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 | |
bafc9b75 EB |
653 | * possible. If there are reasons not to delete it |
654 | * return -EBUSY. On success return 0. | |
1da177e4 LT |
655 | * |
656 | * no dcache lock. | |
657 | */ | |
658 | ||
659 | int d_invalidate(struct dentry * dentry) | |
660 | { | |
661 | /* | |
662 | * If it's already been dropped, return OK. | |
663 | */ | |
da502956 | 664 | spin_lock(&dentry->d_lock); |
1da177e4 | 665 | if (d_unhashed(dentry)) { |
da502956 | 666 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
667 | return 0; |
668 | } | |
1da177e4 | 669 | spin_unlock(&dentry->d_lock); |
bafc9b75 EB |
670 | |
671 | return check_submounts_and_drop(dentry); | |
1da177e4 | 672 | } |
ec4f8605 | 673 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 674 | |
b5c84bf6 | 675 | /* This must be called with d_lock held */ |
dc0474be | 676 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 677 | { |
98474236 | 678 | dentry->d_lockref.count++; |
23044507 NP |
679 | } |
680 | ||
dc0474be | 681 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 682 | { |
98474236 | 683 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
684 | } |
685 | ||
b7ab39f6 NP |
686 | struct dentry *dget_parent(struct dentry *dentry) |
687 | { | |
df3d0bbc | 688 | int gotref; |
b7ab39f6 NP |
689 | struct dentry *ret; |
690 | ||
df3d0bbc WL |
691 | /* |
692 | * Do optimistic parent lookup without any | |
693 | * locking. | |
694 | */ | |
695 | rcu_read_lock(); | |
696 | ret = ACCESS_ONCE(dentry->d_parent); | |
697 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
698 | rcu_read_unlock(); | |
699 | if (likely(gotref)) { | |
700 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
701 | return ret; | |
702 | dput(ret); | |
703 | } | |
704 | ||
b7ab39f6 | 705 | repeat: |
a734eb45 NP |
706 | /* |
707 | * Don't need rcu_dereference because we re-check it was correct under | |
708 | * the lock. | |
709 | */ | |
710 | rcu_read_lock(); | |
b7ab39f6 | 711 | ret = dentry->d_parent; |
a734eb45 NP |
712 | spin_lock(&ret->d_lock); |
713 | if (unlikely(ret != dentry->d_parent)) { | |
714 | spin_unlock(&ret->d_lock); | |
715 | rcu_read_unlock(); | |
b7ab39f6 NP |
716 | goto repeat; |
717 | } | |
a734eb45 | 718 | rcu_read_unlock(); |
98474236 WL |
719 | BUG_ON(!ret->d_lockref.count); |
720 | ret->d_lockref.count++; | |
b7ab39f6 | 721 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
722 | return ret; |
723 | } | |
724 | EXPORT_SYMBOL(dget_parent); | |
725 | ||
1da177e4 LT |
726 | /** |
727 | * d_find_alias - grab a hashed alias of inode | |
728 | * @inode: inode in question | |
1da177e4 LT |
729 | * |
730 | * If inode has a hashed alias, or is a directory and has any alias, | |
731 | * acquire the reference to alias and return it. Otherwise return NULL. | |
732 | * Notice that if inode is a directory there can be only one alias and | |
733 | * it can be unhashed only if it has no children, or if it is the root | |
3ccb354d EB |
734 | * of a filesystem, or if the directory was renamed and d_revalidate |
735 | * was the first vfs operation to notice. | |
1da177e4 | 736 | * |
21c0d8fd | 737 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
52ed46f0 | 738 | * any other hashed alias over that one. |
1da177e4 | 739 | */ |
52ed46f0 | 740 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 741 | { |
da502956 | 742 | struct dentry *alias, *discon_alias; |
1da177e4 | 743 | |
da502956 NP |
744 | again: |
745 | discon_alias = NULL; | |
b67bfe0d | 746 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
da502956 | 747 | spin_lock(&alias->d_lock); |
1da177e4 | 748 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 749 | if (IS_ROOT(alias) && |
da502956 | 750 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 751 | discon_alias = alias; |
52ed46f0 | 752 | } else { |
dc0474be | 753 | __dget_dlock(alias); |
da502956 NP |
754 | spin_unlock(&alias->d_lock); |
755 | return alias; | |
756 | } | |
757 | } | |
758 | spin_unlock(&alias->d_lock); | |
759 | } | |
760 | if (discon_alias) { | |
761 | alias = discon_alias; | |
762 | spin_lock(&alias->d_lock); | |
763 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
8d80d7da BF |
764 | __dget_dlock(alias); |
765 | spin_unlock(&alias->d_lock); | |
766 | return alias; | |
1da177e4 | 767 | } |
da502956 NP |
768 | spin_unlock(&alias->d_lock); |
769 | goto again; | |
1da177e4 | 770 | } |
da502956 | 771 | return NULL; |
1da177e4 LT |
772 | } |
773 | ||
da502956 | 774 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 775 | { |
214fda1f DH |
776 | struct dentry *de = NULL; |
777 | ||
b3d9b7a3 | 778 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 779 | spin_lock(&inode->i_lock); |
52ed46f0 | 780 | de = __d_find_alias(inode); |
873feea0 | 781 | spin_unlock(&inode->i_lock); |
214fda1f | 782 | } |
1da177e4 LT |
783 | return de; |
784 | } | |
ec4f8605 | 785 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
786 | |
787 | /* | |
788 | * Try to kill dentries associated with this inode. | |
789 | * WARNING: you must own a reference to inode. | |
790 | */ | |
791 | void d_prune_aliases(struct inode *inode) | |
792 | { | |
0cdca3f9 | 793 | struct dentry *dentry; |
1da177e4 | 794 | restart: |
873feea0 | 795 | spin_lock(&inode->i_lock); |
b67bfe0d | 796 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 797 | spin_lock(&dentry->d_lock); |
98474236 | 798 | if (!dentry->d_lockref.count) { |
590fb51f YZ |
799 | /* |
800 | * inform the fs via d_prune that this dentry | |
801 | * is about to be unhashed and destroyed. | |
802 | */ | |
803 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && | |
804 | !d_unhashed(dentry)) | |
805 | dentry->d_op->d_prune(dentry); | |
806 | ||
dc0474be | 807 | __dget_dlock(dentry); |
1da177e4 LT |
808 | __d_drop(dentry); |
809 | spin_unlock(&dentry->d_lock); | |
873feea0 | 810 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
811 | dput(dentry); |
812 | goto restart; | |
813 | } | |
814 | spin_unlock(&dentry->d_lock); | |
815 | } | |
873feea0 | 816 | spin_unlock(&inode->i_lock); |
1da177e4 | 817 | } |
ec4f8605 | 818 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 819 | |
3049cfe2 | 820 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 821 | { |
5c47e6d0 | 822 | struct dentry *dentry, *parent; |
da3bbdd4 | 823 | |
60942f2f | 824 | while (!list_empty(list)) { |
ff2fde99 | 825 | struct inode *inode; |
60942f2f | 826 | dentry = list_entry(list->prev, struct dentry, d_lru); |
ec33679d | 827 | spin_lock(&dentry->d_lock); |
046b961b AV |
828 | parent = lock_parent(dentry); |
829 | ||
dd1f6b2e DC |
830 | /* |
831 | * The dispose list is isolated and dentries are not accounted | |
832 | * to the LRU here, so we can simply remove it from the list | |
833 | * here regardless of whether it is referenced or not. | |
834 | */ | |
89dc77bc | 835 | d_shrink_del(dentry); |
dd1f6b2e | 836 | |
1da177e4 LT |
837 | /* |
838 | * We found an inuse dentry which was not removed from | |
dd1f6b2e | 839 | * the LRU because of laziness during lookup. Do not free it. |
1da177e4 | 840 | */ |
41edf278 | 841 | if ((int)dentry->d_lockref.count > 0) { |
da3bbdd4 | 842 | spin_unlock(&dentry->d_lock); |
046b961b AV |
843 | if (parent) |
844 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
845 | continue; |
846 | } | |
77812a1e | 847 | |
64fd72e0 AV |
848 | |
849 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) { | |
850 | bool can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
851 | spin_unlock(&dentry->d_lock); | |
046b961b AV |
852 | if (parent) |
853 | spin_unlock(&parent->d_lock); | |
64fd72e0 AV |
854 | if (can_free) |
855 | dentry_free(dentry); | |
856 | continue; | |
857 | } | |
858 | ||
ff2fde99 AV |
859 | inode = dentry->d_inode; |
860 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
89dc77bc | 861 | d_shrink_add(dentry, list); |
dd1f6b2e | 862 | spin_unlock(&dentry->d_lock); |
046b961b AV |
863 | if (parent) |
864 | spin_unlock(&parent->d_lock); | |
5c47e6d0 | 865 | continue; |
dd1f6b2e | 866 | } |
ff2fde99 | 867 | |
ff2fde99 | 868 | __dentry_kill(dentry); |
046b961b | 869 | |
5c47e6d0 AV |
870 | /* |
871 | * We need to prune ancestors too. This is necessary to prevent | |
872 | * quadratic behavior of shrink_dcache_parent(), but is also | |
873 | * expected to be beneficial in reducing dentry cache | |
874 | * fragmentation. | |
875 | */ | |
876 | dentry = parent; | |
b2b80195 AV |
877 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) { |
878 | parent = lock_parent(dentry); | |
879 | if (dentry->d_lockref.count != 1) { | |
880 | dentry->d_lockref.count--; | |
881 | spin_unlock(&dentry->d_lock); | |
882 | if (parent) | |
883 | spin_unlock(&parent->d_lock); | |
884 | break; | |
885 | } | |
886 | inode = dentry->d_inode; /* can't be NULL */ | |
887 | if (unlikely(!spin_trylock(&inode->i_lock))) { | |
888 | spin_unlock(&dentry->d_lock); | |
889 | if (parent) | |
890 | spin_unlock(&parent->d_lock); | |
891 | cpu_relax(); | |
892 | continue; | |
893 | } | |
894 | __dentry_kill(dentry); | |
895 | dentry = parent; | |
896 | } | |
da3bbdd4 | 897 | } |
3049cfe2 CH |
898 | } |
899 | ||
f6041567 DC |
900 | static enum lru_status |
901 | dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg) | |
902 | { | |
903 | struct list_head *freeable = arg; | |
904 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
905 | ||
906 | ||
907 | /* | |
908 | * we are inverting the lru lock/dentry->d_lock here, | |
909 | * so use a trylock. If we fail to get the lock, just skip | |
910 | * it | |
911 | */ | |
912 | if (!spin_trylock(&dentry->d_lock)) | |
913 | return LRU_SKIP; | |
914 | ||
915 | /* | |
916 | * Referenced dentries are still in use. If they have active | |
917 | * counts, just remove them from the LRU. Otherwise give them | |
918 | * another pass through the LRU. | |
919 | */ | |
920 | if (dentry->d_lockref.count) { | |
89dc77bc | 921 | d_lru_isolate(dentry); |
f6041567 DC |
922 | spin_unlock(&dentry->d_lock); |
923 | return LRU_REMOVED; | |
924 | } | |
925 | ||
926 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
927 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
928 | spin_unlock(&dentry->d_lock); | |
929 | ||
930 | /* | |
931 | * The list move itself will be made by the common LRU code. At | |
932 | * this point, we've dropped the dentry->d_lock but keep the | |
933 | * lru lock. This is safe to do, since every list movement is | |
934 | * protected by the lru lock even if both locks are held. | |
935 | * | |
936 | * This is guaranteed by the fact that all LRU management | |
937 | * functions are intermediated by the LRU API calls like | |
938 | * list_lru_add and list_lru_del. List movement in this file | |
939 | * only ever occur through this functions or through callbacks | |
940 | * like this one, that are called from the LRU API. | |
941 | * | |
942 | * The only exceptions to this are functions like | |
943 | * shrink_dentry_list, and code that first checks for the | |
944 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
945 | * operating only with stack provided lists after they are | |
946 | * properly isolated from the main list. It is thus, always a | |
947 | * local access. | |
948 | */ | |
949 | return LRU_ROTATE; | |
950 | } | |
951 | ||
89dc77bc | 952 | d_lru_shrink_move(dentry, freeable); |
f6041567 DC |
953 | spin_unlock(&dentry->d_lock); |
954 | ||
955 | return LRU_REMOVED; | |
956 | } | |
957 | ||
3049cfe2 | 958 | /** |
b48f03b3 DC |
959 | * prune_dcache_sb - shrink the dcache |
960 | * @sb: superblock | |
f6041567 | 961 | * @nr_to_scan : number of entries to try to free |
9b17c623 | 962 | * @nid: which node to scan for freeable entities |
b48f03b3 | 963 | * |
f6041567 | 964 | * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is |
b48f03b3 DC |
965 | * done when we need more memory an called from the superblock shrinker |
966 | * function. | |
3049cfe2 | 967 | * |
b48f03b3 DC |
968 | * This function may fail to free any resources if all the dentries are in |
969 | * use. | |
3049cfe2 | 970 | */ |
9b17c623 DC |
971 | long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan, |
972 | int nid) | |
3049cfe2 | 973 | { |
f6041567 DC |
974 | LIST_HEAD(dispose); |
975 | long freed; | |
3049cfe2 | 976 | |
9b17c623 DC |
977 | freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate, |
978 | &dispose, &nr_to_scan); | |
f6041567 | 979 | shrink_dentry_list(&dispose); |
0a234c6d | 980 | return freed; |
da3bbdd4 | 981 | } |
23044507 | 982 | |
4e717f5c GC |
983 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
984 | spinlock_t *lru_lock, void *arg) | |
dd1f6b2e | 985 | { |
4e717f5c GC |
986 | struct list_head *freeable = arg; |
987 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 988 | |
4e717f5c GC |
989 | /* |
990 | * we are inverting the lru lock/dentry->d_lock here, | |
991 | * so use a trylock. If we fail to get the lock, just skip | |
992 | * it | |
993 | */ | |
994 | if (!spin_trylock(&dentry->d_lock)) | |
995 | return LRU_SKIP; | |
996 | ||
89dc77bc | 997 | d_lru_shrink_move(dentry, freeable); |
4e717f5c | 998 | spin_unlock(&dentry->d_lock); |
ec33679d | 999 | |
4e717f5c | 1000 | return LRU_REMOVED; |
da3bbdd4 KM |
1001 | } |
1002 | ||
4e717f5c | 1003 | |
1da177e4 LT |
1004 | /** |
1005 | * shrink_dcache_sb - shrink dcache for a superblock | |
1006 | * @sb: superblock | |
1007 | * | |
3049cfe2 CH |
1008 | * Shrink the dcache for the specified super block. This is used to free |
1009 | * the dcache before unmounting a file system. | |
1da177e4 | 1010 | */ |
3049cfe2 | 1011 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1012 | { |
4e717f5c GC |
1013 | long freed; |
1014 | ||
1015 | do { | |
1016 | LIST_HEAD(dispose); | |
1017 | ||
1018 | freed = list_lru_walk(&sb->s_dentry_lru, | |
1019 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); | |
3049cfe2 | 1020 | |
4e717f5c GC |
1021 | this_cpu_sub(nr_dentry_unused, freed); |
1022 | shrink_dentry_list(&dispose); | |
1023 | } while (freed > 0); | |
1da177e4 | 1024 | } |
ec4f8605 | 1025 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1026 | |
db14fc3a MS |
1027 | /** |
1028 | * enum d_walk_ret - action to talke during tree walk | |
1029 | * @D_WALK_CONTINUE: contrinue walk | |
1030 | * @D_WALK_QUIT: quit walk | |
1031 | * @D_WALK_NORETRY: quit when retry is needed | |
1032 | * @D_WALK_SKIP: skip this dentry and its children | |
1033 | */ | |
1034 | enum d_walk_ret { | |
1035 | D_WALK_CONTINUE, | |
1036 | D_WALK_QUIT, | |
1037 | D_WALK_NORETRY, | |
1038 | D_WALK_SKIP, | |
1039 | }; | |
c826cb7d | 1040 | |
1da177e4 | 1041 | /** |
db14fc3a MS |
1042 | * d_walk - walk the dentry tree |
1043 | * @parent: start of walk | |
1044 | * @data: data passed to @enter() and @finish() | |
1045 | * @enter: callback when first entering the dentry | |
1046 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1047 | * |
db14fc3a | 1048 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1049 | */ |
db14fc3a MS |
1050 | static void d_walk(struct dentry *parent, void *data, |
1051 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1052 | void (*finish)(void *)) | |
1da177e4 | 1053 | { |
949854d0 | 1054 | struct dentry *this_parent; |
1da177e4 | 1055 | struct list_head *next; |
48f5ec21 | 1056 | unsigned seq = 0; |
db14fc3a MS |
1057 | enum d_walk_ret ret; |
1058 | bool retry = true; | |
949854d0 | 1059 | |
58db63d0 | 1060 | again: |
48f5ec21 | 1061 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1062 | this_parent = parent; |
2fd6b7f5 | 1063 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1064 | |
1065 | ret = enter(data, this_parent); | |
1066 | switch (ret) { | |
1067 | case D_WALK_CONTINUE: | |
1068 | break; | |
1069 | case D_WALK_QUIT: | |
1070 | case D_WALK_SKIP: | |
1071 | goto out_unlock; | |
1072 | case D_WALK_NORETRY: | |
1073 | retry = false; | |
1074 | break; | |
1075 | } | |
1da177e4 LT |
1076 | repeat: |
1077 | next = this_parent->d_subdirs.next; | |
1078 | resume: | |
1079 | while (next != &this_parent->d_subdirs) { | |
1080 | struct list_head *tmp = next; | |
5160ee6f | 1081 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1082 | next = tmp->next; |
2fd6b7f5 NP |
1083 | |
1084 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
db14fc3a MS |
1085 | |
1086 | ret = enter(data, dentry); | |
1087 | switch (ret) { | |
1088 | case D_WALK_CONTINUE: | |
1089 | break; | |
1090 | case D_WALK_QUIT: | |
2fd6b7f5 | 1091 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1092 | goto out_unlock; |
1093 | case D_WALK_NORETRY: | |
1094 | retry = false; | |
1095 | break; | |
1096 | case D_WALK_SKIP: | |
1097 | spin_unlock(&dentry->d_lock); | |
1098 | continue; | |
2fd6b7f5 | 1099 | } |
db14fc3a | 1100 | |
1da177e4 | 1101 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1102 | spin_unlock(&this_parent->d_lock); |
1103 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1104 | this_parent = dentry; |
2fd6b7f5 | 1105 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1106 | goto repeat; |
1107 | } | |
2fd6b7f5 | 1108 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1109 | } |
1110 | /* | |
1111 | * All done at this level ... ascend and resume the search. | |
1112 | */ | |
1113 | if (this_parent != parent) { | |
c826cb7d | 1114 | struct dentry *child = this_parent; |
31dec132 AV |
1115 | this_parent = child->d_parent; |
1116 | ||
1117 | rcu_read_lock(); | |
1118 | spin_unlock(&child->d_lock); | |
1119 | spin_lock(&this_parent->d_lock); | |
1120 | ||
1121 | /* | |
1122 | * might go back up the wrong parent if we have had a rename | |
1123 | * or deletion | |
1124 | */ | |
1125 | if (this_parent != child->d_parent || | |
1126 | (child->d_flags & DCACHE_DENTRY_KILLED) || | |
1127 | need_seqretry(&rename_lock, seq)) { | |
1128 | spin_unlock(&this_parent->d_lock); | |
1129 | rcu_read_unlock(); | |
949854d0 | 1130 | goto rename_retry; |
31dec132 AV |
1131 | } |
1132 | rcu_read_unlock(); | |
949854d0 | 1133 | next = child->d_u.d_child.next; |
1da177e4 LT |
1134 | goto resume; |
1135 | } | |
48f5ec21 | 1136 | if (need_seqretry(&rename_lock, seq)) { |
db14fc3a | 1137 | spin_unlock(&this_parent->d_lock); |
949854d0 | 1138 | goto rename_retry; |
db14fc3a MS |
1139 | } |
1140 | if (finish) | |
1141 | finish(data); | |
1142 | ||
1143 | out_unlock: | |
1144 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1145 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1146 | return; |
58db63d0 NP |
1147 | |
1148 | rename_retry: | |
db14fc3a MS |
1149 | if (!retry) |
1150 | return; | |
48f5ec21 | 1151 | seq = 1; |
58db63d0 | 1152 | goto again; |
1da177e4 | 1153 | } |
db14fc3a MS |
1154 | |
1155 | /* | |
1156 | * Search for at least 1 mount point in the dentry's subdirs. | |
1157 | * We descend to the next level whenever the d_subdirs | |
1158 | * list is non-empty and continue searching. | |
1159 | */ | |
1160 | ||
db14fc3a MS |
1161 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) |
1162 | { | |
1163 | int *ret = data; | |
1164 | if (d_mountpoint(dentry)) { | |
1165 | *ret = 1; | |
1166 | return D_WALK_QUIT; | |
1167 | } | |
1168 | return D_WALK_CONTINUE; | |
1169 | } | |
1170 | ||
69c88dc7 RD |
1171 | /** |
1172 | * have_submounts - check for mounts over a dentry | |
1173 | * @parent: dentry to check. | |
1174 | * | |
1175 | * Return true if the parent or its subdirectories contain | |
1176 | * a mount point | |
1177 | */ | |
db14fc3a MS |
1178 | int have_submounts(struct dentry *parent) |
1179 | { | |
1180 | int ret = 0; | |
1181 | ||
1182 | d_walk(parent, &ret, check_mount, NULL); | |
1183 | ||
1184 | return ret; | |
1185 | } | |
ec4f8605 | 1186 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1187 | |
eed81007 MS |
1188 | /* |
1189 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1190 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1191 | * subtree can become unreachable). | |
1192 | * | |
1193 | * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For | |
1194 | * this reason take rename_lock and d_lock on dentry and ancestors. | |
1195 | */ | |
1196 | int d_set_mounted(struct dentry *dentry) | |
1197 | { | |
1198 | struct dentry *p; | |
1199 | int ret = -ENOENT; | |
1200 | write_seqlock(&rename_lock); | |
1201 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1202 | /* Need exclusion wrt. check_submounts_and_drop() */ | |
1203 | spin_lock(&p->d_lock); | |
1204 | if (unlikely(d_unhashed(p))) { | |
1205 | spin_unlock(&p->d_lock); | |
1206 | goto out; | |
1207 | } | |
1208 | spin_unlock(&p->d_lock); | |
1209 | } | |
1210 | spin_lock(&dentry->d_lock); | |
1211 | if (!d_unlinked(dentry)) { | |
1212 | dentry->d_flags |= DCACHE_MOUNTED; | |
1213 | ret = 0; | |
1214 | } | |
1215 | spin_unlock(&dentry->d_lock); | |
1216 | out: | |
1217 | write_sequnlock(&rename_lock); | |
1218 | return ret; | |
1219 | } | |
1220 | ||
1da177e4 | 1221 | /* |
fd517909 | 1222 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1223 | * and move any unused dentries to the end of the unused |
1224 | * list for prune_dcache(). We descend to the next level | |
1225 | * whenever the d_subdirs list is non-empty and continue | |
1226 | * searching. | |
1227 | * | |
1228 | * It returns zero iff there are no unused children, | |
1229 | * otherwise it returns the number of children moved to | |
1230 | * the end of the unused list. This may not be the total | |
1231 | * number of unused children, because select_parent can | |
1232 | * drop the lock and return early due to latency | |
1233 | * constraints. | |
1234 | */ | |
1da177e4 | 1235 | |
db14fc3a MS |
1236 | struct select_data { |
1237 | struct dentry *start; | |
1238 | struct list_head dispose; | |
1239 | int found; | |
1240 | }; | |
23044507 | 1241 | |
db14fc3a MS |
1242 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1243 | { | |
1244 | struct select_data *data = _data; | |
1245 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1246 | |
db14fc3a MS |
1247 | if (data->start == dentry) |
1248 | goto out; | |
2fd6b7f5 | 1249 | |
fe91522a | 1250 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1251 | data->found++; |
fe91522a AV |
1252 | } else { |
1253 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1254 | d_lru_del(dentry); | |
1255 | if (!dentry->d_lockref.count) { | |
1256 | d_shrink_add(dentry, &data->dispose); | |
1257 | data->found++; | |
1258 | } | |
1da177e4 | 1259 | } |
db14fc3a MS |
1260 | /* |
1261 | * We can return to the caller if we have found some (this | |
1262 | * ensures forward progress). We'll be coming back to find | |
1263 | * the rest. | |
1264 | */ | |
fe91522a AV |
1265 | if (!list_empty(&data->dispose)) |
1266 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1267 | out: |
db14fc3a | 1268 | return ret; |
1da177e4 LT |
1269 | } |
1270 | ||
1271 | /** | |
1272 | * shrink_dcache_parent - prune dcache | |
1273 | * @parent: parent of entries to prune | |
1274 | * | |
1275 | * Prune the dcache to remove unused children of the parent dentry. | |
1276 | */ | |
db14fc3a | 1277 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1278 | { |
db14fc3a MS |
1279 | for (;;) { |
1280 | struct select_data data; | |
1da177e4 | 1281 | |
db14fc3a MS |
1282 | INIT_LIST_HEAD(&data.dispose); |
1283 | data.start = parent; | |
1284 | data.found = 0; | |
1285 | ||
1286 | d_walk(parent, &data, select_collect, NULL); | |
1287 | if (!data.found) | |
1288 | break; | |
1289 | ||
1290 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1291 | cond_resched(); |
1292 | } | |
1da177e4 | 1293 | } |
ec4f8605 | 1294 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1295 | |
9c8c10e2 | 1296 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1297 | { |
9c8c10e2 AV |
1298 | /* it has busy descendents; complain about those instead */ |
1299 | if (!list_empty(&dentry->d_subdirs)) | |
1300 | return D_WALK_CONTINUE; | |
42c32608 | 1301 | |
9c8c10e2 AV |
1302 | /* root with refcount 1 is fine */ |
1303 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1304 | return D_WALK_CONTINUE; | |
1305 | ||
1306 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1307 | " still in use (%d) [unmount of %s %s]\n", | |
42c32608 AV |
1308 | dentry, |
1309 | dentry->d_inode ? | |
1310 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1311 | dentry, |
42c32608 AV |
1312 | dentry->d_lockref.count, |
1313 | dentry->d_sb->s_type->name, | |
1314 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1315 | WARN_ON(1); |
1316 | return D_WALK_CONTINUE; | |
1317 | } | |
1318 | ||
1319 | static void do_one_tree(struct dentry *dentry) | |
1320 | { | |
1321 | shrink_dcache_parent(dentry); | |
1322 | d_walk(dentry, dentry, umount_check, NULL); | |
1323 | d_drop(dentry); | |
1324 | dput(dentry); | |
42c32608 AV |
1325 | } |
1326 | ||
1327 | /* | |
1328 | * destroy the dentries attached to a superblock on unmounting | |
1329 | */ | |
1330 | void shrink_dcache_for_umount(struct super_block *sb) | |
1331 | { | |
1332 | struct dentry *dentry; | |
1333 | ||
9c8c10e2 | 1334 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1335 | |
1336 | dentry = sb->s_root; | |
1337 | sb->s_root = NULL; | |
9c8c10e2 | 1338 | do_one_tree(dentry); |
42c32608 AV |
1339 | |
1340 | while (!hlist_bl_empty(&sb->s_anon)) { | |
9c8c10e2 AV |
1341 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash)); |
1342 | do_one_tree(dentry); | |
42c32608 AV |
1343 | } |
1344 | } | |
1345 | ||
8ed936b5 EB |
1346 | struct detach_data { |
1347 | struct select_data select; | |
1348 | struct dentry *mountpoint; | |
1349 | }; | |
1350 | static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry) | |
848ac114 | 1351 | { |
8ed936b5 | 1352 | struct detach_data *data = _data; |
848ac114 MS |
1353 | |
1354 | if (d_mountpoint(dentry)) { | |
8ed936b5 EB |
1355 | __dget_dlock(dentry); |
1356 | data->mountpoint = dentry; | |
848ac114 MS |
1357 | return D_WALK_QUIT; |
1358 | } | |
1359 | ||
8ed936b5 | 1360 | return select_collect(&data->select, dentry); |
848ac114 MS |
1361 | } |
1362 | ||
1363 | static void check_and_drop(void *_data) | |
1364 | { | |
8ed936b5 | 1365 | struct detach_data *data = _data; |
848ac114 | 1366 | |
8ed936b5 EB |
1367 | if (!data->mountpoint && !data->select.found) |
1368 | __d_drop(data->select.start); | |
848ac114 MS |
1369 | } |
1370 | ||
1371 | /** | |
8ed936b5 | 1372 | * check_submounts_and_drop - detach submounts, prune dcache, and drop |
848ac114 | 1373 | * |
8ed936b5 EB |
1374 | * The final d_drop is done as an atomic operation relative to |
1375 | * rename_lock ensuring there are no races with d_set_mounted. This | |
1376 | * ensures there are no unhashed dentries on the path to a mountpoint. | |
848ac114 | 1377 | * |
8ed936b5 | 1378 | * @dentry: dentry to detach, prune and drop |
848ac114 MS |
1379 | */ |
1380 | int check_submounts_and_drop(struct dentry *dentry) | |
1381 | { | |
1382 | int ret = 0; | |
1383 | ||
1384 | /* Negative dentries can be dropped without further checks */ | |
1385 | if (!dentry->d_inode) { | |
1386 | d_drop(dentry); | |
1387 | goto out; | |
1388 | } | |
1389 | ||
1390 | for (;;) { | |
8ed936b5 | 1391 | struct detach_data data; |
848ac114 | 1392 | |
8ed936b5 EB |
1393 | data.mountpoint = NULL; |
1394 | INIT_LIST_HEAD(&data.select.dispose); | |
1395 | data.select.start = dentry; | |
1396 | data.select.found = 0; | |
1397 | ||
1398 | d_walk(dentry, &data, detach_and_collect, check_and_drop); | |
848ac114 | 1399 | |
8ed936b5 EB |
1400 | if (data.select.found) |
1401 | shrink_dentry_list(&data.select.dispose); | |
848ac114 | 1402 | |
8ed936b5 EB |
1403 | if (data.mountpoint) { |
1404 | detach_mounts(data.mountpoint); | |
1405 | dput(data.mountpoint); | |
1406 | } | |
848ac114 | 1407 | |
8ed936b5 | 1408 | if (!data.mountpoint && !data.select.found) |
848ac114 MS |
1409 | break; |
1410 | ||
1411 | cond_resched(); | |
1412 | } | |
1413 | ||
1414 | out: | |
1415 | return ret; | |
1416 | } | |
1417 | EXPORT_SYMBOL(check_submounts_and_drop); | |
1418 | ||
1da177e4 | 1419 | /** |
a4464dbc AV |
1420 | * __d_alloc - allocate a dcache entry |
1421 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1422 | * @name: qstr of the name |
1423 | * | |
1424 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1425 | * available. On a success the dentry is returned. The name passed in is | |
1426 | * copied and the copy passed in may be reused after this call. | |
1427 | */ | |
1428 | ||
a4464dbc | 1429 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1430 | { |
1431 | struct dentry *dentry; | |
1432 | char *dname; | |
1433 | ||
e12ba74d | 1434 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1435 | if (!dentry) |
1436 | return NULL; | |
1437 | ||
6326c71f LT |
1438 | /* |
1439 | * We guarantee that the inline name is always NUL-terminated. | |
1440 | * This way the memcpy() done by the name switching in rename | |
1441 | * will still always have a NUL at the end, even if we might | |
1442 | * be overwriting an internal NUL character | |
1443 | */ | |
1444 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 | 1445 | if (name->len > DNAME_INLINE_LEN-1) { |
8d85b484 AV |
1446 | size_t size = offsetof(struct external_name, name[1]); |
1447 | struct external_name *p = kmalloc(size + name->len, GFP_KERNEL); | |
1448 | if (!p) { | |
1da177e4 LT |
1449 | kmem_cache_free(dentry_cache, dentry); |
1450 | return NULL; | |
1451 | } | |
8d85b484 AV |
1452 | atomic_set(&p->u.count, 1); |
1453 | dname = p->name; | |
1da177e4 LT |
1454 | } else { |
1455 | dname = dentry->d_iname; | |
1456 | } | |
1da177e4 LT |
1457 | |
1458 | dentry->d_name.len = name->len; | |
1459 | dentry->d_name.hash = name->hash; | |
1460 | memcpy(dname, name->name, name->len); | |
1461 | dname[name->len] = 0; | |
1462 | ||
6326c71f LT |
1463 | /* Make sure we always see the terminating NUL character */ |
1464 | smp_wmb(); | |
1465 | dentry->d_name.name = dname; | |
1466 | ||
98474236 | 1467 | dentry->d_lockref.count = 1; |
dea3667b | 1468 | dentry->d_flags = 0; |
1da177e4 | 1469 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1470 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1471 | dentry->d_inode = NULL; |
a4464dbc AV |
1472 | dentry->d_parent = dentry; |
1473 | dentry->d_sb = sb; | |
1da177e4 LT |
1474 | dentry->d_op = NULL; |
1475 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1476 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1477 | INIT_LIST_HEAD(&dentry->d_lru); |
1478 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
b3d9b7a3 | 1479 | INIT_HLIST_NODE(&dentry->d_alias); |
2fd6b7f5 | 1480 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1481 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1482 | |
3e880fb5 | 1483 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1484 | |
1da177e4 LT |
1485 | return dentry; |
1486 | } | |
a4464dbc AV |
1487 | |
1488 | /** | |
1489 | * d_alloc - allocate a dcache entry | |
1490 | * @parent: parent of entry to allocate | |
1491 | * @name: qstr of the name | |
1492 | * | |
1493 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1494 | * available. On a success the dentry is returned. The name passed in is | |
1495 | * copied and the copy passed in may be reused after this call. | |
1496 | */ | |
1497 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1498 | { | |
1499 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1500 | if (!dentry) | |
1501 | return NULL; | |
1502 | ||
1503 | spin_lock(&parent->d_lock); | |
1504 | /* | |
1505 | * don't need child lock because it is not subject | |
1506 | * to concurrency here | |
1507 | */ | |
1508 | __dget_dlock(parent); | |
1509 | dentry->d_parent = parent; | |
1510 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1511 | spin_unlock(&parent->d_lock); | |
1512 | ||
1513 | return dentry; | |
1514 | } | |
ec4f8605 | 1515 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1516 | |
e1a24bb0 BF |
1517 | /** |
1518 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1519 | * @sb: the superblock | |
1520 | * @name: qstr of the name | |
1521 | * | |
1522 | * For a filesystem that just pins its dentries in memory and never | |
1523 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1524 | */ | |
4b936885 NP |
1525 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1526 | { | |
e1a24bb0 | 1527 | return __d_alloc(sb, name); |
4b936885 NP |
1528 | } |
1529 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1530 | ||
1da177e4 LT |
1531 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1532 | { | |
1533 | struct qstr q; | |
1534 | ||
1535 | q.name = name; | |
1536 | q.len = strlen(name); | |
1537 | q.hash = full_name_hash(q.name, q.len); | |
1538 | return d_alloc(parent, &q); | |
1539 | } | |
ef26ca97 | 1540 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1541 | |
fb045adb NP |
1542 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1543 | { | |
6f7f7caa LT |
1544 | WARN_ON_ONCE(dentry->d_op); |
1545 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1546 | DCACHE_OP_COMPARE | |
1547 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1548 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1549 | DCACHE_OP_DELETE )); |
1550 | dentry->d_op = op; | |
1551 | if (!op) | |
1552 | return; | |
1553 | if (op->d_hash) | |
1554 | dentry->d_flags |= DCACHE_OP_HASH; | |
1555 | if (op->d_compare) | |
1556 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1557 | if (op->d_revalidate) | |
1558 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1559 | if (op->d_weak_revalidate) |
1560 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1561 | if (op->d_delete) |
1562 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1563 | if (op->d_prune) |
1564 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1565 | |
1566 | } | |
1567 | EXPORT_SYMBOL(d_set_d_op); | |
1568 | ||
b18825a7 DH |
1569 | static unsigned d_flags_for_inode(struct inode *inode) |
1570 | { | |
1571 | unsigned add_flags = DCACHE_FILE_TYPE; | |
1572 | ||
1573 | if (!inode) | |
1574 | return DCACHE_MISS_TYPE; | |
1575 | ||
1576 | if (S_ISDIR(inode->i_mode)) { | |
1577 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1578 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1579 | if (unlikely(!inode->i_op->lookup)) | |
1580 | add_flags = DCACHE_AUTODIR_TYPE; | |
1581 | else | |
1582 | inode->i_opflags |= IOP_LOOKUP; | |
1583 | } | |
1584 | } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1585 | if (unlikely(inode->i_op->follow_link)) | |
1586 | add_flags = DCACHE_SYMLINK_TYPE; | |
1587 | else | |
1588 | inode->i_opflags |= IOP_NOFOLLOW; | |
1589 | } | |
1590 | ||
1591 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1592 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1593 | return add_flags; | |
1594 | } | |
1595 | ||
360da900 OH |
1596 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1597 | { | |
b18825a7 DH |
1598 | unsigned add_flags = d_flags_for_inode(inode); |
1599 | ||
b23fb0a6 | 1600 | spin_lock(&dentry->d_lock); |
22213318 | 1601 | __d_set_type(dentry, add_flags); |
b18825a7 | 1602 | if (inode) |
b3d9b7a3 | 1603 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); |
360da900 | 1604 | dentry->d_inode = inode; |
31e6b01f | 1605 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1606 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1607 | fsnotify_d_instantiate(dentry, inode); |
1608 | } | |
1609 | ||
1da177e4 LT |
1610 | /** |
1611 | * d_instantiate - fill in inode information for a dentry | |
1612 | * @entry: dentry to complete | |
1613 | * @inode: inode to attach to this dentry | |
1614 | * | |
1615 | * Fill in inode information in the entry. | |
1616 | * | |
1617 | * This turns negative dentries into productive full members | |
1618 | * of society. | |
1619 | * | |
1620 | * NOTE! This assumes that the inode count has been incremented | |
1621 | * (or otherwise set) by the caller to indicate that it is now | |
1622 | * in use by the dcache. | |
1623 | */ | |
1624 | ||
1625 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1626 | { | |
b3d9b7a3 | 1627 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
873feea0 NP |
1628 | if (inode) |
1629 | spin_lock(&inode->i_lock); | |
360da900 | 1630 | __d_instantiate(entry, inode); |
873feea0 NP |
1631 | if (inode) |
1632 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1633 | security_d_instantiate(entry, inode); |
1634 | } | |
ec4f8605 | 1635 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1636 | |
1637 | /** | |
1638 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1639 | * @entry: dentry to instantiate | |
1640 | * @inode: inode to attach to this dentry | |
1641 | * | |
1642 | * Fill in inode information in the entry. On success, it returns NULL. | |
1643 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1644 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1645 | * |
1646 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1647 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1648 | * |
1649 | * This also assumes that the inode count has been incremented | |
1650 | * (or otherwise set) by the caller to indicate that it is now | |
1651 | * in use by the dcache. | |
1da177e4 | 1652 | */ |
770bfad8 DH |
1653 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1654 | struct inode *inode) | |
1da177e4 LT |
1655 | { |
1656 | struct dentry *alias; | |
1657 | int len = entry->d_name.len; | |
1658 | const char *name = entry->d_name.name; | |
1659 | unsigned int hash = entry->d_name.hash; | |
1660 | ||
770bfad8 | 1661 | if (!inode) { |
360da900 | 1662 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1663 | return NULL; |
1664 | } | |
1665 | ||
b67bfe0d | 1666 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
9abca360 NP |
1667 | /* |
1668 | * Don't need alias->d_lock here, because aliases with | |
1669 | * d_parent == entry->d_parent are not subject to name or | |
1670 | * parent changes, because the parent inode i_mutex is held. | |
1671 | */ | |
12f8ad4b | 1672 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1673 | continue; |
1674 | if (alias->d_parent != entry->d_parent) | |
1675 | continue; | |
ee983e89 LT |
1676 | if (alias->d_name.len != len) |
1677 | continue; | |
12f8ad4b | 1678 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1679 | continue; |
dc0474be | 1680 | __dget(alias); |
1da177e4 LT |
1681 | return alias; |
1682 | } | |
770bfad8 | 1683 | |
360da900 | 1684 | __d_instantiate(entry, inode); |
1da177e4 LT |
1685 | return NULL; |
1686 | } | |
770bfad8 DH |
1687 | |
1688 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1689 | { | |
1690 | struct dentry *result; | |
1691 | ||
b3d9b7a3 | 1692 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
770bfad8 | 1693 | |
873feea0 NP |
1694 | if (inode) |
1695 | spin_lock(&inode->i_lock); | |
770bfad8 | 1696 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1697 | if (inode) |
1698 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1699 | |
1700 | if (!result) { | |
1701 | security_d_instantiate(entry, inode); | |
1702 | return NULL; | |
1703 | } | |
1704 | ||
1705 | BUG_ON(!d_unhashed(result)); | |
1706 | iput(inode); | |
1707 | return result; | |
1708 | } | |
1709 | ||
1da177e4 LT |
1710 | EXPORT_SYMBOL(d_instantiate_unique); |
1711 | ||
b70a80e7 MS |
1712 | /** |
1713 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1714 | * @entry: dentry to complete | |
1715 | * @inode: inode to attach to this dentry | |
1716 | * | |
1717 | * Fill in inode information in the entry. If a directory alias is found, then | |
1718 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1719 | * guarantees that a directory inode may never have more than one alias. | |
1720 | */ | |
1721 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1722 | { | |
1723 | BUG_ON(!hlist_unhashed(&entry->d_alias)); | |
1724 | ||
1725 | spin_lock(&inode->i_lock); | |
1726 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1727 | spin_unlock(&inode->i_lock); | |
1728 | iput(inode); | |
1729 | return -EBUSY; | |
1730 | } | |
1731 | __d_instantiate(entry, inode); | |
1732 | spin_unlock(&inode->i_lock); | |
1733 | security_d_instantiate(entry, inode); | |
1734 | ||
1735 | return 0; | |
1736 | } | |
1737 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1738 | ||
adc0e91a AV |
1739 | struct dentry *d_make_root(struct inode *root_inode) |
1740 | { | |
1741 | struct dentry *res = NULL; | |
1742 | ||
1743 | if (root_inode) { | |
26fe5750 | 1744 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1745 | |
1746 | res = __d_alloc(root_inode->i_sb, &name); | |
1747 | if (res) | |
1748 | d_instantiate(res, root_inode); | |
1749 | else | |
1750 | iput(root_inode); | |
1751 | } | |
1752 | return res; | |
1753 | } | |
1754 | EXPORT_SYMBOL(d_make_root); | |
1755 | ||
d891eedb BF |
1756 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1757 | { | |
1758 | struct dentry *alias; | |
1759 | ||
b3d9b7a3 | 1760 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1761 | return NULL; |
b3d9b7a3 | 1762 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); |
d891eedb BF |
1763 | __dget(alias); |
1764 | return alias; | |
1765 | } | |
1766 | ||
46f72b34 SW |
1767 | /** |
1768 | * d_find_any_alias - find any alias for a given inode | |
1769 | * @inode: inode to find an alias for | |
1770 | * | |
1771 | * If any aliases exist for the given inode, take and return a | |
1772 | * reference for one of them. If no aliases exist, return %NULL. | |
1773 | */ | |
1774 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1775 | { |
1776 | struct dentry *de; | |
1777 | ||
1778 | spin_lock(&inode->i_lock); | |
1779 | de = __d_find_any_alias(inode); | |
1780 | spin_unlock(&inode->i_lock); | |
1781 | return de; | |
1782 | } | |
46f72b34 | 1783 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1784 | |
49c7dd28 | 1785 | static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected) |
4ea3ada2 | 1786 | { |
b911a6bd | 1787 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1788 | struct dentry *tmp; |
1789 | struct dentry *res; | |
b18825a7 | 1790 | unsigned add_flags; |
4ea3ada2 CH |
1791 | |
1792 | if (!inode) | |
44003728 | 1793 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1794 | if (IS_ERR(inode)) |
1795 | return ERR_CAST(inode); | |
1796 | ||
d891eedb | 1797 | res = d_find_any_alias(inode); |
9308a612 CH |
1798 | if (res) |
1799 | goto out_iput; | |
1800 | ||
a4464dbc | 1801 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1802 | if (!tmp) { |
1803 | res = ERR_PTR(-ENOMEM); | |
1804 | goto out_iput; | |
4ea3ada2 | 1805 | } |
b5c84bf6 | 1806 | |
873feea0 | 1807 | spin_lock(&inode->i_lock); |
d891eedb | 1808 | res = __d_find_any_alias(inode); |
9308a612 | 1809 | if (res) { |
873feea0 | 1810 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1811 | dput(tmp); |
1812 | goto out_iput; | |
1813 | } | |
1814 | ||
1815 | /* attach a disconnected dentry */ | |
1a0a397e BF |
1816 | add_flags = d_flags_for_inode(inode); |
1817 | ||
1818 | if (disconnected) | |
1819 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 1820 | |
9308a612 | 1821 | spin_lock(&tmp->d_lock); |
9308a612 | 1822 | tmp->d_inode = inode; |
b18825a7 | 1823 | tmp->d_flags |= add_flags; |
b3d9b7a3 | 1824 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1825 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1826 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1827 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1828 | spin_unlock(&tmp->d_lock); |
873feea0 | 1829 | spin_unlock(&inode->i_lock); |
24ff6663 | 1830 | security_d_instantiate(tmp, inode); |
9308a612 | 1831 | |
9308a612 CH |
1832 | return tmp; |
1833 | ||
1834 | out_iput: | |
24ff6663 JB |
1835 | if (res && !IS_ERR(res)) |
1836 | security_d_instantiate(res, inode); | |
9308a612 CH |
1837 | iput(inode); |
1838 | return res; | |
4ea3ada2 | 1839 | } |
1a0a397e BF |
1840 | |
1841 | /** | |
1842 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
1843 | * @inode: inode to allocate the dentry for | |
1844 | * | |
1845 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1846 | * similar open by handle operations. The returned dentry may be anonymous, | |
1847 | * or may have a full name (if the inode was already in the cache). | |
1848 | * | |
1849 | * When called on a directory inode, we must ensure that the inode only ever | |
1850 | * has one dentry. If a dentry is found, that is returned instead of | |
1851 | * allocating a new one. | |
1852 | * | |
1853 | * On successful return, the reference to the inode has been transferred | |
1854 | * to the dentry. In case of an error the reference on the inode is released. | |
1855 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1856 | * be passed in and the error will be propagated to the return value, | |
1857 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
1858 | */ | |
1859 | struct dentry *d_obtain_alias(struct inode *inode) | |
1860 | { | |
1861 | return __d_obtain_alias(inode, 1); | |
1862 | } | |
adc48720 | 1863 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 1864 | |
1a0a397e BF |
1865 | /** |
1866 | * d_obtain_root - find or allocate a dentry for a given inode | |
1867 | * @inode: inode to allocate the dentry for | |
1868 | * | |
1869 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
1870 | * | |
1871 | * We must ensure that directory inodes only ever have one dentry. If a | |
1872 | * dentry is found, that is returned instead of 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 | |
1876 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
1877 | * error will be propagate to the return value, with a %NULL @inode | |
1878 | * replaced by ERR_PTR(-ESTALE). | |
1879 | */ | |
1880 | struct dentry *d_obtain_root(struct inode *inode) | |
1881 | { | |
1882 | return __d_obtain_alias(inode, 0); | |
1883 | } | |
1884 | EXPORT_SYMBOL(d_obtain_root); | |
1885 | ||
9403540c BN |
1886 | /** |
1887 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1888 | * @inode: the inode case-insensitive lookup has found | |
1889 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1890 | * @name: the case-exact name to be associated with the returned dentry | |
1891 | * | |
1892 | * This is to avoid filling the dcache with case-insensitive names to the | |
1893 | * same inode, only the actual correct case is stored in the dcache for | |
1894 | * case-insensitive filesystems. | |
1895 | * | |
1896 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1897 | * already exists in in the dcache, use it and return it. | |
1898 | * | |
1899 | * If no entry exists with the exact case name, allocate new dentry with | |
1900 | * the exact case, and return the spliced entry. | |
1901 | */ | |
e45b590b | 1902 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1903 | struct qstr *name) |
1904 | { | |
9403540c BN |
1905 | struct dentry *found; |
1906 | struct dentry *new; | |
1907 | ||
b6520c81 CH |
1908 | /* |
1909 | * First check if a dentry matching the name already exists, | |
1910 | * if not go ahead and create it now. | |
1911 | */ | |
9403540c | 1912 | found = d_hash_and_lookup(dentry->d_parent, name); |
4f522a24 AV |
1913 | if (unlikely(IS_ERR(found))) |
1914 | goto err_out; | |
9403540c BN |
1915 | if (!found) { |
1916 | new = d_alloc(dentry->d_parent, name); | |
1917 | if (!new) { | |
4f522a24 | 1918 | found = ERR_PTR(-ENOMEM); |
9403540c BN |
1919 | goto err_out; |
1920 | } | |
b6520c81 | 1921 | |
9403540c BN |
1922 | found = d_splice_alias(inode, new); |
1923 | if (found) { | |
1924 | dput(new); | |
1925 | return found; | |
1926 | } | |
1927 | return new; | |
1928 | } | |
b6520c81 CH |
1929 | |
1930 | /* | |
1931 | * If a matching dentry exists, and it's not negative use it. | |
1932 | * | |
1933 | * Decrement the reference count to balance the iget() done | |
1934 | * earlier on. | |
1935 | */ | |
9403540c BN |
1936 | if (found->d_inode) { |
1937 | if (unlikely(found->d_inode != inode)) { | |
1938 | /* This can't happen because bad inodes are unhashed. */ | |
1939 | BUG_ON(!is_bad_inode(inode)); | |
1940 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1941 | } | |
9403540c BN |
1942 | iput(inode); |
1943 | return found; | |
1944 | } | |
b6520c81 | 1945 | |
9403540c | 1946 | /* |
9403540c | 1947 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1948 | * already has a dentry. |
9403540c | 1949 | */ |
4513d899 AV |
1950 | new = d_splice_alias(inode, found); |
1951 | if (new) { | |
1952 | dput(found); | |
1953 | found = new; | |
9403540c | 1954 | } |
4513d899 | 1955 | return found; |
9403540c BN |
1956 | |
1957 | err_out: | |
1958 | iput(inode); | |
4f522a24 | 1959 | return found; |
9403540c | 1960 | } |
ec4f8605 | 1961 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1962 | |
12f8ad4b LT |
1963 | /* |
1964 | * Do the slow-case of the dentry name compare. | |
1965 | * | |
1966 | * Unlike the dentry_cmp() function, we need to atomically | |
da53be12 | 1967 | * load the name and length information, so that the |
12f8ad4b LT |
1968 | * filesystem can rely on them, and can use the 'name' and |
1969 | * 'len' information without worrying about walking off the | |
1970 | * end of memory etc. | |
1971 | * | |
1972 | * Thus the read_seqcount_retry() and the "duplicate" info | |
1973 | * in arguments (the low-level filesystem should not look | |
1974 | * at the dentry inode or name contents directly, since | |
1975 | * rename can change them while we're in RCU mode). | |
1976 | */ | |
1977 | enum slow_d_compare { | |
1978 | D_COMP_OK, | |
1979 | D_COMP_NOMATCH, | |
1980 | D_COMP_SEQRETRY, | |
1981 | }; | |
1982 | ||
1983 | static noinline enum slow_d_compare slow_dentry_cmp( | |
1984 | const struct dentry *parent, | |
12f8ad4b LT |
1985 | struct dentry *dentry, |
1986 | unsigned int seq, | |
1987 | const struct qstr *name) | |
1988 | { | |
1989 | int tlen = dentry->d_name.len; | |
1990 | const char *tname = dentry->d_name.name; | |
12f8ad4b LT |
1991 | |
1992 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
1993 | cpu_relax(); | |
1994 | return D_COMP_SEQRETRY; | |
1995 | } | |
da53be12 | 1996 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
12f8ad4b LT |
1997 | return D_COMP_NOMATCH; |
1998 | return D_COMP_OK; | |
1999 | } | |
2000 | ||
31e6b01f NP |
2001 | /** |
2002 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2003 | * @parent: parent dentry | |
2004 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2005 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2006 | * Returns: dentry, or NULL |
2007 | * | |
2008 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2009 | * resolution (store-free path walking) design described in | |
2010 | * Documentation/filesystems/path-lookup.txt. | |
2011 | * | |
2012 | * This is not to be used outside core vfs. | |
2013 | * | |
2014 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2015 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2016 | * without taking d_lock and checking d_seq sequence count against @seq | |
2017 | * returned here. | |
2018 | * | |
15570086 | 2019 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2020 | * function. |
2021 | * | |
2022 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2023 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2024 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2025 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2026 | * |
2027 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2028 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2029 | */ |
8966be90 LT |
2030 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2031 | const struct qstr *name, | |
da53be12 | 2032 | unsigned *seqp) |
31e6b01f | 2033 | { |
26fe5750 | 2034 | u64 hashlen = name->hash_len; |
31e6b01f | 2035 | const unsigned char *str = name->name; |
26fe5750 | 2036 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 2037 | struct hlist_bl_node *node; |
31e6b01f NP |
2038 | struct dentry *dentry; |
2039 | ||
2040 | /* | |
2041 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2042 | * required to prevent single threaded performance regressions | |
2043 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2044 | * Keep the two functions in sync. | |
2045 | */ | |
2046 | ||
2047 | /* | |
2048 | * The hash list is protected using RCU. | |
2049 | * | |
2050 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2051 | * races with d_move(). | |
2052 | * | |
2053 | * It is possible that concurrent renames can mess up our list | |
2054 | * walk here and result in missing our dentry, resulting in the | |
2055 | * false-negative result. d_lookup() protects against concurrent | |
2056 | * renames using rename_lock seqlock. | |
2057 | * | |
b0a4bb83 | 2058 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2059 | */ |
b07ad996 | 2060 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2061 | unsigned seq; |
31e6b01f | 2062 | |
31e6b01f | 2063 | seqretry: |
12f8ad4b LT |
2064 | /* |
2065 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2066 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2067 | * |
2068 | * The caller must perform a seqcount check in order | |
da53be12 | 2069 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2070 | * |
2071 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2072 | * we don't wait for the sequence count to stabilize if it | |
2073 | * is in the middle of a sequence change. If we do the slow | |
2074 | * dentry compare, we will do seqretries until it is stable, | |
2075 | * and if we end up with a successful lookup, we actually | |
2076 | * want to exit RCU lookup anyway. | |
2077 | */ | |
2078 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2079 | if (dentry->d_parent != parent) |
2080 | continue; | |
2e321806 LT |
2081 | if (d_unhashed(dentry)) |
2082 | continue; | |
12f8ad4b | 2083 | |
830c0f0e | 2084 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
2085 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2086 | continue; | |
da53be12 LT |
2087 | *seqp = seq; |
2088 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
12f8ad4b LT |
2089 | case D_COMP_OK: |
2090 | return dentry; | |
2091 | case D_COMP_NOMATCH: | |
31e6b01f | 2092 | continue; |
12f8ad4b LT |
2093 | default: |
2094 | goto seqretry; | |
2095 | } | |
31e6b01f | 2096 | } |
12f8ad4b | 2097 | |
26fe5750 | 2098 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 2099 | continue; |
da53be12 | 2100 | *seqp = seq; |
26fe5750 | 2101 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 2102 | return dentry; |
31e6b01f NP |
2103 | } |
2104 | return NULL; | |
2105 | } | |
2106 | ||
1da177e4 LT |
2107 | /** |
2108 | * d_lookup - search for a dentry | |
2109 | * @parent: parent dentry | |
2110 | * @name: qstr of name we wish to find | |
b04f784e | 2111 | * Returns: dentry, or NULL |
1da177e4 | 2112 | * |
b04f784e NP |
2113 | * d_lookup searches the children of the parent dentry for the name in |
2114 | * question. If the dentry is found its reference count is incremented and the | |
2115 | * dentry is returned. The caller must use dput to free the entry when it has | |
2116 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2117 | */ |
da2d8455 | 2118 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2119 | { |
31e6b01f | 2120 | struct dentry *dentry; |
949854d0 | 2121 | unsigned seq; |
1da177e4 LT |
2122 | |
2123 | do { | |
2124 | seq = read_seqbegin(&rename_lock); | |
2125 | dentry = __d_lookup(parent, name); | |
2126 | if (dentry) | |
2127 | break; | |
2128 | } while (read_seqretry(&rename_lock, seq)); | |
2129 | return dentry; | |
2130 | } | |
ec4f8605 | 2131 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2132 | |
31e6b01f | 2133 | /** |
b04f784e NP |
2134 | * __d_lookup - search for a dentry (racy) |
2135 | * @parent: parent dentry | |
2136 | * @name: qstr of name we wish to find | |
2137 | * Returns: dentry, or NULL | |
2138 | * | |
2139 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2140 | * false-negative result due to unrelated rename activity. | |
2141 | * | |
2142 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2143 | * however it must be used carefully, eg. with a following d_lookup in | |
2144 | * the case of failure. | |
2145 | * | |
2146 | * __d_lookup callers must be commented. | |
2147 | */ | |
a713ca2a | 2148 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
2149 | { |
2150 | unsigned int len = name->len; | |
2151 | unsigned int hash = name->hash; | |
2152 | const unsigned char *str = name->name; | |
b07ad996 | 2153 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 2154 | struct hlist_bl_node *node; |
31e6b01f | 2155 | struct dentry *found = NULL; |
665a7583 | 2156 | struct dentry *dentry; |
1da177e4 | 2157 | |
31e6b01f NP |
2158 | /* |
2159 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2160 | * required to prevent single threaded performance regressions | |
2161 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2162 | * Keep the two functions in sync. | |
2163 | */ | |
2164 | ||
b04f784e NP |
2165 | /* |
2166 | * The hash list is protected using RCU. | |
2167 | * | |
2168 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2169 | * with d_move(). | |
2170 | * | |
2171 | * It is possible that concurrent renames can mess up our list | |
2172 | * walk here and result in missing our dentry, resulting in the | |
2173 | * false-negative result. d_lookup() protects against concurrent | |
2174 | * renames using rename_lock seqlock. | |
2175 | * | |
b0a4bb83 | 2176 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2177 | */ |
1da177e4 LT |
2178 | rcu_read_lock(); |
2179 | ||
b07ad996 | 2180 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2181 | |
1da177e4 LT |
2182 | if (dentry->d_name.hash != hash) |
2183 | continue; | |
1da177e4 LT |
2184 | |
2185 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2186 | if (dentry->d_parent != parent) |
2187 | goto next; | |
d0185c08 LT |
2188 | if (d_unhashed(dentry)) |
2189 | goto next; | |
2190 | ||
1da177e4 LT |
2191 | /* |
2192 | * It is safe to compare names since d_move() cannot | |
2193 | * change the qstr (protected by d_lock). | |
2194 | */ | |
fb045adb | 2195 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
2196 | int tlen = dentry->d_name.len; |
2197 | const char *tname = dentry->d_name.name; | |
da53be12 | 2198 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1da177e4 LT |
2199 | goto next; |
2200 | } else { | |
ee983e89 LT |
2201 | if (dentry->d_name.len != len) |
2202 | goto next; | |
12f8ad4b | 2203 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
2204 | goto next; |
2205 | } | |
2206 | ||
98474236 | 2207 | dentry->d_lockref.count++; |
d0185c08 | 2208 | found = dentry; |
1da177e4 LT |
2209 | spin_unlock(&dentry->d_lock); |
2210 | break; | |
2211 | next: | |
2212 | spin_unlock(&dentry->d_lock); | |
2213 | } | |
2214 | rcu_read_unlock(); | |
2215 | ||
2216 | return found; | |
2217 | } | |
2218 | ||
3e7e241f EB |
2219 | /** |
2220 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2221 | * @dir: Directory to search in | |
2222 | * @name: qstr of name we wish to find | |
2223 | * | |
4f522a24 | 2224 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2225 | */ |
2226 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2227 | { | |
3e7e241f EB |
2228 | /* |
2229 | * Check for a fs-specific hash function. Note that we must | |
2230 | * calculate the standard hash first, as the d_op->d_hash() | |
2231 | * routine may choose to leave the hash value unchanged. | |
2232 | */ | |
2233 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2234 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2235 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2236 | if (unlikely(err < 0)) |
2237 | return ERR_PTR(err); | |
3e7e241f | 2238 | } |
4f522a24 | 2239 | return d_lookup(dir, name); |
3e7e241f | 2240 | } |
4f522a24 | 2241 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2242 | |
1da177e4 | 2243 | /** |
786a5e15 | 2244 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2245 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2246 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2247 | * |
2248 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2249 | * This is used by ncpfs in its readdir implementation. | |
2250 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2251 | * |
2252 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2253 | */ |
d3a23e16 | 2254 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2255 | { |
786a5e15 | 2256 | struct dentry *child; |
d3a23e16 | 2257 | |
2fd6b7f5 | 2258 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2259 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2260 | if (dentry == child) { | |
2fd6b7f5 | 2261 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2262 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2263 | spin_unlock(&dentry->d_lock); |
2264 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2265 | return 1; |
2266 | } | |
2267 | } | |
2fd6b7f5 | 2268 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2269 | |
1da177e4 LT |
2270 | return 0; |
2271 | } | |
ec4f8605 | 2272 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2273 | |
2274 | /* | |
2275 | * When a file is deleted, we have two options: | |
2276 | * - turn this dentry into a negative dentry | |
2277 | * - unhash this dentry and free it. | |
2278 | * | |
2279 | * Usually, we want to just turn this into | |
2280 | * a negative dentry, but if anybody else is | |
2281 | * currently using the dentry or the inode | |
2282 | * we can't do that and we fall back on removing | |
2283 | * it from the hash queues and waiting for | |
2284 | * it to be deleted later when it has no users | |
2285 | */ | |
2286 | ||
2287 | /** | |
2288 | * d_delete - delete a dentry | |
2289 | * @dentry: The dentry to delete | |
2290 | * | |
2291 | * Turn the dentry into a negative dentry if possible, otherwise | |
2292 | * remove it from the hash queues so it can be deleted later | |
2293 | */ | |
2294 | ||
2295 | void d_delete(struct dentry * dentry) | |
2296 | { | |
873feea0 | 2297 | struct inode *inode; |
7a91bf7f | 2298 | int isdir = 0; |
1da177e4 LT |
2299 | /* |
2300 | * Are we the only user? | |
2301 | */ | |
357f8e65 | 2302 | again: |
1da177e4 | 2303 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2304 | inode = dentry->d_inode; |
2305 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2306 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2307 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2308 | spin_unlock(&dentry->d_lock); |
2309 | cpu_relax(); | |
2310 | goto again; | |
2311 | } | |
13e3c5e5 | 2312 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2313 | dentry_unlink_inode(dentry); |
7a91bf7f | 2314 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2315 | return; |
2316 | } | |
2317 | ||
2318 | if (!d_unhashed(dentry)) | |
2319 | __d_drop(dentry); | |
2320 | ||
2321 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2322 | |
2323 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2324 | } |
ec4f8605 | 2325 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2326 | |
b07ad996 | 2327 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2328 | { |
ceb5bdc2 | 2329 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2330 | hlist_bl_lock(b); |
dea3667b | 2331 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2332 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2333 | hlist_bl_unlock(b); |
1da177e4 LT |
2334 | } |
2335 | ||
770bfad8 DH |
2336 | static void _d_rehash(struct dentry * entry) |
2337 | { | |
2338 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2339 | } | |
2340 | ||
1da177e4 LT |
2341 | /** |
2342 | * d_rehash - add an entry back to the hash | |
2343 | * @entry: dentry to add to the hash | |
2344 | * | |
2345 | * Adds a dentry to the hash according to its name. | |
2346 | */ | |
2347 | ||
2348 | void d_rehash(struct dentry * entry) | |
2349 | { | |
1da177e4 | 2350 | spin_lock(&entry->d_lock); |
770bfad8 | 2351 | _d_rehash(entry); |
1da177e4 | 2352 | spin_unlock(&entry->d_lock); |
1da177e4 | 2353 | } |
ec4f8605 | 2354 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2355 | |
fb2d5b86 NP |
2356 | /** |
2357 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2358 | * @dentry: dentry to be updated | |
2359 | * @name: new name | |
2360 | * | |
2361 | * Update a case insensitive dentry with new case of name. | |
2362 | * | |
2363 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2364 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2365 | * lengths). | |
2366 | * | |
2367 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2368 | * keep renames and concurrent inserts, and readdir(2) away). | |
2369 | */ | |
2370 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2371 | { | |
7ebfa57f | 2372 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2373 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2374 | ||
fb2d5b86 | 2375 | spin_lock(&dentry->d_lock); |
31e6b01f | 2376 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2377 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2378 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2379 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2380 | } |
2381 | EXPORT_SYMBOL(dentry_update_name_case); | |
2382 | ||
8d85b484 | 2383 | static void swap_names(struct dentry *dentry, struct dentry *target) |
1da177e4 | 2384 | { |
8d85b484 AV |
2385 | if (unlikely(dname_external(target))) { |
2386 | if (unlikely(dname_external(dentry))) { | |
1da177e4 LT |
2387 | /* |
2388 | * Both external: swap the pointers | |
2389 | */ | |
9a8d5bb4 | 2390 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2391 | } else { |
2392 | /* | |
2393 | * dentry:internal, target:external. Steal target's | |
2394 | * storage and make target internal. | |
2395 | */ | |
321bcf92 BF |
2396 | memcpy(target->d_iname, dentry->d_name.name, |
2397 | dentry->d_name.len + 1); | |
1da177e4 LT |
2398 | dentry->d_name.name = target->d_name.name; |
2399 | target->d_name.name = target->d_iname; | |
2400 | } | |
2401 | } else { | |
8d85b484 | 2402 | if (unlikely(dname_external(dentry))) { |
1da177e4 LT |
2403 | /* |
2404 | * dentry:external, target:internal. Give dentry's | |
2405 | * storage to target and make dentry internal | |
2406 | */ | |
2407 | memcpy(dentry->d_iname, target->d_name.name, | |
2408 | target->d_name.len + 1); | |
2409 | target->d_name.name = dentry->d_name.name; | |
2410 | dentry->d_name.name = dentry->d_iname; | |
2411 | } else { | |
2412 | /* | |
da1ce067 | 2413 | * Both are internal. |
1da177e4 | 2414 | */ |
da1ce067 MS |
2415 | unsigned int i; |
2416 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
2417 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { | |
2418 | swap(((long *) &dentry->d_iname)[i], | |
2419 | ((long *) &target->d_iname)[i]); | |
2420 | } | |
1da177e4 LT |
2421 | } |
2422 | } | |
a28ddb87 | 2423 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
1da177e4 LT |
2424 | } |
2425 | ||
8d85b484 AV |
2426 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2427 | { | |
2428 | struct external_name *old_name = NULL; | |
2429 | if (unlikely(dname_external(dentry))) | |
2430 | old_name = external_name(dentry); | |
2431 | if (unlikely(dname_external(target))) { | |
2432 | atomic_inc(&external_name(target)->u.count); | |
2433 | dentry->d_name = target->d_name; | |
2434 | } else { | |
2435 | memcpy(dentry->d_iname, target->d_name.name, | |
2436 | target->d_name.len + 1); | |
2437 | dentry->d_name.name = dentry->d_iname; | |
2438 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2439 | } | |
2440 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
2441 | kfree_rcu(old_name, u.head); | |
2442 | } | |
2443 | ||
2fd6b7f5 NP |
2444 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2445 | { | |
2446 | /* | |
2447 | * XXXX: do we really need to take target->d_lock? | |
2448 | */ | |
2449 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2450 | spin_lock(&target->d_parent->d_lock); | |
2451 | else { | |
2452 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2453 | spin_lock(&dentry->d_parent->d_lock); | |
2454 | spin_lock_nested(&target->d_parent->d_lock, | |
2455 | DENTRY_D_LOCK_NESTED); | |
2456 | } else { | |
2457 | spin_lock(&target->d_parent->d_lock); | |
2458 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2459 | DENTRY_D_LOCK_NESTED); | |
2460 | } | |
2461 | } | |
2462 | if (target < dentry) { | |
2463 | spin_lock_nested(&target->d_lock, 2); | |
2464 | spin_lock_nested(&dentry->d_lock, 3); | |
2465 | } else { | |
2466 | spin_lock_nested(&dentry->d_lock, 2); | |
2467 | spin_lock_nested(&target->d_lock, 3); | |
2468 | } | |
2469 | } | |
2470 | ||
986c0194 | 2471 | static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target) |
2fd6b7f5 NP |
2472 | { |
2473 | if (target->d_parent != dentry->d_parent) | |
2474 | spin_unlock(&dentry->d_parent->d_lock); | |
2475 | if (target->d_parent != target) | |
2476 | spin_unlock(&target->d_parent->d_lock); | |
986c0194 AV |
2477 | spin_unlock(&target->d_lock); |
2478 | spin_unlock(&dentry->d_lock); | |
2fd6b7f5 NP |
2479 | } |
2480 | ||
1da177e4 | 2481 | /* |
2fd6b7f5 NP |
2482 | * When switching names, the actual string doesn't strictly have to |
2483 | * be preserved in the target - because we're dropping the target | |
2484 | * anyway. As such, we can just do a simple memcpy() to copy over | |
d2fa4a84 ME |
2485 | * the new name before we switch, unless we are going to rehash |
2486 | * it. Note that if we *do* unhash the target, we are not allowed | |
2487 | * to rehash it without giving it a new name/hash key - whether | |
2488 | * we swap or overwrite the names here, resulting name won't match | |
2489 | * the reality in filesystem; it's only there for d_path() purposes. | |
2490 | * Note that all of this is happening under rename_lock, so the | |
2491 | * any hash lookup seeing it in the middle of manipulations will | |
2492 | * be discarded anyway. So we do not care what happens to the hash | |
2493 | * key in that case. | |
1da177e4 | 2494 | */ |
9eaef27b | 2495 | /* |
18367501 | 2496 | * __d_move - move a dentry |
1da177e4 LT |
2497 | * @dentry: entry to move |
2498 | * @target: new dentry | |
da1ce067 | 2499 | * @exchange: exchange the two dentries |
1da177e4 LT |
2500 | * |
2501 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2502 | * dcache entries should not be moved in this way. Caller must hold |
2503 | * rename_lock, the i_mutex of the source and target directories, | |
2504 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2505 | */ |
da1ce067 MS |
2506 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2507 | bool exchange) | |
1da177e4 | 2508 | { |
1da177e4 LT |
2509 | if (!dentry->d_inode) |
2510 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2511 | ||
2fd6b7f5 NP |
2512 | BUG_ON(d_ancestor(dentry, target)); |
2513 | BUG_ON(d_ancestor(target, dentry)); | |
2514 | ||
2fd6b7f5 | 2515 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2516 | |
31e6b01f | 2517 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2518 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2519 | |
ceb5bdc2 NP |
2520 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2521 | ||
2522 | /* | |
2523 | * Move the dentry to the target hash queue. Don't bother checking | |
2524 | * for the same hash queue because of how unlikely it is. | |
2525 | */ | |
2526 | __d_drop(dentry); | |
789680d1 | 2527 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 | 2528 | |
da1ce067 MS |
2529 | /* |
2530 | * Unhash the target (d_delete() is not usable here). If exchanging | |
2531 | * the two dentries, then rehash onto the other's hash queue. | |
2532 | */ | |
1da177e4 | 2533 | __d_drop(target); |
da1ce067 MS |
2534 | if (exchange) { |
2535 | __d_rehash(target, | |
2536 | d_hash(dentry->d_parent, dentry->d_name.hash)); | |
2537 | } | |
1da177e4 | 2538 | |
1da177e4 | 2539 | /* Switch the names.. */ |
8d85b484 AV |
2540 | if (exchange) |
2541 | swap_names(dentry, target); | |
2542 | else | |
2543 | copy_name(dentry, target); | |
1da177e4 | 2544 | |
63cf427a | 2545 | /* ... and switch them in the tree */ |
1da177e4 | 2546 | if (IS_ROOT(dentry)) { |
63cf427a | 2547 | /* splicing a tree */ |
1da177e4 LT |
2548 | dentry->d_parent = target->d_parent; |
2549 | target->d_parent = target; | |
9d8cd306 | 2550 | list_del_init(&target->d_u.d_child); |
63cf427a | 2551 | list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
1da177e4 | 2552 | } else { |
63cf427a | 2553 | /* swapping two dentries */ |
9a8d5bb4 | 2554 | swap(dentry->d_parent, target->d_parent); |
9d8cd306 | 2555 | list_move(&target->d_u.d_child, &target->d_parent->d_subdirs); |
63cf427a AV |
2556 | list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2557 | if (exchange) | |
2558 | fsnotify_d_move(target); | |
2559 | fsnotify_d_move(dentry); | |
1da177e4 LT |
2560 | } |
2561 | ||
31e6b01f NP |
2562 | write_seqcount_end(&target->d_seq); |
2563 | write_seqcount_end(&dentry->d_seq); | |
2564 | ||
986c0194 | 2565 | dentry_unlock_for_move(dentry, target); |
18367501 AV |
2566 | } |
2567 | ||
2568 | /* | |
2569 | * d_move - move a dentry | |
2570 | * @dentry: entry to move | |
2571 | * @target: new dentry | |
2572 | * | |
2573 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2574 | * dcache entries should not be moved in this way. See the locking |
2575 | * requirements for __d_move. | |
18367501 AV |
2576 | */ |
2577 | void d_move(struct dentry *dentry, struct dentry *target) | |
2578 | { | |
2579 | write_seqlock(&rename_lock); | |
da1ce067 | 2580 | __d_move(dentry, target, false); |
1da177e4 | 2581 | write_sequnlock(&rename_lock); |
9eaef27b | 2582 | } |
ec4f8605 | 2583 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2584 | |
da1ce067 MS |
2585 | /* |
2586 | * d_exchange - exchange two dentries | |
2587 | * @dentry1: first dentry | |
2588 | * @dentry2: second dentry | |
2589 | */ | |
2590 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2591 | { | |
2592 | write_seqlock(&rename_lock); | |
2593 | ||
2594 | WARN_ON(!dentry1->d_inode); | |
2595 | WARN_ON(!dentry2->d_inode); | |
2596 | WARN_ON(IS_ROOT(dentry1)); | |
2597 | WARN_ON(IS_ROOT(dentry2)); | |
2598 | ||
2599 | __d_move(dentry1, dentry2, true); | |
2600 | ||
2601 | write_sequnlock(&rename_lock); | |
2602 | } | |
2603 | ||
e2761a11 OH |
2604 | /** |
2605 | * d_ancestor - search for an ancestor | |
2606 | * @p1: ancestor dentry | |
2607 | * @p2: child dentry | |
2608 | * | |
2609 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2610 | * an ancestor of p2, else NULL. | |
9eaef27b | 2611 | */ |
e2761a11 | 2612 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2613 | { |
2614 | struct dentry *p; | |
2615 | ||
871c0067 | 2616 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2617 | if (p->d_parent == p1) |
e2761a11 | 2618 | return p; |
9eaef27b | 2619 | } |
e2761a11 | 2620 | return NULL; |
9eaef27b TM |
2621 | } |
2622 | ||
2623 | /* | |
2624 | * This helper attempts to cope with remotely renamed directories | |
2625 | * | |
2626 | * It assumes that the caller is already holding | |
18367501 | 2627 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2628 | * |
2629 | * Note: If ever the locking in lock_rename() changes, then please | |
2630 | * remember to update this too... | |
9eaef27b | 2631 | */ |
873feea0 NP |
2632 | static struct dentry *__d_unalias(struct inode *inode, |
2633 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2634 | { |
2635 | struct mutex *m1 = NULL, *m2 = NULL; | |
ee3efa91 | 2636 | struct dentry *ret = ERR_PTR(-EBUSY); |
9eaef27b TM |
2637 | |
2638 | /* If alias and dentry share a parent, then no extra locks required */ | |
2639 | if (alias->d_parent == dentry->d_parent) | |
2640 | goto out_unalias; | |
2641 | ||
9eaef27b | 2642 | /* See lock_rename() */ |
9eaef27b TM |
2643 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2644 | goto out_err; | |
2645 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2646 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2647 | goto out_err; | |
2648 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2649 | out_unalias: | |
8ed936b5 EB |
2650 | __d_move(alias, dentry, false); |
2651 | ret = alias; | |
9eaef27b | 2652 | out_err: |
873feea0 | 2653 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2654 | if (m2) |
2655 | mutex_unlock(m2); | |
2656 | if (m1) | |
2657 | mutex_unlock(m1); | |
2658 | return ret; | |
2659 | } | |
2660 | ||
3f70bd51 BF |
2661 | /** |
2662 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
2663 | * @inode: the inode which may have a disconnected dentry | |
2664 | * @dentry: a negative dentry which we want to point to the inode. | |
2665 | * | |
da093a9b BF |
2666 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
2667 | * place of the given dentry and return it, else simply d_add the inode | |
2668 | * to the dentry and return NULL. | |
3f70bd51 | 2669 | * |
908790fa BF |
2670 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
2671 | * we should error out: directories can't have multiple aliases. | |
2672 | * | |
3f70bd51 BF |
2673 | * This is needed in the lookup routine of any filesystem that is exportable |
2674 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
2675 | * | |
2676 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
2677 | * is returned. This matches the expected return value of ->lookup. | |
2678 | * | |
2679 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
2680 | * In that case, we know that the inode will be a regular file, and also this | |
2681 | * will only occur during atomic_open. So we need to check for the dentry | |
2682 | * being already hashed only in the final case. | |
2683 | */ | |
2684 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
2685 | { | |
2686 | struct dentry *new = NULL; | |
2687 | ||
2688 | if (IS_ERR(inode)) | |
2689 | return ERR_CAST(inode); | |
2690 | ||
2691 | if (inode && S_ISDIR(inode->i_mode)) { | |
2692 | spin_lock(&inode->i_lock); | |
908790fa | 2693 | new = __d_find_any_alias(inode); |
3f70bd51 | 2694 | if (new) { |
da093a9b | 2695 | if (!IS_ROOT(new)) { |
908790fa BF |
2696 | spin_unlock(&inode->i_lock); |
2697 | dput(new); | |
2698 | return ERR_PTR(-EIO); | |
2699 | } | |
95ad5c29 BF |
2700 | if (d_ancestor(new, dentry)) { |
2701 | spin_unlock(&inode->i_lock); | |
2702 | dput(new); | |
2703 | return ERR_PTR(-EIO); | |
2704 | } | |
75a2352d | 2705 | write_seqlock(&rename_lock); |
63cf427a | 2706 | __d_move(new, dentry, false); |
75a2352d | 2707 | write_sequnlock(&rename_lock); |
3f70bd51 BF |
2708 | spin_unlock(&inode->i_lock); |
2709 | security_d_instantiate(new, inode); | |
3f70bd51 BF |
2710 | iput(inode); |
2711 | } else { | |
2712 | /* already taking inode->i_lock, so d_add() by hand */ | |
2713 | __d_instantiate(dentry, inode); | |
2714 | spin_unlock(&inode->i_lock); | |
2715 | security_d_instantiate(dentry, inode); | |
2716 | d_rehash(dentry); | |
2717 | } | |
2718 | } else { | |
2719 | d_instantiate(dentry, inode); | |
2720 | if (d_unhashed(dentry)) | |
2721 | d_rehash(dentry); | |
2722 | } | |
2723 | return new; | |
2724 | } | |
2725 | EXPORT_SYMBOL(d_splice_alias); | |
2726 | ||
770bfad8 DH |
2727 | /** |
2728 | * d_materialise_unique - introduce an inode into the tree | |
2729 | * @dentry: candidate dentry | |
2730 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2731 | * | |
2732 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2733 | * root directory alias in its place if there is one. Caller must hold the |
2734 | * i_mutex of the parent directory. | |
770bfad8 DH |
2735 | */ |
2736 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2737 | { | |
9eaef27b | 2738 | struct dentry *actual; |
770bfad8 DH |
2739 | |
2740 | BUG_ON(!d_unhashed(dentry)); | |
2741 | ||
770bfad8 DH |
2742 | if (!inode) { |
2743 | actual = dentry; | |
360da900 | 2744 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2745 | d_rehash(actual); |
2746 | goto out_nolock; | |
770bfad8 DH |
2747 | } |
2748 | ||
873feea0 | 2749 | spin_lock(&inode->i_lock); |
357f8e65 | 2750 | |
9eaef27b TM |
2751 | if (S_ISDIR(inode->i_mode)) { |
2752 | struct dentry *alias; | |
2753 | ||
2754 | /* Does an aliased dentry already exist? */ | |
52ed46f0 | 2755 | alias = __d_find_alias(inode); |
9eaef27b TM |
2756 | if (alias) { |
2757 | actual = alias; | |
18367501 AV |
2758 | write_seqlock(&rename_lock); |
2759 | ||
2760 | if (d_ancestor(alias, dentry)) { | |
2761 | /* Check for loops */ | |
2762 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2763 | spin_unlock(&inode->i_lock); |
18367501 AV |
2764 | } else if (IS_ROOT(alias)) { |
2765 | /* Is this an anonymous mountpoint that we | |
2766 | * could splice into our tree? */ | |
63cf427a | 2767 | __d_move(alias, dentry, false); |
18367501 | 2768 | write_sequnlock(&rename_lock); |
9eaef27b | 2769 | goto found; |
18367501 AV |
2770 | } else { |
2771 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2772 | * aliasing. This drops inode->i_lock */ |
18367501 | 2773 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2774 | } |
18367501 | 2775 | write_sequnlock(&rename_lock); |
dd179946 DH |
2776 | if (IS_ERR(actual)) { |
2777 | if (PTR_ERR(actual) == -ELOOP) | |
2778 | pr_warn_ratelimited( | |
2779 | "VFS: Lookup of '%s' in %s %s" | |
2780 | " would have caused loop\n", | |
2781 | dentry->d_name.name, | |
2782 | inode->i_sb->s_type->name, | |
2783 | inode->i_sb->s_id); | |
9eaef27b | 2784 | dput(alias); |
dd179946 | 2785 | } |
9eaef27b TM |
2786 | goto out_nolock; |
2787 | } | |
770bfad8 DH |
2788 | } |
2789 | ||
2790 | /* Add a unique reference */ | |
2791 | actual = __d_instantiate_unique(dentry, inode); | |
2792 | if (!actual) | |
2793 | actual = dentry; | |
770bfad8 | 2794 | |
8527dd71 | 2795 | d_rehash(actual); |
5cc3821b | 2796 | found: |
873feea0 | 2797 | spin_unlock(&inode->i_lock); |
9eaef27b | 2798 | out_nolock: |
770bfad8 DH |
2799 | if (actual == dentry) { |
2800 | security_d_instantiate(dentry, inode); | |
2801 | return NULL; | |
2802 | } | |
2803 | ||
2804 | iput(inode); | |
2805 | return actual; | |
770bfad8 | 2806 | } |
ec4f8605 | 2807 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2808 | |
cdd16d02 | 2809 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2810 | { |
2811 | *buflen -= namelen; | |
2812 | if (*buflen < 0) | |
2813 | return -ENAMETOOLONG; | |
2814 | *buffer -= namelen; | |
2815 | memcpy(*buffer, str, namelen); | |
2816 | return 0; | |
2817 | } | |
2818 | ||
232d2d60 WL |
2819 | /** |
2820 | * prepend_name - prepend a pathname in front of current buffer pointer | |
18129977 WL |
2821 | * @buffer: buffer pointer |
2822 | * @buflen: allocated length of the buffer | |
2823 | * @name: name string and length qstr structure | |
232d2d60 WL |
2824 | * |
2825 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
2826 | * make sure that either the old or the new name pointer and length are | |
2827 | * fetched. However, there may be mismatch between length and pointer. | |
2828 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
2829 | * the length is reached or a null byte is found. It also prepends "/" at | |
2830 | * the beginning of the name. The sequence number check at the caller will | |
2831 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
2832 | * due to mismatched pointer and length will be discarded. | |
6d13f694 AV |
2833 | * |
2834 | * Data dependency barrier is needed to make sure that we see that terminating | |
2835 | * NUL. Alpha strikes again, film at 11... | |
232d2d60 | 2836 | */ |
cdd16d02 MS |
2837 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2838 | { | |
232d2d60 WL |
2839 | const char *dname = ACCESS_ONCE(name->name); |
2840 | u32 dlen = ACCESS_ONCE(name->len); | |
2841 | char *p; | |
2842 | ||
6d13f694 AV |
2843 | smp_read_barrier_depends(); |
2844 | ||
232d2d60 | 2845 | *buflen -= dlen + 1; |
e825196d AV |
2846 | if (*buflen < 0) |
2847 | return -ENAMETOOLONG; | |
232d2d60 WL |
2848 | p = *buffer -= dlen + 1; |
2849 | *p++ = '/'; | |
2850 | while (dlen--) { | |
2851 | char c = *dname++; | |
2852 | if (!c) | |
2853 | break; | |
2854 | *p++ = c; | |
2855 | } | |
2856 | return 0; | |
cdd16d02 MS |
2857 | } |
2858 | ||
1da177e4 | 2859 | /** |
208898c1 | 2860 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2861 | * @path: the dentry/vfsmount to report |
02125a82 | 2862 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2863 | * @buffer: pointer to the end of the buffer |
2864 | * @buflen: pointer to buffer length | |
552ce544 | 2865 | * |
18129977 WL |
2866 | * The function will first try to write out the pathname without taking any |
2867 | * lock other than the RCU read lock to make sure that dentries won't go away. | |
2868 | * It only checks the sequence number of the global rename_lock as any change | |
2869 | * in the dentry's d_seq will be preceded by changes in the rename_lock | |
2870 | * sequence number. If the sequence number had been changed, it will restart | |
2871 | * the whole pathname back-tracing sequence again by taking the rename_lock. | |
2872 | * In this case, there is no need to take the RCU read lock as the recursive | |
2873 | * parent pointer references will keep the dentry chain alive as long as no | |
2874 | * rename operation is performed. | |
1da177e4 | 2875 | */ |
02125a82 AV |
2876 | static int prepend_path(const struct path *path, |
2877 | const struct path *root, | |
f2eb6575 | 2878 | char **buffer, int *buflen) |
1da177e4 | 2879 | { |
ede4cebc AV |
2880 | struct dentry *dentry; |
2881 | struct vfsmount *vfsmnt; | |
2882 | struct mount *mnt; | |
f2eb6575 | 2883 | int error = 0; |
48a066e7 | 2884 | unsigned seq, m_seq = 0; |
232d2d60 WL |
2885 | char *bptr; |
2886 | int blen; | |
6092d048 | 2887 | |
48f5ec21 | 2888 | rcu_read_lock(); |
48a066e7 AV |
2889 | restart_mnt: |
2890 | read_seqbegin_or_lock(&mount_lock, &m_seq); | |
2891 | seq = 0; | |
4ec6c2ae | 2892 | rcu_read_lock(); |
232d2d60 WL |
2893 | restart: |
2894 | bptr = *buffer; | |
2895 | blen = *buflen; | |
48a066e7 | 2896 | error = 0; |
ede4cebc AV |
2897 | dentry = path->dentry; |
2898 | vfsmnt = path->mnt; | |
2899 | mnt = real_mount(vfsmnt); | |
232d2d60 | 2900 | read_seqbegin_or_lock(&rename_lock, &seq); |
f2eb6575 | 2901 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2902 | struct dentry * parent; |
2903 | ||
1da177e4 | 2904 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
48a066e7 | 2905 | struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); |
552ce544 | 2906 | /* Global root? */ |
48a066e7 AV |
2907 | if (mnt != parent) { |
2908 | dentry = ACCESS_ONCE(mnt->mnt_mountpoint); | |
2909 | mnt = parent; | |
232d2d60 WL |
2910 | vfsmnt = &mnt->mnt; |
2911 | continue; | |
2912 | } | |
2913 | /* | |
2914 | * Filesystems needing to implement special "root names" | |
2915 | * should do so with ->d_dname() | |
2916 | */ | |
2917 | if (IS_ROOT(dentry) && | |
2918 | (dentry->d_name.len != 1 || | |
2919 | dentry->d_name.name[0] != '/')) { | |
2920 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2921 | (int) dentry->d_name.len, | |
2922 | dentry->d_name.name); | |
2923 | } | |
2924 | if (!error) | |
2925 | error = is_mounted(vfsmnt) ? 1 : 2; | |
2926 | break; | |
1da177e4 LT |
2927 | } |
2928 | parent = dentry->d_parent; | |
2929 | prefetch(parent); | |
232d2d60 | 2930 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
2931 | if (error) |
2932 | break; | |
2933 | ||
1da177e4 LT |
2934 | dentry = parent; |
2935 | } | |
48f5ec21 AV |
2936 | if (!(seq & 1)) |
2937 | rcu_read_unlock(); | |
2938 | if (need_seqretry(&rename_lock, seq)) { | |
2939 | seq = 1; | |
232d2d60 | 2940 | goto restart; |
48f5ec21 AV |
2941 | } |
2942 | done_seqretry(&rename_lock, seq); | |
4ec6c2ae LZ |
2943 | |
2944 | if (!(m_seq & 1)) | |
2945 | rcu_read_unlock(); | |
48a066e7 AV |
2946 | if (need_seqretry(&mount_lock, m_seq)) { |
2947 | m_seq = 1; | |
2948 | goto restart_mnt; | |
2949 | } | |
2950 | done_seqretry(&mount_lock, m_seq); | |
1da177e4 | 2951 | |
232d2d60 WL |
2952 | if (error >= 0 && bptr == *buffer) { |
2953 | if (--blen < 0) | |
2954 | error = -ENAMETOOLONG; | |
2955 | else | |
2956 | *--bptr = '/'; | |
2957 | } | |
2958 | *buffer = bptr; | |
2959 | *buflen = blen; | |
7ea600b5 | 2960 | return error; |
f2eb6575 | 2961 | } |
be285c71 | 2962 | |
f2eb6575 MS |
2963 | /** |
2964 | * __d_path - return the path of a dentry | |
2965 | * @path: the dentry/vfsmount to report | |
02125a82 | 2966 | * @root: root vfsmnt/dentry |
cd956a1c | 2967 | * @buf: buffer to return value in |
f2eb6575 MS |
2968 | * @buflen: buffer length |
2969 | * | |
ffd1f4ed | 2970 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2971 | * |
2972 | * Returns a pointer into the buffer or an error code if the | |
2973 | * path was too long. | |
2974 | * | |
be148247 | 2975 | * "buflen" should be positive. |
f2eb6575 | 2976 | * |
02125a82 | 2977 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2978 | */ |
02125a82 AV |
2979 | char *__d_path(const struct path *path, |
2980 | const struct path *root, | |
f2eb6575 MS |
2981 | char *buf, int buflen) |
2982 | { | |
2983 | char *res = buf + buflen; | |
2984 | int error; | |
2985 | ||
2986 | prepend(&res, &buflen, "\0", 1); | |
f2eb6575 | 2987 | error = prepend_path(path, root, &res, &buflen); |
be148247 | 2988 | |
02125a82 AV |
2989 | if (error < 0) |
2990 | return ERR_PTR(error); | |
2991 | if (error > 0) | |
2992 | return NULL; | |
2993 | return res; | |
2994 | } | |
2995 | ||
2996 | char *d_absolute_path(const struct path *path, | |
2997 | char *buf, int buflen) | |
2998 | { | |
2999 | struct path root = {}; | |
3000 | char *res = buf + buflen; | |
3001 | int error; | |
3002 | ||
3003 | prepend(&res, &buflen, "\0", 1); | |
02125a82 | 3004 | error = prepend_path(path, &root, &res, &buflen); |
02125a82 AV |
3005 | |
3006 | if (error > 1) | |
3007 | error = -EINVAL; | |
3008 | if (error < 0) | |
f2eb6575 | 3009 | return ERR_PTR(error); |
f2eb6575 | 3010 | return res; |
1da177e4 LT |
3011 | } |
3012 | ||
ffd1f4ed MS |
3013 | /* |
3014 | * same as __d_path but appends "(deleted)" for unlinked files. | |
3015 | */ | |
02125a82 AV |
3016 | static int path_with_deleted(const struct path *path, |
3017 | const struct path *root, | |
3018 | char **buf, int *buflen) | |
ffd1f4ed MS |
3019 | { |
3020 | prepend(buf, buflen, "\0", 1); | |
3021 | if (d_unlinked(path->dentry)) { | |
3022 | int error = prepend(buf, buflen, " (deleted)", 10); | |
3023 | if (error) | |
3024 | return error; | |
3025 | } | |
3026 | ||
3027 | return prepend_path(path, root, buf, buflen); | |
3028 | } | |
3029 | ||
8df9d1a4 MS |
3030 | static int prepend_unreachable(char **buffer, int *buflen) |
3031 | { | |
3032 | return prepend(buffer, buflen, "(unreachable)", 13); | |
3033 | } | |
3034 | ||
68f0d9d9 LT |
3035 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) |
3036 | { | |
3037 | unsigned seq; | |
3038 | ||
3039 | do { | |
3040 | seq = read_seqcount_begin(&fs->seq); | |
3041 | *root = fs->root; | |
3042 | } while (read_seqcount_retry(&fs->seq, seq)); | |
3043 | } | |
3044 | ||
a03a8a70 JB |
3045 | /** |
3046 | * d_path - return the path of a dentry | |
cf28b486 | 3047 | * @path: path to report |
a03a8a70 JB |
3048 | * @buf: buffer to return value in |
3049 | * @buflen: buffer length | |
3050 | * | |
3051 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
3052 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
3053 | * | |
52afeefb AV |
3054 | * Returns a pointer into the buffer or an error code if the path was |
3055 | * too long. Note: Callers should use the returned pointer, not the passed | |
3056 | * in buffer, to use the name! The implementation often starts at an offset | |
3057 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 3058 | * |
31f3e0b3 | 3059 | * "buflen" should be positive. |
a03a8a70 | 3060 | */ |
20d4fdc1 | 3061 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 3062 | { |
ffd1f4ed | 3063 | char *res = buf + buflen; |
6ac08c39 | 3064 | struct path root; |
ffd1f4ed | 3065 | int error; |
1da177e4 | 3066 | |
c23fbb6b ED |
3067 | /* |
3068 | * We have various synthetic filesystems that never get mounted. On | |
3069 | * these filesystems dentries are never used for lookup purposes, and | |
3070 | * thus don't need to be hashed. They also don't need a name until a | |
3071 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
3072 | * below allows us to generate a name for these objects on demand: | |
f48cfddc EB |
3073 | * |
3074 | * Some pseudo inodes are mountable. When they are mounted | |
3075 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname | |
3076 | * and instead have d_path return the mounted path. | |
c23fbb6b | 3077 | */ |
f48cfddc EB |
3078 | if (path->dentry->d_op && path->dentry->d_op->d_dname && |
3079 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) | |
cf28b486 | 3080 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
c23fbb6b | 3081 | |
68f0d9d9 LT |
3082 | rcu_read_lock(); |
3083 | get_fs_root_rcu(current->fs, &root); | |
02125a82 | 3084 | error = path_with_deleted(path, &root, &res, &buflen); |
68f0d9d9 LT |
3085 | rcu_read_unlock(); |
3086 | ||
02125a82 | 3087 | if (error < 0) |
ffd1f4ed | 3088 | res = ERR_PTR(error); |
1da177e4 LT |
3089 | return res; |
3090 | } | |
ec4f8605 | 3091 | EXPORT_SYMBOL(d_path); |
1da177e4 | 3092 | |
c23fbb6b ED |
3093 | /* |
3094 | * Helper function for dentry_operations.d_dname() members | |
3095 | */ | |
3096 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
3097 | const char *fmt, ...) | |
3098 | { | |
3099 | va_list args; | |
3100 | char temp[64]; | |
3101 | int sz; | |
3102 | ||
3103 | va_start(args, fmt); | |
3104 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
3105 | va_end(args); | |
3106 | ||
3107 | if (sz > sizeof(temp) || sz > buflen) | |
3108 | return ERR_PTR(-ENAMETOOLONG); | |
3109 | ||
3110 | buffer += buflen - sz; | |
3111 | return memcpy(buffer, temp, sz); | |
3112 | } | |
3113 | ||
118b2302 AV |
3114 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
3115 | { | |
3116 | char *end = buffer + buflen; | |
3117 | /* these dentries are never renamed, so d_lock is not needed */ | |
3118 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
232d2d60 | 3119 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
118b2302 AV |
3120 | prepend(&end, &buflen, "/", 1)) |
3121 | end = ERR_PTR(-ENAMETOOLONG); | |
232d2d60 | 3122 | return end; |
118b2302 | 3123 | } |
31bbe16f | 3124 | EXPORT_SYMBOL(simple_dname); |
118b2302 | 3125 | |
6092d048 RP |
3126 | /* |
3127 | * Write full pathname from the root of the filesystem into the buffer. | |
3128 | */ | |
f6500801 | 3129 | static char *__dentry_path(struct dentry *d, char *buf, int buflen) |
6092d048 | 3130 | { |
f6500801 | 3131 | struct dentry *dentry; |
232d2d60 WL |
3132 | char *end, *retval; |
3133 | int len, seq = 0; | |
3134 | int error = 0; | |
6092d048 | 3135 | |
f6500801 AV |
3136 | if (buflen < 2) |
3137 | goto Elong; | |
3138 | ||
48f5ec21 | 3139 | rcu_read_lock(); |
232d2d60 | 3140 | restart: |
f6500801 | 3141 | dentry = d; |
232d2d60 WL |
3142 | end = buf + buflen; |
3143 | len = buflen; | |
3144 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
3145 | /* Get '/' right */ |
3146 | retval = end-1; | |
3147 | *retval = '/'; | |
232d2d60 | 3148 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
3149 | while (!IS_ROOT(dentry)) { |
3150 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 3151 | |
6092d048 | 3152 | prefetch(parent); |
232d2d60 WL |
3153 | error = prepend_name(&end, &len, &dentry->d_name); |
3154 | if (error) | |
3155 | break; | |
6092d048 RP |
3156 | |
3157 | retval = end; | |
3158 | dentry = parent; | |
3159 | } | |
48f5ec21 AV |
3160 | if (!(seq & 1)) |
3161 | rcu_read_unlock(); | |
3162 | if (need_seqretry(&rename_lock, seq)) { | |
3163 | seq = 1; | |
232d2d60 | 3164 | goto restart; |
48f5ec21 AV |
3165 | } |
3166 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3167 | if (error) |
3168 | goto Elong; | |
c103135c AV |
3169 | return retval; |
3170 | Elong: | |
3171 | return ERR_PTR(-ENAMETOOLONG); | |
3172 | } | |
ec2447c2 NP |
3173 | |
3174 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3175 | { | |
232d2d60 | 3176 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3177 | } |
3178 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3179 | |
3180 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3181 | { | |
3182 | char *p = NULL; | |
3183 | char *retval; | |
3184 | ||
c103135c AV |
3185 | if (d_unlinked(dentry)) { |
3186 | p = buf + buflen; | |
3187 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3188 | goto Elong; | |
3189 | buflen++; | |
3190 | } | |
3191 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3192 | if (!IS_ERR(retval) && p) |
3193 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3194 | return retval; |
3195 | Elong: | |
6092d048 RP |
3196 | return ERR_PTR(-ENAMETOOLONG); |
3197 | } | |
3198 | ||
8b19e341 LT |
3199 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, |
3200 | struct path *pwd) | |
5762482f | 3201 | { |
8b19e341 LT |
3202 | unsigned seq; |
3203 | ||
3204 | do { | |
3205 | seq = read_seqcount_begin(&fs->seq); | |
3206 | *root = fs->root; | |
3207 | *pwd = fs->pwd; | |
3208 | } while (read_seqcount_retry(&fs->seq, seq)); | |
5762482f LT |
3209 | } |
3210 | ||
1da177e4 LT |
3211 | /* |
3212 | * NOTE! The user-level library version returns a | |
3213 | * character pointer. The kernel system call just | |
3214 | * returns the length of the buffer filled (which | |
3215 | * includes the ending '\0' character), or a negative | |
3216 | * error value. So libc would do something like | |
3217 | * | |
3218 | * char *getcwd(char * buf, size_t size) | |
3219 | * { | |
3220 | * int retval; | |
3221 | * | |
3222 | * retval = sys_getcwd(buf, size); | |
3223 | * if (retval >= 0) | |
3224 | * return buf; | |
3225 | * errno = -retval; | |
3226 | * return NULL; | |
3227 | * } | |
3228 | */ | |
3cdad428 | 3229 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3230 | { |
552ce544 | 3231 | int error; |
6ac08c39 | 3232 | struct path pwd, root; |
3272c544 | 3233 | char *page = __getname(); |
1da177e4 LT |
3234 | |
3235 | if (!page) | |
3236 | return -ENOMEM; | |
3237 | ||
8b19e341 LT |
3238 | rcu_read_lock(); |
3239 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); | |
1da177e4 | 3240 | |
552ce544 | 3241 | error = -ENOENT; |
f3da392e | 3242 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3243 | unsigned long len; |
3272c544 LT |
3244 | char *cwd = page + PATH_MAX; |
3245 | int buflen = PATH_MAX; | |
1da177e4 | 3246 | |
8df9d1a4 | 3247 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3248 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
ff812d72 | 3249 | rcu_read_unlock(); |
552ce544 | 3250 | |
02125a82 | 3251 | if (error < 0) |
552ce544 LT |
3252 | goto out; |
3253 | ||
8df9d1a4 | 3254 | /* Unreachable from current root */ |
02125a82 | 3255 | if (error > 0) { |
8df9d1a4 MS |
3256 | error = prepend_unreachable(&cwd, &buflen); |
3257 | if (error) | |
3258 | goto out; | |
3259 | } | |
3260 | ||
552ce544 | 3261 | error = -ERANGE; |
3272c544 | 3262 | len = PATH_MAX + page - cwd; |
552ce544 LT |
3263 | if (len <= size) { |
3264 | error = len; | |
3265 | if (copy_to_user(buf, cwd, len)) | |
3266 | error = -EFAULT; | |
3267 | } | |
949854d0 | 3268 | } else { |
ff812d72 | 3269 | rcu_read_unlock(); |
949854d0 | 3270 | } |
1da177e4 LT |
3271 | |
3272 | out: | |
3272c544 | 3273 | __putname(page); |
1da177e4 LT |
3274 | return error; |
3275 | } | |
3276 | ||
3277 | /* | |
3278 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3279 | * | |
3280 | * Trivially implemented using the dcache structure | |
3281 | */ | |
3282 | ||
3283 | /** | |
3284 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3285 | * @new_dentry: new dentry | |
3286 | * @old_dentry: old dentry | |
3287 | * | |
3288 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
3289 | * Returns 0 otherwise. | |
3290 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
3291 | */ | |
3292 | ||
e2761a11 | 3293 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
3294 | { |
3295 | int result; | |
949854d0 | 3296 | unsigned seq; |
1da177e4 | 3297 | |
e2761a11 OH |
3298 | if (new_dentry == old_dentry) |
3299 | return 1; | |
3300 | ||
e2761a11 | 3301 | do { |
1da177e4 | 3302 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3303 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3304 | /* |
3305 | * Need rcu_readlock to protect against the d_parent trashing | |
3306 | * due to d_move | |
3307 | */ | |
3308 | rcu_read_lock(); | |
e2761a11 | 3309 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 3310 | result = 1; |
e2761a11 OH |
3311 | else |
3312 | result = 0; | |
949854d0 | 3313 | rcu_read_unlock(); |
1da177e4 | 3314 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3315 | |
3316 | return result; | |
3317 | } | |
3318 | ||
db14fc3a | 3319 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3320 | { |
db14fc3a MS |
3321 | struct dentry *root = data; |
3322 | if (dentry != root) { | |
3323 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3324 | return D_WALK_SKIP; | |
1da177e4 | 3325 | |
01ddc4ed MS |
3326 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3327 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3328 | dentry->d_lockref.count--; | |
3329 | } | |
1da177e4 | 3330 | } |
db14fc3a MS |
3331 | return D_WALK_CONTINUE; |
3332 | } | |
58db63d0 | 3333 | |
db14fc3a MS |
3334 | void d_genocide(struct dentry *parent) |
3335 | { | |
3336 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3337 | } |
3338 | ||
60545d0d | 3339 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3340 | { |
60545d0d AV |
3341 | inode_dec_link_count(inode); |
3342 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
3343 | !hlist_unhashed(&dentry->d_alias) || | |
3344 | !d_unlinked(dentry)); | |
3345 | spin_lock(&dentry->d_parent->d_lock); | |
3346 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3347 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3348 | (unsigned long long)inode->i_ino); | |
3349 | spin_unlock(&dentry->d_lock); | |
3350 | spin_unlock(&dentry->d_parent->d_lock); | |
3351 | d_instantiate(dentry, inode); | |
1da177e4 | 3352 | } |
60545d0d | 3353 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3354 | |
3355 | static __initdata unsigned long dhash_entries; | |
3356 | static int __init set_dhash_entries(char *str) | |
3357 | { | |
3358 | if (!str) | |
3359 | return 0; | |
3360 | dhash_entries = simple_strtoul(str, &str, 0); | |
3361 | return 1; | |
3362 | } | |
3363 | __setup("dhash_entries=", set_dhash_entries); | |
3364 | ||
3365 | static void __init dcache_init_early(void) | |
3366 | { | |
074b8517 | 3367 | unsigned int loop; |
1da177e4 LT |
3368 | |
3369 | /* If hashes are distributed across NUMA nodes, defer | |
3370 | * hash allocation until vmalloc space is available. | |
3371 | */ | |
3372 | if (hashdist) | |
3373 | return; | |
3374 | ||
3375 | dentry_hashtable = | |
3376 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3377 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3378 | dhash_entries, |
3379 | 13, | |
3380 | HASH_EARLY, | |
3381 | &d_hash_shift, | |
3382 | &d_hash_mask, | |
31fe62b9 | 3383 | 0, |
1da177e4 LT |
3384 | 0); |
3385 | ||
074b8517 | 3386 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3387 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3388 | } |
3389 | ||
74bf17cf | 3390 | static void __init dcache_init(void) |
1da177e4 | 3391 | { |
074b8517 | 3392 | unsigned int loop; |
1da177e4 LT |
3393 | |
3394 | /* | |
3395 | * A constructor could be added for stable state like the lists, | |
3396 | * but it is probably not worth it because of the cache nature | |
3397 | * of the dcache. | |
3398 | */ | |
0a31bd5f CL |
3399 | dentry_cache = KMEM_CACHE(dentry, |
3400 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3401 | |
3402 | /* Hash may have been set up in dcache_init_early */ | |
3403 | if (!hashdist) | |
3404 | return; | |
3405 | ||
3406 | dentry_hashtable = | |
3407 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3408 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3409 | dhash_entries, |
3410 | 13, | |
3411 | 0, | |
3412 | &d_hash_shift, | |
3413 | &d_hash_mask, | |
31fe62b9 | 3414 | 0, |
1da177e4 LT |
3415 | 0); |
3416 | ||
074b8517 | 3417 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3418 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3419 | } |
3420 | ||
3421 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3422 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3423 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3424 | |
1da177e4 LT |
3425 | EXPORT_SYMBOL(d_genocide); |
3426 | ||
1da177e4 LT |
3427 | void __init vfs_caches_init_early(void) |
3428 | { | |
3429 | dcache_init_early(); | |
3430 | inode_init_early(); | |
3431 | } | |
3432 | ||
3433 | void __init vfs_caches_init(unsigned long mempages) | |
3434 | { | |
3435 | unsigned long reserve; | |
3436 | ||
3437 | /* Base hash sizes on available memory, with a reserve equal to | |
3438 | 150% of current kernel size */ | |
3439 | ||
3440 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3441 | mempages -= reserve; | |
3442 | ||
3443 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3444 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3445 | |
74bf17cf DC |
3446 | dcache_init(); |
3447 | inode_init(); | |
1da177e4 | 3448 | files_init(mempages); |
74bf17cf | 3449 | mnt_init(); |
1da177e4 LT |
3450 | bdev_cache_init(); |
3451 | chrdev_init(); | |
3452 | } |