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 INIT_LIST_HEAD(&fi->write_files);
95 INIT_LIST_HEAD(&fi->queued_writes);
96 INIT_LIST_HEAD(&fi->writepages);
97 init_waitqueue_head(&fi->page_waitq);
98 fi->forget = fuse_alloc_forget();
99 if (!fi->forget) {
100 kmem_cache_free(fuse_inode_cachep, inode);
101 return NULL;
102 }
103
104 return inode;
105}
106
107static void fuse_i_callback(struct rcu_head *head)
108{
109 struct inode *inode = container_of(head, struct inode, i_rcu);
110 INIT_LIST_HEAD(&inode->i_dentry);
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 end_writeback(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
143void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
144 u64 attr_valid)
145{
146 struct fuse_conn *fc = get_fuse_conn(inode);
147 struct fuse_inode *fi = get_fuse_inode(inode);
148
149 fi->attr_version = ++fc->attr_version;
150 fi->i_time = attr_valid;
151
152 inode->i_ino = attr->ino;
153 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
154 inode->i_nlink = attr->nlink;
155 inode->i_uid = attr->uid;
156 inode->i_gid = attr->gid;
157 inode->i_blocks = attr->blocks;
158 inode->i_atime.tv_sec = attr->atime;
159 inode->i_atime.tv_nsec = attr->atimensec;
160 inode->i_mtime.tv_sec = attr->mtime;
161 inode->i_mtime.tv_nsec = attr->mtimensec;
162 inode->i_ctime.tv_sec = attr->ctime;
163 inode->i_ctime.tv_nsec = attr->ctimensec;
164
165 if (attr->blksize != 0)
166 inode->i_blkbits = ilog2(attr->blksize);
167 else
168 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
169
170 /*
171 * Don't set the sticky bit in i_mode, unless we want the VFS
172 * to check permissions. This prevents failures due to the
173 * check in may_delete().
174 */
175 fi->orig_i_mode = inode->i_mode;
176 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
177 inode->i_mode &= ~S_ISVTX;
178}
179
180void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
181 u64 attr_valid, u64 attr_version)
182{
183 struct fuse_conn *fc = get_fuse_conn(inode);
184 struct fuse_inode *fi = get_fuse_inode(inode);
185 loff_t oldsize;
186
187 spin_lock(&fc->lock);
188 if (attr_version != 0 && fi->attr_version > attr_version) {
189 spin_unlock(&fc->lock);
190 return;
191 }
192
193 fuse_change_attributes_common(inode, attr, attr_valid);
194
195 oldsize = inode->i_size;
196 i_size_write(inode, attr->size);
197 spin_unlock(&fc->lock);
198
199 if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
200 truncate_pagecache(inode, oldsize, attr->size);
201 invalidate_inode_pages2(inode->i_mapping);
202 }
203}
204
205static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
206{
207 inode->i_mode = attr->mode & S_IFMT;
208 inode->i_size = attr->size;
209 if (S_ISREG(inode->i_mode)) {
210 fuse_init_common(inode);
211 fuse_init_file_inode(inode);
212 } else if (S_ISDIR(inode->i_mode))
213 fuse_init_dir(inode);
214 else if (S_ISLNK(inode->i_mode))
215 fuse_init_symlink(inode);
216 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
217 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
218 fuse_init_common(inode);
219 init_special_inode(inode, inode->i_mode,
220 new_decode_dev(attr->rdev));
221 } else
222 BUG();
223}
224
225int fuse_inode_eq(struct inode *inode, void *_nodeidp)
226{
227 u64 nodeid = *(u64 *) _nodeidp;
228 if (get_node_id(inode) == nodeid)
229 return 1;
230 else
231 return 0;
232}
233
234static int fuse_inode_set(struct inode *inode, void *_nodeidp)
235{
236 u64 nodeid = *(u64 *) _nodeidp;
237 get_fuse_inode(inode)->nodeid = nodeid;
238 return 0;
239}
240
241struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
242 int generation, struct fuse_attr *attr,
243 u64 attr_valid, u64 attr_version)
244{
245 struct inode *inode;
246 struct fuse_inode *fi;
247 struct fuse_conn *fc = get_fuse_conn_super(sb);
248
249 retry:
250 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
251 if (!inode)
252 return NULL;
253
254 if ((inode->i_state & I_NEW)) {
255 inode->i_flags |= S_NOATIME|S_NOCMTIME;
256 inode->i_generation = generation;
257 inode->i_data.backing_dev_info = &fc->bdi;
258 fuse_init_inode(inode, attr);
259 unlock_new_inode(inode);
260 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
261 /* Inode has changed type, any I/O on the old should fail */
262 make_bad_inode(inode);
263 iput(inode);
264 goto retry;
265 }
266
267 fi = get_fuse_inode(inode);
268 spin_lock(&fc->lock);
269 fi->nlookup++;
270 spin_unlock(&fc->lock);
271 fuse_change_attributes(inode, attr, attr_valid, attr_version);
272
273 return inode;
274}
275
276int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
277 loff_t offset, loff_t len)
278{
279 struct inode *inode;
280 pgoff_t pg_start;
281 pgoff_t pg_end;
282
283 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
284 if (!inode)
285 return -ENOENT;
286
287 fuse_invalidate_attr(inode);
288 if (offset >= 0) {
289 pg_start = offset >> PAGE_CACHE_SHIFT;
290 if (len <= 0)
291 pg_end = -1;
292 else
293 pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
294 invalidate_inode_pages2_range(inode->i_mapping,
295 pg_start, pg_end);
296 }
297 iput(inode);
298 return 0;
299}
300
301static void fuse_umount_begin(struct super_block *sb)
302{
303 fuse_abort_conn(get_fuse_conn_super(sb));
304}
305
306static void fuse_send_destroy(struct fuse_conn *fc)
307{
308 struct fuse_req *req = fc->destroy_req;
309 if (req && fc->conn_init) {
310 fc->destroy_req = NULL;
311 req->in.h.opcode = FUSE_DESTROY;
312 req->force = 1;
313 fuse_request_send(fc, req);
314 fuse_put_request(fc, req);
315 }
316}
317
318static void fuse_bdi_destroy(struct fuse_conn *fc)
319{
320 if (fc->bdi_initialized)
321 bdi_destroy(&fc->bdi);
322}
323
324void fuse_conn_kill(struct fuse_conn *fc)
325{
326 spin_lock(&fc->lock);
327 fc->connected = 0;
328 fc->blocked = 0;
329 spin_unlock(&fc->lock);
330 /* Flush all readers on this fs */
331 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
332 wake_up_all(&fc->waitq);
333 wake_up_all(&fc->blocked_waitq);
334 wake_up_all(&fc->reserved_req_waitq);
335 mutex_lock(&fuse_mutex);
336 list_del(&fc->entry);
337 fuse_ctl_remove_conn(fc);
338 mutex_unlock(&fuse_mutex);
339 fuse_bdi_destroy(fc);
340}
341EXPORT_SYMBOL_GPL(fuse_conn_kill);
342
343static void fuse_put_super(struct super_block *sb)
344{
345 struct fuse_conn *fc = get_fuse_conn_super(sb);
346
347 fuse_send_destroy(fc);
348 fuse_conn_kill(fc);
349 fuse_conn_put(fc);
350}
351
352static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
353{
354 stbuf->f_type = FUSE_SUPER_MAGIC;
355 stbuf->f_bsize = attr->bsize;
356 stbuf->f_frsize = attr->frsize;
357 stbuf->f_blocks = attr->blocks;
358 stbuf->f_bfree = attr->bfree;
359 stbuf->f_bavail = attr->bavail;
360 stbuf->f_files = attr->files;
361 stbuf->f_ffree = attr->ffree;
362 stbuf->f_namelen = attr->namelen;
363 /* fsid is left zero */
364}
365
366static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
367{
368 struct super_block *sb = dentry->d_sb;
369 struct fuse_conn *fc = get_fuse_conn_super(sb);
370 struct fuse_req *req;
371 struct fuse_statfs_out outarg;
372 int err;
373
374 if (!fuse_allow_task(fc, current)) {
375 buf->f_type = FUSE_SUPER_MAGIC;
376 return 0;
377 }
378
379 req = fuse_get_req(fc);
380 if (IS_ERR(req))
381 return PTR_ERR(req);
382
383 memset(&outarg, 0, sizeof(outarg));
384 req->in.numargs = 0;
385 req->in.h.opcode = FUSE_STATFS;
386 req->in.h.nodeid = get_node_id(dentry->d_inode);
387 req->out.numargs = 1;
388 req->out.args[0].size =
389 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
390 req->out.args[0].value = &outarg;
391 fuse_request_send(fc, req);
392 err = req->out.h.error;
393 if (!err)
394 convert_fuse_statfs(buf, &outarg.st);
395 fuse_put_request(fc, req);
396 return err;
397}
398
399enum {
400 OPT_FD,
401 OPT_ROOTMODE,
402 OPT_USER_ID,
403 OPT_GROUP_ID,
404 OPT_DEFAULT_PERMISSIONS,
405 OPT_ALLOW_OTHER,
406 OPT_MAX_READ,
407 OPT_BLKSIZE,
408 OPT_ERR
409};
410
411static const match_table_t tokens = {
412 {OPT_FD, "fd=%u"},
413 {OPT_ROOTMODE, "rootmode=%o"},
414 {OPT_USER_ID, "user_id=%u"},
415 {OPT_GROUP_ID, "group_id=%u"},
416 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
417 {OPT_ALLOW_OTHER, "allow_other"},
418 {OPT_MAX_READ, "max_read=%u"},
419 {OPT_BLKSIZE, "blksize=%u"},
420 {OPT_ERR, NULL}
421};
422
423static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
424{
425 char *p;
426 memset(d, 0, sizeof(struct fuse_mount_data));
427 d->max_read = ~0;
428 d->blksize = FUSE_DEFAULT_BLKSIZE;
429
430 while ((p = strsep(&opt, ",")) != NULL) {
431 int token;
432 int value;
433 substring_t args[MAX_OPT_ARGS];
434 if (!*p)
435 continue;
436
437 token = match_token(p, tokens, args);
438 switch (token) {
439 case OPT_FD:
440 if (match_int(&args[0], &value))
441 return 0;
442 d->fd = value;
443 d->fd_present = 1;
444 break;
445
446 case OPT_ROOTMODE:
447 if (match_octal(&args[0], &value))
448 return 0;
449 if (!fuse_valid_type(value))
450 return 0;
451 d->rootmode = value;
452 d->rootmode_present = 1;
453 break;
454
455 case OPT_USER_ID:
456 if (match_int(&args[0], &value))
457 return 0;
458 d->user_id = value;
459 d->user_id_present = 1;
460 break;
461
462 case OPT_GROUP_ID:
463 if (match_int(&args[0], &value))
464 return 0;
465 d->group_id = value;
466 d->group_id_present = 1;
467 break;
468
469 case OPT_DEFAULT_PERMISSIONS:
470 d->flags |= FUSE_DEFAULT_PERMISSIONS;
471 break;
472
473 case OPT_ALLOW_OTHER:
474 d->flags |= FUSE_ALLOW_OTHER;
475 break;
476
477 case OPT_MAX_READ:
478 if (match_int(&args[0], &value))
479 return 0;
480 d->max_read = value;
481 break;
482
483 case OPT_BLKSIZE:
484 if (!is_bdev || match_int(&args[0], &value))
485 return 0;
486 d->blksize = value;
487 break;
488
489 default:
490 return 0;
491 }
492 }
493
494 if (!d->fd_present || !d->rootmode_present ||
495 !d->user_id_present || !d->group_id_present)
496 return 0;
497
498 return 1;
499}
500
501static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
502{
503 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
504
505 seq_printf(m, ",user_id=%u", fc->user_id);
506 seq_printf(m, ",group_id=%u", fc->group_id);
507 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
508 seq_puts(m, ",default_permissions");
509 if (fc->flags & FUSE_ALLOW_OTHER)
510 seq_puts(m, ",allow_other");
511 if (fc->max_read != ~0)
512 seq_printf(m, ",max_read=%u", fc->max_read);
513 if (mnt->mnt_sb->s_bdev &&
514 mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
515 seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
516 return 0;
517}
518
519void fuse_conn_init(struct fuse_conn *fc)
520{
521 memset(fc, 0, sizeof(*fc));
522 spin_lock_init(&fc->lock);
523 mutex_init(&fc->inst_mutex);
524 init_rwsem(&fc->killsb);
525 atomic_set(&fc->count, 1);
526 init_waitqueue_head(&fc->waitq);
527 init_waitqueue_head(&fc->blocked_waitq);
528 init_waitqueue_head(&fc->reserved_req_waitq);
529 INIT_LIST_HEAD(&fc->pending);
530 INIT_LIST_HEAD(&fc->processing);
531 INIT_LIST_HEAD(&fc->io);
532 INIT_LIST_HEAD(&fc->interrupts);
533 INIT_LIST_HEAD(&fc->bg_queue);
534 INIT_LIST_HEAD(&fc->entry);
535 fc->forget_list_tail = &fc->forget_list_head;
536 atomic_set(&fc->num_waiting, 0);
537 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
538 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
539 fc->khctr = 0;
540 fc->polled_files = RB_ROOT;
541 fc->reqctr = 0;
542 fc->blocked = 1;
543 fc->attr_version = 1;
544 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
545}
546EXPORT_SYMBOL_GPL(fuse_conn_init);
547
548void fuse_conn_put(struct fuse_conn *fc)
549{
550 if (atomic_dec_and_test(&fc->count)) {
551 if (fc->destroy_req)
552 fuse_request_free(fc->destroy_req);
553 mutex_destroy(&fc->inst_mutex);
554 fc->release(fc);
555 }
556}
557EXPORT_SYMBOL_GPL(fuse_conn_put);
558
559struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
560{
561 atomic_inc(&fc->count);
562 return fc;
563}
564EXPORT_SYMBOL_GPL(fuse_conn_get);
565
566static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
567{
568 struct fuse_attr attr;
569 memset(&attr, 0, sizeof(attr));
570
571 attr.mode = mode;
572 attr.ino = FUSE_ROOT_ID;
573 attr.nlink = 1;
574 return fuse_iget(sb, 1, 0, &attr, 0, 0);
575}
576
577struct fuse_inode_handle {
578 u64 nodeid;
579 u32 generation;
580};
581
582static struct dentry *fuse_get_dentry(struct super_block *sb,
583 struct fuse_inode_handle *handle)
584{
585 struct fuse_conn *fc = get_fuse_conn_super(sb);
586 struct inode *inode;
587 struct dentry *entry;
588 int err = -ESTALE;
589
590 if (handle->nodeid == 0)
591 goto out_err;
592
593 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
594 if (!inode) {
595 struct fuse_entry_out outarg;
596 struct qstr name;
597
598 if (!fc->export_support)
599 goto out_err;
600
601 name.len = 1;
602 name.name = ".";
603 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
604 &inode);
605 if (err && err != -ENOENT)
606 goto out_err;
607 if (err || !inode) {
608 err = -ESTALE;
609 goto out_err;
610 }
611 err = -EIO;
612 if (get_node_id(inode) != handle->nodeid)
613 goto out_iput;
614 }
615 err = -ESTALE;
616 if (inode->i_generation != handle->generation)
617 goto out_iput;
618
619 entry = d_obtain_alias(inode);
620 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
621 fuse_invalidate_entry_cache(entry);
622
623 return entry;
624
625 out_iput:
626 iput(inode);
627 out_err:
628 return ERR_PTR(err);
629}
630
631static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
632 int connectable)
633{
634 struct inode *inode = dentry->d_inode;
635 bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
636 int len = encode_parent ? 6 : 3;
637 u64 nodeid;
638 u32 generation;
639
640 if (*max_len < len) {
641 *max_len = len;
642 return 255;
643 }
644
645 nodeid = get_fuse_inode(inode)->nodeid;
646 generation = inode->i_generation;
647
648 fh[0] = (u32)(nodeid >> 32);
649 fh[1] = (u32)(nodeid & 0xffffffff);
650 fh[2] = generation;
651
652 if (encode_parent) {
653 struct inode *parent;
654
655 spin_lock(&dentry->d_lock);
656 parent = dentry->d_parent->d_inode;
657 nodeid = get_fuse_inode(parent)->nodeid;
658 generation = parent->i_generation;
659 spin_unlock(&dentry->d_lock);
660
661 fh[3] = (u32)(nodeid >> 32);
662 fh[4] = (u32)(nodeid & 0xffffffff);
663 fh[5] = generation;
664 }
665
666 *max_len = len;
667 return encode_parent ? 0x82 : 0x81;
668}
669
670static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
671 struct fid *fid, int fh_len, int fh_type)
672{
673 struct fuse_inode_handle handle;
674
675 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
676 return NULL;
677
678 handle.nodeid = (u64) fid->raw[0] << 32;
679 handle.nodeid |= (u64) fid->raw[1];
680 handle.generation = fid->raw[2];
681 return fuse_get_dentry(sb, &handle);
682}
683
684static struct dentry *fuse_fh_to_parent(struct super_block *sb,
685 struct fid *fid, int fh_len, int fh_type)
686{
687 struct fuse_inode_handle parent;
688
689 if (fh_type != 0x82 || fh_len < 6)
690 return NULL;
691
692 parent.nodeid = (u64) fid->raw[3] << 32;
693 parent.nodeid |= (u64) fid->raw[4];
694 parent.generation = fid->raw[5];
695 return fuse_get_dentry(sb, &parent);
696}
697
698static struct dentry *fuse_get_parent(struct dentry *child)
699{
700 struct inode *child_inode = child->d_inode;
701 struct fuse_conn *fc = get_fuse_conn(child_inode);
702 struct inode *inode;
703 struct dentry *parent;
704 struct fuse_entry_out outarg;
705 struct qstr name;
706 int err;
707
708 if (!fc->export_support)
709 return ERR_PTR(-ESTALE);
710
711 name.len = 2;
712 name.name = "..";
713 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
714 &name, &outarg, &inode);
715 if (err) {
716 if (err == -ENOENT)
717 return ERR_PTR(-ESTALE);
718 return ERR_PTR(err);
719 }
720
721 parent = d_obtain_alias(inode);
722 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
723 fuse_invalidate_entry_cache(parent);
724
725 return parent;
726}
727
728static const struct export_operations fuse_export_operations = {
729 .fh_to_dentry = fuse_fh_to_dentry,
730 .fh_to_parent = fuse_fh_to_parent,
731 .encode_fh = fuse_encode_fh,
732 .get_parent = fuse_get_parent,
733};
734
735static const struct super_operations fuse_super_operations = {
736 .alloc_inode = fuse_alloc_inode,
737 .destroy_inode = fuse_destroy_inode,
738 .evict_inode = fuse_evict_inode,
739 .drop_inode = generic_delete_inode,
740 .remount_fs = fuse_remount_fs,
741 .put_super = fuse_put_super,
742 .umount_begin = fuse_umount_begin,
743 .statfs = fuse_statfs,
744 .show_options = fuse_show_options,
745};
746
747static void sanitize_global_limit(unsigned *limit)
748{
749 if (*limit == 0)
750 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
751 sizeof(struct fuse_req);
752
753 if (*limit >= 1 << 16)
754 *limit = (1 << 16) - 1;
755}
756
757static int set_global_limit(const char *val, struct kernel_param *kp)
758{
759 int rv;
760
761 rv = param_set_uint(val, kp);
762 if (rv)
763 return rv;
764
765 sanitize_global_limit((unsigned *)kp->arg);
766
767 return 0;
768}
769
770static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
771{
772 int cap_sys_admin = capable(CAP_SYS_ADMIN);
773
774 if (arg->minor < 13)
775 return;
776
777 sanitize_global_limit(&max_user_bgreq);
778 sanitize_global_limit(&max_user_congthresh);
779
780 if (arg->max_background) {
781 fc->max_background = arg->max_background;
782
783 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
784 fc->max_background = max_user_bgreq;
785 }
786 if (arg->congestion_threshold) {
787 fc->congestion_threshold = arg->congestion_threshold;
788
789 if (!cap_sys_admin &&
790 fc->congestion_threshold > max_user_congthresh)
791 fc->congestion_threshold = max_user_congthresh;
792 }
793}
794
795static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
796{
797 struct fuse_init_out *arg = &req->misc.init_out;
798
799 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
800 fc->conn_error = 1;
801 else {
802 unsigned long ra_pages;
803
804 process_init_limits(fc, arg);
805
806 if (arg->minor >= 6) {
807 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
808 if (arg->flags & FUSE_ASYNC_READ)
809 fc->async_read = 1;
810 if (!(arg->flags & FUSE_POSIX_LOCKS))
811 fc->no_lock = 1;
812 if (arg->minor >= 17) {
813 if (!(arg->flags & FUSE_FLOCK_LOCKS))
814 fc->no_flock = 1;
815 } else {
816 if (!(arg->flags & FUSE_POSIX_LOCKS))
817 fc->no_flock = 1;
818 }
819 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
820 fc->atomic_o_trunc = 1;
821 if (arg->minor >= 9) {
822 /* LOOKUP has dependency on proto version */
823 if (arg->flags & FUSE_EXPORT_SUPPORT)
824 fc->export_support = 1;
825 }
826 if (arg->flags & FUSE_BIG_WRITES)
827 fc->big_writes = 1;
828 if (arg->flags & FUSE_DONT_MASK)
829 fc->dont_mask = 1;
830 } else {
831 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
832 fc->no_lock = 1;
833 fc->no_flock = 1;
834 }
835
836 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
837 fc->minor = arg->minor;
838 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
839 fc->max_write = max_t(unsigned, 4096, fc->max_write);
840 fc->conn_init = 1;
841 }
842 fc->blocked = 0;
843 wake_up_all(&fc->blocked_waitq);
844}
845
846static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
847{
848 struct fuse_init_in *arg = &req->misc.init_in;
849
850 arg->major = FUSE_KERNEL_VERSION;
851 arg->minor = FUSE_KERNEL_MINOR_VERSION;
852 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
853 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
854 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
855 FUSE_FLOCK_LOCKS;
856 req->in.h.opcode = FUSE_INIT;
857 req->in.numargs = 1;
858 req->in.args[0].size = sizeof(*arg);
859 req->in.args[0].value = arg;
860 req->out.numargs = 1;
861 /* Variable length argument used for backward compatibility
862 with interface version < 7.5. Rest of init_out is zeroed
863 by do_get_request(), so a short reply is not a problem */
864 req->out.argvar = 1;
865 req->out.args[0].size = sizeof(struct fuse_init_out);
866 req->out.args[0].value = &req->misc.init_out;
867 req->end = process_init_reply;
868 fuse_request_send_background(fc, req);
869}
870
871static void fuse_free_conn(struct fuse_conn *fc)
872{
873 kfree(fc);
874}
875
876static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
877{
878 int err;
879
880 fc->bdi.name = "fuse";
881 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
882 /* fuse does it's own writeback accounting */
883 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
884
885 err = bdi_init(&fc->bdi);
886 if (err)
887 return err;
888
889 fc->bdi_initialized = 1;
890
891 if (sb->s_bdev) {
892 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
893 MAJOR(fc->dev), MINOR(fc->dev));
894 } else {
895 err = bdi_register_dev(&fc->bdi, fc->dev);
896 }
897
898 if (err)
899 return err;
900
901 /*
902 * For a single fuse filesystem use max 1% of dirty +
903 * writeback threshold.
904 *
905 * This gives about 1M of write buffer for memory maps on a
906 * machine with 1G and 10% dirty_ratio, which should be more
907 * than enough.
908 *
909 * Privileged users can raise it by writing to
910 *
911 * /sys/class/bdi/<bdi>/max_ratio
912 */
913 bdi_set_max_ratio(&fc->bdi, 1);
914
915 return 0;
916}
917
918static int fuse_fill_super(struct super_block *sb, void *data, int silent)
919{
920 struct fuse_conn *fc;
921 struct inode *root;
922 struct fuse_mount_data d;
923 struct file *file;
924 struct dentry *root_dentry;
925 struct fuse_req *init_req;
926 int err;
927 int is_bdev = sb->s_bdev != NULL;
928
929 err = -EINVAL;
930 if (sb->s_flags & MS_MANDLOCK)
931 goto err;
932
933 sb->s_flags &= ~MS_NOSEC;
934
935 if (!parse_fuse_opt((char *) data, &d, is_bdev))
936 goto err;
937
938 if (is_bdev) {
939#ifdef CONFIG_BLOCK
940 err = -EINVAL;
941 if (!sb_set_blocksize(sb, d.blksize))
942 goto err;
943#endif
944 } else {
945 sb->s_blocksize = PAGE_CACHE_SIZE;
946 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
947 }
948 sb->s_magic = FUSE_SUPER_MAGIC;
949 sb->s_op = &fuse_super_operations;
950 sb->s_maxbytes = MAX_LFS_FILESIZE;
951 sb->s_export_op = &fuse_export_operations;
952
953 file = fget(d.fd);
954 err = -EINVAL;
955 if (!file)
956 goto err;
957
958 if (file->f_op != &fuse_dev_operations)
959 goto err_fput;
960
961 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
962 err = -ENOMEM;
963 if (!fc)
964 goto err_fput;
965
966 fuse_conn_init(fc);
967
968 fc->dev = sb->s_dev;
969 fc->sb = sb;
970 err = fuse_bdi_init(fc, sb);
971 if (err)
972 goto err_put_conn;
973
974 sb->s_bdi = &fc->bdi;
975
976 /* Handle umasking inside the fuse code */
977 if (sb->s_flags & MS_POSIXACL)
978 fc->dont_mask = 1;
979 sb->s_flags |= MS_POSIXACL;
980
981 fc->release = fuse_free_conn;
982 fc->flags = d.flags;
983 fc->user_id = d.user_id;
984 fc->group_id = d.group_id;
985 fc->max_read = max_t(unsigned, 4096, d.max_read);
986
987 /* Used by get_root_inode() */
988 sb->s_fs_info = fc;
989
990 err = -ENOMEM;
991 root = fuse_get_root_inode(sb, d.rootmode);
992 if (!root)
993 goto err_put_conn;
994
995 root_dentry = d_alloc_root(root);
996 if (!root_dentry) {
997 iput(root);
998 goto err_put_conn;
999 }
1000 /* only now - we want root dentry with NULL ->d_op */
1001 sb->s_d_op = &fuse_dentry_operations;
1002
1003 init_req = fuse_request_alloc();
1004 if (!init_req)
1005 goto err_put_root;
1006
1007 if (is_bdev) {
1008 fc->destroy_req = fuse_request_alloc();
1009 if (!fc->destroy_req)
1010 goto err_free_init_req;
1011 }
1012
1013 mutex_lock(&fuse_mutex);
1014 err = -EINVAL;
1015 if (file->private_data)
1016 goto err_unlock;
1017
1018 err = fuse_ctl_add_conn(fc);
1019 if (err)
1020 goto err_unlock;
1021
1022 list_add_tail(&fc->entry, &fuse_conn_list);
1023 sb->s_root = root_dentry;
1024 fc->connected = 1;
1025 file->private_data = fuse_conn_get(fc);
1026 mutex_unlock(&fuse_mutex);
1027 /*
1028 * atomic_dec_and_test() in fput() provides the necessary
1029 * memory barrier for file->private_data to be visible on all
1030 * CPUs after this
1031 */
1032 fput(file);
1033
1034 fuse_send_init(fc, init_req);
1035
1036 return 0;
1037
1038 err_unlock:
1039 mutex_unlock(&fuse_mutex);
1040 err_free_init_req:
1041 fuse_request_free(init_req);
1042 err_put_root:
1043 dput(root_dentry);
1044 err_put_conn:
1045 fuse_bdi_destroy(fc);
1046 fuse_conn_put(fc);
1047 err_fput:
1048 fput(file);
1049 err:
1050 return err;
1051}
1052
1053static struct dentry *fuse_mount(struct file_system_type *fs_type,
1054 int flags, const char *dev_name,
1055 void *raw_data)
1056{
1057 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1058}
1059
1060static void fuse_kill_sb_anon(struct super_block *sb)
1061{
1062 struct fuse_conn *fc = get_fuse_conn_super(sb);
1063
1064 if (fc) {
1065 down_write(&fc->killsb);
1066 fc->sb = NULL;
1067 up_write(&fc->killsb);
1068 }
1069
1070 kill_anon_super(sb);
1071}
1072
1073static struct file_system_type fuse_fs_type = {
1074 .owner = THIS_MODULE,
1075 .name = "fuse",
1076 .fs_flags = FS_HAS_SUBTYPE,
1077 .mount = fuse_mount,
1078 .kill_sb = fuse_kill_sb_anon,
1079};
1080
1081#ifdef CONFIG_BLOCK
1082static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1083 int flags, const char *dev_name,
1084 void *raw_data)
1085{
1086 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1087}
1088
1089static void fuse_kill_sb_blk(struct super_block *sb)
1090{
1091 struct fuse_conn *fc = get_fuse_conn_super(sb);
1092
1093 if (fc) {
1094 down_write(&fc->killsb);
1095 fc->sb = NULL;
1096 up_write(&fc->killsb);
1097 }
1098
1099 kill_block_super(sb);
1100}
1101
1102static struct file_system_type fuseblk_fs_type = {
1103 .owner = THIS_MODULE,
1104 .name = "fuseblk",
1105 .mount = fuse_mount_blk,
1106 .kill_sb = fuse_kill_sb_blk,
1107 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1108};
1109
1110static inline int register_fuseblk(void)
1111{
1112 return register_filesystem(&fuseblk_fs_type);
1113}
1114
1115static inline void unregister_fuseblk(void)
1116{
1117 unregister_filesystem(&fuseblk_fs_type);
1118}
1119#else
1120static inline int register_fuseblk(void)
1121{
1122 return 0;
1123}
1124
1125static inline void unregister_fuseblk(void)
1126{
1127}
1128#endif
1129
1130static void fuse_inode_init_once(void *foo)
1131{
1132 struct inode *inode = foo;
1133
1134 inode_init_once(inode);
1135}
1136
1137static int __init fuse_fs_init(void)
1138{
1139 int err;
1140
1141 err = register_filesystem(&fuse_fs_type);
1142 if (err)
1143 goto out;
1144
1145 err = register_fuseblk();
1146 if (err)
1147 goto out_unreg;
1148
1149 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1150 sizeof(struct fuse_inode),
1151 0, SLAB_HWCACHE_ALIGN,
1152 fuse_inode_init_once);
1153 err = -ENOMEM;
1154 if (!fuse_inode_cachep)
1155 goto out_unreg2;
1156
1157 return 0;
1158
1159 out_unreg2:
1160 unregister_fuseblk();
1161 out_unreg:
1162 unregister_filesystem(&fuse_fs_type);
1163 out:
1164 return err;
1165}
1166
1167static void fuse_fs_cleanup(void)
1168{
1169 unregister_filesystem(&fuse_fs_type);
1170 unregister_fuseblk();
1171 kmem_cache_destroy(fuse_inode_cachep);
1172}
1173
1174static struct kobject *fuse_kobj;
1175static struct kobject *connections_kobj;
1176
1177static int fuse_sysfs_init(void)
1178{
1179 int err;
1180
1181 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1182 if (!fuse_kobj) {
1183 err = -ENOMEM;
1184 goto out_err;
1185 }
1186
1187 connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1188 if (!connections_kobj) {
1189 err = -ENOMEM;
1190 goto out_fuse_unregister;
1191 }
1192
1193 return 0;
1194
1195 out_fuse_unregister:
1196 kobject_put(fuse_kobj);
1197 out_err:
1198 return err;
1199}
1200
1201static void fuse_sysfs_cleanup(void)
1202{
1203 kobject_put(connections_kobj);
1204 kobject_put(fuse_kobj);
1205}
1206
1207static int __init fuse_init(void)
1208{
1209 int res;
1210
1211 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1212 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1213
1214 INIT_LIST_HEAD(&fuse_conn_list);
1215 res = fuse_fs_init();
1216 if (res)
1217 goto err;
1218
1219 res = fuse_dev_init();
1220 if (res)
1221 goto err_fs_cleanup;
1222
1223 res = fuse_sysfs_init();
1224 if (res)
1225 goto err_dev_cleanup;
1226
1227 res = fuse_ctl_init();
1228 if (res)
1229 goto err_sysfs_cleanup;
1230
1231 sanitize_global_limit(&max_user_bgreq);
1232 sanitize_global_limit(&max_user_congthresh);
1233
1234 return 0;
1235
1236 err_sysfs_cleanup:
1237 fuse_sysfs_cleanup();
1238 err_dev_cleanup:
1239 fuse_dev_cleanup();
1240 err_fs_cleanup:
1241 fuse_fs_cleanup();
1242 err:
1243 return res;
1244}
1245
1246static void __exit fuse_exit(void)
1247{
1248 printk(KERN_DEBUG "fuse exit\n");
1249
1250 fuse_ctl_cleanup();
1251 fuse_sysfs_cleanup();
1252 fuse_fs_cleanup();
1253 fuse_dev_cleanup();
1254}
1255
1256module_init(fuse_init);
1257module_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#include <linux/pid_namespace.h>
25
26MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
27MODULE_DESCRIPTION("Filesystem in Userspace");
28MODULE_LICENSE("GPL");
29
30static struct kmem_cache *fuse_inode_cachep;
31struct list_head fuse_conn_list;
32DEFINE_MUTEX(fuse_mutex);
33
34static int set_global_limit(const char *val, const struct kernel_param *kp);
35
36unsigned max_user_bgreq;
37module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
38 &max_user_bgreq, 0644);
39__MODULE_PARM_TYPE(max_user_bgreq, "uint");
40MODULE_PARM_DESC(max_user_bgreq,
41 "Global limit for the maximum number of backgrounded requests an "
42 "unprivileged user can set");
43
44unsigned max_user_congthresh;
45module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
46 &max_user_congthresh, 0644);
47__MODULE_PARM_TYPE(max_user_congthresh, "uint");
48MODULE_PARM_DESC(max_user_congthresh,
49 "Global limit for the maximum congestion threshold an "
50 "unprivileged user can set");
51
52#define FUSE_SUPER_MAGIC 0x65735546
53
54#define FUSE_DEFAULT_BLKSIZE 512
55
56/** Maximum number of outstanding background requests */
57#define FUSE_DEFAULT_MAX_BACKGROUND 12
58
59/** Congestion starts at 75% of maximum */
60#define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
61
62struct fuse_mount_data {
63 int fd;
64 unsigned rootmode;
65 kuid_t user_id;
66 kgid_t group_id;
67 unsigned fd_present:1;
68 unsigned rootmode_present:1;
69 unsigned user_id_present:1;
70 unsigned group_id_present:1;
71 unsigned default_permissions:1;
72 unsigned allow_other:1;
73 unsigned max_read;
74 unsigned blksize;
75};
76
77struct fuse_forget_link *fuse_alloc_forget(void)
78{
79 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
80}
81
82static struct inode *fuse_alloc_inode(struct super_block *sb)
83{
84 struct inode *inode;
85 struct fuse_inode *fi;
86
87 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
88 if (!inode)
89 return NULL;
90
91 fi = get_fuse_inode(inode);
92 fi->i_time = 0;
93 fi->nodeid = 0;
94 fi->nlookup = 0;
95 fi->attr_version = 0;
96 fi->writectr = 0;
97 fi->orig_ino = 0;
98 fi->state = 0;
99 INIT_LIST_HEAD(&fi->write_files);
100 INIT_LIST_HEAD(&fi->queued_writes);
101 INIT_LIST_HEAD(&fi->writepages);
102 init_waitqueue_head(&fi->page_waitq);
103 mutex_init(&fi->mutex);
104 fi->forget = fuse_alloc_forget();
105 if (!fi->forget) {
106 kmem_cache_free(fuse_inode_cachep, inode);
107 return NULL;
108 }
109
110 return inode;
111}
112
113static void fuse_i_callback(struct rcu_head *head)
114{
115 struct inode *inode = container_of(head, struct inode, i_rcu);
116 kmem_cache_free(fuse_inode_cachep, inode);
117}
118
119static void fuse_destroy_inode(struct inode *inode)
120{
121 struct fuse_inode *fi = get_fuse_inode(inode);
122 BUG_ON(!list_empty(&fi->write_files));
123 BUG_ON(!list_empty(&fi->queued_writes));
124 mutex_destroy(&fi->mutex);
125 kfree(fi->forget);
126 call_rcu(&inode->i_rcu, fuse_i_callback);
127}
128
129static void fuse_evict_inode(struct inode *inode)
130{
131 truncate_inode_pages_final(&inode->i_data);
132 clear_inode(inode);
133 if (inode->i_sb->s_flags & SB_ACTIVE) {
134 struct fuse_conn *fc = get_fuse_conn(inode);
135 struct fuse_inode *fi = get_fuse_inode(inode);
136 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
137 fi->forget = NULL;
138 }
139}
140
141static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
142{
143 sync_filesystem(sb);
144 if (*flags & SB_MANDLOCK)
145 return -EINVAL;
146
147 return 0;
148}
149
150/*
151 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
152 * so that it will fit.
153 */
154static ino_t fuse_squash_ino(u64 ino64)
155{
156 ino_t ino = (ino_t) ino64;
157 if (sizeof(ino_t) < sizeof(u64))
158 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
159 return ino;
160}
161
162void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
163 u64 attr_valid)
164{
165 struct fuse_conn *fc = get_fuse_conn(inode);
166 struct fuse_inode *fi = get_fuse_inode(inode);
167
168 fi->attr_version = ++fc->attr_version;
169 fi->i_time = attr_valid;
170
171 inode->i_ino = fuse_squash_ino(attr->ino);
172 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
173 set_nlink(inode, attr->nlink);
174 inode->i_uid = make_kuid(&init_user_ns, attr->uid);
175 inode->i_gid = make_kgid(&init_user_ns, attr->gid);
176 inode->i_blocks = attr->blocks;
177 inode->i_atime.tv_sec = attr->atime;
178 inode->i_atime.tv_nsec = attr->atimensec;
179 /* mtime from server may be stale due to local buffered write */
180 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
181 inode->i_mtime.tv_sec = attr->mtime;
182 inode->i_mtime.tv_nsec = attr->mtimensec;
183 inode->i_ctime.tv_sec = attr->ctime;
184 inode->i_ctime.tv_nsec = attr->ctimensec;
185 }
186
187 if (attr->blksize != 0)
188 inode->i_blkbits = ilog2(attr->blksize);
189 else
190 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
191
192 /*
193 * Don't set the sticky bit in i_mode, unless we want the VFS
194 * to check permissions. This prevents failures due to the
195 * check in may_delete().
196 */
197 fi->orig_i_mode = inode->i_mode;
198 if (!fc->default_permissions)
199 inode->i_mode &= ~S_ISVTX;
200
201 fi->orig_ino = attr->ino;
202}
203
204void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
205 u64 attr_valid, u64 attr_version)
206{
207 struct fuse_conn *fc = get_fuse_conn(inode);
208 struct fuse_inode *fi = get_fuse_inode(inode);
209 bool is_wb = fc->writeback_cache;
210 loff_t oldsize;
211 struct timespec old_mtime;
212
213 spin_lock(&fc->lock);
214 if ((attr_version != 0 && fi->attr_version > attr_version) ||
215 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
216 spin_unlock(&fc->lock);
217 return;
218 }
219
220 old_mtime = inode->i_mtime;
221 fuse_change_attributes_common(inode, attr, attr_valid);
222
223 oldsize = inode->i_size;
224 /*
225 * In case of writeback_cache enabled, the cached writes beyond EOF
226 * extend local i_size without keeping userspace server in sync. So,
227 * attr->size coming from server can be stale. We cannot trust it.
228 */
229 if (!is_wb || !S_ISREG(inode->i_mode))
230 i_size_write(inode, attr->size);
231 spin_unlock(&fc->lock);
232
233 if (!is_wb && S_ISREG(inode->i_mode)) {
234 bool inval = false;
235
236 if (oldsize != attr->size) {
237 truncate_pagecache(inode, attr->size);
238 inval = true;
239 } else if (fc->auto_inval_data) {
240 struct timespec new_mtime = {
241 .tv_sec = attr->mtime,
242 .tv_nsec = attr->mtimensec,
243 };
244
245 /*
246 * Auto inval mode also checks and invalidates if mtime
247 * has changed.
248 */
249 if (!timespec_equal(&old_mtime, &new_mtime))
250 inval = true;
251 }
252
253 if (inval)
254 invalidate_inode_pages2(inode->i_mapping);
255 }
256}
257
258static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
259{
260 inode->i_mode = attr->mode & S_IFMT;
261 inode->i_size = attr->size;
262 inode->i_mtime.tv_sec = attr->mtime;
263 inode->i_mtime.tv_nsec = attr->mtimensec;
264 inode->i_ctime.tv_sec = attr->ctime;
265 inode->i_ctime.tv_nsec = attr->ctimensec;
266 if (S_ISREG(inode->i_mode)) {
267 fuse_init_common(inode);
268 fuse_init_file_inode(inode);
269 } else if (S_ISDIR(inode->i_mode))
270 fuse_init_dir(inode);
271 else if (S_ISLNK(inode->i_mode))
272 fuse_init_symlink(inode);
273 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
274 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
275 fuse_init_common(inode);
276 init_special_inode(inode, inode->i_mode,
277 new_decode_dev(attr->rdev));
278 } else
279 BUG();
280}
281
282int fuse_inode_eq(struct inode *inode, void *_nodeidp)
283{
284 u64 nodeid = *(u64 *) _nodeidp;
285 if (get_node_id(inode) == nodeid)
286 return 1;
287 else
288 return 0;
289}
290
291static int fuse_inode_set(struct inode *inode, void *_nodeidp)
292{
293 u64 nodeid = *(u64 *) _nodeidp;
294 get_fuse_inode(inode)->nodeid = nodeid;
295 return 0;
296}
297
298struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
299 int generation, struct fuse_attr *attr,
300 u64 attr_valid, u64 attr_version)
301{
302 struct inode *inode;
303 struct fuse_inode *fi;
304 struct fuse_conn *fc = get_fuse_conn_super(sb);
305
306 retry:
307 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
308 if (!inode)
309 return NULL;
310
311 if ((inode->i_state & I_NEW)) {
312 inode->i_flags |= S_NOATIME;
313 if (!fc->writeback_cache || !S_ISREG(attr->mode))
314 inode->i_flags |= S_NOCMTIME;
315 inode->i_generation = generation;
316 fuse_init_inode(inode, attr);
317 unlock_new_inode(inode);
318 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
319 /* Inode has changed type, any I/O on the old should fail */
320 make_bad_inode(inode);
321 iput(inode);
322 goto retry;
323 }
324
325 fi = get_fuse_inode(inode);
326 spin_lock(&fc->lock);
327 fi->nlookup++;
328 spin_unlock(&fc->lock);
329 fuse_change_attributes(inode, attr, attr_valid, attr_version);
330
331 return inode;
332}
333
334int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
335 loff_t offset, loff_t len)
336{
337 struct inode *inode;
338 pgoff_t pg_start;
339 pgoff_t pg_end;
340
341 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
342 if (!inode)
343 return -ENOENT;
344
345 fuse_invalidate_attr(inode);
346 forget_all_cached_acls(inode);
347 if (offset >= 0) {
348 pg_start = offset >> PAGE_SHIFT;
349 if (len <= 0)
350 pg_end = -1;
351 else
352 pg_end = (offset + len - 1) >> PAGE_SHIFT;
353 invalidate_inode_pages2_range(inode->i_mapping,
354 pg_start, pg_end);
355 }
356 iput(inode);
357 return 0;
358}
359
360void fuse_lock_inode(struct inode *inode)
361{
362 if (!get_fuse_conn(inode)->parallel_dirops)
363 mutex_lock(&get_fuse_inode(inode)->mutex);
364}
365
366void fuse_unlock_inode(struct inode *inode)
367{
368 if (!get_fuse_conn(inode)->parallel_dirops)
369 mutex_unlock(&get_fuse_inode(inode)->mutex);
370}
371
372static void fuse_umount_begin(struct super_block *sb)
373{
374 fuse_abort_conn(get_fuse_conn_super(sb));
375}
376
377static void fuse_send_destroy(struct fuse_conn *fc)
378{
379 struct fuse_req *req = fc->destroy_req;
380 if (req && fc->conn_init) {
381 fc->destroy_req = NULL;
382 req->in.h.opcode = FUSE_DESTROY;
383 __set_bit(FR_FORCE, &req->flags);
384 __clear_bit(FR_BACKGROUND, &req->flags);
385 fuse_request_send(fc, req);
386 fuse_put_request(fc, req);
387 }
388}
389
390static void fuse_put_super(struct super_block *sb)
391{
392 struct fuse_conn *fc = get_fuse_conn_super(sb);
393
394 fuse_send_destroy(fc);
395
396 fuse_abort_conn(fc);
397 mutex_lock(&fuse_mutex);
398 list_del(&fc->entry);
399 fuse_ctl_remove_conn(fc);
400 mutex_unlock(&fuse_mutex);
401
402 fuse_conn_put(fc);
403}
404
405static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
406{
407 stbuf->f_type = FUSE_SUPER_MAGIC;
408 stbuf->f_bsize = attr->bsize;
409 stbuf->f_frsize = attr->frsize;
410 stbuf->f_blocks = attr->blocks;
411 stbuf->f_bfree = attr->bfree;
412 stbuf->f_bavail = attr->bavail;
413 stbuf->f_files = attr->files;
414 stbuf->f_ffree = attr->ffree;
415 stbuf->f_namelen = attr->namelen;
416 /* fsid is left zero */
417}
418
419static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
420{
421 struct super_block *sb = dentry->d_sb;
422 struct fuse_conn *fc = get_fuse_conn_super(sb);
423 FUSE_ARGS(args);
424 struct fuse_statfs_out outarg;
425 int err;
426
427 if (!fuse_allow_current_process(fc)) {
428 buf->f_type = FUSE_SUPER_MAGIC;
429 return 0;
430 }
431
432 memset(&outarg, 0, sizeof(outarg));
433 args.in.numargs = 0;
434 args.in.h.opcode = FUSE_STATFS;
435 args.in.h.nodeid = get_node_id(d_inode(dentry));
436 args.out.numargs = 1;
437 args.out.args[0].size = sizeof(outarg);
438 args.out.args[0].value = &outarg;
439 err = fuse_simple_request(fc, &args);
440 if (!err)
441 convert_fuse_statfs(buf, &outarg.st);
442 return err;
443}
444
445enum {
446 OPT_FD,
447 OPT_ROOTMODE,
448 OPT_USER_ID,
449 OPT_GROUP_ID,
450 OPT_DEFAULT_PERMISSIONS,
451 OPT_ALLOW_OTHER,
452 OPT_MAX_READ,
453 OPT_BLKSIZE,
454 OPT_ERR
455};
456
457static const match_table_t tokens = {
458 {OPT_FD, "fd=%u"},
459 {OPT_ROOTMODE, "rootmode=%o"},
460 {OPT_USER_ID, "user_id=%u"},
461 {OPT_GROUP_ID, "group_id=%u"},
462 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
463 {OPT_ALLOW_OTHER, "allow_other"},
464 {OPT_MAX_READ, "max_read=%u"},
465 {OPT_BLKSIZE, "blksize=%u"},
466 {OPT_ERR, NULL}
467};
468
469static int fuse_match_uint(substring_t *s, unsigned int *res)
470{
471 int err = -ENOMEM;
472 char *buf = match_strdup(s);
473 if (buf) {
474 err = kstrtouint(buf, 10, res);
475 kfree(buf);
476 }
477 return err;
478}
479
480static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
481{
482 char *p;
483 memset(d, 0, sizeof(struct fuse_mount_data));
484 d->max_read = ~0;
485 d->blksize = FUSE_DEFAULT_BLKSIZE;
486
487 while ((p = strsep(&opt, ",")) != NULL) {
488 int token;
489 int value;
490 unsigned uv;
491 substring_t args[MAX_OPT_ARGS];
492 if (!*p)
493 continue;
494
495 token = match_token(p, tokens, args);
496 switch (token) {
497 case OPT_FD:
498 if (match_int(&args[0], &value))
499 return 0;
500 d->fd = value;
501 d->fd_present = 1;
502 break;
503
504 case OPT_ROOTMODE:
505 if (match_octal(&args[0], &value))
506 return 0;
507 if (!fuse_valid_type(value))
508 return 0;
509 d->rootmode = value;
510 d->rootmode_present = 1;
511 break;
512
513 case OPT_USER_ID:
514 if (fuse_match_uint(&args[0], &uv))
515 return 0;
516 d->user_id = make_kuid(current_user_ns(), uv);
517 if (!uid_valid(d->user_id))
518 return 0;
519 d->user_id_present = 1;
520 break;
521
522 case OPT_GROUP_ID:
523 if (fuse_match_uint(&args[0], &uv))
524 return 0;
525 d->group_id = make_kgid(current_user_ns(), uv);
526 if (!gid_valid(d->group_id))
527 return 0;
528 d->group_id_present = 1;
529 break;
530
531 case OPT_DEFAULT_PERMISSIONS:
532 d->default_permissions = 1;
533 break;
534
535 case OPT_ALLOW_OTHER:
536 d->allow_other = 1;
537 break;
538
539 case OPT_MAX_READ:
540 if (match_int(&args[0], &value))
541 return 0;
542 d->max_read = value;
543 break;
544
545 case OPT_BLKSIZE:
546 if (!is_bdev || match_int(&args[0], &value))
547 return 0;
548 d->blksize = value;
549 break;
550
551 default:
552 return 0;
553 }
554 }
555
556 if (!d->fd_present || !d->rootmode_present ||
557 !d->user_id_present || !d->group_id_present)
558 return 0;
559
560 return 1;
561}
562
563static int fuse_show_options(struct seq_file *m, struct dentry *root)
564{
565 struct super_block *sb = root->d_sb;
566 struct fuse_conn *fc = get_fuse_conn_super(sb);
567
568 seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
569 seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
570 if (fc->default_permissions)
571 seq_puts(m, ",default_permissions");
572 if (fc->allow_other)
573 seq_puts(m, ",allow_other");
574 if (fc->max_read != ~0)
575 seq_printf(m, ",max_read=%u", fc->max_read);
576 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
577 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
578 return 0;
579}
580
581static void fuse_iqueue_init(struct fuse_iqueue *fiq)
582{
583 memset(fiq, 0, sizeof(struct fuse_iqueue));
584 init_waitqueue_head(&fiq->waitq);
585 INIT_LIST_HEAD(&fiq->pending);
586 INIT_LIST_HEAD(&fiq->interrupts);
587 fiq->forget_list_tail = &fiq->forget_list_head;
588 fiq->connected = 1;
589}
590
591static void fuse_pqueue_init(struct fuse_pqueue *fpq)
592{
593 memset(fpq, 0, sizeof(struct fuse_pqueue));
594 spin_lock_init(&fpq->lock);
595 INIT_LIST_HEAD(&fpq->processing);
596 INIT_LIST_HEAD(&fpq->io);
597 fpq->connected = 1;
598}
599
600void fuse_conn_init(struct fuse_conn *fc)
601{
602 memset(fc, 0, sizeof(*fc));
603 spin_lock_init(&fc->lock);
604 init_rwsem(&fc->killsb);
605 refcount_set(&fc->count, 1);
606 atomic_set(&fc->dev_count, 1);
607 init_waitqueue_head(&fc->blocked_waitq);
608 init_waitqueue_head(&fc->reserved_req_waitq);
609 fuse_iqueue_init(&fc->iq);
610 INIT_LIST_HEAD(&fc->bg_queue);
611 INIT_LIST_HEAD(&fc->entry);
612 INIT_LIST_HEAD(&fc->devices);
613 atomic_set(&fc->num_waiting, 0);
614 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
615 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
616 fc->khctr = 0;
617 fc->polled_files = RB_ROOT;
618 fc->blocked = 0;
619 fc->initialized = 0;
620 fc->connected = 1;
621 fc->attr_version = 1;
622 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
623 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
624}
625EXPORT_SYMBOL_GPL(fuse_conn_init);
626
627void fuse_conn_put(struct fuse_conn *fc)
628{
629 if (refcount_dec_and_test(&fc->count)) {
630 if (fc->destroy_req)
631 fuse_request_free(fc->destroy_req);
632 put_pid_ns(fc->pid_ns);
633 fc->release(fc);
634 }
635}
636EXPORT_SYMBOL_GPL(fuse_conn_put);
637
638struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
639{
640 refcount_inc(&fc->count);
641 return fc;
642}
643EXPORT_SYMBOL_GPL(fuse_conn_get);
644
645static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
646{
647 struct fuse_attr attr;
648 memset(&attr, 0, sizeof(attr));
649
650 attr.mode = mode;
651 attr.ino = FUSE_ROOT_ID;
652 attr.nlink = 1;
653 return fuse_iget(sb, 1, 0, &attr, 0, 0);
654}
655
656struct fuse_inode_handle {
657 u64 nodeid;
658 u32 generation;
659};
660
661static struct dentry *fuse_get_dentry(struct super_block *sb,
662 struct fuse_inode_handle *handle)
663{
664 struct fuse_conn *fc = get_fuse_conn_super(sb);
665 struct inode *inode;
666 struct dentry *entry;
667 int err = -ESTALE;
668
669 if (handle->nodeid == 0)
670 goto out_err;
671
672 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
673 if (!inode) {
674 struct fuse_entry_out outarg;
675 const struct qstr name = QSTR_INIT(".", 1);
676
677 if (!fc->export_support)
678 goto out_err;
679
680 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
681 &inode);
682 if (err && err != -ENOENT)
683 goto out_err;
684 if (err || !inode) {
685 err = -ESTALE;
686 goto out_err;
687 }
688 err = -EIO;
689 if (get_node_id(inode) != handle->nodeid)
690 goto out_iput;
691 }
692 err = -ESTALE;
693 if (inode->i_generation != handle->generation)
694 goto out_iput;
695
696 entry = d_obtain_alias(inode);
697 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
698 fuse_invalidate_entry_cache(entry);
699
700 return entry;
701
702 out_iput:
703 iput(inode);
704 out_err:
705 return ERR_PTR(err);
706}
707
708static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
709 struct inode *parent)
710{
711 int len = parent ? 6 : 3;
712 u64 nodeid;
713 u32 generation;
714
715 if (*max_len < len) {
716 *max_len = len;
717 return FILEID_INVALID;
718 }
719
720 nodeid = get_fuse_inode(inode)->nodeid;
721 generation = inode->i_generation;
722
723 fh[0] = (u32)(nodeid >> 32);
724 fh[1] = (u32)(nodeid & 0xffffffff);
725 fh[2] = generation;
726
727 if (parent) {
728 nodeid = get_fuse_inode(parent)->nodeid;
729 generation = parent->i_generation;
730
731 fh[3] = (u32)(nodeid >> 32);
732 fh[4] = (u32)(nodeid & 0xffffffff);
733 fh[5] = generation;
734 }
735
736 *max_len = len;
737 return parent ? 0x82 : 0x81;
738}
739
740static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
741 struct fid *fid, int fh_len, int fh_type)
742{
743 struct fuse_inode_handle handle;
744
745 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
746 return NULL;
747
748 handle.nodeid = (u64) fid->raw[0] << 32;
749 handle.nodeid |= (u64) fid->raw[1];
750 handle.generation = fid->raw[2];
751 return fuse_get_dentry(sb, &handle);
752}
753
754static struct dentry *fuse_fh_to_parent(struct super_block *sb,
755 struct fid *fid, int fh_len, int fh_type)
756{
757 struct fuse_inode_handle parent;
758
759 if (fh_type != 0x82 || fh_len < 6)
760 return NULL;
761
762 parent.nodeid = (u64) fid->raw[3] << 32;
763 parent.nodeid |= (u64) fid->raw[4];
764 parent.generation = fid->raw[5];
765 return fuse_get_dentry(sb, &parent);
766}
767
768static struct dentry *fuse_get_parent(struct dentry *child)
769{
770 struct inode *child_inode = d_inode(child);
771 struct fuse_conn *fc = get_fuse_conn(child_inode);
772 struct inode *inode;
773 struct dentry *parent;
774 struct fuse_entry_out outarg;
775 const struct qstr name = QSTR_INIT("..", 2);
776 int err;
777
778 if (!fc->export_support)
779 return ERR_PTR(-ESTALE);
780
781 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
782 &name, &outarg, &inode);
783 if (err) {
784 if (err == -ENOENT)
785 return ERR_PTR(-ESTALE);
786 return ERR_PTR(err);
787 }
788
789 parent = d_obtain_alias(inode);
790 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
791 fuse_invalidate_entry_cache(parent);
792
793 return parent;
794}
795
796static const struct export_operations fuse_export_operations = {
797 .fh_to_dentry = fuse_fh_to_dentry,
798 .fh_to_parent = fuse_fh_to_parent,
799 .encode_fh = fuse_encode_fh,
800 .get_parent = fuse_get_parent,
801};
802
803static const struct super_operations fuse_super_operations = {
804 .alloc_inode = fuse_alloc_inode,
805 .destroy_inode = fuse_destroy_inode,
806 .evict_inode = fuse_evict_inode,
807 .write_inode = fuse_write_inode,
808 .drop_inode = generic_delete_inode,
809 .remount_fs = fuse_remount_fs,
810 .put_super = fuse_put_super,
811 .umount_begin = fuse_umount_begin,
812 .statfs = fuse_statfs,
813 .show_options = fuse_show_options,
814};
815
816static void sanitize_global_limit(unsigned *limit)
817{
818 if (*limit == 0)
819 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
820 sizeof(struct fuse_req);
821
822 if (*limit >= 1 << 16)
823 *limit = (1 << 16) - 1;
824}
825
826static int set_global_limit(const char *val, const struct kernel_param *kp)
827{
828 int rv;
829
830 rv = param_set_uint(val, kp);
831 if (rv)
832 return rv;
833
834 sanitize_global_limit((unsigned *)kp->arg);
835
836 return 0;
837}
838
839static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
840{
841 int cap_sys_admin = capable(CAP_SYS_ADMIN);
842
843 if (arg->minor < 13)
844 return;
845
846 sanitize_global_limit(&max_user_bgreq);
847 sanitize_global_limit(&max_user_congthresh);
848
849 if (arg->max_background) {
850 fc->max_background = arg->max_background;
851
852 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
853 fc->max_background = max_user_bgreq;
854 }
855 if (arg->congestion_threshold) {
856 fc->congestion_threshold = arg->congestion_threshold;
857
858 if (!cap_sys_admin &&
859 fc->congestion_threshold > max_user_congthresh)
860 fc->congestion_threshold = max_user_congthresh;
861 }
862}
863
864static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
865{
866 struct fuse_init_out *arg = &req->misc.init_out;
867
868 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
869 fc->conn_error = 1;
870 else {
871 unsigned long ra_pages;
872
873 process_init_limits(fc, arg);
874
875 if (arg->minor >= 6) {
876 ra_pages = arg->max_readahead / PAGE_SIZE;
877 if (arg->flags & FUSE_ASYNC_READ)
878 fc->async_read = 1;
879 if (!(arg->flags & FUSE_POSIX_LOCKS))
880 fc->no_lock = 1;
881 if (arg->minor >= 17) {
882 if (!(arg->flags & FUSE_FLOCK_LOCKS))
883 fc->no_flock = 1;
884 } else {
885 if (!(arg->flags & FUSE_POSIX_LOCKS))
886 fc->no_flock = 1;
887 }
888 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
889 fc->atomic_o_trunc = 1;
890 if (arg->minor >= 9) {
891 /* LOOKUP has dependency on proto version */
892 if (arg->flags & FUSE_EXPORT_SUPPORT)
893 fc->export_support = 1;
894 }
895 if (arg->flags & FUSE_BIG_WRITES)
896 fc->big_writes = 1;
897 if (arg->flags & FUSE_DONT_MASK)
898 fc->dont_mask = 1;
899 if (arg->flags & FUSE_AUTO_INVAL_DATA)
900 fc->auto_inval_data = 1;
901 if (arg->flags & FUSE_DO_READDIRPLUS) {
902 fc->do_readdirplus = 1;
903 if (arg->flags & FUSE_READDIRPLUS_AUTO)
904 fc->readdirplus_auto = 1;
905 }
906 if (arg->flags & FUSE_ASYNC_DIO)
907 fc->async_dio = 1;
908 if (arg->flags & FUSE_WRITEBACK_CACHE)
909 fc->writeback_cache = 1;
910 if (arg->flags & FUSE_PARALLEL_DIROPS)
911 fc->parallel_dirops = 1;
912 if (arg->flags & FUSE_HANDLE_KILLPRIV)
913 fc->handle_killpriv = 1;
914 if (arg->time_gran && arg->time_gran <= 1000000000)
915 fc->sb->s_time_gran = arg->time_gran;
916 if ((arg->flags & FUSE_POSIX_ACL)) {
917 fc->default_permissions = 1;
918 fc->posix_acl = 1;
919 fc->sb->s_xattr = fuse_acl_xattr_handlers;
920 }
921 } else {
922 ra_pages = fc->max_read / PAGE_SIZE;
923 fc->no_lock = 1;
924 fc->no_flock = 1;
925 }
926
927 fc->sb->s_bdi->ra_pages =
928 min(fc->sb->s_bdi->ra_pages, ra_pages);
929 fc->minor = arg->minor;
930 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
931 fc->max_write = max_t(unsigned, 4096, fc->max_write);
932 fc->conn_init = 1;
933 }
934 fuse_set_initialized(fc);
935 wake_up_all(&fc->blocked_waitq);
936}
937
938static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
939{
940 struct fuse_init_in *arg = &req->misc.init_in;
941
942 arg->major = FUSE_KERNEL_VERSION;
943 arg->minor = FUSE_KERNEL_MINOR_VERSION;
944 arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
945 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
946 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
947 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
948 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
949 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
950 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
951 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL;
952 req->in.h.opcode = FUSE_INIT;
953 req->in.numargs = 1;
954 req->in.args[0].size = sizeof(*arg);
955 req->in.args[0].value = arg;
956 req->out.numargs = 1;
957 /* Variable length argument used for backward compatibility
958 with interface version < 7.5. Rest of init_out is zeroed
959 by do_get_request(), so a short reply is not a problem */
960 req->out.argvar = 1;
961 req->out.args[0].size = sizeof(struct fuse_init_out);
962 req->out.args[0].value = &req->misc.init_out;
963 req->end = process_init_reply;
964 fuse_request_send_background(fc, req);
965}
966
967static void fuse_free_conn(struct fuse_conn *fc)
968{
969 WARN_ON(!list_empty(&fc->devices));
970 kfree_rcu(fc, rcu);
971}
972
973static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
974{
975 int err;
976 char *suffix = "";
977
978 if (sb->s_bdev) {
979 suffix = "-fuseblk";
980 /*
981 * sb->s_bdi points to blkdev's bdi however we want to redirect
982 * it to our private bdi...
983 */
984 bdi_put(sb->s_bdi);
985 sb->s_bdi = &noop_backing_dev_info;
986 }
987 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
988 MINOR(fc->dev), suffix);
989 if (err)
990 return err;
991
992 sb->s_bdi->ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
993 /* fuse does it's own writeback accounting */
994 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
995
996 /*
997 * For a single fuse filesystem use max 1% of dirty +
998 * writeback threshold.
999 *
1000 * This gives about 1M of write buffer for memory maps on a
1001 * machine with 1G and 10% dirty_ratio, which should be more
1002 * than enough.
1003 *
1004 * Privileged users can raise it by writing to
1005 *
1006 * /sys/class/bdi/<bdi>/max_ratio
1007 */
1008 bdi_set_max_ratio(sb->s_bdi, 1);
1009
1010 return 0;
1011}
1012
1013struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1014{
1015 struct fuse_dev *fud;
1016
1017 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1018 if (fud) {
1019 fud->fc = fuse_conn_get(fc);
1020 fuse_pqueue_init(&fud->pq);
1021
1022 spin_lock(&fc->lock);
1023 list_add_tail(&fud->entry, &fc->devices);
1024 spin_unlock(&fc->lock);
1025 }
1026
1027 return fud;
1028}
1029EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1030
1031void fuse_dev_free(struct fuse_dev *fud)
1032{
1033 struct fuse_conn *fc = fud->fc;
1034
1035 if (fc) {
1036 spin_lock(&fc->lock);
1037 list_del(&fud->entry);
1038 spin_unlock(&fc->lock);
1039
1040 fuse_conn_put(fc);
1041 }
1042 kfree(fud);
1043}
1044EXPORT_SYMBOL_GPL(fuse_dev_free);
1045
1046static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1047{
1048 struct fuse_dev *fud;
1049 struct fuse_conn *fc;
1050 struct inode *root;
1051 struct fuse_mount_data d;
1052 struct file *file;
1053 struct dentry *root_dentry;
1054 struct fuse_req *init_req;
1055 int err;
1056 int is_bdev = sb->s_bdev != NULL;
1057
1058 err = -EINVAL;
1059 if (sb->s_flags & SB_MANDLOCK)
1060 goto err;
1061
1062 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1063
1064 if (!parse_fuse_opt(data, &d, is_bdev))
1065 goto err;
1066
1067 if (is_bdev) {
1068#ifdef CONFIG_BLOCK
1069 err = -EINVAL;
1070 if (!sb_set_blocksize(sb, d.blksize))
1071 goto err;
1072#endif
1073 } else {
1074 sb->s_blocksize = PAGE_SIZE;
1075 sb->s_blocksize_bits = PAGE_SHIFT;
1076 }
1077 sb->s_magic = FUSE_SUPER_MAGIC;
1078 sb->s_op = &fuse_super_operations;
1079 sb->s_xattr = fuse_xattr_handlers;
1080 sb->s_maxbytes = MAX_LFS_FILESIZE;
1081 sb->s_time_gran = 1;
1082 sb->s_export_op = &fuse_export_operations;
1083 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1084 if (sb->s_user_ns != &init_user_ns)
1085 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1086
1087 file = fget(d.fd);
1088 err = -EINVAL;
1089 if (!file)
1090 goto err;
1091
1092 if ((file->f_op != &fuse_dev_operations) ||
1093 (file->f_cred->user_ns != &init_user_ns))
1094 goto err_fput;
1095
1096 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1097 err = -ENOMEM;
1098 if (!fc)
1099 goto err_fput;
1100
1101 fuse_conn_init(fc);
1102 fc->release = fuse_free_conn;
1103
1104 fud = fuse_dev_alloc(fc);
1105 if (!fud)
1106 goto err_put_conn;
1107
1108 fc->dev = sb->s_dev;
1109 fc->sb = sb;
1110 err = fuse_bdi_init(fc, sb);
1111 if (err)
1112 goto err_dev_free;
1113
1114 /* Handle umasking inside the fuse code */
1115 if (sb->s_flags & SB_POSIXACL)
1116 fc->dont_mask = 1;
1117 sb->s_flags |= SB_POSIXACL;
1118
1119 fc->default_permissions = d.default_permissions;
1120 fc->allow_other = d.allow_other;
1121 fc->user_id = d.user_id;
1122 fc->group_id = d.group_id;
1123 fc->max_read = max_t(unsigned, 4096, d.max_read);
1124
1125 /* Used by get_root_inode() */
1126 sb->s_fs_info = fc;
1127
1128 err = -ENOMEM;
1129 root = fuse_get_root_inode(sb, d.rootmode);
1130 sb->s_d_op = &fuse_root_dentry_operations;
1131 root_dentry = d_make_root(root);
1132 if (!root_dentry)
1133 goto err_dev_free;
1134 /* Root dentry doesn't have .d_revalidate */
1135 sb->s_d_op = &fuse_dentry_operations;
1136
1137 init_req = fuse_request_alloc(0);
1138 if (!init_req)
1139 goto err_put_root;
1140 __set_bit(FR_BACKGROUND, &init_req->flags);
1141
1142 if (is_bdev) {
1143 fc->destroy_req = fuse_request_alloc(0);
1144 if (!fc->destroy_req)
1145 goto err_free_init_req;
1146 }
1147
1148 mutex_lock(&fuse_mutex);
1149 err = -EINVAL;
1150 if (file->private_data)
1151 goto err_unlock;
1152
1153 err = fuse_ctl_add_conn(fc);
1154 if (err)
1155 goto err_unlock;
1156
1157 list_add_tail(&fc->entry, &fuse_conn_list);
1158 sb->s_root = root_dentry;
1159 file->private_data = fud;
1160 mutex_unlock(&fuse_mutex);
1161 /*
1162 * atomic_dec_and_test() in fput() provides the necessary
1163 * memory barrier for file->private_data to be visible on all
1164 * CPUs after this
1165 */
1166 fput(file);
1167
1168 fuse_send_init(fc, init_req);
1169
1170 return 0;
1171
1172 err_unlock:
1173 mutex_unlock(&fuse_mutex);
1174 err_free_init_req:
1175 fuse_request_free(init_req);
1176 err_put_root:
1177 dput(root_dentry);
1178 err_dev_free:
1179 fuse_dev_free(fud);
1180 err_put_conn:
1181 fuse_conn_put(fc);
1182 err_fput:
1183 fput(file);
1184 err:
1185 return err;
1186}
1187
1188static struct dentry *fuse_mount(struct file_system_type *fs_type,
1189 int flags, const char *dev_name,
1190 void *raw_data)
1191{
1192 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1193}
1194
1195static void fuse_kill_sb_anon(struct super_block *sb)
1196{
1197 struct fuse_conn *fc = get_fuse_conn_super(sb);
1198
1199 if (fc) {
1200 down_write(&fc->killsb);
1201 fc->sb = NULL;
1202 up_write(&fc->killsb);
1203 }
1204
1205 kill_anon_super(sb);
1206}
1207
1208static struct file_system_type fuse_fs_type = {
1209 .owner = THIS_MODULE,
1210 .name = "fuse",
1211 .fs_flags = FS_HAS_SUBTYPE,
1212 .mount = fuse_mount,
1213 .kill_sb = fuse_kill_sb_anon,
1214};
1215MODULE_ALIAS_FS("fuse");
1216
1217#ifdef CONFIG_BLOCK
1218static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1219 int flags, const char *dev_name,
1220 void *raw_data)
1221{
1222 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1223}
1224
1225static void fuse_kill_sb_blk(struct super_block *sb)
1226{
1227 struct fuse_conn *fc = get_fuse_conn_super(sb);
1228
1229 if (fc) {
1230 down_write(&fc->killsb);
1231 fc->sb = NULL;
1232 up_write(&fc->killsb);
1233 }
1234
1235 kill_block_super(sb);
1236}
1237
1238static struct file_system_type fuseblk_fs_type = {
1239 .owner = THIS_MODULE,
1240 .name = "fuseblk",
1241 .mount = fuse_mount_blk,
1242 .kill_sb = fuse_kill_sb_blk,
1243 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1244};
1245MODULE_ALIAS_FS("fuseblk");
1246
1247static inline int register_fuseblk(void)
1248{
1249 return register_filesystem(&fuseblk_fs_type);
1250}
1251
1252static inline void unregister_fuseblk(void)
1253{
1254 unregister_filesystem(&fuseblk_fs_type);
1255}
1256#else
1257static inline int register_fuseblk(void)
1258{
1259 return 0;
1260}
1261
1262static inline void unregister_fuseblk(void)
1263{
1264}
1265#endif
1266
1267static void fuse_inode_init_once(void *foo)
1268{
1269 struct inode *inode = foo;
1270
1271 inode_init_once(inode);
1272}
1273
1274static int __init fuse_fs_init(void)
1275{
1276 int err;
1277
1278 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1279 sizeof(struct fuse_inode), 0,
1280 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1281 fuse_inode_init_once);
1282 err = -ENOMEM;
1283 if (!fuse_inode_cachep)
1284 goto out;
1285
1286 err = register_fuseblk();
1287 if (err)
1288 goto out2;
1289
1290 err = register_filesystem(&fuse_fs_type);
1291 if (err)
1292 goto out3;
1293
1294 return 0;
1295
1296 out3:
1297 unregister_fuseblk();
1298 out2:
1299 kmem_cache_destroy(fuse_inode_cachep);
1300 out:
1301 return err;
1302}
1303
1304static void fuse_fs_cleanup(void)
1305{
1306 unregister_filesystem(&fuse_fs_type);
1307 unregister_fuseblk();
1308
1309 /*
1310 * Make sure all delayed rcu free inodes are flushed before we
1311 * destroy cache.
1312 */
1313 rcu_barrier();
1314 kmem_cache_destroy(fuse_inode_cachep);
1315}
1316
1317static struct kobject *fuse_kobj;
1318
1319static int fuse_sysfs_init(void)
1320{
1321 int err;
1322
1323 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1324 if (!fuse_kobj) {
1325 err = -ENOMEM;
1326 goto out_err;
1327 }
1328
1329 err = sysfs_create_mount_point(fuse_kobj, "connections");
1330 if (err)
1331 goto out_fuse_unregister;
1332
1333 return 0;
1334
1335 out_fuse_unregister:
1336 kobject_put(fuse_kobj);
1337 out_err:
1338 return err;
1339}
1340
1341static void fuse_sysfs_cleanup(void)
1342{
1343 sysfs_remove_mount_point(fuse_kobj, "connections");
1344 kobject_put(fuse_kobj);
1345}
1346
1347static int __init fuse_init(void)
1348{
1349 int res;
1350
1351 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1352 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1353
1354 INIT_LIST_HEAD(&fuse_conn_list);
1355 res = fuse_fs_init();
1356 if (res)
1357 goto err;
1358
1359 res = fuse_dev_init();
1360 if (res)
1361 goto err_fs_cleanup;
1362
1363 res = fuse_sysfs_init();
1364 if (res)
1365 goto err_dev_cleanup;
1366
1367 res = fuse_ctl_init();
1368 if (res)
1369 goto err_sysfs_cleanup;
1370
1371 sanitize_global_limit(&max_user_bgreq);
1372 sanitize_global_limit(&max_user_congthresh);
1373
1374 return 0;
1375
1376 err_sysfs_cleanup:
1377 fuse_sysfs_cleanup();
1378 err_dev_cleanup:
1379 fuse_dev_cleanup();
1380 err_fs_cleanup:
1381 fuse_fs_cleanup();
1382 err:
1383 return res;
1384}
1385
1386static void __exit fuse_exit(void)
1387{
1388 printk(KERN_DEBUG "fuse exit\n");
1389
1390 fuse_ctl_cleanup();
1391 fuse_sysfs_cleanup();
1392 fuse_fs_cleanup();
1393 fuse_dev_cleanup();
1394}
1395
1396module_init(fuse_init);
1397module_exit(fuse_exit);