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/parser.h>
19 #include <linux/statfs.h>
20 #include <linux/random.h>
21 #include <linux/sched.h>
22 #include <linux/exportfs.h>
23
24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
25 MODULE_DESCRIPTION("Filesystem in Userspace");
26 MODULE_LICENSE("GPL");
27
28 static struct kmem_cache *fuse_inode_cachep;
29 struct list_head fuse_conn_list;
30 DEFINE_MUTEX(fuse_mutex);
31
32 static int set_global_limit(const char *val, struct kernel_param *kp);
33
34 unsigned max_user_bgreq;
35 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
36 &max_user_bgreq, 0644);
37 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
38 MODULE_PARM_DESC(max_user_bgreq,
39 "Global limit for the maximum number of backgrounded requests an "
40 "unprivileged user can set");
41
42 unsigned max_user_congthresh;
43 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
44 &max_user_congthresh, 0644);
45 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
46 MODULE_PARM_DESC(max_user_congthresh,
47 "Global limit for the maximum congestion threshold an "
48 "unprivileged user can set");
49
50 #define FUSE_SUPER_MAGIC 0x65735546
51
52 #define FUSE_DEFAULT_BLKSIZE 512
53
54 /** Maximum number of outstanding background requests */
55 #define FUSE_DEFAULT_MAX_BACKGROUND 12
56
57 /** Congestion starts at 75% of maximum */
58 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
59
60 struct fuse_mount_data {
61 int fd;
62 unsigned rootmode;
63 unsigned user_id;
64 unsigned group_id;
65 unsigned fd_present:1;
66 unsigned rootmode_present:1;
67 unsigned user_id_present:1;
68 unsigned group_id_present:1;
69 unsigned flags;
70 unsigned max_read;
71 unsigned blksize;
72 };
73
fuse_alloc_forget(void)74 struct fuse_forget_link *fuse_alloc_forget(void)
75 {
76 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
77 }
78
fuse_alloc_inode(struct super_block * sb)79 static struct inode *fuse_alloc_inode(struct super_block *sb)
80 {
81 struct inode *inode;
82 struct fuse_inode *fi;
83
84 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
85 if (!inode)
86 return NULL;
87
88 fi = get_fuse_inode(inode);
89 fi->i_time = 0;
90 fi->nodeid = 0;
91 fi->nlookup = 0;
92 fi->attr_version = 0;
93 fi->writectr = 0;
94 fi->orig_ino = 0;
95 fi->state = 0;
96 INIT_LIST_HEAD(&fi->write_files);
97 INIT_LIST_HEAD(&fi->queued_writes);
98 INIT_LIST_HEAD(&fi->writepages);
99 init_waitqueue_head(&fi->page_waitq);
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
fuse_i_callback(struct rcu_head * head)109 static 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
fuse_destroy_inode(struct inode * inode)115 static void fuse_destroy_inode(struct inode *inode)
116 {
117 struct fuse_inode *fi = get_fuse_inode(inode);
118 BUG_ON(!list_empty(&fi->write_files));
119 BUG_ON(!list_empty(&fi->queued_writes));
120 kfree(fi->forget);
121 call_rcu(&inode->i_rcu, fuse_i_callback);
122 }
123
fuse_evict_inode(struct inode * inode)124 static void fuse_evict_inode(struct inode *inode)
125 {
126 truncate_inode_pages(&inode->i_data, 0);
127 end_writeback(inode);
128 if (inode->i_sb->s_flags & MS_ACTIVE) {
129 struct fuse_conn *fc = get_fuse_conn(inode);
130 struct fuse_inode *fi = get_fuse_inode(inode);
131 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
132 fi->forget = NULL;
133 }
134 }
135
fuse_remount_fs(struct super_block * sb,int * flags,char * data)136 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
137 {
138 if (*flags & MS_MANDLOCK)
139 return -EINVAL;
140
141 return 0;
142 }
143
144 /*
145 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
146 * so that it will fit.
147 */
fuse_squash_ino(u64 ino64)148 static ino_t fuse_squash_ino(u64 ino64)
149 {
150 ino_t ino = (ino_t) ino64;
151 if (sizeof(ino_t) < sizeof(u64))
152 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
153 return ino;
154 }
155
fuse_change_attributes_common(struct inode * inode,struct fuse_attr * attr,u64 attr_valid)156 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
157 u64 attr_valid)
158 {
159 struct fuse_conn *fc = get_fuse_conn(inode);
160 struct fuse_inode *fi = get_fuse_inode(inode);
161
162 fi->attr_version = ++fc->attr_version;
163 fi->i_time = attr_valid;
164
165 inode->i_ino = fuse_squash_ino(attr->ino);
166 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
167 set_nlink(inode, attr->nlink);
168 inode->i_uid = attr->uid;
169 inode->i_gid = attr->gid;
170 inode->i_blocks = attr->blocks;
171 inode->i_atime.tv_sec = attr->atime;
172 inode->i_atime.tv_nsec = attr->atimensec;
173 inode->i_mtime.tv_sec = attr->mtime;
174 inode->i_mtime.tv_nsec = attr->mtimensec;
175 inode->i_ctime.tv_sec = attr->ctime;
176 inode->i_ctime.tv_nsec = attr->ctimensec;
177
178 if (attr->blksize != 0)
179 inode->i_blkbits = ilog2(attr->blksize);
180 else
181 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
182
183 /*
184 * Don't set the sticky bit in i_mode, unless we want the VFS
185 * to check permissions. This prevents failures due to the
186 * check in may_delete().
187 */
188 fi->orig_i_mode = inode->i_mode;
189 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
190 inode->i_mode &= ~S_ISVTX;
191
192 fi->orig_ino = attr->ino;
193 }
194
fuse_change_attributes(struct inode * inode,struct fuse_attr * attr,u64 attr_valid,u64 attr_version)195 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
196 u64 attr_valid, u64 attr_version)
197 {
198 struct fuse_conn *fc = get_fuse_conn(inode);
199 struct fuse_inode *fi = get_fuse_inode(inode);
200 loff_t oldsize;
201
202 spin_lock(&fc->lock);
203 if ((attr_version != 0 && fi->attr_version > attr_version) ||
204 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
205 spin_unlock(&fc->lock);
206 return;
207 }
208
209 fuse_change_attributes_common(inode, attr, attr_valid);
210
211 oldsize = inode->i_size;
212 i_size_write(inode, attr->size);
213 spin_unlock(&fc->lock);
214
215 if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
216 truncate_pagecache(inode, oldsize, attr->size);
217 invalidate_inode_pages2(inode->i_mapping);
218 }
219 }
220
fuse_init_inode(struct inode * inode,struct fuse_attr * attr)221 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
222 {
223 inode->i_mode = attr->mode & S_IFMT;
224 inode->i_size = attr->size;
225 if (S_ISREG(inode->i_mode)) {
226 fuse_init_common(inode);
227 fuse_init_file_inode(inode);
228 } else if (S_ISDIR(inode->i_mode))
229 fuse_init_dir(inode);
230 else if (S_ISLNK(inode->i_mode))
231 fuse_init_symlink(inode);
232 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
233 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
234 fuse_init_common(inode);
235 init_special_inode(inode, inode->i_mode,
236 new_decode_dev(attr->rdev));
237 } else
238 BUG();
239 }
240
fuse_inode_eq(struct inode * inode,void * _nodeidp)241 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
242 {
243 u64 nodeid = *(u64 *) _nodeidp;
244 if (get_node_id(inode) == nodeid)
245 return 1;
246 else
247 return 0;
248 }
249
fuse_inode_set(struct inode * inode,void * _nodeidp)250 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
251 {
252 u64 nodeid = *(u64 *) _nodeidp;
253 get_fuse_inode(inode)->nodeid = nodeid;
254 return 0;
255 }
256
fuse_iget(struct super_block * sb,u64 nodeid,int generation,struct fuse_attr * attr,u64 attr_valid,u64 attr_version)257 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
258 int generation, struct fuse_attr *attr,
259 u64 attr_valid, u64 attr_version)
260 {
261 struct inode *inode;
262 struct fuse_inode *fi;
263 struct fuse_conn *fc = get_fuse_conn_super(sb);
264
265 retry:
266 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
267 if (!inode)
268 return NULL;
269
270 if ((inode->i_state & I_NEW)) {
271 inode->i_flags |= S_NOATIME|S_NOCMTIME;
272 inode->i_generation = generation;
273 inode->i_data.backing_dev_info = &fc->bdi;
274 fuse_init_inode(inode, attr);
275 unlock_new_inode(inode);
276 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
277 /* Inode has changed type, any I/O on the old should fail */
278 make_bad_inode(inode);
279 iput(inode);
280 goto retry;
281 }
282
283 fi = get_fuse_inode(inode);
284 spin_lock(&fc->lock);
285 fi->nlookup++;
286 spin_unlock(&fc->lock);
287 fuse_change_attributes(inode, attr, attr_valid, attr_version);
288
289 return inode;
290 }
291
fuse_reverse_inval_inode(struct super_block * sb,u64 nodeid,loff_t offset,loff_t len)292 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
293 loff_t offset, loff_t len)
294 {
295 struct inode *inode;
296 pgoff_t pg_start;
297 pgoff_t pg_end;
298
299 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
300 if (!inode)
301 return -ENOENT;
302
303 fuse_invalidate_attr(inode);
304 if (offset >= 0) {
305 pg_start = offset >> PAGE_CACHE_SHIFT;
306 if (len <= 0)
307 pg_end = -1;
308 else
309 pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
310 invalidate_inode_pages2_range(inode->i_mapping,
311 pg_start, pg_end);
312 }
313 iput(inode);
314 return 0;
315 }
316
fuse_umount_begin(struct super_block * sb)317 static void fuse_umount_begin(struct super_block *sb)
318 {
319 fuse_abort_conn(get_fuse_conn_super(sb));
320 }
321
fuse_send_destroy(struct fuse_conn * fc)322 static void fuse_send_destroy(struct fuse_conn *fc)
323 {
324 struct fuse_req *req = fc->destroy_req;
325 if (req && fc->conn_init) {
326 fc->destroy_req = NULL;
327 req->in.h.opcode = FUSE_DESTROY;
328 req->force = 1;
329 fuse_request_send(fc, req);
330 fuse_put_request(fc, req);
331 }
332 }
333
fuse_bdi_destroy(struct fuse_conn * fc)334 static void fuse_bdi_destroy(struct fuse_conn *fc)
335 {
336 if (fc->bdi_initialized)
337 bdi_destroy(&fc->bdi);
338 }
339
fuse_conn_kill(struct fuse_conn * fc)340 void fuse_conn_kill(struct fuse_conn *fc)
341 {
342 spin_lock(&fc->lock);
343 fc->connected = 0;
344 fc->blocked = 0;
345 spin_unlock(&fc->lock);
346 /* Flush all readers on this fs */
347 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
348 wake_up_all(&fc->waitq);
349 wake_up_all(&fc->blocked_waitq);
350 wake_up_all(&fc->reserved_req_waitq);
351 mutex_lock(&fuse_mutex);
352 list_del(&fc->entry);
353 fuse_ctl_remove_conn(fc);
354 mutex_unlock(&fuse_mutex);
355 fuse_bdi_destroy(fc);
356 }
357 EXPORT_SYMBOL_GPL(fuse_conn_kill);
358
fuse_put_super(struct super_block * sb)359 static void fuse_put_super(struct super_block *sb)
360 {
361 struct fuse_conn *fc = get_fuse_conn_super(sb);
362
363 fuse_send_destroy(fc);
364 fuse_conn_kill(fc);
365 fuse_conn_put(fc);
366 }
367
convert_fuse_statfs(struct kstatfs * stbuf,struct fuse_kstatfs * attr)368 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
369 {
370 stbuf->f_type = FUSE_SUPER_MAGIC;
371 stbuf->f_bsize = attr->bsize;
372 stbuf->f_frsize = attr->frsize;
373 stbuf->f_blocks = attr->blocks;
374 stbuf->f_bfree = attr->bfree;
375 stbuf->f_bavail = attr->bavail;
376 stbuf->f_files = attr->files;
377 stbuf->f_ffree = attr->ffree;
378 stbuf->f_namelen = attr->namelen;
379 /* fsid is left zero */
380 }
381
fuse_statfs(struct dentry * dentry,struct kstatfs * buf)382 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
383 {
384 struct super_block *sb = dentry->d_sb;
385 struct fuse_conn *fc = get_fuse_conn_super(sb);
386 struct fuse_req *req;
387 struct fuse_statfs_out outarg;
388 int err;
389
390 if (!fuse_allow_task(fc, current)) {
391 buf->f_type = FUSE_SUPER_MAGIC;
392 return 0;
393 }
394
395 req = fuse_get_req(fc);
396 if (IS_ERR(req))
397 return PTR_ERR(req);
398
399 memset(&outarg, 0, sizeof(outarg));
400 req->in.numargs = 0;
401 req->in.h.opcode = FUSE_STATFS;
402 req->in.h.nodeid = get_node_id(dentry->d_inode);
403 req->out.numargs = 1;
404 req->out.args[0].size =
405 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
406 req->out.args[0].value = &outarg;
407 fuse_request_send(fc, req);
408 err = req->out.h.error;
409 if (!err)
410 convert_fuse_statfs(buf, &outarg.st);
411 fuse_put_request(fc, req);
412 return err;
413 }
414
415 enum {
416 OPT_FD,
417 OPT_ROOTMODE,
418 OPT_USER_ID,
419 OPT_GROUP_ID,
420 OPT_DEFAULT_PERMISSIONS,
421 OPT_ALLOW_OTHER,
422 OPT_MAX_READ,
423 OPT_BLKSIZE,
424 OPT_ERR
425 };
426
427 static const match_table_t tokens = {
428 {OPT_FD, "fd=%u"},
429 {OPT_ROOTMODE, "rootmode=%o"},
430 {OPT_USER_ID, "user_id=%u"},
431 {OPT_GROUP_ID, "group_id=%u"},
432 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
433 {OPT_ALLOW_OTHER, "allow_other"},
434 {OPT_MAX_READ, "max_read=%u"},
435 {OPT_BLKSIZE, "blksize=%u"},
436 {OPT_ERR, NULL}
437 };
438
parse_fuse_opt(char * opt,struct fuse_mount_data * d,int is_bdev)439 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
440 {
441 char *p;
442 memset(d, 0, sizeof(struct fuse_mount_data));
443 d->max_read = ~0;
444 d->blksize = FUSE_DEFAULT_BLKSIZE;
445
446 while ((p = strsep(&opt, ",")) != NULL) {
447 int token;
448 int value;
449 substring_t args[MAX_OPT_ARGS];
450 if (!*p)
451 continue;
452
453 token = match_token(p, tokens, args);
454 switch (token) {
455 case OPT_FD:
456 if (match_int(&args[0], &value))
457 return 0;
458 d->fd = value;
459 d->fd_present = 1;
460 break;
461
462 case OPT_ROOTMODE:
463 if (match_octal(&args[0], &value))
464 return 0;
465 if (!fuse_valid_type(value))
466 return 0;
467 d->rootmode = value;
468 d->rootmode_present = 1;
469 break;
470
471 case OPT_USER_ID:
472 if (match_int(&args[0], &value))
473 return 0;
474 d->user_id = value;
475 d->user_id_present = 1;
476 break;
477
478 case OPT_GROUP_ID:
479 if (match_int(&args[0], &value))
480 return 0;
481 d->group_id = value;
482 d->group_id_present = 1;
483 break;
484
485 case OPT_DEFAULT_PERMISSIONS:
486 d->flags |= FUSE_DEFAULT_PERMISSIONS;
487 break;
488
489 case OPT_ALLOW_OTHER:
490 d->flags |= FUSE_ALLOW_OTHER;
491 break;
492
493 case OPT_MAX_READ:
494 if (match_int(&args[0], &value))
495 return 0;
496 d->max_read = value;
497 break;
498
499 case OPT_BLKSIZE:
500 if (!is_bdev || match_int(&args[0], &value))
501 return 0;
502 d->blksize = value;
503 break;
504
505 default:
506 return 0;
507 }
508 }
509
510 if (!d->fd_present || !d->rootmode_present ||
511 !d->user_id_present || !d->group_id_present)
512 return 0;
513
514 return 1;
515 }
516
fuse_show_options(struct seq_file * m,struct dentry * root)517 static int fuse_show_options(struct seq_file *m, struct dentry *root)
518 {
519 struct super_block *sb = root->d_sb;
520 struct fuse_conn *fc = get_fuse_conn_super(sb);
521
522 seq_printf(m, ",user_id=%u", fc->user_id);
523 seq_printf(m, ",group_id=%u", fc->group_id);
524 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
525 seq_puts(m, ",default_permissions");
526 if (fc->flags & FUSE_ALLOW_OTHER)
527 seq_puts(m, ",allow_other");
528 if (fc->max_read != ~0)
529 seq_printf(m, ",max_read=%u", fc->max_read);
530 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
531 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
532 return 0;
533 }
534
fuse_conn_init(struct fuse_conn * fc)535 void fuse_conn_init(struct fuse_conn *fc)
536 {
537 memset(fc, 0, sizeof(*fc));
538 spin_lock_init(&fc->lock);
539 mutex_init(&fc->inst_mutex);
540 init_rwsem(&fc->killsb);
541 atomic_set(&fc->count, 1);
542 init_waitqueue_head(&fc->waitq);
543 init_waitqueue_head(&fc->blocked_waitq);
544 init_waitqueue_head(&fc->reserved_req_waitq);
545 INIT_LIST_HEAD(&fc->pending);
546 INIT_LIST_HEAD(&fc->processing);
547 INIT_LIST_HEAD(&fc->io);
548 INIT_LIST_HEAD(&fc->interrupts);
549 INIT_LIST_HEAD(&fc->bg_queue);
550 INIT_LIST_HEAD(&fc->entry);
551 fc->forget_list_tail = &fc->forget_list_head;
552 atomic_set(&fc->num_waiting, 0);
553 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
554 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
555 fc->khctr = 0;
556 fc->polled_files = RB_ROOT;
557 fc->reqctr = 0;
558 fc->blocked = 1;
559 fc->attr_version = 1;
560 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
561 }
562 EXPORT_SYMBOL_GPL(fuse_conn_init);
563
fuse_conn_put(struct fuse_conn * fc)564 void fuse_conn_put(struct fuse_conn *fc)
565 {
566 if (atomic_dec_and_test(&fc->count)) {
567 if (fc->destroy_req)
568 fuse_request_free(fc->destroy_req);
569 mutex_destroy(&fc->inst_mutex);
570 fc->release(fc);
571 }
572 }
573 EXPORT_SYMBOL_GPL(fuse_conn_put);
574
fuse_conn_get(struct fuse_conn * fc)575 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
576 {
577 atomic_inc(&fc->count);
578 return fc;
579 }
580 EXPORT_SYMBOL_GPL(fuse_conn_get);
581
fuse_get_root_inode(struct super_block * sb,unsigned mode)582 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
583 {
584 struct fuse_attr attr;
585 memset(&attr, 0, sizeof(attr));
586
587 attr.mode = mode;
588 attr.ino = FUSE_ROOT_ID;
589 attr.nlink = 1;
590 return fuse_iget(sb, 1, 0, &attr, 0, 0);
591 }
592
593 struct fuse_inode_handle {
594 u64 nodeid;
595 u32 generation;
596 };
597
fuse_get_dentry(struct super_block * sb,struct fuse_inode_handle * handle)598 static struct dentry *fuse_get_dentry(struct super_block *sb,
599 struct fuse_inode_handle *handle)
600 {
601 struct fuse_conn *fc = get_fuse_conn_super(sb);
602 struct inode *inode;
603 struct dentry *entry;
604 int err = -ESTALE;
605
606 if (handle->nodeid == 0)
607 goto out_err;
608
609 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
610 if (!inode) {
611 struct fuse_entry_out outarg;
612 struct qstr name;
613
614 if (!fc->export_support)
615 goto out_err;
616
617 name.len = 1;
618 name.name = ".";
619 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
620 &inode);
621 if (err && err != -ENOENT)
622 goto out_err;
623 if (err || !inode) {
624 err = -ESTALE;
625 goto out_err;
626 }
627 err = -EIO;
628 if (get_node_id(inode) != handle->nodeid)
629 goto out_iput;
630 }
631 err = -ESTALE;
632 if (inode->i_generation != handle->generation)
633 goto out_iput;
634
635 entry = d_obtain_alias(inode);
636 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
637 fuse_invalidate_entry_cache(entry);
638
639 return entry;
640
641 out_iput:
642 iput(inode);
643 out_err:
644 return ERR_PTR(err);
645 }
646
fuse_encode_fh(struct dentry * dentry,u32 * fh,int * max_len,int connectable)647 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
648 int connectable)
649 {
650 struct inode *inode = dentry->d_inode;
651 bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
652 int len = encode_parent ? 6 : 3;
653 u64 nodeid;
654 u32 generation;
655
656 if (*max_len < len) {
657 *max_len = len;
658 return 255;
659 }
660
661 nodeid = get_fuse_inode(inode)->nodeid;
662 generation = inode->i_generation;
663
664 fh[0] = (u32)(nodeid >> 32);
665 fh[1] = (u32)(nodeid & 0xffffffff);
666 fh[2] = generation;
667
668 if (encode_parent) {
669 struct inode *parent;
670
671 spin_lock(&dentry->d_lock);
672 parent = dentry->d_parent->d_inode;
673 nodeid = get_fuse_inode(parent)->nodeid;
674 generation = parent->i_generation;
675 spin_unlock(&dentry->d_lock);
676
677 fh[3] = (u32)(nodeid >> 32);
678 fh[4] = (u32)(nodeid & 0xffffffff);
679 fh[5] = generation;
680 }
681
682 *max_len = len;
683 return encode_parent ? 0x82 : 0x81;
684 }
685
fuse_fh_to_dentry(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)686 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
687 struct fid *fid, int fh_len, int fh_type)
688 {
689 struct fuse_inode_handle handle;
690
691 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
692 return NULL;
693
694 handle.nodeid = (u64) fid->raw[0] << 32;
695 handle.nodeid |= (u64) fid->raw[1];
696 handle.generation = fid->raw[2];
697 return fuse_get_dentry(sb, &handle);
698 }
699
fuse_fh_to_parent(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)700 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
701 struct fid *fid, int fh_len, int fh_type)
702 {
703 struct fuse_inode_handle parent;
704
705 if (fh_type != 0x82 || fh_len < 6)
706 return NULL;
707
708 parent.nodeid = (u64) fid->raw[3] << 32;
709 parent.nodeid |= (u64) fid->raw[4];
710 parent.generation = fid->raw[5];
711 return fuse_get_dentry(sb, &parent);
712 }
713
fuse_get_parent(struct dentry * child)714 static struct dentry *fuse_get_parent(struct dentry *child)
715 {
716 struct inode *child_inode = child->d_inode;
717 struct fuse_conn *fc = get_fuse_conn(child_inode);
718 struct inode *inode;
719 struct dentry *parent;
720 struct fuse_entry_out outarg;
721 struct qstr name;
722 int err;
723
724 if (!fc->export_support)
725 return ERR_PTR(-ESTALE);
726
727 name.len = 2;
728 name.name = "..";
729 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
730 &name, &outarg, &inode);
731 if (err) {
732 if (err == -ENOENT)
733 return ERR_PTR(-ESTALE);
734 return ERR_PTR(err);
735 }
736
737 parent = d_obtain_alias(inode);
738 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
739 fuse_invalidate_entry_cache(parent);
740
741 return parent;
742 }
743
744 static const struct export_operations fuse_export_operations = {
745 .fh_to_dentry = fuse_fh_to_dentry,
746 .fh_to_parent = fuse_fh_to_parent,
747 .encode_fh = fuse_encode_fh,
748 .get_parent = fuse_get_parent,
749 };
750
751 static const struct super_operations fuse_super_operations = {
752 .alloc_inode = fuse_alloc_inode,
753 .destroy_inode = fuse_destroy_inode,
754 .evict_inode = fuse_evict_inode,
755 .drop_inode = generic_delete_inode,
756 .remount_fs = fuse_remount_fs,
757 .put_super = fuse_put_super,
758 .umount_begin = fuse_umount_begin,
759 .statfs = fuse_statfs,
760 .show_options = fuse_show_options,
761 };
762
sanitize_global_limit(unsigned * limit)763 static void sanitize_global_limit(unsigned *limit)
764 {
765 if (*limit == 0)
766 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
767 sizeof(struct fuse_req);
768
769 if (*limit >= 1 << 16)
770 *limit = (1 << 16) - 1;
771 }
772
set_global_limit(const char * val,struct kernel_param * kp)773 static int set_global_limit(const char *val, struct kernel_param *kp)
774 {
775 int rv;
776
777 rv = param_set_uint(val, kp);
778 if (rv)
779 return rv;
780
781 sanitize_global_limit((unsigned *)kp->arg);
782
783 return 0;
784 }
785
process_init_limits(struct fuse_conn * fc,struct fuse_init_out * arg)786 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
787 {
788 int cap_sys_admin = capable(CAP_SYS_ADMIN);
789
790 if (arg->minor < 13)
791 return;
792
793 sanitize_global_limit(&max_user_bgreq);
794 sanitize_global_limit(&max_user_congthresh);
795
796 if (arg->max_background) {
797 fc->max_background = arg->max_background;
798
799 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
800 fc->max_background = max_user_bgreq;
801 }
802 if (arg->congestion_threshold) {
803 fc->congestion_threshold = arg->congestion_threshold;
804
805 if (!cap_sys_admin &&
806 fc->congestion_threshold > max_user_congthresh)
807 fc->congestion_threshold = max_user_congthresh;
808 }
809 }
810
process_init_reply(struct fuse_conn * fc,struct fuse_req * req)811 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
812 {
813 struct fuse_init_out *arg = &req->misc.init_out;
814
815 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
816 fc->conn_error = 1;
817 else {
818 unsigned long ra_pages;
819
820 process_init_limits(fc, arg);
821
822 if (arg->minor >= 6) {
823 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
824 if (arg->flags & FUSE_ASYNC_READ)
825 fc->async_read = 1;
826 if (!(arg->flags & FUSE_POSIX_LOCKS))
827 fc->no_lock = 1;
828 if (arg->minor >= 17) {
829 if (!(arg->flags & FUSE_FLOCK_LOCKS))
830 fc->no_flock = 1;
831 } else {
832 if (!(arg->flags & FUSE_POSIX_LOCKS))
833 fc->no_flock = 1;
834 }
835 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
836 fc->atomic_o_trunc = 1;
837 if (arg->minor >= 9) {
838 /* LOOKUP has dependency on proto version */
839 if (arg->flags & FUSE_EXPORT_SUPPORT)
840 fc->export_support = 1;
841 }
842 if (arg->flags & FUSE_BIG_WRITES)
843 fc->big_writes = 1;
844 if (arg->flags & FUSE_DONT_MASK)
845 fc->dont_mask = 1;
846 } else {
847 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
848 fc->no_lock = 1;
849 fc->no_flock = 1;
850 }
851
852 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
853 fc->minor = arg->minor;
854 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
855 fc->max_write = max_t(unsigned, 4096, fc->max_write);
856 fc->conn_init = 1;
857 }
858 fc->blocked = 0;
859 wake_up_all(&fc->blocked_waitq);
860 }
861
fuse_send_init(struct fuse_conn * fc,struct fuse_req * req)862 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
863 {
864 struct fuse_init_in *arg = &req->misc.init_in;
865
866 arg->major = FUSE_KERNEL_VERSION;
867 arg->minor = FUSE_KERNEL_MINOR_VERSION;
868 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
869 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
870 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
871 FUSE_FLOCK_LOCKS;
872 req->in.h.opcode = FUSE_INIT;
873 req->in.numargs = 1;
874 req->in.args[0].size = sizeof(*arg);
875 req->in.args[0].value = arg;
876 req->out.numargs = 1;
877 /* Variable length argument used for backward compatibility
878 with interface version < 7.5. Rest of init_out is zeroed
879 by do_get_request(), so a short reply is not a problem */
880 req->out.argvar = 1;
881 req->out.args[0].size = sizeof(struct fuse_init_out);
882 req->out.args[0].value = &req->misc.init_out;
883 req->end = process_init_reply;
884 fuse_request_send_background(fc, req);
885 }
886
fuse_free_conn(struct fuse_conn * fc)887 static void fuse_free_conn(struct fuse_conn *fc)
888 {
889 kfree(fc);
890 }
891
fuse_bdi_init(struct fuse_conn * fc,struct super_block * sb)892 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
893 {
894 int err;
895
896 fc->bdi.name = "fuse";
897 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
898 /* fuse does it's own writeback accounting */
899 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
900
901 err = bdi_init(&fc->bdi);
902 if (err)
903 return err;
904
905 fc->bdi_initialized = 1;
906
907 if (sb->s_bdev) {
908 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
909 MAJOR(fc->dev), MINOR(fc->dev));
910 } else {
911 err = bdi_register_dev(&fc->bdi, fc->dev);
912 }
913
914 if (err)
915 return err;
916
917 /*
918 * For a single fuse filesystem use max 1% of dirty +
919 * writeback threshold.
920 *
921 * This gives about 1M of write buffer for memory maps on a
922 * machine with 1G and 10% dirty_ratio, which should be more
923 * than enough.
924 *
925 * Privileged users can raise it by writing to
926 *
927 * /sys/class/bdi/<bdi>/max_ratio
928 */
929 bdi_set_max_ratio(&fc->bdi, 1);
930
931 return 0;
932 }
933
fuse_fill_super(struct super_block * sb,void * data,int silent)934 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
935 {
936 struct fuse_conn *fc;
937 struct inode *root;
938 struct fuse_mount_data d;
939 struct file *file;
940 struct dentry *root_dentry;
941 struct fuse_req *init_req;
942 int err;
943 int is_bdev = sb->s_bdev != NULL;
944
945 err = -EINVAL;
946 if (sb->s_flags & MS_MANDLOCK)
947 goto err;
948
949 sb->s_flags &= ~MS_NOSEC;
950
951 if (!parse_fuse_opt((char *) data, &d, is_bdev))
952 goto err;
953
954 if (is_bdev) {
955 #ifdef CONFIG_BLOCK
956 err = -EINVAL;
957 if (!sb_set_blocksize(sb, d.blksize))
958 goto err;
959 #endif
960 } else {
961 sb->s_blocksize = PAGE_CACHE_SIZE;
962 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
963 }
964 sb->s_magic = FUSE_SUPER_MAGIC;
965 sb->s_op = &fuse_super_operations;
966 sb->s_maxbytes = MAX_LFS_FILESIZE;
967 sb->s_time_gran = 1;
968 sb->s_export_op = &fuse_export_operations;
969
970 file = fget(d.fd);
971 err = -EINVAL;
972 if (!file)
973 goto err;
974
975 if (file->f_op != &fuse_dev_operations)
976 goto err_fput;
977
978 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
979 err = -ENOMEM;
980 if (!fc)
981 goto err_fput;
982
983 fuse_conn_init(fc);
984
985 fc->dev = sb->s_dev;
986 fc->sb = sb;
987 err = fuse_bdi_init(fc, sb);
988 if (err)
989 goto err_put_conn;
990
991 sb->s_bdi = &fc->bdi;
992
993 /* Handle umasking inside the fuse code */
994 if (sb->s_flags & MS_POSIXACL)
995 fc->dont_mask = 1;
996 sb->s_flags |= MS_POSIXACL;
997
998 fc->release = fuse_free_conn;
999 fc->flags = d.flags;
1000 fc->user_id = d.user_id;
1001 fc->group_id = d.group_id;
1002 fc->max_read = max_t(unsigned, 4096, d.max_read);
1003
1004 /* Used by get_root_inode() */
1005 sb->s_fs_info = fc;
1006
1007 err = -ENOMEM;
1008 root = fuse_get_root_inode(sb, d.rootmode);
1009 root_dentry = d_make_root(root);
1010 if (!root_dentry)
1011 goto err_put_conn;
1012 /* only now - we want root dentry with NULL ->d_op */
1013 sb->s_d_op = &fuse_dentry_operations;
1014
1015 init_req = fuse_request_alloc();
1016 if (!init_req)
1017 goto err_put_root;
1018
1019 if (is_bdev) {
1020 fc->destroy_req = fuse_request_alloc();
1021 if (!fc->destroy_req)
1022 goto err_free_init_req;
1023 }
1024
1025 mutex_lock(&fuse_mutex);
1026 err = -EINVAL;
1027 if (file->private_data)
1028 goto err_unlock;
1029
1030 err = fuse_ctl_add_conn(fc);
1031 if (err)
1032 goto err_unlock;
1033
1034 list_add_tail(&fc->entry, &fuse_conn_list);
1035 sb->s_root = root_dentry;
1036 fc->connected = 1;
1037 file->private_data = fuse_conn_get(fc);
1038 mutex_unlock(&fuse_mutex);
1039 /*
1040 * atomic_dec_and_test() in fput() provides the necessary
1041 * memory barrier for file->private_data to be visible on all
1042 * CPUs after this
1043 */
1044 fput(file);
1045
1046 fuse_send_init(fc, init_req);
1047
1048 return 0;
1049
1050 err_unlock:
1051 mutex_unlock(&fuse_mutex);
1052 err_free_init_req:
1053 fuse_request_free(init_req);
1054 err_put_root:
1055 dput(root_dentry);
1056 err_put_conn:
1057 fuse_bdi_destroy(fc);
1058 fuse_conn_put(fc);
1059 err_fput:
1060 fput(file);
1061 err:
1062 return err;
1063 }
1064
fuse_mount(struct file_system_type * fs_type,int flags,const char * dev_name,void * raw_data)1065 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1066 int flags, const char *dev_name,
1067 void *raw_data)
1068 {
1069 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1070 }
1071
fuse_kill_sb_anon(struct super_block * sb)1072 static void fuse_kill_sb_anon(struct super_block *sb)
1073 {
1074 struct fuse_conn *fc = get_fuse_conn_super(sb);
1075
1076 if (fc) {
1077 down_write(&fc->killsb);
1078 fc->sb = NULL;
1079 up_write(&fc->killsb);
1080 }
1081
1082 kill_anon_super(sb);
1083 }
1084
1085 static struct file_system_type fuse_fs_type = {
1086 .owner = THIS_MODULE,
1087 .name = "fuse",
1088 .fs_flags = FS_HAS_SUBTYPE,
1089 .mount = fuse_mount,
1090 .kill_sb = fuse_kill_sb_anon,
1091 };
1092
1093 #ifdef CONFIG_BLOCK
fuse_mount_blk(struct file_system_type * fs_type,int flags,const char * dev_name,void * raw_data)1094 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1095 int flags, const char *dev_name,
1096 void *raw_data)
1097 {
1098 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1099 }
1100
fuse_kill_sb_blk(struct super_block * sb)1101 static void fuse_kill_sb_blk(struct super_block *sb)
1102 {
1103 struct fuse_conn *fc = get_fuse_conn_super(sb);
1104
1105 if (fc) {
1106 down_write(&fc->killsb);
1107 fc->sb = NULL;
1108 up_write(&fc->killsb);
1109 }
1110
1111 kill_block_super(sb);
1112 }
1113
1114 static struct file_system_type fuseblk_fs_type = {
1115 .owner = THIS_MODULE,
1116 .name = "fuseblk",
1117 .mount = fuse_mount_blk,
1118 .kill_sb = fuse_kill_sb_blk,
1119 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1120 };
1121
register_fuseblk(void)1122 static inline int register_fuseblk(void)
1123 {
1124 return register_filesystem(&fuseblk_fs_type);
1125 }
1126
unregister_fuseblk(void)1127 static inline void unregister_fuseblk(void)
1128 {
1129 unregister_filesystem(&fuseblk_fs_type);
1130 }
1131 #else
register_fuseblk(void)1132 static inline int register_fuseblk(void)
1133 {
1134 return 0;
1135 }
1136
unregister_fuseblk(void)1137 static inline void unregister_fuseblk(void)
1138 {
1139 }
1140 #endif
1141
fuse_inode_init_once(void * foo)1142 static void fuse_inode_init_once(void *foo)
1143 {
1144 struct inode *inode = foo;
1145
1146 inode_init_once(inode);
1147 }
1148
fuse_fs_init(void)1149 static int __init fuse_fs_init(void)
1150 {
1151 int err;
1152
1153 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1154 sizeof(struct fuse_inode),
1155 0, SLAB_HWCACHE_ALIGN,
1156 fuse_inode_init_once);
1157 err = -ENOMEM;
1158 if (!fuse_inode_cachep)
1159 goto out;
1160
1161 err = register_fuseblk();
1162 if (err)
1163 goto out2;
1164
1165 err = register_filesystem(&fuse_fs_type);
1166 if (err)
1167 goto out3;
1168
1169 return 0;
1170
1171 out3:
1172 unregister_fuseblk();
1173 out2:
1174 kmem_cache_destroy(fuse_inode_cachep);
1175 out:
1176 return err;
1177 }
1178
fuse_fs_cleanup(void)1179 static void fuse_fs_cleanup(void)
1180 {
1181 unregister_filesystem(&fuse_fs_type);
1182 unregister_fuseblk();
1183 kmem_cache_destroy(fuse_inode_cachep);
1184 }
1185
1186 static struct kobject *fuse_kobj;
1187 static struct kobject *connections_kobj;
1188
fuse_sysfs_init(void)1189 static int fuse_sysfs_init(void)
1190 {
1191 int err;
1192
1193 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1194 if (!fuse_kobj) {
1195 err = -ENOMEM;
1196 goto out_err;
1197 }
1198
1199 connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1200 if (!connections_kobj) {
1201 err = -ENOMEM;
1202 goto out_fuse_unregister;
1203 }
1204
1205 return 0;
1206
1207 out_fuse_unregister:
1208 kobject_put(fuse_kobj);
1209 out_err:
1210 return err;
1211 }
1212
fuse_sysfs_cleanup(void)1213 static void fuse_sysfs_cleanup(void)
1214 {
1215 kobject_put(connections_kobj);
1216 kobject_put(fuse_kobj);
1217 }
1218
fuse_init(void)1219 static int __init fuse_init(void)
1220 {
1221 int res;
1222
1223 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1224 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1225
1226 INIT_LIST_HEAD(&fuse_conn_list);
1227 res = fuse_fs_init();
1228 if (res)
1229 goto err;
1230
1231 res = fuse_dev_init();
1232 if (res)
1233 goto err_fs_cleanup;
1234
1235 res = fuse_sysfs_init();
1236 if (res)
1237 goto err_dev_cleanup;
1238
1239 res = fuse_ctl_init();
1240 if (res)
1241 goto err_sysfs_cleanup;
1242
1243 sanitize_global_limit(&max_user_bgreq);
1244 sanitize_global_limit(&max_user_congthresh);
1245
1246 return 0;
1247
1248 err_sysfs_cleanup:
1249 fuse_sysfs_cleanup();
1250 err_dev_cleanup:
1251 fuse_dev_cleanup();
1252 err_fs_cleanup:
1253 fuse_fs_cleanup();
1254 err:
1255 return res;
1256 }
1257
fuse_exit(void)1258 static void __exit fuse_exit(void)
1259 {
1260 printk(KERN_DEBUG "fuse exit\n");
1261
1262 fuse_ctl_cleanup();
1263 fuse_sysfs_cleanup();
1264 fuse_fs_cleanup();
1265 fuse_dev_cleanup();
1266 }
1267
1268 module_init(fuse_init);
1269 module_exit(fuse_exit);
1270