1 /* 2 * (C) 1997 Linus Torvalds 3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation) 4 */ 5 #include <linux/export.h> 6 #include <linux/fs.h> 7 #include <linux/mm.h> 8 #include <linux/backing-dev.h> 9 #include <linux/hash.h> 10 #include <linux/swap.h> 11 #include <linux/security.h> 12 #include <linux/cdev.h> 13 #include <linux/bootmem.h> 14 #include <linux/fsnotify.h> 15 #include <linux/mount.h> 16 #include <linux/posix_acl.h> 17 #include <linux/prefetch.h> 18 #include <linux/buffer_head.h> /* for inode_has_buffers */ 19 #include <linux/ratelimit.h> 20 #include "internal.h" 21 22 /* 23 * Inode locking rules: 24 * 25 * inode->i_lock protects: 26 * inode->i_state, inode->i_hash, __iget() 27 * inode->i_sb->s_inode_lru_lock protects: 28 * inode->i_sb->s_inode_lru, inode->i_lru 29 * inode_sb_list_lock protects: 30 * sb->s_inodes, inode->i_sb_list 31 * bdi->wb.list_lock protects: 32 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list 33 * inode_hash_lock protects: 34 * inode_hashtable, inode->i_hash 35 * 36 * Lock ordering: 37 * 38 * inode_sb_list_lock 39 * inode->i_lock 40 * inode->i_sb->s_inode_lru_lock 41 * 42 * bdi->wb.list_lock 43 * inode->i_lock 44 * 45 * inode_hash_lock 46 * inode_sb_list_lock 47 * inode->i_lock 48 * 49 * iunique_lock 50 * inode_hash_lock 51 */ 52 53 static unsigned int i_hash_mask __read_mostly; 54 static unsigned int i_hash_shift __read_mostly; 55 static struct hlist_head *inode_hashtable __read_mostly; 56 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock); 57 58 __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock); 59 60 /* 61 * Empty aops. Can be used for the cases where the user does not 62 * define any of the address_space operations. 63 */ 64 const struct address_space_operations empty_aops = { 65 }; 66 EXPORT_SYMBOL(empty_aops); 67 68 /* 69 * Statistics gathering.. 70 */ 71 struct inodes_stat_t inodes_stat; 72 73 static DEFINE_PER_CPU(unsigned int, nr_inodes); 74 static DEFINE_PER_CPU(unsigned int, nr_unused); 75 76 static struct kmem_cache *inode_cachep __read_mostly; 77 78 static int get_nr_inodes(void) 79 { 80 int i; 81 int sum = 0; 82 for_each_possible_cpu(i) 83 sum += per_cpu(nr_inodes, i); 84 return sum < 0 ? 0 : sum; 85 } 86 87 static inline int get_nr_inodes_unused(void) 88 { 89 int i; 90 int sum = 0; 91 for_each_possible_cpu(i) 92 sum += per_cpu(nr_unused, i); 93 return sum < 0 ? 0 : sum; 94 } 95 96 int get_nr_dirty_inodes(void) 97 { 98 /* not actually dirty inodes, but a wild approximation */ 99 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); 100 return nr_dirty > 0 ? nr_dirty : 0; 101 } 102 103 /* 104 * Handle nr_inode sysctl 105 */ 106 #ifdef CONFIG_SYSCTL 107 int proc_nr_inodes(ctl_table *table, int write, 108 void __user *buffer, size_t *lenp, loff_t *ppos) 109 { 110 inodes_stat.nr_inodes = get_nr_inodes(); 111 inodes_stat.nr_unused = get_nr_inodes_unused(); 112 return proc_dointvec(table, write, buffer, lenp, ppos); 113 } 114 #endif 115 116 /** 117 * inode_init_always - perform inode structure intialisation 118 * @sb: superblock inode belongs to 119 * @inode: inode to initialise 120 * 121 * These are initializations that need to be done on every inode 122 * allocation as the fields are not initialised by slab allocation. 123 */ 124 int inode_init_always(struct super_block *sb, struct inode *inode) 125 { 126 static const struct inode_operations empty_iops; 127 static const struct file_operations empty_fops; 128 struct address_space *const mapping = &inode->i_data; 129 130 inode->i_sb = sb; 131 inode->i_blkbits = sb->s_blocksize_bits; 132 inode->i_flags = 0; 133 atomic_set(&inode->i_count, 1); 134 inode->i_op = &empty_iops; 135 inode->i_fop = &empty_fops; 136 inode->__i_nlink = 1; 137 inode->i_opflags = 0; 138 i_uid_write(inode, 0); 139 i_gid_write(inode, 0); 140 atomic_set(&inode->i_writecount, 0); 141 inode->i_size = 0; 142 inode->i_blocks = 0; 143 inode->i_bytes = 0; 144 inode->i_generation = 0; 145 #ifdef CONFIG_QUOTA 146 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); 147 #endif 148 inode->i_pipe = NULL; 149 inode->i_bdev = NULL; 150 inode->i_cdev = NULL; 151 inode->i_rdev = 0; 152 inode->dirtied_when = 0; 153 154 if (security_inode_alloc(inode)) 155 goto out; 156 spin_lock_init(&inode->i_lock); 157 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); 158 159 mutex_init(&inode->i_mutex); 160 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); 161 162 atomic_set(&inode->i_dio_count, 0); 163 164 mapping->a_ops = &empty_aops; 165 mapping->host = inode; 166 mapping->flags = 0; 167 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); 168 mapping->private_data = NULL; 169 mapping->backing_dev_info = &default_backing_dev_info; 170 mapping->writeback_index = 0; 171 172 /* 173 * If the block_device provides a backing_dev_info for client 174 * inodes then use that. Otherwise the inode share the bdev's 175 * backing_dev_info. 176 */ 177 if (sb->s_bdev) { 178 struct backing_dev_info *bdi; 179 180 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; 181 mapping->backing_dev_info = bdi; 182 } 183 inode->i_private = NULL; 184 inode->i_mapping = mapping; 185 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */ 186 #ifdef CONFIG_FS_POSIX_ACL 187 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; 188 #endif 189 190 #ifdef CONFIG_FSNOTIFY 191 inode->i_fsnotify_mask = 0; 192 #endif 193 194 this_cpu_inc(nr_inodes); 195 196 return 0; 197 out: 198 return -ENOMEM; 199 } 200 EXPORT_SYMBOL(inode_init_always); 201 202 static struct inode *alloc_inode(struct super_block *sb) 203 { 204 struct inode *inode; 205 206 if (sb->s_op->alloc_inode) 207 inode = sb->s_op->alloc_inode(sb); 208 else 209 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); 210 211 if (!inode) 212 return NULL; 213 214 if (unlikely(inode_init_always(sb, inode))) { 215 if (inode->i_sb->s_op->destroy_inode) 216 inode->i_sb->s_op->destroy_inode(inode); 217 else 218 kmem_cache_free(inode_cachep, inode); 219 return NULL; 220 } 221 222 return inode; 223 } 224 225 void free_inode_nonrcu(struct inode *inode) 226 { 227 kmem_cache_free(inode_cachep, inode); 228 } 229 EXPORT_SYMBOL(free_inode_nonrcu); 230 231 void __destroy_inode(struct inode *inode) 232 { 233 BUG_ON(inode_has_buffers(inode)); 234 security_inode_free(inode); 235 fsnotify_inode_delete(inode); 236 if (!inode->i_nlink) { 237 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0); 238 atomic_long_dec(&inode->i_sb->s_remove_count); 239 } 240 241 #ifdef CONFIG_FS_POSIX_ACL 242 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) 243 posix_acl_release(inode->i_acl); 244 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) 245 posix_acl_release(inode->i_default_acl); 246 #endif 247 this_cpu_dec(nr_inodes); 248 } 249 EXPORT_SYMBOL(__destroy_inode); 250 251 static void i_callback(struct rcu_head *head) 252 { 253 struct inode *inode = container_of(head, struct inode, i_rcu); 254 kmem_cache_free(inode_cachep, inode); 255 } 256 257 static void destroy_inode(struct inode *inode) 258 { 259 BUG_ON(!list_empty(&inode->i_lru)); 260 __destroy_inode(inode); 261 if (inode->i_sb->s_op->destroy_inode) 262 inode->i_sb->s_op->destroy_inode(inode); 263 else 264 call_rcu(&inode->i_rcu, i_callback); 265 } 266 267 /** 268 * drop_nlink - directly drop an inode's link count 269 * @inode: inode 270 * 271 * This is a low-level filesystem helper to replace any 272 * direct filesystem manipulation of i_nlink. In cases 273 * where we are attempting to track writes to the 274 * filesystem, a decrement to zero means an imminent 275 * write when the file is truncated and actually unlinked 276 * on the filesystem. 277 */ 278 void drop_nlink(struct inode *inode) 279 { 280 WARN_ON(inode->i_nlink == 0); 281 inode->__i_nlink--; 282 if (!inode->i_nlink) 283 atomic_long_inc(&inode->i_sb->s_remove_count); 284 } 285 EXPORT_SYMBOL(drop_nlink); 286 287 /** 288 * clear_nlink - directly zero an inode's link count 289 * @inode: inode 290 * 291 * This is a low-level filesystem helper to replace any 292 * direct filesystem manipulation of i_nlink. See 293 * drop_nlink() for why we care about i_nlink hitting zero. 294 */ 295 void clear_nlink(struct inode *inode) 296 { 297 if (inode->i_nlink) { 298 inode->__i_nlink = 0; 299 atomic_long_inc(&inode->i_sb->s_remove_count); 300 } 301 } 302 EXPORT_SYMBOL(clear_nlink); 303 304 /** 305 * set_nlink - directly set an inode's link count 306 * @inode: inode 307 * @nlink: new nlink (should be non-zero) 308 * 309 * This is a low-level filesystem helper to replace any 310 * direct filesystem manipulation of i_nlink. 311 */ 312 void set_nlink(struct inode *inode, unsigned int nlink) 313 { 314 if (!nlink) { 315 clear_nlink(inode); 316 } else { 317 /* Yes, some filesystems do change nlink from zero to one */ 318 if (inode->i_nlink == 0) 319 atomic_long_dec(&inode->i_sb->s_remove_count); 320 321 inode->__i_nlink = nlink; 322 } 323 } 324 EXPORT_SYMBOL(set_nlink); 325 326 /** 327 * inc_nlink - directly increment an inode's link count 328 * @inode: inode 329 * 330 * This is a low-level filesystem helper to replace any 331 * direct filesystem manipulation of i_nlink. Currently, 332 * it is only here for parity with dec_nlink(). 333 */ 334 void inc_nlink(struct inode *inode) 335 { 336 if (WARN_ON(inode->i_nlink == 0)) 337 atomic_long_dec(&inode->i_sb->s_remove_count); 338 339 inode->__i_nlink++; 340 } 341 EXPORT_SYMBOL(inc_nlink); 342 343 void address_space_init_once(struct address_space *mapping) 344 { 345 memset(mapping, 0, sizeof(*mapping)); 346 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC); 347 spin_lock_init(&mapping->tree_lock); 348 mutex_init(&mapping->i_mmap_mutex); 349 INIT_LIST_HEAD(&mapping->private_list); 350 spin_lock_init(&mapping->private_lock); 351 mapping->i_mmap = RB_ROOT; 352 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear); 353 } 354 EXPORT_SYMBOL(address_space_init_once); 355 356 /* 357 * These are initializations that only need to be done 358 * once, because the fields are idempotent across use 359 * of the inode, so let the slab aware of that. 360 */ 361 void inode_init_once(struct inode *inode) 362 { 363 memset(inode, 0, sizeof(*inode)); 364 INIT_HLIST_NODE(&inode->i_hash); 365 INIT_LIST_HEAD(&inode->i_devices); 366 INIT_LIST_HEAD(&inode->i_wb_list); 367 INIT_LIST_HEAD(&inode->i_lru); 368 address_space_init_once(&inode->i_data); 369 i_size_ordered_init(inode); 370 #ifdef CONFIG_FSNOTIFY 371 INIT_HLIST_HEAD(&inode->i_fsnotify_marks); 372 #endif 373 } 374 EXPORT_SYMBOL(inode_init_once); 375 376 static void init_once(void *foo) 377 { 378 struct inode *inode = (struct inode *) foo; 379 380 inode_init_once(inode); 381 } 382 383 /* 384 * inode->i_lock must be held 385 */ 386 void __iget(struct inode *inode) 387 { 388 atomic_inc(&inode->i_count); 389 } 390 391 /* 392 * get additional reference to inode; caller must already hold one. 393 */ 394 void ihold(struct inode *inode) 395 { 396 WARN_ON(atomic_inc_return(&inode->i_count) < 2); 397 } 398 EXPORT_SYMBOL(ihold); 399 400 static void inode_lru_list_add(struct inode *inode) 401 { 402 spin_lock(&inode->i_sb->s_inode_lru_lock); 403 if (list_empty(&inode->i_lru)) { 404 list_add(&inode->i_lru, &inode->i_sb->s_inode_lru); 405 inode->i_sb->s_nr_inodes_unused++; 406 this_cpu_inc(nr_unused); 407 } 408 spin_unlock(&inode->i_sb->s_inode_lru_lock); 409 } 410 411 /* 412 * Add inode to LRU if needed (inode is unused and clean). 413 * 414 * Needs inode->i_lock held. 415 */ 416 void inode_add_lru(struct inode *inode) 417 { 418 if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) && 419 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE) 420 inode_lru_list_add(inode); 421 } 422 423 424 static void inode_lru_list_del(struct inode *inode) 425 { 426 spin_lock(&inode->i_sb->s_inode_lru_lock); 427 if (!list_empty(&inode->i_lru)) { 428 list_del_init(&inode->i_lru); 429 inode->i_sb->s_nr_inodes_unused--; 430 this_cpu_dec(nr_unused); 431 } 432 spin_unlock(&inode->i_sb->s_inode_lru_lock); 433 } 434 435 /** 436 * inode_sb_list_add - add inode to the superblock list of inodes 437 * @inode: inode to add 438 */ 439 void inode_sb_list_add(struct inode *inode) 440 { 441 spin_lock(&inode_sb_list_lock); 442 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); 443 spin_unlock(&inode_sb_list_lock); 444 } 445 EXPORT_SYMBOL_GPL(inode_sb_list_add); 446 447 static inline void inode_sb_list_del(struct inode *inode) 448 { 449 if (!list_empty(&inode->i_sb_list)) { 450 spin_lock(&inode_sb_list_lock); 451 list_del_init(&inode->i_sb_list); 452 spin_unlock(&inode_sb_list_lock); 453 } 454 } 455 456 static unsigned long hash(struct super_block *sb, unsigned long hashval) 457 { 458 unsigned long tmp; 459 460 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / 461 L1_CACHE_BYTES; 462 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift); 463 return tmp & i_hash_mask; 464 } 465 466 /** 467 * __insert_inode_hash - hash an inode 468 * @inode: unhashed inode 469 * @hashval: unsigned long value used to locate this object in the 470 * inode_hashtable. 471 * 472 * Add an inode to the inode hash for this superblock. 473 */ 474 void __insert_inode_hash(struct inode *inode, unsigned long hashval) 475 { 476 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); 477 478 spin_lock(&inode_hash_lock); 479 spin_lock(&inode->i_lock); 480 hlist_add_head(&inode->i_hash, b); 481 spin_unlock(&inode->i_lock); 482 spin_unlock(&inode_hash_lock); 483 } 484 EXPORT_SYMBOL(__insert_inode_hash); 485 486 /** 487 * __remove_inode_hash - remove an inode from the hash 488 * @inode: inode to unhash 489 * 490 * Remove an inode from the superblock. 491 */ 492 void __remove_inode_hash(struct inode *inode) 493 { 494 spin_lock(&inode_hash_lock); 495 spin_lock(&inode->i_lock); 496 hlist_del_init(&inode->i_hash); 497 spin_unlock(&inode->i_lock); 498 spin_unlock(&inode_hash_lock); 499 } 500 EXPORT_SYMBOL(__remove_inode_hash); 501 502 void clear_inode(struct inode *inode) 503 { 504 might_sleep(); 505 /* 506 * We have to cycle tree_lock here because reclaim can be still in the 507 * process of removing the last page (in __delete_from_page_cache()) 508 * and we must not free mapping under it. 509 */ 510 spin_lock_irq(&inode->i_data.tree_lock); 511 BUG_ON(inode->i_data.nrpages); 512 spin_unlock_irq(&inode->i_data.tree_lock); 513 BUG_ON(!list_empty(&inode->i_data.private_list)); 514 BUG_ON(!(inode->i_state & I_FREEING)); 515 BUG_ON(inode->i_state & I_CLEAR); 516 /* don't need i_lock here, no concurrent mods to i_state */ 517 inode->i_state = I_FREEING | I_CLEAR; 518 } 519 EXPORT_SYMBOL(clear_inode); 520 521 /* 522 * Free the inode passed in, removing it from the lists it is still connected 523 * to. We remove any pages still attached to the inode and wait for any IO that 524 * is still in progress before finally destroying the inode. 525 * 526 * An inode must already be marked I_FREEING so that we avoid the inode being 527 * moved back onto lists if we race with other code that manipulates the lists 528 * (e.g. writeback_single_inode). The caller is responsible for setting this. 529 * 530 * An inode must already be removed from the LRU list before being evicted from 531 * the cache. This should occur atomically with setting the I_FREEING state 532 * flag, so no inodes here should ever be on the LRU when being evicted. 533 */ 534 static void evict(struct inode *inode) 535 { 536 const struct super_operations *op = inode->i_sb->s_op; 537 538 BUG_ON(!(inode->i_state & I_FREEING)); 539 BUG_ON(!list_empty(&inode->i_lru)); 540 541 if (!list_empty(&inode->i_wb_list)) 542 inode_wb_list_del(inode); 543 544 inode_sb_list_del(inode); 545 546 /* 547 * Wait for flusher thread to be done with the inode so that filesystem 548 * does not start destroying it while writeback is still running. Since 549 * the inode has I_FREEING set, flusher thread won't start new work on 550 * the inode. We just have to wait for running writeback to finish. 551 */ 552 inode_wait_for_writeback(inode); 553 554 if (op->evict_inode) { 555 op->evict_inode(inode); 556 } else { 557 if (inode->i_data.nrpages) 558 truncate_inode_pages(&inode->i_data, 0); 559 clear_inode(inode); 560 } 561 if (S_ISBLK(inode->i_mode) && inode->i_bdev) 562 bd_forget(inode); 563 if (S_ISCHR(inode->i_mode) && inode->i_cdev) 564 cd_forget(inode); 565 566 remove_inode_hash(inode); 567 568 spin_lock(&inode->i_lock); 569 wake_up_bit(&inode->i_state, __I_NEW); 570 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); 571 spin_unlock(&inode->i_lock); 572 573 destroy_inode(inode); 574 } 575 576 /* 577 * dispose_list - dispose of the contents of a local list 578 * @head: the head of the list to free 579 * 580 * Dispose-list gets a local list with local inodes in it, so it doesn't 581 * need to worry about list corruption and SMP locks. 582 */ 583 static void dispose_list(struct list_head *head) 584 { 585 while (!list_empty(head)) { 586 struct inode *inode; 587 588 inode = list_first_entry(head, struct inode, i_lru); 589 list_del_init(&inode->i_lru); 590 591 evict(inode); 592 } 593 } 594 595 /** 596 * evict_inodes - evict all evictable inodes for a superblock 597 * @sb: superblock to operate on 598 * 599 * Make sure that no inodes with zero refcount are retained. This is 600 * called by superblock shutdown after having MS_ACTIVE flag removed, 601 * so any inode reaching zero refcount during or after that call will 602 * be immediately evicted. 603 */ 604 void evict_inodes(struct super_block *sb) 605 { 606 struct inode *inode, *next; 607 LIST_HEAD(dispose); 608 609 spin_lock(&inode_sb_list_lock); 610 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { 611 if (atomic_read(&inode->i_count)) 612 continue; 613 614 spin_lock(&inode->i_lock); 615 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 616 spin_unlock(&inode->i_lock); 617 continue; 618 } 619 620 inode->i_state |= I_FREEING; 621 inode_lru_list_del(inode); 622 spin_unlock(&inode->i_lock); 623 list_add(&inode->i_lru, &dispose); 624 } 625 spin_unlock(&inode_sb_list_lock); 626 627 dispose_list(&dispose); 628 } 629 630 /** 631 * invalidate_inodes - attempt to free all inodes on a superblock 632 * @sb: superblock to operate on 633 * @kill_dirty: flag to guide handling of dirty inodes 634 * 635 * Attempts to free all inodes for a given superblock. If there were any 636 * busy inodes return a non-zero value, else zero. 637 * If @kill_dirty is set, discard dirty inodes too, otherwise treat 638 * them as busy. 639 */ 640 int invalidate_inodes(struct super_block *sb, bool kill_dirty) 641 { 642 int busy = 0; 643 struct inode *inode, *next; 644 LIST_HEAD(dispose); 645 646 spin_lock(&inode_sb_list_lock); 647 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { 648 spin_lock(&inode->i_lock); 649 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 650 spin_unlock(&inode->i_lock); 651 continue; 652 } 653 if (inode->i_state & I_DIRTY && !kill_dirty) { 654 spin_unlock(&inode->i_lock); 655 busy = 1; 656 continue; 657 } 658 if (atomic_read(&inode->i_count)) { 659 spin_unlock(&inode->i_lock); 660 busy = 1; 661 continue; 662 } 663 664 inode->i_state |= I_FREEING; 665 inode_lru_list_del(inode); 666 spin_unlock(&inode->i_lock); 667 list_add(&inode->i_lru, &dispose); 668 } 669 spin_unlock(&inode_sb_list_lock); 670 671 dispose_list(&dispose); 672 673 return busy; 674 } 675 676 static int can_unuse(struct inode *inode) 677 { 678 if (inode->i_state & ~I_REFERENCED) 679 return 0; 680 if (inode_has_buffers(inode)) 681 return 0; 682 if (atomic_read(&inode->i_count)) 683 return 0; 684 if (inode->i_data.nrpages) 685 return 0; 686 return 1; 687 } 688 689 /* 690 * Walk the superblock inode LRU for freeable inodes and attempt to free them. 691 * This is called from the superblock shrinker function with a number of inodes 692 * to trim from the LRU. Inodes to be freed are moved to a temporary list and 693 * then are freed outside inode_lock by dispose_list(). 694 * 695 * Any inodes which are pinned purely because of attached pagecache have their 696 * pagecache removed. If the inode has metadata buffers attached to 697 * mapping->private_list then try to remove them. 698 * 699 * If the inode has the I_REFERENCED flag set, then it means that it has been 700 * used recently - the flag is set in iput_final(). When we encounter such an 701 * inode, clear the flag and move it to the back of the LRU so it gets another 702 * pass through the LRU before it gets reclaimed. This is necessary because of 703 * the fact we are doing lazy LRU updates to minimise lock contention so the 704 * LRU does not have strict ordering. Hence we don't want to reclaim inodes 705 * with this flag set because they are the inodes that are out of order. 706 */ 707 void prune_icache_sb(struct super_block *sb, int nr_to_scan) 708 { 709 LIST_HEAD(freeable); 710 int nr_scanned; 711 unsigned long reap = 0; 712 713 spin_lock(&sb->s_inode_lru_lock); 714 for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) { 715 struct inode *inode; 716 717 if (list_empty(&sb->s_inode_lru)) 718 break; 719 720 inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru); 721 722 /* 723 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here, 724 * so use a trylock. If we fail to get the lock, just move the 725 * inode to the back of the list so we don't spin on it. 726 */ 727 if (!spin_trylock(&inode->i_lock)) { 728 list_move_tail(&inode->i_lru, &sb->s_inode_lru); 729 continue; 730 } 731 732 /* 733 * Referenced or dirty inodes are still in use. Give them 734 * another pass through the LRU as we canot reclaim them now. 735 */ 736 if (atomic_read(&inode->i_count) || 737 (inode->i_state & ~I_REFERENCED)) { 738 list_del_init(&inode->i_lru); 739 spin_unlock(&inode->i_lock); 740 sb->s_nr_inodes_unused--; 741 this_cpu_dec(nr_unused); 742 continue; 743 } 744 745 /* recently referenced inodes get one more pass */ 746 if (inode->i_state & I_REFERENCED) { 747 inode->i_state &= ~I_REFERENCED; 748 list_move(&inode->i_lru, &sb->s_inode_lru); 749 spin_unlock(&inode->i_lock); 750 continue; 751 } 752 if (inode_has_buffers(inode) || inode->i_data.nrpages) { 753 __iget(inode); 754 spin_unlock(&inode->i_lock); 755 spin_unlock(&sb->s_inode_lru_lock); 756 if (remove_inode_buffers(inode)) 757 reap += invalidate_mapping_pages(&inode->i_data, 758 0, -1); 759 iput(inode); 760 spin_lock(&sb->s_inode_lru_lock); 761 762 if (inode != list_entry(sb->s_inode_lru.next, 763 struct inode, i_lru)) 764 continue; /* wrong inode or list_empty */ 765 /* avoid lock inversions with trylock */ 766 if (!spin_trylock(&inode->i_lock)) 767 continue; 768 if (!can_unuse(inode)) { 769 spin_unlock(&inode->i_lock); 770 continue; 771 } 772 } 773 WARN_ON(inode->i_state & I_NEW); 774 inode->i_state |= I_FREEING; 775 spin_unlock(&inode->i_lock); 776 777 list_move(&inode->i_lru, &freeable); 778 sb->s_nr_inodes_unused--; 779 this_cpu_dec(nr_unused); 780 } 781 if (current_is_kswapd()) 782 __count_vm_events(KSWAPD_INODESTEAL, reap); 783 else 784 __count_vm_events(PGINODESTEAL, reap); 785 spin_unlock(&sb->s_inode_lru_lock); 786 if (current->reclaim_state) 787 current->reclaim_state->reclaimed_slab += reap; 788 789 dispose_list(&freeable); 790 } 791 792 static void __wait_on_freeing_inode(struct inode *inode); 793 /* 794 * Called with the inode lock held. 795 */ 796 static struct inode *find_inode(struct super_block *sb, 797 struct hlist_head *head, 798 int (*test)(struct inode *, void *), 799 void *data) 800 { 801 struct inode *inode = NULL; 802 803 repeat: 804 hlist_for_each_entry(inode, head, i_hash) { 805 spin_lock(&inode->i_lock); 806 if (inode->i_sb != sb) { 807 spin_unlock(&inode->i_lock); 808 continue; 809 } 810 if (!test(inode, data)) { 811 spin_unlock(&inode->i_lock); 812 continue; 813 } 814 if (inode->i_state & (I_FREEING|I_WILL_FREE)) { 815 __wait_on_freeing_inode(inode); 816 goto repeat; 817 } 818 __iget(inode); 819 spin_unlock(&inode->i_lock); 820 return inode; 821 } 822 return NULL; 823 } 824 825 /* 826 * find_inode_fast is the fast path version of find_inode, see the comment at 827 * iget_locked for details. 828 */ 829 static struct inode *find_inode_fast(struct super_block *sb, 830 struct hlist_head *head, unsigned long ino) 831 { 832 struct inode *inode = NULL; 833 834 repeat: 835 hlist_for_each_entry(inode, head, i_hash) { 836 spin_lock(&inode->i_lock); 837 if (inode->i_ino != ino) { 838 spin_unlock(&inode->i_lock); 839 continue; 840 } 841 if (inode->i_sb != sb) { 842 spin_unlock(&inode->i_lock); 843 continue; 844 } 845 if (inode->i_state & (I_FREEING|I_WILL_FREE)) { 846 __wait_on_freeing_inode(inode); 847 goto repeat; 848 } 849 __iget(inode); 850 spin_unlock(&inode->i_lock); 851 return inode; 852 } 853 return NULL; 854 } 855 856 /* 857 * Each cpu owns a range of LAST_INO_BATCH numbers. 858 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, 859 * to renew the exhausted range. 860 * 861 * This does not significantly increase overflow rate because every CPU can 862 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is 863 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the 864 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase 865 * overflow rate by 2x, which does not seem too significant. 866 * 867 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW 868 * error if st_ino won't fit in target struct field. Use 32bit counter 869 * here to attempt to avoid that. 870 */ 871 #define LAST_INO_BATCH 1024 872 static DEFINE_PER_CPU(unsigned int, last_ino); 873 874 unsigned int get_next_ino(void) 875 { 876 unsigned int *p = &get_cpu_var(last_ino); 877 unsigned int res = *p; 878 879 #ifdef CONFIG_SMP 880 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { 881 static atomic_t shared_last_ino; 882 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); 883 884 res = next - LAST_INO_BATCH; 885 } 886 #endif 887 888 *p = ++res; 889 put_cpu_var(last_ino); 890 return res; 891 } 892 EXPORT_SYMBOL(get_next_ino); 893 894 /** 895 * new_inode_pseudo - obtain an inode 896 * @sb: superblock 897 * 898 * Allocates a new inode for given superblock. 899 * Inode wont be chained in superblock s_inodes list 900 * This means : 901 * - fs can't be unmount 902 * - quotas, fsnotify, writeback can't work 903 */ 904 struct inode *new_inode_pseudo(struct super_block *sb) 905 { 906 struct inode *inode = alloc_inode(sb); 907 908 if (inode) { 909 spin_lock(&inode->i_lock); 910 inode->i_state = 0; 911 spin_unlock(&inode->i_lock); 912 INIT_LIST_HEAD(&inode->i_sb_list); 913 } 914 return inode; 915 } 916 917 /** 918 * new_inode - obtain an inode 919 * @sb: superblock 920 * 921 * Allocates a new inode for given superblock. The default gfp_mask 922 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. 923 * If HIGHMEM pages are unsuitable or it is known that pages allocated 924 * for the page cache are not reclaimable or migratable, 925 * mapping_set_gfp_mask() must be called with suitable flags on the 926 * newly created inode's mapping 927 * 928 */ 929 struct inode *new_inode(struct super_block *sb) 930 { 931 struct inode *inode; 932 933 spin_lock_prefetch(&inode_sb_list_lock); 934 935 inode = new_inode_pseudo(sb); 936 if (inode) 937 inode_sb_list_add(inode); 938 return inode; 939 } 940 EXPORT_SYMBOL(new_inode); 941 942 #ifdef CONFIG_DEBUG_LOCK_ALLOC 943 void lockdep_annotate_inode_mutex_key(struct inode *inode) 944 { 945 if (S_ISDIR(inode->i_mode)) { 946 struct file_system_type *type = inode->i_sb->s_type; 947 948 /* Set new key only if filesystem hasn't already changed it */ 949 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) { 950 /* 951 * ensure nobody is actually holding i_mutex 952 */ 953 mutex_destroy(&inode->i_mutex); 954 mutex_init(&inode->i_mutex); 955 lockdep_set_class(&inode->i_mutex, 956 &type->i_mutex_dir_key); 957 } 958 } 959 } 960 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key); 961 #endif 962 963 /** 964 * unlock_new_inode - clear the I_NEW state and wake up any waiters 965 * @inode: new inode to unlock 966 * 967 * Called when the inode is fully initialised to clear the new state of the 968 * inode and wake up anyone waiting for the inode to finish initialisation. 969 */ 970 void unlock_new_inode(struct inode *inode) 971 { 972 lockdep_annotate_inode_mutex_key(inode); 973 spin_lock(&inode->i_lock); 974 WARN_ON(!(inode->i_state & I_NEW)); 975 inode->i_state &= ~I_NEW; 976 smp_mb(); 977 wake_up_bit(&inode->i_state, __I_NEW); 978 spin_unlock(&inode->i_lock); 979 } 980 EXPORT_SYMBOL(unlock_new_inode); 981 982 /** 983 * iget5_locked - obtain an inode from a mounted file system 984 * @sb: super block of file system 985 * @hashval: hash value (usually inode number) to get 986 * @test: callback used for comparisons between inodes 987 * @set: callback used to initialize a new struct inode 988 * @data: opaque data pointer to pass to @test and @set 989 * 990 * Search for the inode specified by @hashval and @data in the inode cache, 991 * and if present it is return it with an increased reference count. This is 992 * a generalized version of iget_locked() for file systems where the inode 993 * number is not sufficient for unique identification of an inode. 994 * 995 * If the inode is not in cache, allocate a new inode and return it locked, 996 * hashed, and with the I_NEW flag set. The file system gets to fill it in 997 * before unlocking it via unlock_new_inode(). 998 * 999 * Note both @test and @set are called with the inode_hash_lock held, so can't 1000 * sleep. 1001 */ 1002 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, 1003 int (*test)(struct inode *, void *), 1004 int (*set)(struct inode *, void *), void *data) 1005 { 1006 struct hlist_head *head = inode_hashtable + hash(sb, hashval); 1007 struct inode *inode; 1008 1009 spin_lock(&inode_hash_lock); 1010 inode = find_inode(sb, head, test, data); 1011 spin_unlock(&inode_hash_lock); 1012 1013 if (inode) { 1014 wait_on_inode(inode); 1015 return inode; 1016 } 1017 1018 inode = alloc_inode(sb); 1019 if (inode) { 1020 struct inode *old; 1021 1022 spin_lock(&inode_hash_lock); 1023 /* We released the lock, so.. */ 1024 old = find_inode(sb, head, test, data); 1025 if (!old) { 1026 if (set(inode, data)) 1027 goto set_failed; 1028 1029 spin_lock(&inode->i_lock); 1030 inode->i_state = I_NEW; 1031 hlist_add_head(&inode->i_hash, head); 1032 spin_unlock(&inode->i_lock); 1033 inode_sb_list_add(inode); 1034 spin_unlock(&inode_hash_lock); 1035 1036 /* Return the locked inode with I_NEW set, the 1037 * caller is responsible for filling in the contents 1038 */ 1039 return inode; 1040 } 1041 1042 /* 1043 * Uhhuh, somebody else created the same inode under 1044 * us. Use the old inode instead of the one we just 1045 * allocated. 1046 */ 1047 spin_unlock(&inode_hash_lock); 1048 destroy_inode(inode); 1049 inode = old; 1050 wait_on_inode(inode); 1051 } 1052 return inode; 1053 1054 set_failed: 1055 spin_unlock(&inode_hash_lock); 1056 destroy_inode(inode); 1057 return NULL; 1058 } 1059 EXPORT_SYMBOL(iget5_locked); 1060 1061 /** 1062 * iget_locked - obtain an inode from a mounted file system 1063 * @sb: super block of file system 1064 * @ino: inode number to get 1065 * 1066 * Search for the inode specified by @ino in the inode cache and if present 1067 * return it with an increased reference count. This is for file systems 1068 * where the inode number is sufficient for unique identification of an inode. 1069 * 1070 * If the inode is not in cache, allocate a new inode and return it locked, 1071 * hashed, and with the I_NEW flag set. The file system gets to fill it in 1072 * before unlocking it via unlock_new_inode(). 1073 */ 1074 struct inode *iget_locked(struct super_block *sb, unsigned long ino) 1075 { 1076 struct hlist_head *head = inode_hashtable + hash(sb, ino); 1077 struct inode *inode; 1078 1079 spin_lock(&inode_hash_lock); 1080 inode = find_inode_fast(sb, head, ino); 1081 spin_unlock(&inode_hash_lock); 1082 if (inode) { 1083 wait_on_inode(inode); 1084 return inode; 1085 } 1086 1087 inode = alloc_inode(sb); 1088 if (inode) { 1089 struct inode *old; 1090 1091 spin_lock(&inode_hash_lock); 1092 /* We released the lock, so.. */ 1093 old = find_inode_fast(sb, head, ino); 1094 if (!old) { 1095 inode->i_ino = ino; 1096 spin_lock(&inode->i_lock); 1097 inode->i_state = I_NEW; 1098 hlist_add_head(&inode->i_hash, head); 1099 spin_unlock(&inode->i_lock); 1100 inode_sb_list_add(inode); 1101 spin_unlock(&inode_hash_lock); 1102 1103 /* Return the locked inode with I_NEW set, the 1104 * caller is responsible for filling in the contents 1105 */ 1106 return inode; 1107 } 1108 1109 /* 1110 * Uhhuh, somebody else created the same inode under 1111 * us. Use the old inode instead of the one we just 1112 * allocated. 1113 */ 1114 spin_unlock(&inode_hash_lock); 1115 destroy_inode(inode); 1116 inode = old; 1117 wait_on_inode(inode); 1118 } 1119 return inode; 1120 } 1121 EXPORT_SYMBOL(iget_locked); 1122 1123 /* 1124 * search the inode cache for a matching inode number. 1125 * If we find one, then the inode number we are trying to 1126 * allocate is not unique and so we should not use it. 1127 * 1128 * Returns 1 if the inode number is unique, 0 if it is not. 1129 */ 1130 static int test_inode_iunique(struct super_block *sb, unsigned long ino) 1131 { 1132 struct hlist_head *b = inode_hashtable + hash(sb, ino); 1133 struct inode *inode; 1134 1135 spin_lock(&inode_hash_lock); 1136 hlist_for_each_entry(inode, b, i_hash) { 1137 if (inode->i_ino == ino && inode->i_sb == sb) { 1138 spin_unlock(&inode_hash_lock); 1139 return 0; 1140 } 1141 } 1142 spin_unlock(&inode_hash_lock); 1143 1144 return 1; 1145 } 1146 1147 /** 1148 * iunique - get a unique inode number 1149 * @sb: superblock 1150 * @max_reserved: highest reserved inode number 1151 * 1152 * Obtain an inode number that is unique on the system for a given 1153 * superblock. This is used by file systems that have no natural 1154 * permanent inode numbering system. An inode number is returned that 1155 * is higher than the reserved limit but unique. 1156 * 1157 * BUGS: 1158 * With a large number of inodes live on the file system this function 1159 * currently becomes quite slow. 1160 */ 1161 ino_t iunique(struct super_block *sb, ino_t max_reserved) 1162 { 1163 /* 1164 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW 1165 * error if st_ino won't fit in target struct field. Use 32bit counter 1166 * here to attempt to avoid that. 1167 */ 1168 static DEFINE_SPINLOCK(iunique_lock); 1169 static unsigned int counter; 1170 ino_t res; 1171 1172 spin_lock(&iunique_lock); 1173 do { 1174 if (counter <= max_reserved) 1175 counter = max_reserved + 1; 1176 res = counter++; 1177 } while (!test_inode_iunique(sb, res)); 1178 spin_unlock(&iunique_lock); 1179 1180 return res; 1181 } 1182 EXPORT_SYMBOL(iunique); 1183 1184 struct inode *igrab(struct inode *inode) 1185 { 1186 spin_lock(&inode->i_lock); 1187 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { 1188 __iget(inode); 1189 spin_unlock(&inode->i_lock); 1190 } else { 1191 spin_unlock(&inode->i_lock); 1192 /* 1193 * Handle the case where s_op->clear_inode is not been 1194 * called yet, and somebody is calling igrab 1195 * while the inode is getting freed. 1196 */ 1197 inode = NULL; 1198 } 1199 return inode; 1200 } 1201 EXPORT_SYMBOL(igrab); 1202 1203 /** 1204 * ilookup5_nowait - search for an inode in the inode cache 1205 * @sb: super block of file system to search 1206 * @hashval: hash value (usually inode number) to search for 1207 * @test: callback used for comparisons between inodes 1208 * @data: opaque data pointer to pass to @test 1209 * 1210 * Search for the inode specified by @hashval and @data in the inode cache. 1211 * If the inode is in the cache, the inode is returned with an incremented 1212 * reference count. 1213 * 1214 * Note: I_NEW is not waited upon so you have to be very careful what you do 1215 * with the returned inode. You probably should be using ilookup5() instead. 1216 * 1217 * Note2: @test is called with the inode_hash_lock held, so can't sleep. 1218 */ 1219 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, 1220 int (*test)(struct inode *, void *), void *data) 1221 { 1222 struct hlist_head *head = inode_hashtable + hash(sb, hashval); 1223 struct inode *inode; 1224 1225 spin_lock(&inode_hash_lock); 1226 inode = find_inode(sb, head, test, data); 1227 spin_unlock(&inode_hash_lock); 1228 1229 return inode; 1230 } 1231 EXPORT_SYMBOL(ilookup5_nowait); 1232 1233 /** 1234 * ilookup5 - search for an inode in the inode cache 1235 * @sb: super block of file system to search 1236 * @hashval: hash value (usually inode number) to search for 1237 * @test: callback used for comparisons between inodes 1238 * @data: opaque data pointer to pass to @test 1239 * 1240 * Search for the inode specified by @hashval and @data in the inode cache, 1241 * and if the inode is in the cache, return the inode with an incremented 1242 * reference count. Waits on I_NEW before returning the inode. 1243 * returned with an incremented reference count. 1244 * 1245 * This is a generalized version of ilookup() for file systems where the 1246 * inode number is not sufficient for unique identification of an inode. 1247 * 1248 * Note: @test is called with the inode_hash_lock held, so can't sleep. 1249 */ 1250 struct inode *ilookup5(struct super_block *sb, unsigned long hashval, 1251 int (*test)(struct inode *, void *), void *data) 1252 { 1253 struct inode *inode = ilookup5_nowait(sb, hashval, test, data); 1254 1255 if (inode) 1256 wait_on_inode(inode); 1257 return inode; 1258 } 1259 EXPORT_SYMBOL(ilookup5); 1260 1261 /** 1262 * ilookup - search for an inode in the inode cache 1263 * @sb: super block of file system to search 1264 * @ino: inode number to search for 1265 * 1266 * Search for the inode @ino in the inode cache, and if the inode is in the 1267 * cache, the inode is returned with an incremented reference count. 1268 */ 1269 struct inode *ilookup(struct super_block *sb, unsigned long ino) 1270 { 1271 struct hlist_head *head = inode_hashtable + hash(sb, ino); 1272 struct inode *inode; 1273 1274 spin_lock(&inode_hash_lock); 1275 inode = find_inode_fast(sb, head, ino); 1276 spin_unlock(&inode_hash_lock); 1277 1278 if (inode) 1279 wait_on_inode(inode); 1280 return inode; 1281 } 1282 EXPORT_SYMBOL(ilookup); 1283 1284 int insert_inode_locked(struct inode *inode) 1285 { 1286 struct super_block *sb = inode->i_sb; 1287 ino_t ino = inode->i_ino; 1288 struct hlist_head *head = inode_hashtable + hash(sb, ino); 1289 1290 while (1) { 1291 struct inode *old = NULL; 1292 spin_lock(&inode_hash_lock); 1293 hlist_for_each_entry(old, head, i_hash) { 1294 if (old->i_ino != ino) 1295 continue; 1296 if (old->i_sb != sb) 1297 continue; 1298 spin_lock(&old->i_lock); 1299 if (old->i_state & (I_FREEING|I_WILL_FREE)) { 1300 spin_unlock(&old->i_lock); 1301 continue; 1302 } 1303 break; 1304 } 1305 if (likely(!old)) { 1306 spin_lock(&inode->i_lock); 1307 inode->i_state |= I_NEW; 1308 hlist_add_head(&inode->i_hash, head); 1309 spin_unlock(&inode->i_lock); 1310 spin_unlock(&inode_hash_lock); 1311 return 0; 1312 } 1313 __iget(old); 1314 spin_unlock(&old->i_lock); 1315 spin_unlock(&inode_hash_lock); 1316 wait_on_inode(old); 1317 if (unlikely(!inode_unhashed(old))) { 1318 iput(old); 1319 return -EBUSY; 1320 } 1321 iput(old); 1322 } 1323 } 1324 EXPORT_SYMBOL(insert_inode_locked); 1325 1326 int insert_inode_locked4(struct inode *inode, unsigned long hashval, 1327 int (*test)(struct inode *, void *), void *data) 1328 { 1329 struct super_block *sb = inode->i_sb; 1330 struct hlist_head *head = inode_hashtable + hash(sb, hashval); 1331 1332 while (1) { 1333 struct inode *old = NULL; 1334 1335 spin_lock(&inode_hash_lock); 1336 hlist_for_each_entry(old, head, i_hash) { 1337 if (old->i_sb != sb) 1338 continue; 1339 if (!test(old, data)) 1340 continue; 1341 spin_lock(&old->i_lock); 1342 if (old->i_state & (I_FREEING|I_WILL_FREE)) { 1343 spin_unlock(&old->i_lock); 1344 continue; 1345 } 1346 break; 1347 } 1348 if (likely(!old)) { 1349 spin_lock(&inode->i_lock); 1350 inode->i_state |= I_NEW; 1351 hlist_add_head(&inode->i_hash, head); 1352 spin_unlock(&inode->i_lock); 1353 spin_unlock(&inode_hash_lock); 1354 return 0; 1355 } 1356 __iget(old); 1357 spin_unlock(&old->i_lock); 1358 spin_unlock(&inode_hash_lock); 1359 wait_on_inode(old); 1360 if (unlikely(!inode_unhashed(old))) { 1361 iput(old); 1362 return -EBUSY; 1363 } 1364 iput(old); 1365 } 1366 } 1367 EXPORT_SYMBOL(insert_inode_locked4); 1368 1369 1370 int generic_delete_inode(struct inode *inode) 1371 { 1372 return 1; 1373 } 1374 EXPORT_SYMBOL(generic_delete_inode); 1375 1376 /* 1377 * Called when we're dropping the last reference 1378 * to an inode. 1379 * 1380 * Call the FS "drop_inode()" function, defaulting to 1381 * the legacy UNIX filesystem behaviour. If it tells 1382 * us to evict inode, do so. Otherwise, retain inode 1383 * in cache if fs is alive, sync and evict if fs is 1384 * shutting down. 1385 */ 1386 static void iput_final(struct inode *inode) 1387 { 1388 struct super_block *sb = inode->i_sb; 1389 const struct super_operations *op = inode->i_sb->s_op; 1390 int drop; 1391 1392 WARN_ON(inode->i_state & I_NEW); 1393 1394 if (op->drop_inode) 1395 drop = op->drop_inode(inode); 1396 else 1397 drop = generic_drop_inode(inode); 1398 1399 if (!drop && (sb->s_flags & MS_ACTIVE)) { 1400 inode->i_state |= I_REFERENCED; 1401 inode_add_lru(inode); 1402 spin_unlock(&inode->i_lock); 1403 return; 1404 } 1405 1406 if (!drop) { 1407 inode->i_state |= I_WILL_FREE; 1408 spin_unlock(&inode->i_lock); 1409 write_inode_now(inode, 1); 1410 spin_lock(&inode->i_lock); 1411 WARN_ON(inode->i_state & I_NEW); 1412 inode->i_state &= ~I_WILL_FREE; 1413 } 1414 1415 inode->i_state |= I_FREEING; 1416 if (!list_empty(&inode->i_lru)) 1417 inode_lru_list_del(inode); 1418 spin_unlock(&inode->i_lock); 1419 1420 evict(inode); 1421 } 1422 1423 /** 1424 * iput - put an inode 1425 * @inode: inode to put 1426 * 1427 * Puts an inode, dropping its usage count. If the inode use count hits 1428 * zero, the inode is then freed and may also be destroyed. 1429 * 1430 * Consequently, iput() can sleep. 1431 */ 1432 void iput(struct inode *inode) 1433 { 1434 if (inode) { 1435 BUG_ON(inode->i_state & I_CLEAR); 1436 1437 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) 1438 iput_final(inode); 1439 } 1440 } 1441 EXPORT_SYMBOL(iput); 1442 1443 /** 1444 * bmap - find a block number in a file 1445 * @inode: inode of file 1446 * @block: block to find 1447 * 1448 * Returns the block number on the device holding the inode that 1449 * is the disk block number for the block of the file requested. 1450 * That is, asked for block 4 of inode 1 the function will return the 1451 * disk block relative to the disk start that holds that block of the 1452 * file. 1453 */ 1454 sector_t bmap(struct inode *inode, sector_t block) 1455 { 1456 sector_t res = 0; 1457 if (inode->i_mapping->a_ops->bmap) 1458 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); 1459 return res; 1460 } 1461 EXPORT_SYMBOL(bmap); 1462 1463 /* 1464 * With relative atime, only update atime if the previous atime is 1465 * earlier than either the ctime or mtime or if at least a day has 1466 * passed since the last atime update. 1467 */ 1468 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, 1469 struct timespec now) 1470 { 1471 1472 if (!(mnt->mnt_flags & MNT_RELATIME)) 1473 return 1; 1474 /* 1475 * Is mtime younger than atime? If yes, update atime: 1476 */ 1477 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) 1478 return 1; 1479 /* 1480 * Is ctime younger than atime? If yes, update atime: 1481 */ 1482 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) 1483 return 1; 1484 1485 /* 1486 * Is the previous atime value older than a day? If yes, 1487 * update atime: 1488 */ 1489 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) 1490 return 1; 1491 /* 1492 * Good, we can skip the atime update: 1493 */ 1494 return 0; 1495 } 1496 1497 /* 1498 * This does the actual work of updating an inodes time or version. Must have 1499 * had called mnt_want_write() before calling this. 1500 */ 1501 static int update_time(struct inode *inode, struct timespec *time, int flags) 1502 { 1503 if (inode->i_op->update_time) 1504 return inode->i_op->update_time(inode, time, flags); 1505 1506 if (flags & S_ATIME) 1507 inode->i_atime = *time; 1508 if (flags & S_VERSION) 1509 inode_inc_iversion(inode); 1510 if (flags & S_CTIME) 1511 inode->i_ctime = *time; 1512 if (flags & S_MTIME) 1513 inode->i_mtime = *time; 1514 mark_inode_dirty_sync(inode); 1515 return 0; 1516 } 1517 1518 /** 1519 * touch_atime - update the access time 1520 * @path: the &struct path to update 1521 * 1522 * Update the accessed time on an inode and mark it for writeback. 1523 * This function automatically handles read only file systems and media, 1524 * as well as the "noatime" flag and inode specific "noatime" markers. 1525 */ 1526 void touch_atime(struct path *path) 1527 { 1528 struct vfsmount *mnt = path->mnt; 1529 struct inode *inode = path->dentry->d_inode; 1530 struct timespec now; 1531 1532 if (inode->i_flags & S_NOATIME) 1533 return; 1534 if (IS_NOATIME(inode)) 1535 return; 1536 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) 1537 return; 1538 1539 if (mnt->mnt_flags & MNT_NOATIME) 1540 return; 1541 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) 1542 return; 1543 1544 now = current_fs_time(inode->i_sb); 1545 1546 if (!relatime_need_update(mnt, inode, now)) 1547 return; 1548 1549 if (timespec_equal(&inode->i_atime, &now)) 1550 return; 1551 1552 if (!sb_start_write_trylock(inode->i_sb)) 1553 return; 1554 1555 if (__mnt_want_write(mnt)) 1556 goto skip_update; 1557 /* 1558 * File systems can error out when updating inodes if they need to 1559 * allocate new space to modify an inode (such is the case for 1560 * Btrfs), but since we touch atime while walking down the path we 1561 * really don't care if we failed to update the atime of the file, 1562 * so just ignore the return value. 1563 * We may also fail on filesystems that have the ability to make parts 1564 * of the fs read only, e.g. subvolumes in Btrfs. 1565 */ 1566 update_time(inode, &now, S_ATIME); 1567 __mnt_drop_write(mnt); 1568 skip_update: 1569 sb_end_write(inode->i_sb); 1570 } 1571 EXPORT_SYMBOL(touch_atime); 1572 1573 /* 1574 * The logic we want is 1575 * 1576 * if suid or (sgid and xgrp) 1577 * remove privs 1578 */ 1579 int should_remove_suid(struct dentry *dentry) 1580 { 1581 umode_t mode = dentry->d_inode->i_mode; 1582 int kill = 0; 1583 1584 /* suid always must be killed */ 1585 if (unlikely(mode & S_ISUID)) 1586 kill = ATTR_KILL_SUID; 1587 1588 /* 1589 * sgid without any exec bits is just a mandatory locking mark; leave 1590 * it alone. If some exec bits are set, it's a real sgid; kill it. 1591 */ 1592 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 1593 kill |= ATTR_KILL_SGID; 1594 1595 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) 1596 return kill; 1597 1598 return 0; 1599 } 1600 EXPORT_SYMBOL(should_remove_suid); 1601 1602 static int __remove_suid(struct dentry *dentry, int kill) 1603 { 1604 struct iattr newattrs; 1605 1606 newattrs.ia_valid = ATTR_FORCE | kill; 1607 return notify_change(dentry, &newattrs); 1608 } 1609 1610 int file_remove_suid(struct file *file) 1611 { 1612 struct dentry *dentry = file->f_path.dentry; 1613 struct inode *inode = dentry->d_inode; 1614 int killsuid; 1615 int killpriv; 1616 int error = 0; 1617 1618 /* Fast path for nothing security related */ 1619 if (IS_NOSEC(inode)) 1620 return 0; 1621 1622 killsuid = should_remove_suid(dentry); 1623 killpriv = security_inode_need_killpriv(dentry); 1624 1625 if (killpriv < 0) 1626 return killpriv; 1627 if (killpriv) 1628 error = security_inode_killpriv(dentry); 1629 if (!error && killsuid) 1630 error = __remove_suid(dentry, killsuid); 1631 if (!error && (inode->i_sb->s_flags & MS_NOSEC)) 1632 inode->i_flags |= S_NOSEC; 1633 1634 return error; 1635 } 1636 EXPORT_SYMBOL(file_remove_suid); 1637 1638 /** 1639 * file_update_time - update mtime and ctime time 1640 * @file: file accessed 1641 * 1642 * Update the mtime and ctime members of an inode and mark the inode 1643 * for writeback. Note that this function is meant exclusively for 1644 * usage in the file write path of filesystems, and filesystems may 1645 * choose to explicitly ignore update via this function with the 1646 * S_NOCMTIME inode flag, e.g. for network filesystem where these 1647 * timestamps are handled by the server. This can return an error for 1648 * file systems who need to allocate space in order to update an inode. 1649 */ 1650 1651 int file_update_time(struct file *file) 1652 { 1653 struct inode *inode = file_inode(file); 1654 struct timespec now; 1655 int sync_it = 0; 1656 int ret; 1657 1658 /* First try to exhaust all avenues to not sync */ 1659 if (IS_NOCMTIME(inode)) 1660 return 0; 1661 1662 now = current_fs_time(inode->i_sb); 1663 if (!timespec_equal(&inode->i_mtime, &now)) 1664 sync_it = S_MTIME; 1665 1666 if (!timespec_equal(&inode->i_ctime, &now)) 1667 sync_it |= S_CTIME; 1668 1669 if (IS_I_VERSION(inode)) 1670 sync_it |= S_VERSION; 1671 1672 if (!sync_it) 1673 return 0; 1674 1675 /* Finally allowed to write? Takes lock. */ 1676 if (__mnt_want_write_file(file)) 1677 return 0; 1678 1679 ret = update_time(inode, &now, sync_it); 1680 __mnt_drop_write_file(file); 1681 1682 return ret; 1683 } 1684 EXPORT_SYMBOL(file_update_time); 1685 1686 int inode_needs_sync(struct inode *inode) 1687 { 1688 if (IS_SYNC(inode)) 1689 return 1; 1690 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) 1691 return 1; 1692 return 0; 1693 } 1694 EXPORT_SYMBOL(inode_needs_sync); 1695 1696 int inode_wait(void *word) 1697 { 1698 schedule(); 1699 return 0; 1700 } 1701 EXPORT_SYMBOL(inode_wait); 1702 1703 /* 1704 * If we try to find an inode in the inode hash while it is being 1705 * deleted, we have to wait until the filesystem completes its 1706 * deletion before reporting that it isn't found. This function waits 1707 * until the deletion _might_ have completed. Callers are responsible 1708 * to recheck inode state. 1709 * 1710 * It doesn't matter if I_NEW is not set initially, a call to 1711 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list 1712 * will DTRT. 1713 */ 1714 static void __wait_on_freeing_inode(struct inode *inode) 1715 { 1716 wait_queue_head_t *wq; 1717 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); 1718 wq = bit_waitqueue(&inode->i_state, __I_NEW); 1719 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 1720 spin_unlock(&inode->i_lock); 1721 spin_unlock(&inode_hash_lock); 1722 schedule(); 1723 finish_wait(wq, &wait.wait); 1724 spin_lock(&inode_hash_lock); 1725 } 1726 1727 static __initdata unsigned long ihash_entries; 1728 static int __init set_ihash_entries(char *str) 1729 { 1730 if (!str) 1731 return 0; 1732 ihash_entries = simple_strtoul(str, &str, 0); 1733 return 1; 1734 } 1735 __setup("ihash_entries=", set_ihash_entries); 1736 1737 /* 1738 * Initialize the waitqueues and inode hash table. 1739 */ 1740 void __init inode_init_early(void) 1741 { 1742 unsigned int loop; 1743 1744 /* If hashes are distributed across NUMA nodes, defer 1745 * hash allocation until vmalloc space is available. 1746 */ 1747 if (hashdist) 1748 return; 1749 1750 inode_hashtable = 1751 alloc_large_system_hash("Inode-cache", 1752 sizeof(struct hlist_head), 1753 ihash_entries, 1754 14, 1755 HASH_EARLY, 1756 &i_hash_shift, 1757 &i_hash_mask, 1758 0, 1759 0); 1760 1761 for (loop = 0; loop < (1U << i_hash_shift); loop++) 1762 INIT_HLIST_HEAD(&inode_hashtable[loop]); 1763 } 1764 1765 void __init inode_init(void) 1766 { 1767 unsigned int loop; 1768 1769 /* inode slab cache */ 1770 inode_cachep = kmem_cache_create("inode_cache", 1771 sizeof(struct inode), 1772 0, 1773 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| 1774 SLAB_MEM_SPREAD), 1775 init_once); 1776 1777 /* Hash may have been set up in inode_init_early */ 1778 if (!hashdist) 1779 return; 1780 1781 inode_hashtable = 1782 alloc_large_system_hash("Inode-cache", 1783 sizeof(struct hlist_head), 1784 ihash_entries, 1785 14, 1786 0, 1787 &i_hash_shift, 1788 &i_hash_mask, 1789 0, 1790 0); 1791 1792 for (loop = 0; loop < (1U << i_hash_shift); loop++) 1793 INIT_HLIST_HEAD(&inode_hashtable[loop]); 1794 } 1795 1796 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) 1797 { 1798 inode->i_mode = mode; 1799 if (S_ISCHR(mode)) { 1800 inode->i_fop = &def_chr_fops; 1801 inode->i_rdev = rdev; 1802 } else if (S_ISBLK(mode)) { 1803 inode->i_fop = &def_blk_fops; 1804 inode->i_rdev = rdev; 1805 } else if (S_ISFIFO(mode)) 1806 inode->i_fop = &def_fifo_fops; 1807 else if (S_ISSOCK(mode)) 1808 inode->i_fop = &bad_sock_fops; 1809 else 1810 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" 1811 " inode %s:%lu\n", mode, inode->i_sb->s_id, 1812 inode->i_ino); 1813 } 1814 EXPORT_SYMBOL(init_special_inode); 1815 1816 /** 1817 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards 1818 * @inode: New inode 1819 * @dir: Directory inode 1820 * @mode: mode of the new inode 1821 */ 1822 void inode_init_owner(struct inode *inode, const struct inode *dir, 1823 umode_t mode) 1824 { 1825 inode->i_uid = current_fsuid(); 1826 if (dir && dir->i_mode & S_ISGID) { 1827 inode->i_gid = dir->i_gid; 1828 if (S_ISDIR(mode)) 1829 mode |= S_ISGID; 1830 } else 1831 inode->i_gid = current_fsgid(); 1832 inode->i_mode = mode; 1833 } 1834 EXPORT_SYMBOL(inode_init_owner); 1835 1836 /** 1837 * inode_owner_or_capable - check current task permissions to inode 1838 * @inode: inode being checked 1839 * 1840 * Return true if current either has CAP_FOWNER to the inode, or 1841 * owns the file. 1842 */ 1843 bool inode_owner_or_capable(const struct inode *inode) 1844 { 1845 if (uid_eq(current_fsuid(), inode->i_uid)) 1846 return true; 1847 if (inode_capable(inode, CAP_FOWNER)) 1848 return true; 1849 return false; 1850 } 1851 EXPORT_SYMBOL(inode_owner_or_capable); 1852 1853 /* 1854 * Direct i/o helper functions 1855 */ 1856 static void __inode_dio_wait(struct inode *inode) 1857 { 1858 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP); 1859 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP); 1860 1861 do { 1862 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE); 1863 if (atomic_read(&inode->i_dio_count)) 1864 schedule(); 1865 } while (atomic_read(&inode->i_dio_count)); 1866 finish_wait(wq, &q.wait); 1867 } 1868 1869 /** 1870 * inode_dio_wait - wait for outstanding DIO requests to finish 1871 * @inode: inode to wait for 1872 * 1873 * Waits for all pending direct I/O requests to finish so that we can 1874 * proceed with a truncate or equivalent operation. 1875 * 1876 * Must be called under a lock that serializes taking new references 1877 * to i_dio_count, usually by inode->i_mutex. 1878 */ 1879 void inode_dio_wait(struct inode *inode) 1880 { 1881 if (atomic_read(&inode->i_dio_count)) 1882 __inode_dio_wait(inode); 1883 } 1884 EXPORT_SYMBOL(inode_dio_wait); 1885 1886 /* 1887 * inode_dio_done - signal finish of a direct I/O requests 1888 * @inode: inode the direct I/O happens on 1889 * 1890 * This is called once we've finished processing a direct I/O request, 1891 * and is used to wake up callers waiting for direct I/O to be quiesced. 1892 */ 1893 void inode_dio_done(struct inode *inode) 1894 { 1895 if (atomic_dec_and_test(&inode->i_dio_count)) 1896 wake_up_bit(&inode->i_state, __I_DIO_WAKEUP); 1897 } 1898 EXPORT_SYMBOL(inode_dio_done); 1899