2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/seq_file.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/moduleparam.h>
18 #include <linux/parser.h>
19 #include <linux/statfs.h>
20 #include <linux/random.h>
21 #include <linux/sched.h>
22 #include <linux/exportfs.h>
23 #include <linux/posix_acl.h>
24 #include <linux/pid_namespace.h>
26 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
27 MODULE_DESCRIPTION("Filesystem in Userspace");
28 MODULE_LICENSE("GPL");
30 static struct kmem_cache *fuse_inode_cachep;
31 struct list_head fuse_conn_list;
32 DEFINE_MUTEX(fuse_mutex);
34 static int set_global_limit(const char *val, const struct kernel_param *kp);
36 unsigned max_user_bgreq;
37 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
38 &max_user_bgreq, 0644);
39 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
40 MODULE_PARM_DESC(max_user_bgreq,
41 "Global limit for the maximum number of backgrounded requests an "
42 "unprivileged user can set");
44 unsigned max_user_congthresh;
45 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
46 &max_user_congthresh, 0644);
47 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
48 MODULE_PARM_DESC(max_user_congthresh,
49 "Global limit for the maximum congestion threshold an "
50 "unprivileged user can set");
52 #define FUSE_SUPER_MAGIC 0x65735546
54 #define FUSE_DEFAULT_BLKSIZE 512
56 /** Maximum number of outstanding background requests */
57 #define FUSE_DEFAULT_MAX_BACKGROUND 12
59 /** Congestion starts at 75% of maximum */
60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
62 struct fuse_mount_data {
67 unsigned fd_present:1;
68 unsigned rootmode_present:1;
69 unsigned user_id_present:1;
70 unsigned group_id_present:1;
71 unsigned default_permissions:1;
72 unsigned allow_other:1;
77 struct fuse_forget_link *fuse_alloc_forget(void)
79 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
82 static struct inode *fuse_alloc_inode(struct super_block *sb)
85 struct fuse_inode *fi;
87 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
91 fi = get_fuse_inode(inode);
99 mutex_init(&fi->mutex);
100 spin_lock_init(&fi->lock);
101 fi->forget = fuse_alloc_forget();
103 kmem_cache_free(fuse_inode_cachep, inode);
110 static void fuse_i_callback(struct rcu_head *head)
112 struct inode *inode = container_of(head, struct inode, i_rcu);
113 kmem_cache_free(fuse_inode_cachep, inode);
116 static void fuse_destroy_inode(struct inode *inode)
118 struct fuse_inode *fi = get_fuse_inode(inode);
119 if (S_ISREG(inode->i_mode) && !is_bad_inode(inode)) {
120 WARN_ON(!list_empty(&fi->write_files));
121 WARN_ON(!list_empty(&fi->queued_writes));
123 mutex_destroy(&fi->mutex);
125 call_rcu(&inode->i_rcu, fuse_i_callback);
128 static void fuse_evict_inode(struct inode *inode)
130 truncate_inode_pages_final(&inode->i_data);
132 if (inode->i_sb->s_flags & SB_ACTIVE) {
133 struct fuse_conn *fc = get_fuse_conn(inode);
134 struct fuse_inode *fi = get_fuse_inode(inode);
135 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
140 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
143 if (*flags & SB_MANDLOCK)
150 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
151 * so that it will fit.
153 static ino_t fuse_squash_ino(u64 ino64)
155 ino_t ino = (ino_t) ino64;
156 if (sizeof(ino_t) < sizeof(u64))
157 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
161 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
164 struct fuse_conn *fc = get_fuse_conn(inode);
165 struct fuse_inode *fi = get_fuse_inode(inode);
167 lockdep_assert_held(&fi->lock);
169 fi->attr_version = atomic64_inc_return(&fc->attr_version);
170 fi->i_time = attr_valid;
171 WRITE_ONCE(fi->inval_mask, 0);
173 inode->i_ino = fuse_squash_ino(attr->ino);
174 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
175 set_nlink(inode, attr->nlink);
176 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
177 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
178 inode->i_blocks = attr->blocks;
179 inode->i_atime.tv_sec = attr->atime;
180 inode->i_atime.tv_nsec = attr->atimensec;
181 /* mtime from server may be stale due to local buffered write */
182 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
183 inode->i_mtime.tv_sec = attr->mtime;
184 inode->i_mtime.tv_nsec = attr->mtimensec;
185 inode->i_ctime.tv_sec = attr->ctime;
186 inode->i_ctime.tv_nsec = attr->ctimensec;
189 if (attr->blksize != 0)
190 inode->i_blkbits = ilog2(attr->blksize);
192 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
195 * Don't set the sticky bit in i_mode, unless we want the VFS
196 * to check permissions. This prevents failures due to the
197 * check in may_delete().
199 fi->orig_i_mode = inode->i_mode;
200 if (!fc->default_permissions)
201 inode->i_mode &= ~S_ISVTX;
203 fi->orig_ino = attr->ino;
206 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
207 u64 attr_valid, u64 attr_version)
209 struct fuse_conn *fc = get_fuse_conn(inode);
210 struct fuse_inode *fi = get_fuse_inode(inode);
211 bool is_wb = fc->writeback_cache;
213 struct timespec64 old_mtime;
215 spin_lock(&fi->lock);
216 if ((attr_version != 0 && fi->attr_version > attr_version) ||
217 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
218 spin_unlock(&fi->lock);
222 old_mtime = inode->i_mtime;
223 fuse_change_attributes_common(inode, attr, attr_valid);
225 oldsize = inode->i_size;
227 * In case of writeback_cache enabled, the cached writes beyond EOF
228 * extend local i_size without keeping userspace server in sync. So,
229 * attr->size coming from server can be stale. We cannot trust it.
231 if (!is_wb || !S_ISREG(inode->i_mode))
232 i_size_write(inode, attr->size);
233 spin_unlock(&fi->lock);
235 if (!is_wb && S_ISREG(inode->i_mode)) {
238 if (oldsize != attr->size) {
239 truncate_pagecache(inode, attr->size);
241 } else if (fc->auto_inval_data) {
242 struct timespec64 new_mtime = {
243 .tv_sec = attr->mtime,
244 .tv_nsec = attr->mtimensec,
248 * Auto inval mode also checks and invalidates if mtime
251 if (!timespec64_equal(&old_mtime, &new_mtime))
256 invalidate_inode_pages2(inode->i_mapping);
260 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
262 inode->i_mode = attr->mode & S_IFMT;
263 inode->i_size = attr->size;
264 inode->i_mtime.tv_sec = attr->mtime;
265 inode->i_mtime.tv_nsec = attr->mtimensec;
266 inode->i_ctime.tv_sec = attr->ctime;
267 inode->i_ctime.tv_nsec = attr->ctimensec;
268 if (S_ISREG(inode->i_mode)) {
269 fuse_init_common(inode);
270 fuse_init_file_inode(inode);
271 } else if (S_ISDIR(inode->i_mode))
272 fuse_init_dir(inode);
273 else if (S_ISLNK(inode->i_mode))
274 fuse_init_symlink(inode);
275 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
276 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
277 fuse_init_common(inode);
278 init_special_inode(inode, inode->i_mode,
279 new_decode_dev(attr->rdev));
284 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
286 u64 nodeid = *(u64 *) _nodeidp;
287 if (get_node_id(inode) == nodeid)
293 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
295 u64 nodeid = *(u64 *) _nodeidp;
296 get_fuse_inode(inode)->nodeid = nodeid;
300 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
301 int generation, struct fuse_attr *attr,
302 u64 attr_valid, u64 attr_version)
305 struct fuse_inode *fi;
306 struct fuse_conn *fc = get_fuse_conn_super(sb);
309 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
313 if ((inode->i_state & I_NEW)) {
314 inode->i_flags |= S_NOATIME;
315 if (!fc->writeback_cache || !S_ISREG(attr->mode))
316 inode->i_flags |= S_NOCMTIME;
317 inode->i_generation = generation;
318 fuse_init_inode(inode, attr);
319 unlock_new_inode(inode);
320 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
321 /* Inode has changed type, any I/O on the old should fail */
322 make_bad_inode(inode);
327 fi = get_fuse_inode(inode);
328 spin_lock(&fi->lock);
330 spin_unlock(&fi->lock);
331 fuse_change_attributes(inode, attr, attr_valid, attr_version);
336 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
337 loff_t offset, loff_t len)
343 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
347 fuse_invalidate_attr(inode);
348 forget_all_cached_acls(inode);
350 pg_start = offset >> PAGE_SHIFT;
354 pg_end = (offset + len - 1) >> PAGE_SHIFT;
355 invalidate_inode_pages2_range(inode->i_mapping,
362 bool fuse_lock_inode(struct inode *inode)
366 if (!get_fuse_conn(inode)->parallel_dirops) {
367 mutex_lock(&get_fuse_inode(inode)->mutex);
374 void fuse_unlock_inode(struct inode *inode, bool locked)
377 mutex_unlock(&get_fuse_inode(inode)->mutex);
380 static void fuse_umount_begin(struct super_block *sb)
382 fuse_abort_conn(get_fuse_conn_super(sb));
385 static void fuse_send_destroy(struct fuse_conn *fc)
387 struct fuse_req *req = fc->destroy_req;
388 if (req && fc->conn_init) {
389 fc->destroy_req = NULL;
390 req->in.h.opcode = FUSE_DESTROY;
391 __set_bit(FR_FORCE, &req->flags);
392 __clear_bit(FR_BACKGROUND, &req->flags);
393 fuse_request_send(fc, req);
394 fuse_put_request(fc, req);
398 static void fuse_put_super(struct super_block *sb)
400 struct fuse_conn *fc = get_fuse_conn_super(sb);
402 mutex_lock(&fuse_mutex);
403 list_del(&fc->entry);
404 fuse_ctl_remove_conn(fc);
405 mutex_unlock(&fuse_mutex);
410 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
412 stbuf->f_type = FUSE_SUPER_MAGIC;
413 stbuf->f_bsize = attr->bsize;
414 stbuf->f_frsize = attr->frsize;
415 stbuf->f_blocks = attr->blocks;
416 stbuf->f_bfree = attr->bfree;
417 stbuf->f_bavail = attr->bavail;
418 stbuf->f_files = attr->files;
419 stbuf->f_ffree = attr->ffree;
420 stbuf->f_namelen = attr->namelen;
421 /* fsid is left zero */
424 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
426 struct super_block *sb = dentry->d_sb;
427 struct fuse_conn *fc = get_fuse_conn_super(sb);
429 struct fuse_statfs_out outarg;
432 if (!fuse_allow_current_process(fc)) {
433 buf->f_type = FUSE_SUPER_MAGIC;
437 memset(&outarg, 0, sizeof(outarg));
439 args.in.h.opcode = FUSE_STATFS;
440 args.in.h.nodeid = get_node_id(d_inode(dentry));
441 args.out.numargs = 1;
442 args.out.args[0].size = sizeof(outarg);
443 args.out.args[0].value = &outarg;
444 err = fuse_simple_request(fc, &args);
446 convert_fuse_statfs(buf, &outarg.st);
455 OPT_DEFAULT_PERMISSIONS,
462 static const match_table_t tokens = {
464 {OPT_ROOTMODE, "rootmode=%o"},
465 {OPT_USER_ID, "user_id=%u"},
466 {OPT_GROUP_ID, "group_id=%u"},
467 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
468 {OPT_ALLOW_OTHER, "allow_other"},
469 {OPT_MAX_READ, "max_read=%u"},
470 {OPT_BLKSIZE, "blksize=%u"},
474 static int fuse_match_uint(substring_t *s, unsigned int *res)
477 char *buf = match_strdup(s);
479 err = kstrtouint(buf, 10, res);
485 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev,
486 struct user_namespace *user_ns)
489 memset(d, 0, sizeof(struct fuse_mount_data));
491 d->blksize = FUSE_DEFAULT_BLKSIZE;
493 while ((p = strsep(&opt, ",")) != NULL) {
497 substring_t args[MAX_OPT_ARGS];
501 token = match_token(p, tokens, args);
504 if (match_int(&args[0], &value))
511 if (match_octal(&args[0], &value))
513 if (!fuse_valid_type(value))
516 d->rootmode_present = 1;
520 if (fuse_match_uint(&args[0], &uv))
522 d->user_id = make_kuid(user_ns, uv);
523 if (!uid_valid(d->user_id))
525 d->user_id_present = 1;
529 if (fuse_match_uint(&args[0], &uv))
531 d->group_id = make_kgid(user_ns, uv);
532 if (!gid_valid(d->group_id))
534 d->group_id_present = 1;
537 case OPT_DEFAULT_PERMISSIONS:
538 d->default_permissions = 1;
541 case OPT_ALLOW_OTHER:
546 if (match_int(&args[0], &value))
552 if (!is_bdev || match_int(&args[0], &value))
562 if (!d->fd_present || !d->rootmode_present ||
563 !d->user_id_present || !d->group_id_present)
569 static int fuse_show_options(struct seq_file *m, struct dentry *root)
571 struct super_block *sb = root->d_sb;
572 struct fuse_conn *fc = get_fuse_conn_super(sb);
574 seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
575 seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
576 if (fc->default_permissions)
577 seq_puts(m, ",default_permissions");
579 seq_puts(m, ",allow_other");
580 if (fc->max_read != ~0)
581 seq_printf(m, ",max_read=%u", fc->max_read);
582 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
583 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
587 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
589 memset(fiq, 0, sizeof(struct fuse_iqueue));
590 init_waitqueue_head(&fiq->waitq);
591 INIT_LIST_HEAD(&fiq->pending);
592 INIT_LIST_HEAD(&fiq->interrupts);
593 fiq->forget_list_tail = &fiq->forget_list_head;
597 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
601 spin_lock_init(&fpq->lock);
602 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
603 INIT_LIST_HEAD(&fpq->processing[i]);
604 INIT_LIST_HEAD(&fpq->io);
608 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns)
610 memset(fc, 0, sizeof(*fc));
611 spin_lock_init(&fc->lock);
612 spin_lock_init(&fc->bg_lock);
613 init_rwsem(&fc->killsb);
614 refcount_set(&fc->count, 1);
615 atomic_set(&fc->dev_count, 1);
616 init_waitqueue_head(&fc->blocked_waitq);
617 init_waitqueue_head(&fc->reserved_req_waitq);
618 fuse_iqueue_init(&fc->iq);
619 INIT_LIST_HEAD(&fc->bg_queue);
620 INIT_LIST_HEAD(&fc->entry);
621 INIT_LIST_HEAD(&fc->devices);
622 atomic_set(&fc->num_waiting, 0);
623 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
624 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
625 atomic64_set(&fc->khctr, 0);
626 fc->polled_files = RB_ROOT;
630 atomic64_set(&fc->attr_version, 1);
631 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
632 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
633 fc->user_ns = get_user_ns(user_ns);
634 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
636 EXPORT_SYMBOL_GPL(fuse_conn_init);
638 void fuse_conn_put(struct fuse_conn *fc)
640 if (refcount_dec_and_test(&fc->count)) {
642 fuse_request_free(fc->destroy_req);
643 put_pid_ns(fc->pid_ns);
644 put_user_ns(fc->user_ns);
648 EXPORT_SYMBOL_GPL(fuse_conn_put);
650 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
652 refcount_inc(&fc->count);
655 EXPORT_SYMBOL_GPL(fuse_conn_get);
657 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
659 struct fuse_attr attr;
660 memset(&attr, 0, sizeof(attr));
663 attr.ino = FUSE_ROOT_ID;
665 return fuse_iget(sb, 1, 0, &attr, 0, 0);
668 struct fuse_inode_handle {
673 static struct dentry *fuse_get_dentry(struct super_block *sb,
674 struct fuse_inode_handle *handle)
676 struct fuse_conn *fc = get_fuse_conn_super(sb);
678 struct dentry *entry;
681 if (handle->nodeid == 0)
684 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
686 struct fuse_entry_out outarg;
687 const struct qstr name = QSTR_INIT(".", 1);
689 if (!fc->export_support)
692 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
694 if (err && err != -ENOENT)
701 if (get_node_id(inode) != handle->nodeid)
705 if (inode->i_generation != handle->generation)
708 entry = d_obtain_alias(inode);
709 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
710 fuse_invalidate_entry_cache(entry);
720 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
721 struct inode *parent)
723 int len = parent ? 6 : 3;
727 if (*max_len < len) {
729 return FILEID_INVALID;
732 nodeid = get_fuse_inode(inode)->nodeid;
733 generation = inode->i_generation;
735 fh[0] = (u32)(nodeid >> 32);
736 fh[1] = (u32)(nodeid & 0xffffffff);
740 nodeid = get_fuse_inode(parent)->nodeid;
741 generation = parent->i_generation;
743 fh[3] = (u32)(nodeid >> 32);
744 fh[4] = (u32)(nodeid & 0xffffffff);
749 return parent ? 0x82 : 0x81;
752 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
753 struct fid *fid, int fh_len, int fh_type)
755 struct fuse_inode_handle handle;
757 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
760 handle.nodeid = (u64) fid->raw[0] << 32;
761 handle.nodeid |= (u64) fid->raw[1];
762 handle.generation = fid->raw[2];
763 return fuse_get_dentry(sb, &handle);
766 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
767 struct fid *fid, int fh_len, int fh_type)
769 struct fuse_inode_handle parent;
771 if (fh_type != 0x82 || fh_len < 6)
774 parent.nodeid = (u64) fid->raw[3] << 32;
775 parent.nodeid |= (u64) fid->raw[4];
776 parent.generation = fid->raw[5];
777 return fuse_get_dentry(sb, &parent);
780 static struct dentry *fuse_get_parent(struct dentry *child)
782 struct inode *child_inode = d_inode(child);
783 struct fuse_conn *fc = get_fuse_conn(child_inode);
785 struct dentry *parent;
786 struct fuse_entry_out outarg;
787 const struct qstr name = QSTR_INIT("..", 2);
790 if (!fc->export_support)
791 return ERR_PTR(-ESTALE);
793 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
794 &name, &outarg, &inode);
797 return ERR_PTR(-ESTALE);
801 parent = d_obtain_alias(inode);
802 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
803 fuse_invalidate_entry_cache(parent);
808 static const struct export_operations fuse_export_operations = {
809 .fh_to_dentry = fuse_fh_to_dentry,
810 .fh_to_parent = fuse_fh_to_parent,
811 .encode_fh = fuse_encode_fh,
812 .get_parent = fuse_get_parent,
815 static const struct super_operations fuse_super_operations = {
816 .alloc_inode = fuse_alloc_inode,
817 .destroy_inode = fuse_destroy_inode,
818 .evict_inode = fuse_evict_inode,
819 .write_inode = fuse_write_inode,
820 .drop_inode = generic_delete_inode,
821 .remount_fs = fuse_remount_fs,
822 .put_super = fuse_put_super,
823 .umount_begin = fuse_umount_begin,
824 .statfs = fuse_statfs,
825 .show_options = fuse_show_options,
828 static void sanitize_global_limit(unsigned *limit)
831 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) /
832 sizeof(struct fuse_req);
834 if (*limit >= 1 << 16)
835 *limit = (1 << 16) - 1;
838 static int set_global_limit(const char *val, const struct kernel_param *kp)
842 rv = param_set_uint(val, kp);
846 sanitize_global_limit((unsigned *)kp->arg);
851 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
853 int cap_sys_admin = capable(CAP_SYS_ADMIN);
858 sanitize_global_limit(&max_user_bgreq);
859 sanitize_global_limit(&max_user_congthresh);
861 spin_lock(&fc->bg_lock);
862 if (arg->max_background) {
863 fc->max_background = arg->max_background;
865 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
866 fc->max_background = max_user_bgreq;
868 if (arg->congestion_threshold) {
869 fc->congestion_threshold = arg->congestion_threshold;
871 if (!cap_sys_admin &&
872 fc->congestion_threshold > max_user_congthresh)
873 fc->congestion_threshold = max_user_congthresh;
875 spin_unlock(&fc->bg_lock);
878 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
880 struct fuse_init_out *arg = &req->misc.init_out;
882 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
885 unsigned long ra_pages;
887 process_init_limits(fc, arg);
889 if (arg->minor >= 6) {
890 ra_pages = arg->max_readahead / PAGE_SIZE;
891 if (arg->flags & FUSE_ASYNC_READ)
893 if (!(arg->flags & FUSE_POSIX_LOCKS))
895 if (arg->minor >= 17) {
896 if (!(arg->flags & FUSE_FLOCK_LOCKS))
899 if (!(arg->flags & FUSE_POSIX_LOCKS))
902 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
903 fc->atomic_o_trunc = 1;
904 if (arg->minor >= 9) {
905 /* LOOKUP has dependency on proto version */
906 if (arg->flags & FUSE_EXPORT_SUPPORT)
907 fc->export_support = 1;
909 if (arg->flags & FUSE_BIG_WRITES)
911 if (arg->flags & FUSE_DONT_MASK)
913 if (arg->flags & FUSE_AUTO_INVAL_DATA)
914 fc->auto_inval_data = 1;
915 if (arg->flags & FUSE_DO_READDIRPLUS) {
916 fc->do_readdirplus = 1;
917 if (arg->flags & FUSE_READDIRPLUS_AUTO)
918 fc->readdirplus_auto = 1;
920 if (arg->flags & FUSE_ASYNC_DIO)
922 if (arg->flags & FUSE_WRITEBACK_CACHE)
923 fc->writeback_cache = 1;
924 if (arg->flags & FUSE_PARALLEL_DIROPS)
925 fc->parallel_dirops = 1;
926 if (arg->flags & FUSE_HANDLE_KILLPRIV)
927 fc->handle_killpriv = 1;
928 if (arg->time_gran && arg->time_gran <= 1000000000)
929 fc->sb->s_time_gran = arg->time_gran;
930 if ((arg->flags & FUSE_POSIX_ACL)) {
931 fc->default_permissions = 1;
933 fc->sb->s_xattr = fuse_acl_xattr_handlers;
935 if (arg->flags & FUSE_CACHE_SYMLINKS)
936 fc->cache_symlinks = 1;
937 if (arg->flags & FUSE_ABORT_ERROR)
939 if (arg->flags & FUSE_MAX_PAGES) {
941 min_t(unsigned int, FUSE_MAX_MAX_PAGES,
942 max_t(unsigned int, arg->max_pages, 1));
945 ra_pages = fc->max_read / PAGE_SIZE;
950 fc->sb->s_bdi->ra_pages =
951 min(fc->sb->s_bdi->ra_pages, ra_pages);
952 fc->minor = arg->minor;
953 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
954 fc->max_write = max_t(unsigned, 4096, fc->max_write);
957 fuse_set_initialized(fc);
958 wake_up_all(&fc->blocked_waitq);
961 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
963 struct fuse_init_in *arg = &req->misc.init_in;
965 arg->major = FUSE_KERNEL_VERSION;
966 arg->minor = FUSE_KERNEL_MINOR_VERSION;
967 arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
968 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
969 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
970 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
971 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
972 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
973 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
974 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
975 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
976 FUSE_NO_OPENDIR_SUPPORT;
977 req->in.h.opcode = FUSE_INIT;
979 req->in.args[0].size = sizeof(*arg);
980 req->in.args[0].value = arg;
981 req->out.numargs = 1;
982 /* Variable length argument used for backward compatibility
983 with interface version < 7.5. Rest of init_out is zeroed
984 by do_get_request(), so a short reply is not a problem */
986 req->out.args[0].size = sizeof(struct fuse_init_out);
987 req->out.args[0].value = &req->misc.init_out;
988 req->end = process_init_reply;
989 fuse_request_send_background(fc, req);
992 static void fuse_free_conn(struct fuse_conn *fc)
994 WARN_ON(!list_empty(&fc->devices));
998 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1004 suffix = "-fuseblk";
1006 * sb->s_bdi points to blkdev's bdi however we want to redirect
1007 * it to our private bdi...
1010 sb->s_bdi = &noop_backing_dev_info;
1012 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1013 MINOR(fc->dev), suffix);
1017 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1018 /* fuse does it's own writeback accounting */
1019 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1022 * For a single fuse filesystem use max 1% of dirty +
1023 * writeback threshold.
1025 * This gives about 1M of write buffer for memory maps on a
1026 * machine with 1G and 10% dirty_ratio, which should be more
1029 * Privileged users can raise it by writing to
1031 * /sys/class/bdi/<bdi>/max_ratio
1033 bdi_set_max_ratio(sb->s_bdi, 1);
1038 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1040 struct fuse_dev *fud;
1041 struct list_head *pq;
1043 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1047 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1053 fud->pq.processing = pq;
1054 fud->fc = fuse_conn_get(fc);
1055 fuse_pqueue_init(&fud->pq);
1057 spin_lock(&fc->lock);
1058 list_add_tail(&fud->entry, &fc->devices);
1059 spin_unlock(&fc->lock);
1063 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1065 void fuse_dev_free(struct fuse_dev *fud)
1067 struct fuse_conn *fc = fud->fc;
1070 spin_lock(&fc->lock);
1071 list_del(&fud->entry);
1072 spin_unlock(&fc->lock);
1076 kfree(fud->pq.processing);
1079 EXPORT_SYMBOL_GPL(fuse_dev_free);
1081 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1083 struct fuse_dev *fud;
1084 struct fuse_conn *fc;
1086 struct fuse_mount_data d;
1088 struct dentry *root_dentry;
1089 struct fuse_req *init_req;
1091 int is_bdev = sb->s_bdev != NULL;
1094 if (sb->s_flags & SB_MANDLOCK)
1097 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1099 if (!parse_fuse_opt(data, &d, is_bdev, sb->s_user_ns))
1105 if (!sb_set_blocksize(sb, d.blksize))
1109 sb->s_blocksize = PAGE_SIZE;
1110 sb->s_blocksize_bits = PAGE_SHIFT;
1112 sb->s_magic = FUSE_SUPER_MAGIC;
1113 sb->s_op = &fuse_super_operations;
1114 sb->s_xattr = fuse_xattr_handlers;
1115 sb->s_maxbytes = MAX_LFS_FILESIZE;
1116 sb->s_time_gran = 1;
1117 sb->s_export_op = &fuse_export_operations;
1118 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1119 if (sb->s_user_ns != &init_user_ns)
1120 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1128 * Require mount to happen from the same user namespace which
1129 * opened /dev/fuse to prevent potential attacks.
1131 if (file->f_op != &fuse_dev_operations ||
1132 file->f_cred->user_ns != sb->s_user_ns)
1136 * If we are not in the initial user namespace posix
1137 * acls must be translated.
1139 if (sb->s_user_ns != &init_user_ns)
1140 sb->s_xattr = fuse_no_acl_xattr_handlers;
1142 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1147 fuse_conn_init(fc, sb->s_user_ns);
1148 fc->release = fuse_free_conn;
1150 fud = fuse_dev_alloc(fc);
1154 fc->dev = sb->s_dev;
1156 err = fuse_bdi_init(fc, sb);
1160 /* Handle umasking inside the fuse code */
1161 if (sb->s_flags & SB_POSIXACL)
1163 sb->s_flags |= SB_POSIXACL;
1165 fc->default_permissions = d.default_permissions;
1166 fc->allow_other = d.allow_other;
1167 fc->user_id = d.user_id;
1168 fc->group_id = d.group_id;
1169 fc->max_read = max_t(unsigned, 4096, d.max_read);
1171 /* Used by get_root_inode() */
1175 root = fuse_get_root_inode(sb, d.rootmode);
1176 sb->s_d_op = &fuse_root_dentry_operations;
1177 root_dentry = d_make_root(root);
1180 /* Root dentry doesn't have .d_revalidate */
1181 sb->s_d_op = &fuse_dentry_operations;
1183 init_req = fuse_request_alloc(0);
1186 __set_bit(FR_BACKGROUND, &init_req->flags);
1189 fc->destroy_req = fuse_request_alloc(0);
1190 if (!fc->destroy_req)
1191 goto err_free_init_req;
1194 mutex_lock(&fuse_mutex);
1196 if (file->private_data)
1199 err = fuse_ctl_add_conn(fc);
1203 list_add_tail(&fc->entry, &fuse_conn_list);
1204 sb->s_root = root_dentry;
1205 file->private_data = fud;
1206 mutex_unlock(&fuse_mutex);
1208 * atomic_dec_and_test() in fput() provides the necessary
1209 * memory barrier for file->private_data to be visible on all
1214 fuse_send_init(fc, init_req);
1219 mutex_unlock(&fuse_mutex);
1221 fuse_request_free(init_req);
1228 sb->s_fs_info = NULL;
1235 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1236 int flags, const char *dev_name,
1239 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1242 static void fuse_sb_destroy(struct super_block *sb)
1244 struct fuse_conn *fc = get_fuse_conn_super(sb);
1247 fuse_send_destroy(fc);
1249 fuse_abort_conn(fc);
1250 fuse_wait_aborted(fc);
1252 down_write(&fc->killsb);
1254 up_write(&fc->killsb);
1258 static void fuse_kill_sb_anon(struct super_block *sb)
1260 fuse_sb_destroy(sb);
1261 kill_anon_super(sb);
1264 static struct file_system_type fuse_fs_type = {
1265 .owner = THIS_MODULE,
1267 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1268 .mount = fuse_mount,
1269 .kill_sb = fuse_kill_sb_anon,
1271 MODULE_ALIAS_FS("fuse");
1274 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1275 int flags, const char *dev_name,
1278 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1281 static void fuse_kill_sb_blk(struct super_block *sb)
1283 fuse_sb_destroy(sb);
1284 kill_block_super(sb);
1287 static struct file_system_type fuseblk_fs_type = {
1288 .owner = THIS_MODULE,
1290 .mount = fuse_mount_blk,
1291 .kill_sb = fuse_kill_sb_blk,
1292 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1294 MODULE_ALIAS_FS("fuseblk");
1296 static inline int register_fuseblk(void)
1298 return register_filesystem(&fuseblk_fs_type);
1301 static inline void unregister_fuseblk(void)
1303 unregister_filesystem(&fuseblk_fs_type);
1306 static inline int register_fuseblk(void)
1311 static inline void unregister_fuseblk(void)
1316 static void fuse_inode_init_once(void *foo)
1318 struct inode *inode = foo;
1320 inode_init_once(inode);
1323 static int __init fuse_fs_init(void)
1327 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1328 sizeof(struct fuse_inode), 0,
1329 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1330 fuse_inode_init_once);
1332 if (!fuse_inode_cachep)
1335 err = register_fuseblk();
1339 err = register_filesystem(&fuse_fs_type);
1346 unregister_fuseblk();
1348 kmem_cache_destroy(fuse_inode_cachep);
1353 static void fuse_fs_cleanup(void)
1355 unregister_filesystem(&fuse_fs_type);
1356 unregister_fuseblk();
1359 * Make sure all delayed rcu free inodes are flushed before we
1363 kmem_cache_destroy(fuse_inode_cachep);
1366 static struct kobject *fuse_kobj;
1368 static int fuse_sysfs_init(void)
1372 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1378 err = sysfs_create_mount_point(fuse_kobj, "connections");
1380 goto out_fuse_unregister;
1384 out_fuse_unregister:
1385 kobject_put(fuse_kobj);
1390 static void fuse_sysfs_cleanup(void)
1392 sysfs_remove_mount_point(fuse_kobj, "connections");
1393 kobject_put(fuse_kobj);
1396 static int __init fuse_init(void)
1400 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1401 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1403 INIT_LIST_HEAD(&fuse_conn_list);
1404 res = fuse_fs_init();
1408 res = fuse_dev_init();
1410 goto err_fs_cleanup;
1412 res = fuse_sysfs_init();
1414 goto err_dev_cleanup;
1416 res = fuse_ctl_init();
1418 goto err_sysfs_cleanup;
1420 sanitize_global_limit(&max_user_bgreq);
1421 sanitize_global_limit(&max_user_congthresh);
1426 fuse_sysfs_cleanup();
1435 static void __exit fuse_exit(void)
1437 printk(KERN_DEBUG "fuse exit\n");
1440 fuse_sysfs_cleanup();
1445 module_init(fuse_init);
1446 module_exit(fuse_exit);