Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/dcache.c | |
3 | * | |
4 | * Complete reimplementation | |
5 | * (C) 1997 Thomas Schoebel-Theuer, | |
6 | * with heavy changes by Linus Torvalds | |
7 | */ | |
8 | ||
9 | /* | |
10 | * Notes on the allocation strategy: | |
11 | * | |
12 | * The dcache is a master of the icache - whenever a dcache entry | |
13 | * exists, the inode will always exist. "iput()" is done either when | |
14 | * the dcache entry is deleted or garbage collected. | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/syscalls.h> |
18 | #include <linux/string.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/fs.h> | |
7a91bf7f | 21 | #include <linux/fsnotify.h> |
1da177e4 LT |
22 | #include <linux/slab.h> |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/hash.h> |
25 | #include <linux/cache.h> | |
630d9c47 | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/mount.h> |
28 | #include <linux/file.h> | |
29 | #include <asm/uaccess.h> | |
30 | #include <linux/security.h> | |
31 | #include <linux/seqlock.h> | |
32 | #include <linux/swap.h> | |
33 | #include <linux/bootmem.h> | |
5ad4e53b | 34 | #include <linux/fs_struct.h> |
613afbf8 | 35 | #include <linux/hardirq.h> |
ceb5bdc2 NP |
36 | #include <linux/bit_spinlock.h> |
37 | #include <linux/rculist_bl.h> | |
268bb0ce | 38 | #include <linux/prefetch.h> |
dd179946 | 39 | #include <linux/ratelimit.h> |
f6041567 | 40 | #include <linux/list_lru.h> |
07f3f05c | 41 | #include "internal.h" |
b2dba1af | 42 | #include "mount.h" |
1da177e4 | 43 | |
789680d1 NP |
44 | /* |
45 | * Usage: | |
873feea0 NP |
46 | * dcache->d_inode->i_lock protects: |
47 | * - i_dentry, d_alias, d_inode of aliases | |
ceb5bdc2 NP |
48 | * dcache_hash_bucket lock protects: |
49 | * - the dcache hash table | |
50 | * s_anon bl list spinlock protects: | |
51 | * - the s_anon list (see __d_drop) | |
19156840 | 52 | * dentry->d_sb->s_dentry_lru_lock protects: |
23044507 NP |
53 | * - the dcache lru lists and counters |
54 | * d_lock protects: | |
55 | * - d_flags | |
56 | * - d_name | |
57 | * - d_lru | |
b7ab39f6 | 58 | * - d_count |
da502956 | 59 | * - d_unhashed() |
2fd6b7f5 NP |
60 | * - d_parent and d_subdirs |
61 | * - childrens' d_child and d_parent | |
b23fb0a6 | 62 | * - d_alias, d_inode |
789680d1 NP |
63 | * |
64 | * Ordering: | |
873feea0 | 65 | * dentry->d_inode->i_lock |
b5c84bf6 | 66 | * dentry->d_lock |
19156840 | 67 | * dentry->d_sb->s_dentry_lru_lock |
ceb5bdc2 NP |
68 | * dcache_hash_bucket lock |
69 | * s_anon lock | |
789680d1 | 70 | * |
da502956 NP |
71 | * If there is an ancestor relationship: |
72 | * dentry->d_parent->...->d_parent->d_lock | |
73 | * ... | |
74 | * dentry->d_parent->d_lock | |
75 | * dentry->d_lock | |
76 | * | |
77 | * If no ancestor relationship: | |
789680d1 NP |
78 | * if (dentry1 < dentry2) |
79 | * dentry1->d_lock | |
80 | * dentry2->d_lock | |
81 | */ | |
fa3536cc | 82 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
83 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
84 | ||
74c3cbe3 | 85 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 86 | |
949854d0 | 87 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 88 | |
e18b890b | 89 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 90 | |
1da177e4 LT |
91 | /* |
92 | * This is the single most critical data structure when it comes | |
93 | * to the dcache: the hashtable for lookups. Somebody should try | |
94 | * to make this good - I've just made it work. | |
95 | * | |
96 | * This hash-function tries to avoid losing too many bits of hash | |
97 | * information, yet avoid using a prime hash-size or similar. | |
98 | */ | |
1da177e4 | 99 | |
fa3536cc ED |
100 | static unsigned int d_hash_mask __read_mostly; |
101 | static unsigned int d_hash_shift __read_mostly; | |
ceb5bdc2 | 102 | |
b07ad996 | 103 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 104 | |
8966be90 | 105 | static inline struct hlist_bl_head *d_hash(const struct dentry *parent, |
6d7d1a0d | 106 | unsigned int hash) |
ceb5bdc2 | 107 | { |
6d7d1a0d | 108 | hash += (unsigned long) parent / L1_CACHE_BYTES; |
99d263d4 | 109 | return dentry_hashtable + hash_32(hash, d_hash_shift); |
ceb5bdc2 NP |
110 | } |
111 | ||
1da177e4 LT |
112 | /* Statistics gathering. */ |
113 | struct dentry_stat_t dentry_stat = { | |
114 | .age_limit = 45, | |
115 | }; | |
116 | ||
3942c07c | 117 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 118 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
312d3ca8 CH |
119 | |
120 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
62d36c77 DC |
121 | |
122 | /* | |
123 | * Here we resort to our own counters instead of using generic per-cpu counters | |
124 | * for consistency with what the vfs inode code does. We are expected to harvest | |
125 | * better code and performance by having our own specialized counters. | |
126 | * | |
127 | * Please note that the loop is done over all possible CPUs, not over all online | |
128 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
129 | * on and off. If one of them goes off, we will just keep their counters. | |
130 | * | |
131 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
132 | * please update all vfs counters to match. | |
133 | */ | |
3942c07c | 134 | static long get_nr_dentry(void) |
3e880fb5 NP |
135 | { |
136 | int i; | |
3942c07c | 137 | long sum = 0; |
3e880fb5 NP |
138 | for_each_possible_cpu(i) |
139 | sum += per_cpu(nr_dentry, i); | |
140 | return sum < 0 ? 0 : sum; | |
141 | } | |
142 | ||
62d36c77 DC |
143 | static long get_nr_dentry_unused(void) |
144 | { | |
145 | int i; | |
146 | long sum = 0; | |
147 | for_each_possible_cpu(i) | |
148 | sum += per_cpu(nr_dentry_unused, i); | |
149 | return sum < 0 ? 0 : sum; | |
150 | } | |
151 | ||
1f7e0616 | 152 | int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer, |
312d3ca8 CH |
153 | size_t *lenp, loff_t *ppos) |
154 | { | |
3e880fb5 | 155 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 156 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
3942c07c | 157 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 CH |
158 | } |
159 | #endif | |
160 | ||
5483f18e LT |
161 | /* |
162 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
163 | * The strings are both count bytes long, and count is non-zero. | |
164 | */ | |
e419b4cc LT |
165 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
166 | ||
167 | #include <asm/word-at-a-time.h> | |
168 | /* | |
169 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
170 | * aligned allocation for this particular component. We don't | |
171 | * strictly need the load_unaligned_zeropad() safety, but it | |
172 | * doesn't hurt either. | |
173 | * | |
174 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
175 | * need the careful unaligned handling. | |
176 | */ | |
94753db5 | 177 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 178 | { |
bfcfaa77 | 179 | unsigned long a,b,mask; |
bfcfaa77 LT |
180 | |
181 | for (;;) { | |
12f8ad4b | 182 | a = *(unsigned long *)cs; |
e419b4cc | 183 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
184 | if (tcount < sizeof(unsigned long)) |
185 | break; | |
186 | if (unlikely(a != b)) | |
187 | return 1; | |
188 | cs += sizeof(unsigned long); | |
189 | ct += sizeof(unsigned long); | |
190 | tcount -= sizeof(unsigned long); | |
191 | if (!tcount) | |
192 | return 0; | |
193 | } | |
a5c21dce | 194 | mask = bytemask_from_count(tcount); |
bfcfaa77 | 195 | return unlikely(!!((a ^ b) & mask)); |
e419b4cc LT |
196 | } |
197 | ||
bfcfaa77 | 198 | #else |
e419b4cc | 199 | |
94753db5 | 200 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 201 | { |
5483f18e LT |
202 | do { |
203 | if (*cs != *ct) | |
204 | return 1; | |
205 | cs++; | |
206 | ct++; | |
207 | tcount--; | |
208 | } while (tcount); | |
209 | return 0; | |
210 | } | |
211 | ||
e419b4cc LT |
212 | #endif |
213 | ||
94753db5 LT |
214 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
215 | { | |
6326c71f | 216 | const unsigned char *cs; |
94753db5 LT |
217 | /* |
218 | * Be careful about RCU walk racing with rename: | |
219 | * use ACCESS_ONCE to fetch the name pointer. | |
220 | * | |
221 | * NOTE! Even if a rename will mean that the length | |
222 | * was not loaded atomically, we don't care. The | |
223 | * RCU walk will check the sequence count eventually, | |
224 | * and catch it. And we won't overrun the buffer, | |
225 | * because we're reading the name pointer atomically, | |
226 | * and a dentry name is guaranteed to be properly | |
227 | * terminated with a NUL byte. | |
228 | * | |
229 | * End result: even if 'len' is wrong, we'll exit | |
230 | * early because the data cannot match (there can | |
231 | * be no NUL in the ct/tcount data) | |
232 | */ | |
6326c71f LT |
233 | cs = ACCESS_ONCE(dentry->d_name.name); |
234 | smp_read_barrier_depends(); | |
235 | return dentry_string_cmp(cs, ct, tcount); | |
94753db5 LT |
236 | } |
237 | ||
9c82ab9c | 238 | static void __d_free(struct rcu_head *head) |
1da177e4 | 239 | { |
9c82ab9c CH |
240 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
241 | ||
b3d9b7a3 | 242 | WARN_ON(!hlist_unhashed(&dentry->d_alias)); |
1da177e4 LT |
243 | if (dname_external(dentry)) |
244 | kfree(dentry->d_name.name); | |
245 | kmem_cache_free(dentry_cache, dentry); | |
246 | } | |
247 | ||
b4f0354e AV |
248 | static void dentry_free(struct dentry *dentry) |
249 | { | |
250 | /* if dentry was never visible to RCU, immediate free is OK */ | |
251 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
252 | __d_free(&dentry->d_u.d_rcu); | |
253 | else | |
254 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
255 | } | |
256 | ||
31e6b01f NP |
257 | /** |
258 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
ff5fdb61 | 259 | * @dentry: the target dentry |
31e6b01f NP |
260 | * After this call, in-progress rcu-walk path lookup will fail. This |
261 | * should be called after unhashing, and after changing d_inode (if | |
262 | * the dentry has not already been unhashed). | |
263 | */ | |
264 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
265 | { | |
266 | assert_spin_locked(&dentry->d_lock); | |
267 | /* Go through a barrier */ | |
268 | write_seqcount_barrier(&dentry->d_seq); | |
269 | } | |
270 | ||
1da177e4 LT |
271 | /* |
272 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
273 | * d_iput() operation if defined. Dentry has no refcount |
274 | * and is unhashed. | |
1da177e4 | 275 | */ |
858119e1 | 276 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 277 | __releases(dentry->d_lock) |
873feea0 | 278 | __releases(dentry->d_inode->i_lock) |
1da177e4 LT |
279 | { |
280 | struct inode *inode = dentry->d_inode; | |
281 | if (inode) { | |
282 | dentry->d_inode = NULL; | |
b3d9b7a3 | 283 | hlist_del_init(&dentry->d_alias); |
1da177e4 | 284 | spin_unlock(&dentry->d_lock); |
873feea0 | 285 | spin_unlock(&inode->i_lock); |
f805fbda LT |
286 | if (!inode->i_nlink) |
287 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
288 | if (dentry->d_op && dentry->d_op->d_iput) |
289 | dentry->d_op->d_iput(dentry, inode); | |
290 | else | |
291 | iput(inode); | |
292 | } else { | |
293 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
294 | } |
295 | } | |
296 | ||
31e6b01f NP |
297 | /* |
298 | * Release the dentry's inode, using the filesystem | |
299 | * d_iput() operation if defined. dentry remains in-use. | |
300 | */ | |
301 | static void dentry_unlink_inode(struct dentry * dentry) | |
302 | __releases(dentry->d_lock) | |
873feea0 | 303 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
304 | { |
305 | struct inode *inode = dentry->d_inode; | |
b18825a7 | 306 | __d_clear_type(dentry); |
31e6b01f | 307 | dentry->d_inode = NULL; |
b3d9b7a3 | 308 | hlist_del_init(&dentry->d_alias); |
31e6b01f NP |
309 | dentry_rcuwalk_barrier(dentry); |
310 | spin_unlock(&dentry->d_lock); | |
873feea0 | 311 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
312 | if (!inode->i_nlink) |
313 | fsnotify_inoderemove(inode); | |
314 | if (dentry->d_op && dentry->d_op->d_iput) | |
315 | dentry->d_op->d_iput(dentry, inode); | |
316 | else | |
317 | iput(inode); | |
318 | } | |
319 | ||
89dc77bc LT |
320 | /* |
321 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
322 | * is in use - which includes both the "real" per-superblock | |
323 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
324 | * | |
325 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
326 | * on the shrink list (ie not on the superblock LRU list). | |
327 | * | |
328 | * The per-cpu "nr_dentry_unused" counters are updated with | |
329 | * the DCACHE_LRU_LIST bit. | |
330 | * | |
331 | * These helper functions make sure we always follow the | |
332 | * rules. d_lock must be held by the caller. | |
333 | */ | |
334 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
335 | static void d_lru_add(struct dentry *dentry) | |
336 | { | |
337 | D_FLAG_VERIFY(dentry, 0); | |
338 | dentry->d_flags |= DCACHE_LRU_LIST; | |
339 | this_cpu_inc(nr_dentry_unused); | |
340 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
341 | } | |
342 | ||
343 | static void d_lru_del(struct dentry *dentry) | |
344 | { | |
345 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
346 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
347 | this_cpu_dec(nr_dentry_unused); | |
348 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
349 | } | |
350 | ||
351 | static void d_shrink_del(struct dentry *dentry) | |
352 | { | |
353 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
354 | list_del_init(&dentry->d_lru); | |
355 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
356 | this_cpu_dec(nr_dentry_unused); | |
357 | } | |
358 | ||
359 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
360 | { | |
361 | D_FLAG_VERIFY(dentry, 0); | |
362 | list_add(&dentry->d_lru, list); | |
363 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
364 | this_cpu_inc(nr_dentry_unused); | |
365 | } | |
366 | ||
367 | /* | |
368 | * These can only be called under the global LRU lock, ie during the | |
369 | * callback for freeing the LRU list. "isolate" removes it from the | |
370 | * LRU lists entirely, while shrink_move moves it to the indicated | |
371 | * private list. | |
372 | */ | |
373 | static void d_lru_isolate(struct dentry *dentry) | |
374 | { | |
375 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
376 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
377 | this_cpu_dec(nr_dentry_unused); | |
378 | list_del_init(&dentry->d_lru); | |
379 | } | |
380 | ||
381 | static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list) | |
382 | { | |
383 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
384 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
385 | list_move_tail(&dentry->d_lru, list); | |
386 | } | |
387 | ||
da3bbdd4 | 388 | /* |
f6041567 | 389 | * dentry_lru_(add|del)_list) must be called with d_lock held. |
da3bbdd4 KM |
390 | */ |
391 | static void dentry_lru_add(struct dentry *dentry) | |
392 | { | |
89dc77bc LT |
393 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) |
394 | d_lru_add(dentry); | |
da3bbdd4 KM |
395 | } |
396 | ||
789680d1 NP |
397 | /** |
398 | * d_drop - drop a dentry | |
399 | * @dentry: dentry to drop | |
400 | * | |
401 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
402 | * be found through a VFS lookup any more. Note that this is different from | |
403 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
404 | * possible, giving a successful _negative_ lookup, while d_drop will | |
405 | * just make the cache lookup fail. | |
406 | * | |
407 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
408 | * reason (NFS timeouts or autofs deletes). | |
409 | * | |
410 | * __d_drop requires dentry->d_lock. | |
411 | */ | |
412 | void __d_drop(struct dentry *dentry) | |
413 | { | |
dea3667b | 414 | if (!d_unhashed(dentry)) { |
b61625d2 | 415 | struct hlist_bl_head *b; |
7632e465 BF |
416 | /* |
417 | * Hashed dentries are normally on the dentry hashtable, | |
418 | * with the exception of those newly allocated by | |
419 | * d_obtain_alias, which are always IS_ROOT: | |
420 | */ | |
421 | if (unlikely(IS_ROOT(dentry))) | |
b61625d2 AV |
422 | b = &dentry->d_sb->s_anon; |
423 | else | |
424 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
425 | ||
426 | hlist_bl_lock(b); | |
427 | __hlist_bl_del(&dentry->d_hash); | |
428 | dentry->d_hash.pprev = NULL; | |
429 | hlist_bl_unlock(b); | |
dea3667b | 430 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
431 | } |
432 | } | |
433 | EXPORT_SYMBOL(__d_drop); | |
434 | ||
435 | void d_drop(struct dentry *dentry) | |
436 | { | |
789680d1 NP |
437 | spin_lock(&dentry->d_lock); |
438 | __d_drop(dentry); | |
439 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
440 | } |
441 | EXPORT_SYMBOL(d_drop); | |
442 | ||
e55fd011 | 443 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 444 | { |
41edf278 AV |
445 | struct dentry *parent = NULL; |
446 | bool can_free = true; | |
41edf278 | 447 | if (!IS_ROOT(dentry)) |
77812a1e | 448 | parent = dentry->d_parent; |
31e6b01f | 449 | |
0d98439e LT |
450 | /* |
451 | * The dentry is now unrecoverably dead to the world. | |
452 | */ | |
453 | lockref_mark_dead(&dentry->d_lockref); | |
454 | ||
f0023bc6 | 455 | /* |
f0023bc6 SW |
456 | * inform the fs via d_prune that this dentry is about to be |
457 | * unhashed and destroyed. | |
458 | */ | |
590fb51f | 459 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry)) |
61572bb1 YZ |
460 | dentry->d_op->d_prune(dentry); |
461 | ||
01b60351 AV |
462 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
463 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
464 | d_lru_del(dentry); | |
01b60351 | 465 | } |
77812a1e NP |
466 | /* if it was on the hash then remove it */ |
467 | __d_drop(dentry); | |
03b3b889 AV |
468 | list_del(&dentry->d_u.d_child); |
469 | /* | |
470 | * Inform d_walk() that we are no longer attached to the | |
471 | * dentry tree | |
472 | */ | |
473 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
474 | if (parent) | |
475 | spin_unlock(&parent->d_lock); | |
476 | dentry_iput(dentry); | |
477 | /* | |
478 | * dentry_iput drops the locks, at which point nobody (except | |
479 | * transient RCU lookups) can reach this dentry. | |
480 | */ | |
481 | BUG_ON((int)dentry->d_lockref.count > 0); | |
482 | this_cpu_dec(nr_dentry); | |
483 | if (dentry->d_op && dentry->d_op->d_release) | |
484 | dentry->d_op->d_release(dentry); | |
485 | ||
41edf278 AV |
486 | spin_lock(&dentry->d_lock); |
487 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
488 | dentry->d_flags |= DCACHE_MAY_FREE; | |
489 | can_free = false; | |
490 | } | |
491 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
492 | if (likely(can_free)) |
493 | dentry_free(dentry); | |
e55fd011 AV |
494 | } |
495 | ||
496 | /* | |
497 | * Finish off a dentry we've decided to kill. | |
498 | * dentry->d_lock must be held, returns with it unlocked. | |
499 | * If ref is non-zero, then decrement the refcount too. | |
500 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
501 | */ | |
8cbf74da | 502 | static struct dentry *dentry_kill(struct dentry *dentry) |
e55fd011 AV |
503 | __releases(dentry->d_lock) |
504 | { | |
505 | struct inode *inode = dentry->d_inode; | |
506 | struct dentry *parent = NULL; | |
507 | ||
508 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
509 | goto failed; | |
510 | ||
511 | if (!IS_ROOT(dentry)) { | |
512 | parent = dentry->d_parent; | |
513 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
514 | if (inode) | |
515 | spin_unlock(&inode->i_lock); | |
516 | goto failed; | |
517 | } | |
518 | } | |
519 | ||
520 | __dentry_kill(dentry); | |
03b3b889 | 521 | return parent; |
e55fd011 AV |
522 | |
523 | failed: | |
8cbf74da AV |
524 | spin_unlock(&dentry->d_lock); |
525 | cpu_relax(); | |
e55fd011 | 526 | return dentry; /* try again with same dentry */ |
77812a1e NP |
527 | } |
528 | ||
046b961b AV |
529 | static inline struct dentry *lock_parent(struct dentry *dentry) |
530 | { | |
531 | struct dentry *parent = dentry->d_parent; | |
532 | if (IS_ROOT(dentry)) | |
533 | return NULL; | |
c2338f2d AV |
534 | if (unlikely((int)dentry->d_lockref.count < 0)) |
535 | return NULL; | |
046b961b AV |
536 | if (likely(spin_trylock(&parent->d_lock))) |
537 | return parent; | |
046b961b | 538 | rcu_read_lock(); |
c2338f2d | 539 | spin_unlock(&dentry->d_lock); |
046b961b AV |
540 | again: |
541 | parent = ACCESS_ONCE(dentry->d_parent); | |
542 | spin_lock(&parent->d_lock); | |
543 | /* | |
544 | * We can't blindly lock dentry until we are sure | |
545 | * that we won't violate the locking order. | |
546 | * Any changes of dentry->d_parent must have | |
547 | * been done with parent->d_lock held, so | |
548 | * spin_lock() above is enough of a barrier | |
549 | * for checking if it's still our child. | |
550 | */ | |
551 | if (unlikely(parent != dentry->d_parent)) { | |
552 | spin_unlock(&parent->d_lock); | |
553 | goto again; | |
554 | } | |
555 | rcu_read_unlock(); | |
556 | if (parent != dentry) | |
9f12600f | 557 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
046b961b AV |
558 | else |
559 | parent = NULL; | |
560 | return parent; | |
561 | } | |
562 | ||
1da177e4 LT |
563 | /* |
564 | * This is dput | |
565 | * | |
566 | * This is complicated by the fact that we do not want to put | |
567 | * dentries that are no longer on any hash chain on the unused | |
568 | * list: we'd much rather just get rid of them immediately. | |
569 | * | |
570 | * However, that implies that we have to traverse the dentry | |
571 | * tree upwards to the parents which might _also_ now be | |
572 | * scheduled for deletion (it may have been only waiting for | |
573 | * its last child to go away). | |
574 | * | |
575 | * This tail recursion is done by hand as we don't want to depend | |
576 | * on the compiler to always get this right (gcc generally doesn't). | |
577 | * Real recursion would eat up our stack space. | |
578 | */ | |
579 | ||
580 | /* | |
581 | * dput - release a dentry | |
582 | * @dentry: dentry to release | |
583 | * | |
584 | * Release a dentry. This will drop the usage count and if appropriate | |
585 | * call the dentry unlink method as well as removing it from the queues and | |
586 | * releasing its resources. If the parent dentries were scheduled for release | |
587 | * they too may now get deleted. | |
1da177e4 | 588 | */ |
1da177e4 LT |
589 | void dput(struct dentry *dentry) |
590 | { | |
8aab6a27 | 591 | if (unlikely(!dentry)) |
1da177e4 LT |
592 | return; |
593 | ||
594 | repeat: | |
98474236 | 595 | if (lockref_put_or_lock(&dentry->d_lockref)) |
1da177e4 | 596 | return; |
1da177e4 | 597 | |
8aab6a27 LT |
598 | /* Unreachable? Get rid of it */ |
599 | if (unlikely(d_unhashed(dentry))) | |
600 | goto kill_it; | |
601 | ||
602 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
1da177e4 | 603 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 604 | goto kill_it; |
1da177e4 | 605 | } |
265ac902 | 606 | |
358eec18 LT |
607 | if (!(dentry->d_flags & DCACHE_REFERENCED)) |
608 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 609 | dentry_lru_add(dentry); |
265ac902 | 610 | |
98474236 | 611 | dentry->d_lockref.count--; |
61f3dee4 | 612 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
613 | return; |
614 | ||
d52b9086 | 615 | kill_it: |
8cbf74da | 616 | dentry = dentry_kill(dentry); |
d52b9086 MS |
617 | if (dentry) |
618 | goto repeat; | |
1da177e4 | 619 | } |
ec4f8605 | 620 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
621 | |
622 | /** | |
623 | * d_invalidate - invalidate a dentry | |
624 | * @dentry: dentry to invalidate | |
625 | * | |
626 | * Try to invalidate the dentry if it turns out to be | |
627 | * possible. If there are other dentries that can be | |
628 | * reached through this one we can't delete it and we | |
629 | * return -EBUSY. On success we return 0. | |
630 | * | |
631 | * no dcache lock. | |
632 | */ | |
633 | ||
634 | int d_invalidate(struct dentry * dentry) | |
635 | { | |
636 | /* | |
637 | * If it's already been dropped, return OK. | |
638 | */ | |
da502956 | 639 | spin_lock(&dentry->d_lock); |
1da177e4 | 640 | if (d_unhashed(dentry)) { |
da502956 | 641 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
642 | return 0; |
643 | } | |
644 | /* | |
645 | * Check whether to do a partial shrink_dcache | |
646 | * to get rid of unused child entries. | |
647 | */ | |
648 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 649 | spin_unlock(&dentry->d_lock); |
1da177e4 | 650 | shrink_dcache_parent(dentry); |
da502956 | 651 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
652 | } |
653 | ||
654 | /* | |
655 | * Somebody else still using it? | |
656 | * | |
657 | * If it's a directory, we can't drop it | |
658 | * for fear of somebody re-populating it | |
659 | * with children (even though dropping it | |
660 | * would make it unreachable from the root, | |
661 | * we might still populate it if it was a | |
662 | * working directory or similar). | |
50e69630 AV |
663 | * We also need to leave mountpoints alone, |
664 | * directory or not. | |
1da177e4 | 665 | */ |
98474236 | 666 | if (dentry->d_lockref.count > 1 && dentry->d_inode) { |
50e69630 | 667 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { |
1da177e4 | 668 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
669 | return -EBUSY; |
670 | } | |
671 | } | |
672 | ||
673 | __d_drop(dentry); | |
674 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
675 | return 0; |
676 | } | |
ec4f8605 | 677 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 678 | |
b5c84bf6 | 679 | /* This must be called with d_lock held */ |
dc0474be | 680 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 681 | { |
98474236 | 682 | dentry->d_lockref.count++; |
23044507 NP |
683 | } |
684 | ||
dc0474be | 685 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 686 | { |
98474236 | 687 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
688 | } |
689 | ||
b7ab39f6 NP |
690 | struct dentry *dget_parent(struct dentry *dentry) |
691 | { | |
df3d0bbc | 692 | int gotref; |
b7ab39f6 NP |
693 | struct dentry *ret; |
694 | ||
df3d0bbc WL |
695 | /* |
696 | * Do optimistic parent lookup without any | |
697 | * locking. | |
698 | */ | |
699 | rcu_read_lock(); | |
700 | ret = ACCESS_ONCE(dentry->d_parent); | |
701 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
702 | rcu_read_unlock(); | |
703 | if (likely(gotref)) { | |
704 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
705 | return ret; | |
706 | dput(ret); | |
707 | } | |
708 | ||
b7ab39f6 | 709 | repeat: |
a734eb45 NP |
710 | /* |
711 | * Don't need rcu_dereference because we re-check it was correct under | |
712 | * the lock. | |
713 | */ | |
714 | rcu_read_lock(); | |
b7ab39f6 | 715 | ret = dentry->d_parent; |
a734eb45 NP |
716 | spin_lock(&ret->d_lock); |
717 | if (unlikely(ret != dentry->d_parent)) { | |
718 | spin_unlock(&ret->d_lock); | |
719 | rcu_read_unlock(); | |
b7ab39f6 NP |
720 | goto repeat; |
721 | } | |
a734eb45 | 722 | rcu_read_unlock(); |
98474236 WL |
723 | BUG_ON(!ret->d_lockref.count); |
724 | ret->d_lockref.count++; | |
b7ab39f6 | 725 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
726 | return ret; |
727 | } | |
728 | EXPORT_SYMBOL(dget_parent); | |
729 | ||
1da177e4 LT |
730 | /** |
731 | * d_find_alias - grab a hashed alias of inode | |
732 | * @inode: inode in question | |
1da177e4 LT |
733 | * |
734 | * If inode has a hashed alias, or is a directory and has any alias, | |
735 | * acquire the reference to alias and return it. Otherwise return NULL. | |
736 | * Notice that if inode is a directory there can be only one alias and | |
737 | * it can be unhashed only if it has no children, or if it is the root | |
738 | * of a filesystem. | |
739 | * | |
21c0d8fd | 740 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
52ed46f0 | 741 | * any other hashed alias over that one. |
1da177e4 | 742 | */ |
52ed46f0 | 743 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 744 | { |
da502956 | 745 | struct dentry *alias, *discon_alias; |
1da177e4 | 746 | |
da502956 NP |
747 | again: |
748 | discon_alias = NULL; | |
b67bfe0d | 749 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
da502956 | 750 | spin_lock(&alias->d_lock); |
1da177e4 | 751 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 752 | if (IS_ROOT(alias) && |
da502956 | 753 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 754 | discon_alias = alias; |
52ed46f0 | 755 | } else { |
dc0474be | 756 | __dget_dlock(alias); |
da502956 NP |
757 | spin_unlock(&alias->d_lock); |
758 | return alias; | |
759 | } | |
760 | } | |
761 | spin_unlock(&alias->d_lock); | |
762 | } | |
763 | if (discon_alias) { | |
764 | alias = discon_alias; | |
765 | spin_lock(&alias->d_lock); | |
766 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
8d80d7da BF |
767 | __dget_dlock(alias); |
768 | spin_unlock(&alias->d_lock); | |
769 | return alias; | |
1da177e4 | 770 | } |
da502956 NP |
771 | spin_unlock(&alias->d_lock); |
772 | goto again; | |
1da177e4 | 773 | } |
da502956 | 774 | return NULL; |
1da177e4 LT |
775 | } |
776 | ||
da502956 | 777 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 778 | { |
214fda1f DH |
779 | struct dentry *de = NULL; |
780 | ||
b3d9b7a3 | 781 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 782 | spin_lock(&inode->i_lock); |
52ed46f0 | 783 | de = __d_find_alias(inode); |
873feea0 | 784 | spin_unlock(&inode->i_lock); |
214fda1f | 785 | } |
1da177e4 LT |
786 | return de; |
787 | } | |
ec4f8605 | 788 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
789 | |
790 | /* | |
791 | * Try to kill dentries associated with this inode. | |
792 | * WARNING: you must own a reference to inode. | |
793 | */ | |
794 | void d_prune_aliases(struct inode *inode) | |
795 | { | |
0cdca3f9 | 796 | struct dentry *dentry; |
1da177e4 | 797 | restart: |
873feea0 | 798 | spin_lock(&inode->i_lock); |
b67bfe0d | 799 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 800 | spin_lock(&dentry->d_lock); |
98474236 | 801 | if (!dentry->d_lockref.count) { |
590fb51f YZ |
802 | /* |
803 | * inform the fs via d_prune that this dentry | |
804 | * is about to be unhashed and destroyed. | |
805 | */ | |
806 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && | |
807 | !d_unhashed(dentry)) | |
808 | dentry->d_op->d_prune(dentry); | |
809 | ||
dc0474be | 810 | __dget_dlock(dentry); |
1da177e4 LT |
811 | __d_drop(dentry); |
812 | spin_unlock(&dentry->d_lock); | |
873feea0 | 813 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
814 | dput(dentry); |
815 | goto restart; | |
816 | } | |
817 | spin_unlock(&dentry->d_lock); | |
818 | } | |
873feea0 | 819 | spin_unlock(&inode->i_lock); |
1da177e4 | 820 | } |
ec4f8605 | 821 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 822 | |
3049cfe2 | 823 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 824 | { |
5c47e6d0 | 825 | struct dentry *dentry, *parent; |
da3bbdd4 | 826 | |
60942f2f | 827 | while (!list_empty(list)) { |
ff2fde99 | 828 | struct inode *inode; |
60942f2f | 829 | dentry = list_entry(list->prev, struct dentry, d_lru); |
ec33679d | 830 | spin_lock(&dentry->d_lock); |
046b961b AV |
831 | parent = lock_parent(dentry); |
832 | ||
dd1f6b2e DC |
833 | /* |
834 | * The dispose list is isolated and dentries are not accounted | |
835 | * to the LRU here, so we can simply remove it from the list | |
836 | * here regardless of whether it is referenced or not. | |
837 | */ | |
89dc77bc | 838 | d_shrink_del(dentry); |
dd1f6b2e | 839 | |
1da177e4 LT |
840 | /* |
841 | * We found an inuse dentry which was not removed from | |
dd1f6b2e | 842 | * the LRU because of laziness during lookup. Do not free it. |
1da177e4 | 843 | */ |
41edf278 | 844 | if ((int)dentry->d_lockref.count > 0) { |
da3bbdd4 | 845 | spin_unlock(&dentry->d_lock); |
046b961b AV |
846 | if (parent) |
847 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
848 | continue; |
849 | } | |
77812a1e | 850 | |
64fd72e0 AV |
851 | |
852 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) { | |
853 | bool can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
854 | spin_unlock(&dentry->d_lock); | |
046b961b AV |
855 | if (parent) |
856 | spin_unlock(&parent->d_lock); | |
64fd72e0 AV |
857 | if (can_free) |
858 | dentry_free(dentry); | |
859 | continue; | |
860 | } | |
861 | ||
ff2fde99 AV |
862 | inode = dentry->d_inode; |
863 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
89dc77bc | 864 | d_shrink_add(dentry, list); |
dd1f6b2e | 865 | spin_unlock(&dentry->d_lock); |
046b961b AV |
866 | if (parent) |
867 | spin_unlock(&parent->d_lock); | |
5c47e6d0 | 868 | continue; |
dd1f6b2e | 869 | } |
ff2fde99 | 870 | |
ff2fde99 | 871 | __dentry_kill(dentry); |
046b961b | 872 | |
5c47e6d0 AV |
873 | /* |
874 | * We need to prune ancestors too. This is necessary to prevent | |
875 | * quadratic behavior of shrink_dcache_parent(), but is also | |
876 | * expected to be beneficial in reducing dentry cache | |
877 | * fragmentation. | |
878 | */ | |
879 | dentry = parent; | |
b2b80195 AV |
880 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) { |
881 | parent = lock_parent(dentry); | |
882 | if (dentry->d_lockref.count != 1) { | |
883 | dentry->d_lockref.count--; | |
884 | spin_unlock(&dentry->d_lock); | |
885 | if (parent) | |
886 | spin_unlock(&parent->d_lock); | |
887 | break; | |
888 | } | |
889 | inode = dentry->d_inode; /* can't be NULL */ | |
890 | if (unlikely(!spin_trylock(&inode->i_lock))) { | |
891 | spin_unlock(&dentry->d_lock); | |
892 | if (parent) | |
893 | spin_unlock(&parent->d_lock); | |
894 | cpu_relax(); | |
895 | continue; | |
896 | } | |
897 | __dentry_kill(dentry); | |
898 | dentry = parent; | |
899 | } | |
da3bbdd4 | 900 | } |
3049cfe2 CH |
901 | } |
902 | ||
f6041567 DC |
903 | static enum lru_status |
904 | dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg) | |
905 | { | |
906 | struct list_head *freeable = arg; | |
907 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
908 | ||
909 | ||
910 | /* | |
911 | * we are inverting the lru lock/dentry->d_lock here, | |
912 | * so use a trylock. If we fail to get the lock, just skip | |
913 | * it | |
914 | */ | |
915 | if (!spin_trylock(&dentry->d_lock)) | |
916 | return LRU_SKIP; | |
917 | ||
918 | /* | |
919 | * Referenced dentries are still in use. If they have active | |
920 | * counts, just remove them from the LRU. Otherwise give them | |
921 | * another pass through the LRU. | |
922 | */ | |
923 | if (dentry->d_lockref.count) { | |
89dc77bc | 924 | d_lru_isolate(dentry); |
f6041567 DC |
925 | spin_unlock(&dentry->d_lock); |
926 | return LRU_REMOVED; | |
927 | } | |
928 | ||
929 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
930 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
931 | spin_unlock(&dentry->d_lock); | |
932 | ||
933 | /* | |
934 | * The list move itself will be made by the common LRU code. At | |
935 | * this point, we've dropped the dentry->d_lock but keep the | |
936 | * lru lock. This is safe to do, since every list movement is | |
937 | * protected by the lru lock even if both locks are held. | |
938 | * | |
939 | * This is guaranteed by the fact that all LRU management | |
940 | * functions are intermediated by the LRU API calls like | |
941 | * list_lru_add and list_lru_del. List movement in this file | |
942 | * only ever occur through this functions or through callbacks | |
943 | * like this one, that are called from the LRU API. | |
944 | * | |
945 | * The only exceptions to this are functions like | |
946 | * shrink_dentry_list, and code that first checks for the | |
947 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
948 | * operating only with stack provided lists after they are | |
949 | * properly isolated from the main list. It is thus, always a | |
950 | * local access. | |
951 | */ | |
952 | return LRU_ROTATE; | |
953 | } | |
954 | ||
89dc77bc | 955 | d_lru_shrink_move(dentry, freeable); |
f6041567 DC |
956 | spin_unlock(&dentry->d_lock); |
957 | ||
958 | return LRU_REMOVED; | |
959 | } | |
960 | ||
3049cfe2 | 961 | /** |
b48f03b3 DC |
962 | * prune_dcache_sb - shrink the dcache |
963 | * @sb: superblock | |
f6041567 | 964 | * @nr_to_scan : number of entries to try to free |
9b17c623 | 965 | * @nid: which node to scan for freeable entities |
b48f03b3 | 966 | * |
f6041567 | 967 | * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is |
b48f03b3 DC |
968 | * done when we need more memory an called from the superblock shrinker |
969 | * function. | |
3049cfe2 | 970 | * |
b48f03b3 DC |
971 | * This function may fail to free any resources if all the dentries are in |
972 | * use. | |
3049cfe2 | 973 | */ |
9b17c623 DC |
974 | long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan, |
975 | int nid) | |
3049cfe2 | 976 | { |
f6041567 DC |
977 | LIST_HEAD(dispose); |
978 | long freed; | |
3049cfe2 | 979 | |
9b17c623 DC |
980 | freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate, |
981 | &dispose, &nr_to_scan); | |
f6041567 | 982 | shrink_dentry_list(&dispose); |
0a234c6d | 983 | return freed; |
da3bbdd4 | 984 | } |
23044507 | 985 | |
4e717f5c GC |
986 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
987 | spinlock_t *lru_lock, void *arg) | |
dd1f6b2e | 988 | { |
4e717f5c GC |
989 | struct list_head *freeable = arg; |
990 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 991 | |
4e717f5c GC |
992 | /* |
993 | * we are inverting the lru lock/dentry->d_lock here, | |
994 | * so use a trylock. If we fail to get the lock, just skip | |
995 | * it | |
996 | */ | |
997 | if (!spin_trylock(&dentry->d_lock)) | |
998 | return LRU_SKIP; | |
999 | ||
89dc77bc | 1000 | d_lru_shrink_move(dentry, freeable); |
4e717f5c | 1001 | spin_unlock(&dentry->d_lock); |
ec33679d | 1002 | |
4e717f5c | 1003 | return LRU_REMOVED; |
da3bbdd4 KM |
1004 | } |
1005 | ||
4e717f5c | 1006 | |
1da177e4 LT |
1007 | /** |
1008 | * shrink_dcache_sb - shrink dcache for a superblock | |
1009 | * @sb: superblock | |
1010 | * | |
3049cfe2 CH |
1011 | * Shrink the dcache for the specified super block. This is used to free |
1012 | * the dcache before unmounting a file system. | |
1da177e4 | 1013 | */ |
3049cfe2 | 1014 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1015 | { |
4e717f5c GC |
1016 | long freed; |
1017 | ||
1018 | do { | |
1019 | LIST_HEAD(dispose); | |
1020 | ||
1021 | freed = list_lru_walk(&sb->s_dentry_lru, | |
1022 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); | |
3049cfe2 | 1023 | |
4e717f5c GC |
1024 | this_cpu_sub(nr_dentry_unused, freed); |
1025 | shrink_dentry_list(&dispose); | |
1026 | } while (freed > 0); | |
1da177e4 | 1027 | } |
ec4f8605 | 1028 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1029 | |
db14fc3a MS |
1030 | /** |
1031 | * enum d_walk_ret - action to talke during tree walk | |
1032 | * @D_WALK_CONTINUE: contrinue walk | |
1033 | * @D_WALK_QUIT: quit walk | |
1034 | * @D_WALK_NORETRY: quit when retry is needed | |
1035 | * @D_WALK_SKIP: skip this dentry and its children | |
1036 | */ | |
1037 | enum d_walk_ret { | |
1038 | D_WALK_CONTINUE, | |
1039 | D_WALK_QUIT, | |
1040 | D_WALK_NORETRY, | |
1041 | D_WALK_SKIP, | |
1042 | }; | |
c826cb7d | 1043 | |
1da177e4 | 1044 | /** |
db14fc3a MS |
1045 | * d_walk - walk the dentry tree |
1046 | * @parent: start of walk | |
1047 | * @data: data passed to @enter() and @finish() | |
1048 | * @enter: callback when first entering the dentry | |
1049 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1050 | * |
db14fc3a | 1051 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1052 | */ |
db14fc3a MS |
1053 | static void d_walk(struct dentry *parent, void *data, |
1054 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1055 | void (*finish)(void *)) | |
1da177e4 | 1056 | { |
949854d0 | 1057 | struct dentry *this_parent; |
1da177e4 | 1058 | struct list_head *next; |
48f5ec21 | 1059 | unsigned seq = 0; |
db14fc3a MS |
1060 | enum d_walk_ret ret; |
1061 | bool retry = true; | |
949854d0 | 1062 | |
58db63d0 | 1063 | again: |
48f5ec21 | 1064 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1065 | this_parent = parent; |
2fd6b7f5 | 1066 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1067 | |
1068 | ret = enter(data, this_parent); | |
1069 | switch (ret) { | |
1070 | case D_WALK_CONTINUE: | |
1071 | break; | |
1072 | case D_WALK_QUIT: | |
1073 | case D_WALK_SKIP: | |
1074 | goto out_unlock; | |
1075 | case D_WALK_NORETRY: | |
1076 | retry = false; | |
1077 | break; | |
1078 | } | |
1da177e4 LT |
1079 | repeat: |
1080 | next = this_parent->d_subdirs.next; | |
1081 | resume: | |
1082 | while (next != &this_parent->d_subdirs) { | |
1083 | struct list_head *tmp = next; | |
5160ee6f | 1084 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1085 | next = tmp->next; |
2fd6b7f5 NP |
1086 | |
1087 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
db14fc3a MS |
1088 | |
1089 | ret = enter(data, dentry); | |
1090 | switch (ret) { | |
1091 | case D_WALK_CONTINUE: | |
1092 | break; | |
1093 | case D_WALK_QUIT: | |
2fd6b7f5 | 1094 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1095 | goto out_unlock; |
1096 | case D_WALK_NORETRY: | |
1097 | retry = false; | |
1098 | break; | |
1099 | case D_WALK_SKIP: | |
1100 | spin_unlock(&dentry->d_lock); | |
1101 | continue; | |
2fd6b7f5 | 1102 | } |
db14fc3a | 1103 | |
1da177e4 | 1104 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1105 | spin_unlock(&this_parent->d_lock); |
1106 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1107 | this_parent = dentry; |
2fd6b7f5 | 1108 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1109 | goto repeat; |
1110 | } | |
2fd6b7f5 | 1111 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1112 | } |
1113 | /* | |
1114 | * All done at this level ... ascend and resume the search. | |
1115 | */ | |
1116 | if (this_parent != parent) { | |
c826cb7d | 1117 | struct dentry *child = this_parent; |
31dec132 AV |
1118 | this_parent = child->d_parent; |
1119 | ||
1120 | rcu_read_lock(); | |
1121 | spin_unlock(&child->d_lock); | |
1122 | spin_lock(&this_parent->d_lock); | |
1123 | ||
1124 | /* | |
1125 | * might go back up the wrong parent if we have had a rename | |
1126 | * or deletion | |
1127 | */ | |
1128 | if (this_parent != child->d_parent || | |
1129 | (child->d_flags & DCACHE_DENTRY_KILLED) || | |
1130 | need_seqretry(&rename_lock, seq)) { | |
1131 | spin_unlock(&this_parent->d_lock); | |
1132 | rcu_read_unlock(); | |
949854d0 | 1133 | goto rename_retry; |
31dec132 AV |
1134 | } |
1135 | rcu_read_unlock(); | |
949854d0 | 1136 | next = child->d_u.d_child.next; |
1da177e4 LT |
1137 | goto resume; |
1138 | } | |
48f5ec21 | 1139 | if (need_seqretry(&rename_lock, seq)) { |
db14fc3a | 1140 | spin_unlock(&this_parent->d_lock); |
949854d0 | 1141 | goto rename_retry; |
db14fc3a MS |
1142 | } |
1143 | if (finish) | |
1144 | finish(data); | |
1145 | ||
1146 | out_unlock: | |
1147 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1148 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1149 | return; |
58db63d0 NP |
1150 | |
1151 | rename_retry: | |
db14fc3a MS |
1152 | if (!retry) |
1153 | return; | |
48f5ec21 | 1154 | seq = 1; |
58db63d0 | 1155 | goto again; |
1da177e4 | 1156 | } |
db14fc3a MS |
1157 | |
1158 | /* | |
1159 | * Search for at least 1 mount point in the dentry's subdirs. | |
1160 | * We descend to the next level whenever the d_subdirs | |
1161 | * list is non-empty and continue searching. | |
1162 | */ | |
1163 | ||
db14fc3a MS |
1164 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) |
1165 | { | |
1166 | int *ret = data; | |
1167 | if (d_mountpoint(dentry)) { | |
1168 | *ret = 1; | |
1169 | return D_WALK_QUIT; | |
1170 | } | |
1171 | return D_WALK_CONTINUE; | |
1172 | } | |
1173 | ||
69c88dc7 RD |
1174 | /** |
1175 | * have_submounts - check for mounts over a dentry | |
1176 | * @parent: dentry to check. | |
1177 | * | |
1178 | * Return true if the parent or its subdirectories contain | |
1179 | * a mount point | |
1180 | */ | |
db14fc3a MS |
1181 | int have_submounts(struct dentry *parent) |
1182 | { | |
1183 | int ret = 0; | |
1184 | ||
1185 | d_walk(parent, &ret, check_mount, NULL); | |
1186 | ||
1187 | return ret; | |
1188 | } | |
ec4f8605 | 1189 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1190 | |
eed81007 MS |
1191 | /* |
1192 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1193 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1194 | * subtree can become unreachable). | |
1195 | * | |
1196 | * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For | |
1197 | * this reason take rename_lock and d_lock on dentry and ancestors. | |
1198 | */ | |
1199 | int d_set_mounted(struct dentry *dentry) | |
1200 | { | |
1201 | struct dentry *p; | |
1202 | int ret = -ENOENT; | |
1203 | write_seqlock(&rename_lock); | |
1204 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1205 | /* Need exclusion wrt. check_submounts_and_drop() */ | |
1206 | spin_lock(&p->d_lock); | |
1207 | if (unlikely(d_unhashed(p))) { | |
1208 | spin_unlock(&p->d_lock); | |
1209 | goto out; | |
1210 | } | |
1211 | spin_unlock(&p->d_lock); | |
1212 | } | |
1213 | spin_lock(&dentry->d_lock); | |
1214 | if (!d_unlinked(dentry)) { | |
1215 | dentry->d_flags |= DCACHE_MOUNTED; | |
1216 | ret = 0; | |
1217 | } | |
1218 | spin_unlock(&dentry->d_lock); | |
1219 | out: | |
1220 | write_sequnlock(&rename_lock); | |
1221 | return ret; | |
1222 | } | |
1223 | ||
1da177e4 | 1224 | /* |
fd517909 | 1225 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1226 | * and move any unused dentries to the end of the unused |
1227 | * list for prune_dcache(). We descend to the next level | |
1228 | * whenever the d_subdirs list is non-empty and continue | |
1229 | * searching. | |
1230 | * | |
1231 | * It returns zero iff there are no unused children, | |
1232 | * otherwise it returns the number of children moved to | |
1233 | * the end of the unused list. This may not be the total | |
1234 | * number of unused children, because select_parent can | |
1235 | * drop the lock and return early due to latency | |
1236 | * constraints. | |
1237 | */ | |
1da177e4 | 1238 | |
db14fc3a MS |
1239 | struct select_data { |
1240 | struct dentry *start; | |
1241 | struct list_head dispose; | |
1242 | int found; | |
1243 | }; | |
23044507 | 1244 | |
db14fc3a MS |
1245 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1246 | { | |
1247 | struct select_data *data = _data; | |
1248 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1249 | |
db14fc3a MS |
1250 | if (data->start == dentry) |
1251 | goto out; | |
2fd6b7f5 | 1252 | |
fe91522a | 1253 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1254 | data->found++; |
fe91522a AV |
1255 | } else { |
1256 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1257 | d_lru_del(dentry); | |
1258 | if (!dentry->d_lockref.count) { | |
1259 | d_shrink_add(dentry, &data->dispose); | |
1260 | data->found++; | |
1261 | } | |
1da177e4 | 1262 | } |
db14fc3a MS |
1263 | /* |
1264 | * We can return to the caller if we have found some (this | |
1265 | * ensures forward progress). We'll be coming back to find | |
1266 | * the rest. | |
1267 | */ | |
fe91522a AV |
1268 | if (!list_empty(&data->dispose)) |
1269 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1270 | out: |
db14fc3a | 1271 | return ret; |
1da177e4 LT |
1272 | } |
1273 | ||
1274 | /** | |
1275 | * shrink_dcache_parent - prune dcache | |
1276 | * @parent: parent of entries to prune | |
1277 | * | |
1278 | * Prune the dcache to remove unused children of the parent dentry. | |
1279 | */ | |
db14fc3a | 1280 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1281 | { |
db14fc3a MS |
1282 | for (;;) { |
1283 | struct select_data data; | |
1da177e4 | 1284 | |
db14fc3a MS |
1285 | INIT_LIST_HEAD(&data.dispose); |
1286 | data.start = parent; | |
1287 | data.found = 0; | |
1288 | ||
1289 | d_walk(parent, &data, select_collect, NULL); | |
1290 | if (!data.found) | |
1291 | break; | |
1292 | ||
1293 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1294 | cond_resched(); |
1295 | } | |
1da177e4 | 1296 | } |
ec4f8605 | 1297 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1298 | |
9c8c10e2 | 1299 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1300 | { |
9c8c10e2 AV |
1301 | /* it has busy descendents; complain about those instead */ |
1302 | if (!list_empty(&dentry->d_subdirs)) | |
1303 | return D_WALK_CONTINUE; | |
42c32608 | 1304 | |
9c8c10e2 AV |
1305 | /* root with refcount 1 is fine */ |
1306 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1307 | return D_WALK_CONTINUE; | |
1308 | ||
1309 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1310 | " still in use (%d) [unmount of %s %s]\n", | |
42c32608 AV |
1311 | dentry, |
1312 | dentry->d_inode ? | |
1313 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1314 | dentry, |
42c32608 AV |
1315 | dentry->d_lockref.count, |
1316 | dentry->d_sb->s_type->name, | |
1317 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1318 | WARN_ON(1); |
1319 | return D_WALK_CONTINUE; | |
1320 | } | |
1321 | ||
1322 | static void do_one_tree(struct dentry *dentry) | |
1323 | { | |
1324 | shrink_dcache_parent(dentry); | |
1325 | d_walk(dentry, dentry, umount_check, NULL); | |
1326 | d_drop(dentry); | |
1327 | dput(dentry); | |
42c32608 AV |
1328 | } |
1329 | ||
1330 | /* | |
1331 | * destroy the dentries attached to a superblock on unmounting | |
1332 | */ | |
1333 | void shrink_dcache_for_umount(struct super_block *sb) | |
1334 | { | |
1335 | struct dentry *dentry; | |
1336 | ||
9c8c10e2 | 1337 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1338 | |
1339 | dentry = sb->s_root; | |
1340 | sb->s_root = NULL; | |
9c8c10e2 | 1341 | do_one_tree(dentry); |
42c32608 AV |
1342 | |
1343 | while (!hlist_bl_empty(&sb->s_anon)) { | |
9c8c10e2 AV |
1344 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash)); |
1345 | do_one_tree(dentry); | |
42c32608 AV |
1346 | } |
1347 | } | |
1348 | ||
848ac114 MS |
1349 | static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry) |
1350 | { | |
1351 | struct select_data *data = _data; | |
1352 | ||
1353 | if (d_mountpoint(dentry)) { | |
1354 | data->found = -EBUSY; | |
1355 | return D_WALK_QUIT; | |
1356 | } | |
1357 | ||
1358 | return select_collect(_data, dentry); | |
1359 | } | |
1360 | ||
1361 | static void check_and_drop(void *_data) | |
1362 | { | |
1363 | struct select_data *data = _data; | |
1364 | ||
1365 | if (d_mountpoint(data->start)) | |
1366 | data->found = -EBUSY; | |
1367 | if (!data->found) | |
1368 | __d_drop(data->start); | |
1369 | } | |
1370 | ||
1371 | /** | |
1372 | * check_submounts_and_drop - prune dcache, check for submounts and drop | |
1373 | * | |
1374 | * All done as a single atomic operation relative to has_unlinked_ancestor(). | |
1375 | * Returns 0 if successfully unhashed @parent. If there were submounts then | |
1376 | * return -EBUSY. | |
1377 | * | |
1378 | * @dentry: dentry to prune and drop | |
1379 | */ | |
1380 | int check_submounts_and_drop(struct dentry *dentry) | |
1381 | { | |
1382 | int ret = 0; | |
1383 | ||
1384 | /* Negative dentries can be dropped without further checks */ | |
1385 | if (!dentry->d_inode) { | |
1386 | d_drop(dentry); | |
1387 | goto out; | |
1388 | } | |
1389 | ||
1390 | for (;;) { | |
1391 | struct select_data data; | |
1392 | ||
1393 | INIT_LIST_HEAD(&data.dispose); | |
1394 | data.start = dentry; | |
1395 | data.found = 0; | |
1396 | ||
1397 | d_walk(dentry, &data, check_and_collect, check_and_drop); | |
1398 | ret = data.found; | |
1399 | ||
1400 | if (!list_empty(&data.dispose)) | |
1401 | shrink_dentry_list(&data.dispose); | |
1402 | ||
1403 | if (ret <= 0) | |
1404 | break; | |
1405 | ||
1406 | cond_resched(); | |
1407 | } | |
1408 | ||
1409 | out: | |
1410 | return ret; | |
1411 | } | |
1412 | EXPORT_SYMBOL(check_submounts_and_drop); | |
1413 | ||
1da177e4 | 1414 | /** |
a4464dbc AV |
1415 | * __d_alloc - allocate a dcache entry |
1416 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1417 | * @name: qstr of the name |
1418 | * | |
1419 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1420 | * available. On a success the dentry is returned. The name passed in is | |
1421 | * copied and the copy passed in may be reused after this call. | |
1422 | */ | |
1423 | ||
a4464dbc | 1424 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1425 | { |
1426 | struct dentry *dentry; | |
1427 | char *dname; | |
1428 | ||
e12ba74d | 1429 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1430 | if (!dentry) |
1431 | return NULL; | |
1432 | ||
6326c71f LT |
1433 | /* |
1434 | * We guarantee that the inline name is always NUL-terminated. | |
1435 | * This way the memcpy() done by the name switching in rename | |
1436 | * will still always have a NUL at the end, even if we might | |
1437 | * be overwriting an internal NUL character | |
1438 | */ | |
1439 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 LT |
1440 | if (name->len > DNAME_INLINE_LEN-1) { |
1441 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1442 | if (!dname) { | |
1443 | kmem_cache_free(dentry_cache, dentry); | |
1444 | return NULL; | |
1445 | } | |
1446 | } else { | |
1447 | dname = dentry->d_iname; | |
1448 | } | |
1da177e4 LT |
1449 | |
1450 | dentry->d_name.len = name->len; | |
1451 | dentry->d_name.hash = name->hash; | |
1452 | memcpy(dname, name->name, name->len); | |
1453 | dname[name->len] = 0; | |
1454 | ||
6326c71f LT |
1455 | /* Make sure we always see the terminating NUL character */ |
1456 | smp_wmb(); | |
1457 | dentry->d_name.name = dname; | |
1458 | ||
98474236 | 1459 | dentry->d_lockref.count = 1; |
dea3667b | 1460 | dentry->d_flags = 0; |
1da177e4 | 1461 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1462 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1463 | dentry->d_inode = NULL; |
a4464dbc AV |
1464 | dentry->d_parent = dentry; |
1465 | dentry->d_sb = sb; | |
1da177e4 LT |
1466 | dentry->d_op = NULL; |
1467 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1468 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1469 | INIT_LIST_HEAD(&dentry->d_lru); |
1470 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
b3d9b7a3 | 1471 | INIT_HLIST_NODE(&dentry->d_alias); |
2fd6b7f5 | 1472 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1473 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1474 | |
3e880fb5 | 1475 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1476 | |
1da177e4 LT |
1477 | return dentry; |
1478 | } | |
a4464dbc AV |
1479 | |
1480 | /** | |
1481 | * d_alloc - allocate a dcache entry | |
1482 | * @parent: parent of entry to allocate | |
1483 | * @name: qstr of the name | |
1484 | * | |
1485 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1486 | * available. On a success the dentry is returned. The name passed in is | |
1487 | * copied and the copy passed in may be reused after this call. | |
1488 | */ | |
1489 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1490 | { | |
1491 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1492 | if (!dentry) | |
1493 | return NULL; | |
1494 | ||
1495 | spin_lock(&parent->d_lock); | |
1496 | /* | |
1497 | * don't need child lock because it is not subject | |
1498 | * to concurrency here | |
1499 | */ | |
1500 | __dget_dlock(parent); | |
1501 | dentry->d_parent = parent; | |
1502 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1503 | spin_unlock(&parent->d_lock); | |
1504 | ||
1505 | return dentry; | |
1506 | } | |
ec4f8605 | 1507 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1508 | |
e1a24bb0 BF |
1509 | /** |
1510 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1511 | * @sb: the superblock | |
1512 | * @name: qstr of the name | |
1513 | * | |
1514 | * For a filesystem that just pins its dentries in memory and never | |
1515 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1516 | */ | |
4b936885 NP |
1517 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1518 | { | |
e1a24bb0 | 1519 | return __d_alloc(sb, name); |
4b936885 NP |
1520 | } |
1521 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1522 | ||
1da177e4 LT |
1523 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1524 | { | |
1525 | struct qstr q; | |
1526 | ||
1527 | q.name = name; | |
1528 | q.len = strlen(name); | |
1529 | q.hash = full_name_hash(q.name, q.len); | |
1530 | return d_alloc(parent, &q); | |
1531 | } | |
ef26ca97 | 1532 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1533 | |
fb045adb NP |
1534 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1535 | { | |
6f7f7caa LT |
1536 | WARN_ON_ONCE(dentry->d_op); |
1537 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1538 | DCACHE_OP_COMPARE | |
1539 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1540 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1541 | DCACHE_OP_DELETE )); |
1542 | dentry->d_op = op; | |
1543 | if (!op) | |
1544 | return; | |
1545 | if (op->d_hash) | |
1546 | dentry->d_flags |= DCACHE_OP_HASH; | |
1547 | if (op->d_compare) | |
1548 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1549 | if (op->d_revalidate) | |
1550 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1551 | if (op->d_weak_revalidate) |
1552 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1553 | if (op->d_delete) |
1554 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1555 | if (op->d_prune) |
1556 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1557 | |
1558 | } | |
1559 | EXPORT_SYMBOL(d_set_d_op); | |
1560 | ||
b18825a7 DH |
1561 | static unsigned d_flags_for_inode(struct inode *inode) |
1562 | { | |
1563 | unsigned add_flags = DCACHE_FILE_TYPE; | |
1564 | ||
1565 | if (!inode) | |
1566 | return DCACHE_MISS_TYPE; | |
1567 | ||
1568 | if (S_ISDIR(inode->i_mode)) { | |
1569 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1570 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1571 | if (unlikely(!inode->i_op->lookup)) | |
1572 | add_flags = DCACHE_AUTODIR_TYPE; | |
1573 | else | |
1574 | inode->i_opflags |= IOP_LOOKUP; | |
1575 | } | |
1576 | } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1577 | if (unlikely(inode->i_op->follow_link)) | |
1578 | add_flags = DCACHE_SYMLINK_TYPE; | |
1579 | else | |
1580 | inode->i_opflags |= IOP_NOFOLLOW; | |
1581 | } | |
1582 | ||
1583 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1584 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1585 | return add_flags; | |
1586 | } | |
1587 | ||
360da900 OH |
1588 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1589 | { | |
b18825a7 DH |
1590 | unsigned add_flags = d_flags_for_inode(inode); |
1591 | ||
b23fb0a6 | 1592 | spin_lock(&dentry->d_lock); |
22213318 | 1593 | __d_set_type(dentry, add_flags); |
b18825a7 | 1594 | if (inode) |
b3d9b7a3 | 1595 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); |
360da900 | 1596 | dentry->d_inode = inode; |
31e6b01f | 1597 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1598 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1599 | fsnotify_d_instantiate(dentry, inode); |
1600 | } | |
1601 | ||
1da177e4 LT |
1602 | /** |
1603 | * d_instantiate - fill in inode information for a dentry | |
1604 | * @entry: dentry to complete | |
1605 | * @inode: inode to attach to this dentry | |
1606 | * | |
1607 | * Fill in inode information in the entry. | |
1608 | * | |
1609 | * This turns negative dentries into productive full members | |
1610 | * of society. | |
1611 | * | |
1612 | * NOTE! This assumes that the inode count has been incremented | |
1613 | * (or otherwise set) by the caller to indicate that it is now | |
1614 | * in use by the dcache. | |
1615 | */ | |
1616 | ||
1617 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1618 | { | |
b3d9b7a3 | 1619 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
873feea0 NP |
1620 | if (inode) |
1621 | spin_lock(&inode->i_lock); | |
360da900 | 1622 | __d_instantiate(entry, inode); |
873feea0 NP |
1623 | if (inode) |
1624 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1625 | security_d_instantiate(entry, inode); |
1626 | } | |
ec4f8605 | 1627 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1628 | |
1629 | /** | |
1630 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1631 | * @entry: dentry to instantiate | |
1632 | * @inode: inode to attach to this dentry | |
1633 | * | |
1634 | * Fill in inode information in the entry. On success, it returns NULL. | |
1635 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1636 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1637 | * |
1638 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1639 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1640 | * |
1641 | * This also assumes that the inode count has been incremented | |
1642 | * (or otherwise set) by the caller to indicate that it is now | |
1643 | * in use by the dcache. | |
1da177e4 | 1644 | */ |
770bfad8 DH |
1645 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1646 | struct inode *inode) | |
1da177e4 LT |
1647 | { |
1648 | struct dentry *alias; | |
1649 | int len = entry->d_name.len; | |
1650 | const char *name = entry->d_name.name; | |
1651 | unsigned int hash = entry->d_name.hash; | |
1652 | ||
770bfad8 | 1653 | if (!inode) { |
360da900 | 1654 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1655 | return NULL; |
1656 | } | |
1657 | ||
b67bfe0d | 1658 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
9abca360 NP |
1659 | /* |
1660 | * Don't need alias->d_lock here, because aliases with | |
1661 | * d_parent == entry->d_parent are not subject to name or | |
1662 | * parent changes, because the parent inode i_mutex is held. | |
1663 | */ | |
12f8ad4b | 1664 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1665 | continue; |
1666 | if (alias->d_parent != entry->d_parent) | |
1667 | continue; | |
ee983e89 LT |
1668 | if (alias->d_name.len != len) |
1669 | continue; | |
12f8ad4b | 1670 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1671 | continue; |
dc0474be | 1672 | __dget(alias); |
1da177e4 LT |
1673 | return alias; |
1674 | } | |
770bfad8 | 1675 | |
360da900 | 1676 | __d_instantiate(entry, inode); |
1da177e4 LT |
1677 | return NULL; |
1678 | } | |
770bfad8 DH |
1679 | |
1680 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1681 | { | |
1682 | struct dentry *result; | |
1683 | ||
b3d9b7a3 | 1684 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
770bfad8 | 1685 | |
873feea0 NP |
1686 | if (inode) |
1687 | spin_lock(&inode->i_lock); | |
770bfad8 | 1688 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1689 | if (inode) |
1690 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1691 | |
1692 | if (!result) { | |
1693 | security_d_instantiate(entry, inode); | |
1694 | return NULL; | |
1695 | } | |
1696 | ||
1697 | BUG_ON(!d_unhashed(result)); | |
1698 | iput(inode); | |
1699 | return result; | |
1700 | } | |
1701 | ||
1da177e4 LT |
1702 | EXPORT_SYMBOL(d_instantiate_unique); |
1703 | ||
b70a80e7 MS |
1704 | /** |
1705 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1706 | * @entry: dentry to complete | |
1707 | * @inode: inode to attach to this dentry | |
1708 | * | |
1709 | * Fill in inode information in the entry. If a directory alias is found, then | |
1710 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1711 | * guarantees that a directory inode may never have more than one alias. | |
1712 | */ | |
1713 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1714 | { | |
1715 | BUG_ON(!hlist_unhashed(&entry->d_alias)); | |
1716 | ||
1717 | spin_lock(&inode->i_lock); | |
1718 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1719 | spin_unlock(&inode->i_lock); | |
1720 | iput(inode); | |
1721 | return -EBUSY; | |
1722 | } | |
1723 | __d_instantiate(entry, inode); | |
1724 | spin_unlock(&inode->i_lock); | |
1725 | security_d_instantiate(entry, inode); | |
1726 | ||
1727 | return 0; | |
1728 | } | |
1729 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1730 | ||
adc0e91a AV |
1731 | struct dentry *d_make_root(struct inode *root_inode) |
1732 | { | |
1733 | struct dentry *res = NULL; | |
1734 | ||
1735 | if (root_inode) { | |
26fe5750 | 1736 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1737 | |
1738 | res = __d_alloc(root_inode->i_sb, &name); | |
1739 | if (res) | |
1740 | d_instantiate(res, root_inode); | |
1741 | else | |
1742 | iput(root_inode); | |
1743 | } | |
1744 | return res; | |
1745 | } | |
1746 | EXPORT_SYMBOL(d_make_root); | |
1747 | ||
d891eedb BF |
1748 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1749 | { | |
1750 | struct dentry *alias; | |
1751 | ||
b3d9b7a3 | 1752 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1753 | return NULL; |
b3d9b7a3 | 1754 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); |
d891eedb BF |
1755 | __dget(alias); |
1756 | return alias; | |
1757 | } | |
1758 | ||
46f72b34 SW |
1759 | /** |
1760 | * d_find_any_alias - find any alias for a given inode | |
1761 | * @inode: inode to find an alias for | |
1762 | * | |
1763 | * If any aliases exist for the given inode, take and return a | |
1764 | * reference for one of them. If no aliases exist, return %NULL. | |
1765 | */ | |
1766 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1767 | { |
1768 | struct dentry *de; | |
1769 | ||
1770 | spin_lock(&inode->i_lock); | |
1771 | de = __d_find_any_alias(inode); | |
1772 | spin_unlock(&inode->i_lock); | |
1773 | return de; | |
1774 | } | |
46f72b34 | 1775 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1776 | |
49c7dd28 | 1777 | static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected) |
4ea3ada2 | 1778 | { |
b911a6bd | 1779 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1780 | struct dentry *tmp; |
1781 | struct dentry *res; | |
b18825a7 | 1782 | unsigned add_flags; |
4ea3ada2 CH |
1783 | |
1784 | if (!inode) | |
44003728 | 1785 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1786 | if (IS_ERR(inode)) |
1787 | return ERR_CAST(inode); | |
1788 | ||
d891eedb | 1789 | res = d_find_any_alias(inode); |
9308a612 CH |
1790 | if (res) |
1791 | goto out_iput; | |
1792 | ||
a4464dbc | 1793 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1794 | if (!tmp) { |
1795 | res = ERR_PTR(-ENOMEM); | |
1796 | goto out_iput; | |
4ea3ada2 | 1797 | } |
b5c84bf6 | 1798 | |
873feea0 | 1799 | spin_lock(&inode->i_lock); |
d891eedb | 1800 | res = __d_find_any_alias(inode); |
9308a612 | 1801 | if (res) { |
873feea0 | 1802 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1803 | dput(tmp); |
1804 | goto out_iput; | |
1805 | } | |
1806 | ||
1807 | /* attach a disconnected dentry */ | |
1a0a397e BF |
1808 | add_flags = d_flags_for_inode(inode); |
1809 | ||
1810 | if (disconnected) | |
1811 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 1812 | |
9308a612 | 1813 | spin_lock(&tmp->d_lock); |
9308a612 | 1814 | tmp->d_inode = inode; |
b18825a7 | 1815 | tmp->d_flags |= add_flags; |
b3d9b7a3 | 1816 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1817 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1818 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1819 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1820 | spin_unlock(&tmp->d_lock); |
873feea0 | 1821 | spin_unlock(&inode->i_lock); |
24ff6663 | 1822 | security_d_instantiate(tmp, inode); |
9308a612 | 1823 | |
9308a612 CH |
1824 | return tmp; |
1825 | ||
1826 | out_iput: | |
24ff6663 JB |
1827 | if (res && !IS_ERR(res)) |
1828 | security_d_instantiate(res, inode); | |
9308a612 CH |
1829 | iput(inode); |
1830 | return res; | |
4ea3ada2 | 1831 | } |
1a0a397e BF |
1832 | |
1833 | /** | |
1834 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
1835 | * @inode: inode to allocate the dentry for | |
1836 | * | |
1837 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1838 | * similar open by handle operations. The returned dentry may be anonymous, | |
1839 | * or may have a full name (if the inode was already in the cache). | |
1840 | * | |
1841 | * When called on a directory inode, we must ensure that the inode only ever | |
1842 | * has one dentry. If a dentry is found, that is returned instead of | |
1843 | * allocating a new one. | |
1844 | * | |
1845 | * On successful return, the reference to the inode has been transferred | |
1846 | * to the dentry. In case of an error the reference on the inode is released. | |
1847 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1848 | * be passed in and the error will be propagated to the return value, | |
1849 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
1850 | */ | |
1851 | struct dentry *d_obtain_alias(struct inode *inode) | |
1852 | { | |
1853 | return __d_obtain_alias(inode, 1); | |
1854 | } | |
adc48720 | 1855 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 1856 | |
1a0a397e BF |
1857 | /** |
1858 | * d_obtain_root - find or allocate a dentry for a given inode | |
1859 | * @inode: inode to allocate the dentry for | |
1860 | * | |
1861 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
1862 | * | |
1863 | * We must ensure that directory inodes only ever have one dentry. If a | |
1864 | * dentry is found, that is returned instead of allocating a new one. | |
1865 | * | |
1866 | * On successful return, the reference to the inode has been transferred | |
1867 | * to the dentry. In case of an error the reference on the inode is | |
1868 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
1869 | * error will be propagate to the return value, with a %NULL @inode | |
1870 | * replaced by ERR_PTR(-ESTALE). | |
1871 | */ | |
1872 | struct dentry *d_obtain_root(struct inode *inode) | |
1873 | { | |
1874 | return __d_obtain_alias(inode, 0); | |
1875 | } | |
1876 | EXPORT_SYMBOL(d_obtain_root); | |
1877 | ||
9403540c BN |
1878 | /** |
1879 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1880 | * @inode: the inode case-insensitive lookup has found | |
1881 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1882 | * @name: the case-exact name to be associated with the returned dentry | |
1883 | * | |
1884 | * This is to avoid filling the dcache with case-insensitive names to the | |
1885 | * same inode, only the actual correct case is stored in the dcache for | |
1886 | * case-insensitive filesystems. | |
1887 | * | |
1888 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1889 | * already exists in in the dcache, use it and return it. | |
1890 | * | |
1891 | * If no entry exists with the exact case name, allocate new dentry with | |
1892 | * the exact case, and return the spliced entry. | |
1893 | */ | |
e45b590b | 1894 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1895 | struct qstr *name) |
1896 | { | |
9403540c BN |
1897 | struct dentry *found; |
1898 | struct dentry *new; | |
1899 | ||
b6520c81 CH |
1900 | /* |
1901 | * First check if a dentry matching the name already exists, | |
1902 | * if not go ahead and create it now. | |
1903 | */ | |
9403540c | 1904 | found = d_hash_and_lookup(dentry->d_parent, name); |
4f522a24 AV |
1905 | if (unlikely(IS_ERR(found))) |
1906 | goto err_out; | |
9403540c BN |
1907 | if (!found) { |
1908 | new = d_alloc(dentry->d_parent, name); | |
1909 | if (!new) { | |
4f522a24 | 1910 | found = ERR_PTR(-ENOMEM); |
9403540c BN |
1911 | goto err_out; |
1912 | } | |
b6520c81 | 1913 | |
9403540c BN |
1914 | found = d_splice_alias(inode, new); |
1915 | if (found) { | |
1916 | dput(new); | |
1917 | return found; | |
1918 | } | |
1919 | return new; | |
1920 | } | |
b6520c81 CH |
1921 | |
1922 | /* | |
1923 | * If a matching dentry exists, and it's not negative use it. | |
1924 | * | |
1925 | * Decrement the reference count to balance the iget() done | |
1926 | * earlier on. | |
1927 | */ | |
9403540c BN |
1928 | if (found->d_inode) { |
1929 | if (unlikely(found->d_inode != inode)) { | |
1930 | /* This can't happen because bad inodes are unhashed. */ | |
1931 | BUG_ON(!is_bad_inode(inode)); | |
1932 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1933 | } | |
9403540c BN |
1934 | iput(inode); |
1935 | return found; | |
1936 | } | |
b6520c81 | 1937 | |
9403540c | 1938 | /* |
9403540c | 1939 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1940 | * already has a dentry. |
9403540c | 1941 | */ |
4513d899 AV |
1942 | new = d_splice_alias(inode, found); |
1943 | if (new) { | |
1944 | dput(found); | |
1945 | found = new; | |
9403540c | 1946 | } |
4513d899 | 1947 | return found; |
9403540c BN |
1948 | |
1949 | err_out: | |
1950 | iput(inode); | |
4f522a24 | 1951 | return found; |
9403540c | 1952 | } |
ec4f8605 | 1953 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1954 | |
12f8ad4b LT |
1955 | /* |
1956 | * Do the slow-case of the dentry name compare. | |
1957 | * | |
1958 | * Unlike the dentry_cmp() function, we need to atomically | |
da53be12 | 1959 | * load the name and length information, so that the |
12f8ad4b LT |
1960 | * filesystem can rely on them, and can use the 'name' and |
1961 | * 'len' information without worrying about walking off the | |
1962 | * end of memory etc. | |
1963 | * | |
1964 | * Thus the read_seqcount_retry() and the "duplicate" info | |
1965 | * in arguments (the low-level filesystem should not look | |
1966 | * at the dentry inode or name contents directly, since | |
1967 | * rename can change them while we're in RCU mode). | |
1968 | */ | |
1969 | enum slow_d_compare { | |
1970 | D_COMP_OK, | |
1971 | D_COMP_NOMATCH, | |
1972 | D_COMP_SEQRETRY, | |
1973 | }; | |
1974 | ||
1975 | static noinline enum slow_d_compare slow_dentry_cmp( | |
1976 | const struct dentry *parent, | |
12f8ad4b LT |
1977 | struct dentry *dentry, |
1978 | unsigned int seq, | |
1979 | const struct qstr *name) | |
1980 | { | |
1981 | int tlen = dentry->d_name.len; | |
1982 | const char *tname = dentry->d_name.name; | |
12f8ad4b LT |
1983 | |
1984 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
1985 | cpu_relax(); | |
1986 | return D_COMP_SEQRETRY; | |
1987 | } | |
da53be12 | 1988 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
12f8ad4b LT |
1989 | return D_COMP_NOMATCH; |
1990 | return D_COMP_OK; | |
1991 | } | |
1992 | ||
31e6b01f NP |
1993 | /** |
1994 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1995 | * @parent: parent dentry | |
1996 | * @name: qstr of name we wish to find | |
1f1e6e52 | 1997 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
1998 | * Returns: dentry, or NULL |
1999 | * | |
2000 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2001 | * resolution (store-free path walking) design described in | |
2002 | * Documentation/filesystems/path-lookup.txt. | |
2003 | * | |
2004 | * This is not to be used outside core vfs. | |
2005 | * | |
2006 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2007 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2008 | * without taking d_lock and checking d_seq sequence count against @seq | |
2009 | * returned here. | |
2010 | * | |
15570086 | 2011 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2012 | * function. |
2013 | * | |
2014 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2015 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2016 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2017 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2018 | * |
2019 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2020 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2021 | */ |
8966be90 LT |
2022 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2023 | const struct qstr *name, | |
da53be12 | 2024 | unsigned *seqp) |
31e6b01f | 2025 | { |
26fe5750 | 2026 | u64 hashlen = name->hash_len; |
31e6b01f | 2027 | const unsigned char *str = name->name; |
26fe5750 | 2028 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 2029 | struct hlist_bl_node *node; |
31e6b01f NP |
2030 | struct dentry *dentry; |
2031 | ||
2032 | /* | |
2033 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2034 | * required to prevent single threaded performance regressions | |
2035 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2036 | * Keep the two functions in sync. | |
2037 | */ | |
2038 | ||
2039 | /* | |
2040 | * The hash list is protected using RCU. | |
2041 | * | |
2042 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2043 | * races with d_move(). | |
2044 | * | |
2045 | * It is possible that concurrent renames can mess up our list | |
2046 | * walk here and result in missing our dentry, resulting in the | |
2047 | * false-negative result. d_lookup() protects against concurrent | |
2048 | * renames using rename_lock seqlock. | |
2049 | * | |
b0a4bb83 | 2050 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2051 | */ |
b07ad996 | 2052 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2053 | unsigned seq; |
31e6b01f | 2054 | |
31e6b01f | 2055 | seqretry: |
12f8ad4b LT |
2056 | /* |
2057 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2058 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2059 | * |
2060 | * The caller must perform a seqcount check in order | |
da53be12 | 2061 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2062 | * |
2063 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2064 | * we don't wait for the sequence count to stabilize if it | |
2065 | * is in the middle of a sequence change. If we do the slow | |
2066 | * dentry compare, we will do seqretries until it is stable, | |
2067 | * and if we end up with a successful lookup, we actually | |
2068 | * want to exit RCU lookup anyway. | |
2069 | */ | |
2070 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2071 | if (dentry->d_parent != parent) |
2072 | continue; | |
2e321806 LT |
2073 | if (d_unhashed(dentry)) |
2074 | continue; | |
12f8ad4b | 2075 | |
830c0f0e | 2076 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
2077 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2078 | continue; | |
da53be12 LT |
2079 | *seqp = seq; |
2080 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
12f8ad4b LT |
2081 | case D_COMP_OK: |
2082 | return dentry; | |
2083 | case D_COMP_NOMATCH: | |
31e6b01f | 2084 | continue; |
12f8ad4b LT |
2085 | default: |
2086 | goto seqretry; | |
2087 | } | |
31e6b01f | 2088 | } |
12f8ad4b | 2089 | |
26fe5750 | 2090 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 2091 | continue; |
da53be12 | 2092 | *seqp = seq; |
26fe5750 | 2093 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 2094 | return dentry; |
31e6b01f NP |
2095 | } |
2096 | return NULL; | |
2097 | } | |
2098 | ||
1da177e4 LT |
2099 | /** |
2100 | * d_lookup - search for a dentry | |
2101 | * @parent: parent dentry | |
2102 | * @name: qstr of name we wish to find | |
b04f784e | 2103 | * Returns: dentry, or NULL |
1da177e4 | 2104 | * |
b04f784e NP |
2105 | * d_lookup searches the children of the parent dentry for the name in |
2106 | * question. If the dentry is found its reference count is incremented and the | |
2107 | * dentry is returned. The caller must use dput to free the entry when it has | |
2108 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2109 | */ |
da2d8455 | 2110 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2111 | { |
31e6b01f | 2112 | struct dentry *dentry; |
949854d0 | 2113 | unsigned seq; |
1da177e4 LT |
2114 | |
2115 | do { | |
2116 | seq = read_seqbegin(&rename_lock); | |
2117 | dentry = __d_lookup(parent, name); | |
2118 | if (dentry) | |
2119 | break; | |
2120 | } while (read_seqretry(&rename_lock, seq)); | |
2121 | return dentry; | |
2122 | } | |
ec4f8605 | 2123 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2124 | |
31e6b01f | 2125 | /** |
b04f784e NP |
2126 | * __d_lookup - search for a dentry (racy) |
2127 | * @parent: parent dentry | |
2128 | * @name: qstr of name we wish to find | |
2129 | * Returns: dentry, or NULL | |
2130 | * | |
2131 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2132 | * false-negative result due to unrelated rename activity. | |
2133 | * | |
2134 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2135 | * however it must be used carefully, eg. with a following d_lookup in | |
2136 | * the case of failure. | |
2137 | * | |
2138 | * __d_lookup callers must be commented. | |
2139 | */ | |
a713ca2a | 2140 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
2141 | { |
2142 | unsigned int len = name->len; | |
2143 | unsigned int hash = name->hash; | |
2144 | const unsigned char *str = name->name; | |
b07ad996 | 2145 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 2146 | struct hlist_bl_node *node; |
31e6b01f | 2147 | struct dentry *found = NULL; |
665a7583 | 2148 | struct dentry *dentry; |
1da177e4 | 2149 | |
31e6b01f NP |
2150 | /* |
2151 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2152 | * required to prevent single threaded performance regressions | |
2153 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2154 | * Keep the two functions in sync. | |
2155 | */ | |
2156 | ||
b04f784e NP |
2157 | /* |
2158 | * The hash list is protected using RCU. | |
2159 | * | |
2160 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2161 | * with d_move(). | |
2162 | * | |
2163 | * It is possible that concurrent renames can mess up our list | |
2164 | * walk here and result in missing our dentry, resulting in the | |
2165 | * false-negative result. d_lookup() protects against concurrent | |
2166 | * renames using rename_lock seqlock. | |
2167 | * | |
b0a4bb83 | 2168 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2169 | */ |
1da177e4 LT |
2170 | rcu_read_lock(); |
2171 | ||
b07ad996 | 2172 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2173 | |
1da177e4 LT |
2174 | if (dentry->d_name.hash != hash) |
2175 | continue; | |
1da177e4 LT |
2176 | |
2177 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2178 | if (dentry->d_parent != parent) |
2179 | goto next; | |
d0185c08 LT |
2180 | if (d_unhashed(dentry)) |
2181 | goto next; | |
2182 | ||
1da177e4 LT |
2183 | /* |
2184 | * It is safe to compare names since d_move() cannot | |
2185 | * change the qstr (protected by d_lock). | |
2186 | */ | |
fb045adb | 2187 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
2188 | int tlen = dentry->d_name.len; |
2189 | const char *tname = dentry->d_name.name; | |
da53be12 | 2190 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1da177e4 LT |
2191 | goto next; |
2192 | } else { | |
ee983e89 LT |
2193 | if (dentry->d_name.len != len) |
2194 | goto next; | |
12f8ad4b | 2195 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
2196 | goto next; |
2197 | } | |
2198 | ||
98474236 | 2199 | dentry->d_lockref.count++; |
d0185c08 | 2200 | found = dentry; |
1da177e4 LT |
2201 | spin_unlock(&dentry->d_lock); |
2202 | break; | |
2203 | next: | |
2204 | spin_unlock(&dentry->d_lock); | |
2205 | } | |
2206 | rcu_read_unlock(); | |
2207 | ||
2208 | return found; | |
2209 | } | |
2210 | ||
3e7e241f EB |
2211 | /** |
2212 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2213 | * @dir: Directory to search in | |
2214 | * @name: qstr of name we wish to find | |
2215 | * | |
4f522a24 | 2216 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2217 | */ |
2218 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2219 | { | |
3e7e241f EB |
2220 | /* |
2221 | * Check for a fs-specific hash function. Note that we must | |
2222 | * calculate the standard hash first, as the d_op->d_hash() | |
2223 | * routine may choose to leave the hash value unchanged. | |
2224 | */ | |
2225 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2226 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2227 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2228 | if (unlikely(err < 0)) |
2229 | return ERR_PTR(err); | |
3e7e241f | 2230 | } |
4f522a24 | 2231 | return d_lookup(dir, name); |
3e7e241f | 2232 | } |
4f522a24 | 2233 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2234 | |
1da177e4 | 2235 | /** |
786a5e15 | 2236 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2237 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2238 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2239 | * |
2240 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2241 | * This is used by ncpfs in its readdir implementation. | |
2242 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2243 | * |
2244 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2245 | */ |
d3a23e16 | 2246 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2247 | { |
786a5e15 | 2248 | struct dentry *child; |
d3a23e16 | 2249 | |
2fd6b7f5 | 2250 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2251 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2252 | if (dentry == child) { | |
2fd6b7f5 | 2253 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2254 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2255 | spin_unlock(&dentry->d_lock); |
2256 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2257 | return 1; |
2258 | } | |
2259 | } | |
2fd6b7f5 | 2260 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2261 | |
1da177e4 LT |
2262 | return 0; |
2263 | } | |
ec4f8605 | 2264 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2265 | |
2266 | /* | |
2267 | * When a file is deleted, we have two options: | |
2268 | * - turn this dentry into a negative dentry | |
2269 | * - unhash this dentry and free it. | |
2270 | * | |
2271 | * Usually, we want to just turn this into | |
2272 | * a negative dentry, but if anybody else is | |
2273 | * currently using the dentry or the inode | |
2274 | * we can't do that and we fall back on removing | |
2275 | * it from the hash queues and waiting for | |
2276 | * it to be deleted later when it has no users | |
2277 | */ | |
2278 | ||
2279 | /** | |
2280 | * d_delete - delete a dentry | |
2281 | * @dentry: The dentry to delete | |
2282 | * | |
2283 | * Turn the dentry into a negative dentry if possible, otherwise | |
2284 | * remove it from the hash queues so it can be deleted later | |
2285 | */ | |
2286 | ||
2287 | void d_delete(struct dentry * dentry) | |
2288 | { | |
873feea0 | 2289 | struct inode *inode; |
7a91bf7f | 2290 | int isdir = 0; |
1da177e4 LT |
2291 | /* |
2292 | * Are we the only user? | |
2293 | */ | |
357f8e65 | 2294 | again: |
1da177e4 | 2295 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2296 | inode = dentry->d_inode; |
2297 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2298 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2299 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2300 | spin_unlock(&dentry->d_lock); |
2301 | cpu_relax(); | |
2302 | goto again; | |
2303 | } | |
13e3c5e5 | 2304 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2305 | dentry_unlink_inode(dentry); |
7a91bf7f | 2306 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2307 | return; |
2308 | } | |
2309 | ||
2310 | if (!d_unhashed(dentry)) | |
2311 | __d_drop(dentry); | |
2312 | ||
2313 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2314 | |
2315 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2316 | } |
ec4f8605 | 2317 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2318 | |
b07ad996 | 2319 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2320 | { |
ceb5bdc2 | 2321 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2322 | hlist_bl_lock(b); |
dea3667b | 2323 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2324 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2325 | hlist_bl_unlock(b); |
1da177e4 LT |
2326 | } |
2327 | ||
770bfad8 DH |
2328 | static void _d_rehash(struct dentry * entry) |
2329 | { | |
2330 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2331 | } | |
2332 | ||
1da177e4 LT |
2333 | /** |
2334 | * d_rehash - add an entry back to the hash | |
2335 | * @entry: dentry to add to the hash | |
2336 | * | |
2337 | * Adds a dentry to the hash according to its name. | |
2338 | */ | |
2339 | ||
2340 | void d_rehash(struct dentry * entry) | |
2341 | { | |
1da177e4 | 2342 | spin_lock(&entry->d_lock); |
770bfad8 | 2343 | _d_rehash(entry); |
1da177e4 | 2344 | spin_unlock(&entry->d_lock); |
1da177e4 | 2345 | } |
ec4f8605 | 2346 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2347 | |
fb2d5b86 NP |
2348 | /** |
2349 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2350 | * @dentry: dentry to be updated | |
2351 | * @name: new name | |
2352 | * | |
2353 | * Update a case insensitive dentry with new case of name. | |
2354 | * | |
2355 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2356 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2357 | * lengths). | |
2358 | * | |
2359 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2360 | * keep renames and concurrent inserts, and readdir(2) away). | |
2361 | */ | |
2362 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2363 | { | |
7ebfa57f | 2364 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2365 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2366 | ||
fb2d5b86 | 2367 | spin_lock(&dentry->d_lock); |
31e6b01f | 2368 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2369 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2370 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2371 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2372 | } |
2373 | EXPORT_SYMBOL(dentry_update_name_case); | |
2374 | ||
1da177e4 LT |
2375 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2376 | { | |
2377 | if (dname_external(target)) { | |
2378 | if (dname_external(dentry)) { | |
2379 | /* | |
2380 | * Both external: swap the pointers | |
2381 | */ | |
9a8d5bb4 | 2382 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2383 | } else { |
2384 | /* | |
2385 | * dentry:internal, target:external. Steal target's | |
2386 | * storage and make target internal. | |
2387 | */ | |
321bcf92 BF |
2388 | memcpy(target->d_iname, dentry->d_name.name, |
2389 | dentry->d_name.len + 1); | |
1da177e4 LT |
2390 | dentry->d_name.name = target->d_name.name; |
2391 | target->d_name.name = target->d_iname; | |
2392 | } | |
2393 | } else { | |
2394 | if (dname_external(dentry)) { | |
2395 | /* | |
2396 | * dentry:external, target:internal. Give dentry's | |
2397 | * storage to target and make dentry internal | |
2398 | */ | |
2399 | memcpy(dentry->d_iname, target->d_name.name, | |
2400 | target->d_name.len + 1); | |
2401 | target->d_name.name = dentry->d_name.name; | |
2402 | dentry->d_name.name = dentry->d_iname; | |
2403 | } else { | |
2404 | /* | |
da1ce067 | 2405 | * Both are internal. |
1da177e4 | 2406 | */ |
da1ce067 MS |
2407 | unsigned int i; |
2408 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
2409 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { | |
2410 | swap(((long *) &dentry->d_iname)[i], | |
2411 | ((long *) &target->d_iname)[i]); | |
2412 | } | |
1da177e4 LT |
2413 | } |
2414 | } | |
9a8d5bb4 | 2415 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2416 | } |
2417 | ||
2fd6b7f5 NP |
2418 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2419 | { | |
2420 | /* | |
2421 | * XXXX: do we really need to take target->d_lock? | |
2422 | */ | |
2423 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2424 | spin_lock(&target->d_parent->d_lock); | |
2425 | else { | |
2426 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2427 | spin_lock(&dentry->d_parent->d_lock); | |
2428 | spin_lock_nested(&target->d_parent->d_lock, | |
2429 | DENTRY_D_LOCK_NESTED); | |
2430 | } else { | |
2431 | spin_lock(&target->d_parent->d_lock); | |
2432 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2433 | DENTRY_D_LOCK_NESTED); | |
2434 | } | |
2435 | } | |
2436 | if (target < dentry) { | |
2437 | spin_lock_nested(&target->d_lock, 2); | |
2438 | spin_lock_nested(&dentry->d_lock, 3); | |
2439 | } else { | |
2440 | spin_lock_nested(&dentry->d_lock, 2); | |
2441 | spin_lock_nested(&target->d_lock, 3); | |
2442 | } | |
2443 | } | |
2444 | ||
2445 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2446 | struct dentry *target) | |
2447 | { | |
2448 | if (target->d_parent != dentry->d_parent) | |
2449 | spin_unlock(&dentry->d_parent->d_lock); | |
2450 | if (target->d_parent != target) | |
2451 | spin_unlock(&target->d_parent->d_lock); | |
2452 | } | |
2453 | ||
1da177e4 | 2454 | /* |
2fd6b7f5 NP |
2455 | * When switching names, the actual string doesn't strictly have to |
2456 | * be preserved in the target - because we're dropping the target | |
2457 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2458 | * the new name before we switch. | |
2459 | * | |
2460 | * Note that we have to be a lot more careful about getting the hash | |
2461 | * switched - we have to switch the hash value properly even if it | |
2462 | * then no longer matches the actual (corrupted) string of the target. | |
2463 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2464 | */ |
9eaef27b | 2465 | /* |
18367501 | 2466 | * __d_move - move a dentry |
1da177e4 LT |
2467 | * @dentry: entry to move |
2468 | * @target: new dentry | |
da1ce067 | 2469 | * @exchange: exchange the two dentries |
1da177e4 LT |
2470 | * |
2471 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2472 | * dcache entries should not be moved in this way. Caller must hold |
2473 | * rename_lock, the i_mutex of the source and target directories, | |
2474 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2475 | */ |
da1ce067 MS |
2476 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2477 | bool exchange) | |
1da177e4 | 2478 | { |
1da177e4 LT |
2479 | if (!dentry->d_inode) |
2480 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2481 | ||
2fd6b7f5 NP |
2482 | BUG_ON(d_ancestor(dentry, target)); |
2483 | BUG_ON(d_ancestor(target, dentry)); | |
2484 | ||
2fd6b7f5 | 2485 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2486 | |
31e6b01f | 2487 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2488 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2489 | |
ceb5bdc2 NP |
2490 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2491 | ||
2492 | /* | |
2493 | * Move the dentry to the target hash queue. Don't bother checking | |
2494 | * for the same hash queue because of how unlikely it is. | |
2495 | */ | |
2496 | __d_drop(dentry); | |
789680d1 | 2497 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 | 2498 | |
da1ce067 MS |
2499 | /* |
2500 | * Unhash the target (d_delete() is not usable here). If exchanging | |
2501 | * the two dentries, then rehash onto the other's hash queue. | |
2502 | */ | |
1da177e4 | 2503 | __d_drop(target); |
da1ce067 MS |
2504 | if (exchange) { |
2505 | __d_rehash(target, | |
2506 | d_hash(dentry->d_parent, dentry->d_name.hash)); | |
2507 | } | |
1da177e4 | 2508 | |
5160ee6f ED |
2509 | list_del(&dentry->d_u.d_child); |
2510 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2511 | |
2512 | /* Switch the names.. */ | |
2513 | switch_names(dentry, target); | |
9a8d5bb4 | 2514 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2515 | |
2516 | /* ... and switch the parents */ | |
2517 | if (IS_ROOT(dentry)) { | |
2518 | dentry->d_parent = target->d_parent; | |
2519 | target->d_parent = target; | |
5160ee6f | 2520 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2521 | } else { |
9a8d5bb4 | 2522 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2523 | |
2524 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2525 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2526 | } |
2527 | ||
5160ee6f | 2528 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2529 | |
31e6b01f NP |
2530 | write_seqcount_end(&target->d_seq); |
2531 | write_seqcount_end(&dentry->d_seq); | |
2532 | ||
2fd6b7f5 | 2533 | dentry_unlock_parents_for_move(dentry, target); |
da1ce067 MS |
2534 | if (exchange) |
2535 | fsnotify_d_move(target); | |
1da177e4 | 2536 | spin_unlock(&target->d_lock); |
c32ccd87 | 2537 | fsnotify_d_move(dentry); |
1da177e4 | 2538 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2539 | } |
2540 | ||
2541 | /* | |
2542 | * d_move - move a dentry | |
2543 | * @dentry: entry to move | |
2544 | * @target: new dentry | |
2545 | * | |
2546 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2547 | * dcache entries should not be moved in this way. See the locking |
2548 | * requirements for __d_move. | |
18367501 AV |
2549 | */ |
2550 | void d_move(struct dentry *dentry, struct dentry *target) | |
2551 | { | |
2552 | write_seqlock(&rename_lock); | |
da1ce067 | 2553 | __d_move(dentry, target, false); |
1da177e4 | 2554 | write_sequnlock(&rename_lock); |
9eaef27b | 2555 | } |
ec4f8605 | 2556 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2557 | |
da1ce067 MS |
2558 | /* |
2559 | * d_exchange - exchange two dentries | |
2560 | * @dentry1: first dentry | |
2561 | * @dentry2: second dentry | |
2562 | */ | |
2563 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2564 | { | |
2565 | write_seqlock(&rename_lock); | |
2566 | ||
2567 | WARN_ON(!dentry1->d_inode); | |
2568 | WARN_ON(!dentry2->d_inode); | |
2569 | WARN_ON(IS_ROOT(dentry1)); | |
2570 | WARN_ON(IS_ROOT(dentry2)); | |
2571 | ||
2572 | __d_move(dentry1, dentry2, true); | |
2573 | ||
2574 | write_sequnlock(&rename_lock); | |
2575 | } | |
2576 | ||
e2761a11 OH |
2577 | /** |
2578 | * d_ancestor - search for an ancestor | |
2579 | * @p1: ancestor dentry | |
2580 | * @p2: child dentry | |
2581 | * | |
2582 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2583 | * an ancestor of p2, else NULL. | |
9eaef27b | 2584 | */ |
e2761a11 | 2585 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2586 | { |
2587 | struct dentry *p; | |
2588 | ||
871c0067 | 2589 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2590 | if (p->d_parent == p1) |
e2761a11 | 2591 | return p; |
9eaef27b | 2592 | } |
e2761a11 | 2593 | return NULL; |
9eaef27b TM |
2594 | } |
2595 | ||
2596 | /* | |
2597 | * This helper attempts to cope with remotely renamed directories | |
2598 | * | |
2599 | * It assumes that the caller is already holding | |
18367501 | 2600 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2601 | * |
2602 | * Note: If ever the locking in lock_rename() changes, then please | |
2603 | * remember to update this too... | |
9eaef27b | 2604 | */ |
873feea0 NP |
2605 | static struct dentry *__d_unalias(struct inode *inode, |
2606 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2607 | { |
2608 | struct mutex *m1 = NULL, *m2 = NULL; | |
ee3efa91 | 2609 | struct dentry *ret = ERR_PTR(-EBUSY); |
9eaef27b TM |
2610 | |
2611 | /* If alias and dentry share a parent, then no extra locks required */ | |
2612 | if (alias->d_parent == dentry->d_parent) | |
2613 | goto out_unalias; | |
2614 | ||
9eaef27b | 2615 | /* See lock_rename() */ |
9eaef27b TM |
2616 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2617 | goto out_err; | |
2618 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2619 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2620 | goto out_err; | |
2621 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2622 | out_unalias: | |
ee3efa91 | 2623 | if (likely(!d_mountpoint(alias))) { |
da1ce067 | 2624 | __d_move(alias, dentry, false); |
ee3efa91 AV |
2625 | ret = alias; |
2626 | } | |
9eaef27b | 2627 | out_err: |
873feea0 | 2628 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2629 | if (m2) |
2630 | mutex_unlock(m2); | |
2631 | if (m1) | |
2632 | mutex_unlock(m1); | |
2633 | return ret; | |
2634 | } | |
2635 | ||
770bfad8 DH |
2636 | /* |
2637 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2638 | * named dentry in place of the dentry to be replaced. | |
2639 | */ | |
2640 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2641 | { | |
740da42e | 2642 | struct dentry *dparent; |
770bfad8 | 2643 | |
2fd6b7f5 | 2644 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2645 | |
31e6b01f | 2646 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2647 | write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2648 | |
770bfad8 | 2649 | dparent = dentry->d_parent; |
770bfad8 | 2650 | |
2fd6b7f5 NP |
2651 | switch_names(dentry, anon); |
2652 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2653 | ||
740da42e AV |
2654 | dentry->d_parent = dentry; |
2655 | list_del_init(&dentry->d_u.d_child); | |
2656 | anon->d_parent = dparent; | |
5cc3821b | 2657 | list_move(&anon->d_u.d_child, &dparent->d_subdirs); |
6f18493e AV |
2658 | if (likely(!d_unhashed(anon))) { |
2659 | hlist_bl_lock(&anon->d_sb->s_anon); | |
2660 | __hlist_bl_del(&anon->d_hash); | |
2661 | anon->d_hash.pprev = NULL; | |
2662 | hlist_bl_unlock(&anon->d_sb->s_anon); | |
2663 | } | |
5cc3821b | 2664 | __d_rehash(anon, d_hash(anon->d_parent, anon->d_name.hash)); |
770bfad8 | 2665 | |
31e6b01f NP |
2666 | write_seqcount_end(&dentry->d_seq); |
2667 | write_seqcount_end(&anon->d_seq); | |
2668 | ||
2fd6b7f5 NP |
2669 | dentry_unlock_parents_for_move(anon, dentry); |
2670 | spin_unlock(&dentry->d_lock); | |
5cc3821b | 2671 | spin_unlock(&anon->d_lock); |
770bfad8 DH |
2672 | } |
2673 | ||
3f70bd51 BF |
2674 | /** |
2675 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
2676 | * @inode: the inode which may have a disconnected dentry | |
2677 | * @dentry: a negative dentry which we want to point to the inode. | |
2678 | * | |
da093a9b BF |
2679 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
2680 | * place of the given dentry and return it, else simply d_add the inode | |
2681 | * to the dentry and return NULL. | |
3f70bd51 | 2682 | * |
908790fa BF |
2683 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
2684 | * we should error out: directories can't have multiple aliases. | |
2685 | * | |
3f70bd51 BF |
2686 | * This is needed in the lookup routine of any filesystem that is exportable |
2687 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
2688 | * | |
2689 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
2690 | * is returned. This matches the expected return value of ->lookup. | |
2691 | * | |
2692 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
2693 | * In that case, we know that the inode will be a regular file, and also this | |
2694 | * will only occur during atomic_open. So we need to check for the dentry | |
2695 | * being already hashed only in the final case. | |
2696 | */ | |
2697 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
2698 | { | |
2699 | struct dentry *new = NULL; | |
2700 | ||
2701 | if (IS_ERR(inode)) | |
2702 | return ERR_CAST(inode); | |
2703 | ||
2704 | if (inode && S_ISDIR(inode->i_mode)) { | |
2705 | spin_lock(&inode->i_lock); | |
908790fa | 2706 | new = __d_find_any_alias(inode); |
3f70bd51 | 2707 | if (new) { |
da093a9b | 2708 | if (!IS_ROOT(new)) { |
908790fa BF |
2709 | spin_unlock(&inode->i_lock); |
2710 | dput(new); | |
2711 | return ERR_PTR(-EIO); | |
2712 | } | |
95ad5c29 BF |
2713 | if (d_ancestor(new, dentry)) { |
2714 | spin_unlock(&inode->i_lock); | |
2715 | dput(new); | |
2716 | return ERR_PTR(-EIO); | |
2717 | } | |
75a2352d BF |
2718 | write_seqlock(&rename_lock); |
2719 | __d_materialise_dentry(dentry, new); | |
2720 | write_sequnlock(&rename_lock); | |
3f70bd51 BF |
2721 | spin_unlock(&inode->i_lock); |
2722 | security_d_instantiate(new, inode); | |
3f70bd51 BF |
2723 | iput(inode); |
2724 | } else { | |
2725 | /* already taking inode->i_lock, so d_add() by hand */ | |
2726 | __d_instantiate(dentry, inode); | |
2727 | spin_unlock(&inode->i_lock); | |
2728 | security_d_instantiate(dentry, inode); | |
2729 | d_rehash(dentry); | |
2730 | } | |
2731 | } else { | |
2732 | d_instantiate(dentry, inode); | |
2733 | if (d_unhashed(dentry)) | |
2734 | d_rehash(dentry); | |
2735 | } | |
2736 | return new; | |
2737 | } | |
2738 | EXPORT_SYMBOL(d_splice_alias); | |
2739 | ||
770bfad8 DH |
2740 | /** |
2741 | * d_materialise_unique - introduce an inode into the tree | |
2742 | * @dentry: candidate dentry | |
2743 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2744 | * | |
2745 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2746 | * root directory alias in its place if there is one. Caller must hold the |
2747 | * i_mutex of the parent directory. | |
770bfad8 DH |
2748 | */ |
2749 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2750 | { | |
9eaef27b | 2751 | struct dentry *actual; |
770bfad8 DH |
2752 | |
2753 | BUG_ON(!d_unhashed(dentry)); | |
2754 | ||
770bfad8 DH |
2755 | if (!inode) { |
2756 | actual = dentry; | |
360da900 | 2757 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2758 | d_rehash(actual); |
2759 | goto out_nolock; | |
770bfad8 DH |
2760 | } |
2761 | ||
873feea0 | 2762 | spin_lock(&inode->i_lock); |
357f8e65 | 2763 | |
9eaef27b TM |
2764 | if (S_ISDIR(inode->i_mode)) { |
2765 | struct dentry *alias; | |
2766 | ||
2767 | /* Does an aliased dentry already exist? */ | |
52ed46f0 | 2768 | alias = __d_find_alias(inode); |
9eaef27b TM |
2769 | if (alias) { |
2770 | actual = alias; | |
18367501 AV |
2771 | write_seqlock(&rename_lock); |
2772 | ||
2773 | if (d_ancestor(alias, dentry)) { | |
2774 | /* Check for loops */ | |
2775 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2776 | spin_unlock(&inode->i_lock); |
18367501 AV |
2777 | } else if (IS_ROOT(alias)) { |
2778 | /* Is this an anonymous mountpoint that we | |
2779 | * could splice into our tree? */ | |
9eaef27b | 2780 | __d_materialise_dentry(dentry, alias); |
18367501 | 2781 | write_sequnlock(&rename_lock); |
9eaef27b | 2782 | goto found; |
18367501 AV |
2783 | } else { |
2784 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2785 | * aliasing. This drops inode->i_lock */ |
18367501 | 2786 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2787 | } |
18367501 | 2788 | write_sequnlock(&rename_lock); |
dd179946 DH |
2789 | if (IS_ERR(actual)) { |
2790 | if (PTR_ERR(actual) == -ELOOP) | |
2791 | pr_warn_ratelimited( | |
2792 | "VFS: Lookup of '%s' in %s %s" | |
2793 | " would have caused loop\n", | |
2794 | dentry->d_name.name, | |
2795 | inode->i_sb->s_type->name, | |
2796 | inode->i_sb->s_id); | |
9eaef27b | 2797 | dput(alias); |
dd179946 | 2798 | } |
9eaef27b TM |
2799 | goto out_nolock; |
2800 | } | |
770bfad8 DH |
2801 | } |
2802 | ||
2803 | /* Add a unique reference */ | |
2804 | actual = __d_instantiate_unique(dentry, inode); | |
2805 | if (!actual) | |
2806 | actual = dentry; | |
357f8e65 NP |
2807 | else |
2808 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2809 | |
770bfad8 | 2810 | spin_lock(&actual->d_lock); |
770bfad8 DH |
2811 | _d_rehash(actual); |
2812 | spin_unlock(&actual->d_lock); | |
5cc3821b | 2813 | found: |
873feea0 | 2814 | spin_unlock(&inode->i_lock); |
9eaef27b | 2815 | out_nolock: |
770bfad8 DH |
2816 | if (actual == dentry) { |
2817 | security_d_instantiate(dentry, inode); | |
2818 | return NULL; | |
2819 | } | |
2820 | ||
2821 | iput(inode); | |
2822 | return actual; | |
770bfad8 | 2823 | } |
ec4f8605 | 2824 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2825 | |
cdd16d02 | 2826 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2827 | { |
2828 | *buflen -= namelen; | |
2829 | if (*buflen < 0) | |
2830 | return -ENAMETOOLONG; | |
2831 | *buffer -= namelen; | |
2832 | memcpy(*buffer, str, namelen); | |
2833 | return 0; | |
2834 | } | |
2835 | ||
232d2d60 WL |
2836 | /** |
2837 | * prepend_name - prepend a pathname in front of current buffer pointer | |
18129977 WL |
2838 | * @buffer: buffer pointer |
2839 | * @buflen: allocated length of the buffer | |
2840 | * @name: name string and length qstr structure | |
232d2d60 WL |
2841 | * |
2842 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
2843 | * make sure that either the old or the new name pointer and length are | |
2844 | * fetched. However, there may be mismatch between length and pointer. | |
2845 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
2846 | * the length is reached or a null byte is found. It also prepends "/" at | |
2847 | * the beginning of the name. The sequence number check at the caller will | |
2848 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
2849 | * due to mismatched pointer and length will be discarded. | |
2850 | */ | |
cdd16d02 MS |
2851 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2852 | { | |
232d2d60 WL |
2853 | const char *dname = ACCESS_ONCE(name->name); |
2854 | u32 dlen = ACCESS_ONCE(name->len); | |
2855 | char *p; | |
2856 | ||
232d2d60 | 2857 | *buflen -= dlen + 1; |
e825196d AV |
2858 | if (*buflen < 0) |
2859 | return -ENAMETOOLONG; | |
232d2d60 WL |
2860 | p = *buffer -= dlen + 1; |
2861 | *p++ = '/'; | |
2862 | while (dlen--) { | |
2863 | char c = *dname++; | |
2864 | if (!c) | |
2865 | break; | |
2866 | *p++ = c; | |
2867 | } | |
2868 | return 0; | |
cdd16d02 MS |
2869 | } |
2870 | ||
1da177e4 | 2871 | /** |
208898c1 | 2872 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2873 | * @path: the dentry/vfsmount to report |
02125a82 | 2874 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2875 | * @buffer: pointer to the end of the buffer |
2876 | * @buflen: pointer to buffer length | |
552ce544 | 2877 | * |
18129977 WL |
2878 | * The function will first try to write out the pathname without taking any |
2879 | * lock other than the RCU read lock to make sure that dentries won't go away. | |
2880 | * It only checks the sequence number of the global rename_lock as any change | |
2881 | * in the dentry's d_seq will be preceded by changes in the rename_lock | |
2882 | * sequence number. If the sequence number had been changed, it will restart | |
2883 | * the whole pathname back-tracing sequence again by taking the rename_lock. | |
2884 | * In this case, there is no need to take the RCU read lock as the recursive | |
2885 | * parent pointer references will keep the dentry chain alive as long as no | |
2886 | * rename operation is performed. | |
1da177e4 | 2887 | */ |
02125a82 AV |
2888 | static int prepend_path(const struct path *path, |
2889 | const struct path *root, | |
f2eb6575 | 2890 | char **buffer, int *buflen) |
1da177e4 | 2891 | { |
ede4cebc AV |
2892 | struct dentry *dentry; |
2893 | struct vfsmount *vfsmnt; | |
2894 | struct mount *mnt; | |
f2eb6575 | 2895 | int error = 0; |
48a066e7 | 2896 | unsigned seq, m_seq = 0; |
232d2d60 WL |
2897 | char *bptr; |
2898 | int blen; | |
6092d048 | 2899 | |
48f5ec21 | 2900 | rcu_read_lock(); |
48a066e7 AV |
2901 | restart_mnt: |
2902 | read_seqbegin_or_lock(&mount_lock, &m_seq); | |
2903 | seq = 0; | |
4ec6c2ae | 2904 | rcu_read_lock(); |
232d2d60 WL |
2905 | restart: |
2906 | bptr = *buffer; | |
2907 | blen = *buflen; | |
48a066e7 | 2908 | error = 0; |
ede4cebc AV |
2909 | dentry = path->dentry; |
2910 | vfsmnt = path->mnt; | |
2911 | mnt = real_mount(vfsmnt); | |
232d2d60 | 2912 | read_seqbegin_or_lock(&rename_lock, &seq); |
f2eb6575 | 2913 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2914 | struct dentry * parent; |
2915 | ||
1da177e4 | 2916 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
48a066e7 | 2917 | struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); |
552ce544 | 2918 | /* Global root? */ |
48a066e7 AV |
2919 | if (mnt != parent) { |
2920 | dentry = ACCESS_ONCE(mnt->mnt_mountpoint); | |
2921 | mnt = parent; | |
232d2d60 WL |
2922 | vfsmnt = &mnt->mnt; |
2923 | continue; | |
2924 | } | |
2925 | /* | |
2926 | * Filesystems needing to implement special "root names" | |
2927 | * should do so with ->d_dname() | |
2928 | */ | |
2929 | if (IS_ROOT(dentry) && | |
2930 | (dentry->d_name.len != 1 || | |
2931 | dentry->d_name.name[0] != '/')) { | |
2932 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2933 | (int) dentry->d_name.len, | |
2934 | dentry->d_name.name); | |
2935 | } | |
2936 | if (!error) | |
2937 | error = is_mounted(vfsmnt) ? 1 : 2; | |
2938 | break; | |
1da177e4 LT |
2939 | } |
2940 | parent = dentry->d_parent; | |
2941 | prefetch(parent); | |
232d2d60 | 2942 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
2943 | if (error) |
2944 | break; | |
2945 | ||
1da177e4 LT |
2946 | dentry = parent; |
2947 | } | |
48f5ec21 AV |
2948 | if (!(seq & 1)) |
2949 | rcu_read_unlock(); | |
2950 | if (need_seqretry(&rename_lock, seq)) { | |
2951 | seq = 1; | |
232d2d60 | 2952 | goto restart; |
48f5ec21 AV |
2953 | } |
2954 | done_seqretry(&rename_lock, seq); | |
4ec6c2ae LZ |
2955 | |
2956 | if (!(m_seq & 1)) | |
2957 | rcu_read_unlock(); | |
48a066e7 AV |
2958 | if (need_seqretry(&mount_lock, m_seq)) { |
2959 | m_seq = 1; | |
2960 | goto restart_mnt; | |
2961 | } | |
2962 | done_seqretry(&mount_lock, m_seq); | |
1da177e4 | 2963 | |
232d2d60 WL |
2964 | if (error >= 0 && bptr == *buffer) { |
2965 | if (--blen < 0) | |
2966 | error = -ENAMETOOLONG; | |
2967 | else | |
2968 | *--bptr = '/'; | |
2969 | } | |
2970 | *buffer = bptr; | |
2971 | *buflen = blen; | |
7ea600b5 | 2972 | return error; |
f2eb6575 | 2973 | } |
be285c71 | 2974 | |
f2eb6575 MS |
2975 | /** |
2976 | * __d_path - return the path of a dentry | |
2977 | * @path: the dentry/vfsmount to report | |
02125a82 | 2978 | * @root: root vfsmnt/dentry |
cd956a1c | 2979 | * @buf: buffer to return value in |
f2eb6575 MS |
2980 | * @buflen: buffer length |
2981 | * | |
ffd1f4ed | 2982 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2983 | * |
2984 | * Returns a pointer into the buffer or an error code if the | |
2985 | * path was too long. | |
2986 | * | |
be148247 | 2987 | * "buflen" should be positive. |
f2eb6575 | 2988 | * |
02125a82 | 2989 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2990 | */ |
02125a82 AV |
2991 | char *__d_path(const struct path *path, |
2992 | const struct path *root, | |
f2eb6575 MS |
2993 | char *buf, int buflen) |
2994 | { | |
2995 | char *res = buf + buflen; | |
2996 | int error; | |
2997 | ||
2998 | prepend(&res, &buflen, "\0", 1); | |
f2eb6575 | 2999 | error = prepend_path(path, root, &res, &buflen); |
be148247 | 3000 | |
02125a82 AV |
3001 | if (error < 0) |
3002 | return ERR_PTR(error); | |
3003 | if (error > 0) | |
3004 | return NULL; | |
3005 | return res; | |
3006 | } | |
3007 | ||
3008 | char *d_absolute_path(const struct path *path, | |
3009 | char *buf, int buflen) | |
3010 | { | |
3011 | struct path root = {}; | |
3012 | char *res = buf + buflen; | |
3013 | int error; | |
3014 | ||
3015 | prepend(&res, &buflen, "\0", 1); | |
02125a82 | 3016 | error = prepend_path(path, &root, &res, &buflen); |
02125a82 AV |
3017 | |
3018 | if (error > 1) | |
3019 | error = -EINVAL; | |
3020 | if (error < 0) | |
f2eb6575 | 3021 | return ERR_PTR(error); |
f2eb6575 | 3022 | return res; |
1da177e4 LT |
3023 | } |
3024 | ||
ffd1f4ed MS |
3025 | /* |
3026 | * same as __d_path but appends "(deleted)" for unlinked files. | |
3027 | */ | |
02125a82 AV |
3028 | static int path_with_deleted(const struct path *path, |
3029 | const struct path *root, | |
3030 | char **buf, int *buflen) | |
ffd1f4ed MS |
3031 | { |
3032 | prepend(buf, buflen, "\0", 1); | |
3033 | if (d_unlinked(path->dentry)) { | |
3034 | int error = prepend(buf, buflen, " (deleted)", 10); | |
3035 | if (error) | |
3036 | return error; | |
3037 | } | |
3038 | ||
3039 | return prepend_path(path, root, buf, buflen); | |
3040 | } | |
3041 | ||
8df9d1a4 MS |
3042 | static int prepend_unreachable(char **buffer, int *buflen) |
3043 | { | |
3044 | return prepend(buffer, buflen, "(unreachable)", 13); | |
3045 | } | |
3046 | ||
68f0d9d9 LT |
3047 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) |
3048 | { | |
3049 | unsigned seq; | |
3050 | ||
3051 | do { | |
3052 | seq = read_seqcount_begin(&fs->seq); | |
3053 | *root = fs->root; | |
3054 | } while (read_seqcount_retry(&fs->seq, seq)); | |
3055 | } | |
3056 | ||
a03a8a70 JB |
3057 | /** |
3058 | * d_path - return the path of a dentry | |
cf28b486 | 3059 | * @path: path to report |
a03a8a70 JB |
3060 | * @buf: buffer to return value in |
3061 | * @buflen: buffer length | |
3062 | * | |
3063 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
3064 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
3065 | * | |
52afeefb AV |
3066 | * Returns a pointer into the buffer or an error code if the path was |
3067 | * too long. Note: Callers should use the returned pointer, not the passed | |
3068 | * in buffer, to use the name! The implementation often starts at an offset | |
3069 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 3070 | * |
31f3e0b3 | 3071 | * "buflen" should be positive. |
a03a8a70 | 3072 | */ |
20d4fdc1 | 3073 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 3074 | { |
ffd1f4ed | 3075 | char *res = buf + buflen; |
6ac08c39 | 3076 | struct path root; |
ffd1f4ed | 3077 | int error; |
1da177e4 | 3078 | |
c23fbb6b ED |
3079 | /* |
3080 | * We have various synthetic filesystems that never get mounted. On | |
3081 | * these filesystems dentries are never used for lookup purposes, and | |
3082 | * thus don't need to be hashed. They also don't need a name until a | |
3083 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
3084 | * below allows us to generate a name for these objects on demand: | |
f48cfddc EB |
3085 | * |
3086 | * Some pseudo inodes are mountable. When they are mounted | |
3087 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname | |
3088 | * and instead have d_path return the mounted path. | |
c23fbb6b | 3089 | */ |
f48cfddc EB |
3090 | if (path->dentry->d_op && path->dentry->d_op->d_dname && |
3091 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) | |
cf28b486 | 3092 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
c23fbb6b | 3093 | |
68f0d9d9 LT |
3094 | rcu_read_lock(); |
3095 | get_fs_root_rcu(current->fs, &root); | |
02125a82 | 3096 | error = path_with_deleted(path, &root, &res, &buflen); |
68f0d9d9 LT |
3097 | rcu_read_unlock(); |
3098 | ||
02125a82 | 3099 | if (error < 0) |
ffd1f4ed | 3100 | res = ERR_PTR(error); |
1da177e4 LT |
3101 | return res; |
3102 | } | |
ec4f8605 | 3103 | EXPORT_SYMBOL(d_path); |
1da177e4 | 3104 | |
c23fbb6b ED |
3105 | /* |
3106 | * Helper function for dentry_operations.d_dname() members | |
3107 | */ | |
3108 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
3109 | const char *fmt, ...) | |
3110 | { | |
3111 | va_list args; | |
3112 | char temp[64]; | |
3113 | int sz; | |
3114 | ||
3115 | va_start(args, fmt); | |
3116 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
3117 | va_end(args); | |
3118 | ||
3119 | if (sz > sizeof(temp) || sz > buflen) | |
3120 | return ERR_PTR(-ENAMETOOLONG); | |
3121 | ||
3122 | buffer += buflen - sz; | |
3123 | return memcpy(buffer, temp, sz); | |
3124 | } | |
3125 | ||
118b2302 AV |
3126 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
3127 | { | |
3128 | char *end = buffer + buflen; | |
3129 | /* these dentries are never renamed, so d_lock is not needed */ | |
3130 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
232d2d60 | 3131 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
118b2302 AV |
3132 | prepend(&end, &buflen, "/", 1)) |
3133 | end = ERR_PTR(-ENAMETOOLONG); | |
232d2d60 | 3134 | return end; |
118b2302 | 3135 | } |
31bbe16f | 3136 | EXPORT_SYMBOL(simple_dname); |
118b2302 | 3137 | |
6092d048 RP |
3138 | /* |
3139 | * Write full pathname from the root of the filesystem into the buffer. | |
3140 | */ | |
f6500801 | 3141 | static char *__dentry_path(struct dentry *d, char *buf, int buflen) |
6092d048 | 3142 | { |
f6500801 | 3143 | struct dentry *dentry; |
232d2d60 WL |
3144 | char *end, *retval; |
3145 | int len, seq = 0; | |
3146 | int error = 0; | |
6092d048 | 3147 | |
f6500801 AV |
3148 | if (buflen < 2) |
3149 | goto Elong; | |
3150 | ||
48f5ec21 | 3151 | rcu_read_lock(); |
232d2d60 | 3152 | restart: |
f6500801 | 3153 | dentry = d; |
232d2d60 WL |
3154 | end = buf + buflen; |
3155 | len = buflen; | |
3156 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
3157 | /* Get '/' right */ |
3158 | retval = end-1; | |
3159 | *retval = '/'; | |
232d2d60 | 3160 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
3161 | while (!IS_ROOT(dentry)) { |
3162 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 3163 | |
6092d048 | 3164 | prefetch(parent); |
232d2d60 WL |
3165 | error = prepend_name(&end, &len, &dentry->d_name); |
3166 | if (error) | |
3167 | break; | |
6092d048 RP |
3168 | |
3169 | retval = end; | |
3170 | dentry = parent; | |
3171 | } | |
48f5ec21 AV |
3172 | if (!(seq & 1)) |
3173 | rcu_read_unlock(); | |
3174 | if (need_seqretry(&rename_lock, seq)) { | |
3175 | seq = 1; | |
232d2d60 | 3176 | goto restart; |
48f5ec21 AV |
3177 | } |
3178 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3179 | if (error) |
3180 | goto Elong; | |
c103135c AV |
3181 | return retval; |
3182 | Elong: | |
3183 | return ERR_PTR(-ENAMETOOLONG); | |
3184 | } | |
ec2447c2 NP |
3185 | |
3186 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3187 | { | |
232d2d60 | 3188 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3189 | } |
3190 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3191 | |
3192 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3193 | { | |
3194 | char *p = NULL; | |
3195 | char *retval; | |
3196 | ||
c103135c AV |
3197 | if (d_unlinked(dentry)) { |
3198 | p = buf + buflen; | |
3199 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3200 | goto Elong; | |
3201 | buflen++; | |
3202 | } | |
3203 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3204 | if (!IS_ERR(retval) && p) |
3205 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3206 | return retval; |
3207 | Elong: | |
6092d048 RP |
3208 | return ERR_PTR(-ENAMETOOLONG); |
3209 | } | |
3210 | ||
8b19e341 LT |
3211 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, |
3212 | struct path *pwd) | |
5762482f | 3213 | { |
8b19e341 LT |
3214 | unsigned seq; |
3215 | ||
3216 | do { | |
3217 | seq = read_seqcount_begin(&fs->seq); | |
3218 | *root = fs->root; | |
3219 | *pwd = fs->pwd; | |
3220 | } while (read_seqcount_retry(&fs->seq, seq)); | |
5762482f LT |
3221 | } |
3222 | ||
1da177e4 LT |
3223 | /* |
3224 | * NOTE! The user-level library version returns a | |
3225 | * character pointer. The kernel system call just | |
3226 | * returns the length of the buffer filled (which | |
3227 | * includes the ending '\0' character), or a negative | |
3228 | * error value. So libc would do something like | |
3229 | * | |
3230 | * char *getcwd(char * buf, size_t size) | |
3231 | * { | |
3232 | * int retval; | |
3233 | * | |
3234 | * retval = sys_getcwd(buf, size); | |
3235 | * if (retval >= 0) | |
3236 | * return buf; | |
3237 | * errno = -retval; | |
3238 | * return NULL; | |
3239 | * } | |
3240 | */ | |
3cdad428 | 3241 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3242 | { |
552ce544 | 3243 | int error; |
6ac08c39 | 3244 | struct path pwd, root; |
3272c544 | 3245 | char *page = __getname(); |
1da177e4 LT |
3246 | |
3247 | if (!page) | |
3248 | return -ENOMEM; | |
3249 | ||
8b19e341 LT |
3250 | rcu_read_lock(); |
3251 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); | |
1da177e4 | 3252 | |
552ce544 | 3253 | error = -ENOENT; |
f3da392e | 3254 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3255 | unsigned long len; |
3272c544 LT |
3256 | char *cwd = page + PATH_MAX; |
3257 | int buflen = PATH_MAX; | |
1da177e4 | 3258 | |
8df9d1a4 | 3259 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3260 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
ff812d72 | 3261 | rcu_read_unlock(); |
552ce544 | 3262 | |
02125a82 | 3263 | if (error < 0) |
552ce544 LT |
3264 | goto out; |
3265 | ||
8df9d1a4 | 3266 | /* Unreachable from current root */ |
02125a82 | 3267 | if (error > 0) { |
8df9d1a4 MS |
3268 | error = prepend_unreachable(&cwd, &buflen); |
3269 | if (error) | |
3270 | goto out; | |
3271 | } | |
3272 | ||
552ce544 | 3273 | error = -ERANGE; |
3272c544 | 3274 | len = PATH_MAX + page - cwd; |
552ce544 LT |
3275 | if (len <= size) { |
3276 | error = len; | |
3277 | if (copy_to_user(buf, cwd, len)) | |
3278 | error = -EFAULT; | |
3279 | } | |
949854d0 | 3280 | } else { |
ff812d72 | 3281 | rcu_read_unlock(); |
949854d0 | 3282 | } |
1da177e4 LT |
3283 | |
3284 | out: | |
3272c544 | 3285 | __putname(page); |
1da177e4 LT |
3286 | return error; |
3287 | } | |
3288 | ||
3289 | /* | |
3290 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3291 | * | |
3292 | * Trivially implemented using the dcache structure | |
3293 | */ | |
3294 | ||
3295 | /** | |
3296 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3297 | * @new_dentry: new dentry | |
3298 | * @old_dentry: old dentry | |
3299 | * | |
3300 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
3301 | * Returns 0 otherwise. | |
3302 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
3303 | */ | |
3304 | ||
e2761a11 | 3305 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
3306 | { |
3307 | int result; | |
949854d0 | 3308 | unsigned seq; |
1da177e4 | 3309 | |
e2761a11 OH |
3310 | if (new_dentry == old_dentry) |
3311 | return 1; | |
3312 | ||
e2761a11 | 3313 | do { |
1da177e4 | 3314 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3315 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3316 | /* |
3317 | * Need rcu_readlock to protect against the d_parent trashing | |
3318 | * due to d_move | |
3319 | */ | |
3320 | rcu_read_lock(); | |
e2761a11 | 3321 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 3322 | result = 1; |
e2761a11 OH |
3323 | else |
3324 | result = 0; | |
949854d0 | 3325 | rcu_read_unlock(); |
1da177e4 | 3326 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3327 | |
3328 | return result; | |
3329 | } | |
3330 | ||
db14fc3a | 3331 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3332 | { |
db14fc3a MS |
3333 | struct dentry *root = data; |
3334 | if (dentry != root) { | |
3335 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3336 | return D_WALK_SKIP; | |
1da177e4 | 3337 | |
01ddc4ed MS |
3338 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3339 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3340 | dentry->d_lockref.count--; | |
3341 | } | |
1da177e4 | 3342 | } |
db14fc3a MS |
3343 | return D_WALK_CONTINUE; |
3344 | } | |
58db63d0 | 3345 | |
db14fc3a MS |
3346 | void d_genocide(struct dentry *parent) |
3347 | { | |
3348 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3349 | } |
3350 | ||
60545d0d | 3351 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3352 | { |
60545d0d AV |
3353 | inode_dec_link_count(inode); |
3354 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
3355 | !hlist_unhashed(&dentry->d_alias) || | |
3356 | !d_unlinked(dentry)); | |
3357 | spin_lock(&dentry->d_parent->d_lock); | |
3358 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3359 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3360 | (unsigned long long)inode->i_ino); | |
3361 | spin_unlock(&dentry->d_lock); | |
3362 | spin_unlock(&dentry->d_parent->d_lock); | |
3363 | d_instantiate(dentry, inode); | |
1da177e4 | 3364 | } |
60545d0d | 3365 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3366 | |
3367 | static __initdata unsigned long dhash_entries; | |
3368 | static int __init set_dhash_entries(char *str) | |
3369 | { | |
3370 | if (!str) | |
3371 | return 0; | |
3372 | dhash_entries = simple_strtoul(str, &str, 0); | |
3373 | return 1; | |
3374 | } | |
3375 | __setup("dhash_entries=", set_dhash_entries); | |
3376 | ||
3377 | static void __init dcache_init_early(void) | |
3378 | { | |
074b8517 | 3379 | unsigned int loop; |
1da177e4 LT |
3380 | |
3381 | /* If hashes are distributed across NUMA nodes, defer | |
3382 | * hash allocation until vmalloc space is available. | |
3383 | */ | |
3384 | if (hashdist) | |
3385 | return; | |
3386 | ||
3387 | dentry_hashtable = | |
3388 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3389 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3390 | dhash_entries, |
3391 | 13, | |
3392 | HASH_EARLY, | |
3393 | &d_hash_shift, | |
3394 | &d_hash_mask, | |
31fe62b9 | 3395 | 0, |
1da177e4 LT |
3396 | 0); |
3397 | ||
074b8517 | 3398 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3399 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3400 | } |
3401 | ||
74bf17cf | 3402 | static void __init dcache_init(void) |
1da177e4 | 3403 | { |
074b8517 | 3404 | unsigned int loop; |
1da177e4 LT |
3405 | |
3406 | /* | |
3407 | * A constructor could be added for stable state like the lists, | |
3408 | * but it is probably not worth it because of the cache nature | |
3409 | * of the dcache. | |
3410 | */ | |
0a31bd5f CL |
3411 | dentry_cache = KMEM_CACHE(dentry, |
3412 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3413 | |
3414 | /* Hash may have been set up in dcache_init_early */ | |
3415 | if (!hashdist) | |
3416 | return; | |
3417 | ||
3418 | dentry_hashtable = | |
3419 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3420 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3421 | dhash_entries, |
3422 | 13, | |
3423 | 0, | |
3424 | &d_hash_shift, | |
3425 | &d_hash_mask, | |
31fe62b9 | 3426 | 0, |
1da177e4 LT |
3427 | 0); |
3428 | ||
074b8517 | 3429 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3430 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3431 | } |
3432 | ||
3433 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3434 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3435 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3436 | |
1da177e4 LT |
3437 | EXPORT_SYMBOL(d_genocide); |
3438 | ||
1da177e4 LT |
3439 | void __init vfs_caches_init_early(void) |
3440 | { | |
3441 | dcache_init_early(); | |
3442 | inode_init_early(); | |
3443 | } | |
3444 | ||
3445 | void __init vfs_caches_init(unsigned long mempages) | |
3446 | { | |
3447 | unsigned long reserve; | |
3448 | ||
3449 | /* Base hash sizes on available memory, with a reserve equal to | |
3450 | 150% of current kernel size */ | |
3451 | ||
3452 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3453 | mempages -= reserve; | |
3454 | ||
3455 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3456 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3457 | |
74bf17cf DC |
3458 | dcache_init(); |
3459 | inode_init(); | |
1da177e4 | 3460 | files_init(mempages); |
74bf17cf | 3461 | mnt_init(); |
1da177e4 LT |
3462 | bdev_cache_init(); |
3463 | chrdev_init(); | |
3464 | } |