~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/inode.c

Version: ~ [ linux-5.11 ] ~ [ linux-5.10.17 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.99 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.176 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.221 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.257 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.257 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp