]>
Commit | Line | Data |
---|---|---|
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 | ||
17 | #include <linux/syscalls.h> | |
18 | #include <linux/string.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/fsnotify.h> | |
22 | #include <linux/slab.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/hash.h> | |
25 | #include <linux/cache.h> | |
26 | #include <linux/export.h> | |
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> | |
34 | #include <linux/fs_struct.h> | |
35 | #include <linux/hardirq.h> | |
36 | #include <linux/bit_spinlock.h> | |
37 | #include <linux/rculist_bl.h> | |
38 | #include <linux/prefetch.h> | |
39 | #include <linux/ratelimit.h> | |
40 | #include <linux/list_lru.h> | |
41 | #include "internal.h" | |
42 | #include "mount.h" | |
43 | ||
44 | /* | |
45 | * Usage: | |
46 | * dcache->d_inode->i_lock protects: | |
47 | * - i_dentry, d_alias, d_inode of aliases | |
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) | |
52 | * dentry->d_sb->s_dentry_lru_lock protects: | |
53 | * - the dcache lru lists and counters | |
54 | * d_lock protects: | |
55 | * - d_flags | |
56 | * - d_name | |
57 | * - d_lru | |
58 | * - d_count | |
59 | * - d_unhashed() | |
60 | * - d_parent and d_subdirs | |
61 | * - childrens' d_child and d_parent | |
62 | * - d_alias, d_inode | |
63 | * | |
64 | * Ordering: | |
65 | * dentry->d_inode->i_lock | |
66 | * dentry->d_lock | |
67 | * dentry->d_sb->s_dentry_lru_lock | |
68 | * dcache_hash_bucket lock | |
69 | * s_anon lock | |
70 | * | |
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: | |
78 | * if (dentry1 < dentry2) | |
79 | * dentry1->d_lock | |
80 | * dentry2->d_lock | |
81 | */ | |
82 | int sysctl_vfs_cache_pressure __read_mostly = 100; | |
83 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); | |
84 | ||
85 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); | |
86 | ||
87 | EXPORT_SYMBOL(rename_lock); | |
88 | ||
89 | static struct kmem_cache *dentry_cache __read_mostly; | |
90 | ||
91 | /** | |
92 | * read_seqbegin_or_lock - begin a sequence number check or locking block | |
93 | * @lock: sequence lock | |
94 | * @seq : sequence number to be checked | |
95 | * | |
96 | * First try it once optimistically without taking the lock. If that fails, | |
97 | * take the lock. The sequence number is also used as a marker for deciding | |
98 | * whether to be a reader (even) or writer (odd). | |
99 | * N.B. seq must be initialized to an even number to begin with. | |
100 | */ | |
101 | static inline void read_seqbegin_or_lock(seqlock_t *lock, int *seq) | |
102 | { | |
103 | if (!(*seq & 1)) /* Even */ | |
104 | *seq = read_seqbegin(lock); | |
105 | else /* Odd */ | |
106 | read_seqlock_excl(lock); | |
107 | } | |
108 | ||
109 | static inline int need_seqretry(seqlock_t *lock, int seq) | |
110 | { | |
111 | return !(seq & 1) && read_seqretry(lock, seq); | |
112 | } | |
113 | ||
114 | static inline void done_seqretry(seqlock_t *lock, int seq) | |
115 | { | |
116 | if (seq & 1) | |
117 | read_sequnlock_excl(lock); | |
118 | } | |
119 | ||
120 | /* | |
121 | * This is the single most critical data structure when it comes | |
122 | * to the dcache: the hashtable for lookups. Somebody should try | |
123 | * to make this good - I've just made it work. | |
124 | * | |
125 | * This hash-function tries to avoid losing too many bits of hash | |
126 | * information, yet avoid using a prime hash-size or similar. | |
127 | */ | |
128 | #define D_HASHBITS d_hash_shift | |
129 | #define D_HASHMASK d_hash_mask | |
130 | ||
131 | static unsigned int d_hash_mask __read_mostly; | |
132 | static unsigned int d_hash_shift __read_mostly; | |
133 | ||
134 | static struct hlist_bl_head *dentry_hashtable __read_mostly; | |
135 | ||
136 | static inline struct hlist_bl_head *d_hash(const struct dentry *parent, | |
137 | unsigned int hash) | |
138 | { | |
139 | hash += (unsigned long) parent / L1_CACHE_BYTES; | |
140 | hash = hash + (hash >> D_HASHBITS); | |
141 | return dentry_hashtable + (hash & D_HASHMASK); | |
142 | } | |
143 | ||
144 | /* Statistics gathering. */ | |
145 | struct dentry_stat_t dentry_stat = { | |
146 | .age_limit = 45, | |
147 | }; | |
148 | ||
149 | static DEFINE_PER_CPU(long, nr_dentry); | |
150 | static DEFINE_PER_CPU(long, nr_dentry_unused); | |
151 | ||
152 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
153 | ||
154 | /* | |
155 | * Here we resort to our own counters instead of using generic per-cpu counters | |
156 | * for consistency with what the vfs inode code does. We are expected to harvest | |
157 | * better code and performance by having our own specialized counters. | |
158 | * | |
159 | * Please note that the loop is done over all possible CPUs, not over all online | |
160 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
161 | * on and off. If one of them goes off, we will just keep their counters. | |
162 | * | |
163 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
164 | * please update all vfs counters to match. | |
165 | */ | |
166 | static long get_nr_dentry(void) | |
167 | { | |
168 | int i; | |
169 | long sum = 0; | |
170 | for_each_possible_cpu(i) | |
171 | sum += per_cpu(nr_dentry, i); | |
172 | return sum < 0 ? 0 : sum; | |
173 | } | |
174 | ||
175 | static long get_nr_dentry_unused(void) | |
176 | { | |
177 | int i; | |
178 | long sum = 0; | |
179 | for_each_possible_cpu(i) | |
180 | sum += per_cpu(nr_dentry_unused, i); | |
181 | return sum < 0 ? 0 : sum; | |
182 | } | |
183 | ||
184 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, | |
185 | size_t *lenp, loff_t *ppos) | |
186 | { | |
187 | dentry_stat.nr_dentry = get_nr_dentry(); | |
188 | dentry_stat.nr_unused = get_nr_dentry_unused(); | |
189 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); | |
190 | } | |
191 | #endif | |
192 | ||
193 | /* | |
194 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
195 | * The strings are both count bytes long, and count is non-zero. | |
196 | */ | |
197 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
198 | ||
199 | #include <asm/word-at-a-time.h> | |
200 | /* | |
201 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
202 | * aligned allocation for this particular component. We don't | |
203 | * strictly need the load_unaligned_zeropad() safety, but it | |
204 | * doesn't hurt either. | |
205 | * | |
206 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
207 | * need the careful unaligned handling. | |
208 | */ | |
209 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) | |
210 | { | |
211 | unsigned long a,b,mask; | |
212 | ||
213 | for (;;) { | |
214 | a = *(unsigned long *)cs; | |
215 | b = load_unaligned_zeropad(ct); | |
216 | if (tcount < sizeof(unsigned long)) | |
217 | break; | |
218 | if (unlikely(a != b)) | |
219 | return 1; | |
220 | cs += sizeof(unsigned long); | |
221 | ct += sizeof(unsigned long); | |
222 | tcount -= sizeof(unsigned long); | |
223 | if (!tcount) | |
224 | return 0; | |
225 | } | |
226 | mask = ~(~0ul << tcount*8); | |
227 | return unlikely(!!((a ^ b) & mask)); | |
228 | } | |
229 | ||
230 | #else | |
231 | ||
232 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) | |
233 | { | |
234 | do { | |
235 | if (*cs != *ct) | |
236 | return 1; | |
237 | cs++; | |
238 | ct++; | |
239 | tcount--; | |
240 | } while (tcount); | |
241 | return 0; | |
242 | } | |
243 | ||
244 | #endif | |
245 | ||
246 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) | |
247 | { | |
248 | const unsigned char *cs; | |
249 | /* | |
250 | * Be careful about RCU walk racing with rename: | |
251 | * use ACCESS_ONCE to fetch the name pointer. | |
252 | * | |
253 | * NOTE! Even if a rename will mean that the length | |
254 | * was not loaded atomically, we don't care. The | |
255 | * RCU walk will check the sequence count eventually, | |
256 | * and catch it. And we won't overrun the buffer, | |
257 | * because we're reading the name pointer atomically, | |
258 | * and a dentry name is guaranteed to be properly | |
259 | * terminated with a NUL byte. | |
260 | * | |
261 | * End result: even if 'len' is wrong, we'll exit | |
262 | * early because the data cannot match (there can | |
263 | * be no NUL in the ct/tcount data) | |
264 | */ | |
265 | cs = ACCESS_ONCE(dentry->d_name.name); | |
266 | smp_read_barrier_depends(); | |
267 | return dentry_string_cmp(cs, ct, tcount); | |
268 | } | |
269 | ||
270 | static void __d_free(struct rcu_head *head) | |
271 | { | |
272 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
273 | ||
274 | WARN_ON(!hlist_unhashed(&dentry->d_alias)); | |
275 | if (dname_external(dentry)) | |
276 | kfree(dentry->d_name.name); | |
277 | kmem_cache_free(dentry_cache, dentry); | |
278 | } | |
279 | ||
280 | /* | |
281 | * no locks, please. | |
282 | */ | |
283 | static void d_free(struct dentry *dentry) | |
284 | { | |
285 | BUG_ON((int)dentry->d_lockref.count > 0); | |
286 | this_cpu_dec(nr_dentry); | |
287 | if (dentry->d_op && dentry->d_op->d_release) | |
288 | dentry->d_op->d_release(dentry); | |
289 | ||
290 | /* if dentry was never visible to RCU, immediate free is OK */ | |
291 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
292 | __d_free(&dentry->d_u.d_rcu); | |
293 | else | |
294 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
295 | } | |
296 | ||
297 | /** | |
298 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
299 | * @dentry: the target dentry | |
300 | * After this call, in-progress rcu-walk path lookup will fail. This | |
301 | * should be called after unhashing, and after changing d_inode (if | |
302 | * the dentry has not already been unhashed). | |
303 | */ | |
304 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
305 | { | |
306 | assert_spin_locked(&dentry->d_lock); | |
307 | /* Go through a barrier */ | |
308 | write_seqcount_barrier(&dentry->d_seq); | |
309 | } | |
310 | ||
311 | /* | |
312 | * Release the dentry's inode, using the filesystem | |
313 | * d_iput() operation if defined. Dentry has no refcount | |
314 | * and is unhashed. | |
315 | */ | |
316 | static void dentry_iput(struct dentry * dentry) | |
317 | __releases(dentry->d_lock) | |
318 | __releases(dentry->d_inode->i_lock) | |
319 | { | |
320 | struct inode *inode = dentry->d_inode; | |
321 | if (inode) { | |
322 | dentry->d_inode = NULL; | |
323 | hlist_del_init(&dentry->d_alias); | |
324 | spin_unlock(&dentry->d_lock); | |
325 | spin_unlock(&inode->i_lock); | |
326 | if (!inode->i_nlink) | |
327 | fsnotify_inoderemove(inode); | |
328 | if (dentry->d_op && dentry->d_op->d_iput) | |
329 | dentry->d_op->d_iput(dentry, inode); | |
330 | else | |
331 | iput(inode); | |
332 | } else { | |
333 | spin_unlock(&dentry->d_lock); | |
334 | } | |
335 | } | |
336 | ||
337 | /* | |
338 | * Release the dentry's inode, using the filesystem | |
339 | * d_iput() operation if defined. dentry remains in-use. | |
340 | */ | |
341 | static void dentry_unlink_inode(struct dentry * dentry) | |
342 | __releases(dentry->d_lock) | |
343 | __releases(dentry->d_inode->i_lock) | |
344 | { | |
345 | struct inode *inode = dentry->d_inode; | |
346 | __d_clear_type(dentry); | |
347 | dentry->d_inode = NULL; | |
348 | hlist_del_init(&dentry->d_alias); | |
349 | dentry_rcuwalk_barrier(dentry); | |
350 | spin_unlock(&dentry->d_lock); | |
351 | spin_unlock(&inode->i_lock); | |
352 | if (!inode->i_nlink) | |
353 | fsnotify_inoderemove(inode); | |
354 | if (dentry->d_op && dentry->d_op->d_iput) | |
355 | dentry->d_op->d_iput(dentry, inode); | |
356 | else | |
357 | iput(inode); | |
358 | } | |
359 | ||
360 | /* | |
361 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
362 | * is in use - which includes both the "real" per-superblock | |
363 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
364 | * | |
365 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
366 | * on the shrink list (ie not on the superblock LRU list). | |
367 | * | |
368 | * The per-cpu "nr_dentry_unused" counters are updated with | |
369 | * the DCACHE_LRU_LIST bit. | |
370 | * | |
371 | * These helper functions make sure we always follow the | |
372 | * rules. d_lock must be held by the caller. | |
373 | */ | |
374 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
375 | static void d_lru_add(struct dentry *dentry) | |
376 | { | |
377 | D_FLAG_VERIFY(dentry, 0); | |
378 | dentry->d_flags |= DCACHE_LRU_LIST; | |
379 | this_cpu_inc(nr_dentry_unused); | |
380 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
381 | } | |
382 | ||
383 | static void d_lru_del(struct dentry *dentry) | |
384 | { | |
385 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
386 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
387 | this_cpu_dec(nr_dentry_unused); | |
388 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
389 | } | |
390 | ||
391 | static void d_shrink_del(struct dentry *dentry) | |
392 | { | |
393 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
394 | list_del_init(&dentry->d_lru); | |
395 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
396 | this_cpu_dec(nr_dentry_unused); | |
397 | } | |
398 | ||
399 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
400 | { | |
401 | D_FLAG_VERIFY(dentry, 0); | |
402 | list_add(&dentry->d_lru, list); | |
403 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
404 | this_cpu_inc(nr_dentry_unused); | |
405 | } | |
406 | ||
407 | /* | |
408 | * These can only be called under the global LRU lock, ie during the | |
409 | * callback for freeing the LRU list. "isolate" removes it from the | |
410 | * LRU lists entirely, while shrink_move moves it to the indicated | |
411 | * private list. | |
412 | */ | |
413 | static void d_lru_isolate(struct dentry *dentry) | |
414 | { | |
415 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
416 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
417 | this_cpu_dec(nr_dentry_unused); | |
418 | list_del_init(&dentry->d_lru); | |
419 | } | |
420 | ||
421 | static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list) | |
422 | { | |
423 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
424 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
425 | list_move_tail(&dentry->d_lru, list); | |
426 | } | |
427 | ||
428 | /* | |
429 | * dentry_lru_(add|del)_list) must be called with d_lock held. | |
430 | */ | |
431 | static void dentry_lru_add(struct dentry *dentry) | |
432 | { | |
433 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) | |
434 | d_lru_add(dentry); | |
435 | } | |
436 | ||
437 | /* | |
438 | * Remove a dentry with references from the LRU. | |
439 | * | |
440 | * If we are on the shrink list, then we can get to try_prune_one_dentry() and | |
441 | * lose our last reference through the parent walk. In this case, we need to | |
442 | * remove ourselves from the shrink list, not the LRU. | |
443 | */ | |
444 | static void dentry_lru_del(struct dentry *dentry) | |
445 | { | |
446 | if (dentry->d_flags & DCACHE_LRU_LIST) { | |
447 | if (dentry->d_flags & DCACHE_SHRINK_LIST) | |
448 | return d_shrink_del(dentry); | |
449 | d_lru_del(dentry); | |
450 | } | |
451 | } | |
452 | ||
453 | /** | |
454 | * d_kill - kill dentry and return parent | |
455 | * @dentry: dentry to kill | |
456 | * @parent: parent dentry | |
457 | * | |
458 | * The dentry must already be unhashed and removed from the LRU. | |
459 | * | |
460 | * If this is the root of the dentry tree, return NULL. | |
461 | * | |
462 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by | |
463 | * d_kill. | |
464 | */ | |
465 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) | |
466 | __releases(dentry->d_lock) | |
467 | __releases(parent->d_lock) | |
468 | __releases(dentry->d_inode->i_lock) | |
469 | { | |
470 | list_del(&dentry->d_u.d_child); | |
471 | /* | |
472 | * Inform try_to_ascend() that we are no longer attached to the | |
473 | * dentry tree | |
474 | */ | |
475 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
476 | if (parent) | |
477 | spin_unlock(&parent->d_lock); | |
478 | dentry_iput(dentry); | |
479 | /* | |
480 | * dentry_iput drops the locks, at which point nobody (except | |
481 | * transient RCU lookups) can reach this dentry. | |
482 | */ | |
483 | d_free(dentry); | |
484 | return parent; | |
485 | } | |
486 | ||
487 | /** | |
488 | * d_drop - drop a dentry | |
489 | * @dentry: dentry to drop | |
490 | * | |
491 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
492 | * be found through a VFS lookup any more. Note that this is different from | |
493 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
494 | * possible, giving a successful _negative_ lookup, while d_drop will | |
495 | * just make the cache lookup fail. | |
496 | * | |
497 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
498 | * reason (NFS timeouts or autofs deletes). | |
499 | * | |
500 | * __d_drop requires dentry->d_lock. | |
501 | */ | |
502 | void __d_drop(struct dentry *dentry) | |
503 | { | |
504 | if (!d_unhashed(dentry)) { | |
505 | struct hlist_bl_head *b; | |
506 | /* | |
507 | * Hashed dentries are normally on the dentry hashtable, | |
508 | * with the exception of those newly allocated by | |
509 | * d_obtain_alias, which are always IS_ROOT: | |
510 | */ | |
511 | if (unlikely(IS_ROOT(dentry))) | |
512 | b = &dentry->d_sb->s_anon; | |
513 | else | |
514 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
515 | ||
516 | hlist_bl_lock(b); | |
517 | __hlist_bl_del(&dentry->d_hash); | |
518 | dentry->d_hash.pprev = NULL; | |
519 | hlist_bl_unlock(b); | |
520 | dentry_rcuwalk_barrier(dentry); | |
521 | } | |
522 | } | |
523 | EXPORT_SYMBOL(__d_drop); | |
524 | ||
525 | void d_drop(struct dentry *dentry) | |
526 | { | |
527 | spin_lock(&dentry->d_lock); | |
528 | __d_drop(dentry); | |
529 | spin_unlock(&dentry->d_lock); | |
530 | } | |
531 | EXPORT_SYMBOL(d_drop); | |
532 | ||
533 | /* | |
534 | * Finish off a dentry we've decided to kill. | |
535 | * dentry->d_lock must be held, returns with it unlocked. | |
536 | * If ref is non-zero, then decrement the refcount too. | |
537 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
538 | */ | |
539 | static struct dentry * | |
540 | dentry_kill(struct dentry *dentry, int unlock_on_failure) | |
541 | __releases(dentry->d_lock) | |
542 | { | |
543 | struct inode *inode; | |
544 | struct dentry *parent; | |
545 | ||
546 | inode = dentry->d_inode; | |
547 | if (inode && !spin_trylock(&inode->i_lock)) { | |
548 | relock: | |
549 | if (unlock_on_failure) { | |
550 | spin_unlock(&dentry->d_lock); | |
551 | cpu_relax(); | |
552 | } | |
553 | return dentry; /* try again with same dentry */ | |
554 | } | |
555 | if (IS_ROOT(dentry)) | |
556 | parent = NULL; | |
557 | else | |
558 | parent = dentry->d_parent; | |
559 | if (parent && !spin_trylock(&parent->d_lock)) { | |
560 | if (inode) | |
561 | spin_unlock(&inode->i_lock); | |
562 | goto relock; | |
563 | } | |
564 | ||
565 | /* | |
566 | * The dentry is now unrecoverably dead to the world. | |
567 | */ | |
568 | lockref_mark_dead(&dentry->d_lockref); | |
569 | ||
570 | /* | |
571 | * inform the fs via d_prune that this dentry is about to be | |
572 | * unhashed and destroyed. | |
573 | */ | |
574 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry)) | |
575 | dentry->d_op->d_prune(dentry); | |
576 | ||
577 | dentry_lru_del(dentry); | |
578 | /* if it was on the hash then remove it */ | |
579 | __d_drop(dentry); | |
580 | return d_kill(dentry, parent); | |
581 | } | |
582 | ||
583 | /* | |
584 | * This is dput | |
585 | * | |
586 | * This is complicated by the fact that we do not want to put | |
587 | * dentries that are no longer on any hash chain on the unused | |
588 | * list: we'd much rather just get rid of them immediately. | |
589 | * | |
590 | * However, that implies that we have to traverse the dentry | |
591 | * tree upwards to the parents which might _also_ now be | |
592 | * scheduled for deletion (it may have been only waiting for | |
593 | * its last child to go away). | |
594 | * | |
595 | * This tail recursion is done by hand as we don't want to depend | |
596 | * on the compiler to always get this right (gcc generally doesn't). | |
597 | * Real recursion would eat up our stack space. | |
598 | */ | |
599 | ||
600 | /* | |
601 | * dput - release a dentry | |
602 | * @dentry: dentry to release | |
603 | * | |
604 | * Release a dentry. This will drop the usage count and if appropriate | |
605 | * call the dentry unlink method as well as removing it from the queues and | |
606 | * releasing its resources. If the parent dentries were scheduled for release | |
607 | * they too may now get deleted. | |
608 | */ | |
609 | void dput(struct dentry *dentry) | |
610 | { | |
611 | if (unlikely(!dentry)) | |
612 | return; | |
613 | ||
614 | repeat: | |
615 | if (lockref_put_or_lock(&dentry->d_lockref)) | |
616 | return; | |
617 | ||
618 | /* Unreachable? Get rid of it */ | |
619 | if (unlikely(d_unhashed(dentry))) | |
620 | goto kill_it; | |
621 | ||
622 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
623 | if (dentry->d_op->d_delete(dentry)) | |
624 | goto kill_it; | |
625 | } | |
626 | ||
627 | if (!(dentry->d_flags & DCACHE_REFERENCED)) | |
628 | dentry->d_flags |= DCACHE_REFERENCED; | |
629 | dentry_lru_add(dentry); | |
630 | ||
631 | dentry->d_lockref.count--; | |
632 | spin_unlock(&dentry->d_lock); | |
633 | return; | |
634 | ||
635 | kill_it: | |
636 | dentry = dentry_kill(dentry, 1); | |
637 | if (dentry) | |
638 | goto repeat; | |
639 | } | |
640 | EXPORT_SYMBOL(dput); | |
641 | ||
642 | /** | |
643 | * d_invalidate - invalidate a dentry | |
644 | * @dentry: dentry to invalidate | |
645 | * | |
646 | * Try to invalidate the dentry if it turns out to be | |
647 | * possible. If there are other dentries that can be | |
648 | * reached through this one we can't delete it and we | |
649 | * return -EBUSY. On success we return 0. | |
650 | * | |
651 | * no dcache lock. | |
652 | */ | |
653 | ||
654 | int d_invalidate(struct dentry * dentry) | |
655 | { | |
656 | /* | |
657 | * If it's already been dropped, return OK. | |
658 | */ | |
659 | spin_lock(&dentry->d_lock); | |
660 | if (d_unhashed(dentry)) { | |
661 | spin_unlock(&dentry->d_lock); | |
662 | return 0; | |
663 | } | |
664 | /* | |
665 | * Check whether to do a partial shrink_dcache | |
666 | * to get rid of unused child entries. | |
667 | */ | |
668 | if (!list_empty(&dentry->d_subdirs)) { | |
669 | spin_unlock(&dentry->d_lock); | |
670 | shrink_dcache_parent(dentry); | |
671 | spin_lock(&dentry->d_lock); | |
672 | } | |
673 | ||
674 | /* | |
675 | * Somebody else still using it? | |
676 | * | |
677 | * If it's a directory, we can't drop it | |
678 | * for fear of somebody re-populating it | |
679 | * with children (even though dropping it | |
680 | * would make it unreachable from the root, | |
681 | * we might still populate it if it was a | |
682 | * working directory or similar). | |
683 | * We also need to leave mountpoints alone, | |
684 | * directory or not. | |
685 | */ | |
686 | if (dentry->d_lockref.count > 1 && dentry->d_inode) { | |
687 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { | |
688 | spin_unlock(&dentry->d_lock); | |
689 | return -EBUSY; | |
690 | } | |
691 | } | |
692 | ||
693 | __d_drop(dentry); | |
694 | spin_unlock(&dentry->d_lock); | |
695 | return 0; | |
696 | } | |
697 | EXPORT_SYMBOL(d_invalidate); | |
698 | ||
699 | /* This must be called with d_lock held */ | |
700 | static inline void __dget_dlock(struct dentry *dentry) | |
701 | { | |
702 | dentry->d_lockref.count++; | |
703 | } | |
704 | ||
705 | static inline void __dget(struct dentry *dentry) | |
706 | { | |
707 | lockref_get(&dentry->d_lockref); | |
708 | } | |
709 | ||
710 | struct dentry *dget_parent(struct dentry *dentry) | |
711 | { | |
712 | int gotref; | |
713 | struct dentry *ret; | |
714 | ||
715 | /* | |
716 | * Do optimistic parent lookup without any | |
717 | * locking. | |
718 | */ | |
719 | rcu_read_lock(); | |
720 | ret = ACCESS_ONCE(dentry->d_parent); | |
721 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
722 | rcu_read_unlock(); | |
723 | if (likely(gotref)) { | |
724 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
725 | return ret; | |
726 | dput(ret); | |
727 | } | |
728 | ||
729 | repeat: | |
730 | /* | |
731 | * Don't need rcu_dereference because we re-check it was correct under | |
732 | * the lock. | |
733 | */ | |
734 | rcu_read_lock(); | |
735 | ret = dentry->d_parent; | |
736 | spin_lock(&ret->d_lock); | |
737 | if (unlikely(ret != dentry->d_parent)) { | |
738 | spin_unlock(&ret->d_lock); | |
739 | rcu_read_unlock(); | |
740 | goto repeat; | |
741 | } | |
742 | rcu_read_unlock(); | |
743 | BUG_ON(!ret->d_lockref.count); | |
744 | ret->d_lockref.count++; | |
745 | spin_unlock(&ret->d_lock); | |
746 | return ret; | |
747 | } | |
748 | EXPORT_SYMBOL(dget_parent); | |
749 | ||
750 | /** | |
751 | * d_find_alias - grab a hashed alias of inode | |
752 | * @inode: inode in question | |
753 | * @want_discon: flag, used by d_splice_alias, to request | |
754 | * that only a DISCONNECTED alias be returned. | |
755 | * | |
756 | * If inode has a hashed alias, or is a directory and has any alias, | |
757 | * acquire the reference to alias and return it. Otherwise return NULL. | |
758 | * Notice that if inode is a directory there can be only one alias and | |
759 | * it can be unhashed only if it has no children, or if it is the root | |
760 | * of a filesystem. | |
761 | * | |
762 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer | |
763 | * any other hashed alias over that one unless @want_discon is set, | |
764 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. | |
765 | */ | |
766 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) | |
767 | { | |
768 | struct dentry *alias, *discon_alias; | |
769 | ||
770 | again: | |
771 | discon_alias = NULL; | |
772 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
773 | spin_lock(&alias->d_lock); | |
774 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
775 | if (IS_ROOT(alias) && | |
776 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
777 | discon_alias = alias; | |
778 | } else if (!want_discon) { | |
779 | __dget_dlock(alias); | |
780 | spin_unlock(&alias->d_lock); | |
781 | return alias; | |
782 | } | |
783 | } | |
784 | spin_unlock(&alias->d_lock); | |
785 | } | |
786 | if (discon_alias) { | |
787 | alias = discon_alias; | |
788 | spin_lock(&alias->d_lock); | |
789 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
790 | if (IS_ROOT(alias) && | |
791 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
792 | __dget_dlock(alias); | |
793 | spin_unlock(&alias->d_lock); | |
794 | return alias; | |
795 | } | |
796 | } | |
797 | spin_unlock(&alias->d_lock); | |
798 | goto again; | |
799 | } | |
800 | return NULL; | |
801 | } | |
802 | ||
803 | struct dentry *d_find_alias(struct inode *inode) | |
804 | { | |
805 | struct dentry *de = NULL; | |
806 | ||
807 | if (!hlist_empty(&inode->i_dentry)) { | |
808 | spin_lock(&inode->i_lock); | |
809 | de = __d_find_alias(inode, 0); | |
810 | spin_unlock(&inode->i_lock); | |
811 | } | |
812 | return de; | |
813 | } | |
814 | EXPORT_SYMBOL(d_find_alias); | |
815 | ||
816 | /* | |
817 | * Try to kill dentries associated with this inode. | |
818 | * WARNING: you must own a reference to inode. | |
819 | */ | |
820 | void d_prune_aliases(struct inode *inode) | |
821 | { | |
822 | struct dentry *dentry; | |
823 | restart: | |
824 | spin_lock(&inode->i_lock); | |
825 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { | |
826 | spin_lock(&dentry->d_lock); | |
827 | if (!dentry->d_lockref.count) { | |
828 | /* | |
829 | * inform the fs via d_prune that this dentry | |
830 | * is about to be unhashed and destroyed. | |
831 | */ | |
832 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && | |
833 | !d_unhashed(dentry)) | |
834 | dentry->d_op->d_prune(dentry); | |
835 | ||
836 | __dget_dlock(dentry); | |
837 | __d_drop(dentry); | |
838 | spin_unlock(&dentry->d_lock); | |
839 | spin_unlock(&inode->i_lock); | |
840 | dput(dentry); | |
841 | goto restart; | |
842 | } | |
843 | spin_unlock(&dentry->d_lock); | |
844 | } | |
845 | spin_unlock(&inode->i_lock); | |
846 | } | |
847 | EXPORT_SYMBOL(d_prune_aliases); | |
848 | ||
849 | /* | |
850 | * Try to throw away a dentry - free the inode, dput the parent. | |
851 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
852 | * Releases dentry->d_lock. | |
853 | * | |
854 | * This may fail if locks cannot be acquired no problem, just try again. | |
855 | */ | |
856 | static struct dentry * try_prune_one_dentry(struct dentry *dentry) | |
857 | __releases(dentry->d_lock) | |
858 | { | |
859 | struct dentry *parent; | |
860 | ||
861 | parent = dentry_kill(dentry, 0); | |
862 | /* | |
863 | * If dentry_kill returns NULL, we have nothing more to do. | |
864 | * if it returns the same dentry, trylocks failed. In either | |
865 | * case, just loop again. | |
866 | * | |
867 | * Otherwise, we need to prune ancestors too. This is necessary | |
868 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
869 | * is also expected to be beneficial in reducing dentry cache | |
870 | * fragmentation. | |
871 | */ | |
872 | if (!parent) | |
873 | return NULL; | |
874 | if (parent == dentry) | |
875 | return dentry; | |
876 | ||
877 | /* Prune ancestors. */ | |
878 | dentry = parent; | |
879 | while (dentry) { | |
880 | if (lockref_put_or_lock(&dentry->d_lockref)) | |
881 | return NULL; | |
882 | dentry = dentry_kill(dentry, 1); | |
883 | } | |
884 | return NULL; | |
885 | } | |
886 | ||
887 | static void shrink_dentry_list(struct list_head *list) | |
888 | { | |
889 | struct dentry *dentry; | |
890 | ||
891 | rcu_read_lock(); | |
892 | for (;;) { | |
893 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); | |
894 | if (&dentry->d_lru == list) | |
895 | break; /* empty */ | |
896 | ||
897 | /* | |
898 | * Get the dentry lock, and re-verify that the dentry is | |
899 | * this on the shrinking list. If it is, we know that | |
900 | * DCACHE_SHRINK_LIST and DCACHE_LRU_LIST are set. | |
901 | */ | |
902 | spin_lock(&dentry->d_lock); | |
903 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
904 | spin_unlock(&dentry->d_lock); | |
905 | continue; | |
906 | } | |
907 | ||
908 | /* | |
909 | * The dispose list is isolated and dentries are not accounted | |
910 | * to the LRU here, so we can simply remove it from the list | |
911 | * here regardless of whether it is referenced or not. | |
912 | */ | |
913 | d_shrink_del(dentry); | |
914 | ||
915 | /* | |
916 | * We found an inuse dentry which was not removed from | |
917 | * the LRU because of laziness during lookup. Do not free it. | |
918 | */ | |
919 | if (dentry->d_lockref.count) { | |
920 | spin_unlock(&dentry->d_lock); | |
921 | continue; | |
922 | } | |
923 | rcu_read_unlock(); | |
924 | ||
925 | /* | |
926 | * If 'try_to_prune()' returns a dentry, it will | |
927 | * be the same one we passed in, and d_lock will | |
928 | * have been held the whole time, so it will not | |
929 | * have been added to any other lists. We failed | |
930 | * to get the inode lock. | |
931 | * | |
932 | * We just add it back to the shrink list. | |
933 | */ | |
934 | dentry = try_prune_one_dentry(dentry); | |
935 | ||
936 | rcu_read_lock(); | |
937 | if (dentry) { | |
938 | d_shrink_add(dentry, list); | |
939 | spin_unlock(&dentry->d_lock); | |
940 | } | |
941 | } | |
942 | rcu_read_unlock(); | |
943 | } | |
944 | ||
945 | static enum lru_status | |
946 | dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg) | |
947 | { | |
948 | struct list_head *freeable = arg; | |
949 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
950 | ||
951 | ||
952 | /* | |
953 | * we are inverting the lru lock/dentry->d_lock here, | |
954 | * so use a trylock. If we fail to get the lock, just skip | |
955 | * it | |
956 | */ | |
957 | if (!spin_trylock(&dentry->d_lock)) | |
958 | return LRU_SKIP; | |
959 | ||
960 | /* | |
961 | * Referenced dentries are still in use. If they have active | |
962 | * counts, just remove them from the LRU. Otherwise give them | |
963 | * another pass through the LRU. | |
964 | */ | |
965 | if (dentry->d_lockref.count) { | |
966 | d_lru_isolate(dentry); | |
967 | spin_unlock(&dentry->d_lock); | |
968 | return LRU_REMOVED; | |
969 | } | |
970 | ||
971 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
972 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
973 | spin_unlock(&dentry->d_lock); | |
974 | ||
975 | /* | |
976 | * The list move itself will be made by the common LRU code. At | |
977 | * this point, we've dropped the dentry->d_lock but keep the | |
978 | * lru lock. This is safe to do, since every list movement is | |
979 | * protected by the lru lock even if both locks are held. | |
980 | * | |
981 | * This is guaranteed by the fact that all LRU management | |
982 | * functions are intermediated by the LRU API calls like | |
983 | * list_lru_add and list_lru_del. List movement in this file | |
984 | * only ever occur through this functions or through callbacks | |
985 | * like this one, that are called from the LRU API. | |
986 | * | |
987 | * The only exceptions to this are functions like | |
988 | * shrink_dentry_list, and code that first checks for the | |
989 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
990 | * operating only with stack provided lists after they are | |
991 | * properly isolated from the main list. It is thus, always a | |
992 | * local access. | |
993 | */ | |
994 | return LRU_ROTATE; | |
995 | } | |
996 | ||
997 | d_lru_shrink_move(dentry, freeable); | |
998 | spin_unlock(&dentry->d_lock); | |
999 | ||
1000 | return LRU_REMOVED; | |
1001 | } | |
1002 | ||
1003 | /** | |
1004 | * prune_dcache_sb - shrink the dcache | |
1005 | * @sb: superblock | |
1006 | * @nr_to_scan : number of entries to try to free | |
1007 | * @nid: which node to scan for freeable entities | |
1008 | * | |
1009 | * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is | |
1010 | * done when we need more memory an called from the superblock shrinker | |
1011 | * function. | |
1012 | * | |
1013 | * This function may fail to free any resources if all the dentries are in | |
1014 | * use. | |
1015 | */ | |
1016 | long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan, | |
1017 | int nid) | |
1018 | { | |
1019 | LIST_HEAD(dispose); | |
1020 | long freed; | |
1021 | ||
1022 | freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate, | |
1023 | &dispose, &nr_to_scan); | |
1024 | shrink_dentry_list(&dispose); | |
1025 | return freed; | |
1026 | } | |
1027 | ||
1028 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, | |
1029 | spinlock_t *lru_lock, void *arg) | |
1030 | { | |
1031 | struct list_head *freeable = arg; | |
1032 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
1033 | ||
1034 | /* | |
1035 | * we are inverting the lru lock/dentry->d_lock here, | |
1036 | * so use a trylock. If we fail to get the lock, just skip | |
1037 | * it | |
1038 | */ | |
1039 | if (!spin_trylock(&dentry->d_lock)) | |
1040 | return LRU_SKIP; | |
1041 | ||
1042 | d_lru_shrink_move(dentry, freeable); | |
1043 | spin_unlock(&dentry->d_lock); | |
1044 | ||
1045 | return LRU_REMOVED; | |
1046 | } | |
1047 | ||
1048 | ||
1049 | /** | |
1050 | * shrink_dcache_sb - shrink dcache for a superblock | |
1051 | * @sb: superblock | |
1052 | * | |
1053 | * Shrink the dcache for the specified super block. This is used to free | |
1054 | * the dcache before unmounting a file system. | |
1055 | */ | |
1056 | void shrink_dcache_sb(struct super_block *sb) | |
1057 | { | |
1058 | long freed; | |
1059 | ||
1060 | do { | |
1061 | LIST_HEAD(dispose); | |
1062 | ||
1063 | freed = list_lru_walk(&sb->s_dentry_lru, | |
1064 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); | |
1065 | ||
1066 | this_cpu_sub(nr_dentry_unused, freed); | |
1067 | shrink_dentry_list(&dispose); | |
1068 | } while (freed > 0); | |
1069 | } | |
1070 | EXPORT_SYMBOL(shrink_dcache_sb); | |
1071 | ||
1072 | /* | |
1073 | * This tries to ascend one level of parenthood, but | |
1074 | * we can race with renaming, so we need to re-check | |
1075 | * the parenthood after dropping the lock and check | |
1076 | * that the sequence number still matches. | |
1077 | */ | |
1078 | static struct dentry *try_to_ascend(struct dentry *old, unsigned seq) | |
1079 | { | |
1080 | struct dentry *new = old->d_parent; | |
1081 | ||
1082 | rcu_read_lock(); | |
1083 | spin_unlock(&old->d_lock); | |
1084 | spin_lock(&new->d_lock); | |
1085 | ||
1086 | /* | |
1087 | * might go back up the wrong parent if we have had a rename | |
1088 | * or deletion | |
1089 | */ | |
1090 | if (new != old->d_parent || | |
1091 | (old->d_flags & DCACHE_DENTRY_KILLED) || | |
1092 | need_seqretry(&rename_lock, seq)) { | |
1093 | spin_unlock(&new->d_lock); | |
1094 | new = NULL; | |
1095 | } | |
1096 | rcu_read_unlock(); | |
1097 | return new; | |
1098 | } | |
1099 | ||
1100 | /** | |
1101 | * enum d_walk_ret - action to talke during tree walk | |
1102 | * @D_WALK_CONTINUE: contrinue walk | |
1103 | * @D_WALK_QUIT: quit walk | |
1104 | * @D_WALK_NORETRY: quit when retry is needed | |
1105 | * @D_WALK_SKIP: skip this dentry and its children | |
1106 | */ | |
1107 | enum d_walk_ret { | |
1108 | D_WALK_CONTINUE, | |
1109 | D_WALK_QUIT, | |
1110 | D_WALK_NORETRY, | |
1111 | D_WALK_SKIP, | |
1112 | }; | |
1113 | ||
1114 | /** | |
1115 | * d_walk - walk the dentry tree | |
1116 | * @parent: start of walk | |
1117 | * @data: data passed to @enter() and @finish() | |
1118 | * @enter: callback when first entering the dentry | |
1119 | * @finish: callback when successfully finished the walk | |
1120 | * | |
1121 | * The @enter() and @finish() callbacks are called with d_lock held. | |
1122 | */ | |
1123 | static void d_walk(struct dentry *parent, void *data, | |
1124 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1125 | void (*finish)(void *)) | |
1126 | { | |
1127 | struct dentry *this_parent; | |
1128 | struct list_head *next; | |
1129 | unsigned seq = 0; | |
1130 | enum d_walk_ret ret; | |
1131 | bool retry = true; | |
1132 | ||
1133 | again: | |
1134 | read_seqbegin_or_lock(&rename_lock, &seq); | |
1135 | this_parent = parent; | |
1136 | spin_lock(&this_parent->d_lock); | |
1137 | ||
1138 | ret = enter(data, this_parent); | |
1139 | switch (ret) { | |
1140 | case D_WALK_CONTINUE: | |
1141 | break; | |
1142 | case D_WALK_QUIT: | |
1143 | case D_WALK_SKIP: | |
1144 | goto out_unlock; | |
1145 | case D_WALK_NORETRY: | |
1146 | retry = false; | |
1147 | break; | |
1148 | } | |
1149 | repeat: | |
1150 | next = this_parent->d_subdirs.next; | |
1151 | resume: | |
1152 | while (next != &this_parent->d_subdirs) { | |
1153 | struct list_head *tmp = next; | |
1154 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); | |
1155 | next = tmp->next; | |
1156 | ||
1157 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1158 | ||
1159 | ret = enter(data, dentry); | |
1160 | switch (ret) { | |
1161 | case D_WALK_CONTINUE: | |
1162 | break; | |
1163 | case D_WALK_QUIT: | |
1164 | spin_unlock(&dentry->d_lock); | |
1165 | goto out_unlock; | |
1166 | case D_WALK_NORETRY: | |
1167 | retry = false; | |
1168 | break; | |
1169 | case D_WALK_SKIP: | |
1170 | spin_unlock(&dentry->d_lock); | |
1171 | continue; | |
1172 | } | |
1173 | ||
1174 | if (!list_empty(&dentry->d_subdirs)) { | |
1175 | spin_unlock(&this_parent->d_lock); | |
1176 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1177 | this_parent = dentry; | |
1178 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); | |
1179 | goto repeat; | |
1180 | } | |
1181 | spin_unlock(&dentry->d_lock); | |
1182 | } | |
1183 | /* | |
1184 | * All done at this level ... ascend and resume the search. | |
1185 | */ | |
1186 | if (this_parent != parent) { | |
1187 | struct dentry *child = this_parent; | |
1188 | this_parent = try_to_ascend(this_parent, seq); | |
1189 | if (!this_parent) | |
1190 | goto rename_retry; | |
1191 | next = child->d_u.d_child.next; | |
1192 | goto resume; | |
1193 | } | |
1194 | if (need_seqretry(&rename_lock, seq)) { | |
1195 | spin_unlock(&this_parent->d_lock); | |
1196 | goto rename_retry; | |
1197 | } | |
1198 | if (finish) | |
1199 | finish(data); | |
1200 | ||
1201 | out_unlock: | |
1202 | spin_unlock(&this_parent->d_lock); | |
1203 | done_seqretry(&rename_lock, seq); | |
1204 | return; | |
1205 | ||
1206 | rename_retry: | |
1207 | if (!retry) | |
1208 | return; | |
1209 | seq = 1; | |
1210 | goto again; | |
1211 | } | |
1212 | ||
1213 | /* | |
1214 | * Search for at least 1 mount point in the dentry's subdirs. | |
1215 | * We descend to the next level whenever the d_subdirs | |
1216 | * list is non-empty and continue searching. | |
1217 | */ | |
1218 | ||
1219 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) | |
1220 | { | |
1221 | int *ret = data; | |
1222 | if (d_mountpoint(dentry)) { | |
1223 | *ret = 1; | |
1224 | return D_WALK_QUIT; | |
1225 | } | |
1226 | return D_WALK_CONTINUE; | |
1227 | } | |
1228 | ||
1229 | /** | |
1230 | * have_submounts - check for mounts over a dentry | |
1231 | * @parent: dentry to check. | |
1232 | * | |
1233 | * Return true if the parent or its subdirectories contain | |
1234 | * a mount point | |
1235 | */ | |
1236 | int have_submounts(struct dentry *parent) | |
1237 | { | |
1238 | int ret = 0; | |
1239 | ||
1240 | d_walk(parent, &ret, check_mount, NULL); | |
1241 | ||
1242 | return ret; | |
1243 | } | |
1244 | EXPORT_SYMBOL(have_submounts); | |
1245 | ||
1246 | /* | |
1247 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1248 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1249 | * subtree can become unreachable). | |
1250 | * | |
1251 | * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For | |
1252 | * this reason take rename_lock and d_lock on dentry and ancestors. | |
1253 | */ | |
1254 | int d_set_mounted(struct dentry *dentry) | |
1255 | { | |
1256 | struct dentry *p; | |
1257 | int ret = -ENOENT; | |
1258 | write_seqlock(&rename_lock); | |
1259 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1260 | /* Need exclusion wrt. check_submounts_and_drop() */ | |
1261 | spin_lock(&p->d_lock); | |
1262 | if (unlikely(d_unhashed(p))) { | |
1263 | spin_unlock(&p->d_lock); | |
1264 | goto out; | |
1265 | } | |
1266 | spin_unlock(&p->d_lock); | |
1267 | } | |
1268 | spin_lock(&dentry->d_lock); | |
1269 | if (!d_unlinked(dentry)) { | |
1270 | dentry->d_flags |= DCACHE_MOUNTED; | |
1271 | ret = 0; | |
1272 | } | |
1273 | spin_unlock(&dentry->d_lock); | |
1274 | out: | |
1275 | write_sequnlock(&rename_lock); | |
1276 | return ret; | |
1277 | } | |
1278 | ||
1279 | /* | |
1280 | * Search the dentry child list of the specified parent, | |
1281 | * and move any unused dentries to the end of the unused | |
1282 | * list for prune_dcache(). We descend to the next level | |
1283 | * whenever the d_subdirs list is non-empty and continue | |
1284 | * searching. | |
1285 | * | |
1286 | * It returns zero iff there are no unused children, | |
1287 | * otherwise it returns the number of children moved to | |
1288 | * the end of the unused list. This may not be the total | |
1289 | * number of unused children, because select_parent can | |
1290 | * drop the lock and return early due to latency | |
1291 | * constraints. | |
1292 | */ | |
1293 | ||
1294 | struct select_data { | |
1295 | struct dentry *start; | |
1296 | struct list_head dispose; | |
1297 | int found; | |
1298 | }; | |
1299 | ||
1300 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) | |
1301 | { | |
1302 | struct select_data *data = _data; | |
1303 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1304 | ||
1305 | if (data->start == dentry) | |
1306 | goto out; | |
1307 | ||
1308 | /* | |
1309 | * move only zero ref count dentries to the dispose list. | |
1310 | * | |
1311 | * Those which are presently on the shrink list, being processed | |
1312 | * by shrink_dentry_list(), shouldn't be moved. Otherwise the | |
1313 | * loop in shrink_dcache_parent() might not make any progress | |
1314 | * and loop forever. | |
1315 | */ | |
1316 | if (dentry->d_lockref.count) { | |
1317 | dentry_lru_del(dentry); | |
1318 | } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { | |
1319 | /* | |
1320 | * We can't use d_lru_shrink_move() because we | |
1321 | * need to get the global LRU lock and do the | |
1322 | * LRU accounting. | |
1323 | */ | |
1324 | d_lru_del(dentry); | |
1325 | d_shrink_add(dentry, &data->dispose); | |
1326 | data->found++; | |
1327 | ret = D_WALK_NORETRY; | |
1328 | } | |
1329 | /* | |
1330 | * We can return to the caller if we have found some (this | |
1331 | * ensures forward progress). We'll be coming back to find | |
1332 | * the rest. | |
1333 | */ | |
1334 | if (data->found && need_resched()) | |
1335 | ret = D_WALK_QUIT; | |
1336 | out: | |
1337 | return ret; | |
1338 | } | |
1339 | ||
1340 | /** | |
1341 | * shrink_dcache_parent - prune dcache | |
1342 | * @parent: parent of entries to prune | |
1343 | * | |
1344 | * Prune the dcache to remove unused children of the parent dentry. | |
1345 | */ | |
1346 | void shrink_dcache_parent(struct dentry *parent) | |
1347 | { | |
1348 | for (;;) { | |
1349 | struct select_data data; | |
1350 | ||
1351 | INIT_LIST_HEAD(&data.dispose); | |
1352 | data.start = parent; | |
1353 | data.found = 0; | |
1354 | ||
1355 | d_walk(parent, &data, select_collect, NULL); | |
1356 | if (!data.found) | |
1357 | break; | |
1358 | ||
1359 | shrink_dentry_list(&data.dispose); | |
1360 | cond_resched(); | |
1361 | } | |
1362 | } | |
1363 | EXPORT_SYMBOL(shrink_dcache_parent); | |
1364 | ||
1365 | static enum d_walk_ret umount_collect(void *_data, struct dentry *dentry) | |
1366 | { | |
1367 | struct select_data *data = _data; | |
1368 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1369 | ||
1370 | if (dentry->d_lockref.count) { | |
1371 | dentry_lru_del(dentry); | |
1372 | if (likely(!list_empty(&dentry->d_subdirs))) | |
1373 | goto out; | |
1374 | if (dentry == data->start && dentry->d_lockref.count == 1) | |
1375 | goto out; | |
1376 | printk(KERN_ERR | |
1377 | "BUG: Dentry %p{i=%lx,n=%s}" | |
1378 | " still in use (%d)" | |
1379 | " [unmount of %s %s]\n", | |
1380 | dentry, | |
1381 | dentry->d_inode ? | |
1382 | dentry->d_inode->i_ino : 0UL, | |
1383 | dentry->d_name.name, | |
1384 | dentry->d_lockref.count, | |
1385 | dentry->d_sb->s_type->name, | |
1386 | dentry->d_sb->s_id); | |
1387 | BUG(); | |
1388 | } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { | |
1389 | /* | |
1390 | * We can't use d_lru_shrink_move() because we | |
1391 | * need to get the global LRU lock and do the | |
1392 | * LRU accounting. | |
1393 | */ | |
1394 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1395 | d_lru_del(dentry); | |
1396 | d_shrink_add(dentry, &data->dispose); | |
1397 | data->found++; | |
1398 | ret = D_WALK_NORETRY; | |
1399 | } | |
1400 | out: | |
1401 | if (data->found && need_resched()) | |
1402 | ret = D_WALK_QUIT; | |
1403 | return ret; | |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * destroy the dentries attached to a superblock on unmounting | |
1408 | */ | |
1409 | void shrink_dcache_for_umount(struct super_block *sb) | |
1410 | { | |
1411 | struct dentry *dentry; | |
1412 | ||
1413 | if (down_read_trylock(&sb->s_umount)) | |
1414 | BUG(); | |
1415 | ||
1416 | dentry = sb->s_root; | |
1417 | sb->s_root = NULL; | |
1418 | for (;;) { | |
1419 | struct select_data data; | |
1420 | ||
1421 | INIT_LIST_HEAD(&data.dispose); | |
1422 | data.start = dentry; | |
1423 | data.found = 0; | |
1424 | ||
1425 | d_walk(dentry, &data, umount_collect, NULL); | |
1426 | if (!data.found) | |
1427 | break; | |
1428 | ||
1429 | shrink_dentry_list(&data.dispose); | |
1430 | cond_resched(); | |
1431 | } | |
1432 | d_drop(dentry); | |
1433 | dput(dentry); | |
1434 | ||
1435 | while (!hlist_bl_empty(&sb->s_anon)) { | |
1436 | struct select_data data; | |
1437 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
1438 | ||
1439 | INIT_LIST_HEAD(&data.dispose); | |
1440 | data.start = NULL; | |
1441 | data.found = 0; | |
1442 | ||
1443 | d_walk(dentry, &data, umount_collect, NULL); | |
1444 | if (data.found) | |
1445 | shrink_dentry_list(&data.dispose); | |
1446 | cond_resched(); | |
1447 | } | |
1448 | } | |
1449 | ||
1450 | static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry) | |
1451 | { | |
1452 | struct select_data *data = _data; | |
1453 | ||
1454 | if (d_mountpoint(dentry)) { | |
1455 | data->found = -EBUSY; | |
1456 | return D_WALK_QUIT; | |
1457 | } | |
1458 | ||
1459 | return select_collect(_data, dentry); | |
1460 | } | |
1461 | ||
1462 | static void check_and_drop(void *_data) | |
1463 | { | |
1464 | struct select_data *data = _data; | |
1465 | ||
1466 | if (d_mountpoint(data->start)) | |
1467 | data->found = -EBUSY; | |
1468 | if (!data->found) | |
1469 | __d_drop(data->start); | |
1470 | } | |
1471 | ||
1472 | /** | |
1473 | * check_submounts_and_drop - prune dcache, check for submounts and drop | |
1474 | * | |
1475 | * All done as a single atomic operation relative to has_unlinked_ancestor(). | |
1476 | * Returns 0 if successfully unhashed @parent. If there were submounts then | |
1477 | * return -EBUSY. | |
1478 | * | |
1479 | * @dentry: dentry to prune and drop | |
1480 | */ | |
1481 | int check_submounts_and_drop(struct dentry *dentry) | |
1482 | { | |
1483 | int ret = 0; | |
1484 | ||
1485 | /* Negative dentries can be dropped without further checks */ | |
1486 | if (!dentry->d_inode) { | |
1487 | d_drop(dentry); | |
1488 | goto out; | |
1489 | } | |
1490 | ||
1491 | for (;;) { | |
1492 | struct select_data data; | |
1493 | ||
1494 | INIT_LIST_HEAD(&data.dispose); | |
1495 | data.start = dentry; | |
1496 | data.found = 0; | |
1497 | ||
1498 | d_walk(dentry, &data, check_and_collect, check_and_drop); | |
1499 | ret = data.found; | |
1500 | ||
1501 | if (!list_empty(&data.dispose)) | |
1502 | shrink_dentry_list(&data.dispose); | |
1503 | ||
1504 | if (ret <= 0) | |
1505 | break; | |
1506 | ||
1507 | cond_resched(); | |
1508 | } | |
1509 | ||
1510 | out: | |
1511 | return ret; | |
1512 | } | |
1513 | EXPORT_SYMBOL(check_submounts_and_drop); | |
1514 | ||
1515 | /** | |
1516 | * __d_alloc - allocate a dcache entry | |
1517 | * @sb: filesystem it will belong to | |
1518 | * @name: qstr of the name | |
1519 | * | |
1520 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1521 | * available. On a success the dentry is returned. The name passed in is | |
1522 | * copied and the copy passed in may be reused after this call. | |
1523 | */ | |
1524 | ||
1525 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) | |
1526 | { | |
1527 | struct dentry *dentry; | |
1528 | char *dname; | |
1529 | ||
1530 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); | |
1531 | if (!dentry) | |
1532 | return NULL; | |
1533 | ||
1534 | /* | |
1535 | * We guarantee that the inline name is always NUL-terminated. | |
1536 | * This way the memcpy() done by the name switching in rename | |
1537 | * will still always have a NUL at the end, even if we might | |
1538 | * be overwriting an internal NUL character | |
1539 | */ | |
1540 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1541 | if (name->len > DNAME_INLINE_LEN-1) { | |
1542 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1543 | if (!dname) { | |
1544 | kmem_cache_free(dentry_cache, dentry); | |
1545 | return NULL; | |
1546 | } | |
1547 | } else { | |
1548 | dname = dentry->d_iname; | |
1549 | } | |
1550 | ||
1551 | dentry->d_name.len = name->len; | |
1552 | dentry->d_name.hash = name->hash; | |
1553 | memcpy(dname, name->name, name->len); | |
1554 | dname[name->len] = 0; | |
1555 | ||
1556 | /* Make sure we always see the terminating NUL character */ | |
1557 | smp_wmb(); | |
1558 | dentry->d_name.name = dname; | |
1559 | ||
1560 | dentry->d_lockref.count = 1; | |
1561 | dentry->d_flags = 0; | |
1562 | spin_lock_init(&dentry->d_lock); | |
1563 | seqcount_init(&dentry->d_seq); | |
1564 | dentry->d_inode = NULL; | |
1565 | dentry->d_parent = dentry; | |
1566 | dentry->d_sb = sb; | |
1567 | dentry->d_op = NULL; | |
1568 | dentry->d_fsdata = NULL; | |
1569 | INIT_HLIST_BL_NODE(&dentry->d_hash); | |
1570 | INIT_LIST_HEAD(&dentry->d_lru); | |
1571 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
1572 | INIT_HLIST_NODE(&dentry->d_alias); | |
1573 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
1574 | d_set_d_op(dentry, dentry->d_sb->s_d_op); | |
1575 | ||
1576 | this_cpu_inc(nr_dentry); | |
1577 | ||
1578 | return dentry; | |
1579 | } | |
1580 | ||
1581 | /** | |
1582 | * d_alloc - allocate a dcache entry | |
1583 | * @parent: parent of entry to allocate | |
1584 | * @name: qstr of the name | |
1585 | * | |
1586 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1587 | * available. On a success the dentry is returned. The name passed in is | |
1588 | * copied and the copy passed in may be reused after this call. | |
1589 | */ | |
1590 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1591 | { | |
1592 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1593 | if (!dentry) | |
1594 | return NULL; | |
1595 | ||
1596 | spin_lock(&parent->d_lock); | |
1597 | /* | |
1598 | * don't need child lock because it is not subject | |
1599 | * to concurrency here | |
1600 | */ | |
1601 | __dget_dlock(parent); | |
1602 | dentry->d_parent = parent; | |
1603 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1604 | spin_unlock(&parent->d_lock); | |
1605 | ||
1606 | return dentry; | |
1607 | } | |
1608 | EXPORT_SYMBOL(d_alloc); | |
1609 | ||
1610 | /** | |
1611 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1612 | * @sb: the superblock | |
1613 | * @name: qstr of the name | |
1614 | * | |
1615 | * For a filesystem that just pins its dentries in memory and never | |
1616 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1617 | */ | |
1618 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) | |
1619 | { | |
1620 | return __d_alloc(sb, name); | |
1621 | } | |
1622 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1623 | ||
1624 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | |
1625 | { | |
1626 | struct qstr q; | |
1627 | ||
1628 | q.name = name; | |
1629 | q.len = strlen(name); | |
1630 | q.hash = full_name_hash(q.name, q.len); | |
1631 | return d_alloc(parent, &q); | |
1632 | } | |
1633 | EXPORT_SYMBOL(d_alloc_name); | |
1634 | ||
1635 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) | |
1636 | { | |
1637 | WARN_ON_ONCE(dentry->d_op); | |
1638 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
1639 | DCACHE_OP_COMPARE | | |
1640 | DCACHE_OP_REVALIDATE | | |
1641 | DCACHE_OP_WEAK_REVALIDATE | | |
1642 | DCACHE_OP_DELETE )); | |
1643 | dentry->d_op = op; | |
1644 | if (!op) | |
1645 | return; | |
1646 | if (op->d_hash) | |
1647 | dentry->d_flags |= DCACHE_OP_HASH; | |
1648 | if (op->d_compare) | |
1649 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1650 | if (op->d_revalidate) | |
1651 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
1652 | if (op->d_weak_revalidate) | |
1653 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
1654 | if (op->d_delete) | |
1655 | dentry->d_flags |= DCACHE_OP_DELETE; | |
1656 | if (op->d_prune) | |
1657 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
1658 | ||
1659 | } | |
1660 | EXPORT_SYMBOL(d_set_d_op); | |
1661 | ||
1662 | static unsigned d_flags_for_inode(struct inode *inode) | |
1663 | { | |
1664 | unsigned add_flags = DCACHE_FILE_TYPE; | |
1665 | ||
1666 | if (!inode) | |
1667 | return DCACHE_MISS_TYPE; | |
1668 | ||
1669 | if (S_ISDIR(inode->i_mode)) { | |
1670 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1671 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1672 | if (unlikely(!inode->i_op->lookup)) | |
1673 | add_flags = DCACHE_AUTODIR_TYPE; | |
1674 | else | |
1675 | inode->i_opflags |= IOP_LOOKUP; | |
1676 | } | |
1677 | } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1678 | if (unlikely(inode->i_op->follow_link)) | |
1679 | add_flags = DCACHE_SYMLINK_TYPE; | |
1680 | else | |
1681 | inode->i_opflags |= IOP_NOFOLLOW; | |
1682 | } | |
1683 | ||
1684 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1685 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1686 | return add_flags; | |
1687 | } | |
1688 | ||
1689 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) | |
1690 | { | |
1691 | unsigned add_flags = d_flags_for_inode(inode); | |
1692 | ||
1693 | spin_lock(&dentry->d_lock); | |
1694 | dentry->d_flags &= ~DCACHE_ENTRY_TYPE; | |
1695 | dentry->d_flags |= add_flags; | |
1696 | if (inode) | |
1697 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); | |
1698 | dentry->d_inode = inode; | |
1699 | dentry_rcuwalk_barrier(dentry); | |
1700 | spin_unlock(&dentry->d_lock); | |
1701 | fsnotify_d_instantiate(dentry, inode); | |
1702 | } | |
1703 | ||
1704 | /** | |
1705 | * d_instantiate - fill in inode information for a dentry | |
1706 | * @entry: dentry to complete | |
1707 | * @inode: inode to attach to this dentry | |
1708 | * | |
1709 | * Fill in inode information in the entry. | |
1710 | * | |
1711 | * This turns negative dentries into productive full members | |
1712 | * of society. | |
1713 | * | |
1714 | * NOTE! This assumes that the inode count has been incremented | |
1715 | * (or otherwise set) by the caller to indicate that it is now | |
1716 | * in use by the dcache. | |
1717 | */ | |
1718 | ||
1719 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1720 | { | |
1721 | BUG_ON(!hlist_unhashed(&entry->d_alias)); | |
1722 | if (inode) | |
1723 | spin_lock(&inode->i_lock); | |
1724 | __d_instantiate(entry, inode); | |
1725 | if (inode) | |
1726 | spin_unlock(&inode->i_lock); | |
1727 | security_d_instantiate(entry, inode); | |
1728 | } | |
1729 | EXPORT_SYMBOL(d_instantiate); | |
1730 | ||
1731 | /** | |
1732 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1733 | * @entry: dentry to instantiate | |
1734 | * @inode: inode to attach to this dentry | |
1735 | * | |
1736 | * Fill in inode information in the entry. On success, it returns NULL. | |
1737 | * If an unhashed alias of "entry" already exists, then we return the | |
1738 | * aliased dentry instead and drop one reference to inode. | |
1739 | * | |
1740 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1741 | * had better be holding the parent directory semaphore. | |
1742 | * | |
1743 | * This also assumes that the inode count has been incremented | |
1744 | * (or otherwise set) by the caller to indicate that it is now | |
1745 | * in use by the dcache. | |
1746 | */ | |
1747 | static struct dentry *__d_instantiate_unique(struct dentry *entry, | |
1748 | struct inode *inode) | |
1749 | { | |
1750 | struct dentry *alias; | |
1751 | int len = entry->d_name.len; | |
1752 | const char *name = entry->d_name.name; | |
1753 | unsigned int hash = entry->d_name.hash; | |
1754 | ||
1755 | if (!inode) { | |
1756 | __d_instantiate(entry, NULL); | |
1757 | return NULL; | |
1758 | } | |
1759 | ||
1760 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
1761 | /* | |
1762 | * Don't need alias->d_lock here, because aliases with | |
1763 | * d_parent == entry->d_parent are not subject to name or | |
1764 | * parent changes, because the parent inode i_mutex is held. | |
1765 | */ | |
1766 | if (alias->d_name.hash != hash) | |
1767 | continue; | |
1768 | if (alias->d_parent != entry->d_parent) | |
1769 | continue; | |
1770 | if (alias->d_name.len != len) | |
1771 | continue; | |
1772 | if (dentry_cmp(alias, name, len)) | |
1773 | continue; | |
1774 | __dget(alias); | |
1775 | return alias; | |
1776 | } | |
1777 | ||
1778 | __d_instantiate(entry, inode); | |
1779 | return NULL; | |
1780 | } | |
1781 | ||
1782 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1783 | { | |
1784 | struct dentry *result; | |
1785 | ||
1786 | BUG_ON(!hlist_unhashed(&entry->d_alias)); | |
1787 | ||
1788 | if (inode) | |
1789 | spin_lock(&inode->i_lock); | |
1790 | result = __d_instantiate_unique(entry, inode); | |
1791 | if (inode) | |
1792 | spin_unlock(&inode->i_lock); | |
1793 | ||
1794 | if (!result) { | |
1795 | security_d_instantiate(entry, inode); | |
1796 | return NULL; | |
1797 | } | |
1798 | ||
1799 | BUG_ON(!d_unhashed(result)); | |
1800 | iput(inode); | |
1801 | return result; | |
1802 | } | |
1803 | ||
1804 | EXPORT_SYMBOL(d_instantiate_unique); | |
1805 | ||
1806 | /** | |
1807 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1808 | * @entry: dentry to complete | |
1809 | * @inode: inode to attach to this dentry | |
1810 | * | |
1811 | * Fill in inode information in the entry. If a directory alias is found, then | |
1812 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1813 | * guarantees that a directory inode may never have more than one alias. | |
1814 | */ | |
1815 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1816 | { | |
1817 | BUG_ON(!hlist_unhashed(&entry->d_alias)); | |
1818 | ||
1819 | spin_lock(&inode->i_lock); | |
1820 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1821 | spin_unlock(&inode->i_lock); | |
1822 | iput(inode); | |
1823 | return -EBUSY; | |
1824 | } | |
1825 | __d_instantiate(entry, inode); | |
1826 | spin_unlock(&inode->i_lock); | |
1827 | security_d_instantiate(entry, inode); | |
1828 | ||
1829 | return 0; | |
1830 | } | |
1831 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1832 | ||
1833 | struct dentry *d_make_root(struct inode *root_inode) | |
1834 | { | |
1835 | struct dentry *res = NULL; | |
1836 | ||
1837 | if (root_inode) { | |
1838 | static const struct qstr name = QSTR_INIT("/", 1); | |
1839 | ||
1840 | res = __d_alloc(root_inode->i_sb, &name); | |
1841 | if (res) | |
1842 | d_instantiate(res, root_inode); | |
1843 | else | |
1844 | iput(root_inode); | |
1845 | } | |
1846 | return res; | |
1847 | } | |
1848 | EXPORT_SYMBOL(d_make_root); | |
1849 | ||
1850 | static struct dentry * __d_find_any_alias(struct inode *inode) | |
1851 | { | |
1852 | struct dentry *alias; | |
1853 | ||
1854 | if (hlist_empty(&inode->i_dentry)) | |
1855 | return NULL; | |
1856 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); | |
1857 | __dget(alias); | |
1858 | return alias; | |
1859 | } | |
1860 | ||
1861 | /** | |
1862 | * d_find_any_alias - find any alias for a given inode | |
1863 | * @inode: inode to find an alias for | |
1864 | * | |
1865 | * If any aliases exist for the given inode, take and return a | |
1866 | * reference for one of them. If no aliases exist, return %NULL. | |
1867 | */ | |
1868 | struct dentry *d_find_any_alias(struct inode *inode) | |
1869 | { | |
1870 | struct dentry *de; | |
1871 | ||
1872 | spin_lock(&inode->i_lock); | |
1873 | de = __d_find_any_alias(inode); | |
1874 | spin_unlock(&inode->i_lock); | |
1875 | return de; | |
1876 | } | |
1877 | EXPORT_SYMBOL(d_find_any_alias); | |
1878 | ||
1879 | /** | |
1880 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1881 | * @inode: inode to allocate the dentry for | |
1882 | * | |
1883 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1884 | * similar open by handle operations. The returned dentry may be anonymous, | |
1885 | * or may have a full name (if the inode was already in the cache). | |
1886 | * | |
1887 | * When called on a directory inode, we must ensure that the inode only ever | |
1888 | * has one dentry. If a dentry is found, that is returned instead of | |
1889 | * allocating a new one. | |
1890 | * | |
1891 | * On successful return, the reference to the inode has been transferred | |
1892 | * to the dentry. In case of an error the reference on the inode is released. | |
1893 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1894 | * be passed in and will be the error will be propagate to the return value, | |
1895 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
1896 | */ | |
1897 | struct dentry *d_obtain_alias(struct inode *inode) | |
1898 | { | |
1899 | static const struct qstr anonstring = QSTR_INIT("/", 1); | |
1900 | struct dentry *tmp; | |
1901 | struct dentry *res; | |
1902 | unsigned add_flags; | |
1903 | ||
1904 | if (!inode) | |
1905 | return ERR_PTR(-ESTALE); | |
1906 | if (IS_ERR(inode)) | |
1907 | return ERR_CAST(inode); | |
1908 | ||
1909 | res = d_find_any_alias(inode); | |
1910 | if (res) | |
1911 | goto out_iput; | |
1912 | ||
1913 | tmp = __d_alloc(inode->i_sb, &anonstring); | |
1914 | if (!tmp) { | |
1915 | res = ERR_PTR(-ENOMEM); | |
1916 | goto out_iput; | |
1917 | } | |
1918 | ||
1919 | spin_lock(&inode->i_lock); | |
1920 | res = __d_find_any_alias(inode); | |
1921 | if (res) { | |
1922 | spin_unlock(&inode->i_lock); | |
1923 | dput(tmp); | |
1924 | goto out_iput; | |
1925 | } | |
1926 | ||
1927 | /* attach a disconnected dentry */ | |
1928 | add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED; | |
1929 | ||
1930 | spin_lock(&tmp->d_lock); | |
1931 | tmp->d_inode = inode; | |
1932 | tmp->d_flags |= add_flags; | |
1933 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); | |
1934 | hlist_bl_lock(&tmp->d_sb->s_anon); | |
1935 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); | |
1936 | hlist_bl_unlock(&tmp->d_sb->s_anon); | |
1937 | spin_unlock(&tmp->d_lock); | |
1938 | spin_unlock(&inode->i_lock); | |
1939 | security_d_instantiate(tmp, inode); | |
1940 | ||
1941 | return tmp; | |
1942 | ||
1943 | out_iput: | |
1944 | if (res && !IS_ERR(res)) | |
1945 | security_d_instantiate(res, inode); | |
1946 | iput(inode); | |
1947 | return res; | |
1948 | } | |
1949 | EXPORT_SYMBOL(d_obtain_alias); | |
1950 | ||
1951 | /** | |
1952 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1953 | * @inode: the inode which may have a disconnected dentry | |
1954 | * @dentry: a negative dentry which we want to point to the inode. | |
1955 | * | |
1956 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1957 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1958 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1959 | * | |
1960 | * This is needed in the lookup routine of any filesystem that is exportable | |
1961 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1962 | * | |
1963 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1964 | * is returned. This matches the expected return value of ->lookup. | |
1965 | * | |
1966 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
1967 | * In that case, we know that the inode will be a regular file, and also this | |
1968 | * will only occur during atomic_open. So we need to check for the dentry | |
1969 | * being already hashed only in the final case. | |
1970 | */ | |
1971 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1972 | { | |
1973 | struct dentry *new = NULL; | |
1974 | ||
1975 | if (IS_ERR(inode)) | |
1976 | return ERR_CAST(inode); | |
1977 | ||
1978 | if (inode && S_ISDIR(inode->i_mode)) { | |
1979 | spin_lock(&inode->i_lock); | |
1980 | new = __d_find_alias(inode, 1); | |
1981 | if (new) { | |
1982 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
1983 | spin_unlock(&inode->i_lock); | |
1984 | security_d_instantiate(new, inode); | |
1985 | d_move(new, dentry); | |
1986 | iput(inode); | |
1987 | } else { | |
1988 | /* already taking inode->i_lock, so d_add() by hand */ | |
1989 | __d_instantiate(dentry, inode); | |
1990 | spin_unlock(&inode->i_lock); | |
1991 | security_d_instantiate(dentry, inode); | |
1992 | d_rehash(dentry); | |
1993 | } | |
1994 | } else { | |
1995 | d_instantiate(dentry, inode); | |
1996 | if (d_unhashed(dentry)) | |
1997 | d_rehash(dentry); | |
1998 | } | |
1999 | return new; | |
2000 | } | |
2001 | EXPORT_SYMBOL(d_splice_alias); | |
2002 | ||
2003 | /** | |
2004 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
2005 | * @inode: the inode case-insensitive lookup has found | |
2006 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
2007 | * @name: the case-exact name to be associated with the returned dentry | |
2008 | * | |
2009 | * This is to avoid filling the dcache with case-insensitive names to the | |
2010 | * same inode, only the actual correct case is stored in the dcache for | |
2011 | * case-insensitive filesystems. | |
2012 | * | |
2013 | * For a case-insensitive lookup match and if the the case-exact dentry | |
2014 | * already exists in in the dcache, use it and return it. | |
2015 | * | |
2016 | * If no entry exists with the exact case name, allocate new dentry with | |
2017 | * the exact case, and return the spliced entry. | |
2018 | */ | |
2019 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, | |
2020 | struct qstr *name) | |
2021 | { | |
2022 | struct dentry *found; | |
2023 | struct dentry *new; | |
2024 | ||
2025 | /* | |
2026 | * First check if a dentry matching the name already exists, | |
2027 | * if not go ahead and create it now. | |
2028 | */ | |
2029 | found = d_hash_and_lookup(dentry->d_parent, name); | |
2030 | if (unlikely(IS_ERR(found))) | |
2031 | goto err_out; | |
2032 | if (!found) { | |
2033 | new = d_alloc(dentry->d_parent, name); | |
2034 | if (!new) { | |
2035 | found = ERR_PTR(-ENOMEM); | |
2036 | goto err_out; | |
2037 | } | |
2038 | ||
2039 | found = d_splice_alias(inode, new); | |
2040 | if (found) { | |
2041 | dput(new); | |
2042 | return found; | |
2043 | } | |
2044 | return new; | |
2045 | } | |
2046 | ||
2047 | /* | |
2048 | * If a matching dentry exists, and it's not negative use it. | |
2049 | * | |
2050 | * Decrement the reference count to balance the iget() done | |
2051 | * earlier on. | |
2052 | */ | |
2053 | if (found->d_inode) { | |
2054 | if (unlikely(found->d_inode != inode)) { | |
2055 | /* This can't happen because bad inodes are unhashed. */ | |
2056 | BUG_ON(!is_bad_inode(inode)); | |
2057 | BUG_ON(!is_bad_inode(found->d_inode)); | |
2058 | } | |
2059 | iput(inode); | |
2060 | return found; | |
2061 | } | |
2062 | ||
2063 | /* | |
2064 | * Negative dentry: instantiate it unless the inode is a directory and | |
2065 | * already has a dentry. | |
2066 | */ | |
2067 | new = d_splice_alias(inode, found); | |
2068 | if (new) { | |
2069 | dput(found); | |
2070 | found = new; | |
2071 | } | |
2072 | return found; | |
2073 | ||
2074 | err_out: | |
2075 | iput(inode); | |
2076 | return found; | |
2077 | } | |
2078 | EXPORT_SYMBOL(d_add_ci); | |
2079 | ||
2080 | /* | |
2081 | * Do the slow-case of the dentry name compare. | |
2082 | * | |
2083 | * Unlike the dentry_cmp() function, we need to atomically | |
2084 | * load the name and length information, so that the | |
2085 | * filesystem can rely on them, and can use the 'name' and | |
2086 | * 'len' information without worrying about walking off the | |
2087 | * end of memory etc. | |
2088 | * | |
2089 | * Thus the read_seqcount_retry() and the "duplicate" info | |
2090 | * in arguments (the low-level filesystem should not look | |
2091 | * at the dentry inode or name contents directly, since | |
2092 | * rename can change them while we're in RCU mode). | |
2093 | */ | |
2094 | enum slow_d_compare { | |
2095 | D_COMP_OK, | |
2096 | D_COMP_NOMATCH, | |
2097 | D_COMP_SEQRETRY, | |
2098 | }; | |
2099 | ||
2100 | static noinline enum slow_d_compare slow_dentry_cmp( | |
2101 | const struct dentry *parent, | |
2102 | struct dentry *dentry, | |
2103 | unsigned int seq, | |
2104 | const struct qstr *name) | |
2105 | { | |
2106 | int tlen = dentry->d_name.len; | |
2107 | const char *tname = dentry->d_name.name; | |
2108 | ||
2109 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2110 | cpu_relax(); | |
2111 | return D_COMP_SEQRETRY; | |
2112 | } | |
2113 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) | |
2114 | return D_COMP_NOMATCH; | |
2115 | return D_COMP_OK; | |
2116 | } | |
2117 | ||
2118 | /** | |
2119 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2120 | * @parent: parent dentry | |
2121 | * @name: qstr of name we wish to find | |
2122 | * @seqp: returns d_seq value at the point where the dentry was found | |
2123 | * Returns: dentry, or NULL | |
2124 | * | |
2125 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2126 | * resolution (store-free path walking) design described in | |
2127 | * Documentation/filesystems/path-lookup.txt. | |
2128 | * | |
2129 | * This is not to be used outside core vfs. | |
2130 | * | |
2131 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2132 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2133 | * without taking d_lock and checking d_seq sequence count against @seq | |
2134 | * returned here. | |
2135 | * | |
2136 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount | |
2137 | * function. | |
2138 | * | |
2139 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2140 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2141 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2142 | * is formed, giving integrity down the path walk. | |
2143 | * | |
2144 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2145 | * number we've returned before using any of the resulting dentry state! | |
2146 | */ | |
2147 | struct dentry *__d_lookup_rcu(const struct dentry *parent, | |
2148 | const struct qstr *name, | |
2149 | unsigned *seqp) | |
2150 | { | |
2151 | u64 hashlen = name->hash_len; | |
2152 | const unsigned char *str = name->name; | |
2153 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); | |
2154 | struct hlist_bl_node *node; | |
2155 | struct dentry *dentry; | |
2156 | ||
2157 | /* | |
2158 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2159 | * required to prevent single threaded performance regressions | |
2160 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2161 | * Keep the two functions in sync. | |
2162 | */ | |
2163 | ||
2164 | /* | |
2165 | * The hash list is protected using RCU. | |
2166 | * | |
2167 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2168 | * races with d_move(). | |
2169 | * | |
2170 | * It is possible that concurrent renames can mess up our list | |
2171 | * walk here and result in missing our dentry, resulting in the | |
2172 | * false-negative result. d_lookup() protects against concurrent | |
2173 | * renames using rename_lock seqlock. | |
2174 | * | |
2175 | * See Documentation/filesystems/path-lookup.txt for more details. | |
2176 | */ | |
2177 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { | |
2178 | unsigned seq; | |
2179 | ||
2180 | seqretry: | |
2181 | /* | |
2182 | * The dentry sequence count protects us from concurrent | |
2183 | * renames, and thus protects parent and name fields. | |
2184 | * | |
2185 | * The caller must perform a seqcount check in order | |
2186 | * to do anything useful with the returned dentry. | |
2187 | * | |
2188 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2189 | * we don't wait for the sequence count to stabilize if it | |
2190 | * is in the middle of a sequence change. If we do the slow | |
2191 | * dentry compare, we will do seqretries until it is stable, | |
2192 | * and if we end up with a successful lookup, we actually | |
2193 | * want to exit RCU lookup anyway. | |
2194 | */ | |
2195 | seq = raw_seqcount_begin(&dentry->d_seq); | |
2196 | if (dentry->d_parent != parent) | |
2197 | continue; | |
2198 | if (d_unhashed(dentry)) | |
2199 | continue; | |
2200 | ||
2201 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { | |
2202 | if (dentry->d_name.hash != hashlen_hash(hashlen)) | |
2203 | continue; | |
2204 | *seqp = seq; | |
2205 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
2206 | case D_COMP_OK: | |
2207 | return dentry; | |
2208 | case D_COMP_NOMATCH: | |
2209 | continue; | |
2210 | default: | |
2211 | goto seqretry; | |
2212 | } | |
2213 | } | |
2214 | ||
2215 | if (dentry->d_name.hash_len != hashlen) | |
2216 | continue; | |
2217 | *seqp = seq; | |
2218 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) | |
2219 | return dentry; | |
2220 | } | |
2221 | return NULL; | |
2222 | } | |
2223 | ||
2224 | /** | |
2225 | * d_lookup - search for a dentry | |
2226 | * @parent: parent dentry | |
2227 | * @name: qstr of name we wish to find | |
2228 | * Returns: dentry, or NULL | |
2229 | * | |
2230 | * d_lookup searches the children of the parent dentry for the name in | |
2231 | * question. If the dentry is found its reference count is incremented and the | |
2232 | * dentry is returned. The caller must use dput to free the entry when it has | |
2233 | * finished using it. %NULL is returned if the dentry does not exist. | |
2234 | */ | |
2235 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) | |
2236 | { | |
2237 | struct dentry *dentry; | |
2238 | unsigned seq; | |
2239 | ||
2240 | do { | |
2241 | seq = read_seqbegin(&rename_lock); | |
2242 | dentry = __d_lookup(parent, name); | |
2243 | if (dentry) | |
2244 | break; | |
2245 | } while (read_seqretry(&rename_lock, seq)); | |
2246 | return dentry; | |
2247 | } | |
2248 | EXPORT_SYMBOL(d_lookup); | |
2249 | ||
2250 | /** | |
2251 | * __d_lookup - search for a dentry (racy) | |
2252 | * @parent: parent dentry | |
2253 | * @name: qstr of name we wish to find | |
2254 | * Returns: dentry, or NULL | |
2255 | * | |
2256 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2257 | * false-negative result due to unrelated rename activity. | |
2258 | * | |
2259 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2260 | * however it must be used carefully, eg. with a following d_lookup in | |
2261 | * the case of failure. | |
2262 | * | |
2263 | * __d_lookup callers must be commented. | |
2264 | */ | |
2265 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) | |
2266 | { | |
2267 | unsigned int len = name->len; | |
2268 | unsigned int hash = name->hash; | |
2269 | const unsigned char *str = name->name; | |
2270 | struct hlist_bl_head *b = d_hash(parent, hash); | |
2271 | struct hlist_bl_node *node; | |
2272 | struct dentry *found = NULL; | |
2273 | struct dentry *dentry; | |
2274 | ||
2275 | /* | |
2276 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2277 | * required to prevent single threaded performance regressions | |
2278 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2279 | * Keep the two functions in sync. | |
2280 | */ | |
2281 | ||
2282 | /* | |
2283 | * The hash list is protected using RCU. | |
2284 | * | |
2285 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2286 | * with d_move(). | |
2287 | * | |
2288 | * It is possible that concurrent renames can mess up our list | |
2289 | * walk here and result in missing our dentry, resulting in the | |
2290 | * false-negative result. d_lookup() protects against concurrent | |
2291 | * renames using rename_lock seqlock. | |
2292 | * | |
2293 | * See Documentation/filesystems/path-lookup.txt for more details. | |
2294 | */ | |
2295 | rcu_read_lock(); | |
2296 | ||
2297 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { | |
2298 | ||
2299 | if (dentry->d_name.hash != hash) | |
2300 | continue; | |
2301 | ||
2302 | spin_lock(&dentry->d_lock); | |
2303 | if (dentry->d_parent != parent) | |
2304 | goto next; | |
2305 | if (d_unhashed(dentry)) | |
2306 | goto next; | |
2307 | ||
2308 | /* | |
2309 | * It is safe to compare names since d_move() cannot | |
2310 | * change the qstr (protected by d_lock). | |
2311 | */ | |
2312 | if (parent->d_flags & DCACHE_OP_COMPARE) { | |
2313 | int tlen = dentry->d_name.len; | |
2314 | const char *tname = dentry->d_name.name; | |
2315 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) | |
2316 | goto next; | |
2317 | } else { | |
2318 | if (dentry->d_name.len != len) | |
2319 | goto next; | |
2320 | if (dentry_cmp(dentry, str, len)) | |
2321 | goto next; | |
2322 | } | |
2323 | ||
2324 | dentry->d_lockref.count++; | |
2325 | found = dentry; | |
2326 | spin_unlock(&dentry->d_lock); | |
2327 | break; | |
2328 | next: | |
2329 | spin_unlock(&dentry->d_lock); | |
2330 | } | |
2331 | rcu_read_unlock(); | |
2332 | ||
2333 | return found; | |
2334 | } | |
2335 | ||
2336 | /** | |
2337 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2338 | * @dir: Directory to search in | |
2339 | * @name: qstr of name we wish to find | |
2340 | * | |
2341 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) | |
2342 | */ | |
2343 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2344 | { | |
2345 | /* | |
2346 | * Check for a fs-specific hash function. Note that we must | |
2347 | * calculate the standard hash first, as the d_op->d_hash() | |
2348 | * routine may choose to leave the hash value unchanged. | |
2349 | */ | |
2350 | name->hash = full_name_hash(name->name, name->len); | |
2351 | if (dir->d_flags & DCACHE_OP_HASH) { | |
2352 | int err = dir->d_op->d_hash(dir, name); | |
2353 | if (unlikely(err < 0)) | |
2354 | return ERR_PTR(err); | |
2355 | } | |
2356 | return d_lookup(dir, name); | |
2357 | } | |
2358 | EXPORT_SYMBOL(d_hash_and_lookup); | |
2359 | ||
2360 | /** | |
2361 | * d_validate - verify dentry provided from insecure source (deprecated) | |
2362 | * @dentry: The dentry alleged to be valid child of @dparent | |
2363 | * @dparent: The parent dentry (known to be valid) | |
2364 | * | |
2365 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2366 | * This is used by ncpfs in its readdir implementation. | |
2367 | * Zero is returned in the dentry is invalid. | |
2368 | * | |
2369 | * This function is slow for big directories, and deprecated, do not use it. | |
2370 | */ | |
2371 | int d_validate(struct dentry *dentry, struct dentry *dparent) | |
2372 | { | |
2373 | struct dentry *child; | |
2374 | ||
2375 | spin_lock(&dparent->d_lock); | |
2376 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { | |
2377 | if (dentry == child) { | |
2378 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
2379 | __dget_dlock(dentry); | |
2380 | spin_unlock(&dentry->d_lock); | |
2381 | spin_unlock(&dparent->d_lock); | |
2382 | return 1; | |
2383 | } | |
2384 | } | |
2385 | spin_unlock(&dparent->d_lock); | |
2386 | ||
2387 | return 0; | |
2388 | } | |
2389 | EXPORT_SYMBOL(d_validate); | |
2390 | ||
2391 | /* | |
2392 | * When a file is deleted, we have two options: | |
2393 | * - turn this dentry into a negative dentry | |
2394 | * - unhash this dentry and free it. | |
2395 | * | |
2396 | * Usually, we want to just turn this into | |
2397 | * a negative dentry, but if anybody else is | |
2398 | * currently using the dentry or the inode | |
2399 | * we can't do that and we fall back on removing | |
2400 | * it from the hash queues and waiting for | |
2401 | * it to be deleted later when it has no users | |
2402 | */ | |
2403 | ||
2404 | /** | |
2405 | * d_delete - delete a dentry | |
2406 | * @dentry: The dentry to delete | |
2407 | * | |
2408 | * Turn the dentry into a negative dentry if possible, otherwise | |
2409 | * remove it from the hash queues so it can be deleted later | |
2410 | */ | |
2411 | ||
2412 | void d_delete(struct dentry * dentry) | |
2413 | { | |
2414 | struct inode *inode; | |
2415 | int isdir = 0; | |
2416 | /* | |
2417 | * Are we the only user? | |
2418 | */ | |
2419 | again: | |
2420 | spin_lock(&dentry->d_lock); | |
2421 | inode = dentry->d_inode; | |
2422 | isdir = S_ISDIR(inode->i_mode); | |
2423 | if (dentry->d_lockref.count == 1) { | |
2424 | if (!spin_trylock(&inode->i_lock)) { | |
2425 | spin_unlock(&dentry->d_lock); | |
2426 | cpu_relax(); | |
2427 | goto again; | |
2428 | } | |
2429 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; | |
2430 | dentry_unlink_inode(dentry); | |
2431 | fsnotify_nameremove(dentry, isdir); | |
2432 | return; | |
2433 | } | |
2434 | ||
2435 | if (!d_unhashed(dentry)) | |
2436 | __d_drop(dentry); | |
2437 | ||
2438 | spin_unlock(&dentry->d_lock); | |
2439 | ||
2440 | fsnotify_nameremove(dentry, isdir); | |
2441 | } | |
2442 | EXPORT_SYMBOL(d_delete); | |
2443 | ||
2444 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) | |
2445 | { | |
2446 | BUG_ON(!d_unhashed(entry)); | |
2447 | hlist_bl_lock(b); | |
2448 | entry->d_flags |= DCACHE_RCUACCESS; | |
2449 | hlist_bl_add_head_rcu(&entry->d_hash, b); | |
2450 | hlist_bl_unlock(b); | |
2451 | } | |
2452 | ||
2453 | static void _d_rehash(struct dentry * entry) | |
2454 | { | |
2455 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2456 | } | |
2457 | ||
2458 | /** | |
2459 | * d_rehash - add an entry back to the hash | |
2460 | * @entry: dentry to add to the hash | |
2461 | * | |
2462 | * Adds a dentry to the hash according to its name. | |
2463 | */ | |
2464 | ||
2465 | void d_rehash(struct dentry * entry) | |
2466 | { | |
2467 | spin_lock(&entry->d_lock); | |
2468 | _d_rehash(entry); | |
2469 | spin_unlock(&entry->d_lock); | |
2470 | } | |
2471 | EXPORT_SYMBOL(d_rehash); | |
2472 | ||
2473 | /** | |
2474 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2475 | * @dentry: dentry to be updated | |
2476 | * @name: new name | |
2477 | * | |
2478 | * Update a case insensitive dentry with new case of name. | |
2479 | * | |
2480 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2481 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2482 | * lengths). | |
2483 | * | |
2484 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2485 | * keep renames and concurrent inserts, and readdir(2) away). | |
2486 | */ | |
2487 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2488 | { | |
2489 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); | |
2490 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ | |
2491 | ||
2492 | spin_lock(&dentry->d_lock); | |
2493 | write_seqcount_begin(&dentry->d_seq); | |
2494 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); | |
2495 | write_seqcount_end(&dentry->d_seq); | |
2496 | spin_unlock(&dentry->d_lock); | |
2497 | } | |
2498 | EXPORT_SYMBOL(dentry_update_name_case); | |
2499 | ||
2500 | static void switch_names(struct dentry *dentry, struct dentry *target) | |
2501 | { | |
2502 | if (dname_external(target)) { | |
2503 | if (dname_external(dentry)) { | |
2504 | /* | |
2505 | * Both external: swap the pointers | |
2506 | */ | |
2507 | swap(target->d_name.name, dentry->d_name.name); | |
2508 | } else { | |
2509 | /* | |
2510 | * dentry:internal, target:external. Steal target's | |
2511 | * storage and make target internal. | |
2512 | */ | |
2513 | memcpy(target->d_iname, dentry->d_name.name, | |
2514 | dentry->d_name.len + 1); | |
2515 | dentry->d_name.name = target->d_name.name; | |
2516 | target->d_name.name = target->d_iname; | |
2517 | } | |
2518 | } else { | |
2519 | if (dname_external(dentry)) { | |
2520 | /* | |
2521 | * dentry:external, target:internal. Give dentry's | |
2522 | * storage to target and make dentry internal | |
2523 | */ | |
2524 | memcpy(dentry->d_iname, target->d_name.name, | |
2525 | target->d_name.len + 1); | |
2526 | target->d_name.name = dentry->d_name.name; | |
2527 | dentry->d_name.name = dentry->d_iname; | |
2528 | } else { | |
2529 | /* | |
2530 | * Both are internal. Just copy target to dentry | |
2531 | */ | |
2532 | memcpy(dentry->d_iname, target->d_name.name, | |
2533 | target->d_name.len + 1); | |
2534 | dentry->d_name.len = target->d_name.len; | |
2535 | return; | |
2536 | } | |
2537 | } | |
2538 | swap(dentry->d_name.len, target->d_name.len); | |
2539 | } | |
2540 | ||
2541 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) | |
2542 | { | |
2543 | /* | |
2544 | * XXXX: do we really need to take target->d_lock? | |
2545 | */ | |
2546 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2547 | spin_lock(&target->d_parent->d_lock); | |
2548 | else { | |
2549 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2550 | spin_lock(&dentry->d_parent->d_lock); | |
2551 | spin_lock_nested(&target->d_parent->d_lock, | |
2552 | DENTRY_D_LOCK_NESTED); | |
2553 | } else { | |
2554 | spin_lock(&target->d_parent->d_lock); | |
2555 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2556 | DENTRY_D_LOCK_NESTED); | |
2557 | } | |
2558 | } | |
2559 | if (target < dentry) { | |
2560 | spin_lock_nested(&target->d_lock, 2); | |
2561 | spin_lock_nested(&dentry->d_lock, 3); | |
2562 | } else { | |
2563 | spin_lock_nested(&dentry->d_lock, 2); | |
2564 | spin_lock_nested(&target->d_lock, 3); | |
2565 | } | |
2566 | } | |
2567 | ||
2568 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2569 | struct dentry *target) | |
2570 | { | |
2571 | if (target->d_parent != dentry->d_parent) | |
2572 | spin_unlock(&dentry->d_parent->d_lock); | |
2573 | if (target->d_parent != target) | |
2574 | spin_unlock(&target->d_parent->d_lock); | |
2575 | } | |
2576 | ||
2577 | /* | |
2578 | * When switching names, the actual string doesn't strictly have to | |
2579 | * be preserved in the target - because we're dropping the target | |
2580 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2581 | * the new name before we switch. | |
2582 | * | |
2583 | * Note that we have to be a lot more careful about getting the hash | |
2584 | * switched - we have to switch the hash value properly even if it | |
2585 | * then no longer matches the actual (corrupted) string of the target. | |
2586 | * The hash value has to match the hash queue that the dentry is on.. | |
2587 | */ | |
2588 | /* | |
2589 | * __d_move - move a dentry | |
2590 | * @dentry: entry to move | |
2591 | * @target: new dentry | |
2592 | * | |
2593 | * Update the dcache to reflect the move of a file name. Negative | |
2594 | * dcache entries should not be moved in this way. Caller must hold | |
2595 | * rename_lock, the i_mutex of the source and target directories, | |
2596 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
2597 | */ | |
2598 | static void __d_move(struct dentry * dentry, struct dentry * target) | |
2599 | { | |
2600 | if (!dentry->d_inode) | |
2601 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2602 | ||
2603 | BUG_ON(d_ancestor(dentry, target)); | |
2604 | BUG_ON(d_ancestor(target, dentry)); | |
2605 | ||
2606 | dentry_lock_for_move(dentry, target); | |
2607 | ||
2608 | write_seqcount_begin(&dentry->d_seq); | |
2609 | write_seqcount_begin(&target->d_seq); | |
2610 | ||
2611 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ | |
2612 | ||
2613 | /* | |
2614 | * Move the dentry to the target hash queue. Don't bother checking | |
2615 | * for the same hash queue because of how unlikely it is. | |
2616 | */ | |
2617 | __d_drop(dentry); | |
2618 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); | |
2619 | ||
2620 | /* Unhash the target: dput() will then get rid of it */ | |
2621 | __d_drop(target); | |
2622 | ||
2623 | list_del(&dentry->d_u.d_child); | |
2624 | list_del(&target->d_u.d_child); | |
2625 | ||
2626 | /* Switch the names.. */ | |
2627 | switch_names(dentry, target); | |
2628 | swap(dentry->d_name.hash, target->d_name.hash); | |
2629 | ||
2630 | /* ... and switch the parents */ | |
2631 | if (IS_ROOT(dentry)) { | |
2632 | dentry->d_parent = target->d_parent; | |
2633 | target->d_parent = target; | |
2634 | INIT_LIST_HEAD(&target->d_u.d_child); | |
2635 | } else { | |
2636 | swap(dentry->d_parent, target->d_parent); | |
2637 | ||
2638 | /* And add them back to the (new) parent lists */ | |
2639 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); | |
2640 | } | |
2641 | ||
2642 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
2643 | ||
2644 | write_seqcount_end(&target->d_seq); | |
2645 | write_seqcount_end(&dentry->d_seq); | |
2646 | ||
2647 | dentry_unlock_parents_for_move(dentry, target); | |
2648 | spin_unlock(&target->d_lock); | |
2649 | fsnotify_d_move(dentry); | |
2650 | spin_unlock(&dentry->d_lock); | |
2651 | } | |
2652 | ||
2653 | /* | |
2654 | * d_move - move a dentry | |
2655 | * @dentry: entry to move | |
2656 | * @target: new dentry | |
2657 | * | |
2658 | * Update the dcache to reflect the move of a file name. Negative | |
2659 | * dcache entries should not be moved in this way. See the locking | |
2660 | * requirements for __d_move. | |
2661 | */ | |
2662 | void d_move(struct dentry *dentry, struct dentry *target) | |
2663 | { | |
2664 | write_seqlock(&rename_lock); | |
2665 | __d_move(dentry, target); | |
2666 | write_sequnlock(&rename_lock); | |
2667 | } | |
2668 | EXPORT_SYMBOL(d_move); | |
2669 | ||
2670 | /** | |
2671 | * d_ancestor - search for an ancestor | |
2672 | * @p1: ancestor dentry | |
2673 | * @p2: child dentry | |
2674 | * | |
2675 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2676 | * an ancestor of p2, else NULL. | |
2677 | */ | |
2678 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) | |
2679 | { | |
2680 | struct dentry *p; | |
2681 | ||
2682 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { | |
2683 | if (p->d_parent == p1) | |
2684 | return p; | |
2685 | } | |
2686 | return NULL; | |
2687 | } | |
2688 | ||
2689 | /* | |
2690 | * This helper attempts to cope with remotely renamed directories | |
2691 | * | |
2692 | * It assumes that the caller is already holding | |
2693 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock | |
2694 | * | |
2695 | * Note: If ever the locking in lock_rename() changes, then please | |
2696 | * remember to update this too... | |
2697 | */ | |
2698 | static struct dentry *__d_unalias(struct inode *inode, | |
2699 | struct dentry *dentry, struct dentry *alias) | |
2700 | { | |
2701 | struct mutex *m1 = NULL, *m2 = NULL; | |
2702 | struct dentry *ret = ERR_PTR(-EBUSY); | |
2703 | ||
2704 | /* If alias and dentry share a parent, then no extra locks required */ | |
2705 | if (alias->d_parent == dentry->d_parent) | |
2706 | goto out_unalias; | |
2707 | ||
2708 | /* See lock_rename() */ | |
2709 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
2710 | goto out_err; | |
2711 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2712 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2713 | goto out_err; | |
2714 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2715 | out_unalias: | |
2716 | if (likely(!d_mountpoint(alias))) { | |
2717 | __d_move(alias, dentry); | |
2718 | ret = alias; | |
2719 | } | |
2720 | out_err: | |
2721 | spin_unlock(&inode->i_lock); | |
2722 | if (m2) | |
2723 | mutex_unlock(m2); | |
2724 | if (m1) | |
2725 | mutex_unlock(m1); | |
2726 | return ret; | |
2727 | } | |
2728 | ||
2729 | /* | |
2730 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2731 | * named dentry in place of the dentry to be replaced. | |
2732 | * returns with anon->d_lock held! | |
2733 | */ | |
2734 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2735 | { | |
2736 | struct dentry *dparent; | |
2737 | ||
2738 | dentry_lock_for_move(anon, dentry); | |
2739 | ||
2740 | write_seqcount_begin(&dentry->d_seq); | |
2741 | write_seqcount_begin(&anon->d_seq); | |
2742 | ||
2743 | dparent = dentry->d_parent; | |
2744 | ||
2745 | switch_names(dentry, anon); | |
2746 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2747 | ||
2748 | dentry->d_parent = dentry; | |
2749 | list_del_init(&dentry->d_u.d_child); | |
2750 | anon->d_parent = dparent; | |
2751 | list_move(&anon->d_u.d_child, &dparent->d_subdirs); | |
2752 | ||
2753 | write_seqcount_end(&dentry->d_seq); | |
2754 | write_seqcount_end(&anon->d_seq); | |
2755 | ||
2756 | dentry_unlock_parents_for_move(anon, dentry); | |
2757 | spin_unlock(&dentry->d_lock); | |
2758 | ||
2759 | /* anon->d_lock still locked, returns locked */ | |
2760 | } | |
2761 | ||
2762 | /** | |
2763 | * d_materialise_unique - introduce an inode into the tree | |
2764 | * @dentry: candidate dentry | |
2765 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2766 | * | |
2767 | * Introduces an dentry into the tree, substituting an extant disconnected | |
2768 | * root directory alias in its place if there is one. Caller must hold the | |
2769 | * i_mutex of the parent directory. | |
2770 | */ | |
2771 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2772 | { | |
2773 | struct dentry *actual; | |
2774 | ||
2775 | BUG_ON(!d_unhashed(dentry)); | |
2776 | ||
2777 | if (!inode) { | |
2778 | actual = dentry; | |
2779 | __d_instantiate(dentry, NULL); | |
2780 | d_rehash(actual); | |
2781 | goto out_nolock; | |
2782 | } | |
2783 | ||
2784 | spin_lock(&inode->i_lock); | |
2785 | ||
2786 | if (S_ISDIR(inode->i_mode)) { | |
2787 | struct dentry *alias; | |
2788 | ||
2789 | /* Does an aliased dentry already exist? */ | |
2790 | alias = __d_find_alias(inode, 0); | |
2791 | if (alias) { | |
2792 | actual = alias; | |
2793 | write_seqlock(&rename_lock); | |
2794 | ||
2795 | if (d_ancestor(alias, dentry)) { | |
2796 | /* Check for loops */ | |
2797 | actual = ERR_PTR(-ELOOP); | |
2798 | spin_unlock(&inode->i_lock); | |
2799 | } else if (IS_ROOT(alias)) { | |
2800 | /* Is this an anonymous mountpoint that we | |
2801 | * could splice into our tree? */ | |
2802 | __d_materialise_dentry(dentry, alias); | |
2803 | write_sequnlock(&rename_lock); | |
2804 | __d_drop(alias); | |
2805 | goto found; | |
2806 | } else { | |
2807 | /* Nope, but we must(!) avoid directory | |
2808 | * aliasing. This drops inode->i_lock */ | |
2809 | actual = __d_unalias(inode, dentry, alias); | |
2810 | } | |
2811 | write_sequnlock(&rename_lock); | |
2812 | if (IS_ERR(actual)) { | |
2813 | if (PTR_ERR(actual) == -ELOOP) | |
2814 | pr_warn_ratelimited( | |
2815 | "VFS: Lookup of '%s' in %s %s" | |
2816 | " would have caused loop\n", | |
2817 | dentry->d_name.name, | |
2818 | inode->i_sb->s_type->name, | |
2819 | inode->i_sb->s_id); | |
2820 | dput(alias); | |
2821 | } | |
2822 | goto out_nolock; | |
2823 | } | |
2824 | } | |
2825 | ||
2826 | /* Add a unique reference */ | |
2827 | actual = __d_instantiate_unique(dentry, inode); | |
2828 | if (!actual) | |
2829 | actual = dentry; | |
2830 | else | |
2831 | BUG_ON(!d_unhashed(actual)); | |
2832 | ||
2833 | spin_lock(&actual->d_lock); | |
2834 | found: | |
2835 | _d_rehash(actual); | |
2836 | spin_unlock(&actual->d_lock); | |
2837 | spin_unlock(&inode->i_lock); | |
2838 | out_nolock: | |
2839 | if (actual == dentry) { | |
2840 | security_d_instantiate(dentry, inode); | |
2841 | return NULL; | |
2842 | } | |
2843 | ||
2844 | iput(inode); | |
2845 | return actual; | |
2846 | } | |
2847 | EXPORT_SYMBOL_GPL(d_materialise_unique); | |
2848 | ||
2849 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) | |
2850 | { | |
2851 | *buflen -= namelen; | |
2852 | if (*buflen < 0) | |
2853 | return -ENAMETOOLONG; | |
2854 | *buffer -= namelen; | |
2855 | memcpy(*buffer, str, namelen); | |
2856 | return 0; | |
2857 | } | |
2858 | ||
2859 | /** | |
2860 | * prepend_name - prepend a pathname in front of current buffer pointer | |
2861 | * @buffer: buffer pointer | |
2862 | * @buflen: allocated length of the buffer | |
2863 | * @name: name string and length qstr structure | |
2864 | * | |
2865 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
2866 | * make sure that either the old or the new name pointer and length are | |
2867 | * fetched. However, there may be mismatch between length and pointer. | |
2868 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
2869 | * the length is reached or a null byte is found. It also prepends "/" at | |
2870 | * the beginning of the name. The sequence number check at the caller will | |
2871 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
2872 | * due to mismatched pointer and length will be discarded. | |
2873 | */ | |
2874 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) | |
2875 | { | |
2876 | const char *dname = ACCESS_ONCE(name->name); | |
2877 | u32 dlen = ACCESS_ONCE(name->len); | |
2878 | char *p; | |
2879 | ||
2880 | if (*buflen < dlen + 1) | |
2881 | return -ENAMETOOLONG; | |
2882 | *buflen -= dlen + 1; | |
2883 | p = *buffer -= dlen + 1; | |
2884 | *p++ = '/'; | |
2885 | while (dlen--) { | |
2886 | char c = *dname++; | |
2887 | if (!c) | |
2888 | break; | |
2889 | *p++ = c; | |
2890 | } | |
2891 | return 0; | |
2892 | } | |
2893 | ||
2894 | /** | |
2895 | * prepend_path - Prepend path string to a buffer | |
2896 | * @path: the dentry/vfsmount to report | |
2897 | * @root: root vfsmnt/dentry | |
2898 | * @buffer: pointer to the end of the buffer | |
2899 | * @buflen: pointer to buffer length | |
2900 | * | |
2901 | * The function will first try to write out the pathname without taking any | |
2902 | * lock other than the RCU read lock to make sure that dentries won't go away. | |
2903 | * It only checks the sequence number of the global rename_lock as any change | |
2904 | * in the dentry's d_seq will be preceded by changes in the rename_lock | |
2905 | * sequence number. If the sequence number had been changed, it will restart | |
2906 | * the whole pathname back-tracing sequence again by taking the rename_lock. | |
2907 | * In this case, there is no need to take the RCU read lock as the recursive | |
2908 | * parent pointer references will keep the dentry chain alive as long as no | |
2909 | * rename operation is performed. | |
2910 | */ | |
2911 | static int prepend_path(const struct path *path, | |
2912 | const struct path *root, | |
2913 | char **buffer, int *buflen) | |
2914 | { | |
2915 | struct dentry *dentry = path->dentry; | |
2916 | struct vfsmount *vfsmnt = path->mnt; | |
2917 | struct mount *mnt = real_mount(vfsmnt); | |
2918 | int error = 0; | |
2919 | unsigned seq, m_seq = 0; | |
2920 | char *bptr; | |
2921 | int blen; | |
2922 | ||
2923 | rcu_read_lock(); | |
2924 | restart_mnt: | |
2925 | read_seqbegin_or_lock(&mount_lock, &m_seq); | |
2926 | seq = 0; | |
2927 | rcu_read_lock(); | |
2928 | restart: | |
2929 | bptr = *buffer; | |
2930 | blen = *buflen; | |
2931 | error = 0; | |
2932 | read_seqbegin_or_lock(&rename_lock, &seq); | |
2933 | while (dentry != root->dentry || vfsmnt != root->mnt) { | |
2934 | struct dentry * parent; | |
2935 | ||
2936 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { | |
2937 | struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); | |
2938 | /* Global root? */ | |
2939 | if (mnt != parent) { | |
2940 | dentry = ACCESS_ONCE(mnt->mnt_mountpoint); | |
2941 | mnt = parent; | |
2942 | vfsmnt = &mnt->mnt; | |
2943 | continue; | |
2944 | } | |
2945 | /* | |
2946 | * Filesystems needing to implement special "root names" | |
2947 | * should do so with ->d_dname() | |
2948 | */ | |
2949 | if (IS_ROOT(dentry) && | |
2950 | (dentry->d_name.len != 1 || | |
2951 | dentry->d_name.name[0] != '/')) { | |
2952 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2953 | (int) dentry->d_name.len, | |
2954 | dentry->d_name.name); | |
2955 | } | |
2956 | if (!error) | |
2957 | error = is_mounted(vfsmnt) ? 1 : 2; | |
2958 | break; | |
2959 | } | |
2960 | parent = dentry->d_parent; | |
2961 | prefetch(parent); | |
2962 | error = prepend_name(&bptr, &blen, &dentry->d_name); | |
2963 | if (error) | |
2964 | break; | |
2965 | ||
2966 | dentry = parent; | |
2967 | } | |
2968 | if (!(seq & 1)) | |
2969 | rcu_read_unlock(); | |
2970 | if (need_seqretry(&rename_lock, seq)) { | |
2971 | seq = 1; | |
2972 | goto restart; | |
2973 | } | |
2974 | done_seqretry(&rename_lock, seq); | |
2975 | ||
2976 | if (!(m_seq & 1)) | |
2977 | rcu_read_unlock(); | |
2978 | if (need_seqretry(&mount_lock, m_seq)) { | |
2979 | m_seq = 1; | |
2980 | goto restart_mnt; | |
2981 | } | |
2982 | done_seqretry(&mount_lock, m_seq); | |
2983 | ||
2984 | if (error >= 0 && bptr == *buffer) { | |
2985 | if (--blen < 0) | |
2986 | error = -ENAMETOOLONG; | |
2987 | else | |
2988 | *--bptr = '/'; | |
2989 | } | |
2990 | *buffer = bptr; | |
2991 | *buflen = blen; | |
2992 | return error; | |
2993 | } | |
2994 | ||
2995 | /** | |
2996 | * __d_path - return the path of a dentry | |
2997 | * @path: the dentry/vfsmount to report | |
2998 | * @root: root vfsmnt/dentry | |
2999 | * @buf: buffer to return value in | |
3000 | * @buflen: buffer length | |
3001 | * | |
3002 | * Convert a dentry into an ASCII path name. | |
3003 | * | |
3004 | * Returns a pointer into the buffer or an error code if the | |
3005 | * path was too long. | |
3006 | * | |
3007 | * "buflen" should be positive. | |
3008 | * | |
3009 | * If the path is not reachable from the supplied root, return %NULL. | |
3010 | */ | |
3011 | char *__d_path(const struct path *path, | |
3012 | const struct path *root, | |
3013 | char *buf, int buflen) | |
3014 | { | |
3015 | char *res = buf + buflen; | |
3016 | int error; | |
3017 | ||
3018 | prepend(&res, &buflen, "\0", 1); | |
3019 | error = prepend_path(path, root, &res, &buflen); | |
3020 | ||
3021 | if (error < 0) | |
3022 | return ERR_PTR(error); | |
3023 | if (error > 0) | |
3024 | return NULL; | |
3025 | return res; | |
3026 | } | |
3027 | ||
3028 | char *d_absolute_path(const struct path *path, | |
3029 | char *buf, int buflen) | |
3030 | { | |
3031 | struct path root = {}; | |
3032 | char *res = buf + buflen; | |
3033 | int error; | |
3034 | ||
3035 | prepend(&res, &buflen, "\0", 1); | |
3036 | error = prepend_path(path, &root, &res, &buflen); | |
3037 | ||
3038 | if (error > 1) | |
3039 | error = -EINVAL; | |
3040 | if (error < 0) | |
3041 | return ERR_PTR(error); | |
3042 | return res; | |
3043 | } | |
3044 | ||
3045 | /* | |
3046 | * same as __d_path but appends "(deleted)" for unlinked files. | |
3047 | */ | |
3048 | static int path_with_deleted(const struct path *path, | |
3049 | const struct path *root, | |
3050 | char **buf, int *buflen) | |
3051 | { | |
3052 | prepend(buf, buflen, "\0", 1); | |
3053 | if (d_unlinked(path->dentry)) { | |
3054 | int error = prepend(buf, buflen, " (deleted)", 10); | |
3055 | if (error) | |
3056 | return error; | |
3057 | } | |
3058 | ||
3059 | return prepend_path(path, root, buf, buflen); | |
3060 | } | |
3061 | ||
3062 | static int prepend_unreachable(char **buffer, int *buflen) | |
3063 | { | |
3064 | return prepend(buffer, buflen, "(unreachable)", 13); | |
3065 | } | |
3066 | ||
3067 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) | |
3068 | { | |
3069 | unsigned seq; | |
3070 | ||
3071 | do { | |
3072 | seq = read_seqcount_begin(&fs->seq); | |
3073 | *root = fs->root; | |
3074 | } while (read_seqcount_retry(&fs->seq, seq)); | |
3075 | } | |
3076 | ||
3077 | /** | |
3078 | * d_path - return the path of a dentry | |
3079 | * @path: path to report | |
3080 | * @buf: buffer to return value in | |
3081 | * @buflen: buffer length | |
3082 | * | |
3083 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
3084 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
3085 | * | |
3086 | * Returns a pointer into the buffer or an error code if the path was | |
3087 | * too long. Note: Callers should use the returned pointer, not the passed | |
3088 | * in buffer, to use the name! The implementation often starts at an offset | |
3089 | * into the buffer, and may leave 0 bytes at the start. | |
3090 | * | |
3091 | * "buflen" should be positive. | |
3092 | */ | |
3093 | char *d_path(const struct path *path, char *buf, int buflen) | |
3094 | { | |
3095 | char *res = buf + buflen; | |
3096 | struct path root; | |
3097 | int error; | |
3098 | ||
3099 | /* | |
3100 | * We have various synthetic filesystems that never get mounted. On | |
3101 | * these filesystems dentries are never used for lookup purposes, and | |
3102 | * thus don't need to be hashed. They also don't need a name until a | |
3103 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
3104 | * below allows us to generate a name for these objects on demand: | |
3105 | */ | |
3106 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
3107 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
3108 | ||
3109 | rcu_read_lock(); | |
3110 | get_fs_root_rcu(current->fs, &root); | |
3111 | error = path_with_deleted(path, &root, &res, &buflen); | |
3112 | rcu_read_unlock(); | |
3113 | ||
3114 | if (error < 0) | |
3115 | res = ERR_PTR(error); | |
3116 | return res; | |
3117 | } | |
3118 | EXPORT_SYMBOL(d_path); | |
3119 | ||
3120 | /* | |
3121 | * Helper function for dentry_operations.d_dname() members | |
3122 | */ | |
3123 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
3124 | const char *fmt, ...) | |
3125 | { | |
3126 | va_list args; | |
3127 | char temp[64]; | |
3128 | int sz; | |
3129 | ||
3130 | va_start(args, fmt); | |
3131 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
3132 | va_end(args); | |
3133 | ||
3134 | if (sz > sizeof(temp) || sz > buflen) | |
3135 | return ERR_PTR(-ENAMETOOLONG); | |
3136 | ||
3137 | buffer += buflen - sz; | |
3138 | return memcpy(buffer, temp, sz); | |
3139 | } | |
3140 | ||
3141 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) | |
3142 | { | |
3143 | char *end = buffer + buflen; | |
3144 | /* these dentries are never renamed, so d_lock is not needed */ | |
3145 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
3146 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || | |
3147 | prepend(&end, &buflen, "/", 1)) | |
3148 | end = ERR_PTR(-ENAMETOOLONG); | |
3149 | return end; | |
3150 | } | |
3151 | ||
3152 | /* | |
3153 | * Write full pathname from the root of the filesystem into the buffer. | |
3154 | */ | |
3155 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3156 | { | |
3157 | char *end, *retval; | |
3158 | int len, seq = 0; | |
3159 | int error = 0; | |
3160 | ||
3161 | rcu_read_lock(); | |
3162 | restart: | |
3163 | end = buf + buflen; | |
3164 | len = buflen; | |
3165 | prepend(&end, &len, "\0", 1); | |
3166 | if (buflen < 1) | |
3167 | goto Elong; | |
3168 | /* Get '/' right */ | |
3169 | retval = end-1; | |
3170 | *retval = '/'; | |
3171 | read_seqbegin_or_lock(&rename_lock, &seq); | |
3172 | while (!IS_ROOT(dentry)) { | |
3173 | struct dentry *parent = dentry->d_parent; | |
3174 | int error; | |
3175 | ||
3176 | prefetch(parent); | |
3177 | error = prepend_name(&end, &len, &dentry->d_name); | |
3178 | if (error) | |
3179 | break; | |
3180 | ||
3181 | retval = end; | |
3182 | dentry = parent; | |
3183 | } | |
3184 | if (!(seq & 1)) | |
3185 | rcu_read_unlock(); | |
3186 | if (need_seqretry(&rename_lock, seq)) { | |
3187 | seq = 1; | |
3188 | goto restart; | |
3189 | } | |
3190 | done_seqretry(&rename_lock, seq); | |
3191 | if (error) | |
3192 | goto Elong; | |
3193 | return retval; | |
3194 | Elong: | |
3195 | return ERR_PTR(-ENAMETOOLONG); | |
3196 | } | |
3197 | ||
3198 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3199 | { | |
3200 | return __dentry_path(dentry, buf, buflen); | |
3201 | } | |
3202 | EXPORT_SYMBOL(dentry_path_raw); | |
3203 | ||
3204 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3205 | { | |
3206 | char *p = NULL; | |
3207 | char *retval; | |
3208 | ||
3209 | if (d_unlinked(dentry)) { | |
3210 | p = buf + buflen; | |
3211 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3212 | goto Elong; | |
3213 | buflen++; | |
3214 | } | |
3215 | retval = __dentry_path(dentry, buf, buflen); | |
3216 | if (!IS_ERR(retval) && p) | |
3217 | *p = '/'; /* restore '/' overriden with '\0' */ | |
3218 | return retval; | |
3219 | Elong: | |
3220 | return ERR_PTR(-ENAMETOOLONG); | |
3221 | } | |
3222 | ||
3223 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, | |
3224 | struct path *pwd) | |
3225 | { | |
3226 | unsigned seq; | |
3227 | ||
3228 | do { | |
3229 | seq = read_seqcount_begin(&fs->seq); | |
3230 | *root = fs->root; | |
3231 | *pwd = fs->pwd; | |
3232 | } while (read_seqcount_retry(&fs->seq, seq)); | |
3233 | } | |
3234 | ||
3235 | /* | |
3236 | * NOTE! The user-level library version returns a | |
3237 | * character pointer. The kernel system call just | |
3238 | * returns the length of the buffer filled (which | |
3239 | * includes the ending '\0' character), or a negative | |
3240 | * error value. So libc would do something like | |
3241 | * | |
3242 | * char *getcwd(char * buf, size_t size) | |
3243 | * { | |
3244 | * int retval; | |
3245 | * | |
3246 | * retval = sys_getcwd(buf, size); | |
3247 | * if (retval >= 0) | |
3248 | * return buf; | |
3249 | * errno = -retval; | |
3250 | * return NULL; | |
3251 | * } | |
3252 | */ | |
3253 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) | |
3254 | { | |
3255 | int error; | |
3256 | struct path pwd, root; | |
3257 | char *page = __getname(); | |
3258 | ||
3259 | if (!page) | |
3260 | return -ENOMEM; | |
3261 | ||
3262 | rcu_read_lock(); | |
3263 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); | |
3264 | ||
3265 | error = -ENOENT; | |
3266 | if (!d_unlinked(pwd.dentry)) { | |
3267 | unsigned long len; | |
3268 | char *cwd = page + PATH_MAX; | |
3269 | int buflen = PATH_MAX; | |
3270 | ||
3271 | prepend(&cwd, &buflen, "\0", 1); | |
3272 | error = prepend_path(&pwd, &root, &cwd, &buflen); | |
3273 | rcu_read_unlock(); | |
3274 | ||
3275 | if (error < 0) | |
3276 | goto out; | |
3277 | ||
3278 | /* Unreachable from current root */ | |
3279 | if (error > 0) { | |
3280 | error = prepend_unreachable(&cwd, &buflen); | |
3281 | if (error) | |
3282 | goto out; | |
3283 | } | |
3284 | ||
3285 | error = -ERANGE; | |
3286 | len = PATH_MAX + page - cwd; | |
3287 | if (len <= size) { | |
3288 | error = len; | |
3289 | if (copy_to_user(buf, cwd, len)) | |
3290 | error = -EFAULT; | |
3291 | } | |
3292 | } else { | |
3293 | rcu_read_unlock(); | |
3294 | } | |
3295 | ||
3296 | out: | |
3297 | __putname(page); | |
3298 | return error; | |
3299 | } | |
3300 | ||
3301 | /* | |
3302 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3303 | * | |
3304 | * Trivially implemented using the dcache structure | |
3305 | */ | |
3306 | ||
3307 | /** | |
3308 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3309 | * @new_dentry: new dentry | |
3310 | * @old_dentry: old dentry | |
3311 | * | |
3312 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
3313 | * Returns 0 otherwise. | |
3314 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
3315 | */ | |
3316 | ||
3317 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) | |
3318 | { | |
3319 | int result; | |
3320 | unsigned seq; | |
3321 | ||
3322 | if (new_dentry == old_dentry) | |
3323 | return 1; | |
3324 | ||
3325 | do { | |
3326 | /* for restarting inner loop in case of seq retry */ | |
3327 | seq = read_seqbegin(&rename_lock); | |
3328 | /* | |
3329 | * Need rcu_readlock to protect against the d_parent trashing | |
3330 | * due to d_move | |
3331 | */ | |
3332 | rcu_read_lock(); | |
3333 | if (d_ancestor(old_dentry, new_dentry)) | |
3334 | result = 1; | |
3335 | else | |
3336 | result = 0; | |
3337 | rcu_read_unlock(); | |
3338 | } while (read_seqretry(&rename_lock, seq)); | |
3339 | ||
3340 | return result; | |
3341 | } | |
3342 | ||
3343 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) | |
3344 | { | |
3345 | struct dentry *root = data; | |
3346 | if (dentry != root) { | |
3347 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3348 | return D_WALK_SKIP; | |
3349 | ||
3350 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { | |
3351 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3352 | dentry->d_lockref.count--; | |
3353 | } | |
3354 | } | |
3355 | return D_WALK_CONTINUE; | |
3356 | } | |
3357 | ||
3358 | void d_genocide(struct dentry *parent) | |
3359 | { | |
3360 | d_walk(parent, parent, d_genocide_kill, NULL); | |
3361 | } | |
3362 | ||
3363 | void d_tmpfile(struct dentry *dentry, struct inode *inode) | |
3364 | { | |
3365 | inode_dec_link_count(inode); | |
3366 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
3367 | !hlist_unhashed(&dentry->d_alias) || | |
3368 | !d_unlinked(dentry)); | |
3369 | spin_lock(&dentry->d_parent->d_lock); | |
3370 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3371 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3372 | (unsigned long long)inode->i_ino); | |
3373 | spin_unlock(&dentry->d_lock); | |
3374 | spin_unlock(&dentry->d_parent->d_lock); | |
3375 | d_instantiate(dentry, inode); | |
3376 | } | |
3377 | EXPORT_SYMBOL(d_tmpfile); | |
3378 | ||
3379 | static __initdata unsigned long dhash_entries; | |
3380 | static int __init set_dhash_entries(char *str) | |
3381 | { | |
3382 | if (!str) | |
3383 | return 0; | |
3384 | dhash_entries = simple_strtoul(str, &str, 0); | |
3385 | return 1; | |
3386 | } | |
3387 | __setup("dhash_entries=", set_dhash_entries); | |
3388 | ||
3389 | static void __init dcache_init_early(void) | |
3390 | { | |
3391 | unsigned int loop; | |
3392 | ||
3393 | /* If hashes are distributed across NUMA nodes, defer | |
3394 | * hash allocation until vmalloc space is available. | |
3395 | */ | |
3396 | if (hashdist) | |
3397 | return; | |
3398 | ||
3399 | dentry_hashtable = | |
3400 | alloc_large_system_hash("Dentry cache", | |
3401 | sizeof(struct hlist_bl_head), | |
3402 | dhash_entries, | |
3403 | 13, | |
3404 | HASH_EARLY, | |
3405 | &d_hash_shift, | |
3406 | &d_hash_mask, | |
3407 | 0, | |
3408 | 0); | |
3409 | ||
3410 | for (loop = 0; loop < (1U << d_hash_shift); loop++) | |
3411 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); | |
3412 | } | |
3413 | ||
3414 | static void __init dcache_init(void) | |
3415 | { | |
3416 | unsigned int loop; | |
3417 | ||
3418 | /* | |
3419 | * A constructor could be added for stable state like the lists, | |
3420 | * but it is probably not worth it because of the cache nature | |
3421 | * of the dcache. | |
3422 | */ | |
3423 | dentry_cache = KMEM_CACHE(dentry, | |
3424 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
3425 | ||
3426 | /* Hash may have been set up in dcache_init_early */ | |
3427 | if (!hashdist) | |
3428 | return; | |
3429 | ||
3430 | dentry_hashtable = | |
3431 | alloc_large_system_hash("Dentry cache", | |
3432 | sizeof(struct hlist_bl_head), | |
3433 | dhash_entries, | |
3434 | 13, | |
3435 | 0, | |
3436 | &d_hash_shift, | |
3437 | &d_hash_mask, | |
3438 | 0, | |
3439 | 0); | |
3440 | ||
3441 | for (loop = 0; loop < (1U << d_hash_shift); loop++) | |
3442 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); | |
3443 | } | |
3444 | ||
3445 | /* SLAB cache for __getname() consumers */ | |
3446 | struct kmem_cache *names_cachep __read_mostly; | |
3447 | EXPORT_SYMBOL(names_cachep); | |
3448 | ||
3449 | EXPORT_SYMBOL(d_genocide); | |
3450 | ||
3451 | void __init vfs_caches_init_early(void) | |
3452 | { | |
3453 | dcache_init_early(); | |
3454 | inode_init_early(); | |
3455 | } | |
3456 | ||
3457 | void __init vfs_caches_init(unsigned long mempages) | |
3458 | { | |
3459 | unsigned long reserve; | |
3460 | ||
3461 | /* Base hash sizes on available memory, with a reserve equal to | |
3462 | 150% of current kernel size */ | |
3463 | ||
3464 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3465 | mempages -= reserve; | |
3466 | ||
3467 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
3468 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); | |
3469 | ||
3470 | dcache_init(); | |
3471 | inode_init(); | |
3472 | files_init(mempages); | |
3473 | mnt_init(); | |
3474 | bdev_cache_init(); | |
3475 | chrdev_init(); | |
3476 | } |