fuse: separate out input queue
[cascardo/linux.git] / fs / fuse / inode.c
1 /*
2   FUSE: Filesystem in Userspace
3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
4
5   This program can be distributed under the terms of the GNU GPL.
6   See the file COPYING.
7 */
8
9 #include "fuse_i.h"
10
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/seq_file.h>
15 #include <linux/init.h>
16 #include <linux/module.h>
17 #include <linux/moduleparam.h>
18 #include <linux/parser.h>
19 #include <linux/statfs.h>
20 #include <linux/random.h>
21 #include <linux/sched.h>
22 #include <linux/exportfs.h>
23
24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
25 MODULE_DESCRIPTION("Filesystem in Userspace");
26 MODULE_LICENSE("GPL");
27
28 static struct kmem_cache *fuse_inode_cachep;
29 struct list_head fuse_conn_list;
30 DEFINE_MUTEX(fuse_mutex);
31
32 static int set_global_limit(const char *val, struct kernel_param *kp);
33
34 unsigned max_user_bgreq;
35 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
36                   &max_user_bgreq, 0644);
37 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
38 MODULE_PARM_DESC(max_user_bgreq,
39  "Global limit for the maximum number of backgrounded requests an "
40  "unprivileged user can set");
41
42 unsigned max_user_congthresh;
43 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
44                   &max_user_congthresh, 0644);
45 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
46 MODULE_PARM_DESC(max_user_congthresh,
47  "Global limit for the maximum congestion threshold an "
48  "unprivileged user can set");
49
50 #define FUSE_SUPER_MAGIC 0x65735546
51
52 #define FUSE_DEFAULT_BLKSIZE 512
53
54 /** Maximum number of outstanding background requests */
55 #define FUSE_DEFAULT_MAX_BACKGROUND 12
56
57 /** Congestion starts at 75% of maximum */
58 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
59
60 struct fuse_mount_data {
61         int fd;
62         unsigned rootmode;
63         kuid_t user_id;
64         kgid_t group_id;
65         unsigned fd_present:1;
66         unsigned rootmode_present:1;
67         unsigned user_id_present:1;
68         unsigned group_id_present:1;
69         unsigned flags;
70         unsigned max_read;
71         unsigned blksize;
72 };
73
74 struct fuse_forget_link *fuse_alloc_forget(void)
75 {
76         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
77 }
78
79 static struct inode *fuse_alloc_inode(struct super_block *sb)
80 {
81         struct inode *inode;
82         struct fuse_inode *fi;
83
84         inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
85         if (!inode)
86                 return NULL;
87
88         fi = get_fuse_inode(inode);
89         fi->i_time = 0;
90         fi->nodeid = 0;
91         fi->nlookup = 0;
92         fi->attr_version = 0;
93         fi->writectr = 0;
94         fi->orig_ino = 0;
95         fi->state = 0;
96         INIT_LIST_HEAD(&fi->write_files);
97         INIT_LIST_HEAD(&fi->queued_writes);
98         INIT_LIST_HEAD(&fi->writepages);
99         init_waitqueue_head(&fi->page_waitq);
100         fi->forget = fuse_alloc_forget();
101         if (!fi->forget) {
102                 kmem_cache_free(fuse_inode_cachep, inode);
103                 return NULL;
104         }
105
106         return inode;
107 }
108
109 static void fuse_i_callback(struct rcu_head *head)
110 {
111         struct inode *inode = container_of(head, struct inode, i_rcu);
112         kmem_cache_free(fuse_inode_cachep, inode);
113 }
114
115 static void fuse_destroy_inode(struct inode *inode)
116 {
117         struct fuse_inode *fi = get_fuse_inode(inode);
118         BUG_ON(!list_empty(&fi->write_files));
119         BUG_ON(!list_empty(&fi->queued_writes));
120         kfree(fi->forget);
121         call_rcu(&inode->i_rcu, fuse_i_callback);
122 }
123
124 static void fuse_evict_inode(struct inode *inode)
125 {
126         truncate_inode_pages_final(&inode->i_data);
127         clear_inode(inode);
128         if (inode->i_sb->s_flags & MS_ACTIVE) {
129                 struct fuse_conn *fc = get_fuse_conn(inode);
130                 struct fuse_inode *fi = get_fuse_inode(inode);
131                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
132                 fi->forget = NULL;
133         }
134 }
135
136 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
137 {
138         sync_filesystem(sb);
139         if (*flags & MS_MANDLOCK)
140                 return -EINVAL;
141
142         return 0;
143 }
144
145 /*
146  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
147  * so that it will fit.
148  */
149 static ino_t fuse_squash_ino(u64 ino64)
150 {
151         ino_t ino = (ino_t) ino64;
152         if (sizeof(ino_t) < sizeof(u64))
153                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
154         return ino;
155 }
156
157 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
158                                    u64 attr_valid)
159 {
160         struct fuse_conn *fc = get_fuse_conn(inode);
161         struct fuse_inode *fi = get_fuse_inode(inode);
162
163         fi->attr_version = ++fc->attr_version;
164         fi->i_time = attr_valid;
165
166         inode->i_ino     = fuse_squash_ino(attr->ino);
167         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
168         set_nlink(inode, attr->nlink);
169         inode->i_uid     = make_kuid(&init_user_ns, attr->uid);
170         inode->i_gid     = make_kgid(&init_user_ns, attr->gid);
171         inode->i_blocks  = attr->blocks;
172         inode->i_atime.tv_sec   = attr->atime;
173         inode->i_atime.tv_nsec  = attr->atimensec;
174         /* mtime from server may be stale due to local buffered write */
175         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
176                 inode->i_mtime.tv_sec   = attr->mtime;
177                 inode->i_mtime.tv_nsec  = attr->mtimensec;
178                 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                 fuse_init_inode(inode, attr);
312                 unlock_new_inode(inode);
313         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
314                 /* Inode has changed type, any I/O on the old should fail */
315                 make_bad_inode(inode);
316                 iput(inode);
317                 goto retry;
318         }
319
320         fi = get_fuse_inode(inode);
321         spin_lock(&fc->lock);
322         fi->nlookup++;
323         spin_unlock(&fc->lock);
324         fuse_change_attributes(inode, attr, attr_valid, attr_version);
325
326         return inode;
327 }
328
329 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
330                              loff_t offset, loff_t len)
331 {
332         struct inode *inode;
333         pgoff_t pg_start;
334         pgoff_t pg_end;
335
336         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
337         if (!inode)
338                 return -ENOENT;
339
340         fuse_invalidate_attr(inode);
341         if (offset >= 0) {
342                 pg_start = offset >> PAGE_CACHE_SHIFT;
343                 if (len <= 0)
344                         pg_end = -1;
345                 else
346                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
347                 invalidate_inode_pages2_range(inode->i_mapping,
348                                               pg_start, pg_end);
349         }
350         iput(inode);
351         return 0;
352 }
353
354 static void fuse_umount_begin(struct super_block *sb)
355 {
356         fuse_abort_conn(get_fuse_conn_super(sb));
357 }
358
359 static void fuse_send_destroy(struct fuse_conn *fc)
360 {
361         struct fuse_req *req = fc->destroy_req;
362         if (req && fc->conn_init) {
363                 fc->destroy_req = NULL;
364                 req->in.h.opcode = FUSE_DESTROY;
365                 __set_bit(FR_FORCE, &req->flags);
366                 __clear_bit(FR_BACKGROUND, &req->flags);
367                 fuse_request_send(fc, req);
368                 fuse_put_request(fc, req);
369         }
370 }
371
372 static void fuse_bdi_destroy(struct fuse_conn *fc)
373 {
374         if (fc->bdi_initialized)
375                 bdi_destroy(&fc->bdi);
376 }
377
378 static void fuse_put_super(struct super_block *sb)
379 {
380         struct fuse_conn *fc = get_fuse_conn_super(sb);
381
382         fuse_send_destroy(fc);
383
384         fuse_abort_conn(fc);
385         mutex_lock(&fuse_mutex);
386         list_del(&fc->entry);
387         fuse_ctl_remove_conn(fc);
388         mutex_unlock(&fuse_mutex);
389         fuse_bdi_destroy(fc);
390
391         fuse_conn_put(fc);
392 }
393
394 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
395 {
396         stbuf->f_type    = FUSE_SUPER_MAGIC;
397         stbuf->f_bsize   = attr->bsize;
398         stbuf->f_frsize  = attr->frsize;
399         stbuf->f_blocks  = attr->blocks;
400         stbuf->f_bfree   = attr->bfree;
401         stbuf->f_bavail  = attr->bavail;
402         stbuf->f_files   = attr->files;
403         stbuf->f_ffree   = attr->ffree;
404         stbuf->f_namelen = attr->namelen;
405         /* fsid is left zero */
406 }
407
408 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
409 {
410         struct super_block *sb = dentry->d_sb;
411         struct fuse_conn *fc = get_fuse_conn_super(sb);
412         FUSE_ARGS(args);
413         struct fuse_statfs_out outarg;
414         int err;
415
416         if (!fuse_allow_current_process(fc)) {
417                 buf->f_type = FUSE_SUPER_MAGIC;
418                 return 0;
419         }
420
421         memset(&outarg, 0, sizeof(outarg));
422         args.in.numargs = 0;
423         args.in.h.opcode = FUSE_STATFS;
424         args.in.h.nodeid = get_node_id(d_inode(dentry));
425         args.out.numargs = 1;
426         args.out.args[0].size = sizeof(outarg);
427         args.out.args[0].value = &outarg;
428         err = fuse_simple_request(fc, &args);
429         if (!err)
430                 convert_fuse_statfs(buf, &outarg.st);
431         return err;
432 }
433
434 enum {
435         OPT_FD,
436         OPT_ROOTMODE,
437         OPT_USER_ID,
438         OPT_GROUP_ID,
439         OPT_DEFAULT_PERMISSIONS,
440         OPT_ALLOW_OTHER,
441         OPT_MAX_READ,
442         OPT_BLKSIZE,
443         OPT_ERR
444 };
445
446 static const match_table_t tokens = {
447         {OPT_FD,                        "fd=%u"},
448         {OPT_ROOTMODE,                  "rootmode=%o"},
449         {OPT_USER_ID,                   "user_id=%u"},
450         {OPT_GROUP_ID,                  "group_id=%u"},
451         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
452         {OPT_ALLOW_OTHER,               "allow_other"},
453         {OPT_MAX_READ,                  "max_read=%u"},
454         {OPT_BLKSIZE,                   "blksize=%u"},
455         {OPT_ERR,                       NULL}
456 };
457
458 static int fuse_match_uint(substring_t *s, unsigned int *res)
459 {
460         int err = -ENOMEM;
461         char *buf = match_strdup(s);
462         if (buf) {
463                 err = kstrtouint(buf, 10, res);
464                 kfree(buf);
465         }
466         return err;
467 }
468
469 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
470 {
471         char *p;
472         memset(d, 0, sizeof(struct fuse_mount_data));
473         d->max_read = ~0;
474         d->blksize = FUSE_DEFAULT_BLKSIZE;
475
476         while ((p = strsep(&opt, ",")) != NULL) {
477                 int token;
478                 int value;
479                 unsigned uv;
480                 substring_t args[MAX_OPT_ARGS];
481                 if (!*p)
482                         continue;
483
484                 token = match_token(p, tokens, args);
485                 switch (token) {
486                 case OPT_FD:
487                         if (match_int(&args[0], &value))
488                                 return 0;
489                         d->fd = value;
490                         d->fd_present = 1;
491                         break;
492
493                 case OPT_ROOTMODE:
494                         if (match_octal(&args[0], &value))
495                                 return 0;
496                         if (!fuse_valid_type(value))
497                                 return 0;
498                         d->rootmode = value;
499                         d->rootmode_present = 1;
500                         break;
501
502                 case OPT_USER_ID:
503                         if (fuse_match_uint(&args[0], &uv))
504                                 return 0;
505                         d->user_id = make_kuid(current_user_ns(), uv);
506                         if (!uid_valid(d->user_id))
507                                 return 0;
508                         d->user_id_present = 1;
509                         break;
510
511                 case OPT_GROUP_ID:
512                         if (fuse_match_uint(&args[0], &uv))
513                                 return 0;
514                         d->group_id = make_kgid(current_user_ns(), uv);
515                         if (!gid_valid(d->group_id))
516                                 return 0;
517                         d->group_id_present = 1;
518                         break;
519
520                 case OPT_DEFAULT_PERMISSIONS:
521                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
522                         break;
523
524                 case OPT_ALLOW_OTHER:
525                         d->flags |= FUSE_ALLOW_OTHER;
526                         break;
527
528                 case OPT_MAX_READ:
529                         if (match_int(&args[0], &value))
530                                 return 0;
531                         d->max_read = value;
532                         break;
533
534                 case OPT_BLKSIZE:
535                         if (!is_bdev || match_int(&args[0], &value))
536                                 return 0;
537                         d->blksize = value;
538                         break;
539
540                 default:
541                         return 0;
542                 }
543         }
544
545         if (!d->fd_present || !d->rootmode_present ||
546             !d->user_id_present || !d->group_id_present)
547                 return 0;
548
549         return 1;
550 }
551
552 static int fuse_show_options(struct seq_file *m, struct dentry *root)
553 {
554         struct super_block *sb = root->d_sb;
555         struct fuse_conn *fc = get_fuse_conn_super(sb);
556
557         seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
558         seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
559         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
560                 seq_puts(m, ",default_permissions");
561         if (fc->flags & FUSE_ALLOW_OTHER)
562                 seq_puts(m, ",allow_other");
563         if (fc->max_read != ~0)
564                 seq_printf(m, ",max_read=%u", fc->max_read);
565         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
566                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
567         return 0;
568 }
569
570 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
571 {
572         memset(fiq, 0, sizeof(struct fuse_iqueue));
573         init_waitqueue_head(&fiq->waitq);
574         INIT_LIST_HEAD(&fiq->pending);
575         INIT_LIST_HEAD(&fiq->interrupts);
576         fiq->forget_list_tail = &fiq->forget_list_head;
577 }
578
579 void fuse_conn_init(struct fuse_conn *fc)
580 {
581         memset(fc, 0, sizeof(*fc));
582         spin_lock_init(&fc->lock);
583         init_rwsem(&fc->killsb);
584         atomic_set(&fc->count, 1);
585         init_waitqueue_head(&fc->blocked_waitq);
586         init_waitqueue_head(&fc->reserved_req_waitq);
587         fuse_iqueue_init(&fc->iq);
588         INIT_LIST_HEAD(&fc->processing);
589         INIT_LIST_HEAD(&fc->io);
590         INIT_LIST_HEAD(&fc->bg_queue);
591         INIT_LIST_HEAD(&fc->entry);
592         atomic_set(&fc->num_waiting, 0);
593         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
594         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
595         fc->khctr = 0;
596         fc->polled_files = RB_ROOT;
597         fc->blocked = 0;
598         fc->initialized = 0;
599         fc->attr_version = 1;
600         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
601 }
602 EXPORT_SYMBOL_GPL(fuse_conn_init);
603
604 void fuse_conn_put(struct fuse_conn *fc)
605 {
606         if (atomic_dec_and_test(&fc->count)) {
607                 if (fc->destroy_req)
608                         fuse_request_free(fc->destroy_req);
609                 fc->release(fc);
610         }
611 }
612 EXPORT_SYMBOL_GPL(fuse_conn_put);
613
614 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
615 {
616         atomic_inc(&fc->count);
617         return fc;
618 }
619 EXPORT_SYMBOL_GPL(fuse_conn_get);
620
621 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
622 {
623         struct fuse_attr attr;
624         memset(&attr, 0, sizeof(attr));
625
626         attr.mode = mode;
627         attr.ino = FUSE_ROOT_ID;
628         attr.nlink = 1;
629         return fuse_iget(sb, 1, 0, &attr, 0, 0);
630 }
631
632 struct fuse_inode_handle {
633         u64 nodeid;
634         u32 generation;
635 };
636
637 static struct dentry *fuse_get_dentry(struct super_block *sb,
638                                       struct fuse_inode_handle *handle)
639 {
640         struct fuse_conn *fc = get_fuse_conn_super(sb);
641         struct inode *inode;
642         struct dentry *entry;
643         int err = -ESTALE;
644
645         if (handle->nodeid == 0)
646                 goto out_err;
647
648         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
649         if (!inode) {
650                 struct fuse_entry_out outarg;
651                 struct qstr name;
652
653                 if (!fc->export_support)
654                         goto out_err;
655
656                 name.len = 1;
657                 name.name = ".";
658                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
659                                        &inode);
660                 if (err && err != -ENOENT)
661                         goto out_err;
662                 if (err || !inode) {
663                         err = -ESTALE;
664                         goto out_err;
665                 }
666                 err = -EIO;
667                 if (get_node_id(inode) != handle->nodeid)
668                         goto out_iput;
669         }
670         err = -ESTALE;
671         if (inode->i_generation != handle->generation)
672                 goto out_iput;
673
674         entry = d_obtain_alias(inode);
675         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
676                 fuse_invalidate_entry_cache(entry);
677
678         return entry;
679
680  out_iput:
681         iput(inode);
682  out_err:
683         return ERR_PTR(err);
684 }
685
686 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
687                            struct inode *parent)
688 {
689         int len = parent ? 6 : 3;
690         u64 nodeid;
691         u32 generation;
692
693         if (*max_len < len) {
694                 *max_len = len;
695                 return  FILEID_INVALID;
696         }
697
698         nodeid = get_fuse_inode(inode)->nodeid;
699         generation = inode->i_generation;
700
701         fh[0] = (u32)(nodeid >> 32);
702         fh[1] = (u32)(nodeid & 0xffffffff);
703         fh[2] = generation;
704
705         if (parent) {
706                 nodeid = get_fuse_inode(parent)->nodeid;
707                 generation = parent->i_generation;
708
709                 fh[3] = (u32)(nodeid >> 32);
710                 fh[4] = (u32)(nodeid & 0xffffffff);
711                 fh[5] = generation;
712         }
713
714         *max_len = len;
715         return parent ? 0x82 : 0x81;
716 }
717
718 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
719                 struct fid *fid, int fh_len, int fh_type)
720 {
721         struct fuse_inode_handle handle;
722
723         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
724                 return NULL;
725
726         handle.nodeid = (u64) fid->raw[0] << 32;
727         handle.nodeid |= (u64) fid->raw[1];
728         handle.generation = fid->raw[2];
729         return fuse_get_dentry(sb, &handle);
730 }
731
732 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
733                 struct fid *fid, int fh_len, int fh_type)
734 {
735         struct fuse_inode_handle parent;
736
737         if (fh_type != 0x82 || fh_len < 6)
738                 return NULL;
739
740         parent.nodeid = (u64) fid->raw[3] << 32;
741         parent.nodeid |= (u64) fid->raw[4];
742         parent.generation = fid->raw[5];
743         return fuse_get_dentry(sb, &parent);
744 }
745
746 static struct dentry *fuse_get_parent(struct dentry *child)
747 {
748         struct inode *child_inode = d_inode(child);
749         struct fuse_conn *fc = get_fuse_conn(child_inode);
750         struct inode *inode;
751         struct dentry *parent;
752         struct fuse_entry_out outarg;
753         struct qstr name;
754         int err;
755
756         if (!fc->export_support)
757                 return ERR_PTR(-ESTALE);
758
759         name.len = 2;
760         name.name = "..";
761         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
762                                &name, &outarg, &inode);
763         if (err) {
764                 if (err == -ENOENT)
765                         return ERR_PTR(-ESTALE);
766                 return ERR_PTR(err);
767         }
768
769         parent = d_obtain_alias(inode);
770         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
771                 fuse_invalidate_entry_cache(parent);
772
773         return parent;
774 }
775
776 static const struct export_operations fuse_export_operations = {
777         .fh_to_dentry   = fuse_fh_to_dentry,
778         .fh_to_parent   = fuse_fh_to_parent,
779         .encode_fh      = fuse_encode_fh,
780         .get_parent     = fuse_get_parent,
781 };
782
783 static const struct super_operations fuse_super_operations = {
784         .alloc_inode    = fuse_alloc_inode,
785         .destroy_inode  = fuse_destroy_inode,
786         .evict_inode    = fuse_evict_inode,
787         .write_inode    = fuse_write_inode,
788         .drop_inode     = generic_delete_inode,
789         .remount_fs     = fuse_remount_fs,
790         .put_super      = fuse_put_super,
791         .umount_begin   = fuse_umount_begin,
792         .statfs         = fuse_statfs,
793         .show_options   = fuse_show_options,
794 };
795
796 static void sanitize_global_limit(unsigned *limit)
797 {
798         if (*limit == 0)
799                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
800                          sizeof(struct fuse_req);
801
802         if (*limit >= 1 << 16)
803                 *limit = (1 << 16) - 1;
804 }
805
806 static int set_global_limit(const char *val, struct kernel_param *kp)
807 {
808         int rv;
809
810         rv = param_set_uint(val, kp);
811         if (rv)
812                 return rv;
813
814         sanitize_global_limit((unsigned *)kp->arg);
815
816         return 0;
817 }
818
819 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
820 {
821         int cap_sys_admin = capable(CAP_SYS_ADMIN);
822
823         if (arg->minor < 13)
824                 return;
825
826         sanitize_global_limit(&max_user_bgreq);
827         sanitize_global_limit(&max_user_congthresh);
828
829         if (arg->max_background) {
830                 fc->max_background = arg->max_background;
831
832                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
833                         fc->max_background = max_user_bgreq;
834         }
835         if (arg->congestion_threshold) {
836                 fc->congestion_threshold = arg->congestion_threshold;
837
838                 if (!cap_sys_admin &&
839                     fc->congestion_threshold > max_user_congthresh)
840                         fc->congestion_threshold = max_user_congthresh;
841         }
842 }
843
844 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
845 {
846         struct fuse_init_out *arg = &req->misc.init_out;
847
848         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
849                 fc->conn_error = 1;
850         else {
851                 unsigned long ra_pages;
852
853                 process_init_limits(fc, arg);
854
855                 if (arg->minor >= 6) {
856                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
857                         if (arg->flags & FUSE_ASYNC_READ)
858                                 fc->async_read = 1;
859                         if (!(arg->flags & FUSE_POSIX_LOCKS))
860                                 fc->no_lock = 1;
861                         if (arg->minor >= 17) {
862                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
863                                         fc->no_flock = 1;
864                         } else {
865                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
866                                         fc->no_flock = 1;
867                         }
868                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
869                                 fc->atomic_o_trunc = 1;
870                         if (arg->minor >= 9) {
871                                 /* LOOKUP has dependency on proto version */
872                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
873                                         fc->export_support = 1;
874                         }
875                         if (arg->flags & FUSE_BIG_WRITES)
876                                 fc->big_writes = 1;
877                         if (arg->flags & FUSE_DONT_MASK)
878                                 fc->dont_mask = 1;
879                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
880                                 fc->auto_inval_data = 1;
881                         if (arg->flags & FUSE_DO_READDIRPLUS) {
882                                 fc->do_readdirplus = 1;
883                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
884                                         fc->readdirplus_auto = 1;
885                         }
886                         if (arg->flags & FUSE_ASYNC_DIO)
887                                 fc->async_dio = 1;
888                         if (arg->flags & FUSE_WRITEBACK_CACHE)
889                                 fc->writeback_cache = 1;
890                         if (arg->time_gran && arg->time_gran <= 1000000000)
891                                 fc->sb->s_time_gran = arg->time_gran;
892                 } else {
893                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
894                         fc->no_lock = 1;
895                         fc->no_flock = 1;
896                 }
897
898                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
899                 fc->minor = arg->minor;
900                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
901                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
902                 fc->conn_init = 1;
903         }
904         fuse_set_initialized(fc);
905         wake_up_all(&fc->blocked_waitq);
906 }
907
908 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
909 {
910         struct fuse_init_in *arg = &req->misc.init_in;
911
912         arg->major = FUSE_KERNEL_VERSION;
913         arg->minor = FUSE_KERNEL_MINOR_VERSION;
914         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
915         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
916                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
917                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
918                 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
919                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
920                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT;
921         req->in.h.opcode = FUSE_INIT;
922         req->in.numargs = 1;
923         req->in.args[0].size = sizeof(*arg);
924         req->in.args[0].value = arg;
925         req->out.numargs = 1;
926         /* Variable length argument used for backward compatibility
927            with interface version < 7.5.  Rest of init_out is zeroed
928            by do_get_request(), so a short reply is not a problem */
929         req->out.argvar = 1;
930         req->out.args[0].size = sizeof(struct fuse_init_out);
931         req->out.args[0].value = &req->misc.init_out;
932         req->end = process_init_reply;
933         fuse_request_send_background(fc, req);
934 }
935
936 static void fuse_free_conn(struct fuse_conn *fc)
937 {
938         kfree_rcu(fc, rcu);
939 }
940
941 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
942 {
943         int err;
944
945         fc->bdi.name = "fuse";
946         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
947         /* fuse does it's own writeback accounting */
948         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
949
950         err = bdi_init(&fc->bdi);
951         if (err)
952                 return err;
953
954         fc->bdi_initialized = 1;
955
956         if (sb->s_bdev) {
957                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
958                                     MAJOR(fc->dev), MINOR(fc->dev));
959         } else {
960                 err = bdi_register_dev(&fc->bdi, fc->dev);
961         }
962
963         if (err)
964                 return err;
965
966         /*
967          * For a single fuse filesystem use max 1% of dirty +
968          * writeback threshold.
969          *
970          * This gives about 1M of write buffer for memory maps on a
971          * machine with 1G and 10% dirty_ratio, which should be more
972          * than enough.
973          *
974          * Privileged users can raise it by writing to
975          *
976          *    /sys/class/bdi/<bdi>/max_ratio
977          */
978         bdi_set_max_ratio(&fc->bdi, 1);
979
980         return 0;
981 }
982
983 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
984 {
985         struct fuse_conn *fc;
986         struct inode *root;
987         struct fuse_mount_data d;
988         struct file *file;
989         struct dentry *root_dentry;
990         struct fuse_req *init_req;
991         int err;
992         int is_bdev = sb->s_bdev != NULL;
993
994         err = -EINVAL;
995         if (sb->s_flags & MS_MANDLOCK)
996                 goto err;
997
998         sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
999
1000         if (!parse_fuse_opt(data, &d, is_bdev))
1001                 goto err;
1002
1003         if (is_bdev) {
1004 #ifdef CONFIG_BLOCK
1005                 err = -EINVAL;
1006                 if (!sb_set_blocksize(sb, d.blksize))
1007                         goto err;
1008 #endif
1009         } else {
1010                 sb->s_blocksize = PAGE_CACHE_SIZE;
1011                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1012         }
1013         sb->s_magic = FUSE_SUPER_MAGIC;
1014         sb->s_op = &fuse_super_operations;
1015         sb->s_maxbytes = MAX_LFS_FILESIZE;
1016         sb->s_time_gran = 1;
1017         sb->s_export_op = &fuse_export_operations;
1018
1019         file = fget(d.fd);
1020         err = -EINVAL;
1021         if (!file)
1022                 goto err;
1023
1024         if ((file->f_op != &fuse_dev_operations) ||
1025             (file->f_cred->user_ns != &init_user_ns))
1026                 goto err_fput;
1027
1028         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1029         err = -ENOMEM;
1030         if (!fc)
1031                 goto err_fput;
1032
1033         fuse_conn_init(fc);
1034         fc->release = fuse_free_conn;
1035
1036         fc->dev = sb->s_dev;
1037         fc->sb = sb;
1038         err = fuse_bdi_init(fc, sb);
1039         if (err)
1040                 goto err_put_conn;
1041
1042         sb->s_bdi = &fc->bdi;
1043
1044         /* Handle umasking inside the fuse code */
1045         if (sb->s_flags & MS_POSIXACL)
1046                 fc->dont_mask = 1;
1047         sb->s_flags |= MS_POSIXACL;
1048
1049         fc->flags = d.flags;
1050         fc->user_id = d.user_id;
1051         fc->group_id = d.group_id;
1052         fc->max_read = max_t(unsigned, 4096, d.max_read);
1053
1054         /* Used by get_root_inode() */
1055         sb->s_fs_info = fc;
1056
1057         err = -ENOMEM;
1058         root = fuse_get_root_inode(sb, d.rootmode);
1059         root_dentry = d_make_root(root);
1060         if (!root_dentry)
1061                 goto err_put_conn;
1062         /* only now - we want root dentry with NULL ->d_op */
1063         sb->s_d_op = &fuse_dentry_operations;
1064
1065         init_req = fuse_request_alloc(0);
1066         if (!init_req)
1067                 goto err_put_root;
1068         __set_bit(FR_BACKGROUND, &init_req->flags);
1069
1070         if (is_bdev) {
1071                 fc->destroy_req = fuse_request_alloc(0);
1072                 if (!fc->destroy_req)
1073                         goto err_free_init_req;
1074         }
1075
1076         mutex_lock(&fuse_mutex);
1077         err = -EINVAL;
1078         if (file->private_data)
1079                 goto err_unlock;
1080
1081         err = fuse_ctl_add_conn(fc);
1082         if (err)
1083                 goto err_unlock;
1084
1085         list_add_tail(&fc->entry, &fuse_conn_list);
1086         sb->s_root = root_dentry;
1087         fc->connected = 1;
1088         file->private_data = fuse_conn_get(fc);
1089         mutex_unlock(&fuse_mutex);
1090         /*
1091          * atomic_dec_and_test() in fput() provides the necessary
1092          * memory barrier for file->private_data to be visible on all
1093          * CPUs after this
1094          */
1095         fput(file);
1096
1097         fuse_send_init(fc, init_req);
1098
1099         return 0;
1100
1101  err_unlock:
1102         mutex_unlock(&fuse_mutex);
1103  err_free_init_req:
1104         fuse_request_free(init_req);
1105  err_put_root:
1106         dput(root_dentry);
1107  err_put_conn:
1108         fuse_bdi_destroy(fc);
1109         fuse_conn_put(fc);
1110  err_fput:
1111         fput(file);
1112  err:
1113         return err;
1114 }
1115
1116 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1117                        int flags, const char *dev_name,
1118                        void *raw_data)
1119 {
1120         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1121 }
1122
1123 static void fuse_kill_sb_anon(struct super_block *sb)
1124 {
1125         struct fuse_conn *fc = get_fuse_conn_super(sb);
1126
1127         if (fc) {
1128                 down_write(&fc->killsb);
1129                 fc->sb = NULL;
1130                 up_write(&fc->killsb);
1131         }
1132
1133         kill_anon_super(sb);
1134 }
1135
1136 static struct file_system_type fuse_fs_type = {
1137         .owner          = THIS_MODULE,
1138         .name           = "fuse",
1139         .fs_flags       = FS_HAS_SUBTYPE,
1140         .mount          = fuse_mount,
1141         .kill_sb        = fuse_kill_sb_anon,
1142 };
1143 MODULE_ALIAS_FS("fuse");
1144
1145 #ifdef CONFIG_BLOCK
1146 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1147                            int flags, const char *dev_name,
1148                            void *raw_data)
1149 {
1150         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1151 }
1152
1153 static void fuse_kill_sb_blk(struct super_block *sb)
1154 {
1155         struct fuse_conn *fc = get_fuse_conn_super(sb);
1156
1157         if (fc) {
1158                 down_write(&fc->killsb);
1159                 fc->sb = NULL;
1160                 up_write(&fc->killsb);
1161         }
1162
1163         kill_block_super(sb);
1164 }
1165
1166 static struct file_system_type fuseblk_fs_type = {
1167         .owner          = THIS_MODULE,
1168         .name           = "fuseblk",
1169         .mount          = fuse_mount_blk,
1170         .kill_sb        = fuse_kill_sb_blk,
1171         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1172 };
1173 MODULE_ALIAS_FS("fuseblk");
1174
1175 static inline int register_fuseblk(void)
1176 {
1177         return register_filesystem(&fuseblk_fs_type);
1178 }
1179
1180 static inline void unregister_fuseblk(void)
1181 {
1182         unregister_filesystem(&fuseblk_fs_type);
1183 }
1184 #else
1185 static inline int register_fuseblk(void)
1186 {
1187         return 0;
1188 }
1189
1190 static inline void unregister_fuseblk(void)
1191 {
1192 }
1193 #endif
1194
1195 static void fuse_inode_init_once(void *foo)
1196 {
1197         struct inode *inode = foo;
1198
1199         inode_init_once(inode);
1200 }
1201
1202 static int __init fuse_fs_init(void)
1203 {
1204         int err;
1205
1206         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1207                                               sizeof(struct fuse_inode),
1208                                               0, SLAB_HWCACHE_ALIGN,
1209                                               fuse_inode_init_once);
1210         err = -ENOMEM;
1211         if (!fuse_inode_cachep)
1212                 goto out;
1213
1214         err = register_fuseblk();
1215         if (err)
1216                 goto out2;
1217
1218         err = register_filesystem(&fuse_fs_type);
1219         if (err)
1220                 goto out3;
1221
1222         return 0;
1223
1224  out3:
1225         unregister_fuseblk();
1226  out2:
1227         kmem_cache_destroy(fuse_inode_cachep);
1228  out:
1229         return err;
1230 }
1231
1232 static void fuse_fs_cleanup(void)
1233 {
1234         unregister_filesystem(&fuse_fs_type);
1235         unregister_fuseblk();
1236
1237         /*
1238          * Make sure all delayed rcu free inodes are flushed before we
1239          * destroy cache.
1240          */
1241         rcu_barrier();
1242         kmem_cache_destroy(fuse_inode_cachep);
1243 }
1244
1245 static struct kobject *fuse_kobj;
1246 static struct kobject *connections_kobj;
1247
1248 static int fuse_sysfs_init(void)
1249 {
1250         int err;
1251
1252         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1253         if (!fuse_kobj) {
1254                 err = -ENOMEM;
1255                 goto out_err;
1256         }
1257
1258         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1259         if (!connections_kobj) {
1260                 err = -ENOMEM;
1261                 goto out_fuse_unregister;
1262         }
1263
1264         return 0;
1265
1266  out_fuse_unregister:
1267         kobject_put(fuse_kobj);
1268  out_err:
1269         return err;
1270 }
1271
1272 static void fuse_sysfs_cleanup(void)
1273 {
1274         kobject_put(connections_kobj);
1275         kobject_put(fuse_kobj);
1276 }
1277
1278 static int __init fuse_init(void)
1279 {
1280         int res;
1281
1282         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1283                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1284
1285         INIT_LIST_HEAD(&fuse_conn_list);
1286         res = fuse_fs_init();
1287         if (res)
1288                 goto err;
1289
1290         res = fuse_dev_init();
1291         if (res)
1292                 goto err_fs_cleanup;
1293
1294         res = fuse_sysfs_init();
1295         if (res)
1296                 goto err_dev_cleanup;
1297
1298         res = fuse_ctl_init();
1299         if (res)
1300                 goto err_sysfs_cleanup;
1301
1302         sanitize_global_limit(&max_user_bgreq);
1303         sanitize_global_limit(&max_user_congthresh);
1304
1305         return 0;
1306
1307  err_sysfs_cleanup:
1308         fuse_sysfs_cleanup();
1309  err_dev_cleanup:
1310         fuse_dev_cleanup();
1311  err_fs_cleanup:
1312         fuse_fs_cleanup();
1313  err:
1314         return res;
1315 }
1316
1317 static void __exit fuse_exit(void)
1318 {
1319         printk(KERN_DEBUG "fuse exit\n");
1320
1321         fuse_ctl_cleanup();
1322         fuse_sysfs_cleanup();
1323         fuse_fs_cleanup();
1324         fuse_dev_cleanup();
1325 }
1326
1327 module_init(fuse_init);
1328 module_exit(fuse_exit);