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#include <uapi/linux/magic.h>
27
28MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
29MODULE_DESCRIPTION("Filesystem in Userspace");
30MODULE_LICENSE("GPL");
31
32static struct kmem_cache *fuse_inode_cachep;
33struct list_head fuse_conn_list;
34DEFINE_MUTEX(fuse_mutex);
35
36static int set_global_limit(const char *val, const struct kernel_param *kp);
37
38unsigned max_user_bgreq;
39module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
40 &max_user_bgreq, 0644);
41__MODULE_PARM_TYPE(max_user_bgreq, "uint");
42MODULE_PARM_DESC(max_user_bgreq,
43 "Global limit for the maximum number of backgrounded requests an "
44 "unprivileged user can set");
45
46unsigned max_user_congthresh;
47module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
48 &max_user_congthresh, 0644);
49__MODULE_PARM_TYPE(max_user_congthresh, "uint");
50MODULE_PARM_DESC(max_user_congthresh,
51 "Global limit for the maximum congestion threshold an "
52 "unprivileged user can set");
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
62#ifdef CONFIG_BLOCK
63static struct file_system_type fuseblk_fs_type;
64#endif
65
66struct fuse_forget_link *fuse_alloc_forget(void)
67{
68 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
69}
70
71static struct fuse_submount_lookup *fuse_alloc_submount_lookup(void)
72{
73 struct fuse_submount_lookup *sl;
74
75 sl = kzalloc(sizeof(struct fuse_submount_lookup), GFP_KERNEL_ACCOUNT);
76 if (!sl)
77 return NULL;
78 sl->forget = fuse_alloc_forget();
79 if (!sl->forget)
80 goto out_free;
81
82 return sl;
83
84out_free:
85 kfree(sl);
86 return NULL;
87}
88
89static struct inode *fuse_alloc_inode(struct super_block *sb)
90{
91 struct fuse_inode *fi;
92
93 fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL);
94 if (!fi)
95 return NULL;
96
97 fi->i_time = 0;
98 fi->inval_mask = ~0;
99 fi->nodeid = 0;
100 fi->nlookup = 0;
101 fi->attr_version = 0;
102 fi->orig_ino = 0;
103 fi->state = 0;
104 fi->submount_lookup = NULL;
105 mutex_init(&fi->mutex);
106 spin_lock_init(&fi->lock);
107 fi->forget = fuse_alloc_forget();
108 if (!fi->forget)
109 goto out_free;
110
111 if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
112 goto out_free_forget;
113
114 return &fi->inode;
115
116out_free_forget:
117 kfree(fi->forget);
118out_free:
119 kmem_cache_free(fuse_inode_cachep, fi);
120 return NULL;
121}
122
123static void fuse_free_inode(struct inode *inode)
124{
125 struct fuse_inode *fi = get_fuse_inode(inode);
126
127 mutex_destroy(&fi->mutex);
128 kfree(fi->forget);
129#ifdef CONFIG_FUSE_DAX
130 kfree(fi->dax);
131#endif
132 kmem_cache_free(fuse_inode_cachep, fi);
133}
134
135static void fuse_cleanup_submount_lookup(struct fuse_conn *fc,
136 struct fuse_submount_lookup *sl)
137{
138 if (!refcount_dec_and_test(&sl->count))
139 return;
140
141 fuse_queue_forget(fc, sl->forget, sl->nodeid, 1);
142 sl->forget = NULL;
143 kfree(sl);
144}
145
146static void fuse_evict_inode(struct inode *inode)
147{
148 struct fuse_inode *fi = get_fuse_inode(inode);
149
150 /* Will write inode on close/munmap and in all other dirtiers */
151 WARN_ON(inode->i_state & I_DIRTY_INODE);
152
153 truncate_inode_pages_final(&inode->i_data);
154 clear_inode(inode);
155 if (inode->i_sb->s_flags & SB_ACTIVE) {
156 struct fuse_conn *fc = get_fuse_conn(inode);
157
158 if (FUSE_IS_DAX(inode))
159 fuse_dax_inode_cleanup(inode);
160 if (fi->nlookup) {
161 fuse_queue_forget(fc, fi->forget, fi->nodeid,
162 fi->nlookup);
163 fi->forget = NULL;
164 }
165
166 if (fi->submount_lookup) {
167 fuse_cleanup_submount_lookup(fc, fi->submount_lookup);
168 fi->submount_lookup = NULL;
169 }
170 }
171 if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
172 WARN_ON(!list_empty(&fi->write_files));
173 WARN_ON(!list_empty(&fi->queued_writes));
174 }
175}
176
177static int fuse_reconfigure(struct fs_context *fsc)
178{
179 struct super_block *sb = fsc->root->d_sb;
180
181 sync_filesystem(sb);
182 if (fsc->sb_flags & SB_MANDLOCK)
183 return -EINVAL;
184
185 return 0;
186}
187
188/*
189 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
190 * so that it will fit.
191 */
192static ino_t fuse_squash_ino(u64 ino64)
193{
194 ino_t ino = (ino_t) ino64;
195 if (sizeof(ino_t) < sizeof(u64))
196 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
197 return ino;
198}
199
200void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
201 struct fuse_statx *sx,
202 u64 attr_valid, u32 cache_mask)
203{
204 struct fuse_conn *fc = get_fuse_conn(inode);
205 struct fuse_inode *fi = get_fuse_inode(inode);
206
207 lockdep_assert_held(&fi->lock);
208
209 fi->attr_version = atomic64_inc_return(&fc->attr_version);
210 fi->i_time = attr_valid;
211 /* Clear basic stats from invalid mask */
212 set_mask_bits(&fi->inval_mask, STATX_BASIC_STATS, 0);
213
214 inode->i_ino = fuse_squash_ino(attr->ino);
215 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
216 set_nlink(inode, attr->nlink);
217 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
218 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
219 inode->i_blocks = attr->blocks;
220
221 /* Sanitize nsecs */
222 attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
223 attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
224 attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
225
226 inode_set_atime(inode, attr->atime, attr->atimensec);
227 /* mtime from server may be stale due to local buffered write */
228 if (!(cache_mask & STATX_MTIME)) {
229 inode_set_mtime(inode, attr->mtime, attr->mtimensec);
230 }
231 if (!(cache_mask & STATX_CTIME)) {
232 inode_set_ctime(inode, attr->ctime, attr->ctimensec);
233 }
234 if (sx) {
235 /* Sanitize nsecs */
236 sx->btime.tv_nsec =
237 min_t(u32, sx->btime.tv_nsec, NSEC_PER_SEC - 1);
238
239 /*
240 * Btime has been queried, cache is valid (whether or not btime
241 * is available or not) so clear STATX_BTIME from inval_mask.
242 *
243 * Availability of the btime attribute is indicated in
244 * FUSE_I_BTIME
245 */
246 set_mask_bits(&fi->inval_mask, STATX_BTIME, 0);
247 if (sx->mask & STATX_BTIME) {
248 set_bit(FUSE_I_BTIME, &fi->state);
249 fi->i_btime.tv_sec = sx->btime.tv_sec;
250 fi->i_btime.tv_nsec = sx->btime.tv_nsec;
251 }
252 }
253
254 if (attr->blksize != 0)
255 inode->i_blkbits = ilog2(attr->blksize);
256 else
257 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
258
259 /*
260 * Don't set the sticky bit in i_mode, unless we want the VFS
261 * to check permissions. This prevents failures due to the
262 * check in may_delete().
263 */
264 fi->orig_i_mode = inode->i_mode;
265 if (!fc->default_permissions)
266 inode->i_mode &= ~S_ISVTX;
267
268 fi->orig_ino = attr->ino;
269
270 /*
271 * We are refreshing inode data and it is possible that another
272 * client set suid/sgid or security.capability xattr. So clear
273 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
274 * was set or if security.capability xattr was set. But we don't
275 * know if security.capability has been set or not. So clear it
276 * anyway. Its less efficient but should be safe.
277 */
278 inode->i_flags &= ~S_NOSEC;
279}
280
281u32 fuse_get_cache_mask(struct inode *inode)
282{
283 struct fuse_conn *fc = get_fuse_conn(inode);
284
285 if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
286 return 0;
287
288 return STATX_MTIME | STATX_CTIME | STATX_SIZE;
289}
290
291void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
292 struct fuse_statx *sx,
293 u64 attr_valid, u64 attr_version)
294{
295 struct fuse_conn *fc = get_fuse_conn(inode);
296 struct fuse_inode *fi = get_fuse_inode(inode);
297 u32 cache_mask;
298 loff_t oldsize;
299 struct timespec64 old_mtime;
300
301 spin_lock(&fi->lock);
302 /*
303 * In case of writeback_cache enabled, writes update mtime, ctime and
304 * may update i_size. In these cases trust the cached value in the
305 * inode.
306 */
307 cache_mask = fuse_get_cache_mask(inode);
308 if (cache_mask & STATX_SIZE)
309 attr->size = i_size_read(inode);
310
311 if (cache_mask & STATX_MTIME) {
312 attr->mtime = inode_get_mtime_sec(inode);
313 attr->mtimensec = inode_get_mtime_nsec(inode);
314 }
315 if (cache_mask & STATX_CTIME) {
316 attr->ctime = inode_get_ctime_sec(inode);
317 attr->ctimensec = inode_get_ctime_nsec(inode);
318 }
319
320 if ((attr_version != 0 && fi->attr_version > attr_version) ||
321 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
322 spin_unlock(&fi->lock);
323 return;
324 }
325
326 old_mtime = inode_get_mtime(inode);
327 fuse_change_attributes_common(inode, attr, sx, attr_valid, cache_mask);
328
329 oldsize = inode->i_size;
330 /*
331 * In case of writeback_cache enabled, the cached writes beyond EOF
332 * extend local i_size without keeping userspace server in sync. So,
333 * attr->size coming from server can be stale. We cannot trust it.
334 */
335 if (!(cache_mask & STATX_SIZE))
336 i_size_write(inode, attr->size);
337 spin_unlock(&fi->lock);
338
339 if (!cache_mask && S_ISREG(inode->i_mode)) {
340 bool inval = false;
341
342 if (oldsize != attr->size) {
343 truncate_pagecache(inode, attr->size);
344 if (!fc->explicit_inval_data)
345 inval = true;
346 } else if (fc->auto_inval_data) {
347 struct timespec64 new_mtime = {
348 .tv_sec = attr->mtime,
349 .tv_nsec = attr->mtimensec,
350 };
351
352 /*
353 * Auto inval mode also checks and invalidates if mtime
354 * has changed.
355 */
356 if (!timespec64_equal(&old_mtime, &new_mtime))
357 inval = true;
358 }
359
360 if (inval)
361 invalidate_inode_pages2(inode->i_mapping);
362 }
363
364 if (IS_ENABLED(CONFIG_FUSE_DAX))
365 fuse_dax_dontcache(inode, attr->flags);
366}
367
368static void fuse_init_submount_lookup(struct fuse_submount_lookup *sl,
369 u64 nodeid)
370{
371 sl->nodeid = nodeid;
372 refcount_set(&sl->count, 1);
373}
374
375static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
376 struct fuse_conn *fc)
377{
378 inode->i_mode = attr->mode & S_IFMT;
379 inode->i_size = attr->size;
380 inode_set_mtime(inode, attr->mtime, attr->mtimensec);
381 inode_set_ctime(inode, attr->ctime, attr->ctimensec);
382 if (S_ISREG(inode->i_mode)) {
383 fuse_init_common(inode);
384 fuse_init_file_inode(inode, attr->flags);
385 } else if (S_ISDIR(inode->i_mode))
386 fuse_init_dir(inode);
387 else if (S_ISLNK(inode->i_mode))
388 fuse_init_symlink(inode);
389 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
390 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
391 fuse_init_common(inode);
392 init_special_inode(inode, inode->i_mode,
393 new_decode_dev(attr->rdev));
394 } else
395 BUG();
396 /*
397 * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
398 * so they see the exact same behavior as before.
399 */
400 if (!fc->posix_acl)
401 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
402}
403
404static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
405{
406 u64 nodeid = *(u64 *) _nodeidp;
407 if (get_node_id(inode) == nodeid)
408 return 1;
409 else
410 return 0;
411}
412
413static int fuse_inode_set(struct inode *inode, void *_nodeidp)
414{
415 u64 nodeid = *(u64 *) _nodeidp;
416 get_fuse_inode(inode)->nodeid = nodeid;
417 return 0;
418}
419
420struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
421 int generation, struct fuse_attr *attr,
422 u64 attr_valid, u64 attr_version)
423{
424 struct inode *inode;
425 struct fuse_inode *fi;
426 struct fuse_conn *fc = get_fuse_conn_super(sb);
427
428 /*
429 * Auto mount points get their node id from the submount root, which is
430 * not a unique identifier within this filesystem.
431 *
432 * To avoid conflicts, do not place submount points into the inode hash
433 * table.
434 */
435 if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
436 S_ISDIR(attr->mode)) {
437 struct fuse_inode *fi;
438
439 inode = new_inode(sb);
440 if (!inode)
441 return NULL;
442
443 fuse_init_inode(inode, attr, fc);
444 fi = get_fuse_inode(inode);
445 fi->nodeid = nodeid;
446 fi->submount_lookup = fuse_alloc_submount_lookup();
447 if (!fi->submount_lookup) {
448 iput(inode);
449 return NULL;
450 }
451 /* Sets nlookup = 1 on fi->submount_lookup->nlookup */
452 fuse_init_submount_lookup(fi->submount_lookup, nodeid);
453 inode->i_flags |= S_AUTOMOUNT;
454 goto done;
455 }
456
457retry:
458 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
459 if (!inode)
460 return NULL;
461
462 if ((inode->i_state & I_NEW)) {
463 inode->i_flags |= S_NOATIME;
464 if (!fc->writeback_cache || !S_ISREG(attr->mode))
465 inode->i_flags |= S_NOCMTIME;
466 inode->i_generation = generation;
467 fuse_init_inode(inode, attr, fc);
468 unlock_new_inode(inode);
469 } else if (fuse_stale_inode(inode, generation, attr)) {
470 /* nodeid was reused, any I/O on the old inode should fail */
471 fuse_make_bad(inode);
472 iput(inode);
473 goto retry;
474 }
475 fi = get_fuse_inode(inode);
476 spin_lock(&fi->lock);
477 fi->nlookup++;
478 spin_unlock(&fi->lock);
479done:
480 fuse_change_attributes(inode, attr, NULL, attr_valid, attr_version);
481
482 return inode;
483}
484
485struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
486 struct fuse_mount **fm)
487{
488 struct fuse_mount *fm_iter;
489 struct inode *inode;
490
491 WARN_ON(!rwsem_is_locked(&fc->killsb));
492 list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
493 if (!fm_iter->sb)
494 continue;
495
496 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
497 if (inode) {
498 if (fm)
499 *fm = fm_iter;
500 return inode;
501 }
502 }
503
504 return NULL;
505}
506
507int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
508 loff_t offset, loff_t len)
509{
510 struct fuse_inode *fi;
511 struct inode *inode;
512 pgoff_t pg_start;
513 pgoff_t pg_end;
514
515 inode = fuse_ilookup(fc, nodeid, NULL);
516 if (!inode)
517 return -ENOENT;
518
519 fi = get_fuse_inode(inode);
520 spin_lock(&fi->lock);
521 fi->attr_version = atomic64_inc_return(&fc->attr_version);
522 spin_unlock(&fi->lock);
523
524 fuse_invalidate_attr(inode);
525 forget_all_cached_acls(inode);
526 if (offset >= 0) {
527 pg_start = offset >> PAGE_SHIFT;
528 if (len <= 0)
529 pg_end = -1;
530 else
531 pg_end = (offset + len - 1) >> PAGE_SHIFT;
532 invalidate_inode_pages2_range(inode->i_mapping,
533 pg_start, pg_end);
534 }
535 iput(inode);
536 return 0;
537}
538
539bool fuse_lock_inode(struct inode *inode)
540{
541 bool locked = false;
542
543 if (!get_fuse_conn(inode)->parallel_dirops) {
544 mutex_lock(&get_fuse_inode(inode)->mutex);
545 locked = true;
546 }
547
548 return locked;
549}
550
551void fuse_unlock_inode(struct inode *inode, bool locked)
552{
553 if (locked)
554 mutex_unlock(&get_fuse_inode(inode)->mutex);
555}
556
557static void fuse_umount_begin(struct super_block *sb)
558{
559 struct fuse_conn *fc = get_fuse_conn_super(sb);
560
561 if (fc->no_force_umount)
562 return;
563
564 fuse_abort_conn(fc);
565
566 // Only retire block-device-based superblocks.
567 if (sb->s_bdev != NULL)
568 retire_super(sb);
569}
570
571static void fuse_send_destroy(struct fuse_mount *fm)
572{
573 if (fm->fc->conn_init) {
574 FUSE_ARGS(args);
575
576 args.opcode = FUSE_DESTROY;
577 args.force = true;
578 args.nocreds = true;
579 fuse_simple_request(fm, &args);
580 }
581}
582
583static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
584{
585 stbuf->f_type = FUSE_SUPER_MAGIC;
586 stbuf->f_bsize = attr->bsize;
587 stbuf->f_frsize = attr->frsize;
588 stbuf->f_blocks = attr->blocks;
589 stbuf->f_bfree = attr->bfree;
590 stbuf->f_bavail = attr->bavail;
591 stbuf->f_files = attr->files;
592 stbuf->f_ffree = attr->ffree;
593 stbuf->f_namelen = attr->namelen;
594 /* fsid is left zero */
595}
596
597static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
598{
599 struct super_block *sb = dentry->d_sb;
600 struct fuse_mount *fm = get_fuse_mount_super(sb);
601 FUSE_ARGS(args);
602 struct fuse_statfs_out outarg;
603 int err;
604
605 if (!fuse_allow_current_process(fm->fc)) {
606 buf->f_type = FUSE_SUPER_MAGIC;
607 return 0;
608 }
609
610 memset(&outarg, 0, sizeof(outarg));
611 args.in_numargs = 0;
612 args.opcode = FUSE_STATFS;
613 args.nodeid = get_node_id(d_inode(dentry));
614 args.out_numargs = 1;
615 args.out_args[0].size = sizeof(outarg);
616 args.out_args[0].value = &outarg;
617 err = fuse_simple_request(fm, &args);
618 if (!err)
619 convert_fuse_statfs(buf, &outarg.st);
620 return err;
621}
622
623static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
624{
625 struct fuse_sync_bucket *bucket;
626
627 bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
628 if (bucket) {
629 init_waitqueue_head(&bucket->waitq);
630 /* Initial active count */
631 atomic_set(&bucket->count, 1);
632 }
633 return bucket;
634}
635
636static void fuse_sync_fs_writes(struct fuse_conn *fc)
637{
638 struct fuse_sync_bucket *bucket, *new_bucket;
639 int count;
640
641 new_bucket = fuse_sync_bucket_alloc();
642 spin_lock(&fc->lock);
643 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
644 count = atomic_read(&bucket->count);
645 WARN_ON(count < 1);
646 /* No outstanding writes? */
647 if (count == 1) {
648 spin_unlock(&fc->lock);
649 kfree(new_bucket);
650 return;
651 }
652
653 /*
654 * Completion of new bucket depends on completion of this bucket, so add
655 * one more count.
656 */
657 atomic_inc(&new_bucket->count);
658 rcu_assign_pointer(fc->curr_bucket, new_bucket);
659 spin_unlock(&fc->lock);
660 /*
661 * Drop initial active count. At this point if all writes in this and
662 * ancestor buckets complete, the count will go to zero and this task
663 * will be woken up.
664 */
665 atomic_dec(&bucket->count);
666
667 wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
668
669 /* Drop temp count on descendant bucket */
670 fuse_sync_bucket_dec(new_bucket);
671 kfree_rcu(bucket, rcu);
672}
673
674static int fuse_sync_fs(struct super_block *sb, int wait)
675{
676 struct fuse_mount *fm = get_fuse_mount_super(sb);
677 struct fuse_conn *fc = fm->fc;
678 struct fuse_syncfs_in inarg;
679 FUSE_ARGS(args);
680 int err;
681
682 /*
683 * Userspace cannot handle the wait == 0 case. Avoid a
684 * gratuitous roundtrip.
685 */
686 if (!wait)
687 return 0;
688
689 /* The filesystem is being unmounted. Nothing to do. */
690 if (!sb->s_root)
691 return 0;
692
693 if (!fc->sync_fs)
694 return 0;
695
696 fuse_sync_fs_writes(fc);
697
698 memset(&inarg, 0, sizeof(inarg));
699 args.in_numargs = 1;
700 args.in_args[0].size = sizeof(inarg);
701 args.in_args[0].value = &inarg;
702 args.opcode = FUSE_SYNCFS;
703 args.nodeid = get_node_id(sb->s_root->d_inode);
704 args.out_numargs = 0;
705
706 err = fuse_simple_request(fm, &args);
707 if (err == -ENOSYS) {
708 fc->sync_fs = 0;
709 err = 0;
710 }
711
712 return err;
713}
714
715enum {
716 OPT_SOURCE,
717 OPT_SUBTYPE,
718 OPT_FD,
719 OPT_ROOTMODE,
720 OPT_USER_ID,
721 OPT_GROUP_ID,
722 OPT_DEFAULT_PERMISSIONS,
723 OPT_ALLOW_OTHER,
724 OPT_MAX_READ,
725 OPT_BLKSIZE,
726 OPT_ERR
727};
728
729static const struct fs_parameter_spec fuse_fs_parameters[] = {
730 fsparam_string ("source", OPT_SOURCE),
731 fsparam_u32 ("fd", OPT_FD),
732 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
733 fsparam_u32 ("user_id", OPT_USER_ID),
734 fsparam_u32 ("group_id", OPT_GROUP_ID),
735 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
736 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
737 fsparam_u32 ("max_read", OPT_MAX_READ),
738 fsparam_u32 ("blksize", OPT_BLKSIZE),
739 fsparam_string ("subtype", OPT_SUBTYPE),
740 {}
741};
742
743static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
744{
745 struct fs_parse_result result;
746 struct fuse_fs_context *ctx = fsc->fs_private;
747 int opt;
748
749 if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
750 /*
751 * Ignore options coming from mount(MS_REMOUNT) for backward
752 * compatibility.
753 */
754 if (fsc->oldapi)
755 return 0;
756
757 return invalfc(fsc, "No changes allowed in reconfigure");
758 }
759
760 opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
761 if (opt < 0)
762 return opt;
763
764 switch (opt) {
765 case OPT_SOURCE:
766 if (fsc->source)
767 return invalfc(fsc, "Multiple sources specified");
768 fsc->source = param->string;
769 param->string = NULL;
770 break;
771
772 case OPT_SUBTYPE:
773 if (ctx->subtype)
774 return invalfc(fsc, "Multiple subtypes specified");
775 ctx->subtype = param->string;
776 param->string = NULL;
777 return 0;
778
779 case OPT_FD:
780 ctx->fd = result.uint_32;
781 ctx->fd_present = true;
782 break;
783
784 case OPT_ROOTMODE:
785 if (!fuse_valid_type(result.uint_32))
786 return invalfc(fsc, "Invalid rootmode");
787 ctx->rootmode = result.uint_32;
788 ctx->rootmode_present = true;
789 break;
790
791 case OPT_USER_ID:
792 ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
793 if (!uid_valid(ctx->user_id))
794 return invalfc(fsc, "Invalid user_id");
795 ctx->user_id_present = true;
796 break;
797
798 case OPT_GROUP_ID:
799 ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
800 if (!gid_valid(ctx->group_id))
801 return invalfc(fsc, "Invalid group_id");
802 ctx->group_id_present = true;
803 break;
804
805 case OPT_DEFAULT_PERMISSIONS:
806 ctx->default_permissions = true;
807 break;
808
809 case OPT_ALLOW_OTHER:
810 ctx->allow_other = true;
811 break;
812
813 case OPT_MAX_READ:
814 ctx->max_read = result.uint_32;
815 break;
816
817 case OPT_BLKSIZE:
818 if (!ctx->is_bdev)
819 return invalfc(fsc, "blksize only supported for fuseblk");
820 ctx->blksize = result.uint_32;
821 break;
822
823 default:
824 return -EINVAL;
825 }
826
827 return 0;
828}
829
830static void fuse_free_fsc(struct fs_context *fsc)
831{
832 struct fuse_fs_context *ctx = fsc->fs_private;
833
834 if (ctx) {
835 kfree(ctx->subtype);
836 kfree(ctx);
837 }
838}
839
840static int fuse_show_options(struct seq_file *m, struct dentry *root)
841{
842 struct super_block *sb = root->d_sb;
843 struct fuse_conn *fc = get_fuse_conn_super(sb);
844
845 if (fc->legacy_opts_show) {
846 seq_printf(m, ",user_id=%u",
847 from_kuid_munged(fc->user_ns, fc->user_id));
848 seq_printf(m, ",group_id=%u",
849 from_kgid_munged(fc->user_ns, fc->group_id));
850 if (fc->default_permissions)
851 seq_puts(m, ",default_permissions");
852 if (fc->allow_other)
853 seq_puts(m, ",allow_other");
854 if (fc->max_read != ~0)
855 seq_printf(m, ",max_read=%u", fc->max_read);
856 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
857 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
858 }
859#ifdef CONFIG_FUSE_DAX
860 if (fc->dax_mode == FUSE_DAX_ALWAYS)
861 seq_puts(m, ",dax=always");
862 else if (fc->dax_mode == FUSE_DAX_NEVER)
863 seq_puts(m, ",dax=never");
864 else if (fc->dax_mode == FUSE_DAX_INODE_USER)
865 seq_puts(m, ",dax=inode");
866#endif
867
868 return 0;
869}
870
871static void fuse_iqueue_init(struct fuse_iqueue *fiq,
872 const struct fuse_iqueue_ops *ops,
873 void *priv)
874{
875 memset(fiq, 0, sizeof(struct fuse_iqueue));
876 spin_lock_init(&fiq->lock);
877 init_waitqueue_head(&fiq->waitq);
878 INIT_LIST_HEAD(&fiq->pending);
879 INIT_LIST_HEAD(&fiq->interrupts);
880 fiq->forget_list_tail = &fiq->forget_list_head;
881 fiq->connected = 1;
882 fiq->ops = ops;
883 fiq->priv = priv;
884}
885
886static void fuse_pqueue_init(struct fuse_pqueue *fpq)
887{
888 unsigned int i;
889
890 spin_lock_init(&fpq->lock);
891 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
892 INIT_LIST_HEAD(&fpq->processing[i]);
893 INIT_LIST_HEAD(&fpq->io);
894 fpq->connected = 1;
895}
896
897void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
898 struct user_namespace *user_ns,
899 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
900{
901 memset(fc, 0, sizeof(*fc));
902 spin_lock_init(&fc->lock);
903 spin_lock_init(&fc->bg_lock);
904 init_rwsem(&fc->killsb);
905 refcount_set(&fc->count, 1);
906 atomic_set(&fc->dev_count, 1);
907 init_waitqueue_head(&fc->blocked_waitq);
908 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
909 INIT_LIST_HEAD(&fc->bg_queue);
910 INIT_LIST_HEAD(&fc->entry);
911 INIT_LIST_HEAD(&fc->devices);
912 atomic_set(&fc->num_waiting, 0);
913 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
914 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
915 atomic64_set(&fc->khctr, 0);
916 fc->polled_files = RB_ROOT;
917 fc->blocked = 0;
918 fc->initialized = 0;
919 fc->connected = 1;
920 atomic64_set(&fc->attr_version, 1);
921 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
922 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
923 fc->user_ns = get_user_ns(user_ns);
924 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
925 fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
926
927 INIT_LIST_HEAD(&fc->mounts);
928 list_add(&fm->fc_entry, &fc->mounts);
929 fm->fc = fc;
930}
931EXPORT_SYMBOL_GPL(fuse_conn_init);
932
933static void delayed_release(struct rcu_head *p)
934{
935 struct fuse_conn *fc = container_of(p, struct fuse_conn, rcu);
936
937 put_user_ns(fc->user_ns);
938 fc->release(fc);
939}
940
941void fuse_conn_put(struct fuse_conn *fc)
942{
943 if (refcount_dec_and_test(&fc->count)) {
944 struct fuse_iqueue *fiq = &fc->iq;
945 struct fuse_sync_bucket *bucket;
946
947 if (IS_ENABLED(CONFIG_FUSE_DAX))
948 fuse_dax_conn_free(fc);
949 if (fiq->ops->release)
950 fiq->ops->release(fiq);
951 put_pid_ns(fc->pid_ns);
952 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
953 if (bucket) {
954 WARN_ON(atomic_read(&bucket->count) != 1);
955 kfree(bucket);
956 }
957 call_rcu(&fc->rcu, delayed_release);
958 }
959}
960EXPORT_SYMBOL_GPL(fuse_conn_put);
961
962struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
963{
964 refcount_inc(&fc->count);
965 return fc;
966}
967EXPORT_SYMBOL_GPL(fuse_conn_get);
968
969static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
970{
971 struct fuse_attr attr;
972 memset(&attr, 0, sizeof(attr));
973
974 attr.mode = mode;
975 attr.ino = FUSE_ROOT_ID;
976 attr.nlink = 1;
977 return fuse_iget(sb, 1, 0, &attr, 0, 0);
978}
979
980struct fuse_inode_handle {
981 u64 nodeid;
982 u32 generation;
983};
984
985static struct dentry *fuse_get_dentry(struct super_block *sb,
986 struct fuse_inode_handle *handle)
987{
988 struct fuse_conn *fc = get_fuse_conn_super(sb);
989 struct inode *inode;
990 struct dentry *entry;
991 int err = -ESTALE;
992
993 if (handle->nodeid == 0)
994 goto out_err;
995
996 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
997 if (!inode) {
998 struct fuse_entry_out outarg;
999 const struct qstr name = QSTR_INIT(".", 1);
1000
1001 if (!fc->export_support)
1002 goto out_err;
1003
1004 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
1005 &inode);
1006 if (err && err != -ENOENT)
1007 goto out_err;
1008 if (err || !inode) {
1009 err = -ESTALE;
1010 goto out_err;
1011 }
1012 err = -EIO;
1013 if (get_node_id(inode) != handle->nodeid)
1014 goto out_iput;
1015 }
1016 err = -ESTALE;
1017 if (inode->i_generation != handle->generation)
1018 goto out_iput;
1019
1020 entry = d_obtain_alias(inode);
1021 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
1022 fuse_invalidate_entry_cache(entry);
1023
1024 return entry;
1025
1026 out_iput:
1027 iput(inode);
1028 out_err:
1029 return ERR_PTR(err);
1030}
1031
1032static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
1033 struct inode *parent)
1034{
1035 int len = parent ? 6 : 3;
1036 u64 nodeid;
1037 u32 generation;
1038
1039 if (*max_len < len) {
1040 *max_len = len;
1041 return FILEID_INVALID;
1042 }
1043
1044 nodeid = get_fuse_inode(inode)->nodeid;
1045 generation = inode->i_generation;
1046
1047 fh[0] = (u32)(nodeid >> 32);
1048 fh[1] = (u32)(nodeid & 0xffffffff);
1049 fh[2] = generation;
1050
1051 if (parent) {
1052 nodeid = get_fuse_inode(parent)->nodeid;
1053 generation = parent->i_generation;
1054
1055 fh[3] = (u32)(nodeid >> 32);
1056 fh[4] = (u32)(nodeid & 0xffffffff);
1057 fh[5] = generation;
1058 }
1059
1060 *max_len = len;
1061 return parent ? FILEID_INO64_GEN_PARENT : FILEID_INO64_GEN;
1062}
1063
1064static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
1065 struct fid *fid, int fh_len, int fh_type)
1066{
1067 struct fuse_inode_handle handle;
1068
1069 if ((fh_type != FILEID_INO64_GEN &&
1070 fh_type != FILEID_INO64_GEN_PARENT) || fh_len < 3)
1071 return NULL;
1072
1073 handle.nodeid = (u64) fid->raw[0] << 32;
1074 handle.nodeid |= (u64) fid->raw[1];
1075 handle.generation = fid->raw[2];
1076 return fuse_get_dentry(sb, &handle);
1077}
1078
1079static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1080 struct fid *fid, int fh_len, int fh_type)
1081{
1082 struct fuse_inode_handle parent;
1083
1084 if (fh_type != FILEID_INO64_GEN_PARENT || fh_len < 6)
1085 return NULL;
1086
1087 parent.nodeid = (u64) fid->raw[3] << 32;
1088 parent.nodeid |= (u64) fid->raw[4];
1089 parent.generation = fid->raw[5];
1090 return fuse_get_dentry(sb, &parent);
1091}
1092
1093static struct dentry *fuse_get_parent(struct dentry *child)
1094{
1095 struct inode *child_inode = d_inode(child);
1096 struct fuse_conn *fc = get_fuse_conn(child_inode);
1097 struct inode *inode;
1098 struct dentry *parent;
1099 struct fuse_entry_out outarg;
1100 int err;
1101
1102 if (!fc->export_support)
1103 return ERR_PTR(-ESTALE);
1104
1105 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1106 &dotdot_name, &outarg, &inode);
1107 if (err) {
1108 if (err == -ENOENT)
1109 return ERR_PTR(-ESTALE);
1110 return ERR_PTR(err);
1111 }
1112
1113 parent = d_obtain_alias(inode);
1114 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1115 fuse_invalidate_entry_cache(parent);
1116
1117 return parent;
1118}
1119
1120static const struct export_operations fuse_export_operations = {
1121 .fh_to_dentry = fuse_fh_to_dentry,
1122 .fh_to_parent = fuse_fh_to_parent,
1123 .encode_fh = fuse_encode_fh,
1124 .get_parent = fuse_get_parent,
1125};
1126
1127static const struct super_operations fuse_super_operations = {
1128 .alloc_inode = fuse_alloc_inode,
1129 .free_inode = fuse_free_inode,
1130 .evict_inode = fuse_evict_inode,
1131 .write_inode = fuse_write_inode,
1132 .drop_inode = generic_delete_inode,
1133 .umount_begin = fuse_umount_begin,
1134 .statfs = fuse_statfs,
1135 .sync_fs = fuse_sync_fs,
1136 .show_options = fuse_show_options,
1137};
1138
1139static void sanitize_global_limit(unsigned *limit)
1140{
1141 /*
1142 * The default maximum number of async requests is calculated to consume
1143 * 1/2^13 of the total memory, assuming 392 bytes per request.
1144 */
1145 if (*limit == 0)
1146 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1147
1148 if (*limit >= 1 << 16)
1149 *limit = (1 << 16) - 1;
1150}
1151
1152static int set_global_limit(const char *val, const struct kernel_param *kp)
1153{
1154 int rv;
1155
1156 rv = param_set_uint(val, kp);
1157 if (rv)
1158 return rv;
1159
1160 sanitize_global_limit((unsigned *)kp->arg);
1161
1162 return 0;
1163}
1164
1165static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1166{
1167 int cap_sys_admin = capable(CAP_SYS_ADMIN);
1168
1169 if (arg->minor < 13)
1170 return;
1171
1172 sanitize_global_limit(&max_user_bgreq);
1173 sanitize_global_limit(&max_user_congthresh);
1174
1175 spin_lock(&fc->bg_lock);
1176 if (arg->max_background) {
1177 fc->max_background = arg->max_background;
1178
1179 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1180 fc->max_background = max_user_bgreq;
1181 }
1182 if (arg->congestion_threshold) {
1183 fc->congestion_threshold = arg->congestion_threshold;
1184
1185 if (!cap_sys_admin &&
1186 fc->congestion_threshold > max_user_congthresh)
1187 fc->congestion_threshold = max_user_congthresh;
1188 }
1189 spin_unlock(&fc->bg_lock);
1190}
1191
1192struct fuse_init_args {
1193 struct fuse_args args;
1194 struct fuse_init_in in;
1195 struct fuse_init_out out;
1196};
1197
1198static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1199 int error)
1200{
1201 struct fuse_conn *fc = fm->fc;
1202 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1203 struct fuse_init_out *arg = &ia->out;
1204 bool ok = true;
1205
1206 if (error || arg->major != FUSE_KERNEL_VERSION)
1207 ok = false;
1208 else {
1209 unsigned long ra_pages;
1210
1211 process_init_limits(fc, arg);
1212
1213 if (arg->minor >= 6) {
1214 u64 flags = arg->flags;
1215
1216 if (flags & FUSE_INIT_EXT)
1217 flags |= (u64) arg->flags2 << 32;
1218
1219 ra_pages = arg->max_readahead / PAGE_SIZE;
1220 if (flags & FUSE_ASYNC_READ)
1221 fc->async_read = 1;
1222 if (!(flags & FUSE_POSIX_LOCKS))
1223 fc->no_lock = 1;
1224 if (arg->minor >= 17) {
1225 if (!(flags & FUSE_FLOCK_LOCKS))
1226 fc->no_flock = 1;
1227 } else {
1228 if (!(flags & FUSE_POSIX_LOCKS))
1229 fc->no_flock = 1;
1230 }
1231 if (flags & FUSE_ATOMIC_O_TRUNC)
1232 fc->atomic_o_trunc = 1;
1233 if (arg->minor >= 9) {
1234 /* LOOKUP has dependency on proto version */
1235 if (flags & FUSE_EXPORT_SUPPORT)
1236 fc->export_support = 1;
1237 }
1238 if (flags & FUSE_BIG_WRITES)
1239 fc->big_writes = 1;
1240 if (flags & FUSE_DONT_MASK)
1241 fc->dont_mask = 1;
1242 if (flags & FUSE_AUTO_INVAL_DATA)
1243 fc->auto_inval_data = 1;
1244 else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1245 fc->explicit_inval_data = 1;
1246 if (flags & FUSE_DO_READDIRPLUS) {
1247 fc->do_readdirplus = 1;
1248 if (flags & FUSE_READDIRPLUS_AUTO)
1249 fc->readdirplus_auto = 1;
1250 }
1251 if (flags & FUSE_ASYNC_DIO)
1252 fc->async_dio = 1;
1253 if (flags & FUSE_WRITEBACK_CACHE)
1254 fc->writeback_cache = 1;
1255 if (flags & FUSE_PARALLEL_DIROPS)
1256 fc->parallel_dirops = 1;
1257 if (flags & FUSE_HANDLE_KILLPRIV)
1258 fc->handle_killpriv = 1;
1259 if (arg->time_gran && arg->time_gran <= 1000000000)
1260 fm->sb->s_time_gran = arg->time_gran;
1261 if ((flags & FUSE_POSIX_ACL)) {
1262 fc->default_permissions = 1;
1263 fc->posix_acl = 1;
1264 }
1265 if (flags & FUSE_CACHE_SYMLINKS)
1266 fc->cache_symlinks = 1;
1267 if (flags & FUSE_ABORT_ERROR)
1268 fc->abort_err = 1;
1269 if (flags & FUSE_MAX_PAGES) {
1270 fc->max_pages =
1271 min_t(unsigned int, fc->max_pages_limit,
1272 max_t(unsigned int, arg->max_pages, 1));
1273 }
1274 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1275 if (flags & FUSE_MAP_ALIGNMENT &&
1276 !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1277 ok = false;
1278 }
1279 if (flags & FUSE_HAS_INODE_DAX)
1280 fc->inode_dax = 1;
1281 }
1282 if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1283 fc->handle_killpriv_v2 = 1;
1284 fm->sb->s_flags |= SB_NOSEC;
1285 }
1286 if (flags & FUSE_SETXATTR_EXT)
1287 fc->setxattr_ext = 1;
1288 if (flags & FUSE_SECURITY_CTX)
1289 fc->init_security = 1;
1290 if (flags & FUSE_CREATE_SUPP_GROUP)
1291 fc->create_supp_group = 1;
1292 if (flags & FUSE_DIRECT_IO_ALLOW_MMAP)
1293 fc->direct_io_allow_mmap = 1;
1294 } else {
1295 ra_pages = fc->max_read / PAGE_SIZE;
1296 fc->no_lock = 1;
1297 fc->no_flock = 1;
1298 }
1299
1300 fm->sb->s_bdi->ra_pages =
1301 min(fm->sb->s_bdi->ra_pages, ra_pages);
1302 fc->minor = arg->minor;
1303 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1304 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1305 fc->conn_init = 1;
1306 }
1307 kfree(ia);
1308
1309 if (!ok) {
1310 fc->conn_init = 0;
1311 fc->conn_error = 1;
1312 }
1313
1314 fuse_set_initialized(fc);
1315 wake_up_all(&fc->blocked_waitq);
1316}
1317
1318void fuse_send_init(struct fuse_mount *fm)
1319{
1320 struct fuse_init_args *ia;
1321 u64 flags;
1322
1323 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1324
1325 ia->in.major = FUSE_KERNEL_VERSION;
1326 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1327 ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1328 flags =
1329 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1330 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1331 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1332 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1333 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1334 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1335 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1336 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1337 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1338 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1339 FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP |
1340 FUSE_HAS_EXPIRE_ONLY | FUSE_DIRECT_IO_ALLOW_MMAP;
1341#ifdef CONFIG_FUSE_DAX
1342 if (fm->fc->dax)
1343 flags |= FUSE_MAP_ALIGNMENT;
1344 if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1345 flags |= FUSE_HAS_INODE_DAX;
1346#endif
1347 if (fm->fc->auto_submounts)
1348 flags |= FUSE_SUBMOUNTS;
1349
1350 ia->in.flags = flags;
1351 ia->in.flags2 = flags >> 32;
1352
1353 ia->args.opcode = FUSE_INIT;
1354 ia->args.in_numargs = 1;
1355 ia->args.in_args[0].size = sizeof(ia->in);
1356 ia->args.in_args[0].value = &ia->in;
1357 ia->args.out_numargs = 1;
1358 /* Variable length argument used for backward compatibility
1359 with interface version < 7.5. Rest of init_out is zeroed
1360 by do_get_request(), so a short reply is not a problem */
1361 ia->args.out_argvar = true;
1362 ia->args.out_args[0].size = sizeof(ia->out);
1363 ia->args.out_args[0].value = &ia->out;
1364 ia->args.force = true;
1365 ia->args.nocreds = true;
1366 ia->args.end = process_init_reply;
1367
1368 if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1369 process_init_reply(fm, &ia->args, -ENOTCONN);
1370}
1371EXPORT_SYMBOL_GPL(fuse_send_init);
1372
1373void fuse_free_conn(struct fuse_conn *fc)
1374{
1375 WARN_ON(!list_empty(&fc->devices));
1376 kfree(fc);
1377}
1378EXPORT_SYMBOL_GPL(fuse_free_conn);
1379
1380static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1381{
1382 int err;
1383 char *suffix = "";
1384
1385 if (sb->s_bdev) {
1386 suffix = "-fuseblk";
1387 /*
1388 * sb->s_bdi points to blkdev's bdi however we want to redirect
1389 * it to our private bdi...
1390 */
1391 bdi_put(sb->s_bdi);
1392 sb->s_bdi = &noop_backing_dev_info;
1393 }
1394 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1395 MINOR(fc->dev), suffix);
1396 if (err)
1397 return err;
1398
1399 /* fuse does it's own writeback accounting */
1400 sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1401 sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1402
1403 /*
1404 * For a single fuse filesystem use max 1% of dirty +
1405 * writeback threshold.
1406 *
1407 * This gives about 1M of write buffer for memory maps on a
1408 * machine with 1G and 10% dirty_ratio, which should be more
1409 * than enough.
1410 *
1411 * Privileged users can raise it by writing to
1412 *
1413 * /sys/class/bdi/<bdi>/max_ratio
1414 */
1415 bdi_set_max_ratio(sb->s_bdi, 1);
1416
1417 return 0;
1418}
1419
1420struct fuse_dev *fuse_dev_alloc(void)
1421{
1422 struct fuse_dev *fud;
1423 struct list_head *pq;
1424
1425 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1426 if (!fud)
1427 return NULL;
1428
1429 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1430 if (!pq) {
1431 kfree(fud);
1432 return NULL;
1433 }
1434
1435 fud->pq.processing = pq;
1436 fuse_pqueue_init(&fud->pq);
1437
1438 return fud;
1439}
1440EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1441
1442void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1443{
1444 fud->fc = fuse_conn_get(fc);
1445 spin_lock(&fc->lock);
1446 list_add_tail(&fud->entry, &fc->devices);
1447 spin_unlock(&fc->lock);
1448}
1449EXPORT_SYMBOL_GPL(fuse_dev_install);
1450
1451struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1452{
1453 struct fuse_dev *fud;
1454
1455 fud = fuse_dev_alloc();
1456 if (!fud)
1457 return NULL;
1458
1459 fuse_dev_install(fud, fc);
1460 return fud;
1461}
1462EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1463
1464void fuse_dev_free(struct fuse_dev *fud)
1465{
1466 struct fuse_conn *fc = fud->fc;
1467
1468 if (fc) {
1469 spin_lock(&fc->lock);
1470 list_del(&fud->entry);
1471 spin_unlock(&fc->lock);
1472
1473 fuse_conn_put(fc);
1474 }
1475 kfree(fud->pq.processing);
1476 kfree(fud);
1477}
1478EXPORT_SYMBOL_GPL(fuse_dev_free);
1479
1480static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1481 const struct fuse_inode *fi)
1482{
1483 struct timespec64 atime = inode_get_atime(&fi->inode);
1484 struct timespec64 mtime = inode_get_mtime(&fi->inode);
1485 struct timespec64 ctime = inode_get_ctime(&fi->inode);
1486
1487 *attr = (struct fuse_attr){
1488 .ino = fi->inode.i_ino,
1489 .size = fi->inode.i_size,
1490 .blocks = fi->inode.i_blocks,
1491 .atime = atime.tv_sec,
1492 .mtime = mtime.tv_sec,
1493 .ctime = ctime.tv_sec,
1494 .atimensec = atime.tv_nsec,
1495 .mtimensec = mtime.tv_nsec,
1496 .ctimensec = ctime.tv_nsec,
1497 .mode = fi->inode.i_mode,
1498 .nlink = fi->inode.i_nlink,
1499 .uid = fi->inode.i_uid.val,
1500 .gid = fi->inode.i_gid.val,
1501 .rdev = fi->inode.i_rdev,
1502 .blksize = 1u << fi->inode.i_blkbits,
1503 };
1504}
1505
1506static void fuse_sb_defaults(struct super_block *sb)
1507{
1508 sb->s_magic = FUSE_SUPER_MAGIC;
1509 sb->s_op = &fuse_super_operations;
1510 sb->s_xattr = fuse_xattr_handlers;
1511 sb->s_maxbytes = MAX_LFS_FILESIZE;
1512 sb->s_time_gran = 1;
1513 sb->s_export_op = &fuse_export_operations;
1514 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1515 if (sb->s_user_ns != &init_user_ns)
1516 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1517 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1518}
1519
1520static int fuse_fill_super_submount(struct super_block *sb,
1521 struct fuse_inode *parent_fi)
1522{
1523 struct fuse_mount *fm = get_fuse_mount_super(sb);
1524 struct super_block *parent_sb = parent_fi->inode.i_sb;
1525 struct fuse_attr root_attr;
1526 struct inode *root;
1527 struct fuse_submount_lookup *sl;
1528 struct fuse_inode *fi;
1529
1530 fuse_sb_defaults(sb);
1531 fm->sb = sb;
1532
1533 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1534 sb->s_bdi = bdi_get(parent_sb->s_bdi);
1535
1536 sb->s_xattr = parent_sb->s_xattr;
1537 sb->s_time_gran = parent_sb->s_time_gran;
1538 sb->s_blocksize = parent_sb->s_blocksize;
1539 sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1540 sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1541 if (parent_sb->s_subtype && !sb->s_subtype)
1542 return -ENOMEM;
1543
1544 fuse_fill_attr_from_inode(&root_attr, parent_fi);
1545 root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1546 /*
1547 * This inode is just a duplicate, so it is not looked up and
1548 * its nlookup should not be incremented. fuse_iget() does
1549 * that, though, so undo it here.
1550 */
1551 fi = get_fuse_inode(root);
1552 fi->nlookup--;
1553
1554 sb->s_d_op = &fuse_dentry_operations;
1555 sb->s_root = d_make_root(root);
1556 if (!sb->s_root)
1557 return -ENOMEM;
1558
1559 /*
1560 * Grab the parent's submount_lookup pointer and take a
1561 * reference on the shared nlookup from the parent. This is to
1562 * prevent the last forget for this nodeid from getting
1563 * triggered until all users have finished with it.
1564 */
1565 sl = parent_fi->submount_lookup;
1566 WARN_ON(!sl);
1567 if (sl) {
1568 refcount_inc(&sl->count);
1569 fi->submount_lookup = sl;
1570 }
1571
1572 return 0;
1573}
1574
1575/* Filesystem context private data holds the FUSE inode of the mount point */
1576static int fuse_get_tree_submount(struct fs_context *fsc)
1577{
1578 struct fuse_mount *fm;
1579 struct fuse_inode *mp_fi = fsc->fs_private;
1580 struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1581 struct super_block *sb;
1582 int err;
1583
1584 fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1585 if (!fm)
1586 return -ENOMEM;
1587
1588 fm->fc = fuse_conn_get(fc);
1589 fsc->s_fs_info = fm;
1590 sb = sget_fc(fsc, NULL, set_anon_super_fc);
1591 if (fsc->s_fs_info)
1592 fuse_mount_destroy(fm);
1593 if (IS_ERR(sb))
1594 return PTR_ERR(sb);
1595
1596 /* Initialize superblock, making @mp_fi its root */
1597 err = fuse_fill_super_submount(sb, mp_fi);
1598 if (err) {
1599 deactivate_locked_super(sb);
1600 return err;
1601 }
1602
1603 down_write(&fc->killsb);
1604 list_add_tail(&fm->fc_entry, &fc->mounts);
1605 up_write(&fc->killsb);
1606
1607 sb->s_flags |= SB_ACTIVE;
1608 fsc->root = dget(sb->s_root);
1609
1610 return 0;
1611}
1612
1613static const struct fs_context_operations fuse_context_submount_ops = {
1614 .get_tree = fuse_get_tree_submount,
1615};
1616
1617int fuse_init_fs_context_submount(struct fs_context *fsc)
1618{
1619 fsc->ops = &fuse_context_submount_ops;
1620 return 0;
1621}
1622EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1623
1624int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1625{
1626 struct fuse_dev *fud = NULL;
1627 struct fuse_mount *fm = get_fuse_mount_super(sb);
1628 struct fuse_conn *fc = fm->fc;
1629 struct inode *root;
1630 struct dentry *root_dentry;
1631 int err;
1632
1633 err = -EINVAL;
1634 if (sb->s_flags & SB_MANDLOCK)
1635 goto err;
1636
1637 rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1638 fuse_sb_defaults(sb);
1639
1640 if (ctx->is_bdev) {
1641#ifdef CONFIG_BLOCK
1642 err = -EINVAL;
1643 if (!sb_set_blocksize(sb, ctx->blksize))
1644 goto err;
1645#endif
1646 } else {
1647 sb->s_blocksize = PAGE_SIZE;
1648 sb->s_blocksize_bits = PAGE_SHIFT;
1649 }
1650
1651 sb->s_subtype = ctx->subtype;
1652 ctx->subtype = NULL;
1653 if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1654 err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1655 if (err)
1656 goto err;
1657 }
1658
1659 if (ctx->fudptr) {
1660 err = -ENOMEM;
1661 fud = fuse_dev_alloc_install(fc);
1662 if (!fud)
1663 goto err_free_dax;
1664 }
1665
1666 fc->dev = sb->s_dev;
1667 fm->sb = sb;
1668 err = fuse_bdi_init(fc, sb);
1669 if (err)
1670 goto err_dev_free;
1671
1672 /* Handle umasking inside the fuse code */
1673 if (sb->s_flags & SB_POSIXACL)
1674 fc->dont_mask = 1;
1675 sb->s_flags |= SB_POSIXACL;
1676
1677 fc->default_permissions = ctx->default_permissions;
1678 fc->allow_other = ctx->allow_other;
1679 fc->user_id = ctx->user_id;
1680 fc->group_id = ctx->group_id;
1681 fc->legacy_opts_show = ctx->legacy_opts_show;
1682 fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1683 fc->destroy = ctx->destroy;
1684 fc->no_control = ctx->no_control;
1685 fc->no_force_umount = ctx->no_force_umount;
1686
1687 err = -ENOMEM;
1688 root = fuse_get_root_inode(sb, ctx->rootmode);
1689 sb->s_d_op = &fuse_root_dentry_operations;
1690 root_dentry = d_make_root(root);
1691 if (!root_dentry)
1692 goto err_dev_free;
1693 /* Root dentry doesn't have .d_revalidate */
1694 sb->s_d_op = &fuse_dentry_operations;
1695
1696 mutex_lock(&fuse_mutex);
1697 err = -EINVAL;
1698 if (ctx->fudptr && *ctx->fudptr)
1699 goto err_unlock;
1700
1701 err = fuse_ctl_add_conn(fc);
1702 if (err)
1703 goto err_unlock;
1704
1705 list_add_tail(&fc->entry, &fuse_conn_list);
1706 sb->s_root = root_dentry;
1707 if (ctx->fudptr)
1708 *ctx->fudptr = fud;
1709 mutex_unlock(&fuse_mutex);
1710 return 0;
1711
1712 err_unlock:
1713 mutex_unlock(&fuse_mutex);
1714 dput(root_dentry);
1715 err_dev_free:
1716 if (fud)
1717 fuse_dev_free(fud);
1718 err_free_dax:
1719 if (IS_ENABLED(CONFIG_FUSE_DAX))
1720 fuse_dax_conn_free(fc);
1721 err:
1722 return err;
1723}
1724EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1725
1726static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1727{
1728 struct fuse_fs_context *ctx = fsc->fs_private;
1729 int err;
1730
1731 if (!ctx->file || !ctx->rootmode_present ||
1732 !ctx->user_id_present || !ctx->group_id_present)
1733 return -EINVAL;
1734
1735 /*
1736 * Require mount to happen from the same user namespace which
1737 * opened /dev/fuse to prevent potential attacks.
1738 */
1739 if ((ctx->file->f_op != &fuse_dev_operations) ||
1740 (ctx->file->f_cred->user_ns != sb->s_user_ns))
1741 return -EINVAL;
1742 ctx->fudptr = &ctx->file->private_data;
1743
1744 err = fuse_fill_super_common(sb, ctx);
1745 if (err)
1746 return err;
1747 /* file->private_data shall be visible on all CPUs after this */
1748 smp_mb();
1749 fuse_send_init(get_fuse_mount_super(sb));
1750 return 0;
1751}
1752
1753/*
1754 * This is the path where user supplied an already initialized fuse dev. In
1755 * this case never create a new super if the old one is gone.
1756 */
1757static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1758{
1759 return -ENOTCONN;
1760}
1761
1762static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1763{
1764
1765 return fsc->sget_key == get_fuse_conn_super(sb);
1766}
1767
1768static int fuse_get_tree(struct fs_context *fsc)
1769{
1770 struct fuse_fs_context *ctx = fsc->fs_private;
1771 struct fuse_dev *fud;
1772 struct fuse_conn *fc;
1773 struct fuse_mount *fm;
1774 struct super_block *sb;
1775 int err;
1776
1777 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1778 if (!fc)
1779 return -ENOMEM;
1780
1781 fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1782 if (!fm) {
1783 kfree(fc);
1784 return -ENOMEM;
1785 }
1786
1787 fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1788 fc->release = fuse_free_conn;
1789
1790 fsc->s_fs_info = fm;
1791
1792 if (ctx->fd_present)
1793 ctx->file = fget(ctx->fd);
1794
1795 if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1796 err = get_tree_bdev(fsc, fuse_fill_super);
1797 goto out;
1798 }
1799 /*
1800 * While block dev mount can be initialized with a dummy device fd
1801 * (found by device name), normal fuse mounts can't
1802 */
1803 err = -EINVAL;
1804 if (!ctx->file)
1805 goto out;
1806
1807 /*
1808 * Allow creating a fuse mount with an already initialized fuse
1809 * connection
1810 */
1811 fud = READ_ONCE(ctx->file->private_data);
1812 if (ctx->file->f_op == &fuse_dev_operations && fud) {
1813 fsc->sget_key = fud->fc;
1814 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1815 err = PTR_ERR_OR_ZERO(sb);
1816 if (!IS_ERR(sb))
1817 fsc->root = dget(sb->s_root);
1818 } else {
1819 err = get_tree_nodev(fsc, fuse_fill_super);
1820 }
1821out:
1822 if (fsc->s_fs_info)
1823 fuse_mount_destroy(fm);
1824 if (ctx->file)
1825 fput(ctx->file);
1826 return err;
1827}
1828
1829static const struct fs_context_operations fuse_context_ops = {
1830 .free = fuse_free_fsc,
1831 .parse_param = fuse_parse_param,
1832 .reconfigure = fuse_reconfigure,
1833 .get_tree = fuse_get_tree,
1834};
1835
1836/*
1837 * Set up the filesystem mount context.
1838 */
1839static int fuse_init_fs_context(struct fs_context *fsc)
1840{
1841 struct fuse_fs_context *ctx;
1842
1843 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1844 if (!ctx)
1845 return -ENOMEM;
1846
1847 ctx->max_read = ~0;
1848 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1849 ctx->legacy_opts_show = true;
1850
1851#ifdef CONFIG_BLOCK
1852 if (fsc->fs_type == &fuseblk_fs_type) {
1853 ctx->is_bdev = true;
1854 ctx->destroy = true;
1855 }
1856#endif
1857
1858 fsc->fs_private = ctx;
1859 fsc->ops = &fuse_context_ops;
1860 return 0;
1861}
1862
1863bool fuse_mount_remove(struct fuse_mount *fm)
1864{
1865 struct fuse_conn *fc = fm->fc;
1866 bool last = false;
1867
1868 down_write(&fc->killsb);
1869 list_del_init(&fm->fc_entry);
1870 if (list_empty(&fc->mounts))
1871 last = true;
1872 up_write(&fc->killsb);
1873
1874 return last;
1875}
1876EXPORT_SYMBOL_GPL(fuse_mount_remove);
1877
1878void fuse_conn_destroy(struct fuse_mount *fm)
1879{
1880 struct fuse_conn *fc = fm->fc;
1881
1882 if (fc->destroy)
1883 fuse_send_destroy(fm);
1884
1885 fuse_abort_conn(fc);
1886 fuse_wait_aborted(fc);
1887
1888 if (!list_empty(&fc->entry)) {
1889 mutex_lock(&fuse_mutex);
1890 list_del(&fc->entry);
1891 fuse_ctl_remove_conn(fc);
1892 mutex_unlock(&fuse_mutex);
1893 }
1894}
1895EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1896
1897static void fuse_sb_destroy(struct super_block *sb)
1898{
1899 struct fuse_mount *fm = get_fuse_mount_super(sb);
1900 bool last;
1901
1902 if (sb->s_root) {
1903 last = fuse_mount_remove(fm);
1904 if (last)
1905 fuse_conn_destroy(fm);
1906 }
1907}
1908
1909void fuse_mount_destroy(struct fuse_mount *fm)
1910{
1911 fuse_conn_put(fm->fc);
1912 kfree_rcu(fm, rcu);
1913}
1914EXPORT_SYMBOL(fuse_mount_destroy);
1915
1916static void fuse_kill_sb_anon(struct super_block *sb)
1917{
1918 fuse_sb_destroy(sb);
1919 kill_anon_super(sb);
1920 fuse_mount_destroy(get_fuse_mount_super(sb));
1921}
1922
1923static struct file_system_type fuse_fs_type = {
1924 .owner = THIS_MODULE,
1925 .name = "fuse",
1926 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1927 .init_fs_context = fuse_init_fs_context,
1928 .parameters = fuse_fs_parameters,
1929 .kill_sb = fuse_kill_sb_anon,
1930};
1931MODULE_ALIAS_FS("fuse");
1932
1933#ifdef CONFIG_BLOCK
1934static void fuse_kill_sb_blk(struct super_block *sb)
1935{
1936 fuse_sb_destroy(sb);
1937 kill_block_super(sb);
1938 fuse_mount_destroy(get_fuse_mount_super(sb));
1939}
1940
1941static struct file_system_type fuseblk_fs_type = {
1942 .owner = THIS_MODULE,
1943 .name = "fuseblk",
1944 .init_fs_context = fuse_init_fs_context,
1945 .parameters = fuse_fs_parameters,
1946 .kill_sb = fuse_kill_sb_blk,
1947 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1948};
1949MODULE_ALIAS_FS("fuseblk");
1950
1951static inline int register_fuseblk(void)
1952{
1953 return register_filesystem(&fuseblk_fs_type);
1954}
1955
1956static inline void unregister_fuseblk(void)
1957{
1958 unregister_filesystem(&fuseblk_fs_type);
1959}
1960#else
1961static inline int register_fuseblk(void)
1962{
1963 return 0;
1964}
1965
1966static inline void unregister_fuseblk(void)
1967{
1968}
1969#endif
1970
1971static void fuse_inode_init_once(void *foo)
1972{
1973 struct inode *inode = foo;
1974
1975 inode_init_once(inode);
1976}
1977
1978static int __init fuse_fs_init(void)
1979{
1980 int err;
1981
1982 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1983 sizeof(struct fuse_inode), 0,
1984 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1985 fuse_inode_init_once);
1986 err = -ENOMEM;
1987 if (!fuse_inode_cachep)
1988 goto out;
1989
1990 err = register_fuseblk();
1991 if (err)
1992 goto out2;
1993
1994 err = register_filesystem(&fuse_fs_type);
1995 if (err)
1996 goto out3;
1997
1998 return 0;
1999
2000 out3:
2001 unregister_fuseblk();
2002 out2:
2003 kmem_cache_destroy(fuse_inode_cachep);
2004 out:
2005 return err;
2006}
2007
2008static void fuse_fs_cleanup(void)
2009{
2010 unregister_filesystem(&fuse_fs_type);
2011 unregister_fuseblk();
2012
2013 /*
2014 * Make sure all delayed rcu free inodes are flushed before we
2015 * destroy cache.
2016 */
2017 rcu_barrier();
2018 kmem_cache_destroy(fuse_inode_cachep);
2019}
2020
2021static struct kobject *fuse_kobj;
2022
2023static int fuse_sysfs_init(void)
2024{
2025 int err;
2026
2027 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
2028 if (!fuse_kobj) {
2029 err = -ENOMEM;
2030 goto out_err;
2031 }
2032
2033 err = sysfs_create_mount_point(fuse_kobj, "connections");
2034 if (err)
2035 goto out_fuse_unregister;
2036
2037 return 0;
2038
2039 out_fuse_unregister:
2040 kobject_put(fuse_kobj);
2041 out_err:
2042 return err;
2043}
2044
2045static void fuse_sysfs_cleanup(void)
2046{
2047 sysfs_remove_mount_point(fuse_kobj, "connections");
2048 kobject_put(fuse_kobj);
2049}
2050
2051static int __init fuse_init(void)
2052{
2053 int res;
2054
2055 pr_info("init (API version %i.%i)\n",
2056 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
2057
2058 INIT_LIST_HEAD(&fuse_conn_list);
2059 res = fuse_fs_init();
2060 if (res)
2061 goto err;
2062
2063 res = fuse_dev_init();
2064 if (res)
2065 goto err_fs_cleanup;
2066
2067 res = fuse_sysfs_init();
2068 if (res)
2069 goto err_dev_cleanup;
2070
2071 res = fuse_ctl_init();
2072 if (res)
2073 goto err_sysfs_cleanup;
2074
2075 sanitize_global_limit(&max_user_bgreq);
2076 sanitize_global_limit(&max_user_congthresh);
2077
2078 return 0;
2079
2080 err_sysfs_cleanup:
2081 fuse_sysfs_cleanup();
2082 err_dev_cleanup:
2083 fuse_dev_cleanup();
2084 err_fs_cleanup:
2085 fuse_fs_cleanup();
2086 err:
2087 return res;
2088}
2089
2090static void __exit fuse_exit(void)
2091{
2092 pr_debug("exit\n");
2093
2094 fuse_ctl_cleanup();
2095 fuse_sysfs_cleanup();
2096 fuse_fs_cleanup();
2097 fuse_dev_cleanup();
2098}
2099
2100module_init(fuse_init);
2101module_exit(fuse_exit);