]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - fs/inode.c
Merge tag 'sound-4.19-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
[mirror_ubuntu-eoan-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>
ae5e165d 21#include <linux/iversion.h>
0ae45f63 22#include <trace/events/writeback.h>
a66979ab 23#include "internal.h"
1da177e4 24
250df6ed 25/*
4b4563dc 26 * Inode locking rules:
250df6ed
DC
27 *
28 * inode->i_lock protects:
29 * inode->i_state, inode->i_hash, __iget()
bc3b14cb 30 * Inode LRU list locks protect:
98b745c6 31 * inode->i_sb->s_inode_lru, inode->i_lru
74278da9
DC
32 * inode->i_sb->s_inode_list_lock protects:
33 * inode->i_sb->s_inodes, inode->i_sb_list
f758eeab 34 * bdi->wb.list_lock protects:
c7f54084 35 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
67a23c49
DC
36 * inode_hash_lock protects:
37 * inode_hashtable, inode->i_hash
250df6ed
DC
38 *
39 * Lock ordering:
55fa6091 40 *
74278da9 41 * inode->i_sb->s_inode_list_lock
55fa6091 42 * inode->i_lock
bc3b14cb 43 * Inode LRU list locks
a66979ab 44 *
f758eeab 45 * bdi->wb.list_lock
a66979ab 46 * inode->i_lock
67a23c49
DC
47 *
48 * inode_hash_lock
74278da9 49 * inode->i_sb->s_inode_list_lock
67a23c49
DC
50 * inode->i_lock
51 *
52 * iunique_lock
53 * inode_hash_lock
250df6ed
DC
54 */
55
fa3536cc
ED
56static unsigned int i_hash_mask __read_mostly;
57static unsigned int i_hash_shift __read_mostly;
67a23c49
DC
58static struct hlist_head *inode_hashtable __read_mostly;
59static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
1da177e4 60
7dcda1c9
JA
61/*
62 * Empty aops. Can be used for the cases where the user does not
63 * define any of the address_space operations.
64 */
65const struct address_space_operations empty_aops = {
66};
67EXPORT_SYMBOL(empty_aops);
68
1da177e4
LT
69/*
70 * Statistics gathering..
71 */
72struct inodes_stat_t inodes_stat;
73
3942c07c
GC
74static DEFINE_PER_CPU(unsigned long, nr_inodes);
75static DEFINE_PER_CPU(unsigned long, nr_unused);
cffbc8aa 76
6b3304b5 77static struct kmem_cache *inode_cachep __read_mostly;
1da177e4 78
3942c07c 79static long get_nr_inodes(void)
cffbc8aa 80{
3e880fb5 81 int i;
3942c07c 82 long sum = 0;
3e880fb5
NP
83 for_each_possible_cpu(i)
84 sum += per_cpu(nr_inodes, i);
85 return sum < 0 ? 0 : sum;
cffbc8aa
DC
86}
87
3942c07c 88static inline long get_nr_inodes_unused(void)
cffbc8aa 89{
fcb94f72 90 int i;
3942c07c 91 long sum = 0;
fcb94f72
DC
92 for_each_possible_cpu(i)
93 sum += per_cpu(nr_unused, i);
94 return sum < 0 ? 0 : sum;
cffbc8aa
DC
95}
96
3942c07c 97long get_nr_dirty_inodes(void)
cffbc8aa 98{
3e880fb5 99 /* not actually dirty inodes, but a wild approximation */
3942c07c 100 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
cffbc8aa 101 return nr_dirty > 0 ? nr_dirty : 0;
cffbc8aa
DC
102}
103
104/*
105 * Handle nr_inode sysctl
106 */
107#ifdef CONFIG_SYSCTL
1f7e0616 108int proc_nr_inodes(struct ctl_table *table, int write,
cffbc8aa
DC
109 void __user *buffer, size_t *lenp, loff_t *ppos)
110{
111 inodes_stat.nr_inodes = get_nr_inodes();
fcb94f72 112 inodes_stat.nr_unused = get_nr_inodes_unused();
3942c07c 113 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
cffbc8aa
DC
114}
115#endif
116
bd9b51e7
AV
117static int no_open(struct inode *inode, struct file *file)
118{
119 return -ENXIO;
120}
121
2cb1599f 122/**
6e7c2b4d 123 * inode_init_always - perform inode structure initialisation
0bc02f3f
RD
124 * @sb: superblock inode belongs to
125 * @inode: inode to initialise
2cb1599f
DC
126 *
127 * These are initializations that need to be done on every inode
128 * allocation as the fields are not initialised by slab allocation.
129 */
54e34621 130int inode_init_always(struct super_block *sb, struct inode *inode)
1da177e4 131{
6e1d5dcc 132 static const struct inode_operations empty_iops;
bd9b51e7 133 static const struct file_operations no_open_fops = {.open = no_open};
6b3304b5 134 struct address_space *const mapping = &inode->i_data;
2cb1599f
DC
135
136 inode->i_sb = sb;
137 inode->i_blkbits = sb->s_blocksize_bits;
138 inode->i_flags = 0;
139 atomic_set(&inode->i_count, 1);
140 inode->i_op = &empty_iops;
bd9b51e7 141 inode->i_fop = &no_open_fops;
a78ef704 142 inode->__i_nlink = 1;
3ddcd056 143 inode->i_opflags = 0;
d0a5b995
AG
144 if (sb->s_xattr)
145 inode->i_opflags |= IOP_XATTR;
92361636
EB
146 i_uid_write(inode, 0);
147 i_gid_write(inode, 0);
2cb1599f
DC
148 atomic_set(&inode->i_writecount, 0);
149 inode->i_size = 0;
c75b1d94 150 inode->i_write_hint = WRITE_LIFE_NOT_SET;
2cb1599f
DC
151 inode->i_blocks = 0;
152 inode->i_bytes = 0;
153 inode->i_generation = 0;
2cb1599f
DC
154 inode->i_pipe = NULL;
155 inode->i_bdev = NULL;
156 inode->i_cdev = NULL;
61ba64fc 157 inode->i_link = NULL;
84e710da 158 inode->i_dir_seq = 0;
2cb1599f
DC
159 inode->i_rdev = 0;
160 inode->dirtied_when = 0;
6146f0d5 161
3d65ae46
TE
162#ifdef CONFIG_CGROUP_WRITEBACK
163 inode->i_wb_frn_winner = 0;
164 inode->i_wb_frn_avg_time = 0;
165 inode->i_wb_frn_history = 0;
166#endif
167
6146f0d5 168 if (security_inode_alloc(inode))
54e34621 169 goto out;
2cb1599f
DC
170 spin_lock_init(&inode->i_lock);
171 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
172
9902af79
AV
173 init_rwsem(&inode->i_rwsem);
174 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
2cb1599f 175
bd5fe6c5 176 atomic_set(&inode->i_dio_count, 0);
2cb1599f
DC
177
178 mapping->a_ops = &empty_aops;
179 mapping->host = inode;
180 mapping->flags = 0;
829bc787 181 mapping->wb_err = 0;
4bb5f5d9 182 atomic_set(&mapping->i_mmap_writable, 0);
3c1d4378 183 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
252aa6f5 184 mapping->private_data = NULL;
2cb1599f 185 mapping->writeback_index = 0;
2cb1599f
DC
186 inode->i_private = NULL;
187 inode->i_mapping = mapping;
b3d9b7a3 188 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
f19d4a8f
AV
189#ifdef CONFIG_FS_POSIX_ACL
190 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
191#endif
2cb1599f 192
3be25f49
EP
193#ifdef CONFIG_FSNOTIFY
194 inode->i_fsnotify_mask = 0;
195#endif
4a075e39 196 inode->i_flctx = NULL;
3e880fb5 197 this_cpu_inc(nr_inodes);
cffbc8aa 198
54e34621 199 return 0;
54e34621
CH
200out:
201 return -ENOMEM;
1da177e4 202}
2cb1599f
DC
203EXPORT_SYMBOL(inode_init_always);
204
205static struct inode *alloc_inode(struct super_block *sb)
206{
207 struct inode *inode;
208
209 if (sb->s_op->alloc_inode)
210 inode = sb->s_op->alloc_inode(sb);
211 else
212 inode = kmem_cache_alloc(inode_cachep, GFP_KERNEL);
213
54e34621
CH
214 if (!inode)
215 return NULL;
216
217 if (unlikely(inode_init_always(sb, inode))) {
218 if (inode->i_sb->s_op->destroy_inode)
219 inode->i_sb->s_op->destroy_inode(inode);
220 else
221 kmem_cache_free(inode_cachep, inode);
222 return NULL;
223 }
224
225 return inode;
2cb1599f 226}
1da177e4 227
ff0c7d15
NP
228void free_inode_nonrcu(struct inode *inode)
229{
230 kmem_cache_free(inode_cachep, inode);
231}
232EXPORT_SYMBOL(free_inode_nonrcu);
233
2e00c97e 234void __destroy_inode(struct inode *inode)
1da177e4 235{
b7542f8c 236 BUG_ON(inode_has_buffers(inode));
52ebea74 237 inode_detach_wb(inode);
1da177e4 238 security_inode_free(inode);
3be25f49 239 fsnotify_inode_delete(inode);
f27a0fe0 240 locks_free_lock_context(inode);
7ada4db8
MS
241 if (!inode->i_nlink) {
242 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
243 atomic_long_dec(&inode->i_sb->s_remove_count);
244 }
245
f19d4a8f 246#ifdef CONFIG_FS_POSIX_ACL
b8a7a3a6 247 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
f19d4a8f 248 posix_acl_release(inode->i_acl);
b8a7a3a6 249 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
f19d4a8f
AV
250 posix_acl_release(inode->i_default_acl);
251#endif
3e880fb5 252 this_cpu_dec(nr_inodes);
2e00c97e
CH
253}
254EXPORT_SYMBOL(__destroy_inode);
255
fa0d7e3d
NP
256static void i_callback(struct rcu_head *head)
257{
258 struct inode *inode = container_of(head, struct inode, i_rcu);
fa0d7e3d
NP
259 kmem_cache_free(inode_cachep, inode);
260}
261
56b0dacf 262static void destroy_inode(struct inode *inode)
2e00c97e 263{
7ccf19a8 264 BUG_ON(!list_empty(&inode->i_lru));
2e00c97e 265 __destroy_inode(inode);
1da177e4
LT
266 if (inode->i_sb->s_op->destroy_inode)
267 inode->i_sb->s_op->destroy_inode(inode);
268 else
fa0d7e3d 269 call_rcu(&inode->i_rcu, i_callback);
1da177e4 270}
1da177e4 271
7ada4db8
MS
272/**
273 * drop_nlink - directly drop an inode's link count
274 * @inode: inode
275 *
276 * This is a low-level filesystem helper to replace any
277 * direct filesystem manipulation of i_nlink. In cases
278 * where we are attempting to track writes to the
279 * filesystem, a decrement to zero means an imminent
280 * write when the file is truncated and actually unlinked
281 * on the filesystem.
282 */
283void drop_nlink(struct inode *inode)
284{
285 WARN_ON(inode->i_nlink == 0);
286 inode->__i_nlink--;
287 if (!inode->i_nlink)
288 atomic_long_inc(&inode->i_sb->s_remove_count);
289}
290EXPORT_SYMBOL(drop_nlink);
291
292/**
293 * clear_nlink - directly zero an inode's link count
294 * @inode: inode
295 *
296 * This is a low-level filesystem helper to replace any
297 * direct filesystem manipulation of i_nlink. See
298 * drop_nlink() for why we care about i_nlink hitting zero.
299 */
300void clear_nlink(struct inode *inode)
301{
302 if (inode->i_nlink) {
303 inode->__i_nlink = 0;
304 atomic_long_inc(&inode->i_sb->s_remove_count);
305 }
306}
307EXPORT_SYMBOL(clear_nlink);
308
309/**
310 * set_nlink - directly set an inode's link count
311 * @inode: inode
312 * @nlink: new nlink (should be non-zero)
313 *
314 * This is a low-level filesystem helper to replace any
315 * direct filesystem manipulation of i_nlink.
316 */
317void set_nlink(struct inode *inode, unsigned int nlink)
318{
319 if (!nlink) {
7ada4db8
MS
320 clear_nlink(inode);
321 } else {
322 /* Yes, some filesystems do change nlink from zero to one */
323 if (inode->i_nlink == 0)
324 atomic_long_dec(&inode->i_sb->s_remove_count);
325
326 inode->__i_nlink = nlink;
327 }
328}
329EXPORT_SYMBOL(set_nlink);
330
331/**
332 * inc_nlink - directly increment an inode's link count
333 * @inode: inode
334 *
335 * This is a low-level filesystem helper to replace any
336 * direct filesystem manipulation of i_nlink. Currently,
337 * it is only here for parity with dec_nlink().
338 */
339void inc_nlink(struct inode *inode)
340{
f4e0c30c
AV
341 if (unlikely(inode->i_nlink == 0)) {
342 WARN_ON(!(inode->i_state & I_LINKABLE));
7ada4db8 343 atomic_long_dec(&inode->i_sb->s_remove_count);
f4e0c30c 344 }
7ada4db8
MS
345
346 inode->__i_nlink++;
347}
348EXPORT_SYMBOL(inc_nlink);
349
ae23395d 350static void __address_space_init_once(struct address_space *mapping)
2aa15890 351{
b93b0163 352 INIT_RADIX_TREE(&mapping->i_pages, GFP_ATOMIC | __GFP_ACCOUNT);
c8c06efa 353 init_rwsem(&mapping->i_mmap_rwsem);
2aa15890
MS
354 INIT_LIST_HEAD(&mapping->private_list);
355 spin_lock_init(&mapping->private_lock);
f808c13f 356 mapping->i_mmap = RB_ROOT_CACHED;
2aa15890 357}
ae23395d
DC
358
359void address_space_init_once(struct address_space *mapping)
360{
361 memset(mapping, 0, sizeof(*mapping));
362 __address_space_init_once(mapping);
363}
2aa15890
MS
364EXPORT_SYMBOL(address_space_init_once);
365
1da177e4
LT
366/*
367 * These are initializations that only need to be done
368 * once, because the fields are idempotent across use
369 * of the inode, so let the slab aware of that.
370 */
371void inode_init_once(struct inode *inode)
372{
373 memset(inode, 0, sizeof(*inode));
374 INIT_HLIST_NODE(&inode->i_hash);
1da177e4 375 INIT_LIST_HEAD(&inode->i_devices);
c7f54084 376 INIT_LIST_HEAD(&inode->i_io_list);
6c60d2b5 377 INIT_LIST_HEAD(&inode->i_wb_list);
7ccf19a8 378 INIT_LIST_HEAD(&inode->i_lru);
ae23395d 379 __address_space_init_once(&inode->i_data);
1da177e4
LT
380 i_size_ordered_init(inode);
381}
1da177e4
LT
382EXPORT_SYMBOL(inode_init_once);
383
51cc5068 384static void init_once(void *foo)
1da177e4 385{
6b3304b5 386 struct inode *inode = (struct inode *) foo;
1da177e4 387
a35afb83 388 inode_init_once(inode);
1da177e4
LT
389}
390
391/*
250df6ed 392 * inode->i_lock must be held
1da177e4 393 */
6b3304b5 394void __iget(struct inode *inode)
1da177e4 395{
9e38d86f
NP
396 atomic_inc(&inode->i_count);
397}
2e147f1e 398
7de9c6ee
AV
399/*
400 * get additional reference to inode; caller must already hold one.
401 */
402void ihold(struct inode *inode)
403{
404 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
405}
406EXPORT_SYMBOL(ihold);
407
9e38d86f
NP
408static void inode_lru_list_add(struct inode *inode)
409{
bc3b14cb 410 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
fcb94f72 411 this_cpu_inc(nr_unused);
563f4001
JB
412 else
413 inode->i_state |= I_REFERENCED;
9e38d86f 414}
2e147f1e 415
4eff96dd
JK
416/*
417 * Add inode to LRU if needed (inode is unused and clean).
418 *
419 * Needs inode->i_lock held.
420 */
421void inode_add_lru(struct inode *inode)
422{
0ae45f63
TT
423 if (!(inode->i_state & (I_DIRTY_ALL | I_SYNC |
424 I_FREEING | I_WILL_FREE)) &&
1751e8a6 425 !atomic_read(&inode->i_count) && inode->i_sb->s_flags & SB_ACTIVE)
4eff96dd
JK
426 inode_lru_list_add(inode);
427}
428
429
9e38d86f
NP
430static void inode_lru_list_del(struct inode *inode)
431{
bc3b14cb
DC
432
433 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
fcb94f72 434 this_cpu_dec(nr_unused);
1da177e4
LT
435}
436
646ec461
CH
437/**
438 * inode_sb_list_add - add inode to the superblock list of inodes
439 * @inode: inode to add
440 */
441void inode_sb_list_add(struct inode *inode)
442{
74278da9 443 spin_lock(&inode->i_sb->s_inode_list_lock);
55fa6091 444 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
74278da9 445 spin_unlock(&inode->i_sb->s_inode_list_lock);
646ec461
CH
446}
447EXPORT_SYMBOL_GPL(inode_sb_list_add);
448
55fa6091 449static inline void inode_sb_list_del(struct inode *inode)
646ec461 450{
a209dfc7 451 if (!list_empty(&inode->i_sb_list)) {
74278da9 452 spin_lock(&inode->i_sb->s_inode_list_lock);
a209dfc7 453 list_del_init(&inode->i_sb_list);
74278da9 454 spin_unlock(&inode->i_sb->s_inode_list_lock);
a209dfc7 455 }
646ec461
CH
456}
457
4c51acbc
DC
458static unsigned long hash(struct super_block *sb, unsigned long hashval)
459{
460 unsigned long tmp;
461
462 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
463 L1_CACHE_BYTES;
4b4563dc
CH
464 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
465 return tmp & i_hash_mask;
4c51acbc
DC
466}
467
468/**
469 * __insert_inode_hash - hash an inode
470 * @inode: unhashed inode
471 * @hashval: unsigned long value used to locate this object in the
472 * inode_hashtable.
473 *
474 * Add an inode to the inode hash for this superblock.
475 */
476void __insert_inode_hash(struct inode *inode, unsigned long hashval)
477{
646ec461
CH
478 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
479
67a23c49 480 spin_lock(&inode_hash_lock);
250df6ed 481 spin_lock(&inode->i_lock);
646ec461 482 hlist_add_head(&inode->i_hash, b);
250df6ed 483 spin_unlock(&inode->i_lock);
67a23c49 484 spin_unlock(&inode_hash_lock);
4c51acbc
DC
485}
486EXPORT_SYMBOL(__insert_inode_hash);
487
4c51acbc 488/**
f2ee7abf 489 * __remove_inode_hash - remove an inode from the hash
4c51acbc
DC
490 * @inode: inode to unhash
491 *
492 * Remove an inode from the superblock.
493 */
f2ee7abf 494void __remove_inode_hash(struct inode *inode)
4c51acbc 495{
67a23c49 496 spin_lock(&inode_hash_lock);
250df6ed 497 spin_lock(&inode->i_lock);
4c51acbc 498 hlist_del_init(&inode->i_hash);
250df6ed 499 spin_unlock(&inode->i_lock);
67a23c49 500 spin_unlock(&inode_hash_lock);
4c51acbc 501}
f2ee7abf 502EXPORT_SYMBOL(__remove_inode_hash);
4c51acbc 503
dbd5768f 504void clear_inode(struct inode *inode)
b0683aa6 505{
08142579 506 /*
b93b0163 507 * We have to cycle the i_pages lock here because reclaim can be in the
08142579 508 * process of removing the last page (in __delete_from_page_cache())
b93b0163 509 * and we must not free the mapping under it.
08142579 510 */
b93b0163 511 xa_lock_irq(&inode->i_data.i_pages);
b0683aa6 512 BUG_ON(inode->i_data.nrpages);
f9fe48be 513 BUG_ON(inode->i_data.nrexceptional);
b93b0163 514 xa_unlock_irq(&inode->i_data.i_pages);
b0683aa6
AV
515 BUG_ON(!list_empty(&inode->i_data.private_list));
516 BUG_ON(!(inode->i_state & I_FREEING));
517 BUG_ON(inode->i_state & I_CLEAR);
6c60d2b5 518 BUG_ON(!list_empty(&inode->i_wb_list));
fa0d7e3d 519 /* don't need i_lock here, no concurrent mods to i_state */
b0683aa6
AV
520 inode->i_state = I_FREEING | I_CLEAR;
521}
dbd5768f 522EXPORT_SYMBOL(clear_inode);
b0683aa6 523
b2b2af8e
DC
524/*
525 * Free the inode passed in, removing it from the lists it is still connected
526 * to. We remove any pages still attached to the inode and wait for any IO that
527 * is still in progress before finally destroying the inode.
528 *
529 * An inode must already be marked I_FREEING so that we avoid the inode being
530 * moved back onto lists if we race with other code that manipulates the lists
531 * (e.g. writeback_single_inode). The caller is responsible for setting this.
532 *
533 * An inode must already be removed from the LRU list before being evicted from
534 * the cache. This should occur atomically with setting the I_FREEING state
535 * flag, so no inodes here should ever be on the LRU when being evicted.
536 */
644da596 537static void evict(struct inode *inode)
b4272d4c
AV
538{
539 const struct super_operations *op = inode->i_sb->s_op;
540
b2b2af8e
DC
541 BUG_ON(!(inode->i_state & I_FREEING));
542 BUG_ON(!list_empty(&inode->i_lru));
543
c7f54084
DC
544 if (!list_empty(&inode->i_io_list))
545 inode_io_list_del(inode);
b12362bd 546
55fa6091
DC
547 inode_sb_list_del(inode);
548
169ebd90
JK
549 /*
550 * Wait for flusher thread to be done with the inode so that filesystem
551 * does not start destroying it while writeback is still running. Since
552 * the inode has I_FREEING set, flusher thread won't start new work on
553 * the inode. We just have to wait for running writeback to finish.
554 */
555 inode_wait_for_writeback(inode);
7994e6f7 556
be7ce416
AV
557 if (op->evict_inode) {
558 op->evict_inode(inode);
b4272d4c 559 } else {
91b0abe3 560 truncate_inode_pages_final(&inode->i_data);
dbd5768f 561 clear_inode(inode);
b4272d4c 562 }
661074e9
AV
563 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
564 bd_forget(inode);
565 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
566 cd_forget(inode);
b2b2af8e
DC
567
568 remove_inode_hash(inode);
569
570 spin_lock(&inode->i_lock);
571 wake_up_bit(&inode->i_state, __I_NEW);
572 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
573 spin_unlock(&inode->i_lock);
574
575 destroy_inode(inode);
b4272d4c
AV
576}
577
1da177e4
LT
578/*
579 * dispose_list - dispose of the contents of a local list
580 * @head: the head of the list to free
581 *
582 * Dispose-list gets a local list with local inodes in it, so it doesn't
583 * need to worry about list corruption and SMP locks.
584 */
585static void dispose_list(struct list_head *head)
586{
1da177e4
LT
587 while (!list_empty(head)) {
588 struct inode *inode;
589
7ccf19a8
NP
590 inode = list_first_entry(head, struct inode, i_lru);
591 list_del_init(&inode->i_lru);
1da177e4 592
644da596 593 evict(inode);
ac05fbb4 594 cond_resched();
1da177e4 595 }
1da177e4
LT
596}
597
63997e98
AV
598/**
599 * evict_inodes - evict all evictable inodes for a superblock
600 * @sb: superblock to operate on
601 *
602 * Make sure that no inodes with zero refcount are retained. This is
1751e8a6 603 * called by superblock shutdown after having SB_ACTIVE flag removed,
63997e98
AV
604 * so any inode reaching zero refcount during or after that call will
605 * be immediately evicted.
1da177e4 606 */
63997e98 607void evict_inodes(struct super_block *sb)
1da177e4 608{
63997e98
AV
609 struct inode *inode, *next;
610 LIST_HEAD(dispose);
1da177e4 611
ac05fbb4 612again:
74278da9 613 spin_lock(&sb->s_inode_list_lock);
63997e98
AV
614 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
615 if (atomic_read(&inode->i_count))
aabb8fdb 616 continue;
250df6ed
DC
617
618 spin_lock(&inode->i_lock);
619 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
620 spin_unlock(&inode->i_lock);
1da177e4 621 continue;
250df6ed 622 }
63997e98
AV
623
624 inode->i_state |= I_FREEING;
02afc410 625 inode_lru_list_del(inode);
250df6ed 626 spin_unlock(&inode->i_lock);
02afc410 627 list_add(&inode->i_lru, &dispose);
ac05fbb4
JB
628
629 /*
630 * We can have a ton of inodes to evict at unmount time given
631 * enough memory, check to see if we need to go to sleep for a
632 * bit so we don't livelock.
633 */
634 if (need_resched()) {
635 spin_unlock(&sb->s_inode_list_lock);
636 cond_resched();
637 dispose_list(&dispose);
638 goto again;
639 }
1da177e4 640 }
74278da9 641 spin_unlock(&sb->s_inode_list_lock);
63997e98
AV
642
643 dispose_list(&dispose);
1da177e4 644}
799ea9e9 645EXPORT_SYMBOL_GPL(evict_inodes);
1da177e4 646
1da177e4 647/**
a0318786
CH
648 * invalidate_inodes - attempt to free all inodes on a superblock
649 * @sb: superblock to operate on
93b270f7 650 * @kill_dirty: flag to guide handling of dirty inodes
1da177e4 651 *
a0318786
CH
652 * Attempts to free all inodes for a given superblock. If there were any
653 * busy inodes return a non-zero value, else zero.
93b270f7
N
654 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
655 * them as busy.
1da177e4 656 */
93b270f7 657int invalidate_inodes(struct super_block *sb, bool kill_dirty)
1da177e4 658{
cffbc8aa 659 int busy = 0;
a0318786
CH
660 struct inode *inode, *next;
661 LIST_HEAD(dispose);
1da177e4 662
74278da9 663 spin_lock(&sb->s_inode_list_lock);
a0318786 664 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
250df6ed
DC
665 spin_lock(&inode->i_lock);
666 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
667 spin_unlock(&inode->i_lock);
aabb8fdb 668 continue;
250df6ed 669 }
0ae45f63 670 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
250df6ed 671 spin_unlock(&inode->i_lock);
93b270f7
N
672 busy = 1;
673 continue;
674 }
99a38919 675 if (atomic_read(&inode->i_count)) {
250df6ed 676 spin_unlock(&inode->i_lock);
99a38919 677 busy = 1;
1da177e4
LT
678 continue;
679 }
99a38919 680
99a38919 681 inode->i_state |= I_FREEING;
02afc410 682 inode_lru_list_del(inode);
250df6ed 683 spin_unlock(&inode->i_lock);
02afc410 684 list_add(&inode->i_lru, &dispose);
1da177e4 685 }
74278da9 686 spin_unlock(&sb->s_inode_list_lock);
1da177e4 687
a0318786 688 dispose_list(&dispose);
1da177e4
LT
689
690 return busy;
691}
1da177e4 692
1da177e4 693/*
bc3b14cb 694 * Isolate the inode from the LRU in preparation for freeing it.
1da177e4
LT
695 *
696 * Any inodes which are pinned purely because of attached pagecache have their
9e38d86f
NP
697 * pagecache removed. If the inode has metadata buffers attached to
698 * mapping->private_list then try to remove them.
1da177e4 699 *
9e38d86f
NP
700 * If the inode has the I_REFERENCED flag set, then it means that it has been
701 * used recently - the flag is set in iput_final(). When we encounter such an
702 * inode, clear the flag and move it to the back of the LRU so it gets another
703 * pass through the LRU before it gets reclaimed. This is necessary because of
704 * the fact we are doing lazy LRU updates to minimise lock contention so the
705 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
706 * with this flag set because they are the inodes that are out of order.
1da177e4 707 */
3f97b163
VD
708static enum lru_status inode_lru_isolate(struct list_head *item,
709 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1da177e4 710{
bc3b14cb
DC
711 struct list_head *freeable = arg;
712 struct inode *inode = container_of(item, struct inode, i_lru);
1da177e4 713
bc3b14cb
DC
714 /*
715 * we are inverting the lru lock/inode->i_lock here, so use a trylock.
716 * If we fail to get the lock, just skip it.
717 */
718 if (!spin_trylock(&inode->i_lock))
719 return LRU_SKIP;
1da177e4 720
bc3b14cb
DC
721 /*
722 * Referenced or dirty inodes are still in use. Give them another pass
723 * through the LRU as we canot reclaim them now.
724 */
725 if (atomic_read(&inode->i_count) ||
726 (inode->i_state & ~I_REFERENCED)) {
3f97b163 727 list_lru_isolate(lru, &inode->i_lru);
bc3b14cb
DC
728 spin_unlock(&inode->i_lock);
729 this_cpu_dec(nr_unused);
730 return LRU_REMOVED;
731 }
1da177e4 732
bc3b14cb
DC
733 /* recently referenced inodes get one more pass */
734 if (inode->i_state & I_REFERENCED) {
735 inode->i_state &= ~I_REFERENCED;
736 spin_unlock(&inode->i_lock);
737 return LRU_ROTATE;
738 }
1da177e4 739
bc3b14cb
DC
740 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
741 __iget(inode);
742 spin_unlock(&inode->i_lock);
743 spin_unlock(lru_lock);
744 if (remove_inode_buffers(inode)) {
745 unsigned long reap;
746 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
747 if (current_is_kswapd())
748 __count_vm_events(KSWAPD_INODESTEAL, reap);
749 else
750 __count_vm_events(PGINODESTEAL, reap);
751 if (current->reclaim_state)
752 current->reclaim_state->reclaimed_slab += reap;
02afc410 753 }
bc3b14cb
DC
754 iput(inode);
755 spin_lock(lru_lock);
756 return LRU_RETRY;
757 }
02afc410 758
bc3b14cb
DC
759 WARN_ON(inode->i_state & I_NEW);
760 inode->i_state |= I_FREEING;
3f97b163 761 list_lru_isolate_move(lru, &inode->i_lru, freeable);
bc3b14cb 762 spin_unlock(&inode->i_lock);
9e38d86f 763
bc3b14cb
DC
764 this_cpu_dec(nr_unused);
765 return LRU_REMOVED;
766}
7ccf19a8 767
bc3b14cb
DC
768/*
769 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
770 * This is called from the superblock shrinker function with a number of inodes
771 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
772 * then are freed outside inode_lock by dispose_list().
773 */
503c358c 774long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
bc3b14cb
DC
775{
776 LIST_HEAD(freeable);
777 long freed;
1da177e4 778
503c358c
VD
779 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
780 inode_lru_isolate, &freeable);
1da177e4 781 dispose_list(&freeable);
0a234c6d 782 return freed;
1da177e4
LT
783}
784
1da177e4
LT
785static void __wait_on_freeing_inode(struct inode *inode);
786/*
787 * Called with the inode lock held.
1da177e4 788 */
6b3304b5
MK
789static struct inode *find_inode(struct super_block *sb,
790 struct hlist_head *head,
791 int (*test)(struct inode *, void *),
792 void *data)
1da177e4 793{
6b3304b5 794 struct inode *inode = NULL;
1da177e4
LT
795
796repeat:
b67bfe0d 797 hlist_for_each_entry(inode, head, i_hash) {
5a3cd992 798 if (inode->i_sb != sb)
1da177e4 799 continue;
5a3cd992 800 if (!test(inode, data))
1da177e4 801 continue;
5a3cd992 802 spin_lock(&inode->i_lock);
a4ffdde6 803 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1da177e4
LT
804 __wait_on_freeing_inode(inode);
805 goto repeat;
806 }
c2b6d621
AV
807 if (unlikely(inode->i_state & I_CREATING)) {
808 spin_unlock(&inode->i_lock);
809 return ERR_PTR(-ESTALE);
810 }
f7899bd5 811 __iget(inode);
250df6ed 812 spin_unlock(&inode->i_lock);
f7899bd5 813 return inode;
1da177e4 814 }
f7899bd5 815 return NULL;
1da177e4
LT
816}
817
818/*
819 * find_inode_fast is the fast path version of find_inode, see the comment at
820 * iget_locked for details.
821 */
6b3304b5
MK
822static struct inode *find_inode_fast(struct super_block *sb,
823 struct hlist_head *head, unsigned long ino)
1da177e4 824{
6b3304b5 825 struct inode *inode = NULL;
1da177e4
LT
826
827repeat:
b67bfe0d 828 hlist_for_each_entry(inode, head, i_hash) {
5a3cd992 829 if (inode->i_ino != ino)
1da177e4 830 continue;
5a3cd992 831 if (inode->i_sb != sb)
1da177e4 832 continue;
5a3cd992 833 spin_lock(&inode->i_lock);
a4ffdde6 834 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1da177e4
LT
835 __wait_on_freeing_inode(inode);
836 goto repeat;
837 }
c2b6d621
AV
838 if (unlikely(inode->i_state & I_CREATING)) {
839 spin_unlock(&inode->i_lock);
840 return ERR_PTR(-ESTALE);
841 }
f7899bd5 842 __iget(inode);
250df6ed 843 spin_unlock(&inode->i_lock);
f7899bd5 844 return inode;
1da177e4 845 }
f7899bd5 846 return NULL;
8290c35f
DC
847}
848
f991bd2e
ED
849/*
850 * Each cpu owns a range of LAST_INO_BATCH numbers.
851 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
852 * to renew the exhausted range.
8290c35f 853 *
f991bd2e
ED
854 * This does not significantly increase overflow rate because every CPU can
855 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
856 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
857 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
858 * overflow rate by 2x, which does not seem too significant.
859 *
860 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
861 * error if st_ino won't fit in target struct field. Use 32bit counter
862 * here to attempt to avoid that.
8290c35f 863 */
f991bd2e
ED
864#define LAST_INO_BATCH 1024
865static DEFINE_PER_CPU(unsigned int, last_ino);
866
85fe4025 867unsigned int get_next_ino(void)
8290c35f 868{
f991bd2e
ED
869 unsigned int *p = &get_cpu_var(last_ino);
870 unsigned int res = *p;
8290c35f 871
f991bd2e
ED
872#ifdef CONFIG_SMP
873 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
874 static atomic_t shared_last_ino;
875 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
876
877 res = next - LAST_INO_BATCH;
878 }
879#endif
880
2adc376c
CM
881 res++;
882 /* get_next_ino should not provide a 0 inode number */
883 if (unlikely(!res))
884 res++;
885 *p = res;
f991bd2e
ED
886 put_cpu_var(last_ino);
887 return res;
8290c35f 888}
85fe4025 889EXPORT_SYMBOL(get_next_ino);
8290c35f 890
a209dfc7
ED
891/**
892 * new_inode_pseudo - obtain an inode
893 * @sb: superblock
894 *
895 * Allocates a new inode for given superblock.
896 * Inode wont be chained in superblock s_inodes list
897 * This means :
898 * - fs can't be unmount
899 * - quotas, fsnotify, writeback can't work
900 */
901struct inode *new_inode_pseudo(struct super_block *sb)
902{
903 struct inode *inode = alloc_inode(sb);
904
905 if (inode) {
906 spin_lock(&inode->i_lock);
907 inode->i_state = 0;
908 spin_unlock(&inode->i_lock);
909 INIT_LIST_HEAD(&inode->i_sb_list);
910 }
911 return inode;
912}
913
1da177e4
LT
914/**
915 * new_inode - obtain an inode
916 * @sb: superblock
917 *
769848c0 918 * Allocates a new inode for given superblock. The default gfp_mask
3c1d4378 919 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
769848c0
MG
920 * If HIGHMEM pages are unsuitable or it is known that pages allocated
921 * for the page cache are not reclaimable or migratable,
922 * mapping_set_gfp_mask() must be called with suitable flags on the
923 * newly created inode's mapping
924 *
1da177e4
LT
925 */
926struct inode *new_inode(struct super_block *sb)
927{
6b3304b5 928 struct inode *inode;
1da177e4 929
74278da9 930 spin_lock_prefetch(&sb->s_inode_list_lock);
6b3304b5 931
a209dfc7
ED
932 inode = new_inode_pseudo(sb);
933 if (inode)
55fa6091 934 inode_sb_list_add(inode);
1da177e4
LT
935 return inode;
936}
1da177e4
LT
937EXPORT_SYMBOL(new_inode);
938
14358e6d 939#ifdef CONFIG_DEBUG_LOCK_ALLOC
e096d0c7
JB
940void lockdep_annotate_inode_mutex_key(struct inode *inode)
941{
a3314a0e 942 if (S_ISDIR(inode->i_mode)) {
1e89a5e1
PZ
943 struct file_system_type *type = inode->i_sb->s_type;
944
9a7aa12f 945 /* Set new key only if filesystem hasn't already changed it */
9902af79 946 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
9a7aa12f
JK
947 /*
948 * ensure nobody is actually holding i_mutex
949 */
9902af79
AV
950 // mutex_destroy(&inode->i_mutex);
951 init_rwsem(&inode->i_rwsem);
952 lockdep_set_class(&inode->i_rwsem,
9a7aa12f
JK
953 &type->i_mutex_dir_key);
954 }
1e89a5e1 955 }
e096d0c7
JB
956}
957EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
14358e6d 958#endif
e096d0c7
JB
959
960/**
961 * unlock_new_inode - clear the I_NEW state and wake up any waiters
962 * @inode: new inode to unlock
963 *
964 * Called when the inode is fully initialised to clear the new state of the
965 * inode and wake up anyone waiting for the inode to finish initialisation.
966 */
967void unlock_new_inode(struct inode *inode)
968{
969 lockdep_annotate_inode_mutex_key(inode);
250df6ed 970 spin_lock(&inode->i_lock);
eaff8079 971 WARN_ON(!(inode->i_state & I_NEW));
c2b6d621 972 inode->i_state &= ~I_NEW & ~I_CREATING;
310fa7a3 973 smp_mb();
250df6ed
DC
974 wake_up_bit(&inode->i_state, __I_NEW);
975 spin_unlock(&inode->i_lock);
1da177e4 976}
1da177e4
LT
977EXPORT_SYMBOL(unlock_new_inode);
978
c2b6d621
AV
979void discard_new_inode(struct inode *inode)
980{
981 lockdep_annotate_inode_mutex_key(inode);
982 spin_lock(&inode->i_lock);
983 WARN_ON(!(inode->i_state & I_NEW));
984 inode->i_state &= ~I_NEW;
985 smp_mb();
986 wake_up_bit(&inode->i_state, __I_NEW);
987 spin_unlock(&inode->i_lock);
988 iput(inode);
989}
990EXPORT_SYMBOL(discard_new_inode);
991
375e289e
BF
992/**
993 * lock_two_nondirectories - take two i_mutexes on non-directory objects
4fd699ae
BF
994 *
995 * Lock any non-NULL argument that is not a directory.
996 * Zero, one or two objects may be locked by this function.
997 *
375e289e
BF
998 * @inode1: first inode to lock
999 * @inode2: second inode to lock
1000 */
1001void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1002{
4fd699ae
BF
1003 if (inode1 > inode2)
1004 swap(inode1, inode2);
1005
1006 if (inode1 && !S_ISDIR(inode1->i_mode))
5955102c 1007 inode_lock(inode1);
4fd699ae 1008 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
5955102c 1009 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
375e289e
BF
1010}
1011EXPORT_SYMBOL(lock_two_nondirectories);
1012
1013/**
1014 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1015 * @inode1: first inode to unlock
1016 * @inode2: second inode to unlock
1017 */
1018void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1019{
4fd699ae 1020 if (inode1 && !S_ISDIR(inode1->i_mode))
5955102c 1021 inode_unlock(inode1);
4fd699ae 1022 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
5955102c 1023 inode_unlock(inode2);
375e289e
BF
1024}
1025EXPORT_SYMBOL(unlock_two_nondirectories);
1026
80ea09a0
MS
1027/**
1028 * inode_insert5 - obtain an inode from a mounted file system
1029 * @inode: pre-allocated inode to use for insert to cache
1030 * @hashval: hash value (usually inode number) to get
1031 * @test: callback used for comparisons between inodes
1032 * @set: callback used to initialize a new struct inode
1033 * @data: opaque data pointer to pass to @test and @set
1034 *
1035 * Search for the inode specified by @hashval and @data in the inode cache,
1036 * and if present it is return it with an increased reference count. This is
1037 * a variant of iget5_locked() for callers that don't want to fail on memory
1038 * allocation of inode.
1039 *
1040 * If the inode is not in cache, insert the pre-allocated inode to cache and
1041 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1042 * to fill it in before unlocking it via unlock_new_inode().
1043 *
1044 * Note both @test and @set are called with the inode_hash_lock held, so can't
1045 * sleep.
1046 */
1047struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1048 int (*test)(struct inode *, void *),
1049 int (*set)(struct inode *, void *), void *data)
1050{
1051 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1052 struct inode *old;
e950564b 1053 bool creating = inode->i_state & I_CREATING;
80ea09a0
MS
1054
1055again:
1056 spin_lock(&inode_hash_lock);
1057 old = find_inode(inode->i_sb, head, test, data);
1058 if (unlikely(old)) {
1059 /*
1060 * Uhhuh, somebody else created the same inode under us.
1061 * Use the old inode instead of the preallocated one.
1062 */
1063 spin_unlock(&inode_hash_lock);
c2b6d621
AV
1064 if (IS_ERR(old))
1065 return NULL;
80ea09a0
MS
1066 wait_on_inode(old);
1067 if (unlikely(inode_unhashed(old))) {
1068 iput(old);
1069 goto again;
1070 }
1071 return old;
1072 }
1073
1074 if (set && unlikely(set(inode, data))) {
1075 inode = NULL;
1076 goto unlock;
1077 }
1078
1079 /*
1080 * Return the locked inode with I_NEW set, the
1081 * caller is responsible for filling in the contents
1082 */
1083 spin_lock(&inode->i_lock);
1084 inode->i_state |= I_NEW;
1085 hlist_add_head(&inode->i_hash, head);
1086 spin_unlock(&inode->i_lock);
e950564b
MS
1087 if (!creating)
1088 inode_sb_list_add(inode);
80ea09a0
MS
1089unlock:
1090 spin_unlock(&inode_hash_lock);
1091
1092 return inode;
1093}
1094EXPORT_SYMBOL(inode_insert5);
1095
0b2d0724
CH
1096/**
1097 * iget5_locked - obtain an inode from a mounted file system
1098 * @sb: super block of file system
1099 * @hashval: hash value (usually inode number) to get
1100 * @test: callback used for comparisons between inodes
1101 * @set: callback used to initialize a new struct inode
1102 * @data: opaque data pointer to pass to @test and @set
1103 *
1104 * Search for the inode specified by @hashval and @data in the inode cache,
1105 * and if present it is return it with an increased reference count. This is
1106 * a generalized version of iget_locked() for file systems where the inode
1107 * number is not sufficient for unique identification of an inode.
1108 *
1109 * If the inode is not in cache, allocate a new inode and return it locked,
1110 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1111 * before unlocking it via unlock_new_inode().
1da177e4 1112 *
0b2d0724
CH
1113 * Note both @test and @set are called with the inode_hash_lock held, so can't
1114 * sleep.
1da177e4 1115 */
0b2d0724
CH
1116struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1117 int (*test)(struct inode *, void *),
1118 int (*set)(struct inode *, void *), void *data)
1da177e4 1119{
80ea09a0 1120 struct inode *inode = ilookup5(sb, hashval, test, data);
0b2d0724 1121
80ea09a0 1122 if (!inode) {
e950564b 1123 struct inode *new = alloc_inode(sb);
0b2d0724 1124
80ea09a0 1125 if (new) {
e950564b 1126 new->i_state = 0;
80ea09a0
MS
1127 inode = inode_insert5(new, hashval, test, set, data);
1128 if (unlikely(inode != new))
e950564b 1129 destroy_inode(new);
2864f301 1130 }
1da177e4
LT
1131 }
1132 return inode;
1da177e4 1133}
0b2d0724 1134EXPORT_SYMBOL(iget5_locked);
1da177e4 1135
0b2d0724
CH
1136/**
1137 * iget_locked - obtain an inode from a mounted file system
1138 * @sb: super block of file system
1139 * @ino: inode number to get
1140 *
1141 * Search for the inode specified by @ino in the inode cache and if present
1142 * return it with an increased reference count. This is for file systems
1143 * where the inode number is sufficient for unique identification of an inode.
1144 *
1145 * If the inode is not in cache, allocate a new inode and return it locked,
1146 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1147 * before unlocking it via unlock_new_inode().
1da177e4 1148 */
0b2d0724 1149struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1da177e4 1150{
0b2d0724 1151 struct hlist_head *head = inode_hashtable + hash(sb, ino);
6b3304b5 1152 struct inode *inode;
2864f301 1153again:
0b2d0724
CH
1154 spin_lock(&inode_hash_lock);
1155 inode = find_inode_fast(sb, head, ino);
1156 spin_unlock(&inode_hash_lock);
1157 if (inode) {
c2b6d621
AV
1158 if (IS_ERR(inode))
1159 return NULL;
0b2d0724 1160 wait_on_inode(inode);
2864f301
AV
1161 if (unlikely(inode_unhashed(inode))) {
1162 iput(inode);
1163 goto again;
1164 }
0b2d0724
CH
1165 return inode;
1166 }
1167
1da177e4
LT
1168 inode = alloc_inode(sb);
1169 if (inode) {
6b3304b5 1170 struct inode *old;
1da177e4 1171
67a23c49 1172 spin_lock(&inode_hash_lock);
1da177e4
LT
1173 /* We released the lock, so.. */
1174 old = find_inode_fast(sb, head, ino);
1175 if (!old) {
1176 inode->i_ino = ino;
250df6ed
DC
1177 spin_lock(&inode->i_lock);
1178 inode->i_state = I_NEW;
646ec461 1179 hlist_add_head(&inode->i_hash, head);
250df6ed 1180 spin_unlock(&inode->i_lock);
55fa6091 1181 inode_sb_list_add(inode);
67a23c49 1182 spin_unlock(&inode_hash_lock);
1da177e4
LT
1183
1184 /* Return the locked inode with I_NEW set, the
1185 * caller is responsible for filling in the contents
1186 */
1187 return inode;
1188 }
1189
1190 /*
1191 * Uhhuh, somebody else created the same inode under
1192 * us. Use the old inode instead of the one we just
1193 * allocated.
1194 */
67a23c49 1195 spin_unlock(&inode_hash_lock);
1da177e4 1196 destroy_inode(inode);
c2b6d621
AV
1197 if (IS_ERR(old))
1198 return NULL;
1da177e4
LT
1199 inode = old;
1200 wait_on_inode(inode);
2864f301
AV
1201 if (unlikely(inode_unhashed(inode))) {
1202 iput(inode);
1203 goto again;
1204 }
1da177e4
LT
1205 }
1206 return inode;
1207}
0b2d0724 1208EXPORT_SYMBOL(iget_locked);
1da177e4 1209
ad5e195a
CH
1210/*
1211 * search the inode cache for a matching inode number.
1212 * If we find one, then the inode number we are trying to
1213 * allocate is not unique and so we should not use it.
1214 *
1215 * Returns 1 if the inode number is unique, 0 if it is not.
1216 */
1217static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1218{
1219 struct hlist_head *b = inode_hashtable + hash(sb, ino);
ad5e195a
CH
1220 struct inode *inode;
1221
67a23c49 1222 spin_lock(&inode_hash_lock);
b67bfe0d 1223 hlist_for_each_entry(inode, b, i_hash) {
67a23c49
DC
1224 if (inode->i_ino == ino && inode->i_sb == sb) {
1225 spin_unlock(&inode_hash_lock);
ad5e195a 1226 return 0;
67a23c49 1227 }
ad5e195a 1228 }
67a23c49 1229 spin_unlock(&inode_hash_lock);
ad5e195a
CH
1230
1231 return 1;
1232}
1233
1da177e4
LT
1234/**
1235 * iunique - get a unique inode number
1236 * @sb: superblock
1237 * @max_reserved: highest reserved inode number
1238 *
1239 * Obtain an inode number that is unique on the system for a given
1240 * superblock. This is used by file systems that have no natural
1241 * permanent inode numbering system. An inode number is returned that
1242 * is higher than the reserved limit but unique.
1243 *
1244 * BUGS:
1245 * With a large number of inodes live on the file system this function
1246 * currently becomes quite slow.
1247 */
1248ino_t iunique(struct super_block *sb, ino_t max_reserved)
1249{
866b04fc
JL
1250 /*
1251 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1252 * error if st_ino won't fit in target struct field. Use 32bit counter
1253 * here to attempt to avoid that.
1254 */
ad5e195a 1255 static DEFINE_SPINLOCK(iunique_lock);
866b04fc 1256 static unsigned int counter;
1da177e4 1257 ino_t res;
3361c7be 1258
ad5e195a 1259 spin_lock(&iunique_lock);
3361c7be
JL
1260 do {
1261 if (counter <= max_reserved)
1262 counter = max_reserved + 1;
1da177e4 1263 res = counter++;
ad5e195a
CH
1264 } while (!test_inode_iunique(sb, res));
1265 spin_unlock(&iunique_lock);
1da177e4 1266
3361c7be
JL
1267 return res;
1268}
1da177e4
LT
1269EXPORT_SYMBOL(iunique);
1270
1271struct inode *igrab(struct inode *inode)
1272{
250df6ed
DC
1273 spin_lock(&inode->i_lock);
1274 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1da177e4 1275 __iget(inode);
250df6ed
DC
1276 spin_unlock(&inode->i_lock);
1277 } else {
1278 spin_unlock(&inode->i_lock);
1da177e4
LT
1279 /*
1280 * Handle the case where s_op->clear_inode is not been
1281 * called yet, and somebody is calling igrab
1282 * while the inode is getting freed.
1283 */
1284 inode = NULL;
250df6ed 1285 }
1da177e4
LT
1286 return inode;
1287}
1da177e4
LT
1288EXPORT_SYMBOL(igrab);
1289
1290/**
0b2d0724 1291 * ilookup5_nowait - search for an inode in the inode cache
1da177e4 1292 * @sb: super block of file system to search
0b2d0724 1293 * @hashval: hash value (usually inode number) to search for
1da177e4
LT
1294 * @test: callback used for comparisons between inodes
1295 * @data: opaque data pointer to pass to @test
1da177e4 1296 *
0b2d0724 1297 * Search for the inode specified by @hashval and @data in the inode cache.
1da177e4
LT
1298 * If the inode is in the cache, the inode is returned with an incremented
1299 * reference count.
1300 *
0b2d0724
CH
1301 * Note: I_NEW is not waited upon so you have to be very careful what you do
1302 * with the returned inode. You probably should be using ilookup5() instead.
1da177e4 1303 *
b6d0ad68 1304 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1da177e4 1305 */
0b2d0724
CH
1306struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1307 int (*test)(struct inode *, void *), void *data)
1da177e4 1308{
0b2d0724 1309 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1da177e4
LT
1310 struct inode *inode;
1311
67a23c49 1312 spin_lock(&inode_hash_lock);
1da177e4 1313 inode = find_inode(sb, head, test, data);
67a23c49 1314 spin_unlock(&inode_hash_lock);
88bd5121 1315
c2b6d621 1316 return IS_ERR(inode) ? NULL : inode;
88bd5121 1317}
88bd5121
AA
1318EXPORT_SYMBOL(ilookup5_nowait);
1319
1320/**
1321 * ilookup5 - search for an inode in the inode cache
1322 * @sb: super block of file system to search
1323 * @hashval: hash value (usually inode number) to search for
1324 * @test: callback used for comparisons between inodes
1325 * @data: opaque data pointer to pass to @test
1326 *
0b2d0724
CH
1327 * Search for the inode specified by @hashval and @data in the inode cache,
1328 * and if the inode is in the cache, return the inode with an incremented
1329 * reference count. Waits on I_NEW before returning the inode.
88bd5121 1330 * returned with an incremented reference count.
1da177e4 1331 *
0b2d0724
CH
1332 * This is a generalized version of ilookup() for file systems where the
1333 * inode number is not sufficient for unique identification of an inode.
1da177e4 1334 *
0b2d0724 1335 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1da177e4
LT
1336 */
1337struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1338 int (*test)(struct inode *, void *), void *data)
1339{
2864f301
AV
1340 struct inode *inode;
1341again:
1342 inode = ilookup5_nowait(sb, hashval, test, data);
1343 if (inode) {
0b2d0724 1344 wait_on_inode(inode);
2864f301
AV
1345 if (unlikely(inode_unhashed(inode))) {
1346 iput(inode);
1347 goto again;
1348 }
1349 }
0b2d0724 1350 return inode;
1da177e4 1351}
1da177e4
LT
1352EXPORT_SYMBOL(ilookup5);
1353
1354/**
1355 * ilookup - search for an inode in the inode cache
1356 * @sb: super block of file system to search
1357 * @ino: inode number to search for
1358 *
0b2d0724
CH
1359 * Search for the inode @ino in the inode cache, and if the inode is in the
1360 * cache, the inode is returned with an incremented reference count.
1da177e4
LT
1361 */
1362struct inode *ilookup(struct super_block *sb, unsigned long ino)
1363{
1364 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1da177e4 1365 struct inode *inode;
2864f301 1366again:
0b2d0724
CH
1367 spin_lock(&inode_hash_lock);
1368 inode = find_inode_fast(sb, head, ino);
1369 spin_unlock(&inode_hash_lock);
1da177e4 1370
2864f301 1371 if (inode) {
c2b6d621
AV
1372 if (IS_ERR(inode))
1373 return NULL;
0b2d0724 1374 wait_on_inode(inode);
2864f301
AV
1375 if (unlikely(inode_unhashed(inode))) {
1376 iput(inode);
1377 goto again;
1378 }
1379 }
0b2d0724 1380 return inode;
1da177e4 1381}
0b2d0724 1382EXPORT_SYMBOL(ilookup);
1da177e4 1383
fe032c42
TT
1384/**
1385 * find_inode_nowait - find an inode in the inode cache
1386 * @sb: super block of file system to search
1387 * @hashval: hash value (usually inode number) to search for
1388 * @match: callback used for comparisons between inodes
1389 * @data: opaque data pointer to pass to @match
1390 *
1391 * Search for the inode specified by @hashval and @data in the inode
1392 * cache, where the helper function @match will return 0 if the inode
1393 * does not match, 1 if the inode does match, and -1 if the search
1394 * should be stopped. The @match function must be responsible for
1395 * taking the i_lock spin_lock and checking i_state for an inode being
1396 * freed or being initialized, and incrementing the reference count
1397 * before returning 1. It also must not sleep, since it is called with
1398 * the inode_hash_lock spinlock held.
1399 *
1400 * This is a even more generalized version of ilookup5() when the
1401 * function must never block --- find_inode() can block in
1402 * __wait_on_freeing_inode() --- or when the caller can not increment
1403 * the reference count because the resulting iput() might cause an
1404 * inode eviction. The tradeoff is that the @match funtion must be
1405 * very carefully implemented.
1406 */
1407struct inode *find_inode_nowait(struct super_block *sb,
1408 unsigned long hashval,
1409 int (*match)(struct inode *, unsigned long,
1410 void *),
1411 void *data)
1412{
1413 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1414 struct inode *inode, *ret_inode = NULL;
1415 int mval;
1416
1417 spin_lock(&inode_hash_lock);
1418 hlist_for_each_entry(inode, head, i_hash) {
1419 if (inode->i_sb != sb)
1420 continue;
1421 mval = match(inode, hashval, data);
1422 if (mval == 0)
1423 continue;
1424 if (mval == 1)
1425 ret_inode = inode;
1426 goto out;
1427 }
1428out:
1429 spin_unlock(&inode_hash_lock);
1430 return ret_inode;
1431}
1432EXPORT_SYMBOL(find_inode_nowait);
1433
261bca86
AV
1434int insert_inode_locked(struct inode *inode)
1435{
1436 struct super_block *sb = inode->i_sb;
1437 ino_t ino = inode->i_ino;
1438 struct hlist_head *head = inode_hashtable + hash(sb, ino);
261bca86 1439
261bca86 1440 while (1) {
72a43d63 1441 struct inode *old = NULL;
67a23c49 1442 spin_lock(&inode_hash_lock);
b67bfe0d 1443 hlist_for_each_entry(old, head, i_hash) {
72a43d63
AV
1444 if (old->i_ino != ino)
1445 continue;
1446 if (old->i_sb != sb)
1447 continue;
250df6ed
DC
1448 spin_lock(&old->i_lock);
1449 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1450 spin_unlock(&old->i_lock);
72a43d63 1451 continue;
250df6ed 1452 }
72a43d63
AV
1453 break;
1454 }
b67bfe0d 1455 if (likely(!old)) {
250df6ed 1456 spin_lock(&inode->i_lock);
c2b6d621 1457 inode->i_state |= I_NEW | I_CREATING;
261bca86 1458 hlist_add_head(&inode->i_hash, head);
250df6ed 1459 spin_unlock(&inode->i_lock);
67a23c49 1460 spin_unlock(&inode_hash_lock);
261bca86
AV
1461 return 0;
1462 }
c2b6d621
AV
1463 if (unlikely(old->i_state & I_CREATING)) {
1464 spin_unlock(&old->i_lock);
1465 spin_unlock(&inode_hash_lock);
1466 return -EBUSY;
1467 }
261bca86 1468 __iget(old);
250df6ed 1469 spin_unlock(&old->i_lock);
67a23c49 1470 spin_unlock(&inode_hash_lock);
261bca86 1471 wait_on_inode(old);
1d3382cb 1472 if (unlikely(!inode_unhashed(old))) {
261bca86
AV
1473 iput(old);
1474 return -EBUSY;
1475 }
1476 iput(old);
1477 }
1478}
261bca86
AV
1479EXPORT_SYMBOL(insert_inode_locked);
1480
1481int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1482 int (*test)(struct inode *, void *), void *data)
1483{
c2b6d621
AV
1484 struct inode *old;
1485
1486 inode->i_state |= I_CREATING;
1487 old = inode_insert5(inode, hashval, test, NULL, data);
261bca86 1488
80ea09a0 1489 if (old != inode) {
261bca86 1490 iput(old);
80ea09a0 1491 return -EBUSY;
261bca86 1492 }
80ea09a0 1493 return 0;
261bca86 1494}
261bca86
AV
1495EXPORT_SYMBOL(insert_inode_locked4);
1496
1da177e4 1497
45321ac5
AV
1498int generic_delete_inode(struct inode *inode)
1499{
1500 return 1;
1501}
1502EXPORT_SYMBOL(generic_delete_inode);
1503
45321ac5
AV
1504/*
1505 * Called when we're dropping the last reference
1506 * to an inode.
22fe4042 1507 *
45321ac5
AV
1508 * Call the FS "drop_inode()" function, defaulting to
1509 * the legacy UNIX filesystem behaviour. If it tells
1510 * us to evict inode, do so. Otherwise, retain inode
1511 * in cache if fs is alive, sync and evict if fs is
1512 * shutting down.
22fe4042 1513 */
45321ac5 1514static void iput_final(struct inode *inode)
1da177e4
LT
1515{
1516 struct super_block *sb = inode->i_sb;
45321ac5
AV
1517 const struct super_operations *op = inode->i_sb->s_op;
1518 int drop;
1519
250df6ed
DC
1520 WARN_ON(inode->i_state & I_NEW);
1521
e7f59097 1522 if (op->drop_inode)
45321ac5
AV
1523 drop = op->drop_inode(inode);
1524 else
1525 drop = generic_drop_inode(inode);
1da177e4 1526
1751e8a6 1527 if (!drop && (sb->s_flags & SB_ACTIVE)) {
4eff96dd 1528 inode_add_lru(inode);
b2b2af8e 1529 spin_unlock(&inode->i_lock);
b2b2af8e
DC
1530 return;
1531 }
1532
45321ac5 1533 if (!drop) {
991114c6 1534 inode->i_state |= I_WILL_FREE;
250df6ed 1535 spin_unlock(&inode->i_lock);
1da177e4 1536 write_inode_now(inode, 1);
250df6ed 1537 spin_lock(&inode->i_lock);
7ef0d737 1538 WARN_ON(inode->i_state & I_NEW);
991114c6 1539 inode->i_state &= ~I_WILL_FREE;
1da177e4 1540 }
7ccf19a8 1541
991114c6 1542 inode->i_state |= I_FREEING;
c4ae0c65
ED
1543 if (!list_empty(&inode->i_lru))
1544 inode_lru_list_del(inode);
b2b2af8e 1545 spin_unlock(&inode->i_lock);
b2b2af8e 1546
644da596 1547 evict(inode);
1da177e4
LT
1548}
1549
1da177e4 1550/**
6b3304b5 1551 * iput - put an inode
1da177e4
LT
1552 * @inode: inode to put
1553 *
1554 * Puts an inode, dropping its usage count. If the inode use count hits
1555 * zero, the inode is then freed and may also be destroyed.
1556 *
1557 * Consequently, iput() can sleep.
1558 */
1559void iput(struct inode *inode)
1560{
0ae45f63
TT
1561 if (!inode)
1562 return;
1563 BUG_ON(inode->i_state & I_CLEAR);
1564retry:
1565 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1566 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1567 atomic_inc(&inode->i_count);
0ae45f63
TT
1568 spin_unlock(&inode->i_lock);
1569 trace_writeback_lazytime_iput(inode);
1570 mark_inode_dirty_sync(inode);
1571 goto retry;
1572 }
1573 iput_final(inode);
1da177e4
LT
1574 }
1575}
1da177e4
LT
1576EXPORT_SYMBOL(iput);
1577
1578/**
1579 * bmap - find a block number in a file
1580 * @inode: inode of file
1581 * @block: block to find
1582 *
1583 * Returns the block number on the device holding the inode that
1584 * is the disk block number for the block of the file requested.
1585 * That is, asked for block 4 of inode 1 the function will return the
6b3304b5 1586 * disk block relative to the disk start that holds that block of the
1da177e4
LT
1587 * file.
1588 */
6b3304b5 1589sector_t bmap(struct inode *inode, sector_t block)
1da177e4
LT
1590{
1591 sector_t res = 0;
1592 if (inode->i_mapping->a_ops->bmap)
1593 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1594 return res;
1595}
1da177e4
LT
1596EXPORT_SYMBOL(bmap);
1597
11ff6f05
MG
1598/*
1599 * With relative atime, only update atime if the previous atime is
1600 * earlier than either the ctime or mtime or if at least a day has
1601 * passed since the last atime update.
1602 */
c6718543
MS
1603static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1604 struct timespec now)
11ff6f05
MG
1605{
1606
c6718543 1607 if (!(mnt->mnt_flags & MNT_RELATIME))
11ff6f05
MG
1608 return 1;
1609 /*
1610 * Is mtime younger than atime? If yes, update atime:
1611 */
95582b00 1612 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
11ff6f05
MG
1613 return 1;
1614 /*
1615 * Is ctime younger than atime? If yes, update atime:
1616 */
95582b00 1617 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
11ff6f05
MG
1618 return 1;
1619
1620 /*
1621 * Is the previous atime value older than a day? If yes,
1622 * update atime:
1623 */
1624 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1625 return 1;
1626 /*
1627 * Good, we can skip the atime update:
1628 */
1629 return 0;
1630}
1631
95582b00 1632int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
c3b2da31 1633{
0ae45f63 1634 int iflags = I_DIRTY_TIME;
e38cf302 1635 bool dirty = false;
c3b2da31
JB
1636
1637 if (flags & S_ATIME)
1638 inode->i_atime = *time;
1639 if (flags & S_VERSION)
e38cf302 1640 dirty = inode_maybe_inc_iversion(inode, false);
c3b2da31
JB
1641 if (flags & S_CTIME)
1642 inode->i_ctime = *time;
1643 if (flags & S_MTIME)
1644 inode->i_mtime = *time;
e38cf302
JL
1645 if ((flags & (S_ATIME | S_CTIME | S_MTIME)) &&
1646 !(inode->i_sb->s_flags & SB_LAZYTIME))
1647 dirty = true;
0ae45f63 1648
e38cf302 1649 if (dirty)
0ae45f63
TT
1650 iflags |= I_DIRTY_SYNC;
1651 __mark_inode_dirty(inode, iflags);
c3b2da31
JB
1652 return 0;
1653}
0ae45f63
TT
1654EXPORT_SYMBOL(generic_update_time);
1655
1656/*
1657 * This does the actual work of updating an inodes time or version. Must have
1658 * had called mnt_want_write() before calling this.
1659 */
95582b00 1660static int update_time(struct inode *inode, struct timespec64 *time, int flags)
0ae45f63 1661{
95582b00 1662 int (*update_time)(struct inode *, struct timespec64 *, int);
0ae45f63
TT
1663
1664 update_time = inode->i_op->update_time ? inode->i_op->update_time :
1665 generic_update_time;
1666
1667 return update_time(inode, time, flags);
1668}
c3b2da31 1669
1da177e4 1670/**
869243a0 1671 * touch_atime - update the access time
185553b2 1672 * @path: the &struct path to update
30fdc8ee 1673 * @inode: inode to update
1da177e4
LT
1674 *
1675 * Update the accessed time on an inode and mark it for writeback.
1676 * This function automatically handles read only file systems and media,
1677 * as well as the "noatime" flag and inode specific "noatime" markers.
1678 */
c6718543 1679bool atime_needs_update(const struct path *path, struct inode *inode)
1da177e4 1680{
68ac1234 1681 struct vfsmount *mnt = path->mnt;
95582b00 1682 struct timespec64 now;
1da177e4 1683
cdb70f3f 1684 if (inode->i_flags & S_NOATIME)
8fa9dd24 1685 return false;
0bd23d09
EB
1686
1687 /* Atime updates will likely cause i_uid and i_gid to be written
1688 * back improprely if their true value is unknown to the vfs.
1689 */
1690 if (HAS_UNMAPPED_ID(inode))
1691 return false;
1692
37756ced 1693 if (IS_NOATIME(inode))
8fa9dd24 1694 return false;
1751e8a6 1695 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
8fa9dd24 1696 return false;
47ae32d6 1697
cdb70f3f 1698 if (mnt->mnt_flags & MNT_NOATIME)
8fa9dd24 1699 return false;
cdb70f3f 1700 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
8fa9dd24 1701 return false;
1da177e4 1702
c2050a45 1703 now = current_time(inode);
11ff6f05 1704
c6718543 1705 if (!relatime_need_update(mnt, inode, timespec64_to_timespec(now)))
8fa9dd24 1706 return false;
11ff6f05 1707
95582b00 1708 if (timespec64_equal(&inode->i_atime, &now))
8fa9dd24
N
1709 return false;
1710
1711 return true;
1712}
1713
1714void touch_atime(const struct path *path)
1715{
1716 struct vfsmount *mnt = path->mnt;
1717 struct inode *inode = d_inode(path->dentry);
95582b00 1718 struct timespec64 now;
8fa9dd24 1719
c6718543 1720 if (!atime_needs_update(path, inode))
b12536c2
AK
1721 return;
1722
5d37e9e6 1723 if (!sb_start_write_trylock(inode->i_sb))
b12536c2 1724 return;
47ae32d6 1725
8fa9dd24 1726 if (__mnt_want_write(mnt) != 0)
5d37e9e6 1727 goto skip_update;
c3b2da31
JB
1728 /*
1729 * File systems can error out when updating inodes if they need to
1730 * allocate new space to modify an inode (such is the case for
1731 * Btrfs), but since we touch atime while walking down the path we
1732 * really don't care if we failed to update the atime of the file,
1733 * so just ignore the return value.
2bc55652
AB
1734 * We may also fail on filesystems that have the ability to make parts
1735 * of the fs read only, e.g. subvolumes in Btrfs.
c3b2da31 1736 */
c2050a45 1737 now = current_time(inode);
c3b2da31 1738 update_time(inode, &now, S_ATIME);
5d37e9e6
JK
1739 __mnt_drop_write(mnt);
1740skip_update:
1741 sb_end_write(inode->i_sb);
1da177e4 1742}
869243a0 1743EXPORT_SYMBOL(touch_atime);
1da177e4 1744
3ed37648
CW
1745/*
1746 * The logic we want is
1747 *
1748 * if suid or (sgid and xgrp)
1749 * remove privs
1750 */
1751int should_remove_suid(struct dentry *dentry)
1752{
df2b1afd 1753 umode_t mode = d_inode(dentry)->i_mode;
3ed37648
CW
1754 int kill = 0;
1755
1756 /* suid always must be killed */
1757 if (unlikely(mode & S_ISUID))
1758 kill = ATTR_KILL_SUID;
1759
1760 /*
1761 * sgid without any exec bits is just a mandatory locking mark; leave
1762 * it alone. If some exec bits are set, it's a real sgid; kill it.
1763 */
1764 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1765 kill |= ATTR_KILL_SGID;
1766
1767 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1768 return kill;
1769
1770 return 0;
1771}
1772EXPORT_SYMBOL(should_remove_suid);
1773
dbfae0cd
JK
1774/*
1775 * Return mask of changes for notify_change() that need to be done as a
1776 * response to write or truncate. Return 0 if nothing has to be changed.
1777 * Negative value on error (change should be denied).
1778 */
45f147a1 1779int dentry_needs_remove_privs(struct dentry *dentry)
dbfae0cd 1780{
dbfae0cd
JK
1781 struct inode *inode = d_inode(dentry);
1782 int mask = 0;
1783 int ret;
1784
1785 if (IS_NOSEC(inode))
1786 return 0;
1787
1788 mask = should_remove_suid(dentry);
1789 ret = security_inode_need_killpriv(dentry);
1790 if (ret < 0)
1791 return ret;
1792 if (ret)
1793 mask |= ATTR_KILL_PRIV;
1794 return mask;
1795}
dbfae0cd
JK
1796
1797static int __remove_privs(struct dentry *dentry, int kill)
3ed37648
CW
1798{
1799 struct iattr newattrs;
1800
1801 newattrs.ia_valid = ATTR_FORCE | kill;
27ac0ffe
BF
1802 /*
1803 * Note we call this on write, so notify_change will not
1804 * encounter any conflicting delegations:
1805 */
1806 return notify_change(dentry, &newattrs, NULL);
3ed37648
CW
1807}
1808
5fa8e0a1
JK
1809/*
1810 * Remove special file priviledges (suid, capabilities) when file is written
1811 * to or truncated.
1812 */
1813int file_remove_privs(struct file *file)
3ed37648 1814{
c1892c37
MS
1815 struct dentry *dentry = file_dentry(file);
1816 struct inode *inode = file_inode(file);
dbfae0cd 1817 int kill;
3ed37648
CW
1818 int error = 0;
1819
1820 /* Fast path for nothing security related */
1821 if (IS_NOSEC(inode))
1822 return 0;
1823
c1892c37 1824 kill = dentry_needs_remove_privs(dentry);
dbfae0cd
JK
1825 if (kill < 0)
1826 return kill;
1827 if (kill)
1828 error = __remove_privs(dentry, kill);
2426f391
JK
1829 if (!error)
1830 inode_has_no_xattr(inode);
3ed37648
CW
1831
1832 return error;
1833}
5fa8e0a1 1834EXPORT_SYMBOL(file_remove_privs);
3ed37648 1835
1da177e4 1836/**
870f4817
CH
1837 * file_update_time - update mtime and ctime time
1838 * @file: file accessed
1da177e4 1839 *
870f4817
CH
1840 * Update the mtime and ctime members of an inode and mark the inode
1841 * for writeback. Note that this function is meant exclusively for
1842 * usage in the file write path of filesystems, and filesystems may
1843 * choose to explicitly ignore update via this function with the
2eadfc0e 1844 * S_NOCMTIME inode flag, e.g. for network filesystem where these
c3b2da31
JB
1845 * timestamps are handled by the server. This can return an error for
1846 * file systems who need to allocate space in order to update an inode.
1da177e4
LT
1847 */
1848
c3b2da31 1849int file_update_time(struct file *file)
1da177e4 1850{
496ad9aa 1851 struct inode *inode = file_inode(file);
95582b00 1852 struct timespec64 now;
c3b2da31
JB
1853 int sync_it = 0;
1854 int ret;
1da177e4 1855
ce06e0b2 1856 /* First try to exhaust all avenues to not sync */
1da177e4 1857 if (IS_NOCMTIME(inode))
c3b2da31 1858 return 0;
20ddee2c 1859
c2050a45 1860 now = current_time(inode);
95582b00 1861 if (!timespec64_equal(&inode->i_mtime, &now))
ce06e0b2 1862 sync_it = S_MTIME;
1da177e4 1863
95582b00 1864 if (!timespec64_equal(&inode->i_ctime, &now))
ce06e0b2 1865 sync_it |= S_CTIME;
870f4817 1866
e38cf302 1867 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
ce06e0b2 1868 sync_it |= S_VERSION;
7a224228 1869
ce06e0b2 1870 if (!sync_it)
c3b2da31 1871 return 0;
ce06e0b2
AK
1872
1873 /* Finally allowed to write? Takes lock. */
eb04c282 1874 if (__mnt_want_write_file(file))
c3b2da31 1875 return 0;
ce06e0b2 1876
c3b2da31 1877 ret = update_time(inode, &now, sync_it);
eb04c282 1878 __mnt_drop_write_file(file);
c3b2da31
JB
1879
1880 return ret;
1da177e4 1881}
870f4817 1882EXPORT_SYMBOL(file_update_time);
1da177e4
LT
1883
1884int inode_needs_sync(struct inode *inode)
1885{
1886 if (IS_SYNC(inode))
1887 return 1;
1888 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1889 return 1;
1890 return 0;
1891}
1da177e4
LT
1892EXPORT_SYMBOL(inode_needs_sync);
1893
1da177e4 1894/*
168a9fd6
MS
1895 * If we try to find an inode in the inode hash while it is being
1896 * deleted, we have to wait until the filesystem completes its
1897 * deletion before reporting that it isn't found. This function waits
1898 * until the deletion _might_ have completed. Callers are responsible
1899 * to recheck inode state.
1900 *
eaff8079 1901 * It doesn't matter if I_NEW is not set initially, a call to
250df6ed
DC
1902 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1903 * will DTRT.
1da177e4
LT
1904 */
1905static void __wait_on_freeing_inode(struct inode *inode)
1906{
1907 wait_queue_head_t *wq;
eaff8079
CH
1908 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1909 wq = bit_waitqueue(&inode->i_state, __I_NEW);
21417136 1910 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
250df6ed 1911 spin_unlock(&inode->i_lock);
67a23c49 1912 spin_unlock(&inode_hash_lock);
1da177e4 1913 schedule();
21417136 1914 finish_wait(wq, &wait.wq_entry);
67a23c49 1915 spin_lock(&inode_hash_lock);
1da177e4
LT
1916}
1917
1da177e4
LT
1918static __initdata unsigned long ihash_entries;
1919static int __init set_ihash_entries(char *str)
1920{
1921 if (!str)
1922 return 0;
1923 ihash_entries = simple_strtoul(str, &str, 0);
1924 return 1;
1925}
1926__setup("ihash_entries=", set_ihash_entries);
1927
1928/*
1929 * Initialize the waitqueues and inode hash table.
1930 */
1931void __init inode_init_early(void)
1932{
1da177e4
LT
1933 /* If hashes are distributed across NUMA nodes, defer
1934 * hash allocation until vmalloc space is available.
1935 */
1936 if (hashdist)
1937 return;
1938
1939 inode_hashtable =
1940 alloc_large_system_hash("Inode-cache",
1941 sizeof(struct hlist_head),
1942 ihash_entries,
1943 14,
3d375d78 1944 HASH_EARLY | HASH_ZERO,
1da177e4
LT
1945 &i_hash_shift,
1946 &i_hash_mask,
31fe62b9 1947 0,
1da177e4 1948 0);
1da177e4
LT
1949}
1950
74bf17cf 1951void __init inode_init(void)
1da177e4 1952{
1da177e4 1953 /* inode slab cache */
b0196009
PJ
1954 inode_cachep = kmem_cache_create("inode_cache",
1955 sizeof(struct inode),
1956 0,
1957 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
5d097056 1958 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
20c2df83 1959 init_once);
1da177e4
LT
1960
1961 /* Hash may have been set up in inode_init_early */
1962 if (!hashdist)
1963 return;
1964
1965 inode_hashtable =
1966 alloc_large_system_hash("Inode-cache",
1967 sizeof(struct hlist_head),
1968 ihash_entries,
1969 14,
3d375d78 1970 HASH_ZERO,
1da177e4
LT
1971 &i_hash_shift,
1972 &i_hash_mask,
31fe62b9 1973 0,
1da177e4 1974 0);
1da177e4
LT
1975}
1976
1977void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1978{
1979 inode->i_mode = mode;
1980 if (S_ISCHR(mode)) {
1981 inode->i_fop = &def_chr_fops;
1982 inode->i_rdev = rdev;
1983 } else if (S_ISBLK(mode)) {
1984 inode->i_fop = &def_blk_fops;
1985 inode->i_rdev = rdev;
1986 } else if (S_ISFIFO(mode))
599a0ac1 1987 inode->i_fop = &pipefifo_fops;
1da177e4 1988 else if (S_ISSOCK(mode))
bd9b51e7 1989 ; /* leave it no_open_fops */
1da177e4 1990 else
af0d9ae8
MK
1991 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1992 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1993 inode->i_ino);
1da177e4
LT
1994}
1995EXPORT_SYMBOL(init_special_inode);
a1bd120d
DM
1996
1997/**
eaae668d 1998 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
a1bd120d
DM
1999 * @inode: New inode
2000 * @dir: Directory inode
2001 * @mode: mode of the new inode
2002 */
2003void inode_init_owner(struct inode *inode, const struct inode *dir,
62bb1091 2004 umode_t mode)
a1bd120d
DM
2005{
2006 inode->i_uid = current_fsuid();
2007 if (dir && dir->i_mode & S_ISGID) {
2008 inode->i_gid = dir->i_gid;
0fa3ecd8
LT
2009
2010 /* Directories are special, and always inherit S_ISGID */
a1bd120d
DM
2011 if (S_ISDIR(mode))
2012 mode |= S_ISGID;
0fa3ecd8
LT
2013 else if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP) &&
2014 !in_group_p(inode->i_gid) &&
2015 !capable_wrt_inode_uidgid(dir, CAP_FSETID))
2016 mode &= ~S_ISGID;
a1bd120d
DM
2017 } else
2018 inode->i_gid = current_fsgid();
2019 inode->i_mode = mode;
2020}
2021EXPORT_SYMBOL(inode_init_owner);
e795b717 2022
2e149670
SH
2023/**
2024 * inode_owner_or_capable - check current task permissions to inode
2025 * @inode: inode being checked
2026 *
23adbe12
AL
2027 * Return true if current either has CAP_FOWNER in a namespace with the
2028 * inode owner uid mapped, or owns the file.
e795b717 2029 */
2e149670 2030bool inode_owner_or_capable(const struct inode *inode)
e795b717 2031{
23adbe12
AL
2032 struct user_namespace *ns;
2033
92361636 2034 if (uid_eq(current_fsuid(), inode->i_uid))
e795b717 2035 return true;
23adbe12
AL
2036
2037 ns = current_user_ns();
cc658db4 2038 if (kuid_has_mapping(ns, inode->i_uid) && ns_capable(ns, CAP_FOWNER))
e795b717
SH
2039 return true;
2040 return false;
2041}
2e149670 2042EXPORT_SYMBOL(inode_owner_or_capable);
1d59d61f
TM
2043
2044/*
2045 * Direct i/o helper functions
2046 */
2047static void __inode_dio_wait(struct inode *inode)
2048{
2049 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2050 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2051
2052 do {
21417136 2053 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
1d59d61f
TM
2054 if (atomic_read(&inode->i_dio_count))
2055 schedule();
2056 } while (atomic_read(&inode->i_dio_count));
21417136 2057 finish_wait(wq, &q.wq_entry);
1d59d61f
TM
2058}
2059
2060/**
2061 * inode_dio_wait - wait for outstanding DIO requests to finish
2062 * @inode: inode to wait for
2063 *
2064 * Waits for all pending direct I/O requests to finish so that we can
2065 * proceed with a truncate or equivalent operation.
2066 *
2067 * Must be called under a lock that serializes taking new references
2068 * to i_dio_count, usually by inode->i_mutex.
2069 */
2070void inode_dio_wait(struct inode *inode)
2071{
2072 if (atomic_read(&inode->i_dio_count))
2073 __inode_dio_wait(inode);
2074}
2075EXPORT_SYMBOL(inode_dio_wait);
2076
5f16f322
TT
2077/*
2078 * inode_set_flags - atomically set some inode flags
2079 *
2080 * Note: the caller should be holding i_mutex, or else be sure that
2081 * they have exclusive access to the inode structure (i.e., while the
2082 * inode is being instantiated). The reason for the cmpxchg() loop
2083 * --- which wouldn't be necessary if all code paths which modify
2084 * i_flags actually followed this rule, is that there is at least one
5fa8e0a1
JK
2085 * code path which doesn't today so we use cmpxchg() out of an abundance
2086 * of caution.
5f16f322
TT
2087 *
2088 * In the long run, i_mutex is overkill, and we should probably look
2089 * at using the i_lock spinlock to protect i_flags, and then make sure
2090 * it is so documented in include/linux/fs.h and that all code follows
2091 * the locking convention!!
2092 */
2093void inode_set_flags(struct inode *inode, unsigned int flags,
2094 unsigned int mask)
2095{
2096 unsigned int old_flags, new_flags;
2097
2098 WARN_ON_ONCE(flags & ~mask);
2099 do {
6aa7de05 2100 old_flags = READ_ONCE(inode->i_flags);
5f16f322
TT
2101 new_flags = (old_flags & ~mask) | flags;
2102 } while (unlikely(cmpxchg(&inode->i_flags, old_flags,
2103 new_flags) != old_flags));
2104}
2105EXPORT_SYMBOL(inode_set_flags);
21fc61c7
AV
2106
2107void inode_nohighmem(struct inode *inode)
2108{
2109 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2110}
2111EXPORT_SYMBOL(inode_nohighmem);
3cd88666 2112
8efd6894
DD
2113/**
2114 * timespec64_trunc - Truncate timespec64 to a granularity
2115 * @t: Timespec64
2116 * @gran: Granularity in ns.
2117 *
2118 * Truncate a timespec64 to a granularity. Always rounds down. gran must
2119 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2120 */
2121struct timespec64 timespec64_trunc(struct timespec64 t, unsigned gran)
2122{
2123 /* Avoid division in the common cases 1 ns and 1 s. */
2124 if (gran == 1) {
2125 /* nothing */
2126 } else if (gran == NSEC_PER_SEC) {
2127 t.tv_nsec = 0;
2128 } else if (gran > 1 && gran < NSEC_PER_SEC) {
2129 t.tv_nsec -= t.tv_nsec % gran;
2130 } else {
2131 WARN(1, "illegal file time granularity: %u", gran);
2132 }
2133 return t;
2134}
2135EXPORT_SYMBOL(timespec64_trunc);
2136
3cd88666
DD
2137/**
2138 * current_time - Return FS time
2139 * @inode: inode.
2140 *
2141 * Return the current time truncated to the time granularity supported by
2142 * the fs.
2143 *
2144 * Note that inode and inode->sb cannot be NULL.
2145 * Otherwise, the function warns and returns time without truncation.
2146 */
95582b00 2147struct timespec64 current_time(struct inode *inode)
3cd88666 2148{
95582b00 2149 struct timespec64 now = current_kernel_time64();
3cd88666
DD
2150
2151 if (unlikely(!inode->i_sb)) {
2152 WARN(1, "current_time() called with uninitialized super_block in the inode");
2153 return now;
2154 }
2155
95582b00 2156 return timespec64_trunc(now, inode->i_sb->s_time_gran);
3cd88666
DD
2157}
2158EXPORT_SYMBOL(current_time);