]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/dcache.c | |
3 | * | |
4 | * Complete reimplementation | |
5 | * (C) 1997 Thomas Schoebel-Theuer, | |
6 | * with heavy changes by Linus Torvalds | |
7 | */ | |
8 | ||
9 | /* | |
10 | * Notes on the allocation strategy: | |
11 | * | |
12 | * The dcache is a master of the icache - whenever a dcache entry | |
13 | * exists, the inode will always exist. "iput()" is done either when | |
14 | * the dcache entry is deleted or garbage collected. | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/syscalls.h> |
18 | #include <linux/string.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/fs.h> | |
7a91bf7f | 21 | #include <linux/fsnotify.h> |
1da177e4 LT |
22 | #include <linux/slab.h> |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/hash.h> |
25 | #include <linux/cache.h> | |
630d9c47 | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/mount.h> |
28 | #include <linux/file.h> | |
29 | #include <asm/uaccess.h> | |
30 | #include <linux/security.h> | |
31 | #include <linux/seqlock.h> | |
32 | #include <linux/swap.h> | |
33 | #include <linux/bootmem.h> | |
5ad4e53b | 34 | #include <linux/fs_struct.h> |
613afbf8 | 35 | #include <linux/hardirq.h> |
ceb5bdc2 NP |
36 | #include <linux/bit_spinlock.h> |
37 | #include <linux/rculist_bl.h> | |
268bb0ce | 38 | #include <linux/prefetch.h> |
dd179946 | 39 | #include <linux/ratelimit.h> |
f6041567 | 40 | #include <linux/list_lru.h> |
07f3f05c | 41 | #include "internal.h" |
b2dba1af | 42 | #include "mount.h" |
1da177e4 | 43 | |
789680d1 NP |
44 | /* |
45 | * Usage: | |
873feea0 NP |
46 | * dcache->d_inode->i_lock protects: |
47 | * - i_dentry, d_alias, d_inode of aliases | |
ceb5bdc2 NP |
48 | * dcache_hash_bucket lock protects: |
49 | * - the dcache hash table | |
50 | * s_anon bl list spinlock protects: | |
51 | * - the s_anon list (see __d_drop) | |
19156840 | 52 | * dentry->d_sb->s_dentry_lru_lock protects: |
23044507 NP |
53 | * - the dcache lru lists and counters |
54 | * d_lock protects: | |
55 | * - d_flags | |
56 | * - d_name | |
57 | * - d_lru | |
b7ab39f6 | 58 | * - d_count |
da502956 | 59 | * - d_unhashed() |
2fd6b7f5 NP |
60 | * - d_parent and d_subdirs |
61 | * - childrens' d_child and d_parent | |
b23fb0a6 | 62 | * - d_alias, d_inode |
789680d1 NP |
63 | * |
64 | * Ordering: | |
873feea0 | 65 | * dentry->d_inode->i_lock |
b5c84bf6 | 66 | * dentry->d_lock |
19156840 | 67 | * dentry->d_sb->s_dentry_lru_lock |
ceb5bdc2 NP |
68 | * dcache_hash_bucket lock |
69 | * s_anon lock | |
789680d1 | 70 | * |
da502956 NP |
71 | * If there is an ancestor relationship: |
72 | * dentry->d_parent->...->d_parent->d_lock | |
73 | * ... | |
74 | * dentry->d_parent->d_lock | |
75 | * dentry->d_lock | |
76 | * | |
77 | * If no ancestor relationship: | |
789680d1 NP |
78 | * if (dentry1 < dentry2) |
79 | * dentry1->d_lock | |
80 | * dentry2->d_lock | |
81 | */ | |
fa3536cc | 82 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
83 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
84 | ||
74c3cbe3 | 85 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 86 | |
949854d0 | 87 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 88 | |
e18b890b | 89 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 90 | |
232d2d60 WL |
91 | /** |
92 | * read_seqbegin_or_lock - begin a sequence number check or locking block | |
18129977 WL |
93 | * @lock: sequence lock |
94 | * @seq : sequence number to be checked | |
232d2d60 WL |
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 */ |
18129977 | 106 | read_seqlock_excl(lock); |
232d2d60 WL |
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) | |
18129977 | 117 | read_sequnlock_excl(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 | ||
89dc77bc LT |
359 | /* |
360 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
361 | * is in use - which includes both the "real" per-superblock | |
362 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
363 | * | |
364 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
365 | * on the shrink list (ie not on the superblock LRU list). | |
366 | * | |
367 | * The per-cpu "nr_dentry_unused" counters are updated with | |
368 | * the DCACHE_LRU_LIST bit. | |
369 | * | |
370 | * These helper functions make sure we always follow the | |
371 | * rules. d_lock must be held by the caller. | |
372 | */ | |
373 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
374 | static void d_lru_add(struct dentry *dentry) | |
375 | { | |
376 | D_FLAG_VERIFY(dentry, 0); | |
377 | dentry->d_flags |= DCACHE_LRU_LIST; | |
378 | this_cpu_inc(nr_dentry_unused); | |
379 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
380 | } | |
381 | ||
382 | static void d_lru_del(struct dentry *dentry) | |
383 | { | |
384 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
385 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
386 | this_cpu_dec(nr_dentry_unused); | |
387 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
388 | } | |
389 | ||
390 | static void d_shrink_del(struct dentry *dentry) | |
391 | { | |
392 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
393 | list_del_init(&dentry->d_lru); | |
394 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
395 | this_cpu_dec(nr_dentry_unused); | |
396 | } | |
397 | ||
398 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
399 | { | |
400 | D_FLAG_VERIFY(dentry, 0); | |
401 | list_add(&dentry->d_lru, list); | |
402 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
403 | this_cpu_inc(nr_dentry_unused); | |
404 | } | |
405 | ||
406 | /* | |
407 | * These can only be called under the global LRU lock, ie during the | |
408 | * callback for freeing the LRU list. "isolate" removes it from the | |
409 | * LRU lists entirely, while shrink_move moves it to the indicated | |
410 | * private list. | |
411 | */ | |
412 | static void d_lru_isolate(struct dentry *dentry) | |
413 | { | |
414 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
415 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
416 | this_cpu_dec(nr_dentry_unused); | |
417 | list_del_init(&dentry->d_lru); | |
418 | } | |
419 | ||
420 | static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list) | |
421 | { | |
422 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
423 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
424 | list_move_tail(&dentry->d_lru, list); | |
425 | } | |
426 | ||
da3bbdd4 | 427 | /* |
f6041567 | 428 | * dentry_lru_(add|del)_list) must be called with d_lock held. |
da3bbdd4 KM |
429 | */ |
430 | static void dentry_lru_add(struct dentry *dentry) | |
431 | { | |
89dc77bc LT |
432 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) |
433 | d_lru_add(dentry); | |
da3bbdd4 KM |
434 | } |
435 | ||
f0023bc6 SW |
436 | /* |
437 | * Remove a dentry with references from the LRU. | |
dd1f6b2e DC |
438 | * |
439 | * If we are on the shrink list, then we can get to try_prune_one_dentry() and | |
440 | * lose our last reference through the parent walk. In this case, we need to | |
441 | * remove ourselves from the shrink list, not the LRU. | |
f0023bc6 | 442 | */ |
da3bbdd4 KM |
443 | static void dentry_lru_del(struct dentry *dentry) |
444 | { | |
89dc77bc LT |
445 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
446 | if (dentry->d_flags & DCACHE_SHRINK_LIST) | |
447 | return d_shrink_del(dentry); | |
448 | d_lru_del(dentry); | |
da3bbdd4 | 449 | } |
da3bbdd4 KM |
450 | } |
451 | ||
d52b9086 MS |
452 | /** |
453 | * d_kill - kill dentry and return parent | |
454 | * @dentry: dentry to kill | |
ff5fdb61 | 455 | * @parent: parent dentry |
d52b9086 | 456 | * |
31f3e0b3 | 457 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
458 | * |
459 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 460 | * |
b5c84bf6 NP |
461 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
462 | * d_kill. | |
d52b9086 | 463 | */ |
2fd6b7f5 | 464 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 465 | __releases(dentry->d_lock) |
2fd6b7f5 | 466 | __releases(parent->d_lock) |
873feea0 | 467 | __releases(dentry->d_inode->i_lock) |
d52b9086 | 468 | { |
d52b9086 | 469 | list_del(&dentry->d_u.d_child); |
c83ce989 TM |
470 | /* |
471 | * Inform try_to_ascend() that we are no longer attached to the | |
472 | * dentry tree | |
473 | */ | |
b161dfa6 | 474 | dentry->d_flags |= DCACHE_DENTRY_KILLED; |
2fd6b7f5 NP |
475 | if (parent) |
476 | spin_unlock(&parent->d_lock); | |
d52b9086 | 477 | dentry_iput(dentry); |
b7ab39f6 NP |
478 | /* |
479 | * dentry_iput drops the locks, at which point nobody (except | |
480 | * transient RCU lookups) can reach this dentry. | |
481 | */ | |
d52b9086 | 482 | d_free(dentry); |
871c0067 | 483 | return parent; |
d52b9086 MS |
484 | } |
485 | ||
c6627c60 DH |
486 | /* |
487 | * Unhash a dentry without inserting an RCU walk barrier or checking that | |
488 | * dentry->d_lock is locked. The caller must take care of that, if | |
489 | * appropriate. | |
490 | */ | |
491 | static void __d_shrink(struct dentry *dentry) | |
492 | { | |
493 | if (!d_unhashed(dentry)) { | |
494 | struct hlist_bl_head *b; | |
495 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
496 | b = &dentry->d_sb->s_anon; | |
497 | else | |
498 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
499 | ||
500 | hlist_bl_lock(b); | |
501 | __hlist_bl_del(&dentry->d_hash); | |
502 | dentry->d_hash.pprev = NULL; | |
503 | hlist_bl_unlock(b); | |
504 | } | |
505 | } | |
506 | ||
789680d1 NP |
507 | /** |
508 | * d_drop - drop a dentry | |
509 | * @dentry: dentry to drop | |
510 | * | |
511 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
512 | * be found through a VFS lookup any more. Note that this is different from | |
513 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
514 | * possible, giving a successful _negative_ lookup, while d_drop will | |
515 | * just make the cache lookup fail. | |
516 | * | |
517 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
518 | * reason (NFS timeouts or autofs deletes). | |
519 | * | |
520 | * __d_drop requires dentry->d_lock. | |
521 | */ | |
522 | void __d_drop(struct dentry *dentry) | |
523 | { | |
dea3667b | 524 | if (!d_unhashed(dentry)) { |
c6627c60 | 525 | __d_shrink(dentry); |
dea3667b | 526 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
527 | } |
528 | } | |
529 | EXPORT_SYMBOL(__d_drop); | |
530 | ||
531 | void d_drop(struct dentry *dentry) | |
532 | { | |
789680d1 NP |
533 | spin_lock(&dentry->d_lock); |
534 | __d_drop(dentry); | |
535 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
536 | } |
537 | EXPORT_SYMBOL(d_drop); | |
538 | ||
77812a1e NP |
539 | /* |
540 | * Finish off a dentry we've decided to kill. | |
541 | * dentry->d_lock must be held, returns with it unlocked. | |
542 | * If ref is non-zero, then decrement the refcount too. | |
543 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
544 | */ | |
dd1f6b2e DC |
545 | static inline struct dentry * |
546 | dentry_kill(struct dentry *dentry, int unlock_on_failure) | |
77812a1e NP |
547 | __releases(dentry->d_lock) |
548 | { | |
873feea0 | 549 | struct inode *inode; |
77812a1e NP |
550 | struct dentry *parent; |
551 | ||
873feea0 NP |
552 | inode = dentry->d_inode; |
553 | if (inode && !spin_trylock(&inode->i_lock)) { | |
77812a1e | 554 | relock: |
dd1f6b2e DC |
555 | if (unlock_on_failure) { |
556 | spin_unlock(&dentry->d_lock); | |
557 | cpu_relax(); | |
558 | } | |
77812a1e NP |
559 | return dentry; /* try again with same dentry */ |
560 | } | |
561 | if (IS_ROOT(dentry)) | |
562 | parent = NULL; | |
563 | else | |
564 | parent = dentry->d_parent; | |
565 | if (parent && !spin_trylock(&parent->d_lock)) { | |
873feea0 NP |
566 | if (inode) |
567 | spin_unlock(&inode->i_lock); | |
77812a1e NP |
568 | goto relock; |
569 | } | |
31e6b01f | 570 | |
0d98439e LT |
571 | /* |
572 | * The dentry is now unrecoverably dead to the world. | |
573 | */ | |
574 | lockref_mark_dead(&dentry->d_lockref); | |
575 | ||
f0023bc6 | 576 | /* |
f0023bc6 SW |
577 | * inform the fs via d_prune that this dentry is about to be |
578 | * unhashed and destroyed. | |
579 | */ | |
590fb51f | 580 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry)) |
61572bb1 YZ |
581 | dentry->d_op->d_prune(dentry); |
582 | ||
583 | dentry_lru_del(dentry); | |
77812a1e NP |
584 | /* if it was on the hash then remove it */ |
585 | __d_drop(dentry); | |
586 | return d_kill(dentry, parent); | |
587 | } | |
588 | ||
1da177e4 LT |
589 | /* |
590 | * This is dput | |
591 | * | |
592 | * This is complicated by the fact that we do not want to put | |
593 | * dentries that are no longer on any hash chain on the unused | |
594 | * list: we'd much rather just get rid of them immediately. | |
595 | * | |
596 | * However, that implies that we have to traverse the dentry | |
597 | * tree upwards to the parents which might _also_ now be | |
598 | * scheduled for deletion (it may have been only waiting for | |
599 | * its last child to go away). | |
600 | * | |
601 | * This tail recursion is done by hand as we don't want to depend | |
602 | * on the compiler to always get this right (gcc generally doesn't). | |
603 | * Real recursion would eat up our stack space. | |
604 | */ | |
605 | ||
606 | /* | |
607 | * dput - release a dentry | |
608 | * @dentry: dentry to release | |
609 | * | |
610 | * Release a dentry. This will drop the usage count and if appropriate | |
611 | * call the dentry unlink method as well as removing it from the queues and | |
612 | * releasing its resources. If the parent dentries were scheduled for release | |
613 | * they too may now get deleted. | |
1da177e4 | 614 | */ |
1da177e4 LT |
615 | void dput(struct dentry *dentry) |
616 | { | |
8aab6a27 | 617 | if (unlikely(!dentry)) |
1da177e4 LT |
618 | return; |
619 | ||
620 | repeat: | |
98474236 | 621 | if (lockref_put_or_lock(&dentry->d_lockref)) |
1da177e4 | 622 | return; |
1da177e4 | 623 | |
8aab6a27 LT |
624 | /* Unreachable? Get rid of it */ |
625 | if (unlikely(d_unhashed(dentry))) | |
626 | goto kill_it; | |
627 | ||
628 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
1da177e4 | 629 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 630 | goto kill_it; |
1da177e4 | 631 | } |
265ac902 | 632 | |
39e3c955 | 633 | dentry->d_flags |= DCACHE_REFERENCED; |
a4633357 | 634 | dentry_lru_add(dentry); |
265ac902 | 635 | |
98474236 | 636 | dentry->d_lockref.count--; |
61f3dee4 | 637 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
638 | return; |
639 | ||
d52b9086 | 640 | kill_it: |
dd1f6b2e | 641 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
642 | if (dentry) |
643 | goto repeat; | |
1da177e4 | 644 | } |
ec4f8605 | 645 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
646 | |
647 | /** | |
648 | * d_invalidate - invalidate a dentry | |
649 | * @dentry: dentry to invalidate | |
650 | * | |
651 | * Try to invalidate the dentry if it turns out to be | |
652 | * possible. If there are other dentries that can be | |
653 | * reached through this one we can't delete it and we | |
654 | * return -EBUSY. On success we return 0. | |
655 | * | |
656 | * no dcache lock. | |
657 | */ | |
658 | ||
659 | int d_invalidate(struct dentry * dentry) | |
660 | { | |
661 | /* | |
662 | * If it's already been dropped, return OK. | |
663 | */ | |
da502956 | 664 | spin_lock(&dentry->d_lock); |
1da177e4 | 665 | if (d_unhashed(dentry)) { |
da502956 | 666 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
667 | return 0; |
668 | } | |
669 | /* | |
670 | * Check whether to do a partial shrink_dcache | |
671 | * to get rid of unused child entries. | |
672 | */ | |
673 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 674 | spin_unlock(&dentry->d_lock); |
1da177e4 | 675 | shrink_dcache_parent(dentry); |
da502956 | 676 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
677 | } |
678 | ||
679 | /* | |
680 | * Somebody else still using it? | |
681 | * | |
682 | * If it's a directory, we can't drop it | |
683 | * for fear of somebody re-populating it | |
684 | * with children (even though dropping it | |
685 | * would make it unreachable from the root, | |
686 | * we might still populate it if it was a | |
687 | * working directory or similar). | |
50e69630 AV |
688 | * We also need to leave mountpoints alone, |
689 | * directory or not. | |
1da177e4 | 690 | */ |
98474236 | 691 | if (dentry->d_lockref.count > 1 && dentry->d_inode) { |
50e69630 | 692 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { |
1da177e4 | 693 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
694 | return -EBUSY; |
695 | } | |
696 | } | |
697 | ||
698 | __d_drop(dentry); | |
699 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
700 | return 0; |
701 | } | |
ec4f8605 | 702 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 703 | |
b5c84bf6 | 704 | /* This must be called with d_lock held */ |
dc0474be | 705 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 706 | { |
98474236 | 707 | dentry->d_lockref.count++; |
23044507 NP |
708 | } |
709 | ||
dc0474be | 710 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 711 | { |
98474236 | 712 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
713 | } |
714 | ||
b7ab39f6 NP |
715 | struct dentry *dget_parent(struct dentry *dentry) |
716 | { | |
df3d0bbc | 717 | int gotref; |
b7ab39f6 NP |
718 | struct dentry *ret; |
719 | ||
df3d0bbc WL |
720 | /* |
721 | * Do optimistic parent lookup without any | |
722 | * locking. | |
723 | */ | |
724 | rcu_read_lock(); | |
725 | ret = ACCESS_ONCE(dentry->d_parent); | |
726 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
727 | rcu_read_unlock(); | |
728 | if (likely(gotref)) { | |
729 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
730 | return ret; | |
731 | dput(ret); | |
732 | } | |
733 | ||
b7ab39f6 | 734 | repeat: |
a734eb45 NP |
735 | /* |
736 | * Don't need rcu_dereference because we re-check it was correct under | |
737 | * the lock. | |
738 | */ | |
739 | rcu_read_lock(); | |
b7ab39f6 | 740 | ret = dentry->d_parent; |
a734eb45 NP |
741 | spin_lock(&ret->d_lock); |
742 | if (unlikely(ret != dentry->d_parent)) { | |
743 | spin_unlock(&ret->d_lock); | |
744 | rcu_read_unlock(); | |
b7ab39f6 NP |
745 | goto repeat; |
746 | } | |
a734eb45 | 747 | rcu_read_unlock(); |
98474236 WL |
748 | BUG_ON(!ret->d_lockref.count); |
749 | ret->d_lockref.count++; | |
b7ab39f6 | 750 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
751 | return ret; |
752 | } | |
753 | EXPORT_SYMBOL(dget_parent); | |
754 | ||
1da177e4 LT |
755 | /** |
756 | * d_find_alias - grab a hashed alias of inode | |
757 | * @inode: inode in question | |
32ba9c3f LT |
758 | * @want_discon: flag, used by d_splice_alias, to request |
759 | * that only a DISCONNECTED alias be returned. | |
1da177e4 LT |
760 | * |
761 | * If inode has a hashed alias, or is a directory and has any alias, | |
762 | * acquire the reference to alias and return it. Otherwise return NULL. | |
763 | * Notice that if inode is a directory there can be only one alias and | |
764 | * it can be unhashed only if it has no children, or if it is the root | |
765 | * of a filesystem. | |
766 | * | |
21c0d8fd | 767 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
32ba9c3f LT |
768 | * any other hashed alias over that one unless @want_discon is set, |
769 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. | |
1da177e4 | 770 | */ |
32ba9c3f | 771 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 772 | { |
da502956 | 773 | struct dentry *alias, *discon_alias; |
1da177e4 | 774 | |
da502956 NP |
775 | again: |
776 | discon_alias = NULL; | |
b67bfe0d | 777 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
da502956 | 778 | spin_lock(&alias->d_lock); |
1da177e4 | 779 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 780 | if (IS_ROOT(alias) && |
da502956 | 781 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 782 | discon_alias = alias; |
32ba9c3f | 783 | } else if (!want_discon) { |
dc0474be | 784 | __dget_dlock(alias); |
da502956 NP |
785 | spin_unlock(&alias->d_lock); |
786 | return alias; | |
787 | } | |
788 | } | |
789 | spin_unlock(&alias->d_lock); | |
790 | } | |
791 | if (discon_alias) { | |
792 | alias = discon_alias; | |
793 | spin_lock(&alias->d_lock); | |
794 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
795 | if (IS_ROOT(alias) && | |
796 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 797 | __dget_dlock(alias); |
da502956 | 798 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
799 | return alias; |
800 | } | |
801 | } | |
da502956 NP |
802 | spin_unlock(&alias->d_lock); |
803 | goto again; | |
1da177e4 | 804 | } |
da502956 | 805 | return NULL; |
1da177e4 LT |
806 | } |
807 | ||
da502956 | 808 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 809 | { |
214fda1f DH |
810 | struct dentry *de = NULL; |
811 | ||
b3d9b7a3 | 812 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 813 | spin_lock(&inode->i_lock); |
32ba9c3f | 814 | de = __d_find_alias(inode, 0); |
873feea0 | 815 | spin_unlock(&inode->i_lock); |
214fda1f | 816 | } |
1da177e4 LT |
817 | return de; |
818 | } | |
ec4f8605 | 819 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
820 | |
821 | /* | |
822 | * Try to kill dentries associated with this inode. | |
823 | * WARNING: you must own a reference to inode. | |
824 | */ | |
825 | void d_prune_aliases(struct inode *inode) | |
826 | { | |
0cdca3f9 | 827 | struct dentry *dentry; |
1da177e4 | 828 | restart: |
873feea0 | 829 | spin_lock(&inode->i_lock); |
b67bfe0d | 830 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 831 | spin_lock(&dentry->d_lock); |
98474236 | 832 | if (!dentry->d_lockref.count) { |
590fb51f YZ |
833 | /* |
834 | * inform the fs via d_prune that this dentry | |
835 | * is about to be unhashed and destroyed. | |
836 | */ | |
837 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && | |
838 | !d_unhashed(dentry)) | |
839 | dentry->d_op->d_prune(dentry); | |
840 | ||
dc0474be | 841 | __dget_dlock(dentry); |
1da177e4 LT |
842 | __d_drop(dentry); |
843 | spin_unlock(&dentry->d_lock); | |
873feea0 | 844 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
845 | dput(dentry); |
846 | goto restart; | |
847 | } | |
848 | spin_unlock(&dentry->d_lock); | |
849 | } | |
873feea0 | 850 | spin_unlock(&inode->i_lock); |
1da177e4 | 851 | } |
ec4f8605 | 852 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
853 | |
854 | /* | |
77812a1e NP |
855 | * Try to throw away a dentry - free the inode, dput the parent. |
856 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
857 | * Releases dentry->d_lock. | |
d702ccb3 | 858 | * |
77812a1e | 859 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 860 | */ |
dd1f6b2e | 861 | static struct dentry * try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 862 | __releases(dentry->d_lock) |
1da177e4 | 863 | { |
77812a1e | 864 | struct dentry *parent; |
d52b9086 | 865 | |
dd1f6b2e | 866 | parent = dentry_kill(dentry, 0); |
d52b9086 | 867 | /* |
77812a1e NP |
868 | * If dentry_kill returns NULL, we have nothing more to do. |
869 | * if it returns the same dentry, trylocks failed. In either | |
870 | * case, just loop again. | |
871 | * | |
872 | * Otherwise, we need to prune ancestors too. This is necessary | |
873 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
874 | * is also expected to be beneficial in reducing dentry cache | |
875 | * fragmentation. | |
d52b9086 | 876 | */ |
77812a1e | 877 | if (!parent) |
dd1f6b2e | 878 | return NULL; |
77812a1e | 879 | if (parent == dentry) |
dd1f6b2e | 880 | return dentry; |
77812a1e NP |
881 | |
882 | /* Prune ancestors. */ | |
883 | dentry = parent; | |
d52b9086 | 884 | while (dentry) { |
98474236 | 885 | if (lockref_put_or_lock(&dentry->d_lockref)) |
dd1f6b2e DC |
886 | return NULL; |
887 | dentry = dentry_kill(dentry, 1); | |
d52b9086 | 888 | } |
dd1f6b2e | 889 | return NULL; |
1da177e4 LT |
890 | } |
891 | ||
3049cfe2 | 892 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 893 | { |
da3bbdd4 | 894 | struct dentry *dentry; |
da3bbdd4 | 895 | |
ec33679d NP |
896 | rcu_read_lock(); |
897 | for (;;) { | |
ec33679d NP |
898 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
899 | if (&dentry->d_lru == list) | |
900 | break; /* empty */ | |
89dc77bc LT |
901 | |
902 | /* | |
903 | * Get the dentry lock, and re-verify that the dentry is | |
904 | * this on the shrinking list. If it is, we know that | |
905 | * DCACHE_SHRINK_LIST and DCACHE_LRU_LIST are set. | |
906 | */ | |
ec33679d NP |
907 | spin_lock(&dentry->d_lock); |
908 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
909 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
910 | continue; |
911 | } | |
912 | ||
dd1f6b2e DC |
913 | /* |
914 | * The dispose list is isolated and dentries are not accounted | |
915 | * to the LRU here, so we can simply remove it from the list | |
916 | * here regardless of whether it is referenced or not. | |
917 | */ | |
89dc77bc | 918 | d_shrink_del(dentry); |
dd1f6b2e | 919 | |
1da177e4 LT |
920 | /* |
921 | * We found an inuse dentry which was not removed from | |
dd1f6b2e | 922 | * the LRU because of laziness during lookup. Do not free it. |
1da177e4 | 923 | */ |
98474236 | 924 | if (dentry->d_lockref.count) { |
da3bbdd4 | 925 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
926 | continue; |
927 | } | |
ec33679d | 928 | rcu_read_unlock(); |
77812a1e | 929 | |
89dc77bc LT |
930 | /* |
931 | * If 'try_to_prune()' returns a dentry, it will | |
932 | * be the same one we passed in, and d_lock will | |
933 | * have been held the whole time, so it will not | |
934 | * have been added to any other lists. We failed | |
935 | * to get the inode lock. | |
936 | * | |
937 | * We just add it back to the shrink list. | |
938 | */ | |
dd1f6b2e | 939 | dentry = try_prune_one_dentry(dentry); |
77812a1e | 940 | |
ec33679d | 941 | rcu_read_lock(); |
dd1f6b2e | 942 | if (dentry) { |
89dc77bc | 943 | d_shrink_add(dentry, list); |
dd1f6b2e DC |
944 | spin_unlock(&dentry->d_lock); |
945 | } | |
da3bbdd4 | 946 | } |
ec33679d | 947 | rcu_read_unlock(); |
3049cfe2 CH |
948 | } |
949 | ||
f6041567 DC |
950 | static enum lru_status |
951 | dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg) | |
952 | { | |
953 | struct list_head *freeable = arg; | |
954 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
955 | ||
956 | ||
957 | /* | |
958 | * we are inverting the lru lock/dentry->d_lock here, | |
959 | * so use a trylock. If we fail to get the lock, just skip | |
960 | * it | |
961 | */ | |
962 | if (!spin_trylock(&dentry->d_lock)) | |
963 | return LRU_SKIP; | |
964 | ||
965 | /* | |
966 | * Referenced dentries are still in use. If they have active | |
967 | * counts, just remove them from the LRU. Otherwise give them | |
968 | * another pass through the LRU. | |
969 | */ | |
970 | if (dentry->d_lockref.count) { | |
89dc77bc | 971 | d_lru_isolate(dentry); |
f6041567 DC |
972 | spin_unlock(&dentry->d_lock); |
973 | return LRU_REMOVED; | |
974 | } | |
975 | ||
976 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
977 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
978 | spin_unlock(&dentry->d_lock); | |
979 | ||
980 | /* | |
981 | * The list move itself will be made by the common LRU code. At | |
982 | * this point, we've dropped the dentry->d_lock but keep the | |
983 | * lru lock. This is safe to do, since every list movement is | |
984 | * protected by the lru lock even if both locks are held. | |
985 | * | |
986 | * This is guaranteed by the fact that all LRU management | |
987 | * functions are intermediated by the LRU API calls like | |
988 | * list_lru_add and list_lru_del. List movement in this file | |
989 | * only ever occur through this functions or through callbacks | |
990 | * like this one, that are called from the LRU API. | |
991 | * | |
992 | * The only exceptions to this are functions like | |
993 | * shrink_dentry_list, and code that first checks for the | |
994 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
995 | * operating only with stack provided lists after they are | |
996 | * properly isolated from the main list. It is thus, always a | |
997 | * local access. | |
998 | */ | |
999 | return LRU_ROTATE; | |
1000 | } | |
1001 | ||
89dc77bc | 1002 | d_lru_shrink_move(dentry, freeable); |
f6041567 DC |
1003 | spin_unlock(&dentry->d_lock); |
1004 | ||
1005 | return LRU_REMOVED; | |
1006 | } | |
1007 | ||
3049cfe2 | 1008 | /** |
b48f03b3 DC |
1009 | * prune_dcache_sb - shrink the dcache |
1010 | * @sb: superblock | |
f6041567 | 1011 | * @nr_to_scan : number of entries to try to free |
9b17c623 | 1012 | * @nid: which node to scan for freeable entities |
b48f03b3 | 1013 | * |
f6041567 | 1014 | * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is |
b48f03b3 DC |
1015 | * done when we need more memory an called from the superblock shrinker |
1016 | * function. | |
3049cfe2 | 1017 | * |
b48f03b3 DC |
1018 | * This function may fail to free any resources if all the dentries are in |
1019 | * use. | |
3049cfe2 | 1020 | */ |
9b17c623 DC |
1021 | long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan, |
1022 | int nid) | |
3049cfe2 | 1023 | { |
f6041567 DC |
1024 | LIST_HEAD(dispose); |
1025 | long freed; | |
3049cfe2 | 1026 | |
9b17c623 DC |
1027 | freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate, |
1028 | &dispose, &nr_to_scan); | |
f6041567 | 1029 | shrink_dentry_list(&dispose); |
0a234c6d | 1030 | return freed; |
da3bbdd4 | 1031 | } |
23044507 | 1032 | |
4e717f5c GC |
1033 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
1034 | spinlock_t *lru_lock, void *arg) | |
dd1f6b2e | 1035 | { |
4e717f5c GC |
1036 | struct list_head *freeable = arg; |
1037 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 1038 | |
4e717f5c GC |
1039 | /* |
1040 | * we are inverting the lru lock/dentry->d_lock here, | |
1041 | * so use a trylock. If we fail to get the lock, just skip | |
1042 | * it | |
1043 | */ | |
1044 | if (!spin_trylock(&dentry->d_lock)) | |
1045 | return LRU_SKIP; | |
1046 | ||
89dc77bc | 1047 | d_lru_shrink_move(dentry, freeable); |
4e717f5c | 1048 | spin_unlock(&dentry->d_lock); |
ec33679d | 1049 | |
4e717f5c | 1050 | return LRU_REMOVED; |
da3bbdd4 KM |
1051 | } |
1052 | ||
4e717f5c | 1053 | |
1da177e4 LT |
1054 | /** |
1055 | * shrink_dcache_sb - shrink dcache for a superblock | |
1056 | * @sb: superblock | |
1057 | * | |
3049cfe2 CH |
1058 | * Shrink the dcache for the specified super block. This is used to free |
1059 | * the dcache before unmounting a file system. | |
1da177e4 | 1060 | */ |
3049cfe2 | 1061 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1062 | { |
4e717f5c GC |
1063 | long freed; |
1064 | ||
1065 | do { | |
1066 | LIST_HEAD(dispose); | |
1067 | ||
1068 | freed = list_lru_walk(&sb->s_dentry_lru, | |
1069 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); | |
3049cfe2 | 1070 | |
4e717f5c GC |
1071 | this_cpu_sub(nr_dentry_unused, freed); |
1072 | shrink_dentry_list(&dispose); | |
1073 | } while (freed > 0); | |
1da177e4 | 1074 | } |
ec4f8605 | 1075 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1076 | |
c826cb7d LT |
1077 | /* |
1078 | * This tries to ascend one level of parenthood, but | |
1079 | * we can race with renaming, so we need to re-check | |
1080 | * the parenthood after dropping the lock and check | |
1081 | * that the sequence number still matches. | |
1082 | */ | |
48f5ec21 | 1083 | static struct dentry *try_to_ascend(struct dentry *old, unsigned seq) |
c826cb7d LT |
1084 | { |
1085 | struct dentry *new = old->d_parent; | |
1086 | ||
1087 | rcu_read_lock(); | |
1088 | spin_unlock(&old->d_lock); | |
1089 | spin_lock(&new->d_lock); | |
1090 | ||
1091 | /* | |
1092 | * might go back up the wrong parent if we have had a rename | |
1093 | * or deletion | |
1094 | */ | |
1095 | if (new != old->d_parent || | |
b161dfa6 | 1096 | (old->d_flags & DCACHE_DENTRY_KILLED) || |
48f5ec21 | 1097 | need_seqretry(&rename_lock, seq)) { |
c826cb7d LT |
1098 | spin_unlock(&new->d_lock); |
1099 | new = NULL; | |
1100 | } | |
1101 | rcu_read_unlock(); | |
1102 | return new; | |
1103 | } | |
1104 | ||
db14fc3a MS |
1105 | /** |
1106 | * enum d_walk_ret - action to talke during tree walk | |
1107 | * @D_WALK_CONTINUE: contrinue walk | |
1108 | * @D_WALK_QUIT: quit walk | |
1109 | * @D_WALK_NORETRY: quit when retry is needed | |
1110 | * @D_WALK_SKIP: skip this dentry and its children | |
1111 | */ | |
1112 | enum d_walk_ret { | |
1113 | D_WALK_CONTINUE, | |
1114 | D_WALK_QUIT, | |
1115 | D_WALK_NORETRY, | |
1116 | D_WALK_SKIP, | |
1117 | }; | |
c826cb7d | 1118 | |
1da177e4 | 1119 | /** |
db14fc3a MS |
1120 | * d_walk - walk the dentry tree |
1121 | * @parent: start of walk | |
1122 | * @data: data passed to @enter() and @finish() | |
1123 | * @enter: callback when first entering the dentry | |
1124 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1125 | * |
db14fc3a | 1126 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1127 | */ |
db14fc3a MS |
1128 | static void d_walk(struct dentry *parent, void *data, |
1129 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1130 | void (*finish)(void *)) | |
1da177e4 | 1131 | { |
949854d0 | 1132 | struct dentry *this_parent; |
1da177e4 | 1133 | struct list_head *next; |
48f5ec21 | 1134 | unsigned seq = 0; |
db14fc3a MS |
1135 | enum d_walk_ret ret; |
1136 | bool retry = true; | |
949854d0 | 1137 | |
58db63d0 | 1138 | again: |
48f5ec21 | 1139 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1140 | this_parent = parent; |
2fd6b7f5 | 1141 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1142 | |
1143 | ret = enter(data, this_parent); | |
1144 | switch (ret) { | |
1145 | case D_WALK_CONTINUE: | |
1146 | break; | |
1147 | case D_WALK_QUIT: | |
1148 | case D_WALK_SKIP: | |
1149 | goto out_unlock; | |
1150 | case D_WALK_NORETRY: | |
1151 | retry = false; | |
1152 | break; | |
1153 | } | |
1da177e4 LT |
1154 | repeat: |
1155 | next = this_parent->d_subdirs.next; | |
1156 | resume: | |
1157 | while (next != &this_parent->d_subdirs) { | |
1158 | struct list_head *tmp = next; | |
5160ee6f | 1159 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1160 | next = tmp->next; |
2fd6b7f5 NP |
1161 | |
1162 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
db14fc3a MS |
1163 | |
1164 | ret = enter(data, dentry); | |
1165 | switch (ret) { | |
1166 | case D_WALK_CONTINUE: | |
1167 | break; | |
1168 | case D_WALK_QUIT: | |
2fd6b7f5 | 1169 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1170 | goto out_unlock; |
1171 | case D_WALK_NORETRY: | |
1172 | retry = false; | |
1173 | break; | |
1174 | case D_WALK_SKIP: | |
1175 | spin_unlock(&dentry->d_lock); | |
1176 | continue; | |
2fd6b7f5 | 1177 | } |
db14fc3a | 1178 | |
1da177e4 | 1179 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1180 | spin_unlock(&this_parent->d_lock); |
1181 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1182 | this_parent = dentry; |
2fd6b7f5 | 1183 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1184 | goto repeat; |
1185 | } | |
2fd6b7f5 | 1186 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1187 | } |
1188 | /* | |
1189 | * All done at this level ... ascend and resume the search. | |
1190 | */ | |
1191 | if (this_parent != parent) { | |
c826cb7d | 1192 | struct dentry *child = this_parent; |
48f5ec21 | 1193 | this_parent = try_to_ascend(this_parent, seq); |
c826cb7d | 1194 | if (!this_parent) |
949854d0 | 1195 | goto rename_retry; |
949854d0 | 1196 | next = child->d_u.d_child.next; |
1da177e4 LT |
1197 | goto resume; |
1198 | } | |
48f5ec21 | 1199 | if (need_seqretry(&rename_lock, seq)) { |
db14fc3a | 1200 | spin_unlock(&this_parent->d_lock); |
949854d0 | 1201 | goto rename_retry; |
db14fc3a MS |
1202 | } |
1203 | if (finish) | |
1204 | finish(data); | |
1205 | ||
1206 | out_unlock: | |
1207 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1208 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1209 | return; |
58db63d0 NP |
1210 | |
1211 | rename_retry: | |
db14fc3a MS |
1212 | if (!retry) |
1213 | return; | |
48f5ec21 | 1214 | seq = 1; |
58db63d0 | 1215 | goto again; |
1da177e4 | 1216 | } |
db14fc3a MS |
1217 | |
1218 | /* | |
1219 | * Search for at least 1 mount point in the dentry's subdirs. | |
1220 | * We descend to the next level whenever the d_subdirs | |
1221 | * list is non-empty and continue searching. | |
1222 | */ | |
1223 | ||
1224 | /** | |
1225 | * have_submounts - check for mounts over a dentry | |
1226 | * @parent: dentry to check. | |
1227 | * | |
1228 | * Return true if the parent or its subdirectories contain | |
1229 | * a mount point | |
1230 | */ | |
1231 | ||
1232 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) | |
1233 | { | |
1234 | int *ret = data; | |
1235 | if (d_mountpoint(dentry)) { | |
1236 | *ret = 1; | |
1237 | return D_WALK_QUIT; | |
1238 | } | |
1239 | return D_WALK_CONTINUE; | |
1240 | } | |
1241 | ||
1242 | int have_submounts(struct dentry *parent) | |
1243 | { | |
1244 | int ret = 0; | |
1245 | ||
1246 | d_walk(parent, &ret, check_mount, NULL); | |
1247 | ||
1248 | return ret; | |
1249 | } | |
ec4f8605 | 1250 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1251 | |
eed81007 MS |
1252 | /* |
1253 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1254 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1255 | * subtree can become unreachable). | |
1256 | * | |
1257 | * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For | |
1258 | * this reason take rename_lock and d_lock on dentry and ancestors. | |
1259 | */ | |
1260 | int d_set_mounted(struct dentry *dentry) | |
1261 | { | |
1262 | struct dentry *p; | |
1263 | int ret = -ENOENT; | |
1264 | write_seqlock(&rename_lock); | |
1265 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1266 | /* Need exclusion wrt. check_submounts_and_drop() */ | |
1267 | spin_lock(&p->d_lock); | |
1268 | if (unlikely(d_unhashed(p))) { | |
1269 | spin_unlock(&p->d_lock); | |
1270 | goto out; | |
1271 | } | |
1272 | spin_unlock(&p->d_lock); | |
1273 | } | |
1274 | spin_lock(&dentry->d_lock); | |
1275 | if (!d_unlinked(dentry)) { | |
1276 | dentry->d_flags |= DCACHE_MOUNTED; | |
1277 | ret = 0; | |
1278 | } | |
1279 | spin_unlock(&dentry->d_lock); | |
1280 | out: | |
1281 | write_sequnlock(&rename_lock); | |
1282 | return ret; | |
1283 | } | |
1284 | ||
1da177e4 | 1285 | /* |
fd517909 | 1286 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1287 | * and move any unused dentries to the end of the unused |
1288 | * list for prune_dcache(). We descend to the next level | |
1289 | * whenever the d_subdirs list is non-empty and continue | |
1290 | * searching. | |
1291 | * | |
1292 | * It returns zero iff there are no unused children, | |
1293 | * otherwise it returns the number of children moved to | |
1294 | * the end of the unused list. This may not be the total | |
1295 | * number of unused children, because select_parent can | |
1296 | * drop the lock and return early due to latency | |
1297 | * constraints. | |
1298 | */ | |
1da177e4 | 1299 | |
db14fc3a MS |
1300 | struct select_data { |
1301 | struct dentry *start; | |
1302 | struct list_head dispose; | |
1303 | int found; | |
1304 | }; | |
23044507 | 1305 | |
db14fc3a MS |
1306 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1307 | { | |
1308 | struct select_data *data = _data; | |
1309 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1310 | |
db14fc3a MS |
1311 | if (data->start == dentry) |
1312 | goto out; | |
2fd6b7f5 | 1313 | |
1da177e4 | 1314 | /* |
db14fc3a MS |
1315 | * move only zero ref count dentries to the dispose list. |
1316 | * | |
1317 | * Those which are presently on the shrink list, being processed | |
1318 | * by shrink_dentry_list(), shouldn't be moved. Otherwise the | |
1319 | * loop in shrink_dcache_parent() might not make any progress | |
1320 | * and loop forever. | |
1da177e4 | 1321 | */ |
db14fc3a MS |
1322 | if (dentry->d_lockref.count) { |
1323 | dentry_lru_del(dentry); | |
1324 | } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { | |
89dc77bc LT |
1325 | /* |
1326 | * We can't use d_lru_shrink_move() because we | |
1327 | * need to get the global LRU lock and do the | |
05a8252b | 1328 | * LRU accounting. |
89dc77bc LT |
1329 | */ |
1330 | d_lru_del(dentry); | |
1331 | d_shrink_add(dentry, &data->dispose); | |
db14fc3a MS |
1332 | data->found++; |
1333 | ret = D_WALK_NORETRY; | |
1da177e4 | 1334 | } |
db14fc3a MS |
1335 | /* |
1336 | * We can return to the caller if we have found some (this | |
1337 | * ensures forward progress). We'll be coming back to find | |
1338 | * the rest. | |
1339 | */ | |
1340 | if (data->found && need_resched()) | |
1341 | ret = D_WALK_QUIT; | |
1da177e4 | 1342 | out: |
db14fc3a | 1343 | return ret; |
1da177e4 LT |
1344 | } |
1345 | ||
1346 | /** | |
1347 | * shrink_dcache_parent - prune dcache | |
1348 | * @parent: parent of entries to prune | |
1349 | * | |
1350 | * Prune the dcache to remove unused children of the parent dentry. | |
1351 | */ | |
db14fc3a | 1352 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1353 | { |
db14fc3a MS |
1354 | for (;;) { |
1355 | struct select_data data; | |
1da177e4 | 1356 | |
db14fc3a MS |
1357 | INIT_LIST_HEAD(&data.dispose); |
1358 | data.start = parent; | |
1359 | data.found = 0; | |
1360 | ||
1361 | d_walk(parent, &data, select_collect, NULL); | |
1362 | if (!data.found) | |
1363 | break; | |
1364 | ||
1365 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1366 | cond_resched(); |
1367 | } | |
1da177e4 | 1368 | } |
ec4f8605 | 1369 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1370 | |
42c32608 AV |
1371 | static enum d_walk_ret umount_collect(void *_data, struct dentry *dentry) |
1372 | { | |
1373 | struct select_data *data = _data; | |
1374 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1375 | ||
1376 | if (dentry->d_lockref.count) { | |
1377 | dentry_lru_del(dentry); | |
1378 | if (likely(!list_empty(&dentry->d_subdirs))) | |
1379 | goto out; | |
1380 | if (dentry == data->start && dentry->d_lockref.count == 1) | |
1381 | goto out; | |
1382 | printk(KERN_ERR | |
1383 | "BUG: Dentry %p{i=%lx,n=%s}" | |
1384 | " still in use (%d)" | |
1385 | " [unmount of %s %s]\n", | |
1386 | dentry, | |
1387 | dentry->d_inode ? | |
1388 | dentry->d_inode->i_ino : 0UL, | |
1389 | dentry->d_name.name, | |
1390 | dentry->d_lockref.count, | |
1391 | dentry->d_sb->s_type->name, | |
1392 | dentry->d_sb->s_id); | |
1393 | BUG(); | |
1394 | } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { | |
1395 | /* | |
1396 | * We can't use d_lru_shrink_move() because we | |
1397 | * need to get the global LRU lock and do the | |
1398 | * LRU accounting. | |
1399 | */ | |
1400 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1401 | d_lru_del(dentry); | |
1402 | d_shrink_add(dentry, &data->dispose); | |
1403 | data->found++; | |
1404 | ret = D_WALK_NORETRY; | |
1405 | } | |
1406 | out: | |
1407 | if (data->found && need_resched()) | |
1408 | ret = D_WALK_QUIT; | |
1409 | return ret; | |
1410 | } | |
1411 | ||
1412 | /* | |
1413 | * destroy the dentries attached to a superblock on unmounting | |
1414 | */ | |
1415 | void shrink_dcache_for_umount(struct super_block *sb) | |
1416 | { | |
1417 | struct dentry *dentry; | |
1418 | ||
1419 | if (down_read_trylock(&sb->s_umount)) | |
1420 | BUG(); | |
1421 | ||
1422 | dentry = sb->s_root; | |
1423 | sb->s_root = NULL; | |
1424 | for (;;) { | |
1425 | struct select_data data; | |
1426 | ||
1427 | INIT_LIST_HEAD(&data.dispose); | |
1428 | data.start = dentry; | |
1429 | data.found = 0; | |
1430 | ||
1431 | d_walk(dentry, &data, umount_collect, NULL); | |
1432 | if (!data.found) | |
1433 | break; | |
1434 | ||
1435 | shrink_dentry_list(&data.dispose); | |
1436 | cond_resched(); | |
1437 | } | |
1438 | d_drop(dentry); | |
1439 | dput(dentry); | |
1440 | ||
1441 | while (!hlist_bl_empty(&sb->s_anon)) { | |
1442 | struct select_data data; | |
1443 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
1444 | ||
1445 | INIT_LIST_HEAD(&data.dispose); | |
1446 | data.start = NULL; | |
1447 | data.found = 0; | |
1448 | ||
1449 | d_walk(dentry, &data, umount_collect, NULL); | |
1450 | if (data.found) | |
1451 | shrink_dentry_list(&data.dispose); | |
1452 | cond_resched(); | |
1453 | } | |
1454 | } | |
1455 | ||
848ac114 MS |
1456 | static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry) |
1457 | { | |
1458 | struct select_data *data = _data; | |
1459 | ||
1460 | if (d_mountpoint(dentry)) { | |
1461 | data->found = -EBUSY; | |
1462 | return D_WALK_QUIT; | |
1463 | } | |
1464 | ||
1465 | return select_collect(_data, dentry); | |
1466 | } | |
1467 | ||
1468 | static void check_and_drop(void *_data) | |
1469 | { | |
1470 | struct select_data *data = _data; | |
1471 | ||
1472 | if (d_mountpoint(data->start)) | |
1473 | data->found = -EBUSY; | |
1474 | if (!data->found) | |
1475 | __d_drop(data->start); | |
1476 | } | |
1477 | ||
1478 | /** | |
1479 | * check_submounts_and_drop - prune dcache, check for submounts and drop | |
1480 | * | |
1481 | * All done as a single atomic operation relative to has_unlinked_ancestor(). | |
1482 | * Returns 0 if successfully unhashed @parent. If there were submounts then | |
1483 | * return -EBUSY. | |
1484 | * | |
1485 | * @dentry: dentry to prune and drop | |
1486 | */ | |
1487 | int check_submounts_and_drop(struct dentry *dentry) | |
1488 | { | |
1489 | int ret = 0; | |
1490 | ||
1491 | /* Negative dentries can be dropped without further checks */ | |
1492 | if (!dentry->d_inode) { | |
1493 | d_drop(dentry); | |
1494 | goto out; | |
1495 | } | |
1496 | ||
1497 | for (;;) { | |
1498 | struct select_data data; | |
1499 | ||
1500 | INIT_LIST_HEAD(&data.dispose); | |
1501 | data.start = dentry; | |
1502 | data.found = 0; | |
1503 | ||
1504 | d_walk(dentry, &data, check_and_collect, check_and_drop); | |
1505 | ret = data.found; | |
1506 | ||
1507 | if (!list_empty(&data.dispose)) | |
1508 | shrink_dentry_list(&data.dispose); | |
1509 | ||
1510 | if (ret <= 0) | |
1511 | break; | |
1512 | ||
1513 | cond_resched(); | |
1514 | } | |
1515 | ||
1516 | out: | |
1517 | return ret; | |
1518 | } | |
1519 | EXPORT_SYMBOL(check_submounts_and_drop); | |
1520 | ||
1da177e4 | 1521 | /** |
a4464dbc AV |
1522 | * __d_alloc - allocate a dcache entry |
1523 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1524 | * @name: qstr of the name |
1525 | * | |
1526 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1527 | * available. On a success the dentry is returned. The name passed in is | |
1528 | * copied and the copy passed in may be reused after this call. | |
1529 | */ | |
1530 | ||
a4464dbc | 1531 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1532 | { |
1533 | struct dentry *dentry; | |
1534 | char *dname; | |
1535 | ||
e12ba74d | 1536 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1537 | if (!dentry) |
1538 | return NULL; | |
1539 | ||
6326c71f LT |
1540 | /* |
1541 | * We guarantee that the inline name is always NUL-terminated. | |
1542 | * This way the memcpy() done by the name switching in rename | |
1543 | * will still always have a NUL at the end, even if we might | |
1544 | * be overwriting an internal NUL character | |
1545 | */ | |
1546 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 LT |
1547 | if (name->len > DNAME_INLINE_LEN-1) { |
1548 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1549 | if (!dname) { | |
1550 | kmem_cache_free(dentry_cache, dentry); | |
1551 | return NULL; | |
1552 | } | |
1553 | } else { | |
1554 | dname = dentry->d_iname; | |
1555 | } | |
1da177e4 LT |
1556 | |
1557 | dentry->d_name.len = name->len; | |
1558 | dentry->d_name.hash = name->hash; | |
1559 | memcpy(dname, name->name, name->len); | |
1560 | dname[name->len] = 0; | |
1561 | ||
6326c71f LT |
1562 | /* Make sure we always see the terminating NUL character */ |
1563 | smp_wmb(); | |
1564 | dentry->d_name.name = dname; | |
1565 | ||
98474236 | 1566 | dentry->d_lockref.count = 1; |
dea3667b | 1567 | dentry->d_flags = 0; |
1da177e4 | 1568 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1569 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1570 | dentry->d_inode = NULL; |
a4464dbc AV |
1571 | dentry->d_parent = dentry; |
1572 | dentry->d_sb = sb; | |
1da177e4 LT |
1573 | dentry->d_op = NULL; |
1574 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1575 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1576 | INIT_LIST_HEAD(&dentry->d_lru); |
1577 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
b3d9b7a3 | 1578 | INIT_HLIST_NODE(&dentry->d_alias); |
2fd6b7f5 | 1579 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1580 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1581 | |
3e880fb5 | 1582 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1583 | |
1da177e4 LT |
1584 | return dentry; |
1585 | } | |
a4464dbc AV |
1586 | |
1587 | /** | |
1588 | * d_alloc - allocate a dcache entry | |
1589 | * @parent: parent of entry to allocate | |
1590 | * @name: qstr of the name | |
1591 | * | |
1592 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1593 | * available. On a success the dentry is returned. The name passed in is | |
1594 | * copied and the copy passed in may be reused after this call. | |
1595 | */ | |
1596 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1597 | { | |
1598 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1599 | if (!dentry) | |
1600 | return NULL; | |
1601 | ||
1602 | spin_lock(&parent->d_lock); | |
1603 | /* | |
1604 | * don't need child lock because it is not subject | |
1605 | * to concurrency here | |
1606 | */ | |
1607 | __dget_dlock(parent); | |
1608 | dentry->d_parent = parent; | |
1609 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1610 | spin_unlock(&parent->d_lock); | |
1611 | ||
1612 | return dentry; | |
1613 | } | |
ec4f8605 | 1614 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1615 | |
4b936885 NP |
1616 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1617 | { | |
a4464dbc AV |
1618 | struct dentry *dentry = __d_alloc(sb, name); |
1619 | if (dentry) | |
4b936885 | 1620 | dentry->d_flags |= DCACHE_DISCONNECTED; |
4b936885 NP |
1621 | return dentry; |
1622 | } | |
1623 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1624 | ||
1da177e4 LT |
1625 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1626 | { | |
1627 | struct qstr q; | |
1628 | ||
1629 | q.name = name; | |
1630 | q.len = strlen(name); | |
1631 | q.hash = full_name_hash(q.name, q.len); | |
1632 | return d_alloc(parent, &q); | |
1633 | } | |
ef26ca97 | 1634 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1635 | |
fb045adb NP |
1636 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1637 | { | |
6f7f7caa LT |
1638 | WARN_ON_ONCE(dentry->d_op); |
1639 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1640 | DCACHE_OP_COMPARE | |
1641 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1642 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1643 | DCACHE_OP_DELETE )); |
1644 | dentry->d_op = op; | |
1645 | if (!op) | |
1646 | return; | |
1647 | if (op->d_hash) | |
1648 | dentry->d_flags |= DCACHE_OP_HASH; | |
1649 | if (op->d_compare) | |
1650 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1651 | if (op->d_revalidate) | |
1652 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1653 | if (op->d_weak_revalidate) |
1654 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1655 | if (op->d_delete) |
1656 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1657 | if (op->d_prune) |
1658 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1659 | |
1660 | } | |
1661 | EXPORT_SYMBOL(d_set_d_op); | |
1662 | ||
360da900 OH |
1663 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1664 | { | |
b23fb0a6 | 1665 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1666 | if (inode) { |
1667 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1668 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
b3d9b7a3 | 1669 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1670 | } |
360da900 | 1671 | dentry->d_inode = inode; |
31e6b01f | 1672 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1673 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1674 | fsnotify_d_instantiate(dentry, inode); |
1675 | } | |
1676 | ||
1da177e4 LT |
1677 | /** |
1678 | * d_instantiate - fill in inode information for a dentry | |
1679 | * @entry: dentry to complete | |
1680 | * @inode: inode to attach to this dentry | |
1681 | * | |
1682 | * Fill in inode information in the entry. | |
1683 | * | |
1684 | * This turns negative dentries into productive full members | |
1685 | * of society. | |
1686 | * | |
1687 | * NOTE! This assumes that the inode count has been incremented | |
1688 | * (or otherwise set) by the caller to indicate that it is now | |
1689 | * in use by the dcache. | |
1690 | */ | |
1691 | ||
1692 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1693 | { | |
b3d9b7a3 | 1694 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
873feea0 NP |
1695 | if (inode) |
1696 | spin_lock(&inode->i_lock); | |
360da900 | 1697 | __d_instantiate(entry, inode); |
873feea0 NP |
1698 | if (inode) |
1699 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1700 | security_d_instantiate(entry, inode); |
1701 | } | |
ec4f8605 | 1702 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1703 | |
1704 | /** | |
1705 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1706 | * @entry: dentry to instantiate | |
1707 | * @inode: inode to attach to this dentry | |
1708 | * | |
1709 | * Fill in inode information in the entry. On success, it returns NULL. | |
1710 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1711 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1712 | * |
1713 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1714 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1715 | * |
1716 | * This also assumes that the inode count has been incremented | |
1717 | * (or otherwise set) by the caller to indicate that it is now | |
1718 | * in use by the dcache. | |
1da177e4 | 1719 | */ |
770bfad8 DH |
1720 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1721 | struct inode *inode) | |
1da177e4 LT |
1722 | { |
1723 | struct dentry *alias; | |
1724 | int len = entry->d_name.len; | |
1725 | const char *name = entry->d_name.name; | |
1726 | unsigned int hash = entry->d_name.hash; | |
1727 | ||
770bfad8 | 1728 | if (!inode) { |
360da900 | 1729 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1730 | return NULL; |
1731 | } | |
1732 | ||
b67bfe0d | 1733 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
9abca360 NP |
1734 | /* |
1735 | * Don't need alias->d_lock here, because aliases with | |
1736 | * d_parent == entry->d_parent are not subject to name or | |
1737 | * parent changes, because the parent inode i_mutex is held. | |
1738 | */ | |
12f8ad4b | 1739 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1740 | continue; |
1741 | if (alias->d_parent != entry->d_parent) | |
1742 | continue; | |
ee983e89 LT |
1743 | if (alias->d_name.len != len) |
1744 | continue; | |
12f8ad4b | 1745 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1746 | continue; |
dc0474be | 1747 | __dget(alias); |
1da177e4 LT |
1748 | return alias; |
1749 | } | |
770bfad8 | 1750 | |
360da900 | 1751 | __d_instantiate(entry, inode); |
1da177e4 LT |
1752 | return NULL; |
1753 | } | |
770bfad8 DH |
1754 | |
1755 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1756 | { | |
1757 | struct dentry *result; | |
1758 | ||
b3d9b7a3 | 1759 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
770bfad8 | 1760 | |
873feea0 NP |
1761 | if (inode) |
1762 | spin_lock(&inode->i_lock); | |
770bfad8 | 1763 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1764 | if (inode) |
1765 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1766 | |
1767 | if (!result) { | |
1768 | security_d_instantiate(entry, inode); | |
1769 | return NULL; | |
1770 | } | |
1771 | ||
1772 | BUG_ON(!d_unhashed(result)); | |
1773 | iput(inode); | |
1774 | return result; | |
1775 | } | |
1776 | ||
1da177e4 LT |
1777 | EXPORT_SYMBOL(d_instantiate_unique); |
1778 | ||
b70a80e7 MS |
1779 | /** |
1780 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1781 | * @entry: dentry to complete | |
1782 | * @inode: inode to attach to this dentry | |
1783 | * | |
1784 | * Fill in inode information in the entry. If a directory alias is found, then | |
1785 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1786 | * guarantees that a directory inode may never have more than one alias. | |
1787 | */ | |
1788 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1789 | { | |
1790 | BUG_ON(!hlist_unhashed(&entry->d_alias)); | |
1791 | ||
1792 | spin_lock(&inode->i_lock); | |
1793 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1794 | spin_unlock(&inode->i_lock); | |
1795 | iput(inode); | |
1796 | return -EBUSY; | |
1797 | } | |
1798 | __d_instantiate(entry, inode); | |
1799 | spin_unlock(&inode->i_lock); | |
1800 | security_d_instantiate(entry, inode); | |
1801 | ||
1802 | return 0; | |
1803 | } | |
1804 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1805 | ||
adc0e91a AV |
1806 | struct dentry *d_make_root(struct inode *root_inode) |
1807 | { | |
1808 | struct dentry *res = NULL; | |
1809 | ||
1810 | if (root_inode) { | |
26fe5750 | 1811 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1812 | |
1813 | res = __d_alloc(root_inode->i_sb, &name); | |
1814 | if (res) | |
1815 | d_instantiate(res, root_inode); | |
1816 | else | |
1817 | iput(root_inode); | |
1818 | } | |
1819 | return res; | |
1820 | } | |
1821 | EXPORT_SYMBOL(d_make_root); | |
1822 | ||
d891eedb BF |
1823 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1824 | { | |
1825 | struct dentry *alias; | |
1826 | ||
b3d9b7a3 | 1827 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1828 | return NULL; |
b3d9b7a3 | 1829 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); |
d891eedb BF |
1830 | __dget(alias); |
1831 | return alias; | |
1832 | } | |
1833 | ||
46f72b34 SW |
1834 | /** |
1835 | * d_find_any_alias - find any alias for a given inode | |
1836 | * @inode: inode to find an alias for | |
1837 | * | |
1838 | * If any aliases exist for the given inode, take and return a | |
1839 | * reference for one of them. If no aliases exist, return %NULL. | |
1840 | */ | |
1841 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1842 | { |
1843 | struct dentry *de; | |
1844 | ||
1845 | spin_lock(&inode->i_lock); | |
1846 | de = __d_find_any_alias(inode); | |
1847 | spin_unlock(&inode->i_lock); | |
1848 | return de; | |
1849 | } | |
46f72b34 | 1850 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1851 | |
4ea3ada2 CH |
1852 | /** |
1853 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1854 | * @inode: inode to allocate the dentry for | |
1855 | * | |
1856 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1857 | * similar open by handle operations. The returned dentry may be anonymous, | |
1858 | * or may have a full name (if the inode was already in the cache). | |
1859 | * | |
1860 | * When called on a directory inode, we must ensure that the inode only ever | |
1861 | * has one dentry. If a dentry is found, that is returned instead of | |
1862 | * allocating a new one. | |
1863 | * | |
1864 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1865 | * to the dentry. In case of an error the reference on the inode is released. |
1866 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1867 | * be passed in and will be the error will be propagate to the return value, | |
1868 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1869 | */ |
1870 | struct dentry *d_obtain_alias(struct inode *inode) | |
1871 | { | |
b911a6bd | 1872 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1873 | struct dentry *tmp; |
1874 | struct dentry *res; | |
4ea3ada2 CH |
1875 | |
1876 | if (!inode) | |
44003728 | 1877 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1878 | if (IS_ERR(inode)) |
1879 | return ERR_CAST(inode); | |
1880 | ||
d891eedb | 1881 | res = d_find_any_alias(inode); |
9308a612 CH |
1882 | if (res) |
1883 | goto out_iput; | |
1884 | ||
a4464dbc | 1885 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1886 | if (!tmp) { |
1887 | res = ERR_PTR(-ENOMEM); | |
1888 | goto out_iput; | |
4ea3ada2 | 1889 | } |
b5c84bf6 | 1890 | |
873feea0 | 1891 | spin_lock(&inode->i_lock); |
d891eedb | 1892 | res = __d_find_any_alias(inode); |
9308a612 | 1893 | if (res) { |
873feea0 | 1894 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1895 | dput(tmp); |
1896 | goto out_iput; | |
1897 | } | |
1898 | ||
1899 | /* attach a disconnected dentry */ | |
1900 | spin_lock(&tmp->d_lock); | |
9308a612 CH |
1901 | tmp->d_inode = inode; |
1902 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
b3d9b7a3 | 1903 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1904 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1905 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1906 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1907 | spin_unlock(&tmp->d_lock); |
873feea0 | 1908 | spin_unlock(&inode->i_lock); |
24ff6663 | 1909 | security_d_instantiate(tmp, inode); |
9308a612 | 1910 | |
9308a612 CH |
1911 | return tmp; |
1912 | ||
1913 | out_iput: | |
24ff6663 JB |
1914 | if (res && !IS_ERR(res)) |
1915 | security_d_instantiate(res, inode); | |
9308a612 CH |
1916 | iput(inode); |
1917 | return res; | |
4ea3ada2 | 1918 | } |
adc48720 | 1919 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1920 | |
1921 | /** | |
1922 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1923 | * @inode: the inode which may have a disconnected dentry | |
1924 | * @dentry: a negative dentry which we want to point to the inode. | |
1925 | * | |
1926 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1927 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1928 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1929 | * | |
1930 | * This is needed in the lookup routine of any filesystem that is exportable | |
1931 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1932 | * | |
1933 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1934 | * is returned. This matches the expected return value of ->lookup. | |
1935 | * | |
6d4ade98 SW |
1936 | * Cluster filesystems may call this function with a negative, hashed dentry. |
1937 | * In that case, we know that the inode will be a regular file, and also this | |
1938 | * will only occur during atomic_open. So we need to check for the dentry | |
1939 | * being already hashed only in the final case. | |
1da177e4 LT |
1940 | */ |
1941 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1942 | { | |
1943 | struct dentry *new = NULL; | |
1944 | ||
a9049376 AV |
1945 | if (IS_ERR(inode)) |
1946 | return ERR_CAST(inode); | |
1947 | ||
21c0d8fd | 1948 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1949 | spin_lock(&inode->i_lock); |
32ba9c3f | 1950 | new = __d_find_alias(inode, 1); |
1da177e4 | 1951 | if (new) { |
32ba9c3f | 1952 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); |
873feea0 | 1953 | spin_unlock(&inode->i_lock); |
1da177e4 | 1954 | security_d_instantiate(new, inode); |
1da177e4 LT |
1955 | d_move(new, dentry); |
1956 | iput(inode); | |
1957 | } else { | |
873feea0 | 1958 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1959 | __d_instantiate(dentry, inode); |
873feea0 | 1960 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1961 | security_d_instantiate(dentry, inode); |
1962 | d_rehash(dentry); | |
1963 | } | |
6d4ade98 SW |
1964 | } else { |
1965 | d_instantiate(dentry, inode); | |
1966 | if (d_unhashed(dentry)) | |
1967 | d_rehash(dentry); | |
1968 | } | |
1da177e4 LT |
1969 | return new; |
1970 | } | |
ec4f8605 | 1971 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1972 | |
9403540c BN |
1973 | /** |
1974 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1975 | * @inode: the inode case-insensitive lookup has found | |
1976 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1977 | * @name: the case-exact name to be associated with the returned dentry | |
1978 | * | |
1979 | * This is to avoid filling the dcache with case-insensitive names to the | |
1980 | * same inode, only the actual correct case is stored in the dcache for | |
1981 | * case-insensitive filesystems. | |
1982 | * | |
1983 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1984 | * already exists in in the dcache, use it and return it. | |
1985 | * | |
1986 | * If no entry exists with the exact case name, allocate new dentry with | |
1987 | * the exact case, and return the spliced entry. | |
1988 | */ | |
e45b590b | 1989 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1990 | struct qstr *name) |
1991 | { | |
9403540c BN |
1992 | struct dentry *found; |
1993 | struct dentry *new; | |
1994 | ||
b6520c81 CH |
1995 | /* |
1996 | * First check if a dentry matching the name already exists, | |
1997 | * if not go ahead and create it now. | |
1998 | */ | |
9403540c | 1999 | found = d_hash_and_lookup(dentry->d_parent, name); |
4f522a24 AV |
2000 | if (unlikely(IS_ERR(found))) |
2001 | goto err_out; | |
9403540c BN |
2002 | if (!found) { |
2003 | new = d_alloc(dentry->d_parent, name); | |
2004 | if (!new) { | |
4f522a24 | 2005 | found = ERR_PTR(-ENOMEM); |
9403540c BN |
2006 | goto err_out; |
2007 | } | |
b6520c81 | 2008 | |
9403540c BN |
2009 | found = d_splice_alias(inode, new); |
2010 | if (found) { | |
2011 | dput(new); | |
2012 | return found; | |
2013 | } | |
2014 | return new; | |
2015 | } | |
b6520c81 CH |
2016 | |
2017 | /* | |
2018 | * If a matching dentry exists, and it's not negative use it. | |
2019 | * | |
2020 | * Decrement the reference count to balance the iget() done | |
2021 | * earlier on. | |
2022 | */ | |
9403540c BN |
2023 | if (found->d_inode) { |
2024 | if (unlikely(found->d_inode != inode)) { | |
2025 | /* This can't happen because bad inodes are unhashed. */ | |
2026 | BUG_ON(!is_bad_inode(inode)); | |
2027 | BUG_ON(!is_bad_inode(found->d_inode)); | |
2028 | } | |
9403540c BN |
2029 | iput(inode); |
2030 | return found; | |
2031 | } | |
b6520c81 | 2032 | |
9403540c | 2033 | /* |
9403540c | 2034 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 2035 | * already has a dentry. |
9403540c | 2036 | */ |
4513d899 AV |
2037 | new = d_splice_alias(inode, found); |
2038 | if (new) { | |
2039 | dput(found); | |
2040 | found = new; | |
9403540c | 2041 | } |
4513d899 | 2042 | return found; |
9403540c BN |
2043 | |
2044 | err_out: | |
2045 | iput(inode); | |
4f522a24 | 2046 | return found; |
9403540c | 2047 | } |
ec4f8605 | 2048 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 2049 | |
12f8ad4b LT |
2050 | /* |
2051 | * Do the slow-case of the dentry name compare. | |
2052 | * | |
2053 | * Unlike the dentry_cmp() function, we need to atomically | |
da53be12 | 2054 | * load the name and length information, so that the |
12f8ad4b LT |
2055 | * filesystem can rely on them, and can use the 'name' and |
2056 | * 'len' information without worrying about walking off the | |
2057 | * end of memory etc. | |
2058 | * | |
2059 | * Thus the read_seqcount_retry() and the "duplicate" info | |
2060 | * in arguments (the low-level filesystem should not look | |
2061 | * at the dentry inode or name contents directly, since | |
2062 | * rename can change them while we're in RCU mode). | |
2063 | */ | |
2064 | enum slow_d_compare { | |
2065 | D_COMP_OK, | |
2066 | D_COMP_NOMATCH, | |
2067 | D_COMP_SEQRETRY, | |
2068 | }; | |
2069 | ||
2070 | static noinline enum slow_d_compare slow_dentry_cmp( | |
2071 | const struct dentry *parent, | |
12f8ad4b LT |
2072 | struct dentry *dentry, |
2073 | unsigned int seq, | |
2074 | const struct qstr *name) | |
2075 | { | |
2076 | int tlen = dentry->d_name.len; | |
2077 | const char *tname = dentry->d_name.name; | |
12f8ad4b LT |
2078 | |
2079 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2080 | cpu_relax(); | |
2081 | return D_COMP_SEQRETRY; | |
2082 | } | |
da53be12 | 2083 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
12f8ad4b LT |
2084 | return D_COMP_NOMATCH; |
2085 | return D_COMP_OK; | |
2086 | } | |
2087 | ||
31e6b01f NP |
2088 | /** |
2089 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2090 | * @parent: parent dentry | |
2091 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2092 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2093 | * Returns: dentry, or NULL |
2094 | * | |
2095 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2096 | * resolution (store-free path walking) design described in | |
2097 | * Documentation/filesystems/path-lookup.txt. | |
2098 | * | |
2099 | * This is not to be used outside core vfs. | |
2100 | * | |
2101 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2102 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2103 | * without taking d_lock and checking d_seq sequence count against @seq | |
2104 | * returned here. | |
2105 | * | |
15570086 | 2106 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2107 | * function. |
2108 | * | |
2109 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2110 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2111 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2112 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2113 | * |
2114 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2115 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2116 | */ |
8966be90 LT |
2117 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2118 | const struct qstr *name, | |
da53be12 | 2119 | unsigned *seqp) |
31e6b01f | 2120 | { |
26fe5750 | 2121 | u64 hashlen = name->hash_len; |
31e6b01f | 2122 | const unsigned char *str = name->name; |
26fe5750 | 2123 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 2124 | struct hlist_bl_node *node; |
31e6b01f NP |
2125 | struct dentry *dentry; |
2126 | ||
2127 | /* | |
2128 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2129 | * required to prevent single threaded performance regressions | |
2130 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2131 | * Keep the two functions in sync. | |
2132 | */ | |
2133 | ||
2134 | /* | |
2135 | * The hash list is protected using RCU. | |
2136 | * | |
2137 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2138 | * races with d_move(). | |
2139 | * | |
2140 | * It is possible that concurrent renames can mess up our list | |
2141 | * walk here and result in missing our dentry, resulting in the | |
2142 | * false-negative result. d_lookup() protects against concurrent | |
2143 | * renames using rename_lock seqlock. | |
2144 | * | |
b0a4bb83 | 2145 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2146 | */ |
b07ad996 | 2147 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2148 | unsigned seq; |
31e6b01f | 2149 | |
31e6b01f | 2150 | seqretry: |
12f8ad4b LT |
2151 | /* |
2152 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2153 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2154 | * |
2155 | * The caller must perform a seqcount check in order | |
da53be12 | 2156 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2157 | * |
2158 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2159 | * we don't wait for the sequence count to stabilize if it | |
2160 | * is in the middle of a sequence change. If we do the slow | |
2161 | * dentry compare, we will do seqretries until it is stable, | |
2162 | * and if we end up with a successful lookup, we actually | |
2163 | * want to exit RCU lookup anyway. | |
2164 | */ | |
2165 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2166 | if (dentry->d_parent != parent) |
2167 | continue; | |
2e321806 LT |
2168 | if (d_unhashed(dentry)) |
2169 | continue; | |
12f8ad4b | 2170 | |
830c0f0e | 2171 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
2172 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2173 | continue; | |
da53be12 LT |
2174 | *seqp = seq; |
2175 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
12f8ad4b LT |
2176 | case D_COMP_OK: |
2177 | return dentry; | |
2178 | case D_COMP_NOMATCH: | |
31e6b01f | 2179 | continue; |
12f8ad4b LT |
2180 | default: |
2181 | goto seqretry; | |
2182 | } | |
31e6b01f | 2183 | } |
12f8ad4b | 2184 | |
26fe5750 | 2185 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 2186 | continue; |
da53be12 | 2187 | *seqp = seq; |
26fe5750 | 2188 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 2189 | return dentry; |
31e6b01f NP |
2190 | } |
2191 | return NULL; | |
2192 | } | |
2193 | ||
1da177e4 LT |
2194 | /** |
2195 | * d_lookup - search for a dentry | |
2196 | * @parent: parent dentry | |
2197 | * @name: qstr of name we wish to find | |
b04f784e | 2198 | * Returns: dentry, or NULL |
1da177e4 | 2199 | * |
b04f784e NP |
2200 | * d_lookup searches the children of the parent dentry for the name in |
2201 | * question. If the dentry is found its reference count is incremented and the | |
2202 | * dentry is returned. The caller must use dput to free the entry when it has | |
2203 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2204 | */ |
da2d8455 | 2205 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2206 | { |
31e6b01f | 2207 | struct dentry *dentry; |
949854d0 | 2208 | unsigned seq; |
1da177e4 LT |
2209 | |
2210 | do { | |
2211 | seq = read_seqbegin(&rename_lock); | |
2212 | dentry = __d_lookup(parent, name); | |
2213 | if (dentry) | |
2214 | break; | |
2215 | } while (read_seqretry(&rename_lock, seq)); | |
2216 | return dentry; | |
2217 | } | |
ec4f8605 | 2218 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2219 | |
31e6b01f | 2220 | /** |
b04f784e NP |
2221 | * __d_lookup - search for a dentry (racy) |
2222 | * @parent: parent dentry | |
2223 | * @name: qstr of name we wish to find | |
2224 | * Returns: dentry, or NULL | |
2225 | * | |
2226 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2227 | * false-negative result due to unrelated rename activity. | |
2228 | * | |
2229 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2230 | * however it must be used carefully, eg. with a following d_lookup in | |
2231 | * the case of failure. | |
2232 | * | |
2233 | * __d_lookup callers must be commented. | |
2234 | */ | |
a713ca2a | 2235 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
2236 | { |
2237 | unsigned int len = name->len; | |
2238 | unsigned int hash = name->hash; | |
2239 | const unsigned char *str = name->name; | |
b07ad996 | 2240 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 2241 | struct hlist_bl_node *node; |
31e6b01f | 2242 | struct dentry *found = NULL; |
665a7583 | 2243 | struct dentry *dentry; |
1da177e4 | 2244 | |
31e6b01f NP |
2245 | /* |
2246 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2247 | * required to prevent single threaded performance regressions | |
2248 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2249 | * Keep the two functions in sync. | |
2250 | */ | |
2251 | ||
b04f784e NP |
2252 | /* |
2253 | * The hash list is protected using RCU. | |
2254 | * | |
2255 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2256 | * with d_move(). | |
2257 | * | |
2258 | * It is possible that concurrent renames can mess up our list | |
2259 | * walk here and result in missing our dentry, resulting in the | |
2260 | * false-negative result. d_lookup() protects against concurrent | |
2261 | * renames using rename_lock seqlock. | |
2262 | * | |
b0a4bb83 | 2263 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2264 | */ |
1da177e4 LT |
2265 | rcu_read_lock(); |
2266 | ||
b07ad996 | 2267 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2268 | |
1da177e4 LT |
2269 | if (dentry->d_name.hash != hash) |
2270 | continue; | |
1da177e4 LT |
2271 | |
2272 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2273 | if (dentry->d_parent != parent) |
2274 | goto next; | |
d0185c08 LT |
2275 | if (d_unhashed(dentry)) |
2276 | goto next; | |
2277 | ||
1da177e4 LT |
2278 | /* |
2279 | * It is safe to compare names since d_move() cannot | |
2280 | * change the qstr (protected by d_lock). | |
2281 | */ | |
fb045adb | 2282 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
2283 | int tlen = dentry->d_name.len; |
2284 | const char *tname = dentry->d_name.name; | |
da53be12 | 2285 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1da177e4 LT |
2286 | goto next; |
2287 | } else { | |
ee983e89 LT |
2288 | if (dentry->d_name.len != len) |
2289 | goto next; | |
12f8ad4b | 2290 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
2291 | goto next; |
2292 | } | |
2293 | ||
98474236 | 2294 | dentry->d_lockref.count++; |
d0185c08 | 2295 | found = dentry; |
1da177e4 LT |
2296 | spin_unlock(&dentry->d_lock); |
2297 | break; | |
2298 | next: | |
2299 | spin_unlock(&dentry->d_lock); | |
2300 | } | |
2301 | rcu_read_unlock(); | |
2302 | ||
2303 | return found; | |
2304 | } | |
2305 | ||
3e7e241f EB |
2306 | /** |
2307 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2308 | * @dir: Directory to search in | |
2309 | * @name: qstr of name we wish to find | |
2310 | * | |
4f522a24 | 2311 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2312 | */ |
2313 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2314 | { | |
3e7e241f EB |
2315 | /* |
2316 | * Check for a fs-specific hash function. Note that we must | |
2317 | * calculate the standard hash first, as the d_op->d_hash() | |
2318 | * routine may choose to leave the hash value unchanged. | |
2319 | */ | |
2320 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2321 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2322 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2323 | if (unlikely(err < 0)) |
2324 | return ERR_PTR(err); | |
3e7e241f | 2325 | } |
4f522a24 | 2326 | return d_lookup(dir, name); |
3e7e241f | 2327 | } |
4f522a24 | 2328 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2329 | |
1da177e4 | 2330 | /** |
786a5e15 | 2331 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2332 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2333 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2334 | * |
2335 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2336 | * This is used by ncpfs in its readdir implementation. | |
2337 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2338 | * |
2339 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2340 | */ |
d3a23e16 | 2341 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2342 | { |
786a5e15 | 2343 | struct dentry *child; |
d3a23e16 | 2344 | |
2fd6b7f5 | 2345 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2346 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2347 | if (dentry == child) { | |
2fd6b7f5 | 2348 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2349 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2350 | spin_unlock(&dentry->d_lock); |
2351 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2352 | return 1; |
2353 | } | |
2354 | } | |
2fd6b7f5 | 2355 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2356 | |
1da177e4 LT |
2357 | return 0; |
2358 | } | |
ec4f8605 | 2359 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2360 | |
2361 | /* | |
2362 | * When a file is deleted, we have two options: | |
2363 | * - turn this dentry into a negative dentry | |
2364 | * - unhash this dentry and free it. | |
2365 | * | |
2366 | * Usually, we want to just turn this into | |
2367 | * a negative dentry, but if anybody else is | |
2368 | * currently using the dentry or the inode | |
2369 | * we can't do that and we fall back on removing | |
2370 | * it from the hash queues and waiting for | |
2371 | * it to be deleted later when it has no users | |
2372 | */ | |
2373 | ||
2374 | /** | |
2375 | * d_delete - delete a dentry | |
2376 | * @dentry: The dentry to delete | |
2377 | * | |
2378 | * Turn the dentry into a negative dentry if possible, otherwise | |
2379 | * remove it from the hash queues so it can be deleted later | |
2380 | */ | |
2381 | ||
2382 | void d_delete(struct dentry * dentry) | |
2383 | { | |
873feea0 | 2384 | struct inode *inode; |
7a91bf7f | 2385 | int isdir = 0; |
1da177e4 LT |
2386 | /* |
2387 | * Are we the only user? | |
2388 | */ | |
357f8e65 | 2389 | again: |
1da177e4 | 2390 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2391 | inode = dentry->d_inode; |
2392 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2393 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2394 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2395 | spin_unlock(&dentry->d_lock); |
2396 | cpu_relax(); | |
2397 | goto again; | |
2398 | } | |
13e3c5e5 | 2399 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2400 | dentry_unlink_inode(dentry); |
7a91bf7f | 2401 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2402 | return; |
2403 | } | |
2404 | ||
2405 | if (!d_unhashed(dentry)) | |
2406 | __d_drop(dentry); | |
2407 | ||
2408 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2409 | |
2410 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2411 | } |
ec4f8605 | 2412 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2413 | |
b07ad996 | 2414 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2415 | { |
ceb5bdc2 | 2416 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2417 | hlist_bl_lock(b); |
dea3667b | 2418 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2419 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2420 | hlist_bl_unlock(b); |
1da177e4 LT |
2421 | } |
2422 | ||
770bfad8 DH |
2423 | static void _d_rehash(struct dentry * entry) |
2424 | { | |
2425 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2426 | } | |
2427 | ||
1da177e4 LT |
2428 | /** |
2429 | * d_rehash - add an entry back to the hash | |
2430 | * @entry: dentry to add to the hash | |
2431 | * | |
2432 | * Adds a dentry to the hash according to its name. | |
2433 | */ | |
2434 | ||
2435 | void d_rehash(struct dentry * entry) | |
2436 | { | |
1da177e4 | 2437 | spin_lock(&entry->d_lock); |
770bfad8 | 2438 | _d_rehash(entry); |
1da177e4 | 2439 | spin_unlock(&entry->d_lock); |
1da177e4 | 2440 | } |
ec4f8605 | 2441 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2442 | |
fb2d5b86 NP |
2443 | /** |
2444 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2445 | * @dentry: dentry to be updated | |
2446 | * @name: new name | |
2447 | * | |
2448 | * Update a case insensitive dentry with new case of name. | |
2449 | * | |
2450 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2451 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2452 | * lengths). | |
2453 | * | |
2454 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2455 | * keep renames and concurrent inserts, and readdir(2) away). | |
2456 | */ | |
2457 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2458 | { | |
7ebfa57f | 2459 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2460 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2461 | ||
fb2d5b86 | 2462 | spin_lock(&dentry->d_lock); |
31e6b01f | 2463 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2464 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2465 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2466 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2467 | } |
2468 | EXPORT_SYMBOL(dentry_update_name_case); | |
2469 | ||
1da177e4 LT |
2470 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2471 | { | |
2472 | if (dname_external(target)) { | |
2473 | if (dname_external(dentry)) { | |
2474 | /* | |
2475 | * Both external: swap the pointers | |
2476 | */ | |
9a8d5bb4 | 2477 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2478 | } else { |
2479 | /* | |
2480 | * dentry:internal, target:external. Steal target's | |
2481 | * storage and make target internal. | |
2482 | */ | |
321bcf92 BF |
2483 | memcpy(target->d_iname, dentry->d_name.name, |
2484 | dentry->d_name.len + 1); | |
1da177e4 LT |
2485 | dentry->d_name.name = target->d_name.name; |
2486 | target->d_name.name = target->d_iname; | |
2487 | } | |
2488 | } else { | |
2489 | if (dname_external(dentry)) { | |
2490 | /* | |
2491 | * dentry:external, target:internal. Give dentry's | |
2492 | * storage to target and make dentry internal | |
2493 | */ | |
2494 | memcpy(dentry->d_iname, target->d_name.name, | |
2495 | target->d_name.len + 1); | |
2496 | target->d_name.name = dentry->d_name.name; | |
2497 | dentry->d_name.name = dentry->d_iname; | |
2498 | } else { | |
2499 | /* | |
2500 | * Both are internal. Just copy target to dentry | |
2501 | */ | |
2502 | memcpy(dentry->d_iname, target->d_name.name, | |
2503 | target->d_name.len + 1); | |
dc711ca3 AV |
2504 | dentry->d_name.len = target->d_name.len; |
2505 | return; | |
1da177e4 LT |
2506 | } |
2507 | } | |
9a8d5bb4 | 2508 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2509 | } |
2510 | ||
2fd6b7f5 NP |
2511 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2512 | { | |
2513 | /* | |
2514 | * XXXX: do we really need to take target->d_lock? | |
2515 | */ | |
2516 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2517 | spin_lock(&target->d_parent->d_lock); | |
2518 | else { | |
2519 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2520 | spin_lock(&dentry->d_parent->d_lock); | |
2521 | spin_lock_nested(&target->d_parent->d_lock, | |
2522 | DENTRY_D_LOCK_NESTED); | |
2523 | } else { | |
2524 | spin_lock(&target->d_parent->d_lock); | |
2525 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2526 | DENTRY_D_LOCK_NESTED); | |
2527 | } | |
2528 | } | |
2529 | if (target < dentry) { | |
2530 | spin_lock_nested(&target->d_lock, 2); | |
2531 | spin_lock_nested(&dentry->d_lock, 3); | |
2532 | } else { | |
2533 | spin_lock_nested(&dentry->d_lock, 2); | |
2534 | spin_lock_nested(&target->d_lock, 3); | |
2535 | } | |
2536 | } | |
2537 | ||
2538 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2539 | struct dentry *target) | |
2540 | { | |
2541 | if (target->d_parent != dentry->d_parent) | |
2542 | spin_unlock(&dentry->d_parent->d_lock); | |
2543 | if (target->d_parent != target) | |
2544 | spin_unlock(&target->d_parent->d_lock); | |
2545 | } | |
2546 | ||
1da177e4 | 2547 | /* |
2fd6b7f5 NP |
2548 | * When switching names, the actual string doesn't strictly have to |
2549 | * be preserved in the target - because we're dropping the target | |
2550 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2551 | * the new name before we switch. | |
2552 | * | |
2553 | * Note that we have to be a lot more careful about getting the hash | |
2554 | * switched - we have to switch the hash value properly even if it | |
2555 | * then no longer matches the actual (corrupted) string of the target. | |
2556 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2557 | */ |
9eaef27b | 2558 | /* |
18367501 | 2559 | * __d_move - move a dentry |
1da177e4 LT |
2560 | * @dentry: entry to move |
2561 | * @target: new dentry | |
2562 | * | |
2563 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2564 | * dcache entries should not be moved in this way. Caller must hold |
2565 | * rename_lock, the i_mutex of the source and target directories, | |
2566 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2567 | */ |
18367501 | 2568 | static void __d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2569 | { |
1da177e4 LT |
2570 | if (!dentry->d_inode) |
2571 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2572 | ||
2fd6b7f5 NP |
2573 | BUG_ON(d_ancestor(dentry, target)); |
2574 | BUG_ON(d_ancestor(target, dentry)); | |
2575 | ||
2fd6b7f5 | 2576 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2577 | |
31e6b01f NP |
2578 | write_seqcount_begin(&dentry->d_seq); |
2579 | write_seqcount_begin(&target->d_seq); | |
2580 | ||
ceb5bdc2 NP |
2581 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2582 | ||
2583 | /* | |
2584 | * Move the dentry to the target hash queue. Don't bother checking | |
2585 | * for the same hash queue because of how unlikely it is. | |
2586 | */ | |
2587 | __d_drop(dentry); | |
789680d1 | 2588 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2589 | |
2590 | /* Unhash the target: dput() will then get rid of it */ | |
2591 | __d_drop(target); | |
2592 | ||
5160ee6f ED |
2593 | list_del(&dentry->d_u.d_child); |
2594 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2595 | |
2596 | /* Switch the names.. */ | |
2597 | switch_names(dentry, target); | |
9a8d5bb4 | 2598 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2599 | |
2600 | /* ... and switch the parents */ | |
2601 | if (IS_ROOT(dentry)) { | |
2602 | dentry->d_parent = target->d_parent; | |
2603 | target->d_parent = target; | |
5160ee6f | 2604 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2605 | } else { |
9a8d5bb4 | 2606 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2607 | |
2608 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2609 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2610 | } |
2611 | ||
5160ee6f | 2612 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2613 | |
31e6b01f NP |
2614 | write_seqcount_end(&target->d_seq); |
2615 | write_seqcount_end(&dentry->d_seq); | |
2616 | ||
2fd6b7f5 | 2617 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2618 | spin_unlock(&target->d_lock); |
c32ccd87 | 2619 | fsnotify_d_move(dentry); |
1da177e4 | 2620 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2621 | } |
2622 | ||
2623 | /* | |
2624 | * d_move - move a dentry | |
2625 | * @dentry: entry to move | |
2626 | * @target: new dentry | |
2627 | * | |
2628 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2629 | * dcache entries should not be moved in this way. See the locking |
2630 | * requirements for __d_move. | |
18367501 AV |
2631 | */ |
2632 | void d_move(struct dentry *dentry, struct dentry *target) | |
2633 | { | |
2634 | write_seqlock(&rename_lock); | |
2635 | __d_move(dentry, target); | |
1da177e4 | 2636 | write_sequnlock(&rename_lock); |
9eaef27b | 2637 | } |
ec4f8605 | 2638 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2639 | |
e2761a11 OH |
2640 | /** |
2641 | * d_ancestor - search for an ancestor | |
2642 | * @p1: ancestor dentry | |
2643 | * @p2: child dentry | |
2644 | * | |
2645 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2646 | * an ancestor of p2, else NULL. | |
9eaef27b | 2647 | */ |
e2761a11 | 2648 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2649 | { |
2650 | struct dentry *p; | |
2651 | ||
871c0067 | 2652 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2653 | if (p->d_parent == p1) |
e2761a11 | 2654 | return p; |
9eaef27b | 2655 | } |
e2761a11 | 2656 | return NULL; |
9eaef27b TM |
2657 | } |
2658 | ||
2659 | /* | |
2660 | * This helper attempts to cope with remotely renamed directories | |
2661 | * | |
2662 | * It assumes that the caller is already holding | |
18367501 | 2663 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2664 | * |
2665 | * Note: If ever the locking in lock_rename() changes, then please | |
2666 | * remember to update this too... | |
9eaef27b | 2667 | */ |
873feea0 NP |
2668 | static struct dentry *__d_unalias(struct inode *inode, |
2669 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2670 | { |
2671 | struct mutex *m1 = NULL, *m2 = NULL; | |
ee3efa91 | 2672 | struct dentry *ret = ERR_PTR(-EBUSY); |
9eaef27b TM |
2673 | |
2674 | /* If alias and dentry share a parent, then no extra locks required */ | |
2675 | if (alias->d_parent == dentry->d_parent) | |
2676 | goto out_unalias; | |
2677 | ||
9eaef27b | 2678 | /* See lock_rename() */ |
9eaef27b TM |
2679 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2680 | goto out_err; | |
2681 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2682 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2683 | goto out_err; | |
2684 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2685 | out_unalias: | |
ee3efa91 AV |
2686 | if (likely(!d_mountpoint(alias))) { |
2687 | __d_move(alias, dentry); | |
2688 | ret = alias; | |
2689 | } | |
9eaef27b | 2690 | out_err: |
873feea0 | 2691 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2692 | if (m2) |
2693 | mutex_unlock(m2); | |
2694 | if (m1) | |
2695 | mutex_unlock(m1); | |
2696 | return ret; | |
2697 | } | |
2698 | ||
770bfad8 DH |
2699 | /* |
2700 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2701 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2702 | * returns with anon->d_lock held! |
770bfad8 DH |
2703 | */ |
2704 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2705 | { | |
740da42e | 2706 | struct dentry *dparent; |
770bfad8 | 2707 | |
2fd6b7f5 | 2708 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2709 | |
31e6b01f NP |
2710 | write_seqcount_begin(&dentry->d_seq); |
2711 | write_seqcount_begin(&anon->d_seq); | |
2712 | ||
770bfad8 | 2713 | dparent = dentry->d_parent; |
770bfad8 | 2714 | |
2fd6b7f5 NP |
2715 | switch_names(dentry, anon); |
2716 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2717 | ||
740da42e AV |
2718 | dentry->d_parent = dentry; |
2719 | list_del_init(&dentry->d_u.d_child); | |
2720 | anon->d_parent = dparent; | |
9ed53b12 | 2721 | list_move(&anon->d_u.d_child, &dparent->d_subdirs); |
770bfad8 | 2722 | |
31e6b01f NP |
2723 | write_seqcount_end(&dentry->d_seq); |
2724 | write_seqcount_end(&anon->d_seq); | |
2725 | ||
2fd6b7f5 NP |
2726 | dentry_unlock_parents_for_move(anon, dentry); |
2727 | spin_unlock(&dentry->d_lock); | |
2728 | ||
2729 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2730 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2731 | } | |
2732 | ||
2733 | /** | |
2734 | * d_materialise_unique - introduce an inode into the tree | |
2735 | * @dentry: candidate dentry | |
2736 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2737 | * | |
2738 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2739 | * root directory alias in its place if there is one. Caller must hold the |
2740 | * i_mutex of the parent directory. | |
770bfad8 DH |
2741 | */ |
2742 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2743 | { | |
9eaef27b | 2744 | struct dentry *actual; |
770bfad8 DH |
2745 | |
2746 | BUG_ON(!d_unhashed(dentry)); | |
2747 | ||
770bfad8 DH |
2748 | if (!inode) { |
2749 | actual = dentry; | |
360da900 | 2750 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2751 | d_rehash(actual); |
2752 | goto out_nolock; | |
770bfad8 DH |
2753 | } |
2754 | ||
873feea0 | 2755 | spin_lock(&inode->i_lock); |
357f8e65 | 2756 | |
9eaef27b TM |
2757 | if (S_ISDIR(inode->i_mode)) { |
2758 | struct dentry *alias; | |
2759 | ||
2760 | /* Does an aliased dentry already exist? */ | |
32ba9c3f | 2761 | alias = __d_find_alias(inode, 0); |
9eaef27b TM |
2762 | if (alias) { |
2763 | actual = alias; | |
18367501 AV |
2764 | write_seqlock(&rename_lock); |
2765 | ||
2766 | if (d_ancestor(alias, dentry)) { | |
2767 | /* Check for loops */ | |
2768 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2769 | spin_unlock(&inode->i_lock); |
18367501 AV |
2770 | } else if (IS_ROOT(alias)) { |
2771 | /* Is this an anonymous mountpoint that we | |
2772 | * could splice into our tree? */ | |
9eaef27b | 2773 | __d_materialise_dentry(dentry, alias); |
18367501 | 2774 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2775 | __d_drop(alias); |
2776 | goto found; | |
18367501 AV |
2777 | } else { |
2778 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2779 | * aliasing. This drops inode->i_lock */ |
18367501 | 2780 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2781 | } |
18367501 | 2782 | write_sequnlock(&rename_lock); |
dd179946 DH |
2783 | if (IS_ERR(actual)) { |
2784 | if (PTR_ERR(actual) == -ELOOP) | |
2785 | pr_warn_ratelimited( | |
2786 | "VFS: Lookup of '%s' in %s %s" | |
2787 | " would have caused loop\n", | |
2788 | dentry->d_name.name, | |
2789 | inode->i_sb->s_type->name, | |
2790 | inode->i_sb->s_id); | |
9eaef27b | 2791 | dput(alias); |
dd179946 | 2792 | } |
9eaef27b TM |
2793 | goto out_nolock; |
2794 | } | |
770bfad8 DH |
2795 | } |
2796 | ||
2797 | /* Add a unique reference */ | |
2798 | actual = __d_instantiate_unique(dentry, inode); | |
2799 | if (!actual) | |
2800 | actual = dentry; | |
357f8e65 NP |
2801 | else |
2802 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2803 | |
770bfad8 DH |
2804 | spin_lock(&actual->d_lock); |
2805 | found: | |
2806 | _d_rehash(actual); | |
2807 | spin_unlock(&actual->d_lock); | |
873feea0 | 2808 | spin_unlock(&inode->i_lock); |
9eaef27b | 2809 | out_nolock: |
770bfad8 DH |
2810 | if (actual == dentry) { |
2811 | security_d_instantiate(dentry, inode); | |
2812 | return NULL; | |
2813 | } | |
2814 | ||
2815 | iput(inode); | |
2816 | return actual; | |
770bfad8 | 2817 | } |
ec4f8605 | 2818 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2819 | |
cdd16d02 | 2820 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2821 | { |
2822 | *buflen -= namelen; | |
2823 | if (*buflen < 0) | |
2824 | return -ENAMETOOLONG; | |
2825 | *buffer -= namelen; | |
2826 | memcpy(*buffer, str, namelen); | |
2827 | return 0; | |
2828 | } | |
2829 | ||
232d2d60 WL |
2830 | /** |
2831 | * prepend_name - prepend a pathname in front of current buffer pointer | |
18129977 WL |
2832 | * @buffer: buffer pointer |
2833 | * @buflen: allocated length of the buffer | |
2834 | * @name: name string and length qstr structure | |
232d2d60 WL |
2835 | * |
2836 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
2837 | * make sure that either the old or the new name pointer and length are | |
2838 | * fetched. However, there may be mismatch between length and pointer. | |
2839 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
2840 | * the length is reached or a null byte is found. It also prepends "/" at | |
2841 | * the beginning of the name. The sequence number check at the caller will | |
2842 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
2843 | * due to mismatched pointer and length will be discarded. | |
2844 | */ | |
cdd16d02 MS |
2845 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2846 | { | |
232d2d60 WL |
2847 | const char *dname = ACCESS_ONCE(name->name); |
2848 | u32 dlen = ACCESS_ONCE(name->len); | |
2849 | char *p; | |
2850 | ||
2851 | if (*buflen < dlen + 1) | |
2852 | return -ENAMETOOLONG; | |
2853 | *buflen -= dlen + 1; | |
2854 | p = *buffer -= dlen + 1; | |
2855 | *p++ = '/'; | |
2856 | while (dlen--) { | |
2857 | char c = *dname++; | |
2858 | if (!c) | |
2859 | break; | |
2860 | *p++ = c; | |
2861 | } | |
2862 | return 0; | |
cdd16d02 MS |
2863 | } |
2864 | ||
1da177e4 | 2865 | /** |
208898c1 | 2866 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2867 | * @path: the dentry/vfsmount to report |
02125a82 | 2868 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2869 | * @buffer: pointer to the end of the buffer |
2870 | * @buflen: pointer to buffer length | |
552ce544 | 2871 | * |
18129977 WL |
2872 | * The function will first try to write out the pathname without taking any |
2873 | * lock other than the RCU read lock to make sure that dentries won't go away. | |
2874 | * It only checks the sequence number of the global rename_lock as any change | |
2875 | * in the dentry's d_seq will be preceded by changes in the rename_lock | |
2876 | * sequence number. If the sequence number had been changed, it will restart | |
2877 | * the whole pathname back-tracing sequence again by taking the rename_lock. | |
2878 | * In this case, there is no need to take the RCU read lock as the recursive | |
2879 | * parent pointer references will keep the dentry chain alive as long as no | |
2880 | * rename operation is performed. | |
1da177e4 | 2881 | */ |
02125a82 AV |
2882 | static int prepend_path(const struct path *path, |
2883 | const struct path *root, | |
f2eb6575 | 2884 | char **buffer, int *buflen) |
1da177e4 | 2885 | { |
9d1bc601 MS |
2886 | struct dentry *dentry = path->dentry; |
2887 | struct vfsmount *vfsmnt = path->mnt; | |
0714a533 | 2888 | struct mount *mnt = real_mount(vfsmnt); |
f2eb6575 | 2889 | int error = 0; |
232d2d60 WL |
2890 | unsigned seq = 0; |
2891 | char *bptr; | |
2892 | int blen; | |
6092d048 | 2893 | |
94e92a6e | 2894 | br_read_lock(&vfsmount_lock); |
48f5ec21 | 2895 | rcu_read_lock(); |
232d2d60 WL |
2896 | restart: |
2897 | bptr = *buffer; | |
2898 | blen = *buflen; | |
2899 | read_seqbegin_or_lock(&rename_lock, &seq); | |
f2eb6575 | 2900 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2901 | struct dentry * parent; |
2902 | ||
1da177e4 | 2903 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2904 | /* Global root? */ |
232d2d60 WL |
2905 | if (mnt_has_parent(mnt)) { |
2906 | dentry = mnt->mnt_mountpoint; | |
2907 | mnt = mnt->mnt_parent; | |
2908 | vfsmnt = &mnt->mnt; | |
2909 | continue; | |
2910 | } | |
2911 | /* | |
2912 | * Filesystems needing to implement special "root names" | |
2913 | * should do so with ->d_dname() | |
2914 | */ | |
2915 | if (IS_ROOT(dentry) && | |
2916 | (dentry->d_name.len != 1 || | |
2917 | dentry->d_name.name[0] != '/')) { | |
2918 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2919 | (int) dentry->d_name.len, | |
2920 | dentry->d_name.name); | |
2921 | } | |
2922 | if (!error) | |
2923 | error = is_mounted(vfsmnt) ? 1 : 2; | |
2924 | break; | |
1da177e4 LT |
2925 | } |
2926 | parent = dentry->d_parent; | |
2927 | prefetch(parent); | |
232d2d60 | 2928 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
2929 | if (error) |
2930 | break; | |
2931 | ||
1da177e4 LT |
2932 | dentry = parent; |
2933 | } | |
48f5ec21 AV |
2934 | if (!(seq & 1)) |
2935 | rcu_read_unlock(); | |
2936 | if (need_seqretry(&rename_lock, seq)) { | |
2937 | seq = 1; | |
232d2d60 | 2938 | goto restart; |
48f5ec21 AV |
2939 | } |
2940 | done_seqretry(&rename_lock, seq); | |
94e92a6e | 2941 | br_read_unlock(&vfsmount_lock); |
1da177e4 | 2942 | |
232d2d60 WL |
2943 | if (error >= 0 && bptr == *buffer) { |
2944 | if (--blen < 0) | |
2945 | error = -ENAMETOOLONG; | |
2946 | else | |
2947 | *--bptr = '/'; | |
2948 | } | |
2949 | *buffer = bptr; | |
2950 | *buflen = blen; | |
7ea600b5 | 2951 | return error; |
f2eb6575 | 2952 | } |
be285c71 | 2953 | |
f2eb6575 MS |
2954 | /** |
2955 | * __d_path - return the path of a dentry | |
2956 | * @path: the dentry/vfsmount to report | |
02125a82 | 2957 | * @root: root vfsmnt/dentry |
cd956a1c | 2958 | * @buf: buffer to return value in |
f2eb6575 MS |
2959 | * @buflen: buffer length |
2960 | * | |
ffd1f4ed | 2961 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2962 | * |
2963 | * Returns a pointer into the buffer or an error code if the | |
2964 | * path was too long. | |
2965 | * | |
be148247 | 2966 | * "buflen" should be positive. |
f2eb6575 | 2967 | * |
02125a82 | 2968 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2969 | */ |
02125a82 AV |
2970 | char *__d_path(const struct path *path, |
2971 | const struct path *root, | |
f2eb6575 MS |
2972 | char *buf, int buflen) |
2973 | { | |
2974 | char *res = buf + buflen; | |
2975 | int error; | |
2976 | ||
2977 | prepend(&res, &buflen, "\0", 1); | |
f2eb6575 | 2978 | error = prepend_path(path, root, &res, &buflen); |
be148247 | 2979 | |
02125a82 AV |
2980 | if (error < 0) |
2981 | return ERR_PTR(error); | |
2982 | if (error > 0) | |
2983 | return NULL; | |
2984 | return res; | |
2985 | } | |
2986 | ||
2987 | char *d_absolute_path(const struct path *path, | |
2988 | char *buf, int buflen) | |
2989 | { | |
2990 | struct path root = {}; | |
2991 | char *res = buf + buflen; | |
2992 | int error; | |
2993 | ||
2994 | prepend(&res, &buflen, "\0", 1); | |
02125a82 | 2995 | error = prepend_path(path, &root, &res, &buflen); |
02125a82 AV |
2996 | |
2997 | if (error > 1) | |
2998 | error = -EINVAL; | |
2999 | if (error < 0) | |
f2eb6575 | 3000 | return ERR_PTR(error); |
f2eb6575 | 3001 | return res; |
1da177e4 LT |
3002 | } |
3003 | ||
ffd1f4ed MS |
3004 | /* |
3005 | * same as __d_path but appends "(deleted)" for unlinked files. | |
3006 | */ | |
02125a82 AV |
3007 | static int path_with_deleted(const struct path *path, |
3008 | const struct path *root, | |
3009 | char **buf, int *buflen) | |
ffd1f4ed MS |
3010 | { |
3011 | prepend(buf, buflen, "\0", 1); | |
3012 | if (d_unlinked(path->dentry)) { | |
3013 | int error = prepend(buf, buflen, " (deleted)", 10); | |
3014 | if (error) | |
3015 | return error; | |
3016 | } | |
3017 | ||
3018 | return prepend_path(path, root, buf, buflen); | |
3019 | } | |
3020 | ||
8df9d1a4 MS |
3021 | static int prepend_unreachable(char **buffer, int *buflen) |
3022 | { | |
3023 | return prepend(buffer, buflen, "(unreachable)", 13); | |
3024 | } | |
3025 | ||
68f0d9d9 LT |
3026 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) |
3027 | { | |
3028 | unsigned seq; | |
3029 | ||
3030 | do { | |
3031 | seq = read_seqcount_begin(&fs->seq); | |
3032 | *root = fs->root; | |
3033 | } while (read_seqcount_retry(&fs->seq, seq)); | |
3034 | } | |
3035 | ||
a03a8a70 JB |
3036 | /** |
3037 | * d_path - return the path of a dentry | |
cf28b486 | 3038 | * @path: path to report |
a03a8a70 JB |
3039 | * @buf: buffer to return value in |
3040 | * @buflen: buffer length | |
3041 | * | |
3042 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
3043 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
3044 | * | |
52afeefb AV |
3045 | * Returns a pointer into the buffer or an error code if the path was |
3046 | * too long. Note: Callers should use the returned pointer, not the passed | |
3047 | * in buffer, to use the name! The implementation often starts at an offset | |
3048 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 3049 | * |
31f3e0b3 | 3050 | * "buflen" should be positive. |
a03a8a70 | 3051 | */ |
20d4fdc1 | 3052 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 3053 | { |
ffd1f4ed | 3054 | char *res = buf + buflen; |
6ac08c39 | 3055 | struct path root; |
ffd1f4ed | 3056 | int error; |
1da177e4 | 3057 | |
c23fbb6b ED |
3058 | /* |
3059 | * We have various synthetic filesystems that never get mounted. On | |
3060 | * these filesystems dentries are never used for lookup purposes, and | |
3061 | * thus don't need to be hashed. They also don't need a name until a | |
3062 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
3063 | * below allows us to generate a name for these objects on demand: | |
3064 | */ | |
cf28b486 JB |
3065 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
3066 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 3067 | |
68f0d9d9 LT |
3068 | rcu_read_lock(); |
3069 | get_fs_root_rcu(current->fs, &root); | |
02125a82 | 3070 | error = path_with_deleted(path, &root, &res, &buflen); |
68f0d9d9 LT |
3071 | rcu_read_unlock(); |
3072 | ||
02125a82 | 3073 | if (error < 0) |
ffd1f4ed | 3074 | res = ERR_PTR(error); |
1da177e4 LT |
3075 | return res; |
3076 | } | |
ec4f8605 | 3077 | EXPORT_SYMBOL(d_path); |
1da177e4 | 3078 | |
c23fbb6b ED |
3079 | /* |
3080 | * Helper function for dentry_operations.d_dname() members | |
3081 | */ | |
3082 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
3083 | const char *fmt, ...) | |
3084 | { | |
3085 | va_list args; | |
3086 | char temp[64]; | |
3087 | int sz; | |
3088 | ||
3089 | va_start(args, fmt); | |
3090 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
3091 | va_end(args); | |
3092 | ||
3093 | if (sz > sizeof(temp) || sz > buflen) | |
3094 | return ERR_PTR(-ENAMETOOLONG); | |
3095 | ||
3096 | buffer += buflen - sz; | |
3097 | return memcpy(buffer, temp, sz); | |
3098 | } | |
3099 | ||
118b2302 AV |
3100 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
3101 | { | |
3102 | char *end = buffer + buflen; | |
3103 | /* these dentries are never renamed, so d_lock is not needed */ | |
3104 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
232d2d60 | 3105 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
118b2302 AV |
3106 | prepend(&end, &buflen, "/", 1)) |
3107 | end = ERR_PTR(-ENAMETOOLONG); | |
232d2d60 | 3108 | return end; |
118b2302 AV |
3109 | } |
3110 | ||
6092d048 RP |
3111 | /* |
3112 | * Write full pathname from the root of the filesystem into the buffer. | |
3113 | */ | |
ec2447c2 | 3114 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 | 3115 | { |
232d2d60 WL |
3116 | char *end, *retval; |
3117 | int len, seq = 0; | |
3118 | int error = 0; | |
6092d048 | 3119 | |
48f5ec21 | 3120 | rcu_read_lock(); |
232d2d60 WL |
3121 | restart: |
3122 | end = buf + buflen; | |
3123 | len = buflen; | |
3124 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
3125 | if (buflen < 1) |
3126 | goto Elong; | |
3127 | /* Get '/' right */ | |
3128 | retval = end-1; | |
3129 | *retval = '/'; | |
232d2d60 | 3130 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
3131 | while (!IS_ROOT(dentry)) { |
3132 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 3133 | int error; |
6092d048 | 3134 | |
6092d048 | 3135 | prefetch(parent); |
232d2d60 WL |
3136 | error = prepend_name(&end, &len, &dentry->d_name); |
3137 | if (error) | |
3138 | break; | |
6092d048 RP |
3139 | |
3140 | retval = end; | |
3141 | dentry = parent; | |
3142 | } | |
48f5ec21 AV |
3143 | if (!(seq & 1)) |
3144 | rcu_read_unlock(); | |
3145 | if (need_seqretry(&rename_lock, seq)) { | |
3146 | seq = 1; | |
232d2d60 | 3147 | goto restart; |
48f5ec21 AV |
3148 | } |
3149 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3150 | if (error) |
3151 | goto Elong; | |
c103135c AV |
3152 | return retval; |
3153 | Elong: | |
3154 | return ERR_PTR(-ENAMETOOLONG); | |
3155 | } | |
ec2447c2 NP |
3156 | |
3157 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3158 | { | |
232d2d60 | 3159 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3160 | } |
3161 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3162 | |
3163 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3164 | { | |
3165 | char *p = NULL; | |
3166 | char *retval; | |
3167 | ||
c103135c AV |
3168 | if (d_unlinked(dentry)) { |
3169 | p = buf + buflen; | |
3170 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3171 | goto Elong; | |
3172 | buflen++; | |
3173 | } | |
3174 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3175 | if (!IS_ERR(retval) && p) |
3176 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3177 | return retval; |
3178 | Elong: | |
6092d048 RP |
3179 | return ERR_PTR(-ENAMETOOLONG); |
3180 | } | |
3181 | ||
8b19e341 LT |
3182 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, |
3183 | struct path *pwd) | |
5762482f | 3184 | { |
8b19e341 LT |
3185 | unsigned seq; |
3186 | ||
3187 | do { | |
3188 | seq = read_seqcount_begin(&fs->seq); | |
3189 | *root = fs->root; | |
3190 | *pwd = fs->pwd; | |
3191 | } while (read_seqcount_retry(&fs->seq, seq)); | |
5762482f LT |
3192 | } |
3193 | ||
1da177e4 LT |
3194 | /* |
3195 | * NOTE! The user-level library version returns a | |
3196 | * character pointer. The kernel system call just | |
3197 | * returns the length of the buffer filled (which | |
3198 | * includes the ending '\0' character), or a negative | |
3199 | * error value. So libc would do something like | |
3200 | * | |
3201 | * char *getcwd(char * buf, size_t size) | |
3202 | * { | |
3203 | * int retval; | |
3204 | * | |
3205 | * retval = sys_getcwd(buf, size); | |
3206 | * if (retval >= 0) | |
3207 | * return buf; | |
3208 | * errno = -retval; | |
3209 | * return NULL; | |
3210 | * } | |
3211 | */ | |
3cdad428 | 3212 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3213 | { |
552ce544 | 3214 | int error; |
6ac08c39 | 3215 | struct path pwd, root; |
3272c544 | 3216 | char *page = __getname(); |
1da177e4 LT |
3217 | |
3218 | if (!page) | |
3219 | return -ENOMEM; | |
3220 | ||
8b19e341 LT |
3221 | rcu_read_lock(); |
3222 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); | |
1da177e4 | 3223 | |
552ce544 | 3224 | error = -ENOENT; |
f3da392e | 3225 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3226 | unsigned long len; |
3272c544 LT |
3227 | char *cwd = page + PATH_MAX; |
3228 | int buflen = PATH_MAX; | |
1da177e4 | 3229 | |
8df9d1a4 | 3230 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3231 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
ff812d72 | 3232 | rcu_read_unlock(); |
552ce544 | 3233 | |
02125a82 | 3234 | if (error < 0) |
552ce544 LT |
3235 | goto out; |
3236 | ||
8df9d1a4 | 3237 | /* Unreachable from current root */ |
02125a82 | 3238 | if (error > 0) { |
8df9d1a4 MS |
3239 | error = prepend_unreachable(&cwd, &buflen); |
3240 | if (error) | |
3241 | goto out; | |
3242 | } | |
3243 | ||
552ce544 | 3244 | error = -ERANGE; |
3272c544 | 3245 | len = PATH_MAX + page - cwd; |
552ce544 LT |
3246 | if (len <= size) { |
3247 | error = len; | |
3248 | if (copy_to_user(buf, cwd, len)) | |
3249 | error = -EFAULT; | |
3250 | } | |
949854d0 | 3251 | } else { |
ff812d72 | 3252 | rcu_read_unlock(); |
949854d0 | 3253 | } |
1da177e4 LT |
3254 | |
3255 | out: | |
3272c544 | 3256 | __putname(page); |
1da177e4 LT |
3257 | return error; |
3258 | } | |
3259 | ||
3260 | /* | |
3261 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3262 | * | |
3263 | * Trivially implemented using the dcache structure | |
3264 | */ | |
3265 | ||
3266 | /** | |
3267 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3268 | * @new_dentry: new dentry | |
3269 | * @old_dentry: old dentry | |
3270 | * | |
3271 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
3272 | * Returns 0 otherwise. | |
3273 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
3274 | */ | |
3275 | ||
e2761a11 | 3276 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
3277 | { |
3278 | int result; | |
949854d0 | 3279 | unsigned seq; |
1da177e4 | 3280 | |
e2761a11 OH |
3281 | if (new_dentry == old_dentry) |
3282 | return 1; | |
3283 | ||
e2761a11 | 3284 | do { |
1da177e4 | 3285 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3286 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3287 | /* |
3288 | * Need rcu_readlock to protect against the d_parent trashing | |
3289 | * due to d_move | |
3290 | */ | |
3291 | rcu_read_lock(); | |
e2761a11 | 3292 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 3293 | result = 1; |
e2761a11 OH |
3294 | else |
3295 | result = 0; | |
949854d0 | 3296 | rcu_read_unlock(); |
1da177e4 | 3297 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3298 | |
3299 | return result; | |
3300 | } | |
3301 | ||
db14fc3a | 3302 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3303 | { |
db14fc3a MS |
3304 | struct dentry *root = data; |
3305 | if (dentry != root) { | |
3306 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3307 | return D_WALK_SKIP; | |
1da177e4 | 3308 | |
01ddc4ed MS |
3309 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3310 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3311 | dentry->d_lockref.count--; | |
3312 | } | |
1da177e4 | 3313 | } |
db14fc3a MS |
3314 | return D_WALK_CONTINUE; |
3315 | } | |
58db63d0 | 3316 | |
db14fc3a MS |
3317 | void d_genocide(struct dentry *parent) |
3318 | { | |
3319 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3320 | } |
3321 | ||
60545d0d | 3322 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3323 | { |
60545d0d AV |
3324 | inode_dec_link_count(inode); |
3325 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
3326 | !hlist_unhashed(&dentry->d_alias) || | |
3327 | !d_unlinked(dentry)); | |
3328 | spin_lock(&dentry->d_parent->d_lock); | |
3329 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3330 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3331 | (unsigned long long)inode->i_ino); | |
3332 | spin_unlock(&dentry->d_lock); | |
3333 | spin_unlock(&dentry->d_parent->d_lock); | |
3334 | d_instantiate(dentry, inode); | |
1da177e4 | 3335 | } |
60545d0d | 3336 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3337 | |
3338 | static __initdata unsigned long dhash_entries; | |
3339 | static int __init set_dhash_entries(char *str) | |
3340 | { | |
3341 | if (!str) | |
3342 | return 0; | |
3343 | dhash_entries = simple_strtoul(str, &str, 0); | |
3344 | return 1; | |
3345 | } | |
3346 | __setup("dhash_entries=", set_dhash_entries); | |
3347 | ||
3348 | static void __init dcache_init_early(void) | |
3349 | { | |
074b8517 | 3350 | unsigned int loop; |
1da177e4 LT |
3351 | |
3352 | /* If hashes are distributed across NUMA nodes, defer | |
3353 | * hash allocation until vmalloc space is available. | |
3354 | */ | |
3355 | if (hashdist) | |
3356 | return; | |
3357 | ||
3358 | dentry_hashtable = | |
3359 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3360 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3361 | dhash_entries, |
3362 | 13, | |
3363 | HASH_EARLY, | |
3364 | &d_hash_shift, | |
3365 | &d_hash_mask, | |
31fe62b9 | 3366 | 0, |
1da177e4 LT |
3367 | 0); |
3368 | ||
074b8517 | 3369 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3370 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3371 | } |
3372 | ||
74bf17cf | 3373 | static void __init dcache_init(void) |
1da177e4 | 3374 | { |
074b8517 | 3375 | unsigned int loop; |
1da177e4 LT |
3376 | |
3377 | /* | |
3378 | * A constructor could be added for stable state like the lists, | |
3379 | * but it is probably not worth it because of the cache nature | |
3380 | * of the dcache. | |
3381 | */ | |
0a31bd5f CL |
3382 | dentry_cache = KMEM_CACHE(dentry, |
3383 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3384 | |
3385 | /* Hash may have been set up in dcache_init_early */ | |
3386 | if (!hashdist) | |
3387 | return; | |
3388 | ||
3389 | dentry_hashtable = | |
3390 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3391 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3392 | dhash_entries, |
3393 | 13, | |
3394 | 0, | |
3395 | &d_hash_shift, | |
3396 | &d_hash_mask, | |
31fe62b9 | 3397 | 0, |
1da177e4 LT |
3398 | 0); |
3399 | ||
074b8517 | 3400 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3401 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3402 | } |
3403 | ||
3404 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3405 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3406 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3407 | |
1da177e4 LT |
3408 | EXPORT_SYMBOL(d_genocide); |
3409 | ||
1da177e4 LT |
3410 | void __init vfs_caches_init_early(void) |
3411 | { | |
3412 | dcache_init_early(); | |
3413 | inode_init_early(); | |
3414 | } | |
3415 | ||
3416 | void __init vfs_caches_init(unsigned long mempages) | |
3417 | { | |
3418 | unsigned long reserve; | |
3419 | ||
3420 | /* Base hash sizes on available memory, with a reserve equal to | |
3421 | 150% of current kernel size */ | |
3422 | ||
3423 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3424 | mempages -= reserve; | |
3425 | ||
3426 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3427 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3428 | |
74bf17cf DC |
3429 | dcache_init(); |
3430 | inode_init(); | |
1da177e4 | 3431 | files_init(mempages); |
74bf17cf | 3432 | mnt_init(); |
1da177e4 LT |
3433 | bdev_cache_init(); |
3434 | chrdev_init(); | |
3435 | } |