]> git.karo-electronics.de Git - karo-tx-linux.git/blob - fs/fuse/inode.c
Merge remote-tracking branch 'grant/devicetree/next' into for-next
[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 static void fuse_put_super(struct super_block *sb)
380 {
381         struct fuse_conn *fc = get_fuse_conn_super(sb);
382
383         fuse_send_destroy(fc);
384
385         fuse_abort_conn(fc);
386         mutex_lock(&fuse_mutex);
387         list_del(&fc->entry);
388         fuse_ctl_remove_conn(fc);
389         mutex_unlock(&fuse_mutex);
390         fuse_bdi_destroy(fc);
391
392         fuse_conn_put(fc);
393 }
394
395 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
396 {
397         stbuf->f_type    = FUSE_SUPER_MAGIC;
398         stbuf->f_bsize   = attr->bsize;
399         stbuf->f_frsize  = attr->frsize;
400         stbuf->f_blocks  = attr->blocks;
401         stbuf->f_bfree   = attr->bfree;
402         stbuf->f_bavail  = attr->bavail;
403         stbuf->f_files   = attr->files;
404         stbuf->f_ffree   = attr->ffree;
405         stbuf->f_namelen = attr->namelen;
406         /* fsid is left zero */
407 }
408
409 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
410 {
411         struct super_block *sb = dentry->d_sb;
412         struct fuse_conn *fc = get_fuse_conn_super(sb);
413         FUSE_ARGS(args);
414         struct fuse_statfs_out outarg;
415         int err;
416
417         if (!fuse_allow_current_process(fc)) {
418                 buf->f_type = FUSE_SUPER_MAGIC;
419                 return 0;
420         }
421
422         memset(&outarg, 0, sizeof(outarg));
423         args.in.numargs = 0;
424         args.in.h.opcode = FUSE_STATFS;
425         args.in.h.nodeid = get_node_id(dentry->d_inode);
426         args.out.numargs = 1;
427         args.out.args[0].size = sizeof(outarg);
428         args.out.args[0].value = &outarg;
429         err = fuse_simple_request(fc, &args);
430         if (!err)
431                 convert_fuse_statfs(buf, &outarg.st);
432         return err;
433 }
434
435 enum {
436         OPT_FD,
437         OPT_ROOTMODE,
438         OPT_USER_ID,
439         OPT_GROUP_ID,
440         OPT_DEFAULT_PERMISSIONS,
441         OPT_ALLOW_OTHER,
442         OPT_MAX_READ,
443         OPT_BLKSIZE,
444         OPT_ERR
445 };
446
447 static const match_table_t tokens = {
448         {OPT_FD,                        "fd=%u"},
449         {OPT_ROOTMODE,                  "rootmode=%o"},
450         {OPT_USER_ID,                   "user_id=%u"},
451         {OPT_GROUP_ID,                  "group_id=%u"},
452         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
453         {OPT_ALLOW_OTHER,               "allow_other"},
454         {OPT_MAX_READ,                  "max_read=%u"},
455         {OPT_BLKSIZE,                   "blksize=%u"},
456         {OPT_ERR,                       NULL}
457 };
458
459 static int fuse_match_uint(substring_t *s, unsigned int *res)
460 {
461         int err = -ENOMEM;
462         char *buf = match_strdup(s);
463         if (buf) {
464                 err = kstrtouint(buf, 10, res);
465                 kfree(buf);
466         }
467         return err;
468 }
469
470 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
471 {
472         char *p;
473         memset(d, 0, sizeof(struct fuse_mount_data));
474         d->max_read = ~0;
475         d->blksize = FUSE_DEFAULT_BLKSIZE;
476
477         while ((p = strsep(&opt, ",")) != NULL) {
478                 int token;
479                 int value;
480                 unsigned uv;
481                 substring_t args[MAX_OPT_ARGS];
482                 if (!*p)
483                         continue;
484
485                 token = match_token(p, tokens, args);
486                 switch (token) {
487                 case OPT_FD:
488                         if (match_int(&args[0], &value))
489                                 return 0;
490                         d->fd = value;
491                         d->fd_present = 1;
492                         break;
493
494                 case OPT_ROOTMODE:
495                         if (match_octal(&args[0], &value))
496                                 return 0;
497                         if (!fuse_valid_type(value))
498                                 return 0;
499                         d->rootmode = value;
500                         d->rootmode_present = 1;
501                         break;
502
503                 case OPT_USER_ID:
504                         if (fuse_match_uint(&args[0], &uv))
505                                 return 0;
506                         d->user_id = make_kuid(current_user_ns(), uv);
507                         if (!uid_valid(d->user_id))
508                                 return 0;
509                         d->user_id_present = 1;
510                         break;
511
512                 case OPT_GROUP_ID:
513                         if (fuse_match_uint(&args[0], &uv))
514                                 return 0;
515                         d->group_id = make_kgid(current_user_ns(), uv);
516                         if (!gid_valid(d->group_id))
517                                 return 0;
518                         d->group_id_present = 1;
519                         break;
520
521                 case OPT_DEFAULT_PERMISSIONS:
522                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
523                         break;
524
525                 case OPT_ALLOW_OTHER:
526                         d->flags |= FUSE_ALLOW_OTHER;
527                         break;
528
529                 case OPT_MAX_READ:
530                         if (match_int(&args[0], &value))
531                                 return 0;
532                         d->max_read = value;
533                         break;
534
535                 case OPT_BLKSIZE:
536                         if (!is_bdev || match_int(&args[0], &value))
537                                 return 0;
538                         d->blksize = value;
539                         break;
540
541                 default:
542                         return 0;
543                 }
544         }
545
546         if (!d->fd_present || !d->rootmode_present ||
547             !d->user_id_present || !d->group_id_present)
548                 return 0;
549
550         return 1;
551 }
552
553 static int fuse_show_options(struct seq_file *m, struct dentry *root)
554 {
555         struct super_block *sb = root->d_sb;
556         struct fuse_conn *fc = get_fuse_conn_super(sb);
557
558         seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
559         seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
560         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
561                 seq_puts(m, ",default_permissions");
562         if (fc->flags & FUSE_ALLOW_OTHER)
563                 seq_puts(m, ",allow_other");
564         if (fc->max_read != ~0)
565                 seq_printf(m, ",max_read=%u", fc->max_read);
566         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
567                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
568         return 0;
569 }
570
571 void fuse_conn_init(struct fuse_conn *fc)
572 {
573         memset(fc, 0, sizeof(*fc));
574         spin_lock_init(&fc->lock);
575         init_rwsem(&fc->killsb);
576         atomic_set(&fc->count, 1);
577         init_waitqueue_head(&fc->waitq);
578         init_waitqueue_head(&fc->blocked_waitq);
579         init_waitqueue_head(&fc->reserved_req_waitq);
580         INIT_LIST_HEAD(&fc->pending);
581         INIT_LIST_HEAD(&fc->processing);
582         INIT_LIST_HEAD(&fc->io);
583         INIT_LIST_HEAD(&fc->interrupts);
584         INIT_LIST_HEAD(&fc->bg_queue);
585         INIT_LIST_HEAD(&fc->entry);
586         fc->forget_list_tail = &fc->forget_list_head;
587         atomic_set(&fc->num_waiting, 0);
588         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
589         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
590         fc->khctr = 0;
591         fc->polled_files = RB_ROOT;
592         fc->reqctr = 0;
593         fc->blocked = 0;
594         fc->initialized = 0;
595         fc->attr_version = 1;
596         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
597 }
598 EXPORT_SYMBOL_GPL(fuse_conn_init);
599
600 void fuse_conn_put(struct fuse_conn *fc)
601 {
602         if (atomic_dec_and_test(&fc->count)) {
603                 if (fc->destroy_req)
604                         fuse_request_free(fc->destroy_req);
605                 fc->release(fc);
606         }
607 }
608 EXPORT_SYMBOL_GPL(fuse_conn_put);
609
610 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
611 {
612         atomic_inc(&fc->count);
613         return fc;
614 }
615 EXPORT_SYMBOL_GPL(fuse_conn_get);
616
617 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
618 {
619         struct fuse_attr attr;
620         memset(&attr, 0, sizeof(attr));
621
622         attr.mode = mode;
623         attr.ino = FUSE_ROOT_ID;
624         attr.nlink = 1;
625         return fuse_iget(sb, 1, 0, &attr, 0, 0);
626 }
627
628 struct fuse_inode_handle {
629         u64 nodeid;
630         u32 generation;
631 };
632
633 static struct dentry *fuse_get_dentry(struct super_block *sb,
634                                       struct fuse_inode_handle *handle)
635 {
636         struct fuse_conn *fc = get_fuse_conn_super(sb);
637         struct inode *inode;
638         struct dentry *entry;
639         int err = -ESTALE;
640
641         if (handle->nodeid == 0)
642                 goto out_err;
643
644         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
645         if (!inode) {
646                 struct fuse_entry_out outarg;
647                 struct qstr name;
648
649                 if (!fc->export_support)
650                         goto out_err;
651
652                 name.len = 1;
653                 name.name = ".";
654                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
655                                        &inode);
656                 if (err && err != -ENOENT)
657                         goto out_err;
658                 if (err || !inode) {
659                         err = -ESTALE;
660                         goto out_err;
661                 }
662                 err = -EIO;
663                 if (get_node_id(inode) != handle->nodeid)
664                         goto out_iput;
665         }
666         err = -ESTALE;
667         if (inode->i_generation != handle->generation)
668                 goto out_iput;
669
670         entry = d_obtain_alias(inode);
671         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
672                 fuse_invalidate_entry_cache(entry);
673
674         return entry;
675
676  out_iput:
677         iput(inode);
678  out_err:
679         return ERR_PTR(err);
680 }
681
682 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
683                            struct inode *parent)
684 {
685         int len = parent ? 6 : 3;
686         u64 nodeid;
687         u32 generation;
688
689         if (*max_len < len) {
690                 *max_len = len;
691                 return  FILEID_INVALID;
692         }
693
694         nodeid = get_fuse_inode(inode)->nodeid;
695         generation = inode->i_generation;
696
697         fh[0] = (u32)(nodeid >> 32);
698         fh[1] = (u32)(nodeid & 0xffffffff);
699         fh[2] = generation;
700
701         if (parent) {
702                 nodeid = get_fuse_inode(parent)->nodeid;
703                 generation = parent->i_generation;
704
705                 fh[3] = (u32)(nodeid >> 32);
706                 fh[4] = (u32)(nodeid & 0xffffffff);
707                 fh[5] = generation;
708         }
709
710         *max_len = len;
711         return parent ? 0x82 : 0x81;
712 }
713
714 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
715                 struct fid *fid, int fh_len, int fh_type)
716 {
717         struct fuse_inode_handle handle;
718
719         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
720                 return NULL;
721
722         handle.nodeid = (u64) fid->raw[0] << 32;
723         handle.nodeid |= (u64) fid->raw[1];
724         handle.generation = fid->raw[2];
725         return fuse_get_dentry(sb, &handle);
726 }
727
728 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
729                 struct fid *fid, int fh_len, int fh_type)
730 {
731         struct fuse_inode_handle parent;
732
733         if (fh_type != 0x82 || fh_len < 6)
734                 return NULL;
735
736         parent.nodeid = (u64) fid->raw[3] << 32;
737         parent.nodeid |= (u64) fid->raw[4];
738         parent.generation = fid->raw[5];
739         return fuse_get_dentry(sb, &parent);
740 }
741
742 static struct dentry *fuse_get_parent(struct dentry *child)
743 {
744         struct inode *child_inode = child->d_inode;
745         struct fuse_conn *fc = get_fuse_conn(child_inode);
746         struct inode *inode;
747         struct dentry *parent;
748         struct fuse_entry_out outarg;
749         struct qstr name;
750         int err;
751
752         if (!fc->export_support)
753                 return ERR_PTR(-ESTALE);
754
755         name.len = 2;
756         name.name = "..";
757         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
758                                &name, &outarg, &inode);
759         if (err) {
760                 if (err == -ENOENT)
761                         return ERR_PTR(-ESTALE);
762                 return ERR_PTR(err);
763         }
764
765         parent = d_obtain_alias(inode);
766         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
767                 fuse_invalidate_entry_cache(parent);
768
769         return parent;
770 }
771
772 static const struct export_operations fuse_export_operations = {
773         .fh_to_dentry   = fuse_fh_to_dentry,
774         .fh_to_parent   = fuse_fh_to_parent,
775         .encode_fh      = fuse_encode_fh,
776         .get_parent     = fuse_get_parent,
777 };
778
779 static const struct super_operations fuse_super_operations = {
780         .alloc_inode    = fuse_alloc_inode,
781         .destroy_inode  = fuse_destroy_inode,
782         .evict_inode    = fuse_evict_inode,
783         .write_inode    = fuse_write_inode,
784         .drop_inode     = generic_delete_inode,
785         .remount_fs     = fuse_remount_fs,
786         .put_super      = fuse_put_super,
787         .umount_begin   = fuse_umount_begin,
788         .statfs         = fuse_statfs,
789         .show_options   = fuse_show_options,
790 };
791
792 static void sanitize_global_limit(unsigned *limit)
793 {
794         if (*limit == 0)
795                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
796                          sizeof(struct fuse_req);
797
798         if (*limit >= 1 << 16)
799                 *limit = (1 << 16) - 1;
800 }
801
802 static int set_global_limit(const char *val, struct kernel_param *kp)
803 {
804         int rv;
805
806         rv = param_set_uint(val, kp);
807         if (rv)
808                 return rv;
809
810         sanitize_global_limit((unsigned *)kp->arg);
811
812         return 0;
813 }
814
815 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
816 {
817         int cap_sys_admin = capable(CAP_SYS_ADMIN);
818
819         if (arg->minor < 13)
820                 return;
821
822         sanitize_global_limit(&max_user_bgreq);
823         sanitize_global_limit(&max_user_congthresh);
824
825         if (arg->max_background) {
826                 fc->max_background = arg->max_background;
827
828                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
829                         fc->max_background = max_user_bgreq;
830         }
831         if (arg->congestion_threshold) {
832                 fc->congestion_threshold = arg->congestion_threshold;
833
834                 if (!cap_sys_admin &&
835                     fc->congestion_threshold > max_user_congthresh)
836                         fc->congestion_threshold = max_user_congthresh;
837         }
838 }
839
840 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
841 {
842         struct fuse_init_out *arg = &req->misc.init_out;
843
844         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
845                 fc->conn_error = 1;
846         else {
847                 unsigned long ra_pages;
848
849                 process_init_limits(fc, arg);
850
851                 if (arg->minor >= 6) {
852                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
853                         if (arg->flags & FUSE_ASYNC_READ)
854                                 fc->async_read = 1;
855                         if (!(arg->flags & FUSE_POSIX_LOCKS))
856                                 fc->no_lock = 1;
857                         if (arg->minor >= 17) {
858                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
859                                         fc->no_flock = 1;
860                         } else {
861                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
862                                         fc->no_flock = 1;
863                         }
864                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
865                                 fc->atomic_o_trunc = 1;
866                         if (arg->minor >= 9) {
867                                 /* LOOKUP has dependency on proto version */
868                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
869                                         fc->export_support = 1;
870                         }
871                         if (arg->flags & FUSE_BIG_WRITES)
872                                 fc->big_writes = 1;
873                         if (arg->flags & FUSE_DONT_MASK)
874                                 fc->dont_mask = 1;
875                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
876                                 fc->auto_inval_data = 1;
877                         if (arg->flags & FUSE_DO_READDIRPLUS) {
878                                 fc->do_readdirplus = 1;
879                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
880                                         fc->readdirplus_auto = 1;
881                         }
882                         if (arg->flags & FUSE_ASYNC_DIO)
883                                 fc->async_dio = 1;
884                         if (arg->flags & FUSE_WRITEBACK_CACHE)
885                                 fc->writeback_cache = 1;
886                         if (arg->time_gran && arg->time_gran <= 1000000000)
887                                 fc->sb->s_time_gran = arg->time_gran;
888                 } else {
889                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
890                         fc->no_lock = 1;
891                         fc->no_flock = 1;
892                 }
893
894                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
895                 fc->minor = arg->minor;
896                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
897                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
898                 fc->conn_init = 1;
899         }
900         fuse_set_initialized(fc);
901         wake_up_all(&fc->blocked_waitq);
902 }
903
904 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
905 {
906         struct fuse_init_in *arg = &req->misc.init_in;
907
908         arg->major = FUSE_KERNEL_VERSION;
909         arg->minor = FUSE_KERNEL_MINOR_VERSION;
910         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
911         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
912                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
913                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
914                 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
915                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
916                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT;
917         req->in.h.opcode = FUSE_INIT;
918         req->in.numargs = 1;
919         req->in.args[0].size = sizeof(*arg);
920         req->in.args[0].value = arg;
921         req->out.numargs = 1;
922         /* Variable length argument used for backward compatibility
923            with interface version < 7.5.  Rest of init_out is zeroed
924            by do_get_request(), so a short reply is not a problem */
925         req->out.argvar = 1;
926         req->out.args[0].size = sizeof(struct fuse_init_out);
927         req->out.args[0].value = &req->misc.init_out;
928         req->end = process_init_reply;
929         fuse_request_send_background(fc, req);
930 }
931
932 static void fuse_free_conn(struct fuse_conn *fc)
933 {
934         kfree_rcu(fc, rcu);
935 }
936
937 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
938 {
939         int err;
940
941         fc->bdi.name = "fuse";
942         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
943         /* fuse does it's own writeback accounting */
944         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
945
946         err = bdi_init(&fc->bdi);
947         if (err)
948                 return err;
949
950         fc->bdi_initialized = 1;
951
952         if (sb->s_bdev) {
953                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
954                                     MAJOR(fc->dev), MINOR(fc->dev));
955         } else {
956                 err = bdi_register_dev(&fc->bdi, fc->dev);
957         }
958
959         if (err)
960                 return err;
961
962         /*
963          * For a single fuse filesystem use max 1% of dirty +
964          * writeback threshold.
965          *
966          * This gives about 1M of write buffer for memory maps on a
967          * machine with 1G and 10% dirty_ratio, which should be more
968          * than enough.
969          *
970          * Privileged users can raise it by writing to
971          *
972          *    /sys/class/bdi/<bdi>/max_ratio
973          */
974         bdi_set_max_ratio(&fc->bdi, 1);
975
976         return 0;
977 }
978
979 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
980 {
981         struct fuse_conn *fc;
982         struct inode *root;
983         struct fuse_mount_data d;
984         struct file *file;
985         struct dentry *root_dentry;
986         struct fuse_req *init_req;
987         int err;
988         int is_bdev = sb->s_bdev != NULL;
989
990         err = -EINVAL;
991         if (sb->s_flags & MS_MANDLOCK)
992                 goto err;
993
994         sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
995
996         if (!parse_fuse_opt(data, &d, is_bdev))
997                 goto err;
998
999         if (is_bdev) {
1000 #ifdef CONFIG_BLOCK
1001                 err = -EINVAL;
1002                 if (!sb_set_blocksize(sb, d.blksize))
1003                         goto err;
1004 #endif
1005         } else {
1006                 sb->s_blocksize = PAGE_CACHE_SIZE;
1007                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1008         }
1009         sb->s_magic = FUSE_SUPER_MAGIC;
1010         sb->s_op = &fuse_super_operations;
1011         sb->s_maxbytes = MAX_LFS_FILESIZE;
1012         sb->s_time_gran = 1;
1013         sb->s_export_op = &fuse_export_operations;
1014
1015         file = fget(d.fd);
1016         err = -EINVAL;
1017         if (!file)
1018                 goto err;
1019
1020         if ((file->f_op != &fuse_dev_operations) ||
1021             (file->f_cred->user_ns != &init_user_ns))
1022                 goto err_fput;
1023
1024         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1025         err = -ENOMEM;
1026         if (!fc)
1027                 goto err_fput;
1028
1029         fuse_conn_init(fc);
1030
1031         fc->dev = sb->s_dev;
1032         fc->sb = sb;
1033         err = fuse_bdi_init(fc, sb);
1034         if (err)
1035                 goto err_put_conn;
1036
1037         sb->s_bdi = &fc->bdi;
1038
1039         /* Handle umasking inside the fuse code */
1040         if (sb->s_flags & MS_POSIXACL)
1041                 fc->dont_mask = 1;
1042         sb->s_flags |= MS_POSIXACL;
1043
1044         fc->release = fuse_free_conn;
1045         fc->flags = d.flags;
1046         fc->user_id = d.user_id;
1047         fc->group_id = d.group_id;
1048         fc->max_read = max_t(unsigned, 4096, d.max_read);
1049
1050         /* Used by get_root_inode() */
1051         sb->s_fs_info = fc;
1052
1053         err = -ENOMEM;
1054         root = fuse_get_root_inode(sb, d.rootmode);
1055         root_dentry = d_make_root(root);
1056         if (!root_dentry)
1057                 goto err_put_conn;
1058         /* only now - we want root dentry with NULL ->d_op */
1059         sb->s_d_op = &fuse_dentry_operations;
1060
1061         init_req = fuse_request_alloc(0);
1062         if (!init_req)
1063                 goto err_put_root;
1064         init_req->background = 1;
1065
1066         if (is_bdev) {
1067                 fc->destroy_req = fuse_request_alloc(0);
1068                 if (!fc->destroy_req)
1069                         goto err_free_init_req;
1070         }
1071
1072         mutex_lock(&fuse_mutex);
1073         err = -EINVAL;
1074         if (file->private_data)
1075                 goto err_unlock;
1076
1077         err = fuse_ctl_add_conn(fc);
1078         if (err)
1079                 goto err_unlock;
1080
1081         list_add_tail(&fc->entry, &fuse_conn_list);
1082         sb->s_root = root_dentry;
1083         fc->connected = 1;
1084         file->private_data = fuse_conn_get(fc);
1085         mutex_unlock(&fuse_mutex);
1086         /*
1087          * atomic_dec_and_test() in fput() provides the necessary
1088          * memory barrier for file->private_data to be visible on all
1089          * CPUs after this
1090          */
1091         fput(file);
1092
1093         fuse_send_init(fc, init_req);
1094
1095         return 0;
1096
1097  err_unlock:
1098         mutex_unlock(&fuse_mutex);
1099  err_free_init_req:
1100         fuse_request_free(init_req);
1101  err_put_root:
1102         dput(root_dentry);
1103  err_put_conn:
1104         fuse_bdi_destroy(fc);
1105         fuse_conn_put(fc);
1106  err_fput:
1107         fput(file);
1108  err:
1109         return err;
1110 }
1111
1112 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1113                        int flags, const char *dev_name,
1114                        void *raw_data)
1115 {
1116         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1117 }
1118
1119 static void fuse_kill_sb_anon(struct super_block *sb)
1120 {
1121         struct fuse_conn *fc = get_fuse_conn_super(sb);
1122
1123         if (fc) {
1124                 down_write(&fc->killsb);
1125                 fc->sb = NULL;
1126                 up_write(&fc->killsb);
1127         }
1128
1129         kill_anon_super(sb);
1130 }
1131
1132 static struct file_system_type fuse_fs_type = {
1133         .owner          = THIS_MODULE,
1134         .name           = "fuse",
1135         .fs_flags       = FS_HAS_SUBTYPE,
1136         .mount          = fuse_mount,
1137         .kill_sb        = fuse_kill_sb_anon,
1138 };
1139 MODULE_ALIAS_FS("fuse");
1140
1141 #ifdef CONFIG_BLOCK
1142 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1143                            int flags, const char *dev_name,
1144                            void *raw_data)
1145 {
1146         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1147 }
1148
1149 static void fuse_kill_sb_blk(struct super_block *sb)
1150 {
1151         struct fuse_conn *fc = get_fuse_conn_super(sb);
1152
1153         if (fc) {
1154                 down_write(&fc->killsb);
1155                 fc->sb = NULL;
1156                 up_write(&fc->killsb);
1157         }
1158
1159         kill_block_super(sb);
1160 }
1161
1162 static struct file_system_type fuseblk_fs_type = {
1163         .owner          = THIS_MODULE,
1164         .name           = "fuseblk",
1165         .mount          = fuse_mount_blk,
1166         .kill_sb        = fuse_kill_sb_blk,
1167         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1168 };
1169 MODULE_ALIAS_FS("fuseblk");
1170
1171 static inline int register_fuseblk(void)
1172 {
1173         return register_filesystem(&fuseblk_fs_type);
1174 }
1175
1176 static inline void unregister_fuseblk(void)
1177 {
1178         unregister_filesystem(&fuseblk_fs_type);
1179 }
1180 #else
1181 static inline int register_fuseblk(void)
1182 {
1183         return 0;
1184 }
1185
1186 static inline void unregister_fuseblk(void)
1187 {
1188 }
1189 #endif
1190
1191 static void fuse_inode_init_once(void *foo)
1192 {
1193         struct inode *inode = foo;
1194
1195         inode_init_once(inode);
1196 }
1197
1198 static int __init fuse_fs_init(void)
1199 {
1200         int err;
1201
1202         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1203                                               sizeof(struct fuse_inode),
1204                                               0, SLAB_HWCACHE_ALIGN,
1205                                               fuse_inode_init_once);
1206         err = -ENOMEM;
1207         if (!fuse_inode_cachep)
1208                 goto out;
1209
1210         err = register_fuseblk();
1211         if (err)
1212                 goto out2;
1213
1214         err = register_filesystem(&fuse_fs_type);
1215         if (err)
1216                 goto out3;
1217
1218         return 0;
1219
1220  out3:
1221         unregister_fuseblk();
1222  out2:
1223         kmem_cache_destroy(fuse_inode_cachep);
1224  out:
1225         return err;
1226 }
1227
1228 static void fuse_fs_cleanup(void)
1229 {
1230         unregister_filesystem(&fuse_fs_type);
1231         unregister_fuseblk();
1232
1233         /*
1234          * Make sure all delayed rcu free inodes are flushed before we
1235          * destroy cache.
1236          */
1237         rcu_barrier();
1238         kmem_cache_destroy(fuse_inode_cachep);
1239 }
1240
1241 static struct kobject *fuse_kobj;
1242 static struct kobject *connections_kobj;
1243
1244 static int fuse_sysfs_init(void)
1245 {
1246         int err;
1247
1248         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1249         if (!fuse_kobj) {
1250                 err = -ENOMEM;
1251                 goto out_err;
1252         }
1253
1254         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1255         if (!connections_kobj) {
1256                 err = -ENOMEM;
1257                 goto out_fuse_unregister;
1258         }
1259
1260         return 0;
1261
1262  out_fuse_unregister:
1263         kobject_put(fuse_kobj);
1264  out_err:
1265         return err;
1266 }
1267
1268 static void fuse_sysfs_cleanup(void)
1269 {
1270         kobject_put(connections_kobj);
1271         kobject_put(fuse_kobj);
1272 }
1273
1274 static int __init fuse_init(void)
1275 {
1276         int res;
1277
1278         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1279                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1280
1281         INIT_LIST_HEAD(&fuse_conn_list);
1282         res = fuse_fs_init();
1283         if (res)
1284                 goto err;
1285
1286         res = fuse_dev_init();
1287         if (res)
1288                 goto err_fs_cleanup;
1289
1290         res = fuse_sysfs_init();
1291         if (res)
1292                 goto err_dev_cleanup;
1293
1294         res = fuse_ctl_init();
1295         if (res)
1296                 goto err_sysfs_cleanup;
1297
1298         sanitize_global_limit(&max_user_bgreq);
1299         sanitize_global_limit(&max_user_congthresh);
1300
1301         return 0;
1302
1303  err_sysfs_cleanup:
1304         fuse_sysfs_cleanup();
1305  err_dev_cleanup:
1306         fuse_dev_cleanup();
1307  err_fs_cleanup:
1308         fuse_fs_cleanup();
1309  err:
1310         return res;
1311 }
1312
1313 static void __exit fuse_exit(void)
1314 {
1315         printk(KERN_DEBUG "fuse exit\n");
1316
1317         fuse_ctl_cleanup();
1318         fuse_sysfs_cleanup();
1319         fuse_fs_cleanup();
1320         fuse_dev_cleanup();
1321 }
1322
1323 module_init(fuse_init);
1324 module_exit(fuse_exit);