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