Merge branch 'rcu/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/paulmck...
[pandora-kernel.git] / fs / fuse / inode.c
1 /*
2   FUSE: Filesystem in Userspace
3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
4
5   This program can be distributed under the terms of the GNU GPL.
6   See the file COPYING.
7 */
8
9 #include "fuse_i.h"
10
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
24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
25 MODULE_DESCRIPTION("Filesystem in Userspace");
26 MODULE_LICENSE("GPL");
27
28 static struct kmem_cache *fuse_inode_cachep;
29 struct list_head fuse_conn_list;
30 DEFINE_MUTEX(fuse_mutex);
31
32 static int set_global_limit(const char *val, struct kernel_param *kp);
33
34 unsigned max_user_bgreq;
35 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
36                   &max_user_bgreq, 0644);
37 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
38 MODULE_PARM_DESC(max_user_bgreq,
39  "Global limit for the maximum number of backgrounded requests an "
40  "unprivileged user can set");
41
42 unsigned max_user_congthresh;
43 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
44                   &max_user_congthresh, 0644);
45 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
46 MODULE_PARM_DESC(max_user_congthresh,
47  "Global limit for the maximum congestion threshold an "
48  "unprivileged user can set");
49
50 #define FUSE_SUPER_MAGIC 0x65735546
51
52 #define FUSE_DEFAULT_BLKSIZE 512
53
54 /** Maximum number of outstanding background requests */
55 #define FUSE_DEFAULT_MAX_BACKGROUND 12
56
57 /** Congestion starts at 75% of maximum */
58 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
59
60 struct fuse_mount_data {
61         int fd;
62         unsigned rootmode;
63         unsigned user_id;
64         unsigned group_id;
65         unsigned fd_present:1;
66         unsigned rootmode_present:1;
67         unsigned user_id_present:1;
68         unsigned group_id_present:1;
69         unsigned flags;
70         unsigned max_read;
71         unsigned blksize;
72 };
73
74 struct fuse_forget_link *fuse_alloc_forget(void)
75 {
76         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
77 }
78
79 static struct inode *fuse_alloc_inode(struct super_block *sb)
80 {
81         struct inode *inode;
82         struct fuse_inode *fi;
83
84         inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
85         if (!inode)
86                 return NULL;
87
88         fi = get_fuse_inode(inode);
89         fi->i_time = 0;
90         fi->nodeid = 0;
91         fi->nlookup = 0;
92         fi->attr_version = 0;
93         fi->writectr = 0;
94         fi->orig_ino = 0;
95         INIT_LIST_HEAD(&fi->write_files);
96         INIT_LIST_HEAD(&fi->queued_writes);
97         INIT_LIST_HEAD(&fi->writepages);
98         init_waitqueue_head(&fi->page_waitq);
99         fi->forget = fuse_alloc_forget();
100         if (!fi->forget) {
101                 kmem_cache_free(fuse_inode_cachep, inode);
102                 return NULL;
103         }
104
105         return inode;
106 }
107
108 static void fuse_i_callback(struct rcu_head *head)
109 {
110         struct inode *inode = container_of(head, struct inode, i_rcu);
111         kmem_cache_free(fuse_inode_cachep, inode);
112 }
113
114 static void fuse_destroy_inode(struct inode *inode)
115 {
116         struct fuse_inode *fi = get_fuse_inode(inode);
117         BUG_ON(!list_empty(&fi->write_files));
118         BUG_ON(!list_empty(&fi->queued_writes));
119         kfree(fi->forget);
120         call_rcu(&inode->i_rcu, fuse_i_callback);
121 }
122
123 static void fuse_evict_inode(struct inode *inode)
124 {
125         truncate_inode_pages(&inode->i_data, 0);
126         clear_inode(inode);
127         if (inode->i_sb->s_flags & MS_ACTIVE) {
128                 struct fuse_conn *fc = get_fuse_conn(inode);
129                 struct fuse_inode *fi = get_fuse_inode(inode);
130                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
131                 fi->forget = NULL;
132         }
133 }
134
135 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
136 {
137         if (*flags & MS_MANDLOCK)
138                 return -EINVAL;
139
140         return 0;
141 }
142
143 /*
144  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
145  * so that it will fit.
146  */
147 static ino_t fuse_squash_ino(u64 ino64)
148 {
149         ino_t ino = (ino_t) ino64;
150         if (sizeof(ino_t) < sizeof(u64))
151                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
152         return ino;
153 }
154
155 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
156                                    u64 attr_valid)
157 {
158         struct fuse_conn *fc = get_fuse_conn(inode);
159         struct fuse_inode *fi = get_fuse_inode(inode);
160
161         fi->attr_version = ++fc->attr_version;
162         fi->i_time = attr_valid;
163
164         inode->i_ino     = fuse_squash_ino(attr->ino);
165         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
166         set_nlink(inode, attr->nlink);
167         inode->i_uid     = attr->uid;
168         inode->i_gid     = attr->gid;
169         inode->i_blocks  = attr->blocks;
170         inode->i_atime.tv_sec   = attr->atime;
171         inode->i_atime.tv_nsec  = attr->atimensec;
172         inode->i_mtime.tv_sec   = attr->mtime;
173         inode->i_mtime.tv_nsec  = attr->mtimensec;
174         inode->i_ctime.tv_sec   = attr->ctime;
175         inode->i_ctime.tv_nsec  = attr->ctimensec;
176
177         if (attr->blksize != 0)
178                 inode->i_blkbits = ilog2(attr->blksize);
179         else
180                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
181
182         /*
183          * Don't set the sticky bit in i_mode, unless we want the VFS
184          * to check permissions.  This prevents failures due to the
185          * check in may_delete().
186          */
187         fi->orig_i_mode = inode->i_mode;
188         if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
189                 inode->i_mode &= ~S_ISVTX;
190
191         fi->orig_ino = attr->ino;
192 }
193
194 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
195                             u64 attr_valid, u64 attr_version)
196 {
197         struct fuse_conn *fc = get_fuse_conn(inode);
198         struct fuse_inode *fi = get_fuse_inode(inode);
199         loff_t oldsize;
200
201         spin_lock(&fc->lock);
202         if (attr_version != 0 && fi->attr_version > attr_version) {
203                 spin_unlock(&fc->lock);
204                 return;
205         }
206
207         fuse_change_attributes_common(inode, attr, attr_valid);
208
209         oldsize = inode->i_size;
210         i_size_write(inode, attr->size);
211         spin_unlock(&fc->lock);
212
213         if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
214                 truncate_pagecache(inode, oldsize, attr->size);
215                 invalidate_inode_pages2(inode->i_mapping);
216         }
217 }
218
219 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
220 {
221         inode->i_mode = attr->mode & S_IFMT;
222         inode->i_size = attr->size;
223         if (S_ISREG(inode->i_mode)) {
224                 fuse_init_common(inode);
225                 fuse_init_file_inode(inode);
226         } else if (S_ISDIR(inode->i_mode))
227                 fuse_init_dir(inode);
228         else if (S_ISLNK(inode->i_mode))
229                 fuse_init_symlink(inode);
230         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
231                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
232                 fuse_init_common(inode);
233                 init_special_inode(inode, inode->i_mode,
234                                    new_decode_dev(attr->rdev));
235         } else
236                 BUG();
237 }
238
239 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
240 {
241         u64 nodeid = *(u64 *) _nodeidp;
242         if (get_node_id(inode) == nodeid)
243                 return 1;
244         else
245                 return 0;
246 }
247
248 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
249 {
250         u64 nodeid = *(u64 *) _nodeidp;
251         get_fuse_inode(inode)->nodeid = nodeid;
252         return 0;
253 }
254
255 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
256                         int generation, struct fuse_attr *attr,
257                         u64 attr_valid, u64 attr_version)
258 {
259         struct inode *inode;
260         struct fuse_inode *fi;
261         struct fuse_conn *fc = get_fuse_conn_super(sb);
262
263  retry:
264         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
265         if (!inode)
266                 return NULL;
267
268         if ((inode->i_state & I_NEW)) {
269                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
270                 inode->i_generation = generation;
271                 inode->i_data.backing_dev_info = &fc->bdi;
272                 fuse_init_inode(inode, attr);
273                 unlock_new_inode(inode);
274         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
275                 /* Inode has changed type, any I/O on the old should fail */
276                 make_bad_inode(inode);
277                 iput(inode);
278                 goto retry;
279         }
280
281         fi = get_fuse_inode(inode);
282         spin_lock(&fc->lock);
283         fi->nlookup++;
284         spin_unlock(&fc->lock);
285         fuse_change_attributes(inode, attr, attr_valid, attr_version);
286
287         return inode;
288 }
289
290 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
291                              loff_t offset, loff_t len)
292 {
293         struct inode *inode;
294         pgoff_t pg_start;
295         pgoff_t pg_end;
296
297         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
298         if (!inode)
299                 return -ENOENT;
300
301         fuse_invalidate_attr(inode);
302         if (offset >= 0) {
303                 pg_start = offset >> PAGE_CACHE_SHIFT;
304                 if (len <= 0)
305                         pg_end = -1;
306                 else
307                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
308                 invalidate_inode_pages2_range(inode->i_mapping,
309                                               pg_start, pg_end);
310         }
311         iput(inode);
312         return 0;
313 }
314
315 static void fuse_umount_begin(struct super_block *sb)
316 {
317         fuse_abort_conn(get_fuse_conn_super(sb));
318 }
319
320 static void fuse_send_destroy(struct fuse_conn *fc)
321 {
322         struct fuse_req *req = fc->destroy_req;
323         if (req && fc->conn_init) {
324                 fc->destroy_req = NULL;
325                 req->in.h.opcode = FUSE_DESTROY;
326                 req->force = 1;
327                 fuse_request_send(fc, req);
328                 fuse_put_request(fc, req);
329         }
330 }
331
332 static void fuse_bdi_destroy(struct fuse_conn *fc)
333 {
334         if (fc->bdi_initialized)
335                 bdi_destroy(&fc->bdi);
336 }
337
338 void fuse_conn_kill(struct fuse_conn *fc)
339 {
340         spin_lock(&fc->lock);
341         fc->connected = 0;
342         fc->blocked = 0;
343         spin_unlock(&fc->lock);
344         /* Flush all readers on this fs */
345         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
346         wake_up_all(&fc->waitq);
347         wake_up_all(&fc->blocked_waitq);
348         wake_up_all(&fc->reserved_req_waitq);
349         mutex_lock(&fuse_mutex);
350         list_del(&fc->entry);
351         fuse_ctl_remove_conn(fc);
352         mutex_unlock(&fuse_mutex);
353         fuse_bdi_destroy(fc);
354 }
355 EXPORT_SYMBOL_GPL(fuse_conn_kill);
356
357 static void fuse_put_super(struct super_block *sb)
358 {
359         struct fuse_conn *fc = get_fuse_conn_super(sb);
360
361         fuse_send_destroy(fc);
362         fuse_conn_kill(fc);
363         fuse_conn_put(fc);
364 }
365
366 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
367 {
368         stbuf->f_type    = FUSE_SUPER_MAGIC;
369         stbuf->f_bsize   = attr->bsize;
370         stbuf->f_frsize  = attr->frsize;
371         stbuf->f_blocks  = attr->blocks;
372         stbuf->f_bfree   = attr->bfree;
373         stbuf->f_bavail  = attr->bavail;
374         stbuf->f_files   = attr->files;
375         stbuf->f_ffree   = attr->ffree;
376         stbuf->f_namelen = attr->namelen;
377         /* fsid is left zero */
378 }
379
380 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
381 {
382         struct super_block *sb = dentry->d_sb;
383         struct fuse_conn *fc = get_fuse_conn_super(sb);
384         struct fuse_req *req;
385         struct fuse_statfs_out outarg;
386         int err;
387
388         if (!fuse_allow_task(fc, current)) {
389                 buf->f_type = FUSE_SUPER_MAGIC;
390                 return 0;
391         }
392
393         req = fuse_get_req(fc);
394         if (IS_ERR(req))
395                 return PTR_ERR(req);
396
397         memset(&outarg, 0, sizeof(outarg));
398         req->in.numargs = 0;
399         req->in.h.opcode = FUSE_STATFS;
400         req->in.h.nodeid = get_node_id(dentry->d_inode);
401         req->out.numargs = 1;
402         req->out.args[0].size =
403                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
404         req->out.args[0].value = &outarg;
405         fuse_request_send(fc, req);
406         err = req->out.h.error;
407         if (!err)
408                 convert_fuse_statfs(buf, &outarg.st);
409         fuse_put_request(fc, req);
410         return err;
411 }
412
413 enum {
414         OPT_FD,
415         OPT_ROOTMODE,
416         OPT_USER_ID,
417         OPT_GROUP_ID,
418         OPT_DEFAULT_PERMISSIONS,
419         OPT_ALLOW_OTHER,
420         OPT_MAX_READ,
421         OPT_BLKSIZE,
422         OPT_ERR
423 };
424
425 static const match_table_t tokens = {
426         {OPT_FD,                        "fd=%u"},
427         {OPT_ROOTMODE,                  "rootmode=%o"},
428         {OPT_USER_ID,                   "user_id=%u"},
429         {OPT_GROUP_ID,                  "group_id=%u"},
430         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
431         {OPT_ALLOW_OTHER,               "allow_other"},
432         {OPT_MAX_READ,                  "max_read=%u"},
433         {OPT_BLKSIZE,                   "blksize=%u"},
434         {OPT_ERR,                       NULL}
435 };
436
437 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
438 {
439         char *p;
440         memset(d, 0, sizeof(struct fuse_mount_data));
441         d->max_read = ~0;
442         d->blksize = FUSE_DEFAULT_BLKSIZE;
443
444         while ((p = strsep(&opt, ",")) != NULL) {
445                 int token;
446                 int value;
447                 substring_t args[MAX_OPT_ARGS];
448                 if (!*p)
449                         continue;
450
451                 token = match_token(p, tokens, args);
452                 switch (token) {
453                 case OPT_FD:
454                         if (match_int(&args[0], &value))
455                                 return 0;
456                         d->fd = value;
457                         d->fd_present = 1;
458                         break;
459
460                 case OPT_ROOTMODE:
461                         if (match_octal(&args[0], &value))
462                                 return 0;
463                         if (!fuse_valid_type(value))
464                                 return 0;
465                         d->rootmode = value;
466                         d->rootmode_present = 1;
467                         break;
468
469                 case OPT_USER_ID:
470                         if (match_int(&args[0], &value))
471                                 return 0;
472                         d->user_id = value;
473                         d->user_id_present = 1;
474                         break;
475
476                 case OPT_GROUP_ID:
477                         if (match_int(&args[0], &value))
478                                 return 0;
479                         d->group_id = value;
480                         d->group_id_present = 1;
481                         break;
482
483                 case OPT_DEFAULT_PERMISSIONS:
484                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
485                         break;
486
487                 case OPT_ALLOW_OTHER:
488                         d->flags |= FUSE_ALLOW_OTHER;
489                         break;
490
491                 case OPT_MAX_READ:
492                         if (match_int(&args[0], &value))
493                                 return 0;
494                         d->max_read = value;
495                         break;
496
497                 case OPT_BLKSIZE:
498                         if (!is_bdev || match_int(&args[0], &value))
499                                 return 0;
500                         d->blksize = value;
501                         break;
502
503                 default:
504                         return 0;
505                 }
506         }
507
508         if (!d->fd_present || !d->rootmode_present ||
509             !d->user_id_present || !d->group_id_present)
510                 return 0;
511
512         return 1;
513 }
514
515 static int fuse_show_options(struct seq_file *m, struct dentry *root)
516 {
517         struct super_block *sb = root->d_sb;
518         struct fuse_conn *fc = get_fuse_conn_super(sb);
519
520         seq_printf(m, ",user_id=%u", fc->user_id);
521         seq_printf(m, ",group_id=%u", fc->group_id);
522         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
523                 seq_puts(m, ",default_permissions");
524         if (fc->flags & FUSE_ALLOW_OTHER)
525                 seq_puts(m, ",allow_other");
526         if (fc->max_read != ~0)
527                 seq_printf(m, ",max_read=%u", fc->max_read);
528         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
529                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
530         return 0;
531 }
532
533 void fuse_conn_init(struct fuse_conn *fc)
534 {
535         memset(fc, 0, sizeof(*fc));
536         spin_lock_init(&fc->lock);
537         mutex_init(&fc->inst_mutex);
538         init_rwsem(&fc->killsb);
539         atomic_set(&fc->count, 1);
540         init_waitqueue_head(&fc->waitq);
541         init_waitqueue_head(&fc->blocked_waitq);
542         init_waitqueue_head(&fc->reserved_req_waitq);
543         INIT_LIST_HEAD(&fc->pending);
544         INIT_LIST_HEAD(&fc->processing);
545         INIT_LIST_HEAD(&fc->io);
546         INIT_LIST_HEAD(&fc->interrupts);
547         INIT_LIST_HEAD(&fc->bg_queue);
548         INIT_LIST_HEAD(&fc->entry);
549         fc->forget_list_tail = &fc->forget_list_head;
550         atomic_set(&fc->num_waiting, 0);
551         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
552         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
553         fc->khctr = 0;
554         fc->polled_files = RB_ROOT;
555         fc->reqctr = 0;
556         fc->blocked = 1;
557         fc->attr_version = 1;
558         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
559 }
560 EXPORT_SYMBOL_GPL(fuse_conn_init);
561
562 void fuse_conn_put(struct fuse_conn *fc)
563 {
564         if (atomic_dec_and_test(&fc->count)) {
565                 if (fc->destroy_req)
566                         fuse_request_free(fc->destroy_req);
567                 mutex_destroy(&fc->inst_mutex);
568                 fc->release(fc);
569         }
570 }
571 EXPORT_SYMBOL_GPL(fuse_conn_put);
572
573 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
574 {
575         atomic_inc(&fc->count);
576         return fc;
577 }
578 EXPORT_SYMBOL_GPL(fuse_conn_get);
579
580 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
581 {
582         struct fuse_attr attr;
583         memset(&attr, 0, sizeof(attr));
584
585         attr.mode = mode;
586         attr.ino = FUSE_ROOT_ID;
587         attr.nlink = 1;
588         return fuse_iget(sb, 1, 0, &attr, 0, 0);
589 }
590
591 struct fuse_inode_handle {
592         u64 nodeid;
593         u32 generation;
594 };
595
596 static struct dentry *fuse_get_dentry(struct super_block *sb,
597                                       struct fuse_inode_handle *handle)
598 {
599         struct fuse_conn *fc = get_fuse_conn_super(sb);
600         struct inode *inode;
601         struct dentry *entry;
602         int err = -ESTALE;
603
604         if (handle->nodeid == 0)
605                 goto out_err;
606
607         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
608         if (!inode) {
609                 struct fuse_entry_out outarg;
610                 struct qstr name;
611
612                 if (!fc->export_support)
613                         goto out_err;
614
615                 name.len = 1;
616                 name.name = ".";
617                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
618                                        &inode);
619                 if (err && err != -ENOENT)
620                         goto out_err;
621                 if (err || !inode) {
622                         err = -ESTALE;
623                         goto out_err;
624                 }
625                 err = -EIO;
626                 if (get_node_id(inode) != handle->nodeid)
627                         goto out_iput;
628         }
629         err = -ESTALE;
630         if (inode->i_generation != handle->generation)
631                 goto out_iput;
632
633         entry = d_obtain_alias(inode);
634         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
635                 fuse_invalidate_entry_cache(entry);
636
637         return entry;
638
639  out_iput:
640         iput(inode);
641  out_err:
642         return ERR_PTR(err);
643 }
644
645 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
646                            struct inode *parent)
647 {
648         int len = parent ? 6 : 3;
649         u64 nodeid;
650         u32 generation;
651
652         if (*max_len < len) {
653                 *max_len = len;
654                 return  255;
655         }
656
657         nodeid = get_fuse_inode(inode)->nodeid;
658         generation = inode->i_generation;
659
660         fh[0] = (u32)(nodeid >> 32);
661         fh[1] = (u32)(nodeid & 0xffffffff);
662         fh[2] = generation;
663
664         if (parent) {
665                 nodeid = get_fuse_inode(parent)->nodeid;
666                 generation = parent->i_generation;
667
668                 fh[3] = (u32)(nodeid >> 32);
669                 fh[4] = (u32)(nodeid & 0xffffffff);
670                 fh[5] = generation;
671         }
672
673         *max_len = len;
674         return parent ? 0x82 : 0x81;
675 }
676
677 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
678                 struct fid *fid, int fh_len, int fh_type)
679 {
680         struct fuse_inode_handle handle;
681
682         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
683                 return NULL;
684
685         handle.nodeid = (u64) fid->raw[0] << 32;
686         handle.nodeid |= (u64) fid->raw[1];
687         handle.generation = fid->raw[2];
688         return fuse_get_dentry(sb, &handle);
689 }
690
691 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
692                 struct fid *fid, int fh_len, int fh_type)
693 {
694         struct fuse_inode_handle parent;
695
696         if (fh_type != 0x82 || fh_len < 6)
697                 return NULL;
698
699         parent.nodeid = (u64) fid->raw[3] << 32;
700         parent.nodeid |= (u64) fid->raw[4];
701         parent.generation = fid->raw[5];
702         return fuse_get_dentry(sb, &parent);
703 }
704
705 static struct dentry *fuse_get_parent(struct dentry *child)
706 {
707         struct inode *child_inode = child->d_inode;
708         struct fuse_conn *fc = get_fuse_conn(child_inode);
709         struct inode *inode;
710         struct dentry *parent;
711         struct fuse_entry_out outarg;
712         struct qstr name;
713         int err;
714
715         if (!fc->export_support)
716                 return ERR_PTR(-ESTALE);
717
718         name.len = 2;
719         name.name = "..";
720         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
721                                &name, &outarg, &inode);
722         if (err) {
723                 if (err == -ENOENT)
724                         return ERR_PTR(-ESTALE);
725                 return ERR_PTR(err);
726         }
727
728         parent = d_obtain_alias(inode);
729         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
730                 fuse_invalidate_entry_cache(parent);
731
732         return parent;
733 }
734
735 static const struct export_operations fuse_export_operations = {
736         .fh_to_dentry   = fuse_fh_to_dentry,
737         .fh_to_parent   = fuse_fh_to_parent,
738         .encode_fh      = fuse_encode_fh,
739         .get_parent     = fuse_get_parent,
740 };
741
742 static const struct super_operations fuse_super_operations = {
743         .alloc_inode    = fuse_alloc_inode,
744         .destroy_inode  = fuse_destroy_inode,
745         .evict_inode    = fuse_evict_inode,
746         .drop_inode     = generic_delete_inode,
747         .remount_fs     = fuse_remount_fs,
748         .put_super      = fuse_put_super,
749         .umount_begin   = fuse_umount_begin,
750         .statfs         = fuse_statfs,
751         .show_options   = fuse_show_options,
752 };
753
754 static void sanitize_global_limit(unsigned *limit)
755 {
756         if (*limit == 0)
757                 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
758                          sizeof(struct fuse_req);
759
760         if (*limit >= 1 << 16)
761                 *limit = (1 << 16) - 1;
762 }
763
764 static int set_global_limit(const char *val, struct kernel_param *kp)
765 {
766         int rv;
767
768         rv = param_set_uint(val, kp);
769         if (rv)
770                 return rv;
771
772         sanitize_global_limit((unsigned *)kp->arg);
773
774         return 0;
775 }
776
777 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
778 {
779         int cap_sys_admin = capable(CAP_SYS_ADMIN);
780
781         if (arg->minor < 13)
782                 return;
783
784         sanitize_global_limit(&max_user_bgreq);
785         sanitize_global_limit(&max_user_congthresh);
786
787         if (arg->max_background) {
788                 fc->max_background = arg->max_background;
789
790                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
791                         fc->max_background = max_user_bgreq;
792         }
793         if (arg->congestion_threshold) {
794                 fc->congestion_threshold = arg->congestion_threshold;
795
796                 if (!cap_sys_admin &&
797                     fc->congestion_threshold > max_user_congthresh)
798                         fc->congestion_threshold = max_user_congthresh;
799         }
800 }
801
802 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
803 {
804         struct fuse_init_out *arg = &req->misc.init_out;
805
806         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
807                 fc->conn_error = 1;
808         else {
809                 unsigned long ra_pages;
810
811                 process_init_limits(fc, arg);
812
813                 if (arg->minor >= 6) {
814                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
815                         if (arg->flags & FUSE_ASYNC_READ)
816                                 fc->async_read = 1;
817                         if (!(arg->flags & FUSE_POSIX_LOCKS))
818                                 fc->no_lock = 1;
819                         if (arg->minor >= 17) {
820                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
821                                         fc->no_flock = 1;
822                         } else {
823                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
824                                         fc->no_flock = 1;
825                         }
826                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
827                                 fc->atomic_o_trunc = 1;
828                         if (arg->minor >= 9) {
829                                 /* LOOKUP has dependency on proto version */
830                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
831                                         fc->export_support = 1;
832                         }
833                         if (arg->flags & FUSE_BIG_WRITES)
834                                 fc->big_writes = 1;
835                         if (arg->flags & FUSE_DONT_MASK)
836                                 fc->dont_mask = 1;
837                 } else {
838                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
839                         fc->no_lock = 1;
840                         fc->no_flock = 1;
841                 }
842
843                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
844                 fc->minor = arg->minor;
845                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
846                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
847                 fc->conn_init = 1;
848         }
849         fc->blocked = 0;
850         wake_up_all(&fc->blocked_waitq);
851 }
852
853 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
854 {
855         struct fuse_init_in *arg = &req->misc.init_in;
856
857         arg->major = FUSE_KERNEL_VERSION;
858         arg->minor = FUSE_KERNEL_MINOR_VERSION;
859         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
860         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
861                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
862                 FUSE_FLOCK_LOCKS;
863         req->in.h.opcode = FUSE_INIT;
864         req->in.numargs = 1;
865         req->in.args[0].size = sizeof(*arg);
866         req->in.args[0].value = arg;
867         req->out.numargs = 1;
868         /* Variable length argument used for backward compatibility
869            with interface version < 7.5.  Rest of init_out is zeroed
870            by do_get_request(), so a short reply is not a problem */
871         req->out.argvar = 1;
872         req->out.args[0].size = sizeof(struct fuse_init_out);
873         req->out.args[0].value = &req->misc.init_out;
874         req->end = process_init_reply;
875         fuse_request_send_background(fc, req);
876 }
877
878 static void fuse_free_conn(struct fuse_conn *fc)
879 {
880         kfree(fc);
881 }
882
883 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
884 {
885         int err;
886
887         fc->bdi.name = "fuse";
888         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
889         /* fuse does it's own writeback accounting */
890         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
891
892         err = bdi_init(&fc->bdi);
893         if (err)
894                 return err;
895
896         fc->bdi_initialized = 1;
897
898         if (sb->s_bdev) {
899                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
900                                     MAJOR(fc->dev), MINOR(fc->dev));
901         } else {
902                 err = bdi_register_dev(&fc->bdi, fc->dev);
903         }
904
905         if (err)
906                 return err;
907
908         /*
909          * For a single fuse filesystem use max 1% of dirty +
910          * writeback threshold.
911          *
912          * This gives about 1M of write buffer for memory maps on a
913          * machine with 1G and 10% dirty_ratio, which should be more
914          * than enough.
915          *
916          * Privileged users can raise it by writing to
917          *
918          *    /sys/class/bdi/<bdi>/max_ratio
919          */
920         bdi_set_max_ratio(&fc->bdi, 1);
921
922         return 0;
923 }
924
925 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
926 {
927         struct fuse_conn *fc;
928         struct inode *root;
929         struct fuse_mount_data d;
930         struct file *file;
931         struct dentry *root_dentry;
932         struct fuse_req *init_req;
933         int err;
934         int is_bdev = sb->s_bdev != NULL;
935
936         err = -EINVAL;
937         if (sb->s_flags & MS_MANDLOCK)
938                 goto err;
939
940         sb->s_flags &= ~MS_NOSEC;
941
942         if (!parse_fuse_opt((char *) data, &d, is_bdev))
943                 goto err;
944
945         if (is_bdev) {
946 #ifdef CONFIG_BLOCK
947                 err = -EINVAL;
948                 if (!sb_set_blocksize(sb, d.blksize))
949                         goto err;
950 #endif
951         } else {
952                 sb->s_blocksize = PAGE_CACHE_SIZE;
953                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
954         }
955         sb->s_magic = FUSE_SUPER_MAGIC;
956         sb->s_op = &fuse_super_operations;
957         sb->s_maxbytes = MAX_LFS_FILESIZE;
958         sb->s_time_gran = 1;
959         sb->s_export_op = &fuse_export_operations;
960
961         file = fget(d.fd);
962         err = -EINVAL;
963         if (!file)
964                 goto err;
965
966         if (file->f_op != &fuse_dev_operations)
967                 goto err_fput;
968
969         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
970         err = -ENOMEM;
971         if (!fc)
972                 goto err_fput;
973
974         fuse_conn_init(fc);
975
976         fc->dev = sb->s_dev;
977         fc->sb = sb;
978         err = fuse_bdi_init(fc, sb);
979         if (err)
980                 goto err_put_conn;
981
982         sb->s_bdi = &fc->bdi;
983
984         /* Handle umasking inside the fuse code */
985         if (sb->s_flags & MS_POSIXACL)
986                 fc->dont_mask = 1;
987         sb->s_flags |= MS_POSIXACL;
988
989         fc->release = fuse_free_conn;
990         fc->flags = d.flags;
991         fc->user_id = d.user_id;
992         fc->group_id = d.group_id;
993         fc->max_read = max_t(unsigned, 4096, d.max_read);
994
995         /* Used by get_root_inode() */
996         sb->s_fs_info = fc;
997
998         err = -ENOMEM;
999         root = fuse_get_root_inode(sb, d.rootmode);
1000         root_dentry = d_make_root(root);
1001         if (!root_dentry)
1002                 goto err_put_conn;
1003         /* only now - we want root dentry with NULL ->d_op */
1004         sb->s_d_op = &fuse_dentry_operations;
1005
1006         init_req = fuse_request_alloc();
1007         if (!init_req)
1008                 goto err_put_root;
1009
1010         if (is_bdev) {
1011                 fc->destroy_req = fuse_request_alloc();
1012                 if (!fc->destroy_req)
1013                         goto err_free_init_req;
1014         }
1015
1016         mutex_lock(&fuse_mutex);
1017         err = -EINVAL;
1018         if (file->private_data)
1019                 goto err_unlock;
1020
1021         err = fuse_ctl_add_conn(fc);
1022         if (err)
1023                 goto err_unlock;
1024
1025         list_add_tail(&fc->entry, &fuse_conn_list);
1026         sb->s_root = root_dentry;
1027         fc->connected = 1;
1028         file->private_data = fuse_conn_get(fc);
1029         mutex_unlock(&fuse_mutex);
1030         /*
1031          * atomic_dec_and_test() in fput() provides the necessary
1032          * memory barrier for file->private_data to be visible on all
1033          * CPUs after this
1034          */
1035         fput(file);
1036
1037         fuse_send_init(fc, init_req);
1038
1039         return 0;
1040
1041  err_unlock:
1042         mutex_unlock(&fuse_mutex);
1043  err_free_init_req:
1044         fuse_request_free(init_req);
1045  err_put_root:
1046         dput(root_dentry);
1047  err_put_conn:
1048         fuse_bdi_destroy(fc);
1049         fuse_conn_put(fc);
1050  err_fput:
1051         fput(file);
1052  err:
1053         return err;
1054 }
1055
1056 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1057                        int flags, const char *dev_name,
1058                        void *raw_data)
1059 {
1060         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1061 }
1062
1063 static void fuse_kill_sb_anon(struct super_block *sb)
1064 {
1065         struct fuse_conn *fc = get_fuse_conn_super(sb);
1066
1067         if (fc) {
1068                 down_write(&fc->killsb);
1069                 fc->sb = NULL;
1070                 up_write(&fc->killsb);
1071         }
1072
1073         kill_anon_super(sb);
1074 }
1075
1076 static struct file_system_type fuse_fs_type = {
1077         .owner          = THIS_MODULE,
1078         .name           = "fuse",
1079         .fs_flags       = FS_HAS_SUBTYPE,
1080         .mount          = fuse_mount,
1081         .kill_sb        = fuse_kill_sb_anon,
1082 };
1083
1084 #ifdef CONFIG_BLOCK
1085 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1086                            int flags, const char *dev_name,
1087                            void *raw_data)
1088 {
1089         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1090 }
1091
1092 static void fuse_kill_sb_blk(struct super_block *sb)
1093 {
1094         struct fuse_conn *fc = get_fuse_conn_super(sb);
1095
1096         if (fc) {
1097                 down_write(&fc->killsb);
1098                 fc->sb = NULL;
1099                 up_write(&fc->killsb);
1100         }
1101
1102         kill_block_super(sb);
1103 }
1104
1105 static struct file_system_type fuseblk_fs_type = {
1106         .owner          = THIS_MODULE,
1107         .name           = "fuseblk",
1108         .mount          = fuse_mount_blk,
1109         .kill_sb        = fuse_kill_sb_blk,
1110         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1111 };
1112
1113 static inline int register_fuseblk(void)
1114 {
1115         return register_filesystem(&fuseblk_fs_type);
1116 }
1117
1118 static inline void unregister_fuseblk(void)
1119 {
1120         unregister_filesystem(&fuseblk_fs_type);
1121 }
1122 #else
1123 static inline int register_fuseblk(void)
1124 {
1125         return 0;
1126 }
1127
1128 static inline void unregister_fuseblk(void)
1129 {
1130 }
1131 #endif
1132
1133 static void fuse_inode_init_once(void *foo)
1134 {
1135         struct inode *inode = foo;
1136
1137         inode_init_once(inode);
1138 }
1139
1140 static int __init fuse_fs_init(void)
1141 {
1142         int err;
1143
1144         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1145                                               sizeof(struct fuse_inode),
1146                                               0, SLAB_HWCACHE_ALIGN,
1147                                               fuse_inode_init_once);
1148         err = -ENOMEM;
1149         if (!fuse_inode_cachep)
1150                 goto out;
1151
1152         err = register_fuseblk();
1153         if (err)
1154                 goto out2;
1155
1156         err = register_filesystem(&fuse_fs_type);
1157         if (err)
1158                 goto out3;
1159
1160         return 0;
1161
1162  out3:
1163         unregister_fuseblk();
1164  out2:
1165         kmem_cache_destroy(fuse_inode_cachep);
1166  out:
1167         return err;
1168 }
1169
1170 static void fuse_fs_cleanup(void)
1171 {
1172         unregister_filesystem(&fuse_fs_type);
1173         unregister_fuseblk();
1174         kmem_cache_destroy(fuse_inode_cachep);
1175 }
1176
1177 static struct kobject *fuse_kobj;
1178 static struct kobject *connections_kobj;
1179
1180 static int fuse_sysfs_init(void)
1181 {
1182         int err;
1183
1184         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1185         if (!fuse_kobj) {
1186                 err = -ENOMEM;
1187                 goto out_err;
1188         }
1189
1190         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1191         if (!connections_kobj) {
1192                 err = -ENOMEM;
1193                 goto out_fuse_unregister;
1194         }
1195
1196         return 0;
1197
1198  out_fuse_unregister:
1199         kobject_put(fuse_kobj);
1200  out_err:
1201         return err;
1202 }
1203
1204 static void fuse_sysfs_cleanup(void)
1205 {
1206         kobject_put(connections_kobj);
1207         kobject_put(fuse_kobj);
1208 }
1209
1210 static int __init fuse_init(void)
1211 {
1212         int res;
1213
1214         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1215                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1216
1217         INIT_LIST_HEAD(&fuse_conn_list);
1218         res = fuse_fs_init();
1219         if (res)
1220                 goto err;
1221
1222         res = fuse_dev_init();
1223         if (res)
1224                 goto err_fs_cleanup;
1225
1226         res = fuse_sysfs_init();
1227         if (res)
1228                 goto err_dev_cleanup;
1229
1230         res = fuse_ctl_init();
1231         if (res)
1232                 goto err_sysfs_cleanup;
1233
1234         sanitize_global_limit(&max_user_bgreq);
1235         sanitize_global_limit(&max_user_congthresh);
1236
1237         return 0;
1238
1239  err_sysfs_cleanup:
1240         fuse_sysfs_cleanup();
1241  err_dev_cleanup:
1242         fuse_dev_cleanup();
1243  err_fs_cleanup:
1244         fuse_fs_cleanup();
1245  err:
1246         return res;
1247 }
1248
1249 static void __exit fuse_exit(void)
1250 {
1251         printk(KERN_DEBUG "fuse exit\n");
1252
1253         fuse_ctl_cleanup();
1254         fuse_sysfs_cleanup();
1255         fuse_fs_cleanup();
1256         fuse_dev_cleanup();
1257 }
1258
1259 module_init(fuse_init);
1260 module_exit(fuse_exit);