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->inval_mask = 0;
94 fi->nodeid = 0;
95 fi->nlookup = 0;
96 fi->attr_version = 0;
97 fi->orig_ino = 0;
98 fi->state = 0;
99 mutex_init(&fi->mutex);
100 fi->forget = fuse_alloc_forget();
101 if (!fi->forget) {
102 kmem_cache_free(fuse_inode_cachep, inode);
103 return NULL;
104 }
105
106 return inode;
107}
108
109static void fuse_i_callback(struct rcu_head *head)
110{
111 struct inode *inode = container_of(head, struct inode, i_rcu);
112 kmem_cache_free(fuse_inode_cachep, inode);
113}
114
115static void fuse_destroy_inode(struct inode *inode)
116{
117 struct fuse_inode *fi = get_fuse_inode(inode);
118 if (S_ISREG(inode->i_mode) && !is_bad_inode(inode)) {
119 WARN_ON(!list_empty(&fi->write_files));
120 WARN_ON(!list_empty(&fi->queued_writes));
121 }
122 mutex_destroy(&fi->mutex);
123 kfree(fi->forget);
124 call_rcu(&inode->i_rcu, fuse_i_callback);
125}
126
127static void fuse_evict_inode(struct inode *inode)
128{
129 truncate_inode_pages_final(&inode->i_data);
130 clear_inode(inode);
131 if (inode->i_sb->s_flags & SB_ACTIVE) {
132 struct fuse_conn *fc = get_fuse_conn(inode);
133 struct fuse_inode *fi = get_fuse_inode(inode);
134 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
135 fi->forget = NULL;
136 }
137}
138
139static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
140{
141 sync_filesystem(sb);
142 if (*flags & SB_MANDLOCK)
143 return -EINVAL;
144
145 return 0;
146}
147
148
149
150
151
152static ino_t fuse_squash_ino(u64 ino64)
153{
154 ino_t ino = (ino_t) ino64;
155 if (sizeof(ino_t) < sizeof(u64))
156 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
157 return ino;
158}
159
160void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
161 u64 attr_valid)
162{
163 struct fuse_conn *fc = get_fuse_conn(inode);
164 struct fuse_inode *fi = get_fuse_inode(inode);
165
166 fi->attr_version = ++fc->attr_version;
167 fi->i_time = attr_valid;
168 WRITE_ONCE(fi->inval_mask, 0);
169
170 inode->i_ino = fuse_squash_ino(attr->ino);
171 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
172 set_nlink(inode, attr->nlink);
173 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
174 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
175 inode->i_blocks = attr->blocks;
176 inode->i_atime.tv_sec = attr->atime;
177 inode->i_atime.tv_nsec = attr->atimensec;
178
179 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
180 inode->i_mtime.tv_sec = attr->mtime;
181 inode->i_mtime.tv_nsec = attr->mtimensec;
182 inode->i_ctime.tv_sec = attr->ctime;
183 inode->i_ctime.tv_nsec = attr->ctimensec;
184 }
185
186 if (attr->blksize != 0)
187 inode->i_blkbits = ilog2(attr->blksize);
188 else
189 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
190
191
192
193
194
195
196 fi->orig_i_mode = inode->i_mode;
197 if (!fc->default_permissions)
198 inode->i_mode &= ~S_ISVTX;
199
200 fi->orig_ino = attr->ino;
201}
202
203void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
204 u64 attr_valid, u64 attr_version)
205{
206 struct fuse_conn *fc = get_fuse_conn(inode);
207 struct fuse_inode *fi = get_fuse_inode(inode);
208 bool is_wb = fc->writeback_cache;
209 loff_t oldsize;
210 struct timespec64 old_mtime;
211
212 spin_lock(&fc->lock);
213 if ((attr_version != 0 && fi->attr_version > attr_version) ||
214 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
215 spin_unlock(&fc->lock);
216 return;
217 }
218
219 old_mtime = inode->i_mtime;
220 fuse_change_attributes_common(inode, attr, attr_valid);
221
222 oldsize = inode->i_size;
223
224
225
226
227
228 if (!is_wb || !S_ISREG(inode->i_mode))
229 i_size_write(inode, attr->size);
230 spin_unlock(&fc->lock);
231
232 if (!is_wb && S_ISREG(inode->i_mode)) {
233 bool inval = false;
234
235 if (oldsize != attr->size) {
236 truncate_pagecache(inode, attr->size);
237 inval = true;
238 } else if (fc->auto_inval_data) {
239 struct timespec64 new_mtime = {
240 .tv_sec = attr->mtime,
241 .tv_nsec = attr->mtimensec,
242 };
243
244
245
246
247
248 if (!timespec64_equal(&old_mtime, &new_mtime))
249 inval = true;
250 }
251
252 if (inval)
253 invalidate_inode_pages2(inode->i_mapping);
254 }
255}
256
257static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
258{
259 inode->i_mode = attr->mode & S_IFMT;
260 inode->i_size = attr->size;
261 inode->i_mtime.tv_sec = attr->mtime;
262 inode->i_mtime.tv_nsec = attr->mtimensec;
263 inode->i_ctime.tv_sec = attr->ctime;
264 inode->i_ctime.tv_nsec = attr->ctimensec;
265 if (S_ISREG(inode->i_mode)) {
266 fuse_init_common(inode);
267 fuse_init_file_inode(inode);
268 } else if (S_ISDIR(inode->i_mode))
269 fuse_init_dir(inode);
270 else if (S_ISLNK(inode->i_mode))
271 fuse_init_symlink(inode);
272 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
273 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
274 fuse_init_common(inode);
275 init_special_inode(inode, inode->i_mode,
276 new_decode_dev(attr->rdev));
277 } else
278 BUG();
279}
280
281int fuse_inode_eq(struct inode *inode, void *_nodeidp)
282{
283 u64 nodeid = *(u64 *) _nodeidp;
284 if (get_node_id(inode) == nodeid)
285 return 1;
286 else
287 return 0;
288}
289
290static int fuse_inode_set(struct inode *inode, void *_nodeidp)
291{
292 u64 nodeid = *(u64 *) _nodeidp;
293 get_fuse_inode(inode)->nodeid = nodeid;
294 return 0;
295}
296
297struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
298 int generation, struct fuse_attr *attr,
299 u64 attr_valid, u64 attr_version)
300{
301 struct inode *inode;
302 struct fuse_inode *fi;
303 struct fuse_conn *fc = get_fuse_conn_super(sb);
304
305 retry:
306 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
307 if (!inode)
308 return NULL;
309
310 if ((inode->i_state & I_NEW)) {
311 inode->i_flags |= S_NOATIME;
312 if (!fc->writeback_cache || !S_ISREG(attr->mode))
313 inode->i_flags |= S_NOCMTIME;
314 inode->i_generation = generation;
315 fuse_init_inode(inode, attr);
316 unlock_new_inode(inode);
317 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
318
319 make_bad_inode(inode);
320 iput(inode);
321 goto retry;
322 }
323
324 fi = get_fuse_inode(inode);
325 spin_lock(&fc->lock);
326 fi->nlookup++;
327 spin_unlock(&fc->lock);
328 fuse_change_attributes(inode, attr, attr_valid, attr_version);
329
330 return inode;
331}
332
333int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
334 loff_t offset, loff_t len)
335{
336 struct inode *inode;
337 pgoff_t pg_start;
338 pgoff_t pg_end;
339
340 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
341 if (!inode)
342 return -ENOENT;
343
344 fuse_invalidate_attr(inode);
345 forget_all_cached_acls(inode);
346 if (offset >= 0) {
347 pg_start = offset >> PAGE_SHIFT;
348 if (len <= 0)
349 pg_end = -1;
350 else
351 pg_end = (offset + len - 1) >> PAGE_SHIFT;
352 invalidate_inode_pages2_range(inode->i_mapping,
353 pg_start, pg_end);
354 }
355 iput(inode);
356 return 0;
357}
358
359bool fuse_lock_inode(struct inode *inode)
360{
361 bool locked = false;
362
363 if (!get_fuse_conn(inode)->parallel_dirops) {
364 mutex_lock(&get_fuse_inode(inode)->mutex);
365 locked = true;
366 }
367
368 return locked;
369}
370
371void fuse_unlock_inode(struct inode *inode, bool locked)
372{
373 if (locked)
374 mutex_unlock(&get_fuse_inode(inode)->mutex);
375}
376
377static void fuse_umount_begin(struct super_block *sb)
378{
379 fuse_abort_conn(get_fuse_conn_super(sb), false);
380}
381
382static void fuse_send_destroy(struct fuse_conn *fc)
383{
384 struct fuse_req *req = fc->destroy_req;
385 if (req && fc->conn_init) {
386 fc->destroy_req = NULL;
387 req->in.h.opcode = FUSE_DESTROY;
388 __set_bit(FR_FORCE, &req->flags);
389 __clear_bit(FR_BACKGROUND, &req->flags);
390 fuse_request_send(fc, req);
391 fuse_put_request(fc, req);
392 }
393}
394
395static void fuse_put_super(struct super_block *sb)
396{
397 struct fuse_conn *fc = get_fuse_conn_super(sb);
398
399 mutex_lock(&fuse_mutex);
400 list_del(&fc->entry);
401 fuse_ctl_remove_conn(fc);
402 mutex_unlock(&fuse_mutex);
403
404 fuse_conn_put(fc);
405}
406
407static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
408{
409 stbuf->f_type = FUSE_SUPER_MAGIC;
410 stbuf->f_bsize = attr->bsize;
411 stbuf->f_frsize = attr->frsize;
412 stbuf->f_blocks = attr->blocks;
413 stbuf->f_bfree = attr->bfree;
414 stbuf->f_bavail = attr->bavail;
415 stbuf->f_files = attr->files;
416 stbuf->f_ffree = attr->ffree;
417 stbuf->f_namelen = attr->namelen;
418
419}
420
421static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
422{
423 struct super_block *sb = dentry->d_sb;
424 struct fuse_conn *fc = get_fuse_conn_super(sb);
425 FUSE_ARGS(args);
426 struct fuse_statfs_out outarg;
427 int err;
428
429 if (!fuse_allow_current_process(fc)) {
430 buf->f_type = FUSE_SUPER_MAGIC;
431 return 0;
432 }
433
434 memset(&outarg, 0, sizeof(outarg));
435 args.in.numargs = 0;
436 args.in.h.opcode = FUSE_STATFS;
437 args.in.h.nodeid = get_node_id(d_inode(dentry));
438 args.out.numargs = 1;
439 args.out.args[0].size = sizeof(outarg);
440 args.out.args[0].value = &outarg;
441 err = fuse_simple_request(fc, &args);
442 if (!err)
443 convert_fuse_statfs(buf, &outarg.st);
444 return err;
445}
446
447enum {
448 OPT_FD,
449 OPT_ROOTMODE,
450 OPT_USER_ID,
451 OPT_GROUP_ID,
452 OPT_DEFAULT_PERMISSIONS,
453 OPT_ALLOW_OTHER,
454 OPT_MAX_READ,
455 OPT_BLKSIZE,
456 OPT_ERR
457};
458
459static const match_table_t tokens = {
460 {OPT_FD, "fd=%u"},
461 {OPT_ROOTMODE, "rootmode=%o"},
462 {OPT_USER_ID, "user_id=%u"},
463 {OPT_GROUP_ID, "group_id=%u"},
464 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
465 {OPT_ALLOW_OTHER, "allow_other"},
466 {OPT_MAX_READ, "max_read=%u"},
467 {OPT_BLKSIZE, "blksize=%u"},
468 {OPT_ERR, NULL}
469};
470
471static int fuse_match_uint(substring_t *s, unsigned int *res)
472{
473 int err = -ENOMEM;
474 char *buf = match_strdup(s);
475 if (buf) {
476 err = kstrtouint(buf, 10, res);
477 kfree(buf);
478 }
479 return err;
480}
481
482static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev,
483 struct user_namespace *user_ns)
484{
485 char *p;
486 memset(d, 0, sizeof(struct fuse_mount_data));
487 d->max_read = ~0;
488 d->blksize = FUSE_DEFAULT_BLKSIZE;
489
490 while ((p = strsep(&opt, ",")) != NULL) {
491 int token;
492 int value;
493 unsigned uv;
494 substring_t args[MAX_OPT_ARGS];
495 if (!*p)
496 continue;
497
498 token = match_token(p, tokens, args);
499 switch (token) {
500 case OPT_FD:
501 if (match_int(&args[0], &value))
502 return 0;
503 d->fd = value;
504 d->fd_present = 1;
505 break;
506
507 case OPT_ROOTMODE:
508 if (match_octal(&args[0], &value))
509 return 0;
510 if (!fuse_valid_type(value))
511 return 0;
512 d->rootmode = value;
513 d->rootmode_present = 1;
514 break;
515
516 case OPT_USER_ID:
517 if (fuse_match_uint(&args[0], &uv))
518 return 0;
519 d->user_id = make_kuid(user_ns, uv);
520 if (!uid_valid(d->user_id))
521 return 0;
522 d->user_id_present = 1;
523 break;
524
525 case OPT_GROUP_ID:
526 if (fuse_match_uint(&args[0], &uv))
527 return 0;
528 d->group_id = make_kgid(user_ns, uv);
529 if (!gid_valid(d->group_id))
530 return 0;
531 d->group_id_present = 1;
532 break;
533
534 case OPT_DEFAULT_PERMISSIONS:
535 d->default_permissions = 1;
536 break;
537
538 case OPT_ALLOW_OTHER:
539 d->allow_other = 1;
540 break;
541
542 case OPT_MAX_READ:
543 if (match_int(&args[0], &value))
544 return 0;
545 d->max_read = value;
546 break;
547
548 case OPT_BLKSIZE:
549 if (!is_bdev || match_int(&args[0], &value))
550 return 0;
551 d->blksize = value;
552 break;
553
554 default:
555 return 0;
556 }
557 }
558
559 if (!d->fd_present || !d->rootmode_present ||
560 !d->user_id_present || !d->group_id_present)
561 return 0;
562
563 return 1;
564}
565
566static int fuse_show_options(struct seq_file *m, struct dentry *root)
567{
568 struct super_block *sb = root->d_sb;
569 struct fuse_conn *fc = get_fuse_conn_super(sb);
570
571 seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
572 seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
573 if (fc->default_permissions)
574 seq_puts(m, ",default_permissions");
575 if (fc->allow_other)
576 seq_puts(m, ",allow_other");
577 if (fc->max_read != ~0)
578 seq_printf(m, ",max_read=%u", fc->max_read);
579 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
580 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
581 return 0;
582}
583
584static void fuse_iqueue_init(struct fuse_iqueue *fiq)
585{
586 memset(fiq, 0, sizeof(struct fuse_iqueue));
587 init_waitqueue_head(&fiq->waitq);
588 INIT_LIST_HEAD(&fiq->pending);
589 INIT_LIST_HEAD(&fiq->interrupts);
590 fiq->forget_list_tail = &fiq->forget_list_head;
591 fiq->connected = 1;
592}
593
594static void fuse_pqueue_init(struct fuse_pqueue *fpq)
595{
596 unsigned int i;
597
598 spin_lock_init(&fpq->lock);
599 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
600 INIT_LIST_HEAD(&fpq->processing[i]);
601 INIT_LIST_HEAD(&fpq->io);
602 fpq->connected = 1;
603}
604
605void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns)
606{
607 memset(fc, 0, sizeof(*fc));
608 spin_lock_init(&fc->lock);
609 spin_lock_init(&fc->bg_lock);
610 init_rwsem(&fc->killsb);
611 refcount_set(&fc->count, 1);
612 atomic_set(&fc->dev_count, 1);
613 init_waitqueue_head(&fc->blocked_waitq);
614 init_waitqueue_head(&fc->reserved_req_waitq);
615 fuse_iqueue_init(&fc->iq);
616 INIT_LIST_HEAD(&fc->bg_queue);
617 INIT_LIST_HEAD(&fc->entry);
618 INIT_LIST_HEAD(&fc->devices);
619 atomic_set(&fc->num_waiting, 0);
620 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
621 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
622 fc->khctr = 0;
623 fc->polled_files = RB_ROOT;
624 fc->blocked = 0;
625 fc->initialized = 0;
626 fc->connected = 1;
627 fc->attr_version = 1;
628 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
629 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
630 fc->user_ns = get_user_ns(user_ns);
631}
632EXPORT_SYMBOL_GPL(fuse_conn_init);
633
634void fuse_conn_put(struct fuse_conn *fc)
635{
636 if (refcount_dec_and_test(&fc->count)) {
637 if (fc->destroy_req)
638 fuse_request_free(fc->destroy_req);
639 put_pid_ns(fc->pid_ns);
640 put_user_ns(fc->user_ns);
641 fc->release(fc);
642 }
643}
644EXPORT_SYMBOL_GPL(fuse_conn_put);
645
646struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
647{
648 refcount_inc(&fc->count);
649 return fc;
650}
651EXPORT_SYMBOL_GPL(fuse_conn_get);
652
653static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
654{
655 struct fuse_attr attr;
656 memset(&attr, 0, sizeof(attr));
657
658 attr.mode = mode;
659 attr.ino = FUSE_ROOT_ID;
660 attr.nlink = 1;
661 return fuse_iget(sb, 1, 0, &attr, 0, 0);
662}
663
664struct fuse_inode_handle {
665 u64 nodeid;
666 u32 generation;
667};
668
669static struct dentry *fuse_get_dentry(struct super_block *sb,
670 struct fuse_inode_handle *handle)
671{
672 struct fuse_conn *fc = get_fuse_conn_super(sb);
673 struct inode *inode;
674 struct dentry *entry;
675 int err = -ESTALE;
676
677 if (handle->nodeid == 0)
678 goto out_err;
679
680 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
681 if (!inode) {
682 struct fuse_entry_out outarg;
683 const struct qstr name = QSTR_INIT(".", 1);
684
685 if (!fc->export_support)
686 goto out_err;
687
688 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
689 &inode);
690 if (err && err != -ENOENT)
691 goto out_err;
692 if (err || !inode) {
693 err = -ESTALE;
694 goto out_err;
695 }
696 err = -EIO;
697 if (get_node_id(inode) != handle->nodeid)
698 goto out_iput;
699 }
700 err = -ESTALE;
701 if (inode->i_generation != handle->generation)
702 goto out_iput;
703
704 entry = d_obtain_alias(inode);
705 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
706 fuse_invalidate_entry_cache(entry);
707
708 return entry;
709
710 out_iput:
711 iput(inode);
712 out_err:
713 return ERR_PTR(err);
714}
715
716static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
717 struct inode *parent)
718{
719 int len = parent ? 6 : 3;
720 u64 nodeid;
721 u32 generation;
722
723 if (*max_len < len) {
724 *max_len = len;
725 return FILEID_INVALID;
726 }
727
728 nodeid = get_fuse_inode(inode)->nodeid;
729 generation = inode->i_generation;
730
731 fh[0] = (u32)(nodeid >> 32);
732 fh[1] = (u32)(nodeid & 0xffffffff);
733 fh[2] = generation;
734
735 if (parent) {
736 nodeid = get_fuse_inode(parent)->nodeid;
737 generation = parent->i_generation;
738
739 fh[3] = (u32)(nodeid >> 32);
740 fh[4] = (u32)(nodeid & 0xffffffff);
741 fh[5] = generation;
742 }
743
744 *max_len = len;
745 return parent ? 0x82 : 0x81;
746}
747
748static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
749 struct fid *fid, int fh_len, int fh_type)
750{
751 struct fuse_inode_handle handle;
752
753 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
754 return NULL;
755
756 handle.nodeid = (u64) fid->raw[0] << 32;
757 handle.nodeid |= (u64) fid->raw[1];
758 handle.generation = fid->raw[2];
759 return fuse_get_dentry(sb, &handle);
760}
761
762static struct dentry *fuse_fh_to_parent(struct super_block *sb,
763 struct fid *fid, int fh_len, int fh_type)
764{
765 struct fuse_inode_handle parent;
766
767 if (fh_type != 0x82 || fh_len < 6)
768 return NULL;
769
770 parent.nodeid = (u64) fid->raw[3] << 32;
771 parent.nodeid |= (u64) fid->raw[4];
772 parent.generation = fid->raw[5];
773 return fuse_get_dentry(sb, &parent);
774}
775
776static struct dentry *fuse_get_parent(struct dentry *child)
777{
778 struct inode *child_inode = d_inode(child);
779 struct fuse_conn *fc = get_fuse_conn(child_inode);
780 struct inode *inode;
781 struct dentry *parent;
782 struct fuse_entry_out outarg;
783 const struct qstr name = QSTR_INIT("..", 2);
784 int err;
785
786 if (!fc->export_support)
787 return ERR_PTR(-ESTALE);
788
789 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
790 &name, &outarg, &inode);
791 if (err) {
792 if (err == -ENOENT)
793 return ERR_PTR(-ESTALE);
794 return ERR_PTR(err);
795 }
796
797 parent = d_obtain_alias(inode);
798 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
799 fuse_invalidate_entry_cache(parent);
800
801 return parent;
802}
803
804static const struct export_operations fuse_export_operations = {
805 .fh_to_dentry = fuse_fh_to_dentry,
806 .fh_to_parent = fuse_fh_to_parent,
807 .encode_fh = fuse_encode_fh,
808 .get_parent = fuse_get_parent,
809};
810
811static const struct super_operations fuse_super_operations = {
812 .alloc_inode = fuse_alloc_inode,
813 .destroy_inode = fuse_destroy_inode,
814 .evict_inode = fuse_evict_inode,
815 .write_inode = fuse_write_inode,
816 .drop_inode = generic_delete_inode,
817 .remount_fs = fuse_remount_fs,
818 .put_super = fuse_put_super,
819 .umount_begin = fuse_umount_begin,
820 .statfs = fuse_statfs,
821 .show_options = fuse_show_options,
822};
823
824static void sanitize_global_limit(unsigned *limit)
825{
826 if (*limit == 0)
827 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
828 sizeof(struct fuse_req);
829
830 if (*limit >= 1 << 16)
831 *limit = (1 << 16) - 1;
832}
833
834static int set_global_limit(const char *val, const struct kernel_param *kp)
835{
836 int rv;
837
838 rv = param_set_uint(val, kp);
839 if (rv)
840 return rv;
841
842 sanitize_global_limit((unsigned *)kp->arg);
843
844 return 0;
845}
846
847static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
848{
849 int cap_sys_admin = capable(CAP_SYS_ADMIN);
850
851 if (arg->minor < 13)
852 return;
853
854 sanitize_global_limit(&max_user_bgreq);
855 sanitize_global_limit(&max_user_congthresh);
856
857 spin_lock(&fc->bg_lock);
858 if (arg->max_background) {
859 fc->max_background = arg->max_background;
860
861 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
862 fc->max_background = max_user_bgreq;
863 }
864 if (arg->congestion_threshold) {
865 fc->congestion_threshold = arg->congestion_threshold;
866
867 if (!cap_sys_admin &&
868 fc->congestion_threshold > max_user_congthresh)
869 fc->congestion_threshold = max_user_congthresh;
870 }
871 spin_unlock(&fc->bg_lock);
872}
873
874static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
875{
876 struct fuse_init_out *arg = &req->misc.init_out;
877
878 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
879 fc->conn_error = 1;
880 else {
881 unsigned long ra_pages;
882
883 process_init_limits(fc, arg);
884
885 if (arg->minor >= 6) {
886 ra_pages = arg->max_readahead / PAGE_SIZE;
887 if (arg->flags & FUSE_ASYNC_READ)
888 fc->async_read = 1;
889 if (!(arg->flags & FUSE_POSIX_LOCKS))
890 fc->no_lock = 1;
891 if (arg->minor >= 17) {
892 if (!(arg->flags & FUSE_FLOCK_LOCKS))
893 fc->no_flock = 1;
894 } else {
895 if (!(arg->flags & FUSE_POSIX_LOCKS))
896 fc->no_flock = 1;
897 }
898 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
899 fc->atomic_o_trunc = 1;
900 if (arg->minor >= 9) {
901
902 if (arg->flags & FUSE_EXPORT_SUPPORT)
903 fc->export_support = 1;
904 }
905 if (arg->flags & FUSE_BIG_WRITES)
906 fc->big_writes = 1;
907 if (arg->flags & FUSE_DONT_MASK)
908 fc->dont_mask = 1;
909 if (arg->flags & FUSE_AUTO_INVAL_DATA)
910 fc->auto_inval_data = 1;
911 if (arg->flags & FUSE_DO_READDIRPLUS) {
912 fc->do_readdirplus = 1;
913 if (arg->flags & FUSE_READDIRPLUS_AUTO)
914 fc->readdirplus_auto = 1;
915 }
916 if (arg->flags & FUSE_ASYNC_DIO)
917 fc->async_dio = 1;
918 if (arg->flags & FUSE_WRITEBACK_CACHE)
919 fc->writeback_cache = 1;
920 if (arg->flags & FUSE_PARALLEL_DIROPS)
921 fc->parallel_dirops = 1;
922 if (arg->flags & FUSE_HANDLE_KILLPRIV)
923 fc->handle_killpriv = 1;
924 if (arg->time_gran && arg->time_gran <= 1000000000)
925 fc->sb->s_time_gran = arg->time_gran;
926 if ((arg->flags & FUSE_POSIX_ACL)) {
927 fc->default_permissions = 1;
928 fc->posix_acl = 1;
929 fc->sb->s_xattr = fuse_acl_xattr_handlers;
930 }
931 if (arg->flags & FUSE_CACHE_SYMLINKS)
932 fc->cache_symlinks = 1;
933 if (arg->flags & FUSE_ABORT_ERROR)
934 fc->abort_err = 1;
935 if (arg->flags & FUSE_MAX_PAGES) {
936 fc->max_pages =
937 min_t(unsigned int, FUSE_MAX_MAX_PAGES,
938 max_t(unsigned int, arg->max_pages, 1));
939 }
940 } else {
941 ra_pages = fc->max_read / PAGE_SIZE;
942 fc->no_lock = 1;
943 fc->no_flock = 1;
944 }
945
946 fc->sb->s_bdi->ra_pages =
947 min(fc->sb->s_bdi->ra_pages, ra_pages);
948 fc->minor = arg->minor;
949 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
950 fc->max_write = max_t(unsigned, 4096, fc->max_write);
951 fc->conn_init = 1;
952 }
953 fuse_set_initialized(fc);
954 wake_up_all(&fc->blocked_waitq);
955}
956
957static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
958{
959 struct fuse_init_in *arg = &req->misc.init_in;
960
961 arg->major = FUSE_KERNEL_VERSION;
962 arg->minor = FUSE_KERNEL_MINOR_VERSION;
963 arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
964 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
965 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
966 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
967 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
968 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
969 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
970 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
971 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS;
972 req->in.h.opcode = FUSE_INIT;
973 req->in.numargs = 1;
974 req->in.args[0].size = sizeof(*arg);
975 req->in.args[0].value = arg;
976 req->out.numargs = 1;
977
978
979
980 req->out.argvar = 1;
981 req->out.args[0].size = sizeof(struct fuse_init_out);
982 req->out.args[0].value = &req->misc.init_out;
983 req->end = process_init_reply;
984 fuse_request_send_background(fc, req);
985}
986
987static void fuse_free_conn(struct fuse_conn *fc)
988{
989 WARN_ON(!list_empty(&fc->devices));
990 kfree_rcu(fc, rcu);
991}
992
993static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
994{
995 int err;
996 char *suffix = "";
997
998 if (sb->s_bdev) {
999 suffix = "-fuseblk";
1000
1001
1002
1003
1004 bdi_put(sb->s_bdi);
1005 sb->s_bdi = &noop_backing_dev_info;
1006 }
1007 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1008 MINOR(fc->dev), suffix);
1009 if (err)
1010 return err;
1011
1012 sb->s_bdi->ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
1013
1014 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028 bdi_set_max_ratio(sb->s_bdi, 1);
1029
1030 return 0;
1031}
1032
1033struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1034{
1035 struct fuse_dev *fud;
1036 struct list_head *pq;
1037
1038 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1039 if (!fud)
1040 return NULL;
1041
1042 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1043 if (!pq) {
1044 kfree(fud);
1045 return NULL;
1046 }
1047
1048 fud->pq.processing = pq;
1049 fud->fc = fuse_conn_get(fc);
1050 fuse_pqueue_init(&fud->pq);
1051
1052 spin_lock(&fc->lock);
1053 list_add_tail(&fud->entry, &fc->devices);
1054 spin_unlock(&fc->lock);
1055
1056 return fud;
1057}
1058EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1059
1060void fuse_dev_free(struct fuse_dev *fud)
1061{
1062 struct fuse_conn *fc = fud->fc;
1063
1064 if (fc) {
1065 spin_lock(&fc->lock);
1066 list_del(&fud->entry);
1067 spin_unlock(&fc->lock);
1068
1069 fuse_conn_put(fc);
1070 }
1071 kfree(fud->pq.processing);
1072 kfree(fud);
1073}
1074EXPORT_SYMBOL_GPL(fuse_dev_free);
1075
1076static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1077{
1078 struct fuse_dev *fud;
1079 struct fuse_conn *fc;
1080 struct inode *root;
1081 struct fuse_mount_data d;
1082 struct file *file;
1083 struct dentry *root_dentry;
1084 struct fuse_req *init_req;
1085 int err;
1086 int is_bdev = sb->s_bdev != NULL;
1087
1088 err = -EINVAL;
1089 if (sb->s_flags & SB_MANDLOCK)
1090 goto err;
1091
1092 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1093
1094 if (!parse_fuse_opt(data, &d, is_bdev, sb->s_user_ns))
1095 goto err;
1096
1097 if (is_bdev) {
1098#ifdef CONFIG_BLOCK
1099 err = -EINVAL;
1100 if (!sb_set_blocksize(sb, d.blksize))
1101 goto err;
1102#endif
1103 } else {
1104 sb->s_blocksize = PAGE_SIZE;
1105 sb->s_blocksize_bits = PAGE_SHIFT;
1106 }
1107 sb->s_magic = FUSE_SUPER_MAGIC;
1108 sb->s_op = &fuse_super_operations;
1109 sb->s_xattr = fuse_xattr_handlers;
1110 sb->s_maxbytes = MAX_LFS_FILESIZE;
1111 sb->s_time_gran = 1;
1112 sb->s_export_op = &fuse_export_operations;
1113 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1114 if (sb->s_user_ns != &init_user_ns)
1115 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1116
1117 file = fget(d.fd);
1118 err = -EINVAL;
1119 if (!file)
1120 goto err;
1121
1122
1123
1124
1125
1126 if (file->f_op != &fuse_dev_operations ||
1127 file->f_cred->user_ns != sb->s_user_ns)
1128 goto err_fput;
1129
1130
1131
1132
1133
1134 if (sb->s_user_ns != &init_user_ns)
1135 sb->s_xattr = fuse_no_acl_xattr_handlers;
1136
1137 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1138 err = -ENOMEM;
1139 if (!fc)
1140 goto err_fput;
1141
1142 fuse_conn_init(fc, sb->s_user_ns);
1143 fc->release = fuse_free_conn;
1144
1145 fud = fuse_dev_alloc(fc);
1146 if (!fud)
1147 goto err_put_conn;
1148
1149 fc->dev = sb->s_dev;
1150 fc->sb = sb;
1151 err = fuse_bdi_init(fc, sb);
1152 if (err)
1153 goto err_dev_free;
1154
1155
1156 if (sb->s_flags & SB_POSIXACL)
1157 fc->dont_mask = 1;
1158 sb->s_flags |= SB_POSIXACL;
1159
1160 fc->default_permissions = d.default_permissions;
1161 fc->allow_other = d.allow_other;
1162 fc->user_id = d.user_id;
1163 fc->group_id = d.group_id;
1164 fc->max_read = max_t(unsigned, 4096, d.max_read);
1165 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
1166
1167
1168 sb->s_fs_info = fc;
1169
1170 err = -ENOMEM;
1171 root = fuse_get_root_inode(sb, d.rootmode);
1172 sb->s_d_op = &fuse_root_dentry_operations;
1173 root_dentry = d_make_root(root);
1174 if (!root_dentry)
1175 goto err_dev_free;
1176
1177 sb->s_d_op = &fuse_dentry_operations;
1178
1179 init_req = fuse_request_alloc(0);
1180 if (!init_req)
1181 goto err_put_root;
1182 __set_bit(FR_BACKGROUND, &init_req->flags);
1183
1184 if (is_bdev) {
1185 fc->destroy_req = fuse_request_alloc(0);
1186 if (!fc->destroy_req)
1187 goto err_free_init_req;
1188 }
1189
1190 mutex_lock(&fuse_mutex);
1191 err = -EINVAL;
1192 if (file->private_data)
1193 goto err_unlock;
1194
1195 err = fuse_ctl_add_conn(fc);
1196 if (err)
1197 goto err_unlock;
1198
1199 list_add_tail(&fc->entry, &fuse_conn_list);
1200 sb->s_root = root_dentry;
1201 file->private_data = fud;
1202 mutex_unlock(&fuse_mutex);
1203
1204
1205
1206
1207
1208 fput(file);
1209
1210 fuse_send_init(fc, init_req);
1211
1212 return 0;
1213
1214 err_unlock:
1215 mutex_unlock(&fuse_mutex);
1216 err_free_init_req:
1217 fuse_request_free(init_req);
1218 err_put_root:
1219 dput(root_dentry);
1220 err_dev_free:
1221 fuse_dev_free(fud);
1222 err_put_conn:
1223 fuse_conn_put(fc);
1224 sb->s_fs_info = NULL;
1225 err_fput:
1226 fput(file);
1227 err:
1228 return err;
1229}
1230
1231static struct dentry *fuse_mount(struct file_system_type *fs_type,
1232 int flags, const char *dev_name,
1233 void *raw_data)
1234{
1235 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1236}
1237
1238static void fuse_sb_destroy(struct super_block *sb)
1239{
1240 struct fuse_conn *fc = get_fuse_conn_super(sb);
1241
1242 if (fc) {
1243 fuse_send_destroy(fc);
1244
1245 fuse_abort_conn(fc, false);
1246 fuse_wait_aborted(fc);
1247
1248 down_write(&fc->killsb);
1249 fc->sb = NULL;
1250 up_write(&fc->killsb);
1251 }
1252}
1253
1254static void fuse_kill_sb_anon(struct super_block *sb)
1255{
1256 fuse_sb_destroy(sb);
1257 kill_anon_super(sb);
1258}
1259
1260static struct file_system_type fuse_fs_type = {
1261 .owner = THIS_MODULE,
1262 .name = "fuse",
1263 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1264 .mount = fuse_mount,
1265 .kill_sb = fuse_kill_sb_anon,
1266};
1267MODULE_ALIAS_FS("fuse");
1268
1269#ifdef CONFIG_BLOCK
1270static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1271 int flags, const char *dev_name,
1272 void *raw_data)
1273{
1274 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1275}
1276
1277static void fuse_kill_sb_blk(struct super_block *sb)
1278{
1279 fuse_sb_destroy(sb);
1280 kill_block_super(sb);
1281}
1282
1283static struct file_system_type fuseblk_fs_type = {
1284 .owner = THIS_MODULE,
1285 .name = "fuseblk",
1286 .mount = fuse_mount_blk,
1287 .kill_sb = fuse_kill_sb_blk,
1288 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1289};
1290MODULE_ALIAS_FS("fuseblk");
1291
1292static inline int register_fuseblk(void)
1293{
1294 return register_filesystem(&fuseblk_fs_type);
1295}
1296
1297static inline void unregister_fuseblk(void)
1298{
1299 unregister_filesystem(&fuseblk_fs_type);
1300}
1301#else
1302static inline int register_fuseblk(void)
1303{
1304 return 0;
1305}
1306
1307static inline void unregister_fuseblk(void)
1308{
1309}
1310#endif
1311
1312static void fuse_inode_init_once(void *foo)
1313{
1314 struct inode *inode = foo;
1315
1316 inode_init_once(inode);
1317}
1318
1319static int __init fuse_fs_init(void)
1320{
1321 int err;
1322
1323 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1324 sizeof(struct fuse_inode), 0,
1325 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1326 fuse_inode_init_once);
1327 err = -ENOMEM;
1328 if (!fuse_inode_cachep)
1329 goto out;
1330
1331 err = register_fuseblk();
1332 if (err)
1333 goto out2;
1334
1335 err = register_filesystem(&fuse_fs_type);
1336 if (err)
1337 goto out3;
1338
1339 return 0;
1340
1341 out3:
1342 unregister_fuseblk();
1343 out2:
1344 kmem_cache_destroy(fuse_inode_cachep);
1345 out:
1346 return err;
1347}
1348
1349static void fuse_fs_cleanup(void)
1350{
1351 unregister_filesystem(&fuse_fs_type);
1352 unregister_fuseblk();
1353
1354
1355
1356
1357
1358 rcu_barrier();
1359 kmem_cache_destroy(fuse_inode_cachep);
1360}
1361
1362static struct kobject *fuse_kobj;
1363
1364static int fuse_sysfs_init(void)
1365{
1366 int err;
1367
1368 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1369 if (!fuse_kobj) {
1370 err = -ENOMEM;
1371 goto out_err;
1372 }
1373
1374 err = sysfs_create_mount_point(fuse_kobj, "connections");
1375 if (err)
1376 goto out_fuse_unregister;
1377
1378 return 0;
1379
1380 out_fuse_unregister:
1381 kobject_put(fuse_kobj);
1382 out_err:
1383 return err;
1384}
1385
1386static void fuse_sysfs_cleanup(void)
1387{
1388 sysfs_remove_mount_point(fuse_kobj, "connections");
1389 kobject_put(fuse_kobj);
1390}
1391
1392static int __init fuse_init(void)
1393{
1394 int res;
1395
1396 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1397 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1398
1399 INIT_LIST_HEAD(&fuse_conn_list);
1400 res = fuse_fs_init();
1401 if (res)
1402 goto err;
1403
1404 res = fuse_dev_init();
1405 if (res)
1406 goto err_fs_cleanup;
1407
1408 res = fuse_sysfs_init();
1409 if (res)
1410 goto err_dev_cleanup;
1411
1412 res = fuse_ctl_init();
1413 if (res)
1414 goto err_sysfs_cleanup;
1415
1416 sanitize_global_limit(&max_user_bgreq);
1417 sanitize_global_limit(&max_user_congthresh);
1418
1419 return 0;
1420
1421 err_sysfs_cleanup:
1422 fuse_sysfs_cleanup();
1423 err_dev_cleanup:
1424 fuse_dev_cleanup();
1425 err_fs_cleanup:
1426 fuse_fs_cleanup();
1427 err:
1428 return res;
1429}
1430
1431static void __exit fuse_exit(void)
1432{
1433 printk(KERN_DEBUG "fuse exit\n");
1434
1435 fuse_ctl_cleanup();
1436 fuse_sysfs_cleanup();
1437 fuse_fs_cleanup();
1438 fuse_dev_cleanup();
1439}
1440
1441module_init(fuse_init);
1442module_exit(fuse_exit);
1443
