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/fs_context.h>
19#include <linux/fs_parser.h>
20#include <linux/statfs.h>
21#include <linux/random.h>
22#include <linux/sched.h>
23#include <linux/exportfs.h>
24#include <linux/posix_acl.h>
25#include <linux/pid_namespace.h>
26
27MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
28MODULE_DESCRIPTION("Filesystem in Userspace");
29MODULE_LICENSE("GPL");
30
31static struct kmem_cache *fuse_inode_cachep;
32struct list_head fuse_conn_list;
33DEFINE_MUTEX(fuse_mutex);
34
35static int set_global_limit(const char *val, const struct kernel_param *kp);
36
37unsigned max_user_bgreq;
38module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
39 &max_user_bgreq, 0644);
40__MODULE_PARM_TYPE(max_user_bgreq, "uint");
41MODULE_PARM_DESC(max_user_bgreq,
42 "Global limit for the maximum number of backgrounded requests an "
43 "unprivileged user can set");
44
45unsigned max_user_congthresh;
46module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
47 &max_user_congthresh, 0644);
48__MODULE_PARM_TYPE(max_user_congthresh, "uint");
49MODULE_PARM_DESC(max_user_congthresh,
50 "Global limit for the maximum congestion threshold an "
51 "unprivileged user can set");
52
53#define FUSE_SUPER_MAGIC 0x65735546
54
55#define FUSE_DEFAULT_BLKSIZE 512
56
57/** Maximum number of outstanding background requests */
58#define FUSE_DEFAULT_MAX_BACKGROUND 12
59
60/** Congestion starts at 75% of maximum */
61#define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
62
63#ifdef CONFIG_BLOCK
64static struct file_system_type fuseblk_fs_type;
65#endif
66
67struct fuse_forget_link *fuse_alloc_forget(void)
68{
69 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
70}
71
72static struct inode *fuse_alloc_inode(struct super_block *sb)
73{
74 struct fuse_inode *fi;
75
76 fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
77 if (!fi)
78 return NULL;
79
80 fi->i_time = 0;
81 fi->inval_mask = 0;
82 fi->nodeid = 0;
83 fi->nlookup = 0;
84 fi->attr_version = 0;
85 fi->orig_ino = 0;
86 fi->state = 0;
87 mutex_init(&fi->mutex);
88 spin_lock_init(&fi->lock);
89 fi->forget = fuse_alloc_forget();
90 if (!fi->forget) {
91 kmem_cache_free(fuse_inode_cachep, fi);
92 return NULL;
93 }
94
95 return &fi->inode;
96}
97
98static void fuse_free_inode(struct inode *inode)
99{
100 struct fuse_inode *fi = get_fuse_inode(inode);
101
102 mutex_destroy(&fi->mutex);
103 kfree(fi->forget);
104 kmem_cache_free(fuse_inode_cachep, fi);
105}
106
107static void fuse_evict_inode(struct inode *inode)
108{
109 struct fuse_inode *fi = get_fuse_inode(inode);
110
111 truncate_inode_pages_final(&inode->i_data);
112 clear_inode(inode);
113 if (inode->i_sb->s_flags & SB_ACTIVE) {
114 struct fuse_conn *fc = get_fuse_conn(inode);
115 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
116 fi->forget = NULL;
117 }
118 if (S_ISREG(inode->i_mode) && !is_bad_inode(inode)) {
119 WARN_ON(!list_empty(&fi->write_files));
120 WARN_ON(!list_empty(&fi->queued_writes));
121 }
122}
123
124static int fuse_reconfigure(struct fs_context *fc)
125{
126 struct super_block *sb = fc->root->d_sb;
127
128 sync_filesystem(sb);
129 if (fc->sb_flags & SB_MANDLOCK)
130 return -EINVAL;
131
132 return 0;
133}
134
135/*
136 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
137 * so that it will fit.
138 */
139static ino_t fuse_squash_ino(u64 ino64)
140{
141 ino_t ino = (ino_t) ino64;
142 if (sizeof(ino_t) < sizeof(u64))
143 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
144 return ino;
145}
146
147void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
148 u64 attr_valid)
149{
150 struct fuse_conn *fc = get_fuse_conn(inode);
151 struct fuse_inode *fi = get_fuse_inode(inode);
152
153 lockdep_assert_held(&fi->lock);
154
155 fi->attr_version = atomic64_inc_return(&fc->attr_version);
156 fi->i_time = attr_valid;
157 WRITE_ONCE(fi->inval_mask, 0);
158
159 inode->i_ino = fuse_squash_ino(attr->ino);
160 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
161 set_nlink(inode, attr->nlink);
162 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
163 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
164 inode->i_blocks = attr->blocks;
165 inode->i_atime.tv_sec = attr->atime;
166 inode->i_atime.tv_nsec = attr->atimensec;
167 /* mtime from server may be stale due to local buffered write */
168 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
169 inode->i_mtime.tv_sec = attr->mtime;
170 inode->i_mtime.tv_nsec = attr->mtimensec;
171 inode->i_ctime.tv_sec = attr->ctime;
172 inode->i_ctime.tv_nsec = attr->ctimensec;
173 }
174
175 if (attr->blksize != 0)
176 inode->i_blkbits = ilog2(attr->blksize);
177 else
178 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
179
180 /*
181 * Don't set the sticky bit in i_mode, unless we want the VFS
182 * to check permissions. This prevents failures due to the
183 * check in may_delete().
184 */
185 fi->orig_i_mode = inode->i_mode;
186 if (!fc->default_permissions)
187 inode->i_mode &= ~S_ISVTX;
188
189 fi->orig_ino = attr->ino;
190}
191
192void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
193 u64 attr_valid, u64 attr_version)
194{
195 struct fuse_conn *fc = get_fuse_conn(inode);
196 struct fuse_inode *fi = get_fuse_inode(inode);
197 bool is_wb = fc->writeback_cache;
198 loff_t oldsize;
199 struct timespec64 old_mtime;
200
201 spin_lock(&fi->lock);
202 if ((attr_version != 0 && fi->attr_version > attr_version) ||
203 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
204 spin_unlock(&fi->lock);
205 return;
206 }
207
208 old_mtime = inode->i_mtime;
209 fuse_change_attributes_common(inode, attr, attr_valid);
210
211 oldsize = inode->i_size;
212 /*
213 * In case of writeback_cache enabled, the cached writes beyond EOF
214 * extend local i_size without keeping userspace server in sync. So,
215 * attr->size coming from server can be stale. We cannot trust it.
216 */
217 if (!is_wb || !S_ISREG(inode->i_mode))
218 i_size_write(inode, attr->size);
219 spin_unlock(&fi->lock);
220
221 if (!is_wb && S_ISREG(inode->i_mode)) {
222 bool inval = false;
223
224 if (oldsize != attr->size) {
225 truncate_pagecache(inode, attr->size);
226 if (!fc->explicit_inval_data)
227 inval = true;
228 } else if (fc->auto_inval_data) {
229 struct timespec64 new_mtime = {
230 .tv_sec = attr->mtime,
231 .tv_nsec = attr->mtimensec,
232 };
233
234 /*
235 * Auto inval mode also checks and invalidates if mtime
236 * has changed.
237 */
238 if (!timespec64_equal(&old_mtime, &new_mtime))
239 inval = true;
240 }
241
242 if (inval)
243 invalidate_inode_pages2(inode->i_mapping);
244 }
245}
246
247static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
248{
249 inode->i_mode = attr->mode & S_IFMT;
250 inode->i_size = attr->size;
251 inode->i_mtime.tv_sec = attr->mtime;
252 inode->i_mtime.tv_nsec = attr->mtimensec;
253 inode->i_ctime.tv_sec = attr->ctime;
254 inode->i_ctime.tv_nsec = attr->ctimensec;
255 if (S_ISREG(inode->i_mode)) {
256 fuse_init_common(inode);
257 fuse_init_file_inode(inode);
258 } else if (S_ISDIR(inode->i_mode))
259 fuse_init_dir(inode);
260 else if (S_ISLNK(inode->i_mode))
261 fuse_init_symlink(inode);
262 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
263 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
264 fuse_init_common(inode);
265 init_special_inode(inode, inode->i_mode,
266 new_decode_dev(attr->rdev));
267 } else
268 BUG();
269}
270
271int fuse_inode_eq(struct inode *inode, void *_nodeidp)
272{
273 u64 nodeid = *(u64 *) _nodeidp;
274 if (get_node_id(inode) == nodeid)
275 return 1;
276 else
277 return 0;
278}
279
280static int fuse_inode_set(struct inode *inode, void *_nodeidp)
281{
282 u64 nodeid = *(u64 *) _nodeidp;
283 get_fuse_inode(inode)->nodeid = nodeid;
284 return 0;
285}
286
287struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
288 int generation, struct fuse_attr *attr,
289 u64 attr_valid, u64 attr_version)
290{
291 struct inode *inode;
292 struct fuse_inode *fi;
293 struct fuse_conn *fc = get_fuse_conn_super(sb);
294
295 retry:
296 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
297 if (!inode)
298 return NULL;
299
300 if ((inode->i_state & I_NEW)) {
301 inode->i_flags |= S_NOATIME;
302 if (!fc->writeback_cache || !S_ISREG(attr->mode))
303 inode->i_flags |= S_NOCMTIME;
304 inode->i_generation = generation;
305 fuse_init_inode(inode, attr);
306 unlock_new_inode(inode);
307 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
308 /* Inode has changed type, any I/O on the old should fail */
309 make_bad_inode(inode);
310 iput(inode);
311 goto retry;
312 }
313
314 fi = get_fuse_inode(inode);
315 spin_lock(&fi->lock);
316 fi->nlookup++;
317 spin_unlock(&fi->lock);
318 fuse_change_attributes(inode, attr, attr_valid, attr_version);
319
320 return inode;
321}
322
323int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
324 loff_t offset, loff_t len)
325{
326 struct fuse_conn *fc = get_fuse_conn_super(sb);
327 struct fuse_inode *fi;
328 struct inode *inode;
329 pgoff_t pg_start;
330 pgoff_t pg_end;
331
332 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
333 if (!inode)
334 return -ENOENT;
335
336 fi = get_fuse_inode(inode);
337 spin_lock(&fi->lock);
338 fi->attr_version = atomic64_inc_return(&fc->attr_version);
339 spin_unlock(&fi->lock);
340
341 fuse_invalidate_attr(inode);
342 forget_all_cached_acls(inode);
343 if (offset >= 0) {
344 pg_start = offset >> PAGE_SHIFT;
345 if (len <= 0)
346 pg_end = -1;
347 else
348 pg_end = (offset + len - 1) >> PAGE_SHIFT;
349 invalidate_inode_pages2_range(inode->i_mapping,
350 pg_start, pg_end);
351 }
352 iput(inode);
353 return 0;
354}
355
356bool fuse_lock_inode(struct inode *inode)
357{
358 bool locked = false;
359
360 if (!get_fuse_conn(inode)->parallel_dirops) {
361 mutex_lock(&get_fuse_inode(inode)->mutex);
362 locked = true;
363 }
364
365 return locked;
366}
367
368void fuse_unlock_inode(struct inode *inode, bool locked)
369{
370 if (locked)
371 mutex_unlock(&get_fuse_inode(inode)->mutex);
372}
373
374static void fuse_umount_begin(struct super_block *sb)
375{
376 struct fuse_conn *fc = get_fuse_conn_super(sb);
377
378 if (!fc->no_force_umount)
379 fuse_abort_conn(fc);
380}
381
382static void fuse_send_destroy(struct fuse_conn *fc)
383{
384 if (fc->conn_init) {
385 FUSE_ARGS(args);
386
387 args.opcode = FUSE_DESTROY;
388 args.force = true;
389 args.nocreds = true;
390 fuse_simple_request(fc, &args);
391 }
392}
393
394static void fuse_put_super(struct super_block *sb)
395{
396 struct fuse_conn *fc = get_fuse_conn_super(sb);
397
398 mutex_lock(&fuse_mutex);
399 list_del(&fc->entry);
400 fuse_ctl_remove_conn(fc);
401 mutex_unlock(&fuse_mutex);
402
403 fuse_conn_put(fc);
404}
405
406static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
407{
408 stbuf->f_type = FUSE_SUPER_MAGIC;
409 stbuf->f_bsize = attr->bsize;
410 stbuf->f_frsize = attr->frsize;
411 stbuf->f_blocks = attr->blocks;
412 stbuf->f_bfree = attr->bfree;
413 stbuf->f_bavail = attr->bavail;
414 stbuf->f_files = attr->files;
415 stbuf->f_ffree = attr->ffree;
416 stbuf->f_namelen = attr->namelen;
417 /* fsid is left zero */
418}
419
420static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
421{
422 struct super_block *sb = dentry->d_sb;
423 struct fuse_conn *fc = get_fuse_conn_super(sb);
424 FUSE_ARGS(args);
425 struct fuse_statfs_out outarg;
426 int err;
427
428 if (!fuse_allow_current_process(fc)) {
429 buf->f_type = FUSE_SUPER_MAGIC;
430 return 0;
431 }
432
433 memset(&outarg, 0, sizeof(outarg));
434 args.in_numargs = 0;
435 args.opcode = FUSE_STATFS;
436 args.nodeid = get_node_id(d_inode(dentry));
437 args.out_numargs = 1;
438 args.out_args[0].size = sizeof(outarg);
439 args.out_args[0].value = &outarg;
440 err = fuse_simple_request(fc, &args);
441 if (!err)
442 convert_fuse_statfs(buf, &outarg.st);
443 return err;
444}
445
446enum {
447 OPT_SOURCE,
448 OPT_SUBTYPE,
449 OPT_FD,
450 OPT_ROOTMODE,
451 OPT_USER_ID,
452 OPT_GROUP_ID,
453 OPT_DEFAULT_PERMISSIONS,
454 OPT_ALLOW_OTHER,
455 OPT_MAX_READ,
456 OPT_BLKSIZE,
457 OPT_ERR
458};
459
460static const struct fs_parameter_spec fuse_fs_parameters[] = {
461 fsparam_string ("source", OPT_SOURCE),
462 fsparam_u32 ("fd", OPT_FD),
463 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
464 fsparam_u32 ("user_id", OPT_USER_ID),
465 fsparam_u32 ("group_id", OPT_GROUP_ID),
466 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
467 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
468 fsparam_u32 ("max_read", OPT_MAX_READ),
469 fsparam_u32 ("blksize", OPT_BLKSIZE),
470 fsparam_string ("subtype", OPT_SUBTYPE),
471 {}
472};
473
474static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
475{
476 struct fs_parse_result result;
477 struct fuse_fs_context *ctx = fc->fs_private;
478 int opt;
479
480 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
481 /*
482 * Ignore options coming from mount(MS_REMOUNT) for backward
483 * compatibility.
484 */
485 if (fc->oldapi)
486 return 0;
487
488 return invalfc(fc, "No changes allowed in reconfigure");
489 }
490
491 opt = fs_parse(fc, fuse_fs_parameters, param, &result);
492 if (opt < 0)
493 return opt;
494
495 switch (opt) {
496 case OPT_SOURCE:
497 if (fc->source)
498 return invalfc(fc, "Multiple sources specified");
499 fc->source = param->string;
500 param->string = NULL;
501 break;
502
503 case OPT_SUBTYPE:
504 if (ctx->subtype)
505 return invalfc(fc, "Multiple subtypes specified");
506 ctx->subtype = param->string;
507 param->string = NULL;
508 return 0;
509
510 case OPT_FD:
511 ctx->fd = result.uint_32;
512 ctx->fd_present = true;
513 break;
514
515 case OPT_ROOTMODE:
516 if (!fuse_valid_type(result.uint_32))
517 return invalfc(fc, "Invalid rootmode");
518 ctx->rootmode = result.uint_32;
519 ctx->rootmode_present = true;
520 break;
521
522 case OPT_USER_ID:
523 ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
524 if (!uid_valid(ctx->user_id))
525 return invalfc(fc, "Invalid user_id");
526 ctx->user_id_present = true;
527 break;
528
529 case OPT_GROUP_ID:
530 ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
531 if (!gid_valid(ctx->group_id))
532 return invalfc(fc, "Invalid group_id");
533 ctx->group_id_present = true;
534 break;
535
536 case OPT_DEFAULT_PERMISSIONS:
537 ctx->default_permissions = true;
538 break;
539
540 case OPT_ALLOW_OTHER:
541 ctx->allow_other = true;
542 break;
543
544 case OPT_MAX_READ:
545 ctx->max_read = result.uint_32;
546 break;
547
548 case OPT_BLKSIZE:
549 if (!ctx->is_bdev)
550 return invalfc(fc, "blksize only supported for fuseblk");
551 ctx->blksize = result.uint_32;
552 break;
553
554 default:
555 return -EINVAL;
556 }
557
558 return 0;
559}
560
561static void fuse_free_fc(struct fs_context *fc)
562{
563 struct fuse_fs_context *ctx = fc->fs_private;
564
565 if (ctx) {
566 kfree(ctx->subtype);
567 kfree(ctx);
568 }
569}
570
571static int fuse_show_options(struct seq_file *m, struct dentry *root)
572{
573 struct super_block *sb = root->d_sb;
574 struct fuse_conn *fc = get_fuse_conn_super(sb);
575
576 if (fc->no_mount_options)
577 return 0;
578
579 seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
580 seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
581 if (fc->default_permissions)
582 seq_puts(m, ",default_permissions");
583 if (fc->allow_other)
584 seq_puts(m, ",allow_other");
585 if (fc->max_read != ~0)
586 seq_printf(m, ",max_read=%u", fc->max_read);
587 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
588 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
589 return 0;
590}
591
592static void fuse_iqueue_init(struct fuse_iqueue *fiq,
593 const struct fuse_iqueue_ops *ops,
594 void *priv)
595{
596 memset(fiq, 0, sizeof(struct fuse_iqueue));
597 spin_lock_init(&fiq->lock);
598 init_waitqueue_head(&fiq->waitq);
599 INIT_LIST_HEAD(&fiq->pending);
600 INIT_LIST_HEAD(&fiq->interrupts);
601 fiq->forget_list_tail = &fiq->forget_list_head;
602 fiq->connected = 1;
603 fiq->ops = ops;
604 fiq->priv = priv;
605}
606
607static void fuse_pqueue_init(struct fuse_pqueue *fpq)
608{
609 unsigned int i;
610
611 spin_lock_init(&fpq->lock);
612 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
613 INIT_LIST_HEAD(&fpq->processing[i]);
614 INIT_LIST_HEAD(&fpq->io);
615 fpq->connected = 1;
616}
617
618void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns,
619 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
620{
621 memset(fc, 0, sizeof(*fc));
622 spin_lock_init(&fc->lock);
623 spin_lock_init(&fc->bg_lock);
624 init_rwsem(&fc->killsb);
625 refcount_set(&fc->count, 1);
626 atomic_set(&fc->dev_count, 1);
627 init_waitqueue_head(&fc->blocked_waitq);
628 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
629 INIT_LIST_HEAD(&fc->bg_queue);
630 INIT_LIST_HEAD(&fc->entry);
631 INIT_LIST_HEAD(&fc->devices);
632 atomic_set(&fc->num_waiting, 0);
633 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
634 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
635 atomic64_set(&fc->khctr, 0);
636 fc->polled_files = RB_ROOT;
637 fc->blocked = 0;
638 fc->initialized = 0;
639 fc->connected = 1;
640 atomic64_set(&fc->attr_version, 1);
641 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
642 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
643 fc->user_ns = get_user_ns(user_ns);
644 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
645}
646EXPORT_SYMBOL_GPL(fuse_conn_init);
647
648void fuse_conn_put(struct fuse_conn *fc)
649{
650 if (refcount_dec_and_test(&fc->count)) {
651 struct fuse_iqueue *fiq = &fc->iq;
652
653 if (fiq->ops->release)
654 fiq->ops->release(fiq);
655 put_pid_ns(fc->pid_ns);
656 put_user_ns(fc->user_ns);
657 fc->release(fc);
658 }
659}
660EXPORT_SYMBOL_GPL(fuse_conn_put);
661
662struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
663{
664 refcount_inc(&fc->count);
665 return fc;
666}
667EXPORT_SYMBOL_GPL(fuse_conn_get);
668
669static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
670{
671 struct fuse_attr attr;
672 memset(&attr, 0, sizeof(attr));
673
674 attr.mode = mode;
675 attr.ino = FUSE_ROOT_ID;
676 attr.nlink = 1;
677 return fuse_iget(sb, 1, 0, &attr, 0, 0);
678}
679
680struct fuse_inode_handle {
681 u64 nodeid;
682 u32 generation;
683};
684
685static struct dentry *fuse_get_dentry(struct super_block *sb,
686 struct fuse_inode_handle *handle)
687{
688 struct fuse_conn *fc = get_fuse_conn_super(sb);
689 struct inode *inode;
690 struct dentry *entry;
691 int err = -ESTALE;
692
693 if (handle->nodeid == 0)
694 goto out_err;
695
696 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
697 if (!inode) {
698 struct fuse_entry_out outarg;
699 const struct qstr name = QSTR_INIT(".", 1);
700
701 if (!fc->export_support)
702 goto out_err;
703
704 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
705 &inode);
706 if (err && err != -ENOENT)
707 goto out_err;
708 if (err || !inode) {
709 err = -ESTALE;
710 goto out_err;
711 }
712 err = -EIO;
713 if (get_node_id(inode) != handle->nodeid)
714 goto out_iput;
715 }
716 err = -ESTALE;
717 if (inode->i_generation != handle->generation)
718 goto out_iput;
719
720 entry = d_obtain_alias(inode);
721 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
722 fuse_invalidate_entry_cache(entry);
723
724 return entry;
725
726 out_iput:
727 iput(inode);
728 out_err:
729 return ERR_PTR(err);
730}
731
732static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
733 struct inode *parent)
734{
735 int len = parent ? 6 : 3;
736 u64 nodeid;
737 u32 generation;
738
739 if (*max_len < len) {
740 *max_len = len;
741 return FILEID_INVALID;
742 }
743
744 nodeid = get_fuse_inode(inode)->nodeid;
745 generation = inode->i_generation;
746
747 fh[0] = (u32)(nodeid >> 32);
748 fh[1] = (u32)(nodeid & 0xffffffff);
749 fh[2] = generation;
750
751 if (parent) {
752 nodeid = get_fuse_inode(parent)->nodeid;
753 generation = parent->i_generation;
754
755 fh[3] = (u32)(nodeid >> 32);
756 fh[4] = (u32)(nodeid & 0xffffffff);
757 fh[5] = generation;
758 }
759
760 *max_len = len;
761 return parent ? 0x82 : 0x81;
762}
763
764static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
765 struct fid *fid, int fh_len, int fh_type)
766{
767 struct fuse_inode_handle handle;
768
769 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
770 return NULL;
771
772 handle.nodeid = (u64) fid->raw[0] << 32;
773 handle.nodeid |= (u64) fid->raw[1];
774 handle.generation = fid->raw[2];
775 return fuse_get_dentry(sb, &handle);
776}
777
778static struct dentry *fuse_fh_to_parent(struct super_block *sb,
779 struct fid *fid, int fh_len, int fh_type)
780{
781 struct fuse_inode_handle parent;
782
783 if (fh_type != 0x82 || fh_len < 6)
784 return NULL;
785
786 parent.nodeid = (u64) fid->raw[3] << 32;
787 parent.nodeid |= (u64) fid->raw[4];
788 parent.generation = fid->raw[5];
789 return fuse_get_dentry(sb, &parent);
790}
791
792static struct dentry *fuse_get_parent(struct dentry *child)
793{
794 struct inode *child_inode = d_inode(child);
795 struct fuse_conn *fc = get_fuse_conn(child_inode);
796 struct inode *inode;
797 struct dentry *parent;
798 struct fuse_entry_out outarg;
799 const struct qstr name = QSTR_INIT("..", 2);
800 int err;
801
802 if (!fc->export_support)
803 return ERR_PTR(-ESTALE);
804
805 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
806 &name, &outarg, &inode);
807 if (err) {
808 if (err == -ENOENT)
809 return ERR_PTR(-ESTALE);
810 return ERR_PTR(err);
811 }
812
813 parent = d_obtain_alias(inode);
814 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
815 fuse_invalidate_entry_cache(parent);
816
817 return parent;
818}
819
820static const struct export_operations fuse_export_operations = {
821 .fh_to_dentry = fuse_fh_to_dentry,
822 .fh_to_parent = fuse_fh_to_parent,
823 .encode_fh = fuse_encode_fh,
824 .get_parent = fuse_get_parent,
825};
826
827static const struct super_operations fuse_super_operations = {
828 .alloc_inode = fuse_alloc_inode,
829 .free_inode = fuse_free_inode,
830 .evict_inode = fuse_evict_inode,
831 .write_inode = fuse_write_inode,
832 .drop_inode = generic_delete_inode,
833 .put_super = fuse_put_super,
834 .umount_begin = fuse_umount_begin,
835 .statfs = fuse_statfs,
836 .show_options = fuse_show_options,
837};
838
839static void sanitize_global_limit(unsigned *limit)
840{
841 /*
842 * The default maximum number of async requests is calculated to consume
843 * 1/2^13 of the total memory, assuming 392 bytes per request.
844 */
845 if (*limit == 0)
846 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
847
848 if (*limit >= 1 << 16)
849 *limit = (1 << 16) - 1;
850}
851
852static int set_global_limit(const char *val, const struct kernel_param *kp)
853{
854 int rv;
855
856 rv = param_set_uint(val, kp);
857 if (rv)
858 return rv;
859
860 sanitize_global_limit((unsigned *)kp->arg);
861
862 return 0;
863}
864
865static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
866{
867 int cap_sys_admin = capable(CAP_SYS_ADMIN);
868
869 if (arg->minor < 13)
870 return;
871
872 sanitize_global_limit(&max_user_bgreq);
873 sanitize_global_limit(&max_user_congthresh);
874
875 spin_lock(&fc->bg_lock);
876 if (arg->max_background) {
877 fc->max_background = arg->max_background;
878
879 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
880 fc->max_background = max_user_bgreq;
881 }
882 if (arg->congestion_threshold) {
883 fc->congestion_threshold = arg->congestion_threshold;
884
885 if (!cap_sys_admin &&
886 fc->congestion_threshold > max_user_congthresh)
887 fc->congestion_threshold = max_user_congthresh;
888 }
889 spin_unlock(&fc->bg_lock);
890}
891
892struct fuse_init_args {
893 struct fuse_args args;
894 struct fuse_init_in in;
895 struct fuse_init_out out;
896};
897
898static void process_init_reply(struct fuse_conn *fc, struct fuse_args *args,
899 int error)
900{
901 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
902 struct fuse_init_out *arg = &ia->out;
903
904 if (error || arg->major != FUSE_KERNEL_VERSION)
905 fc->conn_error = 1;
906 else {
907 unsigned long ra_pages;
908
909 process_init_limits(fc, arg);
910
911 if (arg->minor >= 6) {
912 ra_pages = arg->max_readahead / PAGE_SIZE;
913 if (arg->flags & FUSE_ASYNC_READ)
914 fc->async_read = 1;
915 if (!(arg->flags & FUSE_POSIX_LOCKS))
916 fc->no_lock = 1;
917 if (arg->minor >= 17) {
918 if (!(arg->flags & FUSE_FLOCK_LOCKS))
919 fc->no_flock = 1;
920 } else {
921 if (!(arg->flags & FUSE_POSIX_LOCKS))
922 fc->no_flock = 1;
923 }
924 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
925 fc->atomic_o_trunc = 1;
926 if (arg->minor >= 9) {
927 /* LOOKUP has dependency on proto version */
928 if (arg->flags & FUSE_EXPORT_SUPPORT)
929 fc->export_support = 1;
930 }
931 if (arg->flags & FUSE_BIG_WRITES)
932 fc->big_writes = 1;
933 if (arg->flags & FUSE_DONT_MASK)
934 fc->dont_mask = 1;
935 if (arg->flags & FUSE_AUTO_INVAL_DATA)
936 fc->auto_inval_data = 1;
937 else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
938 fc->explicit_inval_data = 1;
939 if (arg->flags & FUSE_DO_READDIRPLUS) {
940 fc->do_readdirplus = 1;
941 if (arg->flags & FUSE_READDIRPLUS_AUTO)
942 fc->readdirplus_auto = 1;
943 }
944 if (arg->flags & FUSE_ASYNC_DIO)
945 fc->async_dio = 1;
946 if (arg->flags & FUSE_WRITEBACK_CACHE)
947 fc->writeback_cache = 1;
948 if (arg->flags & FUSE_PARALLEL_DIROPS)
949 fc->parallel_dirops = 1;
950 if (arg->flags & FUSE_HANDLE_KILLPRIV)
951 fc->handle_killpriv = 1;
952 if (arg->time_gran && arg->time_gran <= 1000000000)
953 fc->sb->s_time_gran = arg->time_gran;
954 if ((arg->flags & FUSE_POSIX_ACL)) {
955 fc->default_permissions = 1;
956 fc->posix_acl = 1;
957 fc->sb->s_xattr = fuse_acl_xattr_handlers;
958 }
959 if (arg->flags & FUSE_CACHE_SYMLINKS)
960 fc->cache_symlinks = 1;
961 if (arg->flags & FUSE_ABORT_ERROR)
962 fc->abort_err = 1;
963 if (arg->flags & FUSE_MAX_PAGES) {
964 fc->max_pages =
965 min_t(unsigned int, FUSE_MAX_MAX_PAGES,
966 max_t(unsigned int, arg->max_pages, 1));
967 }
968 } else {
969 ra_pages = fc->max_read / PAGE_SIZE;
970 fc->no_lock = 1;
971 fc->no_flock = 1;
972 }
973
974 fc->sb->s_bdi->ra_pages =
975 min(fc->sb->s_bdi->ra_pages, ra_pages);
976 fc->minor = arg->minor;
977 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
978 fc->max_write = max_t(unsigned, 4096, fc->max_write);
979 fc->conn_init = 1;
980 }
981 kfree(ia);
982
983 fuse_set_initialized(fc);
984 wake_up_all(&fc->blocked_waitq);
985}
986
987void fuse_send_init(struct fuse_conn *fc)
988{
989 struct fuse_init_args *ia;
990
991 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
992
993 ia->in.major = FUSE_KERNEL_VERSION;
994 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
995 ia->in.max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
996 ia->in.flags |=
997 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
998 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
999 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1000 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1001 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1002 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1003 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1004 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1005 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
1006 ia->args.opcode = FUSE_INIT;
1007 ia->args.in_numargs = 1;
1008 ia->args.in_args[0].size = sizeof(ia->in);
1009 ia->args.in_args[0].value = &ia->in;
1010 ia->args.out_numargs = 1;
1011 /* Variable length argument used for backward compatibility
1012 with interface version < 7.5. Rest of init_out is zeroed
1013 by do_get_request(), so a short reply is not a problem */
1014 ia->args.out_argvar = true;
1015 ia->args.out_args[0].size = sizeof(ia->out);
1016 ia->args.out_args[0].value = &ia->out;
1017 ia->args.force = true;
1018 ia->args.nocreds = true;
1019 ia->args.end = process_init_reply;
1020
1021 if (fuse_simple_background(fc, &ia->args, GFP_KERNEL) != 0)
1022 process_init_reply(fc, &ia->args, -ENOTCONN);
1023}
1024EXPORT_SYMBOL_GPL(fuse_send_init);
1025
1026void fuse_free_conn(struct fuse_conn *fc)
1027{
1028 WARN_ON(!list_empty(&fc->devices));
1029 kfree_rcu(fc, rcu);
1030}
1031EXPORT_SYMBOL_GPL(fuse_free_conn);
1032
1033static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1034{
1035 int err;
1036 char *suffix = "";
1037
1038 if (sb->s_bdev) {
1039 suffix = "-fuseblk";
1040 /*
1041 * sb->s_bdi points to blkdev's bdi however we want to redirect
1042 * it to our private bdi...
1043 */
1044 bdi_put(sb->s_bdi);
1045 sb->s_bdi = &noop_backing_dev_info;
1046 }
1047 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1048 MINOR(fc->dev), suffix);
1049 if (err)
1050 return err;
1051
1052 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1053 /* fuse does it's own writeback accounting */
1054 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1055
1056 /*
1057 * For a single fuse filesystem use max 1% of dirty +
1058 * writeback threshold.
1059 *
1060 * This gives about 1M of write buffer for memory maps on a
1061 * machine with 1G and 10% dirty_ratio, which should be more
1062 * than enough.
1063 *
1064 * Privileged users can raise it by writing to
1065 *
1066 * /sys/class/bdi/<bdi>/max_ratio
1067 */
1068 bdi_set_max_ratio(sb->s_bdi, 1);
1069
1070 return 0;
1071}
1072
1073struct fuse_dev *fuse_dev_alloc(void)
1074{
1075 struct fuse_dev *fud;
1076 struct list_head *pq;
1077
1078 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1079 if (!fud)
1080 return NULL;
1081
1082 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1083 if (!pq) {
1084 kfree(fud);
1085 return NULL;
1086 }
1087
1088 fud->pq.processing = pq;
1089 fuse_pqueue_init(&fud->pq);
1090
1091 return fud;
1092}
1093EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1094
1095void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1096{
1097 fud->fc = fuse_conn_get(fc);
1098 spin_lock(&fc->lock);
1099 list_add_tail(&fud->entry, &fc->devices);
1100 spin_unlock(&fc->lock);
1101}
1102EXPORT_SYMBOL_GPL(fuse_dev_install);
1103
1104struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1105{
1106 struct fuse_dev *fud;
1107
1108 fud = fuse_dev_alloc();
1109 if (!fud)
1110 return NULL;
1111
1112 fuse_dev_install(fud, fc);
1113 return fud;
1114}
1115EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1116
1117void fuse_dev_free(struct fuse_dev *fud)
1118{
1119 struct fuse_conn *fc = fud->fc;
1120
1121 if (fc) {
1122 spin_lock(&fc->lock);
1123 list_del(&fud->entry);
1124 spin_unlock(&fc->lock);
1125
1126 fuse_conn_put(fc);
1127 }
1128 kfree(fud->pq.processing);
1129 kfree(fud);
1130}
1131EXPORT_SYMBOL_GPL(fuse_dev_free);
1132
1133int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1134{
1135 struct fuse_dev *fud = NULL;
1136 struct fuse_conn *fc = get_fuse_conn_super(sb);
1137 struct inode *root;
1138 struct dentry *root_dentry;
1139 int err;
1140
1141 err = -EINVAL;
1142 if (sb->s_flags & SB_MANDLOCK)
1143 goto err;
1144
1145 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1146
1147 if (ctx->is_bdev) {
1148#ifdef CONFIG_BLOCK
1149 err = -EINVAL;
1150 if (!sb_set_blocksize(sb, ctx->blksize))
1151 goto err;
1152#endif
1153 } else {
1154 sb->s_blocksize = PAGE_SIZE;
1155 sb->s_blocksize_bits = PAGE_SHIFT;
1156 }
1157
1158 sb->s_subtype = ctx->subtype;
1159 ctx->subtype = NULL;
1160 sb->s_magic = FUSE_SUPER_MAGIC;
1161 sb->s_op = &fuse_super_operations;
1162 sb->s_xattr = fuse_xattr_handlers;
1163 sb->s_maxbytes = MAX_LFS_FILESIZE;
1164 sb->s_time_gran = 1;
1165 sb->s_export_op = &fuse_export_operations;
1166 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1167 if (sb->s_user_ns != &init_user_ns)
1168 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1169
1170 /*
1171 * If we are not in the initial user namespace posix
1172 * acls must be translated.
1173 */
1174 if (sb->s_user_ns != &init_user_ns)
1175 sb->s_xattr = fuse_no_acl_xattr_handlers;
1176
1177 if (ctx->fudptr) {
1178 err = -ENOMEM;
1179 fud = fuse_dev_alloc_install(fc);
1180 if (!fud)
1181 goto err;
1182 }
1183
1184 fc->dev = sb->s_dev;
1185 fc->sb = sb;
1186 err = fuse_bdi_init(fc, sb);
1187 if (err)
1188 goto err_dev_free;
1189
1190 /* Handle umasking inside the fuse code */
1191 if (sb->s_flags & SB_POSIXACL)
1192 fc->dont_mask = 1;
1193 sb->s_flags |= SB_POSIXACL;
1194
1195 fc->default_permissions = ctx->default_permissions;
1196 fc->allow_other = ctx->allow_other;
1197 fc->user_id = ctx->user_id;
1198 fc->group_id = ctx->group_id;
1199 fc->max_read = max_t(unsigned, 4096, ctx->max_read);
1200 fc->destroy = ctx->destroy;
1201 fc->no_control = ctx->no_control;
1202 fc->no_force_umount = ctx->no_force_umount;
1203 fc->no_mount_options = ctx->no_mount_options;
1204
1205 err = -ENOMEM;
1206 root = fuse_get_root_inode(sb, ctx->rootmode);
1207 sb->s_d_op = &fuse_root_dentry_operations;
1208 root_dentry = d_make_root(root);
1209 if (!root_dentry)
1210 goto err_dev_free;
1211 /* Root dentry doesn't have .d_revalidate */
1212 sb->s_d_op = &fuse_dentry_operations;
1213
1214 mutex_lock(&fuse_mutex);
1215 err = -EINVAL;
1216 if (ctx->fudptr && *ctx->fudptr)
1217 goto err_unlock;
1218
1219 err = fuse_ctl_add_conn(fc);
1220 if (err)
1221 goto err_unlock;
1222
1223 list_add_tail(&fc->entry, &fuse_conn_list);
1224 sb->s_root = root_dentry;
1225 if (ctx->fudptr)
1226 *ctx->fudptr = fud;
1227 mutex_unlock(&fuse_mutex);
1228 return 0;
1229
1230 err_unlock:
1231 mutex_unlock(&fuse_mutex);
1232 dput(root_dentry);
1233 err_dev_free:
1234 if (fud)
1235 fuse_dev_free(fud);
1236 err:
1237 return err;
1238}
1239EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1240
1241static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1242{
1243 struct fuse_fs_context *ctx = fsc->fs_private;
1244 struct file *file;
1245 int err;
1246 struct fuse_conn *fc;
1247
1248 err = -EINVAL;
1249 file = fget(ctx->fd);
1250 if (!file)
1251 goto err;
1252
1253 /*
1254 * Require mount to happen from the same user namespace which
1255 * opened /dev/fuse to prevent potential attacks.
1256 */
1257 if ((file->f_op != &fuse_dev_operations) ||
1258 (file->f_cred->user_ns != sb->s_user_ns))
1259 goto err_fput;
1260 ctx->fudptr = &file->private_data;
1261
1262 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1263 err = -ENOMEM;
1264 if (!fc)
1265 goto err_fput;
1266
1267 fuse_conn_init(fc, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1268 fc->release = fuse_free_conn;
1269 sb->s_fs_info = fc;
1270
1271 err = fuse_fill_super_common(sb, ctx);
1272 if (err)
1273 goto err_put_conn;
1274 /*
1275 * atomic_dec_and_test() in fput() provides the necessary
1276 * memory barrier for file->private_data to be visible on all
1277 * CPUs after this
1278 */
1279 fput(file);
1280 fuse_send_init(get_fuse_conn_super(sb));
1281 return 0;
1282
1283 err_put_conn:
1284 fuse_conn_put(fc);
1285 sb->s_fs_info = NULL;
1286 err_fput:
1287 fput(file);
1288 err:
1289 return err;
1290}
1291
1292static int fuse_get_tree(struct fs_context *fc)
1293{
1294 struct fuse_fs_context *ctx = fc->fs_private;
1295
1296 if (!ctx->fd_present || !ctx->rootmode_present ||
1297 !ctx->user_id_present || !ctx->group_id_present)
1298 return -EINVAL;
1299
1300#ifdef CONFIG_BLOCK
1301 if (ctx->is_bdev)
1302 return get_tree_bdev(fc, fuse_fill_super);
1303#endif
1304
1305 return get_tree_nodev(fc, fuse_fill_super);
1306}
1307
1308static const struct fs_context_operations fuse_context_ops = {
1309 .free = fuse_free_fc,
1310 .parse_param = fuse_parse_param,
1311 .reconfigure = fuse_reconfigure,
1312 .get_tree = fuse_get_tree,
1313};
1314
1315/*
1316 * Set up the filesystem mount context.
1317 */
1318static int fuse_init_fs_context(struct fs_context *fc)
1319{
1320 struct fuse_fs_context *ctx;
1321
1322 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1323 if (!ctx)
1324 return -ENOMEM;
1325
1326 ctx->max_read = ~0;
1327 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1328
1329#ifdef CONFIG_BLOCK
1330 if (fc->fs_type == &fuseblk_fs_type) {
1331 ctx->is_bdev = true;
1332 ctx->destroy = true;
1333 }
1334#endif
1335
1336 fc->fs_private = ctx;
1337 fc->ops = &fuse_context_ops;
1338 return 0;
1339}
1340
1341static void fuse_sb_destroy(struct super_block *sb)
1342{
1343 struct fuse_conn *fc = get_fuse_conn_super(sb);
1344
1345 if (fc) {
1346 if (fc->destroy)
1347 fuse_send_destroy(fc);
1348
1349 fuse_abort_conn(fc);
1350 fuse_wait_aborted(fc);
1351
1352 down_write(&fc->killsb);
1353 fc->sb = NULL;
1354 up_write(&fc->killsb);
1355 }
1356}
1357
1358void fuse_kill_sb_anon(struct super_block *sb)
1359{
1360 fuse_sb_destroy(sb);
1361 kill_anon_super(sb);
1362}
1363EXPORT_SYMBOL_GPL(fuse_kill_sb_anon);
1364
1365static struct file_system_type fuse_fs_type = {
1366 .owner = THIS_MODULE,
1367 .name = "fuse",
1368 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1369 .init_fs_context = fuse_init_fs_context,
1370 .parameters = fuse_fs_parameters,
1371 .kill_sb = fuse_kill_sb_anon,
1372};
1373MODULE_ALIAS_FS("fuse");
1374
1375#ifdef CONFIG_BLOCK
1376static void fuse_kill_sb_blk(struct super_block *sb)
1377{
1378 fuse_sb_destroy(sb);
1379 kill_block_super(sb);
1380}
1381
1382static struct file_system_type fuseblk_fs_type = {
1383 .owner = THIS_MODULE,
1384 .name = "fuseblk",
1385 .init_fs_context = fuse_init_fs_context,
1386 .parameters = fuse_fs_parameters,
1387 .kill_sb = fuse_kill_sb_blk,
1388 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1389};
1390MODULE_ALIAS_FS("fuseblk");
1391
1392static inline int register_fuseblk(void)
1393{
1394 return register_filesystem(&fuseblk_fs_type);
1395}
1396
1397static inline void unregister_fuseblk(void)
1398{
1399 unregister_filesystem(&fuseblk_fs_type);
1400}
1401#else
1402static inline int register_fuseblk(void)
1403{
1404 return 0;
1405}
1406
1407static inline void unregister_fuseblk(void)
1408{
1409}
1410#endif
1411
1412static void fuse_inode_init_once(void *foo)
1413{
1414 struct inode *inode = foo;
1415
1416 inode_init_once(inode);
1417}
1418
1419static int __init fuse_fs_init(void)
1420{
1421 int err;
1422
1423 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1424 sizeof(struct fuse_inode), 0,
1425 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1426 fuse_inode_init_once);
1427 err = -ENOMEM;
1428 if (!fuse_inode_cachep)
1429 goto out;
1430
1431 err = register_fuseblk();
1432 if (err)
1433 goto out2;
1434
1435 err = register_filesystem(&fuse_fs_type);
1436 if (err)
1437 goto out3;
1438
1439 return 0;
1440
1441 out3:
1442 unregister_fuseblk();
1443 out2:
1444 kmem_cache_destroy(fuse_inode_cachep);
1445 out:
1446 return err;
1447}
1448
1449static void fuse_fs_cleanup(void)
1450{
1451 unregister_filesystem(&fuse_fs_type);
1452 unregister_fuseblk();
1453
1454 /*
1455 * Make sure all delayed rcu free inodes are flushed before we
1456 * destroy cache.
1457 */
1458 rcu_barrier();
1459 kmem_cache_destroy(fuse_inode_cachep);
1460}
1461
1462static struct kobject *fuse_kobj;
1463
1464static int fuse_sysfs_init(void)
1465{
1466 int err;
1467
1468 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1469 if (!fuse_kobj) {
1470 err = -ENOMEM;
1471 goto out_err;
1472 }
1473
1474 err = sysfs_create_mount_point(fuse_kobj, "connections");
1475 if (err)
1476 goto out_fuse_unregister;
1477
1478 return 0;
1479
1480 out_fuse_unregister:
1481 kobject_put(fuse_kobj);
1482 out_err:
1483 return err;
1484}
1485
1486static void fuse_sysfs_cleanup(void)
1487{
1488 sysfs_remove_mount_point(fuse_kobj, "connections");
1489 kobject_put(fuse_kobj);
1490}
1491
1492static int __init fuse_init(void)
1493{
1494 int res;
1495
1496 pr_info("init (API version %i.%i)\n",
1497 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1498
1499 INIT_LIST_HEAD(&fuse_conn_list);
1500 res = fuse_fs_init();
1501 if (res)
1502 goto err;
1503
1504 res = fuse_dev_init();
1505 if (res)
1506 goto err_fs_cleanup;
1507
1508 res = fuse_sysfs_init();
1509 if (res)
1510 goto err_dev_cleanup;
1511
1512 res = fuse_ctl_init();
1513 if (res)
1514 goto err_sysfs_cleanup;
1515
1516 sanitize_global_limit(&max_user_bgreq);
1517 sanitize_global_limit(&max_user_congthresh);
1518
1519 return 0;
1520
1521 err_sysfs_cleanup:
1522 fuse_sysfs_cleanup();
1523 err_dev_cleanup:
1524 fuse_dev_cleanup();
1525 err_fs_cleanup:
1526 fuse_fs_cleanup();
1527 err:
1528 return res;
1529}
1530
1531static void __exit fuse_exit(void)
1532{
1533 pr_debug("exit\n");
1534
1535 fuse_ctl_cleanup();
1536 fuse_sysfs_cleanup();
1537 fuse_fs_cleanup();
1538 fuse_dev_cleanup();
1539}
1540
1541module_init(fuse_init);
1542module_exit(fuse_exit);