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