1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * (C) 1997 Linus Torvalds
4 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
5 */
6#include <linux/export.h>
7#include <linux/fs.h>
8#include <linux/filelock.h>
9#include <linux/mm.h>
10#include <linux/backing-dev.h>
11#include <linux/hash.h>
12#include <linux/swap.h>
13#include <linux/security.h>
14#include <linux/cdev.h>
15#include <linux/memblock.h>
16#include <linux/fsnotify.h>
17#include <linux/mount.h>
18#include <linux/posix_acl.h>
19#include <linux/buffer_head.h> /* for inode_has_buffers */
20#include <linux/ratelimit.h>
21#include <linux/list_lru.h>
22#include <linux/iversion.h>
23#include <linux/rw_hint.h>
24#include <linux/seq_file.h>
25#include <linux/debugfs.h>
26#include <trace/events/writeback.h>
27#define CREATE_TRACE_POINTS
28#include <trace/events/timestamp.h>
29
30#include "internal.h"
31
32/*
33 * Inode locking rules:
34 *
35 * inode->i_lock protects:
36 * inode->i_state, inode->i_hash, __iget(), inode->i_io_list
37 * Inode LRU list locks protect:
38 * inode->i_sb->s_inode_lru, inode->i_lru
39 * inode->i_sb->s_inode_list_lock protects:
40 * inode->i_sb->s_inodes, inode->i_sb_list
41 * bdi->wb.list_lock protects:
42 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
43 * inode_hash_lock protects:
44 * inode_hashtable, inode->i_hash
45 *
46 * Lock ordering:
47 *
48 * inode->i_sb->s_inode_list_lock
49 * inode->i_lock
50 * Inode LRU list locks
51 *
52 * bdi->wb.list_lock
53 * inode->i_lock
54 *
55 * inode_hash_lock
56 * inode->i_sb->s_inode_list_lock
57 * inode->i_lock
58 *
59 * iunique_lock
60 * inode_hash_lock
61 */
62
63static unsigned int i_hash_mask __ro_after_init;
64static unsigned int i_hash_shift __ro_after_init;
65static struct hlist_head *inode_hashtable __ro_after_init;
66static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
67
68/*
69 * Empty aops. Can be used for the cases where the user does not
70 * define any of the address_space operations.
71 */
72const struct address_space_operations empty_aops = {
73};
74EXPORT_SYMBOL(empty_aops);
75
76static DEFINE_PER_CPU(unsigned long, nr_inodes);
77static DEFINE_PER_CPU(unsigned long, nr_unused);
78
79static struct kmem_cache *inode_cachep __ro_after_init;
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#ifdef CONFIG_DEBUG_FS
107static DEFINE_PER_CPU(long, mg_ctime_updates);
108static DEFINE_PER_CPU(long, mg_fine_stamps);
109static DEFINE_PER_CPU(long, mg_ctime_swaps);
110
111static unsigned long get_mg_ctime_updates(void)
112{
113 unsigned long sum = 0;
114 int i;
115
116 for_each_possible_cpu(i)
117 sum += data_race(per_cpu(mg_ctime_updates, i));
118 return sum;
119}
120
121static unsigned long get_mg_fine_stamps(void)
122{
123 unsigned long sum = 0;
124 int i;
125
126 for_each_possible_cpu(i)
127 sum += data_race(per_cpu(mg_fine_stamps, i));
128 return sum;
129}
130
131static unsigned long get_mg_ctime_swaps(void)
132{
133 unsigned long sum = 0;
134 int i;
135
136 for_each_possible_cpu(i)
137 sum += data_race(per_cpu(mg_ctime_swaps, i));
138 return sum;
139}
140
141#define mgtime_counter_inc(__var) this_cpu_inc(__var)
142
143static int mgts_show(struct seq_file *s, void *p)
144{
145 unsigned long ctime_updates = get_mg_ctime_updates();
146 unsigned long ctime_swaps = get_mg_ctime_swaps();
147 unsigned long fine_stamps = get_mg_fine_stamps();
148 unsigned long floor_swaps = timekeeping_get_mg_floor_swaps();
149
150 seq_printf(m: s, fmt: "%lu %lu %lu %lu\n",
151 ctime_updates, ctime_swaps, fine_stamps, floor_swaps);
152 return 0;
153}
154
155DEFINE_SHOW_ATTRIBUTE(mgts);
156
157static int __init mg_debugfs_init(void)
158{
159 debugfs_create_file("multigrain_timestamps", S_IFREG | S_IRUGO, NULL, NULL, &mgts_fops);
160 return 0;
161}
162late_initcall(mg_debugfs_init);
163
164#else /* ! CONFIG_DEBUG_FS */
165
166#define mgtime_counter_inc(__var) do { } while (0)
167
168#endif /* CONFIG_DEBUG_FS */
169
170/*
171 * Handle nr_inode sysctl
172 */
173#ifdef CONFIG_SYSCTL
174/*
175 * Statistics gathering..
176 */
177static struct inodes_stat_t inodes_stat;
178
179static int proc_nr_inodes(const struct ctl_table *table, int write, void *buffer,
180 size_t *lenp, loff_t *ppos)
181{
182 inodes_stat.nr_inodes = get_nr_inodes();
183 inodes_stat.nr_unused = get_nr_inodes_unused();
184 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
185}
186
187static const struct ctl_table inodes_sysctls[] = {
188 {
189 .procname = "inode-nr",
190 .data = &inodes_stat,
191 .maxlen = 2*sizeof(long),
192 .mode = 0444,
193 .proc_handler = proc_nr_inodes,
194 },
195 {
196 .procname = "inode-state",
197 .data = &inodes_stat,
198 .maxlen = 7*sizeof(long),
199 .mode = 0444,
200 .proc_handler = proc_nr_inodes,
201 },
202};
203
204static int __init init_fs_inode_sysctls(void)
205{
206 register_sysctl_init("fs", inodes_sysctls);
207 return 0;
208}
209early_initcall(init_fs_inode_sysctls);
210#endif
211
212static int no_open(struct inode *inode, struct file *file)
213{
214 return -ENXIO;
215}
216
217/**
218 * inode_init_always_gfp - perform inode structure initialisation
219 * @sb: superblock inode belongs to
220 * @inode: inode to initialise
221 * @gfp: allocation flags
222 *
223 * These are initializations that need to be done on every inode
224 * allocation as the fields are not initialised by slab allocation.
225 * If there are additional allocations required @gfp is used.
226 */
227int inode_init_always_gfp(struct super_block *sb, struct inode *inode, gfp_t gfp)
228{
229 static const struct inode_operations empty_iops;
230 static const struct file_operations no_open_fops = {.open = no_open};
231 struct address_space *const mapping = &inode->i_data;
232
233 inode->i_sb = sb;
234 inode->i_blkbits = sb->s_blocksize_bits;
235 inode->i_flags = 0;
236 inode->i_state = 0;
237 atomic64_set(v: &inode->i_sequence, i: 0);
238 atomic_set(v: &inode->i_count, i: 1);
239 inode->i_op = &empty_iops;
240 inode->i_fop = &no_open_fops;
241 inode->i_ino = 0;
242 inode->__i_nlink = 1;
243 inode->i_opflags = 0;
244 if (sb->s_xattr)
245 inode->i_opflags |= IOP_XATTR;
246 if (sb->s_type->fs_flags & FS_MGTIME)
247 inode->i_opflags |= IOP_MGTIME;
248 i_uid_write(inode, uid: 0);
249 i_gid_write(inode, gid: 0);
250 atomic_set(v: &inode->i_writecount, i: 0);
251 inode->i_size = 0;
252 inode->i_write_hint = WRITE_LIFE_NOT_SET;
253 inode->i_blocks = 0;
254 inode->i_bytes = 0;
255 inode->i_generation = 0;
256 inode->i_pipe = NULL;
257 inode->i_cdev = NULL;
258 inode->i_link = NULL;
259 inode->i_dir_seq = 0;
260 inode->i_rdev = 0;
261 inode->dirtied_when = 0;
262
263#ifdef CONFIG_CGROUP_WRITEBACK
264 inode->i_wb_frn_winner = 0;
265 inode->i_wb_frn_avg_time = 0;
266 inode->i_wb_frn_history = 0;
267#endif
268
269 spin_lock_init(&inode->i_lock);
270 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
271
272 init_rwsem(&inode->i_rwsem);
273 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
274
275 atomic_set(v: &inode->i_dio_count, i: 0);
276
277 mapping->a_ops = &empty_aops;
278 mapping->host = inode;
279 mapping->flags = 0;
280 mapping->wb_err = 0;
281 atomic_set(v: &mapping->i_mmap_writable, i: 0);
282#ifdef CONFIG_READ_ONLY_THP_FOR_FS
283 atomic_set(&mapping->nr_thps, 0);
284#endif
285 mapping_set_gfp_mask(m: mapping, GFP_HIGHUSER_MOVABLE);
286 mapping->i_private_data = NULL;
287 mapping->writeback_index = 0;
288 init_rwsem(&mapping->invalidate_lock);
289 lockdep_set_class_and_name(&mapping->invalidate_lock,
290 &sb->s_type->invalidate_lock_key,
291 "mapping.invalidate_lock");
292 if (sb->s_iflags & SB_I_STABLE_WRITES)
293 mapping_set_stable_writes(mapping);
294 inode->i_private = NULL;
295 inode->i_mapping = mapping;
296 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
297#ifdef CONFIG_FS_POSIX_ACL
298 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
299#endif
300
301#ifdef CONFIG_FSNOTIFY
302 inode->i_fsnotify_mask = 0;
303#endif
304 inode->i_flctx = NULL;
305
306 if (unlikely(security_inode_alloc(inode, gfp)))
307 return -ENOMEM;
308
309 this_cpu_inc(nr_inodes);
310
311 return 0;
312}
313EXPORT_SYMBOL(inode_init_always_gfp);
314
315void free_inode_nonrcu(struct inode *inode)
316{
317 kmem_cache_free(s: inode_cachep, objp: inode);
318}
319EXPORT_SYMBOL(free_inode_nonrcu);
320
321static void i_callback(struct rcu_head *head)
322{
323 struct inode *inode = container_of(head, struct inode, i_rcu);
324 if (inode->free_inode)
325 inode->free_inode(inode);
326 else
327 free_inode_nonrcu(inode);
328}
329
330/**
331 * alloc_inode - obtain an inode
332 * @sb: superblock
333 *
334 * Allocates a new inode for given superblock.
335 * Inode wont be chained in superblock s_inodes list
336 * This means :
337 * - fs can't be unmount
338 * - quotas, fsnotify, writeback can't work
339 */
340struct inode *alloc_inode(struct super_block *sb)
341{
342 const struct super_operations *ops = sb->s_op;
343 struct inode *inode;
344
345 if (ops->alloc_inode)
346 inode = ops->alloc_inode(sb);
347 else
348 inode = alloc_inode_sb(sb, inode_cachep, GFP_KERNEL);
349
350 if (!inode)
351 return NULL;
352
353 if (unlikely(inode_init_always(sb, inode))) {
354 if (ops->destroy_inode) {
355 ops->destroy_inode(inode);
356 if (!ops->free_inode)
357 return NULL;
358 }
359 inode->free_inode = ops->free_inode;
360 i_callback(head: &inode->i_rcu);
361 return NULL;
362 }
363
364 return inode;
365}
366
367void __destroy_inode(struct inode *inode)
368{
369 BUG_ON(inode_has_buffers(inode));
370 inode_detach_wb(inode);
371 security_inode_free(inode);
372 fsnotify_inode_delete(inode);
373 locks_free_lock_context(inode);
374 if (!inode->i_nlink) {
375 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
376 atomic_long_dec(v: &inode->i_sb->s_remove_count);
377 }
378
379#ifdef CONFIG_FS_POSIX_ACL
380 if (inode->i_acl && !is_uncached_acl(acl: inode->i_acl))
381 posix_acl_release(acl: inode->i_acl);
382 if (inode->i_default_acl && !is_uncached_acl(acl: inode->i_default_acl))
383 posix_acl_release(acl: inode->i_default_acl);
384#endif
385 this_cpu_dec(nr_inodes);
386}
387EXPORT_SYMBOL(__destroy_inode);
388
389static void destroy_inode(struct inode *inode)
390{
391 const struct super_operations *ops = inode->i_sb->s_op;
392
393 BUG_ON(!list_empty(&inode->i_lru));
394 __destroy_inode(inode);
395 if (ops->destroy_inode) {
396 ops->destroy_inode(inode);
397 if (!ops->free_inode)
398 return;
399 }
400 inode->free_inode = ops->free_inode;
401 call_rcu(head: &inode->i_rcu, func: i_callback);
402}
403
404/**
405 * drop_nlink - directly drop an inode's link count
406 * @inode: inode
407 *
408 * This is a low-level filesystem helper to replace any
409 * direct filesystem manipulation of i_nlink. In cases
410 * where we are attempting to track writes to the
411 * filesystem, a decrement to zero means an imminent
412 * write when the file is truncated and actually unlinked
413 * on the filesystem.
414 */
415void drop_nlink(struct inode *inode)
416{
417 WARN_ON(inode->i_nlink == 0);
418 inode->__i_nlink--;
419 if (!inode->i_nlink)
420 atomic_long_inc(v: &inode->i_sb->s_remove_count);
421}
422EXPORT_SYMBOL(drop_nlink);
423
424/**
425 * clear_nlink - directly zero an inode's link count
426 * @inode: inode
427 *
428 * This is a low-level filesystem helper to replace any
429 * direct filesystem manipulation of i_nlink. See
430 * drop_nlink() for why we care about i_nlink hitting zero.
431 */
432void clear_nlink(struct inode *inode)
433{
434 if (inode->i_nlink) {
435 inode->__i_nlink = 0;
436 atomic_long_inc(v: &inode->i_sb->s_remove_count);
437 }
438}
439EXPORT_SYMBOL(clear_nlink);
440
441/**
442 * set_nlink - directly set an inode's link count
443 * @inode: inode
444 * @nlink: new nlink (should be non-zero)
445 *
446 * This is a low-level filesystem helper to replace any
447 * direct filesystem manipulation of i_nlink.
448 */
449void set_nlink(struct inode *inode, unsigned int nlink)
450{
451 if (!nlink) {
452 clear_nlink(inode);
453 } else {
454 /* Yes, some filesystems do change nlink from zero to one */
455 if (inode->i_nlink == 0)
456 atomic_long_dec(v: &inode->i_sb->s_remove_count);
457
458 inode->__i_nlink = nlink;
459 }
460}
461EXPORT_SYMBOL(set_nlink);
462
463/**
464 * inc_nlink - directly increment an inode's link count
465 * @inode: inode
466 *
467 * This is a low-level filesystem helper to replace any
468 * direct filesystem manipulation of i_nlink. Currently,
469 * it is only here for parity with dec_nlink().
470 */
471void inc_nlink(struct inode *inode)
472{
473 if (unlikely(inode->i_nlink == 0)) {
474 WARN_ON(!(inode->i_state & I_LINKABLE));
475 atomic_long_dec(v: &inode->i_sb->s_remove_count);
476 }
477
478 inode->__i_nlink++;
479}
480EXPORT_SYMBOL(inc_nlink);
481
482static void __address_space_init_once(struct address_space *mapping)
483{
484 xa_init_flags(xa: &mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
485 init_rwsem(&mapping->i_mmap_rwsem);
486 INIT_LIST_HEAD(list: &mapping->i_private_list);
487 spin_lock_init(&mapping->i_private_lock);
488 mapping->i_mmap = RB_ROOT_CACHED;
489}
490
491void address_space_init_once(struct address_space *mapping)
492{
493 memset(s: mapping, c: 0, n: sizeof(*mapping));
494 __address_space_init_once(mapping);
495}
496EXPORT_SYMBOL(address_space_init_once);
497
498/*
499 * These are initializations that only need to be done
500 * once, because the fields are idempotent across use
501 * of the inode, so let the slab aware of that.
502 */
503void inode_init_once(struct inode *inode)
504{
505 memset(s: inode, c: 0, n: sizeof(*inode));
506 INIT_HLIST_NODE(h: &inode->i_hash);
507 INIT_LIST_HEAD(list: &inode->i_devices);
508 INIT_LIST_HEAD(list: &inode->i_io_list);
509 INIT_LIST_HEAD(list: &inode->i_wb_list);
510 INIT_LIST_HEAD(list: &inode->i_lru);
511 INIT_LIST_HEAD(list: &inode->i_sb_list);
512 __address_space_init_once(mapping: &inode->i_data);
513 i_size_ordered_init(inode);
514}
515EXPORT_SYMBOL(inode_init_once);
516
517static void init_once(void *foo)
518{
519 struct inode *inode = (struct inode *) foo;
520
521 inode_init_once(inode);
522}
523
524/*
525 * get additional reference to inode; caller must already hold one.
526 */
527void ihold(struct inode *inode)
528{
529 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
530}
531EXPORT_SYMBOL(ihold);
532
533static void __inode_add_lru(struct inode *inode, bool rotate)
534{
535 if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
536 return;
537 if (icount_read(inode))
538 return;
539 if (!(inode->i_sb->s_flags & SB_ACTIVE))
540 return;
541 if (!mapping_shrinkable(mapping: &inode->i_data))
542 return;
543
544 if (list_lru_add_obj(lru: &inode->i_sb->s_inode_lru, item: &inode->i_lru))
545 this_cpu_inc(nr_unused);
546 else if (rotate)
547 inode->i_state |= I_REFERENCED;
548}
549
550struct wait_queue_head *inode_bit_waitqueue(struct wait_bit_queue_entry *wqe,
551 struct inode *inode, u32 bit)
552{
553 void *bit_address;
554
555 bit_address = inode_state_wait_address(inode, bit);
556 init_wait_var_entry(wbq_entry: wqe, var: bit_address, flags: 0);
557 return __var_waitqueue(p: bit_address);
558}
559EXPORT_SYMBOL(inode_bit_waitqueue);
560
561/*
562 * Add inode to LRU if needed (inode is unused and clean).
563 *
564 * Needs inode->i_lock held.
565 */
566void inode_add_lru(struct inode *inode)
567{
568 __inode_add_lru(inode, rotate: false);
569}
570
571static void inode_lru_list_del(struct inode *inode)
572{
573 if (list_lru_del_obj(lru: &inode->i_sb->s_inode_lru, item: &inode->i_lru))
574 this_cpu_dec(nr_unused);
575}
576
577static void inode_pin_lru_isolating(struct inode *inode)
578{
579 lockdep_assert_held(&inode->i_lock);
580 WARN_ON(inode->i_state & (I_LRU_ISOLATING | I_FREEING | I_WILL_FREE));
581 inode->i_state |= I_LRU_ISOLATING;
582}
583
584static void inode_unpin_lru_isolating(struct inode *inode)
585{
586 spin_lock(lock: &inode->i_lock);
587 WARN_ON(!(inode->i_state & I_LRU_ISOLATING));
588 inode->i_state &= ~I_LRU_ISOLATING;
589 /* Called with inode->i_lock which ensures memory ordering. */
590 inode_wake_up_bit(inode, bit: __I_LRU_ISOLATING);
591 spin_unlock(lock: &inode->i_lock);
592}
593
594static void inode_wait_for_lru_isolating(struct inode *inode)
595{
596 struct wait_bit_queue_entry wqe;
597 struct wait_queue_head *wq_head;
598
599 lockdep_assert_held(&inode->i_lock);
600 if (!(inode->i_state & I_LRU_ISOLATING))
601 return;
602
603 wq_head = inode_bit_waitqueue(&wqe, inode, __I_LRU_ISOLATING);
604 for (;;) {
605 prepare_to_wait_event(wq_head, wq_entry: &wqe.wq_entry, TASK_UNINTERRUPTIBLE);
606 /*
607 * Checking I_LRU_ISOLATING with inode->i_lock guarantees
608 * memory ordering.
609 */
610 if (!(inode->i_state & I_LRU_ISOLATING))
611 break;
612 spin_unlock(lock: &inode->i_lock);
613 schedule();
614 spin_lock(lock: &inode->i_lock);
615 }
616 finish_wait(wq_head, wq_entry: &wqe.wq_entry);
617 WARN_ON(inode->i_state & I_LRU_ISOLATING);
618}
619
620/**
621 * inode_sb_list_add - add inode to the superblock list of inodes
622 * @inode: inode to add
623 */
624void inode_sb_list_add(struct inode *inode)
625{
626 struct super_block *sb = inode->i_sb;
627
628 spin_lock(lock: &sb->s_inode_list_lock);
629 list_add(new: &inode->i_sb_list, head: &sb->s_inodes);
630 spin_unlock(lock: &sb->s_inode_list_lock);
631}
632EXPORT_SYMBOL_GPL(inode_sb_list_add);
633
634static inline void inode_sb_list_del(struct inode *inode)
635{
636 struct super_block *sb = inode->i_sb;
637
638 if (!list_empty(head: &inode->i_sb_list)) {
639 spin_lock(lock: &sb->s_inode_list_lock);
640 list_del_init(entry: &inode->i_sb_list);
641 spin_unlock(lock: &sb->s_inode_list_lock);
642 }
643}
644
645static unsigned long hash(struct super_block *sb, unsigned long hashval)
646{
647 unsigned long tmp;
648
649 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
650 L1_CACHE_BYTES;
651 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
652 return tmp & i_hash_mask;
653}
654
655/**
656 * __insert_inode_hash - hash an inode
657 * @inode: unhashed inode
658 * @hashval: unsigned long value used to locate this object in the
659 * inode_hashtable.
660 *
661 * Add an inode to the inode hash for this superblock.
662 */
663void __insert_inode_hash(struct inode *inode, unsigned long hashval)
664{
665 struct hlist_head *b = inode_hashtable + hash(sb: inode->i_sb, hashval);
666
667 spin_lock(lock: &inode_hash_lock);
668 spin_lock(lock: &inode->i_lock);
669 hlist_add_head_rcu(n: &inode->i_hash, h: b);
670 spin_unlock(lock: &inode->i_lock);
671 spin_unlock(lock: &inode_hash_lock);
672}
673EXPORT_SYMBOL(__insert_inode_hash);
674
675/**
676 * __remove_inode_hash - remove an inode from the hash
677 * @inode: inode to unhash
678 *
679 * Remove an inode from the superblock.
680 */
681void __remove_inode_hash(struct inode *inode)
682{
683 spin_lock(lock: &inode_hash_lock);
684 spin_lock(lock: &inode->i_lock);
685 hlist_del_init_rcu(n: &inode->i_hash);
686 spin_unlock(lock: &inode->i_lock);
687 spin_unlock(lock: &inode_hash_lock);
688}
689EXPORT_SYMBOL(__remove_inode_hash);
690
691void dump_mapping(const struct address_space *mapping)
692{
693 struct inode *host;
694 const struct address_space_operations *a_ops;
695 struct hlist_node *dentry_first;
696 struct dentry *dentry_ptr;
697 struct dentry dentry;
698 char fname[64] = {};
699 unsigned long ino;
700
701 /*
702 * If mapping is an invalid pointer, we don't want to crash
703 * accessing it, so probe everything depending on it carefully.
704 */
705 if (get_kernel_nofault(host, &mapping->host) ||
706 get_kernel_nofault(a_ops, &mapping->a_ops)) {
707 pr_warn("invalid mapping:%px\n", mapping);
708 return;
709 }
710
711 if (!host) {
712 pr_warn("aops:%ps\n", a_ops);
713 return;
714 }
715
716 if (get_kernel_nofault(dentry_first, &host->i_dentry.first) ||
717 get_kernel_nofault(ino, &host->i_ino)) {
718 pr_warn("aops:%ps invalid inode:%px\n", a_ops, host);
719 return;
720 }
721
722 if (!dentry_first) {
723 pr_warn("aops:%ps ino:%lx\n", a_ops, ino);
724 return;
725 }
726
727 dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
728 if (get_kernel_nofault(dentry, dentry_ptr) ||
729 !dentry.d_parent || !dentry.d_name.name) {
730 pr_warn("aops:%ps ino:%lx invalid dentry:%px\n",
731 a_ops, ino, dentry_ptr);
732 return;
733 }
734
735 if (strncpy_from_kernel_nofault(dst: fname, unsafe_addr: dentry.d_name.name, count: 63) < 0)
736 strscpy(fname, "<invalid>");
737 /*
738 * Even if strncpy_from_kernel_nofault() succeeded,
739 * the fname could be unreliable
740 */
741 pr_warn("aops:%ps ino:%lx dentry name(?):\"%s\"\n",
742 a_ops, ino, fname);
743}
744
745void clear_inode(struct inode *inode)
746{
747 /*
748 * We have to cycle the i_pages lock here because reclaim can be in the
749 * process of removing the last page (in __filemap_remove_folio())
750 * and we must not free the mapping under it.
751 */
752 xa_lock_irq(&inode->i_data.i_pages);
753 BUG_ON(inode->i_data.nrpages);
754 /*
755 * Almost always, mapping_empty(&inode->i_data) here; but there are
756 * two known and long-standing ways in which nodes may get left behind
757 * (when deep radix-tree node allocation failed partway; or when THP
758 * collapse_file() failed). Until those two known cases are cleaned up,
759 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
760 * nor even WARN_ON(!mapping_empty).
761 */
762 xa_unlock_irq(&inode->i_data.i_pages);
763 BUG_ON(!list_empty(&inode->i_data.i_private_list));
764 BUG_ON(!(inode->i_state & I_FREEING));
765 BUG_ON(inode->i_state & I_CLEAR);
766 BUG_ON(!list_empty(&inode->i_wb_list));
767 /* don't need i_lock here, no concurrent mods to i_state */
768 inode->i_state = I_FREEING | I_CLEAR;
769}
770EXPORT_SYMBOL(clear_inode);
771
772/*
773 * Free the inode passed in, removing it from the lists it is still connected
774 * to. We remove any pages still attached to the inode and wait for any IO that
775 * is still in progress before finally destroying the inode.
776 *
777 * An inode must already be marked I_FREEING so that we avoid the inode being
778 * moved back onto lists if we race with other code that manipulates the lists
779 * (e.g. writeback_single_inode). The caller is responsible for setting this.
780 *
781 * An inode must already be removed from the LRU list before being evicted from
782 * the cache. This should occur atomically with setting the I_FREEING state
783 * flag, so no inodes here should ever be on the LRU when being evicted.
784 */
785static void evict(struct inode *inode)
786{
787 const struct super_operations *op = inode->i_sb->s_op;
788
789 BUG_ON(!(inode->i_state & I_FREEING));
790 BUG_ON(!list_empty(&inode->i_lru));
791
792 if (!list_empty(head: &inode->i_io_list))
793 inode_io_list_del(inode);
794
795 inode_sb_list_del(inode);
796
797 spin_lock(lock: &inode->i_lock);
798 inode_wait_for_lru_isolating(inode);
799
800 /*
801 * Wait for flusher thread to be done with the inode so that filesystem
802 * does not start destroying it while writeback is still running. Since
803 * the inode has I_FREEING set, flusher thread won't start new work on
804 * the inode. We just have to wait for running writeback to finish.
805 */
806 inode_wait_for_writeback(inode);
807 spin_unlock(lock: &inode->i_lock);
808
809 if (op->evict_inode) {
810 op->evict_inode(inode);
811 } else {
812 truncate_inode_pages_final(&inode->i_data);
813 clear_inode(inode);
814 }
815 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
816 cd_forget(inode);
817
818 remove_inode_hash(inode);
819
820 /*
821 * Wake up waiters in __wait_on_freeing_inode().
822 *
823 * It is an invariant that any thread we need to wake up is already
824 * accounted for before remove_inode_hash() acquires ->i_lock -- both
825 * sides take the lock and sleep is aborted if the inode is found
826 * unhashed. Thus either the sleeper wins and goes off CPU, or removal
827 * wins and the sleeper aborts after testing with the lock.
828 *
829 * This also means we don't need any fences for the call below.
830 */
831 inode_wake_up_bit(inode, bit: __I_NEW);
832 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
833
834 destroy_inode(inode);
835}
836
837/*
838 * dispose_list - dispose of the contents of a local list
839 * @head: the head of the list to free
840 *
841 * Dispose-list gets a local list with local inodes in it, so it doesn't
842 * need to worry about list corruption and SMP locks.
843 */
844static void dispose_list(struct list_head *head)
845{
846 while (!list_empty(head)) {
847 struct inode *inode;
848
849 inode = list_first_entry(head, struct inode, i_lru);
850 list_del_init(entry: &inode->i_lru);
851
852 evict(inode);
853 cond_resched();
854 }
855}
856
857/**
858 * evict_inodes - evict all evictable inodes for a superblock
859 * @sb: superblock to operate on
860 *
861 * Make sure that no inodes with zero refcount are retained. This is
862 * called by superblock shutdown after having SB_ACTIVE flag removed,
863 * so any inode reaching zero refcount during or after that call will
864 * be immediately evicted.
865 */
866void evict_inodes(struct super_block *sb)
867{
868 struct inode *inode;
869 LIST_HEAD(dispose);
870
871again:
872 spin_lock(lock: &sb->s_inode_list_lock);
873 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
874 if (icount_read(inode))
875 continue;
876
877 spin_lock(lock: &inode->i_lock);
878 if (icount_read(inode)) {
879 spin_unlock(lock: &inode->i_lock);
880 continue;
881 }
882 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
883 spin_unlock(lock: &inode->i_lock);
884 continue;
885 }
886
887 inode->i_state |= I_FREEING;
888 inode_lru_list_del(inode);
889 spin_unlock(lock: &inode->i_lock);
890 list_add(new: &inode->i_lru, head: &dispose);
891
892 /*
893 * We can have a ton of inodes to evict at unmount time given
894 * enough memory, check to see if we need to go to sleep for a
895 * bit so we don't livelock.
896 */
897 if (need_resched()) {
898 spin_unlock(lock: &sb->s_inode_list_lock);
899 cond_resched();
900 dispose_list(head: &dispose);
901 goto again;
902 }
903 }
904 spin_unlock(lock: &sb->s_inode_list_lock);
905
906 dispose_list(head: &dispose);
907}
908EXPORT_SYMBOL_GPL(evict_inodes);
909
910/*
911 * Isolate the inode from the LRU in preparation for freeing it.
912 *
913 * If the inode has the I_REFERENCED flag set, then it means that it has been
914 * used recently - the flag is set in iput_final(). When we encounter such an
915 * inode, clear the flag and move it to the back of the LRU so it gets another
916 * pass through the LRU before it gets reclaimed. This is necessary because of
917 * the fact we are doing lazy LRU updates to minimise lock contention so the
918 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
919 * with this flag set because they are the inodes that are out of order.
920 */
921static enum lru_status inode_lru_isolate(struct list_head *item,
922 struct list_lru_one *lru, void *arg)
923{
924 struct list_head *freeable = arg;
925 struct inode *inode = container_of(item, struct inode, i_lru);
926
927 /*
928 * We are inverting the lru lock/inode->i_lock here, so use a
929 * trylock. If we fail to get the lock, just skip it.
930 */
931 if (!spin_trylock(lock: &inode->i_lock))
932 return LRU_SKIP;
933
934 /*
935 * Inodes can get referenced, redirtied, or repopulated while
936 * they're already on the LRU, and this can make them
937 * unreclaimable for a while. Remove them lazily here; iput,
938 * sync, or the last page cache deletion will requeue them.
939 */
940 if (icount_read(inode) ||
941 (inode->i_state & ~I_REFERENCED) ||
942 !mapping_shrinkable(mapping: &inode->i_data)) {
943 list_lru_isolate(list: lru, item: &inode->i_lru);
944 spin_unlock(lock: &inode->i_lock);
945 this_cpu_dec(nr_unused);
946 return LRU_REMOVED;
947 }
948
949 /* Recently referenced inodes get one more pass */
950 if (inode->i_state & I_REFERENCED) {
951 inode->i_state &= ~I_REFERENCED;
952 spin_unlock(lock: &inode->i_lock);
953 return LRU_ROTATE;
954 }
955
956 /*
957 * On highmem systems, mapping_shrinkable() permits dropping
958 * page cache in order to free up struct inodes: lowmem might
959 * be under pressure before the cache inside the highmem zone.
960 */
961 if (inode_has_buffers(inode) || !mapping_empty(mapping: &inode->i_data)) {
962 inode_pin_lru_isolating(inode);
963 spin_unlock(lock: &inode->i_lock);
964 spin_unlock(lock: &lru->lock);
965 if (remove_inode_buffers(inode)) {
966 unsigned long reap;
967 reap = invalidate_mapping_pages(mapping: &inode->i_data, start: 0, end: -1);
968 if (current_is_kswapd())
969 __count_vm_events(item: KSWAPD_INODESTEAL, delta: reap);
970 else
971 __count_vm_events(item: PGINODESTEAL, delta: reap);
972 mm_account_reclaimed_pages(pages: reap);
973 }
974 inode_unpin_lru_isolating(inode);
975 return LRU_RETRY;
976 }
977
978 WARN_ON(inode->i_state & I_NEW);
979 inode->i_state |= I_FREEING;
980 list_lru_isolate_move(list: lru, item: &inode->i_lru, head: freeable);
981 spin_unlock(lock: &inode->i_lock);
982
983 this_cpu_dec(nr_unused);
984 return LRU_REMOVED;
985}
986
987/*
988 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
989 * This is called from the superblock shrinker function with a number of inodes
990 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
991 * then are freed outside inode_lock by dispose_list().
992 */
993long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
994{
995 LIST_HEAD(freeable);
996 long freed;
997
998 freed = list_lru_shrink_walk(lru: &sb->s_inode_lru, sc,
999 isolate: inode_lru_isolate, cb_arg: &freeable);
1000 dispose_list(head: &freeable);
1001 return freed;
1002}
1003
1004static void __wait_on_freeing_inode(struct inode *inode, bool is_inode_hash_locked);
1005/*
1006 * Called with the inode lock held.
1007 */
1008static struct inode *find_inode(struct super_block *sb,
1009 struct hlist_head *head,
1010 int (*test)(struct inode *, void *),
1011 void *data, bool is_inode_hash_locked)
1012{
1013 struct inode *inode = NULL;
1014
1015 if (is_inode_hash_locked)
1016 lockdep_assert_held(&inode_hash_lock);
1017 else
1018 lockdep_assert_not_held(&inode_hash_lock);
1019
1020 rcu_read_lock();
1021repeat:
1022 hlist_for_each_entry_rcu(inode, head, i_hash) {
1023 if (inode->i_sb != sb)
1024 continue;
1025 if (!test(inode, data))
1026 continue;
1027 spin_lock(lock: &inode->i_lock);
1028 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1029 __wait_on_freeing_inode(inode, is_inode_hash_locked);
1030 goto repeat;
1031 }
1032 if (unlikely(inode->i_state & I_CREATING)) {
1033 spin_unlock(lock: &inode->i_lock);
1034 rcu_read_unlock();
1035 return ERR_PTR(error: -ESTALE);
1036 }
1037 __iget(inode);
1038 spin_unlock(lock: &inode->i_lock);
1039 rcu_read_unlock();
1040 return inode;
1041 }
1042 rcu_read_unlock();
1043 return NULL;
1044}
1045
1046/*
1047 * find_inode_fast is the fast path version of find_inode, see the comment at
1048 * iget_locked for details.
1049 */
1050static struct inode *find_inode_fast(struct super_block *sb,
1051 struct hlist_head *head, unsigned long ino,
1052 bool is_inode_hash_locked)
1053{
1054 struct inode *inode = NULL;
1055
1056 if (is_inode_hash_locked)
1057 lockdep_assert_held(&inode_hash_lock);
1058 else
1059 lockdep_assert_not_held(&inode_hash_lock);
1060
1061 rcu_read_lock();
1062repeat:
1063 hlist_for_each_entry_rcu(inode, head, i_hash) {
1064 if (inode->i_ino != ino)
1065 continue;
1066 if (inode->i_sb != sb)
1067 continue;
1068 spin_lock(lock: &inode->i_lock);
1069 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
1070 __wait_on_freeing_inode(inode, is_inode_hash_locked);
1071 goto repeat;
1072 }
1073 if (unlikely(inode->i_state & I_CREATING)) {
1074 spin_unlock(lock: &inode->i_lock);
1075 rcu_read_unlock();
1076 return ERR_PTR(error: -ESTALE);
1077 }
1078 __iget(inode);
1079 spin_unlock(lock: &inode->i_lock);
1080 rcu_read_unlock();
1081 return inode;
1082 }
1083 rcu_read_unlock();
1084 return NULL;
1085}
1086
1087/*
1088 * Each cpu owns a range of LAST_INO_BATCH numbers.
1089 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
1090 * to renew the exhausted range.
1091 *
1092 * This does not significantly increase overflow rate because every CPU can
1093 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
1094 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
1095 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
1096 * overflow rate by 2x, which does not seem too significant.
1097 *
1098 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1099 * error if st_ino won't fit in target struct field. Use 32bit counter
1100 * here to attempt to avoid that.
1101 */
1102#define LAST_INO_BATCH 1024
1103static DEFINE_PER_CPU(unsigned int, last_ino);
1104
1105unsigned int get_next_ino(void)
1106{
1107 unsigned int *p = &get_cpu_var(last_ino);
1108 unsigned int res = *p;
1109
1110#ifdef CONFIG_SMP
1111 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
1112 static atomic_t shared_last_ino;
1113 int next = atomic_add_return(LAST_INO_BATCH, v: &shared_last_ino);
1114
1115 res = next - LAST_INO_BATCH;
1116 }
1117#endif
1118
1119 res++;
1120 /* get_next_ino should not provide a 0 inode number */
1121 if (unlikely(!res))
1122 res++;
1123 *p = res;
1124 put_cpu_var(last_ino);
1125 return res;
1126}
1127EXPORT_SYMBOL(get_next_ino);
1128
1129/**
1130 * new_inode - obtain an inode
1131 * @sb: superblock
1132 *
1133 * Allocates a new inode for given superblock. The default gfp_mask
1134 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1135 * If HIGHMEM pages are unsuitable or it is known that pages allocated
1136 * for the page cache are not reclaimable or migratable,
1137 * mapping_set_gfp_mask() must be called with suitable flags on the
1138 * newly created inode's mapping
1139 *
1140 */
1141struct inode *new_inode(struct super_block *sb)
1142{
1143 struct inode *inode;
1144
1145 inode = alloc_inode(sb);
1146 if (inode)
1147 inode_sb_list_add(inode);
1148 return inode;
1149}
1150EXPORT_SYMBOL(new_inode);
1151
1152#ifdef CONFIG_DEBUG_LOCK_ALLOC
1153void lockdep_annotate_inode_mutex_key(struct inode *inode)
1154{
1155 if (S_ISDIR(inode->i_mode)) {
1156 struct file_system_type *type = inode->i_sb->s_type;
1157
1158 /* Set new key only if filesystem hasn't already changed it */
1159 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
1160 /*
1161 * ensure nobody is actually holding i_rwsem
1162 */
1163 init_rwsem(&inode->i_rwsem);
1164 lockdep_set_class(&inode->i_rwsem,
1165 &type->i_mutex_dir_key);
1166 }
1167 }
1168}
1169EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1170#endif
1171
1172/**
1173 * unlock_new_inode - clear the I_NEW state and wake up any waiters
1174 * @inode: new inode to unlock
1175 *
1176 * Called when the inode is fully initialised to clear the new state of the
1177 * inode and wake up anyone waiting for the inode to finish initialisation.
1178 */
1179void unlock_new_inode(struct inode *inode)
1180{
1181 lockdep_annotate_inode_mutex_key(inode);
1182 spin_lock(lock: &inode->i_lock);
1183 WARN_ON(!(inode->i_state & I_NEW));
1184 inode->i_state &= ~I_NEW & ~I_CREATING;
1185 /*
1186 * Pairs with the barrier in prepare_to_wait_event() to make sure
1187 * ___wait_var_event() either sees the bit cleared or
1188 * waitqueue_active() check in wake_up_var() sees the waiter.
1189 */
1190 smp_mb();
1191 inode_wake_up_bit(inode, bit: __I_NEW);
1192 spin_unlock(lock: &inode->i_lock);
1193}
1194EXPORT_SYMBOL(unlock_new_inode);
1195
1196void discard_new_inode(struct inode *inode)
1197{
1198 lockdep_annotate_inode_mutex_key(inode);
1199 spin_lock(lock: &inode->i_lock);
1200 WARN_ON(!(inode->i_state & I_NEW));
1201 inode->i_state &= ~I_NEW;
1202 /*
1203 * Pairs with the barrier in prepare_to_wait_event() to make sure
1204 * ___wait_var_event() either sees the bit cleared or
1205 * waitqueue_active() check in wake_up_var() sees the waiter.
1206 */
1207 smp_mb();
1208 inode_wake_up_bit(inode, bit: __I_NEW);
1209 spin_unlock(lock: &inode->i_lock);
1210 iput(inode);
1211}
1212EXPORT_SYMBOL(discard_new_inode);
1213
1214/**
1215 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1216 *
1217 * Lock any non-NULL argument. Passed objects must not be directories.
1218 * Zero, one or two objects may be locked by this function.
1219 *
1220 * @inode1: first inode to lock
1221 * @inode2: second inode to lock
1222 */
1223void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1224{
1225 if (inode1)
1226 WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1227 if (inode2)
1228 WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1229 if (inode1 > inode2)
1230 swap(inode1, inode2);
1231 if (inode1)
1232 inode_lock(inode: inode1);
1233 if (inode2 && inode2 != inode1)
1234 inode_lock_nested(inode: inode2, subclass: I_MUTEX_NONDIR2);
1235}
1236EXPORT_SYMBOL(lock_two_nondirectories);
1237
1238/**
1239 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1240 * @inode1: first inode to unlock
1241 * @inode2: second inode to unlock
1242 */
1243void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1244{
1245 if (inode1) {
1246 WARN_ON_ONCE(S_ISDIR(inode1->i_mode));
1247 inode_unlock(inode: inode1);
1248 }
1249 if (inode2 && inode2 != inode1) {
1250 WARN_ON_ONCE(S_ISDIR(inode2->i_mode));
1251 inode_unlock(inode: inode2);
1252 }
1253}
1254EXPORT_SYMBOL(unlock_two_nondirectories);
1255
1256/**
1257 * inode_insert5 - obtain an inode from a mounted file system
1258 * @inode: pre-allocated inode to use for insert to cache
1259 * @hashval: hash value (usually inode number) to get
1260 * @test: callback used for comparisons between inodes
1261 * @set: callback used to initialize a new struct inode
1262 * @data: opaque data pointer to pass to @test and @set
1263 *
1264 * Search for the inode specified by @hashval and @data in the inode cache,
1265 * and if present return it with an increased reference count. This is a
1266 * variant of iget5_locked() that doesn't allocate an inode.
1267 *
1268 * If the inode is not present in the cache, insert the pre-allocated inode and
1269 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1270 * to fill it in before unlocking it via unlock_new_inode().
1271 *
1272 * Note that both @test and @set are called with the inode_hash_lock held, so
1273 * they can't sleep.
1274 */
1275struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1276 int (*test)(struct inode *, void *),
1277 int (*set)(struct inode *, void *), void *data)
1278{
1279 struct hlist_head *head = inode_hashtable + hash(sb: inode->i_sb, hashval);
1280 struct inode *old;
1281
1282 might_sleep();
1283
1284again:
1285 spin_lock(lock: &inode_hash_lock);
1286 old = find_inode(sb: inode->i_sb, head, test, data, is_inode_hash_locked: true);
1287 if (unlikely(old)) {
1288 /*
1289 * Uhhuh, somebody else created the same inode under us.
1290 * Use the old inode instead of the preallocated one.
1291 */
1292 spin_unlock(lock: &inode_hash_lock);
1293 if (IS_ERR(ptr: old))
1294 return NULL;
1295 wait_on_inode(inode: old);
1296 if (unlikely(inode_unhashed(old))) {
1297 iput(old);
1298 goto again;
1299 }
1300 return old;
1301 }
1302
1303 if (set && unlikely(set(inode, data))) {
1304 spin_unlock(lock: &inode_hash_lock);
1305 return NULL;
1306 }
1307
1308 /*
1309 * Return the locked inode with I_NEW set, the
1310 * caller is responsible for filling in the contents
1311 */
1312 spin_lock(lock: &inode->i_lock);
1313 inode->i_state |= I_NEW;
1314 hlist_add_head_rcu(n: &inode->i_hash, h: head);
1315 spin_unlock(lock: &inode->i_lock);
1316
1317 spin_unlock(lock: &inode_hash_lock);
1318
1319 /*
1320 * Add inode to the sb list if it's not already. It has I_NEW at this
1321 * point, so it should be safe to test i_sb_list locklessly.
1322 */
1323 if (list_empty(head: &inode->i_sb_list))
1324 inode_sb_list_add(inode);
1325
1326 return inode;
1327}
1328EXPORT_SYMBOL(inode_insert5);
1329
1330/**
1331 * iget5_locked - obtain an inode from a mounted file system
1332 * @sb: super block of file system
1333 * @hashval: hash value (usually inode number) to get
1334 * @test: callback used for comparisons between inodes
1335 * @set: callback used to initialize a new struct inode
1336 * @data: opaque data pointer to pass to @test and @set
1337 *
1338 * Search for the inode specified by @hashval and @data in the inode cache,
1339 * and if present return it with an increased reference count. This is a
1340 * generalized version of iget_locked() for file systems where the inode
1341 * number is not sufficient for unique identification of an inode.
1342 *
1343 * If the inode is not present in the cache, allocate and insert a new inode
1344 * and return it locked, hashed, and with the I_NEW flag set. The file system
1345 * gets to fill it in before unlocking it via unlock_new_inode().
1346 *
1347 * Note that both @test and @set are called with the inode_hash_lock held, so
1348 * they can't sleep.
1349 */
1350struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1351 int (*test)(struct inode *, void *),
1352 int (*set)(struct inode *, void *), void *data)
1353{
1354 struct inode *inode = ilookup5(sb, hashval, test, data);
1355
1356 if (!inode) {
1357 struct inode *new = alloc_inode(sb);
1358
1359 if (new) {
1360 inode = inode_insert5(new, hashval, test, set, data);
1361 if (unlikely(inode != new))
1362 destroy_inode(inode: new);
1363 }
1364 }
1365 return inode;
1366}
1367EXPORT_SYMBOL(iget5_locked);
1368
1369/**
1370 * iget5_locked_rcu - obtain an inode from a mounted file system
1371 * @sb: super block of file system
1372 * @hashval: hash value (usually inode number) to get
1373 * @test: callback used for comparisons between inodes
1374 * @set: callback used to initialize a new struct inode
1375 * @data: opaque data pointer to pass to @test and @set
1376 *
1377 * This is equivalent to iget5_locked, except the @test callback must
1378 * tolerate the inode not being stable, including being mid-teardown.
1379 */
1380struct inode *iget5_locked_rcu(struct super_block *sb, unsigned long hashval,
1381 int (*test)(struct inode *, void *),
1382 int (*set)(struct inode *, void *), void *data)
1383{
1384 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1385 struct inode *inode, *new;
1386
1387 might_sleep();
1388
1389again:
1390 inode = find_inode(sb, head, test, data, is_inode_hash_locked: false);
1391 if (inode) {
1392 if (IS_ERR(ptr: inode))
1393 return NULL;
1394 wait_on_inode(inode);
1395 if (unlikely(inode_unhashed(inode))) {
1396 iput(inode);
1397 goto again;
1398 }
1399 return inode;
1400 }
1401
1402 new = alloc_inode(sb);
1403 if (new) {
1404 inode = inode_insert5(new, hashval, test, set, data);
1405 if (unlikely(inode != new))
1406 destroy_inode(inode: new);
1407 }
1408 return inode;
1409}
1410EXPORT_SYMBOL_GPL(iget5_locked_rcu);
1411
1412/**
1413 * iget_locked - obtain an inode from a mounted file system
1414 * @sb: super block of file system
1415 * @ino: inode number to get
1416 *
1417 * Search for the inode specified by @ino in the inode cache and if present
1418 * return it with an increased reference count. This is for file systems
1419 * where the inode number is sufficient for unique identification of an inode.
1420 *
1421 * If the inode is not in cache, allocate a new inode and return it locked,
1422 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1423 * before unlocking it via unlock_new_inode().
1424 */
1425struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1426{
1427 struct hlist_head *head = inode_hashtable + hash(sb, hashval: ino);
1428 struct inode *inode;
1429
1430 might_sleep();
1431
1432again:
1433 inode = find_inode_fast(sb, head, ino, is_inode_hash_locked: false);
1434 if (inode) {
1435 if (IS_ERR(ptr: inode))
1436 return NULL;
1437 wait_on_inode(inode);
1438 if (unlikely(inode_unhashed(inode))) {
1439 iput(inode);
1440 goto again;
1441 }
1442 return inode;
1443 }
1444
1445 inode = alloc_inode(sb);
1446 if (inode) {
1447 struct inode *old;
1448
1449 spin_lock(lock: &inode_hash_lock);
1450 /* We released the lock, so.. */
1451 old = find_inode_fast(sb, head, ino, is_inode_hash_locked: true);
1452 if (!old) {
1453 inode->i_ino = ino;
1454 spin_lock(lock: &inode->i_lock);
1455 inode->i_state = I_NEW;
1456 hlist_add_head_rcu(n: &inode->i_hash, h: head);
1457 spin_unlock(lock: &inode->i_lock);
1458 spin_unlock(lock: &inode_hash_lock);
1459 inode_sb_list_add(inode);
1460
1461 /* Return the locked inode with I_NEW set, the
1462 * caller is responsible for filling in the contents
1463 */
1464 return inode;
1465 }
1466
1467 /*
1468 * Uhhuh, somebody else created the same inode under
1469 * us. Use the old inode instead of the one we just
1470 * allocated.
1471 */
1472 spin_unlock(lock: &inode_hash_lock);
1473 destroy_inode(inode);
1474 if (IS_ERR(ptr: old))
1475 return NULL;
1476 inode = old;
1477 wait_on_inode(inode);
1478 if (unlikely(inode_unhashed(inode))) {
1479 iput(inode);
1480 goto again;
1481 }
1482 }
1483 return inode;
1484}
1485EXPORT_SYMBOL(iget_locked);
1486
1487/*
1488 * search the inode cache for a matching inode number.
1489 * If we find one, then the inode number we are trying to
1490 * allocate is not unique and so we should not use it.
1491 *
1492 * Returns 1 if the inode number is unique, 0 if it is not.
1493 */
1494static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1495{
1496 struct hlist_head *b = inode_hashtable + hash(sb, hashval: ino);
1497 struct inode *inode;
1498
1499 hlist_for_each_entry_rcu(inode, b, i_hash) {
1500 if (inode->i_ino == ino && inode->i_sb == sb)
1501 return 0;
1502 }
1503 return 1;
1504}
1505
1506/**
1507 * iunique - get a unique inode number
1508 * @sb: superblock
1509 * @max_reserved: highest reserved inode number
1510 *
1511 * Obtain an inode number that is unique on the system for a given
1512 * superblock. This is used by file systems that have no natural
1513 * permanent inode numbering system. An inode number is returned that
1514 * is higher than the reserved limit but unique.
1515 *
1516 * BUGS:
1517 * With a large number of inodes live on the file system this function
1518 * currently becomes quite slow.
1519 */
1520ino_t iunique(struct super_block *sb, ino_t max_reserved)
1521{
1522 /*
1523 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1524 * error if st_ino won't fit in target struct field. Use 32bit counter
1525 * here to attempt to avoid that.
1526 */
1527 static DEFINE_SPINLOCK(iunique_lock);
1528 static unsigned int counter;
1529 ino_t res;
1530
1531 rcu_read_lock();
1532 spin_lock(lock: &iunique_lock);
1533 do {
1534 if (counter <= max_reserved)
1535 counter = max_reserved + 1;
1536 res = counter++;
1537 } while (!test_inode_iunique(sb, ino: res));
1538 spin_unlock(lock: &iunique_lock);
1539 rcu_read_unlock();
1540
1541 return res;
1542}
1543EXPORT_SYMBOL(iunique);
1544
1545struct inode *igrab(struct inode *inode)
1546{
1547 spin_lock(lock: &inode->i_lock);
1548 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1549 __iget(inode);
1550 spin_unlock(lock: &inode->i_lock);
1551 } else {
1552 spin_unlock(lock: &inode->i_lock);
1553 /*
1554 * Handle the case where s_op->clear_inode is not been
1555 * called yet, and somebody is calling igrab
1556 * while the inode is getting freed.
1557 */
1558 inode = NULL;
1559 }
1560 return inode;
1561}
1562EXPORT_SYMBOL(igrab);
1563
1564/**
1565 * ilookup5_nowait - search for an inode in the inode cache
1566 * @sb: super block of file system to search
1567 * @hashval: hash value (usually inode number) to search for
1568 * @test: callback used for comparisons between inodes
1569 * @data: opaque data pointer to pass to @test
1570 *
1571 * Search for the inode specified by @hashval and @data in the inode cache.
1572 * If the inode is in the cache, the inode is returned with an incremented
1573 * reference count.
1574 *
1575 * Note: I_NEW is not waited upon so you have to be very careful what you do
1576 * with the returned inode. You probably should be using ilookup5() instead.
1577 *
1578 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1579 */
1580struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1581 int (*test)(struct inode *, void *), void *data)
1582{
1583 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1584 struct inode *inode;
1585
1586 spin_lock(lock: &inode_hash_lock);
1587 inode = find_inode(sb, head, test, data, is_inode_hash_locked: true);
1588 spin_unlock(lock: &inode_hash_lock);
1589
1590 return IS_ERR(ptr: inode) ? NULL : inode;
1591}
1592EXPORT_SYMBOL(ilookup5_nowait);
1593
1594/**
1595 * ilookup5 - search for an inode in the inode cache
1596 * @sb: super block of file system to search
1597 * @hashval: hash value (usually inode number) to search for
1598 * @test: callback used for comparisons between inodes
1599 * @data: opaque data pointer to pass to @test
1600 *
1601 * Search for the inode specified by @hashval and @data in the inode cache,
1602 * and if the inode is in the cache, return the inode with an incremented
1603 * reference count. Waits on I_NEW before returning the inode.
1604 * returned with an incremented reference count.
1605 *
1606 * This is a generalized version of ilookup() for file systems where the
1607 * inode number is not sufficient for unique identification of an inode.
1608 *
1609 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1610 */
1611struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1612 int (*test)(struct inode *, void *), void *data)
1613{
1614 struct inode *inode;
1615
1616 might_sleep();
1617
1618again:
1619 inode = ilookup5_nowait(sb, hashval, test, data);
1620 if (inode) {
1621 wait_on_inode(inode);
1622 if (unlikely(inode_unhashed(inode))) {
1623 iput(inode);
1624 goto again;
1625 }
1626 }
1627 return inode;
1628}
1629EXPORT_SYMBOL(ilookup5);
1630
1631/**
1632 * ilookup - search for an inode in the inode cache
1633 * @sb: super block of file system to search
1634 * @ino: inode number to search for
1635 *
1636 * Search for the inode @ino in the inode cache, and if the inode is in the
1637 * cache, the inode is returned with an incremented reference count.
1638 */
1639struct inode *ilookup(struct super_block *sb, unsigned long ino)
1640{
1641 struct hlist_head *head = inode_hashtable + hash(sb, hashval: ino);
1642 struct inode *inode;
1643
1644 might_sleep();
1645
1646again:
1647 inode = find_inode_fast(sb, head, ino, is_inode_hash_locked: false);
1648
1649 if (inode) {
1650 if (IS_ERR(ptr: inode))
1651 return NULL;
1652 wait_on_inode(inode);
1653 if (unlikely(inode_unhashed(inode))) {
1654 iput(inode);
1655 goto again;
1656 }
1657 }
1658 return inode;
1659}
1660EXPORT_SYMBOL(ilookup);
1661
1662/**
1663 * find_inode_nowait - find an inode in the inode cache
1664 * @sb: super block of file system to search
1665 * @hashval: hash value (usually inode number) to search for
1666 * @match: callback used for comparisons between inodes
1667 * @data: opaque data pointer to pass to @match
1668 *
1669 * Search for the inode specified by @hashval and @data in the inode
1670 * cache, where the helper function @match will return 0 if the inode
1671 * does not match, 1 if the inode does match, and -1 if the search
1672 * should be stopped. The @match function must be responsible for
1673 * taking the i_lock spin_lock and checking i_state for an inode being
1674 * freed or being initialized, and incrementing the reference count
1675 * before returning 1. It also must not sleep, since it is called with
1676 * the inode_hash_lock spinlock held.
1677 *
1678 * This is a even more generalized version of ilookup5() when the
1679 * function must never block --- find_inode() can block in
1680 * __wait_on_freeing_inode() --- or when the caller can not increment
1681 * the reference count because the resulting iput() might cause an
1682 * inode eviction. The tradeoff is that the @match funtion must be
1683 * very carefully implemented.
1684 */
1685struct inode *find_inode_nowait(struct super_block *sb,
1686 unsigned long hashval,
1687 int (*match)(struct inode *, unsigned long,
1688 void *),
1689 void *data)
1690{
1691 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1692 struct inode *inode, *ret_inode = NULL;
1693 int mval;
1694
1695 spin_lock(lock: &inode_hash_lock);
1696 hlist_for_each_entry(inode, head, i_hash) {
1697 if (inode->i_sb != sb)
1698 continue;
1699 mval = match(inode, hashval, data);
1700 if (mval == 0)
1701 continue;
1702 if (mval == 1)
1703 ret_inode = inode;
1704 goto out;
1705 }
1706out:
1707 spin_unlock(lock: &inode_hash_lock);
1708 return ret_inode;
1709}
1710EXPORT_SYMBOL(find_inode_nowait);
1711
1712/**
1713 * find_inode_rcu - find an inode in the inode cache
1714 * @sb: Super block of file system to search
1715 * @hashval: Key to hash
1716 * @test: Function to test match on an inode
1717 * @data: Data for test function
1718 *
1719 * Search for the inode specified by @hashval and @data in the inode cache,
1720 * where the helper function @test will return 0 if the inode does not match
1721 * and 1 if it does. The @test function must be responsible for taking the
1722 * i_lock spin_lock and checking i_state for an inode being freed or being
1723 * initialized.
1724 *
1725 * If successful, this will return the inode for which the @test function
1726 * returned 1 and NULL otherwise.
1727 *
1728 * The @test function is not permitted to take a ref on any inode presented.
1729 * It is also not permitted to sleep.
1730 *
1731 * The caller must hold the RCU read lock.
1732 */
1733struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1734 int (*test)(struct inode *, void *), void *data)
1735{
1736 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1737 struct inode *inode;
1738
1739 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1740 "suspicious find_inode_rcu() usage");
1741
1742 hlist_for_each_entry_rcu(inode, head, i_hash) {
1743 if (inode->i_sb == sb &&
1744 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1745 test(inode, data))
1746 return inode;
1747 }
1748 return NULL;
1749}
1750EXPORT_SYMBOL(find_inode_rcu);
1751
1752/**
1753 * find_inode_by_ino_rcu - Find an inode in the inode cache
1754 * @sb: Super block of file system to search
1755 * @ino: The inode number to match
1756 *
1757 * Search for the inode specified by @hashval and @data in the inode cache,
1758 * where the helper function @test will return 0 if the inode does not match
1759 * and 1 if it does. The @test function must be responsible for taking the
1760 * i_lock spin_lock and checking i_state for an inode being freed or being
1761 * initialized.
1762 *
1763 * If successful, this will return the inode for which the @test function
1764 * returned 1 and NULL otherwise.
1765 *
1766 * The @test function is not permitted to take a ref on any inode presented.
1767 * It is also not permitted to sleep.
1768 *
1769 * The caller must hold the RCU read lock.
1770 */
1771struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1772 unsigned long ino)
1773{
1774 struct hlist_head *head = inode_hashtable + hash(sb, hashval: ino);
1775 struct inode *inode;
1776
1777 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1778 "suspicious find_inode_by_ino_rcu() usage");
1779
1780 hlist_for_each_entry_rcu(inode, head, i_hash) {
1781 if (inode->i_ino == ino &&
1782 inode->i_sb == sb &&
1783 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1784 return inode;
1785 }
1786 return NULL;
1787}
1788EXPORT_SYMBOL(find_inode_by_ino_rcu);
1789
1790int insert_inode_locked(struct inode *inode)
1791{
1792 struct super_block *sb = inode->i_sb;
1793 ino_t ino = inode->i_ino;
1794 struct hlist_head *head = inode_hashtable + hash(sb, hashval: ino);
1795
1796 might_sleep();
1797
1798 while (1) {
1799 struct inode *old = NULL;
1800 spin_lock(lock: &inode_hash_lock);
1801 hlist_for_each_entry(old, head, i_hash) {
1802 if (old->i_ino != ino)
1803 continue;
1804 if (old->i_sb != sb)
1805 continue;
1806 spin_lock(lock: &old->i_lock);
1807 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1808 spin_unlock(lock: &old->i_lock);
1809 continue;
1810 }
1811 break;
1812 }
1813 if (likely(!old)) {
1814 spin_lock(lock: &inode->i_lock);
1815 inode->i_state |= I_NEW | I_CREATING;
1816 hlist_add_head_rcu(n: &inode->i_hash, h: head);
1817 spin_unlock(lock: &inode->i_lock);
1818 spin_unlock(lock: &inode_hash_lock);
1819 return 0;
1820 }
1821 if (unlikely(old->i_state & I_CREATING)) {
1822 spin_unlock(lock: &old->i_lock);
1823 spin_unlock(lock: &inode_hash_lock);
1824 return -EBUSY;
1825 }
1826 __iget(inode: old);
1827 spin_unlock(lock: &old->i_lock);
1828 spin_unlock(lock: &inode_hash_lock);
1829 wait_on_inode(inode: old);
1830 if (unlikely(!inode_unhashed(old))) {
1831 iput(old);
1832 return -EBUSY;
1833 }
1834 iput(old);
1835 }
1836}
1837EXPORT_SYMBOL(insert_inode_locked);
1838
1839int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1840 int (*test)(struct inode *, void *), void *data)
1841{
1842 struct inode *old;
1843
1844 might_sleep();
1845
1846 inode->i_state |= I_CREATING;
1847 old = inode_insert5(inode, hashval, test, NULL, data);
1848
1849 if (old != inode) {
1850 iput(old);
1851 return -EBUSY;
1852 }
1853 return 0;
1854}
1855EXPORT_SYMBOL(insert_inode_locked4);
1856
1857
1858int inode_just_drop(struct inode *inode)
1859{
1860 return 1;
1861}
1862EXPORT_SYMBOL(inode_just_drop);
1863
1864/*
1865 * Called when we're dropping the last reference
1866 * to an inode.
1867 *
1868 * Call the FS "drop_inode()" function, defaulting to
1869 * the legacy UNIX filesystem behaviour. If it tells
1870 * us to evict inode, do so. Otherwise, retain inode
1871 * in cache if fs is alive, sync and evict if fs is
1872 * shutting down.
1873 */
1874static void iput_final(struct inode *inode)
1875{
1876 struct super_block *sb = inode->i_sb;
1877 const struct super_operations *op = inode->i_sb->s_op;
1878 unsigned long state;
1879 int drop;
1880
1881 WARN_ON(inode->i_state & I_NEW);
1882
1883 if (op->drop_inode)
1884 drop = op->drop_inode(inode);
1885 else
1886 drop = inode_generic_drop(inode);
1887
1888 if (!drop &&
1889 !(inode->i_state & I_DONTCACHE) &&
1890 (sb->s_flags & SB_ACTIVE)) {
1891 __inode_add_lru(inode, rotate: true);
1892 spin_unlock(lock: &inode->i_lock);
1893 return;
1894 }
1895
1896 state = inode->i_state;
1897 if (!drop) {
1898 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1899 spin_unlock(lock: &inode->i_lock);
1900
1901 write_inode_now(inode, sync: 1);
1902
1903 spin_lock(lock: &inode->i_lock);
1904 state = inode->i_state;
1905 WARN_ON(state & I_NEW);
1906 state &= ~I_WILL_FREE;
1907 }
1908
1909 WRITE_ONCE(inode->i_state, state | I_FREEING);
1910 if (!list_empty(head: &inode->i_lru))
1911 inode_lru_list_del(inode);
1912 spin_unlock(lock: &inode->i_lock);
1913
1914 evict(inode);
1915}
1916
1917/**
1918 * iput - put an inode
1919 * @inode: inode to put
1920 *
1921 * Puts an inode, dropping its usage count. If the inode use count hits
1922 * zero, the inode is then freed and may also be destroyed.
1923 *
1924 * Consequently, iput() can sleep.
1925 */
1926void iput(struct inode *inode)
1927{
1928 might_sleep();
1929 if (unlikely(!inode))
1930 return;
1931
1932retry:
1933 lockdep_assert_not_held(&inode->i_lock);
1934 VFS_BUG_ON_INODE(inode->i_state & I_CLEAR, inode);
1935 /*
1936 * Note this assert is technically racy as if the count is bogusly
1937 * equal to one, then two CPUs racing to further drop it can both
1938 * conclude it's fine.
1939 */
1940 VFS_BUG_ON_INODE(atomic_read(&inode->i_count) < 1, inode);
1941
1942 if (atomic_add_unless(v: &inode->i_count, a: -1, u: 1))
1943 return;
1944
1945 if ((inode->i_state & I_DIRTY_TIME) && inode->i_nlink) {
1946 trace_writeback_lazytime_iput(inode);
1947 mark_inode_dirty_sync(inode);
1948 goto retry;
1949 }
1950
1951 spin_lock(lock: &inode->i_lock);
1952 if (unlikely((inode->i_state & I_DIRTY_TIME) && inode->i_nlink)) {
1953 spin_unlock(lock: &inode->i_lock);
1954 goto retry;
1955 }
1956
1957 if (!atomic_dec_and_test(v: &inode->i_count)) {
1958 spin_unlock(lock: &inode->i_lock);
1959 return;
1960 }
1961
1962 /*
1963 * iput_final() drops ->i_lock, we can't assert on it as the inode may
1964 * be deallocated by the time the call returns.
1965 */
1966 iput_final(inode);
1967}
1968EXPORT_SYMBOL(iput);
1969
1970#ifdef CONFIG_BLOCK
1971/**
1972 * bmap - find a block number in a file
1973 * @inode: inode owning the block number being requested
1974 * @block: pointer containing the block to find
1975 *
1976 * Replaces the value in ``*block`` with the block number on the device holding
1977 * corresponding to the requested block number in the file.
1978 * That is, asked for block 4 of inode 1 the function will replace the
1979 * 4 in ``*block``, with disk block relative to the disk start that holds that
1980 * block of the file.
1981 *
1982 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1983 * hole, returns 0 and ``*block`` is also set to 0.
1984 */
1985int bmap(struct inode *inode, sector_t *block)
1986{
1987 if (!inode->i_mapping->a_ops->bmap)
1988 return -EINVAL;
1989
1990 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1991 return 0;
1992}
1993EXPORT_SYMBOL(bmap);
1994#endif
1995
1996/*
1997 * With relative atime, only update atime if the previous atime is
1998 * earlier than or equal to either the ctime or mtime,
1999 * or if at least a day has passed since the last atime update.
2000 */
2001static bool relatime_need_update(struct vfsmount *mnt, struct inode *inode,
2002 struct timespec64 now)
2003{
2004 struct timespec64 atime, mtime, ctime;
2005
2006 if (!(mnt->mnt_flags & MNT_RELATIME))
2007 return true;
2008 /*
2009 * Is mtime younger than or equal to atime? If yes, update atime:
2010 */
2011 atime = inode_get_atime(inode);
2012 mtime = inode_get_mtime(inode);
2013 if (timespec64_compare(lhs: &mtime, rhs: &atime) >= 0)
2014 return true;
2015 /*
2016 * Is ctime younger than or equal to atime? If yes, update atime:
2017 */
2018 ctime = inode_get_ctime(inode);
2019 if (timespec64_compare(lhs: &ctime, rhs: &atime) >= 0)
2020 return true;
2021
2022 /*
2023 * Is the previous atime value older than a day? If yes,
2024 * update atime:
2025 */
2026 if ((long)(now.tv_sec - atime.tv_sec) >= 24*60*60)
2027 return true;
2028 /*
2029 * Good, we can skip the atime update:
2030 */
2031 return false;
2032}
2033
2034/**
2035 * inode_update_timestamps - update the timestamps on the inode
2036 * @inode: inode to be updated
2037 * @flags: S_* flags that needed to be updated
2038 *
2039 * The update_time function is called when an inode's timestamps need to be
2040 * updated for a read or write operation. This function handles updating the
2041 * actual timestamps. It's up to the caller to ensure that the inode is marked
2042 * dirty appropriately.
2043 *
2044 * In the case where any of S_MTIME, S_CTIME, or S_VERSION need to be updated,
2045 * attempt to update all three of them. S_ATIME updates can be handled
2046 * independently of the rest.
2047 *
2048 * Returns a set of S_* flags indicating which values changed.
2049 */
2050int inode_update_timestamps(struct inode *inode, int flags)
2051{
2052 int updated = 0;
2053 struct timespec64 now;
2054
2055 if (flags & (S_MTIME|S_CTIME|S_VERSION)) {
2056 struct timespec64 ctime = inode_get_ctime(inode);
2057 struct timespec64 mtime = inode_get_mtime(inode);
2058
2059 now = inode_set_ctime_current(inode);
2060 if (!timespec64_equal(a: &now, b: &ctime))
2061 updated |= S_CTIME;
2062 if (!timespec64_equal(a: &now, b: &mtime)) {
2063 inode_set_mtime_to_ts(inode, ts: now);
2064 updated |= S_MTIME;
2065 }
2066 if (IS_I_VERSION(inode) && inode_maybe_inc_iversion(inode, force: updated))
2067 updated |= S_VERSION;
2068 } else {
2069 now = current_time(inode);
2070 }
2071
2072 if (flags & S_ATIME) {
2073 struct timespec64 atime = inode_get_atime(inode);
2074
2075 if (!timespec64_equal(a: &now, b: &atime)) {
2076 inode_set_atime_to_ts(inode, ts: now);
2077 updated |= S_ATIME;
2078 }
2079 }
2080 return updated;
2081}
2082EXPORT_SYMBOL(inode_update_timestamps);
2083
2084/**
2085 * generic_update_time - update the timestamps on the inode
2086 * @inode: inode to be updated
2087 * @flags: S_* flags that needed to be updated
2088 *
2089 * The update_time function is called when an inode's timestamps need to be
2090 * updated for a read or write operation. In the case where any of S_MTIME, S_CTIME,
2091 * or S_VERSION need to be updated we attempt to update all three of them. S_ATIME
2092 * updates can be handled done independently of the rest.
2093 *
2094 * Returns a S_* mask indicating which fields were updated.
2095 */
2096int generic_update_time(struct inode *inode, int flags)
2097{
2098 int updated = inode_update_timestamps(inode, flags);
2099 int dirty_flags = 0;
2100
2101 if (updated & (S_ATIME|S_MTIME|S_CTIME))
2102 dirty_flags = inode->i_sb->s_flags & SB_LAZYTIME ? I_DIRTY_TIME : I_DIRTY_SYNC;
2103 if (updated & S_VERSION)
2104 dirty_flags |= I_DIRTY_SYNC;
2105 __mark_inode_dirty(inode, dirty_flags);
2106 return updated;
2107}
2108EXPORT_SYMBOL(generic_update_time);
2109
2110/*
2111 * This does the actual work of updating an inodes time or version. Must have
2112 * had called mnt_want_write() before calling this.
2113 */
2114int inode_update_time(struct inode *inode, int flags)
2115{
2116 if (inode->i_op->update_time)
2117 return inode->i_op->update_time(inode, flags);
2118 generic_update_time(inode, flags);
2119 return 0;
2120}
2121EXPORT_SYMBOL(inode_update_time);
2122
2123/**
2124 * atime_needs_update - update the access time
2125 * @path: the &struct path to update
2126 * @inode: inode to update
2127 *
2128 * Update the accessed time on an inode and mark it for writeback.
2129 * This function automatically handles read only file systems and media,
2130 * as well as the "noatime" flag and inode specific "noatime" markers.
2131 */
2132bool atime_needs_update(const struct path *path, struct inode *inode)
2133{
2134 struct vfsmount *mnt = path->mnt;
2135 struct timespec64 now, atime;
2136
2137 if (inode->i_flags & S_NOATIME)
2138 return false;
2139
2140 /* Atime updates will likely cause i_uid and i_gid to be written
2141 * back improprely if their true value is unknown to the vfs.
2142 */
2143 if (HAS_UNMAPPED_ID(idmap: mnt_idmap(mnt), inode))
2144 return false;
2145
2146 if (IS_NOATIME(inode))
2147 return false;
2148 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
2149 return false;
2150
2151 if (mnt->mnt_flags & MNT_NOATIME)
2152 return false;
2153 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
2154 return false;
2155
2156 now = current_time(inode);
2157
2158 if (!relatime_need_update(mnt, inode, now))
2159 return false;
2160
2161 atime = inode_get_atime(inode);
2162 if (timespec64_equal(a: &atime, b: &now))
2163 return false;
2164
2165 return true;
2166}
2167
2168void touch_atime(const struct path *path)
2169{
2170 struct vfsmount *mnt = path->mnt;
2171 struct inode *inode = d_inode(dentry: path->dentry);
2172
2173 if (!atime_needs_update(path, inode))
2174 return;
2175
2176 if (!sb_start_write_trylock(sb: inode->i_sb))
2177 return;
2178
2179 if (mnt_get_write_access(mnt) != 0)
2180 goto skip_update;
2181 /*
2182 * File systems can error out when updating inodes if they need to
2183 * allocate new space to modify an inode (such is the case for
2184 * Btrfs), but since we touch atime while walking down the path we
2185 * really don't care if we failed to update the atime of the file,
2186 * so just ignore the return value.
2187 * We may also fail on filesystems that have the ability to make parts
2188 * of the fs read only, e.g. subvolumes in Btrfs.
2189 */
2190 inode_update_time(inode, S_ATIME);
2191 mnt_put_write_access(mnt);
2192skip_update:
2193 sb_end_write(sb: inode->i_sb);
2194}
2195EXPORT_SYMBOL(touch_atime);
2196
2197/*
2198 * Return mask of changes for notify_change() that need to be done as a
2199 * response to write or truncate. Return 0 if nothing has to be changed.
2200 * Negative value on error (change should be denied).
2201 */
2202int dentry_needs_remove_privs(struct mnt_idmap *idmap,
2203 struct dentry *dentry)
2204{
2205 struct inode *inode = d_inode(dentry);
2206 int mask = 0;
2207 int ret;
2208
2209 if (IS_NOSEC(inode))
2210 return 0;
2211
2212 mask = setattr_should_drop_suidgid(idmap, inode);
2213 ret = security_inode_need_killpriv(dentry);
2214 if (ret < 0)
2215 return ret;
2216 if (ret)
2217 mask |= ATTR_KILL_PRIV;
2218 return mask;
2219}
2220
2221static int __remove_privs(struct mnt_idmap *idmap,
2222 struct dentry *dentry, int kill)
2223{
2224 struct iattr newattrs;
2225
2226 newattrs.ia_valid = ATTR_FORCE | kill;
2227 /*
2228 * Note we call this on write, so notify_change will not
2229 * encounter any conflicting delegations:
2230 */
2231 return notify_change(idmap, dentry, &newattrs, NULL);
2232}
2233
2234static int file_remove_privs_flags(struct file *file, unsigned int flags)
2235{
2236 struct dentry *dentry = file_dentry(file);
2237 struct inode *inode = file_inode(f: file);
2238 int error = 0;
2239 int kill;
2240
2241 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
2242 return 0;
2243
2244 kill = dentry_needs_remove_privs(idmap: file_mnt_idmap(file), dentry);
2245 if (kill < 0)
2246 return kill;
2247
2248 if (kill) {
2249 if (flags & IOCB_NOWAIT)
2250 return -EAGAIN;
2251
2252 error = __remove_privs(idmap: file_mnt_idmap(file), dentry, kill);
2253 }
2254
2255 if (!error)
2256 inode_has_no_xattr(inode);
2257 return error;
2258}
2259
2260/**
2261 * file_remove_privs - remove special file privileges (suid, capabilities)
2262 * @file: file to remove privileges from
2263 *
2264 * When file is modified by a write or truncation ensure that special
2265 * file privileges are removed.
2266 *
2267 * Return: 0 on success, negative errno on failure.
2268 */
2269int file_remove_privs(struct file *file)
2270{
2271 return file_remove_privs_flags(file, flags: 0);
2272}
2273EXPORT_SYMBOL(file_remove_privs);
2274
2275/**
2276 * current_time - Return FS time (possibly fine-grained)
2277 * @inode: inode.
2278 *
2279 * Return the current time truncated to the time granularity supported by
2280 * the fs, as suitable for a ctime/mtime change. If the ctime is flagged
2281 * as having been QUERIED, get a fine-grained timestamp, but don't update
2282 * the floor.
2283 *
2284 * For a multigrain inode, this is effectively an estimate of the timestamp
2285 * that a file would receive. An actual update must go through
2286 * inode_set_ctime_current().
2287 */
2288struct timespec64 current_time(struct inode *inode)
2289{
2290 struct timespec64 now;
2291 u32 cns;
2292
2293 ktime_get_coarse_real_ts64_mg(ts: &now);
2294
2295 if (!is_mgtime(inode))
2296 goto out;
2297
2298 /* If nothing has queried it, then coarse time is fine */
2299 cns = smp_load_acquire(&inode->i_ctime_nsec);
2300 if (cns & I_CTIME_QUERIED) {
2301 /*
2302 * If there is no apparent change, then get a fine-grained
2303 * timestamp.
2304 */
2305 if (now.tv_nsec == (cns & ~I_CTIME_QUERIED))
2306 ktime_get_real_ts64(tv: &now);
2307 }
2308out:
2309 return timestamp_truncate(t: now, inode);
2310}
2311EXPORT_SYMBOL(current_time);
2312
2313static int inode_needs_update_time(struct inode *inode)
2314{
2315 struct timespec64 now, ts;
2316 int sync_it = 0;
2317
2318 /* First try to exhaust all avenues to not sync */
2319 if (IS_NOCMTIME(inode))
2320 return 0;
2321
2322 now = current_time(inode);
2323
2324 ts = inode_get_mtime(inode);
2325 if (!timespec64_equal(a: &ts, b: &now))
2326 sync_it |= S_MTIME;
2327
2328 ts = inode_get_ctime(inode);
2329 if (!timespec64_equal(a: &ts, b: &now))
2330 sync_it |= S_CTIME;
2331
2332 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2333 sync_it |= S_VERSION;
2334
2335 return sync_it;
2336}
2337
2338static int __file_update_time(struct file *file, int sync_mode)
2339{
2340 int ret = 0;
2341 struct inode *inode = file_inode(f: file);
2342
2343 /* try to update time settings */
2344 if (!mnt_get_write_access_file(file)) {
2345 ret = inode_update_time(inode, sync_mode);
2346 mnt_put_write_access_file(file);
2347 }
2348
2349 return ret;
2350}
2351
2352/**
2353 * file_update_time - update mtime and ctime time
2354 * @file: file accessed
2355 *
2356 * Update the mtime and ctime members of an inode and mark the inode for
2357 * writeback. Note that this function is meant exclusively for usage in
2358 * the file write path of filesystems, and filesystems may choose to
2359 * explicitly ignore updates via this function with the _NOCMTIME inode
2360 * flag, e.g. for network filesystem where these imestamps are handled
2361 * by the server. This can return an error for file systems who need to
2362 * allocate space in order to update an inode.
2363 *
2364 * Return: 0 on success, negative errno on failure.
2365 */
2366int file_update_time(struct file *file)
2367{
2368 int ret;
2369 struct inode *inode = file_inode(f: file);
2370
2371 ret = inode_needs_update_time(inode);
2372 if (ret <= 0)
2373 return ret;
2374
2375 return __file_update_time(file, sync_mode: ret);
2376}
2377EXPORT_SYMBOL(file_update_time);
2378
2379/**
2380 * file_modified_flags - handle mandated vfs changes when modifying a file
2381 * @file: file that was modified
2382 * @flags: kiocb flags
2383 *
2384 * When file has been modified ensure that special
2385 * file privileges are removed and time settings are updated.
2386 *
2387 * If IOCB_NOWAIT is set, special file privileges will not be removed and
2388 * time settings will not be updated. It will return -EAGAIN.
2389 *
2390 * Context: Caller must hold the file's inode lock.
2391 *
2392 * Return: 0 on success, negative errno on failure.
2393 */
2394static int file_modified_flags(struct file *file, int flags)
2395{
2396 int ret;
2397 struct inode *inode = file_inode(f: file);
2398
2399 /*
2400 * Clear the security bits if the process is not being run by root.
2401 * This keeps people from modifying setuid and setgid binaries.
2402 */
2403 ret = file_remove_privs_flags(file, flags);
2404 if (ret)
2405 return ret;
2406
2407 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2408 return 0;
2409
2410 ret = inode_needs_update_time(inode);
2411 if (ret <= 0)
2412 return ret;
2413 if (flags & IOCB_NOWAIT)
2414 return -EAGAIN;
2415
2416 return __file_update_time(file, sync_mode: ret);
2417}
2418
2419/**
2420 * file_modified - handle mandated vfs changes when modifying a file
2421 * @file: file that was modified
2422 *
2423 * When file has been modified ensure that special
2424 * file privileges are removed and time settings are updated.
2425 *
2426 * Context: Caller must hold the file's inode lock.
2427 *
2428 * Return: 0 on success, negative errno on failure.
2429 */
2430int file_modified(struct file *file)
2431{
2432 return file_modified_flags(file, flags: 0);
2433}
2434EXPORT_SYMBOL(file_modified);
2435
2436/**
2437 * kiocb_modified - handle mandated vfs changes when modifying a file
2438 * @iocb: iocb that was modified
2439 *
2440 * When file has been modified ensure that special
2441 * file privileges are removed and time settings are updated.
2442 *
2443 * Context: Caller must hold the file's inode lock.
2444 *
2445 * Return: 0 on success, negative errno on failure.
2446 */
2447int kiocb_modified(struct kiocb *iocb)
2448{
2449 return file_modified_flags(file: iocb->ki_filp, flags: iocb->ki_flags);
2450}
2451EXPORT_SYMBOL_GPL(kiocb_modified);
2452
2453int inode_needs_sync(struct inode *inode)
2454{
2455 if (IS_SYNC(inode))
2456 return 1;
2457 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2458 return 1;
2459 return 0;
2460}
2461EXPORT_SYMBOL(inode_needs_sync);
2462
2463/*
2464 * If we try to find an inode in the inode hash while it is being
2465 * deleted, we have to wait until the filesystem completes its
2466 * deletion before reporting that it isn't found. This function waits
2467 * until the deletion _might_ have completed. Callers are responsible
2468 * to recheck inode state.
2469 *
2470 * It doesn't matter if I_NEW is not set initially, a call to
2471 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2472 * will DTRT.
2473 */
2474static void __wait_on_freeing_inode(struct inode *inode, bool is_inode_hash_locked)
2475{
2476 struct wait_bit_queue_entry wqe;
2477 struct wait_queue_head *wq_head;
2478
2479 /*
2480 * Handle racing against evict(), see that routine for more details.
2481 */
2482 if (unlikely(inode_unhashed(inode))) {
2483 WARN_ON(is_inode_hash_locked);
2484 spin_unlock(lock: &inode->i_lock);
2485 return;
2486 }
2487
2488 wq_head = inode_bit_waitqueue(&wqe, inode, __I_NEW);
2489 prepare_to_wait_event(wq_head, wq_entry: &wqe.wq_entry, TASK_UNINTERRUPTIBLE);
2490 spin_unlock(lock: &inode->i_lock);
2491 rcu_read_unlock();
2492 if (is_inode_hash_locked)
2493 spin_unlock(lock: &inode_hash_lock);
2494 schedule();
2495 finish_wait(wq_head, wq_entry: &wqe.wq_entry);
2496 if (is_inode_hash_locked)
2497 spin_lock(lock: &inode_hash_lock);
2498 rcu_read_lock();
2499}
2500
2501static __initdata unsigned long ihash_entries;
2502static int __init set_ihash_entries(char *str)
2503{
2504 if (!str)
2505 return 0;
2506 ihash_entries = simple_strtoul(str, &str, 0);
2507 return 1;
2508}
2509__setup("ihash_entries=", set_ihash_entries);
2510
2511/*
2512 * Initialize the waitqueues and inode hash table.
2513 */
2514void __init inode_init_early(void)
2515{
2516 /* If hashes are distributed across NUMA nodes, defer
2517 * hash allocation until vmalloc space is available.
2518 */
2519 if (hashdist)
2520 return;
2521
2522 inode_hashtable =
2523 alloc_large_system_hash(tablename: "Inode-cache",
2524 bucketsize: sizeof(struct hlist_head),
2525 numentries: ihash_entries,
2526 scale: 14,
2527 HASH_EARLY | HASH_ZERO,
2528 hash_shift: &i_hash_shift,
2529 hash_mask: &i_hash_mask,
2530 low_limit: 0,
2531 high_limit: 0);
2532}
2533
2534void __init inode_init(void)
2535{
2536 /* inode slab cache */
2537 inode_cachep = kmem_cache_create("inode_cache",
2538 sizeof(struct inode),
2539 0,
2540 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2541 SLAB_ACCOUNT),
2542 init_once);
2543
2544 /* Hash may have been set up in inode_init_early */
2545 if (!hashdist)
2546 return;
2547
2548 inode_hashtable =
2549 alloc_large_system_hash(tablename: "Inode-cache",
2550 bucketsize: sizeof(struct hlist_head),
2551 numentries: ihash_entries,
2552 scale: 14,
2553 HASH_ZERO,
2554 hash_shift: &i_hash_shift,
2555 hash_mask: &i_hash_mask,
2556 low_limit: 0,
2557 high_limit: 0);
2558}
2559
2560void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2561{
2562 inode->i_mode = mode;
2563 switch (inode->i_mode & S_IFMT) {
2564 case S_IFCHR:
2565 inode->i_fop = &def_chr_fops;
2566 inode->i_rdev = rdev;
2567 break;
2568 case S_IFBLK:
2569 if (IS_ENABLED(CONFIG_BLOCK))
2570 inode->i_fop = &def_blk_fops;
2571 inode->i_rdev = rdev;
2572 break;
2573 case S_IFIFO:
2574 inode->i_fop = &pipefifo_fops;
2575 break;
2576 case S_IFSOCK:
2577 /* leave it no_open_fops */
2578 break;
2579 default:
2580 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2581 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2582 inode->i_ino);
2583 break;
2584 }
2585}
2586EXPORT_SYMBOL(init_special_inode);
2587
2588/**
2589 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2590 * @idmap: idmap of the mount the inode was created from
2591 * @inode: New inode
2592 * @dir: Directory inode
2593 * @mode: mode of the new inode
2594 *
2595 * If the inode has been created through an idmapped mount the idmap of
2596 * the vfsmount must be passed through @idmap. This function will then take
2597 * care to map the inode according to @idmap before checking permissions
2598 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2599 * checking is to be performed on the raw inode simply pass @nop_mnt_idmap.
2600 */
2601void inode_init_owner(struct mnt_idmap *idmap, struct inode *inode,
2602 const struct inode *dir, umode_t mode)
2603{
2604 inode_fsuid_set(inode, idmap);
2605 if (dir && dir->i_mode & S_ISGID) {
2606 inode->i_gid = dir->i_gid;
2607
2608 /* Directories are special, and always inherit S_ISGID */
2609 if (S_ISDIR(mode))
2610 mode |= S_ISGID;
2611 } else
2612 inode_fsgid_set(inode, idmap);
2613 inode->i_mode = mode;
2614}
2615EXPORT_SYMBOL(inode_init_owner);
2616
2617/**
2618 * inode_owner_or_capable - check current task permissions to inode
2619 * @idmap: idmap of the mount the inode was found from
2620 * @inode: inode being checked
2621 *
2622 * Return true if current either has CAP_FOWNER in a namespace with the
2623 * inode owner uid mapped, or owns the file.
2624 *
2625 * If the inode has been found through an idmapped mount the idmap of
2626 * the vfsmount must be passed through @idmap. This function will then take
2627 * care to map the inode according to @idmap before checking permissions.
2628 * On non-idmapped mounts or if permission checking is to be performed on the
2629 * raw inode simply pass @nop_mnt_idmap.
2630 */
2631bool inode_owner_or_capable(struct mnt_idmap *idmap,
2632 const struct inode *inode)
2633{
2634 vfsuid_t vfsuid;
2635 struct user_namespace *ns;
2636
2637 vfsuid = i_uid_into_vfsuid(idmap, inode);
2638 if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
2639 return true;
2640
2641 ns = current_user_ns();
2642 if (vfsuid_has_mapping(userns: ns, vfsuid) && ns_capable(ns, CAP_FOWNER))
2643 return true;
2644 return false;
2645}
2646EXPORT_SYMBOL(inode_owner_or_capable);
2647
2648/*
2649 * Direct i/o helper functions
2650 */
2651bool inode_dio_finished(const struct inode *inode)
2652{
2653 return atomic_read(v: &inode->i_dio_count) == 0;
2654}
2655EXPORT_SYMBOL(inode_dio_finished);
2656
2657/**
2658 * inode_dio_wait - wait for outstanding DIO requests to finish
2659 * @inode: inode to wait for
2660 *
2661 * Waits for all pending direct I/O requests to finish so that we can
2662 * proceed with a truncate or equivalent operation.
2663 *
2664 * Must be called under a lock that serializes taking new references
2665 * to i_dio_count, usually by inode->i_rwsem.
2666 */
2667void inode_dio_wait(struct inode *inode)
2668{
2669 wait_var_event(&inode->i_dio_count, inode_dio_finished(inode));
2670}
2671EXPORT_SYMBOL(inode_dio_wait);
2672
2673void inode_dio_wait_interruptible(struct inode *inode)
2674{
2675 wait_var_event_interruptible(&inode->i_dio_count,
2676 inode_dio_finished(inode));
2677}
2678EXPORT_SYMBOL(inode_dio_wait_interruptible);
2679
2680/*
2681 * inode_set_flags - atomically set some inode flags
2682 *
2683 * Note: the caller should be holding i_rwsem exclusively, or else be sure that
2684 * they have exclusive access to the inode structure (i.e., while the
2685 * inode is being instantiated). The reason for the cmpxchg() loop
2686 * --- which wouldn't be necessary if all code paths which modify
2687 * i_flags actually followed this rule, is that there is at least one
2688 * code path which doesn't today so we use cmpxchg() out of an abundance
2689 * of caution.
2690 *
2691 * In the long run, i_rwsem is overkill, and we should probably look
2692 * at using the i_lock spinlock to protect i_flags, and then make sure
2693 * it is so documented in include/linux/fs.h and that all code follows
2694 * the locking convention!!
2695 */
2696void inode_set_flags(struct inode *inode, unsigned int flags,
2697 unsigned int mask)
2698{
2699 WARN_ON_ONCE(flags & ~mask);
2700 set_mask_bits(&inode->i_flags, mask, flags);
2701}
2702EXPORT_SYMBOL(inode_set_flags);
2703
2704void inode_nohighmem(struct inode *inode)
2705{
2706 mapping_set_gfp_mask(m: inode->i_mapping, GFP_USER);
2707}
2708EXPORT_SYMBOL(inode_nohighmem);
2709
2710struct timespec64 inode_set_ctime_to_ts(struct inode *inode, struct timespec64 ts)
2711{
2712 trace_inode_set_ctime_to_ts(inode, ctime: &ts);
2713 set_normalized_timespec64(ts: &ts, sec: ts.tv_sec, nsec: ts.tv_nsec);
2714 inode->i_ctime_sec = ts.tv_sec;
2715 inode->i_ctime_nsec = ts.tv_nsec;
2716 return ts;
2717}
2718EXPORT_SYMBOL(inode_set_ctime_to_ts);
2719
2720/**
2721 * timestamp_truncate - Truncate timespec to a granularity
2722 * @t: Timespec
2723 * @inode: inode being updated
2724 *
2725 * Truncate a timespec to the granularity supported by the fs
2726 * containing the inode. Always rounds down. gran must
2727 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2728 */
2729struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2730{
2731 struct super_block *sb = inode->i_sb;
2732 unsigned int gran = sb->s_time_gran;
2733
2734 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2735 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2736 t.tv_nsec = 0;
2737
2738 /* Avoid division in the common cases 1 ns and 1 s. */
2739 if (gran == 1)
2740 ; /* nothing */
2741 else if (gran == NSEC_PER_SEC)
2742 t.tv_nsec = 0;
2743 else if (gran > 1 && gran < NSEC_PER_SEC)
2744 t.tv_nsec -= t.tv_nsec % gran;
2745 else
2746 WARN(1, "invalid file time granularity: %u", gran);
2747 return t;
2748}
2749EXPORT_SYMBOL(timestamp_truncate);
2750
2751/**
2752 * inode_set_ctime_current - set the ctime to current_time
2753 * @inode: inode
2754 *
2755 * Set the inode's ctime to the current value for the inode. Returns the
2756 * current value that was assigned. If this is not a multigrain inode, then we
2757 * set it to the later of the coarse time and floor value.
2758 *
2759 * If it is multigrain, then we first see if the coarse-grained timestamp is
2760 * distinct from what is already there. If so, then use that. Otherwise, get a
2761 * fine-grained timestamp.
2762 *
2763 * After that, try to swap the new value into i_ctime_nsec. Accept the
2764 * resulting ctime, regardless of the outcome of the swap. If it has
2765 * already been replaced, then that timestamp is later than the earlier
2766 * unacceptable one, and is thus acceptable.
2767 */
2768struct timespec64 inode_set_ctime_current(struct inode *inode)
2769{
2770 struct timespec64 now;
2771 u32 cns, cur;
2772
2773 ktime_get_coarse_real_ts64_mg(ts: &now);
2774 now = timestamp_truncate(now, inode);
2775
2776 /* Just return that if this is not a multigrain fs */
2777 if (!is_mgtime(inode)) {
2778 inode_set_ctime_to_ts(inode, now);
2779 goto out;
2780 }
2781
2782 /*
2783 * A fine-grained time is only needed if someone has queried
2784 * for timestamps, and the current coarse grained time isn't
2785 * later than what's already there.
2786 */
2787 cns = smp_load_acquire(&inode->i_ctime_nsec);
2788 if (cns & I_CTIME_QUERIED) {
2789 struct timespec64 ctime = { .tv_sec = inode->i_ctime_sec,
2790 .tv_nsec = cns & ~I_CTIME_QUERIED };
2791
2792 if (timespec64_compare(lhs: &now, rhs: &ctime) <= 0) {
2793 ktime_get_real_ts64_mg(ts: &now);
2794 now = timestamp_truncate(now, inode);
2795 mgtime_counter_inc(mg_fine_stamps);
2796 }
2797 }
2798 mgtime_counter_inc(mg_ctime_updates);
2799
2800 /* No need to cmpxchg if it's exactly the same */
2801 if (cns == now.tv_nsec && inode->i_ctime_sec == now.tv_sec) {
2802 trace_ctime_xchg_skip(inode, ctime: &now);
2803 goto out;
2804 }
2805 cur = cns;
2806retry:
2807 /* Try to swap the nsec value into place. */
2808 if (try_cmpxchg(&inode->i_ctime_nsec, &cur, now.tv_nsec)) {
2809 /* If swap occurred, then we're (mostly) done */
2810 inode->i_ctime_sec = now.tv_sec;
2811 trace_ctime_ns_xchg(inode, old: cns, new: now.tv_nsec, cur);
2812 mgtime_counter_inc(mg_ctime_swaps);
2813 } else {
2814 /*
2815 * Was the change due to someone marking the old ctime QUERIED?
2816 * If so then retry the swap. This can only happen once since
2817 * the only way to clear I_CTIME_QUERIED is to stamp the inode
2818 * with a new ctime.
2819 */
2820 if (!(cns & I_CTIME_QUERIED) && (cns | I_CTIME_QUERIED) == cur) {
2821 cns = cur;
2822 goto retry;
2823 }
2824 /* Otherwise, keep the existing ctime */
2825 now.tv_sec = inode->i_ctime_sec;
2826 now.tv_nsec = cur & ~I_CTIME_QUERIED;
2827 }
2828out:
2829 return now;
2830}
2831EXPORT_SYMBOL(inode_set_ctime_current);
2832
2833/**
2834 * inode_set_ctime_deleg - try to update the ctime on a delegated inode
2835 * @inode: inode to update
2836 * @update: timespec64 to set the ctime
2837 *
2838 * Attempt to atomically update the ctime on behalf of a delegation holder.
2839 *
2840 * The nfs server can call back the holder of a delegation to get updated
2841 * inode attributes, including the mtime. When updating the mtime, update
2842 * the ctime to a value at least equal to that.
2843 *
2844 * This can race with concurrent updates to the inode, in which
2845 * case the update is skipped.
2846 *
2847 * Note that this works even when multigrain timestamps are not enabled,
2848 * so it is used in either case.
2849 */
2850struct timespec64 inode_set_ctime_deleg(struct inode *inode, struct timespec64 update)
2851{
2852 struct timespec64 now, cur_ts;
2853 u32 cur, old;
2854
2855 /* pairs with try_cmpxchg below */
2856 cur = smp_load_acquire(&inode->i_ctime_nsec);
2857 cur_ts.tv_nsec = cur & ~I_CTIME_QUERIED;
2858 cur_ts.tv_sec = inode->i_ctime_sec;
2859
2860 /* If the update is older than the existing value, skip it. */
2861 if (timespec64_compare(lhs: &update, rhs: &cur_ts) <= 0)
2862 return cur_ts;
2863
2864 ktime_get_coarse_real_ts64_mg(ts: &now);
2865
2866 /* Clamp the update to "now" if it's in the future */
2867 if (timespec64_compare(lhs: &update, rhs: &now) > 0)
2868 update = now;
2869
2870 update = timestamp_truncate(update, inode);
2871
2872 /* No need to update if the values are already the same */
2873 if (timespec64_equal(a: &update, b: &cur_ts))
2874 return cur_ts;
2875
2876 /*
2877 * Try to swap the nsec value into place. If it fails, that means
2878 * it raced with an update due to a write or similar activity. That
2879 * stamp takes precedence, so just skip the update.
2880 */
2881retry:
2882 old = cur;
2883 if (try_cmpxchg(&inode->i_ctime_nsec, &cur, update.tv_nsec)) {
2884 inode->i_ctime_sec = update.tv_sec;
2885 mgtime_counter_inc(mg_ctime_swaps);
2886 return update;
2887 }
2888
2889 /*
2890 * Was the change due to another task marking the old ctime QUERIED?
2891 *
2892 * If so, then retry the swap. This can only happen once since
2893 * the only way to clear I_CTIME_QUERIED is to stamp the inode
2894 * with a new ctime.
2895 */
2896 if (!(old & I_CTIME_QUERIED) && (cur == (old | I_CTIME_QUERIED)))
2897 goto retry;
2898
2899 /* Otherwise, it was a new timestamp. */
2900 cur_ts.tv_sec = inode->i_ctime_sec;
2901 cur_ts.tv_nsec = cur & ~I_CTIME_QUERIED;
2902 return cur_ts;
2903}
2904EXPORT_SYMBOL(inode_set_ctime_deleg);
2905
2906/**
2907 * in_group_or_capable - check whether caller is CAP_FSETID privileged
2908 * @idmap: idmap of the mount @inode was found from
2909 * @inode: inode to check
2910 * @vfsgid: the new/current vfsgid of @inode
2911 *
2912 * Check whether @vfsgid is in the caller's group list or if the caller is
2913 * privileged with CAP_FSETID over @inode. This can be used to determine
2914 * whether the setgid bit can be kept or must be dropped.
2915 *
2916 * Return: true if the caller is sufficiently privileged, false if not.
2917 */
2918bool in_group_or_capable(struct mnt_idmap *idmap,
2919 const struct inode *inode, vfsgid_t vfsgid)
2920{
2921 if (vfsgid_in_group_p(vfsgid))
2922 return true;
2923 if (capable_wrt_inode_uidgid(idmap, inode, CAP_FSETID))
2924 return true;
2925 return false;
2926}
2927EXPORT_SYMBOL(in_group_or_capable);
2928
2929/**
2930 * mode_strip_sgid - handle the sgid bit for non-directories
2931 * @idmap: idmap of the mount the inode was created from
2932 * @dir: parent directory inode
2933 * @mode: mode of the file to be created in @dir
2934 *
2935 * If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2936 * raised and @dir has the S_ISGID bit raised ensure that the caller is
2937 * either in the group of the parent directory or they have CAP_FSETID
2938 * in their user namespace and are privileged over the parent directory.
2939 * In all other cases, strip the S_ISGID bit from @mode.
2940 *
2941 * Return: the new mode to use for the file
2942 */
2943umode_t mode_strip_sgid(struct mnt_idmap *idmap,
2944 const struct inode *dir, umode_t mode)
2945{
2946 if ((mode & (S_ISGID | S_IXGRP)) != (S_ISGID | S_IXGRP))
2947 return mode;
2948 if (S_ISDIR(mode) || !dir || !(dir->i_mode & S_ISGID))
2949 return mode;
2950 if (in_group_or_capable(idmap, dir, i_gid_into_vfsgid(idmap, inode: dir)))
2951 return mode;
2952 return mode & ~S_ISGID;
2953}
2954EXPORT_SYMBOL(mode_strip_sgid);
2955
2956#ifdef CONFIG_DEBUG_VFS
2957/*
2958 * Dump an inode.
2959 *
2960 * TODO: add a proper inode dumping routine, this is a stub to get debug off the
2961 * ground.
2962 *
2963 * TODO: handle getting to fs type with get_kernel_nofault()?
2964 * See dump_mapping() above.
2965 */
2966void dump_inode(struct inode *inode, const char *reason)
2967{
2968 struct super_block *sb = inode->i_sb;
2969
2970 pr_warn("%s encountered for inode %px\n"
2971 "fs %s mode %ho opflags 0x%hx flags 0x%x state 0x%x count %d\n",
2972 reason, inode, sb->s_type->name, inode->i_mode, inode->i_opflags,
2973 inode->i_flags, inode->i_state, atomic_read(&inode->i_count));
2974}
2975
2976EXPORT_SYMBOL(dump_inode);
2977#endif
2978