]>
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> |
07f3f05c | 40 | #include "internal.h" |
b2dba1af | 41 | #include "mount.h" |
1da177e4 | 42 | |
789680d1 NP |
43 | /* |
44 | * Usage: | |
873feea0 NP |
45 | * dcache->d_inode->i_lock protects: |
46 | * - i_dentry, d_alias, d_inode of aliases | |
ceb5bdc2 NP |
47 | * dcache_hash_bucket lock protects: |
48 | * - the dcache hash table | |
49 | * s_anon bl list spinlock protects: | |
50 | * - the s_anon list (see __d_drop) | |
23044507 NP |
51 | * dcache_lru_lock protects: |
52 | * - the dcache lru lists and counters | |
53 | * d_lock protects: | |
54 | * - d_flags | |
55 | * - d_name | |
56 | * - d_lru | |
b7ab39f6 | 57 | * - d_count |
da502956 | 58 | * - d_unhashed() |
2fd6b7f5 NP |
59 | * - d_parent and d_subdirs |
60 | * - childrens' d_child and d_parent | |
b23fb0a6 | 61 | * - d_alias, d_inode |
789680d1 NP |
62 | * |
63 | * Ordering: | |
873feea0 | 64 | * dentry->d_inode->i_lock |
b5c84bf6 NP |
65 | * dentry->d_lock |
66 | * dcache_lru_lock | |
ceb5bdc2 NP |
67 | * dcache_hash_bucket lock |
68 | * s_anon lock | |
789680d1 | 69 | * |
da502956 NP |
70 | * If there is an ancestor relationship: |
71 | * dentry->d_parent->...->d_parent->d_lock | |
72 | * ... | |
73 | * dentry->d_parent->d_lock | |
74 | * dentry->d_lock | |
75 | * | |
76 | * If no ancestor relationship: | |
789680d1 NP |
77 | * if (dentry1 < dentry2) |
78 | * dentry1->d_lock | |
79 | * dentry2->d_lock | |
80 | */ | |
fa3536cc | 81 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
82 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
83 | ||
23044507 | 84 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock); |
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 | /* |
f0023bc6 | 360 | * dentry_lru_(add|del|prune|move_tail) 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))) { |
23044507 | 365 | spin_lock(&dcache_lru_lock); |
8aab6a27 | 366 | dentry->d_flags |= DCACHE_LRU_LIST; |
a4633357 CH |
367 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
368 | dentry->d_sb->s_nr_dentry_unused++; | |
62d36c77 | 369 | this_cpu_inc(nr_dentry_unused); |
23044507 | 370 | spin_unlock(&dcache_lru_lock); |
a4633357 | 371 | } |
da3bbdd4 KM |
372 | } |
373 | ||
23044507 NP |
374 | static void __dentry_lru_del(struct dentry *dentry) |
375 | { | |
376 | list_del_init(&dentry->d_lru); | |
8aab6a27 | 377 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); |
23044507 | 378 | dentry->d_sb->s_nr_dentry_unused--; |
62d36c77 | 379 | this_cpu_dec(nr_dentry_unused); |
23044507 NP |
380 | } |
381 | ||
f0023bc6 SW |
382 | /* |
383 | * Remove a dentry with references from the LRU. | |
384 | */ | |
da3bbdd4 KM |
385 | static void dentry_lru_del(struct dentry *dentry) |
386 | { | |
387 | if (!list_empty(&dentry->d_lru)) { | |
23044507 NP |
388 | spin_lock(&dcache_lru_lock); |
389 | __dentry_lru_del(dentry); | |
390 | spin_unlock(&dcache_lru_lock); | |
da3bbdd4 KM |
391 | } |
392 | } | |
393 | ||
b48f03b3 | 394 | static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list) |
da3bbdd4 | 395 | { |
23044507 | 396 | spin_lock(&dcache_lru_lock); |
a4633357 | 397 | if (list_empty(&dentry->d_lru)) { |
8aab6a27 | 398 | dentry->d_flags |= DCACHE_LRU_LIST; |
b48f03b3 | 399 | list_add_tail(&dentry->d_lru, list); |
a4633357 | 400 | dentry->d_sb->s_nr_dentry_unused++; |
62d36c77 | 401 | this_cpu_inc(nr_dentry_unused); |
a4633357 | 402 | } else { |
b48f03b3 | 403 | list_move_tail(&dentry->d_lru, list); |
da3bbdd4 | 404 | } |
23044507 | 405 | spin_unlock(&dcache_lru_lock); |
da3bbdd4 KM |
406 | } |
407 | ||
d52b9086 MS |
408 | /** |
409 | * d_kill - kill dentry and return parent | |
410 | * @dentry: dentry to kill | |
ff5fdb61 | 411 | * @parent: parent dentry |
d52b9086 | 412 | * |
31f3e0b3 | 413 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
414 | * |
415 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 416 | * |
b5c84bf6 NP |
417 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
418 | * d_kill. | |
d52b9086 | 419 | */ |
2fd6b7f5 | 420 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 421 | __releases(dentry->d_lock) |
2fd6b7f5 | 422 | __releases(parent->d_lock) |
873feea0 | 423 | __releases(dentry->d_inode->i_lock) |
d52b9086 | 424 | { |
d52b9086 | 425 | list_del(&dentry->d_u.d_child); |
c83ce989 TM |
426 | /* |
427 | * Inform try_to_ascend() that we are no longer attached to the | |
428 | * dentry tree | |
429 | */ | |
b161dfa6 | 430 | dentry->d_flags |= DCACHE_DENTRY_KILLED; |
2fd6b7f5 NP |
431 | if (parent) |
432 | spin_unlock(&parent->d_lock); | |
d52b9086 | 433 | dentry_iput(dentry); |
b7ab39f6 NP |
434 | /* |
435 | * dentry_iput drops the locks, at which point nobody (except | |
436 | * transient RCU lookups) can reach this dentry. | |
437 | */ | |
d52b9086 | 438 | d_free(dentry); |
871c0067 | 439 | return parent; |
d52b9086 MS |
440 | } |
441 | ||
c6627c60 DH |
442 | /* |
443 | * Unhash a dentry without inserting an RCU walk barrier or checking that | |
444 | * dentry->d_lock is locked. The caller must take care of that, if | |
445 | * appropriate. | |
446 | */ | |
447 | static void __d_shrink(struct dentry *dentry) | |
448 | { | |
449 | if (!d_unhashed(dentry)) { | |
450 | struct hlist_bl_head *b; | |
451 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
452 | b = &dentry->d_sb->s_anon; | |
453 | else | |
454 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
455 | ||
456 | hlist_bl_lock(b); | |
457 | __hlist_bl_del(&dentry->d_hash); | |
458 | dentry->d_hash.pprev = NULL; | |
459 | hlist_bl_unlock(b); | |
460 | } | |
461 | } | |
462 | ||
789680d1 NP |
463 | /** |
464 | * d_drop - drop a dentry | |
465 | * @dentry: dentry to drop | |
466 | * | |
467 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
468 | * be found through a VFS lookup any more. Note that this is different from | |
469 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
470 | * possible, giving a successful _negative_ lookup, while d_drop will | |
471 | * just make the cache lookup fail. | |
472 | * | |
473 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
474 | * reason (NFS timeouts or autofs deletes). | |
475 | * | |
476 | * __d_drop requires dentry->d_lock. | |
477 | */ | |
478 | void __d_drop(struct dentry *dentry) | |
479 | { | |
dea3667b | 480 | if (!d_unhashed(dentry)) { |
c6627c60 | 481 | __d_shrink(dentry); |
dea3667b | 482 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
483 | } |
484 | } | |
485 | EXPORT_SYMBOL(__d_drop); | |
486 | ||
487 | void d_drop(struct dentry *dentry) | |
488 | { | |
789680d1 NP |
489 | spin_lock(&dentry->d_lock); |
490 | __d_drop(dentry); | |
491 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
492 | } |
493 | EXPORT_SYMBOL(d_drop); | |
494 | ||
77812a1e NP |
495 | /* |
496 | * Finish off a dentry we've decided to kill. | |
497 | * dentry->d_lock must be held, returns with it unlocked. | |
498 | * If ref is non-zero, then decrement the refcount too. | |
499 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
500 | */ | |
0d98439e | 501 | static inline struct dentry *dentry_kill(struct dentry *dentry) |
77812a1e NP |
502 | __releases(dentry->d_lock) |
503 | { | |
873feea0 | 504 | struct inode *inode; |
77812a1e NP |
505 | struct dentry *parent; |
506 | ||
873feea0 NP |
507 | inode = dentry->d_inode; |
508 | if (inode && !spin_trylock(&inode->i_lock)) { | |
77812a1e NP |
509 | relock: |
510 | spin_unlock(&dentry->d_lock); | |
511 | cpu_relax(); | |
512 | return dentry; /* try again with same dentry */ | |
513 | } | |
514 | if (IS_ROOT(dentry)) | |
515 | parent = NULL; | |
516 | else | |
517 | parent = dentry->d_parent; | |
518 | if (parent && !spin_trylock(&parent->d_lock)) { | |
873feea0 NP |
519 | if (inode) |
520 | spin_unlock(&inode->i_lock); | |
77812a1e NP |
521 | goto relock; |
522 | } | |
31e6b01f | 523 | |
0d98439e LT |
524 | /* |
525 | * The dentry is now unrecoverably dead to the world. | |
526 | */ | |
527 | lockref_mark_dead(&dentry->d_lockref); | |
528 | ||
f0023bc6 | 529 | /* |
f0023bc6 SW |
530 | * inform the fs via d_prune that this dentry is about to be |
531 | * unhashed and destroyed. | |
532 | */ | |
590fb51f | 533 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry)) |
61572bb1 YZ |
534 | dentry->d_op->d_prune(dentry); |
535 | ||
536 | dentry_lru_del(dentry); | |
77812a1e NP |
537 | /* if it was on the hash then remove it */ |
538 | __d_drop(dentry); | |
539 | return d_kill(dentry, parent); | |
540 | } | |
541 | ||
1da177e4 LT |
542 | /* |
543 | * This is dput | |
544 | * | |
545 | * This is complicated by the fact that we do not want to put | |
546 | * dentries that are no longer on any hash chain on the unused | |
547 | * list: we'd much rather just get rid of them immediately. | |
548 | * | |
549 | * However, that implies that we have to traverse the dentry | |
550 | * tree upwards to the parents which might _also_ now be | |
551 | * scheduled for deletion (it may have been only waiting for | |
552 | * its last child to go away). | |
553 | * | |
554 | * This tail recursion is done by hand as we don't want to depend | |
555 | * on the compiler to always get this right (gcc generally doesn't). | |
556 | * Real recursion would eat up our stack space. | |
557 | */ | |
558 | ||
559 | /* | |
560 | * dput - release a dentry | |
561 | * @dentry: dentry to release | |
562 | * | |
563 | * Release a dentry. This will drop the usage count and if appropriate | |
564 | * call the dentry unlink method as well as removing it from the queues and | |
565 | * releasing its resources. If the parent dentries were scheduled for release | |
566 | * they too may now get deleted. | |
1da177e4 | 567 | */ |
1da177e4 LT |
568 | void dput(struct dentry *dentry) |
569 | { | |
8aab6a27 | 570 | if (unlikely(!dentry)) |
1da177e4 LT |
571 | return; |
572 | ||
573 | repeat: | |
98474236 | 574 | if (lockref_put_or_lock(&dentry->d_lockref)) |
1da177e4 | 575 | return; |
1da177e4 | 576 | |
8aab6a27 LT |
577 | /* Unreachable? Get rid of it */ |
578 | if (unlikely(d_unhashed(dentry))) | |
579 | goto kill_it; | |
580 | ||
581 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
1da177e4 | 582 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 583 | goto kill_it; |
1da177e4 | 584 | } |
265ac902 | 585 | |
39e3c955 | 586 | dentry->d_flags |= DCACHE_REFERENCED; |
a4633357 | 587 | dentry_lru_add(dentry); |
265ac902 | 588 | |
98474236 | 589 | dentry->d_lockref.count--; |
61f3dee4 | 590 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
591 | return; |
592 | ||
d52b9086 | 593 | kill_it: |
0d98439e | 594 | dentry = dentry_kill(dentry); |
d52b9086 MS |
595 | if (dentry) |
596 | goto repeat; | |
1da177e4 | 597 | } |
ec4f8605 | 598 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
599 | |
600 | /** | |
601 | * d_invalidate - invalidate a dentry | |
602 | * @dentry: dentry to invalidate | |
603 | * | |
604 | * Try to invalidate the dentry if it turns out to be | |
605 | * possible. If there are other dentries that can be | |
606 | * reached through this one we can't delete it and we | |
607 | * return -EBUSY. On success we return 0. | |
608 | * | |
609 | * no dcache lock. | |
610 | */ | |
611 | ||
612 | int d_invalidate(struct dentry * dentry) | |
613 | { | |
614 | /* | |
615 | * If it's already been dropped, return OK. | |
616 | */ | |
da502956 | 617 | spin_lock(&dentry->d_lock); |
1da177e4 | 618 | if (d_unhashed(dentry)) { |
da502956 | 619 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
620 | return 0; |
621 | } | |
622 | /* | |
623 | * Check whether to do a partial shrink_dcache | |
624 | * to get rid of unused child entries. | |
625 | */ | |
626 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 627 | spin_unlock(&dentry->d_lock); |
1da177e4 | 628 | shrink_dcache_parent(dentry); |
da502956 | 629 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
630 | } |
631 | ||
632 | /* | |
633 | * Somebody else still using it? | |
634 | * | |
635 | * If it's a directory, we can't drop it | |
636 | * for fear of somebody re-populating it | |
637 | * with children (even though dropping it | |
638 | * would make it unreachable from the root, | |
639 | * we might still populate it if it was a | |
640 | * working directory or similar). | |
50e69630 AV |
641 | * We also need to leave mountpoints alone, |
642 | * directory or not. | |
1da177e4 | 643 | */ |
98474236 | 644 | if (dentry->d_lockref.count > 1 && dentry->d_inode) { |
50e69630 | 645 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { |
1da177e4 | 646 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
647 | return -EBUSY; |
648 | } | |
649 | } | |
650 | ||
651 | __d_drop(dentry); | |
652 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
653 | return 0; |
654 | } | |
ec4f8605 | 655 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 656 | |
b5c84bf6 | 657 | /* This must be called with d_lock held */ |
dc0474be | 658 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 659 | { |
98474236 | 660 | dentry->d_lockref.count++; |
23044507 NP |
661 | } |
662 | ||
dc0474be | 663 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 664 | { |
98474236 | 665 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
666 | } |
667 | ||
b7ab39f6 NP |
668 | struct dentry *dget_parent(struct dentry *dentry) |
669 | { | |
df3d0bbc | 670 | int gotref; |
b7ab39f6 NP |
671 | struct dentry *ret; |
672 | ||
df3d0bbc WL |
673 | /* |
674 | * Do optimistic parent lookup without any | |
675 | * locking. | |
676 | */ | |
677 | rcu_read_lock(); | |
678 | ret = ACCESS_ONCE(dentry->d_parent); | |
679 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
680 | rcu_read_unlock(); | |
681 | if (likely(gotref)) { | |
682 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
683 | return ret; | |
684 | dput(ret); | |
685 | } | |
686 | ||
b7ab39f6 | 687 | repeat: |
a734eb45 NP |
688 | /* |
689 | * Don't need rcu_dereference because we re-check it was correct under | |
690 | * the lock. | |
691 | */ | |
692 | rcu_read_lock(); | |
b7ab39f6 | 693 | ret = dentry->d_parent; |
a734eb45 NP |
694 | spin_lock(&ret->d_lock); |
695 | if (unlikely(ret != dentry->d_parent)) { | |
696 | spin_unlock(&ret->d_lock); | |
697 | rcu_read_unlock(); | |
b7ab39f6 NP |
698 | goto repeat; |
699 | } | |
a734eb45 | 700 | rcu_read_unlock(); |
98474236 WL |
701 | BUG_ON(!ret->d_lockref.count); |
702 | ret->d_lockref.count++; | |
b7ab39f6 | 703 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
704 | return ret; |
705 | } | |
706 | EXPORT_SYMBOL(dget_parent); | |
707 | ||
1da177e4 LT |
708 | /** |
709 | * d_find_alias - grab a hashed alias of inode | |
710 | * @inode: inode in question | |
32ba9c3f LT |
711 | * @want_discon: flag, used by d_splice_alias, to request |
712 | * that only a DISCONNECTED alias be returned. | |
1da177e4 LT |
713 | * |
714 | * If inode has a hashed alias, or is a directory and has any alias, | |
715 | * acquire the reference to alias and return it. Otherwise return NULL. | |
716 | * Notice that if inode is a directory there can be only one alias and | |
717 | * it can be unhashed only if it has no children, or if it is the root | |
718 | * of a filesystem. | |
719 | * | |
21c0d8fd | 720 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
32ba9c3f LT |
721 | * any other hashed alias over that one unless @want_discon is set, |
722 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. | |
1da177e4 | 723 | */ |
32ba9c3f | 724 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 725 | { |
da502956 | 726 | struct dentry *alias, *discon_alias; |
1da177e4 | 727 | |
da502956 NP |
728 | again: |
729 | discon_alias = NULL; | |
b67bfe0d | 730 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
da502956 | 731 | spin_lock(&alias->d_lock); |
1da177e4 | 732 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 733 | if (IS_ROOT(alias) && |
da502956 | 734 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 735 | discon_alias = alias; |
32ba9c3f | 736 | } else if (!want_discon) { |
dc0474be | 737 | __dget_dlock(alias); |
da502956 NP |
738 | spin_unlock(&alias->d_lock); |
739 | return alias; | |
740 | } | |
741 | } | |
742 | spin_unlock(&alias->d_lock); | |
743 | } | |
744 | if (discon_alias) { | |
745 | alias = discon_alias; | |
746 | spin_lock(&alias->d_lock); | |
747 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
748 | if (IS_ROOT(alias) && | |
749 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 750 | __dget_dlock(alias); |
da502956 | 751 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
752 | return alias; |
753 | } | |
754 | } | |
da502956 NP |
755 | spin_unlock(&alias->d_lock); |
756 | goto again; | |
1da177e4 | 757 | } |
da502956 | 758 | return NULL; |
1da177e4 LT |
759 | } |
760 | ||
da502956 | 761 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 762 | { |
214fda1f DH |
763 | struct dentry *de = NULL; |
764 | ||
b3d9b7a3 | 765 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 766 | spin_lock(&inode->i_lock); |
32ba9c3f | 767 | de = __d_find_alias(inode, 0); |
873feea0 | 768 | spin_unlock(&inode->i_lock); |
214fda1f | 769 | } |
1da177e4 LT |
770 | return de; |
771 | } | |
ec4f8605 | 772 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
773 | |
774 | /* | |
775 | * Try to kill dentries associated with this inode. | |
776 | * WARNING: you must own a reference to inode. | |
777 | */ | |
778 | void d_prune_aliases(struct inode *inode) | |
779 | { | |
0cdca3f9 | 780 | struct dentry *dentry; |
1da177e4 | 781 | restart: |
873feea0 | 782 | spin_lock(&inode->i_lock); |
b67bfe0d | 783 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 784 | spin_lock(&dentry->d_lock); |
98474236 | 785 | if (!dentry->d_lockref.count) { |
590fb51f YZ |
786 | /* |
787 | * inform the fs via d_prune that this dentry | |
788 | * is about to be unhashed and destroyed. | |
789 | */ | |
790 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && | |
791 | !d_unhashed(dentry)) | |
792 | dentry->d_op->d_prune(dentry); | |
793 | ||
dc0474be | 794 | __dget_dlock(dentry); |
1da177e4 LT |
795 | __d_drop(dentry); |
796 | spin_unlock(&dentry->d_lock); | |
873feea0 | 797 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
798 | dput(dentry); |
799 | goto restart; | |
800 | } | |
801 | spin_unlock(&dentry->d_lock); | |
802 | } | |
873feea0 | 803 | spin_unlock(&inode->i_lock); |
1da177e4 | 804 | } |
ec4f8605 | 805 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
806 | |
807 | /* | |
77812a1e NP |
808 | * Try to throw away a dentry - free the inode, dput the parent. |
809 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
810 | * Releases dentry->d_lock. | |
d702ccb3 | 811 | * |
77812a1e | 812 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 813 | */ |
77812a1e | 814 | static void try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 815 | __releases(dentry->d_lock) |
1da177e4 | 816 | { |
77812a1e | 817 | struct dentry *parent; |
d52b9086 | 818 | |
0d98439e | 819 | parent = dentry_kill(dentry); |
d52b9086 | 820 | /* |
77812a1e NP |
821 | * If dentry_kill returns NULL, we have nothing more to do. |
822 | * if it returns the same dentry, trylocks failed. In either | |
823 | * case, just loop again. | |
824 | * | |
825 | * Otherwise, we need to prune ancestors too. This is necessary | |
826 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
827 | * is also expected to be beneficial in reducing dentry cache | |
828 | * fragmentation. | |
d52b9086 | 829 | */ |
77812a1e NP |
830 | if (!parent) |
831 | return; | |
832 | if (parent == dentry) | |
833 | return; | |
834 | ||
835 | /* Prune ancestors. */ | |
836 | dentry = parent; | |
d52b9086 | 837 | while (dentry) { |
98474236 | 838 | if (lockref_put_or_lock(&dentry->d_lockref)) |
89e60548 | 839 | return; |
0d98439e | 840 | dentry = dentry_kill(dentry); |
d52b9086 | 841 | } |
1da177e4 LT |
842 | } |
843 | ||
3049cfe2 | 844 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 845 | { |
da3bbdd4 | 846 | struct dentry *dentry; |
da3bbdd4 | 847 | |
ec33679d NP |
848 | rcu_read_lock(); |
849 | for (;;) { | |
ec33679d NP |
850 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
851 | if (&dentry->d_lru == list) | |
852 | break; /* empty */ | |
853 | spin_lock(&dentry->d_lock); | |
854 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
855 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
856 | continue; |
857 | } | |
858 | ||
1da177e4 LT |
859 | /* |
860 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
861 | * the LRU because of laziness during lookup. Do not free |
862 | * it - just keep it off the LRU list. | |
1da177e4 | 863 | */ |
98474236 | 864 | if (dentry->d_lockref.count) { |
ec33679d | 865 | dentry_lru_del(dentry); |
da3bbdd4 | 866 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
867 | continue; |
868 | } | |
ec33679d | 869 | |
ec33679d | 870 | rcu_read_unlock(); |
77812a1e NP |
871 | |
872 | try_prune_one_dentry(dentry); | |
873 | ||
ec33679d | 874 | rcu_read_lock(); |
da3bbdd4 | 875 | } |
ec33679d | 876 | rcu_read_unlock(); |
3049cfe2 CH |
877 | } |
878 | ||
879 | /** | |
b48f03b3 DC |
880 | * prune_dcache_sb - shrink the dcache |
881 | * @sb: superblock | |
882 | * @count: number of entries to try to free | |
883 | * | |
884 | * Attempt to shrink the superblock dcache LRU by @count entries. This is | |
885 | * done when we need more memory an called from the superblock shrinker | |
886 | * function. | |
3049cfe2 | 887 | * |
b48f03b3 DC |
888 | * This function may fail to free any resources if all the dentries are in |
889 | * use. | |
3049cfe2 | 890 | */ |
b48f03b3 | 891 | void prune_dcache_sb(struct super_block *sb, int count) |
3049cfe2 | 892 | { |
3049cfe2 CH |
893 | struct dentry *dentry; |
894 | LIST_HEAD(referenced); | |
895 | LIST_HEAD(tmp); | |
3049cfe2 | 896 | |
23044507 NP |
897 | relock: |
898 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
899 | while (!list_empty(&sb->s_dentry_lru)) { |
900 | dentry = list_entry(sb->s_dentry_lru.prev, | |
901 | struct dentry, d_lru); | |
902 | BUG_ON(dentry->d_sb != sb); | |
903 | ||
23044507 NP |
904 | if (!spin_trylock(&dentry->d_lock)) { |
905 | spin_unlock(&dcache_lru_lock); | |
906 | cpu_relax(); | |
907 | goto relock; | |
908 | } | |
909 | ||
b48f03b3 | 910 | if (dentry->d_flags & DCACHE_REFERENCED) { |
23044507 NP |
911 | dentry->d_flags &= ~DCACHE_REFERENCED; |
912 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 913 | spin_unlock(&dentry->d_lock); |
23044507 NP |
914 | } else { |
915 | list_move_tail(&dentry->d_lru, &tmp); | |
eaf5f907 | 916 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
23044507 | 917 | spin_unlock(&dentry->d_lock); |
b0d40c92 | 918 | if (!--count) |
23044507 | 919 | break; |
3049cfe2 | 920 | } |
ec33679d | 921 | cond_resched_lock(&dcache_lru_lock); |
3049cfe2 | 922 | } |
da3bbdd4 KM |
923 | if (!list_empty(&referenced)) |
924 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 925 | spin_unlock(&dcache_lru_lock); |
ec33679d NP |
926 | |
927 | shrink_dentry_list(&tmp); | |
da3bbdd4 KM |
928 | } |
929 | ||
1da177e4 LT |
930 | /** |
931 | * shrink_dcache_sb - shrink dcache for a superblock | |
932 | * @sb: superblock | |
933 | * | |
3049cfe2 CH |
934 | * Shrink the dcache for the specified super block. This is used to free |
935 | * the dcache before unmounting a file system. | |
1da177e4 | 936 | */ |
3049cfe2 | 937 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 938 | { |
3049cfe2 CH |
939 | LIST_HEAD(tmp); |
940 | ||
23044507 | 941 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
942 | while (!list_empty(&sb->s_dentry_lru)) { |
943 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
ec33679d | 944 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 945 | shrink_dentry_list(&tmp); |
ec33679d | 946 | spin_lock(&dcache_lru_lock); |
3049cfe2 | 947 | } |
23044507 | 948 | spin_unlock(&dcache_lru_lock); |
1da177e4 | 949 | } |
ec4f8605 | 950 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 951 | |
c636ebdb DH |
952 | /* |
953 | * destroy a single subtree of dentries for unmount | |
954 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
955 | * locking | |
956 | */ | |
957 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
958 | { | |
959 | struct dentry *parent; | |
960 | ||
961 | BUG_ON(!IS_ROOT(dentry)); | |
962 | ||
c636ebdb DH |
963 | for (;;) { |
964 | /* descend to the first leaf in the current subtree */ | |
43c1c9cd | 965 | while (!list_empty(&dentry->d_subdirs)) |
c636ebdb DH |
966 | dentry = list_entry(dentry->d_subdirs.next, |
967 | struct dentry, d_u.d_child); | |
c636ebdb DH |
968 | |
969 | /* consume the dentries from this leaf up through its parents | |
970 | * until we find one with children or run out altogether */ | |
971 | do { | |
972 | struct inode *inode; | |
973 | ||
f0023bc6 | 974 | /* |
61572bb1 | 975 | * inform the fs that this dentry is about to be |
f0023bc6 SW |
976 | * unhashed and destroyed. |
977 | */ | |
590fb51f YZ |
978 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && |
979 | !d_unhashed(dentry)) | |
61572bb1 YZ |
980 | dentry->d_op->d_prune(dentry); |
981 | ||
982 | dentry_lru_del(dentry); | |
43c1c9cd DH |
983 | __d_shrink(dentry); |
984 | ||
98474236 | 985 | if (dentry->d_lockref.count != 0) { |
c636ebdb DH |
986 | printk(KERN_ERR |
987 | "BUG: Dentry %p{i=%lx,n=%s}" | |
988 | " still in use (%d)" | |
989 | " [unmount of %s %s]\n", | |
990 | dentry, | |
991 | dentry->d_inode ? | |
992 | dentry->d_inode->i_ino : 0UL, | |
993 | dentry->d_name.name, | |
98474236 | 994 | dentry->d_lockref.count, |
c636ebdb DH |
995 | dentry->d_sb->s_type->name, |
996 | dentry->d_sb->s_id); | |
997 | BUG(); | |
998 | } | |
999 | ||
2fd6b7f5 | 1000 | if (IS_ROOT(dentry)) { |
c636ebdb | 1001 | parent = NULL; |
2fd6b7f5 NP |
1002 | list_del(&dentry->d_u.d_child); |
1003 | } else { | |
871c0067 | 1004 | parent = dentry->d_parent; |
98474236 | 1005 | parent->d_lockref.count--; |
2fd6b7f5 | 1006 | list_del(&dentry->d_u.d_child); |
871c0067 | 1007 | } |
c636ebdb | 1008 | |
c636ebdb DH |
1009 | inode = dentry->d_inode; |
1010 | if (inode) { | |
1011 | dentry->d_inode = NULL; | |
b3d9b7a3 | 1012 | hlist_del_init(&dentry->d_alias); |
c636ebdb DH |
1013 | if (dentry->d_op && dentry->d_op->d_iput) |
1014 | dentry->d_op->d_iput(dentry, inode); | |
1015 | else | |
1016 | iput(inode); | |
1017 | } | |
1018 | ||
1019 | d_free(dentry); | |
1020 | ||
1021 | /* finished when we fall off the top of the tree, | |
1022 | * otherwise we ascend to the parent and move to the | |
1023 | * next sibling if there is one */ | |
1024 | if (!parent) | |
312d3ca8 | 1025 | return; |
c636ebdb | 1026 | dentry = parent; |
c636ebdb DH |
1027 | } while (list_empty(&dentry->d_subdirs)); |
1028 | ||
1029 | dentry = list_entry(dentry->d_subdirs.next, | |
1030 | struct dentry, d_u.d_child); | |
1031 | } | |
1032 | } | |
1033 | ||
1034 | /* | |
1035 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 1036 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
1037 | * - the superblock is detached from all mountings and open files, so the |
1038 | * dentry trees will not be rearranged by the VFS | |
1039 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
1040 | * any dentries belonging to this superblock that it comes across | |
1041 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
1042 | * in this superblock | |
1043 | */ | |
1044 | void shrink_dcache_for_umount(struct super_block *sb) | |
1045 | { | |
1046 | struct dentry *dentry; | |
1047 | ||
1048 | if (down_read_trylock(&sb->s_umount)) | |
1049 | BUG(); | |
1050 | ||
1051 | dentry = sb->s_root; | |
1052 | sb->s_root = NULL; | |
98474236 | 1053 | dentry->d_lockref.count--; |
c636ebdb DH |
1054 | shrink_dcache_for_umount_subtree(dentry); |
1055 | ||
ceb5bdc2 NP |
1056 | while (!hlist_bl_empty(&sb->s_anon)) { |
1057 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
1058 | shrink_dcache_for_umount_subtree(dentry); |
1059 | } | |
1060 | } | |
1061 | ||
c826cb7d LT |
1062 | /* |
1063 | * This tries to ascend one level of parenthood, but | |
1064 | * we can race with renaming, so we need to re-check | |
1065 | * the parenthood after dropping the lock and check | |
1066 | * that the sequence number still matches. | |
1067 | */ | |
48f5ec21 | 1068 | static struct dentry *try_to_ascend(struct dentry *old, unsigned seq) |
c826cb7d LT |
1069 | { |
1070 | struct dentry *new = old->d_parent; | |
1071 | ||
1072 | rcu_read_lock(); | |
1073 | spin_unlock(&old->d_lock); | |
1074 | spin_lock(&new->d_lock); | |
1075 | ||
1076 | /* | |
1077 | * might go back up the wrong parent if we have had a rename | |
1078 | * or deletion | |
1079 | */ | |
1080 | if (new != old->d_parent || | |
b161dfa6 | 1081 | (old->d_flags & DCACHE_DENTRY_KILLED) || |
48f5ec21 | 1082 | need_seqretry(&rename_lock, seq)) { |
c826cb7d LT |
1083 | spin_unlock(&new->d_lock); |
1084 | new = NULL; | |
1085 | } | |
1086 | rcu_read_unlock(); | |
1087 | return new; | |
1088 | } | |
1089 | ||
db14fc3a MS |
1090 | /** |
1091 | * enum d_walk_ret - action to talke during tree walk | |
1092 | * @D_WALK_CONTINUE: contrinue walk | |
1093 | * @D_WALK_QUIT: quit walk | |
1094 | * @D_WALK_NORETRY: quit when retry is needed | |
1095 | * @D_WALK_SKIP: skip this dentry and its children | |
1096 | */ | |
1097 | enum d_walk_ret { | |
1098 | D_WALK_CONTINUE, | |
1099 | D_WALK_QUIT, | |
1100 | D_WALK_NORETRY, | |
1101 | D_WALK_SKIP, | |
1102 | }; | |
c826cb7d | 1103 | |
1da177e4 | 1104 | /** |
db14fc3a MS |
1105 | * d_walk - walk the dentry tree |
1106 | * @parent: start of walk | |
1107 | * @data: data passed to @enter() and @finish() | |
1108 | * @enter: callback when first entering the dentry | |
1109 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1110 | * |
db14fc3a | 1111 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1112 | */ |
db14fc3a MS |
1113 | static void d_walk(struct dentry *parent, void *data, |
1114 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1115 | void (*finish)(void *)) | |
1da177e4 | 1116 | { |
949854d0 | 1117 | struct dentry *this_parent; |
1da177e4 | 1118 | struct list_head *next; |
48f5ec21 | 1119 | unsigned seq = 0; |
db14fc3a MS |
1120 | enum d_walk_ret ret; |
1121 | bool retry = true; | |
949854d0 | 1122 | |
58db63d0 | 1123 | again: |
48f5ec21 | 1124 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1125 | this_parent = parent; |
2fd6b7f5 | 1126 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1127 | |
1128 | ret = enter(data, this_parent); | |
1129 | switch (ret) { | |
1130 | case D_WALK_CONTINUE: | |
1131 | break; | |
1132 | case D_WALK_QUIT: | |
1133 | case D_WALK_SKIP: | |
1134 | goto out_unlock; | |
1135 | case D_WALK_NORETRY: | |
1136 | retry = false; | |
1137 | break; | |
1138 | } | |
1da177e4 LT |
1139 | repeat: |
1140 | next = this_parent->d_subdirs.next; | |
1141 | resume: | |
1142 | while (next != &this_parent->d_subdirs) { | |
1143 | struct list_head *tmp = next; | |
5160ee6f | 1144 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1145 | next = tmp->next; |
2fd6b7f5 NP |
1146 | |
1147 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
db14fc3a MS |
1148 | |
1149 | ret = enter(data, dentry); | |
1150 | switch (ret) { | |
1151 | case D_WALK_CONTINUE: | |
1152 | break; | |
1153 | case D_WALK_QUIT: | |
2fd6b7f5 | 1154 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1155 | goto out_unlock; |
1156 | case D_WALK_NORETRY: | |
1157 | retry = false; | |
1158 | break; | |
1159 | case D_WALK_SKIP: | |
1160 | spin_unlock(&dentry->d_lock); | |
1161 | continue; | |
2fd6b7f5 | 1162 | } |
db14fc3a | 1163 | |
1da177e4 | 1164 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1165 | spin_unlock(&this_parent->d_lock); |
1166 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1167 | this_parent = dentry; |
2fd6b7f5 | 1168 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1169 | goto repeat; |
1170 | } | |
2fd6b7f5 | 1171 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1172 | } |
1173 | /* | |
1174 | * All done at this level ... ascend and resume the search. | |
1175 | */ | |
1176 | if (this_parent != parent) { | |
c826cb7d | 1177 | struct dentry *child = this_parent; |
48f5ec21 | 1178 | this_parent = try_to_ascend(this_parent, seq); |
c826cb7d | 1179 | if (!this_parent) |
949854d0 | 1180 | goto rename_retry; |
949854d0 | 1181 | next = child->d_u.d_child.next; |
1da177e4 LT |
1182 | goto resume; |
1183 | } | |
48f5ec21 | 1184 | if (need_seqretry(&rename_lock, seq)) { |
db14fc3a | 1185 | spin_unlock(&this_parent->d_lock); |
949854d0 | 1186 | goto rename_retry; |
db14fc3a MS |
1187 | } |
1188 | if (finish) | |
1189 | finish(data); | |
1190 | ||
1191 | out_unlock: | |
1192 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1193 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1194 | return; |
58db63d0 NP |
1195 | |
1196 | rename_retry: | |
db14fc3a MS |
1197 | if (!retry) |
1198 | return; | |
48f5ec21 | 1199 | seq = 1; |
58db63d0 | 1200 | goto again; |
1da177e4 | 1201 | } |
db14fc3a MS |
1202 | |
1203 | /* | |
1204 | * Search for at least 1 mount point in the dentry's subdirs. | |
1205 | * We descend to the next level whenever the d_subdirs | |
1206 | * list is non-empty and continue searching. | |
1207 | */ | |
1208 | ||
1209 | /** | |
1210 | * have_submounts - check for mounts over a dentry | |
1211 | * @parent: dentry to check. | |
1212 | * | |
1213 | * Return true if the parent or its subdirectories contain | |
1214 | * a mount point | |
1215 | */ | |
1216 | ||
1217 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) | |
1218 | { | |
1219 | int *ret = data; | |
1220 | if (d_mountpoint(dentry)) { | |
1221 | *ret = 1; | |
1222 | return D_WALK_QUIT; | |
1223 | } | |
1224 | return D_WALK_CONTINUE; | |
1225 | } | |
1226 | ||
1227 | int have_submounts(struct dentry *parent) | |
1228 | { | |
1229 | int ret = 0; | |
1230 | ||
1231 | d_walk(parent, &ret, check_mount, NULL); | |
1232 | ||
1233 | return ret; | |
1234 | } | |
ec4f8605 | 1235 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1236 | |
eed81007 MS |
1237 | /* |
1238 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1239 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1240 | * subtree can become unreachable). | |
1241 | * | |
1242 | * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For | |
1243 | * this reason take rename_lock and d_lock on dentry and ancestors. | |
1244 | */ | |
1245 | int d_set_mounted(struct dentry *dentry) | |
1246 | { | |
1247 | struct dentry *p; | |
1248 | int ret = -ENOENT; | |
1249 | write_seqlock(&rename_lock); | |
1250 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1251 | /* Need exclusion wrt. check_submounts_and_drop() */ | |
1252 | spin_lock(&p->d_lock); | |
1253 | if (unlikely(d_unhashed(p))) { | |
1254 | spin_unlock(&p->d_lock); | |
1255 | goto out; | |
1256 | } | |
1257 | spin_unlock(&p->d_lock); | |
1258 | } | |
1259 | spin_lock(&dentry->d_lock); | |
1260 | if (!d_unlinked(dentry)) { | |
1261 | dentry->d_flags |= DCACHE_MOUNTED; | |
1262 | ret = 0; | |
1263 | } | |
1264 | spin_unlock(&dentry->d_lock); | |
1265 | out: | |
1266 | write_sequnlock(&rename_lock); | |
1267 | return ret; | |
1268 | } | |
1269 | ||
1da177e4 | 1270 | /* |
fd517909 | 1271 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1272 | * and move any unused dentries to the end of the unused |
1273 | * list for prune_dcache(). We descend to the next level | |
1274 | * whenever the d_subdirs list is non-empty and continue | |
1275 | * searching. | |
1276 | * | |
1277 | * It returns zero iff there are no unused children, | |
1278 | * otherwise it returns the number of children moved to | |
1279 | * the end of the unused list. This may not be the total | |
1280 | * number of unused children, because select_parent can | |
1281 | * drop the lock and return early due to latency | |
1282 | * constraints. | |
1283 | */ | |
1da177e4 | 1284 | |
db14fc3a MS |
1285 | struct select_data { |
1286 | struct dentry *start; | |
1287 | struct list_head dispose; | |
1288 | int found; | |
1289 | }; | |
23044507 | 1290 | |
db14fc3a MS |
1291 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1292 | { | |
1293 | struct select_data *data = _data; | |
1294 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1295 | |
db14fc3a MS |
1296 | if (data->start == dentry) |
1297 | goto out; | |
2fd6b7f5 | 1298 | |
1da177e4 | 1299 | /* |
db14fc3a MS |
1300 | * move only zero ref count dentries to the dispose list. |
1301 | * | |
1302 | * Those which are presently on the shrink list, being processed | |
1303 | * by shrink_dentry_list(), shouldn't be moved. Otherwise the | |
1304 | * loop in shrink_dcache_parent() might not make any progress | |
1305 | * and loop forever. | |
1da177e4 | 1306 | */ |
db14fc3a MS |
1307 | if (dentry->d_lockref.count) { |
1308 | dentry_lru_del(dentry); | |
1309 | } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { | |
1310 | dentry_lru_move_list(dentry, &data->dispose); | |
1311 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
1312 | data->found++; | |
1313 | ret = D_WALK_NORETRY; | |
1da177e4 | 1314 | } |
db14fc3a MS |
1315 | /* |
1316 | * We can return to the caller if we have found some (this | |
1317 | * ensures forward progress). We'll be coming back to find | |
1318 | * the rest. | |
1319 | */ | |
1320 | if (data->found && need_resched()) | |
1321 | ret = D_WALK_QUIT; | |
1da177e4 | 1322 | out: |
db14fc3a | 1323 | return ret; |
1da177e4 LT |
1324 | } |
1325 | ||
1326 | /** | |
1327 | * shrink_dcache_parent - prune dcache | |
1328 | * @parent: parent of entries to prune | |
1329 | * | |
1330 | * Prune the dcache to remove unused children of the parent dentry. | |
1331 | */ | |
db14fc3a | 1332 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1333 | { |
db14fc3a MS |
1334 | for (;;) { |
1335 | struct select_data data; | |
1da177e4 | 1336 | |
db14fc3a MS |
1337 | INIT_LIST_HEAD(&data.dispose); |
1338 | data.start = parent; | |
1339 | data.found = 0; | |
1340 | ||
1341 | d_walk(parent, &data, select_collect, NULL); | |
1342 | if (!data.found) | |
1343 | break; | |
1344 | ||
1345 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1346 | cond_resched(); |
1347 | } | |
1da177e4 | 1348 | } |
ec4f8605 | 1349 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1350 | |
848ac114 MS |
1351 | static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry) |
1352 | { | |
1353 | struct select_data *data = _data; | |
1354 | ||
1355 | if (d_mountpoint(dentry)) { | |
1356 | data->found = -EBUSY; | |
1357 | return D_WALK_QUIT; | |
1358 | } | |
1359 | ||
1360 | return select_collect(_data, dentry); | |
1361 | } | |
1362 | ||
1363 | static void check_and_drop(void *_data) | |
1364 | { | |
1365 | struct select_data *data = _data; | |
1366 | ||
1367 | if (d_mountpoint(data->start)) | |
1368 | data->found = -EBUSY; | |
1369 | if (!data->found) | |
1370 | __d_drop(data->start); | |
1371 | } | |
1372 | ||
1373 | /** | |
1374 | * check_submounts_and_drop - prune dcache, check for submounts and drop | |
1375 | * | |
1376 | * All done as a single atomic operation relative to has_unlinked_ancestor(). | |
1377 | * Returns 0 if successfully unhashed @parent. If there were submounts then | |
1378 | * return -EBUSY. | |
1379 | * | |
1380 | * @dentry: dentry to prune and drop | |
1381 | */ | |
1382 | int check_submounts_and_drop(struct dentry *dentry) | |
1383 | { | |
1384 | int ret = 0; | |
1385 | ||
1386 | /* Negative dentries can be dropped without further checks */ | |
1387 | if (!dentry->d_inode) { | |
1388 | d_drop(dentry); | |
1389 | goto out; | |
1390 | } | |
1391 | ||
1392 | for (;;) { | |
1393 | struct select_data data; | |
1394 | ||
1395 | INIT_LIST_HEAD(&data.dispose); | |
1396 | data.start = dentry; | |
1397 | data.found = 0; | |
1398 | ||
1399 | d_walk(dentry, &data, check_and_collect, check_and_drop); | |
1400 | ret = data.found; | |
1401 | ||
1402 | if (!list_empty(&data.dispose)) | |
1403 | shrink_dentry_list(&data.dispose); | |
1404 | ||
1405 | if (ret <= 0) | |
1406 | break; | |
1407 | ||
1408 | cond_resched(); | |
1409 | } | |
1410 | ||
1411 | out: | |
1412 | return ret; | |
1413 | } | |
1414 | EXPORT_SYMBOL(check_submounts_and_drop); | |
1415 | ||
1da177e4 | 1416 | /** |
a4464dbc AV |
1417 | * __d_alloc - allocate a dcache entry |
1418 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1419 | * @name: qstr of the name |
1420 | * | |
1421 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1422 | * available. On a success the dentry is returned. The name passed in is | |
1423 | * copied and the copy passed in may be reused after this call. | |
1424 | */ | |
1425 | ||
a4464dbc | 1426 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1427 | { |
1428 | struct dentry *dentry; | |
1429 | char *dname; | |
1430 | ||
e12ba74d | 1431 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1432 | if (!dentry) |
1433 | return NULL; | |
1434 | ||
6326c71f LT |
1435 | /* |
1436 | * We guarantee that the inline name is always NUL-terminated. | |
1437 | * This way the memcpy() done by the name switching in rename | |
1438 | * will still always have a NUL at the end, even if we might | |
1439 | * be overwriting an internal NUL character | |
1440 | */ | |
1441 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 LT |
1442 | if (name->len > DNAME_INLINE_LEN-1) { |
1443 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1444 | if (!dname) { | |
1445 | kmem_cache_free(dentry_cache, dentry); | |
1446 | return NULL; | |
1447 | } | |
1448 | } else { | |
1449 | dname = dentry->d_iname; | |
1450 | } | |
1da177e4 LT |
1451 | |
1452 | dentry->d_name.len = name->len; | |
1453 | dentry->d_name.hash = name->hash; | |
1454 | memcpy(dname, name->name, name->len); | |
1455 | dname[name->len] = 0; | |
1456 | ||
6326c71f LT |
1457 | /* Make sure we always see the terminating NUL character */ |
1458 | smp_wmb(); | |
1459 | dentry->d_name.name = dname; | |
1460 | ||
98474236 | 1461 | dentry->d_lockref.count = 1; |
dea3667b | 1462 | dentry->d_flags = 0; |
1da177e4 | 1463 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1464 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1465 | dentry->d_inode = NULL; |
a4464dbc AV |
1466 | dentry->d_parent = dentry; |
1467 | dentry->d_sb = sb; | |
1da177e4 LT |
1468 | dentry->d_op = NULL; |
1469 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1470 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1471 | INIT_LIST_HEAD(&dentry->d_lru); |
1472 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
b3d9b7a3 | 1473 | INIT_HLIST_NODE(&dentry->d_alias); |
2fd6b7f5 | 1474 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1475 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1476 | |
3e880fb5 | 1477 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1478 | |
1da177e4 LT |
1479 | return dentry; |
1480 | } | |
a4464dbc AV |
1481 | |
1482 | /** | |
1483 | * d_alloc - allocate a dcache entry | |
1484 | * @parent: parent of entry to allocate | |
1485 | * @name: qstr of the name | |
1486 | * | |
1487 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1488 | * available. On a success the dentry is returned. The name passed in is | |
1489 | * copied and the copy passed in may be reused after this call. | |
1490 | */ | |
1491 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1492 | { | |
1493 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1494 | if (!dentry) | |
1495 | return NULL; | |
1496 | ||
1497 | spin_lock(&parent->d_lock); | |
1498 | /* | |
1499 | * don't need child lock because it is not subject | |
1500 | * to concurrency here | |
1501 | */ | |
1502 | __dget_dlock(parent); | |
1503 | dentry->d_parent = parent; | |
1504 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1505 | spin_unlock(&parent->d_lock); | |
1506 | ||
1507 | return dentry; | |
1508 | } | |
ec4f8605 | 1509 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1510 | |
4b936885 NP |
1511 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1512 | { | |
a4464dbc AV |
1513 | struct dentry *dentry = __d_alloc(sb, name); |
1514 | if (dentry) | |
4b936885 | 1515 | dentry->d_flags |= DCACHE_DISCONNECTED; |
4b936885 NP |
1516 | return dentry; |
1517 | } | |
1518 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1519 | ||
1da177e4 LT |
1520 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1521 | { | |
1522 | struct qstr q; | |
1523 | ||
1524 | q.name = name; | |
1525 | q.len = strlen(name); | |
1526 | q.hash = full_name_hash(q.name, q.len); | |
1527 | return d_alloc(parent, &q); | |
1528 | } | |
ef26ca97 | 1529 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1530 | |
fb045adb NP |
1531 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1532 | { | |
6f7f7caa LT |
1533 | WARN_ON_ONCE(dentry->d_op); |
1534 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1535 | DCACHE_OP_COMPARE | |
1536 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1537 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1538 | DCACHE_OP_DELETE )); |
1539 | dentry->d_op = op; | |
1540 | if (!op) | |
1541 | return; | |
1542 | if (op->d_hash) | |
1543 | dentry->d_flags |= DCACHE_OP_HASH; | |
1544 | if (op->d_compare) | |
1545 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1546 | if (op->d_revalidate) | |
1547 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1548 | if (op->d_weak_revalidate) |
1549 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1550 | if (op->d_delete) |
1551 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1552 | if (op->d_prune) |
1553 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1554 | |
1555 | } | |
1556 | EXPORT_SYMBOL(d_set_d_op); | |
1557 | ||
360da900 OH |
1558 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1559 | { | |
b23fb0a6 | 1560 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1561 | if (inode) { |
1562 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1563 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
b3d9b7a3 | 1564 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1565 | } |
360da900 | 1566 | dentry->d_inode = inode; |
31e6b01f | 1567 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1568 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1569 | fsnotify_d_instantiate(dentry, inode); |
1570 | } | |
1571 | ||
1da177e4 LT |
1572 | /** |
1573 | * d_instantiate - fill in inode information for a dentry | |
1574 | * @entry: dentry to complete | |
1575 | * @inode: inode to attach to this dentry | |
1576 | * | |
1577 | * Fill in inode information in the entry. | |
1578 | * | |
1579 | * This turns negative dentries into productive full members | |
1580 | * of society. | |
1581 | * | |
1582 | * NOTE! This assumes that the inode count has been incremented | |
1583 | * (or otherwise set) by the caller to indicate that it is now | |
1584 | * in use by the dcache. | |
1585 | */ | |
1586 | ||
1587 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1588 | { | |
b3d9b7a3 | 1589 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
873feea0 NP |
1590 | if (inode) |
1591 | spin_lock(&inode->i_lock); | |
360da900 | 1592 | __d_instantiate(entry, inode); |
873feea0 NP |
1593 | if (inode) |
1594 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1595 | security_d_instantiate(entry, inode); |
1596 | } | |
ec4f8605 | 1597 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1598 | |
1599 | /** | |
1600 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1601 | * @entry: dentry to instantiate | |
1602 | * @inode: inode to attach to this dentry | |
1603 | * | |
1604 | * Fill in inode information in the entry. On success, it returns NULL. | |
1605 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1606 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1607 | * |
1608 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1609 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1610 | * |
1611 | * This also assumes that the inode count has been incremented | |
1612 | * (or otherwise set) by the caller to indicate that it is now | |
1613 | * in use by the dcache. | |
1da177e4 | 1614 | */ |
770bfad8 DH |
1615 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1616 | struct inode *inode) | |
1da177e4 LT |
1617 | { |
1618 | struct dentry *alias; | |
1619 | int len = entry->d_name.len; | |
1620 | const char *name = entry->d_name.name; | |
1621 | unsigned int hash = entry->d_name.hash; | |
1622 | ||
770bfad8 | 1623 | if (!inode) { |
360da900 | 1624 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1625 | return NULL; |
1626 | } | |
1627 | ||
b67bfe0d | 1628 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
9abca360 NP |
1629 | /* |
1630 | * Don't need alias->d_lock here, because aliases with | |
1631 | * d_parent == entry->d_parent are not subject to name or | |
1632 | * parent changes, because the parent inode i_mutex is held. | |
1633 | */ | |
12f8ad4b | 1634 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1635 | continue; |
1636 | if (alias->d_parent != entry->d_parent) | |
1637 | continue; | |
ee983e89 LT |
1638 | if (alias->d_name.len != len) |
1639 | continue; | |
12f8ad4b | 1640 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1641 | continue; |
dc0474be | 1642 | __dget(alias); |
1da177e4 LT |
1643 | return alias; |
1644 | } | |
770bfad8 | 1645 | |
360da900 | 1646 | __d_instantiate(entry, inode); |
1da177e4 LT |
1647 | return NULL; |
1648 | } | |
770bfad8 DH |
1649 | |
1650 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1651 | { | |
1652 | struct dentry *result; | |
1653 | ||
b3d9b7a3 | 1654 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
770bfad8 | 1655 | |
873feea0 NP |
1656 | if (inode) |
1657 | spin_lock(&inode->i_lock); | |
770bfad8 | 1658 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1659 | if (inode) |
1660 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1661 | |
1662 | if (!result) { | |
1663 | security_d_instantiate(entry, inode); | |
1664 | return NULL; | |
1665 | } | |
1666 | ||
1667 | BUG_ON(!d_unhashed(result)); | |
1668 | iput(inode); | |
1669 | return result; | |
1670 | } | |
1671 | ||
1da177e4 LT |
1672 | EXPORT_SYMBOL(d_instantiate_unique); |
1673 | ||
adc0e91a AV |
1674 | struct dentry *d_make_root(struct inode *root_inode) |
1675 | { | |
1676 | struct dentry *res = NULL; | |
1677 | ||
1678 | if (root_inode) { | |
26fe5750 | 1679 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1680 | |
1681 | res = __d_alloc(root_inode->i_sb, &name); | |
1682 | if (res) | |
1683 | d_instantiate(res, root_inode); | |
1684 | else | |
1685 | iput(root_inode); | |
1686 | } | |
1687 | return res; | |
1688 | } | |
1689 | EXPORT_SYMBOL(d_make_root); | |
1690 | ||
d891eedb BF |
1691 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1692 | { | |
1693 | struct dentry *alias; | |
1694 | ||
b3d9b7a3 | 1695 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1696 | return NULL; |
b3d9b7a3 | 1697 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); |
d891eedb BF |
1698 | __dget(alias); |
1699 | return alias; | |
1700 | } | |
1701 | ||
46f72b34 SW |
1702 | /** |
1703 | * d_find_any_alias - find any alias for a given inode | |
1704 | * @inode: inode to find an alias for | |
1705 | * | |
1706 | * If any aliases exist for the given inode, take and return a | |
1707 | * reference for one of them. If no aliases exist, return %NULL. | |
1708 | */ | |
1709 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1710 | { |
1711 | struct dentry *de; | |
1712 | ||
1713 | spin_lock(&inode->i_lock); | |
1714 | de = __d_find_any_alias(inode); | |
1715 | spin_unlock(&inode->i_lock); | |
1716 | return de; | |
1717 | } | |
46f72b34 | 1718 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1719 | |
4ea3ada2 CH |
1720 | /** |
1721 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1722 | * @inode: inode to allocate the dentry for | |
1723 | * | |
1724 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1725 | * similar open by handle operations. The returned dentry may be anonymous, | |
1726 | * or may have a full name (if the inode was already in the cache). | |
1727 | * | |
1728 | * When called on a directory inode, we must ensure that the inode only ever | |
1729 | * has one dentry. If a dentry is found, that is returned instead of | |
1730 | * allocating a new one. | |
1731 | * | |
1732 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1733 | * to the dentry. In case of an error the reference on the inode is released. |
1734 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1735 | * be passed in and will be the error will be propagate to the return value, | |
1736 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1737 | */ |
1738 | struct dentry *d_obtain_alias(struct inode *inode) | |
1739 | { | |
b911a6bd | 1740 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1741 | struct dentry *tmp; |
1742 | struct dentry *res; | |
4ea3ada2 CH |
1743 | |
1744 | if (!inode) | |
44003728 | 1745 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1746 | if (IS_ERR(inode)) |
1747 | return ERR_CAST(inode); | |
1748 | ||
d891eedb | 1749 | res = d_find_any_alias(inode); |
9308a612 CH |
1750 | if (res) |
1751 | goto out_iput; | |
1752 | ||
a4464dbc | 1753 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1754 | if (!tmp) { |
1755 | res = ERR_PTR(-ENOMEM); | |
1756 | goto out_iput; | |
4ea3ada2 | 1757 | } |
b5c84bf6 | 1758 | |
873feea0 | 1759 | spin_lock(&inode->i_lock); |
d891eedb | 1760 | res = __d_find_any_alias(inode); |
9308a612 | 1761 | if (res) { |
873feea0 | 1762 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1763 | dput(tmp); |
1764 | goto out_iput; | |
1765 | } | |
1766 | ||
1767 | /* attach a disconnected dentry */ | |
1768 | spin_lock(&tmp->d_lock); | |
9308a612 CH |
1769 | tmp->d_inode = inode; |
1770 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
b3d9b7a3 | 1771 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1772 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1773 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1774 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1775 | spin_unlock(&tmp->d_lock); |
873feea0 | 1776 | spin_unlock(&inode->i_lock); |
24ff6663 | 1777 | security_d_instantiate(tmp, inode); |
9308a612 | 1778 | |
9308a612 CH |
1779 | return tmp; |
1780 | ||
1781 | out_iput: | |
24ff6663 JB |
1782 | if (res && !IS_ERR(res)) |
1783 | security_d_instantiate(res, inode); | |
9308a612 CH |
1784 | iput(inode); |
1785 | return res; | |
4ea3ada2 | 1786 | } |
adc48720 | 1787 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1788 | |
1789 | /** | |
1790 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1791 | * @inode: the inode which may have a disconnected dentry | |
1792 | * @dentry: a negative dentry which we want to point to the inode. | |
1793 | * | |
1794 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1795 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1796 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1797 | * | |
1798 | * This is needed in the lookup routine of any filesystem that is exportable | |
1799 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1800 | * | |
1801 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1802 | * is returned. This matches the expected return value of ->lookup. | |
1803 | * | |
6d4ade98 SW |
1804 | * Cluster filesystems may call this function with a negative, hashed dentry. |
1805 | * In that case, we know that the inode will be a regular file, and also this | |
1806 | * will only occur during atomic_open. So we need to check for the dentry | |
1807 | * being already hashed only in the final case. | |
1da177e4 LT |
1808 | */ |
1809 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1810 | { | |
1811 | struct dentry *new = NULL; | |
1812 | ||
a9049376 AV |
1813 | if (IS_ERR(inode)) |
1814 | return ERR_CAST(inode); | |
1815 | ||
21c0d8fd | 1816 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1817 | spin_lock(&inode->i_lock); |
32ba9c3f | 1818 | new = __d_find_alias(inode, 1); |
1da177e4 | 1819 | if (new) { |
32ba9c3f | 1820 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); |
873feea0 | 1821 | spin_unlock(&inode->i_lock); |
1da177e4 | 1822 | security_d_instantiate(new, inode); |
1da177e4 LT |
1823 | d_move(new, dentry); |
1824 | iput(inode); | |
1825 | } else { | |
873feea0 | 1826 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1827 | __d_instantiate(dentry, inode); |
873feea0 | 1828 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1829 | security_d_instantiate(dentry, inode); |
1830 | d_rehash(dentry); | |
1831 | } | |
6d4ade98 SW |
1832 | } else { |
1833 | d_instantiate(dentry, inode); | |
1834 | if (d_unhashed(dentry)) | |
1835 | d_rehash(dentry); | |
1836 | } | |
1da177e4 LT |
1837 | return new; |
1838 | } | |
ec4f8605 | 1839 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1840 | |
9403540c BN |
1841 | /** |
1842 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1843 | * @inode: the inode case-insensitive lookup has found | |
1844 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1845 | * @name: the case-exact name to be associated with the returned dentry | |
1846 | * | |
1847 | * This is to avoid filling the dcache with case-insensitive names to the | |
1848 | * same inode, only the actual correct case is stored in the dcache for | |
1849 | * case-insensitive filesystems. | |
1850 | * | |
1851 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1852 | * already exists in in the dcache, use it and return it. | |
1853 | * | |
1854 | * If no entry exists with the exact case name, allocate new dentry with | |
1855 | * the exact case, and return the spliced entry. | |
1856 | */ | |
e45b590b | 1857 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1858 | struct qstr *name) |
1859 | { | |
9403540c BN |
1860 | struct dentry *found; |
1861 | struct dentry *new; | |
1862 | ||
b6520c81 CH |
1863 | /* |
1864 | * First check if a dentry matching the name already exists, | |
1865 | * if not go ahead and create it now. | |
1866 | */ | |
9403540c | 1867 | found = d_hash_and_lookup(dentry->d_parent, name); |
4f522a24 AV |
1868 | if (unlikely(IS_ERR(found))) |
1869 | goto err_out; | |
9403540c BN |
1870 | if (!found) { |
1871 | new = d_alloc(dentry->d_parent, name); | |
1872 | if (!new) { | |
4f522a24 | 1873 | found = ERR_PTR(-ENOMEM); |
9403540c BN |
1874 | goto err_out; |
1875 | } | |
b6520c81 | 1876 | |
9403540c BN |
1877 | found = d_splice_alias(inode, new); |
1878 | if (found) { | |
1879 | dput(new); | |
1880 | return found; | |
1881 | } | |
1882 | return new; | |
1883 | } | |
b6520c81 CH |
1884 | |
1885 | /* | |
1886 | * If a matching dentry exists, and it's not negative use it. | |
1887 | * | |
1888 | * Decrement the reference count to balance the iget() done | |
1889 | * earlier on. | |
1890 | */ | |
9403540c BN |
1891 | if (found->d_inode) { |
1892 | if (unlikely(found->d_inode != inode)) { | |
1893 | /* This can't happen because bad inodes are unhashed. */ | |
1894 | BUG_ON(!is_bad_inode(inode)); | |
1895 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1896 | } | |
9403540c BN |
1897 | iput(inode); |
1898 | return found; | |
1899 | } | |
b6520c81 | 1900 | |
9403540c | 1901 | /* |
9403540c | 1902 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1903 | * already has a dentry. |
9403540c | 1904 | */ |
4513d899 AV |
1905 | new = d_splice_alias(inode, found); |
1906 | if (new) { | |
1907 | dput(found); | |
1908 | found = new; | |
9403540c | 1909 | } |
4513d899 | 1910 | return found; |
9403540c BN |
1911 | |
1912 | err_out: | |
1913 | iput(inode); | |
4f522a24 | 1914 | return found; |
9403540c | 1915 | } |
ec4f8605 | 1916 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1917 | |
12f8ad4b LT |
1918 | /* |
1919 | * Do the slow-case of the dentry name compare. | |
1920 | * | |
1921 | * Unlike the dentry_cmp() function, we need to atomically | |
da53be12 | 1922 | * load the name and length information, so that the |
12f8ad4b LT |
1923 | * filesystem can rely on them, and can use the 'name' and |
1924 | * 'len' information without worrying about walking off the | |
1925 | * end of memory etc. | |
1926 | * | |
1927 | * Thus the read_seqcount_retry() and the "duplicate" info | |
1928 | * in arguments (the low-level filesystem should not look | |
1929 | * at the dentry inode or name contents directly, since | |
1930 | * rename can change them while we're in RCU mode). | |
1931 | */ | |
1932 | enum slow_d_compare { | |
1933 | D_COMP_OK, | |
1934 | D_COMP_NOMATCH, | |
1935 | D_COMP_SEQRETRY, | |
1936 | }; | |
1937 | ||
1938 | static noinline enum slow_d_compare slow_dentry_cmp( | |
1939 | const struct dentry *parent, | |
12f8ad4b LT |
1940 | struct dentry *dentry, |
1941 | unsigned int seq, | |
1942 | const struct qstr *name) | |
1943 | { | |
1944 | int tlen = dentry->d_name.len; | |
1945 | const char *tname = dentry->d_name.name; | |
12f8ad4b LT |
1946 | |
1947 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
1948 | cpu_relax(); | |
1949 | return D_COMP_SEQRETRY; | |
1950 | } | |
da53be12 | 1951 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
12f8ad4b LT |
1952 | return D_COMP_NOMATCH; |
1953 | return D_COMP_OK; | |
1954 | } | |
1955 | ||
31e6b01f NP |
1956 | /** |
1957 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1958 | * @parent: parent dentry | |
1959 | * @name: qstr of name we wish to find | |
1f1e6e52 | 1960 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
1961 | * Returns: dentry, or NULL |
1962 | * | |
1963 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1964 | * resolution (store-free path walking) design described in | |
1965 | * Documentation/filesystems/path-lookup.txt. | |
1966 | * | |
1967 | * This is not to be used outside core vfs. | |
1968 | * | |
1969 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1970 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1971 | * without taking d_lock and checking d_seq sequence count against @seq | |
1972 | * returned here. | |
1973 | * | |
15570086 | 1974 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
1975 | * function. |
1976 | * | |
1977 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1978 | * the returned dentry, so long as its parent's seqlock is checked after the | |
1979 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
1980 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
1981 | * |
1982 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
1983 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 1984 | */ |
8966be90 LT |
1985 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
1986 | const struct qstr *name, | |
da53be12 | 1987 | unsigned *seqp) |
31e6b01f | 1988 | { |
26fe5750 | 1989 | u64 hashlen = name->hash_len; |
31e6b01f | 1990 | const unsigned char *str = name->name; |
26fe5750 | 1991 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 1992 | struct hlist_bl_node *node; |
31e6b01f NP |
1993 | struct dentry *dentry; |
1994 | ||
1995 | /* | |
1996 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1997 | * required to prevent single threaded performance regressions | |
1998 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1999 | * Keep the two functions in sync. | |
2000 | */ | |
2001 | ||
2002 | /* | |
2003 | * The hash list is protected using RCU. | |
2004 | * | |
2005 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2006 | * races with d_move(). | |
2007 | * | |
2008 | * It is possible that concurrent renames can mess up our list | |
2009 | * walk here and result in missing our dentry, resulting in the | |
2010 | * false-negative result. d_lookup() protects against concurrent | |
2011 | * renames using rename_lock seqlock. | |
2012 | * | |
b0a4bb83 | 2013 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2014 | */ |
b07ad996 | 2015 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2016 | unsigned seq; |
31e6b01f | 2017 | |
31e6b01f | 2018 | seqretry: |
12f8ad4b LT |
2019 | /* |
2020 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2021 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2022 | * |
2023 | * The caller must perform a seqcount check in order | |
da53be12 | 2024 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2025 | * |
2026 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2027 | * we don't wait for the sequence count to stabilize if it | |
2028 | * is in the middle of a sequence change. If we do the slow | |
2029 | * dentry compare, we will do seqretries until it is stable, | |
2030 | * and if we end up with a successful lookup, we actually | |
2031 | * want to exit RCU lookup anyway. | |
2032 | */ | |
2033 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2034 | if (dentry->d_parent != parent) |
2035 | continue; | |
2e321806 LT |
2036 | if (d_unhashed(dentry)) |
2037 | continue; | |
12f8ad4b | 2038 | |
830c0f0e | 2039 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
2040 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2041 | continue; | |
da53be12 LT |
2042 | *seqp = seq; |
2043 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
12f8ad4b LT |
2044 | case D_COMP_OK: |
2045 | return dentry; | |
2046 | case D_COMP_NOMATCH: | |
31e6b01f | 2047 | continue; |
12f8ad4b LT |
2048 | default: |
2049 | goto seqretry; | |
2050 | } | |
31e6b01f | 2051 | } |
12f8ad4b | 2052 | |
26fe5750 | 2053 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 2054 | continue; |
da53be12 | 2055 | *seqp = seq; |
26fe5750 | 2056 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 2057 | return dentry; |
31e6b01f NP |
2058 | } |
2059 | return NULL; | |
2060 | } | |
2061 | ||
1da177e4 LT |
2062 | /** |
2063 | * d_lookup - search for a dentry | |
2064 | * @parent: parent dentry | |
2065 | * @name: qstr of name we wish to find | |
b04f784e | 2066 | * Returns: dentry, or NULL |
1da177e4 | 2067 | * |
b04f784e NP |
2068 | * d_lookup searches the children of the parent dentry for the name in |
2069 | * question. If the dentry is found its reference count is incremented and the | |
2070 | * dentry is returned. The caller must use dput to free the entry when it has | |
2071 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2072 | */ |
da2d8455 | 2073 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2074 | { |
31e6b01f | 2075 | struct dentry *dentry; |
949854d0 | 2076 | unsigned seq; |
1da177e4 LT |
2077 | |
2078 | do { | |
2079 | seq = read_seqbegin(&rename_lock); | |
2080 | dentry = __d_lookup(parent, name); | |
2081 | if (dentry) | |
2082 | break; | |
2083 | } while (read_seqretry(&rename_lock, seq)); | |
2084 | return dentry; | |
2085 | } | |
ec4f8605 | 2086 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2087 | |
31e6b01f | 2088 | /** |
b04f784e NP |
2089 | * __d_lookup - search for a dentry (racy) |
2090 | * @parent: parent dentry | |
2091 | * @name: qstr of name we wish to find | |
2092 | * Returns: dentry, or NULL | |
2093 | * | |
2094 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2095 | * false-negative result due to unrelated rename activity. | |
2096 | * | |
2097 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2098 | * however it must be used carefully, eg. with a following d_lookup in | |
2099 | * the case of failure. | |
2100 | * | |
2101 | * __d_lookup callers must be commented. | |
2102 | */ | |
a713ca2a | 2103 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
2104 | { |
2105 | unsigned int len = name->len; | |
2106 | unsigned int hash = name->hash; | |
2107 | const unsigned char *str = name->name; | |
b07ad996 | 2108 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 2109 | struct hlist_bl_node *node; |
31e6b01f | 2110 | struct dentry *found = NULL; |
665a7583 | 2111 | struct dentry *dentry; |
1da177e4 | 2112 | |
31e6b01f NP |
2113 | /* |
2114 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2115 | * required to prevent single threaded performance regressions | |
2116 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2117 | * Keep the two functions in sync. | |
2118 | */ | |
2119 | ||
b04f784e NP |
2120 | /* |
2121 | * The hash list is protected using RCU. | |
2122 | * | |
2123 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2124 | * with d_move(). | |
2125 | * | |
2126 | * It is possible that concurrent renames can mess up our list | |
2127 | * walk here and result in missing our dentry, resulting in the | |
2128 | * false-negative result. d_lookup() protects against concurrent | |
2129 | * renames using rename_lock seqlock. | |
2130 | * | |
b0a4bb83 | 2131 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2132 | */ |
1da177e4 LT |
2133 | rcu_read_lock(); |
2134 | ||
b07ad996 | 2135 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2136 | |
1da177e4 LT |
2137 | if (dentry->d_name.hash != hash) |
2138 | continue; | |
1da177e4 LT |
2139 | |
2140 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2141 | if (dentry->d_parent != parent) |
2142 | goto next; | |
d0185c08 LT |
2143 | if (d_unhashed(dentry)) |
2144 | goto next; | |
2145 | ||
1da177e4 LT |
2146 | /* |
2147 | * It is safe to compare names since d_move() cannot | |
2148 | * change the qstr (protected by d_lock). | |
2149 | */ | |
fb045adb | 2150 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
2151 | int tlen = dentry->d_name.len; |
2152 | const char *tname = dentry->d_name.name; | |
da53be12 | 2153 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1da177e4 LT |
2154 | goto next; |
2155 | } else { | |
ee983e89 LT |
2156 | if (dentry->d_name.len != len) |
2157 | goto next; | |
12f8ad4b | 2158 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
2159 | goto next; |
2160 | } | |
2161 | ||
98474236 | 2162 | dentry->d_lockref.count++; |
d0185c08 | 2163 | found = dentry; |
1da177e4 LT |
2164 | spin_unlock(&dentry->d_lock); |
2165 | break; | |
2166 | next: | |
2167 | spin_unlock(&dentry->d_lock); | |
2168 | } | |
2169 | rcu_read_unlock(); | |
2170 | ||
2171 | return found; | |
2172 | } | |
2173 | ||
3e7e241f EB |
2174 | /** |
2175 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2176 | * @dir: Directory to search in | |
2177 | * @name: qstr of name we wish to find | |
2178 | * | |
4f522a24 | 2179 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2180 | */ |
2181 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2182 | { | |
3e7e241f EB |
2183 | /* |
2184 | * Check for a fs-specific hash function. Note that we must | |
2185 | * calculate the standard hash first, as the d_op->d_hash() | |
2186 | * routine may choose to leave the hash value unchanged. | |
2187 | */ | |
2188 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2189 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2190 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2191 | if (unlikely(err < 0)) |
2192 | return ERR_PTR(err); | |
3e7e241f | 2193 | } |
4f522a24 | 2194 | return d_lookup(dir, name); |
3e7e241f | 2195 | } |
4f522a24 | 2196 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2197 | |
1da177e4 | 2198 | /** |
786a5e15 | 2199 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2200 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2201 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2202 | * |
2203 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2204 | * This is used by ncpfs in its readdir implementation. | |
2205 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2206 | * |
2207 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2208 | */ |
d3a23e16 | 2209 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2210 | { |
786a5e15 | 2211 | struct dentry *child; |
d3a23e16 | 2212 | |
2fd6b7f5 | 2213 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2214 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2215 | if (dentry == child) { | |
2fd6b7f5 | 2216 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2217 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2218 | spin_unlock(&dentry->d_lock); |
2219 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2220 | return 1; |
2221 | } | |
2222 | } | |
2fd6b7f5 | 2223 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2224 | |
1da177e4 LT |
2225 | return 0; |
2226 | } | |
ec4f8605 | 2227 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2228 | |
2229 | /* | |
2230 | * When a file is deleted, we have two options: | |
2231 | * - turn this dentry into a negative dentry | |
2232 | * - unhash this dentry and free it. | |
2233 | * | |
2234 | * Usually, we want to just turn this into | |
2235 | * a negative dentry, but if anybody else is | |
2236 | * currently using the dentry or the inode | |
2237 | * we can't do that and we fall back on removing | |
2238 | * it from the hash queues and waiting for | |
2239 | * it to be deleted later when it has no users | |
2240 | */ | |
2241 | ||
2242 | /** | |
2243 | * d_delete - delete a dentry | |
2244 | * @dentry: The dentry to delete | |
2245 | * | |
2246 | * Turn the dentry into a negative dentry if possible, otherwise | |
2247 | * remove it from the hash queues so it can be deleted later | |
2248 | */ | |
2249 | ||
2250 | void d_delete(struct dentry * dentry) | |
2251 | { | |
873feea0 | 2252 | struct inode *inode; |
7a91bf7f | 2253 | int isdir = 0; |
1da177e4 LT |
2254 | /* |
2255 | * Are we the only user? | |
2256 | */ | |
357f8e65 | 2257 | again: |
1da177e4 | 2258 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2259 | inode = dentry->d_inode; |
2260 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2261 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2262 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2263 | spin_unlock(&dentry->d_lock); |
2264 | cpu_relax(); | |
2265 | goto again; | |
2266 | } | |
13e3c5e5 | 2267 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2268 | dentry_unlink_inode(dentry); |
7a91bf7f | 2269 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2270 | return; |
2271 | } | |
2272 | ||
2273 | if (!d_unhashed(dentry)) | |
2274 | __d_drop(dentry); | |
2275 | ||
2276 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2277 | |
2278 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2279 | } |
ec4f8605 | 2280 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2281 | |
b07ad996 | 2282 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2283 | { |
ceb5bdc2 | 2284 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2285 | hlist_bl_lock(b); |
dea3667b | 2286 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2287 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2288 | hlist_bl_unlock(b); |
1da177e4 LT |
2289 | } |
2290 | ||
770bfad8 DH |
2291 | static void _d_rehash(struct dentry * entry) |
2292 | { | |
2293 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2294 | } | |
2295 | ||
1da177e4 LT |
2296 | /** |
2297 | * d_rehash - add an entry back to the hash | |
2298 | * @entry: dentry to add to the hash | |
2299 | * | |
2300 | * Adds a dentry to the hash according to its name. | |
2301 | */ | |
2302 | ||
2303 | void d_rehash(struct dentry * entry) | |
2304 | { | |
1da177e4 | 2305 | spin_lock(&entry->d_lock); |
770bfad8 | 2306 | _d_rehash(entry); |
1da177e4 | 2307 | spin_unlock(&entry->d_lock); |
1da177e4 | 2308 | } |
ec4f8605 | 2309 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2310 | |
fb2d5b86 NP |
2311 | /** |
2312 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2313 | * @dentry: dentry to be updated | |
2314 | * @name: new name | |
2315 | * | |
2316 | * Update a case insensitive dentry with new case of name. | |
2317 | * | |
2318 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2319 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2320 | * lengths). | |
2321 | * | |
2322 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2323 | * keep renames and concurrent inserts, and readdir(2) away). | |
2324 | */ | |
2325 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2326 | { | |
7ebfa57f | 2327 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2328 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2329 | ||
fb2d5b86 | 2330 | spin_lock(&dentry->d_lock); |
31e6b01f | 2331 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2332 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2333 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2334 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2335 | } |
2336 | EXPORT_SYMBOL(dentry_update_name_case); | |
2337 | ||
1da177e4 LT |
2338 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2339 | { | |
2340 | if (dname_external(target)) { | |
2341 | if (dname_external(dentry)) { | |
2342 | /* | |
2343 | * Both external: swap the pointers | |
2344 | */ | |
9a8d5bb4 | 2345 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2346 | } else { |
2347 | /* | |
2348 | * dentry:internal, target:external. Steal target's | |
2349 | * storage and make target internal. | |
2350 | */ | |
321bcf92 BF |
2351 | memcpy(target->d_iname, dentry->d_name.name, |
2352 | dentry->d_name.len + 1); | |
1da177e4 LT |
2353 | dentry->d_name.name = target->d_name.name; |
2354 | target->d_name.name = target->d_iname; | |
2355 | } | |
2356 | } else { | |
2357 | if (dname_external(dentry)) { | |
2358 | /* | |
2359 | * dentry:external, target:internal. Give dentry's | |
2360 | * storage to target and make dentry internal | |
2361 | */ | |
2362 | memcpy(dentry->d_iname, target->d_name.name, | |
2363 | target->d_name.len + 1); | |
2364 | target->d_name.name = dentry->d_name.name; | |
2365 | dentry->d_name.name = dentry->d_iname; | |
2366 | } else { | |
2367 | /* | |
2368 | * Both are internal. Just copy target to dentry | |
2369 | */ | |
2370 | memcpy(dentry->d_iname, target->d_name.name, | |
2371 | target->d_name.len + 1); | |
dc711ca3 AV |
2372 | dentry->d_name.len = target->d_name.len; |
2373 | return; | |
1da177e4 LT |
2374 | } |
2375 | } | |
9a8d5bb4 | 2376 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2377 | } |
2378 | ||
2fd6b7f5 NP |
2379 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2380 | { | |
2381 | /* | |
2382 | * XXXX: do we really need to take target->d_lock? | |
2383 | */ | |
2384 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2385 | spin_lock(&target->d_parent->d_lock); | |
2386 | else { | |
2387 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2388 | spin_lock(&dentry->d_parent->d_lock); | |
2389 | spin_lock_nested(&target->d_parent->d_lock, | |
2390 | DENTRY_D_LOCK_NESTED); | |
2391 | } else { | |
2392 | spin_lock(&target->d_parent->d_lock); | |
2393 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2394 | DENTRY_D_LOCK_NESTED); | |
2395 | } | |
2396 | } | |
2397 | if (target < dentry) { | |
2398 | spin_lock_nested(&target->d_lock, 2); | |
2399 | spin_lock_nested(&dentry->d_lock, 3); | |
2400 | } else { | |
2401 | spin_lock_nested(&dentry->d_lock, 2); | |
2402 | spin_lock_nested(&target->d_lock, 3); | |
2403 | } | |
2404 | } | |
2405 | ||
2406 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2407 | struct dentry *target) | |
2408 | { | |
2409 | if (target->d_parent != dentry->d_parent) | |
2410 | spin_unlock(&dentry->d_parent->d_lock); | |
2411 | if (target->d_parent != target) | |
2412 | spin_unlock(&target->d_parent->d_lock); | |
2413 | } | |
2414 | ||
1da177e4 | 2415 | /* |
2fd6b7f5 NP |
2416 | * When switching names, the actual string doesn't strictly have to |
2417 | * be preserved in the target - because we're dropping the target | |
2418 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2419 | * the new name before we switch. | |
2420 | * | |
2421 | * Note that we have to be a lot more careful about getting the hash | |
2422 | * switched - we have to switch the hash value properly even if it | |
2423 | * then no longer matches the actual (corrupted) string of the target. | |
2424 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2425 | */ |
9eaef27b | 2426 | /* |
18367501 | 2427 | * __d_move - move a dentry |
1da177e4 LT |
2428 | * @dentry: entry to move |
2429 | * @target: new dentry | |
2430 | * | |
2431 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2432 | * dcache entries should not be moved in this way. Caller must hold |
2433 | * rename_lock, the i_mutex of the source and target directories, | |
2434 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2435 | */ |
18367501 | 2436 | static void __d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2437 | { |
1da177e4 LT |
2438 | if (!dentry->d_inode) |
2439 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2440 | ||
2fd6b7f5 NP |
2441 | BUG_ON(d_ancestor(dentry, target)); |
2442 | BUG_ON(d_ancestor(target, dentry)); | |
2443 | ||
2fd6b7f5 | 2444 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2445 | |
31e6b01f NP |
2446 | write_seqcount_begin(&dentry->d_seq); |
2447 | write_seqcount_begin(&target->d_seq); | |
2448 | ||
ceb5bdc2 NP |
2449 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2450 | ||
2451 | /* | |
2452 | * Move the dentry to the target hash queue. Don't bother checking | |
2453 | * for the same hash queue because of how unlikely it is. | |
2454 | */ | |
2455 | __d_drop(dentry); | |
789680d1 | 2456 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2457 | |
2458 | /* Unhash the target: dput() will then get rid of it */ | |
2459 | __d_drop(target); | |
2460 | ||
5160ee6f ED |
2461 | list_del(&dentry->d_u.d_child); |
2462 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2463 | |
2464 | /* Switch the names.. */ | |
2465 | switch_names(dentry, target); | |
9a8d5bb4 | 2466 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2467 | |
2468 | /* ... and switch the parents */ | |
2469 | if (IS_ROOT(dentry)) { | |
2470 | dentry->d_parent = target->d_parent; | |
2471 | target->d_parent = target; | |
5160ee6f | 2472 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2473 | } else { |
9a8d5bb4 | 2474 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2475 | |
2476 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2477 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2478 | } |
2479 | ||
5160ee6f | 2480 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2481 | |
31e6b01f NP |
2482 | write_seqcount_end(&target->d_seq); |
2483 | write_seqcount_end(&dentry->d_seq); | |
2484 | ||
2fd6b7f5 | 2485 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2486 | spin_unlock(&target->d_lock); |
c32ccd87 | 2487 | fsnotify_d_move(dentry); |
1da177e4 | 2488 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2489 | } |
2490 | ||
2491 | /* | |
2492 | * d_move - move a dentry | |
2493 | * @dentry: entry to move | |
2494 | * @target: new dentry | |
2495 | * | |
2496 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2497 | * dcache entries should not be moved in this way. See the locking |
2498 | * requirements for __d_move. | |
18367501 AV |
2499 | */ |
2500 | void d_move(struct dentry *dentry, struct dentry *target) | |
2501 | { | |
2502 | write_seqlock(&rename_lock); | |
2503 | __d_move(dentry, target); | |
1da177e4 | 2504 | write_sequnlock(&rename_lock); |
9eaef27b | 2505 | } |
ec4f8605 | 2506 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2507 | |
e2761a11 OH |
2508 | /** |
2509 | * d_ancestor - search for an ancestor | |
2510 | * @p1: ancestor dentry | |
2511 | * @p2: child dentry | |
2512 | * | |
2513 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2514 | * an ancestor of p2, else NULL. | |
9eaef27b | 2515 | */ |
e2761a11 | 2516 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2517 | { |
2518 | struct dentry *p; | |
2519 | ||
871c0067 | 2520 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2521 | if (p->d_parent == p1) |
e2761a11 | 2522 | return p; |
9eaef27b | 2523 | } |
e2761a11 | 2524 | return NULL; |
9eaef27b TM |
2525 | } |
2526 | ||
2527 | /* | |
2528 | * This helper attempts to cope with remotely renamed directories | |
2529 | * | |
2530 | * It assumes that the caller is already holding | |
18367501 | 2531 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2532 | * |
2533 | * Note: If ever the locking in lock_rename() changes, then please | |
2534 | * remember to update this too... | |
9eaef27b | 2535 | */ |
873feea0 NP |
2536 | static struct dentry *__d_unalias(struct inode *inode, |
2537 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2538 | { |
2539 | struct mutex *m1 = NULL, *m2 = NULL; | |
ee3efa91 | 2540 | struct dentry *ret = ERR_PTR(-EBUSY); |
9eaef27b TM |
2541 | |
2542 | /* If alias and dentry share a parent, then no extra locks required */ | |
2543 | if (alias->d_parent == dentry->d_parent) | |
2544 | goto out_unalias; | |
2545 | ||
9eaef27b | 2546 | /* See lock_rename() */ |
9eaef27b TM |
2547 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2548 | goto out_err; | |
2549 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2550 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2551 | goto out_err; | |
2552 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2553 | out_unalias: | |
ee3efa91 AV |
2554 | if (likely(!d_mountpoint(alias))) { |
2555 | __d_move(alias, dentry); | |
2556 | ret = alias; | |
2557 | } | |
9eaef27b | 2558 | out_err: |
873feea0 | 2559 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2560 | if (m2) |
2561 | mutex_unlock(m2); | |
2562 | if (m1) | |
2563 | mutex_unlock(m1); | |
2564 | return ret; | |
2565 | } | |
2566 | ||
770bfad8 DH |
2567 | /* |
2568 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2569 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2570 | * returns with anon->d_lock held! |
770bfad8 DH |
2571 | */ |
2572 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2573 | { | |
740da42e | 2574 | struct dentry *dparent; |
770bfad8 | 2575 | |
2fd6b7f5 | 2576 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2577 | |
31e6b01f NP |
2578 | write_seqcount_begin(&dentry->d_seq); |
2579 | write_seqcount_begin(&anon->d_seq); | |
2580 | ||
770bfad8 | 2581 | dparent = dentry->d_parent; |
770bfad8 | 2582 | |
2fd6b7f5 NP |
2583 | switch_names(dentry, anon); |
2584 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2585 | ||
740da42e AV |
2586 | dentry->d_parent = dentry; |
2587 | list_del_init(&dentry->d_u.d_child); | |
2588 | anon->d_parent = dparent; | |
9ed53b12 | 2589 | list_move(&anon->d_u.d_child, &dparent->d_subdirs); |
770bfad8 | 2590 | |
31e6b01f NP |
2591 | write_seqcount_end(&dentry->d_seq); |
2592 | write_seqcount_end(&anon->d_seq); | |
2593 | ||
2fd6b7f5 NP |
2594 | dentry_unlock_parents_for_move(anon, dentry); |
2595 | spin_unlock(&dentry->d_lock); | |
2596 | ||
2597 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2598 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2599 | } | |
2600 | ||
2601 | /** | |
2602 | * d_materialise_unique - introduce an inode into the tree | |
2603 | * @dentry: candidate dentry | |
2604 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2605 | * | |
2606 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2607 | * root directory alias in its place if there is one. Caller must hold the |
2608 | * i_mutex of the parent directory. | |
770bfad8 DH |
2609 | */ |
2610 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2611 | { | |
9eaef27b | 2612 | struct dentry *actual; |
770bfad8 DH |
2613 | |
2614 | BUG_ON(!d_unhashed(dentry)); | |
2615 | ||
770bfad8 DH |
2616 | if (!inode) { |
2617 | actual = dentry; | |
360da900 | 2618 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2619 | d_rehash(actual); |
2620 | goto out_nolock; | |
770bfad8 DH |
2621 | } |
2622 | ||
873feea0 | 2623 | spin_lock(&inode->i_lock); |
357f8e65 | 2624 | |
9eaef27b TM |
2625 | if (S_ISDIR(inode->i_mode)) { |
2626 | struct dentry *alias; | |
2627 | ||
2628 | /* Does an aliased dentry already exist? */ | |
32ba9c3f | 2629 | alias = __d_find_alias(inode, 0); |
9eaef27b TM |
2630 | if (alias) { |
2631 | actual = alias; | |
18367501 AV |
2632 | write_seqlock(&rename_lock); |
2633 | ||
2634 | if (d_ancestor(alias, dentry)) { | |
2635 | /* Check for loops */ | |
2636 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2637 | spin_unlock(&inode->i_lock); |
18367501 AV |
2638 | } else if (IS_ROOT(alias)) { |
2639 | /* Is this an anonymous mountpoint that we | |
2640 | * could splice into our tree? */ | |
9eaef27b | 2641 | __d_materialise_dentry(dentry, alias); |
18367501 | 2642 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2643 | __d_drop(alias); |
2644 | goto found; | |
18367501 AV |
2645 | } else { |
2646 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2647 | * aliasing. This drops inode->i_lock */ |
18367501 | 2648 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2649 | } |
18367501 | 2650 | write_sequnlock(&rename_lock); |
dd179946 DH |
2651 | if (IS_ERR(actual)) { |
2652 | if (PTR_ERR(actual) == -ELOOP) | |
2653 | pr_warn_ratelimited( | |
2654 | "VFS: Lookup of '%s' in %s %s" | |
2655 | " would have caused loop\n", | |
2656 | dentry->d_name.name, | |
2657 | inode->i_sb->s_type->name, | |
2658 | inode->i_sb->s_id); | |
9eaef27b | 2659 | dput(alias); |
dd179946 | 2660 | } |
9eaef27b TM |
2661 | goto out_nolock; |
2662 | } | |
770bfad8 DH |
2663 | } |
2664 | ||
2665 | /* Add a unique reference */ | |
2666 | actual = __d_instantiate_unique(dentry, inode); | |
2667 | if (!actual) | |
2668 | actual = dentry; | |
357f8e65 NP |
2669 | else |
2670 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2671 | |
770bfad8 DH |
2672 | spin_lock(&actual->d_lock); |
2673 | found: | |
2674 | _d_rehash(actual); | |
2675 | spin_unlock(&actual->d_lock); | |
873feea0 | 2676 | spin_unlock(&inode->i_lock); |
9eaef27b | 2677 | out_nolock: |
770bfad8 DH |
2678 | if (actual == dentry) { |
2679 | security_d_instantiate(dentry, inode); | |
2680 | return NULL; | |
2681 | } | |
2682 | ||
2683 | iput(inode); | |
2684 | return actual; | |
770bfad8 | 2685 | } |
ec4f8605 | 2686 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2687 | |
cdd16d02 | 2688 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2689 | { |
2690 | *buflen -= namelen; | |
2691 | if (*buflen < 0) | |
2692 | return -ENAMETOOLONG; | |
2693 | *buffer -= namelen; | |
2694 | memcpy(*buffer, str, namelen); | |
2695 | return 0; | |
2696 | } | |
2697 | ||
232d2d60 WL |
2698 | /** |
2699 | * prepend_name - prepend a pathname in front of current buffer pointer | |
2700 | * buffer: buffer pointer | |
2701 | * buflen: allocated length of the buffer | |
2702 | * name: name string and length qstr structure | |
2703 | * | |
2704 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
2705 | * make sure that either the old or the new name pointer and length are | |
2706 | * fetched. However, there may be mismatch between length and pointer. | |
2707 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
2708 | * the length is reached or a null byte is found. It also prepends "/" at | |
2709 | * the beginning of the name. The sequence number check at the caller will | |
2710 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
2711 | * due to mismatched pointer and length will be discarded. | |
2712 | */ | |
cdd16d02 MS |
2713 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2714 | { | |
232d2d60 WL |
2715 | const char *dname = ACCESS_ONCE(name->name); |
2716 | u32 dlen = ACCESS_ONCE(name->len); | |
2717 | char *p; | |
2718 | ||
2719 | if (*buflen < dlen + 1) | |
2720 | return -ENAMETOOLONG; | |
2721 | *buflen -= dlen + 1; | |
2722 | p = *buffer -= dlen + 1; | |
2723 | *p++ = '/'; | |
2724 | while (dlen--) { | |
2725 | char c = *dname++; | |
2726 | if (!c) | |
2727 | break; | |
2728 | *p++ = c; | |
2729 | } | |
2730 | return 0; | |
cdd16d02 MS |
2731 | } |
2732 | ||
1da177e4 | 2733 | /** |
208898c1 | 2734 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2735 | * @path: the dentry/vfsmount to report |
02125a82 | 2736 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2737 | * @buffer: pointer to the end of the buffer |
2738 | * @buflen: pointer to buffer length | |
552ce544 | 2739 | * |
232d2d60 WL |
2740 | * The function tries to write out the pathname without taking any lock other |
2741 | * than the RCU read lock to make sure that dentries won't go away. It only | |
2742 | * checks the sequence number of the global rename_lock as any change in the | |
2743 | * dentry's d_seq will be preceded by changes in the rename_lock sequence | |
2744 | * number. If the sequence number had been change, it will restart the whole | |
2745 | * pathname back-tracing sequence again. It performs a total of 3 trials of | |
2746 | * lockless back-tracing sequences before falling back to take the | |
2747 | * rename_lock. | |
1da177e4 | 2748 | */ |
02125a82 AV |
2749 | static int prepend_path(const struct path *path, |
2750 | const struct path *root, | |
f2eb6575 | 2751 | char **buffer, int *buflen) |
1da177e4 | 2752 | { |
9d1bc601 MS |
2753 | struct dentry *dentry = path->dentry; |
2754 | struct vfsmount *vfsmnt = path->mnt; | |
0714a533 | 2755 | struct mount *mnt = real_mount(vfsmnt); |
f2eb6575 | 2756 | int error = 0; |
232d2d60 WL |
2757 | unsigned seq = 0; |
2758 | char *bptr; | |
2759 | int blen; | |
6092d048 | 2760 | |
48f5ec21 | 2761 | rcu_read_lock(); |
232d2d60 WL |
2762 | restart: |
2763 | bptr = *buffer; | |
2764 | blen = *buflen; | |
2765 | read_seqbegin_or_lock(&rename_lock, &seq); | |
f2eb6575 | 2766 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2767 | struct dentry * parent; |
2768 | ||
1da177e4 | 2769 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2770 | /* Global root? */ |
232d2d60 WL |
2771 | if (mnt_has_parent(mnt)) { |
2772 | dentry = mnt->mnt_mountpoint; | |
2773 | mnt = mnt->mnt_parent; | |
2774 | vfsmnt = &mnt->mnt; | |
2775 | continue; | |
2776 | } | |
2777 | /* | |
2778 | * Filesystems needing to implement special "root names" | |
2779 | * should do so with ->d_dname() | |
2780 | */ | |
2781 | if (IS_ROOT(dentry) && | |
2782 | (dentry->d_name.len != 1 || | |
2783 | dentry->d_name.name[0] != '/')) { | |
2784 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2785 | (int) dentry->d_name.len, | |
2786 | dentry->d_name.name); | |
2787 | } | |
2788 | if (!error) | |
2789 | error = is_mounted(vfsmnt) ? 1 : 2; | |
2790 | break; | |
1da177e4 LT |
2791 | } |
2792 | parent = dentry->d_parent; | |
2793 | prefetch(parent); | |
232d2d60 | 2794 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
2795 | if (error) |
2796 | break; | |
2797 | ||
1da177e4 LT |
2798 | dentry = parent; |
2799 | } | |
48f5ec21 AV |
2800 | if (!(seq & 1)) |
2801 | rcu_read_unlock(); | |
2802 | if (need_seqretry(&rename_lock, seq)) { | |
2803 | seq = 1; | |
232d2d60 | 2804 | goto restart; |
48f5ec21 AV |
2805 | } |
2806 | done_seqretry(&rename_lock, seq); | |
1da177e4 | 2807 | |
232d2d60 WL |
2808 | if (error >= 0 && bptr == *buffer) { |
2809 | if (--blen < 0) | |
2810 | error = -ENAMETOOLONG; | |
2811 | else | |
2812 | *--bptr = '/'; | |
2813 | } | |
2814 | *buffer = bptr; | |
2815 | *buflen = blen; | |
7ea600b5 | 2816 | return error; |
f2eb6575 | 2817 | } |
be285c71 | 2818 | |
f2eb6575 MS |
2819 | /** |
2820 | * __d_path - return the path of a dentry | |
2821 | * @path: the dentry/vfsmount to report | |
02125a82 | 2822 | * @root: root vfsmnt/dentry |
cd956a1c | 2823 | * @buf: buffer to return value in |
f2eb6575 MS |
2824 | * @buflen: buffer length |
2825 | * | |
ffd1f4ed | 2826 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2827 | * |
2828 | * Returns a pointer into the buffer or an error code if the | |
2829 | * path was too long. | |
2830 | * | |
be148247 | 2831 | * "buflen" should be positive. |
f2eb6575 | 2832 | * |
02125a82 | 2833 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2834 | */ |
02125a82 AV |
2835 | char *__d_path(const struct path *path, |
2836 | const struct path *root, | |
f2eb6575 MS |
2837 | char *buf, int buflen) |
2838 | { | |
2839 | char *res = buf + buflen; | |
2840 | int error; | |
2841 | ||
2842 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2843 | br_read_lock(&vfsmount_lock); |
f2eb6575 | 2844 | error = prepend_path(path, root, &res, &buflen); |
7ea600b5 | 2845 | br_read_unlock(&vfsmount_lock); |
be148247 | 2846 | |
02125a82 AV |
2847 | if (error < 0) |
2848 | return ERR_PTR(error); | |
2849 | if (error > 0) | |
2850 | return NULL; | |
2851 | return res; | |
2852 | } | |
2853 | ||
2854 | char *d_absolute_path(const struct path *path, | |
2855 | char *buf, int buflen) | |
2856 | { | |
2857 | struct path root = {}; | |
2858 | char *res = buf + buflen; | |
2859 | int error; | |
2860 | ||
2861 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2862 | br_read_lock(&vfsmount_lock); |
02125a82 | 2863 | error = prepend_path(path, &root, &res, &buflen); |
7ea600b5 | 2864 | br_read_unlock(&vfsmount_lock); |
02125a82 AV |
2865 | |
2866 | if (error > 1) | |
2867 | error = -EINVAL; | |
2868 | if (error < 0) | |
f2eb6575 | 2869 | return ERR_PTR(error); |
f2eb6575 | 2870 | return res; |
1da177e4 LT |
2871 | } |
2872 | ||
ffd1f4ed MS |
2873 | /* |
2874 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2875 | */ | |
02125a82 AV |
2876 | static int path_with_deleted(const struct path *path, |
2877 | const struct path *root, | |
2878 | char **buf, int *buflen) | |
ffd1f4ed MS |
2879 | { |
2880 | prepend(buf, buflen, "\0", 1); | |
2881 | if (d_unlinked(path->dentry)) { | |
2882 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2883 | if (error) | |
2884 | return error; | |
2885 | } | |
2886 | ||
2887 | return prepend_path(path, root, buf, buflen); | |
2888 | } | |
2889 | ||
8df9d1a4 MS |
2890 | static int prepend_unreachable(char **buffer, int *buflen) |
2891 | { | |
2892 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2893 | } | |
2894 | ||
a03a8a70 JB |
2895 | /** |
2896 | * d_path - return the path of a dentry | |
cf28b486 | 2897 | * @path: path to report |
a03a8a70 JB |
2898 | * @buf: buffer to return value in |
2899 | * @buflen: buffer length | |
2900 | * | |
2901 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2902 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2903 | * | |
52afeefb AV |
2904 | * Returns a pointer into the buffer or an error code if the path was |
2905 | * too long. Note: Callers should use the returned pointer, not the passed | |
2906 | * in buffer, to use the name! The implementation often starts at an offset | |
2907 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2908 | * |
31f3e0b3 | 2909 | * "buflen" should be positive. |
a03a8a70 | 2910 | */ |
20d4fdc1 | 2911 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2912 | { |
ffd1f4ed | 2913 | char *res = buf + buflen; |
6ac08c39 | 2914 | struct path root; |
ffd1f4ed | 2915 | int error; |
1da177e4 | 2916 | |
c23fbb6b ED |
2917 | /* |
2918 | * We have various synthetic filesystems that never get mounted. On | |
2919 | * these filesystems dentries are never used for lookup purposes, and | |
2920 | * thus don't need to be hashed. They also don't need a name until a | |
2921 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2922 | * below allows us to generate a name for these objects on demand: | |
2923 | */ | |
cf28b486 JB |
2924 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2925 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2926 | |
f7ad3c6b | 2927 | get_fs_root(current->fs, &root); |
7ea600b5 | 2928 | br_read_lock(&vfsmount_lock); |
02125a82 | 2929 | error = path_with_deleted(path, &root, &res, &buflen); |
7ea600b5 | 2930 | br_read_unlock(&vfsmount_lock); |
02125a82 | 2931 | if (error < 0) |
ffd1f4ed | 2932 | res = ERR_PTR(error); |
6ac08c39 | 2933 | path_put(&root); |
1da177e4 LT |
2934 | return res; |
2935 | } | |
ec4f8605 | 2936 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2937 | |
c23fbb6b ED |
2938 | /* |
2939 | * Helper function for dentry_operations.d_dname() members | |
2940 | */ | |
2941 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2942 | const char *fmt, ...) | |
2943 | { | |
2944 | va_list args; | |
2945 | char temp[64]; | |
2946 | int sz; | |
2947 | ||
2948 | va_start(args, fmt); | |
2949 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2950 | va_end(args); | |
2951 | ||
2952 | if (sz > sizeof(temp) || sz > buflen) | |
2953 | return ERR_PTR(-ENAMETOOLONG); | |
2954 | ||
2955 | buffer += buflen - sz; | |
2956 | return memcpy(buffer, temp, sz); | |
2957 | } | |
2958 | ||
118b2302 AV |
2959 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
2960 | { | |
2961 | char *end = buffer + buflen; | |
2962 | /* these dentries are never renamed, so d_lock is not needed */ | |
2963 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
232d2d60 | 2964 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
118b2302 AV |
2965 | prepend(&end, &buflen, "/", 1)) |
2966 | end = ERR_PTR(-ENAMETOOLONG); | |
232d2d60 | 2967 | return end; |
118b2302 AV |
2968 | } |
2969 | ||
6092d048 RP |
2970 | /* |
2971 | * Write full pathname from the root of the filesystem into the buffer. | |
2972 | */ | |
ec2447c2 | 2973 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 | 2974 | { |
232d2d60 WL |
2975 | char *end, *retval; |
2976 | int len, seq = 0; | |
2977 | int error = 0; | |
6092d048 | 2978 | |
48f5ec21 | 2979 | rcu_read_lock(); |
232d2d60 WL |
2980 | restart: |
2981 | end = buf + buflen; | |
2982 | len = buflen; | |
2983 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
2984 | if (buflen < 1) |
2985 | goto Elong; | |
2986 | /* Get '/' right */ | |
2987 | retval = end-1; | |
2988 | *retval = '/'; | |
232d2d60 | 2989 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
2990 | while (!IS_ROOT(dentry)) { |
2991 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2992 | int error; |
6092d048 | 2993 | |
6092d048 | 2994 | prefetch(parent); |
232d2d60 WL |
2995 | error = prepend_name(&end, &len, &dentry->d_name); |
2996 | if (error) | |
2997 | break; | |
6092d048 RP |
2998 | |
2999 | retval = end; | |
3000 | dentry = parent; | |
3001 | } | |
48f5ec21 AV |
3002 | if (!(seq & 1)) |
3003 | rcu_read_unlock(); | |
3004 | if (need_seqretry(&rename_lock, seq)) { | |
3005 | seq = 1; | |
232d2d60 | 3006 | goto restart; |
48f5ec21 AV |
3007 | } |
3008 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3009 | if (error) |
3010 | goto Elong; | |
c103135c AV |
3011 | return retval; |
3012 | Elong: | |
3013 | return ERR_PTR(-ENAMETOOLONG); | |
3014 | } | |
ec2447c2 NP |
3015 | |
3016 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3017 | { | |
232d2d60 | 3018 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3019 | } |
3020 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3021 | |
3022 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3023 | { | |
3024 | char *p = NULL; | |
3025 | char *retval; | |
3026 | ||
c103135c AV |
3027 | if (d_unlinked(dentry)) { |
3028 | p = buf + buflen; | |
3029 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3030 | goto Elong; | |
3031 | buflen++; | |
3032 | } | |
3033 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3034 | if (!IS_ERR(retval) && p) |
3035 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3036 | return retval; |
3037 | Elong: | |
6092d048 RP |
3038 | return ERR_PTR(-ENAMETOOLONG); |
3039 | } | |
3040 | ||
1da177e4 LT |
3041 | /* |
3042 | * NOTE! The user-level library version returns a | |
3043 | * character pointer. The kernel system call just | |
3044 | * returns the length of the buffer filled (which | |
3045 | * includes the ending '\0' character), or a negative | |
3046 | * error value. So libc would do something like | |
3047 | * | |
3048 | * char *getcwd(char * buf, size_t size) | |
3049 | * { | |
3050 | * int retval; | |
3051 | * | |
3052 | * retval = sys_getcwd(buf, size); | |
3053 | * if (retval >= 0) | |
3054 | * return buf; | |
3055 | * errno = -retval; | |
3056 | * return NULL; | |
3057 | * } | |
3058 | */ | |
3cdad428 | 3059 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3060 | { |
552ce544 | 3061 | int error; |
6ac08c39 | 3062 | struct path pwd, root; |
552ce544 | 3063 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
3064 | |
3065 | if (!page) | |
3066 | return -ENOMEM; | |
3067 | ||
f7ad3c6b | 3068 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 3069 | |
552ce544 | 3070 | error = -ENOENT; |
7ea600b5 | 3071 | br_read_lock(&vfsmount_lock); |
f3da392e | 3072 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3073 | unsigned long len; |
8df9d1a4 MS |
3074 | char *cwd = page + PAGE_SIZE; |
3075 | int buflen = PAGE_SIZE; | |
1da177e4 | 3076 | |
8df9d1a4 | 3077 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3078 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
7ea600b5 | 3079 | br_read_unlock(&vfsmount_lock); |
552ce544 | 3080 | |
02125a82 | 3081 | if (error < 0) |
552ce544 LT |
3082 | goto out; |
3083 | ||
8df9d1a4 | 3084 | /* Unreachable from current root */ |
02125a82 | 3085 | if (error > 0) { |
8df9d1a4 MS |
3086 | error = prepend_unreachable(&cwd, &buflen); |
3087 | if (error) | |
3088 | goto out; | |
3089 | } | |
3090 | ||
552ce544 LT |
3091 | error = -ERANGE; |
3092 | len = PAGE_SIZE + page - cwd; | |
3093 | if (len <= size) { | |
3094 | error = len; | |
3095 | if (copy_to_user(buf, cwd, len)) | |
3096 | error = -EFAULT; | |
3097 | } | |
949854d0 | 3098 | } else { |
7ea600b5 | 3099 | br_read_unlock(&vfsmount_lock); |
949854d0 | 3100 | } |
1da177e4 LT |
3101 | |
3102 | out: | |
6ac08c39 JB |
3103 | path_put(&pwd); |
3104 | path_put(&root); | |
1da177e4 LT |
3105 | free_page((unsigned long) page); |
3106 | return error; | |
3107 | } | |
3108 | ||
3109 | /* | |
3110 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3111 | * | |
3112 | * Trivially implemented using the dcache structure | |
3113 | */ | |
3114 | ||
3115 | /** | |
3116 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3117 | * @new_dentry: new dentry | |
3118 | * @old_dentry: old dentry | |
3119 | * | |
3120 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
3121 | * Returns 0 otherwise. | |
3122 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
3123 | */ | |
3124 | ||
e2761a11 | 3125 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
3126 | { |
3127 | int result; | |
949854d0 | 3128 | unsigned seq; |
1da177e4 | 3129 | |
e2761a11 OH |
3130 | if (new_dentry == old_dentry) |
3131 | return 1; | |
3132 | ||
e2761a11 | 3133 | do { |
1da177e4 | 3134 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3135 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3136 | /* |
3137 | * Need rcu_readlock to protect against the d_parent trashing | |
3138 | * due to d_move | |
3139 | */ | |
3140 | rcu_read_lock(); | |
e2761a11 | 3141 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 3142 | result = 1; |
e2761a11 OH |
3143 | else |
3144 | result = 0; | |
949854d0 | 3145 | rcu_read_unlock(); |
1da177e4 | 3146 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3147 | |
3148 | return result; | |
3149 | } | |
3150 | ||
db14fc3a | 3151 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3152 | { |
db14fc3a MS |
3153 | struct dentry *root = data; |
3154 | if (dentry != root) { | |
3155 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3156 | return D_WALK_SKIP; | |
1da177e4 | 3157 | |
01ddc4ed MS |
3158 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3159 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3160 | dentry->d_lockref.count--; | |
3161 | } | |
1da177e4 | 3162 | } |
db14fc3a MS |
3163 | return D_WALK_CONTINUE; |
3164 | } | |
58db63d0 | 3165 | |
db14fc3a MS |
3166 | void d_genocide(struct dentry *parent) |
3167 | { | |
3168 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3169 | } |
3170 | ||
60545d0d | 3171 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3172 | { |
60545d0d AV |
3173 | inode_dec_link_count(inode); |
3174 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
3175 | !hlist_unhashed(&dentry->d_alias) || | |
3176 | !d_unlinked(dentry)); | |
3177 | spin_lock(&dentry->d_parent->d_lock); | |
3178 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3179 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3180 | (unsigned long long)inode->i_ino); | |
3181 | spin_unlock(&dentry->d_lock); | |
3182 | spin_unlock(&dentry->d_parent->d_lock); | |
3183 | d_instantiate(dentry, inode); | |
1da177e4 | 3184 | } |
60545d0d | 3185 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3186 | |
3187 | static __initdata unsigned long dhash_entries; | |
3188 | static int __init set_dhash_entries(char *str) | |
3189 | { | |
3190 | if (!str) | |
3191 | return 0; | |
3192 | dhash_entries = simple_strtoul(str, &str, 0); | |
3193 | return 1; | |
3194 | } | |
3195 | __setup("dhash_entries=", set_dhash_entries); | |
3196 | ||
3197 | static void __init dcache_init_early(void) | |
3198 | { | |
074b8517 | 3199 | unsigned int loop; |
1da177e4 LT |
3200 | |
3201 | /* If hashes are distributed across NUMA nodes, defer | |
3202 | * hash allocation until vmalloc space is available. | |
3203 | */ | |
3204 | if (hashdist) | |
3205 | return; | |
3206 | ||
3207 | dentry_hashtable = | |
3208 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3209 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3210 | dhash_entries, |
3211 | 13, | |
3212 | HASH_EARLY, | |
3213 | &d_hash_shift, | |
3214 | &d_hash_mask, | |
31fe62b9 | 3215 | 0, |
1da177e4 LT |
3216 | 0); |
3217 | ||
074b8517 | 3218 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3219 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3220 | } |
3221 | ||
74bf17cf | 3222 | static void __init dcache_init(void) |
1da177e4 | 3223 | { |
074b8517 | 3224 | unsigned int loop; |
1da177e4 LT |
3225 | |
3226 | /* | |
3227 | * A constructor could be added for stable state like the lists, | |
3228 | * but it is probably not worth it because of the cache nature | |
3229 | * of the dcache. | |
3230 | */ | |
0a31bd5f CL |
3231 | dentry_cache = KMEM_CACHE(dentry, |
3232 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3233 | |
3234 | /* Hash may have been set up in dcache_init_early */ | |
3235 | if (!hashdist) | |
3236 | return; | |
3237 | ||
3238 | dentry_hashtable = | |
3239 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3240 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3241 | dhash_entries, |
3242 | 13, | |
3243 | 0, | |
3244 | &d_hash_shift, | |
3245 | &d_hash_mask, | |
31fe62b9 | 3246 | 0, |
1da177e4 LT |
3247 | 0); |
3248 | ||
074b8517 | 3249 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3250 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3251 | } |
3252 | ||
3253 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3254 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3255 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3256 | |
1da177e4 LT |
3257 | EXPORT_SYMBOL(d_genocide); |
3258 | ||
1da177e4 LT |
3259 | void __init vfs_caches_init_early(void) |
3260 | { | |
3261 | dcache_init_early(); | |
3262 | inode_init_early(); | |
3263 | } | |
3264 | ||
3265 | void __init vfs_caches_init(unsigned long mempages) | |
3266 | { | |
3267 | unsigned long reserve; | |
3268 | ||
3269 | /* Base hash sizes on available memory, with a reserve equal to | |
3270 | 150% of current kernel size */ | |
3271 | ||
3272 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3273 | mempages -= reserve; | |
3274 | ||
3275 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3276 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3277 | |
74bf17cf DC |
3278 | dcache_init(); |
3279 | inode_init(); | |
1da177e4 | 3280 | files_init(mempages); |
74bf17cf | 3281 | mnt_init(); |
1da177e4 LT |
3282 | bdev_cache_init(); |
3283 | chrdev_init(); | |
3284 | } |