]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - fs/fuse/inode.c
fuse: add FUSE_NO_OPEN_SUPPORT flag to INIT
[karo-tx-linux.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 fuse_match_uint(substring_t *s, unsigned int *res)
482 {
483         int err = -ENOMEM;
484         char *buf = match_strdup(s);
485         if (buf) {
486                 err = kstrtouint(buf, 10, res);
487                 kfree(buf);
488         }
489         return err;
490 }
491
492 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
493 {
494         char *p;
495         memset(d, 0, sizeof(struct fuse_mount_data));
496         d->max_read = ~0;
497         d->blksize = FUSE_DEFAULT_BLKSIZE;
498
499         while ((p = strsep(&opt, ",")) != NULL) {
500                 int token;
501                 int value;
502                 unsigned uv;
503                 substring_t args[MAX_OPT_ARGS];
504                 if (!*p)
505                         continue;
506
507                 token = match_token(p, tokens, args);
508                 switch (token) {
509                 case OPT_FD:
510                         if (match_int(&args[0], &value))
511                                 return 0;
512                         d->fd = value;
513                         d->fd_present = 1;
514                         break;
515
516                 case OPT_ROOTMODE:
517                         if (match_octal(&args[0], &value))
518                                 return 0;
519                         if (!fuse_valid_type(value))
520                                 return 0;
521                         d->rootmode = value;
522                         d->rootmode_present = 1;
523                         break;
524
525                 case OPT_USER_ID:
526                         if (fuse_match_uint(&args[0], &uv))
527                                 return 0;
528                         d->user_id = make_kuid(current_user_ns(), uv);
529                         if (!uid_valid(d->user_id))
530                                 return 0;
531                         d->user_id_present = 1;
532                         break;
533
534                 case OPT_GROUP_ID:
535                         if (fuse_match_uint(&args[0], &uv))
536                                 return 0;
537                         d->group_id = make_kgid(current_user_ns(), uv);
538                         if (!gid_valid(d->group_id))
539                                 return 0;
540                         d->group_id_present = 1;
541                         break;
542
543                 case OPT_DEFAULT_PERMISSIONS:
544                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
545                         break;
546
547                 case OPT_ALLOW_OTHER:
548                         d->flags |= FUSE_ALLOW_OTHER;
549                         break;
550
551                 case OPT_MAX_READ:
552                         if (match_int(&args[0], &value))
553                                 return 0;
554                         d->max_read = value;
555                         break;
556
557                 case OPT_BLKSIZE:
558                         if (!is_bdev || match_int(&args[0], &value))
559                                 return 0;
560                         d->blksize = value;
561                         break;
562
563                 default:
564                         return 0;
565                 }
566         }
567
568         if (!d->fd_present || !d->rootmode_present ||
569             !d->user_id_present || !d->group_id_present)
570                 return 0;
571
572         return 1;
573 }
574
575 static int fuse_show_options(struct seq_file *m, struct dentry *root)
576 {
577         struct super_block *sb = root->d_sb;
578         struct fuse_conn *fc = get_fuse_conn_super(sb);
579
580         seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
581         seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
582         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
583                 seq_puts(m, ",default_permissions");
584         if (fc->flags & FUSE_ALLOW_OTHER)
585                 seq_puts(m, ",allow_other");
586         if (fc->max_read != ~0)
587                 seq_printf(m, ",max_read=%u", fc->max_read);
588         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
589                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
590         return 0;
591 }
592
593 void fuse_conn_init(struct fuse_conn *fc)
594 {
595         memset(fc, 0, sizeof(*fc));
596         spin_lock_init(&fc->lock);
597         init_rwsem(&fc->killsb);
598         atomic_set(&fc->count, 1);
599         init_waitqueue_head(&fc->waitq);
600         init_waitqueue_head(&fc->blocked_waitq);
601         init_waitqueue_head(&fc->reserved_req_waitq);
602         INIT_LIST_HEAD(&fc->pending);
603         INIT_LIST_HEAD(&fc->processing);
604         INIT_LIST_HEAD(&fc->io);
605         INIT_LIST_HEAD(&fc->interrupts);
606         INIT_LIST_HEAD(&fc->bg_queue);
607         INIT_LIST_HEAD(&fc->entry);
608         fc->forget_list_tail = &fc->forget_list_head;
609         atomic_set(&fc->num_waiting, 0);
610         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
611         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
612         fc->khctr = 0;
613         fc->polled_files = RB_ROOT;
614         fc->reqctr = 0;
615         fc->blocked = 0;
616         fc->initialized = 0;
617         fc->attr_version = 1;
618         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
619 }
620 EXPORT_SYMBOL_GPL(fuse_conn_init);
621
622 void fuse_conn_put(struct fuse_conn *fc)
623 {
624         if (atomic_dec_and_test(&fc->count)) {
625                 if (fc->destroy_req)
626                         fuse_request_free(fc->destroy_req);
627                 fc->release(fc);
628         }
629 }
630 EXPORT_SYMBOL_GPL(fuse_conn_put);
631
632 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
633 {
634         atomic_inc(&fc->count);
635         return fc;
636 }
637 EXPORT_SYMBOL_GPL(fuse_conn_get);
638
639 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
640 {
641         struct fuse_attr attr;
642         memset(&attr, 0, sizeof(attr));
643
644         attr.mode = mode;
645         attr.ino = FUSE_ROOT_ID;
646         attr.nlink = 1;
647         return fuse_iget(sb, 1, 0, &attr, 0, 0);
648 }
649
650 struct fuse_inode_handle {
651         u64 nodeid;
652         u32 generation;
653 };
654
655 static struct dentry *fuse_get_dentry(struct super_block *sb,
656                                       struct fuse_inode_handle *handle)
657 {
658         struct fuse_conn *fc = get_fuse_conn_super(sb);
659         struct inode *inode;
660         struct dentry *entry;
661         int err = -ESTALE;
662
663         if (handle->nodeid == 0)
664                 goto out_err;
665
666         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
667         if (!inode) {
668                 struct fuse_entry_out outarg;
669                 struct qstr name;
670
671                 if (!fc->export_support)
672                         goto out_err;
673
674                 name.len = 1;
675                 name.name = ".";
676                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
677                                        &inode);
678                 if (err && err != -ENOENT)
679                         goto out_err;
680                 if (err || !inode) {
681                         err = -ESTALE;
682                         goto out_err;
683                 }
684                 err = -EIO;
685                 if (get_node_id(inode) != handle->nodeid)
686                         goto out_iput;
687         }
688         err = -ESTALE;
689         if (inode->i_generation != handle->generation)
690                 goto out_iput;
691
692         entry = d_obtain_alias(inode);
693         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
694                 fuse_invalidate_entry_cache(entry);
695
696         return entry;
697
698  out_iput:
699         iput(inode);
700  out_err:
701         return ERR_PTR(err);
702 }
703
704 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
705                            struct inode *parent)
706 {
707         int len = parent ? 6 : 3;
708         u64 nodeid;
709         u32 generation;
710
711         if (*max_len < len) {
712                 *max_len = len;
713                 return  FILEID_INVALID;
714         }
715
716         nodeid = get_fuse_inode(inode)->nodeid;
717         generation = inode->i_generation;
718
719         fh[0] = (u32)(nodeid >> 32);
720         fh[1] = (u32)(nodeid & 0xffffffff);
721         fh[2] = generation;
722
723         if (parent) {
724                 nodeid = get_fuse_inode(parent)->nodeid;
725                 generation = parent->i_generation;
726
727                 fh[3] = (u32)(nodeid >> 32);
728                 fh[4] = (u32)(nodeid & 0xffffffff);
729                 fh[5] = generation;
730         }
731
732         *max_len = len;
733         return parent ? 0x82 : 0x81;
734 }
735
736 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
737                 struct fid *fid, int fh_len, int fh_type)
738 {
739         struct fuse_inode_handle handle;
740
741         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
742                 return NULL;
743
744         handle.nodeid = (u64) fid->raw[0] << 32;
745         handle.nodeid |= (u64) fid->raw[1];
746         handle.generation = fid->raw[2];
747         return fuse_get_dentry(sb, &handle);
748 }
749
750 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
751                 struct fid *fid, int fh_len, int fh_type)
752 {
753         struct fuse_inode_handle parent;
754
755         if (fh_type != 0x82 || fh_len < 6)
756                 return NULL;
757
758         parent.nodeid = (u64) fid->raw[3] << 32;
759         parent.nodeid |= (u64) fid->raw[4];
760         parent.generation = fid->raw[5];
761         return fuse_get_dentry(sb, &parent);
762 }
763
764 static struct dentry *fuse_get_parent(struct dentry *child)
765 {
766         struct inode *child_inode = child->d_inode;
767         struct fuse_conn *fc = get_fuse_conn(child_inode);
768         struct inode *inode;
769         struct dentry *parent;
770         struct fuse_entry_out outarg;
771         struct qstr name;
772         int err;
773
774         if (!fc->export_support)
775                 return ERR_PTR(-ESTALE);
776
777         name.len = 2;
778         name.name = "..";
779         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
780                                &name, &outarg, &inode);
781         if (err) {
782                 if (err == -ENOENT)
783                         return ERR_PTR(-ESTALE);
784                 return ERR_PTR(err);
785         }
786
787         parent = d_obtain_alias(inode);
788         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
789                 fuse_invalidate_entry_cache(parent);
790
791         return parent;
792 }
793
794 static const struct export_operations fuse_export_operations = {
795         .fh_to_dentry   = fuse_fh_to_dentry,
796         .fh_to_parent   = fuse_fh_to_parent,
797         .encode_fh      = fuse_encode_fh,
798         .get_parent     = fuse_get_parent,
799 };
800
801 static const struct super_operations fuse_super_operations = {
802         .alloc_inode    = fuse_alloc_inode,
803         .destroy_inode  = fuse_destroy_inode,
804         .evict_inode    = fuse_evict_inode,
805         .write_inode    = fuse_write_inode,
806         .drop_inode     = generic_delete_inode,
807         .remount_fs     = fuse_remount_fs,
808         .put_super      = fuse_put_super,
809         .umount_begin   = fuse_umount_begin,
810         .statfs         = fuse_statfs,
811         .show_options   = fuse_show_options,
812 };
813
814 static void sanitize_global_limit(unsigned *limit)
815 {
816         if (*limit == 0)
817                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
818                          sizeof(struct fuse_req);
819
820         if (*limit >= 1 << 16)
821                 *limit = (1 << 16) - 1;
822 }
823
824 static int set_global_limit(const char *val, struct kernel_param *kp)
825 {
826         int rv;
827
828         rv = param_set_uint(val, kp);
829         if (rv)
830                 return rv;
831
832         sanitize_global_limit((unsigned *)kp->arg);
833
834         return 0;
835 }
836
837 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
838 {
839         int cap_sys_admin = capable(CAP_SYS_ADMIN);
840
841         if (arg->minor < 13)
842                 return;
843
844         sanitize_global_limit(&max_user_bgreq);
845         sanitize_global_limit(&max_user_congthresh);
846
847         if (arg->max_background) {
848                 fc->max_background = arg->max_background;
849
850                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
851                         fc->max_background = max_user_bgreq;
852         }
853         if (arg->congestion_threshold) {
854                 fc->congestion_threshold = arg->congestion_threshold;
855
856                 if (!cap_sys_admin &&
857                     fc->congestion_threshold > max_user_congthresh)
858                         fc->congestion_threshold = max_user_congthresh;
859         }
860 }
861
862 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
863 {
864         struct fuse_init_out *arg = &req->misc.init_out;
865
866         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
867                 fc->conn_error = 1;
868         else {
869                 unsigned long ra_pages;
870
871                 process_init_limits(fc, arg);
872
873                 if (arg->minor >= 6) {
874                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
875                         if (arg->flags & FUSE_ASYNC_READ)
876                                 fc->async_read = 1;
877                         if (!(arg->flags & FUSE_POSIX_LOCKS))
878                                 fc->no_lock = 1;
879                         if (arg->minor >= 17) {
880                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
881                                         fc->no_flock = 1;
882                         } else {
883                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
884                                         fc->no_flock = 1;
885                         }
886                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
887                                 fc->atomic_o_trunc = 1;
888                         if (arg->minor >= 9) {
889                                 /* LOOKUP has dependency on proto version */
890                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
891                                         fc->export_support = 1;
892                         }
893                         if (arg->flags & FUSE_BIG_WRITES)
894                                 fc->big_writes = 1;
895                         if (arg->flags & FUSE_DONT_MASK)
896                                 fc->dont_mask = 1;
897                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
898                                 fc->auto_inval_data = 1;
899                         if (arg->flags & FUSE_DO_READDIRPLUS) {
900                                 fc->do_readdirplus = 1;
901                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
902                                         fc->readdirplus_auto = 1;
903                         }
904                         if (arg->flags & FUSE_ASYNC_DIO)
905                                 fc->async_dio = 1;
906                         if (arg->flags & FUSE_WRITEBACK_CACHE)
907                                 fc->writeback_cache = 1;
908                         if (arg->time_gran && arg->time_gran <= 1000000000)
909                                 fc->sb->s_time_gran = arg->time_gran;
910                 } else {
911                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
912                         fc->no_lock = 1;
913                         fc->no_flock = 1;
914                 }
915
916                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
917                 fc->minor = arg->minor;
918                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
919                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
920                 fc->conn_init = 1;
921         }
922         fc->initialized = 1;
923         wake_up_all(&fc->blocked_waitq);
924 }
925
926 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
927 {
928         struct fuse_init_in *arg = &req->misc.init_in;
929
930         arg->major = FUSE_KERNEL_VERSION;
931         arg->minor = FUSE_KERNEL_MINOR_VERSION;
932         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
933         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
934                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
935                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
936                 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
937                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
938                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT;
939         req->in.h.opcode = FUSE_INIT;
940         req->in.numargs = 1;
941         req->in.args[0].size = sizeof(*arg);
942         req->in.args[0].value = arg;
943         req->out.numargs = 1;
944         /* Variable length argument used for backward compatibility
945            with interface version < 7.5.  Rest of init_out is zeroed
946            by do_get_request(), so a short reply is not a problem */
947         req->out.argvar = 1;
948         req->out.args[0].size = sizeof(struct fuse_init_out);
949         req->out.args[0].value = &req->misc.init_out;
950         req->end = process_init_reply;
951         fuse_request_send_background(fc, req);
952 }
953
954 static void fuse_free_conn(struct fuse_conn *fc)
955 {
956         kfree_rcu(fc, rcu);
957 }
958
959 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
960 {
961         int err;
962
963         fc->bdi.name = "fuse";
964         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
965         /* fuse does it's own writeback accounting */
966         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
967
968         err = bdi_init(&fc->bdi);
969         if (err)
970                 return err;
971
972         fc->bdi_initialized = 1;
973
974         if (sb->s_bdev) {
975                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
976                                     MAJOR(fc->dev), MINOR(fc->dev));
977         } else {
978                 err = bdi_register_dev(&fc->bdi, fc->dev);
979         }
980
981         if (err)
982                 return err;
983
984         /*
985          * For a single fuse filesystem use max 1% of dirty +
986          * writeback threshold.
987          *
988          * This gives about 1M of write buffer for memory maps on a
989          * machine with 1G and 10% dirty_ratio, which should be more
990          * than enough.
991          *
992          * Privileged users can raise it by writing to
993          *
994          *    /sys/class/bdi/<bdi>/max_ratio
995          */
996         bdi_set_max_ratio(&fc->bdi, 1);
997
998         return 0;
999 }
1000
1001 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1002 {
1003         struct fuse_conn *fc;
1004         struct inode *root;
1005         struct fuse_mount_data d;
1006         struct file *file;
1007         struct dentry *root_dentry;
1008         struct fuse_req *init_req;
1009         int err;
1010         int is_bdev = sb->s_bdev != NULL;
1011
1012         err = -EINVAL;
1013         if (sb->s_flags & MS_MANDLOCK)
1014                 goto err;
1015
1016         sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
1017
1018         if (!parse_fuse_opt(data, &d, is_bdev))
1019                 goto err;
1020
1021         if (is_bdev) {
1022 #ifdef CONFIG_BLOCK
1023                 err = -EINVAL;
1024                 if (!sb_set_blocksize(sb, d.blksize))
1025                         goto err;
1026 #endif
1027         } else {
1028                 sb->s_blocksize = PAGE_CACHE_SIZE;
1029                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1030         }
1031         sb->s_magic = FUSE_SUPER_MAGIC;
1032         sb->s_op = &fuse_super_operations;
1033         sb->s_maxbytes = MAX_LFS_FILESIZE;
1034         sb->s_time_gran = 1;
1035         sb->s_export_op = &fuse_export_operations;
1036
1037         file = fget(d.fd);
1038         err = -EINVAL;
1039         if (!file)
1040                 goto err;
1041
1042         if ((file->f_op != &fuse_dev_operations) ||
1043             (file->f_cred->user_ns != &init_user_ns))
1044                 goto err_fput;
1045
1046         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1047         err = -ENOMEM;
1048         if (!fc)
1049                 goto err_fput;
1050
1051         fuse_conn_init(fc);
1052
1053         fc->dev = sb->s_dev;
1054         fc->sb = sb;
1055         err = fuse_bdi_init(fc, sb);
1056         if (err)
1057                 goto err_put_conn;
1058
1059         sb->s_bdi = &fc->bdi;
1060
1061         /* Handle umasking inside the fuse code */
1062         if (sb->s_flags & MS_POSIXACL)
1063                 fc->dont_mask = 1;
1064         sb->s_flags |= MS_POSIXACL;
1065
1066         fc->release = fuse_free_conn;
1067         fc->flags = d.flags;
1068         fc->user_id = d.user_id;
1069         fc->group_id = d.group_id;
1070         fc->max_read = max_t(unsigned, 4096, d.max_read);
1071
1072         /* Used by get_root_inode() */
1073         sb->s_fs_info = fc;
1074
1075         err = -ENOMEM;
1076         root = fuse_get_root_inode(sb, d.rootmode);
1077         root_dentry = d_make_root(root);
1078         if (!root_dentry)
1079                 goto err_put_conn;
1080         /* only now - we want root dentry with NULL ->d_op */
1081         sb->s_d_op = &fuse_dentry_operations;
1082
1083         init_req = fuse_request_alloc(0);
1084         if (!init_req)
1085                 goto err_put_root;
1086         init_req->background = 1;
1087
1088         if (is_bdev) {
1089                 fc->destroy_req = fuse_request_alloc(0);
1090                 if (!fc->destroy_req)
1091                         goto err_free_init_req;
1092         }
1093
1094         mutex_lock(&fuse_mutex);
1095         err = -EINVAL;
1096         if (file->private_data)
1097                 goto err_unlock;
1098
1099         err = fuse_ctl_add_conn(fc);
1100         if (err)
1101                 goto err_unlock;
1102
1103         list_add_tail(&fc->entry, &fuse_conn_list);
1104         sb->s_root = root_dentry;
1105         fc->connected = 1;
1106         file->private_data = fuse_conn_get(fc);
1107         mutex_unlock(&fuse_mutex);
1108         /*
1109          * atomic_dec_and_test() in fput() provides the necessary
1110          * memory barrier for file->private_data to be visible on all
1111          * CPUs after this
1112          */
1113         fput(file);
1114
1115         fuse_send_init(fc, init_req);
1116
1117         return 0;
1118
1119  err_unlock:
1120         mutex_unlock(&fuse_mutex);
1121  err_free_init_req:
1122         fuse_request_free(init_req);
1123  err_put_root:
1124         dput(root_dentry);
1125  err_put_conn:
1126         fuse_bdi_destroy(fc);
1127         fuse_conn_put(fc);
1128  err_fput:
1129         fput(file);
1130  err:
1131         return err;
1132 }
1133
1134 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1135                        int flags, const char *dev_name,
1136                        void *raw_data)
1137 {
1138         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1139 }
1140
1141 static void fuse_kill_sb_anon(struct super_block *sb)
1142 {
1143         struct fuse_conn *fc = get_fuse_conn_super(sb);
1144
1145         if (fc) {
1146                 down_write(&fc->killsb);
1147                 fc->sb = NULL;
1148                 up_write(&fc->killsb);
1149         }
1150
1151         kill_anon_super(sb);
1152 }
1153
1154 static struct file_system_type fuse_fs_type = {
1155         .owner          = THIS_MODULE,
1156         .name           = "fuse",
1157         .fs_flags       = FS_HAS_SUBTYPE,
1158         .mount          = fuse_mount,
1159         .kill_sb        = fuse_kill_sb_anon,
1160 };
1161 MODULE_ALIAS_FS("fuse");
1162
1163 #ifdef CONFIG_BLOCK
1164 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1165                            int flags, const char *dev_name,
1166                            void *raw_data)
1167 {
1168         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1169 }
1170
1171 static void fuse_kill_sb_blk(struct super_block *sb)
1172 {
1173         struct fuse_conn *fc = get_fuse_conn_super(sb);
1174
1175         if (fc) {
1176                 down_write(&fc->killsb);
1177                 fc->sb = NULL;
1178                 up_write(&fc->killsb);
1179         }
1180
1181         kill_block_super(sb);
1182 }
1183
1184 static struct file_system_type fuseblk_fs_type = {
1185         .owner          = THIS_MODULE,
1186         .name           = "fuseblk",
1187         .mount          = fuse_mount_blk,
1188         .kill_sb        = fuse_kill_sb_blk,
1189         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1190 };
1191 MODULE_ALIAS_FS("fuseblk");
1192
1193 static inline int register_fuseblk(void)
1194 {
1195         return register_filesystem(&fuseblk_fs_type);
1196 }
1197
1198 static inline void unregister_fuseblk(void)
1199 {
1200         unregister_filesystem(&fuseblk_fs_type);
1201 }
1202 #else
1203 static inline int register_fuseblk(void)
1204 {
1205         return 0;
1206 }
1207
1208 static inline void unregister_fuseblk(void)
1209 {
1210 }
1211 #endif
1212
1213 static void fuse_inode_init_once(void *foo)
1214 {
1215         struct inode *inode = foo;
1216
1217         inode_init_once(inode);
1218 }
1219
1220 static int __init fuse_fs_init(void)
1221 {
1222         int err;
1223
1224         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1225                                               sizeof(struct fuse_inode),
1226                                               0, SLAB_HWCACHE_ALIGN,
1227                                               fuse_inode_init_once);
1228         err = -ENOMEM;
1229         if (!fuse_inode_cachep)
1230                 goto out;
1231
1232         err = register_fuseblk();
1233         if (err)
1234                 goto out2;
1235
1236         err = register_filesystem(&fuse_fs_type);
1237         if (err)
1238                 goto out3;
1239
1240         return 0;
1241
1242  out3:
1243         unregister_fuseblk();
1244  out2:
1245         kmem_cache_destroy(fuse_inode_cachep);
1246  out:
1247         return err;
1248 }
1249
1250 static void fuse_fs_cleanup(void)
1251 {
1252         unregister_filesystem(&fuse_fs_type);
1253         unregister_fuseblk();
1254
1255         /*
1256          * Make sure all delayed rcu free inodes are flushed before we
1257          * destroy cache.
1258          */
1259         rcu_barrier();
1260         kmem_cache_destroy(fuse_inode_cachep);
1261 }
1262
1263 static struct kobject *fuse_kobj;
1264 static struct kobject *connections_kobj;
1265
1266 static int fuse_sysfs_init(void)
1267 {
1268         int err;
1269
1270         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1271         if (!fuse_kobj) {
1272                 err = -ENOMEM;
1273                 goto out_err;
1274         }
1275
1276         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1277         if (!connections_kobj) {
1278                 err = -ENOMEM;
1279                 goto out_fuse_unregister;
1280         }
1281
1282         return 0;
1283
1284  out_fuse_unregister:
1285         kobject_put(fuse_kobj);
1286  out_err:
1287         return err;
1288 }
1289
1290 static void fuse_sysfs_cleanup(void)
1291 {
1292         kobject_put(connections_kobj);
1293         kobject_put(fuse_kobj);
1294 }
1295
1296 static int __init fuse_init(void)
1297 {
1298         int res;
1299
1300         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1301                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1302
1303         INIT_LIST_HEAD(&fuse_conn_list);
1304         res = fuse_fs_init();
1305         if (res)
1306                 goto err;
1307
1308         res = fuse_dev_init();
1309         if (res)
1310                 goto err_fs_cleanup;
1311
1312         res = fuse_sysfs_init();
1313         if (res)
1314                 goto err_dev_cleanup;
1315
1316         res = fuse_ctl_init();
1317         if (res)
1318                 goto err_sysfs_cleanup;
1319
1320         sanitize_global_limit(&max_user_bgreq);
1321         sanitize_global_limit(&max_user_congthresh);
1322
1323         return 0;
1324
1325  err_sysfs_cleanup:
1326         fuse_sysfs_cleanup();
1327  err_dev_cleanup:
1328         fuse_dev_cleanup();
1329  err_fs_cleanup:
1330         fuse_fs_cleanup();
1331  err:
1332         return res;
1333 }
1334
1335 static void __exit fuse_exit(void)
1336 {
1337         printk(KERN_DEBUG "fuse exit\n");
1338
1339         fuse_ctl_cleanup();
1340         fuse_sysfs_cleanup();
1341         fuse_fs_cleanup();
1342         fuse_dev_cleanup();
1343 }
1344
1345 module_init(fuse_init);
1346 module_exit(fuse_exit);