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// 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 <linux/rw_hint.h>
24#include <trace/events/writeback.h>
25#include "internal.h"
26
27/*
28 * Inode locking rules:
29 *
30 * inode->i_lock protects:
31 * inode->i_state, inode->i_hash, __iget(), inode->i_io_list
32 * Inode LRU list locks protect:
33 * inode->i_sb->s_inode_lru, inode->i_lru
34 * inode->i_sb->s_inode_list_lock protects:
35 * inode->i_sb->s_inodes, inode->i_sb_list
36 * bdi->wb.list_lock protects:
37 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
38 * inode_hash_lock protects:
39 * inode_hashtable, inode->i_hash
40 *
41 * Lock ordering:
42 *
43 * inode->i_sb->s_inode_list_lock
44 * inode->i_lock
45 * Inode LRU list locks
46 *
47 * bdi->wb.list_lock
48 * inode->i_lock
49 *
50 * inode_hash_lock
51 * inode->i_sb->s_inode_list_lock
52 * inode->i_lock
53 *
54 * iunique_lock
55 * inode_hash_lock
56 */
57
58static unsigned int i_hash_mask __ro_after_init;
59static unsigned int i_hash_shift __ro_after_init;
60static struct hlist_head *inode_hashtable __ro_after_init;
61static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
62
63/*
64 * Empty aops. Can be used for the cases where the user does not
65 * define any of the address_space operations.
66 */
67const struct address_space_operations empty_aops = {
68};
69EXPORT_SYMBOL(empty_aops);
70
71static DEFINE_PER_CPU(unsigned long, nr_inodes);
72static DEFINE_PER_CPU(unsigned long, nr_unused);
73
74static struct kmem_cache *inode_cachep __ro_after_init;
75
76static long get_nr_inodes(void)
77{
78 int i;
79 long sum = 0;
80 for_each_possible_cpu(i)
81 sum += per_cpu(nr_inodes, i);
82 return sum < 0 ? 0 : sum;
83}
84
85static inline long get_nr_inodes_unused(void)
86{
87 int i;
88 long sum = 0;
89 for_each_possible_cpu(i)
90 sum += per_cpu(nr_unused, i);
91 return sum < 0 ? 0 : sum;
92}
93
94long get_nr_dirty_inodes(void)
95{
96 /* not actually dirty inodes, but a wild approximation */
97 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
98 return nr_dirty > 0 ? nr_dirty : 0;
99}
100
101/*
102 * Handle nr_inode sysctl
103 */
104#ifdef CONFIG_SYSCTL
105/*
106 * Statistics gathering..
107 */
108static struct inodes_stat_t inodes_stat;
109
110static int proc_nr_inodes(struct ctl_table *table, int write, void *buffer,
111 size_t *lenp, loff_t *ppos)
112{
113 inodes_stat.nr_inodes = get_nr_inodes();
114 inodes_stat.nr_unused = get_nr_inodes_unused();
115 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
116}
117
118static struct ctl_table inodes_sysctls[] = {
119 {
120 .procname = "inode-nr",
121 .data = &inodes_stat,
122 .maxlen = 2*sizeof(long),
123 .mode = 0444,
124 .proc_handler = proc_nr_inodes,
125 },
126 {
127 .procname = "inode-state",
128 .data = &inodes_stat,
129 .maxlen = 7*sizeof(long),
130 .mode = 0444,
131 .proc_handler = proc_nr_inodes,
132 },
133};
134
135static int __init init_fs_inode_sysctls(void)
136{
137 register_sysctl_init("fs", inodes_sysctls);
138 return 0;
139}
140early_initcall(init_fs_inode_sysctls);
141#endif
142
143static int no_open(struct inode *inode, struct file *file)
144{
145 return -ENXIO;
146}
147
148/**
149 * inode_init_always - perform inode structure initialisation
150 * @sb: superblock inode belongs to
151 * @inode: inode to initialise
152 *
153 * These are initializations that need to be done on every inode
154 * allocation as the fields are not initialised by slab allocation.
155 */
156int inode_init_always(struct super_block *sb, struct inode *inode)
157{
158 static const struct inode_operations empty_iops;
159 static const struct file_operations no_open_fops = {.open = no_open};
160 struct address_space *const mapping = &inode->i_data;
161
162 inode->i_sb = sb;
163 inode->i_blkbits = sb->s_blocksize_bits;
164 inode->i_flags = 0;
165 atomic64_set(&inode->i_sequence, 0);
166 atomic_set(&inode->i_count, 1);
167 inode->i_op = &empty_iops;
168 inode->i_fop = &no_open_fops;
169 inode->i_ino = 0;
170 inode->__i_nlink = 1;
171 inode->i_opflags = 0;
172 if (sb->s_xattr)
173 inode->i_opflags |= IOP_XATTR;
174 i_uid_write(inode, 0);
175 i_gid_write(inode, 0);
176 atomic_set(&inode->i_writecount, 0);
177 inode->i_size = 0;
178 inode->i_write_hint = WRITE_LIFE_NOT_SET;
179 inode->i_blocks = 0;
180 inode->i_bytes = 0;
181 inode->i_generation = 0;
182 inode->i_pipe = NULL;
183 inode->i_cdev = NULL;
184 inode->i_link = NULL;
185 inode->i_dir_seq = 0;
186 inode->i_rdev = 0;
187 inode->dirtied_when = 0;
188
189#ifdef CONFIG_CGROUP_WRITEBACK
190 inode->i_wb_frn_winner = 0;
191 inode->i_wb_frn_avg_time = 0;
192 inode->i_wb_frn_history = 0;
193#endif
194
195 spin_lock_init(&inode->i_lock);
196 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
197
198 init_rwsem(&inode->i_rwsem);
199 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
200
201 atomic_set(&inode->i_dio_count, 0);
202
203 mapping->a_ops = &empty_aops;
204 mapping->host = inode;
205 mapping->flags = 0;
206 mapping->wb_err = 0;
207 atomic_set(&mapping->i_mmap_writable, 0);
208#ifdef CONFIG_READ_ONLY_THP_FOR_FS
209 atomic_set(&mapping->nr_thps, 0);
210#endif
211 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
212 mapping->i_private_data = NULL;
213 mapping->writeback_index = 0;
214 init_rwsem(&mapping->invalidate_lock);
215 lockdep_set_class_and_name(&mapping->invalidate_lock,
216 &sb->s_type->invalidate_lock_key,
217 "mapping.invalidate_lock");
218 if (sb->s_iflags & SB_I_STABLE_WRITES)
219 mapping_set_stable_writes(mapping);
220 inode->i_private = NULL;
221 inode->i_mapping = mapping;
222 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
223#ifdef CONFIG_FS_POSIX_ACL
224 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
225#endif
226
227#ifdef CONFIG_FSNOTIFY
228 inode->i_fsnotify_mask = 0;
229#endif
230 inode->i_flctx = NULL;
231
232 if (unlikely(security_inode_alloc(inode)))
233 return -ENOMEM;
234 this_cpu_inc(nr_inodes);
235
236 return 0;
237}
238EXPORT_SYMBOL(inode_init_always);
239
240void free_inode_nonrcu(struct inode *inode)
241{
242 kmem_cache_free(inode_cachep, inode);
243}
244EXPORT_SYMBOL(free_inode_nonrcu);
245
246static void i_callback(struct rcu_head *head)
247{
248 struct inode *inode = container_of(head, struct inode, i_rcu);
249 if (inode->free_inode)
250 inode->free_inode(inode);
251 else
252 free_inode_nonrcu(inode);
253}
254
255static struct inode *alloc_inode(struct super_block *sb)
256{
257 const struct super_operations *ops = sb->s_op;
258 struct inode *inode;
259
260 if (ops->alloc_inode)
261 inode = ops->alloc_inode(sb);
262 else
263 inode = alloc_inode_sb(sb, inode_cachep, GFP_KERNEL);
264
265 if (!inode)
266 return NULL;
267
268 if (unlikely(inode_init_always(sb, inode))) {
269 if (ops->destroy_inode) {
270 ops->destroy_inode(inode);
271 if (!ops->free_inode)
272 return NULL;
273 }
274 inode->free_inode = ops->free_inode;
275 i_callback(&inode->i_rcu);
276 return NULL;
277 }
278
279 return inode;
280}
281
282void __destroy_inode(struct inode *inode)
283{
284 BUG_ON(inode_has_buffers(inode));
285 inode_detach_wb(inode);
286 security_inode_free(inode);
287 fsnotify_inode_delete(inode);
288 locks_free_lock_context(inode);
289 if (!inode->i_nlink) {
290 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
291 atomic_long_dec(&inode->i_sb->s_remove_count);
292 }
293
294#ifdef CONFIG_FS_POSIX_ACL
295 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
296 posix_acl_release(inode->i_acl);
297 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
298 posix_acl_release(inode->i_default_acl);
299#endif
300 this_cpu_dec(nr_inodes);
301}
302EXPORT_SYMBOL(__destroy_inode);
303
304static void destroy_inode(struct inode *inode)
305{
306 const struct super_operations *ops = inode->i_sb->s_op;
307
308 BUG_ON(!list_empty(&inode->i_lru));
309 __destroy_inode(inode);
310 if (ops->destroy_inode) {
311 ops->destroy_inode(inode);
312 if (!ops->free_inode)
313 return;
314 }
315 inode->free_inode = ops->free_inode;
316 call_rcu(&inode->i_rcu, i_callback);
317}
318
319/**
320 * drop_nlink - directly drop an inode's link count
321 * @inode: inode
322 *
323 * This is a low-level filesystem helper to replace any
324 * direct filesystem manipulation of i_nlink. In cases
325 * where we are attempting to track writes to the
326 * filesystem, a decrement to zero means an imminent
327 * write when the file is truncated and actually unlinked
328 * on the filesystem.
329 */
330void drop_nlink(struct inode *inode)
331{
332 WARN_ON(inode->i_nlink == 0);
333 inode->__i_nlink--;
334 if (!inode->i_nlink)
335 atomic_long_inc(&inode->i_sb->s_remove_count);
336}
337EXPORT_SYMBOL(drop_nlink);
338
339/**
340 * clear_nlink - directly zero an inode's link count
341 * @inode: inode
342 *
343 * This is a low-level filesystem helper to replace any
344 * direct filesystem manipulation of i_nlink. See
345 * drop_nlink() for why we care about i_nlink hitting zero.
346 */
347void clear_nlink(struct inode *inode)
348{
349 if (inode->i_nlink) {
350 inode->__i_nlink = 0;
351 atomic_long_inc(&inode->i_sb->s_remove_count);
352 }
353}
354EXPORT_SYMBOL(clear_nlink);
355
356/**
357 * set_nlink - directly set an inode's link count
358 * @inode: inode
359 * @nlink: new nlink (should be non-zero)
360 *
361 * This is a low-level filesystem helper to replace any
362 * direct filesystem manipulation of i_nlink.
363 */
364void set_nlink(struct inode *inode, unsigned int nlink)
365{
366 if (!nlink) {
367 clear_nlink(inode);
368 } else {
369 /* Yes, some filesystems do change nlink from zero to one */
370 if (inode->i_nlink == 0)
371 atomic_long_dec(&inode->i_sb->s_remove_count);
372
373 inode->__i_nlink = nlink;
374 }
375}
376EXPORT_SYMBOL(set_nlink);
377
378/**
379 * inc_nlink - directly increment an inode's link count
380 * @inode: inode
381 *
382 * This is a low-level filesystem helper to replace any
383 * direct filesystem manipulation of i_nlink. Currently,
384 * it is only here for parity with dec_nlink().
385 */
386void inc_nlink(struct inode *inode)
387{
388 if (unlikely(inode->i_nlink == 0)) {
389 WARN_ON(!(inode->i_state & I_LINKABLE));
390 atomic_long_dec(&inode->i_sb->s_remove_count);
391 }
392
393 inode->__i_nlink++;
394}
395EXPORT_SYMBOL(inc_nlink);
396
397static void __address_space_init_once(struct address_space *mapping)
398{
399 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
400 init_rwsem(&mapping->i_mmap_rwsem);
401 INIT_LIST_HEAD(&mapping->i_private_list);
402 spin_lock_init(&mapping->i_private_lock);
403 mapping->i_mmap = RB_ROOT_CACHED;
404}
405
406void address_space_init_once(struct address_space *mapping)
407{
408 memset(mapping, 0, sizeof(*mapping));
409 __address_space_init_once(mapping);
410}
411EXPORT_SYMBOL(address_space_init_once);
412
413/*
414 * These are initializations that only need to be done
415 * once, because the fields are idempotent across use
416 * of the inode, so let the slab aware of that.
417 */
418void inode_init_once(struct inode *inode)
419{
420 memset(inode, 0, sizeof(*inode));
421 INIT_HLIST_NODE(&inode->i_hash);
422 INIT_LIST_HEAD(&inode->i_devices);
423 INIT_LIST_HEAD(&inode->i_io_list);
424 INIT_LIST_HEAD(&inode->i_wb_list);
425 INIT_LIST_HEAD(&inode->i_lru);
426 INIT_LIST_HEAD(&inode->i_sb_list);
427 __address_space_init_once(&inode->i_data);
428 i_size_ordered_init(inode);
429}
430EXPORT_SYMBOL(inode_init_once);
431
432static void init_once(void *foo)
433{
434 struct inode *inode = (struct inode *) foo;
435
436 inode_init_once(inode);
437}
438
439/*
440 * inode->i_lock must be held
441 */
442void __iget(struct inode *inode)
443{
444 atomic_inc(&inode->i_count);
445}
446
447/*
448 * get additional reference to inode; caller must already hold one.
449 */
450void ihold(struct inode *inode)
451{
452 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
453}
454EXPORT_SYMBOL(ihold);
455
456static void __inode_add_lru(struct inode *inode, bool rotate)
457{
458 if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
459 return;
460 if (atomic_read(&inode->i_count))
461 return;
462 if (!(inode->i_sb->s_flags & SB_ACTIVE))
463 return;
464 if (!mapping_shrinkable(&inode->i_data))
465 return;
466
467 if (list_lru_add_obj(&inode->i_sb->s_inode_lru, &inode->i_lru))
468 this_cpu_inc(nr_unused);
469 else if (rotate)
470 inode->i_state |= I_REFERENCED;
471}
472
473/*
474 * Add inode to LRU if needed (inode is unused and clean).
475 *
476 * Needs inode->i_lock held.
477 */
478void inode_add_lru(struct inode *inode)
479{
480 __inode_add_lru(inode, false);
481}
482
483static void inode_lru_list_del(struct inode *inode)
484{
485 if (list_lru_del_obj(&inode->i_sb->s_inode_lru, &inode->i_lru))
486 this_cpu_dec(nr_unused);
487}
488
489/**
490 * inode_sb_list_add - add inode to the superblock list of inodes
491 * @inode: inode to add
492 */
493void inode_sb_list_add(struct inode *inode)
494{
495 spin_lock(&inode->i_sb->s_inode_list_lock);
496 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
497 spin_unlock(&inode->i_sb->s_inode_list_lock);
498}
499EXPORT_SYMBOL_GPL(inode_sb_list_add);
500
501static inline void inode_sb_list_del(struct inode *inode)
502{
503 if (!list_empty(&inode->i_sb_list)) {
504 spin_lock(&inode->i_sb->s_inode_list_lock);
505 list_del_init(&inode->i_sb_list);
506 spin_unlock(&inode->i_sb->s_inode_list_lock);
507 }
508}
509
510static unsigned long hash(struct super_block *sb, unsigned long hashval)
511{
512 unsigned long tmp;
513
514 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
515 L1_CACHE_BYTES;
516 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
517 return tmp & i_hash_mask;
518}
519
520/**
521 * __insert_inode_hash - hash an inode
522 * @inode: unhashed inode
523 * @hashval: unsigned long value used to locate this object in the
524 * inode_hashtable.
525 *
526 * Add an inode to the inode hash for this superblock.
527 */
528void __insert_inode_hash(struct inode *inode, unsigned long hashval)
529{
530 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
531
532 spin_lock(&inode_hash_lock);
533 spin_lock(&inode->i_lock);
534 hlist_add_head_rcu(&inode->i_hash, b);
535 spin_unlock(&inode->i_lock);
536 spin_unlock(&inode_hash_lock);
537}
538EXPORT_SYMBOL(__insert_inode_hash);
539
540/**
541 * __remove_inode_hash - remove an inode from the hash
542 * @inode: inode to unhash
543 *
544 * Remove an inode from the superblock.
545 */
546void __remove_inode_hash(struct inode *inode)
547{
548 spin_lock(&inode_hash_lock);
549 spin_lock(&inode->i_lock);
550 hlist_del_init_rcu(&inode->i_hash);
551 spin_unlock(&inode->i_lock);
552 spin_unlock(&inode_hash_lock);
553}
554EXPORT_SYMBOL(__remove_inode_hash);
555
556void dump_mapping(const struct address_space *mapping)
557{
558 struct inode *host;
559 const struct address_space_operations *a_ops;
560 struct hlist_node *dentry_first;
561 struct dentry *dentry_ptr;
562 struct dentry dentry;
563 unsigned long ino;
564
565 /*
566 * If mapping is an invalid pointer, we don't want to crash
567 * accessing it, so probe everything depending on it carefully.
568 */
569 if (get_kernel_nofault(host, &mapping->host) ||
570 get_kernel_nofault(a_ops, &mapping->a_ops)) {
571 pr_warn("invalid mapping:%px\n", mapping);
572 return;
573 }
574
575 if (!host) {
576 pr_warn("aops:%ps\n", a_ops);
577 return;
578 }
579
580 if (get_kernel_nofault(dentry_first, &host->i_dentry.first) ||
581 get_kernel_nofault(ino, &host->i_ino)) {
582 pr_warn("aops:%ps invalid inode:%px\n", a_ops, host);
583 return;
584 }
585
586 if (!dentry_first) {
587 pr_warn("aops:%ps ino:%lx\n", a_ops, ino);
588 return;
589 }
590
591 dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
592 if (get_kernel_nofault(dentry, dentry_ptr) ||
593 !dentry.d_parent || !dentry.d_name.name) {
594 pr_warn("aops:%ps ino:%lx invalid dentry:%px\n",
595 a_ops, ino, dentry_ptr);
596 return;
597 }
598
599 /*
600 * if dentry is corrupted, the %pd handler may still crash,
601 * but it's unlikely that we reach here with a corrupt mapping
602 */
603 pr_warn("aops:%ps ino:%lx dentry name:\"%pd\"\n", a_ops, ino, &dentry);
604}
605
606void clear_inode(struct inode *inode)
607{
608 /*
609 * We have to cycle the i_pages lock here because reclaim can be in the
610 * process of removing the last page (in __filemap_remove_folio())
611 * and we must not free the mapping under it.
612 */
613 xa_lock_irq(&inode->i_data.i_pages);
614 BUG_ON(inode->i_data.nrpages);
615 /*
616 * Almost always, mapping_empty(&inode->i_data) here; but there are
617 * two known and long-standing ways in which nodes may get left behind
618 * (when deep radix-tree node allocation failed partway; or when THP
619 * collapse_file() failed). Until those two known cases are cleaned up,
620 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
621 * nor even WARN_ON(!mapping_empty).
622 */
623 xa_unlock_irq(&inode->i_data.i_pages);
624 BUG_ON(!list_empty(&inode->i_data.i_private_list));
625 BUG_ON(!(inode->i_state & I_FREEING));
626 BUG_ON(inode->i_state & I_CLEAR);
627 BUG_ON(!list_empty(&inode->i_wb_list));
628 /* don't need i_lock here, no concurrent mods to i_state */
629 inode->i_state = I_FREEING | I_CLEAR;
630}
631EXPORT_SYMBOL(clear_inode);
632
633/*
634 * Free the inode passed in, removing it from the lists it is still connected
635 * to. We remove any pages still attached to the inode and wait for any IO that
636 * is still in progress before finally destroying the inode.
637 *
638 * An inode must already be marked I_FREEING so that we avoid the inode being
639 * moved back onto lists if we race with other code that manipulates the lists
640 * (e.g. writeback_single_inode). The caller is responsible for setting this.
641 *
642 * An inode must already be removed from the LRU list before being evicted from
643 * the cache. This should occur atomically with setting the I_FREEING state
644 * flag, so no inodes here should ever be on the LRU when being evicted.
645 */
646static void evict(struct inode *inode)
647{
648 const struct super_operations *op = inode->i_sb->s_op;
649
650 BUG_ON(!(inode->i_state & I_FREEING));
651 BUG_ON(!list_empty(&inode->i_lru));
652
653 if (!list_empty(&inode->i_io_list))
654 inode_io_list_del(inode);
655
656 inode_sb_list_del(inode);
657
658 /*
659 * Wait for flusher thread to be done with the inode so that filesystem
660 * does not start destroying it while writeback is still running. Since
661 * the inode has I_FREEING set, flusher thread won't start new work on
662 * the inode. We just have to wait for running writeback to finish.
663 */
664 inode_wait_for_writeback(inode);
665
666 if (op->evict_inode) {
667 op->evict_inode(inode);
668 } else {
669 truncate_inode_pages_final(&inode->i_data);
670 clear_inode(inode);
671 }
672 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
673 cd_forget(inode);
674
675 remove_inode_hash(inode);
676
677 spin_lock(&inode->i_lock);
678 wake_up_bit(&inode->i_state, __I_NEW);
679 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
680 spin_unlock(&inode->i_lock);
681
682 destroy_inode(inode);
683}
684
685/*
686 * dispose_list - dispose of the contents of a local list
687 * @head: the head of the list to free
688 *
689 * Dispose-list gets a local list with local inodes in it, so it doesn't
690 * need to worry about list corruption and SMP locks.
691 */
692static void dispose_list(struct list_head *head)
693{
694 while (!list_empty(head)) {
695 struct inode *inode;
696
697 inode = list_first_entry(head, struct inode, i_lru);
698 list_del_init(&inode->i_lru);
699
700 evict(inode);
701 cond_resched();
702 }
703}
704
705/**
706 * evict_inodes - evict all evictable inodes for a superblock
707 * @sb: superblock to operate on
708 *
709 * Make sure that no inodes with zero refcount are retained. This is
710 * called by superblock shutdown after having SB_ACTIVE flag removed,
711 * so any inode reaching zero refcount during or after that call will
712 * be immediately evicted.
713 */
714void evict_inodes(struct super_block *sb)
715{
716 struct inode *inode, *next;
717 LIST_HEAD(dispose);
718
719again:
720 spin_lock(&sb->s_inode_list_lock);
721 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
722 if (atomic_read(&inode->i_count))
723 continue;
724
725 spin_lock(&inode->i_lock);
726 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
727 spin_unlock(&inode->i_lock);
728 continue;
729 }
730
731 inode->i_state |= I_FREEING;
732 inode_lru_list_del(inode);
733 spin_unlock(&inode->i_lock);
734 list_add(&inode->i_lru, &dispose);
735
736 /*
737 * We can have a ton of inodes to evict at unmount time given
738 * enough memory, check to see if we need to go to sleep for a
739 * bit so we don't livelock.
740 */
741 if (need_resched()) {
742 spin_unlock(&sb->s_inode_list_lock);
743 cond_resched();
744 dispose_list(&dispose);
745 goto again;
746 }
747 }
748 spin_unlock(&sb->s_inode_list_lock);
749
750 dispose_list(&dispose);
751}
752EXPORT_SYMBOL_GPL(evict_inodes);
753
754/**
755 * invalidate_inodes - attempt to free all inodes on a superblock
756 * @sb: superblock to operate on
757 *
758 * Attempts to free all inodes (including dirty inodes) for a given superblock.
759 */
760void invalidate_inodes(struct super_block *sb)
761{
762 struct inode *inode, *next;
763 LIST_HEAD(dispose);
764
765again:
766 spin_lock(&sb->s_inode_list_lock);
767 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
768 spin_lock(&inode->i_lock);
769 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
770 spin_unlock(&inode->i_lock);
771 continue;
772 }
773 if (atomic_read(&inode->i_count)) {
774 spin_unlock(&inode->i_lock);
775 continue;
776 }
777
778 inode->i_state |= I_FREEING;
779 inode_lru_list_del(inode);
780 spin_unlock(&inode->i_lock);
781 list_add(&inode->i_lru, &dispose);
782 if (need_resched()) {
783 spin_unlock(&sb->s_inode_list_lock);
784 cond_resched();
785 dispose_list(&dispose);
786 goto again;
787 }
788 }
789 spin_unlock(&sb->s_inode_list_lock);
790
791 dispose_list(&dispose);
792}
793
794/*
795 * Isolate the inode from the LRU in preparation for freeing it.
796 *
797 * If the inode has the I_REFERENCED flag set, then it means that it has been
798 * used recently - the flag is set in iput_final(). When we encounter such an
799 * inode, clear the flag and move it to the back of the LRU so it gets another
800 * pass through the LRU before it gets reclaimed. This is necessary because of
801 * the fact we are doing lazy LRU updates to minimise lock contention so the
802 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
803 * with this flag set because they are the inodes that are out of order.
804 */
805static enum lru_status inode_lru_isolate(struct list_head *item,
806 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
807{
808 struct list_head *freeable = arg;
809 struct inode *inode = container_of(item, struct inode, i_lru);
810
811 /*
812 * We are inverting the lru lock/inode->i_lock here, so use a
813 * trylock. If we fail to get the lock, just skip it.
814 */
815 if (!spin_trylock(&inode->i_lock))
816 return LRU_SKIP;
817
818 /*
819 * Inodes can get referenced, redirtied, or repopulated while
820 * they're already on the LRU, and this can make them
821 * unreclaimable for a while. Remove them lazily here; iput,
822 * sync, or the last page cache deletion will requeue them.
823 */
824 if (atomic_read(&inode->i_count) ||
825 (inode->i_state & ~I_REFERENCED) ||
826 !mapping_shrinkable(&inode->i_data)) {
827 list_lru_isolate(lru, &inode->i_lru);
828 spin_unlock(&inode->i_lock);
829 this_cpu_dec(nr_unused);
830 return LRU_REMOVED;
831 }
832
833 /* Recently referenced inodes get one more pass */
834 if (inode->i_state & I_REFERENCED) {
835 inode->i_state &= ~I_REFERENCED;
836 spin_unlock(&inode->i_lock);
837 return LRU_ROTATE;
838 }
839
840 /*
841 * On highmem systems, mapping_shrinkable() permits dropping
842 * page cache in order to free up struct inodes: lowmem might
843 * be under pressure before the cache inside the highmem zone.
844 */
845 if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
846 __iget(inode);
847 spin_unlock(&inode->i_lock);
848 spin_unlock(lru_lock);
849 if (remove_inode_buffers(inode)) {
850 unsigned long reap;
851 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
852 if (current_is_kswapd())
853 __count_vm_events(KSWAPD_INODESTEAL, reap);
854 else
855 __count_vm_events(PGINODESTEAL, reap);
856 mm_account_reclaimed_pages(reap);
857 }
858 iput(inode);
859 spin_lock(lru_lock);
860 return LRU_RETRY;
861 }
862
863 WARN_ON(inode->i_state & I_NEW);
864 inode->i_state |= I_FREEING;
865 list_lru_isolate_move(lru, &inode->i_lru, freeable);
866 spin_unlock(&inode->i_lock);
867
868 this_cpu_dec(nr_unused);
869 return LRU_REMOVED;
870}
871
872/*
873 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
874 * This is called from the superblock shrinker function with a number of inodes
875 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
876 * then are freed outside inode_lock by dispose_list().
877 */
878long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
879{
880 LIST_HEAD(freeable);
881 long freed;
882
883 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
884 inode_lru_isolate, &freeable);
885 dispose_list(&freeable);
886 return freed;
887}
888
889static void __wait_on_freeing_inode(struct inode *inode);
890/*
891 * Called with the inode lock held.
892 */
893static struct inode *find_inode(struct super_block *sb,
894 struct hlist_head *head,
895 int (*test)(struct inode *, void *),
896 void *data)
897{
898 struct inode *inode = NULL;
899
900repeat:
901 hlist_for_each_entry(inode, head, i_hash) {
902 if (inode->i_sb != sb)
903 continue;
904 if (!test(inode, data))
905 continue;
906 spin_lock(&inode->i_lock);
907 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
908 __wait_on_freeing_inode(inode);
909 goto repeat;
910 }
911 if (unlikely(inode->i_state & I_CREATING)) {
912 spin_unlock(&inode->i_lock);
913 return ERR_PTR(-ESTALE);
914 }
915 __iget(inode);
916 spin_unlock(&inode->i_lock);
917 return inode;
918 }
919 return NULL;
920}
921
922/*
923 * find_inode_fast is the fast path version of find_inode, see the comment at
924 * iget_locked for details.
925 */
926static struct inode *find_inode_fast(struct super_block *sb,
927 struct hlist_head *head, unsigned long ino)
928{
929 struct inode *inode = NULL;
930
931repeat:
932 hlist_for_each_entry(inode, head, i_hash) {
933 if (inode->i_ino != ino)
934 continue;
935 if (inode->i_sb != sb)
936 continue;
937 spin_lock(&inode->i_lock);
938 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
939 __wait_on_freeing_inode(inode);
940 goto repeat;
941 }
942 if (unlikely(inode->i_state & I_CREATING)) {
943 spin_unlock(&inode->i_lock);
944 return ERR_PTR(-ESTALE);
945 }
946 __iget(inode);
947 spin_unlock(&inode->i_lock);
948 return inode;
949 }
950 return NULL;
951}
952
953/*
954 * Each cpu owns a range of LAST_INO_BATCH numbers.
955 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
956 * to renew the exhausted range.
957 *
958 * This does not significantly increase overflow rate because every CPU can
959 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
960 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
961 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
962 * overflow rate by 2x, which does not seem too significant.
963 *
964 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
965 * error if st_ino won't fit in target struct field. Use 32bit counter
966 * here to attempt to avoid that.
967 */
968#define LAST_INO_BATCH 1024
969static DEFINE_PER_CPU(unsigned int, last_ino);
970
971unsigned int get_next_ino(void)
972{
973 unsigned int *p = &get_cpu_var(last_ino);
974 unsigned int res = *p;
975
976#ifdef CONFIG_SMP
977 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
978 static atomic_t shared_last_ino;
979 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
980
981 res = next - LAST_INO_BATCH;
982 }
983#endif
984
985 res++;
986 /* get_next_ino should not provide a 0 inode number */
987 if (unlikely(!res))
988 res++;
989 *p = res;
990 put_cpu_var(last_ino);
991 return res;
992}
993EXPORT_SYMBOL(get_next_ino);
994
995/**
996 * new_inode_pseudo - obtain an inode
997 * @sb: superblock
998 *
999 * Allocates a new inode for given superblock.
1000 * Inode wont be chained in superblock s_inodes list
1001 * This means :
1002 * - fs can't be unmount
1003 * - quotas, fsnotify, writeback can't work
1004 */
1005struct inode *new_inode_pseudo(struct super_block *sb)
1006{
1007 struct inode *inode = alloc_inode(sb);
1008
1009 if (inode) {
1010 spin_lock(&inode->i_lock);
1011 inode->i_state = 0;
1012 spin_unlock(&inode->i_lock);
1013 }
1014 return inode;
1015}
1016
1017/**
1018 * new_inode - obtain an inode
1019 * @sb: superblock
1020 *
1021 * Allocates a new inode for given superblock. The default gfp_mask
1022 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1023 * If HIGHMEM pages are unsuitable or it is known that pages allocated
1024 * for the page cache are not reclaimable or migratable,
1025 * mapping_set_gfp_mask() must be called with suitable flags on the
1026 * newly created inode's mapping
1027 *
1028 */
1029struct inode *new_inode(struct super_block *sb)
1030{
1031 struct inode *inode;
1032
1033 inode = new_inode_pseudo(sb);
1034 if (inode)
1035 inode_sb_list_add(inode);
1036 return inode;
1037}
1038EXPORT_SYMBOL(new_inode);
1039
1040#ifdef CONFIG_DEBUG_LOCK_ALLOC
1041void lockdep_annotate_inode_mutex_key(struct inode *inode)
1042{
1043 if (S_ISDIR(inode->i_mode)) {
1044 struct file_system_type *type = inode->i_sb->s_type;
1045
1046 /* Set new key only if filesystem hasn't already changed it */
1047 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
1048 /*
1049 * ensure nobody is actually holding i_mutex
1050 */
1051 // mutex_destroy(&inode->i_mutex);
1052 init_rwsem(&inode->i_rwsem);
1053 lockdep_set_class(&inode->i_rwsem,
1054 &type->i_mutex_dir_key);
1055 }
1056 }
1057}
1058EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1059#endif
1060
1061/**
1062 * unlock_new_inode - clear the I_NEW state and wake up any waiters
1063 * @inode: new inode to unlock
1064 *
1065 * Called when the inode is fully initialised to clear the new state of the
1066 * inode and wake up anyone waiting for the inode to finish initialisation.
1067 */
1068void unlock_new_inode(struct inode *inode)
1069{
1070 lockdep_annotate_inode_mutex_key(inode);
1071 spin_lock(&inode->i_lock);
1072 WARN_ON(!(inode->i_state & I_NEW));
1073 inode->i_state &= ~I_NEW & ~I_CREATING;
1074 smp_mb();
1075 wake_up_bit(&inode->i_state, __I_NEW);
1076 spin_unlock(&inode->i_lock);
1077}
1078EXPORT_SYMBOL(unlock_new_inode);
1079
1080void discard_new_inode(struct inode *inode)
1081{
1082 lockdep_annotate_inode_mutex_key(inode);
1083 spin_lock(&inode->i_lock);
1084 WARN_ON(!(inode->i_state & I_NEW));
1085 inode->i_state &= ~I_NEW;
1086 smp_mb();
1087 wake_up_bit(&inode->i_state, __I_NEW);
1088 spin_unlock(&inode->i_lock);
1089 iput(inode);
1090}
1091EXPORT_SYMBOL(discard_new_inode);
1092
1093/**
1094 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1095 *
1096 * Lock any non-NULL argument. Passed objects must not be directories.
1097 * Zero, one or two objects may be locked by this function.
1098 *
1099 * @inode1: first inode to lock
1100 * @inode2: second inode to lock
1101 */
1102void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1103{
1104 if (inode1)
1105 WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1106 if (inode2)
1107 WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1108 if (inode1 > inode2)
1109 swap(inode1, inode2);
1110 if (inode1)
1111 inode_lock(inode1);
1112 if (inode2 && inode2 != inode1)
1113 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1114}
1115EXPORT_SYMBOL(lock_two_nondirectories);
1116
1117/**
1118 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1119 * @inode1: first inode to unlock
1120 * @inode2: second inode to unlock
1121 */
1122void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1123{
1124 if (inode1) {
1125 WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1126 inode_unlock(inode1);
1127 }
1128 if (inode2 && inode2 != inode1) {
1129 WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1130 inode_unlock(inode2);
1131 }
1132}
1133EXPORT_SYMBOL(unlock_two_nondirectories);
1134
1135/**
1136 * inode_insert5 - obtain an inode from a mounted file system
1137 * @inode: pre-allocated inode to use for insert to cache
1138 * @hashval: hash value (usually inode number) to get
1139 * @test: callback used for comparisons between inodes
1140 * @set: callback used to initialize a new struct inode
1141 * @data: opaque data pointer to pass to @test and @set
1142 *
1143 * Search for the inode specified by @hashval and @data in the inode cache,
1144 * and if present it is return it with an increased reference count. This is
1145 * a variant of iget5_locked() for callers that don't want to fail on memory
1146 * allocation of inode.
1147 *
1148 * If the inode is not in cache, insert the pre-allocated inode to cache and
1149 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1150 * to fill it in before unlocking it via unlock_new_inode().
1151 *
1152 * Note both @test and @set are called with the inode_hash_lock held, so can't
1153 * sleep.
1154 */
1155struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1156 int (*test)(struct inode *, void *),
1157 int (*set)(struct inode *, void *), void *data)
1158{
1159 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1160 struct inode *old;
1161
1162again:
1163 spin_lock(&inode_hash_lock);
1164 old = find_inode(inode->i_sb, head, test, data);
1165 if (unlikely(old)) {
1166 /*
1167 * Uhhuh, somebody else created the same inode under us.
1168 * Use the old inode instead of the preallocated one.
1169 */
1170 spin_unlock(&inode_hash_lock);
1171 if (IS_ERR(old))
1172 return NULL;
1173 wait_on_inode(old);
1174 if (unlikely(inode_unhashed(old))) {
1175 iput(old);
1176 goto again;
1177 }
1178 return old;
1179 }
1180
1181 if (set && unlikely(set(inode, data))) {
1182 inode = NULL;
1183 goto unlock;
1184 }
1185
1186 /*
1187 * Return the locked inode with I_NEW set, the
1188 * caller is responsible for filling in the contents
1189 */
1190 spin_lock(&inode->i_lock);
1191 inode->i_state |= I_NEW;
1192 hlist_add_head_rcu(&inode->i_hash, head);
1193 spin_unlock(&inode->i_lock);
1194
1195 /*
1196 * Add inode to the sb list if it's not already. It has I_NEW at this
1197 * point, so it should be safe to test i_sb_list locklessly.
1198 */
1199 if (list_empty(&inode->i_sb_list))
1200 inode_sb_list_add(inode);
1201unlock:
1202 spin_unlock(&inode_hash_lock);
1203
1204 return inode;
1205}
1206EXPORT_SYMBOL(inode_insert5);
1207
1208/**
1209 * iget5_locked - obtain an inode from a mounted file system
1210 * @sb: super block of file system
1211 * @hashval: hash value (usually inode number) to get
1212 * @test: callback used for comparisons between inodes
1213 * @set: callback used to initialize a new struct inode
1214 * @data: opaque data pointer to pass to @test and @set
1215 *
1216 * Search for the inode specified by @hashval and @data in the inode cache,
1217 * and if present it is return it with an increased reference count. This is
1218 * a generalized version of iget_locked() for file systems where the inode
1219 * number is not sufficient for unique identification of an inode.
1220 *
1221 * If the inode is not in cache, allocate a new inode and return it locked,
1222 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1223 * before unlocking it via unlock_new_inode().
1224 *
1225 * Note both @test and @set are called with the inode_hash_lock held, so can't
1226 * sleep.
1227 */
1228struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1229 int (*test)(struct inode *, void *),
1230 int (*set)(struct inode *, void *), void *data)
1231{
1232 struct inode *inode = ilookup5(sb, hashval, test, data);
1233
1234 if (!inode) {
1235 struct inode *new = alloc_inode(sb);
1236
1237 if (new) {
1238 new->i_state = 0;
1239 inode = inode_insert5(new, hashval, test, set, data);
1240 if (unlikely(inode != new))
1241 destroy_inode(new);
1242 }
1243 }
1244 return inode;
1245}
1246EXPORT_SYMBOL(iget5_locked);
1247
1248/**
1249 * iget_locked - obtain an inode from a mounted file system
1250 * @sb: super block of file system
1251 * @ino: inode number to get
1252 *
1253 * Search for the inode specified by @ino in the inode cache and if present
1254 * return it with an increased reference count. This is for file systems
1255 * where the inode number is sufficient for unique identification of an inode.
1256 *
1257 * If the inode is not in cache, allocate a new inode and return it locked,
1258 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1259 * before unlocking it via unlock_new_inode().
1260 */
1261struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1262{
1263 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1264 struct inode *inode;
1265again:
1266 spin_lock(&inode_hash_lock);
1267 inode = find_inode_fast(sb, head, ino);
1268 spin_unlock(&inode_hash_lock);
1269 if (inode) {
1270 if (IS_ERR(inode))
1271 return NULL;
1272 wait_on_inode(inode);
1273 if (unlikely(inode_unhashed(inode))) {
1274 iput(inode);
1275 goto again;
1276 }
1277 return inode;
1278 }
1279
1280 inode = alloc_inode(sb);
1281 if (inode) {
1282 struct inode *old;
1283
1284 spin_lock(&inode_hash_lock);
1285 /* We released the lock, so.. */
1286 old = find_inode_fast(sb, head, ino);
1287 if (!old) {
1288 inode->i_ino = ino;
1289 spin_lock(&inode->i_lock);
1290 inode->i_state = I_NEW;
1291 hlist_add_head_rcu(&inode->i_hash, head);
1292 spin_unlock(&inode->i_lock);
1293 inode_sb_list_add(inode);
1294 spin_unlock(&inode_hash_lock);
1295
1296 /* Return the locked inode with I_NEW set, the
1297 * caller is responsible for filling in the contents
1298 */
1299 return inode;
1300 }
1301
1302 /*
1303 * Uhhuh, somebody else created the same inode under
1304 * us. Use the old inode instead of the one we just
1305 * allocated.
1306 */
1307 spin_unlock(&inode_hash_lock);
1308 destroy_inode(inode);
1309 if (IS_ERR(old))
1310 return NULL;
1311 inode = old;
1312 wait_on_inode(inode);
1313 if (unlikely(inode_unhashed(inode))) {
1314 iput(inode);
1315 goto again;
1316 }
1317 }
1318 return inode;
1319}
1320EXPORT_SYMBOL(iget_locked);
1321
1322/*
1323 * search the inode cache for a matching inode number.
1324 * If we find one, then the inode number we are trying to
1325 * allocate is not unique and so we should not use it.
1326 *
1327 * Returns 1 if the inode number is unique, 0 if it is not.
1328 */
1329static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1330{
1331 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1332 struct inode *inode;
1333
1334 hlist_for_each_entry_rcu(inode, b, i_hash) {
1335 if (inode->i_ino == ino && inode->i_sb == sb)
1336 return 0;
1337 }
1338 return 1;
1339}
1340
1341/**
1342 * iunique - get a unique inode number
1343 * @sb: superblock
1344 * @max_reserved: highest reserved inode number
1345 *
1346 * Obtain an inode number that is unique on the system for a given
1347 * superblock. This is used by file systems that have no natural
1348 * permanent inode numbering system. An inode number is returned that
1349 * is higher than the reserved limit but unique.
1350 *
1351 * BUGS:
1352 * With a large number of inodes live on the file system this function
1353 * currently becomes quite slow.
1354 */
1355ino_t iunique(struct super_block *sb, ino_t max_reserved)
1356{
1357 /*
1358 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1359 * error if st_ino won't fit in target struct field. Use 32bit counter
1360 * here to attempt to avoid that.
1361 */
1362 static DEFINE_SPINLOCK(iunique_lock);
1363 static unsigned int counter;
1364 ino_t res;
1365
1366 rcu_read_lock();
1367 spin_lock(&iunique_lock);
1368 do {
1369 if (counter <= max_reserved)
1370 counter = max_reserved + 1;
1371 res = counter++;
1372 } while (!test_inode_iunique(sb, res));
1373 spin_unlock(&iunique_lock);
1374 rcu_read_unlock();
1375
1376 return res;
1377}
1378EXPORT_SYMBOL(iunique);
1379
1380struct inode *igrab(struct inode *inode)
1381{
1382 spin_lock(&inode->i_lock);
1383 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1384 __iget(inode);
1385 spin_unlock(&inode->i_lock);
1386 } else {
1387 spin_unlock(&inode->i_lock);
1388 /*
1389 * Handle the case where s_op->clear_inode is not been
1390 * called yet, and somebody is calling igrab
1391 * while the inode is getting freed.
1392 */
1393 inode = NULL;
1394 }
1395 return inode;
1396}
1397EXPORT_SYMBOL(igrab);
1398
1399/**
1400 * ilookup5_nowait - search for an inode in the inode cache
1401 * @sb: super block of file system to search
1402 * @hashval: hash value (usually inode number) to search for
1403 * @test: callback used for comparisons between inodes
1404 * @data: opaque data pointer to pass to @test
1405 *
1406 * Search for the inode specified by @hashval and @data in the inode cache.
1407 * If the inode is in the cache, the inode is returned with an incremented
1408 * reference count.
1409 *
1410 * Note: I_NEW is not waited upon so you have to be very careful what you do
1411 * with the returned inode. You probably should be using ilookup5() instead.
1412 *
1413 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1414 */
1415struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1416 int (*test)(struct inode *, void *), void *data)
1417{
1418 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1419 struct inode *inode;
1420
1421 spin_lock(&inode_hash_lock);
1422 inode = find_inode(sb, head, test, data);
1423 spin_unlock(&inode_hash_lock);
1424
1425 return IS_ERR(inode) ? NULL : inode;
1426}
1427EXPORT_SYMBOL(ilookup5_nowait);
1428
1429/**
1430 * ilookup5 - search for an inode in the inode cache
1431 * @sb: super block of file system to search
1432 * @hashval: hash value (usually inode number) to search for
1433 * @test: callback used for comparisons between inodes
1434 * @data: opaque data pointer to pass to @test
1435 *
1436 * Search for the inode specified by @hashval and @data in the inode cache,
1437 * and if the inode is in the cache, return the inode with an incremented
1438 * reference count. Waits on I_NEW before returning the inode.
1439 * returned with an incremented reference count.
1440 *
1441 * This is a generalized version of ilookup() for file systems where the
1442 * inode number is not sufficient for unique identification of an inode.
1443 *
1444 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1445 */
1446struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1447 int (*test)(struct inode *, void *), void *data)
1448{
1449 struct inode *inode;
1450again:
1451 inode = ilookup5_nowait(sb, hashval, test, data);
1452 if (inode) {
1453 wait_on_inode(inode);
1454 if (unlikely(inode_unhashed(inode))) {
1455 iput(inode);
1456 goto again;
1457 }
1458 }
1459 return inode;
1460}
1461EXPORT_SYMBOL(ilookup5);
1462
1463/**
1464 * ilookup - search for an inode in the inode cache
1465 * @sb: super block of file system to search
1466 * @ino: inode number to search for
1467 *
1468 * Search for the inode @ino in the inode cache, and if the inode is in the
1469 * cache, the inode is returned with an incremented reference count.
1470 */
1471struct inode *ilookup(struct super_block *sb, unsigned long ino)
1472{
1473 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1474 struct inode *inode;
1475again:
1476 spin_lock(&inode_hash_lock);
1477 inode = find_inode_fast(sb, head, ino);
1478 spin_unlock(&inode_hash_lock);
1479
1480 if (inode) {
1481 if (IS_ERR(inode))
1482 return NULL;
1483 wait_on_inode(inode);
1484 if (unlikely(inode_unhashed(inode))) {
1485 iput(inode);
1486 goto again;
1487 }
1488 }
1489 return inode;
1490}
1491EXPORT_SYMBOL(ilookup);
1492
1493/**
1494 * find_inode_nowait - find an inode in the inode cache
1495 * @sb: super block of file system to search
1496 * @hashval: hash value (usually inode number) to search for
1497 * @match: callback used for comparisons between inodes
1498 * @data: opaque data pointer to pass to @match
1499 *
1500 * Search for the inode specified by @hashval and @data in the inode
1501 * cache, where the helper function @match will return 0 if the inode
1502 * does not match, 1 if the inode does match, and -1 if the search
1503 * should be stopped. The @match function must be responsible for
1504 * taking the i_lock spin_lock and checking i_state for an inode being
1505 * freed or being initialized, and incrementing the reference count
1506 * before returning 1. It also must not sleep, since it is called with
1507 * the inode_hash_lock spinlock held.
1508 *
1509 * This is a even more generalized version of ilookup5() when the
1510 * function must never block --- find_inode() can block in
1511 * __wait_on_freeing_inode() --- or when the caller can not increment
1512 * the reference count because the resulting iput() might cause an
1513 * inode eviction. The tradeoff is that the @match funtion must be
1514 * very carefully implemented.
1515 */
1516struct inode *find_inode_nowait(struct super_block *sb,
1517 unsigned long hashval,
1518 int (*match)(struct inode *, unsigned long,
1519 void *),
1520 void *data)
1521{
1522 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1523 struct inode *inode, *ret_inode = NULL;
1524 int mval;
1525
1526 spin_lock(&inode_hash_lock);
1527 hlist_for_each_entry(inode, head, i_hash) {
1528 if (inode->i_sb != sb)
1529 continue;
1530 mval = match(inode, hashval, data);
1531 if (mval == 0)
1532 continue;
1533 if (mval == 1)
1534 ret_inode = inode;
1535 goto out;
1536 }
1537out:
1538 spin_unlock(&inode_hash_lock);
1539 return ret_inode;
1540}
1541EXPORT_SYMBOL(find_inode_nowait);
1542
1543/**
1544 * find_inode_rcu - find an inode in the inode cache
1545 * @sb: Super block of file system to search
1546 * @hashval: Key to hash
1547 * @test: Function to test match on an inode
1548 * @data: Data for test function
1549 *
1550 * Search for the inode specified by @hashval and @data in the inode cache,
1551 * where the helper function @test will return 0 if the inode does not match
1552 * and 1 if it does. The @test function must be responsible for taking the
1553 * i_lock spin_lock and checking i_state for an inode being freed or being
1554 * initialized.
1555 *
1556 * If successful, this will return the inode for which the @test function
1557 * returned 1 and NULL otherwise.
1558 *
1559 * The @test function is not permitted to take a ref on any inode presented.
1560 * It is also not permitted to sleep.
1561 *
1562 * The caller must hold the RCU read lock.
1563 */
1564struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1565 int (*test)(struct inode *, void *), void *data)
1566{
1567 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1568 struct inode *inode;
1569
1570 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1571 "suspicious find_inode_rcu() usage");
1572
1573 hlist_for_each_entry_rcu(inode, head, i_hash) {
1574 if (inode->i_sb == sb &&
1575 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1576 test(inode, data))
1577 return inode;
1578 }
1579 return NULL;
1580}
1581EXPORT_SYMBOL(find_inode_rcu);
1582
1583/**
1584 * find_inode_by_ino_rcu - Find an inode in the inode cache
1585 * @sb: Super block of file system to search
1586 * @ino: The inode number to match
1587 *
1588 * Search for the inode specified by @hashval and @data in the inode cache,
1589 * where the helper function @test will return 0 if the inode does not match
1590 * and 1 if it does. The @test function must be responsible for taking the
1591 * i_lock spin_lock and checking i_state for an inode being freed or being
1592 * initialized.
1593 *
1594 * If successful, this will return the inode for which the @test function
1595 * returned 1 and NULL otherwise.
1596 *
1597 * The @test function is not permitted to take a ref on any inode presented.
1598 * It is also not permitted to sleep.
1599 *
1600 * The caller must hold the RCU read lock.
1601 */
1602struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1603 unsigned long ino)
1604{
1605 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1606 struct inode *inode;
1607
1608 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1609 "suspicious find_inode_by_ino_rcu() usage");
1610
1611 hlist_for_each_entry_rcu(inode, head, i_hash) {
1612 if (inode->i_ino == ino &&
1613 inode->i_sb == sb &&
1614 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1615 return inode;
1616 }
1617 return NULL;
1618}
1619EXPORT_SYMBOL(find_inode_by_ino_rcu);
1620
1621int insert_inode_locked(struct inode *inode)
1622{
1623 struct super_block *sb = inode->i_sb;
1624 ino_t ino = inode->i_ino;
1625 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1626
1627 while (1) {
1628 struct inode *old = NULL;
1629 spin_lock(&inode_hash_lock);
1630 hlist_for_each_entry(old, head, i_hash) {
1631 if (old->i_ino != ino)
1632 continue;
1633 if (old->i_sb != sb)
1634 continue;
1635 spin_lock(&old->i_lock);
1636 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1637 spin_unlock(&old->i_lock);
1638 continue;
1639 }
1640 break;
1641 }
1642 if (likely(!old)) {
1643 spin_lock(&inode->i_lock);
1644 inode->i_state |= I_NEW | I_CREATING;
1645 hlist_add_head_rcu(&inode->i_hash, head);
1646 spin_unlock(&inode->i_lock);
1647 spin_unlock(&inode_hash_lock);
1648 return 0;
1649 }
1650 if (unlikely(old->i_state & I_CREATING)) {
1651 spin_unlock(&old->i_lock);
1652 spin_unlock(&inode_hash_lock);
1653 return -EBUSY;
1654 }
1655 __iget(old);
1656 spin_unlock(&old->i_lock);
1657 spin_unlock(&inode_hash_lock);
1658 wait_on_inode(old);
1659 if (unlikely(!inode_unhashed(old))) {
1660 iput(old);
1661 return -EBUSY;
1662 }
1663 iput(old);
1664 }
1665}
1666EXPORT_SYMBOL(insert_inode_locked);
1667
1668int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1669 int (*test)(struct inode *, void *), void *data)
1670{
1671 struct inode *old;
1672
1673 inode->i_state |= I_CREATING;
1674 old = inode_insert5(inode, hashval, test, NULL, data);
1675
1676 if (old != inode) {
1677 iput(old);
1678 return -EBUSY;
1679 }
1680 return 0;
1681}
1682EXPORT_SYMBOL(insert_inode_locked4);
1683
1684
1685int generic_delete_inode(struct inode *inode)
1686{
1687 return 1;
1688}
1689EXPORT_SYMBOL(generic_delete_inode);
1690
1691/*
1692 * Called when we're dropping the last reference
1693 * to an inode.
1694 *
1695 * Call the FS "drop_inode()" function, defaulting to
1696 * the legacy UNIX filesystem behaviour. If it tells
1697 * us to evict inode, do so. Otherwise, retain inode
1698 * in cache if fs is alive, sync and evict if fs is
1699 * shutting down.
1700 */
1701static void iput_final(struct inode *inode)
1702{
1703 struct super_block *sb = inode->i_sb;
1704 const struct super_operations *op = inode->i_sb->s_op;
1705 unsigned long state;
1706 int drop;
1707
1708 WARN_ON(inode->i_state & I_NEW);
1709
1710 if (op->drop_inode)
1711 drop = op->drop_inode(inode);
1712 else
1713 drop = generic_drop_inode(inode);
1714
1715 if (!drop &&
1716 !(inode->i_state & I_DONTCACHE) &&
1717 (sb->s_flags & SB_ACTIVE)) {
1718 __inode_add_lru(inode, true);
1719 spin_unlock(&inode->i_lock);
1720 return;
1721 }
1722
1723 state = inode->i_state;
1724 if (!drop) {
1725 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1726 spin_unlock(&inode->i_lock);
1727
1728 write_inode_now(inode, 1);
1729
1730 spin_lock(&inode->i_lock);
1731 state = inode->i_state;
1732 WARN_ON(state & I_NEW);
1733 state &= ~I_WILL_FREE;
1734 }
1735
1736 WRITE_ONCE(inode->i_state, state | I_FREEING);
1737 if (!list_empty(&inode->i_lru))
1738 inode_lru_list_del(inode);
1739 spin_unlock(&inode->i_lock);
1740
1741 evict(inode);
1742}
1743
1744/**
1745 * iput - put an inode
1746 * @inode: inode to put
1747 *
1748 * Puts an inode, dropping its usage count. If the inode use count hits
1749 * zero, the inode is then freed and may also be destroyed.
1750 *
1751 * Consequently, iput() can sleep.
1752 */
1753void iput(struct inode *inode)
1754{
1755 if (!inode)
1756 return;
1757 BUG_ON(inode->i_state & I_CLEAR);
1758retry:
1759 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1760 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1761 atomic_inc(&inode->i_count);
1762 spin_unlock(&inode->i_lock);
1763 trace_writeback_lazytime_iput(inode);
1764 mark_inode_dirty_sync(inode);
1765 goto retry;
1766 }
1767 iput_final(inode);
1768 }
1769}
1770EXPORT_SYMBOL(iput);
1771
1772#ifdef CONFIG_BLOCK
1773/**
1774 * bmap - find a block number in a file
1775 * @inode: inode owning the block number being requested
1776 * @block: pointer containing the block to find
1777 *
1778 * Replaces the value in ``*block`` with the block number on the device holding
1779 * corresponding to the requested block number in the file.
1780 * That is, asked for block 4 of inode 1 the function will replace the
1781 * 4 in ``*block``, with disk block relative to the disk start that holds that
1782 * block of the file.
1783 *
1784 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1785 * hole, returns 0 and ``*block`` is also set to 0.
1786 */
1787int bmap(struct inode *inode, sector_t *block)
1788{
1789 if (!inode->i_mapping->a_ops->bmap)
1790 return -EINVAL;
1791
1792 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1793 return 0;
1794}
1795EXPORT_SYMBOL(bmap);
1796#endif
1797
1798/*
1799 * With relative atime, only update atime if the previous atime is
1800 * earlier than or equal to either the ctime or mtime,
1801 * or if at least a day has passed since the last atime update.
1802 */
1803static bool relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1804 struct timespec64 now)
1805{
1806 struct timespec64 atime, mtime, ctime;
1807
1808 if (!(mnt->mnt_flags & MNT_RELATIME))
1809 return true;
1810 /*
1811 * Is mtime younger than or equal to atime? If yes, update atime:
1812 */
1813 atime = inode_get_atime(inode);
1814 mtime = inode_get_mtime(inode);
1815 if (timespec64_compare(&mtime, &atime) >= 0)
1816 return true;
1817 /*
1818 * Is ctime younger than or equal to atime? If yes, update atime:
1819 */
1820 ctime = inode_get_ctime(inode);
1821 if (timespec64_compare(&ctime, &atime) >= 0)
1822 return true;
1823
1824 /*
1825 * Is the previous atime value older than a day? If yes,
1826 * update atime:
1827 */
1828 if ((long)(now.tv_sec - atime.tv_sec) >= 24*60*60)
1829 return true;
1830 /*
1831 * Good, we can skip the atime update:
1832 */
1833 return false;
1834}
1835
1836/**
1837 * inode_update_timestamps - update the timestamps on the inode
1838 * @inode: inode to be updated
1839 * @flags: S_* flags that needed to be updated
1840 *
1841 * The update_time function is called when an inode's timestamps need to be
1842 * updated for a read or write operation. This function handles updating the
1843 * actual timestamps. It's up to the caller to ensure that the inode is marked
1844 * dirty appropriately.
1845 *
1846 * In the case where any of S_MTIME, S_CTIME, or S_VERSION need to be updated,
1847 * attempt to update all three of them. S_ATIME updates can be handled
1848 * independently of the rest.
1849 *
1850 * Returns a set of S_* flags indicating which values changed.
1851 */
1852int inode_update_timestamps(struct inode *inode, int flags)
1853{
1854 int updated = 0;
1855 struct timespec64 now;
1856
1857 if (flags & (S_MTIME|S_CTIME|S_VERSION)) {
1858 struct timespec64 ctime = inode_get_ctime(inode);
1859 struct timespec64 mtime = inode_get_mtime(inode);
1860
1861 now = inode_set_ctime_current(inode);
1862 if (!timespec64_equal(&now, &ctime))
1863 updated |= S_CTIME;
1864 if (!timespec64_equal(&now, &mtime)) {
1865 inode_set_mtime_to_ts(inode, now);
1866 updated |= S_MTIME;
1867 }
1868 if (IS_I_VERSION(inode) && inode_maybe_inc_iversion(inode, updated))
1869 updated |= S_VERSION;
1870 } else {
1871 now = current_time(inode);
1872 }
1873
1874 if (flags & S_ATIME) {
1875 struct timespec64 atime = inode_get_atime(inode);
1876
1877 if (!timespec64_equal(&now, &atime)) {
1878 inode_set_atime_to_ts(inode, now);
1879 updated |= S_ATIME;
1880 }
1881 }
1882 return updated;
1883}
1884EXPORT_SYMBOL(inode_update_timestamps);
1885
1886/**
1887 * generic_update_time - update the timestamps on the inode
1888 * @inode: inode to be updated
1889 * @flags: S_* flags that needed to be updated
1890 *
1891 * The update_time function is called when an inode's timestamps need to be
1892 * updated for a read or write operation. In the case where any of S_MTIME, S_CTIME,
1893 * or S_VERSION need to be updated we attempt to update all three of them. S_ATIME
1894 * updates can be handled done independently of the rest.
1895 *
1896 * Returns a S_* mask indicating which fields were updated.
1897 */
1898int generic_update_time(struct inode *inode, int flags)
1899{
1900 int updated = inode_update_timestamps(inode, flags);
1901 int dirty_flags = 0;
1902
1903 if (updated & (S_ATIME|S_MTIME|S_CTIME))
1904 dirty_flags = inode->i_sb->s_flags & SB_LAZYTIME ? I_DIRTY_TIME : I_DIRTY_SYNC;
1905 if (updated & S_VERSION)
1906 dirty_flags |= I_DIRTY_SYNC;
1907 __mark_inode_dirty(inode, dirty_flags);
1908 return updated;
1909}
1910EXPORT_SYMBOL(generic_update_time);
1911
1912/*
1913 * This does the actual work of updating an inodes time or version. Must have
1914 * had called mnt_want_write() before calling this.
1915 */
1916int inode_update_time(struct inode *inode, int flags)
1917{
1918 if (inode->i_op->update_time)
1919 return inode->i_op->update_time(inode, flags);
1920 generic_update_time(inode, flags);
1921 return 0;
1922}
1923EXPORT_SYMBOL(inode_update_time);
1924
1925/**
1926 * atime_needs_update - update the access time
1927 * @path: the &struct path to update
1928 * @inode: inode to update
1929 *
1930 * Update the accessed time on an inode and mark it for writeback.
1931 * This function automatically handles read only file systems and media,
1932 * as well as the "noatime" flag and inode specific "noatime" markers.
1933 */
1934bool atime_needs_update(const struct path *path, struct inode *inode)
1935{
1936 struct vfsmount *mnt = path->mnt;
1937 struct timespec64 now, atime;
1938
1939 if (inode->i_flags & S_NOATIME)
1940 return false;
1941
1942 /* Atime updates will likely cause i_uid and i_gid to be written
1943 * back improprely if their true value is unknown to the vfs.
1944 */
1945 if (HAS_UNMAPPED_ID(mnt_idmap(mnt), inode))
1946 return false;
1947
1948 if (IS_NOATIME(inode))
1949 return false;
1950 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1951 return false;
1952
1953 if (mnt->mnt_flags & MNT_NOATIME)
1954 return false;
1955 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1956 return false;
1957
1958 now = current_time(inode);
1959
1960 if (!relatime_need_update(mnt, inode, now))
1961 return false;
1962
1963 atime = inode_get_atime(inode);
1964 if (timespec64_equal(&atime, &now))
1965 return false;
1966
1967 return true;
1968}
1969
1970void touch_atime(const struct path *path)
1971{
1972 struct vfsmount *mnt = path->mnt;
1973 struct inode *inode = d_inode(path->dentry);
1974
1975 if (!atime_needs_update(path, inode))
1976 return;
1977
1978 if (!sb_start_write_trylock(inode->i_sb))
1979 return;
1980
1981 if (mnt_get_write_access(mnt) != 0)
1982 goto skip_update;
1983 /*
1984 * File systems can error out when updating inodes if they need to
1985 * allocate new space to modify an inode (such is the case for
1986 * Btrfs), but since we touch atime while walking down the path we
1987 * really don't care if we failed to update the atime of the file,
1988 * so just ignore the return value.
1989 * We may also fail on filesystems that have the ability to make parts
1990 * of the fs read only, e.g. subvolumes in Btrfs.
1991 */
1992 inode_update_time(inode, S_ATIME);
1993 mnt_put_write_access(mnt);
1994skip_update:
1995 sb_end_write(inode->i_sb);
1996}
1997EXPORT_SYMBOL(touch_atime);
1998
1999/*
2000 * Return mask of changes for notify_change() that need to be done as a
2001 * response to write or truncate. Return 0 if nothing has to be changed.
2002 * Negative value on error (change should be denied).
2003 */
2004int dentry_needs_remove_privs(struct mnt_idmap *idmap,
2005 struct dentry *dentry)
2006{
2007 struct inode *inode = d_inode(dentry);
2008 int mask = 0;
2009 int ret;
2010
2011 if (IS_NOSEC(inode))
2012 return 0;
2013
2014 mask = setattr_should_drop_suidgid(idmap, inode);
2015 ret = security_inode_need_killpriv(dentry);
2016 if (ret < 0)
2017 return ret;
2018 if (ret)
2019 mask |= ATTR_KILL_PRIV;
2020 return mask;
2021}
2022
2023static int __remove_privs(struct mnt_idmap *idmap,
2024 struct dentry *dentry, int kill)
2025{
2026 struct iattr newattrs;
2027
2028 newattrs.ia_valid = ATTR_FORCE | kill;
2029 /*
2030 * Note we call this on write, so notify_change will not
2031 * encounter any conflicting delegations:
2032 */
2033 return notify_change(idmap, dentry, &newattrs, NULL);
2034}
2035
2036int file_remove_privs_flags(struct file *file, unsigned int flags)
2037{
2038 struct dentry *dentry = file_dentry(file);
2039 struct inode *inode = file_inode(file);
2040 int error = 0;
2041 int kill;
2042
2043 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
2044 return 0;
2045
2046 kill = dentry_needs_remove_privs(file_mnt_idmap(file), dentry);
2047 if (kill < 0)
2048 return kill;
2049
2050 if (kill) {
2051 if (flags & IOCB_NOWAIT)
2052 return -EAGAIN;
2053
2054 error = __remove_privs(file_mnt_idmap(file), dentry, kill);
2055 }
2056
2057 if (!error)
2058 inode_has_no_xattr(inode);
2059 return error;
2060}
2061EXPORT_SYMBOL_GPL(file_remove_privs_flags);
2062
2063/**
2064 * file_remove_privs - remove special file privileges (suid, capabilities)
2065 * @file: file to remove privileges from
2066 *
2067 * When file is modified by a write or truncation ensure that special
2068 * file privileges are removed.
2069 *
2070 * Return: 0 on success, negative errno on failure.
2071 */
2072int file_remove_privs(struct file *file)
2073{
2074 return file_remove_privs_flags(file, 0);
2075}
2076EXPORT_SYMBOL(file_remove_privs);
2077
2078static int inode_needs_update_time(struct inode *inode)
2079{
2080 int sync_it = 0;
2081 struct timespec64 now = current_time(inode);
2082 struct timespec64 ts;
2083
2084 /* First try to exhaust all avenues to not sync */
2085 if (IS_NOCMTIME(inode))
2086 return 0;
2087
2088 ts = inode_get_mtime(inode);
2089 if (!timespec64_equal(&ts, &now))
2090 sync_it = S_MTIME;
2091
2092 ts = inode_get_ctime(inode);
2093 if (!timespec64_equal(&ts, &now))
2094 sync_it |= S_CTIME;
2095
2096 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2097 sync_it |= S_VERSION;
2098
2099 return sync_it;
2100}
2101
2102static int __file_update_time(struct file *file, int sync_mode)
2103{
2104 int ret = 0;
2105 struct inode *inode = file_inode(file);
2106
2107 /* try to update time settings */
2108 if (!mnt_get_write_access_file(file)) {
2109 ret = inode_update_time(inode, sync_mode);
2110 mnt_put_write_access_file(file);
2111 }
2112
2113 return ret;
2114}
2115
2116/**
2117 * file_update_time - update mtime and ctime time
2118 * @file: file accessed
2119 *
2120 * Update the mtime and ctime members of an inode and mark the inode for
2121 * writeback. Note that this function is meant exclusively for usage in
2122 * the file write path of filesystems, and filesystems may choose to
2123 * explicitly ignore updates via this function with the _NOCMTIME inode
2124 * flag, e.g. for network filesystem where these imestamps are handled
2125 * by the server. This can return an error for file systems who need to
2126 * allocate space in order to update an inode.
2127 *
2128 * Return: 0 on success, negative errno on failure.
2129 */
2130int file_update_time(struct file *file)
2131{
2132 int ret;
2133 struct inode *inode = file_inode(file);
2134
2135 ret = inode_needs_update_time(inode);
2136 if (ret <= 0)
2137 return ret;
2138
2139 return __file_update_time(file, ret);
2140}
2141EXPORT_SYMBOL(file_update_time);
2142
2143/**
2144 * file_modified_flags - handle mandated vfs changes when modifying a file
2145 * @file: file that was modified
2146 * @flags: kiocb flags
2147 *
2148 * When file has been modified ensure that special
2149 * file privileges are removed and time settings are updated.
2150 *
2151 * If IOCB_NOWAIT is set, special file privileges will not be removed and
2152 * time settings will not be updated. It will return -EAGAIN.
2153 *
2154 * Context: Caller must hold the file's inode lock.
2155 *
2156 * Return: 0 on success, negative errno on failure.
2157 */
2158static int file_modified_flags(struct file *file, int flags)
2159{
2160 int ret;
2161 struct inode *inode = file_inode(file);
2162
2163 /*
2164 * Clear the security bits if the process is not being run by root.
2165 * This keeps people from modifying setuid and setgid binaries.
2166 */
2167 ret = file_remove_privs_flags(file, flags);
2168 if (ret)
2169 return ret;
2170
2171 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2172 return 0;
2173
2174 ret = inode_needs_update_time(inode);
2175 if (ret <= 0)
2176 return ret;
2177 if (flags & IOCB_NOWAIT)
2178 return -EAGAIN;
2179
2180 return __file_update_time(file, ret);
2181}
2182
2183/**
2184 * file_modified - handle mandated vfs changes when modifying a file
2185 * @file: file that was modified
2186 *
2187 * When file has been modified ensure that special
2188 * file privileges are removed and time settings are updated.
2189 *
2190 * Context: Caller must hold the file's inode lock.
2191 *
2192 * Return: 0 on success, negative errno on failure.
2193 */
2194int file_modified(struct file *file)
2195{
2196 return file_modified_flags(file, 0);
2197}
2198EXPORT_SYMBOL(file_modified);
2199
2200/**
2201 * kiocb_modified - handle mandated vfs changes when modifying a file
2202 * @iocb: iocb that was modified
2203 *
2204 * When file has been modified ensure that special
2205 * file privileges are removed and time settings are updated.
2206 *
2207 * Context: Caller must hold the file's inode lock.
2208 *
2209 * Return: 0 on success, negative errno on failure.
2210 */
2211int kiocb_modified(struct kiocb *iocb)
2212{
2213 return file_modified_flags(iocb->ki_filp, iocb->ki_flags);
2214}
2215EXPORT_SYMBOL_GPL(kiocb_modified);
2216
2217int inode_needs_sync(struct inode *inode)
2218{
2219 if (IS_SYNC(inode))
2220 return 1;
2221 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2222 return 1;
2223 return 0;
2224}
2225EXPORT_SYMBOL(inode_needs_sync);
2226
2227/*
2228 * If we try to find an inode in the inode hash while it is being
2229 * deleted, we have to wait until the filesystem completes its
2230 * deletion before reporting that it isn't found. This function waits
2231 * until the deletion _might_ have completed. Callers are responsible
2232 * to recheck inode state.
2233 *
2234 * It doesn't matter if I_NEW is not set initially, a call to
2235 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2236 * will DTRT.
2237 */
2238static void __wait_on_freeing_inode(struct inode *inode)
2239{
2240 wait_queue_head_t *wq;
2241 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2242 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2243 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2244 spin_unlock(&inode->i_lock);
2245 spin_unlock(&inode_hash_lock);
2246 schedule();
2247 finish_wait(wq, &wait.wq_entry);
2248 spin_lock(&inode_hash_lock);
2249}
2250
2251static __initdata unsigned long ihash_entries;
2252static int __init set_ihash_entries(char *str)
2253{
2254 if (!str)
2255 return 0;
2256 ihash_entries = simple_strtoul(str, &str, 0);
2257 return 1;
2258}
2259__setup("ihash_entries=", set_ihash_entries);
2260
2261/*
2262 * Initialize the waitqueues and inode hash table.
2263 */
2264void __init inode_init_early(void)
2265{
2266 /* If hashes are distributed across NUMA nodes, defer
2267 * hash allocation until vmalloc space is available.
2268 */
2269 if (hashdist)
2270 return;
2271
2272 inode_hashtable =
2273 alloc_large_system_hash("Inode-cache",
2274 sizeof(struct hlist_head),
2275 ihash_entries,
2276 14,
2277 HASH_EARLY | HASH_ZERO,
2278 &i_hash_shift,
2279 &i_hash_mask,
2280 0,
2281 0);
2282}
2283
2284void __init inode_init(void)
2285{
2286 /* inode slab cache */
2287 inode_cachep = kmem_cache_create("inode_cache",
2288 sizeof(struct inode),
2289 0,
2290 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2291 SLAB_ACCOUNT),
2292 init_once);
2293
2294 /* Hash may have been set up in inode_init_early */
2295 if (!hashdist)
2296 return;
2297
2298 inode_hashtable =
2299 alloc_large_system_hash("Inode-cache",
2300 sizeof(struct hlist_head),
2301 ihash_entries,
2302 14,
2303 HASH_ZERO,
2304 &i_hash_shift,
2305 &i_hash_mask,
2306 0,
2307 0);
2308}
2309
2310void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2311{
2312 inode->i_mode = mode;
2313 if (S_ISCHR(mode)) {
2314 inode->i_fop = &def_chr_fops;
2315 inode->i_rdev = rdev;
2316 } else if (S_ISBLK(mode)) {
2317 if (IS_ENABLED(CONFIG_BLOCK))
2318 inode->i_fop = &def_blk_fops;
2319 inode->i_rdev = rdev;
2320 } else if (S_ISFIFO(mode))
2321 inode->i_fop = &pipefifo_fops;
2322 else if (S_ISSOCK(mode))
2323 ; /* leave it no_open_fops */
2324 else
2325 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2326 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2327 inode->i_ino);
2328}
2329EXPORT_SYMBOL(init_special_inode);
2330
2331/**
2332 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2333 * @idmap: idmap of the mount the inode was created from
2334 * @inode: New inode
2335 * @dir: Directory inode
2336 * @mode: mode of the new inode
2337 *
2338 * If the inode has been created through an idmapped mount the idmap of
2339 * the vfsmount must be passed through @idmap. This function will then take
2340 * care to map the inode according to @idmap before checking permissions
2341 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2342 * checking is to be performed on the raw inode simply pass @nop_mnt_idmap.
2343 */
2344void inode_init_owner(struct mnt_idmap *idmap, struct inode *inode,
2345 const struct inode *dir, umode_t mode)
2346{
2347 inode_fsuid_set(inode, idmap);
2348 if (dir && dir->i_mode & S_ISGID) {
2349 inode->i_gid = dir->i_gid;
2350
2351 /* Directories are special, and always inherit S_ISGID */
2352 if (S_ISDIR(mode))
2353 mode |= S_ISGID;
2354 } else
2355 inode_fsgid_set(inode, idmap);
2356 inode->i_mode = mode;
2357}
2358EXPORT_SYMBOL(inode_init_owner);
2359
2360/**
2361 * inode_owner_or_capable - check current task permissions to inode
2362 * @idmap: idmap of the mount the inode was found from
2363 * @inode: inode being checked
2364 *
2365 * Return true if current either has CAP_FOWNER in a namespace with the
2366 * inode owner uid mapped, or owns the file.
2367 *
2368 * If the inode has been found through an idmapped mount the idmap of
2369 * the vfsmount must be passed through @idmap. This function will then take
2370 * care to map the inode according to @idmap before checking permissions.
2371 * On non-idmapped mounts or if permission checking is to be performed on the
2372 * raw inode simply pass @nop_mnt_idmap.
2373 */
2374bool inode_owner_or_capable(struct mnt_idmap *idmap,
2375 const struct inode *inode)
2376{
2377 vfsuid_t vfsuid;
2378 struct user_namespace *ns;
2379
2380 vfsuid = i_uid_into_vfsuid(idmap, inode);
2381 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
2382 return true;
2383
2384 ns = current_user_ns();
2385 if (vfsuid_has_mapping(ns, vfsuid) && ns_capable(ns, CAP_FOWNER))
2386 return true;
2387 return false;
2388}
2389EXPORT_SYMBOL(inode_owner_or_capable);
2390
2391/*
2392 * Direct i/o helper functions
2393 */
2394static void __inode_dio_wait(struct inode *inode)
2395{
2396 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2397 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2398
2399 do {
2400 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2401 if (atomic_read(&inode->i_dio_count))
2402 schedule();
2403 } while (atomic_read(&inode->i_dio_count));
2404 finish_wait(wq, &q.wq_entry);
2405}
2406
2407/**
2408 * inode_dio_wait - wait for outstanding DIO requests to finish
2409 * @inode: inode to wait for
2410 *
2411 * Waits for all pending direct I/O requests to finish so that we can
2412 * proceed with a truncate or equivalent operation.
2413 *
2414 * Must be called under a lock that serializes taking new references
2415 * to i_dio_count, usually by inode->i_mutex.
2416 */
2417void inode_dio_wait(struct inode *inode)
2418{
2419 if (atomic_read(&inode->i_dio_count))
2420 __inode_dio_wait(inode);
2421}
2422EXPORT_SYMBOL(inode_dio_wait);
2423
2424/*
2425 * inode_set_flags - atomically set some inode flags
2426 *
2427 * Note: the caller should be holding i_mutex, or else be sure that
2428 * they have exclusive access to the inode structure (i.e., while the
2429 * inode is being instantiated). The reason for the cmpxchg() loop
2430 * --- which wouldn't be necessary if all code paths which modify
2431 * i_flags actually followed this rule, is that there is at least one
2432 * code path which doesn't today so we use cmpxchg() out of an abundance
2433 * of caution.
2434 *
2435 * In the long run, i_mutex is overkill, and we should probably look
2436 * at using the i_lock spinlock to protect i_flags, and then make sure
2437 * it is so documented in include/linux/fs.h and that all code follows
2438 * the locking convention!!
2439 */
2440void inode_set_flags(struct inode *inode, unsigned int flags,
2441 unsigned int mask)
2442{
2443 WARN_ON_ONCE(flags & ~mask);
2444 set_mask_bits(&inode->i_flags, mask, flags);
2445}
2446EXPORT_SYMBOL(inode_set_flags);
2447
2448void inode_nohighmem(struct inode *inode)
2449{
2450 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2451}
2452EXPORT_SYMBOL(inode_nohighmem);
2453
2454/**
2455 * timestamp_truncate - Truncate timespec to a granularity
2456 * @t: Timespec
2457 * @inode: inode being updated
2458 *
2459 * Truncate a timespec to the granularity supported by the fs
2460 * containing the inode. Always rounds down. gran must
2461 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2462 */
2463struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2464{
2465 struct super_block *sb = inode->i_sb;
2466 unsigned int gran = sb->s_time_gran;
2467
2468 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2469 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2470 t.tv_nsec = 0;
2471
2472 /* Avoid division in the common cases 1 ns and 1 s. */
2473 if (gran == 1)
2474 ; /* nothing */
2475 else if (gran == NSEC_PER_SEC)
2476 t.tv_nsec = 0;
2477 else if (gran > 1 && gran < NSEC_PER_SEC)
2478 t.tv_nsec -= t.tv_nsec % gran;
2479 else
2480 WARN(1, "invalid file time granularity: %u", gran);
2481 return t;
2482}
2483EXPORT_SYMBOL(timestamp_truncate);
2484
2485/**
2486 * current_time - Return FS time
2487 * @inode: inode.
2488 *
2489 * Return the current time truncated to the time granularity supported by
2490 * the fs.
2491 *
2492 * Note that inode and inode->sb cannot be NULL.
2493 * Otherwise, the function warns and returns time without truncation.
2494 */
2495struct timespec64 current_time(struct inode *inode)
2496{
2497 struct timespec64 now;
2498
2499 ktime_get_coarse_real_ts64(&now);
2500 return timestamp_truncate(now, inode);
2501}
2502EXPORT_SYMBOL(current_time);
2503
2504/**
2505 * inode_set_ctime_current - set the ctime to current_time
2506 * @inode: inode
2507 *
2508 * Set the inode->i_ctime to the current value for the inode. Returns
2509 * the current value that was assigned to i_ctime.
2510 */
2511struct timespec64 inode_set_ctime_current(struct inode *inode)
2512{
2513 struct timespec64 now = current_time(inode);
2514
2515 inode_set_ctime_to_ts(inode, now);
2516 return now;
2517}
2518EXPORT_SYMBOL(inode_set_ctime_current);
2519
2520/**
2521 * in_group_or_capable - check whether caller is CAP_FSETID privileged
2522 * @idmap: idmap of the mount @inode was found from
2523 * @inode: inode to check
2524 * @vfsgid: the new/current vfsgid of @inode
2525 *
2526 * Check wether @vfsgid is in the caller's group list or if the caller is
2527 * privileged with CAP_FSETID over @inode. This can be used to determine
2528 * whether the setgid bit can be kept or must be dropped.
2529 *
2530 * Return: true if the caller is sufficiently privileged, false if not.
2531 */
2532bool in_group_or_capable(struct mnt_idmap *idmap,
2533 const struct inode *inode, vfsgid_t vfsgid)
2534{
2535 if (vfsgid_in_group_p(vfsgid))
2536 return true;
2537 if (capable_wrt_inode_uidgid(idmap, inode, CAP_FSETID))
2538 return true;
2539 return false;
2540}
2541
2542/**
2543 * mode_strip_sgid - handle the sgid bit for non-directories
2544 * @idmap: idmap of the mount the inode was created from
2545 * @dir: parent directory inode
2546 * @mode: mode of the file to be created in @dir
2547 *
2548 * If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2549 * raised and @dir has the S_ISGID bit raised ensure that the caller is
2550 * either in the group of the parent directory or they have CAP_FSETID
2551 * in their user namespace and are privileged over the parent directory.
2552 * In all other cases, strip the S_ISGID bit from @mode.
2553 *
2554 * Return: the new mode to use for the file
2555 */
2556umode_t mode_strip_sgid(struct mnt_idmap *idmap,
2557 const struct inode *dir, umode_t mode)
2558{
2559 if ((mode & (S_ISGID | S_IXGRP)) != (S_ISGID | S_IXGRP))
2560 return mode;
2561 if (S_ISDIR(mode) || !dir || !(dir->i_mode & S_ISGID))
2562 return mode;
2563 if (in_group_or_capable(idmap, dir, i_gid_into_vfsgid(idmap, dir)))
2564 return mode;
2565 return mode & ~S_ISGID;
2566}
2567EXPORT_SYMBOL(mode_strip_sgid);