Merge tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso...
[cascardo/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         }
179         inode->i_ctime.tv_sec   = attr->ctime;
180         inode->i_ctime.tv_nsec  = attr->ctimensec;
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         if (S_ISREG(inode->i_mode)) {
260                 fuse_init_common(inode);
261                 fuse_init_file_inode(inode);
262         } else if (S_ISDIR(inode->i_mode))
263                 fuse_init_dir(inode);
264         else if (S_ISLNK(inode->i_mode))
265                 fuse_init_symlink(inode);
266         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
267                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
268                 fuse_init_common(inode);
269                 init_special_inode(inode, inode->i_mode,
270                                    new_decode_dev(attr->rdev));
271         } else
272                 BUG();
273 }
274
275 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
276 {
277         u64 nodeid = *(u64 *) _nodeidp;
278         if (get_node_id(inode) == nodeid)
279                 return 1;
280         else
281                 return 0;
282 }
283
284 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
285 {
286         u64 nodeid = *(u64 *) _nodeidp;
287         get_fuse_inode(inode)->nodeid = nodeid;
288         return 0;
289 }
290
291 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
292                         int generation, struct fuse_attr *attr,
293                         u64 attr_valid, u64 attr_version)
294 {
295         struct inode *inode;
296         struct fuse_inode *fi;
297         struct fuse_conn *fc = get_fuse_conn_super(sb);
298
299  retry:
300         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
301         if (!inode)
302                 return NULL;
303
304         if ((inode->i_state & I_NEW)) {
305                 inode->i_flags |= S_NOATIME;
306                 if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
307                         inode->i_flags |= S_NOCMTIME;
308                 inode->i_generation = generation;
309                 inode->i_data.backing_dev_info = &fc->bdi;
310                 fuse_init_inode(inode, attr);
311                 unlock_new_inode(inode);
312         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
313                 /* Inode has changed type, any I/O on the old should fail */
314                 make_bad_inode(inode);
315                 iput(inode);
316                 goto retry;
317         }
318
319         fi = get_fuse_inode(inode);
320         spin_lock(&fc->lock);
321         fi->nlookup++;
322         spin_unlock(&fc->lock);
323         fuse_change_attributes(inode, attr, attr_valid, attr_version);
324
325         return inode;
326 }
327
328 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
329                              loff_t offset, loff_t len)
330 {
331         struct inode *inode;
332         pgoff_t pg_start;
333         pgoff_t pg_end;
334
335         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
336         if (!inode)
337                 return -ENOENT;
338
339         fuse_invalidate_attr(inode);
340         if (offset >= 0) {
341                 pg_start = offset >> PAGE_CACHE_SHIFT;
342                 if (len <= 0)
343                         pg_end = -1;
344                 else
345                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
346                 invalidate_inode_pages2_range(inode->i_mapping,
347                                               pg_start, pg_end);
348         }
349         iput(inode);
350         return 0;
351 }
352
353 static void fuse_umount_begin(struct super_block *sb)
354 {
355         fuse_abort_conn(get_fuse_conn_super(sb));
356 }
357
358 static void fuse_send_destroy(struct fuse_conn *fc)
359 {
360         struct fuse_req *req = fc->destroy_req;
361         if (req && fc->conn_init) {
362                 fc->destroy_req = NULL;
363                 req->in.h.opcode = FUSE_DESTROY;
364                 req->force = 1;
365                 req->background = 0;
366                 fuse_request_send(fc, req);
367                 fuse_put_request(fc, req);
368         }
369 }
370
371 static void fuse_bdi_destroy(struct fuse_conn *fc)
372 {
373         if (fc->bdi_initialized)
374                 bdi_destroy(&fc->bdi);
375 }
376
377 void fuse_conn_kill(struct fuse_conn *fc)
378 {
379         spin_lock(&fc->lock);
380         fc->connected = 0;
381         fc->blocked = 0;
382         fc->initialized = 1;
383         spin_unlock(&fc->lock);
384         /* Flush all readers on this fs */
385         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
386         wake_up_all(&fc->waitq);
387         wake_up_all(&fc->blocked_waitq);
388         wake_up_all(&fc->reserved_req_waitq);
389 }
390 EXPORT_SYMBOL_GPL(fuse_conn_kill);
391
392 static void fuse_put_super(struct super_block *sb)
393 {
394         struct fuse_conn *fc = get_fuse_conn_super(sb);
395
396         fuse_send_destroy(fc);
397
398         fuse_conn_kill(fc);
399         mutex_lock(&fuse_mutex);
400         list_del(&fc->entry);
401         fuse_ctl_remove_conn(fc);
402         mutex_unlock(&fuse_mutex);
403         fuse_bdi_destroy(fc);
404
405         fuse_conn_put(fc);
406 }
407
408 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
409 {
410         stbuf->f_type    = FUSE_SUPER_MAGIC;
411         stbuf->f_bsize   = attr->bsize;
412         stbuf->f_frsize  = attr->frsize;
413         stbuf->f_blocks  = attr->blocks;
414         stbuf->f_bfree   = attr->bfree;
415         stbuf->f_bavail  = attr->bavail;
416         stbuf->f_files   = attr->files;
417         stbuf->f_ffree   = attr->ffree;
418         stbuf->f_namelen = attr->namelen;
419         /* fsid is left zero */
420 }
421
422 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
423 {
424         struct super_block *sb = dentry->d_sb;
425         struct fuse_conn *fc = get_fuse_conn_super(sb);
426         struct fuse_req *req;
427         struct fuse_statfs_out outarg;
428         int err;
429
430         if (!fuse_allow_current_process(fc)) {
431                 buf->f_type = FUSE_SUPER_MAGIC;
432                 return 0;
433         }
434
435         req = fuse_get_req_nopages(fc);
436         if (IS_ERR(req))
437                 return PTR_ERR(req);
438
439         memset(&outarg, 0, sizeof(outarg));
440         req->in.numargs = 0;
441         req->in.h.opcode = FUSE_STATFS;
442         req->in.h.nodeid = get_node_id(dentry->d_inode);
443         req->out.numargs = 1;
444         req->out.args[0].size =
445                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
446         req->out.args[0].value = &outarg;
447         fuse_request_send(fc, req);
448         err = req->out.h.error;
449         if (!err)
450                 convert_fuse_statfs(buf, &outarg.st);
451         fuse_put_request(fc, req);
452         return err;
453 }
454
455 enum {
456         OPT_FD,
457         OPT_ROOTMODE,
458         OPT_USER_ID,
459         OPT_GROUP_ID,
460         OPT_DEFAULT_PERMISSIONS,
461         OPT_ALLOW_OTHER,
462         OPT_MAX_READ,
463         OPT_BLKSIZE,
464         OPT_ERR
465 };
466
467 static const match_table_t tokens = {
468         {OPT_FD,                        "fd=%u"},
469         {OPT_ROOTMODE,                  "rootmode=%o"},
470         {OPT_USER_ID,                   "user_id=%u"},
471         {OPT_GROUP_ID,                  "group_id=%u"},
472         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
473         {OPT_ALLOW_OTHER,               "allow_other"},
474         {OPT_MAX_READ,                  "max_read=%u"},
475         {OPT_BLKSIZE,                   "blksize=%u"},
476         {OPT_ERR,                       NULL}
477 };
478
479 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
480 {
481         char *p;
482         memset(d, 0, sizeof(struct fuse_mount_data));
483         d->max_read = ~0;
484         d->blksize = FUSE_DEFAULT_BLKSIZE;
485
486         while ((p = strsep(&opt, ",")) != NULL) {
487                 int token;
488                 int value;
489                 substring_t args[MAX_OPT_ARGS];
490                 if (!*p)
491                         continue;
492
493                 token = match_token(p, tokens, args);
494                 switch (token) {
495                 case OPT_FD:
496                         if (match_int(&args[0], &value))
497                                 return 0;
498                         d->fd = value;
499                         d->fd_present = 1;
500                         break;
501
502                 case OPT_ROOTMODE:
503                         if (match_octal(&args[0], &value))
504                                 return 0;
505                         if (!fuse_valid_type(value))
506                                 return 0;
507                         d->rootmode = value;
508                         d->rootmode_present = 1;
509                         break;
510
511                 case OPT_USER_ID:
512                         if (match_int(&args[0], &value))
513                                 return 0;
514                         d->user_id = make_kuid(current_user_ns(), value);
515                         if (!uid_valid(d->user_id))
516                                 return 0;
517                         d->user_id_present = 1;
518                         break;
519
520                 case OPT_GROUP_ID:
521                         if (match_int(&args[0], &value))
522                                 return 0;
523                         d->group_id = make_kgid(current_user_ns(), value);
524                         if (!gid_valid(d->group_id))
525                                 return 0;
526                         d->group_id_present = 1;
527                         break;
528
529                 case OPT_DEFAULT_PERMISSIONS:
530                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
531                         break;
532
533                 case OPT_ALLOW_OTHER:
534                         d->flags |= FUSE_ALLOW_OTHER;
535                         break;
536
537                 case OPT_MAX_READ:
538                         if (match_int(&args[0], &value))
539                                 return 0;
540                         d->max_read = value;
541                         break;
542
543                 case OPT_BLKSIZE:
544                         if (!is_bdev || match_int(&args[0], &value))
545                                 return 0;
546                         d->blksize = value;
547                         break;
548
549                 default:
550                         return 0;
551                 }
552         }
553
554         if (!d->fd_present || !d->rootmode_present ||
555             !d->user_id_present || !d->group_id_present)
556                 return 0;
557
558         return 1;
559 }
560
561 static int fuse_show_options(struct seq_file *m, struct dentry *root)
562 {
563         struct super_block *sb = root->d_sb;
564         struct fuse_conn *fc = get_fuse_conn_super(sb);
565
566         seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
567         seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
568         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
569                 seq_puts(m, ",default_permissions");
570         if (fc->flags & FUSE_ALLOW_OTHER)
571                 seq_puts(m, ",allow_other");
572         if (fc->max_read != ~0)
573                 seq_printf(m, ",max_read=%u", fc->max_read);
574         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
575                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
576         return 0;
577 }
578
579 void fuse_conn_init(struct fuse_conn *fc)
580 {
581         memset(fc, 0, sizeof(*fc));
582         spin_lock_init(&fc->lock);
583         init_rwsem(&fc->killsb);
584         atomic_set(&fc->count, 1);
585         init_waitqueue_head(&fc->waitq);
586         init_waitqueue_head(&fc->blocked_waitq);
587         init_waitqueue_head(&fc->reserved_req_waitq);
588         INIT_LIST_HEAD(&fc->pending);
589         INIT_LIST_HEAD(&fc->processing);
590         INIT_LIST_HEAD(&fc->io);
591         INIT_LIST_HEAD(&fc->interrupts);
592         INIT_LIST_HEAD(&fc->bg_queue);
593         INIT_LIST_HEAD(&fc->entry);
594         fc->forget_list_tail = &fc->forget_list_head;
595         atomic_set(&fc->num_waiting, 0);
596         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
597         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
598         fc->khctr = 0;
599         fc->polled_files = RB_ROOT;
600         fc->reqctr = 0;
601         fc->blocked = 0;
602         fc->initialized = 0;
603         fc->attr_version = 1;
604         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
605 }
606 EXPORT_SYMBOL_GPL(fuse_conn_init);
607
608 void fuse_conn_put(struct fuse_conn *fc)
609 {
610         if (atomic_dec_and_test(&fc->count)) {
611                 if (fc->destroy_req)
612                         fuse_request_free(fc->destroy_req);
613                 fc->release(fc);
614         }
615 }
616 EXPORT_SYMBOL_GPL(fuse_conn_put);
617
618 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
619 {
620         atomic_inc(&fc->count);
621         return fc;
622 }
623 EXPORT_SYMBOL_GPL(fuse_conn_get);
624
625 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
626 {
627         struct fuse_attr attr;
628         memset(&attr, 0, sizeof(attr));
629
630         attr.mode = mode;
631         attr.ino = FUSE_ROOT_ID;
632         attr.nlink = 1;
633         return fuse_iget(sb, 1, 0, &attr, 0, 0);
634 }
635
636 struct fuse_inode_handle {
637         u64 nodeid;
638         u32 generation;
639 };
640
641 static struct dentry *fuse_get_dentry(struct super_block *sb,
642                                       struct fuse_inode_handle *handle)
643 {
644         struct fuse_conn *fc = get_fuse_conn_super(sb);
645         struct inode *inode;
646         struct dentry *entry;
647         int err = -ESTALE;
648
649         if (handle->nodeid == 0)
650                 goto out_err;
651
652         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
653         if (!inode) {
654                 struct fuse_entry_out outarg;
655                 struct qstr name;
656
657                 if (!fc->export_support)
658                         goto out_err;
659
660                 name.len = 1;
661                 name.name = ".";
662                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
663                                        &inode);
664                 if (err && err != -ENOENT)
665                         goto out_err;
666                 if (err || !inode) {
667                         err = -ESTALE;
668                         goto out_err;
669                 }
670                 err = -EIO;
671                 if (get_node_id(inode) != handle->nodeid)
672                         goto out_iput;
673         }
674         err = -ESTALE;
675         if (inode->i_generation != handle->generation)
676                 goto out_iput;
677
678         entry = d_obtain_alias(inode);
679         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
680                 fuse_invalidate_entry_cache(entry);
681
682         return entry;
683
684  out_iput:
685         iput(inode);
686  out_err:
687         return ERR_PTR(err);
688 }
689
690 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
691                            struct inode *parent)
692 {
693         int len = parent ? 6 : 3;
694         u64 nodeid;
695         u32 generation;
696
697         if (*max_len < len) {
698                 *max_len = len;
699                 return  FILEID_INVALID;
700         }
701
702         nodeid = get_fuse_inode(inode)->nodeid;
703         generation = inode->i_generation;
704
705         fh[0] = (u32)(nodeid >> 32);
706         fh[1] = (u32)(nodeid & 0xffffffff);
707         fh[2] = generation;
708
709         if (parent) {
710                 nodeid = get_fuse_inode(parent)->nodeid;
711                 generation = parent->i_generation;
712
713                 fh[3] = (u32)(nodeid >> 32);
714                 fh[4] = (u32)(nodeid & 0xffffffff);
715                 fh[5] = generation;
716         }
717
718         *max_len = len;
719         return parent ? 0x82 : 0x81;
720 }
721
722 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
723                 struct fid *fid, int fh_len, int fh_type)
724 {
725         struct fuse_inode_handle handle;
726
727         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
728                 return NULL;
729
730         handle.nodeid = (u64) fid->raw[0] << 32;
731         handle.nodeid |= (u64) fid->raw[1];
732         handle.generation = fid->raw[2];
733         return fuse_get_dentry(sb, &handle);
734 }
735
736 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
737                 struct fid *fid, int fh_len, int fh_type)
738 {
739         struct fuse_inode_handle parent;
740
741         if (fh_type != 0x82 || fh_len < 6)
742                 return NULL;
743
744         parent.nodeid = (u64) fid->raw[3] << 32;
745         parent.nodeid |= (u64) fid->raw[4];
746         parent.generation = fid->raw[5];
747         return fuse_get_dentry(sb, &parent);
748 }
749
750 static struct dentry *fuse_get_parent(struct dentry *child)
751 {
752         struct inode *child_inode = child->d_inode;
753         struct fuse_conn *fc = get_fuse_conn(child_inode);
754         struct inode *inode;
755         struct dentry *parent;
756         struct fuse_entry_out outarg;
757         struct qstr name;
758         int err;
759
760         if (!fc->export_support)
761                 return ERR_PTR(-ESTALE);
762
763         name.len = 2;
764         name.name = "..";
765         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
766                                &name, &outarg, &inode);
767         if (err) {
768                 if (err == -ENOENT)
769                         return ERR_PTR(-ESTALE);
770                 return ERR_PTR(err);
771         }
772
773         parent = d_obtain_alias(inode);
774         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
775                 fuse_invalidate_entry_cache(parent);
776
777         return parent;
778 }
779
780 static const struct export_operations fuse_export_operations = {
781         .fh_to_dentry   = fuse_fh_to_dentry,
782         .fh_to_parent   = fuse_fh_to_parent,
783         .encode_fh      = fuse_encode_fh,
784         .get_parent     = fuse_get_parent,
785 };
786
787 static const struct super_operations fuse_super_operations = {
788         .alloc_inode    = fuse_alloc_inode,
789         .destroy_inode  = fuse_destroy_inode,
790         .evict_inode    = fuse_evict_inode,
791         .drop_inode     = generic_delete_inode,
792         .remount_fs     = fuse_remount_fs,
793         .put_super      = fuse_put_super,
794         .umount_begin   = fuse_umount_begin,
795         .statfs         = fuse_statfs,
796         .show_options   = fuse_show_options,
797 };
798
799 static void sanitize_global_limit(unsigned *limit)
800 {
801         if (*limit == 0)
802                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
803                          sizeof(struct fuse_req);
804
805         if (*limit >= 1 << 16)
806                 *limit = (1 << 16) - 1;
807 }
808
809 static int set_global_limit(const char *val, struct kernel_param *kp)
810 {
811         int rv;
812
813         rv = param_set_uint(val, kp);
814         if (rv)
815                 return rv;
816
817         sanitize_global_limit((unsigned *)kp->arg);
818
819         return 0;
820 }
821
822 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
823 {
824         int cap_sys_admin = capable(CAP_SYS_ADMIN);
825
826         if (arg->minor < 13)
827                 return;
828
829         sanitize_global_limit(&max_user_bgreq);
830         sanitize_global_limit(&max_user_congthresh);
831
832         if (arg->max_background) {
833                 fc->max_background = arg->max_background;
834
835                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
836                         fc->max_background = max_user_bgreq;
837         }
838         if (arg->congestion_threshold) {
839                 fc->congestion_threshold = arg->congestion_threshold;
840
841                 if (!cap_sys_admin &&
842                     fc->congestion_threshold > max_user_congthresh)
843                         fc->congestion_threshold = max_user_congthresh;
844         }
845 }
846
847 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
848 {
849         struct fuse_init_out *arg = &req->misc.init_out;
850
851         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
852                 fc->conn_error = 1;
853         else {
854                 unsigned long ra_pages;
855
856                 process_init_limits(fc, arg);
857
858                 if (arg->minor >= 6) {
859                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
860                         if (arg->flags & FUSE_ASYNC_READ)
861                                 fc->async_read = 1;
862                         if (!(arg->flags & FUSE_POSIX_LOCKS))
863                                 fc->no_lock = 1;
864                         if (arg->minor >= 17) {
865                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
866                                         fc->no_flock = 1;
867                         } else {
868                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
869                                         fc->no_flock = 1;
870                         }
871                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
872                                 fc->atomic_o_trunc = 1;
873                         if (arg->minor >= 9) {
874                                 /* LOOKUP has dependency on proto version */
875                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
876                                         fc->export_support = 1;
877                         }
878                         if (arg->flags & FUSE_BIG_WRITES)
879                                 fc->big_writes = 1;
880                         if (arg->flags & FUSE_DONT_MASK)
881                                 fc->dont_mask = 1;
882                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
883                                 fc->auto_inval_data = 1;
884                         if (arg->flags & FUSE_DO_READDIRPLUS) {
885                                 fc->do_readdirplus = 1;
886                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
887                                         fc->readdirplus_auto = 1;
888                         }
889                         if (arg->flags & FUSE_ASYNC_DIO)
890                                 fc->async_dio = 1;
891                         if (arg->flags & FUSE_WRITEBACK_CACHE)
892                                 fc->writeback_cache = 1;
893                 } else {
894                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
895                         fc->no_lock = 1;
896                         fc->no_flock = 1;
897                 }
898
899                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
900                 fc->minor = arg->minor;
901                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
902                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
903                 fc->conn_init = 1;
904         }
905         fc->initialized = 1;
906         wake_up_all(&fc->blocked_waitq);
907 }
908
909 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
910 {
911         struct fuse_init_in *arg = &req->misc.init_in;
912
913         arg->major = FUSE_KERNEL_VERSION;
914         arg->minor = FUSE_KERNEL_MINOR_VERSION;
915         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
916         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
917                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
918                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
919                 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
920                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
921                 FUSE_WRITEBACK_CACHE;
922         req->in.h.opcode = FUSE_INIT;
923         req->in.numargs = 1;
924         req->in.args[0].size = sizeof(*arg);
925         req->in.args[0].value = arg;
926         req->out.numargs = 1;
927         /* Variable length argument used for backward compatibility
928            with interface version < 7.5.  Rest of init_out is zeroed
929            by do_get_request(), so a short reply is not a problem */
930         req->out.argvar = 1;
931         req->out.args[0].size = sizeof(struct fuse_init_out);
932         req->out.args[0].value = &req->misc.init_out;
933         req->end = process_init_reply;
934         fuse_request_send_background(fc, req);
935 }
936
937 static void fuse_free_conn(struct fuse_conn *fc)
938 {
939         kfree_rcu(fc, rcu);
940 }
941
942 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
943 {
944         int err;
945
946         fc->bdi.name = "fuse";
947         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
948         /* fuse does it's own writeback accounting */
949         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
950
951         err = bdi_init(&fc->bdi);
952         if (err)
953                 return err;
954
955         fc->bdi_initialized = 1;
956
957         if (sb->s_bdev) {
958                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
959                                     MAJOR(fc->dev), MINOR(fc->dev));
960         } else {
961                 err = bdi_register_dev(&fc->bdi, fc->dev);
962         }
963
964         if (err)
965                 return err;
966
967         /*
968          * For a single fuse filesystem use max 1% of dirty +
969          * writeback threshold.
970          *
971          * This gives about 1M of write buffer for memory maps on a
972          * machine with 1G and 10% dirty_ratio, which should be more
973          * than enough.
974          *
975          * Privileged users can raise it by writing to
976          *
977          *    /sys/class/bdi/<bdi>/max_ratio
978          */
979         bdi_set_max_ratio(&fc->bdi, 1);
980
981         return 0;
982 }
983
984 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
985 {
986         struct fuse_conn *fc;
987         struct inode *root;
988         struct fuse_mount_data d;
989         struct file *file;
990         struct dentry *root_dentry;
991         struct fuse_req *init_req;
992         int err;
993         int is_bdev = sb->s_bdev != NULL;
994
995         err = -EINVAL;
996         if (sb->s_flags & MS_MANDLOCK)
997                 goto err;
998
999         sb->s_flags &= ~MS_NOSEC;
1000
1001         if (!parse_fuse_opt((char *) data, &d, is_bdev))
1002                 goto err;
1003
1004         if (is_bdev) {
1005 #ifdef CONFIG_BLOCK
1006                 err = -EINVAL;
1007                 if (!sb_set_blocksize(sb, d.blksize))
1008                         goto err;
1009 #endif
1010         } else {
1011                 sb->s_blocksize = PAGE_CACHE_SIZE;
1012                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1013         }
1014         sb->s_magic = FUSE_SUPER_MAGIC;
1015         sb->s_op = &fuse_super_operations;
1016         sb->s_maxbytes = MAX_LFS_FILESIZE;
1017         sb->s_time_gran = 1;
1018         sb->s_export_op = &fuse_export_operations;
1019
1020         file = fget(d.fd);
1021         err = -EINVAL;
1022         if (!file)
1023                 goto err;
1024
1025         if ((file->f_op != &fuse_dev_operations) ||
1026             (file->f_cred->user_ns != &init_user_ns))
1027                 goto err_fput;
1028
1029         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1030         err = -ENOMEM;
1031         if (!fc)
1032                 goto err_fput;
1033
1034         fuse_conn_init(fc);
1035
1036         fc->dev = sb->s_dev;
1037         fc->sb = sb;
1038         err = fuse_bdi_init(fc, sb);
1039         if (err)
1040                 goto err_put_conn;
1041
1042         sb->s_bdi = &fc->bdi;
1043
1044         /* Handle umasking inside the fuse code */
1045         if (sb->s_flags & MS_POSIXACL)
1046                 fc->dont_mask = 1;
1047         sb->s_flags |= MS_POSIXACL;
1048
1049         fc->release = fuse_free_conn;
1050         fc->flags = d.flags;
1051         fc->user_id = d.user_id;
1052         fc->group_id = d.group_id;
1053         fc->max_read = max_t(unsigned, 4096, d.max_read);
1054
1055         /* Used by get_root_inode() */
1056         sb->s_fs_info = fc;
1057
1058         err = -ENOMEM;
1059         root = fuse_get_root_inode(sb, d.rootmode);
1060         root_dentry = d_make_root(root);
1061         if (!root_dentry)
1062                 goto err_put_conn;
1063         /* only now - we want root dentry with NULL ->d_op */
1064         sb->s_d_op = &fuse_dentry_operations;
1065
1066         init_req = fuse_request_alloc(0);
1067         if (!init_req)
1068                 goto err_put_root;
1069         init_req->background = 1;
1070
1071         if (is_bdev) {
1072                 fc->destroy_req = fuse_request_alloc(0);
1073                 if (!fc->destroy_req)
1074                         goto err_free_init_req;
1075         }
1076
1077         mutex_lock(&fuse_mutex);
1078         err = -EINVAL;
1079         if (file->private_data)
1080                 goto err_unlock;
1081
1082         err = fuse_ctl_add_conn(fc);
1083         if (err)
1084                 goto err_unlock;
1085
1086         list_add_tail(&fc->entry, &fuse_conn_list);
1087         sb->s_root = root_dentry;
1088         fc->connected = 1;
1089         file->private_data = fuse_conn_get(fc);
1090         mutex_unlock(&fuse_mutex);
1091         /*
1092          * atomic_dec_and_test() in fput() provides the necessary
1093          * memory barrier for file->private_data to be visible on all
1094          * CPUs after this
1095          */
1096         fput(file);
1097
1098         fuse_send_init(fc, init_req);
1099
1100         return 0;
1101
1102  err_unlock:
1103         mutex_unlock(&fuse_mutex);
1104  err_free_init_req:
1105         fuse_request_free(init_req);
1106  err_put_root:
1107         dput(root_dentry);
1108  err_put_conn:
1109         fuse_bdi_destroy(fc);
1110         fuse_conn_put(fc);
1111  err_fput:
1112         fput(file);
1113  err:
1114         return err;
1115 }
1116
1117 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1118                        int flags, const char *dev_name,
1119                        void *raw_data)
1120 {
1121         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1122 }
1123
1124 static void fuse_kill_sb_anon(struct super_block *sb)
1125 {
1126         struct fuse_conn *fc = get_fuse_conn_super(sb);
1127
1128         if (fc) {
1129                 down_write(&fc->killsb);
1130                 fc->sb = NULL;
1131                 up_write(&fc->killsb);
1132         }
1133
1134         kill_anon_super(sb);
1135 }
1136
1137 static struct file_system_type fuse_fs_type = {
1138         .owner          = THIS_MODULE,
1139         .name           = "fuse",
1140         .fs_flags       = FS_HAS_SUBTYPE,
1141         .mount          = fuse_mount,
1142         .kill_sb        = fuse_kill_sb_anon,
1143 };
1144 MODULE_ALIAS_FS("fuse");
1145
1146 #ifdef CONFIG_BLOCK
1147 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1148                            int flags, const char *dev_name,
1149                            void *raw_data)
1150 {
1151         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1152 }
1153
1154 static void fuse_kill_sb_blk(struct super_block *sb)
1155 {
1156         struct fuse_conn *fc = get_fuse_conn_super(sb);
1157
1158         if (fc) {
1159                 down_write(&fc->killsb);
1160                 fc->sb = NULL;
1161                 up_write(&fc->killsb);
1162         }
1163
1164         kill_block_super(sb);
1165 }
1166
1167 static struct file_system_type fuseblk_fs_type = {
1168         .owner          = THIS_MODULE,
1169         .name           = "fuseblk",
1170         .mount          = fuse_mount_blk,
1171         .kill_sb        = fuse_kill_sb_blk,
1172         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1173 };
1174 MODULE_ALIAS_FS("fuseblk");
1175
1176 static inline int register_fuseblk(void)
1177 {
1178         return register_filesystem(&fuseblk_fs_type);
1179 }
1180
1181 static inline void unregister_fuseblk(void)
1182 {
1183         unregister_filesystem(&fuseblk_fs_type);
1184 }
1185 #else
1186 static inline int register_fuseblk(void)
1187 {
1188         return 0;
1189 }
1190
1191 static inline void unregister_fuseblk(void)
1192 {
1193 }
1194 #endif
1195
1196 static void fuse_inode_init_once(void *foo)
1197 {
1198         struct inode *inode = foo;
1199
1200         inode_init_once(inode);
1201 }
1202
1203 static int __init fuse_fs_init(void)
1204 {
1205         int err;
1206
1207         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1208                                               sizeof(struct fuse_inode),
1209                                               0, SLAB_HWCACHE_ALIGN,
1210                                               fuse_inode_init_once);
1211         err = -ENOMEM;
1212         if (!fuse_inode_cachep)
1213                 goto out;
1214
1215         err = register_fuseblk();
1216         if (err)
1217                 goto out2;
1218
1219         err = register_filesystem(&fuse_fs_type);
1220         if (err)
1221                 goto out3;
1222
1223         return 0;
1224
1225  out3:
1226         unregister_fuseblk();
1227  out2:
1228         kmem_cache_destroy(fuse_inode_cachep);
1229  out:
1230         return err;
1231 }
1232
1233 static void fuse_fs_cleanup(void)
1234 {
1235         unregister_filesystem(&fuse_fs_type);
1236         unregister_fuseblk();
1237
1238         /*
1239          * Make sure all delayed rcu free inodes are flushed before we
1240          * destroy cache.
1241          */
1242         rcu_barrier();
1243         kmem_cache_destroy(fuse_inode_cachep);
1244 }
1245
1246 static struct kobject *fuse_kobj;
1247 static struct kobject *connections_kobj;
1248
1249 static int fuse_sysfs_init(void)
1250 {
1251         int err;
1252
1253         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1254         if (!fuse_kobj) {
1255                 err = -ENOMEM;
1256                 goto out_err;
1257         }
1258
1259         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1260         if (!connections_kobj) {
1261                 err = -ENOMEM;
1262                 goto out_fuse_unregister;
1263         }
1264
1265         return 0;
1266
1267  out_fuse_unregister:
1268         kobject_put(fuse_kobj);
1269  out_err:
1270         return err;
1271 }
1272
1273 static void fuse_sysfs_cleanup(void)
1274 {
1275         kobject_put(connections_kobj);
1276         kobject_put(fuse_kobj);
1277 }
1278
1279 static int __init fuse_init(void)
1280 {
1281         int res;
1282
1283         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1284                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1285
1286         INIT_LIST_HEAD(&fuse_conn_list);
1287         res = fuse_fs_init();
1288         if (res)
1289                 goto err;
1290
1291         res = fuse_dev_init();
1292         if (res)
1293                 goto err_fs_cleanup;
1294
1295         res = fuse_sysfs_init();
1296         if (res)
1297                 goto err_dev_cleanup;
1298
1299         res = fuse_ctl_init();
1300         if (res)
1301                 goto err_sysfs_cleanup;
1302
1303         sanitize_global_limit(&max_user_bgreq);
1304         sanitize_global_limit(&max_user_congthresh);
1305
1306         return 0;
1307
1308  err_sysfs_cleanup:
1309         fuse_sysfs_cleanup();
1310  err_dev_cleanup:
1311         fuse_dev_cleanup();
1312  err_fs_cleanup:
1313         fuse_fs_cleanup();
1314  err:
1315         return res;
1316 }
1317
1318 static void __exit fuse_exit(void)
1319 {
1320         printk(KERN_DEBUG "fuse exit\n");
1321
1322         fuse_ctl_cleanup();
1323         fuse_sysfs_cleanup();
1324         fuse_fs_cleanup();
1325         fuse_dev_cleanup();
1326 }
1327
1328 module_init(fuse_init);
1329 module_exit(fuse_exit);