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/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>
27 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
28 MODULE_DESCRIPTION("Filesystem in Userspace");
29 MODULE_LICENSE("GPL");
31 static struct kmem_cache *fuse_inode_cachep;
32 struct list_head fuse_conn_list;
33 DEFINE_MUTEX(fuse_mutex);
35 static int set_global_limit(const char *val, const struct kernel_param *kp);
37 unsigned max_user_bgreq;
38 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
39 &max_user_bgreq, 0644);
40 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
41 MODULE_PARM_DESC(max_user_bgreq,
42 "Global limit for the maximum number of backgrounded requests an "
43 "unprivileged user can set");
45 unsigned max_user_congthresh;
46 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
47 &max_user_congthresh, 0644);
48 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
49 MODULE_PARM_DESC(max_user_congthresh,
50 "Global limit for the maximum congestion threshold an "
51 "unprivileged user can set");
53 #define FUSE_SUPER_MAGIC 0x65735546
55 #define FUSE_DEFAULT_BLKSIZE 512
57 /** Maximum number of outstanding background requests */
58 #define FUSE_DEFAULT_MAX_BACKGROUND 12
60 /** Congestion starts at 75% of maximum */
61 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
64 static struct file_system_type fuseblk_fs_type;
67 struct fuse_fs_context {
74 unsigned fd_present:1;
75 unsigned rootmode_present:1;
76 unsigned user_id_present:1;
77 unsigned group_id_present:1;
78 unsigned default_permissions:1;
79 unsigned allow_other:1;
84 struct fuse_forget_link *fuse_alloc_forget(void)
86 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
89 static struct inode *fuse_alloc_inode(struct super_block *sb)
91 struct fuse_inode *fi;
93 fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
101 fi->attr_version = 0;
104 mutex_init(&fi->mutex);
105 spin_lock_init(&fi->lock);
106 fi->forget = fuse_alloc_forget();
108 kmem_cache_free(fuse_inode_cachep, fi);
115 static void fuse_free_inode(struct inode *inode)
117 struct fuse_inode *fi = get_fuse_inode(inode);
119 mutex_destroy(&fi->mutex);
121 kmem_cache_free(fuse_inode_cachep, fi);
124 static void fuse_evict_inode(struct inode *inode)
126 struct fuse_inode *fi = get_fuse_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 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
135 if (S_ISREG(inode->i_mode) && !is_bad_inode(inode)) {
136 WARN_ON(!list_empty(&fi->write_files));
137 WARN_ON(!list_empty(&fi->queued_writes));
141 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
144 if (*flags & SB_MANDLOCK)
151 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
152 * so that it will fit.
154 static ino_t fuse_squash_ino(u64 ino64)
156 ino_t ino = (ino_t) ino64;
157 if (sizeof(ino_t) < sizeof(u64))
158 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
162 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
165 struct fuse_conn *fc = get_fuse_conn(inode);
166 struct fuse_inode *fi = get_fuse_inode(inode);
168 lockdep_assert_held(&fi->lock);
170 fi->attr_version = atomic64_inc_return(&fc->attr_version);
171 fi->i_time = attr_valid;
172 WRITE_ONCE(fi->inval_mask, 0);
174 inode->i_ino = fuse_squash_ino(attr->ino);
175 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
176 set_nlink(inode, attr->nlink);
177 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
178 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
179 inode->i_blocks = attr->blocks;
180 inode->i_atime.tv_sec = attr->atime;
181 inode->i_atime.tv_nsec = attr->atimensec;
182 /* mtime from server may be stale due to local buffered write */
183 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
184 inode->i_mtime.tv_sec = attr->mtime;
185 inode->i_mtime.tv_nsec = attr->mtimensec;
186 inode->i_ctime.tv_sec = attr->ctime;
187 inode->i_ctime.tv_nsec = attr->ctimensec;
190 if (attr->blksize != 0)
191 inode->i_blkbits = ilog2(attr->blksize);
193 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
196 * Don't set the sticky bit in i_mode, unless we want the VFS
197 * to check permissions. This prevents failures due to the
198 * check in may_delete().
200 fi->orig_i_mode = inode->i_mode;
201 if (!fc->default_permissions)
202 inode->i_mode &= ~S_ISVTX;
204 fi->orig_ino = attr->ino;
207 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
208 u64 attr_valid, u64 attr_version)
210 struct fuse_conn *fc = get_fuse_conn(inode);
211 struct fuse_inode *fi = get_fuse_inode(inode);
212 bool is_wb = fc->writeback_cache;
214 struct timespec64 old_mtime;
216 spin_lock(&fi->lock);
217 if ((attr_version != 0 && fi->attr_version > attr_version) ||
218 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
219 spin_unlock(&fi->lock);
223 old_mtime = inode->i_mtime;
224 fuse_change_attributes_common(inode, attr, attr_valid);
226 oldsize = inode->i_size;
228 * In case of writeback_cache enabled, the cached writes beyond EOF
229 * extend local i_size without keeping userspace server in sync. So,
230 * attr->size coming from server can be stale. We cannot trust it.
232 if (!is_wb || !S_ISREG(inode->i_mode))
233 i_size_write(inode, attr->size);
234 spin_unlock(&fi->lock);
236 if (!is_wb && S_ISREG(inode->i_mode)) {
239 if (oldsize != attr->size) {
240 truncate_pagecache(inode, attr->size);
241 if (!fc->explicit_inval_data)
243 } else if (fc->auto_inval_data) {
244 struct timespec64 new_mtime = {
245 .tv_sec = attr->mtime,
246 .tv_nsec = attr->mtimensec,
250 * Auto inval mode also checks and invalidates if mtime
253 if (!timespec64_equal(&old_mtime, &new_mtime))
258 invalidate_inode_pages2(inode->i_mapping);
262 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
264 inode->i_mode = attr->mode & S_IFMT;
265 inode->i_size = attr->size;
266 inode->i_mtime.tv_sec = attr->mtime;
267 inode->i_mtime.tv_nsec = attr->mtimensec;
268 inode->i_ctime.tv_sec = attr->ctime;
269 inode->i_ctime.tv_nsec = attr->ctimensec;
270 if (S_ISREG(inode->i_mode)) {
271 fuse_init_common(inode);
272 fuse_init_file_inode(inode);
273 } else if (S_ISDIR(inode->i_mode))
274 fuse_init_dir(inode);
275 else if (S_ISLNK(inode->i_mode))
276 fuse_init_symlink(inode);
277 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
278 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
279 fuse_init_common(inode);
280 init_special_inode(inode, inode->i_mode,
281 new_decode_dev(attr->rdev));
286 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
288 u64 nodeid = *(u64 *) _nodeidp;
289 if (get_node_id(inode) == nodeid)
295 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
297 u64 nodeid = *(u64 *) _nodeidp;
298 get_fuse_inode(inode)->nodeid = nodeid;
302 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
303 int generation, struct fuse_attr *attr,
304 u64 attr_valid, u64 attr_version)
307 struct fuse_inode *fi;
308 struct fuse_conn *fc = get_fuse_conn_super(sb);
311 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
315 if ((inode->i_state & I_NEW)) {
316 inode->i_flags |= S_NOATIME;
317 if (!fc->writeback_cache || !S_ISREG(attr->mode))
318 inode->i_flags |= S_NOCMTIME;
319 inode->i_generation = generation;
320 fuse_init_inode(inode, attr);
321 unlock_new_inode(inode);
322 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
323 /* Inode has changed type, any I/O on the old should fail */
324 make_bad_inode(inode);
329 fi = get_fuse_inode(inode);
330 spin_lock(&fi->lock);
332 spin_unlock(&fi->lock);
333 fuse_change_attributes(inode, attr, attr_valid, attr_version);
338 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
339 loff_t offset, loff_t len)
345 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
349 fuse_invalidate_attr(inode);
350 forget_all_cached_acls(inode);
352 pg_start = offset >> PAGE_SHIFT;
356 pg_end = (offset + len - 1) >> PAGE_SHIFT;
357 invalidate_inode_pages2_range(inode->i_mapping,
364 bool fuse_lock_inode(struct inode *inode)
368 if (!get_fuse_conn(inode)->parallel_dirops) {
369 mutex_lock(&get_fuse_inode(inode)->mutex);
376 void fuse_unlock_inode(struct inode *inode, bool locked)
379 mutex_unlock(&get_fuse_inode(inode)->mutex);
382 static void fuse_umount_begin(struct super_block *sb)
384 fuse_abort_conn(get_fuse_conn_super(sb));
387 static void fuse_send_destroy(struct fuse_conn *fc)
392 args.opcode = FUSE_DESTROY;
395 fuse_simple_request(fc, &args);
399 static void fuse_put_super(struct super_block *sb)
401 struct fuse_conn *fc = get_fuse_conn_super(sb);
403 mutex_lock(&fuse_mutex);
404 list_del(&fc->entry);
405 fuse_ctl_remove_conn(fc);
406 mutex_unlock(&fuse_mutex);
411 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
413 stbuf->f_type = FUSE_SUPER_MAGIC;
414 stbuf->f_bsize = attr->bsize;
415 stbuf->f_frsize = attr->frsize;
416 stbuf->f_blocks = attr->blocks;
417 stbuf->f_bfree = attr->bfree;
418 stbuf->f_bavail = attr->bavail;
419 stbuf->f_files = attr->files;
420 stbuf->f_ffree = attr->ffree;
421 stbuf->f_namelen = attr->namelen;
422 /* fsid is left zero */
425 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
427 struct super_block *sb = dentry->d_sb;
428 struct fuse_conn *fc = get_fuse_conn_super(sb);
430 struct fuse_statfs_out outarg;
433 if (!fuse_allow_current_process(fc)) {
434 buf->f_type = FUSE_SUPER_MAGIC;
438 memset(&outarg, 0, sizeof(outarg));
440 args.opcode = FUSE_STATFS;
441 args.nodeid = get_node_id(d_inode(dentry));
442 args.out_numargs = 1;
443 args.out_args[0].size = sizeof(outarg);
444 args.out_args[0].value = &outarg;
445 err = fuse_simple_request(fc, &args);
447 convert_fuse_statfs(buf, &outarg.st);
458 OPT_DEFAULT_PERMISSIONS,
465 static const struct fs_parameter_spec fuse_param_specs[] = {
466 fsparam_string ("source", OPT_SOURCE),
467 fsparam_u32 ("fd", OPT_FD),
468 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
469 fsparam_u32 ("user_id", OPT_USER_ID),
470 fsparam_u32 ("group_id", OPT_GROUP_ID),
471 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
472 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
473 fsparam_u32 ("max_read", OPT_MAX_READ),
474 fsparam_u32 ("blksize", OPT_BLKSIZE),
475 fsparam_string ("subtype", OPT_SUBTYPE),
479 static const struct fs_parameter_description fuse_fs_parameters = {
481 .specs = fuse_param_specs,
484 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
486 struct fs_parse_result result;
487 struct fuse_fs_context *ctx = fc->fs_private;
490 opt = fs_parse(fc, &fuse_fs_parameters, param, &result);
497 return invalf(fc, "fuse: Multiple sources specified");
498 fc->source = param->string;
499 param->string = NULL;
504 return invalf(fc, "fuse: Multiple subtypes specified");
505 ctx->subtype = param->string;
506 param->string = NULL;
510 ctx->fd = result.uint_32;
515 if (!fuse_valid_type(result.uint_32))
516 return invalf(fc, "fuse: Invalid rootmode");
517 ctx->rootmode = result.uint_32;
518 ctx->rootmode_present = 1;
522 ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
523 if (!uid_valid(ctx->user_id))
524 return invalf(fc, "fuse: Invalid user_id");
525 ctx->user_id_present = 1;
529 ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
530 if (!gid_valid(ctx->group_id))
531 return invalf(fc, "fuse: Invalid group_id");
532 ctx->group_id_present = 1;
535 case OPT_DEFAULT_PERMISSIONS:
536 ctx->default_permissions = 1;
539 case OPT_ALLOW_OTHER:
540 ctx->allow_other = 1;
544 ctx->max_read = result.uint_32;
549 return invalf(fc, "fuse: blksize only supported for fuseblk");
550 ctx->blksize = result.uint_32;
560 static void fuse_free_fc(struct fs_context *fc)
562 struct fuse_fs_context *ctx = fc->fs_private;
570 static int fuse_show_options(struct seq_file *m, struct dentry *root)
572 struct super_block *sb = root->d_sb;
573 struct fuse_conn *fc = get_fuse_conn_super(sb);
575 seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
576 seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
577 if (fc->default_permissions)
578 seq_puts(m, ",default_permissions");
580 seq_puts(m, ",allow_other");
581 if (fc->max_read != ~0)
582 seq_printf(m, ",max_read=%u", fc->max_read);
583 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
584 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
588 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
590 memset(fiq, 0, sizeof(struct fuse_iqueue));
591 spin_lock_init(&fiq->lock);
592 init_waitqueue_head(&fiq->waitq);
593 INIT_LIST_HEAD(&fiq->pending);
594 INIT_LIST_HEAD(&fiq->interrupts);
595 fiq->forget_list_tail = &fiq->forget_list_head;
599 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
603 spin_lock_init(&fpq->lock);
604 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
605 INIT_LIST_HEAD(&fpq->processing[i]);
606 INIT_LIST_HEAD(&fpq->io);
610 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns)
612 memset(fc, 0, sizeof(*fc));
613 spin_lock_init(&fc->lock);
614 spin_lock_init(&fc->bg_lock);
615 init_rwsem(&fc->killsb);
616 refcount_set(&fc->count, 1);
617 atomic_set(&fc->dev_count, 1);
618 init_waitqueue_head(&fc->blocked_waitq);
619 fuse_iqueue_init(&fc->iq);
620 INIT_LIST_HEAD(&fc->bg_queue);
621 INIT_LIST_HEAD(&fc->entry);
622 INIT_LIST_HEAD(&fc->devices);
623 atomic_set(&fc->num_waiting, 0);
624 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
625 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
626 atomic64_set(&fc->khctr, 0);
627 fc->polled_files = RB_ROOT;
631 atomic64_set(&fc->attr_version, 1);
632 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
633 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
634 fc->user_ns = get_user_ns(user_ns);
635 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
637 EXPORT_SYMBOL_GPL(fuse_conn_init);
639 void fuse_conn_put(struct fuse_conn *fc)
641 if (refcount_dec_and_test(&fc->count)) {
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 .free_inode = fuse_free_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 * The default maximum number of async requests is calculated to consume
831 * 1/2^13 of the total memory, assuming 392 bytes per request.
834 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
836 if (*limit >= 1 << 16)
837 *limit = (1 << 16) - 1;
840 static int set_global_limit(const char *val, const struct kernel_param *kp)
844 rv = param_set_uint(val, kp);
848 sanitize_global_limit((unsigned *)kp->arg);
853 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
855 int cap_sys_admin = capable(CAP_SYS_ADMIN);
860 sanitize_global_limit(&max_user_bgreq);
861 sanitize_global_limit(&max_user_congthresh);
863 spin_lock(&fc->bg_lock);
864 if (arg->max_background) {
865 fc->max_background = arg->max_background;
867 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
868 fc->max_background = max_user_bgreq;
870 if (arg->congestion_threshold) {
871 fc->congestion_threshold = arg->congestion_threshold;
873 if (!cap_sys_admin &&
874 fc->congestion_threshold > max_user_congthresh)
875 fc->congestion_threshold = max_user_congthresh;
877 spin_unlock(&fc->bg_lock);
880 struct fuse_init_args {
881 struct fuse_args args;
882 struct fuse_init_in in;
883 struct fuse_init_out out;
886 static void process_init_reply(struct fuse_conn *fc, struct fuse_args *args,
889 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
890 struct fuse_init_out *arg = &ia->out;
892 if (error || arg->major != FUSE_KERNEL_VERSION)
895 unsigned long ra_pages;
897 process_init_limits(fc, arg);
899 if (arg->minor >= 6) {
900 ra_pages = arg->max_readahead / PAGE_SIZE;
901 if (arg->flags & FUSE_ASYNC_READ)
903 if (!(arg->flags & FUSE_POSIX_LOCKS))
905 if (arg->minor >= 17) {
906 if (!(arg->flags & FUSE_FLOCK_LOCKS))
909 if (!(arg->flags & FUSE_POSIX_LOCKS))
912 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
913 fc->atomic_o_trunc = 1;
914 if (arg->minor >= 9) {
915 /* LOOKUP has dependency on proto version */
916 if (arg->flags & FUSE_EXPORT_SUPPORT)
917 fc->export_support = 1;
919 if (arg->flags & FUSE_BIG_WRITES)
921 if (arg->flags & FUSE_DONT_MASK)
923 if (arg->flags & FUSE_AUTO_INVAL_DATA)
924 fc->auto_inval_data = 1;
925 else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
926 fc->explicit_inval_data = 1;
927 if (arg->flags & FUSE_DO_READDIRPLUS) {
928 fc->do_readdirplus = 1;
929 if (arg->flags & FUSE_READDIRPLUS_AUTO)
930 fc->readdirplus_auto = 1;
932 if (arg->flags & FUSE_ASYNC_DIO)
934 if (arg->flags & FUSE_WRITEBACK_CACHE)
935 fc->writeback_cache = 1;
936 if (arg->flags & FUSE_PARALLEL_DIROPS)
937 fc->parallel_dirops = 1;
938 if (arg->flags & FUSE_HANDLE_KILLPRIV)
939 fc->handle_killpriv = 1;
940 if (arg->time_gran && arg->time_gran <= 1000000000)
941 fc->sb->s_time_gran = arg->time_gran;
942 if ((arg->flags & FUSE_POSIX_ACL)) {
943 fc->default_permissions = 1;
945 fc->sb->s_xattr = fuse_acl_xattr_handlers;
947 if (arg->flags & FUSE_CACHE_SYMLINKS)
948 fc->cache_symlinks = 1;
949 if (arg->flags & FUSE_ABORT_ERROR)
951 if (arg->flags & FUSE_MAX_PAGES) {
953 min_t(unsigned int, FUSE_MAX_MAX_PAGES,
954 max_t(unsigned int, arg->max_pages, 1));
957 ra_pages = fc->max_read / PAGE_SIZE;
962 fc->sb->s_bdi->ra_pages =
963 min(fc->sb->s_bdi->ra_pages, ra_pages);
964 fc->minor = arg->minor;
965 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
966 fc->max_write = max_t(unsigned, 4096, fc->max_write);
971 fuse_set_initialized(fc);
972 wake_up_all(&fc->blocked_waitq);
975 void fuse_send_init(struct fuse_conn *fc)
977 struct fuse_init_args *ia;
979 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
981 ia->in.major = FUSE_KERNEL_VERSION;
982 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
983 ia->in.max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
985 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
986 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
987 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
988 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
989 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
990 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
991 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
992 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
993 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
994 ia->args.opcode = FUSE_INIT;
995 ia->args.in_numargs = 1;
996 ia->args.in_args[0].size = sizeof(ia->in);
997 ia->args.in_args[0].value = &ia->in;
998 ia->args.out_numargs = 1;
999 /* Variable length argument used for backward compatibility
1000 with interface version < 7.5. Rest of init_out is zeroed
1001 by do_get_request(), so a short reply is not a problem */
1002 ia->args.out_argvar = 1;
1003 ia->args.out_args[0].size = sizeof(ia->out);
1004 ia->args.out_args[0].value = &ia->out;
1005 ia->args.force = true;
1006 ia->args.nocreds = true;
1007 ia->args.end = process_init_reply;
1009 if (fuse_simple_background(fc, &ia->args, GFP_KERNEL) != 0)
1010 process_init_reply(fc, &ia->args, -ENOTCONN);
1012 EXPORT_SYMBOL_GPL(fuse_send_init);
1014 static void fuse_free_conn(struct fuse_conn *fc)
1016 WARN_ON(!list_empty(&fc->devices));
1020 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1026 suffix = "-fuseblk";
1028 * sb->s_bdi points to blkdev's bdi however we want to redirect
1029 * it to our private bdi...
1032 sb->s_bdi = &noop_backing_dev_info;
1034 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1035 MINOR(fc->dev), suffix);
1039 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1040 /* fuse does it's own writeback accounting */
1041 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1044 * For a single fuse filesystem use max 1% of dirty +
1045 * writeback threshold.
1047 * This gives about 1M of write buffer for memory maps on a
1048 * machine with 1G and 10% dirty_ratio, which should be more
1051 * Privileged users can raise it by writing to
1053 * /sys/class/bdi/<bdi>/max_ratio
1055 bdi_set_max_ratio(sb->s_bdi, 1);
1060 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1062 struct fuse_dev *fud;
1063 struct list_head *pq;
1065 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1069 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1075 fud->pq.processing = pq;
1076 fud->fc = fuse_conn_get(fc);
1077 fuse_pqueue_init(&fud->pq);
1079 spin_lock(&fc->lock);
1080 list_add_tail(&fud->entry, &fc->devices);
1081 spin_unlock(&fc->lock);
1085 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1087 void fuse_dev_free(struct fuse_dev *fud)
1089 struct fuse_conn *fc = fud->fc;
1092 spin_lock(&fc->lock);
1093 list_del(&fud->entry);
1094 spin_unlock(&fc->lock);
1098 kfree(fud->pq.processing);
1101 EXPORT_SYMBOL_GPL(fuse_dev_free);
1103 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1105 struct fuse_fs_context *ctx = fsc->fs_private;
1106 struct fuse_dev *fud;
1107 struct fuse_conn *fc;
1110 struct dentry *root_dentry;
1112 int is_bdev = sb->s_bdev != NULL;
1115 if (sb->s_flags & SB_MANDLOCK)
1118 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1123 if (!sb_set_blocksize(sb, ctx->blksize))
1127 sb->s_blocksize = PAGE_SIZE;
1128 sb->s_blocksize_bits = PAGE_SHIFT;
1131 sb->s_subtype = ctx->subtype;
1132 ctx->subtype = NULL;
1133 sb->s_magic = FUSE_SUPER_MAGIC;
1134 sb->s_op = &fuse_super_operations;
1135 sb->s_xattr = fuse_xattr_handlers;
1136 sb->s_maxbytes = MAX_LFS_FILESIZE;
1137 sb->s_time_gran = 1;
1138 sb->s_export_op = &fuse_export_operations;
1139 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1140 if (sb->s_user_ns != &init_user_ns)
1141 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1143 file = fget(ctx->fd);
1149 * Require mount to happen from the same user namespace which
1150 * opened /dev/fuse to prevent potential attacks.
1152 if (file->f_op != &fuse_dev_operations ||
1153 file->f_cred->user_ns != sb->s_user_ns)
1157 * If we are not in the initial user namespace posix
1158 * acls must be translated.
1160 if (sb->s_user_ns != &init_user_ns)
1161 sb->s_xattr = fuse_no_acl_xattr_handlers;
1163 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1168 fuse_conn_init(fc, sb->s_user_ns);
1169 fc->release = fuse_free_conn;
1171 fud = fuse_dev_alloc(fc);
1175 fc->dev = sb->s_dev;
1177 err = fuse_bdi_init(fc, sb);
1181 /* Handle umasking inside the fuse code */
1182 if (sb->s_flags & SB_POSIXACL)
1184 sb->s_flags |= SB_POSIXACL;
1186 fc->default_permissions = ctx->default_permissions;
1187 fc->allow_other = ctx->allow_other;
1188 fc->user_id = ctx->user_id;
1189 fc->group_id = ctx->group_id;
1190 fc->max_read = max_t(unsigned, 4096, ctx->max_read);
1191 fc->destroy = is_bdev;
1193 /* Used by get_root_inode() */
1197 root = fuse_get_root_inode(sb, ctx->rootmode);
1198 sb->s_d_op = &fuse_root_dentry_operations;
1199 root_dentry = d_make_root(root);
1202 /* Root dentry doesn't have .d_revalidate */
1203 sb->s_d_op = &fuse_dentry_operations;
1205 mutex_lock(&fuse_mutex);
1207 if (file->private_data)
1210 err = fuse_ctl_add_conn(fc);
1214 list_add_tail(&fc->entry, &fuse_conn_list);
1215 sb->s_root = root_dentry;
1216 file->private_data = fud;
1217 mutex_unlock(&fuse_mutex);
1219 * atomic_dec_and_test() in fput() provides the necessary
1220 * memory barrier for file->private_data to be visible on all
1230 mutex_unlock(&fuse_mutex);
1236 sb->s_fs_info = NULL;
1243 static int fuse_get_tree(struct fs_context *fc)
1245 struct fuse_fs_context *ctx = fc->fs_private;
1247 if (!ctx->fd_present || !ctx->rootmode_present ||
1248 !ctx->user_id_present || !ctx->group_id_present)
1253 return get_tree_bdev(fc, fuse_fill_super);
1256 return get_tree_nodev(fc, fuse_fill_super);
1259 static const struct fs_context_operations fuse_context_ops = {
1260 .free = fuse_free_fc,
1261 .parse_param = fuse_parse_param,
1262 .get_tree = fuse_get_tree,
1266 * Set up the filesystem mount context.
1268 static int fuse_init_fs_context(struct fs_context *fc)
1270 struct fuse_fs_context *ctx;
1272 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1277 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1280 if (fc->fs_type == &fuseblk_fs_type)
1281 ctx->is_bdev = true;
1284 fc->fs_private = ctx;
1285 fc->ops = &fuse_context_ops;
1289 static void fuse_sb_destroy(struct super_block *sb)
1291 struct fuse_conn *fc = get_fuse_conn_super(sb);
1295 fuse_send_destroy(fc);
1297 fuse_abort_conn(fc);
1298 fuse_wait_aborted(fc);
1300 down_write(&fc->killsb);
1302 up_write(&fc->killsb);
1306 static void fuse_kill_sb_anon(struct super_block *sb)
1308 fuse_sb_destroy(sb);
1309 kill_anon_super(sb);
1312 static struct file_system_type fuse_fs_type = {
1313 .owner = THIS_MODULE,
1315 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1316 .init_fs_context = fuse_init_fs_context,
1317 .parameters = &fuse_fs_parameters,
1318 .kill_sb = fuse_kill_sb_anon,
1320 MODULE_ALIAS_FS("fuse");
1323 static void fuse_kill_sb_blk(struct super_block *sb)
1325 fuse_sb_destroy(sb);
1326 kill_block_super(sb);
1329 static struct file_system_type fuseblk_fs_type = {
1330 .owner = THIS_MODULE,
1332 .init_fs_context = fuse_init_fs_context,
1333 .parameters = &fuse_fs_parameters,
1334 .kill_sb = fuse_kill_sb_blk,
1335 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1337 MODULE_ALIAS_FS("fuseblk");
1339 static inline int register_fuseblk(void)
1341 return register_filesystem(&fuseblk_fs_type);
1344 static inline void unregister_fuseblk(void)
1346 unregister_filesystem(&fuseblk_fs_type);
1349 static inline int register_fuseblk(void)
1354 static inline void unregister_fuseblk(void)
1359 static void fuse_inode_init_once(void *foo)
1361 struct inode *inode = foo;
1363 inode_init_once(inode);
1366 static int __init fuse_fs_init(void)
1370 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1371 sizeof(struct fuse_inode), 0,
1372 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1373 fuse_inode_init_once);
1375 if (!fuse_inode_cachep)
1378 err = register_fuseblk();
1382 err = register_filesystem(&fuse_fs_type);
1389 unregister_fuseblk();
1391 kmem_cache_destroy(fuse_inode_cachep);
1396 static void fuse_fs_cleanup(void)
1398 unregister_filesystem(&fuse_fs_type);
1399 unregister_fuseblk();
1402 * Make sure all delayed rcu free inodes are flushed before we
1406 kmem_cache_destroy(fuse_inode_cachep);
1409 static struct kobject *fuse_kobj;
1411 static int fuse_sysfs_init(void)
1415 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1421 err = sysfs_create_mount_point(fuse_kobj, "connections");
1423 goto out_fuse_unregister;
1427 out_fuse_unregister:
1428 kobject_put(fuse_kobj);
1433 static void fuse_sysfs_cleanup(void)
1435 sysfs_remove_mount_point(fuse_kobj, "connections");
1436 kobject_put(fuse_kobj);
1439 static int __init fuse_init(void)
1443 pr_info("init (API version %i.%i)\n",
1444 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1446 INIT_LIST_HEAD(&fuse_conn_list);
1447 res = fuse_fs_init();
1451 res = fuse_dev_init();
1453 goto err_fs_cleanup;
1455 res = fuse_sysfs_init();
1457 goto err_dev_cleanup;
1459 res = fuse_ctl_init();
1461 goto err_sysfs_cleanup;
1463 sanitize_global_limit(&max_user_bgreq);
1464 sanitize_global_limit(&max_user_congthresh);
1469 fuse_sysfs_cleanup();
1478 static void __exit fuse_exit(void)
1483 fuse_sysfs_cleanup();
1488 module_init(fuse_init);
1489 module_exit(fuse_exit);