | * errno = -retval; | |
3249 | * return NULL; | |
3250 | * } | |
3251 | */ | |
3cdad428 | 3252 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3253 | { |
552ce544 | 3254 | int error; |
6ac08c39 | 3255 | struct path pwd, root; |
3272c544 | 3256 | char *page = __getname(); |
1da177e4 LT |
3257 | |
3258 | if (!page) | |
3259 | return -ENOMEM; | |
3260 | ||
8b19e341 LT |
3261 | rcu_read_lock(); |
3262 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); | |
1da177e4 | 3263 | |
552ce544 | 3264 | error = -ENOENT; |
f3da392e | 3265 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3266 | unsigned long len; |
3272c544 LT |
3267 | char *cwd = page + PATH_MAX; |
3268 | int buflen = PATH_MAX; | |
1da177e4 | 3269 | |
8df9d1a4 | 3270 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3271 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
ff812d72 | 3272 | rcu_read_unlock(); |
552ce544 | 3273 | |
02125a82 | 3274 | if (error < 0) |
552ce544 LT |
3275 | goto out; |
3276 | ||
8df9d1a4 | 3277 | /* Unreachable from current root */ |
02125a82 | 3278 | if (error > 0) { |
8df9d1a4 MS |
3279 | error = prepend_unreachable(&cwd, &buflen); |
3280 | if (error) | |
3281 | goto out; | |
3282 | } | |
3283 | ||
552ce544 | 3284 | error = -ERANGE; |
3272c544 | 3285 | len = PATH_MAX + page - cwd; |
552ce544 LT |
3286 | if (len <= size) { |
3287 | error = len; | |
3288 | if (copy_to_user(buf, cwd, len)) | |
3289 | error = -EFAULT; | |
3290 | } | |
949854d0 | 3291 | } else { |
ff812d72 | 3292 | rcu_read_unlock(); |
949854d0 | 3293 | } |
1da177e4 LT |
3294 | |
3295 | out: | |
3272c544 | 3296 | __putname(page); |
1da177e4 LT |
3297 | return error; |
3298 | } | |
3299 | ||
3300 | /* | |
3301 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3302 | * | |
3303 | * Trivially implemented using the dcache structure | |
3304 | */ | |
3305 | ||
3306 | /** | |
3307 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3308 | * @new_dentry: new dentry | |
3309 | * @old_dentry: old dentry | |
3310 | * | |
3311 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
3312 | * Returns 0 otherwise. | |
3313 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
3314 | */ | |
3315 | ||
e2761a11 | 3316 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
3317 | { |
3318 | int result; | |
949854d0 | 3319 | unsigned seq; |
1da177e4 | 3320 | |
e2761a11 OH |
3321 | if (new_dentry == old_dentry) |
3322 | return 1; | |
3323 | ||
e2761a11 | 3324 | do { |
1da177e4 | 3325 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3326 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3327 | /* |
3328 | * Need rcu_readlock to protect against the d_parent trashing | |
3329 | * due to d_move | |
3330 | */ | |
3331 | rcu_read_lock(); | |
e2761a11 | 3332 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 3333 | result = 1; |
e2761a11 OH |
3334 | else |
3335 | result = 0; | |
949854d0 | 3336 | rcu_read_unlock(); |
1da177e4 | 3337 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3338 | |
3339 | return result; | |
3340 | } | |
3341 | ||
db14fc3a | 3342 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3343 | { |
db14fc3a MS |
3344 | struct dentry *root = data; |
3345 | if (dentry != root) { | |
3346 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3347 | return D_WALK_SKIP; | |
1da177e4 | 3348 | |
01ddc4ed MS |
3349 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3350 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3351 | dentry->d_lockref.count--; | |
3352 | } | |
1da177e4 | 3353 | } |
db14fc3a MS |
3354 | return D_WALK_CONTINUE; |
3355 | } | |
58db63d0 | 3356 | |
db14fc3a MS |
3357 | void d_genocide(struct dentry *parent) |
3358 | { | |
3359 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3360 | } |
3361 | ||
60545d0d | 3362 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3363 | { |
60545d0d AV |
3364 | inode_dec_link_count(inode); |
3365 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
3366 | !hlist_unhashed(&dentry->d_alias) || | |
3367 | !d_unlinked(dentry)); | |
3368 | spin_lock(&dentry->d_parent->d_lock); | |
3369 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3370 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3371 | (unsigned long long)inode->i_ino); | |
3372 | spin_unlock(&dentry->d_lock); | |
3373 | spin_unlock(&dentry->d_parent->d_lock); | |
3374 | d_instantiate(dentry, inode); | |
1da177e4 | 3375 | } |
60545d0d | 3376 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3377 | |
3378 | static __initdata unsigned long dhash_entries; | |
3379 | static int __init set_dhash_entries(char *str) | |
3380 | { | |
3381 | if (!str) | |
3382 | return 0; | |
3383 | dhash_entries = simple_strtoul(str, &str, 0); | |
3384 | return 1; | |
3385 | } | |
3386 | __setup("dhash_entries=", set_dhash_entries); | |
3387 | ||
3388 | static void __init dcache_init_early(void) | |
3389 | { | |
074b8517 | 3390 | unsigned int loop; |
1da177e4 LT |
3391 | |
3392 | /* If hashes are distributed across NUMA nodes, defer | |
3393 | * hash allocation until vmalloc space is available. | |
3394 | */ | |
3395 | if (hashdist) | |
3396 | return; | |
3397 | ||
3398 | dentry_hashtable = | |
3399 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3400 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3401 | dhash_entries, |
3402 | 13, | |
3403 | HASH_EARLY, | |
3404 | &d_hash_shift, | |
3405 | &d_hash_mask, | |
31fe62b9 | 3406 | 0, |
1da177e4 LT |
3407 | 0); |
3408 | ||
074b8517 | 3409 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3410 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3411 | } |
3412 | ||
74bf17cf | 3413 | static void __init dcache_init(void) |
1da177e4 | 3414 | { |
074b8517 | 3415 | unsigned int loop; |
1da177e4 LT |
3416 | |
3417 | /* | |
3418 | * A constructor could be added for stable state like the lists, | |
3419 | * but it is probably not worth it because of the cache nature | |
3420 | * of the dcache. | |
3421 | */ | |
0a31bd5f CL |
3422 | dentry_cache = KMEM_CACHE(dentry, |
3423 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3424 | |
3425 | /* Hash may have been set up in dcache_init_early */ | |
3426 | if (!hashdist) | |
3427 | return; | |
3428 | ||
3429 | dentry_hashtable = | |
3430 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3431 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3432 | dhash_entries, |
3433 | 13, | |
3434 | 0, | |
3435 | &d_hash_shift, | |
3436 | &d_hash_mask, | |
31fe62b9 | 3437 | 0, |
1da177e4 LT |
3438 | 0); |
3439 | ||
074b8517 | 3440 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3441 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3442 | } |
3443 | ||
3444 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3445 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3446 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3447 | |
1da177e4 LT |
3448 | EXPORT_SYMBOL(d_genocide); |
3449 | ||
1da177e4 LT |
3450 | void __init vfs_caches_init_early(void) |
3451 | { | |
3452 | dcache_init_early(); | |
3453 | inode_init_early(); | |
3454 | } | |
3455 | ||
3456 | void __init vfs_caches_init(unsigned long mempages) | |
3457 | { | |
3458 | unsigned long reserve; | |
3459 | ||
3460 | /* Base hash sizes on available memory, with a reserve equal to | |
3461 | 150% of current kernel size */ | |
3462 | ||
3463 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3464 | mempages -= reserve; | |
3465 | ||
3466 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3467 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3468 | |
74bf17cf DC |
3469 | dcache_init(); |
3470 | inode_init(); | |
1da177e4 | 3471 | files_init(mempages); |
74bf17cf | 3472 | mnt_init(); |
1da177e4 LT |
3473 | bdev_cache_init(); |
3474 | chrdev_init(); | |
3475 | } |