2 FUSE: Filesystem in Userspace
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>
25 MODULE_DESCRIPTION("Filesystem in Userspace");
26 MODULE_LICENSE("GPL");
28 static struct kmem_cache *fuse_inode_cachep;
29 struct list_head fuse_conn_list;
30 DEFINE_MUTEX(fuse_mutex);
32 static int set_global_limit(const char *val, struct kernel_param *kp);
34 unsigned max_user_bgreq;
35 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
36 &max_user_bgreq, 0644);
37 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
38 MODULE_PARM_DESC(max_user_bgreq,
39 "Global limit for the maximum number of backgrounded requests an "
40 "unprivileged user can set");
42 unsigned max_user_congthresh;
43 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
44 &max_user_congthresh, 0644);
45 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
46 MODULE_PARM_DESC(max_user_congthresh,
47 "Global limit for the maximum congestion threshold an "
48 "unprivileged user can set");
50 #define FUSE_SUPER_MAGIC 0x65735546
52 #define FUSE_DEFAULT_BLKSIZE 512
54 /** Maximum number of outstanding background requests */
55 #define FUSE_DEFAULT_MAX_BACKGROUND 12
57 /** Congestion starts at 75% of maximum */
58 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
60 struct fuse_mount_data {
65 unsigned fd_present:1;
66 unsigned rootmode_present:1;
67 unsigned user_id_present:1;
68 unsigned group_id_present:1;
74 struct fuse_forget_link *fuse_alloc_forget(void)
76 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
79 static struct inode *fuse_alloc_inode(struct super_block *sb)
82 struct fuse_inode *fi;
84 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
88 fi = get_fuse_inode(inode);
96 INIT_LIST_HEAD(&fi->write_files);
97 INIT_LIST_HEAD(&fi->queued_writes);
98 INIT_LIST_HEAD(&fi->writepages);
99 init_waitqueue_head(&fi->page_waitq);
100 mutex_init(&fi->mutex);
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 BUG_ON(!list_empty(&fi->write_files));
120 BUG_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 & MS_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 & MS_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 fi->attr_version = ++fc->attr_version;
166 fi->i_time = attr_valid;
168 inode->i_ino = fuse_squash_ino(attr->ino);
169 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
170 set_nlink(inode, attr->nlink);
171 inode->i_uid = make_kuid(&init_user_ns, attr->uid);
172 inode->i_gid = make_kgid(&init_user_ns, attr->gid);
173 inode->i_blocks = attr->blocks;
174 inode->i_atime.tv_sec = attr->atime;
175 inode->i_atime.tv_nsec = attr->atimensec;
176 /* mtime from server may be stale due to local buffered write */
177 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
178 inode->i_mtime.tv_sec = attr->mtime;
179 inode->i_mtime.tv_nsec = attr->mtimensec;
180 inode->i_ctime.tv_sec = attr->ctime;
181 inode->i_ctime.tv_nsec = attr->ctimensec;
184 if (attr->blksize != 0)
185 inode->i_blkbits = ilog2(attr->blksize);
187 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
190 * Don't set the sticky bit in i_mode, unless we want the VFS
191 * to check permissions. This prevents failures due to the
192 * check in may_delete().
194 fi->orig_i_mode = inode->i_mode;
195 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
196 inode->i_mode &= ~S_ISVTX;
198 fi->orig_ino = attr->ino;
201 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
202 u64 attr_valid, u64 attr_version)
204 struct fuse_conn *fc = get_fuse_conn(inode);
205 struct fuse_inode *fi = get_fuse_inode(inode);
206 bool is_wb = fc->writeback_cache;
208 struct timespec old_mtime;
210 spin_lock(&fc->lock);
211 if ((attr_version != 0 && fi->attr_version > attr_version) ||
212 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
213 spin_unlock(&fc->lock);
217 old_mtime = inode->i_mtime;
218 fuse_change_attributes_common(inode, attr, attr_valid);
220 oldsize = inode->i_size;
222 * In case of writeback_cache enabled, the cached writes beyond EOF
223 * extend local i_size without keeping userspace server in sync. So,
224 * attr->size coming from server can be stale. We cannot trust it.
226 if (!is_wb || !S_ISREG(inode->i_mode))
227 i_size_write(inode, attr->size);
228 spin_unlock(&fc->lock);
230 if (!is_wb && S_ISREG(inode->i_mode)) {
233 if (oldsize != attr->size) {
234 truncate_pagecache(inode, attr->size);
236 } else if (fc->auto_inval_data) {
237 struct timespec new_mtime = {
238 .tv_sec = attr->mtime,
239 .tv_nsec = attr->mtimensec,
243 * Auto inval mode also checks and invalidates if mtime
246 if (!timespec_equal(&old_mtime, &new_mtime))
251 invalidate_inode_pages2(inode->i_mapping);
255 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
257 inode->i_mode = attr->mode & S_IFMT;
258 inode->i_size = attr->size;
259 inode->i_mtime.tv_sec = attr->mtime;
260 inode->i_mtime.tv_nsec = attr->mtimensec;
261 inode->i_ctime.tv_sec = attr->ctime;
262 inode->i_ctime.tv_nsec = attr->ctimensec;
263 if (S_ISREG(inode->i_mode)) {
264 fuse_init_common(inode);
265 fuse_init_file_inode(inode);
266 } else if (S_ISDIR(inode->i_mode))
267 fuse_init_dir(inode);
268 else if (S_ISLNK(inode->i_mode))
269 fuse_init_symlink(inode);
270 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
271 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
272 fuse_init_common(inode);
273 init_special_inode(inode, inode->i_mode,
274 new_decode_dev(attr->rdev));
279 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
281 u64 nodeid = *(u64 *) _nodeidp;
282 if (get_node_id(inode) == nodeid)
288 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
290 u64 nodeid = *(u64 *) _nodeidp;
291 get_fuse_inode(inode)->nodeid = nodeid;
295 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
296 int generation, struct fuse_attr *attr,
297 u64 attr_valid, u64 attr_version)
300 struct fuse_inode *fi;
301 struct fuse_conn *fc = get_fuse_conn_super(sb);
304 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
308 if ((inode->i_state & I_NEW)) {
309 inode->i_flags |= S_NOATIME;
310 if (!fc->writeback_cache || !S_ISREG(attr->mode))
311 inode->i_flags |= S_NOCMTIME;
312 inode->i_generation = generation;
313 fuse_init_inode(inode, attr);
314 unlock_new_inode(inode);
315 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
316 /* Inode has changed type, any I/O on the old should fail */
317 make_bad_inode(inode);
322 fi = get_fuse_inode(inode);
323 spin_lock(&fc->lock);
325 spin_unlock(&fc->lock);
326 fuse_change_attributes(inode, attr, attr_valid, attr_version);
331 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
332 loff_t offset, loff_t len)
338 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
342 fuse_invalidate_attr(inode);
344 pg_start = offset >> PAGE_SHIFT;
348 pg_end = (offset + len - 1) >> PAGE_SHIFT;
349 invalidate_inode_pages2_range(inode->i_mapping,
356 void fuse_lock_inode(struct inode *inode)
358 if (!get_fuse_conn(inode)->parallel_dirops)
359 mutex_lock(&get_fuse_inode(inode)->mutex);
362 void fuse_unlock_inode(struct inode *inode)
364 if (!get_fuse_conn(inode)->parallel_dirops)
365 mutex_unlock(&get_fuse_inode(inode)->mutex);
368 static void fuse_umount_begin(struct super_block *sb)
370 fuse_abort_conn(get_fuse_conn_super(sb));
373 static void fuse_send_destroy(struct fuse_conn *fc)
375 struct fuse_req *req = fc->destroy_req;
376 if (req && fc->conn_init) {
377 fc->destroy_req = NULL;
378 req->in.h.opcode = FUSE_DESTROY;
379 __set_bit(FR_FORCE, &req->flags);
380 __clear_bit(FR_BACKGROUND, &req->flags);
381 fuse_request_send(fc, req);
382 fuse_put_request(fc, req);
386 static void fuse_bdi_destroy(struct fuse_conn *fc)
388 if (fc->bdi_initialized)
389 bdi_destroy(&fc->bdi);
392 static void fuse_put_super(struct super_block *sb)
394 struct fuse_conn *fc = get_fuse_conn_super(sb);
396 fuse_send_destroy(fc);
399 mutex_lock(&fuse_mutex);
400 list_del(&fc->entry);
401 fuse_ctl_remove_conn(fc);
402 mutex_unlock(&fuse_mutex);
403 fuse_bdi_destroy(fc);
408 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
410 stbuf->f_type = FUSE_SUPER_MAGIC;
411 stbuf->f_bsize = attr->bsize;
412 stbuf->f_frsize = attr->frsize;
413 stbuf->f_blocks = attr->blocks;
414 stbuf->f_bfree = attr->bfree;
415 stbuf->f_bavail = attr->bavail;
416 stbuf->f_files = attr->files;
417 stbuf->f_ffree = attr->ffree;
418 stbuf->f_namelen = attr->namelen;
419 /* fsid is left zero */
422 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
424 struct super_block *sb = dentry->d_sb;
425 struct fuse_conn *fc = get_fuse_conn_super(sb);
427 struct fuse_statfs_out outarg;
430 if (!fuse_allow_current_process(fc)) {
431 buf->f_type = FUSE_SUPER_MAGIC;
435 memset(&outarg, 0, sizeof(outarg));
437 args.in.h.opcode = FUSE_STATFS;
438 args.in.h.nodeid = get_node_id(d_inode(dentry));
439 args.out.numargs = 1;
440 args.out.args[0].size = sizeof(outarg);
441 args.out.args[0].value = &outarg;
442 err = fuse_simple_request(fc, &args);
444 convert_fuse_statfs(buf, &outarg.st);
453 OPT_DEFAULT_PERMISSIONS,
460 static const match_table_t tokens = {
462 {OPT_ROOTMODE, "rootmode=%o"},
463 {OPT_USER_ID, "user_id=%u"},
464 {OPT_GROUP_ID, "group_id=%u"},
465 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
466 {OPT_ALLOW_OTHER, "allow_other"},
467 {OPT_MAX_READ, "max_read=%u"},
468 {OPT_BLKSIZE, "blksize=%u"},
472 static int fuse_match_uint(substring_t *s, unsigned int *res)
475 char *buf = match_strdup(s);
477 err = kstrtouint(buf, 10, res);
483 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
486 memset(d, 0, sizeof(struct fuse_mount_data));
488 d->blksize = FUSE_DEFAULT_BLKSIZE;
490 while ((p = strsep(&opt, ",")) != NULL) {
494 substring_t args[MAX_OPT_ARGS];
498 token = match_token(p, tokens, args);
501 if (match_int(&args[0], &value))
508 if (match_octal(&args[0], &value))
510 if (!fuse_valid_type(value))
513 d->rootmode_present = 1;
517 if (fuse_match_uint(&args[0], &uv))
519 d->user_id = make_kuid(current_user_ns(), uv);
520 if (!uid_valid(d->user_id))
522 d->user_id_present = 1;
526 if (fuse_match_uint(&args[0], &uv))
528 d->group_id = make_kgid(current_user_ns(), uv);
529 if (!gid_valid(d->group_id))
531 d->group_id_present = 1;
534 case OPT_DEFAULT_PERMISSIONS:
535 d->flags |= FUSE_DEFAULT_PERMISSIONS;
538 case OPT_ALLOW_OTHER:
539 d->flags |= FUSE_ALLOW_OTHER;
543 if (match_int(&args[0], &value))
549 if (!is_bdev || match_int(&args[0], &value))
559 if (!d->fd_present || !d->rootmode_present ||
560 !d->user_id_present || !d->group_id_present)
566 static int fuse_show_options(struct seq_file *m, struct dentry *root)
568 struct super_block *sb = root->d_sb;
569 struct fuse_conn *fc = get_fuse_conn_super(sb);
571 seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
572 seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
573 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
574 seq_puts(m, ",default_permissions");
575 if (fc->flags & FUSE_ALLOW_OTHER)
576 seq_puts(m, ",allow_other");
577 if (fc->max_read != ~0)
578 seq_printf(m, ",max_read=%u", fc->max_read);
579 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
580 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
584 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
586 memset(fiq, 0, sizeof(struct fuse_iqueue));
587 init_waitqueue_head(&fiq->waitq);
588 INIT_LIST_HEAD(&fiq->pending);
589 INIT_LIST_HEAD(&fiq->interrupts);
590 fiq->forget_list_tail = &fiq->forget_list_head;
594 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
596 memset(fpq, 0, sizeof(struct fuse_pqueue));
597 spin_lock_init(&fpq->lock);
598 INIT_LIST_HEAD(&fpq->processing);
599 INIT_LIST_HEAD(&fpq->io);
603 void fuse_conn_init(struct fuse_conn *fc)
605 memset(fc, 0, sizeof(*fc));
606 spin_lock_init(&fc->lock);
607 init_rwsem(&fc->killsb);
608 atomic_set(&fc->count, 1);
609 atomic_set(&fc->dev_count, 1);
610 init_waitqueue_head(&fc->blocked_waitq);
611 init_waitqueue_head(&fc->reserved_req_waitq);
612 fuse_iqueue_init(&fc->iq);
613 INIT_LIST_HEAD(&fc->bg_queue);
614 INIT_LIST_HEAD(&fc->entry);
615 INIT_LIST_HEAD(&fc->devices);
616 atomic_set(&fc->num_waiting, 0);
617 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
618 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
620 fc->polled_files = RB_ROOT;
624 fc->attr_version = 1;
625 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
627 EXPORT_SYMBOL_GPL(fuse_conn_init);
629 void fuse_conn_put(struct fuse_conn *fc)
631 if (atomic_dec_and_test(&fc->count)) {
633 fuse_request_free(fc->destroy_req);
637 EXPORT_SYMBOL_GPL(fuse_conn_put);
639 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
641 atomic_inc(&fc->count);
644 EXPORT_SYMBOL_GPL(fuse_conn_get);
646 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
648 struct fuse_attr attr;
649 memset(&attr, 0, sizeof(attr));
652 attr.ino = FUSE_ROOT_ID;
654 return fuse_iget(sb, 1, 0, &attr, 0, 0);
657 struct fuse_inode_handle {
662 static struct dentry *fuse_get_dentry(struct super_block *sb,
663 struct fuse_inode_handle *handle)
665 struct fuse_conn *fc = get_fuse_conn_super(sb);
667 struct dentry *entry;
670 if (handle->nodeid == 0)
673 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
675 struct fuse_entry_out outarg;
678 if (!fc->export_support)
683 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
685 if (err && err != -ENOENT)
692 if (get_node_id(inode) != handle->nodeid)
696 if (inode->i_generation != handle->generation)
699 entry = d_obtain_alias(inode);
700 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
701 fuse_invalidate_entry_cache(entry);
711 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
712 struct inode *parent)
714 int len = parent ? 6 : 3;
718 if (*max_len < len) {
720 return FILEID_INVALID;
723 nodeid = get_fuse_inode(inode)->nodeid;
724 generation = inode->i_generation;
726 fh[0] = (u32)(nodeid >> 32);
727 fh[1] = (u32)(nodeid & 0xffffffff);
731 nodeid = get_fuse_inode(parent)->nodeid;
732 generation = parent->i_generation;
734 fh[3] = (u32)(nodeid >> 32);
735 fh[4] = (u32)(nodeid & 0xffffffff);
740 return parent ? 0x82 : 0x81;
743 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
744 struct fid *fid, int fh_len, int fh_type)
746 struct fuse_inode_handle handle;
748 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
751 handle.nodeid = (u64) fid->raw[0] << 32;
752 handle.nodeid |= (u64) fid->raw[1];
753 handle.generation = fid->raw[2];
754 return fuse_get_dentry(sb, &handle);
757 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
758 struct fid *fid, int fh_len, int fh_type)
760 struct fuse_inode_handle parent;
762 if (fh_type != 0x82 || fh_len < 6)
765 parent.nodeid = (u64) fid->raw[3] << 32;
766 parent.nodeid |= (u64) fid->raw[4];
767 parent.generation = fid->raw[5];
768 return fuse_get_dentry(sb, &parent);
771 static struct dentry *fuse_get_parent(struct dentry *child)
773 struct inode *child_inode = d_inode(child);
774 struct fuse_conn *fc = get_fuse_conn(child_inode);
776 struct dentry *parent;
777 struct fuse_entry_out outarg;
781 if (!fc->export_support)
782 return ERR_PTR(-ESTALE);
786 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
787 &name, &outarg, &inode);
790 return ERR_PTR(-ESTALE);
794 parent = d_obtain_alias(inode);
795 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
796 fuse_invalidate_entry_cache(parent);
801 static const struct export_operations fuse_export_operations = {
802 .fh_to_dentry = fuse_fh_to_dentry,
803 .fh_to_parent = fuse_fh_to_parent,
804 .encode_fh = fuse_encode_fh,
805 .get_parent = fuse_get_parent,
808 static const struct super_operations fuse_super_operations = {
809 .alloc_inode = fuse_alloc_inode,
810 .destroy_inode = fuse_destroy_inode,
811 .evict_inode = fuse_evict_inode,
812 .write_inode = fuse_write_inode,
813 .drop_inode = generic_delete_inode,
814 .remount_fs = fuse_remount_fs,
815 .put_super = fuse_put_super,
816 .umount_begin = fuse_umount_begin,
817 .statfs = fuse_statfs,
818 .show_options = fuse_show_options,
821 static void sanitize_global_limit(unsigned *limit)
824 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
825 sizeof(struct fuse_req);
827 if (*limit >= 1 << 16)
828 *limit = (1 << 16) - 1;
831 static int set_global_limit(const char *val, struct kernel_param *kp)
835 rv = param_set_uint(val, kp);
839 sanitize_global_limit((unsigned *)kp->arg);
844 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
846 int cap_sys_admin = capable(CAP_SYS_ADMIN);
851 sanitize_global_limit(&max_user_bgreq);
852 sanitize_global_limit(&max_user_congthresh);
854 if (arg->max_background) {
855 fc->max_background = arg->max_background;
857 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
858 fc->max_background = max_user_bgreq;
860 if (arg->congestion_threshold) {
861 fc->congestion_threshold = arg->congestion_threshold;
863 if (!cap_sys_admin &&
864 fc->congestion_threshold > max_user_congthresh)
865 fc->congestion_threshold = max_user_congthresh;
869 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
871 struct fuse_init_out *arg = &req->misc.init_out;
873 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
876 unsigned long ra_pages;
878 process_init_limits(fc, arg);
880 if (arg->minor >= 6) {
881 ra_pages = arg->max_readahead / PAGE_SIZE;
882 if (arg->flags & FUSE_ASYNC_READ)
884 if (!(arg->flags & FUSE_POSIX_LOCKS))
886 if (arg->minor >= 17) {
887 if (!(arg->flags & FUSE_FLOCK_LOCKS))
890 if (!(arg->flags & FUSE_POSIX_LOCKS))
893 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
894 fc->atomic_o_trunc = 1;
895 if (arg->minor >= 9) {
896 /* LOOKUP has dependency on proto version */
897 if (arg->flags & FUSE_EXPORT_SUPPORT)
898 fc->export_support = 1;
900 if (arg->flags & FUSE_BIG_WRITES)
902 if (arg->flags & FUSE_DONT_MASK)
904 if (arg->flags & FUSE_AUTO_INVAL_DATA)
905 fc->auto_inval_data = 1;
906 if (arg->flags & FUSE_DO_READDIRPLUS) {
907 fc->do_readdirplus = 1;
908 if (arg->flags & FUSE_READDIRPLUS_AUTO)
909 fc->readdirplus_auto = 1;
911 if (arg->flags & FUSE_ASYNC_DIO)
913 if (arg->flags & FUSE_WRITEBACK_CACHE)
914 fc->writeback_cache = 1;
915 if (arg->flags & FUSE_PARALLEL_DIROPS)
916 fc->parallel_dirops = 1;
917 if (arg->time_gran && arg->time_gran <= 1000000000)
918 fc->sb->s_time_gran = arg->time_gran;
920 ra_pages = fc->max_read / PAGE_SIZE;
925 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
926 fc->minor = arg->minor;
927 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
928 fc->max_write = max_t(unsigned, 4096, fc->max_write);
931 fuse_set_initialized(fc);
932 wake_up_all(&fc->blocked_waitq);
935 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
937 struct fuse_init_in *arg = &req->misc.init_in;
939 arg->major = FUSE_KERNEL_VERSION;
940 arg->minor = FUSE_KERNEL_MINOR_VERSION;
941 arg->max_readahead = fc->bdi.ra_pages * PAGE_SIZE;
942 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
943 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
944 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
945 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
946 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
947 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
948 FUSE_PARALLEL_DIROPS;
949 req->in.h.opcode = FUSE_INIT;
951 req->in.args[0].size = sizeof(*arg);
952 req->in.args[0].value = arg;
953 req->out.numargs = 1;
954 /* Variable length argument used for backward compatibility
955 with interface version < 7.5. Rest of init_out is zeroed
956 by do_get_request(), so a short reply is not a problem */
958 req->out.args[0].size = sizeof(struct fuse_init_out);
959 req->out.args[0].value = &req->misc.init_out;
960 req->end = process_init_reply;
961 fuse_request_send_background(fc, req);
964 static void fuse_free_conn(struct fuse_conn *fc)
966 WARN_ON(!list_empty(&fc->devices));
970 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
974 fc->bdi.name = "fuse";
975 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
976 /* fuse does it's own writeback accounting */
977 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
979 err = bdi_init(&fc->bdi);
983 fc->bdi_initialized = 1;
986 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
987 MAJOR(fc->dev), MINOR(fc->dev));
989 err = bdi_register_dev(&fc->bdi, fc->dev);
996 * For a single fuse filesystem use max 1% of dirty +
997 * writeback threshold.
999 * This gives about 1M of write buffer for memory maps on a
1000 * machine with 1G and 10% dirty_ratio, which should be more
1003 * Privileged users can raise it by writing to
1005 * /sys/class/bdi/<bdi>/max_ratio
1007 bdi_set_max_ratio(&fc->bdi, 1);
1012 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1014 struct fuse_dev *fud;
1016 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1018 fud->fc = fuse_conn_get(fc);
1019 fuse_pqueue_init(&fud->pq);
1021 spin_lock(&fc->lock);
1022 list_add_tail(&fud->entry, &fc->devices);
1023 spin_unlock(&fc->lock);
1028 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1030 void fuse_dev_free(struct fuse_dev *fud)
1032 struct fuse_conn *fc = fud->fc;
1035 spin_lock(&fc->lock);
1036 list_del(&fud->entry);
1037 spin_unlock(&fc->lock);
1043 EXPORT_SYMBOL_GPL(fuse_dev_free);
1045 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1047 struct fuse_dev *fud;
1048 struct fuse_conn *fc;
1050 struct fuse_mount_data d;
1052 struct dentry *root_dentry;
1053 struct fuse_req *init_req;
1055 int is_bdev = sb->s_bdev != NULL;
1058 if (sb->s_flags & MS_MANDLOCK)
1061 sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
1063 if (!parse_fuse_opt(data, &d, is_bdev))
1069 if (!sb_set_blocksize(sb, d.blksize))
1073 sb->s_blocksize = PAGE_SIZE;
1074 sb->s_blocksize_bits = PAGE_SHIFT;
1076 sb->s_magic = FUSE_SUPER_MAGIC;
1077 sb->s_op = &fuse_super_operations;
1078 sb->s_maxbytes = MAX_LFS_FILESIZE;
1079 sb->s_time_gran = 1;
1080 sb->s_export_op = &fuse_export_operations;
1087 if ((file->f_op != &fuse_dev_operations) ||
1088 (file->f_cred->user_ns != &init_user_ns))
1091 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1097 fc->release = fuse_free_conn;
1099 fud = fuse_dev_alloc(fc);
1103 fc->dev = sb->s_dev;
1105 err = fuse_bdi_init(fc, sb);
1109 sb->s_bdi = &fc->bdi;
1111 /* Handle umasking inside the fuse code */
1112 if (sb->s_flags & MS_POSIXACL)
1114 sb->s_flags |= MS_POSIXACL;
1116 fc->flags = d.flags;
1117 fc->user_id = d.user_id;
1118 fc->group_id = d.group_id;
1119 fc->max_read = max_t(unsigned, 4096, d.max_read);
1121 /* Used by get_root_inode() */
1125 root = fuse_get_root_inode(sb, d.rootmode);
1126 root_dentry = d_make_root(root);
1129 /* only now - we want root dentry with NULL ->d_op */
1130 sb->s_d_op = &fuse_dentry_operations;
1132 init_req = fuse_request_alloc(0);
1135 __set_bit(FR_BACKGROUND, &init_req->flags);
1138 fc->destroy_req = fuse_request_alloc(0);
1139 if (!fc->destroy_req)
1140 goto err_free_init_req;
1143 mutex_lock(&fuse_mutex);
1145 if (file->private_data)
1148 err = fuse_ctl_add_conn(fc);
1152 list_add_tail(&fc->entry, &fuse_conn_list);
1153 sb->s_root = root_dentry;
1154 file->private_data = fud;
1155 mutex_unlock(&fuse_mutex);
1157 * atomic_dec_and_test() in fput() provides the necessary
1158 * memory barrier for file->private_data to be visible on all
1163 fuse_send_init(fc, init_req);
1168 mutex_unlock(&fuse_mutex);
1170 fuse_request_free(init_req);
1176 fuse_bdi_destroy(fc);
1184 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1185 int flags, const char *dev_name,
1188 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1191 static void fuse_kill_sb_anon(struct super_block *sb)
1193 struct fuse_conn *fc = get_fuse_conn_super(sb);
1196 down_write(&fc->killsb);
1198 up_write(&fc->killsb);
1201 kill_anon_super(sb);
1204 static struct file_system_type fuse_fs_type = {
1205 .owner = THIS_MODULE,
1207 .fs_flags = FS_HAS_SUBTYPE,
1208 .mount = fuse_mount,
1209 .kill_sb = fuse_kill_sb_anon,
1211 MODULE_ALIAS_FS("fuse");
1214 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1215 int flags, const char *dev_name,
1218 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1221 static void fuse_kill_sb_blk(struct super_block *sb)
1223 struct fuse_conn *fc = get_fuse_conn_super(sb);
1226 down_write(&fc->killsb);
1228 up_write(&fc->killsb);
1231 kill_block_super(sb);
1234 static struct file_system_type fuseblk_fs_type = {
1235 .owner = THIS_MODULE,
1237 .mount = fuse_mount_blk,
1238 .kill_sb = fuse_kill_sb_blk,
1239 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1241 MODULE_ALIAS_FS("fuseblk");
1243 static inline int register_fuseblk(void)
1245 return register_filesystem(&fuseblk_fs_type);
1248 static inline void unregister_fuseblk(void)
1250 unregister_filesystem(&fuseblk_fs_type);
1253 static inline int register_fuseblk(void)
1258 static inline void unregister_fuseblk(void)
1263 static void fuse_inode_init_once(void *foo)
1265 struct inode *inode = foo;
1267 inode_init_once(inode);
1270 static int __init fuse_fs_init(void)
1274 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1275 sizeof(struct fuse_inode), 0,
1276 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT,
1277 fuse_inode_init_once);
1279 if (!fuse_inode_cachep)
1282 err = register_fuseblk();
1286 err = register_filesystem(&fuse_fs_type);
1293 unregister_fuseblk();
1295 kmem_cache_destroy(fuse_inode_cachep);
1300 static void fuse_fs_cleanup(void)
1302 unregister_filesystem(&fuse_fs_type);
1303 unregister_fuseblk();
1306 * Make sure all delayed rcu free inodes are flushed before we
1310 kmem_cache_destroy(fuse_inode_cachep);
1313 static struct kobject *fuse_kobj;
1315 static int fuse_sysfs_init(void)
1319 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1325 err = sysfs_create_mount_point(fuse_kobj, "connections");
1327 goto out_fuse_unregister;
1331 out_fuse_unregister:
1332 kobject_put(fuse_kobj);
1337 static void fuse_sysfs_cleanup(void)
1339 sysfs_remove_mount_point(fuse_kobj, "connections");
1340 kobject_put(fuse_kobj);
1343 static int __init fuse_init(void)
1347 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1348 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1350 INIT_LIST_HEAD(&fuse_conn_list);
1351 res = fuse_fs_init();
1355 res = fuse_dev_init();
1357 goto err_fs_cleanup;
1359 res = fuse_sysfs_init();
1361 goto err_dev_cleanup;
1363 res = fuse_ctl_init();
1365 goto err_sysfs_cleanup;
1367 sanitize_global_limit(&max_user_bgreq);
1368 sanitize_global_limit(&max_user_congthresh);
1373 fuse_sysfs_cleanup();
1382 static void __exit fuse_exit(void)
1384 printk(KERN_DEBUG "fuse exit\n");
1387 fuse_sysfs_cleanup();
1392 module_init(fuse_init);
1393 module_exit(fuse_exit);