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/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>
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_remount_fs(struct super_block *sb, int *flags, char *data)
127 if (*flags & SB_MANDLOCK)
134 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
135 * so that it will fit.
137 static ino_t fuse_squash_ino(u64 ino64)
139 ino_t ino = (ino_t) ino64;
140 if (sizeof(ino_t) < sizeof(u64))
141 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
145 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
148 struct fuse_conn *fc = get_fuse_conn(inode);
149 struct fuse_inode *fi = get_fuse_inode(inode);
151 lockdep_assert_held(&fi->lock);
153 fi->attr_version = atomic64_inc_return(&fc->attr_version);
154 fi->i_time = attr_valid;
155 WRITE_ONCE(fi->inval_mask, 0);
157 inode->i_ino = fuse_squash_ino(attr->ino);
158 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
159 set_nlink(inode, attr->nlink);
160 inode->i_uid = make_kuid(fc->user_ns, attr->uid);
161 inode->i_gid = make_kgid(fc->user_ns, attr->gid);
162 inode->i_blocks = attr->blocks;
163 inode->i_atime.tv_sec = attr->atime;
164 inode->i_atime.tv_nsec = attr->atimensec;
165 /* mtime from server may be stale due to local buffered write */
166 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
167 inode->i_mtime.tv_sec = attr->mtime;
168 inode->i_mtime.tv_nsec = attr->mtimensec;
169 inode->i_ctime.tv_sec = attr->ctime;
170 inode->i_ctime.tv_nsec = attr->ctimensec;
173 if (attr->blksize != 0)
174 inode->i_blkbits = ilog2(attr->blksize);
176 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
179 * Don't set the sticky bit in i_mode, unless we want the VFS
180 * to check permissions. This prevents failures due to the
181 * check in may_delete().
183 fi->orig_i_mode = inode->i_mode;
184 if (!fc->default_permissions)
185 inode->i_mode &= ~S_ISVTX;
187 fi->orig_ino = attr->ino;
190 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
191 u64 attr_valid, u64 attr_version)
193 struct fuse_conn *fc = get_fuse_conn(inode);
194 struct fuse_inode *fi = get_fuse_inode(inode);
195 bool is_wb = fc->writeback_cache;
197 struct timespec64 old_mtime;
199 spin_lock(&fi->lock);
200 if ((attr_version != 0 && fi->attr_version > attr_version) ||
201 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
202 spin_unlock(&fi->lock);
206 old_mtime = inode->i_mtime;
207 fuse_change_attributes_common(inode, attr, attr_valid);
209 oldsize = inode->i_size;
211 * In case of writeback_cache enabled, the cached writes beyond EOF
212 * extend local i_size without keeping userspace server in sync. So,
213 * attr->size coming from server can be stale. We cannot trust it.
215 if (!is_wb || !S_ISREG(inode->i_mode))
216 i_size_write(inode, attr->size);
217 spin_unlock(&fi->lock);
219 if (!is_wb && S_ISREG(inode->i_mode)) {
222 if (oldsize != attr->size) {
223 truncate_pagecache(inode, attr->size);
224 if (!fc->explicit_inval_data)
226 } else if (fc->auto_inval_data) {
227 struct timespec64 new_mtime = {
228 .tv_sec = attr->mtime,
229 .tv_nsec = attr->mtimensec,
233 * Auto inval mode also checks and invalidates if mtime
236 if (!timespec64_equal(&old_mtime, &new_mtime))
241 invalidate_inode_pages2(inode->i_mapping);
245 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
247 inode->i_mode = attr->mode & S_IFMT;
248 inode->i_size = attr->size;
249 inode->i_mtime.tv_sec = attr->mtime;
250 inode->i_mtime.tv_nsec = attr->mtimensec;
251 inode->i_ctime.tv_sec = attr->ctime;
252 inode->i_ctime.tv_nsec = attr->ctimensec;
253 if (S_ISREG(inode->i_mode)) {
254 fuse_init_common(inode);
255 fuse_init_file_inode(inode);
256 } else if (S_ISDIR(inode->i_mode))
257 fuse_init_dir(inode);
258 else if (S_ISLNK(inode->i_mode))
259 fuse_init_symlink(inode);
260 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
261 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
262 fuse_init_common(inode);
263 init_special_inode(inode, inode->i_mode,
264 new_decode_dev(attr->rdev));
269 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
271 u64 nodeid = *(u64 *) _nodeidp;
272 if (get_node_id(inode) == nodeid)
278 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
280 u64 nodeid = *(u64 *) _nodeidp;
281 get_fuse_inode(inode)->nodeid = nodeid;
285 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
286 int generation, struct fuse_attr *attr,
287 u64 attr_valid, u64 attr_version)
290 struct fuse_inode *fi;
291 struct fuse_conn *fc = get_fuse_conn_super(sb);
294 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
298 if ((inode->i_state & I_NEW)) {
299 inode->i_flags |= S_NOATIME;
300 if (!fc->writeback_cache || !S_ISREG(attr->mode))
301 inode->i_flags |= S_NOCMTIME;
302 inode->i_generation = generation;
303 fuse_init_inode(inode, attr);
304 unlock_new_inode(inode);
305 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
306 /* Inode has changed type, any I/O on the old should fail */
307 make_bad_inode(inode);
312 fi = get_fuse_inode(inode);
313 spin_lock(&fi->lock);
315 spin_unlock(&fi->lock);
316 fuse_change_attributes(inode, attr, attr_valid, attr_version);
321 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
322 loff_t offset, loff_t len)
328 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
332 fuse_invalidate_attr(inode);
333 forget_all_cached_acls(inode);
335 pg_start = offset >> PAGE_SHIFT;
339 pg_end = (offset + len - 1) >> PAGE_SHIFT;
340 invalidate_inode_pages2_range(inode->i_mapping,
347 bool fuse_lock_inode(struct inode *inode)
351 if (!get_fuse_conn(inode)->parallel_dirops) {
352 mutex_lock(&get_fuse_inode(inode)->mutex);
359 void fuse_unlock_inode(struct inode *inode, bool locked)
362 mutex_unlock(&get_fuse_inode(inode)->mutex);
365 static void fuse_umount_begin(struct super_block *sb)
367 struct fuse_conn *fc = get_fuse_conn_super(sb);
369 if (!fc->no_force_umount)
373 static void fuse_send_destroy(struct fuse_conn *fc)
378 args.opcode = FUSE_DESTROY;
381 fuse_simple_request(fc, &args);
385 static void fuse_put_super(struct super_block *sb)
387 struct fuse_conn *fc = get_fuse_conn_super(sb);
389 mutex_lock(&fuse_mutex);
390 list_del(&fc->entry);
391 fuse_ctl_remove_conn(fc);
392 mutex_unlock(&fuse_mutex);
397 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
399 stbuf->f_type = FUSE_SUPER_MAGIC;
400 stbuf->f_bsize = attr->bsize;
401 stbuf->f_frsize = attr->frsize;
402 stbuf->f_blocks = attr->blocks;
403 stbuf->f_bfree = attr->bfree;
404 stbuf->f_bavail = attr->bavail;
405 stbuf->f_files = attr->files;
406 stbuf->f_ffree = attr->ffree;
407 stbuf->f_namelen = attr->namelen;
408 /* fsid is left zero */
411 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
413 struct super_block *sb = dentry->d_sb;
414 struct fuse_conn *fc = get_fuse_conn_super(sb);
416 struct fuse_statfs_out outarg;
419 if (!fuse_allow_current_process(fc)) {
420 buf->f_type = FUSE_SUPER_MAGIC;
424 memset(&outarg, 0, sizeof(outarg));
426 args.opcode = FUSE_STATFS;
427 args.nodeid = get_node_id(d_inode(dentry));
428 args.out_numargs = 1;
429 args.out_args[0].size = sizeof(outarg);
430 args.out_args[0].value = &outarg;
431 err = fuse_simple_request(fc, &args);
433 convert_fuse_statfs(buf, &outarg.st);
444 OPT_DEFAULT_PERMISSIONS,
451 static const struct fs_parameter_spec fuse_param_specs[] = {
452 fsparam_string ("source", OPT_SOURCE),
453 fsparam_u32 ("fd", OPT_FD),
454 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
455 fsparam_u32 ("user_id", OPT_USER_ID),
456 fsparam_u32 ("group_id", OPT_GROUP_ID),
457 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
458 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
459 fsparam_u32 ("max_read", OPT_MAX_READ),
460 fsparam_u32 ("blksize", OPT_BLKSIZE),
461 fsparam_string ("subtype", OPT_SUBTYPE),
465 static const struct fs_parameter_description fuse_fs_parameters = {
467 .specs = fuse_param_specs,
470 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
472 struct fs_parse_result result;
473 struct fuse_fs_context *ctx = fc->fs_private;
476 opt = fs_parse(fc, &fuse_fs_parameters, param, &result);
483 return invalf(fc, "fuse: Multiple sources specified");
484 fc->source = param->string;
485 param->string = NULL;
490 return invalf(fc, "fuse: Multiple subtypes specified");
491 ctx->subtype = param->string;
492 param->string = NULL;
496 ctx->fd = result.uint_32;
501 if (!fuse_valid_type(result.uint_32))
502 return invalf(fc, "fuse: Invalid rootmode");
503 ctx->rootmode = result.uint_32;
504 ctx->rootmode_present = 1;
508 ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
509 if (!uid_valid(ctx->user_id))
510 return invalf(fc, "fuse: Invalid user_id");
511 ctx->user_id_present = 1;
515 ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
516 if (!gid_valid(ctx->group_id))
517 return invalf(fc, "fuse: Invalid group_id");
518 ctx->group_id_present = 1;
521 case OPT_DEFAULT_PERMISSIONS:
522 ctx->default_permissions = 1;
525 case OPT_ALLOW_OTHER:
526 ctx->allow_other = 1;
530 ctx->max_read = result.uint_32;
535 return invalf(fc, "fuse: blksize only supported for fuseblk");
536 ctx->blksize = result.uint_32;
546 static void fuse_free_fc(struct fs_context *fc)
548 struct fuse_fs_context *ctx = fc->fs_private;
556 static int fuse_show_options(struct seq_file *m, struct dentry *root)
558 struct super_block *sb = root->d_sb;
559 struct fuse_conn *fc = get_fuse_conn_super(sb);
561 seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
562 seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
563 if (fc->default_permissions)
564 seq_puts(m, ",default_permissions");
566 seq_puts(m, ",allow_other");
567 if (fc->max_read != ~0)
568 seq_printf(m, ",max_read=%u", fc->max_read);
569 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
570 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
574 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
575 const struct fuse_iqueue_ops *ops,
578 memset(fiq, 0, sizeof(struct fuse_iqueue));
579 spin_lock_init(&fiq->lock);
580 init_waitqueue_head(&fiq->waitq);
581 INIT_LIST_HEAD(&fiq->pending);
582 INIT_LIST_HEAD(&fiq->interrupts);
583 fiq->forget_list_tail = &fiq->forget_list_head;
589 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
593 spin_lock_init(&fpq->lock);
594 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
595 INIT_LIST_HEAD(&fpq->processing[i]);
596 INIT_LIST_HEAD(&fpq->io);
600 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns,
601 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
603 memset(fc, 0, sizeof(*fc));
604 spin_lock_init(&fc->lock);
605 spin_lock_init(&fc->bg_lock);
606 init_rwsem(&fc->killsb);
607 refcount_set(&fc->count, 1);
608 atomic_set(&fc->dev_count, 1);
609 init_waitqueue_head(&fc->blocked_waitq);
610 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
611 INIT_LIST_HEAD(&fc->bg_queue);
612 INIT_LIST_HEAD(&fc->entry);
613 INIT_LIST_HEAD(&fc->devices);
614 atomic_set(&fc->num_waiting, 0);
615 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
616 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
617 atomic64_set(&fc->khctr, 0);
618 fc->polled_files = RB_ROOT;
622 atomic64_set(&fc->attr_version, 1);
623 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
624 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
625 fc->user_ns = get_user_ns(user_ns);
626 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
628 EXPORT_SYMBOL_GPL(fuse_conn_init);
630 void fuse_conn_put(struct fuse_conn *fc)
632 if (refcount_dec_and_test(&fc->count)) {
633 struct fuse_iqueue *fiq = &fc->iq;
635 if (fiq->ops->release)
636 fiq->ops->release(fiq);
637 put_pid_ns(fc->pid_ns);
638 put_user_ns(fc->user_ns);
642 EXPORT_SYMBOL_GPL(fuse_conn_put);
644 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
646 refcount_inc(&fc->count);
649 EXPORT_SYMBOL_GPL(fuse_conn_get);
651 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
653 struct fuse_attr attr;
654 memset(&attr, 0, sizeof(attr));
657 attr.ino = FUSE_ROOT_ID;
659 return fuse_iget(sb, 1, 0, &attr, 0, 0);
662 struct fuse_inode_handle {
667 static struct dentry *fuse_get_dentry(struct super_block *sb,
668 struct fuse_inode_handle *handle)
670 struct fuse_conn *fc = get_fuse_conn_super(sb);
672 struct dentry *entry;
675 if (handle->nodeid == 0)
678 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
680 struct fuse_entry_out outarg;
681 const struct qstr name = QSTR_INIT(".", 1);
683 if (!fc->export_support)
686 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
688 if (err && err != -ENOENT)
695 if (get_node_id(inode) != handle->nodeid)
699 if (inode->i_generation != handle->generation)
702 entry = d_obtain_alias(inode);
703 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
704 fuse_invalidate_entry_cache(entry);
714 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
715 struct inode *parent)
717 int len = parent ? 6 : 3;
721 if (*max_len < len) {
723 return FILEID_INVALID;
726 nodeid = get_fuse_inode(inode)->nodeid;
727 generation = inode->i_generation;
729 fh[0] = (u32)(nodeid >> 32);
730 fh[1] = (u32)(nodeid & 0xffffffff);
734 nodeid = get_fuse_inode(parent)->nodeid;
735 generation = parent->i_generation;
737 fh[3] = (u32)(nodeid >> 32);
738 fh[4] = (u32)(nodeid & 0xffffffff);
743 return parent ? 0x82 : 0x81;
746 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
747 struct fid *fid, int fh_len, int fh_type)
749 struct fuse_inode_handle handle;
751 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
754 handle.nodeid = (u64) fid->raw[0] << 32;
755 handle.nodeid |= (u64) fid->raw[1];
756 handle.generation = fid->raw[2];
757 return fuse_get_dentry(sb, &handle);
760 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
761 struct fid *fid, int fh_len, int fh_type)
763 struct fuse_inode_handle parent;
765 if (fh_type != 0x82 || fh_len < 6)
768 parent.nodeid = (u64) fid->raw[3] << 32;
769 parent.nodeid |= (u64) fid->raw[4];
770 parent.generation = fid->raw[5];
771 return fuse_get_dentry(sb, &parent);
774 static struct dentry *fuse_get_parent(struct dentry *child)
776 struct inode *child_inode = d_inode(child);
777 struct fuse_conn *fc = get_fuse_conn(child_inode);
779 struct dentry *parent;
780 struct fuse_entry_out outarg;
781 const struct qstr name = QSTR_INIT("..", 2);
784 if (!fc->export_support)
785 return ERR_PTR(-ESTALE);
787 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
788 &name, &outarg, &inode);
791 return ERR_PTR(-ESTALE);
795 parent = d_obtain_alias(inode);
796 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
797 fuse_invalidate_entry_cache(parent);
802 static const struct export_operations fuse_export_operations = {
803 .fh_to_dentry = fuse_fh_to_dentry,
804 .fh_to_parent = fuse_fh_to_parent,
805 .encode_fh = fuse_encode_fh,
806 .get_parent = fuse_get_parent,
809 static const struct super_operations fuse_super_operations = {
810 .alloc_inode = fuse_alloc_inode,
811 .free_inode = fuse_free_inode,
812 .evict_inode = fuse_evict_inode,
813 .write_inode = fuse_write_inode,
814 .drop_inode = generic_delete_inode,
815 .remount_fs = fuse_remount_fs,
816 .put_super = fuse_put_super,
817 .umount_begin = fuse_umount_begin,
818 .statfs = fuse_statfs,
819 .show_options = fuse_show_options,
822 static void sanitize_global_limit(unsigned *limit)
825 * The default maximum number of async requests is calculated to consume
826 * 1/2^13 of the total memory, assuming 392 bytes per request.
829 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
831 if (*limit >= 1 << 16)
832 *limit = (1 << 16) - 1;
835 static int set_global_limit(const char *val, const struct kernel_param *kp)
839 rv = param_set_uint(val, kp);
843 sanitize_global_limit((unsigned *)kp->arg);
848 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
850 int cap_sys_admin = capable(CAP_SYS_ADMIN);
855 sanitize_global_limit(&max_user_bgreq);
856 sanitize_global_limit(&max_user_congthresh);
858 spin_lock(&fc->bg_lock);
859 if (arg->max_background) {
860 fc->max_background = arg->max_background;
862 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
863 fc->max_background = max_user_bgreq;
865 if (arg->congestion_threshold) {
866 fc->congestion_threshold = arg->congestion_threshold;
868 if (!cap_sys_admin &&
869 fc->congestion_threshold > max_user_congthresh)
870 fc->congestion_threshold = max_user_congthresh;
872 spin_unlock(&fc->bg_lock);
875 struct fuse_init_args {
876 struct fuse_args args;
877 struct fuse_init_in in;
878 struct fuse_init_out out;
881 static void process_init_reply(struct fuse_conn *fc, struct fuse_args *args,
884 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
885 struct fuse_init_out *arg = &ia->out;
887 if (error || arg->major != FUSE_KERNEL_VERSION)
890 unsigned long ra_pages;
892 process_init_limits(fc, arg);
894 if (arg->minor >= 6) {
895 ra_pages = arg->max_readahead / PAGE_SIZE;
896 if (arg->flags & FUSE_ASYNC_READ)
898 if (!(arg->flags & FUSE_POSIX_LOCKS))
900 if (arg->minor >= 17) {
901 if (!(arg->flags & FUSE_FLOCK_LOCKS))
904 if (!(arg->flags & FUSE_POSIX_LOCKS))
907 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
908 fc->atomic_o_trunc = 1;
909 if (arg->minor >= 9) {
910 /* LOOKUP has dependency on proto version */
911 if (arg->flags & FUSE_EXPORT_SUPPORT)
912 fc->export_support = 1;
914 if (arg->flags & FUSE_BIG_WRITES)
916 if (arg->flags & FUSE_DONT_MASK)
918 if (arg->flags & FUSE_AUTO_INVAL_DATA)
919 fc->auto_inval_data = 1;
920 else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
921 fc->explicit_inval_data = 1;
922 if (arg->flags & FUSE_DO_READDIRPLUS) {
923 fc->do_readdirplus = 1;
924 if (arg->flags & FUSE_READDIRPLUS_AUTO)
925 fc->readdirplus_auto = 1;
927 if (arg->flags & FUSE_ASYNC_DIO)
929 if (arg->flags & FUSE_WRITEBACK_CACHE)
930 fc->writeback_cache = 1;
931 if (arg->flags & FUSE_PARALLEL_DIROPS)
932 fc->parallel_dirops = 1;
933 if (arg->flags & FUSE_HANDLE_KILLPRIV)
934 fc->handle_killpriv = 1;
935 if (arg->time_gran && arg->time_gran <= 1000000000)
936 fc->sb->s_time_gran = arg->time_gran;
937 if ((arg->flags & FUSE_POSIX_ACL)) {
938 fc->default_permissions = 1;
940 fc->sb->s_xattr = fuse_acl_xattr_handlers;
942 if (arg->flags & FUSE_CACHE_SYMLINKS)
943 fc->cache_symlinks = 1;
944 if (arg->flags & FUSE_ABORT_ERROR)
946 if (arg->flags & FUSE_MAX_PAGES) {
948 min_t(unsigned int, FUSE_MAX_MAX_PAGES,
949 max_t(unsigned int, arg->max_pages, 1));
952 ra_pages = fc->max_read / PAGE_SIZE;
957 fc->sb->s_bdi->ra_pages =
958 min(fc->sb->s_bdi->ra_pages, ra_pages);
959 fc->minor = arg->minor;
960 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
961 fc->max_write = max_t(unsigned, 4096, fc->max_write);
966 fuse_set_initialized(fc);
967 wake_up_all(&fc->blocked_waitq);
970 void fuse_send_init(struct fuse_conn *fc)
972 struct fuse_init_args *ia;
974 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
976 ia->in.major = FUSE_KERNEL_VERSION;
977 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
978 ia->in.max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
980 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
981 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
982 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
983 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
984 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
985 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
986 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
987 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
988 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
989 ia->args.opcode = FUSE_INIT;
990 ia->args.in_numargs = 1;
991 ia->args.in_args[0].size = sizeof(ia->in);
992 ia->args.in_args[0].value = &ia->in;
993 ia->args.out_numargs = 1;
994 /* Variable length argument used for backward compatibility
995 with interface version < 7.5. Rest of init_out is zeroed
996 by do_get_request(), so a short reply is not a problem */
997 ia->args.out_argvar = 1;
998 ia->args.out_args[0].size = sizeof(ia->out);
999 ia->args.out_args[0].value = &ia->out;
1000 ia->args.force = true;
1001 ia->args.nocreds = true;
1002 ia->args.end = process_init_reply;
1004 if (fuse_simple_background(fc, &ia->args, GFP_KERNEL) != 0)
1005 process_init_reply(fc, &ia->args, -ENOTCONN);
1007 EXPORT_SYMBOL_GPL(fuse_send_init);
1009 void fuse_free_conn(struct fuse_conn *fc)
1011 WARN_ON(!list_empty(&fc->devices));
1014 EXPORT_SYMBOL_GPL(fuse_free_conn);
1016 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1022 suffix = "-fuseblk";
1024 * sb->s_bdi points to blkdev's bdi however we want to redirect
1025 * it to our private bdi...
1028 sb->s_bdi = &noop_backing_dev_info;
1030 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1031 MINOR(fc->dev), suffix);
1035 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1036 /* fuse does it's own writeback accounting */
1037 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1040 * For a single fuse filesystem use max 1% of dirty +
1041 * writeback threshold.
1043 * This gives about 1M of write buffer for memory maps on a
1044 * machine with 1G and 10% dirty_ratio, which should be more
1047 * Privileged users can raise it by writing to
1049 * /sys/class/bdi/<bdi>/max_ratio
1051 bdi_set_max_ratio(sb->s_bdi, 1);
1056 struct fuse_dev *fuse_dev_alloc(void)
1058 struct fuse_dev *fud;
1059 struct list_head *pq;
1061 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1065 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1071 fud->pq.processing = pq;
1072 fuse_pqueue_init(&fud->pq);
1076 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1078 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1080 fud->fc = fuse_conn_get(fc);
1081 spin_lock(&fc->lock);
1082 list_add_tail(&fud->entry, &fc->devices);
1083 spin_unlock(&fc->lock);
1085 EXPORT_SYMBOL_GPL(fuse_dev_install);
1087 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1089 struct fuse_dev *fud;
1091 fud = fuse_dev_alloc();
1095 fuse_dev_install(fud, fc);
1098 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1100 void fuse_dev_free(struct fuse_dev *fud)
1102 struct fuse_conn *fc = fud->fc;
1105 spin_lock(&fc->lock);
1106 list_del(&fud->entry);
1107 spin_unlock(&fc->lock);
1111 kfree(fud->pq.processing);
1114 EXPORT_SYMBOL_GPL(fuse_dev_free);
1116 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1118 struct fuse_dev *fud;
1119 struct fuse_conn *fc = get_fuse_conn_super(sb);
1121 struct dentry *root_dentry;
1125 if (sb->s_flags & SB_MANDLOCK)
1128 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1133 if (!sb_set_blocksize(sb, ctx->blksize))
1137 sb->s_blocksize = PAGE_SIZE;
1138 sb->s_blocksize_bits = PAGE_SHIFT;
1141 sb->s_subtype = ctx->subtype;
1142 ctx->subtype = NULL;
1143 sb->s_magic = FUSE_SUPER_MAGIC;
1144 sb->s_op = &fuse_super_operations;
1145 sb->s_xattr = fuse_xattr_handlers;
1146 sb->s_maxbytes = MAX_LFS_FILESIZE;
1147 sb->s_time_gran = 1;
1148 sb->s_export_op = &fuse_export_operations;
1149 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1150 if (sb->s_user_ns != &init_user_ns)
1151 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1154 * If we are not in the initial user namespace posix
1155 * acls must be translated.
1157 if (sb->s_user_ns != &init_user_ns)
1158 sb->s_xattr = fuse_no_acl_xattr_handlers;
1160 fud = fuse_dev_alloc_install(fc);
1164 fc->dev = sb->s_dev;
1166 err = fuse_bdi_init(fc, sb);
1170 /* Handle umasking inside the fuse code */
1171 if (sb->s_flags & SB_POSIXACL)
1173 sb->s_flags |= SB_POSIXACL;
1175 fc->default_permissions = ctx->default_permissions;
1176 fc->allow_other = ctx->allow_other;
1177 fc->user_id = ctx->user_id;
1178 fc->group_id = ctx->group_id;
1179 fc->max_read = max_t(unsigned, 4096, ctx->max_read);
1180 fc->destroy = ctx->destroy;
1181 fc->no_control = ctx->no_control;
1182 fc->no_force_umount = ctx->no_force_umount;
1185 root = fuse_get_root_inode(sb, ctx->rootmode);
1186 sb->s_d_op = &fuse_root_dentry_operations;
1187 root_dentry = d_make_root(root);
1190 /* Root dentry doesn't have .d_revalidate */
1191 sb->s_d_op = &fuse_dentry_operations;
1193 mutex_lock(&fuse_mutex);
1198 err = fuse_ctl_add_conn(fc);
1202 list_add_tail(&fc->entry, &fuse_conn_list);
1203 sb->s_root = root_dentry;
1205 mutex_unlock(&fuse_mutex);
1209 mutex_unlock(&fuse_mutex);
1216 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1218 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1220 struct fuse_fs_context *ctx = fsc->fs_private;
1223 struct fuse_conn *fc;
1226 file = fget(ctx->fd);
1231 * Require mount to happen from the same user namespace which
1232 * opened /dev/fuse to prevent potential attacks.
1234 if ((file->f_op != &fuse_dev_operations) ||
1235 (file->f_cred->user_ns != sb->s_user_ns))
1237 ctx->fudptr = &file->private_data;
1239 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1244 fuse_conn_init(fc, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1245 fc->release = fuse_free_conn;
1248 err = fuse_fill_super_common(sb, ctx);
1252 * atomic_dec_and_test() in fput() provides the necessary
1253 * memory barrier for file->private_data to be visible on all
1257 fuse_send_init(get_fuse_conn_super(sb));
1262 sb->s_fs_info = NULL;
1269 static int fuse_get_tree(struct fs_context *fc)
1271 struct fuse_fs_context *ctx = fc->fs_private;
1273 if (!ctx->fd_present || !ctx->rootmode_present ||
1274 !ctx->user_id_present || !ctx->group_id_present)
1279 return get_tree_bdev(fc, fuse_fill_super);
1282 return get_tree_nodev(fc, fuse_fill_super);
1285 static const struct fs_context_operations fuse_context_ops = {
1286 .free = fuse_free_fc,
1287 .parse_param = fuse_parse_param,
1288 .get_tree = fuse_get_tree,
1292 * Set up the filesystem mount context.
1294 static int fuse_init_fs_context(struct fs_context *fc)
1296 struct fuse_fs_context *ctx;
1298 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1303 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1306 if (fc->fs_type == &fuseblk_fs_type) {
1307 ctx->is_bdev = true;
1308 ctx->destroy = true;
1312 fc->fs_private = ctx;
1313 fc->ops = &fuse_context_ops;
1317 static void fuse_sb_destroy(struct super_block *sb)
1319 struct fuse_conn *fc = get_fuse_conn_super(sb);
1323 fuse_send_destroy(fc);
1325 fuse_abort_conn(fc);
1326 fuse_wait_aborted(fc);
1328 down_write(&fc->killsb);
1330 up_write(&fc->killsb);
1334 void fuse_kill_sb_anon(struct super_block *sb)
1336 fuse_sb_destroy(sb);
1337 kill_anon_super(sb);
1339 EXPORT_SYMBOL_GPL(fuse_kill_sb_anon);
1341 static struct file_system_type fuse_fs_type = {
1342 .owner = THIS_MODULE,
1344 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1345 .init_fs_context = fuse_init_fs_context,
1346 .parameters = &fuse_fs_parameters,
1347 .kill_sb = fuse_kill_sb_anon,
1349 MODULE_ALIAS_FS("fuse");
1352 static void fuse_kill_sb_blk(struct super_block *sb)
1354 fuse_sb_destroy(sb);
1355 kill_block_super(sb);
1358 static struct file_system_type fuseblk_fs_type = {
1359 .owner = THIS_MODULE,
1361 .init_fs_context = fuse_init_fs_context,
1362 .parameters = &fuse_fs_parameters,
1363 .kill_sb = fuse_kill_sb_blk,
1364 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1366 MODULE_ALIAS_FS("fuseblk");
1368 static inline int register_fuseblk(void)
1370 return register_filesystem(&fuseblk_fs_type);
1373 static inline void unregister_fuseblk(void)
1375 unregister_filesystem(&fuseblk_fs_type);
1378 static inline int register_fuseblk(void)
1383 static inline void unregister_fuseblk(void)
1388 static void fuse_inode_init_once(void *foo)
1390 struct inode *inode = foo;
1392 inode_init_once(inode);
1395 static int __init fuse_fs_init(void)
1399 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1400 sizeof(struct fuse_inode), 0,
1401 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1402 fuse_inode_init_once);
1404 if (!fuse_inode_cachep)
1407 err = register_fuseblk();
1411 err = register_filesystem(&fuse_fs_type);
1418 unregister_fuseblk();
1420 kmem_cache_destroy(fuse_inode_cachep);
1425 static void fuse_fs_cleanup(void)
1427 unregister_filesystem(&fuse_fs_type);
1428 unregister_fuseblk();
1431 * Make sure all delayed rcu free inodes are flushed before we
1435 kmem_cache_destroy(fuse_inode_cachep);
1438 static struct kobject *fuse_kobj;
1440 static int fuse_sysfs_init(void)
1444 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1450 err = sysfs_create_mount_point(fuse_kobj, "connections");
1452 goto out_fuse_unregister;
1456 out_fuse_unregister:
1457 kobject_put(fuse_kobj);
1462 static void fuse_sysfs_cleanup(void)
1464 sysfs_remove_mount_point(fuse_kobj, "connections");
1465 kobject_put(fuse_kobj);
1468 static int __init fuse_init(void)
1472 pr_info("init (API version %i.%i)\n",
1473 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1475 INIT_LIST_HEAD(&fuse_conn_list);
1476 res = fuse_fs_init();
1480 res = fuse_dev_init();
1482 goto err_fs_cleanup;
1484 res = fuse_sysfs_init();
1486 goto err_dev_cleanup;
1488 res = fuse_ctl_init();
1490 goto err_sysfs_cleanup;
1492 sanitize_global_limit(&max_user_bgreq);
1493 sanitize_global_limit(&max_user_congthresh);
1498 fuse_sysfs_cleanup();
1507 static void __exit fuse_exit(void)
1512 fuse_sysfs_cleanup();
1517 module_init(fuse_init);
1518 module_exit(fuse_exit);