Loading...
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
24MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
25MODULE_DESCRIPTION("Filesystem in Userspace");
26MODULE_LICENSE("GPL");
27
28static struct kmem_cache *fuse_inode_cachep;
29struct list_head fuse_conn_list;
30DEFINE_MUTEX(fuse_mutex);
31
32static int set_global_limit(const char *val, struct kernel_param *kp);
33
34unsigned max_user_bgreq;
35module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
36 &max_user_bgreq, 0644);
37__MODULE_PARM_TYPE(max_user_bgreq, "uint");
38MODULE_PARM_DESC(max_user_bgreq,
39 "Global limit for the maximum number of backgrounded requests an "
40 "unprivileged user can set");
41
42unsigned max_user_congthresh;
43module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
44 &max_user_congthresh, 0644);
45__MODULE_PARM_TYPE(max_user_congthresh, "uint");
46MODULE_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
60struct 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
74struct fuse_forget_link *fuse_alloc_forget(void)
75{
76 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
77}
78
79static 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 INIT_LIST_HEAD(&fi->write_files);
96 INIT_LIST_HEAD(&fi->queued_writes);
97 INIT_LIST_HEAD(&fi->writepages);
98 init_waitqueue_head(&fi->page_waitq);
99 fi->forget = fuse_alloc_forget();
100 if (!fi->forget) {
101 kmem_cache_free(fuse_inode_cachep, inode);
102 return NULL;
103 }
104
105 return inode;
106}
107
108static void fuse_i_callback(struct rcu_head *head)
109{
110 struct inode *inode = container_of(head, struct inode, i_rcu);
111 kmem_cache_free(fuse_inode_cachep, inode);
112}
113
114static void fuse_destroy_inode(struct inode *inode)
115{
116 struct fuse_inode *fi = get_fuse_inode(inode);
117 BUG_ON(!list_empty(&fi->write_files));
118 BUG_ON(!list_empty(&fi->queued_writes));
119 kfree(fi->forget);
120 call_rcu(&inode->i_rcu, fuse_i_callback);
121}
122
123static void fuse_evict_inode(struct inode *inode)
124{
125 truncate_inode_pages(&inode->i_data, 0);
126 clear_inode(inode);
127 if (inode->i_sb->s_flags & MS_ACTIVE) {
128 struct fuse_conn *fc = get_fuse_conn(inode);
129 struct fuse_inode *fi = get_fuse_inode(inode);
130 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
131 fi->forget = NULL;
132 }
133}
134
135static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
136{
137 if (*flags & MS_MANDLOCK)
138 return -EINVAL;
139
140 return 0;
141}
142
143/*
144 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
145 * so that it will fit.
146 */
147static ino_t fuse_squash_ino(u64 ino64)
148{
149 ino_t ino = (ino_t) ino64;
150 if (sizeof(ino_t) < sizeof(u64))
151 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
152 return ino;
153}
154
155void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
156 u64 attr_valid)
157{
158 struct fuse_conn *fc = get_fuse_conn(inode);
159 struct fuse_inode *fi = get_fuse_inode(inode);
160
161 fi->attr_version = ++fc->attr_version;
162 fi->i_time = attr_valid;
163
164 inode->i_ino = fuse_squash_ino(attr->ino);
165 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
166 set_nlink(inode, attr->nlink);
167 inode->i_uid = attr->uid;
168 inode->i_gid = attr->gid;
169 inode->i_blocks = attr->blocks;
170 inode->i_atime.tv_sec = attr->atime;
171 inode->i_atime.tv_nsec = attr->atimensec;
172 inode->i_mtime.tv_sec = attr->mtime;
173 inode->i_mtime.tv_nsec = attr->mtimensec;
174 inode->i_ctime.tv_sec = attr->ctime;
175 inode->i_ctime.tv_nsec = attr->ctimensec;
176
177 if (attr->blksize != 0)
178 inode->i_blkbits = ilog2(attr->blksize);
179 else
180 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
181
182 /*
183 * Don't set the sticky bit in i_mode, unless we want the VFS
184 * to check permissions. This prevents failures due to the
185 * check in may_delete().
186 */
187 fi->orig_i_mode = inode->i_mode;
188 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
189 inode->i_mode &= ~S_ISVTX;
190
191 fi->orig_ino = attr->ino;
192}
193
194void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
195 u64 attr_valid, u64 attr_version)
196{
197 struct fuse_conn *fc = get_fuse_conn(inode);
198 struct fuse_inode *fi = get_fuse_inode(inode);
199 loff_t oldsize;
200
201 spin_lock(&fc->lock);
202 if (attr_version != 0 && fi->attr_version > attr_version) {
203 spin_unlock(&fc->lock);
204 return;
205 }
206
207 fuse_change_attributes_common(inode, attr, attr_valid);
208
209 oldsize = inode->i_size;
210 i_size_write(inode, attr->size);
211 spin_unlock(&fc->lock);
212
213 if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
214 truncate_pagecache(inode, oldsize, attr->size);
215 invalidate_inode_pages2(inode->i_mapping);
216 }
217}
218
219static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
220{
221 inode->i_mode = attr->mode & S_IFMT;
222 inode->i_size = attr->size;
223 if (S_ISREG(inode->i_mode)) {
224 fuse_init_common(inode);
225 fuse_init_file_inode(inode);
226 } else if (S_ISDIR(inode->i_mode))
227 fuse_init_dir(inode);
228 else if (S_ISLNK(inode->i_mode))
229 fuse_init_symlink(inode);
230 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
231 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
232 fuse_init_common(inode);
233 init_special_inode(inode, inode->i_mode,
234 new_decode_dev(attr->rdev));
235 } else
236 BUG();
237}
238
239int fuse_inode_eq(struct inode *inode, void *_nodeidp)
240{
241 u64 nodeid = *(u64 *) _nodeidp;
242 if (get_node_id(inode) == nodeid)
243 return 1;
244 else
245 return 0;
246}
247
248static int fuse_inode_set(struct inode *inode, void *_nodeidp)
249{
250 u64 nodeid = *(u64 *) _nodeidp;
251 get_fuse_inode(inode)->nodeid = nodeid;
252 return 0;
253}
254
255struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
256 int generation, struct fuse_attr *attr,
257 u64 attr_valid, u64 attr_version)
258{
259 struct inode *inode;
260 struct fuse_inode *fi;
261 struct fuse_conn *fc = get_fuse_conn_super(sb);
262
263 retry:
264 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
265 if (!inode)
266 return NULL;
267
268 if ((inode->i_state & I_NEW)) {
269 inode->i_flags |= S_NOATIME|S_NOCMTIME;
270 inode->i_generation = generation;
271 inode->i_data.backing_dev_info = &fc->bdi;
272 fuse_init_inode(inode, attr);
273 unlock_new_inode(inode);
274 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
275 /* Inode has changed type, any I/O on the old should fail */
276 make_bad_inode(inode);
277 iput(inode);
278 goto retry;
279 }
280
281 fi = get_fuse_inode(inode);
282 spin_lock(&fc->lock);
283 fi->nlookup++;
284 spin_unlock(&fc->lock);
285 fuse_change_attributes(inode, attr, attr_valid, attr_version);
286
287 return inode;
288}
289
290int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
291 loff_t offset, loff_t len)
292{
293 struct inode *inode;
294 pgoff_t pg_start;
295 pgoff_t pg_end;
296
297 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
298 if (!inode)
299 return -ENOENT;
300
301 fuse_invalidate_attr(inode);
302 if (offset >= 0) {
303 pg_start = offset >> PAGE_CACHE_SHIFT;
304 if (len <= 0)
305 pg_end = -1;
306 else
307 pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
308 invalidate_inode_pages2_range(inode->i_mapping,
309 pg_start, pg_end);
310 }
311 iput(inode);
312 return 0;
313}
314
315static void fuse_umount_begin(struct super_block *sb)
316{
317 fuse_abort_conn(get_fuse_conn_super(sb));
318}
319
320static void fuse_send_destroy(struct fuse_conn *fc)
321{
322 struct fuse_req *req = fc->destroy_req;
323 if (req && fc->conn_init) {
324 fc->destroy_req = NULL;
325 req->in.h.opcode = FUSE_DESTROY;
326 req->force = 1;
327 fuse_request_send(fc, req);
328 fuse_put_request(fc, req);
329 }
330}
331
332static void fuse_bdi_destroy(struct fuse_conn *fc)
333{
334 if (fc->bdi_initialized)
335 bdi_destroy(&fc->bdi);
336}
337
338void fuse_conn_kill(struct fuse_conn *fc)
339{
340 spin_lock(&fc->lock);
341 fc->connected = 0;
342 fc->blocked = 0;
343 spin_unlock(&fc->lock);
344 /* Flush all readers on this fs */
345 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
346 wake_up_all(&fc->waitq);
347 wake_up_all(&fc->blocked_waitq);
348 wake_up_all(&fc->reserved_req_waitq);
349 mutex_lock(&fuse_mutex);
350 list_del(&fc->entry);
351 fuse_ctl_remove_conn(fc);
352 mutex_unlock(&fuse_mutex);
353 fuse_bdi_destroy(fc);
354}
355EXPORT_SYMBOL_GPL(fuse_conn_kill);
356
357static void fuse_put_super(struct super_block *sb)
358{
359 struct fuse_conn *fc = get_fuse_conn_super(sb);
360
361 fuse_send_destroy(fc);
362 fuse_conn_kill(fc);
363 fuse_conn_put(fc);
364}
365
366static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
367{
368 stbuf->f_type = FUSE_SUPER_MAGIC;
369 stbuf->f_bsize = attr->bsize;
370 stbuf->f_frsize = attr->frsize;
371 stbuf->f_blocks = attr->blocks;
372 stbuf->f_bfree = attr->bfree;
373 stbuf->f_bavail = attr->bavail;
374 stbuf->f_files = attr->files;
375 stbuf->f_ffree = attr->ffree;
376 stbuf->f_namelen = attr->namelen;
377 /* fsid is left zero */
378}
379
380static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
381{
382 struct super_block *sb = dentry->d_sb;
383 struct fuse_conn *fc = get_fuse_conn_super(sb);
384 struct fuse_req *req;
385 struct fuse_statfs_out outarg;
386 int err;
387
388 if (!fuse_allow_task(fc, current)) {
389 buf->f_type = FUSE_SUPER_MAGIC;
390 return 0;
391 }
392
393 req = fuse_get_req(fc);
394 if (IS_ERR(req))
395 return PTR_ERR(req);
396
397 memset(&outarg, 0, sizeof(outarg));
398 req->in.numargs = 0;
399 req->in.h.opcode = FUSE_STATFS;
400 req->in.h.nodeid = get_node_id(dentry->d_inode);
401 req->out.numargs = 1;
402 req->out.args[0].size =
403 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
404 req->out.args[0].value = &outarg;
405 fuse_request_send(fc, req);
406 err = req->out.h.error;
407 if (!err)
408 convert_fuse_statfs(buf, &outarg.st);
409 fuse_put_request(fc, req);
410 return err;
411}
412
413enum {
414 OPT_FD,
415 OPT_ROOTMODE,
416 OPT_USER_ID,
417 OPT_GROUP_ID,
418 OPT_DEFAULT_PERMISSIONS,
419 OPT_ALLOW_OTHER,
420 OPT_MAX_READ,
421 OPT_BLKSIZE,
422 OPT_ERR
423};
424
425static const match_table_t tokens = {
426 {OPT_FD, "fd=%u"},
427 {OPT_ROOTMODE, "rootmode=%o"},
428 {OPT_USER_ID, "user_id=%u"},
429 {OPT_GROUP_ID, "group_id=%u"},
430 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
431 {OPT_ALLOW_OTHER, "allow_other"},
432 {OPT_MAX_READ, "max_read=%u"},
433 {OPT_BLKSIZE, "blksize=%u"},
434 {OPT_ERR, NULL}
435};
436
437static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
438{
439 char *p;
440 memset(d, 0, sizeof(struct fuse_mount_data));
441 d->max_read = ~0;
442 d->blksize = FUSE_DEFAULT_BLKSIZE;
443
444 while ((p = strsep(&opt, ",")) != NULL) {
445 int token;
446 int value;
447 substring_t args[MAX_OPT_ARGS];
448 if (!*p)
449 continue;
450
451 token = match_token(p, tokens, args);
452 switch (token) {
453 case OPT_FD:
454 if (match_int(&args[0], &value))
455 return 0;
456 d->fd = value;
457 d->fd_present = 1;
458 break;
459
460 case OPT_ROOTMODE:
461 if (match_octal(&args[0], &value))
462 return 0;
463 if (!fuse_valid_type(value))
464 return 0;
465 d->rootmode = value;
466 d->rootmode_present = 1;
467 break;
468
469 case OPT_USER_ID:
470 if (match_int(&args[0], &value))
471 return 0;
472 d->user_id = value;
473 d->user_id_present = 1;
474 break;
475
476 case OPT_GROUP_ID:
477 if (match_int(&args[0], &value))
478 return 0;
479 d->group_id = value;
480 d->group_id_present = 1;
481 break;
482
483 case OPT_DEFAULT_PERMISSIONS:
484 d->flags |= FUSE_DEFAULT_PERMISSIONS;
485 break;
486
487 case OPT_ALLOW_OTHER:
488 d->flags |= FUSE_ALLOW_OTHER;
489 break;
490
491 case OPT_MAX_READ:
492 if (match_int(&args[0], &value))
493 return 0;
494 d->max_read = value;
495 break;
496
497 case OPT_BLKSIZE:
498 if (!is_bdev || match_int(&args[0], &value))
499 return 0;
500 d->blksize = value;
501 break;
502
503 default:
504 return 0;
505 }
506 }
507
508 if (!d->fd_present || !d->rootmode_present ||
509 !d->user_id_present || !d->group_id_present)
510 return 0;
511
512 return 1;
513}
514
515static int fuse_show_options(struct seq_file *m, struct dentry *root)
516{
517 struct super_block *sb = root->d_sb;
518 struct fuse_conn *fc = get_fuse_conn_super(sb);
519
520 seq_printf(m, ",user_id=%u", fc->user_id);
521 seq_printf(m, ",group_id=%u", fc->group_id);
522 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
523 seq_puts(m, ",default_permissions");
524 if (fc->flags & FUSE_ALLOW_OTHER)
525 seq_puts(m, ",allow_other");
526 if (fc->max_read != ~0)
527 seq_printf(m, ",max_read=%u", fc->max_read);
528 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
529 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
530 return 0;
531}
532
533void fuse_conn_init(struct fuse_conn *fc)
534{
535 memset(fc, 0, sizeof(*fc));
536 spin_lock_init(&fc->lock);
537 mutex_init(&fc->inst_mutex);
538 init_rwsem(&fc->killsb);
539 atomic_set(&fc->count, 1);
540 init_waitqueue_head(&fc->waitq);
541 init_waitqueue_head(&fc->blocked_waitq);
542 init_waitqueue_head(&fc->reserved_req_waitq);
543 INIT_LIST_HEAD(&fc->pending);
544 INIT_LIST_HEAD(&fc->processing);
545 INIT_LIST_HEAD(&fc->io);
546 INIT_LIST_HEAD(&fc->interrupts);
547 INIT_LIST_HEAD(&fc->bg_queue);
548 INIT_LIST_HEAD(&fc->entry);
549 fc->forget_list_tail = &fc->forget_list_head;
550 atomic_set(&fc->num_waiting, 0);
551 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
552 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
553 fc->khctr = 0;
554 fc->polled_files = RB_ROOT;
555 fc->reqctr = 0;
556 fc->blocked = 1;
557 fc->attr_version = 1;
558 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
559}
560EXPORT_SYMBOL_GPL(fuse_conn_init);
561
562void fuse_conn_put(struct fuse_conn *fc)
563{
564 if (atomic_dec_and_test(&fc->count)) {
565 if (fc->destroy_req)
566 fuse_request_free(fc->destroy_req);
567 mutex_destroy(&fc->inst_mutex);
568 fc->release(fc);
569 }
570}
571EXPORT_SYMBOL_GPL(fuse_conn_put);
572
573struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
574{
575 atomic_inc(&fc->count);
576 return fc;
577}
578EXPORT_SYMBOL_GPL(fuse_conn_get);
579
580static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
581{
582 struct fuse_attr attr;
583 memset(&attr, 0, sizeof(attr));
584
585 attr.mode = mode;
586 attr.ino = FUSE_ROOT_ID;
587 attr.nlink = 1;
588 return fuse_iget(sb, 1, 0, &attr, 0, 0);
589}
590
591struct fuse_inode_handle {
592 u64 nodeid;
593 u32 generation;
594};
595
596static struct dentry *fuse_get_dentry(struct super_block *sb,
597 struct fuse_inode_handle *handle)
598{
599 struct fuse_conn *fc = get_fuse_conn_super(sb);
600 struct inode *inode;
601 struct dentry *entry;
602 int err = -ESTALE;
603
604 if (handle->nodeid == 0)
605 goto out_err;
606
607 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
608 if (!inode) {
609 struct fuse_entry_out outarg;
610 struct qstr name;
611
612 if (!fc->export_support)
613 goto out_err;
614
615 name.len = 1;
616 name.name = ".";
617 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
618 &inode);
619 if (err && err != -ENOENT)
620 goto out_err;
621 if (err || !inode) {
622 err = -ESTALE;
623 goto out_err;
624 }
625 err = -EIO;
626 if (get_node_id(inode) != handle->nodeid)
627 goto out_iput;
628 }
629 err = -ESTALE;
630 if (inode->i_generation != handle->generation)
631 goto out_iput;
632
633 entry = d_obtain_alias(inode);
634 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
635 fuse_invalidate_entry_cache(entry);
636
637 return entry;
638
639 out_iput:
640 iput(inode);
641 out_err:
642 return ERR_PTR(err);
643}
644
645static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
646 struct inode *parent)
647{
648 int len = parent ? 6 : 3;
649 u64 nodeid;
650 u32 generation;
651
652 if (*max_len < len) {
653 *max_len = len;
654 return 255;
655 }
656
657 nodeid = get_fuse_inode(inode)->nodeid;
658 generation = inode->i_generation;
659
660 fh[0] = (u32)(nodeid >> 32);
661 fh[1] = (u32)(nodeid & 0xffffffff);
662 fh[2] = generation;
663
664 if (parent) {
665 nodeid = get_fuse_inode(parent)->nodeid;
666 generation = parent->i_generation;
667
668 fh[3] = (u32)(nodeid >> 32);
669 fh[4] = (u32)(nodeid & 0xffffffff);
670 fh[5] = generation;
671 }
672
673 *max_len = len;
674 return parent ? 0x82 : 0x81;
675}
676
677static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
678 struct fid *fid, int fh_len, int fh_type)
679{
680 struct fuse_inode_handle handle;
681
682 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
683 return NULL;
684
685 handle.nodeid = (u64) fid->raw[0] << 32;
686 handle.nodeid |= (u64) fid->raw[1];
687 handle.generation = fid->raw[2];
688 return fuse_get_dentry(sb, &handle);
689}
690
691static struct dentry *fuse_fh_to_parent(struct super_block *sb,
692 struct fid *fid, int fh_len, int fh_type)
693{
694 struct fuse_inode_handle parent;
695
696 if (fh_type != 0x82 || fh_len < 6)
697 return NULL;
698
699 parent.nodeid = (u64) fid->raw[3] << 32;
700 parent.nodeid |= (u64) fid->raw[4];
701 parent.generation = fid->raw[5];
702 return fuse_get_dentry(sb, &parent);
703}
704
705static struct dentry *fuse_get_parent(struct dentry *child)
706{
707 struct inode *child_inode = child->d_inode;
708 struct fuse_conn *fc = get_fuse_conn(child_inode);
709 struct inode *inode;
710 struct dentry *parent;
711 struct fuse_entry_out outarg;
712 struct qstr name;
713 int err;
714
715 if (!fc->export_support)
716 return ERR_PTR(-ESTALE);
717
718 name.len = 2;
719 name.name = "..";
720 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
721 &name, &outarg, &inode);
722 if (err) {
723 if (err == -ENOENT)
724 return ERR_PTR(-ESTALE);
725 return ERR_PTR(err);
726 }
727
728 parent = d_obtain_alias(inode);
729 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
730 fuse_invalidate_entry_cache(parent);
731
732 return parent;
733}
734
735static const struct export_operations fuse_export_operations = {
736 .fh_to_dentry = fuse_fh_to_dentry,
737 .fh_to_parent = fuse_fh_to_parent,
738 .encode_fh = fuse_encode_fh,
739 .get_parent = fuse_get_parent,
740};
741
742static const struct super_operations fuse_super_operations = {
743 .alloc_inode = fuse_alloc_inode,
744 .destroy_inode = fuse_destroy_inode,
745 .evict_inode = fuse_evict_inode,
746 .drop_inode = generic_delete_inode,
747 .remount_fs = fuse_remount_fs,
748 .put_super = fuse_put_super,
749 .umount_begin = fuse_umount_begin,
750 .statfs = fuse_statfs,
751 .show_options = fuse_show_options,
752};
753
754static void sanitize_global_limit(unsigned *limit)
755{
756 if (*limit == 0)
757 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
758 sizeof(struct fuse_req);
759
760 if (*limit >= 1 << 16)
761 *limit = (1 << 16) - 1;
762}
763
764static int set_global_limit(const char *val, struct kernel_param *kp)
765{
766 int rv;
767
768 rv = param_set_uint(val, kp);
769 if (rv)
770 return rv;
771
772 sanitize_global_limit((unsigned *)kp->arg);
773
774 return 0;
775}
776
777static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
778{
779 int cap_sys_admin = capable(CAP_SYS_ADMIN);
780
781 if (arg->minor < 13)
782 return;
783
784 sanitize_global_limit(&max_user_bgreq);
785 sanitize_global_limit(&max_user_congthresh);
786
787 if (arg->max_background) {
788 fc->max_background = arg->max_background;
789
790 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
791 fc->max_background = max_user_bgreq;
792 }
793 if (arg->congestion_threshold) {
794 fc->congestion_threshold = arg->congestion_threshold;
795
796 if (!cap_sys_admin &&
797 fc->congestion_threshold > max_user_congthresh)
798 fc->congestion_threshold = max_user_congthresh;
799 }
800}
801
802static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
803{
804 struct fuse_init_out *arg = &req->misc.init_out;
805
806 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
807 fc->conn_error = 1;
808 else {
809 unsigned long ra_pages;
810
811 process_init_limits(fc, arg);
812
813 if (arg->minor >= 6) {
814 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
815 if (arg->flags & FUSE_ASYNC_READ)
816 fc->async_read = 1;
817 if (!(arg->flags & FUSE_POSIX_LOCKS))
818 fc->no_lock = 1;
819 if (arg->minor >= 17) {
820 if (!(arg->flags & FUSE_FLOCK_LOCKS))
821 fc->no_flock = 1;
822 } else {
823 if (!(arg->flags & FUSE_POSIX_LOCKS))
824 fc->no_flock = 1;
825 }
826 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
827 fc->atomic_o_trunc = 1;
828 if (arg->minor >= 9) {
829 /* LOOKUP has dependency on proto version */
830 if (arg->flags & FUSE_EXPORT_SUPPORT)
831 fc->export_support = 1;
832 }
833 if (arg->flags & FUSE_BIG_WRITES)
834 fc->big_writes = 1;
835 if (arg->flags & FUSE_DONT_MASK)
836 fc->dont_mask = 1;
837 } else {
838 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
839 fc->no_lock = 1;
840 fc->no_flock = 1;
841 }
842
843 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
844 fc->minor = arg->minor;
845 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
846 fc->max_write = max_t(unsigned, 4096, fc->max_write);
847 fc->conn_init = 1;
848 }
849 fc->blocked = 0;
850 wake_up_all(&fc->blocked_waitq);
851}
852
853static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
854{
855 struct fuse_init_in *arg = &req->misc.init_in;
856
857 arg->major = FUSE_KERNEL_VERSION;
858 arg->minor = FUSE_KERNEL_MINOR_VERSION;
859 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
860 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
861 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
862 FUSE_FLOCK_LOCKS;
863 req->in.h.opcode = FUSE_INIT;
864 req->in.numargs = 1;
865 req->in.args[0].size = sizeof(*arg);
866 req->in.args[0].value = arg;
867 req->out.numargs = 1;
868 /* Variable length argument used for backward compatibility
869 with interface version < 7.5. Rest of init_out is zeroed
870 by do_get_request(), so a short reply is not a problem */
871 req->out.argvar = 1;
872 req->out.args[0].size = sizeof(struct fuse_init_out);
873 req->out.args[0].value = &req->misc.init_out;
874 req->end = process_init_reply;
875 fuse_request_send_background(fc, req);
876}
877
878static void fuse_free_conn(struct fuse_conn *fc)
879{
880 kfree(fc);
881}
882
883static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
884{
885 int err;
886
887 fc->bdi.name = "fuse";
888 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
889 /* fuse does it's own writeback accounting */
890 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
891
892 err = bdi_init(&fc->bdi);
893 if (err)
894 return err;
895
896 fc->bdi_initialized = 1;
897
898 if (sb->s_bdev) {
899 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
900 MAJOR(fc->dev), MINOR(fc->dev));
901 } else {
902 err = bdi_register_dev(&fc->bdi, fc->dev);
903 }
904
905 if (err)
906 return err;
907
908 /*
909 * For a single fuse filesystem use max 1% of dirty +
910 * writeback threshold.
911 *
912 * This gives about 1M of write buffer for memory maps on a
913 * machine with 1G and 10% dirty_ratio, which should be more
914 * than enough.
915 *
916 * Privileged users can raise it by writing to
917 *
918 * /sys/class/bdi/<bdi>/max_ratio
919 */
920 bdi_set_max_ratio(&fc->bdi, 1);
921
922 return 0;
923}
924
925static int fuse_fill_super(struct super_block *sb, void *data, int silent)
926{
927 struct fuse_conn *fc;
928 struct inode *root;
929 struct fuse_mount_data d;
930 struct file *file;
931 struct dentry *root_dentry;
932 struct fuse_req *init_req;
933 int err;
934 int is_bdev = sb->s_bdev != NULL;
935
936 err = -EINVAL;
937 if (sb->s_flags & MS_MANDLOCK)
938 goto err;
939
940 sb->s_flags &= ~MS_NOSEC;
941
942 if (!parse_fuse_opt((char *) data, &d, is_bdev))
943 goto err;
944
945 if (is_bdev) {
946#ifdef CONFIG_BLOCK
947 err = -EINVAL;
948 if (!sb_set_blocksize(sb, d.blksize))
949 goto err;
950#endif
951 } else {
952 sb->s_blocksize = PAGE_CACHE_SIZE;
953 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
954 }
955 sb->s_magic = FUSE_SUPER_MAGIC;
956 sb->s_op = &fuse_super_operations;
957 sb->s_maxbytes = MAX_LFS_FILESIZE;
958 sb->s_time_gran = 1;
959 sb->s_export_op = &fuse_export_operations;
960
961 file = fget(d.fd);
962 err = -EINVAL;
963 if (!file)
964 goto err;
965
966 if (file->f_op != &fuse_dev_operations)
967 goto err_fput;
968
969 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
970 err = -ENOMEM;
971 if (!fc)
972 goto err_fput;
973
974 fuse_conn_init(fc);
975
976 fc->dev = sb->s_dev;
977 fc->sb = sb;
978 err = fuse_bdi_init(fc, sb);
979 if (err)
980 goto err_put_conn;
981
982 sb->s_bdi = &fc->bdi;
983
984 /* Handle umasking inside the fuse code */
985 if (sb->s_flags & MS_POSIXACL)
986 fc->dont_mask = 1;
987 sb->s_flags |= MS_POSIXACL;
988
989 fc->release = fuse_free_conn;
990 fc->flags = d.flags;
991 fc->user_id = d.user_id;
992 fc->group_id = d.group_id;
993 fc->max_read = max_t(unsigned, 4096, d.max_read);
994
995 /* Used by get_root_inode() */
996 sb->s_fs_info = fc;
997
998 err = -ENOMEM;
999 root = fuse_get_root_inode(sb, d.rootmode);
1000 root_dentry = d_make_root(root);
1001 if (!root_dentry)
1002 goto err_put_conn;
1003 /* only now - we want root dentry with NULL ->d_op */
1004 sb->s_d_op = &fuse_dentry_operations;
1005
1006 init_req = fuse_request_alloc();
1007 if (!init_req)
1008 goto err_put_root;
1009
1010 if (is_bdev) {
1011 fc->destroy_req = fuse_request_alloc();
1012 if (!fc->destroy_req)
1013 goto err_free_init_req;
1014 }
1015
1016 mutex_lock(&fuse_mutex);
1017 err = -EINVAL;
1018 if (file->private_data)
1019 goto err_unlock;
1020
1021 err = fuse_ctl_add_conn(fc);
1022 if (err)
1023 goto err_unlock;
1024
1025 list_add_tail(&fc->entry, &fuse_conn_list);
1026 sb->s_root = root_dentry;
1027 fc->connected = 1;
1028 file->private_data = fuse_conn_get(fc);
1029 mutex_unlock(&fuse_mutex);
1030 /*
1031 * atomic_dec_and_test() in fput() provides the necessary
1032 * memory barrier for file->private_data to be visible on all
1033 * CPUs after this
1034 */
1035 fput(file);
1036
1037 fuse_send_init(fc, init_req);
1038
1039 return 0;
1040
1041 err_unlock:
1042 mutex_unlock(&fuse_mutex);
1043 err_free_init_req:
1044 fuse_request_free(init_req);
1045 err_put_root:
1046 dput(root_dentry);
1047 err_put_conn:
1048 fuse_bdi_destroy(fc);
1049 fuse_conn_put(fc);
1050 err_fput:
1051 fput(file);
1052 err:
1053 return err;
1054}
1055
1056static struct dentry *fuse_mount(struct file_system_type *fs_type,
1057 int flags, const char *dev_name,
1058 void *raw_data)
1059{
1060 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1061}
1062
1063static void fuse_kill_sb_anon(struct super_block *sb)
1064{
1065 struct fuse_conn *fc = get_fuse_conn_super(sb);
1066
1067 if (fc) {
1068 down_write(&fc->killsb);
1069 fc->sb = NULL;
1070 up_write(&fc->killsb);
1071 }
1072
1073 kill_anon_super(sb);
1074}
1075
1076static struct file_system_type fuse_fs_type = {
1077 .owner = THIS_MODULE,
1078 .name = "fuse",
1079 .fs_flags = FS_HAS_SUBTYPE,
1080 .mount = fuse_mount,
1081 .kill_sb = fuse_kill_sb_anon,
1082};
1083
1084#ifdef CONFIG_BLOCK
1085static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1086 int flags, const char *dev_name,
1087 void *raw_data)
1088{
1089 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1090}
1091
1092static void fuse_kill_sb_blk(struct super_block *sb)
1093{
1094 struct fuse_conn *fc = get_fuse_conn_super(sb);
1095
1096 if (fc) {
1097 down_write(&fc->killsb);
1098 fc->sb = NULL;
1099 up_write(&fc->killsb);
1100 }
1101
1102 kill_block_super(sb);
1103}
1104
1105static struct file_system_type fuseblk_fs_type = {
1106 .owner = THIS_MODULE,
1107 .name = "fuseblk",
1108 .mount = fuse_mount_blk,
1109 .kill_sb = fuse_kill_sb_blk,
1110 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1111};
1112
1113static inline int register_fuseblk(void)
1114{
1115 return register_filesystem(&fuseblk_fs_type);
1116}
1117
1118static inline void unregister_fuseblk(void)
1119{
1120 unregister_filesystem(&fuseblk_fs_type);
1121}
1122#else
1123static inline int register_fuseblk(void)
1124{
1125 return 0;
1126}
1127
1128static inline void unregister_fuseblk(void)
1129{
1130}
1131#endif
1132
1133static void fuse_inode_init_once(void *foo)
1134{
1135 struct inode *inode = foo;
1136
1137 inode_init_once(inode);
1138}
1139
1140static int __init fuse_fs_init(void)
1141{
1142 int err;
1143
1144 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1145 sizeof(struct fuse_inode),
1146 0, SLAB_HWCACHE_ALIGN,
1147 fuse_inode_init_once);
1148 err = -ENOMEM;
1149 if (!fuse_inode_cachep)
1150 goto out;
1151
1152 err = register_fuseblk();
1153 if (err)
1154 goto out2;
1155
1156 err = register_filesystem(&fuse_fs_type);
1157 if (err)
1158 goto out3;
1159
1160 return 0;
1161
1162 out3:
1163 unregister_fuseblk();
1164 out2:
1165 kmem_cache_destroy(fuse_inode_cachep);
1166 out:
1167 return err;
1168}
1169
1170static void fuse_fs_cleanup(void)
1171{
1172 unregister_filesystem(&fuse_fs_type);
1173 unregister_fuseblk();
1174 kmem_cache_destroy(fuse_inode_cachep);
1175}
1176
1177static struct kobject *fuse_kobj;
1178static struct kobject *connections_kobj;
1179
1180static int fuse_sysfs_init(void)
1181{
1182 int err;
1183
1184 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1185 if (!fuse_kobj) {
1186 err = -ENOMEM;
1187 goto out_err;
1188 }
1189
1190 connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1191 if (!connections_kobj) {
1192 err = -ENOMEM;
1193 goto out_fuse_unregister;
1194 }
1195
1196 return 0;
1197
1198 out_fuse_unregister:
1199 kobject_put(fuse_kobj);
1200 out_err:
1201 return err;
1202}
1203
1204static void fuse_sysfs_cleanup(void)
1205{
1206 kobject_put(connections_kobj);
1207 kobject_put(fuse_kobj);
1208}
1209
1210static int __init fuse_init(void)
1211{
1212 int res;
1213
1214 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1215 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1216
1217 INIT_LIST_HEAD(&fuse_conn_list);
1218 res = fuse_fs_init();
1219 if (res)
1220 goto err;
1221
1222 res = fuse_dev_init();
1223 if (res)
1224 goto err_fs_cleanup;
1225
1226 res = fuse_sysfs_init();
1227 if (res)
1228 goto err_dev_cleanup;
1229
1230 res = fuse_ctl_init();
1231 if (res)
1232 goto err_sysfs_cleanup;
1233
1234 sanitize_global_limit(&max_user_bgreq);
1235 sanitize_global_limit(&max_user_congthresh);
1236
1237 return 0;
1238
1239 err_sysfs_cleanup:
1240 fuse_sysfs_cleanup();
1241 err_dev_cleanup:
1242 fuse_dev_cleanup();
1243 err_fs_cleanup:
1244 fuse_fs_cleanup();
1245 err:
1246 return res;
1247}
1248
1249static void __exit fuse_exit(void)
1250{
1251 printk(KERN_DEBUG "fuse exit\n");
1252
1253 fuse_ctl_cleanup();
1254 fuse_sysfs_cleanup();
1255 fuse_fs_cleanup();
1256 fuse_dev_cleanup();
1257}
1258
1259module_init(fuse_init);
1260module_exit(fuse_exit);
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#include <linux/posix_acl.h>
24
25MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
26MODULE_DESCRIPTION("Filesystem in Userspace");
27MODULE_LICENSE("GPL");
28
29static struct kmem_cache *fuse_inode_cachep;
30struct list_head fuse_conn_list;
31DEFINE_MUTEX(fuse_mutex);
32
33static int set_global_limit(const char *val, struct kernel_param *kp);
34
35unsigned max_user_bgreq;
36module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
37 &max_user_bgreq, 0644);
38__MODULE_PARM_TYPE(max_user_bgreq, "uint");
39MODULE_PARM_DESC(max_user_bgreq,
40 "Global limit for the maximum number of backgrounded requests an "
41 "unprivileged user can set");
42
43unsigned max_user_congthresh;
44module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
45 &max_user_congthresh, 0644);
46__MODULE_PARM_TYPE(max_user_congthresh, "uint");
47MODULE_PARM_DESC(max_user_congthresh,
48 "Global limit for the maximum congestion threshold an "
49 "unprivileged user can set");
50
51#define FUSE_SUPER_MAGIC 0x65735546
52
53#define FUSE_DEFAULT_BLKSIZE 512
54
55/** Maximum number of outstanding background requests */
56#define FUSE_DEFAULT_MAX_BACKGROUND 12
57
58/** Congestion starts at 75% of maximum */
59#define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
60
61struct fuse_mount_data {
62 int fd;
63 unsigned rootmode;
64 kuid_t user_id;
65 kgid_t group_id;
66 unsigned fd_present:1;
67 unsigned rootmode_present:1;
68 unsigned user_id_present:1;
69 unsigned group_id_present:1;
70 unsigned default_permissions:1;
71 unsigned allow_other:1;
72 unsigned max_read;
73 unsigned blksize;
74};
75
76struct fuse_forget_link *fuse_alloc_forget(void)
77{
78 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
79}
80
81static struct inode *fuse_alloc_inode(struct super_block *sb)
82{
83 struct inode *inode;
84 struct fuse_inode *fi;
85
86 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
87 if (!inode)
88 return NULL;
89
90 fi = get_fuse_inode(inode);
91 fi->i_time = 0;
92 fi->nodeid = 0;
93 fi->nlookup = 0;
94 fi->attr_version = 0;
95 fi->writectr = 0;
96 fi->orig_ino = 0;
97 fi->state = 0;
98 INIT_LIST_HEAD(&fi->write_files);
99 INIT_LIST_HEAD(&fi->queued_writes);
100 INIT_LIST_HEAD(&fi->writepages);
101 init_waitqueue_head(&fi->page_waitq);
102 mutex_init(&fi->mutex);
103 fi->forget = fuse_alloc_forget();
104 if (!fi->forget) {
105 kmem_cache_free(fuse_inode_cachep, inode);
106 return NULL;
107 }
108
109 return inode;
110}
111
112static void fuse_i_callback(struct rcu_head *head)
113{
114 struct inode *inode = container_of(head, struct inode, i_rcu);
115 kmem_cache_free(fuse_inode_cachep, inode);
116}
117
118static void fuse_destroy_inode(struct inode *inode)
119{
120 struct fuse_inode *fi = get_fuse_inode(inode);
121 BUG_ON(!list_empty(&fi->write_files));
122 BUG_ON(!list_empty(&fi->queued_writes));
123 mutex_destroy(&fi->mutex);
124 kfree(fi->forget);
125 call_rcu(&inode->i_rcu, fuse_i_callback);
126}
127
128static void fuse_evict_inode(struct inode *inode)
129{
130 truncate_inode_pages_final(&inode->i_data);
131 clear_inode(inode);
132 if (inode->i_sb->s_flags & MS_ACTIVE) {
133 struct fuse_conn *fc = get_fuse_conn(inode);
134 struct fuse_inode *fi = get_fuse_inode(inode);
135 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
136 fi->forget = NULL;
137 }
138}
139
140static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
141{
142 sync_filesystem(sb);
143 if (*flags & MS_MANDLOCK)
144 return -EINVAL;
145
146 return 0;
147}
148
149/*
150 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
151 * so that it will fit.
152 */
153static ino_t fuse_squash_ino(u64 ino64)
154{
155 ino_t ino = (ino_t) ino64;
156 if (sizeof(ino_t) < sizeof(u64))
157 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
158 return ino;
159}
160
161void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
162 u64 attr_valid)
163{
164 struct fuse_conn *fc = get_fuse_conn(inode);
165 struct fuse_inode *fi = get_fuse_inode(inode);
166
167 fi->attr_version = ++fc->attr_version;
168 fi->i_time = attr_valid;
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(&init_user_ns, attr->uid);
174 inode->i_gid = make_kgid(&init_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 /* mtime from server may be stale due to local buffered write */
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 * Don't set the sticky bit in i_mode, unless we want the VFS
193 * to check permissions. This prevents failures due to the
194 * check in may_delete().
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 timespec 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 * In case of writeback_cache enabled, the cached writes beyond EOF
225 * extend local i_size without keeping userspace server in sync. So,
226 * attr->size coming from server can be stale. We cannot trust it.
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 timespec new_mtime = {
240 .tv_sec = attr->mtime,
241 .tv_nsec = attr->mtimensec,
242 };
243
244 /*
245 * Auto inval mode also checks and invalidates if mtime
246 * has changed.
247 */
248 if (!timespec_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 /* Inode has changed type, any I/O on the old should fail */
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
359void fuse_lock_inode(struct inode *inode)
360{
361 if (!get_fuse_conn(inode)->parallel_dirops)
362 mutex_lock(&get_fuse_inode(inode)->mutex);
363}
364
365void fuse_unlock_inode(struct inode *inode)
366{
367 if (!get_fuse_conn(inode)->parallel_dirops)
368 mutex_unlock(&get_fuse_inode(inode)->mutex);
369}
370
371static void fuse_umount_begin(struct super_block *sb)
372{
373 fuse_abort_conn(get_fuse_conn_super(sb));
374}
375
376static void fuse_send_destroy(struct fuse_conn *fc)
377{
378 struct fuse_req *req = fc->destroy_req;
379 if (req && fc->conn_init) {
380 fc->destroy_req = NULL;
381 req->in.h.opcode = FUSE_DESTROY;
382 __set_bit(FR_FORCE, &req->flags);
383 __clear_bit(FR_BACKGROUND, &req->flags);
384 fuse_request_send(fc, req);
385 fuse_put_request(fc, req);
386 }
387}
388
389static void fuse_bdi_destroy(struct fuse_conn *fc)
390{
391 if (fc->bdi_initialized)
392 bdi_destroy(&fc->bdi);
393}
394
395static void fuse_put_super(struct super_block *sb)
396{
397 struct fuse_conn *fc = get_fuse_conn_super(sb);
398
399 fuse_send_destroy(fc);
400
401 fuse_abort_conn(fc);
402 mutex_lock(&fuse_mutex);
403 list_del(&fc->entry);
404 fuse_ctl_remove_conn(fc);
405 mutex_unlock(&fuse_mutex);
406 fuse_bdi_destroy(fc);
407
408 fuse_conn_put(fc);
409}
410
411static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
412{
413 stbuf->f_type = FUSE_SUPER_MAGIC;
414 stbuf->f_bsize = attr->bsize;
415 stbuf->f_frsize = attr->frsize;
416 stbuf->f_blocks = attr->blocks;
417 stbuf->f_bfree = attr->bfree;
418 stbuf->f_bavail = attr->bavail;
419 stbuf->f_files = attr->files;
420 stbuf->f_ffree = attr->ffree;
421 stbuf->f_namelen = attr->namelen;
422 /* fsid is left zero */
423}
424
425static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
426{
427 struct super_block *sb = dentry->d_sb;
428 struct fuse_conn *fc = get_fuse_conn_super(sb);
429 FUSE_ARGS(args);
430 struct fuse_statfs_out outarg;
431 int err;
432
433 if (!fuse_allow_current_process(fc)) {
434 buf->f_type = FUSE_SUPER_MAGIC;
435 return 0;
436 }
437
438 memset(&outarg, 0, sizeof(outarg));
439 args.in.numargs = 0;
440 args.in.h.opcode = FUSE_STATFS;
441 args.in.h.nodeid = get_node_id(d_inode(dentry));
442 args.out.numargs = 1;
443 args.out.args[0].size = sizeof(outarg);
444 args.out.args[0].value = &outarg;
445 err = fuse_simple_request(fc, &args);
446 if (!err)
447 convert_fuse_statfs(buf, &outarg.st);
448 return err;
449}
450
451enum {
452 OPT_FD,
453 OPT_ROOTMODE,
454 OPT_USER_ID,
455 OPT_GROUP_ID,
456 OPT_DEFAULT_PERMISSIONS,
457 OPT_ALLOW_OTHER,
458 OPT_MAX_READ,
459 OPT_BLKSIZE,
460 OPT_ERR
461};
462
463static const match_table_t tokens = {
464 {OPT_FD, "fd=%u"},
465 {OPT_ROOTMODE, "rootmode=%o"},
466 {OPT_USER_ID, "user_id=%u"},
467 {OPT_GROUP_ID, "group_id=%u"},
468 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
469 {OPT_ALLOW_OTHER, "allow_other"},
470 {OPT_MAX_READ, "max_read=%u"},
471 {OPT_BLKSIZE, "blksize=%u"},
472 {OPT_ERR, NULL}
473};
474
475static int fuse_match_uint(substring_t *s, unsigned int *res)
476{
477 int err = -ENOMEM;
478 char *buf = match_strdup(s);
479 if (buf) {
480 err = kstrtouint(buf, 10, res);
481 kfree(buf);
482 }
483 return err;
484}
485
486static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
487{
488 char *p;
489 memset(d, 0, sizeof(struct fuse_mount_data));
490 d->max_read = ~0;
491 d->blksize = FUSE_DEFAULT_BLKSIZE;
492
493 while ((p = strsep(&opt, ",")) != NULL) {
494 int token;
495 int value;
496 unsigned uv;
497 substring_t args[MAX_OPT_ARGS];
498 if (!*p)
499 continue;
500
501 token = match_token(p, tokens, args);
502 switch (token) {
503 case OPT_FD:
504 if (match_int(&args[0], &value))
505 return 0;
506 d->fd = value;
507 d->fd_present = 1;
508 break;
509
510 case OPT_ROOTMODE:
511 if (match_octal(&args[0], &value))
512 return 0;
513 if (!fuse_valid_type(value))
514 return 0;
515 d->rootmode = value;
516 d->rootmode_present = 1;
517 break;
518
519 case OPT_USER_ID:
520 if (fuse_match_uint(&args[0], &uv))
521 return 0;
522 d->user_id = make_kuid(current_user_ns(), uv);
523 if (!uid_valid(d->user_id))
524 return 0;
525 d->user_id_present = 1;
526 break;
527
528 case OPT_GROUP_ID:
529 if (fuse_match_uint(&args[0], &uv))
530 return 0;
531 d->group_id = make_kgid(current_user_ns(), uv);
532 if (!gid_valid(d->group_id))
533 return 0;
534 d->group_id_present = 1;
535 break;
536
537 case OPT_DEFAULT_PERMISSIONS:
538 d->default_permissions = 1;
539 break;
540
541 case OPT_ALLOW_OTHER:
542 d->allow_other = 1;
543 break;
544
545 case OPT_MAX_READ:
546 if (match_int(&args[0], &value))
547 return 0;
548 d->max_read = value;
549 break;
550
551 case OPT_BLKSIZE:
552 if (!is_bdev || match_int(&args[0], &value))
553 return 0;
554 d->blksize = value;
555 break;
556
557 default:
558 return 0;
559 }
560 }
561
562 if (!d->fd_present || !d->rootmode_present ||
563 !d->user_id_present || !d->group_id_present)
564 return 0;
565
566 return 1;
567}
568
569static int fuse_show_options(struct seq_file *m, struct dentry *root)
570{
571 struct super_block *sb = root->d_sb;
572 struct fuse_conn *fc = get_fuse_conn_super(sb);
573
574 seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
575 seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
576 if (fc->default_permissions)
577 seq_puts(m, ",default_permissions");
578 if (fc->allow_other)
579 seq_puts(m, ",allow_other");
580 if (fc->max_read != ~0)
581 seq_printf(m, ",max_read=%u", fc->max_read);
582 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
583 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
584 return 0;
585}
586
587static void fuse_iqueue_init(struct fuse_iqueue *fiq)
588{
589 memset(fiq, 0, sizeof(struct fuse_iqueue));
590 init_waitqueue_head(&fiq->waitq);
591 INIT_LIST_HEAD(&fiq->pending);
592 INIT_LIST_HEAD(&fiq->interrupts);
593 fiq->forget_list_tail = &fiq->forget_list_head;
594 fiq->connected = 1;
595}
596
597static void fuse_pqueue_init(struct fuse_pqueue *fpq)
598{
599 memset(fpq, 0, sizeof(struct fuse_pqueue));
600 spin_lock_init(&fpq->lock);
601 INIT_LIST_HEAD(&fpq->processing);
602 INIT_LIST_HEAD(&fpq->io);
603 fpq->connected = 1;
604}
605
606void fuse_conn_init(struct fuse_conn *fc)
607{
608 memset(fc, 0, sizeof(*fc));
609 spin_lock_init(&fc->lock);
610 init_rwsem(&fc->killsb);
611 atomic_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}
630EXPORT_SYMBOL_GPL(fuse_conn_init);
631
632void fuse_conn_put(struct fuse_conn *fc)
633{
634 if (atomic_dec_and_test(&fc->count)) {
635 if (fc->destroy_req)
636 fuse_request_free(fc->destroy_req);
637 fc->release(fc);
638 }
639}
640EXPORT_SYMBOL_GPL(fuse_conn_put);
641
642struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
643{
644 atomic_inc(&fc->count);
645 return fc;
646}
647EXPORT_SYMBOL_GPL(fuse_conn_get);
648
649static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
650{
651 struct fuse_attr attr;
652 memset(&attr, 0, sizeof(attr));
653
654 attr.mode = mode;
655 attr.ino = FUSE_ROOT_ID;
656 attr.nlink = 1;
657 return fuse_iget(sb, 1, 0, &attr, 0, 0);
658}
659
660struct fuse_inode_handle {
661 u64 nodeid;
662 u32 generation;
663};
664
665static struct dentry *fuse_get_dentry(struct super_block *sb,
666 struct fuse_inode_handle *handle)
667{
668 struct fuse_conn *fc = get_fuse_conn_super(sb);
669 struct inode *inode;
670 struct dentry *entry;
671 int err = -ESTALE;
672
673 if (handle->nodeid == 0)
674 goto out_err;
675
676 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
677 if (!inode) {
678 struct fuse_entry_out outarg;
679 const struct qstr name = QSTR_INIT(".", 1);
680
681 if (!fc->export_support)
682 goto out_err;
683
684 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
685 &inode);
686 if (err && err != -ENOENT)
687 goto out_err;
688 if (err || !inode) {
689 err = -ESTALE;
690 goto out_err;
691 }
692 err = -EIO;
693 if (get_node_id(inode) != handle->nodeid)
694 goto out_iput;
695 }
696 err = -ESTALE;
697 if (inode->i_generation != handle->generation)
698 goto out_iput;
699
700 entry = d_obtain_alias(inode);
701 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
702 fuse_invalidate_entry_cache(entry);
703
704 return entry;
705
706 out_iput:
707 iput(inode);
708 out_err:
709 return ERR_PTR(err);
710}
711
712static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
713 struct inode *parent)
714{
715 int len = parent ? 6 : 3;
716 u64 nodeid;
717 u32 generation;
718
719 if (*max_len < len) {
720 *max_len = len;
721 return FILEID_INVALID;
722 }
723
724 nodeid = get_fuse_inode(inode)->nodeid;
725 generation = inode->i_generation;
726
727 fh[0] = (u32)(nodeid >> 32);
728 fh[1] = (u32)(nodeid & 0xffffffff);
729 fh[2] = generation;
730
731 if (parent) {
732 nodeid = get_fuse_inode(parent)->nodeid;
733 generation = parent->i_generation;
734
735 fh[3] = (u32)(nodeid >> 32);
736 fh[4] = (u32)(nodeid & 0xffffffff);
737 fh[5] = generation;
738 }
739
740 *max_len = len;
741 return parent ? 0x82 : 0x81;
742}
743
744static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
745 struct fid *fid, int fh_len, int fh_type)
746{
747 struct fuse_inode_handle handle;
748
749 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
750 return NULL;
751
752 handle.nodeid = (u64) fid->raw[0] << 32;
753 handle.nodeid |= (u64) fid->raw[1];
754 handle.generation = fid->raw[2];
755 return fuse_get_dentry(sb, &handle);
756}
757
758static struct dentry *fuse_fh_to_parent(struct super_block *sb,
759 struct fid *fid, int fh_len, int fh_type)
760{
761 struct fuse_inode_handle parent;
762
763 if (fh_type != 0x82 || fh_len < 6)
764 return NULL;
765
766 parent.nodeid = (u64) fid->raw[3] << 32;
767 parent.nodeid |= (u64) fid->raw[4];
768 parent.generation = fid->raw[5];
769 return fuse_get_dentry(sb, &parent);
770}
771
772static struct dentry *fuse_get_parent(struct dentry *child)
773{
774 struct inode *child_inode = d_inode(child);
775 struct fuse_conn *fc = get_fuse_conn(child_inode);
776 struct inode *inode;
777 struct dentry *parent;
778 struct fuse_entry_out outarg;
779 const struct qstr name = QSTR_INIT("..", 2);
780 int err;
781
782 if (!fc->export_support)
783 return ERR_PTR(-ESTALE);
784
785 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
786 &name, &outarg, &inode);
787 if (err) {
788 if (err == -ENOENT)
789 return ERR_PTR(-ESTALE);
790 return ERR_PTR(err);
791 }
792
793 parent = d_obtain_alias(inode);
794 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
795 fuse_invalidate_entry_cache(parent);
796
797 return parent;
798}
799
800static const struct export_operations fuse_export_operations = {
801 .fh_to_dentry = fuse_fh_to_dentry,
802 .fh_to_parent = fuse_fh_to_parent,
803 .encode_fh = fuse_encode_fh,
804 .get_parent = fuse_get_parent,
805};
806
807static const struct super_operations fuse_super_operations = {
808 .alloc_inode = fuse_alloc_inode,
809 .destroy_inode = fuse_destroy_inode,
810 .evict_inode = fuse_evict_inode,
811 .write_inode = fuse_write_inode,
812 .drop_inode = generic_delete_inode,
813 .remount_fs = fuse_remount_fs,
814 .put_super = fuse_put_super,
815 .umount_begin = fuse_umount_begin,
816 .statfs = fuse_statfs,
817 .show_options = fuse_show_options,
818};
819
820static void sanitize_global_limit(unsigned *limit)
821{
822 if (*limit == 0)
823 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
824 sizeof(struct fuse_req);
825
826 if (*limit >= 1 << 16)
827 *limit = (1 << 16) - 1;
828}
829
830static int set_global_limit(const char *val, struct kernel_param *kp)
831{
832 int rv;
833
834 rv = param_set_uint(val, kp);
835 if (rv)
836 return rv;
837
838 sanitize_global_limit((unsigned *)kp->arg);
839
840 return 0;
841}
842
843static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
844{
845 int cap_sys_admin = capable(CAP_SYS_ADMIN);
846
847 if (arg->minor < 13)
848 return;
849
850 sanitize_global_limit(&max_user_bgreq);
851 sanitize_global_limit(&max_user_congthresh);
852
853 if (arg->max_background) {
854 fc->max_background = arg->max_background;
855
856 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
857 fc->max_background = max_user_bgreq;
858 }
859 if (arg->congestion_threshold) {
860 fc->congestion_threshold = arg->congestion_threshold;
861
862 if (!cap_sys_admin &&
863 fc->congestion_threshold > max_user_congthresh)
864 fc->congestion_threshold = max_user_congthresh;
865 }
866}
867
868static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
869{
870 struct fuse_init_out *arg = &req->misc.init_out;
871
872 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
873 fc->conn_error = 1;
874 else {
875 unsigned long ra_pages;
876
877 process_init_limits(fc, arg);
878
879 if (arg->minor >= 6) {
880 ra_pages = arg->max_readahead / PAGE_SIZE;
881 if (arg->flags & FUSE_ASYNC_READ)
882 fc->async_read = 1;
883 if (!(arg->flags & FUSE_POSIX_LOCKS))
884 fc->no_lock = 1;
885 if (arg->minor >= 17) {
886 if (!(arg->flags & FUSE_FLOCK_LOCKS))
887 fc->no_flock = 1;
888 } else {
889 if (!(arg->flags & FUSE_POSIX_LOCKS))
890 fc->no_flock = 1;
891 }
892 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
893 fc->atomic_o_trunc = 1;
894 if (arg->minor >= 9) {
895 /* LOOKUP has dependency on proto version */
896 if (arg->flags & FUSE_EXPORT_SUPPORT)
897 fc->export_support = 1;
898 }
899 if (arg->flags & FUSE_BIG_WRITES)
900 fc->big_writes = 1;
901 if (arg->flags & FUSE_DONT_MASK)
902 fc->dont_mask = 1;
903 if (arg->flags & FUSE_AUTO_INVAL_DATA)
904 fc->auto_inval_data = 1;
905 if (arg->flags & FUSE_DO_READDIRPLUS) {
906 fc->do_readdirplus = 1;
907 if (arg->flags & FUSE_READDIRPLUS_AUTO)
908 fc->readdirplus_auto = 1;
909 }
910 if (arg->flags & FUSE_ASYNC_DIO)
911 fc->async_dio = 1;
912 if (arg->flags & FUSE_WRITEBACK_CACHE)
913 fc->writeback_cache = 1;
914 if (arg->flags & FUSE_PARALLEL_DIROPS)
915 fc->parallel_dirops = 1;
916 if (arg->flags & FUSE_HANDLE_KILLPRIV)
917 fc->handle_killpriv = 1;
918 if (arg->time_gran && arg->time_gran <= 1000000000)
919 fc->sb->s_time_gran = arg->time_gran;
920 if ((arg->flags & FUSE_POSIX_ACL)) {
921 fc->default_permissions = 1;
922 fc->posix_acl = 1;
923 fc->sb->s_xattr = fuse_acl_xattr_handlers;
924 }
925 } else {
926 ra_pages = fc->max_read / PAGE_SIZE;
927 fc->no_lock = 1;
928 fc->no_flock = 1;
929 }
930
931 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
932 fc->minor = arg->minor;
933 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
934 fc->max_write = max_t(unsigned, 4096, fc->max_write);
935 fc->conn_init = 1;
936 }
937 fuse_set_initialized(fc);
938 wake_up_all(&fc->blocked_waitq);
939}
940
941static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
942{
943 struct fuse_init_in *arg = &req->misc.init_in;
944
945 arg->major = FUSE_KERNEL_VERSION;
946 arg->minor = FUSE_KERNEL_MINOR_VERSION;
947 arg->max_readahead = fc->bdi.ra_pages * PAGE_SIZE;
948 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
949 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
950 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
951 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
952 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
953 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
954 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL;
955 req->in.h.opcode = FUSE_INIT;
956 req->in.numargs = 1;
957 req->in.args[0].size = sizeof(*arg);
958 req->in.args[0].value = arg;
959 req->out.numargs = 1;
960 /* Variable length argument used for backward compatibility
961 with interface version < 7.5. Rest of init_out is zeroed
962 by do_get_request(), so a short reply is not a problem */
963 req->out.argvar = 1;
964 req->out.args[0].size = sizeof(struct fuse_init_out);
965 req->out.args[0].value = &req->misc.init_out;
966 req->end = process_init_reply;
967 fuse_request_send_background(fc, req);
968}
969
970static void fuse_free_conn(struct fuse_conn *fc)
971{
972 WARN_ON(!list_empty(&fc->devices));
973 kfree_rcu(fc, rcu);
974}
975
976static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
977{
978 int err;
979
980 fc->bdi.name = "fuse";
981 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
982 /* fuse does it's own writeback accounting */
983 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
984
985 err = bdi_init(&fc->bdi);
986 if (err)
987 return err;
988
989 fc->bdi_initialized = 1;
990
991 if (sb->s_bdev) {
992 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
993 MAJOR(fc->dev), MINOR(fc->dev));
994 } else {
995 err = bdi_register_dev(&fc->bdi, fc->dev);
996 }
997
998 if (err)
999 return err;
1000
1001 /*
1002 * For a single fuse filesystem use max 1% of dirty +
1003 * writeback threshold.
1004 *
1005 * This gives about 1M of write buffer for memory maps on a
1006 * machine with 1G and 10% dirty_ratio, which should be more
1007 * than enough.
1008 *
1009 * Privileged users can raise it by writing to
1010 *
1011 * /sys/class/bdi/<bdi>/max_ratio
1012 */
1013 bdi_set_max_ratio(&fc->bdi, 1);
1014
1015 return 0;
1016}
1017
1018struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1019{
1020 struct fuse_dev *fud;
1021
1022 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1023 if (fud) {
1024 fud->fc = fuse_conn_get(fc);
1025 fuse_pqueue_init(&fud->pq);
1026
1027 spin_lock(&fc->lock);
1028 list_add_tail(&fud->entry, &fc->devices);
1029 spin_unlock(&fc->lock);
1030 }
1031
1032 return fud;
1033}
1034EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1035
1036void fuse_dev_free(struct fuse_dev *fud)
1037{
1038 struct fuse_conn *fc = fud->fc;
1039
1040 if (fc) {
1041 spin_lock(&fc->lock);
1042 list_del(&fud->entry);
1043 spin_unlock(&fc->lock);
1044
1045 fuse_conn_put(fc);
1046 }
1047 kfree(fud);
1048}
1049EXPORT_SYMBOL_GPL(fuse_dev_free);
1050
1051static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1052{
1053 struct fuse_dev *fud;
1054 struct fuse_conn *fc;
1055 struct inode *root;
1056 struct fuse_mount_data d;
1057 struct file *file;
1058 struct dentry *root_dentry;
1059 struct fuse_req *init_req;
1060 int err;
1061 int is_bdev = sb->s_bdev != NULL;
1062
1063 err = -EINVAL;
1064 if (sb->s_flags & MS_MANDLOCK)
1065 goto err;
1066
1067 sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
1068
1069 if (!parse_fuse_opt(data, &d, is_bdev))
1070 goto err;
1071
1072 if (is_bdev) {
1073#ifdef CONFIG_BLOCK
1074 err = -EINVAL;
1075 if (!sb_set_blocksize(sb, d.blksize))
1076 goto err;
1077#endif
1078 } else {
1079 sb->s_blocksize = PAGE_SIZE;
1080 sb->s_blocksize_bits = PAGE_SHIFT;
1081 }
1082 sb->s_magic = FUSE_SUPER_MAGIC;
1083 sb->s_op = &fuse_super_operations;
1084 sb->s_xattr = fuse_xattr_handlers;
1085 sb->s_maxbytes = MAX_LFS_FILESIZE;
1086 sb->s_time_gran = 1;
1087 sb->s_export_op = &fuse_export_operations;
1088
1089 file = fget(d.fd);
1090 err = -EINVAL;
1091 if (!file)
1092 goto err;
1093
1094 if ((file->f_op != &fuse_dev_operations) ||
1095 (file->f_cred->user_ns != &init_user_ns))
1096 goto err_fput;
1097
1098 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1099 err = -ENOMEM;
1100 if (!fc)
1101 goto err_fput;
1102
1103 fuse_conn_init(fc);
1104 fc->release = fuse_free_conn;
1105
1106 fud = fuse_dev_alloc(fc);
1107 if (!fud)
1108 goto err_put_conn;
1109
1110 fc->dev = sb->s_dev;
1111 fc->sb = sb;
1112 err = fuse_bdi_init(fc, sb);
1113 if (err)
1114 goto err_dev_free;
1115
1116 sb->s_bdi = &fc->bdi;
1117
1118 /* Handle umasking inside the fuse code */
1119 if (sb->s_flags & MS_POSIXACL)
1120 fc->dont_mask = 1;
1121 sb->s_flags |= MS_POSIXACL;
1122
1123 fc->default_permissions = d.default_permissions;
1124 fc->allow_other = d.allow_other;
1125 fc->user_id = d.user_id;
1126 fc->group_id = d.group_id;
1127 fc->max_read = max_t(unsigned, 4096, d.max_read);
1128
1129 /* Used by get_root_inode() */
1130 sb->s_fs_info = fc;
1131
1132 err = -ENOMEM;
1133 root = fuse_get_root_inode(sb, d.rootmode);
1134 sb->s_d_op = &fuse_root_dentry_operations;
1135 root_dentry = d_make_root(root);
1136 if (!root_dentry)
1137 goto err_dev_free;
1138 /* Root dentry doesn't have .d_revalidate */
1139 sb->s_d_op = &fuse_dentry_operations;
1140
1141 init_req = fuse_request_alloc(0);
1142 if (!init_req)
1143 goto err_put_root;
1144 __set_bit(FR_BACKGROUND, &init_req->flags);
1145
1146 if (is_bdev) {
1147 fc->destroy_req = fuse_request_alloc(0);
1148 if (!fc->destroy_req)
1149 goto err_free_init_req;
1150 }
1151
1152 mutex_lock(&fuse_mutex);
1153 err = -EINVAL;
1154 if (file->private_data)
1155 goto err_unlock;
1156
1157 err = fuse_ctl_add_conn(fc);
1158 if (err)
1159 goto err_unlock;
1160
1161 list_add_tail(&fc->entry, &fuse_conn_list);
1162 sb->s_root = root_dentry;
1163 file->private_data = fud;
1164 mutex_unlock(&fuse_mutex);
1165 /*
1166 * atomic_dec_and_test() in fput() provides the necessary
1167 * memory barrier for file->private_data to be visible on all
1168 * CPUs after this
1169 */
1170 fput(file);
1171
1172 fuse_send_init(fc, init_req);
1173
1174 return 0;
1175
1176 err_unlock:
1177 mutex_unlock(&fuse_mutex);
1178 err_free_init_req:
1179 fuse_request_free(init_req);
1180 err_put_root:
1181 dput(root_dentry);
1182 err_dev_free:
1183 fuse_dev_free(fud);
1184 err_put_conn:
1185 fuse_bdi_destroy(fc);
1186 fuse_conn_put(fc);
1187 err_fput:
1188 fput(file);
1189 err:
1190 return err;
1191}
1192
1193static struct dentry *fuse_mount(struct file_system_type *fs_type,
1194 int flags, const char *dev_name,
1195 void *raw_data)
1196{
1197 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1198}
1199
1200static void fuse_kill_sb_anon(struct super_block *sb)
1201{
1202 struct fuse_conn *fc = get_fuse_conn_super(sb);
1203
1204 if (fc) {
1205 down_write(&fc->killsb);
1206 fc->sb = NULL;
1207 up_write(&fc->killsb);
1208 }
1209
1210 kill_anon_super(sb);
1211}
1212
1213static struct file_system_type fuse_fs_type = {
1214 .owner = THIS_MODULE,
1215 .name = "fuse",
1216 .fs_flags = FS_HAS_SUBTYPE,
1217 .mount = fuse_mount,
1218 .kill_sb = fuse_kill_sb_anon,
1219};
1220MODULE_ALIAS_FS("fuse");
1221
1222#ifdef CONFIG_BLOCK
1223static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1224 int flags, const char *dev_name,
1225 void *raw_data)
1226{
1227 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1228}
1229
1230static void fuse_kill_sb_blk(struct super_block *sb)
1231{
1232 struct fuse_conn *fc = get_fuse_conn_super(sb);
1233
1234 if (fc) {
1235 down_write(&fc->killsb);
1236 fc->sb = NULL;
1237 up_write(&fc->killsb);
1238 }
1239
1240 kill_block_super(sb);
1241}
1242
1243static struct file_system_type fuseblk_fs_type = {
1244 .owner = THIS_MODULE,
1245 .name = "fuseblk",
1246 .mount = fuse_mount_blk,
1247 .kill_sb = fuse_kill_sb_blk,
1248 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1249};
1250MODULE_ALIAS_FS("fuseblk");
1251
1252static inline int register_fuseblk(void)
1253{
1254 return register_filesystem(&fuseblk_fs_type);
1255}
1256
1257static inline void unregister_fuseblk(void)
1258{
1259 unregister_filesystem(&fuseblk_fs_type);
1260}
1261#else
1262static inline int register_fuseblk(void)
1263{
1264 return 0;
1265}
1266
1267static inline void unregister_fuseblk(void)
1268{
1269}
1270#endif
1271
1272static void fuse_inode_init_once(void *foo)
1273{
1274 struct inode *inode = foo;
1275
1276 inode_init_once(inode);
1277}
1278
1279static int __init fuse_fs_init(void)
1280{
1281 int err;
1282
1283 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1284 sizeof(struct fuse_inode), 0,
1285 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT,
1286 fuse_inode_init_once);
1287 err = -ENOMEM;
1288 if (!fuse_inode_cachep)
1289 goto out;
1290
1291 err = register_fuseblk();
1292 if (err)
1293 goto out2;
1294
1295 err = register_filesystem(&fuse_fs_type);
1296 if (err)
1297 goto out3;
1298
1299 return 0;
1300
1301 out3:
1302 unregister_fuseblk();
1303 out2:
1304 kmem_cache_destroy(fuse_inode_cachep);
1305 out:
1306 return err;
1307}
1308
1309static void fuse_fs_cleanup(void)
1310{
1311 unregister_filesystem(&fuse_fs_type);
1312 unregister_fuseblk();
1313
1314 /*
1315 * Make sure all delayed rcu free inodes are flushed before we
1316 * destroy cache.
1317 */
1318 rcu_barrier();
1319 kmem_cache_destroy(fuse_inode_cachep);
1320}
1321
1322static struct kobject *fuse_kobj;
1323
1324static int fuse_sysfs_init(void)
1325{
1326 int err;
1327
1328 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1329 if (!fuse_kobj) {
1330 err = -ENOMEM;
1331 goto out_err;
1332 }
1333
1334 err = sysfs_create_mount_point(fuse_kobj, "connections");
1335 if (err)
1336 goto out_fuse_unregister;
1337
1338 return 0;
1339
1340 out_fuse_unregister:
1341 kobject_put(fuse_kobj);
1342 out_err:
1343 return err;
1344}
1345
1346static void fuse_sysfs_cleanup(void)
1347{
1348 sysfs_remove_mount_point(fuse_kobj, "connections");
1349 kobject_put(fuse_kobj);
1350}
1351
1352static int __init fuse_init(void)
1353{
1354 int res;
1355
1356 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1357 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1358
1359 INIT_LIST_HEAD(&fuse_conn_list);
1360 res = fuse_fs_init();
1361 if (res)
1362 goto err;
1363
1364 res = fuse_dev_init();
1365 if (res)
1366 goto err_fs_cleanup;
1367
1368 res = fuse_sysfs_init();
1369 if (res)
1370 goto err_dev_cleanup;
1371
1372 res = fuse_ctl_init();
1373 if (res)
1374 goto err_sysfs_cleanup;
1375
1376 sanitize_global_limit(&max_user_bgreq);
1377 sanitize_global_limit(&max_user_congthresh);
1378
1379 return 0;
1380
1381 err_sysfs_cleanup:
1382 fuse_sysfs_cleanup();
1383 err_dev_cleanup:
1384 fuse_dev_cleanup();
1385 err_fs_cleanup:
1386 fuse_fs_cleanup();
1387 err:
1388 return res;
1389}
1390
1391static void __exit fuse_exit(void)
1392{
1393 printk(KERN_DEBUG "fuse exit\n");
1394
1395 fuse_ctl_cleanup();
1396 fuse_sysfs_cleanup();
1397 fuse_fs_cleanup();
1398 fuse_dev_cleanup();
1399}
1400
1401module_init(fuse_init);
1402module_exit(fuse_exit);