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 | |
232d2d60 WL |
91 | /** |
92 | * read_seqbegin_or_lock - begin a sequence number check or locking block | |
93 | * lock: sequence lock | |
94 | * seq : sequence number to be checked | |
95 | * | |
96 | * First try it once optimistically without taking the lock. If that fails, | |
97 | * take the lock. The sequence number is also used as a marker for deciding | |
98 | * whether to be a reader (even) or writer (odd). | |
99 | * N.B. seq must be initialized to an even number to begin with. | |
100 | */ | |
101 | static inline void read_seqbegin_or_lock(seqlock_t *lock, int *seq) | |
102 | { | |
48f5ec21 | 103 | if (!(*seq & 1)) /* Even */ |
232d2d60 | 104 | *seq = read_seqbegin(lock); |
48f5ec21 | 105 | else /* Odd */ |
232d2d60 WL |
106 | write_seqlock(lock); |
107 | } | |
108 | ||
48f5ec21 | 109 | static inline int need_seqretry(seqlock_t *lock, int seq) |
232d2d60 | 110 | { |
48f5ec21 AV |
111 | return !(seq & 1) && read_seqretry(lock, seq); |
112 | } | |
113 | ||
114 | static inline void done_seqretry(seqlock_t *lock, int seq) | |
115 | { | |
116 | if (seq & 1) | |
232d2d60 | 117 | write_sequnlock(lock); |
232d2d60 WL |
118 | } |
119 | ||
1da177e4 LT |
120 | /* |
121 | * This is the single most critical data structure when it comes | |
122 | * to the dcache: the hashtable for lookups. Somebody should try | |
123 | * to make this good - I've just made it work. | |
124 | * | |
125 | * This hash-function tries to avoid losing too many bits of hash | |
126 | * information, yet avoid using a prime hash-size or similar. | |
127 | */ | |
128 | #define D_HASHBITS d_hash_shift | |
129 | #define D_HASHMASK d_hash_mask | |
130 | ||
fa3536cc ED |
131 | static unsigned int d_hash_mask __read_mostly; |
132 | static unsigned int d_hash_shift __read_mostly; | |
ceb5bdc2 | 133 | |
b07ad996 | 134 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 135 | |
8966be90 | 136 | static inline struct hlist_bl_head *d_hash(const struct dentry *parent, |
6d7d1a0d | 137 | unsigned int hash) |
ceb5bdc2 | 138 | { |
6d7d1a0d LT |
139 | hash += (unsigned long) parent / L1_CACHE_BYTES; |
140 | hash = hash + (hash >> D_HASHBITS); | |
ceb5bdc2 NP |
141 | return dentry_hashtable + (hash & D_HASHMASK); |
142 | } | |
143 | ||
1da177e4 LT |
144 | /* Statistics gathering. */ |
145 | struct dentry_stat_t dentry_stat = { | |
146 | .age_limit = 45, | |
147 | }; | |
148 | ||
3942c07c | 149 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 150 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
312d3ca8 CH |
151 | |
152 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
62d36c77 DC |
153 | |
154 | /* | |
155 | * Here we resort to our own counters instead of using generic per-cpu counters | |
156 | * for consistency with what the vfs inode code does. We are expected to harvest | |
157 | * better code and performance by having our own specialized counters. | |
158 | * | |
159 | * Please note that the loop is done over all possible CPUs, not over all online | |
160 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
161 | * on and off. If one of them goes off, we will just keep their counters. | |
162 | * | |
163 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
164 | * please update all vfs counters to match. | |
165 | */ | |
3942c07c | 166 | static long get_nr_dentry(void) |
3e880fb5 NP |
167 | { |
168 | int i; | |
3942c07c | 169 | long sum = 0; |
3e880fb5 NP |
170 | for_each_possible_cpu(i) |
171 | sum += per_cpu(nr_dentry, i); | |
172 | return sum < 0 ? 0 : sum; | |
173 | } | |
174 | ||
62d36c77 DC |
175 | static long get_nr_dentry_unused(void) |
176 | { | |
177 | int i; | |
178 | long sum = 0; | |
179 | for_each_possible_cpu(i) | |
180 | sum += per_cpu(nr_dentry_unused, i); | |
181 | return sum < 0 ? 0 : sum; | |
182 | } | |
183 | ||
312d3ca8 CH |
184 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
185 | size_t *lenp, loff_t *ppos) | |
186 | { | |
3e880fb5 | 187 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 188 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
3942c07c | 189 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 CH |
190 | } |
191 | #endif | |
192 | ||
5483f18e LT |
193 | /* |
194 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
195 | * The strings are both count bytes long, and count is non-zero. | |
196 | */ | |
e419b4cc LT |
197 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
198 | ||
199 | #include <asm/word-at-a-time.h> | |
200 | /* | |
201 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
202 | * aligned allocation for this particular component. We don't | |
203 | * strictly need the load_unaligned_zeropad() safety, but it | |
204 | * doesn't hurt either. | |
205 | * | |
206 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
207 | * need the careful unaligned handling. | |
208 | */ | |
94753db5 | 209 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 210 | { |
bfcfaa77 | 211 | unsigned long a,b,mask; |
bfcfaa77 LT |
212 | |
213 | for (;;) { | |
12f8ad4b | 214 | a = *(unsigned long *)cs; |
e419b4cc | 215 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
216 | if (tcount < sizeof(unsigned long)) |
217 | break; | |
218 | if (unlikely(a != b)) | |
219 | return 1; | |
220 | cs += sizeof(unsigned long); | |
221 | ct += sizeof(unsigned long); | |
222 | tcount -= sizeof(unsigned long); | |
223 | if (!tcount) | |
224 | return 0; | |
225 | } | |
226 | mask = ~(~0ul << tcount*8); | |
227 | return unlikely(!!((a ^ b) & mask)); | |
e419b4cc LT |
228 | } |
229 | ||
bfcfaa77 | 230 | #else |
e419b4cc | 231 | |
94753db5 | 232 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 233 | { |
5483f18e LT |
234 | do { |
235 | if (*cs != *ct) | |
236 | return 1; | |
237 | cs++; | |
238 | ct++; | |
239 | tcount--; | |
240 | } while (tcount); | |
241 | return 0; | |
242 | } | |
243 | ||
e419b4cc LT |
244 | #endif |
245 | ||
94753db5 LT |
246 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
247 | { | |
6326c71f | 248 | const unsigned char *cs; |
94753db5 LT |
249 | /* |
250 | * Be careful about RCU walk racing with rename: | |
251 | * use ACCESS_ONCE to fetch the name pointer. | |
252 | * | |
253 | * NOTE! Even if a rename will mean that the length | |
254 | * was not loaded atomically, we don't care. The | |
255 | * RCU walk will check the sequence count eventually, | |
256 | * and catch it. And we won't overrun the buffer, | |
257 | * because we're reading the name pointer atomically, | |
258 | * and a dentry name is guaranteed to be properly | |
259 | * terminated with a NUL byte. | |
260 | * | |
261 | * End result: even if 'len' is wrong, we'll exit | |
262 | * early because the data cannot match (there can | |
263 | * be no NUL in the ct/tcount data) | |
264 | */ | |
6326c71f LT |
265 | cs = ACCESS_ONCE(dentry->d_name.name); |
266 | smp_read_barrier_depends(); | |
267 | return dentry_string_cmp(cs, ct, tcount); | |
94753db5 LT |
268 | } |
269 | ||
9c82ab9c | 270 | static void __d_free(struct rcu_head *head) |
1da177e4 | 271 | { |
9c82ab9c CH |
272 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
273 | ||
b3d9b7a3 | 274 | WARN_ON(!hlist_unhashed(&dentry->d_alias)); |
1da177e4 LT |
275 | if (dname_external(dentry)) |
276 | kfree(dentry->d_name.name); | |
277 | kmem_cache_free(dentry_cache, dentry); | |
278 | } | |
279 | ||
280 | /* | |
b5c84bf6 | 281 | * no locks, please. |
1da177e4 LT |
282 | */ |
283 | static void d_free(struct dentry *dentry) | |
284 | { | |
0d98439e | 285 | BUG_ON((int)dentry->d_lockref.count > 0); |
3e880fb5 | 286 | this_cpu_dec(nr_dentry); |
1da177e4 LT |
287 | if (dentry->d_op && dentry->d_op->d_release) |
288 | dentry->d_op->d_release(dentry); | |
312d3ca8 | 289 | |
dea3667b LT |
290 | /* if dentry was never visible to RCU, immediate free is OK */ |
291 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
9c82ab9c | 292 | __d_free(&dentry->d_u.d_rcu); |
b3423415 | 293 | else |
9c82ab9c | 294 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
1da177e4 LT |
295 | } |
296 | ||
31e6b01f NP |
297 | /** |
298 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
ff5fdb61 | 299 | * @dentry: the target dentry |
31e6b01f NP |
300 | * After this call, in-progress rcu-walk path lookup will fail. This |
301 | * should be called after unhashing, and after changing d_inode (if | |
302 | * the dentry has not already been unhashed). | |
303 | */ | |
304 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
305 | { | |
306 | assert_spin_locked(&dentry->d_lock); | |
307 | /* Go through a barrier */ | |
308 | write_seqcount_barrier(&dentry->d_seq); | |
309 | } | |
310 | ||
1da177e4 LT |
311 | /* |
312 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
313 | * d_iput() operation if defined. Dentry has no refcount |
314 | * and is unhashed. | |
1da177e4 | 315 | */ |
858119e1 | 316 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 317 | __releases(dentry->d_lock) |
873feea0 | 318 | __releases(dentry->d_inode->i_lock) |
1da177e4 LT |
319 | { |
320 | struct inode *inode = dentry->d_inode; | |
321 | if (inode) { | |
322 | dentry->d_inode = NULL; | |
b3d9b7a3 | 323 | hlist_del_init(&dentry->d_alias); |
1da177e4 | 324 | spin_unlock(&dentry->d_lock); |
873feea0 | 325 | spin_unlock(&inode->i_lock); |
f805fbda LT |
326 | if (!inode->i_nlink) |
327 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
328 | if (dentry->d_op && dentry->d_op->d_iput) |
329 | dentry->d_op->d_iput(dentry, inode); | |
330 | else | |
331 | iput(inode); | |
332 | } else { | |
333 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
334 | } |
335 | } | |
336 | ||
31e6b01f NP |
337 | /* |
338 | * Release the dentry's inode, using the filesystem | |
339 | * d_iput() operation if defined. dentry remains in-use. | |
340 | */ | |
341 | static void dentry_unlink_inode(struct dentry * dentry) | |
342 | __releases(dentry->d_lock) | |
873feea0 | 343 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
344 | { |
345 | struct inode *inode = dentry->d_inode; | |
346 | dentry->d_inode = NULL; | |
b3d9b7a3 | 347 | hlist_del_init(&dentry->d_alias); |
31e6b01f NP |
348 | dentry_rcuwalk_barrier(dentry); |
349 | spin_unlock(&dentry->d_lock); | |
873feea0 | 350 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
351 | if (!inode->i_nlink) |
352 | fsnotify_inoderemove(inode); | |
353 | if (dentry->d_op && dentry->d_op->d_iput) | |
354 | dentry->d_op->d_iput(dentry, inode); | |
355 | else | |
356 | iput(inode); | |
357 | } | |
358 | ||
da3bbdd4 | 359 | /* |
f6041567 | 360 | * dentry_lru_(add|del)_list) must be called with d_lock held. |
da3bbdd4 KM |
361 | */ |
362 | static void dentry_lru_add(struct dentry *dentry) | |
363 | { | |
8aab6a27 | 364 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) { |
f6041567 DC |
365 | if (list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)) |
366 | this_cpu_inc(nr_dentry_unused); | |
8aab6a27 | 367 | dentry->d_flags |= DCACHE_LRU_LIST; |
a4633357 | 368 | } |
da3bbdd4 KM |
369 | } |
370 | ||
f0023bc6 SW |
371 | /* |
372 | * Remove a dentry with references from the LRU. | |
dd1f6b2e DC |
373 | * |
374 | * If we are on the shrink list, then we can get to try_prune_one_dentry() and | |
375 | * lose our last reference through the parent walk. In this case, we need to | |
376 | * remove ourselves from the shrink list, not the LRU. | |
f0023bc6 | 377 | */ |
da3bbdd4 KM |
378 | static void dentry_lru_del(struct dentry *dentry) |
379 | { | |
dd1f6b2e DC |
380 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
381 | list_del_init(&dentry->d_lru); | |
382 | dentry->d_flags &= ~DCACHE_SHRINK_LIST; | |
383 | return; | |
384 | } | |
385 | ||
f6041567 | 386 | if (list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)) |
dd1f6b2e | 387 | this_cpu_dec(nr_dentry_unused); |
f6041567 | 388 | dentry->d_flags &= ~DCACHE_LRU_LIST; |
da3bbdd4 KM |
389 | } |
390 | ||
d52b9086 MS |
391 | /** |
392 | * d_kill - kill dentry and return parent | |
393 | * @dentry: dentry to kill | |
ff5fdb61 | 394 | * @parent: parent dentry |
d52b9086 | 395 | * |
31f3e0b3 | 396 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
397 | * |
398 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 399 | * |
b5c84bf6 NP |
400 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
401 | * d_kill. | |
d52b9086 | 402 | */ |
2fd6b7f5 | 403 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 404 | __releases(dentry->d_lock) |
2fd6b7f5 | 405 | __releases(parent->d_lock) |
873feea0 | 406 | __releases(dentry->d_inode->i_lock) |
d52b9086 | 407 | { |
d52b9086 | 408 | list_del(&dentry->d_u.d_child); |
c83ce989 TM |
409 | /* |
410 | * Inform try_to_ascend() that we are no longer attached to the | |
411 | * dentry tree | |
412 | */ | |
b161dfa6 | 413 | dentry->d_flags |= DCACHE_DENTRY_KILLED; |
2fd6b7f5 NP |
414 | if (parent) |
415 | spin_unlock(&parent->d_lock); | |
d52b9086 | 416 | dentry_iput(dentry); |
b7ab39f6 NP |
417 | /* |
418 | * dentry_iput drops the locks, at which point nobody (except | |
419 | * transient RCU lookups) can reach this dentry. | |
420 | */ | |
d52b9086 | 421 | d_free(dentry); |
871c0067 | 422 | return parent; |
d52b9086 MS |
423 | } |
424 | ||
c6627c60 DH |
425 | /* |
426 | * Unhash a dentry without inserting an RCU walk barrier or checking that | |
427 | * dentry->d_lock is locked. The caller must take care of that, if | |
428 | * appropriate. | |
429 | */ | |
430 | static void __d_shrink(struct dentry *dentry) | |
431 | { | |
432 | if (!d_unhashed(dentry)) { | |
433 | struct hlist_bl_head *b; | |
434 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
435 | b = &dentry->d_sb->s_anon; | |
436 | else | |
437 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
438 | ||
439 | hlist_bl_lock(b); | |
440 | __hlist_bl_del(&dentry->d_hash); | |
441 | dentry->d_hash.pprev = NULL; | |
442 | hlist_bl_unlock(b); | |
443 | } | |
444 | } | |
445 | ||
789680d1 NP |
446 | /** |
447 | * d_drop - drop a dentry | |
448 | * @dentry: dentry to drop | |
449 | * | |
450 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
451 | * be found through a VFS lookup any more. Note that this is different from | |
452 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
453 | * possible, giving a successful _negative_ lookup, while d_drop will | |
454 | * just make the cache lookup fail. | |
455 | * | |
456 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
457 | * reason (NFS timeouts or autofs deletes). | |
458 | * | |
459 | * __d_drop requires dentry->d_lock. | |
460 | */ | |
461 | void __d_drop(struct dentry *dentry) | |
462 | { | |
dea3667b | 463 | if (!d_unhashed(dentry)) { |
c6627c60 | 464 | __d_shrink(dentry); |
dea3667b | 465 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
466 | } |
467 | } | |
468 | EXPORT_SYMBOL(__d_drop); | |
469 | ||
470 | void d_drop(struct dentry *dentry) | |
471 | { | |
789680d1 NP |
472 | spin_lock(&dentry->d_lock); |
473 | __d_drop(dentry); | |
474 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
475 | } |
476 | EXPORT_SYMBOL(d_drop); | |
477 | ||
77812a1e NP |
478 | /* |
479 | * Finish off a dentry we've decided to kill. | |
480 | * dentry->d_lock must be held, returns with it unlocked. | |
481 | * If ref is non-zero, then decrement the refcount too. | |
482 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
483 | */ | |
dd1f6b2e DC |
484 | static inline struct dentry * |
485 | dentry_kill(struct dentry *dentry, int unlock_on_failure) | |
77812a1e NP |
486 | __releases(dentry->d_lock) |
487 | { | |
873feea0 | 488 | struct inode *inode; |
77812a1e NP |
489 | struct dentry *parent; |
490 | ||
873feea0 NP |
491 | inode = dentry->d_inode; |
492 | if (inode && !spin_trylock(&inode->i_lock)) { | |
77812a1e | 493 | relock: |
dd1f6b2e DC |
494 | if (unlock_on_failure) { |
495 | spin_unlock(&dentry->d_lock); | |
496 | cpu_relax(); | |
497 | } | |
77812a1e NP |
498 | return dentry; /* try again with same dentry */ |
499 | } | |
500 | if (IS_ROOT(dentry)) | |
501 | parent = NULL; | |
502 | else | |
503 | parent = dentry->d_parent; | |
504 | if (parent && !spin_trylock(&parent->d_lock)) { | |
873feea0 NP |
505 | if (inode) |
506 | spin_unlock(&inode->i_lock); | |
77812a1e NP |
507 | goto relock; |
508 | } | |
31e6b01f | 509 | |
0d98439e LT |
510 | /* |
511 | * The dentry is now unrecoverably dead to the world. | |
512 | */ | |
513 | lockref_mark_dead(&dentry->d_lockref); | |
514 | ||
f0023bc6 | 515 | /* |
f0023bc6 SW |
516 | * inform the fs via d_prune that this dentry is about to be |
517 | * unhashed and destroyed. | |
518 | */ | |
590fb51f | 519 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry)) |
61572bb1 YZ |
520 | dentry->d_op->d_prune(dentry); |
521 | ||
522 | dentry_lru_del(dentry); | |
77812a1e NP |
523 | /* if it was on the hash then remove it */ |
524 | __d_drop(dentry); | |
525 | return d_kill(dentry, parent); | |
526 | } | |
527 | ||
1da177e4 LT |
528 | /* |
529 | * This is dput | |
530 | * | |
531 | * This is complicated by the fact that we do not want to put | |
532 | * dentries that are no longer on any hash chain on the unused | |
533 | * list: we'd much rather just get rid of them immediately. | |
534 | * | |
535 | * However, that implies that we have to traverse the dentry | |
536 | * tree upwards to the parents which might _also_ now be | |
537 | * scheduled for deletion (it may have been only waiting for | |
538 | * its last child to go away). | |
539 | * | |
540 | * This tail recursion is done by hand as we don't want to depend | |
541 | * on the compiler to always get this right (gcc generally doesn't). | |
542 | * Real recursion would eat up our stack space. | |
543 | */ | |
544 | ||
545 | /* | |
546 | * dput - release a dentry | |
547 | * @dentry: dentry to release | |
548 | * | |
549 | * Release a dentry. This will drop the usage count and if appropriate | |
550 | * call the dentry unlink method as well as removing it from the queues and | |
551 | * releasing its resources. If the parent dentries were scheduled for release | |
552 | * they too may now get deleted. | |
1da177e4 | 553 | */ |
1da177e4 LT |
554 | void dput(struct dentry *dentry) |
555 | { | |
8aab6a27 | 556 | if (unlikely(!dentry)) |
1da177e4 LT |
557 | return; |
558 | ||
559 | repeat: | |
98474236 | 560 | if (lockref_put_or_lock(&dentry->d_lockref)) |
1da177e4 | 561 | return; |
1da177e4 | 562 | |
8aab6a27 LT |
563 | /* Unreachable? Get rid of it */ |
564 | if (unlikely(d_unhashed(dentry))) | |
565 | goto kill_it; | |
566 | ||
567 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
1da177e4 | 568 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 569 | goto kill_it; |
1da177e4 | 570 | } |
265ac902 | 571 | |
39e3c955 | 572 | dentry->d_flags |= DCACHE_REFERENCED; |
a4633357 | 573 | dentry_lru_add(dentry); |
265ac902 | 574 | |
98474236 | 575 | dentry->d_lockref.count--; |
61f3dee4 | 576 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
577 | return; |
578 | ||
d52b9086 | 579 | kill_it: |
dd1f6b2e | 580 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
581 | if (dentry) |
582 | goto repeat; | |
1da177e4 | 583 | } |
ec4f8605 | 584 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
585 | |
586 | /** | |
587 | * d_invalidate - invalidate a dentry | |
588 | * @dentry: dentry to invalidate | |
589 | * | |
590 | * Try to invalidate the dentry if it turns out to be | |
591 | * possible. If there are other dentries that can be | |
592 | * reached through this one we can't delete it and we | |
593 | * return -EBUSY. On success we return 0. | |
594 | * | |
595 | * no dcache lock. | |
596 | */ | |
597 | ||
598 | int d_invalidate(struct dentry * dentry) | |
599 | { | |
600 | /* | |
601 | * If it's already been dropped, return OK. | |
602 | */ | |
da502956 | 603 | spin_lock(&dentry->d_lock); |
1da177e4 | 604 | if (d_unhashed(dentry)) { |
da502956 | 605 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
606 | return 0; |
607 | } | |
608 | /* | |
609 | * Check whether to do a partial shrink_dcache | |
610 | * to get rid of unused child entries. | |
611 | */ | |
612 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 613 | spin_unlock(&dentry->d_lock); |
1da177e4 | 614 | shrink_dcache_parent(dentry); |
da502956 | 615 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
616 | } |
617 | ||
618 | /* | |
619 | * Somebody else still using it? | |
620 | * | |
621 | * If it's a directory, we can't drop it | |
622 | * for fear of somebody re-populating it | |
623 | * with children (even though dropping it | |
624 | * would make it unreachable from the root, | |
625 | * we might still populate it if it was a | |
626 | * working directory or similar). | |
50e69630 AV |
627 | * We also need to leave mountpoints alone, |
628 | * directory or not. | |
1da177e4 | 629 | */ |
98474236 | 630 | if (dentry->d_lockref.count > 1 && dentry->d_inode) { |
50e69630 | 631 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { |
1da177e4 | 632 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
633 | return -EBUSY; |
634 | } | |
635 | } | |
636 | ||
637 | __d_drop(dentry); | |
638 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
639 | return 0; |
640 | } | |
ec4f8605 | 641 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 642 | |
b5c84bf6 | 643 | /* This must be called with d_lock held */ |
dc0474be | 644 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 645 | { |
98474236 | 646 | dentry->d_lockref.count++; |
23044507 NP |
647 | } |
648 | ||
dc0474be | 649 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 650 | { |
98474236 | 651 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
652 | } |
653 | ||
b7ab39f6 NP |
654 | struct dentry *dget_parent(struct dentry *dentry) |
655 | { | |
df3d0bbc | 656 | int gotref; |
b7ab39f6 NP |
657 | struct dentry *ret; |
658 | ||
df3d0bbc WL |
659 | /* |
660 | * Do optimistic parent lookup without any | |
661 | * locking. | |
662 | */ | |
663 | rcu_read_lock(); | |
664 | ret = ACCESS_ONCE(dentry->d_parent); | |
665 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
666 | rcu_read_unlock(); | |
667 | if (likely(gotref)) { | |
668 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
669 | return ret; | |
670 | dput(ret); | |
671 | } | |
672 | ||
b7ab39f6 | 673 | repeat: |
a734eb45 NP |
674 | /* |
675 | * Don't need rcu_dereference because we re-check it was correct under | |
676 | * the lock. | |
677 | */ | |
678 | rcu_read_lock(); | |
b7ab39f6 | 679 | ret = dentry->d_parent; |
a734eb45 NP |
680 | spin_lock(&ret->d_lock); |
681 | if (unlikely(ret != dentry->d_parent)) { | |
682 | spin_unlock(&ret->d_lock); | |
683 | rcu_read_unlock(); | |
b7ab39f6 NP |
684 | goto repeat; |
685 | } | |
a734eb45 | 686 | rcu_read_unlock(); |
98474236 WL |
687 | BUG_ON(!ret->d_lockref.count); |
688 | ret->d_lockref.count++; | |
b7ab39f6 | 689 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
690 | return ret; |
691 | } | |
692 | EXPORT_SYMBOL(dget_parent); | |
693 | ||
1da177e4 LT |
694 | /** |
695 | * d_find_alias - grab a hashed alias of inode | |
696 | * @inode: inode in question | |
32ba9c3f LT |
697 | * @want_discon: flag, used by d_splice_alias, to request |
698 | * that only a DISCONNECTED alias be returned. | |
1da177e4 LT |
699 | * |
700 | * If inode has a hashed alias, or is a directory and has any alias, | |
701 | * acquire the reference to alias and return it. Otherwise return NULL. | |
702 | * Notice that if inode is a directory there can be only one alias and | |
703 | * it can be unhashed only if it has no children, or if it is the root | |
704 | * of a filesystem. | |
705 | * | |
21c0d8fd | 706 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
32ba9c3f LT |
707 | * any other hashed alias over that one unless @want_discon is set, |
708 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. | |
1da177e4 | 709 | */ |
32ba9c3f | 710 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 711 | { |
da502956 | 712 | struct dentry *alias, *discon_alias; |
1da177e4 | 713 | |
da502956 NP |
714 | again: |
715 | discon_alias = NULL; | |
b67bfe0d | 716 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
da502956 | 717 | spin_lock(&alias->d_lock); |
1da177e4 | 718 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 719 | if (IS_ROOT(alias) && |
da502956 | 720 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 721 | discon_alias = alias; |
32ba9c3f | 722 | } else if (!want_discon) { |
dc0474be | 723 | __dget_dlock(alias); |
da502956 NP |
724 | spin_unlock(&alias->d_lock); |
725 | return alias; | |
726 | } | |
727 | } | |
728 | spin_unlock(&alias->d_lock); | |
729 | } | |
730 | if (discon_alias) { | |
731 | alias = discon_alias; | |
732 | spin_lock(&alias->d_lock); | |
733 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
734 | if (IS_ROOT(alias) && | |
735 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 736 | __dget_dlock(alias); |
da502956 | 737 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
738 | return alias; |
739 | } | |
740 | } | |
da502956 NP |
741 | spin_unlock(&alias->d_lock); |
742 | goto again; | |
1da177e4 | 743 | } |
da502956 | 744 | return NULL; |
1da177e4 LT |
745 | } |
746 | ||
da502956 | 747 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 748 | { |
214fda1f DH |
749 | struct dentry *de = NULL; |
750 | ||
b3d9b7a3 | 751 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 752 | spin_lock(&inode->i_lock); |
32ba9c3f | 753 | de = __d_find_alias(inode, 0); |
873feea0 | 754 | spin_unlock(&inode->i_lock); |
214fda1f | 755 | } |
1da177e4 LT |
756 | return de; |
757 | } | |
ec4f8605 | 758 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
759 | |
760 | /* | |
761 | * Try to kill dentries associated with this inode. | |
762 | * WARNING: you must own a reference to inode. | |
763 | */ | |
764 | void d_prune_aliases(struct inode *inode) | |
765 | { | |
0cdca3f9 | 766 | struct dentry *dentry; |
1da177e4 | 767 | restart: |
873feea0 | 768 | spin_lock(&inode->i_lock); |
b67bfe0d | 769 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 770 | spin_lock(&dentry->d_lock); |
98474236 | 771 | if (!dentry->d_lockref.count) { |
590fb51f YZ |
772 | /* |
773 | * inform the fs via d_prune that this dentry | |
774 | * is about to be unhashed and destroyed. | |
775 | */ | |
776 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && | |
777 | !d_unhashed(dentry)) | |
778 | dentry->d_op->d_prune(dentry); | |
779 | ||
dc0474be | 780 | __dget_dlock(dentry); |
1da177e4 LT |
781 | __d_drop(dentry); |
782 | spin_unlock(&dentry->d_lock); | |
873feea0 | 783 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
784 | dput(dentry); |
785 | goto restart; | |
786 | } | |
787 | spin_unlock(&dentry->d_lock); | |
788 | } | |
873feea0 | 789 | spin_unlock(&inode->i_lock); |
1da177e4 | 790 | } |
ec4f8605 | 791 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
792 | |
793 | /* | |
77812a1e NP |
794 | * Try to throw away a dentry - free the inode, dput the parent. |
795 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
796 | * Releases dentry->d_lock. | |
d702ccb3 | 797 | * |
77812a1e | 798 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 799 | */ |
dd1f6b2e | 800 | static struct dentry * try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 801 | __releases(dentry->d_lock) |
1da177e4 | 802 | { |
77812a1e | 803 | struct dentry *parent; |
d52b9086 | 804 | |
dd1f6b2e | 805 | parent = dentry_kill(dentry, 0); |
d52b9086 | 806 | /* |
77812a1e NP |
807 | * If dentry_kill returns NULL, we have nothing more to do. |
808 | * if it returns the same dentry, trylocks failed. In either | |
809 | * case, just loop again. | |
810 | * | |
811 | * Otherwise, we need to prune ancestors too. This is necessary | |
812 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
813 | * is also expected to be beneficial in reducing dentry cache | |
814 | * fragmentation. | |
d52b9086 | 815 | */ |
77812a1e | 816 | if (!parent) |
dd1f6b2e | 817 | return NULL; |
77812a1e | 818 | if (parent == dentry) |
dd1f6b2e | 819 | return dentry; |
77812a1e NP |
820 | |
821 | /* Prune ancestors. */ | |
822 | dentry = parent; | |
d52b9086 | 823 | while (dentry) { |
98474236 | 824 | if (lockref_put_or_lock(&dentry->d_lockref)) |
dd1f6b2e DC |
825 | return NULL; |
826 | dentry = dentry_kill(dentry, 1); | |
d52b9086 | 827 | } |
dd1f6b2e | 828 | return NULL; |
1da177e4 LT |
829 | } |
830 | ||
3049cfe2 | 831 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 832 | { |
da3bbdd4 | 833 | struct dentry *dentry; |
da3bbdd4 | 834 | |
ec33679d NP |
835 | rcu_read_lock(); |
836 | for (;;) { | |
ec33679d NP |
837 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
838 | if (&dentry->d_lru == list) | |
839 | break; /* empty */ | |
840 | spin_lock(&dentry->d_lock); | |
841 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
842 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
843 | continue; |
844 | } | |
845 | ||
dd1f6b2e DC |
846 | /* |
847 | * The dispose list is isolated and dentries are not accounted | |
848 | * to the LRU here, so we can simply remove it from the list | |
849 | * here regardless of whether it is referenced or not. | |
850 | */ | |
851 | list_del_init(&dentry->d_lru); | |
852 | dentry->d_flags &= ~DCACHE_SHRINK_LIST; | |
853 | ||
1da177e4 LT |
854 | /* |
855 | * We found an inuse dentry which was not removed from | |
dd1f6b2e | 856 | * the LRU because of laziness during lookup. Do not free it. |
1da177e4 | 857 | */ |
98474236 | 858 | if (dentry->d_lockref.count) { |
da3bbdd4 | 859 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
860 | continue; |
861 | } | |
ec33679d | 862 | rcu_read_unlock(); |
77812a1e | 863 | |
dd1f6b2e | 864 | dentry = try_prune_one_dentry(dentry); |
77812a1e | 865 | |
ec33679d | 866 | rcu_read_lock(); |
dd1f6b2e DC |
867 | if (dentry) { |
868 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
869 | list_add(&dentry->d_lru, list); | |
870 | spin_unlock(&dentry->d_lock); | |
871 | } | |
da3bbdd4 | 872 | } |
ec33679d | 873 | rcu_read_unlock(); |
3049cfe2 CH |
874 | } |
875 | ||
f6041567 DC |
876 | static enum lru_status |
877 | dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg) | |
878 | { | |
879 | struct list_head *freeable = arg; | |
880 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
881 | ||
882 | ||
883 | /* | |
884 | * we are inverting the lru lock/dentry->d_lock here, | |
885 | * so use a trylock. If we fail to get the lock, just skip | |
886 | * it | |
887 | */ | |
888 | if (!spin_trylock(&dentry->d_lock)) | |
889 | return LRU_SKIP; | |
890 | ||
891 | /* | |
892 | * Referenced dentries are still in use. If they have active | |
893 | * counts, just remove them from the LRU. Otherwise give them | |
894 | * another pass through the LRU. | |
895 | */ | |
896 | if (dentry->d_lockref.count) { | |
897 | list_del_init(&dentry->d_lru); | |
898 | spin_unlock(&dentry->d_lock); | |
899 | return LRU_REMOVED; | |
900 | } | |
901 | ||
902 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
903 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
904 | spin_unlock(&dentry->d_lock); | |
905 | ||
906 | /* | |
907 | * The list move itself will be made by the common LRU code. At | |
908 | * this point, we've dropped the dentry->d_lock but keep the | |
909 | * lru lock. This is safe to do, since every list movement is | |
910 | * protected by the lru lock even if both locks are held. | |
911 | * | |
912 | * This is guaranteed by the fact that all LRU management | |
913 | * functions are intermediated by the LRU API calls like | |
914 | * list_lru_add and list_lru_del. List movement in this file | |
915 | * only ever occur through this functions or through callbacks | |
916 | * like this one, that are called from the LRU API. | |
917 | * | |
918 | * The only exceptions to this are functions like | |
919 | * shrink_dentry_list, and code that first checks for the | |
920 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
921 | * operating only with stack provided lists after they are | |
922 | * properly isolated from the main list. It is thus, always a | |
923 | * local access. | |
924 | */ | |
925 | return LRU_ROTATE; | |
926 | } | |
927 | ||
928 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
929 | list_move_tail(&dentry->d_lru, freeable); | |
930 | this_cpu_dec(nr_dentry_unused); | |
931 | spin_unlock(&dentry->d_lock); | |
932 | ||
933 | return LRU_REMOVED; | |
934 | } | |
935 | ||
3049cfe2 | 936 | /** |
b48f03b3 DC |
937 | * prune_dcache_sb - shrink the dcache |
938 | * @sb: superblock | |
f6041567 | 939 | * @nr_to_scan : number of entries to try to free |
b48f03b3 | 940 | * |
f6041567 | 941 | * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is |
b48f03b3 DC |
942 | * done when we need more memory an called from the superblock shrinker |
943 | * function. | |
3049cfe2 | 944 | * |
b48f03b3 DC |
945 | * This function may fail to free any resources if all the dentries are in |
946 | * use. | |
3049cfe2 | 947 | */ |
0a234c6d | 948 | long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan) |
3049cfe2 | 949 | { |
f6041567 DC |
950 | LIST_HEAD(dispose); |
951 | long freed; | |
3049cfe2 | 952 | |
f6041567 DC |
953 | freed = list_lru_walk(&sb->s_dentry_lru, dentry_lru_isolate, |
954 | &dispose, nr_to_scan); | |
955 | shrink_dentry_list(&dispose); | |
0a234c6d | 956 | return freed; |
da3bbdd4 KM |
957 | } |
958 | ||
4e717f5c GC |
959 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
960 | spinlock_t *lru_lock, void *arg) | |
dd1f6b2e | 961 | { |
4e717f5c GC |
962 | struct list_head *freeable = arg; |
963 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 964 | |
4e717f5c GC |
965 | /* |
966 | * we are inverting the lru lock/dentry->d_lock here, | |
967 | * so use a trylock. If we fail to get the lock, just skip | |
968 | * it | |
969 | */ | |
970 | if (!spin_trylock(&dentry->d_lock)) | |
971 | return LRU_SKIP; | |
972 | ||
973 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
974 | list_move_tail(&dentry->d_lru, freeable); | |
975 | this_cpu_dec(nr_dentry_unused); | |
976 | spin_unlock(&dentry->d_lock); | |
977 | ||
978 | return LRU_REMOVED; | |
dd1f6b2e DC |
979 | } |
980 | ||
4e717f5c | 981 | |
1da177e4 LT |
982 | /** |
983 | * shrink_dcache_sb - shrink dcache for a superblock | |
984 | * @sb: superblock | |
985 | * | |
3049cfe2 CH |
986 | * Shrink the dcache for the specified super block. This is used to free |
987 | * the dcache before unmounting a file system. | |
1da177e4 | 988 | */ |
3049cfe2 | 989 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 990 | { |
4e717f5c GC |
991 | long freed; |
992 | ||
993 | do { | |
994 | LIST_HEAD(dispose); | |
995 | ||
996 | freed = list_lru_walk(&sb->s_dentry_lru, | |
997 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); | |
3049cfe2 | 998 | |
4e717f5c GC |
999 | this_cpu_sub(nr_dentry_unused, freed); |
1000 | shrink_dentry_list(&dispose); | |
1001 | } while (freed > 0); | |
1da177e4 | 1002 | } |
ec4f8605 | 1003 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1004 | |
c636ebdb DH |
1005 | /* |
1006 | * destroy a single subtree of dentries for unmount | |
1007 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
1008 | * locking | |
1009 | */ | |
1010 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
1011 | { | |
1012 | struct dentry *parent; | |
1013 | ||
1014 | BUG_ON(!IS_ROOT(dentry)); | |
1015 | ||
c636ebdb DH |
1016 | for (;;) { |
1017 | /* descend to the first leaf in the current subtree */ | |
43c1c9cd | 1018 | while (!list_empty(&dentry->d_subdirs)) |
c636ebdb DH |
1019 | dentry = list_entry(dentry->d_subdirs.next, |
1020 | struct dentry, d_u.d_child); | |
c636ebdb DH |
1021 | |
1022 | /* consume the dentries from this leaf up through its parents | |
1023 | * until we find one with children or run out altogether */ | |
1024 | do { | |
1025 | struct inode *inode; | |
1026 | ||
f0023bc6 | 1027 | /* |
61572bb1 | 1028 | * inform the fs that this dentry is about to be |
f0023bc6 SW |
1029 | * unhashed and destroyed. |
1030 | */ | |
590fb51f YZ |
1031 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && |
1032 | !d_unhashed(dentry)) | |
61572bb1 YZ |
1033 | dentry->d_op->d_prune(dentry); |
1034 | ||
1035 | dentry_lru_del(dentry); | |
43c1c9cd DH |
1036 | __d_shrink(dentry); |
1037 | ||
98474236 | 1038 | if (dentry->d_lockref.count != 0) { |
c636ebdb DH |
1039 | printk(KERN_ERR |
1040 | "BUG: Dentry %p{i=%lx,n=%s}" | |
1041 | " still in use (%d)" | |
1042 | " [unmount of %s %s]\n", | |
1043 | dentry, | |
1044 | dentry->d_inode ? | |
1045 | dentry->d_inode->i_ino : 0UL, | |
1046 | dentry->d_name.name, | |
98474236 | 1047 | dentry->d_lockref.count, |
c636ebdb DH |
1048 | dentry->d_sb->s_type->name, |
1049 | dentry->d_sb->s_id); | |
1050 | BUG(); | |
1051 | } | |
1052 | ||
2fd6b7f5 | 1053 | if (IS_ROOT(dentry)) { |
c636ebdb | 1054 | parent = NULL; |
2fd6b7f5 NP |
1055 | list_del(&dentry->d_u.d_child); |
1056 | } else { | |
871c0067 | 1057 | parent = dentry->d_parent; |
98474236 | 1058 | parent->d_lockref.count--; |
2fd6b7f5 | 1059 | list_del(&dentry->d_u.d_child); |
871c0067 | 1060 | } |
c636ebdb | 1061 | |
c636ebdb DH |
1062 | inode = dentry->d_inode; |
1063 | if (inode) { | |
1064 | dentry->d_inode = NULL; | |
b3d9b7a3 | 1065 | hlist_del_init(&dentry->d_alias); |
c636ebdb DH |
1066 | if (dentry->d_op && dentry->d_op->d_iput) |
1067 | dentry->d_op->d_iput(dentry, inode); | |
1068 | else | |
1069 | iput(inode); | |
1070 | } | |
1071 | ||
1072 | d_free(dentry); | |
1073 | ||
1074 | /* finished when we fall off the top of the tree, | |
1075 | * otherwise we ascend to the parent and move to the | |
1076 | * next sibling if there is one */ | |
1077 | if (!parent) | |
312d3ca8 | 1078 | return; |
c636ebdb | 1079 | dentry = parent; |
c636ebdb DH |
1080 | } while (list_empty(&dentry->d_subdirs)); |
1081 | ||
1082 | dentry = list_entry(dentry->d_subdirs.next, | |
1083 | struct dentry, d_u.d_child); | |
1084 | } | |
1085 | } | |
1086 | ||
1087 | /* | |
1088 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 1089 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
1090 | * - the superblock is detached from all mountings and open files, so the |
1091 | * dentry trees will not be rearranged by the VFS | |
1092 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
1093 | * any dentries belonging to this superblock that it comes across | |
1094 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
1095 | * in this superblock | |
1096 | */ | |
1097 | void shrink_dcache_for_umount(struct super_block *sb) | |
1098 | { | |
1099 | struct dentry *dentry; | |
1100 | ||
1101 | if (down_read_trylock(&sb->s_umount)) | |
1102 | BUG(); | |
1103 | ||
1104 | dentry = sb->s_root; | |
1105 | sb->s_root = NULL; | |
98474236 | 1106 | dentry->d_lockref.count--; |
c636ebdb DH |
1107 | shrink_dcache_for_umount_subtree(dentry); |
1108 | ||
ceb5bdc2 NP |
1109 | while (!hlist_bl_empty(&sb->s_anon)) { |
1110 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
1111 | shrink_dcache_for_umount_subtree(dentry); |
1112 | } | |
1113 | } | |
1114 | ||
c826cb7d LT |
1115 | /* |
1116 | * This tries to ascend one level of parenthood, but | |
1117 | * we can race with renaming, so we need to re-check | |
1118 | * the parenthood after dropping the lock and check | |
1119 | * that the sequence number still matches. | |
1120 | */ | |
48f5ec21 | 1121 | static struct dentry *try_to_ascend(struct dentry *old, unsigned seq) |
c826cb7d LT |
1122 | { |
1123 | struct dentry *new = old->d_parent; | |
1124 | ||
1125 | rcu_read_lock(); | |
1126 | spin_unlock(&old->d_lock); | |
1127 | spin_lock(&new->d_lock); | |
1128 | ||
1129 | /* | |
1130 | * might go back up the wrong parent if we have had a rename | |
1131 | * or deletion | |
1132 | */ | |
1133 | if (new != old->d_parent || | |
b161dfa6 | 1134 | (old->d_flags & DCACHE_DENTRY_KILLED) || |
48f5ec21 | 1135 | need_seqretry(&rename_lock, seq)) { |
c826cb7d LT |
1136 | spin_unlock(&new->d_lock); |
1137 | new = NULL; | |
1138 | } | |
1139 | rcu_read_unlock(); | |
1140 | return new; | |
1141 | } | |
1142 | ||
db14fc3a MS |
1143 | /** |
1144 | * enum d_walk_ret - action to talke during tree walk | |
1145 | * @D_WALK_CONTINUE: contrinue walk | |
1146 | * @D_WALK_QUIT: quit walk | |
1147 | * @D_WALK_NORETRY: quit when retry is needed | |
1148 | * @D_WALK_SKIP: skip this dentry and its children | |
1149 | */ | |
1150 | enum d_walk_ret { | |
1151 | D_WALK_CONTINUE, | |
1152 | D_WALK_QUIT, | |
1153 | D_WALK_NORETRY, | |
1154 | D_WALK_SKIP, | |
1155 | }; | |
c826cb7d | 1156 | |
1da177e4 | 1157 | /** |
db14fc3a MS |
1158 | * d_walk - walk the dentry tree |
1159 | * @parent: start of walk | |
1160 | * @data: data passed to @enter() and @finish() | |
1161 | * @enter: callback when first entering the dentry | |
1162 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1163 | * |
db14fc3a | 1164 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1165 | */ |
db14fc3a MS |
1166 | static void d_walk(struct dentry *parent, void *data, |
1167 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1168 | void (*finish)(void *)) | |
1da177e4 | 1169 | { |
949854d0 | 1170 | struct dentry *this_parent; |
1da177e4 | 1171 | struct list_head *next; |
48f5ec21 | 1172 | unsigned seq = 0; |
db14fc3a MS |
1173 | enum d_walk_ret ret; |
1174 | bool retry = true; | |
949854d0 | 1175 | |
58db63d0 | 1176 | again: |
48f5ec21 | 1177 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1178 | this_parent = parent; |
2fd6b7f5 | 1179 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1180 | |
1181 | ret = enter(data, this_parent); | |
1182 | switch (ret) { | |
1183 | case D_WALK_CONTINUE: | |
1184 | break; | |
1185 | case D_WALK_QUIT: | |
1186 | case D_WALK_SKIP: | |
1187 | goto out_unlock; | |
1188 | case D_WALK_NORETRY: | |
1189 | retry = false; | |
1190 | break; | |
1191 | } | |
1da177e4 LT |
1192 | repeat: |
1193 | next = this_parent->d_subdirs.next; | |
1194 | resume: | |
1195 | while (next != &this_parent->d_subdirs) { | |
1196 | struct list_head *tmp = next; | |
5160ee6f | 1197 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1198 | next = tmp->next; |
2fd6b7f5 NP |
1199 | |
1200 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
db14fc3a MS |
1201 | |
1202 | ret = enter(data, dentry); | |
1203 | switch (ret) { | |
1204 | case D_WALK_CONTINUE: | |
1205 | break; | |
1206 | case D_WALK_QUIT: | |
2fd6b7f5 | 1207 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1208 | goto out_unlock; |
1209 | case D_WALK_NORETRY: | |
1210 | retry = false; | |
1211 | break; | |
1212 | case D_WALK_SKIP: | |
1213 | spin_unlock(&dentry->d_lock); | |
1214 | continue; | |
2fd6b7f5 | 1215 | } |
db14fc3a | 1216 | |
1da177e4 | 1217 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1218 | spin_unlock(&this_parent->d_lock); |
1219 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1220 | this_parent = dentry; |
2fd6b7f5 | 1221 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1222 | goto repeat; |
1223 | } | |
2fd6b7f5 | 1224 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1225 | } |
1226 | /* | |
1227 | * All done at this level ... ascend and resume the search. | |
1228 | */ | |
1229 | if (this_parent != parent) { | |
c826cb7d | 1230 | struct dentry *child = this_parent; |
48f5ec21 | 1231 | this_parent = try_to_ascend(this_parent, seq); |
c826cb7d | 1232 | if (!this_parent) |
949854d0 | 1233 | goto rename_retry; |
949854d0 | 1234 | next = child->d_u.d_child.next; |
1da177e4 LT |
1235 | goto resume; |
1236 | } | |
48f5ec21 | 1237 | if (need_seqretry(&rename_lock, seq)) { |
db14fc3a | 1238 | spin_unlock(&this_parent->d_lock); |
949854d0 | 1239 | goto rename_retry; |
db14fc3a MS |
1240 | } |
1241 | if (finish) | |
1242 | finish(data); | |
1243 | ||
1244 | out_unlock: | |
1245 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1246 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1247 | return; |
58db63d0 NP |
1248 | |
1249 | rename_retry: | |
db14fc3a MS |
1250 | if (!retry) |
1251 | return; | |
48f5ec21 | 1252 | seq = 1; |
58db63d0 | 1253 | goto again; |
1da177e4 | 1254 | } |
db14fc3a MS |
1255 | |
1256 | /* | |
1257 | * Search for at least 1 mount point in the dentry's subdirs. | |
1258 | * We descend to the next level whenever the d_subdirs | |
1259 | * list is non-empty and continue searching. | |
1260 | */ | |
1261 | ||
1262 | /** | |
1263 | * have_submounts - check for mounts over a dentry | |
1264 | * @parent: dentry to check. | |
1265 | * | |
1266 | * Return true if the parent or its subdirectories contain | |
1267 | * a mount point | |
1268 | */ | |
1269 | ||
1270 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) | |
1271 | { | |
1272 | int *ret = data; | |
1273 | if (d_mountpoint(dentry)) { | |
1274 | *ret = 1; | |
1275 | return D_WALK_QUIT; | |
1276 | } | |
1277 | return D_WALK_CONTINUE; | |
1278 | } | |
1279 | ||
1280 | int have_submounts(struct dentry *parent) | |
1281 | { | |
1282 | int ret = 0; | |
1283 | ||
1284 | d_walk(parent, &ret, check_mount, NULL); | |
1285 | ||
1286 | return ret; | |
1287 | } | |
ec4f8605 | 1288 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1289 | |
eed81007 MS |
1290 | /* |
1291 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1292 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1293 | * subtree can become unreachable). | |
1294 | * | |
1295 | * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For | |
1296 | * this reason take rename_lock and d_lock on dentry and ancestors. | |
1297 | */ | |
1298 | int d_set_mounted(struct dentry *dentry) | |
1299 | { | |
1300 | struct dentry *p; | |
1301 | int ret = -ENOENT; | |
1302 | write_seqlock(&rename_lock); | |
1303 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1304 | /* Need exclusion wrt. check_submounts_and_drop() */ | |
1305 | spin_lock(&p->d_lock); | |
1306 | if (unlikely(d_unhashed(p))) { | |
1307 | spin_unlock(&p->d_lock); | |
1308 | goto out; | |
1309 | } | |
1310 | spin_unlock(&p->d_lock); | |
1311 | } | |
1312 | spin_lock(&dentry->d_lock); | |
1313 | if (!d_unlinked(dentry)) { | |
1314 | dentry->d_flags |= DCACHE_MOUNTED; | |
1315 | ret = 0; | |
1316 | } | |
1317 | spin_unlock(&dentry->d_lock); | |
1318 | out: | |
1319 | write_sequnlock(&rename_lock); | |
1320 | return ret; | |
1321 | } | |
1322 | ||
1da177e4 | 1323 | /* |
fd517909 | 1324 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1325 | * and move any unused dentries to the end of the unused |
1326 | * list for prune_dcache(). We descend to the next level | |
1327 | * whenever the d_subdirs list is non-empty and continue | |
1328 | * searching. | |
1329 | * | |
1330 | * It returns zero iff there are no unused children, | |
1331 | * otherwise it returns the number of children moved to | |
1332 | * the end of the unused list. This may not be the total | |
1333 | * number of unused children, because select_parent can | |
1334 | * drop the lock and return early due to latency | |
1335 | * constraints. | |
1336 | */ | |
1da177e4 | 1337 | |
db14fc3a MS |
1338 | struct select_data { |
1339 | struct dentry *start; | |
1340 | struct list_head dispose; | |
1341 | int found; | |
1342 | }; | |
23044507 | 1343 | |
db14fc3a MS |
1344 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1345 | { | |
1346 | struct select_data *data = _data; | |
1347 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1348 | |
db14fc3a MS |
1349 | if (data->start == dentry) |
1350 | goto out; | |
2fd6b7f5 | 1351 | |
1da177e4 | 1352 | /* |
db14fc3a MS |
1353 | * move only zero ref count dentries to the dispose list. |
1354 | * | |
1355 | * Those which are presently on the shrink list, being processed | |
1356 | * by shrink_dentry_list(), shouldn't be moved. Otherwise the | |
1357 | * loop in shrink_dcache_parent() might not make any progress | |
1358 | * and loop forever. | |
1da177e4 | 1359 | */ |
db14fc3a MS |
1360 | if (dentry->d_lockref.count) { |
1361 | dentry_lru_del(dentry); | |
1362 | } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { | |
f6041567 DC |
1363 | dentry_lru_del(dentry); |
1364 | list_add_tail(&dentry->d_lru, &data->dispose); | |
db14fc3a MS |
1365 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
1366 | data->found++; | |
1367 | ret = D_WALK_NORETRY; | |
1da177e4 | 1368 | } |
db14fc3a MS |
1369 | /* |
1370 | * We can return to the caller if we have found some (this | |
1371 | * ensures forward progress). We'll be coming back to find | |
1372 | * the rest. | |
1373 | */ | |
1374 | if (data->found && need_resched()) | |
1375 | ret = D_WALK_QUIT; | |
1da177e4 | 1376 | out: |
db14fc3a | 1377 | return ret; |
1da177e4 LT |
1378 | } |
1379 | ||
1380 | /** | |
1381 | * shrink_dcache_parent - prune dcache | |
1382 | * @parent: parent of entries to prune | |
1383 | * | |
1384 | * Prune the dcache to remove unused children of the parent dentry. | |
1385 | */ | |
db14fc3a | 1386 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1387 | { |
db14fc3a MS |
1388 | for (;;) { |
1389 | struct select_data data; | |
1da177e4 | 1390 | |
db14fc3a MS |
1391 | INIT_LIST_HEAD(&data.dispose); |
1392 | data.start = parent; | |
1393 | data.found = 0; | |
1394 | ||
1395 | d_walk(parent, &data, select_collect, NULL); | |
1396 | if (!data.found) | |
1397 | break; | |
1398 | ||
1399 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1400 | cond_resched(); |
1401 | } | |
1da177e4 | 1402 | } |
ec4f8605 | 1403 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1404 | |
848ac114 MS |
1405 | static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry) |
1406 | { | |
1407 | struct select_data *data = _data; | |
1408 | ||
1409 | if (d_mountpoint(dentry)) { | |
1410 | data->found = -EBUSY; | |
1411 | return D_WALK_QUIT; | |
1412 | } | |
1413 | ||
1414 | return select_collect(_data, dentry); | |
1415 | } | |
1416 | ||
1417 | static void check_and_drop(void *_data) | |
1418 | { | |
1419 | struct select_data *data = _data; | |
1420 | ||
1421 | if (d_mountpoint(data->start)) | |
1422 | data->found = -EBUSY; | |
1423 | if (!data->found) | |
1424 | __d_drop(data->start); | |
1425 | } | |
1426 | ||
1427 | /** | |
1428 | * check_submounts_and_drop - prune dcache, check for submounts and drop | |
1429 | * | |
1430 | * All done as a single atomic operation relative to has_unlinked_ancestor(). | |
1431 | * Returns 0 if successfully unhashed @parent. If there were submounts then | |
1432 | * return -EBUSY. | |
1433 | * | |
1434 | * @dentry: dentry to prune and drop | |
1435 | */ | |
1436 | int check_submounts_and_drop(struct dentry *dentry) | |
1437 | { | |
1438 | int ret = 0; | |
1439 | ||
1440 | /* Negative dentries can be dropped without further checks */ | |
1441 | if (!dentry->d_inode) { | |
1442 | d_drop(dentry); | |
1443 | goto out; | |
1444 | } | |
1445 | ||
1446 | for (;;) { | |
1447 | struct select_data data; | |
1448 | ||
1449 | INIT_LIST_HEAD(&data.dispose); | |
1450 | data.start = dentry; | |
1451 | data.found = 0; | |
1452 | ||
1453 | d_walk(dentry, &data, check_and_collect, check_and_drop); | |
1454 | ret = data.found; | |
1455 | ||
1456 | if (!list_empty(&data.dispose)) | |
1457 | shrink_dentry_list(&data.dispose); | |
1458 | ||
1459 | if (ret <= 0) | |
1460 | break; | |
1461 | ||
1462 | cond_resched(); | |
1463 | } | |
1464 | ||
1465 | out: | |
1466 | return ret; | |
1467 | } | |
1468 | EXPORT_SYMBOL(check_submounts_and_drop); | |
1469 | ||
1da177e4 | 1470 | /** |
a4464dbc AV |
1471 | * __d_alloc - allocate a dcache entry |
1472 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1473 | * @name: qstr of the name |
1474 | * | |
1475 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1476 | * available. On a success the dentry is returned. The name passed in is | |
1477 | * copied and the copy passed in may be reused after this call. | |
1478 | */ | |
1479 | ||
a4464dbc | 1480 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1481 | { |
1482 | struct dentry *dentry; | |
1483 | char *dname; | |
1484 | ||
e12ba74d | 1485 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1486 | if (!dentry) |
1487 | return NULL; | |
1488 | ||
6326c71f LT |
1489 | /* |
1490 | * We guarantee that the inline name is always NUL-terminated. | |
1491 | * This way the memcpy() done by the name switching in rename | |
1492 | * will still always have a NUL at the end, even if we might | |
1493 | * be overwriting an internal NUL character | |
1494 | */ | |
1495 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 LT |
1496 | if (name->len > DNAME_INLINE_LEN-1) { |
1497 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1498 | if (!dname) { | |
1499 | kmem_cache_free(dentry_cache, dentry); | |
1500 | return NULL; | |
1501 | } | |
1502 | } else { | |
1503 | dname = dentry->d_iname; | |
1504 | } | |
1da177e4 LT |
1505 | |
1506 | dentry->d_name.len = name->len; | |
1507 | dentry->d_name.hash = name->hash; | |
1508 | memcpy(dname, name->name, name->len); | |
1509 | dname[name->len] = 0; | |
1510 | ||
6326c71f LT |
1511 | /* Make sure we always see the terminating NUL character */ |
1512 | smp_wmb(); | |
1513 | dentry->d_name.name = dname; | |
1514 | ||
98474236 | 1515 | dentry->d_lockref.count = 1; |
dea3667b | 1516 | dentry->d_flags = 0; |
1da177e4 | 1517 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1518 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1519 | dentry->d_inode = NULL; |
a4464dbc AV |
1520 | dentry->d_parent = dentry; |
1521 | dentry->d_sb = sb; | |
1da177e4 LT |
1522 | dentry->d_op = NULL; |
1523 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1524 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1525 | INIT_LIST_HEAD(&dentry->d_lru); |
1526 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
b3d9b7a3 | 1527 | INIT_HLIST_NODE(&dentry->d_alias); |
2fd6b7f5 | 1528 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1529 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1530 | |
3e880fb5 | 1531 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1532 | |
1da177e4 LT |
1533 | return dentry; |
1534 | } | |
a4464dbc AV |
1535 | |
1536 | /** | |
1537 | * d_alloc - allocate a dcache entry | |
1538 | * @parent: parent of entry to allocate | |
1539 | * @name: qstr of the name | |
1540 | * | |
1541 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1542 | * available. On a success the dentry is returned. The name passed in is | |
1543 | * copied and the copy passed in may be reused after this call. | |
1544 | */ | |
1545 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1546 | { | |
1547 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1548 | if (!dentry) | |
1549 | return NULL; | |
1550 | ||
1551 | spin_lock(&parent->d_lock); | |
1552 | /* | |
1553 | * don't need child lock because it is not subject | |
1554 | * to concurrency here | |
1555 | */ | |
1556 | __dget_dlock(parent); | |
1557 | dentry->d_parent = parent; | |
1558 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1559 | spin_unlock(&parent->d_lock); | |
1560 | ||
1561 | return dentry; | |
1562 | } | |
ec4f8605 | 1563 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1564 | |
4b936885 NP |
1565 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1566 | { | |
a4464dbc AV |
1567 | struct dentry *dentry = __d_alloc(sb, name); |
1568 | if (dentry) | |
4b936885 | 1569 | dentry->d_flags |= DCACHE_DISCONNECTED; |
4b936885 NP |
1570 | return dentry; |
1571 | } | |
1572 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1573 | ||
1da177e4 LT |
1574 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1575 | { | |
1576 | struct qstr q; | |
1577 | ||
1578 | q.name = name; | |
1579 | q.len = strlen(name); | |
1580 | q.hash = full_name_hash(q.name, q.len); | |
1581 | return d_alloc(parent, &q); | |
1582 | } | |
ef26ca97 | 1583 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1584 | |
fb045adb NP |
1585 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1586 | { | |
6f7f7caa LT |
1587 | WARN_ON_ONCE(dentry->d_op); |
1588 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1589 | DCACHE_OP_COMPARE | |
1590 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1591 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1592 | DCACHE_OP_DELETE )); |
1593 | dentry->d_op = op; | |
1594 | if (!op) | |
1595 | return; | |
1596 | if (op->d_hash) | |
1597 | dentry->d_flags |= DCACHE_OP_HASH; | |
1598 | if (op->d_compare) | |
1599 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1600 | if (op->d_revalidate) | |
1601 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1602 | if (op->d_weak_revalidate) |
1603 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1604 | if (op->d_delete) |
1605 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1606 | if (op->d_prune) |
1607 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1608 | |
1609 | } | |
1610 | EXPORT_SYMBOL(d_set_d_op); | |
1611 | ||
360da900 OH |
1612 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1613 | { | |
b23fb0a6 | 1614 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1615 | if (inode) { |
1616 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1617 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
b3d9b7a3 | 1618 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1619 | } |
360da900 | 1620 | dentry->d_inode = inode; |
31e6b01f | 1621 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1622 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1623 | fsnotify_d_instantiate(dentry, inode); |
1624 | } | |
1625 | ||
1da177e4 LT |
1626 | /** |
1627 | * d_instantiate - fill in inode information for a dentry | |
1628 | * @entry: dentry to complete | |
1629 | * @inode: inode to attach to this dentry | |
1630 | * | |
1631 | * Fill in inode information in the entry. | |
1632 | * | |
1633 | * This turns negative dentries into productive full members | |
1634 | * of society. | |
1635 | * | |
1636 | * NOTE! This assumes that the inode count has been incremented | |
1637 | * (or otherwise set) by the caller to indicate that it is now | |
1638 | * in use by the dcache. | |
1639 | */ | |
1640 | ||
1641 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1642 | { | |
b3d9b7a3 | 1643 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
873feea0 NP |
1644 | if (inode) |
1645 | spin_lock(&inode->i_lock); | |
360da900 | 1646 | __d_instantiate(entry, inode); |
873feea0 NP |
1647 | if (inode) |
1648 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1649 | security_d_instantiate(entry, inode); |
1650 | } | |
ec4f8605 | 1651 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1652 | |
1653 | /** | |
1654 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1655 | * @entry: dentry to instantiate | |
1656 | * @inode: inode to attach to this dentry | |
1657 | * | |
1658 | * Fill in inode information in the entry. On success, it returns NULL. | |
1659 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1660 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1661 | * |
1662 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1663 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1664 | * |
1665 | * This also assumes that the inode count has been incremented | |
1666 | * (or otherwise set) by the caller to indicate that it is now | |
1667 | * in use by the dcache. | |
1da177e4 | 1668 | */ |
770bfad8 DH |
1669 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1670 | struct inode *inode) | |
1da177e4 LT |
1671 | { |
1672 | struct dentry *alias; | |
1673 | int len = entry->d_name.len; | |
1674 | const char *name = entry->d_name.name; | |
1675 | unsigned int hash = entry->d_name.hash; | |
1676 | ||
770bfad8 | 1677 | if (!inode) { |
360da900 | 1678 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1679 | return NULL; |
1680 | } | |
1681 | ||
b67bfe0d | 1682 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
9abca360 NP |
1683 | /* |
1684 | * Don't need alias->d_lock here, because aliases with | |
1685 | * d_parent == entry->d_parent are not subject to name or | |
1686 | * parent changes, because the parent inode i_mutex is held. | |
1687 | */ | |
12f8ad4b | 1688 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1689 | continue; |
1690 | if (alias->d_parent != entry->d_parent) | |
1691 | continue; | |
ee983e89 LT |
1692 | if (alias->d_name.len != len) |
1693 | continue; | |
12f8ad4b | 1694 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1695 | continue; |
dc0474be | 1696 | __dget(alias); |
1da177e4 LT |
1697 | return alias; |
1698 | } | |
770bfad8 | 1699 | |
360da900 | 1700 | __d_instantiate(entry, inode); |
1da177e4 LT |
1701 | return NULL; |
1702 | } | |
770bfad8 DH |
1703 | |
1704 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1705 | { | |
1706 | struct dentry *result; | |
1707 | ||
b3d9b7a3 | 1708 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
770bfad8 | 1709 | |
873feea0 NP |
1710 | if (inode) |
1711 | spin_lock(&inode->i_lock); | |
770bfad8 | 1712 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1713 | if (inode) |
1714 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1715 | |
1716 | if (!result) { | |
1717 | security_d_instantiate(entry, inode); | |
1718 | return NULL; | |
1719 | } | |
1720 | ||
1721 | BUG_ON(!d_unhashed(result)); | |
1722 | iput(inode); | |
1723 | return result; | |
1724 | } | |
1725 | ||
1da177e4 LT |
1726 | EXPORT_SYMBOL(d_instantiate_unique); |
1727 | ||
adc0e91a AV |
1728 | struct dentry *d_make_root(struct inode *root_inode) |
1729 | { | |
1730 | struct dentry *res = NULL; | |
1731 | ||
1732 | if (root_inode) { | |
26fe5750 | 1733 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1734 | |
1735 | res = __d_alloc(root_inode->i_sb, &name); | |
1736 | if (res) | |
1737 | d_instantiate(res, root_inode); | |
1738 | else | |
1739 | iput(root_inode); | |
1740 | } | |
1741 | return res; | |
1742 | } | |
1743 | EXPORT_SYMBOL(d_make_root); | |
1744 | ||
d891eedb BF |
1745 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1746 | { | |
1747 | struct dentry *alias; | |
1748 | ||
b3d9b7a3 | 1749 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1750 | return NULL; |
b3d9b7a3 | 1751 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); |
d891eedb BF |
1752 | __dget(alias); |
1753 | return alias; | |
1754 | } | |
1755 | ||
46f72b34 SW |
1756 | /** |
1757 | * d_find_any_alias - find any alias for a given inode | |
1758 | * @inode: inode to find an alias for | |
1759 | * | |
1760 | * If any aliases exist for the given inode, take and return a | |
1761 | * reference for one of them. If no aliases exist, return %NULL. | |
1762 | */ | |
1763 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1764 | { |
1765 | struct dentry *de; | |
1766 | ||
1767 | spin_lock(&inode->i_lock); | |
1768 | de = __d_find_any_alias(inode); | |
1769 | spin_unlock(&inode->i_lock); | |
1770 | return de; | |
1771 | } | |
46f72b34 | 1772 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1773 | |
4ea3ada2 CH |
1774 | /** |
1775 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1776 | * @inode: inode to allocate the dentry for | |
1777 | * | |
1778 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1779 | * similar open by handle operations. The returned dentry may be anonymous, | |
1780 | * or may have a full name (if the inode was already in the cache). | |
1781 | * | |
1782 | * When called on a directory inode, we must ensure that the inode only ever | |
1783 | * has one dentry. If a dentry is found, that is returned instead of | |
1784 | * allocating a new one. | |
1785 | * | |
1786 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1787 | * to the dentry. In case of an error the reference on the inode is released. |
1788 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1789 | * be passed in and will be the error will be propagate to the return value, | |
1790 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1791 | */ |
1792 | struct dentry *d_obtain_alias(struct inode *inode) | |
1793 | { | |
b911a6bd | 1794 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1795 | struct dentry *tmp; |
1796 | struct dentry *res; | |
4ea3ada2 CH |
1797 | |
1798 | if (!inode) | |
44003728 | 1799 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1800 | if (IS_ERR(inode)) |
1801 | return ERR_CAST(inode); | |
1802 | ||
d891eedb | 1803 | res = d_find_any_alias(inode); |
9308a612 CH |
1804 | if (res) |
1805 | goto out_iput; | |
1806 | ||
a4464dbc | 1807 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1808 | if (!tmp) { |
1809 | res = ERR_PTR(-ENOMEM); | |
1810 | goto out_iput; | |
4ea3ada2 | 1811 | } |
b5c84bf6 | 1812 | |
873feea0 | 1813 | spin_lock(&inode->i_lock); |
d891eedb | 1814 | res = __d_find_any_alias(inode); |
9308a612 | 1815 | if (res) { |
873feea0 | 1816 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1817 | dput(tmp); |
1818 | goto out_iput; | |
1819 | } | |
1820 | ||
1821 | /* attach a disconnected dentry */ | |
1822 | spin_lock(&tmp->d_lock); | |
9308a612 CH |
1823 | tmp->d_inode = inode; |
1824 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
b3d9b7a3 | 1825 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1826 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1827 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1828 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1829 | spin_unlock(&tmp->d_lock); |
873feea0 | 1830 | spin_unlock(&inode->i_lock); |
24ff6663 | 1831 | security_d_instantiate(tmp, inode); |
9308a612 | 1832 | |
9308a612 CH |
1833 | return tmp; |
1834 | ||
1835 | out_iput: | |
24ff6663 JB |
1836 | if (res && !IS_ERR(res)) |
1837 | security_d_instantiate(res, inode); | |
9308a612 CH |
1838 | iput(inode); |
1839 | return res; | |
4ea3ada2 | 1840 | } |
adc48720 | 1841 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1842 | |
1843 | /** | |
1844 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1845 | * @inode: the inode which may have a disconnected dentry | |
1846 | * @dentry: a negative dentry which we want to point to the inode. | |
1847 | * | |
1848 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1849 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1850 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1851 | * | |
1852 | * This is needed in the lookup routine of any filesystem that is exportable | |
1853 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1854 | * | |
1855 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1856 | * is returned. This matches the expected return value of ->lookup. | |
1857 | * | |
6d4ade98 SW |
1858 | * Cluster filesystems may call this function with a negative, hashed dentry. |
1859 | * In that case, we know that the inode will be a regular file, and also this | |
1860 | * will only occur during atomic_open. So we need to check for the dentry | |
1861 | * being already hashed only in the final case. | |
1da177e4 LT |
1862 | */ |
1863 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1864 | { | |
1865 | struct dentry *new = NULL; | |
1866 | ||
a9049376 AV |
1867 | if (IS_ERR(inode)) |
1868 | return ERR_CAST(inode); | |
1869 | ||
21c0d8fd | 1870 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1871 | spin_lock(&inode->i_lock); |
32ba9c3f | 1872 | new = __d_find_alias(inode, 1); |
1da177e4 | 1873 | if (new) { |
32ba9c3f | 1874 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); |
873feea0 | 1875 | spin_unlock(&inode->i_lock); |
1da177e4 | 1876 | security_d_instantiate(new, inode); |
1da177e4 LT |
1877 | d_move(new, dentry); |
1878 | iput(inode); | |
1879 | } else { | |
873feea0 | 1880 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1881 | __d_instantiate(dentry, inode); |
873feea0 | 1882 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1883 | security_d_instantiate(dentry, inode); |
1884 | d_rehash(dentry); | |
1885 | } | |
6d4ade98 SW |
1886 | } else { |
1887 | d_instantiate(dentry, inode); | |
1888 | if (d_unhashed(dentry)) | |
1889 | d_rehash(dentry); | |
1890 | } | |
1da177e4 LT |
1891 | return new; |
1892 | } | |
ec4f8605 | 1893 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1894 | |
9403540c BN |
1895 | /** |
1896 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1897 | * @inode: the inode case-insensitive lookup has found | |
1898 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1899 | * @name: the case-exact name to be associated with the returned dentry | |
1900 | * | |
1901 | * This is to avoid filling the dcache with case-insensitive names to the | |
1902 | * same inode, only the actual correct case is stored in the dcache for | |
1903 | * case-insensitive filesystems. | |
1904 | * | |
1905 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1906 | * already exists in in the dcache, use it and return it. | |
1907 | * | |
1908 | * If no entry exists with the exact case name, allocate new dentry with | |
1909 | * the exact case, and return the spliced entry. | |
1910 | */ | |
e45b590b | 1911 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1912 | struct qstr *name) |
1913 | { | |
9403540c BN |
1914 | struct dentry *found; |
1915 | struct dentry *new; | |
1916 | ||
b6520c81 CH |
1917 | /* |
1918 | * First check if a dentry matching the name already exists, | |
1919 | * if not go ahead and create it now. | |
1920 | */ | |
9403540c | 1921 | found = d_hash_and_lookup(dentry->d_parent, name); |
4f522a24 AV |
1922 | if (unlikely(IS_ERR(found))) |
1923 | goto err_out; | |
9403540c BN |
1924 | if (!found) { |
1925 | new = d_alloc(dentry->d_parent, name); | |
1926 | if (!new) { | |
4f522a24 | 1927 | found = ERR_PTR(-ENOMEM); |
9403540c BN |
1928 | goto err_out; |
1929 | } | |
b6520c81 | 1930 | |
9403540c BN |
1931 | found = d_splice_alias(inode, new); |
1932 | if (found) { | |
1933 | dput(new); | |
1934 | return found; | |
1935 | } | |
1936 | return new; | |
1937 | } | |
b6520c81 CH |
1938 | |
1939 | /* | |
1940 | * If a matching dentry exists, and it's not negative use it. | |
1941 | * | |
1942 | * Decrement the reference count to balance the iget() done | |
1943 | * earlier on. | |
1944 | */ | |
9403540c BN |
1945 | if (found->d_inode) { |
1946 | if (unlikely(found->d_inode != inode)) { | |
1947 | /* This can't happen because bad inodes are unhashed. */ | |
1948 | BUG_ON(!is_bad_inode(inode)); | |
1949 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1950 | } | |
9403540c BN |
1951 | iput(inode); |
1952 | return found; | |
1953 | } | |
b6520c81 | 1954 | |
9403540c | 1955 | /* |
9403540c | 1956 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1957 | * already has a dentry. |
9403540c | 1958 | */ |
4513d899 AV |
1959 | new = d_splice_alias(inode, found); |
1960 | if (new) { | |
1961 | dput(found); | |
1962 | found = new; | |
9403540c | 1963 | } |
4513d899 | 1964 | return found; |
9403540c BN |
1965 | |
1966 | err_out: | |
1967 | iput(inode); | |
4f522a24 | 1968 | return found; |
9403540c | 1969 | } |
ec4f8605 | 1970 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1971 | |
12f8ad4b LT |
1972 | /* |
1973 | * Do the slow-case of the dentry name compare. | |
1974 | * | |
1975 | * Unlike the dentry_cmp() function, we need to atomically | |
da53be12 | 1976 | * load the name and length information, so that the |
12f8ad4b LT |
1977 | * filesystem can rely on them, and can use the 'name' and |
1978 | * 'len' information without worrying about walking off the | |
1979 | * end of memory etc. | |
1980 | * | |
1981 | * Thus the read_seqcount_retry() and the "duplicate" info | |
1982 | * in arguments (the low-level filesystem should not look | |
1983 | * at the dentry inode or name contents directly, since | |
1984 | * rename can change them while we're in RCU mode). | |
1985 | */ | |
1986 | enum slow_d_compare { | |
1987 | D_COMP_OK, | |
1988 | D_COMP_NOMATCH, | |
1989 | D_COMP_SEQRETRY, | |
1990 | }; | |
1991 | ||
1992 | static noinline enum slow_d_compare slow_dentry_cmp( | |
1993 | const struct dentry *parent, | |
12f8ad4b LT |
1994 | struct dentry *dentry, |
1995 | unsigned int seq, | |
1996 | const struct qstr *name) | |
1997 | { | |
1998 | int tlen = dentry->d_name.len; | |
1999 | const char *tname = dentry->d_name.name; | |
12f8ad4b LT |
2000 | |
2001 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2002 | cpu_relax(); | |
2003 | return D_COMP_SEQRETRY; | |
2004 | } | |
da53be12 | 2005 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
12f8ad4b LT |
2006 | return D_COMP_NOMATCH; |
2007 | return D_COMP_OK; | |
2008 | } | |
2009 | ||
31e6b01f NP |
2010 | /** |
2011 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2012 | * @parent: parent dentry | |
2013 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2014 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2015 | * Returns: dentry, or NULL |
2016 | * | |
2017 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2018 | * resolution (store-free path walking) design described in | |
2019 | * Documentation/filesystems/path-lookup.txt. | |
2020 | * | |
2021 | * This is not to be used outside core vfs. | |
2022 | * | |
2023 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2024 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2025 | * without taking d_lock and checking d_seq sequence count against @seq | |
2026 | * returned here. | |
2027 | * | |
15570086 | 2028 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2029 | * function. |
2030 | * | |
2031 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2032 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2033 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2034 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2035 | * |
2036 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2037 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2038 | */ |
8966be90 LT |
2039 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2040 | const struct qstr *name, | |
da53be12 | 2041 | unsigned *seqp) |
31e6b01f | 2042 | { |
26fe5750 | 2043 | u64 hashlen = name->hash_len; |
31e6b01f | 2044 | const unsigned char *str = name->name; |
26fe5750 | 2045 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 2046 | struct hlist_bl_node *node; |
31e6b01f NP |
2047 | struct dentry *dentry; |
2048 | ||
2049 | /* | |
2050 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2051 | * required to prevent single threaded performance regressions | |
2052 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2053 | * Keep the two functions in sync. | |
2054 | */ | |
2055 | ||
2056 | /* | |
2057 | * The hash list is protected using RCU. | |
2058 | * | |
2059 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2060 | * races with d_move(). | |
2061 | * | |
2062 | * It is possible that concurrent renames can mess up our list | |
2063 | * walk here and result in missing our dentry, resulting in the | |
2064 | * false-negative result. d_lookup() protects against concurrent | |
2065 | * renames using rename_lock seqlock. | |
2066 | * | |
b0a4bb83 | 2067 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2068 | */ |
b07ad996 | 2069 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2070 | unsigned seq; |
31e6b01f | 2071 | |
31e6b01f | 2072 | seqretry: |
12f8ad4b LT |
2073 | /* |
2074 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2075 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2076 | * |
2077 | * The caller must perform a seqcount check in order | |
da53be12 | 2078 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2079 | * |
2080 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2081 | * we don't wait for the sequence count to stabilize if it | |
2082 | * is in the middle of a sequence change. If we do the slow | |
2083 | * dentry compare, we will do seqretries until it is stable, | |
2084 | * and if we end up with a successful lookup, we actually | |
2085 | * want to exit RCU lookup anyway. | |
2086 | */ | |
2087 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2088 | if (dentry->d_parent != parent) |
2089 | continue; | |
2e321806 LT |
2090 | if (d_unhashed(dentry)) |
2091 | continue; | |
12f8ad4b | 2092 | |
830c0f0e | 2093 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
2094 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2095 | continue; | |
da53be12 LT |
2096 | *seqp = seq; |
2097 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
12f8ad4b LT |
2098 | case D_COMP_OK: |
2099 | return dentry; | |
2100 | case D_COMP_NOMATCH: | |
31e6b01f | 2101 | continue; |
12f8ad4b LT |
2102 | default: |
2103 | goto seqretry; | |
2104 | } | |
31e6b01f | 2105 | } |
12f8ad4b | 2106 | |
26fe5750 | 2107 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 2108 | continue; |
da53be12 | 2109 | *seqp = seq; |
26fe5750 | 2110 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 2111 | return dentry; |
31e6b01f NP |
2112 | } |
2113 | return NULL; | |
2114 | } | |
2115 | ||
1da177e4 LT |
2116 | /** |
2117 | * d_lookup - search for a dentry | |
2118 | * @parent: parent dentry | |
2119 | * @name: qstr of name we wish to find | |
b04f784e | 2120 | * Returns: dentry, or NULL |
1da177e4 | 2121 | * |
b04f784e NP |
2122 | * d_lookup searches the children of the parent dentry for the name in |
2123 | * question. If the dentry is found its reference count is incremented and the | |
2124 | * dentry is returned. The caller must use dput to free the entry when it has | |
2125 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2126 | */ |
da2d8455 | 2127 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2128 | { |
31e6b01f | 2129 | struct dentry *dentry; |
949854d0 | 2130 | unsigned seq; |
1da177e4 LT |
2131 | |
2132 | do { | |
2133 | seq = read_seqbegin(&rename_lock); | |
2134 | dentry = __d_lookup(parent, name); | |
2135 | if (dentry) | |
2136 | break; | |
2137 | } while (read_seqretry(&rename_lock, seq)); | |
2138 | return dentry; | |
2139 | } | |
ec4f8605 | 2140 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2141 | |
31e6b01f | 2142 | /** |
b04f784e NP |
2143 | * __d_lookup - search for a dentry (racy) |
2144 | * @parent: parent dentry | |
2145 | * @name: qstr of name we wish to find | |
2146 | * Returns: dentry, or NULL | |
2147 | * | |
2148 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2149 | * false-negative result due to unrelated rename activity. | |
2150 | * | |
2151 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2152 | * however it must be used carefully, eg. with a following d_lookup in | |
2153 | * the case of failure. | |
2154 | * | |
2155 | * __d_lookup callers must be commented. | |
2156 | */ | |
a713ca2a | 2157 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
2158 | { |
2159 | unsigned int len = name->len; | |
2160 | unsigned int hash = name->hash; | |
2161 | const unsigned char *str = name->name; | |
b07ad996 | 2162 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 2163 | struct hlist_bl_node *node; |
31e6b01f | 2164 | struct dentry *found = NULL; |
665a7583 | 2165 | struct dentry *dentry; |
1da177e4 | 2166 | |
31e6b01f NP |
2167 | /* |
2168 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2169 | * required to prevent single threaded performance regressions | |
2170 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2171 | * Keep the two functions in sync. | |
2172 | */ | |
2173 | ||
b04f784e NP |
2174 | /* |
2175 | * The hash list is protected using RCU. | |
2176 | * | |
2177 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2178 | * with d_move(). | |
2179 | * | |
2180 | * It is possible that concurrent renames can mess up our list | |
2181 | * walk here and result in missing our dentry, resulting in the | |
2182 | * false-negative result. d_lookup() protects against concurrent | |
2183 | * renames using rename_lock seqlock. | |
2184 | * | |
b0a4bb83 | 2185 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2186 | */ |
1da177e4 LT |
2187 | rcu_read_lock(); |
2188 | ||
b07ad996 | 2189 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2190 | |
1da177e4 LT |
2191 | if (dentry->d_name.hash != hash) |
2192 | continue; | |
1da177e4 LT |
2193 | |
2194 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2195 | if (dentry->d_parent != parent) |
2196 | goto next; | |
d0185c08 LT |
2197 | if (d_unhashed(dentry)) |
2198 | goto next; | |
2199 | ||
1da177e4 LT |
2200 | /* |
2201 | * It is safe to compare names since d_move() cannot | |
2202 | * change the qstr (protected by d_lock). | |
2203 | */ | |
fb045adb | 2204 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
2205 | int tlen = dentry->d_name.len; |
2206 | const char *tname = dentry->d_name.name; | |
da53be12 | 2207 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1da177e4 LT |
2208 | goto next; |
2209 | } else { | |
ee983e89 LT |
2210 | if (dentry->d_name.len != len) |
2211 | goto next; | |
12f8ad4b | 2212 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
2213 | goto next; |
2214 | } | |
2215 | ||
98474236 | 2216 | dentry->d_lockref.count++; |
d0185c08 | 2217 | found = dentry; |
1da177e4 LT |
2218 | spin_unlock(&dentry->d_lock); |
2219 | break; | |
2220 | next: | |
2221 | spin_unlock(&dentry->d_lock); | |
2222 | } | |
2223 | rcu_read_unlock(); | |
2224 | ||
2225 | return found; | |
2226 | } | |
2227 | ||
3e7e241f EB |
2228 | /** |
2229 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2230 | * @dir: Directory to search in | |
2231 | * @name: qstr of name we wish to find | |
2232 | * | |
4f522a24 | 2233 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2234 | */ |
2235 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2236 | { | |
3e7e241f EB |
2237 | /* |
2238 | * Check for a fs-specific hash function. Note that we must | |
2239 | * calculate the standard hash first, as the d_op->d_hash() | |
2240 | * routine may choose to leave the hash value unchanged. | |
2241 | */ | |
2242 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2243 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2244 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2245 | if (unlikely(err < 0)) |
2246 | return ERR_PTR(err); | |
3e7e241f | 2247 | } |
4f522a24 | 2248 | return d_lookup(dir, name); |
3e7e241f | 2249 | } |
4f522a24 | 2250 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2251 | |
1da177e4 | 2252 | /** |
786a5e15 | 2253 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2254 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2255 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2256 | * |
2257 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2258 | * This is used by ncpfs in its readdir implementation. | |
2259 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2260 | * |
2261 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2262 | */ |
d3a23e16 | 2263 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2264 | { |
786a5e15 | 2265 | struct dentry *child; |
d3a23e16 | 2266 | |
2fd6b7f5 | 2267 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2268 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2269 | if (dentry == child) { | |
2fd6b7f5 | 2270 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2271 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2272 | spin_unlock(&dentry->d_lock); |
2273 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2274 | return 1; |
2275 | } | |
2276 | } | |
2fd6b7f5 | 2277 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2278 | |
1da177e4 LT |
2279 | return 0; |
2280 | } | |
ec4f8605 | 2281 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2282 | |
2283 | /* | |
2284 | * When a file is deleted, we have two options: | |
2285 | * - turn this dentry into a negative dentry | |
2286 | * - unhash this dentry and free it. | |
2287 | * | |
2288 | * Usually, we want to just turn this into | |
2289 | * a negative dentry, but if anybody else is | |
2290 | * currently using the dentry or the inode | |
2291 | * we can't do that and we fall back on removing | |
2292 | * it from the hash queues and waiting for | |
2293 | * it to be deleted later when it has no users | |
2294 | */ | |
2295 | ||
2296 | /** | |
2297 | * d_delete - delete a dentry | |
2298 | * @dentry: The dentry to delete | |
2299 | * | |
2300 | * Turn the dentry into a negative dentry if possible, otherwise | |
2301 | * remove it from the hash queues so it can be deleted later | |
2302 | */ | |
2303 | ||
2304 | void d_delete(struct dentry * dentry) | |
2305 | { | |
873feea0 | 2306 | struct inode *inode; |
7a91bf7f | 2307 | int isdir = 0; |
1da177e4 LT |
2308 | /* |
2309 | * Are we the only user? | |
2310 | */ | |
357f8e65 | 2311 | again: |
1da177e4 | 2312 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2313 | inode = dentry->d_inode; |
2314 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2315 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2316 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2317 | spin_unlock(&dentry->d_lock); |
2318 | cpu_relax(); | |
2319 | goto again; | |
2320 | } | |
13e3c5e5 | 2321 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2322 | dentry_unlink_inode(dentry); |
7a91bf7f | 2323 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2324 | return; |
2325 | } | |
2326 | ||
2327 | if (!d_unhashed(dentry)) | |
2328 | __d_drop(dentry); | |
2329 | ||
2330 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2331 | |
2332 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2333 | } |
ec4f8605 | 2334 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2335 | |
b07ad996 | 2336 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2337 | { |
ceb5bdc2 | 2338 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2339 | hlist_bl_lock(b); |
dea3667b | 2340 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2341 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2342 | hlist_bl_unlock(b); |
1da177e4 LT |
2343 | } |
2344 | ||
770bfad8 DH |
2345 | static void _d_rehash(struct dentry * entry) |
2346 | { | |
2347 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2348 | } | |
2349 | ||
1da177e4 LT |
2350 | /** |
2351 | * d_rehash - add an entry back to the hash | |
2352 | * @entry: dentry to add to the hash | |
2353 | * | |
2354 | * Adds a dentry to the hash according to its name. | |
2355 | */ | |
2356 | ||
2357 | void d_rehash(struct dentry * entry) | |
2358 | { | |
1da177e4 | 2359 | spin_lock(&entry->d_lock); |
770bfad8 | 2360 | _d_rehash(entry); |
1da177e4 | 2361 | spin_unlock(&entry->d_lock); |
1da177e4 | 2362 | } |
ec4f8605 | 2363 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2364 | |
fb2d5b86 NP |
2365 | /** |
2366 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2367 | * @dentry: dentry to be updated | |
2368 | * @name: new name | |
2369 | * | |
2370 | * Update a case insensitive dentry with new case of name. | |
2371 | * | |
2372 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2373 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2374 | * lengths). | |
2375 | * | |
2376 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2377 | * keep renames and concurrent inserts, and readdir(2) away). | |
2378 | */ | |
2379 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2380 | { | |
7ebfa57f | 2381 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2382 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2383 | ||
fb2d5b86 | 2384 | spin_lock(&dentry->d_lock); |
31e6b01f | 2385 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2386 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2387 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2388 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2389 | } |
2390 | EXPORT_SYMBOL(dentry_update_name_case); | |
2391 | ||
1da177e4 LT |
2392 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2393 | { | |
2394 | if (dname_external(target)) { | |
2395 | if (dname_external(dentry)) { | |
2396 | /* | |
2397 | * Both external: swap the pointers | |
2398 | */ | |
9a8d5bb4 | 2399 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2400 | } else { |
2401 | /* | |
2402 | * dentry:internal, target:external. Steal target's | |
2403 | * storage and make target internal. | |
2404 | */ | |
321bcf92 BF |
2405 | memcpy(target->d_iname, dentry->d_name.name, |
2406 | dentry->d_name.len + 1); | |
1da177e4 LT |
2407 | dentry->d_name.name = target->d_name.name; |
2408 | target->d_name.name = target->d_iname; | |
2409 | } | |
2410 | } else { | |
2411 | if (dname_external(dentry)) { | |
2412 | /* | |
2413 | * dentry:external, target:internal. Give dentry's | |
2414 | * storage to target and make dentry internal | |
2415 | */ | |
2416 | memcpy(dentry->d_iname, target->d_name.name, | |
2417 | target->d_name.len + 1); | |
2418 | target->d_name.name = dentry->d_name.name; | |
2419 | dentry->d_name.name = dentry->d_iname; | |
2420 | } else { | |
2421 | /* | |
2422 | * Both are internal. Just copy target to dentry | |
2423 | */ | |
2424 | memcpy(dentry->d_iname, target->d_name.name, | |
2425 | target->d_name.len + 1); | |
dc711ca3 AV |
2426 | dentry->d_name.len = target->d_name.len; |
2427 | return; | |
1da177e4 LT |
2428 | } |
2429 | } | |
9a8d5bb4 | 2430 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2431 | } |
2432 | ||
2fd6b7f5 NP |
2433 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2434 | { | |
2435 | /* | |
2436 | * XXXX: do we really need to take target->d_lock? | |
2437 | */ | |
2438 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2439 | spin_lock(&target->d_parent->d_lock); | |
2440 | else { | |
2441 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2442 | spin_lock(&dentry->d_parent->d_lock); | |
2443 | spin_lock_nested(&target->d_parent->d_lock, | |
2444 | DENTRY_D_LOCK_NESTED); | |
2445 | } else { | |
2446 | spin_lock(&target->d_parent->d_lock); | |
2447 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2448 | DENTRY_D_LOCK_NESTED); | |
2449 | } | |
2450 | } | |
2451 | if (target < dentry) { | |
2452 | spin_lock_nested(&target->d_lock, 2); | |
2453 | spin_lock_nested(&dentry->d_lock, 3); | |
2454 | } else { | |
2455 | spin_lock_nested(&dentry->d_lock, 2); | |
2456 | spin_lock_nested(&target->d_lock, 3); | |
2457 | } | |
2458 | } | |
2459 | ||
2460 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2461 | struct dentry *target) | |
2462 | { | |
2463 | if (target->d_parent != dentry->d_parent) | |
2464 | spin_unlock(&dentry->d_parent->d_lock); | |
2465 | if (target->d_parent != target) | |
2466 | spin_unlock(&target->d_parent->d_lock); | |
2467 | } | |
2468 | ||
1da177e4 | 2469 | /* |
2fd6b7f5 NP |
2470 | * When switching names, the actual string doesn't strictly have to |
2471 | * be preserved in the target - because we're dropping the target | |
2472 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2473 | * the new name before we switch. | |
2474 | * | |
2475 | * Note that we have to be a lot more careful about getting the hash | |
2476 | * switched - we have to switch the hash value properly even if it | |
2477 | * then no longer matches the actual (corrupted) string of the target. | |
2478 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2479 | */ |
9eaef27b | 2480 | /* |
18367501 | 2481 | * __d_move - move a dentry |
1da177e4 LT |
2482 | * @dentry: entry to move |
2483 | * @target: new dentry | |
2484 | * | |
2485 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2486 | * dcache entries should not be moved in this way. Caller must hold |
2487 | * rename_lock, the i_mutex of the source and target directories, | |
2488 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2489 | */ |
18367501 | 2490 | static void __d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2491 | { |
1da177e4 LT |
2492 | if (!dentry->d_inode) |
2493 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2494 | ||
2fd6b7f5 NP |
2495 | BUG_ON(d_ancestor(dentry, target)); |
2496 | BUG_ON(d_ancestor(target, dentry)); | |
2497 | ||
2fd6b7f5 | 2498 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2499 | |
31e6b01f NP |
2500 | write_seqcount_begin(&dentry->d_seq); |
2501 | write_seqcount_begin(&target->d_seq); | |
2502 | ||
ceb5bdc2 NP |
2503 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2504 | ||
2505 | /* | |
2506 | * Move the dentry to the target hash queue. Don't bother checking | |
2507 | * for the same hash queue because of how unlikely it is. | |
2508 | */ | |
2509 | __d_drop(dentry); | |
789680d1 | 2510 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2511 | |
2512 | /* Unhash the target: dput() will then get rid of it */ | |
2513 | __d_drop(target); | |
2514 | ||
5160ee6f ED |
2515 | list_del(&dentry->d_u.d_child); |
2516 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2517 | |
2518 | /* Switch the names.. */ | |
2519 | switch_names(dentry, target); | |
9a8d5bb4 | 2520 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2521 | |
2522 | /* ... and switch the parents */ | |
2523 | if (IS_ROOT(dentry)) { | |
2524 | dentry->d_parent = target->d_parent; | |
2525 | target->d_parent = target; | |
5160ee6f | 2526 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2527 | } else { |
9a8d5bb4 | 2528 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2529 | |
2530 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2531 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2532 | } |
2533 | ||
5160ee6f | 2534 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2535 | |
31e6b01f NP |
2536 | write_seqcount_end(&target->d_seq); |
2537 | write_seqcount_end(&dentry->d_seq); | |
2538 | ||
2fd6b7f5 | 2539 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2540 | spin_unlock(&target->d_lock); |
c32ccd87 | 2541 | fsnotify_d_move(dentry); |
1da177e4 | 2542 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2543 | } |
2544 | ||
2545 | /* | |
2546 | * d_move - move a dentry | |
2547 | * @dentry: entry to move | |
2548 | * @target: new dentry | |
2549 | * | |
2550 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2551 | * dcache entries should not be moved in this way. See the locking |
2552 | * requirements for __d_move. | |
18367501 AV |
2553 | */ |
2554 | void d_move(struct dentry *dentry, struct dentry *target) | |
2555 | { | |
2556 | write_seqlock(&rename_lock); | |
2557 | __d_move(dentry, target); | |
1da177e4 | 2558 | write_sequnlock(&rename_lock); |
9eaef27b | 2559 | } |
ec4f8605 | 2560 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2561 | |
e2761a11 OH |
2562 | /** |
2563 | * d_ancestor - search for an ancestor | |
2564 | * @p1: ancestor dentry | |
2565 | * @p2: child dentry | |
2566 | * | |
2567 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2568 | * an ancestor of p2, else NULL. | |
9eaef27b | 2569 | */ |
e2761a11 | 2570 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2571 | { |
2572 | struct dentry *p; | |
2573 | ||
871c0067 | 2574 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2575 | if (p->d_parent == p1) |
e2761a11 | 2576 | return p; |
9eaef27b | 2577 | } |
e2761a11 | 2578 | return NULL; |
9eaef27b TM |
2579 | } |
2580 | ||
2581 | /* | |
2582 | * This helper attempts to cope with remotely renamed directories | |
2583 | * | |
2584 | * It assumes that the caller is already holding | |
18367501 | 2585 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2586 | * |
2587 | * Note: If ever the locking in lock_rename() changes, then please | |
2588 | * remember to update this too... | |
9eaef27b | 2589 | */ |
873feea0 NP |
2590 | static struct dentry *__d_unalias(struct inode *inode, |
2591 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2592 | { |
2593 | struct mutex *m1 = NULL, *m2 = NULL; | |
ee3efa91 | 2594 | struct dentry *ret = ERR_PTR(-EBUSY); |
9eaef27b TM |
2595 | |
2596 | /* If alias and dentry share a parent, then no extra locks required */ | |
2597 | if (alias->d_parent == dentry->d_parent) | |
2598 | goto out_unalias; | |
2599 | ||
9eaef27b | 2600 | /* See lock_rename() */ |
9eaef27b TM |
2601 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2602 | goto out_err; | |
2603 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2604 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2605 | goto out_err; | |
2606 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2607 | out_unalias: | |
ee3efa91 AV |
2608 | if (likely(!d_mountpoint(alias))) { |
2609 | __d_move(alias, dentry); | |
2610 | ret = alias; | |
2611 | } | |
9eaef27b | 2612 | out_err: |
873feea0 | 2613 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2614 | if (m2) |
2615 | mutex_unlock(m2); | |
2616 | if (m1) | |
2617 | mutex_unlock(m1); | |
2618 | return ret; | |
2619 | } | |
2620 | ||
770bfad8 DH |
2621 | /* |
2622 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2623 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2624 | * returns with anon->d_lock held! |
770bfad8 DH |
2625 | */ |
2626 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2627 | { | |
740da42e | 2628 | struct dentry *dparent; |
770bfad8 | 2629 | |
2fd6b7f5 | 2630 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2631 | |
31e6b01f NP |
2632 | write_seqcount_begin(&dentry->d_seq); |
2633 | write_seqcount_begin(&anon->d_seq); | |
2634 | ||
770bfad8 | 2635 | dparent = dentry->d_parent; |
770bfad8 | 2636 | |
2fd6b7f5 NP |
2637 | switch_names(dentry, anon); |
2638 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2639 | ||
740da42e AV |
2640 | dentry->d_parent = dentry; |
2641 | list_del_init(&dentry->d_u.d_child); | |
2642 | anon->d_parent = dparent; | |
9ed53b12 | 2643 | list_move(&anon->d_u.d_child, &dparent->d_subdirs); |
770bfad8 | 2644 | |
31e6b01f NP |
2645 | write_seqcount_end(&dentry->d_seq); |
2646 | write_seqcount_end(&anon->d_seq); | |
2647 | ||
2fd6b7f5 NP |
2648 | dentry_unlock_parents_for_move(anon, dentry); |
2649 | spin_unlock(&dentry->d_lock); | |
2650 | ||
2651 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2652 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2653 | } | |
2654 | ||
2655 | /** | |
2656 | * d_materialise_unique - introduce an inode into the tree | |
2657 | * @dentry: candidate dentry | |
2658 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2659 | * | |
2660 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2661 | * root directory alias in its place if there is one. Caller must hold the |
2662 | * i_mutex of the parent directory. | |
770bfad8 DH |
2663 | */ |
2664 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2665 | { | |
9eaef27b | 2666 | struct dentry *actual; |
770bfad8 DH |
2667 | |
2668 | BUG_ON(!d_unhashed(dentry)); | |
2669 | ||
770bfad8 DH |
2670 | if (!inode) { |
2671 | actual = dentry; | |
360da900 | 2672 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2673 | d_rehash(actual); |
2674 | goto out_nolock; | |
770bfad8 DH |
2675 | } |
2676 | ||
873feea0 | 2677 | spin_lock(&inode->i_lock); |
357f8e65 | 2678 | |
9eaef27b TM |
2679 | if (S_ISDIR(inode->i_mode)) { |
2680 | struct dentry *alias; | |
2681 | ||
2682 | /* Does an aliased dentry already exist? */ | |
32ba9c3f | 2683 | alias = __d_find_alias(inode, 0); |
9eaef27b TM |
2684 | if (alias) { |
2685 | actual = alias; | |
18367501 AV |
2686 | write_seqlock(&rename_lock); |
2687 | ||
2688 | if (d_ancestor(alias, dentry)) { | |
2689 | /* Check for loops */ | |
2690 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2691 | spin_unlock(&inode->i_lock); |
18367501 AV |
2692 | } else if (IS_ROOT(alias)) { |
2693 | /* Is this an anonymous mountpoint that we | |
2694 | * could splice into our tree? */ | |
9eaef27b | 2695 | __d_materialise_dentry(dentry, alias); |
18367501 | 2696 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2697 | __d_drop(alias); |
2698 | goto found; | |
18367501 AV |
2699 | } else { |
2700 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2701 | * aliasing. This drops inode->i_lock */ |
18367501 | 2702 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2703 | } |
18367501 | 2704 | write_sequnlock(&rename_lock); |
dd179946 DH |
2705 | if (IS_ERR(actual)) { |
2706 | if (PTR_ERR(actual) == -ELOOP) | |
2707 | pr_warn_ratelimited( | |
2708 | "VFS: Lookup of '%s' in %s %s" | |
2709 | " would have caused loop\n", | |
2710 | dentry->d_name.name, | |
2711 | inode->i_sb->s_type->name, | |
2712 | inode->i_sb->s_id); | |
9eaef27b | 2713 | dput(alias); |
dd179946 | 2714 | } |
9eaef27b TM |
2715 | goto out_nolock; |
2716 | } | |
770bfad8 DH |
2717 | } |
2718 | ||
2719 | /* Add a unique reference */ | |
2720 | actual = __d_instantiate_unique(dentry, inode); | |
2721 | if (!actual) | |
2722 | actual = dentry; | |
357f8e65 NP |
2723 | else |
2724 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2725 | |
770bfad8 DH |
2726 | spin_lock(&actual->d_lock); |
2727 | found: | |
2728 | _d_rehash(actual); | |
2729 | spin_unlock(&actual->d_lock); | |
873feea0 | 2730 | spin_unlock(&inode->i_lock); |
9eaef27b | 2731 | out_nolock: |
770bfad8 DH |
2732 | if (actual == dentry) { |
2733 | security_d_instantiate(dentry, inode); | |
2734 | return NULL; | |
2735 | } | |
2736 | ||
2737 | iput(inode); | |
2738 | return actual; | |
770bfad8 | 2739 | } |
ec4f8605 | 2740 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2741 | |
cdd16d02 | 2742 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2743 | { |
2744 | *buflen -= namelen; | |
2745 | if (*buflen < 0) | |
2746 | return -ENAMETOOLONG; | |
2747 | *buffer -= namelen; | |
2748 | memcpy(*buffer, str, namelen); | |
2749 | return 0; | |
2750 | } | |
2751 | ||
232d2d60 WL |
2752 | /** |
2753 | * prepend_name - prepend a pathname in front of current buffer pointer | |
2754 | * buffer: buffer pointer | |
2755 | * buflen: allocated length of the buffer | |
2756 | * name: name string and length qstr structure | |
2757 | * | |
2758 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
2759 | * make sure that either the old or the new name pointer and length are | |
2760 | * fetched. However, there may be mismatch between length and pointer. | |
2761 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
2762 | * the length is reached or a null byte is found. It also prepends "/" at | |
2763 | * the beginning of the name. The sequence number check at the caller will | |
2764 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
2765 | * due to mismatched pointer and length will be discarded. | |
2766 | */ | |
cdd16d02 MS |
2767 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2768 | { | |
232d2d60 WL |
2769 | const char *dname = ACCESS_ONCE(name->name); |
2770 | u32 dlen = ACCESS_ONCE(name->len); | |
2771 | char *p; | |
2772 | ||
2773 | if (*buflen < dlen + 1) | |
2774 | return -ENAMETOOLONG; | |
2775 | *buflen -= dlen + 1; | |
2776 | p = *buffer -= dlen + 1; | |
2777 | *p++ = '/'; | |
2778 | while (dlen--) { | |
2779 | char c = *dname++; | |
2780 | if (!c) | |
2781 | break; | |
2782 | *p++ = c; | |
2783 | } | |
2784 | return 0; | |
cdd16d02 MS |
2785 | } |
2786 | ||
1da177e4 | 2787 | /** |
208898c1 | 2788 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2789 | * @path: the dentry/vfsmount to report |
02125a82 | 2790 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2791 | * @buffer: pointer to the end of the buffer |
2792 | * @buflen: pointer to buffer length | |
552ce544 | 2793 | * |
232d2d60 WL |
2794 | * The function tries to write out the pathname without taking any lock other |
2795 | * than the RCU read lock to make sure that dentries won't go away. It only | |
2796 | * checks the sequence number of the global rename_lock as any change in the | |
2797 | * dentry's d_seq will be preceded by changes in the rename_lock sequence | |
2798 | * number. If the sequence number had been change, it will restart the whole | |
2799 | * pathname back-tracing sequence again. It performs a total of 3 trials of | |
2800 | * lockless back-tracing sequences before falling back to take the | |
2801 | * rename_lock. | |
1da177e4 | 2802 | */ |
02125a82 AV |
2803 | static int prepend_path(const struct path *path, |
2804 | const struct path *root, | |
f2eb6575 | 2805 | char **buffer, int *buflen) |
1da177e4 | 2806 | { |
9d1bc601 MS |
2807 | struct dentry *dentry = path->dentry; |
2808 | struct vfsmount *vfsmnt = path->mnt; | |
0714a533 | 2809 | struct mount *mnt = real_mount(vfsmnt); |
f2eb6575 | 2810 | int error = 0; |
232d2d60 WL |
2811 | unsigned seq = 0; |
2812 | char *bptr; | |
2813 | int blen; | |
6092d048 | 2814 | |
48f5ec21 | 2815 | rcu_read_lock(); |
232d2d60 WL |
2816 | restart: |
2817 | bptr = *buffer; | |
2818 | blen = *buflen; | |
2819 | read_seqbegin_or_lock(&rename_lock, &seq); | |
f2eb6575 | 2820 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2821 | struct dentry * parent; |
2822 | ||
1da177e4 | 2823 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2824 | /* Global root? */ |
232d2d60 WL |
2825 | if (mnt_has_parent(mnt)) { |
2826 | dentry = mnt->mnt_mountpoint; | |
2827 | mnt = mnt->mnt_parent; | |
2828 | vfsmnt = &mnt->mnt; | |
2829 | continue; | |
2830 | } | |
2831 | /* | |
2832 | * Filesystems needing to implement special "root names" | |
2833 | * should do so with ->d_dname() | |
2834 | */ | |
2835 | if (IS_ROOT(dentry) && | |
2836 | (dentry->d_name.len != 1 || | |
2837 | dentry->d_name.name[0] != '/')) { | |
2838 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2839 | (int) dentry->d_name.len, | |
2840 | dentry->d_name.name); | |
2841 | } | |
2842 | if (!error) | |
2843 | error = is_mounted(vfsmnt) ? 1 : 2; | |
2844 | break; | |
1da177e4 LT |
2845 | } |
2846 | parent = dentry->d_parent; | |
2847 | prefetch(parent); | |
232d2d60 | 2848 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
2849 | if (error) |
2850 | break; | |
2851 | ||
1da177e4 LT |
2852 | dentry = parent; |
2853 | } | |
48f5ec21 AV |
2854 | if (!(seq & 1)) |
2855 | rcu_read_unlock(); | |
2856 | if (need_seqretry(&rename_lock, seq)) { | |
2857 | seq = 1; | |
232d2d60 | 2858 | goto restart; |
48f5ec21 AV |
2859 | } |
2860 | done_seqretry(&rename_lock, seq); | |
1da177e4 | 2861 | |
232d2d60 WL |
2862 | if (error >= 0 && bptr == *buffer) { |
2863 | if (--blen < 0) | |
2864 | error = -ENAMETOOLONG; | |
2865 | else | |
2866 | *--bptr = '/'; | |
2867 | } | |
2868 | *buffer = bptr; | |
2869 | *buflen = blen; | |
7ea600b5 | 2870 | return error; |
f2eb6575 | 2871 | } |
be285c71 | 2872 | |
f2eb6575 MS |
2873 | /** |
2874 | * __d_path - return the path of a dentry | |
2875 | * @path: the dentry/vfsmount to report | |
02125a82 | 2876 | * @root: root vfsmnt/dentry |
cd956a1c | 2877 | * @buf: buffer to return value in |
f2eb6575 MS |
2878 | * @buflen: buffer length |
2879 | * | |
ffd1f4ed | 2880 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2881 | * |
2882 | * Returns a pointer into the buffer or an error code if the | |
2883 | * path was too long. | |
2884 | * | |
be148247 | 2885 | * "buflen" should be positive. |
f2eb6575 | 2886 | * |
02125a82 | 2887 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2888 | */ |
02125a82 AV |
2889 | char *__d_path(const struct path *path, |
2890 | const struct path *root, | |
f2eb6575 MS |
2891 | char *buf, int buflen) |
2892 | { | |
2893 | char *res = buf + buflen; | |
2894 | int error; | |
2895 | ||
2896 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2897 | br_read_lock(&vfsmount_lock); |
f2eb6575 | 2898 | error = prepend_path(path, root, &res, &buflen); |
7ea600b5 | 2899 | br_read_unlock(&vfsmount_lock); |
be148247 | 2900 | |
02125a82 AV |
2901 | if (error < 0) |
2902 | return ERR_PTR(error); | |
2903 | if (error > 0) | |
2904 | return NULL; | |
2905 | return res; | |
2906 | } | |
2907 | ||
2908 | char *d_absolute_path(const struct path *path, | |
2909 | char *buf, int buflen) | |
2910 | { | |
2911 | struct path root = {}; | |
2912 | char *res = buf + buflen; | |
2913 | int error; | |
2914 | ||
2915 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2916 | br_read_lock(&vfsmount_lock); |
02125a82 | 2917 | error = prepend_path(path, &root, &res, &buflen); |
7ea600b5 | 2918 | br_read_unlock(&vfsmount_lock); |
02125a82 AV |
2919 | |
2920 | if (error > 1) | |
2921 | error = -EINVAL; | |
2922 | if (error < 0) | |
f2eb6575 | 2923 | return ERR_PTR(error); |
f2eb6575 | 2924 | return res; |
1da177e4 LT |
2925 | } |
2926 | ||
ffd1f4ed MS |
2927 | /* |
2928 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2929 | */ | |
02125a82 AV |
2930 | static int path_with_deleted(const struct path *path, |
2931 | const struct path *root, | |
2932 | char **buf, int *buflen) | |
ffd1f4ed MS |
2933 | { |
2934 | prepend(buf, buflen, "\0", 1); | |
2935 | if (d_unlinked(path->dentry)) { | |
2936 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2937 | if (error) | |
2938 | return error; | |
2939 | } | |
2940 | ||
2941 | return prepend_path(path, root, buf, buflen); | |
2942 | } | |
2943 | ||
8df9d1a4 MS |
2944 | static int prepend_unreachable(char **buffer, int *buflen) |
2945 | { | |
2946 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2947 | } | |
2948 | ||
a03a8a70 JB |
2949 | /** |
2950 | * d_path - return the path of a dentry | |
cf28b486 | 2951 | * @path: path to report |
a03a8a70 JB |
2952 | * @buf: buffer to return value in |
2953 | * @buflen: buffer length | |
2954 | * | |
2955 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2956 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2957 | * | |
52afeefb AV |
2958 | * Returns a pointer into the buffer or an error code if the path was |
2959 | * too long. Note: Callers should use the returned pointer, not the passed | |
2960 | * in buffer, to use the name! The implementation often starts at an offset | |
2961 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2962 | * |
31f3e0b3 | 2963 | * "buflen" should be positive. |
a03a8a70 | 2964 | */ |
20d4fdc1 | 2965 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2966 | { |
ffd1f4ed | 2967 | char *res = buf + buflen; |
6ac08c39 | 2968 | struct path root; |
ffd1f4ed | 2969 | int error; |
1da177e4 | 2970 | |
c23fbb6b ED |
2971 | /* |
2972 | * We have various synthetic filesystems that never get mounted. On | |
2973 | * these filesystems dentries are never used for lookup purposes, and | |
2974 | * thus don't need to be hashed. They also don't need a name until a | |
2975 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2976 | * below allows us to generate a name for these objects on demand: | |
2977 | */ | |
cf28b486 JB |
2978 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2979 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2980 | |
f7ad3c6b | 2981 | get_fs_root(current->fs, &root); |
7ea600b5 | 2982 | br_read_lock(&vfsmount_lock); |
02125a82 | 2983 | error = path_with_deleted(path, &root, &res, &buflen); |
7ea600b5 | 2984 | br_read_unlock(&vfsmount_lock); |
02125a82 | 2985 | if (error < 0) |
ffd1f4ed | 2986 | res = ERR_PTR(error); |
6ac08c39 | 2987 | path_put(&root); |
1da177e4 LT |
2988 | return res; |
2989 | } | |
ec4f8605 | 2990 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2991 | |
c23fbb6b ED |
2992 | /* |
2993 | * Helper function for dentry_operations.d_dname() members | |
2994 | */ | |
2995 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2996 | const char *fmt, ...) | |
2997 | { | |
2998 | va_list args; | |
2999 | char temp[64]; | |
3000 | int sz; | |
3001 | ||
3002 | va_start(args, fmt); | |
3003 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
3004 | va_end(args); | |
3005 | ||
3006 | if (sz > sizeof(temp) || sz > buflen) | |
3007 | return ERR_PTR(-ENAMETOOLONG); | |
3008 | ||
3009 | buffer += buflen - sz; | |
3010 | return memcpy(buffer, temp, sz); | |
3011 | } | |
3012 | ||
118b2302 AV |
3013 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
3014 | { | |
3015 | char *end = buffer + buflen; | |
3016 | /* these dentries are never renamed, so d_lock is not needed */ | |
3017 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
232d2d60 | 3018 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
118b2302 AV |
3019 | prepend(&end, &buflen, "/", 1)) |
3020 | end = ERR_PTR(-ENAMETOOLONG); | |
232d2d60 | 3021 | return end; |
118b2302 AV |
3022 | } |
3023 | ||
6092d048 RP |
3024 | /* |
3025 | * Write full pathname from the root of the filesystem into the buffer. | |
3026 | */ | |
ec2447c2 | 3027 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 | 3028 | { |
232d2d60 WL |
3029 | char *end, *retval; |
3030 | int len, seq = 0; | |
3031 | int error = 0; | |
6092d048 | 3032 | |
48f5ec21 | 3033 | rcu_read_lock(); |
232d2d60 WL |
3034 | restart: |
3035 | end = buf + buflen; | |
3036 | len = buflen; | |
3037 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
3038 | if (buflen < 1) |
3039 | goto Elong; | |
3040 | /* Get '/' right */ | |
3041 | retval = end-1; | |
3042 | *retval = '/'; | |
232d2d60 | 3043 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
3044 | while (!IS_ROOT(dentry)) { |
3045 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 3046 | int error; |
6092d048 | 3047 | |
6092d048 | 3048 | prefetch(parent); |
232d2d60 WL |
3049 | error = prepend_name(&end, &len, &dentry->d_name); |
3050 | if (error) | |
3051 | break; | |
6092d048 RP |
3052 | |
3053 | retval = end; | |
3054 | dentry = parent; | |
3055 | } | |
48f5ec21 AV |
3056 | if (!(seq & 1)) |
3057 | rcu_read_unlock(); | |
3058 | if (need_seqretry(&rename_lock, seq)) { | |
3059 | seq = 1; | |
232d2d60 | 3060 | goto restart; |
48f5ec21 AV |
3061 | } |
3062 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3063 | if (error) |
3064 | goto Elong; | |
c103135c AV |
3065 | return retval; |
3066 | Elong: | |
3067 | return ERR_PTR(-ENAMETOOLONG); | |
3068 | } | |
ec2447c2 NP |
3069 | |
3070 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3071 | { | |
232d2d60 | 3072 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3073 | } |
3074 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3075 | |
3076 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3077 | { | |
3078 | char *p = NULL; | |
3079 | char *retval; | |
3080 | ||
c103135c AV |
3081 | if (d_unlinked(dentry)) { |
3082 | p = buf + buflen; | |
3083 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3084 | goto Elong; | |
3085 | buflen++; | |
3086 | } | |
3087 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3088 | if (!IS_ERR(retval) && p) |
3089 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3090 | return retval; |
3091 | Elong: | |
6092d048 RP |
3092 | return ERR_PTR(-ENAMETOOLONG); |
3093 | } | |
3094 | ||
1da177e4 LT |
3095 | /* |
3096 | * NOTE! The user-level library version returns a | |
3097 | * character pointer. The kernel system call just | |
3098 | * returns the length of the buffer filled (which | |
3099 | * includes the ending '\0' character), or a negative | |
3100 | * error value. So libc would do something like | |
3101 | * | |
3102 | * char *getcwd(char * buf, size_t size) | |
3103 | * { | |
3104 | * int retval; | |
3105 | * | |
3106 | * retval = sys_getcwd(buf, size); | |
3107 | * if (retval >= 0) | |
3108 | * return buf; | |
3109 | * errno = -retval; | |
3110 | * return NULL; | |
3111 | * } | |
3112 | */ | |
3cdad428 | 3113 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3114 | { |
552ce544 | 3115 | int error; |
6ac08c39 | 3116 | struct path pwd, root; |
552ce544 | 3117 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
3118 | |
3119 | if (!page) | |
3120 | return -ENOMEM; | |
3121 | ||
f7ad3c6b | 3122 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 3123 | |
552ce544 | 3124 | error = -ENOENT; |
7ea600b5 | 3125 | br_read_lock(&vfsmount_lock); |
f3da392e | 3126 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3127 | unsigned long len; |
8df9d1a4 MS |
3128 | char *cwd = page + PAGE_SIZE; |
3129 | int buflen = PAGE_SIZE; | |
1da177e4 | 3130 | |
8df9d1a4 | 3131 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3132 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
7ea600b5 | 3133 | br_read_unlock(&vfsmount_lock); |
552ce544 | 3134 | |
02125a82 | 3135 | if (error < 0) |
552ce544 LT |
3136 | goto out; |
3137 | ||
8df9d1a4 | 3138 | /* Unreachable from current root */ |
02125a82 | 3139 | if (error > 0) { |
8df9d1a4 MS |
3140 | error = prepend_unreachable(&cwd, &buflen); |
3141 | if (error) | |
3142 | goto out; | |
3143 | } | |
3144 | ||
552ce544 LT |
3145 | error = -ERANGE; |
3146 | len = PAGE_SIZE + page - cwd; | |
3147 | if (len <= size) { | |
3148 | error = len; | |
3149 | if (copy_to_user(buf, cwd, len)) | |
3150 | error = -EFAULT; | |
3151 | } | |
949854d0 | 3152 | } else { |
7ea600b5 | 3153 | br_read_unlock(&vfsmount_lock); |
949854d0 | 3154 | } |
1da177e4 LT |
3155 | |
3156 | out: | |
6ac08c39 JB |
3157 | path_put(&pwd); |
3158 | path_put(&root); | |
1da177e4 LT |
3159 | free_page((unsigned long) page); |
3160 | return error; | |
3161 | } | |
3162 | ||
3163 | /* | |
3164 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3165 | * | |
3166 | * Trivially implemented using the dcache structure | |
3167 | */ | |
3168 | ||
3169 | /** | |
3170 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3171 | * @new_dentry: new dentry | |
3172 | * @old_dentry: old dentry | |
3173 | * | |
3174 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
3175 | * Returns 0 otherwise. | |
3176 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
3177 | */ | |
3178 | ||
e2761a11 | 3179 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
3180 | { |
3181 | int result; | |
949854d0 | 3182 | unsigned seq; |
1da177e4 | 3183 | |
e2761a11 OH |
3184 | if (new_dentry == old_dentry) |
3185 | return 1; | |
3186 | ||
e2761a11 | 3187 | do { |
1da177e4 | 3188 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3189 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3190 | /* |
3191 | * Need rcu_readlock to protect against the d_parent trashing | |
3192 | * due to d_move | |
3193 | */ | |
3194 | rcu_read_lock(); | |
e2761a11 | 3195 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 3196 | result = 1; |
e2761a11 OH |
3197 | else |
3198 | result = 0; | |
949854d0 | 3199 | rcu_read_unlock(); |
1da177e4 | 3200 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3201 | |
3202 | return result; | |
3203 | } | |
3204 | ||
db14fc3a | 3205 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3206 | { |
db14fc3a MS |
3207 | struct dentry *root = data; |
3208 | if (dentry != root) { | |
3209 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3210 | return D_WALK_SKIP; | |
1da177e4 | 3211 | |
01ddc4ed MS |
3212 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3213 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3214 | dentry->d_lockref.count--; | |
3215 | } | |
1da177e4 | 3216 | } |
db14fc3a MS |
3217 | return D_WALK_CONTINUE; |
3218 | } | |
58db63d0 | 3219 | |
db14fc3a MS |
3220 | void d_genocide(struct dentry *parent) |
3221 | { | |
3222 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3223 | } |
3224 | ||
60545d0d | 3225 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3226 | { |
60545d0d AV |
3227 | inode_dec_link_count(inode); |
3228 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
3229 | !hlist_unhashed(&dentry->d_alias) || | |
3230 | !d_unlinked(dentry)); | |
3231 | spin_lock(&dentry->d_parent->d_lock); | |
3232 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3233 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3234 | (unsigned long long)inode->i_ino); | |
3235 | spin_unlock(&dentry->d_lock); | |
3236 | spin_unlock(&dentry->d_parent->d_lock); | |
3237 | d_instantiate(dentry, inode); | |
1da177e4 | 3238 | } |
60545d0d | 3239 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3240 | |
3241 | static __initdata unsigned long dhash_entries; | |
3242 | static int __init set_dhash_entries(char *str) | |
3243 | { | |
3244 | if (!str) | |
3245 | return 0; | |
3246 | dhash_entries = simple_strtoul(str, &str, 0); | |
3247 | return 1; | |
3248 | } | |
3249 | __setup("dhash_entries=", set_dhash_entries); | |
3250 | ||
3251 | static void __init dcache_init_early(void) | |
3252 | { | |
074b8517 | 3253 | unsigned int loop; |
1da177e4 LT |
3254 | |
3255 | /* If hashes are distributed across NUMA nodes, defer | |
3256 | * hash allocation until vmalloc space is available. | |
3257 | */ | |
3258 | if (hashdist) | |
3259 | return; | |
3260 | ||
3261 | dentry_hashtable = | |
3262 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3263 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3264 | dhash_entries, |
3265 | 13, | |
3266 | HASH_EARLY, | |
3267 | &d_hash_shift, | |
3268 | &d_hash_mask, | |
31fe62b9 | 3269 | 0, |
1da177e4 LT |
3270 | 0); |
3271 | ||
074b8517 | 3272 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3273 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3274 | } |
3275 | ||
74bf17cf | 3276 | static void __init dcache_init(void) |
1da177e4 | 3277 | { |
074b8517 | 3278 | unsigned int loop; |
1da177e4 LT |
3279 | |
3280 | /* | |
3281 | * A constructor could be added for stable state like the lists, | |
3282 | * but it is probably not worth it because of the cache nature | |
3283 | * of the dcache. | |
3284 | */ | |
0a31bd5f CL |
3285 | dentry_cache = KMEM_CACHE(dentry, |
3286 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3287 | |
3288 | /* Hash may have been set up in dcache_init_early */ | |
3289 | if (!hashdist) | |
3290 | return; | |
3291 | ||
3292 | dentry_hashtable = | |
3293 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3294 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3295 | dhash_entries, |
3296 | 13, | |
3297 | 0, | |
3298 | &d_hash_shift, | |
3299 | &d_hash_mask, | |
31fe62b9 | 3300 | 0, |
1da177e4 LT |
3301 | 0); |
3302 | ||
074b8517 | 3303 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3304 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3305 | } |
3306 | ||
3307 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3308 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3309 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3310 | |
1da177e4 LT |
3311 | EXPORT_SYMBOL(d_genocide); |
3312 | ||
1da177e4 LT |
3313 | void __init vfs_caches_init_early(void) |
3314 | { | |
3315 | dcache_init_early(); | |
3316 | inode_init_early(); | |
3317 | } | |
3318 | ||
3319 | void __init vfs_caches_init(unsigned long mempages) | |
3320 | { | |
3321 | unsigned long reserve; | |
3322 | ||
3323 | /* Base hash sizes on available memory, with a reserve equal to | |
3324 | 150% of current kernel size */ | |
3325 | ||
3326 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3327 | mempages -= reserve; | |
3328 | ||
3329 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3330 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3331 | |
74bf17cf DC |
3332 | dcache_init(); |
3333 | inode_init(); | |
1da177e4 | 3334 | files_init(mempages); |
74bf17cf | 3335 | mnt_init(); |
1da177e4 LT |
3336 | bdev_cache_init(); |
3337 | chrdev_init(); | |
3338 | } |