]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 | 2 | * (C) 1997 Linus Torvalds |
4b4563dc | 3 | * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation) |
1da177e4 | 4 | */ |
e59cc473 | 5 | #include <linux/export.h> |
1da177e4 LT |
6 | #include <linux/fs.h> |
7 | #include <linux/mm.h> | |
1da177e4 | 8 | #include <linux/backing-dev.h> |
1da177e4 LT |
9 | #include <linux/hash.h> |
10 | #include <linux/swap.h> | |
11 | #include <linux/security.h> | |
1da177e4 LT |
12 | #include <linux/cdev.h> |
13 | #include <linux/bootmem.h> | |
3be25f49 | 14 | #include <linux/fsnotify.h> |
fc33a7bb | 15 | #include <linux/mount.h> |
f19d4a8f | 16 | #include <linux/posix_acl.h> |
9ce6e0be | 17 | #include <linux/prefetch.h> |
4b4563dc | 18 | #include <linux/buffer_head.h> /* for inode_has_buffers */ |
7ada4db8 | 19 | #include <linux/ratelimit.h> |
bc3b14cb | 20 | #include <linux/list_lru.h> |
0ae45f63 | 21 | #include <trace/events/writeback.h> |
a66979ab | 22 | #include "internal.h" |
1da177e4 | 23 | |
250df6ed | 24 | /* |
4b4563dc | 25 | * Inode locking rules: |
250df6ed DC |
26 | * |
27 | * inode->i_lock protects: | |
28 | * inode->i_state, inode->i_hash, __iget() | |
bc3b14cb | 29 | * Inode LRU list locks protect: |
98b745c6 | 30 | * inode->i_sb->s_inode_lru, inode->i_lru |
74278da9 DC |
31 | * inode->i_sb->s_inode_list_lock protects: |
32 | * inode->i_sb->s_inodes, inode->i_sb_list | |
f758eeab | 33 | * bdi->wb.list_lock protects: |
c7f54084 | 34 | * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list |
67a23c49 DC |
35 | * inode_hash_lock protects: |
36 | * inode_hashtable, inode->i_hash | |
250df6ed DC |
37 | * |
38 | * Lock ordering: | |
55fa6091 | 39 | * |
74278da9 | 40 | * inode->i_sb->s_inode_list_lock |
55fa6091 | 41 | * inode->i_lock |
bc3b14cb | 42 | * Inode LRU list locks |
a66979ab | 43 | * |
f758eeab | 44 | * bdi->wb.list_lock |
a66979ab | 45 | * inode->i_lock |
67a23c49 DC |
46 | * |
47 | * inode_hash_lock | |
74278da9 | 48 | * inode->i_sb->s_inode_list_lock |
67a23c49 DC |
49 | * inode->i_lock |
50 | * | |
51 | * iunique_lock | |
52 | * inode_hash_lock | |
250df6ed DC |
53 | */ |
54 | ||
fa3536cc ED |
55 | static unsigned int i_hash_mask __read_mostly; |
56 | static unsigned int i_hash_shift __read_mostly; | |
67a23c49 DC |
57 | static struct hlist_head *inode_hashtable __read_mostly; |
58 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock); | |
1da177e4 | 59 | |
7dcda1c9 JA |
60 | /* |
61 | * Empty aops. Can be used for the cases where the user does not | |
62 | * define any of the address_space operations. | |
63 | */ | |
64 | const struct address_space_operations empty_aops = { | |
65 | }; | |
66 | EXPORT_SYMBOL(empty_aops); | |
67 | ||
1da177e4 LT |
68 | /* |
69 | * Statistics gathering.. | |
70 | */ | |
71 | struct inodes_stat_t inodes_stat; | |
72 | ||
3942c07c GC |
73 | static DEFINE_PER_CPU(unsigned long, nr_inodes); |
74 | static DEFINE_PER_CPU(unsigned long, nr_unused); | |
cffbc8aa | 75 | |
6b3304b5 | 76 | static struct kmem_cache *inode_cachep __read_mostly; |
1da177e4 | 77 | |
3942c07c | 78 | static long get_nr_inodes(void) |
cffbc8aa | 79 | { |
3e880fb5 | 80 | int i; |
3942c07c | 81 | long sum = 0; |
3e880fb5 NP |
82 | for_each_possible_cpu(i) |
83 | sum += per_cpu(nr_inodes, i); | |
84 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
85 | } |
86 | ||
3942c07c | 87 | static inline long get_nr_inodes_unused(void) |
cffbc8aa | 88 | { |
fcb94f72 | 89 | int i; |
3942c07c | 90 | long sum = 0; |
fcb94f72 DC |
91 | for_each_possible_cpu(i) |
92 | sum += per_cpu(nr_unused, i); | |
93 | return sum < 0 ? 0 : sum; | |
cffbc8aa DC |
94 | } |
95 | ||
3942c07c | 96 | long get_nr_dirty_inodes(void) |
cffbc8aa | 97 | { |
3e880fb5 | 98 | /* not actually dirty inodes, but a wild approximation */ |
3942c07c | 99 | long nr_dirty = get_nr_inodes() - get_nr_inodes_unused(); |
cffbc8aa | 100 | return nr_dirty > 0 ? nr_dirty : 0; |
cffbc8aa DC |
101 | } |
102 | ||
103 | /* | |
104 | * Handle nr_inode sysctl | |
105 | */ | |
106 | #ifdef CONFIG_SYSCTL | |
1f7e0616 | 107 | int proc_nr_inodes(struct ctl_table *table, int write, |
cffbc8aa DC |
108 | void __user *buffer, size_t *lenp, loff_t *ppos) |
109 | { | |
110 | inodes_stat.nr_inodes = get_nr_inodes(); | |
fcb94f72 | 111 | inodes_stat.nr_unused = get_nr_inodes_unused(); |
3942c07c | 112 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
cffbc8aa DC |
113 | } |
114 | #endif | |
115 | ||
bd9b51e7 AV |
116 | static int no_open(struct inode *inode, struct file *file) |
117 | { | |
118 | return -ENXIO; | |
119 | } | |
120 | ||
2cb1599f | 121 | /** |
6e7c2b4d | 122 | * inode_init_always - perform inode structure initialisation |
0bc02f3f RD |
123 | * @sb: superblock inode belongs to |
124 | * @inode: inode to initialise | |
2cb1599f DC |
125 | * |
126 | * These are initializations that need to be done on every inode | |
127 | * allocation as the fields are not initialised by slab allocation. | |
128 | */ | |
54e34621 | 129 | int inode_init_always(struct super_block *sb, struct inode *inode) |
1da177e4 | 130 | { |
6e1d5dcc | 131 | static const struct inode_operations empty_iops; |
bd9b51e7 | 132 | static const struct file_operations no_open_fops = {.open = no_open}; |
6b3304b5 | 133 | struct address_space *const mapping = &inode->i_data; |
2cb1599f DC |
134 | |
135 | inode->i_sb = sb; | |
136 | inode->i_blkbits = sb->s_blocksize_bits; | |
137 | inode->i_flags = 0; | |
138 | atomic_set(&inode->i_count, 1); | |
139 | inode->i_op = &empty_iops; | |
bd9b51e7 | 140 | inode->i_fop = &no_open_fops; |
a78ef704 | 141 | inode->__i_nlink = 1; |
3ddcd056 | 142 | inode->i_opflags = 0; |
d0a5b995 AG |
143 | if (sb->s_xattr) |
144 | inode->i_opflags |= IOP_XATTR; | |
92361636 EB |
145 | i_uid_write(inode, 0); |
146 | i_gid_write(inode, 0); | |
2cb1599f DC |
147 | atomic_set(&inode->i_writecount, 0); |
148 | inode->i_size = 0; | |
149 | inode->i_blocks = 0; | |
150 | inode->i_bytes = 0; | |
151 | inode->i_generation = 0; | |
2cb1599f DC |
152 | inode->i_pipe = NULL; |
153 | inode->i_bdev = NULL; | |
154 | inode->i_cdev = NULL; | |
61ba64fc | 155 | inode->i_link = NULL; |
84e710da | 156 | inode->i_dir_seq = 0; |
2cb1599f DC |
157 | inode->i_rdev = 0; |
158 | inode->dirtied_when = 0; | |
6146f0d5 | 159 | |
3d65ae46 TE |
160 | #ifdef CONFIG_CGROUP_WRITEBACK |
161 | inode->i_wb_frn_winner = 0; | |
162 | inode->i_wb_frn_avg_time = 0; | |
163 | inode->i_wb_frn_history = 0; | |
164 | #endif | |
165 | ||
6146f0d5 | 166 | if (security_inode_alloc(inode)) |
54e34621 | 167 | goto out; |
2cb1599f DC |
168 | spin_lock_init(&inode->i_lock); |
169 | lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key); | |
170 | ||
9902af79 AV |
171 | init_rwsem(&inode->i_rwsem); |
172 | lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key); | |
2cb1599f | 173 | |
bd5fe6c5 | 174 | atomic_set(&inode->i_dio_count, 0); |
2cb1599f DC |
175 | |
176 | mapping->a_ops = &empty_aops; | |
177 | mapping->host = inode; | |
178 | mapping->flags = 0; | |
4bb5f5d9 | 179 | atomic_set(&mapping->i_mmap_writable, 0); |
3c1d4378 | 180 | mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE); |
252aa6f5 | 181 | mapping->private_data = NULL; |
2cb1599f | 182 | mapping->writeback_index = 0; |
2cb1599f DC |
183 | inode->i_private = NULL; |
184 | inode->i_mapping = mapping; | |
b3d9b7a3 | 185 | INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */ |
f19d4a8f AV |
186 | #ifdef CONFIG_FS_POSIX_ACL |
187 | inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED; | |
188 | #endif | |
2cb1599f | 189 | |
3be25f49 EP |
190 | #ifdef CONFIG_FSNOTIFY |
191 | inode->i_fsnotify_mask = 0; | |
192 | #endif | |
4a075e39 | 193 | inode->i_flctx = NULL; |
3e880fb5 | 194 | this_cpu_inc(nr_inodes); |
cffbc8aa | 195 | |
54e34621 | 196 | return 0; |
54e34621 CH |
197 | out: |
198 | return -ENOMEM; | |
1da177e4 | 199 | } |
2cb1599f DC |
200 | EXPORT_SYMBOL(inode_init_always); |
201 | ||
202 | static struct inode *alloc_inode(struct super_block *sb) | |
203 | { | |
204 | struct inode *inode; | |
205 | ||
206 | if (sb->s_op->alloc_inode) | |
207 | inode = sb->s_op->alloc_inode(sb); | |
208 | else | |
209 | inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL); | |
210 | ||
54e34621 CH |
211 | if (!inode) |
212 | return NULL; | |
213 | ||
214 | if (unlikely(inode_init_always(sb, inode))) { | |
215 | if (inode->i_sb->s_op->destroy_inode) | |
216 | inode->i_sb->s_op->destroy_inode(inode); | |
217 | else | |
218 | kmem_cache_free(inode_cachep, inode); | |
219 | return NULL; | |
220 | } | |
221 | ||
222 | return inode; | |
2cb1599f | 223 | } |
1da177e4 | 224 | |
ff0c7d15 NP |
225 | void free_inode_nonrcu(struct inode *inode) |
226 | { | |
227 | kmem_cache_free(inode_cachep, inode); | |
228 | } | |
229 | EXPORT_SYMBOL(free_inode_nonrcu); | |
230 | ||
2e00c97e | 231 | void __destroy_inode(struct inode *inode) |
1da177e4 | 232 | { |
b7542f8c | 233 | BUG_ON(inode_has_buffers(inode)); |
52ebea74 | 234 | inode_detach_wb(inode); |
1da177e4 | 235 | security_inode_free(inode); |
3be25f49 | 236 | fsnotify_inode_delete(inode); |
f27a0fe0 | 237 | locks_free_lock_context(inode); |
7ada4db8 MS |
238 | if (!inode->i_nlink) { |
239 | WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0); | |
240 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
241 | } | |
242 | ||
f19d4a8f | 243 | #ifdef CONFIG_FS_POSIX_ACL |
b8a7a3a6 | 244 | if (inode->i_acl && !is_uncached_acl(inode->i_acl)) |
f19d4a8f | 245 | posix_acl_release(inode->i_acl); |
b8a7a3a6 | 246 | if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl)) |
f19d4a8f AV |
247 | posix_acl_release(inode->i_default_acl); |
248 | #endif | |
3e880fb5 | 249 | this_cpu_dec(nr_inodes); |
2e00c97e CH |
250 | } |
251 | EXPORT_SYMBOL(__destroy_inode); | |
252 | ||
fa0d7e3d NP |
253 | static void i_callback(struct rcu_head *head) |
254 | { | |
255 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
256 | kmem_cache_free(inode_cachep, inode); |
257 | } | |
258 | ||
56b0dacf | 259 | static void destroy_inode(struct inode *inode) |
2e00c97e | 260 | { |
7ccf19a8 | 261 | BUG_ON(!list_empty(&inode->i_lru)); |
2e00c97e | 262 | __destroy_inode(inode); |
1da177e4 LT |
263 | if (inode->i_sb->s_op->destroy_inode) |
264 | inode->i_sb->s_op->destroy_inode(inode); | |
265 | else | |
fa0d7e3d | 266 | call_rcu(&inode->i_rcu, i_callback); |
1da177e4 | 267 | } |
1da177e4 | 268 | |
7ada4db8 MS |
269 | /** |
270 | * drop_nlink - directly drop an inode's link count | |
271 | * @inode: inode | |
272 | * | |
273 | * This is a low-level filesystem helper to replace any | |
274 | * direct filesystem manipulation of i_nlink. In cases | |
275 | * where we are attempting to track writes to the | |
276 | * filesystem, a decrement to zero means an imminent | |
277 | * write when the file is truncated and actually unlinked | |
278 | * on the filesystem. | |
279 | */ | |
280 | void drop_nlink(struct inode *inode) | |
281 | { | |
282 | WARN_ON(inode->i_nlink == 0); | |
283 | inode->__i_nlink--; | |
284 | if (!inode->i_nlink) | |
285 | atomic_long_inc(&inode->i_sb->s_remove_count); | |
286 | } | |
287 | EXPORT_SYMBOL(drop_nlink); | |
288 | ||
289 | /** | |
290 | * clear_nlink - directly zero an inode's link count | |
291 | * @inode: inode | |
292 | * | |
293 | * This is a low-level filesystem helper to replace any | |
294 | * direct filesystem manipulation of i_nlink. See | |
295 | * drop_nlink() for why we care about i_nlink hitting zero. | |
296 | */ | |
297 | void clear_nlink(struct inode *inode) | |
298 | { | |
299 | if (inode->i_nlink) { | |
300 | inode->__i_nlink = 0; | |
301 | atomic_long_inc(&inode->i_sb->s_remove_count); | |
302 | } | |
303 | } | |
304 | EXPORT_SYMBOL(clear_nlink); | |
305 | ||
306 | /** | |
307 | * set_nlink - directly set an inode's link count | |
308 | * @inode: inode | |
309 | * @nlink: new nlink (should be non-zero) | |
310 | * | |
311 | * This is a low-level filesystem helper to replace any | |
312 | * direct filesystem manipulation of i_nlink. | |
313 | */ | |
314 | void set_nlink(struct inode *inode, unsigned int nlink) | |
315 | { | |
316 | if (!nlink) { | |
7ada4db8 MS |
317 | clear_nlink(inode); |
318 | } else { | |
319 | /* Yes, some filesystems do change nlink from zero to one */ | |
320 | if (inode->i_nlink == 0) | |
321 | atomic_long_dec(&inode->i_sb->s_remove_count); | |
322 | ||
323 | inode->__i_nlink = nlink; | |
324 | } | |
325 | } | |
326 | EXPORT_SYMBOL(set_nlink); | |
327 | ||
328 | /** | |
329 | * inc_nlink - directly increment an inode's link count | |
330 | * @inode: inode | |
331 | * | |
332 | * This is a low-level filesystem helper to replace any | |
333 | * direct filesystem manipulation of i_nlink. Currently, | |
334 | * it is only here for parity with dec_nlink(). | |
335 | */ | |
336 | void inc_nlink(struct inode *inode) | |
337 | { | |
f4e0c30c AV |
338 | if (unlikely(inode->i_nlink == 0)) { |
339 | WARN_ON(!(inode->i_state & I_LINKABLE)); | |
7ada4db8 | 340 | atomic_long_dec(&inode->i_sb->s_remove_count); |
f4e0c30c | 341 | } |
7ada4db8 MS |
342 | |
343 | inode->__i_nlink++; | |
344 | } | |
345 | EXPORT_SYMBOL(inc_nlink); | |
346 | ||
2aa15890 MS |
347 | void address_space_init_once(struct address_space *mapping) |
348 | { | |
349 | memset(mapping, 0, sizeof(*mapping)); | |
05eb6e72 | 350 | INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC | __GFP_ACCOUNT); |
2aa15890 | 351 | spin_lock_init(&mapping->tree_lock); |
c8c06efa | 352 | init_rwsem(&mapping->i_mmap_rwsem); |
2aa15890 MS |
353 | INIT_LIST_HEAD(&mapping->private_list); |
354 | spin_lock_init(&mapping->private_lock); | |
6b2dbba8 | 355 | mapping->i_mmap = RB_ROOT; |
2aa15890 MS |
356 | } |
357 | EXPORT_SYMBOL(address_space_init_once); | |
358 | ||
1da177e4 LT |
359 | /* |
360 | * These are initializations that only need to be done | |
361 | * once, because the fields are idempotent across use | |
362 | * of the inode, so let the slab aware of that. | |
363 | */ | |
364 | void inode_init_once(struct inode *inode) | |
365 | { | |
366 | memset(inode, 0, sizeof(*inode)); | |
367 | INIT_HLIST_NODE(&inode->i_hash); | |
1da177e4 | 368 | INIT_LIST_HEAD(&inode->i_devices); |
c7f54084 | 369 | INIT_LIST_HEAD(&inode->i_io_list); |
6c60d2b5 | 370 | INIT_LIST_HEAD(&inode->i_wb_list); |
7ccf19a8 | 371 | INIT_LIST_HEAD(&inode->i_lru); |
2aa15890 | 372 | address_space_init_once(&inode->i_data); |
1da177e4 LT |
373 | i_size_ordered_init(inode); |
374 | } | |
1da177e4 LT |
375 | EXPORT_SYMBOL(inode_init_once); |
376 | ||
51cc5068 | 377 | static void init_once(void *foo) |
1da177e4 | 378 | { |
6b3304b5 | 379 | struct inode *inode = (struct inode *) foo; |
1da177e4 | 380 | |
a35afb83 | 381 | inode_init_once(inode); |
1da177e4 LT |
382 | } |
383 | ||
384 | /* | |
250df6ed | 385 | * inode->i_lock must be held |
1da177e4 | 386 | */ |
6b3304b5 | 387 | void __iget(struct inode *inode) |
1da177e4 | 388 | { |
9e38d86f NP |
389 | atomic_inc(&inode->i_count); |
390 | } | |
2e147f1e | 391 | |
7de9c6ee AV |
392 | /* |
393 | * get additional reference to inode; caller must already hold one. | |
394 | */ | |
395 | void ihold(struct inode *inode) | |
396 | { | |
397 | WARN_ON(atomic_inc_return(&inode->i_count) < 2); | |
398 | } | |
399 | EXPORT_SYMBOL(ihold); | |
400 | ||
9e38d86f NP |
401 | static void inode_lru_list_add(struct inode *inode) |
402 | { | |
bc3b14cb | 403 | if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru)) |
fcb94f72 | 404 | this_cpu_inc(nr_unused); |
9e38d86f | 405 | } |
2e147f1e | 406 | |
4eff96dd JK |
407 | /* |
408 | * Add inode to LRU if needed (inode is unused and clean). | |
409 | * | |
410 | * Needs inode->i_lock held. | |
411 | */ | |
412 | void inode_add_lru(struct inode *inode) | |
413 | { | |
0ae45f63 TT |
414 | if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC | |
415 | I_FREEING | I_WILL_FREE)) && | |
4eff96dd JK |
416 | !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE) |
417 | inode_lru_list_add(inode); | |
418 | } | |
419 | ||
420 | ||
9e38d86f NP |
421 | static void inode_lru_list_del(struct inode *inode) |
422 | { | |
bc3b14cb DC |
423 | |
424 | if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru)) | |
fcb94f72 | 425 | this_cpu_dec(nr_unused); |
1da177e4 LT |
426 | } |
427 | ||
646ec461 CH |
428 | /** |
429 | * inode_sb_list_add - add inode to the superblock list of inodes | |
430 | * @inode: inode to add | |
431 | */ | |
432 | void inode_sb_list_add(struct inode *inode) | |
433 | { | |
74278da9 | 434 | spin_lock(&inode->i_sb->s_inode_list_lock); |
55fa6091 | 435 | list_add(&inode->i_sb_list, &inode->i_sb->s_inodes); |
74278da9 | 436 | spin_unlock(&inode->i_sb->s_inode_list_lock); |
646ec461 CH |
437 | } |
438 | EXPORT_SYMBOL_GPL(inode_sb_list_add); | |
439 | ||
55fa6091 | 440 | static inline void inode_sb_list_del(struct inode *inode) |
646ec461 | 441 | { |
a209dfc7 | 442 | if (!list_empty(&inode->i_sb_list)) { |
74278da9 | 443 | spin_lock(&inode->i_sb->s_inode_list_lock); |
a209dfc7 | 444 | list_del_init(&inode->i_sb_list); |
74278da9 | 445 | spin_unlock(&inode->i_sb->s_inode_list_lock); |
a209dfc7 | 446 | } |
646ec461 CH |
447 | } |
448 | ||
4c51acbc DC |
449 | static unsigned long hash(struct super_block *sb, unsigned long hashval) |
450 | { | |
451 | unsigned long tmp; | |
452 | ||
453 | tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) / | |
454 | L1_CACHE_BYTES; | |
4b4563dc CH |
455 | tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift); |
456 | return tmp & i_hash_mask; | |
4c51acbc DC |
457 | } |
458 | ||
459 | /** | |
460 | * __insert_inode_hash - hash an inode | |
461 | * @inode: unhashed inode | |
462 | * @hashval: unsigned long value used to locate this object in the | |
463 | * inode_hashtable. | |
464 | * | |
465 | * Add an inode to the inode hash for this superblock. | |
466 | */ | |
467 | void __insert_inode_hash(struct inode *inode, unsigned long hashval) | |
468 | { | |
646ec461 CH |
469 | struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval); |
470 | ||
67a23c49 | 471 | spin_lock(&inode_hash_lock); |
250df6ed | 472 | spin_lock(&inode->i_lock); |
646ec461 | 473 | hlist_add_head(&inode->i_hash, b); |
250df6ed | 474 | spin_unlock(&inode->i_lock); |
67a23c49 | 475 | spin_unlock(&inode_hash_lock); |
4c51acbc DC |
476 | } |
477 | EXPORT_SYMBOL(__insert_inode_hash); | |
478 | ||
4c51acbc | 479 | /** |
f2ee7abf | 480 | * __remove_inode_hash - remove an inode from the hash |
4c51acbc DC |
481 | * @inode: inode to unhash |
482 | * | |
483 | * Remove an inode from the superblock. | |
484 | */ | |
f2ee7abf | 485 | void __remove_inode_hash(struct inode *inode) |
4c51acbc | 486 | { |
67a23c49 | 487 | spin_lock(&inode_hash_lock); |
250df6ed | 488 | spin_lock(&inode->i_lock); |
4c51acbc | 489 | hlist_del_init(&inode->i_hash); |
250df6ed | 490 | spin_unlock(&inode->i_lock); |
67a23c49 | 491 | spin_unlock(&inode_hash_lock); |
4c51acbc | 492 | } |
f2ee7abf | 493 | EXPORT_SYMBOL(__remove_inode_hash); |
4c51acbc | 494 | |
dbd5768f | 495 | void clear_inode(struct inode *inode) |
b0683aa6 AV |
496 | { |
497 | might_sleep(); | |
08142579 JK |
498 | /* |
499 | * We have to cycle tree_lock here because reclaim can be still in the | |
500 | * process of removing the last page (in __delete_from_page_cache()) | |
501 | * and we must not free mapping under it. | |
502 | */ | |
503 | spin_lock_irq(&inode->i_data.tree_lock); | |
b0683aa6 | 504 | BUG_ON(inode->i_data.nrpages); |
f9fe48be | 505 | BUG_ON(inode->i_data.nrexceptional); |
08142579 | 506 | spin_unlock_irq(&inode->i_data.tree_lock); |
b0683aa6 AV |
507 | BUG_ON(!list_empty(&inode->i_data.private_list)); |
508 | BUG_ON(!(inode->i_state & I_FREEING)); | |
509 | BUG_ON(inode->i_state & I_CLEAR); | |
6c60d2b5 | 510 | BUG_ON(!list_empty(&inode->i_wb_list)); |
fa0d7e3d | 511 | /* don't need i_lock here, no concurrent mods to i_state */ |
b0683aa6 AV |
512 | inode->i_state = I_FREEING | I_CLEAR; |
513 | } | |
dbd5768f | 514 | EXPORT_SYMBOL(clear_inode); |
b0683aa6 | 515 | |
b2b2af8e DC |
516 | /* |
517 | * Free the inode passed in, removing it from the lists it is still connected | |
518 | * to. We remove any pages still attached to the inode and wait for any IO that | |
519 | * is still in progress before finally destroying the inode. | |
520 | * | |
521 | * An inode must already be marked I_FREEING so that we avoid the inode being | |
522 | * moved back onto lists if we race with other code that manipulates the lists | |
523 | * (e.g. writeback_single_inode). The caller is responsible for setting this. | |
524 | * | |
525 | * An inode must already be removed from the LRU list before being evicted from | |
526 | * the cache. This should occur atomically with setting the I_FREEING state | |
527 | * flag, so no inodes here should ever be on the LRU when being evicted. | |
528 | */ | |
644da596 | 529 | static void evict(struct inode *inode) |
b4272d4c AV |
530 | { |
531 | const struct super_operations *op = inode->i_sb->s_op; | |
532 | ||
b2b2af8e DC |
533 | BUG_ON(!(inode->i_state & I_FREEING)); |
534 | BUG_ON(!list_empty(&inode->i_lru)); | |
535 | ||
c7f54084 DC |
536 | if (!list_empty(&inode->i_io_list)) |
537 | inode_io_list_del(inode); | |
b12362bd | 538 | |
55fa6091 DC |
539 | inode_sb_list_del(inode); |
540 | ||
169ebd90 JK |
541 | /* |
542 | * Wait for flusher thread to be done with the inode so that filesystem | |
543 | * does not start destroying it while writeback is still running. Since | |
544 | * the inode has I_FREEING set, flusher thread won't start new work on | |
545 | * the inode. We just have to wait for running writeback to finish. | |
546 | */ | |
547 | inode_wait_for_writeback(inode); | |
7994e6f7 | 548 | |
be7ce416 AV |
549 | if (op->evict_inode) { |
550 | op->evict_inode(inode); | |
b4272d4c | 551 | } else { |
91b0abe3 | 552 | truncate_inode_pages_final(&inode->i_data); |
dbd5768f | 553 | clear_inode(inode); |
b4272d4c | 554 | } |
661074e9 AV |
555 | if (S_ISBLK(inode->i_mode) && inode->i_bdev) |
556 | bd_forget(inode); | |
557 | if (S_ISCHR(inode->i_mode) && inode->i_cdev) | |
558 | cd_forget(inode); | |
b2b2af8e DC |
559 | |
560 | remove_inode_hash(inode); | |
561 | ||
562 | spin_lock(&inode->i_lock); | |
563 | wake_up_bit(&inode->i_state, __I_NEW); | |
564 | BUG_ON(inode->i_state != (I_FREEING | I_CLEAR)); | |
565 | spin_unlock(&inode->i_lock); | |
566 | ||
567 | destroy_inode(inode); | |
b4272d4c AV |
568 | } |
569 | ||
1da177e4 LT |
570 | /* |
571 | * dispose_list - dispose of the contents of a local list | |
572 | * @head: the head of the list to free | |
573 | * | |
574 | * Dispose-list gets a local list with local inodes in it, so it doesn't | |
575 | * need to worry about list corruption and SMP locks. | |
576 | */ | |
577 | static void dispose_list(struct list_head *head) | |
578 | { | |
1da177e4 LT |
579 | while (!list_empty(head)) { |
580 | struct inode *inode; | |
581 | ||
7ccf19a8 NP |
582 | inode = list_first_entry(head, struct inode, i_lru); |
583 | list_del_init(&inode->i_lru); | |
1da177e4 | 584 | |
644da596 | 585 | evict(inode); |
ac05fbb4 | 586 | cond_resched(); |
1da177e4 | 587 | } |
1da177e4 LT |
588 | } |
589 | ||
63997e98 AV |
590 | /** |
591 | * evict_inodes - evict all evictable inodes for a superblock | |
592 | * @sb: superblock to operate on | |
593 | * | |
594 | * Make sure that no inodes with zero refcount are retained. This is | |
595 | * called by superblock shutdown after having MS_ACTIVE flag removed, | |
596 | * so any inode reaching zero refcount during or after that call will | |
597 | * be immediately evicted. | |
1da177e4 | 598 | */ |
63997e98 | 599 | void evict_inodes(struct super_block *sb) |
1da177e4 | 600 | { |
63997e98 AV |
601 | struct inode *inode, *next; |
602 | LIST_HEAD(dispose); | |
1da177e4 | 603 | |
ac05fbb4 | 604 | again: |
74278da9 | 605 | spin_lock(&sb->s_inode_list_lock); |
63997e98 AV |
606 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
607 | if (atomic_read(&inode->i_count)) | |
aabb8fdb | 608 | continue; |
250df6ed DC |
609 | |
610 | spin_lock(&inode->i_lock); | |
611 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
612 | spin_unlock(&inode->i_lock); | |
1da177e4 | 613 | continue; |
250df6ed | 614 | } |
63997e98 AV |
615 | |
616 | inode->i_state |= I_FREEING; | |
02afc410 | 617 | inode_lru_list_del(inode); |
250df6ed | 618 | spin_unlock(&inode->i_lock); |
02afc410 | 619 | list_add(&inode->i_lru, &dispose); |
ac05fbb4 JB |
620 | |
621 | /* | |
622 | * We can have a ton of inodes to evict at unmount time given | |
623 | * enough memory, check to see if we need to go to sleep for a | |
624 | * bit so we don't livelock. | |
625 | */ | |
626 | if (need_resched()) { | |
627 | spin_unlock(&sb->s_inode_list_lock); | |
628 | cond_resched(); | |
629 | dispose_list(&dispose); | |
630 | goto again; | |
631 | } | |
1da177e4 | 632 | } |
74278da9 | 633 | spin_unlock(&sb->s_inode_list_lock); |
63997e98 AV |
634 | |
635 | dispose_list(&dispose); | |
1da177e4 LT |
636 | } |
637 | ||
1da177e4 | 638 | /** |
a0318786 CH |
639 | * invalidate_inodes - attempt to free all inodes on a superblock |
640 | * @sb: superblock to operate on | |
93b270f7 | 641 | * @kill_dirty: flag to guide handling of dirty inodes |
1da177e4 | 642 | * |
a0318786 CH |
643 | * Attempts to free all inodes for a given superblock. If there were any |
644 | * busy inodes return a non-zero value, else zero. | |
93b270f7 N |
645 | * If @kill_dirty is set, discard dirty inodes too, otherwise treat |
646 | * them as busy. | |
1da177e4 | 647 | */ |
93b270f7 | 648 | int invalidate_inodes(struct super_block *sb, bool kill_dirty) |
1da177e4 | 649 | { |
cffbc8aa | 650 | int busy = 0; |
a0318786 CH |
651 | struct inode *inode, *next; |
652 | LIST_HEAD(dispose); | |
1da177e4 | 653 | |
74278da9 | 654 | spin_lock(&sb->s_inode_list_lock); |
a0318786 | 655 | list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) { |
250df6ed DC |
656 | spin_lock(&inode->i_lock); |
657 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
658 | spin_unlock(&inode->i_lock); | |
aabb8fdb | 659 | continue; |
250df6ed | 660 | } |
0ae45f63 | 661 | if (inode->i_state & I_DIRTY_ALL && !kill_dirty) { |
250df6ed | 662 | spin_unlock(&inode->i_lock); |
93b270f7 N |
663 | busy = 1; |
664 | continue; | |
665 | } | |
99a38919 | 666 | if (atomic_read(&inode->i_count)) { |
250df6ed | 667 | spin_unlock(&inode->i_lock); |
99a38919 | 668 | busy = 1; |
1da177e4 LT |
669 | continue; |
670 | } | |
99a38919 | 671 | |
99a38919 | 672 | inode->i_state |= I_FREEING; |
02afc410 | 673 | inode_lru_list_del(inode); |
250df6ed | 674 | spin_unlock(&inode->i_lock); |
02afc410 | 675 | list_add(&inode->i_lru, &dispose); |
1da177e4 | 676 | } |
74278da9 | 677 | spin_unlock(&sb->s_inode_list_lock); |
1da177e4 | 678 | |
a0318786 | 679 | dispose_list(&dispose); |
1da177e4 LT |
680 | |
681 | return busy; | |
682 | } | |
1da177e4 | 683 | |
1da177e4 | 684 | /* |
bc3b14cb | 685 | * Isolate the inode from the LRU in preparation for freeing it. |
1da177e4 LT |
686 | * |
687 | * Any inodes which are pinned purely because of attached pagecache have their | |
9e38d86f NP |
688 | * pagecache removed. If the inode has metadata buffers attached to |
689 | * mapping->private_list then try to remove them. | |
1da177e4 | 690 | * |
9e38d86f NP |
691 | * If the inode has the I_REFERENCED flag set, then it means that it has been |
692 | * used recently - the flag is set in iput_final(). When we encounter such an | |
693 | * inode, clear the flag and move it to the back of the LRU so it gets another | |
694 | * pass through the LRU before it gets reclaimed. This is necessary because of | |
695 | * the fact we are doing lazy LRU updates to minimise lock contention so the | |
696 | * LRU does not have strict ordering. Hence we don't want to reclaim inodes | |
697 | * with this flag set because they are the inodes that are out of order. | |
1da177e4 | 698 | */ |
3f97b163 VD |
699 | static enum lru_status inode_lru_isolate(struct list_head *item, |
700 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
1da177e4 | 701 | { |
bc3b14cb DC |
702 | struct list_head *freeable = arg; |
703 | struct inode *inode = container_of(item, struct inode, i_lru); | |
1da177e4 | 704 | |
bc3b14cb DC |
705 | /* |
706 | * we are inverting the lru lock/inode->i_lock here, so use a trylock. | |
707 | * If we fail to get the lock, just skip it. | |
708 | */ | |
709 | if (!spin_trylock(&inode->i_lock)) | |
710 | return LRU_SKIP; | |
1da177e4 | 711 | |
bc3b14cb DC |
712 | /* |
713 | * Referenced or dirty inodes are still in use. Give them another pass | |
714 | * through the LRU as we canot reclaim them now. | |
715 | */ | |
716 | if (atomic_read(&inode->i_count) || | |
717 | (inode->i_state & ~I_REFERENCED)) { | |
3f97b163 | 718 | list_lru_isolate(lru, &inode->i_lru); |
bc3b14cb DC |
719 | spin_unlock(&inode->i_lock); |
720 | this_cpu_dec(nr_unused); | |
721 | return LRU_REMOVED; | |
722 | } | |
1da177e4 | 723 | |
bc3b14cb DC |
724 | /* recently referenced inodes get one more pass */ |
725 | if (inode->i_state & I_REFERENCED) { | |
726 | inode->i_state &= ~I_REFERENCED; | |
727 | spin_unlock(&inode->i_lock); | |
728 | return LRU_ROTATE; | |
729 | } | |
1da177e4 | 730 | |
bc3b14cb DC |
731 | if (inode_has_buffers(inode) || inode->i_data.nrpages) { |
732 | __iget(inode); | |
733 | spin_unlock(&inode->i_lock); | |
734 | spin_unlock(lru_lock); | |
735 | if (remove_inode_buffers(inode)) { | |
736 | unsigned long reap; | |
737 | reap = invalidate_mapping_pages(&inode->i_data, 0, -1); | |
738 | if (current_is_kswapd()) | |
739 | __count_vm_events(KSWAPD_INODESTEAL, reap); | |
740 | else | |
741 | __count_vm_events(PGINODESTEAL, reap); | |
742 | if (current->reclaim_state) | |
743 | current->reclaim_state->reclaimed_slab += reap; | |
02afc410 | 744 | } |
bc3b14cb DC |
745 | iput(inode); |
746 | spin_lock(lru_lock); | |
747 | return LRU_RETRY; | |
748 | } | |
02afc410 | 749 | |
bc3b14cb DC |
750 | WARN_ON(inode->i_state & I_NEW); |
751 | inode->i_state |= I_FREEING; | |
3f97b163 | 752 | list_lru_isolate_move(lru, &inode->i_lru, freeable); |
bc3b14cb | 753 | spin_unlock(&inode->i_lock); |
9e38d86f | 754 | |
bc3b14cb DC |
755 | this_cpu_dec(nr_unused); |
756 | return LRU_REMOVED; | |
757 | } | |
7ccf19a8 | 758 | |
bc3b14cb DC |
759 | /* |
760 | * Walk the superblock inode LRU for freeable inodes and attempt to free them. | |
761 | * This is called from the superblock shrinker function with a number of inodes | |
762 | * to trim from the LRU. Inodes to be freed are moved to a temporary list and | |
763 | * then are freed outside inode_lock by dispose_list(). | |
764 | */ | |
503c358c | 765 | long prune_icache_sb(struct super_block *sb, struct shrink_control *sc) |
bc3b14cb DC |
766 | { |
767 | LIST_HEAD(freeable); | |
768 | long freed; | |
1da177e4 | 769 | |
503c358c VD |
770 | freed = list_lru_shrink_walk(&sb->s_inode_lru, sc, |
771 | inode_lru_isolate, &freeable); | |
1da177e4 | 772 | dispose_list(&freeable); |
0a234c6d | 773 | return freed; |
1da177e4 LT |
774 | } |
775 | ||
1da177e4 LT |
776 | static void __wait_on_freeing_inode(struct inode *inode); |
777 | /* | |
778 | * Called with the inode lock held. | |
1da177e4 | 779 | */ |
6b3304b5 MK |
780 | static struct inode *find_inode(struct super_block *sb, |
781 | struct hlist_head *head, | |
782 | int (*test)(struct inode *, void *), | |
783 | void *data) | |
1da177e4 | 784 | { |
6b3304b5 | 785 | struct inode *inode = NULL; |
1da177e4 LT |
786 | |
787 | repeat: | |
b67bfe0d | 788 | hlist_for_each_entry(inode, head, i_hash) { |
5a3cd992 | 789 | if (inode->i_sb != sb) |
1da177e4 | 790 | continue; |
5a3cd992 | 791 | if (!test(inode, data)) |
1da177e4 | 792 | continue; |
5a3cd992 | 793 | spin_lock(&inode->i_lock); |
a4ffdde6 | 794 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
795 | __wait_on_freeing_inode(inode); |
796 | goto repeat; | |
797 | } | |
f7899bd5 | 798 | __iget(inode); |
250df6ed | 799 | spin_unlock(&inode->i_lock); |
f7899bd5 | 800 | return inode; |
1da177e4 | 801 | } |
f7899bd5 | 802 | return NULL; |
1da177e4 LT |
803 | } |
804 | ||
805 | /* | |
806 | * find_inode_fast is the fast path version of find_inode, see the comment at | |
807 | * iget_locked for details. | |
808 | */ | |
6b3304b5 MK |
809 | static struct inode *find_inode_fast(struct super_block *sb, |
810 | struct hlist_head *head, unsigned long ino) | |
1da177e4 | 811 | { |
6b3304b5 | 812 | struct inode *inode = NULL; |
1da177e4 LT |
813 | |
814 | repeat: | |
b67bfe0d | 815 | hlist_for_each_entry(inode, head, i_hash) { |
5a3cd992 | 816 | if (inode->i_ino != ino) |
1da177e4 | 817 | continue; |
5a3cd992 | 818 | if (inode->i_sb != sb) |
1da177e4 | 819 | continue; |
5a3cd992 | 820 | spin_lock(&inode->i_lock); |
a4ffdde6 | 821 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) { |
1da177e4 LT |
822 | __wait_on_freeing_inode(inode); |
823 | goto repeat; | |
824 | } | |
f7899bd5 | 825 | __iget(inode); |
250df6ed | 826 | spin_unlock(&inode->i_lock); |
f7899bd5 | 827 | return inode; |
1da177e4 | 828 | } |
f7899bd5 | 829 | return NULL; |
8290c35f DC |
830 | } |
831 | ||
f991bd2e ED |
832 | /* |
833 | * Each cpu owns a range of LAST_INO_BATCH numbers. | |
834 | * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations, | |
835 | * to renew the exhausted range. | |
8290c35f | 836 | * |
f991bd2e ED |
837 | * This does not significantly increase overflow rate because every CPU can |
838 | * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is | |
839 | * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the | |
840 | * 2^32 range, and is a worst-case. Even a 50% wastage would only increase | |
841 | * overflow rate by 2x, which does not seem too significant. | |
842 | * | |
843 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
844 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
845 | * here to attempt to avoid that. | |
8290c35f | 846 | */ |
f991bd2e ED |
847 | #define LAST_INO_BATCH 1024 |
848 | static DEFINE_PER_CPU(unsigned int, last_ino); | |
849 | ||
85fe4025 | 850 | unsigned int get_next_ino(void) |
8290c35f | 851 | { |
f991bd2e ED |
852 | unsigned int *p = &get_cpu_var(last_ino); |
853 | unsigned int res = *p; | |
8290c35f | 854 | |
f991bd2e ED |
855 | #ifdef CONFIG_SMP |
856 | if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) { | |
857 | static atomic_t shared_last_ino; | |
858 | int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino); | |
859 | ||
860 | res = next - LAST_INO_BATCH; | |
861 | } | |
862 | #endif | |
863 | ||
2adc376c CM |
864 | res++; |
865 | /* get_next_ino should not provide a 0 inode number */ | |
866 | if (unlikely(!res)) | |
867 | res++; | |
868 | *p = res; | |
f991bd2e ED |
869 | put_cpu_var(last_ino); |
870 | return res; | |
8290c35f | 871 | } |
85fe4025 | 872 | EXPORT_SYMBOL(get_next_ino); |
8290c35f | 873 | |
a209dfc7 ED |
874 | /** |
875 | * new_inode_pseudo - obtain an inode | |
876 | * @sb: superblock | |
877 | * | |
878 | * Allocates a new inode for given superblock. | |
879 | * Inode wont be chained in superblock s_inodes list | |
880 | * This means : | |
881 | * - fs can't be unmount | |
882 | * - quotas, fsnotify, writeback can't work | |
883 | */ | |
884 | struct inode *new_inode_pseudo(struct super_block *sb) | |
885 | { | |
886 | struct inode *inode = alloc_inode(sb); | |
887 | ||
888 | if (inode) { | |
889 | spin_lock(&inode->i_lock); | |
890 | inode->i_state = 0; | |
891 | spin_unlock(&inode->i_lock); | |
892 | INIT_LIST_HEAD(&inode->i_sb_list); | |
893 | } | |
894 | return inode; | |
895 | } | |
896 | ||
1da177e4 LT |
897 | /** |
898 | * new_inode - obtain an inode | |
899 | * @sb: superblock | |
900 | * | |
769848c0 | 901 | * Allocates a new inode for given superblock. The default gfp_mask |
3c1d4378 | 902 | * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE. |
769848c0 MG |
903 | * If HIGHMEM pages are unsuitable or it is known that pages allocated |
904 | * for the page cache are not reclaimable or migratable, | |
905 | * mapping_set_gfp_mask() must be called with suitable flags on the | |
906 | * newly created inode's mapping | |
907 | * | |
1da177e4 LT |
908 | */ |
909 | struct inode *new_inode(struct super_block *sb) | |
910 | { | |
6b3304b5 | 911 | struct inode *inode; |
1da177e4 | 912 | |
74278da9 | 913 | spin_lock_prefetch(&sb->s_inode_list_lock); |
6b3304b5 | 914 | |
a209dfc7 ED |
915 | inode = new_inode_pseudo(sb); |
916 | if (inode) | |
55fa6091 | 917 | inode_sb_list_add(inode); |
1da177e4 LT |
918 | return inode; |
919 | } | |
1da177e4 LT |
920 | EXPORT_SYMBOL(new_inode); |
921 | ||
14358e6d | 922 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
e096d0c7 JB |
923 | void lockdep_annotate_inode_mutex_key(struct inode *inode) |
924 | { | |
a3314a0e | 925 | if (S_ISDIR(inode->i_mode)) { |
1e89a5e1 PZ |
926 | struct file_system_type *type = inode->i_sb->s_type; |
927 | ||
9a7aa12f | 928 | /* Set new key only if filesystem hasn't already changed it */ |
9902af79 | 929 | if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) { |
9a7aa12f JK |
930 | /* |
931 | * ensure nobody is actually holding i_mutex | |
932 | */ | |
9902af79 AV |
933 | // mutex_destroy(&inode->i_mutex); |
934 | init_rwsem(&inode->i_rwsem); | |
935 | lockdep_set_class(&inode->i_rwsem, | |
9a7aa12f JK |
936 | &type->i_mutex_dir_key); |
937 | } | |
1e89a5e1 | 938 | } |
e096d0c7 JB |
939 | } |
940 | EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key); | |
14358e6d | 941 | #endif |
e096d0c7 JB |
942 | |
943 | /** | |
944 | * unlock_new_inode - clear the I_NEW state and wake up any waiters | |
945 | * @inode: new inode to unlock | |
946 | * | |
947 | * Called when the inode is fully initialised to clear the new state of the | |
948 | * inode and wake up anyone waiting for the inode to finish initialisation. | |
949 | */ | |
950 | void unlock_new_inode(struct inode *inode) | |
951 | { | |
952 | lockdep_annotate_inode_mutex_key(inode); | |
250df6ed | 953 | spin_lock(&inode->i_lock); |
eaff8079 CH |
954 | WARN_ON(!(inode->i_state & I_NEW)); |
955 | inode->i_state &= ~I_NEW; | |
310fa7a3 | 956 | smp_mb(); |
250df6ed DC |
957 | wake_up_bit(&inode->i_state, __I_NEW); |
958 | spin_unlock(&inode->i_lock); | |
1da177e4 | 959 | } |
1da177e4 LT |
960 | EXPORT_SYMBOL(unlock_new_inode); |
961 | ||
375e289e BF |
962 | /** |
963 | * lock_two_nondirectories - take two i_mutexes on non-directory objects | |
4fd699ae BF |
964 | * |
965 | * Lock any non-NULL argument that is not a directory. | |
966 | * Zero, one or two objects may be locked by this function. | |
967 | * | |
375e289e BF |
968 | * @inode1: first inode to lock |
969 | * @inode2: second inode to lock | |
970 | */ | |
971 | void lock_two_nondirectories(struct inode *inode1, struct inode *inode2) | |
972 | { | |
4fd699ae BF |
973 | if (inode1 > inode2) |
974 | swap(inode1, inode2); | |
975 | ||
976 | if (inode1 && !S_ISDIR(inode1->i_mode)) | |
5955102c | 977 | inode_lock(inode1); |
4fd699ae | 978 | if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1) |
5955102c | 979 | inode_lock_nested(inode2, I_MUTEX_NONDIR2); |
375e289e BF |
980 | } |
981 | EXPORT_SYMBOL(lock_two_nondirectories); | |
982 | ||
983 | /** | |
984 | * unlock_two_nondirectories - release locks from lock_two_nondirectories() | |
985 | * @inode1: first inode to unlock | |
986 | * @inode2: second inode to unlock | |
987 | */ | |
988 | void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2) | |
989 | { | |
4fd699ae | 990 | if (inode1 && !S_ISDIR(inode1->i_mode)) |
5955102c | 991 | inode_unlock(inode1); |
4fd699ae | 992 | if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1) |
5955102c | 993 | inode_unlock(inode2); |
375e289e BF |
994 | } |
995 | EXPORT_SYMBOL(unlock_two_nondirectories); | |
996 | ||
0b2d0724 CH |
997 | /** |
998 | * iget5_locked - obtain an inode from a mounted file system | |
999 | * @sb: super block of file system | |
1000 | * @hashval: hash value (usually inode number) to get | |
1001 | * @test: callback used for comparisons between inodes | |
1002 | * @set: callback used to initialize a new struct inode | |
1003 | * @data: opaque data pointer to pass to @test and @set | |
1004 | * | |
1005 | * Search for the inode specified by @hashval and @data in the inode cache, | |
1006 | * and if present it is return it with an increased reference count. This is | |
1007 | * a generalized version of iget_locked() for file systems where the inode | |
1008 | * number is not sufficient for unique identification of an inode. | |
1009 | * | |
1010 | * If the inode is not in cache, allocate a new inode and return it locked, | |
1011 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
1012 | * before unlocking it via unlock_new_inode(). | |
1da177e4 | 1013 | * |
0b2d0724 CH |
1014 | * Note both @test and @set are called with the inode_hash_lock held, so can't |
1015 | * sleep. | |
1da177e4 | 1016 | */ |
0b2d0724 CH |
1017 | struct inode *iget5_locked(struct super_block *sb, unsigned long hashval, |
1018 | int (*test)(struct inode *, void *), | |
1019 | int (*set)(struct inode *, void *), void *data) | |
1da177e4 | 1020 | { |
0b2d0724 | 1021 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
6b3304b5 | 1022 | struct inode *inode; |
2864f301 | 1023 | again: |
0b2d0724 CH |
1024 | spin_lock(&inode_hash_lock); |
1025 | inode = find_inode(sb, head, test, data); | |
1026 | spin_unlock(&inode_hash_lock); | |
1027 | ||
1028 | if (inode) { | |
1029 | wait_on_inode(inode); | |
2864f301 AV |
1030 | if (unlikely(inode_unhashed(inode))) { |
1031 | iput(inode); | |
1032 | goto again; | |
1033 | } | |
0b2d0724 CH |
1034 | return inode; |
1035 | } | |
1036 | ||
1da177e4 LT |
1037 | inode = alloc_inode(sb); |
1038 | if (inode) { | |
6b3304b5 | 1039 | struct inode *old; |
1da177e4 | 1040 | |
67a23c49 | 1041 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1042 | /* We released the lock, so.. */ |
1043 | old = find_inode(sb, head, test, data); | |
1044 | if (!old) { | |
1045 | if (set(inode, data)) | |
1046 | goto set_failed; | |
1047 | ||
250df6ed DC |
1048 | spin_lock(&inode->i_lock); |
1049 | inode->i_state = I_NEW; | |
646ec461 | 1050 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1051 | spin_unlock(&inode->i_lock); |
55fa6091 | 1052 | inode_sb_list_add(inode); |
67a23c49 | 1053 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1054 | |
1055 | /* Return the locked inode with I_NEW set, the | |
1056 | * caller is responsible for filling in the contents | |
1057 | */ | |
1058 | return inode; | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * Uhhuh, somebody else created the same inode under | |
1063 | * us. Use the old inode instead of the one we just | |
1064 | * allocated. | |
1065 | */ | |
67a23c49 | 1066 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1067 | destroy_inode(inode); |
1068 | inode = old; | |
1069 | wait_on_inode(inode); | |
2864f301 AV |
1070 | if (unlikely(inode_unhashed(inode))) { |
1071 | iput(inode); | |
1072 | goto again; | |
1073 | } | |
1da177e4 LT |
1074 | } |
1075 | return inode; | |
1076 | ||
1077 | set_failed: | |
67a23c49 | 1078 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1079 | destroy_inode(inode); |
1080 | return NULL; | |
1081 | } | |
0b2d0724 | 1082 | EXPORT_SYMBOL(iget5_locked); |
1da177e4 | 1083 | |
0b2d0724 CH |
1084 | /** |
1085 | * iget_locked - obtain an inode from a mounted file system | |
1086 | * @sb: super block of file system | |
1087 | * @ino: inode number to get | |
1088 | * | |
1089 | * Search for the inode specified by @ino in the inode cache and if present | |
1090 | * return it with an increased reference count. This is for file systems | |
1091 | * where the inode number is sufficient for unique identification of an inode. | |
1092 | * | |
1093 | * If the inode is not in cache, allocate a new inode and return it locked, | |
1094 | * hashed, and with the I_NEW flag set. The file system gets to fill it in | |
1095 | * before unlocking it via unlock_new_inode(). | |
1da177e4 | 1096 | */ |
0b2d0724 | 1097 | struct inode *iget_locked(struct super_block *sb, unsigned long ino) |
1da177e4 | 1098 | { |
0b2d0724 | 1099 | struct hlist_head *head = inode_hashtable + hash(sb, ino); |
6b3304b5 | 1100 | struct inode *inode; |
2864f301 | 1101 | again: |
0b2d0724 CH |
1102 | spin_lock(&inode_hash_lock); |
1103 | inode = find_inode_fast(sb, head, ino); | |
1104 | spin_unlock(&inode_hash_lock); | |
1105 | if (inode) { | |
1106 | wait_on_inode(inode); | |
2864f301 AV |
1107 | if (unlikely(inode_unhashed(inode))) { |
1108 | iput(inode); | |
1109 | goto again; | |
1110 | } | |
0b2d0724 CH |
1111 | return inode; |
1112 | } | |
1113 | ||
1da177e4 LT |
1114 | inode = alloc_inode(sb); |
1115 | if (inode) { | |
6b3304b5 | 1116 | struct inode *old; |
1da177e4 | 1117 | |
67a23c49 | 1118 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1119 | /* We released the lock, so.. */ |
1120 | old = find_inode_fast(sb, head, ino); | |
1121 | if (!old) { | |
1122 | inode->i_ino = ino; | |
250df6ed DC |
1123 | spin_lock(&inode->i_lock); |
1124 | inode->i_state = I_NEW; | |
646ec461 | 1125 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1126 | spin_unlock(&inode->i_lock); |
55fa6091 | 1127 | inode_sb_list_add(inode); |
67a23c49 | 1128 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1129 | |
1130 | /* Return the locked inode with I_NEW set, the | |
1131 | * caller is responsible for filling in the contents | |
1132 | */ | |
1133 | return inode; | |
1134 | } | |
1135 | ||
1136 | /* | |
1137 | * Uhhuh, somebody else created the same inode under | |
1138 | * us. Use the old inode instead of the one we just | |
1139 | * allocated. | |
1140 | */ | |
67a23c49 | 1141 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1142 | destroy_inode(inode); |
1143 | inode = old; | |
1144 | wait_on_inode(inode); | |
2864f301 AV |
1145 | if (unlikely(inode_unhashed(inode))) { |
1146 | iput(inode); | |
1147 | goto again; | |
1148 | } | |
1da177e4 LT |
1149 | } |
1150 | return inode; | |
1151 | } | |
0b2d0724 | 1152 | EXPORT_SYMBOL(iget_locked); |
1da177e4 | 1153 | |
ad5e195a CH |
1154 | /* |
1155 | * search the inode cache for a matching inode number. | |
1156 | * If we find one, then the inode number we are trying to | |
1157 | * allocate is not unique and so we should not use it. | |
1158 | * | |
1159 | * Returns 1 if the inode number is unique, 0 if it is not. | |
1160 | */ | |
1161 | static int test_inode_iunique(struct super_block *sb, unsigned long ino) | |
1162 | { | |
1163 | struct hlist_head *b = inode_hashtable + hash(sb, ino); | |
ad5e195a CH |
1164 | struct inode *inode; |
1165 | ||
67a23c49 | 1166 | spin_lock(&inode_hash_lock); |
b67bfe0d | 1167 | hlist_for_each_entry(inode, b, i_hash) { |
67a23c49 DC |
1168 | if (inode->i_ino == ino && inode->i_sb == sb) { |
1169 | spin_unlock(&inode_hash_lock); | |
ad5e195a | 1170 | return 0; |
67a23c49 | 1171 | } |
ad5e195a | 1172 | } |
67a23c49 | 1173 | spin_unlock(&inode_hash_lock); |
ad5e195a CH |
1174 | |
1175 | return 1; | |
1176 | } | |
1177 | ||
1da177e4 LT |
1178 | /** |
1179 | * iunique - get a unique inode number | |
1180 | * @sb: superblock | |
1181 | * @max_reserved: highest reserved inode number | |
1182 | * | |
1183 | * Obtain an inode number that is unique on the system for a given | |
1184 | * superblock. This is used by file systems that have no natural | |
1185 | * permanent inode numbering system. An inode number is returned that | |
1186 | * is higher than the reserved limit but unique. | |
1187 | * | |
1188 | * BUGS: | |
1189 | * With a large number of inodes live on the file system this function | |
1190 | * currently becomes quite slow. | |
1191 | */ | |
1192 | ino_t iunique(struct super_block *sb, ino_t max_reserved) | |
1193 | { | |
866b04fc JL |
1194 | /* |
1195 | * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW | |
1196 | * error if st_ino won't fit in target struct field. Use 32bit counter | |
1197 | * here to attempt to avoid that. | |
1198 | */ | |
ad5e195a | 1199 | static DEFINE_SPINLOCK(iunique_lock); |
866b04fc | 1200 | static unsigned int counter; |
1da177e4 | 1201 | ino_t res; |
3361c7be | 1202 | |
ad5e195a | 1203 | spin_lock(&iunique_lock); |
3361c7be JL |
1204 | do { |
1205 | if (counter <= max_reserved) | |
1206 | counter = max_reserved + 1; | |
1da177e4 | 1207 | res = counter++; |
ad5e195a CH |
1208 | } while (!test_inode_iunique(sb, res)); |
1209 | spin_unlock(&iunique_lock); | |
1da177e4 | 1210 | |
3361c7be JL |
1211 | return res; |
1212 | } | |
1da177e4 LT |
1213 | EXPORT_SYMBOL(iunique); |
1214 | ||
1215 | struct inode *igrab(struct inode *inode) | |
1216 | { | |
250df6ed DC |
1217 | spin_lock(&inode->i_lock); |
1218 | if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) { | |
1da177e4 | 1219 | __iget(inode); |
250df6ed DC |
1220 | spin_unlock(&inode->i_lock); |
1221 | } else { | |
1222 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1223 | /* |
1224 | * Handle the case where s_op->clear_inode is not been | |
1225 | * called yet, and somebody is calling igrab | |
1226 | * while the inode is getting freed. | |
1227 | */ | |
1228 | inode = NULL; | |
250df6ed | 1229 | } |
1da177e4 LT |
1230 | return inode; |
1231 | } | |
1da177e4 LT |
1232 | EXPORT_SYMBOL(igrab); |
1233 | ||
1234 | /** | |
0b2d0724 | 1235 | * ilookup5_nowait - search for an inode in the inode cache |
1da177e4 | 1236 | * @sb: super block of file system to search |
0b2d0724 | 1237 | * @hashval: hash value (usually inode number) to search for |
1da177e4 LT |
1238 | * @test: callback used for comparisons between inodes |
1239 | * @data: opaque data pointer to pass to @test | |
1da177e4 | 1240 | * |
0b2d0724 | 1241 | * Search for the inode specified by @hashval and @data in the inode cache. |
1da177e4 LT |
1242 | * If the inode is in the cache, the inode is returned with an incremented |
1243 | * reference count. | |
1244 | * | |
0b2d0724 CH |
1245 | * Note: I_NEW is not waited upon so you have to be very careful what you do |
1246 | * with the returned inode. You probably should be using ilookup5() instead. | |
1da177e4 | 1247 | * |
b6d0ad68 | 1248 | * Note2: @test is called with the inode_hash_lock held, so can't sleep. |
1da177e4 | 1249 | */ |
0b2d0724 CH |
1250 | struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval, |
1251 | int (*test)(struct inode *, void *), void *data) | |
1da177e4 | 1252 | { |
0b2d0724 | 1253 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); |
1da177e4 LT |
1254 | struct inode *inode; |
1255 | ||
67a23c49 | 1256 | spin_lock(&inode_hash_lock); |
1da177e4 | 1257 | inode = find_inode(sb, head, test, data); |
67a23c49 | 1258 | spin_unlock(&inode_hash_lock); |
88bd5121 | 1259 | |
0b2d0724 | 1260 | return inode; |
88bd5121 | 1261 | } |
88bd5121 AA |
1262 | EXPORT_SYMBOL(ilookup5_nowait); |
1263 | ||
1264 | /** | |
1265 | * ilookup5 - search for an inode in the inode cache | |
1266 | * @sb: super block of file system to search | |
1267 | * @hashval: hash value (usually inode number) to search for | |
1268 | * @test: callback used for comparisons between inodes | |
1269 | * @data: opaque data pointer to pass to @test | |
1270 | * | |
0b2d0724 CH |
1271 | * Search for the inode specified by @hashval and @data in the inode cache, |
1272 | * and if the inode is in the cache, return the inode with an incremented | |
1273 | * reference count. Waits on I_NEW before returning the inode. | |
88bd5121 | 1274 | * returned with an incremented reference count. |
1da177e4 | 1275 | * |
0b2d0724 CH |
1276 | * This is a generalized version of ilookup() for file systems where the |
1277 | * inode number is not sufficient for unique identification of an inode. | |
1da177e4 | 1278 | * |
0b2d0724 | 1279 | * Note: @test is called with the inode_hash_lock held, so can't sleep. |
1da177e4 LT |
1280 | */ |
1281 | struct inode *ilookup5(struct super_block *sb, unsigned long hashval, | |
1282 | int (*test)(struct inode *, void *), void *data) | |
1283 | { | |
2864f301 AV |
1284 | struct inode *inode; |
1285 | again: | |
1286 | inode = ilookup5_nowait(sb, hashval, test, data); | |
1287 | if (inode) { | |
0b2d0724 | 1288 | wait_on_inode(inode); |
2864f301 AV |
1289 | if (unlikely(inode_unhashed(inode))) { |
1290 | iput(inode); | |
1291 | goto again; | |
1292 | } | |
1293 | } | |
0b2d0724 | 1294 | return inode; |
1da177e4 | 1295 | } |
1da177e4 LT |
1296 | EXPORT_SYMBOL(ilookup5); |
1297 | ||
1298 | /** | |
1299 | * ilookup - search for an inode in the inode cache | |
1300 | * @sb: super block of file system to search | |
1301 | * @ino: inode number to search for | |
1302 | * | |
0b2d0724 CH |
1303 | * Search for the inode @ino in the inode cache, and if the inode is in the |
1304 | * cache, the inode is returned with an incremented reference count. | |
1da177e4 LT |
1305 | */ |
1306 | struct inode *ilookup(struct super_block *sb, unsigned long ino) | |
1307 | { | |
1308 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
1da177e4 | 1309 | struct inode *inode; |
2864f301 | 1310 | again: |
0b2d0724 CH |
1311 | spin_lock(&inode_hash_lock); |
1312 | inode = find_inode_fast(sb, head, ino); | |
1313 | spin_unlock(&inode_hash_lock); | |
1da177e4 | 1314 | |
2864f301 | 1315 | if (inode) { |
0b2d0724 | 1316 | wait_on_inode(inode); |
2864f301 AV |
1317 | if (unlikely(inode_unhashed(inode))) { |
1318 | iput(inode); | |
1319 | goto again; | |
1320 | } | |
1321 | } | |
0b2d0724 | 1322 | return inode; |
1da177e4 | 1323 | } |
0b2d0724 | 1324 | EXPORT_SYMBOL(ilookup); |
1da177e4 | 1325 | |
fe032c42 TT |
1326 | /** |
1327 | * find_inode_nowait - find an inode in the inode cache | |
1328 | * @sb: super block of file system to search | |
1329 | * @hashval: hash value (usually inode number) to search for | |
1330 | * @match: callback used for comparisons between inodes | |
1331 | * @data: opaque data pointer to pass to @match | |
1332 | * | |
1333 | * Search for the inode specified by @hashval and @data in the inode | |
1334 | * cache, where the helper function @match will return 0 if the inode | |
1335 | * does not match, 1 if the inode does match, and -1 if the search | |
1336 | * should be stopped. The @match function must be responsible for | |
1337 | * taking the i_lock spin_lock and checking i_state for an inode being | |
1338 | * freed or being initialized, and incrementing the reference count | |
1339 | * before returning 1. It also must not sleep, since it is called with | |
1340 | * the inode_hash_lock spinlock held. | |
1341 | * | |
1342 | * This is a even more generalized version of ilookup5() when the | |
1343 | * function must never block --- find_inode() can block in | |
1344 | * __wait_on_freeing_inode() --- or when the caller can not increment | |
1345 | * the reference count because the resulting iput() might cause an | |
1346 | * inode eviction. The tradeoff is that the @match funtion must be | |
1347 | * very carefully implemented. | |
1348 | */ | |
1349 | struct inode *find_inode_nowait(struct super_block *sb, | |
1350 | unsigned long hashval, | |
1351 | int (*match)(struct inode *, unsigned long, | |
1352 | void *), | |
1353 | void *data) | |
1354 | { | |
1355 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
1356 | struct inode *inode, *ret_inode = NULL; | |
1357 | int mval; | |
1358 | ||
1359 | spin_lock(&inode_hash_lock); | |
1360 | hlist_for_each_entry(inode, head, i_hash) { | |
1361 | if (inode->i_sb != sb) | |
1362 | continue; | |
1363 | mval = match(inode, hashval, data); | |
1364 | if (mval == 0) | |
1365 | continue; | |
1366 | if (mval == 1) | |
1367 | ret_inode = inode; | |
1368 | goto out; | |
1369 | } | |
1370 | out: | |
1371 | spin_unlock(&inode_hash_lock); | |
1372 | return ret_inode; | |
1373 | } | |
1374 | EXPORT_SYMBOL(find_inode_nowait); | |
1375 | ||
261bca86 AV |
1376 | int insert_inode_locked(struct inode *inode) |
1377 | { | |
1378 | struct super_block *sb = inode->i_sb; | |
1379 | ino_t ino = inode->i_ino; | |
1380 | struct hlist_head *head = inode_hashtable + hash(sb, ino); | |
261bca86 | 1381 | |
261bca86 | 1382 | while (1) { |
72a43d63 | 1383 | struct inode *old = NULL; |
67a23c49 | 1384 | spin_lock(&inode_hash_lock); |
b67bfe0d | 1385 | hlist_for_each_entry(old, head, i_hash) { |
72a43d63 AV |
1386 | if (old->i_ino != ino) |
1387 | continue; | |
1388 | if (old->i_sb != sb) | |
1389 | continue; | |
250df6ed DC |
1390 | spin_lock(&old->i_lock); |
1391 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1392 | spin_unlock(&old->i_lock); | |
72a43d63 | 1393 | continue; |
250df6ed | 1394 | } |
72a43d63 AV |
1395 | break; |
1396 | } | |
b67bfe0d | 1397 | if (likely(!old)) { |
250df6ed DC |
1398 | spin_lock(&inode->i_lock); |
1399 | inode->i_state |= I_NEW; | |
261bca86 | 1400 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1401 | spin_unlock(&inode->i_lock); |
67a23c49 | 1402 | spin_unlock(&inode_hash_lock); |
261bca86 AV |
1403 | return 0; |
1404 | } | |
1405 | __iget(old); | |
250df6ed | 1406 | spin_unlock(&old->i_lock); |
67a23c49 | 1407 | spin_unlock(&inode_hash_lock); |
261bca86 | 1408 | wait_on_inode(old); |
1d3382cb | 1409 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1410 | iput(old); |
1411 | return -EBUSY; | |
1412 | } | |
1413 | iput(old); | |
1414 | } | |
1415 | } | |
261bca86 AV |
1416 | EXPORT_SYMBOL(insert_inode_locked); |
1417 | ||
1418 | int insert_inode_locked4(struct inode *inode, unsigned long hashval, | |
1419 | int (*test)(struct inode *, void *), void *data) | |
1420 | { | |
1421 | struct super_block *sb = inode->i_sb; | |
1422 | struct hlist_head *head = inode_hashtable + hash(sb, hashval); | |
261bca86 | 1423 | |
261bca86 | 1424 | while (1) { |
72a43d63 AV |
1425 | struct inode *old = NULL; |
1426 | ||
67a23c49 | 1427 | spin_lock(&inode_hash_lock); |
b67bfe0d | 1428 | hlist_for_each_entry(old, head, i_hash) { |
72a43d63 AV |
1429 | if (old->i_sb != sb) |
1430 | continue; | |
1431 | if (!test(old, data)) | |
1432 | continue; | |
250df6ed DC |
1433 | spin_lock(&old->i_lock); |
1434 | if (old->i_state & (I_FREEING|I_WILL_FREE)) { | |
1435 | spin_unlock(&old->i_lock); | |
72a43d63 | 1436 | continue; |
250df6ed | 1437 | } |
72a43d63 AV |
1438 | break; |
1439 | } | |
b67bfe0d | 1440 | if (likely(!old)) { |
250df6ed DC |
1441 | spin_lock(&inode->i_lock); |
1442 | inode->i_state |= I_NEW; | |
261bca86 | 1443 | hlist_add_head(&inode->i_hash, head); |
250df6ed | 1444 | spin_unlock(&inode->i_lock); |
67a23c49 | 1445 | spin_unlock(&inode_hash_lock); |
261bca86 AV |
1446 | return 0; |
1447 | } | |
1448 | __iget(old); | |
250df6ed | 1449 | spin_unlock(&old->i_lock); |
67a23c49 | 1450 | spin_unlock(&inode_hash_lock); |
261bca86 | 1451 | wait_on_inode(old); |
1d3382cb | 1452 | if (unlikely(!inode_unhashed(old))) { |
261bca86 AV |
1453 | iput(old); |
1454 | return -EBUSY; | |
1455 | } | |
1456 | iput(old); | |
1457 | } | |
1458 | } | |
261bca86 AV |
1459 | EXPORT_SYMBOL(insert_inode_locked4); |
1460 | ||
1da177e4 | 1461 | |
45321ac5 AV |
1462 | int generic_delete_inode(struct inode *inode) |
1463 | { | |
1464 | return 1; | |
1465 | } | |
1466 | EXPORT_SYMBOL(generic_delete_inode); | |
1467 | ||
45321ac5 AV |
1468 | /* |
1469 | * Called when we're dropping the last reference | |
1470 | * to an inode. | |
22fe4042 | 1471 | * |
45321ac5 AV |
1472 | * Call the FS "drop_inode()" function, defaulting to |
1473 | * the legacy UNIX filesystem behaviour. If it tells | |
1474 | * us to evict inode, do so. Otherwise, retain inode | |
1475 | * in cache if fs is alive, sync and evict if fs is | |
1476 | * shutting down. | |
22fe4042 | 1477 | */ |
45321ac5 | 1478 | static void iput_final(struct inode *inode) |
1da177e4 LT |
1479 | { |
1480 | struct super_block *sb = inode->i_sb; | |
45321ac5 AV |
1481 | const struct super_operations *op = inode->i_sb->s_op; |
1482 | int drop; | |
1483 | ||
250df6ed DC |
1484 | WARN_ON(inode->i_state & I_NEW); |
1485 | ||
e7f59097 | 1486 | if (op->drop_inode) |
45321ac5 AV |
1487 | drop = op->drop_inode(inode); |
1488 | else | |
1489 | drop = generic_drop_inode(inode); | |
1da177e4 | 1490 | |
b2b2af8e DC |
1491 | if (!drop && (sb->s_flags & MS_ACTIVE)) { |
1492 | inode->i_state |= I_REFERENCED; | |
4eff96dd | 1493 | inode_add_lru(inode); |
b2b2af8e | 1494 | spin_unlock(&inode->i_lock); |
b2b2af8e DC |
1495 | return; |
1496 | } | |
1497 | ||
45321ac5 | 1498 | if (!drop) { |
991114c6 | 1499 | inode->i_state |= I_WILL_FREE; |
250df6ed | 1500 | spin_unlock(&inode->i_lock); |
1da177e4 | 1501 | write_inode_now(inode, 1); |
250df6ed | 1502 | spin_lock(&inode->i_lock); |
7ef0d737 | 1503 | WARN_ON(inode->i_state & I_NEW); |
991114c6 | 1504 | inode->i_state &= ~I_WILL_FREE; |
1da177e4 | 1505 | } |
7ccf19a8 | 1506 | |
991114c6 | 1507 | inode->i_state |= I_FREEING; |
c4ae0c65 ED |
1508 | if (!list_empty(&inode->i_lru)) |
1509 | inode_lru_list_del(inode); | |
b2b2af8e | 1510 | spin_unlock(&inode->i_lock); |
b2b2af8e | 1511 | |
644da596 | 1512 | evict(inode); |
1da177e4 LT |
1513 | } |
1514 | ||
1da177e4 | 1515 | /** |
6b3304b5 | 1516 | * iput - put an inode |
1da177e4 LT |
1517 | * @inode: inode to put |
1518 | * | |
1519 | * Puts an inode, dropping its usage count. If the inode use count hits | |
1520 | * zero, the inode is then freed and may also be destroyed. | |
1521 | * | |
1522 | * Consequently, iput() can sleep. | |
1523 | */ | |
1524 | void iput(struct inode *inode) | |
1525 | { | |
0ae45f63 TT |
1526 | if (!inode) |
1527 | return; | |
1528 | BUG_ON(inode->i_state & I_CLEAR); | |
1529 | retry: | |
1530 | if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) { | |
1531 | if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) { | |
1532 | atomic_inc(&inode->i_count); | |
1533 | inode->i_state &= ~I_DIRTY_TIME; | |
1534 | spin_unlock(&inode->i_lock); | |
1535 | trace_writeback_lazytime_iput(inode); | |
1536 | mark_inode_dirty_sync(inode); | |
1537 | goto retry; | |
1538 | } | |
1539 | iput_final(inode); | |
1da177e4 LT |
1540 | } |
1541 | } | |
1da177e4 LT |
1542 | EXPORT_SYMBOL(iput); |
1543 | ||
1544 | /** | |
1545 | * bmap - find a block number in a file | |
1546 | * @inode: inode of file | |
1547 | * @block: block to find | |
1548 | * | |
1549 | * Returns the block number on the device holding the inode that | |
1550 | * is the disk block number for the block of the file requested. | |
1551 | * That is, asked for block 4 of inode 1 the function will return the | |
6b3304b5 | 1552 | * disk block relative to the disk start that holds that block of the |
1da177e4 LT |
1553 | * file. |
1554 | */ | |
6b3304b5 | 1555 | sector_t bmap(struct inode *inode, sector_t block) |
1da177e4 LT |
1556 | { |
1557 | sector_t res = 0; | |
1558 | if (inode->i_mapping->a_ops->bmap) | |
1559 | res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block); | |
1560 | return res; | |
1561 | } | |
1da177e4 LT |
1562 | EXPORT_SYMBOL(bmap); |
1563 | ||
598e3c8f MS |
1564 | /* |
1565 | * Update times in overlayed inode from underlying real inode | |
1566 | */ | |
1567 | static void update_ovl_inode_times(struct dentry *dentry, struct inode *inode, | |
1568 | bool rcu) | |
1569 | { | |
1570 | if (!rcu) { | |
1571 | struct inode *realinode = d_real_inode(dentry); | |
1572 | ||
1573 | if (unlikely(inode != realinode) && | |
1574 | (!timespec_equal(&inode->i_mtime, &realinode->i_mtime) || | |
1575 | !timespec_equal(&inode->i_ctime, &realinode->i_ctime))) { | |
1576 | inode->i_mtime = realinode->i_mtime; | |
1577 | inode->i_ctime = realinode->i_ctime; | |
1578 | } | |
1579 | } | |
1580 | } | |
1581 | ||
11ff6f05 MG |
1582 | /* |
1583 | * With relative atime, only update atime if the previous atime is | |
1584 | * earlier than either the ctime or mtime or if at least a day has | |
1585 | * passed since the last atime update. | |
1586 | */ | |
598e3c8f MS |
1587 | static int relatime_need_update(const struct path *path, struct inode *inode, |
1588 | struct timespec now, bool rcu) | |
11ff6f05 MG |
1589 | { |
1590 | ||
598e3c8f | 1591 | if (!(path->mnt->mnt_flags & MNT_RELATIME)) |
11ff6f05 | 1592 | return 1; |
598e3c8f MS |
1593 | |
1594 | update_ovl_inode_times(path->dentry, inode, rcu); | |
11ff6f05 MG |
1595 | /* |
1596 | * Is mtime younger than atime? If yes, update atime: | |
1597 | */ | |
1598 | if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0) | |
1599 | return 1; | |
1600 | /* | |
1601 | * Is ctime younger than atime? If yes, update atime: | |
1602 | */ | |
1603 | if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0) | |
1604 | return 1; | |
1605 | ||
1606 | /* | |
1607 | * Is the previous atime value older than a day? If yes, | |
1608 | * update atime: | |
1609 | */ | |
1610 | if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60) | |
1611 | return 1; | |
1612 | /* | |
1613 | * Good, we can skip the atime update: | |
1614 | */ | |
1615 | return 0; | |
1616 | } | |
1617 | ||
0ae45f63 | 1618 | int generic_update_time(struct inode *inode, struct timespec *time, int flags) |
c3b2da31 | 1619 | { |
0ae45f63 | 1620 | int iflags = I_DIRTY_TIME; |
c3b2da31 JB |
1621 | |
1622 | if (flags & S_ATIME) | |
1623 | inode->i_atime = *time; | |
1624 | if (flags & S_VERSION) | |
1625 | inode_inc_iversion(inode); | |
1626 | if (flags & S_CTIME) | |
1627 | inode->i_ctime = *time; | |
1628 | if (flags & S_MTIME) | |
1629 | inode->i_mtime = *time; | |
0ae45f63 TT |
1630 | |
1631 | if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION)) | |
1632 | iflags |= I_DIRTY_SYNC; | |
1633 | __mark_inode_dirty(inode, iflags); | |
c3b2da31 JB |
1634 | return 0; |
1635 | } | |
0ae45f63 TT |
1636 | EXPORT_SYMBOL(generic_update_time); |
1637 | ||
1638 | /* | |
1639 | * This does the actual work of updating an inodes time or version. Must have | |
1640 | * had called mnt_want_write() before calling this. | |
1641 | */ | |
1642 | static int update_time(struct inode *inode, struct timespec *time, int flags) | |
1643 | { | |
1644 | int (*update_time)(struct inode *, struct timespec *, int); | |
1645 | ||
1646 | update_time = inode->i_op->update_time ? inode->i_op->update_time : | |
1647 | generic_update_time; | |
1648 | ||
1649 | return update_time(inode, time, flags); | |
1650 | } | |
c3b2da31 | 1651 | |
1da177e4 | 1652 | /** |
869243a0 | 1653 | * touch_atime - update the access time |
185553b2 | 1654 | * @path: the &struct path to update |
30fdc8ee | 1655 | * @inode: inode to update |
1da177e4 LT |
1656 | * |
1657 | * Update the accessed time on an inode and mark it for writeback. | |
1658 | * This function automatically handles read only file systems and media, | |
1659 | * as well as the "noatime" flag and inode specific "noatime" markers. | |
1660 | */ | |
598e3c8f MS |
1661 | bool __atime_needs_update(const struct path *path, struct inode *inode, |
1662 | bool rcu) | |
1da177e4 | 1663 | { |
68ac1234 | 1664 | struct vfsmount *mnt = path->mnt; |
1da177e4 LT |
1665 | struct timespec now; |
1666 | ||
cdb70f3f | 1667 | if (inode->i_flags & S_NOATIME) |
8fa9dd24 | 1668 | return false; |
0bd23d09 EB |
1669 | |
1670 | /* Atime updates will likely cause i_uid and i_gid to be written | |
1671 | * back improprely if their true value is unknown to the vfs. | |
1672 | */ | |
1673 | if (HAS_UNMAPPED_ID(inode)) | |
1674 | return false; | |
1675 | ||
37756ced | 1676 | if (IS_NOATIME(inode)) |
8fa9dd24 | 1677 | return false; |
b2276138 | 1678 | if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)) |
8fa9dd24 | 1679 | return false; |
47ae32d6 | 1680 | |
cdb70f3f | 1681 | if (mnt->mnt_flags & MNT_NOATIME) |
8fa9dd24 | 1682 | return false; |
cdb70f3f | 1683 | if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)) |
8fa9dd24 | 1684 | return false; |
1da177e4 | 1685 | |
c2050a45 | 1686 | now = current_time(inode); |
11ff6f05 | 1687 | |
598e3c8f | 1688 | if (!relatime_need_update(path, inode, now, rcu)) |
8fa9dd24 | 1689 | return false; |
11ff6f05 | 1690 | |
47ae32d6 | 1691 | if (timespec_equal(&inode->i_atime, &now)) |
8fa9dd24 N |
1692 | return false; |
1693 | ||
1694 | return true; | |
1695 | } | |
1696 | ||
1697 | void touch_atime(const struct path *path) | |
1698 | { | |
1699 | struct vfsmount *mnt = path->mnt; | |
1700 | struct inode *inode = d_inode(path->dentry); | |
1701 | struct timespec now; | |
1702 | ||
598e3c8f | 1703 | if (!__atime_needs_update(path, inode, false)) |
b12536c2 AK |
1704 | return; |
1705 | ||
5d37e9e6 | 1706 | if (!sb_start_write_trylock(inode->i_sb)) |
b12536c2 | 1707 | return; |
47ae32d6 | 1708 | |
8fa9dd24 | 1709 | if (__mnt_want_write(mnt) != 0) |
5d37e9e6 | 1710 | goto skip_update; |
c3b2da31 JB |
1711 | /* |
1712 | * File systems can error out when updating inodes if they need to | |
1713 | * allocate new space to modify an inode (such is the case for | |
1714 | * Btrfs), but since we touch atime while walking down the path we | |
1715 | * really don't care if we failed to update the atime of the file, | |
1716 | * so just ignore the return value. | |
2bc55652 AB |
1717 | * We may also fail on filesystems that have the ability to make parts |
1718 | * of the fs read only, e.g. subvolumes in Btrfs. | |
c3b2da31 | 1719 | */ |
c2050a45 | 1720 | now = current_time(inode); |
c3b2da31 | 1721 | update_time(inode, &now, S_ATIME); |
5d37e9e6 JK |
1722 | __mnt_drop_write(mnt); |
1723 | skip_update: | |
1724 | sb_end_write(inode->i_sb); | |
1da177e4 | 1725 | } |
869243a0 | 1726 | EXPORT_SYMBOL(touch_atime); |
1da177e4 | 1727 | |
3ed37648 CW |
1728 | /* |
1729 | * The logic we want is | |
1730 | * | |
1731 | * if suid or (sgid and xgrp) | |
1732 | * remove privs | |
1733 | */ | |
1734 | int should_remove_suid(struct dentry *dentry) | |
1735 | { | |
df2b1afd | 1736 | umode_t mode = d_inode(dentry)->i_mode; |
3ed37648 CW |
1737 | int kill = 0; |
1738 | ||
1739 | /* suid always must be killed */ | |
1740 | if (unlikely(mode & S_ISUID)) | |
1741 | kill = ATTR_KILL_SUID; | |
1742 | ||
1743 | /* | |
1744 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1745 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1746 | */ | |
1747 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1748 | kill |= ATTR_KILL_SGID; | |
1749 | ||
1750 | if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) | |
1751 | return kill; | |
1752 | ||
1753 | return 0; | |
1754 | } | |
1755 | EXPORT_SYMBOL(should_remove_suid); | |
1756 | ||
dbfae0cd JK |
1757 | /* |
1758 | * Return mask of changes for notify_change() that need to be done as a | |
1759 | * response to write or truncate. Return 0 if nothing has to be changed. | |
1760 | * Negative value on error (change should be denied). | |
1761 | */ | |
45f147a1 | 1762 | int dentry_needs_remove_privs(struct dentry *dentry) |
dbfae0cd | 1763 | { |
dbfae0cd JK |
1764 | struct inode *inode = d_inode(dentry); |
1765 | int mask = 0; | |
1766 | int ret; | |
1767 | ||
1768 | if (IS_NOSEC(inode)) | |
1769 | return 0; | |
1770 | ||
1771 | mask = should_remove_suid(dentry); | |
1772 | ret = security_inode_need_killpriv(dentry); | |
1773 | if (ret < 0) | |
1774 | return ret; | |
1775 | if (ret) | |
1776 | mask |= ATTR_KILL_PRIV; | |
1777 | return mask; | |
1778 | } | |
dbfae0cd JK |
1779 | |
1780 | static int __remove_privs(struct dentry *dentry, int kill) | |
3ed37648 CW |
1781 | { |
1782 | struct iattr newattrs; | |
1783 | ||
1784 | newattrs.ia_valid = ATTR_FORCE | kill; | |
27ac0ffe BF |
1785 | /* |
1786 | * Note we call this on write, so notify_change will not | |
1787 | * encounter any conflicting delegations: | |
1788 | */ | |
1789 | return notify_change(dentry, &newattrs, NULL); | |
3ed37648 CW |
1790 | } |
1791 | ||
5fa8e0a1 JK |
1792 | /* |
1793 | * Remove special file priviledges (suid, capabilities) when file is written | |
1794 | * to or truncated. | |
1795 | */ | |
1796 | int file_remove_privs(struct file *file) | |
3ed37648 | 1797 | { |
c1892c37 MS |
1798 | struct dentry *dentry = file_dentry(file); |
1799 | struct inode *inode = file_inode(file); | |
dbfae0cd | 1800 | int kill; |
3ed37648 CW |
1801 | int error = 0; |
1802 | ||
1803 | /* Fast path for nothing security related */ | |
1804 | if (IS_NOSEC(inode)) | |
1805 | return 0; | |
1806 | ||
c1892c37 | 1807 | kill = dentry_needs_remove_privs(dentry); |
dbfae0cd JK |
1808 | if (kill < 0) |
1809 | return kill; | |
1810 | if (kill) | |
1811 | error = __remove_privs(dentry, kill); | |
2426f391 JK |
1812 | if (!error) |
1813 | inode_has_no_xattr(inode); | |
3ed37648 CW |
1814 | |
1815 | return error; | |
1816 | } | |
5fa8e0a1 | 1817 | EXPORT_SYMBOL(file_remove_privs); |
3ed37648 | 1818 | |
1da177e4 | 1819 | /** |
870f4817 CH |
1820 | * file_update_time - update mtime and ctime time |
1821 | * @file: file accessed | |
1da177e4 | 1822 | * |
870f4817 CH |
1823 | * Update the mtime and ctime members of an inode and mark the inode |
1824 | * for writeback. Note that this function is meant exclusively for | |
1825 | * usage in the file write path of filesystems, and filesystems may | |
1826 | * choose to explicitly ignore update via this function with the | |
2eadfc0e | 1827 | * S_NOCMTIME inode flag, e.g. for network filesystem where these |
c3b2da31 JB |
1828 | * timestamps are handled by the server. This can return an error for |
1829 | * file systems who need to allocate space in order to update an inode. | |
1da177e4 LT |
1830 | */ |
1831 | ||
c3b2da31 | 1832 | int file_update_time(struct file *file) |
1da177e4 | 1833 | { |
496ad9aa | 1834 | struct inode *inode = file_inode(file); |
1da177e4 | 1835 | struct timespec now; |
c3b2da31 JB |
1836 | int sync_it = 0; |
1837 | int ret; | |
1da177e4 | 1838 | |
ce06e0b2 | 1839 | /* First try to exhaust all avenues to not sync */ |
1da177e4 | 1840 | if (IS_NOCMTIME(inode)) |
c3b2da31 | 1841 | return 0; |
20ddee2c | 1842 | |
c2050a45 | 1843 | now = current_time(inode); |
ce06e0b2 AK |
1844 | if (!timespec_equal(&inode->i_mtime, &now)) |
1845 | sync_it = S_MTIME; | |
1da177e4 | 1846 | |
ce06e0b2 AK |
1847 | if (!timespec_equal(&inode->i_ctime, &now)) |
1848 | sync_it |= S_CTIME; | |
870f4817 | 1849 | |
ce06e0b2 AK |
1850 | if (IS_I_VERSION(inode)) |
1851 | sync_it |= S_VERSION; | |
7a224228 | 1852 | |
ce06e0b2 | 1853 | if (!sync_it) |
c3b2da31 | 1854 | return 0; |
ce06e0b2 AK |
1855 | |
1856 | /* Finally allowed to write? Takes lock. */ | |
eb04c282 | 1857 | if (__mnt_want_write_file(file)) |
c3b2da31 | 1858 | return 0; |
ce06e0b2 | 1859 | |
c3b2da31 | 1860 | ret = update_time(inode, &now, sync_it); |
eb04c282 | 1861 | __mnt_drop_write_file(file); |
c3b2da31 JB |
1862 | |
1863 | return ret; | |
1da177e4 | 1864 | } |
870f4817 | 1865 | EXPORT_SYMBOL(file_update_time); |
1da177e4 LT |
1866 | |
1867 | int inode_needs_sync(struct inode *inode) | |
1868 | { | |
1869 | if (IS_SYNC(inode)) | |
1870 | return 1; | |
1871 | if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode)) | |
1872 | return 1; | |
1873 | return 0; | |
1874 | } | |
1da177e4 LT |
1875 | EXPORT_SYMBOL(inode_needs_sync); |
1876 | ||
1da177e4 | 1877 | /* |
168a9fd6 MS |
1878 | * If we try to find an inode in the inode hash while it is being |
1879 | * deleted, we have to wait until the filesystem completes its | |
1880 | * deletion before reporting that it isn't found. This function waits | |
1881 | * until the deletion _might_ have completed. Callers are responsible | |
1882 | * to recheck inode state. | |
1883 | * | |
eaff8079 | 1884 | * It doesn't matter if I_NEW is not set initially, a call to |
250df6ed DC |
1885 | * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list |
1886 | * will DTRT. | |
1da177e4 LT |
1887 | */ |
1888 | static void __wait_on_freeing_inode(struct inode *inode) | |
1889 | { | |
1890 | wait_queue_head_t *wq; | |
eaff8079 CH |
1891 | DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW); |
1892 | wq = bit_waitqueue(&inode->i_state, __I_NEW); | |
1da177e4 | 1893 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
250df6ed | 1894 | spin_unlock(&inode->i_lock); |
67a23c49 | 1895 | spin_unlock(&inode_hash_lock); |
1da177e4 LT |
1896 | schedule(); |
1897 | finish_wait(wq, &wait.wait); | |
67a23c49 | 1898 | spin_lock(&inode_hash_lock); |
1da177e4 LT |
1899 | } |
1900 | ||
1da177e4 LT |
1901 | static __initdata unsigned long ihash_entries; |
1902 | static int __init set_ihash_entries(char *str) | |
1903 | { | |
1904 | if (!str) | |
1905 | return 0; | |
1906 | ihash_entries = simple_strtoul(str, &str, 0); | |
1907 | return 1; | |
1908 | } | |
1909 | __setup("ihash_entries=", set_ihash_entries); | |
1910 | ||
1911 | /* | |
1912 | * Initialize the waitqueues and inode hash table. | |
1913 | */ | |
1914 | void __init inode_init_early(void) | |
1915 | { | |
074b8517 | 1916 | unsigned int loop; |
1da177e4 LT |
1917 | |
1918 | /* If hashes are distributed across NUMA nodes, defer | |
1919 | * hash allocation until vmalloc space is available. | |
1920 | */ | |
1921 | if (hashdist) | |
1922 | return; | |
1923 | ||
1924 | inode_hashtable = | |
1925 | alloc_large_system_hash("Inode-cache", | |
1926 | sizeof(struct hlist_head), | |
1927 | ihash_entries, | |
1928 | 14, | |
1929 | HASH_EARLY, | |
1930 | &i_hash_shift, | |
1931 | &i_hash_mask, | |
31fe62b9 | 1932 | 0, |
1da177e4 LT |
1933 | 0); |
1934 | ||
074b8517 | 1935 | for (loop = 0; loop < (1U << i_hash_shift); loop++) |
1da177e4 LT |
1936 | INIT_HLIST_HEAD(&inode_hashtable[loop]); |
1937 | } | |
1938 | ||
74bf17cf | 1939 | void __init inode_init(void) |
1da177e4 | 1940 | { |
074b8517 | 1941 | unsigned int loop; |
1da177e4 LT |
1942 | |
1943 | /* inode slab cache */ | |
b0196009 PJ |
1944 | inode_cachep = kmem_cache_create("inode_cache", |
1945 | sizeof(struct inode), | |
1946 | 0, | |
1947 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
5d097056 | 1948 | SLAB_MEM_SPREAD|SLAB_ACCOUNT), |
20c2df83 | 1949 | init_once); |
1da177e4 LT |
1950 | |
1951 | /* Hash may have been set up in inode_init_early */ | |
1952 | if (!hashdist) | |
1953 | return; | |
1954 | ||
1955 | inode_hashtable = | |
1956 | alloc_large_system_hash("Inode-cache", | |
1957 | sizeof(struct hlist_head), | |
1958 | ihash_entries, | |
1959 | 14, | |
1960 | 0, | |
1961 | &i_hash_shift, | |
1962 | &i_hash_mask, | |
31fe62b9 | 1963 | 0, |
1da177e4 LT |
1964 | 0); |
1965 | ||
074b8517 | 1966 | for (loop = 0; loop < (1U << i_hash_shift); loop++) |
1da177e4 LT |
1967 | INIT_HLIST_HEAD(&inode_hashtable[loop]); |
1968 | } | |
1969 | ||
1970 | void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev) | |
1971 | { | |
1972 | inode->i_mode = mode; | |
1973 | if (S_ISCHR(mode)) { | |
1974 | inode->i_fop = &def_chr_fops; | |
1975 | inode->i_rdev = rdev; | |
1976 | } else if (S_ISBLK(mode)) { | |
1977 | inode->i_fop = &def_blk_fops; | |
1978 | inode->i_rdev = rdev; | |
1979 | } else if (S_ISFIFO(mode)) | |
599a0ac1 | 1980 | inode->i_fop = &pipefifo_fops; |
1da177e4 | 1981 | else if (S_ISSOCK(mode)) |
bd9b51e7 | 1982 | ; /* leave it no_open_fops */ |
1da177e4 | 1983 | else |
af0d9ae8 MK |
1984 | printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for" |
1985 | " inode %s:%lu\n", mode, inode->i_sb->s_id, | |
1986 | inode->i_ino); | |
1da177e4 LT |
1987 | } |
1988 | EXPORT_SYMBOL(init_special_inode); | |
a1bd120d DM |
1989 | |
1990 | /** | |
eaae668d | 1991 | * inode_init_owner - Init uid,gid,mode for new inode according to posix standards |
a1bd120d DM |
1992 | * @inode: New inode |
1993 | * @dir: Directory inode | |
1994 | * @mode: mode of the new inode | |
1995 | */ | |
1996 | void inode_init_owner(struct inode *inode, const struct inode *dir, | |
62bb1091 | 1997 | umode_t mode) |
a1bd120d DM |
1998 | { |
1999 | inode->i_uid = current_fsuid(); | |
2000 | if (dir && dir->i_mode & S_ISGID) { | |
2001 | inode->i_gid = dir->i_gid; | |
2002 | if (S_ISDIR(mode)) | |
2003 | mode |= S_ISGID; | |
2004 | } else | |
2005 | inode->i_gid = current_fsgid(); | |
2006 | inode->i_mode = mode; | |
2007 | } | |
2008 | EXPORT_SYMBOL(inode_init_owner); | |
e795b717 | 2009 | |
2e149670 SH |
2010 | /** |
2011 | * inode_owner_or_capable - check current task permissions to inode | |
2012 | * @inode: inode being checked | |
2013 | * | |
23adbe12 AL |
2014 | * Return true if current either has CAP_FOWNER in a namespace with the |
2015 | * inode owner uid mapped, or owns the file. | |
e795b717 | 2016 | */ |
2e149670 | 2017 | bool inode_owner_or_capable(const struct inode *inode) |
e795b717 | 2018 | { |
23adbe12 AL |
2019 | struct user_namespace *ns; |
2020 | ||
92361636 | 2021 | if (uid_eq(current_fsuid(), inode->i_uid)) |
e795b717 | 2022 | return true; |
23adbe12 AL |
2023 | |
2024 | ns = current_user_ns(); | |
2025 | if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid)) | |
e795b717 SH |
2026 | return true; |
2027 | return false; | |
2028 | } | |
2e149670 | 2029 | EXPORT_SYMBOL(inode_owner_or_capable); |
1d59d61f TM |
2030 | |
2031 | /* | |
2032 | * Direct i/o helper functions | |
2033 | */ | |
2034 | static void __inode_dio_wait(struct inode *inode) | |
2035 | { | |
2036 | wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP); | |
2037 | DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP); | |
2038 | ||
2039 | do { | |
2040 | prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE); | |
2041 | if (atomic_read(&inode->i_dio_count)) | |
2042 | schedule(); | |
2043 | } while (atomic_read(&inode->i_dio_count)); | |
2044 | finish_wait(wq, &q.wait); | |
2045 | } | |
2046 | ||
2047 | /** | |
2048 | * inode_dio_wait - wait for outstanding DIO requests to finish | |
2049 | * @inode: inode to wait for | |
2050 | * | |
2051 | * Waits for all pending direct I/O requests to finish so that we can | |
2052 | * proceed with a truncate or equivalent operation. | |
2053 | * | |
2054 | * Must be called under a lock that serializes taking new references | |
2055 | * to i_dio_count, usually by inode->i_mutex. | |
2056 | */ | |
2057 | void inode_dio_wait(struct inode *inode) | |
2058 | { | |
2059 | if (atomic_read(&inode->i_dio_count)) | |
2060 | __inode_dio_wait(inode); | |
2061 | } | |
2062 | EXPORT_SYMBOL(inode_dio_wait); | |
2063 | ||
5f16f322 TT |
2064 | /* |
2065 | * inode_set_flags - atomically set some inode flags | |
2066 | * | |
2067 | * Note: the caller should be holding i_mutex, or else be sure that | |
2068 | * they have exclusive access to the inode structure (i.e., while the | |
2069 | * inode is being instantiated). The reason for the cmpxchg() loop | |
2070 | * --- which wouldn't be necessary if all code paths which modify | |
2071 | * i_flags actually followed this rule, is that there is at least one | |
5fa8e0a1 JK |
2072 | * code path which doesn't today so we use cmpxchg() out of an abundance |
2073 | * of caution. | |
5f16f322 TT |
2074 | * |
2075 | * In the long run, i_mutex is overkill, and we should probably look | |
2076 | * at using the i_lock spinlock to protect i_flags, and then make sure | |
2077 | * it is so documented in include/linux/fs.h and that all code follows | |
2078 | * the locking convention!! | |
2079 | */ | |
2080 | void inode_set_flags(struct inode *inode, unsigned int flags, | |
2081 | unsigned int mask) | |
2082 | { | |
2083 | unsigned int old_flags, new_flags; | |
2084 | ||
2085 | WARN_ON_ONCE(flags & ~mask); | |
2086 | do { | |
2087 | old_flags = ACCESS_ONCE(inode->i_flags); | |
2088 | new_flags = (old_flags & ~mask) | flags; | |
2089 | } while (unlikely(cmpxchg(&inode->i_flags, old_flags, | |
2090 | new_flags) != old_flags)); | |
2091 | } | |
2092 | EXPORT_SYMBOL(inode_set_flags); | |
21fc61c7 AV |
2093 | |
2094 | void inode_nohighmem(struct inode *inode) | |
2095 | { | |
2096 | mapping_set_gfp_mask(inode->i_mapping, GFP_USER); | |
2097 | } | |
2098 | EXPORT_SYMBOL(inode_nohighmem); | |
3cd88666 DD |
2099 | |
2100 | /** | |
2101 | * current_time - Return FS time | |
2102 | * @inode: inode. | |
2103 | * | |
2104 | * Return the current time truncated to the time granularity supported by | |
2105 | * the fs. | |
2106 | * | |
2107 | * Note that inode and inode->sb cannot be NULL. | |
2108 | * Otherwise, the function warns and returns time without truncation. | |
2109 | */ | |
2110 | struct timespec current_time(struct inode *inode) | |
2111 | { | |
2112 | struct timespec now = current_kernel_time(); | |
2113 | ||
2114 | if (unlikely(!inode->i_sb)) { | |
2115 | WARN(1, "current_time() called with uninitialized super_block in the inode"); | |
2116 | return now; | |
2117 | } | |
2118 | ||
2119 | return timespec_trunc(now, inode->i_sb->s_time_gran); | |
2120 | } | |
2121 | EXPORT_SYMBOL(current_time); |