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 #include "dev_uring_i.h"
11
12 #include <linux/pagemap.h>
13 #include <linux/slab.h>
14 #include <linux/file.h>
15 #include <linux/seq_file.h>
16 #include <linux/init.h>
17 #include <linux/module.h>
18 #include <linux/moduleparam.h>
19 #include <linux/fs_context.h>
20 #include <linux/fs_parser.h>
21 #include <linux/statfs.h>
22 #include <linux/random.h>
23 #include <linux/sched.h>
24 #include <linux/exportfs.h>
25 #include <linux/posix_acl.h>
26 #include <linux/pid_namespace.h>
27 #include <uapi/linux/magic.h>
28
29 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
30 MODULE_DESCRIPTION("Filesystem in Userspace");
31 MODULE_LICENSE("GPL");
32
33 static struct kmem_cache *fuse_inode_cachep;
34 struct list_head fuse_conn_list;
35 DEFINE_MUTEX(fuse_mutex);
36
37 static int set_global_limit(const char *val, const struct kernel_param *kp);
38
39 unsigned int fuse_max_pages_limit = 256;
40 /* default is no timeout */
41 unsigned int fuse_default_req_timeout;
42 unsigned int fuse_max_req_timeout;
43
44 unsigned max_user_bgreq;
45 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
46 &max_user_bgreq, 0644);
47 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
48 MODULE_PARM_DESC(max_user_bgreq,
49 "Global limit for the maximum number of backgrounded requests an "
50 "unprivileged user can set");
51
52 unsigned max_user_congthresh;
53 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
54 &max_user_congthresh, 0644);
55 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
56 MODULE_PARM_DESC(max_user_congthresh,
57 "Global limit for the maximum congestion threshold an "
58 "unprivileged user can set");
59
60 #define FUSE_DEFAULT_BLKSIZE 512
61
62 /** Maximum number of outstanding background requests */
63 #define FUSE_DEFAULT_MAX_BACKGROUND 12
64
65 /** Congestion starts at 75% of maximum */
66 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
67
68 #ifdef CONFIG_BLOCK
69 static struct file_system_type fuseblk_fs_type;
70 #endif
71
fuse_alloc_forget(void)72 struct fuse_forget_link *fuse_alloc_forget(void)
73 {
74 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
75 }
76
fuse_alloc_submount_lookup(void)77 static struct fuse_submount_lookup *fuse_alloc_submount_lookup(void)
78 {
79 struct fuse_submount_lookup *sl;
80
81 sl = kzalloc(sizeof(struct fuse_submount_lookup), GFP_KERNEL_ACCOUNT);
82 if (!sl)
83 return NULL;
84 sl->forget = fuse_alloc_forget();
85 if (!sl->forget)
86 goto out_free;
87
88 return sl;
89
90 out_free:
91 kfree(sl);
92 return NULL;
93 }
94
fuse_alloc_inode(struct super_block * sb)95 static struct inode *fuse_alloc_inode(struct super_block *sb)
96 {
97 struct fuse_inode *fi;
98
99 fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL);
100 if (!fi)
101 return NULL;
102
103 fi->i_time = 0;
104 fi->inval_mask = ~0;
105 fi->nodeid = 0;
106 fi->nlookup = 0;
107 fi->attr_version = 0;
108 fi->orig_ino = 0;
109 fi->state = 0;
110 fi->submount_lookup = NULL;
111 mutex_init(&fi->mutex);
112 spin_lock_init(&fi->lock);
113 fi->forget = fuse_alloc_forget();
114 if (!fi->forget)
115 goto out_free;
116
117 if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
118 goto out_free_forget;
119
120 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
121 fuse_inode_backing_set(fi, NULL);
122
123 return &fi->inode;
124
125 out_free_forget:
126 kfree(fi->forget);
127 out_free:
128 kmem_cache_free(fuse_inode_cachep, fi);
129 return NULL;
130 }
131
fuse_free_inode(struct inode * inode)132 static void fuse_free_inode(struct inode *inode)
133 {
134 struct fuse_inode *fi = get_fuse_inode(inode);
135
136 mutex_destroy(&fi->mutex);
137 kfree(fi->forget);
138 #ifdef CONFIG_FUSE_DAX
139 kfree(fi->dax);
140 #endif
141 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
142 fuse_backing_put(fuse_inode_backing(fi));
143
144 kmem_cache_free(fuse_inode_cachep, fi);
145 }
146
fuse_cleanup_submount_lookup(struct fuse_conn * fc,struct fuse_submount_lookup * sl)147 static void fuse_cleanup_submount_lookup(struct fuse_conn *fc,
148 struct fuse_submount_lookup *sl)
149 {
150 if (!refcount_dec_and_test(&sl->count))
151 return;
152
153 fuse_queue_forget(fc, sl->forget, sl->nodeid, 1);
154 sl->forget = NULL;
155 kfree(sl);
156 }
157
fuse_evict_inode(struct inode * inode)158 static void fuse_evict_inode(struct inode *inode)
159 {
160 struct fuse_inode *fi = get_fuse_inode(inode);
161
162 /* Will write inode on close/munmap and in all other dirtiers */
163 WARN_ON(inode->i_state & I_DIRTY_INODE);
164
165 truncate_inode_pages_final(&inode->i_data);
166 clear_inode(inode);
167 if (inode->i_sb->s_flags & SB_ACTIVE) {
168 struct fuse_conn *fc = get_fuse_conn(inode);
169
170 if (FUSE_IS_DAX(inode))
171 fuse_dax_inode_cleanup(inode);
172 if (fi->nlookup) {
173 fuse_queue_forget(fc, fi->forget, fi->nodeid,
174 fi->nlookup);
175 fi->forget = NULL;
176 }
177
178 if (fi->submount_lookup) {
179 fuse_cleanup_submount_lookup(fc, fi->submount_lookup);
180 fi->submount_lookup = NULL;
181 }
182 /*
183 * Evict of non-deleted inode may race with outstanding
184 * LOOKUP/READDIRPLUS requests and result in inconsistency when
185 * the request finishes. Deal with that here by bumping a
186 * counter that can be compared to the starting value.
187 */
188 if (inode->i_nlink > 0)
189 atomic64_inc(&fc->evict_ctr);
190 }
191 if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
192 WARN_ON(fi->iocachectr != 0);
193 WARN_ON(!list_empty(&fi->write_files));
194 WARN_ON(!list_empty(&fi->queued_writes));
195 }
196 }
197
fuse_reconfigure(struct fs_context * fsc)198 static int fuse_reconfigure(struct fs_context *fsc)
199 {
200 struct super_block *sb = fsc->root->d_sb;
201
202 sync_filesystem(sb);
203 if (fsc->sb_flags & SB_MANDLOCK)
204 return -EINVAL;
205
206 return 0;
207 }
208
209 /*
210 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
211 * so that it will fit.
212 */
fuse_squash_ino(u64 ino64)213 static ino_t fuse_squash_ino(u64 ino64)
214 {
215 ino_t ino = (ino_t) ino64;
216 if (sizeof(ino_t) < sizeof(u64))
217 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
218 return ino;
219 }
220
fuse_change_attributes_common(struct inode * inode,struct fuse_attr * attr,struct fuse_statx * sx,u64 attr_valid,u32 cache_mask,u64 evict_ctr)221 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
222 struct fuse_statx *sx,
223 u64 attr_valid, u32 cache_mask,
224 u64 evict_ctr)
225 {
226 struct fuse_conn *fc = get_fuse_conn(inode);
227 struct fuse_inode *fi = get_fuse_inode(inode);
228
229 lockdep_assert_held(&fi->lock);
230
231 /*
232 * Clear basic stats from invalid mask.
233 *
234 * Don't do this if this is coming from a fuse_iget() call and there
235 * might have been a racing evict which would've invalidated the result
236 * if the attr_version would've been preserved.
237 *
238 * !evict_ctr -> this is create
239 * fi->attr_version != 0 -> this is not a new inode
240 * evict_ctr == fuse_get_evict_ctr() -> no evicts while during request
241 */
242 if (!evict_ctr || fi->attr_version || evict_ctr == fuse_get_evict_ctr(fc))
243 set_mask_bits(&fi->inval_mask, STATX_BASIC_STATS, 0);
244
245 fi->attr_version = atomic64_inc_return(&fc->attr_version);
246 fi->i_time = attr_valid;
247
248 inode->i_ino = fuse_squash_ino(attr->ino);
249 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
250 set_nlink(inode, attr->nlink);
251 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
252 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
253 inode->i_blocks = attr->blocks;
254
255 /* Sanitize nsecs */
256 attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
257 attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
258 attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
259
260 inode_set_atime(inode, attr->atime, attr->atimensec);
261 /* mtime from server may be stale due to local buffered write */
262 if (!(cache_mask & STATX_MTIME)) {
263 inode_set_mtime(inode, attr->mtime, attr->mtimensec);
264 }
265 if (!(cache_mask & STATX_CTIME)) {
266 inode_set_ctime(inode, attr->ctime, attr->ctimensec);
267 }
268 if (sx) {
269 /* Sanitize nsecs */
270 sx->btime.tv_nsec =
271 min_t(u32, sx->btime.tv_nsec, NSEC_PER_SEC - 1);
272
273 /*
274 * Btime has been queried, cache is valid (whether or not btime
275 * is available or not) so clear STATX_BTIME from inval_mask.
276 *
277 * Availability of the btime attribute is indicated in
278 * FUSE_I_BTIME
279 */
280 set_mask_bits(&fi->inval_mask, STATX_BTIME, 0);
281 if (sx->mask & STATX_BTIME) {
282 set_bit(FUSE_I_BTIME, &fi->state);
283 fi->i_btime.tv_sec = sx->btime.tv_sec;
284 fi->i_btime.tv_nsec = sx->btime.tv_nsec;
285 }
286 }
287
288 if (attr->blksize != 0)
289 inode->i_blkbits = ilog2(attr->blksize);
290 else
291 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
292
293 /*
294 * Don't set the sticky bit in i_mode, unless we want the VFS
295 * to check permissions. This prevents failures due to the
296 * check in may_delete().
297 */
298 fi->orig_i_mode = inode->i_mode;
299 if (!fc->default_permissions)
300 inode->i_mode &= ~S_ISVTX;
301
302 fi->orig_ino = attr->ino;
303
304 /*
305 * We are refreshing inode data and it is possible that another
306 * client set suid/sgid or security.capability xattr. So clear
307 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
308 * was set or if security.capability xattr was set. But we don't
309 * know if security.capability has been set or not. So clear it
310 * anyway. Its less efficient but should be safe.
311 */
312 inode->i_flags &= ~S_NOSEC;
313 }
314
fuse_get_cache_mask(struct inode * inode)315 u32 fuse_get_cache_mask(struct inode *inode)
316 {
317 struct fuse_conn *fc = get_fuse_conn(inode);
318
319 if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
320 return 0;
321
322 return STATX_MTIME | STATX_CTIME | STATX_SIZE;
323 }
324
fuse_change_attributes_i(struct inode * inode,struct fuse_attr * attr,struct fuse_statx * sx,u64 attr_valid,u64 attr_version,u64 evict_ctr)325 static void fuse_change_attributes_i(struct inode *inode, struct fuse_attr *attr,
326 struct fuse_statx *sx, u64 attr_valid,
327 u64 attr_version, u64 evict_ctr)
328 {
329 struct fuse_conn *fc = get_fuse_conn(inode);
330 struct fuse_inode *fi = get_fuse_inode(inode);
331 u32 cache_mask;
332 loff_t oldsize;
333 struct timespec64 old_mtime;
334
335 spin_lock(&fi->lock);
336 /*
337 * In case of writeback_cache enabled, writes update mtime, ctime and
338 * may update i_size. In these cases trust the cached value in the
339 * inode.
340 */
341 cache_mask = fuse_get_cache_mask(inode);
342 if (cache_mask & STATX_SIZE)
343 attr->size = i_size_read(inode);
344
345 if (cache_mask & STATX_MTIME) {
346 attr->mtime = inode_get_mtime_sec(inode);
347 attr->mtimensec = inode_get_mtime_nsec(inode);
348 }
349 if (cache_mask & STATX_CTIME) {
350 attr->ctime = inode_get_ctime_sec(inode);
351 attr->ctimensec = inode_get_ctime_nsec(inode);
352 }
353
354 if ((attr_version != 0 && fi->attr_version > attr_version) ||
355 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
356 spin_unlock(&fi->lock);
357 return;
358 }
359
360 old_mtime = inode_get_mtime(inode);
361 fuse_change_attributes_common(inode, attr, sx, attr_valid, cache_mask,
362 evict_ctr);
363
364 oldsize = inode->i_size;
365 /*
366 * In case of writeback_cache enabled, the cached writes beyond EOF
367 * extend local i_size without keeping userspace server in sync. So,
368 * attr->size coming from server can be stale. We cannot trust it.
369 */
370 if (!(cache_mask & STATX_SIZE))
371 i_size_write(inode, attr->size);
372 spin_unlock(&fi->lock);
373
374 if (!cache_mask && S_ISREG(inode->i_mode)) {
375 bool inval = false;
376
377 if (oldsize != attr->size) {
378 truncate_pagecache(inode, attr->size);
379 if (!fc->explicit_inval_data)
380 inval = true;
381 } else if (fc->auto_inval_data) {
382 struct timespec64 new_mtime = {
383 .tv_sec = attr->mtime,
384 .tv_nsec = attr->mtimensec,
385 };
386
387 /*
388 * Auto inval mode also checks and invalidates if mtime
389 * has changed.
390 */
391 if (!timespec64_equal(&old_mtime, &new_mtime))
392 inval = true;
393 }
394
395 if (inval)
396 invalidate_inode_pages2(inode->i_mapping);
397 }
398
399 if (IS_ENABLED(CONFIG_FUSE_DAX))
400 fuse_dax_dontcache(inode, attr->flags);
401 }
402
fuse_change_attributes(struct inode * inode,struct fuse_attr * attr,struct fuse_statx * sx,u64 attr_valid,u64 attr_version)403 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
404 struct fuse_statx *sx, u64 attr_valid,
405 u64 attr_version)
406 {
407 fuse_change_attributes_i(inode, attr, sx, attr_valid, attr_version, 0);
408 }
409
fuse_init_submount_lookup(struct fuse_submount_lookup * sl,u64 nodeid)410 static void fuse_init_submount_lookup(struct fuse_submount_lookup *sl,
411 u64 nodeid)
412 {
413 sl->nodeid = nodeid;
414 refcount_set(&sl->count, 1);
415 }
416
fuse_init_inode(struct inode * inode,struct fuse_attr * attr,struct fuse_conn * fc)417 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
418 struct fuse_conn *fc)
419 {
420 inode->i_mode = attr->mode & S_IFMT;
421 inode->i_size = attr->size;
422 inode_set_mtime(inode, attr->mtime, attr->mtimensec);
423 inode_set_ctime(inode, attr->ctime, attr->ctimensec);
424 if (S_ISREG(inode->i_mode)) {
425 fuse_init_common(inode);
426 fuse_init_file_inode(inode, attr->flags);
427 } else if (S_ISDIR(inode->i_mode))
428 fuse_init_dir(inode);
429 else if (S_ISLNK(inode->i_mode))
430 fuse_init_symlink(inode);
431 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
432 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
433 fuse_init_common(inode);
434 init_special_inode(inode, inode->i_mode,
435 new_decode_dev(attr->rdev));
436 } else
437 BUG();
438 /*
439 * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
440 * so they see the exact same behavior as before.
441 */
442 if (!fc->posix_acl)
443 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
444 }
445
fuse_inode_eq(struct inode * inode,void * _nodeidp)446 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
447 {
448 u64 nodeid = *(u64 *) _nodeidp;
449 if (get_node_id(inode) == nodeid)
450 return 1;
451 else
452 return 0;
453 }
454
fuse_inode_set(struct inode * inode,void * _nodeidp)455 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
456 {
457 u64 nodeid = *(u64 *) _nodeidp;
458 get_fuse_inode(inode)->nodeid = nodeid;
459 return 0;
460 }
461
fuse_iget(struct super_block * sb,u64 nodeid,int generation,struct fuse_attr * attr,u64 attr_valid,u64 attr_version,u64 evict_ctr)462 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
463 int generation, struct fuse_attr *attr,
464 u64 attr_valid, u64 attr_version,
465 u64 evict_ctr)
466 {
467 struct inode *inode;
468 struct fuse_inode *fi;
469 struct fuse_conn *fc = get_fuse_conn_super(sb);
470
471 /*
472 * Auto mount points get their node id from the submount root, which is
473 * not a unique identifier within this filesystem.
474 *
475 * To avoid conflicts, do not place submount points into the inode hash
476 * table.
477 */
478 if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
479 S_ISDIR(attr->mode)) {
480 struct fuse_inode *fi;
481
482 inode = new_inode(sb);
483 if (!inode)
484 return NULL;
485
486 fuse_init_inode(inode, attr, fc);
487 fi = get_fuse_inode(inode);
488 fi->nodeid = nodeid;
489 fi->submount_lookup = fuse_alloc_submount_lookup();
490 if (!fi->submount_lookup) {
491 iput(inode);
492 return NULL;
493 }
494 /* Sets nlookup = 1 on fi->submount_lookup->nlookup */
495 fuse_init_submount_lookup(fi->submount_lookup, nodeid);
496 inode->i_flags |= S_AUTOMOUNT;
497 goto done;
498 }
499
500 retry:
501 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
502 if (!inode)
503 return NULL;
504
505 if ((inode->i_state & I_NEW)) {
506 inode->i_flags |= S_NOATIME;
507 if (!fc->writeback_cache || !S_ISREG(attr->mode))
508 inode->i_flags |= S_NOCMTIME;
509 inode->i_generation = generation;
510 fuse_init_inode(inode, attr, fc);
511 unlock_new_inode(inode);
512 } else if (fuse_stale_inode(inode, generation, attr)) {
513 /* nodeid was reused, any I/O on the old inode should fail */
514 fuse_make_bad(inode);
515 if (inode != d_inode(sb->s_root)) {
516 remove_inode_hash(inode);
517 iput(inode);
518 goto retry;
519 }
520 }
521 fi = get_fuse_inode(inode);
522 spin_lock(&fi->lock);
523 fi->nlookup++;
524 spin_unlock(&fi->lock);
525 done:
526 fuse_change_attributes_i(inode, attr, NULL, attr_valid, attr_version,
527 evict_ctr);
528 return inode;
529 }
530
fuse_ilookup(struct fuse_conn * fc,u64 nodeid,struct fuse_mount ** fm)531 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
532 struct fuse_mount **fm)
533 {
534 struct fuse_mount *fm_iter;
535 struct inode *inode;
536
537 WARN_ON(!rwsem_is_locked(&fc->killsb));
538 list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
539 if (!fm_iter->sb)
540 continue;
541
542 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
543 if (inode) {
544 if (fm)
545 *fm = fm_iter;
546 return inode;
547 }
548 }
549
550 return NULL;
551 }
552
fuse_reverse_inval_inode(struct fuse_conn * fc,u64 nodeid,loff_t offset,loff_t len)553 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
554 loff_t offset, loff_t len)
555 {
556 struct fuse_inode *fi;
557 struct inode *inode;
558 pgoff_t pg_start;
559 pgoff_t pg_end;
560
561 inode = fuse_ilookup(fc, nodeid, NULL);
562 if (!inode)
563 return -ENOENT;
564
565 fi = get_fuse_inode(inode);
566 spin_lock(&fi->lock);
567 fi->attr_version = atomic64_inc_return(&fc->attr_version);
568 spin_unlock(&fi->lock);
569
570 fuse_invalidate_attr(inode);
571 forget_all_cached_acls(inode);
572 if (offset >= 0) {
573 pg_start = offset >> PAGE_SHIFT;
574 if (len <= 0)
575 pg_end = -1;
576 else
577 pg_end = (offset + len - 1) >> PAGE_SHIFT;
578 invalidate_inode_pages2_range(inode->i_mapping,
579 pg_start, pg_end);
580 }
581 iput(inode);
582 return 0;
583 }
584
fuse_lock_inode(struct inode * inode)585 bool fuse_lock_inode(struct inode *inode)
586 {
587 bool locked = false;
588
589 if (!get_fuse_conn(inode)->parallel_dirops) {
590 mutex_lock(&get_fuse_inode(inode)->mutex);
591 locked = true;
592 }
593
594 return locked;
595 }
596
fuse_unlock_inode(struct inode * inode,bool locked)597 void fuse_unlock_inode(struct inode *inode, bool locked)
598 {
599 if (locked)
600 mutex_unlock(&get_fuse_inode(inode)->mutex);
601 }
602
fuse_umount_begin(struct super_block * sb)603 static void fuse_umount_begin(struct super_block *sb)
604 {
605 struct fuse_conn *fc = get_fuse_conn_super(sb);
606
607 if (fc->no_force_umount)
608 return;
609
610 fuse_abort_conn(fc);
611
612 // Only retire block-device-based superblocks.
613 if (sb->s_bdev != NULL)
614 retire_super(sb);
615 }
616
fuse_send_destroy(struct fuse_mount * fm)617 static void fuse_send_destroy(struct fuse_mount *fm)
618 {
619 if (fm->fc->conn_init) {
620 FUSE_ARGS(args);
621
622 args.opcode = FUSE_DESTROY;
623 args.force = true;
624 args.nocreds = true;
625 fuse_simple_request(fm, &args);
626 }
627 }
628
convert_fuse_statfs(struct kstatfs * stbuf,struct fuse_kstatfs * attr)629 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
630 {
631 stbuf->f_type = FUSE_SUPER_MAGIC;
632 stbuf->f_bsize = attr->bsize;
633 stbuf->f_frsize = attr->frsize;
634 stbuf->f_blocks = attr->blocks;
635 stbuf->f_bfree = attr->bfree;
636 stbuf->f_bavail = attr->bavail;
637 stbuf->f_files = attr->files;
638 stbuf->f_ffree = attr->ffree;
639 stbuf->f_namelen = attr->namelen;
640 /* fsid is left zero */
641 }
642
fuse_statfs(struct dentry * dentry,struct kstatfs * buf)643 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
644 {
645 struct super_block *sb = dentry->d_sb;
646 struct fuse_mount *fm = get_fuse_mount_super(sb);
647 FUSE_ARGS(args);
648 struct fuse_statfs_out outarg;
649 int err;
650
651 if (!fuse_allow_current_process(fm->fc)) {
652 buf->f_type = FUSE_SUPER_MAGIC;
653 return 0;
654 }
655
656 memset(&outarg, 0, sizeof(outarg));
657 args.in_numargs = 0;
658 args.opcode = FUSE_STATFS;
659 args.nodeid = get_node_id(d_inode(dentry));
660 args.out_numargs = 1;
661 args.out_args[0].size = sizeof(outarg);
662 args.out_args[0].value = &outarg;
663 err = fuse_simple_request(fm, &args);
664 if (!err)
665 convert_fuse_statfs(buf, &outarg.st);
666 return err;
667 }
668
fuse_sync_bucket_alloc(void)669 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
670 {
671 struct fuse_sync_bucket *bucket;
672
673 bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
674 if (bucket) {
675 init_waitqueue_head(&bucket->waitq);
676 /* Initial active count */
677 atomic_set(&bucket->count, 1);
678 }
679 return bucket;
680 }
681
fuse_sync_fs_writes(struct fuse_conn * fc)682 static void fuse_sync_fs_writes(struct fuse_conn *fc)
683 {
684 struct fuse_sync_bucket *bucket, *new_bucket;
685 int count;
686
687 new_bucket = fuse_sync_bucket_alloc();
688 spin_lock(&fc->lock);
689 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
690 count = atomic_read(&bucket->count);
691 WARN_ON(count < 1);
692 /* No outstanding writes? */
693 if (count == 1) {
694 spin_unlock(&fc->lock);
695 kfree(new_bucket);
696 return;
697 }
698
699 /*
700 * Completion of new bucket depends on completion of this bucket, so add
701 * one more count.
702 */
703 atomic_inc(&new_bucket->count);
704 rcu_assign_pointer(fc->curr_bucket, new_bucket);
705 spin_unlock(&fc->lock);
706 /*
707 * Drop initial active count. At this point if all writes in this and
708 * ancestor buckets complete, the count will go to zero and this task
709 * will be woken up.
710 */
711 atomic_dec(&bucket->count);
712
713 wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
714
715 /* Drop temp count on descendant bucket */
716 fuse_sync_bucket_dec(new_bucket);
717 kfree_rcu(bucket, rcu);
718 }
719
fuse_sync_fs(struct super_block * sb,int wait)720 static int fuse_sync_fs(struct super_block *sb, int wait)
721 {
722 struct fuse_mount *fm = get_fuse_mount_super(sb);
723 struct fuse_conn *fc = fm->fc;
724 struct fuse_syncfs_in inarg;
725 FUSE_ARGS(args);
726 int err;
727
728 /*
729 * Userspace cannot handle the wait == 0 case. Avoid a
730 * gratuitous roundtrip.
731 */
732 if (!wait)
733 return 0;
734
735 /* The filesystem is being unmounted. Nothing to do. */
736 if (!sb->s_root)
737 return 0;
738
739 if (!fc->sync_fs)
740 return 0;
741
742 fuse_sync_fs_writes(fc);
743
744 memset(&inarg, 0, sizeof(inarg));
745 args.in_numargs = 1;
746 args.in_args[0].size = sizeof(inarg);
747 args.in_args[0].value = &inarg;
748 args.opcode = FUSE_SYNCFS;
749 args.nodeid = get_node_id(sb->s_root->d_inode);
750 args.out_numargs = 0;
751
752 err = fuse_simple_request(fm, &args);
753 if (err == -ENOSYS) {
754 fc->sync_fs = 0;
755 err = 0;
756 }
757
758 return err;
759 }
760
761 enum {
762 OPT_SOURCE,
763 OPT_SUBTYPE,
764 OPT_FD,
765 OPT_ROOTMODE,
766 OPT_USER_ID,
767 OPT_GROUP_ID,
768 OPT_DEFAULT_PERMISSIONS,
769 OPT_ALLOW_OTHER,
770 OPT_MAX_READ,
771 OPT_BLKSIZE,
772 OPT_ERR
773 };
774
775 static const struct fs_parameter_spec fuse_fs_parameters[] = {
776 fsparam_string ("source", OPT_SOURCE),
777 fsparam_u32 ("fd", OPT_FD),
778 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
779 fsparam_uid ("user_id", OPT_USER_ID),
780 fsparam_gid ("group_id", OPT_GROUP_ID),
781 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
782 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
783 fsparam_u32 ("max_read", OPT_MAX_READ),
784 fsparam_u32 ("blksize", OPT_BLKSIZE),
785 fsparam_string ("subtype", OPT_SUBTYPE),
786 {}
787 };
788
fuse_parse_param(struct fs_context * fsc,struct fs_parameter * param)789 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
790 {
791 struct fs_parse_result result;
792 struct fuse_fs_context *ctx = fsc->fs_private;
793 int opt;
794 kuid_t kuid;
795 kgid_t kgid;
796
797 if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
798 /*
799 * Ignore options coming from mount(MS_REMOUNT) for backward
800 * compatibility.
801 */
802 if (fsc->oldapi)
803 return 0;
804
805 return invalfc(fsc, "No changes allowed in reconfigure");
806 }
807
808 opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
809 if (opt < 0)
810 return opt;
811
812 switch (opt) {
813 case OPT_SOURCE:
814 if (fsc->source)
815 return invalfc(fsc, "Multiple sources specified");
816 fsc->source = param->string;
817 param->string = NULL;
818 break;
819
820 case OPT_SUBTYPE:
821 if (ctx->subtype)
822 return invalfc(fsc, "Multiple subtypes specified");
823 ctx->subtype = param->string;
824 param->string = NULL;
825 return 0;
826
827 case OPT_FD:
828 ctx->fd = result.uint_32;
829 ctx->fd_present = true;
830 break;
831
832 case OPT_ROOTMODE:
833 if (!fuse_valid_type(result.uint_32))
834 return invalfc(fsc, "Invalid rootmode");
835 ctx->rootmode = result.uint_32;
836 ctx->rootmode_present = true;
837 break;
838
839 case OPT_USER_ID:
840 kuid = result.uid;
841 /*
842 * The requested uid must be representable in the
843 * filesystem's idmapping.
844 */
845 if (!kuid_has_mapping(fsc->user_ns, kuid))
846 return invalfc(fsc, "Invalid user_id");
847 ctx->user_id = kuid;
848 ctx->user_id_present = true;
849 break;
850
851 case OPT_GROUP_ID:
852 kgid = result.gid;
853 /*
854 * The requested gid must be representable in the
855 * filesystem's idmapping.
856 */
857 if (!kgid_has_mapping(fsc->user_ns, kgid))
858 return invalfc(fsc, "Invalid group_id");
859 ctx->group_id = kgid;
860 ctx->group_id_present = true;
861 break;
862
863 case OPT_DEFAULT_PERMISSIONS:
864 ctx->default_permissions = true;
865 break;
866
867 case OPT_ALLOW_OTHER:
868 ctx->allow_other = true;
869 break;
870
871 case OPT_MAX_READ:
872 ctx->max_read = result.uint_32;
873 break;
874
875 case OPT_BLKSIZE:
876 if (!ctx->is_bdev)
877 return invalfc(fsc, "blksize only supported for fuseblk");
878 ctx->blksize = result.uint_32;
879 break;
880
881 default:
882 return -EINVAL;
883 }
884
885 return 0;
886 }
887
fuse_free_fsc(struct fs_context * fsc)888 static void fuse_free_fsc(struct fs_context *fsc)
889 {
890 struct fuse_fs_context *ctx = fsc->fs_private;
891
892 if (ctx) {
893 kfree(ctx->subtype);
894 kfree(ctx);
895 }
896 }
897
fuse_show_options(struct seq_file * m,struct dentry * root)898 static int fuse_show_options(struct seq_file *m, struct dentry *root)
899 {
900 struct super_block *sb = root->d_sb;
901 struct fuse_conn *fc = get_fuse_conn_super(sb);
902
903 if (fc->legacy_opts_show) {
904 seq_printf(m, ",user_id=%u",
905 from_kuid_munged(fc->user_ns, fc->user_id));
906 seq_printf(m, ",group_id=%u",
907 from_kgid_munged(fc->user_ns, fc->group_id));
908 if (fc->default_permissions)
909 seq_puts(m, ",default_permissions");
910 if (fc->allow_other)
911 seq_puts(m, ",allow_other");
912 if (fc->max_read != ~0)
913 seq_printf(m, ",max_read=%u", fc->max_read);
914 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
915 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
916 }
917 #ifdef CONFIG_FUSE_DAX
918 if (fc->dax_mode == FUSE_DAX_ALWAYS)
919 seq_puts(m, ",dax=always");
920 else if (fc->dax_mode == FUSE_DAX_NEVER)
921 seq_puts(m, ",dax=never");
922 else if (fc->dax_mode == FUSE_DAX_INODE_USER)
923 seq_puts(m, ",dax=inode");
924 #endif
925
926 return 0;
927 }
928
fuse_iqueue_init(struct fuse_iqueue * fiq,const struct fuse_iqueue_ops * ops,void * priv)929 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
930 const struct fuse_iqueue_ops *ops,
931 void *priv)
932 {
933 memset(fiq, 0, sizeof(struct fuse_iqueue));
934 spin_lock_init(&fiq->lock);
935 init_waitqueue_head(&fiq->waitq);
936 INIT_LIST_HEAD(&fiq->pending);
937 INIT_LIST_HEAD(&fiq->interrupts);
938 fiq->forget_list_tail = &fiq->forget_list_head;
939 fiq->connected = 1;
940 fiq->ops = ops;
941 fiq->priv = priv;
942 }
943
fuse_pqueue_init(struct fuse_pqueue * fpq)944 void fuse_pqueue_init(struct fuse_pqueue *fpq)
945 {
946 unsigned int i;
947
948 spin_lock_init(&fpq->lock);
949 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
950 INIT_LIST_HEAD(&fpq->processing[i]);
951 INIT_LIST_HEAD(&fpq->io);
952 fpq->connected = 1;
953 }
954
fuse_conn_init(struct fuse_conn * fc,struct fuse_mount * fm,struct user_namespace * user_ns,const struct fuse_iqueue_ops * fiq_ops,void * fiq_priv)955 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
956 struct user_namespace *user_ns,
957 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
958 {
959 memset(fc, 0, sizeof(*fc));
960 spin_lock_init(&fc->lock);
961 spin_lock_init(&fc->bg_lock);
962 init_rwsem(&fc->killsb);
963 refcount_set(&fc->count, 1);
964 atomic_set(&fc->dev_count, 1);
965 init_waitqueue_head(&fc->blocked_waitq);
966 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
967 INIT_LIST_HEAD(&fc->bg_queue);
968 INIT_LIST_HEAD(&fc->entry);
969 INIT_LIST_HEAD(&fc->devices);
970 atomic_set(&fc->num_waiting, 0);
971 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
972 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
973 atomic64_set(&fc->khctr, 0);
974 fc->polled_files = RB_ROOT;
975 fc->blocked = 0;
976 fc->initialized = 0;
977 fc->connected = 1;
978 atomic64_set(&fc->attr_version, 1);
979 atomic64_set(&fc->evict_ctr, 1);
980 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
981 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
982 fc->user_ns = get_user_ns(user_ns);
983 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
984 fc->max_pages_limit = fuse_max_pages_limit;
985 fc->name_max = FUSE_NAME_LOW_MAX;
986 fc->timeout.req_timeout = 0;
987
988 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
989 fuse_backing_files_init(fc);
990
991 INIT_LIST_HEAD(&fc->mounts);
992 list_add(&fm->fc_entry, &fc->mounts);
993 fm->fc = fc;
994 }
995 EXPORT_SYMBOL_GPL(fuse_conn_init);
996
delayed_release(struct rcu_head * p)997 static void delayed_release(struct rcu_head *p)
998 {
999 struct fuse_conn *fc = container_of(p, struct fuse_conn, rcu);
1000
1001 fuse_uring_destruct(fc);
1002
1003 put_user_ns(fc->user_ns);
1004 fc->release(fc);
1005 }
1006
fuse_conn_put(struct fuse_conn * fc)1007 void fuse_conn_put(struct fuse_conn *fc)
1008 {
1009 if (refcount_dec_and_test(&fc->count)) {
1010 struct fuse_iqueue *fiq = &fc->iq;
1011 struct fuse_sync_bucket *bucket;
1012
1013 if (IS_ENABLED(CONFIG_FUSE_DAX))
1014 fuse_dax_conn_free(fc);
1015 if (fc->timeout.req_timeout)
1016 cancel_delayed_work_sync(&fc->timeout.work);
1017 if (fiq->ops->release)
1018 fiq->ops->release(fiq);
1019 put_pid_ns(fc->pid_ns);
1020 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
1021 if (bucket) {
1022 WARN_ON(atomic_read(&bucket->count) != 1);
1023 kfree(bucket);
1024 }
1025 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
1026 fuse_backing_files_free(fc);
1027 call_rcu(&fc->rcu, delayed_release);
1028 }
1029 }
1030 EXPORT_SYMBOL_GPL(fuse_conn_put);
1031
fuse_conn_get(struct fuse_conn * fc)1032 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
1033 {
1034 refcount_inc(&fc->count);
1035 return fc;
1036 }
1037 EXPORT_SYMBOL_GPL(fuse_conn_get);
1038
fuse_get_root_inode(struct super_block * sb,unsigned mode)1039 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
1040 {
1041 struct fuse_attr attr;
1042 memset(&attr, 0, sizeof(attr));
1043
1044 attr.mode = mode;
1045 attr.ino = FUSE_ROOT_ID;
1046 attr.nlink = 1;
1047 return fuse_iget(sb, FUSE_ROOT_ID, 0, &attr, 0, 0, 0);
1048 }
1049
1050 struct fuse_inode_handle {
1051 u64 nodeid;
1052 u32 generation;
1053 };
1054
fuse_get_dentry(struct super_block * sb,struct fuse_inode_handle * handle)1055 static struct dentry *fuse_get_dentry(struct super_block *sb,
1056 struct fuse_inode_handle *handle)
1057 {
1058 struct fuse_conn *fc = get_fuse_conn_super(sb);
1059 struct inode *inode;
1060 struct dentry *entry;
1061 int err = -ESTALE;
1062
1063 if (handle->nodeid == 0)
1064 goto out_err;
1065
1066 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
1067 if (!inode) {
1068 struct fuse_entry_out outarg;
1069 const struct qstr name = QSTR_INIT(".", 1);
1070
1071 if (!fc->export_support)
1072 goto out_err;
1073
1074 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
1075 &inode);
1076 if (err && err != -ENOENT)
1077 goto out_err;
1078 if (err || !inode) {
1079 err = -ESTALE;
1080 goto out_err;
1081 }
1082 err = -EIO;
1083 if (get_node_id(inode) != handle->nodeid)
1084 goto out_iput;
1085 }
1086 err = -ESTALE;
1087 if (inode->i_generation != handle->generation)
1088 goto out_iput;
1089
1090 entry = d_obtain_alias(inode);
1091 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
1092 fuse_invalidate_entry_cache(entry);
1093
1094 return entry;
1095
1096 out_iput:
1097 iput(inode);
1098 out_err:
1099 return ERR_PTR(err);
1100 }
1101
fuse_encode_fh(struct inode * inode,u32 * fh,int * max_len,struct inode * parent)1102 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
1103 struct inode *parent)
1104 {
1105 int len = parent ? 6 : 3;
1106 u64 nodeid;
1107 u32 generation;
1108
1109 if (*max_len < len) {
1110 *max_len = len;
1111 return FILEID_INVALID;
1112 }
1113
1114 nodeid = get_fuse_inode(inode)->nodeid;
1115 generation = inode->i_generation;
1116
1117 fh[0] = (u32)(nodeid >> 32);
1118 fh[1] = (u32)(nodeid & 0xffffffff);
1119 fh[2] = generation;
1120
1121 if (parent) {
1122 nodeid = get_fuse_inode(parent)->nodeid;
1123 generation = parent->i_generation;
1124
1125 fh[3] = (u32)(nodeid >> 32);
1126 fh[4] = (u32)(nodeid & 0xffffffff);
1127 fh[5] = generation;
1128 }
1129
1130 *max_len = len;
1131 return parent ? FILEID_INO64_GEN_PARENT : FILEID_INO64_GEN;
1132 }
1133
fuse_fh_to_dentry(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)1134 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
1135 struct fid *fid, int fh_len, int fh_type)
1136 {
1137 struct fuse_inode_handle handle;
1138
1139 if ((fh_type != FILEID_INO64_GEN &&
1140 fh_type != FILEID_INO64_GEN_PARENT) || fh_len < 3)
1141 return NULL;
1142
1143 handle.nodeid = (u64) fid->raw[0] << 32;
1144 handle.nodeid |= (u64) fid->raw[1];
1145 handle.generation = fid->raw[2];
1146 return fuse_get_dentry(sb, &handle);
1147 }
1148
fuse_fh_to_parent(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)1149 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1150 struct fid *fid, int fh_len, int fh_type)
1151 {
1152 struct fuse_inode_handle parent;
1153
1154 if (fh_type != FILEID_INO64_GEN_PARENT || fh_len < 6)
1155 return NULL;
1156
1157 parent.nodeid = (u64) fid->raw[3] << 32;
1158 parent.nodeid |= (u64) fid->raw[4];
1159 parent.generation = fid->raw[5];
1160 return fuse_get_dentry(sb, &parent);
1161 }
1162
fuse_get_parent(struct dentry * child)1163 static struct dentry *fuse_get_parent(struct dentry *child)
1164 {
1165 struct inode *child_inode = d_inode(child);
1166 struct fuse_conn *fc = get_fuse_conn(child_inode);
1167 struct inode *inode;
1168 struct dentry *parent;
1169 struct fuse_entry_out outarg;
1170 int err;
1171
1172 if (!fc->export_support)
1173 return ERR_PTR(-ESTALE);
1174
1175 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1176 &dotdot_name, &outarg, &inode);
1177 if (err) {
1178 if (err == -ENOENT)
1179 return ERR_PTR(-ESTALE);
1180 return ERR_PTR(err);
1181 }
1182
1183 parent = d_obtain_alias(inode);
1184 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1185 fuse_invalidate_entry_cache(parent);
1186
1187 return parent;
1188 }
1189
1190 /* only for fid encoding; no support for file handle */
1191 static const struct export_operations fuse_export_fid_operations = {
1192 .encode_fh = fuse_encode_fh,
1193 };
1194
1195 static const struct export_operations fuse_export_operations = {
1196 .fh_to_dentry = fuse_fh_to_dentry,
1197 .fh_to_parent = fuse_fh_to_parent,
1198 .encode_fh = fuse_encode_fh,
1199 .get_parent = fuse_get_parent,
1200 };
1201
1202 static const struct super_operations fuse_super_operations = {
1203 .alloc_inode = fuse_alloc_inode,
1204 .free_inode = fuse_free_inode,
1205 .evict_inode = fuse_evict_inode,
1206 .write_inode = fuse_write_inode,
1207 .drop_inode = generic_delete_inode,
1208 .umount_begin = fuse_umount_begin,
1209 .statfs = fuse_statfs,
1210 .sync_fs = fuse_sync_fs,
1211 .show_options = fuse_show_options,
1212 };
1213
sanitize_global_limit(unsigned * limit)1214 static void sanitize_global_limit(unsigned *limit)
1215 {
1216 /*
1217 * The default maximum number of async requests is calculated to consume
1218 * 1/2^13 of the total memory, assuming 392 bytes per request.
1219 */
1220 if (*limit == 0)
1221 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1222
1223 if (*limit >= 1 << 16)
1224 *limit = (1 << 16) - 1;
1225 }
1226
set_global_limit(const char * val,const struct kernel_param * kp)1227 static int set_global_limit(const char *val, const struct kernel_param *kp)
1228 {
1229 int rv;
1230
1231 rv = param_set_uint(val, kp);
1232 if (rv)
1233 return rv;
1234
1235 sanitize_global_limit((unsigned *)kp->arg);
1236
1237 return 0;
1238 }
1239
process_init_limits(struct fuse_conn * fc,struct fuse_init_out * arg)1240 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1241 {
1242 int cap_sys_admin = capable(CAP_SYS_ADMIN);
1243
1244 if (arg->minor < 13)
1245 return;
1246
1247 sanitize_global_limit(&max_user_bgreq);
1248 sanitize_global_limit(&max_user_congthresh);
1249
1250 spin_lock(&fc->bg_lock);
1251 if (arg->max_background) {
1252 fc->max_background = arg->max_background;
1253
1254 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1255 fc->max_background = max_user_bgreq;
1256 }
1257 if (arg->congestion_threshold) {
1258 fc->congestion_threshold = arg->congestion_threshold;
1259
1260 if (!cap_sys_admin &&
1261 fc->congestion_threshold > max_user_congthresh)
1262 fc->congestion_threshold = max_user_congthresh;
1263 }
1264 spin_unlock(&fc->bg_lock);
1265 }
1266
set_request_timeout(struct fuse_conn * fc,unsigned int timeout)1267 static void set_request_timeout(struct fuse_conn *fc, unsigned int timeout)
1268 {
1269 fc->timeout.req_timeout = secs_to_jiffies(timeout);
1270 INIT_DELAYED_WORK(&fc->timeout.work, fuse_check_timeout);
1271 queue_delayed_work(system_wq, &fc->timeout.work,
1272 fuse_timeout_timer_freq);
1273 }
1274
init_server_timeout(struct fuse_conn * fc,unsigned int timeout)1275 static void init_server_timeout(struct fuse_conn *fc, unsigned int timeout)
1276 {
1277 if (!timeout && !fuse_max_req_timeout && !fuse_default_req_timeout)
1278 return;
1279
1280 if (!timeout)
1281 timeout = fuse_default_req_timeout;
1282
1283 if (fuse_max_req_timeout) {
1284 if (timeout)
1285 timeout = min(fuse_max_req_timeout, timeout);
1286 else
1287 timeout = fuse_max_req_timeout;
1288 }
1289
1290 timeout = max(FUSE_TIMEOUT_TIMER_FREQ, timeout);
1291
1292 set_request_timeout(fc, timeout);
1293 }
1294
1295 struct fuse_init_args {
1296 struct fuse_args args;
1297 struct fuse_init_in in;
1298 struct fuse_init_out out;
1299 };
1300
process_init_reply(struct fuse_mount * fm,struct fuse_args * args,int error)1301 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1302 int error)
1303 {
1304 struct fuse_conn *fc = fm->fc;
1305 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1306 struct fuse_init_out *arg = &ia->out;
1307 bool ok = true;
1308
1309 if (error || arg->major != FUSE_KERNEL_VERSION)
1310 ok = false;
1311 else {
1312 unsigned long ra_pages;
1313 unsigned int timeout = 0;
1314
1315 process_init_limits(fc, arg);
1316
1317 if (arg->minor >= 6) {
1318 u64 flags = arg->flags;
1319
1320 if (flags & FUSE_INIT_EXT)
1321 flags |= (u64) arg->flags2 << 32;
1322
1323 ra_pages = arg->max_readahead / PAGE_SIZE;
1324 if (flags & FUSE_ASYNC_READ)
1325 fc->async_read = 1;
1326 if (!(flags & FUSE_POSIX_LOCKS))
1327 fc->no_lock = 1;
1328 if (arg->minor >= 17) {
1329 if (!(flags & FUSE_FLOCK_LOCKS))
1330 fc->no_flock = 1;
1331 } else {
1332 if (!(flags & FUSE_POSIX_LOCKS))
1333 fc->no_flock = 1;
1334 }
1335 if (flags & FUSE_ATOMIC_O_TRUNC)
1336 fc->atomic_o_trunc = 1;
1337 if (arg->minor >= 9) {
1338 /* LOOKUP has dependency on proto version */
1339 if (flags & FUSE_EXPORT_SUPPORT)
1340 fc->export_support = 1;
1341 }
1342 if (flags & FUSE_BIG_WRITES)
1343 fc->big_writes = 1;
1344 if (flags & FUSE_DONT_MASK)
1345 fc->dont_mask = 1;
1346 if (flags & FUSE_AUTO_INVAL_DATA)
1347 fc->auto_inval_data = 1;
1348 else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1349 fc->explicit_inval_data = 1;
1350 if (flags & FUSE_DO_READDIRPLUS) {
1351 fc->do_readdirplus = 1;
1352 if (flags & FUSE_READDIRPLUS_AUTO)
1353 fc->readdirplus_auto = 1;
1354 }
1355 if (flags & FUSE_ASYNC_DIO)
1356 fc->async_dio = 1;
1357 if (flags & FUSE_WRITEBACK_CACHE)
1358 fc->writeback_cache = 1;
1359 if (flags & FUSE_PARALLEL_DIROPS)
1360 fc->parallel_dirops = 1;
1361 if (flags & FUSE_HANDLE_KILLPRIV)
1362 fc->handle_killpriv = 1;
1363 if (arg->time_gran && arg->time_gran <= 1000000000)
1364 fm->sb->s_time_gran = arg->time_gran;
1365 if ((flags & FUSE_POSIX_ACL)) {
1366 fc->default_permissions = 1;
1367 fc->posix_acl = 1;
1368 }
1369 if (flags & FUSE_CACHE_SYMLINKS)
1370 fc->cache_symlinks = 1;
1371 if (flags & FUSE_ABORT_ERROR)
1372 fc->abort_err = 1;
1373 if (flags & FUSE_MAX_PAGES) {
1374 fc->max_pages =
1375 min_t(unsigned int, fc->max_pages_limit,
1376 max_t(unsigned int, arg->max_pages, 1));
1377
1378 /*
1379 * PATH_MAX file names might need two pages for
1380 * ops like rename
1381 */
1382 if (fc->max_pages > 1)
1383 fc->name_max = FUSE_NAME_MAX;
1384 }
1385 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1386 if (flags & FUSE_MAP_ALIGNMENT &&
1387 !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1388 ok = false;
1389 }
1390 if (flags & FUSE_HAS_INODE_DAX)
1391 fc->inode_dax = 1;
1392 }
1393 if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1394 fc->handle_killpriv_v2 = 1;
1395 fm->sb->s_flags |= SB_NOSEC;
1396 }
1397 if (flags & FUSE_SETXATTR_EXT)
1398 fc->setxattr_ext = 1;
1399 if (flags & FUSE_SECURITY_CTX)
1400 fc->init_security = 1;
1401 if (flags & FUSE_CREATE_SUPP_GROUP)
1402 fc->create_supp_group = 1;
1403 if (flags & FUSE_DIRECT_IO_ALLOW_MMAP)
1404 fc->direct_io_allow_mmap = 1;
1405 /*
1406 * max_stack_depth is the max stack depth of FUSE fs,
1407 * so it has to be at least 1 to support passthrough
1408 * to backing files.
1409 *
1410 * with max_stack_depth > 1, the backing files can be
1411 * on a stacked fs (e.g. overlayfs) themselves and with
1412 * max_stack_depth == 1, FUSE fs can be stacked as the
1413 * underlying fs of a stacked fs (e.g. overlayfs).
1414 *
1415 * Also don't allow the combination of FUSE_PASSTHROUGH
1416 * and FUSE_WRITEBACK_CACHE, current design doesn't handle
1417 * them together.
1418 */
1419 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH) &&
1420 (flags & FUSE_PASSTHROUGH) &&
1421 arg->max_stack_depth > 0 &&
1422 arg->max_stack_depth <= FILESYSTEM_MAX_STACK_DEPTH &&
1423 !(flags & FUSE_WRITEBACK_CACHE)) {
1424 fc->passthrough = 1;
1425 fc->max_stack_depth = arg->max_stack_depth;
1426 fm->sb->s_stack_depth = arg->max_stack_depth;
1427 }
1428 if (flags & FUSE_NO_EXPORT_SUPPORT)
1429 fm->sb->s_export_op = &fuse_export_fid_operations;
1430 if (flags & FUSE_ALLOW_IDMAP) {
1431 if (fc->default_permissions)
1432 fm->sb->s_iflags &= ~SB_I_NOIDMAP;
1433 else
1434 ok = false;
1435 }
1436 if (flags & FUSE_OVER_IO_URING && fuse_uring_enabled())
1437 fc->io_uring = 1;
1438
1439 if (flags & FUSE_REQUEST_TIMEOUT)
1440 timeout = arg->request_timeout;
1441 } else {
1442 ra_pages = fc->max_read / PAGE_SIZE;
1443 fc->no_lock = 1;
1444 fc->no_flock = 1;
1445 }
1446
1447 init_server_timeout(fc, timeout);
1448
1449 fm->sb->s_bdi->ra_pages =
1450 min(fm->sb->s_bdi->ra_pages, ra_pages);
1451 fc->minor = arg->minor;
1452 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1453 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1454 fc->conn_init = 1;
1455 }
1456 kfree(ia);
1457
1458 if (!ok) {
1459 fc->conn_init = 0;
1460 fc->conn_error = 1;
1461 }
1462
1463 fuse_set_initialized(fc);
1464 wake_up_all(&fc->blocked_waitq);
1465 }
1466
fuse_send_init(struct fuse_mount * fm)1467 void fuse_send_init(struct fuse_mount *fm)
1468 {
1469 struct fuse_init_args *ia;
1470 u64 flags;
1471
1472 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1473
1474 ia->in.major = FUSE_KERNEL_VERSION;
1475 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1476 ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1477 flags =
1478 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1479 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1480 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1481 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1482 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1483 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1484 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1485 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1486 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1487 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1488 FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP |
1489 FUSE_HAS_EXPIRE_ONLY | FUSE_DIRECT_IO_ALLOW_MMAP |
1490 FUSE_NO_EXPORT_SUPPORT | FUSE_HAS_RESEND | FUSE_ALLOW_IDMAP |
1491 FUSE_REQUEST_TIMEOUT;
1492 #ifdef CONFIG_FUSE_DAX
1493 if (fm->fc->dax)
1494 flags |= FUSE_MAP_ALIGNMENT;
1495 if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1496 flags |= FUSE_HAS_INODE_DAX;
1497 #endif
1498 if (fm->fc->auto_submounts)
1499 flags |= FUSE_SUBMOUNTS;
1500 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
1501 flags |= FUSE_PASSTHROUGH;
1502
1503 /*
1504 * This is just an information flag for fuse server. No need to check
1505 * the reply - server is either sending IORING_OP_URING_CMD or not.
1506 */
1507 if (fuse_uring_enabled())
1508 flags |= FUSE_OVER_IO_URING;
1509
1510 ia->in.flags = flags;
1511 ia->in.flags2 = flags >> 32;
1512
1513 ia->args.opcode = FUSE_INIT;
1514 ia->args.in_numargs = 1;
1515 ia->args.in_args[0].size = sizeof(ia->in);
1516 ia->args.in_args[0].value = &ia->in;
1517 ia->args.out_numargs = 1;
1518 /* Variable length argument used for backward compatibility
1519 with interface version < 7.5. Rest of init_out is zeroed
1520 by do_get_request(), so a short reply is not a problem */
1521 ia->args.out_argvar = true;
1522 ia->args.out_args[0].size = sizeof(ia->out);
1523 ia->args.out_args[0].value = &ia->out;
1524 ia->args.force = true;
1525 ia->args.nocreds = true;
1526 ia->args.end = process_init_reply;
1527
1528 if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1529 process_init_reply(fm, &ia->args, -ENOTCONN);
1530 }
1531 EXPORT_SYMBOL_GPL(fuse_send_init);
1532
fuse_free_conn(struct fuse_conn * fc)1533 void fuse_free_conn(struct fuse_conn *fc)
1534 {
1535 WARN_ON(!list_empty(&fc->devices));
1536 kfree(fc);
1537 }
1538 EXPORT_SYMBOL_GPL(fuse_free_conn);
1539
fuse_bdi_init(struct fuse_conn * fc,struct super_block * sb)1540 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1541 {
1542 int err;
1543 char *suffix = "";
1544
1545 if (sb->s_bdev) {
1546 suffix = "-fuseblk";
1547 /*
1548 * sb->s_bdi points to blkdev's bdi however we want to redirect
1549 * it to our private bdi...
1550 */
1551 bdi_put(sb->s_bdi);
1552 sb->s_bdi = &noop_backing_dev_info;
1553 }
1554 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1555 MINOR(fc->dev), suffix);
1556 if (err)
1557 return err;
1558
1559 /* fuse does it's own writeback accounting */
1560 sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1561 sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1562
1563 /*
1564 * For a single fuse filesystem use max 1% of dirty +
1565 * writeback threshold.
1566 *
1567 * This gives about 1M of write buffer for memory maps on a
1568 * machine with 1G and 10% dirty_ratio, which should be more
1569 * than enough.
1570 *
1571 * Privileged users can raise it by writing to
1572 *
1573 * /sys/class/bdi/<bdi>/max_ratio
1574 */
1575 bdi_set_max_ratio(sb->s_bdi, 1);
1576
1577 return 0;
1578 }
1579
fuse_dev_alloc(void)1580 struct fuse_dev *fuse_dev_alloc(void)
1581 {
1582 struct fuse_dev *fud;
1583 struct list_head *pq;
1584
1585 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1586 if (!fud)
1587 return NULL;
1588
1589 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1590 if (!pq) {
1591 kfree(fud);
1592 return NULL;
1593 }
1594
1595 fud->pq.processing = pq;
1596 fuse_pqueue_init(&fud->pq);
1597
1598 return fud;
1599 }
1600 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1601
fuse_dev_install(struct fuse_dev * fud,struct fuse_conn * fc)1602 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1603 {
1604 fud->fc = fuse_conn_get(fc);
1605 spin_lock(&fc->lock);
1606 list_add_tail(&fud->entry, &fc->devices);
1607 spin_unlock(&fc->lock);
1608 }
1609 EXPORT_SYMBOL_GPL(fuse_dev_install);
1610
fuse_dev_alloc_install(struct fuse_conn * fc)1611 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1612 {
1613 struct fuse_dev *fud;
1614
1615 fud = fuse_dev_alloc();
1616 if (!fud)
1617 return NULL;
1618
1619 fuse_dev_install(fud, fc);
1620 return fud;
1621 }
1622 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1623
fuse_dev_free(struct fuse_dev * fud)1624 void fuse_dev_free(struct fuse_dev *fud)
1625 {
1626 struct fuse_conn *fc = fud->fc;
1627
1628 if (fc) {
1629 spin_lock(&fc->lock);
1630 list_del(&fud->entry);
1631 spin_unlock(&fc->lock);
1632
1633 fuse_conn_put(fc);
1634 }
1635 kfree(fud->pq.processing);
1636 kfree(fud);
1637 }
1638 EXPORT_SYMBOL_GPL(fuse_dev_free);
1639
fuse_fill_attr_from_inode(struct fuse_attr * attr,const struct fuse_inode * fi)1640 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1641 const struct fuse_inode *fi)
1642 {
1643 struct timespec64 atime = inode_get_atime(&fi->inode);
1644 struct timespec64 mtime = inode_get_mtime(&fi->inode);
1645 struct timespec64 ctime = inode_get_ctime(&fi->inode);
1646
1647 *attr = (struct fuse_attr){
1648 .ino = fi->inode.i_ino,
1649 .size = fi->inode.i_size,
1650 .blocks = fi->inode.i_blocks,
1651 .atime = atime.tv_sec,
1652 .mtime = mtime.tv_sec,
1653 .ctime = ctime.tv_sec,
1654 .atimensec = atime.tv_nsec,
1655 .mtimensec = mtime.tv_nsec,
1656 .ctimensec = ctime.tv_nsec,
1657 .mode = fi->inode.i_mode,
1658 .nlink = fi->inode.i_nlink,
1659 .uid = __kuid_val(fi->inode.i_uid),
1660 .gid = __kgid_val(fi->inode.i_gid),
1661 .rdev = fi->inode.i_rdev,
1662 .blksize = 1u << fi->inode.i_blkbits,
1663 };
1664 }
1665
fuse_sb_defaults(struct super_block * sb)1666 static void fuse_sb_defaults(struct super_block *sb)
1667 {
1668 sb->s_magic = FUSE_SUPER_MAGIC;
1669 sb->s_op = &fuse_super_operations;
1670 sb->s_xattr = fuse_xattr_handlers;
1671 sb->s_maxbytes = MAX_LFS_FILESIZE;
1672 sb->s_time_gran = 1;
1673 sb->s_export_op = &fuse_export_operations;
1674 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1675 sb->s_iflags |= SB_I_NOIDMAP;
1676 if (sb->s_user_ns != &init_user_ns)
1677 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1678 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1679 }
1680
fuse_fill_super_submount(struct super_block * sb,struct fuse_inode * parent_fi)1681 static int fuse_fill_super_submount(struct super_block *sb,
1682 struct fuse_inode *parent_fi)
1683 {
1684 struct fuse_mount *fm = get_fuse_mount_super(sb);
1685 struct super_block *parent_sb = parent_fi->inode.i_sb;
1686 struct fuse_attr root_attr;
1687 struct inode *root;
1688 struct fuse_submount_lookup *sl;
1689 struct fuse_inode *fi;
1690
1691 fuse_sb_defaults(sb);
1692 fm->sb = sb;
1693
1694 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1695 sb->s_bdi = bdi_get(parent_sb->s_bdi);
1696
1697 sb->s_xattr = parent_sb->s_xattr;
1698 sb->s_export_op = parent_sb->s_export_op;
1699 sb->s_time_gran = parent_sb->s_time_gran;
1700 sb->s_blocksize = parent_sb->s_blocksize;
1701 sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1702 sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1703 if (parent_sb->s_subtype && !sb->s_subtype)
1704 return -ENOMEM;
1705
1706 fuse_fill_attr_from_inode(&root_attr, parent_fi);
1707 root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0,
1708 fuse_get_evict_ctr(fm->fc));
1709 /*
1710 * This inode is just a duplicate, so it is not looked up and
1711 * its nlookup should not be incremented. fuse_iget() does
1712 * that, though, so undo it here.
1713 */
1714 fi = get_fuse_inode(root);
1715 fi->nlookup--;
1716
1717 sb->s_d_op = &fuse_dentry_operations;
1718 sb->s_root = d_make_root(root);
1719 if (!sb->s_root)
1720 return -ENOMEM;
1721
1722 /*
1723 * Grab the parent's submount_lookup pointer and take a
1724 * reference on the shared nlookup from the parent. This is to
1725 * prevent the last forget for this nodeid from getting
1726 * triggered until all users have finished with it.
1727 */
1728 sl = parent_fi->submount_lookup;
1729 WARN_ON(!sl);
1730 if (sl) {
1731 refcount_inc(&sl->count);
1732 fi->submount_lookup = sl;
1733 }
1734
1735 return 0;
1736 }
1737
1738 /* Filesystem context private data holds the FUSE inode of the mount point */
fuse_get_tree_submount(struct fs_context * fsc)1739 static int fuse_get_tree_submount(struct fs_context *fsc)
1740 {
1741 struct fuse_mount *fm;
1742 struct fuse_inode *mp_fi = fsc->fs_private;
1743 struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1744 struct super_block *sb;
1745 int err;
1746
1747 fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1748 if (!fm)
1749 return -ENOMEM;
1750
1751 fm->fc = fuse_conn_get(fc);
1752 fsc->s_fs_info = fm;
1753 sb = sget_fc(fsc, NULL, set_anon_super_fc);
1754 if (fsc->s_fs_info)
1755 fuse_mount_destroy(fm);
1756 if (IS_ERR(sb))
1757 return PTR_ERR(sb);
1758
1759 /* Initialize superblock, making @mp_fi its root */
1760 err = fuse_fill_super_submount(sb, mp_fi);
1761 if (err) {
1762 deactivate_locked_super(sb);
1763 return err;
1764 }
1765
1766 down_write(&fc->killsb);
1767 list_add_tail(&fm->fc_entry, &fc->mounts);
1768 up_write(&fc->killsb);
1769
1770 sb->s_flags |= SB_ACTIVE;
1771 fsc->root = dget(sb->s_root);
1772
1773 return 0;
1774 }
1775
1776 static const struct fs_context_operations fuse_context_submount_ops = {
1777 .get_tree = fuse_get_tree_submount,
1778 };
1779
fuse_init_fs_context_submount(struct fs_context * fsc)1780 int fuse_init_fs_context_submount(struct fs_context *fsc)
1781 {
1782 fsc->ops = &fuse_context_submount_ops;
1783 return 0;
1784 }
1785 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1786
fuse_fill_super_common(struct super_block * sb,struct fuse_fs_context * ctx)1787 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1788 {
1789 struct fuse_dev *fud = NULL;
1790 struct fuse_mount *fm = get_fuse_mount_super(sb);
1791 struct fuse_conn *fc = fm->fc;
1792 struct inode *root;
1793 struct dentry *root_dentry;
1794 int err;
1795
1796 err = -EINVAL;
1797 if (sb->s_flags & SB_MANDLOCK)
1798 goto err;
1799
1800 rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1801 fuse_sb_defaults(sb);
1802
1803 if (ctx->is_bdev) {
1804 #ifdef CONFIG_BLOCK
1805 err = -EINVAL;
1806 if (!sb_set_blocksize(sb, ctx->blksize))
1807 goto err;
1808 #endif
1809 } else {
1810 sb->s_blocksize = PAGE_SIZE;
1811 sb->s_blocksize_bits = PAGE_SHIFT;
1812 }
1813
1814 sb->s_subtype = ctx->subtype;
1815 ctx->subtype = NULL;
1816 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1817 err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1818 if (err)
1819 goto err;
1820 }
1821
1822 if (ctx->fudptr) {
1823 err = -ENOMEM;
1824 fud = fuse_dev_alloc_install(fc);
1825 if (!fud)
1826 goto err_free_dax;
1827 }
1828
1829 fc->dev = sb->s_dev;
1830 fm->sb = sb;
1831 err = fuse_bdi_init(fc, sb);
1832 if (err)
1833 goto err_dev_free;
1834
1835 /* Handle umasking inside the fuse code */
1836 if (sb->s_flags & SB_POSIXACL)
1837 fc->dont_mask = 1;
1838 sb->s_flags |= SB_POSIXACL;
1839
1840 fc->default_permissions = ctx->default_permissions;
1841 fc->allow_other = ctx->allow_other;
1842 fc->user_id = ctx->user_id;
1843 fc->group_id = ctx->group_id;
1844 fc->legacy_opts_show = ctx->legacy_opts_show;
1845 fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1846 fc->destroy = ctx->destroy;
1847 fc->no_control = ctx->no_control;
1848 fc->no_force_umount = ctx->no_force_umount;
1849
1850 err = -ENOMEM;
1851 root = fuse_get_root_inode(sb, ctx->rootmode);
1852 sb->s_d_op = &fuse_root_dentry_operations;
1853 root_dentry = d_make_root(root);
1854 if (!root_dentry)
1855 goto err_dev_free;
1856 /* Root dentry doesn't have .d_revalidate */
1857 sb->s_d_op = &fuse_dentry_operations;
1858
1859 mutex_lock(&fuse_mutex);
1860 err = -EINVAL;
1861 if (ctx->fudptr && *ctx->fudptr)
1862 goto err_unlock;
1863
1864 err = fuse_ctl_add_conn(fc);
1865 if (err)
1866 goto err_unlock;
1867
1868 list_add_tail(&fc->entry, &fuse_conn_list);
1869 sb->s_root = root_dentry;
1870 if (ctx->fudptr)
1871 *ctx->fudptr = fud;
1872 mutex_unlock(&fuse_mutex);
1873 return 0;
1874
1875 err_unlock:
1876 mutex_unlock(&fuse_mutex);
1877 dput(root_dentry);
1878 err_dev_free:
1879 if (fud)
1880 fuse_dev_free(fud);
1881 err_free_dax:
1882 if (IS_ENABLED(CONFIG_FUSE_DAX))
1883 fuse_dax_conn_free(fc);
1884 err:
1885 return err;
1886 }
1887 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1888
fuse_fill_super(struct super_block * sb,struct fs_context * fsc)1889 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1890 {
1891 struct fuse_fs_context *ctx = fsc->fs_private;
1892 int err;
1893
1894 if (!ctx->file || !ctx->rootmode_present ||
1895 !ctx->user_id_present || !ctx->group_id_present)
1896 return -EINVAL;
1897
1898 /*
1899 * Require mount to happen from the same user namespace which
1900 * opened /dev/fuse to prevent potential attacks.
1901 */
1902 if ((ctx->file->f_op != &fuse_dev_operations) ||
1903 (ctx->file->f_cred->user_ns != sb->s_user_ns))
1904 return -EINVAL;
1905 ctx->fudptr = &ctx->file->private_data;
1906
1907 err = fuse_fill_super_common(sb, ctx);
1908 if (err)
1909 return err;
1910 /* file->private_data shall be visible on all CPUs after this */
1911 smp_mb();
1912 fuse_send_init(get_fuse_mount_super(sb));
1913 return 0;
1914 }
1915
1916 /*
1917 * This is the path where user supplied an already initialized fuse dev. In
1918 * this case never create a new super if the old one is gone.
1919 */
fuse_set_no_super(struct super_block * sb,struct fs_context * fsc)1920 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1921 {
1922 return -ENOTCONN;
1923 }
1924
fuse_test_super(struct super_block * sb,struct fs_context * fsc)1925 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1926 {
1927
1928 return fsc->sget_key == get_fuse_conn_super(sb);
1929 }
1930
fuse_get_tree(struct fs_context * fsc)1931 static int fuse_get_tree(struct fs_context *fsc)
1932 {
1933 struct fuse_fs_context *ctx = fsc->fs_private;
1934 struct fuse_dev *fud;
1935 struct fuse_conn *fc;
1936 struct fuse_mount *fm;
1937 struct super_block *sb;
1938 int err;
1939
1940 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1941 if (!fc)
1942 return -ENOMEM;
1943
1944 fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1945 if (!fm) {
1946 kfree(fc);
1947 return -ENOMEM;
1948 }
1949
1950 fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1951 fc->release = fuse_free_conn;
1952
1953 fsc->s_fs_info = fm;
1954
1955 if (ctx->fd_present)
1956 ctx->file = fget(ctx->fd);
1957
1958 if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1959 err = get_tree_bdev(fsc, fuse_fill_super);
1960 goto out;
1961 }
1962 /*
1963 * While block dev mount can be initialized with a dummy device fd
1964 * (found by device name), normal fuse mounts can't
1965 */
1966 err = -EINVAL;
1967 if (!ctx->file)
1968 goto out;
1969
1970 /*
1971 * Allow creating a fuse mount with an already initialized fuse
1972 * connection
1973 */
1974 fud = READ_ONCE(ctx->file->private_data);
1975 if (ctx->file->f_op == &fuse_dev_operations && fud) {
1976 fsc->sget_key = fud->fc;
1977 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1978 err = PTR_ERR_OR_ZERO(sb);
1979 if (!IS_ERR(sb))
1980 fsc->root = dget(sb->s_root);
1981 } else {
1982 err = get_tree_nodev(fsc, fuse_fill_super);
1983 }
1984 out:
1985 if (fsc->s_fs_info)
1986 fuse_mount_destroy(fm);
1987 if (ctx->file)
1988 fput(ctx->file);
1989 return err;
1990 }
1991
1992 static const struct fs_context_operations fuse_context_ops = {
1993 .free = fuse_free_fsc,
1994 .parse_param = fuse_parse_param,
1995 .reconfigure = fuse_reconfigure,
1996 .get_tree = fuse_get_tree,
1997 };
1998
1999 /*
2000 * Set up the filesystem mount context.
2001 */
fuse_init_fs_context(struct fs_context * fsc)2002 static int fuse_init_fs_context(struct fs_context *fsc)
2003 {
2004 struct fuse_fs_context *ctx;
2005
2006 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
2007 if (!ctx)
2008 return -ENOMEM;
2009
2010 ctx->max_read = ~0;
2011 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
2012 ctx->legacy_opts_show = true;
2013
2014 #ifdef CONFIG_BLOCK
2015 if (fsc->fs_type == &fuseblk_fs_type) {
2016 ctx->is_bdev = true;
2017 ctx->destroy = true;
2018 }
2019 #endif
2020
2021 fsc->fs_private = ctx;
2022 fsc->ops = &fuse_context_ops;
2023 return 0;
2024 }
2025
fuse_mount_remove(struct fuse_mount * fm)2026 bool fuse_mount_remove(struct fuse_mount *fm)
2027 {
2028 struct fuse_conn *fc = fm->fc;
2029 bool last = false;
2030
2031 down_write(&fc->killsb);
2032 list_del_init(&fm->fc_entry);
2033 if (list_empty(&fc->mounts))
2034 last = true;
2035 up_write(&fc->killsb);
2036
2037 return last;
2038 }
2039 EXPORT_SYMBOL_GPL(fuse_mount_remove);
2040
fuse_conn_destroy(struct fuse_mount * fm)2041 void fuse_conn_destroy(struct fuse_mount *fm)
2042 {
2043 struct fuse_conn *fc = fm->fc;
2044
2045 if (fc->destroy)
2046 fuse_send_destroy(fm);
2047
2048 fuse_abort_conn(fc);
2049 fuse_wait_aborted(fc);
2050
2051 if (!list_empty(&fc->entry)) {
2052 mutex_lock(&fuse_mutex);
2053 list_del(&fc->entry);
2054 fuse_ctl_remove_conn(fc);
2055 mutex_unlock(&fuse_mutex);
2056 }
2057 }
2058 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
2059
fuse_sb_destroy(struct super_block * sb)2060 static void fuse_sb_destroy(struct super_block *sb)
2061 {
2062 struct fuse_mount *fm = get_fuse_mount_super(sb);
2063 bool last;
2064
2065 if (sb->s_root) {
2066 last = fuse_mount_remove(fm);
2067 if (last)
2068 fuse_conn_destroy(fm);
2069 }
2070 }
2071
fuse_mount_destroy(struct fuse_mount * fm)2072 void fuse_mount_destroy(struct fuse_mount *fm)
2073 {
2074 fuse_conn_put(fm->fc);
2075 kfree_rcu(fm, rcu);
2076 }
2077 EXPORT_SYMBOL(fuse_mount_destroy);
2078
fuse_kill_sb_anon(struct super_block * sb)2079 static void fuse_kill_sb_anon(struct super_block *sb)
2080 {
2081 fuse_sb_destroy(sb);
2082 kill_anon_super(sb);
2083 fuse_mount_destroy(get_fuse_mount_super(sb));
2084 }
2085
2086 static struct file_system_type fuse_fs_type = {
2087 .owner = THIS_MODULE,
2088 .name = "fuse",
2089 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT | FS_ALLOW_IDMAP,
2090 .init_fs_context = fuse_init_fs_context,
2091 .parameters = fuse_fs_parameters,
2092 .kill_sb = fuse_kill_sb_anon,
2093 };
2094 MODULE_ALIAS_FS("fuse");
2095
2096 #ifdef CONFIG_BLOCK
fuse_kill_sb_blk(struct super_block * sb)2097 static void fuse_kill_sb_blk(struct super_block *sb)
2098 {
2099 fuse_sb_destroy(sb);
2100 kill_block_super(sb);
2101 fuse_mount_destroy(get_fuse_mount_super(sb));
2102 }
2103
2104 static struct file_system_type fuseblk_fs_type = {
2105 .owner = THIS_MODULE,
2106 .name = "fuseblk",
2107 .init_fs_context = fuse_init_fs_context,
2108 .parameters = fuse_fs_parameters,
2109 .kill_sb = fuse_kill_sb_blk,
2110 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE | FS_ALLOW_IDMAP,
2111 };
2112 MODULE_ALIAS_FS("fuseblk");
2113
register_fuseblk(void)2114 static inline int register_fuseblk(void)
2115 {
2116 return register_filesystem(&fuseblk_fs_type);
2117 }
2118
unregister_fuseblk(void)2119 static inline void unregister_fuseblk(void)
2120 {
2121 unregister_filesystem(&fuseblk_fs_type);
2122 }
2123 #else
register_fuseblk(void)2124 static inline int register_fuseblk(void)
2125 {
2126 return 0;
2127 }
2128
unregister_fuseblk(void)2129 static inline void unregister_fuseblk(void)
2130 {
2131 }
2132 #endif
2133
fuse_inode_init_once(void * foo)2134 static void fuse_inode_init_once(void *foo)
2135 {
2136 struct inode *inode = foo;
2137
2138 inode_init_once(inode);
2139 }
2140
fuse_fs_init(void)2141 static int __init fuse_fs_init(void)
2142 {
2143 int err;
2144
2145 fuse_inode_cachep = kmem_cache_create("fuse_inode",
2146 sizeof(struct fuse_inode), 0,
2147 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
2148 fuse_inode_init_once);
2149 err = -ENOMEM;
2150 if (!fuse_inode_cachep)
2151 goto out;
2152
2153 err = register_fuseblk();
2154 if (err)
2155 goto out2;
2156
2157 err = register_filesystem(&fuse_fs_type);
2158 if (err)
2159 goto out3;
2160
2161 err = fuse_sysctl_register();
2162 if (err)
2163 goto out4;
2164
2165 return 0;
2166
2167 out4:
2168 unregister_filesystem(&fuse_fs_type);
2169 out3:
2170 unregister_fuseblk();
2171 out2:
2172 kmem_cache_destroy(fuse_inode_cachep);
2173 out:
2174 return err;
2175 }
2176
fuse_fs_cleanup(void)2177 static void fuse_fs_cleanup(void)
2178 {
2179 fuse_sysctl_unregister();
2180 unregister_filesystem(&fuse_fs_type);
2181 unregister_fuseblk();
2182
2183 /*
2184 * Make sure all delayed rcu free inodes are flushed before we
2185 * destroy cache.
2186 */
2187 rcu_barrier();
2188 kmem_cache_destroy(fuse_inode_cachep);
2189 }
2190
2191 static struct kobject *fuse_kobj;
2192
fuse_sysfs_init(void)2193 static int fuse_sysfs_init(void)
2194 {
2195 int err;
2196
2197 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
2198 if (!fuse_kobj) {
2199 err = -ENOMEM;
2200 goto out_err;
2201 }
2202
2203 err = sysfs_create_mount_point(fuse_kobj, "connections");
2204 if (err)
2205 goto out_fuse_unregister;
2206
2207 return 0;
2208
2209 out_fuse_unregister:
2210 kobject_put(fuse_kobj);
2211 out_err:
2212 return err;
2213 }
2214
fuse_sysfs_cleanup(void)2215 static void fuse_sysfs_cleanup(void)
2216 {
2217 sysfs_remove_mount_point(fuse_kobj, "connections");
2218 kobject_put(fuse_kobj);
2219 }
2220
fuse_init(void)2221 static int __init fuse_init(void)
2222 {
2223 int res;
2224
2225 pr_info("init (API version %i.%i)\n",
2226 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
2227
2228 INIT_LIST_HEAD(&fuse_conn_list);
2229 res = fuse_fs_init();
2230 if (res)
2231 goto err;
2232
2233 res = fuse_dev_init();
2234 if (res)
2235 goto err_fs_cleanup;
2236
2237 res = fuse_sysfs_init();
2238 if (res)
2239 goto err_dev_cleanup;
2240
2241 res = fuse_ctl_init();
2242 if (res)
2243 goto err_sysfs_cleanup;
2244
2245 sanitize_global_limit(&max_user_bgreq);
2246 sanitize_global_limit(&max_user_congthresh);
2247
2248 return 0;
2249
2250 err_sysfs_cleanup:
2251 fuse_sysfs_cleanup();
2252 err_dev_cleanup:
2253 fuse_dev_cleanup();
2254 err_fs_cleanup:
2255 fuse_fs_cleanup();
2256 err:
2257 return res;
2258 }
2259
fuse_exit(void)2260 static void __exit fuse_exit(void)
2261 {
2262 pr_debug("exit\n");
2263
2264 fuse_ctl_cleanup();
2265 fuse_sysfs_cleanup();
2266 fuse_fs_cleanup();
2267 fuse_dev_cleanup();
2268 }
2269
2270 module_init(fuse_init);
2271 module_exit(fuse_exit);
2272