| 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/module.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 "internal.h" |
| 40 | |
| 41 | /* |
| 42 | * Usage: |
| 43 | * dcache->d_inode->i_lock protects: |
| 44 | * - i_dentry, d_alias, d_inode of aliases |
| 45 | * dcache_hash_bucket lock protects: |
| 46 | * - the dcache hash table |
| 47 | * s_anon bl list spinlock protects: |
| 48 | * - the s_anon list (see __d_drop) |
| 49 | * dcache_lru_lock protects: |
| 50 | * - the dcache lru lists and counters |
| 51 | * d_lock protects: |
| 52 | * - d_flags |
| 53 | * - d_name |
| 54 | * - d_lru |
| 55 | * - d_count |
| 56 | * - d_unhashed() |
| 57 | * - d_parent and d_subdirs |
| 58 | * - childrens' d_child and d_parent |
| 59 | * - d_alias, d_inode |
| 60 | * |
| 61 | * Ordering: |
| 62 | * dentry->d_inode->i_lock |
| 63 | * dentry->d_lock |
| 64 | * dcache_lru_lock |
| 65 | * dcache_hash_bucket lock |
| 66 | * s_anon lock |
| 67 | * |
| 68 | * If there is an ancestor relationship: |
| 69 | * dentry->d_parent->...->d_parent->d_lock |
| 70 | * ... |
| 71 | * dentry->d_parent->d_lock |
| 72 | * dentry->d_lock |
| 73 | * |
| 74 | * If no ancestor relationship: |
| 75 | * if (dentry1 < dentry2) |
| 76 | * dentry1->d_lock |
| 77 | * dentry2->d_lock |
| 78 | */ |
| 79 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
| 80 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
| 81 | |
| 82 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock); |
| 83 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
| 84 | |
| 85 | EXPORT_SYMBOL(rename_lock); |
| 86 | |
| 87 | static struct kmem_cache *dentry_cache __read_mostly; |
| 88 | |
| 89 | /* |
| 90 | * This is the single most critical data structure when it comes |
| 91 | * to the dcache: the hashtable for lookups. Somebody should try |
| 92 | * to make this good - I've just made it work. |
| 93 | * |
| 94 | * This hash-function tries to avoid losing too many bits of hash |
| 95 | * information, yet avoid using a prime hash-size or similar. |
| 96 | */ |
| 97 | #define D_HASHBITS d_hash_shift |
| 98 | #define D_HASHMASK d_hash_mask |
| 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(struct dentry *parent, |
| 106 | unsigned long hash) |
| 107 | { |
| 108 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; |
| 109 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); |
| 110 | return dentry_hashtable + (hash & D_HASHMASK); |
| 111 | } |
| 112 | |
| 113 | /* Statistics gathering. */ |
| 114 | struct dentry_stat_t dentry_stat = { |
| 115 | .age_limit = 45, |
| 116 | }; |
| 117 | |
| 118 | static DEFINE_PER_CPU(unsigned int, nr_dentry); |
| 119 | |
| 120 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) |
| 121 | static int get_nr_dentry(void) |
| 122 | { |
| 123 | int i; |
| 124 | int sum = 0; |
| 125 | for_each_possible_cpu(i) |
| 126 | sum += per_cpu(nr_dentry, i); |
| 127 | return sum < 0 ? 0 : sum; |
| 128 | } |
| 129 | |
| 130 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
| 131 | size_t *lenp, loff_t *ppos) |
| 132 | { |
| 133 | dentry_stat.nr_dentry = get_nr_dentry(); |
| 134 | return proc_dointvec(table, write, buffer, lenp, ppos); |
| 135 | } |
| 136 | #endif |
| 137 | |
| 138 | static void __d_free(struct rcu_head *head) |
| 139 | { |
| 140 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
| 141 | |
| 142 | WARN_ON(!list_empty(&dentry->d_alias)); |
| 143 | if (dname_external(dentry)) |
| 144 | kfree(dentry->d_name.name); |
| 145 | kmem_cache_free(dentry_cache, dentry); |
| 146 | } |
| 147 | |
| 148 | /* |
| 149 | * no locks, please. |
| 150 | */ |
| 151 | static void d_free(struct dentry *dentry) |
| 152 | { |
| 153 | BUG_ON(dentry->d_count); |
| 154 | this_cpu_dec(nr_dentry); |
| 155 | if (dentry->d_op && dentry->d_op->d_release) |
| 156 | dentry->d_op->d_release(dentry); |
| 157 | |
| 158 | /* if dentry was never visible to RCU, immediate free is OK */ |
| 159 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) |
| 160 | __d_free(&dentry->d_u.d_rcu); |
| 161 | else |
| 162 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
| 163 | } |
| 164 | |
| 165 | /** |
| 166 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups |
| 167 | * @dentry: the target dentry |
| 168 | * After this call, in-progress rcu-walk path lookup will fail. This |
| 169 | * should be called after unhashing, and after changing d_inode (if |
| 170 | * the dentry has not already been unhashed). |
| 171 | */ |
| 172 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) |
| 173 | { |
| 174 | assert_spin_locked(&dentry->d_lock); |
| 175 | /* Go through a barrier */ |
| 176 | write_seqcount_barrier(&dentry->d_seq); |
| 177 | } |
| 178 | |
| 179 | /* |
| 180 | * Release the dentry's inode, using the filesystem |
| 181 | * d_iput() operation if defined. Dentry has no refcount |
| 182 | * and is unhashed. |
| 183 | */ |
| 184 | static void dentry_iput(struct dentry * dentry) |
| 185 | __releases(dentry->d_lock) |
| 186 | __releases(dentry->d_inode->i_lock) |
| 187 | { |
| 188 | struct inode *inode = dentry->d_inode; |
| 189 | if (inode) { |
| 190 | dentry->d_inode = NULL; |
| 191 | list_del_init(&dentry->d_alias); |
| 192 | spin_unlock(&dentry->d_lock); |
| 193 | spin_unlock(&inode->i_lock); |
| 194 | if (!inode->i_nlink) |
| 195 | fsnotify_inoderemove(inode); |
| 196 | if (dentry->d_op && dentry->d_op->d_iput) |
| 197 | dentry->d_op->d_iput(dentry, inode); |
| 198 | else |
| 199 | iput(inode); |
| 200 | } else { |
| 201 | spin_unlock(&dentry->d_lock); |
| 202 | } |
| 203 | } |
| 204 | |
| 205 | /* |
| 206 | * Release the dentry's inode, using the filesystem |
| 207 | * d_iput() operation if defined. dentry remains in-use. |
| 208 | */ |
| 209 | static void dentry_unlink_inode(struct dentry * dentry) |
| 210 | __releases(dentry->d_lock) |
| 211 | __releases(dentry->d_inode->i_lock) |
| 212 | { |
| 213 | struct inode *inode = dentry->d_inode; |
| 214 | dentry->d_inode = NULL; |
| 215 | list_del_init(&dentry->d_alias); |
| 216 | dentry_rcuwalk_barrier(dentry); |
| 217 | spin_unlock(&dentry->d_lock); |
| 218 | spin_unlock(&inode->i_lock); |
| 219 | if (!inode->i_nlink) |
| 220 | fsnotify_inoderemove(inode); |
| 221 | if (dentry->d_op && dentry->d_op->d_iput) |
| 222 | dentry->d_op->d_iput(dentry, inode); |
| 223 | else |
| 224 | iput(inode); |
| 225 | } |
| 226 | |
| 227 | /* |
| 228 | * dentry_lru_(add|del|move_tail) must be called with d_lock held. |
| 229 | */ |
| 230 | static void dentry_lru_add(struct dentry *dentry) |
| 231 | { |
| 232 | if (list_empty(&dentry->d_lru)) { |
| 233 | spin_lock(&dcache_lru_lock); |
| 234 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
| 235 | dentry->d_sb->s_nr_dentry_unused++; |
| 236 | dentry_stat.nr_unused++; |
| 237 | spin_unlock(&dcache_lru_lock); |
| 238 | } |
| 239 | } |
| 240 | |
| 241 | static void __dentry_lru_del(struct dentry *dentry) |
| 242 | { |
| 243 | list_del_init(&dentry->d_lru); |
| 244 | dentry->d_sb->s_nr_dentry_unused--; |
| 245 | dentry_stat.nr_unused--; |
| 246 | } |
| 247 | |
| 248 | static void dentry_lru_del(struct dentry *dentry) |
| 249 | { |
| 250 | if (!list_empty(&dentry->d_lru)) { |
| 251 | spin_lock(&dcache_lru_lock); |
| 252 | __dentry_lru_del(dentry); |
| 253 | spin_unlock(&dcache_lru_lock); |
| 254 | } |
| 255 | } |
| 256 | |
| 257 | static void dentry_lru_move_tail(struct dentry *dentry) |
| 258 | { |
| 259 | spin_lock(&dcache_lru_lock); |
| 260 | if (list_empty(&dentry->d_lru)) { |
| 261 | list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
| 262 | dentry->d_sb->s_nr_dentry_unused++; |
| 263 | dentry_stat.nr_unused++; |
| 264 | } else { |
| 265 | list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
| 266 | } |
| 267 | spin_unlock(&dcache_lru_lock); |
| 268 | } |
| 269 | |
| 270 | /** |
| 271 | * d_kill - kill dentry and return parent |
| 272 | * @dentry: dentry to kill |
| 273 | * @parent: parent dentry |
| 274 | * |
| 275 | * The dentry must already be unhashed and removed from the LRU. |
| 276 | * |
| 277 | * If this is the root of the dentry tree, return NULL. |
| 278 | * |
| 279 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
| 280 | * d_kill. |
| 281 | */ |
| 282 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
| 283 | __releases(dentry->d_lock) |
| 284 | __releases(parent->d_lock) |
| 285 | __releases(dentry->d_inode->i_lock) |
| 286 | { |
| 287 | list_del(&dentry->d_u.d_child); |
| 288 | /* |
| 289 | * Inform try_to_ascend() that we are no longer attached to the |
| 290 | * dentry tree |
| 291 | */ |
| 292 | dentry->d_flags |= DCACHE_DISCONNECTED; |
| 293 | if (parent) |
| 294 | spin_unlock(&parent->d_lock); |
| 295 | dentry_iput(dentry); |
| 296 | /* |
| 297 | * dentry_iput drops the locks, at which point nobody (except |
| 298 | * transient RCU lookups) can reach this dentry. |
| 299 | */ |
| 300 | d_free(dentry); |
| 301 | return parent; |
| 302 | } |
| 303 | |
| 304 | /** |
| 305 | * d_drop - drop a dentry |
| 306 | * @dentry: dentry to drop |
| 307 | * |
| 308 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't |
| 309 | * be found through a VFS lookup any more. Note that this is different from |
| 310 | * deleting the dentry - d_delete will try to mark the dentry negative if |
| 311 | * possible, giving a successful _negative_ lookup, while d_drop will |
| 312 | * just make the cache lookup fail. |
| 313 | * |
| 314 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some |
| 315 | * reason (NFS timeouts or autofs deletes). |
| 316 | * |
| 317 | * __d_drop requires dentry->d_lock. |
| 318 | */ |
| 319 | void __d_drop(struct dentry *dentry) |
| 320 | { |
| 321 | if (!d_unhashed(dentry)) { |
| 322 | struct hlist_bl_head *b; |
| 323 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) |
| 324 | b = &dentry->d_sb->s_anon; |
| 325 | else |
| 326 | b = d_hash(dentry->d_parent, dentry->d_name.hash); |
| 327 | |
| 328 | hlist_bl_lock(b); |
| 329 | __hlist_bl_del(&dentry->d_hash); |
| 330 | dentry->d_hash.pprev = NULL; |
| 331 | hlist_bl_unlock(b); |
| 332 | |
| 333 | dentry_rcuwalk_barrier(dentry); |
| 334 | } |
| 335 | } |
| 336 | EXPORT_SYMBOL(__d_drop); |
| 337 | |
| 338 | void d_drop(struct dentry *dentry) |
| 339 | { |
| 340 | spin_lock(&dentry->d_lock); |
| 341 | __d_drop(dentry); |
| 342 | spin_unlock(&dentry->d_lock); |
| 343 | } |
| 344 | EXPORT_SYMBOL(d_drop); |
| 345 | |
| 346 | /* |
| 347 | * Finish off a dentry we've decided to kill. |
| 348 | * dentry->d_lock must be held, returns with it unlocked. |
| 349 | * If ref is non-zero, then decrement the refcount too. |
| 350 | * Returns dentry requiring refcount drop, or NULL if we're done. |
| 351 | */ |
| 352 | static inline struct dentry *dentry_kill(struct dentry *dentry, int ref) |
| 353 | __releases(dentry->d_lock) |
| 354 | { |
| 355 | struct inode *inode; |
| 356 | struct dentry *parent; |
| 357 | |
| 358 | inode = dentry->d_inode; |
| 359 | if (inode && !spin_trylock(&inode->i_lock)) { |
| 360 | relock: |
| 361 | spin_unlock(&dentry->d_lock); |
| 362 | cpu_relax(); |
| 363 | return dentry; /* try again with same dentry */ |
| 364 | } |
| 365 | if (IS_ROOT(dentry)) |
| 366 | parent = NULL; |
| 367 | else |
| 368 | parent = dentry->d_parent; |
| 369 | if (parent && !spin_trylock(&parent->d_lock)) { |
| 370 | if (inode) |
| 371 | spin_unlock(&inode->i_lock); |
| 372 | goto relock; |
| 373 | } |
| 374 | |
| 375 | if (ref) |
| 376 | dentry->d_count--; |
| 377 | /* if dentry was on the d_lru list delete it from there */ |
| 378 | dentry_lru_del(dentry); |
| 379 | /* if it was on the hash then remove it */ |
| 380 | __d_drop(dentry); |
| 381 | return d_kill(dentry, parent); |
| 382 | } |
| 383 | |
| 384 | /* |
| 385 | * This is dput |
| 386 | * |
| 387 | * This is complicated by the fact that we do not want to put |
| 388 | * dentries that are no longer on any hash chain on the unused |
| 389 | * list: we'd much rather just get rid of them immediately. |
| 390 | * |
| 391 | * However, that implies that we have to traverse the dentry |
| 392 | * tree upwards to the parents which might _also_ now be |
| 393 | * scheduled for deletion (it may have been only waiting for |
| 394 | * its last child to go away). |
| 395 | * |
| 396 | * This tail recursion is done by hand as we don't want to depend |
| 397 | * on the compiler to always get this right (gcc generally doesn't). |
| 398 | * Real recursion would eat up our stack space. |
| 399 | */ |
| 400 | |
| 401 | /* |
| 402 | * dput - release a dentry |
| 403 | * @dentry: dentry to release |
| 404 | * |
| 405 | * Release a dentry. This will drop the usage count and if appropriate |
| 406 | * call the dentry unlink method as well as removing it from the queues and |
| 407 | * releasing its resources. If the parent dentries were scheduled for release |
| 408 | * they too may now get deleted. |
| 409 | */ |
| 410 | void dput(struct dentry *dentry) |
| 411 | { |
| 412 | if (!dentry) |
| 413 | return; |
| 414 | |
| 415 | repeat: |
| 416 | if (dentry->d_count == 1) |
| 417 | might_sleep(); |
| 418 | spin_lock(&dentry->d_lock); |
| 419 | BUG_ON(!dentry->d_count); |
| 420 | if (dentry->d_count > 1) { |
| 421 | dentry->d_count--; |
| 422 | spin_unlock(&dentry->d_lock); |
| 423 | return; |
| 424 | } |
| 425 | |
| 426 | if (dentry->d_flags & DCACHE_OP_DELETE) { |
| 427 | if (dentry->d_op->d_delete(dentry)) |
| 428 | goto kill_it; |
| 429 | } |
| 430 | |
| 431 | /* Unreachable? Get rid of it */ |
| 432 | if (d_unhashed(dentry)) |
| 433 | goto kill_it; |
| 434 | |
| 435 | /* Otherwise leave it cached and ensure it's on the LRU */ |
| 436 | dentry->d_flags |= DCACHE_REFERENCED; |
| 437 | dentry_lru_add(dentry); |
| 438 | |
| 439 | dentry->d_count--; |
| 440 | spin_unlock(&dentry->d_lock); |
| 441 | return; |
| 442 | |
| 443 | kill_it: |
| 444 | dentry = dentry_kill(dentry, 1); |
| 445 | if (dentry) |
| 446 | goto repeat; |
| 447 | } |
| 448 | EXPORT_SYMBOL(dput); |
| 449 | |
| 450 | /** |
| 451 | * d_invalidate - invalidate a dentry |
| 452 | * @dentry: dentry to invalidate |
| 453 | * |
| 454 | * Try to invalidate the dentry if it turns out to be |
| 455 | * possible. If there are other dentries that can be |
| 456 | * reached through this one we can't delete it and we |
| 457 | * return -EBUSY. On success we return 0. |
| 458 | * |
| 459 | * no dcache lock. |
| 460 | */ |
| 461 | |
| 462 | int d_invalidate(struct dentry * dentry) |
| 463 | { |
| 464 | /* |
| 465 | * If it's already been dropped, return OK. |
| 466 | */ |
| 467 | spin_lock(&dentry->d_lock); |
| 468 | if (d_unhashed(dentry)) { |
| 469 | spin_unlock(&dentry->d_lock); |
| 470 | return 0; |
| 471 | } |
| 472 | /* |
| 473 | * Check whether to do a partial shrink_dcache |
| 474 | * to get rid of unused child entries. |
| 475 | */ |
| 476 | if (!list_empty(&dentry->d_subdirs)) { |
| 477 | spin_unlock(&dentry->d_lock); |
| 478 | shrink_dcache_parent(dentry); |
| 479 | spin_lock(&dentry->d_lock); |
| 480 | } |
| 481 | |
| 482 | /* |
| 483 | * Somebody else still using it? |
| 484 | * |
| 485 | * If it's a directory, we can't drop it |
| 486 | * for fear of somebody re-populating it |
| 487 | * with children (even though dropping it |
| 488 | * would make it unreachable from the root, |
| 489 | * we might still populate it if it was a |
| 490 | * working directory or similar). |
| 491 | */ |
| 492 | if (dentry->d_count > 1) { |
| 493 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { |
| 494 | spin_unlock(&dentry->d_lock); |
| 495 | return -EBUSY; |
| 496 | } |
| 497 | } |
| 498 | |
| 499 | __d_drop(dentry); |
| 500 | spin_unlock(&dentry->d_lock); |
| 501 | return 0; |
| 502 | } |
| 503 | EXPORT_SYMBOL(d_invalidate); |
| 504 | |
| 505 | /* This must be called with d_lock held */ |
| 506 | static inline void __dget_dlock(struct dentry *dentry) |
| 507 | { |
| 508 | dentry->d_count++; |
| 509 | } |
| 510 | |
| 511 | static inline void __dget(struct dentry *dentry) |
| 512 | { |
| 513 | spin_lock(&dentry->d_lock); |
| 514 | __dget_dlock(dentry); |
| 515 | spin_unlock(&dentry->d_lock); |
| 516 | } |
| 517 | |
| 518 | struct dentry *dget_parent(struct dentry *dentry) |
| 519 | { |
| 520 | struct dentry *ret; |
| 521 | |
| 522 | repeat: |
| 523 | /* |
| 524 | * Don't need rcu_dereference because we re-check it was correct under |
| 525 | * the lock. |
| 526 | */ |
| 527 | rcu_read_lock(); |
| 528 | ret = dentry->d_parent; |
| 529 | if (!ret) { |
| 530 | rcu_read_unlock(); |
| 531 | goto out; |
| 532 | } |
| 533 | spin_lock(&ret->d_lock); |
| 534 | if (unlikely(ret != dentry->d_parent)) { |
| 535 | spin_unlock(&ret->d_lock); |
| 536 | rcu_read_unlock(); |
| 537 | goto repeat; |
| 538 | } |
| 539 | rcu_read_unlock(); |
| 540 | BUG_ON(!ret->d_count); |
| 541 | ret->d_count++; |
| 542 | spin_unlock(&ret->d_lock); |
| 543 | out: |
| 544 | return ret; |
| 545 | } |
| 546 | EXPORT_SYMBOL(dget_parent); |
| 547 | |
| 548 | /** |
| 549 | * d_find_alias - grab a hashed alias of inode |
| 550 | * @inode: inode in question |
| 551 | * @want_discon: flag, used by d_splice_alias, to request |
| 552 | * that only a DISCONNECTED alias be returned. |
| 553 | * |
| 554 | * If inode has a hashed alias, or is a directory and has any alias, |
| 555 | * acquire the reference to alias and return it. Otherwise return NULL. |
| 556 | * Notice that if inode is a directory there can be only one alias and |
| 557 | * it can be unhashed only if it has no children, or if it is the root |
| 558 | * of a filesystem. |
| 559 | * |
| 560 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
| 561 | * any other hashed alias over that one unless @want_discon is set, |
| 562 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
| 563 | */ |
| 564 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
| 565 | { |
| 566 | struct dentry *alias, *discon_alias; |
| 567 | |
| 568 | again: |
| 569 | discon_alias = NULL; |
| 570 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
| 571 | spin_lock(&alias->d_lock); |
| 572 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
| 573 | if (IS_ROOT(alias) && |
| 574 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
| 575 | discon_alias = alias; |
| 576 | } else if (!want_discon) { |
| 577 | __dget_dlock(alias); |
| 578 | spin_unlock(&alias->d_lock); |
| 579 | return alias; |
| 580 | } |
| 581 | } |
| 582 | spin_unlock(&alias->d_lock); |
| 583 | } |
| 584 | if (discon_alias) { |
| 585 | alias = discon_alias; |
| 586 | spin_lock(&alias->d_lock); |
| 587 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
| 588 | if (IS_ROOT(alias) && |
| 589 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
| 590 | __dget_dlock(alias); |
| 591 | spin_unlock(&alias->d_lock); |
| 592 | return alias; |
| 593 | } |
| 594 | } |
| 595 | spin_unlock(&alias->d_lock); |
| 596 | goto again; |
| 597 | } |
| 598 | return NULL; |
| 599 | } |
| 600 | |
| 601 | struct dentry *d_find_alias(struct inode *inode) |
| 602 | { |
| 603 | struct dentry *de = NULL; |
| 604 | |
| 605 | if (!list_empty(&inode->i_dentry)) { |
| 606 | spin_lock(&inode->i_lock); |
| 607 | de = __d_find_alias(inode, 0); |
| 608 | spin_unlock(&inode->i_lock); |
| 609 | } |
| 610 | return de; |
| 611 | } |
| 612 | EXPORT_SYMBOL(d_find_alias); |
| 613 | |
| 614 | /* |
| 615 | * Try to kill dentries associated with this inode. |
| 616 | * WARNING: you must own a reference to inode. |
| 617 | */ |
| 618 | void d_prune_aliases(struct inode *inode) |
| 619 | { |
| 620 | struct dentry *dentry; |
| 621 | restart: |
| 622 | spin_lock(&inode->i_lock); |
| 623 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
| 624 | spin_lock(&dentry->d_lock); |
| 625 | if (!dentry->d_count) { |
| 626 | __dget_dlock(dentry); |
| 627 | __d_drop(dentry); |
| 628 | spin_unlock(&dentry->d_lock); |
| 629 | spin_unlock(&inode->i_lock); |
| 630 | dput(dentry); |
| 631 | goto restart; |
| 632 | } |
| 633 | spin_unlock(&dentry->d_lock); |
| 634 | } |
| 635 | spin_unlock(&inode->i_lock); |
| 636 | } |
| 637 | EXPORT_SYMBOL(d_prune_aliases); |
| 638 | |
| 639 | /* |
| 640 | * Try to throw away a dentry - free the inode, dput the parent. |
| 641 | * Requires dentry->d_lock is held, and dentry->d_count == 0. |
| 642 | * Releases dentry->d_lock. |
| 643 | * |
| 644 | * This may fail if locks cannot be acquired no problem, just try again. |
| 645 | */ |
| 646 | static void try_prune_one_dentry(struct dentry *dentry) |
| 647 | __releases(dentry->d_lock) |
| 648 | { |
| 649 | struct dentry *parent; |
| 650 | |
| 651 | parent = dentry_kill(dentry, 0); |
| 652 | /* |
| 653 | * If dentry_kill returns NULL, we have nothing more to do. |
| 654 | * if it returns the same dentry, trylocks failed. In either |
| 655 | * case, just loop again. |
| 656 | * |
| 657 | * Otherwise, we need to prune ancestors too. This is necessary |
| 658 | * to prevent quadratic behavior of shrink_dcache_parent(), but |
| 659 | * is also expected to be beneficial in reducing dentry cache |
| 660 | * fragmentation. |
| 661 | */ |
| 662 | if (!parent) |
| 663 | return; |
| 664 | if (parent == dentry) |
| 665 | return; |
| 666 | |
| 667 | /* Prune ancestors. */ |
| 668 | dentry = parent; |
| 669 | while (dentry) { |
| 670 | spin_lock(&dentry->d_lock); |
| 671 | if (dentry->d_count > 1) { |
| 672 | dentry->d_count--; |
| 673 | spin_unlock(&dentry->d_lock); |
| 674 | return; |
| 675 | } |
| 676 | dentry = dentry_kill(dentry, 1); |
| 677 | } |
| 678 | } |
| 679 | |
| 680 | static void shrink_dentry_list(struct list_head *list) |
| 681 | { |
| 682 | struct dentry *dentry; |
| 683 | |
| 684 | rcu_read_lock(); |
| 685 | for (;;) { |
| 686 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
| 687 | if (&dentry->d_lru == list) |
| 688 | break; /* empty */ |
| 689 | spin_lock(&dentry->d_lock); |
| 690 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { |
| 691 | spin_unlock(&dentry->d_lock); |
| 692 | continue; |
| 693 | } |
| 694 | |
| 695 | /* |
| 696 | * We found an inuse dentry which was not removed from |
| 697 | * the LRU because of laziness during lookup. Do not free |
| 698 | * it - just keep it off the LRU list. |
| 699 | */ |
| 700 | if (dentry->d_count) { |
| 701 | dentry_lru_del(dentry); |
| 702 | spin_unlock(&dentry->d_lock); |
| 703 | continue; |
| 704 | } |
| 705 | |
| 706 | rcu_read_unlock(); |
| 707 | |
| 708 | try_prune_one_dentry(dentry); |
| 709 | |
| 710 | rcu_read_lock(); |
| 711 | } |
| 712 | rcu_read_unlock(); |
| 713 | } |
| 714 | |
| 715 | /** |
| 716 | * __shrink_dcache_sb - shrink the dentry LRU on a given superblock |
| 717 | * @sb: superblock to shrink dentry LRU. |
| 718 | * @count: number of entries to prune |
| 719 | * @flags: flags to control the dentry processing |
| 720 | * |
| 721 | * If flags contains DCACHE_REFERENCED reference dentries will not be pruned. |
| 722 | */ |
| 723 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) |
| 724 | { |
| 725 | /* called from prune_dcache() and shrink_dcache_parent() */ |
| 726 | struct dentry *dentry; |
| 727 | LIST_HEAD(referenced); |
| 728 | LIST_HEAD(tmp); |
| 729 | int cnt = *count; |
| 730 | |
| 731 | relock: |
| 732 | spin_lock(&dcache_lru_lock); |
| 733 | while (!list_empty(&sb->s_dentry_lru)) { |
| 734 | dentry = list_entry(sb->s_dentry_lru.prev, |
| 735 | struct dentry, d_lru); |
| 736 | BUG_ON(dentry->d_sb != sb); |
| 737 | |
| 738 | if (!spin_trylock(&dentry->d_lock)) { |
| 739 | spin_unlock(&dcache_lru_lock); |
| 740 | cpu_relax(); |
| 741 | goto relock; |
| 742 | } |
| 743 | |
| 744 | /* |
| 745 | * If we are honouring the DCACHE_REFERENCED flag and the |
| 746 | * dentry has this flag set, don't free it. Clear the flag |
| 747 | * and put it back on the LRU. |
| 748 | */ |
| 749 | if (flags & DCACHE_REFERENCED && |
| 750 | dentry->d_flags & DCACHE_REFERENCED) { |
| 751 | dentry->d_flags &= ~DCACHE_REFERENCED; |
| 752 | list_move(&dentry->d_lru, &referenced); |
| 753 | spin_unlock(&dentry->d_lock); |
| 754 | } else { |
| 755 | list_move_tail(&dentry->d_lru, &tmp); |
| 756 | spin_unlock(&dentry->d_lock); |
| 757 | if (!--cnt) |
| 758 | break; |
| 759 | } |
| 760 | cond_resched_lock(&dcache_lru_lock); |
| 761 | } |
| 762 | if (!list_empty(&referenced)) |
| 763 | list_splice(&referenced, &sb->s_dentry_lru); |
| 764 | spin_unlock(&dcache_lru_lock); |
| 765 | |
| 766 | shrink_dentry_list(&tmp); |
| 767 | |
| 768 | *count = cnt; |
| 769 | } |
| 770 | |
| 771 | /** |
| 772 | * prune_dcache - shrink the dcache |
| 773 | * @count: number of entries to try to free |
| 774 | * |
| 775 | * Shrink the dcache. This is done when we need more memory, or simply when we |
| 776 | * need to unmount something (at which point we need to unuse all dentries). |
| 777 | * |
| 778 | * This function may fail to free any resources if all the dentries are in use. |
| 779 | */ |
| 780 | static void prune_dcache(int count) |
| 781 | { |
| 782 | struct super_block *sb, *p = NULL; |
| 783 | int w_count; |
| 784 | int unused = dentry_stat.nr_unused; |
| 785 | int prune_ratio; |
| 786 | int pruned; |
| 787 | |
| 788 | if (unused == 0 || count == 0) |
| 789 | return; |
| 790 | if (count >= unused) |
| 791 | prune_ratio = 1; |
| 792 | else |
| 793 | prune_ratio = unused / count; |
| 794 | spin_lock(&sb_lock); |
| 795 | list_for_each_entry(sb, &super_blocks, s_list) { |
| 796 | if (list_empty(&sb->s_instances)) |
| 797 | continue; |
| 798 | if (sb->s_nr_dentry_unused == 0) |
| 799 | continue; |
| 800 | sb->s_count++; |
| 801 | /* Now, we reclaim unused dentrins with fairness. |
| 802 | * We reclaim them same percentage from each superblock. |
| 803 | * We calculate number of dentries to scan on this sb |
| 804 | * as follows, but the implementation is arranged to avoid |
| 805 | * overflows: |
| 806 | * number of dentries to scan on this sb = |
| 807 | * count * (number of dentries on this sb / |
| 808 | * number of dentries in the machine) |
| 809 | */ |
| 810 | spin_unlock(&sb_lock); |
| 811 | if (prune_ratio != 1) |
| 812 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; |
| 813 | else |
| 814 | w_count = sb->s_nr_dentry_unused; |
| 815 | pruned = w_count; |
| 816 | /* |
| 817 | * We need to be sure this filesystem isn't being unmounted, |
| 818 | * otherwise we could race with generic_shutdown_super(), and |
| 819 | * end up holding a reference to an inode while the filesystem |
| 820 | * is unmounted. So we try to get s_umount, and make sure |
| 821 | * s_root isn't NULL. |
| 822 | */ |
| 823 | if (down_read_trylock(&sb->s_umount)) { |
| 824 | if ((sb->s_root != NULL) && |
| 825 | (!list_empty(&sb->s_dentry_lru))) { |
| 826 | __shrink_dcache_sb(sb, &w_count, |
| 827 | DCACHE_REFERENCED); |
| 828 | pruned -= w_count; |
| 829 | } |
| 830 | up_read(&sb->s_umount); |
| 831 | } |
| 832 | spin_lock(&sb_lock); |
| 833 | if (p) |
| 834 | __put_super(p); |
| 835 | count -= pruned; |
| 836 | p = sb; |
| 837 | /* more work left to do? */ |
| 838 | if (count <= 0) |
| 839 | break; |
| 840 | } |
| 841 | if (p) |
| 842 | __put_super(p); |
| 843 | spin_unlock(&sb_lock); |
| 844 | } |
| 845 | |
| 846 | /** |
| 847 | * shrink_dcache_sb - shrink dcache for a superblock |
| 848 | * @sb: superblock |
| 849 | * |
| 850 | * Shrink the dcache for the specified super block. This is used to free |
| 851 | * the dcache before unmounting a file system. |
| 852 | */ |
| 853 | void shrink_dcache_sb(struct super_block *sb) |
| 854 | { |
| 855 | LIST_HEAD(tmp); |
| 856 | |
| 857 | spin_lock(&dcache_lru_lock); |
| 858 | while (!list_empty(&sb->s_dentry_lru)) { |
| 859 | list_splice_init(&sb->s_dentry_lru, &tmp); |
| 860 | spin_unlock(&dcache_lru_lock); |
| 861 | shrink_dentry_list(&tmp); |
| 862 | spin_lock(&dcache_lru_lock); |
| 863 | } |
| 864 | spin_unlock(&dcache_lru_lock); |
| 865 | } |
| 866 | EXPORT_SYMBOL(shrink_dcache_sb); |
| 867 | |
| 868 | /* |
| 869 | * destroy a single subtree of dentries for unmount |
| 870 | * - see the comments on shrink_dcache_for_umount() for a description of the |
| 871 | * locking |
| 872 | */ |
| 873 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) |
| 874 | { |
| 875 | struct dentry *parent; |
| 876 | unsigned detached = 0; |
| 877 | |
| 878 | BUG_ON(!IS_ROOT(dentry)); |
| 879 | |
| 880 | /* detach this root from the system */ |
| 881 | spin_lock(&dentry->d_lock); |
| 882 | dentry_lru_del(dentry); |
| 883 | __d_drop(dentry); |
| 884 | spin_unlock(&dentry->d_lock); |
| 885 | |
| 886 | for (;;) { |
| 887 | /* descend to the first leaf in the current subtree */ |
| 888 | while (!list_empty(&dentry->d_subdirs)) { |
| 889 | struct dentry *loop; |
| 890 | |
| 891 | /* this is a branch with children - detach all of them |
| 892 | * from the system in one go */ |
| 893 | spin_lock(&dentry->d_lock); |
| 894 | list_for_each_entry(loop, &dentry->d_subdirs, |
| 895 | d_u.d_child) { |
| 896 | spin_lock_nested(&loop->d_lock, |
| 897 | DENTRY_D_LOCK_NESTED); |
| 898 | dentry_lru_del(loop); |
| 899 | __d_drop(loop); |
| 900 | spin_unlock(&loop->d_lock); |
| 901 | } |
| 902 | spin_unlock(&dentry->d_lock); |
| 903 | |
| 904 | /* move to the first child */ |
| 905 | dentry = list_entry(dentry->d_subdirs.next, |
| 906 | struct dentry, d_u.d_child); |
| 907 | } |
| 908 | |
| 909 | /* consume the dentries from this leaf up through its parents |
| 910 | * until we find one with children or run out altogether */ |
| 911 | do { |
| 912 | struct inode *inode; |
| 913 | |
| 914 | if (dentry->d_count != 0) { |
| 915 | printk(KERN_ERR |
| 916 | "BUG: Dentry %p{i=%lx,n=%s}" |
| 917 | " still in use (%d)" |
| 918 | " [unmount of %s %s]\n", |
| 919 | dentry, |
| 920 | dentry->d_inode ? |
| 921 | dentry->d_inode->i_ino : 0UL, |
| 922 | dentry->d_name.name, |
| 923 | dentry->d_count, |
| 924 | dentry->d_sb->s_type->name, |
| 925 | dentry->d_sb->s_id); |
| 926 | BUG(); |
| 927 | } |
| 928 | |
| 929 | if (IS_ROOT(dentry)) { |
| 930 | parent = NULL; |
| 931 | list_del(&dentry->d_u.d_child); |
| 932 | } else { |
| 933 | parent = dentry->d_parent; |
| 934 | spin_lock(&parent->d_lock); |
| 935 | parent->d_count--; |
| 936 | list_del(&dentry->d_u.d_child); |
| 937 | spin_unlock(&parent->d_lock); |
| 938 | } |
| 939 | |
| 940 | detached++; |
| 941 | |
| 942 | inode = dentry->d_inode; |
| 943 | if (inode) { |
| 944 | dentry->d_inode = NULL; |
| 945 | list_del_init(&dentry->d_alias); |
| 946 | if (dentry->d_op && dentry->d_op->d_iput) |
| 947 | dentry->d_op->d_iput(dentry, inode); |
| 948 | else |
| 949 | iput(inode); |
| 950 | } |
| 951 | |
| 952 | d_free(dentry); |
| 953 | |
| 954 | /* finished when we fall off the top of the tree, |
| 955 | * otherwise we ascend to the parent and move to the |
| 956 | * next sibling if there is one */ |
| 957 | if (!parent) |
| 958 | return; |
| 959 | dentry = parent; |
| 960 | } while (list_empty(&dentry->d_subdirs)); |
| 961 | |
| 962 | dentry = list_entry(dentry->d_subdirs.next, |
| 963 | struct dentry, d_u.d_child); |
| 964 | } |
| 965 | } |
| 966 | |
| 967 | /* |
| 968 | * destroy the dentries attached to a superblock on unmounting |
| 969 | * - we don't need to use dentry->d_lock because: |
| 970 | * - the superblock is detached from all mountings and open files, so the |
| 971 | * dentry trees will not be rearranged by the VFS |
| 972 | * - s_umount is write-locked, so the memory pressure shrinker will ignore |
| 973 | * any dentries belonging to this superblock that it comes across |
| 974 | * - the filesystem itself is no longer permitted to rearrange the dentries |
| 975 | * in this superblock |
| 976 | */ |
| 977 | void shrink_dcache_for_umount(struct super_block *sb) |
| 978 | { |
| 979 | struct dentry *dentry; |
| 980 | |
| 981 | if (down_read_trylock(&sb->s_umount)) |
| 982 | BUG(); |
| 983 | |
| 984 | dentry = sb->s_root; |
| 985 | sb->s_root = NULL; |
| 986 | spin_lock(&dentry->d_lock); |
| 987 | dentry->d_count--; |
| 988 | spin_unlock(&dentry->d_lock); |
| 989 | shrink_dcache_for_umount_subtree(dentry); |
| 990 | |
| 991 | while (!hlist_bl_empty(&sb->s_anon)) { |
| 992 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); |
| 993 | shrink_dcache_for_umount_subtree(dentry); |
| 994 | } |
| 995 | } |
| 996 | |
| 997 | /* |
| 998 | * This tries to ascend one level of parenthood, but |
| 999 | * we can race with renaming, so we need to re-check |
| 1000 | * the parenthood after dropping the lock and check |
| 1001 | * that the sequence number still matches. |
| 1002 | */ |
| 1003 | static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq) |
| 1004 | { |
| 1005 | struct dentry *new = old->d_parent; |
| 1006 | |
| 1007 | rcu_read_lock(); |
| 1008 | spin_unlock(&old->d_lock); |
| 1009 | spin_lock(&new->d_lock); |
| 1010 | |
| 1011 | /* |
| 1012 | * might go back up the wrong parent if we have had a rename |
| 1013 | * or deletion |
| 1014 | */ |
| 1015 | if (new != old->d_parent || |
| 1016 | (old->d_flags & DCACHE_DISCONNECTED) || |
| 1017 | (!locked && read_seqretry(&rename_lock, seq))) { |
| 1018 | spin_unlock(&new->d_lock); |
| 1019 | new = NULL; |
| 1020 | } |
| 1021 | rcu_read_unlock(); |
| 1022 | return new; |
| 1023 | } |
| 1024 | |
| 1025 | |
| 1026 | /* |
| 1027 | * Search for at least 1 mount point in the dentry's subdirs. |
| 1028 | * We descend to the next level whenever the d_subdirs |
| 1029 | * list is non-empty and continue searching. |
| 1030 | */ |
| 1031 | |
| 1032 | /** |
| 1033 | * have_submounts - check for mounts over a dentry |
| 1034 | * @parent: dentry to check. |
| 1035 | * |
| 1036 | * Return true if the parent or its subdirectories contain |
| 1037 | * a mount point |
| 1038 | */ |
| 1039 | int have_submounts(struct dentry *parent) |
| 1040 | { |
| 1041 | struct dentry *this_parent; |
| 1042 | struct list_head *next; |
| 1043 | unsigned seq; |
| 1044 | int locked = 0; |
| 1045 | |
| 1046 | seq = read_seqbegin(&rename_lock); |
| 1047 | again: |
| 1048 | this_parent = parent; |
| 1049 | |
| 1050 | if (d_mountpoint(parent)) |
| 1051 | goto positive; |
| 1052 | spin_lock(&this_parent->d_lock); |
| 1053 | repeat: |
| 1054 | next = this_parent->d_subdirs.next; |
| 1055 | resume: |
| 1056 | while (next != &this_parent->d_subdirs) { |
| 1057 | struct list_head *tmp = next; |
| 1058 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
| 1059 | next = tmp->next; |
| 1060 | |
| 1061 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
| 1062 | /* Have we found a mount point ? */ |
| 1063 | if (d_mountpoint(dentry)) { |
| 1064 | spin_unlock(&dentry->d_lock); |
| 1065 | spin_unlock(&this_parent->d_lock); |
| 1066 | goto positive; |
| 1067 | } |
| 1068 | if (!list_empty(&dentry->d_subdirs)) { |
| 1069 | spin_unlock(&this_parent->d_lock); |
| 1070 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); |
| 1071 | this_parent = dentry; |
| 1072 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
| 1073 | goto repeat; |
| 1074 | } |
| 1075 | spin_unlock(&dentry->d_lock); |
| 1076 | } |
| 1077 | /* |
| 1078 | * All done at this level ... ascend and resume the search. |
| 1079 | */ |
| 1080 | if (this_parent != parent) { |
| 1081 | struct dentry *child = this_parent; |
| 1082 | this_parent = try_to_ascend(this_parent, locked, seq); |
| 1083 | if (!this_parent) |
| 1084 | goto rename_retry; |
| 1085 | next = child->d_u.d_child.next; |
| 1086 | goto resume; |
| 1087 | } |
| 1088 | spin_unlock(&this_parent->d_lock); |
| 1089 | if (!locked && read_seqretry(&rename_lock, seq)) |
| 1090 | goto rename_retry; |
| 1091 | if (locked) |
| 1092 | write_sequnlock(&rename_lock); |
| 1093 | return 0; /* No mount points found in tree */ |
| 1094 | positive: |
| 1095 | if (!locked && read_seqretry(&rename_lock, seq)) |
| 1096 | goto rename_retry; |
| 1097 | if (locked) |
| 1098 | write_sequnlock(&rename_lock); |
| 1099 | return 1; |
| 1100 | |
| 1101 | rename_retry: |
| 1102 | locked = 1; |
| 1103 | write_seqlock(&rename_lock); |
| 1104 | goto again; |
| 1105 | } |
| 1106 | EXPORT_SYMBOL(have_submounts); |
| 1107 | |
| 1108 | /* |
| 1109 | * Search the dentry child list for the specified parent, |
| 1110 | * and move any unused dentries to the end of the unused |
| 1111 | * list for prune_dcache(). We descend to the next level |
| 1112 | * whenever the d_subdirs list is non-empty and continue |
| 1113 | * searching. |
| 1114 | * |
| 1115 | * It returns zero iff there are no unused children, |
| 1116 | * otherwise it returns the number of children moved to |
| 1117 | * the end of the unused list. This may not be the total |
| 1118 | * number of unused children, because select_parent can |
| 1119 | * drop the lock and return early due to latency |
| 1120 | * constraints. |
| 1121 | */ |
| 1122 | static int select_parent(struct dentry * parent) |
| 1123 | { |
| 1124 | struct dentry *this_parent; |
| 1125 | struct list_head *next; |
| 1126 | unsigned seq; |
| 1127 | int found = 0; |
| 1128 | int locked = 0; |
| 1129 | |
| 1130 | seq = read_seqbegin(&rename_lock); |
| 1131 | again: |
| 1132 | this_parent = parent; |
| 1133 | spin_lock(&this_parent->d_lock); |
| 1134 | repeat: |
| 1135 | next = this_parent->d_subdirs.next; |
| 1136 | resume: |
| 1137 | while (next != &this_parent->d_subdirs) { |
| 1138 | struct list_head *tmp = next; |
| 1139 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
| 1140 | next = tmp->next; |
| 1141 | |
| 1142 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
| 1143 | |
| 1144 | /* |
| 1145 | * move only zero ref count dentries to the end |
| 1146 | * of the unused list for prune_dcache |
| 1147 | */ |
| 1148 | if (!dentry->d_count) { |
| 1149 | dentry_lru_move_tail(dentry); |
| 1150 | found++; |
| 1151 | } else { |
| 1152 | dentry_lru_del(dentry); |
| 1153 | } |
| 1154 | |
| 1155 | /* |
| 1156 | * We can return to the caller if we have found some (this |
| 1157 | * ensures forward progress). We'll be coming back to find |
| 1158 | * the rest. |
| 1159 | */ |
| 1160 | if (found && need_resched()) { |
| 1161 | spin_unlock(&dentry->d_lock); |
| 1162 | goto out; |
| 1163 | } |
| 1164 | |
| 1165 | /* |
| 1166 | * Descend a level if the d_subdirs list is non-empty. |
| 1167 | */ |
| 1168 | if (!list_empty(&dentry->d_subdirs)) { |
| 1169 | spin_unlock(&this_parent->d_lock); |
| 1170 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); |
| 1171 | this_parent = dentry; |
| 1172 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
| 1173 | goto repeat; |
| 1174 | } |
| 1175 | |
| 1176 | spin_unlock(&dentry->d_lock); |
| 1177 | } |
| 1178 | /* |
| 1179 | * All done at this level ... ascend and resume the search. |
| 1180 | */ |
| 1181 | if (this_parent != parent) { |
| 1182 | struct dentry *child = this_parent; |
| 1183 | this_parent = try_to_ascend(this_parent, locked, seq); |
| 1184 | if (!this_parent) |
| 1185 | goto rename_retry; |
| 1186 | next = child->d_u.d_child.next; |
| 1187 | goto resume; |
| 1188 | } |
| 1189 | out: |
| 1190 | spin_unlock(&this_parent->d_lock); |
| 1191 | if (!locked && read_seqretry(&rename_lock, seq)) |
| 1192 | goto rename_retry; |
| 1193 | if (locked) |
| 1194 | write_sequnlock(&rename_lock); |
| 1195 | return found; |
| 1196 | |
| 1197 | rename_retry: |
| 1198 | if (found) |
| 1199 | return found; |
| 1200 | locked = 1; |
| 1201 | write_seqlock(&rename_lock); |
| 1202 | goto again; |
| 1203 | } |
| 1204 | |
| 1205 | /** |
| 1206 | * shrink_dcache_parent - prune dcache |
| 1207 | * @parent: parent of entries to prune |
| 1208 | * |
| 1209 | * Prune the dcache to remove unused children of the parent dentry. |
| 1210 | */ |
| 1211 | |
| 1212 | void shrink_dcache_parent(struct dentry * parent) |
| 1213 | { |
| 1214 | struct super_block *sb = parent->d_sb; |
| 1215 | int found; |
| 1216 | |
| 1217 | while ((found = select_parent(parent)) != 0) |
| 1218 | __shrink_dcache_sb(sb, &found, 0); |
| 1219 | } |
| 1220 | EXPORT_SYMBOL(shrink_dcache_parent); |
| 1221 | |
| 1222 | /* |
| 1223 | * Scan `nr' dentries and return the number which remain. |
| 1224 | * |
| 1225 | * We need to avoid reentering the filesystem if the caller is performing a |
| 1226 | * GFP_NOFS allocation attempt. One example deadlock is: |
| 1227 | * |
| 1228 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> |
| 1229 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> |
| 1230 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. |
| 1231 | * |
| 1232 | * In this case we return -1 to tell the caller that we baled. |
| 1233 | */ |
| 1234 | static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
| 1235 | { |
| 1236 | if (nr) { |
| 1237 | if (!(gfp_mask & __GFP_FS)) |
| 1238 | return -1; |
| 1239 | prune_dcache(nr); |
| 1240 | } |
| 1241 | |
| 1242 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; |
| 1243 | } |
| 1244 | |
| 1245 | static struct shrinker dcache_shrinker = { |
| 1246 | .shrink = shrink_dcache_memory, |
| 1247 | .seeks = DEFAULT_SEEKS, |
| 1248 | }; |
| 1249 | |
| 1250 | /** |
| 1251 | * d_alloc - allocate a dcache entry |
| 1252 | * @parent: parent of entry to allocate |
| 1253 | * @name: qstr of the name |
| 1254 | * |
| 1255 | * Allocates a dentry. It returns %NULL if there is insufficient memory |
| 1256 | * available. On a success the dentry is returned. The name passed in is |
| 1257 | * copied and the copy passed in may be reused after this call. |
| 1258 | */ |
| 1259 | |
| 1260 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) |
| 1261 | { |
| 1262 | struct dentry *dentry; |
| 1263 | char *dname; |
| 1264 | |
| 1265 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
| 1266 | if (!dentry) |
| 1267 | return NULL; |
| 1268 | |
| 1269 | if (name->len > DNAME_INLINE_LEN-1) { |
| 1270 | dname = kmalloc(name->len + 1, GFP_KERNEL); |
| 1271 | if (!dname) { |
| 1272 | kmem_cache_free(dentry_cache, dentry); |
| 1273 | return NULL; |
| 1274 | } |
| 1275 | } else { |
| 1276 | dname = dentry->d_iname; |
| 1277 | } |
| 1278 | dentry->d_name.name = dname; |
| 1279 | |
| 1280 | dentry->d_name.len = name->len; |
| 1281 | dentry->d_name.hash = name->hash; |
| 1282 | memcpy(dname, name->name, name->len); |
| 1283 | dname[name->len] = 0; |
| 1284 | |
| 1285 | dentry->d_count = 1; |
| 1286 | dentry->d_flags = 0; |
| 1287 | spin_lock_init(&dentry->d_lock); |
| 1288 | seqcount_init(&dentry->d_seq); |
| 1289 | dentry->d_inode = NULL; |
| 1290 | dentry->d_parent = NULL; |
| 1291 | dentry->d_sb = NULL; |
| 1292 | dentry->d_op = NULL; |
| 1293 | dentry->d_fsdata = NULL; |
| 1294 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
| 1295 | INIT_LIST_HEAD(&dentry->d_lru); |
| 1296 | INIT_LIST_HEAD(&dentry->d_subdirs); |
| 1297 | INIT_LIST_HEAD(&dentry->d_alias); |
| 1298 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
| 1299 | |
| 1300 | if (parent) { |
| 1301 | spin_lock(&parent->d_lock); |
| 1302 | /* |
| 1303 | * don't need child lock because it is not subject |
| 1304 | * to concurrency here |
| 1305 | */ |
| 1306 | __dget_dlock(parent); |
| 1307 | dentry->d_parent = parent; |
| 1308 | dentry->d_sb = parent->d_sb; |
| 1309 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
| 1310 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
| 1311 | spin_unlock(&parent->d_lock); |
| 1312 | } |
| 1313 | |
| 1314 | this_cpu_inc(nr_dentry); |
| 1315 | |
| 1316 | return dentry; |
| 1317 | } |
| 1318 | EXPORT_SYMBOL(d_alloc); |
| 1319 | |
| 1320 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
| 1321 | { |
| 1322 | struct dentry *dentry = d_alloc(NULL, name); |
| 1323 | if (dentry) { |
| 1324 | dentry->d_sb = sb; |
| 1325 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
| 1326 | dentry->d_parent = dentry; |
| 1327 | dentry->d_flags |= DCACHE_DISCONNECTED; |
| 1328 | } |
| 1329 | return dentry; |
| 1330 | } |
| 1331 | EXPORT_SYMBOL(d_alloc_pseudo); |
| 1332 | |
| 1333 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
| 1334 | { |
| 1335 | struct qstr q; |
| 1336 | |
| 1337 | q.name = name; |
| 1338 | q.len = strlen(name); |
| 1339 | q.hash = full_name_hash(q.name, q.len); |
| 1340 | return d_alloc(parent, &q); |
| 1341 | } |
| 1342 | EXPORT_SYMBOL(d_alloc_name); |
| 1343 | |
| 1344 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
| 1345 | { |
| 1346 | WARN_ON_ONCE(dentry->d_op); |
| 1347 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | |
| 1348 | DCACHE_OP_COMPARE | |
| 1349 | DCACHE_OP_REVALIDATE | |
| 1350 | DCACHE_OP_DELETE )); |
| 1351 | dentry->d_op = op; |
| 1352 | if (!op) |
| 1353 | return; |
| 1354 | if (op->d_hash) |
| 1355 | dentry->d_flags |= DCACHE_OP_HASH; |
| 1356 | if (op->d_compare) |
| 1357 | dentry->d_flags |= DCACHE_OP_COMPARE; |
| 1358 | if (op->d_revalidate) |
| 1359 | dentry->d_flags |= DCACHE_OP_REVALIDATE; |
| 1360 | if (op->d_delete) |
| 1361 | dentry->d_flags |= DCACHE_OP_DELETE; |
| 1362 | |
| 1363 | } |
| 1364 | EXPORT_SYMBOL(d_set_d_op); |
| 1365 | |
| 1366 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
| 1367 | { |
| 1368 | spin_lock(&dentry->d_lock); |
| 1369 | if (inode) { |
| 1370 | if (unlikely(IS_AUTOMOUNT(inode))) |
| 1371 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; |
| 1372 | list_add(&dentry->d_alias, &inode->i_dentry); |
| 1373 | } |
| 1374 | dentry->d_inode = inode; |
| 1375 | dentry_rcuwalk_barrier(dentry); |
| 1376 | spin_unlock(&dentry->d_lock); |
| 1377 | fsnotify_d_instantiate(dentry, inode); |
| 1378 | } |
| 1379 | |
| 1380 | /** |
| 1381 | * d_instantiate - fill in inode information for a dentry |
| 1382 | * @entry: dentry to complete |
| 1383 | * @inode: inode to attach to this dentry |
| 1384 | * |
| 1385 | * Fill in inode information in the entry. |
| 1386 | * |
| 1387 | * This turns negative dentries into productive full members |
| 1388 | * of society. |
| 1389 | * |
| 1390 | * NOTE! This assumes that the inode count has been incremented |
| 1391 | * (or otherwise set) by the caller to indicate that it is now |
| 1392 | * in use by the dcache. |
| 1393 | */ |
| 1394 | |
| 1395 | void d_instantiate(struct dentry *entry, struct inode * inode) |
| 1396 | { |
| 1397 | BUG_ON(!list_empty(&entry->d_alias)); |
| 1398 | if (inode) |
| 1399 | spin_lock(&inode->i_lock); |
| 1400 | __d_instantiate(entry, inode); |
| 1401 | if (inode) |
| 1402 | spin_unlock(&inode->i_lock); |
| 1403 | security_d_instantiate(entry, inode); |
| 1404 | } |
| 1405 | EXPORT_SYMBOL(d_instantiate); |
| 1406 | |
| 1407 | /** |
| 1408 | * d_instantiate_unique - instantiate a non-aliased dentry |
| 1409 | * @entry: dentry to instantiate |
| 1410 | * @inode: inode to attach to this dentry |
| 1411 | * |
| 1412 | * Fill in inode information in the entry. On success, it returns NULL. |
| 1413 | * If an unhashed alias of "entry" already exists, then we return the |
| 1414 | * aliased dentry instead and drop one reference to inode. |
| 1415 | * |
| 1416 | * Note that in order to avoid conflicts with rename() etc, the caller |
| 1417 | * had better be holding the parent directory semaphore. |
| 1418 | * |
| 1419 | * This also assumes that the inode count has been incremented |
| 1420 | * (or otherwise set) by the caller to indicate that it is now |
| 1421 | * in use by the dcache. |
| 1422 | */ |
| 1423 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
| 1424 | struct inode *inode) |
| 1425 | { |
| 1426 | struct dentry *alias; |
| 1427 | int len = entry->d_name.len; |
| 1428 | const char *name = entry->d_name.name; |
| 1429 | unsigned int hash = entry->d_name.hash; |
| 1430 | |
| 1431 | if (!inode) { |
| 1432 | __d_instantiate(entry, NULL); |
| 1433 | return NULL; |
| 1434 | } |
| 1435 | |
| 1436 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
| 1437 | struct qstr *qstr = &alias->d_name; |
| 1438 | |
| 1439 | /* |
| 1440 | * Don't need alias->d_lock here, because aliases with |
| 1441 | * d_parent == entry->d_parent are not subject to name or |
| 1442 | * parent changes, because the parent inode i_mutex is held. |
| 1443 | */ |
| 1444 | if (qstr->hash != hash) |
| 1445 | continue; |
| 1446 | if (alias->d_parent != entry->d_parent) |
| 1447 | continue; |
| 1448 | if (dentry_cmp(qstr->name, qstr->len, name, len)) |
| 1449 | continue; |
| 1450 | __dget(alias); |
| 1451 | return alias; |
| 1452 | } |
| 1453 | |
| 1454 | __d_instantiate(entry, inode); |
| 1455 | return NULL; |
| 1456 | } |
| 1457 | |
| 1458 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) |
| 1459 | { |
| 1460 | struct dentry *result; |
| 1461 | |
| 1462 | BUG_ON(!list_empty(&entry->d_alias)); |
| 1463 | |
| 1464 | if (inode) |
| 1465 | spin_lock(&inode->i_lock); |
| 1466 | result = __d_instantiate_unique(entry, inode); |
| 1467 | if (inode) |
| 1468 | spin_unlock(&inode->i_lock); |
| 1469 | |
| 1470 | if (!result) { |
| 1471 | security_d_instantiate(entry, inode); |
| 1472 | return NULL; |
| 1473 | } |
| 1474 | |
| 1475 | BUG_ON(!d_unhashed(result)); |
| 1476 | iput(inode); |
| 1477 | return result; |
| 1478 | } |
| 1479 | |
| 1480 | EXPORT_SYMBOL(d_instantiate_unique); |
| 1481 | |
| 1482 | /** |
| 1483 | * d_alloc_root - allocate root dentry |
| 1484 | * @root_inode: inode to allocate the root for |
| 1485 | * |
| 1486 | * Allocate a root ("/") dentry for the inode given. The inode is |
| 1487 | * instantiated and returned. %NULL is returned if there is insufficient |
| 1488 | * memory or the inode passed is %NULL. |
| 1489 | */ |
| 1490 | |
| 1491 | struct dentry * d_alloc_root(struct inode * root_inode) |
| 1492 | { |
| 1493 | struct dentry *res = NULL; |
| 1494 | |
| 1495 | if (root_inode) { |
| 1496 | static const struct qstr name = { .name = "/", .len = 1 }; |
| 1497 | |
| 1498 | res = d_alloc(NULL, &name); |
| 1499 | if (res) { |
| 1500 | res->d_sb = root_inode->i_sb; |
| 1501 | d_set_d_op(res, res->d_sb->s_d_op); |
| 1502 | res->d_parent = res; |
| 1503 | d_instantiate(res, root_inode); |
| 1504 | } |
| 1505 | } |
| 1506 | return res; |
| 1507 | } |
| 1508 | EXPORT_SYMBOL(d_alloc_root); |
| 1509 | |
| 1510 | static struct dentry * __d_find_any_alias(struct inode *inode) |
| 1511 | { |
| 1512 | struct dentry *alias; |
| 1513 | |
| 1514 | if (list_empty(&inode->i_dentry)) |
| 1515 | return NULL; |
| 1516 | alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias); |
| 1517 | __dget(alias); |
| 1518 | return alias; |
| 1519 | } |
| 1520 | |
| 1521 | static struct dentry * d_find_any_alias(struct inode *inode) |
| 1522 | { |
| 1523 | struct dentry *de; |
| 1524 | |
| 1525 | spin_lock(&inode->i_lock); |
| 1526 | de = __d_find_any_alias(inode); |
| 1527 | spin_unlock(&inode->i_lock); |
| 1528 | return de; |
| 1529 | } |
| 1530 | |
| 1531 | |
| 1532 | /** |
| 1533 | * d_obtain_alias - find or allocate a dentry for a given inode |
| 1534 | * @inode: inode to allocate the dentry for |
| 1535 | * |
| 1536 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or |
| 1537 | * similar open by handle operations. The returned dentry may be anonymous, |
| 1538 | * or may have a full name (if the inode was already in the cache). |
| 1539 | * |
| 1540 | * When called on a directory inode, we must ensure that the inode only ever |
| 1541 | * has one dentry. If a dentry is found, that is returned instead of |
| 1542 | * allocating a new one. |
| 1543 | * |
| 1544 | * On successful return, the reference to the inode has been transferred |
| 1545 | * to the dentry. In case of an error the reference on the inode is released. |
| 1546 | * To make it easier to use in export operations a %NULL or IS_ERR inode may |
| 1547 | * be passed in and will be the error will be propagate to the return value, |
| 1548 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). |
| 1549 | */ |
| 1550 | struct dentry *d_obtain_alias(struct inode *inode) |
| 1551 | { |
| 1552 | static const struct qstr anonstring = { .name = "" }; |
| 1553 | struct dentry *tmp; |
| 1554 | struct dentry *res; |
| 1555 | |
| 1556 | if (!inode) |
| 1557 | return ERR_PTR(-ESTALE); |
| 1558 | if (IS_ERR(inode)) |
| 1559 | return ERR_CAST(inode); |
| 1560 | |
| 1561 | res = d_find_any_alias(inode); |
| 1562 | if (res) |
| 1563 | goto out_iput; |
| 1564 | |
| 1565 | tmp = d_alloc(NULL, &anonstring); |
| 1566 | if (!tmp) { |
| 1567 | res = ERR_PTR(-ENOMEM); |
| 1568 | goto out_iput; |
| 1569 | } |
| 1570 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ |
| 1571 | |
| 1572 | |
| 1573 | spin_lock(&inode->i_lock); |
| 1574 | res = __d_find_any_alias(inode); |
| 1575 | if (res) { |
| 1576 | spin_unlock(&inode->i_lock); |
| 1577 | dput(tmp); |
| 1578 | goto out_iput; |
| 1579 | } |
| 1580 | |
| 1581 | /* attach a disconnected dentry */ |
| 1582 | spin_lock(&tmp->d_lock); |
| 1583 | tmp->d_sb = inode->i_sb; |
| 1584 | d_set_d_op(tmp, tmp->d_sb->s_d_op); |
| 1585 | tmp->d_inode = inode; |
| 1586 | tmp->d_flags |= DCACHE_DISCONNECTED; |
| 1587 | list_add(&tmp->d_alias, &inode->i_dentry); |
| 1588 | hlist_bl_lock(&tmp->d_sb->s_anon); |
| 1589 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
| 1590 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
| 1591 | spin_unlock(&tmp->d_lock); |
| 1592 | spin_unlock(&inode->i_lock); |
| 1593 | security_d_instantiate(tmp, inode); |
| 1594 | |
| 1595 | return tmp; |
| 1596 | |
| 1597 | out_iput: |
| 1598 | if (res && !IS_ERR(res)) |
| 1599 | security_d_instantiate(res, inode); |
| 1600 | iput(inode); |
| 1601 | return res; |
| 1602 | } |
| 1603 | EXPORT_SYMBOL(d_obtain_alias); |
| 1604 | |
| 1605 | /** |
| 1606 | * d_splice_alias - splice a disconnected dentry into the tree if one exists |
| 1607 | * @inode: the inode which may have a disconnected dentry |
| 1608 | * @dentry: a negative dentry which we want to point to the inode. |
| 1609 | * |
| 1610 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and |
| 1611 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry |
| 1612 | * and return it, else simply d_add the inode to the dentry and return NULL. |
| 1613 | * |
| 1614 | * This is needed in the lookup routine of any filesystem that is exportable |
| 1615 | * (via knfsd) so that we can build dcache paths to directories effectively. |
| 1616 | * |
| 1617 | * If a dentry was found and moved, then it is returned. Otherwise NULL |
| 1618 | * is returned. This matches the expected return value of ->lookup. |
| 1619 | * |
| 1620 | */ |
| 1621 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) |
| 1622 | { |
| 1623 | struct dentry *new = NULL; |
| 1624 | |
| 1625 | if (inode && S_ISDIR(inode->i_mode)) { |
| 1626 | spin_lock(&inode->i_lock); |
| 1627 | new = __d_find_alias(inode, 1); |
| 1628 | if (new) { |
| 1629 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); |
| 1630 | spin_unlock(&inode->i_lock); |
| 1631 | security_d_instantiate(new, inode); |
| 1632 | d_move(new, dentry); |
| 1633 | iput(inode); |
| 1634 | } else { |
| 1635 | /* already taking inode->i_lock, so d_add() by hand */ |
| 1636 | __d_instantiate(dentry, inode); |
| 1637 | spin_unlock(&inode->i_lock); |
| 1638 | security_d_instantiate(dentry, inode); |
| 1639 | d_rehash(dentry); |
| 1640 | } |
| 1641 | } else |
| 1642 | d_add(dentry, inode); |
| 1643 | return new; |
| 1644 | } |
| 1645 | EXPORT_SYMBOL(d_splice_alias); |
| 1646 | |
| 1647 | /** |
| 1648 | * d_add_ci - lookup or allocate new dentry with case-exact name |
| 1649 | * @inode: the inode case-insensitive lookup has found |
| 1650 | * @dentry: the negative dentry that was passed to the parent's lookup func |
| 1651 | * @name: the case-exact name to be associated with the returned dentry |
| 1652 | * |
| 1653 | * This is to avoid filling the dcache with case-insensitive names to the |
| 1654 | * same inode, only the actual correct case is stored in the dcache for |
| 1655 | * case-insensitive filesystems. |
| 1656 | * |
| 1657 | * For a case-insensitive lookup match and if the the case-exact dentry |
| 1658 | * already exists in in the dcache, use it and return it. |
| 1659 | * |
| 1660 | * If no entry exists with the exact case name, allocate new dentry with |
| 1661 | * the exact case, and return the spliced entry. |
| 1662 | */ |
| 1663 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
| 1664 | struct qstr *name) |
| 1665 | { |
| 1666 | int error; |
| 1667 | struct dentry *found; |
| 1668 | struct dentry *new; |
| 1669 | |
| 1670 | /* |
| 1671 | * First check if a dentry matching the name already exists, |
| 1672 | * if not go ahead and create it now. |
| 1673 | */ |
| 1674 | found = d_hash_and_lookup(dentry->d_parent, name); |
| 1675 | if (!found) { |
| 1676 | new = d_alloc(dentry->d_parent, name); |
| 1677 | if (!new) { |
| 1678 | error = -ENOMEM; |
| 1679 | goto err_out; |
| 1680 | } |
| 1681 | |
| 1682 | found = d_splice_alias(inode, new); |
| 1683 | if (found) { |
| 1684 | dput(new); |
| 1685 | return found; |
| 1686 | } |
| 1687 | return new; |
| 1688 | } |
| 1689 | |
| 1690 | /* |
| 1691 | * If a matching dentry exists, and it's not negative use it. |
| 1692 | * |
| 1693 | * Decrement the reference count to balance the iget() done |
| 1694 | * earlier on. |
| 1695 | */ |
| 1696 | if (found->d_inode) { |
| 1697 | if (unlikely(found->d_inode != inode)) { |
| 1698 | /* This can't happen because bad inodes are unhashed. */ |
| 1699 | BUG_ON(!is_bad_inode(inode)); |
| 1700 | BUG_ON(!is_bad_inode(found->d_inode)); |
| 1701 | } |
| 1702 | iput(inode); |
| 1703 | return found; |
| 1704 | } |
| 1705 | |
| 1706 | /* |
| 1707 | * Negative dentry: instantiate it unless the inode is a directory and |
| 1708 | * already has a dentry. |
| 1709 | */ |
| 1710 | spin_lock(&inode->i_lock); |
| 1711 | if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) { |
| 1712 | __d_instantiate(found, inode); |
| 1713 | spin_unlock(&inode->i_lock); |
| 1714 | security_d_instantiate(found, inode); |
| 1715 | return found; |
| 1716 | } |
| 1717 | |
| 1718 | /* |
| 1719 | * In case a directory already has a (disconnected) entry grab a |
| 1720 | * reference to it, move it in place and use it. |
| 1721 | */ |
| 1722 | new = list_entry(inode->i_dentry.next, struct dentry, d_alias); |
| 1723 | __dget(new); |
| 1724 | spin_unlock(&inode->i_lock); |
| 1725 | security_d_instantiate(found, inode); |
| 1726 | d_move(new, found); |
| 1727 | iput(inode); |
| 1728 | dput(found); |
| 1729 | return new; |
| 1730 | |
| 1731 | err_out: |
| 1732 | iput(inode); |
| 1733 | return ERR_PTR(error); |
| 1734 | } |
| 1735 | EXPORT_SYMBOL(d_add_ci); |
| 1736 | |
| 1737 | /** |
| 1738 | * __d_lookup_rcu - search for a dentry (racy, store-free) |
| 1739 | * @parent: parent dentry |
| 1740 | * @name: qstr of name we wish to find |
| 1741 | * @seq: returns d_seq value at the point where the dentry was found |
| 1742 | * @inode: returns dentry->d_inode when the inode was found valid. |
| 1743 | * Returns: dentry, or NULL |
| 1744 | * |
| 1745 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name |
| 1746 | * resolution (store-free path walking) design described in |
| 1747 | * Documentation/filesystems/path-lookup.txt. |
| 1748 | * |
| 1749 | * This is not to be used outside core vfs. |
| 1750 | * |
| 1751 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock |
| 1752 | * held, and rcu_read_lock held. The returned dentry must not be stored into |
| 1753 | * without taking d_lock and checking d_seq sequence count against @seq |
| 1754 | * returned here. |
| 1755 | * |
| 1756 | * A refcount may be taken on the found dentry with the __d_rcu_to_refcount |
| 1757 | * function. |
| 1758 | * |
| 1759 | * Alternatively, __d_lookup_rcu may be called again to look up the child of |
| 1760 | * the returned dentry, so long as its parent's seqlock is checked after the |
| 1761 | * child is looked up. Thus, an interlocking stepping of sequence lock checks |
| 1762 | * is formed, giving integrity down the path walk. |
| 1763 | */ |
| 1764 | struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name, |
| 1765 | unsigned *seq, struct inode **inode) |
| 1766 | { |
| 1767 | unsigned int len = name->len; |
| 1768 | unsigned int hash = name->hash; |
| 1769 | const unsigned char *str = name->name; |
| 1770 | struct hlist_bl_head *b = d_hash(parent, hash); |
| 1771 | struct hlist_bl_node *node; |
| 1772 | struct dentry *dentry; |
| 1773 | |
| 1774 | /* |
| 1775 | * Note: There is significant duplication with __d_lookup_rcu which is |
| 1776 | * required to prevent single threaded performance regressions |
| 1777 | * especially on architectures where smp_rmb (in seqcounts) are costly. |
| 1778 | * Keep the two functions in sync. |
| 1779 | */ |
| 1780 | |
| 1781 | /* |
| 1782 | * The hash list is protected using RCU. |
| 1783 | * |
| 1784 | * Carefully use d_seq when comparing a candidate dentry, to avoid |
| 1785 | * races with d_move(). |
| 1786 | * |
| 1787 | * It is possible that concurrent renames can mess up our list |
| 1788 | * walk here and result in missing our dentry, resulting in the |
| 1789 | * false-negative result. d_lookup() protects against concurrent |
| 1790 | * renames using rename_lock seqlock. |
| 1791 | * |
| 1792 | * See Documentation/filesystems/path-lookup.txt for more details. |
| 1793 | */ |
| 1794 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
| 1795 | struct inode *i; |
| 1796 | const char *tname; |
| 1797 | int tlen; |
| 1798 | |
| 1799 | if (dentry->d_name.hash != hash) |
| 1800 | continue; |
| 1801 | |
| 1802 | seqretry: |
| 1803 | *seq = read_seqcount_begin(&dentry->d_seq); |
| 1804 | if (dentry->d_parent != parent) |
| 1805 | continue; |
| 1806 | if (d_unhashed(dentry)) |
| 1807 | continue; |
| 1808 | tlen = dentry->d_name.len; |
| 1809 | tname = dentry->d_name.name; |
| 1810 | i = dentry->d_inode; |
| 1811 | prefetch(tname); |
| 1812 | if (i) |
| 1813 | prefetch(i); |
| 1814 | /* |
| 1815 | * This seqcount check is required to ensure name and |
| 1816 | * len are loaded atomically, so as not to walk off the |
| 1817 | * edge of memory when walking. If we could load this |
| 1818 | * atomically some other way, we could drop this check. |
| 1819 | */ |
| 1820 | if (read_seqcount_retry(&dentry->d_seq, *seq)) |
| 1821 | goto seqretry; |
| 1822 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
| 1823 | if (parent->d_op->d_compare(parent, *inode, |
| 1824 | dentry, i, |
| 1825 | tlen, tname, name)) |
| 1826 | continue; |
| 1827 | } else { |
| 1828 | if (dentry_cmp(tname, tlen, str, len)) |
| 1829 | continue; |
| 1830 | } |
| 1831 | /* |
| 1832 | * No extra seqcount check is required after the name |
| 1833 | * compare. The caller must perform a seqcount check in |
| 1834 | * order to do anything useful with the returned dentry |
| 1835 | * anyway. |
| 1836 | */ |
| 1837 | *inode = i; |
| 1838 | return dentry; |
| 1839 | } |
| 1840 | return NULL; |
| 1841 | } |
| 1842 | |
| 1843 | /** |
| 1844 | * d_lookup - search for a dentry |
| 1845 | * @parent: parent dentry |
| 1846 | * @name: qstr of name we wish to find |
| 1847 | * Returns: dentry, or NULL |
| 1848 | * |
| 1849 | * d_lookup searches the children of the parent dentry for the name in |
| 1850 | * question. If the dentry is found its reference count is incremented and the |
| 1851 | * dentry is returned. The caller must use dput to free the entry when it has |
| 1852 | * finished using it. %NULL is returned if the dentry does not exist. |
| 1853 | */ |
| 1854 | struct dentry *d_lookup(struct dentry *parent, struct qstr *name) |
| 1855 | { |
| 1856 | struct dentry *dentry; |
| 1857 | unsigned seq; |
| 1858 | |
| 1859 | do { |
| 1860 | seq = read_seqbegin(&rename_lock); |
| 1861 | dentry = __d_lookup(parent, name); |
| 1862 | if (dentry) |
| 1863 | break; |
| 1864 | } while (read_seqretry(&rename_lock, seq)); |
| 1865 | return dentry; |
| 1866 | } |
| 1867 | EXPORT_SYMBOL(d_lookup); |
| 1868 | |
| 1869 | /** |
| 1870 | * __d_lookup - search for a dentry (racy) |
| 1871 | * @parent: parent dentry |
| 1872 | * @name: qstr of name we wish to find |
| 1873 | * Returns: dentry, or NULL |
| 1874 | * |
| 1875 | * __d_lookup is like d_lookup, however it may (rarely) return a |
| 1876 | * false-negative result due to unrelated rename activity. |
| 1877 | * |
| 1878 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, |
| 1879 | * however it must be used carefully, eg. with a following d_lookup in |
| 1880 | * the case of failure. |
| 1881 | * |
| 1882 | * __d_lookup callers must be commented. |
| 1883 | */ |
| 1884 | struct dentry *__d_lookup(struct dentry *parent, struct qstr *name) |
| 1885 | { |
| 1886 | unsigned int len = name->len; |
| 1887 | unsigned int hash = name->hash; |
| 1888 | const unsigned char *str = name->name; |
| 1889 | struct hlist_bl_head *b = d_hash(parent, hash); |
| 1890 | struct hlist_bl_node *node; |
| 1891 | struct dentry *found = NULL; |
| 1892 | struct dentry *dentry; |
| 1893 | |
| 1894 | /* |
| 1895 | * Note: There is significant duplication with __d_lookup_rcu which is |
| 1896 | * required to prevent single threaded performance regressions |
| 1897 | * especially on architectures where smp_rmb (in seqcounts) are costly. |
| 1898 | * Keep the two functions in sync. |
| 1899 | */ |
| 1900 | |
| 1901 | /* |
| 1902 | * The hash list is protected using RCU. |
| 1903 | * |
| 1904 | * Take d_lock when comparing a candidate dentry, to avoid races |
| 1905 | * with d_move(). |
| 1906 | * |
| 1907 | * It is possible that concurrent renames can mess up our list |
| 1908 | * walk here and result in missing our dentry, resulting in the |
| 1909 | * false-negative result. d_lookup() protects against concurrent |
| 1910 | * renames using rename_lock seqlock. |
| 1911 | * |
| 1912 | * See Documentation/filesystems/path-lookup.txt for more details. |
| 1913 | */ |
| 1914 | rcu_read_lock(); |
| 1915 | |
| 1916 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
| 1917 | const char *tname; |
| 1918 | int tlen; |
| 1919 | |
| 1920 | if (dentry->d_name.hash != hash) |
| 1921 | continue; |
| 1922 | |
| 1923 | spin_lock(&dentry->d_lock); |
| 1924 | if (dentry->d_parent != parent) |
| 1925 | goto next; |
| 1926 | if (d_unhashed(dentry)) |
| 1927 | goto next; |
| 1928 | |
| 1929 | /* |
| 1930 | * It is safe to compare names since d_move() cannot |
| 1931 | * change the qstr (protected by d_lock). |
| 1932 | */ |
| 1933 | tlen = dentry->d_name.len; |
| 1934 | tname = dentry->d_name.name; |
| 1935 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
| 1936 | if (parent->d_op->d_compare(parent, parent->d_inode, |
| 1937 | dentry, dentry->d_inode, |
| 1938 | tlen, tname, name)) |
| 1939 | goto next; |
| 1940 | } else { |
| 1941 | if (dentry_cmp(tname, tlen, str, len)) |
| 1942 | goto next; |
| 1943 | } |
| 1944 | |
| 1945 | dentry->d_count++; |
| 1946 | found = dentry; |
| 1947 | spin_unlock(&dentry->d_lock); |
| 1948 | break; |
| 1949 | next: |
| 1950 | spin_unlock(&dentry->d_lock); |
| 1951 | } |
| 1952 | rcu_read_unlock(); |
| 1953 | |
| 1954 | return found; |
| 1955 | } |
| 1956 | |
| 1957 | /** |
| 1958 | * d_hash_and_lookup - hash the qstr then search for a dentry |
| 1959 | * @dir: Directory to search in |
| 1960 | * @name: qstr of name we wish to find |
| 1961 | * |
| 1962 | * On hash failure or on lookup failure NULL is returned. |
| 1963 | */ |
| 1964 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) |
| 1965 | { |
| 1966 | struct dentry *dentry = NULL; |
| 1967 | |
| 1968 | /* |
| 1969 | * Check for a fs-specific hash function. Note that we must |
| 1970 | * calculate the standard hash first, as the d_op->d_hash() |
| 1971 | * routine may choose to leave the hash value unchanged. |
| 1972 | */ |
| 1973 | name->hash = full_name_hash(name->name, name->len); |
| 1974 | if (dir->d_flags & DCACHE_OP_HASH) { |
| 1975 | if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0) |
| 1976 | goto out; |
| 1977 | } |
| 1978 | dentry = d_lookup(dir, name); |
| 1979 | out: |
| 1980 | return dentry; |
| 1981 | } |
| 1982 | |
| 1983 | /** |
| 1984 | * d_validate - verify dentry provided from insecure source (deprecated) |
| 1985 | * @dentry: The dentry alleged to be valid child of @dparent |
| 1986 | * @dparent: The parent dentry (known to be valid) |
| 1987 | * |
| 1988 | * An insecure source has sent us a dentry, here we verify it and dget() it. |
| 1989 | * This is used by ncpfs in its readdir implementation. |
| 1990 | * Zero is returned in the dentry is invalid. |
| 1991 | * |
| 1992 | * This function is slow for big directories, and deprecated, do not use it. |
| 1993 | */ |
| 1994 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
| 1995 | { |
| 1996 | struct dentry *child; |
| 1997 | |
| 1998 | spin_lock(&dparent->d_lock); |
| 1999 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
| 2000 | if (dentry == child) { |
| 2001 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
| 2002 | __dget_dlock(dentry); |
| 2003 | spin_unlock(&dentry->d_lock); |
| 2004 | spin_unlock(&dparent->d_lock); |
| 2005 | return 1; |
| 2006 | } |
| 2007 | } |
| 2008 | spin_unlock(&dparent->d_lock); |
| 2009 | |
| 2010 | return 0; |
| 2011 | } |
| 2012 | EXPORT_SYMBOL(d_validate); |
| 2013 | |
| 2014 | /* |
| 2015 | * When a file is deleted, we have two options: |
| 2016 | * - turn this dentry into a negative dentry |
| 2017 | * - unhash this dentry and free it. |
| 2018 | * |
| 2019 | * Usually, we want to just turn this into |
| 2020 | * a negative dentry, but if anybody else is |
| 2021 | * currently using the dentry or the inode |
| 2022 | * we can't do that and we fall back on removing |
| 2023 | * it from the hash queues and waiting for |
| 2024 | * it to be deleted later when it has no users |
| 2025 | */ |
| 2026 | |
| 2027 | /** |
| 2028 | * d_delete - delete a dentry |
| 2029 | * @dentry: The dentry to delete |
| 2030 | * |
| 2031 | * Turn the dentry into a negative dentry if possible, otherwise |
| 2032 | * remove it from the hash queues so it can be deleted later |
| 2033 | */ |
| 2034 | |
| 2035 | void d_delete(struct dentry * dentry) |
| 2036 | { |
| 2037 | struct inode *inode; |
| 2038 | int isdir = 0; |
| 2039 | /* |
| 2040 | * Are we the only user? |
| 2041 | */ |
| 2042 | again: |
| 2043 | spin_lock(&dentry->d_lock); |
| 2044 | inode = dentry->d_inode; |
| 2045 | isdir = S_ISDIR(inode->i_mode); |
| 2046 | if (dentry->d_count == 1) { |
| 2047 | if (inode && !spin_trylock(&inode->i_lock)) { |
| 2048 | spin_unlock(&dentry->d_lock); |
| 2049 | cpu_relax(); |
| 2050 | goto again; |
| 2051 | } |
| 2052 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
| 2053 | dentry_unlink_inode(dentry); |
| 2054 | fsnotify_nameremove(dentry, isdir); |
| 2055 | return; |
| 2056 | } |
| 2057 | |
| 2058 | if (!d_unhashed(dentry)) |
| 2059 | __d_drop(dentry); |
| 2060 | |
| 2061 | spin_unlock(&dentry->d_lock); |
| 2062 | |
| 2063 | fsnotify_nameremove(dentry, isdir); |
| 2064 | } |
| 2065 | EXPORT_SYMBOL(d_delete); |
| 2066 | |
| 2067 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
| 2068 | { |
| 2069 | BUG_ON(!d_unhashed(entry)); |
| 2070 | hlist_bl_lock(b); |
| 2071 | entry->d_flags |= DCACHE_RCUACCESS; |
| 2072 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
| 2073 | hlist_bl_unlock(b); |
| 2074 | } |
| 2075 | |
| 2076 | static void _d_rehash(struct dentry * entry) |
| 2077 | { |
| 2078 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); |
| 2079 | } |
| 2080 | |
| 2081 | /** |
| 2082 | * d_rehash - add an entry back to the hash |
| 2083 | * @entry: dentry to add to the hash |
| 2084 | * |
| 2085 | * Adds a dentry to the hash according to its name. |
| 2086 | */ |
| 2087 | |
| 2088 | void d_rehash(struct dentry * entry) |
| 2089 | { |
| 2090 | spin_lock(&entry->d_lock); |
| 2091 | _d_rehash(entry); |
| 2092 | spin_unlock(&entry->d_lock); |
| 2093 | } |
| 2094 | EXPORT_SYMBOL(d_rehash); |
| 2095 | |
| 2096 | /** |
| 2097 | * dentry_update_name_case - update case insensitive dentry with a new name |
| 2098 | * @dentry: dentry to be updated |
| 2099 | * @name: new name |
| 2100 | * |
| 2101 | * Update a case insensitive dentry with new case of name. |
| 2102 | * |
| 2103 | * dentry must have been returned by d_lookup with name @name. Old and new |
| 2104 | * name lengths must match (ie. no d_compare which allows mismatched name |
| 2105 | * lengths). |
| 2106 | * |
| 2107 | * Parent inode i_mutex must be held over d_lookup and into this call (to |
| 2108 | * keep renames and concurrent inserts, and readdir(2) away). |
| 2109 | */ |
| 2110 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) |
| 2111 | { |
| 2112 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
| 2113 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
| 2114 | |
| 2115 | spin_lock(&dentry->d_lock); |
| 2116 | write_seqcount_begin(&dentry->d_seq); |
| 2117 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
| 2118 | write_seqcount_end(&dentry->d_seq); |
| 2119 | spin_unlock(&dentry->d_lock); |
| 2120 | } |
| 2121 | EXPORT_SYMBOL(dentry_update_name_case); |
| 2122 | |
| 2123 | static void switch_names(struct dentry *dentry, struct dentry *target) |
| 2124 | { |
| 2125 | if (dname_external(target)) { |
| 2126 | if (dname_external(dentry)) { |
| 2127 | /* |
| 2128 | * Both external: swap the pointers |
| 2129 | */ |
| 2130 | swap(target->d_name.name, dentry->d_name.name); |
| 2131 | } else { |
| 2132 | /* |
| 2133 | * dentry:internal, target:external. Steal target's |
| 2134 | * storage and make target internal. |
| 2135 | */ |
| 2136 | memcpy(target->d_iname, dentry->d_name.name, |
| 2137 | dentry->d_name.len + 1); |
| 2138 | dentry->d_name.name = target->d_name.name; |
| 2139 | target->d_name.name = target->d_iname; |
| 2140 | } |
| 2141 | } else { |
| 2142 | if (dname_external(dentry)) { |
| 2143 | /* |
| 2144 | * dentry:external, target:internal. Give dentry's |
| 2145 | * storage to target and make dentry internal |
| 2146 | */ |
| 2147 | memcpy(dentry->d_iname, target->d_name.name, |
| 2148 | target->d_name.len + 1); |
| 2149 | target->d_name.name = dentry->d_name.name; |
| 2150 | dentry->d_name.name = dentry->d_iname; |
| 2151 | } else { |
| 2152 | /* |
| 2153 | * Both are internal. Just copy target to dentry |
| 2154 | */ |
| 2155 | memcpy(dentry->d_iname, target->d_name.name, |
| 2156 | target->d_name.len + 1); |
| 2157 | dentry->d_name.len = target->d_name.len; |
| 2158 | return; |
| 2159 | } |
| 2160 | } |
| 2161 | swap(dentry->d_name.len, target->d_name.len); |
| 2162 | } |
| 2163 | |
| 2164 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
| 2165 | { |
| 2166 | /* |
| 2167 | * XXXX: do we really need to take target->d_lock? |
| 2168 | */ |
| 2169 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) |
| 2170 | spin_lock(&target->d_parent->d_lock); |
| 2171 | else { |
| 2172 | if (d_ancestor(dentry->d_parent, target->d_parent)) { |
| 2173 | spin_lock(&dentry->d_parent->d_lock); |
| 2174 | spin_lock_nested(&target->d_parent->d_lock, |
| 2175 | DENTRY_D_LOCK_NESTED); |
| 2176 | } else { |
| 2177 | spin_lock(&target->d_parent->d_lock); |
| 2178 | spin_lock_nested(&dentry->d_parent->d_lock, |
| 2179 | DENTRY_D_LOCK_NESTED); |
| 2180 | } |
| 2181 | } |
| 2182 | if (target < dentry) { |
| 2183 | spin_lock_nested(&target->d_lock, 2); |
| 2184 | spin_lock_nested(&dentry->d_lock, 3); |
| 2185 | } else { |
| 2186 | spin_lock_nested(&dentry->d_lock, 2); |
| 2187 | spin_lock_nested(&target->d_lock, 3); |
| 2188 | } |
| 2189 | } |
| 2190 | |
| 2191 | static void dentry_unlock_parents_for_move(struct dentry *dentry, |
| 2192 | struct dentry *target) |
| 2193 | { |
| 2194 | if (target->d_parent != dentry->d_parent) |
| 2195 | spin_unlock(&dentry->d_parent->d_lock); |
| 2196 | if (target->d_parent != target) |
| 2197 | spin_unlock(&target->d_parent->d_lock); |
| 2198 | } |
| 2199 | |
| 2200 | /* |
| 2201 | * When switching names, the actual string doesn't strictly have to |
| 2202 | * be preserved in the target - because we're dropping the target |
| 2203 | * anyway. As such, we can just do a simple memcpy() to copy over |
| 2204 | * the new name before we switch. |
| 2205 | * |
| 2206 | * Note that we have to be a lot more careful about getting the hash |
| 2207 | * switched - we have to switch the hash value properly even if it |
| 2208 | * then no longer matches the actual (corrupted) string of the target. |
| 2209 | * The hash value has to match the hash queue that the dentry is on.. |
| 2210 | */ |
| 2211 | /* |
| 2212 | * d_move - move a dentry |
| 2213 | * @dentry: entry to move |
| 2214 | * @target: new dentry |
| 2215 | * |
| 2216 | * Update the dcache to reflect the move of a file name. Negative |
| 2217 | * dcache entries should not be moved in this way. |
| 2218 | */ |
| 2219 | void d_move(struct dentry * dentry, struct dentry * target) |
| 2220 | { |
| 2221 | if (!dentry->d_inode) |
| 2222 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); |
| 2223 | |
| 2224 | BUG_ON(d_ancestor(dentry, target)); |
| 2225 | BUG_ON(d_ancestor(target, dentry)); |
| 2226 | |
| 2227 | write_seqlock(&rename_lock); |
| 2228 | |
| 2229 | dentry_lock_for_move(dentry, target); |
| 2230 | |
| 2231 | write_seqcount_begin(&dentry->d_seq); |
| 2232 | write_seqcount_begin(&target->d_seq); |
| 2233 | |
| 2234 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
| 2235 | |
| 2236 | /* |
| 2237 | * Move the dentry to the target hash queue. Don't bother checking |
| 2238 | * for the same hash queue because of how unlikely it is. |
| 2239 | */ |
| 2240 | __d_drop(dentry); |
| 2241 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
| 2242 | |
| 2243 | /* Unhash the target: dput() will then get rid of it */ |
| 2244 | __d_drop(target); |
| 2245 | |
| 2246 | list_del(&dentry->d_u.d_child); |
| 2247 | list_del(&target->d_u.d_child); |
| 2248 | |
| 2249 | /* Switch the names.. */ |
| 2250 | switch_names(dentry, target); |
| 2251 | swap(dentry->d_name.hash, target->d_name.hash); |
| 2252 | |
| 2253 | /* ... and switch the parents */ |
| 2254 | if (IS_ROOT(dentry)) { |
| 2255 | dentry->d_parent = target->d_parent; |
| 2256 | target->d_parent = target; |
| 2257 | INIT_LIST_HEAD(&target->d_u.d_child); |
| 2258 | } else { |
| 2259 | swap(dentry->d_parent, target->d_parent); |
| 2260 | |
| 2261 | /* And add them back to the (new) parent lists */ |
| 2262 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
| 2263 | } |
| 2264 | |
| 2265 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
| 2266 | |
| 2267 | write_seqcount_end(&target->d_seq); |
| 2268 | write_seqcount_end(&dentry->d_seq); |
| 2269 | |
| 2270 | dentry_unlock_parents_for_move(dentry, target); |
| 2271 | spin_unlock(&target->d_lock); |
| 2272 | fsnotify_d_move(dentry); |
| 2273 | spin_unlock(&dentry->d_lock); |
| 2274 | write_sequnlock(&rename_lock); |
| 2275 | } |
| 2276 | EXPORT_SYMBOL(d_move); |
| 2277 | |
| 2278 | /** |
| 2279 | * d_ancestor - search for an ancestor |
| 2280 | * @p1: ancestor dentry |
| 2281 | * @p2: child dentry |
| 2282 | * |
| 2283 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is |
| 2284 | * an ancestor of p2, else NULL. |
| 2285 | */ |
| 2286 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
| 2287 | { |
| 2288 | struct dentry *p; |
| 2289 | |
| 2290 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
| 2291 | if (p->d_parent == p1) |
| 2292 | return p; |
| 2293 | } |
| 2294 | return NULL; |
| 2295 | } |
| 2296 | |
| 2297 | /* |
| 2298 | * This helper attempts to cope with remotely renamed directories |
| 2299 | * |
| 2300 | * It assumes that the caller is already holding |
| 2301 | * dentry->d_parent->d_inode->i_mutex and the inode->i_lock |
| 2302 | * |
| 2303 | * Note: If ever the locking in lock_rename() changes, then please |
| 2304 | * remember to update this too... |
| 2305 | */ |
| 2306 | static struct dentry *__d_unalias(struct inode *inode, |
| 2307 | struct dentry *dentry, struct dentry *alias) |
| 2308 | { |
| 2309 | struct mutex *m1 = NULL, *m2 = NULL; |
| 2310 | struct dentry *ret; |
| 2311 | |
| 2312 | /* If alias and dentry share a parent, then no extra locks required */ |
| 2313 | if (alias->d_parent == dentry->d_parent) |
| 2314 | goto out_unalias; |
| 2315 | |
| 2316 | /* Check for loops */ |
| 2317 | ret = ERR_PTR(-ELOOP); |
| 2318 | if (d_ancestor(alias, dentry)) |
| 2319 | goto out_err; |
| 2320 | |
| 2321 | /* See lock_rename() */ |
| 2322 | ret = ERR_PTR(-EBUSY); |
| 2323 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
| 2324 | goto out_err; |
| 2325 | m1 = &dentry->d_sb->s_vfs_rename_mutex; |
| 2326 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) |
| 2327 | goto out_err; |
| 2328 | m2 = &alias->d_parent->d_inode->i_mutex; |
| 2329 | out_unalias: |
| 2330 | d_move(alias, dentry); |
| 2331 | ret = alias; |
| 2332 | out_err: |
| 2333 | spin_unlock(&inode->i_lock); |
| 2334 | if (m2) |
| 2335 | mutex_unlock(m2); |
| 2336 | if (m1) |
| 2337 | mutex_unlock(m1); |
| 2338 | return ret; |
| 2339 | } |
| 2340 | |
| 2341 | /* |
| 2342 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a |
| 2343 | * named dentry in place of the dentry to be replaced. |
| 2344 | * returns with anon->d_lock held! |
| 2345 | */ |
| 2346 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) |
| 2347 | { |
| 2348 | struct dentry *dparent, *aparent; |
| 2349 | |
| 2350 | dentry_lock_for_move(anon, dentry); |
| 2351 | |
| 2352 | write_seqcount_begin(&dentry->d_seq); |
| 2353 | write_seqcount_begin(&anon->d_seq); |
| 2354 | |
| 2355 | dparent = dentry->d_parent; |
| 2356 | aparent = anon->d_parent; |
| 2357 | |
| 2358 | switch_names(dentry, anon); |
| 2359 | swap(dentry->d_name.hash, anon->d_name.hash); |
| 2360 | |
| 2361 | dentry->d_parent = (aparent == anon) ? dentry : aparent; |
| 2362 | list_del(&dentry->d_u.d_child); |
| 2363 | if (!IS_ROOT(dentry)) |
| 2364 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
| 2365 | else |
| 2366 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
| 2367 | |
| 2368 | anon->d_parent = (dparent == dentry) ? anon : dparent; |
| 2369 | list_del(&anon->d_u.d_child); |
| 2370 | if (!IS_ROOT(anon)) |
| 2371 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); |
| 2372 | else |
| 2373 | INIT_LIST_HEAD(&anon->d_u.d_child); |
| 2374 | |
| 2375 | write_seqcount_end(&dentry->d_seq); |
| 2376 | write_seqcount_end(&anon->d_seq); |
| 2377 | |
| 2378 | dentry_unlock_parents_for_move(anon, dentry); |
| 2379 | spin_unlock(&dentry->d_lock); |
| 2380 | |
| 2381 | /* anon->d_lock still locked, returns locked */ |
| 2382 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
| 2383 | } |
| 2384 | |
| 2385 | /** |
| 2386 | * d_materialise_unique - introduce an inode into the tree |
| 2387 | * @dentry: candidate dentry |
| 2388 | * @inode: inode to bind to the dentry, to which aliases may be attached |
| 2389 | * |
| 2390 | * Introduces an dentry into the tree, substituting an extant disconnected |
| 2391 | * root directory alias in its place if there is one |
| 2392 | */ |
| 2393 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) |
| 2394 | { |
| 2395 | struct dentry *actual; |
| 2396 | |
| 2397 | BUG_ON(!d_unhashed(dentry)); |
| 2398 | |
| 2399 | if (!inode) { |
| 2400 | actual = dentry; |
| 2401 | __d_instantiate(dentry, NULL); |
| 2402 | d_rehash(actual); |
| 2403 | goto out_nolock; |
| 2404 | } |
| 2405 | |
| 2406 | spin_lock(&inode->i_lock); |
| 2407 | |
| 2408 | if (S_ISDIR(inode->i_mode)) { |
| 2409 | struct dentry *alias; |
| 2410 | |
| 2411 | /* Does an aliased dentry already exist? */ |
| 2412 | alias = __d_find_alias(inode, 0); |
| 2413 | if (alias) { |
| 2414 | actual = alias; |
| 2415 | /* Is this an anonymous mountpoint that we could splice |
| 2416 | * into our tree? */ |
| 2417 | if (IS_ROOT(alias)) { |
| 2418 | __d_materialise_dentry(dentry, alias); |
| 2419 | __d_drop(alias); |
| 2420 | goto found; |
| 2421 | } |
| 2422 | /* Nope, but we must(!) avoid directory aliasing */ |
| 2423 | actual = __d_unalias(inode, dentry, alias); |
| 2424 | if (IS_ERR(actual)) |
| 2425 | dput(alias); |
| 2426 | goto out_nolock; |
| 2427 | } |
| 2428 | } |
| 2429 | |
| 2430 | /* Add a unique reference */ |
| 2431 | actual = __d_instantiate_unique(dentry, inode); |
| 2432 | if (!actual) |
| 2433 | actual = dentry; |
| 2434 | else |
| 2435 | BUG_ON(!d_unhashed(actual)); |
| 2436 | |
| 2437 | spin_lock(&actual->d_lock); |
| 2438 | found: |
| 2439 | _d_rehash(actual); |
| 2440 | spin_unlock(&actual->d_lock); |
| 2441 | spin_unlock(&inode->i_lock); |
| 2442 | out_nolock: |
| 2443 | if (actual == dentry) { |
| 2444 | security_d_instantiate(dentry, inode); |
| 2445 | return NULL; |
| 2446 | } |
| 2447 | |
| 2448 | iput(inode); |
| 2449 | return actual; |
| 2450 | } |
| 2451 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
| 2452 | |
| 2453 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
| 2454 | { |
| 2455 | *buflen -= namelen; |
| 2456 | if (*buflen < 0) |
| 2457 | return -ENAMETOOLONG; |
| 2458 | *buffer -= namelen; |
| 2459 | memcpy(*buffer, str, namelen); |
| 2460 | return 0; |
| 2461 | } |
| 2462 | |
| 2463 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
| 2464 | { |
| 2465 | return prepend(buffer, buflen, name->name, name->len); |
| 2466 | } |
| 2467 | |
| 2468 | /** |
| 2469 | * prepend_path - Prepend path string to a buffer |
| 2470 | * @path: the dentry/vfsmount to report |
| 2471 | * @root: root vfsmnt/dentry (may be modified by this function) |
| 2472 | * @buffer: pointer to the end of the buffer |
| 2473 | * @buflen: pointer to buffer length |
| 2474 | * |
| 2475 | * Caller holds the rename_lock. |
| 2476 | * |
| 2477 | * If path is not reachable from the supplied root, then the value of |
| 2478 | * root is changed (without modifying refcounts). |
| 2479 | */ |
| 2480 | static int prepend_path(const struct path *path, struct path *root, |
| 2481 | char **buffer, int *buflen) |
| 2482 | { |
| 2483 | struct dentry *dentry = path->dentry; |
| 2484 | struct vfsmount *vfsmnt = path->mnt; |
| 2485 | bool slash = false; |
| 2486 | int error = 0; |
| 2487 | |
| 2488 | br_read_lock(vfsmount_lock); |
| 2489 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
| 2490 | struct dentry * parent; |
| 2491 | |
| 2492 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
| 2493 | /* Global root? */ |
| 2494 | if (vfsmnt->mnt_parent == vfsmnt) { |
| 2495 | goto global_root; |
| 2496 | } |
| 2497 | dentry = vfsmnt->mnt_mountpoint; |
| 2498 | vfsmnt = vfsmnt->mnt_parent; |
| 2499 | continue; |
| 2500 | } |
| 2501 | parent = dentry->d_parent; |
| 2502 | prefetch(parent); |
| 2503 | spin_lock(&dentry->d_lock); |
| 2504 | error = prepend_name(buffer, buflen, &dentry->d_name); |
| 2505 | spin_unlock(&dentry->d_lock); |
| 2506 | if (!error) |
| 2507 | error = prepend(buffer, buflen, "/", 1); |
| 2508 | if (error) |
| 2509 | break; |
| 2510 | |
| 2511 | slash = true; |
| 2512 | dentry = parent; |
| 2513 | } |
| 2514 | |
| 2515 | out: |
| 2516 | if (!error && !slash) |
| 2517 | error = prepend(buffer, buflen, "/", 1); |
| 2518 | |
| 2519 | br_read_unlock(vfsmount_lock); |
| 2520 | return error; |
| 2521 | |
| 2522 | global_root: |
| 2523 | /* |
| 2524 | * Filesystems needing to implement special "root names" |
| 2525 | * should do so with ->d_dname() |
| 2526 | */ |
| 2527 | if (IS_ROOT(dentry) && |
| 2528 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { |
| 2529 | WARN(1, "Root dentry has weird name <%.*s>\n", |
| 2530 | (int) dentry->d_name.len, dentry->d_name.name); |
| 2531 | } |
| 2532 | root->mnt = vfsmnt; |
| 2533 | root->dentry = dentry; |
| 2534 | goto out; |
| 2535 | } |
| 2536 | |
| 2537 | /** |
| 2538 | * __d_path - return the path of a dentry |
| 2539 | * @path: the dentry/vfsmount to report |
| 2540 | * @root: root vfsmnt/dentry (may be modified by this function) |
| 2541 | * @buf: buffer to return value in |
| 2542 | * @buflen: buffer length |
| 2543 | * |
| 2544 | * Convert a dentry into an ASCII path name. |
| 2545 | * |
| 2546 | * Returns a pointer into the buffer or an error code if the |
| 2547 | * path was too long. |
| 2548 | * |
| 2549 | * "buflen" should be positive. |
| 2550 | * |
| 2551 | * If path is not reachable from the supplied root, then the value of |
| 2552 | * root is changed (without modifying refcounts). |
| 2553 | */ |
| 2554 | char *__d_path(const struct path *path, struct path *root, |
| 2555 | char *buf, int buflen) |
| 2556 | { |
| 2557 | char *res = buf + buflen; |
| 2558 | int error; |
| 2559 | |
| 2560 | prepend(&res, &buflen, "\0", 1); |
| 2561 | write_seqlock(&rename_lock); |
| 2562 | error = prepend_path(path, root, &res, &buflen); |
| 2563 | write_sequnlock(&rename_lock); |
| 2564 | |
| 2565 | if (error) |
| 2566 | return ERR_PTR(error); |
| 2567 | return res; |
| 2568 | } |
| 2569 | |
| 2570 | /* |
| 2571 | * same as __d_path but appends "(deleted)" for unlinked files. |
| 2572 | */ |
| 2573 | static int path_with_deleted(const struct path *path, struct path *root, |
| 2574 | char **buf, int *buflen) |
| 2575 | { |
| 2576 | prepend(buf, buflen, "\0", 1); |
| 2577 | if (d_unlinked(path->dentry)) { |
| 2578 | int error = prepend(buf, buflen, " (deleted)", 10); |
| 2579 | if (error) |
| 2580 | return error; |
| 2581 | } |
| 2582 | |
| 2583 | return prepend_path(path, root, buf, buflen); |
| 2584 | } |
| 2585 | |
| 2586 | static int prepend_unreachable(char **buffer, int *buflen) |
| 2587 | { |
| 2588 | return prepend(buffer, buflen, "(unreachable)", 13); |
| 2589 | } |
| 2590 | |
| 2591 | /** |
| 2592 | * d_path - return the path of a dentry |
| 2593 | * @path: path to report |
| 2594 | * @buf: buffer to return value in |
| 2595 | * @buflen: buffer length |
| 2596 | * |
| 2597 | * Convert a dentry into an ASCII path name. If the entry has been deleted |
| 2598 | * the string " (deleted)" is appended. Note that this is ambiguous. |
| 2599 | * |
| 2600 | * Returns a pointer into the buffer or an error code if the path was |
| 2601 | * too long. Note: Callers should use the returned pointer, not the passed |
| 2602 | * in buffer, to use the name! The implementation often starts at an offset |
| 2603 | * into the buffer, and may leave 0 bytes at the start. |
| 2604 | * |
| 2605 | * "buflen" should be positive. |
| 2606 | */ |
| 2607 | char *d_path(const struct path *path, char *buf, int buflen) |
| 2608 | { |
| 2609 | char *res = buf + buflen; |
| 2610 | struct path root; |
| 2611 | struct path tmp; |
| 2612 | int error; |
| 2613 | |
| 2614 | /* |
| 2615 | * We have various synthetic filesystems that never get mounted. On |
| 2616 | * these filesystems dentries are never used for lookup purposes, and |
| 2617 | * thus don't need to be hashed. They also don't need a name until a |
| 2618 | * user wants to identify the object in /proc/pid/fd/. The little hack |
| 2619 | * below allows us to generate a name for these objects on demand: |
| 2620 | */ |
| 2621 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
| 2622 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
| 2623 | |
| 2624 | get_fs_root(current->fs, &root); |
| 2625 | write_seqlock(&rename_lock); |
| 2626 | tmp = root; |
| 2627 | error = path_with_deleted(path, &tmp, &res, &buflen); |
| 2628 | if (error) |
| 2629 | res = ERR_PTR(error); |
| 2630 | write_sequnlock(&rename_lock); |
| 2631 | path_put(&root); |
| 2632 | return res; |
| 2633 | } |
| 2634 | EXPORT_SYMBOL(d_path); |
| 2635 | |
| 2636 | /** |
| 2637 | * d_path_with_unreachable - return the path of a dentry |
| 2638 | * @path: path to report |
| 2639 | * @buf: buffer to return value in |
| 2640 | * @buflen: buffer length |
| 2641 | * |
| 2642 | * The difference from d_path() is that this prepends "(unreachable)" |
| 2643 | * to paths which are unreachable from the current process' root. |
| 2644 | */ |
| 2645 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) |
| 2646 | { |
| 2647 | char *res = buf + buflen; |
| 2648 | struct path root; |
| 2649 | struct path tmp; |
| 2650 | int error; |
| 2651 | |
| 2652 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
| 2653 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
| 2654 | |
| 2655 | get_fs_root(current->fs, &root); |
| 2656 | write_seqlock(&rename_lock); |
| 2657 | tmp = root; |
| 2658 | error = path_with_deleted(path, &tmp, &res, &buflen); |
| 2659 | if (!error && !path_equal(&tmp, &root)) |
| 2660 | error = prepend_unreachable(&res, &buflen); |
| 2661 | write_sequnlock(&rename_lock); |
| 2662 | path_put(&root); |
| 2663 | if (error) |
| 2664 | res = ERR_PTR(error); |
| 2665 | |
| 2666 | return res; |
| 2667 | } |
| 2668 | |
| 2669 | /* |
| 2670 | * Helper function for dentry_operations.d_dname() members |
| 2671 | */ |
| 2672 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, |
| 2673 | const char *fmt, ...) |
| 2674 | { |
| 2675 | va_list args; |
| 2676 | char temp[64]; |
| 2677 | int sz; |
| 2678 | |
| 2679 | va_start(args, fmt); |
| 2680 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; |
| 2681 | va_end(args); |
| 2682 | |
| 2683 | if (sz > sizeof(temp) || sz > buflen) |
| 2684 | return ERR_PTR(-ENAMETOOLONG); |
| 2685 | |
| 2686 | buffer += buflen - sz; |
| 2687 | return memcpy(buffer, temp, sz); |
| 2688 | } |
| 2689 | |
| 2690 | /* |
| 2691 | * Write full pathname from the root of the filesystem into the buffer. |
| 2692 | */ |
| 2693 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
| 2694 | { |
| 2695 | char *end = buf + buflen; |
| 2696 | char *retval; |
| 2697 | |
| 2698 | prepend(&end, &buflen, "\0", 1); |
| 2699 | if (buflen < 1) |
| 2700 | goto Elong; |
| 2701 | /* Get '/' right */ |
| 2702 | retval = end-1; |
| 2703 | *retval = '/'; |
| 2704 | |
| 2705 | while (!IS_ROOT(dentry)) { |
| 2706 | struct dentry *parent = dentry->d_parent; |
| 2707 | int error; |
| 2708 | |
| 2709 | prefetch(parent); |
| 2710 | spin_lock(&dentry->d_lock); |
| 2711 | error = prepend_name(&end, &buflen, &dentry->d_name); |
| 2712 | spin_unlock(&dentry->d_lock); |
| 2713 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) |
| 2714 | goto Elong; |
| 2715 | |
| 2716 | retval = end; |
| 2717 | dentry = parent; |
| 2718 | } |
| 2719 | return retval; |
| 2720 | Elong: |
| 2721 | return ERR_PTR(-ENAMETOOLONG); |
| 2722 | } |
| 2723 | |
| 2724 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) |
| 2725 | { |
| 2726 | char *retval; |
| 2727 | |
| 2728 | write_seqlock(&rename_lock); |
| 2729 | retval = __dentry_path(dentry, buf, buflen); |
| 2730 | write_sequnlock(&rename_lock); |
| 2731 | |
| 2732 | return retval; |
| 2733 | } |
| 2734 | EXPORT_SYMBOL(dentry_path_raw); |
| 2735 | |
| 2736 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) |
| 2737 | { |
| 2738 | char *p = NULL; |
| 2739 | char *retval; |
| 2740 | |
| 2741 | write_seqlock(&rename_lock); |
| 2742 | if (d_unlinked(dentry)) { |
| 2743 | p = buf + buflen; |
| 2744 | if (prepend(&p, &buflen, "//deleted", 10) != 0) |
| 2745 | goto Elong; |
| 2746 | buflen++; |
| 2747 | } |
| 2748 | retval = __dentry_path(dentry, buf, buflen); |
| 2749 | write_sequnlock(&rename_lock); |
| 2750 | if (!IS_ERR(retval) && p) |
| 2751 | *p = '/'; /* restore '/' overriden with '\0' */ |
| 2752 | return retval; |
| 2753 | Elong: |
| 2754 | return ERR_PTR(-ENAMETOOLONG); |
| 2755 | } |
| 2756 | |
| 2757 | /* |
| 2758 | * NOTE! The user-level library version returns a |
| 2759 | * character pointer. The kernel system call just |
| 2760 | * returns the length of the buffer filled (which |
| 2761 | * includes the ending '\0' character), or a negative |
| 2762 | * error value. So libc would do something like |
| 2763 | * |
| 2764 | * char *getcwd(char * buf, size_t size) |
| 2765 | * { |
| 2766 | * int retval; |
| 2767 | * |
| 2768 | * retval = sys_getcwd(buf, size); |
| 2769 | * if (retval >= 0) |
| 2770 | * return buf; |
| 2771 | * errno = -retval; |
| 2772 | * return NULL; |
| 2773 | * } |
| 2774 | */ |
| 2775 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
| 2776 | { |
| 2777 | int error; |
| 2778 | struct path pwd, root; |
| 2779 | char *page = (char *) __get_free_page(GFP_USER); |
| 2780 | |
| 2781 | if (!page) |
| 2782 | return -ENOMEM; |
| 2783 | |
| 2784 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
| 2785 | |
| 2786 | error = -ENOENT; |
| 2787 | write_seqlock(&rename_lock); |
| 2788 | if (!d_unlinked(pwd.dentry)) { |
| 2789 | unsigned long len; |
| 2790 | struct path tmp = root; |
| 2791 | char *cwd = page + PAGE_SIZE; |
| 2792 | int buflen = PAGE_SIZE; |
| 2793 | |
| 2794 | prepend(&cwd, &buflen, "\0", 1); |
| 2795 | error = prepend_path(&pwd, &tmp, &cwd, &buflen); |
| 2796 | write_sequnlock(&rename_lock); |
| 2797 | |
| 2798 | if (error) |
| 2799 | goto out; |
| 2800 | |
| 2801 | /* Unreachable from current root */ |
| 2802 | if (!path_equal(&tmp, &root)) { |
| 2803 | error = prepend_unreachable(&cwd, &buflen); |
| 2804 | if (error) |
| 2805 | goto out; |
| 2806 | } |
| 2807 | |
| 2808 | error = -ERANGE; |
| 2809 | len = PAGE_SIZE + page - cwd; |
| 2810 | if (len <= size) { |
| 2811 | error = len; |
| 2812 | if (copy_to_user(buf, cwd, len)) |
| 2813 | error = -EFAULT; |
| 2814 | } |
| 2815 | } else { |
| 2816 | write_sequnlock(&rename_lock); |
| 2817 | } |
| 2818 | |
| 2819 | out: |
| 2820 | path_put(&pwd); |
| 2821 | path_put(&root); |
| 2822 | free_page((unsigned long) page); |
| 2823 | return error; |
| 2824 | } |
| 2825 | |
| 2826 | /* |
| 2827 | * Test whether new_dentry is a subdirectory of old_dentry. |
| 2828 | * |
| 2829 | * Trivially implemented using the dcache structure |
| 2830 | */ |
| 2831 | |
| 2832 | /** |
| 2833 | * is_subdir - is new dentry a subdirectory of old_dentry |
| 2834 | * @new_dentry: new dentry |
| 2835 | * @old_dentry: old dentry |
| 2836 | * |
| 2837 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). |
| 2838 | * Returns 0 otherwise. |
| 2839 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() |
| 2840 | */ |
| 2841 | |
| 2842 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
| 2843 | { |
| 2844 | int result; |
| 2845 | unsigned seq; |
| 2846 | |
| 2847 | if (new_dentry == old_dentry) |
| 2848 | return 1; |
| 2849 | |
| 2850 | do { |
| 2851 | /* for restarting inner loop in case of seq retry */ |
| 2852 | seq = read_seqbegin(&rename_lock); |
| 2853 | /* |
| 2854 | * Need rcu_readlock to protect against the d_parent trashing |
| 2855 | * due to d_move |
| 2856 | */ |
| 2857 | rcu_read_lock(); |
| 2858 | if (d_ancestor(old_dentry, new_dentry)) |
| 2859 | result = 1; |
| 2860 | else |
| 2861 | result = 0; |
| 2862 | rcu_read_unlock(); |
| 2863 | } while (read_seqretry(&rename_lock, seq)); |
| 2864 | |
| 2865 | return result; |
| 2866 | } |
| 2867 | |
| 2868 | int path_is_under(struct path *path1, struct path *path2) |
| 2869 | { |
| 2870 | struct vfsmount *mnt = path1->mnt; |
| 2871 | struct dentry *dentry = path1->dentry; |
| 2872 | int res; |
| 2873 | |
| 2874 | br_read_lock(vfsmount_lock); |
| 2875 | if (mnt != path2->mnt) { |
| 2876 | for (;;) { |
| 2877 | if (mnt->mnt_parent == mnt) { |
| 2878 | br_read_unlock(vfsmount_lock); |
| 2879 | return 0; |
| 2880 | } |
| 2881 | if (mnt->mnt_parent == path2->mnt) |
| 2882 | break; |
| 2883 | mnt = mnt->mnt_parent; |
| 2884 | } |
| 2885 | dentry = mnt->mnt_mountpoint; |
| 2886 | } |
| 2887 | res = is_subdir(dentry, path2->dentry); |
| 2888 | br_read_unlock(vfsmount_lock); |
| 2889 | return res; |
| 2890 | } |
| 2891 | EXPORT_SYMBOL(path_is_under); |
| 2892 | |
| 2893 | void d_genocide(struct dentry *root) |
| 2894 | { |
| 2895 | struct dentry *this_parent; |
| 2896 | struct list_head *next; |
| 2897 | unsigned seq; |
| 2898 | int locked = 0; |
| 2899 | |
| 2900 | seq = read_seqbegin(&rename_lock); |
| 2901 | again: |
| 2902 | this_parent = root; |
| 2903 | spin_lock(&this_parent->d_lock); |
| 2904 | repeat: |
| 2905 | next = this_parent->d_subdirs.next; |
| 2906 | resume: |
| 2907 | while (next != &this_parent->d_subdirs) { |
| 2908 | struct list_head *tmp = next; |
| 2909 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
| 2910 | next = tmp->next; |
| 2911 | |
| 2912 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
| 2913 | if (d_unhashed(dentry) || !dentry->d_inode) { |
| 2914 | spin_unlock(&dentry->d_lock); |
| 2915 | continue; |
| 2916 | } |
| 2917 | if (!list_empty(&dentry->d_subdirs)) { |
| 2918 | spin_unlock(&this_parent->d_lock); |
| 2919 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); |
| 2920 | this_parent = dentry; |
| 2921 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
| 2922 | goto repeat; |
| 2923 | } |
| 2924 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
| 2925 | dentry->d_flags |= DCACHE_GENOCIDE; |
| 2926 | dentry->d_count--; |
| 2927 | } |
| 2928 | spin_unlock(&dentry->d_lock); |
| 2929 | } |
| 2930 | if (this_parent != root) { |
| 2931 | struct dentry *child = this_parent; |
| 2932 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { |
| 2933 | this_parent->d_flags |= DCACHE_GENOCIDE; |
| 2934 | this_parent->d_count--; |
| 2935 | } |
| 2936 | this_parent = try_to_ascend(this_parent, locked, seq); |
| 2937 | if (!this_parent) |
| 2938 | goto rename_retry; |
| 2939 | next = child->d_u.d_child.next; |
| 2940 | goto resume; |
| 2941 | } |
| 2942 | spin_unlock(&this_parent->d_lock); |
| 2943 | if (!locked && read_seqretry(&rename_lock, seq)) |
| 2944 | goto rename_retry; |
| 2945 | if (locked) |
| 2946 | write_sequnlock(&rename_lock); |
| 2947 | return; |
| 2948 | |
| 2949 | rename_retry: |
| 2950 | locked = 1; |
| 2951 | write_seqlock(&rename_lock); |
| 2952 | goto again; |
| 2953 | } |
| 2954 | |
| 2955 | /** |
| 2956 | * find_inode_number - check for dentry with name |
| 2957 | * @dir: directory to check |
| 2958 | * @name: Name to find. |
| 2959 | * |
| 2960 | * Check whether a dentry already exists for the given name, |
| 2961 | * and return the inode number if it has an inode. Otherwise |
| 2962 | * 0 is returned. |
| 2963 | * |
| 2964 | * This routine is used to post-process directory listings for |
| 2965 | * filesystems using synthetic inode numbers, and is necessary |
| 2966 | * to keep getcwd() working. |
| 2967 | */ |
| 2968 | |
| 2969 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) |
| 2970 | { |
| 2971 | struct dentry * dentry; |
| 2972 | ino_t ino = 0; |
| 2973 | |
| 2974 | dentry = d_hash_and_lookup(dir, name); |
| 2975 | if (dentry) { |
| 2976 | if (dentry->d_inode) |
| 2977 | ino = dentry->d_inode->i_ino; |
| 2978 | dput(dentry); |
| 2979 | } |
| 2980 | return ino; |
| 2981 | } |
| 2982 | EXPORT_SYMBOL(find_inode_number); |
| 2983 | |
| 2984 | static __initdata unsigned long dhash_entries; |
| 2985 | static int __init set_dhash_entries(char *str) |
| 2986 | { |
| 2987 | if (!str) |
| 2988 | return 0; |
| 2989 | dhash_entries = simple_strtoul(str, &str, 0); |
| 2990 | return 1; |
| 2991 | } |
| 2992 | __setup("dhash_entries=", set_dhash_entries); |
| 2993 | |
| 2994 | static void __init dcache_init_early(void) |
| 2995 | { |
| 2996 | int loop; |
| 2997 | |
| 2998 | /* If hashes are distributed across NUMA nodes, defer |
| 2999 | * hash allocation until vmalloc space is available. |
| 3000 | */ |
| 3001 | if (hashdist) |
| 3002 | return; |
| 3003 | |
| 3004 | dentry_hashtable = |
| 3005 | alloc_large_system_hash("Dentry cache", |
| 3006 | sizeof(struct hlist_bl_head), |
| 3007 | dhash_entries, |
| 3008 | 13, |
| 3009 | HASH_EARLY, |
| 3010 | &d_hash_shift, |
| 3011 | &d_hash_mask, |
| 3012 | 0); |
| 3013 | |
| 3014 | for (loop = 0; loop < (1 << d_hash_shift); loop++) |
| 3015 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
| 3016 | } |
| 3017 | |
| 3018 | static void __init dcache_init(void) |
| 3019 | { |
| 3020 | int loop; |
| 3021 | |
| 3022 | /* |
| 3023 | * A constructor could be added for stable state like the lists, |
| 3024 | * but it is probably not worth it because of the cache nature |
| 3025 | * of the dcache. |
| 3026 | */ |
| 3027 | dentry_cache = KMEM_CACHE(dentry, |
| 3028 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); |
| 3029 | |
| 3030 | register_shrinker(&dcache_shrinker); |
| 3031 | |
| 3032 | /* Hash may have been set up in dcache_init_early */ |
| 3033 | if (!hashdist) |
| 3034 | return; |
| 3035 | |
| 3036 | dentry_hashtable = |
| 3037 | alloc_large_system_hash("Dentry cache", |
| 3038 | sizeof(struct hlist_bl_head), |
| 3039 | dhash_entries, |
| 3040 | 13, |
| 3041 | 0, |
| 3042 | &d_hash_shift, |
| 3043 | &d_hash_mask, |
| 3044 | 0); |
| 3045 | |
| 3046 | for (loop = 0; loop < (1 << d_hash_shift); loop++) |
| 3047 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
| 3048 | } |
| 3049 | |
| 3050 | /* SLAB cache for __getname() consumers */ |
| 3051 | struct kmem_cache *names_cachep __read_mostly; |
| 3052 | EXPORT_SYMBOL(names_cachep); |
| 3053 | |
| 3054 | EXPORT_SYMBOL(d_genocide); |
| 3055 | |
| 3056 | void __init vfs_caches_init_early(void) |
| 3057 | { |
| 3058 | dcache_init_early(); |
| 3059 | inode_init_early(); |
| 3060 | } |
| 3061 | |
| 3062 | void __init vfs_caches_init(unsigned long mempages) |
| 3063 | { |
| 3064 | unsigned long reserve; |
| 3065 | |
| 3066 | /* Base hash sizes on available memory, with a reserve equal to |
| 3067 | 150% of current kernel size */ |
| 3068 | |
| 3069 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); |
| 3070 | mempages -= reserve; |
| 3071 | |
| 3072 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, |
| 3073 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
| 3074 | |
| 3075 | dcache_init(); |
| 3076 | inode_init(); |
| 3077 | files_init(mempages); |
| 3078 | mnt_init(); |
| 3079 | bdev_cache_init(); |
| 3080 | chrdev_init(); |
| 3081 | } |