fuse: wait for writepages in syncfs
[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         init_rwsem(&fi->i_mmap_sem);
89         spin_lock_init(&fi->lock);
90         fi->forget = fuse_alloc_forget();
91         if (!fi->forget)
92                 goto out_free;
93
94         if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
95                 goto out_free_forget;
96
97         return &fi->inode;
98
99 out_free_forget:
100         kfree(fi->forget);
101 out_free:
102         kmem_cache_free(fuse_inode_cachep, fi);
103         return NULL;
104 }
105
106 static void fuse_free_inode(struct inode *inode)
107 {
108         struct fuse_inode *fi = get_fuse_inode(inode);
109
110         mutex_destroy(&fi->mutex);
111         kfree(fi->forget);
112 #ifdef CONFIG_FUSE_DAX
113         kfree(fi->dax);
114 #endif
115         kmem_cache_free(fuse_inode_cachep, fi);
116 }
117
118 static void fuse_evict_inode(struct inode *inode)
119 {
120         struct fuse_inode *fi = get_fuse_inode(inode);
121
122         truncate_inode_pages_final(&inode->i_data);
123         clear_inode(inode);
124         if (inode->i_sb->s_flags & SB_ACTIVE) {
125                 struct fuse_conn *fc = get_fuse_conn(inode);
126
127                 if (FUSE_IS_DAX(inode))
128                         fuse_dax_inode_cleanup(inode);
129                 if (fi->nlookup) {
130                         fuse_queue_forget(fc, fi->forget, fi->nodeid,
131                                           fi->nlookup);
132                         fi->forget = NULL;
133                 }
134         }
135         if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
136                 WARN_ON(!list_empty(&fi->write_files));
137                 WARN_ON(!list_empty(&fi->queued_writes));
138         }
139 }
140
141 static int fuse_reconfigure(struct fs_context *fsc)
142 {
143         struct super_block *sb = fsc->root->d_sb;
144
145         sync_filesystem(sb);
146         if (fsc->sb_flags & SB_MANDLOCK)
147                 return -EINVAL;
148
149         return 0;
150 }
151
152 /*
153  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
154  * so that it will fit.
155  */
156 static ino_t fuse_squash_ino(u64 ino64)
157 {
158         ino_t ino = (ino_t) ino64;
159         if (sizeof(ino_t) < sizeof(u64))
160                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
161         return ino;
162 }
163
164 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
165                                    u64 attr_valid)
166 {
167         struct fuse_conn *fc = get_fuse_conn(inode);
168         struct fuse_inode *fi = get_fuse_inode(inode);
169
170         lockdep_assert_held(&fi->lock);
171
172         fi->attr_version = atomic64_inc_return(&fc->attr_version);
173         fi->i_time = attr_valid;
174         WRITE_ONCE(fi->inval_mask, 0);
175
176         inode->i_ino     = fuse_squash_ino(attr->ino);
177         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
178         set_nlink(inode, attr->nlink);
179         inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
180         inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
181         inode->i_blocks  = attr->blocks;
182         inode->i_atime.tv_sec   = attr->atime;
183         inode->i_atime.tv_nsec  = attr->atimensec;
184         /* mtime from server may be stale due to local buffered write */
185         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
186                 inode->i_mtime.tv_sec   = attr->mtime;
187                 inode->i_mtime.tv_nsec  = attr->mtimensec;
188                 inode->i_ctime.tv_sec   = attr->ctime;
189                 inode->i_ctime.tv_nsec  = attr->ctimensec;
190         }
191
192         if (attr->blksize != 0)
193                 inode->i_blkbits = ilog2(attr->blksize);
194         else
195                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
196
197         /*
198          * Don't set the sticky bit in i_mode, unless we want the VFS
199          * to check permissions.  This prevents failures due to the
200          * check in may_delete().
201          */
202         fi->orig_i_mode = inode->i_mode;
203         if (!fc->default_permissions)
204                 inode->i_mode &= ~S_ISVTX;
205
206         fi->orig_ino = attr->ino;
207
208         /*
209          * We are refreshing inode data and it is possible that another
210          * client set suid/sgid or security.capability xattr. So clear
211          * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
212          * was set or if security.capability xattr was set. But we don't
213          * know if security.capability has been set or not. So clear it
214          * anyway. Its less efficient but should be safe.
215          */
216         inode->i_flags &= ~S_NOSEC;
217 }
218
219 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
220                             u64 attr_valid, u64 attr_version)
221 {
222         struct fuse_conn *fc = get_fuse_conn(inode);
223         struct fuse_inode *fi = get_fuse_inode(inode);
224         bool is_wb = fc->writeback_cache;
225         loff_t oldsize;
226         struct timespec64 old_mtime;
227
228         spin_lock(&fi->lock);
229         if ((attr_version != 0 && fi->attr_version > attr_version) ||
230             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
231                 spin_unlock(&fi->lock);
232                 return;
233         }
234
235         old_mtime = inode->i_mtime;
236         fuse_change_attributes_common(inode, attr, attr_valid);
237
238         oldsize = inode->i_size;
239         /*
240          * In case of writeback_cache enabled, the cached writes beyond EOF
241          * extend local i_size without keeping userspace server in sync. So,
242          * attr->size coming from server can be stale. We cannot trust it.
243          */
244         if (!is_wb || !S_ISREG(inode->i_mode))
245                 i_size_write(inode, attr->size);
246         spin_unlock(&fi->lock);
247
248         if (!is_wb && S_ISREG(inode->i_mode)) {
249                 bool inval = false;
250
251                 if (oldsize != attr->size) {
252                         truncate_pagecache(inode, attr->size);
253                         if (!fc->explicit_inval_data)
254                                 inval = true;
255                 } else if (fc->auto_inval_data) {
256                         struct timespec64 new_mtime = {
257                                 .tv_sec = attr->mtime,
258                                 .tv_nsec = attr->mtimensec,
259                         };
260
261                         /*
262                          * Auto inval mode also checks and invalidates if mtime
263                          * has changed.
264                          */
265                         if (!timespec64_equal(&old_mtime, &new_mtime))
266                                 inval = true;
267                 }
268
269                 if (inval)
270                         invalidate_inode_pages2(inode->i_mapping);
271         }
272 }
273
274 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
275 {
276         inode->i_mode = attr->mode & S_IFMT;
277         inode->i_size = attr->size;
278         inode->i_mtime.tv_sec  = attr->mtime;
279         inode->i_mtime.tv_nsec = attr->mtimensec;
280         inode->i_ctime.tv_sec  = attr->ctime;
281         inode->i_ctime.tv_nsec = attr->ctimensec;
282         if (S_ISREG(inode->i_mode)) {
283                 fuse_init_common(inode);
284                 fuse_init_file_inode(inode);
285         } else if (S_ISDIR(inode->i_mode))
286                 fuse_init_dir(inode);
287         else if (S_ISLNK(inode->i_mode))
288                 fuse_init_symlink(inode);
289         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
290                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
291                 fuse_init_common(inode);
292                 init_special_inode(inode, inode->i_mode,
293                                    new_decode_dev(attr->rdev));
294         } else
295                 BUG();
296 }
297
298 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
299 {
300         u64 nodeid = *(u64 *) _nodeidp;
301         if (get_node_id(inode) == nodeid)
302                 return 1;
303         else
304                 return 0;
305 }
306
307 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
308 {
309         u64 nodeid = *(u64 *) _nodeidp;
310         get_fuse_inode(inode)->nodeid = nodeid;
311         return 0;
312 }
313
314 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
315                         int generation, struct fuse_attr *attr,
316                         u64 attr_valid, u64 attr_version)
317 {
318         struct inode *inode;
319         struct fuse_inode *fi;
320         struct fuse_conn *fc = get_fuse_conn_super(sb);
321
322         /*
323          * Auto mount points get their node id from the submount root, which is
324          * not a unique identifier within this filesystem.
325          *
326          * To avoid conflicts, do not place submount points into the inode hash
327          * table.
328          */
329         if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
330             S_ISDIR(attr->mode)) {
331                 inode = new_inode(sb);
332                 if (!inode)
333                         return NULL;
334
335                 fuse_init_inode(inode, attr);
336                 get_fuse_inode(inode)->nodeid = nodeid;
337                 inode->i_flags |= S_AUTOMOUNT;
338                 goto done;
339         }
340
341 retry:
342         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
343         if (!inode)
344                 return NULL;
345
346         if ((inode->i_state & I_NEW)) {
347                 inode->i_flags |= S_NOATIME;
348                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
349                         inode->i_flags |= S_NOCMTIME;
350                 inode->i_generation = generation;
351                 fuse_init_inode(inode, attr);
352                 unlock_new_inode(inode);
353         } else if (fuse_stale_inode(inode, generation, attr)) {
354                 /* nodeid was reused, any I/O on the old inode should fail */
355                 fuse_make_bad(inode);
356                 iput(inode);
357                 goto retry;
358         }
359 done:
360         fi = get_fuse_inode(inode);
361         spin_lock(&fi->lock);
362         fi->nlookup++;
363         spin_unlock(&fi->lock);
364         fuse_change_attributes(inode, attr, attr_valid, attr_version);
365
366         return inode;
367 }
368
369 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
370                            struct fuse_mount **fm)
371 {
372         struct fuse_mount *fm_iter;
373         struct inode *inode;
374
375         WARN_ON(!rwsem_is_locked(&fc->killsb));
376         list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
377                 if (!fm_iter->sb)
378                         continue;
379
380                 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
381                 if (inode) {
382                         if (fm)
383                                 *fm = fm_iter;
384                         return inode;
385                 }
386         }
387
388         return NULL;
389 }
390
391 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
392                              loff_t offset, loff_t len)
393 {
394         struct fuse_inode *fi;
395         struct inode *inode;
396         pgoff_t pg_start;
397         pgoff_t pg_end;
398
399         inode = fuse_ilookup(fc, nodeid, NULL);
400         if (!inode)
401                 return -ENOENT;
402
403         fi = get_fuse_inode(inode);
404         spin_lock(&fi->lock);
405         fi->attr_version = atomic64_inc_return(&fc->attr_version);
406         spin_unlock(&fi->lock);
407
408         fuse_invalidate_attr(inode);
409         forget_all_cached_acls(inode);
410         if (offset >= 0) {
411                 pg_start = offset >> PAGE_SHIFT;
412                 if (len <= 0)
413                         pg_end = -1;
414                 else
415                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
416                 invalidate_inode_pages2_range(inode->i_mapping,
417                                               pg_start, pg_end);
418         }
419         iput(inode);
420         return 0;
421 }
422
423 bool fuse_lock_inode(struct inode *inode)
424 {
425         bool locked = false;
426
427         if (!get_fuse_conn(inode)->parallel_dirops) {
428                 mutex_lock(&get_fuse_inode(inode)->mutex);
429                 locked = true;
430         }
431
432         return locked;
433 }
434
435 void fuse_unlock_inode(struct inode *inode, bool locked)
436 {
437         if (locked)
438                 mutex_unlock(&get_fuse_inode(inode)->mutex);
439 }
440
441 static void fuse_umount_begin(struct super_block *sb)
442 {
443         struct fuse_conn *fc = get_fuse_conn_super(sb);
444
445         if (!fc->no_force_umount)
446                 fuse_abort_conn(fc);
447 }
448
449 static void fuse_send_destroy(struct fuse_mount *fm)
450 {
451         if (fm->fc->conn_init) {
452                 FUSE_ARGS(args);
453
454                 args.opcode = FUSE_DESTROY;
455                 args.force = true;
456                 args.nocreds = true;
457                 fuse_simple_request(fm, &args);
458         }
459 }
460
461 static void fuse_put_super(struct super_block *sb)
462 {
463         struct fuse_mount *fm = get_fuse_mount_super(sb);
464
465         fuse_conn_put(fm->fc);
466         kfree(fm);
467 }
468
469 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
470 {
471         stbuf->f_type    = FUSE_SUPER_MAGIC;
472         stbuf->f_bsize   = attr->bsize;
473         stbuf->f_frsize  = attr->frsize;
474         stbuf->f_blocks  = attr->blocks;
475         stbuf->f_bfree   = attr->bfree;
476         stbuf->f_bavail  = attr->bavail;
477         stbuf->f_files   = attr->files;
478         stbuf->f_ffree   = attr->ffree;
479         stbuf->f_namelen = attr->namelen;
480         /* fsid is left zero */
481 }
482
483 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
484 {
485         struct super_block *sb = dentry->d_sb;
486         struct fuse_mount *fm = get_fuse_mount_super(sb);
487         FUSE_ARGS(args);
488         struct fuse_statfs_out outarg;
489         int err;
490
491         if (!fuse_allow_current_process(fm->fc)) {
492                 buf->f_type = FUSE_SUPER_MAGIC;
493                 return 0;
494         }
495
496         memset(&outarg, 0, sizeof(outarg));
497         args.in_numargs = 0;
498         args.opcode = FUSE_STATFS;
499         args.nodeid = get_node_id(d_inode(dentry));
500         args.out_numargs = 1;
501         args.out_args[0].size = sizeof(outarg);
502         args.out_args[0].value = &outarg;
503         err = fuse_simple_request(fm, &args);
504         if (!err)
505                 convert_fuse_statfs(buf, &outarg.st);
506         return err;
507 }
508
509 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
510 {
511         struct fuse_sync_bucket *bucket;
512
513         bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
514         if (bucket) {
515                 init_waitqueue_head(&bucket->waitq);
516                 /* Initial active count */
517                 atomic_set(&bucket->count, 1);
518         }
519         return bucket;
520 }
521
522 static void fuse_sync_fs_writes(struct fuse_conn *fc)
523 {
524         struct fuse_sync_bucket *bucket, *new_bucket;
525         int count;
526
527         new_bucket = fuse_sync_bucket_alloc();
528         spin_lock(&fc->lock);
529         bucket = rcu_dereference_protected(fc->curr_bucket, 1);
530         count = atomic_read(&bucket->count);
531         WARN_ON(count < 1);
532         /* No outstanding writes? */
533         if (count == 1) {
534                 spin_unlock(&fc->lock);
535                 kfree(new_bucket);
536                 return;
537         }
538
539         /*
540          * Completion of new bucket depends on completion of this bucket, so add
541          * one more count.
542          */
543         atomic_inc(&new_bucket->count);
544         rcu_assign_pointer(fc->curr_bucket, new_bucket);
545         spin_unlock(&fc->lock);
546         /*
547          * Drop initial active count.  At this point if all writes in this and
548          * ancestor buckets complete, the count will go to zero and this task
549          * will be woken up.
550          */
551         atomic_dec(&bucket->count);
552
553         wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
554
555         /* Drop temp count on descendant bucket */
556         fuse_sync_bucket_dec(new_bucket);
557         kfree_rcu(bucket, rcu);
558 }
559
560 static int fuse_sync_fs(struct super_block *sb, int wait)
561 {
562         struct fuse_mount *fm = get_fuse_mount_super(sb);
563         struct fuse_conn *fc = fm->fc;
564         struct fuse_syncfs_in inarg;
565         FUSE_ARGS(args);
566         int err;
567
568         /*
569          * Userspace cannot handle the wait == 0 case.  Avoid a
570          * gratuitous roundtrip.
571          */
572         if (!wait)
573                 return 0;
574
575         /* The filesystem is being unmounted.  Nothing to do. */
576         if (!sb->s_root)
577                 return 0;
578
579         if (!fc->sync_fs)
580                 return 0;
581
582         fuse_sync_fs_writes(fc);
583
584         memset(&inarg, 0, sizeof(inarg));
585         args.in_numargs = 1;
586         args.in_args[0].size = sizeof(inarg);
587         args.in_args[0].value = &inarg;
588         args.opcode = FUSE_SYNCFS;
589         args.nodeid = get_node_id(sb->s_root->d_inode);
590         args.out_numargs = 0;
591
592         err = fuse_simple_request(fm, &args);
593         if (err == -ENOSYS) {
594                 fc->sync_fs = 0;
595                 err = 0;
596         }
597
598         return err;
599 }
600
601 enum {
602         OPT_SOURCE,
603         OPT_SUBTYPE,
604         OPT_FD,
605         OPT_ROOTMODE,
606         OPT_USER_ID,
607         OPT_GROUP_ID,
608         OPT_DEFAULT_PERMISSIONS,
609         OPT_ALLOW_OTHER,
610         OPT_MAX_READ,
611         OPT_BLKSIZE,
612         OPT_ERR
613 };
614
615 static const struct fs_parameter_spec fuse_fs_parameters[] = {
616         fsparam_string  ("source",              OPT_SOURCE),
617         fsparam_u32     ("fd",                  OPT_FD),
618         fsparam_u32oct  ("rootmode",            OPT_ROOTMODE),
619         fsparam_u32     ("user_id",             OPT_USER_ID),
620         fsparam_u32     ("group_id",            OPT_GROUP_ID),
621         fsparam_flag    ("default_permissions", OPT_DEFAULT_PERMISSIONS),
622         fsparam_flag    ("allow_other",         OPT_ALLOW_OTHER),
623         fsparam_u32     ("max_read",            OPT_MAX_READ),
624         fsparam_u32     ("blksize",             OPT_BLKSIZE),
625         fsparam_string  ("subtype",             OPT_SUBTYPE),
626         {}
627 };
628
629 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
630 {
631         struct fs_parse_result result;
632         struct fuse_fs_context *ctx = fsc->fs_private;
633         int opt;
634
635         if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
636                 /*
637                  * Ignore options coming from mount(MS_REMOUNT) for backward
638                  * compatibility.
639                  */
640                 if (fsc->oldapi)
641                         return 0;
642
643                 return invalfc(fsc, "No changes allowed in reconfigure");
644         }
645
646         opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
647         if (opt < 0)
648                 return opt;
649
650         switch (opt) {
651         case OPT_SOURCE:
652                 if (fsc->source)
653                         return invalfc(fsc, "Multiple sources specified");
654                 fsc->source = param->string;
655                 param->string = NULL;
656                 break;
657
658         case OPT_SUBTYPE:
659                 if (ctx->subtype)
660                         return invalfc(fsc, "Multiple subtypes specified");
661                 ctx->subtype = param->string;
662                 param->string = NULL;
663                 return 0;
664
665         case OPT_FD:
666                 ctx->fd = result.uint_32;
667                 ctx->fd_present = true;
668                 break;
669
670         case OPT_ROOTMODE:
671                 if (!fuse_valid_type(result.uint_32))
672                         return invalfc(fsc, "Invalid rootmode");
673                 ctx->rootmode = result.uint_32;
674                 ctx->rootmode_present = true;
675                 break;
676
677         case OPT_USER_ID:
678                 ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
679                 if (!uid_valid(ctx->user_id))
680                         return invalfc(fsc, "Invalid user_id");
681                 ctx->user_id_present = true;
682                 break;
683
684         case OPT_GROUP_ID:
685                 ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
686                 if (!gid_valid(ctx->group_id))
687                         return invalfc(fsc, "Invalid group_id");
688                 ctx->group_id_present = true;
689                 break;
690
691         case OPT_DEFAULT_PERMISSIONS:
692                 ctx->default_permissions = true;
693                 break;
694
695         case OPT_ALLOW_OTHER:
696                 ctx->allow_other = true;
697                 break;
698
699         case OPT_MAX_READ:
700                 ctx->max_read = result.uint_32;
701                 break;
702
703         case OPT_BLKSIZE:
704                 if (!ctx->is_bdev)
705                         return invalfc(fsc, "blksize only supported for fuseblk");
706                 ctx->blksize = result.uint_32;
707                 break;
708
709         default:
710                 return -EINVAL;
711         }
712
713         return 0;
714 }
715
716 static void fuse_free_fsc(struct fs_context *fsc)
717 {
718         struct fuse_fs_context *ctx = fsc->fs_private;
719
720         if (ctx) {
721                 kfree(ctx->subtype);
722                 kfree(ctx);
723         }
724 }
725
726 static int fuse_show_options(struct seq_file *m, struct dentry *root)
727 {
728         struct super_block *sb = root->d_sb;
729         struct fuse_conn *fc = get_fuse_conn_super(sb);
730
731         if (fc->legacy_opts_show) {
732                 seq_printf(m, ",user_id=%u",
733                            from_kuid_munged(fc->user_ns, fc->user_id));
734                 seq_printf(m, ",group_id=%u",
735                            from_kgid_munged(fc->user_ns, fc->group_id));
736                 if (fc->default_permissions)
737                         seq_puts(m, ",default_permissions");
738                 if (fc->allow_other)
739                         seq_puts(m, ",allow_other");
740                 if (fc->max_read != ~0)
741                         seq_printf(m, ",max_read=%u", fc->max_read);
742                 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
743                         seq_printf(m, ",blksize=%lu", sb->s_blocksize);
744         }
745 #ifdef CONFIG_FUSE_DAX
746         if (fc->dax)
747                 seq_puts(m, ",dax");
748 #endif
749
750         return 0;
751 }
752
753 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
754                              const struct fuse_iqueue_ops *ops,
755                              void *priv)
756 {
757         memset(fiq, 0, sizeof(struct fuse_iqueue));
758         spin_lock_init(&fiq->lock);
759         init_waitqueue_head(&fiq->waitq);
760         INIT_LIST_HEAD(&fiq->pending);
761         INIT_LIST_HEAD(&fiq->interrupts);
762         fiq->forget_list_tail = &fiq->forget_list_head;
763         fiq->connected = 1;
764         fiq->ops = ops;
765         fiq->priv = priv;
766 }
767
768 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
769 {
770         unsigned int i;
771
772         spin_lock_init(&fpq->lock);
773         for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
774                 INIT_LIST_HEAD(&fpq->processing[i]);
775         INIT_LIST_HEAD(&fpq->io);
776         fpq->connected = 1;
777 }
778
779 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
780                     struct user_namespace *user_ns,
781                     const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
782 {
783         memset(fc, 0, sizeof(*fc));
784         spin_lock_init(&fc->lock);
785         spin_lock_init(&fc->bg_lock);
786         init_rwsem(&fc->killsb);
787         refcount_set(&fc->count, 1);
788         atomic_set(&fc->dev_count, 1);
789         init_waitqueue_head(&fc->blocked_waitq);
790         fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
791         INIT_LIST_HEAD(&fc->bg_queue);
792         INIT_LIST_HEAD(&fc->entry);
793         INIT_LIST_HEAD(&fc->devices);
794         atomic_set(&fc->num_waiting, 0);
795         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
796         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
797         atomic64_set(&fc->khctr, 0);
798         fc->polled_files = RB_ROOT;
799         fc->blocked = 0;
800         fc->initialized = 0;
801         fc->connected = 1;
802         atomic64_set(&fc->attr_version, 1);
803         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
804         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
805         fc->user_ns = get_user_ns(user_ns);
806         fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
807         fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
808
809         INIT_LIST_HEAD(&fc->mounts);
810         list_add(&fm->fc_entry, &fc->mounts);
811         fm->fc = fc;
812 }
813 EXPORT_SYMBOL_GPL(fuse_conn_init);
814
815 void fuse_conn_put(struct fuse_conn *fc)
816 {
817         if (refcount_dec_and_test(&fc->count)) {
818                 struct fuse_iqueue *fiq = &fc->iq;
819                 struct fuse_sync_bucket *bucket;
820
821                 if (IS_ENABLED(CONFIG_FUSE_DAX))
822                         fuse_dax_conn_free(fc);
823                 if (fiq->ops->release)
824                         fiq->ops->release(fiq);
825                 put_pid_ns(fc->pid_ns);
826                 put_user_ns(fc->user_ns);
827                 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
828                 if (bucket) {
829                         WARN_ON(atomic_read(&bucket->count) != 1);
830                         kfree(bucket);
831                 }
832                 fc->release(fc);
833         }
834 }
835 EXPORT_SYMBOL_GPL(fuse_conn_put);
836
837 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
838 {
839         refcount_inc(&fc->count);
840         return fc;
841 }
842 EXPORT_SYMBOL_GPL(fuse_conn_get);
843
844 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
845 {
846         struct fuse_attr attr;
847         memset(&attr, 0, sizeof(attr));
848
849         attr.mode = mode;
850         attr.ino = FUSE_ROOT_ID;
851         attr.nlink = 1;
852         return fuse_iget(sb, 1, 0, &attr, 0, 0);
853 }
854
855 struct fuse_inode_handle {
856         u64 nodeid;
857         u32 generation;
858 };
859
860 static struct dentry *fuse_get_dentry(struct super_block *sb,
861                                       struct fuse_inode_handle *handle)
862 {
863         struct fuse_conn *fc = get_fuse_conn_super(sb);
864         struct inode *inode;
865         struct dentry *entry;
866         int err = -ESTALE;
867
868         if (handle->nodeid == 0)
869                 goto out_err;
870
871         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
872         if (!inode) {
873                 struct fuse_entry_out outarg;
874                 const struct qstr name = QSTR_INIT(".", 1);
875
876                 if (!fc->export_support)
877                         goto out_err;
878
879                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
880                                        &inode);
881                 if (err && err != -ENOENT)
882                         goto out_err;
883                 if (err || !inode) {
884                         err = -ESTALE;
885                         goto out_err;
886                 }
887                 err = -EIO;
888                 if (get_node_id(inode) != handle->nodeid)
889                         goto out_iput;
890         }
891         err = -ESTALE;
892         if (inode->i_generation != handle->generation)
893                 goto out_iput;
894
895         entry = d_obtain_alias(inode);
896         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
897                 fuse_invalidate_entry_cache(entry);
898
899         return entry;
900
901  out_iput:
902         iput(inode);
903  out_err:
904         return ERR_PTR(err);
905 }
906
907 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
908                            struct inode *parent)
909 {
910         int len = parent ? 6 : 3;
911         u64 nodeid;
912         u32 generation;
913
914         if (*max_len < len) {
915                 *max_len = len;
916                 return  FILEID_INVALID;
917         }
918
919         nodeid = get_fuse_inode(inode)->nodeid;
920         generation = inode->i_generation;
921
922         fh[0] = (u32)(nodeid >> 32);
923         fh[1] = (u32)(nodeid & 0xffffffff);
924         fh[2] = generation;
925
926         if (parent) {
927                 nodeid = get_fuse_inode(parent)->nodeid;
928                 generation = parent->i_generation;
929
930                 fh[3] = (u32)(nodeid >> 32);
931                 fh[4] = (u32)(nodeid & 0xffffffff);
932                 fh[5] = generation;
933         }
934
935         *max_len = len;
936         return parent ? 0x82 : 0x81;
937 }
938
939 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
940                 struct fid *fid, int fh_len, int fh_type)
941 {
942         struct fuse_inode_handle handle;
943
944         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
945                 return NULL;
946
947         handle.nodeid = (u64) fid->raw[0] << 32;
948         handle.nodeid |= (u64) fid->raw[1];
949         handle.generation = fid->raw[2];
950         return fuse_get_dentry(sb, &handle);
951 }
952
953 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
954                 struct fid *fid, int fh_len, int fh_type)
955 {
956         struct fuse_inode_handle parent;
957
958         if (fh_type != 0x82 || fh_len < 6)
959                 return NULL;
960
961         parent.nodeid = (u64) fid->raw[3] << 32;
962         parent.nodeid |= (u64) fid->raw[4];
963         parent.generation = fid->raw[5];
964         return fuse_get_dentry(sb, &parent);
965 }
966
967 static struct dentry *fuse_get_parent(struct dentry *child)
968 {
969         struct inode *child_inode = d_inode(child);
970         struct fuse_conn *fc = get_fuse_conn(child_inode);
971         struct inode *inode;
972         struct dentry *parent;
973         struct fuse_entry_out outarg;
974         int err;
975
976         if (!fc->export_support)
977                 return ERR_PTR(-ESTALE);
978
979         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
980                                &dotdot_name, &outarg, &inode);
981         if (err) {
982                 if (err == -ENOENT)
983                         return ERR_PTR(-ESTALE);
984                 return ERR_PTR(err);
985         }
986
987         parent = d_obtain_alias(inode);
988         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
989                 fuse_invalidate_entry_cache(parent);
990
991         return parent;
992 }
993
994 static const struct export_operations fuse_export_operations = {
995         .fh_to_dentry   = fuse_fh_to_dentry,
996         .fh_to_parent   = fuse_fh_to_parent,
997         .encode_fh      = fuse_encode_fh,
998         .get_parent     = fuse_get_parent,
999 };
1000
1001 static const struct super_operations fuse_super_operations = {
1002         .alloc_inode    = fuse_alloc_inode,
1003         .free_inode     = fuse_free_inode,
1004         .evict_inode    = fuse_evict_inode,
1005         .write_inode    = fuse_write_inode,
1006         .drop_inode     = generic_delete_inode,
1007         .put_super      = fuse_put_super,
1008         .umount_begin   = fuse_umount_begin,
1009         .statfs         = fuse_statfs,
1010         .sync_fs        = fuse_sync_fs,
1011         .show_options   = fuse_show_options,
1012 };
1013
1014 static void sanitize_global_limit(unsigned *limit)
1015 {
1016         /*
1017          * The default maximum number of async requests is calculated to consume
1018          * 1/2^13 of the total memory, assuming 392 bytes per request.
1019          */
1020         if (*limit == 0)
1021                 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1022
1023         if (*limit >= 1 << 16)
1024                 *limit = (1 << 16) - 1;
1025 }
1026
1027 static int set_global_limit(const char *val, const struct kernel_param *kp)
1028 {
1029         int rv;
1030
1031         rv = param_set_uint(val, kp);
1032         if (rv)
1033                 return rv;
1034
1035         sanitize_global_limit((unsigned *)kp->arg);
1036
1037         return 0;
1038 }
1039
1040 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1041 {
1042         int cap_sys_admin = capable(CAP_SYS_ADMIN);
1043
1044         if (arg->minor < 13)
1045                 return;
1046
1047         sanitize_global_limit(&max_user_bgreq);
1048         sanitize_global_limit(&max_user_congthresh);
1049
1050         spin_lock(&fc->bg_lock);
1051         if (arg->max_background) {
1052                 fc->max_background = arg->max_background;
1053
1054                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1055                         fc->max_background = max_user_bgreq;
1056         }
1057         if (arg->congestion_threshold) {
1058                 fc->congestion_threshold = arg->congestion_threshold;
1059
1060                 if (!cap_sys_admin &&
1061                     fc->congestion_threshold > max_user_congthresh)
1062                         fc->congestion_threshold = max_user_congthresh;
1063         }
1064         spin_unlock(&fc->bg_lock);
1065 }
1066
1067 struct fuse_init_args {
1068         struct fuse_args args;
1069         struct fuse_init_in in;
1070         struct fuse_init_out out;
1071 };
1072
1073 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1074                                int error)
1075 {
1076         struct fuse_conn *fc = fm->fc;
1077         struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1078         struct fuse_init_out *arg = &ia->out;
1079         bool ok = true;
1080
1081         if (error || arg->major != FUSE_KERNEL_VERSION)
1082                 ok = false;
1083         else {
1084                 unsigned long ra_pages;
1085
1086                 process_init_limits(fc, arg);
1087
1088                 if (arg->minor >= 6) {
1089                         ra_pages = arg->max_readahead / PAGE_SIZE;
1090                         if (arg->flags & FUSE_ASYNC_READ)
1091                                 fc->async_read = 1;
1092                         if (!(arg->flags & FUSE_POSIX_LOCKS))
1093                                 fc->no_lock = 1;
1094                         if (arg->minor >= 17) {
1095                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
1096                                         fc->no_flock = 1;
1097                         } else {
1098                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
1099                                         fc->no_flock = 1;
1100                         }
1101                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
1102                                 fc->atomic_o_trunc = 1;
1103                         if (arg->minor >= 9) {
1104                                 /* LOOKUP has dependency on proto version */
1105                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
1106                                         fc->export_support = 1;
1107                         }
1108                         if (arg->flags & FUSE_BIG_WRITES)
1109                                 fc->big_writes = 1;
1110                         if (arg->flags & FUSE_DONT_MASK)
1111                                 fc->dont_mask = 1;
1112                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
1113                                 fc->auto_inval_data = 1;
1114                         else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
1115                                 fc->explicit_inval_data = 1;
1116                         if (arg->flags & FUSE_DO_READDIRPLUS) {
1117                                 fc->do_readdirplus = 1;
1118                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
1119                                         fc->readdirplus_auto = 1;
1120                         }
1121                         if (arg->flags & FUSE_ASYNC_DIO)
1122                                 fc->async_dio = 1;
1123                         if (arg->flags & FUSE_WRITEBACK_CACHE)
1124                                 fc->writeback_cache = 1;
1125                         if (arg->flags & FUSE_PARALLEL_DIROPS)
1126                                 fc->parallel_dirops = 1;
1127                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
1128                                 fc->handle_killpriv = 1;
1129                         if (arg->time_gran && arg->time_gran <= 1000000000)
1130                                 fm->sb->s_time_gran = arg->time_gran;
1131                         if ((arg->flags & FUSE_POSIX_ACL)) {
1132                                 fc->default_permissions = 1;
1133                                 fc->posix_acl = 1;
1134                                 fm->sb->s_xattr = fuse_acl_xattr_handlers;
1135                         }
1136                         if (arg->flags & FUSE_CACHE_SYMLINKS)
1137                                 fc->cache_symlinks = 1;
1138                         if (arg->flags & FUSE_ABORT_ERROR)
1139                                 fc->abort_err = 1;
1140                         if (arg->flags & FUSE_MAX_PAGES) {
1141                                 fc->max_pages =
1142                                         min_t(unsigned int, fc->max_pages_limit,
1143                                         max_t(unsigned int, arg->max_pages, 1));
1144                         }
1145                         if (IS_ENABLED(CONFIG_FUSE_DAX) &&
1146                             arg->flags & FUSE_MAP_ALIGNMENT &&
1147                             !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1148                                 ok = false;
1149                         }
1150                         if (arg->flags & FUSE_HANDLE_KILLPRIV_V2) {
1151                                 fc->handle_killpriv_v2 = 1;
1152                                 fm->sb->s_flags |= SB_NOSEC;
1153                         }
1154                         if (arg->flags & FUSE_SETXATTR_EXT)
1155                                 fc->setxattr_ext = 1;
1156                 } else {
1157                         ra_pages = fc->max_read / PAGE_SIZE;
1158                         fc->no_lock = 1;
1159                         fc->no_flock = 1;
1160                 }
1161
1162                 fm->sb->s_bdi->ra_pages =
1163                                 min(fm->sb->s_bdi->ra_pages, ra_pages);
1164                 fc->minor = arg->minor;
1165                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1166                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1167                 fc->conn_init = 1;
1168         }
1169         kfree(ia);
1170
1171         if (!ok) {
1172                 fc->conn_init = 0;
1173                 fc->conn_error = 1;
1174         }
1175
1176         fuse_set_initialized(fc);
1177         wake_up_all(&fc->blocked_waitq);
1178 }
1179
1180 void fuse_send_init(struct fuse_mount *fm)
1181 {
1182         struct fuse_init_args *ia;
1183
1184         ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1185
1186         ia->in.major = FUSE_KERNEL_VERSION;
1187         ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1188         ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1189         ia->in.flags |=
1190                 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1191                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1192                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1193                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1194                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1195                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1196                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1197                 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1198                 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1199                 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT;
1200 #ifdef CONFIG_FUSE_DAX
1201         if (fm->fc->dax)
1202                 ia->in.flags |= FUSE_MAP_ALIGNMENT;
1203 #endif
1204         if (fm->fc->auto_submounts)
1205                 ia->in.flags |= FUSE_SUBMOUNTS;
1206
1207         ia->args.opcode = FUSE_INIT;
1208         ia->args.in_numargs = 1;
1209         ia->args.in_args[0].size = sizeof(ia->in);
1210         ia->args.in_args[0].value = &ia->in;
1211         ia->args.out_numargs = 1;
1212         /* Variable length argument used for backward compatibility
1213            with interface version < 7.5.  Rest of init_out is zeroed
1214            by do_get_request(), so a short reply is not a problem */
1215         ia->args.out_argvar = true;
1216         ia->args.out_args[0].size = sizeof(ia->out);
1217         ia->args.out_args[0].value = &ia->out;
1218         ia->args.force = true;
1219         ia->args.nocreds = true;
1220         ia->args.end = process_init_reply;
1221
1222         if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1223                 process_init_reply(fm, &ia->args, -ENOTCONN);
1224 }
1225 EXPORT_SYMBOL_GPL(fuse_send_init);
1226
1227 void fuse_free_conn(struct fuse_conn *fc)
1228 {
1229         WARN_ON(!list_empty(&fc->devices));
1230         kfree_rcu(fc, rcu);
1231 }
1232 EXPORT_SYMBOL_GPL(fuse_free_conn);
1233
1234 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1235 {
1236         int err;
1237         char *suffix = "";
1238
1239         if (sb->s_bdev) {
1240                 suffix = "-fuseblk";
1241                 /*
1242                  * sb->s_bdi points to blkdev's bdi however we want to redirect
1243                  * it to our private bdi...
1244                  */
1245                 bdi_put(sb->s_bdi);
1246                 sb->s_bdi = &noop_backing_dev_info;
1247         }
1248         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1249                                    MINOR(fc->dev), suffix);
1250         if (err)
1251                 return err;
1252
1253         /* fuse does it's own writeback accounting */
1254         sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1255         sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1256
1257         /*
1258          * For a single fuse filesystem use max 1% of dirty +
1259          * writeback threshold.
1260          *
1261          * This gives about 1M of write buffer for memory maps on a
1262          * machine with 1G and 10% dirty_ratio, which should be more
1263          * than enough.
1264          *
1265          * Privileged users can raise it by writing to
1266          *
1267          *    /sys/class/bdi/<bdi>/max_ratio
1268          */
1269         bdi_set_max_ratio(sb->s_bdi, 1);
1270
1271         return 0;
1272 }
1273
1274 struct fuse_dev *fuse_dev_alloc(void)
1275 {
1276         struct fuse_dev *fud;
1277         struct list_head *pq;
1278
1279         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1280         if (!fud)
1281                 return NULL;
1282
1283         pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1284         if (!pq) {
1285                 kfree(fud);
1286                 return NULL;
1287         }
1288
1289         fud->pq.processing = pq;
1290         fuse_pqueue_init(&fud->pq);
1291
1292         return fud;
1293 }
1294 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1295
1296 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1297 {
1298         fud->fc = fuse_conn_get(fc);
1299         spin_lock(&fc->lock);
1300         list_add_tail(&fud->entry, &fc->devices);
1301         spin_unlock(&fc->lock);
1302 }
1303 EXPORT_SYMBOL_GPL(fuse_dev_install);
1304
1305 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1306 {
1307         struct fuse_dev *fud;
1308
1309         fud = fuse_dev_alloc();
1310         if (!fud)
1311                 return NULL;
1312
1313         fuse_dev_install(fud, fc);
1314         return fud;
1315 }
1316 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1317
1318 void fuse_dev_free(struct fuse_dev *fud)
1319 {
1320         struct fuse_conn *fc = fud->fc;
1321
1322         if (fc) {
1323                 spin_lock(&fc->lock);
1324                 list_del(&fud->entry);
1325                 spin_unlock(&fc->lock);
1326
1327                 fuse_conn_put(fc);
1328         }
1329         kfree(fud->pq.processing);
1330         kfree(fud);
1331 }
1332 EXPORT_SYMBOL_GPL(fuse_dev_free);
1333
1334 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1335                                       const struct fuse_inode *fi)
1336 {
1337         *attr = (struct fuse_attr){
1338                 .ino            = fi->inode.i_ino,
1339                 .size           = fi->inode.i_size,
1340                 .blocks         = fi->inode.i_blocks,
1341                 .atime          = fi->inode.i_atime.tv_sec,
1342                 .mtime          = fi->inode.i_mtime.tv_sec,
1343                 .ctime          = fi->inode.i_ctime.tv_sec,
1344                 .atimensec      = fi->inode.i_atime.tv_nsec,
1345                 .mtimensec      = fi->inode.i_mtime.tv_nsec,
1346                 .ctimensec      = fi->inode.i_ctime.tv_nsec,
1347                 .mode           = fi->inode.i_mode,
1348                 .nlink          = fi->inode.i_nlink,
1349                 .uid            = fi->inode.i_uid.val,
1350                 .gid            = fi->inode.i_gid.val,
1351                 .rdev           = fi->inode.i_rdev,
1352                 .blksize        = 1u << fi->inode.i_blkbits,
1353         };
1354 }
1355
1356 static void fuse_sb_defaults(struct super_block *sb)
1357 {
1358         sb->s_magic = FUSE_SUPER_MAGIC;
1359         sb->s_op = &fuse_super_operations;
1360         sb->s_xattr = fuse_xattr_handlers;
1361         sb->s_maxbytes = MAX_LFS_FILESIZE;
1362         sb->s_time_gran = 1;
1363         sb->s_export_op = &fuse_export_operations;
1364         sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1365         if (sb->s_user_ns != &init_user_ns)
1366                 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1367         sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1368
1369         /*
1370          * If we are not in the initial user namespace posix
1371          * acls must be translated.
1372          */
1373         if (sb->s_user_ns != &init_user_ns)
1374                 sb->s_xattr = fuse_no_acl_xattr_handlers;
1375 }
1376
1377 static int fuse_fill_super_submount(struct super_block *sb,
1378                                     struct fuse_inode *parent_fi)
1379 {
1380         struct fuse_mount *fm = get_fuse_mount_super(sb);
1381         struct super_block *parent_sb = parent_fi->inode.i_sb;
1382         struct fuse_attr root_attr;
1383         struct inode *root;
1384
1385         fuse_sb_defaults(sb);
1386         fm->sb = sb;
1387
1388         WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1389         sb->s_bdi = bdi_get(parent_sb->s_bdi);
1390
1391         sb->s_xattr = parent_sb->s_xattr;
1392         sb->s_time_gran = parent_sb->s_time_gran;
1393         sb->s_blocksize = parent_sb->s_blocksize;
1394         sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1395         sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1396         if (parent_sb->s_subtype && !sb->s_subtype)
1397                 return -ENOMEM;
1398
1399         fuse_fill_attr_from_inode(&root_attr, parent_fi);
1400         root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1401         /*
1402          * This inode is just a duplicate, so it is not looked up and
1403          * its nlookup should not be incremented.  fuse_iget() does
1404          * that, though, so undo it here.
1405          */
1406         get_fuse_inode(root)->nlookup--;
1407         sb->s_d_op = &fuse_dentry_operations;
1408         sb->s_root = d_make_root(root);
1409         if (!sb->s_root)
1410                 return -ENOMEM;
1411
1412         return 0;
1413 }
1414
1415 /* Filesystem context private data holds the FUSE inode of the mount point */
1416 static int fuse_get_tree_submount(struct fs_context *fsc)
1417 {
1418         struct fuse_mount *fm;
1419         struct fuse_inode *mp_fi = fsc->fs_private;
1420         struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1421         struct super_block *sb;
1422         int err;
1423
1424         fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1425         if (!fm)
1426                 return -ENOMEM;
1427
1428         fsc->s_fs_info = fm;
1429         sb = sget_fc(fsc, NULL, set_anon_super_fc);
1430         if (IS_ERR(sb)) {
1431                 kfree(fm);
1432                 return PTR_ERR(sb);
1433         }
1434         fm->fc = fuse_conn_get(fc);
1435
1436         /* Initialize superblock, making @mp_fi its root */
1437         err = fuse_fill_super_submount(sb, mp_fi);
1438         if (err) {
1439                 fuse_conn_put(fc);
1440                 kfree(fm);
1441                 sb->s_fs_info = NULL;
1442                 deactivate_locked_super(sb);
1443                 return err;
1444         }
1445
1446         down_write(&fc->killsb);
1447         list_add_tail(&fm->fc_entry, &fc->mounts);
1448         up_write(&fc->killsb);
1449
1450         sb->s_flags |= SB_ACTIVE;
1451         fsc->root = dget(sb->s_root);
1452
1453         return 0;
1454 }
1455
1456 static const struct fs_context_operations fuse_context_submount_ops = {
1457         .get_tree       = fuse_get_tree_submount,
1458 };
1459
1460 int fuse_init_fs_context_submount(struct fs_context *fsc)
1461 {
1462         fsc->ops = &fuse_context_submount_ops;
1463         return 0;
1464 }
1465 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1466
1467 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1468 {
1469         struct fuse_dev *fud = NULL;
1470         struct fuse_mount *fm = get_fuse_mount_super(sb);
1471         struct fuse_conn *fc = fm->fc;
1472         struct inode *root;
1473         struct dentry *root_dentry;
1474         int err;
1475
1476         err = -EINVAL;
1477         if (sb->s_flags & SB_MANDLOCK)
1478                 goto err;
1479
1480         rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1481         fuse_sb_defaults(sb);
1482
1483         if (ctx->is_bdev) {
1484 #ifdef CONFIG_BLOCK
1485                 err = -EINVAL;
1486                 if (!sb_set_blocksize(sb, ctx->blksize))
1487                         goto err;
1488 #endif
1489         } else {
1490                 sb->s_blocksize = PAGE_SIZE;
1491                 sb->s_blocksize_bits = PAGE_SHIFT;
1492         }
1493
1494         sb->s_subtype = ctx->subtype;
1495         ctx->subtype = NULL;
1496         if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1497                 err = fuse_dax_conn_alloc(fc, ctx->dax_dev);
1498                 if (err)
1499                         goto err;
1500         }
1501
1502         if (ctx->fudptr) {
1503                 err = -ENOMEM;
1504                 fud = fuse_dev_alloc_install(fc);
1505                 if (!fud)
1506                         goto err_free_dax;
1507         }
1508
1509         fc->dev = sb->s_dev;
1510         fm->sb = sb;
1511         err = fuse_bdi_init(fc, sb);
1512         if (err)
1513                 goto err_dev_free;
1514
1515         /* Handle umasking inside the fuse code */
1516         if (sb->s_flags & SB_POSIXACL)
1517                 fc->dont_mask = 1;
1518         sb->s_flags |= SB_POSIXACL;
1519
1520         fc->default_permissions = ctx->default_permissions;
1521         fc->allow_other = ctx->allow_other;
1522         fc->user_id = ctx->user_id;
1523         fc->group_id = ctx->group_id;
1524         fc->legacy_opts_show = ctx->legacy_opts_show;
1525         fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1526         fc->destroy = ctx->destroy;
1527         fc->no_control = ctx->no_control;
1528         fc->no_force_umount = ctx->no_force_umount;
1529
1530         err = -ENOMEM;
1531         root = fuse_get_root_inode(sb, ctx->rootmode);
1532         sb->s_d_op = &fuse_root_dentry_operations;
1533         root_dentry = d_make_root(root);
1534         if (!root_dentry)
1535                 goto err_dev_free;
1536         /* Root dentry doesn't have .d_revalidate */
1537         sb->s_d_op = &fuse_dentry_operations;
1538
1539         mutex_lock(&fuse_mutex);
1540         err = -EINVAL;
1541         if (ctx->fudptr && *ctx->fudptr)
1542                 goto err_unlock;
1543
1544         err = fuse_ctl_add_conn(fc);
1545         if (err)
1546                 goto err_unlock;
1547
1548         list_add_tail(&fc->entry, &fuse_conn_list);
1549         sb->s_root = root_dentry;
1550         if (ctx->fudptr)
1551                 *ctx->fudptr = fud;
1552         mutex_unlock(&fuse_mutex);
1553         return 0;
1554
1555  err_unlock:
1556         mutex_unlock(&fuse_mutex);
1557         dput(root_dentry);
1558  err_dev_free:
1559         if (fud)
1560                 fuse_dev_free(fud);
1561  err_free_dax:
1562         if (IS_ENABLED(CONFIG_FUSE_DAX))
1563                 fuse_dax_conn_free(fc);
1564  err:
1565         return err;
1566 }
1567 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1568
1569 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1570 {
1571         struct fuse_fs_context *ctx = fsc->fs_private;
1572         int err;
1573         struct fuse_conn *fc;
1574         struct fuse_mount *fm;
1575
1576         if (!ctx->file || !ctx->rootmode_present ||
1577             !ctx->user_id_present || !ctx->group_id_present)
1578                 return -EINVAL;
1579
1580         /*
1581          * Require mount to happen from the same user namespace which
1582          * opened /dev/fuse to prevent potential attacks.
1583          */
1584         err = -EINVAL;
1585         if ((ctx->file->f_op != &fuse_dev_operations) ||
1586             (ctx->file->f_cred->user_ns != sb->s_user_ns))
1587                 goto err;
1588         ctx->fudptr = &ctx->file->private_data;
1589
1590         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1591         err = -ENOMEM;
1592         if (!fc)
1593                 goto err;
1594
1595         fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1596         if (!fm) {
1597                 kfree(fc);
1598                 goto err;
1599         }
1600
1601         fuse_conn_init(fc, fm, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1602         fc->release = fuse_free_conn;
1603
1604         sb->s_fs_info = fm;
1605
1606         err = fuse_fill_super_common(sb, ctx);
1607         if (err)
1608                 goto err_put_conn;
1609         /* file->private_data shall be visible on all CPUs after this */
1610         smp_mb();
1611         fuse_send_init(get_fuse_mount_super(sb));
1612         return 0;
1613
1614  err_put_conn:
1615         fuse_conn_put(fc);
1616         kfree(fm);
1617         sb->s_fs_info = NULL;
1618  err:
1619         return err;
1620 }
1621
1622 /*
1623  * This is the path where user supplied an already initialized fuse dev.  In
1624  * this case never create a new super if the old one is gone.
1625  */
1626 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1627 {
1628         return -ENOTCONN;
1629 }
1630
1631 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1632 {
1633
1634         return fsc->sget_key == get_fuse_conn_super(sb);
1635 }
1636
1637 static int fuse_get_tree(struct fs_context *fsc)
1638 {
1639         struct fuse_fs_context *ctx = fsc->fs_private;
1640         struct fuse_dev *fud;
1641         struct super_block *sb;
1642         int err;
1643
1644         if (ctx->fd_present)
1645                 ctx->file = fget(ctx->fd);
1646
1647         if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1648                 err = get_tree_bdev(fsc, fuse_fill_super);
1649                 goto out_fput;
1650         }
1651         /*
1652          * While block dev mount can be initialized with a dummy device fd
1653          * (found by device name), normal fuse mounts can't
1654          */
1655         if (!ctx->file)
1656                 return -EINVAL;
1657
1658         /*
1659          * Allow creating a fuse mount with an already initialized fuse
1660          * connection
1661          */
1662         fud = READ_ONCE(ctx->file->private_data);
1663         if (ctx->file->f_op == &fuse_dev_operations && fud) {
1664                 fsc->sget_key = fud->fc;
1665                 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1666                 err = PTR_ERR_OR_ZERO(sb);
1667                 if (!IS_ERR(sb))
1668                         fsc->root = dget(sb->s_root);
1669         } else {
1670                 err = get_tree_nodev(fsc, fuse_fill_super);
1671         }
1672 out_fput:
1673         if (ctx->file)
1674                 fput(ctx->file);
1675         return err;
1676 }
1677
1678 static const struct fs_context_operations fuse_context_ops = {
1679         .free           = fuse_free_fsc,
1680         .parse_param    = fuse_parse_param,
1681         .reconfigure    = fuse_reconfigure,
1682         .get_tree       = fuse_get_tree,
1683 };
1684
1685 /*
1686  * Set up the filesystem mount context.
1687  */
1688 static int fuse_init_fs_context(struct fs_context *fsc)
1689 {
1690         struct fuse_fs_context *ctx;
1691
1692         ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1693         if (!ctx)
1694                 return -ENOMEM;
1695
1696         ctx->max_read = ~0;
1697         ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1698         ctx->legacy_opts_show = true;
1699
1700 #ifdef CONFIG_BLOCK
1701         if (fsc->fs_type == &fuseblk_fs_type) {
1702                 ctx->is_bdev = true;
1703                 ctx->destroy = true;
1704         }
1705 #endif
1706
1707         fsc->fs_private = ctx;
1708         fsc->ops = &fuse_context_ops;
1709         return 0;
1710 }
1711
1712 bool fuse_mount_remove(struct fuse_mount *fm)
1713 {
1714         struct fuse_conn *fc = fm->fc;
1715         bool last = false;
1716
1717         down_write(&fc->killsb);
1718         list_del_init(&fm->fc_entry);
1719         if (list_empty(&fc->mounts))
1720                 last = true;
1721         up_write(&fc->killsb);
1722
1723         return last;
1724 }
1725 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1726
1727 void fuse_conn_destroy(struct fuse_mount *fm)
1728 {
1729         struct fuse_conn *fc = fm->fc;
1730
1731         if (fc->destroy)
1732                 fuse_send_destroy(fm);
1733
1734         fuse_abort_conn(fc);
1735         fuse_wait_aborted(fc);
1736
1737         if (!list_empty(&fc->entry)) {
1738                 mutex_lock(&fuse_mutex);
1739                 list_del(&fc->entry);
1740                 fuse_ctl_remove_conn(fc);
1741                 mutex_unlock(&fuse_mutex);
1742         }
1743 }
1744 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1745
1746 static void fuse_sb_destroy(struct super_block *sb)
1747 {
1748         struct fuse_mount *fm = get_fuse_mount_super(sb);
1749         bool last;
1750
1751         if (fm) {
1752                 last = fuse_mount_remove(fm);
1753                 if (last)
1754                         fuse_conn_destroy(fm);
1755         }
1756 }
1757
1758 static void fuse_kill_sb_anon(struct super_block *sb)
1759 {
1760         fuse_sb_destroy(sb);
1761         kill_anon_super(sb);
1762 }
1763
1764 static struct file_system_type fuse_fs_type = {
1765         .owner          = THIS_MODULE,
1766         .name           = "fuse",
1767         .fs_flags       = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1768         .init_fs_context = fuse_init_fs_context,
1769         .parameters     = fuse_fs_parameters,
1770         .kill_sb        = fuse_kill_sb_anon,
1771 };
1772 MODULE_ALIAS_FS("fuse");
1773
1774 #ifdef CONFIG_BLOCK
1775 static void fuse_kill_sb_blk(struct super_block *sb)
1776 {
1777         fuse_sb_destroy(sb);
1778         kill_block_super(sb);
1779 }
1780
1781 static struct file_system_type fuseblk_fs_type = {
1782         .owner          = THIS_MODULE,
1783         .name           = "fuseblk",
1784         .init_fs_context = fuse_init_fs_context,
1785         .parameters     = fuse_fs_parameters,
1786         .kill_sb        = fuse_kill_sb_blk,
1787         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1788 };
1789 MODULE_ALIAS_FS("fuseblk");
1790
1791 static inline int register_fuseblk(void)
1792 {
1793         return register_filesystem(&fuseblk_fs_type);
1794 }
1795
1796 static inline void unregister_fuseblk(void)
1797 {
1798         unregister_filesystem(&fuseblk_fs_type);
1799 }
1800 #else
1801 static inline int register_fuseblk(void)
1802 {
1803         return 0;
1804 }
1805
1806 static inline void unregister_fuseblk(void)
1807 {
1808 }
1809 #endif
1810
1811 static void fuse_inode_init_once(void *foo)
1812 {
1813         struct inode *inode = foo;
1814
1815         inode_init_once(inode);
1816 }
1817
1818 static int __init fuse_fs_init(void)
1819 {
1820         int err;
1821
1822         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1823                         sizeof(struct fuse_inode), 0,
1824                         SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1825                         fuse_inode_init_once);
1826         err = -ENOMEM;
1827         if (!fuse_inode_cachep)
1828                 goto out;
1829
1830         err = register_fuseblk();
1831         if (err)
1832                 goto out2;
1833
1834         err = register_filesystem(&fuse_fs_type);
1835         if (err)
1836                 goto out3;
1837
1838         return 0;
1839
1840  out3:
1841         unregister_fuseblk();
1842  out2:
1843         kmem_cache_destroy(fuse_inode_cachep);
1844  out:
1845         return err;
1846 }
1847
1848 static void fuse_fs_cleanup(void)
1849 {
1850         unregister_filesystem(&fuse_fs_type);
1851         unregister_fuseblk();
1852
1853         /*
1854          * Make sure all delayed rcu free inodes are flushed before we
1855          * destroy cache.
1856          */
1857         rcu_barrier();
1858         kmem_cache_destroy(fuse_inode_cachep);
1859 }
1860
1861 static struct kobject *fuse_kobj;
1862
1863 static int fuse_sysfs_init(void)
1864 {
1865         int err;
1866
1867         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1868         if (!fuse_kobj) {
1869                 err = -ENOMEM;
1870                 goto out_err;
1871         }
1872
1873         err = sysfs_create_mount_point(fuse_kobj, "connections");
1874         if (err)
1875                 goto out_fuse_unregister;
1876
1877         return 0;
1878
1879  out_fuse_unregister:
1880         kobject_put(fuse_kobj);
1881  out_err:
1882         return err;
1883 }
1884
1885 static void fuse_sysfs_cleanup(void)
1886 {
1887         sysfs_remove_mount_point(fuse_kobj, "connections");
1888         kobject_put(fuse_kobj);
1889 }
1890
1891 static int __init fuse_init(void)
1892 {
1893         int res;
1894
1895         pr_info("init (API version %i.%i)\n",
1896                 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1897
1898         INIT_LIST_HEAD(&fuse_conn_list);
1899         res = fuse_fs_init();
1900         if (res)
1901                 goto err;
1902
1903         res = fuse_dev_init();
1904         if (res)
1905                 goto err_fs_cleanup;
1906
1907         res = fuse_sysfs_init();
1908         if (res)
1909                 goto err_dev_cleanup;
1910
1911         res = fuse_ctl_init();
1912         if (res)
1913                 goto err_sysfs_cleanup;
1914
1915         sanitize_global_limit(&max_user_bgreq);
1916         sanitize_global_limit(&max_user_congthresh);
1917
1918         return 0;
1919
1920  err_sysfs_cleanup:
1921         fuse_sysfs_cleanup();
1922  err_dev_cleanup:
1923         fuse_dev_cleanup();
1924  err_fs_cleanup:
1925         fuse_fs_cleanup();
1926  err:
1927         return res;
1928 }
1929
1930 static void __exit fuse_exit(void)
1931 {
1932         pr_debug("exit\n");
1933
1934         fuse_ctl_cleanup();
1935         fuse_sysfs_cleanup();
1936         fuse_fs_cleanup();
1937         fuse_dev_cleanup();
1938 }
1939
1940 module_init(fuse_init);
1941 module_exit(fuse_exit);