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) && !fuse_is_bad(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;
167 attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
168 attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
169 attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
171 inode->i_atime.tv_sec = attr->atime;
172 inode->i_atime.tv_nsec = attr->atimensec;
173 /* mtime from server may be stale due to local buffered write */
174 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
175 inode->i_mtime.tv_sec = attr->mtime;
176 inode->i_mtime.tv_nsec = attr->mtimensec;
177 inode->i_ctime.tv_sec = attr->ctime;
178 inode->i_ctime.tv_nsec = attr->ctimensec;
181 if (attr->blksize != 0)
182 inode->i_blkbits = ilog2(attr->blksize);
184 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
187 * Don't set the sticky bit in i_mode, unless we want the VFS
188 * to check permissions. This prevents failures due to the
189 * check in may_delete().
191 fi->orig_i_mode = inode->i_mode;
192 if (!fc->default_permissions)
193 inode->i_mode &= ~S_ISVTX;
195 fi->orig_ino = attr->ino;
198 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
199 u64 attr_valid, u64 attr_version)
201 struct fuse_conn *fc = get_fuse_conn(inode);
202 struct fuse_inode *fi = get_fuse_inode(inode);
203 bool is_wb = fc->writeback_cache;
205 struct timespec64 old_mtime;
207 spin_lock(&fi->lock);
208 if ((attr_version != 0 && fi->attr_version > attr_version) ||
209 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
210 spin_unlock(&fi->lock);
214 old_mtime = inode->i_mtime;
215 fuse_change_attributes_common(inode, attr, attr_valid);
217 oldsize = inode->i_size;
219 * In case of writeback_cache enabled, the cached writes beyond EOF
220 * extend local i_size without keeping userspace server in sync. So,
221 * attr->size coming from server can be stale. We cannot trust it.
223 if (!is_wb || !S_ISREG(inode->i_mode))
224 i_size_write(inode, attr->size);
225 spin_unlock(&fi->lock);
227 if (!is_wb && S_ISREG(inode->i_mode)) {
230 if (oldsize != attr->size) {
231 truncate_pagecache(inode, attr->size);
232 if (!fc->explicit_inval_data)
234 } else if (fc->auto_inval_data) {
235 struct timespec64 new_mtime = {
236 .tv_sec = attr->mtime,
237 .tv_nsec = attr->mtimensec,
241 * Auto inval mode also checks and invalidates if mtime
244 if (!timespec64_equal(&old_mtime, &new_mtime))
249 invalidate_inode_pages2(inode->i_mapping);
253 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
255 inode->i_mode = attr->mode & S_IFMT;
256 inode->i_size = attr->size;
257 inode->i_mtime.tv_sec = attr->mtime;
258 inode->i_mtime.tv_nsec = attr->mtimensec;
259 inode->i_ctime.tv_sec = attr->ctime;
260 inode->i_ctime.tv_nsec = attr->ctimensec;
261 if (S_ISREG(inode->i_mode)) {
262 fuse_init_common(inode);
263 fuse_init_file_inode(inode);
264 } else if (S_ISDIR(inode->i_mode))
265 fuse_init_dir(inode);
266 else if (S_ISLNK(inode->i_mode))
267 fuse_init_symlink(inode);
268 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
269 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
270 fuse_init_common(inode);
271 init_special_inode(inode, inode->i_mode,
272 new_decode_dev(attr->rdev));
277 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
279 u64 nodeid = *(u64 *) _nodeidp;
280 if (get_node_id(inode) == nodeid)
286 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
288 u64 nodeid = *(u64 *) _nodeidp;
289 get_fuse_inode(inode)->nodeid = nodeid;
293 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
294 int generation, struct fuse_attr *attr,
295 u64 attr_valid, u64 attr_version)
298 struct fuse_inode *fi;
299 struct fuse_conn *fc = get_fuse_conn_super(sb);
302 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
306 if ((inode->i_state & I_NEW)) {
307 inode->i_flags |= S_NOATIME;
308 if (!fc->writeback_cache || !S_ISREG(attr->mode))
309 inode->i_flags |= S_NOCMTIME;
310 inode->i_generation = generation;
311 fuse_init_inode(inode, attr);
312 unlock_new_inode(inode);
313 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
314 /* Inode has changed type, any I/O on the old should fail */
315 fuse_make_bad(inode);
320 fi = get_fuse_inode(inode);
321 spin_lock(&fi->lock);
323 spin_unlock(&fi->lock);
324 fuse_change_attributes(inode, attr, attr_valid, attr_version);
329 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
330 loff_t offset, loff_t len)
332 struct fuse_conn *fc = get_fuse_conn_super(sb);
333 struct fuse_inode *fi;
338 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
342 fi = get_fuse_inode(inode);
343 spin_lock(&fi->lock);
344 fi->attr_version = atomic64_inc_return(&fc->attr_version);
345 spin_unlock(&fi->lock);
347 fuse_invalidate_attr(inode);
348 forget_all_cached_acls(inode);
350 pg_start = offset >> PAGE_SHIFT;
354 pg_end = (offset + len - 1) >> PAGE_SHIFT;
355 invalidate_inode_pages2_range(inode->i_mapping,
362 bool fuse_lock_inode(struct inode *inode)
366 if (!get_fuse_conn(inode)->parallel_dirops) {
367 mutex_lock(&get_fuse_inode(inode)->mutex);
374 void fuse_unlock_inode(struct inode *inode, bool locked)
377 mutex_unlock(&get_fuse_inode(inode)->mutex);
380 static void fuse_umount_begin(struct super_block *sb)
382 struct fuse_conn *fc = get_fuse_conn_super(sb);
384 if (!fc->no_force_umount)
388 static void fuse_send_destroy(struct fuse_conn *fc)
393 args.opcode = FUSE_DESTROY;
396 fuse_simple_request(fc, &args);
400 static void fuse_put_super(struct super_block *sb)
402 struct fuse_conn *fc = get_fuse_conn_super(sb);
404 mutex_lock(&fuse_mutex);
405 list_del(&fc->entry);
406 fuse_ctl_remove_conn(fc);
407 mutex_unlock(&fuse_mutex);
412 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
414 stbuf->f_type = FUSE_SUPER_MAGIC;
415 stbuf->f_bsize = attr->bsize;
416 stbuf->f_frsize = attr->frsize;
417 stbuf->f_blocks = attr->blocks;
418 stbuf->f_bfree = attr->bfree;
419 stbuf->f_bavail = attr->bavail;
420 stbuf->f_files = attr->files;
421 stbuf->f_ffree = attr->ffree;
422 stbuf->f_namelen = attr->namelen;
423 /* fsid is left zero */
426 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
428 struct super_block *sb = dentry->d_sb;
429 struct fuse_conn *fc = get_fuse_conn_super(sb);
431 struct fuse_statfs_out outarg;
434 if (!fuse_allow_current_process(fc)) {
435 buf->f_type = FUSE_SUPER_MAGIC;
439 memset(&outarg, 0, sizeof(outarg));
441 args.opcode = FUSE_STATFS;
442 args.nodeid = get_node_id(d_inode(dentry));
443 args.out_numargs = 1;
444 args.out_args[0].size = sizeof(outarg);
445 args.out_args[0].value = &outarg;
446 err = fuse_simple_request(fc, &args);
448 convert_fuse_statfs(buf, &outarg.st);
459 OPT_DEFAULT_PERMISSIONS,
466 static const struct fs_parameter_spec fuse_param_specs[] = {
467 fsparam_string ("source", OPT_SOURCE),
468 fsparam_u32 ("fd", OPT_FD),
469 fsparam_u32oct ("rootmode", OPT_ROOTMODE),
470 fsparam_u32 ("user_id", OPT_USER_ID),
471 fsparam_u32 ("group_id", OPT_GROUP_ID),
472 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS),
473 fsparam_flag ("allow_other", OPT_ALLOW_OTHER),
474 fsparam_u32 ("max_read", OPT_MAX_READ),
475 fsparam_u32 ("blksize", OPT_BLKSIZE),
476 fsparam_string ("subtype", OPT_SUBTYPE),
480 static const struct fs_parameter_description fuse_fs_parameters = {
482 .specs = fuse_param_specs,
485 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
487 struct fs_parse_result result;
488 struct fuse_fs_context *ctx = fc->fs_private;
492 * Ignore options coming from mount(MS_REMOUNT) for backward
495 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE)
498 opt = fs_parse(fc, &fuse_fs_parameters, param, &result);
505 return invalf(fc, "fuse: Multiple sources specified");
506 fc->source = param->string;
507 param->string = NULL;
512 return invalf(fc, "fuse: Multiple subtypes specified");
513 ctx->subtype = param->string;
514 param->string = NULL;
518 ctx->fd = result.uint_32;
523 if (!fuse_valid_type(result.uint_32))
524 return invalf(fc, "fuse: Invalid rootmode");
525 ctx->rootmode = result.uint_32;
526 ctx->rootmode_present = 1;
530 ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
531 if (!uid_valid(ctx->user_id))
532 return invalf(fc, "fuse: Invalid user_id");
533 ctx->user_id_present = 1;
537 ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
538 if (!gid_valid(ctx->group_id))
539 return invalf(fc, "fuse: Invalid group_id");
540 ctx->group_id_present = 1;
543 case OPT_DEFAULT_PERMISSIONS:
544 ctx->default_permissions = 1;
547 case OPT_ALLOW_OTHER:
548 ctx->allow_other = 1;
552 ctx->max_read = result.uint_32;
557 return invalf(fc, "fuse: blksize only supported for fuseblk");
558 ctx->blksize = result.uint_32;
568 static void fuse_free_fc(struct fs_context *fc)
570 struct fuse_fs_context *ctx = fc->fs_private;
578 static int fuse_show_options(struct seq_file *m, struct dentry *root)
580 struct super_block *sb = root->d_sb;
581 struct fuse_conn *fc = get_fuse_conn_super(sb);
583 if (fc->no_mount_options)
586 seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
587 seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
588 if (fc->default_permissions)
589 seq_puts(m, ",default_permissions");
591 seq_puts(m, ",allow_other");
592 if (fc->max_read != ~0)
593 seq_printf(m, ",max_read=%u", fc->max_read);
594 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
595 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
599 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
600 const struct fuse_iqueue_ops *ops,
603 memset(fiq, 0, sizeof(struct fuse_iqueue));
604 spin_lock_init(&fiq->lock);
605 init_waitqueue_head(&fiq->waitq);
606 INIT_LIST_HEAD(&fiq->pending);
607 INIT_LIST_HEAD(&fiq->interrupts);
608 fiq->forget_list_tail = &fiq->forget_list_head;
614 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
618 spin_lock_init(&fpq->lock);
619 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
620 INIT_LIST_HEAD(&fpq->processing[i]);
621 INIT_LIST_HEAD(&fpq->io);
625 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns,
626 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
628 memset(fc, 0, sizeof(*fc));
629 spin_lock_init(&fc->lock);
630 spin_lock_init(&fc->bg_lock);
631 init_rwsem(&fc->killsb);
632 refcount_set(&fc->count, 1);
633 atomic_set(&fc->dev_count, 1);
634 init_waitqueue_head(&fc->blocked_waitq);
635 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
636 INIT_LIST_HEAD(&fc->bg_queue);
637 INIT_LIST_HEAD(&fc->entry);
638 INIT_LIST_HEAD(&fc->devices);
639 atomic_set(&fc->num_waiting, 0);
640 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
641 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
642 atomic64_set(&fc->khctr, 0);
643 fc->polled_files = RB_ROOT;
647 atomic64_set(&fc->attr_version, 1);
648 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
649 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
650 fc->user_ns = get_user_ns(user_ns);
651 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
653 EXPORT_SYMBOL_GPL(fuse_conn_init);
655 void fuse_conn_put(struct fuse_conn *fc)
657 if (refcount_dec_and_test(&fc->count)) {
658 struct fuse_iqueue *fiq = &fc->iq;
660 if (fiq->ops->release)
661 fiq->ops->release(fiq);
662 put_pid_ns(fc->pid_ns);
663 put_user_ns(fc->user_ns);
667 EXPORT_SYMBOL_GPL(fuse_conn_put);
669 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
671 refcount_inc(&fc->count);
674 EXPORT_SYMBOL_GPL(fuse_conn_get);
676 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
678 struct fuse_attr attr;
679 memset(&attr, 0, sizeof(attr));
682 attr.ino = FUSE_ROOT_ID;
684 return fuse_iget(sb, 1, 0, &attr, 0, 0);
687 struct fuse_inode_handle {
692 static struct dentry *fuse_get_dentry(struct super_block *sb,
693 struct fuse_inode_handle *handle)
695 struct fuse_conn *fc = get_fuse_conn_super(sb);
697 struct dentry *entry;
700 if (handle->nodeid == 0)
703 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
705 struct fuse_entry_out outarg;
706 const struct qstr name = QSTR_INIT(".", 1);
708 if (!fc->export_support)
711 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
713 if (err && err != -ENOENT)
720 if (get_node_id(inode) != handle->nodeid)
724 if (inode->i_generation != handle->generation)
727 entry = d_obtain_alias(inode);
728 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
729 fuse_invalidate_entry_cache(entry);
739 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
740 struct inode *parent)
742 int len = parent ? 6 : 3;
746 if (*max_len < len) {
748 return FILEID_INVALID;
751 nodeid = get_fuse_inode(inode)->nodeid;
752 generation = inode->i_generation;
754 fh[0] = (u32)(nodeid >> 32);
755 fh[1] = (u32)(nodeid & 0xffffffff);
759 nodeid = get_fuse_inode(parent)->nodeid;
760 generation = parent->i_generation;
762 fh[3] = (u32)(nodeid >> 32);
763 fh[4] = (u32)(nodeid & 0xffffffff);
768 return parent ? 0x82 : 0x81;
771 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
772 struct fid *fid, int fh_len, int fh_type)
774 struct fuse_inode_handle handle;
776 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
779 handle.nodeid = (u64) fid->raw[0] << 32;
780 handle.nodeid |= (u64) fid->raw[1];
781 handle.generation = fid->raw[2];
782 return fuse_get_dentry(sb, &handle);
785 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
786 struct fid *fid, int fh_len, int fh_type)
788 struct fuse_inode_handle parent;
790 if (fh_type != 0x82 || fh_len < 6)
793 parent.nodeid = (u64) fid->raw[3] << 32;
794 parent.nodeid |= (u64) fid->raw[4];
795 parent.generation = fid->raw[5];
796 return fuse_get_dentry(sb, &parent);
799 static struct dentry *fuse_get_parent(struct dentry *child)
801 struct inode *child_inode = d_inode(child);
802 struct fuse_conn *fc = get_fuse_conn(child_inode);
804 struct dentry *parent;
805 struct fuse_entry_out outarg;
806 const struct qstr name = QSTR_INIT("..", 2);
809 if (!fc->export_support)
810 return ERR_PTR(-ESTALE);
812 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
813 &name, &outarg, &inode);
816 return ERR_PTR(-ESTALE);
820 parent = d_obtain_alias(inode);
821 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
822 fuse_invalidate_entry_cache(parent);
827 static const struct export_operations fuse_export_operations = {
828 .fh_to_dentry = fuse_fh_to_dentry,
829 .fh_to_parent = fuse_fh_to_parent,
830 .encode_fh = fuse_encode_fh,
831 .get_parent = fuse_get_parent,
834 static const struct super_operations fuse_super_operations = {
835 .alloc_inode = fuse_alloc_inode,
836 .free_inode = fuse_free_inode,
837 .evict_inode = fuse_evict_inode,
838 .write_inode = fuse_write_inode,
839 .drop_inode = generic_delete_inode,
840 .put_super = fuse_put_super,
841 .umount_begin = fuse_umount_begin,
842 .statfs = fuse_statfs,
843 .show_options = fuse_show_options,
846 static void sanitize_global_limit(unsigned *limit)
849 * The default maximum number of async requests is calculated to consume
850 * 1/2^13 of the total memory, assuming 392 bytes per request.
853 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
855 if (*limit >= 1 << 16)
856 *limit = (1 << 16) - 1;
859 static int set_global_limit(const char *val, const struct kernel_param *kp)
863 rv = param_set_uint(val, kp);
867 sanitize_global_limit((unsigned *)kp->arg);
872 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
874 int cap_sys_admin = capable(CAP_SYS_ADMIN);
879 sanitize_global_limit(&max_user_bgreq);
880 sanitize_global_limit(&max_user_congthresh);
882 spin_lock(&fc->bg_lock);
883 if (arg->max_background) {
884 fc->max_background = arg->max_background;
886 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
887 fc->max_background = max_user_bgreq;
889 if (arg->congestion_threshold) {
890 fc->congestion_threshold = arg->congestion_threshold;
892 if (!cap_sys_admin &&
893 fc->congestion_threshold > max_user_congthresh)
894 fc->congestion_threshold = max_user_congthresh;
896 spin_unlock(&fc->bg_lock);
899 struct fuse_init_args {
900 struct fuse_args args;
901 struct fuse_init_in in;
902 struct fuse_init_out out;
905 static void process_init_reply(struct fuse_conn *fc, struct fuse_args *args,
908 struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
909 struct fuse_init_out *arg = &ia->out;
911 if (error || arg->major != FUSE_KERNEL_VERSION)
914 unsigned long ra_pages;
916 process_init_limits(fc, arg);
918 if (arg->minor >= 6) {
919 ra_pages = arg->max_readahead / PAGE_SIZE;
920 if (arg->flags & FUSE_ASYNC_READ)
922 if (!(arg->flags & FUSE_POSIX_LOCKS))
924 if (arg->minor >= 17) {
925 if (!(arg->flags & FUSE_FLOCK_LOCKS))
928 if (!(arg->flags & FUSE_POSIX_LOCKS))
931 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
932 fc->atomic_o_trunc = 1;
933 if (arg->minor >= 9) {
934 /* LOOKUP has dependency on proto version */
935 if (arg->flags & FUSE_EXPORT_SUPPORT)
936 fc->export_support = 1;
938 if (arg->flags & FUSE_BIG_WRITES)
940 if (arg->flags & FUSE_DONT_MASK)
942 if (arg->flags & FUSE_AUTO_INVAL_DATA)
943 fc->auto_inval_data = 1;
944 else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
945 fc->explicit_inval_data = 1;
946 if (arg->flags & FUSE_DO_READDIRPLUS) {
947 fc->do_readdirplus = 1;
948 if (arg->flags & FUSE_READDIRPLUS_AUTO)
949 fc->readdirplus_auto = 1;
951 if (arg->flags & FUSE_ASYNC_DIO)
953 if (arg->flags & FUSE_WRITEBACK_CACHE)
954 fc->writeback_cache = 1;
955 if (arg->flags & FUSE_PARALLEL_DIROPS)
956 fc->parallel_dirops = 1;
957 if (arg->flags & FUSE_HANDLE_KILLPRIV)
958 fc->handle_killpriv = 1;
959 if (arg->time_gran && arg->time_gran <= 1000000000)
960 fc->sb->s_time_gran = arg->time_gran;
961 if ((arg->flags & FUSE_POSIX_ACL)) {
962 fc->default_permissions = 1;
964 fc->sb->s_xattr = fuse_acl_xattr_handlers;
966 if (arg->flags & FUSE_CACHE_SYMLINKS)
967 fc->cache_symlinks = 1;
968 if (arg->flags & FUSE_ABORT_ERROR)
970 if (arg->flags & FUSE_MAX_PAGES) {
972 min_t(unsigned int, FUSE_MAX_MAX_PAGES,
973 max_t(unsigned int, arg->max_pages, 1));
976 ra_pages = fc->max_read / PAGE_SIZE;
981 fc->sb->s_bdi->ra_pages =
982 min(fc->sb->s_bdi->ra_pages, ra_pages);
983 fc->minor = arg->minor;
984 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
985 fc->max_write = max_t(unsigned, 4096, fc->max_write);
990 fuse_set_initialized(fc);
991 wake_up_all(&fc->blocked_waitq);
994 void fuse_send_init(struct fuse_conn *fc)
996 struct fuse_init_args *ia;
998 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1000 ia->in.major = FUSE_KERNEL_VERSION;
1001 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1002 ia->in.max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
1004 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1005 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1006 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1007 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1008 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1009 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1010 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1011 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1012 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
1013 ia->args.opcode = FUSE_INIT;
1014 ia->args.in_numargs = 1;
1015 ia->args.in_args[0].size = sizeof(ia->in);
1016 ia->args.in_args[0].value = &ia->in;
1017 ia->args.out_numargs = 1;
1018 /* Variable length argument used for backward compatibility
1019 with interface version < 7.5. Rest of init_out is zeroed
1020 by do_get_request(), so a short reply is not a problem */
1021 ia->args.out_argvar = 1;
1022 ia->args.out_args[0].size = sizeof(ia->out);
1023 ia->args.out_args[0].value = &ia->out;
1024 ia->args.force = true;
1025 ia->args.nocreds = true;
1026 ia->args.end = process_init_reply;
1028 if (fuse_simple_background(fc, &ia->args, GFP_KERNEL) != 0)
1029 process_init_reply(fc, &ia->args, -ENOTCONN);
1031 EXPORT_SYMBOL_GPL(fuse_send_init);
1033 void fuse_free_conn(struct fuse_conn *fc)
1035 WARN_ON(!list_empty(&fc->devices));
1038 EXPORT_SYMBOL_GPL(fuse_free_conn);
1040 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1046 suffix = "-fuseblk";
1048 * sb->s_bdi points to blkdev's bdi however we want to redirect
1049 * it to our private bdi...
1052 sb->s_bdi = &noop_backing_dev_info;
1054 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1055 MINOR(fc->dev), suffix);
1059 sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1060 /* fuse does it's own writeback accounting */
1061 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1064 * For a single fuse filesystem use max 1% of dirty +
1065 * writeback threshold.
1067 * This gives about 1M of write buffer for memory maps on a
1068 * machine with 1G and 10% dirty_ratio, which should be more
1071 * Privileged users can raise it by writing to
1073 * /sys/class/bdi/<bdi>/max_ratio
1075 bdi_set_max_ratio(sb->s_bdi, 1);
1080 struct fuse_dev *fuse_dev_alloc(void)
1082 struct fuse_dev *fud;
1083 struct list_head *pq;
1085 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1089 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1095 fud->pq.processing = pq;
1096 fuse_pqueue_init(&fud->pq);
1100 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1102 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1104 fud->fc = fuse_conn_get(fc);
1105 spin_lock(&fc->lock);
1106 list_add_tail(&fud->entry, &fc->devices);
1107 spin_unlock(&fc->lock);
1109 EXPORT_SYMBOL_GPL(fuse_dev_install);
1111 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1113 struct fuse_dev *fud;
1115 fud = fuse_dev_alloc();
1119 fuse_dev_install(fud, fc);
1122 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1124 void fuse_dev_free(struct fuse_dev *fud)
1126 struct fuse_conn *fc = fud->fc;
1129 spin_lock(&fc->lock);
1130 list_del(&fud->entry);
1131 spin_unlock(&fc->lock);
1135 kfree(fud->pq.processing);
1138 EXPORT_SYMBOL_GPL(fuse_dev_free);
1140 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1142 struct fuse_dev *fud;
1143 struct fuse_conn *fc = get_fuse_conn_super(sb);
1145 struct dentry *root_dentry;
1149 if (sb->s_flags & SB_MANDLOCK)
1152 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1157 if (!sb_set_blocksize(sb, ctx->blksize))
1161 sb->s_blocksize = PAGE_SIZE;
1162 sb->s_blocksize_bits = PAGE_SHIFT;
1165 sb->s_subtype = ctx->subtype;
1166 ctx->subtype = NULL;
1167 sb->s_magic = FUSE_SUPER_MAGIC;
1168 sb->s_op = &fuse_super_operations;
1169 sb->s_xattr = fuse_xattr_handlers;
1170 sb->s_maxbytes = MAX_LFS_FILESIZE;
1171 sb->s_time_gran = 1;
1172 sb->s_export_op = &fuse_export_operations;
1173 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1174 if (sb->s_user_ns != &init_user_ns)
1175 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1178 * If we are not in the initial user namespace posix
1179 * acls must be translated.
1181 if (sb->s_user_ns != &init_user_ns)
1182 sb->s_xattr = fuse_no_acl_xattr_handlers;
1184 fud = fuse_dev_alloc_install(fc);
1188 fc->dev = sb->s_dev;
1190 err = fuse_bdi_init(fc, sb);
1194 /* Handle umasking inside the fuse code */
1195 if (sb->s_flags & SB_POSIXACL)
1197 sb->s_flags |= SB_POSIXACL;
1199 fc->default_permissions = ctx->default_permissions;
1200 fc->allow_other = ctx->allow_other;
1201 fc->user_id = ctx->user_id;
1202 fc->group_id = ctx->group_id;
1203 fc->max_read = max_t(unsigned, 4096, ctx->max_read);
1204 fc->destroy = ctx->destroy;
1205 fc->no_control = ctx->no_control;
1206 fc->no_force_umount = ctx->no_force_umount;
1207 fc->no_mount_options = ctx->no_mount_options;
1210 root = fuse_get_root_inode(sb, ctx->rootmode);
1211 sb->s_d_op = &fuse_root_dentry_operations;
1212 root_dentry = d_make_root(root);
1215 /* Root dentry doesn't have .d_revalidate */
1216 sb->s_d_op = &fuse_dentry_operations;
1218 mutex_lock(&fuse_mutex);
1223 err = fuse_ctl_add_conn(fc);
1227 list_add_tail(&fc->entry, &fuse_conn_list);
1228 sb->s_root = root_dentry;
1230 mutex_unlock(&fuse_mutex);
1234 mutex_unlock(&fuse_mutex);
1241 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1243 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1245 struct fuse_fs_context *ctx = fsc->fs_private;
1248 struct fuse_conn *fc;
1251 file = fget(ctx->fd);
1256 * Require mount to happen from the same user namespace which
1257 * opened /dev/fuse to prevent potential attacks.
1259 if ((file->f_op != &fuse_dev_operations) ||
1260 (file->f_cred->user_ns != sb->s_user_ns))
1262 ctx->fudptr = &file->private_data;
1264 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1269 fuse_conn_init(fc, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1270 fc->release = fuse_free_conn;
1273 err = fuse_fill_super_common(sb, ctx);
1277 * atomic_dec_and_test() in fput() provides the necessary
1278 * memory barrier for file->private_data to be visible on all
1282 fuse_send_init(get_fuse_conn_super(sb));
1287 sb->s_fs_info = NULL;
1294 static int fuse_get_tree(struct fs_context *fc)
1296 struct fuse_fs_context *ctx = fc->fs_private;
1298 if (!ctx->fd_present || !ctx->rootmode_present ||
1299 !ctx->user_id_present || !ctx->group_id_present)
1304 return get_tree_bdev(fc, fuse_fill_super);
1307 return get_tree_nodev(fc, fuse_fill_super);
1310 static const struct fs_context_operations fuse_context_ops = {
1311 .free = fuse_free_fc,
1312 .parse_param = fuse_parse_param,
1313 .reconfigure = fuse_reconfigure,
1314 .get_tree = fuse_get_tree,
1318 * Set up the filesystem mount context.
1320 static int fuse_init_fs_context(struct fs_context *fc)
1322 struct fuse_fs_context *ctx;
1324 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1329 ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1332 if (fc->fs_type == &fuseblk_fs_type) {
1333 ctx->is_bdev = true;
1334 ctx->destroy = true;
1338 fc->fs_private = ctx;
1339 fc->ops = &fuse_context_ops;
1343 static void fuse_sb_destroy(struct super_block *sb)
1345 struct fuse_conn *fc = get_fuse_conn_super(sb);
1349 fuse_send_destroy(fc);
1351 fuse_abort_conn(fc);
1352 fuse_wait_aborted(fc);
1354 down_write(&fc->killsb);
1356 up_write(&fc->killsb);
1360 void fuse_kill_sb_anon(struct super_block *sb)
1362 fuse_sb_destroy(sb);
1363 kill_anon_super(sb);
1365 EXPORT_SYMBOL_GPL(fuse_kill_sb_anon);
1367 static struct file_system_type fuse_fs_type = {
1368 .owner = THIS_MODULE,
1370 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1371 .init_fs_context = fuse_init_fs_context,
1372 .parameters = &fuse_fs_parameters,
1373 .kill_sb = fuse_kill_sb_anon,
1375 MODULE_ALIAS_FS("fuse");
1378 static void fuse_kill_sb_blk(struct super_block *sb)
1380 fuse_sb_destroy(sb);
1381 kill_block_super(sb);
1384 static struct file_system_type fuseblk_fs_type = {
1385 .owner = THIS_MODULE,
1387 .init_fs_context = fuse_init_fs_context,
1388 .parameters = &fuse_fs_parameters,
1389 .kill_sb = fuse_kill_sb_blk,
1390 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1392 MODULE_ALIAS_FS("fuseblk");
1394 static inline int register_fuseblk(void)
1396 return register_filesystem(&fuseblk_fs_type);
1399 static inline void unregister_fuseblk(void)
1401 unregister_filesystem(&fuseblk_fs_type);
1404 static inline int register_fuseblk(void)
1409 static inline void unregister_fuseblk(void)
1414 static void fuse_inode_init_once(void *foo)
1416 struct inode *inode = foo;
1418 inode_init_once(inode);
1421 static int __init fuse_fs_init(void)
1425 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1426 sizeof(struct fuse_inode), 0,
1427 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1428 fuse_inode_init_once);
1430 if (!fuse_inode_cachep)
1433 err = register_fuseblk();
1437 err = register_filesystem(&fuse_fs_type);
1444 unregister_fuseblk();
1446 kmem_cache_destroy(fuse_inode_cachep);
1451 static void fuse_fs_cleanup(void)
1453 unregister_filesystem(&fuse_fs_type);
1454 unregister_fuseblk();
1457 * Make sure all delayed rcu free inodes are flushed before we
1461 kmem_cache_destroy(fuse_inode_cachep);
1464 static struct kobject *fuse_kobj;
1466 static int fuse_sysfs_init(void)
1470 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1476 err = sysfs_create_mount_point(fuse_kobj, "connections");
1478 goto out_fuse_unregister;
1482 out_fuse_unregister:
1483 kobject_put(fuse_kobj);
1488 static void fuse_sysfs_cleanup(void)
1490 sysfs_remove_mount_point(fuse_kobj, "connections");
1491 kobject_put(fuse_kobj);
1494 static int __init fuse_init(void)
1498 pr_info("init (API version %i.%i)\n",
1499 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1501 INIT_LIST_HEAD(&fuse_conn_list);
1502 res = fuse_fs_init();
1506 res = fuse_dev_init();
1508 goto err_fs_cleanup;
1510 res = fuse_sysfs_init();
1512 goto err_dev_cleanup;
1514 res = fuse_ctl_init();
1516 goto err_sysfs_cleanup;
1518 sanitize_global_limit(&max_user_bgreq);
1519 sanitize_global_limit(&max_user_congthresh);
1524 fuse_sysfs_cleanup();
1533 static void __exit fuse_exit(void)
1538 fuse_sysfs_cleanup();
1543 module_init(fuse_init);
1544 module_exit(fuse_exit);