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_forget_link *fuse_alloc_forget(void)
69 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
72 static struct inode *fuse_alloc_inode(struct super_block *sb)
74 struct fuse_inode *fi;
76 fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
87 mutex_init(&fi->mutex);
88 spin_lock_init(&fi->lock);
89 fi->forget = fuse_alloc_forget();
91 kmem_cache_free(fuse_inode_cachep, fi);
98 static void fuse_free_inode(struct inode *inode)
100 struct fuse_inode *fi = get_fuse_inode(inode);
102 mutex_destroy(&fi->mutex);
104 kmem_cache_free(fuse_inode_cachep, fi);
107 static void fuse_evict_inode(struct inode *inode)
109 struct fuse_inode *fi = get_fuse_inode(inode);
111 truncate_inode_pages_final(&inode->i_data);
113 if (inode->i_sb->s_flags & SB_ACTIVE) {
114 struct fuse_conn *fc = get_fuse_conn(inode);
115 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
118 if (S_ISREG(inode->i_mode) && !is_bad_inode(inode)) {
119 WARN_ON(!list_empty(&fi->write_files));
120 WARN_ON(!list_empty(&fi->queued_writes));
124 static int fuse_reconfigure(struct fs_context *fc)
126 struct super_block *sb = fc->root->d_sb;
129 if (fc->sb_flags & SB_MANDLOCK)
136 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
137 * so that it will fit.
139 static ino_t fuse_squash_ino(u64 ino64)
141 ino_t ino = (ino_t) ino64;
142 if (sizeof(ino_t) < sizeof(u64))
143 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
147 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
150 struct fuse_conn *fc = get_fuse_conn(inode);
151 struct fuse_inode *fi = get_fuse_inode(inode);
153 lockdep_assert_held(&fi->lock);
155 fi->attr_version = atomic64_inc_return(&fc->attr_version);
156 fi->i_time = attr_valid;
157 WRITE_ONCE(fi->inval_mask, 0);
159 inode->i_ino = fuse_squash_ino(attr->ino);
160 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
161 set_nlink(inode, attr->nlink);
162 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
163 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
164 inode->i_blocks = attr->blocks;
165 inode->i_atime.tv_sec = attr->atime;
166 inode->i_atime.tv_nsec = attr->atimensec;
167 /* mtime from server may be stale due to local buffered write */
168 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
169 inode->i_mtime.tv_sec = attr->mtime;
170 inode->i_mtime.tv_nsec = attr->mtimensec;
171 inode->i_ctime.tv_sec = attr->ctime;
172 inode->i_ctime.tv_nsec = attr->ctimensec;
175 if (attr->blksize != 0)
176 inode->i_blkbits = ilog2(attr->blksize);
178 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
181 * Don't set the sticky bit in i_mode, unless we want the VFS
182 * to check permissions. This prevents failures due to the
183 * check in may_delete().
185 fi->orig_i_mode = inode->i_mode;
186 if (!fc->default_permissions)
187 inode->i_mode &= ~S_ISVTX;
189 fi->orig_ino = attr->ino;
192 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
193 u64 attr_valid, u64 attr_version)
195 struct fuse_conn *fc = get_fuse_conn(inode);
196 struct fuse_inode *fi = get_fuse_inode(inode);
197 bool is_wb = fc->writeback_cache;
199 struct timespec64 old_mtime;
201 spin_lock(&fi->lock);
202 if ((attr_version != 0 && fi->attr_version > attr_version) ||
203 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
204 spin_unlock(&fi->lock);
208 old_mtime = inode->i_mtime;
209 fuse_change_attributes_common(inode, attr, attr_valid);
211 oldsize = inode->i_size;
213 * In case of writeback_cache enabled, the cached writes beyond EOF
214 * extend local i_size without keeping userspace server in sync. So,
215 * attr->size coming from server can be stale. We cannot trust it.
217 if (!is_wb || !S_ISREG(inode->i_mode))
218 i_size_write(inode, attr->size);
219 spin_unlock(&fi->lock);
221 if (!is_wb && S_ISREG(inode->i_mode)) {
224 if (oldsize != attr->size) {
225 truncate_pagecache(inode, attr->size);
226 if (!fc->explicit_inval_data)
228 } else if (fc->auto_inval_data) {
229 struct timespec64 new_mtime = {
230 .tv_sec = attr->mtime,
231 .tv_nsec = attr->mtimensec,
235 * Auto inval mode also checks and invalidates if mtime
238 if (!timespec64_equal(&old_mtime, &new_mtime))
243 invalidate_inode_pages2(inode->i_mapping);
247 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
249 inode->i_mode = attr->mode & S_IFMT;
250 inode->i_size = attr->size;
251 inode->i_mtime.tv_sec = attr->mtime;
252 inode->i_mtime.tv_nsec = attr->mtimensec;
253 inode->i_ctime.tv_sec = attr->ctime;
254 inode->i_ctime.tv_nsec = attr->ctimensec;
255 if (S_ISREG(inode->i_mode)) {
256 fuse_init_common(inode);
257 fuse_init_file_inode(inode);
258 } else if (S_ISDIR(inode->i_mode))
259 fuse_init_dir(inode);
260 else if (S_ISLNK(inode->i_mode))
261 fuse_init_symlink(inode);
262 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
263 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
264 fuse_init_common(inode);
265 init_special_inode(inode, inode->i_mode,
266 new_decode_dev(attr->rdev));
271 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
273 u64 nodeid = *(u64 *) _nodeidp;
274 if (get_node_id(inode) == nodeid)
280 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
282 u64 nodeid = *(u64 *) _nodeidp;
283 get_fuse_inode(inode)->nodeid = nodeid;
287 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
288 int generation, struct fuse_attr *attr,
289 u64 attr_valid, u64 attr_version)
292 struct fuse_inode *fi;
293 struct fuse_conn *fc = get_fuse_conn_super(sb);
296 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
300 if ((inode->i_state & I_NEW)) {
301 inode->i_flags |= S_NOATIME;
302 if (!fc->writeback_cache || !S_ISREG(attr->mode))
303 inode->i_flags |= S_NOCMTIME;
304 inode->i_generation = generation;
305 fuse_init_inode(inode, attr);
306 unlock_new_inode(inode);
307 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
308 /* Inode has changed type, any I/O on the old should fail */
309 make_bad_inode(inode);
314 fi = get_fuse_inode(inode);
315 spin_lock(&fi->lock);
317 spin_unlock(&fi->lock);
318 fuse_change_attributes(inode, attr, attr_valid, attr_version);
323 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
324 loff_t offset, loff_t len)
326 struct fuse_conn *fc = get_fuse_conn_super(sb);
327 struct fuse_inode *fi;
332 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
336 fi = get_fuse_inode(inode);
337 spin_lock(&fi->lock);
338 fi->attr_version = atomic64_inc_return(&fc->attr_version);
339 spin_unlock(&fi->lock);
341 fuse_invalidate_attr(inode);
342 forget_all_cached_acls(inode);
344 pg_start = offset >> PAGE_SHIFT;
348 pg_end = (offset + len - 1) >> PAGE_SHIFT;
349 invalidate_inode_pages2_range(inode->i_mapping,
356 bool fuse_lock_inode(struct inode *inode)
360 if (!get_fuse_conn(inode)->parallel_dirops) {
361 mutex_lock(&get_fuse_inode(inode)->mutex);
368 void fuse_unlock_inode(struct inode *inode, bool locked)
371 mutex_unlock(&get_fuse_inode(inode)->mutex);
374 static void fuse_umount_begin(struct super_block *sb)
376 struct fuse_conn *fc = get_fuse_conn_super(sb);
378 if (!fc->no_force_umount)
382 static void fuse_send_destroy(struct fuse_conn *fc)
387 args.opcode = FUSE_DESTROY;
390 fuse_simple_request(fc, &args);
394 static void fuse_put_super(struct super_block *sb)
396 struct fuse_conn *fc = get_fuse_conn_super(sb);
398 mutex_lock(&fuse_mutex);
399 list_del(&fc->entry);
400 fuse_ctl_remove_conn(fc);
401 mutex_unlock(&fuse_mutex);
406 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
408 stbuf->f_type = FUSE_SUPER_MAGIC;
409 stbuf->f_bsize = attr->bsize;
410 stbuf->f_frsize = attr->frsize;
411 stbuf->f_blocks = attr->blocks;
412 stbuf->f_bfree = attr->bfree;
413 stbuf->f_bavail = attr->bavail;
414 stbuf->f_files = attr->files;
415 stbuf->f_ffree = attr->ffree;
416 stbuf->f_namelen = attr->namelen;
417 /* fsid is left zero */
420 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
422 struct super_block *sb = dentry->d_sb;
423 struct fuse_conn *fc = get_fuse_conn_super(sb);
425 struct fuse_statfs_out outarg;
428 if (!fuse_allow_current_process(fc)) {
429 buf->f_type = FUSE_SUPER_MAGIC;
433 memset(&outarg, 0, sizeof(outarg));
435 args.opcode = FUSE_STATFS;
436 args.nodeid = get_node_id(d_inode(dentry));
437 args.out_numargs = 1;
438 args.out_args[0].size = sizeof(outarg);
439 args.out_args[0].value = &outarg;
440 err = fuse_simple_request(fc, &args);
442 convert_fuse_statfs(buf, &outarg.st);
453 OPT_DEFAULT_PERMISSIONS,
460 static const struct fs_parameter_spec fuse_fs_parameters[] = {
461 fsparam_string ("source", OPT_SOURCE),
462 fsparam_u32 ("fd", OPT_FD),
463 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
464 fsparam_u32 ("user_id", OPT_USER_ID),
465 fsparam_u32 ("group_id", OPT_GROUP_ID),
466 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
467 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
468 fsparam_u32 ("max_read", OPT_MAX_READ),
469 fsparam_u32 ("blksize", OPT_BLKSIZE),
470 fsparam_string ("subtype", OPT_SUBTYPE),
474 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
476 struct fs_parse_result result;
477 struct fuse_fs_context *ctx = fc->fs_private;
480 opt = fs_parse(fc, fuse_fs_parameters, param, &result);
487 return invalfc(fc, "Multiple sources specified");
488 fc->source = param->string;
489 param->string = NULL;
494 return invalfc(fc, "Multiple subtypes specified");
495 ctx->subtype = param->string;
496 param->string = NULL;
500 ctx->fd = result.uint_32;
501 ctx->fd_present = true;
505 if (!fuse_valid_type(result.uint_32))
506 return invalfc(fc, "Invalid rootmode");
507 ctx->rootmode = result.uint_32;
508 ctx->rootmode_present = true;
512 ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
513 if (!uid_valid(ctx->user_id))
514 return invalfc(fc, "Invalid user_id");
515 ctx->user_id_present = true;
519 ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
520 if (!gid_valid(ctx->group_id))
521 return invalfc(fc, "Invalid group_id");
522 ctx->group_id_present = true;
525 case OPT_DEFAULT_PERMISSIONS:
526 ctx->default_permissions = true;
529 case OPT_ALLOW_OTHER:
530 ctx->allow_other = true;
534 ctx->max_read = result.uint_32;
539 return invalfc(fc, "blksize only supported for fuseblk");
540 ctx->blksize = result.uint_32;
550 static void fuse_free_fc(struct fs_context *fc)
552 struct fuse_fs_context *ctx = fc->fs_private;
560 static int fuse_show_options(struct seq_file *m, struct dentry *root)
562 struct super_block *sb = root->d_sb;
563 struct fuse_conn *fc = get_fuse_conn_super(sb);
565 if (fc->no_mount_options)
568 seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
569 seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
570 if (fc->default_permissions)
571 seq_puts(m, ",default_permissions");
573 seq_puts(m, ",allow_other");
574 if (fc->max_read != ~0)
575 seq_printf(m, ",max_read=%u", fc->max_read);
576 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
577 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
581 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
582 const struct fuse_iqueue_ops *ops,
585 memset(fiq, 0, sizeof(struct fuse_iqueue));
586 spin_lock_init(&fiq->lock);
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;
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,
608 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
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 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
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)) {
640 struct fuse_iqueue *fiq = &fc->iq;
642 if (fiq->ops->release)
643 fiq->ops->release(fiq);
644 put_pid_ns(fc->pid_ns);
645 put_user_ns(fc->user_ns);
649 EXPORT_SYMBOL_GPL(fuse_conn_put);
651 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
653 refcount_inc(&fc->count);
656 EXPORT_SYMBOL_GPL(fuse_conn_get);
658 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
660 struct fuse_attr attr;
661 memset(&attr, 0, sizeof(attr));
664 attr.ino = FUSE_ROOT_ID;
666 return fuse_iget(sb, 1, 0, &attr, 0, 0);
669 struct fuse_inode_handle {
674 static struct dentry *fuse_get_dentry(struct super_block *sb,
675 struct fuse_inode_handle *handle)
677 struct fuse_conn *fc = get_fuse_conn_super(sb);
679 struct dentry *entry;
682 if (handle->nodeid == 0)
685 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
687 struct fuse_entry_out outarg;
688 const struct qstr name = QSTR_INIT(".", 1);
690 if (!fc->export_support)
693 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
695 if (err && err != -ENOENT)
702 if (get_node_id(inode) != handle->nodeid)
706 if (inode->i_generation != handle->generation)
709 entry = d_obtain_alias(inode);
710 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
711 fuse_invalidate_entry_cache(entry);
721 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
722 struct inode *parent)
724 int len = parent ? 6 : 3;
728 if (*max_len < len) {
730 return FILEID_INVALID;
733 nodeid = get_fuse_inode(inode)->nodeid;
734 generation = inode->i_generation;
736 fh[0] = (u32)(nodeid >> 32);
737 fh[1] = (u32)(nodeid & 0xffffffff);
741 nodeid = get_fuse_inode(parent)->nodeid;
742 generation = parent->i_generation;
744 fh[3] = (u32)(nodeid >> 32);
745 fh[4] = (u32)(nodeid & 0xffffffff);
750 return parent ? 0x82 : 0x81;
753 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
754 struct fid *fid, int fh_len, int fh_type)
756 struct fuse_inode_handle handle;
758 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
761 handle.nodeid = (u64) fid->raw[0] << 32;
762 handle.nodeid |= (u64) fid->raw[1];
763 handle.generation = fid->raw[2];
764 return fuse_get_dentry(sb, &handle);
767 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
768 struct fid *fid, int fh_len, int fh_type)
770 struct fuse_inode_handle parent;
772 if (fh_type != 0x82 || fh_len < 6)
775 parent.nodeid = (u64) fid->raw[3] << 32;
776 parent.nodeid |= (u64) fid->raw[4];
777 parent.generation = fid->raw[5];
778 return fuse_get_dentry(sb, &parent);
781 static struct dentry *fuse_get_parent(struct dentry *child)
783 struct inode *child_inode = d_inode(child);
784 struct fuse_conn *fc = get_fuse_conn(child_inode);
786 struct dentry *parent;
787 struct fuse_entry_out outarg;
788 const struct qstr name = QSTR_INIT("..", 2);
791 if (!fc->export_support)
792 return ERR_PTR(-ESTALE);
794 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
795 &name, &outarg, &inode);
798 return ERR_PTR(-ESTALE);
802 parent = d_obtain_alias(inode);
803 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
804 fuse_invalidate_entry_cache(parent);
809 static const struct export_operations fuse_export_operations = {
810 .fh_to_dentry = fuse_fh_to_dentry,
811 .fh_to_parent = fuse_fh_to_parent,
812 .encode_fh = fuse_encode_fh,
813 .get_parent = fuse_get_parent,
816 static const struct super_operations fuse_super_operations = {
817 .alloc_inode = fuse_alloc_inode,
818 .free_inode = fuse_free_inode,
819 .evict_inode = fuse_evict_inode,
820 .write_inode = fuse_write_inode,
821 .drop_inode = generic_delete_inode,
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 * The default maximum number of async requests is calculated to consume
832 * 1/2^13 of the total memory, assuming 392 bytes per request.
835 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
837 if (*limit >= 1 << 16)
838 *limit = (1 << 16) - 1;
841 static int set_global_limit(const char *val, const struct kernel_param *kp)
845 rv = param_set_uint(val, kp);
849 sanitize_global_limit((unsigned *)kp->arg);
854 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
856 int cap_sys_admin = capable(CAP_SYS_ADMIN);
861 sanitize_global_limit(&max_user_bgreq);
862 sanitize_global_limit(&max_user_congthresh);
864 spin_lock(&fc->bg_lock);
865 if (arg->max_background) {
866 fc->max_background = arg->max_background;
868 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
869 fc->max_background = max_user_bgreq;
871 if (arg->congestion_threshold) {
872 fc->congestion_threshold = arg->congestion_threshold;
874 if (!cap_sys_admin &&
875 fc->congestion_threshold > max_user_congthresh)
876 fc->congestion_threshold = max_user_congthresh;
878 spin_unlock(&fc->bg_lock);
881 struct fuse_init_args {
882 struct fuse_args args;
883 struct fuse_init_in in;
884 struct fuse_init_out out;
887 static void process_init_reply(struct fuse_conn *fc, struct fuse_args *args,
890 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
891 struct fuse_init_out *arg = &ia->out;
893 if (error || arg->major != FUSE_KERNEL_VERSION)
896 unsigned long ra_pages;
898 process_init_limits(fc, arg);
900 if (arg->minor >= 6) {
901 ra_pages = arg->max_readahead / PAGE_SIZE;
902 if (arg->flags & FUSE_ASYNC_READ)
904 if (!(arg->flags & FUSE_POSIX_LOCKS))
906 if (arg->minor >= 17) {
907 if (!(arg->flags & FUSE_FLOCK_LOCKS))
910 if (!(arg->flags & FUSE_POSIX_LOCKS))
913 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
914 fc->atomic_o_trunc = 1;
915 if (arg->minor >= 9) {
916 /* LOOKUP has dependency on proto version */
917 if (arg->flags & FUSE_EXPORT_SUPPORT)
918 fc->export_support = 1;
920 if (arg->flags & FUSE_BIG_WRITES)
922 if (arg->flags & FUSE_DONT_MASK)
924 if (arg->flags & FUSE_AUTO_INVAL_DATA)
925 fc->auto_inval_data = 1;
926 else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
927 fc->explicit_inval_data = 1;
928 if (arg->flags & FUSE_DO_READDIRPLUS) {
929 fc->do_readdirplus = 1;
930 if (arg->flags & FUSE_READDIRPLUS_AUTO)
931 fc->readdirplus_auto = 1;
933 if (arg->flags & FUSE_ASYNC_DIO)
935 if (arg->flags & FUSE_WRITEBACK_CACHE)
936 fc->writeback_cache = 1;
937 if (arg->flags & FUSE_PARALLEL_DIROPS)
938 fc->parallel_dirops = 1;
939 if (arg->flags & FUSE_HANDLE_KILLPRIV)
940 fc->handle_killpriv = 1;
941 if (arg->time_gran && arg->time_gran <= 1000000000)
942 fc->sb->s_time_gran = arg->time_gran;
943 if ((arg->flags & FUSE_POSIX_ACL)) {
944 fc->default_permissions = 1;
946 fc->sb->s_xattr = fuse_acl_xattr_handlers;
948 if (arg->flags & FUSE_CACHE_SYMLINKS)
949 fc->cache_symlinks = 1;
950 if (arg->flags & FUSE_ABORT_ERROR)
952 if (arg->flags & FUSE_MAX_PAGES) {
954 min_t(unsigned int, FUSE_MAX_MAX_PAGES,
955 max_t(unsigned int, arg->max_pages, 1));
958 ra_pages = fc->max_read / PAGE_SIZE;
963 fc->sb->s_bdi->ra_pages =
964 min(fc->sb->s_bdi->ra_pages, ra_pages);
965 fc->minor = arg->minor;
966 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
967 fc->max_write = max_t(unsigned, 4096, fc->max_write);
972 fuse_set_initialized(fc);
973 wake_up_all(&fc->blocked_waitq);
976 void fuse_send_init(struct fuse_conn *fc)
978 struct fuse_init_args *ia;
980 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
982 ia->in.major = FUSE_KERNEL_VERSION;
983 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
984 ia->in.max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
986 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
987 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
988 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
989 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
990 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
991 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
992 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
993 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
994 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
995 ia->args.opcode = FUSE_INIT;
996 ia->args.in_numargs = 1;
997 ia->args.in_args[0].size = sizeof(ia->in);
998 ia->args.in_args[0].value = &ia->in;
999 ia->args.out_numargs = 1;
1000 /* Variable length argument used for backward compatibility
1001 with interface version < 7.5. Rest of init_out is zeroed
1002 by do_get_request(), so a short reply is not a problem */
1003 ia->args.out_argvar = true;
1004 ia->args.out_args[0].size = sizeof(ia->out);
1005 ia->args.out_args[0].value = &ia->out;
1006 ia->args.force = true;
1007 ia->args.nocreds = true;
1008 ia->args.end = process_init_reply;
1010 if (fuse_simple_background(fc, &ia->args, GFP_KERNEL) != 0)
1011 process_init_reply(fc, &ia->args, -ENOTCONN);
1013 EXPORT_SYMBOL_GPL(fuse_send_init);
1015 void fuse_free_conn(struct fuse_conn *fc)
1017 WARN_ON(!list_empty(&fc->devices));
1020 EXPORT_SYMBOL_GPL(fuse_free_conn);
1022 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1028 suffix = "-fuseblk";
1030 * sb->s_bdi points to blkdev's bdi however we want to redirect
1031 * it to our private bdi...
1034 sb->s_bdi = &noop_backing_dev_info;
1036 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1037 MINOR(fc->dev), suffix);
1041 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1042 /* fuse does it's own writeback accounting */
1043 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1046 * For a single fuse filesystem use max 1% of dirty +
1047 * writeback threshold.
1049 * This gives about 1M of write buffer for memory maps on a
1050 * machine with 1G and 10% dirty_ratio, which should be more
1053 * Privileged users can raise it by writing to
1055 * /sys/class/bdi/<bdi>/max_ratio
1057 bdi_set_max_ratio(sb->s_bdi, 1);
1062 struct fuse_dev *fuse_dev_alloc(void)
1064 struct fuse_dev *fud;
1065 struct list_head *pq;
1067 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1071 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1077 fud->pq.processing = pq;
1078 fuse_pqueue_init(&fud->pq);
1082 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1084 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1086 fud->fc = fuse_conn_get(fc);
1087 spin_lock(&fc->lock);
1088 list_add_tail(&fud->entry, &fc->devices);
1089 spin_unlock(&fc->lock);
1091 EXPORT_SYMBOL_GPL(fuse_dev_install);
1093 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1095 struct fuse_dev *fud;
1097 fud = fuse_dev_alloc();
1101 fuse_dev_install(fud, fc);
1104 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1106 void fuse_dev_free(struct fuse_dev *fud)
1108 struct fuse_conn *fc = fud->fc;
1111 spin_lock(&fc->lock);
1112 list_del(&fud->entry);
1113 spin_unlock(&fc->lock);
1117 kfree(fud->pq.processing);
1120 EXPORT_SYMBOL_GPL(fuse_dev_free);
1122 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1124 struct fuse_dev *fud = NULL;
1125 struct fuse_conn *fc = get_fuse_conn_super(sb);
1127 struct dentry *root_dentry;
1131 if (sb->s_flags & SB_MANDLOCK)
1134 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1139 if (!sb_set_blocksize(sb, ctx->blksize))
1143 sb->s_blocksize = PAGE_SIZE;
1144 sb->s_blocksize_bits = PAGE_SHIFT;
1147 sb->s_subtype = ctx->subtype;
1148 ctx->subtype = NULL;
1149 sb->s_magic = FUSE_SUPER_MAGIC;
1150 sb->s_op = &fuse_super_operations;
1151 sb->s_xattr = fuse_xattr_handlers;
1152 sb->s_maxbytes = MAX_LFS_FILESIZE;
1153 sb->s_time_gran = 1;
1154 sb->s_export_op = &fuse_export_operations;
1155 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1156 if (sb->s_user_ns != &init_user_ns)
1157 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1160 * If we are not in the initial user namespace posix
1161 * acls must be translated.
1163 if (sb->s_user_ns != &init_user_ns)
1164 sb->s_xattr = fuse_no_acl_xattr_handlers;
1168 fud = fuse_dev_alloc_install(fc);
1173 fc->dev = sb->s_dev;
1175 err = fuse_bdi_init(fc, sb);
1179 /* Handle umasking inside the fuse code */
1180 if (sb->s_flags & SB_POSIXACL)
1182 sb->s_flags |= SB_POSIXACL;
1184 fc->default_permissions = ctx->default_permissions;
1185 fc->allow_other = ctx->allow_other;
1186 fc->user_id = ctx->user_id;
1187 fc->group_id = ctx->group_id;
1188 fc->max_read = max_t(unsigned, 4096, ctx->max_read);
1189 fc->destroy = ctx->destroy;
1190 fc->no_control = ctx->no_control;
1191 fc->no_force_umount = ctx->no_force_umount;
1192 fc->no_mount_options = ctx->no_mount_options;
1195 root = fuse_get_root_inode(sb, ctx->rootmode);
1196 sb->s_d_op = &fuse_root_dentry_operations;
1197 root_dentry = d_make_root(root);
1200 /* Root dentry doesn't have .d_revalidate */
1201 sb->s_d_op = &fuse_dentry_operations;
1203 mutex_lock(&fuse_mutex);
1205 if (ctx->fudptr && *ctx->fudptr)
1208 err = fuse_ctl_add_conn(fc);
1212 list_add_tail(&fc->entry, &fuse_conn_list);
1213 sb->s_root = root_dentry;
1216 mutex_unlock(&fuse_mutex);
1220 mutex_unlock(&fuse_mutex);
1228 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1230 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1232 struct fuse_fs_context *ctx = fsc->fs_private;
1235 struct fuse_conn *fc;
1238 file = fget(ctx->fd);
1243 * Require mount to happen from the same user namespace which
1244 * opened /dev/fuse to prevent potential attacks.
1246 if ((file->f_op != &fuse_dev_operations) ||
1247 (file->f_cred->user_ns != sb->s_user_ns))
1249 ctx->fudptr = &file->private_data;
1251 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1256 fuse_conn_init(fc, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1257 fc->release = fuse_free_conn;
1260 err = fuse_fill_super_common(sb, ctx);
1264 * atomic_dec_and_test() in fput() provides the necessary
1265 * memory barrier for file->private_data to be visible on all
1269 fuse_send_init(get_fuse_conn_super(sb));
1274 sb->s_fs_info = NULL;
1281 static int fuse_get_tree(struct fs_context *fc)
1283 struct fuse_fs_context *ctx = fc->fs_private;
1285 if (!ctx->fd_present || !ctx->rootmode_present ||
1286 !ctx->user_id_present || !ctx->group_id_present)
1291 return get_tree_bdev(fc, fuse_fill_super);
1294 return get_tree_nodev(fc, fuse_fill_super);
1297 static const struct fs_context_operations fuse_context_ops = {
1298 .free = fuse_free_fc,
1299 .parse_param = fuse_parse_param,
1300 .reconfigure = fuse_reconfigure,
1301 .get_tree = fuse_get_tree,
1305 * Set up the filesystem mount context.
1307 static int fuse_init_fs_context(struct fs_context *fc)
1309 struct fuse_fs_context *ctx;
1311 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1316 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1319 if (fc->fs_type == &fuseblk_fs_type) {
1320 ctx->is_bdev = true;
1321 ctx->destroy = true;
1325 fc->fs_private = ctx;
1326 fc->ops = &fuse_context_ops;
1330 static void fuse_sb_destroy(struct super_block *sb)
1332 struct fuse_conn *fc = get_fuse_conn_super(sb);
1336 fuse_send_destroy(fc);
1338 fuse_abort_conn(fc);
1339 fuse_wait_aborted(fc);
1341 down_write(&fc->killsb);
1343 up_write(&fc->killsb);
1347 void fuse_kill_sb_anon(struct super_block *sb)
1349 fuse_sb_destroy(sb);
1350 kill_anon_super(sb);
1352 EXPORT_SYMBOL_GPL(fuse_kill_sb_anon);
1354 static struct file_system_type fuse_fs_type = {
1355 .owner = THIS_MODULE,
1357 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1358 .init_fs_context = fuse_init_fs_context,
1359 .parameters = fuse_fs_parameters,
1360 .kill_sb = fuse_kill_sb_anon,
1362 MODULE_ALIAS_FS("fuse");
1365 static void fuse_kill_sb_blk(struct super_block *sb)
1367 fuse_sb_destroy(sb);
1368 kill_block_super(sb);
1371 static struct file_system_type fuseblk_fs_type = {
1372 .owner = THIS_MODULE,
1374 .init_fs_context = fuse_init_fs_context,
1375 .parameters = fuse_fs_parameters,
1376 .kill_sb = fuse_kill_sb_blk,
1377 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1379 MODULE_ALIAS_FS("fuseblk");
1381 static inline int register_fuseblk(void)
1383 return register_filesystem(&fuseblk_fs_type);
1386 static inline void unregister_fuseblk(void)
1388 unregister_filesystem(&fuseblk_fs_type);
1391 static inline int register_fuseblk(void)
1396 static inline void unregister_fuseblk(void)
1401 static void fuse_inode_init_once(void *foo)
1403 struct inode *inode = foo;
1405 inode_init_once(inode);
1408 static int __init fuse_fs_init(void)
1412 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1413 sizeof(struct fuse_inode), 0,
1414 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1415 fuse_inode_init_once);
1417 if (!fuse_inode_cachep)
1420 err = register_fuseblk();
1424 err = register_filesystem(&fuse_fs_type);
1431 unregister_fuseblk();
1433 kmem_cache_destroy(fuse_inode_cachep);
1438 static void fuse_fs_cleanup(void)
1440 unregister_filesystem(&fuse_fs_type);
1441 unregister_fuseblk();
1444 * Make sure all delayed rcu free inodes are flushed before we
1448 kmem_cache_destroy(fuse_inode_cachep);
1451 static struct kobject *fuse_kobj;
1453 static int fuse_sysfs_init(void)
1457 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1463 err = sysfs_create_mount_point(fuse_kobj, "connections");
1465 goto out_fuse_unregister;
1469 out_fuse_unregister:
1470 kobject_put(fuse_kobj);
1475 static void fuse_sysfs_cleanup(void)
1477 sysfs_remove_mount_point(fuse_kobj, "connections");
1478 kobject_put(fuse_kobj);
1481 static int __init fuse_init(void)
1485 pr_info("init (API version %i.%i)\n",
1486 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1488 INIT_LIST_HEAD(&fuse_conn_list);
1489 res = fuse_fs_init();
1493 res = fuse_dev_init();
1495 goto err_fs_cleanup;
1497 res = fuse_sysfs_init();
1499 goto err_dev_cleanup;
1501 res = fuse_ctl_init();
1503 goto err_sysfs_cleanup;
1505 sanitize_global_limit(&max_user_bgreq);
1506 sanitize_global_limit(&max_user_congthresh);
1511 fuse_sysfs_cleanup();
1520 static void __exit fuse_exit(void)
1525 fuse_sysfs_cleanup();
1530 module_init(fuse_init);
1531 module_exit(fuse_exit);