Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * (C) 1997 Linus Torvalds
4 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
5 */
6#include <linux/export.h>
7#include <linux/fs.h>
8#include <linux/filelock.h>
9#include <linux/mm.h>
10#include <linux/backing-dev.h>
11#include <linux/hash.h>
12#include <linux/swap.h>
13#include <linux/security.h>
14#include <linux/cdev.h>
15#include <linux/memblock.h>
16#include <linux/fsnotify.h>
17#include <linux/mount.h>
18#include <linux/posix_acl.h>
19#include <linux/buffer_head.h> /* for inode_has_buffers */
20#include <linux/ratelimit.h>
21#include <linux/list_lru.h>
22#include <linux/iversion.h>
23#include <trace/events/writeback.h>
24#include "internal.h"
25
26/*
27 * Inode locking rules:
28 *
29 * inode->i_lock protects:
30 * inode->i_state, inode->i_hash, __iget(), inode->i_io_list
31 * Inode LRU list locks protect:
32 * inode->i_sb->s_inode_lru, inode->i_lru
33 * inode->i_sb->s_inode_list_lock protects:
34 * inode->i_sb->s_inodes, inode->i_sb_list
35 * bdi->wb.list_lock protects:
36 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
37 * inode_hash_lock protects:
38 * inode_hashtable, inode->i_hash
39 *
40 * Lock ordering:
41 *
42 * inode->i_sb->s_inode_list_lock
43 * inode->i_lock
44 * Inode LRU list locks
45 *
46 * bdi->wb.list_lock
47 * inode->i_lock
48 *
49 * inode_hash_lock
50 * inode->i_sb->s_inode_list_lock
51 * inode->i_lock
52 *
53 * iunique_lock
54 * inode_hash_lock
55 */
56
57static unsigned int i_hash_mask __ro_after_init;
58static unsigned int i_hash_shift __ro_after_init;
59static struct hlist_head *inode_hashtable __ro_after_init;
60static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
61
62/*
63 * Empty aops. Can be used for the cases where the user does not
64 * define any of the address_space operations.
65 */
66const struct address_space_operations empty_aops = {
67};
68EXPORT_SYMBOL(empty_aops);
69
70static DEFINE_PER_CPU(unsigned long, nr_inodes);
71static DEFINE_PER_CPU(unsigned long, nr_unused);
72
73static struct kmem_cache *inode_cachep __ro_after_init;
74
75static long get_nr_inodes(void)
76{
77 int i;
78 long sum = 0;
79 for_each_possible_cpu(i)
80 sum += per_cpu(nr_inodes, i);
81 return sum < 0 ? 0 : sum;
82}
83
84static inline long get_nr_inodes_unused(void)
85{
86 int i;
87 long sum = 0;
88 for_each_possible_cpu(i)
89 sum += per_cpu(nr_unused, i);
90 return sum < 0 ? 0 : sum;
91}
92
93long get_nr_dirty_inodes(void)
94{
95 /* not actually dirty inodes, but a wild approximation */
96 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
97 return nr_dirty > 0 ? nr_dirty : 0;
98}
99
100/*
101 * Handle nr_inode sysctl
102 */
103#ifdef CONFIG_SYSCTL
104/*
105 * Statistics gathering..
106 */
107static struct inodes_stat_t inodes_stat;
108
109static int proc_nr_inodes(struct ctl_table *table, int write, void *buffer,
110 size_t *lenp, loff_t *ppos)
111{
112 inodes_stat.nr_inodes = get_nr_inodes();
113 inodes_stat.nr_unused = get_nr_inodes_unused();
114 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
115}
116
117static struct ctl_table inodes_sysctls[] = {
118 {
119 .procname = "inode-nr",
120 .data = &inodes_stat,
121 .maxlen = 2*sizeof(long),
122 .mode = 0444,
123 .proc_handler = proc_nr_inodes,
124 },
125 {
126 .procname = "inode-state",
127 .data = &inodes_stat,
128 .maxlen = 7*sizeof(long),
129 .mode = 0444,
130 .proc_handler = proc_nr_inodes,
131 },
132};
133
134static int __init init_fs_inode_sysctls(void)
135{
136 register_sysctl_init("fs", inodes_sysctls);
137 return 0;
138}
139early_initcall(init_fs_inode_sysctls);
140#endif
141
142static int no_open(struct inode *inode, struct file *file)
143{
144 return -ENXIO;
145}
146
147/**
148 * inode_init_always - perform inode structure initialisation
149 * @sb: superblock inode belongs to
150 * @inode: inode to initialise
151 *
152 * These are initializations that need to be done on every inode
153 * allocation as the fields are not initialised by slab allocation.
154 */
155int inode_init_always(struct super_block *sb, struct inode *inode)
156{
157 static const struct inode_operations empty_iops;
158 static const struct file_operations no_open_fops = {.open = no_open};
159 struct address_space *const mapping = &inode->i_data;
160
161 inode->i_sb = sb;
162 inode->i_blkbits = sb->s_blocksize_bits;
163 inode->i_flags = 0;
164 atomic64_set(&inode->i_sequence, 0);
165 atomic_set(&inode->i_count, 1);
166 inode->i_op = &empty_iops;
167 inode->i_fop = &no_open_fops;
168 inode->i_ino = 0;
169 inode->__i_nlink = 1;
170 inode->i_opflags = 0;
171 if (sb->s_xattr)
172 inode->i_opflags |= IOP_XATTR;
173 i_uid_write(inode, 0);
174 i_gid_write(inode, 0);
175 atomic_set(&inode->i_writecount, 0);
176 inode->i_size = 0;
177 inode->i_write_hint = WRITE_LIFE_NOT_SET;
178 inode->i_blocks = 0;
179 inode->i_bytes = 0;
180 inode->i_generation = 0;
181 inode->i_pipe = NULL;
182 inode->i_cdev = NULL;
183 inode->i_link = NULL;
184 inode->i_dir_seq = 0;
185 inode->i_rdev = 0;
186 inode->dirtied_when = 0;
187
188#ifdef CONFIG_CGROUP_WRITEBACK
189 inode->i_wb_frn_winner = 0;
190 inode->i_wb_frn_avg_time = 0;
191 inode->i_wb_frn_history = 0;
192#endif
193
194 spin_lock_init(&inode->i_lock);
195 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
196
197 init_rwsem(&inode->i_rwsem);
198 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
199
200 atomic_set(&inode->i_dio_count, 0);
201
202 mapping->a_ops = &empty_aops;
203 mapping->host = inode;
204 mapping->flags = 0;
205 mapping->wb_err = 0;
206 atomic_set(&mapping->i_mmap_writable, 0);
207#ifdef CONFIG_READ_ONLY_THP_FOR_FS
208 atomic_set(&mapping->nr_thps, 0);
209#endif
210 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
211 mapping->i_private_data = NULL;
212 mapping->writeback_index = 0;
213 init_rwsem(&mapping->invalidate_lock);
214 lockdep_set_class_and_name(&mapping->invalidate_lock,
215 &sb->s_type->invalidate_lock_key,
216 "mapping.invalidate_lock");
217 if (sb->s_iflags & SB_I_STABLE_WRITES)
218 mapping_set_stable_writes(mapping);
219 inode->i_private = NULL;
220 inode->i_mapping = mapping;
221 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
222#ifdef CONFIG_FS_POSIX_ACL
223 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
224#endif
225
226#ifdef CONFIG_FSNOTIFY
227 inode->i_fsnotify_mask = 0;
228#endif
229 inode->i_flctx = NULL;
230
231 if (unlikely(security_inode_alloc(inode)))
232 return -ENOMEM;
233 this_cpu_inc(nr_inodes);
234
235 return 0;
236}
237EXPORT_SYMBOL(inode_init_always);
238
239void free_inode_nonrcu(struct inode *inode)
240{
241 kmem_cache_free(inode_cachep, inode);
242}
243EXPORT_SYMBOL(free_inode_nonrcu);
244
245static void i_callback(struct rcu_head *head)
246{
247 struct inode *inode = container_of(head, struct inode, i_rcu);
248 if (inode->free_inode)
249 inode->free_inode(inode);
250 else
251 free_inode_nonrcu(inode);
252}
253
254static struct inode *alloc_inode(struct super_block *sb)
255{
256 const struct super_operations *ops = sb->s_op;
257 struct inode *inode;
258
259 if (ops->alloc_inode)
260 inode = ops->alloc_inode(sb);
261 else
262 inode = alloc_inode_sb(sb, inode_cachep, GFP_KERNEL);
263
264 if (!inode)
265 return NULL;
266
267 if (unlikely(inode_init_always(sb, inode))) {
268 if (ops->destroy_inode) {
269 ops->destroy_inode(inode);
270 if (!ops->free_inode)
271 return NULL;
272 }
273 inode->free_inode = ops->free_inode;
274 i_callback(&inode->i_rcu);
275 return NULL;
276 }
277
278 return inode;
279}
280
281void __destroy_inode(struct inode *inode)
282{
283 BUG_ON(inode_has_buffers(inode));
284 inode_detach_wb(inode);
285 security_inode_free(inode);
286 fsnotify_inode_delete(inode);
287 locks_free_lock_context(inode);
288 if (!inode->i_nlink) {
289 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
290 atomic_long_dec(&inode->i_sb->s_remove_count);
291 }
292
293#ifdef CONFIG_FS_POSIX_ACL
294 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
295 posix_acl_release(inode->i_acl);
296 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
297 posix_acl_release(inode->i_default_acl);
298#endif
299 this_cpu_dec(nr_inodes);
300}
301EXPORT_SYMBOL(__destroy_inode);
302
303static void destroy_inode(struct inode *inode)
304{
305 const struct super_operations *ops = inode->i_sb->s_op;
306
307 BUG_ON(!list_empty(&inode->i_lru));
308 __destroy_inode(inode);
309 if (ops->destroy_inode) {
310 ops->destroy_inode(inode);
311 if (!ops->free_inode)
312 return;
313 }
314 inode->free_inode = ops->free_inode;
315 call_rcu(&inode->i_rcu, i_callback);
316}
317
318/**
319 * drop_nlink - directly drop an inode's link count
320 * @inode: inode
321 *
322 * This is a low-level filesystem helper to replace any
323 * direct filesystem manipulation of i_nlink. In cases
324 * where we are attempting to track writes to the
325 * filesystem, a decrement to zero means an imminent
326 * write when the file is truncated and actually unlinked
327 * on the filesystem.
328 */
329void drop_nlink(struct inode *inode)
330{
331 WARN_ON(inode->i_nlink == 0);
332 inode->__i_nlink--;
333 if (!inode->i_nlink)
334 atomic_long_inc(&inode->i_sb->s_remove_count);
335}
336EXPORT_SYMBOL(drop_nlink);
337
338/**
339 * clear_nlink - directly zero an inode's link count
340 * @inode: inode
341 *
342 * This is a low-level filesystem helper to replace any
343 * direct filesystem manipulation of i_nlink. See
344 * drop_nlink() for why we care about i_nlink hitting zero.
345 */
346void clear_nlink(struct inode *inode)
347{
348 if (inode->i_nlink) {
349 inode->__i_nlink = 0;
350 atomic_long_inc(&inode->i_sb->s_remove_count);
351 }
352}
353EXPORT_SYMBOL(clear_nlink);
354
355/**
356 * set_nlink - directly set an inode's link count
357 * @inode: inode
358 * @nlink: new nlink (should be non-zero)
359 *
360 * This is a low-level filesystem helper to replace any
361 * direct filesystem manipulation of i_nlink.
362 */
363void set_nlink(struct inode *inode, unsigned int nlink)
364{
365 if (!nlink) {
366 clear_nlink(inode);
367 } else {
368 /* Yes, some filesystems do change nlink from zero to one */
369 if (inode->i_nlink == 0)
370 atomic_long_dec(&inode->i_sb->s_remove_count);
371
372 inode->__i_nlink = nlink;
373 }
374}
375EXPORT_SYMBOL(set_nlink);
376
377/**
378 * inc_nlink - directly increment an inode's link count
379 * @inode: inode
380 *
381 * This is a low-level filesystem helper to replace any
382 * direct filesystem manipulation of i_nlink. Currently,
383 * it is only here for parity with dec_nlink().
384 */
385void inc_nlink(struct inode *inode)
386{
387 if (unlikely(inode->i_nlink == 0)) {
388 WARN_ON(!(inode->i_state & I_LINKABLE));
389 atomic_long_dec(&inode->i_sb->s_remove_count);
390 }
391
392 inode->__i_nlink++;
393}
394EXPORT_SYMBOL(inc_nlink);
395
396static void __address_space_init_once(struct address_space *mapping)
397{
398 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
399 init_rwsem(&mapping->i_mmap_rwsem);
400 INIT_LIST_HEAD(&mapping->i_private_list);
401 spin_lock_init(&mapping->i_private_lock);
402 mapping->i_mmap = RB_ROOT_CACHED;
403}
404
405void address_space_init_once(struct address_space *mapping)
406{
407 memset(mapping, 0, sizeof(*mapping));
408 __address_space_init_once(mapping);
409}
410EXPORT_SYMBOL(address_space_init_once);
411
412/*
413 * These are initializations that only need to be done
414 * once, because the fields are idempotent across use
415 * of the inode, so let the slab aware of that.
416 */
417void inode_init_once(struct inode *inode)
418{
419 memset(inode, 0, sizeof(*inode));
420 INIT_HLIST_NODE(&inode->i_hash);
421 INIT_LIST_HEAD(&inode->i_devices);
422 INIT_LIST_HEAD(&inode->i_io_list);
423 INIT_LIST_HEAD(&inode->i_wb_list);
424 INIT_LIST_HEAD(&inode->i_lru);
425 INIT_LIST_HEAD(&inode->i_sb_list);
426 __address_space_init_once(&inode->i_data);
427 i_size_ordered_init(inode);
428}
429EXPORT_SYMBOL(inode_init_once);
430
431static void init_once(void *foo)
432{
433 struct inode *inode = (struct inode *) foo;
434
435 inode_init_once(inode);
436}
437
438/*
439 * inode->i_lock must be held
440 */
441void __iget(struct inode *inode)
442{
443 atomic_inc(&inode->i_count);
444}
445
446/*
447 * get additional reference to inode; caller must already hold one.
448 */
449void ihold(struct inode *inode)
450{
451 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
452}
453EXPORT_SYMBOL(ihold);
454
455static void __inode_add_lru(struct inode *inode, bool rotate)
456{
457 if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
458 return;
459 if (atomic_read(&inode->i_count))
460 return;
461 if (!(inode->i_sb->s_flags & SB_ACTIVE))
462 return;
463 if (!mapping_shrinkable(&inode->i_data))
464 return;
465
466 if (list_lru_add_obj(&inode->i_sb->s_inode_lru, &inode->i_lru))
467 this_cpu_inc(nr_unused);
468 else if (rotate)
469 inode->i_state |= I_REFERENCED;
470}
471
472/*
473 * Add inode to LRU if needed (inode is unused and clean).
474 *
475 * Needs inode->i_lock held.
476 */
477void inode_add_lru(struct inode *inode)
478{
479 __inode_add_lru(inode, false);
480}
481
482static void inode_lru_list_del(struct inode *inode)
483{
484 if (list_lru_del_obj(&inode->i_sb->s_inode_lru, &inode->i_lru))
485 this_cpu_dec(nr_unused);
486}
487
488/**
489 * inode_sb_list_add - add inode to the superblock list of inodes
490 * @inode: inode to add
491 */
492void inode_sb_list_add(struct inode *inode)
493{
494 spin_lock(&inode->i_sb->s_inode_list_lock);
495 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
496 spin_unlock(&inode->i_sb->s_inode_list_lock);
497}
498EXPORT_SYMBOL_GPL(inode_sb_list_add);
499
500static inline void inode_sb_list_del(struct inode *inode)
501{
502 if (!list_empty(&inode->i_sb_list)) {
503 spin_lock(&inode->i_sb->s_inode_list_lock);
504 list_del_init(&inode->i_sb_list);
505 spin_unlock(&inode->i_sb->s_inode_list_lock);
506 }
507}
508
509static unsigned long hash(struct super_block *sb, unsigned long hashval)
510{
511 unsigned long tmp;
512
513 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
514 L1_CACHE_BYTES;
515 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
516 return tmp & i_hash_mask;
517}
518
519/**
520 * __insert_inode_hash - hash an inode
521 * @inode: unhashed inode
522 * @hashval: unsigned long value used to locate this object in the
523 * inode_hashtable.
524 *
525 * Add an inode to the inode hash for this superblock.
526 */
527void __insert_inode_hash(struct inode *inode, unsigned long hashval)
528{
529 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
530
531 spin_lock(&inode_hash_lock);
532 spin_lock(&inode->i_lock);
533 hlist_add_head_rcu(&inode->i_hash, b);
534 spin_unlock(&inode->i_lock);
535 spin_unlock(&inode_hash_lock);
536}
537EXPORT_SYMBOL(__insert_inode_hash);
538
539/**
540 * __remove_inode_hash - remove an inode from the hash
541 * @inode: inode to unhash
542 *
543 * Remove an inode from the superblock.
544 */
545void __remove_inode_hash(struct inode *inode)
546{
547 spin_lock(&inode_hash_lock);
548 spin_lock(&inode->i_lock);
549 hlist_del_init_rcu(&inode->i_hash);
550 spin_unlock(&inode->i_lock);
551 spin_unlock(&inode_hash_lock);
552}
553EXPORT_SYMBOL(__remove_inode_hash);
554
555void dump_mapping(const struct address_space *mapping)
556{
557 struct inode *host;
558 const struct address_space_operations *a_ops;
559 struct hlist_node *dentry_first;
560 struct dentry *dentry_ptr;
561 struct dentry dentry;
562 unsigned long ino;
563
564 /*
565 * If mapping is an invalid pointer, we don't want to crash
566 * accessing it, so probe everything depending on it carefully.
567 */
568 if (get_kernel_nofault(host, &mapping->host) ||
569 get_kernel_nofault(a_ops, &mapping->a_ops)) {
570 pr_warn("invalid mapping:%px\n", mapping);
571 return;
572 }
573
574 if (!host) {
575 pr_warn("aops:%ps\n", a_ops);
576 return;
577 }
578
579 if (get_kernel_nofault(dentry_first, &host->i_dentry.first) ||
580 get_kernel_nofault(ino, &host->i_ino)) {
581 pr_warn("aops:%ps invalid inode:%px\n", a_ops, host);
582 return;
583 }
584
585 if (!dentry_first) {
586 pr_warn("aops:%ps ino:%lx\n", a_ops, ino);
587 return;
588 }
589
590 dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
591 if (get_kernel_nofault(dentry, dentry_ptr)) {
592 pr_warn("aops:%ps ino:%lx invalid dentry:%px\n",
593 a_ops, ino, dentry_ptr);
594 return;
595 }
596
597 /*
598 * if dentry is corrupted, the %pd handler may still crash,
599 * but it's unlikely that we reach here with a corrupt mapping
600 */
601 pr_warn("aops:%ps ino:%lx dentry name:\"%pd\"\n", a_ops, ino, &dentry);
602}
603
604void clear_inode(struct inode *inode)
605{
606 /*
607 * We have to cycle the i_pages lock here because reclaim can be in the
608 * process of removing the last page (in __filemap_remove_folio())
609 * and we must not free the mapping under it.
610 */
611 xa_lock_irq(&inode->i_data.i_pages);
612 BUG_ON(inode->i_data.nrpages);
613 /*
614 * Almost always, mapping_empty(&inode->i_data) here; but there are
615 * two known and long-standing ways in which nodes may get left behind
616 * (when deep radix-tree node allocation failed partway; or when THP
617 * collapse_file() failed). Until those two known cases are cleaned up,
618 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
619 * nor even WARN_ON(!mapping_empty).
620 */
621 xa_unlock_irq(&inode->i_data.i_pages);
622 BUG_ON(!list_empty(&inode->i_data.i_private_list));
623 BUG_ON(!(inode->i_state & I_FREEING));
624 BUG_ON(inode->i_state & I_CLEAR);
625 BUG_ON(!list_empty(&inode->i_wb_list));
626 /* don't need i_lock here, no concurrent mods to i_state */
627 inode->i_state = I_FREEING | I_CLEAR;
628}
629EXPORT_SYMBOL(clear_inode);
630
631/*
632 * Free the inode passed in, removing it from the lists it is still connected
633 * to. We remove any pages still attached to the inode and wait for any IO that
634 * is still in progress before finally destroying the inode.
635 *
636 * An inode must already be marked I_FREEING so that we avoid the inode being
637 * moved back onto lists if we race with other code that manipulates the lists
638 * (e.g. writeback_single_inode). The caller is responsible for setting this.
639 *
640 * An inode must already be removed from the LRU list before being evicted from
641 * the cache. This should occur atomically with setting the I_FREEING state
642 * flag, so no inodes here should ever be on the LRU when being evicted.
643 */
644static void evict(struct inode *inode)
645{
646 const struct super_operations *op = inode->i_sb->s_op;
647
648 BUG_ON(!(inode->i_state & I_FREEING));
649 BUG_ON(!list_empty(&inode->i_lru));
650
651 if (!list_empty(&inode->i_io_list))
652 inode_io_list_del(inode);
653
654 inode_sb_list_del(inode);
655
656 /*
657 * Wait for flusher thread to be done with the inode so that filesystem
658 * does not start destroying it while writeback is still running. Since
659 * the inode has I_FREEING set, flusher thread won't start new work on
660 * the inode. We just have to wait for running writeback to finish.
661 */
662 inode_wait_for_writeback(inode);
663
664 if (op->evict_inode) {
665 op->evict_inode(inode);
666 } else {
667 truncate_inode_pages_final(&inode->i_data);
668 clear_inode(inode);
669 }
670 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
671 cd_forget(inode);
672
673 remove_inode_hash(inode);
674
675 spin_lock(&inode->i_lock);
676 wake_up_bit(&inode->i_state, __I_NEW);
677 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
678 spin_unlock(&inode->i_lock);
679
680 destroy_inode(inode);
681}
682
683/*
684 * dispose_list - dispose of the contents of a local list
685 * @head: the head of the list to free
686 *
687 * Dispose-list gets a local list with local inodes in it, so it doesn't
688 * need to worry about list corruption and SMP locks.
689 */
690static void dispose_list(struct list_head *head)
691{
692 while (!list_empty(head)) {
693 struct inode *inode;
694
695 inode = list_first_entry(head, struct inode, i_lru);
696 list_del_init(&inode->i_lru);
697
698 evict(inode);
699 cond_resched();
700 }
701}
702
703/**
704 * evict_inodes - evict all evictable inodes for a superblock
705 * @sb: superblock to operate on
706 *
707 * Make sure that no inodes with zero refcount are retained. This is
708 * called by superblock shutdown after having SB_ACTIVE flag removed,
709 * so any inode reaching zero refcount during or after that call will
710 * be immediately evicted.
711 */
712void evict_inodes(struct super_block *sb)
713{
714 struct inode *inode, *next;
715 LIST_HEAD(dispose);
716
717again:
718 spin_lock(&sb->s_inode_list_lock);
719 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
720 if (atomic_read(&inode->i_count))
721 continue;
722
723 spin_lock(&inode->i_lock);
724 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
725 spin_unlock(&inode->i_lock);
726 continue;
727 }
728
729 inode->i_state |= I_FREEING;
730 inode_lru_list_del(inode);
731 spin_unlock(&inode->i_lock);
732 list_add(&inode->i_lru, &dispose);
733
734 /*
735 * We can have a ton of inodes to evict at unmount time given
736 * enough memory, check to see if we need to go to sleep for a
737 * bit so we don't livelock.
738 */
739 if (need_resched()) {
740 spin_unlock(&sb->s_inode_list_lock);
741 cond_resched();
742 dispose_list(&dispose);
743 goto again;
744 }
745 }
746 spin_unlock(&sb->s_inode_list_lock);
747
748 dispose_list(&dispose);
749}
750EXPORT_SYMBOL_GPL(evict_inodes);
751
752/**
753 * invalidate_inodes - attempt to free all inodes on a superblock
754 * @sb: superblock to operate on
755 *
756 * Attempts to free all inodes (including dirty inodes) for a given superblock.
757 */
758void invalidate_inodes(struct super_block *sb)
759{
760 struct inode *inode, *next;
761 LIST_HEAD(dispose);
762
763again:
764 spin_lock(&sb->s_inode_list_lock);
765 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
766 spin_lock(&inode->i_lock);
767 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
768 spin_unlock(&inode->i_lock);
769 continue;
770 }
771 if (atomic_read(&inode->i_count)) {
772 spin_unlock(&inode->i_lock);
773 continue;
774 }
775
776 inode->i_state |= I_FREEING;
777 inode_lru_list_del(inode);
778 spin_unlock(&inode->i_lock);
779 list_add(&inode->i_lru, &dispose);
780 if (need_resched()) {
781 spin_unlock(&sb->s_inode_list_lock);
782 cond_resched();
783 dispose_list(&dispose);
784 goto again;
785 }
786 }
787 spin_unlock(&sb->s_inode_list_lock);
788
789 dispose_list(&dispose);
790}
791
792/*
793 * Isolate the inode from the LRU in preparation for freeing it.
794 *
795 * If the inode has the I_REFERENCED flag set, then it means that it has been
796 * used recently - the flag is set in iput_final(). When we encounter such an
797 * inode, clear the flag and move it to the back of the LRU so it gets another
798 * pass through the LRU before it gets reclaimed. This is necessary because of
799 * the fact we are doing lazy LRU updates to minimise lock contention so the
800 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
801 * with this flag set because they are the inodes that are out of order.
802 */
803static enum lru_status inode_lru_isolate(struct list_head *item,
804 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
805{
806 struct list_head *freeable = arg;
807 struct inode *inode = container_of(item, struct inode, i_lru);
808
809 /*
810 * We are inverting the lru lock/inode->i_lock here, so use a
811 * trylock. If we fail to get the lock, just skip it.
812 */
813 if (!spin_trylock(&inode->i_lock))
814 return LRU_SKIP;
815
816 /*
817 * Inodes can get referenced, redirtied, or repopulated while
818 * they're already on the LRU, and this can make them
819 * unreclaimable for a while. Remove them lazily here; iput,
820 * sync, or the last page cache deletion will requeue them.
821 */
822 if (atomic_read(&inode->i_count) ||
823 (inode->i_state & ~I_REFERENCED) ||
824 !mapping_shrinkable(&inode->i_data)) {
825 list_lru_isolate(lru, &inode->i_lru);
826 spin_unlock(&inode->i_lock);
827 this_cpu_dec(nr_unused);
828 return LRU_REMOVED;
829 }
830
831 /* Recently referenced inodes get one more pass */
832 if (inode->i_state & I_REFERENCED) {
833 inode->i_state &= ~I_REFERENCED;
834 spin_unlock(&inode->i_lock);
835 return LRU_ROTATE;
836 }
837
838 /*
839 * On highmem systems, mapping_shrinkable() permits dropping
840 * page cache in order to free up struct inodes: lowmem might
841 * be under pressure before the cache inside the highmem zone.
842 */
843 if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
844 __iget(inode);
845 spin_unlock(&inode->i_lock);
846 spin_unlock(lru_lock);
847 if (remove_inode_buffers(inode)) {
848 unsigned long reap;
849 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
850 if (current_is_kswapd())
851 __count_vm_events(KSWAPD_INODESTEAL, reap);
852 else
853 __count_vm_events(PGINODESTEAL, reap);
854 mm_account_reclaimed_pages(reap);
855 }
856 iput(inode);
857 spin_lock(lru_lock);
858 return LRU_RETRY;
859 }
860
861 WARN_ON(inode->i_state & I_NEW);
862 inode->i_state |= I_FREEING;
863 list_lru_isolate_move(lru, &inode->i_lru, freeable);
864 spin_unlock(&inode->i_lock);
865
866 this_cpu_dec(nr_unused);
867 return LRU_REMOVED;
868}
869
870/*
871 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
872 * This is called from the superblock shrinker function with a number of inodes
873 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
874 * then are freed outside inode_lock by dispose_list().
875 */
876long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
877{
878 LIST_HEAD(freeable);
879 long freed;
880
881 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
882 inode_lru_isolate, &freeable);
883 dispose_list(&freeable);
884 return freed;
885}
886
887static void __wait_on_freeing_inode(struct inode *inode);
888/*
889 * Called with the inode lock held.
890 */
891static struct inode *find_inode(struct super_block *sb,
892 struct hlist_head *head,
893 int (*test)(struct inode *, void *),
894 void *data)
895{
896 struct inode *inode = NULL;
897
898repeat:
899 hlist_for_each_entry(inode, head, i_hash) {
900 if (inode->i_sb != sb)
901 continue;
902 if (!test(inode, data))
903 continue;
904 spin_lock(&inode->i_lock);
905 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
906 __wait_on_freeing_inode(inode);
907 goto repeat;
908 }
909 if (unlikely(inode->i_state & I_CREATING)) {
910 spin_unlock(&inode->i_lock);
911 return ERR_PTR(-ESTALE);
912 }
913 __iget(inode);
914 spin_unlock(&inode->i_lock);
915 return inode;
916 }
917 return NULL;
918}
919
920/*
921 * find_inode_fast is the fast path version of find_inode, see the comment at
922 * iget_locked for details.
923 */
924static struct inode *find_inode_fast(struct super_block *sb,
925 struct hlist_head *head, unsigned long ino)
926{
927 struct inode *inode = NULL;
928
929repeat:
930 hlist_for_each_entry(inode, head, i_hash) {
931 if (inode->i_ino != ino)
932 continue;
933 if (inode->i_sb != sb)
934 continue;
935 spin_lock(&inode->i_lock);
936 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
937 __wait_on_freeing_inode(inode);
938 goto repeat;
939 }
940 if (unlikely(inode->i_state & I_CREATING)) {
941 spin_unlock(&inode->i_lock);
942 return ERR_PTR(-ESTALE);
943 }
944 __iget(inode);
945 spin_unlock(&inode->i_lock);
946 return inode;
947 }
948 return NULL;
949}
950
951/*
952 * Each cpu owns a range of LAST_INO_BATCH numbers.
953 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
954 * to renew the exhausted range.
955 *
956 * This does not significantly increase overflow rate because every CPU can
957 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
958 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
959 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
960 * overflow rate by 2x, which does not seem too significant.
961 *
962 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
963 * error if st_ino won't fit in target struct field. Use 32bit counter
964 * here to attempt to avoid that.
965 */
966#define LAST_INO_BATCH 1024
967static DEFINE_PER_CPU(unsigned int, last_ino);
968
969unsigned int get_next_ino(void)
970{
971 unsigned int *p = &get_cpu_var(last_ino);
972 unsigned int res = *p;
973
974#ifdef CONFIG_SMP
975 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
976 static atomic_t shared_last_ino;
977 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
978
979 res = next - LAST_INO_BATCH;
980 }
981#endif
982
983 res++;
984 /* get_next_ino should not provide a 0 inode number */
985 if (unlikely(!res))
986 res++;
987 *p = res;
988 put_cpu_var(last_ino);
989 return res;
990}
991EXPORT_SYMBOL(get_next_ino);
992
993/**
994 * new_inode_pseudo - obtain an inode
995 * @sb: superblock
996 *
997 * Allocates a new inode for given superblock.
998 * Inode wont be chained in superblock s_inodes list
999 * This means :
1000 * - fs can't be unmount
1001 * - quotas, fsnotify, writeback can't work
1002 */
1003struct inode *new_inode_pseudo(struct super_block *sb)
1004{
1005 struct inode *inode = alloc_inode(sb);
1006
1007 if (inode) {
1008 spin_lock(&inode->i_lock);
1009 inode->i_state = 0;
1010 spin_unlock(&inode->i_lock);
1011 }
1012 return inode;
1013}
1014
1015/**
1016 * new_inode - obtain an inode
1017 * @sb: superblock
1018 *
1019 * Allocates a new inode for given superblock. The default gfp_mask
1020 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1021 * If HIGHMEM pages are unsuitable or it is known that pages allocated
1022 * for the page cache are not reclaimable or migratable,
1023 * mapping_set_gfp_mask() must be called with suitable flags on the
1024 * newly created inode's mapping
1025 *
1026 */
1027struct inode *new_inode(struct super_block *sb)
1028{
1029 struct inode *inode;
1030
1031 inode = new_inode_pseudo(sb);
1032 if (inode)
1033 inode_sb_list_add(inode);
1034 return inode;
1035}
1036EXPORT_SYMBOL(new_inode);
1037
1038#ifdef CONFIG_DEBUG_LOCK_ALLOC
1039void lockdep_annotate_inode_mutex_key(struct inode *inode)
1040{
1041 if (S_ISDIR(inode->i_mode)) {
1042 struct file_system_type *type = inode->i_sb->s_type;
1043
1044 /* Set new key only if filesystem hasn't already changed it */
1045 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
1046 /*
1047 * ensure nobody is actually holding i_mutex
1048 */
1049 // mutex_destroy(&inode->i_mutex);
1050 init_rwsem(&inode->i_rwsem);
1051 lockdep_set_class(&inode->i_rwsem,
1052 &type->i_mutex_dir_key);
1053 }
1054 }
1055}
1056EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1057#endif
1058
1059/**
1060 * unlock_new_inode - clear the I_NEW state and wake up any waiters
1061 * @inode: new inode to unlock
1062 *
1063 * Called when the inode is fully initialised to clear the new state of the
1064 * inode and wake up anyone waiting for the inode to finish initialisation.
1065 */
1066void unlock_new_inode(struct inode *inode)
1067{
1068 lockdep_annotate_inode_mutex_key(inode);
1069 spin_lock(&inode->i_lock);
1070 WARN_ON(!(inode->i_state & I_NEW));
1071 inode->i_state &= ~I_NEW & ~I_CREATING;
1072 smp_mb();
1073 wake_up_bit(&inode->i_state, __I_NEW);
1074 spin_unlock(&inode->i_lock);
1075}
1076EXPORT_SYMBOL(unlock_new_inode);
1077
1078void discard_new_inode(struct inode *inode)
1079{
1080 lockdep_annotate_inode_mutex_key(inode);
1081 spin_lock(&inode->i_lock);
1082 WARN_ON(!(inode->i_state & I_NEW));
1083 inode->i_state &= ~I_NEW;
1084 smp_mb();
1085 wake_up_bit(&inode->i_state, __I_NEW);
1086 spin_unlock(&inode->i_lock);
1087 iput(inode);
1088}
1089EXPORT_SYMBOL(discard_new_inode);
1090
1091/**
1092 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1093 *
1094 * Lock any non-NULL argument. Passed objects must not be directories.
1095 * Zero, one or two objects may be locked by this function.
1096 *
1097 * @inode1: first inode to lock
1098 * @inode2: second inode to lock
1099 */
1100void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1101{
1102 if (inode1)
1103 WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1104 if (inode2)
1105 WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1106 if (inode1 > inode2)
1107 swap(inode1, inode2);
1108 if (inode1)
1109 inode_lock(inode1);
1110 if (inode2 && inode2 != inode1)
1111 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1112}
1113EXPORT_SYMBOL(lock_two_nondirectories);
1114
1115/**
1116 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1117 * @inode1: first inode to unlock
1118 * @inode2: second inode to unlock
1119 */
1120void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1121{
1122 if (inode1) {
1123 WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1124 inode_unlock(inode1);
1125 }
1126 if (inode2 && inode2 != inode1) {
1127 WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1128 inode_unlock(inode2);
1129 }
1130}
1131EXPORT_SYMBOL(unlock_two_nondirectories);
1132
1133/**
1134 * inode_insert5 - obtain an inode from a mounted file system
1135 * @inode: pre-allocated inode to use for insert to cache
1136 * @hashval: hash value (usually inode number) to get
1137 * @test: callback used for comparisons between inodes
1138 * @set: callback used to initialize a new struct inode
1139 * @data: opaque data pointer to pass to @test and @set
1140 *
1141 * Search for the inode specified by @hashval and @data in the inode cache,
1142 * and if present it is return it with an increased reference count. This is
1143 * a variant of iget5_locked() for callers that don't want to fail on memory
1144 * allocation of inode.
1145 *
1146 * If the inode is not in cache, insert the pre-allocated inode to cache and
1147 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1148 * to fill it in before unlocking it via unlock_new_inode().
1149 *
1150 * Note both @test and @set are called with the inode_hash_lock held, so can't
1151 * sleep.
1152 */
1153struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1154 int (*test)(struct inode *, void *),
1155 int (*set)(struct inode *, void *), void *data)
1156{
1157 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1158 struct inode *old;
1159
1160again:
1161 spin_lock(&inode_hash_lock);
1162 old = find_inode(inode->i_sb, head, test, data);
1163 if (unlikely(old)) {
1164 /*
1165 * Uhhuh, somebody else created the same inode under us.
1166 * Use the old inode instead of the preallocated one.
1167 */
1168 spin_unlock(&inode_hash_lock);
1169 if (IS_ERR(old))
1170 return NULL;
1171 wait_on_inode(old);
1172 if (unlikely(inode_unhashed(old))) {
1173 iput(old);
1174 goto again;
1175 }
1176 return old;
1177 }
1178
1179 if (set && unlikely(set(inode, data))) {
1180 inode = NULL;
1181 goto unlock;
1182 }
1183
1184 /*
1185 * Return the locked inode with I_NEW set, the
1186 * caller is responsible for filling in the contents
1187 */
1188 spin_lock(&inode->i_lock);
1189 inode->i_state |= I_NEW;
1190 hlist_add_head_rcu(&inode->i_hash, head);
1191 spin_unlock(&inode->i_lock);
1192
1193 /*
1194 * Add inode to the sb list if it's not already. It has I_NEW at this
1195 * point, so it should be safe to test i_sb_list locklessly.
1196 */
1197 if (list_empty(&inode->i_sb_list))
1198 inode_sb_list_add(inode);
1199unlock:
1200 spin_unlock(&inode_hash_lock);
1201
1202 return inode;
1203}
1204EXPORT_SYMBOL(inode_insert5);
1205
1206/**
1207 * iget5_locked - obtain an inode from a mounted file system
1208 * @sb: super block of file system
1209 * @hashval: hash value (usually inode number) to get
1210 * @test: callback used for comparisons between inodes
1211 * @set: callback used to initialize a new struct inode
1212 * @data: opaque data pointer to pass to @test and @set
1213 *
1214 * Search for the inode specified by @hashval and @data in the inode cache,
1215 * and if present it is return it with an increased reference count. This is
1216 * a generalized version of iget_locked() for file systems where the inode
1217 * number is not sufficient for unique identification of an inode.
1218 *
1219 * If the inode is not in cache, allocate a new inode and return it locked,
1220 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1221 * before unlocking it via unlock_new_inode().
1222 *
1223 * Note both @test and @set are called with the inode_hash_lock held, so can't
1224 * sleep.
1225 */
1226struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1227 int (*test)(struct inode *, void *),
1228 int (*set)(struct inode *, void *), void *data)
1229{
1230 struct inode *inode = ilookup5(sb, hashval, test, data);
1231
1232 if (!inode) {
1233 struct inode *new = alloc_inode(sb);
1234
1235 if (new) {
1236 new->i_state = 0;
1237 inode = inode_insert5(new, hashval, test, set, data);
1238 if (unlikely(inode != new))
1239 destroy_inode(new);
1240 }
1241 }
1242 return inode;
1243}
1244EXPORT_SYMBOL(iget5_locked);
1245
1246/**
1247 * iget_locked - obtain an inode from a mounted file system
1248 * @sb: super block of file system
1249 * @ino: inode number to get
1250 *
1251 * Search for the inode specified by @ino in the inode cache and if present
1252 * return it with an increased reference count. This is for file systems
1253 * where the inode number is sufficient for unique identification of an inode.
1254 *
1255 * If the inode is not in cache, allocate a new inode and return it locked,
1256 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1257 * before unlocking it via unlock_new_inode().
1258 */
1259struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1260{
1261 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1262 struct inode *inode;
1263again:
1264 spin_lock(&inode_hash_lock);
1265 inode = find_inode_fast(sb, head, ino);
1266 spin_unlock(&inode_hash_lock);
1267 if (inode) {
1268 if (IS_ERR(inode))
1269 return NULL;
1270 wait_on_inode(inode);
1271 if (unlikely(inode_unhashed(inode))) {
1272 iput(inode);
1273 goto again;
1274 }
1275 return inode;
1276 }
1277
1278 inode = alloc_inode(sb);
1279 if (inode) {
1280 struct inode *old;
1281
1282 spin_lock(&inode_hash_lock);
1283 /* We released the lock, so.. */
1284 old = find_inode_fast(sb, head, ino);
1285 if (!old) {
1286 inode->i_ino = ino;
1287 spin_lock(&inode->i_lock);
1288 inode->i_state = I_NEW;
1289 hlist_add_head_rcu(&inode->i_hash, head);
1290 spin_unlock(&inode->i_lock);
1291 inode_sb_list_add(inode);
1292 spin_unlock(&inode_hash_lock);
1293
1294 /* Return the locked inode with I_NEW set, the
1295 * caller is responsible for filling in the contents
1296 */
1297 return inode;
1298 }
1299
1300 /*
1301 * Uhhuh, somebody else created the same inode under
1302 * us. Use the old inode instead of the one we just
1303 * allocated.
1304 */
1305 spin_unlock(&inode_hash_lock);
1306 destroy_inode(inode);
1307 if (IS_ERR(old))
1308 return NULL;
1309 inode = old;
1310 wait_on_inode(inode);
1311 if (unlikely(inode_unhashed(inode))) {
1312 iput(inode);
1313 goto again;
1314 }
1315 }
1316 return inode;
1317}
1318EXPORT_SYMBOL(iget_locked);
1319
1320/*
1321 * search the inode cache for a matching inode number.
1322 * If we find one, then the inode number we are trying to
1323 * allocate is not unique and so we should not use it.
1324 *
1325 * Returns 1 if the inode number is unique, 0 if it is not.
1326 */
1327static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1328{
1329 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1330 struct inode *inode;
1331
1332 hlist_for_each_entry_rcu(inode, b, i_hash) {
1333 if (inode->i_ino == ino && inode->i_sb == sb)
1334 return 0;
1335 }
1336 return 1;
1337}
1338
1339/**
1340 * iunique - get a unique inode number
1341 * @sb: superblock
1342 * @max_reserved: highest reserved inode number
1343 *
1344 * Obtain an inode number that is unique on the system for a given
1345 * superblock. This is used by file systems that have no natural
1346 * permanent inode numbering system. An inode number is returned that
1347 * is higher than the reserved limit but unique.
1348 *
1349 * BUGS:
1350 * With a large number of inodes live on the file system this function
1351 * currently becomes quite slow.
1352 */
1353ino_t iunique(struct super_block *sb, ino_t max_reserved)
1354{
1355 /*
1356 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1357 * error if st_ino won't fit in target struct field. Use 32bit counter
1358 * here to attempt to avoid that.
1359 */
1360 static DEFINE_SPINLOCK(iunique_lock);
1361 static unsigned int counter;
1362 ino_t res;
1363
1364 rcu_read_lock();
1365 spin_lock(&iunique_lock);
1366 do {
1367 if (counter <= max_reserved)
1368 counter = max_reserved + 1;
1369 res = counter++;
1370 } while (!test_inode_iunique(sb, res));
1371 spin_unlock(&iunique_lock);
1372 rcu_read_unlock();
1373
1374 return res;
1375}
1376EXPORT_SYMBOL(iunique);
1377
1378struct inode *igrab(struct inode *inode)
1379{
1380 spin_lock(&inode->i_lock);
1381 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1382 __iget(inode);
1383 spin_unlock(&inode->i_lock);
1384 } else {
1385 spin_unlock(&inode->i_lock);
1386 /*
1387 * Handle the case where s_op->clear_inode is not been
1388 * called yet, and somebody is calling igrab
1389 * while the inode is getting freed.
1390 */
1391 inode = NULL;
1392 }
1393 return inode;
1394}
1395EXPORT_SYMBOL(igrab);
1396
1397/**
1398 * ilookup5_nowait - search for an inode in the inode cache
1399 * @sb: super block of file system to search
1400 * @hashval: hash value (usually inode number) to search for
1401 * @test: callback used for comparisons between inodes
1402 * @data: opaque data pointer to pass to @test
1403 *
1404 * Search for the inode specified by @hashval and @data in the inode cache.
1405 * If the inode is in the cache, the inode is returned with an incremented
1406 * reference count.
1407 *
1408 * Note: I_NEW is not waited upon so you have to be very careful what you do
1409 * with the returned inode. You probably should be using ilookup5() instead.
1410 *
1411 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1412 */
1413struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1414 int (*test)(struct inode *, void *), void *data)
1415{
1416 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1417 struct inode *inode;
1418
1419 spin_lock(&inode_hash_lock);
1420 inode = find_inode(sb, head, test, data);
1421 spin_unlock(&inode_hash_lock);
1422
1423 return IS_ERR(inode) ? NULL : inode;
1424}
1425EXPORT_SYMBOL(ilookup5_nowait);
1426
1427/**
1428 * ilookup5 - search for an inode in the inode cache
1429 * @sb: super block of file system to search
1430 * @hashval: hash value (usually inode number) to search for
1431 * @test: callback used for comparisons between inodes
1432 * @data: opaque data pointer to pass to @test
1433 *
1434 * Search for the inode specified by @hashval and @data in the inode cache,
1435 * and if the inode is in the cache, return the inode with an incremented
1436 * reference count. Waits on I_NEW before returning the inode.
1437 * returned with an incremented reference count.
1438 *
1439 * This is a generalized version of ilookup() for file systems where the
1440 * inode number is not sufficient for unique identification of an inode.
1441 *
1442 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1443 */
1444struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1445 int (*test)(struct inode *, void *), void *data)
1446{
1447 struct inode *inode;
1448again:
1449 inode = ilookup5_nowait(sb, hashval, test, data);
1450 if (inode) {
1451 wait_on_inode(inode);
1452 if (unlikely(inode_unhashed(inode))) {
1453 iput(inode);
1454 goto again;
1455 }
1456 }
1457 return inode;
1458}
1459EXPORT_SYMBOL(ilookup5);
1460
1461/**
1462 * ilookup - search for an inode in the inode cache
1463 * @sb: super block of file system to search
1464 * @ino: inode number to search for
1465 *
1466 * Search for the inode @ino in the inode cache, and if the inode is in the
1467 * cache, the inode is returned with an incremented reference count.
1468 */
1469struct inode *ilookup(struct super_block *sb, unsigned long ino)
1470{
1471 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1472 struct inode *inode;
1473again:
1474 spin_lock(&inode_hash_lock);
1475 inode = find_inode_fast(sb, head, ino);
1476 spin_unlock(&inode_hash_lock);
1477
1478 if (inode) {
1479 if (IS_ERR(inode))
1480 return NULL;
1481 wait_on_inode(inode);
1482 if (unlikely(inode_unhashed(inode))) {
1483 iput(inode);
1484 goto again;
1485 }
1486 }
1487 return inode;
1488}
1489EXPORT_SYMBOL(ilookup);
1490
1491/**
1492 * find_inode_nowait - find an inode in the inode cache
1493 * @sb: super block of file system to search
1494 * @hashval: hash value (usually inode number) to search for
1495 * @match: callback used for comparisons between inodes
1496 * @data: opaque data pointer to pass to @match
1497 *
1498 * Search for the inode specified by @hashval and @data in the inode
1499 * cache, where the helper function @match will return 0 if the inode
1500 * does not match, 1 if the inode does match, and -1 if the search
1501 * should be stopped. The @match function must be responsible for
1502 * taking the i_lock spin_lock and checking i_state for an inode being
1503 * freed or being initialized, and incrementing the reference count
1504 * before returning 1. It also must not sleep, since it is called with
1505 * the inode_hash_lock spinlock held.
1506 *
1507 * This is a even more generalized version of ilookup5() when the
1508 * function must never block --- find_inode() can block in
1509 * __wait_on_freeing_inode() --- or when the caller can not increment
1510 * the reference count because the resulting iput() might cause an
1511 * inode eviction. The tradeoff is that the @match funtion must be
1512 * very carefully implemented.
1513 */
1514struct inode *find_inode_nowait(struct super_block *sb,
1515 unsigned long hashval,
1516 int (*match)(struct inode *, unsigned long,
1517 void *),
1518 void *data)
1519{
1520 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1521 struct inode *inode, *ret_inode = NULL;
1522 int mval;
1523
1524 spin_lock(&inode_hash_lock);
1525 hlist_for_each_entry(inode, head, i_hash) {
1526 if (inode->i_sb != sb)
1527 continue;
1528 mval = match(inode, hashval, data);
1529 if (mval == 0)
1530 continue;
1531 if (mval == 1)
1532 ret_inode = inode;
1533 goto out;
1534 }
1535out:
1536 spin_unlock(&inode_hash_lock);
1537 return ret_inode;
1538}
1539EXPORT_SYMBOL(find_inode_nowait);
1540
1541/**
1542 * find_inode_rcu - find an inode in the inode cache
1543 * @sb: Super block of file system to search
1544 * @hashval: Key to hash
1545 * @test: Function to test match on an inode
1546 * @data: Data for test function
1547 *
1548 * Search for the inode specified by @hashval and @data in the inode cache,
1549 * where the helper function @test will return 0 if the inode does not match
1550 * and 1 if it does. The @test function must be responsible for taking the
1551 * i_lock spin_lock and checking i_state for an inode being freed or being
1552 * initialized.
1553 *
1554 * If successful, this will return the inode for which the @test function
1555 * returned 1 and NULL otherwise.
1556 *
1557 * The @test function is not permitted to take a ref on any inode presented.
1558 * It is also not permitted to sleep.
1559 *
1560 * The caller must hold the RCU read lock.
1561 */
1562struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1563 int (*test)(struct inode *, void *), void *data)
1564{
1565 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1566 struct inode *inode;
1567
1568 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1569 "suspicious find_inode_rcu() usage");
1570
1571 hlist_for_each_entry_rcu(inode, head, i_hash) {
1572 if (inode->i_sb == sb &&
1573 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1574 test(inode, data))
1575 return inode;
1576 }
1577 return NULL;
1578}
1579EXPORT_SYMBOL(find_inode_rcu);
1580
1581/**
1582 * find_inode_by_ino_rcu - Find an inode in the inode cache
1583 * @sb: Super block of file system to search
1584 * @ino: The inode number to match
1585 *
1586 * Search for the inode specified by @hashval and @data in the inode cache,
1587 * where the helper function @test will return 0 if the inode does not match
1588 * and 1 if it does. The @test function must be responsible for taking the
1589 * i_lock spin_lock and checking i_state for an inode being freed or being
1590 * initialized.
1591 *
1592 * If successful, this will return the inode for which the @test function
1593 * returned 1 and NULL otherwise.
1594 *
1595 * The @test function is not permitted to take a ref on any inode presented.
1596 * It is also not permitted to sleep.
1597 *
1598 * The caller must hold the RCU read lock.
1599 */
1600struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1601 unsigned long ino)
1602{
1603 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1604 struct inode *inode;
1605
1606 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1607 "suspicious find_inode_by_ino_rcu() usage");
1608
1609 hlist_for_each_entry_rcu(inode, head, i_hash) {
1610 if (inode->i_ino == ino &&
1611 inode->i_sb == sb &&
1612 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1613 return inode;
1614 }
1615 return NULL;
1616}
1617EXPORT_SYMBOL(find_inode_by_ino_rcu);
1618
1619int insert_inode_locked(struct inode *inode)
1620{
1621 struct super_block *sb = inode->i_sb;
1622 ino_t ino = inode->i_ino;
1623 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1624
1625 while (1) {
1626 struct inode *old = NULL;
1627 spin_lock(&inode_hash_lock);
1628 hlist_for_each_entry(old, head, i_hash) {
1629 if (old->i_ino != ino)
1630 continue;
1631 if (old->i_sb != sb)
1632 continue;
1633 spin_lock(&old->i_lock);
1634 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1635 spin_unlock(&old->i_lock);
1636 continue;
1637 }
1638 break;
1639 }
1640 if (likely(!old)) {
1641 spin_lock(&inode->i_lock);
1642 inode->i_state |= I_NEW | I_CREATING;
1643 hlist_add_head_rcu(&inode->i_hash, head);
1644 spin_unlock(&inode->i_lock);
1645 spin_unlock(&inode_hash_lock);
1646 return 0;
1647 }
1648 if (unlikely(old->i_state & I_CREATING)) {
1649 spin_unlock(&old->i_lock);
1650 spin_unlock(&inode_hash_lock);
1651 return -EBUSY;
1652 }
1653 __iget(old);
1654 spin_unlock(&old->i_lock);
1655 spin_unlock(&inode_hash_lock);
1656 wait_on_inode(old);
1657 if (unlikely(!inode_unhashed(old))) {
1658 iput(old);
1659 return -EBUSY;
1660 }
1661 iput(old);
1662 }
1663}
1664EXPORT_SYMBOL(insert_inode_locked);
1665
1666int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1667 int (*test)(struct inode *, void *), void *data)
1668{
1669 struct inode *old;
1670
1671 inode->i_state |= I_CREATING;
1672 old = inode_insert5(inode, hashval, test, NULL, data);
1673
1674 if (old != inode) {
1675 iput(old);
1676 return -EBUSY;
1677 }
1678 return 0;
1679}
1680EXPORT_SYMBOL(insert_inode_locked4);
1681
1682
1683int generic_delete_inode(struct inode *inode)
1684{
1685 return 1;
1686}
1687EXPORT_SYMBOL(generic_delete_inode);
1688
1689/*
1690 * Called when we're dropping the last reference
1691 * to an inode.
1692 *
1693 * Call the FS "drop_inode()" function, defaulting to
1694 * the legacy UNIX filesystem behaviour. If it tells
1695 * us to evict inode, do so. Otherwise, retain inode
1696 * in cache if fs is alive, sync and evict if fs is
1697 * shutting down.
1698 */
1699static void iput_final(struct inode *inode)
1700{
1701 struct super_block *sb = inode->i_sb;
1702 const struct super_operations *op = inode->i_sb->s_op;
1703 unsigned long state;
1704 int drop;
1705
1706 WARN_ON(inode->i_state & I_NEW);
1707
1708 if (op->drop_inode)
1709 drop = op->drop_inode(inode);
1710 else
1711 drop = generic_drop_inode(inode);
1712
1713 if (!drop &&
1714 !(inode->i_state & I_DONTCACHE) &&
1715 (sb->s_flags & SB_ACTIVE)) {
1716 __inode_add_lru(inode, true);
1717 spin_unlock(&inode->i_lock);
1718 return;
1719 }
1720
1721 state = inode->i_state;
1722 if (!drop) {
1723 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1724 spin_unlock(&inode->i_lock);
1725
1726 write_inode_now(inode, 1);
1727
1728 spin_lock(&inode->i_lock);
1729 state = inode->i_state;
1730 WARN_ON(state & I_NEW);
1731 state &= ~I_WILL_FREE;
1732 }
1733
1734 WRITE_ONCE(inode->i_state, state | I_FREEING);
1735 if (!list_empty(&inode->i_lru))
1736 inode_lru_list_del(inode);
1737 spin_unlock(&inode->i_lock);
1738
1739 evict(inode);
1740}
1741
1742/**
1743 * iput - put an inode
1744 * @inode: inode to put
1745 *
1746 * Puts an inode, dropping its usage count. If the inode use count hits
1747 * zero, the inode is then freed and may also be destroyed.
1748 *
1749 * Consequently, iput() can sleep.
1750 */
1751void iput(struct inode *inode)
1752{
1753 if (!inode)
1754 return;
1755 BUG_ON(inode->i_state & I_CLEAR);
1756retry:
1757 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1758 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1759 atomic_inc(&inode->i_count);
1760 spin_unlock(&inode->i_lock);
1761 trace_writeback_lazytime_iput(inode);
1762 mark_inode_dirty_sync(inode);
1763 goto retry;
1764 }
1765 iput_final(inode);
1766 }
1767}
1768EXPORT_SYMBOL(iput);
1769
1770#ifdef CONFIG_BLOCK
1771/**
1772 * bmap - find a block number in a file
1773 * @inode: inode owning the block number being requested
1774 * @block: pointer containing the block to find
1775 *
1776 * Replaces the value in ``*block`` with the block number on the device holding
1777 * corresponding to the requested block number in the file.
1778 * That is, asked for block 4 of inode 1 the function will replace the
1779 * 4 in ``*block``, with disk block relative to the disk start that holds that
1780 * block of the file.
1781 *
1782 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1783 * hole, returns 0 and ``*block`` is also set to 0.
1784 */
1785int bmap(struct inode *inode, sector_t *block)
1786{
1787 if (!inode->i_mapping->a_ops->bmap)
1788 return -EINVAL;
1789
1790 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1791 return 0;
1792}
1793EXPORT_SYMBOL(bmap);
1794#endif
1795
1796/*
1797 * With relative atime, only update atime if the previous atime is
1798 * earlier than or equal to either the ctime or mtime,
1799 * or if at least a day has passed since the last atime update.
1800 */
1801static bool relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1802 struct timespec64 now)
1803{
1804 struct timespec64 atime, mtime, ctime;
1805
1806 if (!(mnt->mnt_flags & MNT_RELATIME))
1807 return true;
1808 /*
1809 * Is mtime younger than or equal to atime? If yes, update atime:
1810 */
1811 atime = inode_get_atime(inode);
1812 mtime = inode_get_mtime(inode);
1813 if (timespec64_compare(&mtime, &atime) >= 0)
1814 return true;
1815 /*
1816 * Is ctime younger than or equal to atime? If yes, update atime:
1817 */
1818 ctime = inode_get_ctime(inode);
1819 if (timespec64_compare(&ctime, &atime) >= 0)
1820 return true;
1821
1822 /*
1823 * Is the previous atime value older than a day? If yes,
1824 * update atime:
1825 */
1826 if ((long)(now.tv_sec - atime.tv_sec) >= 24*60*60)
1827 return true;
1828 /*
1829 * Good, we can skip the atime update:
1830 */
1831 return false;
1832}
1833
1834/**
1835 * inode_update_timestamps - update the timestamps on the inode
1836 * @inode: inode to be updated
1837 * @flags: S_* flags that needed to be updated
1838 *
1839 * The update_time function is called when an inode's timestamps need to be
1840 * updated for a read or write operation. This function handles updating the
1841 * actual timestamps. It's up to the caller to ensure that the inode is marked
1842 * dirty appropriately.
1843 *
1844 * In the case where any of S_MTIME, S_CTIME, or S_VERSION need to be updated,
1845 * attempt to update all three of them. S_ATIME updates can be handled
1846 * independently of the rest.
1847 *
1848 * Returns a set of S_* flags indicating which values changed.
1849 */
1850int inode_update_timestamps(struct inode *inode, int flags)
1851{
1852 int updated = 0;
1853 struct timespec64 now;
1854
1855 if (flags & (S_MTIME|S_CTIME|S_VERSION)) {
1856 struct timespec64 ctime = inode_get_ctime(inode);
1857 struct timespec64 mtime = inode_get_mtime(inode);
1858
1859 now = inode_set_ctime_current(inode);
1860 if (!timespec64_equal(&now, &ctime))
1861 updated |= S_CTIME;
1862 if (!timespec64_equal(&now, &mtime)) {
1863 inode_set_mtime_to_ts(inode, now);
1864 updated |= S_MTIME;
1865 }
1866 if (IS_I_VERSION(inode) && inode_maybe_inc_iversion(inode, updated))
1867 updated |= S_VERSION;
1868 } else {
1869 now = current_time(inode);
1870 }
1871
1872 if (flags & S_ATIME) {
1873 struct timespec64 atime = inode_get_atime(inode);
1874
1875 if (!timespec64_equal(&now, &atime)) {
1876 inode_set_atime_to_ts(inode, now);
1877 updated |= S_ATIME;
1878 }
1879 }
1880 return updated;
1881}
1882EXPORT_SYMBOL(inode_update_timestamps);
1883
1884/**
1885 * generic_update_time - update the timestamps on the inode
1886 * @inode: inode to be updated
1887 * @flags: S_* flags that needed to be updated
1888 *
1889 * The update_time function is called when an inode's timestamps need to be
1890 * updated for a read or write operation. In the case where any of S_MTIME, S_CTIME,
1891 * or S_VERSION need to be updated we attempt to update all three of them. S_ATIME
1892 * updates can be handled done independently of the rest.
1893 *
1894 * Returns a S_* mask indicating which fields were updated.
1895 */
1896int generic_update_time(struct inode *inode, int flags)
1897{
1898 int updated = inode_update_timestamps(inode, flags);
1899 int dirty_flags = 0;
1900
1901 if (updated & (S_ATIME|S_MTIME|S_CTIME))
1902 dirty_flags = inode->i_sb->s_flags & SB_LAZYTIME ? I_DIRTY_TIME : I_DIRTY_SYNC;
1903 if (updated & S_VERSION)
1904 dirty_flags |= I_DIRTY_SYNC;
1905 __mark_inode_dirty(inode, dirty_flags);
1906 return updated;
1907}
1908EXPORT_SYMBOL(generic_update_time);
1909
1910/*
1911 * This does the actual work of updating an inodes time or version. Must have
1912 * had called mnt_want_write() before calling this.
1913 */
1914int inode_update_time(struct inode *inode, int flags)
1915{
1916 if (inode->i_op->update_time)
1917 return inode->i_op->update_time(inode, flags);
1918 generic_update_time(inode, flags);
1919 return 0;
1920}
1921EXPORT_SYMBOL(inode_update_time);
1922
1923/**
1924 * atime_needs_update - update the access time
1925 * @path: the &struct path to update
1926 * @inode: inode to update
1927 *
1928 * Update the accessed time on an inode and mark it for writeback.
1929 * This function automatically handles read only file systems and media,
1930 * as well as the "noatime" flag and inode specific "noatime" markers.
1931 */
1932bool atime_needs_update(const struct path *path, struct inode *inode)
1933{
1934 struct vfsmount *mnt = path->mnt;
1935 struct timespec64 now, atime;
1936
1937 if (inode->i_flags & S_NOATIME)
1938 return false;
1939
1940 /* Atime updates will likely cause i_uid and i_gid to be written
1941 * back improprely if their true value is unknown to the vfs.
1942 */
1943 if (HAS_UNMAPPED_ID(mnt_idmap(mnt), inode))
1944 return false;
1945
1946 if (IS_NOATIME(inode))
1947 return false;
1948 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1949 return false;
1950
1951 if (mnt->mnt_flags & MNT_NOATIME)
1952 return false;
1953 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1954 return false;
1955
1956 now = current_time(inode);
1957
1958 if (!relatime_need_update(mnt, inode, now))
1959 return false;
1960
1961 atime = inode_get_atime(inode);
1962 if (timespec64_equal(&atime, &now))
1963 return false;
1964
1965 return true;
1966}
1967
1968void touch_atime(const struct path *path)
1969{
1970 struct vfsmount *mnt = path->mnt;
1971 struct inode *inode = d_inode(path->dentry);
1972
1973 if (!atime_needs_update(path, inode))
1974 return;
1975
1976 if (!sb_start_write_trylock(inode->i_sb))
1977 return;
1978
1979 if (mnt_get_write_access(mnt) != 0)
1980 goto skip_update;
1981 /*
1982 * File systems can error out when updating inodes if they need to
1983 * allocate new space to modify an inode (such is the case for
1984 * Btrfs), but since we touch atime while walking down the path we
1985 * really don't care if we failed to update the atime of the file,
1986 * so just ignore the return value.
1987 * We may also fail on filesystems that have the ability to make parts
1988 * of the fs read only, e.g. subvolumes in Btrfs.
1989 */
1990 inode_update_time(inode, S_ATIME);
1991 mnt_put_write_access(mnt);
1992skip_update:
1993 sb_end_write(inode->i_sb);
1994}
1995EXPORT_SYMBOL(touch_atime);
1996
1997/*
1998 * Return mask of changes for notify_change() that need to be done as a
1999 * response to write or truncate. Return 0 if nothing has to be changed.
2000 * Negative value on error (change should be denied).
2001 */
2002int dentry_needs_remove_privs(struct mnt_idmap *idmap,
2003 struct dentry *dentry)
2004{
2005 struct inode *inode = d_inode(dentry);
2006 int mask = 0;
2007 int ret;
2008
2009 if (IS_NOSEC(inode))
2010 return 0;
2011
2012 mask = setattr_should_drop_suidgid(idmap, inode);
2013 ret = security_inode_need_killpriv(dentry);
2014 if (ret < 0)
2015 return ret;
2016 if (ret)
2017 mask |= ATTR_KILL_PRIV;
2018 return mask;
2019}
2020
2021static int __remove_privs(struct mnt_idmap *idmap,
2022 struct dentry *dentry, int kill)
2023{
2024 struct iattr newattrs;
2025
2026 newattrs.ia_valid = ATTR_FORCE | kill;
2027 /*
2028 * Note we call this on write, so notify_change will not
2029 * encounter any conflicting delegations:
2030 */
2031 return notify_change(idmap, dentry, &newattrs, NULL);
2032}
2033
2034static int __file_remove_privs(struct file *file, unsigned int flags)
2035{
2036 struct dentry *dentry = file_dentry(file);
2037 struct inode *inode = file_inode(file);
2038 int error = 0;
2039 int kill;
2040
2041 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
2042 return 0;
2043
2044 kill = dentry_needs_remove_privs(file_mnt_idmap(file), dentry);
2045 if (kill < 0)
2046 return kill;
2047
2048 if (kill) {
2049 if (flags & IOCB_NOWAIT)
2050 return -EAGAIN;
2051
2052 error = __remove_privs(file_mnt_idmap(file), dentry, kill);
2053 }
2054
2055 if (!error)
2056 inode_has_no_xattr(inode);
2057 return error;
2058}
2059
2060/**
2061 * file_remove_privs - remove special file privileges (suid, capabilities)
2062 * @file: file to remove privileges from
2063 *
2064 * When file is modified by a write or truncation ensure that special
2065 * file privileges are removed.
2066 *
2067 * Return: 0 on success, negative errno on failure.
2068 */
2069int file_remove_privs(struct file *file)
2070{
2071 return __file_remove_privs(file, 0);
2072}
2073EXPORT_SYMBOL(file_remove_privs);
2074
2075static int inode_needs_update_time(struct inode *inode)
2076{
2077 int sync_it = 0;
2078 struct timespec64 now = current_time(inode);
2079 struct timespec64 ts;
2080
2081 /* First try to exhaust all avenues to not sync */
2082 if (IS_NOCMTIME(inode))
2083 return 0;
2084
2085 ts = inode_get_mtime(inode);
2086 if (!timespec64_equal(&ts, &now))
2087 sync_it = S_MTIME;
2088
2089 ts = inode_get_ctime(inode);
2090 if (!timespec64_equal(&ts, &now))
2091 sync_it |= S_CTIME;
2092
2093 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2094 sync_it |= S_VERSION;
2095
2096 return sync_it;
2097}
2098
2099static int __file_update_time(struct file *file, int sync_mode)
2100{
2101 int ret = 0;
2102 struct inode *inode = file_inode(file);
2103
2104 /* try to update time settings */
2105 if (!mnt_get_write_access_file(file)) {
2106 ret = inode_update_time(inode, sync_mode);
2107 mnt_put_write_access_file(file);
2108 }
2109
2110 return ret;
2111}
2112
2113/**
2114 * file_update_time - update mtime and ctime time
2115 * @file: file accessed
2116 *
2117 * Update the mtime and ctime members of an inode and mark the inode for
2118 * writeback. Note that this function is meant exclusively for usage in
2119 * the file write path of filesystems, and filesystems may choose to
2120 * explicitly ignore updates via this function with the _NOCMTIME inode
2121 * flag, e.g. for network filesystem where these imestamps are handled
2122 * by the server. This can return an error for file systems who need to
2123 * allocate space in order to update an inode.
2124 *
2125 * Return: 0 on success, negative errno on failure.
2126 */
2127int file_update_time(struct file *file)
2128{
2129 int ret;
2130 struct inode *inode = file_inode(file);
2131
2132 ret = inode_needs_update_time(inode);
2133 if (ret <= 0)
2134 return ret;
2135
2136 return __file_update_time(file, ret);
2137}
2138EXPORT_SYMBOL(file_update_time);
2139
2140/**
2141 * file_modified_flags - handle mandated vfs changes when modifying a file
2142 * @file: file that was modified
2143 * @flags: kiocb flags
2144 *
2145 * When file has been modified ensure that special
2146 * file privileges are removed and time settings are updated.
2147 *
2148 * If IOCB_NOWAIT is set, special file privileges will not be removed and
2149 * time settings will not be updated. It will return -EAGAIN.
2150 *
2151 * Context: Caller must hold the file's inode lock.
2152 *
2153 * Return: 0 on success, negative errno on failure.
2154 */
2155static int file_modified_flags(struct file *file, int flags)
2156{
2157 int ret;
2158 struct inode *inode = file_inode(file);
2159
2160 /*
2161 * Clear the security bits if the process is not being run by root.
2162 * This keeps people from modifying setuid and setgid binaries.
2163 */
2164 ret = __file_remove_privs(file, flags);
2165 if (ret)
2166 return ret;
2167
2168 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2169 return 0;
2170
2171 ret = inode_needs_update_time(inode);
2172 if (ret <= 0)
2173 return ret;
2174 if (flags & IOCB_NOWAIT)
2175 return -EAGAIN;
2176
2177 return __file_update_time(file, ret);
2178}
2179
2180/**
2181 * file_modified - handle mandated vfs changes when modifying a file
2182 * @file: file that was modified
2183 *
2184 * When file has been modified ensure that special
2185 * file privileges are removed and time settings are updated.
2186 *
2187 * Context: Caller must hold the file's inode lock.
2188 *
2189 * Return: 0 on success, negative errno on failure.
2190 */
2191int file_modified(struct file *file)
2192{
2193 return file_modified_flags(file, 0);
2194}
2195EXPORT_SYMBOL(file_modified);
2196
2197/**
2198 * kiocb_modified - handle mandated vfs changes when modifying a file
2199 * @iocb: iocb that was modified
2200 *
2201 * When file has been modified ensure that special
2202 * file privileges are removed and time settings are updated.
2203 *
2204 * Context: Caller must hold the file's inode lock.
2205 *
2206 * Return: 0 on success, negative errno on failure.
2207 */
2208int kiocb_modified(struct kiocb *iocb)
2209{
2210 return file_modified_flags(iocb->ki_filp, iocb->ki_flags);
2211}
2212EXPORT_SYMBOL_GPL(kiocb_modified);
2213
2214int inode_needs_sync(struct inode *inode)
2215{
2216 if (IS_SYNC(inode))
2217 return 1;
2218 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2219 return 1;
2220 return 0;
2221}
2222EXPORT_SYMBOL(inode_needs_sync);
2223
2224/*
2225 * If we try to find an inode in the inode hash while it is being
2226 * deleted, we have to wait until the filesystem completes its
2227 * deletion before reporting that it isn't found. This function waits
2228 * until the deletion _might_ have completed. Callers are responsible
2229 * to recheck inode state.
2230 *
2231 * It doesn't matter if I_NEW is not set initially, a call to
2232 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2233 * will DTRT.
2234 */
2235static void __wait_on_freeing_inode(struct inode *inode)
2236{
2237 wait_queue_head_t *wq;
2238 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2239 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2240 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2241 spin_unlock(&inode->i_lock);
2242 spin_unlock(&inode_hash_lock);
2243 schedule();
2244 finish_wait(wq, &wait.wq_entry);
2245 spin_lock(&inode_hash_lock);
2246}
2247
2248static __initdata unsigned long ihash_entries;
2249static int __init set_ihash_entries(char *str)
2250{
2251 if (!str)
2252 return 0;
2253 ihash_entries = simple_strtoul(str, &str, 0);
2254 return 1;
2255}
2256__setup("ihash_entries=", set_ihash_entries);
2257
2258/*
2259 * Initialize the waitqueues and inode hash table.
2260 */
2261void __init inode_init_early(void)
2262{
2263 /* If hashes are distributed across NUMA nodes, defer
2264 * hash allocation until vmalloc space is available.
2265 */
2266 if (hashdist)
2267 return;
2268
2269 inode_hashtable =
2270 alloc_large_system_hash("Inode-cache",
2271 sizeof(struct hlist_head),
2272 ihash_entries,
2273 14,
2274 HASH_EARLY | HASH_ZERO,
2275 &i_hash_shift,
2276 &i_hash_mask,
2277 0,
2278 0);
2279}
2280
2281void __init inode_init(void)
2282{
2283 /* inode slab cache */
2284 inode_cachep = kmem_cache_create("inode_cache",
2285 sizeof(struct inode),
2286 0,
2287 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2288 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2289 init_once);
2290
2291 /* Hash may have been set up in inode_init_early */
2292 if (!hashdist)
2293 return;
2294
2295 inode_hashtable =
2296 alloc_large_system_hash("Inode-cache",
2297 sizeof(struct hlist_head),
2298 ihash_entries,
2299 14,
2300 HASH_ZERO,
2301 &i_hash_shift,
2302 &i_hash_mask,
2303 0,
2304 0);
2305}
2306
2307void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2308{
2309 inode->i_mode = mode;
2310 if (S_ISCHR(mode)) {
2311 inode->i_fop = &def_chr_fops;
2312 inode->i_rdev = rdev;
2313 } else if (S_ISBLK(mode)) {
2314 if (IS_ENABLED(CONFIG_BLOCK))
2315 inode->i_fop = &def_blk_fops;
2316 inode->i_rdev = rdev;
2317 } else if (S_ISFIFO(mode))
2318 inode->i_fop = &pipefifo_fops;
2319 else if (S_ISSOCK(mode))
2320 ; /* leave it no_open_fops */
2321 else
2322 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2323 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2324 inode->i_ino);
2325}
2326EXPORT_SYMBOL(init_special_inode);
2327
2328/**
2329 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2330 * @idmap: idmap of the mount the inode was created from
2331 * @inode: New inode
2332 * @dir: Directory inode
2333 * @mode: mode of the new inode
2334 *
2335 * If the inode has been created through an idmapped mount the idmap of
2336 * the vfsmount must be passed through @idmap. This function will then take
2337 * care to map the inode according to @idmap before checking permissions
2338 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2339 * checking is to be performed on the raw inode simply pass @nop_mnt_idmap.
2340 */
2341void inode_init_owner(struct mnt_idmap *idmap, struct inode *inode,
2342 const struct inode *dir, umode_t mode)
2343{
2344 inode_fsuid_set(inode, idmap);
2345 if (dir && dir->i_mode & S_ISGID) {
2346 inode->i_gid = dir->i_gid;
2347
2348 /* Directories are special, and always inherit S_ISGID */
2349 if (S_ISDIR(mode))
2350 mode |= S_ISGID;
2351 } else
2352 inode_fsgid_set(inode, idmap);
2353 inode->i_mode = mode;
2354}
2355EXPORT_SYMBOL(inode_init_owner);
2356
2357/**
2358 * inode_owner_or_capable - check current task permissions to inode
2359 * @idmap: idmap of the mount the inode was found from
2360 * @inode: inode being checked
2361 *
2362 * Return true if current either has CAP_FOWNER in a namespace with the
2363 * inode owner uid mapped, or owns the file.
2364 *
2365 * If the inode has been found through an idmapped mount the idmap of
2366 * the vfsmount must be passed through @idmap. This function will then take
2367 * care to map the inode according to @idmap before checking permissions.
2368 * On non-idmapped mounts or if permission checking is to be performed on the
2369 * raw inode simply pass @nop_mnt_idmap.
2370 */
2371bool inode_owner_or_capable(struct mnt_idmap *idmap,
2372 const struct inode *inode)
2373{
2374 vfsuid_t vfsuid;
2375 struct user_namespace *ns;
2376
2377 vfsuid = i_uid_into_vfsuid(idmap, inode);
2378 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
2379 return true;
2380
2381 ns = current_user_ns();
2382 if (vfsuid_has_mapping(ns, vfsuid) && ns_capable(ns, CAP_FOWNER))
2383 return true;
2384 return false;
2385}
2386EXPORT_SYMBOL(inode_owner_or_capable);
2387
2388/*
2389 * Direct i/o helper functions
2390 */
2391static void __inode_dio_wait(struct inode *inode)
2392{
2393 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2394 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2395
2396 do {
2397 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2398 if (atomic_read(&inode->i_dio_count))
2399 schedule();
2400 } while (atomic_read(&inode->i_dio_count));
2401 finish_wait(wq, &q.wq_entry);
2402}
2403
2404/**
2405 * inode_dio_wait - wait for outstanding DIO requests to finish
2406 * @inode: inode to wait for
2407 *
2408 * Waits for all pending direct I/O requests to finish so that we can
2409 * proceed with a truncate or equivalent operation.
2410 *
2411 * Must be called under a lock that serializes taking new references
2412 * to i_dio_count, usually by inode->i_mutex.
2413 */
2414void inode_dio_wait(struct inode *inode)
2415{
2416 if (atomic_read(&inode->i_dio_count))
2417 __inode_dio_wait(inode);
2418}
2419EXPORT_SYMBOL(inode_dio_wait);
2420
2421/*
2422 * inode_set_flags - atomically set some inode flags
2423 *
2424 * Note: the caller should be holding i_mutex, or else be sure that
2425 * they have exclusive access to the inode structure (i.e., while the
2426 * inode is being instantiated). The reason for the cmpxchg() loop
2427 * --- which wouldn't be necessary if all code paths which modify
2428 * i_flags actually followed this rule, is that there is at least one
2429 * code path which doesn't today so we use cmpxchg() out of an abundance
2430 * of caution.
2431 *
2432 * In the long run, i_mutex is overkill, and we should probably look
2433 * at using the i_lock spinlock to protect i_flags, and then make sure
2434 * it is so documented in include/linux/fs.h and that all code follows
2435 * the locking convention!!
2436 */
2437void inode_set_flags(struct inode *inode, unsigned int flags,
2438 unsigned int mask)
2439{
2440 WARN_ON_ONCE(flags & ~mask);
2441 set_mask_bits(&inode->i_flags, mask, flags);
2442}
2443EXPORT_SYMBOL(inode_set_flags);
2444
2445void inode_nohighmem(struct inode *inode)
2446{
2447 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2448}
2449EXPORT_SYMBOL(inode_nohighmem);
2450
2451/**
2452 * timestamp_truncate - Truncate timespec to a granularity
2453 * @t: Timespec
2454 * @inode: inode being updated
2455 *
2456 * Truncate a timespec to the granularity supported by the fs
2457 * containing the inode. Always rounds down. gran must
2458 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2459 */
2460struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2461{
2462 struct super_block *sb = inode->i_sb;
2463 unsigned int gran = sb->s_time_gran;
2464
2465 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2466 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2467 t.tv_nsec = 0;
2468
2469 /* Avoid division in the common cases 1 ns and 1 s. */
2470 if (gran == 1)
2471 ; /* nothing */
2472 else if (gran == NSEC_PER_SEC)
2473 t.tv_nsec = 0;
2474 else if (gran > 1 && gran < NSEC_PER_SEC)
2475 t.tv_nsec -= t.tv_nsec % gran;
2476 else
2477 WARN(1, "invalid file time granularity: %u", gran);
2478 return t;
2479}
2480EXPORT_SYMBOL(timestamp_truncate);
2481
2482/**
2483 * current_time - Return FS time
2484 * @inode: inode.
2485 *
2486 * Return the current time truncated to the time granularity supported by
2487 * the fs.
2488 *
2489 * Note that inode and inode->sb cannot be NULL.
2490 * Otherwise, the function warns and returns time without truncation.
2491 */
2492struct timespec64 current_time(struct inode *inode)
2493{
2494 struct timespec64 now;
2495
2496 ktime_get_coarse_real_ts64(&now);
2497 return timestamp_truncate(now, inode);
2498}
2499EXPORT_SYMBOL(current_time);
2500
2501/**
2502 * inode_set_ctime_current - set the ctime to current_time
2503 * @inode: inode
2504 *
2505 * Set the inode->i_ctime to the current value for the inode. Returns
2506 * the current value that was assigned to i_ctime.
2507 */
2508struct timespec64 inode_set_ctime_current(struct inode *inode)
2509{
2510 struct timespec64 now = current_time(inode);
2511
2512 inode_set_ctime(inode, now.tv_sec, now.tv_nsec);
2513 return now;
2514}
2515EXPORT_SYMBOL(inode_set_ctime_current);
2516
2517/**
2518 * in_group_or_capable - check whether caller is CAP_FSETID privileged
2519 * @idmap: idmap of the mount @inode was found from
2520 * @inode: inode to check
2521 * @vfsgid: the new/current vfsgid of @inode
2522 *
2523 * Check wether @vfsgid is in the caller's group list or if the caller is
2524 * privileged with CAP_FSETID over @inode. This can be used to determine
2525 * whether the setgid bit can be kept or must be dropped.
2526 *
2527 * Return: true if the caller is sufficiently privileged, false if not.
2528 */
2529bool in_group_or_capable(struct mnt_idmap *idmap,
2530 const struct inode *inode, vfsgid_t vfsgid)
2531{
2532 if (vfsgid_in_group_p(vfsgid))
2533 return true;
2534 if (capable_wrt_inode_uidgid(idmap, inode, CAP_FSETID))
2535 return true;
2536 return false;
2537}
2538
2539/**
2540 * mode_strip_sgid - handle the sgid bit for non-directories
2541 * @idmap: idmap of the mount the inode was created from
2542 * @dir: parent directory inode
2543 * @mode: mode of the file to be created in @dir
2544 *
2545 * If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2546 * raised and @dir has the S_ISGID bit raised ensure that the caller is
2547 * either in the group of the parent directory or they have CAP_FSETID
2548 * in their user namespace and are privileged over the parent directory.
2549 * In all other cases, strip the S_ISGID bit from @mode.
2550 *
2551 * Return: the new mode to use for the file
2552 */
2553umode_t mode_strip_sgid(struct mnt_idmap *idmap,
2554 const struct inode *dir, umode_t mode)
2555{
2556 if ((mode & (S_ISGID | S_IXGRP)) != (S_ISGID | S_IXGRP))
2557 return mode;
2558 if (S_ISDIR(mode) || !dir || !(dir->i_mode & S_ISGID))
2559 return mode;
2560 if (in_group_or_capable(idmap, dir, i_gid_into_vfsgid(idmap, dir)))
2561 return mode;
2562 return mode & ~S_ISGID;
2563}
2564EXPORT_SYMBOL(mode_strip_sgid);
1/*
2 * (C) 1997 Linus Torvalds
3 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
4 */
5#include <linux/export.h>
6#include <linux/fs.h>
7#include <linux/mm.h>
8#include <linux/backing-dev.h>
9#include <linux/hash.h>
10#include <linux/swap.h>
11#include <linux/security.h>
12#include <linux/cdev.h>
13#include <linux/bootmem.h>
14#include <linux/fsnotify.h>
15#include <linux/mount.h>
16#include <linux/posix_acl.h>
17#include <linux/prefetch.h>
18#include <linux/buffer_head.h> /* for inode_has_buffers */
19#include <linux/ratelimit.h>
20#include "internal.h"
21
22/*
23 * Inode locking rules:
24 *
25 * inode->i_lock protects:
26 * inode->i_state, inode->i_hash, __iget()
27 * inode->i_sb->s_inode_lru_lock protects:
28 * inode->i_sb->s_inode_lru, inode->i_lru
29 * inode_sb_list_lock protects:
30 * sb->s_inodes, inode->i_sb_list
31 * bdi->wb.list_lock protects:
32 * bdi->wb.b_{dirty,io,more_io}, inode->i_wb_list
33 * inode_hash_lock protects:
34 * inode_hashtable, inode->i_hash
35 *
36 * Lock ordering:
37 *
38 * inode_sb_list_lock
39 * inode->i_lock
40 * inode->i_sb->s_inode_lru_lock
41 *
42 * bdi->wb.list_lock
43 * inode->i_lock
44 *
45 * inode_hash_lock
46 * inode_sb_list_lock
47 * inode->i_lock
48 *
49 * iunique_lock
50 * inode_hash_lock
51 */
52
53static unsigned int i_hash_mask __read_mostly;
54static unsigned int i_hash_shift __read_mostly;
55static struct hlist_head *inode_hashtable __read_mostly;
56static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
57
58__cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
59
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
68/*
69 * Statistics gathering..
70 */
71struct inodes_stat_t inodes_stat;
72
73static DEFINE_PER_CPU(unsigned int, nr_inodes);
74static DEFINE_PER_CPU(unsigned int, nr_unused);
75
76static struct kmem_cache *inode_cachep __read_mostly;
77
78static int get_nr_inodes(void)
79{
80 int i;
81 int sum = 0;
82 for_each_possible_cpu(i)
83 sum += per_cpu(nr_inodes, i);
84 return sum < 0 ? 0 : sum;
85}
86
87static inline int get_nr_inodes_unused(void)
88{
89 int i;
90 int sum = 0;
91 for_each_possible_cpu(i)
92 sum += per_cpu(nr_unused, i);
93 return sum < 0 ? 0 : sum;
94}
95
96int get_nr_dirty_inodes(void)
97{
98 /* not actually dirty inodes, but a wild approximation */
99 int nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
100 return nr_dirty > 0 ? nr_dirty : 0;
101}
102
103/*
104 * Handle nr_inode sysctl
105 */
106#ifdef CONFIG_SYSCTL
107int proc_nr_inodes(ctl_table *table, int write,
108 void __user *buffer, size_t *lenp, loff_t *ppos)
109{
110 inodes_stat.nr_inodes = get_nr_inodes();
111 inodes_stat.nr_unused = get_nr_inodes_unused();
112 return proc_dointvec(table, write, buffer, lenp, ppos);
113}
114#endif
115
116/**
117 * inode_init_always - perform inode structure intialisation
118 * @sb: superblock inode belongs to
119 * @inode: inode to initialise
120 *
121 * These are initializations that need to be done on every inode
122 * allocation as the fields are not initialised by slab allocation.
123 */
124int inode_init_always(struct super_block *sb, struct inode *inode)
125{
126 static const struct inode_operations empty_iops;
127 static const struct file_operations empty_fops;
128 struct address_space *const mapping = &inode->i_data;
129
130 inode->i_sb = sb;
131 inode->i_blkbits = sb->s_blocksize_bits;
132 inode->i_flags = 0;
133 atomic_set(&inode->i_count, 1);
134 inode->i_op = &empty_iops;
135 inode->i_fop = &empty_fops;
136 inode->__i_nlink = 1;
137 inode->i_opflags = 0;
138 i_uid_write(inode, 0);
139 i_gid_write(inode, 0);
140 atomic_set(&inode->i_writecount, 0);
141 inode->i_size = 0;
142 inode->i_blocks = 0;
143 inode->i_bytes = 0;
144 inode->i_generation = 0;
145#ifdef CONFIG_QUOTA
146 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
147#endif
148 inode->i_pipe = NULL;
149 inode->i_bdev = NULL;
150 inode->i_cdev = NULL;
151 inode->i_rdev = 0;
152 inode->dirtied_when = 0;
153
154 if (security_inode_alloc(inode))
155 goto out;
156 spin_lock_init(&inode->i_lock);
157 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
158
159 mutex_init(&inode->i_mutex);
160 lockdep_set_class(&inode->i_mutex, &sb->s_type->i_mutex_key);
161
162 atomic_set(&inode->i_dio_count, 0);
163
164 mapping->a_ops = &empty_aops;
165 mapping->host = inode;
166 mapping->flags = 0;
167 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
168 mapping->assoc_mapping = NULL;
169 mapping->backing_dev_info = &default_backing_dev_info;
170 mapping->writeback_index = 0;
171
172 /*
173 * If the block_device provides a backing_dev_info for client
174 * inodes then use that. Otherwise the inode share the bdev's
175 * backing_dev_info.
176 */
177 if (sb->s_bdev) {
178 struct backing_dev_info *bdi;
179
180 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
181 mapping->backing_dev_info = bdi;
182 }
183 inode->i_private = NULL;
184 inode->i_mapping = mapping;
185 INIT_LIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
186#ifdef CONFIG_FS_POSIX_ACL
187 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
188#endif
189
190#ifdef CONFIG_FSNOTIFY
191 inode->i_fsnotify_mask = 0;
192#endif
193
194 this_cpu_inc(nr_inodes);
195
196 return 0;
197out:
198 return -ENOMEM;
199}
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
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;
223}
224
225void free_inode_nonrcu(struct inode *inode)
226{
227 kmem_cache_free(inode_cachep, inode);
228}
229EXPORT_SYMBOL(free_inode_nonrcu);
230
231void __destroy_inode(struct inode *inode)
232{
233 BUG_ON(inode_has_buffers(inode));
234 security_inode_free(inode);
235 fsnotify_inode_delete(inode);
236 if (!inode->i_nlink) {
237 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
238 atomic_long_dec(&inode->i_sb->s_remove_count);
239 }
240
241#ifdef CONFIG_FS_POSIX_ACL
242 if (inode->i_acl && inode->i_acl != ACL_NOT_CACHED)
243 posix_acl_release(inode->i_acl);
244 if (inode->i_default_acl && inode->i_default_acl != ACL_NOT_CACHED)
245 posix_acl_release(inode->i_default_acl);
246#endif
247 this_cpu_dec(nr_inodes);
248}
249EXPORT_SYMBOL(__destroy_inode);
250
251static void i_callback(struct rcu_head *head)
252{
253 struct inode *inode = container_of(head, struct inode, i_rcu);
254 kmem_cache_free(inode_cachep, inode);
255}
256
257static void destroy_inode(struct inode *inode)
258{
259 BUG_ON(!list_empty(&inode->i_lru));
260 __destroy_inode(inode);
261 if (inode->i_sb->s_op->destroy_inode)
262 inode->i_sb->s_op->destroy_inode(inode);
263 else
264 call_rcu(&inode->i_rcu, i_callback);
265}
266
267/**
268 * drop_nlink - directly drop an inode's link count
269 * @inode: inode
270 *
271 * This is a low-level filesystem helper to replace any
272 * direct filesystem manipulation of i_nlink. In cases
273 * where we are attempting to track writes to the
274 * filesystem, a decrement to zero means an imminent
275 * write when the file is truncated and actually unlinked
276 * on the filesystem.
277 */
278void drop_nlink(struct inode *inode)
279{
280 WARN_ON(inode->i_nlink == 0);
281 inode->__i_nlink--;
282 if (!inode->i_nlink)
283 atomic_long_inc(&inode->i_sb->s_remove_count);
284}
285EXPORT_SYMBOL(drop_nlink);
286
287/**
288 * clear_nlink - directly zero an inode's link count
289 * @inode: inode
290 *
291 * This is a low-level filesystem helper to replace any
292 * direct filesystem manipulation of i_nlink. See
293 * drop_nlink() for why we care about i_nlink hitting zero.
294 */
295void clear_nlink(struct inode *inode)
296{
297 if (inode->i_nlink) {
298 inode->__i_nlink = 0;
299 atomic_long_inc(&inode->i_sb->s_remove_count);
300 }
301}
302EXPORT_SYMBOL(clear_nlink);
303
304/**
305 * set_nlink - directly set an inode's link count
306 * @inode: inode
307 * @nlink: new nlink (should be non-zero)
308 *
309 * This is a low-level filesystem helper to replace any
310 * direct filesystem manipulation of i_nlink.
311 */
312void set_nlink(struct inode *inode, unsigned int nlink)
313{
314 if (!nlink) {
315 clear_nlink(inode);
316 } else {
317 /* Yes, some filesystems do change nlink from zero to one */
318 if (inode->i_nlink == 0)
319 atomic_long_dec(&inode->i_sb->s_remove_count);
320
321 inode->__i_nlink = nlink;
322 }
323}
324EXPORT_SYMBOL(set_nlink);
325
326/**
327 * inc_nlink - directly increment an inode's link count
328 * @inode: inode
329 *
330 * This is a low-level filesystem helper to replace any
331 * direct filesystem manipulation of i_nlink. Currently,
332 * it is only here for parity with dec_nlink().
333 */
334void inc_nlink(struct inode *inode)
335{
336 if (WARN_ON(inode->i_nlink == 0))
337 atomic_long_dec(&inode->i_sb->s_remove_count);
338
339 inode->__i_nlink++;
340}
341EXPORT_SYMBOL(inc_nlink);
342
343void address_space_init_once(struct address_space *mapping)
344{
345 memset(mapping, 0, sizeof(*mapping));
346 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
347 spin_lock_init(&mapping->tree_lock);
348 mutex_init(&mapping->i_mmap_mutex);
349 INIT_LIST_HEAD(&mapping->private_list);
350 spin_lock_init(&mapping->private_lock);
351 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
352 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
353}
354EXPORT_SYMBOL(address_space_init_once);
355
356/*
357 * These are initializations that only need to be done
358 * once, because the fields are idempotent across use
359 * of the inode, so let the slab aware of that.
360 */
361void inode_init_once(struct inode *inode)
362{
363 memset(inode, 0, sizeof(*inode));
364 INIT_HLIST_NODE(&inode->i_hash);
365 INIT_LIST_HEAD(&inode->i_devices);
366 INIT_LIST_HEAD(&inode->i_wb_list);
367 INIT_LIST_HEAD(&inode->i_lru);
368 address_space_init_once(&inode->i_data);
369 i_size_ordered_init(inode);
370#ifdef CONFIG_FSNOTIFY
371 INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
372#endif
373}
374EXPORT_SYMBOL(inode_init_once);
375
376static void init_once(void *foo)
377{
378 struct inode *inode = (struct inode *) foo;
379
380 inode_init_once(inode);
381}
382
383/*
384 * inode->i_lock must be held
385 */
386void __iget(struct inode *inode)
387{
388 atomic_inc(&inode->i_count);
389}
390
391/*
392 * get additional reference to inode; caller must already hold one.
393 */
394void ihold(struct inode *inode)
395{
396 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
397}
398EXPORT_SYMBOL(ihold);
399
400static void inode_lru_list_add(struct inode *inode)
401{
402 spin_lock(&inode->i_sb->s_inode_lru_lock);
403 if (list_empty(&inode->i_lru)) {
404 list_add(&inode->i_lru, &inode->i_sb->s_inode_lru);
405 inode->i_sb->s_nr_inodes_unused++;
406 this_cpu_inc(nr_unused);
407 }
408 spin_unlock(&inode->i_sb->s_inode_lru_lock);
409}
410
411static void inode_lru_list_del(struct inode *inode)
412{
413 spin_lock(&inode->i_sb->s_inode_lru_lock);
414 if (!list_empty(&inode->i_lru)) {
415 list_del_init(&inode->i_lru);
416 inode->i_sb->s_nr_inodes_unused--;
417 this_cpu_dec(nr_unused);
418 }
419 spin_unlock(&inode->i_sb->s_inode_lru_lock);
420}
421
422/**
423 * inode_sb_list_add - add inode to the superblock list of inodes
424 * @inode: inode to add
425 */
426void inode_sb_list_add(struct inode *inode)
427{
428 spin_lock(&inode_sb_list_lock);
429 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
430 spin_unlock(&inode_sb_list_lock);
431}
432EXPORT_SYMBOL_GPL(inode_sb_list_add);
433
434static inline void inode_sb_list_del(struct inode *inode)
435{
436 if (!list_empty(&inode->i_sb_list)) {
437 spin_lock(&inode_sb_list_lock);
438 list_del_init(&inode->i_sb_list);
439 spin_unlock(&inode_sb_list_lock);
440 }
441}
442
443static unsigned long hash(struct super_block *sb, unsigned long hashval)
444{
445 unsigned long tmp;
446
447 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
448 L1_CACHE_BYTES;
449 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
450 return tmp & i_hash_mask;
451}
452
453/**
454 * __insert_inode_hash - hash an inode
455 * @inode: unhashed inode
456 * @hashval: unsigned long value used to locate this object in the
457 * inode_hashtable.
458 *
459 * Add an inode to the inode hash for this superblock.
460 */
461void __insert_inode_hash(struct inode *inode, unsigned long hashval)
462{
463 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
464
465 spin_lock(&inode_hash_lock);
466 spin_lock(&inode->i_lock);
467 hlist_add_head(&inode->i_hash, b);
468 spin_unlock(&inode->i_lock);
469 spin_unlock(&inode_hash_lock);
470}
471EXPORT_SYMBOL(__insert_inode_hash);
472
473/**
474 * __remove_inode_hash - remove an inode from the hash
475 * @inode: inode to unhash
476 *
477 * Remove an inode from the superblock.
478 */
479void __remove_inode_hash(struct inode *inode)
480{
481 spin_lock(&inode_hash_lock);
482 spin_lock(&inode->i_lock);
483 hlist_del_init(&inode->i_hash);
484 spin_unlock(&inode->i_lock);
485 spin_unlock(&inode_hash_lock);
486}
487EXPORT_SYMBOL(__remove_inode_hash);
488
489void clear_inode(struct inode *inode)
490{
491 might_sleep();
492 /*
493 * We have to cycle tree_lock here because reclaim can be still in the
494 * process of removing the last page (in __delete_from_page_cache())
495 * and we must not free mapping under it.
496 */
497 spin_lock_irq(&inode->i_data.tree_lock);
498 BUG_ON(inode->i_data.nrpages);
499 spin_unlock_irq(&inode->i_data.tree_lock);
500 BUG_ON(!list_empty(&inode->i_data.private_list));
501 BUG_ON(!(inode->i_state & I_FREEING));
502 BUG_ON(inode->i_state & I_CLEAR);
503 /* don't need i_lock here, no concurrent mods to i_state */
504 inode->i_state = I_FREEING | I_CLEAR;
505}
506EXPORT_SYMBOL(clear_inode);
507
508/*
509 * Free the inode passed in, removing it from the lists it is still connected
510 * to. We remove any pages still attached to the inode and wait for any IO that
511 * is still in progress before finally destroying the inode.
512 *
513 * An inode must already be marked I_FREEING so that we avoid the inode being
514 * moved back onto lists if we race with other code that manipulates the lists
515 * (e.g. writeback_single_inode). The caller is responsible for setting this.
516 *
517 * An inode must already be removed from the LRU list before being evicted from
518 * the cache. This should occur atomically with setting the I_FREEING state
519 * flag, so no inodes here should ever be on the LRU when being evicted.
520 */
521static void evict(struct inode *inode)
522{
523 const struct super_operations *op = inode->i_sb->s_op;
524
525 BUG_ON(!(inode->i_state & I_FREEING));
526 BUG_ON(!list_empty(&inode->i_lru));
527
528 if (!list_empty(&inode->i_wb_list))
529 inode_wb_list_del(inode);
530
531 inode_sb_list_del(inode);
532
533 /*
534 * Wait for flusher thread to be done with the inode so that filesystem
535 * does not start destroying it while writeback is still running. Since
536 * the inode has I_FREEING set, flusher thread won't start new work on
537 * the inode. We just have to wait for running writeback to finish.
538 */
539 inode_wait_for_writeback(inode);
540
541 if (op->evict_inode) {
542 op->evict_inode(inode);
543 } else {
544 if (inode->i_data.nrpages)
545 truncate_inode_pages(&inode->i_data, 0);
546 clear_inode(inode);
547 }
548 if (S_ISBLK(inode->i_mode) && inode->i_bdev)
549 bd_forget(inode);
550 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
551 cd_forget(inode);
552
553 remove_inode_hash(inode);
554
555 spin_lock(&inode->i_lock);
556 wake_up_bit(&inode->i_state, __I_NEW);
557 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
558 spin_unlock(&inode->i_lock);
559
560 destroy_inode(inode);
561}
562
563/*
564 * dispose_list - dispose of the contents of a local list
565 * @head: the head of the list to free
566 *
567 * Dispose-list gets a local list with local inodes in it, so it doesn't
568 * need to worry about list corruption and SMP locks.
569 */
570static void dispose_list(struct list_head *head)
571{
572 while (!list_empty(head)) {
573 struct inode *inode;
574
575 inode = list_first_entry(head, struct inode, i_lru);
576 list_del_init(&inode->i_lru);
577
578 evict(inode);
579 }
580}
581
582/**
583 * evict_inodes - evict all evictable inodes for a superblock
584 * @sb: superblock to operate on
585 *
586 * Make sure that no inodes with zero refcount are retained. This is
587 * called by superblock shutdown after having MS_ACTIVE flag removed,
588 * so any inode reaching zero refcount during or after that call will
589 * be immediately evicted.
590 */
591void evict_inodes(struct super_block *sb)
592{
593 struct inode *inode, *next;
594 LIST_HEAD(dispose);
595
596 spin_lock(&inode_sb_list_lock);
597 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
598 if (atomic_read(&inode->i_count))
599 continue;
600
601 spin_lock(&inode->i_lock);
602 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
603 spin_unlock(&inode->i_lock);
604 continue;
605 }
606
607 inode->i_state |= I_FREEING;
608 inode_lru_list_del(inode);
609 spin_unlock(&inode->i_lock);
610 list_add(&inode->i_lru, &dispose);
611 }
612 spin_unlock(&inode_sb_list_lock);
613
614 dispose_list(&dispose);
615}
616
617/**
618 * invalidate_inodes - attempt to free all inodes on a superblock
619 * @sb: superblock to operate on
620 * @kill_dirty: flag to guide handling of dirty inodes
621 *
622 * Attempts to free all inodes for a given superblock. If there were any
623 * busy inodes return a non-zero value, else zero.
624 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
625 * them as busy.
626 */
627int invalidate_inodes(struct super_block *sb, bool kill_dirty)
628{
629 int busy = 0;
630 struct inode *inode, *next;
631 LIST_HEAD(dispose);
632
633 spin_lock(&inode_sb_list_lock);
634 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
635 spin_lock(&inode->i_lock);
636 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
637 spin_unlock(&inode->i_lock);
638 continue;
639 }
640 if (inode->i_state & I_DIRTY && !kill_dirty) {
641 spin_unlock(&inode->i_lock);
642 busy = 1;
643 continue;
644 }
645 if (atomic_read(&inode->i_count)) {
646 spin_unlock(&inode->i_lock);
647 busy = 1;
648 continue;
649 }
650
651 inode->i_state |= I_FREEING;
652 inode_lru_list_del(inode);
653 spin_unlock(&inode->i_lock);
654 list_add(&inode->i_lru, &dispose);
655 }
656 spin_unlock(&inode_sb_list_lock);
657
658 dispose_list(&dispose);
659
660 return busy;
661}
662
663static int can_unuse(struct inode *inode)
664{
665 if (inode->i_state & ~I_REFERENCED)
666 return 0;
667 if (inode_has_buffers(inode))
668 return 0;
669 if (atomic_read(&inode->i_count))
670 return 0;
671 if (inode->i_data.nrpages)
672 return 0;
673 return 1;
674}
675
676/*
677 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
678 * This is called from the superblock shrinker function with a number of inodes
679 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
680 * then are freed outside inode_lock by dispose_list().
681 *
682 * Any inodes which are pinned purely because of attached pagecache have their
683 * pagecache removed. If the inode has metadata buffers attached to
684 * mapping->private_list then try to remove them.
685 *
686 * If the inode has the I_REFERENCED flag set, then it means that it has been
687 * used recently - the flag is set in iput_final(). When we encounter such an
688 * inode, clear the flag and move it to the back of the LRU so it gets another
689 * pass through the LRU before it gets reclaimed. This is necessary because of
690 * the fact we are doing lazy LRU updates to minimise lock contention so the
691 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
692 * with this flag set because they are the inodes that are out of order.
693 */
694void prune_icache_sb(struct super_block *sb, int nr_to_scan)
695{
696 LIST_HEAD(freeable);
697 int nr_scanned;
698 unsigned long reap = 0;
699
700 spin_lock(&sb->s_inode_lru_lock);
701 for (nr_scanned = nr_to_scan; nr_scanned >= 0; nr_scanned--) {
702 struct inode *inode;
703
704 if (list_empty(&sb->s_inode_lru))
705 break;
706
707 inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
708
709 /*
710 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
711 * so use a trylock. If we fail to get the lock, just move the
712 * inode to the back of the list so we don't spin on it.
713 */
714 if (!spin_trylock(&inode->i_lock)) {
715 list_move_tail(&inode->i_lru, &sb->s_inode_lru);
716 continue;
717 }
718
719 /*
720 * Referenced or dirty inodes are still in use. Give them
721 * another pass through the LRU as we canot reclaim them now.
722 */
723 if (atomic_read(&inode->i_count) ||
724 (inode->i_state & ~I_REFERENCED)) {
725 list_del_init(&inode->i_lru);
726 spin_unlock(&inode->i_lock);
727 sb->s_nr_inodes_unused--;
728 this_cpu_dec(nr_unused);
729 continue;
730 }
731
732 /* recently referenced inodes get one more pass */
733 if (inode->i_state & I_REFERENCED) {
734 inode->i_state &= ~I_REFERENCED;
735 list_move(&inode->i_lru, &sb->s_inode_lru);
736 spin_unlock(&inode->i_lock);
737 continue;
738 }
739 if (inode_has_buffers(inode) || inode->i_data.nrpages) {
740 __iget(inode);
741 spin_unlock(&inode->i_lock);
742 spin_unlock(&sb->s_inode_lru_lock);
743 if (remove_inode_buffers(inode))
744 reap += invalidate_mapping_pages(&inode->i_data,
745 0, -1);
746 iput(inode);
747 spin_lock(&sb->s_inode_lru_lock);
748
749 if (inode != list_entry(sb->s_inode_lru.next,
750 struct inode, i_lru))
751 continue; /* wrong inode or list_empty */
752 /* avoid lock inversions with trylock */
753 if (!spin_trylock(&inode->i_lock))
754 continue;
755 if (!can_unuse(inode)) {
756 spin_unlock(&inode->i_lock);
757 continue;
758 }
759 }
760 WARN_ON(inode->i_state & I_NEW);
761 inode->i_state |= I_FREEING;
762 spin_unlock(&inode->i_lock);
763
764 list_move(&inode->i_lru, &freeable);
765 sb->s_nr_inodes_unused--;
766 this_cpu_dec(nr_unused);
767 }
768 if (current_is_kswapd())
769 __count_vm_events(KSWAPD_INODESTEAL, reap);
770 else
771 __count_vm_events(PGINODESTEAL, reap);
772 spin_unlock(&sb->s_inode_lru_lock);
773 if (current->reclaim_state)
774 current->reclaim_state->reclaimed_slab += reap;
775
776 dispose_list(&freeable);
777}
778
779static void __wait_on_freeing_inode(struct inode *inode);
780/*
781 * Called with the inode lock held.
782 */
783static struct inode *find_inode(struct super_block *sb,
784 struct hlist_head *head,
785 int (*test)(struct inode *, void *),
786 void *data)
787{
788 struct hlist_node *node;
789 struct inode *inode = NULL;
790
791repeat:
792 hlist_for_each_entry(inode, node, head, i_hash) {
793 spin_lock(&inode->i_lock);
794 if (inode->i_sb != sb) {
795 spin_unlock(&inode->i_lock);
796 continue;
797 }
798 if (!test(inode, data)) {
799 spin_unlock(&inode->i_lock);
800 continue;
801 }
802 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
803 __wait_on_freeing_inode(inode);
804 goto repeat;
805 }
806 __iget(inode);
807 spin_unlock(&inode->i_lock);
808 return inode;
809 }
810 return NULL;
811}
812
813/*
814 * find_inode_fast is the fast path version of find_inode, see the comment at
815 * iget_locked for details.
816 */
817static struct inode *find_inode_fast(struct super_block *sb,
818 struct hlist_head *head, unsigned long ino)
819{
820 struct hlist_node *node;
821 struct inode *inode = NULL;
822
823repeat:
824 hlist_for_each_entry(inode, node, head, i_hash) {
825 spin_lock(&inode->i_lock);
826 if (inode->i_ino != ino) {
827 spin_unlock(&inode->i_lock);
828 continue;
829 }
830 if (inode->i_sb != sb) {
831 spin_unlock(&inode->i_lock);
832 continue;
833 }
834 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
835 __wait_on_freeing_inode(inode);
836 goto repeat;
837 }
838 __iget(inode);
839 spin_unlock(&inode->i_lock);
840 return inode;
841 }
842 return NULL;
843}
844
845/*
846 * Each cpu owns a range of LAST_INO_BATCH numbers.
847 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
848 * to renew the exhausted range.
849 *
850 * This does not significantly increase overflow rate because every CPU can
851 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
852 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
853 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
854 * overflow rate by 2x, which does not seem too significant.
855 *
856 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
857 * error if st_ino won't fit in target struct field. Use 32bit counter
858 * here to attempt to avoid that.
859 */
860#define LAST_INO_BATCH 1024
861static DEFINE_PER_CPU(unsigned int, last_ino);
862
863unsigned int get_next_ino(void)
864{
865 unsigned int *p = &get_cpu_var(last_ino);
866 unsigned int res = *p;
867
868#ifdef CONFIG_SMP
869 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
870 static atomic_t shared_last_ino;
871 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
872
873 res = next - LAST_INO_BATCH;
874 }
875#endif
876
877 *p = ++res;
878 put_cpu_var(last_ino);
879 return res;
880}
881EXPORT_SYMBOL(get_next_ino);
882
883/**
884 * new_inode_pseudo - obtain an inode
885 * @sb: superblock
886 *
887 * Allocates a new inode for given superblock.
888 * Inode wont be chained in superblock s_inodes list
889 * This means :
890 * - fs can't be unmount
891 * - quotas, fsnotify, writeback can't work
892 */
893struct inode *new_inode_pseudo(struct super_block *sb)
894{
895 struct inode *inode = alloc_inode(sb);
896
897 if (inode) {
898 spin_lock(&inode->i_lock);
899 inode->i_state = 0;
900 spin_unlock(&inode->i_lock);
901 INIT_LIST_HEAD(&inode->i_sb_list);
902 }
903 return inode;
904}
905
906/**
907 * new_inode - obtain an inode
908 * @sb: superblock
909 *
910 * Allocates a new inode for given superblock. The default gfp_mask
911 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
912 * If HIGHMEM pages are unsuitable or it is known that pages allocated
913 * for the page cache are not reclaimable or migratable,
914 * mapping_set_gfp_mask() must be called with suitable flags on the
915 * newly created inode's mapping
916 *
917 */
918struct inode *new_inode(struct super_block *sb)
919{
920 struct inode *inode;
921
922 spin_lock_prefetch(&inode_sb_list_lock);
923
924 inode = new_inode_pseudo(sb);
925 if (inode)
926 inode_sb_list_add(inode);
927 return inode;
928}
929EXPORT_SYMBOL(new_inode);
930
931#ifdef CONFIG_DEBUG_LOCK_ALLOC
932void lockdep_annotate_inode_mutex_key(struct inode *inode)
933{
934 if (S_ISDIR(inode->i_mode)) {
935 struct file_system_type *type = inode->i_sb->s_type;
936
937 /* Set new key only if filesystem hasn't already changed it */
938 if (lockdep_match_class(&inode->i_mutex, &type->i_mutex_key)) {
939 /*
940 * ensure nobody is actually holding i_mutex
941 */
942 mutex_destroy(&inode->i_mutex);
943 mutex_init(&inode->i_mutex);
944 lockdep_set_class(&inode->i_mutex,
945 &type->i_mutex_dir_key);
946 }
947 }
948}
949EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
950#endif
951
952/**
953 * unlock_new_inode - clear the I_NEW state and wake up any waiters
954 * @inode: new inode to unlock
955 *
956 * Called when the inode is fully initialised to clear the new state of the
957 * inode and wake up anyone waiting for the inode to finish initialisation.
958 */
959void unlock_new_inode(struct inode *inode)
960{
961 lockdep_annotate_inode_mutex_key(inode);
962 spin_lock(&inode->i_lock);
963 WARN_ON(!(inode->i_state & I_NEW));
964 inode->i_state &= ~I_NEW;
965 smp_mb();
966 wake_up_bit(&inode->i_state, __I_NEW);
967 spin_unlock(&inode->i_lock);
968}
969EXPORT_SYMBOL(unlock_new_inode);
970
971/**
972 * iget5_locked - obtain an inode from a mounted file system
973 * @sb: super block of file system
974 * @hashval: hash value (usually inode number) to get
975 * @test: callback used for comparisons between inodes
976 * @set: callback used to initialize a new struct inode
977 * @data: opaque data pointer to pass to @test and @set
978 *
979 * Search for the inode specified by @hashval and @data in the inode cache,
980 * and if present it is return it with an increased reference count. This is
981 * a generalized version of iget_locked() for file systems where the inode
982 * number is not sufficient for unique identification of an inode.
983 *
984 * If the inode is not in cache, allocate a new inode and return it locked,
985 * hashed, and with the I_NEW flag set. The file system gets to fill it in
986 * before unlocking it via unlock_new_inode().
987 *
988 * Note both @test and @set are called with the inode_hash_lock held, so can't
989 * sleep.
990 */
991struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
992 int (*test)(struct inode *, void *),
993 int (*set)(struct inode *, void *), void *data)
994{
995 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
996 struct inode *inode;
997
998 spin_lock(&inode_hash_lock);
999 inode = find_inode(sb, head, test, data);
1000 spin_unlock(&inode_hash_lock);
1001
1002 if (inode) {
1003 wait_on_inode(inode);
1004 return inode;
1005 }
1006
1007 inode = alloc_inode(sb);
1008 if (inode) {
1009 struct inode *old;
1010
1011 spin_lock(&inode_hash_lock);
1012 /* We released the lock, so.. */
1013 old = find_inode(sb, head, test, data);
1014 if (!old) {
1015 if (set(inode, data))
1016 goto set_failed;
1017
1018 spin_lock(&inode->i_lock);
1019 inode->i_state = I_NEW;
1020 hlist_add_head(&inode->i_hash, head);
1021 spin_unlock(&inode->i_lock);
1022 inode_sb_list_add(inode);
1023 spin_unlock(&inode_hash_lock);
1024
1025 /* Return the locked inode with I_NEW set, the
1026 * caller is responsible for filling in the contents
1027 */
1028 return inode;
1029 }
1030
1031 /*
1032 * Uhhuh, somebody else created the same inode under
1033 * us. Use the old inode instead of the one we just
1034 * allocated.
1035 */
1036 spin_unlock(&inode_hash_lock);
1037 destroy_inode(inode);
1038 inode = old;
1039 wait_on_inode(inode);
1040 }
1041 return inode;
1042
1043set_failed:
1044 spin_unlock(&inode_hash_lock);
1045 destroy_inode(inode);
1046 return NULL;
1047}
1048EXPORT_SYMBOL(iget5_locked);
1049
1050/**
1051 * iget_locked - obtain an inode from a mounted file system
1052 * @sb: super block of file system
1053 * @ino: inode number to get
1054 *
1055 * Search for the inode specified by @ino in the inode cache and if present
1056 * return it with an increased reference count. This is for file systems
1057 * where the inode number is sufficient for unique identification of an inode.
1058 *
1059 * If the inode is not in cache, allocate a new inode and return it locked,
1060 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1061 * before unlocking it via unlock_new_inode().
1062 */
1063struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1064{
1065 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1066 struct inode *inode;
1067
1068 spin_lock(&inode_hash_lock);
1069 inode = find_inode_fast(sb, head, ino);
1070 spin_unlock(&inode_hash_lock);
1071 if (inode) {
1072 wait_on_inode(inode);
1073 return inode;
1074 }
1075
1076 inode = alloc_inode(sb);
1077 if (inode) {
1078 struct inode *old;
1079
1080 spin_lock(&inode_hash_lock);
1081 /* We released the lock, so.. */
1082 old = find_inode_fast(sb, head, ino);
1083 if (!old) {
1084 inode->i_ino = ino;
1085 spin_lock(&inode->i_lock);
1086 inode->i_state = I_NEW;
1087 hlist_add_head(&inode->i_hash, head);
1088 spin_unlock(&inode->i_lock);
1089 inode_sb_list_add(inode);
1090 spin_unlock(&inode_hash_lock);
1091
1092 /* Return the locked inode with I_NEW set, the
1093 * caller is responsible for filling in the contents
1094 */
1095 return inode;
1096 }
1097
1098 /*
1099 * Uhhuh, somebody else created the same inode under
1100 * us. Use the old inode instead of the one we just
1101 * allocated.
1102 */
1103 spin_unlock(&inode_hash_lock);
1104 destroy_inode(inode);
1105 inode = old;
1106 wait_on_inode(inode);
1107 }
1108 return inode;
1109}
1110EXPORT_SYMBOL(iget_locked);
1111
1112/*
1113 * search the inode cache for a matching inode number.
1114 * If we find one, then the inode number we are trying to
1115 * allocate is not unique and so we should not use it.
1116 *
1117 * Returns 1 if the inode number is unique, 0 if it is not.
1118 */
1119static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1120{
1121 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1122 struct hlist_node *node;
1123 struct inode *inode;
1124
1125 spin_lock(&inode_hash_lock);
1126 hlist_for_each_entry(inode, node, b, i_hash) {
1127 if (inode->i_ino == ino && inode->i_sb == sb) {
1128 spin_unlock(&inode_hash_lock);
1129 return 0;
1130 }
1131 }
1132 spin_unlock(&inode_hash_lock);
1133
1134 return 1;
1135}
1136
1137/**
1138 * iunique - get a unique inode number
1139 * @sb: superblock
1140 * @max_reserved: highest reserved inode number
1141 *
1142 * Obtain an inode number that is unique on the system for a given
1143 * superblock. This is used by file systems that have no natural
1144 * permanent inode numbering system. An inode number is returned that
1145 * is higher than the reserved limit but unique.
1146 *
1147 * BUGS:
1148 * With a large number of inodes live on the file system this function
1149 * currently becomes quite slow.
1150 */
1151ino_t iunique(struct super_block *sb, ino_t max_reserved)
1152{
1153 /*
1154 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1155 * error if st_ino won't fit in target struct field. Use 32bit counter
1156 * here to attempt to avoid that.
1157 */
1158 static DEFINE_SPINLOCK(iunique_lock);
1159 static unsigned int counter;
1160 ino_t res;
1161
1162 spin_lock(&iunique_lock);
1163 do {
1164 if (counter <= max_reserved)
1165 counter = max_reserved + 1;
1166 res = counter++;
1167 } while (!test_inode_iunique(sb, res));
1168 spin_unlock(&iunique_lock);
1169
1170 return res;
1171}
1172EXPORT_SYMBOL(iunique);
1173
1174struct inode *igrab(struct inode *inode)
1175{
1176 spin_lock(&inode->i_lock);
1177 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1178 __iget(inode);
1179 spin_unlock(&inode->i_lock);
1180 } else {
1181 spin_unlock(&inode->i_lock);
1182 /*
1183 * Handle the case where s_op->clear_inode is not been
1184 * called yet, and somebody is calling igrab
1185 * while the inode is getting freed.
1186 */
1187 inode = NULL;
1188 }
1189 return inode;
1190}
1191EXPORT_SYMBOL(igrab);
1192
1193/**
1194 * ilookup5_nowait - search for an inode in the inode cache
1195 * @sb: super block of file system to search
1196 * @hashval: hash value (usually inode number) to search for
1197 * @test: callback used for comparisons between inodes
1198 * @data: opaque data pointer to pass to @test
1199 *
1200 * Search for the inode specified by @hashval and @data in the inode cache.
1201 * If the inode is in the cache, the inode is returned with an incremented
1202 * reference count.
1203 *
1204 * Note: I_NEW is not waited upon so you have to be very careful what you do
1205 * with the returned inode. You probably should be using ilookup5() instead.
1206 *
1207 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1208 */
1209struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1210 int (*test)(struct inode *, void *), void *data)
1211{
1212 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1213 struct inode *inode;
1214
1215 spin_lock(&inode_hash_lock);
1216 inode = find_inode(sb, head, test, data);
1217 spin_unlock(&inode_hash_lock);
1218
1219 return inode;
1220}
1221EXPORT_SYMBOL(ilookup5_nowait);
1222
1223/**
1224 * ilookup5 - search for an inode in the inode cache
1225 * @sb: super block of file system to search
1226 * @hashval: hash value (usually inode number) to search for
1227 * @test: callback used for comparisons between inodes
1228 * @data: opaque data pointer to pass to @test
1229 *
1230 * Search for the inode specified by @hashval and @data in the inode cache,
1231 * and if the inode is in the cache, return the inode with an incremented
1232 * reference count. Waits on I_NEW before returning the inode.
1233 * returned with an incremented reference count.
1234 *
1235 * This is a generalized version of ilookup() for file systems where the
1236 * inode number is not sufficient for unique identification of an inode.
1237 *
1238 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1239 */
1240struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1241 int (*test)(struct inode *, void *), void *data)
1242{
1243 struct inode *inode = ilookup5_nowait(sb, hashval, test, data);
1244
1245 if (inode)
1246 wait_on_inode(inode);
1247 return inode;
1248}
1249EXPORT_SYMBOL(ilookup5);
1250
1251/**
1252 * ilookup - search for an inode in the inode cache
1253 * @sb: super block of file system to search
1254 * @ino: inode number to search for
1255 *
1256 * Search for the inode @ino in the inode cache, and if the inode is in the
1257 * cache, the inode is returned with an incremented reference count.
1258 */
1259struct inode *ilookup(struct super_block *sb, unsigned long ino)
1260{
1261 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1262 struct inode *inode;
1263
1264 spin_lock(&inode_hash_lock);
1265 inode = find_inode_fast(sb, head, ino);
1266 spin_unlock(&inode_hash_lock);
1267
1268 if (inode)
1269 wait_on_inode(inode);
1270 return inode;
1271}
1272EXPORT_SYMBOL(ilookup);
1273
1274int insert_inode_locked(struct inode *inode)
1275{
1276 struct super_block *sb = inode->i_sb;
1277 ino_t ino = inode->i_ino;
1278 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1279
1280 while (1) {
1281 struct hlist_node *node;
1282 struct inode *old = NULL;
1283 spin_lock(&inode_hash_lock);
1284 hlist_for_each_entry(old, node, head, i_hash) {
1285 if (old->i_ino != ino)
1286 continue;
1287 if (old->i_sb != sb)
1288 continue;
1289 spin_lock(&old->i_lock);
1290 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1291 spin_unlock(&old->i_lock);
1292 continue;
1293 }
1294 break;
1295 }
1296 if (likely(!node)) {
1297 spin_lock(&inode->i_lock);
1298 inode->i_state |= I_NEW;
1299 hlist_add_head(&inode->i_hash, head);
1300 spin_unlock(&inode->i_lock);
1301 spin_unlock(&inode_hash_lock);
1302 return 0;
1303 }
1304 __iget(old);
1305 spin_unlock(&old->i_lock);
1306 spin_unlock(&inode_hash_lock);
1307 wait_on_inode(old);
1308 if (unlikely(!inode_unhashed(old))) {
1309 iput(old);
1310 return -EBUSY;
1311 }
1312 iput(old);
1313 }
1314}
1315EXPORT_SYMBOL(insert_inode_locked);
1316
1317int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1318 int (*test)(struct inode *, void *), void *data)
1319{
1320 struct super_block *sb = inode->i_sb;
1321 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1322
1323 while (1) {
1324 struct hlist_node *node;
1325 struct inode *old = NULL;
1326
1327 spin_lock(&inode_hash_lock);
1328 hlist_for_each_entry(old, node, head, i_hash) {
1329 if (old->i_sb != sb)
1330 continue;
1331 if (!test(old, data))
1332 continue;
1333 spin_lock(&old->i_lock);
1334 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1335 spin_unlock(&old->i_lock);
1336 continue;
1337 }
1338 break;
1339 }
1340 if (likely(!node)) {
1341 spin_lock(&inode->i_lock);
1342 inode->i_state |= I_NEW;
1343 hlist_add_head(&inode->i_hash, head);
1344 spin_unlock(&inode->i_lock);
1345 spin_unlock(&inode_hash_lock);
1346 return 0;
1347 }
1348 __iget(old);
1349 spin_unlock(&old->i_lock);
1350 spin_unlock(&inode_hash_lock);
1351 wait_on_inode(old);
1352 if (unlikely(!inode_unhashed(old))) {
1353 iput(old);
1354 return -EBUSY;
1355 }
1356 iput(old);
1357 }
1358}
1359EXPORT_SYMBOL(insert_inode_locked4);
1360
1361
1362int generic_delete_inode(struct inode *inode)
1363{
1364 return 1;
1365}
1366EXPORT_SYMBOL(generic_delete_inode);
1367
1368/*
1369 * Called when we're dropping the last reference
1370 * to an inode.
1371 *
1372 * Call the FS "drop_inode()" function, defaulting to
1373 * the legacy UNIX filesystem behaviour. If it tells
1374 * us to evict inode, do so. Otherwise, retain inode
1375 * in cache if fs is alive, sync and evict if fs is
1376 * shutting down.
1377 */
1378static void iput_final(struct inode *inode)
1379{
1380 struct super_block *sb = inode->i_sb;
1381 const struct super_operations *op = inode->i_sb->s_op;
1382 int drop;
1383
1384 WARN_ON(inode->i_state & I_NEW);
1385
1386 if (op->drop_inode)
1387 drop = op->drop_inode(inode);
1388 else
1389 drop = generic_drop_inode(inode);
1390
1391 if (!drop && (sb->s_flags & MS_ACTIVE)) {
1392 inode->i_state |= I_REFERENCED;
1393 if (!(inode->i_state & (I_DIRTY|I_SYNC)))
1394 inode_lru_list_add(inode);
1395 spin_unlock(&inode->i_lock);
1396 return;
1397 }
1398
1399 if (!drop) {
1400 inode->i_state |= I_WILL_FREE;
1401 spin_unlock(&inode->i_lock);
1402 write_inode_now(inode, 1);
1403 spin_lock(&inode->i_lock);
1404 WARN_ON(inode->i_state & I_NEW);
1405 inode->i_state &= ~I_WILL_FREE;
1406 }
1407
1408 inode->i_state |= I_FREEING;
1409 if (!list_empty(&inode->i_lru))
1410 inode_lru_list_del(inode);
1411 spin_unlock(&inode->i_lock);
1412
1413 evict(inode);
1414}
1415
1416/**
1417 * iput - put an inode
1418 * @inode: inode to put
1419 *
1420 * Puts an inode, dropping its usage count. If the inode use count hits
1421 * zero, the inode is then freed and may also be destroyed.
1422 *
1423 * Consequently, iput() can sleep.
1424 */
1425void iput(struct inode *inode)
1426{
1427 if (inode) {
1428 BUG_ON(inode->i_state & I_CLEAR);
1429
1430 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock))
1431 iput_final(inode);
1432 }
1433}
1434EXPORT_SYMBOL(iput);
1435
1436/**
1437 * bmap - find a block number in a file
1438 * @inode: inode of file
1439 * @block: block to find
1440 *
1441 * Returns the block number on the device holding the inode that
1442 * is the disk block number for the block of the file requested.
1443 * That is, asked for block 4 of inode 1 the function will return the
1444 * disk block relative to the disk start that holds that block of the
1445 * file.
1446 */
1447sector_t bmap(struct inode *inode, sector_t block)
1448{
1449 sector_t res = 0;
1450 if (inode->i_mapping->a_ops->bmap)
1451 res = inode->i_mapping->a_ops->bmap(inode->i_mapping, block);
1452 return res;
1453}
1454EXPORT_SYMBOL(bmap);
1455
1456/*
1457 * With relative atime, only update atime if the previous atime is
1458 * earlier than either the ctime or mtime or if at least a day has
1459 * passed since the last atime update.
1460 */
1461static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1462 struct timespec now)
1463{
1464
1465 if (!(mnt->mnt_flags & MNT_RELATIME))
1466 return 1;
1467 /*
1468 * Is mtime younger than atime? If yes, update atime:
1469 */
1470 if (timespec_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1471 return 1;
1472 /*
1473 * Is ctime younger than atime? If yes, update atime:
1474 */
1475 if (timespec_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1476 return 1;
1477
1478 /*
1479 * Is the previous atime value older than a day? If yes,
1480 * update atime:
1481 */
1482 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1483 return 1;
1484 /*
1485 * Good, we can skip the atime update:
1486 */
1487 return 0;
1488}
1489
1490/*
1491 * This does the actual work of updating an inodes time or version. Must have
1492 * had called mnt_want_write() before calling this.
1493 */
1494static int update_time(struct inode *inode, struct timespec *time, int flags)
1495{
1496 if (inode->i_op->update_time)
1497 return inode->i_op->update_time(inode, time, flags);
1498
1499 if (flags & S_ATIME)
1500 inode->i_atime = *time;
1501 if (flags & S_VERSION)
1502 inode_inc_iversion(inode);
1503 if (flags & S_CTIME)
1504 inode->i_ctime = *time;
1505 if (flags & S_MTIME)
1506 inode->i_mtime = *time;
1507 mark_inode_dirty_sync(inode);
1508 return 0;
1509}
1510
1511/**
1512 * touch_atime - update the access time
1513 * @path: the &struct path to update
1514 *
1515 * Update the accessed time on an inode and mark it for writeback.
1516 * This function automatically handles read only file systems and media,
1517 * as well as the "noatime" flag and inode specific "noatime" markers.
1518 */
1519void touch_atime(struct path *path)
1520{
1521 struct vfsmount *mnt = path->mnt;
1522 struct inode *inode = path->dentry->d_inode;
1523 struct timespec now;
1524
1525 if (inode->i_flags & S_NOATIME)
1526 return;
1527 if (IS_NOATIME(inode))
1528 return;
1529 if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
1530 return;
1531
1532 if (mnt->mnt_flags & MNT_NOATIME)
1533 return;
1534 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1535 return;
1536
1537 now = current_fs_time(inode->i_sb);
1538
1539 if (!relatime_need_update(mnt, inode, now))
1540 return;
1541
1542 if (timespec_equal(&inode->i_atime, &now))
1543 return;
1544
1545 if (mnt_want_write(mnt))
1546 return;
1547
1548 /*
1549 * File systems can error out when updating inodes if they need to
1550 * allocate new space to modify an inode (such is the case for
1551 * Btrfs), but since we touch atime while walking down the path we
1552 * really don't care if we failed to update the atime of the file,
1553 * so just ignore the return value.
1554 */
1555 update_time(inode, &now, S_ATIME);
1556 mnt_drop_write(mnt);
1557}
1558EXPORT_SYMBOL(touch_atime);
1559
1560/*
1561 * The logic we want is
1562 *
1563 * if suid or (sgid and xgrp)
1564 * remove privs
1565 */
1566int should_remove_suid(struct dentry *dentry)
1567{
1568 umode_t mode = dentry->d_inode->i_mode;
1569 int kill = 0;
1570
1571 /* suid always must be killed */
1572 if (unlikely(mode & S_ISUID))
1573 kill = ATTR_KILL_SUID;
1574
1575 /*
1576 * sgid without any exec bits is just a mandatory locking mark; leave
1577 * it alone. If some exec bits are set, it's a real sgid; kill it.
1578 */
1579 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1580 kill |= ATTR_KILL_SGID;
1581
1582 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
1583 return kill;
1584
1585 return 0;
1586}
1587EXPORT_SYMBOL(should_remove_suid);
1588
1589static int __remove_suid(struct dentry *dentry, int kill)
1590{
1591 struct iattr newattrs;
1592
1593 newattrs.ia_valid = ATTR_FORCE | kill;
1594 return notify_change(dentry, &newattrs);
1595}
1596
1597int file_remove_suid(struct file *file)
1598{
1599 struct dentry *dentry = file->f_path.dentry;
1600 struct inode *inode = dentry->d_inode;
1601 int killsuid;
1602 int killpriv;
1603 int error = 0;
1604
1605 /* Fast path for nothing security related */
1606 if (IS_NOSEC(inode))
1607 return 0;
1608
1609 killsuid = should_remove_suid(dentry);
1610 killpriv = security_inode_need_killpriv(dentry);
1611
1612 if (killpriv < 0)
1613 return killpriv;
1614 if (killpriv)
1615 error = security_inode_killpriv(dentry);
1616 if (!error && killsuid)
1617 error = __remove_suid(dentry, killsuid);
1618 if (!error && (inode->i_sb->s_flags & MS_NOSEC))
1619 inode->i_flags |= S_NOSEC;
1620
1621 return error;
1622}
1623EXPORT_SYMBOL(file_remove_suid);
1624
1625/**
1626 * file_update_time - update mtime and ctime time
1627 * @file: file accessed
1628 *
1629 * Update the mtime and ctime members of an inode and mark the inode
1630 * for writeback. Note that this function is meant exclusively for
1631 * usage in the file write path of filesystems, and filesystems may
1632 * choose to explicitly ignore update via this function with the
1633 * S_NOCMTIME inode flag, e.g. for network filesystem where these
1634 * timestamps are handled by the server. This can return an error for
1635 * file systems who need to allocate space in order to update an inode.
1636 */
1637
1638int file_update_time(struct file *file)
1639{
1640 struct inode *inode = file->f_path.dentry->d_inode;
1641 struct timespec now;
1642 int sync_it = 0;
1643 int ret;
1644
1645 /* First try to exhaust all avenues to not sync */
1646 if (IS_NOCMTIME(inode))
1647 return 0;
1648
1649 now = current_fs_time(inode->i_sb);
1650 if (!timespec_equal(&inode->i_mtime, &now))
1651 sync_it = S_MTIME;
1652
1653 if (!timespec_equal(&inode->i_ctime, &now))
1654 sync_it |= S_CTIME;
1655
1656 if (IS_I_VERSION(inode))
1657 sync_it |= S_VERSION;
1658
1659 if (!sync_it)
1660 return 0;
1661
1662 /* Finally allowed to write? Takes lock. */
1663 if (mnt_want_write_file(file))
1664 return 0;
1665
1666 ret = update_time(inode, &now, sync_it);
1667 mnt_drop_write_file(file);
1668
1669 return ret;
1670}
1671EXPORT_SYMBOL(file_update_time);
1672
1673int inode_needs_sync(struct inode *inode)
1674{
1675 if (IS_SYNC(inode))
1676 return 1;
1677 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
1678 return 1;
1679 return 0;
1680}
1681EXPORT_SYMBOL(inode_needs_sync);
1682
1683int inode_wait(void *word)
1684{
1685 schedule();
1686 return 0;
1687}
1688EXPORT_SYMBOL(inode_wait);
1689
1690/*
1691 * If we try to find an inode in the inode hash while it is being
1692 * deleted, we have to wait until the filesystem completes its
1693 * deletion before reporting that it isn't found. This function waits
1694 * until the deletion _might_ have completed. Callers are responsible
1695 * to recheck inode state.
1696 *
1697 * It doesn't matter if I_NEW is not set initially, a call to
1698 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
1699 * will DTRT.
1700 */
1701static void __wait_on_freeing_inode(struct inode *inode)
1702{
1703 wait_queue_head_t *wq;
1704 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
1705 wq = bit_waitqueue(&inode->i_state, __I_NEW);
1706 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
1707 spin_unlock(&inode->i_lock);
1708 spin_unlock(&inode_hash_lock);
1709 schedule();
1710 finish_wait(wq, &wait.wait);
1711 spin_lock(&inode_hash_lock);
1712}
1713
1714static __initdata unsigned long ihash_entries;
1715static int __init set_ihash_entries(char *str)
1716{
1717 if (!str)
1718 return 0;
1719 ihash_entries = simple_strtoul(str, &str, 0);
1720 return 1;
1721}
1722__setup("ihash_entries=", set_ihash_entries);
1723
1724/*
1725 * Initialize the waitqueues and inode hash table.
1726 */
1727void __init inode_init_early(void)
1728{
1729 unsigned int loop;
1730
1731 /* If hashes are distributed across NUMA nodes, defer
1732 * hash allocation until vmalloc space is available.
1733 */
1734 if (hashdist)
1735 return;
1736
1737 inode_hashtable =
1738 alloc_large_system_hash("Inode-cache",
1739 sizeof(struct hlist_head),
1740 ihash_entries,
1741 14,
1742 HASH_EARLY,
1743 &i_hash_shift,
1744 &i_hash_mask,
1745 0,
1746 0);
1747
1748 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1749 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1750}
1751
1752void __init inode_init(void)
1753{
1754 unsigned int loop;
1755
1756 /* inode slab cache */
1757 inode_cachep = kmem_cache_create("inode_cache",
1758 sizeof(struct inode),
1759 0,
1760 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1761 SLAB_MEM_SPREAD),
1762 init_once);
1763
1764 /* Hash may have been set up in inode_init_early */
1765 if (!hashdist)
1766 return;
1767
1768 inode_hashtable =
1769 alloc_large_system_hash("Inode-cache",
1770 sizeof(struct hlist_head),
1771 ihash_entries,
1772 14,
1773 0,
1774 &i_hash_shift,
1775 &i_hash_mask,
1776 0,
1777 0);
1778
1779 for (loop = 0; loop < (1U << i_hash_shift); loop++)
1780 INIT_HLIST_HEAD(&inode_hashtable[loop]);
1781}
1782
1783void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
1784{
1785 inode->i_mode = mode;
1786 if (S_ISCHR(mode)) {
1787 inode->i_fop = &def_chr_fops;
1788 inode->i_rdev = rdev;
1789 } else if (S_ISBLK(mode)) {
1790 inode->i_fop = &def_blk_fops;
1791 inode->i_rdev = rdev;
1792 } else if (S_ISFIFO(mode))
1793 inode->i_fop = &def_fifo_fops;
1794 else if (S_ISSOCK(mode))
1795 inode->i_fop = &bad_sock_fops;
1796 else
1797 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
1798 " inode %s:%lu\n", mode, inode->i_sb->s_id,
1799 inode->i_ino);
1800}
1801EXPORT_SYMBOL(init_special_inode);
1802
1803/**
1804 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
1805 * @inode: New inode
1806 * @dir: Directory inode
1807 * @mode: mode of the new inode
1808 */
1809void inode_init_owner(struct inode *inode, const struct inode *dir,
1810 umode_t mode)
1811{
1812 inode->i_uid = current_fsuid();
1813 if (dir && dir->i_mode & S_ISGID) {
1814 inode->i_gid = dir->i_gid;
1815 if (S_ISDIR(mode))
1816 mode |= S_ISGID;
1817 } else
1818 inode->i_gid = current_fsgid();
1819 inode->i_mode = mode;
1820}
1821EXPORT_SYMBOL(inode_init_owner);
1822
1823/**
1824 * inode_owner_or_capable - check current task permissions to inode
1825 * @inode: inode being checked
1826 *
1827 * Return true if current either has CAP_FOWNER to the inode, or
1828 * owns the file.
1829 */
1830bool inode_owner_or_capable(const struct inode *inode)
1831{
1832 if (uid_eq(current_fsuid(), inode->i_uid))
1833 return true;
1834 if (inode_capable(inode, CAP_FOWNER))
1835 return true;
1836 return false;
1837}
1838EXPORT_SYMBOL(inode_owner_or_capable);
1839
1840/*
1841 * Direct i/o helper functions
1842 */
1843static void __inode_dio_wait(struct inode *inode)
1844{
1845 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
1846 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
1847
1848 do {
1849 prepare_to_wait(wq, &q.wait, TASK_UNINTERRUPTIBLE);
1850 if (atomic_read(&inode->i_dio_count))
1851 schedule();
1852 } while (atomic_read(&inode->i_dio_count));
1853 finish_wait(wq, &q.wait);
1854}
1855
1856/**
1857 * inode_dio_wait - wait for outstanding DIO requests to finish
1858 * @inode: inode to wait for
1859 *
1860 * Waits for all pending direct I/O requests to finish so that we can
1861 * proceed with a truncate or equivalent operation.
1862 *
1863 * Must be called under a lock that serializes taking new references
1864 * to i_dio_count, usually by inode->i_mutex.
1865 */
1866void inode_dio_wait(struct inode *inode)
1867{
1868 if (atomic_read(&inode->i_dio_count))
1869 __inode_dio_wait(inode);
1870}
1871EXPORT_SYMBOL(inode_dio_wait);
1872
1873/*
1874 * inode_dio_done - signal finish of a direct I/O requests
1875 * @inode: inode the direct I/O happens on
1876 *
1877 * This is called once we've finished processing a direct I/O request,
1878 * and is used to wake up callers waiting for direct I/O to be quiesced.
1879 */
1880void inode_dio_done(struct inode *inode)
1881{
1882 if (atomic_dec_and_test(&inode->i_dio_count))
1883 wake_up_bit(&inode->i_state, __I_DIO_WAKEUP);
1884}
1885EXPORT_SYMBOL(inode_dio_done);