root/fs/inode.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. get_nr_inodes
  2. get_nr_inodes_unused
  3. get_nr_dirty_inodes
  4. proc_nr_inodes
  5. no_open
  6. inode_init_always
  7. free_inode_nonrcu
  8. i_callback
  9. alloc_inode
  10. __destroy_inode
  11. destroy_inode
  12. drop_nlink
  13. clear_nlink
  14. set_nlink
  15. inc_nlink
  16. __address_space_init_once
  17. address_space_init_once
  18. inode_init_once
  19. init_once
  20. __iget
  21. ihold
  22. inode_lru_list_add
  23. inode_add_lru
  24. inode_lru_list_del
  25. inode_sb_list_add
  26. inode_sb_list_del
  27. hash
  28. __insert_inode_hash
  29. __remove_inode_hash
  30. clear_inode
  31. evict
  32. dispose_list
  33. evict_inodes
  34. invalidate_inodes
  35. inode_lru_isolate
  36. prune_icache_sb
  37. find_inode
  38. find_inode_fast
  39. get_next_ino
  40. new_inode_pseudo
  41. new_inode
  42. lockdep_annotate_inode_mutex_key
  43. unlock_new_inode
  44. discard_new_inode
  45. lock_two_nondirectories
  46. unlock_two_nondirectories
  47. inode_insert5
  48. iget5_locked
  49. iget_locked
  50. test_inode_iunique
  51. iunique
  52. igrab
  53. ilookup5_nowait
  54. ilookup5
  55. ilookup
  56. find_inode_nowait
  57. insert_inode_locked
  58. insert_inode_locked4
  59. generic_delete_inode
  60. iput_final
  61. iput
  62. bmap
  63. relatime_need_update
  64. generic_update_time
  65. update_time
  66. atime_needs_update
  67. touch_atime
  68. should_remove_suid
  69. dentry_needs_remove_privs
  70. __remove_privs
  71. file_remove_privs
  72. file_update_time
  73. file_modified
  74. inode_needs_sync
  75. __wait_on_freeing_inode
  76. set_ihash_entries
  77. inode_init_early
  78. inode_init
  79. init_special_inode
  80. inode_init_owner
  81. inode_owner_or_capable
  82. __inode_dio_wait
  83. inode_dio_wait
  84. inode_set_flags
  85. inode_nohighmem
  86. timespec64_trunc
  87. timestamp_truncate
  88. current_time
  89. vfs_ioc_setflags_prepare
  90. vfs_ioc_fssetxattr_check

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * (C) 1997 Linus Torvalds
   4  * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
   5  */
   6 #include <linux/export.h>
   7 #include <linux/fs.h>
   8 #include <linux/mm.h>
   9 #include <linux/backing-dev.h>
  10 #include <linux/hash.h>
  11 #include <linux/swap.h>
  12 #include <linux/security.h>
  13 #include <linux/cdev.h>
  14 #include <linux/memblock.h>
  15 #include <linux/fsnotify.h>
  16 #include <linux/mount.h>
  17 #include <linux/posix_acl.h>
  18 #include <linux/prefetch.h>
  19 #include <linux/buffer_head.h> /* for inode_has_buffers */
  20 #include <linux/ratelimit.h>
  21 #include <linux/list_lru.h>
  22 #include <linux/iversion.h>
  23 #include <trace/events/writeback.h>
  24 #include "internal.h"
  25 
  26 /*
  27  * Inode locking rules:
  28  *
  29  * inode->i_lock protects:
  30  *   inode->i_state, inode->i_hash, __iget()
  31  * Inode LRU list locks protect:
  32  *   inode->i_sb->s_inode_lru, inode->i_lru
  33  * inode->i_sb->s_inode_list_lock protects:
  34  *   inode->i_sb->s_inodes, inode->i_sb_list
  35  * bdi->wb.list_lock protects:
  36  *   bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
  37  * inode_hash_lock protects:
  38  *   inode_hashtable, inode->i_hash
  39  *
  40  * Lock ordering:
  41  *
  42  * inode->i_sb->s_inode_list_lock
  43  *   inode->i_lock
  44  *     Inode LRU list locks
  45  *
  46  * bdi->wb.list_lock
  47  *   inode->i_lock
  48  *
  49  * inode_hash_lock
  50  *   inode->i_sb->s_inode_list_lock
  51  *   inode->i_lock
  52  *
  53  * iunique_lock
  54  *   inode_hash_lock
  55  */
  56 
  57 static unsigned int i_hash_mask __read_mostly;
  58 static unsigned int i_hash_shift __read_mostly;
  59 static struct hlist_head *inode_hashtable __read_mostly;
  60 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
  61 
  62 /*
  63  * Empty aops. Can be used for the cases where the user does not
  64  * define any of the address_space operations.
  65  */
  66 const struct address_space_operations empty_aops = {
  67 };
  68 EXPORT_SYMBOL(empty_aops);
  69 
  70 /*
  71  * Statistics gathering..
  72  */
  73 struct inodes_stat_t inodes_stat;
  74 
  75 static DEFINE_PER_CPU(unsigned long, nr_inodes);
  76 static DEFINE_PER_CPU(unsigned long, nr_unused);
  77 
  78 static struct kmem_cache *inode_cachep __read_mostly;
  79 
  80 static long get_nr_inodes(void)
  81 {
  82         int i;
  83         long sum = 0;
  84         for_each_possible_cpu(i)
  85                 sum += per_cpu(nr_inodes, i);
  86         return sum < 0 ? 0 : sum;
  87 }
  88 
  89 static inline long get_nr_inodes_unused(void)
  90 {
  91         int i;
  92         long sum = 0;
  93         for_each_possible_cpu(i)
  94                 sum += per_cpu(nr_unused, i);
  95         return sum < 0 ? 0 : sum;
  96 }
  97 
  98 long get_nr_dirty_inodes(void)
  99 {
 100         /* not actually dirty inodes, but a wild approximation */
 101         long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
 102         return nr_dirty > 0 ? nr_dirty : 0;
 103 }
 104 
 105 /*
 106  * Handle nr_inode sysctl
 107  */
 108 #ifdef CONFIG_SYSCTL
 109 int proc_nr_inodes(struct ctl_table *table, int write,
 110                    void __user *buffer, size_t *lenp, loff_t *ppos)
 111 {
 112         inodes_stat.nr_inodes = get_nr_inodes();
 113         inodes_stat.nr_unused = get_nr_inodes_unused();
 114         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
 115 }
 116 #endif
 117 
 118 static int no_open(struct inode *inode, struct file *file)
 119 {
 120         return -ENXIO;
 121 }
 122 
 123 /**
 124  * inode_init_always - perform inode structure initialisation
 125  * @sb: superblock inode belongs to
 126  * @inode: inode to initialise
 127  *
 128  * These are initializations that need to be done on every inode
 129  * allocation as the fields are not initialised by slab allocation.
 130  */
 131 int inode_init_always(struct super_block *sb, struct inode *inode)
 132 {
 133         static const struct inode_operations empty_iops;
 134         static const struct file_operations no_open_fops = {.open = no_open};
 135         struct address_space *const mapping = &inode->i_data;
 136 
 137         inode->i_sb = sb;
 138         inode->i_blkbits = sb->s_blocksize_bits;
 139         inode->i_flags = 0;
 140         atomic64_set(&inode->i_sequence, 0);
 141         atomic_set(&inode->i_count, 1);
 142         inode->i_op = &empty_iops;
 143         inode->i_fop = &no_open_fops;
 144         inode->__i_nlink = 1;
 145         inode->i_opflags = 0;
 146         if (sb->s_xattr)
 147                 inode->i_opflags |= IOP_XATTR;
 148         i_uid_write(inode, 0);
 149         i_gid_write(inode, 0);
 150         atomic_set(&inode->i_writecount, 0);
 151         inode->i_size = 0;
 152         inode->i_write_hint = WRITE_LIFE_NOT_SET;
 153         inode->i_blocks = 0;
 154         inode->i_bytes = 0;
 155         inode->i_generation = 0;
 156         inode->i_pipe = NULL;
 157         inode->i_bdev = NULL;
 158         inode->i_cdev = NULL;
 159         inode->i_link = NULL;
 160         inode->i_dir_seq = 0;
 161         inode->i_rdev = 0;
 162         inode->dirtied_when = 0;
 163 
 164 #ifdef CONFIG_CGROUP_WRITEBACK
 165         inode->i_wb_frn_winner = 0;
 166         inode->i_wb_frn_avg_time = 0;
 167         inode->i_wb_frn_history = 0;
 168 #endif
 169 
 170         if (security_inode_alloc(inode))
 171                 goto out;
 172         spin_lock_init(&inode->i_lock);
 173         lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
 174 
 175         init_rwsem(&inode->i_rwsem);
 176         lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
 177 
 178         atomic_set(&inode->i_dio_count, 0);
 179 
 180         mapping->a_ops = &empty_aops;
 181         mapping->host = inode;
 182         mapping->flags = 0;
 183         mapping->wb_err = 0;
 184         atomic_set(&mapping->i_mmap_writable, 0);
 185 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
 186         atomic_set(&mapping->nr_thps, 0);
 187 #endif
 188         mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
 189         mapping->private_data = NULL;
 190         mapping->writeback_index = 0;
 191         inode->i_private = NULL;
 192         inode->i_mapping = mapping;
 193         INIT_HLIST_HEAD(&inode->i_dentry);      /* buggered by rcu freeing */
 194 #ifdef CONFIG_FS_POSIX_ACL
 195         inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
 196 #endif
 197 
 198 #ifdef CONFIG_FSNOTIFY
 199         inode->i_fsnotify_mask = 0;
 200 #endif
 201         inode->i_flctx = NULL;
 202         this_cpu_inc(nr_inodes);
 203 
 204         return 0;
 205 out:
 206         return -ENOMEM;
 207 }
 208 EXPORT_SYMBOL(inode_init_always);
 209 
 210 void free_inode_nonrcu(struct inode *inode)
 211 {
 212         kmem_cache_free(inode_cachep, inode);
 213 }
 214 EXPORT_SYMBOL(free_inode_nonrcu);
 215 
 216 static void i_callback(struct rcu_head *head)
 217 {
 218         struct inode *inode = container_of(head, struct inode, i_rcu);
 219         if (inode->free_inode)
 220                 inode->free_inode(inode);
 221         else
 222                 free_inode_nonrcu(inode);
 223 }
 224 
 225 static struct inode *alloc_inode(struct super_block *sb)
 226 {
 227         const struct super_operations *ops = sb->s_op;
 228         struct inode *inode;
 229 
 230         if (ops->alloc_inode)
 231                 inode = ops->alloc_inode(sb);
 232         else
 233                 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
 234 
 235         if (!inode)
 236                 return NULL;
 237 
 238         if (unlikely(inode_init_always(sb, inode))) {
 239                 if (ops->destroy_inode) {
 240                         ops->destroy_inode(inode);
 241                         if (!ops->free_inode)
 242                                 return NULL;
 243                 }
 244                 inode->free_inode = ops->free_inode;
 245                 i_callback(&inode->i_rcu);
 246                 return NULL;
 247         }
 248 
 249         return inode;
 250 }
 251 
 252 void __destroy_inode(struct inode *inode)
 253 {
 254         BUG_ON(inode_has_buffers(inode));
 255         inode_detach_wb(inode);
 256         security_inode_free(inode);
 257         fsnotify_inode_delete(inode);
 258         locks_free_lock_context(inode);
 259         if (!inode->i_nlink) {
 260                 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
 261                 atomic_long_dec(&inode->i_sb->s_remove_count);
 262         }
 263 
 264 #ifdef CONFIG_FS_POSIX_ACL
 265         if (inode->i_acl && !is_uncached_acl(inode->i_acl))
 266                 posix_acl_release(inode->i_acl);
 267         if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
 268                 posix_acl_release(inode->i_default_acl);
 269 #endif
 270         this_cpu_dec(nr_inodes);
 271 }
 272 EXPORT_SYMBOL(__destroy_inode);
 273 
 274 static void destroy_inode(struct inode *inode)
 275 {
 276         const struct super_operations *ops = inode->i_sb->s_op;
 277 
 278         BUG_ON(!list_empty(&inode->i_lru));
 279         __destroy_inode(inode);
 280         if (ops->destroy_inode) {
 281                 ops->destroy_inode(inode);
 282                 if (!ops->free_inode)
 283                         return;
 284         }
 285         inode->free_inode = ops->free_inode;
 286         call_rcu(&inode->i_rcu, i_callback);
 287 }
 288 
 289 /**
 290  * drop_nlink - directly drop an inode's link count
 291  * @inode: inode
 292  *
 293  * This is a low-level filesystem helper to replace any
 294  * direct filesystem manipulation of i_nlink.  In cases
 295  * where we are attempting to track writes to the
 296  * filesystem, a decrement to zero means an imminent
 297  * write when the file is truncated and actually unlinked
 298  * on the filesystem.
 299  */
 300 void drop_nlink(struct inode *inode)
 301 {
 302         WARN_ON(inode->i_nlink == 0);
 303         inode->__i_nlink--;
 304         if (!inode->i_nlink)
 305                 atomic_long_inc(&inode->i_sb->s_remove_count);
 306 }
 307 EXPORT_SYMBOL(drop_nlink);
 308 
 309 /**
 310  * clear_nlink - directly zero an inode's link count
 311  * @inode: inode
 312  *
 313  * This is a low-level filesystem helper to replace any
 314  * direct filesystem manipulation of i_nlink.  See
 315  * drop_nlink() for why we care about i_nlink hitting zero.
 316  */
 317 void clear_nlink(struct inode *inode)
 318 {
 319         if (inode->i_nlink) {
 320                 inode->__i_nlink = 0;
 321                 atomic_long_inc(&inode->i_sb->s_remove_count);
 322         }
 323 }
 324 EXPORT_SYMBOL(clear_nlink);
 325 
 326 /**
 327  * set_nlink - directly set an inode's link count
 328  * @inode: inode
 329  * @nlink: new nlink (should be non-zero)
 330  *
 331  * This is a low-level filesystem helper to replace any
 332  * direct filesystem manipulation of i_nlink.
 333  */
 334 void set_nlink(struct inode *inode, unsigned int nlink)
 335 {
 336         if (!nlink) {
 337                 clear_nlink(inode);
 338         } else {
 339                 /* Yes, some filesystems do change nlink from zero to one */
 340                 if (inode->i_nlink == 0)
 341                         atomic_long_dec(&inode->i_sb->s_remove_count);
 342 
 343                 inode->__i_nlink = nlink;
 344         }
 345 }
 346 EXPORT_SYMBOL(set_nlink);
 347 
 348 /**
 349  * inc_nlink - directly increment an inode's link count
 350  * @inode: inode
 351  *
 352  * This is a low-level filesystem helper to replace any
 353  * direct filesystem manipulation of i_nlink.  Currently,
 354  * it is only here for parity with dec_nlink().
 355  */
 356 void inc_nlink(struct inode *inode)
 357 {
 358         if (unlikely(inode->i_nlink == 0)) {
 359                 WARN_ON(!(inode->i_state & I_LINKABLE));
 360                 atomic_long_dec(&inode->i_sb->s_remove_count);
 361         }
 362 
 363         inode->__i_nlink++;
 364 }
 365 EXPORT_SYMBOL(inc_nlink);
 366 
 367 static void __address_space_init_once(struct address_space *mapping)
 368 {
 369         xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
 370         init_rwsem(&mapping->i_mmap_rwsem);
 371         INIT_LIST_HEAD(&mapping->private_list);
 372         spin_lock_init(&mapping->private_lock);
 373         mapping->i_mmap = RB_ROOT_CACHED;
 374 }
 375 
 376 void address_space_init_once(struct address_space *mapping)
 377 {
 378         memset(mapping, 0, sizeof(*mapping));
 379         __address_space_init_once(mapping);
 380 }
 381 EXPORT_SYMBOL(address_space_init_once);
 382 
 383 /*
 384  * These are initializations that only need to be done
 385  * once, because the fields are idempotent across use
 386  * of the inode, so let the slab aware of that.
 387  */
 388 void inode_init_once(struct inode *inode)
 389 {
 390         memset(inode, 0, sizeof(*inode));
 391         INIT_HLIST_NODE(&inode->i_hash);
 392         INIT_LIST_HEAD(&inode->i_devices);
 393         INIT_LIST_HEAD(&inode->i_io_list);
 394         INIT_LIST_HEAD(&inode->i_wb_list);
 395         INIT_LIST_HEAD(&inode->i_lru);
 396         __address_space_init_once(&inode->i_data);
 397         i_size_ordered_init(inode);
 398 }
 399 EXPORT_SYMBOL(inode_init_once);
 400 
 401 static void init_once(void *foo)
 402 {
 403         struct inode *inode = (struct inode *) foo;
 404 
 405         inode_init_once(inode);
 406 }
 407 
 408 /*
 409  * inode->i_lock must be held
 410  */
 411 void __iget(struct inode *inode)
 412 {
 413         atomic_inc(&inode->i_count);
 414 }
 415 
 416 /*
 417  * get additional reference to inode; caller must already hold one.
 418  */
 419 void ihold(struct inode *inode)
 420 {
 421         WARN_ON(atomic_inc_return(&inode->i_count) < 2);
 422 }
 423 EXPORT_SYMBOL(ihold);
 424 
 425 static void inode_lru_list_add(struct inode *inode)
 426 {
 427         if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
 428                 this_cpu_inc(nr_unused);
 429         else
 430                 inode->i_state |= I_REFERENCED;
 431 }
 432 
 433 /*
 434  * Add inode to LRU if needed (inode is unused and clean).
 435  *
 436  * Needs inode->i_lock held.
 437  */
 438 void inode_add_lru(struct inode *inode)
 439 {
 440         if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
 441                                 I_FREEING | I_WILL_FREE)) &&
 442             !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
 443                 inode_lru_list_add(inode);
 444 }
 445 
 446 
 447 static void inode_lru_list_del(struct inode *inode)
 448 {
 449 
 450         if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
 451                 this_cpu_dec(nr_unused);
 452 }
 453 
 454 /**
 455  * inode_sb_list_add - add inode to the superblock list of inodes
 456  * @inode: inode to add
 457  */
 458 void inode_sb_list_add(struct inode *inode)
 459 {
 460         spin_lock(&inode->i_sb->s_inode_list_lock);
 461         list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
 462         spin_unlock(&inode->i_sb->s_inode_list_lock);
 463 }
 464 EXPORT_SYMBOL_GPL(inode_sb_list_add);
 465 
 466 static inline void inode_sb_list_del(struct inode *inode)
 467 {
 468         if (!list_empty(&inode->i_sb_list)) {
 469                 spin_lock(&inode->i_sb->s_inode_list_lock);
 470                 list_del_init(&inode->i_sb_list);
 471                 spin_unlock(&inode->i_sb->s_inode_list_lock);
 472         }
 473 }
 474 
 475 static unsigned long hash(struct super_block *sb, unsigned long hashval)
 476 {
 477         unsigned long tmp;
 478 
 479         tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
 480                         L1_CACHE_BYTES;
 481         tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
 482         return tmp & i_hash_mask;
 483 }
 484 
 485 /**
 486  *      __insert_inode_hash - hash an inode
 487  *      @inode: unhashed inode
 488  *      @hashval: unsigned long value used to locate this object in the
 489  *              inode_hashtable.
 490  *
 491  *      Add an inode to the inode hash for this superblock.
 492  */
 493 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
 494 {
 495         struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
 496 
 497         spin_lock(&inode_hash_lock);
 498         spin_lock(&inode->i_lock);
 499         hlist_add_head(&inode->i_hash, b);
 500         spin_unlock(&inode->i_lock);
 501         spin_unlock(&inode_hash_lock);
 502 }
 503 EXPORT_SYMBOL(__insert_inode_hash);
 504 
 505 /**
 506  *      __remove_inode_hash - remove an inode from the hash
 507  *      @inode: inode to unhash
 508  *
 509  *      Remove an inode from the superblock.
 510  */
 511 void __remove_inode_hash(struct inode *inode)
 512 {
 513         spin_lock(&inode_hash_lock);
 514         spin_lock(&inode->i_lock);
 515         hlist_del_init(&inode->i_hash);
 516         spin_unlock(&inode->i_lock);
 517         spin_unlock(&inode_hash_lock);
 518 }
 519 EXPORT_SYMBOL(__remove_inode_hash);
 520 
 521 void clear_inode(struct inode *inode)
 522 {
 523         /*
 524          * We have to cycle the i_pages lock here because reclaim can be in the
 525          * process of removing the last page (in __delete_from_page_cache())
 526          * and we must not free the mapping under it.
 527          */
 528         xa_lock_irq(&inode->i_data.i_pages);
 529         BUG_ON(inode->i_data.nrpages);
 530         BUG_ON(inode->i_data.nrexceptional);
 531         xa_unlock_irq(&inode->i_data.i_pages);
 532         BUG_ON(!list_empty(&inode->i_data.private_list));
 533         BUG_ON(!(inode->i_state & I_FREEING));
 534         BUG_ON(inode->i_state & I_CLEAR);
 535         BUG_ON(!list_empty(&inode->i_wb_list));
 536         /* don't need i_lock here, no concurrent mods to i_state */
 537         inode->i_state = I_FREEING | I_CLEAR;
 538 }
 539 EXPORT_SYMBOL(clear_inode);
 540 
 541 /*
 542  * Free the inode passed in, removing it from the lists it is still connected
 543  * to. We remove any pages still attached to the inode and wait for any IO that
 544  * is still in progress before finally destroying the inode.
 545  *
 546  * An inode must already be marked I_FREEING so that we avoid the inode being
 547  * moved back onto lists if we race with other code that manipulates the lists
 548  * (e.g. writeback_single_inode). The caller is responsible for setting this.
 549  *
 550  * An inode must already be removed from the LRU list before being evicted from
 551  * the cache. This should occur atomically with setting the I_FREEING state
 552  * flag, so no inodes here should ever be on the LRU when being evicted.
 553  */
 554 static void evict(struct inode *inode)
 555 {
 556         const struct super_operations *op = inode->i_sb->s_op;
 557 
 558         BUG_ON(!(inode->i_state & I_FREEING));
 559         BUG_ON(!list_empty(&inode->i_lru));
 560 
 561         if (!list_empty(&inode->i_io_list))
 562                 inode_io_list_del(inode);
 563 
 564         inode_sb_list_del(inode);
 565 
 566         /*
 567          * Wait for flusher thread to be done with the inode so that filesystem
 568          * does not start destroying it while writeback is still running. Since
 569          * the inode has I_FREEING set, flusher thread won't start new work on
 570          * the inode.  We just have to wait for running writeback to finish.
 571          */
 572         inode_wait_for_writeback(inode);
 573 
 574         if (op->evict_inode) {
 575                 op->evict_inode(inode);
 576         } else {
 577                 truncate_inode_pages_final(&inode->i_data);
 578                 clear_inode(inode);
 579         }
 580         if (S_ISBLK(inode->i_mode) && inode->i_bdev)
 581                 bd_forget(inode);
 582         if (S_ISCHR(inode->i_mode) && inode->i_cdev)
 583                 cd_forget(inode);
 584 
 585         remove_inode_hash(inode);
 586 
 587         spin_lock(&inode->i_lock);
 588         wake_up_bit(&inode->i_state, __I_NEW);
 589         BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
 590         spin_unlock(&inode->i_lock);
 591 
 592         destroy_inode(inode);
 593 }
 594 
 595 /*
 596  * dispose_list - dispose of the contents of a local list
 597  * @head: the head of the list to free
 598  *
 599  * Dispose-list gets a local list with local inodes in it, so it doesn't
 600  * need to worry about list corruption and SMP locks.
 601  */
 602 static void dispose_list(struct list_head *head)
 603 {
 604         while (!list_empty(head)) {
 605                 struct inode *inode;
 606 
 607                 inode = list_first_entry(head, struct inode, i_lru);
 608                 list_del_init(&inode->i_lru);
 609 
 610                 evict(inode);
 611                 cond_resched();
 612         }
 613 }
 614 
 615 /**
 616  * evict_inodes - evict all evictable inodes for a superblock
 617  * @sb:         superblock to operate on
 618  *
 619  * Make sure that no inodes with zero refcount are retained.  This is
 620  * called by superblock shutdown after having SB_ACTIVE flag removed,
 621  * so any inode reaching zero refcount during or after that call will
 622  * be immediately evicted.
 623  */
 624 void evict_inodes(struct super_block *sb)
 625 {
 626         struct inode *inode, *next;
 627         LIST_HEAD(dispose);
 628 
 629 again:
 630         spin_lock(&sb->s_inode_list_lock);
 631         list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 632                 if (atomic_read(&inode->i_count))
 633                         continue;
 634 
 635                 spin_lock(&inode->i_lock);
 636                 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 637                         spin_unlock(&inode->i_lock);
 638                         continue;
 639                 }
 640 
 641                 inode->i_state |= I_FREEING;
 642                 inode_lru_list_del(inode);
 643                 spin_unlock(&inode->i_lock);
 644                 list_add(&inode->i_lru, &dispose);
 645 
 646                 /*
 647                  * We can have a ton of inodes to evict at unmount time given
 648                  * enough memory, check to see if we need to go to sleep for a
 649                  * bit so we don't livelock.
 650                  */
 651                 if (need_resched()) {
 652                         spin_unlock(&sb->s_inode_list_lock);
 653                         cond_resched();
 654                         dispose_list(&dispose);
 655                         goto again;
 656                 }
 657         }
 658         spin_unlock(&sb->s_inode_list_lock);
 659 
 660         dispose_list(&dispose);
 661 }
 662 EXPORT_SYMBOL_GPL(evict_inodes);
 663 
 664 /**
 665  * invalidate_inodes    - attempt to free all inodes on a superblock
 666  * @sb:         superblock to operate on
 667  * @kill_dirty: flag to guide handling of dirty inodes
 668  *
 669  * Attempts to free all inodes for a given superblock.  If there were any
 670  * busy inodes return a non-zero value, else zero.
 671  * If @kill_dirty is set, discard dirty inodes too, otherwise treat
 672  * them as busy.
 673  */
 674 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
 675 {
 676         int busy = 0;
 677         struct inode *inode, *next;
 678         LIST_HEAD(dispose);
 679 
 680 again:
 681         spin_lock(&sb->s_inode_list_lock);
 682         list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
 683                 spin_lock(&inode->i_lock);
 684                 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 685                         spin_unlock(&inode->i_lock);
 686                         continue;
 687                 }
 688                 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
 689                         spin_unlock(&inode->i_lock);
 690                         busy = 1;
 691                         continue;
 692                 }
 693                 if (atomic_read(&inode->i_count)) {
 694                         spin_unlock(&inode->i_lock);
 695                         busy = 1;
 696                         continue;
 697                 }
 698 
 699                 inode->i_state |= I_FREEING;
 700                 inode_lru_list_del(inode);
 701                 spin_unlock(&inode->i_lock);
 702                 list_add(&inode->i_lru, &dispose);
 703                 if (need_resched()) {
 704                         spin_unlock(&sb->s_inode_list_lock);
 705                         cond_resched();
 706                         dispose_list(&dispose);
 707                         goto again;
 708                 }
 709         }
 710         spin_unlock(&sb->s_inode_list_lock);
 711 
 712         dispose_list(&dispose);
 713 
 714         return busy;
 715 }
 716 
 717 /*
 718  * Isolate the inode from the LRU in preparation for freeing it.
 719  *
 720  * Any inodes which are pinned purely because of attached pagecache have their
 721  * pagecache removed.  If the inode has metadata buffers attached to
 722  * mapping->private_list then try to remove them.
 723  *
 724  * If the inode has the I_REFERENCED flag set, then it means that it has been
 725  * used recently - the flag is set in iput_final(). When we encounter such an
 726  * inode, clear the flag and move it to the back of the LRU so it gets another
 727  * pass through the LRU before it gets reclaimed. This is necessary because of
 728  * the fact we are doing lazy LRU updates to minimise lock contention so the
 729  * LRU does not have strict ordering. Hence we don't want to reclaim inodes
 730  * with this flag set because they are the inodes that are out of order.
 731  */
 732 static enum lru_status inode_lru_isolate(struct list_head *item,
 733                 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
 734 {
 735         struct list_head *freeable = arg;
 736         struct inode    *inode = container_of(item, struct inode, i_lru);
 737 
 738         /*
 739          * we are inverting the lru lock/inode->i_lock here, so use a trylock.
 740          * If we fail to get the lock, just skip it.
 741          */
 742         if (!spin_trylock(&inode->i_lock))
 743                 return LRU_SKIP;
 744 
 745         /*
 746          * Referenced or dirty inodes are still in use. Give them another pass
 747          * through the LRU as we canot reclaim them now.
 748          */
 749         if (atomic_read(&inode->i_count) ||
 750             (inode->i_state & ~I_REFERENCED)) {
 751                 list_lru_isolate(lru, &inode->i_lru);
 752                 spin_unlock(&inode->i_lock);
 753                 this_cpu_dec(nr_unused);
 754                 return LRU_REMOVED;
 755         }
 756 
 757         /* recently referenced inodes get one more pass */
 758         if (inode->i_state & I_REFERENCED) {
 759                 inode->i_state &= ~I_REFERENCED;
 760                 spin_unlock(&inode->i_lock);
 761                 return LRU_ROTATE;
 762         }
 763 
 764         if (inode_has_buffers(inode) || inode->i_data.nrpages) {
 765                 __iget(inode);
 766                 spin_unlock(&inode->i_lock);
 767                 spin_unlock(lru_lock);
 768                 if (remove_inode_buffers(inode)) {
 769                         unsigned long reap;
 770                         reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
 771                         if (current_is_kswapd())
 772                                 __count_vm_events(KSWAPD_INODESTEAL, reap);
 773                         else
 774                                 __count_vm_events(PGINODESTEAL, reap);
 775                         if (current->reclaim_state)
 776                                 current->reclaim_state->reclaimed_slab += reap;
 777                 }
 778                 iput(inode);
 779                 spin_lock(lru_lock);
 780                 return LRU_RETRY;
 781         }
 782 
 783         WARN_ON(inode->i_state & I_NEW);
 784         inode->i_state |= I_FREEING;
 785         list_lru_isolate_move(lru, &inode->i_lru, freeable);
 786         spin_unlock(&inode->i_lock);
 787 
 788         this_cpu_dec(nr_unused);
 789         return LRU_REMOVED;
 790 }
 791 
 792 /*
 793  * Walk the superblock inode LRU for freeable inodes and attempt to free them.
 794  * This is called from the superblock shrinker function with a number of inodes
 795  * to trim from the LRU. Inodes to be freed are moved to a temporary list and
 796  * then are freed outside inode_lock by dispose_list().
 797  */
 798 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
 799 {
 800         LIST_HEAD(freeable);
 801         long freed;
 802 
 803         freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
 804                                      inode_lru_isolate, &freeable);
 805         dispose_list(&freeable);
 806         return freed;
 807 }
 808 
 809 static void __wait_on_freeing_inode(struct inode *inode);
 810 /*
 811  * Called with the inode lock held.
 812  */
 813 static struct inode *find_inode(struct super_block *sb,
 814                                 struct hlist_head *head,
 815                                 int (*test)(struct inode *, void *),
 816                                 void *data)
 817 {
 818         struct inode *inode = NULL;
 819 
 820 repeat:
 821         hlist_for_each_entry(inode, head, i_hash) {
 822                 if (inode->i_sb != sb)
 823                         continue;
 824                 if (!test(inode, data))
 825                         continue;
 826                 spin_lock(&inode->i_lock);
 827                 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 828                         __wait_on_freeing_inode(inode);
 829                         goto repeat;
 830                 }
 831                 if (unlikely(inode->i_state & I_CREATING)) {
 832                         spin_unlock(&inode->i_lock);
 833                         return ERR_PTR(-ESTALE);
 834                 }
 835                 __iget(inode);
 836                 spin_unlock(&inode->i_lock);
 837                 return inode;
 838         }
 839         return NULL;
 840 }
 841 
 842 /*
 843  * find_inode_fast is the fast path version of find_inode, see the comment at
 844  * iget_locked for details.
 845  */
 846 static struct inode *find_inode_fast(struct super_block *sb,
 847                                 struct hlist_head *head, unsigned long ino)
 848 {
 849         struct inode *inode = NULL;
 850 
 851 repeat:
 852         hlist_for_each_entry(inode, head, i_hash) {
 853                 if (inode->i_ino != ino)
 854                         continue;
 855                 if (inode->i_sb != sb)
 856                         continue;
 857                 spin_lock(&inode->i_lock);
 858                 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
 859                         __wait_on_freeing_inode(inode);
 860                         goto repeat;
 861                 }
 862                 if (unlikely(inode->i_state & I_CREATING)) {
 863                         spin_unlock(&inode->i_lock);
 864                         return ERR_PTR(-ESTALE);
 865                 }
 866                 __iget(inode);
 867                 spin_unlock(&inode->i_lock);
 868                 return inode;
 869         }
 870         return NULL;
 871 }
 872 
 873 /*
 874  * Each cpu owns a range of LAST_INO_BATCH numbers.
 875  * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
 876  * to renew the exhausted range.
 877  *
 878  * This does not significantly increase overflow rate because every CPU can
 879  * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
 880  * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
 881  * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
 882  * overflow rate by 2x, which does not seem too significant.
 883  *
 884  * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
 885  * error if st_ino won't fit in target struct field. Use 32bit counter
 886  * here to attempt to avoid that.
 887  */
 888 #define LAST_INO_BATCH 1024
 889 static DEFINE_PER_CPU(unsigned int, last_ino);
 890 
 891 unsigned int get_next_ino(void)
 892 {
 893         unsigned int *p = &get_cpu_var(last_ino);
 894         unsigned int res = *p;
 895 
 896 #ifdef CONFIG_SMP
 897         if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
 898                 static atomic_t shared_last_ino;
 899                 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
 900 
 901                 res = next - LAST_INO_BATCH;
 902         }
 903 #endif
 904 
 905         res++;
 906         /* get_next_ino should not provide a 0 inode number */
 907         if (unlikely(!res))
 908                 res++;
 909         *p = res;
 910         put_cpu_var(last_ino);
 911         return res;
 912 }
 913 EXPORT_SYMBOL(get_next_ino);
 914 
 915 /**
 916  *      new_inode_pseudo        - obtain an inode
 917  *      @sb: superblock
 918  *
 919  *      Allocates a new inode for given superblock.
 920  *      Inode wont be chained in superblock s_inodes list
 921  *      This means :
 922  *      - fs can't be unmount
 923  *      - quotas, fsnotify, writeback can't work
 924  */
 925 struct inode *new_inode_pseudo(struct super_block *sb)
 926 {
 927         struct inode *inode = alloc_inode(sb);
 928 
 929         if (inode) {
 930                 spin_lock(&inode->i_lock);
 931                 inode->i_state = 0;
 932                 spin_unlock(&inode->i_lock);
 933                 INIT_LIST_HEAD(&inode->i_sb_list);
 934         }
 935         return inode;
 936 }
 937 
 938 /**
 939  *      new_inode       - obtain an inode
 940  *      @sb: superblock
 941  *
 942  *      Allocates a new inode for given superblock. The default gfp_mask
 943  *      for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
 944  *      If HIGHMEM pages are unsuitable or it is known that pages allocated
 945  *      for the page cache are not reclaimable or migratable,
 946  *      mapping_set_gfp_mask() must be called with suitable flags on the
 947  *      newly created inode's mapping
 948  *
 949  */
 950 struct inode *new_inode(struct super_block *sb)
 951 {
 952         struct inode *inode;
 953 
 954         spin_lock_prefetch(&sb->s_inode_list_lock);
 955 
 956         inode = new_inode_pseudo(sb);
 957         if (inode)
 958                 inode_sb_list_add(inode);
 959         return inode;
 960 }
 961 EXPORT_SYMBOL(new_inode);
 962 
 963 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 964 void lockdep_annotate_inode_mutex_key(struct inode *inode)
 965 {
 966         if (S_ISDIR(inode->i_mode)) {
 967                 struct file_system_type *type = inode->i_sb->s_type;
 968 
 969                 /* Set new key only if filesystem hasn't already changed it */
 970                 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
 971                         /*
 972                          * ensure nobody is actually holding i_mutex
 973                          */
 974                         // mutex_destroy(&inode->i_mutex);
 975                         init_rwsem(&inode->i_rwsem);
 976                         lockdep_set_class(&inode->i_rwsem,
 977                                           &type->i_mutex_dir_key);
 978                 }
 979         }
 980 }
 981 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
 982 #endif
 983 
 984 /**
 985  * unlock_new_inode - clear the I_NEW state and wake up any waiters
 986  * @inode:      new inode to unlock
 987  *
 988  * Called when the inode is fully initialised to clear the new state of the
 989  * inode and wake up anyone waiting for the inode to finish initialisation.
 990  */
 991 void unlock_new_inode(struct inode *inode)
 992 {
 993         lockdep_annotate_inode_mutex_key(inode);
 994         spin_lock(&inode->i_lock);
 995         WARN_ON(!(inode->i_state & I_NEW));
 996         inode->i_state &= ~I_NEW & ~I_CREATING;
 997         smp_mb();
 998         wake_up_bit(&inode->i_state, __I_NEW);
 999         spin_unlock(&inode->i_lock);
1000 }
1001 EXPORT_SYMBOL(unlock_new_inode);
1002 
1003 void discard_new_inode(struct inode *inode)
1004 {
1005         lockdep_annotate_inode_mutex_key(inode);
1006         spin_lock(&inode->i_lock);
1007         WARN_ON(!(inode->i_state & I_NEW));
1008         inode->i_state &= ~I_NEW;
1009         smp_mb();
1010         wake_up_bit(&inode->i_state, __I_NEW);
1011         spin_unlock(&inode->i_lock);
1012         iput(inode);
1013 }
1014 EXPORT_SYMBOL(discard_new_inode);
1015 
1016 /**
1017  * lock_two_nondirectories - take two i_mutexes on non-directory objects
1018  *
1019  * Lock any non-NULL argument that is not a directory.
1020  * Zero, one or two objects may be locked by this function.
1021  *
1022  * @inode1: first inode to lock
1023  * @inode2: second inode to lock
1024  */
1025 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1026 {
1027         if (inode1 > inode2)
1028                 swap(inode1, inode2);
1029 
1030         if (inode1 && !S_ISDIR(inode1->i_mode))
1031                 inode_lock(inode1);
1032         if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1033                 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1034 }
1035 EXPORT_SYMBOL(lock_two_nondirectories);
1036 
1037 /**
1038  * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1039  * @inode1: first inode to unlock
1040  * @inode2: second inode to unlock
1041  */
1042 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1043 {
1044         if (inode1 && !S_ISDIR(inode1->i_mode))
1045                 inode_unlock(inode1);
1046         if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1047                 inode_unlock(inode2);
1048 }
1049 EXPORT_SYMBOL(unlock_two_nondirectories);
1050 
1051 /**
1052  * inode_insert5 - obtain an inode from a mounted file system
1053  * @inode:      pre-allocated inode to use for insert to cache
1054  * @hashval:    hash value (usually inode number) to get
1055  * @test:       callback used for comparisons between inodes
1056  * @set:        callback used to initialize a new struct inode
1057  * @data:       opaque data pointer to pass to @test and @set
1058  *
1059  * Search for the inode specified by @hashval and @data in the inode cache,
1060  * and if present it is return it with an increased reference count. This is
1061  * a variant of iget5_locked() for callers that don't want to fail on memory
1062  * allocation of inode.
1063  *
1064  * If the inode is not in cache, insert the pre-allocated inode to cache and
1065  * return it locked, hashed, and with the I_NEW flag set. The file system gets
1066  * to fill it in before unlocking it via unlock_new_inode().
1067  *
1068  * Note both @test and @set are called with the inode_hash_lock held, so can't
1069  * sleep.
1070  */
1071 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1072                             int (*test)(struct inode *, void *),
1073                             int (*set)(struct inode *, void *), void *data)
1074 {
1075         struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1076         struct inode *old;
1077         bool creating = inode->i_state & I_CREATING;
1078 
1079 again:
1080         spin_lock(&inode_hash_lock);
1081         old = find_inode(inode->i_sb, head, test, data);
1082         if (unlikely(old)) {
1083                 /*
1084                  * Uhhuh, somebody else created the same inode under us.
1085                  * Use the old inode instead of the preallocated one.
1086                  */
1087                 spin_unlock(&inode_hash_lock);
1088                 if (IS_ERR(old))
1089                         return NULL;
1090                 wait_on_inode(old);
1091                 if (unlikely(inode_unhashed(old))) {
1092                         iput(old);
1093                         goto again;
1094                 }
1095                 return old;
1096         }
1097 
1098         if (set && unlikely(set(inode, data))) {
1099                 inode = NULL;
1100                 goto unlock;
1101         }
1102 
1103         /*
1104          * Return the locked inode with I_NEW set, the
1105          * caller is responsible for filling in the contents
1106          */
1107         spin_lock(&inode->i_lock);
1108         inode->i_state |= I_NEW;
1109         hlist_add_head(&inode->i_hash, head);
1110         spin_unlock(&inode->i_lock);
1111         if (!creating)
1112                 inode_sb_list_add(inode);
1113 unlock:
1114         spin_unlock(&inode_hash_lock);
1115 
1116         return inode;
1117 }
1118 EXPORT_SYMBOL(inode_insert5);
1119 
1120 /**
1121  * iget5_locked - obtain an inode from a mounted file system
1122  * @sb:         super block of file system
1123  * @hashval:    hash value (usually inode number) to get
1124  * @test:       callback used for comparisons between inodes
1125  * @set:        callback used to initialize a new struct inode
1126  * @data:       opaque data pointer to pass to @test and @set
1127  *
1128  * Search for the inode specified by @hashval and @data in the inode cache,
1129  * and if present it is return it with an increased reference count. This is
1130  * a generalized version of iget_locked() for file systems where the inode
1131  * number is not sufficient for unique identification of an inode.
1132  *
1133  * If the inode is not in cache, allocate a new inode and return it locked,
1134  * hashed, and with the I_NEW flag set. The file system gets to fill it in
1135  * before unlocking it via unlock_new_inode().
1136  *
1137  * Note both @test and @set are called with the inode_hash_lock held, so can't
1138  * sleep.
1139  */
1140 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1141                 int (*test)(struct inode *, void *),
1142                 int (*set)(struct inode *, void *), void *data)
1143 {
1144         struct inode *inode = ilookup5(sb, hashval, test, data);
1145 
1146         if (!inode) {
1147                 struct inode *new = alloc_inode(sb);
1148 
1149                 if (new) {
1150                         new->i_state = 0;
1151                         inode = inode_insert5(new, hashval, test, set, data);
1152                         if (unlikely(inode != new))
1153                                 destroy_inode(new);
1154                 }
1155         }
1156         return inode;
1157 }
1158 EXPORT_SYMBOL(iget5_locked);
1159 
1160 /**
1161  * iget_locked - obtain an inode from a mounted file system
1162  * @sb:         super block of file system
1163  * @ino:        inode number to get
1164  *
1165  * Search for the inode specified by @ino in the inode cache and if present
1166  * return it with an increased reference count. This is for file systems
1167  * where the inode number is sufficient for unique identification of an inode.
1168  *
1169  * If the inode is not in cache, allocate a new inode and return it locked,
1170  * hashed, and with the I_NEW flag set.  The file system gets to fill it in
1171  * before unlocking it via unlock_new_inode().
1172  */
1173 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1174 {
1175         struct hlist_head *head = inode_hashtable + hash(sb, ino);
1176         struct inode *inode;
1177 again:
1178         spin_lock(&inode_hash_lock);
1179         inode = find_inode_fast(sb, head, ino);
1180         spin_unlock(&inode_hash_lock);
1181         if (inode) {
1182                 if (IS_ERR(inode))
1183                         return NULL;
1184                 wait_on_inode(inode);
1185                 if (unlikely(inode_unhashed(inode))) {
1186                         iput(inode);
1187                         goto again;
1188                 }
1189                 return inode;
1190         }
1191 
1192         inode = alloc_inode(sb);
1193         if (inode) {
1194                 struct inode *old;
1195 
1196                 spin_lock(&inode_hash_lock);
1197                 /* We released the lock, so.. */
1198                 old = find_inode_fast(sb, head, ino);
1199                 if (!old) {
1200                         inode->i_ino = ino;
1201                         spin_lock(&inode->i_lock);
1202                         inode->i_state = I_NEW;
1203                         hlist_add_head(&inode->i_hash, head);
1204                         spin_unlock(&inode->i_lock);
1205                         inode_sb_list_add(inode);
1206                         spin_unlock(&inode_hash_lock);
1207 
1208                         /* Return the locked inode with I_NEW set, the
1209                          * caller is responsible for filling in the contents
1210                          */
1211                         return inode;
1212                 }
1213 
1214                 /*
1215                  * Uhhuh, somebody else created the same inode under
1216                  * us. Use the old inode instead of the one we just
1217                  * allocated.
1218                  */
1219                 spin_unlock(&inode_hash_lock);
1220                 destroy_inode(inode);
1221                 if (IS_ERR(old))
1222                         return NULL;
1223                 inode = old;
1224                 wait_on_inode(inode);
1225                 if (unlikely(inode_unhashed(inode))) {
1226                         iput(inode);
1227                         goto again;
1228                 }
1229         }
1230         return inode;
1231 }
1232 EXPORT_SYMBOL(iget_locked);
1233 
1234 /*
1235  * search the inode cache for a matching inode number.
1236  * If we find one, then the inode number we are trying to
1237  * allocate is not unique and so we should not use it.
1238  *
1239  * Returns 1 if the inode number is unique, 0 if it is not.
1240  */
1241 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1242 {
1243         struct hlist_head *b = inode_hashtable + hash(sb, ino);
1244         struct inode *inode;
1245 
1246         spin_lock(&inode_hash_lock);
1247         hlist_for_each_entry(inode, b, i_hash) {
1248                 if (inode->i_ino == ino && inode->i_sb == sb) {
1249                         spin_unlock(&inode_hash_lock);
1250                         return 0;
1251                 }
1252         }
1253         spin_unlock(&inode_hash_lock);
1254 
1255         return 1;
1256 }
1257 
1258 /**
1259  *      iunique - get a unique inode number
1260  *      @sb: superblock
1261  *      @max_reserved: highest reserved inode number
1262  *
1263  *      Obtain an inode number that is unique on the system for a given
1264  *      superblock. This is used by file systems that have no natural
1265  *      permanent inode numbering system. An inode number is returned that
1266  *      is higher than the reserved limit but unique.
1267  *
1268  *      BUGS:
1269  *      With a large number of inodes live on the file system this function
1270  *      currently becomes quite slow.
1271  */
1272 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1273 {
1274         /*
1275          * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1276          * error if st_ino won't fit in target struct field. Use 32bit counter
1277          * here to attempt to avoid that.
1278          */
1279         static DEFINE_SPINLOCK(iunique_lock);
1280         static unsigned int counter;
1281         ino_t res;
1282 
1283         spin_lock(&iunique_lock);
1284         do {
1285                 if (counter <= max_reserved)
1286                         counter = max_reserved + 1;
1287                 res = counter++;
1288         } while (!test_inode_iunique(sb, res));
1289         spin_unlock(&iunique_lock);
1290 
1291         return res;
1292 }
1293 EXPORT_SYMBOL(iunique);
1294 
1295 struct inode *igrab(struct inode *inode)
1296 {
1297         spin_lock(&inode->i_lock);
1298         if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1299                 __iget(inode);
1300                 spin_unlock(&inode->i_lock);
1301         } else {
1302                 spin_unlock(&inode->i_lock);
1303                 /*
1304                  * Handle the case where s_op->clear_inode is not been
1305                  * called yet, and somebody is calling igrab
1306                  * while the inode is getting freed.
1307                  */
1308                 inode = NULL;
1309         }
1310         return inode;
1311 }
1312 EXPORT_SYMBOL(igrab);
1313 
1314 /**
1315  * ilookup5_nowait - search for an inode in the inode cache
1316  * @sb:         super block of file system to search
1317  * @hashval:    hash value (usually inode number) to search for
1318  * @test:       callback used for comparisons between inodes
1319  * @data:       opaque data pointer to pass to @test
1320  *
1321  * Search for the inode specified by @hashval and @data in the inode cache.
1322  * If the inode is in the cache, the inode is returned with an incremented
1323  * reference count.
1324  *
1325  * Note: I_NEW is not waited upon so you have to be very careful what you do
1326  * with the returned inode.  You probably should be using ilookup5() instead.
1327  *
1328  * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1329  */
1330 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1331                 int (*test)(struct inode *, void *), void *data)
1332 {
1333         struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1334         struct inode *inode;
1335 
1336         spin_lock(&inode_hash_lock);
1337         inode = find_inode(sb, head, test, data);
1338         spin_unlock(&inode_hash_lock);
1339 
1340         return IS_ERR(inode) ? NULL : inode;
1341 }
1342 EXPORT_SYMBOL(ilookup5_nowait);
1343 
1344 /**
1345  * ilookup5 - search for an inode in the inode cache
1346  * @sb:         super block of file system to search
1347  * @hashval:    hash value (usually inode number) to search for
1348  * @test:       callback used for comparisons between inodes
1349  * @data:       opaque data pointer to pass to @test
1350  *
1351  * Search for the inode specified by @hashval and @data in the inode cache,
1352  * and if the inode is in the cache, return the inode with an incremented
1353  * reference count.  Waits on I_NEW before returning the inode.
1354  * returned with an incremented reference count.
1355  *
1356  * This is a generalized version of ilookup() for file systems where the
1357  * inode number is not sufficient for unique identification of an inode.
1358  *
1359  * Note: @test is called with the inode_hash_lock held, so can't sleep.
1360  */
1361 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1362                 int (*test)(struct inode *, void *), void *data)
1363 {
1364         struct inode *inode;
1365 again:
1366         inode = ilookup5_nowait(sb, hashval, test, data);
1367         if (inode) {
1368                 wait_on_inode(inode);
1369                 if (unlikely(inode_unhashed(inode))) {
1370                         iput(inode);
1371                         goto again;
1372                 }
1373         }
1374         return inode;
1375 }
1376 EXPORT_SYMBOL(ilookup5);
1377 
1378 /**
1379  * ilookup - search for an inode in the inode cache
1380  * @sb:         super block of file system to search
1381  * @ino:        inode number to search for
1382  *
1383  * Search for the inode @ino in the inode cache, and if the inode is in the
1384  * cache, the inode is returned with an incremented reference count.
1385  */
1386 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1387 {
1388         struct hlist_head *head = inode_hashtable + hash(sb, ino);
1389         struct inode *inode;
1390 again:
1391         spin_lock(&inode_hash_lock);
1392         inode = find_inode_fast(sb, head, ino);
1393         spin_unlock(&inode_hash_lock);
1394 
1395         if (inode) {
1396                 if (IS_ERR(inode))
1397                         return NULL;
1398                 wait_on_inode(inode);
1399                 if (unlikely(inode_unhashed(inode))) {
1400                         iput(inode);
1401                         goto again;
1402                 }
1403         }
1404         return inode;
1405 }
1406 EXPORT_SYMBOL(ilookup);
1407 
1408 /**
1409  * find_inode_nowait - find an inode in the inode cache
1410  * @sb:         super block of file system to search
1411  * @hashval:    hash value (usually inode number) to search for
1412  * @match:      callback used for comparisons between inodes
1413  * @data:       opaque data pointer to pass to @match
1414  *
1415  * Search for the inode specified by @hashval and @data in the inode
1416  * cache, where the helper function @match will return 0 if the inode
1417  * does not match, 1 if the inode does match, and -1 if the search
1418  * should be stopped.  The @match function must be responsible for
1419  * taking the i_lock spin_lock and checking i_state for an inode being
1420  * freed or being initialized, and incrementing the reference count
1421  * before returning 1.  It also must not sleep, since it is called with
1422  * the inode_hash_lock spinlock held.
1423  *
1424  * This is a even more generalized version of ilookup5() when the
1425  * function must never block --- find_inode() can block in
1426  * __wait_on_freeing_inode() --- or when the caller can not increment
1427  * the reference count because the resulting iput() might cause an
1428  * inode eviction.  The tradeoff is that the @match funtion must be
1429  * very carefully implemented.
1430  */
1431 struct inode *find_inode_nowait(struct super_block *sb,
1432                                 unsigned long hashval,
1433                                 int (*match)(struct inode *, unsigned long,
1434                                              void *),
1435                                 void *data)
1436 {
1437         struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1438         struct inode *inode, *ret_inode = NULL;
1439         int mval;
1440 
1441         spin_lock(&inode_hash_lock);
1442         hlist_for_each_entry(inode, head, i_hash) {
1443                 if (inode->i_sb != sb)
1444                         continue;
1445                 mval = match(inode, hashval, data);
1446                 if (mval == 0)
1447                         continue;
1448                 if (mval == 1)
1449                         ret_inode = inode;
1450                 goto out;
1451         }
1452 out:
1453         spin_unlock(&inode_hash_lock);
1454         return ret_inode;
1455 }
1456 EXPORT_SYMBOL(find_inode_nowait);
1457 
1458 int insert_inode_locked(struct inode *inode)
1459 {
1460         struct super_block *sb = inode->i_sb;
1461         ino_t ino = inode->i_ino;
1462         struct hlist_head *head = inode_hashtable + hash(sb, ino);
1463 
1464         while (1) {
1465                 struct inode *old = NULL;
1466                 spin_lock(&inode_hash_lock);
1467                 hlist_for_each_entry(old, head, i_hash) {
1468                         if (old->i_ino != ino)
1469                                 continue;
1470                         if (old->i_sb != sb)
1471                                 continue;
1472                         spin_lock(&old->i_lock);
1473                         if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1474                                 spin_unlock(&old->i_lock);
1475                                 continue;
1476                         }
1477                         break;
1478                 }
1479                 if (likely(!old)) {
1480                         spin_lock(&inode->i_lock);
1481                         inode->i_state |= I_NEW | I_CREATING;
1482                         hlist_add_head(&inode->i_hash, head);
1483                         spin_unlock(&inode->i_lock);
1484                         spin_unlock(&inode_hash_lock);
1485                         return 0;
1486                 }
1487                 if (unlikely(old->i_state & I_CREATING)) {
1488                         spin_unlock(&old->i_lock);
1489                         spin_unlock(&inode_hash_lock);
1490                         return -EBUSY;
1491                 }
1492                 __iget(old);
1493                 spin_unlock(&old->i_lock);
1494                 spin_unlock(&inode_hash_lock);
1495                 wait_on_inode(old);
1496                 if (unlikely(!inode_unhashed(old))) {
1497                         iput(old);
1498                         return -EBUSY;
1499                 }
1500                 iput(old);
1501         }
1502 }
1503 EXPORT_SYMBOL(insert_inode_locked);
1504 
1505 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1506                 int (*test)(struct inode *, void *), void *data)
1507 {
1508         struct inode *old;
1509 
1510         inode->i_state |= I_CREATING;
1511         old = inode_insert5(inode, hashval, test, NULL, data);
1512 
1513         if (old != inode) {
1514                 iput(old);
1515                 return -EBUSY;
1516         }
1517         return 0;
1518 }
1519 EXPORT_SYMBOL(insert_inode_locked4);
1520 
1521 
1522 int generic_delete_inode(struct inode *inode)
1523 {
1524         return 1;
1525 }
1526 EXPORT_SYMBOL(generic_delete_inode);
1527 
1528 /*
1529  * Called when we're dropping the last reference
1530  * to an inode.
1531  *
1532  * Call the FS "drop_inode()" function, defaulting to
1533  * the legacy UNIX filesystem behaviour.  If it tells
1534  * us to evict inode, do so.  Otherwise, retain inode
1535  * in cache if fs is alive, sync and evict if fs is
1536  * shutting down.
1537  */
1538 static void iput_final(struct inode *inode)
1539 {
1540         struct super_block *sb = inode->i_sb;
1541         const struct super_operations *op = inode->i_sb->s_op;
1542         int drop;
1543 
1544         WARN_ON(inode->i_state & I_NEW);
1545 
1546         if (op->drop_inode)
1547                 drop = op->drop_inode(inode);
1548         else
1549                 drop = generic_drop_inode(inode);
1550 
1551         if (!drop && (sb->s_flags & SB_ACTIVE)) {
1552                 inode_add_lru(inode);
1553                 spin_unlock(&inode->i_lock);
1554                 return;
1555         }
1556 
1557         if (!drop) {
1558                 inode->i_state |= I_WILL_FREE;
1559                 spin_unlock(&inode->i_lock);
1560                 write_inode_now(inode, 1);
1561                 spin_lock(&inode->i_lock);
1562                 WARN_ON(inode->i_state & I_NEW);
1563                 inode->i_state &= ~I_WILL_FREE;
1564         }
1565 
1566         inode->i_state |= I_FREEING;
1567         if (!list_empty(&inode->i_lru))
1568                 inode_lru_list_del(inode);
1569         spin_unlock(&inode->i_lock);
1570 
1571         evict(inode);
1572 }
1573 
1574 /**
1575  *      iput    - put an inode
1576  *      @inode: inode to put
1577  *
1578  *      Puts an inode, dropping its usage count. If the inode use count hits
1579  *      zero, the inode is then freed and may also be destroyed.
1580  *
1581  *      Consequently, iput() can sleep.
1582  */
1583 void iput(struct inode *inode)
1584 {
1585         if (!inode)
1586                 return;
1587         BUG_ON(inode->i_state & I_CLEAR);
1588 retry:
1589         if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1590                 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1591                         atomic_inc(&inode->i_count);
1592                         spin_unlock(&inode->i_lock);
1593                         trace_writeback_lazytime_iput(inode);
1594                         mark_inode_dirty_sync(inode);
1595                         goto retry;
1596                 }
1597                 iput_final(inode);
1598         }
1599 }
1600 EXPORT_SYMBOL(iput);
1601 
1602 /**
1603  *      bmap    - find a block number in a file
1604  *      @inode: inode of file
1605  *      @block: block to find
1606  *
1607  *      Returns the block number on the device holding the inode that
1608  *      is the disk block number for the block of the file requested.
1609  *      That is, asked for block 4 of inode 1 the function will return the
1610  *      disk block relative to the disk start that holds that block of the
1611  *      file.
1612  */
1613 sector_t bmap(struct inode *inode, sector_t block)
1614 {
1615         sector_t res = 0;
1616         if (inode->i_mapping->a_ops->bmap)
1617                 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1618         return res;
1619 }
1620 EXPORT_SYMBOL(bmap);
1621 
1622 /*
1623  * With relative atime, only update atime if the previous atime is
1624  * earlier than either the ctime or mtime or if at least a day has
1625  * passed since the last atime update.
1626  */
1627 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1628                              struct timespec64 now)
1629 {
1630 
1631         if (!(mnt->mnt_flags & MNT_RELATIME))
1632                 return 1;
1633         /*
1634          * Is mtime younger than atime? If yes, update atime:
1635          */
1636         if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1637                 return 1;
1638         /*
1639          * Is ctime younger than atime? If yes, update atime:
1640          */
1641         if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1642                 return 1;
1643 
1644         /*
1645          * Is the previous atime value older than a day? If yes,
1646          * update atime:
1647          */
1648         if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1649                 return 1;
1650         /*
1651          * Good, we can skip the atime update:
1652          */
1653         return 0;
1654 }
1655 
1656 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1657 {
1658         int iflags = I_DIRTY_TIME;
1659         bool dirty = false;
1660 
1661         if (flags & S_ATIME)
1662                 inode->i_atime = *time;
1663         if (flags & S_VERSION)
1664                 dirty = inode_maybe_inc_iversion(inode, false);
1665         if (flags & S_CTIME)
1666                 inode->i_ctime = *time;
1667         if (flags & S_MTIME)
1668                 inode->i_mtime = *time;
1669         if ((flags & (S_ATIME | S_CTIME | S_MTIME)) &&
1670             !(inode->i_sb->s_flags & SB_LAZYTIME))
1671                 dirty = true;
1672 
1673         if (dirty)
1674                 iflags |= I_DIRTY_SYNC;
1675         __mark_inode_dirty(inode, iflags);
1676         return 0;
1677 }
1678 EXPORT_SYMBOL(generic_update_time);
1679 
1680 /*
1681  * This does the actual work of updating an inodes time or version.  Must have
1682  * had called mnt_want_write() before calling this.
1683  */
1684 static int update_time(struct inode *inode, struct timespec64 *time, int flags)
1685 {
1686         int (*update_time)(struct inode *, struct timespec64 *, int);
1687 
1688         update_time = inode->i_op->update_time ? inode->i_op->update_time :
1689                 generic_update_time;
1690 
1691         return update_time(inode, time, flags);
1692 }
1693 
1694 /**
1695  *      touch_atime     -       update the access time
1696  *      @path: the &struct path to update
1697  *      @inode: inode to update
1698  *
1699  *      Update the accessed time on an inode and mark it for writeback.
1700  *      This function automatically handles read only file systems and media,
1701  *      as well as the "noatime" flag and inode specific "noatime" markers.
1702  */
1703 bool atime_needs_update(const struct path *path, struct inode *inode)
1704 {
1705         struct vfsmount *mnt = path->mnt;
1706         struct timespec64 now;
1707 
1708         if (inode->i_flags & S_NOATIME)
1709                 return false;
1710 
1711         /* Atime updates will likely cause i_uid and i_gid to be written
1712          * back improprely if their true value is unknown to the vfs.
1713          */
1714         if (HAS_UNMAPPED_ID(inode))
1715                 return false;
1716 
1717         if (IS_NOATIME(inode))
1718                 return false;
1719         if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1720                 return false;
1721 
1722         if (mnt->mnt_flags & MNT_NOATIME)
1723                 return false;
1724         if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1725                 return false;
1726 
1727         now = current_time(inode);
1728 
1729         if (!relatime_need_update(mnt, inode, now))
1730                 return false;
1731 
1732         if (timespec64_equal(&inode->i_atime, &now))
1733                 return false;
1734 
1735         return true;
1736 }
1737 
1738 void touch_atime(const struct path *path)
1739 {
1740         struct vfsmount *mnt = path->mnt;
1741         struct inode *inode = d_inode(path->dentry);
1742         struct timespec64 now;
1743 
1744         if (!atime_needs_update(path, inode))
1745                 return;
1746 
1747         if (!sb_start_write_trylock(inode->i_sb))
1748                 return;
1749 
1750         if (__mnt_want_write(mnt) != 0)
1751                 goto skip_update;
1752         /*
1753          * File systems can error out when updating inodes if they need to
1754          * allocate new space to modify an inode (such is the case for
1755          * Btrfs), but since we touch atime while walking down the path we
1756          * really don't care if we failed to update the atime of the file,
1757          * so just ignore the return value.
1758          * We may also fail on filesystems that have the ability to make parts
1759          * of the fs read only, e.g. subvolumes in Btrfs.
1760          */
1761         now = current_time(inode);
1762         update_time(inode, &now, S_ATIME);
1763         __mnt_drop_write(mnt);
1764 skip_update:
1765         sb_end_write(inode->i_sb);
1766 }
1767 EXPORT_SYMBOL(touch_atime);
1768 
1769 /*
1770  * The logic we want is
1771  *
1772  *      if suid or (sgid and xgrp)
1773  *              remove privs
1774  */
1775 int should_remove_suid(struct dentry *dentry)
1776 {
1777         umode_t mode = d_inode(dentry)->i_mode;
1778         int kill = 0;
1779 
1780         /* suid always must be killed */
1781         if (unlikely(mode & S_ISUID))
1782                 kill = ATTR_KILL_SUID;
1783 
1784         /*
1785          * sgid without any exec bits is just a mandatory locking mark; leave
1786          * it alone.  If some exec bits are set, it's a real sgid; kill it.
1787          */
1788         if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1789                 kill |= ATTR_KILL_SGID;
1790 
1791         if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1792                 return kill;
1793 
1794         return 0;
1795 }
1796 EXPORT_SYMBOL(should_remove_suid);
1797 
1798 /*
1799  * Return mask of changes for notify_change() that need to be done as a
1800  * response to write or truncate. Return 0 if nothing has to be changed.
1801  * Negative value on error (change should be denied).
1802  */
1803 int dentry_needs_remove_privs(struct dentry *dentry)
1804 {
1805         struct inode *inode = d_inode(dentry);
1806         int mask = 0;
1807         int ret;
1808 
1809         if (IS_NOSEC(inode))
1810                 return 0;
1811 
1812         mask = should_remove_suid(dentry);
1813         ret = security_inode_need_killpriv(dentry);
1814         if (ret < 0)
1815                 return ret;
1816         if (ret)
1817                 mask |= ATTR_KILL_PRIV;
1818         return mask;
1819 }
1820 
1821 static int __remove_privs(struct dentry *dentry, int kill)
1822 {
1823         struct iattr newattrs;
1824 
1825         newattrs.ia_valid = ATTR_FORCE | kill;
1826         /*
1827          * Note we call this on write, so notify_change will not
1828          * encounter any conflicting delegations:
1829          */
1830         return notify_change(dentry, &newattrs, NULL);
1831 }
1832 
1833 /*
1834  * Remove special file priviledges (suid, capabilities) when file is written
1835  * to or truncated.
1836  */
1837 int file_remove_privs(struct file *file)
1838 {
1839         struct dentry *dentry = file_dentry(file);
1840         struct inode *inode = file_inode(file);
1841         int kill;
1842         int error = 0;
1843 
1844         /*
1845          * Fast path for nothing security related.
1846          * As well for non-regular files, e.g. blkdev inodes.
1847          * For example, blkdev_write_iter() might get here
1848          * trying to remove privs which it is not allowed to.
1849          */
1850         if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
1851                 return 0;
1852 
1853         kill = dentry_needs_remove_privs(dentry);
1854         if (kill < 0)
1855                 return kill;
1856         if (kill)
1857                 error = __remove_privs(dentry, kill);
1858         if (!error)
1859                 inode_has_no_xattr(inode);
1860 
1861         return error;
1862 }
1863 EXPORT_SYMBOL(file_remove_privs);
1864 
1865 /**
1866  *      file_update_time        -       update mtime and ctime time
1867  *      @file: file accessed
1868  *
1869  *      Update the mtime and ctime members of an inode and mark the inode
1870  *      for writeback.  Note that this function is meant exclusively for
1871  *      usage in the file write path of filesystems, and filesystems may
1872  *      choose to explicitly ignore update via this function with the
1873  *      S_NOCMTIME inode flag, e.g. for network filesystem where these
1874  *      timestamps are handled by the server.  This can return an error for
1875  *      file systems who need to allocate space in order to update an inode.
1876  */
1877 
1878 int file_update_time(struct file *file)
1879 {
1880         struct inode *inode = file_inode(file);
1881         struct timespec64 now;
1882         int sync_it = 0;
1883         int ret;
1884 
1885         /* First try to exhaust all avenues to not sync */
1886         if (IS_NOCMTIME(inode))
1887                 return 0;
1888 
1889         now = current_time(inode);
1890         if (!timespec64_equal(&inode->i_mtime, &now))
1891                 sync_it = S_MTIME;
1892 
1893         if (!timespec64_equal(&inode->i_ctime, &now))
1894                 sync_it |= S_CTIME;
1895 
1896         if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
1897                 sync_it |= S_VERSION;
1898 
1899         if (!sync_it)
1900                 return 0;
1901 
1902         /* Finally allowed to write? Takes lock. */
1903         if (__mnt_want_write_file(file))
1904                 return 0;
1905 
1906         ret = update_time(inode, &now, sync_it);
1907         __mnt_drop_write_file(file);
1908 
1909         return ret;
1910 }
1911 EXPORT_SYMBOL(file_update_time);
1912 
1913 /* Caller must hold the file's inode lock */
1914 int file_modified(struct file *file)
1915 {
1916         int err;
1917 
1918         /*
1919          * Clear the security bits if the process is not being run by root.
1920          * This keeps people from modifying setuid and setgid binaries.
1921          */
1922         err = file_remove_privs(file);
1923         if (err)
1924                 return err;
1925 
1926         if (unlikely(file->f_mode & FMODE_NOCMTIME))
1927                 return 0;
1928 
1929         return file_update_time(file);
1930 }
1931 EXPORT_SYMBOL(file_modified);
1932 
1933 int inode_needs_sync(struct inode *inode)
1934 {
1935         if (IS_SYNC(inode))
1936                 return 1;
1937         if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1938                 return 1;
1939         return 0;
1940 }
1941 EXPORT_SYMBOL(inode_needs_sync);
1942 
1943 /*
1944  * If we try to find an inode in the inode hash while it is being
1945  * deleted, we have to wait until the filesystem completes its
1946  * deletion before reporting that it isn't found.  This function waits
1947  * until the deletion _might_ have completed.  Callers are responsible
1948  * to recheck inode state.
1949  *
1950  * It doesn't matter if I_NEW is not set initially, a call to
1951  * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1952  * will DTRT.
1953  */
1954 static void __wait_on_freeing_inode(struct inode *inode)
1955 {
1956         wait_queue_head_t *wq;
1957         DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1958         wq = bit_waitqueue(&inode->i_state, __I_NEW);
1959         prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
1960         spin_unlock(&inode->i_lock);
1961         spin_unlock(&inode_hash_lock);
1962         schedule();
1963         finish_wait(wq, &wait.wq_entry);
1964         spin_lock(&inode_hash_lock);
1965 }
1966 
1967 static __initdata unsigned long ihash_entries;
1968 static int __init set_ihash_entries(char *str)
1969 {
1970         if (!str)
1971                 return 0;
1972         ihash_entries = simple_strtoul(str, &str, 0);
1973         return 1;
1974 }
1975 __setup("ihash_entries=", set_ihash_entries);
1976 
1977 /*
1978  * Initialize the waitqueues and inode hash table.
1979  */
1980 void __init inode_init_early(void)
1981 {
1982         /* If hashes are distributed across NUMA nodes, defer
1983          * hash allocation until vmalloc space is available.
1984          */
1985         if (hashdist)
1986                 return;
1987 
1988         inode_hashtable =
1989                 alloc_large_system_hash("Inode-cache",
1990                                         sizeof(struct hlist_head),
1991                                         ihash_entries,
1992                                         14,
1993                                         HASH_EARLY | HASH_ZERO,
1994                                         &i_hash_shift,
1995                                         &i_hash_mask,
1996                                         0,
1997                                         0);
1998 }
1999 
2000 void __init inode_init(void)
2001 {
2002         /* inode slab cache */
2003         inode_cachep = kmem_cache_create("inode_cache",
2004                                          sizeof(struct inode),
2005                                          0,
2006                                          (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2007                                          SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2008                                          init_once);
2009 
2010         /* Hash may have been set up in inode_init_early */
2011         if (!hashdist)
2012                 return;
2013 
2014         inode_hashtable =
2015                 alloc_large_system_hash("Inode-cache",
2016                                         sizeof(struct hlist_head),
2017                                         ihash_entries,
2018                                         14,
2019                                         HASH_ZERO,
2020                                         &i_hash_shift,
2021                                         &i_hash_mask,
2022                                         0,
2023                                         0);
2024 }
2025 
2026 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2027 {
2028         inode->i_mode = mode;
2029         if (S_ISCHR(mode)) {
2030                 inode->i_fop = &def_chr_fops;
2031                 inode->i_rdev = rdev;
2032         } else if (S_ISBLK(mode)) {
2033                 inode->i_fop = &def_blk_fops;
2034                 inode->i_rdev = rdev;
2035         } else if (S_ISFIFO(mode))
2036                 inode->i_fop = &pipefifo_fops;
2037         else if (S_ISSOCK(mode))
2038                 ;       /* leave it no_open_fops */
2039         else
2040                 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2041                                   " inode %s:%lu\n", mode, inode->i_sb->s_id,
2042                                   inode->i_ino);
2043 }
2044 EXPORT_SYMBOL(init_special_inode);
2045 
2046 /**
2047  * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2048  * @inode: New inode
2049  * @dir: Directory inode
2050  * @mode: mode of the new inode
2051  */
2052 void inode_init_owner(struct inode *inode, const struct inode *dir,
2053                         umode_t mode)
2054 {
2055         inode->i_uid = current_fsuid();
2056         if (dir && dir->i_mode & S_ISGID) {
2057                 inode->i_gid = dir->i_gid;
2058 
2059                 /* Directories are special, and always inherit S_ISGID */
2060                 if (S_ISDIR(mode))
2061                         mode |= S_ISGID;
2062                 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2063                          !in_group_p(inode->i_gid) &&
2064                          !capable_wrt_inode_uidgid(dir, CAP_FSETID))
2065                         mode &= ~S_ISGID;
2066         } else
2067                 inode->i_gid = current_fsgid();
2068         inode->i_mode = mode;
2069 }
2070 EXPORT_SYMBOL(inode_init_owner);
2071 
2072 /**
2073  * inode_owner_or_capable - check current task permissions to inode
2074  * @inode: inode being checked
2075  *
2076  * Return true if current either has CAP_FOWNER in a namespace with the
2077  * inode owner uid mapped, or owns the file.
2078  */
2079 bool inode_owner_or_capable(const struct inode *inode)
2080 {
2081         struct user_namespace *ns;
2082 
2083         if (uid_eq(current_fsuid(), inode->i_uid))
2084                 return true;
2085 
2086         ns = current_user_ns();
2087         if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
2088                 return true;
2089         return false;
2090 }
2091 EXPORT_SYMBOL(inode_owner_or_capable);
2092 
2093 /*
2094  * Direct i/o helper functions
2095  */
2096 static void __inode_dio_wait(struct inode *inode)
2097 {
2098         wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2099         DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2100 
2101         do {
2102                 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2103                 if (atomic_read(&inode->i_dio_count))
2104                         schedule();
2105         } while (atomic_read(&inode->i_dio_count));
2106         finish_wait(wq, &q.wq_entry);
2107 }
2108 
2109 /**
2110  * inode_dio_wait - wait for outstanding DIO requests to finish
2111  * @inode: inode to wait for
2112  *
2113  * Waits for all pending direct I/O requests to finish so that we can
2114  * proceed with a truncate or equivalent operation.
2115  *
2116  * Must be called under a lock that serializes taking new references
2117  * to i_dio_count, usually by inode->i_mutex.
2118  */
2119 void inode_dio_wait(struct inode *inode)
2120 {
2121         if (atomic_read(&inode->i_dio_count))
2122                 __inode_dio_wait(inode);
2123 }
2124 EXPORT_SYMBOL(inode_dio_wait);
2125 
2126 /*
2127  * inode_set_flags - atomically set some inode flags
2128  *
2129  * Note: the caller should be holding i_mutex, or else be sure that
2130  * they have exclusive access to the inode structure (i.e., while the
2131  * inode is being instantiated).  The reason for the cmpxchg() loop
2132  * --- which wouldn't be necessary if all code paths which modify
2133  * i_flags actually followed this rule, is that there is at least one
2134  * code path which doesn't today so we use cmpxchg() out of an abundance
2135  * of caution.
2136  *
2137  * In the long run, i_mutex is overkill, and we should probably look
2138  * at using the i_lock spinlock to protect i_flags, and then make sure
2139  * it is so documented in include/linux/fs.h and that all code follows
2140  * the locking convention!!
2141  */
2142 void inode_set_flags(struct inode *inode, unsigned int flags,
2143                      unsigned int mask)
2144 {
2145         WARN_ON_ONCE(flags & ~mask);
2146         set_mask_bits(&inode->i_flags, mask, flags);
2147 }
2148 EXPORT_SYMBOL(inode_set_flags);
2149 
2150 void inode_nohighmem(struct inode *inode)
2151 {
2152         mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2153 }
2154 EXPORT_SYMBOL(inode_nohighmem);
2155 
2156 /**
2157  * timespec64_trunc - Truncate timespec64 to a granularity
2158  * @t: Timespec64
2159  * @gran: Granularity in ns.
2160  *
2161  * Truncate a timespec64 to a granularity. Always rounds down. gran must
2162  * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2163  */
2164 struct timespec64 timespec64_trunc(struct timespec64 t, unsigned gran)
2165 {
2166         /* Avoid division in the common cases 1 ns and 1 s. */
2167         if (gran == 1) {
2168                 /* nothing */
2169         } else if (gran == NSEC_PER_SEC) {
2170                 t.tv_nsec = 0;
2171         } else if (gran > 1 && gran < NSEC_PER_SEC) {
2172                 t.tv_nsec -= t.tv_nsec % gran;
2173         } else {
2174                 WARN(1, "illegal file time granularity: %u", gran);
2175         }
2176         return t;
2177 }
2178 EXPORT_SYMBOL(timespec64_trunc);
2179 
2180 /**
2181  * timestamp_truncate - Truncate timespec to a granularity
2182  * @t: Timespec
2183  * @inode: inode being updated
2184  *
2185  * Truncate a timespec to the granularity supported by the fs
2186  * containing the inode. Always rounds down. gran must
2187  * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2188  */
2189 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2190 {
2191         struct super_block *sb = inode->i_sb;
2192         unsigned int gran = sb->s_time_gran;
2193 
2194         t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2195         if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2196                 t.tv_nsec = 0;
2197 
2198         /* Avoid division in the common cases 1 ns and 1 s. */
2199         if (gran == 1)
2200                 ; /* nothing */
2201         else if (gran == NSEC_PER_SEC)
2202                 t.tv_nsec = 0;
2203         else if (gran > 1 && gran < NSEC_PER_SEC)
2204                 t.tv_nsec -= t.tv_nsec % gran;
2205         else
2206                 WARN(1, "invalid file time granularity: %u", gran);
2207         return t;
2208 }
2209 EXPORT_SYMBOL(timestamp_truncate);
2210 
2211 /**
2212  * current_time - Return FS time
2213  * @inode: inode.
2214  *
2215  * Return the current time truncated to the time granularity supported by
2216  * the fs.
2217  *
2218  * Note that inode and inode->sb cannot be NULL.
2219  * Otherwise, the function warns and returns time without truncation.
2220  */
2221 struct timespec64 current_time(struct inode *inode)
2222 {
2223         struct timespec64 now;
2224 
2225         ktime_get_coarse_real_ts64(&now);
2226 
2227         if (unlikely(!inode->i_sb)) {
2228                 WARN(1, "current_time() called with uninitialized super_block in the inode");
2229                 return now;
2230         }
2231 
2232         return timestamp_truncate(now, inode);
2233 }
2234 EXPORT_SYMBOL(current_time);
2235 
2236 /*
2237  * Generic function to check FS_IOC_SETFLAGS values and reject any invalid
2238  * configurations.
2239  *
2240  * Note: the caller should be holding i_mutex, or else be sure that they have
2241  * exclusive access to the inode structure.
2242  */
2243 int vfs_ioc_setflags_prepare(struct inode *inode, unsigned int oldflags,
2244                              unsigned int flags)
2245 {
2246         /*
2247          * The IMMUTABLE and APPEND_ONLY flags can only be changed by
2248          * the relevant capability.
2249          *
2250          * This test looks nicer. Thanks to Pauline Middelink
2251          */
2252         if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL) &&
2253             !capable(CAP_LINUX_IMMUTABLE))
2254                 return -EPERM;
2255 
2256         return 0;
2257 }
2258 EXPORT_SYMBOL(vfs_ioc_setflags_prepare);
2259 
2260 /*
2261  * Generic function to check FS_IOC_FSSETXATTR values and reject any invalid
2262  * configurations.
2263  *
2264  * Note: the caller should be holding i_mutex, or else be sure that they have
2265  * exclusive access to the inode structure.
2266  */
2267 int vfs_ioc_fssetxattr_check(struct inode *inode, const struct fsxattr *old_fa,
2268                              struct fsxattr *fa)
2269 {
2270         /*
2271          * Can't modify an immutable/append-only file unless we have
2272          * appropriate permission.
2273          */
2274         if ((old_fa->fsx_xflags ^ fa->fsx_xflags) &
2275                         (FS_XFLAG_IMMUTABLE | FS_XFLAG_APPEND) &&
2276             !capable(CAP_LINUX_IMMUTABLE))
2277                 return -EPERM;
2278 
2279         /*
2280          * Project Quota ID state is only allowed to change from within the init
2281          * namespace. Enforce that restriction only if we are trying to change
2282          * the quota ID state. Everything else is allowed in user namespaces.
2283          */
2284         if (current_user_ns() != &init_user_ns) {
2285                 if (old_fa->fsx_projid != fa->fsx_projid)
2286                         return -EINVAL;
2287                 if ((old_fa->fsx_xflags ^ fa->fsx_xflags) &
2288                                 FS_XFLAG_PROJINHERIT)
2289                         return -EINVAL;
2290         }
2291 
2292         /* Check extent size hints. */
2293         if ((fa->fsx_xflags & FS_XFLAG_EXTSIZE) && !S_ISREG(inode->i_mode))
2294                 return -EINVAL;
2295 
2296         if ((fa->fsx_xflags & FS_XFLAG_EXTSZINHERIT) &&
2297                         !S_ISDIR(inode->i_mode))
2298                 return -EINVAL;
2299 
2300         if ((fa->fsx_xflags & FS_XFLAG_COWEXTSIZE) &&
2301             !S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode))
2302                 return -EINVAL;
2303 
2304         /*
2305          * It is only valid to set the DAX flag on regular files and
2306          * directories on filesystems.
2307          */
2308         if ((fa->fsx_xflags & FS_XFLAG_DAX) &&
2309             !(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
2310                 return -EINVAL;
2311 
2312         /* Extent size hints of zero turn off the flags. */
2313         if (fa->fsx_extsize == 0)
2314                 fa->fsx_xflags &= ~(FS_XFLAG_EXTSIZE | FS_XFLAG_EXTSZINHERIT);
2315         if (fa->fsx_cowextsize == 0)
2316                 fa->fsx_xflags &= ~FS_XFLAG_COWEXTSIZE;
2317 
2318         return 0;
2319 }
2320 EXPORT_SYMBOL(vfs_ioc_fssetxattr_check);

/* [<][>][^][v][top][bottom][index][help] */