GNU Linux-libre 4.4.284-gnu1
[releases.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                 fuse_init_inode(inode, attr);
312                 unlock_new_inode(inode);
313         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
314                 /* Inode has changed type, any I/O on the old should fail */
315                 make_bad_inode(inode);
316                 iput(inode);
317                 goto retry;
318         }
319
320         fi = get_fuse_inode(inode);
321         spin_lock(&fc->lock);
322         fi->nlookup++;
323         spin_unlock(&fc->lock);
324         fuse_change_attributes(inode, attr, attr_valid, attr_version);
325
326         return inode;
327 }
328
329 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
330                              loff_t offset, loff_t len)
331 {
332         struct inode *inode;
333         pgoff_t pg_start;
334         pgoff_t pg_end;
335
336         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
337         if (!inode)
338                 return -ENOENT;
339
340         fuse_invalidate_attr(inode);
341         if (offset >= 0) {
342                 pg_start = offset >> PAGE_CACHE_SHIFT;
343                 if (len <= 0)
344                         pg_end = -1;
345                 else
346                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
347                 invalidate_inode_pages2_range(inode->i_mapping,
348                                               pg_start, pg_end);
349         }
350         iput(inode);
351         return 0;
352 }
353
354 static void fuse_umount_begin(struct super_block *sb)
355 {
356         fuse_abort_conn(get_fuse_conn_super(sb));
357 }
358
359 static void fuse_send_destroy(struct fuse_conn *fc)
360 {
361         struct fuse_req *req = fc->destroy_req;
362         if (req && fc->conn_init) {
363                 fc->destroy_req = NULL;
364                 req->in.h.opcode = FUSE_DESTROY;
365                 __set_bit(FR_FORCE, &req->flags);
366                 __clear_bit(FR_BACKGROUND, &req->flags);
367                 fuse_request_send(fc, req);
368                 fuse_put_request(fc, req);
369         }
370 }
371
372 static void fuse_bdi_destroy(struct fuse_conn *fc)
373 {
374         if (fc->bdi_initialized)
375                 bdi_destroy(&fc->bdi);
376 }
377
378 static void fuse_put_super(struct super_block *sb)
379 {
380         struct fuse_conn *fc = get_fuse_conn_super(sb);
381
382         mutex_lock(&fuse_mutex);
383         list_del(&fc->entry);
384         fuse_ctl_remove_conn(fc);
385         mutex_unlock(&fuse_mutex);
386         fuse_bdi_destroy(fc);
387
388         fuse_conn_put(fc);
389 }
390
391 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
392 {
393         stbuf->f_type    = FUSE_SUPER_MAGIC;
394         stbuf->f_bsize   = attr->bsize;
395         stbuf->f_frsize  = attr->frsize;
396         stbuf->f_blocks  = attr->blocks;
397         stbuf->f_bfree   = attr->bfree;
398         stbuf->f_bavail  = attr->bavail;
399         stbuf->f_files   = attr->files;
400         stbuf->f_ffree   = attr->ffree;
401         stbuf->f_namelen = attr->namelen;
402         /* fsid is left zero */
403 }
404
405 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
406 {
407         struct super_block *sb = dentry->d_sb;
408         struct fuse_conn *fc = get_fuse_conn_super(sb);
409         FUSE_ARGS(args);
410         struct fuse_statfs_out outarg;
411         int err;
412
413         if (!fuse_allow_current_process(fc)) {
414                 buf->f_type = FUSE_SUPER_MAGIC;
415                 return 0;
416         }
417
418         memset(&outarg, 0, sizeof(outarg));
419         args.in.numargs = 0;
420         args.in.h.opcode = FUSE_STATFS;
421         args.in.h.nodeid = get_node_id(d_inode(dentry));
422         args.out.numargs = 1;
423         args.out.args[0].size = sizeof(outarg);
424         args.out.args[0].value = &outarg;
425         err = fuse_simple_request(fc, &args);
426         if (!err)
427                 convert_fuse_statfs(buf, &outarg.st);
428         return err;
429 }
430
431 enum {
432         OPT_FD,
433         OPT_ROOTMODE,
434         OPT_USER_ID,
435         OPT_GROUP_ID,
436         OPT_DEFAULT_PERMISSIONS,
437         OPT_ALLOW_OTHER,
438         OPT_MAX_READ,
439         OPT_BLKSIZE,
440         OPT_ERR
441 };
442
443 static const match_table_t tokens = {
444         {OPT_FD,                        "fd=%u"},
445         {OPT_ROOTMODE,                  "rootmode=%o"},
446         {OPT_USER_ID,                   "user_id=%u"},
447         {OPT_GROUP_ID,                  "group_id=%u"},
448         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
449         {OPT_ALLOW_OTHER,               "allow_other"},
450         {OPT_MAX_READ,                  "max_read=%u"},
451         {OPT_BLKSIZE,                   "blksize=%u"},
452         {OPT_ERR,                       NULL}
453 };
454
455 static int fuse_match_uint(substring_t *s, unsigned int *res)
456 {
457         int err = -ENOMEM;
458         char *buf = match_strdup(s);
459         if (buf) {
460                 err = kstrtouint(buf, 10, res);
461                 kfree(buf);
462         }
463         return err;
464 }
465
466 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
467 {
468         char *p;
469         memset(d, 0, sizeof(struct fuse_mount_data));
470         d->max_read = ~0;
471         d->blksize = FUSE_DEFAULT_BLKSIZE;
472
473         while ((p = strsep(&opt, ",")) != NULL) {
474                 int token;
475                 int value;
476                 unsigned uv;
477                 substring_t args[MAX_OPT_ARGS];
478                 if (!*p)
479                         continue;
480
481                 token = match_token(p, tokens, args);
482                 switch (token) {
483                 case OPT_FD:
484                         if (match_int(&args[0], &value))
485                                 return 0;
486                         d->fd = value;
487                         d->fd_present = 1;
488                         break;
489
490                 case OPT_ROOTMODE:
491                         if (match_octal(&args[0], &value))
492                                 return 0;
493                         if (!fuse_valid_type(value))
494                                 return 0;
495                         d->rootmode = value;
496                         d->rootmode_present = 1;
497                         break;
498
499                 case OPT_USER_ID:
500                         if (fuse_match_uint(&args[0], &uv))
501                                 return 0;
502                         d->user_id = make_kuid(current_user_ns(), uv);
503                         if (!uid_valid(d->user_id))
504                                 return 0;
505                         d->user_id_present = 1;
506                         break;
507
508                 case OPT_GROUP_ID:
509                         if (fuse_match_uint(&args[0], &uv))
510                                 return 0;
511                         d->group_id = make_kgid(current_user_ns(), uv);
512                         if (!gid_valid(d->group_id))
513                                 return 0;
514                         d->group_id_present = 1;
515                         break;
516
517                 case OPT_DEFAULT_PERMISSIONS:
518                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
519                         break;
520
521                 case OPT_ALLOW_OTHER:
522                         d->flags |= FUSE_ALLOW_OTHER;
523                         break;
524
525                 case OPT_MAX_READ:
526                         if (match_int(&args[0], &value))
527                                 return 0;
528                         d->max_read = value;
529                         break;
530
531                 case OPT_BLKSIZE:
532                         if (!is_bdev || match_int(&args[0], &value))
533                                 return 0;
534                         d->blksize = value;
535                         break;
536
537                 default:
538                         return 0;
539                 }
540         }
541
542         if (!d->fd_present || !d->rootmode_present ||
543             !d->user_id_present || !d->group_id_present)
544                 return 0;
545
546         return 1;
547 }
548
549 static int fuse_show_options(struct seq_file *m, struct dentry *root)
550 {
551         struct super_block *sb = root->d_sb;
552         struct fuse_conn *fc = get_fuse_conn_super(sb);
553
554         seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
555         seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
556         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
557                 seq_puts(m, ",default_permissions");
558         if (fc->flags & FUSE_ALLOW_OTHER)
559                 seq_puts(m, ",allow_other");
560         if (fc->max_read != ~0)
561                 seq_printf(m, ",max_read=%u", fc->max_read);
562         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
563                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
564         return 0;
565 }
566
567 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
568 {
569         memset(fiq, 0, sizeof(struct fuse_iqueue));
570         init_waitqueue_head(&fiq->waitq);
571         INIT_LIST_HEAD(&fiq->pending);
572         INIT_LIST_HEAD(&fiq->interrupts);
573         fiq->forget_list_tail = &fiq->forget_list_head;
574         fiq->connected = 1;
575 }
576
577 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
578 {
579         memset(fpq, 0, sizeof(struct fuse_pqueue));
580         spin_lock_init(&fpq->lock);
581         INIT_LIST_HEAD(&fpq->processing);
582         INIT_LIST_HEAD(&fpq->io);
583         fpq->connected = 1;
584 }
585
586 void fuse_conn_init(struct fuse_conn *fc)
587 {
588         memset(fc, 0, sizeof(*fc));
589         spin_lock_init(&fc->lock);
590         init_rwsem(&fc->killsb);
591         atomic_set(&fc->count, 1);
592         atomic_set(&fc->dev_count, 1);
593         init_waitqueue_head(&fc->blocked_waitq);
594         init_waitqueue_head(&fc->reserved_req_waitq);
595         fuse_iqueue_init(&fc->iq);
596         INIT_LIST_HEAD(&fc->bg_queue);
597         INIT_LIST_HEAD(&fc->entry);
598         INIT_LIST_HEAD(&fc->devices);
599         atomic_set(&fc->num_waiting, 0);
600         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
601         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
602         fc->khctr = 0;
603         fc->polled_files = RB_ROOT;
604         fc->blocked = 0;
605         fc->initialized = 0;
606         fc->connected = 1;
607         fc->attr_version = 1;
608         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
609 }
610 EXPORT_SYMBOL_GPL(fuse_conn_init);
611
612 void fuse_conn_put(struct fuse_conn *fc)
613 {
614         if (atomic_dec_and_test(&fc->count)) {
615                 if (fc->destroy_req)
616                         fuse_request_free(fc->destroy_req);
617                 fc->release(fc);
618         }
619 }
620 EXPORT_SYMBOL_GPL(fuse_conn_put);
621
622 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
623 {
624         atomic_inc(&fc->count);
625         return fc;
626 }
627 EXPORT_SYMBOL_GPL(fuse_conn_get);
628
629 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
630 {
631         struct fuse_attr attr;
632         memset(&attr, 0, sizeof(attr));
633
634         attr.mode = mode;
635         attr.ino = FUSE_ROOT_ID;
636         attr.nlink = 1;
637         return fuse_iget(sb, 1, 0, &attr, 0, 0);
638 }
639
640 struct fuse_inode_handle {
641         u64 nodeid;
642         u32 generation;
643 };
644
645 static struct dentry *fuse_get_dentry(struct super_block *sb,
646                                       struct fuse_inode_handle *handle)
647 {
648         struct fuse_conn *fc = get_fuse_conn_super(sb);
649         struct inode *inode;
650         struct dentry *entry;
651         int err = -ESTALE;
652
653         if (handle->nodeid == 0)
654                 goto out_err;
655
656         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
657         if (!inode) {
658                 struct fuse_entry_out outarg;
659                 struct qstr name;
660
661                 if (!fc->export_support)
662                         goto out_err;
663
664                 name.len = 1;
665                 name.name = ".";
666                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
667                                        &inode);
668                 if (err && err != -ENOENT)
669                         goto out_err;
670                 if (err || !inode) {
671                         err = -ESTALE;
672                         goto out_err;
673                 }
674                 err = -EIO;
675                 if (get_node_id(inode) != handle->nodeid)
676                         goto out_iput;
677         }
678         err = -ESTALE;
679         if (inode->i_generation != handle->generation)
680                 goto out_iput;
681
682         entry = d_obtain_alias(inode);
683         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
684                 fuse_invalidate_entry_cache(entry);
685
686         return entry;
687
688  out_iput:
689         iput(inode);
690  out_err:
691         return ERR_PTR(err);
692 }
693
694 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
695                            struct inode *parent)
696 {
697         int len = parent ? 6 : 3;
698         u64 nodeid;
699         u32 generation;
700
701         if (*max_len < len) {
702                 *max_len = len;
703                 return  FILEID_INVALID;
704         }
705
706         nodeid = get_fuse_inode(inode)->nodeid;
707         generation = inode->i_generation;
708
709         fh[0] = (u32)(nodeid >> 32);
710         fh[1] = (u32)(nodeid & 0xffffffff);
711         fh[2] = generation;
712
713         if (parent) {
714                 nodeid = get_fuse_inode(parent)->nodeid;
715                 generation = parent->i_generation;
716
717                 fh[3] = (u32)(nodeid >> 32);
718                 fh[4] = (u32)(nodeid & 0xffffffff);
719                 fh[5] = generation;
720         }
721
722         *max_len = len;
723         return parent ? 0x82 : 0x81;
724 }
725
726 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
727                 struct fid *fid, int fh_len, int fh_type)
728 {
729         struct fuse_inode_handle handle;
730
731         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
732                 return NULL;
733
734         handle.nodeid = (u64) fid->raw[0] << 32;
735         handle.nodeid |= (u64) fid->raw[1];
736         handle.generation = fid->raw[2];
737         return fuse_get_dentry(sb, &handle);
738 }
739
740 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
741                 struct fid *fid, int fh_len, int fh_type)
742 {
743         struct fuse_inode_handle parent;
744
745         if (fh_type != 0x82 || fh_len < 6)
746                 return NULL;
747
748         parent.nodeid = (u64) fid->raw[3] << 32;
749         parent.nodeid |= (u64) fid->raw[4];
750         parent.generation = fid->raw[5];
751         return fuse_get_dentry(sb, &parent);
752 }
753
754 static struct dentry *fuse_get_parent(struct dentry *child)
755 {
756         struct inode *child_inode = d_inode(child);
757         struct fuse_conn *fc = get_fuse_conn(child_inode);
758         struct inode *inode;
759         struct dentry *parent;
760         struct fuse_entry_out outarg;
761         struct qstr name;
762         int err;
763
764         if (!fc->export_support)
765                 return ERR_PTR(-ESTALE);
766
767         name.len = 2;
768         name.name = "..";
769         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
770                                &name, &outarg, &inode);
771         if (err) {
772                 if (err == -ENOENT)
773                         return ERR_PTR(-ESTALE);
774                 return ERR_PTR(err);
775         }
776
777         parent = d_obtain_alias(inode);
778         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
779                 fuse_invalidate_entry_cache(parent);
780
781         return parent;
782 }
783
784 static const struct export_operations fuse_export_operations = {
785         .fh_to_dentry   = fuse_fh_to_dentry,
786         .fh_to_parent   = fuse_fh_to_parent,
787         .encode_fh      = fuse_encode_fh,
788         .get_parent     = fuse_get_parent,
789 };
790
791 static const struct super_operations fuse_super_operations = {
792         .alloc_inode    = fuse_alloc_inode,
793         .destroy_inode  = fuse_destroy_inode,
794         .evict_inode    = fuse_evict_inode,
795         .write_inode    = fuse_write_inode,
796         .drop_inode     = generic_delete_inode,
797         .remount_fs     = fuse_remount_fs,
798         .put_super      = fuse_put_super,
799         .umount_begin   = fuse_umount_begin,
800         .statfs         = fuse_statfs,
801         .show_options   = fuse_show_options,
802 };
803
804 static void sanitize_global_limit(unsigned *limit)
805 {
806         if (*limit == 0)
807                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
808                          sizeof(struct fuse_req);
809
810         if (*limit >= 1 << 16)
811                 *limit = (1 << 16) - 1;
812 }
813
814 static int set_global_limit(const char *val, struct kernel_param *kp)
815 {
816         int rv;
817
818         rv = param_set_uint(val, kp);
819         if (rv)
820                 return rv;
821
822         sanitize_global_limit((unsigned *)kp->arg);
823
824         return 0;
825 }
826
827 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
828 {
829         int cap_sys_admin = capable(CAP_SYS_ADMIN);
830
831         if (arg->minor < 13)
832                 return;
833
834         sanitize_global_limit(&max_user_bgreq);
835         sanitize_global_limit(&max_user_congthresh);
836
837         if (arg->max_background) {
838                 fc->max_background = arg->max_background;
839
840                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
841                         fc->max_background = max_user_bgreq;
842         }
843         if (arg->congestion_threshold) {
844                 fc->congestion_threshold = arg->congestion_threshold;
845
846                 if (!cap_sys_admin &&
847                     fc->congestion_threshold > max_user_congthresh)
848                         fc->congestion_threshold = max_user_congthresh;
849         }
850 }
851
852 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
853 {
854         struct fuse_init_out *arg = &req->misc.init_out;
855
856         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
857                 fc->conn_error = 1;
858         else {
859                 unsigned long ra_pages;
860
861                 process_init_limits(fc, arg);
862
863                 if (arg->minor >= 6) {
864                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
865                         if (arg->flags & FUSE_ASYNC_READ)
866                                 fc->async_read = 1;
867                         if (!(arg->flags & FUSE_POSIX_LOCKS))
868                                 fc->no_lock = 1;
869                         if (arg->minor >= 17) {
870                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
871                                         fc->no_flock = 1;
872                         } else {
873                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
874                                         fc->no_flock = 1;
875                         }
876                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
877                                 fc->atomic_o_trunc = 1;
878                         if (arg->minor >= 9) {
879                                 /* LOOKUP has dependency on proto version */
880                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
881                                         fc->export_support = 1;
882                         }
883                         if (arg->flags & FUSE_BIG_WRITES)
884                                 fc->big_writes = 1;
885                         if (arg->flags & FUSE_DONT_MASK)
886                                 fc->dont_mask = 1;
887                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
888                                 fc->auto_inval_data = 1;
889                         if (arg->flags & FUSE_DO_READDIRPLUS) {
890                                 fc->do_readdirplus = 1;
891                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
892                                         fc->readdirplus_auto = 1;
893                         }
894                         if (arg->flags & FUSE_ASYNC_DIO)
895                                 fc->async_dio = 1;
896                         if (arg->flags & FUSE_WRITEBACK_CACHE)
897                                 fc->writeback_cache = 1;
898                         if (arg->time_gran && arg->time_gran <= 1000000000)
899                                 fc->sb->s_time_gran = arg->time_gran;
900                 } else {
901                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
902                         fc->no_lock = 1;
903                         fc->no_flock = 1;
904                 }
905
906                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
907                 fc->minor = arg->minor;
908                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
909                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
910                 fc->conn_init = 1;
911         }
912         fuse_set_initialized(fc);
913         wake_up_all(&fc->blocked_waitq);
914 }
915
916 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
917 {
918         struct fuse_init_in *arg = &req->misc.init_in;
919
920         arg->major = FUSE_KERNEL_VERSION;
921         arg->minor = FUSE_KERNEL_MINOR_VERSION;
922         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
923         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
924                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
925                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
926                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
927                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
928                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT;
929         req->in.h.opcode = FUSE_INIT;
930         req->in.numargs = 1;
931         req->in.args[0].size = sizeof(*arg);
932         req->in.args[0].value = arg;
933         req->out.numargs = 1;
934         /* Variable length argument used for backward compatibility
935            with interface version < 7.5.  Rest of init_out is zeroed
936            by do_get_request(), so a short reply is not a problem */
937         req->out.argvar = 1;
938         req->out.args[0].size = sizeof(struct fuse_init_out);
939         req->out.args[0].value = &req->misc.init_out;
940         req->end = process_init_reply;
941         fuse_request_send_background(fc, req);
942 }
943
944 static void fuse_free_conn(struct fuse_conn *fc)
945 {
946         WARN_ON(!list_empty(&fc->devices));
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 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
993 {
994         struct fuse_dev *fud;
995
996         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
997         if (fud) {
998                 fud->fc = fuse_conn_get(fc);
999                 fuse_pqueue_init(&fud->pq);
1000
1001                 spin_lock(&fc->lock);
1002                 list_add_tail(&fud->entry, &fc->devices);
1003                 spin_unlock(&fc->lock);
1004         }
1005
1006         return fud;
1007 }
1008 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1009
1010 void fuse_dev_free(struct fuse_dev *fud)
1011 {
1012         struct fuse_conn *fc = fud->fc;
1013
1014         if (fc) {
1015                 spin_lock(&fc->lock);
1016                 list_del(&fud->entry);
1017                 spin_unlock(&fc->lock);
1018
1019                 fuse_conn_put(fc);
1020         }
1021         kfree(fud);
1022 }
1023 EXPORT_SYMBOL_GPL(fuse_dev_free);
1024
1025 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1026 {
1027         struct fuse_dev *fud;
1028         struct fuse_conn *fc;
1029         struct inode *root;
1030         struct fuse_mount_data d;
1031         struct file *file;
1032         struct dentry *root_dentry;
1033         struct fuse_req *init_req;
1034         int err;
1035         int is_bdev = sb->s_bdev != NULL;
1036
1037         err = -EINVAL;
1038         if (sb->s_flags & MS_MANDLOCK)
1039                 goto err;
1040
1041         sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
1042
1043         if (!parse_fuse_opt(data, &d, is_bdev))
1044                 goto err;
1045
1046         if (is_bdev) {
1047 #ifdef CONFIG_BLOCK
1048                 err = -EINVAL;
1049                 if (!sb_set_blocksize(sb, d.blksize))
1050                         goto err;
1051 #endif
1052         } else {
1053                 sb->s_blocksize = PAGE_CACHE_SIZE;
1054                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1055         }
1056         sb->s_magic = FUSE_SUPER_MAGIC;
1057         sb->s_op = &fuse_super_operations;
1058         sb->s_maxbytes = MAX_LFS_FILESIZE;
1059         sb->s_time_gran = 1;
1060         sb->s_export_op = &fuse_export_operations;
1061
1062         file = fget(d.fd);
1063         err = -EINVAL;
1064         if (!file)
1065                 goto err;
1066
1067         if ((file->f_op != &fuse_dev_operations) ||
1068             (file->f_cred->user_ns != &init_user_ns))
1069                 goto err_fput;
1070
1071         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1072         err = -ENOMEM;
1073         if (!fc)
1074                 goto err_fput;
1075
1076         fuse_conn_init(fc);
1077         fc->release = fuse_free_conn;
1078
1079         fud = fuse_dev_alloc(fc);
1080         if (!fud)
1081                 goto err_put_conn;
1082
1083         fc->dev = sb->s_dev;
1084         fc->sb = sb;
1085         err = fuse_bdi_init(fc, sb);
1086         if (err)
1087                 goto err_dev_free;
1088
1089         sb->s_bdi = &fc->bdi;
1090
1091         /* Handle umasking inside the fuse code */
1092         if (sb->s_flags & MS_POSIXACL)
1093                 fc->dont_mask = 1;
1094         sb->s_flags |= MS_POSIXACL;
1095
1096         fc->flags = d.flags;
1097         fc->user_id = d.user_id;
1098         fc->group_id = d.group_id;
1099         fc->max_read = max_t(unsigned, 4096, d.max_read);
1100
1101         /* Used by get_root_inode() */
1102         sb->s_fs_info = fc;
1103
1104         err = -ENOMEM;
1105         root = fuse_get_root_inode(sb, d.rootmode);
1106         root_dentry = d_make_root(root);
1107         if (!root_dentry)
1108                 goto err_dev_free;
1109         /* only now - we want root dentry with NULL ->d_op */
1110         sb->s_d_op = &fuse_dentry_operations;
1111
1112         init_req = fuse_request_alloc(0);
1113         if (!init_req)
1114                 goto err_put_root;
1115         __set_bit(FR_BACKGROUND, &init_req->flags);
1116
1117         if (is_bdev) {
1118                 fc->destroy_req = fuse_request_alloc(0);
1119                 if (!fc->destroy_req)
1120                         goto err_free_init_req;
1121         }
1122
1123         mutex_lock(&fuse_mutex);
1124         err = -EINVAL;
1125         if (file->private_data)
1126                 goto err_unlock;
1127
1128         err = fuse_ctl_add_conn(fc);
1129         if (err)
1130                 goto err_unlock;
1131
1132         list_add_tail(&fc->entry, &fuse_conn_list);
1133         sb->s_root = root_dentry;
1134         file->private_data = fud;
1135         mutex_unlock(&fuse_mutex);
1136         /*
1137          * atomic_dec_and_test() in fput() provides the necessary
1138          * memory barrier for file->private_data to be visible on all
1139          * CPUs after this
1140          */
1141         fput(file);
1142
1143         fuse_send_init(fc, init_req);
1144
1145         return 0;
1146
1147  err_unlock:
1148         mutex_unlock(&fuse_mutex);
1149  err_free_init_req:
1150         fuse_request_free(init_req);
1151  err_put_root:
1152         dput(root_dentry);
1153  err_dev_free:
1154         fuse_dev_free(fud);
1155  err_put_conn:
1156         fuse_bdi_destroy(fc);
1157         fuse_conn_put(fc);
1158         sb->s_fs_info = NULL;
1159  err_fput:
1160         fput(file);
1161  err:
1162         return err;
1163 }
1164
1165 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1166                        int flags, const char *dev_name,
1167                        void *raw_data)
1168 {
1169         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1170 }
1171
1172 static void fuse_sb_destroy(struct super_block *sb)
1173 {
1174         struct fuse_conn *fc = get_fuse_conn_super(sb);
1175
1176         if (fc) {
1177                 fuse_send_destroy(fc);
1178
1179                 fuse_abort_conn(fc);
1180                 fuse_wait_aborted(fc);
1181
1182                 down_write(&fc->killsb);
1183                 fc->sb = NULL;
1184                 up_write(&fc->killsb);
1185         }
1186 }
1187
1188 static void fuse_kill_sb_anon(struct super_block *sb)
1189 {
1190         fuse_sb_destroy(sb);
1191         kill_anon_super(sb);
1192 }
1193
1194 static struct file_system_type fuse_fs_type = {
1195         .owner          = THIS_MODULE,
1196         .name           = "fuse",
1197         .fs_flags       = FS_HAS_SUBTYPE,
1198         .mount          = fuse_mount,
1199         .kill_sb        = fuse_kill_sb_anon,
1200 };
1201 MODULE_ALIAS_FS("fuse");
1202
1203 #ifdef CONFIG_BLOCK
1204 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1205                            int flags, const char *dev_name,
1206                            void *raw_data)
1207 {
1208         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1209 }
1210
1211 static void fuse_kill_sb_blk(struct super_block *sb)
1212 {
1213         fuse_sb_destroy(sb);
1214         kill_block_super(sb);
1215 }
1216
1217 static struct file_system_type fuseblk_fs_type = {
1218         .owner          = THIS_MODULE,
1219         .name           = "fuseblk",
1220         .mount          = fuse_mount_blk,
1221         .kill_sb        = fuse_kill_sb_blk,
1222         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1223 };
1224 MODULE_ALIAS_FS("fuseblk");
1225
1226 static inline int register_fuseblk(void)
1227 {
1228         return register_filesystem(&fuseblk_fs_type);
1229 }
1230
1231 static inline void unregister_fuseblk(void)
1232 {
1233         unregister_filesystem(&fuseblk_fs_type);
1234 }
1235 #else
1236 static inline int register_fuseblk(void)
1237 {
1238         return 0;
1239 }
1240
1241 static inline void unregister_fuseblk(void)
1242 {
1243 }
1244 #endif
1245
1246 static void fuse_inode_init_once(void *foo)
1247 {
1248         struct inode *inode = foo;
1249
1250         inode_init_once(inode);
1251 }
1252
1253 static int __init fuse_fs_init(void)
1254 {
1255         int err;
1256
1257         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1258                                               sizeof(struct fuse_inode),
1259                                               0, SLAB_HWCACHE_ALIGN,
1260                                               fuse_inode_init_once);
1261         err = -ENOMEM;
1262         if (!fuse_inode_cachep)
1263                 goto out;
1264
1265         err = register_fuseblk();
1266         if (err)
1267                 goto out2;
1268
1269         err = register_filesystem(&fuse_fs_type);
1270         if (err)
1271                 goto out3;
1272
1273         return 0;
1274
1275  out3:
1276         unregister_fuseblk();
1277  out2:
1278         kmem_cache_destroy(fuse_inode_cachep);
1279  out:
1280         return err;
1281 }
1282
1283 static void fuse_fs_cleanup(void)
1284 {
1285         unregister_filesystem(&fuse_fs_type);
1286         unregister_fuseblk();
1287
1288         /*
1289          * Make sure all delayed rcu free inodes are flushed before we
1290          * destroy cache.
1291          */
1292         rcu_barrier();
1293         kmem_cache_destroy(fuse_inode_cachep);
1294 }
1295
1296 static struct kobject *fuse_kobj;
1297
1298 static int fuse_sysfs_init(void)
1299 {
1300         int err;
1301
1302         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1303         if (!fuse_kobj) {
1304                 err = -ENOMEM;
1305                 goto out_err;
1306         }
1307
1308         err = sysfs_create_mount_point(fuse_kobj, "connections");
1309         if (err)
1310                 goto out_fuse_unregister;
1311
1312         return 0;
1313
1314  out_fuse_unregister:
1315         kobject_put(fuse_kobj);
1316  out_err:
1317         return err;
1318 }
1319
1320 static void fuse_sysfs_cleanup(void)
1321 {
1322         sysfs_remove_mount_point(fuse_kobj, "connections");
1323         kobject_put(fuse_kobj);
1324 }
1325
1326 static int __init fuse_init(void)
1327 {
1328         int res;
1329
1330         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1331                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1332
1333         INIT_LIST_HEAD(&fuse_conn_list);
1334         res = fuse_fs_init();
1335         if (res)
1336                 goto err;
1337
1338         res = fuse_dev_init();
1339         if (res)
1340                 goto err_fs_cleanup;
1341
1342         res = fuse_sysfs_init();
1343         if (res)
1344                 goto err_dev_cleanup;
1345
1346         res = fuse_ctl_init();
1347         if (res)
1348                 goto err_sysfs_cleanup;
1349
1350         sanitize_global_limit(&max_user_bgreq);
1351         sanitize_global_limit(&max_user_congthresh);
1352
1353         return 0;
1354
1355  err_sysfs_cleanup:
1356         fuse_sysfs_cleanup();
1357  err_dev_cleanup:
1358         fuse_dev_cleanup();
1359  err_fs_cleanup:
1360         fuse_fs_cleanup();
1361  err:
1362         return res;
1363 }
1364
1365 static void __exit fuse_exit(void)
1366 {
1367         printk(KERN_DEBUG "fuse exit\n");
1368
1369         fuse_ctl_cleanup();
1370         fuse_sysfs_cleanup();
1371         fuse_fs_cleanup();
1372         fuse_dev_cleanup();
1373 }
1374
1375 module_init(fuse_init);
1376 module_exit(fuse_exit);