Merge branch 'upstream-davem' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
[linux-2.6] / fs / fuse / inode.c
1 /*
2   FUSE: Filesystem in Userspace
3   Copyright (C) 2001-2006  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         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                 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         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 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 static struct fuse_conn *new_conn(struct super_block *sb)
466 {
467         struct fuse_conn *fc;
468         int err;
469
470         fc = kzalloc(sizeof(*fc), GFP_KERNEL);
471         if (fc) {
472                 spin_lock_init(&fc->lock);
473                 mutex_init(&fc->inst_mutex);
474                 atomic_set(&fc->count, 1);
475                 init_waitqueue_head(&fc->waitq);
476                 init_waitqueue_head(&fc->blocked_waitq);
477                 init_waitqueue_head(&fc->reserved_req_waitq);
478                 INIT_LIST_HEAD(&fc->pending);
479                 INIT_LIST_HEAD(&fc->processing);
480                 INIT_LIST_HEAD(&fc->io);
481                 INIT_LIST_HEAD(&fc->interrupts);
482                 INIT_LIST_HEAD(&fc->bg_queue);
483                 atomic_set(&fc->num_waiting, 0);
484                 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
485                 fc->bdi.unplug_io_fn = default_unplug_io_fn;
486                 /* fuse does it's own writeback accounting */
487                 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
488                 fc->dev = sb->s_dev;
489                 err = bdi_init(&fc->bdi);
490                 if (err)
491                         goto error_kfree;
492                 if (sb->s_bdev) {
493                         err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
494                                            MAJOR(fc->dev), MINOR(fc->dev));
495                 } else {
496                         err = bdi_register_dev(&fc->bdi, fc->dev);
497                 }
498                 if (err)
499                         goto error_bdi_destroy;
500                 /*
501                  * For a single fuse filesystem use max 1% of dirty +
502                  * writeback threshold.
503                  *
504                  * This gives about 1M of write buffer for memory maps on a
505                  * machine with 1G and 10% dirty_ratio, which should be more
506                  * than enough.
507                  *
508                  * Privileged users can raise it by writing to
509                  *
510                  *    /sys/class/bdi/<bdi>/max_ratio
511                  */
512                 bdi_set_max_ratio(&fc->bdi, 1);
513                 fc->reqctr = 0;
514                 fc->blocked = 1;
515                 fc->attr_version = 1;
516                 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
517         }
518         return fc;
519
520 error_bdi_destroy:
521         bdi_destroy(&fc->bdi);
522 error_kfree:
523         mutex_destroy(&fc->inst_mutex);
524         kfree(fc);
525         return NULL;
526 }
527
528 void fuse_conn_put(struct fuse_conn *fc)
529 {
530         if (atomic_dec_and_test(&fc->count)) {
531                 if (fc->destroy_req)
532                         fuse_request_free(fc->destroy_req);
533                 mutex_destroy(&fc->inst_mutex);
534                 bdi_destroy(&fc->bdi);
535                 kfree(fc);
536         }
537 }
538
539 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
540 {
541         atomic_inc(&fc->count);
542         return fc;
543 }
544
545 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
546 {
547         struct fuse_attr attr;
548         memset(&attr, 0, sizeof(attr));
549
550         attr.mode = mode;
551         attr.ino = FUSE_ROOT_ID;
552         attr.nlink = 1;
553         return fuse_iget(sb, 1, 0, &attr, 0, 0);
554 }
555
556 struct fuse_inode_handle
557 {
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_alloc_anon(inode);
600         err = -ENOMEM;
601         if (!entry)
602                 goto out_iput;
603
604         if (get_node_id(inode) != FUSE_ROOT_ID) {
605                 entry->d_op = &fuse_dentry_operations;
606                 fuse_invalidate_entry_cache(entry);
607         }
608
609         return entry;
610
611  out_iput:
612         iput(inode);
613  out_err:
614         return ERR_PTR(err);
615 }
616
617 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
618                            int connectable)
619 {
620         struct inode *inode = dentry->d_inode;
621         bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
622         int len = encode_parent ? 6 : 3;
623         u64 nodeid;
624         u32 generation;
625
626         if (*max_len < len)
627                 return  255;
628
629         nodeid = get_fuse_inode(inode)->nodeid;
630         generation = inode->i_generation;
631
632         fh[0] = (u32)(nodeid >> 32);
633         fh[1] = (u32)(nodeid & 0xffffffff);
634         fh[2] = generation;
635
636         if (encode_parent) {
637                 struct inode *parent;
638
639                 spin_lock(&dentry->d_lock);
640                 parent = dentry->d_parent->d_inode;
641                 nodeid = get_fuse_inode(parent)->nodeid;
642                 generation = parent->i_generation;
643                 spin_unlock(&dentry->d_lock);
644
645                 fh[3] = (u32)(nodeid >> 32);
646                 fh[4] = (u32)(nodeid & 0xffffffff);
647                 fh[5] = generation;
648         }
649
650         *max_len = len;
651         return encode_parent ? 0x82 : 0x81;
652 }
653
654 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
655                 struct fid *fid, int fh_len, int fh_type)
656 {
657         struct fuse_inode_handle handle;
658
659         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
660                 return NULL;
661
662         handle.nodeid = (u64) fid->raw[0] << 32;
663         handle.nodeid |= (u64) fid->raw[1];
664         handle.generation = fid->raw[2];
665         return fuse_get_dentry(sb, &handle);
666 }
667
668 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
669                 struct fid *fid, int fh_len, int fh_type)
670 {
671         struct fuse_inode_handle parent;
672
673         if (fh_type != 0x82 || fh_len < 6)
674                 return NULL;
675
676         parent.nodeid = (u64) fid->raw[3] << 32;
677         parent.nodeid |= (u64) fid->raw[4];
678         parent.generation = fid->raw[5];
679         return fuse_get_dentry(sb, &parent);
680 }
681
682 static struct dentry *fuse_get_parent(struct dentry *child)
683 {
684         struct inode *child_inode = child->d_inode;
685         struct fuse_conn *fc = get_fuse_conn(child_inode);
686         struct inode *inode;
687         struct dentry *parent;
688         struct fuse_entry_out outarg;
689         struct qstr name;
690         int err;
691
692         if (!fc->export_support)
693                 return ERR_PTR(-ESTALE);
694
695         name.len = 2;
696         name.name = "..";
697         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
698                                &name, &outarg, &inode);
699         if (err && err != -ENOENT)
700                 return ERR_PTR(err);
701         if (err || !inode)
702                 return ERR_PTR(-ESTALE);
703
704         parent = d_alloc_anon(inode);
705         if (!parent) {
706                 iput(inode);
707                 return ERR_PTR(-ENOMEM);
708         }
709         if (get_node_id(inode) != FUSE_ROOT_ID) {
710                 parent->d_op = &fuse_dentry_operations;
711                 fuse_invalidate_entry_cache(parent);
712         }
713
714         return parent;
715 }
716
717 static const struct export_operations fuse_export_operations = {
718         .fh_to_dentry   = fuse_fh_to_dentry,
719         .fh_to_parent   = fuse_fh_to_parent,
720         .encode_fh      = fuse_encode_fh,
721         .get_parent     = fuse_get_parent,
722 };
723
724 static const struct super_operations fuse_super_operations = {
725         .alloc_inode    = fuse_alloc_inode,
726         .destroy_inode  = fuse_destroy_inode,
727         .clear_inode    = fuse_clear_inode,
728         .drop_inode     = generic_delete_inode,
729         .remount_fs     = fuse_remount_fs,
730         .put_super      = fuse_put_super,
731         .umount_begin   = fuse_umount_begin,
732         .statfs         = fuse_statfs,
733         .show_options   = fuse_show_options,
734 };
735
736 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
737 {
738         struct fuse_init_out *arg = &req->misc.init_out;
739
740         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
741                 fc->conn_error = 1;
742         else {
743                 unsigned long ra_pages;
744
745                 if (arg->minor >= 6) {
746                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
747                         if (arg->flags & FUSE_ASYNC_READ)
748                                 fc->async_read = 1;
749                         if (!(arg->flags & FUSE_POSIX_LOCKS))
750                                 fc->no_lock = 1;
751                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
752                                 fc->atomic_o_trunc = 1;
753                         if (arg->minor >= 9) {
754                                 /* LOOKUP has dependency on proto version */
755                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
756                                         fc->export_support = 1;
757                         }
758                         if (arg->flags & FUSE_BIG_WRITES)
759                                 fc->big_writes = 1;
760                 } else {
761                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
762                         fc->no_lock = 1;
763                 }
764
765                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
766                 fc->minor = arg->minor;
767                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
768                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
769                 fc->conn_init = 1;
770         }
771         fuse_put_request(fc, req);
772         fc->blocked = 0;
773         wake_up_all(&fc->blocked_waitq);
774 }
775
776 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
777 {
778         struct fuse_init_in *arg = &req->misc.init_in;
779
780         arg->major = FUSE_KERNEL_VERSION;
781         arg->minor = FUSE_KERNEL_MINOR_VERSION;
782         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
783         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
784                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES;
785         req->in.h.opcode = FUSE_INIT;
786         req->in.numargs = 1;
787         req->in.args[0].size = sizeof(*arg);
788         req->in.args[0].value = arg;
789         req->out.numargs = 1;
790         /* Variable length arguement used for backward compatibility
791            with interface version < 7.5.  Rest of init_out is zeroed
792            by do_get_request(), so a short reply is not a problem */
793         req->out.argvar = 1;
794         req->out.args[0].size = sizeof(struct fuse_init_out);
795         req->out.args[0].value = &req->misc.init_out;
796         req->end = process_init_reply;
797         request_send_background(fc, req);
798 }
799
800 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
801 {
802         struct fuse_conn *fc;
803         struct inode *root;
804         struct fuse_mount_data d;
805         struct file *file;
806         struct dentry *root_dentry;
807         struct fuse_req *init_req;
808         int err;
809         int is_bdev = sb->s_bdev != NULL;
810
811         if (sb->s_flags & MS_MANDLOCK)
812                 return -EINVAL;
813
814         if (!parse_fuse_opt((char *) data, &d, is_bdev))
815                 return -EINVAL;
816
817         if (is_bdev) {
818 #ifdef CONFIG_BLOCK
819                 if (!sb_set_blocksize(sb, d.blksize))
820                         return -EINVAL;
821 #endif
822         } else {
823                 sb->s_blocksize = PAGE_CACHE_SIZE;
824                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
825         }
826         sb->s_magic = FUSE_SUPER_MAGIC;
827         sb->s_op = &fuse_super_operations;
828         sb->s_maxbytes = MAX_LFS_FILESIZE;
829         sb->s_export_op = &fuse_export_operations;
830
831         file = fget(d.fd);
832         if (!file)
833                 return -EINVAL;
834
835         if (file->f_op != &fuse_dev_operations)
836                 return -EINVAL;
837
838         fc = new_conn(sb);
839         if (!fc)
840                 return -ENOMEM;
841
842         fc->flags = d.flags;
843         fc->user_id = d.user_id;
844         fc->group_id = d.group_id;
845         fc->max_read = max_t(unsigned, 4096, d.max_read);
846
847         /* Used by get_root_inode() */
848         sb->s_fs_info = fc;
849
850         err = -ENOMEM;
851         root = get_root_inode(sb, d.rootmode);
852         if (!root)
853                 goto err;
854
855         root_dentry = d_alloc_root(root);
856         if (!root_dentry) {
857                 iput(root);
858                 goto err;
859         }
860
861         init_req = fuse_request_alloc();
862         if (!init_req)
863                 goto err_put_root;
864
865         if (is_bdev) {
866                 fc->destroy_req = fuse_request_alloc();
867                 if (!fc->destroy_req)
868                         goto err_put_root;
869         }
870
871         mutex_lock(&fuse_mutex);
872         err = -EINVAL;
873         if (file->private_data)
874                 goto err_unlock;
875
876         err = fuse_ctl_add_conn(fc);
877         if (err)
878                 goto err_unlock;
879
880         list_add_tail(&fc->entry, &fuse_conn_list);
881         sb->s_root = root_dentry;
882         fc->connected = 1;
883         file->private_data = fuse_conn_get(fc);
884         mutex_unlock(&fuse_mutex);
885         /*
886          * atomic_dec_and_test() in fput() provides the necessary
887          * memory barrier for file->private_data to be visible on all
888          * CPUs after this
889          */
890         fput(file);
891
892         fuse_send_init(fc, init_req);
893
894         return 0;
895
896  err_unlock:
897         mutex_unlock(&fuse_mutex);
898         fuse_request_free(init_req);
899  err_put_root:
900         dput(root_dentry);
901  err:
902         fput(file);
903         fuse_conn_put(fc);
904         return err;
905 }
906
907 static int fuse_get_sb(struct file_system_type *fs_type,
908                        int flags, const char *dev_name,
909                        void *raw_data, struct vfsmount *mnt)
910 {
911         return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
912 }
913
914 static struct file_system_type fuse_fs_type = {
915         .owner          = THIS_MODULE,
916         .name           = "fuse",
917         .fs_flags       = FS_HAS_SUBTYPE,
918         .get_sb         = fuse_get_sb,
919         .kill_sb        = kill_anon_super,
920 };
921
922 #ifdef CONFIG_BLOCK
923 static int fuse_get_sb_blk(struct file_system_type *fs_type,
924                            int flags, const char *dev_name,
925                            void *raw_data, struct vfsmount *mnt)
926 {
927         return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
928                            mnt);
929 }
930
931 static struct file_system_type fuseblk_fs_type = {
932         .owner          = THIS_MODULE,
933         .name           = "fuseblk",
934         .get_sb         = fuse_get_sb_blk,
935         .kill_sb        = kill_block_super,
936         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
937 };
938
939 static inline int register_fuseblk(void)
940 {
941         return register_filesystem(&fuseblk_fs_type);
942 }
943
944 static inline void unregister_fuseblk(void)
945 {
946         unregister_filesystem(&fuseblk_fs_type);
947 }
948 #else
949 static inline int register_fuseblk(void)
950 {
951         return 0;
952 }
953
954 static inline void unregister_fuseblk(void)
955 {
956 }
957 #endif
958
959 static void fuse_inode_init_once(void *foo)
960 {
961         struct inode * inode = foo;
962
963         inode_init_once(inode);
964 }
965
966 static int __init fuse_fs_init(void)
967 {
968         int err;
969
970         err = register_filesystem(&fuse_fs_type);
971         if (err)
972                 goto out;
973
974         err = register_fuseblk();
975         if (err)
976                 goto out_unreg;
977
978         fuse_inode_cachep = kmem_cache_create("fuse_inode",
979                                               sizeof(struct fuse_inode),
980                                               0, SLAB_HWCACHE_ALIGN,
981                                               fuse_inode_init_once);
982         err = -ENOMEM;
983         if (!fuse_inode_cachep)
984                 goto out_unreg2;
985
986         return 0;
987
988  out_unreg2:
989         unregister_fuseblk();
990  out_unreg:
991         unregister_filesystem(&fuse_fs_type);
992  out:
993         return err;
994 }
995
996 static void fuse_fs_cleanup(void)
997 {
998         unregister_filesystem(&fuse_fs_type);
999         unregister_fuseblk();
1000         kmem_cache_destroy(fuse_inode_cachep);
1001 }
1002
1003 static struct kobject *fuse_kobj;
1004 static struct kobject *connections_kobj;
1005
1006 static int fuse_sysfs_init(void)
1007 {
1008         int err;
1009
1010         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1011         if (!fuse_kobj) {
1012                 err = -ENOMEM;
1013                 goto out_err;
1014         }
1015
1016         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1017         if (!connections_kobj) {
1018                 err = -ENOMEM;
1019                 goto out_fuse_unregister;
1020         }
1021
1022         return 0;
1023
1024  out_fuse_unregister:
1025         kobject_put(fuse_kobj);
1026  out_err:
1027         return err;
1028 }
1029
1030 static void fuse_sysfs_cleanup(void)
1031 {
1032         kobject_put(connections_kobj);
1033         kobject_put(fuse_kobj);
1034 }
1035
1036 static int __init fuse_init(void)
1037 {
1038         int res;
1039
1040         printk("fuse init (API version %i.%i)\n",
1041                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1042
1043         INIT_LIST_HEAD(&fuse_conn_list);
1044         res = fuse_fs_init();
1045         if (res)
1046                 goto err;
1047
1048         res = fuse_dev_init();
1049         if (res)
1050                 goto err_fs_cleanup;
1051
1052         res = fuse_sysfs_init();
1053         if (res)
1054                 goto err_dev_cleanup;
1055
1056         res = fuse_ctl_init();
1057         if (res)
1058                 goto err_sysfs_cleanup;
1059
1060         return 0;
1061
1062  err_sysfs_cleanup:
1063         fuse_sysfs_cleanup();
1064  err_dev_cleanup:
1065         fuse_dev_cleanup();
1066  err_fs_cleanup:
1067         fuse_fs_cleanup();
1068  err:
1069         return res;
1070 }
1071
1072 static void __exit fuse_exit(void)
1073 {
1074         printk(KERN_DEBUG "fuse exit\n");
1075
1076         fuse_ctl_cleanup();
1077         fuse_sysfs_cleanup();
1078         fuse_fs_cleanup();
1079         fuse_dev_cleanup();
1080 }
1081
1082 module_init(fuse_init);
1083 module_exit(fuse_exit);