Merge tag 'for-linus' of git://git.armlinux.org.uk/~rmk/linux-arm
[platform/kernel/linux-rpi.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/fs_context.h>
19 #include <linux/fs_parser.h>
20 #include <linux/statfs.h>
21 #include <linux/random.h>
22 #include <linux/sched.h>
23 #include <linux/exportfs.h>
24 #include <linux/posix_acl.h>
25 #include <linux/pid_namespace.h>
26
27 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
28 MODULE_DESCRIPTION("Filesystem in Userspace");
29 MODULE_LICENSE("GPL");
30
31 static struct kmem_cache *fuse_inode_cachep;
32 struct list_head fuse_conn_list;
33 DEFINE_MUTEX(fuse_mutex);
34
35 static int set_global_limit(const char *val, const struct kernel_param *kp);
36
37 unsigned max_user_bgreq;
38 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
39                   &max_user_bgreq, 0644);
40 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
41 MODULE_PARM_DESC(max_user_bgreq,
42  "Global limit for the maximum number of backgrounded requests an "
43  "unprivileged user can set");
44
45 unsigned max_user_congthresh;
46 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
47                   &max_user_congthresh, 0644);
48 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
49 MODULE_PARM_DESC(max_user_congthresh,
50  "Global limit for the maximum congestion threshold an "
51  "unprivileged user can set");
52
53 #define FUSE_SUPER_MAGIC 0x65735546
54
55 #define FUSE_DEFAULT_BLKSIZE 512
56
57 /** Maximum number of outstanding background requests */
58 #define FUSE_DEFAULT_MAX_BACKGROUND 12
59
60 /** Congestion starts at 75% of maximum */
61 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
62
63 #ifdef CONFIG_BLOCK
64 static struct file_system_type fuseblk_fs_type;
65 #endif
66
67 struct fuse_forget_link *fuse_alloc_forget(void)
68 {
69         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
70 }
71
72 static struct inode *fuse_alloc_inode(struct super_block *sb)
73 {
74         struct fuse_inode *fi;
75
76         fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
77         if (!fi)
78                 return NULL;
79
80         fi->i_time = 0;
81         fi->inval_mask = 0;
82         fi->nodeid = 0;
83         fi->nlookup = 0;
84         fi->attr_version = 0;
85         fi->orig_ino = 0;
86         fi->state = 0;
87         mutex_init(&fi->mutex);
88         spin_lock_init(&fi->lock);
89         fi->forget = fuse_alloc_forget();
90         if (!fi->forget)
91                 goto out_free;
92
93         if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
94                 goto out_free_forget;
95
96         return &fi->inode;
97
98 out_free_forget:
99         kfree(fi->forget);
100 out_free:
101         kmem_cache_free(fuse_inode_cachep, fi);
102         return NULL;
103 }
104
105 static void fuse_free_inode(struct inode *inode)
106 {
107         struct fuse_inode *fi = get_fuse_inode(inode);
108
109         mutex_destroy(&fi->mutex);
110         kfree(fi->forget);
111 #ifdef CONFIG_FUSE_DAX
112         kfree(fi->dax);
113 #endif
114         kmem_cache_free(fuse_inode_cachep, fi);
115 }
116
117 static void fuse_evict_inode(struct inode *inode)
118 {
119         struct fuse_inode *fi = get_fuse_inode(inode);
120
121         truncate_inode_pages_final(&inode->i_data);
122         clear_inode(inode);
123         if (inode->i_sb->s_flags & SB_ACTIVE) {
124                 struct fuse_conn *fc = get_fuse_conn(inode);
125
126                 if (FUSE_IS_DAX(inode))
127                         fuse_dax_inode_cleanup(inode);
128                 if (fi->nlookup) {
129                         fuse_queue_forget(fc, fi->forget, fi->nodeid,
130                                           fi->nlookup);
131                         fi->forget = NULL;
132                 }
133         }
134         if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
135                 WARN_ON(!list_empty(&fi->write_files));
136                 WARN_ON(!list_empty(&fi->queued_writes));
137         }
138 }
139
140 static int fuse_reconfigure(struct fs_context *fsc)
141 {
142         struct super_block *sb = fsc->root->d_sb;
143
144         sync_filesystem(sb);
145         if (fsc->sb_flags & SB_MANDLOCK)
146                 return -EINVAL;
147
148         return 0;
149 }
150
151 /*
152  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
153  * so that it will fit.
154  */
155 static ino_t fuse_squash_ino(u64 ino64)
156 {
157         ino_t ino = (ino_t) ino64;
158         if (sizeof(ino_t) < sizeof(u64))
159                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
160         return ino;
161 }
162
163 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
164                                    u64 attr_valid)
165 {
166         struct fuse_conn *fc = get_fuse_conn(inode);
167         struct fuse_inode *fi = get_fuse_inode(inode);
168
169         lockdep_assert_held(&fi->lock);
170
171         fi->attr_version = atomic64_inc_return(&fc->attr_version);
172         fi->i_time = attr_valid;
173         WRITE_ONCE(fi->inval_mask, 0);
174
175         inode->i_ino     = fuse_squash_ino(attr->ino);
176         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
177         set_nlink(inode, attr->nlink);
178         inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
179         inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
180         inode->i_blocks  = attr->blocks;
181         inode->i_atime.tv_sec   = attr->atime;
182         inode->i_atime.tv_nsec  = attr->atimensec;
183         /* mtime from server may be stale due to local buffered write */
184         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
185                 inode->i_mtime.tv_sec   = attr->mtime;
186                 inode->i_mtime.tv_nsec  = attr->mtimensec;
187                 inode->i_ctime.tv_sec   = attr->ctime;
188                 inode->i_ctime.tv_nsec  = attr->ctimensec;
189         }
190
191         if (attr->blksize != 0)
192                 inode->i_blkbits = ilog2(attr->blksize);
193         else
194                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
195
196         /*
197          * Don't set the sticky bit in i_mode, unless we want the VFS
198          * to check permissions.  This prevents failures due to the
199          * check in may_delete().
200          */
201         fi->orig_i_mode = inode->i_mode;
202         if (!fc->default_permissions)
203                 inode->i_mode &= ~S_ISVTX;
204
205         fi->orig_ino = attr->ino;
206
207         /*
208          * We are refreshing inode data and it is possible that another
209          * client set suid/sgid or security.capability xattr. So clear
210          * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
211          * was set or if security.capability xattr was set. But we don't
212          * know if security.capability has been set or not. So clear it
213          * anyway. Its less efficient but should be safe.
214          */
215         inode->i_flags &= ~S_NOSEC;
216 }
217
218 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
219                             u64 attr_valid, u64 attr_version)
220 {
221         struct fuse_conn *fc = get_fuse_conn(inode);
222         struct fuse_inode *fi = get_fuse_inode(inode);
223         bool is_wb = fc->writeback_cache;
224         loff_t oldsize;
225         struct timespec64 old_mtime;
226
227         spin_lock(&fi->lock);
228         if ((attr_version != 0 && fi->attr_version > attr_version) ||
229             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
230                 spin_unlock(&fi->lock);
231                 return;
232         }
233
234         old_mtime = inode->i_mtime;
235         fuse_change_attributes_common(inode, attr, attr_valid);
236
237         oldsize = inode->i_size;
238         /*
239          * In case of writeback_cache enabled, the cached writes beyond EOF
240          * extend local i_size without keeping userspace server in sync. So,
241          * attr->size coming from server can be stale. We cannot trust it.
242          */
243         if (!is_wb || !S_ISREG(inode->i_mode))
244                 i_size_write(inode, attr->size);
245         spin_unlock(&fi->lock);
246
247         if (!is_wb && S_ISREG(inode->i_mode)) {
248                 bool inval = false;
249
250                 if (oldsize != attr->size) {
251                         truncate_pagecache(inode, attr->size);
252                         if (!fc->explicit_inval_data)
253                                 inval = true;
254                 } else if (fc->auto_inval_data) {
255                         struct timespec64 new_mtime = {
256                                 .tv_sec = attr->mtime,
257                                 .tv_nsec = attr->mtimensec,
258                         };
259
260                         /*
261                          * Auto inval mode also checks and invalidates if mtime
262                          * has changed.
263                          */
264                         if (!timespec64_equal(&old_mtime, &new_mtime))
265                                 inval = true;
266                 }
267
268                 if (inval)
269                         invalidate_inode_pages2(inode->i_mapping);
270         }
271 }
272
273 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
274 {
275         inode->i_mode = attr->mode & S_IFMT;
276         inode->i_size = attr->size;
277         inode->i_mtime.tv_sec  = attr->mtime;
278         inode->i_mtime.tv_nsec = attr->mtimensec;
279         inode->i_ctime.tv_sec  = attr->ctime;
280         inode->i_ctime.tv_nsec = attr->ctimensec;
281         if (S_ISREG(inode->i_mode)) {
282                 fuse_init_common(inode);
283                 fuse_init_file_inode(inode);
284         } else if (S_ISDIR(inode->i_mode))
285                 fuse_init_dir(inode);
286         else if (S_ISLNK(inode->i_mode))
287                 fuse_init_symlink(inode);
288         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
289                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
290                 fuse_init_common(inode);
291                 init_special_inode(inode, inode->i_mode,
292                                    new_decode_dev(attr->rdev));
293         } else
294                 BUG();
295 }
296
297 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
298 {
299         u64 nodeid = *(u64 *) _nodeidp;
300         if (get_node_id(inode) == nodeid)
301                 return 1;
302         else
303                 return 0;
304 }
305
306 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
307 {
308         u64 nodeid = *(u64 *) _nodeidp;
309         get_fuse_inode(inode)->nodeid = nodeid;
310         return 0;
311 }
312
313 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
314                         int generation, struct fuse_attr *attr,
315                         u64 attr_valid, u64 attr_version)
316 {
317         struct inode *inode;
318         struct fuse_inode *fi;
319         struct fuse_conn *fc = get_fuse_conn_super(sb);
320
321         /*
322          * Auto mount points get their node id from the submount root, which is
323          * not a unique identifier within this filesystem.
324          *
325          * To avoid conflicts, do not place submount points into the inode hash
326          * table.
327          */
328         if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
329             S_ISDIR(attr->mode)) {
330                 inode = new_inode(sb);
331                 if (!inode)
332                         return NULL;
333
334                 fuse_init_inode(inode, attr);
335                 get_fuse_inode(inode)->nodeid = nodeid;
336                 inode->i_flags |= S_AUTOMOUNT;
337                 goto done;
338         }
339
340 retry:
341         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
342         if (!inode)
343                 return NULL;
344
345         if ((inode->i_state & I_NEW)) {
346                 inode->i_flags |= S_NOATIME;
347                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
348                         inode->i_flags |= S_NOCMTIME;
349                 inode->i_generation = generation;
350                 fuse_init_inode(inode, attr);
351                 unlock_new_inode(inode);
352         } else if (fuse_stale_inode(inode, generation, attr)) {
353                 /* nodeid was reused, any I/O on the old inode should fail */
354                 fuse_make_bad(inode);
355                 iput(inode);
356                 goto retry;
357         }
358 done:
359         fi = get_fuse_inode(inode);
360         spin_lock(&fi->lock);
361         fi->nlookup++;
362         spin_unlock(&fi->lock);
363         fuse_change_attributes(inode, attr, attr_valid, attr_version);
364
365         return inode;
366 }
367
368 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
369                            struct fuse_mount **fm)
370 {
371         struct fuse_mount *fm_iter;
372         struct inode *inode;
373
374         WARN_ON(!rwsem_is_locked(&fc->killsb));
375         list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
376                 if (!fm_iter->sb)
377                         continue;
378
379                 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
380                 if (inode) {
381                         if (fm)
382                                 *fm = fm_iter;
383                         return inode;
384                 }
385         }
386
387         return NULL;
388 }
389
390 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
391                              loff_t offset, loff_t len)
392 {
393         struct fuse_inode *fi;
394         struct inode *inode;
395         pgoff_t pg_start;
396         pgoff_t pg_end;
397
398         inode = fuse_ilookup(fc, nodeid, NULL);
399         if (!inode)
400                 return -ENOENT;
401
402         fi = get_fuse_inode(inode);
403         spin_lock(&fi->lock);
404         fi->attr_version = atomic64_inc_return(&fc->attr_version);
405         spin_unlock(&fi->lock);
406
407         fuse_invalidate_attr(inode);
408         forget_all_cached_acls(inode);
409         if (offset >= 0) {
410                 pg_start = offset >> PAGE_SHIFT;
411                 if (len <= 0)
412                         pg_end = -1;
413                 else
414                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
415                 invalidate_inode_pages2_range(inode->i_mapping,
416                                               pg_start, pg_end);
417         }
418         iput(inode);
419         return 0;
420 }
421
422 bool fuse_lock_inode(struct inode *inode)
423 {
424         bool locked = false;
425
426         if (!get_fuse_conn(inode)->parallel_dirops) {
427                 mutex_lock(&get_fuse_inode(inode)->mutex);
428                 locked = true;
429         }
430
431         return locked;
432 }
433
434 void fuse_unlock_inode(struct inode *inode, bool locked)
435 {
436         if (locked)
437                 mutex_unlock(&get_fuse_inode(inode)->mutex);
438 }
439
440 static void fuse_umount_begin(struct super_block *sb)
441 {
442         struct fuse_conn *fc = get_fuse_conn_super(sb);
443
444         if (!fc->no_force_umount)
445                 fuse_abort_conn(fc);
446 }
447
448 static void fuse_send_destroy(struct fuse_mount *fm)
449 {
450         if (fm->fc->conn_init) {
451                 FUSE_ARGS(args);
452
453                 args.opcode = FUSE_DESTROY;
454                 args.force = true;
455                 args.nocreds = true;
456                 fuse_simple_request(fm, &args);
457         }
458 }
459
460 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
461 {
462         stbuf->f_type    = FUSE_SUPER_MAGIC;
463         stbuf->f_bsize   = attr->bsize;
464         stbuf->f_frsize  = attr->frsize;
465         stbuf->f_blocks  = attr->blocks;
466         stbuf->f_bfree   = attr->bfree;
467         stbuf->f_bavail  = attr->bavail;
468         stbuf->f_files   = attr->files;
469         stbuf->f_ffree   = attr->ffree;
470         stbuf->f_namelen = attr->namelen;
471         /* fsid is left zero */
472 }
473
474 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
475 {
476         struct super_block *sb = dentry->d_sb;
477         struct fuse_mount *fm = get_fuse_mount_super(sb);
478         FUSE_ARGS(args);
479         struct fuse_statfs_out outarg;
480         int err;
481
482         if (!fuse_allow_current_process(fm->fc)) {
483                 buf->f_type = FUSE_SUPER_MAGIC;
484                 return 0;
485         }
486
487         memset(&outarg, 0, sizeof(outarg));
488         args.in_numargs = 0;
489         args.opcode = FUSE_STATFS;
490         args.nodeid = get_node_id(d_inode(dentry));
491         args.out_numargs = 1;
492         args.out_args[0].size = sizeof(outarg);
493         args.out_args[0].value = &outarg;
494         err = fuse_simple_request(fm, &args);
495         if (!err)
496                 convert_fuse_statfs(buf, &outarg.st);
497         return err;
498 }
499
500 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
501 {
502         struct fuse_sync_bucket *bucket;
503
504         bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
505         if (bucket) {
506                 init_waitqueue_head(&bucket->waitq);
507                 /* Initial active count */
508                 atomic_set(&bucket->count, 1);
509         }
510         return bucket;
511 }
512
513 static void fuse_sync_fs_writes(struct fuse_conn *fc)
514 {
515         struct fuse_sync_bucket *bucket, *new_bucket;
516         int count;
517
518         new_bucket = fuse_sync_bucket_alloc();
519         spin_lock(&fc->lock);
520         bucket = rcu_dereference_protected(fc->curr_bucket, 1);
521         count = atomic_read(&bucket->count);
522         WARN_ON(count < 1);
523         /* No outstanding writes? */
524         if (count == 1) {
525                 spin_unlock(&fc->lock);
526                 kfree(new_bucket);
527                 return;
528         }
529
530         /*
531          * Completion of new bucket depends on completion of this bucket, so add
532          * one more count.
533          */
534         atomic_inc(&new_bucket->count);
535         rcu_assign_pointer(fc->curr_bucket, new_bucket);
536         spin_unlock(&fc->lock);
537         /*
538          * Drop initial active count.  At this point if all writes in this and
539          * ancestor buckets complete, the count will go to zero and this task
540          * will be woken up.
541          */
542         atomic_dec(&bucket->count);
543
544         wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
545
546         /* Drop temp count on descendant bucket */
547         fuse_sync_bucket_dec(new_bucket);
548         kfree_rcu(bucket, rcu);
549 }
550
551 static int fuse_sync_fs(struct super_block *sb, int wait)
552 {
553         struct fuse_mount *fm = get_fuse_mount_super(sb);
554         struct fuse_conn *fc = fm->fc;
555         struct fuse_syncfs_in inarg;
556         FUSE_ARGS(args);
557         int err;
558
559         /*
560          * Userspace cannot handle the wait == 0 case.  Avoid a
561          * gratuitous roundtrip.
562          */
563         if (!wait)
564                 return 0;
565
566         /* The filesystem is being unmounted.  Nothing to do. */
567         if (!sb->s_root)
568                 return 0;
569
570         if (!fc->sync_fs)
571                 return 0;
572
573         fuse_sync_fs_writes(fc);
574
575         memset(&inarg, 0, sizeof(inarg));
576         args.in_numargs = 1;
577         args.in_args[0].size = sizeof(inarg);
578         args.in_args[0].value = &inarg;
579         args.opcode = FUSE_SYNCFS;
580         args.nodeid = get_node_id(sb->s_root->d_inode);
581         args.out_numargs = 0;
582
583         err = fuse_simple_request(fm, &args);
584         if (err == -ENOSYS) {
585                 fc->sync_fs = 0;
586                 err = 0;
587         }
588
589         return err;
590 }
591
592 enum {
593         OPT_SOURCE,
594         OPT_SUBTYPE,
595         OPT_FD,
596         OPT_ROOTMODE,
597         OPT_USER_ID,
598         OPT_GROUP_ID,
599         OPT_DEFAULT_PERMISSIONS,
600         OPT_ALLOW_OTHER,
601         OPT_MAX_READ,
602         OPT_BLKSIZE,
603         OPT_ERR
604 };
605
606 static const struct fs_parameter_spec fuse_fs_parameters[] = {
607         fsparam_string  ("source",              OPT_SOURCE),
608         fsparam_u32     ("fd",                  OPT_FD),
609         fsparam_u32oct  ("rootmode",            OPT_ROOTMODE),
610         fsparam_u32     ("user_id",             OPT_USER_ID),
611         fsparam_u32     ("group_id",            OPT_GROUP_ID),
612         fsparam_flag    ("default_permissions", OPT_DEFAULT_PERMISSIONS),
613         fsparam_flag    ("allow_other",         OPT_ALLOW_OTHER),
614         fsparam_u32     ("max_read",            OPT_MAX_READ),
615         fsparam_u32     ("blksize",             OPT_BLKSIZE),
616         fsparam_string  ("subtype",             OPT_SUBTYPE),
617         {}
618 };
619
620 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
621 {
622         struct fs_parse_result result;
623         struct fuse_fs_context *ctx = fsc->fs_private;
624         int opt;
625
626         if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
627                 /*
628                  * Ignore options coming from mount(MS_REMOUNT) for backward
629                  * compatibility.
630                  */
631                 if (fsc->oldapi)
632                         return 0;
633
634                 return invalfc(fsc, "No changes allowed in reconfigure");
635         }
636
637         opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
638         if (opt < 0)
639                 return opt;
640
641         switch (opt) {
642         case OPT_SOURCE:
643                 if (fsc->source)
644                         return invalfc(fsc, "Multiple sources specified");
645                 fsc->source = param->string;
646                 param->string = NULL;
647                 break;
648
649         case OPT_SUBTYPE:
650                 if (ctx->subtype)
651                         return invalfc(fsc, "Multiple subtypes specified");
652                 ctx->subtype = param->string;
653                 param->string = NULL;
654                 return 0;
655
656         case OPT_FD:
657                 ctx->fd = result.uint_32;
658                 ctx->fd_present = true;
659                 break;
660
661         case OPT_ROOTMODE:
662                 if (!fuse_valid_type(result.uint_32))
663                         return invalfc(fsc, "Invalid rootmode");
664                 ctx->rootmode = result.uint_32;
665                 ctx->rootmode_present = true;
666                 break;
667
668         case OPT_USER_ID:
669                 ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
670                 if (!uid_valid(ctx->user_id))
671                         return invalfc(fsc, "Invalid user_id");
672                 ctx->user_id_present = true;
673                 break;
674
675         case OPT_GROUP_ID:
676                 ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
677                 if (!gid_valid(ctx->group_id))
678                         return invalfc(fsc, "Invalid group_id");
679                 ctx->group_id_present = true;
680                 break;
681
682         case OPT_DEFAULT_PERMISSIONS:
683                 ctx->default_permissions = true;
684                 break;
685
686         case OPT_ALLOW_OTHER:
687                 ctx->allow_other = true;
688                 break;
689
690         case OPT_MAX_READ:
691                 ctx->max_read = result.uint_32;
692                 break;
693
694         case OPT_BLKSIZE:
695                 if (!ctx->is_bdev)
696                         return invalfc(fsc, "blksize only supported for fuseblk");
697                 ctx->blksize = result.uint_32;
698                 break;
699
700         default:
701                 return -EINVAL;
702         }
703
704         return 0;
705 }
706
707 static void fuse_free_fsc(struct fs_context *fsc)
708 {
709         struct fuse_fs_context *ctx = fsc->fs_private;
710
711         if (ctx) {
712                 kfree(ctx->subtype);
713                 kfree(ctx);
714         }
715 }
716
717 static int fuse_show_options(struct seq_file *m, struct dentry *root)
718 {
719         struct super_block *sb = root->d_sb;
720         struct fuse_conn *fc = get_fuse_conn_super(sb);
721
722         if (fc->legacy_opts_show) {
723                 seq_printf(m, ",user_id=%u",
724                            from_kuid_munged(fc->user_ns, fc->user_id));
725                 seq_printf(m, ",group_id=%u",
726                            from_kgid_munged(fc->user_ns, fc->group_id));
727                 if (fc->default_permissions)
728                         seq_puts(m, ",default_permissions");
729                 if (fc->allow_other)
730                         seq_puts(m, ",allow_other");
731                 if (fc->max_read != ~0)
732                         seq_printf(m, ",max_read=%u", fc->max_read);
733                 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
734                         seq_printf(m, ",blksize=%lu", sb->s_blocksize);
735         }
736 #ifdef CONFIG_FUSE_DAX
737         if (fc->dax)
738                 seq_puts(m, ",dax");
739 #endif
740
741         return 0;
742 }
743
744 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
745                              const struct fuse_iqueue_ops *ops,
746                              void *priv)
747 {
748         memset(fiq, 0, sizeof(struct fuse_iqueue));
749         spin_lock_init(&fiq->lock);
750         init_waitqueue_head(&fiq->waitq);
751         INIT_LIST_HEAD(&fiq->pending);
752         INIT_LIST_HEAD(&fiq->interrupts);
753         fiq->forget_list_tail = &fiq->forget_list_head;
754         fiq->connected = 1;
755         fiq->ops = ops;
756         fiq->priv = priv;
757 }
758
759 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
760 {
761         unsigned int i;
762
763         spin_lock_init(&fpq->lock);
764         for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
765                 INIT_LIST_HEAD(&fpq->processing[i]);
766         INIT_LIST_HEAD(&fpq->io);
767         fpq->connected = 1;
768 }
769
770 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
771                     struct user_namespace *user_ns,
772                     const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
773 {
774         memset(fc, 0, sizeof(*fc));
775         spin_lock_init(&fc->lock);
776         spin_lock_init(&fc->bg_lock);
777         init_rwsem(&fc->killsb);
778         refcount_set(&fc->count, 1);
779         atomic_set(&fc->dev_count, 1);
780         init_waitqueue_head(&fc->blocked_waitq);
781         fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
782         INIT_LIST_HEAD(&fc->bg_queue);
783         INIT_LIST_HEAD(&fc->entry);
784         INIT_LIST_HEAD(&fc->devices);
785         atomic_set(&fc->num_waiting, 0);
786         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
787         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
788         atomic64_set(&fc->khctr, 0);
789         fc->polled_files = RB_ROOT;
790         fc->blocked = 0;
791         fc->initialized = 0;
792         fc->connected = 1;
793         atomic64_set(&fc->attr_version, 1);
794         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
795         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
796         fc->user_ns = get_user_ns(user_ns);
797         fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
798         fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
799
800         INIT_LIST_HEAD(&fc->mounts);
801         list_add(&fm->fc_entry, &fc->mounts);
802         fm->fc = fc;
803 }
804 EXPORT_SYMBOL_GPL(fuse_conn_init);
805
806 void fuse_conn_put(struct fuse_conn *fc)
807 {
808         if (refcount_dec_and_test(&fc->count)) {
809                 struct fuse_iqueue *fiq = &fc->iq;
810                 struct fuse_sync_bucket *bucket;
811
812                 if (IS_ENABLED(CONFIG_FUSE_DAX))
813                         fuse_dax_conn_free(fc);
814                 if (fiq->ops->release)
815                         fiq->ops->release(fiq);
816                 put_pid_ns(fc->pid_ns);
817                 put_user_ns(fc->user_ns);
818                 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
819                 if (bucket) {
820                         WARN_ON(atomic_read(&bucket->count) != 1);
821                         kfree(bucket);
822                 }
823                 fc->release(fc);
824         }
825 }
826 EXPORT_SYMBOL_GPL(fuse_conn_put);
827
828 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
829 {
830         refcount_inc(&fc->count);
831         return fc;
832 }
833 EXPORT_SYMBOL_GPL(fuse_conn_get);
834
835 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
836 {
837         struct fuse_attr attr;
838         memset(&attr, 0, sizeof(attr));
839
840         attr.mode = mode;
841         attr.ino = FUSE_ROOT_ID;
842         attr.nlink = 1;
843         return fuse_iget(sb, 1, 0, &attr, 0, 0);
844 }
845
846 struct fuse_inode_handle {
847         u64 nodeid;
848         u32 generation;
849 };
850
851 static struct dentry *fuse_get_dentry(struct super_block *sb,
852                                       struct fuse_inode_handle *handle)
853 {
854         struct fuse_conn *fc = get_fuse_conn_super(sb);
855         struct inode *inode;
856         struct dentry *entry;
857         int err = -ESTALE;
858
859         if (handle->nodeid == 0)
860                 goto out_err;
861
862         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
863         if (!inode) {
864                 struct fuse_entry_out outarg;
865                 const struct qstr name = QSTR_INIT(".", 1);
866
867                 if (!fc->export_support)
868                         goto out_err;
869
870                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
871                                        &inode);
872                 if (err && err != -ENOENT)
873                         goto out_err;
874                 if (err || !inode) {
875                         err = -ESTALE;
876                         goto out_err;
877                 }
878                 err = -EIO;
879                 if (get_node_id(inode) != handle->nodeid)
880                         goto out_iput;
881         }
882         err = -ESTALE;
883         if (inode->i_generation != handle->generation)
884                 goto out_iput;
885
886         entry = d_obtain_alias(inode);
887         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
888                 fuse_invalidate_entry_cache(entry);
889
890         return entry;
891
892  out_iput:
893         iput(inode);
894  out_err:
895         return ERR_PTR(err);
896 }
897
898 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
899                            struct inode *parent)
900 {
901         int len = parent ? 6 : 3;
902         u64 nodeid;
903         u32 generation;
904
905         if (*max_len < len) {
906                 *max_len = len;
907                 return  FILEID_INVALID;
908         }
909
910         nodeid = get_fuse_inode(inode)->nodeid;
911         generation = inode->i_generation;
912
913         fh[0] = (u32)(nodeid >> 32);
914         fh[1] = (u32)(nodeid & 0xffffffff);
915         fh[2] = generation;
916
917         if (parent) {
918                 nodeid = get_fuse_inode(parent)->nodeid;
919                 generation = parent->i_generation;
920
921                 fh[3] = (u32)(nodeid >> 32);
922                 fh[4] = (u32)(nodeid & 0xffffffff);
923                 fh[5] = generation;
924         }
925
926         *max_len = len;
927         return parent ? 0x82 : 0x81;
928 }
929
930 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
931                 struct fid *fid, int fh_len, int fh_type)
932 {
933         struct fuse_inode_handle handle;
934
935         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
936                 return NULL;
937
938         handle.nodeid = (u64) fid->raw[0] << 32;
939         handle.nodeid |= (u64) fid->raw[1];
940         handle.generation = fid->raw[2];
941         return fuse_get_dentry(sb, &handle);
942 }
943
944 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
945                 struct fid *fid, int fh_len, int fh_type)
946 {
947         struct fuse_inode_handle parent;
948
949         if (fh_type != 0x82 || fh_len < 6)
950                 return NULL;
951
952         parent.nodeid = (u64) fid->raw[3] << 32;
953         parent.nodeid |= (u64) fid->raw[4];
954         parent.generation = fid->raw[5];
955         return fuse_get_dentry(sb, &parent);
956 }
957
958 static struct dentry *fuse_get_parent(struct dentry *child)
959 {
960         struct inode *child_inode = d_inode(child);
961         struct fuse_conn *fc = get_fuse_conn(child_inode);
962         struct inode *inode;
963         struct dentry *parent;
964         struct fuse_entry_out outarg;
965         int err;
966
967         if (!fc->export_support)
968                 return ERR_PTR(-ESTALE);
969
970         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
971                                &dotdot_name, &outarg, &inode);
972         if (err) {
973                 if (err == -ENOENT)
974                         return ERR_PTR(-ESTALE);
975                 return ERR_PTR(err);
976         }
977
978         parent = d_obtain_alias(inode);
979         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
980                 fuse_invalidate_entry_cache(parent);
981
982         return parent;
983 }
984
985 static const struct export_operations fuse_export_operations = {
986         .fh_to_dentry   = fuse_fh_to_dentry,
987         .fh_to_parent   = fuse_fh_to_parent,
988         .encode_fh      = fuse_encode_fh,
989         .get_parent     = fuse_get_parent,
990 };
991
992 static const struct super_operations fuse_super_operations = {
993         .alloc_inode    = fuse_alloc_inode,
994         .free_inode     = fuse_free_inode,
995         .evict_inode    = fuse_evict_inode,
996         .write_inode    = fuse_write_inode,
997         .drop_inode     = generic_delete_inode,
998         .umount_begin   = fuse_umount_begin,
999         .statfs         = fuse_statfs,
1000         .sync_fs        = fuse_sync_fs,
1001         .show_options   = fuse_show_options,
1002 };
1003
1004 static void sanitize_global_limit(unsigned *limit)
1005 {
1006         /*
1007          * The default maximum number of async requests is calculated to consume
1008          * 1/2^13 of the total memory, assuming 392 bytes per request.
1009          */
1010         if (*limit == 0)
1011                 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1012
1013         if (*limit >= 1 << 16)
1014                 *limit = (1 << 16) - 1;
1015 }
1016
1017 static int set_global_limit(const char *val, const struct kernel_param *kp)
1018 {
1019         int rv;
1020
1021         rv = param_set_uint(val, kp);
1022         if (rv)
1023                 return rv;
1024
1025         sanitize_global_limit((unsigned *)kp->arg);
1026
1027         return 0;
1028 }
1029
1030 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1031 {
1032         int cap_sys_admin = capable(CAP_SYS_ADMIN);
1033
1034         if (arg->minor < 13)
1035                 return;
1036
1037         sanitize_global_limit(&max_user_bgreq);
1038         sanitize_global_limit(&max_user_congthresh);
1039
1040         spin_lock(&fc->bg_lock);
1041         if (arg->max_background) {
1042                 fc->max_background = arg->max_background;
1043
1044                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1045                         fc->max_background = max_user_bgreq;
1046         }
1047         if (arg->congestion_threshold) {
1048                 fc->congestion_threshold = arg->congestion_threshold;
1049
1050                 if (!cap_sys_admin &&
1051                     fc->congestion_threshold > max_user_congthresh)
1052                         fc->congestion_threshold = max_user_congthresh;
1053         }
1054         spin_unlock(&fc->bg_lock);
1055 }
1056
1057 struct fuse_init_args {
1058         struct fuse_args args;
1059         struct fuse_init_in in;
1060         struct fuse_init_out out;
1061 };
1062
1063 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1064                                int error)
1065 {
1066         struct fuse_conn *fc = fm->fc;
1067         struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1068         struct fuse_init_out *arg = &ia->out;
1069         bool ok = true;
1070
1071         if (error || arg->major != FUSE_KERNEL_VERSION)
1072                 ok = false;
1073         else {
1074                 unsigned long ra_pages;
1075
1076                 process_init_limits(fc, arg);
1077
1078                 if (arg->minor >= 6) {
1079                         ra_pages = arg->max_readahead / PAGE_SIZE;
1080                         if (arg->flags & FUSE_ASYNC_READ)
1081                                 fc->async_read = 1;
1082                         if (!(arg->flags & FUSE_POSIX_LOCKS))
1083                                 fc->no_lock = 1;
1084                         if (arg->minor >= 17) {
1085                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
1086                                         fc->no_flock = 1;
1087                         } else {
1088                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
1089                                         fc->no_flock = 1;
1090                         }
1091                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
1092                                 fc->atomic_o_trunc = 1;
1093                         if (arg->minor >= 9) {
1094                                 /* LOOKUP has dependency on proto version */
1095                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
1096                                         fc->export_support = 1;
1097                         }
1098                         if (arg->flags & FUSE_BIG_WRITES)
1099                                 fc->big_writes = 1;
1100                         if (arg->flags & FUSE_DONT_MASK)
1101                                 fc->dont_mask = 1;
1102                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
1103                                 fc->auto_inval_data = 1;
1104                         else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
1105                                 fc->explicit_inval_data = 1;
1106                         if (arg->flags & FUSE_DO_READDIRPLUS) {
1107                                 fc->do_readdirplus = 1;
1108                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
1109                                         fc->readdirplus_auto = 1;
1110                         }
1111                         if (arg->flags & FUSE_ASYNC_DIO)
1112                                 fc->async_dio = 1;
1113                         if (arg->flags & FUSE_WRITEBACK_CACHE)
1114                                 fc->writeback_cache = 1;
1115                         if (arg->flags & FUSE_PARALLEL_DIROPS)
1116                                 fc->parallel_dirops = 1;
1117                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
1118                                 fc->handle_killpriv = 1;
1119                         if (arg->time_gran && arg->time_gran <= 1000000000)
1120                                 fm->sb->s_time_gran = arg->time_gran;
1121                         if ((arg->flags & FUSE_POSIX_ACL)) {
1122                                 fc->default_permissions = 1;
1123                                 fc->posix_acl = 1;
1124                                 fm->sb->s_xattr = fuse_acl_xattr_handlers;
1125                         }
1126                         if (arg->flags & FUSE_CACHE_SYMLINKS)
1127                                 fc->cache_symlinks = 1;
1128                         if (arg->flags & FUSE_ABORT_ERROR)
1129                                 fc->abort_err = 1;
1130                         if (arg->flags & FUSE_MAX_PAGES) {
1131                                 fc->max_pages =
1132                                         min_t(unsigned int, fc->max_pages_limit,
1133                                         max_t(unsigned int, arg->max_pages, 1));
1134                         }
1135                         if (IS_ENABLED(CONFIG_FUSE_DAX) &&
1136                             arg->flags & FUSE_MAP_ALIGNMENT &&
1137                             !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1138                                 ok = false;
1139                         }
1140                         if (arg->flags & FUSE_HANDLE_KILLPRIV_V2) {
1141                                 fc->handle_killpriv_v2 = 1;
1142                                 fm->sb->s_flags |= SB_NOSEC;
1143                         }
1144                         if (arg->flags & FUSE_SETXATTR_EXT)
1145                                 fc->setxattr_ext = 1;
1146                 } else {
1147                         ra_pages = fc->max_read / PAGE_SIZE;
1148                         fc->no_lock = 1;
1149                         fc->no_flock = 1;
1150                 }
1151
1152                 fm->sb->s_bdi->ra_pages =
1153                                 min(fm->sb->s_bdi->ra_pages, ra_pages);
1154                 fc->minor = arg->minor;
1155                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1156                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1157                 fc->conn_init = 1;
1158         }
1159         kfree(ia);
1160
1161         if (!ok) {
1162                 fc->conn_init = 0;
1163                 fc->conn_error = 1;
1164         }
1165
1166         fuse_set_initialized(fc);
1167         wake_up_all(&fc->blocked_waitq);
1168 }
1169
1170 void fuse_send_init(struct fuse_mount *fm)
1171 {
1172         struct fuse_init_args *ia;
1173
1174         ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1175
1176         ia->in.major = FUSE_KERNEL_VERSION;
1177         ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1178         ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1179         ia->in.flags |=
1180                 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1181                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1182                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1183                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1184                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1185                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1186                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1187                 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1188                 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1189                 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT;
1190 #ifdef CONFIG_FUSE_DAX
1191         if (fm->fc->dax)
1192                 ia->in.flags |= FUSE_MAP_ALIGNMENT;
1193 #endif
1194         if (fm->fc->auto_submounts)
1195                 ia->in.flags |= FUSE_SUBMOUNTS;
1196
1197         ia->args.opcode = FUSE_INIT;
1198         ia->args.in_numargs = 1;
1199         ia->args.in_args[0].size = sizeof(ia->in);
1200         ia->args.in_args[0].value = &ia->in;
1201         ia->args.out_numargs = 1;
1202         /* Variable length argument used for backward compatibility
1203            with interface version < 7.5.  Rest of init_out is zeroed
1204            by do_get_request(), so a short reply is not a problem */
1205         ia->args.out_argvar = true;
1206         ia->args.out_args[0].size = sizeof(ia->out);
1207         ia->args.out_args[0].value = &ia->out;
1208         ia->args.force = true;
1209         ia->args.nocreds = true;
1210         ia->args.end = process_init_reply;
1211
1212         if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1213                 process_init_reply(fm, &ia->args, -ENOTCONN);
1214 }
1215 EXPORT_SYMBOL_GPL(fuse_send_init);
1216
1217 void fuse_free_conn(struct fuse_conn *fc)
1218 {
1219         WARN_ON(!list_empty(&fc->devices));
1220         kfree_rcu(fc, rcu);
1221 }
1222 EXPORT_SYMBOL_GPL(fuse_free_conn);
1223
1224 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1225 {
1226         int err;
1227         char *suffix = "";
1228
1229         if (sb->s_bdev) {
1230                 suffix = "-fuseblk";
1231                 /*
1232                  * sb->s_bdi points to blkdev's bdi however we want to redirect
1233                  * it to our private bdi...
1234                  */
1235                 bdi_put(sb->s_bdi);
1236                 sb->s_bdi = &noop_backing_dev_info;
1237         }
1238         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1239                                    MINOR(fc->dev), suffix);
1240         if (err)
1241                 return err;
1242
1243         /* fuse does it's own writeback accounting */
1244         sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1245         sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1246
1247         /*
1248          * For a single fuse filesystem use max 1% of dirty +
1249          * writeback threshold.
1250          *
1251          * This gives about 1M of write buffer for memory maps on a
1252          * machine with 1G and 10% dirty_ratio, which should be more
1253          * than enough.
1254          *
1255          * Privileged users can raise it by writing to
1256          *
1257          *    /sys/class/bdi/<bdi>/max_ratio
1258          */
1259         bdi_set_max_ratio(sb->s_bdi, 1);
1260
1261         return 0;
1262 }
1263
1264 struct fuse_dev *fuse_dev_alloc(void)
1265 {
1266         struct fuse_dev *fud;
1267         struct list_head *pq;
1268
1269         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1270         if (!fud)
1271                 return NULL;
1272
1273         pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1274         if (!pq) {
1275                 kfree(fud);
1276                 return NULL;
1277         }
1278
1279         fud->pq.processing = pq;
1280         fuse_pqueue_init(&fud->pq);
1281
1282         return fud;
1283 }
1284 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1285
1286 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1287 {
1288         fud->fc = fuse_conn_get(fc);
1289         spin_lock(&fc->lock);
1290         list_add_tail(&fud->entry, &fc->devices);
1291         spin_unlock(&fc->lock);
1292 }
1293 EXPORT_SYMBOL_GPL(fuse_dev_install);
1294
1295 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1296 {
1297         struct fuse_dev *fud;
1298
1299         fud = fuse_dev_alloc();
1300         if (!fud)
1301                 return NULL;
1302
1303         fuse_dev_install(fud, fc);
1304         return fud;
1305 }
1306 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1307
1308 void fuse_dev_free(struct fuse_dev *fud)
1309 {
1310         struct fuse_conn *fc = fud->fc;
1311
1312         if (fc) {
1313                 spin_lock(&fc->lock);
1314                 list_del(&fud->entry);
1315                 spin_unlock(&fc->lock);
1316
1317                 fuse_conn_put(fc);
1318         }
1319         kfree(fud->pq.processing);
1320         kfree(fud);
1321 }
1322 EXPORT_SYMBOL_GPL(fuse_dev_free);
1323
1324 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1325                                       const struct fuse_inode *fi)
1326 {
1327         *attr = (struct fuse_attr){
1328                 .ino            = fi->inode.i_ino,
1329                 .size           = fi->inode.i_size,
1330                 .blocks         = fi->inode.i_blocks,
1331                 .atime          = fi->inode.i_atime.tv_sec,
1332                 .mtime          = fi->inode.i_mtime.tv_sec,
1333                 .ctime          = fi->inode.i_ctime.tv_sec,
1334                 .atimensec      = fi->inode.i_atime.tv_nsec,
1335                 .mtimensec      = fi->inode.i_mtime.tv_nsec,
1336                 .ctimensec      = fi->inode.i_ctime.tv_nsec,
1337                 .mode           = fi->inode.i_mode,
1338                 .nlink          = fi->inode.i_nlink,
1339                 .uid            = fi->inode.i_uid.val,
1340                 .gid            = fi->inode.i_gid.val,
1341                 .rdev           = fi->inode.i_rdev,
1342                 .blksize        = 1u << fi->inode.i_blkbits,
1343         };
1344 }
1345
1346 static void fuse_sb_defaults(struct super_block *sb)
1347 {
1348         sb->s_magic = FUSE_SUPER_MAGIC;
1349         sb->s_op = &fuse_super_operations;
1350         sb->s_xattr = fuse_xattr_handlers;
1351         sb->s_maxbytes = MAX_LFS_FILESIZE;
1352         sb->s_time_gran = 1;
1353         sb->s_export_op = &fuse_export_operations;
1354         sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1355         if (sb->s_user_ns != &init_user_ns)
1356                 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1357         sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1358
1359         /*
1360          * If we are not in the initial user namespace posix
1361          * acls must be translated.
1362          */
1363         if (sb->s_user_ns != &init_user_ns)
1364                 sb->s_xattr = fuse_no_acl_xattr_handlers;
1365 }
1366
1367 static int fuse_fill_super_submount(struct super_block *sb,
1368                                     struct fuse_inode *parent_fi)
1369 {
1370         struct fuse_mount *fm = get_fuse_mount_super(sb);
1371         struct super_block *parent_sb = parent_fi->inode.i_sb;
1372         struct fuse_attr root_attr;
1373         struct inode *root;
1374
1375         fuse_sb_defaults(sb);
1376         fm->sb = sb;
1377
1378         WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1379         sb->s_bdi = bdi_get(parent_sb->s_bdi);
1380
1381         sb->s_xattr = parent_sb->s_xattr;
1382         sb->s_time_gran = parent_sb->s_time_gran;
1383         sb->s_blocksize = parent_sb->s_blocksize;
1384         sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1385         sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1386         if (parent_sb->s_subtype && !sb->s_subtype)
1387                 return -ENOMEM;
1388
1389         fuse_fill_attr_from_inode(&root_attr, parent_fi);
1390         root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1391         /*
1392          * This inode is just a duplicate, so it is not looked up and
1393          * its nlookup should not be incremented.  fuse_iget() does
1394          * that, though, so undo it here.
1395          */
1396         get_fuse_inode(root)->nlookup--;
1397         sb->s_d_op = &fuse_dentry_operations;
1398         sb->s_root = d_make_root(root);
1399         if (!sb->s_root)
1400                 return -ENOMEM;
1401
1402         return 0;
1403 }
1404
1405 /* Filesystem context private data holds the FUSE inode of the mount point */
1406 static int fuse_get_tree_submount(struct fs_context *fsc)
1407 {
1408         struct fuse_mount *fm;
1409         struct fuse_inode *mp_fi = fsc->fs_private;
1410         struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1411         struct super_block *sb;
1412         int err;
1413
1414         fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1415         if (!fm)
1416                 return -ENOMEM;
1417
1418         fm->fc = fuse_conn_get(fc);
1419         fsc->s_fs_info = fm;
1420         sb = sget_fc(fsc, NULL, set_anon_super_fc);
1421         if (fsc->s_fs_info)
1422                 fuse_mount_destroy(fm);
1423         if (IS_ERR(sb))
1424                 return PTR_ERR(sb);
1425
1426         /* Initialize superblock, making @mp_fi its root */
1427         err = fuse_fill_super_submount(sb, mp_fi);
1428         if (err) {
1429                 deactivate_locked_super(sb);
1430                 return err;
1431         }
1432
1433         down_write(&fc->killsb);
1434         list_add_tail(&fm->fc_entry, &fc->mounts);
1435         up_write(&fc->killsb);
1436
1437         sb->s_flags |= SB_ACTIVE;
1438         fsc->root = dget(sb->s_root);
1439
1440         return 0;
1441 }
1442
1443 static const struct fs_context_operations fuse_context_submount_ops = {
1444         .get_tree       = fuse_get_tree_submount,
1445 };
1446
1447 int fuse_init_fs_context_submount(struct fs_context *fsc)
1448 {
1449         fsc->ops = &fuse_context_submount_ops;
1450         return 0;
1451 }
1452 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1453
1454 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1455 {
1456         struct fuse_dev *fud = NULL;
1457         struct fuse_mount *fm = get_fuse_mount_super(sb);
1458         struct fuse_conn *fc = fm->fc;
1459         struct inode *root;
1460         struct dentry *root_dentry;
1461         int err;
1462
1463         err = -EINVAL;
1464         if (sb->s_flags & SB_MANDLOCK)
1465                 goto err;
1466
1467         rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1468         fuse_sb_defaults(sb);
1469
1470         if (ctx->is_bdev) {
1471 #ifdef CONFIG_BLOCK
1472                 err = -EINVAL;
1473                 if (!sb_set_blocksize(sb, ctx->blksize))
1474                         goto err;
1475 #endif
1476         } else {
1477                 sb->s_blocksize = PAGE_SIZE;
1478                 sb->s_blocksize_bits = PAGE_SHIFT;
1479         }
1480
1481         sb->s_subtype = ctx->subtype;
1482         ctx->subtype = NULL;
1483         if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1484                 err = fuse_dax_conn_alloc(fc, ctx->dax_dev);
1485                 if (err)
1486                         goto err;
1487         }
1488
1489         if (ctx->fudptr) {
1490                 err = -ENOMEM;
1491                 fud = fuse_dev_alloc_install(fc);
1492                 if (!fud)
1493                         goto err_free_dax;
1494         }
1495
1496         fc->dev = sb->s_dev;
1497         fm->sb = sb;
1498         err = fuse_bdi_init(fc, sb);
1499         if (err)
1500                 goto err_dev_free;
1501
1502         /* Handle umasking inside the fuse code */
1503         if (sb->s_flags & SB_POSIXACL)
1504                 fc->dont_mask = 1;
1505         sb->s_flags |= SB_POSIXACL;
1506
1507         fc->default_permissions = ctx->default_permissions;
1508         fc->allow_other = ctx->allow_other;
1509         fc->user_id = ctx->user_id;
1510         fc->group_id = ctx->group_id;
1511         fc->legacy_opts_show = ctx->legacy_opts_show;
1512         fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1513         fc->destroy = ctx->destroy;
1514         fc->no_control = ctx->no_control;
1515         fc->no_force_umount = ctx->no_force_umount;
1516
1517         err = -ENOMEM;
1518         root = fuse_get_root_inode(sb, ctx->rootmode);
1519         sb->s_d_op = &fuse_root_dentry_operations;
1520         root_dentry = d_make_root(root);
1521         if (!root_dentry)
1522                 goto err_dev_free;
1523         /* Root dentry doesn't have .d_revalidate */
1524         sb->s_d_op = &fuse_dentry_operations;
1525
1526         mutex_lock(&fuse_mutex);
1527         err = -EINVAL;
1528         if (ctx->fudptr && *ctx->fudptr)
1529                 goto err_unlock;
1530
1531         err = fuse_ctl_add_conn(fc);
1532         if (err)
1533                 goto err_unlock;
1534
1535         list_add_tail(&fc->entry, &fuse_conn_list);
1536         sb->s_root = root_dentry;
1537         if (ctx->fudptr)
1538                 *ctx->fudptr = fud;
1539         mutex_unlock(&fuse_mutex);
1540         return 0;
1541
1542  err_unlock:
1543         mutex_unlock(&fuse_mutex);
1544         dput(root_dentry);
1545  err_dev_free:
1546         if (fud)
1547                 fuse_dev_free(fud);
1548  err_free_dax:
1549         if (IS_ENABLED(CONFIG_FUSE_DAX))
1550                 fuse_dax_conn_free(fc);
1551  err:
1552         return err;
1553 }
1554 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1555
1556 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1557 {
1558         struct fuse_fs_context *ctx = fsc->fs_private;
1559         int err;
1560
1561         if (!ctx->file || !ctx->rootmode_present ||
1562             !ctx->user_id_present || !ctx->group_id_present)
1563                 return -EINVAL;
1564
1565         /*
1566          * Require mount to happen from the same user namespace which
1567          * opened /dev/fuse to prevent potential attacks.
1568          */
1569         if ((ctx->file->f_op != &fuse_dev_operations) ||
1570             (ctx->file->f_cred->user_ns != sb->s_user_ns))
1571                 return -EINVAL;
1572         ctx->fudptr = &ctx->file->private_data;
1573
1574         err = fuse_fill_super_common(sb, ctx);
1575         if (err)
1576                 return err;
1577         /* file->private_data shall be visible on all CPUs after this */
1578         smp_mb();
1579         fuse_send_init(get_fuse_mount_super(sb));
1580         return 0;
1581 }
1582
1583 /*
1584  * This is the path where user supplied an already initialized fuse dev.  In
1585  * this case never create a new super if the old one is gone.
1586  */
1587 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1588 {
1589         return -ENOTCONN;
1590 }
1591
1592 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1593 {
1594
1595         return fsc->sget_key == get_fuse_conn_super(sb);
1596 }
1597
1598 static int fuse_get_tree(struct fs_context *fsc)
1599 {
1600         struct fuse_fs_context *ctx = fsc->fs_private;
1601         struct fuse_dev *fud;
1602         struct fuse_conn *fc;
1603         struct fuse_mount *fm;
1604         struct super_block *sb;
1605         int err;
1606
1607         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1608         if (!fc)
1609                 return -ENOMEM;
1610
1611         fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1612         if (!fm) {
1613                 kfree(fc);
1614                 return -ENOMEM;
1615         }
1616
1617         fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1618         fc->release = fuse_free_conn;
1619
1620         fsc->s_fs_info = fm;
1621
1622         if (ctx->fd_present)
1623                 ctx->file = fget(ctx->fd);
1624
1625         if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1626                 err = get_tree_bdev(fsc, fuse_fill_super);
1627                 goto out;
1628         }
1629         /*
1630          * While block dev mount can be initialized with a dummy device fd
1631          * (found by device name), normal fuse mounts can't
1632          */
1633         err = -EINVAL;
1634         if (!ctx->file)
1635                 goto out;
1636
1637         /*
1638          * Allow creating a fuse mount with an already initialized fuse
1639          * connection
1640          */
1641         fud = READ_ONCE(ctx->file->private_data);
1642         if (ctx->file->f_op == &fuse_dev_operations && fud) {
1643                 fsc->sget_key = fud->fc;
1644                 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1645                 err = PTR_ERR_OR_ZERO(sb);
1646                 if (!IS_ERR(sb))
1647                         fsc->root = dget(sb->s_root);
1648         } else {
1649                 err = get_tree_nodev(fsc, fuse_fill_super);
1650         }
1651 out:
1652         if (fsc->s_fs_info)
1653                 fuse_mount_destroy(fm);
1654         if (ctx->file)
1655                 fput(ctx->file);
1656         return err;
1657 }
1658
1659 static const struct fs_context_operations fuse_context_ops = {
1660         .free           = fuse_free_fsc,
1661         .parse_param    = fuse_parse_param,
1662         .reconfigure    = fuse_reconfigure,
1663         .get_tree       = fuse_get_tree,
1664 };
1665
1666 /*
1667  * Set up the filesystem mount context.
1668  */
1669 static int fuse_init_fs_context(struct fs_context *fsc)
1670 {
1671         struct fuse_fs_context *ctx;
1672
1673         ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1674         if (!ctx)
1675                 return -ENOMEM;
1676
1677         ctx->max_read = ~0;
1678         ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1679         ctx->legacy_opts_show = true;
1680
1681 #ifdef CONFIG_BLOCK
1682         if (fsc->fs_type == &fuseblk_fs_type) {
1683                 ctx->is_bdev = true;
1684                 ctx->destroy = true;
1685         }
1686 #endif
1687
1688         fsc->fs_private = ctx;
1689         fsc->ops = &fuse_context_ops;
1690         return 0;
1691 }
1692
1693 bool fuse_mount_remove(struct fuse_mount *fm)
1694 {
1695         struct fuse_conn *fc = fm->fc;
1696         bool last = false;
1697
1698         down_write(&fc->killsb);
1699         list_del_init(&fm->fc_entry);
1700         if (list_empty(&fc->mounts))
1701                 last = true;
1702         up_write(&fc->killsb);
1703
1704         return last;
1705 }
1706 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1707
1708 void fuse_conn_destroy(struct fuse_mount *fm)
1709 {
1710         struct fuse_conn *fc = fm->fc;
1711
1712         if (fc->destroy)
1713                 fuse_send_destroy(fm);
1714
1715         fuse_abort_conn(fc);
1716         fuse_wait_aborted(fc);
1717
1718         if (!list_empty(&fc->entry)) {
1719                 mutex_lock(&fuse_mutex);
1720                 list_del(&fc->entry);
1721                 fuse_ctl_remove_conn(fc);
1722                 mutex_unlock(&fuse_mutex);
1723         }
1724 }
1725 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1726
1727 static void fuse_sb_destroy(struct super_block *sb)
1728 {
1729         struct fuse_mount *fm = get_fuse_mount_super(sb);
1730         bool last;
1731
1732         if (sb->s_root) {
1733                 last = fuse_mount_remove(fm);
1734                 if (last)
1735                         fuse_conn_destroy(fm);
1736         }
1737 }
1738
1739 void fuse_mount_destroy(struct fuse_mount *fm)
1740 {
1741         fuse_conn_put(fm->fc);
1742         kfree(fm);
1743 }
1744 EXPORT_SYMBOL(fuse_mount_destroy);
1745
1746 static void fuse_kill_sb_anon(struct super_block *sb)
1747 {
1748         fuse_sb_destroy(sb);
1749         kill_anon_super(sb);
1750         fuse_mount_destroy(get_fuse_mount_super(sb));
1751 }
1752
1753 static struct file_system_type fuse_fs_type = {
1754         .owner          = THIS_MODULE,
1755         .name           = "fuse",
1756         .fs_flags       = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1757         .init_fs_context = fuse_init_fs_context,
1758         .parameters     = fuse_fs_parameters,
1759         .kill_sb        = fuse_kill_sb_anon,
1760 };
1761 MODULE_ALIAS_FS("fuse");
1762
1763 #ifdef CONFIG_BLOCK
1764 static void fuse_kill_sb_blk(struct super_block *sb)
1765 {
1766         fuse_sb_destroy(sb);
1767         kill_block_super(sb);
1768         fuse_mount_destroy(get_fuse_mount_super(sb));
1769 }
1770
1771 static struct file_system_type fuseblk_fs_type = {
1772         .owner          = THIS_MODULE,
1773         .name           = "fuseblk",
1774         .init_fs_context = fuse_init_fs_context,
1775         .parameters     = fuse_fs_parameters,
1776         .kill_sb        = fuse_kill_sb_blk,
1777         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1778 };
1779 MODULE_ALIAS_FS("fuseblk");
1780
1781 static inline int register_fuseblk(void)
1782 {
1783         return register_filesystem(&fuseblk_fs_type);
1784 }
1785
1786 static inline void unregister_fuseblk(void)
1787 {
1788         unregister_filesystem(&fuseblk_fs_type);
1789 }
1790 #else
1791 static inline int register_fuseblk(void)
1792 {
1793         return 0;
1794 }
1795
1796 static inline void unregister_fuseblk(void)
1797 {
1798 }
1799 #endif
1800
1801 static void fuse_inode_init_once(void *foo)
1802 {
1803         struct inode *inode = foo;
1804
1805         inode_init_once(inode);
1806 }
1807
1808 static int __init fuse_fs_init(void)
1809 {
1810         int err;
1811
1812         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1813                         sizeof(struct fuse_inode), 0,
1814                         SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1815                         fuse_inode_init_once);
1816         err = -ENOMEM;
1817         if (!fuse_inode_cachep)
1818                 goto out;
1819
1820         err = register_fuseblk();
1821         if (err)
1822                 goto out2;
1823
1824         err = register_filesystem(&fuse_fs_type);
1825         if (err)
1826                 goto out3;
1827
1828         return 0;
1829
1830  out3:
1831         unregister_fuseblk();
1832  out2:
1833         kmem_cache_destroy(fuse_inode_cachep);
1834  out:
1835         return err;
1836 }
1837
1838 static void fuse_fs_cleanup(void)
1839 {
1840         unregister_filesystem(&fuse_fs_type);
1841         unregister_fuseblk();
1842
1843         /*
1844          * Make sure all delayed rcu free inodes are flushed before we
1845          * destroy cache.
1846          */
1847         rcu_barrier();
1848         kmem_cache_destroy(fuse_inode_cachep);
1849 }
1850
1851 static struct kobject *fuse_kobj;
1852
1853 static int fuse_sysfs_init(void)
1854 {
1855         int err;
1856
1857         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1858         if (!fuse_kobj) {
1859                 err = -ENOMEM;
1860                 goto out_err;
1861         }
1862
1863         err = sysfs_create_mount_point(fuse_kobj, "connections");
1864         if (err)
1865                 goto out_fuse_unregister;
1866
1867         return 0;
1868
1869  out_fuse_unregister:
1870         kobject_put(fuse_kobj);
1871  out_err:
1872         return err;
1873 }
1874
1875 static void fuse_sysfs_cleanup(void)
1876 {
1877         sysfs_remove_mount_point(fuse_kobj, "connections");
1878         kobject_put(fuse_kobj);
1879 }
1880
1881 static int __init fuse_init(void)
1882 {
1883         int res;
1884
1885         pr_info("init (API version %i.%i)\n",
1886                 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1887
1888         INIT_LIST_HEAD(&fuse_conn_list);
1889         res = fuse_fs_init();
1890         if (res)
1891                 goto err;
1892
1893         res = fuse_dev_init();
1894         if (res)
1895                 goto err_fs_cleanup;
1896
1897         res = fuse_sysfs_init();
1898         if (res)
1899                 goto err_dev_cleanup;
1900
1901         res = fuse_ctl_init();
1902         if (res)
1903                 goto err_sysfs_cleanup;
1904
1905         sanitize_global_limit(&max_user_bgreq);
1906         sanitize_global_limit(&max_user_congthresh);
1907
1908         return 0;
1909
1910  err_sysfs_cleanup:
1911         fuse_sysfs_cleanup();
1912  err_dev_cleanup:
1913         fuse_dev_cleanup();
1914  err_fs_cleanup:
1915         fuse_fs_cleanup();
1916  err:
1917         return res;
1918 }
1919
1920 static void __exit fuse_exit(void)
1921 {
1922         pr_debug("exit\n");
1923
1924         fuse_ctl_cleanup();
1925         fuse_sysfs_cleanup();
1926         fuse_fs_cleanup();
1927         fuse_dev_cleanup();
1928 }
1929
1930 module_init(fuse_init);
1931 module_exit(fuse_exit);