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

TOMOYO Linux Cross Reference
Linux/fs/inode.c

Version: ~ [ linux-5.14-rc3 ] ~ [ linux-5.13.5 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.53 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.135 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.198 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.240 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.276 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.276 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ 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 /*
  2  * (C) 1997 Linus Torvalds
  3  * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
  4  */
  5 #include <linux/export.h>
  6 #include <linux/fs.h>
  7 #include <linux/mm.h>
  8 #include <linux/backing-dev.h>
  9 #include <linux/hash.h>
 10 #include <linux/swap.h>
 11 #include <linux/security.h>
 12 #include <linux/cdev.h>
 13 #include <linux/bootmem.h>
 14 #include <linux/fsnotify.h>
 15 #include <linux/mount.h>
 16 #include <linux/posix_acl.h>
 17 #include <linux/prefetch.h>
 18 #include <linux/buffer_head.h> /* for inode_has_buffers */
 19 #include <linux/ratelimit.h>
 20 #include <linux/list_lru.h>
 21 #include "internal.h"
 22 
 23 /*
 24  * Inode locking rules:
 25  *
 26  * inode->i_lock protects:
 27  *   inode->i_state, inode->i_hash, __iget()
 28  * Inode LRU list locks protect:
 29  *   inode->i_sb->s_inode_lru, inode->i_lru
 30  * inode_sb_list_lock protects:
 31  *   sb->s_inodes, inode->i_sb_list
 32  * bdi->wb.list_lock protects:
 33  *   bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
 34  * inode_hash_lock protects:
 35  *   inode_hashtable, inode->i_hash
 36  *
 37  * Lock ordering:
 38  *
 39  * inode_sb_list_lock
 40  *   inode->i_lock
 41  *     Inode LRU list locks
 42  *
 43  * bdi->wb.list_lock
 44  *   inode->i_lock
 45  *
 46  * inode_hash_lock
 47  *   inode_sb_list_lock
 48  *   inode->i_lock
 49  *
 50  * iunique_lock
 51  *   inode_hash_lock
 52  */
 53 
 54 static unsigned int i_hash_mask __read_mostly;
 55 static unsigned int i_hash_shift __read_mostly;
 56 static struct hlist_head *inode_hashtable __read_mostly;
 57 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
 58 
 59 __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
 60 
 61 /*
 62  * Empty aops. Can be used for the cases where the user does not
 63  * define any of the address_space operations.
 64  */
 65 const struct address_space_operations empty_aops = {
 66 };
 67 EXPORT_SYMBOL(empty_aops);
 68 
 69 /*
 70  * Statistics gathering..
 71  */
 72 struct inodes_stat_t inodes_stat;
 73 
 74 static DEFINE_PER_CPU(unsigned long, nr_inodes);
 75 static DEFINE_PER_CPU(unsigned long, nr_unused);
 76 
 77 static struct kmem_cache *inode_cachep __read_mostly;
 78 
 79 static long get_nr_inodes(void)
 80 {
 81         int i;
 82         long sum = 0;
 83         for_each_possible_cpu(i)
 84                 sum += per_cpu(nr_inodes, i);
 85         return sum < 0 ? 0 : sum;
 86 }
 87 
 88 static inline long get_nr_inodes_unused(void)
 89 {
 90         int i;
 91         long sum = 0;
 92         for_each_possible_cpu(i)
 93                 sum += per_cpu(nr_unused, i);
 94         return sum < 0 ? 0 : sum;
 95 }
 96 
 97 long get_nr_dirty_inodes(void)
 98 {
 99         /* not actually dirty inodes, but a wild approximation */
100         long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
101         return nr_dirty > 0 ? nr_dirty : 0;
102 }
103 
104 /*
105  * Handle nr_inode sysctl
106  */
107 #ifdef CONFIG_SYSCTL
108 int proc_nr_inodes(struct ctl_table *table, int write,
109                    void __user *buffer, size_t *lenp, loff_t *ppos)
110 {
111         inodes_stat.nr_inodes = get_nr_inodes();
112         inodes_stat.nr_unused = get_nr_inodes_unused();
113         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
114 }
115 #endif
116 
117 /**
118  * inode_init_always - perform inode structure intialisation
119  * @sb: superblock inode belongs to
120  * @inode: inode to initialise
121  *
122  * These are initializations that need to be done on every inode
123  * allocation as the fields are not initialised by slab allocation.
124  */
125 int inode_init_always(struct super_block *sb, struct inode *inode)
126 {
127         static const struct inode_operations empty_iops;
128         static const struct file_operations empty_fops;
129         struct address_space *const mapping = &inode->i_data;
130 
131         inode->i_sb = sb;
132         inode->i_blkbits = sb->s_blocksize_bits;
133         inode->i_flags = 0;
134         atomic_set(&inode->i_count, 1);
135         inode->i_op = &empty_iops;
136         inode->i_fop = &empty_fops;
137         inode->__i_nlink = 1;
138         inode->i_opflags = 0;
139         i_uid_write(inode, 0);
140         i_gid_write(inode, 0);
141         atomic_set(&inode->i_writecount, 0);
142         inode->i_size = 0;
143         inode->i_blocks = 0;
144         inode->i_bytes = 0;
145         inode->i_generation = 0;
146 #ifdef CONFIG_QUOTA
147         memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
148 #endif
149         inode->i_pipe = NULL;
150         inode->i_bdev = NULL;
151         inode->i_cdev = NULL;
152         inode->i_rdev = 0;
153         inode->dirtied_when = 0;
154 
155         if (security_inode_alloc(inode))
156                 goto out;
157         spin_lock_init(&inode->i_lock);
158         lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
159 
160         mutex_init(&inode->i_mutex);
161         lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
162 
163         atomic_set(&inode->i_dio_count, 0);
164 
165         mapping->a_ops = &empty_aops;
166         mapping->host = inode;
167         mapping->flags = 0;
168         atomic_set(&mapping->i_mmap_writable, 0);
169         mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
170         mapping->private_data = NULL;
171         mapping->backing_dev_info = &default_backing_dev_info;
172         mapping->writeback_index = 0;
173 
174         /*
175          * If the block_device provides a backing_dev_info for client
176          * inodes then use that.  Otherwise the inode share the bdev's
177          * backing_dev_info.
178          */
179         if (sb->s_bdev) {
180                 struct backing_dev_info *bdi;
181 
182                 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
183                 mapping->backing_dev_info = bdi;
184         }
185         inode->i_private = NULL;
186         inode->i_mapping = mapping;
187         INIT_HLIST_HEAD(&inode->i_dentry);      /* buggered by rcu freeing */
188 #ifdef CONFIG_FS_POSIX_ACL
189         inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
190 #endif
191 
192 #ifdef CONFIG_FSNOTIFY
193         inode->i_fsnotify_mask = 0;
194 #endif
195 
196         this_cpu_inc(nr_inodes);
197 
198         return 0;
199 out:
200         return -ENOMEM;
201 }
202 EXPORT_SYMBOL(inode_init_always);
203 
204 static struct inode *alloc_inode(struct super_block *sb)
205 {
206         struct inode *inode;
207 
208         if (sb->s_op->alloc_inode)
209                 inode = sb->s_op->alloc_inode(sb);
210         else
211                 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
212 
213         if (!inode)
214                 return NULL;
215 
216         if (unlikely(inode_init_always(sb, inode))) {
217                 if (inode->i_sb->s_op->destroy_inode)
218                         inode->i_sb->s_op->destroy_inode(inode);
219                 else
220                         kmem_cache_free(inode_cachep, inode);
221                 return NULL;
222         }
223 
224         return inode;
225 }
226 
227 void free_inode_nonrcu(struct inode *inode)
228 {
229         kmem_cache_free(inode_cachep, inode);
230 }
231 EXPORT_SYMBOL(free_inode_nonrcu);
232 
233 void __destroy_inode(struct inode *inode)
234 {
235         BUG_ON(inode_has_buffers(inode));
236         security_inode_free(inode);
237         fsnotify_inode_delete(inode);
238         if (!inode->i_nlink) {
239                 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
240                 atomic_long_dec(&inode->i_sb->s_remove_count);
241         }
242 
243 #ifdef CONFIG_FS_POSIX_ACL
244         if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
245                 posix_acl_release(inode->i_acl);
246         if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
247                 posix_acl_release(inode->i_default_acl);
248 #endif
249         this_cpu_dec(nr_inodes);
250 }
251 EXPORT_SYMBOL(__destroy_inode);
252 
253 static void i_callback(struct rcu_head *head)
254 {
255         struct inode *inode = container_of(head, struct inode, i_rcu);
256         kmem_cache_free(inode_cachep, inode);
257 }
258 
259 static void destroy_inode(struct inode *inode)
260 {
261         BUG_ON(!list_empty(&inode->i_lru));
262         __destroy_inode(inode);
263         if (inode->i_sb->s_op->destroy_inode)
264                 inode->i_sb->s_op->destroy_inode(inode);
265         else
266                 call_rcu(&inode->i_rcu, i_callback);
267 }
268 
269 /**
270  * drop_nlink - directly drop an inode's link count
271  * @inode: inode
272  *
273  * This is a low-level filesystem helper to replace any
274  * direct filesystem manipulation of i_nlink.  In cases
275  * where we are attempting to track writes to the
276  * filesystem, a decrement to zero means an imminent
277  * write when the file is truncated and actually unlinked
278  * on the filesystem.
279  */
280 void drop_nlink(struct inode *inode)
281 {
282         WARN_ON(inode->i_nlink == 0);
283         inode->__i_nlink--;
284         if (!inode->i_nlink)
285                 atomic_long_inc(&inode->i_sb->s_remove_count);
286 }
287 EXPORT_SYMBOL(drop_nlink);
288 
289 /**
290  * clear_nlink - directly zero 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.  See
295  * drop_nlink() for why we care about i_nlink hitting zero.
296  */
297 void clear_nlink(struct inode *inode)
298 {
299         if (inode->i_nlink) {
300                 inode->__i_nlink = 0;
301                 atomic_long_inc(&inode->i_sb->s_remove_count);
302         }
303 }
304 EXPORT_SYMBOL(clear_nlink);
305 
306 /**
307  * set_nlink - directly set an inode's link count
308  * @inode: inode
309  * @nlink: new nlink (should be non-zero)
310  *
311  * This is a low-level filesystem helper to replace any
312  * direct filesystem manipulation of i_nlink.
313  */
314 void set_nlink(struct inode *inode, unsigned int nlink)
315 {
316         if (!nlink) {
317                 clear_nlink(inode);
318         } else {
319                 /* Yes, some filesystems do change nlink from zero to one */
320                 if (inode->i_nlink == 0)
321                         atomic_long_dec(&inode->i_sb->s_remove_count);
322 
323                 inode->__i_nlink = nlink;
324         }
325 }
326 EXPORT_SYMBOL(set_nlink);
327 
328 /**
329  * inc_nlink - directly increment an inode's link count
330  * @inode: inode
331  *
332  * This is a low-level filesystem helper to replace any
333  * direct filesystem manipulation of i_nlink.  Currently,
334  * it is only here for parity with dec_nlink().
335  */
336 void inc_nlink(struct inode *inode)
337 {
338         if (unlikely(inode->i_nlink == 0)) {
339                 WARN_ON(!(inode->i_state & I_LINKABLE));
340                 atomic_long_dec(&inode->i_sb->s_remove_count);
341         }
342 
343         inode->__i_nlink++;
344 }
345 EXPORT_SYMBOL(inc_nlink);
346 
347 void address_space_init_once(struct address_space *mapping)
348 {
349         memset(mapping, 0, sizeof(*mapping));
350         INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
351         spin_lock_init(&mapping->tree_lock);
352         mutex_init(&mapping->i_mmap_mutex);
353         INIT_LIST_HEAD(&mapping->private_list);
354         spin_lock_init(&mapping->private_lock);
355         mapping->i_mmap = RB_ROOT;
356         INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
357 }
358 EXPORT_SYMBOL(address_space_init_once);
359 
360 /*
361  * These are initializations that only need to be done
362  * once, because the fields are idempotent across use
363  * of the inode, so let the slab aware of that.
364  */
365 void inode_init_once(struct inode *inode)
366 {
367         memset(inode, 0, sizeof(*inode));
368         INIT_HLIST_NODE(&inode->i_hash);
369         INIT_LIST_HEAD(&inode->i_devices);
370         INIT_LIST_HEAD(&inode->i_wb_list);
371         INIT_LIST_HEAD(&inode->i_lru);
372         address_space_init_once(&inode->i_data);
373         i_size_ordered_init(inode);
374 #ifdef CONFIG_FSNOTIFY
375         INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
376 #endif
377 }
378 EXPORT_SYMBOL(inode_init_once);
379 
380 static void init_once(void *foo)
381 {
382         struct inode *inode = (struct inode *) foo;
383 
384         inode_init_once(inode);
385 }
386 
387 /*
388  * inode->i_lock must be held
389  */
390 void __iget(struct inode *inode)
391 {
392         atomic_inc(&inode->i_count);
393 }
394 
395 /*
396  * get additional reference to inode; caller must already hold one.
397  */
398 void ihold(struct inode *inode)
399 {
400         WARN_ON(atomic_inc_return(&inode->i_count) < 2);
401 }
402 EXPORT_SYMBOL(ihold);
403 
404 static void inode_lru_list_add(struct inode *inode)
405 {
406         if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
407                 this_cpu_inc(nr_unused);
408 }
409 
410 /*
411  * Add inode to LRU if needed (inode is unused and clean).
412  *
413  * Needs inode->i_lock held.
414  */
415 void inode_add_lru(struct inode *inode)
416 {
417         if (!(inode->i_state & (I_DIRTY | I_SYNC | I_FREEING | I_WILL_FREE)) &&
418             !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
419                 inode_lru_list_add(inode);
420 }
421 
422 
423 static void inode_lru_list_del(struct inode *inode)
424 {
425 
426         if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
427                 this_cpu_dec(nr_unused);
428 }
429 
430 /**
431  * inode_sb_list_add - add inode to the superblock list of inodes
432  * @inode: inode to add
433  */
434 void inode_sb_list_add(struct inode *inode)
435 {
436         spin_lock(&inode_sb_list_lock);
437         list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
438         spin_unlock(&inode_sb_list_lock);
439 }
440 EXPORT_SYMBOL_GPL(inode_sb_list_add);
441 
442 static inline void inode_sb_list_del(struct inode *inode)
443 {
444         if (!list_empty(&inode->i_sb_list)) {
445                 spin_lock(&inode_sb_list_lock);
446                 list_del_init(&inode->i_sb_list);
447                 spin_unlock(&inode_sb_list_lock);
448         }
449 }
450 
451 static unsigned long hash(struct super_block *sb, unsigned long hashval)
452 {
453         unsigned long tmp;
454 
455         tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
456                         L1_CACHE_BYTES;
457         tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
458         return tmp & i_hash_mask;
459 }
460 
461 /**
462  *      __insert_inode_hash - hash an inode
463  *      @inode: unhashed inode
464  *      @hashval: unsigned long value used to locate this object in the
465  *              inode_hashtable.
466  *
467  *      Add an inode to the inode hash for this superblock.
468  */
469 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
470 {
471         struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
472 
473         spin_lock(&inode_hash_lock);
474         spin_lock(&inode->i_lock);
475         hlist_add_head(&inode->i_hash, b);
476         spin_unlock(&inode->i_lock);
477         spin_unlock(&inode_hash_lock);
478 }
479 EXPORT_SYMBOL(__insert_inode_hash);
480 
481 /**
482  *      __remove_inode_hash - remove an inode from the hash
483  *      @inode: inode to unhash
484  *
485  *      Remove an inode from the superblock.
486  */
487 void __remove_inode_hash(struct inode *inode)
488 {
489         spin_lock(&inode_hash_lock);
490         spin_lock(&inode->i_lock);
491         hlist_del_init(&inode->i_hash);
492         spin_unlock(&inode->i_lock);
493         spin_unlock(&inode_hash_lock);
494 }
495 EXPORT_SYMBOL(__remove_inode_hash);
496 
497 void clear_inode(struct inode *inode)
498 {
499         might_sleep();
500         /*
501          * We have to cycle tree_lock here because reclaim can be still in the
502          * process of removing the last page (in __delete_from_page_cache())
503          * and we must not free mapping under it.
504          */
505         spin_lock_irq(&inode->i_data.tree_lock);
506         BUG_ON(inode->i_data.nrpages);
507         BUG_ON(inode->i_data.nrshadows);
508         spin_unlock_irq(&inode->i_data.tree_lock);
509         BUG_ON(!list_empty(&inode->i_data.private_list));
510         BUG_ON(!(inode->i_state & I_FREEING));
511         BUG_ON(inode->i_state & I_CLEAR);
512         /* don't need i_lock here, no concurrent mods to i_state */
513         inode->i_state = I_FREEING | I_CLEAR;
514 }
515 EXPORT_SYMBOL(clear_inode);
516 
517 /*
518  * Free the inode passed in, removing it from the lists it is still connected
519  * to. We remove any pages still attached to the inode and wait for any IO that
520  * is still in progress before finally destroying the inode.
521  *
522  * An inode must already be marked I_FREEING so that we avoid the inode being
523  * moved back onto lists if we race with other code that manipulates the lists
524  * (e.g. writeback_single_inode). The caller is responsible for setting this.
525  *
526  * An inode must already be removed from the LRU list before being evicted from
527  * the cache. This should occur atomically with setting the I_FREEING state
528  * flag, so no inodes here should ever be on the LRU when being evicted.
529  */
530 static void evict(struct inode *inode)
531 {
532         const struct super_operations *op = inode->i_sb->s_op;
533 
534         BUG_ON(!(inode->i_state & I_FREEING));
535         BUG_ON(!list_empty(&inode->i_lru));
536 
537         if (!list_empty(&inode->i_wb_list))
538                 inode_wb_list_del(inode);
539 
540         inode_sb_list_del(inode);
541 
542         /*
543          * Wait for flusher thread to be done with the inode so that filesystem
544          * does not start destroying it while writeback is still running. Since
545          * the inode has I_FREEING set, flusher thread won't start new work on
546          * the inode.  We just have to wait for running writeback to finish.
547          */
548         inode_wait_for_writeback(inode);
549 
550         if (op->evict_inode) {
551                 op->evict_inode(inode);
552         } else {
553                 truncate_inode_pages_final(&inode->i_data);
554                 clear_inode(inode);
555         }
556         if (S_ISBLK(inode->i_mode) && inode->i_bdev)
557                 bd_forget(inode);
558         if (S_ISCHR(inode->i_mode) && inode->i_cdev)
559                 cd_forget(inode);
560 
561         remove_inode_hash(inode);
562 
563         spin_lock(&inode->i_lock);
564         wake_up_bit(&inode->i_state, __I_NEW);
565         BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
566         spin_unlock(&inode->i_lock);
567 
568         destroy_inode(inode);
569 }
570 
571 /*
572  * dispose_list - dispose of the contents of a local list
573  * @head: the head of the list to free
574  *
575  * Dispose-list gets a local list with local inodes in it, so it doesn't
576  * need to worry about list corruption and SMP locks.
577  */
578 static void dispose_list(struct list_head *head)
579 {
580         while (!list_empty(head)) {
581                 struct inode *inode;
582 
583                 inode = list_first_entry(head, struct inode, i_lru);
584                 list_del_init(&inode->i_lru);
585 
586                 evict(inode);
587         }
588 }
589 
590 /**
591  * evict_inodes - evict all evictable inodes for a superblock
592  * @sb:         superblock to operate on
593  *
594  * Make sure that no inodes with zero refcount are retained.  This is
595  * called by superblock shutdown after having MS_ACTIVE flag removed,
596  * so any inode reaching zero refcount during or after that call will
597  * be immediately evicted.
598  */
599 void evict_inodes(struct super_block *sb)
600 {
601         struct inode *inode, *next;
602         LIST_HEAD(dispose);
603 
604         spin_lock(&inode_sb_list_lock);
605         list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
606                 if (atomic_read(&inode->i_count))
607                         continue;
608 
609                 spin_lock(&inode->i_lock);
610                 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
611                         spin_unlock(&inode->i_lock);
612                         continue;
613                 }
614 
615                 inode->i_state |= I_FREEING;
616                 inode_lru_list_del(inode);
617                 spin_unlock(&inode->i_lock);
618                 list_add(&inode->i_lru, &dispose);
619         }
620         spin_unlock(&inode_sb_list_lock);
621 
622         dispose_list(&dispose);
623 }
624 
625 /**
626  * invalidate_inodes    - attempt to free all inodes on a superblock
627  * @sb:         superblock to operate on
628  * @kill_dirty: flag to guide handling of dirty inodes
629  *
630  * Attempts to free all inodes for a given superblock.  If there were any
631  * busy inodes return a non-zero value, else zero.
632  * If @kill_dirty is set, discard dirty inodes too, otherwise treat
633  * them as busy.
634  */
635 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
636 {
637         int busy = 0;
638         struct inode *inode, *next;
639         LIST_HEAD(dispose);
640 
641         spin_lock(&inode_sb_list_lock);
642         list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
643                 spin_lock(&inode->i_lock);
644                 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
645                         spin_unlock(&inode->i_lock);
646                         continue;
647                 }
648                 if (inode->i_state & I_DIRTY && !kill_dirty) {
649                         spin_unlock(&inode->i_lock);
650                         busy = 1;
651                         continue;
652                 }
653                 if (atomic_read(&inode->i_count)) {
654                         spin_unlock(&inode->i_lock);
655                         busy = 1;
656                         continue;
657                 }
658 
659                 inode->i_state |= I_FREEING;
660                 inode_lru_list_del(inode);
661                 spin_unlock(&inode->i_lock);
662                 list_add(&inode->i_lru, &dispose);
663         }
664         spin_unlock(&inode_sb_list_lock);
665 
666         dispose_list(&dispose);
667 
668         return busy;
669 }
670 
671 /*
672  * Isolate the inode from the LRU in preparation for freeing it.
673  *
674  * Any inodes which are pinned purely because of attached pagecache have their
675  * pagecache removed.  If the inode has metadata buffers attached to
676  * mapping->private_list then try to remove them.
677  *
678  * If the inode has the I_REFERENCED flag set, then it means that it has been
679  * used recently - the flag is set in iput_final(). When we encounter such an
680  * inode, clear the flag and move it to the back of the LRU so it gets another
681  * pass through the LRU before it gets reclaimed. This is necessary because of
682  * the fact we are doing lazy LRU updates to minimise lock contention so the
683  * LRU does not have strict ordering. Hence we don't want to reclaim inodes
684  * with this flag set because they are the inodes that are out of order.
685  */
686 static enum lru_status
687 inode_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
688 {
689         struct list_head *freeable = arg;
690         struct inode    *inode = container_of(item, struct inode, i_lru);
691 
692         /*
693          * we are inverting the lru lock/inode->i_lock here, so use a trylock.
694          * If we fail to get the lock, just skip it.
695          */
696         if (!spin_trylock(&inode->i_lock))
697                 return LRU_SKIP;
698 
699         /*
700          * Referenced or dirty inodes are still in use. Give them another pass
701          * through the LRU as we canot reclaim them now.
702          */
703         if (atomic_read(&inode->i_count) ||
704             (inode->i_state & ~I_REFERENCED)) {
705                 list_del_init(&inode->i_lru);
706                 spin_unlock(&inode->i_lock);
707                 this_cpu_dec(nr_unused);
708                 return LRU_REMOVED;
709         }
710 
711         /* recently referenced inodes get one more pass */
712         if (inode->i_state & I_REFERENCED) {
713                 inode->i_state &= ~I_REFERENCED;
714                 spin_unlock(&inode->i_lock);
715                 return LRU_ROTATE;
716         }
717 
718         if (inode_has_buffers(inode) || inode->i_data.nrpages) {
719                 __iget(inode);
720                 spin_unlock(&inode->i_lock);
721                 spin_unlock(lru_lock);
722                 if (remove_inode_buffers(inode)) {
723                         unsigned long reap;
724                         reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
725                         if (current_is_kswapd())
726                                 __count_vm_events(KSWAPD_INODESTEAL, reap);
727                         else
728                                 __count_vm_events(PGINODESTEAL, reap);
729                         if (current->reclaim_state)
730                                 current->reclaim_state->reclaimed_slab += reap;
731                 }
732                 iput(inode);
733                 spin_lock(lru_lock);
734                 return LRU_RETRY;
735         }
736 
737         WARN_ON(inode->i_state & I_NEW);
738         inode->i_state |= I_FREEING;
739         list_move(&inode->i_lru, freeable);
740         spin_unlock(&inode->i_lock);
741 
742         this_cpu_dec(nr_unused);
743         return LRU_REMOVED;
744 }
745 
746 /*
747  * Walk the superblock inode LRU for freeable inodes and attempt to free them.
748  * This is called from the superblock shrinker function with a number of inodes
749  * to trim from the LRU. Inodes to be freed are moved to a temporary list and
750  * then are freed outside inode_lock by dispose_list().
751  */
752 long prune_icache_sb(struct super_block *sb, unsigned long nr_to_scan,
753                      int nid)
754 {
755         LIST_HEAD(freeable);
756         long freed;
757 
758         freed = list_lru_walk_node(&sb->s_inode_lru, nid, inode_lru_isolate,
759                                        &freeable, &nr_to_scan);
760         dispose_list(&freeable);
761         return freed;
762 }
763 
764 static void __wait_on_freeing_inode(struct inode *inode);
765 /*
766  * Called with the inode lock held.
767  */
768 static struct inode *find_inode(struct super_block *sb,
769                                 struct hlist_head *head,
770                                 int (*test)(struct inode *, void *),
771                                 void *data)
772 {
773         struct inode *inode = NULL;
774 
775 repeat:
776         hlist_for_each_entry(inode, head, i_hash) {
777                 if (inode->i_sb != sb)
778                         continue;
779                 if (!test(inode, data))
780                         continue;
781                 spin_lock(&inode->i_lock);
782                 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
783                         __wait_on_freeing_inode(inode);
784                         goto repeat;
785                 }
786                 __iget(inode);
787                 spin_unlock(&inode->i_lock);
788                 return inode;
789         }
790         return NULL;
791 }
792 
793 /*
794  * find_inode_fast is the fast path version of find_inode, see the comment at
795  * iget_locked for details.
796  */
797 static struct inode *find_inode_fast(struct super_block *sb,
798                                 struct hlist_head *head, unsigned long ino)
799 {
800         struct inode *inode = NULL;
801 
802 repeat:
803         hlist_for_each_entry(inode, head, i_hash) {
804                 if (inode->i_ino != ino)
805                         continue;
806                 if (inode->i_sb != sb)
807                         continue;
808                 spin_lock(&inode->i_lock);
809                 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
810                         __wait_on_freeing_inode(inode);
811                         goto repeat;
812                 }
813                 __iget(inode);
814                 spin_unlock(&inode->i_lock);
815                 return inode;
816         }
817         return NULL;
818 }
819 
820 /*
821  * Each cpu owns a range of LAST_INO_BATCH numbers.
822  * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
823  * to renew the exhausted range.
824  *
825  * This does not significantly increase overflow rate because every CPU can
826  * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
827  * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
828  * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
829  * overflow rate by 2x, which does not seem too significant.
830  *
831  * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
832  * error if st_ino won't fit in target struct field. Use 32bit counter
833  * here to attempt to avoid that.
834  */
835 #define LAST_INO_BATCH 1024
836 static DEFINE_PER_CPU(unsigned int, last_ino);
837 
838 unsigned int get_next_ino(void)
839 {
840         unsigned int *p = &get_cpu_var(last_ino);
841         unsigned int res = *p;
842 
843 #ifdef CONFIG_SMP
844         if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
845                 static atomic_t shared_last_ino;
846                 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
847 
848                 res = next - LAST_INO_BATCH;
849         }
850 #endif
851 
852         *p = ++res;
853         put_cpu_var(last_ino);
854         return res;
855 }
856 EXPORT_SYMBOL(get_next_ino);
857 
858 /**
859  *      new_inode_pseudo        - obtain an inode
860  *      @sb: superblock
861  *
862  *      Allocates a new inode for given superblock.
863  *      Inode wont be chained in superblock s_inodes list
864  *      This means :
865  *      - fs can't be unmount
866  *      - quotas, fsnotify, writeback can't work
867  */
868 struct inode *new_inode_pseudo(struct super_block *sb)
869 {
870         struct inode *inode = alloc_inode(sb);
871 
872         if (inode) {
873                 spin_lock(&inode->i_lock);
874                 inode->i_state = 0;
875                 spin_unlock(&inode->i_lock);
876                 INIT_LIST_HEAD(&inode->i_sb_list);
877         }
878         return inode;
879 }
880 
881 /**
882  *      new_inode       - obtain an inode
883  *      @sb: superblock
884  *
885  *      Allocates a new inode for given superblock. The default gfp_mask
886  *      for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
887  *      If HIGHMEM pages are unsuitable or it is known that pages allocated
888  *      for the page cache are not reclaimable or migratable,
889  *      mapping_set_gfp_mask() must be called with suitable flags on the
890  *      newly created inode's mapping
891  *
892  */
893 struct inode *new_inode(struct super_block *sb)
894 {
895         struct inode *inode;
896 
897         spin_lock_prefetch(&inode_sb_list_lock);
898 
899         inode = new_inode_pseudo(sb);
900         if (inode)
901                 inode_sb_list_add(inode);
902         return inode;
903 }
904 EXPORT_SYMBOL(new_inode);
905 
906 #ifdef CONFIG_DEBUG_LOCK_ALLOC
907 void lockdep_annotate_inode_mutex_key(struct inode *inode)
908 {
909         if (S_ISDIR(inode->i_mode)) {
910                 struct file_system_type *type = inode->i_sb->s_type;
911 
912                 /* Set new key only if filesystem hasn't already changed it */
913                 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
914                         /*
915                          * ensure nobody is actually holding i_mutex
916                          */
917                         mutex_destroy(&inode->i_mutex);
918                         mutex_init(&inode->i_mutex);
919                         lockdep_set_class(&inode->i_mutex,
920                                           &type->i_mutex_dir_key);
921                 }
922         }
923 }
924 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
925 #endif
926 
927 /**
928  * unlock_new_inode - clear the I_NEW state and wake up any waiters
929  * @inode:      new inode to unlock
930  *
931  * Called when the inode is fully initialised to clear the new state of the
932  * inode and wake up anyone waiting for the inode to finish initialisation.
933  */
934 void unlock_new_inode(struct inode *inode)
935 {
936         lockdep_annotate_inode_mutex_key(inode);
937         spin_lock(&inode->i_lock);
938         WARN_ON(!(inode->i_state & I_NEW));
939         inode->i_state &= ~I_NEW;
940         smp_mb();
941         wake_up_bit(&inode->i_state, __I_NEW);
942         spin_unlock(&inode->i_lock);
943 }
944 EXPORT_SYMBOL(unlock_new_inode);
945 
946 /**
947  * lock_two_nondirectories - take two i_mutexes on non-directory objects
948  *
949  * Lock any non-NULL argument that is not a directory.
950  * Zero, one or two objects may be locked by this function.
951  *
952  * @inode1: first inode to lock
953  * @inode2: second inode to lock
954  */
955 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
956 {
957         if (inode1 > inode2)
958                 swap(inode1, inode2);
959 
960         if (inode1 && !S_ISDIR(inode1->i_mode))
961                 mutex_lock(&inode1->i_mutex);
962         if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
963                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_NONDIR2);
964 }
965 EXPORT_SYMBOL(lock_two_nondirectories);
966 
967 /**
968  * unlock_two_nondirectories - release locks from lock_two_nondirectories()
969  * @inode1: first inode to unlock
970  * @inode2: second inode to unlock
971  */
972 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
973 {
974         if (inode1 && !S_ISDIR(inode1->i_mode))
975                 mutex_unlock(&inode1->i_mutex);
976         if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
977                 mutex_unlock(&inode2->i_mutex);
978 }
979 EXPORT_SYMBOL(unlock_two_nondirectories);
980 
981 /**
982  * iget5_locked - obtain an inode from a mounted file system
983  * @sb:         super block of file system
984  * @hashval:    hash value (usually inode number) to get
985  * @test:       callback used for comparisons between inodes
986  * @set:        callback used to initialize a new struct inode
987  * @data:       opaque data pointer to pass to @test and @set
988  *
989  * Search for the inode specified by @hashval and @data in the inode cache,
990  * and if present it is return it with an increased reference count. This is
991  * a generalized version of iget_locked() for file systems where the inode
992  * number is not sufficient for unique identification of an inode.
993  *
994  * If the inode is not in cache, allocate a new inode and return it locked,
995  * hashed, and with the I_NEW flag set. The file system gets to fill it in
996  * before unlocking it via unlock_new_inode().
997  *
998  * Note both @test and @set are called with the inode_hash_lock held, so can't
999  * sleep.
1000  */
1001 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1002                 int (*test)(struct inode *, void *),
1003                 int (*set)(struct inode *, void *), void *data)
1004 {
1005         struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1006         struct inode *inode;
1007 
1008         spin_lock(&inode_hash_lock);
1009         inode = find_inode(sb, head, test, data);
1010         spin_unlock(&inode_hash_lock);
1011 
1012         if (inode) {
1013                 wait_on_inode(inode);
1014                 return inode;
1015         }
1016 
1017         inode = alloc_inode(sb);
1018         if (inode) {
1019                 struct inode *old;
1020 
1021                 spin_lock(&inode_hash_lock);
1022                 /* We released the lock, so.. */
1023                 old = find_inode(sb, head, test, data);
1024                 if (!old) {
1025                         if (set(inode, data))
1026                                 goto set_failed;
1027 
1028                         spin_lock(&inode->i_lock);
1029                         inode->i_state = I_NEW;
1030                         hlist_add_head(&inode->i_hash, head);
1031                         spin_unlock(&inode->i_lock);
1032                         inode_sb_list_add(inode);
1033                         spin_unlock(&inode_hash_lock);
1034 
1035                         /* Return the locked inode with I_NEW set, the
1036                          * caller is responsible for filling in the contents
1037                          */
1038                         return inode;
1039                 }
1040 
1041                 /*
1042                  * Uhhuh, somebody else created the same inode under
1043                  * us. Use the old inode instead of the one we just
1044                  * allocated.
1045                  */
1046                 spin_unlock(&inode_hash_lock);
1047                 destroy_inode(inode);
1048                 inode = old;
1049                 wait_on_inode(inode);
1050         }
1051         return inode;
1052 
1053 set_failed:
1054         spin_unlock(&inode_hash_lock);
1055         destroy_inode(inode);
1056         return NULL;
1057 }
1058 EXPORT_SYMBOL(iget5_locked);
1059 
1060 /**
1061  * iget_locked - obtain an inode from a mounted file system
1062  * @sb:         super block of file system
1063  * @ino:        inode number to get
1064  *
1065  * Search for the inode specified by @ino in the inode cache and if present
1066  * return it with an increased reference count. This is for file systems
1067  * where the inode number is sufficient for unique identification of an inode.
1068  *
1069  * If the inode is not in cache, allocate a new inode and return it locked,
1070  * hashed, and with the I_NEW flag set.  The file system gets to fill it in
1071  * before unlocking it via unlock_new_inode().
1072  */
1073 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1074 {
1075         struct hlist_head *head = inode_hashtable + hash(sb, ino);
1076         struct inode *inode;
1077 
1078         spin_lock(&inode_hash_lock);
1079         inode = find_inode_fast(sb, head, ino);
1080         spin_unlock(&inode_hash_lock);
1081         if (inode) {
1082                 wait_on_inode(inode);
1083                 return inode;
1084         }
1085 
1086         inode = alloc_inode(sb);
1087         if (inode) {
1088                 struct inode *old;
1089 
1090                 spin_lock(&inode_hash_lock);
1091                 /* We released the lock, so.. */
1092                 old = find_inode_fast(sb, head, ino);
1093                 if (!old) {
1094                         inode->i_ino = ino;
1095                         spin_lock(&inode->i_lock);
1096                         inode->i_state = I_NEW;
1097                         hlist_add_head(&inode->i_hash, head);
1098                         spin_unlock(&inode->i_lock);
1099                         inode_sb_list_add(inode);
1100                         spin_unlock(&inode_hash_lock);
1101 
1102                         /* Return the locked inode with I_NEW set, the
1103                          * caller is responsible for filling in the contents
1104                          */
1105                         return inode;
1106                 }
1107 
1108                 /*
1109                  * Uhhuh, somebody else created the same inode under
1110                  * us. Use the old inode instead of the one we just
1111                  * allocated.
1112                  */
1113                 spin_unlock(&inode_hash_lock);
1114                 destroy_inode(inode);
1115                 inode = old;
1116                 wait_on_inode(inode);
1117         }
1118         return inode;
1119 }
1120 EXPORT_SYMBOL(iget_locked);
1121 
1122 /*
1123  * search the inode cache for a matching inode number.
1124  * If we find one, then the inode number we are trying to
1125  * allocate is not unique and so we should not use it.
1126  *
1127  * Returns 1 if the inode number is unique, 0 if it is not.
1128  */
1129 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1130 {
1131         struct hlist_head *b = inode_hashtable + hash(sb, ino);
1132         struct inode *inode;
1133 
1134         spin_lock(&inode_hash_lock);
1135         hlist_for_each_entry(inode, b, i_hash) {
1136                 if (inode->i_ino == ino && inode->i_sb == sb) {
1137                         spin_unlock(&inode_hash_lock);
1138                         return 0;
1139                 }
1140         }
1141         spin_unlock(&inode_hash_lock);
1142 
1143         return 1;
1144 }
1145 
1146 /**
1147  *      iunique - get a unique inode number
1148  *      @sb: superblock
1149  *      @max_reserved: highest reserved inode number
1150  *
1151  *      Obtain an inode number that is unique on the system for a given
1152  *      superblock. This is used by file systems that have no natural
1153  *      permanent inode numbering system. An inode number is returned that
1154  *      is higher than the reserved limit but unique.
1155  *
1156  *      BUGS:
1157  *      With a large number of inodes live on the file system this function
1158  *      currently becomes quite slow.
1159  */
1160 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1161 {
1162         /*
1163          * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1164          * error if st_ino won't fit in target struct field. Use 32bit counter
1165          * here to attempt to avoid that.
1166          */
1167         static DEFINE_SPINLOCK(iunique_lock);
1168         static unsigned int counter;
1169         ino_t res;
1170 
1171         spin_lock(&iunique_lock);
1172         do {
1173                 if (counter <= max_reserved)
1174                         counter = max_reserved + 1;
1175                 res = counter++;
1176         } while (!test_inode_iunique(sb, res));
1177         spin_unlock(&iunique_lock);
1178 
1179         return res;
1180 }
1181 EXPORT_SYMBOL(iunique);
1182 
1183 struct inode *igrab(struct inode *inode)
1184 {
1185         spin_lock(&inode->i_lock);
1186         if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1187                 __iget(inode);
1188                 spin_unlock(&inode->i_lock);
1189         } else {
1190                 spin_unlock(&inode->i_lock);
1191                 /*
1192                  * Handle the case where s_op->clear_inode is not been
1193                  * called yet, and somebody is calling igrab
1194                  * while the inode is getting freed.
1195                  */
1196                 inode = NULL;
1197         }
1198         return inode;
1199 }
1200 EXPORT_SYMBOL(igrab);
1201 
1202 /**
1203  * ilookup5_nowait - search for an inode in the inode cache
1204  * @sb:         super block of file system to search
1205  * @hashval:    hash value (usually inode number) to search for
1206  * @test:       callback used for comparisons between inodes
1207  * @data:       opaque data pointer to pass to @test
1208  *
1209  * Search for the inode specified by @hashval and @data in the inode cache.
1210  * If the inode is in the cache, the inode is returned with an incremented
1211  * reference count.
1212  *
1213  * Note: I_NEW is not waited upon so you have to be very careful what you do
1214  * with the returned inode.  You probably should be using ilookup5() instead.
1215  *
1216  * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1217  */
1218 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1219                 int (*test)(struct inode *, void *), void *data)
1220 {
1221         struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1222         struct inode *inode;
1223 
1224         spin_lock(&inode_hash_lock);
1225         inode = find_inode(sb, head, test, data);
1226         spin_unlock(&inode_hash_lock);
1227 
1228         return inode;
1229 }
1230 EXPORT_SYMBOL(ilookup5_nowait);
1231 
1232 /**
1233  * ilookup5 - search for an inode in the inode cache
1234  * @sb:         super block of file system to search
1235  * @hashval:    hash value (usually inode number) to search for
1236  * @test:       callback used for comparisons between inodes
1237  * @data:       opaque data pointer to pass to @test
1238  *
1239  * Search for the inode specified by @hashval and @data in the inode cache,
1240  * and if the inode is in the cache, return the inode with an incremented
1241  * reference count.  Waits on I_NEW before returning the inode.
1242  * returned with an incremented reference count.
1243  *
1244  * This is a generalized version of ilookup() for file systems where the
1245  * inode number is not sufficient for unique identification of an inode.
1246  *
1247  * Note: @test is called with the inode_hash_lock held, so can't sleep.
1248  */
1249 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1250                 int (*test)(struct inode *, void *), void *data)
1251 {
1252         struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1253 
1254         if (inode)
1255                 wait_on_inode(inode);
1256         return inode;
1257 }
1258 EXPORT_SYMBOL(ilookup5);
1259 
1260 /**
1261  * ilookup - search for an inode in the inode cache
1262  * @sb:         super block of file system to search
1263  * @ino:        inode number to search for
1264  *
1265  * Search for the inode @ino in the inode cache, and if the inode is in the
1266  * cache, the inode is returned with an incremented reference count.
1267  */
1268 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1269 {
1270         struct hlist_head *head = inode_hashtable + hash(sb, ino);
1271         struct inode *inode;
1272 
1273         spin_lock(&inode_hash_lock);
1274         inode = find_inode_fast(sb, head, ino);
1275         spin_unlock(&inode_hash_lock);
1276 
1277         if (inode)
1278                 wait_on_inode(inode);
1279         return inode;
1280 }
1281 EXPORT_SYMBOL(ilookup);
1282 
1283 int insert_inode_locked(struct inode *inode)
1284 {
1285         struct super_block *sb = inode->i_sb;
1286         ino_t ino = inode->i_ino;
1287         struct hlist_head *head = inode_hashtable + hash(sb, ino);
1288 
1289         while (1) {
1290                 struct inode *old = NULL;
1291                 spin_lock(&inode_hash_lock);
1292                 hlist_for_each_entry(old, head, i_hash) {
1293                         if (old->i_ino != ino)
1294                                 continue;
1295                         if (old->i_sb != sb)
1296                                 continue;
1297                         spin_lock(&old->i_lock);
1298                         if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1299                                 spin_unlock(&old->i_lock);
1300                                 continue;
1301                         }
1302                         break;
1303                 }
1304                 if (likely(!old)) {
1305                         spin_lock(&inode->i_lock);
1306                         inode->i_state |= I_NEW;
1307                         hlist_add_head(&inode->i_hash, head);
1308                         spin_unlock(&inode->i_lock);
1309                         spin_unlock(&inode_hash_lock);
1310                         return 0;
1311                 }
1312                 __iget(old);
1313                 spin_unlock(&old->i_lock);
1314                 spin_unlock(&inode_hash_lock);
1315                 wait_on_inode(old);
1316                 if (unlikely(!inode_unhashed(old))) {
1317                         iput(old);
1318                         return -EBUSY;
1319                 }
1320                 iput(old);
1321         }
1322 }
1323 EXPORT_SYMBOL(insert_inode_locked);
1324 
1325 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1326                 int (*test)(struct inode *, void *), void *data)
1327 {
1328         struct super_block *sb = inode->i_sb;
1329         struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1330 
1331         while (1) {
1332                 struct inode *old = NULL;
1333 
1334                 spin_lock(&inode_hash_lock);
1335                 hlist_for_each_entry(old, head, i_hash) {
1336                         if (old->i_sb != sb)
1337                                 continue;
1338                         if (!test(old, data))
1339                                 continue;
1340                         spin_lock(&old->i_lock);
1341                         if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1342                                 spin_unlock(&old->i_lock);
1343                                 continue;
1344                         }
1345                         break;
1346                 }
1347                 if (likely(!old)) {
1348                         spin_lock(&inode->i_lock);
1349                         inode->i_state |= I_NEW;
1350                         hlist_add_head(&inode->i_hash, head);
1351                         spin_unlock(&inode->i_lock);
1352                         spin_unlock(&inode_hash_lock);
1353                         return 0;
1354                 }
1355                 __iget(old);
1356                 spin_unlock(&old->i_lock);
1357                 spin_unlock(&inode_hash_lock);
1358                 wait_on_inode(old);
1359                 if (unlikely(!inode_unhashed(old))) {
1360                         iput(old);
1361                         return -EBUSY;
1362                 }
1363                 iput(old);
1364         }
1365 }
1366 EXPORT_SYMBOL(insert_inode_locked4);
1367 
1368 
1369 int generic_delete_inode(struct inode *inode)
1370 {
1371         return 1;
1372 }
1373 EXPORT_SYMBOL(generic_delete_inode);
1374 
1375 /*
1376  * Called when we're dropping the last reference
1377  * to an inode.
1378  *
1379  * Call the FS "drop_inode()" function, defaulting to
1380  * the legacy UNIX filesystem behaviour.  If it tells
1381  * us to evict inode, do so.  Otherwise, retain inode
1382  * in cache if fs is alive, sync and evict if fs is
1383  * shutting down.
1384  */
1385 static void iput_final(struct inode *inode)
1386 {
1387         struct super_block *sb = inode->i_sb;
1388         const struct super_operations *op = inode->i_sb->s_op;
1389         int drop;
1390 
1391         WARN_ON(inode->i_state & I_NEW);
1392 
1393         if (op->drop_inode)
1394                 drop = op->drop_inode(inode);
1395         else
1396                 drop = generic_drop_inode(inode);
1397 
1398         if (!drop && (sb->s_flags & MS_ACTIVE)) {
1399                 inode->i_state |= I_REFERENCED;
1400                 inode_add_lru(inode);
1401                 spin_unlock(&inode->i_lock);
1402                 return;
1403         }
1404 
1405         if (!drop) {
1406                 inode->i_state |= I_WILL_FREE;
1407                 spin_unlock(&inode->i_lock);
1408                 write_inode_now(inode, 1);
1409                 spin_lock(&inode->i_lock);
1410                 WARN_ON(inode->i_state & I_NEW);
1411                 inode->i_state &= ~I_WILL_FREE;
1412         }
1413 
1414         inode->i_state |= I_FREEING;
1415         if (!list_empty(&inode->i_lru))
1416                 inode_lru_list_del(inode);
1417         spin_unlock(&inode->i_lock);
1418 
1419         evict(inode);
1420 }
1421 
1422 /**
1423  *      iput    - put an inode
1424  *      @inode: inode to put
1425  *
1426  *      Puts an inode, dropping its usage count. If the inode use count hits
1427  *      zero, the inode is then freed and may also be destroyed.
1428  *
1429  *      Consequently, iput() can sleep.
1430  */
1431 void iput(struct inode *inode)
1432 {
1433         if (inode) {
1434                 BUG_ON(inode->i_state & I_CLEAR);
1435 
1436                 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1437                         iput_final(inode);
1438         }
1439 }
1440 EXPORT_SYMBOL(iput);
1441 
1442 /**
1443  *      bmap    - find a block number in a file
1444  *      @inode: inode of file
1445  *      @block: block to find
1446  *
1447  *      Returns the block number on the device holding the inode that
1448  *      is the disk block number for the block of the file requested.
1449  *      That is, asked for block 4 of inode 1 the function will return the
1450  *      disk block relative to the disk start that holds that block of the
1451  *      file.
1452  */
1453 sector_t bmap(struct inode *inode, sector_t block)
1454 {
1455         sector_t res = 0;
1456         if (inode->i_mapping->a_ops->bmap)
1457                 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1458         return res;
1459 }
1460 EXPORT_SYMBOL(bmap);
1461 
1462 /*
1463  * With relative atime, only update atime if the previous atime is
1464  * earlier than either the ctime or mtime or if at least a day has
1465  * passed since the last atime update.
1466  */
1467 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1468                              struct timespec now)
1469 {
1470 
1471         if (!(mnt->mnt_flags & MNT_RELATIME))
1472                 return 1;
1473         /*
1474          * Is mtime younger than atime? If yes, update atime:
1475          */
1476         if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1477                 return 1;
1478         /*
1479          * Is ctime younger than atime? If yes, update atime:
1480          */
1481         if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1482                 return 1;
1483 
1484         /*
1485          * Is the previous atime value older than a day? If yes,
1486          * update atime:
1487          */
1488         if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1489                 return 1;
1490         /*
1491          * Good, we can skip the atime update:
1492          */
1493         return 0;
1494 }
1495 
1496 /*
1497  * This does the actual work of updating an inodes time or version.  Must have
1498  * had called mnt_want_write() before calling this.
1499  */
1500 static int update_time(struct inode *inode, struct timespec *time, int flags)
1501 {
1502         if (inode->i_op->update_time)
1503                 return inode->i_op->update_time(inode, time, flags);
1504 
1505         if (flags & S_ATIME)
1506                 inode->i_atime = *time;
1507         if (flags & S_VERSION)
1508                 inode_inc_iversion(inode);
1509         if (flags & S_CTIME)
1510                 inode->i_ctime = *time;
1511         if (flags & S_MTIME)
1512                 inode->i_mtime = *time;
1513         mark_inode_dirty_sync(inode);
1514         return 0;
1515 }
1516 
1517 /**
1518  *      touch_atime     -       update the access time
1519  *      @path: the &struct path to update
1520  *
1521  *      Update the accessed time on an inode and mark it for writeback.
1522  *      This function automatically handles read only file systems and media,
1523  *      as well as the "noatime" flag and inode specific "noatime" markers.
1524  */
1525 void touch_atime(const struct path *path)
1526 {
1527         struct vfsmount *mnt = path->mnt;
1528         struct inode *inode = path->dentry->d_inode;
1529         struct timespec now;
1530 
1531         if (inode->i_flags & S_NOATIME)
1532                 return;
1533         if (IS_NOATIME(inode))
1534                 return;
1535         if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1536                 return;
1537 
1538         if (mnt->mnt_flags & MNT_NOATIME)
1539                 return;
1540         if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1541                 return;
1542 
1543         now = current_fs_time(inode->i_sb);
1544 
1545         if (!relatime_need_update(mnt, inode, now))
1546                 return;
1547 
1548         if (timespec_equal(&inode->i_atime, &now))
1549                 return;
1550 
1551         if (!sb_start_write_trylock(inode->i_sb))
1552                 return;
1553 
1554         if (__mnt_want_write(mnt))
1555                 goto skip_update;
1556         /*
1557          * File systems can error out when updating inodes if they need to
1558          * allocate new space to modify an inode (such is the case for
1559          * Btrfs), but since we touch atime while walking down the path we
1560          * really don't care if we failed to update the atime of the file,
1561          * so just ignore the return value.
1562          * We may also fail on filesystems that have the ability to make parts
1563          * of the fs read only, e.g. subvolumes in Btrfs.
1564          */
1565         update_time(inode, &now, S_ATIME);
1566         __mnt_drop_write(mnt);
1567 skip_update:
1568         sb_end_write(inode->i_sb);
1569 }
1570 EXPORT_SYMBOL(touch_atime);
1571 
1572 /*
1573  * The logic we want is
1574  *
1575  *      if suid or (sgid and xgrp)
1576  *              remove privs
1577  */
1578 int should_remove_suid(struct dentry *dentry)
1579 {
1580         umode_t mode = dentry->d_inode->i_mode;
1581         int kill = 0;
1582 
1583         /* suid always must be killed */
1584         if (unlikely(mode & S_ISUID))
1585                 kill = ATTR_KILL_SUID;
1586 
1587         /*
1588          * sgid without any exec bits is just a mandatory locking mark; leave
1589          * it alone.  If some exec bits are set, it's a real sgid; kill it.
1590          */
1591         if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1592                 kill |= ATTR_KILL_SGID;
1593 
1594         if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1595                 return kill;
1596 
1597         return 0;
1598 }
1599 EXPORT_SYMBOL(should_remove_suid);
1600 
1601 static int __remove_suid(struct dentry *dentry, int kill)
1602 {
1603         struct iattr newattrs;
1604 
1605         newattrs.ia_valid = ATTR_FORCE | kill;
1606         /*
1607          * Note we call this on write, so notify_change will not
1608          * encounter any conflicting delegations:
1609          */
1610         return notify_change(dentry, &newattrs, NULL);
1611 }
1612 
1613 int file_remove_suid(struct file *file)
1614 {
1615         struct dentry *dentry = file->f_path.dentry;
1616         struct inode *inode = dentry->d_inode;
1617         int killsuid;
1618         int killpriv;
1619         int error = 0;
1620 
1621         /* Fast path for nothing security related */
1622         if (IS_NOSEC(inode))
1623                 return 0;
1624 
1625         killsuid = should_remove_suid(dentry);
1626         killpriv = security_inode_need_killpriv(dentry);
1627 
1628         if (killpriv < 0)
1629                 return killpriv;
1630         if (killpriv)
1631                 error = security_inode_killpriv(dentry);
1632         if (!error && killsuid)
1633                 error = __remove_suid(dentry, killsuid);
1634         if (!error)
1635                 inode_has_no_xattr(inode);
1636 
1637         return error;
1638 }
1639 EXPORT_SYMBOL(file_remove_suid);
1640 
1641 /**
1642  *      file_update_time        -       update mtime and ctime time
1643  *      @file: file accessed
1644  *
1645  *      Update the mtime and ctime members of an inode and mark the inode
1646  *      for writeback.  Note that this function is meant exclusively for
1647  *      usage in the file write path of filesystems, and filesystems may
1648  *      choose to explicitly ignore update via this function with the
1649  *      S_NOCMTIME inode flag, e.g. for network filesystem where these
1650  *      timestamps are handled by the server.  This can return an error for
1651  *      file systems who need to allocate space in order to update an inode.
1652  */
1653 
1654 int file_update_time(struct file *file)
1655 {
1656         struct inode *inode = file_inode(file);
1657         struct timespec now;
1658         int sync_it = 0;
1659         int ret;
1660 
1661         /* First try to exhaust all avenues to not sync */
1662         if (IS_NOCMTIME(inode))
1663                 return 0;
1664 
1665         now = current_fs_time(inode->i_sb);
1666         if (!timespec_equal(&inode->i_mtime, &now))
1667                 sync_it = S_MTIME;
1668 
1669         if (!timespec_equal(&inode->i_ctime, &now))
1670                 sync_it |= S_CTIME;
1671 
1672         if (IS_I_VERSION(inode))
1673                 sync_it |= S_VERSION;
1674 
1675         if (!sync_it)
1676                 return 0;
1677 
1678         /* Finally allowed to write? Takes lock. */
1679         if (__mnt_want_write_file(file))
1680                 return 0;
1681 
1682         ret = update_time(inode, &now, sync_it);
1683         __mnt_drop_write_file(file);
1684 
1685         return ret;
1686 }
1687 EXPORT_SYMBOL(file_update_time);
1688 
1689 int inode_needs_sync(struct inode *inode)
1690 {
1691         if (IS_SYNC(inode))
1692                 return 1;
1693         if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1694                 return 1;
1695         return 0;
1696 }
1697 EXPORT_SYMBOL(inode_needs_sync);
1698 
1699 /*
1700  * If we try to find an inode in the inode hash while it is being
1701  * deleted, we have to wait until the filesystem completes its
1702  * deletion before reporting that it isn't found.  This function waits
1703  * until the deletion _might_ have completed.  Callers are responsible
1704  * to recheck inode state.
1705  *
1706  * It doesn't matter if I_NEW is not set initially, a call to
1707  * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1708  * will DTRT.
1709  */
1710 static void __wait_on_freeing_inode(struct inode *inode)
1711 {
1712         wait_queue_head_t *wq;
1713         DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1714         wq = bit_waitqueue(&inode->i_state, __I_NEW);
1715         prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1716         spin_unlock(&inode->i_lock);
1717         spin_unlock(&inode_hash_lock);
1718         schedule();
1719         finish_wait(wq, &wait.wait);
1720         spin_lock(&inode_hash_lock);
1721 }
1722 
1723 static __initdata unsigned long ihash_entries;
1724 static int __init set_ihash_entries(char *str)
1725 {
1726         if (!str)
1727                 return 0;
1728         ihash_entries = simple_strtoul(str, &str, 0);
1729         return 1;
1730 }
1731 __setup("ihash_entries=", set_ihash_entries);
1732 
1733 /*
1734  * Initialize the waitqueues and inode hash table.
1735  */
1736 void __init inode_init_early(void)
1737 {
1738         unsigned int loop;
1739 
1740         /* If hashes are distributed across NUMA nodes, defer
1741          * hash allocation until vmalloc space is available.
1742          */
1743         if (hashdist)
1744                 return;
1745 
1746         inode_hashtable =
1747                 alloc_large_system_hash("Inode-cache",
1748                                         sizeof(struct hlist_head),
1749                                         ihash_entries,
1750                                         14,
1751                                         HASH_EARLY,
1752                                         &i_hash_shift,
1753                                         &i_hash_mask,
1754                                         0,
1755                                         0);
1756 
1757         for (loop = 0; loop < (1U << i_hash_shift); loop++)
1758                 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1759 }
1760 
1761 void __init inode_init(void)
1762 {
1763         unsigned int loop;
1764 
1765         /* inode slab cache */
1766         inode_cachep = kmem_cache_create("inode_cache",
1767                                          sizeof(struct inode),
1768                                          0,
1769                                          (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1770                                          SLAB_MEM_SPREAD),
1771                                          init_once);
1772 
1773         /* Hash may have been set up in inode_init_early */
1774         if (!hashdist)
1775                 return;
1776 
1777         inode_hashtable =
1778                 alloc_large_system_hash("Inode-cache",
1779                                         sizeof(struct hlist_head),
1780                                         ihash_entries,
1781                                         14,
1782                                         0,
1783                                         &i_hash_shift,
1784                                         &i_hash_mask,
1785                                         0,
1786                                         0);
1787 
1788         for (loop = 0; loop < (1U << i_hash_shift); loop++)
1789                 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1790 }
1791 
1792 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1793 {
1794         inode->i_mode = mode;
1795         if (S_ISCHR(mode)) {
1796                 inode->i_fop = &def_chr_fops;
1797                 inode->i_rdev = rdev;
1798         } else if (S_ISBLK(mode)) {
1799                 inode->i_fop = &def_blk_fops;
1800                 inode->i_rdev = rdev;
1801         } else if (S_ISFIFO(mode))
1802                 inode->i_fop = &pipefifo_fops;
1803         else if (S_ISSOCK(mode))
1804                 inode->i_fop = &bad_sock_fops;
1805         else
1806                 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1807                                   " inode %s:%lu\n", mode, inode->i_sb->s_id,
1808                                   inode->i_ino);
1809 }
1810 EXPORT_SYMBOL(init_special_inode);
1811 
1812 /**
1813  * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1814  * @inode: New inode
1815  * @dir: Directory inode
1816  * @mode: mode of the new inode
1817  */
1818 void inode_init_owner(struct inode *inode, const struct inode *dir,
1819                         umode_t mode)
1820 {
1821         inode->i_uid = current_fsuid();
1822         if (dir && dir->i_mode & S_ISGID) {
1823                 inode->i_gid = dir->i_gid;
1824 
1825                 /* Directories are special, and always inherit S_ISGID */
1826                 if (S_ISDIR(mode))
1827                         mode |= S_ISGID;
1828                 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
1829                          !in_group_p(inode->i_gid) &&
1830                          !capable_wrt_inode_uidgid(dir, CAP_FSETID))
1831                         mode &= ~S_ISGID;
1832         } else
1833                 inode->i_gid = current_fsgid();
1834         inode->i_mode = mode;
1835 }
1836 EXPORT_SYMBOL(inode_init_owner);
1837 
1838 /**
1839  * inode_owner_or_capable - check current task permissions to inode
1840  * @inode: inode being checked
1841  *
1842  * Return true if current either has CAP_FOWNER in a namespace with the
1843  * inode owner uid mapped, or owns the file.
1844  */
1845 bool inode_owner_or_capable(const struct inode *inode)
1846 {
1847         struct user_namespace *ns;
1848 
1849         if (uid_eq(current_fsuid(), inode->i_uid))
1850                 return true;
1851 
1852         ns = current_user_ns();
1853         if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
1854                 return true;
1855         return false;
1856 }
1857 EXPORT_SYMBOL(inode_owner_or_capable);
1858 
1859 /*
1860  * Direct i/o helper functions
1861  */
1862 static void __inode_dio_wait(struct inode *inode)
1863 {
1864         wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1865         DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1866 
1867         do {
1868                 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1869                 if (atomic_read(&inode->i_dio_count))
1870                         schedule();
1871         } while (atomic_read(&inode->i_dio_count));
1872         finish_wait(wq, &q.wait);
1873 }
1874 
1875 /**
1876  * inode_dio_wait - wait for outstanding DIO requests to finish
1877  * @inode: inode to wait for
1878  *
1879  * Waits for all pending direct I/O requests to finish so that we can
1880  * proceed with a truncate or equivalent operation.
1881  *
1882  * Must be called under a lock that serializes taking new references
1883  * to i_dio_count, usually by inode->i_mutex.
1884  */
1885 void inode_dio_wait(struct inode *inode)
1886 {
1887         if (atomic_read(&inode->i_dio_count))
1888                 __inode_dio_wait(inode);
1889 }
1890 EXPORT_SYMBOL(inode_dio_wait);
1891 
1892 /*
1893  * inode_dio_done - signal finish of a direct I/O requests
1894  * @inode: inode the direct I/O happens on
1895  *
1896  * This is called once we've finished processing a direct I/O request,
1897  * and is used to wake up callers waiting for direct I/O to be quiesced.
1898  */
1899 void inode_dio_done(struct inode *inode)
1900 {
1901         if (atomic_dec_and_test(&inode->i_dio_count))
1902                 wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1903 }
1904 EXPORT_SYMBOL(inode_dio_done);
1905 
1906 /*
1907  * inode_set_flags - atomically set some inode flags
1908  *
1909  * Note: the caller should be holding i_mutex, or else be sure that
1910  * they have exclusive access to the inode structure (i.e., while the
1911  * inode is being instantiated).  The reason for the cmpxchg() loop
1912  * --- which wouldn't be necessary if all code paths which modify
1913  * i_flags actually followed this rule, is that there is at least one
1914  * code path which doesn't today --- for example,
1915  * __generic_file_aio_write() calls file_remove_suid() without holding
1916  * i_mutex --- so we use cmpxchg() out of an abundance of caution.
1917  *
1918  * In the long run, i_mutex is overkill, and we should probably look
1919  * at using the i_lock spinlock to protect i_flags, and then make sure
1920  * it is so documented in include/linux/fs.h and that all code follows
1921  * the locking convention!!
1922  */
1923 void inode_set_flags(struct inode *inode, unsigned int flags,
1924                      unsigned int mask)
1925 {
1926         unsigned int old_flags, new_flags;
1927 
1928         WARN_ON_ONCE(flags & ~mask);
1929         do {
1930                 old_flags = ACCESS_ONCE(inode->i_flags);
1931                 new_flags = (old_flags & ~mask) | flags;
1932         } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
1933                                   new_flags) != old_flags));
1934 }
1935 EXPORT_SYMBOL(inode_set_flags);
1936 

~ [ 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