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 | ||
e18b890b | 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 | |
2cb1599f DC |
127 | struct address_space * const mapping = &inode->i_data; |
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 LT |
218 | |
219 | void destroy_inode(struct inode *inode) | |
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 LT |
254 | } |
255 | ||
256 | EXPORT_SYMBOL(inode_init_once); | |
257 | ||
51cc5068 | 258 | static void init_once(void *foo) |
1da177e4 LT |
259 | { |
260 | struct inode * inode = (struct inode *) foo; | |
261 | ||
a35afb83 | 262 | inode_init_once(inode); |
1da177e4 LT |
263 | } |
264 | ||
265 | /* | |
266 | * inode_lock must be held | |
267 | */ | |
268 | void __iget(struct inode * inode) | |
269 | { | |
270 | if (atomic_read(&inode->i_count)) { | |
271 | atomic_inc(&inode->i_count); | |
272 | return; | |
273 | } | |
274 | atomic_inc(&inode->i_count); | |
1c0eeaf5 | 275 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) |
1da177e4 LT |
276 | list_move(&inode->i_list, &inode_in_use); |
277 | inodes_stat.nr_unused--; | |
278 | } | |
279 | ||
280 | /** | |
281 | * clear_inode - clear an inode | |
282 | * @inode: inode to clear | |
283 | * | |
284 | * This is called by the filesystem to tell us | |
285 | * that the inode is no longer useful. We just | |
286 | * terminate it with extreme prejudice. | |
287 | */ | |
288 | void clear_inode(struct inode *inode) | |
289 | { | |
290 | might_sleep(); | |
291 | invalidate_inode_buffers(inode); | |
292 | ||
b7542f8c ES |
293 | BUG_ON(inode->i_data.nrpages); |
294 | BUG_ON(!(inode->i_state & I_FREEING)); | |
295 | BUG_ON(inode->i_state & I_CLEAR); | |
1c0eeaf5 | 296 | inode_sync_wait(inode); |
1da177e4 | 297 | DQUOT_DROP(inode); |
acb0c854 | 298 | if (inode->i_sb->s_op->clear_inode) |
1da177e4 | 299 | inode->i_sb->s_op->clear_inode(inode); |
eaf796e7 | 300 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
1da177e4 | 301 | bd_forget(inode); |
577c4eb0 | 302 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) |
1da177e4 LT |
303 | cd_forget(inode); |
304 | inode->i_state = I_CLEAR; | |
305 | } | |
306 | ||
307 | EXPORT_SYMBOL(clear_inode); | |
308 | ||
309 | /* | |
310 | * dispose_list - dispose of the contents of a local list | |
311 | * @head: the head of the list to free | |
312 | * | |
313 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
314 | * need to worry about list corruption and SMP locks. | |
315 | */ | |
316 | static void dispose_list(struct list_head *head) | |
317 | { | |
318 | int nr_disposed = 0; | |
319 | ||
320 | while (!list_empty(head)) { | |
321 | struct inode *inode; | |
322 | ||
b5e61818 | 323 | inode = list_first_entry(head, struct inode, i_list); |
1da177e4 LT |
324 | list_del(&inode->i_list); |
325 | ||
326 | if (inode->i_data.nrpages) | |
327 | truncate_inode_pages(&inode->i_data, 0); | |
328 | clear_inode(inode); | |
4120db47 AB |
329 | |
330 | spin_lock(&inode_lock); | |
331 | hlist_del_init(&inode->i_hash); | |
332 | list_del_init(&inode->i_sb_list); | |
333 | spin_unlock(&inode_lock); | |
334 | ||
335 | wake_up_inode(inode); | |
1da177e4 LT |
336 | destroy_inode(inode); |
337 | nr_disposed++; | |
338 | } | |
339 | spin_lock(&inode_lock); | |
340 | inodes_stat.nr_inodes -= nr_disposed; | |
341 | spin_unlock(&inode_lock); | |
342 | } | |
343 | ||
344 | /* | |
345 | * Invalidate all inodes for a device. | |
346 | */ | |
347 | static int invalidate_list(struct list_head *head, struct list_head *dispose) | |
348 | { | |
349 | struct list_head *next; | |
350 | int busy = 0, count = 0; | |
351 | ||
352 | next = head->next; | |
353 | for (;;) { | |
354 | struct list_head * tmp = next; | |
355 | struct inode * inode; | |
356 | ||
357 | /* | |
358 | * We can reschedule here without worrying about the list's | |
359 | * consistency because the per-sb list of inodes must not | |
f24075bd | 360 | * change during umount anymore, and because iprune_mutex keeps |
1da177e4 LT |
361 | * shrink_icache_memory() away. |
362 | */ | |
363 | cond_resched_lock(&inode_lock); | |
364 | ||
365 | next = next->next; | |
366 | if (tmp == head) | |
367 | break; | |
368 | inode = list_entry(tmp, struct inode, i_sb_list); | |
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 | */ | |
392 | int invalidate_inodes(struct super_block * sb) | |
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 | } | |
408 | ||
409 | EXPORT_SYMBOL(invalidate_inodes); | |
1da177e4 LT |
410 | |
411 | static int can_unuse(struct inode *inode) | |
412 | { | |
413 | if (inode->i_state) | |
414 | return 0; | |
415 | if (inode_has_buffers(inode)) | |
416 | return 0; | |
417 | if (atomic_read(&inode->i_count)) | |
418 | return 0; | |
419 | if (inode->i_data.nrpages) | |
420 | return 0; | |
421 | return 1; | |
422 | } | |
423 | ||
424 | /* | |
425 | * Scan `goal' inodes on the unused list for freeable ones. They are moved to | |
426 | * a temporary list and then are freed outside inode_lock by dispose_list(). | |
427 | * | |
428 | * Any inodes which are pinned purely because of attached pagecache have their | |
429 | * pagecache removed. We expect the final iput() on that inode to add it to | |
430 | * the front of the inode_unused list. So look for it there and if the | |
431 | * inode is still freeable, proceed. The right inode is found 99.9% of the | |
432 | * time in testing on a 4-way. | |
433 | * | |
434 | * If the inode has metadata buffers attached to mapping->private_list then | |
435 | * try to remove them. | |
436 | */ | |
437 | static void prune_icache(int nr_to_scan) | |
438 | { | |
439 | LIST_HEAD(freeable); | |
440 | int nr_pruned = 0; | |
441 | int nr_scanned; | |
442 | unsigned long reap = 0; | |
443 | ||
f24075bd | 444 | mutex_lock(&iprune_mutex); |
1da177e4 LT |
445 | spin_lock(&inode_lock); |
446 | for (nr_scanned = 0; nr_scanned < nr_to_scan; nr_scanned++) { | |
447 | struct inode *inode; | |
448 | ||
449 | if (list_empty(&inode_unused)) | |
450 | break; | |
451 | ||
452 | inode = list_entry(inode_unused.prev, struct inode, i_list); | |
453 | ||
454 | if (inode->i_state || atomic_read(&inode->i_count)) { | |
455 | list_move(&inode->i_list, &inode_unused); | |
456 | continue; | |
457 | } | |
458 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { | |
459 | __iget(inode); | |
460 | spin_unlock(&inode_lock); | |
461 | if (remove_inode_buffers(inode)) | |
fc0ecff6 AM |
462 | reap += invalidate_mapping_pages(&inode->i_data, |
463 | 0, -1); | |
1da177e4 LT |
464 | iput(inode); |
465 | spin_lock(&inode_lock); | |
466 | ||
467 | if (inode != list_entry(inode_unused.next, | |
468 | struct inode, i_list)) | |
469 | continue; /* wrong inode or list_empty */ | |
470 | if (!can_unuse(inode)) | |
471 | continue; | |
472 | } | |
1da177e4 | 473 | list_move(&inode->i_list, &freeable); |
7ef0d737 | 474 | WARN_ON(inode->i_state & I_NEW); |
1da177e4 LT |
475 | inode->i_state |= I_FREEING; |
476 | nr_pruned++; | |
477 | } | |
478 | inodes_stat.nr_unused -= nr_pruned; | |
f8891e5e CL |
479 | if (current_is_kswapd()) |
480 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
481 | else | |
482 | __count_vm_events(PGINODESTEAL, reap); | |
1da177e4 LT |
483 | spin_unlock(&inode_lock); |
484 | ||
485 | dispose_list(&freeable); | |
f24075bd | 486 | mutex_unlock(&iprune_mutex); |
1da177e4 LT |
487 | } |
488 | ||
489 | /* | |
490 | * shrink_icache_memory() will attempt to reclaim some unused inodes. Here, | |
491 | * "unused" means that no dentries are referring to the inodes: the files are | |
492 | * not open and the dcache references to those inodes have already been | |
493 | * reclaimed. | |
494 | * | |
495 | * This function is passed the number of inodes to scan, and it returns the | |
496 | * total number of remaining possibly-reclaimable inodes. | |
497 | */ | |
27496a8c | 498 | static int shrink_icache_memory(int nr, gfp_t gfp_mask) |
1da177e4 LT |
499 | { |
500 | if (nr) { | |
501 | /* | |
502 | * Nasty deadlock avoidance. We may hold various FS locks, | |
503 | * and we don't want to recurse into the FS that called us | |
504 | * in clear_inode() and friends.. | |
505 | */ | |
506 | if (!(gfp_mask & __GFP_FS)) | |
507 | return -1; | |
508 | prune_icache(nr); | |
509 | } | |
510 | return (inodes_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; | |
511 | } | |
512 | ||
8e1f936b RR |
513 | static struct shrinker icache_shrinker = { |
514 | .shrink = shrink_icache_memory, | |
515 | .seeks = DEFAULT_SEEKS, | |
516 | }; | |
517 | ||
1da177e4 LT |
518 | static void __wait_on_freeing_inode(struct inode *inode); |
519 | /* | |
520 | * Called with the inode lock held. | |
521 | * NOTE: we are not increasing the inode-refcount, you must call __iget() | |
522 | * by hand after calling find_inode now! This simplifies iunique and won't | |
523 | * add any additional branch in the common code. | |
524 | */ | |
525 | static struct inode * find_inode(struct super_block * sb, struct hlist_head *head, int (*test)(struct inode *, void *), void *data) | |
526 | { | |
527 | struct hlist_node *node; | |
528 | struct inode * inode = NULL; | |
529 | ||
530 | repeat: | |
c5c8be3c | 531 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
532 | if (inode->i_sb != sb) |
533 | continue; | |
534 | if (!test(inode, data)) | |
535 | continue; | |
991114c6 | 536 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) { |
1da177e4 LT |
537 | __wait_on_freeing_inode(inode); |
538 | goto repeat; | |
539 | } | |
540 | break; | |
541 | } | |
542 | return node ? inode : NULL; | |
543 | } | |
544 | ||
545 | /* | |
546 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
547 | * iget_locked for details. | |
548 | */ | |
549 | static struct inode * find_inode_fast(struct super_block * sb, struct hlist_head *head, unsigned long ino) | |
550 | { | |
551 | struct hlist_node *node; | |
552 | struct inode * inode = NULL; | |
553 | ||
554 | repeat: | |
c5c8be3c | 555 | hlist_for_each_entry(inode, node, head, i_hash) { |
1da177e4 LT |
556 | if (inode->i_ino != ino) |
557 | continue; | |
558 | if (inode->i_sb != sb) | |
559 | continue; | |
991114c6 | 560 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE)) { |
1da177e4 LT |
561 | __wait_on_freeing_inode(inode); |
562 | goto repeat; | |
563 | } | |
564 | break; | |
565 | } | |
566 | return node ? inode : NULL; | |
567 | } | |
568 | ||
8290c35f DC |
569 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
570 | { | |
571 | unsigned long tmp; | |
572 | ||
573 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
574 | L1_CACHE_BYTES; | |
575 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> I_HASHBITS); | |
576 | return tmp & I_HASHMASK; | |
577 | } | |
578 | ||
579 | static inline void | |
580 | __inode_add_to_lists(struct super_block *sb, struct hlist_head *head, | |
581 | struct inode *inode) | |
582 | { | |
583 | inodes_stat.nr_inodes++; | |
584 | list_add(&inode->i_list, &inode_in_use); | |
585 | list_add(&inode->i_sb_list, &sb->s_inodes); | |
586 | if (head) | |
587 | hlist_add_head(&inode->i_hash, head); | |
588 | } | |
589 | ||
590 | /** | |
591 | * inode_add_to_lists - add a new inode to relevant lists | |
0bc02f3f RD |
592 | * @sb: superblock inode belongs to |
593 | * @inode: inode to mark in use | |
8290c35f DC |
594 | * |
595 | * When an inode is allocated it needs to be accounted for, added to the in use | |
596 | * list, the owning superblock and the inode hash. This needs to be done under | |
597 | * the inode_lock, so export a function to do this rather than the inode lock | |
598 | * itself. We calculate the hash list to add to here so it is all internal | |
599 | * which requires the caller to have already set up the inode number in the | |
600 | * inode to add. | |
601 | */ | |
602 | void inode_add_to_lists(struct super_block *sb, struct inode *inode) | |
603 | { | |
604 | struct hlist_head *head = inode_hashtable + hash(sb, inode->i_ino); | |
605 | ||
606 | spin_lock(&inode_lock); | |
607 | __inode_add_to_lists(sb, head, inode); | |
608 | spin_unlock(&inode_lock); | |
609 | } | |
610 | EXPORT_SYMBOL_GPL(inode_add_to_lists); | |
611 | ||
1da177e4 LT |
612 | /** |
613 | * new_inode - obtain an inode | |
614 | * @sb: superblock | |
615 | * | |
769848c0 | 616 | * Allocates a new inode for given superblock. The default gfp_mask |
3c1d4378 | 617 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
769848c0 MG |
618 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
619 | * for the page cache are not reclaimable or migratable, | |
620 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
621 | * newly created inode's mapping | |
622 | * | |
1da177e4 LT |
623 | */ |
624 | struct inode *new_inode(struct super_block *sb) | |
625 | { | |
866b04fc JL |
626 | /* |
627 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
628 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
629 | * here to attempt to avoid that. | |
630 | */ | |
631 | static unsigned int last_ino; | |
1da177e4 LT |
632 | struct inode * inode; |
633 | ||
634 | spin_lock_prefetch(&inode_lock); | |
635 | ||
636 | inode = alloc_inode(sb); | |
637 | if (inode) { | |
638 | spin_lock(&inode_lock); | |
8290c35f | 639 | __inode_add_to_lists(sb, NULL, inode); |
1da177e4 LT |
640 | inode->i_ino = ++last_ino; |
641 | inode->i_state = 0; | |
642 | spin_unlock(&inode_lock); | |
643 | } | |
644 | return inode; | |
645 | } | |
646 | ||
647 | EXPORT_SYMBOL(new_inode); | |
648 | ||
649 | void unlock_new_inode(struct inode *inode) | |
650 | { | |
14358e6d | 651 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
1e89a5e1 PZ |
652 | if (inode->i_mode & S_IFDIR) { |
653 | struct file_system_type *type = inode->i_sb->s_type; | |
654 | ||
655 | /* | |
656 | * ensure nobody is actually holding i_mutex | |
657 | */ | |
658 | mutex_destroy(&inode->i_mutex); | |
659 | mutex_init(&inode->i_mutex); | |
14358e6d | 660 | lockdep_set_class(&inode->i_mutex, &type->i_mutex_dir_key); |
1e89a5e1 | 661 | } |
14358e6d | 662 | #endif |
1da177e4 LT |
663 | /* |
664 | * This is special! We do not need the spinlock | |
665 | * when clearing I_LOCK, because we're guaranteed | |
666 | * that nobody else tries to do anything about the | |
667 | * state of the inode when it is locked, as we | |
668 | * just created it (so there can be no old holders | |
669 | * that haven't tested I_LOCK). | |
670 | */ | |
7ef0d737 | 671 | WARN_ON((inode->i_state & (I_LOCK|I_NEW)) != (I_LOCK|I_NEW)); |
1da177e4 LT |
672 | inode->i_state &= ~(I_LOCK|I_NEW); |
673 | wake_up_inode(inode); | |
674 | } | |
675 | ||
676 | EXPORT_SYMBOL(unlock_new_inode); | |
677 | ||
678 | /* | |
679 | * This is called without the inode lock held.. Be careful. | |
680 | * | |
681 | * We no longer cache the sb_flags in i_flags - see fs.h | |
682 | * -- rmk@arm.uk.linux.org | |
683 | */ | |
684 | static struct inode * get_new_inode(struct super_block *sb, struct hlist_head *head, int (*test)(struct inode *, void *), int (*set)(struct inode *, void *), void *data) | |
685 | { | |
686 | struct inode * inode; | |
687 | ||
688 | inode = alloc_inode(sb); | |
689 | if (inode) { | |
690 | struct inode * old; | |
691 | ||
692 | spin_lock(&inode_lock); | |
693 | /* We released the lock, so.. */ | |
694 | old = find_inode(sb, head, test, data); | |
695 | if (!old) { | |
696 | if (set(inode, data)) | |
697 | goto set_failed; | |
698 | ||
8290c35f | 699 | __inode_add_to_lists(sb, head, inode); |
1da177e4 LT |
700 | inode->i_state = I_LOCK|I_NEW; |
701 | spin_unlock(&inode_lock); | |
702 | ||
703 | /* Return the locked inode with I_NEW set, the | |
704 | * caller is responsible for filling in the contents | |
705 | */ | |
706 | return inode; | |
707 | } | |
708 | ||
709 | /* | |
710 | * Uhhuh, somebody else created the same inode under | |
711 | * us. Use the old inode instead of the one we just | |
712 | * allocated. | |
713 | */ | |
714 | __iget(old); | |
715 | spin_unlock(&inode_lock); | |
716 | destroy_inode(inode); | |
717 | inode = old; | |
718 | wait_on_inode(inode); | |
719 | } | |
720 | return inode; | |
721 | ||
722 | set_failed: | |
723 | spin_unlock(&inode_lock); | |
724 | destroy_inode(inode); | |
725 | return NULL; | |
726 | } | |
727 | ||
728 | /* | |
729 | * get_new_inode_fast is the fast path version of get_new_inode, see the | |
730 | * comment at iget_locked for details. | |
731 | */ | |
732 | static struct inode * get_new_inode_fast(struct super_block *sb, struct hlist_head *head, unsigned long ino) | |
733 | { | |
734 | struct inode * inode; | |
735 | ||
736 | inode = alloc_inode(sb); | |
737 | if (inode) { | |
738 | struct inode * old; | |
739 | ||
740 | spin_lock(&inode_lock); | |
741 | /* We released the lock, so.. */ | |
742 | old = find_inode_fast(sb, head, ino); | |
743 | if (!old) { | |
744 | inode->i_ino = ino; | |
8290c35f | 745 | __inode_add_to_lists(sb, head, inode); |
1da177e4 LT |
746 | inode->i_state = I_LOCK|I_NEW; |
747 | spin_unlock(&inode_lock); | |
748 | ||
749 | /* Return the locked inode with I_NEW set, the | |
750 | * caller is responsible for filling in the contents | |
751 | */ | |
752 | return inode; | |
753 | } | |
754 | ||
755 | /* | |
756 | * Uhhuh, somebody else created the same inode under | |
757 | * us. Use the old inode instead of the one we just | |
758 | * allocated. | |
759 | */ | |
760 | __iget(old); | |
761 | spin_unlock(&inode_lock); | |
762 | destroy_inode(inode); | |
763 | inode = old; | |
764 | wait_on_inode(inode); | |
765 | } | |
766 | return inode; | |
767 | } | |
768 | ||
1da177e4 LT |
769 | /** |
770 | * iunique - get a unique inode number | |
771 | * @sb: superblock | |
772 | * @max_reserved: highest reserved inode number | |
773 | * | |
774 | * Obtain an inode number that is unique on the system for a given | |
775 | * superblock. This is used by file systems that have no natural | |
776 | * permanent inode numbering system. An inode number is returned that | |
777 | * is higher than the reserved limit but unique. | |
778 | * | |
779 | * BUGS: | |
780 | * With a large number of inodes live on the file system this function | |
781 | * currently becomes quite slow. | |
782 | */ | |
783 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
784 | { | |
866b04fc JL |
785 | /* |
786 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
787 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
788 | * here to attempt to avoid that. | |
789 | */ | |
790 | static unsigned int counter; | |
1da177e4 | 791 | struct inode *inode; |
3361c7be | 792 | struct hlist_head *head; |
1da177e4 | 793 | ino_t res; |
3361c7be | 794 | |
1da177e4 | 795 | spin_lock(&inode_lock); |
3361c7be JL |
796 | do { |
797 | if (counter <= max_reserved) | |
798 | counter = max_reserved + 1; | |
1da177e4 | 799 | res = counter++; |
3361c7be | 800 | head = inode_hashtable + hash(sb, res); |
1da177e4 | 801 | inode = find_inode_fast(sb, head, res); |
3361c7be JL |
802 | } while (inode != NULL); |
803 | spin_unlock(&inode_lock); | |
1da177e4 | 804 | |
3361c7be JL |
805 | return res; |
806 | } | |
1da177e4 LT |
807 | EXPORT_SYMBOL(iunique); |
808 | ||
809 | struct inode *igrab(struct inode *inode) | |
810 | { | |
811 | spin_lock(&inode_lock); | |
4a3b0a49 | 812 | if (!(inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE))) |
1da177e4 LT |
813 | __iget(inode); |
814 | else | |
815 | /* | |
816 | * Handle the case where s_op->clear_inode is not been | |
817 | * called yet, and somebody is calling igrab | |
818 | * while the inode is getting freed. | |
819 | */ | |
820 | inode = NULL; | |
821 | spin_unlock(&inode_lock); | |
822 | return inode; | |
823 | } | |
824 | ||
825 | EXPORT_SYMBOL(igrab); | |
826 | ||
827 | /** | |
828 | * ifind - internal function, you want ilookup5() or iget5(). | |
829 | * @sb: super block of file system to search | |
830 | * @head: the head of the list to search | |
831 | * @test: callback used for comparisons between inodes | |
832 | * @data: opaque data pointer to pass to @test | |
88bd5121 | 833 | * @wait: if true wait for the inode to be unlocked, if false do not |
1da177e4 LT |
834 | * |
835 | * ifind() searches for the inode specified by @data in the inode | |
836 | * cache. This is a generalized version of ifind_fast() for file systems where | |
837 | * the inode number is not sufficient for unique identification of an inode. | |
838 | * | |
839 | * If the inode is in the cache, the inode is returned with an incremented | |
840 | * reference count. | |
841 | * | |
842 | * Otherwise NULL is returned. | |
843 | * | |
844 | * Note, @test is called with the inode_lock held, so can't sleep. | |
845 | */ | |
5d2bea45 | 846 | static struct inode *ifind(struct super_block *sb, |
1da177e4 | 847 | struct hlist_head *head, int (*test)(struct inode *, void *), |
88bd5121 | 848 | void *data, const int wait) |
1da177e4 LT |
849 | { |
850 | struct inode *inode; | |
851 | ||
852 | spin_lock(&inode_lock); | |
853 | inode = find_inode(sb, head, test, data); | |
854 | if (inode) { | |
855 | __iget(inode); | |
856 | spin_unlock(&inode_lock); | |
88bd5121 AA |
857 | if (likely(wait)) |
858 | wait_on_inode(inode); | |
1da177e4 LT |
859 | return inode; |
860 | } | |
861 | spin_unlock(&inode_lock); | |
862 | return NULL; | |
863 | } | |
864 | ||
865 | /** | |
866 | * ifind_fast - internal function, you want ilookup() or iget(). | |
867 | * @sb: super block of file system to search | |
868 | * @head: head of the list to search | |
869 | * @ino: inode number to search for | |
870 | * | |
871 | * ifind_fast() searches for the inode @ino in the inode cache. This is for | |
872 | * file systems where the inode number is sufficient for unique identification | |
873 | * of an inode. | |
874 | * | |
875 | * If the inode is in the cache, the inode is returned with an incremented | |
876 | * reference count. | |
877 | * | |
878 | * Otherwise NULL is returned. | |
879 | */ | |
5d2bea45 | 880 | static struct inode *ifind_fast(struct super_block *sb, |
1da177e4 LT |
881 | struct hlist_head *head, unsigned long ino) |
882 | { | |
883 | struct inode *inode; | |
884 | ||
885 | spin_lock(&inode_lock); | |
886 | inode = find_inode_fast(sb, head, ino); | |
887 | if (inode) { | |
888 | __iget(inode); | |
889 | spin_unlock(&inode_lock); | |
890 | wait_on_inode(inode); | |
891 | return inode; | |
892 | } | |
893 | spin_unlock(&inode_lock); | |
894 | return NULL; | |
895 | } | |
896 | ||
897 | /** | |
88bd5121 | 898 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 LT |
899 | * @sb: super block of file system to search |
900 | * @hashval: hash value (usually inode number) to search for | |
901 | * @test: callback used for comparisons between inodes | |
902 | * @data: opaque data pointer to pass to @test | |
903 | * | |
904 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
905 | * @data in the inode cache. This is a generalized version of ilookup() for | |
906 | * file systems where the inode number is not sufficient for unique | |
907 | * identification of an inode. | |
908 | * | |
909 | * If the inode is in the cache, the inode is returned with an incremented | |
88bd5121 AA |
910 | * reference count. Note, the inode lock is not waited upon so you have to be |
911 | * very careful what you do with the returned inode. You probably should be | |
912 | * using ilookup5() instead. | |
913 | * | |
914 | * Otherwise NULL is returned. | |
915 | * | |
916 | * Note, @test is called with the inode_lock held, so can't sleep. | |
917 | */ | |
918 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, | |
919 | int (*test)(struct inode *, void *), void *data) | |
920 | { | |
921 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
922 | ||
923 | return ifind(sb, head, test, data, 0); | |
924 | } | |
925 | ||
926 | EXPORT_SYMBOL(ilookup5_nowait); | |
927 | ||
928 | /** | |
929 | * ilookup5 - search for an inode in the inode cache | |
930 | * @sb: super block of file system to search | |
931 | * @hashval: hash value (usually inode number) to search for | |
932 | * @test: callback used for comparisons between inodes | |
933 | * @data: opaque data pointer to pass to @test | |
934 | * | |
935 | * ilookup5() uses ifind() to search for the inode specified by @hashval and | |
936 | * @data in the inode cache. This is a generalized version of ilookup() for | |
937 | * file systems where the inode number is not sufficient for unique | |
938 | * identification of an inode. | |
939 | * | |
940 | * If the inode is in the cache, the inode lock is waited upon and the inode is | |
941 | * returned with an incremented reference count. | |
1da177e4 LT |
942 | * |
943 | * Otherwise NULL is returned. | |
944 | * | |
945 | * Note, @test is called with the inode_lock held, so can't sleep. | |
946 | */ | |
947 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
948 | int (*test)(struct inode *, void *), void *data) | |
949 | { | |
950 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
951 | ||
88bd5121 | 952 | return ifind(sb, head, test, data, 1); |
1da177e4 LT |
953 | } |
954 | ||
955 | EXPORT_SYMBOL(ilookup5); | |
956 | ||
957 | /** | |
958 | * ilookup - search for an inode in the inode cache | |
959 | * @sb: super block of file system to search | |
960 | * @ino: inode number to search for | |
961 | * | |
962 | * ilookup() uses ifind_fast() to search for the inode @ino in the inode cache. | |
963 | * This is for file systems where the inode number is sufficient for unique | |
964 | * identification of an inode. | |
965 | * | |
966 | * If the inode is in the cache, the inode is returned with an incremented | |
967 | * reference count. | |
968 | * | |
969 | * Otherwise NULL is returned. | |
970 | */ | |
971 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
972 | { | |
973 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
974 | ||
975 | return ifind_fast(sb, head, ino); | |
976 | } | |
977 | ||
978 | EXPORT_SYMBOL(ilookup); | |
979 | ||
980 | /** | |
981 | * iget5_locked - obtain an inode from a mounted file system | |
982 | * @sb: super block of file system | |
983 | * @hashval: hash value (usually inode number) to get | |
984 | * @test: callback used for comparisons between inodes | |
985 | * @set: callback used to initialize a new struct inode | |
986 | * @data: opaque data pointer to pass to @test and @set | |
987 | * | |
1da177e4 LT |
988 | * iget5_locked() uses ifind() to search for the inode specified by @hashval |
989 | * and @data in the inode cache and if present it is returned with an increased | |
990 | * reference count. This is a generalized version of iget_locked() for file | |
991 | * systems where the inode number is not sufficient for unique identification | |
992 | * of an inode. | |
993 | * | |
994 | * If the inode is not in cache, get_new_inode() is called to allocate a new | |
995 | * inode and this is returned locked, hashed, and with the I_NEW flag set. The | |
996 | * file system gets to fill it in before unlocking it via unlock_new_inode(). | |
997 | * | |
998 | * Note both @test and @set are called with the inode_lock held, so can't sleep. | |
999 | */ | |
1000 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, | |
1001 | int (*test)(struct inode *, void *), | |
1002 | int (*set)(struct inode *, void *), void *data) | |
1003 | { | |
1004 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1005 | struct inode *inode; | |
1006 | ||
88bd5121 | 1007 | inode = ifind(sb, head, test, data, 1); |
1da177e4 LT |
1008 | if (inode) |
1009 | return inode; | |
1010 | /* | |
1011 | * get_new_inode() will do the right thing, re-trying the search | |
1012 | * in case it had to block at any point. | |
1013 | */ | |
1014 | return get_new_inode(sb, head, test, set, data); | |
1015 | } | |
1016 | ||
1017 | EXPORT_SYMBOL(iget5_locked); | |
1018 | ||
1019 | /** | |
1020 | * iget_locked - obtain an inode from a mounted file system | |
1021 | * @sb: super block of file system | |
1022 | * @ino: inode number to get | |
1023 | * | |
1da177e4 LT |
1024 | * iget_locked() uses ifind_fast() to search for the inode specified by @ino in |
1025 | * the inode cache and if present it is returned with an increased reference | |
1026 | * count. This is for file systems where the inode number is sufficient for | |
1027 | * unique identification of an inode. | |
1028 | * | |
1029 | * If the inode is not in cache, get_new_inode_fast() is called to allocate a | |
1030 | * new inode and this is returned locked, hashed, and with the I_NEW flag set. | |
1031 | * The file system gets to fill it in before unlocking it via | |
1032 | * unlock_new_inode(). | |
1033 | */ | |
1034 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) | |
1035 | { | |
1036 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1037 | struct inode *inode; | |
1038 | ||
1039 | inode = ifind_fast(sb, head, ino); | |
1040 | if (inode) | |
1041 | return inode; | |
1042 | /* | |
1043 | * get_new_inode_fast() will do the right thing, re-trying the search | |
1044 | * in case it had to block at any point. | |
1045 | */ | |
1046 | return get_new_inode_fast(sb, head, ino); | |
1047 | } | |
1048 | ||
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 | } | |
1077 | ||
1078 | EXPORT_SYMBOL(insert_inode_locked); | |
1079 | ||
1080 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1081 | int (*test)(struct inode *, void *), void *data) | |
1082 | { | |
1083 | struct super_block *sb = inode->i_sb; | |
1084 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1085 | struct inode *old; | |
1086 | ||
1087 | inode->i_state |= I_LOCK|I_NEW; | |
1088 | ||
1089 | while (1) { | |
1090 | spin_lock(&inode_lock); | |
1091 | old = find_inode(sb, head, test, data); | |
1092 | if (likely(!old)) { | |
1093 | hlist_add_head(&inode->i_hash, head); | |
1094 | spin_unlock(&inode_lock); | |
1095 | return 0; | |
1096 | } | |
1097 | __iget(old); | |
1098 | spin_unlock(&inode_lock); | |
1099 | wait_on_inode(old); | |
1100 | if (unlikely(!hlist_unhashed(&old->i_hash))) { | |
1101 | iput(old); | |
1102 | return -EBUSY; | |
1103 | } | |
1104 | iput(old); | |
1105 | } | |
1106 | } | |
1107 | ||
1108 | EXPORT_SYMBOL(insert_inode_locked4); | |
1109 | ||
1da177e4 LT |
1110 | /** |
1111 | * __insert_inode_hash - hash an inode | |
1112 | * @inode: unhashed inode | |
1113 | * @hashval: unsigned long value used to locate this object in the | |
1114 | * inode_hashtable. | |
1115 | * | |
1116 | * Add an inode to the inode hash for this superblock. | |
1117 | */ | |
1118 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
1119 | { | |
1120 | struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval); | |
1121 | spin_lock(&inode_lock); | |
1122 | hlist_add_head(&inode->i_hash, head); | |
1123 | spin_unlock(&inode_lock); | |
1124 | } | |
1125 | ||
1126 | EXPORT_SYMBOL(__insert_inode_hash); | |
1127 | ||
1128 | /** | |
1129 | * remove_inode_hash - remove an inode from the hash | |
1130 | * @inode: inode to unhash | |
1131 | * | |
1132 | * Remove an inode from the superblock. | |
1133 | */ | |
1134 | void remove_inode_hash(struct inode *inode) | |
1135 | { | |
1136 | spin_lock(&inode_lock); | |
1137 | hlist_del_init(&inode->i_hash); | |
1138 | spin_unlock(&inode_lock); | |
1139 | } | |
1140 | ||
1141 | EXPORT_SYMBOL(remove_inode_hash); | |
1142 | ||
1143 | /* | |
1144 | * Tell the filesystem that this inode is no longer of any interest and should | |
1145 | * be completely destroyed. | |
1146 | * | |
1147 | * We leave the inode in the inode hash table until *after* the filesystem's | |
1148 | * ->delete_inode completes. This ensures that an iget (such as nfsd might | |
1149 | * instigate) will always find up-to-date information either in the hash or on | |
1150 | * disk. | |
1151 | * | |
1152 | * I_FREEING is set so that no-one will take a new reference to the inode while | |
1153 | * it is being deleted. | |
1154 | */ | |
b32714ba | 1155 | void generic_delete_inode(struct inode *inode) |
1da177e4 | 1156 | { |
ee9b6d61 | 1157 | const struct super_operations *op = inode->i_sb->s_op; |
1da177e4 | 1158 | |
b32714ba AV |
1159 | list_del_init(&inode->i_list); |
1160 | list_del_init(&inode->i_sb_list); | |
7ef0d737 | 1161 | WARN_ON(inode->i_state & I_NEW); |
b32714ba AV |
1162 | inode->i_state |= I_FREEING; |
1163 | inodes_stat.nr_inodes--; | |
1164 | spin_unlock(&inode_lock); | |
1165 | ||
1da177e4 LT |
1166 | security_inode_delete(inode); |
1167 | ||
1168 | if (op->delete_inode) { | |
1169 | void (*delete)(struct inode *) = op->delete_inode; | |
1170 | if (!is_bad_inode(inode)) | |
1171 | DQUOT_INIT(inode); | |
e85b5652 MF |
1172 | /* Filesystems implementing their own |
1173 | * s_op->delete_inode are required to call | |
1174 | * truncate_inode_pages and clear_inode() | |
1175 | * internally */ | |
1da177e4 | 1176 | delete(inode); |
e85b5652 MF |
1177 | } else { |
1178 | truncate_inode_pages(&inode->i_data, 0); | |
1da177e4 | 1179 | clear_inode(inode); |
e85b5652 | 1180 | } |
1da177e4 LT |
1181 | spin_lock(&inode_lock); |
1182 | hlist_del_init(&inode->i_hash); | |
1183 | spin_unlock(&inode_lock); | |
1184 | wake_up_inode(inode); | |
b7542f8c | 1185 | BUG_ON(inode->i_state != I_CLEAR); |
1da177e4 LT |
1186 | destroy_inode(inode); |
1187 | } | |
1188 | ||
1189 | EXPORT_SYMBOL(generic_delete_inode); | |
1190 | ||
1191 | static void generic_forget_inode(struct inode *inode) | |
1192 | { | |
1193 | struct super_block *sb = inode->i_sb; | |
1194 | ||
1195 | if (!hlist_unhashed(&inode->i_hash)) { | |
1c0eeaf5 | 1196 | if (!(inode->i_state & (I_DIRTY|I_SYNC))) |
1da177e4 LT |
1197 | list_move(&inode->i_list, &inode_unused); |
1198 | inodes_stat.nr_unused++; | |
acb0c854 | 1199 | if (sb->s_flags & MS_ACTIVE) { |
991114c6 | 1200 | spin_unlock(&inode_lock); |
1da177e4 | 1201 | return; |
991114c6 | 1202 | } |
7ef0d737 | 1203 | WARN_ON(inode->i_state & I_NEW); |
991114c6 AV |
1204 | inode->i_state |= I_WILL_FREE; |
1205 | spin_unlock(&inode_lock); | |
1da177e4 LT |
1206 | write_inode_now(inode, 1); |
1207 | spin_lock(&inode_lock); | |
7ef0d737 | 1208 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1209 | inode->i_state &= ~I_WILL_FREE; |
1da177e4 LT |
1210 | inodes_stat.nr_unused--; |
1211 | hlist_del_init(&inode->i_hash); | |
1212 | } | |
1213 | list_del_init(&inode->i_list); | |
1214 | list_del_init(&inode->i_sb_list); | |
7ef0d737 | 1215 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1216 | inode->i_state |= I_FREEING; |
1da177e4 LT |
1217 | inodes_stat.nr_inodes--; |
1218 | spin_unlock(&inode_lock); | |
1219 | if (inode->i_data.nrpages) | |
1220 | truncate_inode_pages(&inode->i_data, 0); | |
1221 | clear_inode(inode); | |
7f04c26d | 1222 | wake_up_inode(inode); |
1da177e4 LT |
1223 | destroy_inode(inode); |
1224 | } | |
1225 | ||
1226 | /* | |
1227 | * Normal UNIX filesystem behaviour: delete the | |
1228 | * inode when the usage count drops to zero, and | |
1229 | * i_nlink is zero. | |
1230 | */ | |
cb2c0233 | 1231 | void generic_drop_inode(struct inode *inode) |
1da177e4 LT |
1232 | { |
1233 | if (!inode->i_nlink) | |
1234 | generic_delete_inode(inode); | |
1235 | else | |
1236 | generic_forget_inode(inode); | |
1237 | } | |
1238 | ||
cb2c0233 MF |
1239 | EXPORT_SYMBOL_GPL(generic_drop_inode); |
1240 | ||
1da177e4 LT |
1241 | /* |
1242 | * Called when we're dropping the last reference | |
1243 | * to an inode. | |
1244 | * | |
1245 | * Call the FS "drop()" function, defaulting to | |
1246 | * the legacy UNIX filesystem behaviour.. | |
1247 | * | |
1248 | * NOTE! NOTE! NOTE! We're called with the inode lock | |
1249 | * held, and the drop function is supposed to release | |
1250 | * the lock! | |
1251 | */ | |
1252 | static inline void iput_final(struct inode *inode) | |
1253 | { | |
ee9b6d61 | 1254 | const struct super_operations *op = inode->i_sb->s_op; |
1da177e4 LT |
1255 | void (*drop)(struct inode *) = generic_drop_inode; |
1256 | ||
1257 | if (op && op->drop_inode) | |
1258 | drop = op->drop_inode; | |
1259 | drop(inode); | |
1260 | } | |
1261 | ||
1262 | /** | |
1263 | * iput - put an inode | |
1264 | * @inode: inode to put | |
1265 | * | |
1266 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1267 | * zero, the inode is then freed and may also be destroyed. | |
1268 | * | |
1269 | * Consequently, iput() can sleep. | |
1270 | */ | |
1271 | void iput(struct inode *inode) | |
1272 | { | |
1273 | if (inode) { | |
1da177e4 LT |
1274 | BUG_ON(inode->i_state == I_CLEAR); |
1275 | ||
1da177e4 LT |
1276 | if (atomic_dec_and_lock(&inode->i_count, &inode_lock)) |
1277 | iput_final(inode); | |
1278 | } | |
1279 | } | |
1280 | ||
1281 | EXPORT_SYMBOL(iput); | |
1282 | ||
1283 | /** | |
1284 | * bmap - find a block number in a file | |
1285 | * @inode: inode of file | |
1286 | * @block: block to find | |
1287 | * | |
1288 | * Returns the block number on the device holding the inode that | |
1289 | * is the disk block number for the block of the file requested. | |
1290 | * That is, asked for block 4 of inode 1 the function will return the | |
1291 | * disk block relative to the disk start that holds that block of the | |
1292 | * file. | |
1293 | */ | |
1294 | sector_t bmap(struct inode * inode, sector_t block) | |
1295 | { | |
1296 | sector_t res = 0; | |
1297 | if (inode->i_mapping->a_ops->bmap) | |
1298 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1299 | return res; | |
1300 | } | |
1da177e4 LT |
1301 | EXPORT_SYMBOL(bmap); |
1302 | ||
11ff6f05 MG |
1303 | /* |
1304 | * With relative atime, only update atime if the previous atime is | |
1305 | * earlier than either the ctime or mtime or if at least a day has | |
1306 | * passed since the last atime update. | |
1307 | */ | |
1308 | static int relatime_need_update(struct vfsmount *mnt, struct inode *inode, | |
1309 | struct timespec now) | |
1310 | { | |
1311 | ||
1312 | if (!(mnt->mnt_flags & MNT_RELATIME)) | |
1313 | return 1; | |
1314 | /* | |
1315 | * Is mtime younger than atime? If yes, update atime: | |
1316 | */ | |
1317 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1318 | return 1; | |
1319 | /* | |
1320 | * Is ctime younger than atime? If yes, update atime: | |
1321 | */ | |
1322 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1323 | return 1; | |
1324 | ||
1325 | /* | |
1326 | * Is the previous atime value older than a day? If yes, | |
1327 | * update atime: | |
1328 | */ | |
1329 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1330 | return 1; | |
1331 | /* | |
1332 | * Good, we can skip the atime update: | |
1333 | */ | |
1334 | return 0; | |
1335 | } | |
1336 | ||
1da177e4 | 1337 | /** |
869243a0 CH |
1338 | * touch_atime - update the access time |
1339 | * @mnt: mount the inode is accessed on | |
7045f37b | 1340 | * @dentry: dentry accessed |
1da177e4 LT |
1341 | * |
1342 | * Update the accessed time on an inode and mark it for writeback. | |
1343 | * This function automatically handles read only file systems and media, | |
1344 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1345 | */ | |
869243a0 | 1346 | void touch_atime(struct vfsmount *mnt, struct dentry *dentry) |
1da177e4 | 1347 | { |
869243a0 | 1348 | struct inode *inode = dentry->d_inode; |
1da177e4 LT |
1349 | struct timespec now; |
1350 | ||
cdb70f3f | 1351 | if (mnt_want_write(mnt)) |
b2276138 | 1352 | return; |
cdb70f3f DH |
1353 | if (inode->i_flags & S_NOATIME) |
1354 | goto out; | |
37756ced | 1355 | if (IS_NOATIME(inode)) |
cdb70f3f | 1356 | goto out; |
b2276138 | 1357 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
cdb70f3f | 1358 | goto out; |
47ae32d6 | 1359 | |
cdb70f3f DH |
1360 | if (mnt->mnt_flags & MNT_NOATIME) |
1361 | goto out; | |
1362 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) | |
1363 | goto out; | |
1da177e4 LT |
1364 | |
1365 | now = current_fs_time(inode->i_sb); | |
11ff6f05 MG |
1366 | |
1367 | if (!relatime_need_update(mnt, inode, now)) | |
1368 | goto out; | |
1369 | ||
47ae32d6 | 1370 | if (timespec_equal(&inode->i_atime, &now)) |
cdb70f3f | 1371 | goto out; |
47ae32d6 VH |
1372 | |
1373 | inode->i_atime = now; | |
1374 | mark_inode_dirty_sync(inode); | |
cdb70f3f DH |
1375 | out: |
1376 | mnt_drop_write(mnt); | |
1da177e4 | 1377 | } |
869243a0 | 1378 | EXPORT_SYMBOL(touch_atime); |
1da177e4 LT |
1379 | |
1380 | /** | |
870f4817 CH |
1381 | * file_update_time - update mtime and ctime time |
1382 | * @file: file accessed | |
1da177e4 | 1383 | * |
870f4817 CH |
1384 | * Update the mtime and ctime members of an inode and mark the inode |
1385 | * for writeback. Note that this function is meant exclusively for | |
1386 | * usage in the file write path of filesystems, and filesystems may | |
1387 | * choose to explicitly ignore update via this function with the | |
1388 | * S_NOCTIME inode flag, e.g. for network filesystem where these | |
1389 | * timestamps are handled by the server. | |
1da177e4 LT |
1390 | */ |
1391 | ||
870f4817 | 1392 | void file_update_time(struct file *file) |
1da177e4 | 1393 | { |
0f7fc9e4 | 1394 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
1395 | struct timespec now; |
1396 | int sync_it = 0; | |
20ddee2c | 1397 | int err; |
1da177e4 LT |
1398 | |
1399 | if (IS_NOCMTIME(inode)) | |
1400 | return; | |
20ddee2c DH |
1401 | |
1402 | err = mnt_want_write(file->f_path.mnt); | |
1403 | if (err) | |
1da177e4 LT |
1404 | return; |
1405 | ||
1406 | now = current_fs_time(inode->i_sb); | |
ed97bd37 AM |
1407 | if (!timespec_equal(&inode->i_mtime, &now)) { |
1408 | inode->i_mtime = now; | |
1da177e4 | 1409 | sync_it = 1; |
ed97bd37 | 1410 | } |
1da177e4 | 1411 | |
ed97bd37 AM |
1412 | if (!timespec_equal(&inode->i_ctime, &now)) { |
1413 | inode->i_ctime = now; | |
870f4817 | 1414 | sync_it = 1; |
ed97bd37 | 1415 | } |
870f4817 | 1416 | |
7a224228 JNC |
1417 | if (IS_I_VERSION(inode)) { |
1418 | inode_inc_iversion(inode); | |
1419 | sync_it = 1; | |
1420 | } | |
1421 | ||
1da177e4 LT |
1422 | if (sync_it) |
1423 | mark_inode_dirty_sync(inode); | |
20ddee2c | 1424 | mnt_drop_write(file->f_path.mnt); |
1da177e4 LT |
1425 | } |
1426 | ||
870f4817 | 1427 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1428 | |
1429 | int inode_needs_sync(struct inode *inode) | |
1430 | { | |
1431 | if (IS_SYNC(inode)) | |
1432 | return 1; | |
1433 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1434 | return 1; | |
1435 | return 0; | |
1436 | } | |
1437 | ||
1438 | EXPORT_SYMBOL(inode_needs_sync); | |
1439 | ||
1da177e4 LT |
1440 | int inode_wait(void *word) |
1441 | { | |
1442 | schedule(); | |
1443 | return 0; | |
1444 | } | |
d44dab8d | 1445 | EXPORT_SYMBOL(inode_wait); |
1da177e4 LT |
1446 | |
1447 | /* | |
168a9fd6 MS |
1448 | * If we try to find an inode in the inode hash while it is being |
1449 | * deleted, we have to wait until the filesystem completes its | |
1450 | * deletion before reporting that it isn't found. This function waits | |
1451 | * until the deletion _might_ have completed. Callers are responsible | |
1452 | * to recheck inode state. | |
1453 | * | |
1454 | * It doesn't matter if I_LOCK is not set initially, a call to | |
1455 | * wake_up_inode() after removing from the hash list will DTRT. | |
1456 | * | |
1da177e4 LT |
1457 | * This is called with inode_lock held. |
1458 | */ | |
1459 | static void __wait_on_freeing_inode(struct inode *inode) | |
1460 | { | |
1461 | wait_queue_head_t *wq; | |
1462 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_LOCK); | |
1da177e4 LT |
1463 | wq = bit_waitqueue(&inode->i_state, __I_LOCK); |
1464 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
1465 | spin_unlock(&inode_lock); | |
1466 | schedule(); | |
1467 | finish_wait(wq, &wait.wait); | |
1468 | spin_lock(&inode_lock); | |
1469 | } | |
1470 | ||
62752ee1 MF |
1471 | /* |
1472 | * We rarely want to lock two inodes that do not have a parent/child | |
1473 | * relationship (such as directory, child inode) simultaneously. The | |
1474 | * vast majority of file systems should be able to get along fine | |
1475 | * without this. Do not use these functions except as a last resort. | |
1476 | */ | |
1477 | void inode_double_lock(struct inode *inode1, struct inode *inode2) | |
1478 | { | |
1479 | if (inode1 == NULL || inode2 == NULL || inode1 == inode2) { | |
1480 | if (inode1) | |
1481 | mutex_lock(&inode1->i_mutex); | |
1482 | else if (inode2) | |
1483 | mutex_lock(&inode2->i_mutex); | |
1484 | return; | |
1485 | } | |
1486 | ||
1487 | if (inode1 < inode2) { | |
1488 | mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT); | |
1489 | mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD); | |
1490 | } else { | |
1491 | mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT); | |
1492 | mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD); | |
1493 | } | |
1494 | } | |
1495 | EXPORT_SYMBOL(inode_double_lock); | |
1496 | ||
1497 | void inode_double_unlock(struct inode *inode1, struct inode *inode2) | |
1498 | { | |
1499 | if (inode1) | |
1500 | mutex_unlock(&inode1->i_mutex); | |
1501 | ||
1502 | if (inode2 && inode2 != inode1) | |
1503 | mutex_unlock(&inode2->i_mutex); | |
1504 | } | |
1505 | EXPORT_SYMBOL(inode_double_unlock); | |
1506 | ||
1da177e4 LT |
1507 | static __initdata unsigned long ihash_entries; |
1508 | static int __init set_ihash_entries(char *str) | |
1509 | { | |
1510 | if (!str) | |
1511 | return 0; | |
1512 | ihash_entries = simple_strtoul(str, &str, 0); | |
1513 | return 1; | |
1514 | } | |
1515 | __setup("ihash_entries=", set_ihash_entries); | |
1516 | ||
1517 | /* | |
1518 | * Initialize the waitqueues and inode hash table. | |
1519 | */ | |
1520 | void __init inode_init_early(void) | |
1521 | { | |
1522 | int loop; | |
1523 | ||
1524 | /* If hashes are distributed across NUMA nodes, defer | |
1525 | * hash allocation until vmalloc space is available. | |
1526 | */ | |
1527 | if (hashdist) | |
1528 | return; | |
1529 | ||
1530 | inode_hashtable = | |
1531 | alloc_large_system_hash("Inode-cache", | |
1532 | sizeof(struct hlist_head), | |
1533 | ihash_entries, | |
1534 | 14, | |
1535 | HASH_EARLY, | |
1536 | &i_hash_shift, | |
1537 | &i_hash_mask, | |
1538 | 0); | |
1539 | ||
1540 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1541 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1542 | } | |
1543 | ||
74bf17cf | 1544 | void __init inode_init(void) |
1da177e4 LT |
1545 | { |
1546 | int loop; | |
1547 | ||
1548 | /* inode slab cache */ | |
b0196009 PJ |
1549 | inode_cachep = kmem_cache_create("inode_cache", |
1550 | sizeof(struct inode), | |
1551 | 0, | |
1552 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
1553 | SLAB_MEM_SPREAD), | |
20c2df83 | 1554 | init_once); |
8e1f936b | 1555 | register_shrinker(&icache_shrinker); |
1da177e4 LT |
1556 | |
1557 | /* Hash may have been set up in inode_init_early */ | |
1558 | if (!hashdist) | |
1559 | return; | |
1560 | ||
1561 | inode_hashtable = | |
1562 | alloc_large_system_hash("Inode-cache", | |
1563 | sizeof(struct hlist_head), | |
1564 | ihash_entries, | |
1565 | 14, | |
1566 | 0, | |
1567 | &i_hash_shift, | |
1568 | &i_hash_mask, | |
1569 | 0); | |
1570 | ||
1571 | for (loop = 0; loop < (1 << i_hash_shift); loop++) | |
1572 | INIT_HLIST_HEAD(&inode_hashtable[loop]); | |
1573 | } | |
1574 | ||
1575 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1576 | { | |
1577 | inode->i_mode = mode; | |
1578 | if (S_ISCHR(mode)) { | |
1579 | inode->i_fop = &def_chr_fops; | |
1580 | inode->i_rdev = rdev; | |
1581 | } else if (S_ISBLK(mode)) { | |
1582 | inode->i_fop = &def_blk_fops; | |
1583 | inode->i_rdev = rdev; | |
1584 | } else if (S_ISFIFO(mode)) | |
1585 | inode->i_fop = &def_fifo_fops; | |
1586 | else if (S_ISSOCK(mode)) | |
1587 | inode->i_fop = &bad_sock_fops; | |
1588 | else | |
1589 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o)\n", | |
1590 | mode); | |
1591 | } | |
1592 | EXPORT_SYMBOL(init_special_inode); |