Merge tag 'fuse-update-5.15' of git://git.kernel.org/pub/scm/linux/kernel/git/mszered...
[linux-2.6-microblaze.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 fuse_put_super(struct super_block *sb)
461 {
462         struct fuse_mount *fm = get_fuse_mount_super(sb);
463
464         fuse_conn_put(fm->fc);
465         kfree(fm);
466 }
467
468 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
469 {
470         stbuf->f_type    = FUSE_SUPER_MAGIC;
471         stbuf->f_bsize   = attr->bsize;
472         stbuf->f_frsize  = attr->frsize;
473         stbuf->f_blocks  = attr->blocks;
474         stbuf->f_bfree   = attr->bfree;
475         stbuf->f_bavail  = attr->bavail;
476         stbuf->f_files   = attr->files;
477         stbuf->f_ffree   = attr->ffree;
478         stbuf->f_namelen = attr->namelen;
479         /* fsid is left zero */
480 }
481
482 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
483 {
484         struct super_block *sb = dentry->d_sb;
485         struct fuse_mount *fm = get_fuse_mount_super(sb);
486         FUSE_ARGS(args);
487         struct fuse_statfs_out outarg;
488         int err;
489
490         if (!fuse_allow_current_process(fm->fc)) {
491                 buf->f_type = FUSE_SUPER_MAGIC;
492                 return 0;
493         }
494
495         memset(&outarg, 0, sizeof(outarg));
496         args.in_numargs = 0;
497         args.opcode = FUSE_STATFS;
498         args.nodeid = get_node_id(d_inode(dentry));
499         args.out_numargs = 1;
500         args.out_args[0].size = sizeof(outarg);
501         args.out_args[0].value = &outarg;
502         err = fuse_simple_request(fm, &args);
503         if (!err)
504                 convert_fuse_statfs(buf, &outarg.st);
505         return err;
506 }
507
508 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
509 {
510         struct fuse_sync_bucket *bucket;
511
512         bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
513         if (bucket) {
514                 init_waitqueue_head(&bucket->waitq);
515                 /* Initial active count */
516                 atomic_set(&bucket->count, 1);
517         }
518         return bucket;
519 }
520
521 static void fuse_sync_fs_writes(struct fuse_conn *fc)
522 {
523         struct fuse_sync_bucket *bucket, *new_bucket;
524         int count;
525
526         new_bucket = fuse_sync_bucket_alloc();
527         spin_lock(&fc->lock);
528         bucket = rcu_dereference_protected(fc->curr_bucket, 1);
529         count = atomic_read(&bucket->count);
530         WARN_ON(count < 1);
531         /* No outstanding writes? */
532         if (count == 1) {
533                 spin_unlock(&fc->lock);
534                 kfree(new_bucket);
535                 return;
536         }
537
538         /*
539          * Completion of new bucket depends on completion of this bucket, so add
540          * one more count.
541          */
542         atomic_inc(&new_bucket->count);
543         rcu_assign_pointer(fc->curr_bucket, new_bucket);
544         spin_unlock(&fc->lock);
545         /*
546          * Drop initial active count.  At this point if all writes in this and
547          * ancestor buckets complete, the count will go to zero and this task
548          * will be woken up.
549          */
550         atomic_dec(&bucket->count);
551
552         wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
553
554         /* Drop temp count on descendant bucket */
555         fuse_sync_bucket_dec(new_bucket);
556         kfree_rcu(bucket, rcu);
557 }
558
559 static int fuse_sync_fs(struct super_block *sb, int wait)
560 {
561         struct fuse_mount *fm = get_fuse_mount_super(sb);
562         struct fuse_conn *fc = fm->fc;
563         struct fuse_syncfs_in inarg;
564         FUSE_ARGS(args);
565         int err;
566
567         /*
568          * Userspace cannot handle the wait == 0 case.  Avoid a
569          * gratuitous roundtrip.
570          */
571         if (!wait)
572                 return 0;
573
574         /* The filesystem is being unmounted.  Nothing to do. */
575         if (!sb->s_root)
576                 return 0;
577
578         if (!fc->sync_fs)
579                 return 0;
580
581         fuse_sync_fs_writes(fc);
582
583         memset(&inarg, 0, sizeof(inarg));
584         args.in_numargs = 1;
585         args.in_args[0].size = sizeof(inarg);
586         args.in_args[0].value = &inarg;
587         args.opcode = FUSE_SYNCFS;
588         args.nodeid = get_node_id(sb->s_root->d_inode);
589         args.out_numargs = 0;
590
591         err = fuse_simple_request(fm, &args);
592         if (err == -ENOSYS) {
593                 fc->sync_fs = 0;
594                 err = 0;
595         }
596
597         return err;
598 }
599
600 enum {
601         OPT_SOURCE,
602         OPT_SUBTYPE,
603         OPT_FD,
604         OPT_ROOTMODE,
605         OPT_USER_ID,
606         OPT_GROUP_ID,
607         OPT_DEFAULT_PERMISSIONS,
608         OPT_ALLOW_OTHER,
609         OPT_MAX_READ,
610         OPT_BLKSIZE,
611         OPT_ERR
612 };
613
614 static const struct fs_parameter_spec fuse_fs_parameters[] = {
615         fsparam_string  ("source",              OPT_SOURCE),
616         fsparam_u32     ("fd",                  OPT_FD),
617         fsparam_u32oct  ("rootmode",            OPT_ROOTMODE),
618         fsparam_u32     ("user_id",             OPT_USER_ID),
619         fsparam_u32     ("group_id",            OPT_GROUP_ID),
620         fsparam_flag    ("default_permissions", OPT_DEFAULT_PERMISSIONS),
621         fsparam_flag    ("allow_other",         OPT_ALLOW_OTHER),
622         fsparam_u32     ("max_read",            OPT_MAX_READ),
623         fsparam_u32     ("blksize",             OPT_BLKSIZE),
624         fsparam_string  ("subtype",             OPT_SUBTYPE),
625         {}
626 };
627
628 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
629 {
630         struct fs_parse_result result;
631         struct fuse_fs_context *ctx = fsc->fs_private;
632         int opt;
633
634         if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
635                 /*
636                  * Ignore options coming from mount(MS_REMOUNT) for backward
637                  * compatibility.
638                  */
639                 if (fsc->oldapi)
640                         return 0;
641
642                 return invalfc(fsc, "No changes allowed in reconfigure");
643         }
644
645         opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
646         if (opt < 0)
647                 return opt;
648
649         switch (opt) {
650         case OPT_SOURCE:
651                 if (fsc->source)
652                         return invalfc(fsc, "Multiple sources specified");
653                 fsc->source = param->string;
654                 param->string = NULL;
655                 break;
656
657         case OPT_SUBTYPE:
658                 if (ctx->subtype)
659                         return invalfc(fsc, "Multiple subtypes specified");
660                 ctx->subtype = param->string;
661                 param->string = NULL;
662                 return 0;
663
664         case OPT_FD:
665                 ctx->fd = result.uint_32;
666                 ctx->fd_present = true;
667                 break;
668
669         case OPT_ROOTMODE:
670                 if (!fuse_valid_type(result.uint_32))
671                         return invalfc(fsc, "Invalid rootmode");
672                 ctx->rootmode = result.uint_32;
673                 ctx->rootmode_present = true;
674                 break;
675
676         case OPT_USER_ID:
677                 ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
678                 if (!uid_valid(ctx->user_id))
679                         return invalfc(fsc, "Invalid user_id");
680                 ctx->user_id_present = true;
681                 break;
682
683         case OPT_GROUP_ID:
684                 ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
685                 if (!gid_valid(ctx->group_id))
686                         return invalfc(fsc, "Invalid group_id");
687                 ctx->group_id_present = true;
688                 break;
689
690         case OPT_DEFAULT_PERMISSIONS:
691                 ctx->default_permissions = true;
692                 break;
693
694         case OPT_ALLOW_OTHER:
695                 ctx->allow_other = true;
696                 break;
697
698         case OPT_MAX_READ:
699                 ctx->max_read = result.uint_32;
700                 break;
701
702         case OPT_BLKSIZE:
703                 if (!ctx->is_bdev)
704                         return invalfc(fsc, "blksize only supported for fuseblk");
705                 ctx->blksize = result.uint_32;
706                 break;
707
708         default:
709                 return -EINVAL;
710         }
711
712         return 0;
713 }
714
715 static void fuse_free_fsc(struct fs_context *fsc)
716 {
717         struct fuse_fs_context *ctx = fsc->fs_private;
718
719         if (ctx) {
720                 kfree(ctx->subtype);
721                 kfree(ctx);
722         }
723 }
724
725 static int fuse_show_options(struct seq_file *m, struct dentry *root)
726 {
727         struct super_block *sb = root->d_sb;
728         struct fuse_conn *fc = get_fuse_conn_super(sb);
729
730         if (fc->legacy_opts_show) {
731                 seq_printf(m, ",user_id=%u",
732                            from_kuid_munged(fc->user_ns, fc->user_id));
733                 seq_printf(m, ",group_id=%u",
734                            from_kgid_munged(fc->user_ns, fc->group_id));
735                 if (fc->default_permissions)
736                         seq_puts(m, ",default_permissions");
737                 if (fc->allow_other)
738                         seq_puts(m, ",allow_other");
739                 if (fc->max_read != ~0)
740                         seq_printf(m, ",max_read=%u", fc->max_read);
741                 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
742                         seq_printf(m, ",blksize=%lu", sb->s_blocksize);
743         }
744 #ifdef CONFIG_FUSE_DAX
745         if (fc->dax)
746                 seq_puts(m, ",dax");
747 #endif
748
749         return 0;
750 }
751
752 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
753                              const struct fuse_iqueue_ops *ops,
754                              void *priv)
755 {
756         memset(fiq, 0, sizeof(struct fuse_iqueue));
757         spin_lock_init(&fiq->lock);
758         init_waitqueue_head(&fiq->waitq);
759         INIT_LIST_HEAD(&fiq->pending);
760         INIT_LIST_HEAD(&fiq->interrupts);
761         fiq->forget_list_tail = &fiq->forget_list_head;
762         fiq->connected = 1;
763         fiq->ops = ops;
764         fiq->priv = priv;
765 }
766
767 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
768 {
769         unsigned int i;
770
771         spin_lock_init(&fpq->lock);
772         for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
773                 INIT_LIST_HEAD(&fpq->processing[i]);
774         INIT_LIST_HEAD(&fpq->io);
775         fpq->connected = 1;
776 }
777
778 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
779                     struct user_namespace *user_ns,
780                     const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
781 {
782         memset(fc, 0, sizeof(*fc));
783         spin_lock_init(&fc->lock);
784         spin_lock_init(&fc->bg_lock);
785         init_rwsem(&fc->killsb);
786         refcount_set(&fc->count, 1);
787         atomic_set(&fc->dev_count, 1);
788         init_waitqueue_head(&fc->blocked_waitq);
789         fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
790         INIT_LIST_HEAD(&fc->bg_queue);
791         INIT_LIST_HEAD(&fc->entry);
792         INIT_LIST_HEAD(&fc->devices);
793         atomic_set(&fc->num_waiting, 0);
794         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
795         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
796         atomic64_set(&fc->khctr, 0);
797         fc->polled_files = RB_ROOT;
798         fc->blocked = 0;
799         fc->initialized = 0;
800         fc->connected = 1;
801         atomic64_set(&fc->attr_version, 1);
802         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
803         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
804         fc->user_ns = get_user_ns(user_ns);
805         fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
806         fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
807
808         INIT_LIST_HEAD(&fc->mounts);
809         list_add(&fm->fc_entry, &fc->mounts);
810         fm->fc = fc;
811 }
812 EXPORT_SYMBOL_GPL(fuse_conn_init);
813
814 void fuse_conn_put(struct fuse_conn *fc)
815 {
816         if (refcount_dec_and_test(&fc->count)) {
817                 struct fuse_iqueue *fiq = &fc->iq;
818                 struct fuse_sync_bucket *bucket;
819
820                 if (IS_ENABLED(CONFIG_FUSE_DAX))
821                         fuse_dax_conn_free(fc);
822                 if (fiq->ops->release)
823                         fiq->ops->release(fiq);
824                 put_pid_ns(fc->pid_ns);
825                 put_user_ns(fc->user_ns);
826                 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
827                 if (bucket) {
828                         WARN_ON(atomic_read(&bucket->count) != 1);
829                         kfree(bucket);
830                 }
831                 fc->release(fc);
832         }
833 }
834 EXPORT_SYMBOL_GPL(fuse_conn_put);
835
836 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
837 {
838         refcount_inc(&fc->count);
839         return fc;
840 }
841 EXPORT_SYMBOL_GPL(fuse_conn_get);
842
843 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
844 {
845         struct fuse_attr attr;
846         memset(&attr, 0, sizeof(attr));
847
848         attr.mode = mode;
849         attr.ino = FUSE_ROOT_ID;
850         attr.nlink = 1;
851         return fuse_iget(sb, 1, 0, &attr, 0, 0);
852 }
853
854 struct fuse_inode_handle {
855         u64 nodeid;
856         u32 generation;
857 };
858
859 static struct dentry *fuse_get_dentry(struct super_block *sb,
860                                       struct fuse_inode_handle *handle)
861 {
862         struct fuse_conn *fc = get_fuse_conn_super(sb);
863         struct inode *inode;
864         struct dentry *entry;
865         int err = -ESTALE;
866
867         if (handle->nodeid == 0)
868                 goto out_err;
869
870         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
871         if (!inode) {
872                 struct fuse_entry_out outarg;
873                 const struct qstr name = QSTR_INIT(".", 1);
874
875                 if (!fc->export_support)
876                         goto out_err;
877
878                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
879                                        &inode);
880                 if (err && err != -ENOENT)
881                         goto out_err;
882                 if (err || !inode) {
883                         err = -ESTALE;
884                         goto out_err;
885                 }
886                 err = -EIO;
887                 if (get_node_id(inode) != handle->nodeid)
888                         goto out_iput;
889         }
890         err = -ESTALE;
891         if (inode->i_generation != handle->generation)
892                 goto out_iput;
893
894         entry = d_obtain_alias(inode);
895         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
896                 fuse_invalidate_entry_cache(entry);
897
898         return entry;
899
900  out_iput:
901         iput(inode);
902  out_err:
903         return ERR_PTR(err);
904 }
905
906 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
907                            struct inode *parent)
908 {
909         int len = parent ? 6 : 3;
910         u64 nodeid;
911         u32 generation;
912
913         if (*max_len < len) {
914                 *max_len = len;
915                 return  FILEID_INVALID;
916         }
917
918         nodeid = get_fuse_inode(inode)->nodeid;
919         generation = inode->i_generation;
920
921         fh[0] = (u32)(nodeid >> 32);
922         fh[1] = (u32)(nodeid & 0xffffffff);
923         fh[2] = generation;
924
925         if (parent) {
926                 nodeid = get_fuse_inode(parent)->nodeid;
927                 generation = parent->i_generation;
928
929                 fh[3] = (u32)(nodeid >> 32);
930                 fh[4] = (u32)(nodeid & 0xffffffff);
931                 fh[5] = generation;
932         }
933
934         *max_len = len;
935         return parent ? 0x82 : 0x81;
936 }
937
938 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
939                 struct fid *fid, int fh_len, int fh_type)
940 {
941         struct fuse_inode_handle handle;
942
943         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
944                 return NULL;
945
946         handle.nodeid = (u64) fid->raw[0] << 32;
947         handle.nodeid |= (u64) fid->raw[1];
948         handle.generation = fid->raw[2];
949         return fuse_get_dentry(sb, &handle);
950 }
951
952 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
953                 struct fid *fid, int fh_len, int fh_type)
954 {
955         struct fuse_inode_handle parent;
956
957         if (fh_type != 0x82 || fh_len < 6)
958                 return NULL;
959
960         parent.nodeid = (u64) fid->raw[3] << 32;
961         parent.nodeid |= (u64) fid->raw[4];
962         parent.generation = fid->raw[5];
963         return fuse_get_dentry(sb, &parent);
964 }
965
966 static struct dentry *fuse_get_parent(struct dentry *child)
967 {
968         struct inode *child_inode = d_inode(child);
969         struct fuse_conn *fc = get_fuse_conn(child_inode);
970         struct inode *inode;
971         struct dentry *parent;
972         struct fuse_entry_out outarg;
973         int err;
974
975         if (!fc->export_support)
976                 return ERR_PTR(-ESTALE);
977
978         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
979                                &dotdot_name, &outarg, &inode);
980         if (err) {
981                 if (err == -ENOENT)
982                         return ERR_PTR(-ESTALE);
983                 return ERR_PTR(err);
984         }
985
986         parent = d_obtain_alias(inode);
987         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
988                 fuse_invalidate_entry_cache(parent);
989
990         return parent;
991 }
992
993 static const struct export_operations fuse_export_operations = {
994         .fh_to_dentry   = fuse_fh_to_dentry,
995         .fh_to_parent   = fuse_fh_to_parent,
996         .encode_fh      = fuse_encode_fh,
997         .get_parent     = fuse_get_parent,
998 };
999
1000 static const struct super_operations fuse_super_operations = {
1001         .alloc_inode    = fuse_alloc_inode,
1002         .free_inode     = fuse_free_inode,
1003         .evict_inode    = fuse_evict_inode,
1004         .write_inode    = fuse_write_inode,
1005         .drop_inode     = generic_delete_inode,
1006         .put_super      = fuse_put_super,
1007         .umount_begin   = fuse_umount_begin,
1008         .statfs         = fuse_statfs,
1009         .sync_fs        = fuse_sync_fs,
1010         .show_options   = fuse_show_options,
1011 };
1012
1013 static void sanitize_global_limit(unsigned *limit)
1014 {
1015         /*
1016          * The default maximum number of async requests is calculated to consume
1017          * 1/2^13 of the total memory, assuming 392 bytes per request.
1018          */
1019         if (*limit == 0)
1020                 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1021
1022         if (*limit >= 1 << 16)
1023                 *limit = (1 << 16) - 1;
1024 }
1025
1026 static int set_global_limit(const char *val, const struct kernel_param *kp)
1027 {
1028         int rv;
1029
1030         rv = param_set_uint(val, kp);
1031         if (rv)
1032                 return rv;
1033
1034         sanitize_global_limit((unsigned *)kp->arg);
1035
1036         return 0;
1037 }
1038
1039 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1040 {
1041         int cap_sys_admin = capable(CAP_SYS_ADMIN);
1042
1043         if (arg->minor < 13)
1044                 return;
1045
1046         sanitize_global_limit(&max_user_bgreq);
1047         sanitize_global_limit(&max_user_congthresh);
1048
1049         spin_lock(&fc->bg_lock);
1050         if (arg->max_background) {
1051                 fc->max_background = arg->max_background;
1052
1053                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1054                         fc->max_background = max_user_bgreq;
1055         }
1056         if (arg->congestion_threshold) {
1057                 fc->congestion_threshold = arg->congestion_threshold;
1058
1059                 if (!cap_sys_admin &&
1060                     fc->congestion_threshold > max_user_congthresh)
1061                         fc->congestion_threshold = max_user_congthresh;
1062         }
1063         spin_unlock(&fc->bg_lock);
1064 }
1065
1066 struct fuse_init_args {
1067         struct fuse_args args;
1068         struct fuse_init_in in;
1069         struct fuse_init_out out;
1070 };
1071
1072 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1073                                int error)
1074 {
1075         struct fuse_conn *fc = fm->fc;
1076         struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1077         struct fuse_init_out *arg = &ia->out;
1078         bool ok = true;
1079
1080         if (error || arg->major != FUSE_KERNEL_VERSION)
1081                 ok = false;
1082         else {
1083                 unsigned long ra_pages;
1084
1085                 process_init_limits(fc, arg);
1086
1087                 if (arg->minor >= 6) {
1088                         ra_pages = arg->max_readahead / PAGE_SIZE;
1089                         if (arg->flags & FUSE_ASYNC_READ)
1090                                 fc->async_read = 1;
1091                         if (!(arg->flags & FUSE_POSIX_LOCKS))
1092                                 fc->no_lock = 1;
1093                         if (arg->minor >= 17) {
1094                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
1095                                         fc->no_flock = 1;
1096                         } else {
1097                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
1098                                         fc->no_flock = 1;
1099                         }
1100                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
1101                                 fc->atomic_o_trunc = 1;
1102                         if (arg->minor >= 9) {
1103                                 /* LOOKUP has dependency on proto version */
1104                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
1105                                         fc->export_support = 1;
1106                         }
1107                         if (arg->flags & FUSE_BIG_WRITES)
1108                                 fc->big_writes = 1;
1109                         if (arg->flags & FUSE_DONT_MASK)
1110                                 fc->dont_mask = 1;
1111                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
1112                                 fc->auto_inval_data = 1;
1113                         else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
1114                                 fc->explicit_inval_data = 1;
1115                         if (arg->flags & FUSE_DO_READDIRPLUS) {
1116                                 fc->do_readdirplus = 1;
1117                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
1118                                         fc->readdirplus_auto = 1;
1119                         }
1120                         if (arg->flags & FUSE_ASYNC_DIO)
1121                                 fc->async_dio = 1;
1122                         if (arg->flags & FUSE_WRITEBACK_CACHE)
1123                                 fc->writeback_cache = 1;
1124                         if (arg->flags & FUSE_PARALLEL_DIROPS)
1125                                 fc->parallel_dirops = 1;
1126                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
1127                                 fc->handle_killpriv = 1;
1128                         if (arg->time_gran && arg->time_gran <= 1000000000)
1129                                 fm->sb->s_time_gran = arg->time_gran;
1130                         if ((arg->flags & FUSE_POSIX_ACL)) {
1131                                 fc->default_permissions = 1;
1132                                 fc->posix_acl = 1;
1133                                 fm->sb->s_xattr = fuse_acl_xattr_handlers;
1134                         }
1135                         if (arg->flags & FUSE_CACHE_SYMLINKS)
1136                                 fc->cache_symlinks = 1;
1137                         if (arg->flags & FUSE_ABORT_ERROR)
1138                                 fc->abort_err = 1;
1139                         if (arg->flags & FUSE_MAX_PAGES) {
1140                                 fc->max_pages =
1141                                         min_t(unsigned int, fc->max_pages_limit,
1142                                         max_t(unsigned int, arg->max_pages, 1));
1143                         }
1144                         if (IS_ENABLED(CONFIG_FUSE_DAX) &&
1145                             arg->flags & FUSE_MAP_ALIGNMENT &&
1146                             !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1147                                 ok = false;
1148                         }
1149                         if (arg->flags & FUSE_HANDLE_KILLPRIV_V2) {
1150                                 fc->handle_killpriv_v2 = 1;
1151                                 fm->sb->s_flags |= SB_NOSEC;
1152                         }
1153                         if (arg->flags & FUSE_SETXATTR_EXT)
1154                                 fc->setxattr_ext = 1;
1155                 } else {
1156                         ra_pages = fc->max_read / PAGE_SIZE;
1157                         fc->no_lock = 1;
1158                         fc->no_flock = 1;
1159                 }
1160
1161                 fm->sb->s_bdi->ra_pages =
1162                                 min(fm->sb->s_bdi->ra_pages, ra_pages);
1163                 fc->minor = arg->minor;
1164                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1165                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1166                 fc->conn_init = 1;
1167         }
1168         kfree(ia);
1169
1170         if (!ok) {
1171                 fc->conn_init = 0;
1172                 fc->conn_error = 1;
1173         }
1174
1175         fuse_set_initialized(fc);
1176         wake_up_all(&fc->blocked_waitq);
1177 }
1178
1179 void fuse_send_init(struct fuse_mount *fm)
1180 {
1181         struct fuse_init_args *ia;
1182
1183         ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1184
1185         ia->in.major = FUSE_KERNEL_VERSION;
1186         ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1187         ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1188         ia->in.flags |=
1189                 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1190                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1191                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1192                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1193                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1194                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1195                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1196                 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1197                 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1198                 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT;
1199 #ifdef CONFIG_FUSE_DAX
1200         if (fm->fc->dax)
1201                 ia->in.flags |= FUSE_MAP_ALIGNMENT;
1202 #endif
1203         if (fm->fc->auto_submounts)
1204                 ia->in.flags |= FUSE_SUBMOUNTS;
1205
1206         ia->args.opcode = FUSE_INIT;
1207         ia->args.in_numargs = 1;
1208         ia->args.in_args[0].size = sizeof(ia->in);
1209         ia->args.in_args[0].value = &ia->in;
1210         ia->args.out_numargs = 1;
1211         /* Variable length argument used for backward compatibility
1212            with interface version < 7.5.  Rest of init_out is zeroed
1213            by do_get_request(), so a short reply is not a problem */
1214         ia->args.out_argvar = true;
1215         ia->args.out_args[0].size = sizeof(ia->out);
1216         ia->args.out_args[0].value = &ia->out;
1217         ia->args.force = true;
1218         ia->args.nocreds = true;
1219         ia->args.end = process_init_reply;
1220
1221         if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1222                 process_init_reply(fm, &ia->args, -ENOTCONN);
1223 }
1224 EXPORT_SYMBOL_GPL(fuse_send_init);
1225
1226 void fuse_free_conn(struct fuse_conn *fc)
1227 {
1228         WARN_ON(!list_empty(&fc->devices));
1229         kfree_rcu(fc, rcu);
1230 }
1231 EXPORT_SYMBOL_GPL(fuse_free_conn);
1232
1233 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1234 {
1235         int err;
1236         char *suffix = "";
1237
1238         if (sb->s_bdev) {
1239                 suffix = "-fuseblk";
1240                 /*
1241                  * sb->s_bdi points to blkdev's bdi however we want to redirect
1242                  * it to our private bdi...
1243                  */
1244                 bdi_put(sb->s_bdi);
1245                 sb->s_bdi = &noop_backing_dev_info;
1246         }
1247         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1248                                    MINOR(fc->dev), suffix);
1249         if (err)
1250                 return err;
1251
1252         /* fuse does it's own writeback accounting */
1253         sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1254         sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1255
1256         /*
1257          * For a single fuse filesystem use max 1% of dirty +
1258          * writeback threshold.
1259          *
1260          * This gives about 1M of write buffer for memory maps on a
1261          * machine with 1G and 10% dirty_ratio, which should be more
1262          * than enough.
1263          *
1264          * Privileged users can raise it by writing to
1265          *
1266          *    /sys/class/bdi/<bdi>/max_ratio
1267          */
1268         bdi_set_max_ratio(sb->s_bdi, 1);
1269
1270         return 0;
1271 }
1272
1273 struct fuse_dev *fuse_dev_alloc(void)
1274 {
1275         struct fuse_dev *fud;
1276         struct list_head *pq;
1277
1278         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1279         if (!fud)
1280                 return NULL;
1281
1282         pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1283         if (!pq) {
1284                 kfree(fud);
1285                 return NULL;
1286         }
1287
1288         fud->pq.processing = pq;
1289         fuse_pqueue_init(&fud->pq);
1290
1291         return fud;
1292 }
1293 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1294
1295 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1296 {
1297         fud->fc = fuse_conn_get(fc);
1298         spin_lock(&fc->lock);
1299         list_add_tail(&fud->entry, &fc->devices);
1300         spin_unlock(&fc->lock);
1301 }
1302 EXPORT_SYMBOL_GPL(fuse_dev_install);
1303
1304 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1305 {
1306         struct fuse_dev *fud;
1307
1308         fud = fuse_dev_alloc();
1309         if (!fud)
1310                 return NULL;
1311
1312         fuse_dev_install(fud, fc);
1313         return fud;
1314 }
1315 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1316
1317 void fuse_dev_free(struct fuse_dev *fud)
1318 {
1319         struct fuse_conn *fc = fud->fc;
1320
1321         if (fc) {
1322                 spin_lock(&fc->lock);
1323                 list_del(&fud->entry);
1324                 spin_unlock(&fc->lock);
1325
1326                 fuse_conn_put(fc);
1327         }
1328         kfree(fud->pq.processing);
1329         kfree(fud);
1330 }
1331 EXPORT_SYMBOL_GPL(fuse_dev_free);
1332
1333 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1334                                       const struct fuse_inode *fi)
1335 {
1336         *attr = (struct fuse_attr){
1337                 .ino            = fi->inode.i_ino,
1338                 .size           = fi->inode.i_size,
1339                 .blocks         = fi->inode.i_blocks,
1340                 .atime          = fi->inode.i_atime.tv_sec,
1341                 .mtime          = fi->inode.i_mtime.tv_sec,
1342                 .ctime          = fi->inode.i_ctime.tv_sec,
1343                 .atimensec      = fi->inode.i_atime.tv_nsec,
1344                 .mtimensec      = fi->inode.i_mtime.tv_nsec,
1345                 .ctimensec      = fi->inode.i_ctime.tv_nsec,
1346                 .mode           = fi->inode.i_mode,
1347                 .nlink          = fi->inode.i_nlink,
1348                 .uid            = fi->inode.i_uid.val,
1349                 .gid            = fi->inode.i_gid.val,
1350                 .rdev           = fi->inode.i_rdev,
1351                 .blksize        = 1u << fi->inode.i_blkbits,
1352         };
1353 }
1354
1355 static void fuse_sb_defaults(struct super_block *sb)
1356 {
1357         sb->s_magic = FUSE_SUPER_MAGIC;
1358         sb->s_op = &fuse_super_operations;
1359         sb->s_xattr = fuse_xattr_handlers;
1360         sb->s_maxbytes = MAX_LFS_FILESIZE;
1361         sb->s_time_gran = 1;
1362         sb->s_export_op = &fuse_export_operations;
1363         sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1364         if (sb->s_user_ns != &init_user_ns)
1365                 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1366         sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1367
1368         /*
1369          * If we are not in the initial user namespace posix
1370          * acls must be translated.
1371          */
1372         if (sb->s_user_ns != &init_user_ns)
1373                 sb->s_xattr = fuse_no_acl_xattr_handlers;
1374 }
1375
1376 static int fuse_fill_super_submount(struct super_block *sb,
1377                                     struct fuse_inode *parent_fi)
1378 {
1379         struct fuse_mount *fm = get_fuse_mount_super(sb);
1380         struct super_block *parent_sb = parent_fi->inode.i_sb;
1381         struct fuse_attr root_attr;
1382         struct inode *root;
1383
1384         fuse_sb_defaults(sb);
1385         fm->sb = sb;
1386
1387         WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1388         sb->s_bdi = bdi_get(parent_sb->s_bdi);
1389
1390         sb->s_xattr = parent_sb->s_xattr;
1391         sb->s_time_gran = parent_sb->s_time_gran;
1392         sb->s_blocksize = parent_sb->s_blocksize;
1393         sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1394         sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1395         if (parent_sb->s_subtype && !sb->s_subtype)
1396                 return -ENOMEM;
1397
1398         fuse_fill_attr_from_inode(&root_attr, parent_fi);
1399         root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1400         /*
1401          * This inode is just a duplicate, so it is not looked up and
1402          * its nlookup should not be incremented.  fuse_iget() does
1403          * that, though, so undo it here.
1404          */
1405         get_fuse_inode(root)->nlookup--;
1406         sb->s_d_op = &fuse_dentry_operations;
1407         sb->s_root = d_make_root(root);
1408         if (!sb->s_root)
1409                 return -ENOMEM;
1410
1411         return 0;
1412 }
1413
1414 /* Filesystem context private data holds the FUSE inode of the mount point */
1415 static int fuse_get_tree_submount(struct fs_context *fsc)
1416 {
1417         struct fuse_mount *fm;
1418         struct fuse_inode *mp_fi = fsc->fs_private;
1419         struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1420         struct super_block *sb;
1421         int err;
1422
1423         fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1424         if (!fm)
1425                 return -ENOMEM;
1426
1427         fsc->s_fs_info = fm;
1428         sb = sget_fc(fsc, NULL, set_anon_super_fc);
1429         if (IS_ERR(sb)) {
1430                 kfree(fm);
1431                 return PTR_ERR(sb);
1432         }
1433         fm->fc = fuse_conn_get(fc);
1434
1435         /* Initialize superblock, making @mp_fi its root */
1436         err = fuse_fill_super_submount(sb, mp_fi);
1437         if (err) {
1438                 fuse_conn_put(fc);
1439                 kfree(fm);
1440                 sb->s_fs_info = NULL;
1441                 deactivate_locked_super(sb);
1442                 return err;
1443         }
1444
1445         down_write(&fc->killsb);
1446         list_add_tail(&fm->fc_entry, &fc->mounts);
1447         up_write(&fc->killsb);
1448
1449         sb->s_flags |= SB_ACTIVE;
1450         fsc->root = dget(sb->s_root);
1451
1452         return 0;
1453 }
1454
1455 static const struct fs_context_operations fuse_context_submount_ops = {
1456         .get_tree       = fuse_get_tree_submount,
1457 };
1458
1459 int fuse_init_fs_context_submount(struct fs_context *fsc)
1460 {
1461         fsc->ops = &fuse_context_submount_ops;
1462         return 0;
1463 }
1464 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1465
1466 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1467 {
1468         struct fuse_dev *fud = NULL;
1469         struct fuse_mount *fm = get_fuse_mount_super(sb);
1470         struct fuse_conn *fc = fm->fc;
1471         struct inode *root;
1472         struct dentry *root_dentry;
1473         int err;
1474
1475         err = -EINVAL;
1476         if (sb->s_flags & SB_MANDLOCK)
1477                 goto err;
1478
1479         rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1480         fuse_sb_defaults(sb);
1481
1482         if (ctx->is_bdev) {
1483 #ifdef CONFIG_BLOCK
1484                 err = -EINVAL;
1485                 if (!sb_set_blocksize(sb, ctx->blksize))
1486                         goto err;
1487 #endif
1488         } else {
1489                 sb->s_blocksize = PAGE_SIZE;
1490                 sb->s_blocksize_bits = PAGE_SHIFT;
1491         }
1492
1493         sb->s_subtype = ctx->subtype;
1494         ctx->subtype = NULL;
1495         if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1496                 err = fuse_dax_conn_alloc(fc, ctx->dax_dev);
1497                 if (err)
1498                         goto err;
1499         }
1500
1501         if (ctx->fudptr) {
1502                 err = -ENOMEM;
1503                 fud = fuse_dev_alloc_install(fc);
1504                 if (!fud)
1505                         goto err_free_dax;
1506         }
1507
1508         fc->dev = sb->s_dev;
1509         fm->sb = sb;
1510         err = fuse_bdi_init(fc, sb);
1511         if (err)
1512                 goto err_dev_free;
1513
1514         /* Handle umasking inside the fuse code */
1515         if (sb->s_flags & SB_POSIXACL)
1516                 fc->dont_mask = 1;
1517         sb->s_flags |= SB_POSIXACL;
1518
1519         fc->default_permissions = ctx->default_permissions;
1520         fc->allow_other = ctx->allow_other;
1521         fc->user_id = ctx->user_id;
1522         fc->group_id = ctx->group_id;
1523         fc->legacy_opts_show = ctx->legacy_opts_show;
1524         fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1525         fc->destroy = ctx->destroy;
1526         fc->no_control = ctx->no_control;
1527         fc->no_force_umount = ctx->no_force_umount;
1528
1529         err = -ENOMEM;
1530         root = fuse_get_root_inode(sb, ctx->rootmode);
1531         sb->s_d_op = &fuse_root_dentry_operations;
1532         root_dentry = d_make_root(root);
1533         if (!root_dentry)
1534                 goto err_dev_free;
1535         /* Root dentry doesn't have .d_revalidate */
1536         sb->s_d_op = &fuse_dentry_operations;
1537
1538         mutex_lock(&fuse_mutex);
1539         err = -EINVAL;
1540         if (ctx->fudptr && *ctx->fudptr)
1541                 goto err_unlock;
1542
1543         err = fuse_ctl_add_conn(fc);
1544         if (err)
1545                 goto err_unlock;
1546
1547         list_add_tail(&fc->entry, &fuse_conn_list);
1548         sb->s_root = root_dentry;
1549         if (ctx->fudptr)
1550                 *ctx->fudptr = fud;
1551         mutex_unlock(&fuse_mutex);
1552         return 0;
1553
1554  err_unlock:
1555         mutex_unlock(&fuse_mutex);
1556         dput(root_dentry);
1557  err_dev_free:
1558         if (fud)
1559                 fuse_dev_free(fud);
1560  err_free_dax:
1561         if (IS_ENABLED(CONFIG_FUSE_DAX))
1562                 fuse_dax_conn_free(fc);
1563  err:
1564         return err;
1565 }
1566 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1567
1568 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1569 {
1570         struct fuse_fs_context *ctx = fsc->fs_private;
1571         int err;
1572         struct fuse_conn *fc;
1573         struct fuse_mount *fm;
1574
1575         if (!ctx->file || !ctx->rootmode_present ||
1576             !ctx->user_id_present || !ctx->group_id_present)
1577                 return -EINVAL;
1578
1579         /*
1580          * Require mount to happen from the same user namespace which
1581          * opened /dev/fuse to prevent potential attacks.
1582          */
1583         err = -EINVAL;
1584         if ((ctx->file->f_op != &fuse_dev_operations) ||
1585             (ctx->file->f_cred->user_ns != sb->s_user_ns))
1586                 goto err;
1587         ctx->fudptr = &ctx->file->private_data;
1588
1589         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1590         err = -ENOMEM;
1591         if (!fc)
1592                 goto err;
1593
1594         fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1595         if (!fm) {
1596                 kfree(fc);
1597                 goto err;
1598         }
1599
1600         fuse_conn_init(fc, fm, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1601         fc->release = fuse_free_conn;
1602
1603         sb->s_fs_info = fm;
1604
1605         err = fuse_fill_super_common(sb, ctx);
1606         if (err)
1607                 goto err_put_conn;
1608         /* file->private_data shall be visible on all CPUs after this */
1609         smp_mb();
1610         fuse_send_init(get_fuse_mount_super(sb));
1611         return 0;
1612
1613  err_put_conn:
1614         fuse_conn_put(fc);
1615         kfree(fm);
1616         sb->s_fs_info = NULL;
1617  err:
1618         return err;
1619 }
1620
1621 /*
1622  * This is the path where user supplied an already initialized fuse dev.  In
1623  * this case never create a new super if the old one is gone.
1624  */
1625 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1626 {
1627         return -ENOTCONN;
1628 }
1629
1630 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1631 {
1632
1633         return fsc->sget_key == get_fuse_conn_super(sb);
1634 }
1635
1636 static int fuse_get_tree(struct fs_context *fsc)
1637 {
1638         struct fuse_fs_context *ctx = fsc->fs_private;
1639         struct fuse_dev *fud;
1640         struct super_block *sb;
1641         int err;
1642
1643         if (ctx->fd_present)
1644                 ctx->file = fget(ctx->fd);
1645
1646         if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1647                 err = get_tree_bdev(fsc, fuse_fill_super);
1648                 goto out_fput;
1649         }
1650         /*
1651          * While block dev mount can be initialized with a dummy device fd
1652          * (found by device name), normal fuse mounts can't
1653          */
1654         if (!ctx->file)
1655                 return -EINVAL;
1656
1657         /*
1658          * Allow creating a fuse mount with an already initialized fuse
1659          * connection
1660          */
1661         fud = READ_ONCE(ctx->file->private_data);
1662         if (ctx->file->f_op == &fuse_dev_operations && fud) {
1663                 fsc->sget_key = fud->fc;
1664                 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1665                 err = PTR_ERR_OR_ZERO(sb);
1666                 if (!IS_ERR(sb))
1667                         fsc->root = dget(sb->s_root);
1668         } else {
1669                 err = get_tree_nodev(fsc, fuse_fill_super);
1670         }
1671 out_fput:
1672         if (ctx->file)
1673                 fput(ctx->file);
1674         return err;
1675 }
1676
1677 static const struct fs_context_operations fuse_context_ops = {
1678         .free           = fuse_free_fsc,
1679         .parse_param    = fuse_parse_param,
1680         .reconfigure    = fuse_reconfigure,
1681         .get_tree       = fuse_get_tree,
1682 };
1683
1684 /*
1685  * Set up the filesystem mount context.
1686  */
1687 static int fuse_init_fs_context(struct fs_context *fsc)
1688 {
1689         struct fuse_fs_context *ctx;
1690
1691         ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1692         if (!ctx)
1693                 return -ENOMEM;
1694
1695         ctx->max_read = ~0;
1696         ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1697         ctx->legacy_opts_show = true;
1698
1699 #ifdef CONFIG_BLOCK
1700         if (fsc->fs_type == &fuseblk_fs_type) {
1701                 ctx->is_bdev = true;
1702                 ctx->destroy = true;
1703         }
1704 #endif
1705
1706         fsc->fs_private = ctx;
1707         fsc->ops = &fuse_context_ops;
1708         return 0;
1709 }
1710
1711 bool fuse_mount_remove(struct fuse_mount *fm)
1712 {
1713         struct fuse_conn *fc = fm->fc;
1714         bool last = false;
1715
1716         down_write(&fc->killsb);
1717         list_del_init(&fm->fc_entry);
1718         if (list_empty(&fc->mounts))
1719                 last = true;
1720         up_write(&fc->killsb);
1721
1722         return last;
1723 }
1724 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1725
1726 void fuse_conn_destroy(struct fuse_mount *fm)
1727 {
1728         struct fuse_conn *fc = fm->fc;
1729
1730         if (fc->destroy)
1731                 fuse_send_destroy(fm);
1732
1733         fuse_abort_conn(fc);
1734         fuse_wait_aborted(fc);
1735
1736         if (!list_empty(&fc->entry)) {
1737                 mutex_lock(&fuse_mutex);
1738                 list_del(&fc->entry);
1739                 fuse_ctl_remove_conn(fc);
1740                 mutex_unlock(&fuse_mutex);
1741         }
1742 }
1743 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1744
1745 static void fuse_sb_destroy(struct super_block *sb)
1746 {
1747         struct fuse_mount *fm = get_fuse_mount_super(sb);
1748         bool last;
1749
1750         if (fm) {
1751                 last = fuse_mount_remove(fm);
1752                 if (last)
1753                         fuse_conn_destroy(fm);
1754         }
1755 }
1756
1757 static void fuse_kill_sb_anon(struct super_block *sb)
1758 {
1759         fuse_sb_destroy(sb);
1760         kill_anon_super(sb);
1761 }
1762
1763 static struct file_system_type fuse_fs_type = {
1764         .owner          = THIS_MODULE,
1765         .name           = "fuse",
1766         .fs_flags       = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1767         .init_fs_context = fuse_init_fs_context,
1768         .parameters     = fuse_fs_parameters,
1769         .kill_sb        = fuse_kill_sb_anon,
1770 };
1771 MODULE_ALIAS_FS("fuse");
1772
1773 #ifdef CONFIG_BLOCK
1774 static void fuse_kill_sb_blk(struct super_block *sb)
1775 {
1776         fuse_sb_destroy(sb);
1777         kill_block_super(sb);
1778 }
1779
1780 static struct file_system_type fuseblk_fs_type = {
1781         .owner          = THIS_MODULE,
1782         .name           = "fuseblk",
1783         .init_fs_context = fuse_init_fs_context,
1784         .parameters     = fuse_fs_parameters,
1785         .kill_sb        = fuse_kill_sb_blk,
1786         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1787 };
1788 MODULE_ALIAS_FS("fuseblk");
1789
1790 static inline int register_fuseblk(void)
1791 {
1792         return register_filesystem(&fuseblk_fs_type);
1793 }
1794
1795 static inline void unregister_fuseblk(void)
1796 {
1797         unregister_filesystem(&fuseblk_fs_type);
1798 }
1799 #else
1800 static inline int register_fuseblk(void)
1801 {
1802         return 0;
1803 }
1804
1805 static inline void unregister_fuseblk(void)
1806 {
1807 }
1808 #endif
1809
1810 static void fuse_inode_init_once(void *foo)
1811 {
1812         struct inode *inode = foo;
1813
1814         inode_init_once(inode);
1815 }
1816
1817 static int __init fuse_fs_init(void)
1818 {
1819         int err;
1820
1821         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1822                         sizeof(struct fuse_inode), 0,
1823                         SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1824                         fuse_inode_init_once);
1825         err = -ENOMEM;
1826         if (!fuse_inode_cachep)
1827                 goto out;
1828
1829         err = register_fuseblk();
1830         if (err)
1831                 goto out2;
1832
1833         err = register_filesystem(&fuse_fs_type);
1834         if (err)
1835                 goto out3;
1836
1837         return 0;
1838
1839  out3:
1840         unregister_fuseblk();
1841  out2:
1842         kmem_cache_destroy(fuse_inode_cachep);
1843  out:
1844         return err;
1845 }
1846
1847 static void fuse_fs_cleanup(void)
1848 {
1849         unregister_filesystem(&fuse_fs_type);
1850         unregister_fuseblk();
1851
1852         /*
1853          * Make sure all delayed rcu free inodes are flushed before we
1854          * destroy cache.
1855          */
1856         rcu_barrier();
1857         kmem_cache_destroy(fuse_inode_cachep);
1858 }
1859
1860 static struct kobject *fuse_kobj;
1861
1862 static int fuse_sysfs_init(void)
1863 {
1864         int err;
1865
1866         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1867         if (!fuse_kobj) {
1868                 err = -ENOMEM;
1869                 goto out_err;
1870         }
1871
1872         err = sysfs_create_mount_point(fuse_kobj, "connections");
1873         if (err)
1874                 goto out_fuse_unregister;
1875
1876         return 0;
1877
1878  out_fuse_unregister:
1879         kobject_put(fuse_kobj);
1880  out_err:
1881         return err;
1882 }
1883
1884 static void fuse_sysfs_cleanup(void)
1885 {
1886         sysfs_remove_mount_point(fuse_kobj, "connections");
1887         kobject_put(fuse_kobj);
1888 }
1889
1890 static int __init fuse_init(void)
1891 {
1892         int res;
1893
1894         pr_info("init (API version %i.%i)\n",
1895                 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1896
1897         INIT_LIST_HEAD(&fuse_conn_list);
1898         res = fuse_fs_init();
1899         if (res)
1900                 goto err;
1901
1902         res = fuse_dev_init();
1903         if (res)
1904                 goto err_fs_cleanup;
1905
1906         res = fuse_sysfs_init();
1907         if (res)
1908                 goto err_dev_cleanup;
1909
1910         res = fuse_ctl_init();
1911         if (res)
1912                 goto err_sysfs_cleanup;
1913
1914         sanitize_global_limit(&max_user_bgreq);
1915         sanitize_global_limit(&max_user_congthresh);
1916
1917         return 0;
1918
1919  err_sysfs_cleanup:
1920         fuse_sysfs_cleanup();
1921  err_dev_cleanup:
1922         fuse_dev_cleanup();
1923  err_fs_cleanup:
1924         fuse_fs_cleanup();
1925  err:
1926         return res;
1927 }
1928
1929 static void __exit fuse_exit(void)
1930 {
1931         pr_debug("exit\n");
1932
1933         fuse_ctl_cleanup();
1934         fuse_sysfs_cleanup();
1935         fuse_fs_cleanup();
1936         fuse_dev_cleanup();
1937 }
1938
1939 module_init(fuse_init);
1940 module_exit(fuse_exit);