1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * (C) 1997 Linus Torvalds
4 * (C) 1999 Andrea Arcangeli <andrea@suse.de> (dynamic inode allocation)
5 */
6 #include <linux/export.h>
7 #include <linux/fs.h>
8 #include <linux/mm.h>
9 #include <linux/backing-dev.h>
10 #include <linux/hash.h>
11 #include <linux/swap.h>
12 #include <linux/security.h>
13 #include <linux/cdev.h>
14 #include <linux/memblock.h>
15 #include <linux/fsnotify.h>
16 #include <linux/mount.h>
17 #include <linux/posix_acl.h>
18 #include <linux/prefetch.h>
19 #include <linux/buffer_head.h> /* for inode_has_buffers */
20 #include <linux/ratelimit.h>
21 #include <linux/list_lru.h>
22 #include <linux/iversion.h>
23 #include <trace/events/writeback.h>
24 #include "internal.h"
25
26 /*
27 * Inode locking rules:
28 *
29 * inode->i_lock protects:
30 * inode->i_state, inode->i_hash, __iget(), inode->i_io_list
31 * Inode LRU list locks protect:
32 * inode->i_sb->s_inode_lru, inode->i_lru
33 * inode->i_sb->s_inode_list_lock protects:
34 * inode->i_sb->s_inodes, inode->i_sb_list
35 * bdi->wb.list_lock protects:
36 * bdi->wb.b_{dirty,io,more_io,dirty_time}, inode->i_io_list
37 * inode_hash_lock protects:
38 * inode_hashtable, inode->i_hash
39 *
40 * Lock ordering:
41 *
42 * inode->i_sb->s_inode_list_lock
43 * inode->i_lock
44 * Inode LRU list locks
45 *
46 * bdi->wb.list_lock
47 * inode->i_lock
48 *
49 * inode_hash_lock
50 * inode->i_sb->s_inode_list_lock
51 * inode->i_lock
52 *
53 * iunique_lock
54 * inode_hash_lock
55 */
56
57 static unsigned int i_hash_mask __read_mostly;
58 static unsigned int i_hash_shift __read_mostly;
59 static struct hlist_head *inode_hashtable __read_mostly;
60 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
61
62 /*
63 * Empty aops. Can be used for the cases where the user does not
64 * define any of the address_space operations.
65 */
66 const struct address_space_operations empty_aops = {
67 };
68 EXPORT_SYMBOL(empty_aops);
69
70 static DEFINE_PER_CPU(unsigned long, nr_inodes);
71 static DEFINE_PER_CPU(unsigned long, nr_unused);
72
73 static struct kmem_cache *inode_cachep __read_mostly;
74
get_nr_inodes(void)75 static long get_nr_inodes(void)
76 {
77 int i;
78 long sum = 0;
79 for_each_possible_cpu(i)
80 sum += per_cpu(nr_inodes, i);
81 return sum < 0 ? 0 : sum;
82 }
83
get_nr_inodes_unused(void)84 static inline long get_nr_inodes_unused(void)
85 {
86 int i;
87 long sum = 0;
88 for_each_possible_cpu(i)
89 sum += per_cpu(nr_unused, i);
90 return sum < 0 ? 0 : sum;
91 }
92
get_nr_dirty_inodes(void)93 long get_nr_dirty_inodes(void)
94 {
95 /* not actually dirty inodes, but a wild approximation */
96 long nr_dirty = get_nr_inodes() - get_nr_inodes_unused();
97 return nr_dirty > 0 ? nr_dirty : 0;
98 }
99
100 /*
101 * Handle nr_inode sysctl
102 */
103 #ifdef CONFIG_SYSCTL
104 /*
105 * Statistics gathering..
106 */
107 static struct inodes_stat_t inodes_stat;
108
proc_nr_inodes(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)109 static int proc_nr_inodes(struct ctl_table *table, int write, void *buffer,
110 size_t *lenp, loff_t *ppos)
111 {
112 inodes_stat.nr_inodes = get_nr_inodes();
113 inodes_stat.nr_unused = get_nr_inodes_unused();
114 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
115 }
116
117 static struct ctl_table inodes_sysctls[] = {
118 {
119 .procname = "inode-nr",
120 .data = &inodes_stat,
121 .maxlen = 2*sizeof(long),
122 .mode = 0444,
123 .proc_handler = proc_nr_inodes,
124 },
125 {
126 .procname = "inode-state",
127 .data = &inodes_stat,
128 .maxlen = 7*sizeof(long),
129 .mode = 0444,
130 .proc_handler = proc_nr_inodes,
131 },
132 { }
133 };
134
init_fs_inode_sysctls(void)135 static int __init init_fs_inode_sysctls(void)
136 {
137 register_sysctl_init("fs", inodes_sysctls);
138 return 0;
139 }
140 early_initcall(init_fs_inode_sysctls);
141 #endif
142
no_open(struct inode * inode,struct file * file)143 static int no_open(struct inode *inode, struct file *file)
144 {
145 return -ENXIO;
146 }
147
148 /**
149 * inode_init_always - perform inode structure initialisation
150 * @sb: superblock inode belongs to
151 * @inode: inode to initialise
152 *
153 * These are initializations that need to be done on every inode
154 * allocation as the fields are not initialised by slab allocation.
155 */
inode_init_always(struct super_block * sb,struct inode * inode)156 int inode_init_always(struct super_block *sb, struct inode *inode)
157 {
158 static const struct inode_operations empty_iops;
159 static const struct file_operations no_open_fops = {.open = no_open};
160 struct address_space *const mapping = &inode->i_data;
161
162 inode->i_sb = sb;
163 inode->i_blkbits = sb->s_blocksize_bits;
164 inode->i_flags = 0;
165 atomic64_set(&inode->i_sequence, 0);
166 atomic_set(&inode->i_count, 1);
167 inode->i_op = &empty_iops;
168 inode->i_fop = &no_open_fops;
169 inode->i_ino = 0;
170 inode->__i_nlink = 1;
171 inode->i_opflags = 0;
172 if (sb->s_xattr)
173 inode->i_opflags |= IOP_XATTR;
174 i_uid_write(inode, 0);
175 i_gid_write(inode, 0);
176 atomic_set(&inode->i_writecount, 0);
177 inode->i_size = 0;
178 inode->i_write_hint = WRITE_LIFE_NOT_SET;
179 inode->i_blocks = 0;
180 inode->i_bytes = 0;
181 inode->i_generation = 0;
182 inode->i_pipe = NULL;
183 inode->i_cdev = NULL;
184 inode->i_link = NULL;
185 inode->i_dir_seq = 0;
186 inode->i_rdev = 0;
187 inode->dirtied_when = 0;
188
189 #ifdef CONFIG_CGROUP_WRITEBACK
190 inode->i_wb_frn_winner = 0;
191 inode->i_wb_frn_avg_time = 0;
192 inode->i_wb_frn_history = 0;
193 #endif
194
195 spin_lock_init(&inode->i_lock);
196 lockdep_set_class(&inode->i_lock, &sb->s_type->i_lock_key);
197
198 init_rwsem(&inode->i_rwsem);
199 lockdep_set_class(&inode->i_rwsem, &sb->s_type->i_mutex_key);
200
201 atomic_set(&inode->i_dio_count, 0);
202
203 mapping->a_ops = &empty_aops;
204 mapping->host = inode;
205 mapping->flags = 0;
206 mapping->wb_err = 0;
207 atomic_set(&mapping->i_mmap_writable, 0);
208 #ifdef CONFIG_READ_ONLY_THP_FOR_FS
209 atomic_set(&mapping->nr_thps, 0);
210 #endif
211 mapping_set_gfp_mask(mapping, GFP_HIGHUSER_MOVABLE);
212 mapping->private_data = NULL;
213 mapping->writeback_index = 0;
214 init_rwsem(&mapping->invalidate_lock);
215 lockdep_set_class_and_name(&mapping->invalidate_lock,
216 &sb->s_type->invalidate_lock_key,
217 "mapping.invalidate_lock");
218 if (sb->s_iflags & SB_I_STABLE_WRITES)
219 mapping_set_stable_writes(mapping);
220 inode->i_private = NULL;
221 inode->i_mapping = mapping;
222 INIT_HLIST_HEAD(&inode->i_dentry); /* buggered by rcu freeing */
223 #ifdef CONFIG_FS_POSIX_ACL
224 inode->i_acl = inode->i_default_acl = ACL_NOT_CACHED;
225 #endif
226
227 #ifdef CONFIG_FSNOTIFY
228 inode->i_fsnotify_mask = 0;
229 #endif
230 inode->i_flctx = NULL;
231
232 if (unlikely(security_inode_alloc(inode)))
233 return -ENOMEM;
234 this_cpu_inc(nr_inodes);
235
236 return 0;
237 }
238 EXPORT_SYMBOL(inode_init_always);
239
free_inode_nonrcu(struct inode * inode)240 void free_inode_nonrcu(struct inode *inode)
241 {
242 kmem_cache_free(inode_cachep, inode);
243 }
244 EXPORT_SYMBOL(free_inode_nonrcu);
245
i_callback(struct rcu_head * head)246 static void i_callback(struct rcu_head *head)
247 {
248 struct inode *inode = container_of(head, struct inode, i_rcu);
249 if (inode->free_inode)
250 inode->free_inode(inode);
251 else
252 free_inode_nonrcu(inode);
253 }
254
alloc_inode(struct super_block * sb)255 static struct inode *alloc_inode(struct super_block *sb)
256 {
257 const struct super_operations *ops = sb->s_op;
258 struct inode *inode;
259
260 if (ops->alloc_inode)
261 inode = ops->alloc_inode(sb);
262 else
263 inode = alloc_inode_sb(sb, inode_cachep, GFP_KERNEL);
264
265 if (!inode)
266 return NULL;
267
268 if (unlikely(inode_init_always(sb, inode))) {
269 if (ops->destroy_inode) {
270 ops->destroy_inode(inode);
271 if (!ops->free_inode)
272 return NULL;
273 }
274 inode->free_inode = ops->free_inode;
275 i_callback(&inode->i_rcu);
276 return NULL;
277 }
278
279 return inode;
280 }
281
__destroy_inode(struct inode * inode)282 void __destroy_inode(struct inode *inode)
283 {
284 BUG_ON(inode_has_buffers(inode));
285 inode_detach_wb(inode);
286 security_inode_free(inode);
287 fsnotify_inode_delete(inode);
288 locks_free_lock_context(inode);
289 if (!inode->i_nlink) {
290 WARN_ON(atomic_long_read(&inode->i_sb->s_remove_count) == 0);
291 atomic_long_dec(&inode->i_sb->s_remove_count);
292 }
293
294 #ifdef CONFIG_FS_POSIX_ACL
295 if (inode->i_acl && !is_uncached_acl(inode->i_acl))
296 posix_acl_release(inode->i_acl);
297 if (inode->i_default_acl && !is_uncached_acl(inode->i_default_acl))
298 posix_acl_release(inode->i_default_acl);
299 #endif
300 this_cpu_dec(nr_inodes);
301 }
302 EXPORT_SYMBOL(__destroy_inode);
303
destroy_inode(struct inode * inode)304 static void destroy_inode(struct inode *inode)
305 {
306 const struct super_operations *ops = inode->i_sb->s_op;
307
308 BUG_ON(!list_empty(&inode->i_lru));
309 __destroy_inode(inode);
310 if (ops->destroy_inode) {
311 ops->destroy_inode(inode);
312 if (!ops->free_inode)
313 return;
314 }
315 inode->free_inode = ops->free_inode;
316 call_rcu(&inode->i_rcu, i_callback);
317 }
318
319 /**
320 * drop_nlink - directly drop an inode's link count
321 * @inode: inode
322 *
323 * This is a low-level filesystem helper to replace any
324 * direct filesystem manipulation of i_nlink. In cases
325 * where we are attempting to track writes to the
326 * filesystem, a decrement to zero means an imminent
327 * write when the file is truncated and actually unlinked
328 * on the filesystem.
329 */
drop_nlink(struct inode * inode)330 void drop_nlink(struct inode *inode)
331 {
332 WARN_ON(inode->i_nlink == 0);
333 inode->__i_nlink--;
334 if (!inode->i_nlink)
335 atomic_long_inc(&inode->i_sb->s_remove_count);
336 }
337 EXPORT_SYMBOL(drop_nlink);
338
339 /**
340 * clear_nlink - directly zero an inode's link count
341 * @inode: inode
342 *
343 * This is a low-level filesystem helper to replace any
344 * direct filesystem manipulation of i_nlink. See
345 * drop_nlink() for why we care about i_nlink hitting zero.
346 */
clear_nlink(struct inode * inode)347 void clear_nlink(struct inode *inode)
348 {
349 if (inode->i_nlink) {
350 inode->__i_nlink = 0;
351 atomic_long_inc(&inode->i_sb->s_remove_count);
352 }
353 }
354 EXPORT_SYMBOL(clear_nlink);
355
356 /**
357 * set_nlink - directly set an inode's link count
358 * @inode: inode
359 * @nlink: new nlink (should be non-zero)
360 *
361 * This is a low-level filesystem helper to replace any
362 * direct filesystem manipulation of i_nlink.
363 */
set_nlink(struct inode * inode,unsigned int nlink)364 void set_nlink(struct inode *inode, unsigned int nlink)
365 {
366 if (!nlink) {
367 clear_nlink(inode);
368 } else {
369 /* Yes, some filesystems do change nlink from zero to one */
370 if (inode->i_nlink == 0)
371 atomic_long_dec(&inode->i_sb->s_remove_count);
372
373 inode->__i_nlink = nlink;
374 }
375 }
376 EXPORT_SYMBOL(set_nlink);
377
378 /**
379 * inc_nlink - directly increment an inode's link count
380 * @inode: inode
381 *
382 * This is a low-level filesystem helper to replace any
383 * direct filesystem manipulation of i_nlink. Currently,
384 * it is only here for parity with dec_nlink().
385 */
inc_nlink(struct inode * inode)386 void inc_nlink(struct inode *inode)
387 {
388 if (unlikely(inode->i_nlink == 0)) {
389 WARN_ON(!(inode->i_state & I_LINKABLE));
390 atomic_long_dec(&inode->i_sb->s_remove_count);
391 }
392
393 inode->__i_nlink++;
394 }
395 EXPORT_SYMBOL(inc_nlink);
396
__address_space_init_once(struct address_space * mapping)397 static void __address_space_init_once(struct address_space *mapping)
398 {
399 xa_init_flags(&mapping->i_pages, XA_FLAGS_LOCK_IRQ | XA_FLAGS_ACCOUNT);
400 init_rwsem(&mapping->i_mmap_rwsem);
401 INIT_LIST_HEAD(&mapping->private_list);
402 spin_lock_init(&mapping->private_lock);
403 mapping->i_mmap = RB_ROOT_CACHED;
404 }
405
address_space_init_once(struct address_space * mapping)406 void address_space_init_once(struct address_space *mapping)
407 {
408 memset(mapping, 0, sizeof(*mapping));
409 __address_space_init_once(mapping);
410 }
411 EXPORT_SYMBOL(address_space_init_once);
412
413 /*
414 * These are initializations that only need to be done
415 * once, because the fields are idempotent across use
416 * of the inode, so let the slab aware of that.
417 */
inode_init_once(struct inode * inode)418 void inode_init_once(struct inode *inode)
419 {
420 memset(inode, 0, sizeof(*inode));
421 INIT_HLIST_NODE(&inode->i_hash);
422 INIT_LIST_HEAD(&inode->i_devices);
423 INIT_LIST_HEAD(&inode->i_io_list);
424 INIT_LIST_HEAD(&inode->i_wb_list);
425 INIT_LIST_HEAD(&inode->i_lru);
426 INIT_LIST_HEAD(&inode->i_sb_list);
427 __address_space_init_once(&inode->i_data);
428 i_size_ordered_init(inode);
429 }
430 EXPORT_SYMBOL(inode_init_once);
431
init_once(void * foo)432 static void init_once(void *foo)
433 {
434 struct inode *inode = (struct inode *) foo;
435
436 inode_init_once(inode);
437 }
438
439 /*
440 * inode->i_lock must be held
441 */
__iget(struct inode * inode)442 void __iget(struct inode *inode)
443 {
444 atomic_inc(&inode->i_count);
445 }
446
447 /*
448 * get additional reference to inode; caller must already hold one.
449 */
ihold(struct inode * inode)450 void ihold(struct inode *inode)
451 {
452 WARN_ON(atomic_inc_return(&inode->i_count) < 2);
453 }
454 EXPORT_SYMBOL(ihold);
455
__inode_add_lru(struct inode * inode,bool rotate)456 static void __inode_add_lru(struct inode *inode, bool rotate)
457 {
458 if (inode->i_state & (I_DIRTY_ALL | I_SYNC | I_FREEING | I_WILL_FREE))
459 return;
460 if (atomic_read(&inode->i_count))
461 return;
462 if (!(inode->i_sb->s_flags & SB_ACTIVE))
463 return;
464 if (!mapping_shrinkable(&inode->i_data))
465 return;
466
467 if (list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru))
468 this_cpu_inc(nr_unused);
469 else if (rotate)
470 inode->i_state |= I_REFERENCED;
471 }
472
473 /*
474 * Add inode to LRU if needed (inode is unused and clean).
475 *
476 * Needs inode->i_lock held.
477 */
inode_add_lru(struct inode * inode)478 void inode_add_lru(struct inode *inode)
479 {
480 __inode_add_lru(inode, false);
481 }
482
inode_lru_list_del(struct inode * inode)483 static void inode_lru_list_del(struct inode *inode)
484 {
485 if (list_lru_del(&inode->i_sb->s_inode_lru, &inode->i_lru))
486 this_cpu_dec(nr_unused);
487 }
488
489 /**
490 * inode_sb_list_add - add inode to the superblock list of inodes
491 * @inode: inode to add
492 */
inode_sb_list_add(struct inode * inode)493 void inode_sb_list_add(struct inode *inode)
494 {
495 spin_lock(&inode->i_sb->s_inode_list_lock);
496 list_add(&inode->i_sb_list, &inode->i_sb->s_inodes);
497 spin_unlock(&inode->i_sb->s_inode_list_lock);
498 }
499 EXPORT_SYMBOL_GPL(inode_sb_list_add);
500
inode_sb_list_del(struct inode * inode)501 static inline void inode_sb_list_del(struct inode *inode)
502 {
503 if (!list_empty(&inode->i_sb_list)) {
504 spin_lock(&inode->i_sb->s_inode_list_lock);
505 list_del_init(&inode->i_sb_list);
506 spin_unlock(&inode->i_sb->s_inode_list_lock);
507 }
508 }
509
hash(struct super_block * sb,unsigned long hashval)510 static unsigned long hash(struct super_block *sb, unsigned long hashval)
511 {
512 unsigned long tmp;
513
514 tmp = (hashval * (unsigned long)sb) ^ (GOLDEN_RATIO_PRIME + hashval) /
515 L1_CACHE_BYTES;
516 tmp = tmp ^ ((tmp ^ GOLDEN_RATIO_PRIME) >> i_hash_shift);
517 return tmp & i_hash_mask;
518 }
519
520 /**
521 * __insert_inode_hash - hash an inode
522 * @inode: unhashed inode
523 * @hashval: unsigned long value used to locate this object in the
524 * inode_hashtable.
525 *
526 * Add an inode to the inode hash for this superblock.
527 */
__insert_inode_hash(struct inode * inode,unsigned long hashval)528 void __insert_inode_hash(struct inode *inode, unsigned long hashval)
529 {
530 struct hlist_head *b = inode_hashtable + hash(inode->i_sb, hashval);
531
532 spin_lock(&inode_hash_lock);
533 spin_lock(&inode->i_lock);
534 hlist_add_head_rcu(&inode->i_hash, b);
535 spin_unlock(&inode->i_lock);
536 spin_unlock(&inode_hash_lock);
537 }
538 EXPORT_SYMBOL(__insert_inode_hash);
539
540 /**
541 * __remove_inode_hash - remove an inode from the hash
542 * @inode: inode to unhash
543 *
544 * Remove an inode from the superblock.
545 */
__remove_inode_hash(struct inode * inode)546 void __remove_inode_hash(struct inode *inode)
547 {
548 spin_lock(&inode_hash_lock);
549 spin_lock(&inode->i_lock);
550 hlist_del_init_rcu(&inode->i_hash);
551 spin_unlock(&inode->i_lock);
552 spin_unlock(&inode_hash_lock);
553 }
554 EXPORT_SYMBOL(__remove_inode_hash);
555
dump_mapping(const struct address_space * mapping)556 void dump_mapping(const struct address_space *mapping)
557 {
558 struct inode *host;
559 const struct address_space_operations *a_ops;
560 struct hlist_node *dentry_first;
561 struct dentry *dentry_ptr;
562 struct dentry dentry;
563 unsigned long ino;
564
565 /*
566 * If mapping is an invalid pointer, we don't want to crash
567 * accessing it, so probe everything depending on it carefully.
568 */
569 if (get_kernel_nofault(host, &mapping->host) ||
570 get_kernel_nofault(a_ops, &mapping->a_ops)) {
571 pr_warn("invalid mapping:%px\n", mapping);
572 return;
573 }
574
575 if (!host) {
576 pr_warn("aops:%ps\n", a_ops);
577 return;
578 }
579
580 if (get_kernel_nofault(dentry_first, &host->i_dentry.first) ||
581 get_kernel_nofault(ino, &host->i_ino)) {
582 pr_warn("aops:%ps invalid inode:%px\n", a_ops, host);
583 return;
584 }
585
586 if (!dentry_first) {
587 pr_warn("aops:%ps ino:%lx\n", a_ops, ino);
588 return;
589 }
590
591 dentry_ptr = container_of(dentry_first, struct dentry, d_u.d_alias);
592 if (get_kernel_nofault(dentry, dentry_ptr)) {
593 pr_warn("aops:%ps ino:%lx invalid dentry:%px\n",
594 a_ops, ino, dentry_ptr);
595 return;
596 }
597
598 /*
599 * if dentry is corrupted, the %pd handler may still crash,
600 * but it's unlikely that we reach here with a corrupt mapping
601 */
602 pr_warn("aops:%ps ino:%lx dentry name:\"%pd\"\n", a_ops, ino, &dentry);
603 }
604
clear_inode(struct inode * inode)605 void clear_inode(struct inode *inode)
606 {
607 /*
608 * We have to cycle the i_pages lock here because reclaim can be in the
609 * process of removing the last page (in __filemap_remove_folio())
610 * and we must not free the mapping under it.
611 */
612 xa_lock_irq(&inode->i_data.i_pages);
613 BUG_ON(inode->i_data.nrpages);
614 /*
615 * Almost always, mapping_empty(&inode->i_data) here; but there are
616 * two known and long-standing ways in which nodes may get left behind
617 * (when deep radix-tree node allocation failed partway; or when THP
618 * collapse_file() failed). Until those two known cases are cleaned up,
619 * or a cleanup function is called here, do not BUG_ON(!mapping_empty),
620 * nor even WARN_ON(!mapping_empty).
621 */
622 xa_unlock_irq(&inode->i_data.i_pages);
623 BUG_ON(!list_empty(&inode->i_data.private_list));
624 BUG_ON(!(inode->i_state & I_FREEING));
625 BUG_ON(inode->i_state & I_CLEAR);
626 BUG_ON(!list_empty(&inode->i_wb_list));
627 /* don't need i_lock here, no concurrent mods to i_state */
628 inode->i_state = I_FREEING | I_CLEAR;
629 }
630 EXPORT_SYMBOL(clear_inode);
631
632 /*
633 * Free the inode passed in, removing it from the lists it is still connected
634 * to. We remove any pages still attached to the inode and wait for any IO that
635 * is still in progress before finally destroying the inode.
636 *
637 * An inode must already be marked I_FREEING so that we avoid the inode being
638 * moved back onto lists if we race with other code that manipulates the lists
639 * (e.g. writeback_single_inode). The caller is responsible for setting this.
640 *
641 * An inode must already be removed from the LRU list before being evicted from
642 * the cache. This should occur atomically with setting the I_FREEING state
643 * flag, so no inodes here should ever be on the LRU when being evicted.
644 */
evict(struct inode * inode)645 static void evict(struct inode *inode)
646 {
647 const struct super_operations *op = inode->i_sb->s_op;
648
649 BUG_ON(!(inode->i_state & I_FREEING));
650 BUG_ON(!list_empty(&inode->i_lru));
651
652 if (!list_empty(&inode->i_io_list))
653 inode_io_list_del(inode);
654
655 inode_sb_list_del(inode);
656
657 /*
658 * Wait for flusher thread to be done with the inode so that filesystem
659 * does not start destroying it while writeback is still running. Since
660 * the inode has I_FREEING set, flusher thread won't start new work on
661 * the inode. We just have to wait for running writeback to finish.
662 */
663 inode_wait_for_writeback(inode);
664
665 if (op->evict_inode) {
666 op->evict_inode(inode);
667 } else {
668 truncate_inode_pages_final(&inode->i_data);
669 clear_inode(inode);
670 }
671 if (S_ISCHR(inode->i_mode) && inode->i_cdev)
672 cd_forget(inode);
673
674 remove_inode_hash(inode);
675
676 spin_lock(&inode->i_lock);
677 wake_up_bit(&inode->i_state, __I_NEW);
678 BUG_ON(inode->i_state != (I_FREEING | I_CLEAR));
679 spin_unlock(&inode->i_lock);
680
681 destroy_inode(inode);
682 }
683
684 /*
685 * dispose_list - dispose of the contents of a local list
686 * @head: the head of the list to free
687 *
688 * Dispose-list gets a local list with local inodes in it, so it doesn't
689 * need to worry about list corruption and SMP locks.
690 */
dispose_list(struct list_head * head)691 static void dispose_list(struct list_head *head)
692 {
693 while (!list_empty(head)) {
694 struct inode *inode;
695
696 inode = list_first_entry(head, struct inode, i_lru);
697 list_del_init(&inode->i_lru);
698
699 evict(inode);
700 cond_resched();
701 }
702 }
703
704 /**
705 * evict_inodes - evict all evictable inodes for a superblock
706 * @sb: superblock to operate on
707 *
708 * Make sure that no inodes with zero refcount are retained. This is
709 * called by superblock shutdown after having SB_ACTIVE flag removed,
710 * so any inode reaching zero refcount during or after that call will
711 * be immediately evicted.
712 */
evict_inodes(struct super_block * sb)713 void evict_inodes(struct super_block *sb)
714 {
715 struct inode *inode, *next;
716 LIST_HEAD(dispose);
717
718 again:
719 spin_lock(&sb->s_inode_list_lock);
720 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
721 if (atomic_read(&inode->i_count))
722 continue;
723
724 spin_lock(&inode->i_lock);
725 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
726 spin_unlock(&inode->i_lock);
727 continue;
728 }
729
730 inode->i_state |= I_FREEING;
731 inode_lru_list_del(inode);
732 spin_unlock(&inode->i_lock);
733 list_add(&inode->i_lru, &dispose);
734
735 /*
736 * We can have a ton of inodes to evict at unmount time given
737 * enough memory, check to see if we need to go to sleep for a
738 * bit so we don't livelock.
739 */
740 if (need_resched()) {
741 spin_unlock(&sb->s_inode_list_lock);
742 cond_resched();
743 dispose_list(&dispose);
744 goto again;
745 }
746 }
747 spin_unlock(&sb->s_inode_list_lock);
748
749 dispose_list(&dispose);
750 }
751 EXPORT_SYMBOL_GPL(evict_inodes);
752
753 /**
754 * invalidate_inodes - attempt to free all inodes on a superblock
755 * @sb: superblock to operate on
756 * @kill_dirty: flag to guide handling of dirty inodes
757 *
758 * Attempts to free all inodes for a given superblock. If there were any
759 * busy inodes return a non-zero value, else zero.
760 * If @kill_dirty is set, discard dirty inodes too, otherwise treat
761 * them as busy.
762 */
invalidate_inodes(struct super_block * sb,bool kill_dirty)763 int invalidate_inodes(struct super_block *sb, bool kill_dirty)
764 {
765 int busy = 0;
766 struct inode *inode, *next;
767 LIST_HEAD(dispose);
768
769 again:
770 spin_lock(&sb->s_inode_list_lock);
771 list_for_each_entry_safe(inode, next, &sb->s_inodes, i_sb_list) {
772 spin_lock(&inode->i_lock);
773 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
774 spin_unlock(&inode->i_lock);
775 continue;
776 }
777 if (inode->i_state & I_DIRTY_ALL && !kill_dirty) {
778 spin_unlock(&inode->i_lock);
779 busy = 1;
780 continue;
781 }
782 if (atomic_read(&inode->i_count)) {
783 spin_unlock(&inode->i_lock);
784 busy = 1;
785 continue;
786 }
787
788 inode->i_state |= I_FREEING;
789 inode_lru_list_del(inode);
790 spin_unlock(&inode->i_lock);
791 list_add(&inode->i_lru, &dispose);
792 if (need_resched()) {
793 spin_unlock(&sb->s_inode_list_lock);
794 cond_resched();
795 dispose_list(&dispose);
796 goto again;
797 }
798 }
799 spin_unlock(&sb->s_inode_list_lock);
800
801 dispose_list(&dispose);
802
803 return busy;
804 }
805
806 /*
807 * Isolate the inode from the LRU in preparation for freeing it.
808 *
809 * If the inode has the I_REFERENCED flag set, then it means that it has been
810 * used recently - the flag is set in iput_final(). When we encounter such an
811 * inode, clear the flag and move it to the back of the LRU so it gets another
812 * pass through the LRU before it gets reclaimed. This is necessary because of
813 * the fact we are doing lazy LRU updates to minimise lock contention so the
814 * LRU does not have strict ordering. Hence we don't want to reclaim inodes
815 * with this flag set because they are the inodes that are out of order.
816 */
inode_lru_isolate(struct list_head * item,struct list_lru_one * lru,spinlock_t * lru_lock,void * arg)817 static enum lru_status inode_lru_isolate(struct list_head *item,
818 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
819 {
820 struct list_head *freeable = arg;
821 struct inode *inode = container_of(item, struct inode, i_lru);
822
823 /*
824 * We are inverting the lru lock/inode->i_lock here, so use a
825 * trylock. If we fail to get the lock, just skip it.
826 */
827 if (!spin_trylock(&inode->i_lock))
828 return LRU_SKIP;
829
830 /*
831 * Inodes can get referenced, redirtied, or repopulated while
832 * they're already on the LRU, and this can make them
833 * unreclaimable for a while. Remove them lazily here; iput,
834 * sync, or the last page cache deletion will requeue them.
835 */
836 if (atomic_read(&inode->i_count) ||
837 (inode->i_state & ~I_REFERENCED) ||
838 !mapping_shrinkable(&inode->i_data)) {
839 list_lru_isolate(lru, &inode->i_lru);
840 spin_unlock(&inode->i_lock);
841 this_cpu_dec(nr_unused);
842 return LRU_REMOVED;
843 }
844
845 /* Recently referenced inodes get one more pass */
846 if (inode->i_state & I_REFERENCED) {
847 inode->i_state &= ~I_REFERENCED;
848 spin_unlock(&inode->i_lock);
849 return LRU_ROTATE;
850 }
851
852 /*
853 * On highmem systems, mapping_shrinkable() permits dropping
854 * page cache in order to free up struct inodes: lowmem might
855 * be under pressure before the cache inside the highmem zone.
856 */
857 if (inode_has_buffers(inode) || !mapping_empty(&inode->i_data)) {
858 __iget(inode);
859 spin_unlock(&inode->i_lock);
860 spin_unlock(lru_lock);
861 if (remove_inode_buffers(inode)) {
862 unsigned long reap;
863 reap = invalidate_mapping_pages(&inode->i_data, 0, -1);
864 if (current_is_kswapd())
865 __count_vm_events(KSWAPD_INODESTEAL, reap);
866 else
867 __count_vm_events(PGINODESTEAL, reap);
868 if (current->reclaim_state)
869 current->reclaim_state->reclaimed_slab += reap;
870 }
871 iput(inode);
872 spin_lock(lru_lock);
873 return LRU_RETRY;
874 }
875
876 WARN_ON(inode->i_state & I_NEW);
877 inode->i_state |= I_FREEING;
878 list_lru_isolate_move(lru, &inode->i_lru, freeable);
879 spin_unlock(&inode->i_lock);
880
881 this_cpu_dec(nr_unused);
882 return LRU_REMOVED;
883 }
884
885 /*
886 * Walk the superblock inode LRU for freeable inodes and attempt to free them.
887 * This is called from the superblock shrinker function with a number of inodes
888 * to trim from the LRU. Inodes to be freed are moved to a temporary list and
889 * then are freed outside inode_lock by dispose_list().
890 */
prune_icache_sb(struct super_block * sb,struct shrink_control * sc)891 long prune_icache_sb(struct super_block *sb, struct shrink_control *sc)
892 {
893 LIST_HEAD(freeable);
894 long freed;
895
896 freed = list_lru_shrink_walk(&sb->s_inode_lru, sc,
897 inode_lru_isolate, &freeable);
898 dispose_list(&freeable);
899 return freed;
900 }
901
902 static void __wait_on_freeing_inode(struct inode *inode);
903 /*
904 * Called with the inode lock held.
905 */
find_inode(struct super_block * sb,struct hlist_head * head,int (* test)(struct inode *,void *),void * data)906 static struct inode *find_inode(struct super_block *sb,
907 struct hlist_head *head,
908 int (*test)(struct inode *, void *),
909 void *data)
910 {
911 struct inode *inode = NULL;
912
913 repeat:
914 hlist_for_each_entry(inode, head, i_hash) {
915 if (inode->i_sb != sb)
916 continue;
917 if (!test(inode, data))
918 continue;
919 spin_lock(&inode->i_lock);
920 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
921 __wait_on_freeing_inode(inode);
922 goto repeat;
923 }
924 if (unlikely(inode->i_state & I_CREATING)) {
925 spin_unlock(&inode->i_lock);
926 return ERR_PTR(-ESTALE);
927 }
928 __iget(inode);
929 spin_unlock(&inode->i_lock);
930 return inode;
931 }
932 return NULL;
933 }
934
935 /*
936 * find_inode_fast is the fast path version of find_inode, see the comment at
937 * iget_locked for details.
938 */
find_inode_fast(struct super_block * sb,struct hlist_head * head,unsigned long ino)939 static struct inode *find_inode_fast(struct super_block *sb,
940 struct hlist_head *head, unsigned long ino)
941 {
942 struct inode *inode = NULL;
943
944 repeat:
945 hlist_for_each_entry(inode, head, i_hash) {
946 if (inode->i_ino != ino)
947 continue;
948 if (inode->i_sb != sb)
949 continue;
950 spin_lock(&inode->i_lock);
951 if (inode->i_state & (I_FREEING|I_WILL_FREE)) {
952 __wait_on_freeing_inode(inode);
953 goto repeat;
954 }
955 if (unlikely(inode->i_state & I_CREATING)) {
956 spin_unlock(&inode->i_lock);
957 return ERR_PTR(-ESTALE);
958 }
959 __iget(inode);
960 spin_unlock(&inode->i_lock);
961 return inode;
962 }
963 return NULL;
964 }
965
966 /*
967 * Each cpu owns a range of LAST_INO_BATCH numbers.
968 * 'shared_last_ino' is dirtied only once out of LAST_INO_BATCH allocations,
969 * to renew the exhausted range.
970 *
971 * This does not significantly increase overflow rate because every CPU can
972 * consume at most LAST_INO_BATCH-1 unused inode numbers. So there is
973 * NR_CPUS*(LAST_INO_BATCH-1) wastage. At 4096 and 1024, this is ~0.1% of the
974 * 2^32 range, and is a worst-case. Even a 50% wastage would only increase
975 * overflow rate by 2x, which does not seem too significant.
976 *
977 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
978 * error if st_ino won't fit in target struct field. Use 32bit counter
979 * here to attempt to avoid that.
980 */
981 #define LAST_INO_BATCH 1024
982 static DEFINE_PER_CPU(unsigned int, last_ino);
983
get_next_ino(void)984 unsigned int get_next_ino(void)
985 {
986 unsigned int *p = &get_cpu_var(last_ino);
987 unsigned int res = *p;
988
989 #ifdef CONFIG_SMP
990 if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
991 static atomic_t shared_last_ino;
992 int next = atomic_add_return(LAST_INO_BATCH, &shared_last_ino);
993
994 res = next - LAST_INO_BATCH;
995 }
996 #endif
997
998 res++;
999 /* get_next_ino should not provide a 0 inode number */
1000 if (unlikely(!res))
1001 res++;
1002 *p = res;
1003 put_cpu_var(last_ino);
1004 return res;
1005 }
1006 EXPORT_SYMBOL(get_next_ino);
1007
1008 /**
1009 * new_inode_pseudo - obtain an inode
1010 * @sb: superblock
1011 *
1012 * Allocates a new inode for given superblock.
1013 * Inode wont be chained in superblock s_inodes list
1014 * This means :
1015 * - fs can't be unmount
1016 * - quotas, fsnotify, writeback can't work
1017 */
new_inode_pseudo(struct super_block * sb)1018 struct inode *new_inode_pseudo(struct super_block *sb)
1019 {
1020 struct inode *inode = alloc_inode(sb);
1021
1022 if (inode) {
1023 spin_lock(&inode->i_lock);
1024 inode->i_state = 0;
1025 spin_unlock(&inode->i_lock);
1026 }
1027 return inode;
1028 }
1029
1030 /**
1031 * new_inode - obtain an inode
1032 * @sb: superblock
1033 *
1034 * Allocates a new inode for given superblock. The default gfp_mask
1035 * for allocations related to inode->i_mapping is GFP_HIGHUSER_MOVABLE.
1036 * If HIGHMEM pages are unsuitable or it is known that pages allocated
1037 * for the page cache are not reclaimable or migratable,
1038 * mapping_set_gfp_mask() must be called with suitable flags on the
1039 * newly created inode's mapping
1040 *
1041 */
new_inode(struct super_block * sb)1042 struct inode *new_inode(struct super_block *sb)
1043 {
1044 struct inode *inode;
1045
1046 spin_lock_prefetch(&sb->s_inode_list_lock);
1047
1048 inode = new_inode_pseudo(sb);
1049 if (inode)
1050 inode_sb_list_add(inode);
1051 return inode;
1052 }
1053 EXPORT_SYMBOL(new_inode);
1054
1055 #ifdef CONFIG_DEBUG_LOCK_ALLOC
lockdep_annotate_inode_mutex_key(struct inode * inode)1056 void lockdep_annotate_inode_mutex_key(struct inode *inode)
1057 {
1058 if (S_ISDIR(inode->i_mode)) {
1059 struct file_system_type *type = inode->i_sb->s_type;
1060
1061 /* Set new key only if filesystem hasn't already changed it */
1062 if (lockdep_match_class(&inode->i_rwsem, &type->i_mutex_key)) {
1063 /*
1064 * ensure nobody is actually holding i_mutex
1065 */
1066 // mutex_destroy(&inode->i_mutex);
1067 init_rwsem(&inode->i_rwsem);
1068 lockdep_set_class(&inode->i_rwsem,
1069 &type->i_mutex_dir_key);
1070 }
1071 }
1072 }
1073 EXPORT_SYMBOL(lockdep_annotate_inode_mutex_key);
1074 #endif
1075
1076 /**
1077 * unlock_new_inode - clear the I_NEW state and wake up any waiters
1078 * @inode: new inode to unlock
1079 *
1080 * Called when the inode is fully initialised to clear the new state of the
1081 * inode and wake up anyone waiting for the inode to finish initialisation.
1082 */
unlock_new_inode(struct inode * inode)1083 void unlock_new_inode(struct inode *inode)
1084 {
1085 lockdep_annotate_inode_mutex_key(inode);
1086 spin_lock(&inode->i_lock);
1087 WARN_ON(!(inode->i_state & I_NEW));
1088 inode->i_state &= ~I_NEW & ~I_CREATING;
1089 smp_mb();
1090 wake_up_bit(&inode->i_state, __I_NEW);
1091 spin_unlock(&inode->i_lock);
1092 }
1093 EXPORT_SYMBOL(unlock_new_inode);
1094
discard_new_inode(struct inode * inode)1095 void discard_new_inode(struct inode *inode)
1096 {
1097 lockdep_annotate_inode_mutex_key(inode);
1098 spin_lock(&inode->i_lock);
1099 WARN_ON(!(inode->i_state & I_NEW));
1100 inode->i_state &= ~I_NEW;
1101 smp_mb();
1102 wake_up_bit(&inode->i_state, __I_NEW);
1103 spin_unlock(&inode->i_lock);
1104 iput(inode);
1105 }
1106 EXPORT_SYMBOL(discard_new_inode);
1107
1108 /**
1109 * lock_two_inodes - lock two inodes (may be regular files but also dirs)
1110 *
1111 * Lock any non-NULL argument. The caller must make sure that if he is passing
1112 * in two directories, one is not ancestor of the other. Zero, one or two
1113 * objects may be locked by this function.
1114 *
1115 * @inode1: first inode to lock
1116 * @inode2: second inode to lock
1117 * @subclass1: inode lock subclass for the first lock obtained
1118 * @subclass2: inode lock subclass for the second lock obtained
1119 */
lock_two_inodes(struct inode * inode1,struct inode * inode2,unsigned subclass1,unsigned subclass2)1120 void lock_two_inodes(struct inode *inode1, struct inode *inode2,
1121 unsigned subclass1, unsigned subclass2)
1122 {
1123 if (!inode1 || !inode2) {
1124 /*
1125 * Make sure @subclass1 will be used for the acquired lock.
1126 * This is not strictly necessary (no current caller cares) but
1127 * let's keep things consistent.
1128 */
1129 if (!inode1)
1130 swap(inode1, inode2);
1131 goto lock;
1132 }
1133
1134 /*
1135 * If one object is directory and the other is not, we must make sure
1136 * to lock directory first as the other object may be its child.
1137 */
1138 if (S_ISDIR(inode2->i_mode) == S_ISDIR(inode1->i_mode)) {
1139 if (inode1 > inode2)
1140 swap(inode1, inode2);
1141 } else if (!S_ISDIR(inode1->i_mode))
1142 swap(inode1, inode2);
1143 lock:
1144 if (inode1)
1145 inode_lock_nested(inode1, subclass1);
1146 if (inode2 && inode2 != inode1)
1147 inode_lock_nested(inode2, subclass2);
1148 }
1149
1150 /**
1151 * lock_two_nondirectories - take two i_mutexes on non-directory objects
1152 *
1153 * Lock any non-NULL argument that is not a directory.
1154 * Zero, one or two objects may be locked by this function.
1155 *
1156 * @inode1: first inode to lock
1157 * @inode2: second inode to lock
1158 */
lock_two_nondirectories(struct inode * inode1,struct inode * inode2)1159 void lock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1160 {
1161 if (inode1 > inode2)
1162 swap(inode1, inode2);
1163
1164 if (inode1 && !S_ISDIR(inode1->i_mode))
1165 inode_lock(inode1);
1166 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1167 inode_lock_nested(inode2, I_MUTEX_NONDIR2);
1168 }
1169 EXPORT_SYMBOL(lock_two_nondirectories);
1170
1171 /**
1172 * unlock_two_nondirectories - release locks from lock_two_nondirectories()
1173 * @inode1: first inode to unlock
1174 * @inode2: second inode to unlock
1175 */
unlock_two_nondirectories(struct inode * inode1,struct inode * inode2)1176 void unlock_two_nondirectories(struct inode *inode1, struct inode *inode2)
1177 {
1178 if (inode1 && !S_ISDIR(inode1->i_mode))
1179 inode_unlock(inode1);
1180 if (inode2 && !S_ISDIR(inode2->i_mode) && inode2 != inode1)
1181 inode_unlock(inode2);
1182 }
1183 EXPORT_SYMBOL(unlock_two_nondirectories);
1184
1185 /**
1186 * inode_insert5 - obtain an inode from a mounted file system
1187 * @inode: pre-allocated inode to use for insert to cache
1188 * @hashval: hash value (usually inode number) to get
1189 * @test: callback used for comparisons between inodes
1190 * @set: callback used to initialize a new struct inode
1191 * @data: opaque data pointer to pass to @test and @set
1192 *
1193 * Search for the inode specified by @hashval and @data in the inode cache,
1194 * and if present it is return it with an increased reference count. This is
1195 * a variant of iget5_locked() for callers that don't want to fail on memory
1196 * allocation of inode.
1197 *
1198 * If the inode is not in cache, insert the pre-allocated inode to cache and
1199 * return it locked, hashed, and with the I_NEW flag set. The file system gets
1200 * to fill it in before unlocking it via unlock_new_inode().
1201 *
1202 * Note both @test and @set are called with the inode_hash_lock held, so can't
1203 * sleep.
1204 */
inode_insert5(struct inode * inode,unsigned long hashval,int (* test)(struct inode *,void *),int (* set)(struct inode *,void *),void * data)1205 struct inode *inode_insert5(struct inode *inode, unsigned long hashval,
1206 int (*test)(struct inode *, void *),
1207 int (*set)(struct inode *, void *), void *data)
1208 {
1209 struct hlist_head *head = inode_hashtable + hash(inode->i_sb, hashval);
1210 struct inode *old;
1211
1212 again:
1213 spin_lock(&inode_hash_lock);
1214 old = find_inode(inode->i_sb, head, test, data);
1215 if (unlikely(old)) {
1216 /*
1217 * Uhhuh, somebody else created the same inode under us.
1218 * Use the old inode instead of the preallocated one.
1219 */
1220 spin_unlock(&inode_hash_lock);
1221 if (IS_ERR(old))
1222 return NULL;
1223 wait_on_inode(old);
1224 if (unlikely(inode_unhashed(old))) {
1225 iput(old);
1226 goto again;
1227 }
1228 return old;
1229 }
1230
1231 if (set && unlikely(set(inode, data))) {
1232 inode = NULL;
1233 goto unlock;
1234 }
1235
1236 /*
1237 * Return the locked inode with I_NEW set, the
1238 * caller is responsible for filling in the contents
1239 */
1240 spin_lock(&inode->i_lock);
1241 inode->i_state |= I_NEW;
1242 hlist_add_head_rcu(&inode->i_hash, head);
1243 spin_unlock(&inode->i_lock);
1244
1245 /*
1246 * Add inode to the sb list if it's not already. It has I_NEW at this
1247 * point, so it should be safe to test i_sb_list locklessly.
1248 */
1249 if (list_empty(&inode->i_sb_list))
1250 inode_sb_list_add(inode);
1251 unlock:
1252 spin_unlock(&inode_hash_lock);
1253
1254 return inode;
1255 }
1256 EXPORT_SYMBOL(inode_insert5);
1257
1258 /**
1259 * iget5_locked - obtain an inode from a mounted file system
1260 * @sb: super block of file system
1261 * @hashval: hash value (usually inode number) to get
1262 * @test: callback used for comparisons between inodes
1263 * @set: callback used to initialize a new struct inode
1264 * @data: opaque data pointer to pass to @test and @set
1265 *
1266 * Search for the inode specified by @hashval and @data in the inode cache,
1267 * and if present it is return it with an increased reference count. This is
1268 * a generalized version of iget_locked() for file systems where the inode
1269 * number is not sufficient for unique identification of an inode.
1270 *
1271 * If the inode is not in cache, allocate a new inode and return it locked,
1272 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1273 * before unlocking it via unlock_new_inode().
1274 *
1275 * Note both @test and @set are called with the inode_hash_lock held, so can't
1276 * sleep.
1277 */
iget5_locked(struct super_block * sb,unsigned long hashval,int (* test)(struct inode *,void *),int (* set)(struct inode *,void *),void * data)1278 struct inode *iget5_locked(struct super_block *sb, unsigned long hashval,
1279 int (*test)(struct inode *, void *),
1280 int (*set)(struct inode *, void *), void *data)
1281 {
1282 struct inode *inode = ilookup5(sb, hashval, test, data);
1283
1284 if (!inode) {
1285 struct inode *new = alloc_inode(sb);
1286
1287 if (new) {
1288 new->i_state = 0;
1289 inode = inode_insert5(new, hashval, test, set, data);
1290 if (unlikely(inode != new))
1291 destroy_inode(new);
1292 }
1293 }
1294 return inode;
1295 }
1296 EXPORT_SYMBOL(iget5_locked);
1297
1298 /**
1299 * iget_locked - obtain an inode from a mounted file system
1300 * @sb: super block of file system
1301 * @ino: inode number to get
1302 *
1303 * Search for the inode specified by @ino in the inode cache and if present
1304 * return it with an increased reference count. This is for file systems
1305 * where the inode number is sufficient for unique identification of an inode.
1306 *
1307 * If the inode is not in cache, allocate a new inode and return it locked,
1308 * hashed, and with the I_NEW flag set. The file system gets to fill it in
1309 * before unlocking it via unlock_new_inode().
1310 */
iget_locked(struct super_block * sb,unsigned long ino)1311 struct inode *iget_locked(struct super_block *sb, unsigned long ino)
1312 {
1313 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1314 struct inode *inode;
1315 again:
1316 spin_lock(&inode_hash_lock);
1317 inode = find_inode_fast(sb, head, ino);
1318 spin_unlock(&inode_hash_lock);
1319 if (inode) {
1320 if (IS_ERR(inode))
1321 return NULL;
1322 wait_on_inode(inode);
1323 if (unlikely(inode_unhashed(inode))) {
1324 iput(inode);
1325 goto again;
1326 }
1327 return inode;
1328 }
1329
1330 inode = alloc_inode(sb);
1331 if (inode) {
1332 struct inode *old;
1333
1334 spin_lock(&inode_hash_lock);
1335 /* We released the lock, so.. */
1336 old = find_inode_fast(sb, head, ino);
1337 if (!old) {
1338 inode->i_ino = ino;
1339 spin_lock(&inode->i_lock);
1340 inode->i_state = I_NEW;
1341 hlist_add_head_rcu(&inode->i_hash, head);
1342 spin_unlock(&inode->i_lock);
1343 inode_sb_list_add(inode);
1344 spin_unlock(&inode_hash_lock);
1345
1346 /* Return the locked inode with I_NEW set, the
1347 * caller is responsible for filling in the contents
1348 */
1349 return inode;
1350 }
1351
1352 /*
1353 * Uhhuh, somebody else created the same inode under
1354 * us. Use the old inode instead of the one we just
1355 * allocated.
1356 */
1357 spin_unlock(&inode_hash_lock);
1358 destroy_inode(inode);
1359 if (IS_ERR(old))
1360 return NULL;
1361 inode = old;
1362 wait_on_inode(inode);
1363 if (unlikely(inode_unhashed(inode))) {
1364 iput(inode);
1365 goto again;
1366 }
1367 }
1368 return inode;
1369 }
1370 EXPORT_SYMBOL(iget_locked);
1371
1372 /*
1373 * search the inode cache for a matching inode number.
1374 * If we find one, then the inode number we are trying to
1375 * allocate is not unique and so we should not use it.
1376 *
1377 * Returns 1 if the inode number is unique, 0 if it is not.
1378 */
test_inode_iunique(struct super_block * sb,unsigned long ino)1379 static int test_inode_iunique(struct super_block *sb, unsigned long ino)
1380 {
1381 struct hlist_head *b = inode_hashtable + hash(sb, ino);
1382 struct inode *inode;
1383
1384 hlist_for_each_entry_rcu(inode, b, i_hash) {
1385 if (inode->i_ino == ino && inode->i_sb == sb)
1386 return 0;
1387 }
1388 return 1;
1389 }
1390
1391 /**
1392 * iunique - get a unique inode number
1393 * @sb: superblock
1394 * @max_reserved: highest reserved inode number
1395 *
1396 * Obtain an inode number that is unique on the system for a given
1397 * superblock. This is used by file systems that have no natural
1398 * permanent inode numbering system. An inode number is returned that
1399 * is higher than the reserved limit but unique.
1400 *
1401 * BUGS:
1402 * With a large number of inodes live on the file system this function
1403 * currently becomes quite slow.
1404 */
iunique(struct super_block * sb,ino_t max_reserved)1405 ino_t iunique(struct super_block *sb, ino_t max_reserved)
1406 {
1407 /*
1408 * On a 32bit, non LFS stat() call, glibc will generate an EOVERFLOW
1409 * error if st_ino won't fit in target struct field. Use 32bit counter
1410 * here to attempt to avoid that.
1411 */
1412 static DEFINE_SPINLOCK(iunique_lock);
1413 static unsigned int counter;
1414 ino_t res;
1415
1416 rcu_read_lock();
1417 spin_lock(&iunique_lock);
1418 do {
1419 if (counter <= max_reserved)
1420 counter = max_reserved + 1;
1421 res = counter++;
1422 } while (!test_inode_iunique(sb, res));
1423 spin_unlock(&iunique_lock);
1424 rcu_read_unlock();
1425
1426 return res;
1427 }
1428 EXPORT_SYMBOL(iunique);
1429
igrab(struct inode * inode)1430 struct inode *igrab(struct inode *inode)
1431 {
1432 spin_lock(&inode->i_lock);
1433 if (!(inode->i_state & (I_FREEING|I_WILL_FREE))) {
1434 __iget(inode);
1435 spin_unlock(&inode->i_lock);
1436 } else {
1437 spin_unlock(&inode->i_lock);
1438 /*
1439 * Handle the case where s_op->clear_inode is not been
1440 * called yet, and somebody is calling igrab
1441 * while the inode is getting freed.
1442 */
1443 inode = NULL;
1444 }
1445 return inode;
1446 }
1447 EXPORT_SYMBOL(igrab);
1448
1449 /**
1450 * ilookup5_nowait - search for an inode in the inode cache
1451 * @sb: super block of file system to search
1452 * @hashval: hash value (usually inode number) to search for
1453 * @test: callback used for comparisons between inodes
1454 * @data: opaque data pointer to pass to @test
1455 *
1456 * Search for the inode specified by @hashval and @data in the inode cache.
1457 * If the inode is in the cache, the inode is returned with an incremented
1458 * reference count.
1459 *
1460 * Note: I_NEW is not waited upon so you have to be very careful what you do
1461 * with the returned inode. You probably should be using ilookup5() instead.
1462 *
1463 * Note2: @test is called with the inode_hash_lock held, so can't sleep.
1464 */
ilookup5_nowait(struct super_block * sb,unsigned long hashval,int (* test)(struct inode *,void *),void * data)1465 struct inode *ilookup5_nowait(struct super_block *sb, unsigned long hashval,
1466 int (*test)(struct inode *, void *), void *data)
1467 {
1468 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1469 struct inode *inode;
1470
1471 spin_lock(&inode_hash_lock);
1472 inode = find_inode(sb, head, test, data);
1473 spin_unlock(&inode_hash_lock);
1474
1475 return IS_ERR(inode) ? NULL : inode;
1476 }
1477 EXPORT_SYMBOL(ilookup5_nowait);
1478
1479 /**
1480 * ilookup5 - search for an inode in the inode cache
1481 * @sb: super block of file system to search
1482 * @hashval: hash value (usually inode number) to search for
1483 * @test: callback used for comparisons between inodes
1484 * @data: opaque data pointer to pass to @test
1485 *
1486 * Search for the inode specified by @hashval and @data in the inode cache,
1487 * and if the inode is in the cache, return the inode with an incremented
1488 * reference count. Waits on I_NEW before returning the inode.
1489 * returned with an incremented reference count.
1490 *
1491 * This is a generalized version of ilookup() for file systems where the
1492 * inode number is not sufficient for unique identification of an inode.
1493 *
1494 * Note: @test is called with the inode_hash_lock held, so can't sleep.
1495 */
ilookup5(struct super_block * sb,unsigned long hashval,int (* test)(struct inode *,void *),void * data)1496 struct inode *ilookup5(struct super_block *sb, unsigned long hashval,
1497 int (*test)(struct inode *, void *), void *data)
1498 {
1499 struct inode *inode;
1500 again:
1501 inode = ilookup5_nowait(sb, hashval, test, data);
1502 if (inode) {
1503 wait_on_inode(inode);
1504 if (unlikely(inode_unhashed(inode))) {
1505 iput(inode);
1506 goto again;
1507 }
1508 }
1509 return inode;
1510 }
1511 EXPORT_SYMBOL(ilookup5);
1512
1513 /**
1514 * ilookup - search for an inode in the inode cache
1515 * @sb: super block of file system to search
1516 * @ino: inode number to search for
1517 *
1518 * Search for the inode @ino in the inode cache, and if the inode is in the
1519 * cache, the inode is returned with an incremented reference count.
1520 */
ilookup(struct super_block * sb,unsigned long ino)1521 struct inode *ilookup(struct super_block *sb, unsigned long ino)
1522 {
1523 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1524 struct inode *inode;
1525 again:
1526 spin_lock(&inode_hash_lock);
1527 inode = find_inode_fast(sb, head, ino);
1528 spin_unlock(&inode_hash_lock);
1529
1530 if (inode) {
1531 if (IS_ERR(inode))
1532 return NULL;
1533 wait_on_inode(inode);
1534 if (unlikely(inode_unhashed(inode))) {
1535 iput(inode);
1536 goto again;
1537 }
1538 }
1539 return inode;
1540 }
1541 EXPORT_SYMBOL(ilookup);
1542
1543 /**
1544 * find_inode_nowait - find an inode in the inode cache
1545 * @sb: super block of file system to search
1546 * @hashval: hash value (usually inode number) to search for
1547 * @match: callback used for comparisons between inodes
1548 * @data: opaque data pointer to pass to @match
1549 *
1550 * Search for the inode specified by @hashval and @data in the inode
1551 * cache, where the helper function @match will return 0 if the inode
1552 * does not match, 1 if the inode does match, and -1 if the search
1553 * should be stopped. The @match function must be responsible for
1554 * taking the i_lock spin_lock and checking i_state for an inode being
1555 * freed or being initialized, and incrementing the reference count
1556 * before returning 1. It also must not sleep, since it is called with
1557 * the inode_hash_lock spinlock held.
1558 *
1559 * This is a even more generalized version of ilookup5() when the
1560 * function must never block --- find_inode() can block in
1561 * __wait_on_freeing_inode() --- or when the caller can not increment
1562 * the reference count because the resulting iput() might cause an
1563 * inode eviction. The tradeoff is that the @match funtion must be
1564 * very carefully implemented.
1565 */
find_inode_nowait(struct super_block * sb,unsigned long hashval,int (* match)(struct inode *,unsigned long,void *),void * data)1566 struct inode *find_inode_nowait(struct super_block *sb,
1567 unsigned long hashval,
1568 int (*match)(struct inode *, unsigned long,
1569 void *),
1570 void *data)
1571 {
1572 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1573 struct inode *inode, *ret_inode = NULL;
1574 int mval;
1575
1576 spin_lock(&inode_hash_lock);
1577 hlist_for_each_entry(inode, head, i_hash) {
1578 if (inode->i_sb != sb)
1579 continue;
1580 mval = match(inode, hashval, data);
1581 if (mval == 0)
1582 continue;
1583 if (mval == 1)
1584 ret_inode = inode;
1585 goto out;
1586 }
1587 out:
1588 spin_unlock(&inode_hash_lock);
1589 return ret_inode;
1590 }
1591 EXPORT_SYMBOL(find_inode_nowait);
1592
1593 /**
1594 * find_inode_rcu - find an inode in the inode cache
1595 * @sb: Super block of file system to search
1596 * @hashval: Key to hash
1597 * @test: Function to test match on an inode
1598 * @data: Data for test function
1599 *
1600 * Search for the inode specified by @hashval and @data in the inode cache,
1601 * where the helper function @test will return 0 if the inode does not match
1602 * and 1 if it does. The @test function must be responsible for taking the
1603 * i_lock spin_lock and checking i_state for an inode being freed or being
1604 * initialized.
1605 *
1606 * If successful, this will return the inode for which the @test function
1607 * returned 1 and NULL otherwise.
1608 *
1609 * The @test function is not permitted to take a ref on any inode presented.
1610 * It is also not permitted to sleep.
1611 *
1612 * The caller must hold the RCU read lock.
1613 */
find_inode_rcu(struct super_block * sb,unsigned long hashval,int (* test)(struct inode *,void *),void * data)1614 struct inode *find_inode_rcu(struct super_block *sb, unsigned long hashval,
1615 int (*test)(struct inode *, void *), void *data)
1616 {
1617 struct hlist_head *head = inode_hashtable + hash(sb, hashval);
1618 struct inode *inode;
1619
1620 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1621 "suspicious find_inode_rcu() usage");
1622
1623 hlist_for_each_entry_rcu(inode, head, i_hash) {
1624 if (inode->i_sb == sb &&
1625 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)) &&
1626 test(inode, data))
1627 return inode;
1628 }
1629 return NULL;
1630 }
1631 EXPORT_SYMBOL(find_inode_rcu);
1632
1633 /**
1634 * find_inode_by_ino_rcu - Find an inode in the inode cache
1635 * @sb: Super block of file system to search
1636 * @ino: The inode number to match
1637 *
1638 * Search for the inode specified by @hashval and @data in the inode cache,
1639 * where the helper function @test will return 0 if the inode does not match
1640 * and 1 if it does. The @test function must be responsible for taking the
1641 * i_lock spin_lock and checking i_state for an inode being freed or being
1642 * initialized.
1643 *
1644 * If successful, this will return the inode for which the @test function
1645 * returned 1 and NULL otherwise.
1646 *
1647 * The @test function is not permitted to take a ref on any inode presented.
1648 * It is also not permitted to sleep.
1649 *
1650 * The caller must hold the RCU read lock.
1651 */
find_inode_by_ino_rcu(struct super_block * sb,unsigned long ino)1652 struct inode *find_inode_by_ino_rcu(struct super_block *sb,
1653 unsigned long ino)
1654 {
1655 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1656 struct inode *inode;
1657
1658 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
1659 "suspicious find_inode_by_ino_rcu() usage");
1660
1661 hlist_for_each_entry_rcu(inode, head, i_hash) {
1662 if (inode->i_ino == ino &&
1663 inode->i_sb == sb &&
1664 !(READ_ONCE(inode->i_state) & (I_FREEING | I_WILL_FREE)))
1665 return inode;
1666 }
1667 return NULL;
1668 }
1669 EXPORT_SYMBOL(find_inode_by_ino_rcu);
1670
insert_inode_locked(struct inode * inode)1671 int insert_inode_locked(struct inode *inode)
1672 {
1673 struct super_block *sb = inode->i_sb;
1674 ino_t ino = inode->i_ino;
1675 struct hlist_head *head = inode_hashtable + hash(sb, ino);
1676
1677 while (1) {
1678 struct inode *old = NULL;
1679 spin_lock(&inode_hash_lock);
1680 hlist_for_each_entry(old, head, i_hash) {
1681 if (old->i_ino != ino)
1682 continue;
1683 if (old->i_sb != sb)
1684 continue;
1685 spin_lock(&old->i_lock);
1686 if (old->i_state & (I_FREEING|I_WILL_FREE)) {
1687 spin_unlock(&old->i_lock);
1688 continue;
1689 }
1690 break;
1691 }
1692 if (likely(!old)) {
1693 spin_lock(&inode->i_lock);
1694 inode->i_state |= I_NEW | I_CREATING;
1695 hlist_add_head_rcu(&inode->i_hash, head);
1696 spin_unlock(&inode->i_lock);
1697 spin_unlock(&inode_hash_lock);
1698 return 0;
1699 }
1700 if (unlikely(old->i_state & I_CREATING)) {
1701 spin_unlock(&old->i_lock);
1702 spin_unlock(&inode_hash_lock);
1703 return -EBUSY;
1704 }
1705 __iget(old);
1706 spin_unlock(&old->i_lock);
1707 spin_unlock(&inode_hash_lock);
1708 wait_on_inode(old);
1709 if (unlikely(!inode_unhashed(old))) {
1710 iput(old);
1711 return -EBUSY;
1712 }
1713 iput(old);
1714 }
1715 }
1716 EXPORT_SYMBOL(insert_inode_locked);
1717
insert_inode_locked4(struct inode * inode,unsigned long hashval,int (* test)(struct inode *,void *),void * data)1718 int insert_inode_locked4(struct inode *inode, unsigned long hashval,
1719 int (*test)(struct inode *, void *), void *data)
1720 {
1721 struct inode *old;
1722
1723 inode->i_state |= I_CREATING;
1724 old = inode_insert5(inode, hashval, test, NULL, data);
1725
1726 if (old != inode) {
1727 iput(old);
1728 return -EBUSY;
1729 }
1730 return 0;
1731 }
1732 EXPORT_SYMBOL(insert_inode_locked4);
1733
1734
generic_delete_inode(struct inode * inode)1735 int generic_delete_inode(struct inode *inode)
1736 {
1737 return 1;
1738 }
1739 EXPORT_SYMBOL(generic_delete_inode);
1740
1741 /*
1742 * Called when we're dropping the last reference
1743 * to an inode.
1744 *
1745 * Call the FS "drop_inode()" function, defaulting to
1746 * the legacy UNIX filesystem behaviour. If it tells
1747 * us to evict inode, do so. Otherwise, retain inode
1748 * in cache if fs is alive, sync and evict if fs is
1749 * shutting down.
1750 */
iput_final(struct inode * inode)1751 static void iput_final(struct inode *inode)
1752 {
1753 struct super_block *sb = inode->i_sb;
1754 const struct super_operations *op = inode->i_sb->s_op;
1755 unsigned long state;
1756 int drop;
1757
1758 WARN_ON(inode->i_state & I_NEW);
1759
1760 if (op->drop_inode)
1761 drop = op->drop_inode(inode);
1762 else
1763 drop = generic_drop_inode(inode);
1764
1765 if (!drop &&
1766 !(inode->i_state & I_DONTCACHE) &&
1767 (sb->s_flags & SB_ACTIVE)) {
1768 __inode_add_lru(inode, true);
1769 spin_unlock(&inode->i_lock);
1770 return;
1771 }
1772
1773 state = inode->i_state;
1774 if (!drop) {
1775 WRITE_ONCE(inode->i_state, state | I_WILL_FREE);
1776 spin_unlock(&inode->i_lock);
1777
1778 write_inode_now(inode, 1);
1779
1780 spin_lock(&inode->i_lock);
1781 state = inode->i_state;
1782 WARN_ON(state & I_NEW);
1783 state &= ~I_WILL_FREE;
1784 }
1785
1786 WRITE_ONCE(inode->i_state, state | I_FREEING);
1787 if (!list_empty(&inode->i_lru))
1788 inode_lru_list_del(inode);
1789 spin_unlock(&inode->i_lock);
1790
1791 evict(inode);
1792 }
1793
1794 /**
1795 * iput - put an inode
1796 * @inode: inode to put
1797 *
1798 * Puts an inode, dropping its usage count. If the inode use count hits
1799 * zero, the inode is then freed and may also be destroyed.
1800 *
1801 * Consequently, iput() can sleep.
1802 */
iput(struct inode * inode)1803 void iput(struct inode *inode)
1804 {
1805 if (!inode)
1806 return;
1807 BUG_ON(inode->i_state & I_CLEAR);
1808 retry:
1809 if (atomic_dec_and_lock(&inode->i_count, &inode->i_lock)) {
1810 if (inode->i_nlink && (inode->i_state & I_DIRTY_TIME)) {
1811 atomic_inc(&inode->i_count);
1812 spin_unlock(&inode->i_lock);
1813 trace_writeback_lazytime_iput(inode);
1814 mark_inode_dirty_sync(inode);
1815 goto retry;
1816 }
1817 iput_final(inode);
1818 }
1819 }
1820 EXPORT_SYMBOL(iput);
1821
1822 #ifdef CONFIG_BLOCK
1823 /**
1824 * bmap - find a block number in a file
1825 * @inode: inode owning the block number being requested
1826 * @block: pointer containing the block to find
1827 *
1828 * Replaces the value in ``*block`` with the block number on the device holding
1829 * corresponding to the requested block number in the file.
1830 * That is, asked for block 4 of inode 1 the function will replace the
1831 * 4 in ``*block``, with disk block relative to the disk start that holds that
1832 * block of the file.
1833 *
1834 * Returns -EINVAL in case of error, 0 otherwise. If mapping falls into a
1835 * hole, returns 0 and ``*block`` is also set to 0.
1836 */
bmap(struct inode * inode,sector_t * block)1837 int bmap(struct inode *inode, sector_t *block)
1838 {
1839 if (!inode->i_mapping->a_ops->bmap)
1840 return -EINVAL;
1841
1842 *block = inode->i_mapping->a_ops->bmap(inode->i_mapping, *block);
1843 return 0;
1844 }
1845 EXPORT_SYMBOL(bmap);
1846 #endif
1847
1848 /*
1849 * With relative atime, only update atime if the previous atime is
1850 * earlier than either the ctime or mtime or if at least a day has
1851 * passed since the last atime update.
1852 */
relatime_need_update(struct vfsmount * mnt,struct inode * inode,struct timespec64 now)1853 static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
1854 struct timespec64 now)
1855 {
1856
1857 if (!(mnt->mnt_flags & MNT_RELATIME))
1858 return 1;
1859 /*
1860 * Is mtime younger than atime? If yes, update atime:
1861 */
1862 if (timespec64_compare(&inode->i_mtime, &inode->i_atime) >= 0)
1863 return 1;
1864 /*
1865 * Is ctime younger than atime? If yes, update atime:
1866 */
1867 if (timespec64_compare(&inode->i_ctime, &inode->i_atime) >= 0)
1868 return 1;
1869
1870 /*
1871 * Is the previous atime value older than a day? If yes,
1872 * update atime:
1873 */
1874 if ((long)(now.tv_sec - inode->i_atime.tv_sec) >= 24*60*60)
1875 return 1;
1876 /*
1877 * Good, we can skip the atime update:
1878 */
1879 return 0;
1880 }
1881
generic_update_time(struct inode * inode,struct timespec64 * time,int flags)1882 int generic_update_time(struct inode *inode, struct timespec64 *time, int flags)
1883 {
1884 int dirty_flags = 0;
1885
1886 if (flags & (S_ATIME | S_CTIME | S_MTIME)) {
1887 if (flags & S_ATIME)
1888 inode->i_atime = *time;
1889 if (flags & S_CTIME)
1890 inode->i_ctime = *time;
1891 if (flags & S_MTIME)
1892 inode->i_mtime = *time;
1893
1894 if (inode->i_sb->s_flags & SB_LAZYTIME)
1895 dirty_flags |= I_DIRTY_TIME;
1896 else
1897 dirty_flags |= I_DIRTY_SYNC;
1898 }
1899
1900 if ((flags & S_VERSION) && inode_maybe_inc_iversion(inode, false))
1901 dirty_flags |= I_DIRTY_SYNC;
1902
1903 __mark_inode_dirty(inode, dirty_flags);
1904 return 0;
1905 }
1906 EXPORT_SYMBOL(generic_update_time);
1907
1908 /*
1909 * This does the actual work of updating an inodes time or version. Must have
1910 * had called mnt_want_write() before calling this.
1911 */
inode_update_time(struct inode * inode,struct timespec64 * time,int flags)1912 int inode_update_time(struct inode *inode, struct timespec64 *time, int flags)
1913 {
1914 if (inode->i_op->update_time)
1915 return inode->i_op->update_time(inode, time, flags);
1916 return generic_update_time(inode, time, flags);
1917 }
1918 EXPORT_SYMBOL(inode_update_time);
1919
1920 /**
1921 * atime_needs_update - update the access time
1922 * @path: the &struct path to update
1923 * @inode: inode to update
1924 *
1925 * Update the accessed time on an inode and mark it for writeback.
1926 * This function automatically handles read only file systems and media,
1927 * as well as the "noatime" flag and inode specific "noatime" markers.
1928 */
atime_needs_update(const struct path * path,struct inode * inode)1929 bool atime_needs_update(const struct path *path, struct inode *inode)
1930 {
1931 struct vfsmount *mnt = path->mnt;
1932 struct timespec64 now;
1933
1934 if (inode->i_flags & S_NOATIME)
1935 return false;
1936
1937 /* Atime updates will likely cause i_uid and i_gid to be written
1938 * back improprely if their true value is unknown to the vfs.
1939 */
1940 if (HAS_UNMAPPED_ID(mnt_user_ns(mnt), inode))
1941 return false;
1942
1943 if (IS_NOATIME(inode))
1944 return false;
1945 if ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))
1946 return false;
1947
1948 if (mnt->mnt_flags & MNT_NOATIME)
1949 return false;
1950 if ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))
1951 return false;
1952
1953 now = current_time(inode);
1954
1955 if (!relatime_need_update(mnt, inode, now))
1956 return false;
1957
1958 if (timespec64_equal(&inode->i_atime, &now))
1959 return false;
1960
1961 return true;
1962 }
1963
touch_atime(const struct path * path)1964 void touch_atime(const struct path *path)
1965 {
1966 struct vfsmount *mnt = path->mnt;
1967 struct inode *inode = d_inode(path->dentry);
1968 struct timespec64 now;
1969
1970 if (!atime_needs_update(path, inode))
1971 return;
1972
1973 if (!sb_start_write_trylock(inode->i_sb))
1974 return;
1975
1976 if (__mnt_want_write(mnt) != 0)
1977 goto skip_update;
1978 /*
1979 * File systems can error out when updating inodes if they need to
1980 * allocate new space to modify an inode (such is the case for
1981 * Btrfs), but since we touch atime while walking down the path we
1982 * really don't care if we failed to update the atime of the file,
1983 * so just ignore the return value.
1984 * We may also fail on filesystems that have the ability to make parts
1985 * of the fs read only, e.g. subvolumes in Btrfs.
1986 */
1987 now = current_time(inode);
1988 inode_update_time(inode, &now, S_ATIME);
1989 __mnt_drop_write(mnt);
1990 skip_update:
1991 sb_end_write(inode->i_sb);
1992 }
1993 EXPORT_SYMBOL(touch_atime);
1994
1995 /*
1996 * Return mask of changes for notify_change() that need to be done as a
1997 * response to write or truncate. Return 0 if nothing has to be changed.
1998 * Negative value on error (change should be denied).
1999 */
dentry_needs_remove_privs(struct user_namespace * mnt_userns,struct dentry * dentry)2000 int dentry_needs_remove_privs(struct user_namespace *mnt_userns,
2001 struct dentry *dentry)
2002 {
2003 struct inode *inode = d_inode(dentry);
2004 int mask = 0;
2005 int ret;
2006
2007 if (IS_NOSEC(inode))
2008 return 0;
2009
2010 mask = setattr_should_drop_suidgid(mnt_userns, inode);
2011 ret = security_inode_need_killpriv(dentry);
2012 if (ret < 0)
2013 return ret;
2014 if (ret)
2015 mask |= ATTR_KILL_PRIV;
2016 return mask;
2017 }
2018
__remove_privs(struct user_namespace * mnt_userns,struct dentry * dentry,int kill)2019 static int __remove_privs(struct user_namespace *mnt_userns,
2020 struct dentry *dentry, int kill)
2021 {
2022 struct iattr newattrs;
2023
2024 newattrs.ia_valid = ATTR_FORCE | kill;
2025 /*
2026 * Note we call this on write, so notify_change will not
2027 * encounter any conflicting delegations:
2028 */
2029 return notify_change(mnt_userns, dentry, &newattrs, NULL);
2030 }
2031
__file_remove_privs(struct file * file,unsigned int flags)2032 static int __file_remove_privs(struct file *file, unsigned int flags)
2033 {
2034 struct dentry *dentry = file_dentry(file);
2035 struct inode *inode = file_inode(file);
2036 int error = 0;
2037 int kill;
2038
2039 if (IS_NOSEC(inode) || !S_ISREG(inode->i_mode))
2040 return 0;
2041
2042 kill = dentry_needs_remove_privs(file_mnt_user_ns(file), dentry);
2043 if (kill < 0)
2044 return kill;
2045
2046 if (kill) {
2047 if (flags & IOCB_NOWAIT)
2048 return -EAGAIN;
2049
2050 error = __remove_privs(file_mnt_user_ns(file), dentry, kill);
2051 }
2052
2053 if (!error)
2054 inode_has_no_xattr(inode);
2055 return error;
2056 }
2057
2058 /**
2059 * file_remove_privs - remove special file privileges (suid, capabilities)
2060 * @file: file to remove privileges from
2061 *
2062 * When file is modified by a write or truncation ensure that special
2063 * file privileges are removed.
2064 *
2065 * Return: 0 on success, negative errno on failure.
2066 */
file_remove_privs(struct file * file)2067 int file_remove_privs(struct file *file)
2068 {
2069 return __file_remove_privs(file, 0);
2070 }
2071 EXPORT_SYMBOL(file_remove_privs);
2072
inode_needs_update_time(struct inode * inode,struct timespec64 * now)2073 static int inode_needs_update_time(struct inode *inode, struct timespec64 *now)
2074 {
2075 int sync_it = 0;
2076
2077 /* First try to exhaust all avenues to not sync */
2078 if (IS_NOCMTIME(inode))
2079 return 0;
2080
2081 if (!timespec64_equal(&inode->i_mtime, now))
2082 sync_it = S_MTIME;
2083
2084 if (!timespec64_equal(&inode->i_ctime, now))
2085 sync_it |= S_CTIME;
2086
2087 if (IS_I_VERSION(inode) && inode_iversion_need_inc(inode))
2088 sync_it |= S_VERSION;
2089
2090 if (!sync_it)
2091 return 0;
2092
2093 return sync_it;
2094 }
2095
__file_update_time(struct file * file,struct timespec64 * now,int sync_mode)2096 static int __file_update_time(struct file *file, struct timespec64 *now,
2097 int sync_mode)
2098 {
2099 int ret = 0;
2100 struct inode *inode = file_inode(file);
2101
2102 /* try to update time settings */
2103 if (!__mnt_want_write_file(file)) {
2104 ret = inode_update_time(inode, now, sync_mode);
2105 __mnt_drop_write_file(file);
2106 }
2107
2108 return ret;
2109 }
2110
2111 /**
2112 * file_update_time - update mtime and ctime time
2113 * @file: file accessed
2114 *
2115 * Update the mtime and ctime members of an inode and mark the inode for
2116 * writeback. Note that this function is meant exclusively for usage in
2117 * the file write path of filesystems, and filesystems may choose to
2118 * explicitly ignore updates via this function with the _NOCMTIME inode
2119 * flag, e.g. for network filesystem where these imestamps are handled
2120 * by the server. This can return an error for file systems who need to
2121 * allocate space in order to update an inode.
2122 *
2123 * Return: 0 on success, negative errno on failure.
2124 */
file_update_time(struct file * file)2125 int file_update_time(struct file *file)
2126 {
2127 int ret;
2128 struct inode *inode = file_inode(file);
2129 struct timespec64 now = current_time(inode);
2130
2131 ret = inode_needs_update_time(inode, &now);
2132 if (ret <= 0)
2133 return ret;
2134
2135 return __file_update_time(file, &now, ret);
2136 }
2137 EXPORT_SYMBOL(file_update_time);
2138
2139 /**
2140 * file_modified_flags - handle mandated vfs changes when modifying a file
2141 * @file: file that was modified
2142 * @flags: kiocb flags
2143 *
2144 * When file has been modified ensure that special
2145 * file privileges are removed and time settings are updated.
2146 *
2147 * If IOCB_NOWAIT is set, special file privileges will not be removed and
2148 * time settings will not be updated. It will return -EAGAIN.
2149 *
2150 * Context: Caller must hold the file's inode lock.
2151 *
2152 * Return: 0 on success, negative errno on failure.
2153 */
file_modified_flags(struct file * file,int flags)2154 static int file_modified_flags(struct file *file, int flags)
2155 {
2156 int ret;
2157 struct inode *inode = file_inode(file);
2158 struct timespec64 now = current_time(inode);
2159
2160 /*
2161 * Clear the security bits if the process is not being run by root.
2162 * This keeps people from modifying setuid and setgid binaries.
2163 */
2164 ret = __file_remove_privs(file, flags);
2165 if (ret)
2166 return ret;
2167
2168 if (unlikely(file->f_mode & FMODE_NOCMTIME))
2169 return 0;
2170
2171 ret = inode_needs_update_time(inode, &now);
2172 if (ret <= 0)
2173 return ret;
2174 if (flags & IOCB_NOWAIT)
2175 return -EAGAIN;
2176
2177 return __file_update_time(file, &now, ret);
2178 }
2179
2180 /**
2181 * file_modified - handle mandated vfs changes when modifying a file
2182 * @file: file that was modified
2183 *
2184 * When file has been modified ensure that special
2185 * file privileges are removed and time settings are updated.
2186 *
2187 * Context: Caller must hold the file's inode lock.
2188 *
2189 * Return: 0 on success, negative errno on failure.
2190 */
file_modified(struct file * file)2191 int file_modified(struct file *file)
2192 {
2193 return file_modified_flags(file, 0);
2194 }
2195 EXPORT_SYMBOL(file_modified);
2196
2197 /**
2198 * kiocb_modified - handle mandated vfs changes when modifying a file
2199 * @iocb: iocb that was modified
2200 *
2201 * When file has been modified ensure that special
2202 * file privileges are removed and time settings are updated.
2203 *
2204 * Context: Caller must hold the file's inode lock.
2205 *
2206 * Return: 0 on success, negative errno on failure.
2207 */
kiocb_modified(struct kiocb * iocb)2208 int kiocb_modified(struct kiocb *iocb)
2209 {
2210 return file_modified_flags(iocb->ki_filp, iocb->ki_flags);
2211 }
2212 EXPORT_SYMBOL_GPL(kiocb_modified);
2213
inode_needs_sync(struct inode * inode)2214 int inode_needs_sync(struct inode *inode)
2215 {
2216 if (IS_SYNC(inode))
2217 return 1;
2218 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
2219 return 1;
2220 return 0;
2221 }
2222 EXPORT_SYMBOL(inode_needs_sync);
2223
2224 /*
2225 * If we try to find an inode in the inode hash while it is being
2226 * deleted, we have to wait until the filesystem completes its
2227 * deletion before reporting that it isn't found. This function waits
2228 * until the deletion _might_ have completed. Callers are responsible
2229 * to recheck inode state.
2230 *
2231 * It doesn't matter if I_NEW is not set initially, a call to
2232 * wake_up_bit(&inode->i_state, __I_NEW) after removing from the hash list
2233 * will DTRT.
2234 */
__wait_on_freeing_inode(struct inode * inode)2235 static void __wait_on_freeing_inode(struct inode *inode)
2236 {
2237 wait_queue_head_t *wq;
2238 DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
2239 wq = bit_waitqueue(&inode->i_state, __I_NEW);
2240 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2241 spin_unlock(&inode->i_lock);
2242 spin_unlock(&inode_hash_lock);
2243 schedule();
2244 finish_wait(wq, &wait.wq_entry);
2245 spin_lock(&inode_hash_lock);
2246 }
2247
2248 static __initdata unsigned long ihash_entries;
set_ihash_entries(char * str)2249 static int __init set_ihash_entries(char *str)
2250 {
2251 if (!str)
2252 return 0;
2253 ihash_entries = simple_strtoul(str, &str, 0);
2254 return 1;
2255 }
2256 __setup("ihash_entries=", set_ihash_entries);
2257
2258 /*
2259 * Initialize the waitqueues and inode hash table.
2260 */
inode_init_early(void)2261 void __init inode_init_early(void)
2262 {
2263 /* If hashes are distributed across NUMA nodes, defer
2264 * hash allocation until vmalloc space is available.
2265 */
2266 if (hashdist)
2267 return;
2268
2269 inode_hashtable =
2270 alloc_large_system_hash("Inode-cache",
2271 sizeof(struct hlist_head),
2272 ihash_entries,
2273 14,
2274 HASH_EARLY | HASH_ZERO,
2275 &i_hash_shift,
2276 &i_hash_mask,
2277 0,
2278 0);
2279 }
2280
inode_init(void)2281 void __init inode_init(void)
2282 {
2283 /* inode slab cache */
2284 inode_cachep = kmem_cache_create("inode_cache",
2285 sizeof(struct inode),
2286 0,
2287 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
2288 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2289 init_once);
2290
2291 /* Hash may have been set up in inode_init_early */
2292 if (!hashdist)
2293 return;
2294
2295 inode_hashtable =
2296 alloc_large_system_hash("Inode-cache",
2297 sizeof(struct hlist_head),
2298 ihash_entries,
2299 14,
2300 HASH_ZERO,
2301 &i_hash_shift,
2302 &i_hash_mask,
2303 0,
2304 0);
2305 }
2306
init_special_inode(struct inode * inode,umode_t mode,dev_t rdev)2307 void init_special_inode(struct inode *inode, umode_t mode, dev_t rdev)
2308 {
2309 inode->i_mode = mode;
2310 if (S_ISCHR(mode)) {
2311 inode->i_fop = &def_chr_fops;
2312 inode->i_rdev = rdev;
2313 } else if (S_ISBLK(mode)) {
2314 inode->i_fop = &def_blk_fops;
2315 inode->i_rdev = rdev;
2316 } else if (S_ISFIFO(mode))
2317 inode->i_fop = &pipefifo_fops;
2318 else if (S_ISSOCK(mode))
2319 ; /* leave it no_open_fops */
2320 else
2321 printk(KERN_DEBUG "init_special_inode: bogus i_mode (%o) for"
2322 " inode %s:%lu\n", mode, inode->i_sb->s_id,
2323 inode->i_ino);
2324 }
2325 EXPORT_SYMBOL(init_special_inode);
2326
2327 /**
2328 * inode_init_owner - Init uid,gid,mode for new inode according to posix standards
2329 * @mnt_userns: User namespace of the mount the inode was created from
2330 * @inode: New inode
2331 * @dir: Directory inode
2332 * @mode: mode of the new inode
2333 *
2334 * If the inode has been created through an idmapped mount the user namespace of
2335 * the vfsmount must be passed through @mnt_userns. This function will then take
2336 * care to map the inode according to @mnt_userns before checking permissions
2337 * and initializing i_uid and i_gid. On non-idmapped mounts or if permission
2338 * checking is to be performed on the raw inode simply passs init_user_ns.
2339 */
inode_init_owner(struct user_namespace * mnt_userns,struct inode * inode,const struct inode * dir,umode_t mode)2340 void inode_init_owner(struct user_namespace *mnt_userns, struct inode *inode,
2341 const struct inode *dir, umode_t mode)
2342 {
2343 inode_fsuid_set(inode, mnt_userns);
2344 if (dir && dir->i_mode & S_ISGID) {
2345 inode->i_gid = dir->i_gid;
2346
2347 /* Directories are special, and always inherit S_ISGID */
2348 if (S_ISDIR(mode))
2349 mode |= S_ISGID;
2350 } else
2351 inode_fsgid_set(inode, mnt_userns);
2352 inode->i_mode = mode;
2353 }
2354 EXPORT_SYMBOL(inode_init_owner);
2355
2356 /**
2357 * inode_owner_or_capable - check current task permissions to inode
2358 * @mnt_userns: user namespace of the mount the inode was found from
2359 * @inode: inode being checked
2360 *
2361 * Return true if current either has CAP_FOWNER in a namespace with the
2362 * inode owner uid mapped, or owns the file.
2363 *
2364 * If the inode has been found through an idmapped mount the user namespace of
2365 * the vfsmount must be passed through @mnt_userns. This function will then take
2366 * care to map the inode according to @mnt_userns before checking permissions.
2367 * On non-idmapped mounts or if permission checking is to be performed on the
2368 * raw inode simply passs init_user_ns.
2369 */
inode_owner_or_capable(struct user_namespace * mnt_userns,const struct inode * inode)2370 bool inode_owner_or_capable(struct user_namespace *mnt_userns,
2371 const struct inode *inode)
2372 {
2373 kuid_t i_uid;
2374 struct user_namespace *ns;
2375
2376 i_uid = i_uid_into_mnt(mnt_userns, inode);
2377 if (uid_eq(current_fsuid(), i_uid))
2378 return true;
2379
2380 ns = current_user_ns();
2381 if (kuid_has_mapping(ns, i_uid) && ns_capable(ns, CAP_FOWNER))
2382 return true;
2383 return false;
2384 }
2385 EXPORT_SYMBOL(inode_owner_or_capable);
2386
2387 /*
2388 * Direct i/o helper functions
2389 */
__inode_dio_wait(struct inode * inode)2390 static void __inode_dio_wait(struct inode *inode)
2391 {
2392 wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_DIO_WAKEUP);
2393 DEFINE_WAIT_BIT(q, &inode->i_state, __I_DIO_WAKEUP);
2394
2395 do {
2396 prepare_to_wait(wq, &q.wq_entry, TASK_UNINTERRUPTIBLE);
2397 if (atomic_read(&inode->i_dio_count))
2398 schedule();
2399 } while (atomic_read(&inode->i_dio_count));
2400 finish_wait(wq, &q.wq_entry);
2401 }
2402
2403 /**
2404 * inode_dio_wait - wait for outstanding DIO requests to finish
2405 * @inode: inode to wait for
2406 *
2407 * Waits for all pending direct I/O requests to finish so that we can
2408 * proceed with a truncate or equivalent operation.
2409 *
2410 * Must be called under a lock that serializes taking new references
2411 * to i_dio_count, usually by inode->i_mutex.
2412 */
inode_dio_wait(struct inode * inode)2413 void inode_dio_wait(struct inode *inode)
2414 {
2415 if (atomic_read(&inode->i_dio_count))
2416 __inode_dio_wait(inode);
2417 }
2418 EXPORT_SYMBOL(inode_dio_wait);
2419
2420 /*
2421 * inode_set_flags - atomically set some inode flags
2422 *
2423 * Note: the caller should be holding i_mutex, or else be sure that
2424 * they have exclusive access to the inode structure (i.e., while the
2425 * inode is being instantiated). The reason for the cmpxchg() loop
2426 * --- which wouldn't be necessary if all code paths which modify
2427 * i_flags actually followed this rule, is that there is at least one
2428 * code path which doesn't today so we use cmpxchg() out of an abundance
2429 * of caution.
2430 *
2431 * In the long run, i_mutex is overkill, and we should probably look
2432 * at using the i_lock spinlock to protect i_flags, and then make sure
2433 * it is so documented in include/linux/fs.h and that all code follows
2434 * the locking convention!!
2435 */
inode_set_flags(struct inode * inode,unsigned int flags,unsigned int mask)2436 void inode_set_flags(struct inode *inode, unsigned int flags,
2437 unsigned int mask)
2438 {
2439 WARN_ON_ONCE(flags & ~mask);
2440 set_mask_bits(&inode->i_flags, mask, flags);
2441 }
2442 EXPORT_SYMBOL(inode_set_flags);
2443
inode_nohighmem(struct inode * inode)2444 void inode_nohighmem(struct inode *inode)
2445 {
2446 mapping_set_gfp_mask(inode->i_mapping, GFP_USER);
2447 }
2448 EXPORT_SYMBOL(inode_nohighmem);
2449
2450 /**
2451 * timestamp_truncate - Truncate timespec to a granularity
2452 * @t: Timespec
2453 * @inode: inode being updated
2454 *
2455 * Truncate a timespec to the granularity supported by the fs
2456 * containing the inode. Always rounds down. gran must
2457 * not be 0 nor greater than a second (NSEC_PER_SEC, or 10^9 ns).
2458 */
timestamp_truncate(struct timespec64 t,struct inode * inode)2459 struct timespec64 timestamp_truncate(struct timespec64 t, struct inode *inode)
2460 {
2461 struct super_block *sb = inode->i_sb;
2462 unsigned int gran = sb->s_time_gran;
2463
2464 t.tv_sec = clamp(t.tv_sec, sb->s_time_min, sb->s_time_max);
2465 if (unlikely(t.tv_sec == sb->s_time_max || t.tv_sec == sb->s_time_min))
2466 t.tv_nsec = 0;
2467
2468 /* Avoid division in the common cases 1 ns and 1 s. */
2469 if (gran == 1)
2470 ; /* nothing */
2471 else if (gran == NSEC_PER_SEC)
2472 t.tv_nsec = 0;
2473 else if (gran > 1 && gran < NSEC_PER_SEC)
2474 t.tv_nsec -= t.tv_nsec % gran;
2475 else
2476 WARN(1, "invalid file time granularity: %u", gran);
2477 return t;
2478 }
2479 EXPORT_SYMBOL(timestamp_truncate);
2480
2481 /**
2482 * current_time - Return FS time
2483 * @inode: inode.
2484 *
2485 * Return the current time truncated to the time granularity supported by
2486 * the fs.
2487 *
2488 * Note that inode and inode->sb cannot be NULL.
2489 * Otherwise, the function warns and returns time without truncation.
2490 */
current_time(struct inode * inode)2491 struct timespec64 current_time(struct inode *inode)
2492 {
2493 struct timespec64 now;
2494
2495 ktime_get_coarse_real_ts64(&now);
2496
2497 if (unlikely(!inode->i_sb)) {
2498 WARN(1, "current_time() called with uninitialized super_block in the inode");
2499 return now;
2500 }
2501
2502 return timestamp_truncate(now, inode);
2503 }
2504 EXPORT_SYMBOL(current_time);
2505
2506 /**
2507 * inode_set_ctime_current - set the ctime to current_time
2508 * @inode: inode
2509 *
2510 * Set the inode->i_ctime to the current value for the inode. Returns
2511 * the current value that was assigned to i_ctime.
2512 */
inode_set_ctime_current(struct inode * inode)2513 struct timespec64 inode_set_ctime_current(struct inode *inode)
2514 {
2515 struct timespec64 now = current_time(inode);
2516
2517 inode_set_ctime(inode, now.tv_sec, now.tv_nsec);
2518 return now;
2519 }
2520 EXPORT_SYMBOL(inode_set_ctime_current);
2521
2522 /**
2523 * in_group_or_capable - check whether caller is CAP_FSETID privileged
2524 * @mnt_userns: user namespace of the mount @inode was found from
2525 * @inode: inode to check
2526 * @vfsgid: the new/current vfsgid of @inode
2527 *
2528 * Check wether @vfsgid is in the caller's group list or if the caller is
2529 * privileged with CAP_FSETID over @inode. This can be used to determine
2530 * whether the setgid bit can be kept or must be dropped.
2531 *
2532 * Return: true if the caller is sufficiently privileged, false if not.
2533 */
in_group_or_capable(struct user_namespace * mnt_userns,const struct inode * inode,vfsgid_t vfsgid)2534 bool in_group_or_capable(struct user_namespace *mnt_userns,
2535 const struct inode *inode, vfsgid_t vfsgid)
2536 {
2537 if (vfsgid_in_group_p(vfsgid))
2538 return true;
2539 if (capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FSETID))
2540 return true;
2541 return false;
2542 }
2543
2544 /**
2545 * mode_strip_sgid - handle the sgid bit for non-directories
2546 * @mnt_userns: User namespace of the mount the inode was created from
2547 * @dir: parent directory inode
2548 * @mode: mode of the file to be created in @dir
2549 *
2550 * If the @mode of the new file has both the S_ISGID and S_IXGRP bit
2551 * raised and @dir has the S_ISGID bit raised ensure that the caller is
2552 * either in the group of the parent directory or they have CAP_FSETID
2553 * in their user namespace and are privileged over the parent directory.
2554 * In all other cases, strip the S_ISGID bit from @mode.
2555 *
2556 * Return: the new mode to use for the file
2557 */
mode_strip_sgid(struct user_namespace * mnt_userns,const struct inode * dir,umode_t mode)2558 umode_t mode_strip_sgid(struct user_namespace *mnt_userns,
2559 const struct inode *dir, umode_t mode)
2560 {
2561 if ((mode & (S_ISGID | S_IXGRP)) != (S_ISGID | S_IXGRP))
2562 return mode;
2563 if (S_ISDIR(mode) || !dir || !(dir->i_mode & S_ISGID))
2564 return mode;
2565 if (in_group_or_capable(mnt_userns, dir,
2566 i_gid_into_vfsgid(mnt_userns, dir)))
2567 return mode;
2568 return mode & ~S_ISGID;
2569 }
2570 EXPORT_SYMBOL(mode_strip_sgid);
2571