]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/inode.c | |
3 | * | |
4 | * (C) 1997 Linus Torvalds | |
5 | */ | |
6 | ||
1da177e4 LT |
7 | #include <linux/fs.h> |
8 | #include <linux/mm.h> | |
9 | #include <linux/dcache.h> | |
10 | #include <linux/init.h> | |
11 | #include <linux/quotaops.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/writeback.h> | |
14 | #include <linux/module.h> | |
15 | #include <linux/backing-dev.h> | |
16 | #include <linux/wait.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> | |
0eeca283 | 23 | #include <linux/inotify.h> |
fc33a7bb | 24 | #include <linux/mount.h> |
1da177e4 LT |
25 | |
26 | /* | |
27 | * This is needed for the following functions: | |
28 | * - inode_has_buffers | |
29 | * - invalidate_inode_buffers | |
1da177e4 LT |
30 | * - invalidate_bdev |
31 | * | |
32 | * FIXME: remove all knowledge of the buffer layer from this file | |
33 | */ | |
34 | #include <linux/buffer_head.h> | |
35 | ||
36 | /* | |
37 | * New inode.c implementation. | |
38 | * | |
39 | * This implementation has the basic premise of trying | |
40 | * to be extremely low-overhead and SMP-safe, yet be | |
41 | * simple enough to be "obviously correct". | |
42 | * | |
43 | * Famous last words. | |
44 | */ | |
45 | ||
46 | /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */ | |
47 | ||
48 | /* #define INODE_PARANOIA 1 */ | |
49 | /* #define INODE_DEBUG 1 */ | |
50 | ||
51 | /* | |
52 | * Inode lookup is no longer as critical as it used to be: | |
53 | * most of the lookups are going to be through the dcache. | |
54 | */ | |
55 | #define I_HASHBITS i_hash_shift | |
56 | #define I_HASHMASK i_hash_mask | |
57 | ||
fa3536cc ED |
58 | static unsigned int i_hash_mask __read_mostly; |
59 | static unsigned int i_hash_shift __read_mostly; | |
1da177e4 LT |
60 | |
61 | /* | |
62 | * Each inode can be on two separate lists. One is | |
63 | * the hash list of the inode, used for lookups. The | |
64 | * other linked list is the "type" list: | |
65 | * "in_use" - valid inode, i_count > 0, i_nlink > 0 | |
66 | * "dirty" - as "in_use" but also dirty | |
67 | * "unused" - valid inode, i_count = 0 | |
68 | * | |
69 | * A "dirty" list is maintained for each super block, | |
70 | * allowing for low-overhead inode sync() operations. | |
71 | */ | |
72 | ||
73 | LIST_HEAD(inode_in_use); | |
74 | LIST_HEAD(inode_unused); | |
fa3536cc | 75 | static struct hlist_head *inode_hashtable __read_mostly; |
1da177e4 LT |
76 | |
77 | /* | |
78 | * A simple spinlock to protect the list manipulations. | |
79 | * | |
80 | * NOTE! You also have to own the lock if you change | |
81 | * the i_state of an inode while it is in use.. | |
82 | */ | |
83 | DEFINE_SPINLOCK(inode_lock); | |
84 | ||
85 | /* | |
f24075bd | 86 | * iprune_mutex provides exclusion between the kswapd or try_to_free_pages |
1da177e4 LT |
87 | * icache shrinking path, and the umount path. Without this exclusion, |
88 | * by the time prune_icache calls iput for the inode whose pages it has | |
89 | * been invalidating, or by the time it calls clear_inode & destroy_inode | |
90 | * from its final dispose_list, the struct super_block they refer to | |
91 | * (for inode->i_sb->s_op) may already have been freed and reused. | |
92 | */ | |
bdfc3266 | 93 | static DEFINE_MUTEX(iprune_mutex); |
1da177e4 LT |
94 | |
95 | /* | |
96 | * Statistics gathering.. | |
97 | */ | |
98 | struct inodes_stat_t inodes_stat; | |
99 | ||
e18b890b | 100 | static struct kmem_cache * inode_cachep __read_mostly; |
1da177e4 | 101 | |
1c0eeaf5 JE |
102 | static void wake_up_inode(struct inode *inode) |
103 | { | |
104 | /* | |
105 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
106 | */ | |
107 | smp_mb(); | |
108 | wake_up_bit(&inode->i_state, __I_LOCK); | |
109 | } | |
110 | ||
2cb1599f DC |
111 | /** |
112 | * inode_init_always - perform inode structure intialisation | |
113 | * @sb - superblock inode belongs to. | |
114 | * @inode - inode to initialise | |
115 | * | |
116 | * These are initializations that need to be done on every inode | |
117 | * allocation as the fields are not initialised by slab allocation. | |
118 | */ | |
119 | struct inode *inode_init_always(struct super_block *sb, struct inode *inode) | |
1da177e4 | 120 | { |
f5e54d6e | 121 | static const struct address_space_operations empty_aops; |
1da177e4 | 122 | static struct inode_operations empty_iops; |
99ac48f5 | 123 | static const struct file_operations empty_fops; |
1da177e4 | 124 | |
2cb1599f DC |
125 | struct address_space * const mapping = &inode->i_data; |
126 | ||
127 | inode->i_sb = sb; | |
128 | inode->i_blkbits = sb->s_blocksize_bits; | |
129 | inode->i_flags = 0; | |
130 | atomic_set(&inode->i_count, 1); | |
131 | inode->i_op = &empty_iops; | |
132 | inode->i_fop = &empty_fops; | |
133 | inode->i_nlink = 1; | |
134 | atomic_set(&inode->i_writecount, 0); | |
135 | inode->i_size = 0; | |
136 | inode->i_blocks = 0; | |
137 | inode->i_bytes = 0; | |
138 | inode->i_generation = 0; | |
1da177e4 | 139 | #ifdef CONFIG_QUOTA |
2cb1599f | 140 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
1da177e4 | 141 | #endif |
2cb1599f DC |
142 | inode->i_pipe = NULL; |
143 | inode->i_bdev = NULL; | |
144 | inode->i_cdev = NULL; | |
145 | inode->i_rdev = 0; | |
146 | inode->dirtied_when = 0; | |
147 | if (security_inode_alloc(inode)) { | |
148 | if (inode->i_sb->s_op->destroy_inode) | |
149 | inode->i_sb->s_op->destroy_inode(inode); | |
150 | else | |
151 | kmem_cache_free(inode_cachep, (inode)); | |
152 | return NULL; | |
1da177e4 | 153 | } |
2cb1599f DC |
154 | |
155 | spin_lock_init(&inode->i_lock); | |
156 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
157 | ||
158 | mutex_init(&inode->i_mutex); | |
159 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
160 | ||
161 | init_rwsem(&inode->i_alloc_sem); | |
162 | lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key); | |
163 | ||
164 | mapping->a_ops = &empty_aops; | |
165 | mapping->host = inode; | |
166 | mapping->flags = 0; | |
167 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_PAGECACHE); | |
168 | mapping->assoc_mapping = NULL; | |
169 | mapping->backing_dev_info = &default_backing_dev_info; | |
170 | mapping->writeback_index = 0; | |
171 | ||
172 | /* | |
173 | * If the block_device provides a backing_dev_info for client | |
174 | * inodes then use that. Otherwise the inode share the bdev's | |
175 | * backing_dev_info. | |
176 | */ | |
177 | if (sb->s_bdev) { | |
178 | struct backing_dev_info *bdi; | |
179 | ||
180 | bdi = sb->s_bdev->bd_inode_backing_dev_info; | |
181 | if (!bdi) | |
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 | ||
1da177e4 LT |
188 | return inode; |
189 | } | |
2cb1599f DC |
190 | EXPORT_SYMBOL(inode_init_always); |
191 | ||
192 | static struct inode *alloc_inode(struct super_block *sb) | |
193 | { | |
194 | struct inode *inode; | |
195 | ||
196 | if (sb->s_op->alloc_inode) | |
197 | inode = sb->s_op->alloc_inode(sb); | |
198 | else | |
199 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
200 | ||
201 | if (inode) | |
202 | return inode_init_always(sb, inode); | |
203 | return NULL; | |
204 | } | |
1da177e4 LT |
205 | |
206 | void destroy_inode(struct inode *inode) | |
207 | { | |
b7542f8c | 208 | BUG_ON(inode_has_buffers(inode)); |
1da177e4 LT |
209 | security_inode_free(inode); |
210 | if (inode->i_sb->s_op->destroy_inode) | |
211 | inode->i_sb->s_op->destroy_inode(inode); | |
212 | else | |
213 | kmem_cache_free(inode_cachep, (inode)); | |
214 | } | |
215 | ||
216 | ||
217 | /* | |
218 | * These are initializations that only need to be done | |
219 | * once, because the fields are idempotent across use | |
220 | * of the inode, so let the slab aware of that. | |
221 | */ | |
222 | void inode_init_once(struct inode *inode) | |
223 | { | |
224 | memset(inode, 0, sizeof(*inode)); | |
225 | INIT_HLIST_NODE(&inode->i_hash); | |
226 | INIT_LIST_HEAD(&inode->i_dentry); | |
227 | INIT_LIST_HEAD(&inode->i_devices); | |
1da177e4 | 228 | INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC); |
19fd6231 | 229 | spin_lock_init(&inode->i_data.tree_lock); |
1da177e4 LT |
230 | spin_lock_init(&inode->i_data.i_mmap_lock); |
231 | INIT_LIST_HEAD(&inode->i_data.private_list); | |
232 | spin_lock_init(&inode->i_data.private_lock); | |
233 | INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap); | |
234 | INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear); | |
1da177e4 | 235 | i_size_ordered_init(inode); |
0eeca283 RL |
236 | #ifdef CONFIG_INOTIFY |
237 | INIT_LIST_HEAD(&inode->inotify_watches); | |
d4f9af9d | 238 | mutex_init(&inode->inotify_mutex); |
0eeca283 | 239 | #endif |
1da177e4 LT |
240 | } |
241 | ||
242 | EXPORT_SYMBOL(inode_init_once); | |
243 | ||
51cc5068 | 244 | static void init_once(void *foo) |
1da177e4 LT |
245 | { |
246 | struct inode * inode = (struct inode *) foo; | |
247 | ||
a35afb83 | 248 | inode_init_once(inode); |
1da177e4 LT |
249 | } |
250 | ||
251 | /* | |
252 | * inode_lock must be held | |
253 | */ | |
254 | void __iget(struct inode * inode) | |
255 | { | |
256 | if (atomic_read(&inode->i_count)) { | |
257 | atomic_inc(&inode->i_count); | |
258 | return; | |
259 | } | |
260 | atomic_inc(&inode->i_count); | |
1c0eeaf5 | 261 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) |
1da177e4 LT |
262 | list_move(&inode->i_list, &inode_in_use); |
263 | inodes_stat.nr_unused--; | |
264 | } | |
265 | ||
266 | /** | |
267 | * clear_inode - clear an inode | |
268 | * @inode: inode to clear | |
269 | * | |
270 | * This is called by the filesystem to tell us | |
271 | * that the inode is no longer useful. We just | |
272 | * terminate it with extreme prejudice. | |
273 | */ | |
274 | void clear_inode(struct inode *inode) | |
275 | { | |
276 | might_sleep(); | |
277 | invalidate_inode_buffers(inode); | |
278 | ||
b7542f8c ES |
279 | BUG_ON(inode->i_data.nrpages); |
280 | BUG_ON(!(inode->i_state & I_FREEING)); | |
281 | BUG_ON(inode->i_state & I_CLEAR); | |
1c0eeaf5 | 282 | inode_sync_wait(inode); |
1da177e4 | 283 | DQUOT_DROP(inode); |
acb0c854 | 284 | if (inode->i_sb->s_op->clear_inode) |
1da177e4 | 285 | inode->i_sb->s_op->clear_inode(inode); |
eaf796e7 | 286 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
1da177e4 | 287 | bd_forget(inode); |
577c4eb0 | 288 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) |
1da177e4 LT |
289 | cd_forget(inode); |
290 | inode->i_state = I_CLEAR; | |
291 | } | |
292 | ||
293 | EXPORT_SYMBOL(clear_inode); | |
294 | ||
295 | /* | |
296 | * dispose_list - dispose of the contents of a local list | |
297 | * @head: the head of the list to free | |
298 | * | |
299 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
300 | * need to worry about list corruption and SMP locks. | |
301 | */ | |
302 | static void dispose_list(struct list_head *head) | |
303 | { | |
304 | int nr_disposed = 0; | |
305 | ||
306 | while (!list_empty(head)) { | |
307 | struct inode *inode; | |
308 | ||
b5e61818 | 309 | inode = list_first_entry(head, struct inode, i_list); |
1da177e4 LT |
310 | list_del(&inode->i_list); |
311 | ||
312 | if (inode->i_data.nrpages) | |
313 | truncate_inode_pages(&inode->i_data, 0); | |
314 | clear_inode(inode); | |
4120db47 AB |
315 | |
316 | spin_lock(&inode_lock); | |
317 | hlist_del_init(&inode->i_hash); | |
318 | list_del_init(&inode->i_sb_list); | |
319 | spin_unlock(&inode_lock); | |
320 | ||
321 | wake_up_inode(inode); | |
1da177e4 LT |
322 | destroy_inode(inode); |
323 | nr_disposed++; | |
324 | } | |
325 | spin_lock(&inode_lock); | |
326 | inodes_stat.nr_inodes -= nr_disposed; | |
327 | spin_unlock(&inode_lock); | |
328 | } | |
329 | ||
330 | /* | |
331 | * Invalidate all inodes for a device. | |
332 | */ | |
333 | static int invalidate_list(struct list_head *head, struct list_head *dispose) | |
334 | { | |
335 | struct list_head *next; | |
336 | int busy = 0, count = 0; | |
337 | ||
338 | next = head->next; | |
339 | for (;;) { | |
340 | struct list_head * tmp = next; | |
341 | struct inode * inode; | |
342 | ||
343 | /* | |
344 | * We can reschedule here without worrying about the list's | |
345 | * consistency because the per-sb list of inodes must not | |
f24075bd | 346 | * change during umount anymore, and because iprune_mutex keeps |
1da177e4 LT |
347 | * shrink_icache_memory() away. |
348 | */ | |
349 | cond_resched_lock(&inode_lock); | |
350 | ||
351 | next = next->next; | |
352 | if (tmp == head) | |
353 | break; | |
354 | inode = list_entry(tmp, struct inode, i_sb_list); | |
355 | invalidate_inode_buffers(inode); | |
356 | if (!atomic_read(&inode->i_count)) { | |
1da177e4 LT |
357 | list_move(&inode->i_list, dispose); |
358 | inode->i_state |= I_FREEING; | |
359 | count++; | |
360 | continue; | |
361 | } | |
362 | busy = 1; | |
363 | } | |
364 | /* only unused inodes may be cached with i_count zero */ | |
365 | inodes_stat.nr_unused -= count; | |
366 | return busy; | |
367 | } | |
368 | ||
1da177e4 LT |
369 | /** |
370 | * invalidate_inodes - discard the inodes on a device | |
371 | * @sb: superblock | |
372 | * | |
373 | * Discard all of the inodes for a given superblock. If the discard | |
374 | * fails because there are busy inodes then a non zero value is returned. | |
375 | * If the discard is successful all the inodes have been discarded. | |
376 | */ | |
377 | int invalidate_inodes(struct super_block * sb) | |
378 | { | |
379 | int busy; | |
380 | LIST_HEAD(throw_away); | |
381 | ||
f24075bd | 382 | mutex_lock(&iprune_mutex); |
1da177e4 | 383 | spin_lock(&inode_lock); |
0eeca283 | 384 | inotify_unmount_inodes(&sb->s_inodes); |
1da177e4 LT |
385 | busy = invalidate_list(&sb->s_inodes, &throw_away); |
386 | spin_unlock(&inode_lock); | |
387 | ||
388 | dispose_list(&throw_away); | |
f24075bd | 389 | mutex_unlock(&iprune_mutex); |
1da177e4 LT |
390 | |
391 | return busy; | |
392 | } | |
393 | ||
394 | EXPORT_SYMBOL(invalidate_inodes); | |
1da177e4 LT |
395 | |
396 | static int can_unuse(struct inode *inode) | |
397 | { | |
398 | if (inode->i_state) | |
399 | return 0; | |
400 | if (inode_has_buffers(inode)) | |
401 | return 0; | |
402 | if (atomic_read(&inode->i_count)) | |
403 | return 0; | |
404 | if (inode->i_data.nrpages) | |
405 | return 0; | |
406 | return 1; | |
407 | } | |
408 | ||
409 | /* | |
410 | * Scan `goal' inodes on the unused list for freeable ones. They are moved to | |
411 | * a temporary list and then are freed outside inode_lock by dispose_list(). | |
412 | * | |
413 | * Any inodes which are pinned purely because of attached pagecache have their | |
414 | * pagecache removed. We expect the final iput() on that inode to add it to | |
415 | * the front of the inode_unused list. So look for it there and if the | |
416 | * inode is still freeable, proceed. The right inode is found 99.9% of the | |
417 | * time in testing on a 4-way. | |
418 | * | |
419 | * If the inode has metadata buffers attached to mapping->private_list then | |
420 | * try to remove them. | |
421 | */ | |
422 | static void prune_icache(int nr_to_scan) | |
423 | { | |
424 | LIST_HEAD(freeable); | |
425 | int nr_pruned = 0; | |
426 | int nr_scanned; | |
427 | unsigned long reap = 0; | |
428 | ||
f24075bd | 429 | mutex_lock(&iprune_mutex); |
1da177e4 LT |
430 | spin_lock(&inode_lock); |
431 | for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) { | |
432 | struct inode *inode; | |
433 | ||
434 | if (list_empty(&inode_unused)) | |
435 | break; | |
436 | ||
437 | inode = list_entry(inode_unused.prev, struct inode, i_list); | |
438 | ||
439 | if (inode->i_state || atomic_read(&inode->i_count)) { | |
440 | list_move(&inode->i_list, &inode_unused); | |
441 | continue; | |
442 | } | |
443 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
444 | __iget(inode); | |
445 | spin_unlock(&inode_lock); | |
446 | if (remove_inode_buffers(inode)) | |
fc0ecff6 AM |
447 | reap += invalidate_mapping_pages(&inode->i_data, |
448 | 0, -1); | |
1da177e4 LT |
449 | iput(inode); |
450 | spin_lock(&inode_lock); | |
451 | ||
452 | if (inode != list_entry(inode_unused.next, | |
453 | struct inode, i_list)) | |
454 | continue; /* wrong inode or list_empty */ | |
455 | if (!can_unuse(inode)) | |
456 | continue; | |
457 | } | |
1da177e4 LT |
458 | list_move(&inode->i_list, &freeable); |
459 | inode->i_state |= I_FREEING; | |
460 | nr_pruned++; | |
461 | } | |
462 | inodes_stat.nr_unused -= nr_pruned; | |
f8891e5e CL |
463 | if (current_is_kswapd()) |
464 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
465 | else | |
466 | __count_vm_events(PGINODESTEAL, reap); | |
1da177e4 LT |
467 | spin_unlock(&inode_lock); |
468 | ||
469 | dispose_list(&freeable); | |
f24075bd | 470 | mutex_unlock(&iprune_mutex); |
1da177e4 LT |
471 | } |
472 | ||
473 | /* | |
474 | * shrink_icache_memory() will attempt to reclaim some unused inodes. Here, | |
475 | * "unused" means that no dentries are referring to the inodes: the files are | |
476 | * not open and the dcache references to those inodes have already been | |
477 | * reclaimed. | |
478 | * | |
479 | * This function is passed the number of inodes to scan, and it returns the | |
480 | * total number of remaining possibly-reclaimable inodes. | |
481 | */ | |
27496a8c | 482 | static int shrink_icache_memory(int nr, gfp_t gfp_mask) |
1da177e4 LT |
483 | { |
484 | if (nr) { | |
485 | /* | |
486 | * Nasty deadlock avoidance. We may hold various FS locks, | |
487 | * and we don't want to recurse into the FS that called us | |
488 | * in clear_inode() and friends.. | |
489 | */ | |
490 | if (!(gfp_mask & __GFP_FS)) | |
491 | return -1; | |
492 | prune_icache(nr); | |
493 | } | |
494 | return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; | |
495 | } | |
496 | ||
8e1f936b RR |
497 | static struct shrinker icache_shrinker = { |
498 | .shrink = shrink_icache_memory, | |
499 | .seeks = DEFAULT_SEEKS, | |
500 | }; | |
501 | ||
1da177e4 LT |
502 | static void __wait_on_freeing_inode(struct inode *inode); |
503 | /* | |
504 | * Called with the inode lock held. | |
505 | * NOTE: we are not increasing the inode-refcount, you must call __iget() | |
506 | * by hand after calling find_inode now! This simplifies iunique and won't | |
507 | * add any additional branch in the common code. | |
508 | */ | |
509 | static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data) | |
510 | { | |
511 | struct hlist_node *node; | |
512 | struct inode * inode = NULL; | |
513 | ||
514 | repeat: | |
c5c8be3c | 515 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
516 | if (inode->i_sb != sb) |
517 | continue; | |
518 | if (!test(inode, data)) | |
519 | continue; | |
991114c6 | 520 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) { |
1da177e4 LT |
521 | __wait_on_freeing_inode(inode); |
522 | goto repeat; | |
523 | } | |
524 | break; | |
525 | } | |
526 | return node ? inode : NULL; | |
527 | } | |
528 | ||
529 | /* | |
530 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
531 | * iget_locked for details. | |
532 | */ | |
533 | static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino) | |
534 | { | |
535 | struct hlist_node *node; | |
536 | struct inode * inode = NULL; | |
537 | ||
538 | repeat: | |
c5c8be3c | 539 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
540 | if (inode->i_ino != ino) |
541 | continue; | |
542 | if (inode->i_sb != sb) | |
543 | continue; | |
991114c6 | 544 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) { |
1da177e4 LT |
545 | __wait_on_freeing_inode(inode); |
546 | goto repeat; | |
547 | } | |
548 | break; | |
549 | } | |
550 | return node ? inode : NULL; | |
551 | } | |
552 | ||
553 | /** | |
554 | * new_inode - obtain an inode | |
555 | * @sb: superblock | |
556 | * | |
769848c0 MG |
557 | * Allocates a new inode for given superblock. The default gfp_mask |
558 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_PAGECACHE. | |
559 | * If HIGHMEM pages are unsuitable or it is known that pages allocated | |
560 | * for the page cache are not reclaimable or migratable, | |
561 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
562 | * newly created inode's mapping | |
563 | * | |
1da177e4 LT |
564 | */ |
565 | struct inode *new_inode(struct super_block *sb) | |
566 | { | |
866b04fc JL |
567 | /* |
568 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
569 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
570 | * here to attempt to avoid that. | |
571 | */ | |
572 | static unsigned int last_ino; | |
1da177e4 LT |
573 | struct inode * inode; |
574 | ||
575 | spin_lock_prefetch(&inode_lock); | |
576 | ||
577 | inode = alloc_inode(sb); | |
578 | if (inode) { | |
579 | spin_lock(&inode_lock); | |
580 | inodes_stat.nr_inodes++; | |
581 | list_add(&inode->i_list, &inode_in_use); | |
582 | list_add(&inode->i_sb_list, &sb->s_inodes); | |
583 | inode->i_ino = ++last_ino; | |
584 | inode->i_state = 0; | |
585 | spin_unlock(&inode_lock); | |
586 | } | |
587 | return inode; | |
588 | } | |
589 | ||
590 | EXPORT_SYMBOL(new_inode); | |
591 | ||
592 | void unlock_new_inode(struct inode *inode) | |
593 | { | |
14358e6d | 594 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
1e89a5e1 PZ |
595 | if (inode->i_mode & S_IFDIR) { |
596 | struct file_system_type *type = inode->i_sb->s_type; | |
597 | ||
598 | /* | |
599 | * ensure nobody is actually holding i_mutex | |
600 | */ | |
601 | mutex_destroy(&inode->i_mutex); | |
602 | mutex_init(&inode->i_mutex); | |
14358e6d | 603 | lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key); |
1e89a5e1 | 604 | } |
14358e6d | 605 | #endif |
1da177e4 LT |
606 | /* |
607 | * This is special! We do not need the spinlock | |
608 | * when clearing I_LOCK, because we're guaranteed | |
609 | * that nobody else tries to do anything about the | |
610 | * state of the inode when it is locked, as we | |
611 | * just created it (so there can be no old holders | |
612 | * that haven't tested I_LOCK). | |
613 | */ | |
614 | inode->i_state &= ~(I_LOCK|I_NEW); | |
615 | wake_up_inode(inode); | |
616 | } | |
617 | ||
618 | EXPORT_SYMBOL(unlock_new_inode); | |
619 | ||
620 | /* | |
621 | * This is called without the inode lock held.. Be careful. | |
622 | * | |
623 | * We no longer cache the sb_flags in i_flags - see fs.h | |
624 | * -- rmk@arm.uk.linux.org | |
625 | */ | |
626 | static struct inode * get_new_inode(struct super_block *sb, struct hlist_head *head, int (*test)(struct inode *, void *), int (*set)(struct inode *, void *), void *data) | |
627 | { | |
628 | struct inode * inode; | |
629 | ||
630 | inode = alloc_inode(sb); | |
631 | if (inode) { | |
632 | struct inode * old; | |
633 | ||
634 | spin_lock(&inode_lock); | |
635 | /* We released the lock, so.. */ | |
636 | old = find_inode(sb, head, test, data); | |
637 | if (!old) { | |
638 | if (set(inode, data)) | |
639 | goto set_failed; | |
640 | ||
641 | inodes_stat.nr_inodes++; | |
642 | list_add(&inode->i_list, &inode_in_use); | |
643 | list_add(&inode->i_sb_list, &sb->s_inodes); | |
644 | hlist_add_head(&inode->i_hash, head); | |
645 | inode->i_state = I_LOCK|I_NEW; | |
646 | spin_unlock(&inode_lock); | |
647 | ||
648 | /* Return the locked inode with I_NEW set, the | |
649 | * caller is responsible for filling in the contents | |
650 | */ | |
651 | return inode; | |
652 | } | |
653 | ||
654 | /* | |
655 | * Uhhuh, somebody else created the same inode under | |
656 | * us. Use the old inode instead of the one we just | |
657 | * allocated. | |
658 | */ | |
659 | __iget(old); | |
660 | spin_unlock(&inode_lock); | |
661 | destroy_inode(inode); | |
662 | inode = old; | |
663 | wait_on_inode(inode); | |
664 | } | |
665 | return inode; | |
666 | ||
667 | set_failed: | |
668 | spin_unlock(&inode_lock); | |
669 | destroy_inode(inode); | |
670 | return NULL; | |
671 | } | |
672 | ||
673 | /* | |
674 | * get_new_inode_fast is the fast path version of get_new_inode, see the | |
675 | * comment at iget_locked for details. | |
676 | */ | |
677 | static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino) | |
678 | { | |
679 | struct inode * inode; | |
680 | ||
681 | inode = alloc_inode(sb); | |
682 | if (inode) { | |
683 | struct inode * old; | |
684 | ||
685 | spin_lock(&inode_lock); | |
686 | /* We released the lock, so.. */ | |
687 | old = find_inode_fast(sb, head, ino); | |
688 | if (!old) { | |
689 | inode->i_ino = ino; | |
690 | inodes_stat.nr_inodes++; | |
691 | list_add(&inode->i_list, &inode_in_use); | |
692 | list_add(&inode->i_sb_list, &sb->s_inodes); | |
693 | hlist_add_head(&inode->i_hash, head); | |
694 | inode->i_state = I_LOCK|I_NEW; | |
695 | spin_unlock(&inode_lock); | |
696 | ||
697 | /* Return the locked inode with I_NEW set, the | |
698 | * caller is responsible for filling in the contents | |
699 | */ | |
700 | return inode; | |
701 | } | |
702 | ||
703 | /* | |
704 | * Uhhuh, somebody else created the same inode under | |
705 | * us. Use the old inode instead of the one we just | |
706 | * allocated. | |
707 | */ | |
708 | __iget(old); | |
709 | spin_unlock(&inode_lock); | |
710 | destroy_inode(inode); | |
711 | inode = old; | |
712 | wait_on_inode(inode); | |
713 | } | |
714 | return inode; | |
715 | } | |
716 | ||
ed97bd37 | 717 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
1da177e4 LT |
718 | { |
719 | unsigned long tmp; | |
720 | ||
721 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
722 | L1_CACHE_BYTES; | |
723 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS); | |
724 | return tmp & I_HASHMASK; | |
725 | } | |
726 | ||
727 | /** | |
728 | * iunique - get a unique inode number | |
729 | * @sb: superblock | |
730 | * @max_reserved: highest reserved inode number | |
731 | * | |
732 | * Obtain an inode number that is unique on the system for a given | |
733 | * superblock. This is used by file systems that have no natural | |
734 | * permanent inode numbering system. An inode number is returned that | |
735 | * is higher than the reserved limit but unique. | |
736 | * | |
737 | * BUGS: | |
738 | * With a large number of inodes live on the file system this function | |
739 | * currently becomes quite slow. | |
740 | */ | |
741 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
742 | { | |
866b04fc JL |
743 | /* |
744 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
745 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
746 | * here to attempt to avoid that. | |
747 | */ | |
748 | static unsigned int counter; | |
1da177e4 | 749 | struct inode *inode; |
3361c7be | 750 | struct hlist_head *head; |
1da177e4 | 751 | ino_t res; |
3361c7be | 752 | |
1da177e4 | 753 | spin_lock(&inode_lock); |
3361c7be JL |
754 | do { |
755 | if (counter <= max_reserved) | |
756 | counter = max_reserved + 1; | |
1da177e4 | 757 | res = counter++; |
3361c7be | 758 | head = inode_hashtable + hash(sb, res); |
1da177e4 | 759 | inode = find_inode_fast(sb, head, res); |
3361c7be JL |
760 | } while (inode != NULL); |
761 | spin_unlock(&inode_lock); | |
1da177e4 | 762 | |
3361c7be JL |
763 | return res; |
764 | } | |
1da177e4 LT |
765 | EXPORT_SYMBOL(iunique); |
766 | ||
767 | struct inode *igrab(struct inode *inode) | |
768 | { | |
769 | spin_lock(&inode_lock); | |
4a3b0a49 | 770 | if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))) |
1da177e4 LT |
771 | __iget(inode); |
772 | else | |
773 | /* | |
774 | * Handle the case where s_op->clear_inode is not been | |
775 | * called yet, and somebody is calling igrab | |
776 | * while the inode is getting freed. | |
777 | */ | |
778 | inode = NULL; | |
779 | spin_unlock(&inode_lock); | |
780 | return inode; | |
781 | } | |
782 | ||
783 | EXPORT_SYMBOL(igrab); | |
784 | ||
785 | /** | |
786 | * ifind - internal function, you want ilookup5() or iget5(). | |
787 | * @sb: super block of file system to search | |
788 | * @head: the head of the list to search | |
789 | * @test: callback used for comparisons between inodes | |
790 | * @data: opaque data pointer to pass to @test | |
88bd5121 | 791 | * @wait: if true wait for the inode to be unlocked, if false do not |
1da177e4 LT |
792 | * |
793 | * ifind() searches for the inode specified by @data in the inode | |
794 | * cache. This is a generalized version of ifind_fast() for file systems where | |
795 | * the inode number is not sufficient for unique identification of an inode. | |
796 | * | |
797 | * If the inode is in the cache, the inode is returned with an incremented | |
798 | * reference count. | |
799 | * | |
800 | * Otherwise NULL is returned. | |
801 | * | |
802 | * Note, @test is called with the inode_lock held, so can't sleep. | |
803 | */ | |
5d2bea45 | 804 | static struct inode *ifind(struct super_block *sb, |
1da177e4 | 805 | struct hlist_head *head, int (*test)(struct inode *, void *), |
88bd5121 | 806 | void *data, const int wait) |
1da177e4 LT |
807 | { |
808 | struct inode *inode; | |
809 | ||
810 | spin_lock(&inode_lock); | |
811 | inode = find_inode(sb, head, test, data); | |
812 | if (inode) { | |
813 | __iget(inode); | |
814 | spin_unlock(&inode_lock); | |
88bd5121 AA |
815 | if (likely(wait)) |
816 | wait_on_inode(inode); | |
1da177e4 LT |
817 | return inode; |
818 | } | |
819 | spin_unlock(&inode_lock); | |
820 | return NULL; | |
821 | } | |
822 | ||
823 | /** | |
824 | * ifind_fast - internal function, you want ilookup() or iget(). | |
825 | * @sb: super block of file system to search | |
826 | * @head: head of the list to search | |
827 | * @ino: inode number to search for | |
828 | * | |
829 | * ifind_fast() searches for the inode @ino in the inode cache. This is for | |
830 | * file systems where the inode number is sufficient for unique identification | |
831 | * of an inode. | |
832 | * | |
833 | * If the inode is in the cache, the inode is returned with an incremented | |
834 | * reference count. | |
835 | * | |
836 | * Otherwise NULL is returned. | |
837 | */ | |
5d2bea45 | 838 | static struct inode *ifind_fast(struct super_block *sb, |
1da177e4 LT |
839 | struct hlist_head *head, unsigned long ino) |
840 | { | |
841 | struct inode *inode; | |
842 | ||
843 | spin_lock(&inode_lock); | |
844 | inode = find_inode_fast(sb, head, ino); | |
845 | if (inode) { | |
846 | __iget(inode); | |
847 | spin_unlock(&inode_lock); | |
848 | wait_on_inode(inode); | |
849 | return inode; | |
850 | } | |
851 | spin_unlock(&inode_lock); | |
852 | return NULL; | |
853 | } | |
854 | ||
855 | /** | |
88bd5121 | 856 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 LT |
857 | * @sb: super block of file system to search |
858 | * @hashval: hash value (usually inode number) to search for | |
859 | * @test: callback used for comparisons between inodes | |
860 | * @data: opaque data pointer to pass to @test | |
861 | * | |
862 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
863 | * @data in the inode cache. This is a generalized version of ilookup() for | |
864 | * file systems where the inode number is not sufficient for unique | |
865 | * identification of an inode. | |
866 | * | |
867 | * If the inode is in the cache, the inode is returned with an incremented | |
88bd5121 AA |
868 | * reference count. Note, the inode lock is not waited upon so you have to be |
869 | * very careful what you do with the returned inode. You probably should be | |
870 | * using ilookup5() instead. | |
871 | * | |
872 | * Otherwise NULL is returned. | |
873 | * | |
874 | * Note, @test is called with the inode_lock held, so can't sleep. | |
875 | */ | |
876 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, | |
877 | int (*test)(struct inode *, void *), void *data) | |
878 | { | |
879 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
880 | ||
881 | return ifind(sb, head, test, data, 0); | |
882 | } | |
883 | ||
884 | EXPORT_SYMBOL(ilookup5_nowait); | |
885 | ||
886 | /** | |
887 | * ilookup5 - search for an inode in the inode cache | |
888 | * @sb: super block of file system to search | |
889 | * @hashval: hash value (usually inode number) to search for | |
890 | * @test: callback used for comparisons between inodes | |
891 | * @data: opaque data pointer to pass to @test | |
892 | * | |
893 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
894 | * @data in the inode cache. This is a generalized version of ilookup() for | |
895 | * file systems where the inode number is not sufficient for unique | |
896 | * identification of an inode. | |
897 | * | |
898 | * If the inode is in the cache, the inode lock is waited upon and the inode is | |
899 | * returned with an incremented reference count. | |
1da177e4 LT |
900 | * |
901 | * Otherwise NULL is returned. | |
902 | * | |
903 | * Note, @test is called with the inode_lock held, so can't sleep. | |
904 | */ | |
905 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
906 | int (*test)(struct inode *, void *), void *data) | |
907 | { | |
908 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
909 | ||
88bd5121 | 910 | return ifind(sb, head, test, data, 1); |
1da177e4 LT |
911 | } |
912 | ||
913 | EXPORT_SYMBOL(ilookup5); | |
914 | ||
915 | /** | |
916 | * ilookup - search for an inode in the inode cache | |
917 | * @sb: super block of file system to search | |
918 | * @ino: inode number to search for | |
919 | * | |
920 | * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache. | |
921 | * This is for file systems where the inode number is sufficient for unique | |
922 | * identification of an inode. | |
923 | * | |
924 | * If the inode is in the cache, the inode is returned with an incremented | |
925 | * reference count. | |
926 | * | |
927 | * Otherwise NULL is returned. | |
928 | */ | |
929 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
930 | { | |
931 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
932 | ||
933 | return ifind_fast(sb, head, ino); | |
934 | } | |
935 | ||
936 | EXPORT_SYMBOL(ilookup); | |
937 | ||
938 | /** | |
939 | * iget5_locked - obtain an inode from a mounted file system | |
940 | * @sb: super block of file system | |
941 | * @hashval: hash value (usually inode number) to get | |
942 | * @test: callback used for comparisons between inodes | |
943 | * @set: callback used to initialize a new struct inode | |
944 | * @data: opaque data pointer to pass to @test and @set | |
945 | * | |
1da177e4 LT |
946 | * iget5_locked() uses ifind() to search for the inode specified by @hashval |
947 | * and @data in the inode cache and if present it is returned with an increased | |
948 | * reference count. This is a generalized version of iget_locked() for file | |
949 | * systems where the inode number is not sufficient for unique identification | |
950 | * of an inode. | |
951 | * | |
952 | * If the inode is not in cache, get_new_inode() is called to allocate a new | |
953 | * inode and this is returned locked, hashed, and with the I_NEW flag set. The | |
954 | * file system gets to fill it in before unlocking it via unlock_new_inode(). | |
955 | * | |
956 | * Note both @test and @set are called with the inode_lock held, so can't sleep. | |
957 | */ | |
958 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, | |
959 | int (*test)(struct inode *, void *), | |
960 | int (*set)(struct inode *, void *), void *data) | |
961 | { | |
962 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
963 | struct inode *inode; | |
964 | ||
88bd5121 | 965 | inode = ifind(sb, head, test, data, 1); |
1da177e4 LT |
966 | if (inode) |
967 | return inode; | |
968 | /* | |
969 | * get_new_inode() will do the right thing, re-trying the search | |
970 | * in case it had to block at any point. | |
971 | */ | |
972 | return get_new_inode(sb, head, test, set, data); | |
973 | } | |
974 | ||
975 | EXPORT_SYMBOL(iget5_locked); | |
976 | ||
977 | /** | |
978 | * iget_locked - obtain an inode from a mounted file system | |
979 | * @sb: super block of file system | |
980 | * @ino: inode number to get | |
981 | * | |
1da177e4 LT |
982 | * iget_locked() uses ifind_fast() to search for the inode specified by @ino in |
983 | * the inode cache and if present it is returned with an increased reference | |
984 | * count. This is for file systems where the inode number is sufficient for | |
985 | * unique identification of an inode. | |
986 | * | |
987 | * If the inode is not in cache, get_new_inode_fast() is called to allocate a | |
988 | * new inode and this is returned locked, hashed, and with the I_NEW flag set. | |
989 | * The file system gets to fill it in before unlocking it via | |
990 | * unlock_new_inode(). | |
991 | */ | |
992 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) | |
993 | { | |
994 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
995 | struct inode *inode; | |
996 | ||
997 | inode = ifind_fast(sb, head, ino); | |
998 | if (inode) | |
999 | return inode; | |
1000 | /* | |
1001 | * get_new_inode_fast() will do the right thing, re-trying the search | |
1002 | * in case it had to block at any point. | |
1003 | */ | |
1004 | return get_new_inode_fast(sb, head, ino); | |
1005 | } | |
1006 | ||
1007 | EXPORT_SYMBOL(iget_locked); | |
1008 | ||
1009 | /** | |
1010 | * __insert_inode_hash - hash an inode | |
1011 | * @inode: unhashed inode | |
1012 | * @hashval: unsigned long value used to locate this object in the | |
1013 | * inode_hashtable. | |
1014 | * | |
1015 | * Add an inode to the inode hash for this superblock. | |
1016 | */ | |
1017 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
1018 | { | |
1019 | struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval); | |
1020 | spin_lock(&inode_lock); | |
1021 | hlist_add_head(&inode->i_hash, head); | |
1022 | spin_unlock(&inode_lock); | |
1023 | } | |
1024 | ||
1025 | EXPORT_SYMBOL(__insert_inode_hash); | |
1026 | ||
1027 | /** | |
1028 | * remove_inode_hash - remove an inode from the hash | |
1029 | * @inode: inode to unhash | |
1030 | * | |
1031 | * Remove an inode from the superblock. | |
1032 | */ | |
1033 | void remove_inode_hash(struct inode *inode) | |
1034 | { | |
1035 | spin_lock(&inode_lock); | |
1036 | hlist_del_init(&inode->i_hash); | |
1037 | spin_unlock(&inode_lock); | |
1038 | } | |
1039 | ||
1040 | EXPORT_SYMBOL(remove_inode_hash); | |
1041 | ||
1042 | /* | |
1043 | * Tell the filesystem that this inode is no longer of any interest and should | |
1044 | * be completely destroyed. | |
1045 | * | |
1046 | * We leave the inode in the inode hash table until *after* the filesystem's | |
1047 | * ->delete_inode completes. This ensures that an iget (such as nfsd might | |
1048 | * instigate) will always find up-to-date information either in the hash or on | |
1049 | * disk. | |
1050 | * | |
1051 | * I_FREEING is set so that no-one will take a new reference to the inode while | |
1052 | * it is being deleted. | |
1053 | */ | |
1054 | void generic_delete_inode(struct inode *inode) | |
1055 | { | |
ee9b6d61 | 1056 | const struct super_operations *op = inode->i_sb->s_op; |
1da177e4 LT |
1057 | |
1058 | list_del_init(&inode->i_list); | |
1059 | list_del_init(&inode->i_sb_list); | |
ed97bd37 | 1060 | inode->i_state |= I_FREEING; |
1da177e4 LT |
1061 | inodes_stat.nr_inodes--; |
1062 | spin_unlock(&inode_lock); | |
1063 | ||
1da177e4 LT |
1064 | security_inode_delete(inode); |
1065 | ||
1066 | if (op->delete_inode) { | |
1067 | void (*delete)(struct inode *) = op->delete_inode; | |
1068 | if (!is_bad_inode(inode)) | |
1069 | DQUOT_INIT(inode); | |
e85b5652 MF |
1070 | /* Filesystems implementing their own |
1071 | * s_op->delete_inode are required to call | |
1072 | * truncate_inode_pages and clear_inode() | |
1073 | * internally */ | |
1da177e4 | 1074 | delete(inode); |
e85b5652 MF |
1075 | } else { |
1076 | truncate_inode_pages(&inode->i_data, 0); | |
1da177e4 | 1077 | clear_inode(inode); |
e85b5652 | 1078 | } |
1da177e4 LT |
1079 | spin_lock(&inode_lock); |
1080 | hlist_del_init(&inode->i_hash); | |
1081 | spin_unlock(&inode_lock); | |
1082 | wake_up_inode(inode); | |
b7542f8c | 1083 | BUG_ON(inode->i_state != I_CLEAR); |
1da177e4 LT |
1084 | destroy_inode(inode); |
1085 | } | |
1086 | ||
1087 | EXPORT_SYMBOL(generic_delete_inode); | |
1088 | ||
1089 | static void generic_forget_inode(struct inode *inode) | |
1090 | { | |
1091 | struct super_block *sb = inode->i_sb; | |
1092 | ||
1093 | if (!hlist_unhashed(&inode->i_hash)) { | |
1c0eeaf5 | 1094 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) |
1da177e4 LT |
1095 | list_move(&inode->i_list, &inode_unused); |
1096 | inodes_stat.nr_unused++; | |
acb0c854 | 1097 | if (sb->s_flags & MS_ACTIVE) { |
991114c6 | 1098 | spin_unlock(&inode_lock); |
1da177e4 | 1099 | return; |
991114c6 AV |
1100 | } |
1101 | inode->i_state |= I_WILL_FREE; | |
1102 | spin_unlock(&inode_lock); | |
1da177e4 LT |
1103 | write_inode_now(inode, 1); |
1104 | spin_lock(&inode_lock); | |
991114c6 | 1105 | inode->i_state &= ~I_WILL_FREE; |
1da177e4 LT |
1106 | inodes_stat.nr_unused--; |
1107 | hlist_del_init(&inode->i_hash); | |
1108 | } | |
1109 | list_del_init(&inode->i_list); | |
1110 | list_del_init(&inode->i_sb_list); | |
991114c6 | 1111 | inode->i_state |= I_FREEING; |
1da177e4 LT |
1112 | inodes_stat.nr_inodes--; |
1113 | spin_unlock(&inode_lock); | |
1114 | if (inode->i_data.nrpages) | |
1115 | truncate_inode_pages(&inode->i_data, 0); | |
1116 | clear_inode(inode); | |
7f04c26d | 1117 | wake_up_inode(inode); |
1da177e4 LT |
1118 | destroy_inode(inode); |
1119 | } | |
1120 | ||
1121 | /* | |
1122 | * Normal UNIX filesystem behaviour: delete the | |
1123 | * inode when the usage count drops to zero, and | |
1124 | * i_nlink is zero. | |
1125 | */ | |
cb2c0233 | 1126 | void generic_drop_inode(struct inode *inode) |
1da177e4 LT |
1127 | { |
1128 | if (!inode->i_nlink) | |
1129 | generic_delete_inode(inode); | |
1130 | else | |
1131 | generic_forget_inode(inode); | |
1132 | } | |
1133 | ||
cb2c0233 MF |
1134 | EXPORT_SYMBOL_GPL(generic_drop_inode); |
1135 | ||
1da177e4 LT |
1136 | /* |
1137 | * Called when we're dropping the last reference | |
1138 | * to an inode. | |
1139 | * | |
1140 | * Call the FS "drop()" function, defaulting to | |
1141 | * the legacy UNIX filesystem behaviour.. | |
1142 | * | |
1143 | * NOTE! NOTE! NOTE! We're called with the inode lock | |
1144 | * held, and the drop function is supposed to release | |
1145 | * the lock! | |
1146 | */ | |
1147 | static inline void iput_final(struct inode *inode) | |
1148 | { | |
ee9b6d61 | 1149 | const struct super_operations *op = inode->i_sb->s_op; |
1da177e4 LT |
1150 | void (*drop)(struct inode *) = generic_drop_inode; |
1151 | ||
1152 | if (op && op->drop_inode) | |
1153 | drop = op->drop_inode; | |
1154 | drop(inode); | |
1155 | } | |
1156 | ||
1157 | /** | |
1158 | * iput - put an inode | |
1159 | * @inode: inode to put | |
1160 | * | |
1161 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1162 | * zero, the inode is then freed and may also be destroyed. | |
1163 | * | |
1164 | * Consequently, iput() can sleep. | |
1165 | */ | |
1166 | void iput(struct inode *inode) | |
1167 | { | |
1168 | if (inode) { | |
1da177e4 LT |
1169 | BUG_ON(inode->i_state == I_CLEAR); |
1170 | ||
1da177e4 LT |
1171 | if (atomic_dec_and_lock(&inode->i_count, &inode_lock)) |
1172 | iput_final(inode); | |
1173 | } | |
1174 | } | |
1175 | ||
1176 | EXPORT_SYMBOL(iput); | |
1177 | ||
1178 | /** | |
1179 | * bmap - find a block number in a file | |
1180 | * @inode: inode of file | |
1181 | * @block: block to find | |
1182 | * | |
1183 | * Returns the block number on the device holding the inode that | |
1184 | * is the disk block number for the block of the file requested. | |
1185 | * That is, asked for block 4 of inode 1 the function will return the | |
1186 | * disk block relative to the disk start that holds that block of the | |
1187 | * file. | |
1188 | */ | |
1189 | sector_t bmap(struct inode * inode, sector_t block) | |
1190 | { | |
1191 | sector_t res = 0; | |
1192 | if (inode->i_mapping->a_ops->bmap) | |
1193 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1194 | return res; | |
1195 | } | |
1da177e4 LT |
1196 | EXPORT_SYMBOL(bmap); |
1197 | ||
1198 | /** | |
869243a0 CH |
1199 | * touch_atime - update the access time |
1200 | * @mnt: mount the inode is accessed on | |
7045f37b | 1201 | * @dentry: dentry accessed |
1da177e4 LT |
1202 | * |
1203 | * Update the accessed time on an inode and mark it for writeback. | |
1204 | * This function automatically handles read only file systems and media, | |
1205 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1206 | */ | |
869243a0 | 1207 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 1208 | { |
869243a0 | 1209 | struct inode *inode = dentry->d_inode; |
1da177e4 LT |
1210 | struct timespec now; |
1211 | ||
cdb70f3f | 1212 | if (mnt_want_write(mnt)) |
b2276138 | 1213 | return; |
cdb70f3f DH |
1214 | if (inode->i_flags & S_NOATIME) |
1215 | goto out; | |
37756ced | 1216 | if (IS_NOATIME(inode)) |
cdb70f3f | 1217 | goto out; |
b2276138 | 1218 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
cdb70f3f | 1219 | goto out; |
47ae32d6 | 1220 | |
cdb70f3f DH |
1221 | if (mnt->mnt_flags & MNT_NOATIME) |
1222 | goto out; | |
1223 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1224 | goto out; | |
1225 | if (mnt->mnt_flags & MNT_RELATIME) { | |
1226 | /* | |
1227 | * With relative atime, only update atime if the previous | |
1228 | * atime is earlier than either the ctime or mtime. | |
1229 | */ | |
1230 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) < 0 && | |
1231 | timespec_compare(&inode->i_ctime, &inode->i_atime) < 0) | |
1232 | goto out; | |
b2276138 | 1233 | } |
1da177e4 LT |
1234 | |
1235 | now = current_fs_time(inode->i_sb); | |
47ae32d6 | 1236 | if (timespec_equal(&inode->i_atime, &now)) |
cdb70f3f | 1237 | goto out; |
47ae32d6 VH |
1238 | |
1239 | inode->i_atime = now; | |
1240 | mark_inode_dirty_sync(inode); | |
cdb70f3f DH |
1241 | out: |
1242 | mnt_drop_write(mnt); | |
1da177e4 | 1243 | } |
869243a0 | 1244 | EXPORT_SYMBOL(touch_atime); |
1da177e4 LT |
1245 | |
1246 | /** | |
870f4817 CH |
1247 | * file_update_time - update mtime and ctime time |
1248 | * @file: file accessed | |
1da177e4 | 1249 | * |
870f4817 CH |
1250 | * Update the mtime and ctime members of an inode and mark the inode |
1251 | * for writeback. Note that this function is meant exclusively for | |
1252 | * usage in the file write path of filesystems, and filesystems may | |
1253 | * choose to explicitly ignore update via this function with the | |
1254 | * S_NOCTIME inode flag, e.g. for network filesystem where these | |
1255 | * timestamps are handled by the server. | |
1da177e4 LT |
1256 | */ |
1257 | ||
870f4817 | 1258 | void file_update_time(struct file *file) |
1da177e4 | 1259 | { |
0f7fc9e4 | 1260 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
1261 | struct timespec now; |
1262 | int sync_it = 0; | |
20ddee2c | 1263 | int err; |
1da177e4 LT |
1264 | |
1265 | if (IS_NOCMTIME(inode)) | |
1266 | return; | |
20ddee2c DH |
1267 | |
1268 | err = mnt_want_write(file->f_path.mnt); | |
1269 | if (err) | |
1da177e4 LT |
1270 | return; |
1271 | ||
1272 | now = current_fs_time(inode->i_sb); | |
ed97bd37 AM |
1273 | if (!timespec_equal(&inode->i_mtime, &now)) { |
1274 | inode->i_mtime = now; | |
1da177e4 | 1275 | sync_it = 1; |
ed97bd37 | 1276 | } |
1da177e4 | 1277 | |
ed97bd37 AM |
1278 | if (!timespec_equal(&inode->i_ctime, &now)) { |
1279 | inode->i_ctime = now; | |
870f4817 | 1280 | sync_it = 1; |
ed97bd37 | 1281 | } |
870f4817 | 1282 | |
7a224228 JNC |
1283 | if (IS_I_VERSION(inode)) { |
1284 | inode_inc_iversion(inode); | |
1285 | sync_it = 1; | |
1286 | } | |
1287 | ||
1da177e4 LT |
1288 | if (sync_it) |
1289 | mark_inode_dirty_sync(inode); | |
20ddee2c | 1290 | mnt_drop_write(file->f_path.mnt); |
1da177e4 LT |
1291 | } |
1292 | ||
870f4817 | 1293 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1294 | |
1295 | int inode_needs_sync(struct inode *inode) | |
1296 | { | |
1297 | if (IS_SYNC(inode)) | |
1298 | return 1; | |
1299 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1300 | return 1; | |
1301 | return 0; | |
1302 | } | |
1303 | ||
1304 | EXPORT_SYMBOL(inode_needs_sync); | |
1305 | ||
1da177e4 LT |
1306 | int inode_wait(void *word) |
1307 | { | |
1308 | schedule(); | |
1309 | return 0; | |
1310 | } | |
1311 | ||
1312 | /* | |
168a9fd6 MS |
1313 | * If we try to find an inode in the inode hash while it is being |
1314 | * deleted, we have to wait until the filesystem completes its | |
1315 | * deletion before reporting that it isn't found. This function waits | |
1316 | * until the deletion _might_ have completed. Callers are responsible | |
1317 | * to recheck inode state. | |
1318 | * | |
1319 | * It doesn't matter if I_LOCK is not set initially, a call to | |
1320 | * wake_up_inode() after removing from the hash list will DTRT. | |
1321 | * | |
1da177e4 LT |
1322 | * This is called with inode_lock held. |
1323 | */ | |
1324 | static void __wait_on_freeing_inode(struct inode *inode) | |
1325 | { | |
1326 | wait_queue_head_t *wq; | |
1327 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK); | |
1da177e4 LT |
1328 | wq = bit_waitqueue(&inode->i_state, __I_LOCK); |
1329 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
1330 | spin_unlock(&inode_lock); | |
1331 | schedule(); | |
1332 | finish_wait(wq, &wait.wait); | |
1333 | spin_lock(&inode_lock); | |
1334 | } | |
1335 | ||
62752ee1 MF |
1336 | /* |
1337 | * We rarely want to lock two inodes that do not have a parent/child | |
1338 | * relationship (such as directory, child inode) simultaneously. The | |
1339 | * vast majority of file systems should be able to get along fine | |
1340 | * without this. Do not use these functions except as a last resort. | |
1341 | */ | |
1342 | void inode_double_lock(struct inode *inode1, struct inode *inode2) | |
1343 | { | |
1344 | if (inode1 == NULL || inode2 == NULL || inode1 == inode2) { | |
1345 | if (inode1) | |
1346 | mutex_lock(&inode1->i_mutex); | |
1347 | else if (inode2) | |
1348 | mutex_lock(&inode2->i_mutex); | |
1349 | return; | |
1350 | } | |
1351 | ||
1352 | if (inode1 < inode2) { | |
1353 | mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT); | |
1354 | mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD); | |
1355 | } else { | |
1356 | mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT); | |
1357 | mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD); | |
1358 | } | |
1359 | } | |
1360 | EXPORT_SYMBOL(inode_double_lock); | |
1361 | ||
1362 | void inode_double_unlock(struct inode *inode1, struct inode *inode2) | |
1363 | { | |
1364 | if (inode1) | |
1365 | mutex_unlock(&inode1->i_mutex); | |
1366 | ||
1367 | if (inode2 && inode2 != inode1) | |
1368 | mutex_unlock(&inode2->i_mutex); | |
1369 | } | |
1370 | EXPORT_SYMBOL(inode_double_unlock); | |
1371 | ||
1da177e4 LT |
1372 | static __initdata unsigned long ihash_entries; |
1373 | static int __init set_ihash_entries(char *str) | |
1374 | { | |
1375 | if (!str) | |
1376 | return 0; | |
1377 | ihash_entries = simple_strtoul(str, &str, 0); | |
1378 | return 1; | |
1379 | } | |
1380 | __setup("ihash_entries=", set_ihash_entries); | |
1381 | ||
1382 | /* | |
1383 | * Initialize the waitqueues and inode hash table. | |
1384 | */ | |
1385 | void __init inode_init_early(void) | |
1386 | { | |
1387 | int loop; | |
1388 | ||
1389 | /* If hashes are distributed across NUMA nodes, defer | |
1390 | * hash allocation until vmalloc space is available. | |
1391 | */ | |
1392 | if (hashdist) | |
1393 | return; | |
1394 | ||
1395 | inode_hashtable = | |
1396 | alloc_large_system_hash("Inode-cache", | |
1397 | sizeof(struct hlist_head), | |
1398 | ihash_entries, | |
1399 | 14, | |
1400 | HASH_EARLY, | |
1401 | &i_hash_shift, | |
1402 | &i_hash_mask, | |
1403 | 0); | |
1404 | ||
1405 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1406 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1407 | } | |
1408 | ||
74bf17cf | 1409 | void __init inode_init(void) |
1da177e4 LT |
1410 | { |
1411 | int loop; | |
1412 | ||
1413 | /* inode slab cache */ | |
b0196009 PJ |
1414 | inode_cachep = kmem_cache_create("inode_cache", |
1415 | sizeof(struct inode), | |
1416 | 0, | |
1417 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1418 | SLAB_MEM_SPREAD), | |
20c2df83 | 1419 | init_once); |
8e1f936b | 1420 | register_shrinker(&icache_shrinker); |
1da177e4 LT |
1421 | |
1422 | /* Hash may have been set up in inode_init_early */ | |
1423 | if (!hashdist) | |
1424 | return; | |
1425 | ||
1426 | inode_hashtable = | |
1427 | alloc_large_system_hash("Inode-cache", | |
1428 | sizeof(struct hlist_head), | |
1429 | ihash_entries, | |
1430 | 14, | |
1431 | 0, | |
1432 | &i_hash_shift, | |
1433 | &i_hash_mask, | |
1434 | 0); | |
1435 | ||
1436 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1437 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1438 | } | |
1439 | ||
1440 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1441 | { | |
1442 | inode->i_mode = mode; | |
1443 | if (S_ISCHR(mode)) { | |
1444 | inode->i_fop = &def_chr_fops; | |
1445 | inode->i_rdev = rdev; | |
1446 | } else if (S_ISBLK(mode)) { | |
1447 | inode->i_fop = &def_blk_fops; | |
1448 | inode->i_rdev = rdev; | |
1449 | } else if (S_ISFIFO(mode)) | |
1450 | inode->i_fop = &def_fifo_fops; | |
1451 | else if (S_ISSOCK(mode)) | |
1452 | inode->i_fop = &bad_sock_fops; | |
1453 | else | |
1454 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n", | |
1455 | mode); | |
1456 | } | |
1457 | EXPORT_SYMBOL(init_special_inode); |