]>
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> | |
6146f0d5 | 20 | #include <linux/ima.h> |
1da177e4 LT |
21 | #include <linux/pagemap.h> |
22 | #include <linux/cdev.h> | |
23 | #include <linux/bootmem.h> | |
0eeca283 | 24 | #include <linux/inotify.h> |
fc33a7bb | 25 | #include <linux/mount.h> |
efaee192 | 26 | #include <linux/async.h> |
1da177e4 LT |
27 | |
28 | /* | |
29 | * This is needed for the following functions: | |
30 | * - inode_has_buffers | |
31 | * - invalidate_inode_buffers | |
1da177e4 LT |
32 | * - invalidate_bdev |
33 | * | |
34 | * FIXME: remove all knowledge of the buffer layer from this file | |
35 | */ | |
36 | #include <linux/buffer_head.h> | |
37 | ||
38 | /* | |
39 | * New inode.c implementation. | |
40 | * | |
41 | * This implementation has the basic premise of trying | |
42 | * to be extremely low-overhead and SMP-safe, yet be | |
43 | * simple enough to be "obviously correct". | |
44 | * | |
45 | * Famous last words. | |
46 | */ | |
47 | ||
48 | /* inode dynamic allocation 1999, Andrea Arcangeli <andrea@suse.de> */ | |
49 | ||
50 | /* #define INODE_PARANOIA 1 */ | |
51 | /* #define INODE_DEBUG 1 */ | |
52 | ||
53 | /* | |
54 | * Inode lookup is no longer as critical as it used to be: | |
55 | * most of the lookups are going to be through the dcache. | |
56 | */ | |
57 | #define I_HASHBITS i_hash_shift | |
58 | #define I_HASHMASK i_hash_mask | |
59 | ||
fa3536cc ED |
60 | static unsigned int i_hash_mask __read_mostly; |
61 | static unsigned int i_hash_shift __read_mostly; | |
1da177e4 LT |
62 | |
63 | /* | |
64 | * Each inode can be on two separate lists. One is | |
65 | * the hash list of the inode, used for lookups. The | |
66 | * other linked list is the "type" list: | |
67 | * "in_use" - valid inode, i_count > 0, i_nlink > 0 | |
68 | * "dirty" - as "in_use" but also dirty | |
69 | * "unused" - valid inode, i_count = 0 | |
70 | * | |
71 | * A "dirty" list is maintained for each super block, | |
72 | * allowing for low-overhead inode sync() operations. | |
73 | */ | |
74 | ||
75 | LIST_HEAD(inode_in_use); | |
76 | LIST_HEAD(inode_unused); | |
fa3536cc | 77 | static struct hlist_head *inode_hashtable __read_mostly; |
1da177e4 LT |
78 | |
79 | /* | |
80 | * A simple spinlock to protect the list manipulations. | |
81 | * | |
82 | * NOTE! You also have to own the lock if you change | |
83 | * the i_state of an inode while it is in use.. | |
84 | */ | |
85 | DEFINE_SPINLOCK(inode_lock); | |
86 | ||
87 | /* | |
f24075bd | 88 | * iprune_mutex provides exclusion between the kswapd or try_to_free_pages |
1da177e4 LT |
89 | * icache shrinking path, and the umount path. Without this exclusion, |
90 | * by the time prune_icache calls iput for the inode whose pages it has | |
91 | * been invalidating, or by the time it calls clear_inode & destroy_inode | |
92 | * from its final dispose_list, the struct super_block they refer to | |
93 | * (for inode->i_sb->s_op) may already have been freed and reused. | |
94 | */ | |
bdfc3266 | 95 | static DEFINE_MUTEX(iprune_mutex); |
1da177e4 LT |
96 | |
97 | /* | |
98 | * Statistics gathering.. | |
99 | */ | |
100 | struct inodes_stat_t inodes_stat; | |
101 | ||
6b3304b5 | 102 | static struct kmem_cache *inode_cachep __read_mostly; |
1da177e4 | 103 | |
1c0eeaf5 JE |
104 | static void wake_up_inode(struct inode *inode) |
105 | { | |
106 | /* | |
107 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
108 | */ | |
109 | smp_mb(); | |
110 | wake_up_bit(&inode->i_state, __I_LOCK); | |
111 | } | |
112 | ||
2cb1599f DC |
113 | /** |
114 | * inode_init_always - perform inode structure intialisation | |
0bc02f3f RD |
115 | * @sb: superblock inode belongs to |
116 | * @inode: inode to initialise | |
2cb1599f DC |
117 | * |
118 | * These are initializations that need to be done on every inode | |
119 | * allocation as the fields are not initialised by slab allocation. | |
120 | */ | |
121 | struct inode *inode_init_always(struct super_block *sb, struct inode *inode) | |
1da177e4 | 122 | { |
f5e54d6e | 123 | static const struct address_space_operations empty_aops; |
1da177e4 | 124 | static struct inode_operations empty_iops; |
99ac48f5 | 125 | static const struct file_operations empty_fops; |
1da177e4 | 126 | |
6b3304b5 | 127 | struct address_space *const mapping = &inode->i_data; |
2cb1599f DC |
128 | |
129 | inode->i_sb = sb; | |
130 | inode->i_blkbits = sb->s_blocksize_bits; | |
131 | inode->i_flags = 0; | |
132 | atomic_set(&inode->i_count, 1); | |
133 | inode->i_op = &empty_iops; | |
134 | inode->i_fop = &empty_fops; | |
135 | inode->i_nlink = 1; | |
56ff5efa AV |
136 | inode->i_uid = 0; |
137 | inode->i_gid = 0; | |
2cb1599f DC |
138 | atomic_set(&inode->i_writecount, 0); |
139 | inode->i_size = 0; | |
140 | inode->i_blocks = 0; | |
141 | inode->i_bytes = 0; | |
142 | inode->i_generation = 0; | |
1da177e4 | 143 | #ifdef CONFIG_QUOTA |
2cb1599f | 144 | memset(&inode->i_dquot, 0, sizeof(inode->i_dquot)); |
1da177e4 | 145 | #endif |
2cb1599f DC |
146 | inode->i_pipe = NULL; |
147 | inode->i_bdev = NULL; | |
148 | inode->i_cdev = NULL; | |
149 | inode->i_rdev = 0; | |
150 | inode->dirtied_when = 0; | |
6146f0d5 MZ |
151 | |
152 | if (security_inode_alloc(inode)) | |
153 | goto out_free_inode; | |
154 | ||
155 | /* allocate and initialize an i_integrity */ | |
156 | if (ima_inode_alloc(inode)) | |
157 | goto out_free_security; | |
2cb1599f DC |
158 | |
159 | spin_lock_init(&inode->i_lock); | |
160 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
161 | ||
162 | mutex_init(&inode->i_mutex); | |
163 | lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key); | |
164 | ||
165 | init_rwsem(&inode->i_alloc_sem); | |
166 | lockdep_set_class(&inode->i_alloc_sem, &sb->s_type->i_alloc_sem_key); | |
167 | ||
168 | mapping->a_ops = &empty_aops; | |
169 | mapping->host = inode; | |
170 | mapping->flags = 0; | |
3c1d4378 | 171 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); |
2cb1599f DC |
172 | mapping->assoc_mapping = NULL; |
173 | mapping->backing_dev_info = &default_backing_dev_info; | |
174 | mapping->writeback_index = 0; | |
175 | ||
176 | /* | |
177 | * If the block_device provides a backing_dev_info for client | |
178 | * inodes then use that. Otherwise the inode share the bdev's | |
179 | * backing_dev_info. | |
180 | */ | |
181 | if (sb->s_bdev) { | |
182 | struct backing_dev_info *bdi; | |
183 | ||
184 | bdi = sb->s_bdev->bd_inode_backing_dev_info; | |
185 | if (!bdi) | |
186 | bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info; | |
187 | mapping->backing_dev_info = bdi; | |
188 | } | |
189 | inode->i_private = NULL; | |
190 | inode->i_mapping = mapping; | |
191 | ||
1da177e4 | 192 | return inode; |
6146f0d5 MZ |
193 | |
194 | out_free_security: | |
195 | security_inode_free(inode); | |
196 | out_free_inode: | |
197 | if (inode->i_sb->s_op->destroy_inode) | |
198 | inode->i_sb->s_op->destroy_inode(inode); | |
199 | else | |
200 | kmem_cache_free(inode_cachep, (inode)); | |
201 | return NULL; | |
1da177e4 | 202 | } |
2cb1599f DC |
203 | EXPORT_SYMBOL(inode_init_always); |
204 | ||
205 | static struct inode *alloc_inode(struct super_block *sb) | |
206 | { | |
207 | struct inode *inode; | |
208 | ||
209 | if (sb->s_op->alloc_inode) | |
210 | inode = sb->s_op->alloc_inode(sb); | |
211 | else | |
212 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
213 | ||
214 | if (inode) | |
215 | return inode_init_always(sb, inode); | |
216 | return NULL; | |
217 | } | |
1da177e4 | 218 | |
6b3304b5 | 219 | void destroy_inode(struct inode *inode) |
1da177e4 | 220 | { |
b7542f8c | 221 | BUG_ON(inode_has_buffers(inode)); |
1da177e4 LT |
222 | security_inode_free(inode); |
223 | if (inode->i_sb->s_op->destroy_inode) | |
224 | inode->i_sb->s_op->destroy_inode(inode); | |
225 | else | |
226 | kmem_cache_free(inode_cachep, (inode)); | |
227 | } | |
087e3b04 | 228 | EXPORT_SYMBOL(destroy_inode); |
1da177e4 LT |
229 | |
230 | ||
231 | /* | |
232 | * These are initializations that only need to be done | |
233 | * once, because the fields are idempotent across use | |
234 | * of the inode, so let the slab aware of that. | |
235 | */ | |
236 | void inode_init_once(struct inode *inode) | |
237 | { | |
238 | memset(inode, 0, sizeof(*inode)); | |
239 | INIT_HLIST_NODE(&inode->i_hash); | |
240 | INIT_LIST_HEAD(&inode->i_dentry); | |
241 | INIT_LIST_HEAD(&inode->i_devices); | |
1da177e4 | 242 | INIT_RADIX_TREE(&inode->i_data.page_tree, GFP_ATOMIC); |
19fd6231 | 243 | spin_lock_init(&inode->i_data.tree_lock); |
1da177e4 LT |
244 | spin_lock_init(&inode->i_data.i_mmap_lock); |
245 | INIT_LIST_HEAD(&inode->i_data.private_list); | |
246 | spin_lock_init(&inode->i_data.private_lock); | |
247 | INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap); | |
248 | INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear); | |
1da177e4 | 249 | i_size_ordered_init(inode); |
0eeca283 RL |
250 | #ifdef CONFIG_INOTIFY |
251 | INIT_LIST_HEAD(&inode->inotify_watches); | |
d4f9af9d | 252 | mutex_init(&inode->inotify_mutex); |
0eeca283 | 253 | #endif |
1da177e4 | 254 | } |
1da177e4 LT |
255 | EXPORT_SYMBOL(inode_init_once); |
256 | ||
51cc5068 | 257 | static void init_once(void *foo) |
1da177e4 | 258 | { |
6b3304b5 | 259 | struct inode *inode = (struct inode *) foo; |
1da177e4 | 260 | |
a35afb83 | 261 | inode_init_once(inode); |
1da177e4 LT |
262 | } |
263 | ||
264 | /* | |
265 | * inode_lock must be held | |
266 | */ | |
6b3304b5 | 267 | void __iget(struct inode *inode) |
1da177e4 LT |
268 | { |
269 | if (atomic_read(&inode->i_count)) { | |
270 | atomic_inc(&inode->i_count); | |
271 | return; | |
272 | } | |
273 | atomic_inc(&inode->i_count); | |
1c0eeaf5 | 274 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) |
1da177e4 LT |
275 | list_move(&inode->i_list, &inode_in_use); |
276 | inodes_stat.nr_unused--; | |
277 | } | |
278 | ||
279 | /** | |
280 | * clear_inode - clear an inode | |
281 | * @inode: inode to clear | |
282 | * | |
283 | * This is called by the filesystem to tell us | |
284 | * that the inode is no longer useful. We just | |
285 | * terminate it with extreme prejudice. | |
286 | */ | |
287 | void clear_inode(struct inode *inode) | |
288 | { | |
289 | might_sleep(); | |
290 | invalidate_inode_buffers(inode); | |
6b3304b5 | 291 | |
b7542f8c ES |
292 | BUG_ON(inode->i_data.nrpages); |
293 | BUG_ON(!(inode->i_state & I_FREEING)); | |
294 | BUG_ON(inode->i_state & I_CLEAR); | |
1c0eeaf5 | 295 | inode_sync_wait(inode); |
9e3509e2 | 296 | vfs_dq_drop(inode); |
acb0c854 | 297 | if (inode->i_sb->s_op->clear_inode) |
1da177e4 | 298 | inode->i_sb->s_op->clear_inode(inode); |
eaf796e7 | 299 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
1da177e4 | 300 | bd_forget(inode); |
577c4eb0 | 301 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) |
1da177e4 LT |
302 | cd_forget(inode); |
303 | inode->i_state = I_CLEAR; | |
304 | } | |
1da177e4 LT |
305 | EXPORT_SYMBOL(clear_inode); |
306 | ||
307 | /* | |
308 | * dispose_list - dispose of the contents of a local list | |
309 | * @head: the head of the list to free | |
310 | * | |
311 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
312 | * need to worry about list corruption and SMP locks. | |
313 | */ | |
314 | static void dispose_list(struct list_head *head) | |
315 | { | |
316 | int nr_disposed = 0; | |
317 | ||
318 | while (!list_empty(head)) { | |
319 | struct inode *inode; | |
320 | ||
b5e61818 | 321 | inode = list_first_entry(head, struct inode, i_list); |
1da177e4 LT |
322 | list_del(&inode->i_list); |
323 | ||
324 | if (inode->i_data.nrpages) | |
325 | truncate_inode_pages(&inode->i_data, 0); | |
326 | clear_inode(inode); | |
4120db47 AB |
327 | |
328 | spin_lock(&inode_lock); | |
329 | hlist_del_init(&inode->i_hash); | |
330 | list_del_init(&inode->i_sb_list); | |
331 | spin_unlock(&inode_lock); | |
332 | ||
333 | wake_up_inode(inode); | |
1da177e4 LT |
334 | destroy_inode(inode); |
335 | nr_disposed++; | |
336 | } | |
337 | spin_lock(&inode_lock); | |
338 | inodes_stat.nr_inodes -= nr_disposed; | |
339 | spin_unlock(&inode_lock); | |
340 | } | |
341 | ||
342 | /* | |
343 | * Invalidate all inodes for a device. | |
344 | */ | |
345 | static int invalidate_list(struct list_head *head, struct list_head *dispose) | |
346 | { | |
347 | struct list_head *next; | |
348 | int busy = 0, count = 0; | |
349 | ||
350 | next = head->next; | |
351 | for (;;) { | |
6b3304b5 MK |
352 | struct list_head *tmp = next; |
353 | struct inode *inode; | |
1da177e4 LT |
354 | |
355 | /* | |
356 | * We can reschedule here without worrying about the list's | |
357 | * consistency because the per-sb list of inodes must not | |
f24075bd | 358 | * change during umount anymore, and because iprune_mutex keeps |
1da177e4 LT |
359 | * shrink_icache_memory() away. |
360 | */ | |
361 | cond_resched_lock(&inode_lock); | |
362 | ||
363 | next = next->next; | |
364 | if (tmp == head) | |
365 | break; | |
366 | inode = list_entry(tmp, struct inode, i_sb_list); | |
aabb8fdb NP |
367 | if (inode->i_state & I_NEW) |
368 | continue; | |
1da177e4 LT |
369 | invalidate_inode_buffers(inode); |
370 | if (!atomic_read(&inode->i_count)) { | |
1da177e4 | 371 | list_move(&inode->i_list, dispose); |
7ef0d737 | 372 | WARN_ON(inode->i_state & I_NEW); |
1da177e4 LT |
373 | inode->i_state |= I_FREEING; |
374 | count++; | |
375 | continue; | |
376 | } | |
377 | busy = 1; | |
378 | } | |
379 | /* only unused inodes may be cached with i_count zero */ | |
380 | inodes_stat.nr_unused -= count; | |
381 | return busy; | |
382 | } | |
383 | ||
1da177e4 LT |
384 | /** |
385 | * invalidate_inodes - discard the inodes on a device | |
386 | * @sb: superblock | |
387 | * | |
388 | * Discard all of the inodes for a given superblock. If the discard | |
389 | * fails because there are busy inodes then a non zero value is returned. | |
390 | * If the discard is successful all the inodes have been discarded. | |
391 | */ | |
6b3304b5 | 392 | int invalidate_inodes(struct super_block *sb) |
1da177e4 LT |
393 | { |
394 | int busy; | |
395 | LIST_HEAD(throw_away); | |
396 | ||
f24075bd | 397 | mutex_lock(&iprune_mutex); |
1da177e4 | 398 | spin_lock(&inode_lock); |
0eeca283 | 399 | inotify_unmount_inodes(&sb->s_inodes); |
1da177e4 LT |
400 | busy = invalidate_list(&sb->s_inodes, &throw_away); |
401 | spin_unlock(&inode_lock); | |
402 | ||
403 | dispose_list(&throw_away); | |
f24075bd | 404 | mutex_unlock(&iprune_mutex); |
1da177e4 LT |
405 | |
406 | return busy; | |
407 | } | |
1da177e4 | 408 | EXPORT_SYMBOL(invalidate_inodes); |
1da177e4 LT |
409 | |
410 | static int can_unuse(struct inode *inode) | |
411 | { | |
412 | if (inode->i_state) | |
413 | return 0; | |
414 | if (inode_has_buffers(inode)) | |
415 | return 0; | |
416 | if (atomic_read(&inode->i_count)) | |
417 | return 0; | |
418 | if (inode->i_data.nrpages) | |
419 | return 0; | |
420 | return 1; | |
421 | } | |
422 | ||
423 | /* | |
424 | * Scan `goal' inodes on the unused list for freeable ones. They are moved to | |
425 | * a temporary list and then are freed outside inode_lock by dispose_list(). | |
426 | * | |
427 | * Any inodes which are pinned purely because of attached pagecache have their | |
428 | * pagecache removed. We expect the final iput() on that inode to add it to | |
429 | * the front of the inode_unused list. So look for it there and if the | |
430 | * inode is still freeable, proceed. The right inode is found 99.9% of the | |
431 | * time in testing on a 4-way. | |
432 | * | |
433 | * If the inode has metadata buffers attached to mapping->private_list then | |
434 | * try to remove them. | |
435 | */ | |
436 | static void prune_icache(int nr_to_scan) | |
437 | { | |
438 | LIST_HEAD(freeable); | |
439 | int nr_pruned = 0; | |
440 | int nr_scanned; | |
441 | unsigned long reap = 0; | |
442 | ||
f24075bd | 443 | mutex_lock(&iprune_mutex); |
1da177e4 LT |
444 | spin_lock(&inode_lock); |
445 | for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) { | |
446 | struct inode *inode; | |
447 | ||
448 | if (list_empty(&inode_unused)) | |
449 | break; | |
450 | ||
451 | inode = list_entry(inode_unused.prev, struct inode, i_list); | |
452 | ||
453 | if (inode->i_state || atomic_read(&inode->i_count)) { | |
454 | list_move(&inode->i_list, &inode_unused); | |
455 | continue; | |
456 | } | |
457 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
458 | __iget(inode); | |
459 | spin_unlock(&inode_lock); | |
460 | if (remove_inode_buffers(inode)) | |
fc0ecff6 AM |
461 | reap += invalidate_mapping_pages(&inode->i_data, |
462 | 0, -1); | |
1da177e4 LT |
463 | iput(inode); |
464 | spin_lock(&inode_lock); | |
465 | ||
466 | if (inode != list_entry(inode_unused.next, | |
467 | struct inode, i_list)) | |
468 | continue; /* wrong inode or list_empty */ | |
469 | if (!can_unuse(inode)) | |
470 | continue; | |
471 | } | |
1da177e4 | 472 | list_move(&inode->i_list, &freeable); |
7ef0d737 | 473 | WARN_ON(inode->i_state & I_NEW); |
1da177e4 LT |
474 | inode->i_state |= I_FREEING; |
475 | nr_pruned++; | |
476 | } | |
477 | inodes_stat.nr_unused -= nr_pruned; | |
f8891e5e CL |
478 | if (current_is_kswapd()) |
479 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
480 | else | |
481 | __count_vm_events(PGINODESTEAL, reap); | |
1da177e4 LT |
482 | spin_unlock(&inode_lock); |
483 | ||
484 | dispose_list(&freeable); | |
f24075bd | 485 | mutex_unlock(&iprune_mutex); |
1da177e4 LT |
486 | } |
487 | ||
488 | /* | |
489 | * shrink_icache_memory() will attempt to reclaim some unused inodes. Here, | |
490 | * "unused" means that no dentries are referring to the inodes: the files are | |
491 | * not open and the dcache references to those inodes have already been | |
492 | * reclaimed. | |
493 | * | |
494 | * This function is passed the number of inodes to scan, and it returns the | |
495 | * total number of remaining possibly-reclaimable inodes. | |
496 | */ | |
27496a8c | 497 | static int shrink_icache_memory(int nr, gfp_t gfp_mask) |
1da177e4 LT |
498 | { |
499 | if (nr) { | |
500 | /* | |
501 | * Nasty deadlock avoidance. We may hold various FS locks, | |
502 | * and we don't want to recurse into the FS that called us | |
503 | * in clear_inode() and friends.. | |
6b3304b5 | 504 | */ |
1da177e4 LT |
505 | if (!(gfp_mask & __GFP_FS)) |
506 | return -1; | |
507 | prune_icache(nr); | |
508 | } | |
509 | return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; | |
510 | } | |
511 | ||
8e1f936b RR |
512 | static struct shrinker icache_shrinker = { |
513 | .shrink = shrink_icache_memory, | |
514 | .seeks = DEFAULT_SEEKS, | |
515 | }; | |
516 | ||
1da177e4 LT |
517 | static void __wait_on_freeing_inode(struct inode *inode); |
518 | /* | |
519 | * Called with the inode lock held. | |
520 | * NOTE: we are not increasing the inode-refcount, you must call __iget() | |
521 | * by hand after calling find_inode now! This simplifies iunique and won't | |
522 | * add any additional branch in the common code. | |
523 | */ | |
6b3304b5 MK |
524 | static struct inode *find_inode(struct super_block *sb, |
525 | struct hlist_head *head, | |
526 | int (*test)(struct inode *, void *), | |
527 | void *data) | |
1da177e4 LT |
528 | { |
529 | struct hlist_node *node; | |
6b3304b5 | 530 | struct inode *inode = NULL; |
1da177e4 LT |
531 | |
532 | repeat: | |
c5c8be3c | 533 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
534 | if (inode->i_sb != sb) |
535 | continue; | |
536 | if (!test(inode, data)) | |
537 | continue; | |
991114c6 | 538 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) { |
1da177e4 LT |
539 | __wait_on_freeing_inode(inode); |
540 | goto repeat; | |
541 | } | |
542 | break; | |
543 | } | |
544 | return node ? inode : NULL; | |
545 | } | |
546 | ||
547 | /* | |
548 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
549 | * iget_locked for details. | |
550 | */ | |
6b3304b5 MK |
551 | static struct inode *find_inode_fast(struct super_block *sb, |
552 | struct hlist_head *head, unsigned long ino) | |
1da177e4 LT |
553 | { |
554 | struct hlist_node *node; | |
6b3304b5 | 555 | struct inode *inode = NULL; |
1da177e4 LT |
556 | |
557 | repeat: | |
c5c8be3c | 558 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
559 | if (inode->i_ino != ino) |
560 | continue; | |
561 | if (inode->i_sb != sb) | |
562 | continue; | |
991114c6 | 563 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) { |
1da177e4 LT |
564 | __wait_on_freeing_inode(inode); |
565 | goto repeat; | |
566 | } | |
567 | break; | |
568 | } | |
569 | return node ? inode : NULL; | |
570 | } | |
571 | ||
8290c35f DC |
572 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
573 | { | |
574 | unsigned long tmp; | |
575 | ||
576 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
577 | L1_CACHE_BYTES; | |
578 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS); | |
579 | return tmp & I_HASHMASK; | |
580 | } | |
581 | ||
582 | static inline void | |
583 | __inode_add_to_lists(struct super_block *sb, struct hlist_head *head, | |
584 | struct inode *inode) | |
585 | { | |
586 | inodes_stat.nr_inodes++; | |
587 | list_add(&inode->i_list, &inode_in_use); | |
588 | list_add(&inode->i_sb_list, &sb->s_inodes); | |
589 | if (head) | |
590 | hlist_add_head(&inode->i_hash, head); | |
591 | } | |
592 | ||
593 | /** | |
594 | * inode_add_to_lists - add a new inode to relevant lists | |
0bc02f3f RD |
595 | * @sb: superblock inode belongs to |
596 | * @inode: inode to mark in use | |
8290c35f DC |
597 | * |
598 | * When an inode is allocated it needs to be accounted for, added to the in use | |
599 | * list, the owning superblock and the inode hash. This needs to be done under | |
600 | * the inode_lock, so export a function to do this rather than the inode lock | |
601 | * itself. We calculate the hash list to add to here so it is all internal | |
602 | * which requires the caller to have already set up the inode number in the | |
603 | * inode to add. | |
604 | */ | |
605 | void inode_add_to_lists(struct super_block *sb, struct inode *inode) | |
606 | { | |
607 | struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino); | |
608 | ||
609 | spin_lock(&inode_lock); | |
610 | __inode_add_to_lists(sb, head, inode); | |
611 | spin_unlock(&inode_lock); | |
612 | } | |
613 | EXPORT_SYMBOL_GPL(inode_add_to_lists); | |
614 | ||
1da177e4 LT |
615 | /** |
616 | * new_inode - obtain an inode | |
617 | * @sb: superblock | |
618 | * | |
769848c0 | 619 | * Allocates a new inode for given superblock. The default gfp_mask |
3c1d4378 | 620 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
769848c0 MG |
621 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
622 | * for the page cache are not reclaimable or migratable, | |
623 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
624 | * newly created inode's mapping | |
625 | * | |
1da177e4 LT |
626 | */ |
627 | struct inode *new_inode(struct super_block *sb) | |
628 | { | |
866b04fc JL |
629 | /* |
630 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
631 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
632 | * here to attempt to avoid that. | |
633 | */ | |
634 | static unsigned int last_ino; | |
6b3304b5 | 635 | struct inode *inode; |
1da177e4 LT |
636 | |
637 | spin_lock_prefetch(&inode_lock); | |
6b3304b5 | 638 | |
1da177e4 LT |
639 | inode = alloc_inode(sb); |
640 | if (inode) { | |
641 | spin_lock(&inode_lock); | |
8290c35f | 642 | __inode_add_to_lists(sb, NULL, inode); |
1da177e4 LT |
643 | inode->i_ino = ++last_ino; |
644 | inode->i_state = 0; | |
645 | spin_unlock(&inode_lock); | |
646 | } | |
647 | return inode; | |
648 | } | |
1da177e4 LT |
649 | EXPORT_SYMBOL(new_inode); |
650 | ||
651 | void unlock_new_inode(struct inode *inode) | |
652 | { | |
14358e6d | 653 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
1e89a5e1 PZ |
654 | if (inode->i_mode & S_IFDIR) { |
655 | struct file_system_type *type = inode->i_sb->s_type; | |
656 | ||
657 | /* | |
658 | * ensure nobody is actually holding i_mutex | |
659 | */ | |
660 | mutex_destroy(&inode->i_mutex); | |
661 | mutex_init(&inode->i_mutex); | |
14358e6d | 662 | lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key); |
1e89a5e1 | 663 | } |
14358e6d | 664 | #endif |
1da177e4 LT |
665 | /* |
666 | * This is special! We do not need the spinlock | |
667 | * when clearing I_LOCK, because we're guaranteed | |
668 | * that nobody else tries to do anything about the | |
669 | * state of the inode when it is locked, as we | |
670 | * just created it (so there can be no old holders | |
671 | * that haven't tested I_LOCK). | |
672 | */ | |
7ef0d737 | 673 | WARN_ON((inode->i_state & (I_LOCK|I_NEW)) != (I_LOCK|I_NEW)); |
1da177e4 LT |
674 | inode->i_state &= ~(I_LOCK|I_NEW); |
675 | wake_up_inode(inode); | |
676 | } | |
1da177e4 LT |
677 | EXPORT_SYMBOL(unlock_new_inode); |
678 | ||
679 | /* | |
680 | * This is called without the inode lock held.. Be careful. | |
681 | * | |
682 | * We no longer cache the sb_flags in i_flags - see fs.h | |
683 | * -- rmk@arm.uk.linux.org | |
684 | */ | |
6b3304b5 MK |
685 | static struct inode *get_new_inode(struct super_block *sb, |
686 | struct hlist_head *head, | |
687 | int (*test)(struct inode *, void *), | |
688 | int (*set)(struct inode *, void *), | |
689 | void *data) | |
1da177e4 | 690 | { |
6b3304b5 | 691 | struct inode *inode; |
1da177e4 LT |
692 | |
693 | inode = alloc_inode(sb); | |
694 | if (inode) { | |
6b3304b5 | 695 | struct inode *old; |
1da177e4 LT |
696 | |
697 | spin_lock(&inode_lock); | |
698 | /* We released the lock, so.. */ | |
699 | old = find_inode(sb, head, test, data); | |
700 | if (!old) { | |
701 | if (set(inode, data)) | |
702 | goto set_failed; | |
703 | ||
8290c35f | 704 | __inode_add_to_lists(sb, head, inode); |
1da177e4 LT |
705 | inode->i_state = I_LOCK|I_NEW; |
706 | spin_unlock(&inode_lock); | |
707 | ||
708 | /* Return the locked inode with I_NEW set, the | |
709 | * caller is responsible for filling in the contents | |
710 | */ | |
711 | return inode; | |
712 | } | |
713 | ||
714 | /* | |
715 | * Uhhuh, somebody else created the same inode under | |
716 | * us. Use the old inode instead of the one we just | |
717 | * allocated. | |
718 | */ | |
719 | __iget(old); | |
720 | spin_unlock(&inode_lock); | |
721 | destroy_inode(inode); | |
722 | inode = old; | |
723 | wait_on_inode(inode); | |
724 | } | |
725 | return inode; | |
726 | ||
727 | set_failed: | |
728 | spin_unlock(&inode_lock); | |
729 | destroy_inode(inode); | |
730 | return NULL; | |
731 | } | |
732 | ||
733 | /* | |
734 | * get_new_inode_fast is the fast path version of get_new_inode, see the | |
735 | * comment at iget_locked for details. | |
736 | */ | |
6b3304b5 MK |
737 | static struct inode *get_new_inode_fast(struct super_block *sb, |
738 | struct hlist_head *head, unsigned long ino) | |
1da177e4 | 739 | { |
6b3304b5 | 740 | struct inode *inode; |
1da177e4 LT |
741 | |
742 | inode = alloc_inode(sb); | |
743 | if (inode) { | |
6b3304b5 | 744 | struct inode *old; |
1da177e4 LT |
745 | |
746 | spin_lock(&inode_lock); | |
747 | /* We released the lock, so.. */ | |
748 | old = find_inode_fast(sb, head, ino); | |
749 | if (!old) { | |
750 | inode->i_ino = ino; | |
8290c35f | 751 | __inode_add_to_lists(sb, head, inode); |
1da177e4 LT |
752 | inode->i_state = I_LOCK|I_NEW; |
753 | spin_unlock(&inode_lock); | |
754 | ||
755 | /* Return the locked inode with I_NEW set, the | |
756 | * caller is responsible for filling in the contents | |
757 | */ | |
758 | return inode; | |
759 | } | |
760 | ||
761 | /* | |
762 | * Uhhuh, somebody else created the same inode under | |
763 | * us. Use the old inode instead of the one we just | |
764 | * allocated. | |
765 | */ | |
766 | __iget(old); | |
767 | spin_unlock(&inode_lock); | |
768 | destroy_inode(inode); | |
769 | inode = old; | |
770 | wait_on_inode(inode); | |
771 | } | |
772 | return inode; | |
773 | } | |
774 | ||
1da177e4 LT |
775 | /** |
776 | * iunique - get a unique inode number | |
777 | * @sb: superblock | |
778 | * @max_reserved: highest reserved inode number | |
779 | * | |
780 | * Obtain an inode number that is unique on the system for a given | |
781 | * superblock. This is used by file systems that have no natural | |
782 | * permanent inode numbering system. An inode number is returned that | |
783 | * is higher than the reserved limit but unique. | |
784 | * | |
785 | * BUGS: | |
786 | * With a large number of inodes live on the file system this function | |
787 | * currently becomes quite slow. | |
788 | */ | |
789 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
790 | { | |
866b04fc JL |
791 | /* |
792 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
793 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
794 | * here to attempt to avoid that. | |
795 | */ | |
796 | static unsigned int counter; | |
1da177e4 | 797 | struct inode *inode; |
3361c7be | 798 | struct hlist_head *head; |
1da177e4 | 799 | ino_t res; |
3361c7be | 800 | |
1da177e4 | 801 | spin_lock(&inode_lock); |
3361c7be JL |
802 | do { |
803 | if (counter <= max_reserved) | |
804 | counter = max_reserved + 1; | |
1da177e4 | 805 | res = counter++; |
3361c7be | 806 | head = inode_hashtable + hash(sb, res); |
1da177e4 | 807 | inode = find_inode_fast(sb, head, res); |
3361c7be JL |
808 | } while (inode != NULL); |
809 | spin_unlock(&inode_lock); | |
1da177e4 | 810 | |
3361c7be JL |
811 | return res; |
812 | } | |
1da177e4 LT |
813 | EXPORT_SYMBOL(iunique); |
814 | ||
815 | struct inode *igrab(struct inode *inode) | |
816 | { | |
817 | spin_lock(&inode_lock); | |
4a3b0a49 | 818 | if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))) |
1da177e4 LT |
819 | __iget(inode); |
820 | else | |
821 | /* | |
822 | * Handle the case where s_op->clear_inode is not been | |
823 | * called yet, and somebody is calling igrab | |
824 | * while the inode is getting freed. | |
825 | */ | |
826 | inode = NULL; | |
827 | spin_unlock(&inode_lock); | |
828 | return inode; | |
829 | } | |
1da177e4 LT |
830 | EXPORT_SYMBOL(igrab); |
831 | ||
832 | /** | |
833 | * ifind - internal function, you want ilookup5() or iget5(). | |
834 | * @sb: super block of file system to search | |
835 | * @head: the head of the list to search | |
836 | * @test: callback used for comparisons between inodes | |
837 | * @data: opaque data pointer to pass to @test | |
88bd5121 | 838 | * @wait: if true wait for the inode to be unlocked, if false do not |
1da177e4 LT |
839 | * |
840 | * ifind() searches for the inode specified by @data in the inode | |
841 | * cache. This is a generalized version of ifind_fast() for file systems where | |
842 | * the inode number is not sufficient for unique identification of an inode. | |
843 | * | |
844 | * If the inode is in the cache, the inode is returned with an incremented | |
845 | * reference count. | |
846 | * | |
847 | * Otherwise NULL is returned. | |
848 | * | |
849 | * Note, @test is called with the inode_lock held, so can't sleep. | |
850 | */ | |
5d2bea45 | 851 | static struct inode *ifind(struct super_block *sb, |
1da177e4 | 852 | struct hlist_head *head, int (*test)(struct inode *, void *), |
88bd5121 | 853 | void *data, const int wait) |
1da177e4 LT |
854 | { |
855 | struct inode *inode; | |
856 | ||
857 | spin_lock(&inode_lock); | |
858 | inode = find_inode(sb, head, test, data); | |
859 | if (inode) { | |
860 | __iget(inode); | |
861 | spin_unlock(&inode_lock); | |
88bd5121 AA |
862 | if (likely(wait)) |
863 | wait_on_inode(inode); | |
1da177e4 LT |
864 | return inode; |
865 | } | |
866 | spin_unlock(&inode_lock); | |
867 | return NULL; | |
868 | } | |
869 | ||
870 | /** | |
871 | * ifind_fast - internal function, you want ilookup() or iget(). | |
872 | * @sb: super block of file system to search | |
873 | * @head: head of the list to search | |
874 | * @ino: inode number to search for | |
875 | * | |
876 | * ifind_fast() searches for the inode @ino in the inode cache. This is for | |
877 | * file systems where the inode number is sufficient for unique identification | |
878 | * of an inode. | |
879 | * | |
880 | * If the inode is in the cache, the inode is returned with an incremented | |
881 | * reference count. | |
882 | * | |
883 | * Otherwise NULL is returned. | |
884 | */ | |
5d2bea45 | 885 | static struct inode *ifind_fast(struct super_block *sb, |
1da177e4 LT |
886 | struct hlist_head *head, unsigned long ino) |
887 | { | |
888 | struct inode *inode; | |
889 | ||
890 | spin_lock(&inode_lock); | |
891 | inode = find_inode_fast(sb, head, ino); | |
892 | if (inode) { | |
893 | __iget(inode); | |
894 | spin_unlock(&inode_lock); | |
895 | wait_on_inode(inode); | |
896 | return inode; | |
897 | } | |
898 | spin_unlock(&inode_lock); | |
899 | return NULL; | |
900 | } | |
901 | ||
902 | /** | |
88bd5121 | 903 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 LT |
904 | * @sb: super block of file system to search |
905 | * @hashval: hash value (usually inode number) to search for | |
906 | * @test: callback used for comparisons between inodes | |
907 | * @data: opaque data pointer to pass to @test | |
908 | * | |
909 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
910 | * @data in the inode cache. This is a generalized version of ilookup() for | |
911 | * file systems where the inode number is not sufficient for unique | |
912 | * identification of an inode. | |
913 | * | |
914 | * If the inode is in the cache, the inode is returned with an incremented | |
88bd5121 AA |
915 | * reference count. Note, the inode lock is not waited upon so you have to be |
916 | * very careful what you do with the returned inode. You probably should be | |
917 | * using ilookup5() instead. | |
918 | * | |
919 | * Otherwise NULL is returned. | |
920 | * | |
921 | * Note, @test is called with the inode_lock held, so can't sleep. | |
922 | */ | |
923 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, | |
924 | int (*test)(struct inode *, void *), void *data) | |
925 | { | |
926 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
927 | ||
928 | return ifind(sb, head, test, data, 0); | |
929 | } | |
88bd5121 AA |
930 | EXPORT_SYMBOL(ilookup5_nowait); |
931 | ||
932 | /** | |
933 | * ilookup5 - search for an inode in the inode cache | |
934 | * @sb: super block of file system to search | |
935 | * @hashval: hash value (usually inode number) to search for | |
936 | * @test: callback used for comparisons between inodes | |
937 | * @data: opaque data pointer to pass to @test | |
938 | * | |
939 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
940 | * @data in the inode cache. This is a generalized version of ilookup() for | |
941 | * file systems where the inode number is not sufficient for unique | |
942 | * identification of an inode. | |
943 | * | |
944 | * If the inode is in the cache, the inode lock is waited upon and the inode is | |
945 | * returned with an incremented reference count. | |
1da177e4 LT |
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(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 | ||
88bd5121 | 956 | return ifind(sb, head, test, data, 1); |
1da177e4 | 957 | } |
1da177e4 LT |
958 | EXPORT_SYMBOL(ilookup5); |
959 | ||
960 | /** | |
961 | * ilookup - search for an inode in the inode cache | |
962 | * @sb: super block of file system to search | |
963 | * @ino: inode number to search for | |
964 | * | |
965 | * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache. | |
966 | * This is for file systems where the inode number is sufficient for unique | |
967 | * identification of an inode. | |
968 | * | |
969 | * If the inode is in the cache, the inode is returned with an incremented | |
970 | * reference count. | |
971 | * | |
972 | * Otherwise NULL is returned. | |
973 | */ | |
974 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
975 | { | |
976 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
977 | ||
978 | return ifind_fast(sb, head, ino); | |
979 | } | |
1da177e4 LT |
980 | EXPORT_SYMBOL(ilookup); |
981 | ||
982 | /** | |
983 | * iget5_locked - obtain an inode from a mounted file system | |
984 | * @sb: super block of file system | |
985 | * @hashval: hash value (usually inode number) to get | |
986 | * @test: callback used for comparisons between inodes | |
987 | * @set: callback used to initialize a new struct inode | |
988 | * @data: opaque data pointer to pass to @test and @set | |
989 | * | |
1da177e4 LT |
990 | * iget5_locked() uses ifind() to search for the inode specified by @hashval |
991 | * and @data in the inode cache and if present it is returned with an increased | |
992 | * reference count. This is a generalized version of iget_locked() for file | |
993 | * systems where the inode number is not sufficient for unique identification | |
994 | * of an inode. | |
995 | * | |
996 | * If the inode is not in cache, get_new_inode() is called to allocate a new | |
997 | * inode and this is returned locked, hashed, and with the I_NEW flag set. The | |
998 | * file system gets to fill it in before unlocking it via unlock_new_inode(). | |
999 | * | |
1000 | * Note both @test and @set are called with the inode_lock held, so can't sleep. | |
1001 | */ | |
1002 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, | |
1003 | int (*test)(struct inode *, void *), | |
1004 | int (*set)(struct inode *, void *), void *data) | |
1005 | { | |
1006 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1007 | struct inode *inode; | |
1008 | ||
88bd5121 | 1009 | inode = ifind(sb, head, test, data, 1); |
1da177e4 LT |
1010 | if (inode) |
1011 | return inode; | |
1012 | /* | |
1013 | * get_new_inode() will do the right thing, re-trying the search | |
1014 | * in case it had to block at any point. | |
1015 | */ | |
1016 | return get_new_inode(sb, head, test, set, data); | |
1017 | } | |
1da177e4 LT |
1018 | EXPORT_SYMBOL(iget5_locked); |
1019 | ||
1020 | /** | |
1021 | * iget_locked - obtain an inode from a mounted file system | |
1022 | * @sb: super block of file system | |
1023 | * @ino: inode number to get | |
1024 | * | |
1da177e4 LT |
1025 | * iget_locked() uses ifind_fast() to search for the inode specified by @ino in |
1026 | * the inode cache and if present it is returned with an increased reference | |
1027 | * count. This is for file systems where the inode number is sufficient for | |
1028 | * unique identification of an inode. | |
1029 | * | |
1030 | * If the inode is not in cache, get_new_inode_fast() is called to allocate a | |
1031 | * new inode and this is returned locked, hashed, and with the I_NEW flag set. | |
1032 | * The file system gets to fill it in before unlocking it via | |
1033 | * unlock_new_inode(). | |
1034 | */ | |
1035 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) | |
1036 | { | |
1037 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1038 | struct inode *inode; | |
1039 | ||
1040 | inode = ifind_fast(sb, head, ino); | |
1041 | if (inode) | |
1042 | return inode; | |
1043 | /* | |
1044 | * get_new_inode_fast() will do the right thing, re-trying the search | |
1045 | * in case it had to block at any point. | |
1046 | */ | |
1047 | return get_new_inode_fast(sb, head, ino); | |
1048 | } | |
1da177e4 LT |
1049 | EXPORT_SYMBOL(iget_locked); |
1050 | ||
261bca86 AV |
1051 | int insert_inode_locked(struct inode *inode) |
1052 | { | |
1053 | struct super_block *sb = inode->i_sb; | |
1054 | ino_t ino = inode->i_ino; | |
1055 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1056 | struct inode *old; | |
1057 | ||
1058 | inode->i_state |= I_LOCK|I_NEW; | |
1059 | while (1) { | |
1060 | spin_lock(&inode_lock); | |
1061 | old = find_inode_fast(sb, head, ino); | |
1062 | if (likely(!old)) { | |
1063 | hlist_add_head(&inode->i_hash, head); | |
1064 | spin_unlock(&inode_lock); | |
1065 | return 0; | |
1066 | } | |
1067 | __iget(old); | |
1068 | spin_unlock(&inode_lock); | |
1069 | wait_on_inode(old); | |
1070 | if (unlikely(!hlist_unhashed(&old->i_hash))) { | |
1071 | iput(old); | |
1072 | return -EBUSY; | |
1073 | } | |
1074 | iput(old); | |
1075 | } | |
1076 | } | |
261bca86 AV |
1077 | EXPORT_SYMBOL(insert_inode_locked); |
1078 | ||
1079 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1080 | int (*test)(struct inode *, void *), void *data) | |
1081 | { | |
1082 | struct super_block *sb = inode->i_sb; | |
1083 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1084 | struct inode *old; | |
1085 | ||
1086 | inode->i_state |= I_LOCK|I_NEW; | |
1087 | ||
1088 | while (1) { | |
1089 | spin_lock(&inode_lock); | |
1090 | old = find_inode(sb, head, test, data); | |
1091 | if (likely(!old)) { | |
1092 | hlist_add_head(&inode->i_hash, head); | |
1093 | spin_unlock(&inode_lock); | |
1094 | return 0; | |
1095 | } | |
1096 | __iget(old); | |
1097 | spin_unlock(&inode_lock); | |
1098 | wait_on_inode(old); | |
1099 | if (unlikely(!hlist_unhashed(&old->i_hash))) { | |
1100 | iput(old); | |
1101 | return -EBUSY; | |
1102 | } | |
1103 | iput(old); | |
1104 | } | |
1105 | } | |
261bca86 AV |
1106 | EXPORT_SYMBOL(insert_inode_locked4); |
1107 | ||
1da177e4 LT |
1108 | /** |
1109 | * __insert_inode_hash - hash an inode | |
1110 | * @inode: unhashed inode | |
1111 | * @hashval: unsigned long value used to locate this object in the | |
1112 | * inode_hashtable. | |
1113 | * | |
1114 | * Add an inode to the inode hash for this superblock. | |
1115 | */ | |
1116 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
1117 | { | |
1118 | struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval); | |
1119 | spin_lock(&inode_lock); | |
1120 | hlist_add_head(&inode->i_hash, head); | |
1121 | spin_unlock(&inode_lock); | |
1122 | } | |
1da177e4 LT |
1123 | EXPORT_SYMBOL(__insert_inode_hash); |
1124 | ||
1125 | /** | |
1126 | * remove_inode_hash - remove an inode from the hash | |
1127 | * @inode: inode to unhash | |
1128 | * | |
1129 | * Remove an inode from the superblock. | |
1130 | */ | |
1131 | void remove_inode_hash(struct inode *inode) | |
1132 | { | |
1133 | spin_lock(&inode_lock); | |
1134 | hlist_del_init(&inode->i_hash); | |
1135 | spin_unlock(&inode_lock); | |
1136 | } | |
1da177e4 LT |
1137 | EXPORT_SYMBOL(remove_inode_hash); |
1138 | ||
1139 | /* | |
1140 | * Tell the filesystem that this inode is no longer of any interest and should | |
1141 | * be completely destroyed. | |
1142 | * | |
1143 | * We leave the inode in the inode hash table until *after* the filesystem's | |
1144 | * ->delete_inode completes. This ensures that an iget (such as nfsd might | |
1145 | * instigate) will always find up-to-date information either in the hash or on | |
1146 | * disk. | |
1147 | * | |
1148 | * I_FREEING is set so that no-one will take a new reference to the inode while | |
1149 | * it is being deleted. | |
1150 | */ | |
b32714ba | 1151 | void generic_delete_inode(struct inode *inode) |
1da177e4 | 1152 | { |
ee9b6d61 | 1153 | const struct super_operations *op = inode->i_sb->s_op; |
1da177e4 | 1154 | |
b32714ba AV |
1155 | list_del_init(&inode->i_list); |
1156 | list_del_init(&inode->i_sb_list); | |
7ef0d737 | 1157 | WARN_ON(inode->i_state & I_NEW); |
b32714ba AV |
1158 | inode->i_state |= I_FREEING; |
1159 | inodes_stat.nr_inodes--; | |
1160 | spin_unlock(&inode_lock); | |
1161 | ||
1da177e4 LT |
1162 | security_inode_delete(inode); |
1163 | ||
1164 | if (op->delete_inode) { | |
1165 | void (*delete)(struct inode *) = op->delete_inode; | |
1166 | if (!is_bad_inode(inode)) | |
9e3509e2 | 1167 | vfs_dq_init(inode); |
e85b5652 MF |
1168 | /* Filesystems implementing their own |
1169 | * s_op->delete_inode are required to call | |
1170 | * truncate_inode_pages and clear_inode() | |
1171 | * internally */ | |
1da177e4 | 1172 | delete(inode); |
e85b5652 MF |
1173 | } else { |
1174 | truncate_inode_pages(&inode->i_data, 0); | |
1da177e4 | 1175 | clear_inode(inode); |
e85b5652 | 1176 | } |
1da177e4 LT |
1177 | spin_lock(&inode_lock); |
1178 | hlist_del_init(&inode->i_hash); | |
1179 | spin_unlock(&inode_lock); | |
1180 | wake_up_inode(inode); | |
b7542f8c | 1181 | BUG_ON(inode->i_state != I_CLEAR); |
1da177e4 LT |
1182 | destroy_inode(inode); |
1183 | } | |
1da177e4 LT |
1184 | EXPORT_SYMBOL(generic_delete_inode); |
1185 | ||
1186 | static void generic_forget_inode(struct inode *inode) | |
1187 | { | |
1188 | struct super_block *sb = inode->i_sb; | |
1189 | ||
1190 | if (!hlist_unhashed(&inode->i_hash)) { | |
1c0eeaf5 | 1191 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) |
1da177e4 LT |
1192 | list_move(&inode->i_list, &inode_unused); |
1193 | inodes_stat.nr_unused++; | |
acb0c854 | 1194 | if (sb->s_flags & MS_ACTIVE) { |
991114c6 | 1195 | spin_unlock(&inode_lock); |
1da177e4 | 1196 | return; |
991114c6 | 1197 | } |
7ef0d737 | 1198 | WARN_ON(inode->i_state & I_NEW); |
991114c6 AV |
1199 | inode->i_state |= I_WILL_FREE; |
1200 | spin_unlock(&inode_lock); | |
1da177e4 LT |
1201 | write_inode_now(inode, 1); |
1202 | spin_lock(&inode_lock); | |
7ef0d737 | 1203 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1204 | inode->i_state &= ~I_WILL_FREE; |
1da177e4 LT |
1205 | inodes_stat.nr_unused--; |
1206 | hlist_del_init(&inode->i_hash); | |
1207 | } | |
1208 | list_del_init(&inode->i_list); | |
1209 | list_del_init(&inode->i_sb_list); | |
7ef0d737 | 1210 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1211 | inode->i_state |= I_FREEING; |
1da177e4 LT |
1212 | inodes_stat.nr_inodes--; |
1213 | spin_unlock(&inode_lock); | |
1214 | if (inode->i_data.nrpages) | |
1215 | truncate_inode_pages(&inode->i_data, 0); | |
1216 | clear_inode(inode); | |
7f04c26d | 1217 | wake_up_inode(inode); |
1da177e4 LT |
1218 | destroy_inode(inode); |
1219 | } | |
1220 | ||
1221 | /* | |
1222 | * Normal UNIX filesystem behaviour: delete the | |
1223 | * inode when the usage count drops to zero, and | |
1224 | * i_nlink is zero. | |
1225 | */ | |
cb2c0233 | 1226 | void generic_drop_inode(struct inode *inode) |
1da177e4 LT |
1227 | { |
1228 | if (!inode->i_nlink) | |
1229 | generic_delete_inode(inode); | |
1230 | else | |
1231 | generic_forget_inode(inode); | |
1232 | } | |
cb2c0233 MF |
1233 | EXPORT_SYMBOL_GPL(generic_drop_inode); |
1234 | ||
1da177e4 LT |
1235 | /* |
1236 | * Called when we're dropping the last reference | |
6b3304b5 | 1237 | * to an inode. |
1da177e4 LT |
1238 | * |
1239 | * Call the FS "drop()" function, defaulting to | |
1240 | * the legacy UNIX filesystem behaviour.. | |
1241 | * | |
1242 | * NOTE! NOTE! NOTE! We're called with the inode lock | |
1243 | * held, and the drop function is supposed to release | |
1244 | * the lock! | |
1245 | */ | |
1246 | static inline void iput_final(struct inode *inode) | |
1247 | { | |
ee9b6d61 | 1248 | const struct super_operations *op = inode->i_sb->s_op; |
1da177e4 LT |
1249 | void (*drop)(struct inode *) = generic_drop_inode; |
1250 | ||
1251 | if (op && op->drop_inode) | |
1252 | drop = op->drop_inode; | |
1253 | drop(inode); | |
1254 | } | |
1255 | ||
1256 | /** | |
6b3304b5 | 1257 | * iput - put an inode |
1da177e4 LT |
1258 | * @inode: inode to put |
1259 | * | |
1260 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1261 | * zero, the inode is then freed and may also be destroyed. | |
1262 | * | |
1263 | * Consequently, iput() can sleep. | |
1264 | */ | |
1265 | void iput(struct inode *inode) | |
1266 | { | |
1267 | if (inode) { | |
1da177e4 LT |
1268 | BUG_ON(inode->i_state == I_CLEAR); |
1269 | ||
1da177e4 LT |
1270 | if (atomic_dec_and_lock(&inode->i_count, &inode_lock)) |
1271 | iput_final(inode); | |
1272 | } | |
1273 | } | |
1da177e4 LT |
1274 | EXPORT_SYMBOL(iput); |
1275 | ||
1276 | /** | |
1277 | * bmap - find a block number in a file | |
1278 | * @inode: inode of file | |
1279 | * @block: block to find | |
1280 | * | |
1281 | * Returns the block number on the device holding the inode that | |
1282 | * is the disk block number for the block of the file requested. | |
1283 | * That is, asked for block 4 of inode 1 the function will return the | |
6b3304b5 | 1284 | * disk block relative to the disk start that holds that block of the |
1da177e4 LT |
1285 | * file. |
1286 | */ | |
6b3304b5 | 1287 | sector_t bmap(struct inode *inode, sector_t block) |
1da177e4 LT |
1288 | { |
1289 | sector_t res = 0; | |
1290 | if (inode->i_mapping->a_ops->bmap) | |
1291 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1292 | return res; | |
1293 | } | |
1da177e4 LT |
1294 | EXPORT_SYMBOL(bmap); |
1295 | ||
11ff6f05 MG |
1296 | /* |
1297 | * With relative atime, only update atime if the previous atime is | |
1298 | * earlier than either the ctime or mtime or if at least a day has | |
1299 | * passed since the last atime update. | |
1300 | */ | |
1301 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1302 | struct timespec now) | |
1303 | { | |
1304 | ||
1305 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1306 | return 1; | |
1307 | /* | |
1308 | * Is mtime younger than atime? If yes, update atime: | |
1309 | */ | |
1310 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1311 | return 1; | |
1312 | /* | |
1313 | * Is ctime younger than atime? If yes, update atime: | |
1314 | */ | |
1315 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1316 | return 1; | |
1317 | ||
1318 | /* | |
1319 | * Is the previous atime value older than a day? If yes, | |
1320 | * update atime: | |
1321 | */ | |
1322 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1323 | return 1; | |
1324 | /* | |
1325 | * Good, we can skip the atime update: | |
1326 | */ | |
1327 | return 0; | |
1328 | } | |
1329 | ||
1da177e4 | 1330 | /** |
869243a0 CH |
1331 | * touch_atime - update the access time |
1332 | * @mnt: mount the inode is accessed on | |
7045f37b | 1333 | * @dentry: dentry accessed |
1da177e4 LT |
1334 | * |
1335 | * Update the accessed time on an inode and mark it for writeback. | |
1336 | * This function automatically handles read only file systems and media, | |
1337 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1338 | */ | |
869243a0 | 1339 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 1340 | { |
869243a0 | 1341 | struct inode *inode = dentry->d_inode; |
1da177e4 LT |
1342 | struct timespec now; |
1343 | ||
cdb70f3f | 1344 | if (mnt_want_write(mnt)) |
b2276138 | 1345 | return; |
cdb70f3f DH |
1346 | if (inode->i_flags & S_NOATIME) |
1347 | goto out; | |
37756ced | 1348 | if (IS_NOATIME(inode)) |
cdb70f3f | 1349 | goto out; |
b2276138 | 1350 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
cdb70f3f | 1351 | goto out; |
47ae32d6 | 1352 | |
cdb70f3f DH |
1353 | if (mnt->mnt_flags & MNT_NOATIME) |
1354 | goto out; | |
1355 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1356 | goto out; | |
1da177e4 LT |
1357 | |
1358 | now = current_fs_time(inode->i_sb); | |
11ff6f05 MG |
1359 | |
1360 | if (!relatime_need_update(mnt, inode, now)) | |
1361 | goto out; | |
1362 | ||
47ae32d6 | 1363 | if (timespec_equal(&inode->i_atime, &now)) |
cdb70f3f | 1364 | goto out; |
47ae32d6 VH |
1365 | |
1366 | inode->i_atime = now; | |
1367 | mark_inode_dirty_sync(inode); | |
cdb70f3f DH |
1368 | out: |
1369 | mnt_drop_write(mnt); | |
1da177e4 | 1370 | } |
869243a0 | 1371 | EXPORT_SYMBOL(touch_atime); |
1da177e4 LT |
1372 | |
1373 | /** | |
870f4817 CH |
1374 | * file_update_time - update mtime and ctime time |
1375 | * @file: file accessed | |
1da177e4 | 1376 | * |
870f4817 CH |
1377 | * Update the mtime and ctime members of an inode and mark the inode |
1378 | * for writeback. Note that this function is meant exclusively for | |
1379 | * usage in the file write path of filesystems, and filesystems may | |
1380 | * choose to explicitly ignore update via this function with the | |
1381 | * S_NOCTIME inode flag, e.g. for network filesystem where these | |
1382 | * timestamps are handled by the server. | |
1da177e4 LT |
1383 | */ |
1384 | ||
870f4817 | 1385 | void file_update_time(struct file *file) |
1da177e4 | 1386 | { |
0f7fc9e4 | 1387 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
1388 | struct timespec now; |
1389 | int sync_it = 0; | |
20ddee2c | 1390 | int err; |
1da177e4 LT |
1391 | |
1392 | if (IS_NOCMTIME(inode)) | |
1393 | return; | |
20ddee2c DH |
1394 | |
1395 | err = mnt_want_write(file->f_path.mnt); | |
1396 | if (err) | |
1da177e4 LT |
1397 | return; |
1398 | ||
1399 | now = current_fs_time(inode->i_sb); | |
ed97bd37 AM |
1400 | if (!timespec_equal(&inode->i_mtime, &now)) { |
1401 | inode->i_mtime = now; | |
1da177e4 | 1402 | sync_it = 1; |
ed97bd37 | 1403 | } |
1da177e4 | 1404 | |
ed97bd37 AM |
1405 | if (!timespec_equal(&inode->i_ctime, &now)) { |
1406 | inode->i_ctime = now; | |
870f4817 | 1407 | sync_it = 1; |
ed97bd37 | 1408 | } |
870f4817 | 1409 | |
7a224228 JNC |
1410 | if (IS_I_VERSION(inode)) { |
1411 | inode_inc_iversion(inode); | |
1412 | sync_it = 1; | |
1413 | } | |
1414 | ||
1da177e4 LT |
1415 | if (sync_it) |
1416 | mark_inode_dirty_sync(inode); | |
20ddee2c | 1417 | mnt_drop_write(file->f_path.mnt); |
1da177e4 | 1418 | } |
870f4817 | 1419 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1420 | |
1421 | int inode_needs_sync(struct inode *inode) | |
1422 | { | |
1423 | if (IS_SYNC(inode)) | |
1424 | return 1; | |
1425 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1426 | return 1; | |
1427 | return 0; | |
1428 | } | |
1da177e4 LT |
1429 | EXPORT_SYMBOL(inode_needs_sync); |
1430 | ||
1da177e4 LT |
1431 | int inode_wait(void *word) |
1432 | { | |
1433 | schedule(); | |
1434 | return 0; | |
1435 | } | |
d44dab8d | 1436 | EXPORT_SYMBOL(inode_wait); |
1da177e4 LT |
1437 | |
1438 | /* | |
168a9fd6 MS |
1439 | * If we try to find an inode in the inode hash while it is being |
1440 | * deleted, we have to wait until the filesystem completes its | |
1441 | * deletion before reporting that it isn't found. This function waits | |
1442 | * until the deletion _might_ have completed. Callers are responsible | |
1443 | * to recheck inode state. | |
1444 | * | |
1445 | * It doesn't matter if I_LOCK is not set initially, a call to | |
1446 | * wake_up_inode() after removing from the hash list will DTRT. | |
1447 | * | |
1da177e4 LT |
1448 | * This is called with inode_lock held. |
1449 | */ | |
1450 | static void __wait_on_freeing_inode(struct inode *inode) | |
1451 | { | |
1452 | wait_queue_head_t *wq; | |
1453 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK); | |
1da177e4 LT |
1454 | wq = bit_waitqueue(&inode->i_state, __I_LOCK); |
1455 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
1456 | spin_unlock(&inode_lock); | |
1457 | schedule(); | |
1458 | finish_wait(wq, &wait.wait); | |
1459 | spin_lock(&inode_lock); | |
1460 | } | |
1461 | ||
1da177e4 LT |
1462 | static __initdata unsigned long ihash_entries; |
1463 | static int __init set_ihash_entries(char *str) | |
1464 | { | |
1465 | if (!str) | |
1466 | return 0; | |
1467 | ihash_entries = simple_strtoul(str, &str, 0); | |
1468 | return 1; | |
1469 | } | |
1470 | __setup("ihash_entries=", set_ihash_entries); | |
1471 | ||
1472 | /* | |
1473 | * Initialize the waitqueues and inode hash table. | |
1474 | */ | |
1475 | void __init inode_init_early(void) | |
1476 | { | |
1477 | int loop; | |
1478 | ||
1479 | /* If hashes are distributed across NUMA nodes, defer | |
1480 | * hash allocation until vmalloc space is available. | |
1481 | */ | |
1482 | if (hashdist) | |
1483 | return; | |
1484 | ||
1485 | inode_hashtable = | |
1486 | alloc_large_system_hash("Inode-cache", | |
1487 | sizeof(struct hlist_head), | |
1488 | ihash_entries, | |
1489 | 14, | |
1490 | HASH_EARLY, | |
1491 | &i_hash_shift, | |
1492 | &i_hash_mask, | |
1493 | 0); | |
1494 | ||
1495 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1496 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1497 | } | |
1498 | ||
74bf17cf | 1499 | void __init inode_init(void) |
1da177e4 LT |
1500 | { |
1501 | int loop; | |
1502 | ||
1503 | /* inode slab cache */ | |
b0196009 PJ |
1504 | inode_cachep = kmem_cache_create("inode_cache", |
1505 | sizeof(struct inode), | |
1506 | 0, | |
1507 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1508 | SLAB_MEM_SPREAD), | |
20c2df83 | 1509 | init_once); |
8e1f936b | 1510 | register_shrinker(&icache_shrinker); |
1da177e4 LT |
1511 | |
1512 | /* Hash may have been set up in inode_init_early */ | |
1513 | if (!hashdist) | |
1514 | return; | |
1515 | ||
1516 | inode_hashtable = | |
1517 | alloc_large_system_hash("Inode-cache", | |
1518 | sizeof(struct hlist_head), | |
1519 | ihash_entries, | |
1520 | 14, | |
1521 | 0, | |
1522 | &i_hash_shift, | |
1523 | &i_hash_mask, | |
1524 | 0); | |
1525 | ||
1526 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1527 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1528 | } | |
1529 | ||
1530 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1531 | { | |
1532 | inode->i_mode = mode; | |
1533 | if (S_ISCHR(mode)) { | |
1534 | inode->i_fop = &def_chr_fops; | |
1535 | inode->i_rdev = rdev; | |
1536 | } else if (S_ISBLK(mode)) { | |
1537 | inode->i_fop = &def_blk_fops; | |
1538 | inode->i_rdev = rdev; | |
1539 | } else if (S_ISFIFO(mode)) | |
1540 | inode->i_fop = &def_fifo_fops; | |
1541 | else if (S_ISSOCK(mode)) | |
1542 | inode->i_fop = &bad_sock_fops; | |
1543 | else | |
1544 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n", | |
1545 | mode); | |
1546 | } | |
1547 | EXPORT_SYMBOL(init_special_inode); |