selftests: ftrace: Introduce TMPDIR for temporary files
[linux-stable.git] / fs / fuse / inode.c
blob17141099f2e783fc6f2c10eabda326e13728b2f5
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>
25 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
26 MODULE_DESCRIPTION("Filesystem in Userspace");
27 MODULE_LICENSE("GPL");
29 static struct kmem_cache *fuse_inode_cachep;
30 struct list_head fuse_conn_list;
31 DEFINE_MUTEX(fuse_mutex);
33 static int set_global_limit(const char *val, struct kernel_param *kp);
35 unsigned max_user_bgreq;
36 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
37 &max_user_bgreq, 0644);
38 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
39 MODULE_PARM_DESC(max_user_bgreq,
40 "Global limit for the maximum number of backgrounded requests an "
41 "unprivileged user can set");
43 unsigned max_user_congthresh;
44 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
45 &max_user_congthresh, 0644);
46 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
47 MODULE_PARM_DESC(max_user_congthresh,
48 "Global limit for the maximum congestion threshold an "
49 "unprivileged user can set");
51 #define FUSE_SUPER_MAGIC 0x65735546
53 #define FUSE_DEFAULT_BLKSIZE 512
55 /** Maximum number of outstanding background requests */
56 #define FUSE_DEFAULT_MAX_BACKGROUND 12
58 /** Congestion starts at 75% of maximum */
59 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
61 struct fuse_mount_data {
62 int fd;
63 unsigned rootmode;
64 kuid_t user_id;
65 kgid_t group_id;
66 unsigned fd_present:1;
67 unsigned rootmode_present:1;
68 unsigned user_id_present:1;
69 unsigned group_id_present:1;
70 unsigned default_permissions:1;
71 unsigned allow_other:1;
72 unsigned max_read;
73 unsigned blksize;
76 struct fuse_forget_link *fuse_alloc_forget(void)
78 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
81 static struct inode *fuse_alloc_inode(struct super_block *sb)
83 struct inode *inode;
84 struct fuse_inode *fi;
86 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
87 if (!inode)
88 return NULL;
90 fi = get_fuse_inode(inode);
91 fi->i_time = 0;
92 fi->nodeid = 0;
93 fi->nlookup = 0;
94 fi->attr_version = 0;
95 fi->writectr = 0;
96 fi->orig_ino = 0;
97 fi->state = 0;
98 INIT_LIST_HEAD(&fi->write_files);
99 INIT_LIST_HEAD(&fi->queued_writes);
100 INIT_LIST_HEAD(&fi->writepages);
101 init_waitqueue_head(&fi->page_waitq);
102 mutex_init(&fi->mutex);
103 fi->forget = fuse_alloc_forget();
104 if (!fi->forget) {
105 kmem_cache_free(fuse_inode_cachep, inode);
106 return NULL;
109 return inode;
112 static void fuse_i_callback(struct rcu_head *head)
114 struct inode *inode = container_of(head, struct inode, i_rcu);
115 kmem_cache_free(fuse_inode_cachep, inode);
118 static void fuse_destroy_inode(struct inode *inode)
120 struct fuse_inode *fi = get_fuse_inode(inode);
121 BUG_ON(!list_empty(&fi->write_files));
122 BUG_ON(!list_empty(&fi->queued_writes));
123 mutex_destroy(&fi->mutex);
124 kfree(fi->forget);
125 call_rcu(&inode->i_rcu, fuse_i_callback);
128 static void fuse_evict_inode(struct inode *inode)
130 truncate_inode_pages_final(&inode->i_data);
131 clear_inode(inode);
132 if (inode->i_sb->s_flags & MS_ACTIVE) {
133 struct fuse_conn *fc = get_fuse_conn(inode);
134 struct fuse_inode *fi = get_fuse_inode(inode);
135 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
136 fi->forget = NULL;
140 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
142 sync_filesystem(sb);
143 if (*flags & MS_MANDLOCK)
144 return -EINVAL;
146 return 0;
150 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
151 * so that it will fit.
153 static ino_t fuse_squash_ino(u64 ino64)
155 ino_t ino = (ino_t) ino64;
156 if (sizeof(ino_t) < sizeof(u64))
157 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
158 return ino;
161 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
162 u64 attr_valid)
164 struct fuse_conn *fc = get_fuse_conn(inode);
165 struct fuse_inode *fi = get_fuse_inode(inode);
167 fi->attr_version = ++fc->attr_version;
168 fi->i_time = attr_valid;
170 inode->i_ino = fuse_squash_ino(attr->ino);
171 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
172 set_nlink(inode, attr->nlink);
173 inode->i_uid = make_kuid(&init_user_ns, attr->uid);
174 inode->i_gid = make_kgid(&init_user_ns, attr->gid);
175 inode->i_blocks = attr->blocks;
176 inode->i_atime.tv_sec = attr->atime;
177 inode->i_atime.tv_nsec = attr->atimensec;
178 /* mtime from server may be stale due to local buffered write */
179 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
180 inode->i_mtime.tv_sec = attr->mtime;
181 inode->i_mtime.tv_nsec = attr->mtimensec;
182 inode->i_ctime.tv_sec = attr->ctime;
183 inode->i_ctime.tv_nsec = attr->ctimensec;
186 if (attr->blksize != 0)
187 inode->i_blkbits = ilog2(attr->blksize);
188 else
189 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
192 * Don't set the sticky bit in i_mode, unless we want the VFS
193 * to check permissions. This prevents failures due to the
194 * check in may_delete().
196 fi->orig_i_mode = inode->i_mode;
197 if (!fc->default_permissions)
198 inode->i_mode &= ~S_ISVTX;
200 fi->orig_ino = attr->ino;
203 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
204 u64 attr_valid, u64 attr_version)
206 struct fuse_conn *fc = get_fuse_conn(inode);
207 struct fuse_inode *fi = get_fuse_inode(inode);
208 bool is_wb = fc->writeback_cache;
209 loff_t oldsize;
210 struct timespec old_mtime;
212 spin_lock(&fc->lock);
213 if ((attr_version != 0 && fi->attr_version > attr_version) ||
214 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
215 spin_unlock(&fc->lock);
216 return;
219 old_mtime = inode->i_mtime;
220 fuse_change_attributes_common(inode, attr, attr_valid);
222 oldsize = inode->i_size;
224 * In case of writeback_cache enabled, the cached writes beyond EOF
225 * extend local i_size without keeping userspace server in sync. So,
226 * attr->size coming from server can be stale. We cannot trust it.
228 if (!is_wb || !S_ISREG(inode->i_mode))
229 i_size_write(inode, attr->size);
230 spin_unlock(&fc->lock);
232 if (!is_wb && S_ISREG(inode->i_mode)) {
233 bool inval = false;
235 if (oldsize != attr->size) {
236 truncate_pagecache(inode, attr->size);
237 inval = true;
238 } else if (fc->auto_inval_data) {
239 struct timespec new_mtime = {
240 .tv_sec = attr->mtime,
241 .tv_nsec = attr->mtimensec,
245 * Auto inval mode also checks and invalidates if mtime
246 * has changed.
248 if (!timespec_equal(&old_mtime, &new_mtime))
249 inval = true;
252 if (inval)
253 invalidate_inode_pages2(inode->i_mapping);
257 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
259 inode->i_mode = attr->mode & S_IFMT;
260 inode->i_size = attr->size;
261 inode->i_mtime.tv_sec = attr->mtime;
262 inode->i_mtime.tv_nsec = attr->mtimensec;
263 inode->i_ctime.tv_sec = attr->ctime;
264 inode->i_ctime.tv_nsec = attr->ctimensec;
265 if (S_ISREG(inode->i_mode)) {
266 fuse_init_common(inode);
267 fuse_init_file_inode(inode);
268 } else if (S_ISDIR(inode->i_mode))
269 fuse_init_dir(inode);
270 else if (S_ISLNK(inode->i_mode))
271 fuse_init_symlink(inode);
272 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
273 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
274 fuse_init_common(inode);
275 init_special_inode(inode, inode->i_mode,
276 new_decode_dev(attr->rdev));
277 } else
278 BUG();
281 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
283 u64 nodeid = *(u64 *) _nodeidp;
284 if (get_node_id(inode) == nodeid)
285 return 1;
286 else
287 return 0;
290 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
292 u64 nodeid = *(u64 *) _nodeidp;
293 get_fuse_inode(inode)->nodeid = nodeid;
294 return 0;
297 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
298 int generation, struct fuse_attr *attr,
299 u64 attr_valid, u64 attr_version)
301 struct inode *inode;
302 struct fuse_inode *fi;
303 struct fuse_conn *fc = get_fuse_conn_super(sb);
305 retry:
306 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
307 if (!inode)
308 return NULL;
310 if ((inode->i_state & I_NEW)) {
311 inode->i_flags |= S_NOATIME;
312 if (!fc->writeback_cache || !S_ISREG(attr->mode))
313 inode->i_flags |= S_NOCMTIME;
314 inode->i_generation = generation;
315 fuse_init_inode(inode, attr);
316 unlock_new_inode(inode);
317 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
318 /* Inode has changed type, any I/O on the old should fail */
319 make_bad_inode(inode);
320 iput(inode);
321 goto retry;
324 fi = get_fuse_inode(inode);
325 spin_lock(&fc->lock);
326 fi->nlookup++;
327 spin_unlock(&fc->lock);
328 fuse_change_attributes(inode, attr, attr_valid, attr_version);
330 return inode;
333 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
334 loff_t offset, loff_t len)
336 struct inode *inode;
337 pgoff_t pg_start;
338 pgoff_t pg_end;
340 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
341 if (!inode)
342 return -ENOENT;
344 fuse_invalidate_attr(inode);
345 forget_all_cached_acls(inode);
346 if (offset >= 0) {
347 pg_start = offset >> PAGE_SHIFT;
348 if (len <= 0)
349 pg_end = -1;
350 else
351 pg_end = (offset + len - 1) >> PAGE_SHIFT;
352 invalidate_inode_pages2_range(inode->i_mapping,
353 pg_start, pg_end);
355 iput(inode);
356 return 0;
359 void fuse_lock_inode(struct inode *inode)
361 if (!get_fuse_conn(inode)->parallel_dirops)
362 mutex_lock(&get_fuse_inode(inode)->mutex);
365 void fuse_unlock_inode(struct inode *inode)
367 if (!get_fuse_conn(inode)->parallel_dirops)
368 mutex_unlock(&get_fuse_inode(inode)->mutex);
371 static void fuse_umount_begin(struct super_block *sb)
373 fuse_abort_conn(get_fuse_conn_super(sb));
376 static void fuse_send_destroy(struct fuse_conn *fc)
378 struct fuse_req *req = fc->destroy_req;
379 if (req && fc->conn_init) {
380 fc->destroy_req = NULL;
381 req->in.h.opcode = FUSE_DESTROY;
382 __set_bit(FR_FORCE, &req->flags);
383 __clear_bit(FR_BACKGROUND, &req->flags);
384 fuse_request_send(fc, req);
385 fuse_put_request(fc, req);
389 static void fuse_bdi_destroy(struct fuse_conn *fc)
391 if (fc->bdi_initialized)
392 bdi_destroy(&fc->bdi);
395 static void fuse_put_super(struct super_block *sb)
397 struct fuse_conn *fc = get_fuse_conn_super(sb);
399 fuse_send_destroy(fc);
401 fuse_abort_conn(fc);
402 mutex_lock(&fuse_mutex);
403 list_del(&fc->entry);
404 fuse_ctl_remove_conn(fc);
405 mutex_unlock(&fuse_mutex);
406 fuse_bdi_destroy(fc);
408 fuse_conn_put(fc);
411 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
413 stbuf->f_type = FUSE_SUPER_MAGIC;
414 stbuf->f_bsize = attr->bsize;
415 stbuf->f_frsize = attr->frsize;
416 stbuf->f_blocks = attr->blocks;
417 stbuf->f_bfree = attr->bfree;
418 stbuf->f_bavail = attr->bavail;
419 stbuf->f_files = attr->files;
420 stbuf->f_ffree = attr->ffree;
421 stbuf->f_namelen = attr->namelen;
422 /* fsid is left zero */
425 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
427 struct super_block *sb = dentry->d_sb;
428 struct fuse_conn *fc = get_fuse_conn_super(sb);
429 FUSE_ARGS(args);
430 struct fuse_statfs_out outarg;
431 int err;
433 if (!fuse_allow_current_process(fc)) {
434 buf->f_type = FUSE_SUPER_MAGIC;
435 return 0;
438 memset(&outarg, 0, sizeof(outarg));
439 args.in.numargs = 0;
440 args.in.h.opcode = FUSE_STATFS;
441 args.in.h.nodeid = get_node_id(d_inode(dentry));
442 args.out.numargs = 1;
443 args.out.args[0].size = sizeof(outarg);
444 args.out.args[0].value = &outarg;
445 err = fuse_simple_request(fc, &args);
446 if (!err)
447 convert_fuse_statfs(buf, &outarg.st);
448 return err;
451 enum {
452 OPT_FD,
453 OPT_ROOTMODE,
454 OPT_USER_ID,
455 OPT_GROUP_ID,
456 OPT_DEFAULT_PERMISSIONS,
457 OPT_ALLOW_OTHER,
458 OPT_MAX_READ,
459 OPT_BLKSIZE,
460 OPT_ERR
463 static const match_table_t tokens = {
464 {OPT_FD, "fd=%u"},
465 {OPT_ROOTMODE, "rootmode=%o"},
466 {OPT_USER_ID, "user_id=%u"},
467 {OPT_GROUP_ID, "group_id=%u"},
468 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
469 {OPT_ALLOW_OTHER, "allow_other"},
470 {OPT_MAX_READ, "max_read=%u"},
471 {OPT_BLKSIZE, "blksize=%u"},
472 {OPT_ERR, NULL}
475 static int fuse_match_uint(substring_t *s, unsigned int *res)
477 int err = -ENOMEM;
478 char *buf = match_strdup(s);
479 if (buf) {
480 err = kstrtouint(buf, 10, res);
481 kfree(buf);
483 return err;
486 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
488 char *p;
489 memset(d, 0, sizeof(struct fuse_mount_data));
490 d->max_read = ~0;
491 d->blksize = FUSE_DEFAULT_BLKSIZE;
493 while ((p = strsep(&opt, ",")) != NULL) {
494 int token;
495 int value;
496 unsigned uv;
497 substring_t args[MAX_OPT_ARGS];
498 if (!*p)
499 continue;
501 token = match_token(p, tokens, args);
502 switch (token) {
503 case OPT_FD:
504 if (match_int(&args[0], &value))
505 return 0;
506 d->fd = value;
507 d->fd_present = 1;
508 break;
510 case OPT_ROOTMODE:
511 if (match_octal(&args[0], &value))
512 return 0;
513 if (!fuse_valid_type(value))
514 return 0;
515 d->rootmode = value;
516 d->rootmode_present = 1;
517 break;
519 case OPT_USER_ID:
520 if (fuse_match_uint(&args[0], &uv))
521 return 0;
522 d->user_id = make_kuid(current_user_ns(), uv);
523 if (!uid_valid(d->user_id))
524 return 0;
525 d->user_id_present = 1;
526 break;
528 case OPT_GROUP_ID:
529 if (fuse_match_uint(&args[0], &uv))
530 return 0;
531 d->group_id = make_kgid(current_user_ns(), uv);
532 if (!gid_valid(d->group_id))
533 return 0;
534 d->group_id_present = 1;
535 break;
537 case OPT_DEFAULT_PERMISSIONS:
538 d->default_permissions = 1;
539 break;
541 case OPT_ALLOW_OTHER:
542 d->allow_other = 1;
543 break;
545 case OPT_MAX_READ:
546 if (match_int(&args[0], &value))
547 return 0;
548 d->max_read = value;
549 break;
551 case OPT_BLKSIZE:
552 if (!is_bdev || match_int(&args[0], &value))
553 return 0;
554 d->blksize = value;
555 break;
557 default:
558 return 0;
562 if (!d->fd_present || !d->rootmode_present ||
563 !d->user_id_present || !d->group_id_present)
564 return 0;
566 return 1;
569 static int fuse_show_options(struct seq_file *m, struct dentry *root)
571 struct super_block *sb = root->d_sb;
572 struct fuse_conn *fc = get_fuse_conn_super(sb);
574 seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
575 seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
576 if (fc->default_permissions)
577 seq_puts(m, ",default_permissions");
578 if (fc->allow_other)
579 seq_puts(m, ",allow_other");
580 if (fc->max_read != ~0)
581 seq_printf(m, ",max_read=%u", fc->max_read);
582 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
583 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
584 return 0;
587 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
589 memset(fiq, 0, sizeof(struct fuse_iqueue));
590 init_waitqueue_head(&fiq->waitq);
591 INIT_LIST_HEAD(&fiq->pending);
592 INIT_LIST_HEAD(&fiq->interrupts);
593 fiq->forget_list_tail = &fiq->forget_list_head;
594 fiq->connected = 1;
597 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
599 memset(fpq, 0, sizeof(struct fuse_pqueue));
600 spin_lock_init(&fpq->lock);
601 INIT_LIST_HEAD(&fpq->processing);
602 INIT_LIST_HEAD(&fpq->io);
603 fpq->connected = 1;
606 void fuse_conn_init(struct fuse_conn *fc)
608 memset(fc, 0, sizeof(*fc));
609 spin_lock_init(&fc->lock);
610 init_rwsem(&fc->killsb);
611 atomic_set(&fc->count, 1);
612 atomic_set(&fc->dev_count, 1);
613 init_waitqueue_head(&fc->blocked_waitq);
614 init_waitqueue_head(&fc->reserved_req_waitq);
615 fuse_iqueue_init(&fc->iq);
616 INIT_LIST_HEAD(&fc->bg_queue);
617 INIT_LIST_HEAD(&fc->entry);
618 INIT_LIST_HEAD(&fc->devices);
619 atomic_set(&fc->num_waiting, 0);
620 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
621 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
622 fc->khctr = 0;
623 fc->polled_files = RB_ROOT;
624 fc->blocked = 0;
625 fc->initialized = 0;
626 fc->connected = 1;
627 fc->attr_version = 1;
628 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
630 EXPORT_SYMBOL_GPL(fuse_conn_init);
632 void fuse_conn_put(struct fuse_conn *fc)
634 if (atomic_dec_and_test(&fc->count)) {
635 if (fc->destroy_req)
636 fuse_request_free(fc->destroy_req);
637 fc->release(fc);
640 EXPORT_SYMBOL_GPL(fuse_conn_put);
642 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
644 atomic_inc(&fc->count);
645 return fc;
647 EXPORT_SYMBOL_GPL(fuse_conn_get);
649 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
651 struct fuse_attr attr;
652 memset(&attr, 0, sizeof(attr));
654 attr.mode = mode;
655 attr.ino = FUSE_ROOT_ID;
656 attr.nlink = 1;
657 return fuse_iget(sb, 1, 0, &attr, 0, 0);
660 struct fuse_inode_handle {
661 u64 nodeid;
662 u32 generation;
665 static struct dentry *fuse_get_dentry(struct super_block *sb,
666 struct fuse_inode_handle *handle)
668 struct fuse_conn *fc = get_fuse_conn_super(sb);
669 struct inode *inode;
670 struct dentry *entry;
671 int err = -ESTALE;
673 if (handle->nodeid == 0)
674 goto out_err;
676 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
677 if (!inode) {
678 struct fuse_entry_out outarg;
679 const struct qstr name = QSTR_INIT(".", 1);
681 if (!fc->export_support)
682 goto out_err;
684 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
685 &inode);
686 if (err && err != -ENOENT)
687 goto out_err;
688 if (err || !inode) {
689 err = -ESTALE;
690 goto out_err;
692 err = -EIO;
693 if (get_node_id(inode) != handle->nodeid)
694 goto out_iput;
696 err = -ESTALE;
697 if (inode->i_generation != handle->generation)
698 goto out_iput;
700 entry = d_obtain_alias(inode);
701 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
702 fuse_invalidate_entry_cache(entry);
704 return entry;
706 out_iput:
707 iput(inode);
708 out_err:
709 return ERR_PTR(err);
712 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
713 struct inode *parent)
715 int len = parent ? 6 : 3;
716 u64 nodeid;
717 u32 generation;
719 if (*max_len < len) {
720 *max_len = len;
721 return FILEID_INVALID;
724 nodeid = get_fuse_inode(inode)->nodeid;
725 generation = inode->i_generation;
727 fh[0] = (u32)(nodeid >> 32);
728 fh[1] = (u32)(nodeid & 0xffffffff);
729 fh[2] = generation;
731 if (parent) {
732 nodeid = get_fuse_inode(parent)->nodeid;
733 generation = parent->i_generation;
735 fh[3] = (u32)(nodeid >> 32);
736 fh[4] = (u32)(nodeid & 0xffffffff);
737 fh[5] = generation;
740 *max_len = len;
741 return parent ? 0x82 : 0x81;
744 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
745 struct fid *fid, int fh_len, int fh_type)
747 struct fuse_inode_handle handle;
749 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
750 return NULL;
752 handle.nodeid = (u64) fid->raw[0] << 32;
753 handle.nodeid |= (u64) fid->raw[1];
754 handle.generation = fid->raw[2];
755 return fuse_get_dentry(sb, &handle);
758 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
759 struct fid *fid, int fh_len, int fh_type)
761 struct fuse_inode_handle parent;
763 if (fh_type != 0x82 || fh_len < 6)
764 return NULL;
766 parent.nodeid = (u64) fid->raw[3] << 32;
767 parent.nodeid |= (u64) fid->raw[4];
768 parent.generation = fid->raw[5];
769 return fuse_get_dentry(sb, &parent);
772 static struct dentry *fuse_get_parent(struct dentry *child)
774 struct inode *child_inode = d_inode(child);
775 struct fuse_conn *fc = get_fuse_conn(child_inode);
776 struct inode *inode;
777 struct dentry *parent;
778 struct fuse_entry_out outarg;
779 const struct qstr name = QSTR_INIT("..", 2);
780 int err;
782 if (!fc->export_support)
783 return ERR_PTR(-ESTALE);
785 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
786 &name, &outarg, &inode);
787 if (err) {
788 if (err == -ENOENT)
789 return ERR_PTR(-ESTALE);
790 return ERR_PTR(err);
793 parent = d_obtain_alias(inode);
794 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
795 fuse_invalidate_entry_cache(parent);
797 return parent;
800 static const struct export_operations fuse_export_operations = {
801 .fh_to_dentry = fuse_fh_to_dentry,
802 .fh_to_parent = fuse_fh_to_parent,
803 .encode_fh = fuse_encode_fh,
804 .get_parent = fuse_get_parent,
807 static const struct super_operations fuse_super_operations = {
808 .alloc_inode = fuse_alloc_inode,
809 .destroy_inode = fuse_destroy_inode,
810 .evict_inode = fuse_evict_inode,
811 .write_inode = fuse_write_inode,
812 .drop_inode = generic_delete_inode,
813 .remount_fs = fuse_remount_fs,
814 .put_super = fuse_put_super,
815 .umount_begin = fuse_umount_begin,
816 .statfs = fuse_statfs,
817 .show_options = fuse_show_options,
820 static void sanitize_global_limit(unsigned *limit)
822 if (*limit == 0)
823 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
824 sizeof(struct fuse_req);
826 if (*limit >= 1 << 16)
827 *limit = (1 << 16) - 1;
830 static int set_global_limit(const char *val, struct kernel_param *kp)
832 int rv;
834 rv = param_set_uint(val, kp);
835 if (rv)
836 return rv;
838 sanitize_global_limit((unsigned *)kp->arg);
840 return 0;
843 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
845 int cap_sys_admin = capable(CAP_SYS_ADMIN);
847 if (arg->minor < 13)
848 return;
850 sanitize_global_limit(&max_user_bgreq);
851 sanitize_global_limit(&max_user_congthresh);
853 if (arg->max_background) {
854 fc->max_background = arg->max_background;
856 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
857 fc->max_background = max_user_bgreq;
859 if (arg->congestion_threshold) {
860 fc->congestion_threshold = arg->congestion_threshold;
862 if (!cap_sys_admin &&
863 fc->congestion_threshold > max_user_congthresh)
864 fc->congestion_threshold = max_user_congthresh;
868 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
870 struct fuse_init_out *arg = &req->misc.init_out;
872 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
873 fc->conn_error = 1;
874 else {
875 unsigned long ra_pages;
877 process_init_limits(fc, arg);
879 if (arg->minor >= 6) {
880 ra_pages = arg->max_readahead / PAGE_SIZE;
881 if (arg->flags & FUSE_ASYNC_READ)
882 fc->async_read = 1;
883 if (!(arg->flags & FUSE_POSIX_LOCKS))
884 fc->no_lock = 1;
885 if (arg->minor >= 17) {
886 if (!(arg->flags & FUSE_FLOCK_LOCKS))
887 fc->no_flock = 1;
888 } else {
889 if (!(arg->flags & FUSE_POSIX_LOCKS))
890 fc->no_flock = 1;
892 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
893 fc->atomic_o_trunc = 1;
894 if (arg->minor >= 9) {
895 /* LOOKUP has dependency on proto version */
896 if (arg->flags & FUSE_EXPORT_SUPPORT)
897 fc->export_support = 1;
899 if (arg->flags & FUSE_BIG_WRITES)
900 fc->big_writes = 1;
901 if (arg->flags & FUSE_DONT_MASK)
902 fc->dont_mask = 1;
903 if (arg->flags & FUSE_AUTO_INVAL_DATA)
904 fc->auto_inval_data = 1;
905 if (arg->flags & FUSE_DO_READDIRPLUS) {
906 fc->do_readdirplus = 1;
907 if (arg->flags & FUSE_READDIRPLUS_AUTO)
908 fc->readdirplus_auto = 1;
910 if (arg->flags & FUSE_ASYNC_DIO)
911 fc->async_dio = 1;
912 if (arg->flags & FUSE_WRITEBACK_CACHE)
913 fc->writeback_cache = 1;
914 if (arg->flags & FUSE_PARALLEL_DIROPS)
915 fc->parallel_dirops = 1;
916 if (arg->flags & FUSE_HANDLE_KILLPRIV)
917 fc->handle_killpriv = 1;
918 if (arg->time_gran && arg->time_gran <= 1000000000)
919 fc->sb->s_time_gran = arg->time_gran;
920 if ((arg->flags & FUSE_POSIX_ACL)) {
921 fc->default_permissions = 1;
922 fc->posix_acl = 1;
923 fc->sb->s_xattr = fuse_acl_xattr_handlers;
925 } else {
926 ra_pages = fc->max_read / PAGE_SIZE;
927 fc->no_lock = 1;
928 fc->no_flock = 1;
931 fc->bdi.ra_pages = min(fc->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->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;
980 fc->bdi.name = "fuse";
981 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
982 /* fuse does it's own writeback accounting */
983 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
985 err = bdi_init(&fc->bdi);
986 if (err)
987 return err;
989 fc->bdi_initialized = 1;
991 if (sb->s_bdev) {
992 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
993 MAJOR(fc->dev), MINOR(fc->dev));
994 } else {
995 err = bdi_register_dev(&fc->bdi, fc->dev);
998 if (err)
999 return err;
1002 * For a single fuse filesystem use max 1% of dirty +
1003 * writeback threshold.
1005 * This gives about 1M of write buffer for memory maps on a
1006 * machine with 1G and 10% dirty_ratio, which should be more
1007 * than enough.
1009 * Privileged users can raise it by writing to
1011 * /sys/class/bdi/<bdi>/max_ratio
1013 bdi_set_max_ratio(&fc->bdi, 1);
1015 return 0;
1018 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1020 struct fuse_dev *fud;
1022 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1023 if (fud) {
1024 fud->fc = fuse_conn_get(fc);
1025 fuse_pqueue_init(&fud->pq);
1027 spin_lock(&fc->lock);
1028 list_add_tail(&fud->entry, &fc->devices);
1029 spin_unlock(&fc->lock);
1032 return fud;
1034 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1036 void fuse_dev_free(struct fuse_dev *fud)
1038 struct fuse_conn *fc = fud->fc;
1040 if (fc) {
1041 spin_lock(&fc->lock);
1042 list_del(&fud->entry);
1043 spin_unlock(&fc->lock);
1045 fuse_conn_put(fc);
1047 kfree(fud);
1049 EXPORT_SYMBOL_GPL(fuse_dev_free);
1051 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1053 struct fuse_dev *fud;
1054 struct fuse_conn *fc;
1055 struct inode *root;
1056 struct fuse_mount_data d;
1057 struct file *file;
1058 struct dentry *root_dentry;
1059 struct fuse_req *init_req;
1060 int err;
1061 int is_bdev = sb->s_bdev != NULL;
1063 err = -EINVAL;
1064 if (sb->s_flags & MS_MANDLOCK)
1065 goto err;
1067 sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
1069 if (!parse_fuse_opt(data, &d, is_bdev))
1070 goto err;
1072 if (is_bdev) {
1073 #ifdef CONFIG_BLOCK
1074 err = -EINVAL;
1075 if (!sb_set_blocksize(sb, d.blksize))
1076 goto err;
1077 #endif
1078 } else {
1079 sb->s_blocksize = PAGE_SIZE;
1080 sb->s_blocksize_bits = PAGE_SHIFT;
1082 sb->s_magic = FUSE_SUPER_MAGIC;
1083 sb->s_op = &fuse_super_operations;
1084 sb->s_xattr = fuse_xattr_handlers;
1085 sb->s_maxbytes = MAX_LFS_FILESIZE;
1086 sb->s_time_gran = 1;
1087 sb->s_export_op = &fuse_export_operations;
1089 file = fget(d.fd);
1090 err = -EINVAL;
1091 if (!file)
1092 goto err;
1094 if ((file->f_op != &fuse_dev_operations) ||
1095 (file->f_cred->user_ns != &init_user_ns))
1096 goto err_fput;
1098 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1099 err = -ENOMEM;
1100 if (!fc)
1101 goto err_fput;
1103 fuse_conn_init(fc);
1104 fc->release = fuse_free_conn;
1106 fud = fuse_dev_alloc(fc);
1107 if (!fud)
1108 goto err_put_conn;
1110 fc->dev = sb->s_dev;
1111 fc->sb = sb;
1112 err = fuse_bdi_init(fc, sb);
1113 if (err)
1114 goto err_dev_free;
1116 sb->s_bdi = &fc->bdi;
1118 /* Handle umasking inside the fuse code */
1119 if (sb->s_flags & MS_POSIXACL)
1120 fc->dont_mask = 1;
1121 sb->s_flags |= MS_POSIXACL;
1123 fc->default_permissions = d.default_permissions;
1124 fc->allow_other = d.allow_other;
1125 fc->user_id = d.user_id;
1126 fc->group_id = d.group_id;
1127 fc->max_read = max_t(unsigned, 4096, d.max_read);
1129 /* Used by get_root_inode() */
1130 sb->s_fs_info = fc;
1132 err = -ENOMEM;
1133 root = fuse_get_root_inode(sb, d.rootmode);
1134 root_dentry = d_make_root(root);
1135 if (!root_dentry)
1136 goto err_dev_free;
1137 /* only now - we want root dentry with NULL ->d_op */
1138 sb->s_d_op = &fuse_dentry_operations;
1140 init_req = fuse_request_alloc(0);
1141 if (!init_req)
1142 goto err_put_root;
1143 __set_bit(FR_BACKGROUND, &init_req->flags);
1145 if (is_bdev) {
1146 fc->destroy_req = fuse_request_alloc(0);
1147 if (!fc->destroy_req)
1148 goto err_free_init_req;
1151 mutex_lock(&fuse_mutex);
1152 err = -EINVAL;
1153 if (file->private_data)
1154 goto err_unlock;
1156 err = fuse_ctl_add_conn(fc);
1157 if (err)
1158 goto err_unlock;
1160 list_add_tail(&fc->entry, &fuse_conn_list);
1161 sb->s_root = root_dentry;
1162 file->private_data = fud;
1163 mutex_unlock(&fuse_mutex);
1165 * atomic_dec_and_test() in fput() provides the necessary
1166 * memory barrier for file->private_data to be visible on all
1167 * CPUs after this
1169 fput(file);
1171 fuse_send_init(fc, init_req);
1173 return 0;
1175 err_unlock:
1176 mutex_unlock(&fuse_mutex);
1177 err_free_init_req:
1178 fuse_request_free(init_req);
1179 err_put_root:
1180 dput(root_dentry);
1181 err_dev_free:
1182 fuse_dev_free(fud);
1183 err_put_conn:
1184 fuse_bdi_destroy(fc);
1185 fuse_conn_put(fc);
1186 err_fput:
1187 fput(file);
1188 err:
1189 return err;
1192 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1193 int flags, const char *dev_name,
1194 void *raw_data)
1196 return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1199 static void fuse_kill_sb_anon(struct super_block *sb)
1201 struct fuse_conn *fc = get_fuse_conn_super(sb);
1203 if (fc) {
1204 down_write(&fc->killsb);
1205 fc->sb = NULL;
1206 up_write(&fc->killsb);
1209 kill_anon_super(sb);
1212 static struct file_system_type fuse_fs_type = {
1213 .owner = THIS_MODULE,
1214 .name = "fuse",
1215 .fs_flags = FS_HAS_SUBTYPE,
1216 .mount = fuse_mount,
1217 .kill_sb = fuse_kill_sb_anon,
1219 MODULE_ALIAS_FS("fuse");
1221 #ifdef CONFIG_BLOCK
1222 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1223 int flags, const char *dev_name,
1224 void *raw_data)
1226 return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1229 static void fuse_kill_sb_blk(struct super_block *sb)
1231 struct fuse_conn *fc = get_fuse_conn_super(sb);
1233 if (fc) {
1234 down_write(&fc->killsb);
1235 fc->sb = NULL;
1236 up_write(&fc->killsb);
1239 kill_block_super(sb);
1242 static struct file_system_type fuseblk_fs_type = {
1243 .owner = THIS_MODULE,
1244 .name = "fuseblk",
1245 .mount = fuse_mount_blk,
1246 .kill_sb = fuse_kill_sb_blk,
1247 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1249 MODULE_ALIAS_FS("fuseblk");
1251 static inline int register_fuseblk(void)
1253 return register_filesystem(&fuseblk_fs_type);
1256 static inline void unregister_fuseblk(void)
1258 unregister_filesystem(&fuseblk_fs_type);
1260 #else
1261 static inline int register_fuseblk(void)
1263 return 0;
1266 static inline void unregister_fuseblk(void)
1269 #endif
1271 static void fuse_inode_init_once(void *foo)
1273 struct inode *inode = foo;
1275 inode_init_once(inode);
1278 static int __init fuse_fs_init(void)
1280 int err;
1282 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1283 sizeof(struct fuse_inode), 0,
1284 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT,
1285 fuse_inode_init_once);
1286 err = -ENOMEM;
1287 if (!fuse_inode_cachep)
1288 goto out;
1290 err = register_fuseblk();
1291 if (err)
1292 goto out2;
1294 err = register_filesystem(&fuse_fs_type);
1295 if (err)
1296 goto out3;
1298 return 0;
1300 out3:
1301 unregister_fuseblk();
1302 out2:
1303 kmem_cache_destroy(fuse_inode_cachep);
1304 out:
1305 return err;
1308 static void fuse_fs_cleanup(void)
1310 unregister_filesystem(&fuse_fs_type);
1311 unregister_fuseblk();
1314 * Make sure all delayed rcu free inodes are flushed before we
1315 * destroy cache.
1317 rcu_barrier();
1318 kmem_cache_destroy(fuse_inode_cachep);
1321 static struct kobject *fuse_kobj;
1323 static int fuse_sysfs_init(void)
1325 int err;
1327 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1328 if (!fuse_kobj) {
1329 err = -ENOMEM;
1330 goto out_err;
1333 err = sysfs_create_mount_point(fuse_kobj, "connections");
1334 if (err)
1335 goto out_fuse_unregister;
1337 return 0;
1339 out_fuse_unregister:
1340 kobject_put(fuse_kobj);
1341 out_err:
1342 return err;
1345 static void fuse_sysfs_cleanup(void)
1347 sysfs_remove_mount_point(fuse_kobj, "connections");
1348 kobject_put(fuse_kobj);
1351 static int __init fuse_init(void)
1353 int res;
1355 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1356 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1358 INIT_LIST_HEAD(&fuse_conn_list);
1359 res = fuse_fs_init();
1360 if (res)
1361 goto err;
1363 res = fuse_dev_init();
1364 if (res)
1365 goto err_fs_cleanup;
1367 res = fuse_sysfs_init();
1368 if (res)
1369 goto err_dev_cleanup;
1371 res = fuse_ctl_init();
1372 if (res)
1373 goto err_sysfs_cleanup;
1375 sanitize_global_limit(&max_user_bgreq);
1376 sanitize_global_limit(&max_user_congthresh);
1378 return 0;
1380 err_sysfs_cleanup:
1381 fuse_sysfs_cleanup();
1382 err_dev_cleanup:
1383 fuse_dev_cleanup();
1384 err_fs_cleanup:
1385 fuse_fs_cleanup();
1386 err:
1387 return res;
1390 static void __exit fuse_exit(void)
1392 printk(KERN_DEBUG "fuse exit\n");
1394 fuse_ctl_cleanup();
1395 fuse_sysfs_cleanup();
1396 fuse_fs_cleanup();
1397 fuse_dev_cleanup();
1400 module_init(fuse_init);
1401 module_exit(fuse_exit);