]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/fs/inode.c | |
3 | * | |
4 | * (C) 1997 Linus Torvalds | |
5 | */ | |
6 | ||
7 | #include <linux/fs.h> | |
8 | #include <linux/mm.h> | |
9 | #include <linux/dcache.h> | |
10 | #include <linux/init.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/writeback.h> | |
13 | #include <linux/module.h> | |
14 | #include <linux/backing-dev.h> | |
15 | #include <linux/wait.h> | |
16 | #include <linux/rwsem.h> | |
17 | #include <linux/hash.h> | |
18 | #include <linux/swap.h> | |
19 | #include <linux/security.h> | |
20 | #include <linux/pagemap.h> | |
21 | #include <linux/cdev.h> | |
22 | #include <linux/bootmem.h> | |
23 | #include <linux/inotify.h> | |
24 | #include <linux/fsnotify.h> | |
25 | #include <linux/mount.h> | |
26 | #include <linux/async.h> | |
27 | #include <linux/posix_acl.h> | |
28 | ||
29 | /* | |
30 | * This is needed for the following functions: | |
31 | * - inode_has_buffers | |
32 | * - invalidate_inode_buffers | |
33 | * - invalidate_bdev | |
34 | * | |
35 | * FIXME: remove all knowledge of the buffer layer from this file | |
36 | */ | |
37 | #include <linux/buffer_head.h> | |
38 | ||
39 | /* | |
40 | * New inode.c implementation. | |
41 | * | |
42 | * This implementation has the basic premise of trying | |
43 | * to be extremely low-overhead and SMP-safe, yet be | |
44 | * simple enough to be "obviously correct". | |
45 | * | |
46 | * Famous last words. | |
47 | */ | |
48 | ||
49 | /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */ | |
50 | ||
51 | /* #define INODE_PARANOIA 1 */ | |
52 | /* #define INODE_DEBUG 1 */ | |
53 | ||
54 | /* | |
55 | * Inode lookup is no longer as critical as it used to be: | |
56 | * most of the lookups are going to be through the dcache. | |
57 | */ | |
58 | #define I_HASHBITS i_hash_shift | |
59 | #define I_HASHMASK i_hash_mask | |
60 | ||
61 | static unsigned int i_hash_mask __read_mostly; | |
62 | static unsigned int i_hash_shift __read_mostly; | |
63 | ||
64 | /* | |
65 | * Each inode can be on two separate lists. One is | |
66 | * the hash list of the inode, used for lookups. The | |
67 | * other linked list is the "type" list: | |
68 | * "in_use" - valid inode, i_count > 0, i_nlink > 0 | |
69 | * "dirty" - as "in_use" but also dirty | |
70 | * "unused" - valid inode, i_count = 0 | |
71 | * | |
72 | * A "dirty" list is maintained for each super block, | |
73 | * allowing for low-overhead inode sync() operations. | |
74 | */ | |
75 | ||
76 | LIST_HEAD(inode_in_use); | |
77 | LIST_HEAD(inode_unused); | |
78 | static struct hlist_head *inode_hashtable __read_mostly; | |
79 | ||
80 | /* | |
81 | * A simple spinlock to protect the list manipulations. | |
82 | * | |
83 | * NOTE! You also have to own the lock if you change | |
84 | * the i_state of an inode while it is in use.. | |
85 | */ | |
86 | DEFINE_SPINLOCK(inode_lock); | |
87 | ||
88 | /* | |
89 | * iprune_sem provides exclusion between the kswapd or try_to_free_pages | |
90 | * icache shrinking path, and the umount path. Without this exclusion, | |
91 | * by the time prune_icache calls iput for the inode whose pages it has | |
92 | * been invalidating, or by the time it calls clear_inode & destroy_inode | |
93 | * from its final dispose_list, the struct super_block they refer to | |
94 | * (for inode->i_sb->s_op) may already have been freed and reused. | |
95 | * | |
96 | * We make this an rwsem because the fastpath is icache shrinking. In | |
97 | * some cases a filesystem may be doing a significant amount of work in | |
98 | * its inode reclaim code, so this should improve parallelism. | |
99 | */ | |
100 | static DECLARE_RWSEM(iprune_sem); | |
101 | ||
102 | /* | |
103 | * Statistics gathering.. | |
104 | */ | |
105 | struct inodes_stat_t inodes_stat; | |
106 | ||
107 | static struct kmem_cache *inode_cachep __read_mostly; | |
108 | ||
109 | static void wake_up_inode(struct inode *inode) | |
110 | { | |
111 | /* | |
112 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
113 | */ | |
114 | smp_mb(); | |
115 | wake_up_bit(&inode->i_state, __I_NEW); | |
116 | } | |
117 | ||
118 | /** | |
119 | * inode_init_always - perform inode structure intialisation | |
120 | * @sb: superblock inode belongs to | |
121 | * @inode: inode to initialise | |
122 | * | |
123 | * These are initializations that need to be done on every inode | |
124 | * allocation as the fields are not initialised by slab allocation. | |
125 | */ | |
126 | int inode_init_always(struct super_block *sb, struct inode *inode) | |
127 | { | |
128 | static const struct address_space_operations empty_aops; | |
129 | static const struct inode_operations empty_iops; | |
130 | static const struct file_operations empty_fops; | |
131 | struct address_space *const mapping = &inode->i_data; | |
132 | ||
133 | inode->i_sb = sb; | |
134 | inode->i_blkbits = sb->s_blocksize_bits; | |
135 | inode->i_flags = 0; | |
136 | atomic_set(&inode->i_count, 1); | |
137 | inode->i_op = &empty_iops; | |
138 | inode->i_fop = &empty_fops; | |
139 | inode->i_nlink = 1; | |
140 | inode->i_uid = 0; | |
141 | inode->i_gid = 0; | |
142 | atomic_set(&inode->i_writecount, 0); | |
143 | inode->i_size = 0; | |
144 | inode->i_blocks = 0; | |
145 | inode->i_bytes = 0; | |
146 | inode->i_generation = 0; | |
147 | #ifdef CONFIG_QUOTA | |
148 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); | |
149 | #endif | |
150 | inode->i_pipe = NULL; | |
151 | inode->i_bdev = NULL; | |
152 | inode->i_cdev = NULL; | |
153 | inode->i_rdev = 0; | |
154 | inode->dirtied_when = 0; | |
155 | ||
156 | if (security_inode_alloc(inode)) | |
157 | goto out; | |
158 | spin_lock_init(&inode->i_lock); | |
159 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
160 | ||
161 | mutex_init(&inode->i_mutex); | |
162 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
163 | ||
164 | init_rwsem(&inode->i_alloc_sem); | |
165 | lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key); | |
166 | ||
167 | mapping->a_ops = &empty_aops; | |
168 | mapping->host = inode; | |
169 | mapping->flags = 0; | |
170 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); | |
171 | mapping->assoc_mapping = NULL; | |
172 | mapping->backing_dev_info = &default_backing_dev_info; | |
173 | mapping->writeback_index = 0; | |
174 | ||
175 | /* | |
176 | * If the block_device provides a backing_dev_info for client | |
177 | * inodes then use that. Otherwise the inode share the bdev's | |
178 | * backing_dev_info. | |
179 | */ | |
180 | if (sb->s_bdev) { | |
181 | struct backing_dev_info *bdi; | |
182 | ||
183 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; | |
184 | mapping->backing_dev_info = bdi; | |
185 | } | |
186 | inode->i_private = NULL; | |
187 | inode->i_mapping = mapping; | |
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 | return 0; | |
197 | out: | |
198 | return -ENOMEM; | |
199 | } | |
200 | EXPORT_SYMBOL(inode_init_always); | |
201 | ||
202 | static struct inode *alloc_inode(struct super_block *sb) | |
203 | { | |
204 | struct inode *inode; | |
205 | ||
206 | if (sb->s_op->alloc_inode) | |
207 | inode = sb->s_op->alloc_inode(sb); | |
208 | else | |
209 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
210 | ||
211 | if (!inode) | |
212 | return NULL; | |
213 | ||
214 | if (unlikely(inode_init_always(sb, inode))) { | |
215 | if (inode->i_sb->s_op->destroy_inode) | |
216 | inode->i_sb->s_op->destroy_inode(inode); | |
217 | else | |
218 | kmem_cache_free(inode_cachep, inode); | |
219 | return NULL; | |
220 | } | |
221 | ||
222 | return inode; | |
223 | } | |
224 | ||
225 | void __destroy_inode(struct inode *inode) | |
226 | { | |
227 | BUG_ON(inode_has_buffers(inode)); | |
228 | security_inode_free(inode); | |
229 | fsnotify_inode_delete(inode); | |
230 | #ifdef CONFIG_FS_POSIX_ACL | |
231 | if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED) | |
232 | posix_acl_release(inode->i_acl); | |
233 | if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED) | |
234 | posix_acl_release(inode->i_default_acl); | |
235 | #endif | |
236 | } | |
237 | EXPORT_SYMBOL(__destroy_inode); | |
238 | ||
239 | void destroy_inode(struct inode *inode) | |
240 | { | |
241 | __destroy_inode(inode); | |
242 | if (inode->i_sb->s_op->destroy_inode) | |
243 | inode->i_sb->s_op->destroy_inode(inode); | |
244 | else | |
245 | kmem_cache_free(inode_cachep, (inode)); | |
246 | } | |
247 | ||
248 | /* | |
249 | * These are initializations that only need to be done | |
250 | * once, because the fields are idempotent across use | |
251 | * of the inode, so let the slab aware of that. | |
252 | */ | |
253 | void inode_init_once(struct inode *inode) | |
254 | { | |
255 | memset(inode, 0, sizeof(*inode)); | |
256 | INIT_HLIST_NODE(&inode->i_hash); | |
257 | INIT_LIST_HEAD(&inode->i_dentry); | |
258 | INIT_LIST_HEAD(&inode->i_devices); | |
259 | INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC); | |
260 | spin_lock_init(&inode->i_data.tree_lock); | |
261 | spin_lock_init(&inode->i_data.i_mmap_lock); | |
262 | INIT_LIST_HEAD(&inode->i_data.private_list); | |
263 | spin_lock_init(&inode->i_data.private_lock); | |
264 | INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap); | |
265 | INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear); | |
266 | i_size_ordered_init(inode); | |
267 | #ifdef CONFIG_INOTIFY | |
268 | INIT_LIST_HEAD(&inode->inotify_watches); | |
269 | mutex_init(&inode->inotify_mutex); | |
270 | #endif | |
271 | #ifdef CONFIG_FSNOTIFY | |
272 | INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries); | |
273 | #endif | |
274 | } | |
275 | EXPORT_SYMBOL(inode_init_once); | |
276 | ||
277 | static void init_once(void *foo) | |
278 | { | |
279 | struct inode *inode = (struct inode *) foo; | |
280 | ||
281 | inode_init_once(inode); | |
282 | } | |
283 | ||
284 | /* | |
285 | * inode_lock must be held | |
286 | */ | |
287 | void __iget(struct inode *inode) | |
288 | { | |
289 | if (atomic_inc_return(&inode->i_count) != 1) | |
290 | return; | |
291 | ||
292 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) | |
293 | list_move(&inode->i_list, &inode_in_use); | |
294 | inodes_stat.nr_unused--; | |
295 | } | |
296 | ||
297 | /** | |
298 | * clear_inode - clear an inode | |
299 | * @inode: inode to clear | |
300 | * | |
301 | * This is called by the filesystem to tell us | |
302 | * that the inode is no longer useful. We just | |
303 | * terminate it with extreme prejudice. | |
304 | */ | |
305 | void clear_inode(struct inode *inode) | |
306 | { | |
307 | might_sleep(); | |
308 | invalidate_inode_buffers(inode); | |
309 | ||
310 | BUG_ON(inode->i_data.nrpages); | |
311 | BUG_ON(!(inode->i_state & I_FREEING)); | |
312 | BUG_ON(inode->i_state & I_CLEAR); | |
313 | inode_sync_wait(inode); | |
314 | if (inode->i_sb->s_op->clear_inode) | |
315 | inode->i_sb->s_op->clear_inode(inode); | |
316 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) | |
317 | bd_forget(inode); | |
318 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
319 | cd_forget(inode); | |
320 | inode->i_state = I_FREEING | I_CLEAR; | |
321 | } | |
322 | EXPORT_SYMBOL(clear_inode); | |
323 | ||
324 | static void evict(struct inode *inode, int delete) | |
325 | { | |
326 | const struct super_operations *op = inode->i_sb->s_op; | |
327 | ||
328 | if (op->evict_inode) { | |
329 | op->evict_inode(inode); | |
330 | } else if (delete && op->delete_inode) { | |
331 | op->delete_inode(inode); | |
332 | } else { | |
333 | if (inode->i_data.nrpages) | |
334 | truncate_inode_pages(&inode->i_data, 0); | |
335 | clear_inode(inode); | |
336 | } | |
337 | } | |
338 | ||
339 | /* | |
340 | * dispose_list - dispose of the contents of a local list | |
341 | * @head: the head of the list to free | |
342 | * | |
343 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
344 | * need to worry about list corruption and SMP locks. | |
345 | */ | |
346 | static void dispose_list(struct list_head *head) | |
347 | { | |
348 | int nr_disposed = 0; | |
349 | ||
350 | while (!list_empty(head)) { | |
351 | struct inode *inode; | |
352 | ||
353 | inode = list_first_entry(head, struct inode, i_list); | |
354 | list_del(&inode->i_list); | |
355 | ||
356 | evict(inode, 0); | |
357 | ||
358 | spin_lock(&inode_lock); | |
359 | hlist_del_init(&inode->i_hash); | |
360 | list_del_init(&inode->i_sb_list); | |
361 | spin_unlock(&inode_lock); | |
362 | ||
363 | wake_up_inode(inode); | |
364 | destroy_inode(inode); | |
365 | nr_disposed++; | |
366 | } | |
367 | spin_lock(&inode_lock); | |
368 | inodes_stat.nr_inodes -= nr_disposed; | |
369 | spin_unlock(&inode_lock); | |
370 | } | |
371 | ||
372 | /* | |
373 | * Invalidate all inodes for a device. | |
374 | */ | |
375 | static int invalidate_list(struct list_head *head, struct list_head *dispose) | |
376 | { | |
377 | struct list_head *next; | |
378 | int busy = 0, count = 0; | |
379 | ||
380 | next = head->next; | |
381 | for (;;) { | |
382 | struct list_head *tmp = next; | |
383 | struct inode *inode; | |
384 | ||
385 | /* | |
386 | * We can reschedule here without worrying about the list's | |
387 | * consistency because the per-sb list of inodes must not | |
388 | * change during umount anymore, and because iprune_sem keeps | |
389 | * shrink_icache_memory() away. | |
390 | */ | |
391 | cond_resched_lock(&inode_lock); | |
392 | ||
393 | next = next->next; | |
394 | if (tmp == head) | |
395 | break; | |
396 | inode = list_entry(tmp, struct inode, i_sb_list); | |
397 | if (inode->i_state & I_NEW) | |
398 | continue; | |
399 | invalidate_inode_buffers(inode); | |
400 | if (!atomic_read(&inode->i_count)) { | |
401 | list_move(&inode->i_list, dispose); | |
402 | WARN_ON(inode->i_state & I_NEW); | |
403 | inode->i_state |= I_FREEING; | |
404 | count++; | |
405 | continue; | |
406 | } | |
407 | busy = 1; | |
408 | } | |
409 | /* only unused inodes may be cached with i_count zero */ | |
410 | inodes_stat.nr_unused -= count; | |
411 | return busy; | |
412 | } | |
413 | ||
414 | /** | |
415 | * invalidate_inodes - discard the inodes on a device | |
416 | * @sb: superblock | |
417 | * | |
418 | * Discard all of the inodes for a given superblock. If the discard | |
419 | * fails because there are busy inodes then a non zero value is returned. | |
420 | * If the discard is successful all the inodes have been discarded. | |
421 | */ | |
422 | int invalidate_inodes(struct super_block *sb) | |
423 | { | |
424 | int busy; | |
425 | LIST_HEAD(throw_away); | |
426 | ||
427 | down_write(&iprune_sem); | |
428 | spin_lock(&inode_lock); | |
429 | inotify_unmount_inodes(&sb->s_inodes); | |
430 | fsnotify_unmount_inodes(&sb->s_inodes); | |
431 | busy = invalidate_list(&sb->s_inodes, &throw_away); | |
432 | spin_unlock(&inode_lock); | |
433 | ||
434 | dispose_list(&throw_away); | |
435 | up_write(&iprune_sem); | |
436 | ||
437 | return busy; | |
438 | } | |
439 | EXPORT_SYMBOL(invalidate_inodes); | |
440 | ||
441 | static int can_unuse(struct inode *inode) | |
442 | { | |
443 | if (inode->i_state) | |
444 | return 0; | |
445 | if (inode_has_buffers(inode)) | |
446 | return 0; | |
447 | if (atomic_read(&inode->i_count)) | |
448 | return 0; | |
449 | if (inode->i_data.nrpages) | |
450 | return 0; | |
451 | return 1; | |
452 | } | |
453 | ||
454 | /* | |
455 | * Scan `goal' inodes on the unused list for freeable ones. They are moved to | |
456 | * a temporary list and then are freed outside inode_lock by dispose_list(). | |
457 | * | |
458 | * Any inodes which are pinned purely because of attached pagecache have their | |
459 | * pagecache removed. We expect the final iput() on that inode to add it to | |
460 | * the front of the inode_unused list. So look for it there and if the | |
461 | * inode is still freeable, proceed. The right inode is found 99.9% of the | |
462 | * time in testing on a 4-way. | |
463 | * | |
464 | * If the inode has metadata buffers attached to mapping->private_list then | |
465 | * try to remove them. | |
466 | */ | |
467 | static void prune_icache(int nr_to_scan) | |
468 | { | |
469 | LIST_HEAD(freeable); | |
470 | int nr_pruned = 0; | |
471 | int nr_scanned; | |
472 | unsigned long reap = 0; | |
473 | ||
474 | down_read(&iprune_sem); | |
475 | spin_lock(&inode_lock); | |
476 | for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) { | |
477 | struct inode *inode; | |
478 | ||
479 | if (list_empty(&inode_unused)) | |
480 | break; | |
481 | ||
482 | inode = list_entry(inode_unused.prev, struct inode, i_list); | |
483 | ||
484 | if (inode->i_state || atomic_read(&inode->i_count)) { | |
485 | list_move(&inode->i_list, &inode_unused); | |
486 | continue; | |
487 | } | |
488 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
489 | __iget(inode); | |
490 | spin_unlock(&inode_lock); | |
491 | if (remove_inode_buffers(inode)) | |
492 | reap += invalidate_mapping_pages(&inode->i_data, | |
493 | 0, -1); | |
494 | iput(inode); | |
495 | spin_lock(&inode_lock); | |
496 | ||
497 | if (inode != list_entry(inode_unused.next, | |
498 | struct inode, i_list)) | |
499 | continue; /* wrong inode or list_empty */ | |
500 | if (!can_unuse(inode)) | |
501 | continue; | |
502 | } | |
503 | list_move(&inode->i_list, &freeable); | |
504 | WARN_ON(inode->i_state & I_NEW); | |
505 | inode->i_state |= I_FREEING; | |
506 | nr_pruned++; | |
507 | } | |
508 | inodes_stat.nr_unused -= nr_pruned; | |
509 | if (current_is_kswapd()) | |
510 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
511 | else | |
512 | __count_vm_events(PGINODESTEAL, reap); | |
513 | spin_unlock(&inode_lock); | |
514 | ||
515 | dispose_list(&freeable); | |
516 | up_read(&iprune_sem); | |
517 | } | |
518 | ||
519 | /* | |
520 | * shrink_icache_memory() will attempt to reclaim some unused inodes. Here, | |
521 | * "unused" means that no dentries are referring to the inodes: the files are | |
522 | * not open and the dcache references to those inodes have already been | |
523 | * reclaimed. | |
524 | * | |
525 | * This function is passed the number of inodes to scan, and it returns the | |
526 | * total number of remaining possibly-reclaimable inodes. | |
527 | */ | |
528 | static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) | |
529 | { | |
530 | if (nr) { | |
531 | /* | |
532 | * Nasty deadlock avoidance. We may hold various FS locks, | |
533 | * and we don't want to recurse into the FS that called us | |
534 | * in clear_inode() and friends.. | |
535 | */ | |
536 | if (!(gfp_mask & __GFP_FS)) | |
537 | return -1; | |
538 | prune_icache(nr); | |
539 | } | |
540 | return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; | |
541 | } | |
542 | ||
543 | static struct shrinker icache_shrinker = { | |
544 | .shrink = shrink_icache_memory, | |
545 | .seeks = DEFAULT_SEEKS, | |
546 | }; | |
547 | ||
548 | static void __wait_on_freeing_inode(struct inode *inode); | |
549 | /* | |
550 | * Called with the inode lock held. | |
551 | * NOTE: we are not increasing the inode-refcount, you must call __iget() | |
552 | * by hand after calling find_inode now! This simplifies iunique and won't | |
553 | * add any additional branch in the common code. | |
554 | */ | |
555 | static struct inode *find_inode(struct super_block *sb, | |
556 | struct hlist_head *head, | |
557 | int (*test)(struct inode *, void *), | |
558 | void *data) | |
559 | { | |
560 | struct hlist_node *node; | |
561 | struct inode *inode = NULL; | |
562 | ||
563 | repeat: | |
564 | hlist_for_each_entry(inode, node, head, i_hash) { | |
565 | if (inode->i_sb != sb) | |
566 | continue; | |
567 | if (!test(inode, data)) | |
568 | continue; | |
569 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { | |
570 | __wait_on_freeing_inode(inode); | |
571 | goto repeat; | |
572 | } | |
573 | break; | |
574 | } | |
575 | return node ? inode : NULL; | |
576 | } | |
577 | ||
578 | /* | |
579 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
580 | * iget_locked for details. | |
581 | */ | |
582 | static struct inode *find_inode_fast(struct super_block *sb, | |
583 | struct hlist_head *head, unsigned long ino) | |
584 | { | |
585 | struct hlist_node *node; | |
586 | struct inode *inode = NULL; | |
587 | ||
588 | repeat: | |
589 | hlist_for_each_entry(inode, node, head, i_hash) { | |
590 | if (inode->i_ino != ino) | |
591 | continue; | |
592 | if (inode->i_sb != sb) | |
593 | continue; | |
594 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { | |
595 | __wait_on_freeing_inode(inode); | |
596 | goto repeat; | |
597 | } | |
598 | break; | |
599 | } | |
600 | return node ? inode : NULL; | |
601 | } | |
602 | ||
603 | static unsigned long hash(struct super_block *sb, unsigned long hashval) | |
604 | { | |
605 | unsigned long tmp; | |
606 | ||
607 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
608 | L1_CACHE_BYTES; | |
609 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS); | |
610 | return tmp & I_HASHMASK; | |
611 | } | |
612 | ||
613 | static inline void | |
614 | __inode_add_to_lists(struct super_block *sb, struct hlist_head *head, | |
615 | struct inode *inode) | |
616 | { | |
617 | inodes_stat.nr_inodes++; | |
618 | list_add(&inode->i_list, &inode_in_use); | |
619 | list_add(&inode->i_sb_list, &sb->s_inodes); | |
620 | if (head) | |
621 | hlist_add_head(&inode->i_hash, head); | |
622 | } | |
623 | ||
624 | /** | |
625 | * inode_add_to_lists - add a new inode to relevant lists | |
626 | * @sb: superblock inode belongs to | |
627 | * @inode: inode to mark in use | |
628 | * | |
629 | * When an inode is allocated it needs to be accounted for, added to the in use | |
630 | * list, the owning superblock and the inode hash. This needs to be done under | |
631 | * the inode_lock, so export a function to do this rather than the inode lock | |
632 | * itself. We calculate the hash list to add to here so it is all internal | |
633 | * which requires the caller to have already set up the inode number in the | |
634 | * inode to add. | |
635 | */ | |
636 | void inode_add_to_lists(struct super_block *sb, struct inode *inode) | |
637 | { | |
638 | struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino); | |
639 | ||
640 | spin_lock(&inode_lock); | |
641 | __inode_add_to_lists(sb, head, inode); | |
642 | spin_unlock(&inode_lock); | |
643 | } | |
644 | EXPORT_SYMBOL_GPL(inode_add_to_lists); | |
645 | ||
646 | /** | |
647 | * new_inode - obtain an inode | |
648 | * @sb: superblock | |
649 | * | |
650 | * Allocates a new inode for given superblock. The default gfp_mask | |
651 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. | |
652 | * If HIGHMEM pages are unsuitable or it is known that pages allocated | |
653 | * for the page cache are not reclaimable or migratable, | |
654 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
655 | * newly created inode's mapping | |
656 | * | |
657 | */ | |
658 | struct inode *new_inode(struct super_block *sb) | |
659 | { | |
660 | /* | |
661 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
662 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
663 | * here to attempt to avoid that. | |
664 | */ | |
665 | static unsigned int last_ino; | |
666 | struct inode *inode; | |
667 | ||
668 | spin_lock_prefetch(&inode_lock); | |
669 | ||
670 | inode = alloc_inode(sb); | |
671 | if (inode) { | |
672 | spin_lock(&inode_lock); | |
673 | __inode_add_to_lists(sb, NULL, inode); | |
674 | inode->i_ino = ++last_ino; | |
675 | inode->i_state = 0; | |
676 | spin_unlock(&inode_lock); | |
677 | } | |
678 | return inode; | |
679 | } | |
680 | EXPORT_SYMBOL(new_inode); | |
681 | ||
682 | void unlock_new_inode(struct inode *inode) | |
683 | { | |
684 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
685 | if (inode->i_mode & S_IFDIR) { | |
686 | struct file_system_type *type = inode->i_sb->s_type; | |
687 | ||
688 | /* Set new key only if filesystem hasn't already changed it */ | |
689 | if (!lockdep_match_class(&inode->i_mutex, | |
690 | &type->i_mutex_key)) { | |
691 | /* | |
692 | * ensure nobody is actually holding i_mutex | |
693 | */ | |
694 | mutex_destroy(&inode->i_mutex); | |
695 | mutex_init(&inode->i_mutex); | |
696 | lockdep_set_class(&inode->i_mutex, | |
697 | &type->i_mutex_dir_key); | |
698 | } | |
699 | } | |
700 | #endif | |
701 | /* | |
702 | * This is special! We do not need the spinlock when clearing I_NEW, | |
703 | * because we're guaranteed that nobody else tries to do anything about | |
704 | * the state of the inode when it is locked, as we just created it (so | |
705 | * there can be no old holders that haven't tested I_NEW). | |
706 | * However we must emit the memory barrier so that other CPUs reliably | |
707 | * see the clearing of I_NEW after the other inode initialisation has | |
708 | * completed. | |
709 | */ | |
710 | smp_mb(); | |
711 | WARN_ON(!(inode->i_state & I_NEW)); | |
712 | inode->i_state &= ~I_NEW; | |
713 | wake_up_inode(inode); | |
714 | } | |
715 | EXPORT_SYMBOL(unlock_new_inode); | |
716 | ||
717 | /* | |
718 | * This is called without the inode lock held.. Be careful. | |
719 | * | |
720 | * We no longer cache the sb_flags in i_flags - see fs.h | |
721 | * -- rmk@arm.uk.linux.org | |
722 | */ | |
723 | static struct inode *get_new_inode(struct super_block *sb, | |
724 | struct hlist_head *head, | |
725 | int (*test)(struct inode *, void *), | |
726 | int (*set)(struct inode *, void *), | |
727 | void *data) | |
728 | { | |
729 | struct inode *inode; | |
730 | ||
731 | inode = alloc_inode(sb); | |
732 | if (inode) { | |
733 | struct inode *old; | |
734 | ||
735 | spin_lock(&inode_lock); | |
736 | /* We released the lock, so.. */ | |
737 | old = find_inode(sb, head, test, data); | |
738 | if (!old) { | |
739 | if (set(inode, data)) | |
740 | goto set_failed; | |
741 | ||
742 | __inode_add_to_lists(sb, head, inode); | |
743 | inode->i_state = I_NEW; | |
744 | spin_unlock(&inode_lock); | |
745 | ||
746 | /* Return the locked inode with I_NEW set, the | |
747 | * caller is responsible for filling in the contents | |
748 | */ | |
749 | return inode; | |
750 | } | |
751 | ||
752 | /* | |
753 | * Uhhuh, somebody else created the same inode under | |
754 | * us. Use the old inode instead of the one we just | |
755 | * allocated. | |
756 | */ | |
757 | __iget(old); | |
758 | spin_unlock(&inode_lock); | |
759 | destroy_inode(inode); | |
760 | inode = old; | |
761 | wait_on_inode(inode); | |
762 | } | |
763 | return inode; | |
764 | ||
765 | set_failed: | |
766 | spin_unlock(&inode_lock); | |
767 | destroy_inode(inode); | |
768 | return NULL; | |
769 | } | |
770 | ||
771 | /* | |
772 | * get_new_inode_fast is the fast path version of get_new_inode, see the | |
773 | * comment at iget_locked for details. | |
774 | */ | |
775 | static struct inode *get_new_inode_fast(struct super_block *sb, | |
776 | struct hlist_head *head, unsigned long ino) | |
777 | { | |
778 | struct inode *inode; | |
779 | ||
780 | inode = alloc_inode(sb); | |
781 | if (inode) { | |
782 | struct inode *old; | |
783 | ||
784 | spin_lock(&inode_lock); | |
785 | /* We released the lock, so.. */ | |
786 | old = find_inode_fast(sb, head, ino); | |
787 | if (!old) { | |
788 | inode->i_ino = ino; | |
789 | __inode_add_to_lists(sb, head, inode); | |
790 | inode->i_state = I_NEW; | |
791 | spin_unlock(&inode_lock); | |
792 | ||
793 | /* Return the locked inode with I_NEW set, the | |
794 | * caller is responsible for filling in the contents | |
795 | */ | |
796 | return inode; | |
797 | } | |
798 | ||
799 | /* | |
800 | * Uhhuh, somebody else created the same inode under | |
801 | * us. Use the old inode instead of the one we just | |
802 | * allocated. | |
803 | */ | |
804 | __iget(old); | |
805 | spin_unlock(&inode_lock); | |
806 | destroy_inode(inode); | |
807 | inode = old; | |
808 | wait_on_inode(inode); | |
809 | } | |
810 | return inode; | |
811 | } | |
812 | ||
813 | /** | |
814 | * iunique - get a unique inode number | |
815 | * @sb: superblock | |
816 | * @max_reserved: highest reserved inode number | |
817 | * | |
818 | * Obtain an inode number that is unique on the system for a given | |
819 | * superblock. This is used by file systems that have no natural | |
820 | * permanent inode numbering system. An inode number is returned that | |
821 | * is higher than the reserved limit but unique. | |
822 | * | |
823 | * BUGS: | |
824 | * With a large number of inodes live on the file system this function | |
825 | * currently becomes quite slow. | |
826 | */ | |
827 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
828 | { | |
829 | /* | |
830 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
831 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
832 | * here to attempt to avoid that. | |
833 | */ | |
834 | static unsigned int counter; | |
835 | struct inode *inode; | |
836 | struct hlist_head *head; | |
837 | ino_t res; | |
838 | ||
839 | spin_lock(&inode_lock); | |
840 | do { | |
841 | if (counter <= max_reserved) | |
842 | counter = max_reserved + 1; | |
843 | res = counter++; | |
844 | head = inode_hashtable + hash(sb, res); | |
845 | inode = find_inode_fast(sb, head, res); | |
846 | } while (inode != NULL); | |
847 | spin_unlock(&inode_lock); | |
848 | ||
849 | return res; | |
850 | } | |
851 | EXPORT_SYMBOL(iunique); | |
852 | ||
853 | struct inode *igrab(struct inode *inode) | |
854 | { | |
855 | spin_lock(&inode_lock); | |
856 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) | |
857 | __iget(inode); | |
858 | else | |
859 | /* | |
860 | * Handle the case where s_op->clear_inode is not been | |
861 | * called yet, and somebody is calling igrab | |
862 | * while the inode is getting freed. | |
863 | */ | |
864 | inode = NULL; | |
865 | spin_unlock(&inode_lock); | |
866 | return inode; | |
867 | } | |
868 | EXPORT_SYMBOL(igrab); | |
869 | ||
870 | /** | |
871 | * ifind - internal function, you want ilookup5() or iget5(). | |
872 | * @sb: super block of file system to search | |
873 | * @head: the head of the list to search | |
874 | * @test: callback used for comparisons between inodes | |
875 | * @data: opaque data pointer to pass to @test | |
876 | * @wait: if true wait for the inode to be unlocked, if false do not | |
877 | * | |
878 | * ifind() searches for the inode specified by @data in the inode | |
879 | * cache. This is a generalized version of ifind_fast() for file systems where | |
880 | * the inode number is not sufficient for unique identification of an inode. | |
881 | * | |
882 | * If the inode is in the cache, the inode is returned with an incremented | |
883 | * reference count. | |
884 | * | |
885 | * Otherwise NULL is returned. | |
886 | * | |
887 | * Note, @test is called with the inode_lock held, so can't sleep. | |
888 | */ | |
889 | static struct inode *ifind(struct super_block *sb, | |
890 | struct hlist_head *head, int (*test)(struct inode *, void *), | |
891 | void *data, const int wait) | |
892 | { | |
893 | struct inode *inode; | |
894 | ||
895 | spin_lock(&inode_lock); | |
896 | inode = find_inode(sb, head, test, data); | |
897 | if (inode) { | |
898 | __iget(inode); | |
899 | spin_unlock(&inode_lock); | |
900 | if (likely(wait)) | |
901 | wait_on_inode(inode); | |
902 | return inode; | |
903 | } | |
904 | spin_unlock(&inode_lock); | |
905 | return NULL; | |
906 | } | |
907 | ||
908 | /** | |
909 | * ifind_fast - internal function, you want ilookup() or iget(). | |
910 | * @sb: super block of file system to search | |
911 | * @head: head of the list to search | |
912 | * @ino: inode number to search for | |
913 | * | |
914 | * ifind_fast() searches for the inode @ino in the inode cache. This is for | |
915 | * file systems where the inode number is sufficient for unique identification | |
916 | * of an inode. | |
917 | * | |
918 | * If the inode is in the cache, the inode is returned with an incremented | |
919 | * reference count. | |
920 | * | |
921 | * Otherwise NULL is returned. | |
922 | */ | |
923 | static struct inode *ifind_fast(struct super_block *sb, | |
924 | struct hlist_head *head, unsigned long ino) | |
925 | { | |
926 | struct inode *inode; | |
927 | ||
928 | spin_lock(&inode_lock); | |
929 | inode = find_inode_fast(sb, head, ino); | |
930 | if (inode) { | |
931 | __iget(inode); | |
932 | spin_unlock(&inode_lock); | |
933 | wait_on_inode(inode); | |
934 | return inode; | |
935 | } | |
936 | spin_unlock(&inode_lock); | |
937 | return NULL; | |
938 | } | |
939 | ||
940 | /** | |
941 | * ilookup5_nowait - search for an inode in the inode cache | |
942 | * @sb: super block of file system to search | |
943 | * @hashval: hash value (usually inode number) to search for | |
944 | * @test: callback used for comparisons between inodes | |
945 | * @data: opaque data pointer to pass to @test | |
946 | * | |
947 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
948 | * @data in the inode cache. This is a generalized version of ilookup() for | |
949 | * file systems where the inode number is not sufficient for unique | |
950 | * identification of an inode. | |
951 | * | |
952 | * If the inode is in the cache, the inode is returned with an incremented | |
953 | * reference count. Note, the inode lock is not waited upon so you have to be | |
954 | * very careful what you do with the returned inode. You probably should be | |
955 | * using ilookup5() instead. | |
956 | * | |
957 | * Otherwise NULL is returned. | |
958 | * | |
959 | * Note, @test is called with the inode_lock held, so can't sleep. | |
960 | */ | |
961 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, | |
962 | int (*test)(struct inode *, void *), void *data) | |
963 | { | |
964 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
965 | ||
966 | return ifind(sb, head, test, data, 0); | |
967 | } | |
968 | EXPORT_SYMBOL(ilookup5_nowait); | |
969 | ||
970 | /** | |
971 | * ilookup5 - search for an inode in the inode cache | |
972 | * @sb: super block of file system to search | |
973 | * @hashval: hash value (usually inode number) to search for | |
974 | * @test: callback used for comparisons between inodes | |
975 | * @data: opaque data pointer to pass to @test | |
976 | * | |
977 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
978 | * @data in the inode cache. This is a generalized version of ilookup() for | |
979 | * file systems where the inode number is not sufficient for unique | |
980 | * identification of an inode. | |
981 | * | |
982 | * If the inode is in the cache, the inode lock is waited upon and the inode is | |
983 | * returned with an incremented reference count. | |
984 | * | |
985 | * Otherwise NULL is returned. | |
986 | * | |
987 | * Note, @test is called with the inode_lock held, so can't sleep. | |
988 | */ | |
989 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
990 | int (*test)(struct inode *, void *), void *data) | |
991 | { | |
992 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
993 | ||
994 | return ifind(sb, head, test, data, 1); | |
995 | } | |
996 | EXPORT_SYMBOL(ilookup5); | |
997 | ||
998 | /** | |
999 | * ilookup - search for an inode in the inode cache | |
1000 | * @sb: super block of file system to search | |
1001 | * @ino: inode number to search for | |
1002 | * | |
1003 | * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache. | |
1004 | * This is for file systems where the inode number is sufficient for unique | |
1005 | * identification of an inode. | |
1006 | * | |
1007 | * If the inode is in the cache, the inode is returned with an incremented | |
1008 | * reference count. | |
1009 | * | |
1010 | * Otherwise NULL is returned. | |
1011 | */ | |
1012 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1013 | { | |
1014 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1015 | ||
1016 | return ifind_fast(sb, head, ino); | |
1017 | } | |
1018 | EXPORT_SYMBOL(ilookup); | |
1019 | ||
1020 | /** | |
1021 | * iget5_locked - obtain an inode from a mounted file system | |
1022 | * @sb: super block of file system | |
1023 | * @hashval: hash value (usually inode number) to get | |
1024 | * @test: callback used for comparisons between inodes | |
1025 | * @set: callback used to initialize a new struct inode | |
1026 | * @data: opaque data pointer to pass to @test and @set | |
1027 | * | |
1028 | * iget5_locked() uses ifind() to search for the inode specified by @hashval | |
1029 | * and @data in the inode cache and if present it is returned with an increased | |
1030 | * reference count. This is a generalized version of iget_locked() for file | |
1031 | * systems where the inode number is not sufficient for unique identification | |
1032 | * of an inode. | |
1033 | * | |
1034 | * If the inode is not in cache, get_new_inode() is called to allocate a new | |
1035 | * inode and this is returned locked, hashed, and with the I_NEW flag set. The | |
1036 | * file system gets to fill it in before unlocking it via unlock_new_inode(). | |
1037 | * | |
1038 | * Note both @test and @set are called with the inode_lock held, so can't sleep. | |
1039 | */ | |
1040 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, | |
1041 | int (*test)(struct inode *, void *), | |
1042 | int (*set)(struct inode *, void *), void *data) | |
1043 | { | |
1044 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1045 | struct inode *inode; | |
1046 | ||
1047 | inode = ifind(sb, head, test, data, 1); | |
1048 | if (inode) | |
1049 | return inode; | |
1050 | /* | |
1051 | * get_new_inode() will do the right thing, re-trying the search | |
1052 | * in case it had to block at any point. | |
1053 | */ | |
1054 | return get_new_inode(sb, head, test, set, data); | |
1055 | } | |
1056 | EXPORT_SYMBOL(iget5_locked); | |
1057 | ||
1058 | /** | |
1059 | * iget_locked - obtain an inode from a mounted file system | |
1060 | * @sb: super block of file system | |
1061 | * @ino: inode number to get | |
1062 | * | |
1063 | * iget_locked() uses ifind_fast() to search for the inode specified by @ino in | |
1064 | * the inode cache and if present it is returned with an increased reference | |
1065 | * count. This is for file systems where the inode number is sufficient for | |
1066 | * unique identification of an inode. | |
1067 | * | |
1068 | * If the inode is not in cache, get_new_inode_fast() is called to allocate a | |
1069 | * new inode and this is returned locked, hashed, and with the I_NEW flag set. | |
1070 | * The file system gets to fill it in before unlocking it via | |
1071 | * 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 | inode = ifind_fast(sb, head, ino); | |
1079 | if (inode) | |
1080 | return inode; | |
1081 | /* | |
1082 | * get_new_inode_fast() will do the right thing, re-trying the search | |
1083 | * in case it had to block at any point. | |
1084 | */ | |
1085 | return get_new_inode_fast(sb, head, ino); | |
1086 | } | |
1087 | EXPORT_SYMBOL(iget_locked); | |
1088 | ||
1089 | int insert_inode_locked(struct inode *inode) | |
1090 | { | |
1091 | struct super_block *sb = inode->i_sb; | |
1092 | ino_t ino = inode->i_ino; | |
1093 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1094 | ||
1095 | inode->i_state |= I_NEW; | |
1096 | while (1) { | |
1097 | struct hlist_node *node; | |
1098 | struct inode *old = NULL; | |
1099 | spin_lock(&inode_lock); | |
1100 | hlist_for_each_entry(old, node, head, i_hash) { | |
1101 | if (old->i_ino != ino) | |
1102 | continue; | |
1103 | if (old->i_sb != sb) | |
1104 | continue; | |
1105 | if (old->i_state & (I_FREEING|I_WILL_FREE)) | |
1106 | continue; | |
1107 | break; | |
1108 | } | |
1109 | if (likely(!node)) { | |
1110 | hlist_add_head(&inode->i_hash, head); | |
1111 | spin_unlock(&inode_lock); | |
1112 | return 0; | |
1113 | } | |
1114 | __iget(old); | |
1115 | spin_unlock(&inode_lock); | |
1116 | wait_on_inode(old); | |
1117 | if (unlikely(!hlist_unhashed(&old->i_hash))) { | |
1118 | iput(old); | |
1119 | return -EBUSY; | |
1120 | } | |
1121 | iput(old); | |
1122 | } | |
1123 | } | |
1124 | EXPORT_SYMBOL(insert_inode_locked); | |
1125 | ||
1126 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1127 | int (*test)(struct inode *, void *), void *data) | |
1128 | { | |
1129 | struct super_block *sb = inode->i_sb; | |
1130 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1131 | ||
1132 | inode->i_state |= I_NEW; | |
1133 | ||
1134 | while (1) { | |
1135 | struct hlist_node *node; | |
1136 | struct inode *old = NULL; | |
1137 | ||
1138 | spin_lock(&inode_lock); | |
1139 | hlist_for_each_entry(old, node, head, i_hash) { | |
1140 | if (old->i_sb != sb) | |
1141 | continue; | |
1142 | if (!test(old, data)) | |
1143 | continue; | |
1144 | if (old->i_state & (I_FREEING|I_WILL_FREE)) | |
1145 | continue; | |
1146 | break; | |
1147 | } | |
1148 | if (likely(!node)) { | |
1149 | hlist_add_head(&inode->i_hash, head); | |
1150 | spin_unlock(&inode_lock); | |
1151 | return 0; | |
1152 | } | |
1153 | __iget(old); | |
1154 | spin_unlock(&inode_lock); | |
1155 | wait_on_inode(old); | |
1156 | if (unlikely(!hlist_unhashed(&old->i_hash))) { | |
1157 | iput(old); | |
1158 | return -EBUSY; | |
1159 | } | |
1160 | iput(old); | |
1161 | } | |
1162 | } | |
1163 | EXPORT_SYMBOL(insert_inode_locked4); | |
1164 | ||
1165 | /** | |
1166 | * __insert_inode_hash - hash an inode | |
1167 | * @inode: unhashed inode | |
1168 | * @hashval: unsigned long value used to locate this object in the | |
1169 | * inode_hashtable. | |
1170 | * | |
1171 | * Add an inode to the inode hash for this superblock. | |
1172 | */ | |
1173 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
1174 | { | |
1175 | struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval); | |
1176 | spin_lock(&inode_lock); | |
1177 | hlist_add_head(&inode->i_hash, head); | |
1178 | spin_unlock(&inode_lock); | |
1179 | } | |
1180 | EXPORT_SYMBOL(__insert_inode_hash); | |
1181 | ||
1182 | /** | |
1183 | * remove_inode_hash - remove an inode from the hash | |
1184 | * @inode: inode to unhash | |
1185 | * | |
1186 | * Remove an inode from the superblock. | |
1187 | */ | |
1188 | void remove_inode_hash(struct inode *inode) | |
1189 | { | |
1190 | spin_lock(&inode_lock); | |
1191 | hlist_del_init(&inode->i_hash); | |
1192 | spin_unlock(&inode_lock); | |
1193 | } | |
1194 | EXPORT_SYMBOL(remove_inode_hash); | |
1195 | ||
1196 | /* | |
1197 | * Tell the filesystem that this inode is no longer of any interest and should | |
1198 | * be completely destroyed. | |
1199 | * | |
1200 | * We leave the inode in the inode hash table until *after* the filesystem's | |
1201 | * ->delete_inode completes. This ensures that an iget (such as nfsd might | |
1202 | * instigate) will always find up-to-date information either in the hash or on | |
1203 | * disk. | |
1204 | * | |
1205 | * I_FREEING is set so that no-one will take a new reference to the inode while | |
1206 | * it is being deleted. | |
1207 | */ | |
1208 | void generic_delete_inode(struct inode *inode) | |
1209 | { | |
1210 | list_del_init(&inode->i_list); | |
1211 | list_del_init(&inode->i_sb_list); | |
1212 | WARN_ON(inode->i_state & I_NEW); | |
1213 | inode->i_state |= I_FREEING; | |
1214 | inodes_stat.nr_inodes--; | |
1215 | spin_unlock(&inode_lock); | |
1216 | ||
1217 | evict(inode, 1); | |
1218 | ||
1219 | spin_lock(&inode_lock); | |
1220 | hlist_del_init(&inode->i_hash); | |
1221 | spin_unlock(&inode_lock); | |
1222 | wake_up_inode(inode); | |
1223 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); | |
1224 | destroy_inode(inode); | |
1225 | } | |
1226 | EXPORT_SYMBOL(generic_delete_inode); | |
1227 | ||
1228 | /** | |
1229 | * generic_detach_inode - remove inode from inode lists | |
1230 | * @inode: inode to remove | |
1231 | * | |
1232 | * Remove inode from inode lists, write it if it's dirty. This is just an | |
1233 | * internal VFS helper exported for hugetlbfs. Do not use! | |
1234 | * | |
1235 | * Returns 1 if inode should be completely destroyed. | |
1236 | */ | |
1237 | static int generic_detach_inode(struct inode *inode) | |
1238 | { | |
1239 | struct super_block *sb = inode->i_sb; | |
1240 | ||
1241 | if (!hlist_unhashed(&inode->i_hash)) { | |
1242 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) | |
1243 | list_move(&inode->i_list, &inode_unused); | |
1244 | inodes_stat.nr_unused++; | |
1245 | if (sb->s_flags & MS_ACTIVE) { | |
1246 | spin_unlock(&inode_lock); | |
1247 | return 0; | |
1248 | } | |
1249 | WARN_ON(inode->i_state & I_NEW); | |
1250 | inode->i_state |= I_WILL_FREE; | |
1251 | spin_unlock(&inode_lock); | |
1252 | write_inode_now(inode, 1); | |
1253 | spin_lock(&inode_lock); | |
1254 | WARN_ON(inode->i_state & I_NEW); | |
1255 | inode->i_state &= ~I_WILL_FREE; | |
1256 | inodes_stat.nr_unused--; | |
1257 | hlist_del_init(&inode->i_hash); | |
1258 | } | |
1259 | list_del_init(&inode->i_list); | |
1260 | list_del_init(&inode->i_sb_list); | |
1261 | WARN_ON(inode->i_state & I_NEW); | |
1262 | inode->i_state |= I_FREEING; | |
1263 | inodes_stat.nr_inodes--; | |
1264 | spin_unlock(&inode_lock); | |
1265 | return 1; | |
1266 | } | |
1267 | ||
1268 | static void generic_forget_inode(struct inode *inode) | |
1269 | { | |
1270 | if (!generic_detach_inode(inode)) | |
1271 | return; | |
1272 | evict(inode, 0); | |
1273 | wake_up_inode(inode); | |
1274 | destroy_inode(inode); | |
1275 | } | |
1276 | ||
1277 | /* | |
1278 | * Normal UNIX filesystem behaviour: delete the | |
1279 | * inode when the usage count drops to zero, and | |
1280 | * i_nlink is zero. | |
1281 | */ | |
1282 | void generic_drop_inode(struct inode *inode) | |
1283 | { | |
1284 | if (!inode->i_nlink) | |
1285 | generic_delete_inode(inode); | |
1286 | else | |
1287 | generic_forget_inode(inode); | |
1288 | } | |
1289 | EXPORT_SYMBOL_GPL(generic_drop_inode); | |
1290 | ||
1291 | /* | |
1292 | * Called when we're dropping the last reference | |
1293 | * to an inode. | |
1294 | * | |
1295 | * Call the FS "drop()" function, defaulting to | |
1296 | * the legacy UNIX filesystem behaviour.. | |
1297 | * | |
1298 | * NOTE! NOTE! NOTE! We're called with the inode lock | |
1299 | * held, and the drop function is supposed to release | |
1300 | * the lock! | |
1301 | */ | |
1302 | static inline void iput_final(struct inode *inode) | |
1303 | { | |
1304 | const struct super_operations *op = inode->i_sb->s_op; | |
1305 | void (*drop)(struct inode *) = generic_drop_inode; | |
1306 | ||
1307 | if (op && op->drop_inode) | |
1308 | drop = op->drop_inode; | |
1309 | drop(inode); | |
1310 | } | |
1311 | ||
1312 | /** | |
1313 | * iput - put an inode | |
1314 | * @inode: inode to put | |
1315 | * | |
1316 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1317 | * zero, the inode is then freed and may also be destroyed. | |
1318 | * | |
1319 | * Consequently, iput() can sleep. | |
1320 | */ | |
1321 | void iput(struct inode *inode) | |
1322 | { | |
1323 | if (inode) { | |
1324 | BUG_ON(inode->i_state & I_CLEAR); | |
1325 | ||
1326 | if (atomic_dec_and_lock(&inode->i_count, &inode_lock)) | |
1327 | iput_final(inode); | |
1328 | } | |
1329 | } | |
1330 | EXPORT_SYMBOL(iput); | |
1331 | ||
1332 | /** | |
1333 | * bmap - find a block number in a file | |
1334 | * @inode: inode of file | |
1335 | * @block: block to find | |
1336 | * | |
1337 | * Returns the block number on the device holding the inode that | |
1338 | * is the disk block number for the block of the file requested. | |
1339 | * That is, asked for block 4 of inode 1 the function will return the | |
1340 | * disk block relative to the disk start that holds that block of the | |
1341 | * file. | |
1342 | */ | |
1343 | sector_t bmap(struct inode *inode, sector_t block) | |
1344 | { | |
1345 | sector_t res = 0; | |
1346 | if (inode->i_mapping->a_ops->bmap) | |
1347 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1348 | return res; | |
1349 | } | |
1350 | EXPORT_SYMBOL(bmap); | |
1351 | ||
1352 | /* | |
1353 | * With relative atime, only update atime if the previous atime is | |
1354 | * earlier than either the ctime or mtime or if at least a day has | |
1355 | * passed since the last atime update. | |
1356 | */ | |
1357 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1358 | struct timespec now) | |
1359 | { | |
1360 | ||
1361 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1362 | return 1; | |
1363 | /* | |
1364 | * Is mtime younger than atime? If yes, update atime: | |
1365 | */ | |
1366 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1367 | return 1; | |
1368 | /* | |
1369 | * Is ctime younger than atime? If yes, update atime: | |
1370 | */ | |
1371 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1372 | return 1; | |
1373 | ||
1374 | /* | |
1375 | * Is the previous atime value older than a day? If yes, | |
1376 | * update atime: | |
1377 | */ | |
1378 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1379 | return 1; | |
1380 | /* | |
1381 | * Good, we can skip the atime update: | |
1382 | */ | |
1383 | return 0; | |
1384 | } | |
1385 | ||
1386 | /** | |
1387 | * touch_atime - update the access time | |
1388 | * @mnt: mount the inode is accessed on | |
1389 | * @dentry: dentry accessed | |
1390 | * | |
1391 | * Update the accessed time on an inode and mark it for writeback. | |
1392 | * This function automatically handles read only file systems and media, | |
1393 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1394 | */ | |
1395 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) | |
1396 | { | |
1397 | struct inode *inode = dentry->d_inode; | |
1398 | struct timespec now; | |
1399 | ||
1400 | if (inode->i_flags & S_NOATIME) | |
1401 | return; | |
1402 | if (IS_NOATIME(inode)) | |
1403 | return; | |
1404 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1405 | return; | |
1406 | ||
1407 | if (mnt->mnt_flags & MNT_NOATIME) | |
1408 | return; | |
1409 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1410 | return; | |
1411 | ||
1412 | now = current_fs_time(inode->i_sb); | |
1413 | ||
1414 | if (!relatime_need_update(mnt, inode, now)) | |
1415 | return; | |
1416 | ||
1417 | if (timespec_equal(&inode->i_atime, &now)) | |
1418 | return; | |
1419 | ||
1420 | if (mnt_want_write(mnt)) | |
1421 | return; | |
1422 | ||
1423 | inode->i_atime = now; | |
1424 | mark_inode_dirty_sync(inode); | |
1425 | mnt_drop_write(mnt); | |
1426 | } | |
1427 | EXPORT_SYMBOL(touch_atime); | |
1428 | ||
1429 | /** | |
1430 | * file_update_time - update mtime and ctime time | |
1431 | * @file: file accessed | |
1432 | * | |
1433 | * Update the mtime and ctime members of an inode and mark the inode | |
1434 | * for writeback. Note that this function is meant exclusively for | |
1435 | * usage in the file write path of filesystems, and filesystems may | |
1436 | * choose to explicitly ignore update via this function with the | |
1437 | * S_NOCMTIME inode flag, e.g. for network filesystem where these | |
1438 | * timestamps are handled by the server. | |
1439 | */ | |
1440 | ||
1441 | void file_update_time(struct file *file) | |
1442 | { | |
1443 | struct inode *inode = file->f_path.dentry->d_inode; | |
1444 | struct timespec now; | |
1445 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; | |
1446 | ||
1447 | /* First try to exhaust all avenues to not sync */ | |
1448 | if (IS_NOCMTIME(inode)) | |
1449 | return; | |
1450 | ||
1451 | now = current_fs_time(inode->i_sb); | |
1452 | if (!timespec_equal(&inode->i_mtime, &now)) | |
1453 | sync_it = S_MTIME; | |
1454 | ||
1455 | if (!timespec_equal(&inode->i_ctime, &now)) | |
1456 | sync_it |= S_CTIME; | |
1457 | ||
1458 | if (IS_I_VERSION(inode)) | |
1459 | sync_it |= S_VERSION; | |
1460 | ||
1461 | if (!sync_it) | |
1462 | return; | |
1463 | ||
1464 | /* Finally allowed to write? Takes lock. */ | |
1465 | if (mnt_want_write_file(file)) | |
1466 | return; | |
1467 | ||
1468 | /* Only change inode inside the lock region */ | |
1469 | if (sync_it & S_VERSION) | |
1470 | inode_inc_iversion(inode); | |
1471 | if (sync_it & S_CTIME) | |
1472 | inode->i_ctime = now; | |
1473 | if (sync_it & S_MTIME) | |
1474 | inode->i_mtime = now; | |
1475 | mark_inode_dirty_sync(inode); | |
1476 | mnt_drop_write(file->f_path.mnt); | |
1477 | } | |
1478 | EXPORT_SYMBOL(file_update_time); | |
1479 | ||
1480 | int inode_needs_sync(struct inode *inode) | |
1481 | { | |
1482 | if (IS_SYNC(inode)) | |
1483 | return 1; | |
1484 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1485 | return 1; | |
1486 | return 0; | |
1487 | } | |
1488 | EXPORT_SYMBOL(inode_needs_sync); | |
1489 | ||
1490 | int inode_wait(void *word) | |
1491 | { | |
1492 | schedule(); | |
1493 | return 0; | |
1494 | } | |
1495 | EXPORT_SYMBOL(inode_wait); | |
1496 | ||
1497 | /* | |
1498 | * If we try to find an inode in the inode hash while it is being | |
1499 | * deleted, we have to wait until the filesystem completes its | |
1500 | * deletion before reporting that it isn't found. This function waits | |
1501 | * until the deletion _might_ have completed. Callers are responsible | |
1502 | * to recheck inode state. | |
1503 | * | |
1504 | * It doesn't matter if I_NEW is not set initially, a call to | |
1505 | * wake_up_inode() after removing from the hash list will DTRT. | |
1506 | * | |
1507 | * This is called with inode_lock held. | |
1508 | */ | |
1509 | static void __wait_on_freeing_inode(struct inode *inode) | |
1510 | { | |
1511 | wait_queue_head_t *wq; | |
1512 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); | |
1513 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1514 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
1515 | spin_unlock(&inode_lock); | |
1516 | schedule(); | |
1517 | finish_wait(wq, &wait.wait); | |
1518 | spin_lock(&inode_lock); | |
1519 | } | |
1520 | ||
1521 | static __initdata unsigned long ihash_entries; | |
1522 | static int __init set_ihash_entries(char *str) | |
1523 | { | |
1524 | if (!str) | |
1525 | return 0; | |
1526 | ihash_entries = simple_strtoul(str, &str, 0); | |
1527 | return 1; | |
1528 | } | |
1529 | __setup("ihash_entries=", set_ihash_entries); | |
1530 | ||
1531 | /* | |
1532 | * Initialize the waitqueues and inode hash table. | |
1533 | */ | |
1534 | void __init inode_init_early(void) | |
1535 | { | |
1536 | int loop; | |
1537 | ||
1538 | /* If hashes are distributed across NUMA nodes, defer | |
1539 | * hash allocation until vmalloc space is available. | |
1540 | */ | |
1541 | if (hashdist) | |
1542 | return; | |
1543 | ||
1544 | inode_hashtable = | |
1545 | alloc_large_system_hash("Inode-cache", | |
1546 | sizeof(struct hlist_head), | |
1547 | ihash_entries, | |
1548 | 14, | |
1549 | HASH_EARLY, | |
1550 | &i_hash_shift, | |
1551 | &i_hash_mask, | |
1552 | 0); | |
1553 | ||
1554 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1555 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1556 | } | |
1557 | ||
1558 | void __init inode_init(void) | |
1559 | { | |
1560 | int loop; | |
1561 | ||
1562 | /* inode slab cache */ | |
1563 | inode_cachep = kmem_cache_create("inode_cache", | |
1564 | sizeof(struct inode), | |
1565 | 0, | |
1566 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1567 | SLAB_MEM_SPREAD), | |
1568 | init_once); | |
1569 | register_shrinker(&icache_shrinker); | |
1570 | ||
1571 | /* Hash may have been set up in inode_init_early */ | |
1572 | if (!hashdist) | |
1573 | return; | |
1574 | ||
1575 | inode_hashtable = | |
1576 | alloc_large_system_hash("Inode-cache", | |
1577 | sizeof(struct hlist_head), | |
1578 | ihash_entries, | |
1579 | 14, | |
1580 | 0, | |
1581 | &i_hash_shift, | |
1582 | &i_hash_mask, | |
1583 | 0); | |
1584 | ||
1585 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1586 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1587 | } | |
1588 | ||
1589 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1590 | { | |
1591 | inode->i_mode = mode; | |
1592 | if (S_ISCHR(mode)) { | |
1593 | inode->i_fop = &def_chr_fops; | |
1594 | inode->i_rdev = rdev; | |
1595 | } else if (S_ISBLK(mode)) { | |
1596 | inode->i_fop = &def_blk_fops; | |
1597 | inode->i_rdev = rdev; | |
1598 | } else if (S_ISFIFO(mode)) | |
1599 | inode->i_fop = &def_fifo_fops; | |
1600 | else if (S_ISSOCK(mode)) | |
1601 | inode->i_fop = &bad_sock_fops; | |
1602 | else | |
1603 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" | |
1604 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1605 | inode->i_ino); | |
1606 | } | |
1607 | EXPORT_SYMBOL(init_special_inode); | |
1608 | ||
1609 | /** | |
1610 | * Init uid,gid,mode for new inode according to posix standards | |
1611 | * @inode: New inode | |
1612 | * @dir: Directory inode | |
1613 | * @mode: mode of the new inode | |
1614 | */ | |
1615 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
1616 | mode_t mode) | |
1617 | { | |
1618 | inode->i_uid = current_fsuid(); | |
1619 | if (dir && dir->i_mode & S_ISGID) { | |
1620 | inode->i_gid = dir->i_gid; | |
1621 | if (S_ISDIR(mode)) | |
1622 | mode |= S_ISGID; | |
1623 | } else | |
1624 | inode->i_gid = current_fsgid(); | |
1625 | inode->i_mode = mode; | |
1626 | } | |
1627 | EXPORT_SYMBOL(inode_init_owner); |