projects / linux-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge branches 'pm-cpuidle', 'pm-core' and 'pm-sleep'
[linux-block.git] / fs / fuse / inode.c
... / ...
CommitLineData
1/*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include "fuse_i.h"
10
11#include <linux/pagemap.h>
12#include <linux/slab.h>
13#include <linux/file.h>
14#include <linux/seq_file.h>
15#include <linux/init.h>
16#include <linux/module.h>
17#include <linux/moduleparam.h>
18#include <linux/fs_context.h>
19#include <linux/fs_parser.h>
20#include <linux/statfs.h>
21#include <linux/random.h>
22#include <linux/sched.h>
23#include <linux/exportfs.h>
24#include <linux/posix_acl.h>
25#include <linux/pid_namespace.h>
26#include <uapi/linux/magic.h>
27
28MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
29MODULE_DESCRIPTION("Filesystem in Userspace");
30MODULE_LICENSE("GPL");
31
32static struct kmem_cache *fuse_inode_cachep;
33struct list_head fuse_conn_list;
34DEFINE_MUTEX(fuse_mutex);
35
36static int set_global_limit(const char *val, const struct kernel_param *kp);
37
38unsigned max_user_bgreq;
39module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
40 &max_user_bgreq, 0644);
41__MODULE_PARM_TYPE(max_user_bgreq, "uint");
42MODULE_PARM_DESC(max_user_bgreq,
43 "Global limit for the maximum number of backgrounded requests an "
44 "unprivileged user can set");
45
46unsigned max_user_congthresh;
47module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
48 &max_user_congthresh, 0644);
49__MODULE_PARM_TYPE(max_user_congthresh, "uint");
50MODULE_PARM_DESC(max_user_congthresh,
51 "Global limit for the maximum congestion threshold an "
52 "unprivileged user can set");
53
54#define FUSE_DEFAULT_BLKSIZE 512
55
56/** Maximum number of outstanding background requests */
57#define FUSE_DEFAULT_MAX_BACKGROUND 12
58
59/** Congestion starts at 75% of maximum */
60#define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
61
62#ifdef CONFIG_BLOCK
63static struct file_system_type fuseblk_fs_type;
64#endif
65
66struct fuse_forget_link *fuse_alloc_forget(void)
67{
68 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
69}
70
71static struct inode *fuse_alloc_inode(struct super_block *sb)
72{
73 struct fuse_inode *fi;
74
75 fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL);
76 if (!fi)
77 return NULL;
78
79 fi->i_time = 0;
80 fi->inval_mask = 0;
81 fi->nodeid = 0;
82 fi->nlookup = 0;
83 fi->attr_version = 0;
84 fi->orig_ino = 0;
85 fi->state = 0;
86 mutex_init(&fi->mutex);
87 spin_lock_init(&fi->lock);
88 fi->forget = fuse_alloc_forget();
89 if (!fi->forget)
90 goto out_free;
91
92 if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
93 goto out_free_forget;
94
95 return &fi->inode;
96
97out_free_forget:
98 kfree(fi->forget);
99out_free:
100 kmem_cache_free(fuse_inode_cachep, fi);
101 return NULL;
102}
103
104static void fuse_free_inode(struct inode *inode)
105{
106 struct fuse_inode *fi = get_fuse_inode(inode);
107
108 mutex_destroy(&fi->mutex);
109 kfree(fi->forget);
110#ifdef CONFIG_FUSE_DAX
111 kfree(fi->dax);
112#endif
113 kmem_cache_free(fuse_inode_cachep, fi);
114}
115
116static void fuse_evict_inode(struct inode *inode)
117{
118 struct fuse_inode *fi = get_fuse_inode(inode);
119
120 /* Will write inode on close/munmap and in all other dirtiers */
121 WARN_ON(inode->i_state & I_DIRTY_INODE);
122
123 truncate_inode_pages_final(&inode->i_data);
124 clear_inode(inode);
125 if (inode->i_sb->s_flags & SB_ACTIVE) {
126 struct fuse_conn *fc = get_fuse_conn(inode);
127
128 if (FUSE_IS_DAX(inode))
129 fuse_dax_inode_cleanup(inode);
130 if (fi->nlookup) {
131 fuse_queue_forget(fc, fi->forget, fi->nodeid,
132 fi->nlookup);
133 fi->forget = NULL;
134 }
135 }
136 if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
137 WARN_ON(!list_empty(&fi->write_files));
138 WARN_ON(!list_empty(&fi->queued_writes));
139 }
140}
141
142static int fuse_reconfigure(struct fs_context *fsc)
143{
144 struct super_block *sb = fsc->root->d_sb;
145
146 sync_filesystem(sb);
147 if (fsc->sb_flags & SB_MANDLOCK)
148 return -EINVAL;
149
150 return 0;
151}
152
153/*
154 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
155 * so that it will fit.
156 */
157static ino_t fuse_squash_ino(u64 ino64)
158{
159 ino_t ino = (ino_t) ino64;
160 if (sizeof(ino_t) < sizeof(u64))
161 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
162 return ino;
163}
164
165void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
166 u64 attr_valid, u32 cache_mask)
167{
168 struct fuse_conn *fc = get_fuse_conn(inode);
169 struct fuse_inode *fi = get_fuse_inode(inode);
170
171 lockdep_assert_held(&fi->lock);
172
173 fi->attr_version = atomic64_inc_return(&fc->attr_version);
174 fi->i_time = attr_valid;
175 WRITE_ONCE(fi->inval_mask, 0);
176
177 inode->i_ino = fuse_squash_ino(attr->ino);
178 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
179 set_nlink(inode, attr->nlink);
180 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
181 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
182 inode->i_blocks = attr->blocks;
183
184 /* Sanitize nsecs */
185 attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
186 attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
187 attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
188
189 inode->i_atime.tv_sec = attr->atime;
190 inode->i_atime.tv_nsec = attr->atimensec;
191 /* mtime from server may be stale due to local buffered write */
192 if (!(cache_mask & STATX_MTIME)) {
193 inode->i_mtime.tv_sec = attr->mtime;
194 inode->i_mtime.tv_nsec = attr->mtimensec;
195 }
196 if (!(cache_mask & STATX_CTIME)) {
197 inode->i_ctime.tv_sec = attr->ctime;
198 inode->i_ctime.tv_nsec = attr->ctimensec;
199 }
200
201 if (attr->blksize != 0)
202 inode->i_blkbits = ilog2(attr->blksize);
203 else
204 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
205
206 /*
207 * Don't set the sticky bit in i_mode, unless we want the VFS
208 * to check permissions. This prevents failures due to the
209 * check in may_delete().
210 */
211 fi->orig_i_mode = inode->i_mode;
212 if (!fc->default_permissions)
213 inode->i_mode &= ~S_ISVTX;
214
215 fi->orig_ino = attr->ino;
216
217 /*
218 * We are refreshing inode data and it is possible that another
219 * client set suid/sgid or security.capability xattr. So clear
220 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
221 * was set or if security.capability xattr was set. But we don't
222 * know if security.capability has been set or not. So clear it
223 * anyway. Its less efficient but should be safe.
224 */
225 inode->i_flags &= ~S_NOSEC;
226}
227
228u32 fuse_get_cache_mask(struct inode *inode)
229{
230 struct fuse_conn *fc = get_fuse_conn(inode);
231
232 if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
233 return 0;
234
235 return STATX_MTIME | STATX_CTIME | STATX_SIZE;
236}
237
238void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
239 u64 attr_valid, u64 attr_version)
240{
241 struct fuse_conn *fc = get_fuse_conn(inode);
242 struct fuse_inode *fi = get_fuse_inode(inode);
243 u32 cache_mask;
244 loff_t oldsize;
245 struct timespec64 old_mtime;
246
247 spin_lock(&fi->lock);
248 /*
249 * In case of writeback_cache enabled, writes update mtime, ctime and
250 * may update i_size. In these cases trust the cached value in the
251 * inode.
252 */
253 cache_mask = fuse_get_cache_mask(inode);
254 if (cache_mask & STATX_SIZE)
255 attr->size = i_size_read(inode);
256
257 if (cache_mask & STATX_MTIME) {
258 attr->mtime = inode->i_mtime.tv_sec;
259 attr->mtimensec = inode->i_mtime.tv_nsec;
260 }
261 if (cache_mask & STATX_CTIME) {
262 attr->ctime = inode->i_ctime.tv_sec;
263 attr->ctimensec = inode->i_ctime.tv_nsec;
264 }
265
266 if ((attr_version != 0 && fi->attr_version > attr_version) ||
267 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
268 spin_unlock(&fi->lock);
269 return;
270 }
271
272 old_mtime = inode->i_mtime;
273 fuse_change_attributes_common(inode, attr, attr_valid, cache_mask);
274
275 oldsize = inode->i_size;
276 /*
277 * In case of writeback_cache enabled, the cached writes beyond EOF
278 * extend local i_size without keeping userspace server in sync. So,
279 * attr->size coming from server can be stale. We cannot trust it.
280 */
281 if (!(cache_mask & STATX_SIZE))
282 i_size_write(inode, attr->size);
283 spin_unlock(&fi->lock);
284
285 if (!cache_mask && S_ISREG(inode->i_mode)) {
286 bool inval = false;
287
288 if (oldsize != attr->size) {
289 truncate_pagecache(inode, attr->size);
290 if (!fc->explicit_inval_data)
291 inval = true;
292 } else if (fc->auto_inval_data) {
293 struct timespec64 new_mtime = {
294 .tv_sec = attr->mtime,
295 .tv_nsec = attr->mtimensec,
296 };
297
298 /*
299 * Auto inval mode also checks and invalidates if mtime
300 * has changed.
301 */
302 if (!timespec64_equal(&old_mtime, &new_mtime))
303 inval = true;
304 }
305
306 if (inval)
307 invalidate_inode_pages2(inode->i_mapping);
308 }
309
310 if (IS_ENABLED(CONFIG_FUSE_DAX))
311 fuse_dax_dontcache(inode, attr->flags);
312}
313
314static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
315 struct fuse_conn *fc)
316{
317 inode->i_mode = attr->mode & S_IFMT;
318 inode->i_size = attr->size;
319 inode->i_mtime.tv_sec = attr->mtime;
320 inode->i_mtime.tv_nsec = attr->mtimensec;
321 inode->i_ctime.tv_sec = attr->ctime;
322 inode->i_ctime.tv_nsec = attr->ctimensec;
323 if (S_ISREG(inode->i_mode)) {
324 fuse_init_common(inode);
325 fuse_init_file_inode(inode, attr->flags);
326 } else if (S_ISDIR(inode->i_mode))
327 fuse_init_dir(inode);
328 else if (S_ISLNK(inode->i_mode))
329 fuse_init_symlink(inode);
330 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
331 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
332 fuse_init_common(inode);
333 init_special_inode(inode, inode->i_mode,
334 new_decode_dev(attr->rdev));
335 } else
336 BUG();
337 /*
338 * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
339 * so they see the exact same behavior as before.
340 */
341 if (!fc->posix_acl)
342 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
343}
344
345static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
346{
347 u64 nodeid = *(u64 *) _nodeidp;
348 if (get_node_id(inode) == nodeid)
349 return 1;
350 else
351 return 0;
352}
353
354static int fuse_inode_set(struct inode *inode, void *_nodeidp)
355{
356 u64 nodeid = *(u64 *) _nodeidp;
357 get_fuse_inode(inode)->nodeid = nodeid;
358 return 0;
359}
360
361struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
362 int generation, struct fuse_attr *attr,
363 u64 attr_valid, u64 attr_version)
364{
365 struct inode *inode;
366 struct fuse_inode *fi;
367 struct fuse_conn *fc = get_fuse_conn_super(sb);
368
369 /*
370 * Auto mount points get their node id from the submount root, which is
371 * not a unique identifier within this filesystem.
372 *
373 * To avoid conflicts, do not place submount points into the inode hash
374 * table.
375 */
376 if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
377 S_ISDIR(attr->mode)) {
378 inode = new_inode(sb);
379 if (!inode)
380 return NULL;
381
382 fuse_init_inode(inode, attr, fc);
383 get_fuse_inode(inode)->nodeid = nodeid;
384 inode->i_flags |= S_AUTOMOUNT;
385 goto done;
386 }
387
388retry:
389 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
390 if (!inode)
391 return NULL;
392
393 if ((inode->i_state & I_NEW)) {
394 inode->i_flags |= S_NOATIME;
395 if (!fc->writeback_cache || !S_ISREG(attr->mode))
396 inode->i_flags |= S_NOCMTIME;
397 inode->i_generation = generation;
398 fuse_init_inode(inode, attr, fc);
399 unlock_new_inode(inode);
400 } else if (fuse_stale_inode(inode, generation, attr)) {
401 /* nodeid was reused, any I/O on the old inode should fail */
402 fuse_make_bad(inode);
403 iput(inode);
404 goto retry;
405 }
406done:
407 fi = get_fuse_inode(inode);
408 spin_lock(&fi->lock);
409 fi->nlookup++;
410 spin_unlock(&fi->lock);
411 fuse_change_attributes(inode, attr, attr_valid, attr_version);
412
413 return inode;
414}
415
416struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
417 struct fuse_mount **fm)
418{
419 struct fuse_mount *fm_iter;
420 struct inode *inode;
421
422 WARN_ON(!rwsem_is_locked(&fc->killsb));
423 list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
424 if (!fm_iter->sb)
425 continue;
426
427 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
428 if (inode) {
429 if (fm)
430 *fm = fm_iter;
431 return inode;
432 }
433 }
434
435 return NULL;
436}
437
438int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
439 loff_t offset, loff_t len)
440{
441 struct fuse_inode *fi;
442 struct inode *inode;
443 pgoff_t pg_start;
444 pgoff_t pg_end;
445
446 inode = fuse_ilookup(fc, nodeid, NULL);
447 if (!inode)
448 return -ENOENT;
449
450 fi = get_fuse_inode(inode);
451 spin_lock(&fi->lock);
452 fi->attr_version = atomic64_inc_return(&fc->attr_version);
453 spin_unlock(&fi->lock);
454
455 fuse_invalidate_attr(inode);
456 forget_all_cached_acls(inode);
457 if (offset >= 0) {
458 pg_start = offset >> PAGE_SHIFT;
459 if (len <= 0)
460 pg_end = -1;
461 else
462 pg_end = (offset + len - 1) >> PAGE_SHIFT;
463 invalidate_inode_pages2_range(inode->i_mapping,
464 pg_start, pg_end);
465 }
466 iput(inode);
467 return 0;
468}
469
470bool fuse_lock_inode(struct inode *inode)
471{
472 bool locked = false;
473
474 if (!get_fuse_conn(inode)->parallel_dirops) {
475 mutex_lock(&get_fuse_inode(inode)->mutex);
476 locked = true;
477 }
478
479 return locked;
480}
481
482void fuse_unlock_inode(struct inode *inode, bool locked)
483{
484 if (locked)
485 mutex_unlock(&get_fuse_inode(inode)->mutex);
486}
487
488static void fuse_umount_begin(struct super_block *sb)
489{
490 struct fuse_conn *fc = get_fuse_conn_super(sb);
491
492 if (fc->no_force_umount)
493 return;
494
495 fuse_abort_conn(fc);
496
497 // Only retire block-device-based superblocks.
498 if (sb->s_bdev != NULL)
499 retire_super(sb);
500}
501
502static void fuse_send_destroy(struct fuse_mount *fm)
503{
504 if (fm->fc->conn_init) {
505 FUSE_ARGS(args);
506
507 args.opcode = FUSE_DESTROY;
508 args.force = true;
509 args.nocreds = true;
510 fuse_simple_request(fm, &args);
511 }
512}
513
514static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
515{
516 stbuf->f_type = FUSE_SUPER_MAGIC;
517 stbuf->f_bsize = attr->bsize;
518 stbuf->f_frsize = attr->frsize;
519 stbuf->f_blocks = attr->blocks;
520 stbuf->f_bfree = attr->bfree;
521 stbuf->f_bavail = attr->bavail;
522 stbuf->f_files = attr->files;
523 stbuf->f_ffree = attr->ffree;
524 stbuf->f_namelen = attr->namelen;
525 /* fsid is left zero */
526}
527
528static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
529{
530 struct super_block *sb = dentry->d_sb;
531 struct fuse_mount *fm = get_fuse_mount_super(sb);
532 FUSE_ARGS(args);
533 struct fuse_statfs_out outarg;
534 int err;
535
536 if (!fuse_allow_current_process(fm->fc)) {
537 buf->f_type = FUSE_SUPER_MAGIC;
538 return 0;
539 }
540
541 memset(&outarg, 0, sizeof(outarg));
542 args.in_numargs = 0;
543 args.opcode = FUSE_STATFS;
544 args.nodeid = get_node_id(d_inode(dentry));
545 args.out_numargs = 1;
546 args.out_args[0].size = sizeof(outarg);
547 args.out_args[0].value = &outarg;
548 err = fuse_simple_request(fm, &args);
549 if (!err)
550 convert_fuse_statfs(buf, &outarg.st);
551 return err;
552}
553
554static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
555{
556 struct fuse_sync_bucket *bucket;
557
558 bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
559 if (bucket) {
560 init_waitqueue_head(&bucket->waitq);
561 /* Initial active count */
562 atomic_set(&bucket->count, 1);
563 }
564 return bucket;
565}
566
567static void fuse_sync_fs_writes(struct fuse_conn *fc)
568{
569 struct fuse_sync_bucket *bucket, *new_bucket;
570 int count;
571
572 new_bucket = fuse_sync_bucket_alloc();
573 spin_lock(&fc->lock);
574 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
575 count = atomic_read(&bucket->count);
576 WARN_ON(count < 1);
577 /* No outstanding writes? */
578 if (count == 1) {
579 spin_unlock(&fc->lock);
580 kfree(new_bucket);
581 return;
582 }
583
584 /*
585 * Completion of new bucket depends on completion of this bucket, so add
586 * one more count.
587 */
588 atomic_inc(&new_bucket->count);
589 rcu_assign_pointer(fc->curr_bucket, new_bucket);
590 spin_unlock(&fc->lock);
591 /*
592 * Drop initial active count. At this point if all writes in this and
593 * ancestor buckets complete, the count will go to zero and this task
594 * will be woken up.
595 */
596 atomic_dec(&bucket->count);
597
598 wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
599
600 /* Drop temp count on descendant bucket */
601 fuse_sync_bucket_dec(new_bucket);
602 kfree_rcu(bucket, rcu);
603}
604
605static int fuse_sync_fs(struct super_block *sb, int wait)
606{
607 struct fuse_mount *fm = get_fuse_mount_super(sb);
608 struct fuse_conn *fc = fm->fc;
609 struct fuse_syncfs_in inarg;
610 FUSE_ARGS(args);
611 int err;
612
613 /*
614 * Userspace cannot handle the wait == 0 case. Avoid a
615 * gratuitous roundtrip.
616 */
617 if (!wait)
618 return 0;
619
620 /* The filesystem is being unmounted. Nothing to do. */
621 if (!sb->s_root)
622 return 0;
623
624 if (!fc->sync_fs)
625 return 0;
626
627 fuse_sync_fs_writes(fc);
628
629 memset(&inarg, 0, sizeof(inarg));
630 args.in_numargs = 1;
631 args.in_args[0].size = sizeof(inarg);
632 args.in_args[0].value = &inarg;
633 args.opcode = FUSE_SYNCFS;
634 args.nodeid = get_node_id(sb->s_root->d_inode);
635 args.out_numargs = 0;
636
637 err = fuse_simple_request(fm, &args);
638 if (err == -ENOSYS) {
639 fc->sync_fs = 0;
640 err = 0;
641 }
642
643 return err;
644}
645
646enum {
647 OPT_SOURCE,
648 OPT_SUBTYPE,
649 OPT_FD,
650 OPT_ROOTMODE,
651 OPT_USER_ID,
652 OPT_GROUP_ID,
653 OPT_DEFAULT_PERMISSIONS,
654 OPT_ALLOW_OTHER,
655 OPT_MAX_READ,
656 OPT_BLKSIZE,
657 OPT_ERR
658};
659
660static const struct fs_parameter_spec fuse_fs_parameters[] = {
661 fsparam_string ("source", OPT_SOURCE),
662 fsparam_u32 ("fd", OPT_FD),
663 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
664 fsparam_u32 ("user_id", OPT_USER_ID),
665 fsparam_u32 ("group_id", OPT_GROUP_ID),
666 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
667 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
668 fsparam_u32 ("max_read", OPT_MAX_READ),
669 fsparam_u32 ("blksize", OPT_BLKSIZE),
670 fsparam_string ("subtype", OPT_SUBTYPE),
671 {}
672};
673
674static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
675{
676 struct fs_parse_result result;
677 struct fuse_fs_context *ctx = fsc->fs_private;
678 int opt;
679
680 if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
681 /*
682 * Ignore options coming from mount(MS_REMOUNT) for backward
683 * compatibility.
684 */
685 if (fsc->oldapi)
686 return 0;
687
688 return invalfc(fsc, "No changes allowed in reconfigure");
689 }
690
691 opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
692 if (opt < 0)
693 return opt;
694
695 switch (opt) {
696 case OPT_SOURCE:
697 if (fsc->source)
698 return invalfc(fsc, "Multiple sources specified");
699 fsc->source = param->string;
700 param->string = NULL;
701 break;
702
703 case OPT_SUBTYPE:
704 if (ctx->subtype)
705 return invalfc(fsc, "Multiple subtypes specified");
706 ctx->subtype = param->string;
707 param->string = NULL;
708 return 0;
709
710 case OPT_FD:
711 ctx->fd = result.uint_32;
712 ctx->fd_present = true;
713 break;
714
715 case OPT_ROOTMODE:
716 if (!fuse_valid_type(result.uint_32))
717 return invalfc(fsc, "Invalid rootmode");
718 ctx->rootmode = result.uint_32;
719 ctx->rootmode_present = true;
720 break;
721
722 case OPT_USER_ID:
723 ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
724 if (!uid_valid(ctx->user_id))
725 return invalfc(fsc, "Invalid user_id");
726 ctx->user_id_present = true;
727 break;
728
729 case OPT_GROUP_ID:
730 ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
731 if (!gid_valid(ctx->group_id))
732 return invalfc(fsc, "Invalid group_id");
733 ctx->group_id_present = true;
734 break;
735
736 case OPT_DEFAULT_PERMISSIONS:
737 ctx->default_permissions = true;
738 break;
739
740 case OPT_ALLOW_OTHER:
741 ctx->allow_other = true;
742 break;
743
744 case OPT_MAX_READ:
745 ctx->max_read = result.uint_32;
746 break;
747
748 case OPT_BLKSIZE:
749 if (!ctx->is_bdev)
750 return invalfc(fsc, "blksize only supported for fuseblk");
751 ctx->blksize = result.uint_32;
752 break;
753
754 default:
755 return -EINVAL;
756 }
757
758 return 0;
759}
760
761static void fuse_free_fsc(struct fs_context *fsc)
762{
763 struct fuse_fs_context *ctx = fsc->fs_private;
764
765 if (ctx) {
766 kfree(ctx->subtype);
767 kfree(ctx);
768 }
769}
770
771static int fuse_show_options(struct seq_file *m, struct dentry *root)
772{
773 struct super_block *sb = root->d_sb;
774 struct fuse_conn *fc = get_fuse_conn_super(sb);
775
776 if (fc->legacy_opts_show) {
777 seq_printf(m, ",user_id=%u",
778 from_kuid_munged(fc->user_ns, fc->user_id));
779 seq_printf(m, ",group_id=%u",
780 from_kgid_munged(fc->user_ns, fc->group_id));
781 if (fc->default_permissions)
782 seq_puts(m, ",default_permissions");
783 if (fc->allow_other)
784 seq_puts(m, ",allow_other");
785 if (fc->max_read != ~0)
786 seq_printf(m, ",max_read=%u", fc->max_read);
787 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
788 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
789 }
790#ifdef CONFIG_FUSE_DAX
791 if (fc->dax_mode == FUSE_DAX_ALWAYS)
792 seq_puts(m, ",dax=always");
793 else if (fc->dax_mode == FUSE_DAX_NEVER)
794 seq_puts(m, ",dax=never");
795 else if (fc->dax_mode == FUSE_DAX_INODE_USER)
796 seq_puts(m, ",dax=inode");
797#endif
798
799 return 0;
800}
801
802static void fuse_iqueue_init(struct fuse_iqueue *fiq,
803 const struct fuse_iqueue_ops *ops,
804 void *priv)
805{
806 memset(fiq, 0, sizeof(struct fuse_iqueue));
807 spin_lock_init(&fiq->lock);
808 init_waitqueue_head(&fiq->waitq);
809 INIT_LIST_HEAD(&fiq->pending);
810 INIT_LIST_HEAD(&fiq->interrupts);
811 fiq->forget_list_tail = &fiq->forget_list_head;
812 fiq->connected = 1;
813 fiq->ops = ops;
814 fiq->priv = priv;
815}
816
817static void fuse_pqueue_init(struct fuse_pqueue *fpq)
818{
819 unsigned int i;
820
821 spin_lock_init(&fpq->lock);
822 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
823 INIT_LIST_HEAD(&fpq->processing[i]);
824 INIT_LIST_HEAD(&fpq->io);
825 fpq->connected = 1;
826}
827
828void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
829 struct user_namespace *user_ns,
830 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
831{
832 memset(fc, 0, sizeof(*fc));
833 spin_lock_init(&fc->lock);
834 spin_lock_init(&fc->bg_lock);
835 init_rwsem(&fc->killsb);
836 refcount_set(&fc->count, 1);
837 atomic_set(&fc->dev_count, 1);
838 init_waitqueue_head(&fc->blocked_waitq);
839 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
840 INIT_LIST_HEAD(&fc->bg_queue);
841 INIT_LIST_HEAD(&fc->entry);
842 INIT_LIST_HEAD(&fc->devices);
843 atomic_set(&fc->num_waiting, 0);
844 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
845 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
846 atomic64_set(&fc->khctr, 0);
847 fc->polled_files = RB_ROOT;
848 fc->blocked = 0;
849 fc->initialized = 0;
850 fc->connected = 1;
851 atomic64_set(&fc->attr_version, 1);
852 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
853 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
854 fc->user_ns = get_user_ns(user_ns);
855 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
856 fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
857
858 INIT_LIST_HEAD(&fc->mounts);
859 list_add(&fm->fc_entry, &fc->mounts);
860 fm->fc = fc;
861}
862EXPORT_SYMBOL_GPL(fuse_conn_init);
863
864void fuse_conn_put(struct fuse_conn *fc)
865{
866 if (refcount_dec_and_test(&fc->count)) {
867 struct fuse_iqueue *fiq = &fc->iq;
868 struct fuse_sync_bucket *bucket;
869
870 if (IS_ENABLED(CONFIG_FUSE_DAX))
871 fuse_dax_conn_free(fc);
872 if (fiq->ops->release)
873 fiq->ops->release(fiq);
874 put_pid_ns(fc->pid_ns);
875 put_user_ns(fc->user_ns);
876 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
877 if (bucket) {
878 WARN_ON(atomic_read(&bucket->count) != 1);
879 kfree(bucket);
880 }
881 fc->release(fc);
882 }
883}
884EXPORT_SYMBOL_GPL(fuse_conn_put);
885
886struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
887{
888 refcount_inc(&fc->count);
889 return fc;
890}
891EXPORT_SYMBOL_GPL(fuse_conn_get);
892
893static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
894{
895 struct fuse_attr attr;
896 memset(&attr, 0, sizeof(attr));
897
898 attr.mode = mode;
899 attr.ino = FUSE_ROOT_ID;
900 attr.nlink = 1;
901 return fuse_iget(sb, 1, 0, &attr, 0, 0);
902}
903
904struct fuse_inode_handle {
905 u64 nodeid;
906 u32 generation;
907};
908
909static struct dentry *fuse_get_dentry(struct super_block *sb,
910 struct fuse_inode_handle *handle)
911{
912 struct fuse_conn *fc = get_fuse_conn_super(sb);
913 struct inode *inode;
914 struct dentry *entry;
915 int err = -ESTALE;
916
917 if (handle->nodeid == 0)
918 goto out_err;
919
920 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
921 if (!inode) {
922 struct fuse_entry_out outarg;
923 const struct qstr name = QSTR_INIT(".", 1);
924
925 if (!fc->export_support)
926 goto out_err;
927
928 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
929 &inode);
930 if (err && err != -ENOENT)
931 goto out_err;
932 if (err || !inode) {
933 err = -ESTALE;
934 goto out_err;
935 }
936 err = -EIO;
937 if (get_node_id(inode) != handle->nodeid)
938 goto out_iput;
939 }
940 err = -ESTALE;
941 if (inode->i_generation != handle->generation)
942 goto out_iput;
943
944 entry = d_obtain_alias(inode);
945 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
946 fuse_invalidate_entry_cache(entry);
947
948 return entry;
949
950 out_iput:
951 iput(inode);
952 out_err:
953 return ERR_PTR(err);
954}
955
956static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
957 struct inode *parent)
958{
959 int len = parent ? 6 : 3;
960 u64 nodeid;
961 u32 generation;
962
963 if (*max_len < len) {
964 *max_len = len;
965 return FILEID_INVALID;
966 }
967
968 nodeid = get_fuse_inode(inode)->nodeid;
969 generation = inode->i_generation;
970
971 fh[0] = (u32)(nodeid >> 32);
972 fh[1] = (u32)(nodeid & 0xffffffff);
973 fh[2] = generation;
974
975 if (parent) {
976 nodeid = get_fuse_inode(parent)->nodeid;
977 generation = parent->i_generation;
978
979 fh[3] = (u32)(nodeid >> 32);
980 fh[4] = (u32)(nodeid & 0xffffffff);
981 fh[5] = generation;
982 }
983
984 *max_len = len;
985 return parent ? 0x82 : 0x81;
986}
987
988static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
989 struct fid *fid, int fh_len, int fh_type)
990{
991 struct fuse_inode_handle handle;
992
993 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
994 return NULL;
995
996 handle.nodeid = (u64) fid->raw[0] << 32;
997 handle.nodeid |= (u64) fid->raw[1];
998 handle.generation = fid->raw[2];
999 return fuse_get_dentry(sb, &handle);
1000}
1001
1002static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1003 struct fid *fid, int fh_len, int fh_type)
1004{
1005 struct fuse_inode_handle parent;
1006
1007 if (fh_type != 0x82 || fh_len < 6)
1008 return NULL;
1009
1010 parent.nodeid = (u64) fid->raw[3] << 32;
1011 parent.nodeid |= (u64) fid->raw[4];
1012 parent.generation = fid->raw[5];
1013 return fuse_get_dentry(sb, &parent);
1014}
1015
1016static struct dentry *fuse_get_parent(struct dentry *child)
1017{
1018 struct inode *child_inode = d_inode(child);
1019 struct fuse_conn *fc = get_fuse_conn(child_inode);
1020 struct inode *inode;
1021 struct dentry *parent;
1022 struct fuse_entry_out outarg;
1023 int err;
1024
1025 if (!fc->export_support)
1026 return ERR_PTR(-ESTALE);
1027
1028 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1029 &dotdot_name, &outarg, &inode);
1030 if (err) {
1031 if (err == -ENOENT)
1032 return ERR_PTR(-ESTALE);
1033 return ERR_PTR(err);
1034 }
1035
1036 parent = d_obtain_alias(inode);
1037 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1038 fuse_invalidate_entry_cache(parent);
1039
1040 return parent;
1041}
1042
1043static const struct export_operations fuse_export_operations = {
1044 .fh_to_dentry = fuse_fh_to_dentry,
1045 .fh_to_parent = fuse_fh_to_parent,
1046 .encode_fh = fuse_encode_fh,
1047 .get_parent = fuse_get_parent,
1048};
1049
1050static const struct super_operations fuse_super_operations = {
1051 .alloc_inode = fuse_alloc_inode,
1052 .free_inode = fuse_free_inode,
1053 .evict_inode = fuse_evict_inode,
1054 .write_inode = fuse_write_inode,
1055 .drop_inode = generic_delete_inode,
1056 .umount_begin = fuse_umount_begin,
1057 .statfs = fuse_statfs,
1058 .sync_fs = fuse_sync_fs,
1059 .show_options = fuse_show_options,
1060};
1061
1062static void sanitize_global_limit(unsigned *limit)
1063{
1064 /*
1065 * The default maximum number of async requests is calculated to consume
1066 * 1/2^13 of the total memory, assuming 392 bytes per request.
1067 */
1068 if (*limit == 0)
1069 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1070
1071 if (*limit >= 1 << 16)
1072 *limit = (1 << 16) - 1;
1073}
1074
1075static int set_global_limit(const char *val, const struct kernel_param *kp)
1076{
1077 int rv;
1078
1079 rv = param_set_uint(val, kp);
1080 if (rv)
1081 return rv;
1082
1083 sanitize_global_limit((unsigned *)kp->arg);
1084
1085 return 0;
1086}
1087
1088static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1089{
1090 int cap_sys_admin = capable(CAP_SYS_ADMIN);
1091
1092 if (arg->minor < 13)
1093 return;
1094
1095 sanitize_global_limit(&max_user_bgreq);
1096 sanitize_global_limit(&max_user_congthresh);
1097
1098 spin_lock(&fc->bg_lock);
1099 if (arg->max_background) {
1100 fc->max_background = arg->max_background;
1101
1102 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1103 fc->max_background = max_user_bgreq;
1104 }
1105 if (arg->congestion_threshold) {
1106 fc->congestion_threshold = arg->congestion_threshold;
1107
1108 if (!cap_sys_admin &&
1109 fc->congestion_threshold > max_user_congthresh)
1110 fc->congestion_threshold = max_user_congthresh;
1111 }
1112 spin_unlock(&fc->bg_lock);
1113}
1114
1115struct fuse_init_args {
1116 struct fuse_args args;
1117 struct fuse_init_in in;
1118 struct fuse_init_out out;
1119};
1120
1121static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1122 int error)
1123{
1124 struct fuse_conn *fc = fm->fc;
1125 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1126 struct fuse_init_out *arg = &ia->out;
1127 bool ok = true;
1128
1129 if (error || arg->major != FUSE_KERNEL_VERSION)
1130 ok = false;
1131 else {
1132 unsigned long ra_pages;
1133
1134 process_init_limits(fc, arg);
1135
1136 if (arg->minor >= 6) {
1137 u64 flags = arg->flags | (u64) arg->flags2 << 32;
1138
1139 ra_pages = arg->max_readahead / PAGE_SIZE;
1140 if (flags & FUSE_ASYNC_READ)
1141 fc->async_read = 1;
1142 if (!(flags & FUSE_POSIX_LOCKS))
1143 fc->no_lock = 1;
1144 if (arg->minor >= 17) {
1145 if (!(flags & FUSE_FLOCK_LOCKS))
1146 fc->no_flock = 1;
1147 } else {
1148 if (!(flags & FUSE_POSIX_LOCKS))
1149 fc->no_flock = 1;
1150 }
1151 if (flags & FUSE_ATOMIC_O_TRUNC)
1152 fc->atomic_o_trunc = 1;
1153 if (arg->minor >= 9) {
1154 /* LOOKUP has dependency on proto version */
1155 if (flags & FUSE_EXPORT_SUPPORT)
1156 fc->export_support = 1;
1157 }
1158 if (flags & FUSE_BIG_WRITES)
1159 fc->big_writes = 1;
1160 if (flags & FUSE_DONT_MASK)
1161 fc->dont_mask = 1;
1162 if (flags & FUSE_AUTO_INVAL_DATA)
1163 fc->auto_inval_data = 1;
1164 else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1165 fc->explicit_inval_data = 1;
1166 if (flags & FUSE_DO_READDIRPLUS) {
1167 fc->do_readdirplus = 1;
1168 if (flags & FUSE_READDIRPLUS_AUTO)
1169 fc->readdirplus_auto = 1;
1170 }
1171 if (flags & FUSE_ASYNC_DIO)
1172 fc->async_dio = 1;
1173 if (flags & FUSE_WRITEBACK_CACHE)
1174 fc->writeback_cache = 1;
1175 if (flags & FUSE_PARALLEL_DIROPS)
1176 fc->parallel_dirops = 1;
1177 if (flags & FUSE_HANDLE_KILLPRIV)
1178 fc->handle_killpriv = 1;
1179 if (arg->time_gran && arg->time_gran <= 1000000000)
1180 fm->sb->s_time_gran = arg->time_gran;
1181 if ((flags & FUSE_POSIX_ACL)) {
1182 fc->default_permissions = 1;
1183 fc->posix_acl = 1;
1184 }
1185 if (flags & FUSE_CACHE_SYMLINKS)
1186 fc->cache_symlinks = 1;
1187 if (flags & FUSE_ABORT_ERROR)
1188 fc->abort_err = 1;
1189 if (flags & FUSE_MAX_PAGES) {
1190 fc->max_pages =
1191 min_t(unsigned int, fc->max_pages_limit,
1192 max_t(unsigned int, arg->max_pages, 1));
1193 }
1194 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1195 if (flags & FUSE_MAP_ALIGNMENT &&
1196 !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1197 ok = false;
1198 }
1199 if (flags & FUSE_HAS_INODE_DAX)
1200 fc->inode_dax = 1;
1201 }
1202 if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1203 fc->handle_killpriv_v2 = 1;
1204 fm->sb->s_flags |= SB_NOSEC;
1205 }
1206 if (flags & FUSE_SETXATTR_EXT)
1207 fc->setxattr_ext = 1;
1208 if (flags & FUSE_SECURITY_CTX)
1209 fc->init_security = 1;
1210 } else {
1211 ra_pages = fc->max_read / PAGE_SIZE;
1212 fc->no_lock = 1;
1213 fc->no_flock = 1;
1214 }
1215
1216 fm->sb->s_bdi->ra_pages =
1217 min(fm->sb->s_bdi->ra_pages, ra_pages);
1218 fc->minor = arg->minor;
1219 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1220 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1221 fc->conn_init = 1;
1222 }
1223 kfree(ia);
1224
1225 if (!ok) {
1226 fc->conn_init = 0;
1227 fc->conn_error = 1;
1228 }
1229
1230 fuse_set_initialized(fc);
1231 wake_up_all(&fc->blocked_waitq);
1232}
1233
1234void fuse_send_init(struct fuse_mount *fm)
1235{
1236 struct fuse_init_args *ia;
1237 u64 flags;
1238
1239 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1240
1241 ia->in.major = FUSE_KERNEL_VERSION;
1242 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1243 ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1244 flags =
1245 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1246 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1247 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1248 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1249 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1250 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1251 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1252 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1253 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1254 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1255 FUSE_SECURITY_CTX;
1256#ifdef CONFIG_FUSE_DAX
1257 if (fm->fc->dax)
1258 flags |= FUSE_MAP_ALIGNMENT;
1259 if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1260 flags |= FUSE_HAS_INODE_DAX;
1261#endif
1262 if (fm->fc->auto_submounts)
1263 flags |= FUSE_SUBMOUNTS;
1264
1265 ia->in.flags = flags;
1266 ia->in.flags2 = flags >> 32;
1267
1268 ia->args.opcode = FUSE_INIT;
1269 ia->args.in_numargs = 1;
1270 ia->args.in_args[0].size = sizeof(ia->in);
1271 ia->args.in_args[0].value = &ia->in;
1272 ia->args.out_numargs = 1;
1273 /* Variable length argument used for backward compatibility
1274 with interface version < 7.5. Rest of init_out is zeroed
1275 by do_get_request(), so a short reply is not a problem */
1276 ia->args.out_argvar = true;
1277 ia->args.out_args[0].size = sizeof(ia->out);
1278 ia->args.out_args[0].value = &ia->out;
1279 ia->args.force = true;
1280 ia->args.nocreds = true;
1281 ia->args.end = process_init_reply;
1282
1283 if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1284 process_init_reply(fm, &ia->args, -ENOTCONN);
1285}
1286EXPORT_SYMBOL_GPL(fuse_send_init);
1287
1288void fuse_free_conn(struct fuse_conn *fc)
1289{
1290 WARN_ON(!list_empty(&fc->devices));
1291 kfree_rcu(fc, rcu);
1292}
1293EXPORT_SYMBOL_GPL(fuse_free_conn);
1294
1295static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1296{
1297 int err;
1298 char *suffix = "";
1299
1300 if (sb->s_bdev) {
1301 suffix = "-fuseblk";
1302 /*
1303 * sb->s_bdi points to blkdev's bdi however we want to redirect
1304 * it to our private bdi...
1305 */
1306 bdi_put(sb->s_bdi);
1307 sb->s_bdi = &noop_backing_dev_info;
1308 }
1309 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1310 MINOR(fc->dev), suffix);
1311 if (err)
1312 return err;
1313
1314 /* fuse does it's own writeback accounting */
1315 sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1316 sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1317
1318 /*
1319 * For a single fuse filesystem use max 1% of dirty +
1320 * writeback threshold.
1321 *
1322 * This gives about 1M of write buffer for memory maps on a
1323 * machine with 1G and 10% dirty_ratio, which should be more
1324 * than enough.
1325 *
1326 * Privileged users can raise it by writing to
1327 *
1328 * /sys/class/bdi/<bdi>/max_ratio
1329 */
1330 bdi_set_max_ratio(sb->s_bdi, 1);
1331
1332 return 0;
1333}
1334
1335struct fuse_dev *fuse_dev_alloc(void)
1336{
1337 struct fuse_dev *fud;
1338 struct list_head *pq;
1339
1340 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1341 if (!fud)
1342 return NULL;
1343
1344 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1345 if (!pq) {
1346 kfree(fud);
1347 return NULL;
1348 }
1349
1350 fud->pq.processing = pq;
1351 fuse_pqueue_init(&fud->pq);
1352
1353 return fud;
1354}
1355EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1356
1357void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1358{
1359 fud->fc = fuse_conn_get(fc);
1360 spin_lock(&fc->lock);
1361 list_add_tail(&fud->entry, &fc->devices);
1362 spin_unlock(&fc->lock);
1363}
1364EXPORT_SYMBOL_GPL(fuse_dev_install);
1365
1366struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1367{
1368 struct fuse_dev *fud;
1369
1370 fud = fuse_dev_alloc();
1371 if (!fud)
1372 return NULL;
1373
1374 fuse_dev_install(fud, fc);
1375 return fud;
1376}
1377EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1378
1379void fuse_dev_free(struct fuse_dev *fud)
1380{
1381 struct fuse_conn *fc = fud->fc;
1382
1383 if (fc) {
1384 spin_lock(&fc->lock);
1385 list_del(&fud->entry);
1386 spin_unlock(&fc->lock);
1387
1388 fuse_conn_put(fc);
1389 }
1390 kfree(fud->pq.processing);
1391 kfree(fud);
1392}
1393EXPORT_SYMBOL_GPL(fuse_dev_free);
1394
1395static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1396 const struct fuse_inode *fi)
1397{
1398 *attr = (struct fuse_attr){
1399 .ino = fi->inode.i_ino,
1400 .size = fi->inode.i_size,
1401 .blocks = fi->inode.i_blocks,
1402 .atime = fi->inode.i_atime.tv_sec,
1403 .mtime = fi->inode.i_mtime.tv_sec,
1404 .ctime = fi->inode.i_ctime.tv_sec,
1405 .atimensec = fi->inode.i_atime.tv_nsec,
1406 .mtimensec = fi->inode.i_mtime.tv_nsec,
1407 .ctimensec = fi->inode.i_ctime.tv_nsec,
1408 .mode = fi->inode.i_mode,
1409 .nlink = fi->inode.i_nlink,
1410 .uid = fi->inode.i_uid.val,
1411 .gid = fi->inode.i_gid.val,
1412 .rdev = fi->inode.i_rdev,
1413 .blksize = 1u << fi->inode.i_blkbits,
1414 };
1415}
1416
1417static void fuse_sb_defaults(struct super_block *sb)
1418{
1419 sb->s_magic = FUSE_SUPER_MAGIC;
1420 sb->s_op = &fuse_super_operations;
1421 sb->s_xattr = fuse_xattr_handlers;
1422 sb->s_maxbytes = MAX_LFS_FILESIZE;
1423 sb->s_time_gran = 1;
1424 sb->s_export_op = &fuse_export_operations;
1425 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1426 if (sb->s_user_ns != &init_user_ns)
1427 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1428 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1429}
1430
1431static int fuse_fill_super_submount(struct super_block *sb,
1432 struct fuse_inode *parent_fi)
1433{
1434 struct fuse_mount *fm = get_fuse_mount_super(sb);
1435 struct super_block *parent_sb = parent_fi->inode.i_sb;
1436 struct fuse_attr root_attr;
1437 struct inode *root;
1438
1439 fuse_sb_defaults(sb);
1440 fm->sb = sb;
1441
1442 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1443 sb->s_bdi = bdi_get(parent_sb->s_bdi);
1444
1445 sb->s_xattr = parent_sb->s_xattr;
1446 sb->s_time_gran = parent_sb->s_time_gran;
1447 sb->s_blocksize = parent_sb->s_blocksize;
1448 sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1449 sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1450 if (parent_sb->s_subtype && !sb->s_subtype)
1451 return -ENOMEM;
1452
1453 fuse_fill_attr_from_inode(&root_attr, parent_fi);
1454 root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1455 /*
1456 * This inode is just a duplicate, so it is not looked up and
1457 * its nlookup should not be incremented. fuse_iget() does
1458 * that, though, so undo it here.
1459 */
1460 get_fuse_inode(root)->nlookup--;
1461 sb->s_d_op = &fuse_dentry_operations;
1462 sb->s_root = d_make_root(root);
1463 if (!sb->s_root)
1464 return -ENOMEM;
1465
1466 return 0;
1467}
1468
1469/* Filesystem context private data holds the FUSE inode of the mount point */
1470static int fuse_get_tree_submount(struct fs_context *fsc)
1471{
1472 struct fuse_mount *fm;
1473 struct fuse_inode *mp_fi = fsc->fs_private;
1474 struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1475 struct super_block *sb;
1476 int err;
1477
1478 fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1479 if (!fm)
1480 return -ENOMEM;
1481
1482 fm->fc = fuse_conn_get(fc);
1483 fsc->s_fs_info = fm;
1484 sb = sget_fc(fsc, NULL, set_anon_super_fc);
1485 if (fsc->s_fs_info)
1486 fuse_mount_destroy(fm);
1487 if (IS_ERR(sb))
1488 return PTR_ERR(sb);
1489
1490 /* Initialize superblock, making @mp_fi its root */
1491 err = fuse_fill_super_submount(sb, mp_fi);
1492 if (err) {
1493 deactivate_locked_super(sb);
1494 return err;
1495 }
1496
1497 down_write(&fc->killsb);
1498 list_add_tail(&fm->fc_entry, &fc->mounts);
1499 up_write(&fc->killsb);
1500
1501 sb->s_flags |= SB_ACTIVE;
1502 fsc->root = dget(sb->s_root);
1503
1504 return 0;
1505}
1506
1507static const struct fs_context_operations fuse_context_submount_ops = {
1508 .get_tree = fuse_get_tree_submount,
1509};
1510
1511int fuse_init_fs_context_submount(struct fs_context *fsc)
1512{
1513 fsc->ops = &fuse_context_submount_ops;
1514 return 0;
1515}
1516EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1517
1518int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1519{
1520 struct fuse_dev *fud = NULL;
1521 struct fuse_mount *fm = get_fuse_mount_super(sb);
1522 struct fuse_conn *fc = fm->fc;
1523 struct inode *root;
1524 struct dentry *root_dentry;
1525 int err;
1526
1527 err = -EINVAL;
1528 if (sb->s_flags & SB_MANDLOCK)
1529 goto err;
1530
1531 rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1532 fuse_sb_defaults(sb);
1533
1534 if (ctx->is_bdev) {
1535#ifdef CONFIG_BLOCK
1536 err = -EINVAL;
1537 if (!sb_set_blocksize(sb, ctx->blksize))
1538 goto err;
1539#endif
1540 } else {
1541 sb->s_blocksize = PAGE_SIZE;
1542 sb->s_blocksize_bits = PAGE_SHIFT;
1543 }
1544
1545 sb->s_subtype = ctx->subtype;
1546 ctx->subtype = NULL;
1547 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1548 err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1549 if (err)
1550 goto err;
1551 }
1552
1553 if (ctx->fudptr) {
1554 err = -ENOMEM;
1555 fud = fuse_dev_alloc_install(fc);
1556 if (!fud)
1557 goto err_free_dax;
1558 }
1559
1560 fc->dev = sb->s_dev;
1561 fm->sb = sb;
1562 err = fuse_bdi_init(fc, sb);
1563 if (err)
1564 goto err_dev_free;
1565
1566 /* Handle umasking inside the fuse code */
1567 if (sb->s_flags & SB_POSIXACL)
1568 fc->dont_mask = 1;
1569 sb->s_flags |= SB_POSIXACL;
1570
1571 fc->default_permissions = ctx->default_permissions;
1572 fc->allow_other = ctx->allow_other;
1573 fc->user_id = ctx->user_id;
1574 fc->group_id = ctx->group_id;
1575 fc->legacy_opts_show = ctx->legacy_opts_show;
1576 fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1577 fc->destroy = ctx->destroy;
1578 fc->no_control = ctx->no_control;
1579 fc->no_force_umount = ctx->no_force_umount;
1580
1581 err = -ENOMEM;
1582 root = fuse_get_root_inode(sb, ctx->rootmode);
1583 sb->s_d_op = &fuse_root_dentry_operations;
1584 root_dentry = d_make_root(root);
1585 if (!root_dentry)
1586 goto err_dev_free;
1587 /* Root dentry doesn't have .d_revalidate */
1588 sb->s_d_op = &fuse_dentry_operations;
1589
1590 mutex_lock(&fuse_mutex);
1591 err = -EINVAL;
1592 if (ctx->fudptr && *ctx->fudptr)
1593 goto err_unlock;
1594
1595 err = fuse_ctl_add_conn(fc);
1596 if (err)
1597 goto err_unlock;
1598
1599 list_add_tail(&fc->entry, &fuse_conn_list);
1600 sb->s_root = root_dentry;
1601 if (ctx->fudptr)
1602 *ctx->fudptr = fud;
1603 mutex_unlock(&fuse_mutex);
1604 return 0;
1605
1606 err_unlock:
1607 mutex_unlock(&fuse_mutex);
1608 dput(root_dentry);
1609 err_dev_free:
1610 if (fud)
1611 fuse_dev_free(fud);
1612 err_free_dax:
1613 if (IS_ENABLED(CONFIG_FUSE_DAX))
1614 fuse_dax_conn_free(fc);
1615 err:
1616 return err;
1617}
1618EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1619
1620static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1621{
1622 struct fuse_fs_context *ctx = fsc->fs_private;
1623 int err;
1624
1625 if (!ctx->file || !ctx->rootmode_present ||
1626 !ctx->user_id_present || !ctx->group_id_present)
1627 return -EINVAL;
1628
1629 /*
1630 * Require mount to happen from the same user namespace which
1631 * opened /dev/fuse to prevent potential attacks.
1632 */
1633 if ((ctx->file->f_op != &fuse_dev_operations) ||
1634 (ctx->file->f_cred->user_ns != sb->s_user_ns))
1635 return -EINVAL;
1636 ctx->fudptr = &ctx->file->private_data;
1637
1638 err = fuse_fill_super_common(sb, ctx);
1639 if (err)
1640 return err;
1641 /* file->private_data shall be visible on all CPUs after this */
1642 smp_mb();
1643 fuse_send_init(get_fuse_mount_super(sb));
1644 return 0;
1645}
1646
1647/*
1648 * This is the path where user supplied an already initialized fuse dev. In
1649 * this case never create a new super if the old one is gone.
1650 */
1651static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1652{
1653 return -ENOTCONN;
1654}
1655
1656static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1657{
1658
1659 return fsc->sget_key == get_fuse_conn_super(sb);
1660}
1661
1662static int fuse_get_tree(struct fs_context *fsc)
1663{
1664 struct fuse_fs_context *ctx = fsc->fs_private;
1665 struct fuse_dev *fud;
1666 struct fuse_conn *fc;
1667 struct fuse_mount *fm;
1668 struct super_block *sb;
1669 int err;
1670
1671 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1672 if (!fc)
1673 return -ENOMEM;
1674
1675 fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1676 if (!fm) {
1677 kfree(fc);
1678 return -ENOMEM;
1679 }
1680
1681 fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1682 fc->release = fuse_free_conn;
1683
1684 fsc->s_fs_info = fm;
1685
1686 if (ctx->fd_present)
1687 ctx->file = fget(ctx->fd);
1688
1689 if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1690 err = get_tree_bdev(fsc, fuse_fill_super);
1691 goto out;
1692 }
1693 /*
1694 * While block dev mount can be initialized with a dummy device fd
1695 * (found by device name), normal fuse mounts can't
1696 */
1697 err = -EINVAL;
1698 if (!ctx->file)
1699 goto out;
1700
1701 /*
1702 * Allow creating a fuse mount with an already initialized fuse
1703 * connection
1704 */
1705 fud = READ_ONCE(ctx->file->private_data);
1706 if (ctx->file->f_op == &fuse_dev_operations && fud) {
1707 fsc->sget_key = fud->fc;
1708 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1709 err = PTR_ERR_OR_ZERO(sb);
1710 if (!IS_ERR(sb))
1711 fsc->root = dget(sb->s_root);
1712 } else {
1713 err = get_tree_nodev(fsc, fuse_fill_super);
1714 }
1715out:
1716 if (fsc->s_fs_info)
1717 fuse_mount_destroy(fm);
1718 if (ctx->file)
1719 fput(ctx->file);
1720 return err;
1721}
1722
1723static const struct fs_context_operations fuse_context_ops = {
1724 .free = fuse_free_fsc,
1725 .parse_param = fuse_parse_param,
1726 .reconfigure = fuse_reconfigure,
1727 .get_tree = fuse_get_tree,
1728};
1729
1730/*
1731 * Set up the filesystem mount context.
1732 */
1733static int fuse_init_fs_context(struct fs_context *fsc)
1734{
1735 struct fuse_fs_context *ctx;
1736
1737 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1738 if (!ctx)
1739 return -ENOMEM;
1740
1741 ctx->max_read = ~0;
1742 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1743 ctx->legacy_opts_show = true;
1744
1745#ifdef CONFIG_BLOCK
1746 if (fsc->fs_type == &fuseblk_fs_type) {
1747 ctx->is_bdev = true;
1748 ctx->destroy = true;
1749 }
1750#endif
1751
1752 fsc->fs_private = ctx;
1753 fsc->ops = &fuse_context_ops;
1754 return 0;
1755}
1756
1757bool fuse_mount_remove(struct fuse_mount *fm)
1758{
1759 struct fuse_conn *fc = fm->fc;
1760 bool last = false;
1761
1762 down_write(&fc->killsb);
1763 list_del_init(&fm->fc_entry);
1764 if (list_empty(&fc->mounts))
1765 last = true;
1766 up_write(&fc->killsb);
1767
1768 return last;
1769}
1770EXPORT_SYMBOL_GPL(fuse_mount_remove);
1771
1772void fuse_conn_destroy(struct fuse_mount *fm)
1773{
1774 struct fuse_conn *fc = fm->fc;
1775
1776 if (fc->destroy)
1777 fuse_send_destroy(fm);
1778
1779 fuse_abort_conn(fc);
1780 fuse_wait_aborted(fc);
1781
1782 if (!list_empty(&fc->entry)) {
1783 mutex_lock(&fuse_mutex);
1784 list_del(&fc->entry);
1785 fuse_ctl_remove_conn(fc);
1786 mutex_unlock(&fuse_mutex);
1787 }
1788}
1789EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1790
1791static void fuse_sb_destroy(struct super_block *sb)
1792{
1793 struct fuse_mount *fm = get_fuse_mount_super(sb);
1794 bool last;
1795
1796 if (sb->s_root) {
1797 last = fuse_mount_remove(fm);
1798 if (last)
1799 fuse_conn_destroy(fm);
1800 }
1801}
1802
1803void fuse_mount_destroy(struct fuse_mount *fm)
1804{
1805 fuse_conn_put(fm->fc);
1806 kfree(fm);
1807}
1808EXPORT_SYMBOL(fuse_mount_destroy);
1809
1810static void fuse_kill_sb_anon(struct super_block *sb)
1811{
1812 fuse_sb_destroy(sb);
1813 kill_anon_super(sb);
1814 fuse_mount_destroy(get_fuse_mount_super(sb));
1815}
1816
1817static struct file_system_type fuse_fs_type = {
1818 .owner = THIS_MODULE,
1819 .name = "fuse",
1820 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1821 .init_fs_context = fuse_init_fs_context,
1822 .parameters = fuse_fs_parameters,
1823 .kill_sb = fuse_kill_sb_anon,
1824};
1825MODULE_ALIAS_FS("fuse");
1826
1827#ifdef CONFIG_BLOCK
1828static void fuse_kill_sb_blk(struct super_block *sb)
1829{
1830 fuse_sb_destroy(sb);
1831 kill_block_super(sb);
1832 fuse_mount_destroy(get_fuse_mount_super(sb));
1833}
1834
1835static struct file_system_type fuseblk_fs_type = {
1836 .owner = THIS_MODULE,
1837 .name = "fuseblk",
1838 .init_fs_context = fuse_init_fs_context,
1839 .parameters = fuse_fs_parameters,
1840 .kill_sb = fuse_kill_sb_blk,
1841 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1842};
1843MODULE_ALIAS_FS("fuseblk");
1844
1845static inline int register_fuseblk(void)
1846{
1847 return register_filesystem(&fuseblk_fs_type);
1848}
1849
1850static inline void unregister_fuseblk(void)
1851{
1852 unregister_filesystem(&fuseblk_fs_type);
1853}
1854#else
1855static inline int register_fuseblk(void)
1856{
1857 return 0;
1858}
1859
1860static inline void unregister_fuseblk(void)
1861{
1862}
1863#endif
1864
1865static void fuse_inode_init_once(void *foo)
1866{
1867 struct inode *inode = foo;
1868
1869 inode_init_once(inode);
1870}
1871
1872static int __init fuse_fs_init(void)
1873{
1874 int err;
1875
1876 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1877 sizeof(struct fuse_inode), 0,
1878 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1879 fuse_inode_init_once);
1880 err = -ENOMEM;
1881 if (!fuse_inode_cachep)
1882 goto out;
1883
1884 err = register_fuseblk();
1885 if (err)
1886 goto out2;
1887
1888 err = register_filesystem(&fuse_fs_type);
1889 if (err)
1890 goto out3;
1891
1892 return 0;
1893
1894 out3:
1895 unregister_fuseblk();
1896 out2:
1897 kmem_cache_destroy(fuse_inode_cachep);
1898 out:
1899 return err;
1900}
1901
1902static void fuse_fs_cleanup(void)
1903{
1904 unregister_filesystem(&fuse_fs_type);
1905 unregister_fuseblk();
1906
1907 /*
1908 * Make sure all delayed rcu free inodes are flushed before we
1909 * destroy cache.
1910 */
1911 rcu_barrier();
1912 kmem_cache_destroy(fuse_inode_cachep);
1913}
1914
1915static struct kobject *fuse_kobj;
1916
1917static int fuse_sysfs_init(void)
1918{
1919 int err;
1920
1921 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1922 if (!fuse_kobj) {
1923 err = -ENOMEM;
1924 goto out_err;
1925 }
1926
1927 err = sysfs_create_mount_point(fuse_kobj, "connections");
1928 if (err)
1929 goto out_fuse_unregister;
1930
1931 return 0;
1932
1933 out_fuse_unregister:
1934 kobject_put(fuse_kobj);
1935 out_err:
1936 return err;
1937}
1938
1939static void fuse_sysfs_cleanup(void)
1940{
1941 sysfs_remove_mount_point(fuse_kobj, "connections");
1942 kobject_put(fuse_kobj);
1943}
1944
1945static int __init fuse_init(void)
1946{
1947 int res;
1948
1949 pr_info("init (API version %i.%i)\n",
1950 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1951
1952 INIT_LIST_HEAD(&fuse_conn_list);
1953 res = fuse_fs_init();
1954 if (res)
1955 goto err;
1956
1957 res = fuse_dev_init();
1958 if (res)
1959 goto err_fs_cleanup;
1960
1961 res = fuse_sysfs_init();
1962 if (res)
1963 goto err_dev_cleanup;
1964
1965 res = fuse_ctl_init();
1966 if (res)
1967 goto err_sysfs_cleanup;
1968
1969 sanitize_global_limit(&max_user_bgreq);
1970 sanitize_global_limit(&max_user_congthresh);
1971
1972 return 0;
1973
1974 err_sysfs_cleanup:
1975 fuse_sysfs_cleanup();
1976 err_dev_cleanup:
1977 fuse_dev_cleanup();
1978 err_fs_cleanup:
1979 fuse_fs_cleanup();
1980 err:
1981 return res;
1982}
1983
1984static void __exit fuse_exit(void)
1985{
1986 pr_debug("exit\n");
1987
1988 fuse_ctl_cleanup();
1989 fuse_sysfs_cleanup();
1990 fuse_fs_cleanup();
1991 fuse_dev_cleanup();
1992}
1993
1994module_init(fuse_init);
1995module_exit(fuse_exit);
Linux 6.x block layer and io_uring tree(s)