Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6/mini2440.git] / fs / fuse / inode.c
blobf91ccc4a189de18b52450f8d847b4983ac5e4311
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/parser.h>
18 #include <linux/statfs.h>
19 #include <linux/random.h>
20 #include <linux/sched.h>
21 #include <linux/exportfs.h>
23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
24 MODULE_DESCRIPTION("Filesystem in Userspace");
25 MODULE_LICENSE("GPL");
27 static struct kmem_cache *fuse_inode_cachep;
28 struct list_head fuse_conn_list;
29 DEFINE_MUTEX(fuse_mutex);
31 #define FUSE_SUPER_MAGIC 0x65735546
33 #define FUSE_DEFAULT_BLKSIZE 512
35 struct fuse_mount_data {
36 int fd;
37 unsigned rootmode;
38 unsigned user_id;
39 unsigned group_id;
40 unsigned fd_present:1;
41 unsigned rootmode_present:1;
42 unsigned user_id_present:1;
43 unsigned group_id_present:1;
44 unsigned flags;
45 unsigned max_read;
46 unsigned blksize;
49 static struct inode *fuse_alloc_inode(struct super_block *sb)
51 struct inode *inode;
52 struct fuse_inode *fi;
54 inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
55 if (!inode)
56 return NULL;
58 fi = get_fuse_inode(inode);
59 fi->i_time = 0;
60 fi->nodeid = 0;
61 fi->nlookup = 0;
62 fi->attr_version = 0;
63 fi->writectr = 0;
64 INIT_LIST_HEAD(&fi->write_files);
65 INIT_LIST_HEAD(&fi->queued_writes);
66 INIT_LIST_HEAD(&fi->writepages);
67 init_waitqueue_head(&fi->page_waitq);
68 fi->forget_req = fuse_request_alloc();
69 if (!fi->forget_req) {
70 kmem_cache_free(fuse_inode_cachep, inode);
71 return NULL;
74 return inode;
77 static void fuse_destroy_inode(struct inode *inode)
79 struct fuse_inode *fi = get_fuse_inode(inode);
80 BUG_ON(!list_empty(&fi->write_files));
81 BUG_ON(!list_empty(&fi->queued_writes));
82 if (fi->forget_req)
83 fuse_request_free(fi->forget_req);
84 kmem_cache_free(fuse_inode_cachep, inode);
87 void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
88 u64 nodeid, u64 nlookup)
90 struct fuse_forget_in *inarg = &req->misc.forget_in;
91 inarg->nlookup = nlookup;
92 req->in.h.opcode = FUSE_FORGET;
93 req->in.h.nodeid = nodeid;
94 req->in.numargs = 1;
95 req->in.args[0].size = sizeof(struct fuse_forget_in);
96 req->in.args[0].value = inarg;
97 fuse_request_send_noreply(fc, req);
100 static void fuse_clear_inode(struct inode *inode)
102 if (inode->i_sb->s_flags & MS_ACTIVE) {
103 struct fuse_conn *fc = get_fuse_conn(inode);
104 struct fuse_inode *fi = get_fuse_inode(inode);
105 fuse_send_forget(fc, fi->forget_req, fi->nodeid, fi->nlookup);
106 fi->forget_req = NULL;
110 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
112 if (*flags & MS_MANDLOCK)
113 return -EINVAL;
115 return 0;
118 void fuse_truncate(struct address_space *mapping, loff_t offset)
120 /* See vmtruncate() */
121 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
122 truncate_inode_pages(mapping, offset);
123 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
126 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
127 u64 attr_valid)
129 struct fuse_conn *fc = get_fuse_conn(inode);
130 struct fuse_inode *fi = get_fuse_inode(inode);
132 fi->attr_version = ++fc->attr_version;
133 fi->i_time = attr_valid;
135 inode->i_ino = attr->ino;
136 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
137 inode->i_nlink = attr->nlink;
138 inode->i_uid = attr->uid;
139 inode->i_gid = attr->gid;
140 inode->i_blocks = attr->blocks;
141 inode->i_atime.tv_sec = attr->atime;
142 inode->i_atime.tv_nsec = attr->atimensec;
143 inode->i_mtime.tv_sec = attr->mtime;
144 inode->i_mtime.tv_nsec = attr->mtimensec;
145 inode->i_ctime.tv_sec = attr->ctime;
146 inode->i_ctime.tv_nsec = attr->ctimensec;
148 if (attr->blksize != 0)
149 inode->i_blkbits = ilog2(attr->blksize);
150 else
151 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
154 * Don't set the sticky bit in i_mode, unless we want the VFS
155 * to check permissions. This prevents failures due to the
156 * check in may_delete().
158 fi->orig_i_mode = inode->i_mode;
159 if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
160 inode->i_mode &= ~S_ISVTX;
163 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
164 u64 attr_valid, u64 attr_version)
166 struct fuse_conn *fc = get_fuse_conn(inode);
167 struct fuse_inode *fi = get_fuse_inode(inode);
168 loff_t oldsize;
170 spin_lock(&fc->lock);
171 if (attr_version != 0 && fi->attr_version > attr_version) {
172 spin_unlock(&fc->lock);
173 return;
176 fuse_change_attributes_common(inode, attr, attr_valid);
178 oldsize = inode->i_size;
179 i_size_write(inode, attr->size);
180 spin_unlock(&fc->lock);
182 if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
183 if (attr->size < oldsize)
184 fuse_truncate(inode->i_mapping, attr->size);
185 invalidate_inode_pages2(inode->i_mapping);
189 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
191 inode->i_mode = attr->mode & S_IFMT;
192 inode->i_size = attr->size;
193 if (S_ISREG(inode->i_mode)) {
194 fuse_init_common(inode);
195 fuse_init_file_inode(inode);
196 } else if (S_ISDIR(inode->i_mode))
197 fuse_init_dir(inode);
198 else if (S_ISLNK(inode->i_mode))
199 fuse_init_symlink(inode);
200 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
201 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
202 fuse_init_common(inode);
203 init_special_inode(inode, inode->i_mode,
204 new_decode_dev(attr->rdev));
205 } else
206 BUG();
209 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
211 u64 nodeid = *(u64 *) _nodeidp;
212 if (get_node_id(inode) == nodeid)
213 return 1;
214 else
215 return 0;
218 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
220 u64 nodeid = *(u64 *) _nodeidp;
221 get_fuse_inode(inode)->nodeid = nodeid;
222 return 0;
225 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
226 int generation, struct fuse_attr *attr,
227 u64 attr_valid, u64 attr_version)
229 struct inode *inode;
230 struct fuse_inode *fi;
231 struct fuse_conn *fc = get_fuse_conn_super(sb);
233 retry:
234 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
235 if (!inode)
236 return NULL;
238 if ((inode->i_state & I_NEW)) {
239 inode->i_flags |= S_NOATIME|S_NOCMTIME;
240 inode->i_generation = generation;
241 inode->i_data.backing_dev_info = &fc->bdi;
242 fuse_init_inode(inode, attr);
243 unlock_new_inode(inode);
244 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
245 /* Inode has changed type, any I/O on the old should fail */
246 make_bad_inode(inode);
247 iput(inode);
248 goto retry;
251 fi = get_fuse_inode(inode);
252 spin_lock(&fc->lock);
253 fi->nlookup++;
254 spin_unlock(&fc->lock);
255 fuse_change_attributes(inode, attr, attr_valid, attr_version);
257 return inode;
260 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
261 loff_t offset, loff_t len)
263 struct inode *inode;
264 pgoff_t pg_start;
265 pgoff_t pg_end;
267 inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
268 if (!inode)
269 return -ENOENT;
271 fuse_invalidate_attr(inode);
272 if (offset >= 0) {
273 pg_start = offset >> PAGE_CACHE_SHIFT;
274 if (len <= 0)
275 pg_end = -1;
276 else
277 pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
278 invalidate_inode_pages2_range(inode->i_mapping,
279 pg_start, pg_end);
281 iput(inode);
282 return 0;
285 static void fuse_umount_begin(struct super_block *sb)
287 fuse_abort_conn(get_fuse_conn_super(sb));
290 static void fuse_send_destroy(struct fuse_conn *fc)
292 struct fuse_req *req = fc->destroy_req;
293 if (req && fc->conn_init) {
294 fc->destroy_req = NULL;
295 req->in.h.opcode = FUSE_DESTROY;
296 req->force = 1;
297 fuse_request_send(fc, req);
298 fuse_put_request(fc, req);
302 static void fuse_bdi_destroy(struct fuse_conn *fc)
304 if (fc->bdi_initialized)
305 bdi_destroy(&fc->bdi);
308 void fuse_conn_kill(struct fuse_conn *fc)
310 spin_lock(&fc->lock);
311 fc->connected = 0;
312 fc->blocked = 0;
313 spin_unlock(&fc->lock);
314 /* Flush all readers on this fs */
315 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
316 wake_up_all(&fc->waitq);
317 wake_up_all(&fc->blocked_waitq);
318 wake_up_all(&fc->reserved_req_waitq);
319 mutex_lock(&fuse_mutex);
320 list_del(&fc->entry);
321 fuse_ctl_remove_conn(fc);
322 mutex_unlock(&fuse_mutex);
323 fuse_bdi_destroy(fc);
325 EXPORT_SYMBOL_GPL(fuse_conn_kill);
327 static void fuse_put_super(struct super_block *sb)
329 struct fuse_conn *fc = get_fuse_conn_super(sb);
331 fuse_send_destroy(fc);
332 fuse_conn_kill(fc);
333 fuse_conn_put(fc);
336 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
338 stbuf->f_type = FUSE_SUPER_MAGIC;
339 stbuf->f_bsize = attr->bsize;
340 stbuf->f_frsize = attr->frsize;
341 stbuf->f_blocks = attr->blocks;
342 stbuf->f_bfree = attr->bfree;
343 stbuf->f_bavail = attr->bavail;
344 stbuf->f_files = attr->files;
345 stbuf->f_ffree = attr->ffree;
346 stbuf->f_namelen = attr->namelen;
347 /* fsid is left zero */
350 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
352 struct super_block *sb = dentry->d_sb;
353 struct fuse_conn *fc = get_fuse_conn_super(sb);
354 struct fuse_req *req;
355 struct fuse_statfs_out outarg;
356 int err;
358 if (!fuse_allow_task(fc, current)) {
359 buf->f_type = FUSE_SUPER_MAGIC;
360 return 0;
363 req = fuse_get_req(fc);
364 if (IS_ERR(req))
365 return PTR_ERR(req);
367 memset(&outarg, 0, sizeof(outarg));
368 req->in.numargs = 0;
369 req->in.h.opcode = FUSE_STATFS;
370 req->in.h.nodeid = get_node_id(dentry->d_inode);
371 req->out.numargs = 1;
372 req->out.args[0].size =
373 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
374 req->out.args[0].value = &outarg;
375 fuse_request_send(fc, req);
376 err = req->out.h.error;
377 if (!err)
378 convert_fuse_statfs(buf, &outarg.st);
379 fuse_put_request(fc, req);
380 return err;
383 enum {
384 OPT_FD,
385 OPT_ROOTMODE,
386 OPT_USER_ID,
387 OPT_GROUP_ID,
388 OPT_DEFAULT_PERMISSIONS,
389 OPT_ALLOW_OTHER,
390 OPT_MAX_READ,
391 OPT_BLKSIZE,
392 OPT_ERR
395 static const match_table_t tokens = {
396 {OPT_FD, "fd=%u"},
397 {OPT_ROOTMODE, "rootmode=%o"},
398 {OPT_USER_ID, "user_id=%u"},
399 {OPT_GROUP_ID, "group_id=%u"},
400 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
401 {OPT_ALLOW_OTHER, "allow_other"},
402 {OPT_MAX_READ, "max_read=%u"},
403 {OPT_BLKSIZE, "blksize=%u"},
404 {OPT_ERR, NULL}
407 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
409 char *p;
410 memset(d, 0, sizeof(struct fuse_mount_data));
411 d->max_read = ~0;
412 d->blksize = FUSE_DEFAULT_BLKSIZE;
414 while ((p = strsep(&opt, ",")) != NULL) {
415 int token;
416 int value;
417 substring_t args[MAX_OPT_ARGS];
418 if (!*p)
419 continue;
421 token = match_token(p, tokens, args);
422 switch (token) {
423 case OPT_FD:
424 if (match_int(&args[0], &value))
425 return 0;
426 d->fd = value;
427 d->fd_present = 1;
428 break;
430 case OPT_ROOTMODE:
431 if (match_octal(&args[0], &value))
432 return 0;
433 if (!fuse_valid_type(value))
434 return 0;
435 d->rootmode = value;
436 d->rootmode_present = 1;
437 break;
439 case OPT_USER_ID:
440 if (match_int(&args[0], &value))
441 return 0;
442 d->user_id = value;
443 d->user_id_present = 1;
444 break;
446 case OPT_GROUP_ID:
447 if (match_int(&args[0], &value))
448 return 0;
449 d->group_id = value;
450 d->group_id_present = 1;
451 break;
453 case OPT_DEFAULT_PERMISSIONS:
454 d->flags |= FUSE_DEFAULT_PERMISSIONS;
455 break;
457 case OPT_ALLOW_OTHER:
458 d->flags |= FUSE_ALLOW_OTHER;
459 break;
461 case OPT_MAX_READ:
462 if (match_int(&args[0], &value))
463 return 0;
464 d->max_read = value;
465 break;
467 case OPT_BLKSIZE:
468 if (!is_bdev || match_int(&args[0], &value))
469 return 0;
470 d->blksize = value;
471 break;
473 default:
474 return 0;
478 if (!d->fd_present || !d->rootmode_present ||
479 !d->user_id_present || !d->group_id_present)
480 return 0;
482 return 1;
485 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
487 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
489 seq_printf(m, ",user_id=%u", fc->user_id);
490 seq_printf(m, ",group_id=%u", fc->group_id);
491 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
492 seq_puts(m, ",default_permissions");
493 if (fc->flags & FUSE_ALLOW_OTHER)
494 seq_puts(m, ",allow_other");
495 if (fc->max_read != ~0)
496 seq_printf(m, ",max_read=%u", fc->max_read);
497 if (mnt->mnt_sb->s_bdev &&
498 mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
499 seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
500 return 0;
503 void fuse_conn_init(struct fuse_conn *fc)
505 memset(fc, 0, sizeof(*fc));
506 spin_lock_init(&fc->lock);
507 mutex_init(&fc->inst_mutex);
508 init_rwsem(&fc->killsb);
509 atomic_set(&fc->count, 1);
510 init_waitqueue_head(&fc->waitq);
511 init_waitqueue_head(&fc->blocked_waitq);
512 init_waitqueue_head(&fc->reserved_req_waitq);
513 INIT_LIST_HEAD(&fc->pending);
514 INIT_LIST_HEAD(&fc->processing);
515 INIT_LIST_HEAD(&fc->io);
516 INIT_LIST_HEAD(&fc->interrupts);
517 INIT_LIST_HEAD(&fc->bg_queue);
518 INIT_LIST_HEAD(&fc->entry);
519 atomic_set(&fc->num_waiting, 0);
520 fc->khctr = 0;
521 fc->polled_files = RB_ROOT;
522 fc->reqctr = 0;
523 fc->blocked = 1;
524 fc->attr_version = 1;
525 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
527 EXPORT_SYMBOL_GPL(fuse_conn_init);
529 void fuse_conn_put(struct fuse_conn *fc)
531 if (atomic_dec_and_test(&fc->count)) {
532 if (fc->destroy_req)
533 fuse_request_free(fc->destroy_req);
534 mutex_destroy(&fc->inst_mutex);
535 fc->release(fc);
538 EXPORT_SYMBOL_GPL(fuse_conn_put);
540 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
542 atomic_inc(&fc->count);
543 return fc;
545 EXPORT_SYMBOL_GPL(fuse_conn_get);
547 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
549 struct fuse_attr attr;
550 memset(&attr, 0, sizeof(attr));
552 attr.mode = mode;
553 attr.ino = FUSE_ROOT_ID;
554 attr.nlink = 1;
555 return fuse_iget(sb, 1, 0, &attr, 0, 0);
558 struct fuse_inode_handle {
559 u64 nodeid;
560 u32 generation;
563 static struct dentry *fuse_get_dentry(struct super_block *sb,
564 struct fuse_inode_handle *handle)
566 struct fuse_conn *fc = get_fuse_conn_super(sb);
567 struct inode *inode;
568 struct dentry *entry;
569 int err = -ESTALE;
571 if (handle->nodeid == 0)
572 goto out_err;
574 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
575 if (!inode) {
576 struct fuse_entry_out outarg;
577 struct qstr name;
579 if (!fc->export_support)
580 goto out_err;
582 name.len = 1;
583 name.name = ".";
584 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
585 &inode);
586 if (err && err != -ENOENT)
587 goto out_err;
588 if (err || !inode) {
589 err = -ESTALE;
590 goto out_err;
592 err = -EIO;
593 if (get_node_id(inode) != handle->nodeid)
594 goto out_iput;
596 err = -ESTALE;
597 if (inode->i_generation != handle->generation)
598 goto out_iput;
600 entry = d_obtain_alias(inode);
601 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) {
602 entry->d_op = &fuse_dentry_operations;
603 fuse_invalidate_entry_cache(entry);
606 return entry;
608 out_iput:
609 iput(inode);
610 out_err:
611 return ERR_PTR(err);
614 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
615 int connectable)
617 struct inode *inode = dentry->d_inode;
618 bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
619 int len = encode_parent ? 6 : 3;
620 u64 nodeid;
621 u32 generation;
623 if (*max_len < len)
624 return 255;
626 nodeid = get_fuse_inode(inode)->nodeid;
627 generation = inode->i_generation;
629 fh[0] = (u32)(nodeid >> 32);
630 fh[1] = (u32)(nodeid & 0xffffffff);
631 fh[2] = generation;
633 if (encode_parent) {
634 struct inode *parent;
636 spin_lock(&dentry->d_lock);
637 parent = dentry->d_parent->d_inode;
638 nodeid = get_fuse_inode(parent)->nodeid;
639 generation = parent->i_generation;
640 spin_unlock(&dentry->d_lock);
642 fh[3] = (u32)(nodeid >> 32);
643 fh[4] = (u32)(nodeid & 0xffffffff);
644 fh[5] = generation;
647 *max_len = len;
648 return encode_parent ? 0x82 : 0x81;
651 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
652 struct fid *fid, int fh_len, int fh_type)
654 struct fuse_inode_handle handle;
656 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
657 return NULL;
659 handle.nodeid = (u64) fid->raw[0] << 32;
660 handle.nodeid |= (u64) fid->raw[1];
661 handle.generation = fid->raw[2];
662 return fuse_get_dentry(sb, &handle);
665 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
666 struct fid *fid, int fh_len, int fh_type)
668 struct fuse_inode_handle parent;
670 if (fh_type != 0x82 || fh_len < 6)
671 return NULL;
673 parent.nodeid = (u64) fid->raw[3] << 32;
674 parent.nodeid |= (u64) fid->raw[4];
675 parent.generation = fid->raw[5];
676 return fuse_get_dentry(sb, &parent);
679 static struct dentry *fuse_get_parent(struct dentry *child)
681 struct inode *child_inode = child->d_inode;
682 struct fuse_conn *fc = get_fuse_conn(child_inode);
683 struct inode *inode;
684 struct dentry *parent;
685 struct fuse_entry_out outarg;
686 struct qstr name;
687 int err;
689 if (!fc->export_support)
690 return ERR_PTR(-ESTALE);
692 name.len = 2;
693 name.name = "..";
694 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
695 &name, &outarg, &inode);
696 if (err) {
697 if (err == -ENOENT)
698 return ERR_PTR(-ESTALE);
699 return ERR_PTR(err);
702 parent = d_obtain_alias(inode);
703 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) {
704 parent->d_op = &fuse_dentry_operations;
705 fuse_invalidate_entry_cache(parent);
708 return parent;
711 static const struct export_operations fuse_export_operations = {
712 .fh_to_dentry = fuse_fh_to_dentry,
713 .fh_to_parent = fuse_fh_to_parent,
714 .encode_fh = fuse_encode_fh,
715 .get_parent = fuse_get_parent,
718 static const struct super_operations fuse_super_operations = {
719 .alloc_inode = fuse_alloc_inode,
720 .destroy_inode = fuse_destroy_inode,
721 .clear_inode = fuse_clear_inode,
722 .drop_inode = generic_delete_inode,
723 .remount_fs = fuse_remount_fs,
724 .put_super = fuse_put_super,
725 .umount_begin = fuse_umount_begin,
726 .statfs = fuse_statfs,
727 .show_options = fuse_show_options,
730 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
732 struct fuse_init_out *arg = &req->misc.init_out;
734 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
735 fc->conn_error = 1;
736 else {
737 unsigned long ra_pages;
739 if (arg->minor >= 6) {
740 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
741 if (arg->flags & FUSE_ASYNC_READ)
742 fc->async_read = 1;
743 if (!(arg->flags & FUSE_POSIX_LOCKS))
744 fc->no_lock = 1;
745 if (arg->flags & FUSE_ATOMIC_O_TRUNC)
746 fc->atomic_o_trunc = 1;
747 if (arg->minor >= 9) {
748 /* LOOKUP has dependency on proto version */
749 if (arg->flags & FUSE_EXPORT_SUPPORT)
750 fc->export_support = 1;
752 if (arg->flags & FUSE_BIG_WRITES)
753 fc->big_writes = 1;
754 if (arg->flags & FUSE_DONT_MASK)
755 fc->dont_mask = 1;
756 } else {
757 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
758 fc->no_lock = 1;
761 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
762 fc->minor = arg->minor;
763 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
764 fc->max_write = max_t(unsigned, 4096, fc->max_write);
765 fc->conn_init = 1;
767 fc->blocked = 0;
768 wake_up_all(&fc->blocked_waitq);
771 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
773 struct fuse_init_in *arg = &req->misc.init_in;
775 arg->major = FUSE_KERNEL_VERSION;
776 arg->minor = FUSE_KERNEL_MINOR_VERSION;
777 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
778 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
779 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK;
780 req->in.h.opcode = FUSE_INIT;
781 req->in.numargs = 1;
782 req->in.args[0].size = sizeof(*arg);
783 req->in.args[0].value = arg;
784 req->out.numargs = 1;
785 /* Variable length arguement used for backward compatibility
786 with interface version < 7.5. Rest of init_out is zeroed
787 by do_get_request(), so a short reply is not a problem */
788 req->out.argvar = 1;
789 req->out.args[0].size = sizeof(struct fuse_init_out);
790 req->out.args[0].value = &req->misc.init_out;
791 req->end = process_init_reply;
792 fuse_request_send_background(fc, req);
795 static void fuse_free_conn(struct fuse_conn *fc)
797 kfree(fc);
800 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
802 int err;
804 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
805 fc->bdi.unplug_io_fn = default_unplug_io_fn;
806 /* fuse does it's own writeback accounting */
807 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
809 err = bdi_init(&fc->bdi);
810 if (err)
811 return err;
813 fc->bdi_initialized = 1;
815 if (sb->s_bdev) {
816 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
817 MAJOR(fc->dev), MINOR(fc->dev));
818 } else {
819 err = bdi_register_dev(&fc->bdi, fc->dev);
822 if (err)
823 return err;
826 * For a single fuse filesystem use max 1% of dirty +
827 * writeback threshold.
829 * This gives about 1M of write buffer for memory maps on a
830 * machine with 1G and 10% dirty_ratio, which should be more
831 * than enough.
833 * Privileged users can raise it by writing to
835 * /sys/class/bdi/<bdi>/max_ratio
837 bdi_set_max_ratio(&fc->bdi, 1);
839 return 0;
842 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
844 struct fuse_conn *fc;
845 struct inode *root;
846 struct fuse_mount_data d;
847 struct file *file;
848 struct dentry *root_dentry;
849 struct fuse_req *init_req;
850 int err;
851 int is_bdev = sb->s_bdev != NULL;
853 err = -EINVAL;
854 if (sb->s_flags & MS_MANDLOCK)
855 goto err;
857 if (!parse_fuse_opt((char *) data, &d, is_bdev))
858 goto err;
860 if (is_bdev) {
861 #ifdef CONFIG_BLOCK
862 err = -EINVAL;
863 if (!sb_set_blocksize(sb, d.blksize))
864 goto err;
865 #endif
866 } else {
867 sb->s_blocksize = PAGE_CACHE_SIZE;
868 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
870 sb->s_magic = FUSE_SUPER_MAGIC;
871 sb->s_op = &fuse_super_operations;
872 sb->s_maxbytes = MAX_LFS_FILESIZE;
873 sb->s_export_op = &fuse_export_operations;
875 file = fget(d.fd);
876 err = -EINVAL;
877 if (!file)
878 goto err;
880 if (file->f_op != &fuse_dev_operations)
881 goto err_fput;
883 fc = kmalloc(sizeof(*fc), GFP_KERNEL);
884 err = -ENOMEM;
885 if (!fc)
886 goto err_fput;
888 fuse_conn_init(fc);
890 fc->dev = sb->s_dev;
891 fc->sb = sb;
892 err = fuse_bdi_init(fc, sb);
893 if (err)
894 goto err_put_conn;
896 /* Handle umasking inside the fuse code */
897 if (sb->s_flags & MS_POSIXACL)
898 fc->dont_mask = 1;
899 sb->s_flags |= MS_POSIXACL;
901 fc->release = fuse_free_conn;
902 fc->flags = d.flags;
903 fc->user_id = d.user_id;
904 fc->group_id = d.group_id;
905 fc->max_read = max_t(unsigned, 4096, d.max_read);
907 /* Used by get_root_inode() */
908 sb->s_fs_info = fc;
910 err = -ENOMEM;
911 root = fuse_get_root_inode(sb, d.rootmode);
912 if (!root)
913 goto err_put_conn;
915 root_dentry = d_alloc_root(root);
916 if (!root_dentry) {
917 iput(root);
918 goto err_put_conn;
921 init_req = fuse_request_alloc();
922 if (!init_req)
923 goto err_put_root;
925 if (is_bdev) {
926 fc->destroy_req = fuse_request_alloc();
927 if (!fc->destroy_req)
928 goto err_free_init_req;
931 mutex_lock(&fuse_mutex);
932 err = -EINVAL;
933 if (file->private_data)
934 goto err_unlock;
936 err = fuse_ctl_add_conn(fc);
937 if (err)
938 goto err_unlock;
940 list_add_tail(&fc->entry, &fuse_conn_list);
941 sb->s_root = root_dentry;
942 fc->connected = 1;
943 file->private_data = fuse_conn_get(fc);
944 mutex_unlock(&fuse_mutex);
946 * atomic_dec_and_test() in fput() provides the necessary
947 * memory barrier for file->private_data to be visible on all
948 * CPUs after this
950 fput(file);
952 fuse_send_init(fc, init_req);
954 return 0;
956 err_unlock:
957 mutex_unlock(&fuse_mutex);
958 err_free_init_req:
959 fuse_request_free(init_req);
960 err_put_root:
961 dput(root_dentry);
962 err_put_conn:
963 fuse_bdi_destroy(fc);
964 fuse_conn_put(fc);
965 err_fput:
966 fput(file);
967 err:
968 return err;
971 static int fuse_get_sb(struct file_system_type *fs_type,
972 int flags, const char *dev_name,
973 void *raw_data, struct vfsmount *mnt)
975 return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
978 static void fuse_kill_sb_anon(struct super_block *sb)
980 struct fuse_conn *fc = get_fuse_conn_super(sb);
982 if (fc) {
983 down_write(&fc->killsb);
984 fc->sb = NULL;
985 up_write(&fc->killsb);
988 kill_anon_super(sb);
991 static struct file_system_type fuse_fs_type = {
992 .owner = THIS_MODULE,
993 .name = "fuse",
994 .fs_flags = FS_HAS_SUBTYPE,
995 .get_sb = fuse_get_sb,
996 .kill_sb = fuse_kill_sb_anon,
999 #ifdef CONFIG_BLOCK
1000 static int fuse_get_sb_blk(struct file_system_type *fs_type,
1001 int flags, const char *dev_name,
1002 void *raw_data, struct vfsmount *mnt)
1004 return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
1005 mnt);
1008 static void fuse_kill_sb_blk(struct super_block *sb)
1010 struct fuse_conn *fc = get_fuse_conn_super(sb);
1012 if (fc) {
1013 down_write(&fc->killsb);
1014 fc->sb = NULL;
1015 up_write(&fc->killsb);
1018 kill_block_super(sb);
1021 static struct file_system_type fuseblk_fs_type = {
1022 .owner = THIS_MODULE,
1023 .name = "fuseblk",
1024 .get_sb = fuse_get_sb_blk,
1025 .kill_sb = fuse_kill_sb_blk,
1026 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1029 static inline int register_fuseblk(void)
1031 return register_filesystem(&fuseblk_fs_type);
1034 static inline void unregister_fuseblk(void)
1036 unregister_filesystem(&fuseblk_fs_type);
1038 #else
1039 static inline int register_fuseblk(void)
1041 return 0;
1044 static inline void unregister_fuseblk(void)
1047 #endif
1049 static void fuse_inode_init_once(void *foo)
1051 struct inode *inode = foo;
1053 inode_init_once(inode);
1056 static int __init fuse_fs_init(void)
1058 int err;
1060 err = register_filesystem(&fuse_fs_type);
1061 if (err)
1062 goto out;
1064 err = register_fuseblk();
1065 if (err)
1066 goto out_unreg;
1068 fuse_inode_cachep = kmem_cache_create("fuse_inode",
1069 sizeof(struct fuse_inode),
1070 0, SLAB_HWCACHE_ALIGN,
1071 fuse_inode_init_once);
1072 err = -ENOMEM;
1073 if (!fuse_inode_cachep)
1074 goto out_unreg2;
1076 return 0;
1078 out_unreg2:
1079 unregister_fuseblk();
1080 out_unreg:
1081 unregister_filesystem(&fuse_fs_type);
1082 out:
1083 return err;
1086 static void fuse_fs_cleanup(void)
1088 unregister_filesystem(&fuse_fs_type);
1089 unregister_fuseblk();
1090 kmem_cache_destroy(fuse_inode_cachep);
1093 static struct kobject *fuse_kobj;
1094 static struct kobject *connections_kobj;
1096 static int fuse_sysfs_init(void)
1098 int err;
1100 fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1101 if (!fuse_kobj) {
1102 err = -ENOMEM;
1103 goto out_err;
1106 connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1107 if (!connections_kobj) {
1108 err = -ENOMEM;
1109 goto out_fuse_unregister;
1112 return 0;
1114 out_fuse_unregister:
1115 kobject_put(fuse_kobj);
1116 out_err:
1117 return err;
1120 static void fuse_sysfs_cleanup(void)
1122 kobject_put(connections_kobj);
1123 kobject_put(fuse_kobj);
1126 static int __init fuse_init(void)
1128 int res;
1130 printk(KERN_INFO "fuse init (API version %i.%i)\n",
1131 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1133 INIT_LIST_HEAD(&fuse_conn_list);
1134 res = fuse_fs_init();
1135 if (res)
1136 goto err;
1138 res = fuse_dev_init();
1139 if (res)
1140 goto err_fs_cleanup;
1142 res = fuse_sysfs_init();
1143 if (res)
1144 goto err_dev_cleanup;
1146 res = fuse_ctl_init();
1147 if (res)
1148 goto err_sysfs_cleanup;
1150 return 0;
1152 err_sysfs_cleanup:
1153 fuse_sysfs_cleanup();
1154 err_dev_cleanup:
1155 fuse_dev_cleanup();
1156 err_fs_cleanup:
1157 fuse_fs_cleanup();
1158 err:
1159 return res;
1162 static void __exit fuse_exit(void)
1164 printk(KERN_DEBUG "fuse exit\n");
1166 fuse_ctl_cleanup();
1167 fuse_sysfs_cleanup();
1168 fuse_fs_cleanup();
1169 fuse_dev_cleanup();
1172 module_init(fuse_init);
1173 module_exit(fuse_exit);