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 17 #include <linux/syscalls.h> 18 #include <linux/string.h> 19 #include <linux/mm.h> 20 #include <linux/fs.h> 21 #include <linux/fsnotify.h> 22 #include <linux/slab.h> 23 #include <linux/init.h> 24 #include <linux/hash.h> 25 #include <linux/cache.h> 26 #include <linux/export.h> 27 #include <linux/mount.h> 28 #include <linux/file.h> 29 #include <asm/uaccess.h> 30 #include <linux/security.h> 31 #include <linux/seqlock.h> 32 #include <linux/swap.h> 33 #include <linux/bootmem.h> 34 #include <linux/fs_struct.h> 35 #include <linux/hardirq.h> 36 #include <linux/bit_spinlock.h> 37 #include <linux/rculist_bl.h> 38 #include <linux/prefetch.h> 39 #include <linux/ratelimit.h> 40 #include <linux/list_lru.h> 41 #include "internal.h" 42 #include "mount.h" 43 44 /* 45 * Usage: 46 * dcache->d_inode->i_lock protects: 47 * - i_dentry, d_alias, d_inode of aliases 48 * dcache_hash_bucket lock protects: 49 * - the dcache hash table 50 * s_anon bl list spinlock protects: 51 * - the s_anon list (see __d_drop) 52 * dentry->d_sb->s_dentry_lru_lock protects: 53 * - the dcache lru lists and counters 54 * d_lock protects: 55 * - d_flags 56 * - d_name 57 * - d_lru 58 * - d_count 59 * - d_unhashed() 60 * - d_parent and d_subdirs 61 * - childrens' d_child and d_parent 62 * - d_alias, d_inode 63 * 64 * Ordering: 65 * dentry->d_inode->i_lock 66 * dentry->d_lock 67 * dentry->d_sb->s_dentry_lru_lock 68 * dcache_hash_bucket lock 69 * s_anon lock 70 * 71 * If there is an ancestor relationship: 72 * dentry->d_parent->...->d_parent->d_lock 73 * ... 74 * dentry->d_parent->d_lock 75 * dentry->d_lock 76 * 77 * If no ancestor relationship: 78 * if (dentry1 < dentry2) 79 * dentry1->d_lock 80 * dentry2->d_lock 81 */ 82 int sysctl_vfs_cache_pressure __read_mostly = 100; 83 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); 84 85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); 86 87 EXPORT_SYMBOL(rename_lock); 88 89 static struct kmem_cache *dentry_cache __read_mostly; 90 91 /* 92 * This is the single most critical data structure when it comes 93 * to the dcache: the hashtable for lookups. Somebody should try 94 * to make this good - I've just made it work. 95 * 96 * This hash-function tries to avoid losing too many bits of hash 97 * information, yet avoid using a prime hash-size or similar. 98 */ 99 100 static unsigned int d_hash_mask __read_mostly; 101 static unsigned int d_hash_shift __read_mostly; 102 103 static struct hlist_bl_head *dentry_hashtable __read_mostly; 104 105 static inline struct hlist_bl_head *d_hash(const struct dentry *parent, 106 unsigned int hash) 107 { 108 hash += (unsigned long) parent / L1_CACHE_BYTES; 109 return dentry_hashtable + hash_32(hash, d_hash_shift); 110 } 111 112 /* Statistics gathering. */ 113 struct dentry_stat_t dentry_stat = { 114 .age_limit = 45, 115 }; 116 117 static DEFINE_PER_CPU(long, nr_dentry); 118 static DEFINE_PER_CPU(long, nr_dentry_unused); 119 120 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) 121 122 /* 123 * Here we resort to our own counters instead of using generic per-cpu counters 124 * for consistency with what the vfs inode code does. We are expected to harvest 125 * better code and performance by having our own specialized counters. 126 * 127 * Please note that the loop is done over all possible CPUs, not over all online 128 * CPUs. The reason for this is that we don't want to play games with CPUs going 129 * on and off. If one of them goes off, we will just keep their counters. 130 * 131 * glommer: See cffbc8a for details, and if you ever intend to change this, 132 * please update all vfs counters to match. 133 */ 134 static long get_nr_dentry(void) 135 { 136 int i; 137 long sum = 0; 138 for_each_possible_cpu(i) 139 sum += per_cpu(nr_dentry, i); 140 return sum < 0 ? 0 : sum; 141 } 142 143 static long get_nr_dentry_unused(void) 144 { 145 int i; 146 long sum = 0; 147 for_each_possible_cpu(i) 148 sum += per_cpu(nr_dentry_unused, i); 149 return sum < 0 ? 0 : sum; 150 } 151 152 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer, 153 size_t *lenp, loff_t *ppos) 154 { 155 dentry_stat.nr_dentry = get_nr_dentry(); 156 dentry_stat.nr_unused = get_nr_dentry_unused(); 157 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); 158 } 159 #endif 160 161 /* 162 * Compare 2 name strings, return 0 if they match, otherwise non-zero. 163 * The strings are both count bytes long, and count is non-zero. 164 */ 165 #ifdef CONFIG_DCACHE_WORD_ACCESS 166 167 #include <asm/word-at-a-time.h> 168 /* 169 * NOTE! 'cs' and 'scount' come from a dentry, so it has a 170 * aligned allocation for this particular component. We don't 171 * strictly need the load_unaligned_zeropad() safety, but it 172 * doesn't hurt either. 173 * 174 * In contrast, 'ct' and 'tcount' can be from a pathname, and do 175 * need the careful unaligned handling. 176 */ 177 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) 178 { 179 unsigned long a,b,mask; 180 181 for (;;) { 182 a = *(unsigned long *)cs; 183 b = load_unaligned_zeropad(ct); 184 if (tcount < sizeof(unsigned long)) 185 break; 186 if (unlikely(a != b)) 187 return 1; 188 cs += sizeof(unsigned long); 189 ct += sizeof(unsigned long); 190 tcount -= sizeof(unsigned long); 191 if (!tcount) 192 return 0; 193 } 194 mask = bytemask_from_count(tcount); 195 return unlikely(!!((a ^ b) & mask)); 196 } 197 198 #else 199 200 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) 201 { 202 do { 203 if (*cs != *ct) 204 return 1; 205 cs++; 206 ct++; 207 tcount--; 208 } while (tcount); 209 return 0; 210 } 211 212 #endif 213 214 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) 215 { 216 const unsigned char *cs; 217 /* 218 * Be careful about RCU walk racing with rename: 219 * use ACCESS_ONCE to fetch the name pointer. 220 * 221 * NOTE! Even if a rename will mean that the length 222 * was not loaded atomically, we don't care. The 223 * RCU walk will check the sequence count eventually, 224 * and catch it. And we won't overrun the buffer, 225 * because we're reading the name pointer atomically, 226 * and a dentry name is guaranteed to be properly 227 * terminated with a NUL byte. 228 * 229 * End result: even if 'len' is wrong, we'll exit 230 * early because the data cannot match (there can 231 * be no NUL in the ct/tcount data) 232 */ 233 cs = ACCESS_ONCE(dentry->d_name.name); 234 smp_read_barrier_depends(); 235 return dentry_string_cmp(cs, ct, tcount); 236 } 237 238 struct external_name { 239 union { 240 atomic_t count; 241 struct rcu_head head; 242 } u; 243 unsigned char name[]; 244 }; 245 246 static inline struct external_name *external_name(struct dentry *dentry) 247 { 248 return container_of(dentry->d_name.name, struct external_name, name[0]); 249 } 250 251 static void __d_free(struct rcu_head *head) 252 { 253 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); 254 255 WARN_ON(!hlist_unhashed(&dentry->d_alias)); 256 kmem_cache_free(dentry_cache, dentry); 257 } 258 259 static void __d_free_external(struct rcu_head *head) 260 { 261 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); 262 WARN_ON(!hlist_unhashed(&dentry->d_alias)); 263 kfree(external_name(dentry)); 264 kmem_cache_free(dentry_cache, dentry); 265 } 266 267 static inline int dname_external(const struct dentry *dentry) 268 { 269 return dentry->d_name.name != dentry->d_iname; 270 } 271 272 static void dentry_free(struct dentry *dentry) 273 { 274 if (unlikely(dname_external(dentry))) { 275 struct external_name *p = external_name(dentry); 276 if (likely(atomic_dec_and_test(&p->u.count))) { 277 call_rcu(&dentry->d_u.d_rcu, __d_free_external); 278 return; 279 } 280 } 281 /* if dentry was never visible to RCU, immediate free is OK */ 282 if (!(dentry->d_flags & DCACHE_RCUACCESS)) 283 __d_free(&dentry->d_u.d_rcu); 284 else 285 call_rcu(&dentry->d_u.d_rcu, __d_free); 286 } 287 288 /** 289 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups 290 * @dentry: the target dentry 291 * After this call, in-progress rcu-walk path lookup will fail. This 292 * should be called after unhashing, and after changing d_inode (if 293 * the dentry has not already been unhashed). 294 */ 295 static inline void dentry_rcuwalk_barrier(struct dentry *dentry) 296 { 297 assert_spin_locked(&dentry->d_lock); 298 /* Go through a barrier */ 299 write_seqcount_barrier(&dentry->d_seq); 300 } 301 302 /* 303 * Release the dentry's inode, using the filesystem 304 * d_iput() operation if defined. Dentry has no refcount 305 * and is unhashed. 306 */ 307 static void dentry_iput(struct dentry * dentry) 308 __releases(dentry->d_lock) 309 __releases(dentry->d_inode->i_lock) 310 { 311 struct inode *inode = dentry->d_inode; 312 if (inode) { 313 dentry->d_inode = NULL; 314 hlist_del_init(&dentry->d_alias); 315 spin_unlock(&dentry->d_lock); 316 spin_unlock(&inode->i_lock); 317 if (!inode->i_nlink) 318 fsnotify_inoderemove(inode); 319 if (dentry->d_op && dentry->d_op->d_iput) 320 dentry->d_op->d_iput(dentry, inode); 321 else 322 iput(inode); 323 } else { 324 spin_unlock(&dentry->d_lock); 325 } 326 } 327 328 /* 329 * Release the dentry's inode, using the filesystem 330 * d_iput() operation if defined. dentry remains in-use. 331 */ 332 static void dentry_unlink_inode(struct dentry * dentry) 333 __releases(dentry->d_lock) 334 __releases(dentry->d_inode->i_lock) 335 { 336 struct inode *inode = dentry->d_inode; 337 __d_clear_type(dentry); 338 dentry->d_inode = NULL; 339 hlist_del_init(&dentry->d_alias); 340 dentry_rcuwalk_barrier(dentry); 341 spin_unlock(&dentry->d_lock); 342 spin_unlock(&inode->i_lock); 343 if (!inode->i_nlink) 344 fsnotify_inoderemove(inode); 345 if (dentry->d_op && dentry->d_op->d_iput) 346 dentry->d_op->d_iput(dentry, inode); 347 else 348 iput(inode); 349 } 350 351 /* 352 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry 353 * is in use - which includes both the "real" per-superblock 354 * LRU list _and_ the DCACHE_SHRINK_LIST use. 355 * 356 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is 357 * on the shrink list (ie not on the superblock LRU list). 358 * 359 * The per-cpu "nr_dentry_unused" counters are updated with 360 * the DCACHE_LRU_LIST bit. 361 * 362 * These helper functions make sure we always follow the 363 * rules. d_lock must be held by the caller. 364 */ 365 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) 366 static void d_lru_add(struct dentry *dentry) 367 { 368 D_FLAG_VERIFY(dentry, 0); 369 dentry->d_flags |= DCACHE_LRU_LIST; 370 this_cpu_inc(nr_dentry_unused); 371 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); 372 } 373 374 static void d_lru_del(struct dentry *dentry) 375 { 376 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); 377 dentry->d_flags &= ~DCACHE_LRU_LIST; 378 this_cpu_dec(nr_dentry_unused); 379 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); 380 } 381 382 static void d_shrink_del(struct dentry *dentry) 383 { 384 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); 385 list_del_init(&dentry->d_lru); 386 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); 387 this_cpu_dec(nr_dentry_unused); 388 } 389 390 static void d_shrink_add(struct dentry *dentry, struct list_head *list) 391 { 392 D_FLAG_VERIFY(dentry, 0); 393 list_add(&dentry->d_lru, list); 394 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; 395 this_cpu_inc(nr_dentry_unused); 396 } 397 398 /* 399 * These can only be called under the global LRU lock, ie during the 400 * callback for freeing the LRU list. "isolate" removes it from the 401 * LRU lists entirely, while shrink_move moves it to the indicated 402 * private list. 403 */ 404 static void d_lru_isolate(struct dentry *dentry) 405 { 406 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); 407 dentry->d_flags &= ~DCACHE_LRU_LIST; 408 this_cpu_dec(nr_dentry_unused); 409 list_del_init(&dentry->d_lru); 410 } 411 412 static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list) 413 { 414 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); 415 dentry->d_flags |= DCACHE_SHRINK_LIST; 416 list_move_tail(&dentry->d_lru, list); 417 } 418 419 /* 420 * dentry_lru_(add|del)_list) must be called with d_lock held. 421 */ 422 static void dentry_lru_add(struct dentry *dentry) 423 { 424 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) 425 d_lru_add(dentry); 426 } 427 428 /** 429 * d_drop - drop a dentry 430 * @dentry: dentry to drop 431 * 432 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't 433 * be found through a VFS lookup any more. Note that this is different from 434 * deleting the dentry - d_delete will try to mark the dentry negative if 435 * possible, giving a successful _negative_ lookup, while d_drop will 436 * just make the cache lookup fail. 437 * 438 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some 439 * reason (NFS timeouts or autofs deletes). 440 * 441 * __d_drop requires dentry->d_lock. 442 */ 443 void __d_drop(struct dentry *dentry) 444 { 445 if (!d_unhashed(dentry)) { 446 struct hlist_bl_head *b; 447 /* 448 * Hashed dentries are normally on the dentry hashtable, 449 * with the exception of those newly allocated by 450 * d_obtain_alias, which are always IS_ROOT: 451 */ 452 if (unlikely(IS_ROOT(dentry))) 453 b = &dentry->d_sb->s_anon; 454 else 455 b = d_hash(dentry->d_parent, dentry->d_name.hash); 456 457 hlist_bl_lock(b); 458 __hlist_bl_del(&dentry->d_hash); 459 dentry->d_hash.pprev = NULL; 460 hlist_bl_unlock(b); 461 dentry_rcuwalk_barrier(dentry); 462 } 463 } 464 EXPORT_SYMBOL(__d_drop); 465 466 void d_drop(struct dentry *dentry) 467 { 468 spin_lock(&dentry->d_lock); 469 __d_drop(dentry); 470 spin_unlock(&dentry->d_lock); 471 } 472 EXPORT_SYMBOL(d_drop); 473 474 static void __dentry_kill(struct dentry *dentry) 475 { 476 struct dentry *parent = NULL; 477 bool can_free = true; 478 if (!IS_ROOT(dentry)) 479 parent = dentry->d_parent; 480 481 /* 482 * The dentry is now unrecoverably dead to the world. 483 */ 484 lockref_mark_dead(&dentry->d_lockref); 485 486 /* 487 * inform the fs via d_prune that this dentry is about to be 488 * unhashed and destroyed. 489 */ 490 if (dentry->d_flags & DCACHE_OP_PRUNE) 491 dentry->d_op->d_prune(dentry); 492 493 if (dentry->d_flags & DCACHE_LRU_LIST) { 494 if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) 495 d_lru_del(dentry); 496 } 497 /* if it was on the hash then remove it */ 498 __d_drop(dentry); 499 list_del(&dentry->d_u.d_child); 500 /* 501 * Inform d_walk() that we are no longer attached to the 502 * dentry tree 503 */ 504 dentry->d_flags |= DCACHE_DENTRY_KILLED; 505 if (parent) 506 spin_unlock(&parent->d_lock); 507 dentry_iput(dentry); 508 /* 509 * dentry_iput drops the locks, at which point nobody (except 510 * transient RCU lookups) can reach this dentry. 511 */ 512 BUG_ON((int)dentry->d_lockref.count > 0); 513 this_cpu_dec(nr_dentry); 514 if (dentry->d_op && dentry->d_op->d_release) 515 dentry->d_op->d_release(dentry); 516 517 spin_lock(&dentry->d_lock); 518 if (dentry->d_flags & DCACHE_SHRINK_LIST) { 519 dentry->d_flags |= DCACHE_MAY_FREE; 520 can_free = false; 521 } 522 spin_unlock(&dentry->d_lock); 523 if (likely(can_free)) 524 dentry_free(dentry); 525 } 526 527 /* 528 * Finish off a dentry we've decided to kill. 529 * dentry->d_lock must be held, returns with it unlocked. 530 * If ref is non-zero, then decrement the refcount too. 531 * Returns dentry requiring refcount drop, or NULL if we're done. 532 */ 533 static struct dentry *dentry_kill(struct dentry *dentry) 534 __releases(dentry->d_lock) 535 { 536 struct inode *inode = dentry->d_inode; 537 struct dentry *parent = NULL; 538 539 if (inode && unlikely(!spin_trylock(&inode->i_lock))) 540 goto failed; 541 542 if (!IS_ROOT(dentry)) { 543 parent = dentry->d_parent; 544 if (unlikely(!spin_trylock(&parent->d_lock))) { 545 if (inode) 546 spin_unlock(&inode->i_lock); 547 goto failed; 548 } 549 } 550 551 __dentry_kill(dentry); 552 return parent; 553 554 failed: 555 spin_unlock(&dentry->d_lock); 556 cpu_relax(); 557 return dentry; /* try again with same dentry */ 558 } 559 560 static inline struct dentry *lock_parent(struct dentry *dentry) 561 { 562 struct dentry *parent = dentry->d_parent; 563 if (IS_ROOT(dentry)) 564 return NULL; 565 if (unlikely((int)dentry->d_lockref.count < 0)) 566 return NULL; 567 if (likely(spin_trylock(&parent->d_lock))) 568 return parent; 569 rcu_read_lock(); 570 spin_unlock(&dentry->d_lock); 571 again: 572 parent = ACCESS_ONCE(dentry->d_parent); 573 spin_lock(&parent->d_lock); 574 /* 575 * We can't blindly lock dentry until we are sure 576 * that we won't violate the locking order. 577 * Any changes of dentry->d_parent must have 578 * been done with parent->d_lock held, so 579 * spin_lock() above is enough of a barrier 580 * for checking if it's still our child. 581 */ 582 if (unlikely(parent != dentry->d_parent)) { 583 spin_unlock(&parent->d_lock); 584 goto again; 585 } 586 rcu_read_unlock(); 587 if (parent != dentry) 588 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 589 else 590 parent = NULL; 591 return parent; 592 } 593 594 /* 595 * This is dput 596 * 597 * This is complicated by the fact that we do not want to put 598 * dentries that are no longer on any hash chain on the unused 599 * list: we'd much rather just get rid of them immediately. 600 * 601 * However, that implies that we have to traverse the dentry 602 * tree upwards to the parents which might _also_ now be 603 * scheduled for deletion (it may have been only waiting for 604 * its last child to go away). 605 * 606 * This tail recursion is done by hand as we don't want to depend 607 * on the compiler to always get this right (gcc generally doesn't). 608 * Real recursion would eat up our stack space. 609 */ 610 611 /* 612 * dput - release a dentry 613 * @dentry: dentry to release 614 * 615 * Release a dentry. This will drop the usage count and if appropriate 616 * call the dentry unlink method as well as removing it from the queues and 617 * releasing its resources. If the parent dentries were scheduled for release 618 * they too may now get deleted. 619 */ 620 void dput(struct dentry *dentry) 621 { 622 if (unlikely(!dentry)) 623 return; 624 625 repeat: 626 if (lockref_put_or_lock(&dentry->d_lockref)) 627 return; 628 629 /* Unreachable? Get rid of it */ 630 if (unlikely(d_unhashed(dentry))) 631 goto kill_it; 632 633 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { 634 if (dentry->d_op->d_delete(dentry)) 635 goto kill_it; 636 } 637 638 if (!(dentry->d_flags & DCACHE_REFERENCED)) 639 dentry->d_flags |= DCACHE_REFERENCED; 640 dentry_lru_add(dentry); 641 642 dentry->d_lockref.count--; 643 spin_unlock(&dentry->d_lock); 644 return; 645 646 kill_it: 647 dentry = dentry_kill(dentry); 648 if (dentry) 649 goto repeat; 650 } 651 EXPORT_SYMBOL(dput); 652 653 654 /* This must be called with d_lock held */ 655 static inline void __dget_dlock(struct dentry *dentry) 656 { 657 dentry->d_lockref.count++; 658 } 659 660 static inline void __dget(struct dentry *dentry) 661 { 662 lockref_get(&dentry->d_lockref); 663 } 664 665 struct dentry *dget_parent(struct dentry *dentry) 666 { 667 int gotref; 668 struct dentry *ret; 669 670 /* 671 * Do optimistic parent lookup without any 672 * locking. 673 */ 674 rcu_read_lock(); 675 ret = ACCESS_ONCE(dentry->d_parent); 676 gotref = lockref_get_not_zero(&ret->d_lockref); 677 rcu_read_unlock(); 678 if (likely(gotref)) { 679 if (likely(ret == ACCESS_ONCE(dentry->d_parent))) 680 return ret; 681 dput(ret); 682 } 683 684 repeat: 685 /* 686 * Don't need rcu_dereference because we re-check it was correct under 687 * the lock. 688 */ 689 rcu_read_lock(); 690 ret = dentry->d_parent; 691 spin_lock(&ret->d_lock); 692 if (unlikely(ret != dentry->d_parent)) { 693 spin_unlock(&ret->d_lock); 694 rcu_read_unlock(); 695 goto repeat; 696 } 697 rcu_read_unlock(); 698 BUG_ON(!ret->d_lockref.count); 699 ret->d_lockref.count++; 700 spin_unlock(&ret->d_lock); 701 return ret; 702 } 703 EXPORT_SYMBOL(dget_parent); 704 705 /** 706 * d_find_alias - grab a hashed alias of inode 707 * @inode: inode in question 708 * 709 * If inode has a hashed alias, or is a directory and has any alias, 710 * acquire the reference to alias and return it. Otherwise return NULL. 711 * Notice that if inode is a directory there can be only one alias and 712 * it can be unhashed only if it has no children, or if it is the root 713 * of a filesystem, or if the directory was renamed and d_revalidate 714 * was the first vfs operation to notice. 715 * 716 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer 717 * any other hashed alias over that one. 718 */ 719 static struct dentry *__d_find_alias(struct inode *inode) 720 { 721 struct dentry *alias, *discon_alias; 722 723 again: 724 discon_alias = NULL; 725 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { 726 spin_lock(&alias->d_lock); 727 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { 728 if (IS_ROOT(alias) && 729 (alias->d_flags & DCACHE_DISCONNECTED)) { 730 discon_alias = alias; 731 } else { 732 __dget_dlock(alias); 733 spin_unlock(&alias->d_lock); 734 return alias; 735 } 736 } 737 spin_unlock(&alias->d_lock); 738 } 739 if (discon_alias) { 740 alias = discon_alias; 741 spin_lock(&alias->d_lock); 742 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { 743 __dget_dlock(alias); 744 spin_unlock(&alias->d_lock); 745 return alias; 746 } 747 spin_unlock(&alias->d_lock); 748 goto again; 749 } 750 return NULL; 751 } 752 753 struct dentry *d_find_alias(struct inode *inode) 754 { 755 struct dentry *de = NULL; 756 757 if (!hlist_empty(&inode->i_dentry)) { 758 spin_lock(&inode->i_lock); 759 de = __d_find_alias(inode); 760 spin_unlock(&inode->i_lock); 761 } 762 return de; 763 } 764 EXPORT_SYMBOL(d_find_alias); 765 766 /* 767 * Try to kill dentries associated with this inode. 768 * WARNING: you must own a reference to inode. 769 */ 770 void d_prune_aliases(struct inode *inode) 771 { 772 struct dentry *dentry; 773 restart: 774 spin_lock(&inode->i_lock); 775 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { 776 spin_lock(&dentry->d_lock); 777 if (!dentry->d_lockref.count) { 778 struct dentry *parent = lock_parent(dentry); 779 if (likely(!dentry->d_lockref.count)) { 780 __dentry_kill(dentry); 781 goto restart; 782 } 783 if (parent) 784 spin_unlock(&parent->d_lock); 785 } 786 spin_unlock(&dentry->d_lock); 787 } 788 spin_unlock(&inode->i_lock); 789 } 790 EXPORT_SYMBOL(d_prune_aliases); 791 792 static void shrink_dentry_list(struct list_head *list) 793 { 794 struct dentry *dentry, *parent; 795 796 while (!list_empty(list)) { 797 struct inode *inode; 798 dentry = list_entry(list->prev, struct dentry, d_lru); 799 spin_lock(&dentry->d_lock); 800 parent = lock_parent(dentry); 801 802 /* 803 * The dispose list is isolated and dentries are not accounted 804 * to the LRU here, so we can simply remove it from the list 805 * here regardless of whether it is referenced or not. 806 */ 807 d_shrink_del(dentry); 808 809 /* 810 * We found an inuse dentry which was not removed from 811 * the LRU because of laziness during lookup. Do not free it. 812 */ 813 if ((int)dentry->d_lockref.count > 0) { 814 spin_unlock(&dentry->d_lock); 815 if (parent) 816 spin_unlock(&parent->d_lock); 817 continue; 818 } 819 820 821 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) { 822 bool can_free = dentry->d_flags & DCACHE_MAY_FREE; 823 spin_unlock(&dentry->d_lock); 824 if (parent) 825 spin_unlock(&parent->d_lock); 826 if (can_free) 827 dentry_free(dentry); 828 continue; 829 } 830 831 inode = dentry->d_inode; 832 if (inode && unlikely(!spin_trylock(&inode->i_lock))) { 833 d_shrink_add(dentry, list); 834 spin_unlock(&dentry->d_lock); 835 if (parent) 836 spin_unlock(&parent->d_lock); 837 continue; 838 } 839 840 __dentry_kill(dentry); 841 842 /* 843 * We need to prune ancestors too. This is necessary to prevent 844 * quadratic behavior of shrink_dcache_parent(), but is also 845 * expected to be beneficial in reducing dentry cache 846 * fragmentation. 847 */ 848 dentry = parent; 849 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) { 850 parent = lock_parent(dentry); 851 if (dentry->d_lockref.count != 1) { 852 dentry->d_lockref.count--; 853 spin_unlock(&dentry->d_lock); 854 if (parent) 855 spin_unlock(&parent->d_lock); 856 break; 857 } 858 inode = dentry->d_inode; /* can't be NULL */ 859 if (unlikely(!spin_trylock(&inode->i_lock))) { 860 spin_unlock(&dentry->d_lock); 861 if (parent) 862 spin_unlock(&parent->d_lock); 863 cpu_relax(); 864 continue; 865 } 866 __dentry_kill(dentry); 867 dentry = parent; 868 } 869 } 870 } 871 872 static enum lru_status 873 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg) 874 { 875 struct list_head *freeable = arg; 876 struct dentry *dentry = container_of(item, struct dentry, d_lru); 877 878 879 /* 880 * we are inverting the lru lock/dentry->d_lock here, 881 * so use a trylock. If we fail to get the lock, just skip 882 * it 883 */ 884 if (!spin_trylock(&dentry->d_lock)) 885 return LRU_SKIP; 886 887 /* 888 * Referenced dentries are still in use. If they have active 889 * counts, just remove them from the LRU. Otherwise give them 890 * another pass through the LRU. 891 */ 892 if (dentry->d_lockref.count) { 893 d_lru_isolate(dentry); 894 spin_unlock(&dentry->d_lock); 895 return LRU_REMOVED; 896 } 897 898 if (dentry->d_flags & DCACHE_REFERENCED) { 899 dentry->d_flags &= ~DCACHE_REFERENCED; 900 spin_unlock(&dentry->d_lock); 901 902 /* 903 * The list move itself will be made by the common LRU code. At 904 * this point, we've dropped the dentry->d_lock but keep the 905 * lru lock. This is safe to do, since every list movement is 906 * protected by the lru lock even if both locks are held. 907 * 908 * This is guaranteed by the fact that all LRU management 909 * functions are intermediated by the LRU API calls like 910 * list_lru_add and list_lru_del. List movement in this file 911 * only ever occur through this functions or through callbacks 912 * like this one, that are called from the LRU API. 913 * 914 * The only exceptions to this are functions like 915 * shrink_dentry_list, and code that first checks for the 916 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be 917 * operating only with stack provided lists after they are 918 * properly isolated from the main list. It is thus, always a 919 * local access. 920 */ 921 return LRU_ROTATE; 922 } 923 924 d_lru_shrink_move(dentry, freeable); 925 spin_unlock(&dentry->d_lock); 926 927 return LRU_REMOVED; 928 } 929 930 /** 931 * prune_dcache_sb - shrink the dcache 932 * @sb: superblock 933 * @nr_to_scan : number of entries to try to free 934 * @nid: which node to scan for freeable entities 935 * 936 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is 937 * done when we need more memory an called from the superblock shrinker 938 * function. 939 * 940 * This function may fail to free any resources if all the dentries are in 941 * use. 942 */ 943 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan, 944 int nid) 945 { 946 LIST_HEAD(dispose); 947 long freed; 948 949 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate, 950 &dispose, &nr_to_scan); 951 shrink_dentry_list(&dispose); 952 return freed; 953 } 954 955 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, 956 spinlock_t *lru_lock, void *arg) 957 { 958 struct list_head *freeable = arg; 959 struct dentry *dentry = container_of(item, struct dentry, d_lru); 960 961 /* 962 * we are inverting the lru lock/dentry->d_lock here, 963 * so use a trylock. If we fail to get the lock, just skip 964 * it 965 */ 966 if (!spin_trylock(&dentry->d_lock)) 967 return LRU_SKIP; 968 969 d_lru_shrink_move(dentry, freeable); 970 spin_unlock(&dentry->d_lock); 971 972 return LRU_REMOVED; 973 } 974 975 976 /** 977 * shrink_dcache_sb - shrink dcache for a superblock 978 * @sb: superblock 979 * 980 * Shrink the dcache for the specified super block. This is used to free 981 * the dcache before unmounting a file system. 982 */ 983 void shrink_dcache_sb(struct super_block *sb) 984 { 985 long freed; 986 987 do { 988 LIST_HEAD(dispose); 989 990 freed = list_lru_walk(&sb->s_dentry_lru, 991 dentry_lru_isolate_shrink, &dispose, UINT_MAX); 992 993 this_cpu_sub(nr_dentry_unused, freed); 994 shrink_dentry_list(&dispose); 995 } while (freed > 0); 996 } 997 EXPORT_SYMBOL(shrink_dcache_sb); 998 999 /** 1000 * enum d_walk_ret - action to talke during tree walk 1001 * @D_WALK_CONTINUE: contrinue walk 1002 * @D_WALK_QUIT: quit walk 1003 * @D_WALK_NORETRY: quit when retry is needed 1004 * @D_WALK_SKIP: skip this dentry and its children 1005 */ 1006 enum d_walk_ret { 1007 D_WALK_CONTINUE, 1008 D_WALK_QUIT, 1009 D_WALK_NORETRY, 1010 D_WALK_SKIP, 1011 }; 1012 1013 /** 1014 * d_walk - walk the dentry tree 1015 * @parent: start of walk 1016 * @data: data passed to @enter() and @finish() 1017 * @enter: callback when first entering the dentry 1018 * @finish: callback when successfully finished the walk 1019 * 1020 * The @enter() and @finish() callbacks are called with d_lock held. 1021 */ 1022 static void d_walk(struct dentry *parent, void *data, 1023 enum d_walk_ret (*enter)(void *, struct dentry *), 1024 void (*finish)(void *)) 1025 { 1026 struct dentry *this_parent; 1027 struct list_head *next; 1028 unsigned seq = 0; 1029 enum d_walk_ret ret; 1030 bool retry = true; 1031 1032 again: 1033 read_seqbegin_or_lock(&rename_lock, &seq); 1034 this_parent = parent; 1035 spin_lock(&this_parent->d_lock); 1036 1037 ret = enter(data, this_parent); 1038 switch (ret) { 1039 case D_WALK_CONTINUE: 1040 break; 1041 case D_WALK_QUIT: 1042 case D_WALK_SKIP: 1043 goto out_unlock; 1044 case D_WALK_NORETRY: 1045 retry = false; 1046 break; 1047 } 1048 repeat: 1049 next = this_parent->d_subdirs.next; 1050 resume: 1051 while (next != &this_parent->d_subdirs) { 1052 struct list_head *tmp = next; 1053 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); 1054 next = tmp->next; 1055 1056 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 1057 1058 ret = enter(data, dentry); 1059 switch (ret) { 1060 case D_WALK_CONTINUE: 1061 break; 1062 case D_WALK_QUIT: 1063 spin_unlock(&dentry->d_lock); 1064 goto out_unlock; 1065 case D_WALK_NORETRY: 1066 retry = false; 1067 break; 1068 case D_WALK_SKIP: 1069 spin_unlock(&dentry->d_lock); 1070 continue; 1071 } 1072 1073 if (!list_empty(&dentry->d_subdirs)) { 1074 spin_unlock(&this_parent->d_lock); 1075 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); 1076 this_parent = dentry; 1077 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); 1078 goto repeat; 1079 } 1080 spin_unlock(&dentry->d_lock); 1081 } 1082 /* 1083 * All done at this level ... ascend and resume the search. 1084 */ 1085 if (this_parent != parent) { 1086 struct dentry *child = this_parent; 1087 this_parent = child->d_parent; 1088 1089 rcu_read_lock(); 1090 spin_unlock(&child->d_lock); 1091 spin_lock(&this_parent->d_lock); 1092 1093 /* 1094 * might go back up the wrong parent if we have had a rename 1095 * or deletion 1096 */ 1097 if (this_parent != child->d_parent || 1098 (child->d_flags & DCACHE_DENTRY_KILLED) || 1099 need_seqretry(&rename_lock, seq)) { 1100 spin_unlock(&this_parent->d_lock); 1101 rcu_read_unlock(); 1102 goto rename_retry; 1103 } 1104 rcu_read_unlock(); 1105 next = child->d_u.d_child.next; 1106 goto resume; 1107 } 1108 if (need_seqretry(&rename_lock, seq)) { 1109 spin_unlock(&this_parent->d_lock); 1110 goto rename_retry; 1111 } 1112 if (finish) 1113 finish(data); 1114 1115 out_unlock: 1116 spin_unlock(&this_parent->d_lock); 1117 done_seqretry(&rename_lock, seq); 1118 return; 1119 1120 rename_retry: 1121 if (!retry) 1122 return; 1123 seq = 1; 1124 goto again; 1125 } 1126 1127 /* 1128 * Search for at least 1 mount point in the dentry's subdirs. 1129 * We descend to the next level whenever the d_subdirs 1130 * list is non-empty and continue searching. 1131 */ 1132 1133 static enum d_walk_ret check_mount(void *data, struct dentry *dentry) 1134 { 1135 int *ret = data; 1136 if (d_mountpoint(dentry)) { 1137 *ret = 1; 1138 return D_WALK_QUIT; 1139 } 1140 return D_WALK_CONTINUE; 1141 } 1142 1143 /** 1144 * have_submounts - check for mounts over a dentry 1145 * @parent: dentry to check. 1146 * 1147 * Return true if the parent or its subdirectories contain 1148 * a mount point 1149 */ 1150 int have_submounts(struct dentry *parent) 1151 { 1152 int ret = 0; 1153 1154 d_walk(parent, &ret, check_mount, NULL); 1155 1156 return ret; 1157 } 1158 EXPORT_SYMBOL(have_submounts); 1159 1160 /* 1161 * Called by mount code to set a mountpoint and check if the mountpoint is 1162 * reachable (e.g. NFS can unhash a directory dentry and then the complete 1163 * subtree can become unreachable). 1164 * 1165 * Only one of d_invalidate() and d_set_mounted() must succeed. For 1166 * this reason take rename_lock and d_lock on dentry and ancestors. 1167 */ 1168 int d_set_mounted(struct dentry *dentry) 1169 { 1170 struct dentry *p; 1171 int ret = -ENOENT; 1172 write_seqlock(&rename_lock); 1173 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { 1174 /* Need exclusion wrt. d_invalidate() */ 1175 spin_lock(&p->d_lock); 1176 if (unlikely(d_unhashed(p))) { 1177 spin_unlock(&p->d_lock); 1178 goto out; 1179 } 1180 spin_unlock(&p->d_lock); 1181 } 1182 spin_lock(&dentry->d_lock); 1183 if (!d_unlinked(dentry)) { 1184 dentry->d_flags |= DCACHE_MOUNTED; 1185 ret = 0; 1186 } 1187 spin_unlock(&dentry->d_lock); 1188 out: 1189 write_sequnlock(&rename_lock); 1190 return ret; 1191 } 1192 1193 /* 1194 * Search the dentry child list of the specified parent, 1195 * and move any unused dentries to the end of the unused 1196 * list for prune_dcache(). We descend to the next level 1197 * whenever the d_subdirs list is non-empty and continue 1198 * searching. 1199 * 1200 * It returns zero iff there are no unused children, 1201 * otherwise it returns the number of children moved to 1202 * the end of the unused list. This may not be the total 1203 * number of unused children, because select_parent can 1204 * drop the lock and return early due to latency 1205 * constraints. 1206 */ 1207 1208 struct select_data { 1209 struct dentry *start; 1210 struct list_head dispose; 1211 int found; 1212 }; 1213 1214 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) 1215 { 1216 struct select_data *data = _data; 1217 enum d_walk_ret ret = D_WALK_CONTINUE; 1218 1219 if (data->start == dentry) 1220 goto out; 1221 1222 if (dentry->d_flags & DCACHE_SHRINK_LIST) { 1223 data->found++; 1224 } else { 1225 if (dentry->d_flags & DCACHE_LRU_LIST) 1226 d_lru_del(dentry); 1227 if (!dentry->d_lockref.count) { 1228 d_shrink_add(dentry, &data->dispose); 1229 data->found++; 1230 } 1231 } 1232 /* 1233 * We can return to the caller if we have found some (this 1234 * ensures forward progress). We'll be coming back to find 1235 * the rest. 1236 */ 1237 if (!list_empty(&data->dispose)) 1238 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; 1239 out: 1240 return ret; 1241 } 1242 1243 /** 1244 * shrink_dcache_parent - prune dcache 1245 * @parent: parent of entries to prune 1246 * 1247 * Prune the dcache to remove unused children of the parent dentry. 1248 */ 1249 void shrink_dcache_parent(struct dentry *parent) 1250 { 1251 for (;;) { 1252 struct select_data data; 1253 1254 INIT_LIST_HEAD(&data.dispose); 1255 data.start = parent; 1256 data.found = 0; 1257 1258 d_walk(parent, &data, select_collect, NULL); 1259 if (!data.found) 1260 break; 1261 1262 shrink_dentry_list(&data.dispose); 1263 cond_resched(); 1264 } 1265 } 1266 EXPORT_SYMBOL(shrink_dcache_parent); 1267 1268 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) 1269 { 1270 /* it has busy descendents; complain about those instead */ 1271 if (!list_empty(&dentry->d_subdirs)) 1272 return D_WALK_CONTINUE; 1273 1274 /* root with refcount 1 is fine */ 1275 if (dentry == _data && dentry->d_lockref.count == 1) 1276 return D_WALK_CONTINUE; 1277 1278 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " 1279 " still in use (%d) [unmount of %s %s]\n", 1280 dentry, 1281 dentry->d_inode ? 1282 dentry->d_inode->i_ino : 0UL, 1283 dentry, 1284 dentry->d_lockref.count, 1285 dentry->d_sb->s_type->name, 1286 dentry->d_sb->s_id); 1287 WARN_ON(1); 1288 return D_WALK_CONTINUE; 1289 } 1290 1291 static void do_one_tree(struct dentry *dentry) 1292 { 1293 shrink_dcache_parent(dentry); 1294 d_walk(dentry, dentry, umount_check, NULL); 1295 d_drop(dentry); 1296 dput(dentry); 1297 } 1298 1299 /* 1300 * destroy the dentries attached to a superblock on unmounting 1301 */ 1302 void shrink_dcache_for_umount(struct super_block *sb) 1303 { 1304 struct dentry *dentry; 1305 1306 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); 1307 1308 dentry = sb->s_root; 1309 sb->s_root = NULL; 1310 do_one_tree(dentry); 1311 1312 while (!hlist_bl_empty(&sb->s_anon)) { 1313 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash)); 1314 do_one_tree(dentry); 1315 } 1316 } 1317 1318 struct detach_data { 1319 struct select_data select; 1320 struct dentry *mountpoint; 1321 }; 1322 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry) 1323 { 1324 struct detach_data *data = _data; 1325 1326 if (d_mountpoint(dentry)) { 1327 __dget_dlock(dentry); 1328 data->mountpoint = dentry; 1329 return D_WALK_QUIT; 1330 } 1331 1332 return select_collect(&data->select, dentry); 1333 } 1334 1335 static void check_and_drop(void *_data) 1336 { 1337 struct detach_data *data = _data; 1338 1339 if (!data->mountpoint && !data->select.found) 1340 __d_drop(data->select.start); 1341 } 1342 1343 /** 1344 * d_invalidate - detach submounts, prune dcache, and drop 1345 * @dentry: dentry to invalidate (aka detach, prune and drop) 1346 * 1347 * no dcache lock. 1348 * 1349 * The final d_drop is done as an atomic operation relative to 1350 * rename_lock ensuring there are no races with d_set_mounted. This 1351 * ensures there are no unhashed dentries on the path to a mountpoint. 1352 */ 1353 void d_invalidate(struct dentry *dentry) 1354 { 1355 /* 1356 * If it's already been dropped, return OK. 1357 */ 1358 spin_lock(&dentry->d_lock); 1359 if (d_unhashed(dentry)) { 1360 spin_unlock(&dentry->d_lock); 1361 return; 1362 } 1363 spin_unlock(&dentry->d_lock); 1364 1365 /* Negative dentries can be dropped without further checks */ 1366 if (!dentry->d_inode) { 1367 d_drop(dentry); 1368 return; 1369 } 1370 1371 for (;;) { 1372 struct detach_data data; 1373 1374 data.mountpoint = NULL; 1375 INIT_LIST_HEAD(&data.select.dispose); 1376 data.select.start = dentry; 1377 data.select.found = 0; 1378 1379 d_walk(dentry, &data, detach_and_collect, check_and_drop); 1380 1381 if (data.select.found) 1382 shrink_dentry_list(&data.select.dispose); 1383 1384 if (data.mountpoint) { 1385 detach_mounts(data.mountpoint); 1386 dput(data.mountpoint); 1387 } 1388 1389 if (!data.mountpoint && !data.select.found) 1390 break; 1391 1392 cond_resched(); 1393 } 1394 } 1395 EXPORT_SYMBOL(d_invalidate); 1396 1397 /** 1398 * __d_alloc - allocate a dcache entry 1399 * @sb: filesystem it will belong to 1400 * @name: qstr of the name 1401 * 1402 * Allocates a dentry. It returns %NULL if there is insufficient memory 1403 * available. On a success the dentry is returned. The name passed in is 1404 * copied and the copy passed in may be reused after this call. 1405 */ 1406 1407 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) 1408 { 1409 struct dentry *dentry; 1410 char *dname; 1411 1412 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); 1413 if (!dentry) 1414 return NULL; 1415 1416 /* 1417 * We guarantee that the inline name is always NUL-terminated. 1418 * This way the memcpy() done by the name switching in rename 1419 * will still always have a NUL at the end, even if we might 1420 * be overwriting an internal NUL character 1421 */ 1422 dentry->d_iname[DNAME_INLINE_LEN-1] = 0; 1423 if (name->len > DNAME_INLINE_LEN-1) { 1424 size_t size = offsetof(struct external_name, name[1]); 1425 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL); 1426 if (!p) { 1427 kmem_cache_free(dentry_cache, dentry); 1428 return NULL; 1429 } 1430 atomic_set(&p->u.count, 1); 1431 dname = p->name; 1432 } else { 1433 dname = dentry->d_iname; 1434 } 1435 1436 dentry->d_name.len = name->len; 1437 dentry->d_name.hash = name->hash; 1438 memcpy(dname, name->name, name->len); 1439 dname[name->len] = 0; 1440 1441 /* Make sure we always see the terminating NUL character */ 1442 smp_wmb(); 1443 dentry->d_name.name = dname; 1444 1445 dentry->d_lockref.count = 1; 1446 dentry->d_flags = 0; 1447 spin_lock_init(&dentry->d_lock); 1448 seqcount_init(&dentry->d_seq); 1449 dentry->d_inode = NULL; 1450 dentry->d_parent = dentry; 1451 dentry->d_sb = sb; 1452 dentry->d_op = NULL; 1453 dentry->d_fsdata = NULL; 1454 INIT_HLIST_BL_NODE(&dentry->d_hash); 1455 INIT_LIST_HEAD(&dentry->d_lru); 1456 INIT_LIST_HEAD(&dentry->d_subdirs); 1457 INIT_HLIST_NODE(&dentry->d_alias); 1458 INIT_LIST_HEAD(&dentry->d_u.d_child); 1459 d_set_d_op(dentry, dentry->d_sb->s_d_op); 1460 1461 this_cpu_inc(nr_dentry); 1462 1463 return dentry; 1464 } 1465 1466 /** 1467 * d_alloc - allocate a dcache entry 1468 * @parent: parent of entry to allocate 1469 * @name: qstr of the name 1470 * 1471 * Allocates a dentry. It returns %NULL if there is insufficient memory 1472 * available. On a success the dentry is returned. The name passed in is 1473 * copied and the copy passed in may be reused after this call. 1474 */ 1475 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) 1476 { 1477 struct dentry *dentry = __d_alloc(parent->d_sb, name); 1478 if (!dentry) 1479 return NULL; 1480 1481 spin_lock(&parent->d_lock); 1482 /* 1483 * don't need child lock because it is not subject 1484 * to concurrency here 1485 */ 1486 __dget_dlock(parent); 1487 dentry->d_parent = parent; 1488 list_add(&dentry->d_u.d_child, &parent->d_subdirs); 1489 spin_unlock(&parent->d_lock); 1490 1491 return dentry; 1492 } 1493 EXPORT_SYMBOL(d_alloc); 1494 1495 /** 1496 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) 1497 * @sb: the superblock 1498 * @name: qstr of the name 1499 * 1500 * For a filesystem that just pins its dentries in memory and never 1501 * performs lookups at all, return an unhashed IS_ROOT dentry. 1502 */ 1503 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) 1504 { 1505 return __d_alloc(sb, name); 1506 } 1507 EXPORT_SYMBOL(d_alloc_pseudo); 1508 1509 struct dentry *d_alloc_name(struct dentry *parent, const char *name) 1510 { 1511 struct qstr q; 1512 1513 q.name = name; 1514 q.len = strlen(name); 1515 q.hash = full_name_hash(q.name, q.len); 1516 return d_alloc(parent, &q); 1517 } 1518 EXPORT_SYMBOL(d_alloc_name); 1519 1520 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) 1521 { 1522 WARN_ON_ONCE(dentry->d_op); 1523 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | 1524 DCACHE_OP_COMPARE | 1525 DCACHE_OP_REVALIDATE | 1526 DCACHE_OP_WEAK_REVALIDATE | 1527 DCACHE_OP_DELETE )); 1528 dentry->d_op = op; 1529 if (!op) 1530 return; 1531 if (op->d_hash) 1532 dentry->d_flags |= DCACHE_OP_HASH; 1533 if (op->d_compare) 1534 dentry->d_flags |= DCACHE_OP_COMPARE; 1535 if (op->d_revalidate) 1536 dentry->d_flags |= DCACHE_OP_REVALIDATE; 1537 if (op->d_weak_revalidate) 1538 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; 1539 if (op->d_delete) 1540 dentry->d_flags |= DCACHE_OP_DELETE; 1541 if (op->d_prune) 1542 dentry->d_flags |= DCACHE_OP_PRUNE; 1543 1544 } 1545 EXPORT_SYMBOL(d_set_d_op); 1546 1547 static unsigned d_flags_for_inode(struct inode *inode) 1548 { 1549 unsigned add_flags = DCACHE_FILE_TYPE; 1550 1551 if (!inode) 1552 return DCACHE_MISS_TYPE; 1553 1554 if (S_ISDIR(inode->i_mode)) { 1555 add_flags = DCACHE_DIRECTORY_TYPE; 1556 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { 1557 if (unlikely(!inode->i_op->lookup)) 1558 add_flags = DCACHE_AUTODIR_TYPE; 1559 else 1560 inode->i_opflags |= IOP_LOOKUP; 1561 } 1562 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { 1563 if (unlikely(inode->i_op->follow_link)) 1564 add_flags = DCACHE_SYMLINK_TYPE; 1565 else 1566 inode->i_opflags |= IOP_NOFOLLOW; 1567 } 1568 1569 if (unlikely(IS_AUTOMOUNT(inode))) 1570 add_flags |= DCACHE_NEED_AUTOMOUNT; 1571 return add_flags; 1572 } 1573 1574 static void __d_instantiate(struct dentry *dentry, struct inode *inode) 1575 { 1576 unsigned add_flags = d_flags_for_inode(inode); 1577 1578 spin_lock(&dentry->d_lock); 1579 __d_set_type(dentry, add_flags); 1580 if (inode) 1581 hlist_add_head(&dentry->d_alias, &inode->i_dentry); 1582 dentry->d_inode = inode; 1583 dentry_rcuwalk_barrier(dentry); 1584 spin_unlock(&dentry->d_lock); 1585 fsnotify_d_instantiate(dentry, inode); 1586 } 1587 1588 /** 1589 * d_instantiate - fill in inode information for a dentry 1590 * @entry: dentry to complete 1591 * @inode: inode to attach to this dentry 1592 * 1593 * Fill in inode information in the entry. 1594 * 1595 * This turns negative dentries into productive full members 1596 * of society. 1597 * 1598 * NOTE! This assumes that the inode count has been incremented 1599 * (or otherwise set) by the caller to indicate that it is now 1600 * in use by the dcache. 1601 */ 1602 1603 void d_instantiate(struct dentry *entry, struct inode * inode) 1604 { 1605 BUG_ON(!hlist_unhashed(&entry->d_alias)); 1606 if (inode) 1607 spin_lock(&inode->i_lock); 1608 __d_instantiate(entry, inode); 1609 if (inode) 1610 spin_unlock(&inode->i_lock); 1611 security_d_instantiate(entry, inode); 1612 } 1613 EXPORT_SYMBOL(d_instantiate); 1614 1615 /** 1616 * d_instantiate_unique - instantiate a non-aliased dentry 1617 * @entry: dentry to instantiate 1618 * @inode: inode to attach to this dentry 1619 * 1620 * Fill in inode information in the entry. On success, it returns NULL. 1621 * If an unhashed alias of "entry" already exists, then we return the 1622 * aliased dentry instead and drop one reference to inode. 1623 * 1624 * Note that in order to avoid conflicts with rename() etc, the caller 1625 * had better be holding the parent directory semaphore. 1626 * 1627 * This also assumes that the inode count has been incremented 1628 * (or otherwise set) by the caller to indicate that it is now 1629 * in use by the dcache. 1630 */ 1631 static struct dentry *__d_instantiate_unique(struct dentry *entry, 1632 struct inode *inode) 1633 { 1634 struct dentry *alias; 1635 int len = entry->d_name.len; 1636 const char *name = entry->d_name.name; 1637 unsigned int hash = entry->d_name.hash; 1638 1639 if (!inode) { 1640 __d_instantiate(entry, NULL); 1641 return NULL; 1642 } 1643 1644 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { 1645 /* 1646 * Don't need alias->d_lock here, because aliases with 1647 * d_parent == entry->d_parent are not subject to name or 1648 * parent changes, because the parent inode i_mutex is held. 1649 */ 1650 if (alias->d_name.hash != hash) 1651 continue; 1652 if (alias->d_parent != entry->d_parent) 1653 continue; 1654 if (alias->d_name.len != len) 1655 continue; 1656 if (dentry_cmp(alias, name, len)) 1657 continue; 1658 __dget(alias); 1659 return alias; 1660 } 1661 1662 __d_instantiate(entry, inode); 1663 return NULL; 1664 } 1665 1666 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) 1667 { 1668 struct dentry *result; 1669 1670 BUG_ON(!hlist_unhashed(&entry->d_alias)); 1671 1672 if (inode) 1673 spin_lock(&inode->i_lock); 1674 result = __d_instantiate_unique(entry, inode); 1675 if (inode) 1676 spin_unlock(&inode->i_lock); 1677 1678 if (!result) { 1679 security_d_instantiate(entry, inode); 1680 return NULL; 1681 } 1682 1683 BUG_ON(!d_unhashed(result)); 1684 iput(inode); 1685 return result; 1686 } 1687 1688 EXPORT_SYMBOL(d_instantiate_unique); 1689 1690 /** 1691 * d_instantiate_no_diralias - instantiate a non-aliased dentry 1692 * @entry: dentry to complete 1693 * @inode: inode to attach to this dentry 1694 * 1695 * Fill in inode information in the entry. If a directory alias is found, then 1696 * return an error (and drop inode). Together with d_materialise_unique() this 1697 * guarantees that a directory inode may never have more than one alias. 1698 */ 1699 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) 1700 { 1701 BUG_ON(!hlist_unhashed(&entry->d_alias)); 1702 1703 spin_lock(&inode->i_lock); 1704 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { 1705 spin_unlock(&inode->i_lock); 1706 iput(inode); 1707 return -EBUSY; 1708 } 1709 __d_instantiate(entry, inode); 1710 spin_unlock(&inode->i_lock); 1711 security_d_instantiate(entry, inode); 1712 1713 return 0; 1714 } 1715 EXPORT_SYMBOL(d_instantiate_no_diralias); 1716 1717 struct dentry *d_make_root(struct inode *root_inode) 1718 { 1719 struct dentry *res = NULL; 1720 1721 if (root_inode) { 1722 static const struct qstr name = QSTR_INIT("/", 1); 1723 1724 res = __d_alloc(root_inode->i_sb, &name); 1725 if (res) 1726 d_instantiate(res, root_inode); 1727 else 1728 iput(root_inode); 1729 } 1730 return res; 1731 } 1732 EXPORT_SYMBOL(d_make_root); 1733 1734 static struct dentry * __d_find_any_alias(struct inode *inode) 1735 { 1736 struct dentry *alias; 1737 1738 if (hlist_empty(&inode->i_dentry)) 1739 return NULL; 1740 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); 1741 __dget(alias); 1742 return alias; 1743 } 1744 1745 /** 1746 * d_find_any_alias - find any alias for a given inode 1747 * @inode: inode to find an alias for 1748 * 1749 * If any aliases exist for the given inode, take and return a 1750 * reference for one of them. If no aliases exist, return %NULL. 1751 */ 1752 struct dentry *d_find_any_alias(struct inode *inode) 1753 { 1754 struct dentry *de; 1755 1756 spin_lock(&inode->i_lock); 1757 de = __d_find_any_alias(inode); 1758 spin_unlock(&inode->i_lock); 1759 return de; 1760 } 1761 EXPORT_SYMBOL(d_find_any_alias); 1762 1763 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected) 1764 { 1765 static const struct qstr anonstring = QSTR_INIT("/", 1); 1766 struct dentry *tmp; 1767 struct dentry *res; 1768 unsigned add_flags; 1769 1770 if (!inode) 1771 return ERR_PTR(-ESTALE); 1772 if (IS_ERR(inode)) 1773 return ERR_CAST(inode); 1774 1775 res = d_find_any_alias(inode); 1776 if (res) 1777 goto out_iput; 1778 1779 tmp = __d_alloc(inode->i_sb, &anonstring); 1780 if (!tmp) { 1781 res = ERR_PTR(-ENOMEM); 1782 goto out_iput; 1783 } 1784 1785 spin_lock(&inode->i_lock); 1786 res = __d_find_any_alias(inode); 1787 if (res) { 1788 spin_unlock(&inode->i_lock); 1789 dput(tmp); 1790 goto out_iput; 1791 } 1792 1793 /* attach a disconnected dentry */ 1794 add_flags = d_flags_for_inode(inode); 1795 1796 if (disconnected) 1797 add_flags |= DCACHE_DISCONNECTED; 1798 1799 spin_lock(&tmp->d_lock); 1800 tmp->d_inode = inode; 1801 tmp->d_flags |= add_flags; 1802 hlist_add_head(&tmp->d_alias, &inode->i_dentry); 1803 hlist_bl_lock(&tmp->d_sb->s_anon); 1804 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); 1805 hlist_bl_unlock(&tmp->d_sb->s_anon); 1806 spin_unlock(&tmp->d_lock); 1807 spin_unlock(&inode->i_lock); 1808 security_d_instantiate(tmp, inode); 1809 1810 return tmp; 1811 1812 out_iput: 1813 if (res && !IS_ERR(res)) 1814 security_d_instantiate(res, inode); 1815 iput(inode); 1816 return res; 1817 } 1818 1819 /** 1820 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode 1821 * @inode: inode to allocate the dentry for 1822 * 1823 * Obtain a dentry for an inode resulting from NFS filehandle conversion or 1824 * similar open by handle operations. The returned dentry may be anonymous, 1825 * or may have a full name (if the inode was already in the cache). 1826 * 1827 * When called on a directory inode, we must ensure that the inode only ever 1828 * has one dentry. If a dentry is found, that is returned instead of 1829 * allocating a new one. 1830 * 1831 * On successful return, the reference to the inode has been transferred 1832 * to the dentry. In case of an error the reference on the inode is released. 1833 * To make it easier to use in export operations a %NULL or IS_ERR inode may 1834 * be passed in and the error will be propagated to the return value, 1835 * with a %NULL @inode replaced by ERR_PTR(-ESTALE). 1836 */ 1837 struct dentry *d_obtain_alias(struct inode *inode) 1838 { 1839 return __d_obtain_alias(inode, 1); 1840 } 1841 EXPORT_SYMBOL(d_obtain_alias); 1842 1843 /** 1844 * d_obtain_root - find or allocate a dentry for a given inode 1845 * @inode: inode to allocate the dentry for 1846 * 1847 * Obtain an IS_ROOT dentry for the root of a filesystem. 1848 * 1849 * We must ensure that directory inodes only ever have one dentry. If a 1850 * dentry is found, that is returned instead of allocating a new one. 1851 * 1852 * On successful return, the reference to the inode has been transferred 1853 * to the dentry. In case of an error the reference on the inode is 1854 * released. A %NULL or IS_ERR inode may be passed in and will be the 1855 * error will be propagate to the return value, with a %NULL @inode 1856 * replaced by ERR_PTR(-ESTALE). 1857 */ 1858 struct dentry *d_obtain_root(struct inode *inode) 1859 { 1860 return __d_obtain_alias(inode, 0); 1861 } 1862 EXPORT_SYMBOL(d_obtain_root); 1863 1864 /** 1865 * d_add_ci - lookup or allocate new dentry with case-exact name 1866 * @inode: the inode case-insensitive lookup has found 1867 * @dentry: the negative dentry that was passed to the parent's lookup func 1868 * @name: the case-exact name to be associated with the returned dentry 1869 * 1870 * This is to avoid filling the dcache with case-insensitive names to the 1871 * same inode, only the actual correct case is stored in the dcache for 1872 * case-insensitive filesystems. 1873 * 1874 * For a case-insensitive lookup match and if the the case-exact dentry 1875 * already exists in in the dcache, use it and return it. 1876 * 1877 * If no entry exists with the exact case name, allocate new dentry with 1878 * the exact case, and return the spliced entry. 1879 */ 1880 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, 1881 struct qstr *name) 1882 { 1883 struct dentry *found; 1884 struct dentry *new; 1885 1886 /* 1887 * First check if a dentry matching the name already exists, 1888 * if not go ahead and create it now. 1889 */ 1890 found = d_hash_and_lookup(dentry->d_parent, name); 1891 if (unlikely(IS_ERR(found))) 1892 goto err_out; 1893 if (!found) { 1894 new = d_alloc(dentry->d_parent, name); 1895 if (!new) { 1896 found = ERR_PTR(-ENOMEM); 1897 goto err_out; 1898 } 1899 1900 found = d_splice_alias(inode, new); 1901 if (found) { 1902 dput(new); 1903 return found; 1904 } 1905 return new; 1906 } 1907 1908 /* 1909 * If a matching dentry exists, and it's not negative use it. 1910 * 1911 * Decrement the reference count to balance the iget() done 1912 * earlier on. 1913 */ 1914 if (found->d_inode) { 1915 if (unlikely(found->d_inode != inode)) { 1916 /* This can't happen because bad inodes are unhashed. */ 1917 BUG_ON(!is_bad_inode(inode)); 1918 BUG_ON(!is_bad_inode(found->d_inode)); 1919 } 1920 iput(inode); 1921 return found; 1922 } 1923 1924 /* 1925 * Negative dentry: instantiate it unless the inode is a directory and 1926 * already has a dentry. 1927 */ 1928 new = d_splice_alias(inode, found); 1929 if (new) { 1930 dput(found); 1931 found = new; 1932 } 1933 return found; 1934 1935 err_out: 1936 iput(inode); 1937 return found; 1938 } 1939 EXPORT_SYMBOL(d_add_ci); 1940 1941 /* 1942 * Do the slow-case of the dentry name compare. 1943 * 1944 * Unlike the dentry_cmp() function, we need to atomically 1945 * load the name and length information, so that the 1946 * filesystem can rely on them, and can use the 'name' and 1947 * 'len' information without worrying about walking off the 1948 * end of memory etc. 1949 * 1950 * Thus the read_seqcount_retry() and the "duplicate" info 1951 * in arguments (the low-level filesystem should not look 1952 * at the dentry inode or name contents directly, since 1953 * rename can change them while we're in RCU mode). 1954 */ 1955 enum slow_d_compare { 1956 D_COMP_OK, 1957 D_COMP_NOMATCH, 1958 D_COMP_SEQRETRY, 1959 }; 1960 1961 static noinline enum slow_d_compare slow_dentry_cmp( 1962 const struct dentry *parent, 1963 struct dentry *dentry, 1964 unsigned int seq, 1965 const struct qstr *name) 1966 { 1967 int tlen = dentry->d_name.len; 1968 const char *tname = dentry->d_name.name; 1969 1970 if (read_seqcount_retry(&dentry->d_seq, seq)) { 1971 cpu_relax(); 1972 return D_COMP_SEQRETRY; 1973 } 1974 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) 1975 return D_COMP_NOMATCH; 1976 return D_COMP_OK; 1977 } 1978 1979 /** 1980 * __d_lookup_rcu - search for a dentry (racy, store-free) 1981 * @parent: parent dentry 1982 * @name: qstr of name we wish to find 1983 * @seqp: returns d_seq value at the point where the dentry was found 1984 * Returns: dentry, or NULL 1985 * 1986 * __d_lookup_rcu is the dcache lookup function for rcu-walk name 1987 * resolution (store-free path walking) design described in 1988 * Documentation/filesystems/path-lookup.txt. 1989 * 1990 * This is not to be used outside core vfs. 1991 * 1992 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock 1993 * held, and rcu_read_lock held. The returned dentry must not be stored into 1994 * without taking d_lock and checking d_seq sequence count against @seq 1995 * returned here. 1996 * 1997 * A refcount may be taken on the found dentry with the d_rcu_to_refcount 1998 * function. 1999 * 2000 * Alternatively, __d_lookup_rcu may be called again to look up the child of 2001 * the returned dentry, so long as its parent's seqlock is checked after the 2002 * child is looked up. Thus, an interlocking stepping of sequence lock checks 2003 * is formed, giving integrity down the path walk. 2004 * 2005 * NOTE! The caller *has* to check the resulting dentry against the sequence 2006 * number we've returned before using any of the resulting dentry state! 2007 */ 2008 struct dentry *__d_lookup_rcu(const struct dentry *parent, 2009 const struct qstr *name, 2010 unsigned *seqp) 2011 { 2012 u64 hashlen = name->hash_len; 2013 const unsigned char *str = name->name; 2014 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); 2015 struct hlist_bl_node *node; 2016 struct dentry *dentry; 2017 2018 /* 2019 * Note: There is significant duplication with __d_lookup_rcu which is 2020 * required to prevent single threaded performance regressions 2021 * especially on architectures where smp_rmb (in seqcounts) are costly. 2022 * Keep the two functions in sync. 2023 */ 2024 2025 /* 2026 * The hash list is protected using RCU. 2027 * 2028 * Carefully use d_seq when comparing a candidate dentry, to avoid 2029 * races with d_move(). 2030 * 2031 * It is possible that concurrent renames can mess up our list 2032 * walk here and result in missing our dentry, resulting in the 2033 * false-negative result. d_lookup() protects against concurrent 2034 * renames using rename_lock seqlock. 2035 * 2036 * See Documentation/filesystems/path-lookup.txt for more details. 2037 */ 2038 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { 2039 unsigned seq; 2040 2041 seqretry: 2042 /* 2043 * The dentry sequence count protects us from concurrent 2044 * renames, and thus protects parent and name fields. 2045 * 2046 * The caller must perform a seqcount check in order 2047 * to do anything useful with the returned dentry. 2048 * 2049 * NOTE! We do a "raw" seqcount_begin here. That means that 2050 * we don't wait for the sequence count to stabilize if it 2051 * is in the middle of a sequence change. If we do the slow 2052 * dentry compare, we will do seqretries until it is stable, 2053 * and if we end up with a successful lookup, we actually 2054 * want to exit RCU lookup anyway. 2055 */ 2056 seq = raw_seqcount_begin(&dentry->d_seq); 2057 if (dentry->d_parent != parent) 2058 continue; 2059 if (d_unhashed(dentry)) 2060 continue; 2061 2062 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { 2063 if (dentry->d_name.hash != hashlen_hash(hashlen)) 2064 continue; 2065 *seqp = seq; 2066 switch (slow_dentry_cmp(parent, dentry, seq, name)) { 2067 case D_COMP_OK: 2068 return dentry; 2069 case D_COMP_NOMATCH: 2070 continue; 2071 default: 2072 goto seqretry; 2073 } 2074 } 2075 2076 if (dentry->d_name.hash_len != hashlen) 2077 continue; 2078 *seqp = seq; 2079 if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) 2080 return dentry; 2081 } 2082 return NULL; 2083 } 2084 2085 /** 2086 * d_lookup - search for a dentry 2087 * @parent: parent dentry 2088 * @name: qstr of name we wish to find 2089 * Returns: dentry, or NULL 2090 * 2091 * d_lookup searches the children of the parent dentry for the name in 2092 * question. If the dentry is found its reference count is incremented and the 2093 * dentry is returned. The caller must use dput to free the entry when it has 2094 * finished using it. %NULL is returned if the dentry does not exist. 2095 */ 2096 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) 2097 { 2098 struct dentry *dentry; 2099 unsigned seq; 2100 2101 do { 2102 seq = read_seqbegin(&rename_lock); 2103 dentry = __d_lookup(parent, name); 2104 if (dentry) 2105 break; 2106 } while (read_seqretry(&rename_lock, seq)); 2107 return dentry; 2108 } 2109 EXPORT_SYMBOL(d_lookup); 2110 2111 /** 2112 * __d_lookup - search for a dentry (racy) 2113 * @parent: parent dentry 2114 * @name: qstr of name we wish to find 2115 * Returns: dentry, or NULL 2116 * 2117 * __d_lookup is like d_lookup, however it may (rarely) return a 2118 * false-negative result due to unrelated rename activity. 2119 * 2120 * __d_lookup is slightly faster by avoiding rename_lock read seqlock, 2121 * however it must be used carefully, eg. with a following d_lookup in 2122 * the case of failure. 2123 * 2124 * __d_lookup callers must be commented. 2125 */ 2126 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) 2127 { 2128 unsigned int len = name->len; 2129 unsigned int hash = name->hash; 2130 const unsigned char *str = name->name; 2131 struct hlist_bl_head *b = d_hash(parent, hash); 2132 struct hlist_bl_node *node; 2133 struct dentry *found = NULL; 2134 struct dentry *dentry; 2135 2136 /* 2137 * Note: There is significant duplication with __d_lookup_rcu which is 2138 * required to prevent single threaded performance regressions 2139 * especially on architectures where smp_rmb (in seqcounts) are costly. 2140 * Keep the two functions in sync. 2141 */ 2142 2143 /* 2144 * The hash list is protected using RCU. 2145 * 2146 * Take d_lock when comparing a candidate dentry, to avoid races 2147 * with d_move(). 2148 * 2149 * It is possible that concurrent renames can mess up our list 2150 * walk here and result in missing our dentry, resulting in the 2151 * false-negative result. d_lookup() protects against concurrent 2152 * renames using rename_lock seqlock. 2153 * 2154 * See Documentation/filesystems/path-lookup.txt for more details. 2155 */ 2156 rcu_read_lock(); 2157 2158 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { 2159 2160 if (dentry->d_name.hash != hash) 2161 continue; 2162 2163 spin_lock(&dentry->d_lock); 2164 if (dentry->d_parent != parent) 2165 goto next; 2166 if (d_unhashed(dentry)) 2167 goto next; 2168 2169 /* 2170 * It is safe to compare names since d_move() cannot 2171 * change the qstr (protected by d_lock). 2172 */ 2173 if (parent->d_flags & DCACHE_OP_COMPARE) { 2174 int tlen = dentry->d_name.len; 2175 const char *tname = dentry->d_name.name; 2176 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) 2177 goto next; 2178 } else { 2179 if (dentry->d_name.len != len) 2180 goto next; 2181 if (dentry_cmp(dentry, str, len)) 2182 goto next; 2183 } 2184 2185 dentry->d_lockref.count++; 2186 found = dentry; 2187 spin_unlock(&dentry->d_lock); 2188 break; 2189 next: 2190 spin_unlock(&dentry->d_lock); 2191 } 2192 rcu_read_unlock(); 2193 2194 return found; 2195 } 2196 2197 /** 2198 * d_hash_and_lookup - hash the qstr then search for a dentry 2199 * @dir: Directory to search in 2200 * @name: qstr of name we wish to find 2201 * 2202 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) 2203 */ 2204 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) 2205 { 2206 /* 2207 * Check for a fs-specific hash function. Note that we must 2208 * calculate the standard hash first, as the d_op->d_hash() 2209 * routine may choose to leave the hash value unchanged. 2210 */ 2211 name->hash = full_name_hash(name->name, name->len); 2212 if (dir->d_flags & DCACHE_OP_HASH) { 2213 int err = dir->d_op->d_hash(dir, name); 2214 if (unlikely(err < 0)) 2215 return ERR_PTR(err); 2216 } 2217 return d_lookup(dir, name); 2218 } 2219 EXPORT_SYMBOL(d_hash_and_lookup); 2220 2221 /** 2222 * d_validate - verify dentry provided from insecure source (deprecated) 2223 * @dentry: The dentry alleged to be valid child of @dparent 2224 * @dparent: The parent dentry (known to be valid) 2225 * 2226 * An insecure source has sent us a dentry, here we verify it and dget() it. 2227 * This is used by ncpfs in its readdir implementation. 2228 * Zero is returned in the dentry is invalid. 2229 * 2230 * This function is slow for big directories, and deprecated, do not use it. 2231 */ 2232 int d_validate(struct dentry *dentry, struct dentry *dparent) 2233 { 2234 struct dentry *child; 2235 2236 spin_lock(&dparent->d_lock); 2237 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { 2238 if (dentry == child) { 2239 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 2240 __dget_dlock(dentry); 2241 spin_unlock(&dentry->d_lock); 2242 spin_unlock(&dparent->d_lock); 2243 return 1; 2244 } 2245 } 2246 spin_unlock(&dparent->d_lock); 2247 2248 return 0; 2249 } 2250 EXPORT_SYMBOL(d_validate); 2251 2252 /* 2253 * When a file is deleted, we have two options: 2254 * - turn this dentry into a negative dentry 2255 * - unhash this dentry and free it. 2256 * 2257 * Usually, we want to just turn this into 2258 * a negative dentry, but if anybody else is 2259 * currently using the dentry or the inode 2260 * we can't do that and we fall back on removing 2261 * it from the hash queues and waiting for 2262 * it to be deleted later when it has no users 2263 */ 2264 2265 /** 2266 * d_delete - delete a dentry 2267 * @dentry: The dentry to delete 2268 * 2269 * Turn the dentry into a negative dentry if possible, otherwise 2270 * remove it from the hash queues so it can be deleted later 2271 */ 2272 2273 void d_delete(struct dentry * dentry) 2274 { 2275 struct inode *inode; 2276 int isdir = 0; 2277 /* 2278 * Are we the only user? 2279 */ 2280 again: 2281 spin_lock(&dentry->d_lock); 2282 inode = dentry->d_inode; 2283 isdir = S_ISDIR(inode->i_mode); 2284 if (dentry->d_lockref.count == 1) { 2285 if (!spin_trylock(&inode->i_lock)) { 2286 spin_unlock(&dentry->d_lock); 2287 cpu_relax(); 2288 goto again; 2289 } 2290 dentry->d_flags &= ~DCACHE_CANT_MOUNT; 2291 dentry_unlink_inode(dentry); 2292 fsnotify_nameremove(dentry, isdir); 2293 return; 2294 } 2295 2296 if (!d_unhashed(dentry)) 2297 __d_drop(dentry); 2298 2299 spin_unlock(&dentry->d_lock); 2300 2301 fsnotify_nameremove(dentry, isdir); 2302 } 2303 EXPORT_SYMBOL(d_delete); 2304 2305 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) 2306 { 2307 BUG_ON(!d_unhashed(entry)); 2308 hlist_bl_lock(b); 2309 entry->d_flags |= DCACHE_RCUACCESS; 2310 hlist_bl_add_head_rcu(&entry->d_hash, b); 2311 hlist_bl_unlock(b); 2312 } 2313 2314 static void _d_rehash(struct dentry * entry) 2315 { 2316 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); 2317 } 2318 2319 /** 2320 * d_rehash - add an entry back to the hash 2321 * @entry: dentry to add to the hash 2322 * 2323 * Adds a dentry to the hash according to its name. 2324 */ 2325 2326 void d_rehash(struct dentry * entry) 2327 { 2328 spin_lock(&entry->d_lock); 2329 _d_rehash(entry); 2330 spin_unlock(&entry->d_lock); 2331 } 2332 EXPORT_SYMBOL(d_rehash); 2333 2334 /** 2335 * dentry_update_name_case - update case insensitive dentry with a new name 2336 * @dentry: dentry to be updated 2337 * @name: new name 2338 * 2339 * Update a case insensitive dentry with new case of name. 2340 * 2341 * dentry must have been returned by d_lookup with name @name. Old and new 2342 * name lengths must match (ie. no d_compare which allows mismatched name 2343 * lengths). 2344 * 2345 * Parent inode i_mutex must be held over d_lookup and into this call (to 2346 * keep renames and concurrent inserts, and readdir(2) away). 2347 */ 2348 void dentry_update_name_case(struct dentry *dentry, struct qstr *name) 2349 { 2350 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); 2351 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ 2352 2353 spin_lock(&dentry->d_lock); 2354 write_seqcount_begin(&dentry->d_seq); 2355 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); 2356 write_seqcount_end(&dentry->d_seq); 2357 spin_unlock(&dentry->d_lock); 2358 } 2359 EXPORT_SYMBOL(dentry_update_name_case); 2360 2361 static void swap_names(struct dentry *dentry, struct dentry *target) 2362 { 2363 if (unlikely(dname_external(target))) { 2364 if (unlikely(dname_external(dentry))) { 2365 /* 2366 * Both external: swap the pointers 2367 */ 2368 swap(target->d_name.name, dentry->d_name.name); 2369 } else { 2370 /* 2371 * dentry:internal, target:external. Steal target's 2372 * storage and make target internal. 2373 */ 2374 memcpy(target->d_iname, dentry->d_name.name, 2375 dentry->d_name.len + 1); 2376 dentry->d_name.name = target->d_name.name; 2377 target->d_name.name = target->d_iname; 2378 } 2379 } else { 2380 if (unlikely(dname_external(dentry))) { 2381 /* 2382 * dentry:external, target:internal. Give dentry's 2383 * storage to target and make dentry internal 2384 */ 2385 memcpy(dentry->d_iname, target->d_name.name, 2386 target->d_name.len + 1); 2387 target->d_name.name = dentry->d_name.name; 2388 dentry->d_name.name = dentry->d_iname; 2389 } else { 2390 /* 2391 * Both are internal. 2392 */ 2393 unsigned int i; 2394 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); 2395 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { 2396 swap(((long *) &dentry->d_iname)[i], 2397 ((long *) &target->d_iname)[i]); 2398 } 2399 } 2400 } 2401 swap(dentry->d_name.hash_len, target->d_name.hash_len); 2402 } 2403 2404 static void copy_name(struct dentry *dentry, struct dentry *target) 2405 { 2406 struct external_name *old_name = NULL; 2407 if (unlikely(dname_external(dentry))) 2408 old_name = external_name(dentry); 2409 if (unlikely(dname_external(target))) { 2410 atomic_inc(&external_name(target)->u.count); 2411 dentry->d_name = target->d_name; 2412 } else { 2413 memcpy(dentry->d_iname, target->d_name.name, 2414 target->d_name.len + 1); 2415 dentry->d_name.name = dentry->d_iname; 2416 dentry->d_name.hash_len = target->d_name.hash_len; 2417 } 2418 if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) 2419 kfree_rcu(old_name, u.head); 2420 } 2421 2422 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) 2423 { 2424 /* 2425 * XXXX: do we really need to take target->d_lock? 2426 */ 2427 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) 2428 spin_lock(&target->d_parent->d_lock); 2429 else { 2430 if (d_ancestor(dentry->d_parent, target->d_parent)) { 2431 spin_lock(&dentry->d_parent->d_lock); 2432 spin_lock_nested(&target->d_parent->d_lock, 2433 DENTRY_D_LOCK_NESTED); 2434 } else { 2435 spin_lock(&target->d_parent->d_lock); 2436 spin_lock_nested(&dentry->d_parent->d_lock, 2437 DENTRY_D_LOCK_NESTED); 2438 } 2439 } 2440 if (target < dentry) { 2441 spin_lock_nested(&target->d_lock, 2); 2442 spin_lock_nested(&dentry->d_lock, 3); 2443 } else { 2444 spin_lock_nested(&dentry->d_lock, 2); 2445 spin_lock_nested(&target->d_lock, 3); 2446 } 2447 } 2448 2449 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target) 2450 { 2451 if (target->d_parent != dentry->d_parent) 2452 spin_unlock(&dentry->d_parent->d_lock); 2453 if (target->d_parent != target) 2454 spin_unlock(&target->d_parent->d_lock); 2455 spin_unlock(&target->d_lock); 2456 spin_unlock(&dentry->d_lock); 2457 } 2458 2459 /* 2460 * When switching names, the actual string doesn't strictly have to 2461 * be preserved in the target - because we're dropping the target 2462 * anyway. As such, we can just do a simple memcpy() to copy over 2463 * the new name before we switch, unless we are going to rehash 2464 * it. Note that if we *do* unhash the target, we are not allowed 2465 * to rehash it without giving it a new name/hash key - whether 2466 * we swap or overwrite the names here, resulting name won't match 2467 * the reality in filesystem; it's only there for d_path() purposes. 2468 * Note that all of this is happening under rename_lock, so the 2469 * any hash lookup seeing it in the middle of manipulations will 2470 * be discarded anyway. So we do not care what happens to the hash 2471 * key in that case. 2472 */ 2473 /* 2474 * __d_move - move a dentry 2475 * @dentry: entry to move 2476 * @target: new dentry 2477 * @exchange: exchange the two dentries 2478 * 2479 * Update the dcache to reflect the move of a file name. Negative 2480 * dcache entries should not be moved in this way. Caller must hold 2481 * rename_lock, the i_mutex of the source and target directories, 2482 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). 2483 */ 2484 static void __d_move(struct dentry *dentry, struct dentry *target, 2485 bool exchange) 2486 { 2487 if (!dentry->d_inode) 2488 printk(KERN_WARNING "VFS: moving negative dcache entry\n"); 2489 2490 BUG_ON(d_ancestor(dentry, target)); 2491 BUG_ON(d_ancestor(target, dentry)); 2492 2493 dentry_lock_for_move(dentry, target); 2494 2495 write_seqcount_begin(&dentry->d_seq); 2496 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); 2497 2498 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ 2499 2500 /* 2501 * Move the dentry to the target hash queue. Don't bother checking 2502 * for the same hash queue because of how unlikely it is. 2503 */ 2504 __d_drop(dentry); 2505 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); 2506 2507 /* 2508 * Unhash the target (d_delete() is not usable here). If exchanging 2509 * the two dentries, then rehash onto the other's hash queue. 2510 */ 2511 __d_drop(target); 2512 if (exchange) { 2513 __d_rehash(target, 2514 d_hash(dentry->d_parent, dentry->d_name.hash)); 2515 } 2516 2517 /* Switch the names.. */ 2518 if (exchange) 2519 swap_names(dentry, target); 2520 else 2521 copy_name(dentry, target); 2522 2523 /* ... and switch them in the tree */ 2524 if (IS_ROOT(dentry)) { 2525 /* splicing a tree */ 2526 dentry->d_parent = target->d_parent; 2527 target->d_parent = target; 2528 list_del_init(&target->d_u.d_child); 2529 list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); 2530 } else { 2531 /* swapping two dentries */ 2532 swap(dentry->d_parent, target->d_parent); 2533 list_move(&target->d_u.d_child, &target->d_parent->d_subdirs); 2534 list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); 2535 if (exchange) 2536 fsnotify_d_move(target); 2537 fsnotify_d_move(dentry); 2538 } 2539 2540 write_seqcount_end(&target->d_seq); 2541 write_seqcount_end(&dentry->d_seq); 2542 2543 dentry_unlock_for_move(dentry, target); 2544 } 2545 2546 /* 2547 * d_move - move a dentry 2548 * @dentry: entry to move 2549 * @target: new dentry 2550 * 2551 * Update the dcache to reflect the move of a file name. Negative 2552 * dcache entries should not be moved in this way. See the locking 2553 * requirements for __d_move. 2554 */ 2555 void d_move(struct dentry *dentry, struct dentry *target) 2556 { 2557 write_seqlock(&rename_lock); 2558 __d_move(dentry, target, false); 2559 write_sequnlock(&rename_lock); 2560 } 2561 EXPORT_SYMBOL(d_move); 2562 2563 /* 2564 * d_exchange - exchange two dentries 2565 * @dentry1: first dentry 2566 * @dentry2: second dentry 2567 */ 2568 void d_exchange(struct dentry *dentry1, struct dentry *dentry2) 2569 { 2570 write_seqlock(&rename_lock); 2571 2572 WARN_ON(!dentry1->d_inode); 2573 WARN_ON(!dentry2->d_inode); 2574 WARN_ON(IS_ROOT(dentry1)); 2575 WARN_ON(IS_ROOT(dentry2)); 2576 2577 __d_move(dentry1, dentry2, true); 2578 2579 write_sequnlock(&rename_lock); 2580 } 2581 2582 /** 2583 * d_ancestor - search for an ancestor 2584 * @p1: ancestor dentry 2585 * @p2: child dentry 2586 * 2587 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is 2588 * an ancestor of p2, else NULL. 2589 */ 2590 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) 2591 { 2592 struct dentry *p; 2593 2594 for (p = p2; !IS_ROOT(p); p = p->d_parent) { 2595 if (p->d_parent == p1) 2596 return p; 2597 } 2598 return NULL; 2599 } 2600 2601 /* 2602 * This helper attempts to cope with remotely renamed directories 2603 * 2604 * It assumes that the caller is already holding 2605 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock 2606 * 2607 * Note: If ever the locking in lock_rename() changes, then please 2608 * remember to update this too... 2609 */ 2610 static struct dentry *__d_unalias(struct inode *inode, 2611 struct dentry *dentry, struct dentry *alias) 2612 { 2613 struct mutex *m1 = NULL, *m2 = NULL; 2614 struct dentry *ret = ERR_PTR(-EBUSY); 2615 2616 /* If alias and dentry share a parent, then no extra locks required */ 2617 if (alias->d_parent == dentry->d_parent) 2618 goto out_unalias; 2619 2620 /* See lock_rename() */ 2621 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) 2622 goto out_err; 2623 m1 = &dentry->d_sb->s_vfs_rename_mutex; 2624 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) 2625 goto out_err; 2626 m2 = &alias->d_parent->d_inode->i_mutex; 2627 out_unalias: 2628 __d_move(alias, dentry, false); 2629 ret = alias; 2630 out_err: 2631 spin_unlock(&inode->i_lock); 2632 if (m2) 2633 mutex_unlock(m2); 2634 if (m1) 2635 mutex_unlock(m1); 2636 return ret; 2637 } 2638 2639 /** 2640 * d_splice_alias - splice a disconnected dentry into the tree if one exists 2641 * @inode: the inode which may have a disconnected dentry 2642 * @dentry: a negative dentry which we want to point to the inode. 2643 * 2644 * If inode is a directory and has an IS_ROOT alias, then d_move that in 2645 * place of the given dentry and return it, else simply d_add the inode 2646 * to the dentry and return NULL. 2647 * 2648 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and 2649 * we should error out: directories can't have multiple aliases. 2650 * 2651 * This is needed in the lookup routine of any filesystem that is exportable 2652 * (via knfsd) so that we can build dcache paths to directories effectively. 2653 * 2654 * If a dentry was found and moved, then it is returned. Otherwise NULL 2655 * is returned. This matches the expected return value of ->lookup. 2656 * 2657 * Cluster filesystems may call this function with a negative, hashed dentry. 2658 * In that case, we know that the inode will be a regular file, and also this 2659 * will only occur during atomic_open. So we need to check for the dentry 2660 * being already hashed only in the final case. 2661 */ 2662 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) 2663 { 2664 struct dentry *new = NULL; 2665 2666 if (IS_ERR(inode)) 2667 return ERR_CAST(inode); 2668 2669 if (inode && S_ISDIR(inode->i_mode)) { 2670 spin_lock(&inode->i_lock); 2671 new = __d_find_any_alias(inode); 2672 if (new) { 2673 if (!IS_ROOT(new)) { 2674 spin_unlock(&inode->i_lock); 2675 dput(new); 2676 iput(inode); 2677 return ERR_PTR(-EIO); 2678 } 2679 if (d_ancestor(new, dentry)) { 2680 spin_unlock(&inode->i_lock); 2681 dput(new); 2682 iput(inode); 2683 return ERR_PTR(-EIO); 2684 } 2685 write_seqlock(&rename_lock); 2686 __d_move(new, dentry, false); 2687 write_sequnlock(&rename_lock); 2688 spin_unlock(&inode->i_lock); 2689 security_d_instantiate(new, inode); 2690 iput(inode); 2691 } else { 2692 /* already taking inode->i_lock, so d_add() by hand */ 2693 __d_instantiate(dentry, inode); 2694 spin_unlock(&inode->i_lock); 2695 security_d_instantiate(dentry, inode); 2696 d_rehash(dentry); 2697 } 2698 } else { 2699 d_instantiate(dentry, inode); 2700 if (d_unhashed(dentry)) 2701 d_rehash(dentry); 2702 } 2703 return new; 2704 } 2705 EXPORT_SYMBOL(d_splice_alias); 2706 2707 /** 2708 * d_materialise_unique - introduce an inode into the tree 2709 * @dentry: candidate dentry 2710 * @inode: inode to bind to the dentry, to which aliases may be attached 2711 * 2712 * Introduces an dentry into the tree, substituting an extant disconnected 2713 * root directory alias in its place if there is one. Caller must hold the 2714 * i_mutex of the parent directory. 2715 */ 2716 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) 2717 { 2718 struct dentry *actual; 2719 2720 BUG_ON(!d_unhashed(dentry)); 2721 2722 if (!inode) { 2723 actual = dentry; 2724 __d_instantiate(dentry, NULL); 2725 d_rehash(actual); 2726 goto out_nolock; 2727 } 2728 2729 spin_lock(&inode->i_lock); 2730 2731 if (S_ISDIR(inode->i_mode)) { 2732 struct dentry *alias; 2733 2734 /* Does an aliased dentry already exist? */ 2735 alias = __d_find_alias(inode); 2736 if (alias) { 2737 actual = alias; 2738 write_seqlock(&rename_lock); 2739 2740 if (d_ancestor(alias, dentry)) { 2741 /* Check for loops */ 2742 actual = ERR_PTR(-ELOOP); 2743 spin_unlock(&inode->i_lock); 2744 } else if (IS_ROOT(alias)) { 2745 /* Is this an anonymous mountpoint that we 2746 * could splice into our tree? */ 2747 __d_move(alias, dentry, false); 2748 write_sequnlock(&rename_lock); 2749 goto found; 2750 } else { 2751 /* Nope, but we must(!) avoid directory 2752 * aliasing. This drops inode->i_lock */ 2753 actual = __d_unalias(inode, dentry, alias); 2754 } 2755 write_sequnlock(&rename_lock); 2756 if (IS_ERR(actual)) { 2757 if (PTR_ERR(actual) == -ELOOP) 2758 pr_warn_ratelimited( 2759 "VFS: Lookup of '%s' in %s %s" 2760 " would have caused loop\n", 2761 dentry->d_name.name, 2762 inode->i_sb->s_type->name, 2763 inode->i_sb->s_id); 2764 dput(alias); 2765 } 2766 goto out_nolock; 2767 } 2768 } 2769 2770 /* Add a unique reference */ 2771 actual = __d_instantiate_unique(dentry, inode); 2772 if (!actual) 2773 actual = dentry; 2774 2775 d_rehash(actual); 2776 found: 2777 spin_unlock(&inode->i_lock); 2778 out_nolock: 2779 if (actual == dentry) { 2780 security_d_instantiate(dentry, inode); 2781 return NULL; 2782 } 2783 2784 iput(inode); 2785 return actual; 2786 } 2787 EXPORT_SYMBOL_GPL(d_materialise_unique); 2788 2789 static int prepend(char **buffer, int *buflen, const char *str, int namelen) 2790 { 2791 *buflen -= namelen; 2792 if (*buflen < 0) 2793 return -ENAMETOOLONG; 2794 *buffer -= namelen; 2795 memcpy(*buffer, str, namelen); 2796 return 0; 2797 } 2798 2799 /** 2800 * prepend_name - prepend a pathname in front of current buffer pointer 2801 * @buffer: buffer pointer 2802 * @buflen: allocated length of the buffer 2803 * @name: name string and length qstr structure 2804 * 2805 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to 2806 * make sure that either the old or the new name pointer and length are 2807 * fetched. However, there may be mismatch between length and pointer. 2808 * The length cannot be trusted, we need to copy it byte-by-byte until 2809 * the length is reached or a null byte is found. It also prepends "/" at 2810 * the beginning of the name. The sequence number check at the caller will 2811 * retry it again when a d_move() does happen. So any garbage in the buffer 2812 * due to mismatched pointer and length will be discarded. 2813 * 2814 * Data dependency barrier is needed to make sure that we see that terminating 2815 * NUL. Alpha strikes again, film at 11... 2816 */ 2817 static int prepend_name(char **buffer, int *buflen, struct qstr *name) 2818 { 2819 const char *dname = ACCESS_ONCE(name->name); 2820 u32 dlen = ACCESS_ONCE(name->len); 2821 char *p; 2822 2823 smp_read_barrier_depends(); 2824 2825 *buflen -= dlen + 1; 2826 if (*buflen < 0) 2827 return -ENAMETOOLONG; 2828 p = *buffer -= dlen + 1; 2829 *p++ = '/'; 2830 while (dlen--) { 2831 char c = *dname++; 2832 if (!c) 2833 break; 2834 *p++ = c; 2835 } 2836 return 0; 2837 } 2838 2839 /** 2840 * prepend_path - Prepend path string to a buffer 2841 * @path: the dentry/vfsmount to report 2842 * @root: root vfsmnt/dentry 2843 * @buffer: pointer to the end of the buffer 2844 * @buflen: pointer to buffer length 2845 * 2846 * The function will first try to write out the pathname without taking any 2847 * lock other than the RCU read lock to make sure that dentries won't go away. 2848 * It only checks the sequence number of the global rename_lock as any change 2849 * in the dentry's d_seq will be preceded by changes in the rename_lock 2850 * sequence number. If the sequence number had been changed, it will restart 2851 * the whole pathname back-tracing sequence again by taking the rename_lock. 2852 * In this case, there is no need to take the RCU read lock as the recursive 2853 * parent pointer references will keep the dentry chain alive as long as no 2854 * rename operation is performed. 2855 */ 2856 static int prepend_path(const struct path *path, 2857 const struct path *root, 2858 char **buffer, int *buflen) 2859 { 2860 struct dentry *dentry; 2861 struct vfsmount *vfsmnt; 2862 struct mount *mnt; 2863 int error = 0; 2864 unsigned seq, m_seq = 0; 2865 char *bptr; 2866 int blen; 2867 2868 rcu_read_lock(); 2869 restart_mnt: 2870 read_seqbegin_or_lock(&mount_lock, &m_seq); 2871 seq = 0; 2872 rcu_read_lock(); 2873 restart: 2874 bptr = *buffer; 2875 blen = *buflen; 2876 error = 0; 2877 dentry = path->dentry; 2878 vfsmnt = path->mnt; 2879 mnt = real_mount(vfsmnt); 2880 read_seqbegin_or_lock(&rename_lock, &seq); 2881 while (dentry != root->dentry || vfsmnt != root->mnt) { 2882 struct dentry * parent; 2883 2884 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { 2885 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); 2886 /* Global root? */ 2887 if (mnt != parent) { 2888 dentry = ACCESS_ONCE(mnt->mnt_mountpoint); 2889 mnt = parent; 2890 vfsmnt = &mnt->mnt; 2891 continue; 2892 } 2893 /* 2894 * Filesystems needing to implement special "root names" 2895 * should do so with ->d_dname() 2896 */ 2897 if (IS_ROOT(dentry) && 2898 (dentry->d_name.len != 1 || 2899 dentry->d_name.name[0] != '/')) { 2900 WARN(1, "Root dentry has weird name <%.*s>\n", 2901 (int) dentry->d_name.len, 2902 dentry->d_name.name); 2903 } 2904 if (!error) 2905 error = is_mounted(vfsmnt) ? 1 : 2; 2906 break; 2907 } 2908 parent = dentry->d_parent; 2909 prefetch(parent); 2910 error = prepend_name(&bptr, &blen, &dentry->d_name); 2911 if (error) 2912 break; 2913 2914 dentry = parent; 2915 } 2916 if (!(seq & 1)) 2917 rcu_read_unlock(); 2918 if (need_seqretry(&rename_lock, seq)) { 2919 seq = 1; 2920 goto restart; 2921 } 2922 done_seqretry(&rename_lock, seq); 2923 2924 if (!(m_seq & 1)) 2925 rcu_read_unlock(); 2926 if (need_seqretry(&mount_lock, m_seq)) { 2927 m_seq = 1; 2928 goto restart_mnt; 2929 } 2930 done_seqretry(&mount_lock, m_seq); 2931 2932 if (error >= 0 && bptr == *buffer) { 2933 if (--blen < 0) 2934 error = -ENAMETOOLONG; 2935 else 2936 *--bptr = '/'; 2937 } 2938 *buffer = bptr; 2939 *buflen = blen; 2940 return error; 2941 } 2942 2943 /** 2944 * __d_path - return the path of a dentry 2945 * @path: the dentry/vfsmount to report 2946 * @root: root vfsmnt/dentry 2947 * @buf: buffer to return value in 2948 * @buflen: buffer length 2949 * 2950 * Convert a dentry into an ASCII path name. 2951 * 2952 * Returns a pointer into the buffer or an error code if the 2953 * path was too long. 2954 * 2955 * "buflen" should be positive. 2956 * 2957 * If the path is not reachable from the supplied root, return %NULL. 2958 */ 2959 char *__d_path(const struct path *path, 2960 const struct path *root, 2961 char *buf, int buflen) 2962 { 2963 char *res = buf + buflen; 2964 int error; 2965 2966 prepend(&res, &buflen, "\0", 1); 2967 error = prepend_path(path, root, &res, &buflen); 2968 2969 if (error < 0) 2970 return ERR_PTR(error); 2971 if (error > 0) 2972 return NULL; 2973 return res; 2974 } 2975 2976 char *d_absolute_path(const struct path *path, 2977 char *buf, int buflen) 2978 { 2979 struct path root = {}; 2980 char *res = buf + buflen; 2981 int error; 2982 2983 prepend(&res, &buflen, "\0", 1); 2984 error = prepend_path(path, &root, &res, &buflen); 2985 2986 if (error > 1) 2987 error = -EINVAL; 2988 if (error < 0) 2989 return ERR_PTR(error); 2990 return res; 2991 } 2992 2993 /* 2994 * same as __d_path but appends "(deleted)" for unlinked files. 2995 */ 2996 static int path_with_deleted(const struct path *path, 2997 const struct path *root, 2998 char **buf, int *buflen) 2999 { 3000 prepend(buf, buflen, "\0", 1); 3001 if (d_unlinked(path->dentry)) { 3002 int error = prepend(buf, buflen, " (deleted)", 10); 3003 if (error) 3004 return error; 3005 } 3006 3007 return prepend_path(path, root, buf, buflen); 3008 } 3009 3010 static int prepend_unreachable(char **buffer, int *buflen) 3011 { 3012 return prepend(buffer, buflen, "(unreachable)", 13); 3013 } 3014 3015 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) 3016 { 3017 unsigned seq; 3018 3019 do { 3020 seq = read_seqcount_begin(&fs->seq); 3021 *root = fs->root; 3022 } while (read_seqcount_retry(&fs->seq, seq)); 3023 } 3024 3025 /** 3026 * d_path - return the path of a dentry 3027 * @path: path to report 3028 * @buf: buffer to return value in 3029 * @buflen: buffer length 3030 * 3031 * Convert a dentry into an ASCII path name. If the entry has been deleted 3032 * the string " (deleted)" is appended. Note that this is ambiguous. 3033 * 3034 * Returns a pointer into the buffer or an error code if the path was 3035 * too long. Note: Callers should use the returned pointer, not the passed 3036 * in buffer, to use the name! The implementation often starts at an offset 3037 * into the buffer, and may leave 0 bytes at the start. 3038 * 3039 * "buflen" should be positive. 3040 */ 3041 char *d_path(const struct path *path, char *buf, int buflen) 3042 { 3043 char *res = buf + buflen; 3044 struct path root; 3045 int error; 3046 3047 /* 3048 * We have various synthetic filesystems that never get mounted. On 3049 * these filesystems dentries are never used for lookup purposes, and 3050 * thus don't need to be hashed. They also don't need a name until a 3051 * user wants to identify the object in /proc/pid/fd/. The little hack 3052 * below allows us to generate a name for these objects on demand: 3053 * 3054 * Some pseudo inodes are mountable. When they are mounted 3055 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname 3056 * and instead have d_path return the mounted path. 3057 */ 3058 if (path->dentry->d_op && path->dentry->d_op->d_dname && 3059 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) 3060 return path->dentry->d_op->d_dname(path->dentry, buf, buflen); 3061 3062 rcu_read_lock(); 3063 get_fs_root_rcu(current->fs, &root); 3064 error = path_with_deleted(path, &root, &res, &buflen); 3065 rcu_read_unlock(); 3066 3067 if (error < 0) 3068 res = ERR_PTR(error); 3069 return res; 3070 } 3071 EXPORT_SYMBOL(d_path); 3072 3073 /* 3074 * Helper function for dentry_operations.d_dname() members 3075 */ 3076 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, 3077 const char *fmt, ...) 3078 { 3079 va_list args; 3080 char temp[64]; 3081 int sz; 3082 3083 va_start(args, fmt); 3084 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; 3085 va_end(args); 3086 3087 if (sz > sizeof(temp) || sz > buflen) 3088 return ERR_PTR(-ENAMETOOLONG); 3089 3090 buffer += buflen - sz; 3091 return memcpy(buffer, temp, sz); 3092 } 3093 3094 char *simple_dname(struct dentry *dentry, char *buffer, int buflen) 3095 { 3096 char *end = buffer + buflen; 3097 /* these dentries are never renamed, so d_lock is not needed */ 3098 if (prepend(&end, &buflen, " (deleted)", 11) || 3099 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || 3100 prepend(&end, &buflen, "/", 1)) 3101 end = ERR_PTR(-ENAMETOOLONG); 3102 return end; 3103 } 3104 EXPORT_SYMBOL(simple_dname); 3105 3106 /* 3107 * Write full pathname from the root of the filesystem into the buffer. 3108 */ 3109 static char *__dentry_path(struct dentry *d, char *buf, int buflen) 3110 { 3111 struct dentry *dentry; 3112 char *end, *retval; 3113 int len, seq = 0; 3114 int error = 0; 3115 3116 if (buflen < 2) 3117 goto Elong; 3118 3119 rcu_read_lock(); 3120 restart: 3121 dentry = d; 3122 end = buf + buflen; 3123 len = buflen; 3124 prepend(&end, &len, "\0", 1); 3125 /* Get '/' right */ 3126 retval = end-1; 3127 *retval = '/'; 3128 read_seqbegin_or_lock(&rename_lock, &seq); 3129 while (!IS_ROOT(dentry)) { 3130 struct dentry *parent = dentry->d_parent; 3131 3132 prefetch(parent); 3133 error = prepend_name(&end, &len, &dentry->d_name); 3134 if (error) 3135 break; 3136 3137 retval = end; 3138 dentry = parent; 3139 } 3140 if (!(seq & 1)) 3141 rcu_read_unlock(); 3142 if (need_seqretry(&rename_lock, seq)) { 3143 seq = 1; 3144 goto restart; 3145 } 3146 done_seqretry(&rename_lock, seq); 3147 if (error) 3148 goto Elong; 3149 return retval; 3150 Elong: 3151 return ERR_PTR(-ENAMETOOLONG); 3152 } 3153 3154 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) 3155 { 3156 return __dentry_path(dentry, buf, buflen); 3157 } 3158 EXPORT_SYMBOL(dentry_path_raw); 3159 3160 char *dentry_path(struct dentry *dentry, char *buf, int buflen) 3161 { 3162 char *p = NULL; 3163 char *retval; 3164 3165 if (d_unlinked(dentry)) { 3166 p = buf + buflen; 3167 if (prepend(&p, &buflen, "//deleted", 10) != 0) 3168 goto Elong; 3169 buflen++; 3170 } 3171 retval = __dentry_path(dentry, buf, buflen); 3172 if (!IS_ERR(retval) && p) 3173 *p = '/'; /* restore '/' overriden with '\0' */ 3174 return retval; 3175 Elong: 3176 return ERR_PTR(-ENAMETOOLONG); 3177 } 3178 3179 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, 3180 struct path *pwd) 3181 { 3182 unsigned seq; 3183 3184 do { 3185 seq = read_seqcount_begin(&fs->seq); 3186 *root = fs->root; 3187 *pwd = fs->pwd; 3188 } while (read_seqcount_retry(&fs->seq, seq)); 3189 } 3190 3191 /* 3192 * NOTE! The user-level library version returns a 3193 * character pointer. The kernel system call just 3194 * returns the length of the buffer filled (which 3195 * includes the ending '\0' character), or a negative 3196 * error value. So libc would do something like 3197 * 3198 * char *getcwd(char * buf, size_t size) 3199 * { 3200 * int retval; 3201 * 3202 * retval = sys_getcwd(buf, size); 3203 * if (retval >= 0) 3204 * return buf; 3205 * errno = -retval; 3206 * return NULL; 3207 * } 3208 */ 3209 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) 3210 { 3211 int error; 3212 struct path pwd, root; 3213 char *page = __getname(); 3214 3215 if (!page) 3216 return -ENOMEM; 3217 3218 rcu_read_lock(); 3219 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); 3220 3221 error = -ENOENT; 3222 if (!d_unlinked(pwd.dentry)) { 3223 unsigned long len; 3224 char *cwd = page + PATH_MAX; 3225 int buflen = PATH_MAX; 3226 3227 prepend(&cwd, &buflen, "\0", 1); 3228 error = prepend_path(&pwd, &root, &cwd, &buflen); 3229 rcu_read_unlock(); 3230 3231 if (error < 0) 3232 goto out; 3233 3234 /* Unreachable from current root */ 3235 if (error > 0) { 3236 error = prepend_unreachable(&cwd, &buflen); 3237 if (error) 3238 goto out; 3239 } 3240 3241 error = -ERANGE; 3242 len = PATH_MAX + page - cwd; 3243 if (len <= size) { 3244 error = len; 3245 if (copy_to_user(buf, cwd, len)) 3246 error = -EFAULT; 3247 } 3248 } else { 3249 rcu_read_unlock(); 3250 } 3251 3252 out: 3253 __putname(page); 3254 return error; 3255 } 3256 3257 /* 3258 * Test whether new_dentry is a subdirectory of old_dentry. 3259 * 3260 * Trivially implemented using the dcache structure 3261 */ 3262 3263 /** 3264 * is_subdir - is new dentry a subdirectory of old_dentry 3265 * @new_dentry: new dentry 3266 * @old_dentry: old dentry 3267 * 3268 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). 3269 * Returns 0 otherwise. 3270 * Caller must ensure that "new_dentry" is pinned before calling is_subdir() 3271 */ 3272 3273 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) 3274 { 3275 int result; 3276 unsigned seq; 3277 3278 if (new_dentry == old_dentry) 3279 return 1; 3280 3281 do { 3282 /* for restarting inner loop in case of seq retry */ 3283 seq = read_seqbegin(&rename_lock); 3284 /* 3285 * Need rcu_readlock to protect against the d_parent trashing 3286 * due to d_move 3287 */ 3288 rcu_read_lock(); 3289 if (d_ancestor(old_dentry, new_dentry)) 3290 result = 1; 3291 else 3292 result = 0; 3293 rcu_read_unlock(); 3294 } while (read_seqretry(&rename_lock, seq)); 3295 3296 return result; 3297 } 3298 3299 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) 3300 { 3301 struct dentry *root = data; 3302 if (dentry != root) { 3303 if (d_unhashed(dentry) || !dentry->d_inode) 3304 return D_WALK_SKIP; 3305 3306 if (!(dentry->d_flags & DCACHE_GENOCIDE)) { 3307 dentry->d_flags |= DCACHE_GENOCIDE; 3308 dentry->d_lockref.count--; 3309 } 3310 } 3311 return D_WALK_CONTINUE; 3312 } 3313 3314 void d_genocide(struct dentry *parent) 3315 { 3316 d_walk(parent, parent, d_genocide_kill, NULL); 3317 } 3318 3319 void d_tmpfile(struct dentry *dentry, struct inode *inode) 3320 { 3321 inode_dec_link_count(inode); 3322 BUG_ON(dentry->d_name.name != dentry->d_iname || 3323 !hlist_unhashed(&dentry->d_alias) || 3324 !d_unlinked(dentry)); 3325 spin_lock(&dentry->d_parent->d_lock); 3326 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 3327 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", 3328 (unsigned long long)inode->i_ino); 3329 spin_unlock(&dentry->d_lock); 3330 spin_unlock(&dentry->d_parent->d_lock); 3331 d_instantiate(dentry, inode); 3332 } 3333 EXPORT_SYMBOL(d_tmpfile); 3334 3335 static __initdata unsigned long dhash_entries; 3336 static int __init set_dhash_entries(char *str) 3337 { 3338 if (!str) 3339 return 0; 3340 dhash_entries = simple_strtoul(str, &str, 0); 3341 return 1; 3342 } 3343 __setup("dhash_entries=", set_dhash_entries); 3344 3345 static void __init dcache_init_early(void) 3346 { 3347 unsigned int loop; 3348 3349 /* If hashes are distributed across NUMA nodes, defer 3350 * hash allocation until vmalloc space is available. 3351 */ 3352 if (hashdist) 3353 return; 3354 3355 dentry_hashtable = 3356 alloc_large_system_hash("Dentry cache", 3357 sizeof(struct hlist_bl_head), 3358 dhash_entries, 3359 13, 3360 HASH_EARLY, 3361 &d_hash_shift, 3362 &d_hash_mask, 3363 0, 3364 0); 3365 3366 for (loop = 0; loop < (1U << d_hash_shift); loop++) 3367 INIT_HLIST_BL_HEAD(dentry_hashtable + loop); 3368 } 3369 3370 static void __init dcache_init(void) 3371 { 3372 unsigned int loop; 3373 3374 /* 3375 * A constructor could be added for stable state like the lists, 3376 * but it is probably not worth it because of the cache nature 3377 * of the dcache. 3378 */ 3379 dentry_cache = KMEM_CACHE(dentry, 3380 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); 3381 3382 /* Hash may have been set up in dcache_init_early */ 3383 if (!hashdist) 3384 return; 3385 3386 dentry_hashtable = 3387 alloc_large_system_hash("Dentry cache", 3388 sizeof(struct hlist_bl_head), 3389 dhash_entries, 3390 13, 3391 0, 3392 &d_hash_shift, 3393 &d_hash_mask, 3394 0, 3395 0); 3396 3397 for (loop = 0; loop < (1U << d_hash_shift); loop++) 3398 INIT_HLIST_BL_HEAD(dentry_hashtable + loop); 3399 } 3400 3401 /* SLAB cache for __getname() consumers */ 3402 struct kmem_cache *names_cachep __read_mostly; 3403 EXPORT_SYMBOL(names_cachep); 3404 3405 EXPORT_SYMBOL(d_genocide); 3406 3407 void __init vfs_caches_init_early(void) 3408 { 3409 dcache_init_early(); 3410 inode_init_early(); 3411 } 3412 3413 void __init vfs_caches_init(unsigned long mempages) 3414 { 3415 unsigned long reserve; 3416 3417 /* Base hash sizes on available memory, with a reserve equal to 3418 150% of current kernel size */ 3419 3420 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); 3421 mempages -= reserve; 3422 3423 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, 3424 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 3425 3426 dcache_init(); 3427 inode_init(); 3428 files_init(mempages); 3429 mnt_init(); 3430 bdev_cache_init(); 3431 chrdev_init(); 3432 } 3433