]>
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> | |
1da177e4 LT |
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> | |
88e0fbc4 | 16 | #include <linux/rwsem.h> |
1da177e4 LT |
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> |
3be25f49 | 24 | #include <linux/fsnotify.h> |
fc33a7bb | 25 | #include <linux/mount.h> |
efaee192 | 26 | #include <linux/async.h> |
f19d4a8f | 27 | #include <linux/posix_acl.h> |
1da177e4 LT |
28 | |
29 | /* | |
30 | * This is needed for the following functions: | |
31 | * - inode_has_buffers | |
32 | * - invalidate_inode_buffers | |
1da177e4 LT |
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 | ||
fa3536cc ED |
61 | static unsigned int i_hash_mask __read_mostly; |
62 | static unsigned int i_hash_shift __read_mostly; | |
1da177e4 LT |
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); | |
fa3536cc | 78 | static struct hlist_head *inode_hashtable __read_mostly; |
1da177e4 LT |
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 | /* | |
88e0fbc4 | 89 | * iprune_sem provides exclusion between the kswapd or try_to_free_pages |
1da177e4 LT |
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. | |
88e0fbc4 NP |
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. | |
1da177e4 | 99 | */ |
88e0fbc4 | 100 | static DECLARE_RWSEM(iprune_sem); |
1da177e4 LT |
101 | |
102 | /* | |
103 | * Statistics gathering.. | |
104 | */ | |
105 | struct inodes_stat_t inodes_stat; | |
106 | ||
6b3304b5 | 107 | static struct kmem_cache *inode_cachep __read_mostly; |
1da177e4 | 108 | |
1c0eeaf5 JE |
109 | static void wake_up_inode(struct inode *inode) |
110 | { | |
111 | /* | |
112 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
113 | */ | |
114 | smp_mb(); | |
eaff8079 | 115 | wake_up_bit(&inode->i_state, __I_NEW); |
1c0eeaf5 JE |
116 | } |
117 | ||
2cb1599f DC |
118 | /** |
119 | * inode_init_always - perform inode structure intialisation | |
0bc02f3f RD |
120 | * @sb: superblock inode belongs to |
121 | * @inode: inode to initialise | |
2cb1599f DC |
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 | */ | |
54e34621 | 126 | int inode_init_always(struct super_block *sb, struct inode *inode) |
1da177e4 | 127 | { |
f5e54d6e | 128 | static const struct address_space_operations empty_aops; |
6e1d5dcc | 129 | static const struct inode_operations empty_iops; |
99ac48f5 | 130 | static const struct file_operations empty_fops; |
6b3304b5 | 131 | struct address_space *const mapping = &inode->i_data; |
2cb1599f DC |
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; | |
56ff5efa AV |
140 | inode->i_uid = 0; |
141 | inode->i_gid = 0; | |
2cb1599f DC |
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; | |
1da177e4 | 147 | #ifdef CONFIG_QUOTA |
2cb1599f | 148 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
1da177e4 | 149 | #endif |
2cb1599f DC |
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; | |
6146f0d5 MZ |
155 | |
156 | if (security_inode_alloc(inode)) | |
54e34621 | 157 | goto out; |
2cb1599f DC |
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; | |
3c1d4378 | 170 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); |
2cb1599f DC |
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 | ||
2c96ce9f | 183 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; |
2cb1599f DC |
184 | mapping->backing_dev_info = bdi; |
185 | } | |
186 | inode->i_private = NULL; | |
187 | inode->i_mapping = mapping; | |
f19d4a8f AV |
188 | #ifdef CONFIG_FS_POSIX_ACL |
189 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; | |
190 | #endif | |
2cb1599f | 191 | |
3be25f49 EP |
192 | #ifdef CONFIG_FSNOTIFY |
193 | inode->i_fsnotify_mask = 0; | |
194 | #endif | |
195 | ||
54e34621 | 196 | return 0; |
54e34621 CH |
197 | out: |
198 | return -ENOMEM; | |
1da177e4 | 199 | } |
2cb1599f DC |
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 | ||
54e34621 CH |
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; | |
2cb1599f | 223 | } |
1da177e4 | 224 | |
2e00c97e | 225 | void __destroy_inode(struct inode *inode) |
1da177e4 | 226 | { |
b7542f8c | 227 | BUG_ON(inode_has_buffers(inode)); |
1da177e4 | 228 | security_inode_free(inode); |
3be25f49 | 229 | fsnotify_inode_delete(inode); |
f19d4a8f AV |
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 | |
2e00c97e CH |
236 | } |
237 | EXPORT_SYMBOL(__destroy_inode); | |
238 | ||
239 | void destroy_inode(struct inode *inode) | |
240 | { | |
241 | __destroy_inode(inode); | |
1da177e4 LT |
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 | } | |
1da177e4 LT |
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); | |
1da177e4 | 259 | INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC); |
19fd6231 | 260 | spin_lock_init(&inode->i_data.tree_lock); |
1da177e4 LT |
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); | |
1da177e4 | 266 | i_size_ordered_init(inode); |
0eeca283 RL |
267 | #ifdef CONFIG_INOTIFY |
268 | INIT_LIST_HEAD(&inode->inotify_watches); | |
d4f9af9d | 269 | mutex_init(&inode->inotify_mutex); |
0eeca283 | 270 | #endif |
3be25f49 EP |
271 | #ifdef CONFIG_FSNOTIFY |
272 | INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries); | |
273 | #endif | |
1da177e4 | 274 | } |
1da177e4 LT |
275 | EXPORT_SYMBOL(inode_init_once); |
276 | ||
51cc5068 | 277 | static void init_once(void *foo) |
1da177e4 | 278 | { |
6b3304b5 | 279 | struct inode *inode = (struct inode *) foo; |
1da177e4 | 280 | |
a35afb83 | 281 | inode_init_once(inode); |
1da177e4 LT |
282 | } |
283 | ||
284 | /* | |
285 | * inode_lock must be held | |
286 | */ | |
6b3304b5 | 287 | void __iget(struct inode *inode) |
1da177e4 | 288 | { |
2e147f1e | 289 | if (atomic_inc_return(&inode->i_count) != 1) |
1da177e4 | 290 | return; |
2e147f1e | 291 | |
1c0eeaf5 | 292 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) |
1da177e4 LT |
293 | list_move(&inode->i_list, &inode_in_use); |
294 | inodes_stat.nr_unused--; | |
295 | } | |
296 | ||
b0683aa6 AV |
297 | void end_writeback(struct inode *inode) |
298 | { | |
299 | might_sleep(); | |
300 | BUG_ON(inode->i_data.nrpages); | |
301 | BUG_ON(!list_empty(&inode->i_data.private_list)); | |
302 | BUG_ON(!(inode->i_state & I_FREEING)); | |
303 | BUG_ON(inode->i_state & I_CLEAR); | |
304 | inode_sync_wait(inode); | |
305 | inode->i_state = I_FREEING | I_CLEAR; | |
306 | } | |
307 | EXPORT_SYMBOL(end_writeback); | |
308 | ||
644da596 | 309 | static void evict(struct inode *inode) |
b4272d4c AV |
310 | { |
311 | const struct super_operations *op = inode->i_sb->s_op; | |
312 | ||
be7ce416 AV |
313 | if (op->evict_inode) { |
314 | op->evict_inode(inode); | |
b4272d4c AV |
315 | } else { |
316 | if (inode->i_data.nrpages) | |
317 | truncate_inode_pages(&inode->i_data, 0); | |
30140837 AV |
318 | invalidate_inode_buffers(inode); |
319 | end_writeback(inode); | |
320 | if (op->clear_inode) | |
321 | op->clear_inode(inode); | |
b4272d4c | 322 | } |
661074e9 AV |
323 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
324 | bd_forget(inode); | |
325 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
326 | cd_forget(inode); | |
b4272d4c AV |
327 | } |
328 | ||
1da177e4 LT |
329 | /* |
330 | * dispose_list - dispose of the contents of a local list | |
331 | * @head: the head of the list to free | |
332 | * | |
333 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
334 | * need to worry about list corruption and SMP locks. | |
335 | */ | |
336 | static void dispose_list(struct list_head *head) | |
337 | { | |
338 | int nr_disposed = 0; | |
339 | ||
340 | while (!list_empty(head)) { | |
341 | struct inode *inode; | |
342 | ||
b5e61818 | 343 | inode = list_first_entry(head, struct inode, i_list); |
1da177e4 LT |
344 | list_del(&inode->i_list); |
345 | ||
644da596 | 346 | evict(inode); |
4120db47 AB |
347 | |
348 | spin_lock(&inode_lock); | |
349 | hlist_del_init(&inode->i_hash); | |
350 | list_del_init(&inode->i_sb_list); | |
351 | spin_unlock(&inode_lock); | |
352 | ||
353 | wake_up_inode(inode); | |
1da177e4 LT |
354 | destroy_inode(inode); |
355 | nr_disposed++; | |
356 | } | |
357 | spin_lock(&inode_lock); | |
358 | inodes_stat.nr_inodes -= nr_disposed; | |
359 | spin_unlock(&inode_lock); | |
360 | } | |
361 | ||
362 | /* | |
363 | * Invalidate all inodes for a device. | |
364 | */ | |
365 | static int invalidate_list(struct list_head *head, struct list_head *dispose) | |
366 | { | |
367 | struct list_head *next; | |
368 | int busy = 0, count = 0; | |
369 | ||
370 | next = head->next; | |
371 | for (;;) { | |
6b3304b5 MK |
372 | struct list_head *tmp = next; |
373 | struct inode *inode; | |
1da177e4 LT |
374 | |
375 | /* | |
376 | * We can reschedule here without worrying about the list's | |
377 | * consistency because the per-sb list of inodes must not | |
88e0fbc4 | 378 | * change during umount anymore, and because iprune_sem keeps |
1da177e4 LT |
379 | * shrink_icache_memory() away. |
380 | */ | |
381 | cond_resched_lock(&inode_lock); | |
382 | ||
383 | next = next->next; | |
384 | if (tmp == head) | |
385 | break; | |
386 | inode = list_entry(tmp, struct inode, i_sb_list); | |
aabb8fdb NP |
387 | if (inode->i_state & I_NEW) |
388 | continue; | |
1da177e4 LT |
389 | invalidate_inode_buffers(inode); |
390 | if (!atomic_read(&inode->i_count)) { | |
1da177e4 | 391 | list_move(&inode->i_list, dispose); |
7ef0d737 | 392 | WARN_ON(inode->i_state & I_NEW); |
1da177e4 LT |
393 | inode->i_state |= I_FREEING; |
394 | count++; | |
395 | continue; | |
396 | } | |
397 | busy = 1; | |
398 | } | |
399 | /* only unused inodes may be cached with i_count zero */ | |
400 | inodes_stat.nr_unused -= count; | |
401 | return busy; | |
402 | } | |
403 | ||
1da177e4 LT |
404 | /** |
405 | * invalidate_inodes - discard the inodes on a device | |
406 | * @sb: superblock | |
407 | * | |
408 | * Discard all of the inodes for a given superblock. If the discard | |
409 | * fails because there are busy inodes then a non zero value is returned. | |
410 | * If the discard is successful all the inodes have been discarded. | |
411 | */ | |
6b3304b5 | 412 | int invalidate_inodes(struct super_block *sb) |
1da177e4 LT |
413 | { |
414 | int busy; | |
415 | LIST_HEAD(throw_away); | |
416 | ||
88e0fbc4 | 417 | down_write(&iprune_sem); |
1da177e4 | 418 | spin_lock(&inode_lock); |
0eeca283 | 419 | inotify_unmount_inodes(&sb->s_inodes); |
164bc619 | 420 | fsnotify_unmount_inodes(&sb->s_inodes); |
1da177e4 LT |
421 | busy = invalidate_list(&sb->s_inodes, &throw_away); |
422 | spin_unlock(&inode_lock); | |
423 | ||
424 | dispose_list(&throw_away); | |
88e0fbc4 | 425 | up_write(&iprune_sem); |
1da177e4 LT |
426 | |
427 | return busy; | |
428 | } | |
1da177e4 | 429 | EXPORT_SYMBOL(invalidate_inodes); |
1da177e4 LT |
430 | |
431 | static int can_unuse(struct inode *inode) | |
432 | { | |
433 | if (inode->i_state) | |
434 | return 0; | |
435 | if (inode_has_buffers(inode)) | |
436 | return 0; | |
437 | if (atomic_read(&inode->i_count)) | |
438 | return 0; | |
439 | if (inode->i_data.nrpages) | |
440 | return 0; | |
441 | return 1; | |
442 | } | |
443 | ||
444 | /* | |
445 | * Scan `goal' inodes on the unused list for freeable ones. They are moved to | |
446 | * a temporary list and then are freed outside inode_lock by dispose_list(). | |
447 | * | |
448 | * Any inodes which are pinned purely because of attached pagecache have their | |
449 | * pagecache removed. We expect the final iput() on that inode to add it to | |
450 | * the front of the inode_unused list. So look for it there and if the | |
451 | * inode is still freeable, proceed. The right inode is found 99.9% of the | |
452 | * time in testing on a 4-way. | |
453 | * | |
454 | * If the inode has metadata buffers attached to mapping->private_list then | |
455 | * try to remove them. | |
456 | */ | |
457 | static void prune_icache(int nr_to_scan) | |
458 | { | |
459 | LIST_HEAD(freeable); | |
460 | int nr_pruned = 0; | |
461 | int nr_scanned; | |
462 | unsigned long reap = 0; | |
463 | ||
88e0fbc4 | 464 | down_read(&iprune_sem); |
1da177e4 LT |
465 | spin_lock(&inode_lock); |
466 | for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) { | |
467 | struct inode *inode; | |
468 | ||
469 | if (list_empty(&inode_unused)) | |
470 | break; | |
471 | ||
472 | inode = list_entry(inode_unused.prev, struct inode, i_list); | |
473 | ||
474 | if (inode->i_state || atomic_read(&inode->i_count)) { | |
475 | list_move(&inode->i_list, &inode_unused); | |
476 | continue; | |
477 | } | |
478 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
479 | __iget(inode); | |
480 | spin_unlock(&inode_lock); | |
481 | if (remove_inode_buffers(inode)) | |
fc0ecff6 AM |
482 | reap += invalidate_mapping_pages(&inode->i_data, |
483 | 0, -1); | |
1da177e4 LT |
484 | iput(inode); |
485 | spin_lock(&inode_lock); | |
486 | ||
487 | if (inode != list_entry(inode_unused.next, | |
488 | struct inode, i_list)) | |
489 | continue; /* wrong inode or list_empty */ | |
490 | if (!can_unuse(inode)) | |
491 | continue; | |
492 | } | |
1da177e4 | 493 | list_move(&inode->i_list, &freeable); |
7ef0d737 | 494 | WARN_ON(inode->i_state & I_NEW); |
1da177e4 LT |
495 | inode->i_state |= I_FREEING; |
496 | nr_pruned++; | |
497 | } | |
498 | inodes_stat.nr_unused -= nr_pruned; | |
f8891e5e CL |
499 | if (current_is_kswapd()) |
500 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
501 | else | |
502 | __count_vm_events(PGINODESTEAL, reap); | |
1da177e4 LT |
503 | spin_unlock(&inode_lock); |
504 | ||
505 | dispose_list(&freeable); | |
88e0fbc4 | 506 | up_read(&iprune_sem); |
1da177e4 LT |
507 | } |
508 | ||
509 | /* | |
510 | * shrink_icache_memory() will attempt to reclaim some unused inodes. Here, | |
511 | * "unused" means that no dentries are referring to the inodes: the files are | |
512 | * not open and the dcache references to those inodes have already been | |
513 | * reclaimed. | |
514 | * | |
515 | * This function is passed the number of inodes to scan, and it returns the | |
516 | * total number of remaining possibly-reclaimable inodes. | |
517 | */ | |
7f8275d0 | 518 | static int shrink_icache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
519 | { |
520 | if (nr) { | |
521 | /* | |
522 | * Nasty deadlock avoidance. We may hold various FS locks, | |
523 | * and we don't want to recurse into the FS that called us | |
524 | * in clear_inode() and friends.. | |
6b3304b5 | 525 | */ |
1da177e4 LT |
526 | if (!(gfp_mask & __GFP_FS)) |
527 | return -1; | |
528 | prune_icache(nr); | |
529 | } | |
530 | return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; | |
531 | } | |
532 | ||
8e1f936b RR |
533 | static struct shrinker icache_shrinker = { |
534 | .shrink = shrink_icache_memory, | |
535 | .seeks = DEFAULT_SEEKS, | |
536 | }; | |
537 | ||
1da177e4 LT |
538 | static void __wait_on_freeing_inode(struct inode *inode); |
539 | /* | |
540 | * Called with the inode lock held. | |
541 | * NOTE: we are not increasing the inode-refcount, you must call __iget() | |
542 | * by hand after calling find_inode now! This simplifies iunique and won't | |
543 | * add any additional branch in the common code. | |
544 | */ | |
6b3304b5 MK |
545 | static struct inode *find_inode(struct super_block *sb, |
546 | struct hlist_head *head, | |
547 | int (*test)(struct inode *, void *), | |
548 | void *data) | |
1da177e4 LT |
549 | { |
550 | struct hlist_node *node; | |
6b3304b5 | 551 | struct inode *inode = NULL; |
1da177e4 LT |
552 | |
553 | repeat: | |
c5c8be3c | 554 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
555 | if (inode->i_sb != sb) |
556 | continue; | |
557 | if (!test(inode, data)) | |
558 | continue; | |
a4ffdde6 | 559 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
560 | __wait_on_freeing_inode(inode); |
561 | goto repeat; | |
562 | } | |
563 | break; | |
564 | } | |
565 | return node ? inode : NULL; | |
566 | } | |
567 | ||
568 | /* | |
569 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
570 | * iget_locked for details. | |
571 | */ | |
6b3304b5 MK |
572 | static struct inode *find_inode_fast(struct super_block *sb, |
573 | struct hlist_head *head, unsigned long ino) | |
1da177e4 LT |
574 | { |
575 | struct hlist_node *node; | |
6b3304b5 | 576 | struct inode *inode = NULL; |
1da177e4 LT |
577 | |
578 | repeat: | |
c5c8be3c | 579 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
580 | if (inode->i_ino != ino) |
581 | continue; | |
582 | if (inode->i_sb != sb) | |
583 | continue; | |
a4ffdde6 | 584 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
585 | __wait_on_freeing_inode(inode); |
586 | goto repeat; | |
587 | } | |
588 | break; | |
589 | } | |
590 | return node ? inode : NULL; | |
591 | } | |
592 | ||
8290c35f DC |
593 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
594 | { | |
595 | unsigned long tmp; | |
596 | ||
597 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
598 | L1_CACHE_BYTES; | |
599 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS); | |
600 | return tmp & I_HASHMASK; | |
601 | } | |
602 | ||
603 | static inline void | |
604 | __inode_add_to_lists(struct super_block *sb, struct hlist_head *head, | |
605 | struct inode *inode) | |
606 | { | |
607 | inodes_stat.nr_inodes++; | |
608 | list_add(&inode->i_list, &inode_in_use); | |
609 | list_add(&inode->i_sb_list, &sb->s_inodes); | |
610 | if (head) | |
611 | hlist_add_head(&inode->i_hash, head); | |
612 | } | |
613 | ||
614 | /** | |
615 | * inode_add_to_lists - add a new inode to relevant lists | |
0bc02f3f RD |
616 | * @sb: superblock inode belongs to |
617 | * @inode: inode to mark in use | |
8290c35f DC |
618 | * |
619 | * When an inode is allocated it needs to be accounted for, added to the in use | |
620 | * list, the owning superblock and the inode hash. This needs to be done under | |
621 | * the inode_lock, so export a function to do this rather than the inode lock | |
622 | * itself. We calculate the hash list to add to here so it is all internal | |
623 | * which requires the caller to have already set up the inode number in the | |
624 | * inode to add. | |
625 | */ | |
626 | void inode_add_to_lists(struct super_block *sb, struct inode *inode) | |
627 | { | |
628 | struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino); | |
629 | ||
630 | spin_lock(&inode_lock); | |
631 | __inode_add_to_lists(sb, head, inode); | |
632 | spin_unlock(&inode_lock); | |
633 | } | |
634 | EXPORT_SYMBOL_GPL(inode_add_to_lists); | |
635 | ||
1da177e4 LT |
636 | /** |
637 | * new_inode - obtain an inode | |
638 | * @sb: superblock | |
639 | * | |
769848c0 | 640 | * Allocates a new inode for given superblock. The default gfp_mask |
3c1d4378 | 641 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
769848c0 MG |
642 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
643 | * for the page cache are not reclaimable or migratable, | |
644 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
645 | * newly created inode's mapping | |
646 | * | |
1da177e4 LT |
647 | */ |
648 | struct inode *new_inode(struct super_block *sb) | |
649 | { | |
866b04fc JL |
650 | /* |
651 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
652 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
653 | * here to attempt to avoid that. | |
654 | */ | |
655 | static unsigned int last_ino; | |
6b3304b5 | 656 | struct inode *inode; |
1da177e4 LT |
657 | |
658 | spin_lock_prefetch(&inode_lock); | |
6b3304b5 | 659 | |
1da177e4 LT |
660 | inode = alloc_inode(sb); |
661 | if (inode) { | |
662 | spin_lock(&inode_lock); | |
8290c35f | 663 | __inode_add_to_lists(sb, NULL, inode); |
1da177e4 LT |
664 | inode->i_ino = ++last_ino; |
665 | inode->i_state = 0; | |
666 | spin_unlock(&inode_lock); | |
667 | } | |
668 | return inode; | |
669 | } | |
1da177e4 LT |
670 | EXPORT_SYMBOL(new_inode); |
671 | ||
672 | void unlock_new_inode(struct inode *inode) | |
673 | { | |
14358e6d | 674 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
1e89a5e1 PZ |
675 | if (inode->i_mode & S_IFDIR) { |
676 | struct file_system_type *type = inode->i_sb->s_type; | |
677 | ||
9a7aa12f JK |
678 | /* Set new key only if filesystem hasn't already changed it */ |
679 | if (!lockdep_match_class(&inode->i_mutex, | |
680 | &type->i_mutex_key)) { | |
681 | /* | |
682 | * ensure nobody is actually holding i_mutex | |
683 | */ | |
684 | mutex_destroy(&inode->i_mutex); | |
685 | mutex_init(&inode->i_mutex); | |
686 | lockdep_set_class(&inode->i_mutex, | |
687 | &type->i_mutex_dir_key); | |
688 | } | |
1e89a5e1 | 689 | } |
14358e6d | 690 | #endif |
1da177e4 | 691 | /* |
eaff8079 | 692 | * This is special! We do not need the spinlock when clearing I_NEW, |
580be083 JK |
693 | * because we're guaranteed that nobody else tries to do anything about |
694 | * the state of the inode when it is locked, as we just created it (so | |
eaff8079 | 695 | * there can be no old holders that haven't tested I_NEW). |
580be083 | 696 | * However we must emit the memory barrier so that other CPUs reliably |
eaff8079 | 697 | * see the clearing of I_NEW after the other inode initialisation has |
580be083 | 698 | * completed. |
1da177e4 | 699 | */ |
580be083 | 700 | smp_mb(); |
eaff8079 CH |
701 | WARN_ON(!(inode->i_state & I_NEW)); |
702 | inode->i_state &= ~I_NEW; | |
1da177e4 LT |
703 | wake_up_inode(inode); |
704 | } | |
1da177e4 LT |
705 | EXPORT_SYMBOL(unlock_new_inode); |
706 | ||
707 | /* | |
708 | * This is called without the inode lock held.. Be careful. | |
709 | * | |
710 | * We no longer cache the sb_flags in i_flags - see fs.h | |
711 | * -- rmk@arm.uk.linux.org | |
712 | */ | |
6b3304b5 MK |
713 | static struct inode *get_new_inode(struct super_block *sb, |
714 | struct hlist_head *head, | |
715 | int (*test)(struct inode *, void *), | |
716 | int (*set)(struct inode *, void *), | |
717 | void *data) | |
1da177e4 | 718 | { |
6b3304b5 | 719 | struct inode *inode; |
1da177e4 LT |
720 | |
721 | inode = alloc_inode(sb); | |
722 | if (inode) { | |
6b3304b5 | 723 | struct inode *old; |
1da177e4 LT |
724 | |
725 | spin_lock(&inode_lock); | |
726 | /* We released the lock, so.. */ | |
727 | old = find_inode(sb, head, test, data); | |
728 | if (!old) { | |
729 | if (set(inode, data)) | |
730 | goto set_failed; | |
731 | ||
8290c35f | 732 | __inode_add_to_lists(sb, head, inode); |
eaff8079 | 733 | inode->i_state = I_NEW; |
1da177e4 LT |
734 | spin_unlock(&inode_lock); |
735 | ||
736 | /* Return the locked inode with I_NEW set, the | |
737 | * caller is responsible for filling in the contents | |
738 | */ | |
739 | return inode; | |
740 | } | |
741 | ||
742 | /* | |
743 | * Uhhuh, somebody else created the same inode under | |
744 | * us. Use the old inode instead of the one we just | |
745 | * allocated. | |
746 | */ | |
747 | __iget(old); | |
748 | spin_unlock(&inode_lock); | |
749 | destroy_inode(inode); | |
750 | inode = old; | |
751 | wait_on_inode(inode); | |
752 | } | |
753 | return inode; | |
754 | ||
755 | set_failed: | |
756 | spin_unlock(&inode_lock); | |
757 | destroy_inode(inode); | |
758 | return NULL; | |
759 | } | |
760 | ||
761 | /* | |
762 | * get_new_inode_fast is the fast path version of get_new_inode, see the | |
763 | * comment at iget_locked for details. | |
764 | */ | |
6b3304b5 MK |
765 | static struct inode *get_new_inode_fast(struct super_block *sb, |
766 | struct hlist_head *head, unsigned long ino) | |
1da177e4 | 767 | { |
6b3304b5 | 768 | struct inode *inode; |
1da177e4 LT |
769 | |
770 | inode = alloc_inode(sb); | |
771 | if (inode) { | |
6b3304b5 | 772 | struct inode *old; |
1da177e4 LT |
773 | |
774 | spin_lock(&inode_lock); | |
775 | /* We released the lock, so.. */ | |
776 | old = find_inode_fast(sb, head, ino); | |
777 | if (!old) { | |
778 | inode->i_ino = ino; | |
8290c35f | 779 | __inode_add_to_lists(sb, head, inode); |
eaff8079 | 780 | inode->i_state = I_NEW; |
1da177e4 LT |
781 | spin_unlock(&inode_lock); |
782 | ||
783 | /* Return the locked inode with I_NEW set, the | |
784 | * caller is responsible for filling in the contents | |
785 | */ | |
786 | return inode; | |
787 | } | |
788 | ||
789 | /* | |
790 | * Uhhuh, somebody else created the same inode under | |
791 | * us. Use the old inode instead of the one we just | |
792 | * allocated. | |
793 | */ | |
794 | __iget(old); | |
795 | spin_unlock(&inode_lock); | |
796 | destroy_inode(inode); | |
797 | inode = old; | |
798 | wait_on_inode(inode); | |
799 | } | |
800 | return inode; | |
801 | } | |
802 | ||
1da177e4 LT |
803 | /** |
804 | * iunique - get a unique inode number | |
805 | * @sb: superblock | |
806 | * @max_reserved: highest reserved inode number | |
807 | * | |
808 | * Obtain an inode number that is unique on the system for a given | |
809 | * superblock. This is used by file systems that have no natural | |
810 | * permanent inode numbering system. An inode number is returned that | |
811 | * is higher than the reserved limit but unique. | |
812 | * | |
813 | * BUGS: | |
814 | * With a large number of inodes live on the file system this function | |
815 | * currently becomes quite slow. | |
816 | */ | |
817 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
818 | { | |
866b04fc JL |
819 | /* |
820 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
821 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
822 | * here to attempt to avoid that. | |
823 | */ | |
824 | static unsigned int counter; | |
1da177e4 | 825 | struct inode *inode; |
3361c7be | 826 | struct hlist_head *head; |
1da177e4 | 827 | ino_t res; |
3361c7be | 828 | |
1da177e4 | 829 | spin_lock(&inode_lock); |
3361c7be JL |
830 | do { |
831 | if (counter <= max_reserved) | |
832 | counter = max_reserved + 1; | |
1da177e4 | 833 | res = counter++; |
3361c7be | 834 | head = inode_hashtable + hash(sb, res); |
1da177e4 | 835 | inode = find_inode_fast(sb, head, res); |
3361c7be JL |
836 | } while (inode != NULL); |
837 | spin_unlock(&inode_lock); | |
1da177e4 | 838 | |
3361c7be JL |
839 | return res; |
840 | } | |
1da177e4 LT |
841 | EXPORT_SYMBOL(iunique); |
842 | ||
843 | struct inode *igrab(struct inode *inode) | |
844 | { | |
845 | spin_lock(&inode_lock); | |
a4ffdde6 | 846 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) |
1da177e4 LT |
847 | __iget(inode); |
848 | else | |
849 | /* | |
850 | * Handle the case where s_op->clear_inode is not been | |
851 | * called yet, and somebody is calling igrab | |
852 | * while the inode is getting freed. | |
853 | */ | |
854 | inode = NULL; | |
855 | spin_unlock(&inode_lock); | |
856 | return inode; | |
857 | } | |
1da177e4 LT |
858 | EXPORT_SYMBOL(igrab); |
859 | ||
860 | /** | |
861 | * ifind - internal function, you want ilookup5() or iget5(). | |
862 | * @sb: super block of file system to search | |
863 | * @head: the head of the list to search | |
864 | * @test: callback used for comparisons between inodes | |
865 | * @data: opaque data pointer to pass to @test | |
88bd5121 | 866 | * @wait: if true wait for the inode to be unlocked, if false do not |
1da177e4 LT |
867 | * |
868 | * ifind() searches for the inode specified by @data in the inode | |
869 | * cache. This is a generalized version of ifind_fast() for file systems where | |
870 | * the inode number is not sufficient for unique identification of an inode. | |
871 | * | |
872 | * If the inode is in the cache, the inode is returned with an incremented | |
873 | * reference count. | |
874 | * | |
875 | * Otherwise NULL is returned. | |
876 | * | |
877 | * Note, @test is called with the inode_lock held, so can't sleep. | |
878 | */ | |
5d2bea45 | 879 | static struct inode *ifind(struct super_block *sb, |
1da177e4 | 880 | struct hlist_head *head, int (*test)(struct inode *, void *), |
88bd5121 | 881 | void *data, const int wait) |
1da177e4 LT |
882 | { |
883 | struct inode *inode; | |
884 | ||
885 | spin_lock(&inode_lock); | |
886 | inode = find_inode(sb, head, test, data); | |
887 | if (inode) { | |
888 | __iget(inode); | |
889 | spin_unlock(&inode_lock); | |
88bd5121 AA |
890 | if (likely(wait)) |
891 | wait_on_inode(inode); | |
1da177e4 LT |
892 | return inode; |
893 | } | |
894 | spin_unlock(&inode_lock); | |
895 | return NULL; | |
896 | } | |
897 | ||
898 | /** | |
899 | * ifind_fast - internal function, you want ilookup() or iget(). | |
900 | * @sb: super block of file system to search | |
901 | * @head: head of the list to search | |
902 | * @ino: inode number to search for | |
903 | * | |
904 | * ifind_fast() searches for the inode @ino in the inode cache. This is for | |
905 | * file systems where the inode number is sufficient for unique identification | |
906 | * of an inode. | |
907 | * | |
908 | * If the inode is in the cache, the inode is returned with an incremented | |
909 | * reference count. | |
910 | * | |
911 | * Otherwise NULL is returned. | |
912 | */ | |
5d2bea45 | 913 | static struct inode *ifind_fast(struct super_block *sb, |
1da177e4 LT |
914 | struct hlist_head *head, unsigned long ino) |
915 | { | |
916 | struct inode *inode; | |
917 | ||
918 | spin_lock(&inode_lock); | |
919 | inode = find_inode_fast(sb, head, ino); | |
920 | if (inode) { | |
921 | __iget(inode); | |
922 | spin_unlock(&inode_lock); | |
923 | wait_on_inode(inode); | |
924 | return inode; | |
925 | } | |
926 | spin_unlock(&inode_lock); | |
927 | return NULL; | |
928 | } | |
929 | ||
930 | /** | |
88bd5121 | 931 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 LT |
932 | * @sb: super block of file system to search |
933 | * @hashval: hash value (usually inode number) to search for | |
934 | * @test: callback used for comparisons between inodes | |
935 | * @data: opaque data pointer to pass to @test | |
936 | * | |
937 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
938 | * @data in the inode cache. This is a generalized version of ilookup() for | |
939 | * file systems where the inode number is not sufficient for unique | |
940 | * identification of an inode. | |
941 | * | |
942 | * If the inode is in the cache, the inode is returned with an incremented | |
88bd5121 AA |
943 | * reference count. Note, the inode lock is not waited upon so you have to be |
944 | * very careful what you do with the returned inode. You probably should be | |
945 | * using ilookup5() instead. | |
946 | * | |
947 | * Otherwise NULL is returned. | |
948 | * | |
949 | * Note, @test is called with the inode_lock held, so can't sleep. | |
950 | */ | |
951 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, | |
952 | int (*test)(struct inode *, void *), void *data) | |
953 | { | |
954 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
955 | ||
956 | return ifind(sb, head, test, data, 0); | |
957 | } | |
88bd5121 AA |
958 | EXPORT_SYMBOL(ilookup5_nowait); |
959 | ||
960 | /** | |
961 | * ilookup5 - search for an inode in the inode cache | |
962 | * @sb: super block of file system to search | |
963 | * @hashval: hash value (usually inode number) to search for | |
964 | * @test: callback used for comparisons between inodes | |
965 | * @data: opaque data pointer to pass to @test | |
966 | * | |
967 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
968 | * @data in the inode cache. This is a generalized version of ilookup() for | |
969 | * file systems where the inode number is not sufficient for unique | |
970 | * identification of an inode. | |
971 | * | |
972 | * If the inode is in the cache, the inode lock is waited upon and the inode is | |
973 | * returned with an incremented reference count. | |
1da177e4 LT |
974 | * |
975 | * Otherwise NULL is returned. | |
976 | * | |
977 | * Note, @test is called with the inode_lock held, so can't sleep. | |
978 | */ | |
979 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
980 | int (*test)(struct inode *, void *), void *data) | |
981 | { | |
982 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
983 | ||
88bd5121 | 984 | return ifind(sb, head, test, data, 1); |
1da177e4 | 985 | } |
1da177e4 LT |
986 | EXPORT_SYMBOL(ilookup5); |
987 | ||
988 | /** | |
989 | * ilookup - search for an inode in the inode cache | |
990 | * @sb: super block of file system to search | |
991 | * @ino: inode number to search for | |
992 | * | |
993 | * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache. | |
994 | * This is for file systems where the inode number is sufficient for unique | |
995 | * identification of an inode. | |
996 | * | |
997 | * If the inode is in the cache, the inode is returned with an incremented | |
998 | * reference count. | |
999 | * | |
1000 | * Otherwise NULL is returned. | |
1001 | */ | |
1002 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1003 | { | |
1004 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1005 | ||
1006 | return ifind_fast(sb, head, ino); | |
1007 | } | |
1da177e4 LT |
1008 | EXPORT_SYMBOL(ilookup); |
1009 | ||
1010 | /** | |
1011 | * iget5_locked - obtain an inode from a mounted file system | |
1012 | * @sb: super block of file system | |
1013 | * @hashval: hash value (usually inode number) to get | |
1014 | * @test: callback used for comparisons between inodes | |
1015 | * @set: callback used to initialize a new struct inode | |
1016 | * @data: opaque data pointer to pass to @test and @set | |
1017 | * | |
1da177e4 LT |
1018 | * iget5_locked() uses ifind() to search for the inode specified by @hashval |
1019 | * and @data in the inode cache and if present it is returned with an increased | |
1020 | * reference count. This is a generalized version of iget_locked() for file | |
1021 | * systems where the inode number is not sufficient for unique identification | |
1022 | * of an inode. | |
1023 | * | |
1024 | * If the inode is not in cache, get_new_inode() is called to allocate a new | |
1025 | * inode and this is returned locked, hashed, and with the I_NEW flag set. The | |
1026 | * file system gets to fill it in before unlocking it via unlock_new_inode(). | |
1027 | * | |
1028 | * Note both @test and @set are called with the inode_lock held, so can't sleep. | |
1029 | */ | |
1030 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, | |
1031 | int (*test)(struct inode *, void *), | |
1032 | int (*set)(struct inode *, void *), void *data) | |
1033 | { | |
1034 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1035 | struct inode *inode; | |
1036 | ||
88bd5121 | 1037 | inode = ifind(sb, head, test, data, 1); |
1da177e4 LT |
1038 | if (inode) |
1039 | return inode; | |
1040 | /* | |
1041 | * get_new_inode() will do the right thing, re-trying the search | |
1042 | * in case it had to block at any point. | |
1043 | */ | |
1044 | return get_new_inode(sb, head, test, set, data); | |
1045 | } | |
1da177e4 LT |
1046 | EXPORT_SYMBOL(iget5_locked); |
1047 | ||
1048 | /** | |
1049 | * iget_locked - obtain an inode from a mounted file system | |
1050 | * @sb: super block of file system | |
1051 | * @ino: inode number to get | |
1052 | * | |
1da177e4 LT |
1053 | * iget_locked() uses ifind_fast() to search for the inode specified by @ino in |
1054 | * the inode cache and if present it is returned with an increased reference | |
1055 | * count. This is for file systems where the inode number is sufficient for | |
1056 | * unique identification of an inode. | |
1057 | * | |
1058 | * If the inode is not in cache, get_new_inode_fast() is called to allocate a | |
1059 | * new inode and this is returned locked, hashed, and with the I_NEW flag set. | |
1060 | * The file system gets to fill it in before unlocking it via | |
1061 | * unlock_new_inode(). | |
1062 | */ | |
1063 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) | |
1064 | { | |
1065 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1066 | struct inode *inode; | |
1067 | ||
1068 | inode = ifind_fast(sb, head, ino); | |
1069 | if (inode) | |
1070 | return inode; | |
1071 | /* | |
1072 | * get_new_inode_fast() will do the right thing, re-trying the search | |
1073 | * in case it had to block at any point. | |
1074 | */ | |
1075 | return get_new_inode_fast(sb, head, ino); | |
1076 | } | |
1da177e4 LT |
1077 | EXPORT_SYMBOL(iget_locked); |
1078 | ||
261bca86 AV |
1079 | int insert_inode_locked(struct inode *inode) |
1080 | { | |
1081 | struct super_block *sb = inode->i_sb; | |
1082 | ino_t ino = inode->i_ino; | |
1083 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
261bca86 | 1084 | |
eaff8079 | 1085 | inode->i_state |= I_NEW; |
261bca86 | 1086 | while (1) { |
72a43d63 AV |
1087 | struct hlist_node *node; |
1088 | struct inode *old = NULL; | |
261bca86 | 1089 | spin_lock(&inode_lock); |
72a43d63 AV |
1090 | hlist_for_each_entry(old, node, head, i_hash) { |
1091 | if (old->i_ino != ino) | |
1092 | continue; | |
1093 | if (old->i_sb != sb) | |
1094 | continue; | |
a4ffdde6 | 1095 | if (old->i_state & (I_FREEING|I_WILL_FREE)) |
72a43d63 AV |
1096 | continue; |
1097 | break; | |
1098 | } | |
1099 | if (likely(!node)) { | |
261bca86 AV |
1100 | hlist_add_head(&inode->i_hash, head); |
1101 | spin_unlock(&inode_lock); | |
1102 | return 0; | |
1103 | } | |
1104 | __iget(old); | |
1105 | spin_unlock(&inode_lock); | |
1106 | wait_on_inode(old); | |
1107 | if (unlikely(!hlist_unhashed(&old->i_hash))) { | |
1108 | iput(old); | |
1109 | return -EBUSY; | |
1110 | } | |
1111 | iput(old); | |
1112 | } | |
1113 | } | |
261bca86 AV |
1114 | EXPORT_SYMBOL(insert_inode_locked); |
1115 | ||
1116 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1117 | int (*test)(struct inode *, void *), void *data) | |
1118 | { | |
1119 | struct super_block *sb = inode->i_sb; | |
1120 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
261bca86 | 1121 | |
eaff8079 | 1122 | inode->i_state |= I_NEW; |
261bca86 AV |
1123 | |
1124 | while (1) { | |
72a43d63 AV |
1125 | struct hlist_node *node; |
1126 | struct inode *old = NULL; | |
1127 | ||
261bca86 | 1128 | spin_lock(&inode_lock); |
72a43d63 AV |
1129 | hlist_for_each_entry(old, node, head, i_hash) { |
1130 | if (old->i_sb != sb) | |
1131 | continue; | |
1132 | if (!test(old, data)) | |
1133 | continue; | |
a4ffdde6 | 1134 | if (old->i_state & (I_FREEING|I_WILL_FREE)) |
72a43d63 AV |
1135 | continue; |
1136 | break; | |
1137 | } | |
1138 | if (likely(!node)) { | |
261bca86 AV |
1139 | hlist_add_head(&inode->i_hash, head); |
1140 | spin_unlock(&inode_lock); | |
1141 | return 0; | |
1142 | } | |
1143 | __iget(old); | |
1144 | spin_unlock(&inode_lock); | |
1145 | wait_on_inode(old); | |
1146 | if (unlikely(!hlist_unhashed(&old->i_hash))) { | |
1147 | iput(old); | |
1148 | return -EBUSY; | |
1149 | } | |
1150 | iput(old); | |
1151 | } | |
1152 | } | |
261bca86 AV |
1153 | EXPORT_SYMBOL(insert_inode_locked4); |
1154 | ||
1da177e4 LT |
1155 | /** |
1156 | * __insert_inode_hash - hash an inode | |
1157 | * @inode: unhashed inode | |
1158 | * @hashval: unsigned long value used to locate this object in the | |
1159 | * inode_hashtable. | |
1160 | * | |
1161 | * Add an inode to the inode hash for this superblock. | |
1162 | */ | |
1163 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
1164 | { | |
1165 | struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval); | |
1166 | spin_lock(&inode_lock); | |
1167 | hlist_add_head(&inode->i_hash, head); | |
1168 | spin_unlock(&inode_lock); | |
1169 | } | |
1da177e4 LT |
1170 | EXPORT_SYMBOL(__insert_inode_hash); |
1171 | ||
1172 | /** | |
1173 | * remove_inode_hash - remove an inode from the hash | |
1174 | * @inode: inode to unhash | |
1175 | * | |
1176 | * Remove an inode from the superblock. | |
1177 | */ | |
1178 | void remove_inode_hash(struct inode *inode) | |
1179 | { | |
1180 | spin_lock(&inode_lock); | |
1181 | hlist_del_init(&inode->i_hash); | |
1182 | spin_unlock(&inode_lock); | |
1183 | } | |
1da177e4 LT |
1184 | EXPORT_SYMBOL(remove_inode_hash); |
1185 | ||
45321ac5 AV |
1186 | int generic_delete_inode(struct inode *inode) |
1187 | { | |
1188 | return 1; | |
1189 | } | |
1190 | EXPORT_SYMBOL(generic_delete_inode); | |
1191 | ||
1da177e4 | 1192 | /* |
45321ac5 AV |
1193 | * Normal UNIX filesystem behaviour: delete the |
1194 | * inode when the usage count drops to zero, and | |
1195 | * i_nlink is zero. | |
1da177e4 | 1196 | */ |
45321ac5 | 1197 | int generic_drop_inode(struct inode *inode) |
1da177e4 | 1198 | { |
45321ac5 | 1199 | return !inode->i_nlink || hlist_unhashed(&inode->i_hash); |
1da177e4 | 1200 | } |
45321ac5 | 1201 | EXPORT_SYMBOL_GPL(generic_drop_inode); |
1da177e4 | 1202 | |
45321ac5 AV |
1203 | /* |
1204 | * Called when we're dropping the last reference | |
1205 | * to an inode. | |
22fe4042 | 1206 | * |
45321ac5 AV |
1207 | * Call the FS "drop_inode()" function, defaulting to |
1208 | * the legacy UNIX filesystem behaviour. If it tells | |
1209 | * us to evict inode, do so. Otherwise, retain inode | |
1210 | * in cache if fs is alive, sync and evict if fs is | |
1211 | * shutting down. | |
22fe4042 | 1212 | */ |
45321ac5 | 1213 | static void iput_final(struct inode *inode) |
1da177e4 LT |
1214 | { |
1215 | struct super_block *sb = inode->i_sb; | |
45321ac5 AV |
1216 | const struct super_operations *op = inode->i_sb->s_op; |
1217 | int drop; | |
1218 | ||
1219 | if (op && op->drop_inode) | |
1220 | drop = op->drop_inode(inode); | |
1221 | else | |
1222 | drop = generic_drop_inode(inode); | |
1da177e4 | 1223 | |
45321ac5 | 1224 | if (!drop) { |
1c0eeaf5 | 1225 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) |
1da177e4 LT |
1226 | list_move(&inode->i_list, &inode_unused); |
1227 | inodes_stat.nr_unused++; | |
acb0c854 | 1228 | if (sb->s_flags & MS_ACTIVE) { |
991114c6 | 1229 | spin_unlock(&inode_lock); |
45321ac5 | 1230 | return; |
991114c6 | 1231 | } |
7ef0d737 | 1232 | WARN_ON(inode->i_state & I_NEW); |
991114c6 AV |
1233 | inode->i_state |= I_WILL_FREE; |
1234 | spin_unlock(&inode_lock); | |
1da177e4 LT |
1235 | write_inode_now(inode, 1); |
1236 | spin_lock(&inode_lock); | |
7ef0d737 | 1237 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1238 | inode->i_state &= ~I_WILL_FREE; |
1da177e4 LT |
1239 | inodes_stat.nr_unused--; |
1240 | hlist_del_init(&inode->i_hash); | |
1241 | } | |
1242 | list_del_init(&inode->i_list); | |
1243 | list_del_init(&inode->i_sb_list); | |
7ef0d737 | 1244 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1245 | inode->i_state |= I_FREEING; |
1da177e4 LT |
1246 | inodes_stat.nr_inodes--; |
1247 | spin_unlock(&inode_lock); | |
644da596 | 1248 | evict(inode); |
45321ac5 AV |
1249 | spin_lock(&inode_lock); |
1250 | hlist_del_init(&inode->i_hash); | |
1251 | spin_unlock(&inode_lock); | |
7f04c26d | 1252 | wake_up_inode(inode); |
45321ac5 | 1253 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); |
1da177e4 LT |
1254 | destroy_inode(inode); |
1255 | } | |
1256 | ||
1da177e4 | 1257 | /** |
6b3304b5 | 1258 | * iput - put an inode |
1da177e4 LT |
1259 | * @inode: inode to put |
1260 | * | |
1261 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1262 | * zero, the inode is then freed and may also be destroyed. | |
1263 | * | |
1264 | * Consequently, iput() can sleep. | |
1265 | */ | |
1266 | void iput(struct inode *inode) | |
1267 | { | |
1268 | if (inode) { | |
a4ffdde6 | 1269 | BUG_ON(inode->i_state & I_CLEAR); |
1da177e4 | 1270 | |
1da177e4 LT |
1271 | if (atomic_dec_and_lock(&inode->i_count, &inode_lock)) |
1272 | iput_final(inode); | |
1273 | } | |
1274 | } | |
1da177e4 LT |
1275 | EXPORT_SYMBOL(iput); |
1276 | ||
1277 | /** | |
1278 | * bmap - find a block number in a file | |
1279 | * @inode: inode of file | |
1280 | * @block: block to find | |
1281 | * | |
1282 | * Returns the block number on the device holding the inode that | |
1283 | * is the disk block number for the block of the file requested. | |
1284 | * That is, asked for block 4 of inode 1 the function will return the | |
6b3304b5 | 1285 | * disk block relative to the disk start that holds that block of the |
1da177e4 LT |
1286 | * file. |
1287 | */ | |
6b3304b5 | 1288 | sector_t bmap(struct inode *inode, sector_t block) |
1da177e4 LT |
1289 | { |
1290 | sector_t res = 0; | |
1291 | if (inode->i_mapping->a_ops->bmap) | |
1292 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1293 | return res; | |
1294 | } | |
1da177e4 LT |
1295 | EXPORT_SYMBOL(bmap); |
1296 | ||
11ff6f05 MG |
1297 | /* |
1298 | * With relative atime, only update atime if the previous atime is | |
1299 | * earlier than either the ctime or mtime or if at least a day has | |
1300 | * passed since the last atime update. | |
1301 | */ | |
1302 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1303 | struct timespec now) | |
1304 | { | |
1305 | ||
1306 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1307 | return 1; | |
1308 | /* | |
1309 | * Is mtime younger than atime? If yes, update atime: | |
1310 | */ | |
1311 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1312 | return 1; | |
1313 | /* | |
1314 | * Is ctime younger than atime? If yes, update atime: | |
1315 | */ | |
1316 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1317 | return 1; | |
1318 | ||
1319 | /* | |
1320 | * Is the previous atime value older than a day? If yes, | |
1321 | * update atime: | |
1322 | */ | |
1323 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1324 | return 1; | |
1325 | /* | |
1326 | * Good, we can skip the atime update: | |
1327 | */ | |
1328 | return 0; | |
1329 | } | |
1330 | ||
1da177e4 | 1331 | /** |
869243a0 CH |
1332 | * touch_atime - update the access time |
1333 | * @mnt: mount the inode is accessed on | |
7045f37b | 1334 | * @dentry: dentry accessed |
1da177e4 LT |
1335 | * |
1336 | * Update the accessed time on an inode and mark it for writeback. | |
1337 | * This function automatically handles read only file systems and media, | |
1338 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1339 | */ | |
869243a0 | 1340 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 1341 | { |
869243a0 | 1342 | struct inode *inode = dentry->d_inode; |
1da177e4 LT |
1343 | struct timespec now; |
1344 | ||
cdb70f3f | 1345 | if (inode->i_flags & S_NOATIME) |
b12536c2 | 1346 | return; |
37756ced | 1347 | if (IS_NOATIME(inode)) |
b12536c2 | 1348 | return; |
b2276138 | 1349 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1350 | return; |
47ae32d6 | 1351 | |
cdb70f3f | 1352 | if (mnt->mnt_flags & MNT_NOATIME) |
b12536c2 | 1353 | return; |
cdb70f3f | 1354 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) |
b12536c2 | 1355 | return; |
1da177e4 LT |
1356 | |
1357 | now = current_fs_time(inode->i_sb); | |
11ff6f05 MG |
1358 | |
1359 | if (!relatime_need_update(mnt, inode, now)) | |
b12536c2 | 1360 | return; |
11ff6f05 | 1361 | |
47ae32d6 | 1362 | if (timespec_equal(&inode->i_atime, &now)) |
b12536c2 AK |
1363 | return; |
1364 | ||
1365 | if (mnt_want_write(mnt)) | |
1366 | return; | |
47ae32d6 VH |
1367 | |
1368 | inode->i_atime = now; | |
1369 | mark_inode_dirty_sync(inode); | |
cdb70f3f | 1370 | mnt_drop_write(mnt); |
1da177e4 | 1371 | } |
869243a0 | 1372 | EXPORT_SYMBOL(touch_atime); |
1da177e4 LT |
1373 | |
1374 | /** | |
870f4817 CH |
1375 | * file_update_time - update mtime and ctime time |
1376 | * @file: file accessed | |
1da177e4 | 1377 | * |
870f4817 CH |
1378 | * Update the mtime and ctime members of an inode and mark the inode |
1379 | * for writeback. Note that this function is meant exclusively for | |
1380 | * usage in the file write path of filesystems, and filesystems may | |
1381 | * choose to explicitly ignore update via this function with the | |
2eadfc0e | 1382 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
870f4817 | 1383 | * timestamps are handled by the server. |
1da177e4 LT |
1384 | */ |
1385 | ||
870f4817 | 1386 | void file_update_time(struct file *file) |
1da177e4 | 1387 | { |
0f7fc9e4 | 1388 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 | 1389 | struct timespec now; |
ce06e0b2 | 1390 | enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0; |
1da177e4 | 1391 | |
ce06e0b2 | 1392 | /* First try to exhaust all avenues to not sync */ |
1da177e4 LT |
1393 | if (IS_NOCMTIME(inode)) |
1394 | return; | |
20ddee2c | 1395 | |
1da177e4 | 1396 | now = current_fs_time(inode->i_sb); |
ce06e0b2 AK |
1397 | if (!timespec_equal(&inode->i_mtime, &now)) |
1398 | sync_it = S_MTIME; | |
1da177e4 | 1399 | |
ce06e0b2 AK |
1400 | if (!timespec_equal(&inode->i_ctime, &now)) |
1401 | sync_it |= S_CTIME; | |
870f4817 | 1402 | |
ce06e0b2 AK |
1403 | if (IS_I_VERSION(inode)) |
1404 | sync_it |= S_VERSION; | |
7a224228 | 1405 | |
ce06e0b2 AK |
1406 | if (!sync_it) |
1407 | return; | |
1408 | ||
1409 | /* Finally allowed to write? Takes lock. */ | |
1410 | if (mnt_want_write_file(file)) | |
1411 | return; | |
1412 | ||
1413 | /* Only change inode inside the lock region */ | |
1414 | if (sync_it & S_VERSION) | |
1415 | inode_inc_iversion(inode); | |
1416 | if (sync_it & S_CTIME) | |
1417 | inode->i_ctime = now; | |
1418 | if (sync_it & S_MTIME) | |
1419 | inode->i_mtime = now; | |
1420 | mark_inode_dirty_sync(inode); | |
20ddee2c | 1421 | mnt_drop_write(file->f_path.mnt); |
1da177e4 | 1422 | } |
870f4817 | 1423 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1424 | |
1425 | int inode_needs_sync(struct inode *inode) | |
1426 | { | |
1427 | if (IS_SYNC(inode)) | |
1428 | return 1; | |
1429 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1430 | return 1; | |
1431 | return 0; | |
1432 | } | |
1da177e4 LT |
1433 | EXPORT_SYMBOL(inode_needs_sync); |
1434 | ||
1da177e4 LT |
1435 | int inode_wait(void *word) |
1436 | { | |
1437 | schedule(); | |
1438 | return 0; | |
1439 | } | |
d44dab8d | 1440 | EXPORT_SYMBOL(inode_wait); |
1da177e4 LT |
1441 | |
1442 | /* | |
168a9fd6 MS |
1443 | * If we try to find an inode in the inode hash while it is being |
1444 | * deleted, we have to wait until the filesystem completes its | |
1445 | * deletion before reporting that it isn't found. This function waits | |
1446 | * until the deletion _might_ have completed. Callers are responsible | |
1447 | * to recheck inode state. | |
1448 | * | |
eaff8079 | 1449 | * It doesn't matter if I_NEW is not set initially, a call to |
168a9fd6 MS |
1450 | * wake_up_inode() after removing from the hash list will DTRT. |
1451 | * | |
1da177e4 LT |
1452 | * This is called with inode_lock held. |
1453 | */ | |
1454 | static void __wait_on_freeing_inode(struct inode *inode) | |
1455 | { | |
1456 | wait_queue_head_t *wq; | |
eaff8079 CH |
1457 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1458 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1da177e4 LT |
1459 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
1460 | spin_unlock(&inode_lock); | |
1461 | schedule(); | |
1462 | finish_wait(wq, &wait.wait); | |
1463 | spin_lock(&inode_lock); | |
1464 | } | |
1465 | ||
1da177e4 LT |
1466 | static __initdata unsigned long ihash_entries; |
1467 | static int __init set_ihash_entries(char *str) | |
1468 | { | |
1469 | if (!str) | |
1470 | return 0; | |
1471 | ihash_entries = simple_strtoul(str, &str, 0); | |
1472 | return 1; | |
1473 | } | |
1474 | __setup("ihash_entries=", set_ihash_entries); | |
1475 | ||
1476 | /* | |
1477 | * Initialize the waitqueues and inode hash table. | |
1478 | */ | |
1479 | void __init inode_init_early(void) | |
1480 | { | |
1481 | int loop; | |
1482 | ||
1483 | /* If hashes are distributed across NUMA nodes, defer | |
1484 | * hash allocation until vmalloc space is available. | |
1485 | */ | |
1486 | if (hashdist) | |
1487 | return; | |
1488 | ||
1489 | inode_hashtable = | |
1490 | alloc_large_system_hash("Inode-cache", | |
1491 | sizeof(struct hlist_head), | |
1492 | ihash_entries, | |
1493 | 14, | |
1494 | HASH_EARLY, | |
1495 | &i_hash_shift, | |
1496 | &i_hash_mask, | |
1497 | 0); | |
1498 | ||
1499 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1500 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1501 | } | |
1502 | ||
74bf17cf | 1503 | void __init inode_init(void) |
1da177e4 LT |
1504 | { |
1505 | int loop; | |
1506 | ||
1507 | /* inode slab cache */ | |
b0196009 PJ |
1508 | inode_cachep = kmem_cache_create("inode_cache", |
1509 | sizeof(struct inode), | |
1510 | 0, | |
1511 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1512 | SLAB_MEM_SPREAD), | |
20c2df83 | 1513 | init_once); |
8e1f936b | 1514 | register_shrinker(&icache_shrinker); |
1da177e4 LT |
1515 | |
1516 | /* Hash may have been set up in inode_init_early */ | |
1517 | if (!hashdist) | |
1518 | return; | |
1519 | ||
1520 | inode_hashtable = | |
1521 | alloc_large_system_hash("Inode-cache", | |
1522 | sizeof(struct hlist_head), | |
1523 | ihash_entries, | |
1524 | 14, | |
1525 | 0, | |
1526 | &i_hash_shift, | |
1527 | &i_hash_mask, | |
1528 | 0); | |
1529 | ||
1530 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1531 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1532 | } | |
1533 | ||
1534 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1535 | { | |
1536 | inode->i_mode = mode; | |
1537 | if (S_ISCHR(mode)) { | |
1538 | inode->i_fop = &def_chr_fops; | |
1539 | inode->i_rdev = rdev; | |
1540 | } else if (S_ISBLK(mode)) { | |
1541 | inode->i_fop = &def_blk_fops; | |
1542 | inode->i_rdev = rdev; | |
1543 | } else if (S_ISFIFO(mode)) | |
1544 | inode->i_fop = &def_fifo_fops; | |
1545 | else if (S_ISSOCK(mode)) | |
1546 | inode->i_fop = &bad_sock_fops; | |
1547 | else | |
af0d9ae8 MK |
1548 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1549 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1550 | inode->i_ino); | |
1da177e4 LT |
1551 | } |
1552 | EXPORT_SYMBOL(init_special_inode); | |
a1bd120d DM |
1553 | |
1554 | /** | |
1555 | * Init uid,gid,mode for new inode according to posix standards | |
1556 | * @inode: New inode | |
1557 | * @dir: Directory inode | |
1558 | * @mode: mode of the new inode | |
1559 | */ | |
1560 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
1561 | mode_t mode) | |
1562 | { | |
1563 | inode->i_uid = current_fsuid(); | |
1564 | if (dir && dir->i_mode & S_ISGID) { | |
1565 | inode->i_gid = dir->i_gid; | |
1566 | if (S_ISDIR(mode)) | |
1567 | mode |= S_ISGID; | |
1568 | } else | |
1569 | inode->i_gid = current_fsgid(); | |
1570 | inode->i_mode = mode; | |
1571 | } | |
1572 | EXPORT_SYMBOL(inode_init_owner); |