Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
[linux-2.6] / 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/parser.h>
18 #include <linux/statfs.h>
19 #include <linux/random.h>
20 #include <linux/sched.h>
21 #include <linux/exportfs.h>
22
23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
24 MODULE_DESCRIPTION("Filesystem in Userspace");
25 MODULE_LICENSE("GPL");
26
27 static struct kmem_cache *fuse_inode_cachep;
28 struct list_head fuse_conn_list;
29 DEFINE_MUTEX(fuse_mutex);
30
31 #define FUSE_SUPER_MAGIC 0x65735546
32
33 #define FUSE_DEFAULT_BLKSIZE 512
34
35 struct fuse_mount_data {
36         int fd;
37         unsigned rootmode;
38         unsigned user_id;
39         unsigned group_id;
40         unsigned fd_present:1;
41         unsigned rootmode_present:1;
42         unsigned user_id_present:1;
43         unsigned group_id_present:1;
44         unsigned flags;
45         unsigned max_read;
46         unsigned blksize;
47 };
48
49 static struct inode *fuse_alloc_inode(struct super_block *sb)
50 {
51         struct inode *inode;
52         struct fuse_inode *fi;
53
54         inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
55         if (!inode)
56                 return NULL;
57
58         fi = get_fuse_inode(inode);
59         fi->i_time = 0;
60         fi->nodeid = 0;
61         fi->nlookup = 0;
62         fi->attr_version = 0;
63         fi->writectr = 0;
64         INIT_LIST_HEAD(&fi->write_files);
65         INIT_LIST_HEAD(&fi->queued_writes);
66         INIT_LIST_HEAD(&fi->writepages);
67         init_waitqueue_head(&fi->page_waitq);
68         fi->forget_req = fuse_request_alloc();
69         if (!fi->forget_req) {
70                 kmem_cache_free(fuse_inode_cachep, inode);
71                 return NULL;
72         }
73
74         return inode;
75 }
76
77 static void fuse_destroy_inode(struct inode *inode)
78 {
79         struct fuse_inode *fi = get_fuse_inode(inode);
80         BUG_ON(!list_empty(&fi->write_files));
81         BUG_ON(!list_empty(&fi->queued_writes));
82         if (fi->forget_req)
83                 fuse_request_free(fi->forget_req);
84         kmem_cache_free(fuse_inode_cachep, inode);
85 }
86
87 void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
88                       u64 nodeid, u64 nlookup)
89 {
90         struct fuse_forget_in *inarg = &req->misc.forget_in;
91         inarg->nlookup = nlookup;
92         req->in.h.opcode = FUSE_FORGET;
93         req->in.h.nodeid = nodeid;
94         req->in.numargs = 1;
95         req->in.args[0].size = sizeof(struct fuse_forget_in);
96         req->in.args[0].value = inarg;
97         fuse_request_send_noreply(fc, req);
98 }
99
100 static void fuse_clear_inode(struct inode *inode)
101 {
102         if (inode->i_sb->s_flags & MS_ACTIVE) {
103                 struct fuse_conn *fc = get_fuse_conn(inode);
104                 struct fuse_inode *fi = get_fuse_inode(inode);
105                 fuse_send_forget(fc, fi->forget_req, fi->nodeid, fi->nlookup);
106                 fi->forget_req = NULL;
107         }
108 }
109
110 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
111 {
112         if (*flags & MS_MANDLOCK)
113                 return -EINVAL;
114
115         return 0;
116 }
117
118 void fuse_truncate(struct address_space *mapping, loff_t offset)
119 {
120         /* See vmtruncate() */
121         unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
122         truncate_inode_pages(mapping, offset);
123         unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
124 }
125
126 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
127                                    u64 attr_valid)
128 {
129         struct fuse_conn *fc = get_fuse_conn(inode);
130         struct fuse_inode *fi = get_fuse_inode(inode);
131
132         fi->attr_version = ++fc->attr_version;
133         fi->i_time = attr_valid;
134
135         inode->i_ino     = attr->ino;
136         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
137         inode->i_nlink   = attr->nlink;
138         inode->i_uid     = attr->uid;
139         inode->i_gid     = attr->gid;
140         inode->i_blocks  = attr->blocks;
141         inode->i_atime.tv_sec   = attr->atime;
142         inode->i_atime.tv_nsec  = attr->atimensec;
143         inode->i_mtime.tv_sec   = attr->mtime;
144         inode->i_mtime.tv_nsec  = attr->mtimensec;
145         inode->i_ctime.tv_sec   = attr->ctime;
146         inode->i_ctime.tv_nsec  = attr->ctimensec;
147
148         if (attr->blksize != 0)
149                 inode->i_blkbits = ilog2(attr->blksize);
150         else
151                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
152
153         /*
154          * Don't set the sticky bit in i_mode, unless we want the VFS
155          * to check permissions.  This prevents failures due to the
156          * check in may_delete().
157          */
158         fi->orig_i_mode = inode->i_mode;
159         if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
160                 inode->i_mode &= ~S_ISVTX;
161 }
162
163 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
164                             u64 attr_valid, u64 attr_version)
165 {
166         struct fuse_conn *fc = get_fuse_conn(inode);
167         struct fuse_inode *fi = get_fuse_inode(inode);
168         loff_t oldsize;
169
170         spin_lock(&fc->lock);
171         if (attr_version != 0 && fi->attr_version > attr_version) {
172                 spin_unlock(&fc->lock);
173                 return;
174         }
175
176         fuse_change_attributes_common(inode, attr, attr_valid);
177
178         oldsize = inode->i_size;
179         i_size_write(inode, attr->size);
180         spin_unlock(&fc->lock);
181
182         if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
183                 if (attr->size < oldsize)
184                         fuse_truncate(inode->i_mapping, attr->size);
185                 invalidate_inode_pages2(inode->i_mapping);
186         }
187 }
188
189 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
190 {
191         inode->i_mode = attr->mode & S_IFMT;
192         inode->i_size = attr->size;
193         if (S_ISREG(inode->i_mode)) {
194                 fuse_init_common(inode);
195                 fuse_init_file_inode(inode);
196         } else if (S_ISDIR(inode->i_mode))
197                 fuse_init_dir(inode);
198         else if (S_ISLNK(inode->i_mode))
199                 fuse_init_symlink(inode);
200         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
201                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
202                 fuse_init_common(inode);
203                 init_special_inode(inode, inode->i_mode,
204                                    new_decode_dev(attr->rdev));
205         } else
206                 BUG();
207 }
208
209 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
210 {
211         u64 nodeid = *(u64 *) _nodeidp;
212         if (get_node_id(inode) == nodeid)
213                 return 1;
214         else
215                 return 0;
216 }
217
218 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
219 {
220         u64 nodeid = *(u64 *) _nodeidp;
221         get_fuse_inode(inode)->nodeid = nodeid;
222         return 0;
223 }
224
225 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
226                         int generation, struct fuse_attr *attr,
227                         u64 attr_valid, u64 attr_version)
228 {
229         struct inode *inode;
230         struct fuse_inode *fi;
231         struct fuse_conn *fc = get_fuse_conn_super(sb);
232
233  retry:
234         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
235         if (!inode)
236                 return NULL;
237
238         if ((inode->i_state & I_NEW)) {
239                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
240                 inode->i_generation = generation;
241                 inode->i_data.backing_dev_info = &fc->bdi;
242                 fuse_init_inode(inode, attr);
243                 unlock_new_inode(inode);
244         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
245                 /* Inode has changed type, any I/O on the old should fail */
246                 make_bad_inode(inode);
247                 iput(inode);
248                 goto retry;
249         }
250
251         fi = get_fuse_inode(inode);
252         spin_lock(&fc->lock);
253         fi->nlookup++;
254         spin_unlock(&fc->lock);
255         fuse_change_attributes(inode, attr, attr_valid, attr_version);
256
257         return inode;
258 }
259
260 static void fuse_umount_begin(struct super_block *sb)
261 {
262         fuse_abort_conn(get_fuse_conn_super(sb));
263 }
264
265 static void fuse_send_destroy(struct fuse_conn *fc)
266 {
267         struct fuse_req *req = fc->destroy_req;
268         if (req && fc->conn_init) {
269                 fc->destroy_req = NULL;
270                 req->in.h.opcode = FUSE_DESTROY;
271                 req->force = 1;
272                 fuse_request_send(fc, req);
273                 fuse_put_request(fc, req);
274         }
275 }
276
277 static void fuse_put_super(struct super_block *sb)
278 {
279         struct fuse_conn *fc = get_fuse_conn_super(sb);
280
281         fuse_send_destroy(fc);
282         spin_lock(&fc->lock);
283         fc->connected = 0;
284         fc->blocked = 0;
285         spin_unlock(&fc->lock);
286         /* Flush all readers on this fs */
287         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
288         wake_up_all(&fc->waitq);
289         wake_up_all(&fc->blocked_waitq);
290         wake_up_all(&fc->reserved_req_waitq);
291         mutex_lock(&fuse_mutex);
292         list_del(&fc->entry);
293         fuse_ctl_remove_conn(fc);
294         mutex_unlock(&fuse_mutex);
295         fuse_conn_put(fc);
296 }
297
298 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
299 {
300         stbuf->f_type    = FUSE_SUPER_MAGIC;
301         stbuf->f_bsize   = attr->bsize;
302         stbuf->f_frsize  = attr->frsize;
303         stbuf->f_blocks  = attr->blocks;
304         stbuf->f_bfree   = attr->bfree;
305         stbuf->f_bavail  = attr->bavail;
306         stbuf->f_files   = attr->files;
307         stbuf->f_ffree   = attr->ffree;
308         stbuf->f_namelen = attr->namelen;
309         /* fsid is left zero */
310 }
311
312 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
313 {
314         struct super_block *sb = dentry->d_sb;
315         struct fuse_conn *fc = get_fuse_conn_super(sb);
316         struct fuse_req *req;
317         struct fuse_statfs_out outarg;
318         int err;
319
320         if (!fuse_allow_task(fc, current)) {
321                 buf->f_type = FUSE_SUPER_MAGIC;
322                 return 0;
323         }
324
325         req = fuse_get_req(fc);
326         if (IS_ERR(req))
327                 return PTR_ERR(req);
328
329         memset(&outarg, 0, sizeof(outarg));
330         req->in.numargs = 0;
331         req->in.h.opcode = FUSE_STATFS;
332         req->in.h.nodeid = get_node_id(dentry->d_inode);
333         req->out.numargs = 1;
334         req->out.args[0].size =
335                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
336         req->out.args[0].value = &outarg;
337         fuse_request_send(fc, req);
338         err = req->out.h.error;
339         if (!err)
340                 convert_fuse_statfs(buf, &outarg.st);
341         fuse_put_request(fc, req);
342         return err;
343 }
344
345 enum {
346         OPT_FD,
347         OPT_ROOTMODE,
348         OPT_USER_ID,
349         OPT_GROUP_ID,
350         OPT_DEFAULT_PERMISSIONS,
351         OPT_ALLOW_OTHER,
352         OPT_MAX_READ,
353         OPT_BLKSIZE,
354         OPT_ERR
355 };
356
357 static const match_table_t tokens = {
358         {OPT_FD,                        "fd=%u"},
359         {OPT_ROOTMODE,                  "rootmode=%o"},
360         {OPT_USER_ID,                   "user_id=%u"},
361         {OPT_GROUP_ID,                  "group_id=%u"},
362         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
363         {OPT_ALLOW_OTHER,               "allow_other"},
364         {OPT_MAX_READ,                  "max_read=%u"},
365         {OPT_BLKSIZE,                   "blksize=%u"},
366         {OPT_ERR,                       NULL}
367 };
368
369 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
370 {
371         char *p;
372         memset(d, 0, sizeof(struct fuse_mount_data));
373         d->max_read = ~0;
374         d->blksize = FUSE_DEFAULT_BLKSIZE;
375
376         while ((p = strsep(&opt, ",")) != NULL) {
377                 int token;
378                 int value;
379                 substring_t args[MAX_OPT_ARGS];
380                 if (!*p)
381                         continue;
382
383                 token = match_token(p, tokens, args);
384                 switch (token) {
385                 case OPT_FD:
386                         if (match_int(&args[0], &value))
387                                 return 0;
388                         d->fd = value;
389                         d->fd_present = 1;
390                         break;
391
392                 case OPT_ROOTMODE:
393                         if (match_octal(&args[0], &value))
394                                 return 0;
395                         if (!fuse_valid_type(value))
396                                 return 0;
397                         d->rootmode = value;
398                         d->rootmode_present = 1;
399                         break;
400
401                 case OPT_USER_ID:
402                         if (match_int(&args[0], &value))
403                                 return 0;
404                         d->user_id = value;
405                         d->user_id_present = 1;
406                         break;
407
408                 case OPT_GROUP_ID:
409                         if (match_int(&args[0], &value))
410                                 return 0;
411                         d->group_id = value;
412                         d->group_id_present = 1;
413                         break;
414
415                 case OPT_DEFAULT_PERMISSIONS:
416                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
417                         break;
418
419                 case OPT_ALLOW_OTHER:
420                         d->flags |= FUSE_ALLOW_OTHER;
421                         break;
422
423                 case OPT_MAX_READ:
424                         if (match_int(&args[0], &value))
425                                 return 0;
426                         d->max_read = value;
427                         break;
428
429                 case OPT_BLKSIZE:
430                         if (!is_bdev || match_int(&args[0], &value))
431                                 return 0;
432                         d->blksize = value;
433                         break;
434
435                 default:
436                         return 0;
437                 }
438         }
439
440         if (!d->fd_present || !d->rootmode_present ||
441             !d->user_id_present || !d->group_id_present)
442                 return 0;
443
444         return 1;
445 }
446
447 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
448 {
449         struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
450
451         seq_printf(m, ",user_id=%u", fc->user_id);
452         seq_printf(m, ",group_id=%u", fc->group_id);
453         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
454                 seq_puts(m, ",default_permissions");
455         if (fc->flags & FUSE_ALLOW_OTHER)
456                 seq_puts(m, ",allow_other");
457         if (fc->max_read != ~0)
458                 seq_printf(m, ",max_read=%u", fc->max_read);
459         if (mnt->mnt_sb->s_bdev &&
460             mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
461                 seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
462         return 0;
463 }
464
465 int fuse_conn_init(struct fuse_conn *fc, struct super_block *sb)
466 {
467         int err;
468
469         memset(fc, 0, sizeof(*fc));
470         spin_lock_init(&fc->lock);
471         mutex_init(&fc->inst_mutex);
472         atomic_set(&fc->count, 1);
473         init_waitqueue_head(&fc->waitq);
474         init_waitqueue_head(&fc->blocked_waitq);
475         init_waitqueue_head(&fc->reserved_req_waitq);
476         INIT_LIST_HEAD(&fc->pending);
477         INIT_LIST_HEAD(&fc->processing);
478         INIT_LIST_HEAD(&fc->io);
479         INIT_LIST_HEAD(&fc->interrupts);
480         INIT_LIST_HEAD(&fc->bg_queue);
481         INIT_LIST_HEAD(&fc->entry);
482         atomic_set(&fc->num_waiting, 0);
483         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
484         fc->bdi.unplug_io_fn = default_unplug_io_fn;
485         /* fuse does it's own writeback accounting */
486         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
487         fc->khctr = 0;
488         fc->polled_files = RB_ROOT;
489         fc->dev = sb->s_dev;
490         err = bdi_init(&fc->bdi);
491         if (err)
492                 goto error_mutex_destroy;
493         if (sb->s_bdev) {
494                 err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
495                                    MAJOR(fc->dev), MINOR(fc->dev));
496         } else {
497                 err = bdi_register_dev(&fc->bdi, fc->dev);
498         }
499         if (err)
500                 goto error_bdi_destroy;
501         /*
502          * For a single fuse filesystem use max 1% of dirty +
503          * writeback threshold.
504          *
505          * This gives about 1M of write buffer for memory maps on a
506          * machine with 1G and 10% dirty_ratio, which should be more
507          * than enough.
508          *
509          * Privileged users can raise it by writing to
510          *
511          *    /sys/class/bdi/<bdi>/max_ratio
512          */
513         bdi_set_max_ratio(&fc->bdi, 1);
514         fc->reqctr = 0;
515         fc->blocked = 1;
516         fc->attr_version = 1;
517         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
518
519         return 0;
520
521  error_bdi_destroy:
522         bdi_destroy(&fc->bdi);
523  error_mutex_destroy:
524         mutex_destroy(&fc->inst_mutex);
525         return err;
526 }
527 EXPORT_SYMBOL_GPL(fuse_conn_init);
528
529 void fuse_conn_put(struct fuse_conn *fc)
530 {
531         if (atomic_dec_and_test(&fc->count)) {
532                 if (fc->destroy_req)
533                         fuse_request_free(fc->destroy_req);
534                 mutex_destroy(&fc->inst_mutex);
535                 bdi_destroy(&fc->bdi);
536                 fc->release(fc);
537         }
538 }
539
540 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
541 {
542         atomic_inc(&fc->count);
543         return fc;
544 }
545
546 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
547 {
548         struct fuse_attr attr;
549         memset(&attr, 0, sizeof(attr));
550
551         attr.mode = mode;
552         attr.ino = FUSE_ROOT_ID;
553         attr.nlink = 1;
554         return fuse_iget(sb, 1, 0, &attr, 0, 0);
555 }
556
557 struct fuse_inode_handle {
558         u64 nodeid;
559         u32 generation;
560 };
561
562 static struct dentry *fuse_get_dentry(struct super_block *sb,
563                                       struct fuse_inode_handle *handle)
564 {
565         struct fuse_conn *fc = get_fuse_conn_super(sb);
566         struct inode *inode;
567         struct dentry *entry;
568         int err = -ESTALE;
569
570         if (handle->nodeid == 0)
571                 goto out_err;
572
573         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
574         if (!inode) {
575                 struct fuse_entry_out outarg;
576                 struct qstr name;
577
578                 if (!fc->export_support)
579                         goto out_err;
580
581                 name.len = 1;
582                 name.name = ".";
583                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
584                                        &inode);
585                 if (err && err != -ENOENT)
586                         goto out_err;
587                 if (err || !inode) {
588                         err = -ESTALE;
589                         goto out_err;
590                 }
591                 err = -EIO;
592                 if (get_node_id(inode) != handle->nodeid)
593                         goto out_iput;
594         }
595         err = -ESTALE;
596         if (inode->i_generation != handle->generation)
597                 goto out_iput;
598
599         entry = d_obtain_alias(inode);
600         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) {
601                 entry->d_op = &fuse_dentry_operations;
602                 fuse_invalidate_entry_cache(entry);
603         }
604
605         return entry;
606
607  out_iput:
608         iput(inode);
609  out_err:
610         return ERR_PTR(err);
611 }
612
613 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
614                            int connectable)
615 {
616         struct inode *inode = dentry->d_inode;
617         bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
618         int len = encode_parent ? 6 : 3;
619         u64 nodeid;
620         u32 generation;
621
622         if (*max_len < len)
623                 return  255;
624
625         nodeid = get_fuse_inode(inode)->nodeid;
626         generation = inode->i_generation;
627
628         fh[0] = (u32)(nodeid >> 32);
629         fh[1] = (u32)(nodeid & 0xffffffff);
630         fh[2] = generation;
631
632         if (encode_parent) {
633                 struct inode *parent;
634
635                 spin_lock(&dentry->d_lock);
636                 parent = dentry->d_parent->d_inode;
637                 nodeid = get_fuse_inode(parent)->nodeid;
638                 generation = parent->i_generation;
639                 spin_unlock(&dentry->d_lock);
640
641                 fh[3] = (u32)(nodeid >> 32);
642                 fh[4] = (u32)(nodeid & 0xffffffff);
643                 fh[5] = generation;
644         }
645
646         *max_len = len;
647         return encode_parent ? 0x82 : 0x81;
648 }
649
650 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
651                 struct fid *fid, int fh_len, int fh_type)
652 {
653         struct fuse_inode_handle handle;
654
655         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
656                 return NULL;
657
658         handle.nodeid = (u64) fid->raw[0] << 32;
659         handle.nodeid |= (u64) fid->raw[1];
660         handle.generation = fid->raw[2];
661         return fuse_get_dentry(sb, &handle);
662 }
663
664 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
665                 struct fid *fid, int fh_len, int fh_type)
666 {
667         struct fuse_inode_handle parent;
668
669         if (fh_type != 0x82 || fh_len < 6)
670                 return NULL;
671
672         parent.nodeid = (u64) fid->raw[3] << 32;
673         parent.nodeid |= (u64) fid->raw[4];
674         parent.generation = fid->raw[5];
675         return fuse_get_dentry(sb, &parent);
676 }
677
678 static struct dentry *fuse_get_parent(struct dentry *child)
679 {
680         struct inode *child_inode = child->d_inode;
681         struct fuse_conn *fc = get_fuse_conn(child_inode);
682         struct inode *inode;
683         struct dentry *parent;
684         struct fuse_entry_out outarg;
685         struct qstr name;
686         int err;
687
688         if (!fc->export_support)
689                 return ERR_PTR(-ESTALE);
690
691         name.len = 2;
692         name.name = "..";
693         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
694                                &name, &outarg, &inode);
695         if (err) {
696                 if (err == -ENOENT)
697                         return ERR_PTR(-ESTALE);
698                 return ERR_PTR(err);
699         }
700
701         parent = d_obtain_alias(inode);
702         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) {
703                 parent->d_op = &fuse_dentry_operations;
704                 fuse_invalidate_entry_cache(parent);
705         }
706
707         return parent;
708 }
709
710 static const struct export_operations fuse_export_operations = {
711         .fh_to_dentry   = fuse_fh_to_dentry,
712         .fh_to_parent   = fuse_fh_to_parent,
713         .encode_fh      = fuse_encode_fh,
714         .get_parent     = fuse_get_parent,
715 };
716
717 static const struct super_operations fuse_super_operations = {
718         .alloc_inode    = fuse_alloc_inode,
719         .destroy_inode  = fuse_destroy_inode,
720         .clear_inode    = fuse_clear_inode,
721         .drop_inode     = generic_delete_inode,
722         .remount_fs     = fuse_remount_fs,
723         .put_super      = fuse_put_super,
724         .umount_begin   = fuse_umount_begin,
725         .statfs         = fuse_statfs,
726         .show_options   = fuse_show_options,
727 };
728
729 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
730 {
731         struct fuse_init_out *arg = &req->misc.init_out;
732
733         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
734                 fc->conn_error = 1;
735         else {
736                 unsigned long ra_pages;
737
738                 if (arg->minor >= 6) {
739                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
740                         if (arg->flags & FUSE_ASYNC_READ)
741                                 fc->async_read = 1;
742                         if (!(arg->flags & FUSE_POSIX_LOCKS))
743                                 fc->no_lock = 1;
744                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
745                                 fc->atomic_o_trunc = 1;
746                         if (arg->minor >= 9) {
747                                 /* LOOKUP has dependency on proto version */
748                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
749                                         fc->export_support = 1;
750                         }
751                         if (arg->flags & FUSE_BIG_WRITES)
752                                 fc->big_writes = 1;
753                 } else {
754                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
755                         fc->no_lock = 1;
756                 }
757
758                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
759                 fc->minor = arg->minor;
760                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
761                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
762                 fc->conn_init = 1;
763         }
764         fc->blocked = 0;
765         wake_up_all(&fc->blocked_waitq);
766 }
767
768 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
769 {
770         struct fuse_init_in *arg = &req->misc.init_in;
771
772         arg->major = FUSE_KERNEL_VERSION;
773         arg->minor = FUSE_KERNEL_MINOR_VERSION;
774         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
775         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
776                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES;
777         req->in.h.opcode = FUSE_INIT;
778         req->in.numargs = 1;
779         req->in.args[0].size = sizeof(*arg);
780         req->in.args[0].value = arg;
781         req->out.numargs = 1;
782         /* Variable length arguement used for backward compatibility
783            with interface version < 7.5.  Rest of init_out is zeroed
784            by do_get_request(), so a short reply is not a problem */
785         req->out.argvar = 1;
786         req->out.args[0].size = sizeof(struct fuse_init_out);
787         req->out.args[0].value = &req->misc.init_out;
788         req->end = process_init_reply;
789         fuse_request_send_background(fc, req);
790 }
791
792 static void fuse_free_conn(struct fuse_conn *fc)
793 {
794         kfree(fc);
795 }
796
797 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
798 {
799         struct fuse_conn *fc;
800         struct inode *root;
801         struct fuse_mount_data d;
802         struct file *file;
803         struct dentry *root_dentry;
804         struct fuse_req *init_req;
805         int err;
806         int is_bdev = sb->s_bdev != NULL;
807
808         if (sb->s_flags & MS_MANDLOCK)
809                 return -EINVAL;
810
811         if (!parse_fuse_opt((char *) data, &d, is_bdev))
812                 return -EINVAL;
813
814         if (is_bdev) {
815 #ifdef CONFIG_BLOCK
816                 if (!sb_set_blocksize(sb, d.blksize))
817                         return -EINVAL;
818 #endif
819         } else {
820                 sb->s_blocksize = PAGE_CACHE_SIZE;
821                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
822         }
823         sb->s_magic = FUSE_SUPER_MAGIC;
824         sb->s_op = &fuse_super_operations;
825         sb->s_maxbytes = MAX_LFS_FILESIZE;
826         sb->s_export_op = &fuse_export_operations;
827
828         file = fget(d.fd);
829         if (!file)
830                 return -EINVAL;
831
832         if (file->f_op != &fuse_dev_operations)
833                 return -EINVAL;
834
835         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
836         if (!fc)
837                 return -ENOMEM;
838
839         err = fuse_conn_init(fc, sb);
840         if (err) {
841                 kfree(fc);
842                 return err;
843         }
844
845         fc->release = fuse_free_conn;
846         fc->flags = d.flags;
847         fc->user_id = d.user_id;
848         fc->group_id = d.group_id;
849         fc->max_read = max_t(unsigned, 4096, d.max_read);
850
851         /* Used by get_root_inode() */
852         sb->s_fs_info = fc;
853
854         err = -ENOMEM;
855         root = fuse_get_root_inode(sb, d.rootmode);
856         if (!root)
857                 goto err;
858
859         root_dentry = d_alloc_root(root);
860         if (!root_dentry) {
861                 iput(root);
862                 goto err;
863         }
864
865         init_req = fuse_request_alloc();
866         if (!init_req)
867                 goto err_put_root;
868
869         if (is_bdev) {
870                 fc->destroy_req = fuse_request_alloc();
871                 if (!fc->destroy_req)
872                         goto err_free_init_req;
873         }
874
875         mutex_lock(&fuse_mutex);
876         err = -EINVAL;
877         if (file->private_data)
878                 goto err_unlock;
879
880         err = fuse_ctl_add_conn(fc);
881         if (err)
882                 goto err_unlock;
883
884         list_add_tail(&fc->entry, &fuse_conn_list);
885         sb->s_root = root_dentry;
886         fc->connected = 1;
887         file->private_data = fuse_conn_get(fc);
888         mutex_unlock(&fuse_mutex);
889         /*
890          * atomic_dec_and_test() in fput() provides the necessary
891          * memory barrier for file->private_data to be visible on all
892          * CPUs after this
893          */
894         fput(file);
895
896         fuse_send_init(fc, init_req);
897
898         return 0;
899
900  err_unlock:
901         mutex_unlock(&fuse_mutex);
902  err_free_init_req:
903         fuse_request_free(init_req);
904  err_put_root:
905         dput(root_dentry);
906  err:
907         fput(file);
908         fuse_conn_put(fc);
909         return err;
910 }
911
912 static int fuse_get_sb(struct file_system_type *fs_type,
913                        int flags, const char *dev_name,
914                        void *raw_data, struct vfsmount *mnt)
915 {
916         return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
917 }
918
919 static struct file_system_type fuse_fs_type = {
920         .owner          = THIS_MODULE,
921         .name           = "fuse",
922         .fs_flags       = FS_HAS_SUBTYPE,
923         .get_sb         = fuse_get_sb,
924         .kill_sb        = kill_anon_super,
925 };
926
927 #ifdef CONFIG_BLOCK
928 static int fuse_get_sb_blk(struct file_system_type *fs_type,
929                            int flags, const char *dev_name,
930                            void *raw_data, struct vfsmount *mnt)
931 {
932         return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
933                            mnt);
934 }
935
936 static struct file_system_type fuseblk_fs_type = {
937         .owner          = THIS_MODULE,
938         .name           = "fuseblk",
939         .get_sb         = fuse_get_sb_blk,
940         .kill_sb        = kill_block_super,
941         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
942 };
943
944 static inline int register_fuseblk(void)
945 {
946         return register_filesystem(&fuseblk_fs_type);
947 }
948
949 static inline void unregister_fuseblk(void)
950 {
951         unregister_filesystem(&fuseblk_fs_type);
952 }
953 #else
954 static inline int register_fuseblk(void)
955 {
956         return 0;
957 }
958
959 static inline void unregister_fuseblk(void)
960 {
961 }
962 #endif
963
964 static void fuse_inode_init_once(void *foo)
965 {
966         struct inode *inode = foo;
967
968         inode_init_once(inode);
969 }
970
971 static int __init fuse_fs_init(void)
972 {
973         int err;
974
975         err = register_filesystem(&fuse_fs_type);
976         if (err)
977                 goto out;
978
979         err = register_fuseblk();
980         if (err)
981                 goto out_unreg;
982
983         fuse_inode_cachep = kmem_cache_create("fuse_inode",
984                                               sizeof(struct fuse_inode),
985                                               0, SLAB_HWCACHE_ALIGN,
986                                               fuse_inode_init_once);
987         err = -ENOMEM;
988         if (!fuse_inode_cachep)
989                 goto out_unreg2;
990
991         return 0;
992
993  out_unreg2:
994         unregister_fuseblk();
995  out_unreg:
996         unregister_filesystem(&fuse_fs_type);
997  out:
998         return err;
999 }
1000
1001 static void fuse_fs_cleanup(void)
1002 {
1003         unregister_filesystem(&fuse_fs_type);
1004         unregister_fuseblk();
1005         kmem_cache_destroy(fuse_inode_cachep);
1006 }
1007
1008 static struct kobject *fuse_kobj;
1009 static struct kobject *connections_kobj;
1010
1011 static int fuse_sysfs_init(void)
1012 {
1013         int err;
1014
1015         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1016         if (!fuse_kobj) {
1017                 err = -ENOMEM;
1018                 goto out_err;
1019         }
1020
1021         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1022         if (!connections_kobj) {
1023                 err = -ENOMEM;
1024                 goto out_fuse_unregister;
1025         }
1026
1027         return 0;
1028
1029  out_fuse_unregister:
1030         kobject_put(fuse_kobj);
1031  out_err:
1032         return err;
1033 }
1034
1035 static void fuse_sysfs_cleanup(void)
1036 {
1037         kobject_put(connections_kobj);
1038         kobject_put(fuse_kobj);
1039 }
1040
1041 static int __init fuse_init(void)
1042 {
1043         int res;
1044
1045         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1046                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1047
1048         INIT_LIST_HEAD(&fuse_conn_list);
1049         res = fuse_fs_init();
1050         if (res)
1051                 goto err;
1052
1053         res = fuse_dev_init();
1054         if (res)
1055                 goto err_fs_cleanup;
1056
1057         res = fuse_sysfs_init();
1058         if (res)
1059                 goto err_dev_cleanup;
1060
1061         res = fuse_ctl_init();
1062         if (res)
1063                 goto err_sysfs_cleanup;
1064
1065         return 0;
1066
1067  err_sysfs_cleanup:
1068         fuse_sysfs_cleanup();
1069  err_dev_cleanup:
1070         fuse_dev_cleanup();
1071  err_fs_cleanup:
1072         fuse_fs_cleanup();
1073  err:
1074         return res;
1075 }
1076
1077 static void __exit fuse_exit(void)
1078 {
1079         printk(KERN_DEBUG "fuse exit\n");
1080
1081         fuse_ctl_cleanup();
1082         fuse_sysfs_cleanup();
1083         fuse_fs_cleanup();
1084         fuse_dev_cleanup();
1085 }
1086
1087 module_init(fuse_init);
1088 module_exit(fuse_exit);