]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/inode.c
bus: brcmstb_gisb: update to support new revision
[mirror_ubuntu-bionic-kernel.git] / fs / inode.c
CommitLineData
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
55static unsigned int i_hash_mask __read_mostly;
56static unsigned int i_hash_shift __read_mostly;
67a23c49
DC
57static struct hlist_head *inode_hashtable __read_mostly;
58static __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 */
64const struct address_space_operations empty_aops = {
65};
66EXPORT_SYMBOL(empty_aops);
67
1da177e4
LT
68/*
69 * Statistics gathering..
70 */
71struct inodes_stat_t inodes_stat;
72
3942c07c
GC
73static DEFINE_PER_CPU(unsigned long, nr_inodes);
74static DEFINE_PER_CPU(unsigned long, nr_unused);
cffbc8aa 75
6b3304b5 76static struct kmem_cache *inode_cachep __read_mostly;
1da177e4 77
3942c07c 78static 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 87static 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 96long 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 107int 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
116static 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 129int 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
197out:
198 return -ENOMEM;
1da177e4 199}
2cb1599f
DC
200EXPORT_SYMBOL(inode_init_always);
201
202static 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
225void free_inode_nonrcu(struct inode *inode)
226{
227 kmem_cache_free(inode_cachep, inode);
228}
229EXPORT_SYMBOL(free_inode_nonrcu);
230
2e00c97e 231void __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}
251EXPORT_SYMBOL(__destroy_inode);
252
fa0d7e3d
NP
253static 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 259static 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 */
280void 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}
287EXPORT_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 */
297void 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}
304EXPORT_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 */
314void 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}
326EXPORT_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 */
336void 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}
345EXPORT_SYMBOL(inc_nlink);
346
2aa15890
MS
347void 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}
357EXPORT_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 */
364void 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
375EXPORT_SYMBOL(inode_init_once);
376
51cc5068 377static 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 387void __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 */
395void ihold(struct inode *inode)
396{
397 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
398}
399EXPORT_SYMBOL(ihold);
400
9e38d86f
NP
401static 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);
563f4001
JB
405 else
406 inode->i_state |= I_REFERENCED;
9e38d86f 407}
2e147f1e 408
4eff96dd
JK
409/*
410 * Add inode to LRU if needed (inode is unused and clean).
411 *
412 * Needs inode->i_lock held.
413 */
414void inode_add_lru(struct inode *inode)
415{
0ae45f63
TT
416 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
417 I_FREEING | I_WILL_FREE)) &&
4eff96dd
JK
418 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & MS_ACTIVE)
419 inode_lru_list_add(inode);
420}
421
422
9e38d86f
NP
423static void inode_lru_list_del(struct inode *inode)
424{
bc3b14cb
DC
425
426 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
fcb94f72 427 this_cpu_dec(nr_unused);
1da177e4
LT
428}
429
646ec461
CH
430/**
431 * inode_sb_list_add - add inode to the superblock list of inodes
432 * @inode: inode to add
433 */
434void inode_sb_list_add(struct inode *inode)
435{
74278da9 436 spin_lock(&inode->i_sb->s_inode_list_lock);
55fa6091 437 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
74278da9 438 spin_unlock(&inode->i_sb->s_inode_list_lock);
646ec461
CH
439}
440EXPORT_SYMBOL_GPL(inode_sb_list_add);
441
55fa6091 442static inline void inode_sb_list_del(struct inode *inode)
646ec461 443{
a209dfc7 444 if (!list_empty(&inode->i_sb_list)) {
74278da9 445 spin_lock(&inode->i_sb->s_inode_list_lock);
a209dfc7 446 list_del_init(&inode->i_sb_list);
74278da9 447 spin_unlock(&inode->i_sb->s_inode_list_lock);
a209dfc7 448 }
646ec461
CH
449}
450
4c51acbc
DC
451static unsigned long hash(struct super_block *sb, unsigned long hashval)
452{
453 unsigned long tmp;
454
455 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
456 L1_CACHE_BYTES;
4b4563dc
CH
457 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
458 return tmp & i_hash_mask;
4c51acbc
DC
459}
460
461/**
462 * __insert_inode_hash - hash an inode
463 * @inode: unhashed inode
464 * @hashval: unsigned long value used to locate this object in the
465 * inode_hashtable.
466 *
467 * Add an inode to the inode hash for this superblock.
468 */
469void __insert_inode_hash(struct inode *inode, unsigned long hashval)
470{
646ec461
CH
471 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
472
67a23c49 473 spin_lock(&inode_hash_lock);
250df6ed 474 spin_lock(&inode->i_lock);
646ec461 475 hlist_add_head(&inode->i_hash, b);
250df6ed 476 spin_unlock(&inode->i_lock);
67a23c49 477 spin_unlock(&inode_hash_lock);
4c51acbc
DC
478}
479EXPORT_SYMBOL(__insert_inode_hash);
480
4c51acbc 481/**
f2ee7abf 482 * __remove_inode_hash - remove an inode from the hash
4c51acbc
DC
483 * @inode: inode to unhash
484 *
485 * Remove an inode from the superblock.
486 */
f2ee7abf 487void __remove_inode_hash(struct inode *inode)
4c51acbc 488{
67a23c49 489 spin_lock(&inode_hash_lock);
250df6ed 490 spin_lock(&inode->i_lock);
4c51acbc 491 hlist_del_init(&inode->i_hash);
250df6ed 492 spin_unlock(&inode->i_lock);
67a23c49 493 spin_unlock(&inode_hash_lock);
4c51acbc 494}
f2ee7abf 495EXPORT_SYMBOL(__remove_inode_hash);
4c51acbc 496
dbd5768f 497void clear_inode(struct inode *inode)
b0683aa6
AV
498{
499 might_sleep();
08142579
JK
500 /*
501 * We have to cycle tree_lock here because reclaim can be still in the
502 * process of removing the last page (in __delete_from_page_cache())
503 * and we must not free mapping under it.
504 */
505 spin_lock_irq(&inode->i_data.tree_lock);
b0683aa6 506 BUG_ON(inode->i_data.nrpages);
f9fe48be 507 BUG_ON(inode->i_data.nrexceptional);
08142579 508 spin_unlock_irq(&inode->i_data.tree_lock);
b0683aa6
AV
509 BUG_ON(!list_empty(&inode->i_data.private_list));
510 BUG_ON(!(inode->i_state & I_FREEING));
511 BUG_ON(inode->i_state & I_CLEAR);
6c60d2b5 512 BUG_ON(!list_empty(&inode->i_wb_list));
fa0d7e3d 513 /* don't need i_lock here, no concurrent mods to i_state */
b0683aa6
AV
514 inode->i_state = I_FREEING | I_CLEAR;
515}
dbd5768f 516EXPORT_SYMBOL(clear_inode);
b0683aa6 517
b2b2af8e
DC
518/*
519 * Free the inode passed in, removing it from the lists it is still connected
520 * to. We remove any pages still attached to the inode and wait for any IO that
521 * is still in progress before finally destroying the inode.
522 *
523 * An inode must already be marked I_FREEING so that we avoid the inode being
524 * moved back onto lists if we race with other code that manipulates the lists
525 * (e.g. writeback_single_inode). The caller is responsible for setting this.
526 *
527 * An inode must already be removed from the LRU list before being evicted from
528 * the cache. This should occur atomically with setting the I_FREEING state
529 * flag, so no inodes here should ever be on the LRU when being evicted.
530 */
644da596 531static void evict(struct inode *inode)
b4272d4c
AV
532{
533 const struct super_operations *op = inode->i_sb->s_op;
534
b2b2af8e
DC
535 BUG_ON(!(inode->i_state & I_FREEING));
536 BUG_ON(!list_empty(&inode->i_lru));
537
c7f54084
DC
538 if (!list_empty(&inode->i_io_list))
539 inode_io_list_del(inode);
b12362bd 540
55fa6091
DC
541 inode_sb_list_del(inode);
542
169ebd90
JK
543 /*
544 * Wait for flusher thread to be done with the inode so that filesystem
545 * does not start destroying it while writeback is still running. Since
546 * the inode has I_FREEING set, flusher thread won't start new work on
547 * the inode. We just have to wait for running writeback to finish.
548 */
549 inode_wait_for_writeback(inode);
7994e6f7 550
be7ce416
AV
551 if (op->evict_inode) {
552 op->evict_inode(inode);
b4272d4c 553 } else {
91b0abe3 554 truncate_inode_pages_final(&inode->i_data);
dbd5768f 555 clear_inode(inode);
b4272d4c 556 }
661074e9
AV
557 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
558 bd_forget(inode);
559 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
560 cd_forget(inode);
b2b2af8e
DC
561
562 remove_inode_hash(inode);
563
564 spin_lock(&inode->i_lock);
565 wake_up_bit(&inode->i_state, __I_NEW);
566 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
567 spin_unlock(&inode->i_lock);
568
569 destroy_inode(inode);
b4272d4c
AV
570}
571
1da177e4
LT
572/*
573 * dispose_list - dispose of the contents of a local list
574 * @head: the head of the list to free
575 *
576 * Dispose-list gets a local list with local inodes in it, so it doesn't
577 * need to worry about list corruption and SMP locks.
578 */
579static void dispose_list(struct list_head *head)
580{
1da177e4
LT
581 while (!list_empty(head)) {
582 struct inode *inode;
583
7ccf19a8
NP
584 inode = list_first_entry(head, struct inode, i_lru);
585 list_del_init(&inode->i_lru);
1da177e4 586
644da596 587 evict(inode);
ac05fbb4 588 cond_resched();
1da177e4 589 }
1da177e4
LT
590}
591
63997e98
AV
592/**
593 * evict_inodes - evict all evictable inodes for a superblock
594 * @sb: superblock to operate on
595 *
596 * Make sure that no inodes with zero refcount are retained. This is
597 * called by superblock shutdown after having MS_ACTIVE flag removed,
598 * so any inode reaching zero refcount during or after that call will
599 * be immediately evicted.
1da177e4 600 */
63997e98 601void evict_inodes(struct super_block *sb)
1da177e4 602{
63997e98
AV
603 struct inode *inode, *next;
604 LIST_HEAD(dispose);
1da177e4 605
ac05fbb4 606again:
74278da9 607 spin_lock(&sb->s_inode_list_lock);
63997e98
AV
608 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
609 if (atomic_read(&inode->i_count))
aabb8fdb 610 continue;
250df6ed
DC
611
612 spin_lock(&inode->i_lock);
613 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
614 spin_unlock(&inode->i_lock);
1da177e4 615 continue;
250df6ed 616 }
63997e98
AV
617
618 inode->i_state |= I_FREEING;
02afc410 619 inode_lru_list_del(inode);
250df6ed 620 spin_unlock(&inode->i_lock);
02afc410 621 list_add(&inode->i_lru, &dispose);
ac05fbb4
JB
622
623 /*
624 * We can have a ton of inodes to evict at unmount time given
625 * enough memory, check to see if we need to go to sleep for a
626 * bit so we don't livelock.
627 */
628 if (need_resched()) {
629 spin_unlock(&sb->s_inode_list_lock);
630 cond_resched();
631 dispose_list(&dispose);
632 goto again;
633 }
1da177e4 634 }
74278da9 635 spin_unlock(&sb->s_inode_list_lock);
63997e98
AV
636
637 dispose_list(&dispose);
1da177e4
LT
638}
639
1da177e4 640/**
a0318786
CH
641 * invalidate_inodes - attempt to free all inodes on a superblock
642 * @sb: superblock to operate on
93b270f7 643 * @kill_dirty: flag to guide handling of dirty inodes
1da177e4 644 *
a0318786
CH
645 * Attempts to free all inodes for a given superblock. If there were any
646 * busy inodes return a non-zero value, else zero.
93b270f7
N
647 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
648 * them as busy.
1da177e4 649 */
93b270f7 650int invalidate_inodes(struct super_block *sb, bool kill_dirty)
1da177e4 651{
cffbc8aa 652 int busy = 0;
a0318786
CH
653 struct inode *inode, *next;
654 LIST_HEAD(dispose);
1da177e4 655
74278da9 656 spin_lock(&sb->s_inode_list_lock);
a0318786 657 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
250df6ed
DC
658 spin_lock(&inode->i_lock);
659 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
660 spin_unlock(&inode->i_lock);
aabb8fdb 661 continue;
250df6ed 662 }
0ae45f63 663 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
250df6ed 664 spin_unlock(&inode->i_lock);
93b270f7
N
665 busy = 1;
666 continue;
667 }
99a38919 668 if (atomic_read(&inode->i_count)) {
250df6ed 669 spin_unlock(&inode->i_lock);
99a38919 670 busy = 1;
1da177e4
LT
671 continue;
672 }
99a38919 673
99a38919 674 inode->i_state |= I_FREEING;
02afc410 675 inode_lru_list_del(inode);
250df6ed 676 spin_unlock(&inode->i_lock);
02afc410 677 list_add(&inode->i_lru, &dispose);
1da177e4 678 }
74278da9 679 spin_unlock(&sb->s_inode_list_lock);
1da177e4 680
a0318786 681 dispose_list(&dispose);
1da177e4
LT
682
683 return busy;
684}
1da177e4 685
1da177e4 686/*
bc3b14cb 687 * Isolate the inode from the LRU in preparation for freeing it.
1da177e4
LT
688 *
689 * Any inodes which are pinned purely because of attached pagecache have their
9e38d86f
NP
690 * pagecache removed. If the inode has metadata buffers attached to
691 * mapping->private_list then try to remove them.
1da177e4 692 *
9e38d86f
NP
693 * If the inode has the I_REFERENCED flag set, then it means that it has been
694 * used recently - the flag is set in iput_final(). When we encounter such an
695 * inode, clear the flag and move it to the back of the LRU so it gets another
696 * pass through the LRU before it gets reclaimed. This is necessary because of
697 * the fact we are doing lazy LRU updates to minimise lock contention so the
698 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
699 * with this flag set because they are the inodes that are out of order.
1da177e4 700 */
3f97b163
VD
701static enum lru_status inode_lru_isolate(struct list_head *item,
702 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1da177e4 703{
bc3b14cb
DC
704 struct list_head *freeable = arg;
705 struct inode *inode = container_of(item, struct inode, i_lru);
1da177e4 706
bc3b14cb
DC
707 /*
708 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
709 * If we fail to get the lock, just skip it.
710 */
711 if (!spin_trylock(&inode->i_lock))
712 return LRU_SKIP;
1da177e4 713
bc3b14cb
DC
714 /*
715 * Referenced or dirty inodes are still in use. Give them another pass
716 * through the LRU as we canot reclaim them now.
717 */
718 if (atomic_read(&inode->i_count) ||
719 (inode->i_state & ~I_REFERENCED)) {
3f97b163 720 list_lru_isolate(lru, &inode->i_lru);
bc3b14cb
DC
721 spin_unlock(&inode->i_lock);
722 this_cpu_dec(nr_unused);
723 return LRU_REMOVED;
724 }
1da177e4 725
bc3b14cb
DC
726 /* recently referenced inodes get one more pass */
727 if (inode->i_state & I_REFERENCED) {
728 inode->i_state &= ~I_REFERENCED;
729 spin_unlock(&inode->i_lock);
730 return LRU_ROTATE;
731 }
1da177e4 732
bc3b14cb
DC
733 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
734 __iget(inode);
735 spin_unlock(&inode->i_lock);
736 spin_unlock(lru_lock);
737 if (remove_inode_buffers(inode)) {
738 unsigned long reap;
739 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
740 if (current_is_kswapd())
741 __count_vm_events(KSWAPD_INODESTEAL, reap);
742 else
743 __count_vm_events(PGINODESTEAL, reap);
744 if (current->reclaim_state)
745 current->reclaim_state->reclaimed_slab += reap;
02afc410 746 }
bc3b14cb
DC
747 iput(inode);
748 spin_lock(lru_lock);
749 return LRU_RETRY;
750 }
02afc410 751
bc3b14cb
DC
752 WARN_ON(inode->i_state & I_NEW);
753 inode->i_state |= I_FREEING;
3f97b163 754 list_lru_isolate_move(lru, &inode->i_lru, freeable);
bc3b14cb 755 spin_unlock(&inode->i_lock);
9e38d86f 756
bc3b14cb
DC
757 this_cpu_dec(nr_unused);
758 return LRU_REMOVED;
759}
7ccf19a8 760
bc3b14cb
DC
761/*
762 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
763 * This is called from the superblock shrinker function with a number of inodes
764 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
765 * then are freed outside inode_lock by dispose_list().
766 */
503c358c 767long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
bc3b14cb
DC
768{
769 LIST_HEAD(freeable);
770 long freed;
1da177e4 771
503c358c
VD
772 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
773 inode_lru_isolate, &freeable);
1da177e4 774 dispose_list(&freeable);
0a234c6d 775 return freed;
1da177e4
LT
776}
777
1da177e4
LT
778static void __wait_on_freeing_inode(struct inode *inode);
779/*
780 * Called with the inode lock held.
1da177e4 781 */
6b3304b5
MK
782static struct inode *find_inode(struct super_block *sb,
783 struct hlist_head *head,
784 int (*test)(struct inode *, void *),
785 void *data)
1da177e4 786{
6b3304b5 787 struct inode *inode = NULL;
1da177e4
LT
788
789repeat:
b67bfe0d 790 hlist_for_each_entry(inode, head, i_hash) {
5a3cd992 791 if (inode->i_sb != sb)
1da177e4 792 continue;
5a3cd992 793 if (!test(inode, data))
1da177e4 794 continue;
5a3cd992 795 spin_lock(&inode->i_lock);
a4ffdde6 796 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1da177e4
LT
797 __wait_on_freeing_inode(inode);
798 goto repeat;
799 }
f7899bd5 800 __iget(inode);
250df6ed 801 spin_unlock(&inode->i_lock);
f7899bd5 802 return inode;
1da177e4 803 }
f7899bd5 804 return NULL;
1da177e4
LT
805}
806
807/*
808 * find_inode_fast is the fast path version of find_inode, see the comment at
809 * iget_locked for details.
810 */
6b3304b5
MK
811static struct inode *find_inode_fast(struct super_block *sb,
812 struct hlist_head *head, unsigned long ino)
1da177e4 813{
6b3304b5 814 struct inode *inode = NULL;
1da177e4
LT
815
816repeat:
b67bfe0d 817 hlist_for_each_entry(inode, head, i_hash) {
5a3cd992 818 if (inode->i_ino != ino)
1da177e4 819 continue;
5a3cd992 820 if (inode->i_sb != sb)
1da177e4 821 continue;
5a3cd992 822 spin_lock(&inode->i_lock);
a4ffdde6 823 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1da177e4
LT
824 __wait_on_freeing_inode(inode);
825 goto repeat;
826 }
f7899bd5 827 __iget(inode);
250df6ed 828 spin_unlock(&inode->i_lock);
f7899bd5 829 return inode;
1da177e4 830 }
f7899bd5 831 return NULL;
8290c35f
DC
832}
833
f991bd2e
ED
834/*
835 * Each cpu owns a range of LAST_INO_BATCH numbers.
836 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
837 * to renew the exhausted range.
8290c35f 838 *
f991bd2e
ED
839 * This does not significantly increase overflow rate because every CPU can
840 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
841 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
842 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
843 * overflow rate by 2x, which does not seem too significant.
844 *
845 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
846 * error if st_ino won't fit in target struct field. Use 32bit counter
847 * here to attempt to avoid that.
8290c35f 848 */
f991bd2e
ED
849#define LAST_INO_BATCH 1024
850static DEFINE_PER_CPU(unsigned int, last_ino);
851
85fe4025 852unsigned int get_next_ino(void)
8290c35f 853{
f991bd2e
ED
854 unsigned int *p = &get_cpu_var(last_ino);
855 unsigned int res = *p;
8290c35f 856
f991bd2e
ED
857#ifdef CONFIG_SMP
858 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
859 static atomic_t shared_last_ino;
860 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
861
862 res = next - LAST_INO_BATCH;
863 }
864#endif
865
2adc376c
CM
866 res++;
867 /* get_next_ino should not provide a 0 inode number */
868 if (unlikely(!res))
869 res++;
870 *p = res;
f991bd2e
ED
871 put_cpu_var(last_ino);
872 return res;
8290c35f 873}
85fe4025 874EXPORT_SYMBOL(get_next_ino);
8290c35f 875
a209dfc7
ED
876/**
877 * new_inode_pseudo - obtain an inode
878 * @sb: superblock
879 *
880 * Allocates a new inode for given superblock.
881 * Inode wont be chained in superblock s_inodes list
882 * This means :
883 * - fs can't be unmount
884 * - quotas, fsnotify, writeback can't work
885 */
886struct inode *new_inode_pseudo(struct super_block *sb)
887{
888 struct inode *inode = alloc_inode(sb);
889
890 if (inode) {
891 spin_lock(&inode->i_lock);
892 inode->i_state = 0;
893 spin_unlock(&inode->i_lock);
894 INIT_LIST_HEAD(&inode->i_sb_list);
895 }
896 return inode;
897}
898
1da177e4
LT
899/**
900 * new_inode - obtain an inode
901 * @sb: superblock
902 *
769848c0 903 * Allocates a new inode for given superblock. The default gfp_mask
3c1d4378 904 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
769848c0
MG
905 * If HIGHMEM pages are unsuitable or it is known that pages allocated
906 * for the page cache are not reclaimable or migratable,
907 * mapping_set_gfp_mask() must be called with suitable flags on the
908 * newly created inode's mapping
909 *
1da177e4
LT
910 */
911struct inode *new_inode(struct super_block *sb)
912{
6b3304b5 913 struct inode *inode;
1da177e4 914
74278da9 915 spin_lock_prefetch(&sb->s_inode_list_lock);
6b3304b5 916
a209dfc7
ED
917 inode = new_inode_pseudo(sb);
918 if (inode)
55fa6091 919 inode_sb_list_add(inode);
1da177e4
LT
920 return inode;
921}
1da177e4
LT
922EXPORT_SYMBOL(new_inode);
923
14358e6d 924#ifdef CONFIG_DEBUG_LOCK_ALLOC
e096d0c7
JB
925void lockdep_annotate_inode_mutex_key(struct inode *inode)
926{
a3314a0e 927 if (S_ISDIR(inode->i_mode)) {
1e89a5e1
PZ
928 struct file_system_type *type = inode->i_sb->s_type;
929
9a7aa12f 930 /* Set new key only if filesystem hasn't already changed it */
9902af79 931 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
9a7aa12f
JK
932 /*
933 * ensure nobody is actually holding i_mutex
934 */
9902af79
AV
935 // mutex_destroy(&inode->i_mutex);
936 init_rwsem(&inode->i_rwsem);
937 lockdep_set_class(&inode->i_rwsem,
9a7aa12f
JK
938 &type->i_mutex_dir_key);
939 }
1e89a5e1 940 }
e096d0c7
JB
941}
942EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
14358e6d 943#endif
e096d0c7
JB
944
945/**
946 * unlock_new_inode - clear the I_NEW state and wake up any waiters
947 * @inode: new inode to unlock
948 *
949 * Called when the inode is fully initialised to clear the new state of the
950 * inode and wake up anyone waiting for the inode to finish initialisation.
951 */
952void unlock_new_inode(struct inode *inode)
953{
954 lockdep_annotate_inode_mutex_key(inode);
250df6ed 955 spin_lock(&inode->i_lock);
eaff8079
CH
956 WARN_ON(!(inode->i_state & I_NEW));
957 inode->i_state &= ~I_NEW;
310fa7a3 958 smp_mb();
250df6ed
DC
959 wake_up_bit(&inode->i_state, __I_NEW);
960 spin_unlock(&inode->i_lock);
1da177e4 961}
1da177e4
LT
962EXPORT_SYMBOL(unlock_new_inode);
963
375e289e
BF
964/**
965 * lock_two_nondirectories - take two i_mutexes on non-directory objects
4fd699ae
BF
966 *
967 * Lock any non-NULL argument that is not a directory.
968 * Zero, one or two objects may be locked by this function.
969 *
375e289e
BF
970 * @inode1: first inode to lock
971 * @inode2: second inode to lock
972 */
973void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
974{
4fd699ae
BF
975 if (inode1 > inode2)
976 swap(inode1, inode2);
977
978 if (inode1 && !S_ISDIR(inode1->i_mode))
5955102c 979 inode_lock(inode1);
4fd699ae 980 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
5955102c 981 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
375e289e
BF
982}
983EXPORT_SYMBOL(lock_two_nondirectories);
984
985/**
986 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
987 * @inode1: first inode to unlock
988 * @inode2: second inode to unlock
989 */
990void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
991{
4fd699ae 992 if (inode1 && !S_ISDIR(inode1->i_mode))
5955102c 993 inode_unlock(inode1);
4fd699ae 994 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
5955102c 995 inode_unlock(inode2);
375e289e
BF
996}
997EXPORT_SYMBOL(unlock_two_nondirectories);
998
0b2d0724
CH
999/**
1000 * iget5_locked - obtain an inode from a mounted file system
1001 * @sb: super block of file system
1002 * @hashval: hash value (usually inode number) to get
1003 * @test: callback used for comparisons between inodes
1004 * @set: callback used to initialize a new struct inode
1005 * @data: opaque data pointer to pass to @test and @set
1006 *
1007 * Search for the inode specified by @hashval and @data in the inode cache,
1008 * and if present it is return it with an increased reference count. This is
1009 * a generalized version of iget_locked() for file systems where the inode
1010 * number is not sufficient for unique identification of an inode.
1011 *
1012 * If the inode is not in cache, allocate a new inode and return it locked,
1013 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1014 * before unlocking it via unlock_new_inode().
1da177e4 1015 *
0b2d0724
CH
1016 * Note both @test and @set are called with the inode_hash_lock held, so can't
1017 * sleep.
1da177e4 1018 */
0b2d0724
CH
1019struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1020 int (*test)(struct inode *, void *),
1021 int (*set)(struct inode *, void *), void *data)
1da177e4 1022{
0b2d0724 1023 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
6b3304b5 1024 struct inode *inode;
2864f301 1025again:
0b2d0724
CH
1026 spin_lock(&inode_hash_lock);
1027 inode = find_inode(sb, head, test, data);
1028 spin_unlock(&inode_hash_lock);
1029
1030 if (inode) {
1031 wait_on_inode(inode);
2864f301
AV
1032 if (unlikely(inode_unhashed(inode))) {
1033 iput(inode);
1034 goto again;
1035 }
0b2d0724
CH
1036 return inode;
1037 }
1038
1da177e4
LT
1039 inode = alloc_inode(sb);
1040 if (inode) {
6b3304b5 1041 struct inode *old;
1da177e4 1042
67a23c49 1043 spin_lock(&inode_hash_lock);
1da177e4
LT
1044 /* We released the lock, so.. */
1045 old = find_inode(sb, head, test, data);
1046 if (!old) {
1047 if (set(inode, data))
1048 goto set_failed;
1049
250df6ed
DC
1050 spin_lock(&inode->i_lock);
1051 inode->i_state = I_NEW;
646ec461 1052 hlist_add_head(&inode->i_hash, head);
250df6ed 1053 spin_unlock(&inode->i_lock);
55fa6091 1054 inode_sb_list_add(inode);
67a23c49 1055 spin_unlock(&inode_hash_lock);
1da177e4
LT
1056
1057 /* Return the locked inode with I_NEW set, the
1058 * caller is responsible for filling in the contents
1059 */
1060 return inode;
1061 }
1062
1063 /*
1064 * Uhhuh, somebody else created the same inode under
1065 * us. Use the old inode instead of the one we just
1066 * allocated.
1067 */
67a23c49 1068 spin_unlock(&inode_hash_lock);
1da177e4
LT
1069 destroy_inode(inode);
1070 inode = old;
1071 wait_on_inode(inode);
2864f301
AV
1072 if (unlikely(inode_unhashed(inode))) {
1073 iput(inode);
1074 goto again;
1075 }
1da177e4
LT
1076 }
1077 return inode;
1078
1079set_failed:
67a23c49 1080 spin_unlock(&inode_hash_lock);
1da177e4
LT
1081 destroy_inode(inode);
1082 return NULL;
1083}
0b2d0724 1084EXPORT_SYMBOL(iget5_locked);
1da177e4 1085
0b2d0724
CH
1086/**
1087 * iget_locked - obtain an inode from a mounted file system
1088 * @sb: super block of file system
1089 * @ino: inode number to get
1090 *
1091 * Search for the inode specified by @ino in the inode cache and if present
1092 * return it with an increased reference count. This is for file systems
1093 * where the inode number is sufficient for unique identification of an inode.
1094 *
1095 * If the inode is not in cache, allocate a new inode and return it locked,
1096 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1097 * before unlocking it via unlock_new_inode().
1da177e4 1098 */
0b2d0724 1099struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1da177e4 1100{
0b2d0724 1101 struct hlist_head *head = inode_hashtable + hash(sb, ino);
6b3304b5 1102 struct inode *inode;
2864f301 1103again:
0b2d0724
CH
1104 spin_lock(&inode_hash_lock);
1105 inode = find_inode_fast(sb, head, ino);
1106 spin_unlock(&inode_hash_lock);
1107 if (inode) {
1108 wait_on_inode(inode);
2864f301
AV
1109 if (unlikely(inode_unhashed(inode))) {
1110 iput(inode);
1111 goto again;
1112 }
0b2d0724
CH
1113 return inode;
1114 }
1115
1da177e4
LT
1116 inode = alloc_inode(sb);
1117 if (inode) {
6b3304b5 1118 struct inode *old;
1da177e4 1119
67a23c49 1120 spin_lock(&inode_hash_lock);
1da177e4
LT
1121 /* We released the lock, so.. */
1122 old = find_inode_fast(sb, head, ino);
1123 if (!old) {
1124 inode->i_ino = ino;
250df6ed
DC
1125 spin_lock(&inode->i_lock);
1126 inode->i_state = I_NEW;
646ec461 1127 hlist_add_head(&inode->i_hash, head);
250df6ed 1128 spin_unlock(&inode->i_lock);
55fa6091 1129 inode_sb_list_add(inode);
67a23c49 1130 spin_unlock(&inode_hash_lock);
1da177e4
LT
1131
1132 /* Return the locked inode with I_NEW set, the
1133 * caller is responsible for filling in the contents
1134 */
1135 return inode;
1136 }
1137
1138 /*
1139 * Uhhuh, somebody else created the same inode under
1140 * us. Use the old inode instead of the one we just
1141 * allocated.
1142 */
67a23c49 1143 spin_unlock(&inode_hash_lock);
1da177e4
LT
1144 destroy_inode(inode);
1145 inode = old;
1146 wait_on_inode(inode);
2864f301
AV
1147 if (unlikely(inode_unhashed(inode))) {
1148 iput(inode);
1149 goto again;
1150 }
1da177e4
LT
1151 }
1152 return inode;
1153}
0b2d0724 1154EXPORT_SYMBOL(iget_locked);
1da177e4 1155
ad5e195a
CH
1156/*
1157 * search the inode cache for a matching inode number.
1158 * If we find one, then the inode number we are trying to
1159 * allocate is not unique and so we should not use it.
1160 *
1161 * Returns 1 if the inode number is unique, 0 if it is not.
1162 */
1163static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1164{
1165 struct hlist_head *b = inode_hashtable + hash(sb, ino);
ad5e195a
CH
1166 struct inode *inode;
1167
67a23c49 1168 spin_lock(&inode_hash_lock);
b67bfe0d 1169 hlist_for_each_entry(inode, b, i_hash) {
67a23c49
DC
1170 if (inode->i_ino == ino && inode->i_sb == sb) {
1171 spin_unlock(&inode_hash_lock);
ad5e195a 1172 return 0;
67a23c49 1173 }
ad5e195a 1174 }
67a23c49 1175 spin_unlock(&inode_hash_lock);
ad5e195a
CH
1176
1177 return 1;
1178}
1179
1da177e4
LT
1180/**
1181 * iunique - get a unique inode number
1182 * @sb: superblock
1183 * @max_reserved: highest reserved inode number
1184 *
1185 * Obtain an inode number that is unique on the system for a given
1186 * superblock. This is used by file systems that have no natural
1187 * permanent inode numbering system. An inode number is returned that
1188 * is higher than the reserved limit but unique.
1189 *
1190 * BUGS:
1191 * With a large number of inodes live on the file system this function
1192 * currently becomes quite slow.
1193 */
1194ino_t iunique(struct super_block *sb, ino_t max_reserved)
1195{
866b04fc
JL
1196 /*
1197 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1198 * error if st_ino won't fit in target struct field. Use 32bit counter
1199 * here to attempt to avoid that.
1200 */
ad5e195a 1201 static DEFINE_SPINLOCK(iunique_lock);
866b04fc 1202 static unsigned int counter;
1da177e4 1203 ino_t res;
3361c7be 1204
ad5e195a 1205 spin_lock(&iunique_lock);
3361c7be
JL
1206 do {
1207 if (counter <= max_reserved)
1208 counter = max_reserved + 1;
1da177e4 1209 res = counter++;
ad5e195a
CH
1210 } while (!test_inode_iunique(sb, res));
1211 spin_unlock(&iunique_lock);
1da177e4 1212
3361c7be
JL
1213 return res;
1214}
1da177e4
LT
1215EXPORT_SYMBOL(iunique);
1216
1217struct inode *igrab(struct inode *inode)
1218{
250df6ed
DC
1219 spin_lock(&inode->i_lock);
1220 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1da177e4 1221 __iget(inode);
250df6ed
DC
1222 spin_unlock(&inode->i_lock);
1223 } else {
1224 spin_unlock(&inode->i_lock);
1da177e4
LT
1225 /*
1226 * Handle the case where s_op->clear_inode is not been
1227 * called yet, and somebody is calling igrab
1228 * while the inode is getting freed.
1229 */
1230 inode = NULL;
250df6ed 1231 }
1da177e4
LT
1232 return inode;
1233}
1da177e4
LT
1234EXPORT_SYMBOL(igrab);
1235
1236/**
0b2d0724 1237 * ilookup5_nowait - search for an inode in the inode cache
1da177e4 1238 * @sb: super block of file system to search
0b2d0724 1239 * @hashval: hash value (usually inode number) to search for
1da177e4
LT
1240 * @test: callback used for comparisons between inodes
1241 * @data: opaque data pointer to pass to @test
1da177e4 1242 *
0b2d0724 1243 * Search for the inode specified by @hashval and @data in the inode cache.
1da177e4
LT
1244 * If the inode is in the cache, the inode is returned with an incremented
1245 * reference count.
1246 *
0b2d0724
CH
1247 * Note: I_NEW is not waited upon so you have to be very careful what you do
1248 * with the returned inode. You probably should be using ilookup5() instead.
1da177e4 1249 *
b6d0ad68 1250 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1da177e4 1251 */
0b2d0724
CH
1252struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1253 int (*test)(struct inode *, void *), void *data)
1da177e4 1254{
0b2d0724 1255 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1da177e4
LT
1256 struct inode *inode;
1257
67a23c49 1258 spin_lock(&inode_hash_lock);
1da177e4 1259 inode = find_inode(sb, head, test, data);
67a23c49 1260 spin_unlock(&inode_hash_lock);
88bd5121 1261
0b2d0724 1262 return inode;
88bd5121 1263}
88bd5121
AA
1264EXPORT_SYMBOL(ilookup5_nowait);
1265
1266/**
1267 * ilookup5 - search for an inode in the inode cache
1268 * @sb: super block of file system to search
1269 * @hashval: hash value (usually inode number) to search for
1270 * @test: callback used for comparisons between inodes
1271 * @data: opaque data pointer to pass to @test
1272 *
0b2d0724
CH
1273 * Search for the inode specified by @hashval and @data in the inode cache,
1274 * and if the inode is in the cache, return the inode with an incremented
1275 * reference count. Waits on I_NEW before returning the inode.
88bd5121 1276 * returned with an incremented reference count.
1da177e4 1277 *
0b2d0724
CH
1278 * This is a generalized version of ilookup() for file systems where the
1279 * inode number is not sufficient for unique identification of an inode.
1da177e4 1280 *
0b2d0724 1281 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1da177e4
LT
1282 */
1283struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1284 int (*test)(struct inode *, void *), void *data)
1285{
2864f301
AV
1286 struct inode *inode;
1287again:
1288 inode = ilookup5_nowait(sb, hashval, test, data);
1289 if (inode) {
0b2d0724 1290 wait_on_inode(inode);
2864f301
AV
1291 if (unlikely(inode_unhashed(inode))) {
1292 iput(inode);
1293 goto again;
1294 }
1295 }
0b2d0724 1296 return inode;
1da177e4 1297}
1da177e4
LT
1298EXPORT_SYMBOL(ilookup5);
1299
1300/**
1301 * ilookup - search for an inode in the inode cache
1302 * @sb: super block of file system to search
1303 * @ino: inode number to search for
1304 *
0b2d0724
CH
1305 * Search for the inode @ino in the inode cache, and if the inode is in the
1306 * cache, the inode is returned with an incremented reference count.
1da177e4
LT
1307 */
1308struct inode *ilookup(struct super_block *sb, unsigned long ino)
1309{
1310 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1da177e4 1311 struct inode *inode;
2864f301 1312again:
0b2d0724
CH
1313 spin_lock(&inode_hash_lock);
1314 inode = find_inode_fast(sb, head, ino);
1315 spin_unlock(&inode_hash_lock);
1da177e4 1316
2864f301 1317 if (inode) {
0b2d0724 1318 wait_on_inode(inode);
2864f301
AV
1319 if (unlikely(inode_unhashed(inode))) {
1320 iput(inode);
1321 goto again;
1322 }
1323 }
0b2d0724 1324 return inode;
1da177e4 1325}
0b2d0724 1326EXPORT_SYMBOL(ilookup);
1da177e4 1327
fe032c42
TT
1328/**
1329 * find_inode_nowait - find an inode in the inode cache
1330 * @sb: super block of file system to search
1331 * @hashval: hash value (usually inode number) to search for
1332 * @match: callback used for comparisons between inodes
1333 * @data: opaque data pointer to pass to @match
1334 *
1335 * Search for the inode specified by @hashval and @data in the inode
1336 * cache, where the helper function @match will return 0 if the inode
1337 * does not match, 1 if the inode does match, and -1 if the search
1338 * should be stopped. The @match function must be responsible for
1339 * taking the i_lock spin_lock and checking i_state for an inode being
1340 * freed or being initialized, and incrementing the reference count
1341 * before returning 1. It also must not sleep, since it is called with
1342 * the inode_hash_lock spinlock held.
1343 *
1344 * This is a even more generalized version of ilookup5() when the
1345 * function must never block --- find_inode() can block in
1346 * __wait_on_freeing_inode() --- or when the caller can not increment
1347 * the reference count because the resulting iput() might cause an
1348 * inode eviction. The tradeoff is that the @match funtion must be
1349 * very carefully implemented.
1350 */
1351struct inode *find_inode_nowait(struct super_block *sb,
1352 unsigned long hashval,
1353 int (*match)(struct inode *, unsigned long,
1354 void *),
1355 void *data)
1356{
1357 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1358 struct inode *inode, *ret_inode = NULL;
1359 int mval;
1360
1361 spin_lock(&inode_hash_lock);
1362 hlist_for_each_entry(inode, head, i_hash) {
1363 if (inode->i_sb != sb)
1364 continue;
1365 mval = match(inode, hashval, data);
1366 if (mval == 0)
1367 continue;
1368 if (mval == 1)
1369 ret_inode = inode;
1370 goto out;
1371 }
1372out:
1373 spin_unlock(&inode_hash_lock);
1374 return ret_inode;
1375}
1376EXPORT_SYMBOL(find_inode_nowait);
1377
261bca86
AV
1378int insert_inode_locked(struct inode *inode)
1379{
1380 struct super_block *sb = inode->i_sb;
1381 ino_t ino = inode->i_ino;
1382 struct hlist_head *head = inode_hashtable + hash(sb, ino);
261bca86 1383
261bca86 1384 while (1) {
72a43d63 1385 struct inode *old = NULL;
67a23c49 1386 spin_lock(&inode_hash_lock);
b67bfe0d 1387 hlist_for_each_entry(old, head, i_hash) {
72a43d63
AV
1388 if (old->i_ino != ino)
1389 continue;
1390 if (old->i_sb != sb)
1391 continue;
250df6ed
DC
1392 spin_lock(&old->i_lock);
1393 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1394 spin_unlock(&old->i_lock);
72a43d63 1395 continue;
250df6ed 1396 }
72a43d63
AV
1397 break;
1398 }
b67bfe0d 1399 if (likely(!old)) {
250df6ed
DC
1400 spin_lock(&inode->i_lock);
1401 inode->i_state |= I_NEW;
261bca86 1402 hlist_add_head(&inode->i_hash, head);
250df6ed 1403 spin_unlock(&inode->i_lock);
67a23c49 1404 spin_unlock(&inode_hash_lock);
261bca86
AV
1405 return 0;
1406 }
1407 __iget(old);
250df6ed 1408 spin_unlock(&old->i_lock);
67a23c49 1409 spin_unlock(&inode_hash_lock);
261bca86 1410 wait_on_inode(old);
1d3382cb 1411 if (unlikely(!inode_unhashed(old))) {
261bca86
AV
1412 iput(old);
1413 return -EBUSY;
1414 }
1415 iput(old);
1416 }
1417}
261bca86
AV
1418EXPORT_SYMBOL(insert_inode_locked);
1419
1420int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1421 int (*test)(struct inode *, void *), void *data)
1422{
1423 struct super_block *sb = inode->i_sb;
1424 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
261bca86 1425
261bca86 1426 while (1) {
72a43d63
AV
1427 struct inode *old = NULL;
1428
67a23c49 1429 spin_lock(&inode_hash_lock);
b67bfe0d 1430 hlist_for_each_entry(old, head, i_hash) {
72a43d63
AV
1431 if (old->i_sb != sb)
1432 continue;
1433 if (!test(old, data))
1434 continue;
250df6ed
DC
1435 spin_lock(&old->i_lock);
1436 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1437 spin_unlock(&old->i_lock);
72a43d63 1438 continue;
250df6ed 1439 }
72a43d63
AV
1440 break;
1441 }
b67bfe0d 1442 if (likely(!old)) {
250df6ed
DC
1443 spin_lock(&inode->i_lock);
1444 inode->i_state |= I_NEW;
261bca86 1445 hlist_add_head(&inode->i_hash, head);
250df6ed 1446 spin_unlock(&inode->i_lock);
67a23c49 1447 spin_unlock(&inode_hash_lock);
261bca86
AV
1448 return 0;
1449 }
1450 __iget(old);
250df6ed 1451 spin_unlock(&old->i_lock);
67a23c49 1452 spin_unlock(&inode_hash_lock);
261bca86 1453 wait_on_inode(old);
1d3382cb 1454 if (unlikely(!inode_unhashed(old))) {
261bca86
AV
1455 iput(old);
1456 return -EBUSY;
1457 }
1458 iput(old);
1459 }
1460}
261bca86
AV
1461EXPORT_SYMBOL(insert_inode_locked4);
1462
1da177e4 1463
45321ac5
AV
1464int generic_delete_inode(struct inode *inode)
1465{
1466 return 1;
1467}
1468EXPORT_SYMBOL(generic_delete_inode);
1469
45321ac5
AV
1470/*
1471 * Called when we're dropping the last reference
1472 * to an inode.
22fe4042 1473 *
45321ac5
AV
1474 * Call the FS "drop_inode()" function, defaulting to
1475 * the legacy UNIX filesystem behaviour. If it tells
1476 * us to evict inode, do so. Otherwise, retain inode
1477 * in cache if fs is alive, sync and evict if fs is
1478 * shutting down.
22fe4042 1479 */
45321ac5 1480static void iput_final(struct inode *inode)
1da177e4
LT
1481{
1482 struct super_block *sb = inode->i_sb;
45321ac5
AV
1483 const struct super_operations *op = inode->i_sb->s_op;
1484 int drop;
1485
250df6ed
DC
1486 WARN_ON(inode->i_state & I_NEW);
1487
e7f59097 1488 if (op->drop_inode)
45321ac5
AV
1489 drop = op->drop_inode(inode);
1490 else
1491 drop = generic_drop_inode(inode);
1da177e4 1492
b2b2af8e 1493 if (!drop && (sb->s_flags & MS_ACTIVE)) {
4eff96dd 1494 inode_add_lru(inode);
b2b2af8e 1495 spin_unlock(&inode->i_lock);
b2b2af8e
DC
1496 return;
1497 }
1498
45321ac5 1499 if (!drop) {
991114c6 1500 inode->i_state |= I_WILL_FREE;
250df6ed 1501 spin_unlock(&inode->i_lock);
1da177e4 1502 write_inode_now(inode, 1);
250df6ed 1503 spin_lock(&inode->i_lock);
7ef0d737 1504 WARN_ON(inode->i_state & I_NEW);
991114c6 1505 inode->i_state &= ~I_WILL_FREE;
1da177e4 1506 }
7ccf19a8 1507
991114c6 1508 inode->i_state |= I_FREEING;
c4ae0c65
ED
1509 if (!list_empty(&inode->i_lru))
1510 inode_lru_list_del(inode);
b2b2af8e 1511 spin_unlock(&inode->i_lock);
b2b2af8e 1512
644da596 1513 evict(inode);
1da177e4
LT
1514}
1515
1da177e4 1516/**
6b3304b5 1517 * iput - put an inode
1da177e4
LT
1518 * @inode: inode to put
1519 *
1520 * Puts an inode, dropping its usage count. If the inode use count hits
1521 * zero, the inode is then freed and may also be destroyed.
1522 *
1523 * Consequently, iput() can sleep.
1524 */
1525void iput(struct inode *inode)
1526{
0ae45f63
TT
1527 if (!inode)
1528 return;
1529 BUG_ON(inode->i_state & I_CLEAR);
1530retry:
1531 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1532 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1533 atomic_inc(&inode->i_count);
1534 inode->i_state &= ~I_DIRTY_TIME;
1535 spin_unlock(&inode->i_lock);
1536 trace_writeback_lazytime_iput(inode);
1537 mark_inode_dirty_sync(inode);
1538 goto retry;
1539 }
1540 iput_final(inode);
1da177e4
LT
1541 }
1542}
1da177e4
LT
1543EXPORT_SYMBOL(iput);
1544
1545/**
1546 * bmap - find a block number in a file
1547 * @inode: inode of file
1548 * @block: block to find
1549 *
1550 * Returns the block number on the device holding the inode that
1551 * is the disk block number for the block of the file requested.
1552 * That is, asked for block 4 of inode 1 the function will return the
6b3304b5 1553 * disk block relative to the disk start that holds that block of the
1da177e4
LT
1554 * file.
1555 */
6b3304b5 1556sector_t bmap(struct inode *inode, sector_t block)
1da177e4
LT
1557{
1558 sector_t res = 0;
1559 if (inode->i_mapping->a_ops->bmap)
1560 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1561 return res;
1562}
1da177e4
LT
1563EXPORT_SYMBOL(bmap);
1564
598e3c8f
MS
1565/*
1566 * Update times in overlayed inode from underlying real inode
1567 */
1568static void update_ovl_inode_times(struct dentry *dentry, struct inode *inode,
1569 bool rcu)
1570{
1571 if (!rcu) {
1572 struct inode *realinode = d_real_inode(dentry);
1573
1574 if (unlikely(inode != realinode) &&
1575 (!timespec_equal(&inode->i_mtime, &realinode->i_mtime) ||
1576 !timespec_equal(&inode->i_ctime, &realinode->i_ctime))) {
1577 inode->i_mtime = realinode->i_mtime;
1578 inode->i_ctime = realinode->i_ctime;
1579 }
1580 }
1581}
1582
11ff6f05
MG
1583/*
1584 * With relative atime, only update atime if the previous atime is
1585 * earlier than either the ctime or mtime or if at least a day has
1586 * passed since the last atime update.
1587 */
598e3c8f
MS
1588static int relatime_need_update(const struct path *path, struct inode *inode,
1589 struct timespec now, bool rcu)
11ff6f05
MG
1590{
1591
598e3c8f 1592 if (!(path->mnt->mnt_flags & MNT_RELATIME))
11ff6f05 1593 return 1;
598e3c8f
MS
1594
1595 update_ovl_inode_times(path->dentry, inode, rcu);
11ff6f05
MG
1596 /*
1597 * Is mtime younger than atime? If yes, update atime:
1598 */
1599 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1600 return 1;
1601 /*
1602 * Is ctime younger than atime? If yes, update atime:
1603 */
1604 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1605 return 1;
1606
1607 /*
1608 * Is the previous atime value older than a day? If yes,
1609 * update atime:
1610 */
1611 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1612 return 1;
1613 /*
1614 * Good, we can skip the atime update:
1615 */
1616 return 0;
1617}
1618
0ae45f63 1619int generic_update_time(struct inode *inode, struct timespec *time, int flags)
c3b2da31 1620{
0ae45f63 1621 int iflags = I_DIRTY_TIME;
c3b2da31
JB
1622
1623 if (flags & S_ATIME)
1624 inode->i_atime = *time;
1625 if (flags & S_VERSION)
1626 inode_inc_iversion(inode);
1627 if (flags & S_CTIME)
1628 inode->i_ctime = *time;
1629 if (flags & S_MTIME)
1630 inode->i_mtime = *time;
0ae45f63
TT
1631
1632 if (!(inode->i_sb->s_flags & MS_LAZYTIME) || (flags & S_VERSION))
1633 iflags |= I_DIRTY_SYNC;
1634 __mark_inode_dirty(inode, iflags);
c3b2da31
JB
1635 return 0;
1636}
0ae45f63
TT
1637EXPORT_SYMBOL(generic_update_time);
1638
1639/*
1640 * This does the actual work of updating an inodes time or version. Must have
1641 * had called mnt_want_write() before calling this.
1642 */
1643static int update_time(struct inode *inode, struct timespec *time, int flags)
1644{
1645 int (*update_time)(struct inode *, struct timespec *, int);
1646
1647 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1648 generic_update_time;
1649
1650 return update_time(inode, time, flags);
1651}
c3b2da31 1652
1da177e4 1653/**
869243a0 1654 * touch_atime - update the access time
185553b2 1655 * @path: the &struct path to update
30fdc8ee 1656 * @inode: inode to update
1da177e4
LT
1657 *
1658 * Update the accessed time on an inode and mark it for writeback.
1659 * This function automatically handles read only file systems and media,
1660 * as well as the "noatime" flag and inode specific "noatime" markers.
1661 */
598e3c8f
MS
1662bool __atime_needs_update(const struct path *path, struct inode *inode,
1663 bool rcu)
1da177e4 1664{
68ac1234 1665 struct vfsmount *mnt = path->mnt;
1da177e4
LT
1666 struct timespec now;
1667
cdb70f3f 1668 if (inode->i_flags & S_NOATIME)
8fa9dd24 1669 return false;
0bd23d09
EB
1670
1671 /* Atime updates will likely cause i_uid and i_gid to be written
1672 * back improprely if their true value is unknown to the vfs.
1673 */
1674 if (HAS_UNMAPPED_ID(inode))
1675 return false;
1676
37756ced 1677 if (IS_NOATIME(inode))
8fa9dd24 1678 return false;
b2276138 1679 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
8fa9dd24 1680 return false;
47ae32d6 1681
cdb70f3f 1682 if (mnt->mnt_flags & MNT_NOATIME)
8fa9dd24 1683 return false;
cdb70f3f 1684 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
8fa9dd24 1685 return false;
1da177e4 1686
c2050a45 1687 now = current_time(inode);
11ff6f05 1688
598e3c8f 1689 if (!relatime_need_update(path, inode, now, rcu))
8fa9dd24 1690 return false;
11ff6f05 1691
47ae32d6 1692 if (timespec_equal(&inode->i_atime, &now))
8fa9dd24
N
1693 return false;
1694
1695 return true;
1696}
1697
1698void touch_atime(const struct path *path)
1699{
1700 struct vfsmount *mnt = path->mnt;
1701 struct inode *inode = d_inode(path->dentry);
1702 struct timespec now;
1703
598e3c8f 1704 if (!__atime_needs_update(path, inode, false))
b12536c2
AK
1705 return;
1706
5d37e9e6 1707 if (!sb_start_write_trylock(inode->i_sb))
b12536c2 1708 return;
47ae32d6 1709
8fa9dd24 1710 if (__mnt_want_write(mnt) != 0)
5d37e9e6 1711 goto skip_update;
c3b2da31
JB
1712 /*
1713 * File systems can error out when updating inodes if they need to
1714 * allocate new space to modify an inode (such is the case for
1715 * Btrfs), but since we touch atime while walking down the path we
1716 * really don't care if we failed to update the atime of the file,
1717 * so just ignore the return value.
2bc55652
AB
1718 * We may also fail on filesystems that have the ability to make parts
1719 * of the fs read only, e.g. subvolumes in Btrfs.
c3b2da31 1720 */
c2050a45 1721 now = current_time(inode);
c3b2da31 1722 update_time(inode, &now, S_ATIME);
5d37e9e6
JK
1723 __mnt_drop_write(mnt);
1724skip_update:
1725 sb_end_write(inode->i_sb);
1da177e4 1726}
869243a0 1727EXPORT_SYMBOL(touch_atime);
1da177e4 1728
3ed37648
CW
1729/*
1730 * The logic we want is
1731 *
1732 * if suid or (sgid and xgrp)
1733 * remove privs
1734 */
1735int should_remove_suid(struct dentry *dentry)
1736{
df2b1afd 1737 umode_t mode = d_inode(dentry)->i_mode;
3ed37648
CW
1738 int kill = 0;
1739
1740 /* suid always must be killed */
1741 if (unlikely(mode & S_ISUID))
1742 kill = ATTR_KILL_SUID;
1743
1744 /*
1745 * sgid without any exec bits is just a mandatory locking mark; leave
1746 * it alone. If some exec bits are set, it's a real sgid; kill it.
1747 */
1748 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1749 kill |= ATTR_KILL_SGID;
1750
1751 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1752 return kill;
1753
1754 return 0;
1755}
1756EXPORT_SYMBOL(should_remove_suid);
1757
dbfae0cd
JK
1758/*
1759 * Return mask of changes for notify_change() that need to be done as a
1760 * response to write or truncate. Return 0 if nothing has to be changed.
1761 * Negative value on error (change should be denied).
1762 */
45f147a1 1763int dentry_needs_remove_privs(struct dentry *dentry)
dbfae0cd 1764{
dbfae0cd
JK
1765 struct inode *inode = d_inode(dentry);
1766 int mask = 0;
1767 int ret;
1768
1769 if (IS_NOSEC(inode))
1770 return 0;
1771
1772 mask = should_remove_suid(dentry);
1773 ret = security_inode_need_killpriv(dentry);
1774 if (ret < 0)
1775 return ret;
1776 if (ret)
1777 mask |= ATTR_KILL_PRIV;
1778 return mask;
1779}
dbfae0cd
JK
1780
1781static int __remove_privs(struct dentry *dentry, int kill)
3ed37648
CW
1782{
1783 struct iattr newattrs;
1784
1785 newattrs.ia_valid = ATTR_FORCE | kill;
27ac0ffe
BF
1786 /*
1787 * Note we call this on write, so notify_change will not
1788 * encounter any conflicting delegations:
1789 */
1790 return notify_change(dentry, &newattrs, NULL);
3ed37648
CW
1791}
1792
5fa8e0a1
JK
1793/*
1794 * Remove special file priviledges (suid, capabilities) when file is written
1795 * to or truncated.
1796 */
1797int file_remove_privs(struct file *file)
3ed37648 1798{
c1892c37
MS
1799 struct dentry *dentry = file_dentry(file);
1800 struct inode *inode = file_inode(file);
dbfae0cd 1801 int kill;
3ed37648
CW
1802 int error = 0;
1803
1804 /* Fast path for nothing security related */
1805 if (IS_NOSEC(inode))
1806 return 0;
1807
c1892c37 1808 kill = dentry_needs_remove_privs(dentry);
dbfae0cd
JK
1809 if (kill < 0)
1810 return kill;
1811 if (kill)
1812 error = __remove_privs(dentry, kill);
2426f391
JK
1813 if (!error)
1814 inode_has_no_xattr(inode);
3ed37648
CW
1815
1816 return error;
1817}
5fa8e0a1 1818EXPORT_SYMBOL(file_remove_privs);
3ed37648 1819
1da177e4 1820/**
870f4817
CH
1821 * file_update_time - update mtime and ctime time
1822 * @file: file accessed
1da177e4 1823 *
870f4817
CH
1824 * Update the mtime and ctime members of an inode and mark the inode
1825 * for writeback. Note that this function is meant exclusively for
1826 * usage in the file write path of filesystems, and filesystems may
1827 * choose to explicitly ignore update via this function with the
2eadfc0e 1828 * S_NOCMTIME inode flag, e.g. for network filesystem where these
c3b2da31
JB
1829 * timestamps are handled by the server. This can return an error for
1830 * file systems who need to allocate space in order to update an inode.
1da177e4
LT
1831 */
1832
c3b2da31 1833int file_update_time(struct file *file)
1da177e4 1834{
496ad9aa 1835 struct inode *inode = file_inode(file);
1da177e4 1836 struct timespec now;
c3b2da31
JB
1837 int sync_it = 0;
1838 int ret;
1da177e4 1839
ce06e0b2 1840 /* First try to exhaust all avenues to not sync */
1da177e4 1841 if (IS_NOCMTIME(inode))
c3b2da31 1842 return 0;
20ddee2c 1843
c2050a45 1844 now = current_time(inode);
ce06e0b2
AK
1845 if (!timespec_equal(&inode->i_mtime, &now))
1846 sync_it = S_MTIME;
1da177e4 1847
ce06e0b2
AK
1848 if (!timespec_equal(&inode->i_ctime, &now))
1849 sync_it |= S_CTIME;
870f4817 1850
ce06e0b2
AK
1851 if (IS_I_VERSION(inode))
1852 sync_it |= S_VERSION;
7a224228 1853
ce06e0b2 1854 if (!sync_it)
c3b2da31 1855 return 0;
ce06e0b2
AK
1856
1857 /* Finally allowed to write? Takes lock. */
eb04c282 1858 if (__mnt_want_write_file(file))
c3b2da31 1859 return 0;
ce06e0b2 1860
c3b2da31 1861 ret = update_time(inode, &now, sync_it);
eb04c282 1862 __mnt_drop_write_file(file);
c3b2da31
JB
1863
1864 return ret;
1da177e4 1865}
870f4817 1866EXPORT_SYMBOL(file_update_time);
1da177e4
LT
1867
1868int inode_needs_sync(struct inode *inode)
1869{
1870 if (IS_SYNC(inode))
1871 return 1;
1872 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1873 return 1;
1874 return 0;
1875}
1da177e4
LT
1876EXPORT_SYMBOL(inode_needs_sync);
1877
1da177e4 1878/*
168a9fd6
MS
1879 * If we try to find an inode in the inode hash while it is being
1880 * deleted, we have to wait until the filesystem completes its
1881 * deletion before reporting that it isn't found. This function waits
1882 * until the deletion _might_ have completed. Callers are responsible
1883 * to recheck inode state.
1884 *
eaff8079 1885 * It doesn't matter if I_NEW is not set initially, a call to
250df6ed
DC
1886 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1887 * will DTRT.
1da177e4
LT
1888 */
1889static void __wait_on_freeing_inode(struct inode *inode)
1890{
1891 wait_queue_head_t *wq;
eaff8079
CH
1892 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1893 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1da177e4 1894 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
250df6ed 1895 spin_unlock(&inode->i_lock);
67a23c49 1896 spin_unlock(&inode_hash_lock);
1da177e4
LT
1897 schedule();
1898 finish_wait(wq, &wait.wait);
67a23c49 1899 spin_lock(&inode_hash_lock);
1da177e4
LT
1900}
1901
1da177e4
LT
1902static __initdata unsigned long ihash_entries;
1903static int __init set_ihash_entries(char *str)
1904{
1905 if (!str)
1906 return 0;
1907 ihash_entries = simple_strtoul(str, &str, 0);
1908 return 1;
1909}
1910__setup("ihash_entries=", set_ihash_entries);
1911
1912/*
1913 * Initialize the waitqueues and inode hash table.
1914 */
1915void __init inode_init_early(void)
1916{
074b8517 1917 unsigned int loop;
1da177e4
LT
1918
1919 /* If hashes are distributed across NUMA nodes, defer
1920 * hash allocation until vmalloc space is available.
1921 */
1922 if (hashdist)
1923 return;
1924
1925 inode_hashtable =
1926 alloc_large_system_hash("Inode-cache",
1927 sizeof(struct hlist_head),
1928 ihash_entries,
1929 14,
1930 HASH_EARLY,
1931 &i_hash_shift,
1932 &i_hash_mask,
31fe62b9 1933 0,
1da177e4
LT
1934 0);
1935
074b8517 1936 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1da177e4
LT
1937 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1938}
1939
74bf17cf 1940void __init inode_init(void)
1da177e4 1941{
074b8517 1942 unsigned int loop;
1da177e4
LT
1943
1944 /* inode slab cache */
b0196009
PJ
1945 inode_cachep = kmem_cache_create("inode_cache",
1946 sizeof(struct inode),
1947 0,
1948 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
5d097056 1949 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
20c2df83 1950 init_once);
1da177e4
LT
1951
1952 /* Hash may have been set up in inode_init_early */
1953 if (!hashdist)
1954 return;
1955
1956 inode_hashtable =
1957 alloc_large_system_hash("Inode-cache",
1958 sizeof(struct hlist_head),
1959 ihash_entries,
1960 14,
1961 0,
1962 &i_hash_shift,
1963 &i_hash_mask,
31fe62b9 1964 0,
1da177e4
LT
1965 0);
1966
074b8517 1967 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1da177e4
LT
1968 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1969}
1970
1971void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1972{
1973 inode->i_mode = mode;
1974 if (S_ISCHR(mode)) {
1975 inode->i_fop = &def_chr_fops;
1976 inode->i_rdev = rdev;
1977 } else if (S_ISBLK(mode)) {
1978 inode->i_fop = &def_blk_fops;
1979 inode->i_rdev = rdev;
1980 } else if (S_ISFIFO(mode))
599a0ac1 1981 inode->i_fop = &pipefifo_fops;
1da177e4 1982 else if (S_ISSOCK(mode))
bd9b51e7 1983 ; /* leave it no_open_fops */
1da177e4 1984 else
af0d9ae8
MK
1985 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1986 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1987 inode->i_ino);
1da177e4
LT
1988}
1989EXPORT_SYMBOL(init_special_inode);
a1bd120d
DM
1990
1991/**
eaae668d 1992 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
a1bd120d
DM
1993 * @inode: New inode
1994 * @dir: Directory inode
1995 * @mode: mode of the new inode
1996 */
1997void inode_init_owner(struct inode *inode, const struct inode *dir,
62bb1091 1998 umode_t mode)
a1bd120d
DM
1999{
2000 inode->i_uid = current_fsuid();
2001 if (dir && dir->i_mode & S_ISGID) {
2002 inode->i_gid = dir->i_gid;
2003 if (S_ISDIR(mode))
2004 mode |= S_ISGID;
2005 } else
2006 inode->i_gid = current_fsgid();
2007 inode->i_mode = mode;
2008}
2009EXPORT_SYMBOL(inode_init_owner);
e795b717 2010
2e149670
SH
2011/**
2012 * inode_owner_or_capable - check current task permissions to inode
2013 * @inode: inode being checked
2014 *
23adbe12
AL
2015 * Return true if current either has CAP_FOWNER in a namespace with the
2016 * inode owner uid mapped, or owns the file.
e795b717 2017 */
2e149670 2018bool inode_owner_or_capable(const struct inode *inode)
e795b717 2019{
23adbe12
AL
2020 struct user_namespace *ns;
2021
92361636 2022 if (uid_eq(current_fsuid(), inode->i_uid))
e795b717 2023 return true;
23adbe12
AL
2024
2025 ns = current_user_ns();
2026 if (ns_capable(ns, CAP_FOWNER) && kuid_has_mapping(ns, inode->i_uid))
e795b717
SH
2027 return true;
2028 return false;
2029}
2e149670 2030EXPORT_SYMBOL(inode_owner_or_capable);
1d59d61f
TM
2031
2032/*
2033 * Direct i/o helper functions
2034 */
2035static void __inode_dio_wait(struct inode *inode)
2036{
2037 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2038 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2039
2040 do {
2041 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
2042 if (atomic_read(&inode->i_dio_count))
2043 schedule();
2044 } while (atomic_read(&inode->i_dio_count));
2045 finish_wait(wq, &q.wait);
2046}
2047
2048/**
2049 * inode_dio_wait - wait for outstanding DIO requests to finish
2050 * @inode: inode to wait for
2051 *
2052 * Waits for all pending direct I/O requests to finish so that we can
2053 * proceed with a truncate or equivalent operation.
2054 *
2055 * Must be called under a lock that serializes taking new references
2056 * to i_dio_count, usually by inode->i_mutex.
2057 */
2058void inode_dio_wait(struct inode *inode)
2059{
2060 if (atomic_read(&inode->i_dio_count))
2061 __inode_dio_wait(inode);
2062}
2063EXPORT_SYMBOL(inode_dio_wait);
2064
5f16f322
TT
2065/*
2066 * inode_set_flags - atomically set some inode flags
2067 *
2068 * Note: the caller should be holding i_mutex, or else be sure that
2069 * they have exclusive access to the inode structure (i.e., while the
2070 * inode is being instantiated). The reason for the cmpxchg() loop
2071 * --- which wouldn't be necessary if all code paths which modify
2072 * i_flags actually followed this rule, is that there is at least one
5fa8e0a1
JK
2073 * code path which doesn't today so we use cmpxchg() out of an abundance
2074 * of caution.
5f16f322
TT
2075 *
2076 * In the long run, i_mutex is overkill, and we should probably look
2077 * at using the i_lock spinlock to protect i_flags, and then make sure
2078 * it is so documented in include/linux/fs.h and that all code follows
2079 * the locking convention!!
2080 */
2081void inode_set_flags(struct inode *inode, unsigned int flags,
2082 unsigned int mask)
2083{
2084 unsigned int old_flags, new_flags;
2085
2086 WARN_ON_ONCE(flags & ~mask);
2087 do {
2088 old_flags = ACCESS_ONCE(inode->i_flags);
2089 new_flags = (old_flags & ~mask) | flags;
2090 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2091 new_flags) != old_flags));
2092}
2093EXPORT_SYMBOL(inode_set_flags);
21fc61c7
AV
2094
2095void inode_nohighmem(struct inode *inode)
2096{
2097 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2098}
2099EXPORT_SYMBOL(inode_nohighmem);
3cd88666
DD
2100
2101/**
2102 * current_time - Return FS time
2103 * @inode: inode.
2104 *
2105 * Return the current time truncated to the time granularity supported by
2106 * the fs.
2107 *
2108 * Note that inode and inode->sb cannot be NULL.
2109 * Otherwise, the function warns and returns time without truncation.
2110 */
2111struct timespec current_time(struct inode *inode)
2112{
2113 struct timespec now = current_kernel_time();
2114
2115 if (unlikely(!inode->i_sb)) {
2116 WARN(1, "current_time() called with uninitialized super_block in the inode");
2117 return now;
2118 }
2119
2120 return timespec_trunc(now, inode->i_sb->s_time_gran);
2121}
2122EXPORT_SYMBOL(current_time);