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; |
482db906 AV |
109 | hash = hash + (hash >> d_hash_shift); |
110 | return dentry_hashtable + (hash & d_hash_mask); | |
ceb5bdc2 NP |
111 | } |
112 | ||
1da177e4 LT |
113 | /* Statistics gathering. */ |
114 | struct dentry_stat_t dentry_stat = { | |
115 | .age_limit = 45, | |
116 | }; | |
117 | ||
3942c07c | 118 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 119 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
312d3ca8 CH |
120 | |
121 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
62d36c77 DC |
122 | |
123 | /* | |
124 | * Here we resort to our own counters instead of using generic per-cpu counters | |
125 | * for consistency with what the vfs inode code does. We are expected to harvest | |
126 | * better code and performance by having our own specialized counters. | |
127 | * | |
128 | * Please note that the loop is done over all possible CPUs, not over all online | |
129 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
130 | * on and off. If one of them goes off, we will just keep their counters. | |
131 | * | |
132 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
133 | * please update all vfs counters to match. | |
134 | */ | |
3942c07c | 135 | static long get_nr_dentry(void) |
3e880fb5 NP |
136 | { |
137 | int i; | |
3942c07c | 138 | long sum = 0; |
3e880fb5 NP |
139 | for_each_possible_cpu(i) |
140 | sum += per_cpu(nr_dentry, i); | |
141 | return sum < 0 ? 0 : sum; | |
142 | } | |
143 | ||
62d36c77 DC |
144 | static long get_nr_dentry_unused(void) |
145 | { | |
146 | int i; | |
147 | long sum = 0; | |
148 | for_each_possible_cpu(i) | |
149 | sum += per_cpu(nr_dentry_unused, i); | |
150 | return sum < 0 ? 0 : sum; | |
151 | } | |
152 | ||
312d3ca8 CH |
153 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
154 | size_t *lenp, loff_t *ppos) | |
155 | { | |
3e880fb5 | 156 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 157 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
3942c07c | 158 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 CH |
159 | } |
160 | #endif | |
161 | ||
5483f18e LT |
162 | /* |
163 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
164 | * The strings are both count bytes long, and count is non-zero. | |
165 | */ | |
e419b4cc LT |
166 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
167 | ||
168 | #include <asm/word-at-a-time.h> | |
169 | /* | |
170 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
171 | * aligned allocation for this particular component. We don't | |
172 | * strictly need the load_unaligned_zeropad() safety, but it | |
173 | * doesn't hurt either. | |
174 | * | |
175 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
176 | * need the careful unaligned handling. | |
177 | */ | |
94753db5 | 178 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 179 | { |
bfcfaa77 | 180 | unsigned long a,b,mask; |
bfcfaa77 LT |
181 | |
182 | for (;;) { | |
12f8ad4b | 183 | a = *(unsigned long *)cs; |
e419b4cc | 184 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
185 | if (tcount < sizeof(unsigned long)) |
186 | break; | |
187 | if (unlikely(a != b)) | |
188 | return 1; | |
189 | cs += sizeof(unsigned long); | |
190 | ct += sizeof(unsigned long); | |
191 | tcount -= sizeof(unsigned long); | |
192 | if (!tcount) | |
193 | return 0; | |
194 | } | |
a5c21dce | 195 | mask = bytemask_from_count(tcount); |
bfcfaa77 | 196 | return unlikely(!!((a ^ b) & mask)); |
e419b4cc LT |
197 | } |
198 | ||
bfcfaa77 | 199 | #else |
e419b4cc | 200 | |
94753db5 | 201 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 202 | { |
5483f18e LT |
203 | do { |
204 | if (*cs != *ct) | |
205 | return 1; | |
206 | cs++; | |
207 | ct++; | |
208 | tcount--; | |
209 | } while (tcount); | |
210 | return 0; | |
211 | } | |
212 | ||
e419b4cc LT |
213 | #endif |
214 | ||
94753db5 LT |
215 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
216 | { | |
6326c71f | 217 | const unsigned char *cs; |
94753db5 LT |
218 | /* |
219 | * Be careful about RCU walk racing with rename: | |
220 | * use ACCESS_ONCE to fetch the name pointer. | |
221 | * | |
222 | * NOTE! Even if a rename will mean that the length | |
223 | * was not loaded atomically, we don't care. The | |
224 | * RCU walk will check the sequence count eventually, | |
225 | * and catch it. And we won't overrun the buffer, | |
226 | * because we're reading the name pointer atomically, | |
227 | * and a dentry name is guaranteed to be properly | |
228 | * terminated with a NUL byte. | |
229 | * | |
230 | * End result: even if 'len' is wrong, we'll exit | |
231 | * early because the data cannot match (there can | |
232 | * be no NUL in the ct/tcount data) | |
233 | */ | |
6326c71f LT |
234 | cs = ACCESS_ONCE(dentry->d_name.name); |
235 | smp_read_barrier_depends(); | |
236 | return dentry_string_cmp(cs, ct, tcount); | |
94753db5 LT |
237 | } |
238 | ||
9c82ab9c | 239 | static void __d_free(struct rcu_head *head) |
1da177e4 | 240 | { |
9c82ab9c CH |
241 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
242 | ||
b3d9b7a3 | 243 | WARN_ON(!hlist_unhashed(&dentry->d_alias)); |
1da177e4 LT |
244 | if (dname_external(dentry)) |
245 | kfree(dentry->d_name.name); | |
246 | kmem_cache_free(dentry_cache, dentry); | |
247 | } | |
248 | ||
b4f0354e AV |
249 | static void dentry_free(struct dentry *dentry) |
250 | { | |
251 | /* if dentry was never visible to RCU, immediate free is OK */ | |
252 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
253 | __d_free(&dentry->d_u.d_rcu); | |
254 | else | |
255 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
256 | } | |
257 | ||
31e6b01f NP |
258 | /** |
259 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
ff5fdb61 | 260 | * @dentry: the target dentry |
31e6b01f NP |
261 | * After this call, in-progress rcu-walk path lookup will fail. This |
262 | * should be called after unhashing, and after changing d_inode (if | |
263 | * the dentry has not already been unhashed). | |
264 | */ | |
265 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
266 | { | |
267 | assert_spin_locked(&dentry->d_lock); | |
268 | /* Go through a barrier */ | |
269 | write_seqcount_barrier(&dentry->d_seq); | |
270 | } | |
271 | ||
1da177e4 LT |
272 | /* |
273 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
274 | * d_iput() operation if defined. Dentry has no refcount |
275 | * and is unhashed. | |
1da177e4 | 276 | */ |
858119e1 | 277 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 278 | __releases(dentry->d_lock) |
873feea0 | 279 | __releases(dentry->d_inode->i_lock) |
1da177e4 LT |
280 | { |
281 | struct inode *inode = dentry->d_inode; | |
282 | if (inode) { | |
283 | dentry->d_inode = NULL; | |
b3d9b7a3 | 284 | hlist_del_init(&dentry->d_alias); |
1da177e4 | 285 | spin_unlock(&dentry->d_lock); |
873feea0 | 286 | spin_unlock(&inode->i_lock); |
f805fbda LT |
287 | if (!inode->i_nlink) |
288 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
289 | if (dentry->d_op && dentry->d_op->d_iput) |
290 | dentry->d_op->d_iput(dentry, inode); | |
291 | else | |
292 | iput(inode); | |
293 | } else { | |
294 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
295 | } |
296 | } | |
297 | ||
31e6b01f NP |
298 | /* |
299 | * Release the dentry's inode, using the filesystem | |
300 | * d_iput() operation if defined. dentry remains in-use. | |
301 | */ | |
302 | static void dentry_unlink_inode(struct dentry * dentry) | |
303 | __releases(dentry->d_lock) | |
873feea0 | 304 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
305 | { |
306 | struct inode *inode = dentry->d_inode; | |
b18825a7 | 307 | __d_clear_type(dentry); |
31e6b01f | 308 | dentry->d_inode = NULL; |
b3d9b7a3 | 309 | hlist_del_init(&dentry->d_alias); |
31e6b01f NP |
310 | dentry_rcuwalk_barrier(dentry); |
311 | spin_unlock(&dentry->d_lock); | |
873feea0 | 312 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
313 | if (!inode->i_nlink) |
314 | fsnotify_inoderemove(inode); | |
315 | if (dentry->d_op && dentry->d_op->d_iput) | |
316 | dentry->d_op->d_iput(dentry, inode); | |
317 | else | |
318 | iput(inode); | |
319 | } | |
320 | ||
89dc77bc LT |
321 | /* |
322 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
323 | * is in use - which includes both the "real" per-superblock | |
324 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
325 | * | |
326 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
327 | * on the shrink list (ie not on the superblock LRU list). | |
328 | * | |
329 | * The per-cpu "nr_dentry_unused" counters are updated with | |
330 | * the DCACHE_LRU_LIST bit. | |
331 | * | |
332 | * These helper functions make sure we always follow the | |
333 | * rules. d_lock must be held by the caller. | |
334 | */ | |
335 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
336 | static void d_lru_add(struct dentry *dentry) | |
337 | { | |
338 | D_FLAG_VERIFY(dentry, 0); | |
339 | dentry->d_flags |= DCACHE_LRU_LIST; | |
340 | this_cpu_inc(nr_dentry_unused); | |
341 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
342 | } | |
343 | ||
344 | static void d_lru_del(struct dentry *dentry) | |
345 | { | |
346 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
347 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
348 | this_cpu_dec(nr_dentry_unused); | |
349 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
350 | } | |
351 | ||
352 | static void d_shrink_del(struct dentry *dentry) | |
353 | { | |
354 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
355 | list_del_init(&dentry->d_lru); | |
356 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
357 | this_cpu_dec(nr_dentry_unused); | |
358 | } | |
359 | ||
360 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
361 | { | |
362 | D_FLAG_VERIFY(dentry, 0); | |
363 | list_add(&dentry->d_lru, list); | |
364 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
365 | this_cpu_inc(nr_dentry_unused); | |
366 | } | |
367 | ||
368 | /* | |
369 | * These can only be called under the global LRU lock, ie during the | |
370 | * callback for freeing the LRU list. "isolate" removes it from the | |
371 | * LRU lists entirely, while shrink_move moves it to the indicated | |
372 | * private list. | |
373 | */ | |
374 | static void d_lru_isolate(struct dentry *dentry) | |
375 | { | |
376 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
377 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
378 | this_cpu_dec(nr_dentry_unused); | |
379 | list_del_init(&dentry->d_lru); | |
380 | } | |
381 | ||
382 | static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list) | |
383 | { | |
384 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
385 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
386 | list_move_tail(&dentry->d_lru, list); | |
387 | } | |
388 | ||
da3bbdd4 | 389 | /* |
f6041567 | 390 | * dentry_lru_(add|del)_list) must be called with d_lock held. |
da3bbdd4 KM |
391 | */ |
392 | static void dentry_lru_add(struct dentry *dentry) | |
393 | { | |
89dc77bc LT |
394 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) |
395 | d_lru_add(dentry); | |
da3bbdd4 KM |
396 | } |
397 | ||
789680d1 NP |
398 | /** |
399 | * d_drop - drop a dentry | |
400 | * @dentry: dentry to drop | |
401 | * | |
402 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
403 | * be found through a VFS lookup any more. Note that this is different from | |
404 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
405 | * possible, giving a successful _negative_ lookup, while d_drop will | |
406 | * just make the cache lookup fail. | |
407 | * | |
408 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
409 | * reason (NFS timeouts or autofs deletes). | |
410 | * | |
411 | * __d_drop requires dentry->d_lock. | |
412 | */ | |
413 | void __d_drop(struct dentry *dentry) | |
414 | { | |
dea3667b | 415 | if (!d_unhashed(dentry)) { |
b61625d2 | 416 | struct hlist_bl_head *b; |
7632e465 BF |
417 | /* |
418 | * Hashed dentries are normally on the dentry hashtable, | |
419 | * with the exception of those newly allocated by | |
420 | * d_obtain_alias, which are always IS_ROOT: | |
421 | */ | |
422 | if (unlikely(IS_ROOT(dentry))) | |
b61625d2 AV |
423 | b = &dentry->d_sb->s_anon; |
424 | else | |
425 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
426 | ||
427 | hlist_bl_lock(b); | |
428 | __hlist_bl_del(&dentry->d_hash); | |
429 | dentry->d_hash.pprev = NULL; | |
430 | hlist_bl_unlock(b); | |
dea3667b | 431 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
432 | } |
433 | } | |
434 | EXPORT_SYMBOL(__d_drop); | |
435 | ||
436 | void d_drop(struct dentry *dentry) | |
437 | { | |
789680d1 NP |
438 | spin_lock(&dentry->d_lock); |
439 | __d_drop(dentry); | |
440 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
441 | } |
442 | EXPORT_SYMBOL(d_drop); | |
443 | ||
e55fd011 | 444 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 445 | { |
41edf278 AV |
446 | struct dentry *parent = NULL; |
447 | bool can_free = true; | |
41edf278 | 448 | if (!IS_ROOT(dentry)) |
77812a1e | 449 | parent = dentry->d_parent; |
31e6b01f | 450 | |
0d98439e LT |
451 | /* |
452 | * The dentry is now unrecoverably dead to the world. | |
453 | */ | |
454 | lockref_mark_dead(&dentry->d_lockref); | |
455 | ||
f0023bc6 | 456 | /* |
f0023bc6 SW |
457 | * inform the fs via d_prune that this dentry is about to be |
458 | * unhashed and destroyed. | |
459 | */ | |
590fb51f | 460 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry)) |
61572bb1 YZ |
461 | dentry->d_op->d_prune(dentry); |
462 | ||
01b60351 AV |
463 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
464 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
465 | d_lru_del(dentry); | |
01b60351 | 466 | } |
77812a1e NP |
467 | /* if it was on the hash then remove it */ |
468 | __d_drop(dentry); | |
03b3b889 AV |
469 | list_del(&dentry->d_u.d_child); |
470 | /* | |
471 | * Inform d_walk() that we are no longer attached to the | |
472 | * dentry tree | |
473 | */ | |
474 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
475 | if (parent) | |
476 | spin_unlock(&parent->d_lock); | |
477 | dentry_iput(dentry); | |
478 | /* | |
479 | * dentry_iput drops the locks, at which point nobody (except | |
480 | * transient RCU lookups) can reach this dentry. | |
481 | */ | |
482 | BUG_ON((int)dentry->d_lockref.count > 0); | |
483 | this_cpu_dec(nr_dentry); | |
484 | if (dentry->d_op && dentry->d_op->d_release) | |
485 | dentry->d_op->d_release(dentry); | |
486 | ||
41edf278 AV |
487 | spin_lock(&dentry->d_lock); |
488 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
489 | dentry->d_flags |= DCACHE_MAY_FREE; | |
490 | can_free = false; | |
491 | } | |
492 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
493 | if (likely(can_free)) |
494 | dentry_free(dentry); | |
e55fd011 AV |
495 | } |
496 | ||
497 | /* | |
498 | * Finish off a dentry we've decided to kill. | |
499 | * dentry->d_lock must be held, returns with it unlocked. | |
500 | * If ref is non-zero, then decrement the refcount too. | |
501 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
502 | */ | |
503 | static struct dentry * | |
504 | dentry_kill(struct dentry *dentry, int unlock_on_failure) | |
505 | __releases(dentry->d_lock) | |
506 | { | |
507 | struct inode *inode = dentry->d_inode; | |
508 | struct dentry *parent = NULL; | |
509 | ||
510 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
511 | goto failed; | |
512 | ||
513 | if (!IS_ROOT(dentry)) { | |
514 | parent = dentry->d_parent; | |
515 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
516 | if (inode) | |
517 | spin_unlock(&inode->i_lock); | |
518 | goto failed; | |
519 | } | |
520 | } | |
521 | ||
522 | __dentry_kill(dentry); | |
03b3b889 | 523 | return parent; |
e55fd011 AV |
524 | |
525 | failed: | |
526 | if (unlock_on_failure) { | |
527 | spin_unlock(&dentry->d_lock); | |
528 | cpu_relax(); | |
529 | } | |
530 | return dentry; /* try again with same dentry */ | |
77812a1e NP |
531 | } |
532 | ||
1da177e4 LT |
533 | /* |
534 | * This is dput | |
535 | * | |
536 | * This is complicated by the fact that we do not want to put | |
537 | * dentries that are no longer on any hash chain on the unused | |
538 | * list: we'd much rather just get rid of them immediately. | |
539 | * | |
540 | * However, that implies that we have to traverse the dentry | |
541 | * tree upwards to the parents which might _also_ now be | |
542 | * scheduled for deletion (it may have been only waiting for | |
543 | * its last child to go away). | |
544 | * | |
545 | * This tail recursion is done by hand as we don't want to depend | |
546 | * on the compiler to always get this right (gcc generally doesn't). | |
547 | * Real recursion would eat up our stack space. | |
548 | */ | |
549 | ||
550 | /* | |
551 | * dput - release a dentry | |
552 | * @dentry: dentry to release | |
553 | * | |
554 | * Release a dentry. This will drop the usage count and if appropriate | |
555 | * call the dentry unlink method as well as removing it from the queues and | |
556 | * releasing its resources. If the parent dentries were scheduled for release | |
557 | * they too may now get deleted. | |
1da177e4 | 558 | */ |
1da177e4 LT |
559 | void dput(struct dentry *dentry) |
560 | { | |
8aab6a27 | 561 | if (unlikely(!dentry)) |
1da177e4 LT |
562 | return; |
563 | ||
564 | repeat: | |
98474236 | 565 | if (lockref_put_or_lock(&dentry->d_lockref)) |
1da177e4 | 566 | return; |
1da177e4 | 567 | |
8aab6a27 LT |
568 | /* Unreachable? Get rid of it */ |
569 | if (unlikely(d_unhashed(dentry))) | |
570 | goto kill_it; | |
571 | ||
572 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
1da177e4 | 573 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 574 | goto kill_it; |
1da177e4 | 575 | } |
265ac902 | 576 | |
358eec18 LT |
577 | if (!(dentry->d_flags & DCACHE_REFERENCED)) |
578 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 579 | dentry_lru_add(dentry); |
265ac902 | 580 | |
98474236 | 581 | dentry->d_lockref.count--; |
61f3dee4 | 582 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
583 | return; |
584 | ||
d52b9086 | 585 | kill_it: |
dd1f6b2e | 586 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
587 | if (dentry) |
588 | goto repeat; | |
1da177e4 | 589 | } |
ec4f8605 | 590 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
591 | |
592 | /** | |
593 | * d_invalidate - invalidate a dentry | |
594 | * @dentry: dentry to invalidate | |
595 | * | |
596 | * Try to invalidate the dentry if it turns out to be | |
597 | * possible. If there are other dentries that can be | |
598 | * reached through this one we can't delete it and we | |
599 | * return -EBUSY. On success we return 0. | |
600 | * | |
601 | * no dcache lock. | |
602 | */ | |
603 | ||
604 | int d_invalidate(struct dentry * dentry) | |
605 | { | |
606 | /* | |
607 | * If it's already been dropped, return OK. | |
608 | */ | |
da502956 | 609 | spin_lock(&dentry->d_lock); |
1da177e4 | 610 | if (d_unhashed(dentry)) { |
da502956 | 611 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
612 | return 0; |
613 | } | |
614 | /* | |
615 | * Check whether to do a partial shrink_dcache | |
616 | * to get rid of unused child entries. | |
617 | */ | |
618 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 619 | spin_unlock(&dentry->d_lock); |
1da177e4 | 620 | shrink_dcache_parent(dentry); |
da502956 | 621 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
622 | } |
623 | ||
624 | /* | |
625 | * Somebody else still using it? | |
626 | * | |
627 | * If it's a directory, we can't drop it | |
628 | * for fear of somebody re-populating it | |
629 | * with children (even though dropping it | |
630 | * would make it unreachable from the root, | |
631 | * we might still populate it if it was a | |
632 | * working directory or similar). | |
50e69630 AV |
633 | * We also need to leave mountpoints alone, |
634 | * directory or not. | |
1da177e4 | 635 | */ |
98474236 | 636 | if (dentry->d_lockref.count > 1 && dentry->d_inode) { |
50e69630 | 637 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { |
1da177e4 | 638 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
639 | return -EBUSY; |
640 | } | |
641 | } | |
642 | ||
643 | __d_drop(dentry); | |
644 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
645 | return 0; |
646 | } | |
ec4f8605 | 647 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 648 | |
b5c84bf6 | 649 | /* This must be called with d_lock held */ |
dc0474be | 650 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 651 | { |
98474236 | 652 | dentry->d_lockref.count++; |
23044507 NP |
653 | } |
654 | ||
dc0474be | 655 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 656 | { |
98474236 | 657 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
658 | } |
659 | ||
b7ab39f6 NP |
660 | struct dentry *dget_parent(struct dentry *dentry) |
661 | { | |
df3d0bbc | 662 | int gotref; |
b7ab39f6 NP |
663 | struct dentry *ret; |
664 | ||
df3d0bbc WL |
665 | /* |
666 | * Do optimistic parent lookup without any | |
667 | * locking. | |
668 | */ | |
669 | rcu_read_lock(); | |
670 | ret = ACCESS_ONCE(dentry->d_parent); | |
671 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
672 | rcu_read_unlock(); | |
673 | if (likely(gotref)) { | |
674 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
675 | return ret; | |
676 | dput(ret); | |
677 | } | |
678 | ||
b7ab39f6 | 679 | repeat: |
a734eb45 NP |
680 | /* |
681 | * Don't need rcu_dereference because we re-check it was correct under | |
682 | * the lock. | |
683 | */ | |
684 | rcu_read_lock(); | |
b7ab39f6 | 685 | ret = dentry->d_parent; |
a734eb45 NP |
686 | spin_lock(&ret->d_lock); |
687 | if (unlikely(ret != dentry->d_parent)) { | |
688 | spin_unlock(&ret->d_lock); | |
689 | rcu_read_unlock(); | |
b7ab39f6 NP |
690 | goto repeat; |
691 | } | |
a734eb45 | 692 | rcu_read_unlock(); |
98474236 WL |
693 | BUG_ON(!ret->d_lockref.count); |
694 | ret->d_lockref.count++; | |
b7ab39f6 | 695 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
696 | return ret; |
697 | } | |
698 | EXPORT_SYMBOL(dget_parent); | |
699 | ||
1da177e4 LT |
700 | /** |
701 | * d_find_alias - grab a hashed alias of inode | |
702 | * @inode: inode in question | |
32ba9c3f LT |
703 | * @want_discon: flag, used by d_splice_alias, to request |
704 | * that only a DISCONNECTED alias be returned. | |
1da177e4 LT |
705 | * |
706 | * If inode has a hashed alias, or is a directory and has any alias, | |
707 | * acquire the reference to alias and return it. Otherwise return NULL. | |
708 | * Notice that if inode is a directory there can be only one alias and | |
709 | * it can be unhashed only if it has no children, or if it is the root | |
710 | * of a filesystem. | |
711 | * | |
21c0d8fd | 712 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
32ba9c3f LT |
713 | * any other hashed alias over that one unless @want_discon is set, |
714 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. | |
1da177e4 | 715 | */ |
32ba9c3f | 716 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 717 | { |
da502956 | 718 | struct dentry *alias, *discon_alias; |
1da177e4 | 719 | |
da502956 NP |
720 | again: |
721 | discon_alias = NULL; | |
b67bfe0d | 722 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
da502956 | 723 | spin_lock(&alias->d_lock); |
1da177e4 | 724 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 725 | if (IS_ROOT(alias) && |
da502956 | 726 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 727 | discon_alias = alias; |
32ba9c3f | 728 | } else if (!want_discon) { |
dc0474be | 729 | __dget_dlock(alias); |
da502956 NP |
730 | spin_unlock(&alias->d_lock); |
731 | return alias; | |
732 | } | |
733 | } | |
734 | spin_unlock(&alias->d_lock); | |
735 | } | |
736 | if (discon_alias) { | |
737 | alias = discon_alias; | |
738 | spin_lock(&alias->d_lock); | |
739 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
740 | if (IS_ROOT(alias) && | |
741 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 742 | __dget_dlock(alias); |
da502956 | 743 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
744 | return alias; |
745 | } | |
746 | } | |
da502956 NP |
747 | spin_unlock(&alias->d_lock); |
748 | goto again; | |
1da177e4 | 749 | } |
da502956 | 750 | return NULL; |
1da177e4 LT |
751 | } |
752 | ||
da502956 | 753 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 754 | { |
214fda1f DH |
755 | struct dentry *de = NULL; |
756 | ||
b3d9b7a3 | 757 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 758 | spin_lock(&inode->i_lock); |
32ba9c3f | 759 | de = __d_find_alias(inode, 0); |
873feea0 | 760 | spin_unlock(&inode->i_lock); |
214fda1f | 761 | } |
1da177e4 LT |
762 | return de; |
763 | } | |
ec4f8605 | 764 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
765 | |
766 | /* | |
767 | * Try to kill dentries associated with this inode. | |
768 | * WARNING: you must own a reference to inode. | |
769 | */ | |
770 | void d_prune_aliases(struct inode *inode) | |
771 | { | |
0cdca3f9 | 772 | struct dentry *dentry; |
1da177e4 | 773 | restart: |
873feea0 | 774 | spin_lock(&inode->i_lock); |
b67bfe0d | 775 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 776 | spin_lock(&dentry->d_lock); |
98474236 | 777 | if (!dentry->d_lockref.count) { |
590fb51f YZ |
778 | /* |
779 | * inform the fs via d_prune that this dentry | |
780 | * is about to be unhashed and destroyed. | |
781 | */ | |
782 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && | |
783 | !d_unhashed(dentry)) | |
784 | dentry->d_op->d_prune(dentry); | |
785 | ||
dc0474be | 786 | __dget_dlock(dentry); |
1da177e4 LT |
787 | __d_drop(dentry); |
788 | spin_unlock(&dentry->d_lock); | |
873feea0 | 789 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
790 | dput(dentry); |
791 | goto restart; | |
792 | } | |
793 | spin_unlock(&dentry->d_lock); | |
794 | } | |
873feea0 | 795 | spin_unlock(&inode->i_lock); |
1da177e4 | 796 | } |
ec4f8605 | 797 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 798 | |
3049cfe2 | 799 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 800 | { |
5c47e6d0 | 801 | struct dentry *dentry, *parent; |
da3bbdd4 | 802 | |
60942f2f | 803 | while (!list_empty(list)) { |
ff2fde99 | 804 | struct inode *inode; |
60942f2f | 805 | dentry = list_entry(list->prev, struct dentry, d_lru); |
ec33679d | 806 | spin_lock(&dentry->d_lock); |
dd1f6b2e DC |
807 | /* |
808 | * The dispose list is isolated and dentries are not accounted | |
809 | * to the LRU here, so we can simply remove it from the list | |
810 | * here regardless of whether it is referenced or not. | |
811 | */ | |
89dc77bc | 812 | d_shrink_del(dentry); |
dd1f6b2e | 813 | |
1da177e4 LT |
814 | /* |
815 | * We found an inuse dentry which was not removed from | |
dd1f6b2e | 816 | * the LRU because of laziness during lookup. Do not free it. |
1da177e4 | 817 | */ |
41edf278 | 818 | if ((int)dentry->d_lockref.count > 0) { |
da3bbdd4 | 819 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
820 | continue; |
821 | } | |
77812a1e | 822 | |
64fd72e0 AV |
823 | |
824 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) { | |
825 | bool can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
826 | spin_unlock(&dentry->d_lock); | |
827 | if (can_free) | |
828 | dentry_free(dentry); | |
829 | continue; | |
830 | } | |
831 | ||
ff2fde99 AV |
832 | inode = dentry->d_inode; |
833 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
89dc77bc | 834 | d_shrink_add(dentry, list); |
dd1f6b2e | 835 | spin_unlock(&dentry->d_lock); |
5c47e6d0 | 836 | continue; |
dd1f6b2e | 837 | } |
ff2fde99 AV |
838 | |
839 | parent = NULL; | |
840 | if (!IS_ROOT(dentry)) { | |
841 | parent = dentry->d_parent; | |
842 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
843 | if (inode) | |
844 | spin_unlock(&inode->i_lock); | |
845 | d_shrink_add(dentry, list); | |
846 | spin_unlock(&dentry->d_lock); | |
847 | continue; | |
848 | } | |
849 | } | |
850 | ||
851 | __dentry_kill(dentry); | |
5c47e6d0 AV |
852 | /* |
853 | * We need to prune ancestors too. This is necessary to prevent | |
854 | * quadratic behavior of shrink_dcache_parent(), but is also | |
855 | * expected to be beneficial in reducing dentry cache | |
856 | * fragmentation. | |
857 | */ | |
858 | dentry = parent; | |
859 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) | |
860 | dentry = dentry_kill(dentry, 1); | |
da3bbdd4 | 861 | } |
3049cfe2 CH |
862 | } |
863 | ||
f6041567 DC |
864 | static enum lru_status |
865 | dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg) | |
866 | { | |
867 | struct list_head *freeable = arg; | |
868 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
869 | ||
870 | ||
871 | /* | |
872 | * we are inverting the lru lock/dentry->d_lock here, | |
873 | * so use a trylock. If we fail to get the lock, just skip | |
874 | * it | |
875 | */ | |
876 | if (!spin_trylock(&dentry->d_lock)) | |
877 | return LRU_SKIP; | |
878 | ||
879 | /* | |
880 | * Referenced dentries are still in use. If they have active | |
881 | * counts, just remove them from the LRU. Otherwise give them | |
882 | * another pass through the LRU. | |
883 | */ | |
884 | if (dentry->d_lockref.count) { | |
89dc77bc | 885 | d_lru_isolate(dentry); |
f6041567 DC |
886 | spin_unlock(&dentry->d_lock); |
887 | return LRU_REMOVED; | |
888 | } | |
889 | ||
890 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
891 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
892 | spin_unlock(&dentry->d_lock); | |
893 | ||
894 | /* | |
895 | * The list move itself will be made by the common LRU code. At | |
896 | * this point, we've dropped the dentry->d_lock but keep the | |
897 | * lru lock. This is safe to do, since every list movement is | |
898 | * protected by the lru lock even if both locks are held. | |
899 | * | |
900 | * This is guaranteed by the fact that all LRU management | |
901 | * functions are intermediated by the LRU API calls like | |
902 | * list_lru_add and list_lru_del. List movement in this file | |
903 | * only ever occur through this functions or through callbacks | |
904 | * like this one, that are called from the LRU API. | |
905 | * | |
906 | * The only exceptions to this are functions like | |
907 | * shrink_dentry_list, and code that first checks for the | |
908 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
909 | * operating only with stack provided lists after they are | |
910 | * properly isolated from the main list. It is thus, always a | |
911 | * local access. | |
912 | */ | |
913 | return LRU_ROTATE; | |
914 | } | |
915 | ||
89dc77bc | 916 | d_lru_shrink_move(dentry, freeable); |
f6041567 DC |
917 | spin_unlock(&dentry->d_lock); |
918 | ||
919 | return LRU_REMOVED; | |
920 | } | |
921 | ||
3049cfe2 | 922 | /** |
b48f03b3 DC |
923 | * prune_dcache_sb - shrink the dcache |
924 | * @sb: superblock | |
f6041567 | 925 | * @nr_to_scan : number of entries to try to free |
9b17c623 | 926 | * @nid: which node to scan for freeable entities |
b48f03b3 | 927 | * |
f6041567 | 928 | * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is |
b48f03b3 DC |
929 | * done when we need more memory an called from the superblock shrinker |
930 | * function. | |
3049cfe2 | 931 | * |
b48f03b3 DC |
932 | * This function may fail to free any resources if all the dentries are in |
933 | * use. | |
3049cfe2 | 934 | */ |
9b17c623 DC |
935 | long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan, |
936 | int nid) | |
3049cfe2 | 937 | { |
f6041567 DC |
938 | LIST_HEAD(dispose); |
939 | long freed; | |
3049cfe2 | 940 | |
9b17c623 DC |
941 | freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate, |
942 | &dispose, &nr_to_scan); | |
f6041567 | 943 | shrink_dentry_list(&dispose); |
0a234c6d | 944 | return freed; |
da3bbdd4 | 945 | } |
23044507 | 946 | |
4e717f5c GC |
947 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
948 | spinlock_t *lru_lock, void *arg) | |
dd1f6b2e | 949 | { |
4e717f5c GC |
950 | struct list_head *freeable = arg; |
951 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 952 | |
4e717f5c GC |
953 | /* |
954 | * we are inverting the lru lock/dentry->d_lock here, | |
955 | * so use a trylock. If we fail to get the lock, just skip | |
956 | * it | |
957 | */ | |
958 | if (!spin_trylock(&dentry->d_lock)) | |
959 | return LRU_SKIP; | |
960 | ||
89dc77bc | 961 | d_lru_shrink_move(dentry, freeable); |
4e717f5c | 962 | spin_unlock(&dentry->d_lock); |
ec33679d | 963 | |
4e717f5c | 964 | return LRU_REMOVED; |
da3bbdd4 KM |
965 | } |
966 | ||
4e717f5c | 967 | |
1da177e4 LT |
968 | /** |
969 | * shrink_dcache_sb - shrink dcache for a superblock | |
970 | * @sb: superblock | |
971 | * | |
3049cfe2 CH |
972 | * Shrink the dcache for the specified super block. This is used to free |
973 | * the dcache before unmounting a file system. | |
1da177e4 | 974 | */ |
3049cfe2 | 975 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 976 | { |
4e717f5c GC |
977 | long freed; |
978 | ||
979 | do { | |
980 | LIST_HEAD(dispose); | |
981 | ||
982 | freed = list_lru_walk(&sb->s_dentry_lru, | |
983 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); | |
3049cfe2 | 984 | |
4e717f5c GC |
985 | this_cpu_sub(nr_dentry_unused, freed); |
986 | shrink_dentry_list(&dispose); | |
987 | } while (freed > 0); | |
1da177e4 | 988 | } |
ec4f8605 | 989 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 990 | |
db14fc3a MS |
991 | /** |
992 | * enum d_walk_ret - action to talke during tree walk | |
993 | * @D_WALK_CONTINUE: contrinue walk | |
994 | * @D_WALK_QUIT: quit walk | |
995 | * @D_WALK_NORETRY: quit when retry is needed | |
996 | * @D_WALK_SKIP: skip this dentry and its children | |
997 | */ | |
998 | enum d_walk_ret { | |
999 | D_WALK_CONTINUE, | |
1000 | D_WALK_QUIT, | |
1001 | D_WALK_NORETRY, | |
1002 | D_WALK_SKIP, | |
1003 | }; | |
c826cb7d | 1004 | |
1da177e4 | 1005 | /** |
db14fc3a MS |
1006 | * d_walk - walk the dentry tree |
1007 | * @parent: start of walk | |
1008 | * @data: data passed to @enter() and @finish() | |
1009 | * @enter: callback when first entering the dentry | |
1010 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1011 | * |
db14fc3a | 1012 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1013 | */ |
db14fc3a MS |
1014 | static void d_walk(struct dentry *parent, void *data, |
1015 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1016 | void (*finish)(void *)) | |
1da177e4 | 1017 | { |
949854d0 | 1018 | struct dentry *this_parent; |
1da177e4 | 1019 | struct list_head *next; |
48f5ec21 | 1020 | unsigned seq = 0; |
db14fc3a MS |
1021 | enum d_walk_ret ret; |
1022 | bool retry = true; | |
949854d0 | 1023 | |
58db63d0 | 1024 | again: |
48f5ec21 | 1025 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1026 | this_parent = parent; |
2fd6b7f5 | 1027 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1028 | |
1029 | ret = enter(data, this_parent); | |
1030 | switch (ret) { | |
1031 | case D_WALK_CONTINUE: | |
1032 | break; | |
1033 | case D_WALK_QUIT: | |
1034 | case D_WALK_SKIP: | |
1035 | goto out_unlock; | |
1036 | case D_WALK_NORETRY: | |
1037 | retry = false; | |
1038 | break; | |
1039 | } | |
1da177e4 LT |
1040 | repeat: |
1041 | next = this_parent->d_subdirs.next; | |
1042 | resume: | |
1043 | while (next != &this_parent->d_subdirs) { | |
1044 | struct list_head *tmp = next; | |
5160ee6f | 1045 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1046 | next = tmp->next; |
2fd6b7f5 NP |
1047 | |
1048 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
db14fc3a MS |
1049 | |
1050 | ret = enter(data, dentry); | |
1051 | switch (ret) { | |
1052 | case D_WALK_CONTINUE: | |
1053 | break; | |
1054 | case D_WALK_QUIT: | |
2fd6b7f5 | 1055 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1056 | goto out_unlock; |
1057 | case D_WALK_NORETRY: | |
1058 | retry = false; | |
1059 | break; | |
1060 | case D_WALK_SKIP: | |
1061 | spin_unlock(&dentry->d_lock); | |
1062 | continue; | |
2fd6b7f5 | 1063 | } |
db14fc3a | 1064 | |
1da177e4 | 1065 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1066 | spin_unlock(&this_parent->d_lock); |
1067 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1068 | this_parent = dentry; |
2fd6b7f5 | 1069 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1070 | goto repeat; |
1071 | } | |
2fd6b7f5 | 1072 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1073 | } |
1074 | /* | |
1075 | * All done at this level ... ascend and resume the search. | |
1076 | */ | |
1077 | if (this_parent != parent) { | |
c826cb7d | 1078 | struct dentry *child = this_parent; |
31dec132 AV |
1079 | this_parent = child->d_parent; |
1080 | ||
1081 | rcu_read_lock(); | |
1082 | spin_unlock(&child->d_lock); | |
1083 | spin_lock(&this_parent->d_lock); | |
1084 | ||
1085 | /* | |
1086 | * might go back up the wrong parent if we have had a rename | |
1087 | * or deletion | |
1088 | */ | |
1089 | if (this_parent != child->d_parent || | |
1090 | (child->d_flags & DCACHE_DENTRY_KILLED) || | |
1091 | need_seqretry(&rename_lock, seq)) { | |
1092 | spin_unlock(&this_parent->d_lock); | |
1093 | rcu_read_unlock(); | |
949854d0 | 1094 | goto rename_retry; |
31dec132 AV |
1095 | } |
1096 | rcu_read_unlock(); | |
949854d0 | 1097 | next = child->d_u.d_child.next; |
1da177e4 LT |
1098 | goto resume; |
1099 | } | |
48f5ec21 | 1100 | if (need_seqretry(&rename_lock, seq)) { |
db14fc3a | 1101 | spin_unlock(&this_parent->d_lock); |
949854d0 | 1102 | goto rename_retry; |
db14fc3a MS |
1103 | } |
1104 | if (finish) | |
1105 | finish(data); | |
1106 | ||
1107 | out_unlock: | |
1108 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1109 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1110 | return; |
58db63d0 NP |
1111 | |
1112 | rename_retry: | |
db14fc3a MS |
1113 | if (!retry) |
1114 | return; | |
48f5ec21 | 1115 | seq = 1; |
58db63d0 | 1116 | goto again; |
1da177e4 | 1117 | } |
db14fc3a MS |
1118 | |
1119 | /* | |
1120 | * Search for at least 1 mount point in the dentry's subdirs. | |
1121 | * We descend to the next level whenever the d_subdirs | |
1122 | * list is non-empty and continue searching. | |
1123 | */ | |
1124 | ||
db14fc3a MS |
1125 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) |
1126 | { | |
1127 | int *ret = data; | |
1128 | if (d_mountpoint(dentry)) { | |
1129 | *ret = 1; | |
1130 | return D_WALK_QUIT; | |
1131 | } | |
1132 | return D_WALK_CONTINUE; | |
1133 | } | |
1134 | ||
69c88dc7 RD |
1135 | /** |
1136 | * have_submounts - check for mounts over a dentry | |
1137 | * @parent: dentry to check. | |
1138 | * | |
1139 | * Return true if the parent or its subdirectories contain | |
1140 | * a mount point | |
1141 | */ | |
db14fc3a MS |
1142 | int have_submounts(struct dentry *parent) |
1143 | { | |
1144 | int ret = 0; | |
1145 | ||
1146 | d_walk(parent, &ret, check_mount, NULL); | |
1147 | ||
1148 | return ret; | |
1149 | } | |
ec4f8605 | 1150 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1151 | |
eed81007 MS |
1152 | /* |
1153 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1154 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1155 | * subtree can become unreachable). | |
1156 | * | |
1157 | * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For | |
1158 | * this reason take rename_lock and d_lock on dentry and ancestors. | |
1159 | */ | |
1160 | int d_set_mounted(struct dentry *dentry) | |
1161 | { | |
1162 | struct dentry *p; | |
1163 | int ret = -ENOENT; | |
1164 | write_seqlock(&rename_lock); | |
1165 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1166 | /* Need exclusion wrt. check_submounts_and_drop() */ | |
1167 | spin_lock(&p->d_lock); | |
1168 | if (unlikely(d_unhashed(p))) { | |
1169 | spin_unlock(&p->d_lock); | |
1170 | goto out; | |
1171 | } | |
1172 | spin_unlock(&p->d_lock); | |
1173 | } | |
1174 | spin_lock(&dentry->d_lock); | |
1175 | if (!d_unlinked(dentry)) { | |
1176 | dentry->d_flags |= DCACHE_MOUNTED; | |
1177 | ret = 0; | |
1178 | } | |
1179 | spin_unlock(&dentry->d_lock); | |
1180 | out: | |
1181 | write_sequnlock(&rename_lock); | |
1182 | return ret; | |
1183 | } | |
1184 | ||
1da177e4 | 1185 | /* |
fd517909 | 1186 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1187 | * and move any unused dentries to the end of the unused |
1188 | * list for prune_dcache(). We descend to the next level | |
1189 | * whenever the d_subdirs list is non-empty and continue | |
1190 | * searching. | |
1191 | * | |
1192 | * It returns zero iff there are no unused children, | |
1193 | * otherwise it returns the number of children moved to | |
1194 | * the end of the unused list. This may not be the total | |
1195 | * number of unused children, because select_parent can | |
1196 | * drop the lock and return early due to latency | |
1197 | * constraints. | |
1198 | */ | |
1da177e4 | 1199 | |
db14fc3a MS |
1200 | struct select_data { |
1201 | struct dentry *start; | |
1202 | struct list_head dispose; | |
1203 | int found; | |
1204 | }; | |
23044507 | 1205 | |
db14fc3a MS |
1206 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1207 | { | |
1208 | struct select_data *data = _data; | |
1209 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1210 | |
db14fc3a MS |
1211 | if (data->start == dentry) |
1212 | goto out; | |
2fd6b7f5 | 1213 | |
fe91522a | 1214 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1215 | data->found++; |
fe91522a AV |
1216 | } else { |
1217 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1218 | d_lru_del(dentry); | |
1219 | if (!dentry->d_lockref.count) { | |
1220 | d_shrink_add(dentry, &data->dispose); | |
1221 | data->found++; | |
1222 | } | |
1da177e4 | 1223 | } |
db14fc3a MS |
1224 | /* |
1225 | * We can return to the caller if we have found some (this | |
1226 | * ensures forward progress). We'll be coming back to find | |
1227 | * the rest. | |
1228 | */ | |
fe91522a AV |
1229 | if (!list_empty(&data->dispose)) |
1230 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1231 | out: |
db14fc3a | 1232 | return ret; |
1da177e4 LT |
1233 | } |
1234 | ||
1235 | /** | |
1236 | * shrink_dcache_parent - prune dcache | |
1237 | * @parent: parent of entries to prune | |
1238 | * | |
1239 | * Prune the dcache to remove unused children of the parent dentry. | |
1240 | */ | |
db14fc3a | 1241 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1242 | { |
db14fc3a MS |
1243 | for (;;) { |
1244 | struct select_data data; | |
1da177e4 | 1245 | |
db14fc3a MS |
1246 | INIT_LIST_HEAD(&data.dispose); |
1247 | data.start = parent; | |
1248 | data.found = 0; | |
1249 | ||
1250 | d_walk(parent, &data, select_collect, NULL); | |
1251 | if (!data.found) | |
1252 | break; | |
1253 | ||
1254 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1255 | cond_resched(); |
1256 | } | |
1da177e4 | 1257 | } |
ec4f8605 | 1258 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1259 | |
9c8c10e2 | 1260 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1261 | { |
9c8c10e2 AV |
1262 | /* it has busy descendents; complain about those instead */ |
1263 | if (!list_empty(&dentry->d_subdirs)) | |
1264 | return D_WALK_CONTINUE; | |
42c32608 | 1265 | |
9c8c10e2 AV |
1266 | /* root with refcount 1 is fine */ |
1267 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1268 | return D_WALK_CONTINUE; | |
1269 | ||
1270 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1271 | " still in use (%d) [unmount of %s %s]\n", | |
42c32608 AV |
1272 | dentry, |
1273 | dentry->d_inode ? | |
1274 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1275 | dentry, |
42c32608 AV |
1276 | dentry->d_lockref.count, |
1277 | dentry->d_sb->s_type->name, | |
1278 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1279 | WARN_ON(1); |
1280 | return D_WALK_CONTINUE; | |
1281 | } | |
1282 | ||
1283 | static void do_one_tree(struct dentry *dentry) | |
1284 | { | |
1285 | shrink_dcache_parent(dentry); | |
1286 | d_walk(dentry, dentry, umount_check, NULL); | |
1287 | d_drop(dentry); | |
1288 | dput(dentry); | |
42c32608 AV |
1289 | } |
1290 | ||
1291 | /* | |
1292 | * destroy the dentries attached to a superblock on unmounting | |
1293 | */ | |
1294 | void shrink_dcache_for_umount(struct super_block *sb) | |
1295 | { | |
1296 | struct dentry *dentry; | |
1297 | ||
9c8c10e2 | 1298 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1299 | |
1300 | dentry = sb->s_root; | |
1301 | sb->s_root = NULL; | |
9c8c10e2 | 1302 | do_one_tree(dentry); |
42c32608 AV |
1303 | |
1304 | while (!hlist_bl_empty(&sb->s_anon)) { | |
9c8c10e2 AV |
1305 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash)); |
1306 | do_one_tree(dentry); | |
42c32608 AV |
1307 | } |
1308 | } | |
1309 | ||
848ac114 MS |
1310 | static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry) |
1311 | { | |
1312 | struct select_data *data = _data; | |
1313 | ||
1314 | if (d_mountpoint(dentry)) { | |
1315 | data->found = -EBUSY; | |
1316 | return D_WALK_QUIT; | |
1317 | } | |
1318 | ||
1319 | return select_collect(_data, dentry); | |
1320 | } | |
1321 | ||
1322 | static void check_and_drop(void *_data) | |
1323 | { | |
1324 | struct select_data *data = _data; | |
1325 | ||
1326 | if (d_mountpoint(data->start)) | |
1327 | data->found = -EBUSY; | |
1328 | if (!data->found) | |
1329 | __d_drop(data->start); | |
1330 | } | |
1331 | ||
1332 | /** | |
1333 | * check_submounts_and_drop - prune dcache, check for submounts and drop | |
1334 | * | |
1335 | * All done as a single atomic operation relative to has_unlinked_ancestor(). | |
1336 | * Returns 0 if successfully unhashed @parent. If there were submounts then | |
1337 | * return -EBUSY. | |
1338 | * | |
1339 | * @dentry: dentry to prune and drop | |
1340 | */ | |
1341 | int check_submounts_and_drop(struct dentry *dentry) | |
1342 | { | |
1343 | int ret = 0; | |
1344 | ||
1345 | /* Negative dentries can be dropped without further checks */ | |
1346 | if (!dentry->d_inode) { | |
1347 | d_drop(dentry); | |
1348 | goto out; | |
1349 | } | |
1350 | ||
1351 | for (;;) { | |
1352 | struct select_data data; | |
1353 | ||
1354 | INIT_LIST_HEAD(&data.dispose); | |
1355 | data.start = dentry; | |
1356 | data.found = 0; | |
1357 | ||
1358 | d_walk(dentry, &data, check_and_collect, check_and_drop); | |
1359 | ret = data.found; | |
1360 | ||
1361 | if (!list_empty(&data.dispose)) | |
1362 | shrink_dentry_list(&data.dispose); | |
1363 | ||
1364 | if (ret <= 0) | |
1365 | break; | |
1366 | ||
1367 | cond_resched(); | |
1368 | } | |
1369 | ||
1370 | out: | |
1371 | return ret; | |
1372 | } | |
1373 | EXPORT_SYMBOL(check_submounts_and_drop); | |
1374 | ||
1da177e4 | 1375 | /** |
a4464dbc AV |
1376 | * __d_alloc - allocate a dcache entry |
1377 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1378 | * @name: qstr of the name |
1379 | * | |
1380 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1381 | * available. On a success the dentry is returned. The name passed in is | |
1382 | * copied and the copy passed in may be reused after this call. | |
1383 | */ | |
1384 | ||
a4464dbc | 1385 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1386 | { |
1387 | struct dentry *dentry; | |
1388 | char *dname; | |
1389 | ||
e12ba74d | 1390 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1391 | if (!dentry) |
1392 | return NULL; | |
1393 | ||
6326c71f LT |
1394 | /* |
1395 | * We guarantee that the inline name is always NUL-terminated. | |
1396 | * This way the memcpy() done by the name switching in rename | |
1397 | * will still always have a NUL at the end, even if we might | |
1398 | * be overwriting an internal NUL character | |
1399 | */ | |
1400 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 LT |
1401 | if (name->len > DNAME_INLINE_LEN-1) { |
1402 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1403 | if (!dname) { | |
1404 | kmem_cache_free(dentry_cache, dentry); | |
1405 | return NULL; | |
1406 | } | |
1407 | } else { | |
1408 | dname = dentry->d_iname; | |
1409 | } | |
1da177e4 LT |
1410 | |
1411 | dentry->d_name.len = name->len; | |
1412 | dentry->d_name.hash = name->hash; | |
1413 | memcpy(dname, name->name, name->len); | |
1414 | dname[name->len] = 0; | |
1415 | ||
6326c71f LT |
1416 | /* Make sure we always see the terminating NUL character */ |
1417 | smp_wmb(); | |
1418 | dentry->d_name.name = dname; | |
1419 | ||
98474236 | 1420 | dentry->d_lockref.count = 1; |
dea3667b | 1421 | dentry->d_flags = 0; |
1da177e4 | 1422 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1423 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1424 | dentry->d_inode = NULL; |
a4464dbc AV |
1425 | dentry->d_parent = dentry; |
1426 | dentry->d_sb = sb; | |
1da177e4 LT |
1427 | dentry->d_op = NULL; |
1428 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1429 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1430 | INIT_LIST_HEAD(&dentry->d_lru); |
1431 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
b3d9b7a3 | 1432 | INIT_HLIST_NODE(&dentry->d_alias); |
2fd6b7f5 | 1433 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1434 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1435 | |
3e880fb5 | 1436 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1437 | |
1da177e4 LT |
1438 | return dentry; |
1439 | } | |
a4464dbc AV |
1440 | |
1441 | /** | |
1442 | * d_alloc - allocate a dcache entry | |
1443 | * @parent: parent of entry to allocate | |
1444 | * @name: qstr of the name | |
1445 | * | |
1446 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1447 | * available. On a success the dentry is returned. The name passed in is | |
1448 | * copied and the copy passed in may be reused after this call. | |
1449 | */ | |
1450 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1451 | { | |
1452 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1453 | if (!dentry) | |
1454 | return NULL; | |
1455 | ||
1456 | spin_lock(&parent->d_lock); | |
1457 | /* | |
1458 | * don't need child lock because it is not subject | |
1459 | * to concurrency here | |
1460 | */ | |
1461 | __dget_dlock(parent); | |
1462 | dentry->d_parent = parent; | |
1463 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1464 | spin_unlock(&parent->d_lock); | |
1465 | ||
1466 | return dentry; | |
1467 | } | |
ec4f8605 | 1468 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1469 | |
e1a24bb0 BF |
1470 | /** |
1471 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1472 | * @sb: the superblock | |
1473 | * @name: qstr of the name | |
1474 | * | |
1475 | * For a filesystem that just pins its dentries in memory and never | |
1476 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1477 | */ | |
4b936885 NP |
1478 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1479 | { | |
e1a24bb0 | 1480 | return __d_alloc(sb, name); |
4b936885 NP |
1481 | } |
1482 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1483 | ||
1da177e4 LT |
1484 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1485 | { | |
1486 | struct qstr q; | |
1487 | ||
1488 | q.name = name; | |
1489 | q.len = strlen(name); | |
1490 | q.hash = full_name_hash(q.name, q.len); | |
1491 | return d_alloc(parent, &q); | |
1492 | } | |
ef26ca97 | 1493 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1494 | |
fb045adb NP |
1495 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1496 | { | |
6f7f7caa LT |
1497 | WARN_ON_ONCE(dentry->d_op); |
1498 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1499 | DCACHE_OP_COMPARE | |
1500 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1501 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1502 | DCACHE_OP_DELETE )); |
1503 | dentry->d_op = op; | |
1504 | if (!op) | |
1505 | return; | |
1506 | if (op->d_hash) | |
1507 | dentry->d_flags |= DCACHE_OP_HASH; | |
1508 | if (op->d_compare) | |
1509 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1510 | if (op->d_revalidate) | |
1511 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1512 | if (op->d_weak_revalidate) |
1513 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1514 | if (op->d_delete) |
1515 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1516 | if (op->d_prune) |
1517 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1518 | |
1519 | } | |
1520 | EXPORT_SYMBOL(d_set_d_op); | |
1521 | ||
b18825a7 DH |
1522 | static unsigned d_flags_for_inode(struct inode *inode) |
1523 | { | |
1524 | unsigned add_flags = DCACHE_FILE_TYPE; | |
1525 | ||
1526 | if (!inode) | |
1527 | return DCACHE_MISS_TYPE; | |
1528 | ||
1529 | if (S_ISDIR(inode->i_mode)) { | |
1530 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1531 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1532 | if (unlikely(!inode->i_op->lookup)) | |
1533 | add_flags = DCACHE_AUTODIR_TYPE; | |
1534 | else | |
1535 | inode->i_opflags |= IOP_LOOKUP; | |
1536 | } | |
1537 | } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1538 | if (unlikely(inode->i_op->follow_link)) | |
1539 | add_flags = DCACHE_SYMLINK_TYPE; | |
1540 | else | |
1541 | inode->i_opflags |= IOP_NOFOLLOW; | |
1542 | } | |
1543 | ||
1544 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1545 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1546 | return add_flags; | |
1547 | } | |
1548 | ||
360da900 OH |
1549 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1550 | { | |
b18825a7 DH |
1551 | unsigned add_flags = d_flags_for_inode(inode); |
1552 | ||
b23fb0a6 | 1553 | spin_lock(&dentry->d_lock); |
22213318 | 1554 | __d_set_type(dentry, add_flags); |
b18825a7 | 1555 | if (inode) |
b3d9b7a3 | 1556 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); |
360da900 | 1557 | dentry->d_inode = inode; |
31e6b01f | 1558 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1559 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1560 | fsnotify_d_instantiate(dentry, inode); |
1561 | } | |
1562 | ||
1da177e4 LT |
1563 | /** |
1564 | * d_instantiate - fill in inode information for a dentry | |
1565 | * @entry: dentry to complete | |
1566 | * @inode: inode to attach to this dentry | |
1567 | * | |
1568 | * Fill in inode information in the entry. | |
1569 | * | |
1570 | * This turns negative dentries into productive full members | |
1571 | * of society. | |
1572 | * | |
1573 | * NOTE! This assumes that the inode count has been incremented | |
1574 | * (or otherwise set) by the caller to indicate that it is now | |
1575 | * in use by the dcache. | |
1576 | */ | |
1577 | ||
1578 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1579 | { | |
b3d9b7a3 | 1580 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
873feea0 NP |
1581 | if (inode) |
1582 | spin_lock(&inode->i_lock); | |
360da900 | 1583 | __d_instantiate(entry, inode); |
873feea0 NP |
1584 | if (inode) |
1585 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1586 | security_d_instantiate(entry, inode); |
1587 | } | |
ec4f8605 | 1588 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1589 | |
1590 | /** | |
1591 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1592 | * @entry: dentry to instantiate | |
1593 | * @inode: inode to attach to this dentry | |
1594 | * | |
1595 | * Fill in inode information in the entry. On success, it returns NULL. | |
1596 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1597 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1598 | * |
1599 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1600 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1601 | * |
1602 | * This also assumes that the inode count has been incremented | |
1603 | * (or otherwise set) by the caller to indicate that it is now | |
1604 | * in use by the dcache. | |
1da177e4 | 1605 | */ |
770bfad8 DH |
1606 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1607 | struct inode *inode) | |
1da177e4 LT |
1608 | { |
1609 | struct dentry *alias; | |
1610 | int len = entry->d_name.len; | |
1611 | const char *name = entry->d_name.name; | |
1612 | unsigned int hash = entry->d_name.hash; | |
1613 | ||
770bfad8 | 1614 | if (!inode) { |
360da900 | 1615 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1616 | return NULL; |
1617 | } | |
1618 | ||
b67bfe0d | 1619 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
9abca360 NP |
1620 | /* |
1621 | * Don't need alias->d_lock here, because aliases with | |
1622 | * d_parent == entry->d_parent are not subject to name or | |
1623 | * parent changes, because the parent inode i_mutex is held. | |
1624 | */ | |
12f8ad4b | 1625 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1626 | continue; |
1627 | if (alias->d_parent != entry->d_parent) | |
1628 | continue; | |
ee983e89 LT |
1629 | if (alias->d_name.len != len) |
1630 | continue; | |
12f8ad4b | 1631 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1632 | continue; |
dc0474be | 1633 | __dget(alias); |
1da177e4 LT |
1634 | return alias; |
1635 | } | |
770bfad8 | 1636 | |
360da900 | 1637 | __d_instantiate(entry, inode); |
1da177e4 LT |
1638 | return NULL; |
1639 | } | |
770bfad8 DH |
1640 | |
1641 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1642 | { | |
1643 | struct dentry *result; | |
1644 | ||
b3d9b7a3 | 1645 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
770bfad8 | 1646 | |
873feea0 NP |
1647 | if (inode) |
1648 | spin_lock(&inode->i_lock); | |
770bfad8 | 1649 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1650 | if (inode) |
1651 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1652 | |
1653 | if (!result) { | |
1654 | security_d_instantiate(entry, inode); | |
1655 | return NULL; | |
1656 | } | |
1657 | ||
1658 | BUG_ON(!d_unhashed(result)); | |
1659 | iput(inode); | |
1660 | return result; | |
1661 | } | |
1662 | ||
1da177e4 LT |
1663 | EXPORT_SYMBOL(d_instantiate_unique); |
1664 | ||
b70a80e7 MS |
1665 | /** |
1666 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1667 | * @entry: dentry to complete | |
1668 | * @inode: inode to attach to this dentry | |
1669 | * | |
1670 | * Fill in inode information in the entry. If a directory alias is found, then | |
1671 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1672 | * guarantees that a directory inode may never have more than one alias. | |
1673 | */ | |
1674 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1675 | { | |
1676 | BUG_ON(!hlist_unhashed(&entry->d_alias)); | |
1677 | ||
1678 | spin_lock(&inode->i_lock); | |
1679 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1680 | spin_unlock(&inode->i_lock); | |
1681 | iput(inode); | |
1682 | return -EBUSY; | |
1683 | } | |
1684 | __d_instantiate(entry, inode); | |
1685 | spin_unlock(&inode->i_lock); | |
1686 | security_d_instantiate(entry, inode); | |
1687 | ||
1688 | return 0; | |
1689 | } | |
1690 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1691 | ||
adc0e91a AV |
1692 | struct dentry *d_make_root(struct inode *root_inode) |
1693 | { | |
1694 | struct dentry *res = NULL; | |
1695 | ||
1696 | if (root_inode) { | |
26fe5750 | 1697 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1698 | |
1699 | res = __d_alloc(root_inode->i_sb, &name); | |
1700 | if (res) | |
1701 | d_instantiate(res, root_inode); | |
1702 | else | |
1703 | iput(root_inode); | |
1704 | } | |
1705 | return res; | |
1706 | } | |
1707 | EXPORT_SYMBOL(d_make_root); | |
1708 | ||
d891eedb BF |
1709 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1710 | { | |
1711 | struct dentry *alias; | |
1712 | ||
b3d9b7a3 | 1713 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1714 | return NULL; |
b3d9b7a3 | 1715 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); |
d891eedb BF |
1716 | __dget(alias); |
1717 | return alias; | |
1718 | } | |
1719 | ||
46f72b34 SW |
1720 | /** |
1721 | * d_find_any_alias - find any alias for a given inode | |
1722 | * @inode: inode to find an alias for | |
1723 | * | |
1724 | * If any aliases exist for the given inode, take and return a | |
1725 | * reference for one of them. If no aliases exist, return %NULL. | |
1726 | */ | |
1727 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1728 | { |
1729 | struct dentry *de; | |
1730 | ||
1731 | spin_lock(&inode->i_lock); | |
1732 | de = __d_find_any_alias(inode); | |
1733 | spin_unlock(&inode->i_lock); | |
1734 | return de; | |
1735 | } | |
46f72b34 | 1736 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1737 | |
4ea3ada2 CH |
1738 | /** |
1739 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1740 | * @inode: inode to allocate the dentry for | |
1741 | * | |
1742 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1743 | * similar open by handle operations. The returned dentry may be anonymous, | |
1744 | * or may have a full name (if the inode was already in the cache). | |
1745 | * | |
1746 | * When called on a directory inode, we must ensure that the inode only ever | |
1747 | * has one dentry. If a dentry is found, that is returned instead of | |
1748 | * allocating a new one. | |
1749 | * | |
1750 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1751 | * to the dentry. In case of an error the reference on the inode is released. |
1752 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1753 | * be passed in and will be the error will be propagate to the return value, | |
1754 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1755 | */ |
1756 | struct dentry *d_obtain_alias(struct inode *inode) | |
1757 | { | |
b911a6bd | 1758 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1759 | struct dentry *tmp; |
1760 | struct dentry *res; | |
b18825a7 | 1761 | unsigned add_flags; |
4ea3ada2 CH |
1762 | |
1763 | if (!inode) | |
44003728 | 1764 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1765 | if (IS_ERR(inode)) |
1766 | return ERR_CAST(inode); | |
1767 | ||
d891eedb | 1768 | res = d_find_any_alias(inode); |
9308a612 CH |
1769 | if (res) |
1770 | goto out_iput; | |
1771 | ||
a4464dbc | 1772 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1773 | if (!tmp) { |
1774 | res = ERR_PTR(-ENOMEM); | |
1775 | goto out_iput; | |
4ea3ada2 | 1776 | } |
b5c84bf6 | 1777 | |
873feea0 | 1778 | spin_lock(&inode->i_lock); |
d891eedb | 1779 | res = __d_find_any_alias(inode); |
9308a612 | 1780 | if (res) { |
873feea0 | 1781 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1782 | dput(tmp); |
1783 | goto out_iput; | |
1784 | } | |
1785 | ||
1786 | /* attach a disconnected dentry */ | |
b18825a7 DH |
1787 | add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED; |
1788 | ||
9308a612 | 1789 | spin_lock(&tmp->d_lock); |
9308a612 | 1790 | tmp->d_inode = inode; |
b18825a7 | 1791 | tmp->d_flags |= add_flags; |
b3d9b7a3 | 1792 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1793 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1794 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1795 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1796 | spin_unlock(&tmp->d_lock); |
873feea0 | 1797 | spin_unlock(&inode->i_lock); |
24ff6663 | 1798 | security_d_instantiate(tmp, inode); |
9308a612 | 1799 | |
9308a612 CH |
1800 | return tmp; |
1801 | ||
1802 | out_iput: | |
24ff6663 JB |
1803 | if (res && !IS_ERR(res)) |
1804 | security_d_instantiate(res, inode); | |
9308a612 CH |
1805 | iput(inode); |
1806 | return res; | |
4ea3ada2 | 1807 | } |
adc48720 | 1808 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1809 | |
1810 | /** | |
1811 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1812 | * @inode: the inode which may have a disconnected dentry | |
1813 | * @dentry: a negative dentry which we want to point to the inode. | |
1814 | * | |
1815 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1816 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1817 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1818 | * | |
1819 | * This is needed in the lookup routine of any filesystem that is exportable | |
1820 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1821 | * | |
1822 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1823 | * is returned. This matches the expected return value of ->lookup. | |
1824 | * | |
6d4ade98 SW |
1825 | * Cluster filesystems may call this function with a negative, hashed dentry. |
1826 | * In that case, we know that the inode will be a regular file, and also this | |
1827 | * will only occur during atomic_open. So we need to check for the dentry | |
1828 | * being already hashed only in the final case. | |
1da177e4 LT |
1829 | */ |
1830 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1831 | { | |
1832 | struct dentry *new = NULL; | |
1833 | ||
a9049376 AV |
1834 | if (IS_ERR(inode)) |
1835 | return ERR_CAST(inode); | |
1836 | ||
21c0d8fd | 1837 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1838 | spin_lock(&inode->i_lock); |
32ba9c3f | 1839 | new = __d_find_alias(inode, 1); |
1da177e4 | 1840 | if (new) { |
32ba9c3f | 1841 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); |
873feea0 | 1842 | spin_unlock(&inode->i_lock); |
1da177e4 | 1843 | security_d_instantiate(new, inode); |
1da177e4 LT |
1844 | d_move(new, dentry); |
1845 | iput(inode); | |
1846 | } else { | |
873feea0 | 1847 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1848 | __d_instantiate(dentry, inode); |
873feea0 | 1849 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1850 | security_d_instantiate(dentry, inode); |
1851 | d_rehash(dentry); | |
1852 | } | |
6d4ade98 SW |
1853 | } else { |
1854 | d_instantiate(dentry, inode); | |
1855 | if (d_unhashed(dentry)) | |
1856 | d_rehash(dentry); | |
1857 | } | |
1da177e4 LT |
1858 | return new; |
1859 | } | |
ec4f8605 | 1860 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1861 | |
9403540c BN |
1862 | /** |
1863 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1864 | * @inode: the inode case-insensitive lookup has found | |
1865 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1866 | * @name: the case-exact name to be associated with the returned dentry | |
1867 | * | |
1868 | * This is to avoid filling the dcache with case-insensitive names to the | |
1869 | * same inode, only the actual correct case is stored in the dcache for | |
1870 | * case-insensitive filesystems. | |
1871 | * | |
1872 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1873 | * already exists in in the dcache, use it and return it. | |
1874 | * | |
1875 | * If no entry exists with the exact case name, allocate new dentry with | |
1876 | * the exact case, and return the spliced entry. | |
1877 | */ | |
e45b590b | 1878 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1879 | struct qstr *name) |
1880 | { | |
9403540c BN |
1881 | struct dentry *found; |
1882 | struct dentry *new; | |
1883 | ||
b6520c81 CH |
1884 | /* |
1885 | * First check if a dentry matching the name already exists, | |
1886 | * if not go ahead and create it now. | |
1887 | */ | |
9403540c | 1888 | found = d_hash_and_lookup(dentry->d_parent, name); |
4f522a24 AV |
1889 | if (unlikely(IS_ERR(found))) |
1890 | goto err_out; | |
9403540c BN |
1891 | if (!found) { |
1892 | new = d_alloc(dentry->d_parent, name); | |
1893 | if (!new) { | |
4f522a24 | 1894 | found = ERR_PTR(-ENOMEM); |
9403540c BN |
1895 | goto err_out; |
1896 | } | |
b6520c81 | 1897 | |
9403540c BN |
1898 | found = d_splice_alias(inode, new); |
1899 | if (found) { | |
1900 | dput(new); | |
1901 | return found; | |
1902 | } | |
1903 | return new; | |
1904 | } | |
b6520c81 CH |
1905 | |
1906 | /* | |
1907 | * If a matching dentry exists, and it's not negative use it. | |
1908 | * | |
1909 | * Decrement the reference count to balance the iget() done | |
1910 | * earlier on. | |
1911 | */ | |
9403540c BN |
1912 | if (found->d_inode) { |
1913 | if (unlikely(found->d_inode != inode)) { | |
1914 | /* This can't happen because bad inodes are unhashed. */ | |
1915 | BUG_ON(!is_bad_inode(inode)); | |
1916 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1917 | } | |
9403540c BN |
1918 | iput(inode); |
1919 | return found; | |
1920 | } | |
b6520c81 | 1921 | |
9403540c | 1922 | /* |
9403540c | 1923 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1924 | * already has a dentry. |
9403540c | 1925 | */ |
4513d899 AV |
1926 | new = d_splice_alias(inode, found); |
1927 | if (new) { | |
1928 | dput(found); | |
1929 | found = new; | |
9403540c | 1930 | } |
4513d899 | 1931 | return found; |
9403540c BN |
1932 | |
1933 | err_out: | |
1934 | iput(inode); | |
4f522a24 | 1935 | return found; |
9403540c | 1936 | } |
ec4f8605 | 1937 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1938 | |
12f8ad4b LT |
1939 | /* |
1940 | * Do the slow-case of the dentry name compare. | |
1941 | * | |
1942 | * Unlike the dentry_cmp() function, we need to atomically | |
da53be12 | 1943 | * load the name and length information, so that the |
12f8ad4b LT |
1944 | * filesystem can rely on them, and can use the 'name' and |
1945 | * 'len' information without worrying about walking off the | |
1946 | * end of memory etc. | |
1947 | * | |
1948 | * Thus the read_seqcount_retry() and the "duplicate" info | |
1949 | * in arguments (the low-level filesystem should not look | |
1950 | * at the dentry inode or name contents directly, since | |
1951 | * rename can change them while we're in RCU mode). | |
1952 | */ | |
1953 | enum slow_d_compare { | |
1954 | D_COMP_OK, | |
1955 | D_COMP_NOMATCH, | |
1956 | D_COMP_SEQRETRY, | |
1957 | }; | |
1958 | ||
1959 | static noinline enum slow_d_compare slow_dentry_cmp( | |
1960 | const struct dentry *parent, | |
12f8ad4b LT |
1961 | struct dentry *dentry, |
1962 | unsigned int seq, | |
1963 | const struct qstr *name) | |
1964 | { | |
1965 | int tlen = dentry->d_name.len; | |
1966 | const char *tname = dentry->d_name.name; | |
12f8ad4b LT |
1967 | |
1968 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
1969 | cpu_relax(); | |
1970 | return D_COMP_SEQRETRY; | |
1971 | } | |
da53be12 | 1972 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
12f8ad4b LT |
1973 | return D_COMP_NOMATCH; |
1974 | return D_COMP_OK; | |
1975 | } | |
1976 | ||
31e6b01f NP |
1977 | /** |
1978 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1979 | * @parent: parent dentry | |
1980 | * @name: qstr of name we wish to find | |
1f1e6e52 | 1981 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
1982 | * Returns: dentry, or NULL |
1983 | * | |
1984 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1985 | * resolution (store-free path walking) design described in | |
1986 | * Documentation/filesystems/path-lookup.txt. | |
1987 | * | |
1988 | * This is not to be used outside core vfs. | |
1989 | * | |
1990 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1991 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1992 | * without taking d_lock and checking d_seq sequence count against @seq | |
1993 | * returned here. | |
1994 | * | |
15570086 | 1995 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
1996 | * function. |
1997 | * | |
1998 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1999 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2000 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2001 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2002 | * |
2003 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2004 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2005 | */ |
8966be90 LT |
2006 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2007 | const struct qstr *name, | |
da53be12 | 2008 | unsigned *seqp) |
31e6b01f | 2009 | { |
26fe5750 | 2010 | u64 hashlen = name->hash_len; |
31e6b01f | 2011 | const unsigned char *str = name->name; |
26fe5750 | 2012 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 2013 | struct hlist_bl_node *node; |
31e6b01f NP |
2014 | struct dentry *dentry; |
2015 | ||
2016 | /* | |
2017 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2018 | * required to prevent single threaded performance regressions | |
2019 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2020 | * Keep the two functions in sync. | |
2021 | */ | |
2022 | ||
2023 | /* | |
2024 | * The hash list is protected using RCU. | |
2025 | * | |
2026 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2027 | * races with d_move(). | |
2028 | * | |
2029 | * It is possible that concurrent renames can mess up our list | |
2030 | * walk here and result in missing our dentry, resulting in the | |
2031 | * false-negative result. d_lookup() protects against concurrent | |
2032 | * renames using rename_lock seqlock. | |
2033 | * | |
b0a4bb83 | 2034 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2035 | */ |
b07ad996 | 2036 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2037 | unsigned seq; |
31e6b01f | 2038 | |
31e6b01f | 2039 | seqretry: |
12f8ad4b LT |
2040 | /* |
2041 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2042 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2043 | * |
2044 | * The caller must perform a seqcount check in order | |
da53be12 | 2045 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2046 | * |
2047 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2048 | * we don't wait for the sequence count to stabilize if it | |
2049 | * is in the middle of a sequence change. If we do the slow | |
2050 | * dentry compare, we will do seqretries until it is stable, | |
2051 | * and if we end up with a successful lookup, we actually | |
2052 | * want to exit RCU lookup anyway. | |
2053 | */ | |
2054 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2055 | if (dentry->d_parent != parent) |
2056 | continue; | |
2e321806 LT |
2057 | if (d_unhashed(dentry)) |
2058 | continue; | |
12f8ad4b | 2059 | |
830c0f0e | 2060 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
2061 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2062 | continue; | |
da53be12 LT |
2063 | *seqp = seq; |
2064 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
12f8ad4b LT |
2065 | case D_COMP_OK: |
2066 | return dentry; | |
2067 | case D_COMP_NOMATCH: | |
31e6b01f | 2068 | continue; |
12f8ad4b LT |
2069 | default: |
2070 | goto seqretry; | |
2071 | } | |
31e6b01f | 2072 | } |
12f8ad4b | 2073 | |
26fe5750 | 2074 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 2075 | continue; |
da53be12 | 2076 | *seqp = seq; |
26fe5750 | 2077 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 2078 | return dentry; |
31e6b01f NP |
2079 | } |
2080 | return NULL; | |
2081 | } | |
2082 | ||
1da177e4 LT |
2083 | /** |
2084 | * d_lookup - search for a dentry | |
2085 | * @parent: parent dentry | |
2086 | * @name: qstr of name we wish to find | |
b04f784e | 2087 | * Returns: dentry, or NULL |
1da177e4 | 2088 | * |
b04f784e NP |
2089 | * d_lookup searches the children of the parent dentry for the name in |
2090 | * question. If the dentry is found its reference count is incremented and the | |
2091 | * dentry is returned. The caller must use dput to free the entry when it has | |
2092 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2093 | */ |
da2d8455 | 2094 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2095 | { |
31e6b01f | 2096 | struct dentry *dentry; |
949854d0 | 2097 | unsigned seq; |
1da177e4 LT |
2098 | |
2099 | do { | |
2100 | seq = read_seqbegin(&rename_lock); | |
2101 | dentry = __d_lookup(parent, name); | |
2102 | if (dentry) | |
2103 | break; | |
2104 | } while (read_seqretry(&rename_lock, seq)); | |
2105 | return dentry; | |
2106 | } | |
ec4f8605 | 2107 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2108 | |
31e6b01f | 2109 | /** |
b04f784e NP |
2110 | * __d_lookup - search for a dentry (racy) |
2111 | * @parent: parent dentry | |
2112 | * @name: qstr of name we wish to find | |
2113 | * Returns: dentry, or NULL | |
2114 | * | |
2115 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2116 | * false-negative result due to unrelated rename activity. | |
2117 | * | |
2118 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2119 | * however it must be used carefully, eg. with a following d_lookup in | |
2120 | * the case of failure. | |
2121 | * | |
2122 | * __d_lookup callers must be commented. | |
2123 | */ | |
a713ca2a | 2124 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
2125 | { |
2126 | unsigned int len = name->len; | |
2127 | unsigned int hash = name->hash; | |
2128 | const unsigned char *str = name->name; | |
b07ad996 | 2129 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 2130 | struct hlist_bl_node *node; |
31e6b01f | 2131 | struct dentry *found = NULL; |
665a7583 | 2132 | struct dentry *dentry; |
1da177e4 | 2133 | |
31e6b01f NP |
2134 | /* |
2135 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2136 | * required to prevent single threaded performance regressions | |
2137 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2138 | * Keep the two functions in sync. | |
2139 | */ | |
2140 | ||
b04f784e NP |
2141 | /* |
2142 | * The hash list is protected using RCU. | |
2143 | * | |
2144 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2145 | * with d_move(). | |
2146 | * | |
2147 | * It is possible that concurrent renames can mess up our list | |
2148 | * walk here and result in missing our dentry, resulting in the | |
2149 | * false-negative result. d_lookup() protects against concurrent | |
2150 | * renames using rename_lock seqlock. | |
2151 | * | |
b0a4bb83 | 2152 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2153 | */ |
1da177e4 LT |
2154 | rcu_read_lock(); |
2155 | ||
b07ad996 | 2156 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2157 | |
1da177e4 LT |
2158 | if (dentry->d_name.hash != hash) |
2159 | continue; | |
1da177e4 LT |
2160 | |
2161 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2162 | if (dentry->d_parent != parent) |
2163 | goto next; | |
d0185c08 LT |
2164 | if (d_unhashed(dentry)) |
2165 | goto next; | |
2166 | ||
1da177e4 LT |
2167 | /* |
2168 | * It is safe to compare names since d_move() cannot | |
2169 | * change the qstr (protected by d_lock). | |
2170 | */ | |
fb045adb | 2171 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
2172 | int tlen = dentry->d_name.len; |
2173 | const char *tname = dentry->d_name.name; | |
da53be12 | 2174 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1da177e4 LT |
2175 | goto next; |
2176 | } else { | |
ee983e89 LT |
2177 | if (dentry->d_name.len != len) |
2178 | goto next; | |
12f8ad4b | 2179 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
2180 | goto next; |
2181 | } | |
2182 | ||
98474236 | 2183 | dentry->d_lockref.count++; |
d0185c08 | 2184 | found = dentry; |
1da177e4 LT |
2185 | spin_unlock(&dentry->d_lock); |
2186 | break; | |
2187 | next: | |
2188 | spin_unlock(&dentry->d_lock); | |
2189 | } | |
2190 | rcu_read_unlock(); | |
2191 | ||
2192 | return found; | |
2193 | } | |
2194 | ||
3e7e241f EB |
2195 | /** |
2196 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2197 | * @dir: Directory to search in | |
2198 | * @name: qstr of name we wish to find | |
2199 | * | |
4f522a24 | 2200 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2201 | */ |
2202 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2203 | { | |
3e7e241f EB |
2204 | /* |
2205 | * Check for a fs-specific hash function. Note that we must | |
2206 | * calculate the standard hash first, as the d_op->d_hash() | |
2207 | * routine may choose to leave the hash value unchanged. | |
2208 | */ | |
2209 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2210 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2211 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2212 | if (unlikely(err < 0)) |
2213 | return ERR_PTR(err); | |
3e7e241f | 2214 | } |
4f522a24 | 2215 | return d_lookup(dir, name); |
3e7e241f | 2216 | } |
4f522a24 | 2217 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2218 | |
1da177e4 | 2219 | /** |
786a5e15 | 2220 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2221 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2222 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2223 | * |
2224 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2225 | * This is used by ncpfs in its readdir implementation. | |
2226 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2227 | * |
2228 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2229 | */ |
d3a23e16 | 2230 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2231 | { |
786a5e15 | 2232 | struct dentry *child; |
d3a23e16 | 2233 | |
2fd6b7f5 | 2234 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2235 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2236 | if (dentry == child) { | |
2fd6b7f5 | 2237 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2238 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2239 | spin_unlock(&dentry->d_lock); |
2240 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2241 | return 1; |
2242 | } | |
2243 | } | |
2fd6b7f5 | 2244 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2245 | |
1da177e4 LT |
2246 | return 0; |
2247 | } | |
ec4f8605 | 2248 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2249 | |
2250 | /* | |
2251 | * When a file is deleted, we have two options: | |
2252 | * - turn this dentry into a negative dentry | |
2253 | * - unhash this dentry and free it. | |
2254 | * | |
2255 | * Usually, we want to just turn this into | |
2256 | * a negative dentry, but if anybody else is | |
2257 | * currently using the dentry or the inode | |
2258 | * we can't do that and we fall back on removing | |
2259 | * it from the hash queues and waiting for | |
2260 | * it to be deleted later when it has no users | |
2261 | */ | |
2262 | ||
2263 | /** | |
2264 | * d_delete - delete a dentry | |
2265 | * @dentry: The dentry to delete | |
2266 | * | |
2267 | * Turn the dentry into a negative dentry if possible, otherwise | |
2268 | * remove it from the hash queues so it can be deleted later | |
2269 | */ | |
2270 | ||
2271 | void d_delete(struct dentry * dentry) | |
2272 | { | |
873feea0 | 2273 | struct inode *inode; |
7a91bf7f | 2274 | int isdir = 0; |
1da177e4 LT |
2275 | /* |
2276 | * Are we the only user? | |
2277 | */ | |
357f8e65 | 2278 | again: |
1da177e4 | 2279 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2280 | inode = dentry->d_inode; |
2281 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2282 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2283 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2284 | spin_unlock(&dentry->d_lock); |
2285 | cpu_relax(); | |
2286 | goto again; | |
2287 | } | |
13e3c5e5 | 2288 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2289 | dentry_unlink_inode(dentry); |
7a91bf7f | 2290 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2291 | return; |
2292 | } | |
2293 | ||
2294 | if (!d_unhashed(dentry)) | |
2295 | __d_drop(dentry); | |
2296 | ||
2297 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2298 | |
2299 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2300 | } |
ec4f8605 | 2301 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2302 | |
b07ad996 | 2303 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2304 | { |
ceb5bdc2 | 2305 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2306 | hlist_bl_lock(b); |
dea3667b | 2307 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2308 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2309 | hlist_bl_unlock(b); |
1da177e4 LT |
2310 | } |
2311 | ||
770bfad8 DH |
2312 | static void _d_rehash(struct dentry * entry) |
2313 | { | |
2314 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2315 | } | |
2316 | ||
1da177e4 LT |
2317 | /** |
2318 | * d_rehash - add an entry back to the hash | |
2319 | * @entry: dentry to add to the hash | |
2320 | * | |
2321 | * Adds a dentry to the hash according to its name. | |
2322 | */ | |
2323 | ||
2324 | void d_rehash(struct dentry * entry) | |
2325 | { | |
1da177e4 | 2326 | spin_lock(&entry->d_lock); |
770bfad8 | 2327 | _d_rehash(entry); |
1da177e4 | 2328 | spin_unlock(&entry->d_lock); |
1da177e4 | 2329 | } |
ec4f8605 | 2330 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2331 | |
fb2d5b86 NP |
2332 | /** |
2333 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2334 | * @dentry: dentry to be updated | |
2335 | * @name: new name | |
2336 | * | |
2337 | * Update a case insensitive dentry with new case of name. | |
2338 | * | |
2339 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2340 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2341 | * lengths). | |
2342 | * | |
2343 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2344 | * keep renames and concurrent inserts, and readdir(2) away). | |
2345 | */ | |
2346 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2347 | { | |
7ebfa57f | 2348 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2349 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2350 | ||
fb2d5b86 | 2351 | spin_lock(&dentry->d_lock); |
31e6b01f | 2352 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2353 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2354 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2355 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2356 | } |
2357 | EXPORT_SYMBOL(dentry_update_name_case); | |
2358 | ||
1da177e4 LT |
2359 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2360 | { | |
2361 | if (dname_external(target)) { | |
2362 | if (dname_external(dentry)) { | |
2363 | /* | |
2364 | * Both external: swap the pointers | |
2365 | */ | |
9a8d5bb4 | 2366 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2367 | } else { |
2368 | /* | |
2369 | * dentry:internal, target:external. Steal target's | |
2370 | * storage and make target internal. | |
2371 | */ | |
321bcf92 BF |
2372 | memcpy(target->d_iname, dentry->d_name.name, |
2373 | dentry->d_name.len + 1); | |
1da177e4 LT |
2374 | dentry->d_name.name = target->d_name.name; |
2375 | target->d_name.name = target->d_iname; | |
2376 | } | |
2377 | } else { | |
2378 | if (dname_external(dentry)) { | |
2379 | /* | |
2380 | * dentry:external, target:internal. Give dentry's | |
2381 | * storage to target and make dentry internal | |
2382 | */ | |
2383 | memcpy(dentry->d_iname, target->d_name.name, | |
2384 | target->d_name.len + 1); | |
2385 | target->d_name.name = dentry->d_name.name; | |
2386 | dentry->d_name.name = dentry->d_iname; | |
2387 | } else { | |
2388 | /* | |
da1ce067 | 2389 | * Both are internal. |
1da177e4 | 2390 | */ |
da1ce067 MS |
2391 | unsigned int i; |
2392 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
2393 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { | |
2394 | swap(((long *) &dentry->d_iname)[i], | |
2395 | ((long *) &target->d_iname)[i]); | |
2396 | } | |
1da177e4 LT |
2397 | } |
2398 | } | |
9a8d5bb4 | 2399 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2400 | } |
2401 | ||
2fd6b7f5 NP |
2402 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2403 | { | |
2404 | /* | |
2405 | * XXXX: do we really need to take target->d_lock? | |
2406 | */ | |
2407 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2408 | spin_lock(&target->d_parent->d_lock); | |
2409 | else { | |
2410 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2411 | spin_lock(&dentry->d_parent->d_lock); | |
2412 | spin_lock_nested(&target->d_parent->d_lock, | |
2413 | DENTRY_D_LOCK_NESTED); | |
2414 | } else { | |
2415 | spin_lock(&target->d_parent->d_lock); | |
2416 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2417 | DENTRY_D_LOCK_NESTED); | |
2418 | } | |
2419 | } | |
2420 | if (target < dentry) { | |
2421 | spin_lock_nested(&target->d_lock, 2); | |
2422 | spin_lock_nested(&dentry->d_lock, 3); | |
2423 | } else { | |
2424 | spin_lock_nested(&dentry->d_lock, 2); | |
2425 | spin_lock_nested(&target->d_lock, 3); | |
2426 | } | |
2427 | } | |
2428 | ||
2429 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2430 | struct dentry *target) | |
2431 | { | |
2432 | if (target->d_parent != dentry->d_parent) | |
2433 | spin_unlock(&dentry->d_parent->d_lock); | |
2434 | if (target->d_parent != target) | |
2435 | spin_unlock(&target->d_parent->d_lock); | |
2436 | } | |
2437 | ||
1da177e4 | 2438 | /* |
2fd6b7f5 NP |
2439 | * When switching names, the actual string doesn't strictly have to |
2440 | * be preserved in the target - because we're dropping the target | |
2441 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2442 | * the new name before we switch. | |
2443 | * | |
2444 | * Note that we have to be a lot more careful about getting the hash | |
2445 | * switched - we have to switch the hash value properly even if it | |
2446 | * then no longer matches the actual (corrupted) string of the target. | |
2447 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2448 | */ |
9eaef27b | 2449 | /* |
18367501 | 2450 | * __d_move - move a dentry |
1da177e4 LT |
2451 | * @dentry: entry to move |
2452 | * @target: new dentry | |
da1ce067 | 2453 | * @exchange: exchange the two dentries |
1da177e4 LT |
2454 | * |
2455 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2456 | * dcache entries should not be moved in this way. Caller must hold |
2457 | * rename_lock, the i_mutex of the source and target directories, | |
2458 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2459 | */ |
da1ce067 MS |
2460 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2461 | bool exchange) | |
1da177e4 | 2462 | { |
1da177e4 LT |
2463 | if (!dentry->d_inode) |
2464 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2465 | ||
2fd6b7f5 NP |
2466 | BUG_ON(d_ancestor(dentry, target)); |
2467 | BUG_ON(d_ancestor(target, dentry)); | |
2468 | ||
2fd6b7f5 | 2469 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2470 | |
31e6b01f | 2471 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2472 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2473 | |
ceb5bdc2 NP |
2474 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2475 | ||
2476 | /* | |
2477 | * Move the dentry to the target hash queue. Don't bother checking | |
2478 | * for the same hash queue because of how unlikely it is. | |
2479 | */ | |
2480 | __d_drop(dentry); | |
789680d1 | 2481 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 | 2482 | |
da1ce067 MS |
2483 | /* |
2484 | * Unhash the target (d_delete() is not usable here). If exchanging | |
2485 | * the two dentries, then rehash onto the other's hash queue. | |
2486 | */ | |
1da177e4 | 2487 | __d_drop(target); |
da1ce067 MS |
2488 | if (exchange) { |
2489 | __d_rehash(target, | |
2490 | d_hash(dentry->d_parent, dentry->d_name.hash)); | |
2491 | } | |
1da177e4 | 2492 | |
5160ee6f ED |
2493 | list_del(&dentry->d_u.d_child); |
2494 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2495 | |
2496 | /* Switch the names.. */ | |
2497 | switch_names(dentry, target); | |
9a8d5bb4 | 2498 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2499 | |
2500 | /* ... and switch the parents */ | |
2501 | if (IS_ROOT(dentry)) { | |
2502 | dentry->d_parent = target->d_parent; | |
2503 | target->d_parent = target; | |
5160ee6f | 2504 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2505 | } else { |
9a8d5bb4 | 2506 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2507 | |
2508 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2509 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2510 | } |
2511 | ||
5160ee6f | 2512 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2513 | |
31e6b01f NP |
2514 | write_seqcount_end(&target->d_seq); |
2515 | write_seqcount_end(&dentry->d_seq); | |
2516 | ||
2fd6b7f5 | 2517 | dentry_unlock_parents_for_move(dentry, target); |
da1ce067 MS |
2518 | if (exchange) |
2519 | fsnotify_d_move(target); | |
1da177e4 | 2520 | spin_unlock(&target->d_lock); |
c32ccd87 | 2521 | fsnotify_d_move(dentry); |
1da177e4 | 2522 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2523 | } |
2524 | ||
2525 | /* | |
2526 | * d_move - move a dentry | |
2527 | * @dentry: entry to move | |
2528 | * @target: new dentry | |
2529 | * | |
2530 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2531 | * dcache entries should not be moved in this way. See the locking |
2532 | * requirements for __d_move. | |
18367501 AV |
2533 | */ |
2534 | void d_move(struct dentry *dentry, struct dentry *target) | |
2535 | { | |
2536 | write_seqlock(&rename_lock); | |
da1ce067 | 2537 | __d_move(dentry, target, false); |
1da177e4 | 2538 | write_sequnlock(&rename_lock); |
9eaef27b | 2539 | } |
ec4f8605 | 2540 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2541 | |
da1ce067 MS |
2542 | /* |
2543 | * d_exchange - exchange two dentries | |
2544 | * @dentry1: first dentry | |
2545 | * @dentry2: second dentry | |
2546 | */ | |
2547 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2548 | { | |
2549 | write_seqlock(&rename_lock); | |
2550 | ||
2551 | WARN_ON(!dentry1->d_inode); | |
2552 | WARN_ON(!dentry2->d_inode); | |
2553 | WARN_ON(IS_ROOT(dentry1)); | |
2554 | WARN_ON(IS_ROOT(dentry2)); | |
2555 | ||
2556 | __d_move(dentry1, dentry2, true); | |
2557 | ||
2558 | write_sequnlock(&rename_lock); | |
2559 | } | |
2560 | ||
e2761a11 OH |
2561 | /** |
2562 | * d_ancestor - search for an ancestor | |
2563 | * @p1: ancestor dentry | |
2564 | * @p2: child dentry | |
2565 | * | |
2566 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2567 | * an ancestor of p2, else NULL. | |
9eaef27b | 2568 | */ |
e2761a11 | 2569 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2570 | { |
2571 | struct dentry *p; | |
2572 | ||
871c0067 | 2573 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2574 | if (p->d_parent == p1) |
e2761a11 | 2575 | return p; |
9eaef27b | 2576 | } |
e2761a11 | 2577 | return NULL; |
9eaef27b TM |
2578 | } |
2579 | ||
2580 | /* | |
2581 | * This helper attempts to cope with remotely renamed directories | |
2582 | * | |
2583 | * It assumes that the caller is already holding | |
18367501 | 2584 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2585 | * |
2586 | * Note: If ever the locking in lock_rename() changes, then please | |
2587 | * remember to update this too... | |
9eaef27b | 2588 | */ |
873feea0 NP |
2589 | static struct dentry *__d_unalias(struct inode *inode, |
2590 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2591 | { |
2592 | struct mutex *m1 = NULL, *m2 = NULL; | |
ee3efa91 | 2593 | struct dentry *ret = ERR_PTR(-EBUSY); |
9eaef27b TM |
2594 | |
2595 | /* If alias and dentry share a parent, then no extra locks required */ | |
2596 | if (alias->d_parent == dentry->d_parent) | |
2597 | goto out_unalias; | |
2598 | ||
9eaef27b | 2599 | /* See lock_rename() */ |
9eaef27b TM |
2600 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2601 | goto out_err; | |
2602 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2603 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2604 | goto out_err; | |
2605 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2606 | out_unalias: | |
ee3efa91 | 2607 | if (likely(!d_mountpoint(alias))) { |
da1ce067 | 2608 | __d_move(alias, dentry, false); |
ee3efa91 AV |
2609 | ret = alias; |
2610 | } | |
9eaef27b | 2611 | out_err: |
873feea0 | 2612 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2613 | if (m2) |
2614 | mutex_unlock(m2); | |
2615 | if (m1) | |
2616 | mutex_unlock(m1); | |
2617 | return ret; | |
2618 | } | |
2619 | ||
770bfad8 DH |
2620 | /* |
2621 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2622 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2623 | * returns with anon->d_lock held! |
770bfad8 DH |
2624 | */ |
2625 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2626 | { | |
740da42e | 2627 | struct dentry *dparent; |
770bfad8 | 2628 | |
2fd6b7f5 | 2629 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2630 | |
31e6b01f | 2631 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2632 | write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2633 | |
770bfad8 | 2634 | dparent = dentry->d_parent; |
770bfad8 | 2635 | |
2fd6b7f5 NP |
2636 | switch_names(dentry, anon); |
2637 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2638 | ||
740da42e AV |
2639 | dentry->d_parent = dentry; |
2640 | list_del_init(&dentry->d_u.d_child); | |
2641 | anon->d_parent = dparent; | |
9ed53b12 | 2642 | list_move(&anon->d_u.d_child, &dparent->d_subdirs); |
770bfad8 | 2643 | |
31e6b01f NP |
2644 | write_seqcount_end(&dentry->d_seq); |
2645 | write_seqcount_end(&anon->d_seq); | |
2646 | ||
2fd6b7f5 NP |
2647 | dentry_unlock_parents_for_move(anon, dentry); |
2648 | spin_unlock(&dentry->d_lock); | |
2649 | ||
2650 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2651 | } |
2652 | ||
2653 | /** | |
2654 | * d_materialise_unique - introduce an inode into the tree | |
2655 | * @dentry: candidate dentry | |
2656 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2657 | * | |
2658 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2659 | * root directory alias in its place if there is one. Caller must hold the |
2660 | * i_mutex of the parent directory. | |
770bfad8 DH |
2661 | */ |
2662 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2663 | { | |
9eaef27b | 2664 | struct dentry *actual; |
770bfad8 DH |
2665 | |
2666 | BUG_ON(!d_unhashed(dentry)); | |
2667 | ||
770bfad8 DH |
2668 | if (!inode) { |
2669 | actual = dentry; | |
360da900 | 2670 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2671 | d_rehash(actual); |
2672 | goto out_nolock; | |
770bfad8 DH |
2673 | } |
2674 | ||
873feea0 | 2675 | spin_lock(&inode->i_lock); |
357f8e65 | 2676 | |
9eaef27b TM |
2677 | if (S_ISDIR(inode->i_mode)) { |
2678 | struct dentry *alias; | |
2679 | ||
2680 | /* Does an aliased dentry already exist? */ | |
32ba9c3f | 2681 | alias = __d_find_alias(inode, 0); |
9eaef27b TM |
2682 | if (alias) { |
2683 | actual = alias; | |
18367501 AV |
2684 | write_seqlock(&rename_lock); |
2685 | ||
2686 | if (d_ancestor(alias, dentry)) { | |
2687 | /* Check for loops */ | |
2688 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2689 | spin_unlock(&inode->i_lock); |
18367501 AV |
2690 | } else if (IS_ROOT(alias)) { |
2691 | /* Is this an anonymous mountpoint that we | |
2692 | * could splice into our tree? */ | |
9eaef27b | 2693 | __d_materialise_dentry(dentry, alias); |
18367501 | 2694 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2695 | __d_drop(alias); |
2696 | goto found; | |
18367501 AV |
2697 | } else { |
2698 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2699 | * aliasing. This drops inode->i_lock */ |
18367501 | 2700 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2701 | } |
18367501 | 2702 | write_sequnlock(&rename_lock); |
dd179946 DH |
2703 | if (IS_ERR(actual)) { |
2704 | if (PTR_ERR(actual) == -ELOOP) | |
2705 | pr_warn_ratelimited( | |
2706 | "VFS: Lookup of '%s' in %s %s" | |
2707 | " would have caused loop\n", | |
2708 | dentry->d_name.name, | |
2709 | inode->i_sb->s_type->name, | |
2710 | inode->i_sb->s_id); | |
9eaef27b | 2711 | dput(alias); |
dd179946 | 2712 | } |
9eaef27b TM |
2713 | goto out_nolock; |
2714 | } | |
770bfad8 DH |
2715 | } |
2716 | ||
2717 | /* Add a unique reference */ | |
2718 | actual = __d_instantiate_unique(dentry, inode); | |
2719 | if (!actual) | |
2720 | actual = dentry; | |
357f8e65 NP |
2721 | else |
2722 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2723 | |
770bfad8 DH |
2724 | spin_lock(&actual->d_lock); |
2725 | found: | |
2726 | _d_rehash(actual); | |
2727 | spin_unlock(&actual->d_lock); | |
873feea0 | 2728 | spin_unlock(&inode->i_lock); |
9eaef27b | 2729 | out_nolock: |
770bfad8 DH |
2730 | if (actual == dentry) { |
2731 | security_d_instantiate(dentry, inode); | |
2732 | return NULL; | |
2733 | } | |
2734 | ||
2735 | iput(inode); | |
2736 | return actual; | |
770bfad8 | 2737 | } |
ec4f8605 | 2738 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2739 | |
cdd16d02 | 2740 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2741 | { |
2742 | *buflen -= namelen; | |
2743 | if (*buflen < 0) | |
2744 | return -ENAMETOOLONG; | |
2745 | *buffer -= namelen; | |
2746 | memcpy(*buffer, str, namelen); | |
2747 | return 0; | |
2748 | } | |
2749 | ||
232d2d60 WL |
2750 | /** |
2751 | * prepend_name - prepend a pathname in front of current buffer pointer | |
18129977 WL |
2752 | * @buffer: buffer pointer |
2753 | * @buflen: allocated length of the buffer | |
2754 | * @name: name string and length qstr structure | |
232d2d60 WL |
2755 | * |
2756 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
2757 | * make sure that either the old or the new name pointer and length are | |
2758 | * fetched. However, there may be mismatch between length and pointer. | |
2759 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
2760 | * the length is reached or a null byte is found. It also prepends "/" at | |
2761 | * the beginning of the name. The sequence number check at the caller will | |
2762 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
2763 | * due to mismatched pointer and length will be discarded. | |
2764 | */ | |
cdd16d02 MS |
2765 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2766 | { | |
232d2d60 WL |
2767 | const char *dname = ACCESS_ONCE(name->name); |
2768 | u32 dlen = ACCESS_ONCE(name->len); | |
2769 | char *p; | |
2770 | ||
232d2d60 | 2771 | *buflen -= dlen + 1; |
e825196d AV |
2772 | if (*buflen < 0) |
2773 | return -ENAMETOOLONG; | |
232d2d60 WL |
2774 | p = *buffer -= dlen + 1; |
2775 | *p++ = '/'; | |
2776 | while (dlen--) { | |
2777 | char c = *dname++; | |
2778 | if (!c) | |
2779 | break; | |
2780 | *p++ = c; | |
2781 | } | |
2782 | return 0; | |
cdd16d02 MS |
2783 | } |
2784 | ||
1da177e4 | 2785 | /** |
208898c1 | 2786 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2787 | * @path: the dentry/vfsmount to report |
02125a82 | 2788 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2789 | * @buffer: pointer to the end of the buffer |
2790 | * @buflen: pointer to buffer length | |
552ce544 | 2791 | * |
18129977 WL |
2792 | * The function will first try to write out the pathname without taking any |
2793 | * lock other than the RCU read lock to make sure that dentries won't go away. | |
2794 | * It only checks the sequence number of the global rename_lock as any change | |
2795 | * in the dentry's d_seq will be preceded by changes in the rename_lock | |
2796 | * sequence number. If the sequence number had been changed, it will restart | |
2797 | * the whole pathname back-tracing sequence again by taking the rename_lock. | |
2798 | * In this case, there is no need to take the RCU read lock as the recursive | |
2799 | * parent pointer references will keep the dentry chain alive as long as no | |
2800 | * rename operation is performed. | |
1da177e4 | 2801 | */ |
02125a82 AV |
2802 | static int prepend_path(const struct path *path, |
2803 | const struct path *root, | |
f2eb6575 | 2804 | char **buffer, int *buflen) |
1da177e4 | 2805 | { |
ede4cebc AV |
2806 | struct dentry *dentry; |
2807 | struct vfsmount *vfsmnt; | |
2808 | struct mount *mnt; | |
f2eb6575 | 2809 | int error = 0; |
48a066e7 | 2810 | unsigned seq, m_seq = 0; |
232d2d60 WL |
2811 | char *bptr; |
2812 | int blen; | |
6092d048 | 2813 | |
48f5ec21 | 2814 | rcu_read_lock(); |
48a066e7 AV |
2815 | restart_mnt: |
2816 | read_seqbegin_or_lock(&mount_lock, &m_seq); | |
2817 | seq = 0; | |
4ec6c2ae | 2818 | rcu_read_lock(); |
232d2d60 WL |
2819 | restart: |
2820 | bptr = *buffer; | |
2821 | blen = *buflen; | |
48a066e7 | 2822 | error = 0; |
ede4cebc AV |
2823 | dentry = path->dentry; |
2824 | vfsmnt = path->mnt; | |
2825 | mnt = real_mount(vfsmnt); | |
232d2d60 | 2826 | read_seqbegin_or_lock(&rename_lock, &seq); |
f2eb6575 | 2827 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2828 | struct dentry * parent; |
2829 | ||
1da177e4 | 2830 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
48a066e7 | 2831 | struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); |
552ce544 | 2832 | /* Global root? */ |
48a066e7 AV |
2833 | if (mnt != parent) { |
2834 | dentry = ACCESS_ONCE(mnt->mnt_mountpoint); | |
2835 | mnt = parent; | |
232d2d60 WL |
2836 | vfsmnt = &mnt->mnt; |
2837 | continue; | |
2838 | } | |
2839 | /* | |
2840 | * Filesystems needing to implement special "root names" | |
2841 | * should do so with ->d_dname() | |
2842 | */ | |
2843 | if (IS_ROOT(dentry) && | |
2844 | (dentry->d_name.len != 1 || | |
2845 | dentry->d_name.name[0] != '/')) { | |
2846 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2847 | (int) dentry->d_name.len, | |
2848 | dentry->d_name.name); | |
2849 | } | |
2850 | if (!error) | |
2851 | error = is_mounted(vfsmnt) ? 1 : 2; | |
2852 | break; | |
1da177e4 LT |
2853 | } |
2854 | parent = dentry->d_parent; | |
2855 | prefetch(parent); | |
232d2d60 | 2856 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
2857 | if (error) |
2858 | break; | |
2859 | ||
1da177e4 LT |
2860 | dentry = parent; |
2861 | } | |
48f5ec21 AV |
2862 | if (!(seq & 1)) |
2863 | rcu_read_unlock(); | |
2864 | if (need_seqretry(&rename_lock, seq)) { | |
2865 | seq = 1; | |
232d2d60 | 2866 | goto restart; |
48f5ec21 AV |
2867 | } |
2868 | done_seqretry(&rename_lock, seq); | |
4ec6c2ae LZ |
2869 | |
2870 | if (!(m_seq & 1)) | |
2871 | rcu_read_unlock(); | |
48a066e7 AV |
2872 | if (need_seqretry(&mount_lock, m_seq)) { |
2873 | m_seq = 1; | |
2874 | goto restart_mnt; | |
2875 | } | |
2876 | done_seqretry(&mount_lock, m_seq); | |
1da177e4 | 2877 | |
232d2d60 WL |
2878 | if (error >= 0 && bptr == *buffer) { |
2879 | if (--blen < 0) | |
2880 | error = -ENAMETOOLONG; | |
2881 | else | |
2882 | *--bptr = '/'; | |
2883 | } | |
2884 | *buffer = bptr; | |
2885 | *buflen = blen; | |
7ea600b5 | 2886 | return error; |
f2eb6575 | 2887 | } |
be285c71 | 2888 | |
f2eb6575 MS |
2889 | /** |
2890 | * __d_path - return the path of a dentry | |
2891 | * @path: the dentry/vfsmount to report | |
02125a82 | 2892 | * @root: root vfsmnt/dentry |
cd956a1c | 2893 | * @buf: buffer to return value in |
f2eb6575 MS |
2894 | * @buflen: buffer length |
2895 | * | |
ffd1f4ed | 2896 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2897 | * |
2898 | * Returns a pointer into the buffer or an error code if the | |
2899 | * path was too long. | |
2900 | * | |
be148247 | 2901 | * "buflen" should be positive. |
f2eb6575 | 2902 | * |
02125a82 | 2903 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2904 | */ |
02125a82 AV |
2905 | char *__d_path(const struct path *path, |
2906 | const struct path *root, | |
f2eb6575 MS |
2907 | char *buf, int buflen) |
2908 | { | |
2909 | char *res = buf + buflen; | |
2910 | int error; | |
2911 | ||
2912 | prepend(&res, &buflen, "\0", 1); | |
f2eb6575 | 2913 | error = prepend_path(path, root, &res, &buflen); |
be148247 | 2914 | |
02125a82 AV |
2915 | if (error < 0) |
2916 | return ERR_PTR(error); | |
2917 | if (error > 0) | |
2918 | return NULL; | |
2919 | return res; | |
2920 | } | |
2921 | ||
2922 | char *d_absolute_path(const struct path *path, | |
2923 | char *buf, int buflen) | |
2924 | { | |
2925 | struct path root = {}; | |
2926 | char *res = buf + buflen; | |
2927 | int error; | |
2928 | ||
2929 | prepend(&res, &buflen, "\0", 1); | |
02125a82 | 2930 | error = prepend_path(path, &root, &res, &buflen); |
02125a82 AV |
2931 | |
2932 | if (error > 1) | |
2933 | error = -EINVAL; | |
2934 | if (error < 0) | |
f2eb6575 | 2935 | return ERR_PTR(error); |
f2eb6575 | 2936 | return res; |
1da177e4 LT |
2937 | } |
2938 | ||
ffd1f4ed MS |
2939 | /* |
2940 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2941 | */ | |
02125a82 AV |
2942 | static int path_with_deleted(const struct path *path, |
2943 | const struct path *root, | |
2944 | char **buf, int *buflen) | |
ffd1f4ed MS |
2945 | { |
2946 | prepend(buf, buflen, "\0", 1); | |
2947 | if (d_unlinked(path->dentry)) { | |
2948 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2949 | if (error) | |
2950 | return error; | |
2951 | } | |
2952 | ||
2953 | return prepend_path(path, root, buf, buflen); | |
2954 | } | |
2955 | ||
8df9d1a4 MS |
2956 | static int prepend_unreachable(char **buffer, int *buflen) |
2957 | { | |
2958 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2959 | } | |
2960 | ||
68f0d9d9 LT |
2961 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) |
2962 | { | |
2963 | unsigned seq; | |
2964 | ||
2965 | do { | |
2966 | seq = read_seqcount_begin(&fs->seq); | |
2967 | *root = fs->root; | |
2968 | } while (read_seqcount_retry(&fs->seq, seq)); | |
2969 | } | |
2970 | ||
a03a8a70 JB |
2971 | /** |
2972 | * d_path - return the path of a dentry | |
cf28b486 | 2973 | * @path: path to report |
a03a8a70 JB |
2974 | * @buf: buffer to return value in |
2975 | * @buflen: buffer length | |
2976 | * | |
2977 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2978 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2979 | * | |
52afeefb AV |
2980 | * Returns a pointer into the buffer or an error code if the path was |
2981 | * too long. Note: Callers should use the returned pointer, not the passed | |
2982 | * in buffer, to use the name! The implementation often starts at an offset | |
2983 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2984 | * |
31f3e0b3 | 2985 | * "buflen" should be positive. |
a03a8a70 | 2986 | */ |
20d4fdc1 | 2987 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2988 | { |
ffd1f4ed | 2989 | char *res = buf + buflen; |
6ac08c39 | 2990 | struct path root; |
ffd1f4ed | 2991 | int error; |
1da177e4 | 2992 | |
c23fbb6b ED |
2993 | /* |
2994 | * We have various synthetic filesystems that never get mounted. On | |
2995 | * these filesystems dentries are never used for lookup purposes, and | |
2996 | * thus don't need to be hashed. They also don't need a name until a | |
2997 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2998 | * below allows us to generate a name for these objects on demand: | |
f48cfddc EB |
2999 | * |
3000 | * Some pseudo inodes are mountable. When they are mounted | |
3001 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname | |
3002 | * and instead have d_path return the mounted path. | |
c23fbb6b | 3003 | */ |
f48cfddc EB |
3004 | if (path->dentry->d_op && path->dentry->d_op->d_dname && |
3005 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) | |
cf28b486 | 3006 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
c23fbb6b | 3007 | |
68f0d9d9 LT |
3008 | rcu_read_lock(); |
3009 | get_fs_root_rcu(current->fs, &root); | |
02125a82 | 3010 | error = path_with_deleted(path, &root, &res, &buflen); |
68f0d9d9 LT |
3011 | rcu_read_unlock(); |
3012 | ||
02125a82 | 3013 | if (error < 0) |
ffd1f4ed | 3014 | res = ERR_PTR(error); |
1da177e4 LT |
3015 | return res; |
3016 | } | |
ec4f8605 | 3017 | EXPORT_SYMBOL(d_path); |
1da177e4 | 3018 | |
c23fbb6b ED |
3019 | /* |
3020 | * Helper function for dentry_operations.d_dname() members | |
3021 | */ | |
3022 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
3023 | const char *fmt, ...) | |
3024 | { | |
3025 | va_list args; | |
3026 | char temp[64]; | |
3027 | int sz; | |
3028 | ||
3029 | va_start(args, fmt); | |
3030 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
3031 | va_end(args); | |
3032 | ||
3033 | if (sz > sizeof(temp) || sz > buflen) | |
3034 | return ERR_PTR(-ENAMETOOLONG); | |
3035 | ||
3036 | buffer += buflen - sz; | |
3037 | return memcpy(buffer, temp, sz); | |
3038 | } | |
3039 | ||
118b2302 AV |
3040 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
3041 | { | |
3042 | char *end = buffer + buflen; | |
3043 | /* these dentries are never renamed, so d_lock is not needed */ | |
3044 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
232d2d60 | 3045 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
118b2302 AV |
3046 | prepend(&end, &buflen, "/", 1)) |
3047 | end = ERR_PTR(-ENAMETOOLONG); | |
232d2d60 | 3048 | return end; |
118b2302 | 3049 | } |
31bbe16f | 3050 | EXPORT_SYMBOL(simple_dname); |
118b2302 | 3051 | |
6092d048 RP |
3052 | /* |
3053 | * Write full pathname from the root of the filesystem into the buffer. | |
3054 | */ | |
f6500801 | 3055 | static char *__dentry_path(struct dentry *d, char *buf, int buflen) |
6092d048 | 3056 | { |
f6500801 | 3057 | struct dentry *dentry; |
232d2d60 WL |
3058 | char *end, *retval; |
3059 | int len, seq = 0; | |
3060 | int error = 0; | |
6092d048 | 3061 | |
f6500801 AV |
3062 | if (buflen < 2) |
3063 | goto Elong; | |
3064 | ||
48f5ec21 | 3065 | rcu_read_lock(); |
232d2d60 | 3066 | restart: |
f6500801 | 3067 | dentry = d; |
232d2d60 WL |
3068 | end = buf + buflen; |
3069 | len = buflen; | |
3070 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
3071 | /* Get '/' right */ |
3072 | retval = end-1; | |
3073 | *retval = '/'; | |
232d2d60 | 3074 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
3075 | while (!IS_ROOT(dentry)) { |
3076 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 3077 | |
6092d048 | 3078 | prefetch(parent); |
232d2d60 WL |
3079 | error = prepend_name(&end, &len, &dentry->d_name); |
3080 | if (error) | |
3081 | break; | |
6092d048 RP |
3082 | |
3083 | retval = end; | |
3084 | dentry = parent; | |
3085 | } | |
48f5ec21 AV |
3086 | if (!(seq & 1)) |
3087 | rcu_read_unlock(); | |
3088 | if (need_seqretry(&rename_lock, seq)) { | |
3089 | seq = 1; | |
232d2d60 | 3090 | goto restart; |
48f5ec21 AV |
3091 | } |
3092 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3093 | if (error) |
3094 | goto Elong; | |
c103135c AV |
3095 | return retval; |
3096 | Elong: | |
3097 | return ERR_PTR(-ENAMETOOLONG); | |
3098 | } | |
ec2447c2 NP |
3099 | |
3100 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3101 | { | |
232d2d60 | 3102 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3103 | } |
3104 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3105 | |
3106 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3107 | { | |
3108 | char *p = NULL; | |
3109 | char *retval; | |
3110 | ||
c103135c AV |
3111 | if (d_unlinked(dentry)) { |
3112 | p = buf + buflen; | |
3113 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3114 | goto Elong; | |
3115 | buflen++; | |
3116 | } | |
3117 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3118 | if (!IS_ERR(retval) && p) |
3119 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3120 | return retval; |
3121 | Elong: | |
6092d048 RP |
3122 | return ERR_PTR(-ENAMETOOLONG); |
3123 | } | |
3124 | ||
8b19e341 LT |
3125 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, |
3126 | struct path *pwd) | |
5762482f | 3127 | { |
8b19e341 LT |
3128 | unsigned seq; |
3129 | ||
3130 | do { | |
3131 | seq = read_seqcount_begin(&fs->seq); | |
3132 | *root = fs->root; | |
3133 | *pwd = fs->pwd; | |
3134 | } while (read_seqcount_retry(&fs->seq, seq)); | |
5762482f LT |
3135 | } |
3136 | ||
1da177e4 LT |
3137 | /* |
3138 | * NOTE! The user-level library version returns a | |
3139 | * character pointer. The kernel system call just | |
3140 | * returns the length of the buffer filled (which | |
3141 | * includes the ending '\0' character), or a negative | |
3142 | * error value. So libc would do something like | |
3143 | * | |
3144 | * char *getcwd(char * buf, size_t size) | |
3145 | * { | |
3146 | * int retval; | |
3147 | * | |
3148 | * retval = sys_getcwd(buf, size); | |
3149 | * if (retval >= 0) | |
3150 | * return buf; | |
3151 | * errno = -retval; | |
3152 | * return NULL; | |
3153 | * } | |
3154 | */ | |
3cdad428 | 3155 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3156 | { |
552ce544 | 3157 | int error; |
6ac08c39 | 3158 | struct path pwd, root; |
3272c544 | 3159 | char *page = __getname(); |
1da177e4 LT |
3160 | |
3161 | if (!page) | |
3162 | return -ENOMEM; | |
3163 | ||
8b19e341 LT |
3164 | rcu_read_lock(); |
3165 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); | |
1da177e4 | 3166 | |
552ce544 | 3167 | error = -ENOENT; |
f3da392e | 3168 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3169 | unsigned long len; |
3272c544 LT |
3170 | char *cwd = page + PATH_MAX; |
3171 | int buflen = PATH_MAX; | |
1da177e4 | 3172 | |
8df9d1a4 | 3173 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3174 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
ff812d72 | 3175 | rcu_read_unlock(); |
552ce544 | 3176 | |
02125a82 | 3177 | if (error < 0) |
552ce544 LT |
3178 | goto out; |
3179 | ||
8df9d1a4 | 3180 | /* Unreachable from current root */ |
02125a82 | 3181 | if (error > 0) { |
8df9d1a4 MS |
3182 | error = prepend_unreachable(&cwd, &buflen); |
3183 | if (error) | |
3184 | goto out; | |
3185 | } | |
3186 | ||
552ce544 | 3187 | error = -ERANGE; |
3272c544 | 3188 | len = PATH_MAX + page - cwd; |
552ce544 LT |
3189 | if (len <= size) { |
3190 | error = len; | |
3191 | if (copy_to_user(buf, cwd, len)) | |
3192 | error = -EFAULT; | |
3193 | } | |
949854d0 | 3194 | } else { |
ff812d72 | 3195 | rcu_read_unlock(); |
949854d0 | 3196 | } |
1da177e4 LT |
3197 | |
3198 | out: | |
3272c544 | 3199 | __putname(page); |
1da177e4 LT |
3200 | return error; |
3201 | } | |
3202 | ||
3203 | /* | |
3204 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3205 | * | |
3206 | * Trivially implemented using the dcache structure | |
3207 | */ | |
3208 | ||
3209 | /** | |
3210 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3211 | * @new_dentry: new dentry | |
3212 | * @old_dentry: old dentry | |
3213 | * | |
3214 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
3215 | * Returns 0 otherwise. | |
3216 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
3217 | */ | |
3218 | ||
e2761a11 | 3219 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
3220 | { |
3221 | int result; | |
949854d0 | 3222 | unsigned seq; |
1da177e4 | 3223 | |
e2761a11 OH |
3224 | if (new_dentry == old_dentry) |
3225 | return 1; | |
3226 | ||
e2761a11 | 3227 | do { |
1da177e4 | 3228 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3229 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3230 | /* |
3231 | * Need rcu_readlock to protect against the d_parent trashing | |
3232 | * due to d_move | |
3233 | */ | |
3234 | rcu_read_lock(); | |
e2761a11 | 3235 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 3236 | result = 1; |
e2761a11 OH |
3237 | else |
3238 | result = 0; | |
949854d0 | 3239 | rcu_read_unlock(); |
1da177e4 | 3240 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3241 | |
3242 | return result; | |
3243 | } | |
3244 | ||
db14fc3a | 3245 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3246 | { |
db14fc3a MS |
3247 | struct dentry *root = data; |
3248 | if (dentry != root) { | |
3249 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3250 | return D_WALK_SKIP; | |
1da177e4 | 3251 | |
01ddc4ed MS |
3252 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3253 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3254 | dentry->d_lockref.count--; | |
3255 | } | |
1da177e4 | 3256 | } |
db14fc3a MS |
3257 | return D_WALK_CONTINUE; |
3258 | } | |
58db63d0 | 3259 | |
db14fc3a MS |
3260 | void d_genocide(struct dentry *parent) |
3261 | { | |
3262 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3263 | } |
3264 | ||
60545d0d | 3265 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3266 | { |
60545d0d AV |
3267 | inode_dec_link_count(inode); |
3268 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
3269 | !hlist_unhashed(&dentry->d_alias) || | |
3270 | !d_unlinked(dentry)); | |
3271 | spin_lock(&dentry->d_parent->d_lock); | |
3272 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3273 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3274 | (unsigned long long)inode->i_ino); | |
3275 | spin_unlock(&dentry->d_lock); | |
3276 | spin_unlock(&dentry->d_parent->d_lock); | |
3277 | d_instantiate(dentry, inode); | |
1da177e4 | 3278 | } |
60545d0d | 3279 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3280 | |
3281 | static __initdata unsigned long dhash_entries; | |
3282 | static int __init set_dhash_entries(char *str) | |
3283 | { | |
3284 | if (!str) | |
3285 | return 0; | |
3286 | dhash_entries = simple_strtoul(str, &str, 0); | |
3287 | return 1; | |
3288 | } | |
3289 | __setup("dhash_entries=", set_dhash_entries); | |
3290 | ||
3291 | static void __init dcache_init_early(void) | |
3292 | { | |
074b8517 | 3293 | unsigned int loop; |
1da177e4 LT |
3294 | |
3295 | /* If hashes are distributed across NUMA nodes, defer | |
3296 | * hash allocation until vmalloc space is available. | |
3297 | */ | |
3298 | if (hashdist) | |
3299 | return; | |
3300 | ||
3301 | dentry_hashtable = | |
3302 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3303 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3304 | dhash_entries, |
3305 | 13, | |
3306 | HASH_EARLY, | |
3307 | &d_hash_shift, | |
3308 | &d_hash_mask, | |
31fe62b9 | 3309 | 0, |
1da177e4 LT |
3310 | 0); |
3311 | ||
074b8517 | 3312 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3313 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3314 | } |
3315 | ||
74bf17cf | 3316 | static void __init dcache_init(void) |
1da177e4 | 3317 | { |
074b8517 | 3318 | unsigned int loop; |
1da177e4 LT |
3319 | |
3320 | /* | |
3321 | * A constructor could be added for stable state like the lists, | |
3322 | * but it is probably not worth it because of the cache nature | |
3323 | * of the dcache. | |
3324 | */ | |
0a31bd5f CL |
3325 | dentry_cache = KMEM_CACHE(dentry, |
3326 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3327 | |
3328 | /* Hash may have been set up in dcache_init_early */ | |
3329 | if (!hashdist) | |
3330 | return; | |
3331 | ||
3332 | dentry_hashtable = | |
3333 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3334 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3335 | dhash_entries, |
3336 | 13, | |
3337 | 0, | |
3338 | &d_hash_shift, | |
3339 | &d_hash_mask, | |
31fe62b9 | 3340 | 0, |
1da177e4 LT |
3341 | 0); |
3342 | ||
074b8517 | 3343 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3344 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3345 | } |
3346 | ||
3347 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3348 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3349 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3350 | |
1da177e4 LT |
3351 | EXPORT_SYMBOL(d_genocide); |
3352 | ||
1da177e4 LT |
3353 | void __init vfs_caches_init_early(void) |
3354 | { | |
3355 | dcache_init_early(); | |
3356 | inode_init_early(); | |
3357 | } | |
3358 | ||
3359 | void __init vfs_caches_init(unsigned long mempages) | |
3360 | { | |
3361 | unsigned long reserve; | |
3362 | ||
3363 | /* Base hash sizes on available memory, with a reserve equal to | |
3364 | 150% of current kernel size */ | |
3365 | ||
3366 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3367 | mempages -= reserve; | |
3368 | ||
3369 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3370 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3371 | |
74bf17cf DC |
3372 | dcache_init(); |
3373 | inode_init(); | |
1da177e4 | 3374 | files_init(mempages); |
74bf17cf | 3375 | mnt_init(); |
1da177e4 LT |
3376 | bdev_cache_init(); |
3377 | chrdev_init(); | |
3378 | } |