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