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> |
df4c0e36 AR |
41 | #include <linux/kasan.h> |
42 | ||
07f3f05c | 43 | #include "internal.h" |
b2dba1af | 44 | #include "mount.h" |
1da177e4 | 45 | |
789680d1 NP |
46 | /* |
47 | * Usage: | |
873feea0 | 48 | * dcache->d_inode->i_lock protects: |
946e51f2 | 49 | * - i_dentry, d_u.d_alias, d_inode of aliases |
ceb5bdc2 NP |
50 | * dcache_hash_bucket lock protects: |
51 | * - the dcache hash table | |
52 | * s_anon bl list spinlock protects: | |
53 | * - the s_anon list (see __d_drop) | |
19156840 | 54 | * dentry->d_sb->s_dentry_lru_lock protects: |
23044507 NP |
55 | * - the dcache lru lists and counters |
56 | * d_lock protects: | |
57 | * - d_flags | |
58 | * - d_name | |
59 | * - d_lru | |
b7ab39f6 | 60 | * - d_count |
da502956 | 61 | * - d_unhashed() |
2fd6b7f5 NP |
62 | * - d_parent and d_subdirs |
63 | * - childrens' d_child and d_parent | |
946e51f2 | 64 | * - d_u.d_alias, d_inode |
789680d1 NP |
65 | * |
66 | * Ordering: | |
873feea0 | 67 | * dentry->d_inode->i_lock |
b5c84bf6 | 68 | * dentry->d_lock |
19156840 | 69 | * dentry->d_sb->s_dentry_lru_lock |
ceb5bdc2 NP |
70 | * dcache_hash_bucket lock |
71 | * s_anon lock | |
789680d1 | 72 | * |
da502956 NP |
73 | * If there is an ancestor relationship: |
74 | * dentry->d_parent->...->d_parent->d_lock | |
75 | * ... | |
76 | * dentry->d_parent->d_lock | |
77 | * dentry->d_lock | |
78 | * | |
79 | * If no ancestor relationship: | |
789680d1 NP |
80 | * if (dentry1 < dentry2) |
81 | * dentry1->d_lock | |
82 | * dentry2->d_lock | |
83 | */ | |
fa3536cc | 84 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
85 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
86 | ||
74c3cbe3 | 87 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 88 | |
949854d0 | 89 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 90 | |
e18b890b | 91 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 92 | |
1da177e4 LT |
93 | /* |
94 | * This is the single most critical data structure when it comes | |
95 | * to the dcache: the hashtable for lookups. Somebody should try | |
96 | * to make this good - I've just made it work. | |
97 | * | |
98 | * This hash-function tries to avoid losing too many bits of hash | |
99 | * information, yet avoid using a prime hash-size or similar. | |
100 | */ | |
1da177e4 | 101 | |
fa3536cc ED |
102 | static unsigned int d_hash_mask __read_mostly; |
103 | static unsigned int d_hash_shift __read_mostly; | |
ceb5bdc2 | 104 | |
b07ad996 | 105 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 106 | |
8966be90 | 107 | static inline struct hlist_bl_head *d_hash(const struct dentry *parent, |
6d7d1a0d | 108 | unsigned int hash) |
ceb5bdc2 | 109 | { |
6d7d1a0d | 110 | hash += (unsigned long) parent / L1_CACHE_BYTES; |
99d263d4 | 111 | return dentry_hashtable + hash_32(hash, d_hash_shift); |
ceb5bdc2 NP |
112 | } |
113 | ||
1da177e4 LT |
114 | /* Statistics gathering. */ |
115 | struct dentry_stat_t dentry_stat = { | |
116 | .age_limit = 45, | |
117 | }; | |
118 | ||
3942c07c | 119 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 120 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
312d3ca8 CH |
121 | |
122 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
62d36c77 DC |
123 | |
124 | /* | |
125 | * Here we resort to our own counters instead of using generic per-cpu counters | |
126 | * for consistency with what the vfs inode code does. We are expected to harvest | |
127 | * better code and performance by having our own specialized counters. | |
128 | * | |
129 | * Please note that the loop is done over all possible CPUs, not over all online | |
130 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
131 | * on and off. If one of them goes off, we will just keep their counters. | |
132 | * | |
133 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
134 | * please update all vfs counters to match. | |
135 | */ | |
3942c07c | 136 | static long get_nr_dentry(void) |
3e880fb5 NP |
137 | { |
138 | int i; | |
3942c07c | 139 | long sum = 0; |
3e880fb5 NP |
140 | for_each_possible_cpu(i) |
141 | sum += per_cpu(nr_dentry, i); | |
142 | return sum < 0 ? 0 : sum; | |
143 | } | |
144 | ||
62d36c77 DC |
145 | static long get_nr_dentry_unused(void) |
146 | { | |
147 | int i; | |
148 | long sum = 0; | |
149 | for_each_possible_cpu(i) | |
150 | sum += per_cpu(nr_dentry_unused, i); | |
151 | return sum < 0 ? 0 : sum; | |
152 | } | |
153 | ||
1f7e0616 | 154 | int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer, |
312d3ca8 CH |
155 | size_t *lenp, loff_t *ppos) |
156 | { | |
3e880fb5 | 157 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 158 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
3942c07c | 159 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 CH |
160 | } |
161 | #endif | |
162 | ||
5483f18e LT |
163 | /* |
164 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
165 | * The strings are both count bytes long, and count is non-zero. | |
166 | */ | |
e419b4cc LT |
167 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
168 | ||
169 | #include <asm/word-at-a-time.h> | |
170 | /* | |
171 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
172 | * aligned allocation for this particular component. We don't | |
173 | * strictly need the load_unaligned_zeropad() safety, but it | |
174 | * doesn't hurt either. | |
175 | * | |
176 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
177 | * need the careful unaligned handling. | |
178 | */ | |
94753db5 | 179 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 180 | { |
bfcfaa77 | 181 | unsigned long a,b,mask; |
bfcfaa77 LT |
182 | |
183 | for (;;) { | |
12f8ad4b | 184 | a = *(unsigned long *)cs; |
e419b4cc | 185 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
186 | if (tcount < sizeof(unsigned long)) |
187 | break; | |
188 | if (unlikely(a != b)) | |
189 | return 1; | |
190 | cs += sizeof(unsigned long); | |
191 | ct += sizeof(unsigned long); | |
192 | tcount -= sizeof(unsigned long); | |
193 | if (!tcount) | |
194 | return 0; | |
195 | } | |
a5c21dce | 196 | mask = bytemask_from_count(tcount); |
bfcfaa77 | 197 | return unlikely(!!((a ^ b) & mask)); |
e419b4cc LT |
198 | } |
199 | ||
bfcfaa77 | 200 | #else |
e419b4cc | 201 | |
94753db5 | 202 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 203 | { |
5483f18e LT |
204 | do { |
205 | if (*cs != *ct) | |
206 | return 1; | |
207 | cs++; | |
208 | ct++; | |
209 | tcount--; | |
210 | } while (tcount); | |
211 | return 0; | |
212 | } | |
213 | ||
e419b4cc LT |
214 | #endif |
215 | ||
94753db5 LT |
216 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
217 | { | |
6326c71f | 218 | const unsigned char *cs; |
94753db5 LT |
219 | /* |
220 | * Be careful about RCU walk racing with rename: | |
221 | * use ACCESS_ONCE to fetch the name pointer. | |
222 | * | |
223 | * NOTE! Even if a rename will mean that the length | |
224 | * was not loaded atomically, we don't care. The | |
225 | * RCU walk will check the sequence count eventually, | |
226 | * and catch it. And we won't overrun the buffer, | |
227 | * because we're reading the name pointer atomically, | |
228 | * and a dentry name is guaranteed to be properly | |
229 | * terminated with a NUL byte. | |
230 | * | |
231 | * End result: even if 'len' is wrong, we'll exit | |
232 | * early because the data cannot match (there can | |
233 | * be no NUL in the ct/tcount data) | |
234 | */ | |
6326c71f LT |
235 | cs = ACCESS_ONCE(dentry->d_name.name); |
236 | smp_read_barrier_depends(); | |
237 | return dentry_string_cmp(cs, ct, tcount); | |
94753db5 LT |
238 | } |
239 | ||
8d85b484 AV |
240 | struct external_name { |
241 | union { | |
242 | atomic_t count; | |
243 | struct rcu_head head; | |
244 | } u; | |
245 | unsigned char name[]; | |
246 | }; | |
247 | ||
248 | static inline struct external_name *external_name(struct dentry *dentry) | |
249 | { | |
250 | return container_of(dentry->d_name.name, struct external_name, name[0]); | |
251 | } | |
252 | ||
9c82ab9c | 253 | static void __d_free(struct rcu_head *head) |
1da177e4 | 254 | { |
9c82ab9c CH |
255 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
256 | ||
8d85b484 AV |
257 | kmem_cache_free(dentry_cache, dentry); |
258 | } | |
259 | ||
260 | static void __d_free_external(struct rcu_head *head) | |
261 | { | |
262 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
8d85b484 | 263 | kfree(external_name(dentry)); |
1da177e4 LT |
264 | kmem_cache_free(dentry_cache, dentry); |
265 | } | |
266 | ||
810bb172 AV |
267 | static inline int dname_external(const struct dentry *dentry) |
268 | { | |
269 | return dentry->d_name.name != dentry->d_iname; | |
270 | } | |
271 | ||
b4f0354e AV |
272 | static void dentry_free(struct dentry *dentry) |
273 | { | |
946e51f2 | 274 | WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias)); |
8d85b484 AV |
275 | if (unlikely(dname_external(dentry))) { |
276 | struct external_name *p = external_name(dentry); | |
277 | if (likely(atomic_dec_and_test(&p->u.count))) { | |
278 | call_rcu(&dentry->d_u.d_rcu, __d_free_external); | |
279 | return; | |
280 | } | |
281 | } | |
b4f0354e AV |
282 | /* if dentry was never visible to RCU, immediate free is OK */ |
283 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
284 | __d_free(&dentry->d_u.d_rcu); | |
285 | else | |
286 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
287 | } | |
288 | ||
31e6b01f NP |
289 | /** |
290 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
ff5fdb61 | 291 | * @dentry: the target dentry |
31e6b01f NP |
292 | * After this call, in-progress rcu-walk path lookup will fail. This |
293 | * should be called after unhashing, and after changing d_inode (if | |
294 | * the dentry has not already been unhashed). | |
295 | */ | |
296 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
297 | { | |
298 | assert_spin_locked(&dentry->d_lock); | |
299 | /* Go through a barrier */ | |
300 | write_seqcount_barrier(&dentry->d_seq); | |
301 | } | |
302 | ||
1da177e4 LT |
303 | /* |
304 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
305 | * d_iput() operation if defined. Dentry has no refcount |
306 | * and is unhashed. | |
1da177e4 | 307 | */ |
858119e1 | 308 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 309 | __releases(dentry->d_lock) |
873feea0 | 310 | __releases(dentry->d_inode->i_lock) |
1da177e4 LT |
311 | { |
312 | struct inode *inode = dentry->d_inode; | |
313 | if (inode) { | |
314 | dentry->d_inode = NULL; | |
946e51f2 | 315 | hlist_del_init(&dentry->d_u.d_alias); |
1da177e4 | 316 | spin_unlock(&dentry->d_lock); |
873feea0 | 317 | spin_unlock(&inode->i_lock); |
f805fbda LT |
318 | if (!inode->i_nlink) |
319 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
320 | if (dentry->d_op && dentry->d_op->d_iput) |
321 | dentry->d_op->d_iput(dentry, inode); | |
322 | else | |
323 | iput(inode); | |
324 | } else { | |
325 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
326 | } |
327 | } | |
328 | ||
31e6b01f NP |
329 | /* |
330 | * Release the dentry's inode, using the filesystem | |
331 | * d_iput() operation if defined. dentry remains in-use. | |
332 | */ | |
333 | static void dentry_unlink_inode(struct dentry * dentry) | |
334 | __releases(dentry->d_lock) | |
873feea0 | 335 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
336 | { |
337 | struct inode *inode = dentry->d_inode; | |
b18825a7 | 338 | __d_clear_type(dentry); |
31e6b01f | 339 | dentry->d_inode = NULL; |
946e51f2 | 340 | hlist_del_init(&dentry->d_u.d_alias); |
31e6b01f NP |
341 | dentry_rcuwalk_barrier(dentry); |
342 | spin_unlock(&dentry->d_lock); | |
873feea0 | 343 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
344 | if (!inode->i_nlink) |
345 | fsnotify_inoderemove(inode); | |
346 | if (dentry->d_op && dentry->d_op->d_iput) | |
347 | dentry->d_op->d_iput(dentry, inode); | |
348 | else | |
349 | iput(inode); | |
350 | } | |
351 | ||
89dc77bc LT |
352 | /* |
353 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
354 | * is in use - which includes both the "real" per-superblock | |
355 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
356 | * | |
357 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
358 | * on the shrink list (ie not on the superblock LRU list). | |
359 | * | |
360 | * The per-cpu "nr_dentry_unused" counters are updated with | |
361 | * the DCACHE_LRU_LIST bit. | |
362 | * | |
363 | * These helper functions make sure we always follow the | |
364 | * rules. d_lock must be held by the caller. | |
365 | */ | |
366 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
367 | static void d_lru_add(struct dentry *dentry) | |
368 | { | |
369 | D_FLAG_VERIFY(dentry, 0); | |
370 | dentry->d_flags |= DCACHE_LRU_LIST; | |
371 | this_cpu_inc(nr_dentry_unused); | |
372 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
373 | } | |
374 | ||
375 | static void d_lru_del(struct dentry *dentry) | |
376 | { | |
377 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
378 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
379 | this_cpu_dec(nr_dentry_unused); | |
380 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
381 | } | |
382 | ||
383 | static void d_shrink_del(struct dentry *dentry) | |
384 | { | |
385 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
386 | list_del_init(&dentry->d_lru); | |
387 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
388 | this_cpu_dec(nr_dentry_unused); | |
389 | } | |
390 | ||
391 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
392 | { | |
393 | D_FLAG_VERIFY(dentry, 0); | |
394 | list_add(&dentry->d_lru, list); | |
395 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
396 | this_cpu_inc(nr_dentry_unused); | |
397 | } | |
398 | ||
399 | /* | |
400 | * These can only be called under the global LRU lock, ie during the | |
401 | * callback for freeing the LRU list. "isolate" removes it from the | |
402 | * LRU lists entirely, while shrink_move moves it to the indicated | |
403 | * private list. | |
404 | */ | |
3f97b163 | 405 | static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry) |
89dc77bc LT |
406 | { |
407 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
408 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
409 | this_cpu_dec(nr_dentry_unused); | |
3f97b163 | 410 | list_lru_isolate(lru, &dentry->d_lru); |
89dc77bc LT |
411 | } |
412 | ||
3f97b163 VD |
413 | static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry, |
414 | struct list_head *list) | |
89dc77bc LT |
415 | { |
416 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
417 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
3f97b163 | 418 | list_lru_isolate_move(lru, &dentry->d_lru, list); |
89dc77bc LT |
419 | } |
420 | ||
da3bbdd4 | 421 | /* |
f6041567 | 422 | * dentry_lru_(add|del)_list) must be called with d_lock held. |
da3bbdd4 KM |
423 | */ |
424 | static void dentry_lru_add(struct dentry *dentry) | |
425 | { | |
89dc77bc LT |
426 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) |
427 | d_lru_add(dentry); | |
da3bbdd4 KM |
428 | } |
429 | ||
789680d1 NP |
430 | /** |
431 | * d_drop - drop a dentry | |
432 | * @dentry: dentry to drop | |
433 | * | |
434 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
435 | * be found through a VFS lookup any more. Note that this is different from | |
436 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
437 | * possible, giving a successful _negative_ lookup, while d_drop will | |
438 | * just make the cache lookup fail. | |
439 | * | |
440 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
441 | * reason (NFS timeouts or autofs deletes). | |
442 | * | |
443 | * __d_drop requires dentry->d_lock. | |
444 | */ | |
445 | void __d_drop(struct dentry *dentry) | |
446 | { | |
dea3667b | 447 | if (!d_unhashed(dentry)) { |
b61625d2 | 448 | struct hlist_bl_head *b; |
7632e465 BF |
449 | /* |
450 | * Hashed dentries are normally on the dentry hashtable, | |
451 | * with the exception of those newly allocated by | |
452 | * d_obtain_alias, which are always IS_ROOT: | |
453 | */ | |
454 | if (unlikely(IS_ROOT(dentry))) | |
b61625d2 AV |
455 | b = &dentry->d_sb->s_anon; |
456 | else | |
457 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
458 | ||
459 | hlist_bl_lock(b); | |
460 | __hlist_bl_del(&dentry->d_hash); | |
461 | dentry->d_hash.pprev = NULL; | |
462 | hlist_bl_unlock(b); | |
dea3667b | 463 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
464 | } |
465 | } | |
466 | EXPORT_SYMBOL(__d_drop); | |
467 | ||
468 | void d_drop(struct dentry *dentry) | |
469 | { | |
789680d1 NP |
470 | spin_lock(&dentry->d_lock); |
471 | __d_drop(dentry); | |
472 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
473 | } |
474 | EXPORT_SYMBOL(d_drop); | |
475 | ||
e55fd011 | 476 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 477 | { |
41edf278 AV |
478 | struct dentry *parent = NULL; |
479 | bool can_free = true; | |
41edf278 | 480 | if (!IS_ROOT(dentry)) |
77812a1e | 481 | parent = dentry->d_parent; |
31e6b01f | 482 | |
0d98439e LT |
483 | /* |
484 | * The dentry is now unrecoverably dead to the world. | |
485 | */ | |
486 | lockref_mark_dead(&dentry->d_lockref); | |
487 | ||
f0023bc6 | 488 | /* |
f0023bc6 SW |
489 | * inform the fs via d_prune that this dentry is about to be |
490 | * unhashed and destroyed. | |
491 | */ | |
29266201 | 492 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
61572bb1 YZ |
493 | dentry->d_op->d_prune(dentry); |
494 | ||
01b60351 AV |
495 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
496 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
497 | d_lru_del(dentry); | |
01b60351 | 498 | } |
77812a1e NP |
499 | /* if it was on the hash then remove it */ |
500 | __d_drop(dentry); | |
ca5358ef | 501 | __list_del_entry(&dentry->d_child); |
03b3b889 AV |
502 | /* |
503 | * Inform d_walk() that we are no longer attached to the | |
504 | * dentry tree | |
505 | */ | |
506 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
507 | if (parent) | |
508 | spin_unlock(&parent->d_lock); | |
509 | dentry_iput(dentry); | |
510 | /* | |
511 | * dentry_iput drops the locks, at which point nobody (except | |
512 | * transient RCU lookups) can reach this dentry. | |
513 | */ | |
360f5479 | 514 | BUG_ON(dentry->d_lockref.count > 0); |
03b3b889 AV |
515 | this_cpu_dec(nr_dentry); |
516 | if (dentry->d_op && dentry->d_op->d_release) | |
517 | dentry->d_op->d_release(dentry); | |
518 | ||
41edf278 AV |
519 | spin_lock(&dentry->d_lock); |
520 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
521 | dentry->d_flags |= DCACHE_MAY_FREE; | |
522 | can_free = false; | |
523 | } | |
524 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
525 | if (likely(can_free)) |
526 | dentry_free(dentry); | |
e55fd011 AV |
527 | } |
528 | ||
529 | /* | |
530 | * Finish off a dentry we've decided to kill. | |
531 | * dentry->d_lock must be held, returns with it unlocked. | |
532 | * If ref is non-zero, then decrement the refcount too. | |
533 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
534 | */ | |
8cbf74da | 535 | static struct dentry *dentry_kill(struct dentry *dentry) |
e55fd011 AV |
536 | __releases(dentry->d_lock) |
537 | { | |
538 | struct inode *inode = dentry->d_inode; | |
539 | struct dentry *parent = NULL; | |
540 | ||
541 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
542 | goto failed; | |
543 | ||
544 | if (!IS_ROOT(dentry)) { | |
545 | parent = dentry->d_parent; | |
546 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
547 | if (inode) | |
548 | spin_unlock(&inode->i_lock); | |
549 | goto failed; | |
550 | } | |
551 | } | |
552 | ||
553 | __dentry_kill(dentry); | |
03b3b889 | 554 | return parent; |
e55fd011 AV |
555 | |
556 | failed: | |
8cbf74da AV |
557 | spin_unlock(&dentry->d_lock); |
558 | cpu_relax(); | |
e55fd011 | 559 | return dentry; /* try again with same dentry */ |
77812a1e NP |
560 | } |
561 | ||
046b961b AV |
562 | static inline struct dentry *lock_parent(struct dentry *dentry) |
563 | { | |
564 | struct dentry *parent = dentry->d_parent; | |
565 | if (IS_ROOT(dentry)) | |
566 | return NULL; | |
360f5479 | 567 | if (unlikely(dentry->d_lockref.count < 0)) |
c2338f2d | 568 | return NULL; |
046b961b AV |
569 | if (likely(spin_trylock(&parent->d_lock))) |
570 | return parent; | |
046b961b | 571 | rcu_read_lock(); |
c2338f2d | 572 | spin_unlock(&dentry->d_lock); |
046b961b AV |
573 | again: |
574 | parent = ACCESS_ONCE(dentry->d_parent); | |
575 | spin_lock(&parent->d_lock); | |
576 | /* | |
577 | * We can't blindly lock dentry until we are sure | |
578 | * that we won't violate the locking order. | |
579 | * Any changes of dentry->d_parent must have | |
580 | * been done with parent->d_lock held, so | |
581 | * spin_lock() above is enough of a barrier | |
582 | * for checking if it's still our child. | |
583 | */ | |
584 | if (unlikely(parent != dentry->d_parent)) { | |
585 | spin_unlock(&parent->d_lock); | |
586 | goto again; | |
587 | } | |
588 | rcu_read_unlock(); | |
589 | if (parent != dentry) | |
9f12600f | 590 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
046b961b AV |
591 | else |
592 | parent = NULL; | |
593 | return parent; | |
594 | } | |
595 | ||
360f5479 LT |
596 | /* |
597 | * Try to do a lockless dput(), and return whether that was successful. | |
598 | * | |
599 | * If unsuccessful, we return false, having already taken the dentry lock. | |
600 | * | |
601 | * The caller needs to hold the RCU read lock, so that the dentry is | |
602 | * guaranteed to stay around even if the refcount goes down to zero! | |
603 | */ | |
604 | static inline bool fast_dput(struct dentry *dentry) | |
605 | { | |
606 | int ret; | |
607 | unsigned int d_flags; | |
608 | ||
609 | /* | |
610 | * If we have a d_op->d_delete() operation, we sould not | |
611 | * let the dentry count go to zero, so use "put__or_lock". | |
612 | */ | |
613 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) | |
614 | return lockref_put_or_lock(&dentry->d_lockref); | |
615 | ||
616 | /* | |
617 | * .. otherwise, we can try to just decrement the | |
618 | * lockref optimistically. | |
619 | */ | |
620 | ret = lockref_put_return(&dentry->d_lockref); | |
621 | ||
622 | /* | |
623 | * If the lockref_put_return() failed due to the lock being held | |
624 | * by somebody else, the fast path has failed. We will need to | |
625 | * get the lock, and then check the count again. | |
626 | */ | |
627 | if (unlikely(ret < 0)) { | |
628 | spin_lock(&dentry->d_lock); | |
629 | if (dentry->d_lockref.count > 1) { | |
630 | dentry->d_lockref.count--; | |
631 | spin_unlock(&dentry->d_lock); | |
632 | return 1; | |
633 | } | |
634 | return 0; | |
635 | } | |
636 | ||
637 | /* | |
638 | * If we weren't the last ref, we're done. | |
639 | */ | |
640 | if (ret) | |
641 | return 1; | |
642 | ||
643 | /* | |
644 | * Careful, careful. The reference count went down | |
645 | * to zero, but we don't hold the dentry lock, so | |
646 | * somebody else could get it again, and do another | |
647 | * dput(), and we need to not race with that. | |
648 | * | |
649 | * However, there is a very special and common case | |
650 | * where we don't care, because there is nothing to | |
651 | * do: the dentry is still hashed, it does not have | |
652 | * a 'delete' op, and it's referenced and already on | |
653 | * the LRU list. | |
654 | * | |
655 | * NOTE! Since we aren't locked, these values are | |
656 | * not "stable". However, it is sufficient that at | |
657 | * some point after we dropped the reference the | |
658 | * dentry was hashed and the flags had the proper | |
659 | * value. Other dentry users may have re-gotten | |
660 | * a reference to the dentry and change that, but | |
661 | * our work is done - we can leave the dentry | |
662 | * around with a zero refcount. | |
663 | */ | |
664 | smp_rmb(); | |
665 | d_flags = ACCESS_ONCE(dentry->d_flags); | |
666 | d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST; | |
667 | ||
668 | /* Nothing to do? Dropping the reference was all we needed? */ | |
669 | if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry)) | |
670 | return 1; | |
671 | ||
672 | /* | |
673 | * Not the fast normal case? Get the lock. We've already decremented | |
674 | * the refcount, but we'll need to re-check the situation after | |
675 | * getting the lock. | |
676 | */ | |
677 | spin_lock(&dentry->d_lock); | |
678 | ||
679 | /* | |
680 | * Did somebody else grab a reference to it in the meantime, and | |
681 | * we're no longer the last user after all? Alternatively, somebody | |
682 | * else could have killed it and marked it dead. Either way, we | |
683 | * don't need to do anything else. | |
684 | */ | |
685 | if (dentry->d_lockref.count) { | |
686 | spin_unlock(&dentry->d_lock); | |
687 | return 1; | |
688 | } | |
689 | ||
690 | /* | |
691 | * Re-get the reference we optimistically dropped. We hold the | |
692 | * lock, and we just tested that it was zero, so we can just | |
693 | * set it to 1. | |
694 | */ | |
695 | dentry->d_lockref.count = 1; | |
696 | return 0; | |
697 | } | |
698 | ||
699 | ||
1da177e4 LT |
700 | /* |
701 | * This is dput | |
702 | * | |
703 | * This is complicated by the fact that we do not want to put | |
704 | * dentries that are no longer on any hash chain on the unused | |
705 | * list: we'd much rather just get rid of them immediately. | |
706 | * | |
707 | * However, that implies that we have to traverse the dentry | |
708 | * tree upwards to the parents which might _also_ now be | |
709 | * scheduled for deletion (it may have been only waiting for | |
710 | * its last child to go away). | |
711 | * | |
712 | * This tail recursion is done by hand as we don't want to depend | |
713 | * on the compiler to always get this right (gcc generally doesn't). | |
714 | * Real recursion would eat up our stack space. | |
715 | */ | |
716 | ||
717 | /* | |
718 | * dput - release a dentry | |
719 | * @dentry: dentry to release | |
720 | * | |
721 | * Release a dentry. This will drop the usage count and if appropriate | |
722 | * call the dentry unlink method as well as removing it from the queues and | |
723 | * releasing its resources. If the parent dentries were scheduled for release | |
724 | * they too may now get deleted. | |
1da177e4 | 725 | */ |
1da177e4 LT |
726 | void dput(struct dentry *dentry) |
727 | { | |
8aab6a27 | 728 | if (unlikely(!dentry)) |
1da177e4 LT |
729 | return; |
730 | ||
731 | repeat: | |
360f5479 LT |
732 | rcu_read_lock(); |
733 | if (likely(fast_dput(dentry))) { | |
734 | rcu_read_unlock(); | |
1da177e4 | 735 | return; |
360f5479 LT |
736 | } |
737 | ||
738 | /* Slow case: now with the dentry lock held */ | |
739 | rcu_read_unlock(); | |
1da177e4 | 740 | |
8aab6a27 LT |
741 | /* Unreachable? Get rid of it */ |
742 | if (unlikely(d_unhashed(dentry))) | |
743 | goto kill_it; | |
744 | ||
745 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
1da177e4 | 746 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 747 | goto kill_it; |
1da177e4 | 748 | } |
265ac902 | 749 | |
358eec18 LT |
750 | if (!(dentry->d_flags & DCACHE_REFERENCED)) |
751 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 752 | dentry_lru_add(dentry); |
265ac902 | 753 | |
98474236 | 754 | dentry->d_lockref.count--; |
61f3dee4 | 755 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
756 | return; |
757 | ||
d52b9086 | 758 | kill_it: |
8cbf74da | 759 | dentry = dentry_kill(dentry); |
d52b9086 MS |
760 | if (dentry) |
761 | goto repeat; | |
1da177e4 | 762 | } |
ec4f8605 | 763 | EXPORT_SYMBOL(dput); |
1da177e4 | 764 | |
1da177e4 | 765 | |
b5c84bf6 | 766 | /* This must be called with d_lock held */ |
dc0474be | 767 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 768 | { |
98474236 | 769 | dentry->d_lockref.count++; |
23044507 NP |
770 | } |
771 | ||
dc0474be | 772 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 773 | { |
98474236 | 774 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
775 | } |
776 | ||
b7ab39f6 NP |
777 | struct dentry *dget_parent(struct dentry *dentry) |
778 | { | |
df3d0bbc | 779 | int gotref; |
b7ab39f6 NP |
780 | struct dentry *ret; |
781 | ||
df3d0bbc WL |
782 | /* |
783 | * Do optimistic parent lookup without any | |
784 | * locking. | |
785 | */ | |
786 | rcu_read_lock(); | |
787 | ret = ACCESS_ONCE(dentry->d_parent); | |
788 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
789 | rcu_read_unlock(); | |
790 | if (likely(gotref)) { | |
791 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
792 | return ret; | |
793 | dput(ret); | |
794 | } | |
795 | ||
b7ab39f6 | 796 | repeat: |
a734eb45 NP |
797 | /* |
798 | * Don't need rcu_dereference because we re-check it was correct under | |
799 | * the lock. | |
800 | */ | |
801 | rcu_read_lock(); | |
b7ab39f6 | 802 | ret = dentry->d_parent; |
a734eb45 NP |
803 | spin_lock(&ret->d_lock); |
804 | if (unlikely(ret != dentry->d_parent)) { | |
805 | spin_unlock(&ret->d_lock); | |
806 | rcu_read_unlock(); | |
b7ab39f6 NP |
807 | goto repeat; |
808 | } | |
a734eb45 | 809 | rcu_read_unlock(); |
98474236 WL |
810 | BUG_ON(!ret->d_lockref.count); |
811 | ret->d_lockref.count++; | |
b7ab39f6 | 812 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
813 | return ret; |
814 | } | |
815 | EXPORT_SYMBOL(dget_parent); | |
816 | ||
1da177e4 LT |
817 | /** |
818 | * d_find_alias - grab a hashed alias of inode | |
819 | * @inode: inode in question | |
1da177e4 LT |
820 | * |
821 | * If inode has a hashed alias, or is a directory and has any alias, | |
822 | * acquire the reference to alias and return it. Otherwise return NULL. | |
823 | * Notice that if inode is a directory there can be only one alias and | |
824 | * it can be unhashed only if it has no children, or if it is the root | |
3ccb354d EB |
825 | * of a filesystem, or if the directory was renamed and d_revalidate |
826 | * was the first vfs operation to notice. | |
1da177e4 | 827 | * |
21c0d8fd | 828 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
52ed46f0 | 829 | * any other hashed alias over that one. |
1da177e4 | 830 | */ |
52ed46f0 | 831 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 832 | { |
da502956 | 833 | struct dentry *alias, *discon_alias; |
1da177e4 | 834 | |
da502956 NP |
835 | again: |
836 | discon_alias = NULL; | |
946e51f2 | 837 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
da502956 | 838 | spin_lock(&alias->d_lock); |
1da177e4 | 839 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 840 | if (IS_ROOT(alias) && |
da502956 | 841 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 842 | discon_alias = alias; |
52ed46f0 | 843 | } else { |
dc0474be | 844 | __dget_dlock(alias); |
da502956 NP |
845 | spin_unlock(&alias->d_lock); |
846 | return alias; | |
847 | } | |
848 | } | |
849 | spin_unlock(&alias->d_lock); | |
850 | } | |
851 | if (discon_alias) { | |
852 | alias = discon_alias; | |
853 | spin_lock(&alias->d_lock); | |
854 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
8d80d7da BF |
855 | __dget_dlock(alias); |
856 | spin_unlock(&alias->d_lock); | |
857 | return alias; | |
1da177e4 | 858 | } |
da502956 NP |
859 | spin_unlock(&alias->d_lock); |
860 | goto again; | |
1da177e4 | 861 | } |
da502956 | 862 | return NULL; |
1da177e4 LT |
863 | } |
864 | ||
da502956 | 865 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 866 | { |
214fda1f DH |
867 | struct dentry *de = NULL; |
868 | ||
b3d9b7a3 | 869 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 870 | spin_lock(&inode->i_lock); |
52ed46f0 | 871 | de = __d_find_alias(inode); |
873feea0 | 872 | spin_unlock(&inode->i_lock); |
214fda1f | 873 | } |
1da177e4 LT |
874 | return de; |
875 | } | |
ec4f8605 | 876 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
877 | |
878 | /* | |
879 | * Try to kill dentries associated with this inode. | |
880 | * WARNING: you must own a reference to inode. | |
881 | */ | |
882 | void d_prune_aliases(struct inode *inode) | |
883 | { | |
0cdca3f9 | 884 | struct dentry *dentry; |
1da177e4 | 885 | restart: |
873feea0 | 886 | spin_lock(&inode->i_lock); |
946e51f2 | 887 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { |
1da177e4 | 888 | spin_lock(&dentry->d_lock); |
98474236 | 889 | if (!dentry->d_lockref.count) { |
29355c39 AV |
890 | struct dentry *parent = lock_parent(dentry); |
891 | if (likely(!dentry->d_lockref.count)) { | |
892 | __dentry_kill(dentry); | |
4a7795d3 | 893 | dput(parent); |
29355c39 AV |
894 | goto restart; |
895 | } | |
896 | if (parent) | |
897 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
898 | } |
899 | spin_unlock(&dentry->d_lock); | |
900 | } | |
873feea0 | 901 | spin_unlock(&inode->i_lock); |
1da177e4 | 902 | } |
ec4f8605 | 903 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 904 | |
3049cfe2 | 905 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 906 | { |
5c47e6d0 | 907 | struct dentry *dentry, *parent; |
da3bbdd4 | 908 | |
60942f2f | 909 | while (!list_empty(list)) { |
ff2fde99 | 910 | struct inode *inode; |
60942f2f | 911 | dentry = list_entry(list->prev, struct dentry, d_lru); |
ec33679d | 912 | spin_lock(&dentry->d_lock); |
046b961b AV |
913 | parent = lock_parent(dentry); |
914 | ||
dd1f6b2e DC |
915 | /* |
916 | * The dispose list is isolated and dentries are not accounted | |
917 | * to the LRU here, so we can simply remove it from the list | |
918 | * here regardless of whether it is referenced or not. | |
919 | */ | |
89dc77bc | 920 | d_shrink_del(dentry); |
dd1f6b2e | 921 | |
1da177e4 LT |
922 | /* |
923 | * We found an inuse dentry which was not removed from | |
dd1f6b2e | 924 | * the LRU because of laziness during lookup. Do not free it. |
1da177e4 | 925 | */ |
360f5479 | 926 | if (dentry->d_lockref.count > 0) { |
da3bbdd4 | 927 | spin_unlock(&dentry->d_lock); |
046b961b AV |
928 | if (parent) |
929 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
930 | continue; |
931 | } | |
77812a1e | 932 | |
64fd72e0 AV |
933 | |
934 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) { | |
935 | bool can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
936 | spin_unlock(&dentry->d_lock); | |
046b961b AV |
937 | if (parent) |
938 | spin_unlock(&parent->d_lock); | |
64fd72e0 AV |
939 | if (can_free) |
940 | dentry_free(dentry); | |
941 | continue; | |
942 | } | |
943 | ||
ff2fde99 AV |
944 | inode = dentry->d_inode; |
945 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
89dc77bc | 946 | d_shrink_add(dentry, list); |
dd1f6b2e | 947 | spin_unlock(&dentry->d_lock); |
046b961b AV |
948 | if (parent) |
949 | spin_unlock(&parent->d_lock); | |
5c47e6d0 | 950 | continue; |
dd1f6b2e | 951 | } |
ff2fde99 | 952 | |
ff2fde99 | 953 | __dentry_kill(dentry); |
046b961b | 954 | |
5c47e6d0 AV |
955 | /* |
956 | * We need to prune ancestors too. This is necessary to prevent | |
957 | * quadratic behavior of shrink_dcache_parent(), but is also | |
958 | * expected to be beneficial in reducing dentry cache | |
959 | * fragmentation. | |
960 | */ | |
961 | dentry = parent; | |
b2b80195 AV |
962 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) { |
963 | parent = lock_parent(dentry); | |
964 | if (dentry->d_lockref.count != 1) { | |
965 | dentry->d_lockref.count--; | |
966 | spin_unlock(&dentry->d_lock); | |
967 | if (parent) | |
968 | spin_unlock(&parent->d_lock); | |
969 | break; | |
970 | } | |
971 | inode = dentry->d_inode; /* can't be NULL */ | |
972 | if (unlikely(!spin_trylock(&inode->i_lock))) { | |
973 | spin_unlock(&dentry->d_lock); | |
974 | if (parent) | |
975 | spin_unlock(&parent->d_lock); | |
976 | cpu_relax(); | |
977 | continue; | |
978 | } | |
979 | __dentry_kill(dentry); | |
980 | dentry = parent; | |
981 | } | |
da3bbdd4 | 982 | } |
3049cfe2 CH |
983 | } |
984 | ||
3f97b163 VD |
985 | static enum lru_status dentry_lru_isolate(struct list_head *item, |
986 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
f6041567 DC |
987 | { |
988 | struct list_head *freeable = arg; | |
989 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
990 | ||
991 | ||
992 | /* | |
993 | * we are inverting the lru lock/dentry->d_lock here, | |
994 | * so use a trylock. If we fail to get the lock, just skip | |
995 | * it | |
996 | */ | |
997 | if (!spin_trylock(&dentry->d_lock)) | |
998 | return LRU_SKIP; | |
999 | ||
1000 | /* | |
1001 | * Referenced dentries are still in use. If they have active | |
1002 | * counts, just remove them from the LRU. Otherwise give them | |
1003 | * another pass through the LRU. | |
1004 | */ | |
1005 | if (dentry->d_lockref.count) { | |
3f97b163 | 1006 | d_lru_isolate(lru, dentry); |
f6041567 DC |
1007 | spin_unlock(&dentry->d_lock); |
1008 | return LRU_REMOVED; | |
1009 | } | |
1010 | ||
1011 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
1012 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
1013 | spin_unlock(&dentry->d_lock); | |
1014 | ||
1015 | /* | |
1016 | * The list move itself will be made by the common LRU code. At | |
1017 | * this point, we've dropped the dentry->d_lock but keep the | |
1018 | * lru lock. This is safe to do, since every list movement is | |
1019 | * protected by the lru lock even if both locks are held. | |
1020 | * | |
1021 | * This is guaranteed by the fact that all LRU management | |
1022 | * functions are intermediated by the LRU API calls like | |
1023 | * list_lru_add and list_lru_del. List movement in this file | |
1024 | * only ever occur through this functions or through callbacks | |
1025 | * like this one, that are called from the LRU API. | |
1026 | * | |
1027 | * The only exceptions to this are functions like | |
1028 | * shrink_dentry_list, and code that first checks for the | |
1029 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
1030 | * operating only with stack provided lists after they are | |
1031 | * properly isolated from the main list. It is thus, always a | |
1032 | * local access. | |
1033 | */ | |
1034 | return LRU_ROTATE; | |
1035 | } | |
1036 | ||
3f97b163 | 1037 | d_lru_shrink_move(lru, dentry, freeable); |
f6041567 DC |
1038 | spin_unlock(&dentry->d_lock); |
1039 | ||
1040 | return LRU_REMOVED; | |
1041 | } | |
1042 | ||
3049cfe2 | 1043 | /** |
b48f03b3 DC |
1044 | * prune_dcache_sb - shrink the dcache |
1045 | * @sb: superblock | |
503c358c | 1046 | * @sc: shrink control, passed to list_lru_shrink_walk() |
b48f03b3 | 1047 | * |
503c358c VD |
1048 | * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This |
1049 | * is done when we need more memory and called from the superblock shrinker | |
b48f03b3 | 1050 | * function. |
3049cfe2 | 1051 | * |
b48f03b3 DC |
1052 | * This function may fail to free any resources if all the dentries are in |
1053 | * use. | |
3049cfe2 | 1054 | */ |
503c358c | 1055 | long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc) |
3049cfe2 | 1056 | { |
f6041567 DC |
1057 | LIST_HEAD(dispose); |
1058 | long freed; | |
3049cfe2 | 1059 | |
503c358c VD |
1060 | freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc, |
1061 | dentry_lru_isolate, &dispose); | |
f6041567 | 1062 | shrink_dentry_list(&dispose); |
0a234c6d | 1063 | return freed; |
da3bbdd4 | 1064 | } |
23044507 | 1065 | |
4e717f5c | 1066 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
3f97b163 | 1067 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
dd1f6b2e | 1068 | { |
4e717f5c GC |
1069 | struct list_head *freeable = arg; |
1070 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 1071 | |
4e717f5c GC |
1072 | /* |
1073 | * we are inverting the lru lock/dentry->d_lock here, | |
1074 | * so use a trylock. If we fail to get the lock, just skip | |
1075 | * it | |
1076 | */ | |
1077 | if (!spin_trylock(&dentry->d_lock)) | |
1078 | return LRU_SKIP; | |
1079 | ||
3f97b163 | 1080 | d_lru_shrink_move(lru, dentry, freeable); |
4e717f5c | 1081 | spin_unlock(&dentry->d_lock); |
ec33679d | 1082 | |
4e717f5c | 1083 | return LRU_REMOVED; |
da3bbdd4 KM |
1084 | } |
1085 | ||
4e717f5c | 1086 | |
1da177e4 LT |
1087 | /** |
1088 | * shrink_dcache_sb - shrink dcache for a superblock | |
1089 | * @sb: superblock | |
1090 | * | |
3049cfe2 CH |
1091 | * Shrink the dcache for the specified super block. This is used to free |
1092 | * the dcache before unmounting a file system. | |
1da177e4 | 1093 | */ |
3049cfe2 | 1094 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1095 | { |
4e717f5c GC |
1096 | long freed; |
1097 | ||
1098 | do { | |
1099 | LIST_HEAD(dispose); | |
1100 | ||
1101 | freed = list_lru_walk(&sb->s_dentry_lru, | |
1102 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); | |
3049cfe2 | 1103 | |
4e717f5c GC |
1104 | this_cpu_sub(nr_dentry_unused, freed); |
1105 | shrink_dentry_list(&dispose); | |
1106 | } while (freed > 0); | |
1da177e4 | 1107 | } |
ec4f8605 | 1108 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1109 | |
db14fc3a MS |
1110 | /** |
1111 | * enum d_walk_ret - action to talke during tree walk | |
1112 | * @D_WALK_CONTINUE: contrinue walk | |
1113 | * @D_WALK_QUIT: quit walk | |
1114 | * @D_WALK_NORETRY: quit when retry is needed | |
1115 | * @D_WALK_SKIP: skip this dentry and its children | |
1116 | */ | |
1117 | enum d_walk_ret { | |
1118 | D_WALK_CONTINUE, | |
1119 | D_WALK_QUIT, | |
1120 | D_WALK_NORETRY, | |
1121 | D_WALK_SKIP, | |
1122 | }; | |
c826cb7d | 1123 | |
1da177e4 | 1124 | /** |
db14fc3a MS |
1125 | * d_walk - walk the dentry tree |
1126 | * @parent: start of walk | |
1127 | * @data: data passed to @enter() and @finish() | |
1128 | * @enter: callback when first entering the dentry | |
1129 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1130 | * |
db14fc3a | 1131 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1132 | */ |
db14fc3a MS |
1133 | static void d_walk(struct dentry *parent, void *data, |
1134 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1135 | void (*finish)(void *)) | |
1da177e4 | 1136 | { |
949854d0 | 1137 | struct dentry *this_parent; |
1da177e4 | 1138 | struct list_head *next; |
48f5ec21 | 1139 | unsigned seq = 0; |
db14fc3a MS |
1140 | enum d_walk_ret ret; |
1141 | bool retry = true; | |
949854d0 | 1142 | |
58db63d0 | 1143 | again: |
48f5ec21 | 1144 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1145 | this_parent = parent; |
2fd6b7f5 | 1146 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1147 | |
1148 | ret = enter(data, this_parent); | |
1149 | switch (ret) { | |
1150 | case D_WALK_CONTINUE: | |
1151 | break; | |
1152 | case D_WALK_QUIT: | |
1153 | case D_WALK_SKIP: | |
1154 | goto out_unlock; | |
1155 | case D_WALK_NORETRY: | |
1156 | retry = false; | |
1157 | break; | |
1158 | } | |
1da177e4 LT |
1159 | repeat: |
1160 | next = this_parent->d_subdirs.next; | |
1161 | resume: | |
1162 | while (next != &this_parent->d_subdirs) { | |
1163 | struct list_head *tmp = next; | |
946e51f2 | 1164 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1da177e4 | 1165 | next = tmp->next; |
2fd6b7f5 NP |
1166 | |
1167 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
db14fc3a MS |
1168 | |
1169 | ret = enter(data, dentry); | |
1170 | switch (ret) { | |
1171 | case D_WALK_CONTINUE: | |
1172 | break; | |
1173 | case D_WALK_QUIT: | |
2fd6b7f5 | 1174 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1175 | goto out_unlock; |
1176 | case D_WALK_NORETRY: | |
1177 | retry = false; | |
1178 | break; | |
1179 | case D_WALK_SKIP: | |
1180 | spin_unlock(&dentry->d_lock); | |
1181 | continue; | |
2fd6b7f5 | 1182 | } |
db14fc3a | 1183 | |
1da177e4 | 1184 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1185 | spin_unlock(&this_parent->d_lock); |
1186 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1187 | this_parent = dentry; |
2fd6b7f5 | 1188 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1189 | goto repeat; |
1190 | } | |
2fd6b7f5 | 1191 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1192 | } |
1193 | /* | |
1194 | * All done at this level ... ascend and resume the search. | |
1195 | */ | |
ca5358ef AV |
1196 | rcu_read_lock(); |
1197 | ascend: | |
1da177e4 | 1198 | if (this_parent != parent) { |
c826cb7d | 1199 | struct dentry *child = this_parent; |
31dec132 AV |
1200 | this_parent = child->d_parent; |
1201 | ||
31dec132 AV |
1202 | spin_unlock(&child->d_lock); |
1203 | spin_lock(&this_parent->d_lock); | |
1204 | ||
ca5358ef AV |
1205 | /* might go back up the wrong parent if we have had a rename. */ |
1206 | if (need_seqretry(&rename_lock, seq)) | |
949854d0 | 1207 | goto rename_retry; |
ca5358ef AV |
1208 | next = child->d_child.next; |
1209 | while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) { | |
1210 | if (next == &this_parent->d_subdirs) | |
1211 | goto ascend; | |
1212 | child = list_entry(next, struct dentry, d_child); | |
1213 | next = next->next; | |
31dec132 AV |
1214 | } |
1215 | rcu_read_unlock(); | |
1da177e4 LT |
1216 | goto resume; |
1217 | } | |
ca5358ef | 1218 | if (need_seqretry(&rename_lock, seq)) |
949854d0 | 1219 | goto rename_retry; |
ca5358ef | 1220 | rcu_read_unlock(); |
db14fc3a MS |
1221 | if (finish) |
1222 | finish(data); | |
1223 | ||
1224 | out_unlock: | |
1225 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1226 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1227 | return; |
58db63d0 NP |
1228 | |
1229 | rename_retry: | |
ca5358ef AV |
1230 | spin_unlock(&this_parent->d_lock); |
1231 | rcu_read_unlock(); | |
1232 | BUG_ON(seq & 1); | |
db14fc3a MS |
1233 | if (!retry) |
1234 | return; | |
48f5ec21 | 1235 | seq = 1; |
58db63d0 | 1236 | goto again; |
1da177e4 | 1237 | } |
db14fc3a MS |
1238 | |
1239 | /* | |
1240 | * Search for at least 1 mount point in the dentry's subdirs. | |
1241 | * We descend to the next level whenever the d_subdirs | |
1242 | * list is non-empty and continue searching. | |
1243 | */ | |
1244 | ||
db14fc3a MS |
1245 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) |
1246 | { | |
1247 | int *ret = data; | |
1248 | if (d_mountpoint(dentry)) { | |
1249 | *ret = 1; | |
1250 | return D_WALK_QUIT; | |
1251 | } | |
1252 | return D_WALK_CONTINUE; | |
1253 | } | |
1254 | ||
69c88dc7 RD |
1255 | /** |
1256 | * have_submounts - check for mounts over a dentry | |
1257 | * @parent: dentry to check. | |
1258 | * | |
1259 | * Return true if the parent or its subdirectories contain | |
1260 | * a mount point | |
1261 | */ | |
db14fc3a MS |
1262 | int have_submounts(struct dentry *parent) |
1263 | { | |
1264 | int ret = 0; | |
1265 | ||
1266 | d_walk(parent, &ret, check_mount, NULL); | |
1267 | ||
1268 | return ret; | |
1269 | } | |
ec4f8605 | 1270 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1271 | |
eed81007 MS |
1272 | /* |
1273 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1274 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1275 | * subtree can become unreachable). | |
1276 | * | |
1ffe46d1 | 1277 | * Only one of d_invalidate() and d_set_mounted() must succeed. For |
eed81007 MS |
1278 | * this reason take rename_lock and d_lock on dentry and ancestors. |
1279 | */ | |
1280 | int d_set_mounted(struct dentry *dentry) | |
1281 | { | |
1282 | struct dentry *p; | |
1283 | int ret = -ENOENT; | |
1284 | write_seqlock(&rename_lock); | |
1285 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1ffe46d1 | 1286 | /* Need exclusion wrt. d_invalidate() */ |
eed81007 MS |
1287 | spin_lock(&p->d_lock); |
1288 | if (unlikely(d_unhashed(p))) { | |
1289 | spin_unlock(&p->d_lock); | |
1290 | goto out; | |
1291 | } | |
1292 | spin_unlock(&p->d_lock); | |
1293 | } | |
1294 | spin_lock(&dentry->d_lock); | |
1295 | if (!d_unlinked(dentry)) { | |
1296 | dentry->d_flags |= DCACHE_MOUNTED; | |
1297 | ret = 0; | |
1298 | } | |
1299 | spin_unlock(&dentry->d_lock); | |
1300 | out: | |
1301 | write_sequnlock(&rename_lock); | |
1302 | return ret; | |
1303 | } | |
1304 | ||
1da177e4 | 1305 | /* |
fd517909 | 1306 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1307 | * and move any unused dentries to the end of the unused |
1308 | * list for prune_dcache(). We descend to the next level | |
1309 | * whenever the d_subdirs list is non-empty and continue | |
1310 | * searching. | |
1311 | * | |
1312 | * It returns zero iff there are no unused children, | |
1313 | * otherwise it returns the number of children moved to | |
1314 | * the end of the unused list. This may not be the total | |
1315 | * number of unused children, because select_parent can | |
1316 | * drop the lock and return early due to latency | |
1317 | * constraints. | |
1318 | */ | |
1da177e4 | 1319 | |
db14fc3a MS |
1320 | struct select_data { |
1321 | struct dentry *start; | |
1322 | struct list_head dispose; | |
1323 | int found; | |
1324 | }; | |
23044507 | 1325 | |
db14fc3a MS |
1326 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1327 | { | |
1328 | struct select_data *data = _data; | |
1329 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1330 | |
db14fc3a MS |
1331 | if (data->start == dentry) |
1332 | goto out; | |
2fd6b7f5 | 1333 | |
fe91522a | 1334 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1335 | data->found++; |
fe91522a AV |
1336 | } else { |
1337 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1338 | d_lru_del(dentry); | |
1339 | if (!dentry->d_lockref.count) { | |
1340 | d_shrink_add(dentry, &data->dispose); | |
1341 | data->found++; | |
1342 | } | |
1da177e4 | 1343 | } |
db14fc3a MS |
1344 | /* |
1345 | * We can return to the caller if we have found some (this | |
1346 | * ensures forward progress). We'll be coming back to find | |
1347 | * the rest. | |
1348 | */ | |
fe91522a AV |
1349 | if (!list_empty(&data->dispose)) |
1350 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1351 | out: |
db14fc3a | 1352 | return ret; |
1da177e4 LT |
1353 | } |
1354 | ||
1355 | /** | |
1356 | * shrink_dcache_parent - prune dcache | |
1357 | * @parent: parent of entries to prune | |
1358 | * | |
1359 | * Prune the dcache to remove unused children of the parent dentry. | |
1360 | */ | |
db14fc3a | 1361 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1362 | { |
db14fc3a MS |
1363 | for (;;) { |
1364 | struct select_data data; | |
1da177e4 | 1365 | |
db14fc3a MS |
1366 | INIT_LIST_HEAD(&data.dispose); |
1367 | data.start = parent; | |
1368 | data.found = 0; | |
1369 | ||
1370 | d_walk(parent, &data, select_collect, NULL); | |
1371 | if (!data.found) | |
1372 | break; | |
1373 | ||
1374 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1375 | cond_resched(); |
1376 | } | |
1da177e4 | 1377 | } |
ec4f8605 | 1378 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1379 | |
9c8c10e2 | 1380 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1381 | { |
9c8c10e2 AV |
1382 | /* it has busy descendents; complain about those instead */ |
1383 | if (!list_empty(&dentry->d_subdirs)) | |
1384 | return D_WALK_CONTINUE; | |
42c32608 | 1385 | |
9c8c10e2 AV |
1386 | /* root with refcount 1 is fine */ |
1387 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1388 | return D_WALK_CONTINUE; | |
1389 | ||
1390 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1391 | " still in use (%d) [unmount of %s %s]\n", | |
42c32608 AV |
1392 | dentry, |
1393 | dentry->d_inode ? | |
1394 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1395 | dentry, |
42c32608 AV |
1396 | dentry->d_lockref.count, |
1397 | dentry->d_sb->s_type->name, | |
1398 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1399 | WARN_ON(1); |
1400 | return D_WALK_CONTINUE; | |
1401 | } | |
1402 | ||
1403 | static void do_one_tree(struct dentry *dentry) | |
1404 | { | |
1405 | shrink_dcache_parent(dentry); | |
1406 | d_walk(dentry, dentry, umount_check, NULL); | |
1407 | d_drop(dentry); | |
1408 | dput(dentry); | |
42c32608 AV |
1409 | } |
1410 | ||
1411 | /* | |
1412 | * destroy the dentries attached to a superblock on unmounting | |
1413 | */ | |
1414 | void shrink_dcache_for_umount(struct super_block *sb) | |
1415 | { | |
1416 | struct dentry *dentry; | |
1417 | ||
9c8c10e2 | 1418 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1419 | |
1420 | dentry = sb->s_root; | |
1421 | sb->s_root = NULL; | |
9c8c10e2 | 1422 | do_one_tree(dentry); |
42c32608 AV |
1423 | |
1424 | while (!hlist_bl_empty(&sb->s_anon)) { | |
9c8c10e2 AV |
1425 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash)); |
1426 | do_one_tree(dentry); | |
42c32608 AV |
1427 | } |
1428 | } | |
1429 | ||
8ed936b5 EB |
1430 | struct detach_data { |
1431 | struct select_data select; | |
1432 | struct dentry *mountpoint; | |
1433 | }; | |
1434 | static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry) | |
848ac114 | 1435 | { |
8ed936b5 | 1436 | struct detach_data *data = _data; |
848ac114 MS |
1437 | |
1438 | if (d_mountpoint(dentry)) { | |
8ed936b5 EB |
1439 | __dget_dlock(dentry); |
1440 | data->mountpoint = dentry; | |
848ac114 MS |
1441 | return D_WALK_QUIT; |
1442 | } | |
1443 | ||
8ed936b5 | 1444 | return select_collect(&data->select, dentry); |
848ac114 MS |
1445 | } |
1446 | ||
1447 | static void check_and_drop(void *_data) | |
1448 | { | |
8ed936b5 | 1449 | struct detach_data *data = _data; |
848ac114 | 1450 | |
8ed936b5 EB |
1451 | if (!data->mountpoint && !data->select.found) |
1452 | __d_drop(data->select.start); | |
848ac114 MS |
1453 | } |
1454 | ||
1455 | /** | |
1ffe46d1 EB |
1456 | * d_invalidate - detach submounts, prune dcache, and drop |
1457 | * @dentry: dentry to invalidate (aka detach, prune and drop) | |
1458 | * | |
1ffe46d1 | 1459 | * no dcache lock. |
848ac114 | 1460 | * |
8ed936b5 EB |
1461 | * The final d_drop is done as an atomic operation relative to |
1462 | * rename_lock ensuring there are no races with d_set_mounted. This | |
1463 | * ensures there are no unhashed dentries on the path to a mountpoint. | |
848ac114 | 1464 | */ |
5542aa2f | 1465 | void d_invalidate(struct dentry *dentry) |
848ac114 | 1466 | { |
1ffe46d1 EB |
1467 | /* |
1468 | * If it's already been dropped, return OK. | |
1469 | */ | |
1470 | spin_lock(&dentry->d_lock); | |
1471 | if (d_unhashed(dentry)) { | |
1472 | spin_unlock(&dentry->d_lock); | |
5542aa2f | 1473 | return; |
1ffe46d1 EB |
1474 | } |
1475 | spin_unlock(&dentry->d_lock); | |
1476 | ||
848ac114 MS |
1477 | /* Negative dentries can be dropped without further checks */ |
1478 | if (!dentry->d_inode) { | |
1479 | d_drop(dentry); | |
5542aa2f | 1480 | return; |
848ac114 MS |
1481 | } |
1482 | ||
1483 | for (;;) { | |
8ed936b5 | 1484 | struct detach_data data; |
848ac114 | 1485 | |
8ed936b5 EB |
1486 | data.mountpoint = NULL; |
1487 | INIT_LIST_HEAD(&data.select.dispose); | |
1488 | data.select.start = dentry; | |
1489 | data.select.found = 0; | |
1490 | ||
1491 | d_walk(dentry, &data, detach_and_collect, check_and_drop); | |
848ac114 | 1492 | |
8ed936b5 EB |
1493 | if (data.select.found) |
1494 | shrink_dentry_list(&data.select.dispose); | |
848ac114 | 1495 | |
8ed936b5 EB |
1496 | if (data.mountpoint) { |
1497 | detach_mounts(data.mountpoint); | |
1498 | dput(data.mountpoint); | |
1499 | } | |
848ac114 | 1500 | |
8ed936b5 | 1501 | if (!data.mountpoint && !data.select.found) |
848ac114 MS |
1502 | break; |
1503 | ||
1504 | cond_resched(); | |
1505 | } | |
848ac114 | 1506 | } |
1ffe46d1 | 1507 | EXPORT_SYMBOL(d_invalidate); |
848ac114 | 1508 | |
1da177e4 | 1509 | /** |
a4464dbc AV |
1510 | * __d_alloc - allocate a dcache entry |
1511 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1512 | * @name: qstr of the name |
1513 | * | |
1514 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1515 | * available. On a success the dentry is returned. The name passed in is | |
1516 | * copied and the copy passed in may be reused after this call. | |
1517 | */ | |
1518 | ||
a4464dbc | 1519 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1520 | { |
1521 | struct dentry *dentry; | |
1522 | char *dname; | |
1523 | ||
e12ba74d | 1524 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1525 | if (!dentry) |
1526 | return NULL; | |
1527 | ||
6326c71f LT |
1528 | /* |
1529 | * We guarantee that the inline name is always NUL-terminated. | |
1530 | * This way the memcpy() done by the name switching in rename | |
1531 | * will still always have a NUL at the end, even if we might | |
1532 | * be overwriting an internal NUL character | |
1533 | */ | |
1534 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 | 1535 | if (name->len > DNAME_INLINE_LEN-1) { |
8d85b484 AV |
1536 | size_t size = offsetof(struct external_name, name[1]); |
1537 | struct external_name *p = kmalloc(size + name->len, GFP_KERNEL); | |
1538 | if (!p) { | |
1da177e4 LT |
1539 | kmem_cache_free(dentry_cache, dentry); |
1540 | return NULL; | |
1541 | } | |
8d85b484 AV |
1542 | atomic_set(&p->u.count, 1); |
1543 | dname = p->name; | |
df4c0e36 AR |
1544 | if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS)) |
1545 | kasan_unpoison_shadow(dname, | |
1546 | round_up(name->len + 1, sizeof(unsigned long))); | |
1da177e4 LT |
1547 | } else { |
1548 | dname = dentry->d_iname; | |
1549 | } | |
1da177e4 LT |
1550 | |
1551 | dentry->d_name.len = name->len; | |
1552 | dentry->d_name.hash = name->hash; | |
1553 | memcpy(dname, name->name, name->len); | |
1554 | dname[name->len] = 0; | |
1555 | ||
6326c71f LT |
1556 | /* Make sure we always see the terminating NUL character */ |
1557 | smp_wmb(); | |
1558 | dentry->d_name.name = dname; | |
1559 | ||
98474236 | 1560 | dentry->d_lockref.count = 1; |
dea3667b | 1561 | dentry->d_flags = 0; |
1da177e4 | 1562 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1563 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1564 | dentry->d_inode = NULL; |
a4464dbc AV |
1565 | dentry->d_parent = dentry; |
1566 | dentry->d_sb = sb; | |
1da177e4 LT |
1567 | dentry->d_op = NULL; |
1568 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1569 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1570 | INIT_LIST_HEAD(&dentry->d_lru); |
1571 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
946e51f2 AV |
1572 | INIT_HLIST_NODE(&dentry->d_u.d_alias); |
1573 | INIT_LIST_HEAD(&dentry->d_child); | |
a4464dbc | 1574 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1575 | |
3e880fb5 | 1576 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1577 | |
1da177e4 LT |
1578 | return dentry; |
1579 | } | |
a4464dbc AV |
1580 | |
1581 | /** | |
1582 | * d_alloc - allocate a dcache entry | |
1583 | * @parent: parent of entry to allocate | |
1584 | * @name: qstr of the name | |
1585 | * | |
1586 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1587 | * available. On a success the dentry is returned. The name passed in is | |
1588 | * copied and the copy passed in may be reused after this call. | |
1589 | */ | |
1590 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1591 | { | |
1592 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1593 | if (!dentry) | |
1594 | return NULL; | |
1595 | ||
1596 | spin_lock(&parent->d_lock); | |
1597 | /* | |
1598 | * don't need child lock because it is not subject | |
1599 | * to concurrency here | |
1600 | */ | |
1601 | __dget_dlock(parent); | |
1602 | dentry->d_parent = parent; | |
946e51f2 | 1603 | list_add(&dentry->d_child, &parent->d_subdirs); |
a4464dbc AV |
1604 | spin_unlock(&parent->d_lock); |
1605 | ||
1606 | return dentry; | |
1607 | } | |
ec4f8605 | 1608 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1609 | |
e1a24bb0 BF |
1610 | /** |
1611 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1612 | * @sb: the superblock | |
1613 | * @name: qstr of the name | |
1614 | * | |
1615 | * For a filesystem that just pins its dentries in memory and never | |
1616 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1617 | */ | |
4b936885 NP |
1618 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1619 | { | |
e1a24bb0 | 1620 | return __d_alloc(sb, name); |
4b936885 NP |
1621 | } |
1622 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1623 | ||
1da177e4 LT |
1624 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1625 | { | |
1626 | struct qstr q; | |
1627 | ||
1628 | q.name = name; | |
1629 | q.len = strlen(name); | |
1630 | q.hash = full_name_hash(q.name, q.len); | |
1631 | return d_alloc(parent, &q); | |
1632 | } | |
ef26ca97 | 1633 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1634 | |
fb045adb NP |
1635 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1636 | { | |
6f7f7caa LT |
1637 | WARN_ON_ONCE(dentry->d_op); |
1638 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1639 | DCACHE_OP_COMPARE | |
1640 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1641 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1642 | DCACHE_OP_DELETE )); |
1643 | dentry->d_op = op; | |
1644 | if (!op) | |
1645 | return; | |
1646 | if (op->d_hash) | |
1647 | dentry->d_flags |= DCACHE_OP_HASH; | |
1648 | if (op->d_compare) | |
1649 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1650 | if (op->d_revalidate) | |
1651 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1652 | if (op->d_weak_revalidate) |
1653 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1654 | if (op->d_delete) |
1655 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1656 | if (op->d_prune) |
1657 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1658 | |
1659 | } | |
1660 | EXPORT_SYMBOL(d_set_d_op); | |
1661 | ||
df1a085a DH |
1662 | |
1663 | /* | |
1664 | * d_set_fallthru - Mark a dentry as falling through to a lower layer | |
1665 | * @dentry - The dentry to mark | |
1666 | * | |
1667 | * Mark a dentry as falling through to the lower layer (as set with | |
1668 | * d_pin_lower()). This flag may be recorded on the medium. | |
1669 | */ | |
1670 | void d_set_fallthru(struct dentry *dentry) | |
1671 | { | |
1672 | spin_lock(&dentry->d_lock); | |
1673 | dentry->d_flags |= DCACHE_FALLTHRU; | |
1674 | spin_unlock(&dentry->d_lock); | |
1675 | } | |
1676 | EXPORT_SYMBOL(d_set_fallthru); | |
1677 | ||
b18825a7 DH |
1678 | static unsigned d_flags_for_inode(struct inode *inode) |
1679 | { | |
44bdb5e5 | 1680 | unsigned add_flags = DCACHE_REGULAR_TYPE; |
b18825a7 DH |
1681 | |
1682 | if (!inode) | |
1683 | return DCACHE_MISS_TYPE; | |
1684 | ||
1685 | if (S_ISDIR(inode->i_mode)) { | |
1686 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1687 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1688 | if (unlikely(!inode->i_op->lookup)) | |
1689 | add_flags = DCACHE_AUTODIR_TYPE; | |
1690 | else | |
1691 | inode->i_opflags |= IOP_LOOKUP; | |
1692 | } | |
44bdb5e5 DH |
1693 | goto type_determined; |
1694 | } | |
1695 | ||
1696 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1697 | if (unlikely(inode->i_op->follow_link)) { | |
b18825a7 | 1698 | add_flags = DCACHE_SYMLINK_TYPE; |
44bdb5e5 DH |
1699 | goto type_determined; |
1700 | } | |
1701 | inode->i_opflags |= IOP_NOFOLLOW; | |
b18825a7 DH |
1702 | } |
1703 | ||
44bdb5e5 DH |
1704 | if (unlikely(!S_ISREG(inode->i_mode))) |
1705 | add_flags = DCACHE_SPECIAL_TYPE; | |
1706 | ||
1707 | type_determined: | |
b18825a7 DH |
1708 | if (unlikely(IS_AUTOMOUNT(inode))) |
1709 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1710 | return add_flags; | |
1711 | } | |
1712 | ||
360da900 OH |
1713 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1714 | { | |
b18825a7 DH |
1715 | unsigned add_flags = d_flags_for_inode(inode); |
1716 | ||
b23fb0a6 | 1717 | spin_lock(&dentry->d_lock); |
df1a085a DH |
1718 | dentry->d_flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); |
1719 | dentry->d_flags |= add_flags; | |
b18825a7 | 1720 | if (inode) |
946e51f2 | 1721 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); |
360da900 | 1722 | dentry->d_inode = inode; |
31e6b01f | 1723 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1724 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1725 | fsnotify_d_instantiate(dentry, inode); |
1726 | } | |
1727 | ||
1da177e4 LT |
1728 | /** |
1729 | * d_instantiate - fill in inode information for a dentry | |
1730 | * @entry: dentry to complete | |
1731 | * @inode: inode to attach to this dentry | |
1732 | * | |
1733 | * Fill in inode information in the entry. | |
1734 | * | |
1735 | * This turns negative dentries into productive full members | |
1736 | * of society. | |
1737 | * | |
1738 | * NOTE! This assumes that the inode count has been incremented | |
1739 | * (or otherwise set) by the caller to indicate that it is now | |
1740 | * in use by the dcache. | |
1741 | */ | |
1742 | ||
1743 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1744 | { | |
946e51f2 | 1745 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
873feea0 NP |
1746 | if (inode) |
1747 | spin_lock(&inode->i_lock); | |
360da900 | 1748 | __d_instantiate(entry, inode); |
873feea0 NP |
1749 | if (inode) |
1750 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1751 | security_d_instantiate(entry, inode); |
1752 | } | |
ec4f8605 | 1753 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1754 | |
1755 | /** | |
1756 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1757 | * @entry: dentry to instantiate | |
1758 | * @inode: inode to attach to this dentry | |
1759 | * | |
1760 | * Fill in inode information in the entry. On success, it returns NULL. | |
1761 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1762 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1763 | * |
1764 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1765 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1766 | * |
1767 | * This also assumes that the inode count has been incremented | |
1768 | * (or otherwise set) by the caller to indicate that it is now | |
1769 | * in use by the dcache. | |
1da177e4 | 1770 | */ |
770bfad8 DH |
1771 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1772 | struct inode *inode) | |
1da177e4 LT |
1773 | { |
1774 | struct dentry *alias; | |
1775 | int len = entry->d_name.len; | |
1776 | const char *name = entry->d_name.name; | |
1777 | unsigned int hash = entry->d_name.hash; | |
1778 | ||
770bfad8 | 1779 | if (!inode) { |
360da900 | 1780 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1781 | return NULL; |
1782 | } | |
1783 | ||
946e51f2 | 1784 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
9abca360 NP |
1785 | /* |
1786 | * Don't need alias->d_lock here, because aliases with | |
1787 | * d_parent == entry->d_parent are not subject to name or | |
1788 | * parent changes, because the parent inode i_mutex is held. | |
1789 | */ | |
12f8ad4b | 1790 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1791 | continue; |
1792 | if (alias->d_parent != entry->d_parent) | |
1793 | continue; | |
ee983e89 LT |
1794 | if (alias->d_name.len != len) |
1795 | continue; | |
12f8ad4b | 1796 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1797 | continue; |
dc0474be | 1798 | __dget(alias); |
1da177e4 LT |
1799 | return alias; |
1800 | } | |
770bfad8 | 1801 | |
360da900 | 1802 | __d_instantiate(entry, inode); |
1da177e4 LT |
1803 | return NULL; |
1804 | } | |
770bfad8 DH |
1805 | |
1806 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1807 | { | |
1808 | struct dentry *result; | |
1809 | ||
946e51f2 | 1810 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
770bfad8 | 1811 | |
873feea0 NP |
1812 | if (inode) |
1813 | spin_lock(&inode->i_lock); | |
770bfad8 | 1814 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1815 | if (inode) |
1816 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1817 | |
1818 | if (!result) { | |
1819 | security_d_instantiate(entry, inode); | |
1820 | return NULL; | |
1821 | } | |
1822 | ||
1823 | BUG_ON(!d_unhashed(result)); | |
1824 | iput(inode); | |
1825 | return result; | |
1826 | } | |
1827 | ||
1da177e4 LT |
1828 | EXPORT_SYMBOL(d_instantiate_unique); |
1829 | ||
b70a80e7 MS |
1830 | /** |
1831 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1832 | * @entry: dentry to complete | |
1833 | * @inode: inode to attach to this dentry | |
1834 | * | |
1835 | * Fill in inode information in the entry. If a directory alias is found, then | |
1836 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1837 | * guarantees that a directory inode may never have more than one alias. | |
1838 | */ | |
1839 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1840 | { | |
946e51f2 | 1841 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
b70a80e7 MS |
1842 | |
1843 | spin_lock(&inode->i_lock); | |
1844 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1845 | spin_unlock(&inode->i_lock); | |
1846 | iput(inode); | |
1847 | return -EBUSY; | |
1848 | } | |
1849 | __d_instantiate(entry, inode); | |
1850 | spin_unlock(&inode->i_lock); | |
1851 | security_d_instantiate(entry, inode); | |
1852 | ||
1853 | return 0; | |
1854 | } | |
1855 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1856 | ||
adc0e91a AV |
1857 | struct dentry *d_make_root(struct inode *root_inode) |
1858 | { | |
1859 | struct dentry *res = NULL; | |
1860 | ||
1861 | if (root_inode) { | |
26fe5750 | 1862 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1863 | |
1864 | res = __d_alloc(root_inode->i_sb, &name); | |
1865 | if (res) | |
1866 | d_instantiate(res, root_inode); | |
1867 | else | |
1868 | iput(root_inode); | |
1869 | } | |
1870 | return res; | |
1871 | } | |
1872 | EXPORT_SYMBOL(d_make_root); | |
1873 | ||
d891eedb BF |
1874 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1875 | { | |
1876 | struct dentry *alias; | |
1877 | ||
b3d9b7a3 | 1878 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1879 | return NULL; |
946e51f2 | 1880 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); |
d891eedb BF |
1881 | __dget(alias); |
1882 | return alias; | |
1883 | } | |
1884 | ||
46f72b34 SW |
1885 | /** |
1886 | * d_find_any_alias - find any alias for a given inode | |
1887 | * @inode: inode to find an alias for | |
1888 | * | |
1889 | * If any aliases exist for the given inode, take and return a | |
1890 | * reference for one of them. If no aliases exist, return %NULL. | |
1891 | */ | |
1892 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1893 | { |
1894 | struct dentry *de; | |
1895 | ||
1896 | spin_lock(&inode->i_lock); | |
1897 | de = __d_find_any_alias(inode); | |
1898 | spin_unlock(&inode->i_lock); | |
1899 | return de; | |
1900 | } | |
46f72b34 | 1901 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1902 | |
49c7dd28 | 1903 | static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected) |
4ea3ada2 | 1904 | { |
b911a6bd | 1905 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1906 | struct dentry *tmp; |
1907 | struct dentry *res; | |
b18825a7 | 1908 | unsigned add_flags; |
4ea3ada2 CH |
1909 | |
1910 | if (!inode) | |
44003728 | 1911 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1912 | if (IS_ERR(inode)) |
1913 | return ERR_CAST(inode); | |
1914 | ||
d891eedb | 1915 | res = d_find_any_alias(inode); |
9308a612 CH |
1916 | if (res) |
1917 | goto out_iput; | |
1918 | ||
a4464dbc | 1919 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1920 | if (!tmp) { |
1921 | res = ERR_PTR(-ENOMEM); | |
1922 | goto out_iput; | |
4ea3ada2 | 1923 | } |
b5c84bf6 | 1924 | |
873feea0 | 1925 | spin_lock(&inode->i_lock); |
d891eedb | 1926 | res = __d_find_any_alias(inode); |
9308a612 | 1927 | if (res) { |
873feea0 | 1928 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1929 | dput(tmp); |
1930 | goto out_iput; | |
1931 | } | |
1932 | ||
1933 | /* attach a disconnected dentry */ | |
1a0a397e BF |
1934 | add_flags = d_flags_for_inode(inode); |
1935 | ||
1936 | if (disconnected) | |
1937 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 1938 | |
9308a612 | 1939 | spin_lock(&tmp->d_lock); |
9308a612 | 1940 | tmp->d_inode = inode; |
b18825a7 | 1941 | tmp->d_flags |= add_flags; |
946e51f2 | 1942 | hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry); |
1879fd6a | 1943 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1944 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1945 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1946 | spin_unlock(&tmp->d_lock); |
873feea0 | 1947 | spin_unlock(&inode->i_lock); |
24ff6663 | 1948 | security_d_instantiate(tmp, inode); |
9308a612 | 1949 | |
9308a612 CH |
1950 | return tmp; |
1951 | ||
1952 | out_iput: | |
24ff6663 JB |
1953 | if (res && !IS_ERR(res)) |
1954 | security_d_instantiate(res, inode); | |
9308a612 CH |
1955 | iput(inode); |
1956 | return res; | |
4ea3ada2 | 1957 | } |
1a0a397e BF |
1958 | |
1959 | /** | |
1960 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
1961 | * @inode: inode to allocate the dentry for | |
1962 | * | |
1963 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1964 | * similar open by handle operations. The returned dentry may be anonymous, | |
1965 | * or may have a full name (if the inode was already in the cache). | |
1966 | * | |
1967 | * When called on a directory inode, we must ensure that the inode only ever | |
1968 | * has one dentry. If a dentry is found, that is returned instead of | |
1969 | * allocating a new one. | |
1970 | * | |
1971 | * On successful return, the reference to the inode has been transferred | |
1972 | * to the dentry. In case of an error the reference on the inode is released. | |
1973 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1974 | * be passed in and the error will be propagated to the return value, | |
1975 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
1976 | */ | |
1977 | struct dentry *d_obtain_alias(struct inode *inode) | |
1978 | { | |
1979 | return __d_obtain_alias(inode, 1); | |
1980 | } | |
adc48720 | 1981 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 1982 | |
1a0a397e BF |
1983 | /** |
1984 | * d_obtain_root - find or allocate a dentry for a given inode | |
1985 | * @inode: inode to allocate the dentry for | |
1986 | * | |
1987 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
1988 | * | |
1989 | * We must ensure that directory inodes only ever have one dentry. If a | |
1990 | * dentry is found, that is returned instead of allocating a new one. | |
1991 | * | |
1992 | * On successful return, the reference to the inode has been transferred | |
1993 | * to the dentry. In case of an error the reference on the inode is | |
1994 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
1995 | * error will be propagate to the return value, with a %NULL @inode | |
1996 | * replaced by ERR_PTR(-ESTALE). | |
1997 | */ | |
1998 | struct dentry *d_obtain_root(struct inode *inode) | |
1999 | { | |
2000 | return __d_obtain_alias(inode, 0); | |
2001 | } | |
2002 | EXPORT_SYMBOL(d_obtain_root); | |
2003 | ||
9403540c BN |
2004 | /** |
2005 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
2006 | * @inode: the inode case-insensitive lookup has found | |
2007 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
2008 | * @name: the case-exact name to be associated with the returned dentry | |
2009 | * | |
2010 | * This is to avoid filling the dcache with case-insensitive names to the | |
2011 | * same inode, only the actual correct case is stored in the dcache for | |
2012 | * case-insensitive filesystems. | |
2013 | * | |
2014 | * For a case-insensitive lookup match and if the the case-exact dentry | |
2015 | * already exists in in the dcache, use it and return it. | |
2016 | * | |
2017 | * If no entry exists with the exact case name, allocate new dentry with | |
2018 | * the exact case, and return the spliced entry. | |
2019 | */ | |
e45b590b | 2020 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
2021 | struct qstr *name) |
2022 | { | |
9403540c BN |
2023 | struct dentry *found; |
2024 | struct dentry *new; | |
2025 | ||
b6520c81 CH |
2026 | /* |
2027 | * First check if a dentry matching the name already exists, | |
2028 | * if not go ahead and create it now. | |
2029 | */ | |
9403540c | 2030 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
2031 | if (!found) { |
2032 | new = d_alloc(dentry->d_parent, name); | |
2033 | if (!new) { | |
4f522a24 | 2034 | found = ERR_PTR(-ENOMEM); |
427c77d4 AV |
2035 | } else { |
2036 | found = d_splice_alias(inode, new); | |
2037 | if (found) { | |
2038 | dput(new); | |
2039 | return found; | |
2040 | } | |
2041 | return new; | |
9403540c | 2042 | } |
9403540c | 2043 | } |
9403540c | 2044 | iput(inode); |
4f522a24 | 2045 | return found; |
9403540c | 2046 | } |
ec4f8605 | 2047 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 2048 | |
12f8ad4b LT |
2049 | /* |
2050 | * Do the slow-case of the dentry name compare. | |
2051 | * | |
2052 | * Unlike the dentry_cmp() function, we need to atomically | |
da53be12 | 2053 | * load the name and length information, so that the |
12f8ad4b LT |
2054 | * filesystem can rely on them, and can use the 'name' and |
2055 | * 'len' information without worrying about walking off the | |
2056 | * end of memory etc. | |
2057 | * | |
2058 | * Thus the read_seqcount_retry() and the "duplicate" info | |
2059 | * in arguments (the low-level filesystem should not look | |
2060 | * at the dentry inode or name contents directly, since | |
2061 | * rename can change them while we're in RCU mode). | |
2062 | */ | |
2063 | enum slow_d_compare { | |
2064 | D_COMP_OK, | |
2065 | D_COMP_NOMATCH, | |
2066 | D_COMP_SEQRETRY, | |
2067 | }; | |
2068 | ||
2069 | static noinline enum slow_d_compare slow_dentry_cmp( | |
2070 | const struct dentry *parent, | |
12f8ad4b LT |
2071 | struct dentry *dentry, |
2072 | unsigned int seq, | |
2073 | const struct qstr *name) | |
2074 | { | |
2075 | int tlen = dentry->d_name.len; | |
2076 | const char *tname = dentry->d_name.name; | |
12f8ad4b LT |
2077 | |
2078 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2079 | cpu_relax(); | |
2080 | return D_COMP_SEQRETRY; | |
2081 | } | |
da53be12 | 2082 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
12f8ad4b LT |
2083 | return D_COMP_NOMATCH; |
2084 | return D_COMP_OK; | |
2085 | } | |
2086 | ||
31e6b01f NP |
2087 | /** |
2088 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2089 | * @parent: parent dentry | |
2090 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2091 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2092 | * Returns: dentry, or NULL |
2093 | * | |
2094 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2095 | * resolution (store-free path walking) design described in | |
2096 | * Documentation/filesystems/path-lookup.txt. | |
2097 | * | |
2098 | * This is not to be used outside core vfs. | |
2099 | * | |
2100 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2101 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2102 | * without taking d_lock and checking d_seq sequence count against @seq | |
2103 | * returned here. | |
2104 | * | |
15570086 | 2105 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2106 | * function. |
2107 | * | |
2108 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2109 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2110 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2111 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2112 | * |
2113 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2114 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2115 | */ |
8966be90 LT |
2116 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2117 | const struct qstr *name, | |
da53be12 | 2118 | unsigned *seqp) |
31e6b01f | 2119 | { |
26fe5750 | 2120 | u64 hashlen = name->hash_len; |
31e6b01f | 2121 | const unsigned char *str = name->name; |
26fe5750 | 2122 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 2123 | struct hlist_bl_node *node; |
31e6b01f NP |
2124 | struct dentry *dentry; |
2125 | ||
2126 | /* | |
2127 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2128 | * required to prevent single threaded performance regressions | |
2129 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2130 | * Keep the two functions in sync. | |
2131 | */ | |
2132 | ||
2133 | /* | |
2134 | * The hash list is protected using RCU. | |
2135 | * | |
2136 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2137 | * races with d_move(). | |
2138 | * | |
2139 | * It is possible that concurrent renames can mess up our list | |
2140 | * walk here and result in missing our dentry, resulting in the | |
2141 | * false-negative result. d_lookup() protects against concurrent | |
2142 | * renames using rename_lock seqlock. | |
2143 | * | |
b0a4bb83 | 2144 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2145 | */ |
b07ad996 | 2146 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2147 | unsigned seq; |
31e6b01f | 2148 | |
31e6b01f | 2149 | seqretry: |
12f8ad4b LT |
2150 | /* |
2151 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2152 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2153 | * |
2154 | * The caller must perform a seqcount check in order | |
da53be12 | 2155 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2156 | * |
2157 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2158 | * we don't wait for the sequence count to stabilize if it | |
2159 | * is in the middle of a sequence change. If we do the slow | |
2160 | * dentry compare, we will do seqretries until it is stable, | |
2161 | * and if we end up with a successful lookup, we actually | |
2162 | * want to exit RCU lookup anyway. | |
2163 | */ | |
2164 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2165 | if (dentry->d_parent != parent) |
2166 | continue; | |
2e321806 LT |
2167 | if (d_unhashed(dentry)) |
2168 | continue; | |
12f8ad4b | 2169 | |
830c0f0e | 2170 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
2171 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2172 | continue; | |
da53be12 LT |
2173 | *seqp = seq; |
2174 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
12f8ad4b LT |
2175 | case D_COMP_OK: |
2176 | return dentry; | |
2177 | case D_COMP_NOMATCH: | |
31e6b01f | 2178 | continue; |
12f8ad4b LT |
2179 | default: |
2180 | goto seqretry; | |
2181 | } | |
31e6b01f | 2182 | } |
12f8ad4b | 2183 | |
26fe5750 | 2184 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 2185 | continue; |
da53be12 | 2186 | *seqp = seq; |
26fe5750 | 2187 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 2188 | return dentry; |
31e6b01f NP |
2189 | } |
2190 | return NULL; | |
2191 | } | |
2192 | ||
1da177e4 LT |
2193 | /** |
2194 | * d_lookup - search for a dentry | |
2195 | * @parent: parent dentry | |
2196 | * @name: qstr of name we wish to find | |
b04f784e | 2197 | * Returns: dentry, or NULL |
1da177e4 | 2198 | * |
b04f784e NP |
2199 | * d_lookup searches the children of the parent dentry for the name in |
2200 | * question. If the dentry is found its reference count is incremented and the | |
2201 | * dentry is returned. The caller must use dput to free the entry when it has | |
2202 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2203 | */ |
da2d8455 | 2204 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2205 | { |
31e6b01f | 2206 | struct dentry *dentry; |
949854d0 | 2207 | unsigned seq; |
1da177e4 | 2208 | |
b8314f93 DY |
2209 | do { |
2210 | seq = read_seqbegin(&rename_lock); | |
2211 | dentry = __d_lookup(parent, name); | |
2212 | if (dentry) | |
1da177e4 LT |
2213 | break; |
2214 | } while (read_seqretry(&rename_lock, seq)); | |
2215 | return dentry; | |
2216 | } | |
ec4f8605 | 2217 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2218 | |
31e6b01f | 2219 | /** |
b04f784e NP |
2220 | * __d_lookup - search for a dentry (racy) |
2221 | * @parent: parent dentry | |
2222 | * @name: qstr of name we wish to find | |
2223 | * Returns: dentry, or NULL | |
2224 | * | |
2225 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2226 | * false-negative result due to unrelated rename activity. | |
2227 | * | |
2228 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2229 | * however it must be used carefully, eg. with a following d_lookup in | |
2230 | * the case of failure. | |
2231 | * | |
2232 | * __d_lookup callers must be commented. | |
2233 | */ | |
a713ca2a | 2234 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
2235 | { |
2236 | unsigned int len = name->len; | |
2237 | unsigned int hash = name->hash; | |
2238 | const unsigned char *str = name->name; | |
b07ad996 | 2239 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 2240 | struct hlist_bl_node *node; |
31e6b01f | 2241 | struct dentry *found = NULL; |
665a7583 | 2242 | struct dentry *dentry; |
1da177e4 | 2243 | |
31e6b01f NP |
2244 | /* |
2245 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2246 | * required to prevent single threaded performance regressions | |
2247 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2248 | * Keep the two functions in sync. | |
2249 | */ | |
2250 | ||
b04f784e NP |
2251 | /* |
2252 | * The hash list is protected using RCU. | |
2253 | * | |
2254 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2255 | * with d_move(). | |
2256 | * | |
2257 | * It is possible that concurrent renames can mess up our list | |
2258 | * walk here and result in missing our dentry, resulting in the | |
2259 | * false-negative result. d_lookup() protects against concurrent | |
2260 | * renames using rename_lock seqlock. | |
2261 | * | |
b0a4bb83 | 2262 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2263 | */ |
1da177e4 LT |
2264 | rcu_read_lock(); |
2265 | ||
b07ad996 | 2266 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2267 | |
1da177e4 LT |
2268 | if (dentry->d_name.hash != hash) |
2269 | continue; | |
1da177e4 LT |
2270 | |
2271 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2272 | if (dentry->d_parent != parent) |
2273 | goto next; | |
d0185c08 LT |
2274 | if (d_unhashed(dentry)) |
2275 | goto next; | |
2276 | ||
1da177e4 LT |
2277 | /* |
2278 | * It is safe to compare names since d_move() cannot | |
2279 | * change the qstr (protected by d_lock). | |
2280 | */ | |
fb045adb | 2281 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
2282 | int tlen = dentry->d_name.len; |
2283 | const char *tname = dentry->d_name.name; | |
da53be12 | 2284 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1da177e4 LT |
2285 | goto next; |
2286 | } else { | |
ee983e89 LT |
2287 | if (dentry->d_name.len != len) |
2288 | goto next; | |
12f8ad4b | 2289 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
2290 | goto next; |
2291 | } | |
2292 | ||
98474236 | 2293 | dentry->d_lockref.count++; |
d0185c08 | 2294 | found = dentry; |
1da177e4 LT |
2295 | spin_unlock(&dentry->d_lock); |
2296 | break; | |
2297 | next: | |
2298 | spin_unlock(&dentry->d_lock); | |
2299 | } | |
2300 | rcu_read_unlock(); | |
2301 | ||
2302 | return found; | |
2303 | } | |
2304 | ||
3e7e241f EB |
2305 | /** |
2306 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2307 | * @dir: Directory to search in | |
2308 | * @name: qstr of name we wish to find | |
2309 | * | |
4f522a24 | 2310 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2311 | */ |
2312 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2313 | { | |
3e7e241f EB |
2314 | /* |
2315 | * Check for a fs-specific hash function. Note that we must | |
2316 | * calculate the standard hash first, as the d_op->d_hash() | |
2317 | * routine may choose to leave the hash value unchanged. | |
2318 | */ | |
2319 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2320 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2321 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2322 | if (unlikely(err < 0)) |
2323 | return ERR_PTR(err); | |
3e7e241f | 2324 | } |
4f522a24 | 2325 | return d_lookup(dir, name); |
3e7e241f | 2326 | } |
4f522a24 | 2327 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2328 | |
1da177e4 LT |
2329 | /* |
2330 | * When a file is deleted, we have two options: | |
2331 | * - turn this dentry into a negative dentry | |
2332 | * - unhash this dentry and free it. | |
2333 | * | |
2334 | * Usually, we want to just turn this into | |
2335 | * a negative dentry, but if anybody else is | |
2336 | * currently using the dentry or the inode | |
2337 | * we can't do that and we fall back on removing | |
2338 | * it from the hash queues and waiting for | |
2339 | * it to be deleted later when it has no users | |
2340 | */ | |
2341 | ||
2342 | /** | |
2343 | * d_delete - delete a dentry | |
2344 | * @dentry: The dentry to delete | |
2345 | * | |
2346 | * Turn the dentry into a negative dentry if possible, otherwise | |
2347 | * remove it from the hash queues so it can be deleted later | |
2348 | */ | |
2349 | ||
2350 | void d_delete(struct dentry * dentry) | |
2351 | { | |
873feea0 | 2352 | struct inode *inode; |
7a91bf7f | 2353 | int isdir = 0; |
1da177e4 LT |
2354 | /* |
2355 | * Are we the only user? | |
2356 | */ | |
357f8e65 | 2357 | again: |
1da177e4 | 2358 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2359 | inode = dentry->d_inode; |
2360 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2361 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2362 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2363 | spin_unlock(&dentry->d_lock); |
2364 | cpu_relax(); | |
2365 | goto again; | |
2366 | } | |
13e3c5e5 | 2367 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2368 | dentry_unlink_inode(dentry); |
7a91bf7f | 2369 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2370 | return; |
2371 | } | |
2372 | ||
2373 | if (!d_unhashed(dentry)) | |
2374 | __d_drop(dentry); | |
2375 | ||
2376 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2377 | |
2378 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2379 | } |
ec4f8605 | 2380 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2381 | |
b07ad996 | 2382 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2383 | { |
ceb5bdc2 | 2384 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2385 | hlist_bl_lock(b); |
dea3667b | 2386 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2387 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2388 | hlist_bl_unlock(b); |
1da177e4 LT |
2389 | } |
2390 | ||
770bfad8 DH |
2391 | static void _d_rehash(struct dentry * entry) |
2392 | { | |
2393 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2394 | } | |
2395 | ||
1da177e4 LT |
2396 | /** |
2397 | * d_rehash - add an entry back to the hash | |
2398 | * @entry: dentry to add to the hash | |
2399 | * | |
2400 | * Adds a dentry to the hash according to its name. | |
2401 | */ | |
2402 | ||
2403 | void d_rehash(struct dentry * entry) | |
2404 | { | |
1da177e4 | 2405 | spin_lock(&entry->d_lock); |
770bfad8 | 2406 | _d_rehash(entry); |
1da177e4 | 2407 | spin_unlock(&entry->d_lock); |
1da177e4 | 2408 | } |
ec4f8605 | 2409 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2410 | |
fb2d5b86 NP |
2411 | /** |
2412 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2413 | * @dentry: dentry to be updated | |
2414 | * @name: new name | |
2415 | * | |
2416 | * Update a case insensitive dentry with new case of name. | |
2417 | * | |
2418 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2419 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2420 | * lengths). | |
2421 | * | |
2422 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2423 | * keep renames and concurrent inserts, and readdir(2) away). | |
2424 | */ | |
2425 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2426 | { | |
7ebfa57f | 2427 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2428 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2429 | ||
fb2d5b86 | 2430 | spin_lock(&dentry->d_lock); |
31e6b01f | 2431 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2432 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2433 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2434 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2435 | } |
2436 | EXPORT_SYMBOL(dentry_update_name_case); | |
2437 | ||
8d85b484 | 2438 | static void swap_names(struct dentry *dentry, struct dentry *target) |
1da177e4 | 2439 | { |
8d85b484 AV |
2440 | if (unlikely(dname_external(target))) { |
2441 | if (unlikely(dname_external(dentry))) { | |
1da177e4 LT |
2442 | /* |
2443 | * Both external: swap the pointers | |
2444 | */ | |
9a8d5bb4 | 2445 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2446 | } else { |
2447 | /* | |
2448 | * dentry:internal, target:external. Steal target's | |
2449 | * storage and make target internal. | |
2450 | */ | |
321bcf92 BF |
2451 | memcpy(target->d_iname, dentry->d_name.name, |
2452 | dentry->d_name.len + 1); | |
1da177e4 LT |
2453 | dentry->d_name.name = target->d_name.name; |
2454 | target->d_name.name = target->d_iname; | |
2455 | } | |
2456 | } else { | |
8d85b484 | 2457 | if (unlikely(dname_external(dentry))) { |
1da177e4 LT |
2458 | /* |
2459 | * dentry:external, target:internal. Give dentry's | |
2460 | * storage to target and make dentry internal | |
2461 | */ | |
2462 | memcpy(dentry->d_iname, target->d_name.name, | |
2463 | target->d_name.len + 1); | |
2464 | target->d_name.name = dentry->d_name.name; | |
2465 | dentry->d_name.name = dentry->d_iname; | |
2466 | } else { | |
2467 | /* | |
da1ce067 | 2468 | * Both are internal. |
1da177e4 | 2469 | */ |
da1ce067 MS |
2470 | unsigned int i; |
2471 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
08d4f772 MP |
2472 | kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN); |
2473 | kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN); | |
da1ce067 MS |
2474 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { |
2475 | swap(((long *) &dentry->d_iname)[i], | |
2476 | ((long *) &target->d_iname)[i]); | |
2477 | } | |
1da177e4 LT |
2478 | } |
2479 | } | |
a28ddb87 | 2480 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
1da177e4 LT |
2481 | } |
2482 | ||
8d85b484 AV |
2483 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2484 | { | |
2485 | struct external_name *old_name = NULL; | |
2486 | if (unlikely(dname_external(dentry))) | |
2487 | old_name = external_name(dentry); | |
2488 | if (unlikely(dname_external(target))) { | |
2489 | atomic_inc(&external_name(target)->u.count); | |
2490 | dentry->d_name = target->d_name; | |
2491 | } else { | |
2492 | memcpy(dentry->d_iname, target->d_name.name, | |
2493 | target->d_name.len + 1); | |
2494 | dentry->d_name.name = dentry->d_iname; | |
2495 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2496 | } | |
2497 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
2498 | kfree_rcu(old_name, u.head); | |
2499 | } | |
2500 | ||
2fd6b7f5 NP |
2501 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2502 | { | |
2503 | /* | |
2504 | * XXXX: do we really need to take target->d_lock? | |
2505 | */ | |
2506 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2507 | spin_lock(&target->d_parent->d_lock); | |
2508 | else { | |
2509 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2510 | spin_lock(&dentry->d_parent->d_lock); | |
2511 | spin_lock_nested(&target->d_parent->d_lock, | |
2512 | DENTRY_D_LOCK_NESTED); | |
2513 | } else { | |
2514 | spin_lock(&target->d_parent->d_lock); | |
2515 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2516 | DENTRY_D_LOCK_NESTED); | |
2517 | } | |
2518 | } | |
2519 | if (target < dentry) { | |
2520 | spin_lock_nested(&target->d_lock, 2); | |
2521 | spin_lock_nested(&dentry->d_lock, 3); | |
2522 | } else { | |
2523 | spin_lock_nested(&dentry->d_lock, 2); | |
2524 | spin_lock_nested(&target->d_lock, 3); | |
2525 | } | |
2526 | } | |
2527 | ||
986c0194 | 2528 | static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target) |
2fd6b7f5 NP |
2529 | { |
2530 | if (target->d_parent != dentry->d_parent) | |
2531 | spin_unlock(&dentry->d_parent->d_lock); | |
2532 | if (target->d_parent != target) | |
2533 | spin_unlock(&target->d_parent->d_lock); | |
986c0194 AV |
2534 | spin_unlock(&target->d_lock); |
2535 | spin_unlock(&dentry->d_lock); | |
2fd6b7f5 NP |
2536 | } |
2537 | ||
1da177e4 | 2538 | /* |
2fd6b7f5 NP |
2539 | * When switching names, the actual string doesn't strictly have to |
2540 | * be preserved in the target - because we're dropping the target | |
2541 | * anyway. As such, we can just do a simple memcpy() to copy over | |
d2fa4a84 ME |
2542 | * the new name before we switch, unless we are going to rehash |
2543 | * it. Note that if we *do* unhash the target, we are not allowed | |
2544 | * to rehash it without giving it a new name/hash key - whether | |
2545 | * we swap or overwrite the names here, resulting name won't match | |
2546 | * the reality in filesystem; it's only there for d_path() purposes. | |
2547 | * Note that all of this is happening under rename_lock, so the | |
2548 | * any hash lookup seeing it in the middle of manipulations will | |
2549 | * be discarded anyway. So we do not care what happens to the hash | |
2550 | * key in that case. | |
1da177e4 | 2551 | */ |
9eaef27b | 2552 | /* |
18367501 | 2553 | * __d_move - move a dentry |
1da177e4 LT |
2554 | * @dentry: entry to move |
2555 | * @target: new dentry | |
da1ce067 | 2556 | * @exchange: exchange the two dentries |
1da177e4 LT |
2557 | * |
2558 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2559 | * dcache entries should not be moved in this way. Caller must hold |
2560 | * rename_lock, the i_mutex of the source and target directories, | |
2561 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2562 | */ |
da1ce067 MS |
2563 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2564 | bool exchange) | |
1da177e4 | 2565 | { |
1da177e4 LT |
2566 | if (!dentry->d_inode) |
2567 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2568 | ||
2fd6b7f5 NP |
2569 | BUG_ON(d_ancestor(dentry, target)); |
2570 | BUG_ON(d_ancestor(target, dentry)); | |
2571 | ||
2fd6b7f5 | 2572 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2573 | |
31e6b01f | 2574 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2575 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2576 | |
ceb5bdc2 NP |
2577 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2578 | ||
2579 | /* | |
2580 | * Move the dentry to the target hash queue. Don't bother checking | |
2581 | * for the same hash queue because of how unlikely it is. | |
2582 | */ | |
2583 | __d_drop(dentry); | |
789680d1 | 2584 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 | 2585 | |
da1ce067 MS |
2586 | /* |
2587 | * Unhash the target (d_delete() is not usable here). If exchanging | |
2588 | * the two dentries, then rehash onto the other's hash queue. | |
2589 | */ | |
1da177e4 | 2590 | __d_drop(target); |
da1ce067 MS |
2591 | if (exchange) { |
2592 | __d_rehash(target, | |
2593 | d_hash(dentry->d_parent, dentry->d_name.hash)); | |
2594 | } | |
1da177e4 | 2595 | |
1da177e4 | 2596 | /* Switch the names.. */ |
8d85b484 AV |
2597 | if (exchange) |
2598 | swap_names(dentry, target); | |
2599 | else | |
2600 | copy_name(dentry, target); | |
1da177e4 | 2601 | |
63cf427a | 2602 | /* ... and switch them in the tree */ |
1da177e4 | 2603 | if (IS_ROOT(dentry)) { |
63cf427a | 2604 | /* splicing a tree */ |
1da177e4 LT |
2605 | dentry->d_parent = target->d_parent; |
2606 | target->d_parent = target; | |
946e51f2 AV |
2607 | list_del_init(&target->d_child); |
2608 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
1da177e4 | 2609 | } else { |
63cf427a | 2610 | /* swapping two dentries */ |
9a8d5bb4 | 2611 | swap(dentry->d_parent, target->d_parent); |
946e51f2 AV |
2612 | list_move(&target->d_child, &target->d_parent->d_subdirs); |
2613 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
63cf427a AV |
2614 | if (exchange) |
2615 | fsnotify_d_move(target); | |
2616 | fsnotify_d_move(dentry); | |
1da177e4 LT |
2617 | } |
2618 | ||
31e6b01f NP |
2619 | write_seqcount_end(&target->d_seq); |
2620 | write_seqcount_end(&dentry->d_seq); | |
2621 | ||
986c0194 | 2622 | dentry_unlock_for_move(dentry, target); |
18367501 AV |
2623 | } |
2624 | ||
2625 | /* | |
2626 | * d_move - move a dentry | |
2627 | * @dentry: entry to move | |
2628 | * @target: new dentry | |
2629 | * | |
2630 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2631 | * dcache entries should not be moved in this way. See the locking |
2632 | * requirements for __d_move. | |
18367501 AV |
2633 | */ |
2634 | void d_move(struct dentry *dentry, struct dentry *target) | |
2635 | { | |
2636 | write_seqlock(&rename_lock); | |
da1ce067 | 2637 | __d_move(dentry, target, false); |
1da177e4 | 2638 | write_sequnlock(&rename_lock); |
9eaef27b | 2639 | } |
ec4f8605 | 2640 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2641 | |
da1ce067 MS |
2642 | /* |
2643 | * d_exchange - exchange two dentries | |
2644 | * @dentry1: first dentry | |
2645 | * @dentry2: second dentry | |
2646 | */ | |
2647 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2648 | { | |
2649 | write_seqlock(&rename_lock); | |
2650 | ||
2651 | WARN_ON(!dentry1->d_inode); | |
2652 | WARN_ON(!dentry2->d_inode); | |
2653 | WARN_ON(IS_ROOT(dentry1)); | |
2654 | WARN_ON(IS_ROOT(dentry2)); | |
2655 | ||
2656 | __d_move(dentry1, dentry2, true); | |
2657 | ||
2658 | write_sequnlock(&rename_lock); | |
2659 | } | |
2660 | ||
e2761a11 OH |
2661 | /** |
2662 | * d_ancestor - search for an ancestor | |
2663 | * @p1: ancestor dentry | |
2664 | * @p2: child dentry | |
2665 | * | |
2666 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2667 | * an ancestor of p2, else NULL. | |
9eaef27b | 2668 | */ |
e2761a11 | 2669 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2670 | { |
2671 | struct dentry *p; | |
2672 | ||
871c0067 | 2673 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2674 | if (p->d_parent == p1) |
e2761a11 | 2675 | return p; |
9eaef27b | 2676 | } |
e2761a11 | 2677 | return NULL; |
9eaef27b TM |
2678 | } |
2679 | ||
2680 | /* | |
2681 | * This helper attempts to cope with remotely renamed directories | |
2682 | * | |
2683 | * It assumes that the caller is already holding | |
18367501 | 2684 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2685 | * |
2686 | * Note: If ever the locking in lock_rename() changes, then please | |
2687 | * remember to update this too... | |
9eaef27b | 2688 | */ |
b5ae6b15 | 2689 | static int __d_unalias(struct inode *inode, |
873feea0 | 2690 | struct dentry *dentry, struct dentry *alias) |
9eaef27b TM |
2691 | { |
2692 | struct mutex *m1 = NULL, *m2 = NULL; | |
b5ae6b15 | 2693 | int ret = -EBUSY; |
9eaef27b TM |
2694 | |
2695 | /* If alias and dentry share a parent, then no extra locks required */ | |
2696 | if (alias->d_parent == dentry->d_parent) | |
2697 | goto out_unalias; | |
2698 | ||
9eaef27b | 2699 | /* See lock_rename() */ |
9eaef27b TM |
2700 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2701 | goto out_err; | |
2702 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2703 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2704 | goto out_err; | |
2705 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2706 | out_unalias: | |
8ed936b5 | 2707 | __d_move(alias, dentry, false); |
b5ae6b15 | 2708 | ret = 0; |
9eaef27b | 2709 | out_err: |
873feea0 | 2710 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2711 | if (m2) |
2712 | mutex_unlock(m2); | |
2713 | if (m1) | |
2714 | mutex_unlock(m1); | |
2715 | return ret; | |
2716 | } | |
2717 | ||
3f70bd51 BF |
2718 | /** |
2719 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
2720 | * @inode: the inode which may have a disconnected dentry | |
2721 | * @dentry: a negative dentry which we want to point to the inode. | |
2722 | * | |
da093a9b BF |
2723 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
2724 | * place of the given dentry and return it, else simply d_add the inode | |
2725 | * to the dentry and return NULL. | |
3f70bd51 | 2726 | * |
908790fa BF |
2727 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
2728 | * we should error out: directories can't have multiple aliases. | |
2729 | * | |
3f70bd51 BF |
2730 | * This is needed in the lookup routine of any filesystem that is exportable |
2731 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
2732 | * | |
2733 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
2734 | * is returned. This matches the expected return value of ->lookup. | |
2735 | * | |
2736 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
2737 | * In that case, we know that the inode will be a regular file, and also this | |
2738 | * will only occur during atomic_open. So we need to check for the dentry | |
2739 | * being already hashed only in the final case. | |
2740 | */ | |
2741 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
2742 | { | |
3f70bd51 BF |
2743 | if (IS_ERR(inode)) |
2744 | return ERR_CAST(inode); | |
2745 | ||
770bfad8 DH |
2746 | BUG_ON(!d_unhashed(dentry)); |
2747 | ||
770bfad8 | 2748 | if (!inode) { |
360da900 | 2749 | __d_instantiate(dentry, NULL); |
b5ae6b15 | 2750 | goto out; |
770bfad8 | 2751 | } |
873feea0 | 2752 | spin_lock(&inode->i_lock); |
9eaef27b | 2753 | if (S_ISDIR(inode->i_mode)) { |
b5ae6b15 AV |
2754 | struct dentry *new = __d_find_any_alias(inode); |
2755 | if (unlikely(new)) { | |
18367501 | 2756 | write_seqlock(&rename_lock); |
b5ae6b15 AV |
2757 | if (unlikely(d_ancestor(new, dentry))) { |
2758 | write_sequnlock(&rename_lock); | |
b18dafc8 | 2759 | spin_unlock(&inode->i_lock); |
b5ae6b15 AV |
2760 | dput(new); |
2761 | new = ERR_PTR(-ELOOP); | |
2762 | pr_warn_ratelimited( | |
2763 | "VFS: Lookup of '%s' in %s %s" | |
2764 | " would have caused loop\n", | |
2765 | dentry->d_name.name, | |
2766 | inode->i_sb->s_type->name, | |
2767 | inode->i_sb->s_id); | |
2768 | } else if (!IS_ROOT(new)) { | |
2769 | int err = __d_unalias(inode, dentry, new); | |
18367501 | 2770 | write_sequnlock(&rename_lock); |
b5ae6b15 AV |
2771 | if (err) { |
2772 | dput(new); | |
2773 | new = ERR_PTR(err); | |
2774 | } | |
18367501 | 2775 | } else { |
b5ae6b15 AV |
2776 | __d_move(new, dentry, false); |
2777 | write_sequnlock(&rename_lock); | |
2778 | spin_unlock(&inode->i_lock); | |
2779 | security_d_instantiate(new, inode); | |
dd179946 | 2780 | } |
b5ae6b15 AV |
2781 | iput(inode); |
2782 | return new; | |
9eaef27b | 2783 | } |
770bfad8 | 2784 | } |
b5ae6b15 AV |
2785 | /* already taking inode->i_lock, so d_add() by hand */ |
2786 | __d_instantiate(dentry, inode); | |
873feea0 | 2787 | spin_unlock(&inode->i_lock); |
b5ae6b15 AV |
2788 | out: |
2789 | security_d_instantiate(dentry, inode); | |
2790 | d_rehash(dentry); | |
2791 | return NULL; | |
770bfad8 | 2792 | } |
b5ae6b15 | 2793 | EXPORT_SYMBOL(d_splice_alias); |
770bfad8 | 2794 | |
cdd16d02 | 2795 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2796 | { |
2797 | *buflen -= namelen; | |
2798 | if (*buflen < 0) | |
2799 | return -ENAMETOOLONG; | |
2800 | *buffer -= namelen; | |
2801 | memcpy(*buffer, str, namelen); | |
2802 | return 0; | |
2803 | } | |
2804 | ||
232d2d60 WL |
2805 | /** |
2806 | * prepend_name - prepend a pathname in front of current buffer pointer | |
18129977 WL |
2807 | * @buffer: buffer pointer |
2808 | * @buflen: allocated length of the buffer | |
2809 | * @name: name string and length qstr structure | |
232d2d60 WL |
2810 | * |
2811 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
2812 | * make sure that either the old or the new name pointer and length are | |
2813 | * fetched. However, there may be mismatch between length and pointer. | |
2814 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
2815 | * the length is reached or a null byte is found. It also prepends "/" at | |
2816 | * the beginning of the name. The sequence number check at the caller will | |
2817 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
2818 | * due to mismatched pointer and length will be discarded. | |
6d13f694 AV |
2819 | * |
2820 | * Data dependency barrier is needed to make sure that we see that terminating | |
2821 | * NUL. Alpha strikes again, film at 11... | |
232d2d60 | 2822 | */ |
cdd16d02 MS |
2823 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2824 | { | |
232d2d60 WL |
2825 | const char *dname = ACCESS_ONCE(name->name); |
2826 | u32 dlen = ACCESS_ONCE(name->len); | |
2827 | char *p; | |
2828 | ||
6d13f694 AV |
2829 | smp_read_barrier_depends(); |
2830 | ||
232d2d60 | 2831 | *buflen -= dlen + 1; |
e825196d AV |
2832 | if (*buflen < 0) |
2833 | return -ENAMETOOLONG; | |
232d2d60 WL |
2834 | p = *buffer -= dlen + 1; |
2835 | *p++ = '/'; | |
2836 | while (dlen--) { | |
2837 | char c = *dname++; | |
2838 | if (!c) | |
2839 | break; | |
2840 | *p++ = c; | |
2841 | } | |
2842 | return 0; | |
cdd16d02 MS |
2843 | } |
2844 | ||
1da177e4 | 2845 | /** |
208898c1 | 2846 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2847 | * @path: the dentry/vfsmount to report |
02125a82 | 2848 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2849 | * @buffer: pointer to the end of the buffer |
2850 | * @buflen: pointer to buffer length | |
552ce544 | 2851 | * |
18129977 WL |
2852 | * The function will first try to write out the pathname without taking any |
2853 | * lock other than the RCU read lock to make sure that dentries won't go away. | |
2854 | * It only checks the sequence number of the global rename_lock as any change | |
2855 | * in the dentry's d_seq will be preceded by changes in the rename_lock | |
2856 | * sequence number. If the sequence number had been changed, it will restart | |
2857 | * the whole pathname back-tracing sequence again by taking the rename_lock. | |
2858 | * In this case, there is no need to take the RCU read lock as the recursive | |
2859 | * parent pointer references will keep the dentry chain alive as long as no | |
2860 | * rename operation is performed. | |
1da177e4 | 2861 | */ |
02125a82 AV |
2862 | static int prepend_path(const struct path *path, |
2863 | const struct path *root, | |
f2eb6575 | 2864 | char **buffer, int *buflen) |
1da177e4 | 2865 | { |
ede4cebc AV |
2866 | struct dentry *dentry; |
2867 | struct vfsmount *vfsmnt; | |
2868 | struct mount *mnt; | |
f2eb6575 | 2869 | int error = 0; |
48a066e7 | 2870 | unsigned seq, m_seq = 0; |
232d2d60 WL |
2871 | char *bptr; |
2872 | int blen; | |
6092d048 | 2873 | |
48f5ec21 | 2874 | rcu_read_lock(); |
48a066e7 AV |
2875 | restart_mnt: |
2876 | read_seqbegin_or_lock(&mount_lock, &m_seq); | |
2877 | seq = 0; | |
4ec6c2ae | 2878 | rcu_read_lock(); |
232d2d60 WL |
2879 | restart: |
2880 | bptr = *buffer; | |
2881 | blen = *buflen; | |
48a066e7 | 2882 | error = 0; |
ede4cebc AV |
2883 | dentry = path->dentry; |
2884 | vfsmnt = path->mnt; | |
2885 | mnt = real_mount(vfsmnt); | |
232d2d60 | 2886 | read_seqbegin_or_lock(&rename_lock, &seq); |
f2eb6575 | 2887 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2888 | struct dentry * parent; |
2889 | ||
1da177e4 | 2890 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
48a066e7 | 2891 | struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); |
552ce544 | 2892 | /* Global root? */ |
48a066e7 AV |
2893 | if (mnt != parent) { |
2894 | dentry = ACCESS_ONCE(mnt->mnt_mountpoint); | |
2895 | mnt = parent; | |
232d2d60 WL |
2896 | vfsmnt = &mnt->mnt; |
2897 | continue; | |
2898 | } | |
2899 | /* | |
2900 | * Filesystems needing to implement special "root names" | |
2901 | * should do so with ->d_dname() | |
2902 | */ | |
2903 | if (IS_ROOT(dentry) && | |
2904 | (dentry->d_name.len != 1 || | |
2905 | dentry->d_name.name[0] != '/')) { | |
2906 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2907 | (int) dentry->d_name.len, | |
2908 | dentry->d_name.name); | |
2909 | } | |
2910 | if (!error) | |
2911 | error = is_mounted(vfsmnt) ? 1 : 2; | |
2912 | break; | |
1da177e4 LT |
2913 | } |
2914 | parent = dentry->d_parent; | |
2915 | prefetch(parent); | |
232d2d60 | 2916 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
2917 | if (error) |
2918 | break; | |
2919 | ||
1da177e4 LT |
2920 | dentry = parent; |
2921 | } | |
48f5ec21 AV |
2922 | if (!(seq & 1)) |
2923 | rcu_read_unlock(); | |
2924 | if (need_seqretry(&rename_lock, seq)) { | |
2925 | seq = 1; | |
232d2d60 | 2926 | goto restart; |
48f5ec21 AV |
2927 | } |
2928 | done_seqretry(&rename_lock, seq); | |
4ec6c2ae LZ |
2929 | |
2930 | if (!(m_seq & 1)) | |
2931 | rcu_read_unlock(); | |
48a066e7 AV |
2932 | if (need_seqretry(&mount_lock, m_seq)) { |
2933 | m_seq = 1; | |
2934 | goto restart_mnt; | |
2935 | } | |
2936 | done_seqretry(&mount_lock, m_seq); | |
1da177e4 | 2937 | |
232d2d60 WL |
2938 | if (error >= 0 && bptr == *buffer) { |
2939 | if (--blen < 0) | |
2940 | error = -ENAMETOOLONG; | |
2941 | else | |
2942 | *--bptr = '/'; | |
2943 | } | |
2944 | *buffer = bptr; | |
2945 | *buflen = blen; | |
7ea600b5 | 2946 | return error; |
f2eb6575 | 2947 | } |
be285c71 | 2948 | |
f2eb6575 MS |
2949 | /** |
2950 | * __d_path - return the path of a dentry | |
2951 | * @path: the dentry/vfsmount to report | |
02125a82 | 2952 | * @root: root vfsmnt/dentry |
cd956a1c | 2953 | * @buf: buffer to return value in |
f2eb6575 MS |
2954 | * @buflen: buffer length |
2955 | * | |
ffd1f4ed | 2956 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2957 | * |
2958 | * Returns a pointer into the buffer or an error code if the | |
2959 | * path was too long. | |
2960 | * | |
be148247 | 2961 | * "buflen" should be positive. |
f2eb6575 | 2962 | * |
02125a82 | 2963 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2964 | */ |
02125a82 AV |
2965 | char *__d_path(const struct path *path, |
2966 | const struct path *root, | |
f2eb6575 MS |
2967 | char *buf, int buflen) |
2968 | { | |
2969 | char *res = buf + buflen; | |
2970 | int error; | |
2971 | ||
2972 | prepend(&res, &buflen, "\0", 1); | |
f2eb6575 | 2973 | error = prepend_path(path, root, &res, &buflen); |
be148247 | 2974 | |
02125a82 AV |
2975 | if (error < 0) |
2976 | return ERR_PTR(error); | |
2977 | if (error > 0) | |
2978 | return NULL; | |
2979 | return res; | |
2980 | } | |
2981 | ||
2982 | char *d_absolute_path(const struct path *path, | |
2983 | char *buf, int buflen) | |
2984 | { | |
2985 | struct path root = {}; | |
2986 | char *res = buf + buflen; | |
2987 | int error; | |
2988 | ||
2989 | prepend(&res, &buflen, "\0", 1); | |
02125a82 | 2990 | error = prepend_path(path, &root, &res, &buflen); |
02125a82 AV |
2991 | |
2992 | if (error > 1) | |
2993 | error = -EINVAL; | |
2994 | if (error < 0) | |
f2eb6575 | 2995 | return ERR_PTR(error); |
f2eb6575 | 2996 | return res; |
1da177e4 LT |
2997 | } |
2998 | ||
ffd1f4ed MS |
2999 | /* |
3000 | * same as __d_path but appends "(deleted)" for unlinked files. | |
3001 | */ | |
02125a82 AV |
3002 | static int path_with_deleted(const struct path *path, |
3003 | const struct path *root, | |
3004 | char **buf, int *buflen) | |
ffd1f4ed MS |
3005 | { |
3006 | prepend(buf, buflen, "\0", 1); | |
3007 | if (d_unlinked(path->dentry)) { | |
3008 | int error = prepend(buf, buflen, " (deleted)", 10); | |
3009 | if (error) | |
3010 | return error; | |
3011 | } | |
3012 | ||
3013 | return prepend_path(path, root, buf, buflen); | |
3014 | } | |
3015 | ||
8df9d1a4 MS |
3016 | static int prepend_unreachable(char **buffer, int *buflen) |
3017 | { | |
3018 | return prepend(buffer, buflen, "(unreachable)", 13); | |
3019 | } | |
3020 | ||
68f0d9d9 LT |
3021 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) |
3022 | { | |
3023 | unsigned seq; | |
3024 | ||
3025 | do { | |
3026 | seq = read_seqcount_begin(&fs->seq); | |
3027 | *root = fs->root; | |
3028 | } while (read_seqcount_retry(&fs->seq, seq)); | |
3029 | } | |
3030 | ||
a03a8a70 JB |
3031 | /** |
3032 | * d_path - return the path of a dentry | |
cf28b486 | 3033 | * @path: path to report |
a03a8a70 JB |
3034 | * @buf: buffer to return value in |
3035 | * @buflen: buffer length | |
3036 | * | |
3037 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
3038 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
3039 | * | |
52afeefb AV |
3040 | * Returns a pointer into the buffer or an error code if the path was |
3041 | * too long. Note: Callers should use the returned pointer, not the passed | |
3042 | * in buffer, to use the name! The implementation often starts at an offset | |
3043 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 3044 | * |
31f3e0b3 | 3045 | * "buflen" should be positive. |
a03a8a70 | 3046 | */ |
20d4fdc1 | 3047 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 3048 | { |
ffd1f4ed | 3049 | char *res = buf + buflen; |
6ac08c39 | 3050 | struct path root; |
ffd1f4ed | 3051 | int error; |
1da177e4 | 3052 | |
c23fbb6b ED |
3053 | /* |
3054 | * We have various synthetic filesystems that never get mounted. On | |
3055 | * these filesystems dentries are never used for lookup purposes, and | |
3056 | * thus don't need to be hashed. They also don't need a name until a | |
3057 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
3058 | * below allows us to generate a name for these objects on demand: | |
f48cfddc EB |
3059 | * |
3060 | * Some pseudo inodes are mountable. When they are mounted | |
3061 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname | |
3062 | * and instead have d_path return the mounted path. | |
c23fbb6b | 3063 | */ |
f48cfddc EB |
3064 | if (path->dentry->d_op && path->dentry->d_op->d_dname && |
3065 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) | |
cf28b486 | 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 | 3109 | } |
31bbe16f | 3110 | EXPORT_SYMBOL(simple_dname); |
118b2302 | 3111 | |
6092d048 RP |
3112 | /* |
3113 | * Write full pathname from the root of the filesystem into the buffer. | |
3114 | */ | |
f6500801 | 3115 | static char *__dentry_path(struct dentry *d, char *buf, int buflen) |
6092d048 | 3116 | { |
f6500801 | 3117 | struct dentry *dentry; |
232d2d60 WL |
3118 | char *end, *retval; |
3119 | int len, seq = 0; | |
3120 | int error = 0; | |
6092d048 | 3121 | |
f6500801 AV |
3122 | if (buflen < 2) |
3123 | goto Elong; | |
3124 | ||
48f5ec21 | 3125 | rcu_read_lock(); |
232d2d60 | 3126 | restart: |
f6500801 | 3127 | dentry = d; |
232d2d60 WL |
3128 | end = buf + buflen; |
3129 | len = buflen; | |
3130 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
3131 | /* Get '/' right */ |
3132 | retval = end-1; | |
3133 | *retval = '/'; | |
232d2d60 | 3134 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
3135 | while (!IS_ROOT(dentry)) { |
3136 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 3137 | |
6092d048 | 3138 | prefetch(parent); |
232d2d60 WL |
3139 | error = prepend_name(&end, &len, &dentry->d_name); |
3140 | if (error) | |
3141 | break; | |
6092d048 RP |
3142 | |
3143 | retval = end; | |
3144 | dentry = parent; | |
3145 | } | |
48f5ec21 AV |
3146 | if (!(seq & 1)) |
3147 | rcu_read_unlock(); | |
3148 | if (need_seqretry(&rename_lock, seq)) { | |
3149 | seq = 1; | |
232d2d60 | 3150 | goto restart; |
48f5ec21 AV |
3151 | } |
3152 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3153 | if (error) |
3154 | goto Elong; | |
c103135c AV |
3155 | return retval; |
3156 | Elong: | |
3157 | return ERR_PTR(-ENAMETOOLONG); | |
3158 | } | |
ec2447c2 NP |
3159 | |
3160 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3161 | { | |
232d2d60 | 3162 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3163 | } |
3164 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3165 | |
3166 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3167 | { | |
3168 | char *p = NULL; | |
3169 | char *retval; | |
3170 | ||
c103135c AV |
3171 | if (d_unlinked(dentry)) { |
3172 | p = buf + buflen; | |
3173 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3174 | goto Elong; | |
3175 | buflen++; | |
3176 | } | |
3177 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3178 | if (!IS_ERR(retval) && p) |
3179 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3180 | return retval; |
3181 | Elong: | |
6092d048 RP |
3182 | return ERR_PTR(-ENAMETOOLONG); |
3183 | } | |
3184 | ||
8b19e341 LT |
3185 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, |
3186 | struct path *pwd) | |
5762482f | 3187 | { |
8b19e341 LT |
3188 | unsigned seq; |
3189 | ||
3190 | do { | |
3191 | seq = read_seqcount_begin(&fs->seq); | |
3192 | *root = fs->root; | |
3193 | *pwd = fs->pwd; | |
3194 | } while (read_seqcount_retry(&fs->seq, seq)); | |
5762482f LT |
3195 | } |
3196 | ||
1da177e4 LT |
3197 | /* |
3198 | * NOTE! The user-level library version returns a | |
3199 | * character pointer. The kernel system call just | |
3200 | * returns the length of the buffer filled (which | |
3201 | * includes the ending '\0' character), or a negative | |
3202 | * error value. So libc would do something like | |
3203 | * | |
3204 | * char *getcwd(char * buf, size_t size) | |
3205 | * { | |
3206 | * int retval; | |
3207 | * | |
3208 | * retval = sys_getcwd(buf, size); | |
3209 | * if (retval >= 0) | |
3210 | * return buf; | |
3211 | * errno = -retval; | |
3212 | * return NULL; | |
3213 | * } | |
3214 | */ | |
3cdad428 | 3215 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3216 | { |
552ce544 | 3217 | int error; |
6ac08c39 | 3218 | struct path pwd, root; |
3272c544 | 3219 | char *page = __getname(); |
1da177e4 LT |
3220 | |
3221 | if (!page) | |
3222 | return -ENOMEM; | |
3223 | ||
8b19e341 LT |
3224 | rcu_read_lock(); |
3225 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); | |
1da177e4 | 3226 | |
552ce544 | 3227 | error = -ENOENT; |
f3da392e | 3228 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3229 | unsigned long len; |
3272c544 LT |
3230 | char *cwd = page + PATH_MAX; |
3231 | int buflen = PATH_MAX; | |
1da177e4 | 3232 | |
8df9d1a4 | 3233 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3234 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
ff812d72 | 3235 | rcu_read_unlock(); |
552ce544 | 3236 | |
02125a82 | 3237 | if (error < 0) |
552ce544 LT |
3238 | goto out; |
3239 | ||
8df9d1a4 | 3240 | /* Unreachable from current root */ |
02125a82 | 3241 | if (error > 0) { |
8df9d1a4 MS |
3242 | error = prepend_unreachable(&cwd, &buflen); |
3243 | if (error) | |
3244 | goto out; | |
3245 | } | |
3246 | ||
552ce544 | 3247 | error = -ERANGE; |
3272c544 | 3248 | len = PATH_MAX + page - cwd; |
552ce544 LT |
3249 | if (len <= size) { |
3250 | error = len; | |
3251 | if (copy_to_user(buf, cwd, len)) | |
3252 | error = -EFAULT; | |
3253 | } | |
949854d0 | 3254 | } else { |
ff812d72 | 3255 | rcu_read_unlock(); |
949854d0 | 3256 | } |
1da177e4 LT |
3257 | |
3258 | out: | |
3272c544 | 3259 | __putname(page); |
1da177e4 LT |
3260 | return error; |
3261 | } | |
3262 | ||
3263 | /* | |
3264 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3265 | * | |
3266 | * Trivially implemented using the dcache structure | |
3267 | */ | |
3268 | ||
3269 | /** | |
3270 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3271 | * @new_dentry: new dentry | |
3272 | * @old_dentry: old dentry | |
3273 | * | |
3274 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
3275 | * Returns 0 otherwise. | |
3276 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
3277 | */ | |
3278 | ||
e2761a11 | 3279 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
3280 | { |
3281 | int result; | |
949854d0 | 3282 | unsigned seq; |
1da177e4 | 3283 | |
e2761a11 OH |
3284 | if (new_dentry == old_dentry) |
3285 | return 1; | |
3286 | ||
e2761a11 | 3287 | do { |
1da177e4 | 3288 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3289 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3290 | /* |
3291 | * Need rcu_readlock to protect against the d_parent trashing | |
3292 | * due to d_move | |
3293 | */ | |
3294 | rcu_read_lock(); | |
e2761a11 | 3295 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 3296 | result = 1; |
e2761a11 OH |
3297 | else |
3298 | result = 0; | |
949854d0 | 3299 | rcu_read_unlock(); |
1da177e4 | 3300 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3301 | |
3302 | return result; | |
3303 | } | |
3304 | ||
db14fc3a | 3305 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3306 | { |
db14fc3a MS |
3307 | struct dentry *root = data; |
3308 | if (dentry != root) { | |
3309 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3310 | return D_WALK_SKIP; | |
1da177e4 | 3311 | |
01ddc4ed MS |
3312 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3313 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3314 | dentry->d_lockref.count--; | |
3315 | } | |
1da177e4 | 3316 | } |
db14fc3a MS |
3317 | return D_WALK_CONTINUE; |
3318 | } | |
58db63d0 | 3319 | |
db14fc3a MS |
3320 | void d_genocide(struct dentry *parent) |
3321 | { | |
3322 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3323 | } |
3324 | ||
60545d0d | 3325 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3326 | { |
60545d0d AV |
3327 | inode_dec_link_count(inode); |
3328 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
946e51f2 | 3329 | !hlist_unhashed(&dentry->d_u.d_alias) || |
60545d0d AV |
3330 | !d_unlinked(dentry)); |
3331 | spin_lock(&dentry->d_parent->d_lock); | |
3332 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3333 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3334 | (unsigned long long)inode->i_ino); | |
3335 | spin_unlock(&dentry->d_lock); | |
3336 | spin_unlock(&dentry->d_parent->d_lock); | |
3337 | d_instantiate(dentry, inode); | |
1da177e4 | 3338 | } |
60545d0d | 3339 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3340 | |
3341 | static __initdata unsigned long dhash_entries; | |
3342 | static int __init set_dhash_entries(char *str) | |
3343 | { | |
3344 | if (!str) | |
3345 | return 0; | |
3346 | dhash_entries = simple_strtoul(str, &str, 0); | |
3347 | return 1; | |
3348 | } | |
3349 | __setup("dhash_entries=", set_dhash_entries); | |
3350 | ||
3351 | static void __init dcache_init_early(void) | |
3352 | { | |
074b8517 | 3353 | unsigned int loop; |
1da177e4 LT |
3354 | |
3355 | /* If hashes are distributed across NUMA nodes, defer | |
3356 | * hash allocation until vmalloc space is available. | |
3357 | */ | |
3358 | if (hashdist) | |
3359 | return; | |
3360 | ||
3361 | dentry_hashtable = | |
3362 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3363 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3364 | dhash_entries, |
3365 | 13, | |
3366 | HASH_EARLY, | |
3367 | &d_hash_shift, | |
3368 | &d_hash_mask, | |
31fe62b9 | 3369 | 0, |
1da177e4 LT |
3370 | 0); |
3371 | ||
074b8517 | 3372 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3373 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3374 | } |
3375 | ||
74bf17cf | 3376 | static void __init dcache_init(void) |
1da177e4 | 3377 | { |
074b8517 | 3378 | unsigned int loop; |
1da177e4 LT |
3379 | |
3380 | /* | |
3381 | * A constructor could be added for stable state like the lists, | |
3382 | * but it is probably not worth it because of the cache nature | |
3383 | * of the dcache. | |
3384 | */ | |
0a31bd5f CL |
3385 | dentry_cache = KMEM_CACHE(dentry, |
3386 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3387 | |
3388 | /* Hash may have been set up in dcache_init_early */ | |
3389 | if (!hashdist) | |
3390 | return; | |
3391 | ||
3392 | dentry_hashtable = | |
3393 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3394 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3395 | dhash_entries, |
3396 | 13, | |
3397 | 0, | |
3398 | &d_hash_shift, | |
3399 | &d_hash_mask, | |
31fe62b9 | 3400 | 0, |
1da177e4 LT |
3401 | 0); |
3402 | ||
074b8517 | 3403 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3404 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3405 | } |
3406 | ||
3407 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3408 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3409 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3410 | |
1da177e4 LT |
3411 | EXPORT_SYMBOL(d_genocide); |
3412 | ||
1da177e4 LT |
3413 | void __init vfs_caches_init_early(void) |
3414 | { | |
3415 | dcache_init_early(); | |
3416 | inode_init_early(); | |
3417 | } | |
3418 | ||
3419 | void __init vfs_caches_init(unsigned long mempages) | |
3420 | { | |
3421 | unsigned long reserve; | |
3422 | ||
3423 | /* Base hash sizes on available memory, with a reserve equal to | |
3424 | 150% of current kernel size */ | |
3425 | ||
3426 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3427 | mempages -= reserve; | |
3428 | ||
3429 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3430 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3431 | |
74bf17cf DC |
3432 | dcache_init(); |
3433 | inode_init(); | |
1da177e4 | 3434 | files_init(mempages); |
74bf17cf | 3435 | mnt_init(); |
1da177e4 LT |
3436 | bdev_cache_init(); |
3437 | chrdev_init(); | |
3438 | } |