1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
4
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7 */
8
9 #include "fuse_i.h"
10
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/seq_file.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/moduleparam.h>
18 #include <linux/fs_context.h>
19 #include <linux/fs_parser.h>
20 #include <linux/statfs.h>
21 #include <linux/random.h>
22 #include <linux/sched.h>
23 #include <linux/exportfs.h>
24 #include <linux/posix_acl.h>
25 #include <linux/pid_namespace.h>
26
27 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
28 MODULE_DESCRIPTION("Filesystem in Userspace");
29 MODULE_LICENSE("GPL");
30
31 static struct kmem_cache *fuse_inode_cachep;
32 struct list_head fuse_conn_list;
33 DEFINE_MUTEX(fuse_mutex);
34
35 static int set_global_limit(const char *val, const struct kernel_param *kp);
36
37 unsigned max_user_bgreq;
38 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
39 &max_user_bgreq, 0644);
40 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
41 MODULE_PARM_DESC(max_user_bgreq,
42 "Global limit for the maximum number of backgrounded requests an "
43 "unprivileged user can set");
44
45 unsigned max_user_congthresh;
46 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
47 &max_user_congthresh, 0644);
48 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
49 MODULE_PARM_DESC(max_user_congthresh,
50 "Global limit for the maximum congestion threshold an "
51 "unprivileged user can set");
52
53 #define FUSE_SUPER_MAGIC 0x65735546
54
55 #define FUSE_DEFAULT_BLKSIZE 512
56
57 /** Maximum number of outstanding background requests */
58 #define FUSE_DEFAULT_MAX_BACKGROUND 12
59
60 /** Congestion starts at 75% of maximum */
61 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
62
63 #ifdef CONFIG_BLOCK
64 static struct file_system_type fuseblk_fs_type;
65 #endif
66
fuse_alloc_forget(void)67 struct fuse_forget_link *fuse_alloc_forget(void)
68 {
69 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
70 }
71
fuse_alloc_inode(struct super_block * sb)72 static struct inode *fuse_alloc_inode(struct super_block *sb)
73 {
74 struct fuse_inode *fi;
75
76 fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
77 if (!fi)
78 return NULL;
79
80 fi->i_time = 0;
81 fi->inval_mask = 0;
82 fi->nodeid = 0;
83 fi->nlookup = 0;
84 fi->attr_version = 0;
85 fi->orig_ino = 0;
86 fi->state = 0;
87 mutex_init(&fi->mutex);
88 init_rwsem(&fi->i_mmap_sem);
89 spin_lock_init(&fi->lock);
90 fi->forget = fuse_alloc_forget();
91 if (!fi->forget)
92 goto out_free;
93
94 if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
95 goto out_free_forget;
96
97 return &fi->inode;
98
99 out_free_forget:
100 kfree(fi->forget);
101 out_free:
102 kmem_cache_free(fuse_inode_cachep, fi);
103 return NULL;
104 }
105
fuse_free_inode(struct inode * inode)106 static void fuse_free_inode(struct inode *inode)
107 {
108 struct fuse_inode *fi = get_fuse_inode(inode);
109
110 mutex_destroy(&fi->mutex);
111 kfree(fi->forget);
112 #ifdef CONFIG_FUSE_DAX
113 kfree(fi->dax);
114 #endif
115 kmem_cache_free(fuse_inode_cachep, fi);
116 }
117
fuse_evict_inode(struct inode * inode)118 static void fuse_evict_inode(struct inode *inode)
119 {
120 struct fuse_inode *fi = get_fuse_inode(inode);
121
122 truncate_inode_pages_final(&inode->i_data);
123 clear_inode(inode);
124 if (inode->i_sb->s_flags & SB_ACTIVE) {
125 struct fuse_conn *fc = get_fuse_conn(inode);
126
127 if (FUSE_IS_DAX(inode))
128 fuse_dax_inode_cleanup(inode);
129 if (fi->nlookup) {
130 fuse_queue_forget(fc, fi->forget, fi->nodeid,
131 fi->nlookup);
132 fi->forget = NULL;
133 }
134 }
135 if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
136 WARN_ON(!list_empty(&fi->write_files));
137 WARN_ON(!list_empty(&fi->queued_writes));
138 }
139 }
140
fuse_reconfigure(struct fs_context * fc)141 static int fuse_reconfigure(struct fs_context *fc)
142 {
143 struct super_block *sb = fc->root->d_sb;
144
145 sync_filesystem(sb);
146 if (fc->sb_flags & SB_MANDLOCK)
147 return -EINVAL;
148
149 return 0;
150 }
151
152 /*
153 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
154 * so that it will fit.
155 */
fuse_squash_ino(u64 ino64)156 static ino_t fuse_squash_ino(u64 ino64)
157 {
158 ino_t ino = (ino_t) ino64;
159 if (sizeof(ino_t) < sizeof(u64))
160 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
161 return ino;
162 }
163
fuse_change_attributes_common(struct inode * inode,struct fuse_attr * attr,u64 attr_valid)164 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
165 u64 attr_valid)
166 {
167 struct fuse_conn *fc = get_fuse_conn(inode);
168 struct fuse_inode *fi = get_fuse_inode(inode);
169
170 lockdep_assert_held(&fi->lock);
171
172 fi->attr_version = atomic64_inc_return(&fc->attr_version);
173 fi->i_time = attr_valid;
174 WRITE_ONCE(fi->inval_mask, 0);
175
176 inode->i_ino = fuse_squash_ino(attr->ino);
177 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
178 set_nlink(inode, attr->nlink);
179 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
180 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
181 inode->i_blocks = attr->blocks;
182 inode->i_atime.tv_sec = attr->atime;
183 inode->i_atime.tv_nsec = attr->atimensec;
184 /* mtime from server may be stale due to local buffered write */
185 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
186 inode->i_mtime.tv_sec = attr->mtime;
187 inode->i_mtime.tv_nsec = attr->mtimensec;
188 inode->i_ctime.tv_sec = attr->ctime;
189 inode->i_ctime.tv_nsec = attr->ctimensec;
190 }
191
192 if (attr->blksize != 0)
193 inode->i_blkbits = ilog2(attr->blksize);
194 else
195 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
196
197 /*
198 * Don't set the sticky bit in i_mode, unless we want the VFS
199 * to check permissions. This prevents failures due to the
200 * check in may_delete().
201 */
202 fi->orig_i_mode = inode->i_mode;
203 if (!fc->default_permissions)
204 inode->i_mode &= ~S_ISVTX;
205
206 fi->orig_ino = attr->ino;
207
208 /*
209 * We are refreshing inode data and it is possible that another
210 * client set suid/sgid or security.capability xattr. So clear
211 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
212 * was set or if security.capability xattr was set. But we don't
213 * know if security.capability has been set or not. So clear it
214 * anyway. Its less efficient but should be safe.
215 */
216 inode->i_flags &= ~S_NOSEC;
217 }
218
fuse_change_attributes(struct inode * inode,struct fuse_attr * attr,u64 attr_valid,u64 attr_version)219 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
220 u64 attr_valid, u64 attr_version)
221 {
222 struct fuse_conn *fc = get_fuse_conn(inode);
223 struct fuse_inode *fi = get_fuse_inode(inode);
224 bool is_wb = fc->writeback_cache;
225 loff_t oldsize;
226 struct timespec64 old_mtime;
227
228 spin_lock(&fi->lock);
229 if ((attr_version != 0 && fi->attr_version > attr_version) ||
230 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
231 spin_unlock(&fi->lock);
232 return;
233 }
234
235 old_mtime = inode->i_mtime;
236 fuse_change_attributes_common(inode, attr, attr_valid);
237
238 oldsize = inode->i_size;
239 /*
240 * In case of writeback_cache enabled, the cached writes beyond EOF
241 * extend local i_size without keeping userspace server in sync. So,
242 * attr->size coming from server can be stale. We cannot trust it.
243 */
244 if (!is_wb || !S_ISREG(inode->i_mode))
245 i_size_write(inode, attr->size);
246 spin_unlock(&fi->lock);
247
248 if (!is_wb && S_ISREG(inode->i_mode)) {
249 bool inval = false;
250
251 if (oldsize != attr->size) {
252 truncate_pagecache(inode, attr->size);
253 if (!fc->explicit_inval_data)
254 inval = true;
255 } else if (fc->auto_inval_data) {
256 struct timespec64 new_mtime = {
257 .tv_sec = attr->mtime,
258 .tv_nsec = attr->mtimensec,
259 };
260
261 /*
262 * Auto inval mode also checks and invalidates if mtime
263 * has changed.
264 */
265 if (!timespec64_equal(&old_mtime, &new_mtime))
266 inval = true;
267 }
268
269 if (inval)
270 invalidate_inode_pages2(inode->i_mapping);
271 }
272 }
273
fuse_init_inode(struct inode * inode,struct fuse_attr * attr)274 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
275 {
276 inode->i_mode = attr->mode & S_IFMT;
277 inode->i_size = attr->size;
278 inode->i_mtime.tv_sec = attr->mtime;
279 inode->i_mtime.tv_nsec = attr->mtimensec;
280 inode->i_ctime.tv_sec = attr->ctime;
281 inode->i_ctime.tv_nsec = attr->ctimensec;
282 if (S_ISREG(inode->i_mode)) {
283 fuse_init_common(inode);
284 fuse_init_file_inode(inode);
285 } else if (S_ISDIR(inode->i_mode))
286 fuse_init_dir(inode);
287 else if (S_ISLNK(inode->i_mode))
288 fuse_init_symlink(inode);
289 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
290 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
291 fuse_init_common(inode);
292 init_special_inode(inode, inode->i_mode,
293 new_decode_dev(attr->rdev));
294 } else
295 BUG();
296 }
297
fuse_inode_eq(struct inode * inode,void * _nodeidp)298 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
299 {
300 u64 nodeid = *(u64 *) _nodeidp;
301 if (get_node_id(inode) == nodeid)
302 return 1;
303 else
304 return 0;
305 }
306
fuse_inode_set(struct inode * inode,void * _nodeidp)307 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
308 {
309 u64 nodeid = *(u64 *) _nodeidp;
310 get_fuse_inode(inode)->nodeid = nodeid;
311 return 0;
312 }
313
fuse_iget(struct super_block * sb,u64 nodeid,int generation,struct fuse_attr * attr,u64 attr_valid,u64 attr_version)314 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
315 int generation, struct fuse_attr *attr,
316 u64 attr_valid, u64 attr_version)
317 {
318 struct inode *inode;
319 struct fuse_inode *fi;
320 struct fuse_conn *fc = get_fuse_conn_super(sb);
321
322 /*
323 * Auto mount points get their node id from the submount root, which is
324 * not a unique identifier within this filesystem.
325 *
326 * To avoid conflicts, do not place submount points into the inode hash
327 * table.
328 */
329 if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
330 S_ISDIR(attr->mode)) {
331 inode = new_inode(sb);
332 if (!inode)
333 return NULL;
334
335 fuse_init_inode(inode, attr);
336 get_fuse_inode(inode)->nodeid = nodeid;
337 inode->i_flags |= S_AUTOMOUNT;
338 goto done;
339 }
340
341 retry:
342 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
343 if (!inode)
344 return NULL;
345
346 if ((inode->i_state & I_NEW)) {
347 inode->i_flags |= S_NOATIME;
348 if (!fc->writeback_cache || !S_ISREG(attr->mode))
349 inode->i_flags |= S_NOCMTIME;
350 inode->i_generation = generation;
351 fuse_init_inode(inode, attr);
352 unlock_new_inode(inode);
353 } else if (inode_wrong_type(inode, attr->mode)) {
354 /* Inode has changed type, any I/O on the old should fail */
355 fuse_make_bad(inode);
356 iput(inode);
357 goto retry;
358 }
359 done:
360 fi = get_fuse_inode(inode);
361 spin_lock(&fi->lock);
362 fi->nlookup++;
363 spin_unlock(&fi->lock);
364 fuse_change_attributes(inode, attr, attr_valid, attr_version);
365
366 return inode;
367 }
368
fuse_ilookup(struct fuse_conn * fc,u64 nodeid,struct fuse_mount ** fm)369 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
370 struct fuse_mount **fm)
371 {
372 struct fuse_mount *fm_iter;
373 struct inode *inode;
374
375 WARN_ON(!rwsem_is_locked(&fc->killsb));
376 list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
377 if (!fm_iter->sb)
378 continue;
379
380 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
381 if (inode) {
382 if (fm)
383 *fm = fm_iter;
384 return inode;
385 }
386 }
387
388 return NULL;
389 }
390
fuse_reverse_inval_inode(struct fuse_conn * fc,u64 nodeid,loff_t offset,loff_t len)391 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
392 loff_t offset, loff_t len)
393 {
394 struct fuse_inode *fi;
395 struct inode *inode;
396 pgoff_t pg_start;
397 pgoff_t pg_end;
398
399 inode = fuse_ilookup(fc, nodeid, NULL);
400 if (!inode)
401 return -ENOENT;
402
403 fi = get_fuse_inode(inode);
404 spin_lock(&fi->lock);
405 fi->attr_version = atomic64_inc_return(&fc->attr_version);
406 spin_unlock(&fi->lock);
407
408 fuse_invalidate_attr(inode);
409 forget_all_cached_acls(inode);
410 if (offset >= 0) {
411 pg_start = offset >> PAGE_SHIFT;
412 if (len <= 0)
413 pg_end = -1;
414 else
415 pg_end = (offset + len - 1) >> PAGE_SHIFT;
416 invalidate_inode_pages2_range(inode->i_mapping,
417 pg_start, pg_end);
418 }
419 iput(inode);
420 return 0;
421 }
422
fuse_lock_inode(struct inode * inode)423 bool fuse_lock_inode(struct inode *inode)
424 {
425 bool locked = false;
426
427 if (!get_fuse_conn(inode)->parallel_dirops) {
428 mutex_lock(&get_fuse_inode(inode)->mutex);
429 locked = true;
430 }
431
432 return locked;
433 }
434
fuse_unlock_inode(struct inode * inode,bool locked)435 void fuse_unlock_inode(struct inode *inode, bool locked)
436 {
437 if (locked)
438 mutex_unlock(&get_fuse_inode(inode)->mutex);
439 }
440
fuse_umount_begin(struct super_block * sb)441 static void fuse_umount_begin(struct super_block *sb)
442 {
443 struct fuse_conn *fc = get_fuse_conn_super(sb);
444
445 if (!fc->no_force_umount)
446 fuse_abort_conn(fc);
447 }
448
fuse_send_destroy(struct fuse_mount * fm)449 static void fuse_send_destroy(struct fuse_mount *fm)
450 {
451 if (fm->fc->conn_init) {
452 FUSE_ARGS(args);
453
454 args.opcode = FUSE_DESTROY;
455 args.force = true;
456 args.nocreds = true;
457 fuse_simple_request(fm, &args);
458 }
459 }
460
fuse_put_super(struct super_block * sb)461 static void fuse_put_super(struct super_block *sb)
462 {
463 struct fuse_mount *fm = get_fuse_mount_super(sb);
464
465 fuse_conn_put(fm->fc);
466 kfree(fm);
467 }
468
convert_fuse_statfs(struct kstatfs * stbuf,struct fuse_kstatfs * attr)469 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
470 {
471 stbuf->f_type = FUSE_SUPER_MAGIC;
472 stbuf->f_bsize = attr->bsize;
473 stbuf->f_frsize = attr->frsize;
474 stbuf->f_blocks = attr->blocks;
475 stbuf->f_bfree = attr->bfree;
476 stbuf->f_bavail = attr->bavail;
477 stbuf->f_files = attr->files;
478 stbuf->f_ffree = attr->ffree;
479 stbuf->f_namelen = attr->namelen;
480 /* fsid is left zero */
481 }
482
fuse_statfs(struct dentry * dentry,struct kstatfs * buf)483 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
484 {
485 struct super_block *sb = dentry->d_sb;
486 struct fuse_mount *fm = get_fuse_mount_super(sb);
487 FUSE_ARGS(args);
488 struct fuse_statfs_out outarg;
489 int err;
490
491 if (!fuse_allow_current_process(fm->fc)) {
492 buf->f_type = FUSE_SUPER_MAGIC;
493 return 0;
494 }
495
496 memset(&outarg, 0, sizeof(outarg));
497 args.in_numargs = 0;
498 args.opcode = FUSE_STATFS;
499 args.nodeid = get_node_id(d_inode(dentry));
500 args.out_numargs = 1;
501 args.out_args[0].size = sizeof(outarg);
502 args.out_args[0].value = &outarg;
503 err = fuse_simple_request(fm, &args);
504 if (!err)
505 convert_fuse_statfs(buf, &outarg.st);
506 return err;
507 }
508
509 enum {
510 OPT_SOURCE,
511 OPT_SUBTYPE,
512 OPT_FD,
513 OPT_ROOTMODE,
514 OPT_USER_ID,
515 OPT_GROUP_ID,
516 OPT_DEFAULT_PERMISSIONS,
517 OPT_ALLOW_OTHER,
518 OPT_MAX_READ,
519 OPT_BLKSIZE,
520 OPT_ERR
521 };
522
523 static const struct fs_parameter_spec fuse_fs_parameters[] = {
524 fsparam_string ("source", OPT_SOURCE),
525 fsparam_u32 ("fd", OPT_FD),
526 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
527 fsparam_u32 ("user_id", OPT_USER_ID),
528 fsparam_u32 ("group_id", OPT_GROUP_ID),
529 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
530 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
531 fsparam_u32 ("max_read", OPT_MAX_READ),
532 fsparam_u32 ("blksize", OPT_BLKSIZE),
533 fsparam_string ("subtype", OPT_SUBTYPE),
534 {}
535 };
536
fuse_parse_param(struct fs_context * fc,struct fs_parameter * param)537 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
538 {
539 struct fs_parse_result result;
540 struct fuse_fs_context *ctx = fc->fs_private;
541 int opt;
542
543 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
544 /*
545 * Ignore options coming from mount(MS_REMOUNT) for backward
546 * compatibility.
547 */
548 if (fc->oldapi)
549 return 0;
550
551 return invalfc(fc, "No changes allowed in reconfigure");
552 }
553
554 opt = fs_parse(fc, fuse_fs_parameters, param, &result);
555 if (opt < 0)
556 return opt;
557
558 switch (opt) {
559 case OPT_SOURCE:
560 if (fc->source)
561 return invalfc(fc, "Multiple sources specified");
562 fc->source = param->string;
563 param->string = NULL;
564 break;
565
566 case OPT_SUBTYPE:
567 if (ctx->subtype)
568 return invalfc(fc, "Multiple subtypes specified");
569 ctx->subtype = param->string;
570 param->string = NULL;
571 return 0;
572
573 case OPT_FD:
574 ctx->fd = result.uint_32;
575 ctx->fd_present = true;
576 break;
577
578 case OPT_ROOTMODE:
579 if (!fuse_valid_type(result.uint_32))
580 return invalfc(fc, "Invalid rootmode");
581 ctx->rootmode = result.uint_32;
582 ctx->rootmode_present = true;
583 break;
584
585 case OPT_USER_ID:
586 ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
587 if (!uid_valid(ctx->user_id))
588 return invalfc(fc, "Invalid user_id");
589 ctx->user_id_present = true;
590 break;
591
592 case OPT_GROUP_ID:
593 ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
594 if (!gid_valid(ctx->group_id))
595 return invalfc(fc, "Invalid group_id");
596 ctx->group_id_present = true;
597 break;
598
599 case OPT_DEFAULT_PERMISSIONS:
600 ctx->default_permissions = true;
601 break;
602
603 case OPT_ALLOW_OTHER:
604 ctx->allow_other = true;
605 break;
606
607 case OPT_MAX_READ:
608 ctx->max_read = result.uint_32;
609 break;
610
611 case OPT_BLKSIZE:
612 if (!ctx->is_bdev)
613 return invalfc(fc, "blksize only supported for fuseblk");
614 ctx->blksize = result.uint_32;
615 break;
616
617 default:
618 return -EINVAL;
619 }
620
621 return 0;
622 }
623
fuse_free_fc(struct fs_context * fc)624 static void fuse_free_fc(struct fs_context *fc)
625 {
626 struct fuse_fs_context *ctx = fc->fs_private;
627
628 if (ctx) {
629 kfree(ctx->subtype);
630 kfree(ctx);
631 }
632 }
633
fuse_show_options(struct seq_file * m,struct dentry * root)634 static int fuse_show_options(struct seq_file *m, struct dentry *root)
635 {
636 struct super_block *sb = root->d_sb;
637 struct fuse_conn *fc = get_fuse_conn_super(sb);
638
639 if (fc->legacy_opts_show) {
640 seq_printf(m, ",user_id=%u",
641 from_kuid_munged(fc->user_ns, fc->user_id));
642 seq_printf(m, ",group_id=%u",
643 from_kgid_munged(fc->user_ns, fc->group_id));
644 if (fc->default_permissions)
645 seq_puts(m, ",default_permissions");
646 if (fc->allow_other)
647 seq_puts(m, ",allow_other");
648 if (fc->max_read != ~0)
649 seq_printf(m, ",max_read=%u", fc->max_read);
650 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
651 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
652 }
653 #ifdef CONFIG_FUSE_DAX
654 if (fc->dax)
655 seq_puts(m, ",dax");
656 #endif
657
658 return 0;
659 }
660
fuse_iqueue_init(struct fuse_iqueue * fiq,const struct fuse_iqueue_ops * ops,void * priv)661 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
662 const struct fuse_iqueue_ops *ops,
663 void *priv)
664 {
665 memset(fiq, 0, sizeof(struct fuse_iqueue));
666 spin_lock_init(&fiq->lock);
667 init_waitqueue_head(&fiq->waitq);
668 INIT_LIST_HEAD(&fiq->pending);
669 INIT_LIST_HEAD(&fiq->interrupts);
670 fiq->forget_list_tail = &fiq->forget_list_head;
671 fiq->connected = 1;
672 fiq->ops = ops;
673 fiq->priv = priv;
674 }
675
fuse_pqueue_init(struct fuse_pqueue * fpq)676 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
677 {
678 unsigned int i;
679
680 spin_lock_init(&fpq->lock);
681 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
682 INIT_LIST_HEAD(&fpq->processing[i]);
683 INIT_LIST_HEAD(&fpq->io);
684 fpq->connected = 1;
685 }
686
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)687 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
688 struct user_namespace *user_ns,
689 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
690 {
691 memset(fc, 0, sizeof(*fc));
692 spin_lock_init(&fc->lock);
693 spin_lock_init(&fc->bg_lock);
694 init_rwsem(&fc->killsb);
695 refcount_set(&fc->count, 1);
696 atomic_set(&fc->dev_count, 1);
697 init_waitqueue_head(&fc->blocked_waitq);
698 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
699 INIT_LIST_HEAD(&fc->bg_queue);
700 INIT_LIST_HEAD(&fc->entry);
701 INIT_LIST_HEAD(&fc->devices);
702 atomic_set(&fc->num_waiting, 0);
703 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
704 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
705 atomic64_set(&fc->khctr, 0);
706 fc->polled_files = RB_ROOT;
707 fc->blocked = 0;
708 fc->initialized = 0;
709 fc->connected = 1;
710 atomic64_set(&fc->attr_version, 1);
711 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
712 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
713 fc->user_ns = get_user_ns(user_ns);
714 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
715 fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
716
717 INIT_LIST_HEAD(&fc->mounts);
718 list_add(&fm->fc_entry, &fc->mounts);
719 fm->fc = fc;
720 }
721 EXPORT_SYMBOL_GPL(fuse_conn_init);
722
fuse_conn_put(struct fuse_conn * fc)723 void fuse_conn_put(struct fuse_conn *fc)
724 {
725 if (refcount_dec_and_test(&fc->count)) {
726 struct fuse_iqueue *fiq = &fc->iq;
727
728 if (IS_ENABLED(CONFIG_FUSE_DAX))
729 fuse_dax_conn_free(fc);
730 if (fiq->ops->release)
731 fiq->ops->release(fiq);
732 put_pid_ns(fc->pid_ns);
733 put_user_ns(fc->user_ns);
734 fc->release(fc);
735 }
736 }
737 EXPORT_SYMBOL_GPL(fuse_conn_put);
738
fuse_conn_get(struct fuse_conn * fc)739 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
740 {
741 refcount_inc(&fc->count);
742 return fc;
743 }
744 EXPORT_SYMBOL_GPL(fuse_conn_get);
745
fuse_get_root_inode(struct super_block * sb,unsigned mode)746 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
747 {
748 struct fuse_attr attr;
749 memset(&attr, 0, sizeof(attr));
750
751 attr.mode = mode;
752 attr.ino = FUSE_ROOT_ID;
753 attr.nlink = 1;
754 return fuse_iget(sb, 1, 0, &attr, 0, 0);
755 }
756
757 struct fuse_inode_handle {
758 u64 nodeid;
759 u32 generation;
760 };
761
fuse_get_dentry(struct super_block * sb,struct fuse_inode_handle * handle)762 static struct dentry *fuse_get_dentry(struct super_block *sb,
763 struct fuse_inode_handle *handle)
764 {
765 struct fuse_conn *fc = get_fuse_conn_super(sb);
766 struct inode *inode;
767 struct dentry *entry;
768 int err = -ESTALE;
769
770 if (handle->nodeid == 0)
771 goto out_err;
772
773 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
774 if (!inode) {
775 struct fuse_entry_out outarg;
776 const struct qstr name = QSTR_INIT(".", 1);
777
778 if (!fc->export_support)
779 goto out_err;
780
781 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
782 &inode);
783 if (err && err != -ENOENT)
784 goto out_err;
785 if (err || !inode) {
786 err = -ESTALE;
787 goto out_err;
788 }
789 err = -EIO;
790 if (get_node_id(inode) != handle->nodeid)
791 goto out_iput;
792 }
793 err = -ESTALE;
794 if (inode->i_generation != handle->generation)
795 goto out_iput;
796
797 entry = d_obtain_alias(inode);
798 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
799 fuse_invalidate_entry_cache(entry);
800
801 return entry;
802
803 out_iput:
804 iput(inode);
805 out_err:
806 return ERR_PTR(err);
807 }
808
fuse_encode_fh(struct inode * inode,u32 * fh,int * max_len,struct inode * parent)809 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
810 struct inode *parent)
811 {
812 int len = parent ? 6 : 3;
813 u64 nodeid;
814 u32 generation;
815
816 if (*max_len < len) {
817 *max_len = len;
818 return FILEID_INVALID;
819 }
820
821 nodeid = get_fuse_inode(inode)->nodeid;
822 generation = inode->i_generation;
823
824 fh[0] = (u32)(nodeid >> 32);
825 fh[1] = (u32)(nodeid & 0xffffffff);
826 fh[2] = generation;
827
828 if (parent) {
829 nodeid = get_fuse_inode(parent)->nodeid;
830 generation = parent->i_generation;
831
832 fh[3] = (u32)(nodeid >> 32);
833 fh[4] = (u32)(nodeid & 0xffffffff);
834 fh[5] = generation;
835 }
836
837 *max_len = len;
838 return parent ? 0x82 : 0x81;
839 }
840
fuse_fh_to_dentry(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)841 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
842 struct fid *fid, int fh_len, int fh_type)
843 {
844 struct fuse_inode_handle handle;
845
846 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
847 return NULL;
848
849 handle.nodeid = (u64) fid->raw[0] << 32;
850 handle.nodeid |= (u64) fid->raw[1];
851 handle.generation = fid->raw[2];
852 return fuse_get_dentry(sb, &handle);
853 }
854
fuse_fh_to_parent(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)855 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
856 struct fid *fid, int fh_len, int fh_type)
857 {
858 struct fuse_inode_handle parent;
859
860 if (fh_type != 0x82 || fh_len < 6)
861 return NULL;
862
863 parent.nodeid = (u64) fid->raw[3] << 32;
864 parent.nodeid |= (u64) fid->raw[4];
865 parent.generation = fid->raw[5];
866 return fuse_get_dentry(sb, &parent);
867 }
868
fuse_get_parent(struct dentry * child)869 static struct dentry *fuse_get_parent(struct dentry *child)
870 {
871 struct inode *child_inode = d_inode(child);
872 struct fuse_conn *fc = get_fuse_conn(child_inode);
873 struct inode *inode;
874 struct dentry *parent;
875 struct fuse_entry_out outarg;
876 int err;
877
878 if (!fc->export_support)
879 return ERR_PTR(-ESTALE);
880
881 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
882 &dotdot_name, &outarg, &inode);
883 if (err) {
884 if (err == -ENOENT)
885 return ERR_PTR(-ESTALE);
886 return ERR_PTR(err);
887 }
888
889 parent = d_obtain_alias(inode);
890 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
891 fuse_invalidate_entry_cache(parent);
892
893 return parent;
894 }
895
896 static const struct export_operations fuse_export_operations = {
897 .fh_to_dentry = fuse_fh_to_dentry,
898 .fh_to_parent = fuse_fh_to_parent,
899 .encode_fh = fuse_encode_fh,
900 .get_parent = fuse_get_parent,
901 };
902
903 static const struct super_operations fuse_super_operations = {
904 .alloc_inode = fuse_alloc_inode,
905 .free_inode = fuse_free_inode,
906 .evict_inode = fuse_evict_inode,
907 .write_inode = fuse_write_inode,
908 .drop_inode = generic_delete_inode,
909 .put_super = fuse_put_super,
910 .umount_begin = fuse_umount_begin,
911 .statfs = fuse_statfs,
912 .show_options = fuse_show_options,
913 };
914
sanitize_global_limit(unsigned * limit)915 static void sanitize_global_limit(unsigned *limit)
916 {
917 /*
918 * The default maximum number of async requests is calculated to consume
919 * 1/2^13 of the total memory, assuming 392 bytes per request.
920 */
921 if (*limit == 0)
922 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
923
924 if (*limit >= 1 << 16)
925 *limit = (1 << 16) - 1;
926 }
927
set_global_limit(const char * val,const struct kernel_param * kp)928 static int set_global_limit(const char *val, const struct kernel_param *kp)
929 {
930 int rv;
931
932 rv = param_set_uint(val, kp);
933 if (rv)
934 return rv;
935
936 sanitize_global_limit((unsigned *)kp->arg);
937
938 return 0;
939 }
940
process_init_limits(struct fuse_conn * fc,struct fuse_init_out * arg)941 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
942 {
943 int cap_sys_admin = capable(CAP_SYS_ADMIN);
944
945 if (arg->minor < 13)
946 return;
947
948 sanitize_global_limit(&max_user_bgreq);
949 sanitize_global_limit(&max_user_congthresh);
950
951 spin_lock(&fc->bg_lock);
952 if (arg->max_background) {
953 fc->max_background = arg->max_background;
954
955 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
956 fc->max_background = max_user_bgreq;
957 }
958 if (arg->congestion_threshold) {
959 fc->congestion_threshold = arg->congestion_threshold;
960
961 if (!cap_sys_admin &&
962 fc->congestion_threshold > max_user_congthresh)
963 fc->congestion_threshold = max_user_congthresh;
964 }
965 spin_unlock(&fc->bg_lock);
966 }
967
968 struct fuse_init_args {
969 struct fuse_args args;
970 struct fuse_init_in in;
971 struct fuse_init_out out;
972 };
973
process_init_reply(struct fuse_mount * fm,struct fuse_args * args,int error)974 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
975 int error)
976 {
977 struct fuse_conn *fc = fm->fc;
978 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
979 struct fuse_init_out *arg = &ia->out;
980 bool ok = true;
981
982 if (error || arg->major != FUSE_KERNEL_VERSION)
983 ok = false;
984 else {
985 unsigned long ra_pages;
986
987 process_init_limits(fc, arg);
988
989 if (arg->minor >= 6) {
990 ra_pages = arg->max_readahead / PAGE_SIZE;
991 if (arg->flags & FUSE_ASYNC_READ)
992 fc->async_read = 1;
993 if (!(arg->flags & FUSE_POSIX_LOCKS))
994 fc->no_lock = 1;
995 if (arg->minor >= 17) {
996 if (!(arg->flags & FUSE_FLOCK_LOCKS))
997 fc->no_flock = 1;
998 } else {
999 if (!(arg->flags & FUSE_POSIX_LOCKS))
1000 fc->no_flock = 1;
1001 }
1002 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
1003 fc->atomic_o_trunc = 1;
1004 if (arg->minor >= 9) {
1005 /* LOOKUP has dependency on proto version */
1006 if (arg->flags & FUSE_EXPORT_SUPPORT)
1007 fc->export_support = 1;
1008 }
1009 if (arg->flags & FUSE_BIG_WRITES)
1010 fc->big_writes = 1;
1011 if (arg->flags & FUSE_DONT_MASK)
1012 fc->dont_mask = 1;
1013 if (arg->flags & FUSE_AUTO_INVAL_DATA)
1014 fc->auto_inval_data = 1;
1015 else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
1016 fc->explicit_inval_data = 1;
1017 if (arg->flags & FUSE_DO_READDIRPLUS) {
1018 fc->do_readdirplus = 1;
1019 if (arg->flags & FUSE_READDIRPLUS_AUTO)
1020 fc->readdirplus_auto = 1;
1021 }
1022 if (arg->flags & FUSE_ASYNC_DIO)
1023 fc->async_dio = 1;
1024 if (arg->flags & FUSE_WRITEBACK_CACHE)
1025 fc->writeback_cache = 1;
1026 if (arg->flags & FUSE_PARALLEL_DIROPS)
1027 fc->parallel_dirops = 1;
1028 if (arg->flags & FUSE_HANDLE_KILLPRIV)
1029 fc->handle_killpriv = 1;
1030 if (arg->time_gran && arg->time_gran <= 1000000000)
1031 fm->sb->s_time_gran = arg->time_gran;
1032 if ((arg->flags & FUSE_POSIX_ACL)) {
1033 fc->default_permissions = 1;
1034 fc->posix_acl = 1;
1035 fm->sb->s_xattr = fuse_acl_xattr_handlers;
1036 }
1037 if (arg->flags & FUSE_CACHE_SYMLINKS)
1038 fc->cache_symlinks = 1;
1039 if (arg->flags & FUSE_ABORT_ERROR)
1040 fc->abort_err = 1;
1041 if (arg->flags & FUSE_MAX_PAGES) {
1042 fc->max_pages =
1043 min_t(unsigned int, fc->max_pages_limit,
1044 max_t(unsigned int, arg->max_pages, 1));
1045 }
1046 if (IS_ENABLED(CONFIG_FUSE_DAX) &&
1047 arg->flags & FUSE_MAP_ALIGNMENT &&
1048 !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1049 ok = false;
1050 }
1051 if (arg->flags & FUSE_HANDLE_KILLPRIV_V2) {
1052 fc->handle_killpriv_v2 = 1;
1053 fm->sb->s_flags |= SB_NOSEC;
1054 }
1055 if (arg->flags & FUSE_SETXATTR_EXT)
1056 fc->setxattr_ext = 1;
1057 } else {
1058 ra_pages = fc->max_read / PAGE_SIZE;
1059 fc->no_lock = 1;
1060 fc->no_flock = 1;
1061 }
1062
1063 fm->sb->s_bdi->ra_pages =
1064 min(fm->sb->s_bdi->ra_pages, ra_pages);
1065 fc->minor = arg->minor;
1066 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1067 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1068 fc->conn_init = 1;
1069 }
1070 kfree(ia);
1071
1072 if (!ok) {
1073 fc->conn_init = 0;
1074 fc->conn_error = 1;
1075 }
1076
1077 fuse_set_initialized(fc);
1078 wake_up_all(&fc->blocked_waitq);
1079 }
1080
fuse_send_init(struct fuse_mount * fm)1081 void fuse_send_init(struct fuse_mount *fm)
1082 {
1083 struct fuse_init_args *ia;
1084
1085 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1086
1087 ia->in.major = FUSE_KERNEL_VERSION;
1088 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1089 ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1090 ia->in.flags |=
1091 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1092 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1093 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1094 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1095 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1096 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1097 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1098 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1099 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1100 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT;
1101 #ifdef CONFIG_FUSE_DAX
1102 if (fm->fc->dax)
1103 ia->in.flags |= FUSE_MAP_ALIGNMENT;
1104 #endif
1105 if (fm->fc->auto_submounts)
1106 ia->in.flags |= FUSE_SUBMOUNTS;
1107
1108 ia->args.opcode = FUSE_INIT;
1109 ia->args.in_numargs = 1;
1110 ia->args.in_args[0].size = sizeof(ia->in);
1111 ia->args.in_args[0].value = &ia->in;
1112 ia->args.out_numargs = 1;
1113 /* Variable length argument used for backward compatibility
1114 with interface version < 7.5. Rest of init_out is zeroed
1115 by do_get_request(), so a short reply is not a problem */
1116 ia->args.out_argvar = true;
1117 ia->args.out_args[0].size = sizeof(ia->out);
1118 ia->args.out_args[0].value = &ia->out;
1119 ia->args.force = true;
1120 ia->args.nocreds = true;
1121 ia->args.end = process_init_reply;
1122
1123 if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1124 process_init_reply(fm, &ia->args, -ENOTCONN);
1125 }
1126 EXPORT_SYMBOL_GPL(fuse_send_init);
1127
fuse_free_conn(struct fuse_conn * fc)1128 void fuse_free_conn(struct fuse_conn *fc)
1129 {
1130 WARN_ON(!list_empty(&fc->devices));
1131 kfree_rcu(fc, rcu);
1132 }
1133 EXPORT_SYMBOL_GPL(fuse_free_conn);
1134
fuse_bdi_init(struct fuse_conn * fc,struct super_block * sb)1135 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1136 {
1137 int err;
1138 char *suffix = "";
1139
1140 if (sb->s_bdev) {
1141 suffix = "-fuseblk";
1142 /*
1143 * sb->s_bdi points to blkdev's bdi however we want to redirect
1144 * it to our private bdi...
1145 */
1146 bdi_put(sb->s_bdi);
1147 sb->s_bdi = &noop_backing_dev_info;
1148 }
1149 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1150 MINOR(fc->dev), suffix);
1151 if (err)
1152 return err;
1153
1154 /* fuse does it's own writeback accounting */
1155 sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1156 sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1157
1158 /*
1159 * For a single fuse filesystem use max 1% of dirty +
1160 * writeback threshold.
1161 *
1162 * This gives about 1M of write buffer for memory maps on a
1163 * machine with 1G and 10% dirty_ratio, which should be more
1164 * than enough.
1165 *
1166 * Privileged users can raise it by writing to
1167 *
1168 * /sys/class/bdi/<bdi>/max_ratio
1169 */
1170 bdi_set_max_ratio(sb->s_bdi, 1);
1171
1172 return 0;
1173 }
1174
fuse_dev_alloc(void)1175 struct fuse_dev *fuse_dev_alloc(void)
1176 {
1177 struct fuse_dev *fud;
1178 struct list_head *pq;
1179
1180 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1181 if (!fud)
1182 return NULL;
1183
1184 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1185 if (!pq) {
1186 kfree(fud);
1187 return NULL;
1188 }
1189
1190 fud->pq.processing = pq;
1191 fuse_pqueue_init(&fud->pq);
1192
1193 return fud;
1194 }
1195 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1196
fuse_dev_install(struct fuse_dev * fud,struct fuse_conn * fc)1197 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1198 {
1199 fud->fc = fuse_conn_get(fc);
1200 spin_lock(&fc->lock);
1201 list_add_tail(&fud->entry, &fc->devices);
1202 spin_unlock(&fc->lock);
1203 }
1204 EXPORT_SYMBOL_GPL(fuse_dev_install);
1205
fuse_dev_alloc_install(struct fuse_conn * fc)1206 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1207 {
1208 struct fuse_dev *fud;
1209
1210 fud = fuse_dev_alloc();
1211 if (!fud)
1212 return NULL;
1213
1214 fuse_dev_install(fud, fc);
1215 return fud;
1216 }
1217 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1218
fuse_dev_free(struct fuse_dev * fud)1219 void fuse_dev_free(struct fuse_dev *fud)
1220 {
1221 struct fuse_conn *fc = fud->fc;
1222
1223 if (fc) {
1224 spin_lock(&fc->lock);
1225 list_del(&fud->entry);
1226 spin_unlock(&fc->lock);
1227
1228 fuse_conn_put(fc);
1229 }
1230 kfree(fud->pq.processing);
1231 kfree(fud);
1232 }
1233 EXPORT_SYMBOL_GPL(fuse_dev_free);
1234
fuse_fill_attr_from_inode(struct fuse_attr * attr,const struct fuse_inode * fi)1235 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1236 const struct fuse_inode *fi)
1237 {
1238 *attr = (struct fuse_attr){
1239 .ino = fi->inode.i_ino,
1240 .size = fi->inode.i_size,
1241 .blocks = fi->inode.i_blocks,
1242 .atime = fi->inode.i_atime.tv_sec,
1243 .mtime = fi->inode.i_mtime.tv_sec,
1244 .ctime = fi->inode.i_ctime.tv_sec,
1245 .atimensec = fi->inode.i_atime.tv_nsec,
1246 .mtimensec = fi->inode.i_mtime.tv_nsec,
1247 .ctimensec = fi->inode.i_ctime.tv_nsec,
1248 .mode = fi->inode.i_mode,
1249 .nlink = fi->inode.i_nlink,
1250 .uid = fi->inode.i_uid.val,
1251 .gid = fi->inode.i_gid.val,
1252 .rdev = fi->inode.i_rdev,
1253 .blksize = 1u << fi->inode.i_blkbits,
1254 };
1255 }
1256
fuse_sb_defaults(struct super_block * sb)1257 static void fuse_sb_defaults(struct super_block *sb)
1258 {
1259 sb->s_magic = FUSE_SUPER_MAGIC;
1260 sb->s_op = &fuse_super_operations;
1261 sb->s_xattr = fuse_xattr_handlers;
1262 sb->s_maxbytes = MAX_LFS_FILESIZE;
1263 sb->s_time_gran = 1;
1264 sb->s_export_op = &fuse_export_operations;
1265 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1266 if (sb->s_user_ns != &init_user_ns)
1267 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1268 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1269
1270 /*
1271 * If we are not in the initial user namespace posix
1272 * acls must be translated.
1273 */
1274 if (sb->s_user_ns != &init_user_ns)
1275 sb->s_xattr = fuse_no_acl_xattr_handlers;
1276 }
1277
fuse_fill_super_submount(struct super_block * sb,struct fuse_inode * parent_fi)1278 int fuse_fill_super_submount(struct super_block *sb,
1279 struct fuse_inode *parent_fi)
1280 {
1281 struct fuse_mount *fm = get_fuse_mount_super(sb);
1282 struct super_block *parent_sb = parent_fi->inode.i_sb;
1283 struct fuse_attr root_attr;
1284 struct inode *root;
1285
1286 fuse_sb_defaults(sb);
1287 fm->sb = sb;
1288
1289 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1290 sb->s_bdi = bdi_get(parent_sb->s_bdi);
1291
1292 sb->s_xattr = parent_sb->s_xattr;
1293 sb->s_time_gran = parent_sb->s_time_gran;
1294 sb->s_blocksize = parent_sb->s_blocksize;
1295 sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1296 sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1297 if (parent_sb->s_subtype && !sb->s_subtype)
1298 return -ENOMEM;
1299
1300 fuse_fill_attr_from_inode(&root_attr, parent_fi);
1301 root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1302 /*
1303 * This inode is just a duplicate, so it is not looked up and
1304 * its nlookup should not be incremented. fuse_iget() does
1305 * that, though, so undo it here.
1306 */
1307 get_fuse_inode(root)->nlookup--;
1308 sb->s_d_op = &fuse_dentry_operations;
1309 sb->s_root = d_make_root(root);
1310 if (!sb->s_root)
1311 return -ENOMEM;
1312
1313 return 0;
1314 }
1315
fuse_fill_super_common(struct super_block * sb,struct fuse_fs_context * ctx)1316 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1317 {
1318 struct fuse_dev *fud = NULL;
1319 struct fuse_mount *fm = get_fuse_mount_super(sb);
1320 struct fuse_conn *fc = fm->fc;
1321 struct inode *root;
1322 struct dentry *root_dentry;
1323 int err;
1324
1325 err = -EINVAL;
1326 if (sb->s_flags & SB_MANDLOCK)
1327 goto err;
1328
1329 fuse_sb_defaults(sb);
1330
1331 if (ctx->is_bdev) {
1332 #ifdef CONFIG_BLOCK
1333 err = -EINVAL;
1334 if (!sb_set_blocksize(sb, ctx->blksize))
1335 goto err;
1336 #endif
1337 } else {
1338 sb->s_blocksize = PAGE_SIZE;
1339 sb->s_blocksize_bits = PAGE_SHIFT;
1340 }
1341
1342 sb->s_subtype = ctx->subtype;
1343 ctx->subtype = NULL;
1344 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1345 err = fuse_dax_conn_alloc(fc, ctx->dax_dev);
1346 if (err)
1347 goto err;
1348 }
1349
1350 if (ctx->fudptr) {
1351 err = -ENOMEM;
1352 fud = fuse_dev_alloc_install(fc);
1353 if (!fud)
1354 goto err_free_dax;
1355 }
1356
1357 fc->dev = sb->s_dev;
1358 fm->sb = sb;
1359 err = fuse_bdi_init(fc, sb);
1360 if (err)
1361 goto err_dev_free;
1362
1363 /* Handle umasking inside the fuse code */
1364 if (sb->s_flags & SB_POSIXACL)
1365 fc->dont_mask = 1;
1366 sb->s_flags |= SB_POSIXACL;
1367
1368 fc->default_permissions = ctx->default_permissions;
1369 fc->allow_other = ctx->allow_other;
1370 fc->user_id = ctx->user_id;
1371 fc->group_id = ctx->group_id;
1372 fc->legacy_opts_show = ctx->legacy_opts_show;
1373 fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1374 fc->destroy = ctx->destroy;
1375 fc->no_control = ctx->no_control;
1376 fc->no_force_umount = ctx->no_force_umount;
1377
1378 err = -ENOMEM;
1379 root = fuse_get_root_inode(sb, ctx->rootmode);
1380 sb->s_d_op = &fuse_root_dentry_operations;
1381 root_dentry = d_make_root(root);
1382 if (!root_dentry)
1383 goto err_dev_free;
1384 /* Root dentry doesn't have .d_revalidate */
1385 sb->s_d_op = &fuse_dentry_operations;
1386
1387 mutex_lock(&fuse_mutex);
1388 err = -EINVAL;
1389 if (ctx->fudptr && *ctx->fudptr)
1390 goto err_unlock;
1391
1392 err = fuse_ctl_add_conn(fc);
1393 if (err)
1394 goto err_unlock;
1395
1396 list_add_tail(&fc->entry, &fuse_conn_list);
1397 sb->s_root = root_dentry;
1398 if (ctx->fudptr)
1399 *ctx->fudptr = fud;
1400 mutex_unlock(&fuse_mutex);
1401 return 0;
1402
1403 err_unlock:
1404 mutex_unlock(&fuse_mutex);
1405 dput(root_dentry);
1406 err_dev_free:
1407 if (fud)
1408 fuse_dev_free(fud);
1409 err_free_dax:
1410 if (IS_ENABLED(CONFIG_FUSE_DAX))
1411 fuse_dax_conn_free(fc);
1412 err:
1413 return err;
1414 }
1415 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1416
fuse_fill_super(struct super_block * sb,struct fs_context * fsc)1417 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1418 {
1419 struct fuse_fs_context *ctx = fsc->fs_private;
1420 struct file *file;
1421 int err;
1422 struct fuse_conn *fc;
1423 struct fuse_mount *fm;
1424
1425 err = -EINVAL;
1426 file = fget(ctx->fd);
1427 if (!file)
1428 goto err;
1429
1430 /*
1431 * Require mount to happen from the same user namespace which
1432 * opened /dev/fuse to prevent potential attacks.
1433 */
1434 if ((file->f_op != &fuse_dev_operations) ||
1435 (file->f_cred->user_ns != sb->s_user_ns))
1436 goto err_fput;
1437 ctx->fudptr = &file->private_data;
1438
1439 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1440 err = -ENOMEM;
1441 if (!fc)
1442 goto err_fput;
1443
1444 fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1445 if (!fm) {
1446 kfree(fc);
1447 goto err_fput;
1448 }
1449
1450 fuse_conn_init(fc, fm, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1451 fc->release = fuse_free_conn;
1452
1453 sb->s_fs_info = fm;
1454
1455 err = fuse_fill_super_common(sb, ctx);
1456 if (err)
1457 goto err_put_conn;
1458 /*
1459 * atomic_dec_and_test() in fput() provides the necessary
1460 * memory barrier for file->private_data to be visible on all
1461 * CPUs after this
1462 */
1463 fput(file);
1464 fuse_send_init(get_fuse_mount_super(sb));
1465 return 0;
1466
1467 err_put_conn:
1468 fuse_conn_put(fc);
1469 kfree(fm);
1470 sb->s_fs_info = NULL;
1471 err_fput:
1472 fput(file);
1473 err:
1474 return err;
1475 }
1476
fuse_get_tree(struct fs_context * fc)1477 static int fuse_get_tree(struct fs_context *fc)
1478 {
1479 struct fuse_fs_context *ctx = fc->fs_private;
1480
1481 if (!ctx->fd_present || !ctx->rootmode_present ||
1482 !ctx->user_id_present || !ctx->group_id_present)
1483 return -EINVAL;
1484
1485 #ifdef CONFIG_BLOCK
1486 if (ctx->is_bdev)
1487 return get_tree_bdev(fc, fuse_fill_super);
1488 #endif
1489
1490 return get_tree_nodev(fc, fuse_fill_super);
1491 }
1492
1493 static const struct fs_context_operations fuse_context_ops = {
1494 .free = fuse_free_fc,
1495 .parse_param = fuse_parse_param,
1496 .reconfigure = fuse_reconfigure,
1497 .get_tree = fuse_get_tree,
1498 };
1499
1500 /*
1501 * Set up the filesystem mount context.
1502 */
fuse_init_fs_context(struct fs_context * fc)1503 static int fuse_init_fs_context(struct fs_context *fc)
1504 {
1505 struct fuse_fs_context *ctx;
1506
1507 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1508 if (!ctx)
1509 return -ENOMEM;
1510
1511 ctx->max_read = ~0;
1512 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1513 ctx->legacy_opts_show = true;
1514
1515 #ifdef CONFIG_BLOCK
1516 if (fc->fs_type == &fuseblk_fs_type) {
1517 ctx->is_bdev = true;
1518 ctx->destroy = true;
1519 }
1520 #endif
1521
1522 fc->fs_private = ctx;
1523 fc->ops = &fuse_context_ops;
1524 return 0;
1525 }
1526
fuse_mount_remove(struct fuse_mount * fm)1527 bool fuse_mount_remove(struct fuse_mount *fm)
1528 {
1529 struct fuse_conn *fc = fm->fc;
1530 bool last = false;
1531
1532 down_write(&fc->killsb);
1533 list_del_init(&fm->fc_entry);
1534 if (list_empty(&fc->mounts))
1535 last = true;
1536 up_write(&fc->killsb);
1537
1538 return last;
1539 }
1540 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1541
fuse_conn_destroy(struct fuse_mount * fm)1542 void fuse_conn_destroy(struct fuse_mount *fm)
1543 {
1544 struct fuse_conn *fc = fm->fc;
1545
1546 if (fc->destroy)
1547 fuse_send_destroy(fm);
1548
1549 fuse_abort_conn(fc);
1550 fuse_wait_aborted(fc);
1551
1552 if (!list_empty(&fc->entry)) {
1553 mutex_lock(&fuse_mutex);
1554 list_del(&fc->entry);
1555 fuse_ctl_remove_conn(fc);
1556 mutex_unlock(&fuse_mutex);
1557 }
1558 }
1559 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1560
fuse_sb_destroy(struct super_block * sb)1561 static void fuse_sb_destroy(struct super_block *sb)
1562 {
1563 struct fuse_mount *fm = get_fuse_mount_super(sb);
1564 bool last;
1565
1566 if (fm) {
1567 last = fuse_mount_remove(fm);
1568 if (last)
1569 fuse_conn_destroy(fm);
1570 }
1571 }
1572
fuse_kill_sb_anon(struct super_block * sb)1573 static void fuse_kill_sb_anon(struct super_block *sb)
1574 {
1575 fuse_sb_destroy(sb);
1576 kill_anon_super(sb);
1577 }
1578
1579 static struct file_system_type fuse_fs_type = {
1580 .owner = THIS_MODULE,
1581 .name = "fuse",
1582 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1583 .init_fs_context = fuse_init_fs_context,
1584 .parameters = fuse_fs_parameters,
1585 .kill_sb = fuse_kill_sb_anon,
1586 };
1587 MODULE_ALIAS_FS("fuse");
1588
1589 #ifdef CONFIG_BLOCK
fuse_kill_sb_blk(struct super_block * sb)1590 static void fuse_kill_sb_blk(struct super_block *sb)
1591 {
1592 fuse_sb_destroy(sb);
1593 kill_block_super(sb);
1594 }
1595
1596 static struct file_system_type fuseblk_fs_type = {
1597 .owner = THIS_MODULE,
1598 .name = "fuseblk",
1599 .init_fs_context = fuse_init_fs_context,
1600 .parameters = fuse_fs_parameters,
1601 .kill_sb = fuse_kill_sb_blk,
1602 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1603 };
1604 MODULE_ALIAS_FS("fuseblk");
1605
register_fuseblk(void)1606 static inline int register_fuseblk(void)
1607 {
1608 return register_filesystem(&fuseblk_fs_type);
1609 }
1610
unregister_fuseblk(void)1611 static inline void unregister_fuseblk(void)
1612 {
1613 unregister_filesystem(&fuseblk_fs_type);
1614 }
1615 #else
register_fuseblk(void)1616 static inline int register_fuseblk(void)
1617 {
1618 return 0;
1619 }
1620
unregister_fuseblk(void)1621 static inline void unregister_fuseblk(void)
1622 {
1623 }
1624 #endif
1625
fuse_inode_init_once(void * foo)1626 static void fuse_inode_init_once(void *foo)
1627 {
1628 struct inode *inode = foo;
1629
1630 inode_init_once(inode);
1631 }
1632
fuse_fs_init(void)1633 static int __init fuse_fs_init(void)
1634 {
1635 int err;
1636
1637 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1638 sizeof(struct fuse_inode), 0,
1639 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1640 fuse_inode_init_once);
1641 err = -ENOMEM;
1642 if (!fuse_inode_cachep)
1643 goto out;
1644
1645 err = register_fuseblk();
1646 if (err)
1647 goto out2;
1648
1649 err = register_filesystem(&fuse_fs_type);
1650 if (err)
1651 goto out3;
1652
1653 return 0;
1654
1655 out3:
1656 unregister_fuseblk();
1657 out2:
1658 kmem_cache_destroy(fuse_inode_cachep);
1659 out:
1660 return err;
1661 }
1662
fuse_fs_cleanup(void)1663 static void fuse_fs_cleanup(void)
1664 {
1665 unregister_filesystem(&fuse_fs_type);
1666 unregister_fuseblk();
1667
1668 /*
1669 * Make sure all delayed rcu free inodes are flushed before we
1670 * destroy cache.
1671 */
1672 rcu_barrier();
1673 kmem_cache_destroy(fuse_inode_cachep);
1674 }
1675
1676 static struct kobject *fuse_kobj;
1677
fuse_sysfs_init(void)1678 static int fuse_sysfs_init(void)
1679 {
1680 int err;
1681
1682 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1683 if (!fuse_kobj) {
1684 err = -ENOMEM;
1685 goto out_err;
1686 }
1687
1688 err = sysfs_create_mount_point(fuse_kobj, "connections");
1689 if (err)
1690 goto out_fuse_unregister;
1691
1692 return 0;
1693
1694 out_fuse_unregister:
1695 kobject_put(fuse_kobj);
1696 out_err:
1697 return err;
1698 }
1699
fuse_sysfs_cleanup(void)1700 static void fuse_sysfs_cleanup(void)
1701 {
1702 sysfs_remove_mount_point(fuse_kobj, "connections");
1703 kobject_put(fuse_kobj);
1704 }
1705
fuse_init(void)1706 static int __init fuse_init(void)
1707 {
1708 int res;
1709
1710 pr_info("init (API version %i.%i)\n",
1711 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1712
1713 INIT_LIST_HEAD(&fuse_conn_list);
1714 res = fuse_fs_init();
1715 if (res)
1716 goto err;
1717
1718 res = fuse_dev_init();
1719 if (res)
1720 goto err_fs_cleanup;
1721
1722 res = fuse_sysfs_init();
1723 if (res)
1724 goto err_dev_cleanup;
1725
1726 res = fuse_ctl_init();
1727 if (res)
1728 goto err_sysfs_cleanup;
1729
1730 sanitize_global_limit(&max_user_bgreq);
1731 sanitize_global_limit(&max_user_congthresh);
1732
1733 return 0;
1734
1735 err_sysfs_cleanup:
1736 fuse_sysfs_cleanup();
1737 err_dev_cleanup:
1738 fuse_dev_cleanup();
1739 err_fs_cleanup:
1740 fuse_fs_cleanup();
1741 err:
1742 return res;
1743 }
1744
fuse_exit(void)1745 static void __exit fuse_exit(void)
1746 {
1747 pr_debug("exit\n");
1748
1749 fuse_ctl_cleanup();
1750 fuse_sysfs_cleanup();
1751 fuse_fs_cleanup();
1752 fuse_dev_cleanup();
1753 }
1754
1755 module_init(fuse_init);
1756 module_exit(fuse_exit);
1757