Linux 4.14.215
[linux/fpc-iii.git] / fs / fuse / inode.c
blobffb61787d77af84b041ea61a267c67779504dcfc
1 /*
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.
6 See the file COPYING.
7 */
9 #include "fuse_i.h"
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/seq_file.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/moduleparam.h>
18 #include <linux/parser.h>
19 #include <linux/statfs.h>
20 #include <linux/random.h>
21 #include <linux/sched.h>
22 #include <linux/exportfs.h>
23 #include <linux/posix_acl.h>
24 #include <linux/pid_namespace.h>
26 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
27 MODULE_DESCRIPTION("Filesystem in Userspace");
28 MODULE_LICENSE("GPL");
30 static struct kmem_cache *fuse_inode_cachep;
31 struct list_head fuse_conn_list;
32 DEFINE_MUTEX(fuse_mutex);
34 static int set_global_limit(const char *val, struct kernel_param *kp);
36 unsigned max_user_bgreq;
37 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
38 &max_user_bgreq, 0644);
39 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
40 MODULE_PARM_DESC(max_user_bgreq,
41 "Global limit for the maximum number of backgrounded requests an "
42 "unprivileged user can set");
44 unsigned max_user_congthresh;
45 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
46 &max_user_congthresh, 0644);
47 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
48 MODULE_PARM_DESC(max_user_congthresh,
49 "Global limit for the maximum congestion threshold an "
50 "unprivileged user can set");
52 #define FUSE_SUPER_MAGIC 0x65735546
54 #define FUSE_DEFAULT_BLKSIZE 512
56 /** Maximum number of outstanding background requests */
57 #define FUSE_DEFAULT_MAX_BACKGROUND 12
59 /** Congestion starts at 75% of maximum */
60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
62 struct fuse_mount_data {
63 int fd;
64 unsigned rootmode;
65 kuid_t user_id;
66 kgid_t group_id;
67 unsigned fd_present:1;
68 unsigned rootmode_present:1;
69 unsigned user_id_present:1;
70 unsigned group_id_present:1;
71 unsigned default_permissions:1;
72 unsigned allow_other:1;
73 unsigned max_read;
74 unsigned blksize;
77 struct fuse_forget_link *fuse_alloc_forget(void)
79 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
82 static struct inode *fuse_alloc_inode(struct super_block *sb)
84 struct inode *inode;
85 struct fuse_inode *fi;
87 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
88 if (!inode)
89 return NULL;
91 fi = get_fuse_inode(inode);
92 fi->i_time = 0;
93 fi->nodeid = 0;
94 fi->nlookup = 0;
95 fi->attr_version = 0;
96 fi->writectr = 0;
97 fi->orig_ino = 0;
98 fi->state = 0;
99 INIT_LIST_HEAD(&fi->write_files);
100 INIT_LIST_HEAD(&fi->queued_writes);
101 INIT_LIST_HEAD(&fi->writepages);
102 init_waitqueue_head(&fi->page_waitq);
103 mutex_init(&fi->mutex);
104 fi->forget = fuse_alloc_forget();
105 if (!fi->forget) {
106 kmem_cache_free(fuse_inode_cachep, inode);
107 return NULL;
110 return inode;
113 static void fuse_i_callback(struct rcu_head *head)
115 struct inode *inode = container_of(head, struct inode, i_rcu);
116 kmem_cache_free(fuse_inode_cachep, inode);
119 static void fuse_destroy_inode(struct inode *inode)
121 struct fuse_inode *fi = get_fuse_inode(inode);
122 BUG_ON(!list_empty(&fi->write_files));
123 BUG_ON(!list_empty(&fi->queued_writes));
124 mutex_destroy(&fi->mutex);
125 kfree(fi->forget);
126 call_rcu(&inode->i_rcu, fuse_i_callback);
129 static void fuse_evict_inode(struct inode *inode)
131 truncate_inode_pages_final(&inode->i_data);
132 clear_inode(inode);
133 if (inode->i_sb->s_flags & MS_ACTIVE) {
134 struct fuse_conn *fc = get_fuse_conn(inode);
135 struct fuse_inode *fi = get_fuse_inode(inode);
136 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
137 fi->forget = NULL;
141 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
143 sync_filesystem(sb);
144 if (*flags & MS_MANDLOCK)
145 return -EINVAL;
147 return 0;
151 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
152 * so that it will fit.
154 static ino_t fuse_squash_ino(u64 ino64)
156 ino_t ino = (ino_t) ino64;
157 if (sizeof(ino_t) < sizeof(u64))
158 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
159 return ino;
162 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
163 u64 attr_valid)
165 struct fuse_conn *fc = get_fuse_conn(inode);
166 struct fuse_inode *fi = get_fuse_inode(inode);
168 fi->attr_version = ++fc->attr_version;
169 fi->i_time = attr_valid;
171 inode->i_ino = fuse_squash_ino(attr->ino);
172 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
173 set_nlink(inode, attr->nlink);
174 inode->i_uid = make_kuid(&init_user_ns, attr->uid);
175 inode->i_gid = make_kgid(&init_user_ns, attr->gid);
176 inode->i_blocks = attr->blocks;
177 inode->i_atime.tv_sec = attr->atime;
178 inode->i_atime.tv_nsec = attr->atimensec;
179 /* mtime from server may be stale due to local buffered write */
180 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
181 inode->i_mtime.tv_sec = attr->mtime;
182 inode->i_mtime.tv_nsec = attr->mtimensec;
183 inode->i_ctime.tv_sec = attr->ctime;
184 inode->i_ctime.tv_nsec = attr->ctimensec;
187 if (attr->blksize != 0)
188 inode->i_blkbits = ilog2(attr->blksize);
189 else
190 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
193 * Don't set the sticky bit in i_mode, unless we want the VFS
194 * to check permissions. This prevents failures due to the
195 * check in may_delete().
197 fi->orig_i_mode = inode->i_mode;
198 if (!fc->default_permissions)
199 inode->i_mode &= ~S_ISVTX;
201 fi->orig_ino = attr->ino;
204 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
205 u64 attr_valid, u64 attr_version)
207 struct fuse_conn *fc = get_fuse_conn(inode);
208 struct fuse_inode *fi = get_fuse_inode(inode);
209 bool is_wb = fc->writeback_cache;
210 loff_t oldsize;
211 struct timespec old_mtime;
213 spin_lock(&fc->lock);
214 if ((attr_version != 0 && fi->attr_version > attr_version) ||
215 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
216 spin_unlock(&fc->lock);
217 return;
220 old_mtime = inode->i_mtime;
221 fuse_change_attributes_common(inode, attr, attr_valid);
223 oldsize = inode->i_size;
225 * In case of writeback_cache enabled, the cached writes beyond EOF
226 * extend local i_size without keeping userspace server in sync. So,
227 * attr->size coming from server can be stale. We cannot trust it.
229 if (!is_wb || !S_ISREG(inode->i_mode))
230 i_size_write(inode, attr->size);
231 spin_unlock(&fc->lock);
233 if (!is_wb && S_ISREG(inode->i_mode)) {
234 bool inval = false;
236 if (oldsize != attr->size) {
237 truncate_pagecache(inode, attr->size);
238 inval = true;
239 } else if (fc->auto_inval_data) {
240 struct timespec new_mtime = {
241 .tv_sec = attr->mtime,
242 .tv_nsec = attr->mtimensec,
246 * Auto inval mode also checks and invalidates if mtime
247 * has changed.
249 if (!timespec_equal(&old_mtime, &new_mtime))
250 inval = true;
253 if (inval)
254 invalidate_inode_pages2(inode->i_mapping);
258 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
260 inode->i_mode = attr->mode & S_IFMT;
261 inode->i_size = attr->size;
262 inode->i_mtime.tv_sec = attr->mtime;
263 inode->i_mtime.tv_nsec = attr->mtimensec;
264 inode->i_ctime.tv_sec = attr->ctime;
265 inode->i_ctime.tv_nsec = attr->ctimensec;
266 if (S_ISREG(inode->i_mode)) {
267 fuse_init_common(inode);
268 fuse_init_file_inode(inode);
269 } else if (S_ISDIR(inode->i_mode))
270 fuse_init_dir(inode);
271 else if (S_ISLNK(inode->i_mode))
272 fuse_init_symlink(inode);
273 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
274 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
275 fuse_init_common(inode);
276 init_special_inode(inode, inode->i_mode,
277 new_decode_dev(attr->rdev));
278 } else
279 BUG();
282 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
284 u64 nodeid = *(u64 *) _nodeidp;
285 if (get_node_id(inode) == nodeid)
286 return 1;
287 else
288 return 0;
291 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
293 u64 nodeid = *(u64 *) _nodeidp;
294 get_fuse_inode(inode)->nodeid = nodeid;
295 return 0;
298 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
299 int generation, struct fuse_attr *attr,
300 u64 attr_valid, u64 attr_version)
302 struct inode *inode;
303 struct fuse_inode *fi;
304 struct fuse_conn *fc = get_fuse_conn_super(sb);
306 retry:
307 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
308 if (!inode)
309 return NULL;
311 if ((inode->i_state & I_NEW)) {
312 inode->i_flags |= S_NOATIME;
313 if (!fc->writeback_cache || !S_ISREG(attr->mode))
314 inode->i_flags |= S_NOCMTIME;
315 inode->i_generation = generation;
316 fuse_init_inode(inode, attr);
317 unlock_new_inode(inode);
318 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
319 /* Inode has changed type, any I/O on the old should fail */
320 make_bad_inode(inode);
321 iput(inode);
322 goto retry;
325 fi = get_fuse_inode(inode);
326 spin_lock(&fc->lock);
327 fi->nlookup++;
328 spin_unlock(&fc->lock);
329 fuse_change_attributes(inode, attr, attr_valid, attr_version);
331 return inode;
334 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
335 loff_t offset, loff_t len)
337 struct inode *inode;
338 pgoff_t pg_start;
339 pgoff_t pg_end;
341 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
342 if (!inode)
343 return -ENOENT;
345 fuse_invalidate_attr(inode);
346 forget_all_cached_acls(inode);
347 if (offset >= 0) {
348 pg_start = offset >> PAGE_SHIFT;
349 if (len <= 0)
350 pg_end = -1;
351 else
352 pg_end = (offset + len - 1) >> PAGE_SHIFT;
353 invalidate_inode_pages2_range(inode->i_mapping,
354 pg_start, pg_end);
356 iput(inode);
357 return 0;
360 bool fuse_lock_inode(struct inode *inode)
362 bool locked = false;
364 if (!get_fuse_conn(inode)->parallel_dirops) {
365 mutex_lock(&get_fuse_inode(inode)->mutex);
366 locked = true;
369 return locked;
372 void fuse_unlock_inode(struct inode *inode, bool locked)
374 if (locked)
375 mutex_unlock(&get_fuse_inode(inode)->mutex);
378 static void fuse_umount_begin(struct super_block *sb)
380 fuse_abort_conn(get_fuse_conn_super(sb));
383 static void fuse_send_destroy(struct fuse_conn *fc)
385 struct fuse_req *req = fc->destroy_req;
386 if (req && fc->conn_init) {
387 fc->destroy_req = NULL;
388 req->in.h.opcode = FUSE_DESTROY;
389 __set_bit(FR_FORCE, &req->flags);
390 __clear_bit(FR_BACKGROUND, &req->flags);
391 fuse_request_send(fc, req);
392 fuse_put_request(fc, req);
396 static void fuse_put_super(struct super_block *sb)
398 struct fuse_conn *fc = get_fuse_conn_super(sb);
400 mutex_lock(&fuse_mutex);
401 list_del(&fc->entry);
402 fuse_ctl_remove_conn(fc);
403 mutex_unlock(&fuse_mutex);
405 fuse_conn_put(fc);
408 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
410 stbuf->f_type = FUSE_SUPER_MAGIC;
411 stbuf->f_bsize = attr->bsize;
412 stbuf->f_frsize = attr->frsize;
413 stbuf->f_blocks = attr->blocks;
414 stbuf->f_bfree = attr->bfree;
415 stbuf->f_bavail = attr->bavail;
416 stbuf->f_files = attr->files;
417 stbuf->f_ffree = attr->ffree;
418 stbuf->f_namelen = attr->namelen;
419 /* fsid is left zero */
422 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
424 struct super_block *sb = dentry->d_sb;
425 struct fuse_conn *fc = get_fuse_conn_super(sb);
426 FUSE_ARGS(args);
427 struct fuse_statfs_out outarg;
428 int err;
430 if (!fuse_allow_current_process(fc)) {
431 buf->f_type = FUSE_SUPER_MAGIC;
432 return 0;
435 memset(&outarg, 0, sizeof(outarg));
436 args.in.numargs = 0;
437 args.in.h.opcode = FUSE_STATFS;
438 args.in.h.nodeid = get_node_id(d_inode(dentry));
439 args.out.numargs = 1;
440 args.out.args[0].size = sizeof(outarg);
441 args.out.args[0].value = &outarg;
442 err = fuse_simple_request(fc, &args);
443 if (!err)
444 convert_fuse_statfs(buf, &outarg.st);
445 return err;
448 enum {
449 OPT_FD,
450 OPT_ROOTMODE,
451 OPT_USER_ID,
452 OPT_GROUP_ID,
453 OPT_DEFAULT_PERMISSIONS,
454 OPT_ALLOW_OTHER,
455 OPT_MAX_READ,
456 OPT_BLKSIZE,
457 OPT_ERR
460 static const match_table_t tokens = {
461 {OPT_FD, "fd=%u"},
462 {OPT_ROOTMODE, "rootmode=%o"},
463 {OPT_USER_ID, "user_id=%u"},
464 {OPT_GROUP_ID, "group_id=%u"},
465 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
466 {OPT_ALLOW_OTHER, "allow_other"},
467 {OPT_MAX_READ, "max_read=%u"},
468 {OPT_BLKSIZE, "blksize=%u"},
469 {OPT_ERR, NULL}
472 static int fuse_match_uint(substring_t *s, unsigned int *res)
474 int err = -ENOMEM;
475 char *buf = match_strdup(s);
476 if (buf) {
477 err = kstrtouint(buf, 10, res);
478 kfree(buf);
480 return err;
483 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
485 char *p;
486 memset(d, 0, sizeof(struct fuse_mount_data));
487 d->max_read = ~0;
488 d->blksize = FUSE_DEFAULT_BLKSIZE;
490 while ((p = strsep(&opt, ",")) != NULL) {
491 int token;
492 int value;
493 unsigned uv;
494 substring_t args[MAX_OPT_ARGS];
495 if (!*p)
496 continue;
498 token = match_token(p, tokens, args);
499 switch (token) {
500 case OPT_FD:
501 if (match_int(&args[0], &value))
502 return 0;
503 d->fd = value;
504 d->fd_present = 1;
505 break;
507 case OPT_ROOTMODE:
508 if (match_octal(&args[0], &value))
509 return 0;
510 if (!fuse_valid_type(value))
511 return 0;
512 d->rootmode = value;
513 d->rootmode_present = 1;
514 break;
516 case OPT_USER_ID:
517 if (fuse_match_uint(&args[0], &uv))
518 return 0;
519 d->user_id = make_kuid(current_user_ns(), uv);
520 if (!uid_valid(d->user_id))
521 return 0;
522 d->user_id_present = 1;
523 break;
525 case OPT_GROUP_ID:
526 if (fuse_match_uint(&args[0], &uv))
527 return 0;
528 d->group_id = make_kgid(current_user_ns(), uv);
529 if (!gid_valid(d->group_id))
530 return 0;
531 d->group_id_present = 1;
532 break;
534 case OPT_DEFAULT_PERMISSIONS:
535 d->default_permissions = 1;
536 break;
538 case OPT_ALLOW_OTHER:
539 d->allow_other = 1;
540 break;
542 case OPT_MAX_READ:
543 if (match_int(&args[0], &value))
544 return 0;
545 d->max_read = value;
546 break;
548 case OPT_BLKSIZE:
549 if (!is_bdev || match_int(&args[0], &value))
550 return 0;
551 d->blksize = value;
552 break;
554 default:
555 return 0;
559 if (!d->fd_present || !d->rootmode_present ||
560 !d->user_id_present || !d->group_id_present)
561 return 0;
563 return 1;
566 static int fuse_show_options(struct seq_file *m, struct dentry *root)
568 struct super_block *sb = root->d_sb;
569 struct fuse_conn *fc = get_fuse_conn_super(sb);
571 seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
572 seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
573 if (fc->default_permissions)
574 seq_puts(m, ",default_permissions");
575 if (fc->allow_other)
576 seq_puts(m, ",allow_other");
577 if (fc->max_read != ~0)
578 seq_printf(m, ",max_read=%u", fc->max_read);
579 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
580 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
581 return 0;
584 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
586 memset(fiq, 0, sizeof(struct fuse_iqueue));
587 init_waitqueue_head(&fiq->waitq);
588 INIT_LIST_HEAD(&fiq->pending);
589 INIT_LIST_HEAD(&fiq->interrupts);
590 fiq->forget_list_tail = &fiq->forget_list_head;
591 fiq->connected = 1;
594 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
596 memset(fpq, 0, sizeof(struct fuse_pqueue));
597 spin_lock_init(&fpq->lock);
598 INIT_LIST_HEAD(&fpq->processing);
599 INIT_LIST_HEAD(&fpq->io);
600 fpq->connected = 1;
603 void fuse_conn_init(struct fuse_conn *fc)
605 memset(fc, 0, sizeof(*fc));
606 spin_lock_init(&fc->lock);
607 init_rwsem(&fc->killsb);
608 refcount_set(&fc->count, 1);
609 atomic_set(&fc->dev_count, 1);
610 init_waitqueue_head(&fc->blocked_waitq);
611 init_waitqueue_head(&fc->reserved_req_waitq);
612 fuse_iqueue_init(&fc->iq);
613 INIT_LIST_HEAD(&fc->bg_queue);
614 INIT_LIST_HEAD(&fc->entry);
615 INIT_LIST_HEAD(&fc->devices);
616 atomic_set(&fc->num_waiting, 0);
617 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
618 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
619 fc->khctr = 0;
620 fc->polled_files = RB_ROOT;
621 fc->blocked = 0;
622 fc->initialized = 0;
623 fc->connected = 1;
624 fc->attr_version = 1;
625 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
626 fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
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 if (fc->destroy_req)
634 fuse_request_free(fc->destroy_req);
635 put_pid_ns(fc->pid_ns);
636 fc->release(fc);
639 EXPORT_SYMBOL_GPL(fuse_conn_put);
641 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
643 refcount_inc(&fc->count);
644 return fc;
646 EXPORT_SYMBOL_GPL(fuse_conn_get);
648 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
650 struct fuse_attr attr;
651 memset(&attr, 0, sizeof(attr));
653 attr.mode = mode;
654 attr.ino = FUSE_ROOT_ID;
655 attr.nlink = 1;
656 return fuse_iget(sb, 1, 0, &attr, 0, 0);
659 struct fuse_inode_handle {
660 u64 nodeid;
661 u32 generation;
664 static struct dentry *fuse_get_dentry(struct super_block *sb,
665 struct fuse_inode_handle *handle)
667 struct fuse_conn *fc = get_fuse_conn_super(sb);
668 struct inode *inode;
669 struct dentry *entry;
670 int err = -ESTALE;
672 if (handle->nodeid == 0)
673 goto out_err;
675 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
676 if (!inode) {
677 struct fuse_entry_out outarg;
678 const struct qstr name = QSTR_INIT(".", 1);
680 if (!fc->export_support)
681 goto out_err;
683 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
684 &inode);
685 if (err && err != -ENOENT)
686 goto out_err;
687 if (err || !inode) {
688 err = -ESTALE;
689 goto out_err;
691 err = -EIO;
692 if (get_node_id(inode) != handle->nodeid)
693 goto out_iput;
695 err = -ESTALE;
696 if (inode->i_generation != handle->generation)
697 goto out_iput;
699 entry = d_obtain_alias(inode);
700 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
701 fuse_invalidate_entry_cache(entry);
703 return entry;
705 out_iput:
706 iput(inode);
707 out_err:
708 return ERR_PTR(err);
711 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
712 struct inode *parent)
714 int len = parent ? 6 : 3;
715 u64 nodeid;
716 u32 generation;
718 if (*max_len < len) {
719 *max_len = len;
720 return FILEID_INVALID;
723 nodeid = get_fuse_inode(inode)->nodeid;
724 generation = inode->i_generation;
726 fh[0] = (u32)(nodeid >> 32);
727 fh[1] = (u32)(nodeid & 0xffffffff);
728 fh[2] = generation;
730 if (parent) {
731 nodeid = get_fuse_inode(parent)->nodeid;
732 generation = parent->i_generation;
734 fh[3] = (u32)(nodeid >> 32);
735 fh[4] = (u32)(nodeid & 0xffffffff);
736 fh[5] = generation;
739 *max_len = len;
740 return parent ? 0x82 : 0x81;
743 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
744 struct fid *fid, int fh_len, int fh_type)
746 struct fuse_inode_handle handle;
748 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
749 return NULL;
751 handle.nodeid = (u64) fid->raw[0] << 32;
752 handle.nodeid |= (u64) fid->raw[1];
753 handle.generation = fid->raw[2];
754 return fuse_get_dentry(sb, &handle);
757 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
758 struct fid *fid, int fh_len, int fh_type)
760 struct fuse_inode_handle parent;
762 if (fh_type != 0x82 || fh_len < 6)
763 return NULL;
765 parent.nodeid = (u64) fid->raw[3] << 32;
766 parent.nodeid |= (u64) fid->raw[4];
767 parent.generation = fid->raw[5];
768 return fuse_get_dentry(sb, &parent);
771 static struct dentry *fuse_get_parent(struct dentry *child)
773 struct inode *child_inode = d_inode(child);
774 struct fuse_conn *fc = get_fuse_conn(child_inode);
775 struct inode *inode;
776 struct dentry *parent;
777 struct fuse_entry_out outarg;
778 const struct qstr name = QSTR_INIT("..", 2);
779 int err;
781 if (!fc->export_support)
782 return ERR_PTR(-ESTALE);
784 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
785 &name, &outarg, &inode);
786 if (err) {
787 if (err == -ENOENT)
788 return ERR_PTR(-ESTALE);
789 return ERR_PTR(err);
792 parent = d_obtain_alias(inode);
793 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
794 fuse_invalidate_entry_cache(parent);
796 return parent;
799 static const struct export_operations fuse_export_operations = {
800 .fh_to_dentry = fuse_fh_to_dentry,
801 .fh_to_parent = fuse_fh_to_parent,
802 .encode_fh = fuse_encode_fh,
803 .get_parent = fuse_get_parent,
806 static const struct super_operations fuse_super_operations = {
807 .alloc_inode = fuse_alloc_inode,
808 .destroy_inode = fuse_destroy_inode,
809 .evict_inode = fuse_evict_inode,
810 .write_inode = fuse_write_inode,
811 .drop_inode = generic_delete_inode,
812 .remount_fs = fuse_remount_fs,
813 .put_super = fuse_put_super,
814 .umount_begin = fuse_umount_begin,
815 .statfs = fuse_statfs,
816 .show_options = fuse_show_options,
819 static void sanitize_global_limit(unsigned *limit)
821 if (*limit == 0)
822 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
823 sizeof(struct fuse_req);
825 if (*limit >= 1 << 16)
826 *limit = (1 << 16) - 1;
829 static int set_global_limit(const char *val, struct kernel_param *kp)
831 int rv;
833 rv = param_set_uint(val, kp);
834 if (rv)
835 return rv;
837 sanitize_global_limit((unsigned *)kp->arg);
839 return 0;
842 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
844 int cap_sys_admin = capable(CAP_SYS_ADMIN);
846 if (arg->minor < 13)
847 return;
849 sanitize_global_limit(&max_user_bgreq);
850 sanitize_global_limit(&max_user_congthresh);
852 if (arg->max_background) {
853 fc->max_background = arg->max_background;
855 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
856 fc->max_background = max_user_bgreq;
858 if (arg->congestion_threshold) {
859 fc->congestion_threshold = arg->congestion_threshold;
861 if (!cap_sys_admin &&
862 fc->congestion_threshold > max_user_congthresh)
863 fc->congestion_threshold = max_user_congthresh;
867 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
869 struct fuse_init_out *arg = &req->misc.init_out;
871 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
872 fc->conn_error = 1;
873 else {
874 unsigned long ra_pages;
876 process_init_limits(fc, arg);
878 if (arg->minor >= 6) {
879 ra_pages = arg->max_readahead / PAGE_SIZE;
880 if (arg->flags & FUSE_ASYNC_READ)
881 fc->async_read = 1;
882 if (!(arg->flags & FUSE_POSIX_LOCKS))
883 fc->no_lock = 1;
884 if (arg->minor >= 17) {
885 if (!(arg->flags & FUSE_FLOCK_LOCKS))
886 fc->no_flock = 1;
887 } else {
888 if (!(arg->flags & FUSE_POSIX_LOCKS))
889 fc->no_flock = 1;
891 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
892 fc->atomic_o_trunc = 1;
893 if (arg->minor >= 9) {
894 /* LOOKUP has dependency on proto version */
895 if (arg->flags & FUSE_EXPORT_SUPPORT)
896 fc->export_support = 1;
898 if (arg->flags & FUSE_BIG_WRITES)
899 fc->big_writes = 1;
900 if (arg->flags & FUSE_DONT_MASK)
901 fc->dont_mask = 1;
902 if (arg->flags & FUSE_AUTO_INVAL_DATA)
903 fc->auto_inval_data = 1;
904 if (arg->flags & FUSE_DO_READDIRPLUS) {
905 fc->do_readdirplus = 1;
906 if (arg->flags & FUSE_READDIRPLUS_AUTO)
907 fc->readdirplus_auto = 1;
909 if (arg->flags & FUSE_ASYNC_DIO)
910 fc->async_dio = 1;
911 if (arg->flags & FUSE_WRITEBACK_CACHE)
912 fc->writeback_cache = 1;
913 if (arg->flags & FUSE_PARALLEL_DIROPS)
914 fc->parallel_dirops = 1;
915 if (arg->flags & FUSE_HANDLE_KILLPRIV)
916 fc->handle_killpriv = 1;
917 if (arg->time_gran && arg->time_gran <= 1000000000)
918 fc->sb->s_time_gran = arg->time_gran;
919 if ((arg->flags & FUSE_POSIX_ACL)) {
920 fc->default_permissions = 1;
921 fc->posix_acl = 1;
922 fc->sb->s_xattr = fuse_acl_xattr_handlers;
924 } else {
925 ra_pages = fc->max_read / PAGE_SIZE;
926 fc->no_lock = 1;
927 fc->no_flock = 1;
930 fc->sb->s_bdi->ra_pages =
931 min(fc->sb->s_bdi->ra_pages, ra_pages);
932 fc->minor = arg->minor;
933 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
934 fc->max_write = max_t(unsigned, 4096, fc->max_write);
935 fc->conn_init = 1;
937 fuse_set_initialized(fc);
938 wake_up_all(&fc->blocked_waitq);
941 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
943 struct fuse_init_in *arg = &req->misc.init_in;
945 arg->major = FUSE_KERNEL_VERSION;
946 arg->minor = FUSE_KERNEL_MINOR_VERSION;
947 arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
948 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
949 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
950 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
951 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
952 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
953 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
954 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL;
955 req->in.h.opcode = FUSE_INIT;
956 req->in.numargs = 1;
957 req->in.args[0].size = sizeof(*arg);
958 req->in.args[0].value = arg;
959 req->out.numargs = 1;
960 /* Variable length argument used for backward compatibility
961 with interface version < 7.5. Rest of init_out is zeroed
962 by do_get_request(), so a short reply is not a problem */
963 req->out.argvar = 1;
964 req->out.args[0].size = sizeof(struct fuse_init_out);
965 req->out.args[0].value = &req->misc.init_out;
966 req->end = process_init_reply;
967 fuse_request_send_background(fc, req);
970 static void fuse_free_conn(struct fuse_conn *fc)
972 WARN_ON(!list_empty(&fc->devices));
973 kfree_rcu(fc, rcu);
976 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
978 int err;
979 char *suffix = "";
981 if (sb->s_bdev) {
982 suffix = "-fuseblk";
984 * sb->s_bdi points to blkdev's bdi however we want to redirect
985 * it to our private bdi...
987 bdi_put(sb->s_bdi);
988 sb->s_bdi = &noop_backing_dev_info;
990 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
991 MINOR(fc->dev), suffix);
992 if (err)
993 return err;
995 sb->s_bdi->ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
996 /* fuse does it's own writeback accounting */
997 sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1000 * For a single fuse filesystem use max 1% of dirty +
1001 * writeback threshold.
1003 * This gives about 1M of write buffer for memory maps on a
1004 * machine with 1G and 10% dirty_ratio, which should be more
1005 * than enough.
1007 * Privileged users can raise it by writing to
1009 * /sys/class/bdi/<bdi>/max_ratio
1011 bdi_set_max_ratio(sb->s_bdi, 1);
1013 return 0;
1016 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1018 struct fuse_dev *fud;
1020 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1021 if (fud) {
1022 fud->fc = fuse_conn_get(fc);
1023 fuse_pqueue_init(&fud->pq);
1025 spin_lock(&fc->lock);
1026 list_add_tail(&fud->entry, &fc->devices);
1027 spin_unlock(&fc->lock);
1030 return fud;
1032 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1034 void fuse_dev_free(struct fuse_dev *fud)
1036 struct fuse_conn *fc = fud->fc;
1038 if (fc) {
1039 spin_lock(&fc->lock);
1040 list_del(&fud->entry);
1041 spin_unlock(&fc->lock);
1043 fuse_conn_put(fc);
1045 kfree(fud);
1047 EXPORT_SYMBOL_GPL(fuse_dev_free);
1049 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1051 struct fuse_dev *fud;
1052 struct fuse_conn *fc;
1053 struct inode *root;
1054 struct fuse_mount_data d;
1055 struct file *file;
1056 struct dentry *root_dentry;
1057 struct fuse_req *init_req;
1058 int err;
1059 int is_bdev = sb->s_bdev != NULL;
1061 err = -EINVAL;
1062 if (sb->s_flags & MS_MANDLOCK)
1063 goto err;
1065 sb->s_flags &= ~(MS_NOSEC | SB_I_VERSION);
1067 if (!parse_fuse_opt(data, &d, is_bdev))
1068 goto err;
1070 if (is_bdev) {
1071 #ifdef CONFIG_BLOCK
1072 err = -EINVAL;
1073 if (!sb_set_blocksize(sb, d.blksize))
1074 goto err;
1075 #endif
1076 } else {
1077 sb->s_blocksize = PAGE_SIZE;
1078 sb->s_blocksize_bits = PAGE_SHIFT;
1080 sb->s_magic = FUSE_SUPER_MAGIC;
1081 sb->s_op = &fuse_super_operations;
1082 sb->s_xattr = fuse_xattr_handlers;
1083 sb->s_maxbytes = MAX_LFS_FILESIZE;
1084 sb->s_time_gran = 1;
1085 sb->s_export_op = &fuse_export_operations;
1087 file = fget(d.fd);
1088 err = -EINVAL;
1089 if (!file)
1090 goto err;
1092 if ((file->f_op != &fuse_dev_operations) ||
1093 (file->f_cred->user_ns != &init_user_ns))
1094 goto err_fput;
1096 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1097 err = -ENOMEM;
1098 if (!fc)
1099 goto err_fput;
1101 fuse_conn_init(fc);
1102 fc->release = fuse_free_conn;
1104 fud = fuse_dev_alloc(fc);
1105 if (!fud)
1106 goto err_put_conn;
1108 fc->dev = sb->s_dev;
1109 fc->sb = sb;
1110 err = fuse_bdi_init(fc, sb);
1111 if (err)
1112 goto err_dev_free;
1114 /* Handle umasking inside the fuse code */
1115 if (sb->s_flags & MS_POSIXACL)
1116 fc->dont_mask = 1;
1117 sb->s_flags |= MS_POSIXACL;
1119 fc->default_permissions = d.default_permissions;
1120 fc->allow_other = d.allow_other;
1121 fc->user_id = d.user_id;
1122 fc->group_id = d.group_id;
1123 fc->max_read = max_t(unsigned, 4096, d.max_read);
1125 /* Used by get_root_inode() */
1126 sb->s_fs_info = fc;
1128 err = -ENOMEM;
1129 root = fuse_get_root_inode(sb, d.rootmode);
1130 sb->s_d_op = &fuse_root_dentry_operations;
1131 root_dentry = d_make_root(root);
1132 if (!root_dentry)
1133 goto err_dev_free;
1134 /* Root dentry doesn't have .d_revalidate */
1135 sb->s_d_op = &fuse_dentry_operations;
1137 init_req = fuse_request_alloc(0);
1138 if (!init_req)
1139 goto err_put_root;
1140 __set_bit(FR_BACKGROUND, &init_req->flags);
1142 if (is_bdev) {
1143 fc->destroy_req = fuse_request_alloc(0);
1144 if (!fc->destroy_req)
1145 goto err_free_init_req;
1148 mutex_lock(&fuse_mutex);
1149 err = -EINVAL;
1150 if (file->private_data)
1151 goto err_unlock;
1153 err = fuse_ctl_add_conn(fc);
1154 if (err)
1155 goto err_unlock;
1157 list_add_tail(&fc->entry, &fuse_conn_list);
1158 sb->s_root = root_dentry;
1159 file->private_data = fud;
1160 mutex_unlock(&fuse_mutex);
1162 * atomic_dec_and_test() in fput() provides the necessary
1163 * memory barrier for file->private_data to be visible on all
1164 * CPUs after this
1166 fput(file);
1168 fuse_send_init(fc, init_req);
1170 return 0;
1172 err_unlock:
1173 mutex_unlock(&fuse_mutex);
1174 err_free_init_req:
1175 fuse_request_free(init_req);
1176 err_put_root:
1177 dput(root_dentry);
1178 err_dev_free:
1179 fuse_dev_free(fud);
1180 err_put_conn:
1181 fuse_conn_put(fc);
1182 sb->s_fs_info = NULL;
1183 err_fput:
1184 fput(file);
1185 err:
1186 return err;
1189 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1190 int flags, const char *dev_name,
1191 void *raw_data)
1193 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1196 static void fuse_sb_destroy(struct super_block *sb)
1198 struct fuse_conn *fc = get_fuse_conn_super(sb);
1200 if (fc) {
1201 fuse_send_destroy(fc);
1203 fuse_abort_conn(fc);
1204 fuse_wait_aborted(fc);
1206 down_write(&fc->killsb);
1207 fc->sb = NULL;
1208 up_write(&fc->killsb);
1212 static void fuse_kill_sb_anon(struct super_block *sb)
1214 fuse_sb_destroy(sb);
1215 kill_anon_super(sb);
1218 static struct file_system_type fuse_fs_type = {
1219 .owner = THIS_MODULE,
1220 .name = "fuse",
1221 .fs_flags = FS_HAS_SUBTYPE,
1222 .mount = fuse_mount,
1223 .kill_sb = fuse_kill_sb_anon,
1225 MODULE_ALIAS_FS("fuse");
1227 #ifdef CONFIG_BLOCK
1228 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1229 int flags, const char *dev_name,
1230 void *raw_data)
1232 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1235 static void fuse_kill_sb_blk(struct super_block *sb)
1237 fuse_sb_destroy(sb);
1238 kill_block_super(sb);
1241 static struct file_system_type fuseblk_fs_type = {
1242 .owner = THIS_MODULE,
1243 .name = "fuseblk",
1244 .mount = fuse_mount_blk,
1245 .kill_sb = fuse_kill_sb_blk,
1246 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1248 MODULE_ALIAS_FS("fuseblk");
1250 static inline int register_fuseblk(void)
1252 return register_filesystem(&fuseblk_fs_type);
1255 static inline void unregister_fuseblk(void)
1257 unregister_filesystem(&fuseblk_fs_type);
1259 #else
1260 static inline int register_fuseblk(void)
1262 return 0;
1265 static inline void unregister_fuseblk(void)
1268 #endif
1270 static void fuse_inode_init_once(void *foo)
1272 struct inode *inode = foo;
1274 inode_init_once(inode);
1277 static int __init fuse_fs_init(void)
1279 int err;
1281 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1282 sizeof(struct fuse_inode), 0,
1283 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT,
1284 fuse_inode_init_once);
1285 err = -ENOMEM;
1286 if (!fuse_inode_cachep)
1287 goto out;
1289 err = register_fuseblk();
1290 if (err)
1291 goto out2;
1293 err = register_filesystem(&fuse_fs_type);
1294 if (err)
1295 goto out3;
1297 return 0;
1299 out3:
1300 unregister_fuseblk();
1301 out2:
1302 kmem_cache_destroy(fuse_inode_cachep);
1303 out:
1304 return err;
1307 static void fuse_fs_cleanup(void)
1309 unregister_filesystem(&fuse_fs_type);
1310 unregister_fuseblk();
1313 * Make sure all delayed rcu free inodes are flushed before we
1314 * destroy cache.
1316 rcu_barrier();
1317 kmem_cache_destroy(fuse_inode_cachep);
1320 static struct kobject *fuse_kobj;
1322 static int fuse_sysfs_init(void)
1324 int err;
1326 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1327 if (!fuse_kobj) {
1328 err = -ENOMEM;
1329 goto out_err;
1332 err = sysfs_create_mount_point(fuse_kobj, "connections");
1333 if (err)
1334 goto out_fuse_unregister;
1336 return 0;
1338 out_fuse_unregister:
1339 kobject_put(fuse_kobj);
1340 out_err:
1341 return err;
1344 static void fuse_sysfs_cleanup(void)
1346 sysfs_remove_mount_point(fuse_kobj, "connections");
1347 kobject_put(fuse_kobj);
1350 static int __init fuse_init(void)
1352 int res;
1354 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1355 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1357 INIT_LIST_HEAD(&fuse_conn_list);
1358 res = fuse_fs_init();
1359 if (res)
1360 goto err;
1362 res = fuse_dev_init();
1363 if (res)
1364 goto err_fs_cleanup;
1366 res = fuse_sysfs_init();
1367 if (res)
1368 goto err_dev_cleanup;
1370 res = fuse_ctl_init();
1371 if (res)
1372 goto err_sysfs_cleanup;
1374 sanitize_global_limit(&max_user_bgreq);
1375 sanitize_global_limit(&max_user_congthresh);
1377 return 0;
1379 err_sysfs_cleanup:
1380 fuse_sysfs_cleanup();
1381 err_dev_cleanup:
1382 fuse_dev_cleanup();
1383 err_fs_cleanup:
1384 fuse_fs_cleanup();
1385 err:
1386 return res;
1389 static void __exit fuse_exit(void)
1391 printk(KERN_DEBUG "fuse exit\n");
1393 fuse_ctl_cleanup();
1394 fuse_sysfs_cleanup();
1395 fuse_fs_cleanup();
1396 fuse_dev_cleanup();
1399 module_init(fuse_init);
1400 module_exit(fuse_exit);