projects/linux·blob master · refs

2193 lines · 52.7 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 FUSE: Filesystem in Userspace
4 Copyright (C) 2001-2008 Miklos Szeredi <[email protected]>
5*/
6
7#include "dev.h"
8#include "fuse_i.h"
9
10#include <linux/dax.h>
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 <[email protected]>");
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 int fuse_max_pages_limit = 256;
39/* default is no timeout */
40
41unsigned int max_user_bgreq;
42module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
43 &max_user_bgreq, 0644);
44__MODULE_PARM_TYPE(max_user_bgreq, "uint");
45MODULE_PARM_DESC(max_user_bgreq,
46 "Global limit for the maximum number of backgrounded requests an "
47 "unprivileged user can set");
48
49unsigned int max_user_congthresh;
50module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
51 &max_user_congthresh, 0644);
52__MODULE_PARM_TYPE(max_user_congthresh, "uint");
53MODULE_PARM_DESC(max_user_congthresh,
54 "Global limit for the maximum congestion threshold an "
55 "unprivileged user can set");
56
57#define FUSE_DEFAULT_BLKSIZE 512
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
66static struct fuse_submount_lookup *fuse_alloc_submount_lookup(void)
67{
68 struct fuse_submount_lookup *sl;
69
70 sl = kzalloc_obj(struct fuse_submount_lookup, GFP_KERNEL_ACCOUNT);
71 if (!sl)
72 return NULL;
73 sl->forget = fuse_alloc_forget();
74 if (!sl->forget)
75 goto out_free;
76
77 return sl;
78
79out_free:
80 kfree(sl);
81 return NULL;
82}
83
84static struct inode *fuse_alloc_inode(struct super_block *sb)
85{
86 struct fuse_inode *fi;
87
88 fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL);
89 if (!fi)
90 return NULL;
91
92 /* Initialize private data (i.e. everything except fi->inode) */
93 BUILD_BUG_ON(offsetof(struct fuse_inode, inode) != 0);
94 memset((void *) fi + sizeof(fi->inode), 0, sizeof(*fi) - sizeof(fi->inode));
95
96 fi->inval_mask = ~0;
97 mutex_init(&fi->mutex);
98 spin_lock_init(&fi->lock);
99 fi->forget = fuse_alloc_forget();
100 if (!fi->forget)
101 goto out_free;
102
103 if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
104 goto out_free_forget;
105
106 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
107 fuse_inode_backing_set(fi, NULL);
108
109 return &fi->inode;
110
111out_free_forget:
112 kfree(fi->forget);
113out_free:
114 kmem_cache_free(fuse_inode_cachep, fi);
115 return NULL;
116}
117
118static void fuse_free_inode(struct inode *inode)
119{
120 struct fuse_inode *fi = get_fuse_inode(inode);
121
122 mutex_destroy(&fi->mutex);
123 kfree(fi->forget);
124#ifdef CONFIG_FUSE_DAX
125 kfree(fi->dax);
126#endif
127 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
128 fuse_backing_put(fuse_inode_backing(fi));
129
130 kmem_cache_free(fuse_inode_cachep, fi);
131}
132
133static void fuse_cleanup_submount_lookup(struct fuse_conn *fc,
134 struct fuse_submount_lookup *sl)
135{
136 if (!refcount_dec_and_test(&sl->count))
137 return;
138
139 fuse_chan_queue_forget(fc->chan, sl->forget, sl->nodeid, 1);
140 sl->forget = NULL;
141 kfree(sl);
142}
143
144static void fuse_evict_inode(struct inode *inode)
145{
146 struct fuse_inode *fi = get_fuse_inode(inode);
147
148 /* Will write inode on close/munmap and in all other dirtiers */
149 WARN_ON(inode_state_read_once(inode) & I_DIRTY_INODE);
150
151 if (FUSE_IS_DAX(inode))
152 dax_break_layout_final(inode);
153
154 truncate_inode_pages_final(&inode->i_data);
155 clear_inode(inode);
156 if (inode->i_sb->s_flags & SB_ACTIVE) {
157 struct fuse_conn *fc = get_fuse_conn(inode);
158
159 if (FUSE_IS_DAX(inode))
160 fuse_dax_inode_cleanup(inode);
161 if (fi->nlookup) {
162 fuse_chan_queue_forget(fc->chan, fi->forget, fi->nodeid,
163 fi->nlookup);
164 fi->forget = NULL;
165 }
166
167 if (fi->submount_lookup) {
168 fuse_cleanup_submount_lookup(fc, fi->submount_lookup);
169 fi->submount_lookup = NULL;
170 }
171 /*
172 * Evict of non-deleted inode may race with outstanding
173 * LOOKUP/READDIRPLUS requests and result in inconsistency when
174 * the request finishes. Deal with that here by bumping a
175 * counter that can be compared to the starting value.
176 */
177 if (inode->i_nlink > 0)
178 atomic64_inc(&fc->evict_ctr);
179 }
180 if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
181 WARN_ON(fi->iocachectr != 0);
182 WARN_ON(!list_empty(&fi->write_files));
183 WARN_ON(!list_empty(&fi->queued_writes));
184 }
185}
186
187static int fuse_reconfigure(struct fs_context *fsc)
188{
189 struct super_block *sb = fsc->root->d_sb;
190
191 sync_filesystem(sb);
192 if (fsc->sb_flags & SB_MANDLOCK)
193 return -EINVAL;
194
195 return 0;
196}
197
198/*
199 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
200 * so that it will fit.
201 */
202static ino_t fuse_squash_ino(u64 ino64)
203{
204 ino_t ino = (ino_t) ino64;
205 if (sizeof(ino_t) < sizeof(u64))
206 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
207 return ino;
208}
209
210void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
211 struct fuse_statx *sx,
212 u64 attr_valid, u32 cache_mask,
213 u64 evict_ctr)
214{
215 struct fuse_conn *fc = get_fuse_conn(inode);
216 struct fuse_inode *fi = get_fuse_inode(inode);
217
218 lockdep_assert_held(&fi->lock);
219
220 /*
221 * Clear basic stats from invalid mask.
222 *
223 * Don't do this if this is coming from a fuse_iget() call and there
224 * might have been a racing evict which would've invalidated the result
225 * if the attr_version would've been preserved.
226 *
227 * !evict_ctr -> this is create
228 * fi->attr_version != 0 -> this is not a new inode
229 * evict_ctr == fuse_get_evict_ctr() -> no evicts while during request
230 */
231 if (!evict_ctr || fi->attr_version || evict_ctr == fuse_get_evict_ctr(fc))
232 set_mask_bits(&fi->inval_mask, STATX_BASIC_STATS, 0);
233
234 fi->attr_version = atomic64_inc_return(&fc->attr_version);
235 fi->i_time = attr_valid;
236
237 inode->i_ino = fuse_squash_ino(attr->ino);
238 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
239 set_nlink(inode, attr->nlink);
240 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
241 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
242 inode->i_blocks = attr->blocks;
243
244 /* Sanitize nsecs */
245 attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
246 attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
247 attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
248
249 inode_set_atime(inode, attr->atime, attr->atimensec);
250 /* mtime from server may be stale due to local buffered write */
251 if (!(cache_mask & STATX_MTIME)) {
252 inode_set_mtime(inode, attr->mtime, attr->mtimensec);
253 }
254 if (!(cache_mask & STATX_CTIME)) {
255 inode_set_ctime(inode, attr->ctime, attr->ctimensec);
256 }
257 if (sx) {
258 /* Sanitize nsecs */
259 sx->btime.tv_nsec =
260 min_t(u32, sx->btime.tv_nsec, NSEC_PER_SEC - 1);
261
262 /*
263 * Btime has been queried, cache is valid (whether or not btime
264 * is available or not) so clear STATX_BTIME from inval_mask.
265 *
266 * Availability of the btime attribute is indicated in
267 * FUSE_I_BTIME
268 */
269 set_mask_bits(&fi->inval_mask, STATX_BTIME, 0);
270 if (sx->mask & STATX_BTIME) {
271 set_bit(FUSE_I_BTIME, &fi->state);
272 fi->i_btime.tv_sec = sx->btime.tv_sec;
273 fi->i_btime.tv_nsec = sx->btime.tv_nsec;
274 }
275 }
276
277 if (attr->blksize)
278 fi->cached_i_blkbits = ilog2(attr->blksize);
279 else
280 fi->cached_i_blkbits = inode->i_sb->s_blocksize_bits;
281
282 /*
283 * Don't set the sticky bit in i_mode, unless we want the VFS
284 * to check permissions. This prevents failures due to the
285 * check in may_delete().
286 */
287 fi->orig_i_mode = inode->i_mode;
288 if (!fc->default_permissions)
289 inode->i_mode &= ~S_ISVTX;
290
291 fi->orig_ino = attr->ino;
292
293 /*
294 * We are refreshing inode data and it is possible that another
295 * client set suid/sgid or security.capability xattr. So clear
296 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
297 * was set or if security.capability xattr was set. But we don't
298 * know if security.capability has been set or not. So clear it
299 * anyway. Its less efficient but should be safe.
300 */
301 inode->i_flags &= ~S_NOSEC;
302}
303
304u32 fuse_get_cache_mask(struct inode *inode)
305{
306 struct fuse_conn *fc = get_fuse_conn(inode);
307
308 if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
309 return 0;
310
311 return STATX_MTIME | STATX_CTIME | STATX_SIZE;
312}
313
314static void fuse_change_attributes_i(struct inode *inode, struct fuse_attr *attr,
315 struct fuse_statx *sx, u64 attr_valid,
316 u64 attr_version, u64 evict_ctr)
317{
318 struct fuse_conn *fc = get_fuse_conn(inode);
319 struct fuse_inode *fi = get_fuse_inode(inode);
320 u32 cache_mask;
321 loff_t oldsize;
322 struct timespec64 old_mtime;
323
324 spin_lock(&fi->lock);
325 /*
326 * In case of writeback_cache enabled, writes update mtime, ctime and
327 * may update i_size. In these cases trust the cached value in the
328 * inode.
329 */
330 cache_mask = fuse_get_cache_mask(inode);
331 if (cache_mask & STATX_SIZE)
332 attr->size = i_size_read(inode);
333
334 if (cache_mask & STATX_MTIME) {
335 attr->mtime = inode_get_mtime_sec(inode);
336 attr->mtimensec = inode_get_mtime_nsec(inode);
337 }
338 if (cache_mask & STATX_CTIME) {
339 attr->ctime = inode_get_ctime_sec(inode);
340 attr->ctimensec = inode_get_ctime_nsec(inode);
341 }
342
343 if ((attr_version != 0 && fi->attr_version > attr_version) ||
344 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
345 spin_unlock(&fi->lock);
346 return;
347 }
348
349 old_mtime = inode_get_mtime(inode);
350 fuse_change_attributes_common(inode, attr, sx, attr_valid, cache_mask,
351 evict_ctr);
352
353 oldsize = inode->i_size;
354 /*
355 * In case of writeback_cache enabled, the cached writes beyond EOF
356 * extend local i_size without keeping userspace server in sync. So,
357 * attr->size coming from server can be stale. We cannot trust it.
358 */
359 if (!(cache_mask & STATX_SIZE))
360 i_size_write(inode, attr->size);
361 spin_unlock(&fi->lock);
362
363 if (!cache_mask && S_ISREG(inode->i_mode)) {
364 bool inval = false;
365
366 if (oldsize != attr->size) {
367 truncate_pagecache(inode, attr->size);
368 if (!fc->explicit_inval_data)
369 inval = true;
370 } else if (fc->auto_inval_data) {
371 struct timespec64 new_mtime = {
372 .tv_sec = attr->mtime,
373 .tv_nsec = attr->mtimensec,
374 };
375
376 /*
377 * Auto inval mode also checks and invalidates if mtime
378 * has changed.
379 */
380 if (!timespec64_equal(&old_mtime, &new_mtime))
381 inval = true;
382 }
383
384 if (inval)
385 invalidate_inode_pages2(inode->i_mapping);
386 }
387
388 if (IS_ENABLED(CONFIG_FUSE_DAX))
389 fuse_dax_dontcache(inode, attr->flags);
390}
391
392void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
393 struct fuse_statx *sx, u64 attr_valid,
394 u64 attr_version)
395{
396 fuse_change_attributes_i(inode, attr, sx, attr_valid, attr_version, 0);
397}
398
399static void fuse_init_submount_lookup(struct fuse_submount_lookup *sl,
400 u64 nodeid)
401{
402 sl->nodeid = nodeid;
403 refcount_set(&sl->count, 1);
404}
405
406static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
407 struct fuse_conn *fc)
408{
409 inode->i_mode = attr->mode & S_IFMT;
410 inode->i_size = attr->size;
411 inode_set_mtime(inode, attr->mtime, attr->mtimensec);
412 inode_set_ctime(inode, attr->ctime, attr->ctimensec);
413 if (S_ISREG(inode->i_mode)) {
414 fuse_init_common(inode);
415 fuse_init_file_inode(inode, attr->flags);
416 } else if (S_ISDIR(inode->i_mode))
417 fuse_init_dir(inode);
418 else if (S_ISLNK(inode->i_mode))
419 fuse_init_symlink(inode);
420 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
421 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
422 fuse_init_common(inode);
423 init_special_inode(inode, inode->i_mode,
424 new_decode_dev(attr->rdev));
425 } else
426 BUG();
427 /*
428 * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
429 * so they see the exact same behavior as before.
430 */
431 if (!fc->posix_acl)
432 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
433}
434
435static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
436{
437 u64 nodeid = *(u64 *) _nodeidp;
438 if (get_node_id(inode) == nodeid)
439 return 1;
440 else
441 return 0;
442}
443
444static int fuse_inode_set(struct inode *inode, void *_nodeidp)
445{
446 u64 nodeid = *(u64 *) _nodeidp;
447 get_fuse_inode(inode)->nodeid = nodeid;
448 return 0;
449}
450
451struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
452 int generation, struct fuse_attr *attr,
453 u64 attr_valid, u64 attr_version,
454 u64 evict_ctr)
455{
456 struct inode *inode;
457 struct fuse_inode *fi;
458 struct fuse_conn *fc = get_fuse_conn_super(sb);
459 bool is_new_inode = false;
460
461 /*
462 * Auto mount points get their node id from the submount root, which is
463 * not a unique identifier within this filesystem.
464 *
465 * To avoid conflicts, do not place submount points into the inode hash
466 * table.
467 */
468 if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
469 S_ISDIR(attr->mode)) {
470 struct fuse_inode *fi;
471
472 inode = new_inode(sb);
473 if (!inode)
474 return NULL;
475
476 fuse_init_inode(inode, attr, fc);
477 fi = get_fuse_inode(inode);
478 fi->nodeid = nodeid;
479 fi->submount_lookup = fuse_alloc_submount_lookup();
480 if (!fi->submount_lookup) {
481 iput(inode);
482 return NULL;
483 }
484 /* Sets nlookup = 1 on fi->submount_lookup->nlookup */
485 fuse_init_submount_lookup(fi->submount_lookup, nodeid);
486 inode->i_flags |= S_AUTOMOUNT;
487 goto done;
488 }
489
490retry:
491 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
492 if (!inode)
493 return NULL;
494
495 is_new_inode = inode_state_read_once(inode) & I_NEW;
496 if (is_new_inode) {
497 inode->i_flags |= S_NOATIME;
498 if (!fc->writeback_cache || !S_ISREG(attr->mode))
499 inode->i_flags |= S_NOCMTIME;
500 inode->i_generation = generation;
501 fuse_init_inode(inode, attr, fc);
502 } else if (fuse_stale_inode(inode, generation, attr)) {
503 /* nodeid was reused, any I/O on the old inode should fail */
504 fuse_make_bad(inode);
505 if (inode != d_inode(sb->s_root)) {
506 remove_inode_hash(inode);
507 iput(inode);
508 goto retry;
509 }
510 }
511 fi = get_fuse_inode(inode);
512 spin_lock(&fi->lock);
513 fi->nlookup++;
514 spin_unlock(&fi->lock);
515done:
516 fuse_change_attributes_i(inode, attr, NULL, attr_valid, attr_version,
517 evict_ctr);
518 if (is_new_inode)
519 unlock_new_inode(inode);
520 return inode;
521}
522
523struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
524 struct fuse_mount **fm)
525{
526 struct fuse_mount *fm_iter;
527 struct inode *inode;
528
529 WARN_ON(!rwsem_is_locked(&fc->killsb));
530 list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
531 if (!fm_iter->sb)
532 continue;
533
534 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
535 if (inode) {
536 if (fm)
537 *fm = fm_iter;
538 return inode;
539 }
540 }
541
542 return NULL;
543}
544
545int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
546 loff_t offset, loff_t len)
547{
548 struct fuse_inode *fi;
549 struct inode *inode;
550 pgoff_t pg_start;
551 pgoff_t pg_end;
552
553 inode = fuse_ilookup(fc, nodeid, NULL);
554 if (!inode)
555 return -ENOENT;
556
557 fi = get_fuse_inode(inode);
558 spin_lock(&fi->lock);
559 fi->attr_version = atomic64_inc_return(&fc->attr_version);
560 spin_unlock(&fi->lock);
561
562 fuse_invalidate_attr(inode);
563 forget_all_cached_acls(inode);
564 if (offset >= 0) {
565 pg_start = offset >> PAGE_SHIFT;
566 if (len <= 0)
567 pg_end = -1;
568 else
569 pg_end = (offset + len - 1) >> PAGE_SHIFT;
570 invalidate_inode_pages2_range(inode->i_mapping,
571 pg_start, pg_end);
572 }
573 iput(inode);
574 return 0;
575}
576
577void fuse_try_prune_one_inode(struct fuse_conn *fc, u64 nodeid)
578{
579 struct inode *inode;
580
581 inode = fuse_ilookup(fc, nodeid, NULL);
582 if (!inode)
583 return;
584 d_prune_aliases(inode);
585 iput(inode);
586}
587
588bool fuse_lock_inode(struct inode *inode)
589{
590 bool locked = false;
591
592 if (!get_fuse_conn(inode)->parallel_dirops) {
593 mutex_lock(&get_fuse_inode(inode)->mutex);
594 locked = true;
595 }
596
597 return locked;
598}
599
600void fuse_unlock_inode(struct inode *inode, bool locked)
601{
602 if (locked)
603 mutex_unlock(&get_fuse_inode(inode)->mutex);
604}
605
606static void fuse_umount_begin(struct super_block *sb)
607{
608 struct fuse_conn *fc = get_fuse_conn_super(sb);
609
610 if (fc->no_force_umount)
611 return;
612
613 fuse_chan_abort(fc->chan, false);
614
615 // Only retire block-device-based superblocks.
616 if (sb->s_bdev != NULL)
617 retire_super(sb);
618}
619
620static void fuse_send_destroy(struct fuse_mount *fm)
621{
622 if (fm->fc->conn_init) {
623 FUSE_ARGS(args);
624
625 args.opcode = FUSE_DESTROY;
626 args.force = true;
627 args.nocreds = true;
628 fuse_simple_request(fm, &args);
629 }
630}
631
632static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
633{
634 stbuf->f_type = FUSE_SUPER_MAGIC;
635 stbuf->f_bsize = attr->bsize;
636 stbuf->f_frsize = attr->frsize;
637 stbuf->f_blocks = attr->blocks;
638 stbuf->f_bfree = attr->bfree;
639 stbuf->f_bavail = attr->bavail;
640 stbuf->f_files = attr->files;
641 stbuf->f_ffree = attr->ffree;
642 stbuf->f_namelen = attr->namelen;
643 /* fsid is left zero */
644}
645
646static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
647{
648 struct super_block *sb = dentry->d_sb;
649 struct fuse_mount *fm = get_fuse_mount_super(sb);
650 FUSE_ARGS(args);
651 struct fuse_statfs_out outarg;
652 int err;
653
654 if (!fuse_allow_current_process(fm->fc)) {
655 buf->f_type = FUSE_SUPER_MAGIC;
656 return 0;
657 }
658
659 memset(&outarg, 0, sizeof(outarg));
660 args.in_numargs = 0;
661 args.opcode = FUSE_STATFS;
662 args.nodeid = get_node_id(d_inode(dentry));
663 args.out_numargs = 1;
664 args.out_args[0].size = sizeof(outarg);
665 args.out_args[0].value = &outarg;
666 err = fuse_simple_request(fm, &args);
667 if (!err)
668 convert_fuse_statfs(buf, &outarg.st);
669 return err;
670}
671
672static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
673{
674 struct fuse_sync_bucket *bucket;
675
676 bucket = kzalloc_obj(*bucket, GFP_KERNEL | __GFP_NOFAIL);
677 init_waitqueue_head(&bucket->waitq);
678 /* Initial active count */
679 atomic_set(&bucket->count, 1);
680 return bucket;
681}
682
683static void fuse_sync_fs_writes(struct fuse_conn *fc)
684{
685 struct fuse_sync_bucket *bucket, *new_bucket;
686 int count;
687
688 new_bucket = fuse_sync_bucket_alloc();
689 spin_lock(&fc->lock);
690 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
691 count = atomic_read(&bucket->count);
692 WARN_ON(count < 1);
693 /* No outstanding writes? */
694 if (count == 1) {
695 spin_unlock(&fc->lock);
696 kfree(new_bucket);
697 return;
698 }
699
700 /*
701 * Completion of new bucket depends on completion of this bucket, so add
702 * one more count.
703 */
704 atomic_inc(&new_bucket->count);
705 rcu_assign_pointer(fc->curr_bucket, new_bucket);
706 spin_unlock(&fc->lock);
707 /*
708 * Drop initial active count. At this point if all writes in this and
709 * ancestor buckets complete, the count will go to zero and this task
710 * will be woken up.
711 */
712 atomic_dec(&bucket->count);
713
714 wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
715
716 /* Drop temp count on descendant bucket */
717 fuse_sync_bucket_dec(new_bucket);
718 kfree_rcu(bucket, rcu);
719}
720
721static int fuse_sync_fs(struct super_block *sb, int wait)
722{
723 struct fuse_mount *fm = get_fuse_mount_super(sb);
724 struct fuse_conn *fc = fm->fc;
725 struct fuse_syncfs_in inarg;
726 FUSE_ARGS(args);
727 int err;
728
729 /*
730 * Userspace cannot handle the wait == 0 case. Avoid a
731 * gratuitous roundtrip.
732 */
733 if (!wait)
734 return 0;
735
736 /* The filesystem is being unmounted. Nothing to do. */
737 if (!sb->s_root)
738 return 0;
739
740 if (!fc->sync_fs)
741 return 0;
742
743 fuse_sync_fs_writes(fc);
744
745 memset(&inarg, 0, sizeof(inarg));
746 args.in_numargs = 1;
747 args.in_args[0].size = sizeof(inarg);
748 args.in_args[0].value = &inarg;
749 args.opcode = FUSE_SYNCFS;
750 args.nodeid = get_node_id(sb->s_root->d_inode);
751 args.out_numargs = 0;
752
753 err = fuse_simple_request(fm, &args);
754 if (err == -ENOSYS) {
755 fc->sync_fs = 0;
756 err = 0;
757 }
758
759 return err;
760}
761
762enum {
763 OPT_SOURCE,
764 OPT_SUBTYPE,
765 OPT_FD,
766 OPT_ROOTMODE,
767 OPT_USER_ID,
768 OPT_GROUP_ID,
769 OPT_DEFAULT_PERMISSIONS,
770 OPT_ALLOW_OTHER,
771 OPT_MAX_READ,
772 OPT_BLKSIZE,
773 OPT_ERR
774};
775
776static const struct fs_parameter_spec fuse_fs_parameters[] = {
777 fsparam_string ("source", OPT_SOURCE),
778 fsparam_fd ("fd", OPT_FD),
779 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
780 fsparam_uid ("user_id", OPT_USER_ID),
781 fsparam_gid ("group_id", OPT_GROUP_ID),
782 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
783 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
784 fsparam_u32 ("max_read", OPT_MAX_READ),
785 fsparam_u32 ("blksize", OPT_BLKSIZE),
786 fsparam_string ("subtype", OPT_SUBTYPE),
787 {}
788};
789
790static int fuse_opt_fd(struct fs_context *fsc, struct file *file)
791{
792 struct fuse_fs_context *ctx = fsc->fs_private;
793
794 if (ctx->fud)
795 return invalfc(fsc, "Multiple fd specified");
796
797 if (file->f_op != &fuse_dev_operations)
798 return invalfc(fsc, "fd is not a fuse device");
799 /*
800 * Require mount to happen from the same user namespace which
801 * opened /dev/fuse to prevent potential attacks.
802 */
803 if (file->f_cred->user_ns != fsc->user_ns)
804 return invalfc(fsc, "wrong user namespace for fuse device");
805
806 ctx->fud = fuse_dev_grab(file);
807
808 return 0;
809}
810
811static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
812{
813 struct fs_parse_result result;
814 struct fuse_fs_context *ctx = fsc->fs_private;
815 int opt;
816 kuid_t kuid;
817 kgid_t kgid;
818
819 if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
820 /*
821 * Ignore options coming from mount(MS_REMOUNT) for backward
822 * compatibility.
823 */
824 if (fsc->oldapi)
825 return 0;
826
827 return invalfc(fsc, "No changes allowed in reconfigure");
828 }
829
830 opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
831 if (opt < 0)
832 return opt;
833
834 switch (opt) {
835 case OPT_SOURCE:
836 if (fsc->source)
837 return invalfc(fsc, "Multiple sources specified");
838 fsc->source = param->string;
839 param->string = NULL;
840 break;
841
842 case OPT_SUBTYPE:
843 if (ctx->subtype)
844 return invalfc(fsc, "Multiple subtypes specified");
845 ctx->subtype = param->string;
846 param->string = NULL;
847 return 0;
848
849 case OPT_FD:
850 if (param->type == fs_value_is_file) {
851 return fuse_opt_fd(fsc, param->file);
852 } else {
853 struct file *file __free(fput) = fget(result.uint_32);
854 if (!file)
855 return -EBADF;
856
857 return fuse_opt_fd(fsc, file);
858 }
859
860 case OPT_ROOTMODE:
861 if (!fuse_valid_type(result.uint_32))
862 return invalfc(fsc, "Invalid rootmode");
863 ctx->rootmode = result.uint_32;
864 ctx->rootmode_present = true;
865 break;
866
867 case OPT_USER_ID:
868 kuid = result.uid;
869 /*
870 * The requested uid must be representable in the
871 * filesystem's idmapping.
872 */
873 if (!kuid_has_mapping(fsc->user_ns, kuid))
874 return invalfc(fsc, "Invalid user_id");
875 ctx->user_id = kuid;
876 ctx->user_id_present = true;
877 break;
878
879 case OPT_GROUP_ID:
880 kgid = result.gid;
881 /*
882 * The requested gid must be representable in the
883 * filesystem's idmapping.
884 */
885 if (!kgid_has_mapping(fsc->user_ns, kgid))
886 return invalfc(fsc, "Invalid group_id");
887 ctx->group_id = kgid;
888 ctx->group_id_present = true;
889 break;
890
891 case OPT_DEFAULT_PERMISSIONS:
892 ctx->default_permissions = true;
893 break;
894
895 case OPT_ALLOW_OTHER:
896 ctx->allow_other = true;
897 break;
898
899 case OPT_MAX_READ:
900 ctx->max_read = result.uint_32;
901 break;
902
903 case OPT_BLKSIZE:
904 if (!ctx->is_bdev)
905 return invalfc(fsc, "blksize only supported for fuseblk");
906 ctx->blksize = result.uint_32;
907 break;
908
909 default:
910 return -EINVAL;
911 }
912
913 return 0;
914}
915
916static void fuse_free_fsc(struct fs_context *fsc)
917{
918 struct fuse_fs_context *ctx = fsc->fs_private;
919
920 if (ctx) {
921 if (ctx->fud)
922 fuse_dev_put(ctx->fud);
923 kfree(ctx->subtype);
924 kfree(ctx);
925 }
926}
927
928static int fuse_show_options(struct seq_file *m, struct dentry *root)
929{
930 struct super_block *sb = root->d_sb;
931 struct fuse_conn *fc = get_fuse_conn_super(sb);
932
933 if (fc->legacy_opts_show) {
934 seq_printf(m, ",user_id=%u",
935 from_kuid_munged(fc->user_ns, fc->user_id));
936 seq_printf(m, ",group_id=%u",
937 from_kgid_munged(fc->user_ns, fc->group_id));
938 if (fc->default_permissions)
939 seq_puts(m, ",default_permissions");
940 if (fc->allow_other)
941 seq_puts(m, ",allow_other");
942 if (fc->max_read != ~0)
943 seq_printf(m, ",max_read=%u", fc->max_read);
944 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
945 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
946 }
947#ifdef CONFIG_FUSE_DAX
948 if (fc->dax_mode == FUSE_DAX_ALWAYS)
949 seq_puts(m, ",dax=always");
950 else if (fc->dax_mode == FUSE_DAX_NEVER)
951 seq_puts(m, ",dax=never");
952 else if (fc->dax_mode == FUSE_DAX_INODE_USER)
953 seq_puts(m, ",dax=inode");
954#endif
955
956 return 0;
957}
958
959void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
960 struct user_namespace *user_ns, struct fuse_chan *fch)
961{
962 memset(fc, 0, sizeof(*fc));
963 spin_lock_init(&fc->lock);
964 init_rwsem(&fc->killsb);
965 refcount_set(&fc->count, 1);
966 atomic_set(&fc->epoch, 1);
967 INIT_WORK(&fc->epoch_work, fuse_epoch_work);
968 INIT_LIST_HEAD(&fc->entry);
969 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
970 atomic64_set(&fc->khctr, 0);
971 fc->polled_files = RB_ROOT;
972 atomic64_set(&fc->attr_version, 1);
973 atomic64_set(&fc->evict_ctr, 1);
974 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
975 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
976 fc->user_ns = get_user_ns(user_ns);
977 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
978 fc->max_pages_limit = fuse_max_pages_limit;
979 fc->name_max = FUSE_NAME_LOW_MAX;
980
981 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
982 fuse_backing_files_init(fc);
983
984 INIT_LIST_HEAD(&fc->mounts);
985 list_add(&fm->fc_entry, &fc->mounts);
986 fm->fc = fc;
987 fuse_chan_set_fc(fch, fc);
988 fc->chan = fch;
989}
990EXPORT_SYMBOL_GPL(fuse_conn_init);
991
992static void delayed_release(struct rcu_head *p)
993{
994 struct fuse_conn *fc = container_of(p, struct fuse_conn, rcu);
995
996 fuse_uring_destruct(fc->chan);
997 fuse_chan_free(fc->chan);
998
999 put_user_ns(fc->user_ns);
1000 fc->release(fc);
1001}
1002
1003void fuse_conn_put(struct fuse_conn *fc)
1004{
1005 struct fuse_sync_bucket *bucket;
1006
1007 if (!refcount_dec_and_test(&fc->count))
1008 return;
1009
1010 if (IS_ENABLED(CONFIG_FUSE_DAX))
1011 fuse_dax_conn_free(fc);
1012 cancel_work_sync(&fc->epoch_work);
1013 fuse_chan_release(fc->chan);
1014 put_pid_ns(fc->pid_ns);
1015 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
1016 if (bucket) {
1017 WARN_ON(atomic_read(&bucket->count) != 1);
1018 kfree(bucket);
1019 }
1020 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
1021 fuse_backing_files_free(fc);
1022 call_rcu(&fc->rcu, delayed_release);
1023}
1024EXPORT_SYMBOL_GPL(fuse_conn_put);
1025
1026struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
1027{
1028 refcount_inc(&fc->count);
1029 return fc;
1030}
1031EXPORT_SYMBOL_GPL(fuse_conn_get);
1032
1033dev_t fuse_conn_get_id(struct fuse_conn *fc)
1034{
1035 return fc->dev;
1036}
1037
1038static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned int mode)
1039{
1040 struct fuse_attr attr;
1041 memset(&attr, 0, sizeof(attr));
1042
1043 attr.mode = mode;
1044 attr.ino = FUSE_ROOT_ID;
1045 attr.nlink = 1;
1046 return fuse_iget(sb, FUSE_ROOT_ID, 0, &attr, 0, 0, 0);
1047}
1048
1049struct fuse_inode_handle {
1050 u64 nodeid;
1051 u32 generation;
1052};
1053
1054static struct dentry *fuse_get_dentry(struct super_block *sb,
1055 struct fuse_inode_handle *handle)
1056{
1057 struct fuse_conn *fc = get_fuse_conn_super(sb);
1058 struct inode *inode;
1059 struct dentry *entry;
1060 int err = -ESTALE;
1061
1062 if (handle->nodeid == 0)
1063 goto out_err;
1064
1065 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
1066 if (!inode) {
1067 struct fuse_entry_out outarg;
1068
1069 if (!fc->export_support)
1070 goto out_err;
1071
1072 err = fuse_lookup_name(sb, handle->nodeid, &QSTR("."), &outarg,
1073 &inode);
1074 if (err && err != -ENOENT)
1075 goto out_err;
1076 if (err || !inode) {
1077 err = -ESTALE;
1078 goto out_err;
1079 }
1080 err = -EIO;
1081 if (get_node_id(inode) != handle->nodeid)
1082 goto out_iput;
1083 }
1084 err = -ESTALE;
1085 if (inode->i_generation != handle->generation)
1086 goto out_iput;
1087
1088 entry = d_obtain_alias(inode);
1089 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
1090 fuse_invalidate_entry_cache(entry);
1091
1092 return entry;
1093
1094 out_iput:
1095 iput(inode);
1096 out_err:
1097 return ERR_PTR(err);
1098}
1099
1100static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
1101 struct inode *parent)
1102{
1103 int len = parent ? 6 : 3;
1104 u64 nodeid;
1105 u32 generation;
1106
1107 if (*max_len < len) {
1108 *max_len = len;
1109 return FILEID_INVALID;
1110 }
1111
1112 nodeid = get_fuse_inode(inode)->nodeid;
1113 generation = inode->i_generation;
1114
1115 fh[0] = (u32)(nodeid >> 32);
1116 fh[1] = (u32)(nodeid & 0xffffffff);
1117 fh[2] = generation;
1118
1119 if (parent) {
1120 nodeid = get_fuse_inode(parent)->nodeid;
1121 generation = parent->i_generation;
1122
1123 fh[3] = (u32)(nodeid >> 32);
1124 fh[4] = (u32)(nodeid & 0xffffffff);
1125 fh[5] = generation;
1126 }
1127
1128 *max_len = len;
1129 return parent ? FILEID_INO64_GEN_PARENT : FILEID_INO64_GEN;
1130}
1131
1132static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
1133 struct fid *fid, int fh_len, int fh_type)
1134{
1135 struct fuse_inode_handle handle;
1136
1137 if ((fh_type != FILEID_INO64_GEN &&
1138 fh_type != FILEID_INO64_GEN_PARENT) || fh_len < 3)
1139 return NULL;
1140
1141 handle.nodeid = (u64) fid->raw[0] << 32;
1142 handle.nodeid |= (u64) fid->raw[1];
1143 handle.generation = fid->raw[2];
1144 return fuse_get_dentry(sb, &handle);
1145}
1146
1147static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1148 struct fid *fid, int fh_len, int fh_type)
1149{
1150 struct fuse_inode_handle parent;
1151
1152 if (fh_type != FILEID_INO64_GEN_PARENT || fh_len < 6)
1153 return NULL;
1154
1155 parent.nodeid = (u64) fid->raw[3] << 32;
1156 parent.nodeid |= (u64) fid->raw[4];
1157 parent.generation = fid->raw[5];
1158 return fuse_get_dentry(sb, &parent);
1159}
1160
1161static struct dentry *fuse_get_parent(struct dentry *child)
1162{
1163 struct inode *child_inode = d_inode(child);
1164 struct fuse_conn *fc = get_fuse_conn(child_inode);
1165 struct inode *inode;
1166 struct dentry *parent;
1167 struct fuse_entry_out outarg;
1168 int err;
1169
1170 if (!fc->export_support)
1171 return ERR_PTR(-ESTALE);
1172
1173 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1174 &dotdot_name, &outarg, &inode);
1175 if (err) {
1176 if (err == -ENOENT)
1177 return ERR_PTR(-ESTALE);
1178 return ERR_PTR(err);
1179 }
1180
1181 parent = d_obtain_alias(inode);
1182 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1183 fuse_invalidate_entry_cache(parent);
1184
1185 return parent;
1186}
1187
1188/* only for fid encoding; no support for file handle */
1189static const struct export_operations fuse_export_fid_operations = {
1190 .encode_fh = fuse_encode_fh,
1191};
1192
1193static const struct export_operations fuse_export_operations = {
1194 .fh_to_dentry = fuse_fh_to_dentry,
1195 .fh_to_parent = fuse_fh_to_parent,
1196 .encode_fh = fuse_encode_fh,
1197 .get_parent = fuse_get_parent,
1198};
1199
1200static const struct super_operations fuse_super_operations = {
1201 .alloc_inode = fuse_alloc_inode,
1202 .free_inode = fuse_free_inode,
1203 .evict_inode = fuse_evict_inode,
1204 .write_inode = fuse_write_inode,
1205 .drop_inode = inode_just_drop,
1206 .umount_begin = fuse_umount_begin,
1207 .statfs = fuse_statfs,
1208 .sync_fs = fuse_sync_fs,
1209 .show_options = fuse_show_options,
1210};
1211
1212static void sanitize_global_limit(unsigned int *limit)
1213{
1214 /*
1215 * The default maximum number of async requests is calculated to consume
1216 * 1/2^13 of the total memory, assuming 392 bytes per request.
1217 */
1218 if (*limit == 0)
1219 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1220
1221 if (*limit >= 1 << 16)
1222 *limit = (1 << 16) - 1;
1223}
1224
1225static int set_global_limit(const char *val, const struct kernel_param *kp)
1226{
1227 int rv;
1228
1229 rv = param_set_uint(val, kp);
1230 if (rv)
1231 return rv;
1232
1233 sanitize_global_limit((unsigned int *)kp->arg);
1234
1235 return 0;
1236}
1237
1238static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1239{
1240 int cap_sys_admin = capable(CAP_SYS_ADMIN);
1241
1242 if (arg->minor < 13)
1243 return;
1244
1245 sanitize_global_limit(&max_user_bgreq);
1246 sanitize_global_limit(&max_user_congthresh);
1247
1248 if (arg->max_background) {
1249 unsigned int max_background = arg->max_background;
1250
1251 if (!cap_sys_admin && max_background > max_user_bgreq)
1252 max_background = max_user_bgreq;
1253
1254 fuse_chan_max_background_set(fc->chan, max_background);
1255 }
1256 if (arg->congestion_threshold) {
1257 fc->congestion_threshold = arg->congestion_threshold;
1258
1259 if (!cap_sys_admin &&
1260 fc->congestion_threshold > max_user_congthresh)
1261 fc->congestion_threshold = max_user_congthresh;
1262 }
1263}
1264
1265struct fuse_init_args {
1266 struct fuse_args args;
1267 struct fuse_init_in in;
1268 struct fuse_init_out out;
1269 struct fuse_mount *fm;
1270};
1271
1272static void process_init_reply(struct fuse_args *args, int error)
1273{
1274 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1275 struct fuse_mount *fm = ia->fm;
1276 struct fuse_conn *fc = fm->fc;
1277 struct fuse_init_out *arg = &ia->out;
1278 bool io_uring_enabled = false;
1279 bool ok = true;
1280
1281 if (error || arg->major != FUSE_KERNEL_VERSION)
1282 ok = false;
1283 else {
1284 unsigned long ra_pages;
1285 unsigned int timeout = 0;
1286
1287 process_init_limits(fc, arg);
1288
1289 if (arg->minor >= 6) {
1290 u64 flags = arg->flags;
1291
1292 if (flags & FUSE_INIT_EXT)
1293 flags |= (u64) arg->flags2 << 32;
1294
1295 ra_pages = arg->max_readahead / PAGE_SIZE;
1296 if (flags & FUSE_ASYNC_READ)
1297 fc->async_read = 1;
1298 if (!(flags & FUSE_POSIX_LOCKS))
1299 fc->no_lock = 1;
1300 if (arg->minor >= 17) {
1301 if (!(flags & FUSE_FLOCK_LOCKS))
1302 fc->no_flock = 1;
1303 } else {
1304 if (!(flags & FUSE_POSIX_LOCKS))
1305 fc->no_flock = 1;
1306 }
1307 if (flags & FUSE_ATOMIC_O_TRUNC)
1308 fc->atomic_o_trunc = 1;
1309 if (arg->minor >= 9) {
1310 /* LOOKUP has dependency on proto version */
1311 if (flags & FUSE_EXPORT_SUPPORT)
1312 fc->export_support = 1;
1313 }
1314 if (flags & FUSE_BIG_WRITES)
1315 fc->big_writes = 1;
1316 if (flags & FUSE_DONT_MASK)
1317 fc->dont_mask = 1;
1318 if (flags & FUSE_AUTO_INVAL_DATA)
1319 fc->auto_inval_data = 1;
1320 else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1321 fc->explicit_inval_data = 1;
1322 if (flags & FUSE_DO_READDIRPLUS) {
1323 fc->do_readdirplus = 1;
1324 if (flags & FUSE_READDIRPLUS_AUTO)
1325 fc->readdirplus_auto = 1;
1326 }
1327 if (flags & FUSE_ASYNC_DIO)
1328 fc->async_dio = 1;
1329 if (flags & FUSE_WRITEBACK_CACHE)
1330 fc->writeback_cache = 1;
1331 if (flags & FUSE_PARALLEL_DIROPS)
1332 fc->parallel_dirops = 1;
1333 if (flags & FUSE_HANDLE_KILLPRIV)
1334 fc->handle_killpriv = 1;
1335 if (arg->time_gran && arg->time_gran <= 1000000000)
1336 fm->sb->s_time_gran = arg->time_gran;
1337 if ((flags & FUSE_POSIX_ACL)) {
1338 fc->default_permissions = 1;
1339 fc->posix_acl = 1;
1340 }
1341 if (flags & FUSE_CACHE_SYMLINKS)
1342 fc->cache_symlinks = 1;
1343 if (flags & FUSE_ABORT_ERROR)
1344 fc->abort_err = 1;
1345 if (flags & FUSE_MAX_PAGES) {
1346 fc->max_pages =
1347 min_t(unsigned int, fc->max_pages_limit,
1348 max_t(unsigned int, arg->max_pages, 1));
1349
1350 /*
1351 * PATH_MAX file names might need two pages for
1352 * ops like rename
1353 */
1354 if (fc->max_pages > 1)
1355 fc->name_max = FUSE_NAME_MAX;
1356 }
1357 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1358 if (flags & FUSE_MAP_ALIGNMENT &&
1359 !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1360 ok = false;
1361 }
1362 if (flags & FUSE_HAS_INODE_DAX)
1363 fc->inode_dax = 1;
1364 }
1365 if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1366 fc->handle_killpriv_v2 = 1;
1367 fm->sb->s_flags |= SB_NOSEC;
1368 }
1369 if (flags & FUSE_SETXATTR_EXT)
1370 fc->setxattr_ext = 1;
1371 if (flags & FUSE_SECURITY_CTX)
1372 fc->init_security = 1;
1373 if (flags & FUSE_CREATE_SUPP_GROUP)
1374 fc->create_supp_group = 1;
1375 if (flags & FUSE_DIRECT_IO_ALLOW_MMAP)
1376 fc->direct_io_allow_mmap = 1;
1377 /*
1378 * max_stack_depth is the max stack depth of FUSE fs,
1379 * so it has to be at least 1 to support passthrough
1380 * to backing files.
1381 *
1382 * with max_stack_depth > 1, the backing files can be
1383 * on a stacked fs (e.g. overlayfs) themselves and with
1384 * max_stack_depth == 1, FUSE fs can be stacked as the
1385 * underlying fs of a stacked fs (e.g. overlayfs).
1386 *
1387 * Also don't allow the combination of FUSE_PASSTHROUGH
1388 * and FUSE_WRITEBACK_CACHE, current design doesn't handle
1389 * them together.
1390 */
1391 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH) &&
1392 (flags & FUSE_PASSTHROUGH) &&
1393 arg->max_stack_depth > 0 &&
1394 arg->max_stack_depth <= FILESYSTEM_MAX_STACK_DEPTH &&
1395 !(flags & FUSE_WRITEBACK_CACHE)) {
1396 fc->passthrough = 1;
1397 fc->max_stack_depth = arg->max_stack_depth;
1398 fm->sb->s_stack_depth = arg->max_stack_depth;
1399 }
1400 if (flags & FUSE_NO_EXPORT_SUPPORT)
1401 fm->sb->s_export_op = &fuse_export_fid_operations;
1402 if (flags & FUSE_ALLOW_IDMAP) {
1403 if (fc->default_permissions)
1404 fm->sb->s_iflags &= ~SB_I_NOIDMAP;
1405 else
1406 ok = false;
1407 }
1408 if (flags & FUSE_OVER_IO_URING && fuse_uring_enabled())
1409 io_uring_enabled = true;
1410
1411 if (flags & FUSE_REQUEST_TIMEOUT)
1412 timeout = arg->request_timeout;
1413 } else {
1414 ra_pages = fc->max_read / PAGE_SIZE;
1415 fc->no_lock = 1;
1416 fc->no_flock = 1;
1417 }
1418
1419 fuse_init_server_timeout(fc->chan, timeout);
1420
1421 fm->sb->s_bdi->ra_pages =
1422 min(fm->sb->s_bdi->ra_pages, ra_pages);
1423 fm->sb->s_bdi->io_pages = fc->max_pages;
1424 fc->minor = arg->minor;
1425 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1426 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1427 fc->conn_init = 1;
1428 }
1429 kfree(ia);
1430
1431 if (!ok) {
1432 fc->conn_init = 0;
1433 fc->conn_error = 1;
1434 fuse_chan_set_initialized(fc->chan, NULL);
1435 } else {
1436 struct fuse_chan_param cp = {
1437 .minor = fc->minor,
1438 .max_write = fc->max_write,
1439 .max_pages = fc->max_pages,
1440 .io_uring_enabled = io_uring_enabled,
1441 };
1442 fuse_chan_set_initialized(fc->chan, &cp);
1443 }
1444}
1445
1446static struct fuse_init_args *fuse_new_init(struct fuse_mount *fm)
1447{
1448 struct fuse_init_args *ia;
1449 u64 flags;
1450
1451 ia = kzalloc_obj(*ia, GFP_KERNEL | __GFP_NOFAIL);
1452
1453 ia->fm = fm;
1454 ia->in.major = FUSE_KERNEL_VERSION;
1455 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1456 ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1457 flags =
1458 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1459 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1460 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1461 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1462 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1463 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1464 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1465 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1466 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1467 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1468 FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP |
1469 FUSE_HAS_EXPIRE_ONLY | FUSE_DIRECT_IO_ALLOW_MMAP |
1470 FUSE_NO_EXPORT_SUPPORT | FUSE_HAS_RESEND | FUSE_ALLOW_IDMAP |
1471 FUSE_REQUEST_TIMEOUT;
1472#ifdef CONFIG_FUSE_DAX
1473 if (fm->fc->dax)
1474 flags |= FUSE_MAP_ALIGNMENT;
1475 if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1476 flags |= FUSE_HAS_INODE_DAX;
1477#endif
1478 if (fm->fc->auto_submounts)
1479 flags |= FUSE_SUBMOUNTS;
1480 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
1481 flags |= FUSE_PASSTHROUGH;
1482
1483 if (fuse_uring_enabled())
1484 flags |= FUSE_OVER_IO_URING | FUSE_HAS_IO_URING_BUFPOOL;
1485
1486 ia->in.flags = flags;
1487 ia->in.flags2 = flags >> 32;
1488
1489 ia->args.opcode = FUSE_INIT;
1490 ia->args.in_numargs = 1;
1491 ia->args.in_args[0].size = sizeof(ia->in);
1492 ia->args.in_args[0].value = &ia->in;
1493 ia->args.out_numargs = 1;
1494 /* Variable length argument used for backward compatibility
1495 with interface version < 7.5. Rest of init_out is zeroed
1496 by do_get_request(), so a short reply is not a problem */
1497 ia->args.out_argvar = true;
1498 ia->args.out_args[0].size = sizeof(ia->out);
1499 ia->args.out_args[0].value = &ia->out;
1500 ia->args.force = true;
1501 ia->args.nocreds = true;
1502
1503 return ia;
1504}
1505
1506int fuse_send_init(struct fuse_mount *fm)
1507{
1508 struct fuse_init_args *ia = fuse_new_init(fm);
1509 int err;
1510
1511 if (fm->fc->sync_init) {
1512 ia->args.abort_on_kill = true;
1513 err = fuse_simple_request(fm, &ia->args);
1514 /* Ignore size of init reply */
1515 if (err > 0)
1516 err = 0;
1517 } else {
1518 ia->args.end = process_init_reply;
1519 err = fuse_simple_background(fm, &ia->args, GFP_KERNEL);
1520 if (!err)
1521 return 0;
1522 }
1523 process_init_reply(&ia->args, err);
1524 if (fm->fc->conn_error)
1525 return -ENOTCONN;
1526 return 0;
1527}
1528EXPORT_SYMBOL_GPL(fuse_send_init);
1529
1530void fuse_free_conn(struct fuse_conn *fc)
1531{
1532 kfree(fc);
1533}
1534EXPORT_SYMBOL_GPL(fuse_free_conn);
1535
1536static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1537{
1538 int err;
1539 char *suffix = "";
1540
1541 if (sb->s_bdev) {
1542 suffix = "-fuseblk";
1543 /*
1544 * sb->s_bdi points to blkdev's bdi however we want to redirect
1545 * it to our private bdi...
1546 */
1547 bdi_put(sb->s_bdi);
1548 sb->s_bdi = &noop_backing_dev_info;
1549 }
1550 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1551 MINOR(fc->dev), suffix);
1552 if (err)
1553 return err;
1554
1555 sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1556
1557 /*
1558 * For a single fuse filesystem use max 1% of dirty +
1559 * writeback threshold.
1560 *
1561 * This gives about 1M of write buffer for memory maps on a
1562 * machine with 1G and 10% dirty_ratio, which should be more
1563 * than enough.
1564 *
1565 * Privileged users can raise it by writing to
1566 *
1567 * /sys/class/bdi/<bdi>/max_ratio
1568 */
1569 bdi_set_max_ratio(sb->s_bdi, 1);
1570
1571 return 0;
1572}
1573
1574static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1575 const struct fuse_inode *fi)
1576{
1577 struct timespec64 atime = inode_get_atime(&fi->inode);
1578 struct timespec64 mtime = inode_get_mtime(&fi->inode);
1579 struct timespec64 ctime = inode_get_ctime(&fi->inode);
1580
1581 *attr = (struct fuse_attr){
1582 .ino = fi->inode.i_ino,
1583 .size = fi->inode.i_size,
1584 .blocks = fi->inode.i_blocks,
1585 .atime = atime.tv_sec,
1586 .mtime = mtime.tv_sec,
1587 .ctime = ctime.tv_sec,
1588 .atimensec = atime.tv_nsec,
1589 .mtimensec = mtime.tv_nsec,
1590 .ctimensec = ctime.tv_nsec,
1591 .mode = fi->inode.i_mode,
1592 .nlink = fi->inode.i_nlink,
1593 .uid = __kuid_val(fi->inode.i_uid),
1594 .gid = __kgid_val(fi->inode.i_gid),
1595 .rdev = fi->inode.i_rdev,
1596 .blksize = 1u << fi->inode.i_blkbits,
1597 };
1598}
1599
1600static void fuse_sb_defaults(struct super_block *sb)
1601{
1602 sb->s_magic = FUSE_SUPER_MAGIC;
1603 sb->s_op = &fuse_super_operations;
1604 sb->s_xattr = fuse_xattr_handlers;
1605 sb->s_maxbytes = MAX_LFS_FILESIZE;
1606 sb->s_time_gran = 1;
1607 sb->s_export_op = &fuse_export_operations;
1608 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1609 sb->s_iflags |= SB_I_NOIDMAP;
1610 sb->s_iflags |= SB_I_NO_DATA_INTEGRITY;
1611 if (sb->s_user_ns != &init_user_ns)
1612 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1613 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1614}
1615
1616static int fuse_fill_super_submount(struct super_block *sb,
1617 struct fuse_inode *parent_fi)
1618{
1619 struct fuse_mount *fm = get_fuse_mount_super(sb);
1620 struct super_block *parent_sb = parent_fi->inode.i_sb;
1621 struct fuse_attr root_attr;
1622 struct inode *root;
1623 struct fuse_submount_lookup *sl;
1624 struct fuse_inode *fi;
1625
1626 fuse_sb_defaults(sb);
1627 fm->sb = sb;
1628
1629 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1630 sb->s_bdi = bdi_get(parent_sb->s_bdi);
1631
1632 sb->s_xattr = parent_sb->s_xattr;
1633 sb->s_export_op = parent_sb->s_export_op;
1634 sb->s_time_gran = parent_sb->s_time_gran;
1635 sb->s_blocksize = parent_sb->s_blocksize;
1636 sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1637 sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1638 if (parent_sb->s_subtype && !sb->s_subtype)
1639 return -ENOMEM;
1640
1641 fuse_fill_attr_from_inode(&root_attr, parent_fi);
1642 root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0,
1643 fuse_get_evict_ctr(fm->fc));
1644 if (!root)
1645 return -ENOMEM;
1646 /*
1647 * This inode is just a duplicate, so it is not looked up and
1648 * its nlookup should not be incremented. fuse_iget() does
1649 * that, though, so undo it here.
1650 */
1651 fi = get_fuse_inode(root);
1652 fi->nlookup--;
1653
1654 set_default_d_op(sb, &fuse_dentry_operations);
1655 sb->s_root = d_make_root(root);
1656 if (!sb->s_root)
1657 return -ENOMEM;
1658
1659 /*
1660 * Grab the parent's submount_lookup pointer and take a
1661 * reference on the shared nlookup from the parent. This is to
1662 * prevent the last forget for this nodeid from getting
1663 * triggered until all users have finished with it.
1664 */
1665 sl = parent_fi->submount_lookup;
1666 WARN_ON(!sl);
1667 if (sl) {
1668 refcount_inc(&sl->count);
1669 fi->submount_lookup = sl;
1670 }
1671
1672 return 0;
1673}
1674
1675/* Filesystem context private data holds the FUSE inode of the mount point */
1676static int fuse_get_tree_submount(struct fs_context *fsc)
1677{
1678 struct fuse_mount *fm;
1679 struct fuse_inode *mp_fi = fsc->fs_private;
1680 struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1681 struct super_block *sb;
1682 int err;
1683
1684 fm = kzalloc_obj(struct fuse_mount);
1685 if (!fm)
1686 return -ENOMEM;
1687
1688 fm->fc = fuse_conn_get(fc);
1689 fsc->s_fs_info = fm;
1690 sb = sget_fc(fsc, NULL, set_anon_super_fc);
1691 if (fsc->s_fs_info)
1692 fuse_mount_destroy(fm);
1693 if (IS_ERR(sb))
1694 return PTR_ERR(sb);
1695
1696 /* Initialize superblock, making @mp_fi its root */
1697 err = fuse_fill_super_submount(sb, mp_fi);
1698 if (err) {
1699 deactivate_locked_super(sb);
1700 return err;
1701 }
1702
1703 down_write(&fc->killsb);
1704 list_add_tail(&fm->fc_entry, &fc->mounts);
1705 up_write(&fc->killsb);
1706
1707 sb->s_flags |= SB_ACTIVE;
1708 fsc->root = dget(sb->s_root);
1709
1710 return 0;
1711}
1712
1713static const struct fs_context_operations fuse_context_submount_ops = {
1714 .get_tree = fuse_get_tree_submount,
1715};
1716
1717int fuse_init_fs_context_submount(struct fs_context *fsc)
1718{
1719 fsc->ops = &fuse_context_submount_ops;
1720 return 0;
1721}
1722EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1723
1724int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1725{
1726 struct fuse_dev *fud = ctx->fud;
1727 struct fuse_mount *fm = get_fuse_mount_super(sb);
1728 struct fuse_conn *fc = fm->fc;
1729 struct inode *root;
1730 struct dentry *root_dentry;
1731 int err;
1732
1733 err = -EINVAL;
1734 if (sb->s_flags & SB_MANDLOCK)
1735 goto err;
1736
1737 rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1738 fuse_sb_defaults(sb);
1739
1740 if (ctx->is_bdev) {
1741#ifdef CONFIG_BLOCK
1742 err = -EINVAL;
1743 if (!sb_set_blocksize(sb, ctx->blksize))
1744 goto err;
1745#endif
1746 fc->sync_fs = 1;
1747 } else {
1748 sb->s_blocksize = PAGE_SIZE;
1749 sb->s_blocksize_bits = PAGE_SHIFT;
1750 }
1751
1752 sb->s_subtype = ctx->subtype;
1753 ctx->subtype = NULL;
1754 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1755 err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1756 if (err)
1757 goto err;
1758 }
1759
1760 fc->dev = sb->s_dev;
1761 fm->sb = sb;
1762 err = fuse_bdi_init(fc, sb);
1763 if (err)
1764 goto err_free_dax;
1765
1766 /* Handle umasking inside the fuse code */
1767 if (sb->s_flags & SB_POSIXACL)
1768 fc->dont_mask = 1;
1769 sb->s_flags |= SB_POSIXACL;
1770
1771 fc->default_permissions = ctx->default_permissions;
1772 fc->allow_other = ctx->allow_other;
1773 fc->user_id = ctx->user_id;
1774 fc->group_id = ctx->group_id;
1775 fc->legacy_opts_show = ctx->legacy_opts_show;
1776 fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1777 fc->destroy = ctx->destroy;
1778 fc->no_control = ctx->no_control;
1779 fc->no_force_umount = ctx->no_force_umount;
1780
1781 err = -ENOMEM;
1782 root = fuse_get_root_inode(sb, ctx->rootmode);
1783 set_default_d_op(sb, &fuse_dentry_operations);
1784 root_dentry = d_make_root(root);
1785 if (!root_dentry)
1786 goto err_free_dax;
1787
1788 mutex_lock(&fuse_mutex);
1789 err = -EINVAL;
1790 if (fud) {
1791 if (fuse_dev_is_installed(fud))
1792 goto err_unlock;
1793 if (fuse_dev_is_sync_init(fud))
1794 fc->sync_init = 1;
1795 }
1796
1797 err = fuse_ctl_add_conn(fc);
1798 if (err)
1799 goto err_unlock;
1800
1801 list_add_tail(&fc->entry, &fuse_conn_list);
1802 sb->s_root = root_dentry;
1803 if (fud)
1804 fuse_dev_install(fud, fc->chan);
1805
1806 mutex_unlock(&fuse_mutex);
1807 return 0;
1808
1809 err_unlock:
1810 mutex_unlock(&fuse_mutex);
1811 dput(root_dentry);
1812 err_free_dax:
1813 if (IS_ENABLED(CONFIG_FUSE_DAX))
1814 fuse_dax_conn_free(fc);
1815 err:
1816 return err;
1817}
1818EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1819
1820static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1821{
1822 struct fuse_fs_context *ctx = fsc->fs_private;
1823 struct fuse_mount *fm;
1824 int err;
1825
1826 if (!ctx->fud || !ctx->rootmode_present ||
1827 !ctx->user_id_present || !ctx->group_id_present)
1828 return -EINVAL;
1829
1830 err = fuse_fill_super_common(sb, ctx);
1831 if (err)
1832 return err;
1833
1834 fm = get_fuse_mount_super(sb);
1835
1836 return fuse_send_init(fm);
1837}
1838
1839/*
1840 * This is the path where user supplied an already initialized fuse dev. In
1841 * this case never create a new super if the old one is gone.
1842 */
1843static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1844{
1845 return -ENOTCONN;
1846}
1847
1848static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1849{
1850 return fuse_dev_verify(fsc->sget_key, get_fuse_conn_super(sb)->chan);
1851}
1852
1853static int fuse_get_tree(struct fs_context *fsc)
1854{
1855 struct fuse_fs_context *ctx = fsc->fs_private;
1856 struct fuse_conn *fc;
1857 struct fuse_mount *fm;
1858 struct super_block *sb;
1859 struct fuse_chan *fch __free(fuse_chan_free) = fuse_dev_chan_new();
1860 int err;
1861
1862 if (!fch)
1863 return -ENOMEM;
1864
1865 fc = kmalloc_obj(*fc);
1866 if (!fc)
1867 return -ENOMEM;
1868
1869 fm = kzalloc_obj(*fm);
1870 if (!fm) {
1871 kfree(fc);
1872 return -ENOMEM;
1873 }
1874
1875 fuse_conn_init(fc, fm, fsc->user_ns, no_free_ptr(fch));
1876 fc->release = fuse_free_conn;
1877
1878 fsc->s_fs_info = fm;
1879
1880 if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1881 err = get_tree_bdev(fsc, fuse_fill_super);
1882 goto out;
1883 }
1884 /*
1885 * While block dev mount can be initialized with a dummy device fd
1886 * (found by device name), normal fuse mounts can't
1887 */
1888 err = -EINVAL;
1889 if (!ctx->fud)
1890 goto out;
1891
1892 /*
1893 * Allow creating a fuse mount with an already initialized fuse
1894 * connection
1895 */
1896 if (fuse_dev_is_installed(ctx->fud)) {
1897 fsc->sget_key = ctx->fud;
1898 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1899 err = PTR_ERR_OR_ZERO(sb);
1900 if (!IS_ERR(sb))
1901 fsc->root = dget(sb->s_root);
1902 } else {
1903 err = get_tree_nodev(fsc, fuse_fill_super);
1904 }
1905out:
1906 if (fsc->s_fs_info)
1907 fuse_mount_destroy(fm);
1908 return err;
1909}
1910
1911static const struct fs_context_operations fuse_context_ops = {
1912 .free = fuse_free_fsc,
1913 .parse_param = fuse_parse_param,
1914 .reconfigure = fuse_reconfigure,
1915 .get_tree = fuse_get_tree,
1916};
1917
1918/*
1919 * Set up the filesystem mount context.
1920 */
1921static int fuse_init_fs_context(struct fs_context *fsc)
1922{
1923 struct fuse_fs_context *ctx;
1924
1925 ctx = kzalloc_obj(struct fuse_fs_context);
1926 if (!ctx)
1927 return -ENOMEM;
1928
1929 ctx->max_read = ~0;
1930 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1931 ctx->legacy_opts_show = true;
1932
1933#ifdef CONFIG_BLOCK
1934 if (fsc->fs_type == &fuseblk_fs_type) {
1935 ctx->is_bdev = true;
1936 ctx->destroy = true;
1937 }
1938#endif
1939
1940 fsc->fs_private = ctx;
1941 fsc->ops = &fuse_context_ops;
1942 return 0;
1943}
1944
1945bool fuse_mount_remove(struct fuse_mount *fm)
1946{
1947 struct fuse_conn *fc = fm->fc;
1948 bool last = false;
1949
1950 down_write(&fc->killsb);
1951 list_del_init(&fm->fc_entry);
1952 if (list_empty(&fc->mounts))
1953 last = true;
1954 up_write(&fc->killsb);
1955
1956 return last;
1957}
1958EXPORT_SYMBOL_GPL(fuse_mount_remove);
1959
1960void fuse_conn_destroy(struct fuse_mount *fm)
1961{
1962 struct fuse_conn *fc = fm->fc;
1963
1964 if (fc->destroy)
1965 fuse_send_destroy(fm);
1966
1967 fuse_chan_abort(fc->chan, false);
1968 fuse_chan_wait_aborted(fc->chan);
1969
1970 if (!list_empty(&fc->entry)) {
1971 mutex_lock(&fuse_mutex);
1972 list_del(&fc->entry);
1973 fuse_ctl_remove_conn(fc);
1974 mutex_unlock(&fuse_mutex);
1975 }
1976}
1977EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1978
1979static void fuse_sb_destroy(struct super_block *sb)
1980{
1981 struct fuse_mount *fm = get_fuse_mount_super(sb);
1982 bool last;
1983
1984 if (sb->s_root) {
1985 last = fuse_mount_remove(fm);
1986 if (last)
1987 fuse_conn_destroy(fm);
1988 }
1989}
1990
1991void fuse_mount_destroy(struct fuse_mount *fm)
1992{
1993 fuse_conn_put(fm->fc);
1994 kfree_rcu(fm, rcu);
1995}
1996EXPORT_SYMBOL(fuse_mount_destroy);
1997
1998static void fuse_kill_sb_anon(struct super_block *sb)
1999{
2000 fuse_sb_destroy(sb);
2001 kill_anon_super(sb);
2002 fuse_mount_destroy(get_fuse_mount_super(sb));
2003}
2004
2005static struct file_system_type fuse_fs_type = {
2006 .owner = THIS_MODULE,
2007 .name = "fuse",
2008 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
2009 .init_fs_context = fuse_init_fs_context,
2010 .parameters = fuse_fs_parameters,
2011 .kill_sb = fuse_kill_sb_anon,
2012};
2013MODULE_ALIAS_FS("fuse");
2014
2015#ifdef CONFIG_BLOCK
2016static void fuse_kill_sb_blk(struct super_block *sb)
2017{
2018 fuse_sb_destroy(sb);
2019 kill_block_super(sb);
2020 fuse_mount_destroy(get_fuse_mount_super(sb));
2021}
2022
2023static struct file_system_type fuseblk_fs_type = {
2024 .owner = THIS_MODULE,
2025 .name = "fuseblk",
2026 .init_fs_context = fuse_init_fs_context,
2027 .parameters = fuse_fs_parameters,
2028 .kill_sb = fuse_kill_sb_blk,
2029 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE | FS_ALLOW_IDMAP,
2030};
2031MODULE_ALIAS_FS("fuseblk");
2032
2033static inline int register_fuseblk(void)
2034{
2035 return register_filesystem(&fuseblk_fs_type);
2036}
2037
2038static inline void unregister_fuseblk(void)
2039{
2040 unregister_filesystem(&fuseblk_fs_type);
2041}
2042#else
2043static inline int register_fuseblk(void)
2044{
2045 return 0;
2046}
2047
2048static inline void unregister_fuseblk(void)
2049{
2050}
2051#endif
2052
2053static void fuse_inode_init_once(void *foo)
2054{
2055 struct inode *inode = foo;
2056
2057 inode_init_once(inode);
2058}
2059
2060static int __init fuse_fs_init(void)
2061{
2062 int err;
2063
2064 fuse_inode_cachep = kmem_cache_create("fuse_inode",
2065 sizeof(struct fuse_inode), 0,
2066 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
2067 fuse_inode_init_once);
2068 err = -ENOMEM;
2069 if (!fuse_inode_cachep)
2070 goto out;
2071
2072 err = register_fuseblk();
2073 if (err)
2074 goto out2;
2075
2076 err = register_filesystem(&fuse_fs_type);
2077 if (err)
2078 goto out3;
2079
2080 err = fuse_sysctl_register();
2081 if (err)
2082 goto out4;
2083
2084 return 0;
2085
2086 out4:
2087 unregister_filesystem(&fuse_fs_type);
2088 out3:
2089 unregister_fuseblk();
2090 out2:
2091 kmem_cache_destroy(fuse_inode_cachep);
2092 out:
2093 return err;
2094}
2095
2096static void fuse_fs_cleanup(void)
2097{
2098 fuse_sysctl_unregister();
2099 unregister_filesystem(&fuse_fs_type);
2100 unregister_fuseblk();
2101
2102 /*
2103 * Make sure all delayed rcu free inodes are flushed before we
2104 * destroy cache.
2105 */
2106 rcu_barrier();
2107 kmem_cache_destroy(fuse_inode_cachep);
2108}
2109
2110static struct kobject *fuse_kobj;
2111
2112static int fuse_sysfs_init(void)
2113{
2114 int err;
2115
2116 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
2117 if (!fuse_kobj) {
2118 err = -ENOMEM;
2119 goto out_err;
2120 }
2121
2122 err = sysfs_create_mount_point(fuse_kobj, "connections");
2123 if (err)
2124 goto out_fuse_unregister;
2125
2126 return 0;
2127
2128 out_fuse_unregister:
2129 kobject_put(fuse_kobj);
2130 out_err:
2131 return err;
2132}
2133
2134static void fuse_sysfs_cleanup(void)
2135{
2136 sysfs_remove_mount_point(fuse_kobj, "connections");
2137 kobject_put(fuse_kobj);
2138}
2139
2140static int __init fuse_init(void)
2141{
2142 int res;
2143
2144 pr_info("init (API version %i.%i)\n",
2145 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
2146
2147 INIT_LIST_HEAD(&fuse_conn_list);
2148 res = fuse_fs_init();
2149 if (res)
2150 goto err;
2151
2152 res = fuse_dev_init();
2153 if (res)
2154 goto err_fs_cleanup;
2155
2156 res = fuse_sysfs_init();
2157 if (res)
2158 goto err_dev_cleanup;
2159
2160 res = fuse_ctl_init();
2161 if (res)
2162 goto err_sysfs_cleanup;
2163
2164 fuse_dentry_tree_init();
2165
2166 sanitize_global_limit(&max_user_bgreq);
2167 sanitize_global_limit(&max_user_congthresh);
2168
2169 return 0;
2170
2171 err_sysfs_cleanup:
2172 fuse_sysfs_cleanup();
2173 err_dev_cleanup:
2174 fuse_dev_cleanup();
2175 err_fs_cleanup:
2176 fuse_fs_cleanup();
2177 err:
2178 return res;
2179}
2180
2181static void __exit fuse_exit(void)
2182{
2183 pr_debug("exit\n");
2184
2185 fuse_dentry_tree_cleanup();
2186 fuse_ctl_cleanup();
2187 fuse_sysfs_cleanup();
2188 fuse_fs_cleanup();
2189 fuse_dev_cleanup();
2190}
2191
2192module_init(fuse_init);
2193module_exit(fuse_exit);