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)
84 struct fuse_inode *fi;
86 fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
97 mutex_init(&fi->mutex);
98 spin_lock_init(&fi->lock);
99 fi->forget = fuse_alloc_forget();
101 kmem_cache_free(fuse_inode_cachep, fi);
108 static void fuse_i_callback(struct rcu_head *head)
110 struct inode *inode = container_of(head, struct inode, i_rcu);
111 kmem_cache_free(fuse_inode_cachep, inode);
114 static void fuse_destroy_inode(struct inode *inode)
116 struct fuse_inode *fi = get_fuse_inode(inode);
117 if (S_ISREG(inode->i_mode) && !is_bad_inode(inode)) {
118 WARN_ON(!list_empty(&fi->write_files));
119 WARN_ON(!list_empty(&fi->queued_writes));
121 mutex_destroy(&fi->mutex);
123 call_rcu(&inode->i_rcu, fuse_i_callback);
126 static void fuse_evict_inode(struct inode *inode)
128 truncate_inode_pages_final(&inode->i_data);
130 if (inode->i_sb->s_flags & SB_ACTIVE) {
131 struct fuse_conn *fc = get_fuse_conn(inode);
132 struct fuse_inode *fi = get_fuse_inode(inode);
133 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
138 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
141 if (*flags & SB_MANDLOCK)
148 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
149 * so that it will fit.
151 static ino_t fuse_squash_ino(u64 ino64)
153 ino_t ino = (ino_t) ino64;
154 if (sizeof(ino_t) < sizeof(u64))
155 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
159 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
162 struct fuse_conn *fc = get_fuse_conn(inode);
163 struct fuse_inode *fi = get_fuse_inode(inode);
165 lockdep_assert_held(&fi->lock);
167 fi->attr_version = atomic64_inc_return(&fc->attr_version);
168 fi->i_time = attr_valid;
169 WRITE_ONCE(fi->inval_mask, 0);
171 inode->i_ino = fuse_squash_ino(attr->ino);
172 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
173 set_nlink(inode, attr->nlink);
174 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
175 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
176 inode->i_blocks = attr->blocks;
177 inode->i_atime.tv_sec = attr->atime;
178 inode->i_atime.tv_nsec = attr->atimensec;
179 /* mtime from server may be stale due to local buffered write */
180 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
181 inode->i_mtime.tv_sec = attr->mtime;
182 inode->i_mtime.tv_nsec = attr->mtimensec;
183 inode->i_ctime.tv_sec = attr->ctime;
184 inode->i_ctime.tv_nsec = attr->ctimensec;
187 if (attr->blksize != 0)
188 inode->i_blkbits = ilog2(attr->blksize);
190 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
193 * Don't set the sticky bit in i_mode, unless we want the VFS
194 * to check permissions. This prevents failures due to the
195 * check in may_delete().
197 fi->orig_i_mode = inode->i_mode;
198 if (!fc->default_permissions)
199 inode->i_mode &= ~S_ISVTX;
201 fi->orig_ino = attr->ino;
204 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
205 u64 attr_valid, u64 attr_version)
207 struct fuse_conn *fc = get_fuse_conn(inode);
208 struct fuse_inode *fi = get_fuse_inode(inode);
209 bool is_wb = fc->writeback_cache;
211 struct timespec64 old_mtime;
213 spin_lock(&fi->lock);
214 if ((attr_version != 0 && fi->attr_version > attr_version) ||
215 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
216 spin_unlock(&fi->lock);
220 old_mtime = inode->i_mtime;
221 fuse_change_attributes_common(inode, attr, attr_valid);
223 oldsize = inode->i_size;
225 * In case of writeback_cache enabled, the cached writes beyond EOF
226 * extend local i_size without keeping userspace server in sync. So,
227 * attr->size coming from server can be stale. We cannot trust it.
229 if (!is_wb || !S_ISREG(inode->i_mode))
230 i_size_write(inode, attr->size);
231 spin_unlock(&fi->lock);
233 if (!is_wb && S_ISREG(inode->i_mode)) {
236 if (oldsize != attr->size) {
237 truncate_pagecache(inode, attr->size);
238 if (!fc->explicit_inval_data)
240 } else if (fc->auto_inval_data) {
241 struct timespec64 new_mtime = {
242 .tv_sec = attr->mtime,
243 .tv_nsec = attr->mtimensec,
247 * Auto inval mode also checks and invalidates if mtime
250 if (!timespec64_equal(&old_mtime, &new_mtime))
255 invalidate_inode_pages2(inode->i_mapping);
259 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
261 inode->i_mode = attr->mode & S_IFMT;
262 inode->i_size = attr->size;
263 inode->i_mtime.tv_sec = attr->mtime;
264 inode->i_mtime.tv_nsec = attr->mtimensec;
265 inode->i_ctime.tv_sec = attr->ctime;
266 inode->i_ctime.tv_nsec = attr->ctimensec;
267 if (S_ISREG(inode->i_mode)) {
268 fuse_init_common(inode);
269 fuse_init_file_inode(inode);
270 } else if (S_ISDIR(inode->i_mode))
271 fuse_init_dir(inode);
272 else if (S_ISLNK(inode->i_mode))
273 fuse_init_symlink(inode);
274 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
275 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
276 fuse_init_common(inode);
277 init_special_inode(inode, inode->i_mode,
278 new_decode_dev(attr->rdev));
283 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
285 u64 nodeid = *(u64 *) _nodeidp;
286 if (get_node_id(inode) == nodeid)
292 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
294 u64 nodeid = *(u64 *) _nodeidp;
295 get_fuse_inode(inode)->nodeid = nodeid;
299 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
300 int generation, struct fuse_attr *attr,
301 u64 attr_valid, u64 attr_version)
304 struct fuse_inode *fi;
305 struct fuse_conn *fc = get_fuse_conn_super(sb);
308 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
312 if ((inode->i_state & I_NEW)) {
313 inode->i_flags |= S_NOATIME;
314 if (!fc->writeback_cache || !S_ISREG(attr->mode))
315 inode->i_flags |= S_NOCMTIME;
316 inode->i_generation = generation;
317 fuse_init_inode(inode, attr);
318 unlock_new_inode(inode);
319 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
320 /* Inode has changed type, any I/O on the old should fail */
321 make_bad_inode(inode);
326 fi = get_fuse_inode(inode);
327 spin_lock(&fi->lock);
329 spin_unlock(&fi->lock);
330 fuse_change_attributes(inode, attr, attr_valid, attr_version);
335 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
336 loff_t offset, loff_t len)
342 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
346 fuse_invalidate_attr(inode);
347 forget_all_cached_acls(inode);
349 pg_start = offset >> PAGE_SHIFT;
353 pg_end = (offset + len - 1) >> PAGE_SHIFT;
354 invalidate_inode_pages2_range(inode->i_mapping,
361 bool fuse_lock_inode(struct inode *inode)
365 if (!get_fuse_conn(inode)->parallel_dirops) {
366 mutex_lock(&get_fuse_inode(inode)->mutex);
373 void fuse_unlock_inode(struct inode *inode, bool locked)
376 mutex_unlock(&get_fuse_inode(inode)->mutex);
379 static void fuse_umount_begin(struct super_block *sb)
381 fuse_abort_conn(get_fuse_conn_super(sb));
384 static void fuse_send_destroy(struct fuse_conn *fc)
386 struct fuse_req *req = fc->destroy_req;
387 if (req && fc->conn_init) {
388 fc->destroy_req = NULL;
389 req->in.h.opcode = FUSE_DESTROY;
390 __set_bit(FR_FORCE, &req->flags);
391 __clear_bit(FR_BACKGROUND, &req->flags);
392 fuse_request_send(fc, req);
393 fuse_put_request(fc, req);
397 static void fuse_put_super(struct super_block *sb)
399 struct fuse_conn *fc = get_fuse_conn_super(sb);
401 mutex_lock(&fuse_mutex);
402 list_del(&fc->entry);
403 fuse_ctl_remove_conn(fc);
404 mutex_unlock(&fuse_mutex);
409 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
411 stbuf->f_type = FUSE_SUPER_MAGIC;
412 stbuf->f_bsize = attr->bsize;
413 stbuf->f_frsize = attr->frsize;
414 stbuf->f_blocks = attr->blocks;
415 stbuf->f_bfree = attr->bfree;
416 stbuf->f_bavail = attr->bavail;
417 stbuf->f_files = attr->files;
418 stbuf->f_ffree = attr->ffree;
419 stbuf->f_namelen = attr->namelen;
420 /* fsid is left zero */
423 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
425 struct super_block *sb = dentry->d_sb;
426 struct fuse_conn *fc = get_fuse_conn_super(sb);
428 struct fuse_statfs_out outarg;
431 if (!fuse_allow_current_process(fc)) {
432 buf->f_type = FUSE_SUPER_MAGIC;
436 memset(&outarg, 0, sizeof(outarg));
438 args.in.h.opcode = FUSE_STATFS;
439 args.in.h.nodeid = get_node_id(d_inode(dentry));
440 args.out.numargs = 1;
441 args.out.args[0].size = sizeof(outarg);
442 args.out.args[0].value = &outarg;
443 err = fuse_simple_request(fc, &args);
445 convert_fuse_statfs(buf, &outarg.st);
454 OPT_DEFAULT_PERMISSIONS,
461 static const match_table_t tokens = {
463 {OPT_ROOTMODE, "rootmode=%o"},
464 {OPT_USER_ID, "user_id=%u"},
465 {OPT_GROUP_ID, "group_id=%u"},
466 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
467 {OPT_ALLOW_OTHER, "allow_other"},
468 {OPT_MAX_READ, "max_read=%u"},
469 {OPT_BLKSIZE, "blksize=%u"},
473 static int fuse_match_uint(substring_t *s, unsigned int *res)
476 char *buf = match_strdup(s);
478 err = kstrtouint(buf, 10, res);
484 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev,
485 struct user_namespace *user_ns)
488 memset(d, 0, sizeof(struct fuse_mount_data));
490 d->blksize = FUSE_DEFAULT_BLKSIZE;
492 while ((p = strsep(&opt, ",")) != NULL) {
496 substring_t args[MAX_OPT_ARGS];
500 token = match_token(p, tokens, args);
503 if (match_int(&args[0], &value))
510 if (match_octal(&args[0], &value))
512 if (!fuse_valid_type(value))
515 d->rootmode_present = 1;
519 if (fuse_match_uint(&args[0], &uv))
521 d->user_id = make_kuid(user_ns, uv);
522 if (!uid_valid(d->user_id))
524 d->user_id_present = 1;
528 if (fuse_match_uint(&args[0], &uv))
530 d->group_id = make_kgid(user_ns, uv);
531 if (!gid_valid(d->group_id))
533 d->group_id_present = 1;
536 case OPT_DEFAULT_PERMISSIONS:
537 d->default_permissions = 1;
540 case OPT_ALLOW_OTHER:
545 if (match_int(&args[0], &value))
551 if (!is_bdev || match_int(&args[0], &value))
561 if (!d->fd_present || !d->rootmode_present ||
562 !d->user_id_present || !d->group_id_present)
568 static int fuse_show_options(struct seq_file *m, struct dentry *root)
570 struct super_block *sb = root->d_sb;
571 struct fuse_conn *fc = get_fuse_conn_super(sb);
573 seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
574 seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
575 if (fc->default_permissions)
576 seq_puts(m, ",default_permissions");
578 seq_puts(m, ",allow_other");
579 if (fc->max_read != ~0)
580 seq_printf(m, ",max_read=%u", fc->max_read);
581 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
582 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
586 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
588 memset(fiq, 0, sizeof(struct fuse_iqueue));
589 init_waitqueue_head(&fiq->waitq);
590 INIT_LIST_HEAD(&fiq->pending);
591 INIT_LIST_HEAD(&fiq->interrupts);
592 fiq->forget_list_tail = &fiq->forget_list_head;
596 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
600 spin_lock_init(&fpq->lock);
601 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
602 INIT_LIST_HEAD(&fpq->processing[i]);
603 INIT_LIST_HEAD(&fpq->io);
607 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns)
609 memset(fc, 0, sizeof(*fc));
610 spin_lock_init(&fc->lock);
611 spin_lock_init(&fc->bg_lock);
612 init_rwsem(&fc->killsb);
613 refcount_set(&fc->count, 1);
614 atomic_set(&fc->dev_count, 1);
615 init_waitqueue_head(&fc->blocked_waitq);
616 init_waitqueue_head(&fc->reserved_req_waitq);
617 fuse_iqueue_init(&fc->iq);
618 INIT_LIST_HEAD(&fc->bg_queue);
619 INIT_LIST_HEAD(&fc->entry);
620 INIT_LIST_HEAD(&fc->devices);
621 atomic_set(&fc->num_waiting, 0);
622 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
623 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
624 atomic64_set(&fc->khctr, 0);
625 fc->polled_files = RB_ROOT;
629 atomic64_set(&fc->attr_version, 1);
630 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
631 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
632 fc->user_ns = get_user_ns(user_ns);
633 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
635 EXPORT_SYMBOL_GPL(fuse_conn_init);
637 void fuse_conn_put(struct fuse_conn *fc)
639 if (refcount_dec_and_test(&fc->count)) {
641 fuse_request_free(fc->destroy_req);
642 put_pid_ns(fc->pid_ns);
643 put_user_ns(fc->user_ns);
647 EXPORT_SYMBOL_GPL(fuse_conn_put);
649 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
651 refcount_inc(&fc->count);
654 EXPORT_SYMBOL_GPL(fuse_conn_get);
656 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
658 struct fuse_attr attr;
659 memset(&attr, 0, sizeof(attr));
662 attr.ino = FUSE_ROOT_ID;
664 return fuse_iget(sb, 1, 0, &attr, 0, 0);
667 struct fuse_inode_handle {
672 static struct dentry *fuse_get_dentry(struct super_block *sb,
673 struct fuse_inode_handle *handle)
675 struct fuse_conn *fc = get_fuse_conn_super(sb);
677 struct dentry *entry;
680 if (handle->nodeid == 0)
683 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
685 struct fuse_entry_out outarg;
686 const struct qstr name = QSTR_INIT(".", 1);
688 if (!fc->export_support)
691 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
693 if (err && err != -ENOENT)
700 if (get_node_id(inode) != handle->nodeid)
704 if (inode->i_generation != handle->generation)
707 entry = d_obtain_alias(inode);
708 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
709 fuse_invalidate_entry_cache(entry);
719 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
720 struct inode *parent)
722 int len = parent ? 6 : 3;
726 if (*max_len < len) {
728 return FILEID_INVALID;
731 nodeid = get_fuse_inode(inode)->nodeid;
732 generation = inode->i_generation;
734 fh[0] = (u32)(nodeid >> 32);
735 fh[1] = (u32)(nodeid & 0xffffffff);
739 nodeid = get_fuse_inode(parent)->nodeid;
740 generation = parent->i_generation;
742 fh[3] = (u32)(nodeid >> 32);
743 fh[4] = (u32)(nodeid & 0xffffffff);
748 return parent ? 0x82 : 0x81;
751 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
752 struct fid *fid, int fh_len, int fh_type)
754 struct fuse_inode_handle handle;
756 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
759 handle.nodeid = (u64) fid->raw[0] << 32;
760 handle.nodeid |= (u64) fid->raw[1];
761 handle.generation = fid->raw[2];
762 return fuse_get_dentry(sb, &handle);
765 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
766 struct fid *fid, int fh_len, int fh_type)
768 struct fuse_inode_handle parent;
770 if (fh_type != 0x82 || fh_len < 6)
773 parent.nodeid = (u64) fid->raw[3] << 32;
774 parent.nodeid |= (u64) fid->raw[4];
775 parent.generation = fid->raw[5];
776 return fuse_get_dentry(sb, &parent);
779 static struct dentry *fuse_get_parent(struct dentry *child)
781 struct inode *child_inode = d_inode(child);
782 struct fuse_conn *fc = get_fuse_conn(child_inode);
784 struct dentry *parent;
785 struct fuse_entry_out outarg;
786 const struct qstr name = QSTR_INIT("..", 2);
789 if (!fc->export_support)
790 return ERR_PTR(-ESTALE);
792 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
793 &name, &outarg, &inode);
796 return ERR_PTR(-ESTALE);
800 parent = d_obtain_alias(inode);
801 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
802 fuse_invalidate_entry_cache(parent);
807 static const struct export_operations fuse_export_operations = {
808 .fh_to_dentry = fuse_fh_to_dentry,
809 .fh_to_parent = fuse_fh_to_parent,
810 .encode_fh = fuse_encode_fh,
811 .get_parent = fuse_get_parent,
814 static const struct super_operations fuse_super_operations = {
815 .alloc_inode = fuse_alloc_inode,
816 .destroy_inode = fuse_destroy_inode,
817 .evict_inode = fuse_evict_inode,
818 .write_inode = fuse_write_inode,
819 .drop_inode = generic_delete_inode,
820 .remount_fs = fuse_remount_fs,
821 .put_super = fuse_put_super,
822 .umount_begin = fuse_umount_begin,
823 .statfs = fuse_statfs,
824 .show_options = fuse_show_options,
827 static void sanitize_global_limit(unsigned *limit)
830 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) /
831 sizeof(struct fuse_req);
833 if (*limit >= 1 << 16)
834 *limit = (1 << 16) - 1;
837 static int set_global_limit(const char *val, const struct kernel_param *kp)
841 rv = param_set_uint(val, kp);
845 sanitize_global_limit((unsigned *)kp->arg);
850 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
852 int cap_sys_admin = capable(CAP_SYS_ADMIN);
857 sanitize_global_limit(&max_user_bgreq);
858 sanitize_global_limit(&max_user_congthresh);
860 spin_lock(&fc->bg_lock);
861 if (arg->max_background) {
862 fc->max_background = arg->max_background;
864 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
865 fc->max_background = max_user_bgreq;
867 if (arg->congestion_threshold) {
868 fc->congestion_threshold = arg->congestion_threshold;
870 if (!cap_sys_admin &&
871 fc->congestion_threshold > max_user_congthresh)
872 fc->congestion_threshold = max_user_congthresh;
874 spin_unlock(&fc->bg_lock);
877 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
879 struct fuse_init_out *arg = &req->misc.init_out;
881 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
884 unsigned long ra_pages;
886 process_init_limits(fc, arg);
888 if (arg->minor >= 6) {
889 ra_pages = arg->max_readahead / PAGE_SIZE;
890 if (arg->flags & FUSE_ASYNC_READ)
892 if (!(arg->flags & FUSE_POSIX_LOCKS))
894 if (arg->minor >= 17) {
895 if (!(arg->flags & FUSE_FLOCK_LOCKS))
898 if (!(arg->flags & FUSE_POSIX_LOCKS))
901 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
902 fc->atomic_o_trunc = 1;
903 if (arg->minor >= 9) {
904 /* LOOKUP has dependency on proto version */
905 if (arg->flags & FUSE_EXPORT_SUPPORT)
906 fc->export_support = 1;
908 if (arg->flags & FUSE_BIG_WRITES)
910 if (arg->flags & FUSE_DONT_MASK)
912 if (arg->flags & FUSE_AUTO_INVAL_DATA)
913 fc->auto_inval_data = 1;
914 else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
915 fc->explicit_inval_data = 1;
916 if (arg->flags & FUSE_DO_READDIRPLUS) {
917 fc->do_readdirplus = 1;
918 if (arg->flags & FUSE_READDIRPLUS_AUTO)
919 fc->readdirplus_auto = 1;
921 if (arg->flags & FUSE_ASYNC_DIO)
923 if (arg->flags & FUSE_WRITEBACK_CACHE)
924 fc->writeback_cache = 1;
925 if (arg->flags & FUSE_PARALLEL_DIROPS)
926 fc->parallel_dirops = 1;
927 if (arg->flags & FUSE_HANDLE_KILLPRIV)
928 fc->handle_killpriv = 1;
929 if (arg->time_gran && arg->time_gran <= 1000000000)
930 fc->sb->s_time_gran = arg->time_gran;
931 if ((arg->flags & FUSE_POSIX_ACL)) {
932 fc->default_permissions = 1;
934 fc->sb->s_xattr = fuse_acl_xattr_handlers;
936 if (arg->flags & FUSE_CACHE_SYMLINKS)
937 fc->cache_symlinks = 1;
938 if (arg->flags & FUSE_ABORT_ERROR)
940 if (arg->flags & FUSE_MAX_PAGES) {
942 min_t(unsigned int, FUSE_MAX_MAX_PAGES,
943 max_t(unsigned int, arg->max_pages, 1));
946 ra_pages = fc->max_read / PAGE_SIZE;
951 fc->sb->s_bdi->ra_pages =
952 min(fc->sb->s_bdi->ra_pages, ra_pages);
953 fc->minor = arg->minor;
954 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
955 fc->max_write = max_t(unsigned, 4096, fc->max_write);
958 fuse_set_initialized(fc);
959 wake_up_all(&fc->blocked_waitq);
962 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
964 struct fuse_init_in *arg = &req->misc.init_in;
966 arg->major = FUSE_KERNEL_VERSION;
967 arg->minor = FUSE_KERNEL_MINOR_VERSION;
968 arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
969 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
970 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
971 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
972 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
973 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
974 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
975 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
976 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
977 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
978 req->in.h.opcode = FUSE_INIT;
980 req->in.args[0].size = sizeof(*arg);
981 req->in.args[0].value = arg;
982 req->out.numargs = 1;
983 /* Variable length argument used for backward compatibility
984 with interface version < 7.5. Rest of init_out is zeroed
985 by do_get_request(), so a short reply is not a problem */
987 req->out.args[0].size = sizeof(struct fuse_init_out);
988 req->out.args[0].value = &req->misc.init_out;
989 req->end = process_init_reply;
990 fuse_request_send_background(fc, req);
993 static void fuse_free_conn(struct fuse_conn *fc)
995 WARN_ON(!list_empty(&fc->devices));
999 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1005 suffix = "-fuseblk";
1007 * sb->s_bdi points to blkdev's bdi however we want to redirect
1008 * it to our private bdi...
1011 sb->s_bdi = &noop_backing_dev_info;
1013 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1014 MINOR(fc->dev), suffix);
1018 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1019 /* fuse does it's own writeback accounting */
1020 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1023 * For a single fuse filesystem use max 1% of dirty +
1024 * writeback threshold.
1026 * This gives about 1M of write buffer for memory maps on a
1027 * machine with 1G and 10% dirty_ratio, which should be more
1030 * Privileged users can raise it by writing to
1032 * /sys/class/bdi/<bdi>/max_ratio
1034 bdi_set_max_ratio(sb->s_bdi, 1);
1039 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1041 struct fuse_dev *fud;
1042 struct list_head *pq;
1044 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1048 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1054 fud->pq.processing = pq;
1055 fud->fc = fuse_conn_get(fc);
1056 fuse_pqueue_init(&fud->pq);
1058 spin_lock(&fc->lock);
1059 list_add_tail(&fud->entry, &fc->devices);
1060 spin_unlock(&fc->lock);
1064 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1066 void fuse_dev_free(struct fuse_dev *fud)
1068 struct fuse_conn *fc = fud->fc;
1071 spin_lock(&fc->lock);
1072 list_del(&fud->entry);
1073 spin_unlock(&fc->lock);
1077 kfree(fud->pq.processing);
1080 EXPORT_SYMBOL_GPL(fuse_dev_free);
1082 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1084 struct fuse_dev *fud;
1085 struct fuse_conn *fc;
1087 struct fuse_mount_data d;
1089 struct dentry *root_dentry;
1090 struct fuse_req *init_req;
1092 int is_bdev = sb->s_bdev != NULL;
1095 if (sb->s_flags & SB_MANDLOCK)
1098 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1100 if (!parse_fuse_opt(data, &d, is_bdev, sb->s_user_ns))
1106 if (!sb_set_blocksize(sb, d.blksize))
1110 sb->s_blocksize = PAGE_SIZE;
1111 sb->s_blocksize_bits = PAGE_SHIFT;
1113 sb->s_magic = FUSE_SUPER_MAGIC;
1114 sb->s_op = &fuse_super_operations;
1115 sb->s_xattr = fuse_xattr_handlers;
1116 sb->s_maxbytes = MAX_LFS_FILESIZE;
1117 sb->s_time_gran = 1;
1118 sb->s_export_op = &fuse_export_operations;
1119 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1120 if (sb->s_user_ns != &init_user_ns)
1121 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1129 * Require mount to happen from the same user namespace which
1130 * opened /dev/fuse to prevent potential attacks.
1132 if (file->f_op != &fuse_dev_operations ||
1133 file->f_cred->user_ns != sb->s_user_ns)
1137 * If we are not in the initial user namespace posix
1138 * acls must be translated.
1140 if (sb->s_user_ns != &init_user_ns)
1141 sb->s_xattr = fuse_no_acl_xattr_handlers;
1143 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1148 fuse_conn_init(fc, sb->s_user_ns);
1149 fc->release = fuse_free_conn;
1151 fud = fuse_dev_alloc(fc);
1155 fc->dev = sb->s_dev;
1157 err = fuse_bdi_init(fc, sb);
1161 /* Handle umasking inside the fuse code */
1162 if (sb->s_flags & SB_POSIXACL)
1164 sb->s_flags |= SB_POSIXACL;
1166 fc->default_permissions = d.default_permissions;
1167 fc->allow_other = d.allow_other;
1168 fc->user_id = d.user_id;
1169 fc->group_id = d.group_id;
1170 fc->max_read = max_t(unsigned, 4096, d.max_read);
1172 /* Used by get_root_inode() */
1176 root = fuse_get_root_inode(sb, d.rootmode);
1177 sb->s_d_op = &fuse_root_dentry_operations;
1178 root_dentry = d_make_root(root);
1181 /* Root dentry doesn't have .d_revalidate */
1182 sb->s_d_op = &fuse_dentry_operations;
1184 init_req = fuse_request_alloc(0);
1187 __set_bit(FR_BACKGROUND, &init_req->flags);
1190 fc->destroy_req = fuse_request_alloc(0);
1191 if (!fc->destroy_req)
1192 goto err_free_init_req;
1195 mutex_lock(&fuse_mutex);
1197 if (file->private_data)
1200 err = fuse_ctl_add_conn(fc);
1204 list_add_tail(&fc->entry, &fuse_conn_list);
1205 sb->s_root = root_dentry;
1206 file->private_data = fud;
1207 mutex_unlock(&fuse_mutex);
1209 * atomic_dec_and_test() in fput() provides the necessary
1210 * memory barrier for file->private_data to be visible on all
1215 fuse_send_init(fc, init_req);
1220 mutex_unlock(&fuse_mutex);
1222 fuse_request_free(init_req);
1229 sb->s_fs_info = NULL;
1236 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1237 int flags, const char *dev_name,
1240 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1243 static void fuse_sb_destroy(struct super_block *sb)
1245 struct fuse_conn *fc = get_fuse_conn_super(sb);
1248 fuse_send_destroy(fc);
1250 fuse_abort_conn(fc);
1251 fuse_wait_aborted(fc);
1253 down_write(&fc->killsb);
1255 up_write(&fc->killsb);
1259 static void fuse_kill_sb_anon(struct super_block *sb)
1261 fuse_sb_destroy(sb);
1262 kill_anon_super(sb);
1265 static struct file_system_type fuse_fs_type = {
1266 .owner = THIS_MODULE,
1268 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1269 .mount = fuse_mount,
1270 .kill_sb = fuse_kill_sb_anon,
1272 MODULE_ALIAS_FS("fuse");
1275 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1276 int flags, const char *dev_name,
1279 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1282 static void fuse_kill_sb_blk(struct super_block *sb)
1284 fuse_sb_destroy(sb);
1285 kill_block_super(sb);
1288 static struct file_system_type fuseblk_fs_type = {
1289 .owner = THIS_MODULE,
1291 .mount = fuse_mount_blk,
1292 .kill_sb = fuse_kill_sb_blk,
1293 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1295 MODULE_ALIAS_FS("fuseblk");
1297 static inline int register_fuseblk(void)
1299 return register_filesystem(&fuseblk_fs_type);
1302 static inline void unregister_fuseblk(void)
1304 unregister_filesystem(&fuseblk_fs_type);
1307 static inline int register_fuseblk(void)
1312 static inline void unregister_fuseblk(void)
1317 static void fuse_inode_init_once(void *foo)
1319 struct inode *inode = foo;
1321 inode_init_once(inode);
1324 static int __init fuse_fs_init(void)
1328 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1329 sizeof(struct fuse_inode), 0,
1330 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1331 fuse_inode_init_once);
1333 if (!fuse_inode_cachep)
1336 err = register_fuseblk();
1340 err = register_filesystem(&fuse_fs_type);
1347 unregister_fuseblk();
1349 kmem_cache_destroy(fuse_inode_cachep);
1354 static void fuse_fs_cleanup(void)
1356 unregister_filesystem(&fuse_fs_type);
1357 unregister_fuseblk();
1360 * Make sure all delayed rcu free inodes are flushed before we
1364 kmem_cache_destroy(fuse_inode_cachep);
1367 static struct kobject *fuse_kobj;
1369 static int fuse_sysfs_init(void)
1373 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1379 err = sysfs_create_mount_point(fuse_kobj, "connections");
1381 goto out_fuse_unregister;
1385 out_fuse_unregister:
1386 kobject_put(fuse_kobj);
1391 static void fuse_sysfs_cleanup(void)
1393 sysfs_remove_mount_point(fuse_kobj, "connections");
1394 kobject_put(fuse_kobj);
1397 static int __init fuse_init(void)
1401 pr_info("init (API version %i.%i)\n",
1402 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1404 INIT_LIST_HEAD(&fuse_conn_list);
1405 res = fuse_fs_init();
1409 res = fuse_dev_init();
1411 goto err_fs_cleanup;
1413 res = fuse_sysfs_init();
1415 goto err_dev_cleanup;
1417 res = fuse_ctl_init();
1419 goto err_sysfs_cleanup;
1421 sanitize_global_limit(&max_user_bgreq);
1422 sanitize_global_limit(&max_user_congthresh);
1427 fuse_sysfs_cleanup();
1436 static void __exit fuse_exit(void)
1441 fuse_sysfs_cleanup();
1446 module_init(fuse_init);
1447 module_exit(fuse_exit);