Merge tag 'trace-fixes-v3.15-rc4-v2' of git://git.kernel.org/pub/scm/linux/kernel...
[firefly-linux-kernel-4.4.55.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         sync_filesystem(sb);
139         if (*flags & MS_MANDLOCK)
140                 return -EINVAL;
141
142         return 0;
143 }
144
145 /*
146  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
147  * so that it will fit.
148  */
149 static ino_t fuse_squash_ino(u64 ino64)
150 {
151         ino_t ino = (ino_t) ino64;
152         if (sizeof(ino_t) < sizeof(u64))
153                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
154         return ino;
155 }
156
157 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
158                                    u64 attr_valid)
159 {
160         struct fuse_conn *fc = get_fuse_conn(inode);
161         struct fuse_inode *fi = get_fuse_inode(inode);
162
163         fi->attr_version = ++fc->attr_version;
164         fi->i_time = attr_valid;
165
166         inode->i_ino     = fuse_squash_ino(attr->ino);
167         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
168         set_nlink(inode, attr->nlink);
169         inode->i_uid     = make_kuid(&init_user_ns, attr->uid);
170         inode->i_gid     = make_kgid(&init_user_ns, attr->gid);
171         inode->i_blocks  = attr->blocks;
172         inode->i_atime.tv_sec   = attr->atime;
173         inode->i_atime.tv_nsec  = attr->atimensec;
174         /* mtime from server may be stale due to local buffered write */
175         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
176                 inode->i_mtime.tv_sec   = attr->mtime;
177                 inode->i_mtime.tv_nsec  = attr->mtimensec;
178                 inode->i_ctime.tv_sec   = attr->ctime;
179                 inode->i_ctime.tv_nsec  = attr->ctimensec;
180         }
181
182         if (attr->blksize != 0)
183                 inode->i_blkbits = ilog2(attr->blksize);
184         else
185                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
186
187         /*
188          * Don't set the sticky bit in i_mode, unless we want the VFS
189          * to check permissions.  This prevents failures due to the
190          * check in may_delete().
191          */
192         fi->orig_i_mode = inode->i_mode;
193         if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
194                 inode->i_mode &= ~S_ISVTX;
195
196         fi->orig_ino = attr->ino;
197 }
198
199 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
200                             u64 attr_valid, u64 attr_version)
201 {
202         struct fuse_conn *fc = get_fuse_conn(inode);
203         struct fuse_inode *fi = get_fuse_inode(inode);
204         bool is_wb = fc->writeback_cache;
205         loff_t oldsize;
206         struct timespec old_mtime;
207
208         spin_lock(&fc->lock);
209         if ((attr_version != 0 && fi->attr_version > attr_version) ||
210             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
211                 spin_unlock(&fc->lock);
212                 return;
213         }
214
215         old_mtime = inode->i_mtime;
216         fuse_change_attributes_common(inode, attr, attr_valid);
217
218         oldsize = inode->i_size;
219         /*
220          * In case of writeback_cache enabled, the cached writes beyond EOF
221          * extend local i_size without keeping userspace server in sync. So,
222          * attr->size coming from server can be stale. We cannot trust it.
223          */
224         if (!is_wb || !S_ISREG(inode->i_mode))
225                 i_size_write(inode, attr->size);
226         spin_unlock(&fc->lock);
227
228         if (!is_wb && S_ISREG(inode->i_mode)) {
229                 bool inval = false;
230
231                 if (oldsize != attr->size) {
232                         truncate_pagecache(inode, attr->size);
233                         inval = true;
234                 } else if (fc->auto_inval_data) {
235                         struct timespec new_mtime = {
236                                 .tv_sec = attr->mtime,
237                                 .tv_nsec = attr->mtimensec,
238                         };
239
240                         /*
241                          * Auto inval mode also checks and invalidates if mtime
242                          * has changed.
243                          */
244                         if (!timespec_equal(&old_mtime, &new_mtime))
245                                 inval = true;
246                 }
247
248                 if (inval)
249                         invalidate_inode_pages2(inode->i_mapping);
250         }
251 }
252
253 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
254 {
255         inode->i_mode = attr->mode & S_IFMT;
256         inode->i_size = attr->size;
257         inode->i_mtime.tv_sec  = attr->mtime;
258         inode->i_mtime.tv_nsec = attr->mtimensec;
259         inode->i_ctime.tv_sec  = attr->ctime;
260         inode->i_ctime.tv_nsec = attr->ctimensec;
261         if (S_ISREG(inode->i_mode)) {
262                 fuse_init_common(inode);
263                 fuse_init_file_inode(inode);
264         } else if (S_ISDIR(inode->i_mode))
265                 fuse_init_dir(inode);
266         else if (S_ISLNK(inode->i_mode))
267                 fuse_init_symlink(inode);
268         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
269                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
270                 fuse_init_common(inode);
271                 init_special_inode(inode, inode->i_mode,
272                                    new_decode_dev(attr->rdev));
273         } else
274                 BUG();
275 }
276
277 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
278 {
279         u64 nodeid = *(u64 *) _nodeidp;
280         if (get_node_id(inode) == nodeid)
281                 return 1;
282         else
283                 return 0;
284 }
285
286 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
287 {
288         u64 nodeid = *(u64 *) _nodeidp;
289         get_fuse_inode(inode)->nodeid = nodeid;
290         return 0;
291 }
292
293 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
294                         int generation, struct fuse_attr *attr,
295                         u64 attr_valid, u64 attr_version)
296 {
297         struct inode *inode;
298         struct fuse_inode *fi;
299         struct fuse_conn *fc = get_fuse_conn_super(sb);
300
301  retry:
302         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
303         if (!inode)
304                 return NULL;
305
306         if ((inode->i_state & I_NEW)) {
307                 inode->i_flags |= S_NOATIME;
308                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
309                         inode->i_flags |= S_NOCMTIME;
310                 inode->i_generation = generation;
311                 inode->i_data.backing_dev_info = &fc->bdi;
312                 fuse_init_inode(inode, attr);
313                 unlock_new_inode(inode);
314         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
315                 /* Inode has changed type, any I/O on the old should fail */
316                 make_bad_inode(inode);
317                 iput(inode);
318                 goto retry;
319         }
320
321         fi = get_fuse_inode(inode);
322         spin_lock(&fc->lock);
323         fi->nlookup++;
324         spin_unlock(&fc->lock);
325         fuse_change_attributes(inode, attr, attr_valid, attr_version);
326
327         return inode;
328 }
329
330 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
331                              loff_t offset, loff_t len)
332 {
333         struct inode *inode;
334         pgoff_t pg_start;
335         pgoff_t pg_end;
336
337         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
338         if (!inode)
339                 return -ENOENT;
340
341         fuse_invalidate_attr(inode);
342         if (offset >= 0) {
343                 pg_start = offset >> PAGE_CACHE_SHIFT;
344                 if (len <= 0)
345                         pg_end = -1;
346                 else
347                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
348                 invalidate_inode_pages2_range(inode->i_mapping,
349                                               pg_start, pg_end);
350         }
351         iput(inode);
352         return 0;
353 }
354
355 static void fuse_umount_begin(struct super_block *sb)
356 {
357         fuse_abort_conn(get_fuse_conn_super(sb));
358 }
359
360 static void fuse_send_destroy(struct fuse_conn *fc)
361 {
362         struct fuse_req *req = fc->destroy_req;
363         if (req && fc->conn_init) {
364                 fc->destroy_req = NULL;
365                 req->in.h.opcode = FUSE_DESTROY;
366                 req->force = 1;
367                 req->background = 0;
368                 fuse_request_send(fc, req);
369                 fuse_put_request(fc, req);
370         }
371 }
372
373 static void fuse_bdi_destroy(struct fuse_conn *fc)
374 {
375         if (fc->bdi_initialized)
376                 bdi_destroy(&fc->bdi);
377 }
378
379 void fuse_conn_kill(struct fuse_conn *fc)
380 {
381         spin_lock(&fc->lock);
382         fc->connected = 0;
383         fc->blocked = 0;
384         fc->initialized = 1;
385         spin_unlock(&fc->lock);
386         /* Flush all readers on this fs */
387         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
388         wake_up_all(&fc->waitq);
389         wake_up_all(&fc->blocked_waitq);
390         wake_up_all(&fc->reserved_req_waitq);
391 }
392 EXPORT_SYMBOL_GPL(fuse_conn_kill);
393
394 static void fuse_put_super(struct super_block *sb)
395 {
396         struct fuse_conn *fc = get_fuse_conn_super(sb);
397
398         fuse_send_destroy(fc);
399
400         fuse_conn_kill(fc);
401         mutex_lock(&fuse_mutex);
402         list_del(&fc->entry);
403         fuse_ctl_remove_conn(fc);
404         mutex_unlock(&fuse_mutex);
405         fuse_bdi_destroy(fc);
406
407         fuse_conn_put(fc);
408 }
409
410 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
411 {
412         stbuf->f_type    = FUSE_SUPER_MAGIC;
413         stbuf->f_bsize   = attr->bsize;
414         stbuf->f_frsize  = attr->frsize;
415         stbuf->f_blocks  = attr->blocks;
416         stbuf->f_bfree   = attr->bfree;
417         stbuf->f_bavail  = attr->bavail;
418         stbuf->f_files   = attr->files;
419         stbuf->f_ffree   = attr->ffree;
420         stbuf->f_namelen = attr->namelen;
421         /* fsid is left zero */
422 }
423
424 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
425 {
426         struct super_block *sb = dentry->d_sb;
427         struct fuse_conn *fc = get_fuse_conn_super(sb);
428         struct fuse_req *req;
429         struct fuse_statfs_out outarg;
430         int err;
431
432         if (!fuse_allow_current_process(fc)) {
433                 buf->f_type = FUSE_SUPER_MAGIC;
434                 return 0;
435         }
436
437         req = fuse_get_req_nopages(fc);
438         if (IS_ERR(req))
439                 return PTR_ERR(req);
440
441         memset(&outarg, 0, sizeof(outarg));
442         req->in.numargs = 0;
443         req->in.h.opcode = FUSE_STATFS;
444         req->in.h.nodeid = get_node_id(dentry->d_inode);
445         req->out.numargs = 1;
446         req->out.args[0].size =
447                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
448         req->out.args[0].value = &outarg;
449         fuse_request_send(fc, req);
450         err = req->out.h.error;
451         if (!err)
452                 convert_fuse_statfs(buf, &outarg.st);
453         fuse_put_request(fc, req);
454         return err;
455 }
456
457 enum {
458         OPT_FD,
459         OPT_ROOTMODE,
460         OPT_USER_ID,
461         OPT_GROUP_ID,
462         OPT_DEFAULT_PERMISSIONS,
463         OPT_ALLOW_OTHER,
464         OPT_MAX_READ,
465         OPT_BLKSIZE,
466         OPT_ERR
467 };
468
469 static const match_table_t tokens = {
470         {OPT_FD,                        "fd=%u"},
471         {OPT_ROOTMODE,                  "rootmode=%o"},
472         {OPT_USER_ID,                   "user_id=%u"},
473         {OPT_GROUP_ID,                  "group_id=%u"},
474         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
475         {OPT_ALLOW_OTHER,               "allow_other"},
476         {OPT_MAX_READ,                  "max_read=%u"},
477         {OPT_BLKSIZE,                   "blksize=%u"},
478         {OPT_ERR,                       NULL}
479 };
480
481 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
482 {
483         char *p;
484         memset(d, 0, sizeof(struct fuse_mount_data));
485         d->max_read = ~0;
486         d->blksize = FUSE_DEFAULT_BLKSIZE;
487
488         while ((p = strsep(&opt, ",")) != NULL) {
489                 int token;
490                 int value;
491                 substring_t args[MAX_OPT_ARGS];
492                 if (!*p)
493                         continue;
494
495                 token = match_token(p, tokens, args);
496                 switch (token) {
497                 case OPT_FD:
498                         if (match_int(&args[0], &value))
499                                 return 0;
500                         d->fd = value;
501                         d->fd_present = 1;
502                         break;
503
504                 case OPT_ROOTMODE:
505                         if (match_octal(&args[0], &value))
506                                 return 0;
507                         if (!fuse_valid_type(value))
508                                 return 0;
509                         d->rootmode = value;
510                         d->rootmode_present = 1;
511                         break;
512
513                 case OPT_USER_ID:
514                         if (match_int(&args[0], &value))
515                                 return 0;
516                         d->user_id = make_kuid(current_user_ns(), value);
517                         if (!uid_valid(d->user_id))
518                                 return 0;
519                         d->user_id_present = 1;
520                         break;
521
522                 case OPT_GROUP_ID:
523                         if (match_int(&args[0], &value))
524                                 return 0;
525                         d->group_id = make_kgid(current_user_ns(), value);
526                         if (!gid_valid(d->group_id))
527                                 return 0;
528                         d->group_id_present = 1;
529                         break;
530
531                 case OPT_DEFAULT_PERMISSIONS:
532                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
533                         break;
534
535                 case OPT_ALLOW_OTHER:
536                         d->flags |= FUSE_ALLOW_OTHER;
537                         break;
538
539                 case OPT_MAX_READ:
540                         if (match_int(&args[0], &value))
541                                 return 0;
542                         d->max_read = value;
543                         break;
544
545                 case OPT_BLKSIZE:
546                         if (!is_bdev || match_int(&args[0], &value))
547                                 return 0;
548                         d->blksize = value;
549                         break;
550
551                 default:
552                         return 0;
553                 }
554         }
555
556         if (!d->fd_present || !d->rootmode_present ||
557             !d->user_id_present || !d->group_id_present)
558                 return 0;
559
560         return 1;
561 }
562
563 static int fuse_show_options(struct seq_file *m, struct dentry *root)
564 {
565         struct super_block *sb = root->d_sb;
566         struct fuse_conn *fc = get_fuse_conn_super(sb);
567
568         seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
569         seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
570         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
571                 seq_puts(m, ",default_permissions");
572         if (fc->flags & FUSE_ALLOW_OTHER)
573                 seq_puts(m, ",allow_other");
574         if (fc->max_read != ~0)
575                 seq_printf(m, ",max_read=%u", fc->max_read);
576         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
577                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
578         return 0;
579 }
580
581 void fuse_conn_init(struct fuse_conn *fc)
582 {
583         memset(fc, 0, sizeof(*fc));
584         spin_lock_init(&fc->lock);
585         init_rwsem(&fc->killsb);
586         atomic_set(&fc->count, 1);
587         init_waitqueue_head(&fc->waitq);
588         init_waitqueue_head(&fc->blocked_waitq);
589         init_waitqueue_head(&fc->reserved_req_waitq);
590         INIT_LIST_HEAD(&fc->pending);
591         INIT_LIST_HEAD(&fc->processing);
592         INIT_LIST_HEAD(&fc->io);
593         INIT_LIST_HEAD(&fc->interrupts);
594         INIT_LIST_HEAD(&fc->bg_queue);
595         INIT_LIST_HEAD(&fc->entry);
596         fc->forget_list_tail = &fc->forget_list_head;
597         atomic_set(&fc->num_waiting, 0);
598         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
599         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
600         fc->khctr = 0;
601         fc->polled_files = RB_ROOT;
602         fc->reqctr = 0;
603         fc->blocked = 0;
604         fc->initialized = 0;
605         fc->attr_version = 1;
606         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
607 }
608 EXPORT_SYMBOL_GPL(fuse_conn_init);
609
610 void fuse_conn_put(struct fuse_conn *fc)
611 {
612         if (atomic_dec_and_test(&fc->count)) {
613                 if (fc->destroy_req)
614                         fuse_request_free(fc->destroy_req);
615                 fc->release(fc);
616         }
617 }
618 EXPORT_SYMBOL_GPL(fuse_conn_put);
619
620 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
621 {
622         atomic_inc(&fc->count);
623         return fc;
624 }
625 EXPORT_SYMBOL_GPL(fuse_conn_get);
626
627 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
628 {
629         struct fuse_attr attr;
630         memset(&attr, 0, sizeof(attr));
631
632         attr.mode = mode;
633         attr.ino = FUSE_ROOT_ID;
634         attr.nlink = 1;
635         return fuse_iget(sb, 1, 0, &attr, 0, 0);
636 }
637
638 struct fuse_inode_handle {
639         u64 nodeid;
640         u32 generation;
641 };
642
643 static struct dentry *fuse_get_dentry(struct super_block *sb,
644                                       struct fuse_inode_handle *handle)
645 {
646         struct fuse_conn *fc = get_fuse_conn_super(sb);
647         struct inode *inode;
648         struct dentry *entry;
649         int err = -ESTALE;
650
651         if (handle->nodeid == 0)
652                 goto out_err;
653
654         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
655         if (!inode) {
656                 struct fuse_entry_out outarg;
657                 struct qstr name;
658
659                 if (!fc->export_support)
660                         goto out_err;
661
662                 name.len = 1;
663                 name.name = ".";
664                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
665                                        &inode);
666                 if (err && err != -ENOENT)
667                         goto out_err;
668                 if (err || !inode) {
669                         err = -ESTALE;
670                         goto out_err;
671                 }
672                 err = -EIO;
673                 if (get_node_id(inode) != handle->nodeid)
674                         goto out_iput;
675         }
676         err = -ESTALE;
677         if (inode->i_generation != handle->generation)
678                 goto out_iput;
679
680         entry = d_obtain_alias(inode);
681         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
682                 fuse_invalidate_entry_cache(entry);
683
684         return entry;
685
686  out_iput:
687         iput(inode);
688  out_err:
689         return ERR_PTR(err);
690 }
691
692 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
693                            struct inode *parent)
694 {
695         int len = parent ? 6 : 3;
696         u64 nodeid;
697         u32 generation;
698
699         if (*max_len < len) {
700                 *max_len = len;
701                 return  FILEID_INVALID;
702         }
703
704         nodeid = get_fuse_inode(inode)->nodeid;
705         generation = inode->i_generation;
706
707         fh[0] = (u32)(nodeid >> 32);
708         fh[1] = (u32)(nodeid & 0xffffffff);
709         fh[2] = generation;
710
711         if (parent) {
712                 nodeid = get_fuse_inode(parent)->nodeid;
713                 generation = parent->i_generation;
714
715                 fh[3] = (u32)(nodeid >> 32);
716                 fh[4] = (u32)(nodeid & 0xffffffff);
717                 fh[5] = generation;
718         }
719
720         *max_len = len;
721         return parent ? 0x82 : 0x81;
722 }
723
724 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
725                 struct fid *fid, int fh_len, int fh_type)
726 {
727         struct fuse_inode_handle handle;
728
729         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
730                 return NULL;
731
732         handle.nodeid = (u64) fid->raw[0] << 32;
733         handle.nodeid |= (u64) fid->raw[1];
734         handle.generation = fid->raw[2];
735         return fuse_get_dentry(sb, &handle);
736 }
737
738 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
739                 struct fid *fid, int fh_len, int fh_type)
740 {
741         struct fuse_inode_handle parent;
742
743         if (fh_type != 0x82 || fh_len < 6)
744                 return NULL;
745
746         parent.nodeid = (u64) fid->raw[3] << 32;
747         parent.nodeid |= (u64) fid->raw[4];
748         parent.generation = fid->raw[5];
749         return fuse_get_dentry(sb, &parent);
750 }
751
752 static struct dentry *fuse_get_parent(struct dentry *child)
753 {
754         struct inode *child_inode = child->d_inode;
755         struct fuse_conn *fc = get_fuse_conn(child_inode);
756         struct inode *inode;
757         struct dentry *parent;
758         struct fuse_entry_out outarg;
759         struct qstr name;
760         int err;
761
762         if (!fc->export_support)
763                 return ERR_PTR(-ESTALE);
764
765         name.len = 2;
766         name.name = "..";
767         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
768                                &name, &outarg, &inode);
769         if (err) {
770                 if (err == -ENOENT)
771                         return ERR_PTR(-ESTALE);
772                 return ERR_PTR(err);
773         }
774
775         parent = d_obtain_alias(inode);
776         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
777                 fuse_invalidate_entry_cache(parent);
778
779         return parent;
780 }
781
782 static const struct export_operations fuse_export_operations = {
783         .fh_to_dentry   = fuse_fh_to_dentry,
784         .fh_to_parent   = fuse_fh_to_parent,
785         .encode_fh      = fuse_encode_fh,
786         .get_parent     = fuse_get_parent,
787 };
788
789 static const struct super_operations fuse_super_operations = {
790         .alloc_inode    = fuse_alloc_inode,
791         .destroy_inode  = fuse_destroy_inode,
792         .evict_inode    = fuse_evict_inode,
793         .write_inode    = fuse_write_inode,
794         .drop_inode     = generic_delete_inode,
795         .remount_fs     = fuse_remount_fs,
796         .put_super      = fuse_put_super,
797         .umount_begin   = fuse_umount_begin,
798         .statfs         = fuse_statfs,
799         .show_options   = fuse_show_options,
800 };
801
802 static void sanitize_global_limit(unsigned *limit)
803 {
804         if (*limit == 0)
805                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
806                          sizeof(struct fuse_req);
807
808         if (*limit >= 1 << 16)
809                 *limit = (1 << 16) - 1;
810 }
811
812 static int set_global_limit(const char *val, struct kernel_param *kp)
813 {
814         int rv;
815
816         rv = param_set_uint(val, kp);
817         if (rv)
818                 return rv;
819
820         sanitize_global_limit((unsigned *)kp->arg);
821
822         return 0;
823 }
824
825 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
826 {
827         int cap_sys_admin = capable(CAP_SYS_ADMIN);
828
829         if (arg->minor < 13)
830                 return;
831
832         sanitize_global_limit(&max_user_bgreq);
833         sanitize_global_limit(&max_user_congthresh);
834
835         if (arg->max_background) {
836                 fc->max_background = arg->max_background;
837
838                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
839                         fc->max_background = max_user_bgreq;
840         }
841         if (arg->congestion_threshold) {
842                 fc->congestion_threshold = arg->congestion_threshold;
843
844                 if (!cap_sys_admin &&
845                     fc->congestion_threshold > max_user_congthresh)
846                         fc->congestion_threshold = max_user_congthresh;
847         }
848 }
849
850 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
851 {
852         struct fuse_init_out *arg = &req->misc.init_out;
853
854         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
855                 fc->conn_error = 1;
856         else {
857                 unsigned long ra_pages;
858
859                 process_init_limits(fc, arg);
860
861                 if (arg->minor >= 6) {
862                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
863                         if (arg->flags & FUSE_ASYNC_READ)
864                                 fc->async_read = 1;
865                         if (!(arg->flags & FUSE_POSIX_LOCKS))
866                                 fc->no_lock = 1;
867                         if (arg->minor >= 17) {
868                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
869                                         fc->no_flock = 1;
870                         } else {
871                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
872                                         fc->no_flock = 1;
873                         }
874                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
875                                 fc->atomic_o_trunc = 1;
876                         if (arg->minor >= 9) {
877                                 /* LOOKUP has dependency on proto version */
878                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
879                                         fc->export_support = 1;
880                         }
881                         if (arg->flags & FUSE_BIG_WRITES)
882                                 fc->big_writes = 1;
883                         if (arg->flags & FUSE_DONT_MASK)
884                                 fc->dont_mask = 1;
885                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
886                                 fc->auto_inval_data = 1;
887                         if (arg->flags & FUSE_DO_READDIRPLUS) {
888                                 fc->do_readdirplus = 1;
889                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
890                                         fc->readdirplus_auto = 1;
891                         }
892                         if (arg->flags & FUSE_ASYNC_DIO)
893                                 fc->async_dio = 1;
894                         if (arg->flags & FUSE_WRITEBACK_CACHE)
895                                 fc->writeback_cache = 1;
896                         if (arg->time_gran && arg->time_gran <= 1000000000)
897                                 fc->sb->s_time_gran = arg->time_gran;
898                         else
899                                 fc->sb->s_time_gran = 1000000000;
900
901                 } else {
902                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
903                         fc->no_lock = 1;
904                         fc->no_flock = 1;
905                 }
906
907                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
908                 fc->minor = arg->minor;
909                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
910                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
911                 fc->conn_init = 1;
912         }
913         fc->initialized = 1;
914         wake_up_all(&fc->blocked_waitq);
915 }
916
917 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
918 {
919         struct fuse_init_in *arg = &req->misc.init_in;
920
921         arg->major = FUSE_KERNEL_VERSION;
922         arg->minor = FUSE_KERNEL_MINOR_VERSION;
923         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
924         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
925                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
926                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
927                 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
928                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
929                 FUSE_WRITEBACK_CACHE;
930         req->in.h.opcode = FUSE_INIT;
931         req->in.numargs = 1;
932         req->in.args[0].size = sizeof(*arg);
933         req->in.args[0].value = arg;
934         req->out.numargs = 1;
935         /* Variable length argument used for backward compatibility
936            with interface version < 7.5.  Rest of init_out is zeroed
937            by do_get_request(), so a short reply is not a problem */
938         req->out.argvar = 1;
939         req->out.args[0].size = sizeof(struct fuse_init_out);
940         req->out.args[0].value = &req->misc.init_out;
941         req->end = process_init_reply;
942         fuse_request_send_background(fc, req);
943 }
944
945 static void fuse_free_conn(struct fuse_conn *fc)
946 {
947         kfree_rcu(fc, rcu);
948 }
949
950 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
951 {
952         int err;
953
954         fc->bdi.name = "fuse";
955         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
956         /* fuse does it's own writeback accounting */
957         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
958
959         err = bdi_init(&fc->bdi);
960         if (err)
961                 return err;
962
963         fc->bdi_initialized = 1;
964
965         if (sb->s_bdev) {
966                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
967                                     MAJOR(fc->dev), MINOR(fc->dev));
968         } else {
969                 err = bdi_register_dev(&fc->bdi, fc->dev);
970         }
971
972         if (err)
973                 return err;
974
975         /*
976          * For a single fuse filesystem use max 1% of dirty +
977          * writeback threshold.
978          *
979          * This gives about 1M of write buffer for memory maps on a
980          * machine with 1G and 10% dirty_ratio, which should be more
981          * than enough.
982          *
983          * Privileged users can raise it by writing to
984          *
985          *    /sys/class/bdi/<bdi>/max_ratio
986          */
987         bdi_set_max_ratio(&fc->bdi, 1);
988
989         return 0;
990 }
991
992 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
993 {
994         struct fuse_conn *fc;
995         struct inode *root;
996         struct fuse_mount_data d;
997         struct file *file;
998         struct dentry *root_dentry;
999         struct fuse_req *init_req;
1000         int err;
1001         int is_bdev = sb->s_bdev != NULL;
1002
1003         err = -EINVAL;
1004         if (sb->s_flags & MS_MANDLOCK)
1005                 goto err;
1006
1007         sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
1008
1009         if (!parse_fuse_opt((char *) data, &d, is_bdev))
1010                 goto err;
1011
1012         if (is_bdev) {
1013 #ifdef CONFIG_BLOCK
1014                 err = -EINVAL;
1015                 if (!sb_set_blocksize(sb, d.blksize))
1016                         goto err;
1017 #endif
1018         } else {
1019                 sb->s_blocksize = PAGE_CACHE_SIZE;
1020                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1021         }
1022         sb->s_magic = FUSE_SUPER_MAGIC;
1023         sb->s_op = &fuse_super_operations;
1024         sb->s_maxbytes = MAX_LFS_FILESIZE;
1025         sb->s_time_gran = 1;
1026         sb->s_export_op = &fuse_export_operations;
1027
1028         file = fget(d.fd);
1029         err = -EINVAL;
1030         if (!file)
1031                 goto err;
1032
1033         if ((file->f_op != &fuse_dev_operations) ||
1034             (file->f_cred->user_ns != &init_user_ns))
1035                 goto err_fput;
1036
1037         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1038         err = -ENOMEM;
1039         if (!fc)
1040                 goto err_fput;
1041
1042         fuse_conn_init(fc);
1043
1044         fc->dev = sb->s_dev;
1045         fc->sb = sb;
1046         err = fuse_bdi_init(fc, sb);
1047         if (err)
1048                 goto err_put_conn;
1049
1050         sb->s_bdi = &fc->bdi;
1051
1052         /* Handle umasking inside the fuse code */
1053         if (sb->s_flags & MS_POSIXACL)
1054                 fc->dont_mask = 1;
1055         sb->s_flags |= MS_POSIXACL;
1056
1057         fc->release = fuse_free_conn;
1058         fc->flags = d.flags;
1059         fc->user_id = d.user_id;
1060         fc->group_id = d.group_id;
1061         fc->max_read = max_t(unsigned, 4096, d.max_read);
1062
1063         /* Used by get_root_inode() */
1064         sb->s_fs_info = fc;
1065
1066         err = -ENOMEM;
1067         root = fuse_get_root_inode(sb, d.rootmode);
1068         root_dentry = d_make_root(root);
1069         if (!root_dentry)
1070                 goto err_put_conn;
1071         /* only now - we want root dentry with NULL ->d_op */
1072         sb->s_d_op = &fuse_dentry_operations;
1073
1074         init_req = fuse_request_alloc(0);
1075         if (!init_req)
1076                 goto err_put_root;
1077         init_req->background = 1;
1078
1079         if (is_bdev) {
1080                 fc->destroy_req = fuse_request_alloc(0);
1081                 if (!fc->destroy_req)
1082                         goto err_free_init_req;
1083         }
1084
1085         mutex_lock(&fuse_mutex);
1086         err = -EINVAL;
1087         if (file->private_data)
1088                 goto err_unlock;
1089
1090         err = fuse_ctl_add_conn(fc);
1091         if (err)
1092                 goto err_unlock;
1093
1094         list_add_tail(&fc->entry, &fuse_conn_list);
1095         sb->s_root = root_dentry;
1096         fc->connected = 1;
1097         file->private_data = fuse_conn_get(fc);
1098         mutex_unlock(&fuse_mutex);
1099         /*
1100          * atomic_dec_and_test() in fput() provides the necessary
1101          * memory barrier for file->private_data to be visible on all
1102          * CPUs after this
1103          */
1104         fput(file);
1105
1106         fuse_send_init(fc, init_req);
1107
1108         return 0;
1109
1110  err_unlock:
1111         mutex_unlock(&fuse_mutex);
1112  err_free_init_req:
1113         fuse_request_free(init_req);
1114  err_put_root:
1115         dput(root_dentry);
1116  err_put_conn:
1117         fuse_bdi_destroy(fc);
1118         fuse_conn_put(fc);
1119  err_fput:
1120         fput(file);
1121  err:
1122         return err;
1123 }
1124
1125 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1126                        int flags, const char *dev_name,
1127                        void *raw_data)
1128 {
1129         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1130 }
1131
1132 static void fuse_kill_sb_anon(struct super_block *sb)
1133 {
1134         struct fuse_conn *fc = get_fuse_conn_super(sb);
1135
1136         if (fc) {
1137                 down_write(&fc->killsb);
1138                 fc->sb = NULL;
1139                 up_write(&fc->killsb);
1140         }
1141
1142         kill_anon_super(sb);
1143 }
1144
1145 static struct file_system_type fuse_fs_type = {
1146         .owner          = THIS_MODULE,
1147         .name           = "fuse",
1148         .fs_flags       = FS_HAS_SUBTYPE,
1149         .mount          = fuse_mount,
1150         .kill_sb        = fuse_kill_sb_anon,
1151 };
1152 MODULE_ALIAS_FS("fuse");
1153
1154 #ifdef CONFIG_BLOCK
1155 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1156                            int flags, const char *dev_name,
1157                            void *raw_data)
1158 {
1159         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1160 }
1161
1162 static void fuse_kill_sb_blk(struct super_block *sb)
1163 {
1164         struct fuse_conn *fc = get_fuse_conn_super(sb);
1165
1166         if (fc) {
1167                 down_write(&fc->killsb);
1168                 fc->sb = NULL;
1169                 up_write(&fc->killsb);
1170         }
1171
1172         kill_block_super(sb);
1173 }
1174
1175 static struct file_system_type fuseblk_fs_type = {
1176         .owner          = THIS_MODULE,
1177         .name           = "fuseblk",
1178         .mount          = fuse_mount_blk,
1179         .kill_sb        = fuse_kill_sb_blk,
1180         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1181 };
1182 MODULE_ALIAS_FS("fuseblk");
1183
1184 static inline int register_fuseblk(void)
1185 {
1186         return register_filesystem(&fuseblk_fs_type);
1187 }
1188
1189 static inline void unregister_fuseblk(void)
1190 {
1191         unregister_filesystem(&fuseblk_fs_type);
1192 }
1193 #else
1194 static inline int register_fuseblk(void)
1195 {
1196         return 0;
1197 }
1198
1199 static inline void unregister_fuseblk(void)
1200 {
1201 }
1202 #endif
1203
1204 static void fuse_inode_init_once(void *foo)
1205 {
1206         struct inode *inode = foo;
1207
1208         inode_init_once(inode);
1209 }
1210
1211 static int __init fuse_fs_init(void)
1212 {
1213         int err;
1214
1215         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1216                                               sizeof(struct fuse_inode),
1217                                               0, SLAB_HWCACHE_ALIGN,
1218                                               fuse_inode_init_once);
1219         err = -ENOMEM;
1220         if (!fuse_inode_cachep)
1221                 goto out;
1222
1223         err = register_fuseblk();
1224         if (err)
1225                 goto out2;
1226
1227         err = register_filesystem(&fuse_fs_type);
1228         if (err)
1229                 goto out3;
1230
1231         return 0;
1232
1233  out3:
1234         unregister_fuseblk();
1235  out2:
1236         kmem_cache_destroy(fuse_inode_cachep);
1237  out:
1238         return err;
1239 }
1240
1241 static void fuse_fs_cleanup(void)
1242 {
1243         unregister_filesystem(&fuse_fs_type);
1244         unregister_fuseblk();
1245
1246         /*
1247          * Make sure all delayed rcu free inodes are flushed before we
1248          * destroy cache.
1249          */
1250         rcu_barrier();
1251         kmem_cache_destroy(fuse_inode_cachep);
1252 }
1253
1254 static struct kobject *fuse_kobj;
1255 static struct kobject *connections_kobj;
1256
1257 static int fuse_sysfs_init(void)
1258 {
1259         int err;
1260
1261         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1262         if (!fuse_kobj) {
1263                 err = -ENOMEM;
1264                 goto out_err;
1265         }
1266
1267         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1268         if (!connections_kobj) {
1269                 err = -ENOMEM;
1270                 goto out_fuse_unregister;
1271         }
1272
1273         return 0;
1274
1275  out_fuse_unregister:
1276         kobject_put(fuse_kobj);
1277  out_err:
1278         return err;
1279 }
1280
1281 static void fuse_sysfs_cleanup(void)
1282 {
1283         kobject_put(connections_kobj);
1284         kobject_put(fuse_kobj);
1285 }
1286
1287 static int __init fuse_init(void)
1288 {
1289         int res;
1290
1291         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1292                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1293
1294         INIT_LIST_HEAD(&fuse_conn_list);
1295         res = fuse_fs_init();
1296         if (res)
1297                 goto err;
1298
1299         res = fuse_dev_init();
1300         if (res)
1301                 goto err_fs_cleanup;
1302
1303         res = fuse_sysfs_init();
1304         if (res)
1305                 goto err_dev_cleanup;
1306
1307         res = fuse_ctl_init();
1308         if (res)
1309                 goto err_sysfs_cleanup;
1310
1311         sanitize_global_limit(&max_user_bgreq);
1312         sanitize_global_limit(&max_user_congthresh);
1313
1314         return 0;
1315
1316  err_sysfs_cleanup:
1317         fuse_sysfs_cleanup();
1318  err_dev_cleanup:
1319         fuse_dev_cleanup();
1320  err_fs_cleanup:
1321         fuse_fs_cleanup();
1322  err:
1323         return res;
1324 }
1325
1326 static void __exit fuse_exit(void)
1327 {
1328         printk(KERN_DEBUG "fuse exit\n");
1329
1330         fuse_ctl_cleanup();
1331         fuse_sysfs_cleanup();
1332         fuse_fs_cleanup();
1333         fuse_dev_cleanup();
1334 }
1335
1336 module_init(fuse_init);
1337 module_exit(fuse_exit);