Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi...
[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         kuid_t user_id;
64         kgid_t 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         fi->state = 0;
96         INIT_LIST_HEAD(&fi->write_files);
97         INIT_LIST_HEAD(&fi->queued_writes);
98         INIT_LIST_HEAD(&fi->writepages);
99         init_waitqueue_head(&fi->page_waitq);
100         fi->forget = fuse_alloc_forget();
101         if (!fi->forget) {
102                 kmem_cache_free(fuse_inode_cachep, inode);
103                 return NULL;
104         }
105
106         return inode;
107 }
108
109 static void fuse_i_callback(struct rcu_head *head)
110 {
111         struct inode *inode = container_of(head, struct inode, i_rcu);
112         kmem_cache_free(fuse_inode_cachep, inode);
113 }
114
115 static void fuse_destroy_inode(struct inode *inode)
116 {
117         struct fuse_inode *fi = get_fuse_inode(inode);
118         BUG_ON(!list_empty(&fi->write_files));
119         BUG_ON(!list_empty(&fi->queued_writes));
120         kfree(fi->forget);
121         call_rcu(&inode->i_rcu, fuse_i_callback);
122 }
123
124 static void fuse_evict_inode(struct inode *inode)
125 {
126         truncate_inode_pages_final(&inode->i_data);
127         clear_inode(inode);
128         if (inode->i_sb->s_flags & MS_ACTIVE) {
129                 struct fuse_conn *fc = get_fuse_conn(inode);
130                 struct fuse_inode *fi = get_fuse_inode(inode);
131                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
132                 fi->forget = NULL;
133         }
134 }
135
136 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
137 {
138         if (*flags & MS_MANDLOCK)
139                 return -EINVAL;
140
141         return 0;
142 }
143
144 /*
145  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
146  * so that it will fit.
147  */
148 static ino_t fuse_squash_ino(u64 ino64)
149 {
150         ino_t ino = (ino_t) ino64;
151         if (sizeof(ino_t) < sizeof(u64))
152                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
153         return ino;
154 }
155
156 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
157                                    u64 attr_valid)
158 {
159         struct fuse_conn *fc = get_fuse_conn(inode);
160         struct fuse_inode *fi = get_fuse_inode(inode);
161
162         fi->attr_version = ++fc->attr_version;
163         fi->i_time = attr_valid;
164
165         inode->i_ino     = fuse_squash_ino(attr->ino);
166         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
167         set_nlink(inode, attr->nlink);
168         inode->i_uid     = make_kuid(&init_user_ns, attr->uid);
169         inode->i_gid     = make_kgid(&init_user_ns, attr->gid);
170         inode->i_blocks  = attr->blocks;
171         inode->i_atime.tv_sec   = attr->atime;
172         inode->i_atime.tv_nsec  = attr->atimensec;
173         /* mtime from server may be stale due to local buffered write */
174         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
175                 inode->i_mtime.tv_sec   = attr->mtime;
176                 inode->i_mtime.tv_nsec  = attr->mtimensec;
177         }
178         inode->i_ctime.tv_sec   = attr->ctime;
179         inode->i_ctime.tv_nsec  = attr->ctimensec;
180
181         if (attr->blksize != 0)
182                 inode->i_blkbits = ilog2(attr->blksize);
183         else
184                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
185
186         /*
187          * Don't set the sticky bit in i_mode, unless we want the VFS
188          * to check permissions.  This prevents failures due to the
189          * check in may_delete().
190          */
191         fi->orig_i_mode = inode->i_mode;
192         if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
193                 inode->i_mode &= ~S_ISVTX;
194
195         fi->orig_ino = attr->ino;
196 }
197
198 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
199                             u64 attr_valid, u64 attr_version)
200 {
201         struct fuse_conn *fc = get_fuse_conn(inode);
202         struct fuse_inode *fi = get_fuse_inode(inode);
203         bool is_wb = fc->writeback_cache;
204         loff_t oldsize;
205         struct timespec old_mtime;
206
207         spin_lock(&fc->lock);
208         if ((attr_version != 0 && fi->attr_version > attr_version) ||
209             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
210                 spin_unlock(&fc->lock);
211                 return;
212         }
213
214         old_mtime = inode->i_mtime;
215         fuse_change_attributes_common(inode, attr, attr_valid);
216
217         oldsize = inode->i_size;
218         /*
219          * In case of writeback_cache enabled, the cached writes beyond EOF
220          * extend local i_size without keeping userspace server in sync. So,
221          * attr->size coming from server can be stale. We cannot trust it.
222          */
223         if (!is_wb || !S_ISREG(inode->i_mode))
224                 i_size_write(inode, attr->size);
225         spin_unlock(&fc->lock);
226
227         if (!is_wb && S_ISREG(inode->i_mode)) {
228                 bool inval = false;
229
230                 if (oldsize != attr->size) {
231                         truncate_pagecache(inode, attr->size);
232                         inval = true;
233                 } else if (fc->auto_inval_data) {
234                         struct timespec new_mtime = {
235                                 .tv_sec = attr->mtime,
236                                 .tv_nsec = attr->mtimensec,
237                         };
238
239                         /*
240                          * Auto inval mode also checks and invalidates if mtime
241                          * has changed.
242                          */
243                         if (!timespec_equal(&old_mtime, &new_mtime))
244                                 inval = true;
245                 }
246
247                 if (inval)
248                         invalidate_inode_pages2(inode->i_mapping);
249         }
250 }
251
252 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
253 {
254         inode->i_mode = attr->mode & S_IFMT;
255         inode->i_size = attr->size;
256         inode->i_mtime.tv_sec  = attr->mtime;
257         inode->i_mtime.tv_nsec = attr->mtimensec;
258         if (S_ISREG(inode->i_mode)) {
259                 fuse_init_common(inode);
260                 fuse_init_file_inode(inode);
261         } else if (S_ISDIR(inode->i_mode))
262                 fuse_init_dir(inode);
263         else if (S_ISLNK(inode->i_mode))
264                 fuse_init_symlink(inode);
265         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
266                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
267                 fuse_init_common(inode);
268                 init_special_inode(inode, inode->i_mode,
269                                    new_decode_dev(attr->rdev));
270         } else
271                 BUG();
272 }
273
274 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
275 {
276         u64 nodeid = *(u64 *) _nodeidp;
277         if (get_node_id(inode) == nodeid)
278                 return 1;
279         else
280                 return 0;
281 }
282
283 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
284 {
285         u64 nodeid = *(u64 *) _nodeidp;
286         get_fuse_inode(inode)->nodeid = nodeid;
287         return 0;
288 }
289
290 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
291                         int generation, struct fuse_attr *attr,
292                         u64 attr_valid, u64 attr_version)
293 {
294         struct inode *inode;
295         struct fuse_inode *fi;
296         struct fuse_conn *fc = get_fuse_conn_super(sb);
297
298  retry:
299         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
300         if (!inode)
301                 return NULL;
302
303         if ((inode->i_state & I_NEW)) {
304                 inode->i_flags |= S_NOATIME;
305                 if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
306                         inode->i_flags |= S_NOCMTIME;
307                 inode->i_generation = generation;
308                 inode->i_data.backing_dev_info = &fc->bdi;
309                 fuse_init_inode(inode, attr);
310                 unlock_new_inode(inode);
311         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
312                 /* Inode has changed type, any I/O on the old should fail */
313                 make_bad_inode(inode);
314                 iput(inode);
315                 goto retry;
316         }
317
318         fi = get_fuse_inode(inode);
319         spin_lock(&fc->lock);
320         fi->nlookup++;
321         spin_unlock(&fc->lock);
322         fuse_change_attributes(inode, attr, attr_valid, attr_version);
323
324         return inode;
325 }
326
327 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
328                              loff_t offset, loff_t len)
329 {
330         struct inode *inode;
331         pgoff_t pg_start;
332         pgoff_t pg_end;
333
334         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
335         if (!inode)
336                 return -ENOENT;
337
338         fuse_invalidate_attr(inode);
339         if (offset >= 0) {
340                 pg_start = offset >> PAGE_CACHE_SHIFT;
341                 if (len <= 0)
342                         pg_end = -1;
343                 else
344                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
345                 invalidate_inode_pages2_range(inode->i_mapping,
346                                               pg_start, pg_end);
347         }
348         iput(inode);
349         return 0;
350 }
351
352 static void fuse_umount_begin(struct super_block *sb)
353 {
354         fuse_abort_conn(get_fuse_conn_super(sb));
355 }
356
357 static void fuse_send_destroy(struct fuse_conn *fc)
358 {
359         struct fuse_req *req = fc->destroy_req;
360         if (req && fc->conn_init) {
361                 fc->destroy_req = NULL;
362                 req->in.h.opcode = FUSE_DESTROY;
363                 req->force = 1;
364                 req->background = 0;
365                 fuse_request_send(fc, req);
366                 fuse_put_request(fc, req);
367         }
368 }
369
370 static void fuse_bdi_destroy(struct fuse_conn *fc)
371 {
372         if (fc->bdi_initialized)
373                 bdi_destroy(&fc->bdi);
374 }
375
376 void fuse_conn_kill(struct fuse_conn *fc)
377 {
378         spin_lock(&fc->lock);
379         fc->connected = 0;
380         fc->blocked = 0;
381         fc->initialized = 1;
382         spin_unlock(&fc->lock);
383         /* Flush all readers on this fs */
384         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
385         wake_up_all(&fc->waitq);
386         wake_up_all(&fc->blocked_waitq);
387         wake_up_all(&fc->reserved_req_waitq);
388 }
389 EXPORT_SYMBOL_GPL(fuse_conn_kill);
390
391 static void fuse_put_super(struct super_block *sb)
392 {
393         struct fuse_conn *fc = get_fuse_conn_super(sb);
394
395         fuse_send_destroy(fc);
396
397         fuse_conn_kill(fc);
398         mutex_lock(&fuse_mutex);
399         list_del(&fc->entry);
400         fuse_ctl_remove_conn(fc);
401         mutex_unlock(&fuse_mutex);
402         fuse_bdi_destroy(fc);
403
404         fuse_conn_put(fc);
405 }
406
407 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
408 {
409         stbuf->f_type    = FUSE_SUPER_MAGIC;
410         stbuf->f_bsize   = attr->bsize;
411         stbuf->f_frsize  = attr->frsize;
412         stbuf->f_blocks  = attr->blocks;
413         stbuf->f_bfree   = attr->bfree;
414         stbuf->f_bavail  = attr->bavail;
415         stbuf->f_files   = attr->files;
416         stbuf->f_ffree   = attr->ffree;
417         stbuf->f_namelen = attr->namelen;
418         /* fsid is left zero */
419 }
420
421 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
422 {
423         struct super_block *sb = dentry->d_sb;
424         struct fuse_conn *fc = get_fuse_conn_super(sb);
425         struct fuse_req *req;
426         struct fuse_statfs_out outarg;
427         int err;
428
429         if (!fuse_allow_current_process(fc)) {
430                 buf->f_type = FUSE_SUPER_MAGIC;
431                 return 0;
432         }
433
434         req = fuse_get_req_nopages(fc);
435         if (IS_ERR(req))
436                 return PTR_ERR(req);
437
438         memset(&outarg, 0, sizeof(outarg));
439         req->in.numargs = 0;
440         req->in.h.opcode = FUSE_STATFS;
441         req->in.h.nodeid = get_node_id(dentry->d_inode);
442         req->out.numargs = 1;
443         req->out.args[0].size =
444                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
445         req->out.args[0].value = &outarg;
446         fuse_request_send(fc, req);
447         err = req->out.h.error;
448         if (!err)
449                 convert_fuse_statfs(buf, &outarg.st);
450         fuse_put_request(fc, req);
451         return err;
452 }
453
454 enum {
455         OPT_FD,
456         OPT_ROOTMODE,
457         OPT_USER_ID,
458         OPT_GROUP_ID,
459         OPT_DEFAULT_PERMISSIONS,
460         OPT_ALLOW_OTHER,
461         OPT_MAX_READ,
462         OPT_BLKSIZE,
463         OPT_ERR
464 };
465
466 static const match_table_t tokens = {
467         {OPT_FD,                        "fd=%u"},
468         {OPT_ROOTMODE,                  "rootmode=%o"},
469         {OPT_USER_ID,                   "user_id=%u"},
470         {OPT_GROUP_ID,                  "group_id=%u"},
471         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
472         {OPT_ALLOW_OTHER,               "allow_other"},
473         {OPT_MAX_READ,                  "max_read=%u"},
474         {OPT_BLKSIZE,                   "blksize=%u"},
475         {OPT_ERR,                       NULL}
476 };
477
478 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
479 {
480         char *p;
481         memset(d, 0, sizeof(struct fuse_mount_data));
482         d->max_read = ~0;
483         d->blksize = FUSE_DEFAULT_BLKSIZE;
484
485         while ((p = strsep(&opt, ",")) != NULL) {
486                 int token;
487                 int value;
488                 substring_t args[MAX_OPT_ARGS];
489                 if (!*p)
490                         continue;
491
492                 token = match_token(p, tokens, args);
493                 switch (token) {
494                 case OPT_FD:
495                         if (match_int(&args[0], &value))
496                                 return 0;
497                         d->fd = value;
498                         d->fd_present = 1;
499                         break;
500
501                 case OPT_ROOTMODE:
502                         if (match_octal(&args[0], &value))
503                                 return 0;
504                         if (!fuse_valid_type(value))
505                                 return 0;
506                         d->rootmode = value;
507                         d->rootmode_present = 1;
508                         break;
509
510                 case OPT_USER_ID:
511                         if (match_int(&args[0], &value))
512                                 return 0;
513                         d->user_id = make_kuid(current_user_ns(), value);
514                         if (!uid_valid(d->user_id))
515                                 return 0;
516                         d->user_id_present = 1;
517                         break;
518
519                 case OPT_GROUP_ID:
520                         if (match_int(&args[0], &value))
521                                 return 0;
522                         d->group_id = make_kgid(current_user_ns(), value);
523                         if (!gid_valid(d->group_id))
524                                 return 0;
525                         d->group_id_present = 1;
526                         break;
527
528                 case OPT_DEFAULT_PERMISSIONS:
529                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
530                         break;
531
532                 case OPT_ALLOW_OTHER:
533                         d->flags |= FUSE_ALLOW_OTHER;
534                         break;
535
536                 case OPT_MAX_READ:
537                         if (match_int(&args[0], &value))
538                                 return 0;
539                         d->max_read = value;
540                         break;
541
542                 case OPT_BLKSIZE:
543                         if (!is_bdev || match_int(&args[0], &value))
544                                 return 0;
545                         d->blksize = value;
546                         break;
547
548                 default:
549                         return 0;
550                 }
551         }
552
553         if (!d->fd_present || !d->rootmode_present ||
554             !d->user_id_present || !d->group_id_present)
555                 return 0;
556
557         return 1;
558 }
559
560 static int fuse_show_options(struct seq_file *m, struct dentry *root)
561 {
562         struct super_block *sb = root->d_sb;
563         struct fuse_conn *fc = get_fuse_conn_super(sb);
564
565         seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
566         seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
567         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
568                 seq_puts(m, ",default_permissions");
569         if (fc->flags & FUSE_ALLOW_OTHER)
570                 seq_puts(m, ",allow_other");
571         if (fc->max_read != ~0)
572                 seq_printf(m, ",max_read=%u", fc->max_read);
573         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
574                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
575         return 0;
576 }
577
578 void fuse_conn_init(struct fuse_conn *fc)
579 {
580         memset(fc, 0, sizeof(*fc));
581         spin_lock_init(&fc->lock);
582         init_rwsem(&fc->killsb);
583         atomic_set(&fc->count, 1);
584         init_waitqueue_head(&fc->waitq);
585         init_waitqueue_head(&fc->blocked_waitq);
586         init_waitqueue_head(&fc->reserved_req_waitq);
587         INIT_LIST_HEAD(&fc->pending);
588         INIT_LIST_HEAD(&fc->processing);
589         INIT_LIST_HEAD(&fc->io);
590         INIT_LIST_HEAD(&fc->interrupts);
591         INIT_LIST_HEAD(&fc->bg_queue);
592         INIT_LIST_HEAD(&fc->entry);
593         fc->forget_list_tail = &fc->forget_list_head;
594         atomic_set(&fc->num_waiting, 0);
595         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
596         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
597         fc->khctr = 0;
598         fc->polled_files = RB_ROOT;
599         fc->reqctr = 0;
600         fc->blocked = 0;
601         fc->initialized = 0;
602         fc->attr_version = 1;
603         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
604 }
605 EXPORT_SYMBOL_GPL(fuse_conn_init);
606
607 void fuse_conn_put(struct fuse_conn *fc)
608 {
609         if (atomic_dec_and_test(&fc->count)) {
610                 if (fc->destroy_req)
611                         fuse_request_free(fc->destroy_req);
612                 fc->release(fc);
613         }
614 }
615 EXPORT_SYMBOL_GPL(fuse_conn_put);
616
617 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
618 {
619         atomic_inc(&fc->count);
620         return fc;
621 }
622 EXPORT_SYMBOL_GPL(fuse_conn_get);
623
624 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
625 {
626         struct fuse_attr attr;
627         memset(&attr, 0, sizeof(attr));
628
629         attr.mode = mode;
630         attr.ino = FUSE_ROOT_ID;
631         attr.nlink = 1;
632         return fuse_iget(sb, 1, 0, &attr, 0, 0);
633 }
634
635 struct fuse_inode_handle {
636         u64 nodeid;
637         u32 generation;
638 };
639
640 static struct dentry *fuse_get_dentry(struct super_block *sb,
641                                       struct fuse_inode_handle *handle)
642 {
643         struct fuse_conn *fc = get_fuse_conn_super(sb);
644         struct inode *inode;
645         struct dentry *entry;
646         int err = -ESTALE;
647
648         if (handle->nodeid == 0)
649                 goto out_err;
650
651         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
652         if (!inode) {
653                 struct fuse_entry_out outarg;
654                 struct qstr name;
655
656                 if (!fc->export_support)
657                         goto out_err;
658
659                 name.len = 1;
660                 name.name = ".";
661                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
662                                        &inode);
663                 if (err && err != -ENOENT)
664                         goto out_err;
665                 if (err || !inode) {
666                         err = -ESTALE;
667                         goto out_err;
668                 }
669                 err = -EIO;
670                 if (get_node_id(inode) != handle->nodeid)
671                         goto out_iput;
672         }
673         err = -ESTALE;
674         if (inode->i_generation != handle->generation)
675                 goto out_iput;
676
677         entry = d_obtain_alias(inode);
678         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
679                 fuse_invalidate_entry_cache(entry);
680
681         return entry;
682
683  out_iput:
684         iput(inode);
685  out_err:
686         return ERR_PTR(err);
687 }
688
689 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
690                            struct inode *parent)
691 {
692         int len = parent ? 6 : 3;
693         u64 nodeid;
694         u32 generation;
695
696         if (*max_len < len) {
697                 *max_len = len;
698                 return  FILEID_INVALID;
699         }
700
701         nodeid = get_fuse_inode(inode)->nodeid;
702         generation = inode->i_generation;
703
704         fh[0] = (u32)(nodeid >> 32);
705         fh[1] = (u32)(nodeid & 0xffffffff);
706         fh[2] = generation;
707
708         if (parent) {
709                 nodeid = get_fuse_inode(parent)->nodeid;
710                 generation = parent->i_generation;
711
712                 fh[3] = (u32)(nodeid >> 32);
713                 fh[4] = (u32)(nodeid & 0xffffffff);
714                 fh[5] = generation;
715         }
716
717         *max_len = len;
718         return parent ? 0x82 : 0x81;
719 }
720
721 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
722                 struct fid *fid, int fh_len, int fh_type)
723 {
724         struct fuse_inode_handle handle;
725
726         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
727                 return NULL;
728
729         handle.nodeid = (u64) fid->raw[0] << 32;
730         handle.nodeid |= (u64) fid->raw[1];
731         handle.generation = fid->raw[2];
732         return fuse_get_dentry(sb, &handle);
733 }
734
735 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
736                 struct fid *fid, int fh_len, int fh_type)
737 {
738         struct fuse_inode_handle parent;
739
740         if (fh_type != 0x82 || fh_len < 6)
741                 return NULL;
742
743         parent.nodeid = (u64) fid->raw[3] << 32;
744         parent.nodeid |= (u64) fid->raw[4];
745         parent.generation = fid->raw[5];
746         return fuse_get_dentry(sb, &parent);
747 }
748
749 static struct dentry *fuse_get_parent(struct dentry *child)
750 {
751         struct inode *child_inode = child->d_inode;
752         struct fuse_conn *fc = get_fuse_conn(child_inode);
753         struct inode *inode;
754         struct dentry *parent;
755         struct fuse_entry_out outarg;
756         struct qstr name;
757         int err;
758
759         if (!fc->export_support)
760                 return ERR_PTR(-ESTALE);
761
762         name.len = 2;
763         name.name = "..";
764         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
765                                &name, &outarg, &inode);
766         if (err) {
767                 if (err == -ENOENT)
768                         return ERR_PTR(-ESTALE);
769                 return ERR_PTR(err);
770         }
771
772         parent = d_obtain_alias(inode);
773         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
774                 fuse_invalidate_entry_cache(parent);
775
776         return parent;
777 }
778
779 static const struct export_operations fuse_export_operations = {
780         .fh_to_dentry   = fuse_fh_to_dentry,
781         .fh_to_parent   = fuse_fh_to_parent,
782         .encode_fh      = fuse_encode_fh,
783         .get_parent     = fuse_get_parent,
784 };
785
786 static const struct super_operations fuse_super_operations = {
787         .alloc_inode    = fuse_alloc_inode,
788         .destroy_inode  = fuse_destroy_inode,
789         .evict_inode    = fuse_evict_inode,
790         .drop_inode     = generic_delete_inode,
791         .remount_fs     = fuse_remount_fs,
792         .put_super      = fuse_put_super,
793         .umount_begin   = fuse_umount_begin,
794         .statfs         = fuse_statfs,
795         .show_options   = fuse_show_options,
796 };
797
798 static void sanitize_global_limit(unsigned *limit)
799 {
800         if (*limit == 0)
801                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
802                          sizeof(struct fuse_req);
803
804         if (*limit >= 1 << 16)
805                 *limit = (1 << 16) - 1;
806 }
807
808 static int set_global_limit(const char *val, struct kernel_param *kp)
809 {
810         int rv;
811
812         rv = param_set_uint(val, kp);
813         if (rv)
814                 return rv;
815
816         sanitize_global_limit((unsigned *)kp->arg);
817
818         return 0;
819 }
820
821 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
822 {
823         int cap_sys_admin = capable(CAP_SYS_ADMIN);
824
825         if (arg->minor < 13)
826                 return;
827
828         sanitize_global_limit(&max_user_bgreq);
829         sanitize_global_limit(&max_user_congthresh);
830
831         if (arg->max_background) {
832                 fc->max_background = arg->max_background;
833
834                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
835                         fc->max_background = max_user_bgreq;
836         }
837         if (arg->congestion_threshold) {
838                 fc->congestion_threshold = arg->congestion_threshold;
839
840                 if (!cap_sys_admin &&
841                     fc->congestion_threshold > max_user_congthresh)
842                         fc->congestion_threshold = max_user_congthresh;
843         }
844 }
845
846 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
847 {
848         struct fuse_init_out *arg = &req->misc.init_out;
849
850         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
851                 fc->conn_error = 1;
852         else {
853                 unsigned long ra_pages;
854
855                 process_init_limits(fc, arg);
856
857                 if (arg->minor >= 6) {
858                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
859                         if (arg->flags & FUSE_ASYNC_READ)
860                                 fc->async_read = 1;
861                         if (!(arg->flags & FUSE_POSIX_LOCKS))
862                                 fc->no_lock = 1;
863                         if (arg->minor >= 17) {
864                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
865                                         fc->no_flock = 1;
866                         } else {
867                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
868                                         fc->no_flock = 1;
869                         }
870                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
871                                 fc->atomic_o_trunc = 1;
872                         if (arg->minor >= 9) {
873                                 /* LOOKUP has dependency on proto version */
874                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
875                                         fc->export_support = 1;
876                         }
877                         if (arg->flags & FUSE_BIG_WRITES)
878                                 fc->big_writes = 1;
879                         if (arg->flags & FUSE_DONT_MASK)
880                                 fc->dont_mask = 1;
881                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
882                                 fc->auto_inval_data = 1;
883                         if (arg->flags & FUSE_DO_READDIRPLUS) {
884                                 fc->do_readdirplus = 1;
885                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
886                                         fc->readdirplus_auto = 1;
887                         }
888                         if (arg->flags & FUSE_ASYNC_DIO)
889                                 fc->async_dio = 1;
890                         if (arg->flags & FUSE_WRITEBACK_CACHE)
891                                 fc->writeback_cache = 1;
892                 } else {
893                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
894                         fc->no_lock = 1;
895                         fc->no_flock = 1;
896                 }
897
898                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
899                 fc->minor = arg->minor;
900                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
901                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
902                 fc->conn_init = 1;
903         }
904         fc->initialized = 1;
905         wake_up_all(&fc->blocked_waitq);
906 }
907
908 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
909 {
910         struct fuse_init_in *arg = &req->misc.init_in;
911
912         arg->major = FUSE_KERNEL_VERSION;
913         arg->minor = FUSE_KERNEL_MINOR_VERSION;
914         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
915         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
916                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
917                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
918                 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
919                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
920                 FUSE_WRITEBACK_CACHE;
921         req->in.h.opcode = FUSE_INIT;
922         req->in.numargs = 1;
923         req->in.args[0].size = sizeof(*arg);
924         req->in.args[0].value = arg;
925         req->out.numargs = 1;
926         /* Variable length argument used for backward compatibility
927            with interface version < 7.5.  Rest of init_out is zeroed
928            by do_get_request(), so a short reply is not a problem */
929         req->out.argvar = 1;
930         req->out.args[0].size = sizeof(struct fuse_init_out);
931         req->out.args[0].value = &req->misc.init_out;
932         req->end = process_init_reply;
933         fuse_request_send_background(fc, req);
934 }
935
936 static void fuse_free_conn(struct fuse_conn *fc)
937 {
938         kfree_rcu(fc, rcu);
939 }
940
941 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
942 {
943         int err;
944
945         fc->bdi.name = "fuse";
946         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
947         /* fuse does it's own writeback accounting */
948         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
949
950         err = bdi_init(&fc->bdi);
951         if (err)
952                 return err;
953
954         fc->bdi_initialized = 1;
955
956         if (sb->s_bdev) {
957                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
958                                     MAJOR(fc->dev), MINOR(fc->dev));
959         } else {
960                 err = bdi_register_dev(&fc->bdi, fc->dev);
961         }
962
963         if (err)
964                 return err;
965
966         /*
967          * For a single fuse filesystem use max 1% of dirty +
968          * writeback threshold.
969          *
970          * This gives about 1M of write buffer for memory maps on a
971          * machine with 1G and 10% dirty_ratio, which should be more
972          * than enough.
973          *
974          * Privileged users can raise it by writing to
975          *
976          *    /sys/class/bdi/<bdi>/max_ratio
977          */
978         bdi_set_max_ratio(&fc->bdi, 1);
979
980         return 0;
981 }
982
983 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
984 {
985         struct fuse_conn *fc;
986         struct inode *root;
987         struct fuse_mount_data d;
988         struct file *file;
989         struct dentry *root_dentry;
990         struct fuse_req *init_req;
991         int err;
992         int is_bdev = sb->s_bdev != NULL;
993
994         err = -EINVAL;
995         if (sb->s_flags & MS_MANDLOCK)
996                 goto err;
997
998         sb->s_flags &= ~MS_NOSEC;
999
1000         if (!parse_fuse_opt((char *) data, &d, is_bdev))
1001                 goto err;
1002
1003         if (is_bdev) {
1004 #ifdef CONFIG_BLOCK
1005                 err = -EINVAL;
1006                 if (!sb_set_blocksize(sb, d.blksize))
1007                         goto err;
1008 #endif
1009         } else {
1010                 sb->s_blocksize = PAGE_CACHE_SIZE;
1011                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1012         }
1013         sb->s_magic = FUSE_SUPER_MAGIC;
1014         sb->s_op = &fuse_super_operations;
1015         sb->s_maxbytes = MAX_LFS_FILESIZE;
1016         sb->s_time_gran = 1;
1017         sb->s_export_op = &fuse_export_operations;
1018
1019         file = fget(d.fd);
1020         err = -EINVAL;
1021         if (!file)
1022                 goto err;
1023
1024         if ((file->f_op != &fuse_dev_operations) ||
1025             (file->f_cred->user_ns != &init_user_ns))
1026                 goto err_fput;
1027
1028         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1029         err = -ENOMEM;
1030         if (!fc)
1031                 goto err_fput;
1032
1033         fuse_conn_init(fc);
1034
1035         fc->dev = sb->s_dev;
1036         fc->sb = sb;
1037         err = fuse_bdi_init(fc, sb);
1038         if (err)
1039                 goto err_put_conn;
1040
1041         sb->s_bdi = &fc->bdi;
1042
1043         /* Handle umasking inside the fuse code */
1044         if (sb->s_flags & MS_POSIXACL)
1045                 fc->dont_mask = 1;
1046         sb->s_flags |= MS_POSIXACL;
1047
1048         fc->release = fuse_free_conn;
1049         fc->flags = d.flags;
1050         fc->user_id = d.user_id;
1051         fc->group_id = d.group_id;
1052         fc->max_read = max_t(unsigned, 4096, d.max_read);
1053
1054         /* Used by get_root_inode() */
1055         sb->s_fs_info = fc;
1056
1057         err = -ENOMEM;
1058         root = fuse_get_root_inode(sb, d.rootmode);
1059         root_dentry = d_make_root(root);
1060         if (!root_dentry)
1061                 goto err_put_conn;
1062         /* only now - we want root dentry with NULL ->d_op */
1063         sb->s_d_op = &fuse_dentry_operations;
1064
1065         init_req = fuse_request_alloc(0);
1066         if (!init_req)
1067                 goto err_put_root;
1068         init_req->background = 1;
1069
1070         if (is_bdev) {
1071                 fc->destroy_req = fuse_request_alloc(0);
1072                 if (!fc->destroy_req)
1073                         goto err_free_init_req;
1074         }
1075
1076         mutex_lock(&fuse_mutex);
1077         err = -EINVAL;
1078         if (file->private_data)
1079                 goto err_unlock;
1080
1081         err = fuse_ctl_add_conn(fc);
1082         if (err)
1083                 goto err_unlock;
1084
1085         list_add_tail(&fc->entry, &fuse_conn_list);
1086         sb->s_root = root_dentry;
1087         fc->connected = 1;
1088         file->private_data = fuse_conn_get(fc);
1089         mutex_unlock(&fuse_mutex);
1090         /*
1091          * atomic_dec_and_test() in fput() provides the necessary
1092          * memory barrier for file->private_data to be visible on all
1093          * CPUs after this
1094          */
1095         fput(file);
1096
1097         fuse_send_init(fc, init_req);
1098
1099         return 0;
1100
1101  err_unlock:
1102         mutex_unlock(&fuse_mutex);
1103  err_free_init_req:
1104         fuse_request_free(init_req);
1105  err_put_root:
1106         dput(root_dentry);
1107  err_put_conn:
1108         fuse_bdi_destroy(fc);
1109         fuse_conn_put(fc);
1110  err_fput:
1111         fput(file);
1112  err:
1113         return err;
1114 }
1115
1116 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1117                        int flags, const char *dev_name,
1118                        void *raw_data)
1119 {
1120         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1121 }
1122
1123 static void fuse_kill_sb_anon(struct super_block *sb)
1124 {
1125         struct fuse_conn *fc = get_fuse_conn_super(sb);
1126
1127         if (fc) {
1128                 down_write(&fc->killsb);
1129                 fc->sb = NULL;
1130                 up_write(&fc->killsb);
1131         }
1132
1133         kill_anon_super(sb);
1134 }
1135
1136 static struct file_system_type fuse_fs_type = {
1137         .owner          = THIS_MODULE,
1138         .name           = "fuse",
1139         .fs_flags       = FS_HAS_SUBTYPE,
1140         .mount          = fuse_mount,
1141         .kill_sb        = fuse_kill_sb_anon,
1142 };
1143 MODULE_ALIAS_FS("fuse");
1144
1145 #ifdef CONFIG_BLOCK
1146 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1147                            int flags, const char *dev_name,
1148                            void *raw_data)
1149 {
1150         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1151 }
1152
1153 static void fuse_kill_sb_blk(struct super_block *sb)
1154 {
1155         struct fuse_conn *fc = get_fuse_conn_super(sb);
1156
1157         if (fc) {
1158                 down_write(&fc->killsb);
1159                 fc->sb = NULL;
1160                 up_write(&fc->killsb);
1161         }
1162
1163         kill_block_super(sb);
1164 }
1165
1166 static struct file_system_type fuseblk_fs_type = {
1167         .owner          = THIS_MODULE,
1168         .name           = "fuseblk",
1169         .mount          = fuse_mount_blk,
1170         .kill_sb        = fuse_kill_sb_blk,
1171         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1172 };
1173 MODULE_ALIAS_FS("fuseblk");
1174
1175 static inline int register_fuseblk(void)
1176 {
1177         return register_filesystem(&fuseblk_fs_type);
1178 }
1179
1180 static inline void unregister_fuseblk(void)
1181 {
1182         unregister_filesystem(&fuseblk_fs_type);
1183 }
1184 #else
1185 static inline int register_fuseblk(void)
1186 {
1187         return 0;
1188 }
1189
1190 static inline void unregister_fuseblk(void)
1191 {
1192 }
1193 #endif
1194
1195 static void fuse_inode_init_once(void *foo)
1196 {
1197         struct inode *inode = foo;
1198
1199         inode_init_once(inode);
1200 }
1201
1202 static int __init fuse_fs_init(void)
1203 {
1204         int err;
1205
1206         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1207                                               sizeof(struct fuse_inode),
1208                                               0, SLAB_HWCACHE_ALIGN,
1209                                               fuse_inode_init_once);
1210         err = -ENOMEM;
1211         if (!fuse_inode_cachep)
1212                 goto out;
1213
1214         err = register_fuseblk();
1215         if (err)
1216                 goto out2;
1217
1218         err = register_filesystem(&fuse_fs_type);
1219         if (err)
1220                 goto out3;
1221
1222         return 0;
1223
1224  out3:
1225         unregister_fuseblk();
1226  out2:
1227         kmem_cache_destroy(fuse_inode_cachep);
1228  out:
1229         return err;
1230 }
1231
1232 static void fuse_fs_cleanup(void)
1233 {
1234         unregister_filesystem(&fuse_fs_type);
1235         unregister_fuseblk();
1236
1237         /*
1238          * Make sure all delayed rcu free inodes are flushed before we
1239          * destroy cache.
1240          */
1241         rcu_barrier();
1242         kmem_cache_destroy(fuse_inode_cachep);
1243 }
1244
1245 static struct kobject *fuse_kobj;
1246 static struct kobject *connections_kobj;
1247
1248 static int fuse_sysfs_init(void)
1249 {
1250         int err;
1251
1252         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1253         if (!fuse_kobj) {
1254                 err = -ENOMEM;
1255                 goto out_err;
1256         }
1257
1258         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1259         if (!connections_kobj) {
1260                 err = -ENOMEM;
1261                 goto out_fuse_unregister;
1262         }
1263
1264         return 0;
1265
1266  out_fuse_unregister:
1267         kobject_put(fuse_kobj);
1268  out_err:
1269         return err;
1270 }
1271
1272 static void fuse_sysfs_cleanup(void)
1273 {
1274         kobject_put(connections_kobj);
1275         kobject_put(fuse_kobj);
1276 }
1277
1278 static int __init fuse_init(void)
1279 {
1280         int res;
1281
1282         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1283                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1284
1285         INIT_LIST_HEAD(&fuse_conn_list);
1286         res = fuse_fs_init();
1287         if (res)
1288                 goto err;
1289
1290         res = fuse_dev_init();
1291         if (res)
1292                 goto err_fs_cleanup;
1293
1294         res = fuse_sysfs_init();
1295         if (res)
1296                 goto err_dev_cleanup;
1297
1298         res = fuse_ctl_init();
1299         if (res)
1300                 goto err_sysfs_cleanup;
1301
1302         sanitize_global_limit(&max_user_bgreq);
1303         sanitize_global_limit(&max_user_congthresh);
1304
1305         return 0;
1306
1307  err_sysfs_cleanup:
1308         fuse_sysfs_cleanup();
1309  err_dev_cleanup:
1310         fuse_dev_cleanup();
1311  err_fs_cleanup:
1312         fuse_fs_cleanup();
1313  err:
1314         return res;
1315 }
1316
1317 static void __exit fuse_exit(void)
1318 {
1319         printk(KERN_DEBUG "fuse exit\n");
1320
1321         fuse_ctl_cleanup();
1322         fuse_sysfs_cleanup();
1323         fuse_fs_cleanup();
1324         fuse_dev_cleanup();
1325 }
1326
1327 module_init(fuse_init);
1328 module_exit(fuse_exit);