]>
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> | |
7c0f6ba6 | 29 | #include <linux/uaccess.h> |
1da177e4 LT |
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 | |
cdf01226 DH |
93 | const struct qstr empty_name = QSTR_INIT("", 0); |
94 | EXPORT_SYMBOL(empty_name); | |
95 | const struct qstr slash_name = QSTR_INIT("/", 1); | |
96 | EXPORT_SYMBOL(slash_name); | |
97 | ||
1da177e4 LT |
98 | /* |
99 | * This is the single most critical data structure when it comes | |
100 | * to the dcache: the hashtable for lookups. Somebody should try | |
101 | * to make this good - I've just made it work. | |
102 | * | |
103 | * This hash-function tries to avoid losing too many bits of hash | |
104 | * information, yet avoid using a prime hash-size or similar. | |
105 | */ | |
1da177e4 | 106 | |
fa3536cc ED |
107 | static unsigned int d_hash_mask __read_mostly; |
108 | static unsigned int d_hash_shift __read_mostly; | |
ceb5bdc2 | 109 | |
b07ad996 | 110 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 111 | |
8387ff25 | 112 | static inline struct hlist_bl_head *d_hash(unsigned int hash) |
ceb5bdc2 | 113 | { |
703b5faf | 114 | return dentry_hashtable + (hash >> (32 - d_hash_shift)); |
ceb5bdc2 NP |
115 | } |
116 | ||
94bdd655 AV |
117 | #define IN_LOOKUP_SHIFT 10 |
118 | static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT]; | |
119 | ||
120 | static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent, | |
121 | unsigned int hash) | |
122 | { | |
123 | hash += (unsigned long) parent / L1_CACHE_BYTES; | |
124 | return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT); | |
125 | } | |
126 | ||
127 | ||
1da177e4 LT |
128 | /* Statistics gathering. */ |
129 | struct dentry_stat_t dentry_stat = { | |
130 | .age_limit = 45, | |
131 | }; | |
132 | ||
3942c07c | 133 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 134 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
312d3ca8 CH |
135 | |
136 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
62d36c77 DC |
137 | |
138 | /* | |
139 | * Here we resort to our own counters instead of using generic per-cpu counters | |
140 | * for consistency with what the vfs inode code does. We are expected to harvest | |
141 | * better code and performance by having our own specialized counters. | |
142 | * | |
143 | * Please note that the loop is done over all possible CPUs, not over all online | |
144 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
145 | * on and off. If one of them goes off, we will just keep their counters. | |
146 | * | |
147 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
148 | * please update all vfs counters to match. | |
149 | */ | |
3942c07c | 150 | static long get_nr_dentry(void) |
3e880fb5 NP |
151 | { |
152 | int i; | |
3942c07c | 153 | long sum = 0; |
3e880fb5 NP |
154 | for_each_possible_cpu(i) |
155 | sum += per_cpu(nr_dentry, i); | |
156 | return sum < 0 ? 0 : sum; | |
157 | } | |
158 | ||
62d36c77 DC |
159 | static long get_nr_dentry_unused(void) |
160 | { | |
161 | int i; | |
162 | long sum = 0; | |
163 | for_each_possible_cpu(i) | |
164 | sum += per_cpu(nr_dentry_unused, i); | |
165 | return sum < 0 ? 0 : sum; | |
166 | } | |
167 | ||
1f7e0616 | 168 | int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer, |
312d3ca8 CH |
169 | size_t *lenp, loff_t *ppos) |
170 | { | |
3e880fb5 | 171 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 172 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
3942c07c | 173 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 CH |
174 | } |
175 | #endif | |
176 | ||
5483f18e LT |
177 | /* |
178 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
179 | * The strings are both count bytes long, and count is non-zero. | |
180 | */ | |
e419b4cc LT |
181 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
182 | ||
183 | #include <asm/word-at-a-time.h> | |
184 | /* | |
185 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
186 | * aligned allocation for this particular component. We don't | |
187 | * strictly need the load_unaligned_zeropad() safety, but it | |
188 | * doesn't hurt either. | |
189 | * | |
190 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
191 | * need the careful unaligned handling. | |
192 | */ | |
94753db5 | 193 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 194 | { |
bfcfaa77 | 195 | unsigned long a,b,mask; |
bfcfaa77 LT |
196 | |
197 | for (;;) { | |
12f8ad4b | 198 | a = *(unsigned long *)cs; |
e419b4cc | 199 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
200 | if (tcount < sizeof(unsigned long)) |
201 | break; | |
202 | if (unlikely(a != b)) | |
203 | return 1; | |
204 | cs += sizeof(unsigned long); | |
205 | ct += sizeof(unsigned long); | |
206 | tcount -= sizeof(unsigned long); | |
207 | if (!tcount) | |
208 | return 0; | |
209 | } | |
a5c21dce | 210 | mask = bytemask_from_count(tcount); |
bfcfaa77 | 211 | return unlikely(!!((a ^ b) & mask)); |
e419b4cc LT |
212 | } |
213 | ||
bfcfaa77 | 214 | #else |
e419b4cc | 215 | |
94753db5 | 216 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 217 | { |
5483f18e LT |
218 | do { |
219 | if (*cs != *ct) | |
220 | return 1; | |
221 | cs++; | |
222 | ct++; | |
223 | tcount--; | |
224 | } while (tcount); | |
225 | return 0; | |
226 | } | |
227 | ||
e419b4cc LT |
228 | #endif |
229 | ||
94753db5 LT |
230 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
231 | { | |
94753db5 LT |
232 | /* |
233 | * Be careful about RCU walk racing with rename: | |
5383f45d | 234 | * use 'READ_ONCE' to fetch the name pointer. |
94753db5 LT |
235 | * |
236 | * NOTE! Even if a rename will mean that the length | |
237 | * was not loaded atomically, we don't care. The | |
238 | * RCU walk will check the sequence count eventually, | |
239 | * and catch it. And we won't overrun the buffer, | |
240 | * because we're reading the name pointer atomically, | |
241 | * and a dentry name is guaranteed to be properly | |
242 | * terminated with a NUL byte. | |
243 | * | |
244 | * End result: even if 'len' is wrong, we'll exit | |
245 | * early because the data cannot match (there can | |
246 | * be no NUL in the ct/tcount data) | |
247 | */ | |
5383f45d | 248 | const unsigned char *cs = READ_ONCE(dentry->d_name.name); |
ae0a843c | 249 | |
6326c71f | 250 | return dentry_string_cmp(cs, ct, tcount); |
94753db5 LT |
251 | } |
252 | ||
8d85b484 AV |
253 | struct external_name { |
254 | union { | |
255 | atomic_t count; | |
256 | struct rcu_head head; | |
257 | } u; | |
258 | unsigned char name[]; | |
259 | }; | |
260 | ||
261 | static inline struct external_name *external_name(struct dentry *dentry) | |
262 | { | |
263 | return container_of(dentry->d_name.name, struct external_name, name[0]); | |
264 | } | |
265 | ||
9c82ab9c | 266 | static void __d_free(struct rcu_head *head) |
1da177e4 | 267 | { |
9c82ab9c CH |
268 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
269 | ||
8d85b484 AV |
270 | kmem_cache_free(dentry_cache, dentry); |
271 | } | |
272 | ||
6d794237 RG |
273 | static void __d_free_external_name(struct rcu_head *head) |
274 | { | |
275 | struct external_name *name = container_of(head, struct external_name, | |
276 | u.head); | |
277 | ||
278 | mod_node_page_state(page_pgdat(virt_to_page(name)), | |
279 | NR_INDIRECTLY_RECLAIMABLE_BYTES, | |
280 | -ksize(name)); | |
281 | ||
282 | kfree(name); | |
283 | } | |
284 | ||
8d85b484 AV |
285 | static void __d_free_external(struct rcu_head *head) |
286 | { | |
287 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
6d794237 RG |
288 | |
289 | __d_free_external_name(&external_name(dentry)->u.head); | |
290 | ||
291 | kmem_cache_free(dentry_cache, dentry); | |
1da177e4 LT |
292 | } |
293 | ||
810bb172 AV |
294 | static inline int dname_external(const struct dentry *dentry) |
295 | { | |
296 | return dentry->d_name.name != dentry->d_iname; | |
297 | } | |
298 | ||
49d31c2f AV |
299 | void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry) |
300 | { | |
301 | spin_lock(&dentry->d_lock); | |
302 | if (unlikely(dname_external(dentry))) { | |
303 | struct external_name *p = external_name(dentry); | |
304 | atomic_inc(&p->u.count); | |
305 | spin_unlock(&dentry->d_lock); | |
306 | name->name = p->name; | |
307 | } else { | |
e20c4abf TH |
308 | memcpy(name->inline_name, dentry->d_iname, |
309 | dentry->d_name.len + 1); | |
49d31c2f AV |
310 | spin_unlock(&dentry->d_lock); |
311 | name->name = name->inline_name; | |
312 | } | |
313 | } | |
314 | EXPORT_SYMBOL(take_dentry_name_snapshot); | |
315 | ||
316 | void release_dentry_name_snapshot(struct name_snapshot *name) | |
317 | { | |
318 | if (unlikely(name->name != name->inline_name)) { | |
319 | struct external_name *p; | |
320 | p = container_of(name->name, struct external_name, name[0]); | |
321 | if (unlikely(atomic_dec_and_test(&p->u.count))) | |
6d794237 | 322 | call_rcu(&p->u.head, __d_free_external_name); |
49d31c2f AV |
323 | } |
324 | } | |
325 | EXPORT_SYMBOL(release_dentry_name_snapshot); | |
326 | ||
4bf46a27 DH |
327 | static inline void __d_set_inode_and_type(struct dentry *dentry, |
328 | struct inode *inode, | |
329 | unsigned type_flags) | |
330 | { | |
331 | unsigned flags; | |
332 | ||
333 | dentry->d_inode = inode; | |
4bf46a27 DH |
334 | flags = READ_ONCE(dentry->d_flags); |
335 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
336 | flags |= type_flags; | |
337 | WRITE_ONCE(dentry->d_flags, flags); | |
338 | } | |
339 | ||
4bf46a27 DH |
340 | static inline void __d_clear_type_and_inode(struct dentry *dentry) |
341 | { | |
342 | unsigned flags = READ_ONCE(dentry->d_flags); | |
343 | ||
344 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
345 | WRITE_ONCE(dentry->d_flags, flags); | |
4bf46a27 DH |
346 | dentry->d_inode = NULL; |
347 | } | |
348 | ||
b4f0354e AV |
349 | static void dentry_free(struct dentry *dentry) |
350 | { | |
946e51f2 | 351 | WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias)); |
8d85b484 AV |
352 | if (unlikely(dname_external(dentry))) { |
353 | struct external_name *p = external_name(dentry); | |
354 | if (likely(atomic_dec_and_test(&p->u.count))) { | |
355 | call_rcu(&dentry->d_u.d_rcu, __d_free_external); | |
356 | return; | |
357 | } | |
358 | } | |
b4f0354e AV |
359 | /* if dentry was never visible to RCU, immediate free is OK */ |
360 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
361 | __d_free(&dentry->d_u.d_rcu); | |
362 | else | |
363 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
364 | } | |
365 | ||
1da177e4 LT |
366 | /* |
367 | * Release the dentry's inode, using the filesystem | |
550dce01 | 368 | * d_iput() operation if defined. |
31e6b01f NP |
369 | */ |
370 | static void dentry_unlink_inode(struct dentry * dentry) | |
371 | __releases(dentry->d_lock) | |
873feea0 | 372 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
373 | { |
374 | struct inode *inode = dentry->d_inode; | |
a528aca7 | 375 | |
d5426a38 | 376 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 377 | __d_clear_type_and_inode(dentry); |
946e51f2 | 378 | hlist_del_init(&dentry->d_u.d_alias); |
d5426a38 | 379 | raw_write_seqcount_end(&dentry->d_seq); |
31e6b01f | 380 | spin_unlock(&dentry->d_lock); |
873feea0 | 381 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
382 | if (!inode->i_nlink) |
383 | fsnotify_inoderemove(inode); | |
384 | if (dentry->d_op && dentry->d_op->d_iput) | |
385 | dentry->d_op->d_iput(dentry, inode); | |
386 | else | |
387 | iput(inode); | |
388 | } | |
389 | ||
89dc77bc LT |
390 | /* |
391 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
392 | * is in use - which includes both the "real" per-superblock | |
393 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
394 | * | |
395 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
396 | * on the shrink list (ie not on the superblock LRU list). | |
397 | * | |
398 | * The per-cpu "nr_dentry_unused" counters are updated with | |
399 | * the DCACHE_LRU_LIST bit. | |
400 | * | |
401 | * These helper functions make sure we always follow the | |
402 | * rules. d_lock must be held by the caller. | |
403 | */ | |
404 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
405 | static void d_lru_add(struct dentry *dentry) | |
406 | { | |
407 | D_FLAG_VERIFY(dentry, 0); | |
408 | dentry->d_flags |= DCACHE_LRU_LIST; | |
409 | this_cpu_inc(nr_dentry_unused); | |
410 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
411 | } | |
412 | ||
413 | static void d_lru_del(struct dentry *dentry) | |
414 | { | |
415 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
416 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
417 | this_cpu_dec(nr_dentry_unused); | |
418 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
419 | } | |
420 | ||
421 | static void d_shrink_del(struct dentry *dentry) | |
422 | { | |
423 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
424 | list_del_init(&dentry->d_lru); | |
425 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
426 | this_cpu_dec(nr_dentry_unused); | |
427 | } | |
428 | ||
429 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
430 | { | |
431 | D_FLAG_VERIFY(dentry, 0); | |
432 | list_add(&dentry->d_lru, list); | |
433 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
434 | this_cpu_inc(nr_dentry_unused); | |
435 | } | |
436 | ||
437 | /* | |
438 | * These can only be called under the global LRU lock, ie during the | |
439 | * callback for freeing the LRU list. "isolate" removes it from the | |
440 | * LRU lists entirely, while shrink_move moves it to the indicated | |
441 | * private list. | |
442 | */ | |
3f97b163 | 443 | static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry) |
89dc77bc LT |
444 | { |
445 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
446 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
447 | this_cpu_dec(nr_dentry_unused); | |
3f97b163 | 448 | list_lru_isolate(lru, &dentry->d_lru); |
89dc77bc LT |
449 | } |
450 | ||
3f97b163 VD |
451 | static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry, |
452 | struct list_head *list) | |
89dc77bc LT |
453 | { |
454 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
455 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
3f97b163 | 456 | list_lru_isolate_move(lru, &dentry->d_lru, list); |
89dc77bc LT |
457 | } |
458 | ||
da3bbdd4 | 459 | /* |
f6041567 | 460 | * dentry_lru_(add|del)_list) must be called with d_lock held. |
da3bbdd4 KM |
461 | */ |
462 | static void dentry_lru_add(struct dentry *dentry) | |
463 | { | |
89dc77bc LT |
464 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) |
465 | d_lru_add(dentry); | |
563f4001 JB |
466 | else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED))) |
467 | dentry->d_flags |= DCACHE_REFERENCED; | |
da3bbdd4 KM |
468 | } |
469 | ||
789680d1 NP |
470 | /** |
471 | * d_drop - drop a dentry | |
472 | * @dentry: dentry to drop | |
473 | * | |
474 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
475 | * be found through a VFS lookup any more. Note that this is different from | |
476 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
477 | * possible, giving a successful _negative_ lookup, while d_drop will | |
478 | * just make the cache lookup fail. | |
479 | * | |
480 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
481 | * reason (NFS timeouts or autofs deletes). | |
482 | * | |
db470ce8 N |
483 | * __d_drop requires dentry->d_lock |
484 | * ___d_drop doesn't mark dentry as "unhashed" | |
485 | * (dentry->d_hash.pprev will be LIST_POISON2, not NULL). | |
789680d1 | 486 | */ |
db470ce8 | 487 | static void ___d_drop(struct dentry *dentry) |
789680d1 | 488 | { |
dea3667b | 489 | if (!d_unhashed(dentry)) { |
b61625d2 | 490 | struct hlist_bl_head *b; |
7632e465 BF |
491 | /* |
492 | * Hashed dentries are normally on the dentry hashtable, | |
493 | * with the exception of those newly allocated by | |
494 | * d_obtain_alias, which are always IS_ROOT: | |
495 | */ | |
496 | if (unlikely(IS_ROOT(dentry))) | |
b61625d2 AV |
497 | b = &dentry->d_sb->s_anon; |
498 | else | |
8387ff25 | 499 | b = d_hash(dentry->d_name.hash); |
b61625d2 AV |
500 | |
501 | hlist_bl_lock(b); | |
502 | __hlist_bl_del(&dentry->d_hash); | |
b61625d2 | 503 | hlist_bl_unlock(b); |
d614146d AV |
504 | /* After this call, in-progress rcu-walk path lookup will fail. */ |
505 | write_seqcount_invalidate(&dentry->d_seq); | |
789680d1 NP |
506 | } |
507 | } | |
db470ce8 N |
508 | |
509 | void __d_drop(struct dentry *dentry) | |
510 | { | |
511 | ___d_drop(dentry); | |
512 | dentry->d_hash.pprev = NULL; | |
513 | } | |
789680d1 NP |
514 | EXPORT_SYMBOL(__d_drop); |
515 | ||
516 | void d_drop(struct dentry *dentry) | |
517 | { | |
789680d1 NP |
518 | spin_lock(&dentry->d_lock); |
519 | __d_drop(dentry); | |
520 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
521 | } |
522 | EXPORT_SYMBOL(d_drop); | |
523 | ||
ba65dc5e AV |
524 | static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent) |
525 | { | |
526 | struct dentry *next; | |
527 | /* | |
528 | * Inform d_walk() and shrink_dentry_list() that we are no longer | |
529 | * attached to the dentry tree | |
530 | */ | |
531 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
532 | if (unlikely(list_empty(&dentry->d_child))) | |
533 | return; | |
534 | __list_del_entry(&dentry->d_child); | |
535 | /* | |
536 | * Cursors can move around the list of children. While we'd been | |
537 | * a normal list member, it didn't matter - ->d_child.next would've | |
538 | * been updated. However, from now on it won't be and for the | |
539 | * things like d_walk() it might end up with a nasty surprise. | |
540 | * Normally d_walk() doesn't care about cursors moving around - | |
541 | * ->d_lock on parent prevents that and since a cursor has no children | |
542 | * of its own, we get through it without ever unlocking the parent. | |
543 | * There is one exception, though - if we ascend from a child that | |
544 | * gets killed as soon as we unlock it, the next sibling is found | |
545 | * using the value left in its ->d_child.next. And if _that_ | |
546 | * pointed to a cursor, and cursor got moved (e.g. by lseek()) | |
547 | * before d_walk() regains parent->d_lock, we'll end up skipping | |
548 | * everything the cursor had been moved past. | |
549 | * | |
550 | * Solution: make sure that the pointer left behind in ->d_child.next | |
551 | * points to something that won't be moving around. I.e. skip the | |
552 | * cursors. | |
553 | */ | |
554 | while (dentry->d_child.next != &parent->d_subdirs) { | |
555 | next = list_entry(dentry->d_child.next, struct dentry, d_child); | |
556 | if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR))) | |
557 | break; | |
558 | dentry->d_child.next = next->d_child.next; | |
559 | } | |
560 | } | |
561 | ||
e55fd011 | 562 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 563 | { |
41edf278 AV |
564 | struct dentry *parent = NULL; |
565 | bool can_free = true; | |
41edf278 | 566 | if (!IS_ROOT(dentry)) |
77812a1e | 567 | parent = dentry->d_parent; |
31e6b01f | 568 | |
0d98439e LT |
569 | /* |
570 | * The dentry is now unrecoverably dead to the world. | |
571 | */ | |
572 | lockref_mark_dead(&dentry->d_lockref); | |
573 | ||
f0023bc6 | 574 | /* |
f0023bc6 SW |
575 | * inform the fs via d_prune that this dentry is about to be |
576 | * unhashed and destroyed. | |
577 | */ | |
29266201 | 578 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
61572bb1 YZ |
579 | dentry->d_op->d_prune(dentry); |
580 | ||
01b60351 AV |
581 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
582 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
583 | d_lru_del(dentry); | |
01b60351 | 584 | } |
77812a1e NP |
585 | /* if it was on the hash then remove it */ |
586 | __d_drop(dentry); | |
ba65dc5e | 587 | dentry_unlist(dentry, parent); |
03b3b889 AV |
588 | if (parent) |
589 | spin_unlock(&parent->d_lock); | |
550dce01 AV |
590 | if (dentry->d_inode) |
591 | dentry_unlink_inode(dentry); | |
592 | else | |
593 | spin_unlock(&dentry->d_lock); | |
03b3b889 AV |
594 | this_cpu_dec(nr_dentry); |
595 | if (dentry->d_op && dentry->d_op->d_release) | |
596 | dentry->d_op->d_release(dentry); | |
597 | ||
41edf278 AV |
598 | spin_lock(&dentry->d_lock); |
599 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
600 | dentry->d_flags |= DCACHE_MAY_FREE; | |
601 | can_free = false; | |
602 | } | |
603 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
604 | if (likely(can_free)) |
605 | dentry_free(dentry); | |
e55fd011 AV |
606 | } |
607 | ||
608 | /* | |
609 | * Finish off a dentry we've decided to kill. | |
610 | * dentry->d_lock must be held, returns with it unlocked. | |
611 | * If ref is non-zero, then decrement the refcount too. | |
612 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
613 | */ | |
8cbf74da | 614 | static struct dentry *dentry_kill(struct dentry *dentry) |
e55fd011 AV |
615 | __releases(dentry->d_lock) |
616 | { | |
617 | struct inode *inode = dentry->d_inode; | |
618 | struct dentry *parent = NULL; | |
619 | ||
620 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
621 | goto failed; | |
622 | ||
623 | if (!IS_ROOT(dentry)) { | |
624 | parent = dentry->d_parent; | |
625 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
626 | if (inode) | |
627 | spin_unlock(&inode->i_lock); | |
628 | goto failed; | |
629 | } | |
630 | } | |
631 | ||
632 | __dentry_kill(dentry); | |
03b3b889 | 633 | return parent; |
e55fd011 AV |
634 | |
635 | failed: | |
8cbf74da | 636 | spin_unlock(&dentry->d_lock); |
e55fd011 | 637 | return dentry; /* try again with same dentry */ |
77812a1e NP |
638 | } |
639 | ||
046b961b AV |
640 | static inline struct dentry *lock_parent(struct dentry *dentry) |
641 | { | |
642 | struct dentry *parent = dentry->d_parent; | |
643 | if (IS_ROOT(dentry)) | |
644 | return NULL; | |
360f5479 | 645 | if (unlikely(dentry->d_lockref.count < 0)) |
c2338f2d | 646 | return NULL; |
046b961b AV |
647 | if (likely(spin_trylock(&parent->d_lock))) |
648 | return parent; | |
046b961b | 649 | rcu_read_lock(); |
c2338f2d | 650 | spin_unlock(&dentry->d_lock); |
046b961b AV |
651 | again: |
652 | parent = ACCESS_ONCE(dentry->d_parent); | |
653 | spin_lock(&parent->d_lock); | |
654 | /* | |
655 | * We can't blindly lock dentry until we are sure | |
656 | * that we won't violate the locking order. | |
657 | * Any changes of dentry->d_parent must have | |
658 | * been done with parent->d_lock held, so | |
659 | * spin_lock() above is enough of a barrier | |
660 | * for checking if it's still our child. | |
661 | */ | |
662 | if (unlikely(parent != dentry->d_parent)) { | |
663 | spin_unlock(&parent->d_lock); | |
664 | goto again; | |
665 | } | |
b071bce3 | 666 | if (parent != dentry) { |
9f12600f | 667 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
b071bce3 AV |
668 | if (unlikely(dentry->d_lockref.count < 0)) { |
669 | spin_unlock(&parent->d_lock); | |
670 | parent = NULL; | |
671 | } | |
672 | } else { | |
046b961b | 673 | parent = NULL; |
b071bce3 AV |
674 | } |
675 | rcu_read_unlock(); | |
046b961b AV |
676 | return parent; |
677 | } | |
678 | ||
360f5479 LT |
679 | /* |
680 | * Try to do a lockless dput(), and return whether that was successful. | |
681 | * | |
682 | * If unsuccessful, we return false, having already taken the dentry lock. | |
683 | * | |
684 | * The caller needs to hold the RCU read lock, so that the dentry is | |
685 | * guaranteed to stay around even if the refcount goes down to zero! | |
686 | */ | |
687 | static inline bool fast_dput(struct dentry *dentry) | |
688 | { | |
689 | int ret; | |
690 | unsigned int d_flags; | |
691 | ||
692 | /* | |
693 | * If we have a d_op->d_delete() operation, we sould not | |
75a6f82a | 694 | * let the dentry count go to zero, so use "put_or_lock". |
360f5479 LT |
695 | */ |
696 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) | |
697 | return lockref_put_or_lock(&dentry->d_lockref); | |
698 | ||
699 | /* | |
700 | * .. otherwise, we can try to just decrement the | |
701 | * lockref optimistically. | |
702 | */ | |
703 | ret = lockref_put_return(&dentry->d_lockref); | |
704 | ||
705 | /* | |
706 | * If the lockref_put_return() failed due to the lock being held | |
707 | * by somebody else, the fast path has failed. We will need to | |
708 | * get the lock, and then check the count again. | |
709 | */ | |
710 | if (unlikely(ret < 0)) { | |
711 | spin_lock(&dentry->d_lock); | |
712 | if (dentry->d_lockref.count > 1) { | |
713 | dentry->d_lockref.count--; | |
714 | spin_unlock(&dentry->d_lock); | |
715 | return 1; | |
716 | } | |
717 | return 0; | |
718 | } | |
719 | ||
720 | /* | |
721 | * If we weren't the last ref, we're done. | |
722 | */ | |
723 | if (ret) | |
724 | return 1; | |
725 | ||
726 | /* | |
727 | * Careful, careful. The reference count went down | |
728 | * to zero, but we don't hold the dentry lock, so | |
729 | * somebody else could get it again, and do another | |
730 | * dput(), and we need to not race with that. | |
731 | * | |
732 | * However, there is a very special and common case | |
733 | * where we don't care, because there is nothing to | |
734 | * do: the dentry is still hashed, it does not have | |
735 | * a 'delete' op, and it's referenced and already on | |
736 | * the LRU list. | |
737 | * | |
738 | * NOTE! Since we aren't locked, these values are | |
739 | * not "stable". However, it is sufficient that at | |
740 | * some point after we dropped the reference the | |
741 | * dentry was hashed and the flags had the proper | |
742 | * value. Other dentry users may have re-gotten | |
743 | * a reference to the dentry and change that, but | |
744 | * our work is done - we can leave the dentry | |
745 | * around with a zero refcount. | |
746 | */ | |
747 | smp_rmb(); | |
748 | d_flags = ACCESS_ONCE(dentry->d_flags); | |
75a6f82a | 749 | d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED; |
360f5479 LT |
750 | |
751 | /* Nothing to do? Dropping the reference was all we needed? */ | |
752 | if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry)) | |
753 | return 1; | |
754 | ||
755 | /* | |
756 | * Not the fast normal case? Get the lock. We've already decremented | |
757 | * the refcount, but we'll need to re-check the situation after | |
758 | * getting the lock. | |
759 | */ | |
760 | spin_lock(&dentry->d_lock); | |
761 | ||
762 | /* | |
763 | * Did somebody else grab a reference to it in the meantime, and | |
764 | * we're no longer the last user after all? Alternatively, somebody | |
765 | * else could have killed it and marked it dead. Either way, we | |
766 | * don't need to do anything else. | |
767 | */ | |
768 | if (dentry->d_lockref.count) { | |
769 | spin_unlock(&dentry->d_lock); | |
770 | return 1; | |
771 | } | |
772 | ||
773 | /* | |
774 | * Re-get the reference we optimistically dropped. We hold the | |
775 | * lock, and we just tested that it was zero, so we can just | |
776 | * set it to 1. | |
777 | */ | |
778 | dentry->d_lockref.count = 1; | |
779 | return 0; | |
780 | } | |
781 | ||
782 | ||
1da177e4 LT |
783 | /* |
784 | * This is dput | |
785 | * | |
786 | * This is complicated by the fact that we do not want to put | |
787 | * dentries that are no longer on any hash chain on the unused | |
788 | * list: we'd much rather just get rid of them immediately. | |
789 | * | |
790 | * However, that implies that we have to traverse the dentry | |
791 | * tree upwards to the parents which might _also_ now be | |
792 | * scheduled for deletion (it may have been only waiting for | |
793 | * its last child to go away). | |
794 | * | |
795 | * This tail recursion is done by hand as we don't want to depend | |
796 | * on the compiler to always get this right (gcc generally doesn't). | |
797 | * Real recursion would eat up our stack space. | |
798 | */ | |
799 | ||
800 | /* | |
801 | * dput - release a dentry | |
802 | * @dentry: dentry to release | |
803 | * | |
804 | * Release a dentry. This will drop the usage count and if appropriate | |
805 | * call the dentry unlink method as well as removing it from the queues and | |
806 | * releasing its resources. If the parent dentries were scheduled for release | |
807 | * they too may now get deleted. | |
1da177e4 | 808 | */ |
1da177e4 LT |
809 | void dput(struct dentry *dentry) |
810 | { | |
8aab6a27 | 811 | if (unlikely(!dentry)) |
1da177e4 LT |
812 | return; |
813 | ||
814 | repeat: | |
47be6184 WF |
815 | might_sleep(); |
816 | ||
360f5479 LT |
817 | rcu_read_lock(); |
818 | if (likely(fast_dput(dentry))) { | |
819 | rcu_read_unlock(); | |
1da177e4 | 820 | return; |
360f5479 LT |
821 | } |
822 | ||
823 | /* Slow case: now with the dentry lock held */ | |
824 | rcu_read_unlock(); | |
1da177e4 | 825 | |
85c7f810 AV |
826 | WARN_ON(d_in_lookup(dentry)); |
827 | ||
8aab6a27 LT |
828 | /* Unreachable? Get rid of it */ |
829 | if (unlikely(d_unhashed(dentry))) | |
830 | goto kill_it; | |
831 | ||
75a6f82a AV |
832 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) |
833 | goto kill_it; | |
834 | ||
8aab6a27 | 835 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { |
1da177e4 | 836 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 837 | goto kill_it; |
1da177e4 | 838 | } |
265ac902 | 839 | |
a4633357 | 840 | dentry_lru_add(dentry); |
265ac902 | 841 | |
98474236 | 842 | dentry->d_lockref.count--; |
61f3dee4 | 843 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
844 | return; |
845 | ||
d52b9086 | 846 | kill_it: |
8cbf74da | 847 | dentry = dentry_kill(dentry); |
47be6184 WF |
848 | if (dentry) { |
849 | cond_resched(); | |
d52b9086 | 850 | goto repeat; |
47be6184 | 851 | } |
1da177e4 | 852 | } |
ec4f8605 | 853 | EXPORT_SYMBOL(dput); |
1da177e4 | 854 | |
1da177e4 | 855 | |
b5c84bf6 | 856 | /* This must be called with d_lock held */ |
dc0474be | 857 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 858 | { |
98474236 | 859 | dentry->d_lockref.count++; |
23044507 NP |
860 | } |
861 | ||
dc0474be | 862 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 863 | { |
98474236 | 864 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
865 | } |
866 | ||
b7ab39f6 NP |
867 | struct dentry *dget_parent(struct dentry *dentry) |
868 | { | |
df3d0bbc | 869 | int gotref; |
b7ab39f6 NP |
870 | struct dentry *ret; |
871 | ||
df3d0bbc WL |
872 | /* |
873 | * Do optimistic parent lookup without any | |
874 | * locking. | |
875 | */ | |
876 | rcu_read_lock(); | |
877 | ret = ACCESS_ONCE(dentry->d_parent); | |
878 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
879 | rcu_read_unlock(); | |
880 | if (likely(gotref)) { | |
881 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
882 | return ret; | |
883 | dput(ret); | |
884 | } | |
885 | ||
b7ab39f6 | 886 | repeat: |
a734eb45 NP |
887 | /* |
888 | * Don't need rcu_dereference because we re-check it was correct under | |
889 | * the lock. | |
890 | */ | |
891 | rcu_read_lock(); | |
b7ab39f6 | 892 | ret = dentry->d_parent; |
a734eb45 NP |
893 | spin_lock(&ret->d_lock); |
894 | if (unlikely(ret != dentry->d_parent)) { | |
895 | spin_unlock(&ret->d_lock); | |
896 | rcu_read_unlock(); | |
b7ab39f6 NP |
897 | goto repeat; |
898 | } | |
a734eb45 | 899 | rcu_read_unlock(); |
98474236 WL |
900 | BUG_ON(!ret->d_lockref.count); |
901 | ret->d_lockref.count++; | |
b7ab39f6 | 902 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
903 | return ret; |
904 | } | |
905 | EXPORT_SYMBOL(dget_parent); | |
906 | ||
1da177e4 LT |
907 | /** |
908 | * d_find_alias - grab a hashed alias of inode | |
909 | * @inode: inode in question | |
1da177e4 LT |
910 | * |
911 | * If inode has a hashed alias, or is a directory and has any alias, | |
912 | * acquire the reference to alias and return it. Otherwise return NULL. | |
913 | * Notice that if inode is a directory there can be only one alias and | |
914 | * it can be unhashed only if it has no children, or if it is the root | |
3ccb354d EB |
915 | * of a filesystem, or if the directory was renamed and d_revalidate |
916 | * was the first vfs operation to notice. | |
1da177e4 | 917 | * |
21c0d8fd | 918 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
52ed46f0 | 919 | * any other hashed alias over that one. |
1da177e4 | 920 | */ |
52ed46f0 | 921 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 922 | { |
da502956 | 923 | struct dentry *alias, *discon_alias; |
1da177e4 | 924 | |
da502956 NP |
925 | again: |
926 | discon_alias = NULL; | |
946e51f2 | 927 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
da502956 | 928 | spin_lock(&alias->d_lock); |
1da177e4 | 929 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 930 | if (IS_ROOT(alias) && |
da502956 | 931 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 932 | discon_alias = alias; |
52ed46f0 | 933 | } else { |
dc0474be | 934 | __dget_dlock(alias); |
da502956 NP |
935 | spin_unlock(&alias->d_lock); |
936 | return alias; | |
937 | } | |
938 | } | |
939 | spin_unlock(&alias->d_lock); | |
940 | } | |
941 | if (discon_alias) { | |
942 | alias = discon_alias; | |
943 | spin_lock(&alias->d_lock); | |
944 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
8d80d7da BF |
945 | __dget_dlock(alias); |
946 | spin_unlock(&alias->d_lock); | |
947 | return alias; | |
1da177e4 | 948 | } |
da502956 NP |
949 | spin_unlock(&alias->d_lock); |
950 | goto again; | |
1da177e4 | 951 | } |
da502956 | 952 | return NULL; |
1da177e4 LT |
953 | } |
954 | ||
da502956 | 955 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 956 | { |
214fda1f DH |
957 | struct dentry *de = NULL; |
958 | ||
b3d9b7a3 | 959 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 960 | spin_lock(&inode->i_lock); |
52ed46f0 | 961 | de = __d_find_alias(inode); |
873feea0 | 962 | spin_unlock(&inode->i_lock); |
214fda1f | 963 | } |
1da177e4 LT |
964 | return de; |
965 | } | |
ec4f8605 | 966 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
967 | |
968 | /* | |
969 | * Try to kill dentries associated with this inode. | |
970 | * WARNING: you must own a reference to inode. | |
971 | */ | |
972 | void d_prune_aliases(struct inode *inode) | |
973 | { | |
0cdca3f9 | 974 | struct dentry *dentry; |
1da177e4 | 975 | restart: |
873feea0 | 976 | spin_lock(&inode->i_lock); |
946e51f2 | 977 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { |
1da177e4 | 978 | spin_lock(&dentry->d_lock); |
98474236 | 979 | if (!dentry->d_lockref.count) { |
29355c39 AV |
980 | struct dentry *parent = lock_parent(dentry); |
981 | if (likely(!dentry->d_lockref.count)) { | |
982 | __dentry_kill(dentry); | |
4a7795d3 | 983 | dput(parent); |
29355c39 AV |
984 | goto restart; |
985 | } | |
986 | if (parent) | |
987 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
988 | } |
989 | spin_unlock(&dentry->d_lock); | |
990 | } | |
873feea0 | 991 | spin_unlock(&inode->i_lock); |
1da177e4 | 992 | } |
ec4f8605 | 993 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 994 | |
3049cfe2 | 995 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 996 | { |
5c47e6d0 | 997 | struct dentry *dentry, *parent; |
da3bbdd4 | 998 | |
60942f2f | 999 | while (!list_empty(list)) { |
ff2fde99 | 1000 | struct inode *inode; |
60942f2f | 1001 | dentry = list_entry(list->prev, struct dentry, d_lru); |
ec33679d | 1002 | spin_lock(&dentry->d_lock); |
046b961b AV |
1003 | parent = lock_parent(dentry); |
1004 | ||
dd1f6b2e DC |
1005 | /* |
1006 | * The dispose list is isolated and dentries are not accounted | |
1007 | * to the LRU here, so we can simply remove it from the list | |
1008 | * here regardless of whether it is referenced or not. | |
1009 | */ | |
89dc77bc | 1010 | d_shrink_del(dentry); |
dd1f6b2e | 1011 | |
1da177e4 LT |
1012 | /* |
1013 | * We found an inuse dentry which was not removed from | |
dd1f6b2e | 1014 | * the LRU because of laziness during lookup. Do not free it. |
1da177e4 | 1015 | */ |
360f5479 | 1016 | if (dentry->d_lockref.count > 0) { |
da3bbdd4 | 1017 | spin_unlock(&dentry->d_lock); |
046b961b AV |
1018 | if (parent) |
1019 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
1020 | continue; |
1021 | } | |
77812a1e | 1022 | |
64fd72e0 AV |
1023 | |
1024 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) { | |
1025 | bool can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
1026 | spin_unlock(&dentry->d_lock); | |
046b961b AV |
1027 | if (parent) |
1028 | spin_unlock(&parent->d_lock); | |
64fd72e0 AV |
1029 | if (can_free) |
1030 | dentry_free(dentry); | |
1031 | continue; | |
1032 | } | |
1033 | ||
ff2fde99 AV |
1034 | inode = dentry->d_inode; |
1035 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
89dc77bc | 1036 | d_shrink_add(dentry, list); |
dd1f6b2e | 1037 | spin_unlock(&dentry->d_lock); |
046b961b AV |
1038 | if (parent) |
1039 | spin_unlock(&parent->d_lock); | |
5c47e6d0 | 1040 | continue; |
dd1f6b2e | 1041 | } |
ff2fde99 | 1042 | |
ff2fde99 | 1043 | __dentry_kill(dentry); |
046b961b | 1044 | |
5c47e6d0 AV |
1045 | /* |
1046 | * We need to prune ancestors too. This is necessary to prevent | |
1047 | * quadratic behavior of shrink_dcache_parent(), but is also | |
1048 | * expected to be beneficial in reducing dentry cache | |
1049 | * fragmentation. | |
1050 | */ | |
1051 | dentry = parent; | |
b2b80195 AV |
1052 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) { |
1053 | parent = lock_parent(dentry); | |
1054 | if (dentry->d_lockref.count != 1) { | |
1055 | dentry->d_lockref.count--; | |
1056 | spin_unlock(&dentry->d_lock); | |
1057 | if (parent) | |
1058 | spin_unlock(&parent->d_lock); | |
1059 | break; | |
1060 | } | |
1061 | inode = dentry->d_inode; /* can't be NULL */ | |
1062 | if (unlikely(!spin_trylock(&inode->i_lock))) { | |
1063 | spin_unlock(&dentry->d_lock); | |
1064 | if (parent) | |
1065 | spin_unlock(&parent->d_lock); | |
1066 | cpu_relax(); | |
1067 | continue; | |
1068 | } | |
1069 | __dentry_kill(dentry); | |
1070 | dentry = parent; | |
1071 | } | |
da3bbdd4 | 1072 | } |
3049cfe2 CH |
1073 | } |
1074 | ||
3f97b163 VD |
1075 | static enum lru_status dentry_lru_isolate(struct list_head *item, |
1076 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
f6041567 DC |
1077 | { |
1078 | struct list_head *freeable = arg; | |
1079 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
1080 | ||
1081 | ||
1082 | /* | |
1083 | * we are inverting the lru lock/dentry->d_lock here, | |
1084 | * so use a trylock. If we fail to get the lock, just skip | |
1085 | * it | |
1086 | */ | |
1087 | if (!spin_trylock(&dentry->d_lock)) | |
1088 | return LRU_SKIP; | |
1089 | ||
1090 | /* | |
1091 | * Referenced dentries are still in use. If they have active | |
1092 | * counts, just remove them from the LRU. Otherwise give them | |
1093 | * another pass through the LRU. | |
1094 | */ | |
1095 | if (dentry->d_lockref.count) { | |
3f97b163 | 1096 | d_lru_isolate(lru, dentry); |
f6041567 DC |
1097 | spin_unlock(&dentry->d_lock); |
1098 | return LRU_REMOVED; | |
1099 | } | |
1100 | ||
1101 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
1102 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
1103 | spin_unlock(&dentry->d_lock); | |
1104 | ||
1105 | /* | |
1106 | * The list move itself will be made by the common LRU code. At | |
1107 | * this point, we've dropped the dentry->d_lock but keep the | |
1108 | * lru lock. This is safe to do, since every list movement is | |
1109 | * protected by the lru lock even if both locks are held. | |
1110 | * | |
1111 | * This is guaranteed by the fact that all LRU management | |
1112 | * functions are intermediated by the LRU API calls like | |
1113 | * list_lru_add and list_lru_del. List movement in this file | |
1114 | * only ever occur through this functions or through callbacks | |
1115 | * like this one, that are called from the LRU API. | |
1116 | * | |
1117 | * The only exceptions to this are functions like | |
1118 | * shrink_dentry_list, and code that first checks for the | |
1119 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
1120 | * operating only with stack provided lists after they are | |
1121 | * properly isolated from the main list. It is thus, always a | |
1122 | * local access. | |
1123 | */ | |
1124 | return LRU_ROTATE; | |
1125 | } | |
1126 | ||
3f97b163 | 1127 | d_lru_shrink_move(lru, dentry, freeable); |
f6041567 DC |
1128 | spin_unlock(&dentry->d_lock); |
1129 | ||
1130 | return LRU_REMOVED; | |
1131 | } | |
1132 | ||
3049cfe2 | 1133 | /** |
b48f03b3 DC |
1134 | * prune_dcache_sb - shrink the dcache |
1135 | * @sb: superblock | |
503c358c | 1136 | * @sc: shrink control, passed to list_lru_shrink_walk() |
b48f03b3 | 1137 | * |
503c358c VD |
1138 | * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This |
1139 | * is done when we need more memory and called from the superblock shrinker | |
b48f03b3 | 1140 | * function. |
3049cfe2 | 1141 | * |
b48f03b3 DC |
1142 | * This function may fail to free any resources if all the dentries are in |
1143 | * use. | |
3049cfe2 | 1144 | */ |
503c358c | 1145 | long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc) |
3049cfe2 | 1146 | { |
f6041567 DC |
1147 | LIST_HEAD(dispose); |
1148 | long freed; | |
3049cfe2 | 1149 | |
503c358c VD |
1150 | freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc, |
1151 | dentry_lru_isolate, &dispose); | |
f6041567 | 1152 | shrink_dentry_list(&dispose); |
0a234c6d | 1153 | return freed; |
da3bbdd4 | 1154 | } |
23044507 | 1155 | |
4e717f5c | 1156 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
3f97b163 | 1157 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
dd1f6b2e | 1158 | { |
4e717f5c GC |
1159 | struct list_head *freeable = arg; |
1160 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 1161 | |
4e717f5c GC |
1162 | /* |
1163 | * we are inverting the lru lock/dentry->d_lock here, | |
1164 | * so use a trylock. If we fail to get the lock, just skip | |
1165 | * it | |
1166 | */ | |
1167 | if (!spin_trylock(&dentry->d_lock)) | |
1168 | return LRU_SKIP; | |
1169 | ||
3f97b163 | 1170 | d_lru_shrink_move(lru, dentry, freeable); |
4e717f5c | 1171 | spin_unlock(&dentry->d_lock); |
ec33679d | 1172 | |
4e717f5c | 1173 | return LRU_REMOVED; |
da3bbdd4 KM |
1174 | } |
1175 | ||
4e717f5c | 1176 | |
1da177e4 LT |
1177 | /** |
1178 | * shrink_dcache_sb - shrink dcache for a superblock | |
1179 | * @sb: superblock | |
1180 | * | |
3049cfe2 CH |
1181 | * Shrink the dcache for the specified super block. This is used to free |
1182 | * the dcache before unmounting a file system. | |
1da177e4 | 1183 | */ |
3049cfe2 | 1184 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1185 | { |
4e717f5c GC |
1186 | do { |
1187 | LIST_HEAD(dispose); | |
1188 | ||
b30b0fda | 1189 | list_lru_walk(&sb->s_dentry_lru, |
b17c070f | 1190 | dentry_lru_isolate_shrink, &dispose, 1024); |
4e717f5c | 1191 | shrink_dentry_list(&dispose); |
b17c070f ST |
1192 | cond_resched(); |
1193 | } while (list_lru_count(&sb->s_dentry_lru) > 0); | |
1da177e4 | 1194 | } |
ec4f8605 | 1195 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1196 | |
db14fc3a MS |
1197 | /** |
1198 | * enum d_walk_ret - action to talke during tree walk | |
1199 | * @D_WALK_CONTINUE: contrinue walk | |
1200 | * @D_WALK_QUIT: quit walk | |
1201 | * @D_WALK_NORETRY: quit when retry is needed | |
1202 | * @D_WALK_SKIP: skip this dentry and its children | |
1203 | */ | |
1204 | enum d_walk_ret { | |
1205 | D_WALK_CONTINUE, | |
1206 | D_WALK_QUIT, | |
1207 | D_WALK_NORETRY, | |
1208 | D_WALK_SKIP, | |
1209 | }; | |
c826cb7d | 1210 | |
1da177e4 | 1211 | /** |
db14fc3a MS |
1212 | * d_walk - walk the dentry tree |
1213 | * @parent: start of walk | |
1214 | * @data: data passed to @enter() and @finish() | |
1215 | * @enter: callback when first entering the dentry | |
1216 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1217 | * |
db14fc3a | 1218 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1219 | */ |
db14fc3a MS |
1220 | static void d_walk(struct dentry *parent, void *data, |
1221 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1222 | void (*finish)(void *)) | |
1da177e4 | 1223 | { |
949854d0 | 1224 | struct dentry *this_parent; |
1da177e4 | 1225 | struct list_head *next; |
48f5ec21 | 1226 | unsigned seq = 0; |
db14fc3a MS |
1227 | enum d_walk_ret ret; |
1228 | bool retry = true; | |
949854d0 | 1229 | |
58db63d0 | 1230 | again: |
48f5ec21 | 1231 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1232 | this_parent = parent; |
2fd6b7f5 | 1233 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1234 | |
1235 | ret = enter(data, this_parent); | |
1236 | switch (ret) { | |
1237 | case D_WALK_CONTINUE: | |
1238 | break; | |
1239 | case D_WALK_QUIT: | |
1240 | case D_WALK_SKIP: | |
1241 | goto out_unlock; | |
1242 | case D_WALK_NORETRY: | |
1243 | retry = false; | |
1244 | break; | |
1245 | } | |
1da177e4 LT |
1246 | repeat: |
1247 | next = this_parent->d_subdirs.next; | |
1248 | resume: | |
1249 | while (next != &this_parent->d_subdirs) { | |
1250 | struct list_head *tmp = next; | |
946e51f2 | 1251 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1da177e4 | 1252 | next = tmp->next; |
2fd6b7f5 | 1253 | |
ba65dc5e AV |
1254 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR)) |
1255 | continue; | |
1256 | ||
2fd6b7f5 | 1257 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
db14fc3a MS |
1258 | |
1259 | ret = enter(data, dentry); | |
1260 | switch (ret) { | |
1261 | case D_WALK_CONTINUE: | |
1262 | break; | |
1263 | case D_WALK_QUIT: | |
2fd6b7f5 | 1264 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1265 | goto out_unlock; |
1266 | case D_WALK_NORETRY: | |
1267 | retry = false; | |
1268 | break; | |
1269 | case D_WALK_SKIP: | |
1270 | spin_unlock(&dentry->d_lock); | |
1271 | continue; | |
2fd6b7f5 | 1272 | } |
db14fc3a | 1273 | |
1da177e4 | 1274 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1275 | spin_unlock(&this_parent->d_lock); |
1276 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1277 | this_parent = dentry; |
2fd6b7f5 | 1278 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1279 | goto repeat; |
1280 | } | |
2fd6b7f5 | 1281 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1282 | } |
1283 | /* | |
1284 | * All done at this level ... ascend and resume the search. | |
1285 | */ | |
ca5358ef AV |
1286 | rcu_read_lock(); |
1287 | ascend: | |
1da177e4 | 1288 | if (this_parent != parent) { |
c826cb7d | 1289 | struct dentry *child = this_parent; |
31dec132 AV |
1290 | this_parent = child->d_parent; |
1291 | ||
31dec132 AV |
1292 | spin_unlock(&child->d_lock); |
1293 | spin_lock(&this_parent->d_lock); | |
1294 | ||
ca5358ef AV |
1295 | /* might go back up the wrong parent if we have had a rename. */ |
1296 | if (need_seqretry(&rename_lock, seq)) | |
949854d0 | 1297 | goto rename_retry; |
2159184e AV |
1298 | /* go into the first sibling still alive */ |
1299 | do { | |
1300 | next = child->d_child.next; | |
ca5358ef AV |
1301 | if (next == &this_parent->d_subdirs) |
1302 | goto ascend; | |
1303 | child = list_entry(next, struct dentry, d_child); | |
2159184e | 1304 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); |
31dec132 | 1305 | rcu_read_unlock(); |
1da177e4 LT |
1306 | goto resume; |
1307 | } | |
ca5358ef | 1308 | if (need_seqretry(&rename_lock, seq)) |
949854d0 | 1309 | goto rename_retry; |
ca5358ef | 1310 | rcu_read_unlock(); |
db14fc3a MS |
1311 | if (finish) |
1312 | finish(data); | |
1313 | ||
1314 | out_unlock: | |
1315 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1316 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1317 | return; |
58db63d0 NP |
1318 | |
1319 | rename_retry: | |
ca5358ef AV |
1320 | spin_unlock(&this_parent->d_lock); |
1321 | rcu_read_unlock(); | |
1322 | BUG_ON(seq & 1); | |
db14fc3a MS |
1323 | if (!retry) |
1324 | return; | |
48f5ec21 | 1325 | seq = 1; |
58db63d0 | 1326 | goto again; |
1da177e4 | 1327 | } |
db14fc3a | 1328 | |
01619491 IK |
1329 | struct check_mount { |
1330 | struct vfsmount *mnt; | |
1331 | unsigned int mounted; | |
1332 | }; | |
1333 | ||
1334 | static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry) | |
1335 | { | |
1336 | struct check_mount *info = data; | |
1337 | struct path path = { .mnt = info->mnt, .dentry = dentry }; | |
1338 | ||
1339 | if (likely(!d_mountpoint(dentry))) | |
1340 | return D_WALK_CONTINUE; | |
1341 | if (__path_is_mountpoint(&path)) { | |
1342 | info->mounted = 1; | |
1343 | return D_WALK_QUIT; | |
1344 | } | |
1345 | return D_WALK_CONTINUE; | |
1346 | } | |
1347 | ||
1348 | /** | |
1349 | * path_has_submounts - check for mounts over a dentry in the | |
1350 | * current namespace. | |
1351 | * @parent: path to check. | |
1352 | * | |
1353 | * Return true if the parent or its subdirectories contain | |
1354 | * a mount point in the current namespace. | |
1355 | */ | |
1356 | int path_has_submounts(const struct path *parent) | |
1357 | { | |
1358 | struct check_mount data = { .mnt = parent->mnt, .mounted = 0 }; | |
1359 | ||
1360 | read_seqlock_excl(&mount_lock); | |
1361 | d_walk(parent->dentry, &data, path_check_mount, NULL); | |
1362 | read_sequnlock_excl(&mount_lock); | |
1363 | ||
1364 | return data.mounted; | |
1365 | } | |
1366 | EXPORT_SYMBOL(path_has_submounts); | |
1367 | ||
eed81007 MS |
1368 | /* |
1369 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1370 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1371 | * subtree can become unreachable). | |
1372 | * | |
1ffe46d1 | 1373 | * Only one of d_invalidate() and d_set_mounted() must succeed. For |
eed81007 MS |
1374 | * this reason take rename_lock and d_lock on dentry and ancestors. |
1375 | */ | |
1376 | int d_set_mounted(struct dentry *dentry) | |
1377 | { | |
1378 | struct dentry *p; | |
1379 | int ret = -ENOENT; | |
1380 | write_seqlock(&rename_lock); | |
1381 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1ffe46d1 | 1382 | /* Need exclusion wrt. d_invalidate() */ |
eed81007 MS |
1383 | spin_lock(&p->d_lock); |
1384 | if (unlikely(d_unhashed(p))) { | |
1385 | spin_unlock(&p->d_lock); | |
1386 | goto out; | |
1387 | } | |
1388 | spin_unlock(&p->d_lock); | |
1389 | } | |
1390 | spin_lock(&dentry->d_lock); | |
1391 | if (!d_unlinked(dentry)) { | |
3895dbf8 EB |
1392 | ret = -EBUSY; |
1393 | if (!d_mountpoint(dentry)) { | |
1394 | dentry->d_flags |= DCACHE_MOUNTED; | |
1395 | ret = 0; | |
1396 | } | |
eed81007 MS |
1397 | } |
1398 | spin_unlock(&dentry->d_lock); | |
1399 | out: | |
1400 | write_sequnlock(&rename_lock); | |
1401 | return ret; | |
1402 | } | |
1403 | ||
1da177e4 | 1404 | /* |
fd517909 | 1405 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1406 | * and move any unused dentries to the end of the unused |
1407 | * list for prune_dcache(). We descend to the next level | |
1408 | * whenever the d_subdirs list is non-empty and continue | |
1409 | * searching. | |
1410 | * | |
1411 | * It returns zero iff there are no unused children, | |
1412 | * otherwise it returns the number of children moved to | |
1413 | * the end of the unused list. This may not be the total | |
1414 | * number of unused children, because select_parent can | |
1415 | * drop the lock and return early due to latency | |
1416 | * constraints. | |
1417 | */ | |
1da177e4 | 1418 | |
db14fc3a MS |
1419 | struct select_data { |
1420 | struct dentry *start; | |
1421 | struct list_head dispose; | |
1422 | int found; | |
1423 | }; | |
23044507 | 1424 | |
db14fc3a MS |
1425 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1426 | { | |
1427 | struct select_data *data = _data; | |
1428 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1429 | |
db14fc3a MS |
1430 | if (data->start == dentry) |
1431 | goto out; | |
2fd6b7f5 | 1432 | |
fe91522a | 1433 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1434 | data->found++; |
fe91522a AV |
1435 | } else { |
1436 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1437 | d_lru_del(dentry); | |
1438 | if (!dentry->d_lockref.count) { | |
1439 | d_shrink_add(dentry, &data->dispose); | |
1440 | data->found++; | |
1441 | } | |
1da177e4 | 1442 | } |
db14fc3a MS |
1443 | /* |
1444 | * We can return to the caller if we have found some (this | |
1445 | * ensures forward progress). We'll be coming back to find | |
1446 | * the rest. | |
1447 | */ | |
fe91522a AV |
1448 | if (!list_empty(&data->dispose)) |
1449 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1450 | out: |
db14fc3a | 1451 | return ret; |
1da177e4 LT |
1452 | } |
1453 | ||
1454 | /** | |
1455 | * shrink_dcache_parent - prune dcache | |
1456 | * @parent: parent of entries to prune | |
1457 | * | |
1458 | * Prune the dcache to remove unused children of the parent dentry. | |
1459 | */ | |
db14fc3a | 1460 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1461 | { |
db14fc3a MS |
1462 | for (;;) { |
1463 | struct select_data data; | |
1da177e4 | 1464 | |
db14fc3a MS |
1465 | INIT_LIST_HEAD(&data.dispose); |
1466 | data.start = parent; | |
1467 | data.found = 0; | |
1468 | ||
1469 | d_walk(parent, &data, select_collect, NULL); | |
1470 | if (!data.found) | |
1471 | break; | |
1472 | ||
1473 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1474 | cond_resched(); |
1475 | } | |
1da177e4 | 1476 | } |
ec4f8605 | 1477 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1478 | |
9c8c10e2 | 1479 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1480 | { |
9c8c10e2 AV |
1481 | /* it has busy descendents; complain about those instead */ |
1482 | if (!list_empty(&dentry->d_subdirs)) | |
1483 | return D_WALK_CONTINUE; | |
42c32608 | 1484 | |
9c8c10e2 AV |
1485 | /* root with refcount 1 is fine */ |
1486 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1487 | return D_WALK_CONTINUE; | |
1488 | ||
1489 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1490 | " still in use (%d) [unmount of %s %s]\n", | |
42c32608 AV |
1491 | dentry, |
1492 | dentry->d_inode ? | |
1493 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1494 | dentry, |
42c32608 AV |
1495 | dentry->d_lockref.count, |
1496 | dentry->d_sb->s_type->name, | |
1497 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1498 | WARN_ON(1); |
1499 | return D_WALK_CONTINUE; | |
1500 | } | |
1501 | ||
1502 | static void do_one_tree(struct dentry *dentry) | |
1503 | { | |
1504 | shrink_dcache_parent(dentry); | |
1505 | d_walk(dentry, dentry, umount_check, NULL); | |
1506 | d_drop(dentry); | |
1507 | dput(dentry); | |
42c32608 AV |
1508 | } |
1509 | ||
1510 | /* | |
1511 | * destroy the dentries attached to a superblock on unmounting | |
1512 | */ | |
1513 | void shrink_dcache_for_umount(struct super_block *sb) | |
1514 | { | |
1515 | struct dentry *dentry; | |
1516 | ||
9c8c10e2 | 1517 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1518 | |
1519 | dentry = sb->s_root; | |
1520 | sb->s_root = NULL; | |
9c8c10e2 | 1521 | do_one_tree(dentry); |
42c32608 AV |
1522 | |
1523 | while (!hlist_bl_empty(&sb->s_anon)) { | |
9c8c10e2 AV |
1524 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash)); |
1525 | do_one_tree(dentry); | |
42c32608 AV |
1526 | } |
1527 | } | |
1528 | ||
8ed936b5 EB |
1529 | struct detach_data { |
1530 | struct select_data select; | |
1531 | struct dentry *mountpoint; | |
1532 | }; | |
1533 | static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry) | |
848ac114 | 1534 | { |
8ed936b5 | 1535 | struct detach_data *data = _data; |
848ac114 MS |
1536 | |
1537 | if (d_mountpoint(dentry)) { | |
8ed936b5 EB |
1538 | __dget_dlock(dentry); |
1539 | data->mountpoint = dentry; | |
848ac114 MS |
1540 | return D_WALK_QUIT; |
1541 | } | |
1542 | ||
8ed936b5 | 1543 | return select_collect(&data->select, dentry); |
848ac114 MS |
1544 | } |
1545 | ||
1546 | static void check_and_drop(void *_data) | |
1547 | { | |
8ed936b5 | 1548 | struct detach_data *data = _data; |
848ac114 | 1549 | |
81be24d2 | 1550 | if (!data->mountpoint && list_empty(&data->select.dispose)) |
8ed936b5 | 1551 | __d_drop(data->select.start); |
848ac114 MS |
1552 | } |
1553 | ||
1554 | /** | |
1ffe46d1 EB |
1555 | * d_invalidate - detach submounts, prune dcache, and drop |
1556 | * @dentry: dentry to invalidate (aka detach, prune and drop) | |
1557 | * | |
1ffe46d1 | 1558 | * no dcache lock. |
848ac114 | 1559 | * |
8ed936b5 EB |
1560 | * The final d_drop is done as an atomic operation relative to |
1561 | * rename_lock ensuring there are no races with d_set_mounted. This | |
1562 | * ensures there are no unhashed dentries on the path to a mountpoint. | |
848ac114 | 1563 | */ |
5542aa2f | 1564 | void d_invalidate(struct dentry *dentry) |
848ac114 | 1565 | { |
1ffe46d1 EB |
1566 | /* |
1567 | * If it's already been dropped, return OK. | |
1568 | */ | |
1569 | spin_lock(&dentry->d_lock); | |
1570 | if (d_unhashed(dentry)) { | |
1571 | spin_unlock(&dentry->d_lock); | |
5542aa2f | 1572 | return; |
1ffe46d1 EB |
1573 | } |
1574 | spin_unlock(&dentry->d_lock); | |
1575 | ||
848ac114 MS |
1576 | /* Negative dentries can be dropped without further checks */ |
1577 | if (!dentry->d_inode) { | |
1578 | d_drop(dentry); | |
5542aa2f | 1579 | return; |
848ac114 MS |
1580 | } |
1581 | ||
1582 | for (;;) { | |
8ed936b5 | 1583 | struct detach_data data; |
848ac114 | 1584 | |
8ed936b5 EB |
1585 | data.mountpoint = NULL; |
1586 | INIT_LIST_HEAD(&data.select.dispose); | |
1587 | data.select.start = dentry; | |
1588 | data.select.found = 0; | |
1589 | ||
1590 | d_walk(dentry, &data, detach_and_collect, check_and_drop); | |
848ac114 | 1591 | |
81be24d2 | 1592 | if (!list_empty(&data.select.dispose)) |
8ed936b5 | 1593 | shrink_dentry_list(&data.select.dispose); |
81be24d2 AV |
1594 | else if (!data.mountpoint) |
1595 | return; | |
848ac114 | 1596 | |
8ed936b5 EB |
1597 | if (data.mountpoint) { |
1598 | detach_mounts(data.mountpoint); | |
1599 | dput(data.mountpoint); | |
1600 | } | |
848ac114 MS |
1601 | cond_resched(); |
1602 | } | |
848ac114 | 1603 | } |
1ffe46d1 | 1604 | EXPORT_SYMBOL(d_invalidate); |
848ac114 | 1605 | |
1da177e4 | 1606 | /** |
a4464dbc AV |
1607 | * __d_alloc - allocate a dcache entry |
1608 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1609 | * @name: qstr of the name |
1610 | * | |
1611 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1612 | * available. On a success the dentry is returned. The name passed in is | |
1613 | * copied and the copy passed in may be reused after this call. | |
1614 | */ | |
1615 | ||
a4464dbc | 1616 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 | 1617 | { |
6d794237 | 1618 | struct external_name *ext = NULL; |
1da177e4 LT |
1619 | struct dentry *dentry; |
1620 | char *dname; | |
285b102d | 1621 | int err; |
1da177e4 | 1622 | |
e12ba74d | 1623 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1624 | if (!dentry) |
1625 | return NULL; | |
1626 | ||
6326c71f LT |
1627 | /* |
1628 | * We guarantee that the inline name is always NUL-terminated. | |
1629 | * This way the memcpy() done by the name switching in rename | |
1630 | * will still always have a NUL at the end, even if we might | |
1631 | * be overwriting an internal NUL character | |
1632 | */ | |
1633 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
798434bd | 1634 | if (unlikely(!name)) { |
cdf01226 | 1635 | name = &slash_name; |
798434bd AV |
1636 | dname = dentry->d_iname; |
1637 | } else if (name->len > DNAME_INLINE_LEN-1) { | |
8d85b484 | 1638 | size_t size = offsetof(struct external_name, name[1]); |
6d794237 RG |
1639 | ext = kmalloc(size + name->len, GFP_KERNEL_ACCOUNT); |
1640 | if (!ext) { | |
1da177e4 LT |
1641 | kmem_cache_free(dentry_cache, dentry); |
1642 | return NULL; | |
1643 | } | |
6d794237 RG |
1644 | atomic_set(&ext->u.count, 1); |
1645 | dname = ext->name; | |
df4c0e36 AR |
1646 | if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS)) |
1647 | kasan_unpoison_shadow(dname, | |
1648 | round_up(name->len + 1, sizeof(unsigned long))); | |
1da177e4 LT |
1649 | } else { |
1650 | dname = dentry->d_iname; | |
1651 | } | |
1da177e4 LT |
1652 | |
1653 | dentry->d_name.len = name->len; | |
1654 | dentry->d_name.hash = name->hash; | |
1655 | memcpy(dname, name->name, name->len); | |
1656 | dname[name->len] = 0; | |
1657 | ||
6326c71f LT |
1658 | /* Make sure we always see the terminating NUL character */ |
1659 | smp_wmb(); | |
1660 | dentry->d_name.name = dname; | |
1661 | ||
98474236 | 1662 | dentry->d_lockref.count = 1; |
dea3667b | 1663 | dentry->d_flags = 0; |
1da177e4 | 1664 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1665 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1666 | dentry->d_inode = NULL; |
a4464dbc AV |
1667 | dentry->d_parent = dentry; |
1668 | dentry->d_sb = sb; | |
1da177e4 LT |
1669 | dentry->d_op = NULL; |
1670 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1671 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1672 | INIT_LIST_HEAD(&dentry->d_lru); |
1673 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
946e51f2 AV |
1674 | INIT_HLIST_NODE(&dentry->d_u.d_alias); |
1675 | INIT_LIST_HEAD(&dentry->d_child); | |
a4464dbc | 1676 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1677 | |
285b102d MS |
1678 | if (dentry->d_op && dentry->d_op->d_init) { |
1679 | err = dentry->d_op->d_init(dentry); | |
1680 | if (err) { | |
1681 | if (dname_external(dentry)) | |
1682 | kfree(external_name(dentry)); | |
1683 | kmem_cache_free(dentry_cache, dentry); | |
1684 | return NULL; | |
1685 | } | |
1686 | } | |
1687 | ||
6d794237 RG |
1688 | if (unlikely(ext)) { |
1689 | pg_data_t *pgdat = page_pgdat(virt_to_page(ext)); | |
1690 | mod_node_page_state(pgdat, NR_INDIRECTLY_RECLAIMABLE_BYTES, | |
1691 | ksize(ext)); | |
1692 | } | |
1693 | ||
3e880fb5 | 1694 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1695 | |
1da177e4 LT |
1696 | return dentry; |
1697 | } | |
a4464dbc AV |
1698 | |
1699 | /** | |
1700 | * d_alloc - allocate a dcache entry | |
1701 | * @parent: parent of entry to allocate | |
1702 | * @name: qstr of the name | |
1703 | * | |
1704 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1705 | * available. On a success the dentry is returned. The name passed in is | |
1706 | * copied and the copy passed in may be reused after this call. | |
1707 | */ | |
1708 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1709 | { | |
1710 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1711 | if (!dentry) | |
1712 | return NULL; | |
3d56c25e | 1713 | dentry->d_flags |= DCACHE_RCUACCESS; |
a4464dbc AV |
1714 | spin_lock(&parent->d_lock); |
1715 | /* | |
1716 | * don't need child lock because it is not subject | |
1717 | * to concurrency here | |
1718 | */ | |
1719 | __dget_dlock(parent); | |
1720 | dentry->d_parent = parent; | |
946e51f2 | 1721 | list_add(&dentry->d_child, &parent->d_subdirs); |
a4464dbc AV |
1722 | spin_unlock(&parent->d_lock); |
1723 | ||
1724 | return dentry; | |
1725 | } | |
ec4f8605 | 1726 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1727 | |
ba65dc5e AV |
1728 | struct dentry *d_alloc_cursor(struct dentry * parent) |
1729 | { | |
1730 | struct dentry *dentry = __d_alloc(parent->d_sb, NULL); | |
1731 | if (dentry) { | |
1732 | dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR; | |
1733 | dentry->d_parent = dget(parent); | |
1734 | } | |
1735 | return dentry; | |
1736 | } | |
1737 | ||
e1a24bb0 BF |
1738 | /** |
1739 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1740 | * @sb: the superblock | |
1741 | * @name: qstr of the name | |
1742 | * | |
1743 | * For a filesystem that just pins its dentries in memory and never | |
1744 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1745 | */ | |
4b936885 NP |
1746 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1747 | { | |
e1a24bb0 | 1748 | return __d_alloc(sb, name); |
4b936885 NP |
1749 | } |
1750 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1751 | ||
1da177e4 LT |
1752 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1753 | { | |
1754 | struct qstr q; | |
1755 | ||
1756 | q.name = name; | |
8387ff25 | 1757 | q.hash_len = hashlen_string(parent, name); |
1da177e4 LT |
1758 | return d_alloc(parent, &q); |
1759 | } | |
ef26ca97 | 1760 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1761 | |
fb045adb NP |
1762 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1763 | { | |
6f7f7caa LT |
1764 | WARN_ON_ONCE(dentry->d_op); |
1765 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1766 | DCACHE_OP_COMPARE | |
1767 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1768 | DCACHE_OP_WEAK_REVALIDATE | |
4bacc9c9 | 1769 | DCACHE_OP_DELETE | |
d101a125 | 1770 | DCACHE_OP_REAL)); |
fb045adb NP |
1771 | dentry->d_op = op; |
1772 | if (!op) | |
1773 | return; | |
1774 | if (op->d_hash) | |
1775 | dentry->d_flags |= DCACHE_OP_HASH; | |
1776 | if (op->d_compare) | |
1777 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1778 | if (op->d_revalidate) | |
1779 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1780 | if (op->d_weak_revalidate) |
1781 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1782 | if (op->d_delete) |
1783 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1784 | if (op->d_prune) |
1785 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
d101a125 MS |
1786 | if (op->d_real) |
1787 | dentry->d_flags |= DCACHE_OP_REAL; | |
fb045adb NP |
1788 | |
1789 | } | |
1790 | EXPORT_SYMBOL(d_set_d_op); | |
1791 | ||
df1a085a DH |
1792 | |
1793 | /* | |
1794 | * d_set_fallthru - Mark a dentry as falling through to a lower layer | |
1795 | * @dentry - The dentry to mark | |
1796 | * | |
1797 | * Mark a dentry as falling through to the lower layer (as set with | |
1798 | * d_pin_lower()). This flag may be recorded on the medium. | |
1799 | */ | |
1800 | void d_set_fallthru(struct dentry *dentry) | |
1801 | { | |
1802 | spin_lock(&dentry->d_lock); | |
1803 | dentry->d_flags |= DCACHE_FALLTHRU; | |
1804 | spin_unlock(&dentry->d_lock); | |
1805 | } | |
1806 | EXPORT_SYMBOL(d_set_fallthru); | |
1807 | ||
b18825a7 DH |
1808 | static unsigned d_flags_for_inode(struct inode *inode) |
1809 | { | |
44bdb5e5 | 1810 | unsigned add_flags = DCACHE_REGULAR_TYPE; |
b18825a7 DH |
1811 | |
1812 | if (!inode) | |
1813 | return DCACHE_MISS_TYPE; | |
1814 | ||
1815 | if (S_ISDIR(inode->i_mode)) { | |
1816 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1817 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1818 | if (unlikely(!inode->i_op->lookup)) | |
1819 | add_flags = DCACHE_AUTODIR_TYPE; | |
1820 | else | |
1821 | inode->i_opflags |= IOP_LOOKUP; | |
1822 | } | |
44bdb5e5 DH |
1823 | goto type_determined; |
1824 | } | |
1825 | ||
1826 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
6b255391 | 1827 | if (unlikely(inode->i_op->get_link)) { |
b18825a7 | 1828 | add_flags = DCACHE_SYMLINK_TYPE; |
44bdb5e5 DH |
1829 | goto type_determined; |
1830 | } | |
1831 | inode->i_opflags |= IOP_NOFOLLOW; | |
b18825a7 DH |
1832 | } |
1833 | ||
44bdb5e5 DH |
1834 | if (unlikely(!S_ISREG(inode->i_mode))) |
1835 | add_flags = DCACHE_SPECIAL_TYPE; | |
1836 | ||
1837 | type_determined: | |
b18825a7 DH |
1838 | if (unlikely(IS_AUTOMOUNT(inode))) |
1839 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1840 | return add_flags; | |
1841 | } | |
1842 | ||
360da900 OH |
1843 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1844 | { | |
b18825a7 | 1845 | unsigned add_flags = d_flags_for_inode(inode); |
85c7f810 | 1846 | WARN_ON(d_in_lookup(dentry)); |
b18825a7 | 1847 | |
b23fb0a6 | 1848 | spin_lock(&dentry->d_lock); |
de689f5e | 1849 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); |
a528aca7 | 1850 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 1851 | __d_set_inode_and_type(dentry, inode, add_flags); |
a528aca7 | 1852 | raw_write_seqcount_end(&dentry->d_seq); |
affda484 | 1853 | fsnotify_update_flags(dentry); |
b23fb0a6 | 1854 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1855 | } |
1856 | ||
1da177e4 LT |
1857 | /** |
1858 | * d_instantiate - fill in inode information for a dentry | |
1859 | * @entry: dentry to complete | |
1860 | * @inode: inode to attach to this dentry | |
1861 | * | |
1862 | * Fill in inode information in the entry. | |
1863 | * | |
1864 | * This turns negative dentries into productive full members | |
1865 | * of society. | |
1866 | * | |
1867 | * NOTE! This assumes that the inode count has been incremented | |
1868 | * (or otherwise set) by the caller to indicate that it is now | |
1869 | * in use by the dcache. | |
1870 | */ | |
1871 | ||
1872 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1873 | { | |
946e51f2 | 1874 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
de689f5e | 1875 | if (inode) { |
b9680917 | 1876 | security_d_instantiate(entry, inode); |
873feea0 | 1877 | spin_lock(&inode->i_lock); |
de689f5e | 1878 | __d_instantiate(entry, inode); |
873feea0 | 1879 | spin_unlock(&inode->i_lock); |
de689f5e | 1880 | } |
1da177e4 | 1881 | } |
ec4f8605 | 1882 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 | 1883 | |
f440ea85 AV |
1884 | /* |
1885 | * This should be equivalent to d_instantiate() + unlock_new_inode(), | |
1886 | * with lockdep-related part of unlock_new_inode() done before | |
1887 | * anything else. Use that instead of open-coding d_instantiate()/ | |
1888 | * unlock_new_inode() combinations. | |
1889 | */ | |
1890 | void d_instantiate_new(struct dentry *entry, struct inode *inode) | |
1891 | { | |
1892 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); | |
1893 | BUG_ON(!inode); | |
1894 | lockdep_annotate_inode_mutex_key(inode); | |
1895 | security_d_instantiate(entry, inode); | |
1896 | spin_lock(&inode->i_lock); | |
1897 | __d_instantiate(entry, inode); | |
1898 | WARN_ON(!(inode->i_state & I_NEW)); | |
1899 | inode->i_state &= ~I_NEW; | |
1900 | smp_mb(); | |
1901 | wake_up_bit(&inode->i_state, __I_NEW); | |
1902 | spin_unlock(&inode->i_lock); | |
1903 | } | |
1904 | EXPORT_SYMBOL(d_instantiate_new); | |
1905 | ||
b70a80e7 MS |
1906 | /** |
1907 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1908 | * @entry: dentry to complete | |
1909 | * @inode: inode to attach to this dentry | |
1910 | * | |
1911 | * Fill in inode information in the entry. If a directory alias is found, then | |
1912 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1913 | * guarantees that a directory inode may never have more than one alias. | |
1914 | */ | |
1915 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1916 | { | |
946e51f2 | 1917 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
b70a80e7 | 1918 | |
b9680917 | 1919 | security_d_instantiate(entry, inode); |
b70a80e7 MS |
1920 | spin_lock(&inode->i_lock); |
1921 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1922 | spin_unlock(&inode->i_lock); | |
1923 | iput(inode); | |
1924 | return -EBUSY; | |
1925 | } | |
1926 | __d_instantiate(entry, inode); | |
1927 | spin_unlock(&inode->i_lock); | |
b70a80e7 MS |
1928 | |
1929 | return 0; | |
1930 | } | |
1931 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1932 | ||
adc0e91a AV |
1933 | struct dentry *d_make_root(struct inode *root_inode) |
1934 | { | |
1935 | struct dentry *res = NULL; | |
1936 | ||
1937 | if (root_inode) { | |
798434bd | 1938 | res = __d_alloc(root_inode->i_sb, NULL); |
abfc0ec6 AV |
1939 | if (res) { |
1940 | res->d_flags |= DCACHE_RCUACCESS; | |
adc0e91a | 1941 | d_instantiate(res, root_inode); |
abfc0ec6 | 1942 | } else { |
adc0e91a | 1943 | iput(root_inode); |
abfc0ec6 | 1944 | } |
adc0e91a AV |
1945 | } |
1946 | return res; | |
1947 | } | |
1948 | EXPORT_SYMBOL(d_make_root); | |
1949 | ||
d891eedb BF |
1950 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1951 | { | |
1952 | struct dentry *alias; | |
1953 | ||
b3d9b7a3 | 1954 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1955 | return NULL; |
946e51f2 | 1956 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); |
d891eedb BF |
1957 | __dget(alias); |
1958 | return alias; | |
1959 | } | |
1960 | ||
46f72b34 SW |
1961 | /** |
1962 | * d_find_any_alias - find any alias for a given inode | |
1963 | * @inode: inode to find an alias for | |
1964 | * | |
1965 | * If any aliases exist for the given inode, take and return a | |
1966 | * reference for one of them. If no aliases exist, return %NULL. | |
1967 | */ | |
1968 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1969 | { |
1970 | struct dentry *de; | |
1971 | ||
1972 | spin_lock(&inode->i_lock); | |
1973 | de = __d_find_any_alias(inode); | |
1974 | spin_unlock(&inode->i_lock); | |
1975 | return de; | |
1976 | } | |
46f72b34 | 1977 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1978 | |
49c7dd28 | 1979 | static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected) |
4ea3ada2 | 1980 | { |
9308a612 CH |
1981 | struct dentry *tmp; |
1982 | struct dentry *res; | |
b18825a7 | 1983 | unsigned add_flags; |
4ea3ada2 CH |
1984 | |
1985 | if (!inode) | |
44003728 | 1986 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1987 | if (IS_ERR(inode)) |
1988 | return ERR_CAST(inode); | |
1989 | ||
d891eedb | 1990 | res = d_find_any_alias(inode); |
9308a612 CH |
1991 | if (res) |
1992 | goto out_iput; | |
1993 | ||
798434bd | 1994 | tmp = __d_alloc(inode->i_sb, NULL); |
9308a612 CH |
1995 | if (!tmp) { |
1996 | res = ERR_PTR(-ENOMEM); | |
1997 | goto out_iput; | |
4ea3ada2 | 1998 | } |
b5c84bf6 | 1999 | |
b9680917 | 2000 | security_d_instantiate(tmp, inode); |
873feea0 | 2001 | spin_lock(&inode->i_lock); |
d891eedb | 2002 | res = __d_find_any_alias(inode); |
9308a612 | 2003 | if (res) { |
873feea0 | 2004 | spin_unlock(&inode->i_lock); |
9308a612 CH |
2005 | dput(tmp); |
2006 | goto out_iput; | |
2007 | } | |
2008 | ||
2009 | /* attach a disconnected dentry */ | |
1a0a397e BF |
2010 | add_flags = d_flags_for_inode(inode); |
2011 | ||
2012 | if (disconnected) | |
2013 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 2014 | |
9308a612 | 2015 | spin_lock(&tmp->d_lock); |
4bf46a27 | 2016 | __d_set_inode_and_type(tmp, inode, add_flags); |
946e51f2 | 2017 | hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry); |
1879fd6a | 2018 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 2019 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 2020 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 2021 | spin_unlock(&tmp->d_lock); |
873feea0 | 2022 | spin_unlock(&inode->i_lock); |
9308a612 | 2023 | |
9308a612 CH |
2024 | return tmp; |
2025 | ||
2026 | out_iput: | |
2027 | iput(inode); | |
2028 | return res; | |
4ea3ada2 | 2029 | } |
1a0a397e BF |
2030 | |
2031 | /** | |
2032 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
2033 | * @inode: inode to allocate the dentry for | |
2034 | * | |
2035 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
2036 | * similar open by handle operations. The returned dentry may be anonymous, | |
2037 | * or may have a full name (if the inode was already in the cache). | |
2038 | * | |
2039 | * When called on a directory inode, we must ensure that the inode only ever | |
2040 | * has one dentry. If a dentry is found, that is returned instead of | |
2041 | * allocating a new one. | |
2042 | * | |
2043 | * On successful return, the reference to the inode has been transferred | |
2044 | * to the dentry. In case of an error the reference on the inode is released. | |
2045 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
2046 | * be passed in and the error will be propagated to the return value, | |
2047 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
2048 | */ | |
2049 | struct dentry *d_obtain_alias(struct inode *inode) | |
2050 | { | |
2051 | return __d_obtain_alias(inode, 1); | |
2052 | } | |
adc48720 | 2053 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 2054 | |
1a0a397e BF |
2055 | /** |
2056 | * d_obtain_root - find or allocate a dentry for a given inode | |
2057 | * @inode: inode to allocate the dentry for | |
2058 | * | |
2059 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
2060 | * | |
2061 | * We must ensure that directory inodes only ever have one dentry. If a | |
2062 | * dentry is found, that is returned instead of allocating a new one. | |
2063 | * | |
2064 | * On successful return, the reference to the inode has been transferred | |
2065 | * to the dentry. In case of an error the reference on the inode is | |
2066 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
2067 | * error will be propagate to the return value, with a %NULL @inode | |
2068 | * replaced by ERR_PTR(-ESTALE). | |
2069 | */ | |
2070 | struct dentry *d_obtain_root(struct inode *inode) | |
2071 | { | |
2072 | return __d_obtain_alias(inode, 0); | |
2073 | } | |
2074 | EXPORT_SYMBOL(d_obtain_root); | |
2075 | ||
9403540c BN |
2076 | /** |
2077 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
2078 | * @inode: the inode case-insensitive lookup has found | |
2079 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
2080 | * @name: the case-exact name to be associated with the returned dentry | |
2081 | * | |
2082 | * This is to avoid filling the dcache with case-insensitive names to the | |
2083 | * same inode, only the actual correct case is stored in the dcache for | |
2084 | * case-insensitive filesystems. | |
2085 | * | |
2086 | * For a case-insensitive lookup match and if the the case-exact dentry | |
2087 | * already exists in in the dcache, use it and return it. | |
2088 | * | |
2089 | * If no entry exists with the exact case name, allocate new dentry with | |
2090 | * the exact case, and return the spliced entry. | |
2091 | */ | |
e45b590b | 2092 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
2093 | struct qstr *name) |
2094 | { | |
d9171b93 | 2095 | struct dentry *found, *res; |
9403540c | 2096 | |
b6520c81 CH |
2097 | /* |
2098 | * First check if a dentry matching the name already exists, | |
2099 | * if not go ahead and create it now. | |
2100 | */ | |
9403540c | 2101 | found = d_hash_and_lookup(dentry->d_parent, name); |
d9171b93 AV |
2102 | if (found) { |
2103 | iput(inode); | |
2104 | return found; | |
2105 | } | |
2106 | if (d_in_lookup(dentry)) { | |
2107 | found = d_alloc_parallel(dentry->d_parent, name, | |
2108 | dentry->d_wait); | |
2109 | if (IS_ERR(found) || !d_in_lookup(found)) { | |
2110 | iput(inode); | |
2111 | return found; | |
9403540c | 2112 | } |
d9171b93 AV |
2113 | } else { |
2114 | found = d_alloc(dentry->d_parent, name); | |
2115 | if (!found) { | |
2116 | iput(inode); | |
2117 | return ERR_PTR(-ENOMEM); | |
2118 | } | |
2119 | } | |
2120 | res = d_splice_alias(inode, found); | |
2121 | if (res) { | |
2122 | dput(found); | |
2123 | return res; | |
9403540c | 2124 | } |
4f522a24 | 2125 | return found; |
9403540c | 2126 | } |
ec4f8605 | 2127 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 2128 | |
12f8ad4b | 2129 | |
d4c91a8f AV |
2130 | static inline bool d_same_name(const struct dentry *dentry, |
2131 | const struct dentry *parent, | |
2132 | const struct qstr *name) | |
12f8ad4b | 2133 | { |
d4c91a8f AV |
2134 | if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) { |
2135 | if (dentry->d_name.len != name->len) | |
2136 | return false; | |
2137 | return dentry_cmp(dentry, name->name, name->len) == 0; | |
12f8ad4b | 2138 | } |
6fa67e70 | 2139 | return parent->d_op->d_compare(dentry, |
d4c91a8f AV |
2140 | dentry->d_name.len, dentry->d_name.name, |
2141 | name) == 0; | |
12f8ad4b LT |
2142 | } |
2143 | ||
31e6b01f NP |
2144 | /** |
2145 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2146 | * @parent: parent dentry | |
2147 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2148 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2149 | * Returns: dentry, or NULL |
2150 | * | |
2151 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2152 | * resolution (store-free path walking) design described in | |
2153 | * Documentation/filesystems/path-lookup.txt. | |
2154 | * | |
2155 | * This is not to be used outside core vfs. | |
2156 | * | |
2157 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2158 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2159 | * without taking d_lock and checking d_seq sequence count against @seq | |
2160 | * returned here. | |
2161 | * | |
15570086 | 2162 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2163 | * function. |
2164 | * | |
2165 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2166 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2167 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2168 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2169 | * |
2170 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2171 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2172 | */ |
8966be90 LT |
2173 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2174 | const struct qstr *name, | |
da53be12 | 2175 | unsigned *seqp) |
31e6b01f | 2176 | { |
26fe5750 | 2177 | u64 hashlen = name->hash_len; |
31e6b01f | 2178 | const unsigned char *str = name->name; |
8387ff25 | 2179 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); |
ceb5bdc2 | 2180 | struct hlist_bl_node *node; |
31e6b01f NP |
2181 | struct dentry *dentry; |
2182 | ||
2183 | /* | |
2184 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2185 | * required to prevent single threaded performance regressions | |
2186 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2187 | * Keep the two functions in sync. | |
2188 | */ | |
2189 | ||
2190 | /* | |
2191 | * The hash list is protected using RCU. | |
2192 | * | |
2193 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2194 | * races with d_move(). | |
2195 | * | |
2196 | * It is possible that concurrent renames can mess up our list | |
2197 | * walk here and result in missing our dentry, resulting in the | |
2198 | * false-negative result. d_lookup() protects against concurrent | |
2199 | * renames using rename_lock seqlock. | |
2200 | * | |
b0a4bb83 | 2201 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2202 | */ |
b07ad996 | 2203 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2204 | unsigned seq; |
31e6b01f | 2205 | |
31e6b01f | 2206 | seqretry: |
12f8ad4b LT |
2207 | /* |
2208 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2209 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2210 | * |
2211 | * The caller must perform a seqcount check in order | |
da53be12 | 2212 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2213 | * |
2214 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2215 | * we don't wait for the sequence count to stabilize if it | |
2216 | * is in the middle of a sequence change. If we do the slow | |
2217 | * dentry compare, we will do seqretries until it is stable, | |
2218 | * and if we end up with a successful lookup, we actually | |
2219 | * want to exit RCU lookup anyway. | |
d4c91a8f AV |
2220 | * |
2221 | * Note that raw_seqcount_begin still *does* smp_rmb(), so | |
2222 | * we are still guaranteed NUL-termination of ->d_name.name. | |
12f8ad4b LT |
2223 | */ |
2224 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2225 | if (dentry->d_parent != parent) |
2226 | continue; | |
2e321806 LT |
2227 | if (d_unhashed(dentry)) |
2228 | continue; | |
12f8ad4b | 2229 | |
830c0f0e | 2230 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
d4c91a8f AV |
2231 | int tlen; |
2232 | const char *tname; | |
26fe5750 LT |
2233 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2234 | continue; | |
d4c91a8f AV |
2235 | tlen = dentry->d_name.len; |
2236 | tname = dentry->d_name.name; | |
2237 | /* we want a consistent (name,len) pair */ | |
2238 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2239 | cpu_relax(); | |
12f8ad4b LT |
2240 | goto seqretry; |
2241 | } | |
6fa67e70 | 2242 | if (parent->d_op->d_compare(dentry, |
d4c91a8f AV |
2243 | tlen, tname, name) != 0) |
2244 | continue; | |
2245 | } else { | |
2246 | if (dentry->d_name.hash_len != hashlen) | |
2247 | continue; | |
2248 | if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0) | |
2249 | continue; | |
31e6b01f | 2250 | } |
da53be12 | 2251 | *seqp = seq; |
d4c91a8f | 2252 | return dentry; |
31e6b01f NP |
2253 | } |
2254 | return NULL; | |
2255 | } | |
2256 | ||
1da177e4 LT |
2257 | /** |
2258 | * d_lookup - search for a dentry | |
2259 | * @parent: parent dentry | |
2260 | * @name: qstr of name we wish to find | |
b04f784e | 2261 | * Returns: dentry, or NULL |
1da177e4 | 2262 | * |
b04f784e NP |
2263 | * d_lookup searches the children of the parent dentry for the name in |
2264 | * question. If the dentry is found its reference count is incremented and the | |
2265 | * dentry is returned. The caller must use dput to free the entry when it has | |
2266 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2267 | */ |
da2d8455 | 2268 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2269 | { |
31e6b01f | 2270 | struct dentry *dentry; |
949854d0 | 2271 | unsigned seq; |
1da177e4 | 2272 | |
b8314f93 DY |
2273 | do { |
2274 | seq = read_seqbegin(&rename_lock); | |
2275 | dentry = __d_lookup(parent, name); | |
2276 | if (dentry) | |
1da177e4 LT |
2277 | break; |
2278 | } while (read_seqretry(&rename_lock, seq)); | |
2279 | return dentry; | |
2280 | } | |
ec4f8605 | 2281 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2282 | |
31e6b01f | 2283 | /** |
b04f784e NP |
2284 | * __d_lookup - search for a dentry (racy) |
2285 | * @parent: parent dentry | |
2286 | * @name: qstr of name we wish to find | |
2287 | * Returns: dentry, or NULL | |
2288 | * | |
2289 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2290 | * false-negative result due to unrelated rename activity. | |
2291 | * | |
2292 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2293 | * however it must be used carefully, eg. with a following d_lookup in | |
2294 | * the case of failure. | |
2295 | * | |
2296 | * __d_lookup callers must be commented. | |
2297 | */ | |
a713ca2a | 2298 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2299 | { |
1da177e4 | 2300 | unsigned int hash = name->hash; |
8387ff25 | 2301 | struct hlist_bl_head *b = d_hash(hash); |
ceb5bdc2 | 2302 | struct hlist_bl_node *node; |
31e6b01f | 2303 | struct dentry *found = NULL; |
665a7583 | 2304 | struct dentry *dentry; |
1da177e4 | 2305 | |
31e6b01f NP |
2306 | /* |
2307 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2308 | * required to prevent single threaded performance regressions | |
2309 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2310 | * Keep the two functions in sync. | |
2311 | */ | |
2312 | ||
b04f784e NP |
2313 | /* |
2314 | * The hash list is protected using RCU. | |
2315 | * | |
2316 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2317 | * with d_move(). | |
2318 | * | |
2319 | * It is possible that concurrent renames can mess up our list | |
2320 | * walk here and result in missing our dentry, resulting in the | |
2321 | * false-negative result. d_lookup() protects against concurrent | |
2322 | * renames using rename_lock seqlock. | |
2323 | * | |
b0a4bb83 | 2324 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2325 | */ |
1da177e4 LT |
2326 | rcu_read_lock(); |
2327 | ||
b07ad996 | 2328 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2329 | |
1da177e4 LT |
2330 | if (dentry->d_name.hash != hash) |
2331 | continue; | |
1da177e4 LT |
2332 | |
2333 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2334 | if (dentry->d_parent != parent) |
2335 | goto next; | |
d0185c08 LT |
2336 | if (d_unhashed(dentry)) |
2337 | goto next; | |
2338 | ||
d4c91a8f AV |
2339 | if (!d_same_name(dentry, parent, name)) |
2340 | goto next; | |
1da177e4 | 2341 | |
98474236 | 2342 | dentry->d_lockref.count++; |
d0185c08 | 2343 | found = dentry; |
1da177e4 LT |
2344 | spin_unlock(&dentry->d_lock); |
2345 | break; | |
2346 | next: | |
2347 | spin_unlock(&dentry->d_lock); | |
2348 | } | |
2349 | rcu_read_unlock(); | |
2350 | ||
2351 | return found; | |
2352 | } | |
2353 | ||
3e7e241f EB |
2354 | /** |
2355 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2356 | * @dir: Directory to search in | |
2357 | * @name: qstr of name we wish to find | |
2358 | * | |
4f522a24 | 2359 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2360 | */ |
2361 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2362 | { | |
3e7e241f EB |
2363 | /* |
2364 | * Check for a fs-specific hash function. Note that we must | |
2365 | * calculate the standard hash first, as the d_op->d_hash() | |
2366 | * routine may choose to leave the hash value unchanged. | |
2367 | */ | |
8387ff25 | 2368 | name->hash = full_name_hash(dir, name->name, name->len); |
fb045adb | 2369 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2370 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2371 | if (unlikely(err < 0)) |
2372 | return ERR_PTR(err); | |
3e7e241f | 2373 | } |
4f522a24 | 2374 | return d_lookup(dir, name); |
3e7e241f | 2375 | } |
4f522a24 | 2376 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2377 | |
1da177e4 LT |
2378 | /* |
2379 | * When a file is deleted, we have two options: | |
2380 | * - turn this dentry into a negative dentry | |
2381 | * - unhash this dentry and free it. | |
2382 | * | |
2383 | * Usually, we want to just turn this into | |
2384 | * a negative dentry, but if anybody else is | |
2385 | * currently using the dentry or the inode | |
2386 | * we can't do that and we fall back on removing | |
2387 | * it from the hash queues and waiting for | |
2388 | * it to be deleted later when it has no users | |
2389 | */ | |
2390 | ||
2391 | /** | |
2392 | * d_delete - delete a dentry | |
2393 | * @dentry: The dentry to delete | |
2394 | * | |
2395 | * Turn the dentry into a negative dentry if possible, otherwise | |
2396 | * remove it from the hash queues so it can be deleted later | |
2397 | */ | |
2398 | ||
2399 | void d_delete(struct dentry * dentry) | |
2400 | { | |
873feea0 | 2401 | struct inode *inode; |
7a91bf7f | 2402 | int isdir = 0; |
1da177e4 LT |
2403 | /* |
2404 | * Are we the only user? | |
2405 | */ | |
357f8e65 | 2406 | again: |
1da177e4 | 2407 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2408 | inode = dentry->d_inode; |
2409 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2410 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2411 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2412 | spin_unlock(&dentry->d_lock); |
2413 | cpu_relax(); | |
2414 | goto again; | |
2415 | } | |
13e3c5e5 | 2416 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2417 | dentry_unlink_inode(dentry); |
7a91bf7f | 2418 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2419 | return; |
2420 | } | |
2421 | ||
2422 | if (!d_unhashed(dentry)) | |
2423 | __d_drop(dentry); | |
2424 | ||
2425 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2426 | |
2427 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2428 | } |
ec4f8605 | 2429 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2430 | |
15d3c589 | 2431 | static void __d_rehash(struct dentry *entry) |
1da177e4 | 2432 | { |
15d3c589 | 2433 | struct hlist_bl_head *b = d_hash(entry->d_name.hash); |
db470ce8 | 2434 | |
1879fd6a | 2435 | hlist_bl_lock(b); |
b07ad996 | 2436 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2437 | hlist_bl_unlock(b); |
1da177e4 LT |
2438 | } |
2439 | ||
2440 | /** | |
2441 | * d_rehash - add an entry back to the hash | |
2442 | * @entry: dentry to add to the hash | |
2443 | * | |
2444 | * Adds a dentry to the hash according to its name. | |
2445 | */ | |
2446 | ||
2447 | void d_rehash(struct dentry * entry) | |
2448 | { | |
1da177e4 | 2449 | spin_lock(&entry->d_lock); |
15d3c589 | 2450 | __d_rehash(entry); |
1da177e4 | 2451 | spin_unlock(&entry->d_lock); |
1da177e4 | 2452 | } |
ec4f8605 | 2453 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2454 | |
84e710da AV |
2455 | static inline unsigned start_dir_add(struct inode *dir) |
2456 | { | |
2457 | ||
2458 | for (;;) { | |
2459 | unsigned n = dir->i_dir_seq; | |
2460 | if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n) | |
2461 | return n; | |
2462 | cpu_relax(); | |
2463 | } | |
2464 | } | |
2465 | ||
2466 | static inline void end_dir_add(struct inode *dir, unsigned n) | |
2467 | { | |
2468 | smp_store_release(&dir->i_dir_seq, n + 2); | |
2469 | } | |
2470 | ||
d9171b93 AV |
2471 | static void d_wait_lookup(struct dentry *dentry) |
2472 | { | |
2473 | if (d_in_lookup(dentry)) { | |
2474 | DECLARE_WAITQUEUE(wait, current); | |
2475 | add_wait_queue(dentry->d_wait, &wait); | |
2476 | do { | |
2477 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2478 | spin_unlock(&dentry->d_lock); | |
2479 | schedule(); | |
2480 | spin_lock(&dentry->d_lock); | |
2481 | } while (d_in_lookup(dentry)); | |
2482 | } | |
2483 | } | |
2484 | ||
94bdd655 | 2485 | struct dentry *d_alloc_parallel(struct dentry *parent, |
d9171b93 AV |
2486 | const struct qstr *name, |
2487 | wait_queue_head_t *wq) | |
94bdd655 | 2488 | { |
94bdd655 | 2489 | unsigned int hash = name->hash; |
94bdd655 AV |
2490 | struct hlist_bl_head *b = in_lookup_hash(parent, hash); |
2491 | struct hlist_bl_node *node; | |
2492 | struct dentry *new = d_alloc(parent, name); | |
2493 | struct dentry *dentry; | |
2494 | unsigned seq, r_seq, d_seq; | |
2495 | ||
2496 | if (unlikely(!new)) | |
2497 | return ERR_PTR(-ENOMEM); | |
2498 | ||
2499 | retry: | |
2500 | rcu_read_lock(); | |
3623c1f3 | 2501 | seq = smp_load_acquire(&parent->d_inode->i_dir_seq); |
94bdd655 AV |
2502 | r_seq = read_seqbegin(&rename_lock); |
2503 | dentry = __d_lookup_rcu(parent, name, &d_seq); | |
2504 | if (unlikely(dentry)) { | |
2505 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2506 | rcu_read_unlock(); | |
2507 | goto retry; | |
2508 | } | |
2509 | if (read_seqcount_retry(&dentry->d_seq, d_seq)) { | |
2510 | rcu_read_unlock(); | |
2511 | dput(dentry); | |
2512 | goto retry; | |
2513 | } | |
2514 | rcu_read_unlock(); | |
2515 | dput(new); | |
2516 | return dentry; | |
2517 | } | |
2518 | if (unlikely(read_seqretry(&rename_lock, r_seq))) { | |
2519 | rcu_read_unlock(); | |
2520 | goto retry; | |
2521 | } | |
3623c1f3 WD |
2522 | |
2523 | if (unlikely(seq & 1)) { | |
2524 | rcu_read_unlock(); | |
2525 | goto retry; | |
2526 | } | |
2527 | ||
94bdd655 | 2528 | hlist_bl_lock(b); |
2565b271 | 2529 | if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) { |
94bdd655 AV |
2530 | hlist_bl_unlock(b); |
2531 | rcu_read_unlock(); | |
2532 | goto retry; | |
2533 | } | |
94bdd655 AV |
2534 | /* |
2535 | * No changes for the parent since the beginning of d_lookup(). | |
2536 | * Since all removals from the chain happen with hlist_bl_lock(), | |
2537 | * any potential in-lookup matches are going to stay here until | |
2538 | * we unlock the chain. All fields are stable in everything | |
2539 | * we encounter. | |
2540 | */ | |
2541 | hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) { | |
2542 | if (dentry->d_name.hash != hash) | |
2543 | continue; | |
2544 | if (dentry->d_parent != parent) | |
2545 | continue; | |
d4c91a8f AV |
2546 | if (!d_same_name(dentry, parent, name)) |
2547 | continue; | |
94bdd655 | 2548 | hlist_bl_unlock(b); |
e7d6ef97 AV |
2549 | /* now we can try to grab a reference */ |
2550 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2551 | rcu_read_unlock(); | |
2552 | goto retry; | |
2553 | } | |
2554 | ||
2555 | rcu_read_unlock(); | |
2556 | /* | |
2557 | * somebody is likely to be still doing lookup for it; | |
2558 | * wait for them to finish | |
2559 | */ | |
d9171b93 AV |
2560 | spin_lock(&dentry->d_lock); |
2561 | d_wait_lookup(dentry); | |
2562 | /* | |
2563 | * it's not in-lookup anymore; in principle we should repeat | |
2564 | * everything from dcache lookup, but it's likely to be what | |
2565 | * d_lookup() would've found anyway. If it is, just return it; | |
2566 | * otherwise we really have to repeat the whole thing. | |
2567 | */ | |
2568 | if (unlikely(dentry->d_name.hash != hash)) | |
2569 | goto mismatch; | |
2570 | if (unlikely(dentry->d_parent != parent)) | |
2571 | goto mismatch; | |
2572 | if (unlikely(d_unhashed(dentry))) | |
2573 | goto mismatch; | |
d4c91a8f AV |
2574 | if (unlikely(!d_same_name(dentry, parent, name))) |
2575 | goto mismatch; | |
d9171b93 AV |
2576 | /* OK, it *is* a hashed match; return it */ |
2577 | spin_unlock(&dentry->d_lock); | |
94bdd655 AV |
2578 | dput(new); |
2579 | return dentry; | |
2580 | } | |
e7d6ef97 | 2581 | rcu_read_unlock(); |
94bdd655 AV |
2582 | /* we can't take ->d_lock here; it's OK, though. */ |
2583 | new->d_flags |= DCACHE_PAR_LOOKUP; | |
d9171b93 | 2584 | new->d_wait = wq; |
94bdd655 AV |
2585 | hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b); |
2586 | hlist_bl_unlock(b); | |
2587 | return new; | |
d9171b93 AV |
2588 | mismatch: |
2589 | spin_unlock(&dentry->d_lock); | |
2590 | dput(dentry); | |
2591 | goto retry; | |
94bdd655 AV |
2592 | } |
2593 | EXPORT_SYMBOL(d_alloc_parallel); | |
2594 | ||
85c7f810 AV |
2595 | void __d_lookup_done(struct dentry *dentry) |
2596 | { | |
94bdd655 AV |
2597 | struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent, |
2598 | dentry->d_name.hash); | |
2599 | hlist_bl_lock(b); | |
85c7f810 | 2600 | dentry->d_flags &= ~DCACHE_PAR_LOOKUP; |
94bdd655 | 2601 | __hlist_bl_del(&dentry->d_u.d_in_lookup_hash); |
d9171b93 AV |
2602 | wake_up_all(dentry->d_wait); |
2603 | dentry->d_wait = NULL; | |
94bdd655 AV |
2604 | hlist_bl_unlock(b); |
2605 | INIT_HLIST_NODE(&dentry->d_u.d_alias); | |
d9171b93 | 2606 | INIT_LIST_HEAD(&dentry->d_lru); |
85c7f810 AV |
2607 | } |
2608 | EXPORT_SYMBOL(__d_lookup_done); | |
ed782b5a AV |
2609 | |
2610 | /* inode->i_lock held if inode is non-NULL */ | |
2611 | ||
2612 | static inline void __d_add(struct dentry *dentry, struct inode *inode) | |
2613 | { | |
84e710da AV |
2614 | struct inode *dir = NULL; |
2615 | unsigned n; | |
0568d705 | 2616 | spin_lock(&dentry->d_lock); |
84e710da AV |
2617 | if (unlikely(d_in_lookup(dentry))) { |
2618 | dir = dentry->d_parent->d_inode; | |
2619 | n = start_dir_add(dir); | |
85c7f810 | 2620 | __d_lookup_done(dentry); |
84e710da | 2621 | } |
ed782b5a | 2622 | if (inode) { |
0568d705 AV |
2623 | unsigned add_flags = d_flags_for_inode(inode); |
2624 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
2625 | raw_write_seqcount_begin(&dentry->d_seq); | |
2626 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2627 | raw_write_seqcount_end(&dentry->d_seq); | |
affda484 | 2628 | fsnotify_update_flags(dentry); |
ed782b5a | 2629 | } |
15d3c589 | 2630 | __d_rehash(dentry); |
84e710da AV |
2631 | if (dir) |
2632 | end_dir_add(dir, n); | |
0568d705 AV |
2633 | spin_unlock(&dentry->d_lock); |
2634 | if (inode) | |
2635 | spin_unlock(&inode->i_lock); | |
ed782b5a AV |
2636 | } |
2637 | ||
34d0d19d AV |
2638 | /** |
2639 | * d_add - add dentry to hash queues | |
2640 | * @entry: dentry to add | |
2641 | * @inode: The inode to attach to this dentry | |
2642 | * | |
2643 | * This adds the entry to the hash queues and initializes @inode. | |
2644 | * The entry was actually filled in earlier during d_alloc(). | |
2645 | */ | |
2646 | ||
2647 | void d_add(struct dentry *entry, struct inode *inode) | |
2648 | { | |
b9680917 AV |
2649 | if (inode) { |
2650 | security_d_instantiate(entry, inode); | |
ed782b5a | 2651 | spin_lock(&inode->i_lock); |
b9680917 | 2652 | } |
ed782b5a | 2653 | __d_add(entry, inode); |
34d0d19d AV |
2654 | } |
2655 | EXPORT_SYMBOL(d_add); | |
2656 | ||
668d0cd5 AV |
2657 | /** |
2658 | * d_exact_alias - find and hash an exact unhashed alias | |
2659 | * @entry: dentry to add | |
2660 | * @inode: The inode to go with this dentry | |
2661 | * | |
2662 | * If an unhashed dentry with the same name/parent and desired | |
2663 | * inode already exists, hash and return it. Otherwise, return | |
2664 | * NULL. | |
2665 | * | |
2666 | * Parent directory should be locked. | |
2667 | */ | |
2668 | struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode) | |
2669 | { | |
2670 | struct dentry *alias; | |
668d0cd5 AV |
2671 | unsigned int hash = entry->d_name.hash; |
2672 | ||
2673 | spin_lock(&inode->i_lock); | |
2674 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { | |
2675 | /* | |
2676 | * Don't need alias->d_lock here, because aliases with | |
2677 | * d_parent == entry->d_parent are not subject to name or | |
2678 | * parent changes, because the parent inode i_mutex is held. | |
2679 | */ | |
2680 | if (alias->d_name.hash != hash) | |
2681 | continue; | |
2682 | if (alias->d_parent != entry->d_parent) | |
2683 | continue; | |
d4c91a8f | 2684 | if (!d_same_name(alias, entry->d_parent, &entry->d_name)) |
668d0cd5 AV |
2685 | continue; |
2686 | spin_lock(&alias->d_lock); | |
2687 | if (!d_unhashed(alias)) { | |
2688 | spin_unlock(&alias->d_lock); | |
2689 | alias = NULL; | |
2690 | } else { | |
2691 | __dget_dlock(alias); | |
15d3c589 | 2692 | __d_rehash(alias); |
668d0cd5 AV |
2693 | spin_unlock(&alias->d_lock); |
2694 | } | |
2695 | spin_unlock(&inode->i_lock); | |
2696 | return alias; | |
2697 | } | |
2698 | spin_unlock(&inode->i_lock); | |
2699 | return NULL; | |
2700 | } | |
2701 | EXPORT_SYMBOL(d_exact_alias); | |
2702 | ||
fb2d5b86 NP |
2703 | /** |
2704 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2705 | * @dentry: dentry to be updated | |
2706 | * @name: new name | |
2707 | * | |
2708 | * Update a case insensitive dentry with new case of name. | |
2709 | * | |
2710 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2711 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2712 | * lengths). | |
2713 | * | |
2714 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2715 | * keep renames and concurrent inserts, and readdir(2) away). | |
2716 | */ | |
9aba36de | 2717 | void dentry_update_name_case(struct dentry *dentry, const struct qstr *name) |
fb2d5b86 | 2718 | { |
5955102c | 2719 | BUG_ON(!inode_is_locked(dentry->d_parent->d_inode)); |
fb2d5b86 NP |
2720 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2721 | ||
fb2d5b86 | 2722 | spin_lock(&dentry->d_lock); |
31e6b01f | 2723 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2724 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2725 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2726 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2727 | } |
2728 | EXPORT_SYMBOL(dentry_update_name_case); | |
2729 | ||
8d85b484 | 2730 | static void swap_names(struct dentry *dentry, struct dentry *target) |
1da177e4 | 2731 | { |
8d85b484 AV |
2732 | if (unlikely(dname_external(target))) { |
2733 | if (unlikely(dname_external(dentry))) { | |
1da177e4 LT |
2734 | /* |
2735 | * Both external: swap the pointers | |
2736 | */ | |
9a8d5bb4 | 2737 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2738 | } else { |
2739 | /* | |
2740 | * dentry:internal, target:external. Steal target's | |
2741 | * storage and make target internal. | |
2742 | */ | |
321bcf92 BF |
2743 | memcpy(target->d_iname, dentry->d_name.name, |
2744 | dentry->d_name.len + 1); | |
1da177e4 LT |
2745 | dentry->d_name.name = target->d_name.name; |
2746 | target->d_name.name = target->d_iname; | |
2747 | } | |
2748 | } else { | |
8d85b484 | 2749 | if (unlikely(dname_external(dentry))) { |
1da177e4 LT |
2750 | /* |
2751 | * dentry:external, target:internal. Give dentry's | |
2752 | * storage to target and make dentry internal | |
2753 | */ | |
2754 | memcpy(dentry->d_iname, target->d_name.name, | |
2755 | target->d_name.len + 1); | |
2756 | target->d_name.name = dentry->d_name.name; | |
2757 | dentry->d_name.name = dentry->d_iname; | |
2758 | } else { | |
2759 | /* | |
da1ce067 | 2760 | * Both are internal. |
1da177e4 | 2761 | */ |
da1ce067 MS |
2762 | unsigned int i; |
2763 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
2764 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { | |
2765 | swap(((long *) &dentry->d_iname)[i], | |
2766 | ((long *) &target->d_iname)[i]); | |
2767 | } | |
1da177e4 LT |
2768 | } |
2769 | } | |
a28ddb87 | 2770 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
1da177e4 LT |
2771 | } |
2772 | ||
8d85b484 AV |
2773 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2774 | { | |
2775 | struct external_name *old_name = NULL; | |
2776 | if (unlikely(dname_external(dentry))) | |
2777 | old_name = external_name(dentry); | |
2778 | if (unlikely(dname_external(target))) { | |
2779 | atomic_inc(&external_name(target)->u.count); | |
2780 | dentry->d_name = target->d_name; | |
2781 | } else { | |
2782 | memcpy(dentry->d_iname, target->d_name.name, | |
2783 | target->d_name.len + 1); | |
2784 | dentry->d_name.name = dentry->d_iname; | |
2785 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2786 | } | |
2787 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
6d794237 | 2788 | call_rcu(&old_name->u.head, __d_free_external_name); |
8d85b484 AV |
2789 | } |
2790 | ||
2fd6b7f5 NP |
2791 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2792 | { | |
2793 | /* | |
2794 | * XXXX: do we really need to take target->d_lock? | |
2795 | */ | |
2796 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2797 | spin_lock(&target->d_parent->d_lock); | |
2798 | else { | |
2799 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2800 | spin_lock(&dentry->d_parent->d_lock); | |
2801 | spin_lock_nested(&target->d_parent->d_lock, | |
2802 | DENTRY_D_LOCK_NESTED); | |
2803 | } else { | |
2804 | spin_lock(&target->d_parent->d_lock); | |
2805 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2806 | DENTRY_D_LOCK_NESTED); | |
2807 | } | |
2808 | } | |
2809 | if (target < dentry) { | |
2810 | spin_lock_nested(&target->d_lock, 2); | |
2811 | spin_lock_nested(&dentry->d_lock, 3); | |
2812 | } else { | |
2813 | spin_lock_nested(&dentry->d_lock, 2); | |
2814 | spin_lock_nested(&target->d_lock, 3); | |
2815 | } | |
2816 | } | |
2817 | ||
986c0194 | 2818 | static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target) |
2fd6b7f5 NP |
2819 | { |
2820 | if (target->d_parent != dentry->d_parent) | |
2821 | spin_unlock(&dentry->d_parent->d_lock); | |
2822 | if (target->d_parent != target) | |
2823 | spin_unlock(&target->d_parent->d_lock); | |
986c0194 AV |
2824 | spin_unlock(&target->d_lock); |
2825 | spin_unlock(&dentry->d_lock); | |
2fd6b7f5 NP |
2826 | } |
2827 | ||
1da177e4 | 2828 | /* |
2fd6b7f5 NP |
2829 | * When switching names, the actual string doesn't strictly have to |
2830 | * be preserved in the target - because we're dropping the target | |
2831 | * anyway. As such, we can just do a simple memcpy() to copy over | |
d2fa4a84 ME |
2832 | * the new name before we switch, unless we are going to rehash |
2833 | * it. Note that if we *do* unhash the target, we are not allowed | |
2834 | * to rehash it without giving it a new name/hash key - whether | |
2835 | * we swap or overwrite the names here, resulting name won't match | |
2836 | * the reality in filesystem; it's only there for d_path() purposes. | |
2837 | * Note that all of this is happening under rename_lock, so the | |
2838 | * any hash lookup seeing it in the middle of manipulations will | |
2839 | * be discarded anyway. So we do not care what happens to the hash | |
2840 | * key in that case. | |
1da177e4 | 2841 | */ |
9eaef27b | 2842 | /* |
18367501 | 2843 | * __d_move - move a dentry |
1da177e4 LT |
2844 | * @dentry: entry to move |
2845 | * @target: new dentry | |
da1ce067 | 2846 | * @exchange: exchange the two dentries |
1da177e4 LT |
2847 | * |
2848 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2849 | * dcache entries should not be moved in this way. Caller must hold |
2850 | * rename_lock, the i_mutex of the source and target directories, | |
2851 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2852 | */ |
da1ce067 MS |
2853 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2854 | bool exchange) | |
1da177e4 | 2855 | { |
84e710da AV |
2856 | struct inode *dir = NULL; |
2857 | unsigned n; | |
1da177e4 LT |
2858 | if (!dentry->d_inode) |
2859 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2860 | ||
2fd6b7f5 NP |
2861 | BUG_ON(d_ancestor(dentry, target)); |
2862 | BUG_ON(d_ancestor(target, dentry)); | |
2863 | ||
2fd6b7f5 | 2864 | dentry_lock_for_move(dentry, target); |
84e710da AV |
2865 | if (unlikely(d_in_lookup(target))) { |
2866 | dir = target->d_parent->d_inode; | |
2867 | n = start_dir_add(dir); | |
85c7f810 | 2868 | __d_lookup_done(target); |
84e710da | 2869 | } |
1da177e4 | 2870 | |
31e6b01f | 2871 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2872 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2873 | |
15d3c589 | 2874 | /* unhash both */ |
db470ce8 N |
2875 | /* ___d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2876 | ___d_drop(dentry); | |
2877 | ___d_drop(target); | |
1da177e4 | 2878 | |
1da177e4 | 2879 | /* Switch the names.. */ |
8d85b484 AV |
2880 | if (exchange) |
2881 | swap_names(dentry, target); | |
2882 | else | |
2883 | copy_name(dentry, target); | |
1da177e4 | 2884 | |
15d3c589 AV |
2885 | /* rehash in new place(s) */ |
2886 | __d_rehash(dentry); | |
2887 | if (exchange) | |
2888 | __d_rehash(target); | |
db470ce8 N |
2889 | else |
2890 | target->d_hash.pprev = NULL; | |
15d3c589 | 2891 | |
63cf427a | 2892 | /* ... and switch them in the tree */ |
1da177e4 | 2893 | if (IS_ROOT(dentry)) { |
63cf427a | 2894 | /* splicing a tree */ |
3d56c25e | 2895 | dentry->d_flags |= DCACHE_RCUACCESS; |
1da177e4 LT |
2896 | dentry->d_parent = target->d_parent; |
2897 | target->d_parent = target; | |
946e51f2 AV |
2898 | list_del_init(&target->d_child); |
2899 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
1da177e4 | 2900 | } else { |
63cf427a | 2901 | /* swapping two dentries */ |
9a8d5bb4 | 2902 | swap(dentry->d_parent, target->d_parent); |
946e51f2 AV |
2903 | list_move(&target->d_child, &target->d_parent->d_subdirs); |
2904 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
63cf427a | 2905 | if (exchange) |
affda484 AV |
2906 | fsnotify_update_flags(target); |
2907 | fsnotify_update_flags(dentry); | |
1da177e4 LT |
2908 | } |
2909 | ||
31e6b01f NP |
2910 | write_seqcount_end(&target->d_seq); |
2911 | write_seqcount_end(&dentry->d_seq); | |
2912 | ||
84e710da AV |
2913 | if (dir) |
2914 | end_dir_add(dir, n); | |
986c0194 | 2915 | dentry_unlock_for_move(dentry, target); |
18367501 AV |
2916 | } |
2917 | ||
2918 | /* | |
2919 | * d_move - move a dentry | |
2920 | * @dentry: entry to move | |
2921 | * @target: new dentry | |
2922 | * | |
2923 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2924 | * dcache entries should not be moved in this way. See the locking |
2925 | * requirements for __d_move. | |
18367501 AV |
2926 | */ |
2927 | void d_move(struct dentry *dentry, struct dentry *target) | |
2928 | { | |
2929 | write_seqlock(&rename_lock); | |
da1ce067 | 2930 | __d_move(dentry, target, false); |
1da177e4 | 2931 | write_sequnlock(&rename_lock); |
9eaef27b | 2932 | } |
ec4f8605 | 2933 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2934 | |
da1ce067 MS |
2935 | /* |
2936 | * d_exchange - exchange two dentries | |
2937 | * @dentry1: first dentry | |
2938 | * @dentry2: second dentry | |
2939 | */ | |
2940 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2941 | { | |
2942 | write_seqlock(&rename_lock); | |
2943 | ||
2944 | WARN_ON(!dentry1->d_inode); | |
2945 | WARN_ON(!dentry2->d_inode); | |
2946 | WARN_ON(IS_ROOT(dentry1)); | |
2947 | WARN_ON(IS_ROOT(dentry2)); | |
2948 | ||
2949 | __d_move(dentry1, dentry2, true); | |
2950 | ||
2951 | write_sequnlock(&rename_lock); | |
2952 | } | |
2953 | ||
e2761a11 OH |
2954 | /** |
2955 | * d_ancestor - search for an ancestor | |
2956 | * @p1: ancestor dentry | |
2957 | * @p2: child dentry | |
2958 | * | |
2959 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2960 | * an ancestor of p2, else NULL. | |
9eaef27b | 2961 | */ |
e2761a11 | 2962 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2963 | { |
2964 | struct dentry *p; | |
2965 | ||
871c0067 | 2966 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2967 | if (p->d_parent == p1) |
e2761a11 | 2968 | return p; |
9eaef27b | 2969 | } |
e2761a11 | 2970 | return NULL; |
9eaef27b TM |
2971 | } |
2972 | ||
2973 | /* | |
2974 | * This helper attempts to cope with remotely renamed directories | |
2975 | * | |
2976 | * It assumes that the caller is already holding | |
a03e283b | 2977 | * dentry->d_parent->d_inode->i_mutex, and rename_lock |
9eaef27b TM |
2978 | * |
2979 | * Note: If ever the locking in lock_rename() changes, then please | |
2980 | * remember to update this too... | |
9eaef27b | 2981 | */ |
b5ae6b15 | 2982 | static int __d_unalias(struct inode *inode, |
873feea0 | 2983 | struct dentry *dentry, struct dentry *alias) |
9eaef27b | 2984 | { |
9902af79 AV |
2985 | struct mutex *m1 = NULL; |
2986 | struct rw_semaphore *m2 = NULL; | |
3d330dc1 | 2987 | int ret = -ESTALE; |
9eaef27b TM |
2988 | |
2989 | /* If alias and dentry share a parent, then no extra locks required */ | |
2990 | if (alias->d_parent == dentry->d_parent) | |
2991 | goto out_unalias; | |
2992 | ||
9eaef27b | 2993 | /* See lock_rename() */ |
9eaef27b TM |
2994 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2995 | goto out_err; | |
2996 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
9902af79 | 2997 | if (!inode_trylock_shared(alias->d_parent->d_inode)) |
9eaef27b | 2998 | goto out_err; |
9902af79 | 2999 | m2 = &alias->d_parent->d_inode->i_rwsem; |
9eaef27b | 3000 | out_unalias: |
8ed936b5 | 3001 | __d_move(alias, dentry, false); |
b5ae6b15 | 3002 | ret = 0; |
9eaef27b | 3003 | out_err: |
9eaef27b | 3004 | if (m2) |
9902af79 | 3005 | up_read(m2); |
9eaef27b TM |
3006 | if (m1) |
3007 | mutex_unlock(m1); | |
3008 | return ret; | |
3009 | } | |
3010 | ||
3f70bd51 BF |
3011 | /** |
3012 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
3013 | * @inode: the inode which may have a disconnected dentry | |
3014 | * @dentry: a negative dentry which we want to point to the inode. | |
3015 | * | |
da093a9b BF |
3016 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
3017 | * place of the given dentry and return it, else simply d_add the inode | |
3018 | * to the dentry and return NULL. | |
3f70bd51 | 3019 | * |
908790fa BF |
3020 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
3021 | * we should error out: directories can't have multiple aliases. | |
3022 | * | |
3f70bd51 BF |
3023 | * This is needed in the lookup routine of any filesystem that is exportable |
3024 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
3025 | * | |
3026 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
3027 | * is returned. This matches the expected return value of ->lookup. | |
3028 | * | |
3029 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
3030 | * In that case, we know that the inode will be a regular file, and also this | |
3031 | * will only occur during atomic_open. So we need to check for the dentry | |
3032 | * being already hashed only in the final case. | |
3033 | */ | |
3034 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
3035 | { | |
3f70bd51 BF |
3036 | if (IS_ERR(inode)) |
3037 | return ERR_CAST(inode); | |
3038 | ||
770bfad8 DH |
3039 | BUG_ON(!d_unhashed(dentry)); |
3040 | ||
de689f5e | 3041 | if (!inode) |
b5ae6b15 | 3042 | goto out; |
de689f5e | 3043 | |
b9680917 | 3044 | security_d_instantiate(dentry, inode); |
873feea0 | 3045 | spin_lock(&inode->i_lock); |
9eaef27b | 3046 | if (S_ISDIR(inode->i_mode)) { |
b5ae6b15 AV |
3047 | struct dentry *new = __d_find_any_alias(inode); |
3048 | if (unlikely(new)) { | |
a03e283b EB |
3049 | /* The reference to new ensures it remains an alias */ |
3050 | spin_unlock(&inode->i_lock); | |
18367501 | 3051 | write_seqlock(&rename_lock); |
b5ae6b15 AV |
3052 | if (unlikely(d_ancestor(new, dentry))) { |
3053 | write_sequnlock(&rename_lock); | |
b5ae6b15 AV |
3054 | dput(new); |
3055 | new = ERR_PTR(-ELOOP); | |
3056 | pr_warn_ratelimited( | |
3057 | "VFS: Lookup of '%s' in %s %s" | |
3058 | " would have caused loop\n", | |
3059 | dentry->d_name.name, | |
3060 | inode->i_sb->s_type->name, | |
3061 | inode->i_sb->s_id); | |
3062 | } else if (!IS_ROOT(new)) { | |
3063 | int err = __d_unalias(inode, dentry, new); | |
18367501 | 3064 | write_sequnlock(&rename_lock); |
b5ae6b15 AV |
3065 | if (err) { |
3066 | dput(new); | |
3067 | new = ERR_PTR(err); | |
3068 | } | |
18367501 | 3069 | } else { |
b5ae6b15 AV |
3070 | __d_move(new, dentry, false); |
3071 | write_sequnlock(&rename_lock); | |
dd179946 | 3072 | } |
b5ae6b15 AV |
3073 | iput(inode); |
3074 | return new; | |
9eaef27b | 3075 | } |
770bfad8 | 3076 | } |
b5ae6b15 | 3077 | out: |
ed782b5a | 3078 | __d_add(dentry, inode); |
b5ae6b15 | 3079 | return NULL; |
770bfad8 | 3080 | } |
b5ae6b15 | 3081 | EXPORT_SYMBOL(d_splice_alias); |
770bfad8 | 3082 | |
cdd16d02 | 3083 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
3084 | { |
3085 | *buflen -= namelen; | |
3086 | if (*buflen < 0) | |
3087 | return -ENAMETOOLONG; | |
3088 | *buffer -= namelen; | |
3089 | memcpy(*buffer, str, namelen); | |
3090 | return 0; | |
3091 | } | |
3092 | ||
232d2d60 WL |
3093 | /** |
3094 | * prepend_name - prepend a pathname in front of current buffer pointer | |
18129977 WL |
3095 | * @buffer: buffer pointer |
3096 | * @buflen: allocated length of the buffer | |
3097 | * @name: name string and length qstr structure | |
232d2d60 WL |
3098 | * |
3099 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
3100 | * make sure that either the old or the new name pointer and length are | |
3101 | * fetched. However, there may be mismatch between length and pointer. | |
3102 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
3103 | * the length is reached or a null byte is found. It also prepends "/" at | |
3104 | * the beginning of the name. The sequence number check at the caller will | |
3105 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
3106 | * due to mismatched pointer and length will be discarded. | |
6d13f694 AV |
3107 | * |
3108 | * Data dependency barrier is needed to make sure that we see that terminating | |
3109 | * NUL. Alpha strikes again, film at 11... | |
232d2d60 | 3110 | */ |
9aba36de | 3111 | static int prepend_name(char **buffer, int *buflen, const struct qstr *name) |
cdd16d02 | 3112 | { |
232d2d60 WL |
3113 | const char *dname = ACCESS_ONCE(name->name); |
3114 | u32 dlen = ACCESS_ONCE(name->len); | |
3115 | char *p; | |
3116 | ||
6d13f694 AV |
3117 | smp_read_barrier_depends(); |
3118 | ||
232d2d60 | 3119 | *buflen -= dlen + 1; |
e825196d AV |
3120 | if (*buflen < 0) |
3121 | return -ENAMETOOLONG; | |
232d2d60 WL |
3122 | p = *buffer -= dlen + 1; |
3123 | *p++ = '/'; | |
3124 | while (dlen--) { | |
3125 | char c = *dname++; | |
3126 | if (!c) | |
3127 | break; | |
3128 | *p++ = c; | |
3129 | } | |
3130 | return 0; | |
cdd16d02 MS |
3131 | } |
3132 | ||
1da177e4 | 3133 | /** |
208898c1 | 3134 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 3135 | * @path: the dentry/vfsmount to report |
02125a82 | 3136 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
3137 | * @buffer: pointer to the end of the buffer |
3138 | * @buflen: pointer to buffer length | |
552ce544 | 3139 | * |
18129977 WL |
3140 | * The function will first try to write out the pathname without taking any |
3141 | * lock other than the RCU read lock to make sure that dentries won't go away. | |
3142 | * It only checks the sequence number of the global rename_lock as any change | |
3143 | * in the dentry's d_seq will be preceded by changes in the rename_lock | |
3144 | * sequence number. If the sequence number had been changed, it will restart | |
3145 | * the whole pathname back-tracing sequence again by taking the rename_lock. | |
3146 | * In this case, there is no need to take the RCU read lock as the recursive | |
3147 | * parent pointer references will keep the dentry chain alive as long as no | |
3148 | * rename operation is performed. | |
1da177e4 | 3149 | */ |
02125a82 AV |
3150 | static int prepend_path(const struct path *path, |
3151 | const struct path *root, | |
f2eb6575 | 3152 | char **buffer, int *buflen) |
1da177e4 | 3153 | { |
ede4cebc AV |
3154 | struct dentry *dentry; |
3155 | struct vfsmount *vfsmnt; | |
3156 | struct mount *mnt; | |
f2eb6575 | 3157 | int error = 0; |
48a066e7 | 3158 | unsigned seq, m_seq = 0; |
232d2d60 WL |
3159 | char *bptr; |
3160 | int blen; | |
6092d048 | 3161 | |
48f5ec21 | 3162 | rcu_read_lock(); |
48a066e7 AV |
3163 | restart_mnt: |
3164 | read_seqbegin_or_lock(&mount_lock, &m_seq); | |
3165 | seq = 0; | |
4ec6c2ae | 3166 | rcu_read_lock(); |
232d2d60 WL |
3167 | restart: |
3168 | bptr = *buffer; | |
3169 | blen = *buflen; | |
48a066e7 | 3170 | error = 0; |
ede4cebc AV |
3171 | dentry = path->dentry; |
3172 | vfsmnt = path->mnt; | |
3173 | mnt = real_mount(vfsmnt); | |
232d2d60 | 3174 | read_seqbegin_or_lock(&rename_lock, &seq); |
f2eb6575 | 3175 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
3176 | struct dentry * parent; |
3177 | ||
1da177e4 | 3178 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
48a066e7 | 3179 | struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); |
cde93be4 EB |
3180 | /* Escaped? */ |
3181 | if (dentry != vfsmnt->mnt_root) { | |
3182 | bptr = *buffer; | |
3183 | blen = *buflen; | |
3184 | error = 3; | |
3185 | break; | |
3186 | } | |
552ce544 | 3187 | /* Global root? */ |
48a066e7 AV |
3188 | if (mnt != parent) { |
3189 | dentry = ACCESS_ONCE(mnt->mnt_mountpoint); | |
3190 | mnt = parent; | |
232d2d60 WL |
3191 | vfsmnt = &mnt->mnt; |
3192 | continue; | |
3193 | } | |
232d2d60 WL |
3194 | if (!error) |
3195 | error = is_mounted(vfsmnt) ? 1 : 2; | |
3196 | break; | |
1da177e4 LT |
3197 | } |
3198 | parent = dentry->d_parent; | |
3199 | prefetch(parent); | |
232d2d60 | 3200 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
3201 | if (error) |
3202 | break; | |
3203 | ||
1da177e4 LT |
3204 | dentry = parent; |
3205 | } | |
48f5ec21 AV |
3206 | if (!(seq & 1)) |
3207 | rcu_read_unlock(); | |
3208 | if (need_seqretry(&rename_lock, seq)) { | |
3209 | seq = 1; | |
232d2d60 | 3210 | goto restart; |
48f5ec21 AV |
3211 | } |
3212 | done_seqretry(&rename_lock, seq); | |
4ec6c2ae LZ |
3213 | |
3214 | if (!(m_seq & 1)) | |
3215 | rcu_read_unlock(); | |
48a066e7 AV |
3216 | if (need_seqretry(&mount_lock, m_seq)) { |
3217 | m_seq = 1; | |
3218 | goto restart_mnt; | |
3219 | } | |
3220 | done_seqretry(&mount_lock, m_seq); | |
1da177e4 | 3221 | |
232d2d60 WL |
3222 | if (error >= 0 && bptr == *buffer) { |
3223 | if (--blen < 0) | |
3224 | error = -ENAMETOOLONG; | |
3225 | else | |
3226 | *--bptr = '/'; | |
3227 | } | |
3228 | *buffer = bptr; | |
3229 | *buflen = blen; | |
7ea600b5 | 3230 | return error; |
f2eb6575 | 3231 | } |
be285c71 | 3232 | |
f2eb6575 MS |
3233 | /** |
3234 | * __d_path - return the path of a dentry | |
3235 | * @path: the dentry/vfsmount to report | |
02125a82 | 3236 | * @root: root vfsmnt/dentry |
cd956a1c | 3237 | * @buf: buffer to return value in |
f2eb6575 MS |
3238 | * @buflen: buffer length |
3239 | * | |
ffd1f4ed | 3240 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
3241 | * |
3242 | * Returns a pointer into the buffer or an error code if the | |
3243 | * path was too long. | |
3244 | * | |
be148247 | 3245 | * "buflen" should be positive. |
f2eb6575 | 3246 | * |
02125a82 | 3247 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 3248 | */ |
02125a82 AV |
3249 | char *__d_path(const struct path *path, |
3250 | const struct path *root, | |
f2eb6575 MS |
3251 | char *buf, int buflen) |
3252 | { | |
3253 | char *res = buf + buflen; | |
3254 | int error; | |
3255 | ||
3256 | prepend(&res, &buflen, "\0", 1); | |
f2eb6575 | 3257 | error = prepend_path(path, root, &res, &buflen); |
be148247 | 3258 | |
02125a82 AV |
3259 | if (error < 0) |
3260 | return ERR_PTR(error); | |
3261 | if (error > 0) | |
3262 | return NULL; | |
3263 | return res; | |
3264 | } | |
3265 | ||
3266 | char *d_absolute_path(const struct path *path, | |
3267 | char *buf, int buflen) | |
3268 | { | |
3269 | struct path root = {}; | |
3270 | char *res = buf + buflen; | |
3271 | int error; | |
3272 | ||
3273 | prepend(&res, &buflen, "\0", 1); | |
02125a82 | 3274 | error = prepend_path(path, &root, &res, &buflen); |
02125a82 AV |
3275 | |
3276 | if (error > 1) | |
3277 | error = -EINVAL; | |
3278 | if (error < 0) | |
f2eb6575 | 3279 | return ERR_PTR(error); |
f2eb6575 | 3280 | return res; |
1da177e4 LT |
3281 | } |
3282 | ||
ffd1f4ed MS |
3283 | /* |
3284 | * same as __d_path but appends "(deleted)" for unlinked files. | |
3285 | */ | |
02125a82 AV |
3286 | static int path_with_deleted(const struct path *path, |
3287 | const struct path *root, | |
3288 | char **buf, int *buflen) | |
ffd1f4ed MS |
3289 | { |
3290 | prepend(buf, buflen, "\0", 1); | |
3291 | if (d_unlinked(path->dentry)) { | |
3292 | int error = prepend(buf, buflen, " (deleted)", 10); | |
3293 | if (error) | |
3294 | return error; | |
3295 | } | |
3296 | ||
3297 | return prepend_path(path, root, buf, buflen); | |
3298 | } | |
3299 | ||
8df9d1a4 MS |
3300 | static int prepend_unreachable(char **buffer, int *buflen) |
3301 | { | |
3302 | return prepend(buffer, buflen, "(unreachable)", 13); | |
3303 | } | |
3304 | ||
68f0d9d9 LT |
3305 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) |
3306 | { | |
3307 | unsigned seq; | |
3308 | ||
3309 | do { | |
3310 | seq = read_seqcount_begin(&fs->seq); | |
3311 | *root = fs->root; | |
3312 | } while (read_seqcount_retry(&fs->seq, seq)); | |
3313 | } | |
3314 | ||
a03a8a70 JB |
3315 | /** |
3316 | * d_path - return the path of a dentry | |
cf28b486 | 3317 | * @path: path to report |
a03a8a70 JB |
3318 | * @buf: buffer to return value in |
3319 | * @buflen: buffer length | |
3320 | * | |
3321 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
3322 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
3323 | * | |
52afeefb AV |
3324 | * Returns a pointer into the buffer or an error code if the path was |
3325 | * too long. Note: Callers should use the returned pointer, not the passed | |
3326 | * in buffer, to use the name! The implementation often starts at an offset | |
3327 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 3328 | * |
31f3e0b3 | 3329 | * "buflen" should be positive. |
a03a8a70 | 3330 | */ |
20d4fdc1 | 3331 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 3332 | { |
ffd1f4ed | 3333 | char *res = buf + buflen; |
6ac08c39 | 3334 | struct path root; |
ffd1f4ed | 3335 | int error; |
1da177e4 | 3336 | |
c23fbb6b ED |
3337 | /* |
3338 | * We have various synthetic filesystems that never get mounted. On | |
3339 | * these filesystems dentries are never used for lookup purposes, and | |
3340 | * thus don't need to be hashed. They also don't need a name until a | |
3341 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
3342 | * below allows us to generate a name for these objects on demand: | |
f48cfddc EB |
3343 | * |
3344 | * Some pseudo inodes are mountable. When they are mounted | |
3345 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname | |
3346 | * and instead have d_path return the mounted path. | |
c23fbb6b | 3347 | */ |
f48cfddc EB |
3348 | if (path->dentry->d_op && path->dentry->d_op->d_dname && |
3349 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) | |
cf28b486 | 3350 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
c23fbb6b | 3351 | |
68f0d9d9 LT |
3352 | rcu_read_lock(); |
3353 | get_fs_root_rcu(current->fs, &root); | |
02125a82 | 3354 | error = path_with_deleted(path, &root, &res, &buflen); |
68f0d9d9 LT |
3355 | rcu_read_unlock(); |
3356 | ||
02125a82 | 3357 | if (error < 0) |
ffd1f4ed | 3358 | res = ERR_PTR(error); |
1da177e4 LT |
3359 | return res; |
3360 | } | |
ec4f8605 | 3361 | EXPORT_SYMBOL(d_path); |
1da177e4 | 3362 | |
c23fbb6b ED |
3363 | /* |
3364 | * Helper function for dentry_operations.d_dname() members | |
3365 | */ | |
3366 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
3367 | const char *fmt, ...) | |
3368 | { | |
3369 | va_list args; | |
3370 | char temp[64]; | |
3371 | int sz; | |
3372 | ||
3373 | va_start(args, fmt); | |
3374 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
3375 | va_end(args); | |
3376 | ||
3377 | if (sz > sizeof(temp) || sz > buflen) | |
3378 | return ERR_PTR(-ENAMETOOLONG); | |
3379 | ||
3380 | buffer += buflen - sz; | |
3381 | return memcpy(buffer, temp, sz); | |
3382 | } | |
3383 | ||
118b2302 AV |
3384 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
3385 | { | |
3386 | char *end = buffer + buflen; | |
3387 | /* these dentries are never renamed, so d_lock is not needed */ | |
3388 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
232d2d60 | 3389 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
118b2302 AV |
3390 | prepend(&end, &buflen, "/", 1)) |
3391 | end = ERR_PTR(-ENAMETOOLONG); | |
232d2d60 | 3392 | return end; |
118b2302 | 3393 | } |
31bbe16f | 3394 | EXPORT_SYMBOL(simple_dname); |
118b2302 | 3395 | |
6092d048 RP |
3396 | /* |
3397 | * Write full pathname from the root of the filesystem into the buffer. | |
3398 | */ | |
f6500801 | 3399 | static char *__dentry_path(struct dentry *d, char *buf, int buflen) |
6092d048 | 3400 | { |
f6500801 | 3401 | struct dentry *dentry; |
232d2d60 WL |
3402 | char *end, *retval; |
3403 | int len, seq = 0; | |
3404 | int error = 0; | |
6092d048 | 3405 | |
f6500801 AV |
3406 | if (buflen < 2) |
3407 | goto Elong; | |
3408 | ||
48f5ec21 | 3409 | rcu_read_lock(); |
232d2d60 | 3410 | restart: |
f6500801 | 3411 | dentry = d; |
232d2d60 WL |
3412 | end = buf + buflen; |
3413 | len = buflen; | |
3414 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
3415 | /* Get '/' right */ |
3416 | retval = end-1; | |
3417 | *retval = '/'; | |
232d2d60 | 3418 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
3419 | while (!IS_ROOT(dentry)) { |
3420 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 3421 | |
6092d048 | 3422 | prefetch(parent); |
232d2d60 WL |
3423 | error = prepend_name(&end, &len, &dentry->d_name); |
3424 | if (error) | |
3425 | break; | |
6092d048 RP |
3426 | |
3427 | retval = end; | |
3428 | dentry = parent; | |
3429 | } | |
48f5ec21 AV |
3430 | if (!(seq & 1)) |
3431 | rcu_read_unlock(); | |
3432 | if (need_seqretry(&rename_lock, seq)) { | |
3433 | seq = 1; | |
232d2d60 | 3434 | goto restart; |
48f5ec21 AV |
3435 | } |
3436 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3437 | if (error) |
3438 | goto Elong; | |
c103135c AV |
3439 | return retval; |
3440 | Elong: | |
3441 | return ERR_PTR(-ENAMETOOLONG); | |
3442 | } | |
ec2447c2 NP |
3443 | |
3444 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3445 | { | |
232d2d60 | 3446 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3447 | } |
3448 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3449 | |
3450 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3451 | { | |
3452 | char *p = NULL; | |
3453 | char *retval; | |
3454 | ||
c103135c AV |
3455 | if (d_unlinked(dentry)) { |
3456 | p = buf + buflen; | |
3457 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3458 | goto Elong; | |
3459 | buflen++; | |
3460 | } | |
3461 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3462 | if (!IS_ERR(retval) && p) |
3463 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3464 | return retval; |
3465 | Elong: | |
6092d048 RP |
3466 | return ERR_PTR(-ENAMETOOLONG); |
3467 | } | |
3468 | ||
8b19e341 LT |
3469 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, |
3470 | struct path *pwd) | |
5762482f | 3471 | { |
8b19e341 LT |
3472 | unsigned seq; |
3473 | ||
3474 | do { | |
3475 | seq = read_seqcount_begin(&fs->seq); | |
3476 | *root = fs->root; | |
3477 | *pwd = fs->pwd; | |
3478 | } while (read_seqcount_retry(&fs->seq, seq)); | |
5762482f LT |
3479 | } |
3480 | ||
1da177e4 LT |
3481 | /* |
3482 | * NOTE! The user-level library version returns a | |
3483 | * character pointer. The kernel system call just | |
3484 | * returns the length of the buffer filled (which | |
3485 | * includes the ending '\0' character), or a negative | |
3486 | * error value. So libc would do something like | |
3487 | * | |
3488 | * char *getcwd(char * buf, size_t size) | |
3489 | * { | |
3490 | * int retval; | |
3491 | * | |
3492 | * retval = sys_getcwd(buf, size); | |
3493 | * if (retval >= 0) | |
3494 | * return buf; | |
3495 | * errno = -retval; | |
3496 | * return NULL; | |
3497 | * } | |
3498 | */ | |
3cdad428 | 3499 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3500 | { |
552ce544 | 3501 | int error; |
6ac08c39 | 3502 | struct path pwd, root; |
3272c544 | 3503 | char *page = __getname(); |
1da177e4 LT |
3504 | |
3505 | if (!page) | |
3506 | return -ENOMEM; | |
3507 | ||
8b19e341 LT |
3508 | rcu_read_lock(); |
3509 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); | |
1da177e4 | 3510 | |
552ce544 | 3511 | error = -ENOENT; |
f3da392e | 3512 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3513 | unsigned long len; |
3272c544 LT |
3514 | char *cwd = page + PATH_MAX; |
3515 | int buflen = PATH_MAX; | |
1da177e4 | 3516 | |
8df9d1a4 | 3517 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3518 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
ff812d72 | 3519 | rcu_read_unlock(); |
552ce544 | 3520 | |
02125a82 | 3521 | if (error < 0) |
552ce544 LT |
3522 | goto out; |
3523 | ||
8df9d1a4 | 3524 | /* Unreachable from current root */ |
02125a82 | 3525 | if (error > 0) { |
8df9d1a4 MS |
3526 | error = prepend_unreachable(&cwd, &buflen); |
3527 | if (error) | |
3528 | goto out; | |
3529 | } | |
3530 | ||
552ce544 | 3531 | error = -ERANGE; |
3272c544 | 3532 | len = PATH_MAX + page - cwd; |
552ce544 LT |
3533 | if (len <= size) { |
3534 | error = len; | |
3535 | if (copy_to_user(buf, cwd, len)) | |
3536 | error = -EFAULT; | |
3537 | } | |
949854d0 | 3538 | } else { |
ff812d72 | 3539 | rcu_read_unlock(); |
949854d0 | 3540 | } |
1da177e4 LT |
3541 | |
3542 | out: | |
3272c544 | 3543 | __putname(page); |
1da177e4 LT |
3544 | return error; |
3545 | } | |
3546 | ||
3547 | /* | |
3548 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3549 | * | |
3550 | * Trivially implemented using the dcache structure | |
3551 | */ | |
3552 | ||
3553 | /** | |
3554 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3555 | * @new_dentry: new dentry | |
3556 | * @old_dentry: old dentry | |
3557 | * | |
a6e5787f YB |
3558 | * Returns true if new_dentry is a subdirectory of the parent (at any depth). |
3559 | * Returns false otherwise. | |
1da177e4 LT |
3560 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() |
3561 | */ | |
3562 | ||
a6e5787f | 3563 | bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 | 3564 | { |
a6e5787f | 3565 | bool result; |
949854d0 | 3566 | unsigned seq; |
1da177e4 | 3567 | |
e2761a11 | 3568 | if (new_dentry == old_dentry) |
a6e5787f | 3569 | return true; |
e2761a11 | 3570 | |
e2761a11 | 3571 | do { |
1da177e4 | 3572 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3573 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3574 | /* |
3575 | * Need rcu_readlock to protect against the d_parent trashing | |
3576 | * due to d_move | |
3577 | */ | |
3578 | rcu_read_lock(); | |
e2761a11 | 3579 | if (d_ancestor(old_dentry, new_dentry)) |
a6e5787f | 3580 | result = true; |
e2761a11 | 3581 | else |
a6e5787f | 3582 | result = false; |
949854d0 | 3583 | rcu_read_unlock(); |
1da177e4 | 3584 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3585 | |
3586 | return result; | |
3587 | } | |
3588 | ||
db14fc3a | 3589 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3590 | { |
db14fc3a MS |
3591 | struct dentry *root = data; |
3592 | if (dentry != root) { | |
3593 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3594 | return D_WALK_SKIP; | |
1da177e4 | 3595 | |
01ddc4ed MS |
3596 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3597 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3598 | dentry->d_lockref.count--; | |
3599 | } | |
1da177e4 | 3600 | } |
db14fc3a MS |
3601 | return D_WALK_CONTINUE; |
3602 | } | |
58db63d0 | 3603 | |
db14fc3a MS |
3604 | void d_genocide(struct dentry *parent) |
3605 | { | |
3606 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3607 | } |
3608 | ||
60545d0d | 3609 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3610 | { |
60545d0d AV |
3611 | inode_dec_link_count(inode); |
3612 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
946e51f2 | 3613 | !hlist_unhashed(&dentry->d_u.d_alias) || |
60545d0d AV |
3614 | !d_unlinked(dentry)); |
3615 | spin_lock(&dentry->d_parent->d_lock); | |
3616 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3617 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3618 | (unsigned long long)inode->i_ino); | |
3619 | spin_unlock(&dentry->d_lock); | |
3620 | spin_unlock(&dentry->d_parent->d_lock); | |
3621 | d_instantiate(dentry, inode); | |
1da177e4 | 3622 | } |
60545d0d | 3623 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3624 | |
3625 | static __initdata unsigned long dhash_entries; | |
3626 | static int __init set_dhash_entries(char *str) | |
3627 | { | |
3628 | if (!str) | |
3629 | return 0; | |
3630 | dhash_entries = simple_strtoul(str, &str, 0); | |
3631 | return 1; | |
3632 | } | |
3633 | __setup("dhash_entries=", set_dhash_entries); | |
3634 | ||
3635 | static void __init dcache_init_early(void) | |
3636 | { | |
1da177e4 LT |
3637 | /* If hashes are distributed across NUMA nodes, defer |
3638 | * hash allocation until vmalloc space is available. | |
3639 | */ | |
3640 | if (hashdist) | |
3641 | return; | |
3642 | ||
3643 | dentry_hashtable = | |
3644 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3645 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3646 | dhash_entries, |
3647 | 13, | |
3d375d78 | 3648 | HASH_EARLY | HASH_ZERO, |
1da177e4 LT |
3649 | &d_hash_shift, |
3650 | &d_hash_mask, | |
31fe62b9 | 3651 | 0, |
1da177e4 | 3652 | 0); |
1da177e4 LT |
3653 | } |
3654 | ||
74bf17cf | 3655 | static void __init dcache_init(void) |
1da177e4 | 3656 | { |
3d375d78 | 3657 | /* |
1da177e4 LT |
3658 | * A constructor could be added for stable state like the lists, |
3659 | * but it is probably not worth it because of the cache nature | |
3d375d78 | 3660 | * of the dcache. |
1da177e4 | 3661 | */ |
0a31bd5f | 3662 | dentry_cache = KMEM_CACHE(dentry, |
5d097056 | 3663 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT); |
1da177e4 LT |
3664 | |
3665 | /* Hash may have been set up in dcache_init_early */ | |
3666 | if (!hashdist) | |
3667 | return; | |
3668 | ||
3669 | dentry_hashtable = | |
3670 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3671 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3672 | dhash_entries, |
3673 | 13, | |
3d375d78 | 3674 | HASH_ZERO, |
1da177e4 LT |
3675 | &d_hash_shift, |
3676 | &d_hash_mask, | |
31fe62b9 | 3677 | 0, |
1da177e4 | 3678 | 0); |
1da177e4 LT |
3679 | } |
3680 | ||
3681 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3682 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3683 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3684 | |
1da177e4 LT |
3685 | EXPORT_SYMBOL(d_genocide); |
3686 | ||
1da177e4 LT |
3687 | void __init vfs_caches_init_early(void) |
3688 | { | |
6916363f SAS |
3689 | int i; |
3690 | ||
3691 | for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++) | |
3692 | INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]); | |
3693 | ||
1da177e4 LT |
3694 | dcache_init_early(); |
3695 | inode_init_early(); | |
3696 | } | |
3697 | ||
4248b0da | 3698 | void __init vfs_caches_init(void) |
1da177e4 | 3699 | { |
1da177e4 | 3700 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, |
20c2df83 | 3701 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3702 | |
74bf17cf DC |
3703 | dcache_init(); |
3704 | inode_init(); | |
4248b0da MG |
3705 | files_init(); |
3706 | files_maxfiles_init(); | |
74bf17cf | 3707 | mnt_init(); |
1da177e4 LT |
3708 | bdev_cache_init(); |
3709 | chrdev_init(); | |
3710 | } |