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