1
2
3
4
5
6
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/parser.h>
19#include <linux/statfs.h>
20#include <linux/random.h>
21#include <linux/sched.h>
22#include <linux/exportfs.h>
23#include <linux/posix_acl.h>
24#include <linux/pid_namespace.h>
25
26MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
27MODULE_DESCRIPTION("Filesystem in Userspace");
28MODULE_LICENSE("GPL");
29
30static struct kmem_cache *fuse_inode_cachep;
31struct list_head fuse_conn_list;
32DEFINE_MUTEX(fuse_mutex);
33
34static int set_global_limit(const char *val, const struct kernel_param *kp);
35
36unsigned max_user_bgreq;
37module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
38 &max_user_bgreq, 0644);
39__MODULE_PARM_TYPE(max_user_bgreq, "uint");
40MODULE_PARM_DESC(max_user_bgreq,
41 "Global limit for the maximum number of backgrounded requests an "
42 "unprivileged user can set");
43
44unsigned max_user_congthresh;
45module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
46 &max_user_congthresh, 0644);
47__MODULE_PARM_TYPE(max_user_congthresh, "uint");
48MODULE_PARM_DESC(max_user_congthresh,
49 "Global limit for the maximum congestion threshold an "
50 "unprivileged user can set");
51
52#define FUSE_SUPER_MAGIC 0x65735546
53
54#define FUSE_DEFAULT_BLKSIZE 512
55
56
57#define FUSE_DEFAULT_MAX_BACKGROUND 12
58
59
60#define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
61
62struct fuse_mount_data {
63 int fd;
64 unsigned rootmode;
65 kuid_t user_id;
66 kgid_t group_id;
67 unsigned fd_present:1;
68 unsigned rootmode_present:1;
69 unsigned user_id_present:1;
70 unsigned group_id_present:1;
71 unsigned default_permissions:1;
72 unsigned allow_other:1;
73 unsigned max_read;
74 unsigned blksize;
75};
76
77struct fuse_forget_link *fuse_alloc_forget(void)
78{
79 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
80}
81
82static struct inode *fuse_alloc_inode(struct super_block *sb)
83{
84 struct inode *inode;
85 struct fuse_inode *fi;
86
87 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
88 if (!inode)
89 return NULL;
90
91 fi = get_fuse_inode(inode);
92 fi->i_time = 0;
93 fi->nodeid = 0;
94 fi->nlookup = 0;
95 fi->attr_version = 0;
96 fi->writectr = 0;
97 fi->orig_ino = 0;
98 fi->state = 0;
99 INIT_LIST_HEAD(&fi->write_files);
100 INIT_LIST_HEAD(&fi->queued_writes);
101 INIT_LIST_HEAD(&fi->writepages);
102 init_waitqueue_head(&fi->page_waitq);
103 mutex_init(&fi->mutex);
104 fi->forget = fuse_alloc_forget();
105 if (!fi->forget) {
106 kmem_cache_free(fuse_inode_cachep, inode);
107 return NULL;
108 }
109
110 return inode;
111}
112
113static void fuse_i_callback(struct rcu_head *head)
114{
115 struct inode *inode = container_of(head, struct inode, i_rcu);
116 kmem_cache_free(fuse_inode_cachep, inode);
117}
118
119static void fuse_destroy_inode(struct inode *inode)
120{
121 struct fuse_inode *fi = get_fuse_inode(inode);
122 BUG_ON(!list_empty(&fi->write_files));
123 BUG_ON(!list_empty(&fi->queued_writes));
124 mutex_destroy(&fi->mutex);
125 kfree(fi->forget);
126 call_rcu(&inode->i_rcu, fuse_i_callback);
127}
128
129static void fuse_evict_inode(struct inode *inode)
130{
131 truncate_inode_pages_final(&inode->i_data);
132 clear_inode(inode);
133 if (inode->i_sb->s_flags & SB_ACTIVE) {
134 struct fuse_conn *fc = get_fuse_conn(inode);
135 struct fuse_inode *fi = get_fuse_inode(inode);
136 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
137 fi->forget = NULL;
138 }
139}
140
141static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
142{
143 sync_filesystem(sb);
144 if (*flags & SB_MANDLOCK)
145 return -EINVAL;
146
147 return 0;
148}
149
150
151
152
153
154static ino_t fuse_squash_ino(u64 ino64)
155{
156 ino_t ino = (ino_t) ino64;
157 if (sizeof(ino_t) < sizeof(u64))
158 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
159 return ino;
160}
161
162void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
163 u64 attr_valid)
164{
165 struct fuse_conn *fc = get_fuse_conn(inode);
166 struct fuse_inode *fi = get_fuse_inode(inode);
167
168 fi->attr_version = ++fc->attr_version;
169 fi->i_time = attr_valid;
170
171 inode->i_ino = fuse_squash_ino(attr->ino);
172 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
173 set_nlink(inode, attr->nlink);
174 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
175 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
176 inode->i_blocks = attr->blocks;
177 inode->i_atime.tv_sec = attr->atime;
178 inode->i_atime.tv_nsec = attr->atimensec;
179
180 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
181 inode->i_mtime.tv_sec = attr->mtime;
182 inode->i_mtime.tv_nsec = attr->mtimensec;
183 inode->i_ctime.tv_sec = attr->ctime;
184 inode->i_ctime.tv_nsec = attr->ctimensec;
185 }
186
187 if (attr->blksize != 0)
188 inode->i_blkbits = ilog2(attr->blksize);
189 else
190 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
191
192
193
194
195
196
197 fi->orig_i_mode = inode->i_mode;
198 if (!fc->default_permissions)
199 inode->i_mode &= ~S_ISVTX;
200
201 fi->orig_ino = attr->ino;
202}
203
204void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
205 u64 attr_valid, u64 attr_version)
206{
207 struct fuse_conn *fc = get_fuse_conn(inode);
208 struct fuse_inode *fi = get_fuse_inode(inode);
209 bool is_wb = fc->writeback_cache;
210 loff_t oldsize;
211 struct timespec64 old_mtime;
212
213 spin_lock(&fc->lock);
214 if ((attr_version != 0 && fi->attr_version > attr_version) ||
215 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
216 spin_unlock(&fc->lock);
217 return;
218 }
219
220 old_mtime = inode->i_mtime;
221 fuse_change_attributes_common(inode, attr, attr_valid);
222
223 oldsize = inode->i_size;
224
225
226
227
228
229 if (!is_wb || !S_ISREG(inode->i_mode))
230 i_size_write(inode, attr->size);
231 spin_unlock(&fc->lock);
232
233 if (!is_wb && S_ISREG(inode->i_mode)) {
234 bool inval = false;
235
236 if (oldsize != attr->size) {
237 truncate_pagecache(inode, attr->size);
238 inval = true;
239 } else if (fc->auto_inval_data) {
240 struct timespec64 new_mtime = {
241 .tv_sec = attr->mtime,
242 .tv_nsec = attr->mtimensec,
243 };
244
245
246
247
248
249 if (!timespec64_equal(&old_mtime, &new_mtime))
250 inval = true;
251 }
252
253 if (inval)
254 invalidate_inode_pages2(inode->i_mapping);
255 }
256}
257
258static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
259{
260 inode->i_mode = attr->mode & S_IFMT;
261 inode->i_size = attr->size;
262 inode->i_mtime.tv_sec = attr->mtime;
263 inode->i_mtime.tv_nsec = attr->mtimensec;
264 inode->i_ctime.tv_sec = attr->ctime;
265 inode->i_ctime.tv_nsec = attr->ctimensec;
266 if (S_ISREG(inode->i_mode)) {
267 fuse_init_common(inode);
268 fuse_init_file_inode(inode);
269 } else if (S_ISDIR(inode->i_mode))
270 fuse_init_dir(inode);
271 else if (S_ISLNK(inode->i_mode))
272 fuse_init_symlink(inode);
273 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
274 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
275 fuse_init_common(inode);
276 init_special_inode(inode, inode->i_mode,
277 new_decode_dev(attr->rdev));
278 } else
279 BUG();
280}
281
282int fuse_inode_eq(struct inode *inode, void *_nodeidp)
283{
284 u64 nodeid = *(u64 *) _nodeidp;
285 if (get_node_id(inode) == nodeid)
286 return 1;
287 else
288 return 0;
289}
290
291static int fuse_inode_set(struct inode *inode, void *_nodeidp)
292{
293 u64 nodeid = *(u64 *) _nodeidp;
294 get_fuse_inode(inode)->nodeid = nodeid;
295 return 0;
296}
297
298struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
299 int generation, struct fuse_attr *attr,
300 u64 attr_valid, u64 attr_version)
301{
302 struct inode *inode;
303 struct fuse_inode *fi;
304 struct fuse_conn *fc = get_fuse_conn_super(sb);
305
306 retry:
307 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
308 if (!inode)
309 return NULL;
310
311 if ((inode->i_state & I_NEW)) {
312 inode->i_flags |= S_NOATIME;
313 if (!fc->writeback_cache || !S_ISREG(attr->mode))
314 inode->i_flags |= S_NOCMTIME;
315 inode->i_generation = generation;
316 fuse_init_inode(inode, attr);
317 unlock_new_inode(inode);
318 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
319
320 make_bad_inode(inode);
321 iput(inode);
322 goto retry;
323 }
324
325 fi = get_fuse_inode(inode);
326 spin_lock(&fc->lock);
327 fi->nlookup++;
328 spin_unlock(&fc->lock);
329 fuse_change_attributes(inode, attr, attr_valid, attr_version);
330
331 return inode;
332}
333
334int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
335 loff_t offset, loff_t len)
336{
337 struct inode *inode;
338 pgoff_t pg_start;
339 pgoff_t pg_end;
340
341 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
342 if (!inode)
343 return -ENOENT;
344
345 fuse_invalidate_attr(inode);
346 forget_all_cached_acls(inode);
347 if (offset >= 0) {
348 pg_start = offset >> PAGE_SHIFT;
349 if (len <= 0)
350 pg_end = -1;
351 else
352 pg_end = (offset + len - 1) >> PAGE_SHIFT;
353 invalidate_inode_pages2_range(inode->i_mapping,
354 pg_start, pg_end);
355 }
356 iput(inode);
357 return 0;
358}
359
360bool fuse_lock_inode(struct inode *inode)
361{
362 bool locked = false;
363
364 if (!get_fuse_conn(inode)->parallel_dirops) {
365 mutex_lock(&get_fuse_inode(inode)->mutex);
366 locked = true;
367 }
368
369 return locked;
370}
371
372void fuse_unlock_inode(struct inode *inode, bool locked)
373{
374 if (locked)
375 mutex_unlock(&get_fuse_inode(inode)->mutex);
376}
377
378static void fuse_umount_begin(struct super_block *sb)
379{
380 fuse_abort_conn(get_fuse_conn_super(sb), false);
381}
382
383static void fuse_send_destroy(struct fuse_conn *fc)
384{
385 struct fuse_req *req = fc->destroy_req;
386 if (req && fc->conn_init) {
387 fc->destroy_req = NULL;
388 req->in.h.opcode = FUSE_DESTROY;
389 __set_bit(FR_FORCE, &req->flags);
390 __clear_bit(FR_BACKGROUND, &req->flags);
391 fuse_request_send(fc, req);
392 fuse_put_request(fc, req);
393 }
394}
395
396static void fuse_put_super(struct super_block *sb)
397{
398 struct fuse_conn *fc = get_fuse_conn_super(sb);
399
400 mutex_lock(&fuse_mutex);
401 list_del(&fc->entry);
402 fuse_ctl_remove_conn(fc);
403 mutex_unlock(&fuse_mutex);
404
405 fuse_conn_put(fc);
406}
407
408static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
409{
410 stbuf->f_type = FUSE_SUPER_MAGIC;
411 stbuf->f_bsize = attr->bsize;
412 stbuf->f_frsize = attr->frsize;
413 stbuf->f_blocks = attr->blocks;
414 stbuf->f_bfree = attr->bfree;
415 stbuf->f_bavail = attr->bavail;
416 stbuf->f_files = attr->files;
417 stbuf->f_ffree = attr->ffree;
418 stbuf->f_namelen = attr->namelen;
419
420}
421
422static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
423{
424 struct super_block *sb = dentry->d_sb;
425 struct fuse_conn *fc = get_fuse_conn_super(sb);
426 FUSE_ARGS(args);
427 struct fuse_statfs_out outarg;
428 int err;
429
430 if (!fuse_allow_current_process(fc)) {
431 buf->f_type = FUSE_SUPER_MAGIC;
432 return 0;
433 }
434
435 memset(&outarg, 0, sizeof(outarg));
436 args.in.numargs = 0;
437 args.in.h.opcode = FUSE_STATFS;
438 args.in.h.nodeid = get_node_id(d_inode(dentry));
439 args.out.numargs = 1;
440 args.out.args[0].size = sizeof(outarg);
441 args.out.args[0].value = &outarg;
442 err = fuse_simple_request(fc, &args);
443 if (!err)
444 convert_fuse_statfs(buf, &outarg.st);
445 return err;
446}
447
448enum {
449 OPT_FD,
450 OPT_ROOTMODE,
451 OPT_USER_ID,
452 OPT_GROUP_ID,
453 OPT_DEFAULT_PERMISSIONS,
454 OPT_ALLOW_OTHER,
455 OPT_MAX_READ,
456 OPT_BLKSIZE,
457 OPT_ERR
458};
459
460static const match_table_t tokens = {
461 {OPT_FD, "fd=%u"},
462 {OPT_ROOTMODE, "rootmode=%o"},
463 {OPT_USER_ID, "user_id=%u"},
464 {OPT_GROUP_ID, "group_id=%u"},
465 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
466 {OPT_ALLOW_OTHER, "allow_other"},
467 {OPT_MAX_READ, "max_read=%u"},
468 {OPT_BLKSIZE, "blksize=%u"},
469 {OPT_ERR, NULL}
470};
471
472static int fuse_match_uint(substring_t *s, unsigned int *res)
473{
474 int err = -ENOMEM;
475 char *buf = match_strdup(s);
476 if (buf) {
477 err = kstrtouint(buf, 10, res);
478 kfree(buf);
479 }
480 return err;
481}
482
483static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev,
484 struct user_namespace *user_ns)
485{
486 char *p;
487 memset(d, 0, sizeof(struct fuse_mount_data));
488 d->max_read = ~0;
489 d->blksize = FUSE_DEFAULT_BLKSIZE;
490
491 while ((p = strsep(&opt, ",")) != NULL) {
492 int token;
493 int value;
494 unsigned uv;
495 substring_t args[MAX_OPT_ARGS];
496 if (!*p)
497 continue;
498
499 token = match_token(p, tokens, args);
500 switch (token) {
501 case OPT_FD:
502 if (match_int(&args[0], &value))
503 return 0;
504 d->fd = value;
505 d->fd_present = 1;
506 break;
507
508 case OPT_ROOTMODE:
509 if (match_octal(&args[0], &value))
510 return 0;
511 if (!fuse_valid_type(value))
512 return 0;
513 d->rootmode = value;
514 d->rootmode_present = 1;
515 break;
516
517 case OPT_USER_ID:
518 if (fuse_match_uint(&args[0], &uv))
519 return 0;
520 d->user_id = make_kuid(user_ns, uv);
521 if (!uid_valid(d->user_id))
522 return 0;
523 d->user_id_present = 1;
524 break;
525
526 case OPT_GROUP_ID:
527 if (fuse_match_uint(&args[0], &uv))
528 return 0;
529 d->group_id = make_kgid(user_ns, uv);
530 if (!gid_valid(d->group_id))
531 return 0;
532 d->group_id_present = 1;
533 break;
534
535 case OPT_DEFAULT_PERMISSIONS:
536 d->default_permissions = 1;
537 break;
538
539 case OPT_ALLOW_OTHER:
540 d->allow_other = 1;
541 break;
542
543 case OPT_MAX_READ:
544 if (match_int(&args[0], &value))
545 return 0;
546 d->max_read = value;
547 break;
548
549 case OPT_BLKSIZE:
550 if (!is_bdev || match_int(&args[0], &value))
551 return 0;
552 d->blksize = value;
553 break;
554
555 default:
556 return 0;
557 }
558 }
559
560 if (!d->fd_present || !d->rootmode_present ||
561 !d->user_id_present || !d->group_id_present)
562 return 0;
563
564 return 1;
565}
566
567static int fuse_show_options(struct seq_file *m, struct dentry *root)
568{
569 struct super_block *sb = root->d_sb;
570 struct fuse_conn *fc = get_fuse_conn_super(sb);
571
572 seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
573 seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
574 if (fc->default_permissions)
575 seq_puts(m, ",default_permissions");
576 if (fc->allow_other)
577 seq_puts(m, ",allow_other");
578 if (fc->max_read != ~0)
579 seq_printf(m, ",max_read=%u", fc->max_read);
580 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
581 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
582 return 0;
583}
584
585static void fuse_iqueue_init(struct fuse_iqueue *fiq)
586{
587 memset(fiq, 0, sizeof(struct fuse_iqueue));
588 init_waitqueue_head(&fiq->waitq);
589 INIT_LIST_HEAD(&fiq->pending);
590 INIT_LIST_HEAD(&fiq->interrupts);
591 fiq->forget_list_tail = &fiq->forget_list_head;
592 fiq->connected = 1;
593}
594
595static void fuse_pqueue_init(struct fuse_pqueue *fpq)
596{
597 memset(fpq, 0, sizeof(struct fuse_pqueue));
598 spin_lock_init(&fpq->lock);
599 INIT_LIST_HEAD(&fpq->processing);
600 INIT_LIST_HEAD(&fpq->io);
601 fpq->connected = 1;
602}
603
604void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns)
605{
606 memset(fc, 0, sizeof(*fc));
607 spin_lock_init(&fc->lock);
608 init_rwsem(&fc->killsb);
609 refcount_set(&fc->count, 1);
610 atomic_set(&fc->dev_count, 1);
611 init_waitqueue_head(&fc->blocked_waitq);
612 init_waitqueue_head(&fc->reserved_req_waitq);
613 fuse_iqueue_init(&fc->iq);
614 INIT_LIST_HEAD(&fc->bg_queue);
615 INIT_LIST_HEAD(&fc->entry);
616 INIT_LIST_HEAD(&fc->devices);
617 atomic_set(&fc->num_waiting, 0);
618 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
619 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
620 fc->khctr = 0;
621 fc->polled_files = RB_ROOT;
622 fc->blocked = 0;
623 fc->initialized = 0;
624 fc->connected = 1;
625 fc->attr_version = 1;
626 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
627 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
628 fc->user_ns = get_user_ns(user_ns);
629}
630EXPORT_SYMBOL_GPL(fuse_conn_init);
631
632void fuse_conn_put(struct fuse_conn *fc)
633{
634 if (refcount_dec_and_test(&fc->count)) {
635 if (fc->destroy_req)
636 fuse_request_free(fc->destroy_req);
637 put_pid_ns(fc->pid_ns);
638 put_user_ns(fc->user_ns);
639 fc->release(fc);
640 }
641}
642EXPORT_SYMBOL_GPL(fuse_conn_put);
643
644struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
645{
646 refcount_inc(&fc->count);
647 return fc;
648}
649EXPORT_SYMBOL_GPL(fuse_conn_get);
650
651static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
652{
653 struct fuse_attr attr;
654 memset(&attr, 0, sizeof(attr));
655
656 attr.mode = mode;
657 attr.ino = FUSE_ROOT_ID;
658 attr.nlink = 1;
659 return fuse_iget(sb, 1, 0, &attr, 0, 0);
660}
661
662struct fuse_inode_handle {
663 u64 nodeid;
664 u32 generation;
665};
666
667static struct dentry *fuse_get_dentry(struct super_block *sb,
668 struct fuse_inode_handle *handle)
669{
670 struct fuse_conn *fc = get_fuse_conn_super(sb);
671 struct inode *inode;
672 struct dentry *entry;
673 int err = -ESTALE;
674
675 if (handle->nodeid == 0)
676 goto out_err;
677
678 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
679 if (!inode) {
680 struct fuse_entry_out outarg;
681 const struct qstr name = QSTR_INIT(".", 1);
682
683 if (!fc->export_support)
684 goto out_err;
685
686 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
687 &inode);
688 if (err && err != -ENOENT)
689 goto out_err;
690 if (err || !inode) {
691 err = -ESTALE;
692 goto out_err;
693 }
694 err = -EIO;
695 if (get_node_id(inode) != handle->nodeid)
696 goto out_iput;
697 }
698 err = -ESTALE;
699 if (inode->i_generation != handle->generation)
700 goto out_iput;
701
702 entry = d_obtain_alias(inode);
703 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
704 fuse_invalidate_entry_cache(entry);
705
706 return entry;
707
708 out_iput:
709 iput(inode);
710 out_err:
711 return ERR_PTR(err);
712}
713
714static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
715 struct inode *parent)
716{
717 int len = parent ? 6 : 3;
718 u64 nodeid;
719 u32 generation;
720
721 if (*max_len < len) {
722 *max_len = len;
723 return FILEID_INVALID;
724 }
725
726 nodeid = get_fuse_inode(inode)->nodeid;
727 generation = inode->i_generation;
728
729 fh[0] = (u32)(nodeid >> 32);
730 fh[1] = (u32)(nodeid & 0xffffffff);
731 fh[2] = generation;
732
733 if (parent) {
734 nodeid = get_fuse_inode(parent)->nodeid;
735 generation = parent->i_generation;
736
737 fh[3] = (u32)(nodeid >> 32);
738 fh[4] = (u32)(nodeid & 0xffffffff);
739 fh[5] = generation;
740 }
741
742 *max_len = len;
743 return parent ? 0x82 : 0x81;
744}
745
746static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
747 struct fid *fid, int fh_len, int fh_type)
748{
749 struct fuse_inode_handle handle;
750
751 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
752 return NULL;
753
754 handle.nodeid = (u64) fid->raw[0] << 32;
755 handle.nodeid |= (u64) fid->raw[1];
756 handle.generation = fid->raw[2];
757 return fuse_get_dentry(sb, &handle);
758}
759
760static struct dentry *fuse_fh_to_parent(struct super_block *sb,
761 struct fid *fid, int fh_len, int fh_type)
762{
763 struct fuse_inode_handle parent;
764
765 if (fh_type != 0x82 || fh_len < 6)
766 return NULL;
767
768 parent.nodeid = (u64) fid->raw[3] << 32;
769 parent.nodeid |= (u64) fid->raw[4];
770 parent.generation = fid->raw[5];
771 return fuse_get_dentry(sb, &parent);
772}
773
774static struct dentry *fuse_get_parent(struct dentry *child)
775{
776 struct inode *child_inode = d_inode(child);
777 struct fuse_conn *fc = get_fuse_conn(child_inode);
778 struct inode *inode;
779 struct dentry *parent;
780 struct fuse_entry_out outarg;
781 const struct qstr name = QSTR_INIT("..", 2);
782 int err;
783
784 if (!fc->export_support)
785 return ERR_PTR(-ESTALE);
786
787 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
788 &name, &outarg, &inode);
789 if (err) {
790 if (err == -ENOENT)
791 return ERR_PTR(-ESTALE);
792 return ERR_PTR(err);
793 }
794
795 parent = d_obtain_alias(inode);
796 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
797 fuse_invalidate_entry_cache(parent);
798
799 return parent;
800}
801
802static const struct export_operations fuse_export_operations = {
803 .fh_to_dentry = fuse_fh_to_dentry,
804 .fh_to_parent = fuse_fh_to_parent,
805 .encode_fh = fuse_encode_fh,
806 .get_parent = fuse_get_parent,
807};
808
809static const struct super_operations fuse_super_operations = {
810 .alloc_inode = fuse_alloc_inode,
811 .destroy_inode = fuse_destroy_inode,
812 .evict_inode = fuse_evict_inode,
813 .write_inode = fuse_write_inode,
814 .drop_inode = generic_delete_inode,
815 .remount_fs = fuse_remount_fs,
816 .put_super = fuse_put_super,
817 .umount_begin = fuse_umount_begin,
818 .statfs = fuse_statfs,
819 .show_options = fuse_show_options,
820};
821
822static void sanitize_global_limit(unsigned *limit)
823{
824 if (*limit == 0)
825 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
826 sizeof(struct fuse_req);
827
828 if (*limit >= 1 << 16)
829 *limit = (1 << 16) - 1;
830}
831
832static int set_global_limit(const char *val, const struct kernel_param *kp)
833{
834 int rv;
835
836 rv = param_set_uint(val, kp);
837 if (rv)
838 return rv;
839
840 sanitize_global_limit((unsigned *)kp->arg);
841
842 return 0;
843}
844
845static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
846{
847 int cap_sys_admin = capable(CAP_SYS_ADMIN);
848
849 if (arg->minor < 13)
850 return;
851
852 sanitize_global_limit(&max_user_bgreq);
853 sanitize_global_limit(&max_user_congthresh);
854
855 if (arg->max_background) {
856 fc->max_background = arg->max_background;
857
858 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
859 fc->max_background = max_user_bgreq;
860 }
861 if (arg->congestion_threshold) {
862 fc->congestion_threshold = arg->congestion_threshold;
863
864 if (!cap_sys_admin &&
865 fc->congestion_threshold > max_user_congthresh)
866 fc->congestion_threshold = max_user_congthresh;
867 }
868}
869
870static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
871{
872 struct fuse_init_out *arg = &req->misc.init_out;
873
874 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
875 fc->conn_error = 1;
876 else {
877 unsigned long ra_pages;
878
879 process_init_limits(fc, arg);
880
881 if (arg->minor >= 6) {
882 ra_pages = arg->max_readahead / PAGE_SIZE;
883 if (arg->flags & FUSE_ASYNC_READ)
884 fc->async_read = 1;
885 if (!(arg->flags & FUSE_POSIX_LOCKS))
886 fc->no_lock = 1;
887 if (arg->minor >= 17) {
888 if (!(arg->flags & FUSE_FLOCK_LOCKS))
889 fc->no_flock = 1;
890 } else {
891 if (!(arg->flags & FUSE_POSIX_LOCKS))
892 fc->no_flock = 1;
893 }
894 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
895 fc->atomic_o_trunc = 1;
896 if (arg->minor >= 9) {
897
898 if (arg->flags & FUSE_EXPORT_SUPPORT)
899 fc->export_support = 1;
900 }
901 if (arg->flags & FUSE_BIG_WRITES)
902 fc->big_writes = 1;
903 if (arg->flags & FUSE_DONT_MASK)
904 fc->dont_mask = 1;
905 if (arg->flags & FUSE_AUTO_INVAL_DATA)
906 fc->auto_inval_data = 1;
907 if (arg->flags & FUSE_DO_READDIRPLUS) {
908 fc->do_readdirplus = 1;
909 if (arg->flags & FUSE_READDIRPLUS_AUTO)
910 fc->readdirplus_auto = 1;
911 }
912 if (arg->flags & FUSE_ASYNC_DIO)
913 fc->async_dio = 1;
914 if (arg->flags & FUSE_WRITEBACK_CACHE)
915 fc->writeback_cache = 1;
916 if (arg->flags & FUSE_PARALLEL_DIROPS)
917 fc->parallel_dirops = 1;
918 if (arg->flags & FUSE_HANDLE_KILLPRIV)
919 fc->handle_killpriv = 1;
920 if (arg->time_gran && arg->time_gran <= 1000000000)
921 fc->sb->s_time_gran = arg->time_gran;
922 if ((arg->flags & FUSE_POSIX_ACL)) {
923 fc->default_permissions = 1;
924 fc->posix_acl = 1;
925 fc->sb->s_xattr = fuse_acl_xattr_handlers;
926 }
927 if (arg->flags & FUSE_ABORT_ERROR)
928 fc->abort_err = 1;
929 } else {
930 ra_pages = fc->max_read / PAGE_SIZE;
931 fc->no_lock = 1;
932 fc->no_flock = 1;
933 }
934
935 fc->sb->s_bdi->ra_pages =
936 min(fc->sb->s_bdi->ra_pages, ra_pages);
937 fc->minor = arg->minor;
938 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
939 fc->max_write = max_t(unsigned, 4096, fc->max_write);
940 fc->conn_init = 1;
941 }
942 fuse_set_initialized(fc);
943 wake_up_all(&fc->blocked_waitq);
944}
945
946static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
947{
948 struct fuse_init_in *arg = &req->misc.init_in;
949
950 arg->major = FUSE_KERNEL_VERSION;
951 arg->minor = FUSE_KERNEL_MINOR_VERSION;
952 arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
953 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
954 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
955 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
956 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
957 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
958 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
959 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
960 FUSE_ABORT_ERROR;
961 req->in.h.opcode = FUSE_INIT;
962 req->in.numargs = 1;
963 req->in.args[0].size = sizeof(*arg);
964 req->in.args[0].value = arg;
965 req->out.numargs = 1;
966
967
968
969 req->out.argvar = 1;
970 req->out.args[0].size = sizeof(struct fuse_init_out);
971 req->out.args[0].value = &req->misc.init_out;
972 req->end = process_init_reply;
973 fuse_request_send_background(fc, req);
974}
975
976static void fuse_free_conn(struct fuse_conn *fc)
977{
978 WARN_ON(!list_empty(&fc->devices));
979 kfree_rcu(fc, rcu);
980}
981
982static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
983{
984 int err;
985 char *suffix = "";
986
987 if (sb->s_bdev) {
988 suffix = "-fuseblk";
989
990
991
992
993 bdi_put(sb->s_bdi);
994 sb->s_bdi = &noop_backing_dev_info;
995 }
996 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
997 MINOR(fc->dev), suffix);
998 if (err)
999 return err;
1000
1001 sb->s_bdi->ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
1002
1003 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017 bdi_set_max_ratio(sb->s_bdi, 1);
1018
1019 return 0;
1020}
1021
1022struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1023{
1024 struct fuse_dev *fud;
1025
1026 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1027 if (fud) {
1028 fud->fc = fuse_conn_get(fc);
1029 fuse_pqueue_init(&fud->pq);
1030
1031 spin_lock(&fc->lock);
1032 list_add_tail(&fud->entry, &fc->devices);
1033 spin_unlock(&fc->lock);
1034 }
1035
1036 return fud;
1037}
1038EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1039
1040void fuse_dev_free(struct fuse_dev *fud)
1041{
1042 struct fuse_conn *fc = fud->fc;
1043
1044 if (fc) {
1045 spin_lock(&fc->lock);
1046 list_del(&fud->entry);
1047 spin_unlock(&fc->lock);
1048
1049 fuse_conn_put(fc);
1050 }
1051 kfree(fud);
1052}
1053EXPORT_SYMBOL_GPL(fuse_dev_free);
1054
1055static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1056{
1057 struct fuse_dev *fud;
1058 struct fuse_conn *fc;
1059 struct inode *root;
1060 struct fuse_mount_data d;
1061 struct file *file;
1062 struct dentry *root_dentry;
1063 struct fuse_req *init_req;
1064 int err;
1065 int is_bdev = sb->s_bdev != NULL;
1066
1067 err = -EINVAL;
1068 if (sb->s_flags & SB_MANDLOCK)
1069 goto err;
1070
1071 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1072
1073 if (!parse_fuse_opt(data, &d, is_bdev, sb->s_user_ns))
1074 goto err;
1075
1076 if (is_bdev) {
1077#ifdef CONFIG_BLOCK
1078 err = -EINVAL;
1079 if (!sb_set_blocksize(sb, d.blksize))
1080 goto err;
1081#endif
1082 } else {
1083 sb->s_blocksize = PAGE_SIZE;
1084 sb->s_blocksize_bits = PAGE_SHIFT;
1085 }
1086 sb->s_magic = FUSE_SUPER_MAGIC;
1087 sb->s_op = &fuse_super_operations;
1088 sb->s_xattr = fuse_xattr_handlers;
1089 sb->s_maxbytes = MAX_LFS_FILESIZE;
1090 sb->s_time_gran = 1;
1091 sb->s_export_op = &fuse_export_operations;
1092 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1093 if (sb->s_user_ns != &init_user_ns)
1094 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1095
1096 file = fget(d.fd);
1097 err = -EINVAL;
1098 if (!file)
1099 goto err;
1100
1101
1102
1103
1104
1105 if (file->f_op != &fuse_dev_operations ||
1106 file->f_cred->user_ns != sb->s_user_ns)
1107 goto err_fput;
1108
1109
1110
1111
1112
1113 if (sb->s_user_ns != &init_user_ns)
1114 sb->s_xattr = fuse_no_acl_xattr_handlers;
1115
1116 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1117 err = -ENOMEM;
1118 if (!fc)
1119 goto err_fput;
1120
1121 fuse_conn_init(fc, sb->s_user_ns);
1122 fc->release = fuse_free_conn;
1123
1124 fud = fuse_dev_alloc(fc);
1125 if (!fud)
1126 goto err_put_conn;
1127
1128 fc->dev = sb->s_dev;
1129 fc->sb = sb;
1130 err = fuse_bdi_init(fc, sb);
1131 if (err)
1132 goto err_dev_free;
1133
1134
1135 if (sb->s_flags & SB_POSIXACL)
1136 fc->dont_mask = 1;
1137 sb->s_flags |= SB_POSIXACL;
1138
1139 fc->default_permissions = d.default_permissions;
1140 fc->allow_other = d.allow_other;
1141 fc->user_id = d.user_id;
1142 fc->group_id = d.group_id;
1143 fc->max_read = max_t(unsigned, 4096, d.max_read);
1144
1145
1146 sb->s_fs_info = fc;
1147
1148 err = -ENOMEM;
1149 root = fuse_get_root_inode(sb, d.rootmode);
1150 sb->s_d_op = &fuse_root_dentry_operations;
1151 root_dentry = d_make_root(root);
1152 if (!root_dentry)
1153 goto err_dev_free;
1154
1155 sb->s_d_op = &fuse_dentry_operations;
1156
1157 init_req = fuse_request_alloc(0);
1158 if (!init_req)
1159 goto err_put_root;
1160 __set_bit(FR_BACKGROUND, &init_req->flags);
1161
1162 if (is_bdev) {
1163 fc->destroy_req = fuse_request_alloc(0);
1164 if (!fc->destroy_req)
1165 goto err_free_init_req;
1166 }
1167
1168 mutex_lock(&fuse_mutex);
1169 err = -EINVAL;
1170 if (file->private_data)
1171 goto err_unlock;
1172
1173 err = fuse_ctl_add_conn(fc);
1174 if (err)
1175 goto err_unlock;
1176
1177 list_add_tail(&fc->entry, &fuse_conn_list);
1178 sb->s_root = root_dentry;
1179 file->private_data = fud;
1180 mutex_unlock(&fuse_mutex);
1181
1182
1183
1184
1185
1186 fput(file);
1187
1188 fuse_send_init(fc, init_req);
1189
1190 return 0;
1191
1192 err_unlock:
1193 mutex_unlock(&fuse_mutex);
1194 err_free_init_req:
1195 fuse_request_free(init_req);
1196 err_put_root:
1197 dput(root_dentry);
1198 err_dev_free:
1199 fuse_dev_free(fud);
1200 err_put_conn:
1201 fuse_conn_put(fc);
1202 sb->s_fs_info = NULL;
1203 err_fput:
1204 fput(file);
1205 err:
1206 return err;
1207}
1208
1209static struct dentry *fuse_mount(struct file_system_type *fs_type,
1210 int flags, const char *dev_name,
1211 void *raw_data)
1212{
1213 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1214}
1215
1216static void fuse_sb_destroy(struct super_block *sb)
1217{
1218 struct fuse_conn *fc = get_fuse_conn_super(sb);
1219
1220 if (fc) {
1221 fuse_send_destroy(fc);
1222
1223 fuse_abort_conn(fc, false);
1224 fuse_wait_aborted(fc);
1225
1226 down_write(&fc->killsb);
1227 fc->sb = NULL;
1228 up_write(&fc->killsb);
1229 }
1230}
1231
1232static void fuse_kill_sb_anon(struct super_block *sb)
1233{
1234 fuse_sb_destroy(sb);
1235 kill_anon_super(sb);
1236}
1237
1238static struct file_system_type fuse_fs_type = {
1239 .owner = THIS_MODULE,
1240 .name = "fuse",
1241 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1242 .mount = fuse_mount,
1243 .kill_sb = fuse_kill_sb_anon,
1244};
1245MODULE_ALIAS_FS("fuse");
1246
1247#ifdef CONFIG_BLOCK
1248static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1249 int flags, const char *dev_name,
1250 void *raw_data)
1251{
1252 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1253}
1254
1255static void fuse_kill_sb_blk(struct super_block *sb)
1256{
1257 fuse_sb_destroy(sb);
1258 kill_block_super(sb);
1259}
1260
1261static struct file_system_type fuseblk_fs_type = {
1262 .owner = THIS_MODULE,
1263 .name = "fuseblk",
1264 .mount = fuse_mount_blk,
1265 .kill_sb = fuse_kill_sb_blk,
1266 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1267};
1268MODULE_ALIAS_FS("fuseblk");
1269
1270static inline int register_fuseblk(void)
1271{
1272 return register_filesystem(&fuseblk_fs_type);
1273}
1274
1275static inline void unregister_fuseblk(void)
1276{
1277 unregister_filesystem(&fuseblk_fs_type);
1278}
1279#else
1280static inline int register_fuseblk(void)
1281{
1282 return 0;
1283}
1284
1285static inline void unregister_fuseblk(void)
1286{
1287}
1288#endif
1289
1290static void fuse_inode_init_once(void *foo)
1291{
1292 struct inode *inode = foo;
1293
1294 inode_init_once(inode);
1295}
1296
1297static int __init fuse_fs_init(void)
1298{
1299 int err;
1300
1301 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1302 sizeof(struct fuse_inode), 0,
1303 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1304 fuse_inode_init_once);
1305 err = -ENOMEM;
1306 if (!fuse_inode_cachep)
1307 goto out;
1308
1309 err = register_fuseblk();
1310 if (err)
1311 goto out2;
1312
1313 err = register_filesystem(&fuse_fs_type);
1314 if (err)
1315 goto out3;
1316
1317 return 0;
1318
1319 out3:
1320 unregister_fuseblk();
1321 out2:
1322 kmem_cache_destroy(fuse_inode_cachep);
1323 out:
1324 return err;
1325}
1326
1327static void fuse_fs_cleanup(void)
1328{
1329 unregister_filesystem(&fuse_fs_type);
1330 unregister_fuseblk();
1331
1332
1333
1334
1335
1336 rcu_barrier();
1337 kmem_cache_destroy(fuse_inode_cachep);
1338}
1339
1340static struct kobject *fuse_kobj;
1341
1342static int fuse_sysfs_init(void)
1343{
1344 int err;
1345
1346 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1347 if (!fuse_kobj) {
1348 err = -ENOMEM;
1349 goto out_err;
1350 }
1351
1352 err = sysfs_create_mount_point(fuse_kobj, "connections");
1353 if (err)
1354 goto out_fuse_unregister;
1355
1356 return 0;
1357
1358 out_fuse_unregister:
1359 kobject_put(fuse_kobj);
1360 out_err:
1361 return err;
1362}
1363
1364static void fuse_sysfs_cleanup(void)
1365{
1366 sysfs_remove_mount_point(fuse_kobj, "connections");
1367 kobject_put(fuse_kobj);
1368}
1369
1370static int __init fuse_init(void)
1371{
1372 int res;
1373
1374 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1375 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1376
1377 INIT_LIST_HEAD(&fuse_conn_list);
1378 res = fuse_fs_init();
1379 if (res)
1380 goto err;
1381
1382 res = fuse_dev_init();
1383 if (res)
1384 goto err_fs_cleanup;
1385
1386 res = fuse_sysfs_init();
1387 if (res)
1388 goto err_dev_cleanup;
1389
1390 res = fuse_ctl_init();
1391 if (res)
1392 goto err_sysfs_cleanup;
1393
1394 sanitize_global_limit(&max_user_bgreq);
1395 sanitize_global_limit(&max_user_congthresh);
1396
1397 return 0;
1398
1399 err_sysfs_cleanup:
1400 fuse_sysfs_cleanup();
1401 err_dev_cleanup:
1402 fuse_dev_cleanup();
1403 err_fs_cleanup:
1404 fuse_fs_cleanup();
1405 err:
1406 return res;
1407}
1408
1409static void __exit fuse_exit(void)
1410{
1411 printk(KERN_DEBUG "fuse exit\n");
1412
1413 fuse_ctl_cleanup();
1414 fuse_sysfs_cleanup();
1415 fuse_fs_cleanup();
1416 fuse_dev_cleanup();
1417}
1418
1419module_init(fuse_init);
1420module_exit(fuse_exit);
1421
