64cf8d07393e4e55d2101039525fbaa0a1008e35
[linux-2.6.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         INIT_LIST_HEAD(&fi->write_files);
95         INIT_LIST_HEAD(&fi->queued_writes);
96         INIT_LIST_HEAD(&fi->writepages);
97         init_waitqueue_head(&fi->page_waitq);
98         fi->forget = fuse_alloc_forget();
99         if (!fi->forget) {
100                 kmem_cache_free(fuse_inode_cachep, inode);
101                 return NULL;
102         }
103
104         return inode;
105 }
106
107 static void fuse_i_callback(struct rcu_head *head)
108 {
109         struct inode *inode = container_of(head, struct inode, i_rcu);
110         kmem_cache_free(fuse_inode_cachep, inode);
111 }
112
113 static void fuse_destroy_inode(struct inode *inode)
114 {
115         struct fuse_inode *fi = get_fuse_inode(inode);
116         BUG_ON(!list_empty(&fi->write_files));
117         BUG_ON(!list_empty(&fi->queued_writes));
118         kfree(fi->forget);
119         call_rcu(&inode->i_rcu, fuse_i_callback);
120 }
121
122 static void fuse_evict_inode(struct inode *inode)
123 {
124         truncate_inode_pages(&inode->i_data, 0);
125         end_writeback(inode);
126         if (inode->i_sb->s_flags & MS_ACTIVE) {
127                 struct fuse_conn *fc = get_fuse_conn(inode);
128                 struct fuse_inode *fi = get_fuse_inode(inode);
129                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
130                 fi->forget = NULL;
131         }
132 }
133
134 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
135 {
136         if (*flags & MS_MANDLOCK)
137                 return -EINVAL;
138
139         return 0;
140 }
141
142 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
143                                    u64 attr_valid)
144 {
145         struct fuse_conn *fc = get_fuse_conn(inode);
146         struct fuse_inode *fi = get_fuse_inode(inode);
147
148         fi->attr_version = ++fc->attr_version;
149         fi->i_time = attr_valid;
150
151         inode->i_ino     = attr->ino;
152         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
153         set_nlink(inode, attr->nlink);
154         inode->i_uid     = attr->uid;
155         inode->i_gid     = attr->gid;
156         inode->i_blocks  = attr->blocks;
157         inode->i_atime.tv_sec   = attr->atime;
158         inode->i_atime.tv_nsec  = attr->atimensec;
159         inode->i_mtime.tv_sec   = attr->mtime;
160         inode->i_mtime.tv_nsec  = attr->mtimensec;
161         inode->i_ctime.tv_sec   = attr->ctime;
162         inode->i_ctime.tv_nsec  = attr->ctimensec;
163
164         if (attr->blksize != 0)
165                 inode->i_blkbits = ilog2(attr->blksize);
166         else
167                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
168
169         /*
170          * Don't set the sticky bit in i_mode, unless we want the VFS
171          * to check permissions.  This prevents failures due to the
172          * check in may_delete().
173          */
174         fi->orig_i_mode = inode->i_mode;
175         if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
176                 inode->i_mode &= ~S_ISVTX;
177 }
178
179 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
180                             u64 attr_valid, u64 attr_version)
181 {
182         struct fuse_conn *fc = get_fuse_conn(inode);
183         struct fuse_inode *fi = get_fuse_inode(inode);
184         loff_t oldsize;
185
186         spin_lock(&fc->lock);
187         if (attr_version != 0 && fi->attr_version > attr_version) {
188                 spin_unlock(&fc->lock);
189                 return;
190         }
191
192         fuse_change_attributes_common(inode, attr, attr_valid);
193
194         oldsize = inode->i_size;
195         i_size_write(inode, attr->size);
196         spin_unlock(&fc->lock);
197
198         if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
199                 truncate_pagecache(inode, oldsize, attr->size);
200                 invalidate_inode_pages2(inode->i_mapping);
201         }
202 }
203
204 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
205 {
206         inode->i_mode = attr->mode & S_IFMT;
207         inode->i_size = attr->size;
208         if (S_ISREG(inode->i_mode)) {
209                 fuse_init_common(inode);
210                 fuse_init_file_inode(inode);
211         } else if (S_ISDIR(inode->i_mode))
212                 fuse_init_dir(inode);
213         else if (S_ISLNK(inode->i_mode))
214                 fuse_init_symlink(inode);
215         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
216                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
217                 fuse_init_common(inode);
218                 init_special_inode(inode, inode->i_mode,
219                                    new_decode_dev(attr->rdev));
220         } else
221                 BUG();
222 }
223
224 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
225 {
226         u64 nodeid = *(u64 *) _nodeidp;
227         if (get_node_id(inode) == nodeid)
228                 return 1;
229         else
230                 return 0;
231 }
232
233 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
234 {
235         u64 nodeid = *(u64 *) _nodeidp;
236         get_fuse_inode(inode)->nodeid = nodeid;
237         return 0;
238 }
239
240 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
241                         int generation, struct fuse_attr *attr,
242                         u64 attr_valid, u64 attr_version)
243 {
244         struct inode *inode;
245         struct fuse_inode *fi;
246         struct fuse_conn *fc = get_fuse_conn_super(sb);
247
248  retry:
249         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
250         if (!inode)
251                 return NULL;
252
253         if ((inode->i_state & I_NEW)) {
254                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
255                 inode->i_generation = generation;
256                 inode->i_data.backing_dev_info = &fc->bdi;
257                 fuse_init_inode(inode, attr);
258                 unlock_new_inode(inode);
259         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
260                 /* Inode has changed type, any I/O on the old should fail */
261                 make_bad_inode(inode);
262                 iput(inode);
263                 goto retry;
264         }
265
266         fi = get_fuse_inode(inode);
267         spin_lock(&fc->lock);
268         fi->nlookup++;
269         spin_unlock(&fc->lock);
270         fuse_change_attributes(inode, attr, attr_valid, attr_version);
271
272         return inode;
273 }
274
275 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
276                              loff_t offset, loff_t len)
277 {
278         struct inode *inode;
279         pgoff_t pg_start;
280         pgoff_t pg_end;
281
282         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
283         if (!inode)
284                 return -ENOENT;
285
286         fuse_invalidate_attr(inode);
287         if (offset >= 0) {
288                 pg_start = offset >> PAGE_CACHE_SHIFT;
289                 if (len <= 0)
290                         pg_end = -1;
291                 else
292                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
293                 invalidate_inode_pages2_range(inode->i_mapping,
294                                               pg_start, pg_end);
295         }
296         iput(inode);
297         return 0;
298 }
299
300 static void fuse_umount_begin(struct super_block *sb)
301 {
302         fuse_abort_conn(get_fuse_conn_super(sb));
303 }
304
305 static void fuse_send_destroy(struct fuse_conn *fc)
306 {
307         struct fuse_req *req = fc->destroy_req;
308         if (req && fc->conn_init) {
309                 fc->destroy_req = NULL;
310                 req->in.h.opcode = FUSE_DESTROY;
311                 req->force = 1;
312                 fuse_request_send(fc, req);
313                 fuse_put_request(fc, req);
314         }
315 }
316
317 static void fuse_bdi_destroy(struct fuse_conn *fc)
318 {
319         if (fc->bdi_initialized)
320                 bdi_destroy(&fc->bdi);
321 }
322
323 void fuse_conn_kill(struct fuse_conn *fc)
324 {
325         spin_lock(&fc->lock);
326         fc->connected = 0;
327         fc->blocked = 0;
328         spin_unlock(&fc->lock);
329         /* Flush all readers on this fs */
330         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
331         wake_up_all(&fc->waitq);
332         wake_up_all(&fc->blocked_waitq);
333         wake_up_all(&fc->reserved_req_waitq);
334         mutex_lock(&fuse_mutex);
335         list_del(&fc->entry);
336         fuse_ctl_remove_conn(fc);
337         mutex_unlock(&fuse_mutex);
338         fuse_bdi_destroy(fc);
339 }
340 EXPORT_SYMBOL_GPL(fuse_conn_kill);
341
342 static void fuse_put_super(struct super_block *sb)
343 {
344         struct fuse_conn *fc = get_fuse_conn_super(sb);
345
346         fuse_send_destroy(fc);
347         fuse_conn_kill(fc);
348         fuse_conn_put(fc);
349 }
350
351 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
352 {
353         stbuf->f_type    = FUSE_SUPER_MAGIC;
354         stbuf->f_bsize   = attr->bsize;
355         stbuf->f_frsize  = attr->frsize;
356         stbuf->f_blocks  = attr->blocks;
357         stbuf->f_bfree   = attr->bfree;
358         stbuf->f_bavail  = attr->bavail;
359         stbuf->f_files   = attr->files;
360         stbuf->f_ffree   = attr->ffree;
361         stbuf->f_namelen = attr->namelen;
362         /* fsid is left zero */
363 }
364
365 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
366 {
367         struct super_block *sb = dentry->d_sb;
368         struct fuse_conn *fc = get_fuse_conn_super(sb);
369         struct fuse_req *req;
370         struct fuse_statfs_out outarg;
371         int err;
372
373         if (!fuse_allow_task(fc, current)) {
374                 buf->f_type = FUSE_SUPER_MAGIC;
375                 return 0;
376         }
377
378         req = fuse_get_req(fc);
379         if (IS_ERR(req))
380                 return PTR_ERR(req);
381
382         memset(&outarg, 0, sizeof(outarg));
383         req->in.numargs = 0;
384         req->in.h.opcode = FUSE_STATFS;
385         req->in.h.nodeid = get_node_id(dentry->d_inode);
386         req->out.numargs = 1;
387         req->out.args[0].size =
388                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
389         req->out.args[0].value = &outarg;
390         fuse_request_send(fc, req);
391         err = req->out.h.error;
392         if (!err)
393                 convert_fuse_statfs(buf, &outarg.st);
394         fuse_put_request(fc, req);
395         return err;
396 }
397
398 enum {
399         OPT_FD,
400         OPT_ROOTMODE,
401         OPT_USER_ID,
402         OPT_GROUP_ID,
403         OPT_DEFAULT_PERMISSIONS,
404         OPT_ALLOW_OTHER,
405         OPT_MAX_READ,
406         OPT_BLKSIZE,
407         OPT_ERR
408 };
409
410 static const match_table_t tokens = {
411         {OPT_FD,                        "fd=%u"},
412         {OPT_ROOTMODE,                  "rootmode=%o"},
413         {OPT_USER_ID,                   "user_id=%u"},
414         {OPT_GROUP_ID,                  "group_id=%u"},
415         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
416         {OPT_ALLOW_OTHER,               "allow_other"},
417         {OPT_MAX_READ,                  "max_read=%u"},
418         {OPT_BLKSIZE,                   "blksize=%u"},
419         {OPT_ERR,                       NULL}
420 };
421
422 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
423 {
424         char *p;
425         memset(d, 0, sizeof(struct fuse_mount_data));
426         d->max_read = ~0;
427         d->blksize = FUSE_DEFAULT_BLKSIZE;
428
429         while ((p = strsep(&opt, ",")) != NULL) {
430                 int token;
431                 int value;
432                 substring_t args[MAX_OPT_ARGS];
433                 if (!*p)
434                         continue;
435
436                 token = match_token(p, tokens, args);
437                 switch (token) {
438                 case OPT_FD:
439                         if (match_int(&args[0], &value))
440                                 return 0;
441                         d->fd = value;
442                         d->fd_present = 1;
443                         break;
444
445                 case OPT_ROOTMODE:
446                         if (match_octal(&args[0], &value))
447                                 return 0;
448                         if (!fuse_valid_type(value))
449                                 return 0;
450                         d->rootmode = value;
451                         d->rootmode_present = 1;
452                         break;
453
454                 case OPT_USER_ID:
455                         if (match_int(&args[0], &value))
456                                 return 0;
457                         d->user_id = value;
458                         d->user_id_present = 1;
459                         break;
460
461                 case OPT_GROUP_ID:
462                         if (match_int(&args[0], &value))
463                                 return 0;
464                         d->group_id = value;
465                         d->group_id_present = 1;
466                         break;
467
468                 case OPT_DEFAULT_PERMISSIONS:
469                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
470                         break;
471
472                 case OPT_ALLOW_OTHER:
473                         d->flags |= FUSE_ALLOW_OTHER;
474                         break;
475
476                 case OPT_MAX_READ:
477                         if (match_int(&args[0], &value))
478                                 return 0;
479                         d->max_read = value;
480                         break;
481
482                 case OPT_BLKSIZE:
483                         if (!is_bdev || match_int(&args[0], &value))
484                                 return 0;
485                         d->blksize = value;
486                         break;
487
488                 default:
489                         return 0;
490                 }
491         }
492
493         if (!d->fd_present || !d->rootmode_present ||
494             !d->user_id_present || !d->group_id_present)
495                 return 0;
496
497         return 1;
498 }
499
500 static int fuse_show_options(struct seq_file *m, struct dentry *root)
501 {
502         struct super_block *sb = root->d_sb;
503         struct fuse_conn *fc = get_fuse_conn_super(sb);
504
505         seq_printf(m, ",user_id=%u", fc->user_id);
506         seq_printf(m, ",group_id=%u", fc->group_id);
507         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
508                 seq_puts(m, ",default_permissions");
509         if (fc->flags & FUSE_ALLOW_OTHER)
510                 seq_puts(m, ",allow_other");
511         if (fc->max_read != ~0)
512                 seq_printf(m, ",max_read=%u", fc->max_read);
513         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
514                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
515         return 0;
516 }
517
518 void fuse_conn_init(struct fuse_conn *fc)
519 {
520         memset(fc, 0, sizeof(*fc));
521         spin_lock_init(&fc->lock);
522         mutex_init(&fc->inst_mutex);
523         init_rwsem(&fc->killsb);
524         atomic_set(&fc->count, 1);
525         init_waitqueue_head(&fc->waitq);
526         init_waitqueue_head(&fc->blocked_waitq);
527         init_waitqueue_head(&fc->reserved_req_waitq);
528         INIT_LIST_HEAD(&fc->pending);
529         INIT_LIST_HEAD(&fc->processing);
530         INIT_LIST_HEAD(&fc->io);
531         INIT_LIST_HEAD(&fc->interrupts);
532         INIT_LIST_HEAD(&fc->bg_queue);
533         INIT_LIST_HEAD(&fc->entry);
534         fc->forget_list_tail = &fc->forget_list_head;
535         atomic_set(&fc->num_waiting, 0);
536         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
537         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
538         fc->khctr = 0;
539         fc->polled_files = RB_ROOT;
540         fc->reqctr = 0;
541         fc->blocked = 1;
542         fc->attr_version = 1;
543         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
544 }
545 EXPORT_SYMBOL_GPL(fuse_conn_init);
546
547 void fuse_conn_put(struct fuse_conn *fc)
548 {
549         if (atomic_dec_and_test(&fc->count)) {
550                 if (fc->destroy_req)
551                         fuse_request_free(fc->destroy_req);
552                 mutex_destroy(&fc->inst_mutex);
553                 fc->release(fc);
554         }
555 }
556 EXPORT_SYMBOL_GPL(fuse_conn_put);
557
558 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
559 {
560         atomic_inc(&fc->count);
561         return fc;
562 }
563 EXPORT_SYMBOL_GPL(fuse_conn_get);
564
565 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
566 {
567         struct fuse_attr attr;
568         memset(&attr, 0, sizeof(attr));
569
570         attr.mode = mode;
571         attr.ino = FUSE_ROOT_ID;
572         attr.nlink = 1;
573         return fuse_iget(sb, 1, 0, &attr, 0, 0);
574 }
575
576 struct fuse_inode_handle {
577         u64 nodeid;
578         u32 generation;
579 };
580
581 static struct dentry *fuse_get_dentry(struct super_block *sb,
582                                       struct fuse_inode_handle *handle)
583 {
584         struct fuse_conn *fc = get_fuse_conn_super(sb);
585         struct inode *inode;
586         struct dentry *entry;
587         int err = -ESTALE;
588
589         if (handle->nodeid == 0)
590                 goto out_err;
591
592         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
593         if (!inode) {
594                 struct fuse_entry_out outarg;
595                 struct qstr name;
596
597                 if (!fc->export_support)
598                         goto out_err;
599
600                 name.len = 1;
601                 name.name = ".";
602                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
603                                        &inode);
604                 if (err && err != -ENOENT)
605                         goto out_err;
606                 if (err || !inode) {
607                         err = -ESTALE;
608                         goto out_err;
609                 }
610                 err = -EIO;
611                 if (get_node_id(inode) != handle->nodeid)
612                         goto out_iput;
613         }
614         err = -ESTALE;
615         if (inode->i_generation != handle->generation)
616                 goto out_iput;
617
618         entry = d_obtain_alias(inode);
619         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
620                 fuse_invalidate_entry_cache(entry);
621
622         return entry;
623
624  out_iput:
625         iput(inode);
626  out_err:
627         return ERR_PTR(err);
628 }
629
630 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
631                            int connectable)
632 {
633         struct inode *inode = dentry->d_inode;
634         bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
635         int len = encode_parent ? 6 : 3;
636         u64 nodeid;
637         u32 generation;
638
639         if (*max_len < len) {
640                 *max_len = len;
641                 return  255;
642         }
643
644         nodeid = get_fuse_inode(inode)->nodeid;
645         generation = inode->i_generation;
646
647         fh[0] = (u32)(nodeid >> 32);
648         fh[1] = (u32)(nodeid & 0xffffffff);
649         fh[2] = generation;
650
651         if (encode_parent) {
652                 struct inode *parent;
653
654                 spin_lock(&dentry->d_lock);
655                 parent = dentry->d_parent->d_inode;
656                 nodeid = get_fuse_inode(parent)->nodeid;
657                 generation = parent->i_generation;
658                 spin_unlock(&dentry->d_lock);
659
660                 fh[3] = (u32)(nodeid >> 32);
661                 fh[4] = (u32)(nodeid & 0xffffffff);
662                 fh[5] = generation;
663         }
664
665         *max_len = len;
666         return encode_parent ? 0x82 : 0x81;
667 }
668
669 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
670                 struct fid *fid, int fh_len, int fh_type)
671 {
672         struct fuse_inode_handle handle;
673
674         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
675                 return NULL;
676
677         handle.nodeid = (u64) fid->raw[0] << 32;
678         handle.nodeid |= (u64) fid->raw[1];
679         handle.generation = fid->raw[2];
680         return fuse_get_dentry(sb, &handle);
681 }
682
683 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
684                 struct fid *fid, int fh_len, int fh_type)
685 {
686         struct fuse_inode_handle parent;
687
688         if (fh_type != 0x82 || fh_len < 6)
689                 return NULL;
690
691         parent.nodeid = (u64) fid->raw[3] << 32;
692         parent.nodeid |= (u64) fid->raw[4];
693         parent.generation = fid->raw[5];
694         return fuse_get_dentry(sb, &parent);
695 }
696
697 static struct dentry *fuse_get_parent(struct dentry *child)
698 {
699         struct inode *child_inode = child->d_inode;
700         struct fuse_conn *fc = get_fuse_conn(child_inode);
701         struct inode *inode;
702         struct dentry *parent;
703         struct fuse_entry_out outarg;
704         struct qstr name;
705         int err;
706
707         if (!fc->export_support)
708                 return ERR_PTR(-ESTALE);
709
710         name.len = 2;
711         name.name = "..";
712         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
713                                &name, &outarg, &inode);
714         if (err) {
715                 if (err == -ENOENT)
716                         return ERR_PTR(-ESTALE);
717                 return ERR_PTR(err);
718         }
719
720         parent = d_obtain_alias(inode);
721         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
722                 fuse_invalidate_entry_cache(parent);
723
724         return parent;
725 }
726
727 static const struct export_operations fuse_export_operations = {
728         .fh_to_dentry   = fuse_fh_to_dentry,
729         .fh_to_parent   = fuse_fh_to_parent,
730         .encode_fh      = fuse_encode_fh,
731         .get_parent     = fuse_get_parent,
732 };
733
734 static const struct super_operations fuse_super_operations = {
735         .alloc_inode    = fuse_alloc_inode,
736         .destroy_inode  = fuse_destroy_inode,
737         .evict_inode    = fuse_evict_inode,
738         .drop_inode     = generic_delete_inode,
739         .remount_fs     = fuse_remount_fs,
740         .put_super      = fuse_put_super,
741         .umount_begin   = fuse_umount_begin,
742         .statfs         = fuse_statfs,
743         .show_options   = fuse_show_options,
744 };
745
746 static void sanitize_global_limit(unsigned *limit)
747 {
748         if (*limit == 0)
749                 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
750                          sizeof(struct fuse_req);
751
752         if (*limit >= 1 << 16)
753                 *limit = (1 << 16) - 1;
754 }
755
756 static int set_global_limit(const char *val, struct kernel_param *kp)
757 {
758         int rv;
759
760         rv = param_set_uint(val, kp);
761         if (rv)
762                 return rv;
763
764         sanitize_global_limit((unsigned *)kp->arg);
765
766         return 0;
767 }
768
769 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
770 {
771         int cap_sys_admin = capable(CAP_SYS_ADMIN);
772
773         if (arg->minor < 13)
774                 return;
775
776         sanitize_global_limit(&max_user_bgreq);
777         sanitize_global_limit(&max_user_congthresh);
778
779         if (arg->max_background) {
780                 fc->max_background = arg->max_background;
781
782                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
783                         fc->max_background = max_user_bgreq;
784         }
785         if (arg->congestion_threshold) {
786                 fc->congestion_threshold = arg->congestion_threshold;
787
788                 if (!cap_sys_admin &&
789                     fc->congestion_threshold > max_user_congthresh)
790                         fc->congestion_threshold = max_user_congthresh;
791         }
792 }
793
794 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
795 {
796         struct fuse_init_out *arg = &req->misc.init_out;
797
798         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
799                 fc->conn_error = 1;
800         else {
801                 unsigned long ra_pages;
802
803                 process_init_limits(fc, arg);
804
805                 if (arg->minor >= 6) {
806                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
807                         if (arg->flags & FUSE_ASYNC_READ)
808                                 fc->async_read = 1;
809                         if (!(arg->flags & FUSE_POSIX_LOCKS))
810                                 fc->no_lock = 1;
811                         if (arg->minor >= 17) {
812                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
813                                         fc->no_flock = 1;
814                         } else {
815                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
816                                         fc->no_flock = 1;
817                         }
818                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
819                                 fc->atomic_o_trunc = 1;
820                         if (arg->minor >= 9) {
821                                 /* LOOKUP has dependency on proto version */
822                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
823                                         fc->export_support = 1;
824                         }
825                         if (arg->flags & FUSE_BIG_WRITES)
826                                 fc->big_writes = 1;
827                         if (arg->flags & FUSE_DONT_MASK)
828                                 fc->dont_mask = 1;
829                 } else {
830                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
831                         fc->no_lock = 1;
832                         fc->no_flock = 1;
833                 }
834
835                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
836                 fc->minor = arg->minor;
837                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
838                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
839                 fc->conn_init = 1;
840         }
841         fc->blocked = 0;
842         wake_up_all(&fc->blocked_waitq);
843 }
844
845 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
846 {
847         struct fuse_init_in *arg = &req->misc.init_in;
848
849         arg->major = FUSE_KERNEL_VERSION;
850         arg->minor = FUSE_KERNEL_MINOR_VERSION;
851         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
852         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
853                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
854                 FUSE_FLOCK_LOCKS;
855         req->in.h.opcode = FUSE_INIT;
856         req->in.numargs = 1;
857         req->in.args[0].size = sizeof(*arg);
858         req->in.args[0].value = arg;
859         req->out.numargs = 1;
860         /* Variable length argument used for backward compatibility
861            with interface version < 7.5.  Rest of init_out is zeroed
862            by do_get_request(), so a short reply is not a problem */
863         req->out.argvar = 1;
864         req->out.args[0].size = sizeof(struct fuse_init_out);
865         req->out.args[0].value = &req->misc.init_out;
866         req->end = process_init_reply;
867         fuse_request_send_background(fc, req);
868 }
869
870 static void fuse_free_conn(struct fuse_conn *fc)
871 {
872         kfree(fc);
873 }
874
875 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
876 {
877         int err;
878
879         fc->bdi.name = "fuse";
880         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
881         /* fuse does it's own writeback accounting */
882         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
883
884         err = bdi_init(&fc->bdi);
885         if (err)
886                 return err;
887
888         fc->bdi_initialized = 1;
889
890         if (sb->s_bdev) {
891                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
892                                     MAJOR(fc->dev), MINOR(fc->dev));
893         } else {
894                 err = bdi_register_dev(&fc->bdi, fc->dev);
895         }
896
897         if (err)
898                 return err;
899
900         /*
901          * For a single fuse filesystem use max 1% of dirty +
902          * writeback threshold.
903          *
904          * This gives about 1M of write buffer for memory maps on a
905          * machine with 1G and 10% dirty_ratio, which should be more
906          * than enough.
907          *
908          * Privileged users can raise it by writing to
909          *
910          *    /sys/class/bdi/<bdi>/max_ratio
911          */
912         bdi_set_max_ratio(&fc->bdi, 1);
913
914         return 0;
915 }
916
917 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
918 {
919         struct fuse_conn *fc;
920         struct inode *root;
921         struct fuse_mount_data d;
922         struct file *file;
923         struct dentry *root_dentry;
924         struct fuse_req *init_req;
925         int err;
926         int is_bdev = sb->s_bdev != NULL;
927
928         err = -EINVAL;
929         if (sb->s_flags & MS_MANDLOCK)
930                 goto err;
931
932         sb->s_flags &= ~MS_NOSEC;
933
934         if (!parse_fuse_opt((char *) data, &d, is_bdev))
935                 goto err;
936
937         if (is_bdev) {
938 #ifdef CONFIG_BLOCK
939                 err = -EINVAL;
940                 if (!sb_set_blocksize(sb, d.blksize))
941                         goto err;
942 #endif
943         } else {
944                 sb->s_blocksize = PAGE_CACHE_SIZE;
945                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
946         }
947         sb->s_magic = FUSE_SUPER_MAGIC;
948         sb->s_op = &fuse_super_operations;
949         sb->s_maxbytes = MAX_LFS_FILESIZE;
950         sb->s_export_op = &fuse_export_operations;
951
952         file = fget(d.fd);
953         err = -EINVAL;
954         if (!file)
955                 goto err;
956
957         if (file->f_op != &fuse_dev_operations)
958                 goto err_fput;
959
960         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
961         err = -ENOMEM;
962         if (!fc)
963                 goto err_fput;
964
965         fuse_conn_init(fc);
966
967         fc->dev = sb->s_dev;
968         fc->sb = sb;
969         err = fuse_bdi_init(fc, sb);
970         if (err)
971                 goto err_put_conn;
972
973         sb->s_bdi = &fc->bdi;
974
975         /* Handle umasking inside the fuse code */
976         if (sb->s_flags & MS_POSIXACL)
977                 fc->dont_mask = 1;
978         sb->s_flags |= MS_POSIXACL;
979
980         fc->release = fuse_free_conn;
981         fc->flags = d.flags;
982         fc->user_id = d.user_id;
983         fc->group_id = d.group_id;
984         fc->max_read = max_t(unsigned, 4096, d.max_read);
985
986         /* Used by get_root_inode() */
987         sb->s_fs_info = fc;
988
989         err = -ENOMEM;
990         root = fuse_get_root_inode(sb, d.rootmode);
991         if (!root)
992                 goto err_put_conn;
993
994         root_dentry = d_alloc_root(root);
995         if (!root_dentry) {
996                 iput(root);
997                 goto err_put_conn;
998         }
999         /* only now - we want root dentry with NULL ->d_op */
1000         sb->s_d_op = &fuse_dentry_operations;
1001
1002         init_req = fuse_request_alloc();
1003         if (!init_req)
1004                 goto err_put_root;
1005
1006         if (is_bdev) {
1007                 fc->destroy_req = fuse_request_alloc();
1008                 if (!fc->destroy_req)
1009                         goto err_free_init_req;
1010         }
1011
1012         mutex_lock(&fuse_mutex);
1013         err = -EINVAL;
1014         if (file->private_data)
1015                 goto err_unlock;
1016
1017         err = fuse_ctl_add_conn(fc);
1018         if (err)
1019                 goto err_unlock;
1020
1021         list_add_tail(&fc->entry, &fuse_conn_list);
1022         sb->s_root = root_dentry;
1023         fc->connected = 1;
1024         file->private_data = fuse_conn_get(fc);
1025         mutex_unlock(&fuse_mutex);
1026         /*
1027          * atomic_dec_and_test() in fput() provides the necessary
1028          * memory barrier for file->private_data to be visible on all
1029          * CPUs after this
1030          */
1031         fput(file);
1032
1033         fuse_send_init(fc, init_req);
1034
1035         return 0;
1036
1037  err_unlock:
1038         mutex_unlock(&fuse_mutex);
1039  err_free_init_req:
1040         fuse_request_free(init_req);
1041  err_put_root:
1042         dput(root_dentry);
1043  err_put_conn:
1044         fuse_bdi_destroy(fc);
1045         fuse_conn_put(fc);
1046  err_fput:
1047         fput(file);
1048  err:
1049         return err;
1050 }
1051
1052 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1053                        int flags, const char *dev_name,
1054                        void *raw_data)
1055 {
1056         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1057 }
1058
1059 static void fuse_kill_sb_anon(struct super_block *sb)
1060 {
1061         struct fuse_conn *fc = get_fuse_conn_super(sb);
1062
1063         if (fc) {
1064                 down_write(&fc->killsb);
1065                 fc->sb = NULL;
1066                 up_write(&fc->killsb);
1067         }
1068
1069         kill_anon_super(sb);
1070 }
1071
1072 static struct file_system_type fuse_fs_type = {
1073         .owner          = THIS_MODULE,
1074         .name           = "fuse",
1075         .fs_flags       = FS_HAS_SUBTYPE,
1076         .mount          = fuse_mount,
1077         .kill_sb        = fuse_kill_sb_anon,
1078 };
1079
1080 #ifdef CONFIG_BLOCK
1081 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1082                            int flags, const char *dev_name,
1083                            void *raw_data)
1084 {
1085         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1086 }
1087
1088 static void fuse_kill_sb_blk(struct super_block *sb)
1089 {
1090         struct fuse_conn *fc = get_fuse_conn_super(sb);
1091
1092         if (fc) {
1093                 down_write(&fc->killsb);
1094                 fc->sb = NULL;
1095                 up_write(&fc->killsb);
1096         }
1097
1098         kill_block_super(sb);
1099 }
1100
1101 static struct file_system_type fuseblk_fs_type = {
1102         .owner          = THIS_MODULE,
1103         .name           = "fuseblk",
1104         .mount          = fuse_mount_blk,
1105         .kill_sb        = fuse_kill_sb_blk,
1106         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1107 };
1108
1109 static inline int register_fuseblk(void)
1110 {
1111         return register_filesystem(&fuseblk_fs_type);
1112 }
1113
1114 static inline void unregister_fuseblk(void)
1115 {
1116         unregister_filesystem(&fuseblk_fs_type);
1117 }
1118 #else
1119 static inline int register_fuseblk(void)
1120 {
1121         return 0;
1122 }
1123
1124 static inline void unregister_fuseblk(void)
1125 {
1126 }
1127 #endif
1128
1129 static void fuse_inode_init_once(void *foo)
1130 {
1131         struct inode *inode = foo;
1132
1133         inode_init_once(inode);
1134 }
1135
1136 static int __init fuse_fs_init(void)
1137 {
1138         int err;
1139
1140         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1141                                               sizeof(struct fuse_inode),
1142                                               0, SLAB_HWCACHE_ALIGN,
1143                                               fuse_inode_init_once);
1144         err = -ENOMEM;
1145         if (!fuse_inode_cachep)
1146                 goto out;
1147
1148         err = register_fuseblk();
1149         if (err)
1150                 goto out2;
1151
1152         err = register_filesystem(&fuse_fs_type);
1153         if (err)
1154                 goto out3;
1155
1156         return 0;
1157
1158  out3:
1159         unregister_fuseblk();
1160  out2:
1161         kmem_cache_destroy(fuse_inode_cachep);
1162  out:
1163         return err;
1164 }
1165
1166 static void fuse_fs_cleanup(void)
1167 {
1168         unregister_filesystem(&fuse_fs_type);
1169         unregister_fuseblk();
1170         kmem_cache_destroy(fuse_inode_cachep);
1171 }
1172
1173 static struct kobject *fuse_kobj;
1174 static struct kobject *connections_kobj;
1175
1176 static int fuse_sysfs_init(void)
1177 {
1178         int err;
1179
1180         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1181         if (!fuse_kobj) {
1182                 err = -ENOMEM;
1183                 goto out_err;
1184         }
1185
1186         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1187         if (!connections_kobj) {
1188                 err = -ENOMEM;
1189                 goto out_fuse_unregister;
1190         }
1191
1192         return 0;
1193
1194  out_fuse_unregister:
1195         kobject_put(fuse_kobj);
1196  out_err:
1197         return err;
1198 }
1199
1200 static void fuse_sysfs_cleanup(void)
1201 {
1202         kobject_put(connections_kobj);
1203         kobject_put(fuse_kobj);
1204 }
1205
1206 static int __init fuse_init(void)
1207 {
1208         int res;
1209
1210         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1211                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1212
1213         INIT_LIST_HEAD(&fuse_conn_list);
1214         res = fuse_fs_init();
1215         if (res)
1216                 goto err;
1217
1218         res = fuse_dev_init();
1219         if (res)
1220                 goto err_fs_cleanup;
1221
1222         res = fuse_sysfs_init();
1223         if (res)
1224                 goto err_dev_cleanup;
1225
1226         res = fuse_ctl_init();
1227         if (res)
1228                 goto err_sysfs_cleanup;
1229
1230         sanitize_global_limit(&max_user_bgreq);
1231         sanitize_global_limit(&max_user_congthresh);
1232
1233         return 0;
1234
1235  err_sysfs_cleanup:
1236         fuse_sysfs_cleanup();
1237  err_dev_cleanup:
1238         fuse_dev_cleanup();
1239  err_fs_cleanup:
1240         fuse_fs_cleanup();
1241  err:
1242         return res;
1243 }
1244
1245 static void __exit fuse_exit(void)
1246 {
1247         printk(KERN_DEBUG "fuse exit\n");
1248
1249         fuse_ctl_cleanup();
1250         fuse_sysfs_cleanup();
1251         fuse_fs_cleanup();
1252         fuse_dev_cleanup();
1253 }
1254
1255 module_init(fuse_init);
1256 module_exit(fuse_exit);