[PATCH] fuse: scramble lock owner ID
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / fuse / inode.c
blob5ceb8bd7a189ed0967451b56d655083fe5df6467
1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2006 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>
21 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
22 MODULE_DESCRIPTION("Filesystem in Userspace");
23 MODULE_LICENSE("GPL");
25 static kmem_cache_t *fuse_inode_cachep;
26 struct list_head fuse_conn_list;
27 DEFINE_MUTEX(fuse_mutex);
29 #define FUSE_SUPER_MAGIC 0x65735546
31 struct fuse_mount_data {
32 int fd;
33 unsigned rootmode;
34 unsigned user_id;
35 unsigned group_id;
36 unsigned fd_present : 1;
37 unsigned rootmode_present : 1;
38 unsigned user_id_present : 1;
39 unsigned group_id_present : 1;
40 unsigned flags;
41 unsigned max_read;
44 static struct inode *fuse_alloc_inode(struct super_block *sb)
46 struct inode *inode;
47 struct fuse_inode *fi;
49 inode = kmem_cache_alloc(fuse_inode_cachep, SLAB_KERNEL);
50 if (!inode)
51 return NULL;
53 fi = get_fuse_inode(inode);
54 fi->i_time = jiffies - 1;
55 fi->nodeid = 0;
56 fi->nlookup = 0;
57 fi->forget_req = fuse_request_alloc();
58 if (!fi->forget_req) {
59 kmem_cache_free(fuse_inode_cachep, inode);
60 return NULL;
63 return inode;
66 static void fuse_destroy_inode(struct inode *inode)
68 struct fuse_inode *fi = get_fuse_inode(inode);
69 if (fi->forget_req)
70 fuse_request_free(fi->forget_req);
71 kmem_cache_free(fuse_inode_cachep, inode);
74 static void fuse_read_inode(struct inode *inode)
76 /* No op */
79 void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
80 unsigned long nodeid, u64 nlookup)
82 struct fuse_forget_in *inarg = &req->misc.forget_in;
83 inarg->nlookup = nlookup;
84 req->in.h.opcode = FUSE_FORGET;
85 req->in.h.nodeid = nodeid;
86 req->in.numargs = 1;
87 req->in.args[0].size = sizeof(struct fuse_forget_in);
88 req->in.args[0].value = inarg;
89 request_send_noreply(fc, req);
92 static void fuse_clear_inode(struct inode *inode)
94 if (inode->i_sb->s_flags & MS_ACTIVE) {
95 struct fuse_conn *fc = get_fuse_conn(inode);
96 struct fuse_inode *fi = get_fuse_inode(inode);
97 fuse_send_forget(fc, fi->forget_req, fi->nodeid, fi->nlookup);
98 fi->forget_req = NULL;
102 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
104 if (*flags & MS_MANDLOCK)
105 return -EINVAL;
107 return 0;
110 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr)
112 if (S_ISREG(inode->i_mode) && i_size_read(inode) != attr->size)
113 invalidate_inode_pages(inode->i_mapping);
115 inode->i_ino = attr->ino;
116 inode->i_mode = (inode->i_mode & S_IFMT) + (attr->mode & 07777);
117 inode->i_nlink = attr->nlink;
118 inode->i_uid = attr->uid;
119 inode->i_gid = attr->gid;
120 i_size_write(inode, attr->size);
121 inode->i_blksize = PAGE_CACHE_SIZE;
122 inode->i_blocks = attr->blocks;
123 inode->i_atime.tv_sec = attr->atime;
124 inode->i_atime.tv_nsec = attr->atimensec;
125 inode->i_mtime.tv_sec = attr->mtime;
126 inode->i_mtime.tv_nsec = attr->mtimensec;
127 inode->i_ctime.tv_sec = attr->ctime;
128 inode->i_ctime.tv_nsec = attr->ctimensec;
131 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
133 inode->i_mode = attr->mode & S_IFMT;
134 i_size_write(inode, attr->size);
135 if (S_ISREG(inode->i_mode)) {
136 fuse_init_common(inode);
137 fuse_init_file_inode(inode);
138 } else if (S_ISDIR(inode->i_mode))
139 fuse_init_dir(inode);
140 else if (S_ISLNK(inode->i_mode))
141 fuse_init_symlink(inode);
142 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
143 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
144 fuse_init_common(inode);
145 init_special_inode(inode, inode->i_mode,
146 new_decode_dev(attr->rdev));
147 } else
148 BUG();
151 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
153 unsigned long nodeid = *(unsigned long *) _nodeidp;
154 if (get_node_id(inode) == nodeid)
155 return 1;
156 else
157 return 0;
160 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
162 unsigned long nodeid = *(unsigned long *) _nodeidp;
163 get_fuse_inode(inode)->nodeid = nodeid;
164 return 0;
167 struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
168 int generation, struct fuse_attr *attr)
170 struct inode *inode;
171 struct fuse_inode *fi;
172 struct fuse_conn *fc = get_fuse_conn_super(sb);
173 int retried = 0;
175 retry:
176 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
177 if (!inode)
178 return NULL;
180 if ((inode->i_state & I_NEW)) {
181 inode->i_flags |= S_NOATIME|S_NOCMTIME;
182 inode->i_generation = generation;
183 inode->i_data.backing_dev_info = &fc->bdi;
184 fuse_init_inode(inode, attr);
185 unlock_new_inode(inode);
186 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
187 BUG_ON(retried);
188 /* Inode has changed type, any I/O on the old should fail */
189 make_bad_inode(inode);
190 iput(inode);
191 retried = 1;
192 goto retry;
195 fi = get_fuse_inode(inode);
196 fi->nlookup ++;
197 fuse_change_attributes(inode, attr);
198 return inode;
201 static void fuse_umount_begin(struct super_block *sb)
203 fuse_abort_conn(get_fuse_conn_super(sb));
206 static void fuse_put_super(struct super_block *sb)
208 struct fuse_conn *fc = get_fuse_conn_super(sb);
210 spin_lock(&fc->lock);
211 fc->connected = 0;
212 fc->blocked = 0;
213 spin_unlock(&fc->lock);
214 /* Flush all readers on this fs */
215 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
216 wake_up_all(&fc->waitq);
217 wake_up_all(&fc->blocked_waitq);
218 mutex_lock(&fuse_mutex);
219 list_del(&fc->entry);
220 fuse_ctl_remove_conn(fc);
221 mutex_unlock(&fuse_mutex);
222 fuse_conn_put(fc);
225 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
227 stbuf->f_type = FUSE_SUPER_MAGIC;
228 stbuf->f_bsize = attr->bsize;
229 stbuf->f_frsize = attr->frsize;
230 stbuf->f_blocks = attr->blocks;
231 stbuf->f_bfree = attr->bfree;
232 stbuf->f_bavail = attr->bavail;
233 stbuf->f_files = attr->files;
234 stbuf->f_ffree = attr->ffree;
235 stbuf->f_namelen = attr->namelen;
236 /* fsid is left zero */
239 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
241 struct super_block *sb = dentry->d_sb;
242 struct fuse_conn *fc = get_fuse_conn_super(sb);
243 struct fuse_req *req;
244 struct fuse_statfs_out outarg;
245 int err;
247 req = fuse_get_req(fc);
248 if (IS_ERR(req))
249 return PTR_ERR(req);
251 memset(&outarg, 0, sizeof(outarg));
252 req->in.numargs = 0;
253 req->in.h.opcode = FUSE_STATFS;
254 req->out.numargs = 1;
255 req->out.args[0].size =
256 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
257 req->out.args[0].value = &outarg;
258 request_send(fc, req);
259 err = req->out.h.error;
260 if (!err)
261 convert_fuse_statfs(buf, &outarg.st);
262 fuse_put_request(fc, req);
263 return err;
266 enum {
267 OPT_FD,
268 OPT_ROOTMODE,
269 OPT_USER_ID,
270 OPT_GROUP_ID,
271 OPT_DEFAULT_PERMISSIONS,
272 OPT_ALLOW_OTHER,
273 OPT_MAX_READ,
274 OPT_ERR
277 static match_table_t tokens = {
278 {OPT_FD, "fd=%u"},
279 {OPT_ROOTMODE, "rootmode=%o"},
280 {OPT_USER_ID, "user_id=%u"},
281 {OPT_GROUP_ID, "group_id=%u"},
282 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
283 {OPT_ALLOW_OTHER, "allow_other"},
284 {OPT_MAX_READ, "max_read=%u"},
285 {OPT_ERR, NULL}
288 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d)
290 char *p;
291 memset(d, 0, sizeof(struct fuse_mount_data));
292 d->max_read = ~0;
294 while ((p = strsep(&opt, ",")) != NULL) {
295 int token;
296 int value;
297 substring_t args[MAX_OPT_ARGS];
298 if (!*p)
299 continue;
301 token = match_token(p, tokens, args);
302 switch (token) {
303 case OPT_FD:
304 if (match_int(&args[0], &value))
305 return 0;
306 d->fd = value;
307 d->fd_present = 1;
308 break;
310 case OPT_ROOTMODE:
311 if (match_octal(&args[0], &value))
312 return 0;
313 d->rootmode = value;
314 d->rootmode_present = 1;
315 break;
317 case OPT_USER_ID:
318 if (match_int(&args[0], &value))
319 return 0;
320 d->user_id = value;
321 d->user_id_present = 1;
322 break;
324 case OPT_GROUP_ID:
325 if (match_int(&args[0], &value))
326 return 0;
327 d->group_id = value;
328 d->group_id_present = 1;
329 break;
331 case OPT_DEFAULT_PERMISSIONS:
332 d->flags |= FUSE_DEFAULT_PERMISSIONS;
333 break;
335 case OPT_ALLOW_OTHER:
336 d->flags |= FUSE_ALLOW_OTHER;
337 break;
339 case OPT_MAX_READ:
340 if (match_int(&args[0], &value))
341 return 0;
342 d->max_read = value;
343 break;
345 default:
346 return 0;
350 if (!d->fd_present || !d->rootmode_present ||
351 !d->user_id_present || !d->group_id_present)
352 return 0;
354 return 1;
357 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
359 struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
361 seq_printf(m, ",user_id=%u", fc->user_id);
362 seq_printf(m, ",group_id=%u", fc->group_id);
363 if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
364 seq_puts(m, ",default_permissions");
365 if (fc->flags & FUSE_ALLOW_OTHER)
366 seq_puts(m, ",allow_other");
367 if (fc->max_read != ~0)
368 seq_printf(m, ",max_read=%u", fc->max_read);
369 return 0;
372 static struct fuse_conn *new_conn(void)
374 struct fuse_conn *fc;
376 fc = kzalloc(sizeof(*fc), GFP_KERNEL);
377 if (fc) {
378 spin_lock_init(&fc->lock);
379 atomic_set(&fc->count, 1);
380 init_waitqueue_head(&fc->waitq);
381 init_waitqueue_head(&fc->blocked_waitq);
382 INIT_LIST_HEAD(&fc->pending);
383 INIT_LIST_HEAD(&fc->processing);
384 INIT_LIST_HEAD(&fc->io);
385 INIT_LIST_HEAD(&fc->interrupts);
386 atomic_set(&fc->num_waiting, 0);
387 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
388 fc->bdi.unplug_io_fn = default_unplug_io_fn;
389 fc->reqctr = 0;
390 fc->blocked = 1;
391 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
393 return fc;
396 void fuse_conn_put(struct fuse_conn *fc)
398 if (atomic_dec_and_test(&fc->count))
399 kfree(fc);
402 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
404 atomic_inc(&fc->count);
405 return fc;
408 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
410 struct fuse_attr attr;
411 memset(&attr, 0, sizeof(attr));
413 attr.mode = mode;
414 attr.ino = FUSE_ROOT_ID;
415 return fuse_iget(sb, 1, 0, &attr);
418 static struct super_operations fuse_super_operations = {
419 .alloc_inode = fuse_alloc_inode,
420 .destroy_inode = fuse_destroy_inode,
421 .read_inode = fuse_read_inode,
422 .clear_inode = fuse_clear_inode,
423 .remount_fs = fuse_remount_fs,
424 .put_super = fuse_put_super,
425 .umount_begin = fuse_umount_begin,
426 .statfs = fuse_statfs,
427 .show_options = fuse_show_options,
430 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
432 struct fuse_init_out *arg = &req->misc.init_out;
434 if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
435 fc->conn_error = 1;
436 else {
437 unsigned long ra_pages;
439 if (arg->minor >= 6) {
440 ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
441 if (arg->flags & FUSE_ASYNC_READ)
442 fc->async_read = 1;
443 if (!(arg->flags & FUSE_POSIX_LOCKS))
444 fc->no_lock = 1;
445 } else {
446 ra_pages = fc->max_read / PAGE_CACHE_SIZE;
447 fc->no_lock = 1;
450 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
451 fc->minor = arg->minor;
452 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
454 fuse_put_request(fc, req);
455 fc->blocked = 0;
456 wake_up_all(&fc->blocked_waitq);
459 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
461 struct fuse_init_in *arg = &req->misc.init_in;
463 arg->major = FUSE_KERNEL_VERSION;
464 arg->minor = FUSE_KERNEL_MINOR_VERSION;
465 arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
466 arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS;
467 req->in.h.opcode = FUSE_INIT;
468 req->in.numargs = 1;
469 req->in.args[0].size = sizeof(*arg);
470 req->in.args[0].value = arg;
471 req->out.numargs = 1;
472 /* Variable length arguement used for backward compatibility
473 with interface version < 7.5. Rest of init_out is zeroed
474 by do_get_request(), so a short reply is not a problem */
475 req->out.argvar = 1;
476 req->out.args[0].size = sizeof(struct fuse_init_out);
477 req->out.args[0].value = &req->misc.init_out;
478 req->end = process_init_reply;
479 request_send_background(fc, req);
482 static u64 conn_id(void)
484 static u64 ctr = 1;
485 return ctr++;
488 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
490 struct fuse_conn *fc;
491 struct inode *root;
492 struct fuse_mount_data d;
493 struct file *file;
494 struct dentry *root_dentry;
495 struct fuse_req *init_req;
496 int err;
498 if (sb->s_flags & MS_MANDLOCK)
499 return -EINVAL;
501 if (!parse_fuse_opt((char *) data, &d))
502 return -EINVAL;
504 sb->s_blocksize = PAGE_CACHE_SIZE;
505 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
506 sb->s_magic = FUSE_SUPER_MAGIC;
507 sb->s_op = &fuse_super_operations;
508 sb->s_maxbytes = MAX_LFS_FILESIZE;
510 file = fget(d.fd);
511 if (!file)
512 return -EINVAL;
514 if (file->f_op != &fuse_dev_operations)
515 return -EINVAL;
517 fc = new_conn();
518 if (!fc)
519 return -ENOMEM;
521 fc->flags = d.flags;
522 fc->user_id = d.user_id;
523 fc->group_id = d.group_id;
524 fc->max_read = d.max_read;
526 /* Used by get_root_inode() */
527 sb->s_fs_info = fc;
529 err = -ENOMEM;
530 root = get_root_inode(sb, d.rootmode);
531 if (!root)
532 goto err;
534 root_dentry = d_alloc_root(root);
535 if (!root_dentry) {
536 iput(root);
537 goto err;
540 init_req = fuse_request_alloc();
541 if (!init_req)
542 goto err_put_root;
544 mutex_lock(&fuse_mutex);
545 err = -EINVAL;
546 if (file->private_data)
547 goto err_unlock;
549 fc->id = conn_id();
550 err = fuse_ctl_add_conn(fc);
551 if (err)
552 goto err_unlock;
554 list_add_tail(&fc->entry, &fuse_conn_list);
555 sb->s_root = root_dentry;
556 fc->connected = 1;
557 file->private_data = fuse_conn_get(fc);
558 mutex_unlock(&fuse_mutex);
560 * atomic_dec_and_test() in fput() provides the necessary
561 * memory barrier for file->private_data to be visible on all
562 * CPUs after this
564 fput(file);
566 fuse_send_init(fc, init_req);
568 return 0;
570 err_unlock:
571 mutex_unlock(&fuse_mutex);
572 fuse_request_free(init_req);
573 err_put_root:
574 dput(root_dentry);
575 err:
576 fput(file);
577 fuse_conn_put(fc);
578 return err;
581 static int fuse_get_sb(struct file_system_type *fs_type,
582 int flags, const char *dev_name,
583 void *raw_data, struct vfsmount *mnt)
585 return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
588 static struct file_system_type fuse_fs_type = {
589 .owner = THIS_MODULE,
590 .name = "fuse",
591 .get_sb = fuse_get_sb,
592 .kill_sb = kill_anon_super,
595 static decl_subsys(fuse, NULL, NULL);
596 static decl_subsys(connections, NULL, NULL);
598 static void fuse_inode_init_once(void *foo, kmem_cache_t *cachep,
599 unsigned long flags)
601 struct inode * inode = foo;
603 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
604 SLAB_CTOR_CONSTRUCTOR)
605 inode_init_once(inode);
608 static int __init fuse_fs_init(void)
610 int err;
612 err = register_filesystem(&fuse_fs_type);
613 if (err)
614 printk("fuse: failed to register filesystem\n");
615 else {
616 fuse_inode_cachep = kmem_cache_create("fuse_inode",
617 sizeof(struct fuse_inode),
618 0, SLAB_HWCACHE_ALIGN,
619 fuse_inode_init_once, NULL);
620 if (!fuse_inode_cachep) {
621 unregister_filesystem(&fuse_fs_type);
622 err = -ENOMEM;
626 return err;
629 static void fuse_fs_cleanup(void)
631 unregister_filesystem(&fuse_fs_type);
632 kmem_cache_destroy(fuse_inode_cachep);
635 static int fuse_sysfs_init(void)
637 int err;
639 kset_set_kset_s(&fuse_subsys, fs_subsys);
640 err = subsystem_register(&fuse_subsys);
641 if (err)
642 goto out_err;
644 kset_set_kset_s(&connections_subsys, fuse_subsys);
645 err = subsystem_register(&connections_subsys);
646 if (err)
647 goto out_fuse_unregister;
649 return 0;
651 out_fuse_unregister:
652 subsystem_unregister(&fuse_subsys);
653 out_err:
654 return err;
657 static void fuse_sysfs_cleanup(void)
659 subsystem_unregister(&connections_subsys);
660 subsystem_unregister(&fuse_subsys);
663 static int __init fuse_init(void)
665 int res;
667 printk("fuse init (API version %i.%i)\n",
668 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
670 INIT_LIST_HEAD(&fuse_conn_list);
671 res = fuse_fs_init();
672 if (res)
673 goto err;
675 res = fuse_dev_init();
676 if (res)
677 goto err_fs_cleanup;
679 res = fuse_sysfs_init();
680 if (res)
681 goto err_dev_cleanup;
683 res = fuse_ctl_init();
684 if (res)
685 goto err_sysfs_cleanup;
687 return 0;
689 err_sysfs_cleanup:
690 fuse_sysfs_cleanup();
691 err_dev_cleanup:
692 fuse_dev_cleanup();
693 err_fs_cleanup:
694 fuse_fs_cleanup();
695 err:
696 return res;
699 static void __exit fuse_exit(void)
701 printk(KERN_DEBUG "fuse exit\n");
703 fuse_ctl_cleanup();
704 fuse_sysfs_cleanup();
705 fuse_fs_cleanup();
706 fuse_dev_cleanup();
709 module_init(fuse_init);
710 module_exit(fuse_exit);