xref: /kernel/linux/linux-5.10/fs/fuse/inode.c (revision 8c2ecf20)
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
27MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
28MODULE_DESCRIPTION("Filesystem in Userspace");
29MODULE_LICENSE("GPL");
30
31static struct kmem_cache *fuse_inode_cachep;
32struct list_head fuse_conn_list;
33DEFINE_MUTEX(fuse_mutex);
34
35static int set_global_limit(const char *val, const struct kernel_param *kp);
36
37unsigned max_user_bgreq;
38module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
39		  &max_user_bgreq, 0644);
40__MODULE_PARM_TYPE(max_user_bgreq, "uint");
41MODULE_PARM_DESC(max_user_bgreq,
42 "Global limit for the maximum number of backgrounded requests an "
43 "unprivileged user can set");
44
45unsigned max_user_congthresh;
46module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
47		  &max_user_congthresh, 0644);
48__MODULE_PARM_TYPE(max_user_congthresh, "uint");
49MODULE_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
64static struct file_system_type fuseblk_fs_type;
65#endif
66
67struct fuse_forget_link *fuse_alloc_forget(void)
68{
69	return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
70}
71
72static 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
99out_free_forget:
100	kfree(fi->forget);
101out_free:
102	kmem_cache_free(fuse_inode_cachep, fi);
103	return NULL;
104}
105
106static 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
118static void fuse_evict_inode(struct inode *inode)
119{
120	struct fuse_inode *fi = get_fuse_inode(inode);
121
122	/* Will write inode on close/munmap and in all other dirtiers */
123	WARN_ON(inode->i_state & I_DIRTY_INODE);
124
125	truncate_inode_pages_final(&inode->i_data);
126	clear_inode(inode);
127	if (inode->i_sb->s_flags & SB_ACTIVE) {
128		struct fuse_conn *fc = get_fuse_conn(inode);
129
130		if (FUSE_IS_DAX(inode))
131			fuse_dax_inode_cleanup(inode);
132		if (fi->nlookup) {
133			fuse_queue_forget(fc, fi->forget, fi->nodeid,
134					  fi->nlookup);
135			fi->forget = NULL;
136		}
137	}
138	if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
139		WARN_ON(!list_empty(&fi->write_files));
140		WARN_ON(!list_empty(&fi->queued_writes));
141	}
142}
143
144static int fuse_reconfigure(struct fs_context *fc)
145{
146	struct super_block *sb = fc->root->d_sb;
147
148	sync_filesystem(sb);
149	if (fc->sb_flags & SB_MANDLOCK)
150		return -EINVAL;
151
152	return 0;
153}
154
155/*
156 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
157 * so that it will fit.
158 */
159static ino_t fuse_squash_ino(u64 ino64)
160{
161	ino_t ino = (ino_t) ino64;
162	if (sizeof(ino_t) < sizeof(u64))
163		ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
164	return ino;
165}
166
167void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
168				   u64 attr_valid)
169{
170	struct fuse_conn *fc = get_fuse_conn(inode);
171	struct fuse_inode *fi = get_fuse_inode(inode);
172
173	lockdep_assert_held(&fi->lock);
174
175	fi->attr_version = atomic64_inc_return(&fc->attr_version);
176	fi->i_time = attr_valid;
177	WRITE_ONCE(fi->inval_mask, 0);
178
179	inode->i_ino     = fuse_squash_ino(attr->ino);
180	inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
181	set_nlink(inode, attr->nlink);
182	inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
183	inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
184	inode->i_blocks  = attr->blocks;
185
186	/* Sanitize nsecs */
187	attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
188	attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
189	attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
190
191	inode->i_atime.tv_sec   = attr->atime;
192	inode->i_atime.tv_nsec  = attr->atimensec;
193	/* mtime from server may be stale due to local buffered write */
194	if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
195		inode->i_mtime.tv_sec   = attr->mtime;
196		inode->i_mtime.tv_nsec  = attr->mtimensec;
197		inode->i_ctime.tv_sec   = attr->ctime;
198		inode->i_ctime.tv_nsec  = attr->ctimensec;
199	}
200
201	if (attr->blksize != 0)
202		inode->i_blkbits = ilog2(attr->blksize);
203	else
204		inode->i_blkbits = inode->i_sb->s_blocksize_bits;
205
206	/*
207	 * Don't set the sticky bit in i_mode, unless we want the VFS
208	 * to check permissions.  This prevents failures due to the
209	 * check in may_delete().
210	 */
211	fi->orig_i_mode = inode->i_mode;
212	if (!fc->default_permissions)
213		inode->i_mode &= ~S_ISVTX;
214
215	fi->orig_ino = attr->ino;
216}
217
218void 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
273static 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
297static 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
306static 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
313struct 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
340retry:
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	}
358done:
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
368struct 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
390int 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
422bool 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
434void fuse_unlock_inode(struct inode *inode, bool locked)
435{
436	if (locked)
437		mutex_unlock(&get_fuse_inode(inode)->mutex);
438}
439
440static 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
448static 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
460static void fuse_put_super(struct super_block *sb)
461{
462	struct fuse_mount *fm = get_fuse_mount_super(sb);
463
464	fuse_mount_put(fm);
465}
466
467static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
468{
469	stbuf->f_type    = FUSE_SUPER_MAGIC;
470	stbuf->f_bsize   = attr->bsize;
471	stbuf->f_frsize  = attr->frsize;
472	stbuf->f_blocks  = attr->blocks;
473	stbuf->f_bfree   = attr->bfree;
474	stbuf->f_bavail  = attr->bavail;
475	stbuf->f_files   = attr->files;
476	stbuf->f_ffree   = attr->ffree;
477	stbuf->f_namelen = attr->namelen;
478	/* fsid is left zero */
479}
480
481static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
482{
483	struct super_block *sb = dentry->d_sb;
484	struct fuse_mount *fm = get_fuse_mount_super(sb);
485	FUSE_ARGS(args);
486	struct fuse_statfs_out outarg;
487	int err;
488
489	if (!fuse_allow_current_process(fm->fc)) {
490		buf->f_type = FUSE_SUPER_MAGIC;
491		return 0;
492	}
493
494	memset(&outarg, 0, sizeof(outarg));
495	args.in_numargs = 0;
496	args.opcode = FUSE_STATFS;
497	args.nodeid = get_node_id(d_inode(dentry));
498	args.out_numargs = 1;
499	args.out_args[0].size = sizeof(outarg);
500	args.out_args[0].value = &outarg;
501	err = fuse_simple_request(fm, &args);
502	if (!err)
503		convert_fuse_statfs(buf, &outarg.st);
504	return err;
505}
506
507enum {
508	OPT_SOURCE,
509	OPT_SUBTYPE,
510	OPT_FD,
511	OPT_ROOTMODE,
512	OPT_USER_ID,
513	OPT_GROUP_ID,
514	OPT_DEFAULT_PERMISSIONS,
515	OPT_ALLOW_OTHER,
516	OPT_MAX_READ,
517	OPT_BLKSIZE,
518	OPT_ERR
519};
520
521static const struct fs_parameter_spec fuse_fs_parameters[] = {
522	fsparam_string	("source",		OPT_SOURCE),
523	fsparam_u32	("fd",			OPT_FD),
524	fsparam_u32oct	("rootmode",		OPT_ROOTMODE),
525	fsparam_u32	("user_id",		OPT_USER_ID),
526	fsparam_u32	("group_id",		OPT_GROUP_ID),
527	fsparam_flag	("default_permissions",	OPT_DEFAULT_PERMISSIONS),
528	fsparam_flag	("allow_other",		OPT_ALLOW_OTHER),
529	fsparam_u32	("max_read",		OPT_MAX_READ),
530	fsparam_u32	("blksize",		OPT_BLKSIZE),
531	fsparam_string	("subtype",		OPT_SUBTYPE),
532	{}
533};
534
535static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
536{
537	struct fs_parse_result result;
538	struct fuse_fs_context *ctx = fc->fs_private;
539	int opt;
540
541	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
542		/*
543		 * Ignore options coming from mount(MS_REMOUNT) for backward
544		 * compatibility.
545		 */
546		if (fc->oldapi)
547			return 0;
548
549		return invalfc(fc, "No changes allowed in reconfigure");
550	}
551
552	opt = fs_parse(fc, fuse_fs_parameters, param, &result);
553	if (opt < 0)
554		return opt;
555
556	switch (opt) {
557	case OPT_SOURCE:
558		if (fc->source)
559			return invalfc(fc, "Multiple sources specified");
560		fc->source = param->string;
561		param->string = NULL;
562		break;
563
564	case OPT_SUBTYPE:
565		if (ctx->subtype)
566			return invalfc(fc, "Multiple subtypes specified");
567		ctx->subtype = param->string;
568		param->string = NULL;
569		return 0;
570
571	case OPT_FD:
572		ctx->fd = result.uint_32;
573		ctx->fd_present = true;
574		break;
575
576	case OPT_ROOTMODE:
577		if (!fuse_valid_type(result.uint_32))
578			return invalfc(fc, "Invalid rootmode");
579		ctx->rootmode = result.uint_32;
580		ctx->rootmode_present = true;
581		break;
582
583	case OPT_USER_ID:
584		ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
585		if (!uid_valid(ctx->user_id))
586			return invalfc(fc, "Invalid user_id");
587		ctx->user_id_present = true;
588		break;
589
590	case OPT_GROUP_ID:
591		ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
592		if (!gid_valid(ctx->group_id))
593			return invalfc(fc, "Invalid group_id");
594		ctx->group_id_present = true;
595		break;
596
597	case OPT_DEFAULT_PERMISSIONS:
598		ctx->default_permissions = true;
599		break;
600
601	case OPT_ALLOW_OTHER:
602		ctx->allow_other = true;
603		break;
604
605	case OPT_MAX_READ:
606		ctx->max_read = result.uint_32;
607		break;
608
609	case OPT_BLKSIZE:
610		if (!ctx->is_bdev)
611			return invalfc(fc, "blksize only supported for fuseblk");
612		ctx->blksize = result.uint_32;
613		break;
614
615	default:
616		return -EINVAL;
617	}
618
619	return 0;
620}
621
622static void fuse_free_fc(struct fs_context *fc)
623{
624	struct fuse_fs_context *ctx = fc->fs_private;
625
626	if (ctx) {
627		kfree(ctx->subtype);
628		kfree(ctx);
629	}
630}
631
632static int fuse_show_options(struct seq_file *m, struct dentry *root)
633{
634	struct super_block *sb = root->d_sb;
635	struct fuse_conn *fc = get_fuse_conn_super(sb);
636
637	if (fc->legacy_opts_show) {
638		seq_printf(m, ",user_id=%u",
639			   from_kuid_munged(fc->user_ns, fc->user_id));
640		seq_printf(m, ",group_id=%u",
641			   from_kgid_munged(fc->user_ns, fc->group_id));
642		if (fc->default_permissions)
643			seq_puts(m, ",default_permissions");
644		if (fc->allow_other)
645			seq_puts(m, ",allow_other");
646		if (fc->max_read != ~0)
647			seq_printf(m, ",max_read=%u", fc->max_read);
648		if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
649			seq_printf(m, ",blksize=%lu", sb->s_blocksize);
650	}
651#ifdef CONFIG_FUSE_DAX
652	if (fc->dax)
653		seq_puts(m, ",dax");
654#endif
655
656	return 0;
657}
658
659static void fuse_iqueue_init(struct fuse_iqueue *fiq,
660			     const struct fuse_iqueue_ops *ops,
661			     void *priv)
662{
663	memset(fiq, 0, sizeof(struct fuse_iqueue));
664	spin_lock_init(&fiq->lock);
665	init_waitqueue_head(&fiq->waitq);
666	INIT_LIST_HEAD(&fiq->pending);
667	INIT_LIST_HEAD(&fiq->interrupts);
668	fiq->forget_list_tail = &fiq->forget_list_head;
669	fiq->connected = 1;
670	fiq->ops = ops;
671	fiq->priv = priv;
672}
673
674static void fuse_pqueue_init(struct fuse_pqueue *fpq)
675{
676	unsigned int i;
677
678	spin_lock_init(&fpq->lock);
679	for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
680		INIT_LIST_HEAD(&fpq->processing[i]);
681	INIT_LIST_HEAD(&fpq->io);
682	fpq->connected = 1;
683}
684
685void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
686		    struct user_namespace *user_ns,
687		    const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
688{
689	memset(fc, 0, sizeof(*fc));
690	spin_lock_init(&fc->lock);
691	spin_lock_init(&fc->bg_lock);
692	init_rwsem(&fc->killsb);
693	refcount_set(&fc->count, 1);
694	atomic_set(&fc->dev_count, 1);
695	init_waitqueue_head(&fc->blocked_waitq);
696	fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
697	INIT_LIST_HEAD(&fc->bg_queue);
698	INIT_LIST_HEAD(&fc->entry);
699	INIT_LIST_HEAD(&fc->devices);
700	atomic_set(&fc->num_waiting, 0);
701	fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
702	fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
703	atomic64_set(&fc->khctr, 0);
704	fc->polled_files = RB_ROOT;
705	fc->blocked = 0;
706	fc->initialized = 0;
707	fc->connected = 1;
708	atomic64_set(&fc->attr_version, 1);
709	get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
710	fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
711	fc->user_ns = get_user_ns(user_ns);
712	fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
713	fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
714
715	INIT_LIST_HEAD(&fc->mounts);
716	list_add(&fm->fc_entry, &fc->mounts);
717	fm->fc = fc;
718	refcount_set(&fm->count, 1);
719}
720EXPORT_SYMBOL_GPL(fuse_conn_init);
721
722void fuse_conn_put(struct fuse_conn *fc)
723{
724	if (refcount_dec_and_test(&fc->count)) {
725		struct fuse_iqueue *fiq = &fc->iq;
726
727		if (IS_ENABLED(CONFIG_FUSE_DAX))
728			fuse_dax_conn_free(fc);
729		if (fiq->ops->release)
730			fiq->ops->release(fiq);
731		put_pid_ns(fc->pid_ns);
732		put_user_ns(fc->user_ns);
733		fc->release(fc);
734	}
735}
736EXPORT_SYMBOL_GPL(fuse_conn_put);
737
738struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
739{
740	refcount_inc(&fc->count);
741	return fc;
742}
743EXPORT_SYMBOL_GPL(fuse_conn_get);
744
745void fuse_mount_put(struct fuse_mount *fm)
746{
747	if (refcount_dec_and_test(&fm->count)) {
748		if (fm->fc)
749			fuse_conn_put(fm->fc);
750		kfree(fm);
751	}
752}
753EXPORT_SYMBOL_GPL(fuse_mount_put);
754
755struct fuse_mount *fuse_mount_get(struct fuse_mount *fm)
756{
757	refcount_inc(&fm->count);
758	return fm;
759}
760EXPORT_SYMBOL_GPL(fuse_mount_get);
761
762static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
763{
764	struct fuse_attr attr;
765	memset(&attr, 0, sizeof(attr));
766
767	attr.mode = mode;
768	attr.ino = FUSE_ROOT_ID;
769	attr.nlink = 1;
770	return fuse_iget(sb, 1, 0, &attr, 0, 0);
771}
772
773struct fuse_inode_handle {
774	u64 nodeid;
775	u32 generation;
776};
777
778static struct dentry *fuse_get_dentry(struct super_block *sb,
779				      struct fuse_inode_handle *handle)
780{
781	struct fuse_conn *fc = get_fuse_conn_super(sb);
782	struct inode *inode;
783	struct dentry *entry;
784	int err = -ESTALE;
785
786	if (handle->nodeid == 0)
787		goto out_err;
788
789	inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
790	if (!inode) {
791		struct fuse_entry_out outarg;
792		const struct qstr name = QSTR_INIT(".", 1);
793
794		if (!fc->export_support)
795			goto out_err;
796
797		err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
798				       &inode);
799		if (err && err != -ENOENT)
800			goto out_err;
801		if (err || !inode) {
802			err = -ESTALE;
803			goto out_err;
804		}
805		err = -EIO;
806		if (get_node_id(inode) != handle->nodeid)
807			goto out_iput;
808	}
809	err = -ESTALE;
810	if (inode->i_generation != handle->generation)
811		goto out_iput;
812
813	entry = d_obtain_alias(inode);
814	if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
815		fuse_invalidate_entry_cache(entry);
816
817	return entry;
818
819 out_iput:
820	iput(inode);
821 out_err:
822	return ERR_PTR(err);
823}
824
825static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
826			   struct inode *parent)
827{
828	int len = parent ? 6 : 3;
829	u64 nodeid;
830	u32 generation;
831
832	if (*max_len < len) {
833		*max_len = len;
834		return  FILEID_INVALID;
835	}
836
837	nodeid = get_fuse_inode(inode)->nodeid;
838	generation = inode->i_generation;
839
840	fh[0] = (u32)(nodeid >> 32);
841	fh[1] = (u32)(nodeid & 0xffffffff);
842	fh[2] = generation;
843
844	if (parent) {
845		nodeid = get_fuse_inode(parent)->nodeid;
846		generation = parent->i_generation;
847
848		fh[3] = (u32)(nodeid >> 32);
849		fh[4] = (u32)(nodeid & 0xffffffff);
850		fh[5] = generation;
851	}
852
853	*max_len = len;
854	return parent ? 0x82 : 0x81;
855}
856
857static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
858		struct fid *fid, int fh_len, int fh_type)
859{
860	struct fuse_inode_handle handle;
861
862	if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
863		return NULL;
864
865	handle.nodeid = (u64) fid->raw[0] << 32;
866	handle.nodeid |= (u64) fid->raw[1];
867	handle.generation = fid->raw[2];
868	return fuse_get_dentry(sb, &handle);
869}
870
871static struct dentry *fuse_fh_to_parent(struct super_block *sb,
872		struct fid *fid, int fh_len, int fh_type)
873{
874	struct fuse_inode_handle parent;
875
876	if (fh_type != 0x82 || fh_len < 6)
877		return NULL;
878
879	parent.nodeid = (u64) fid->raw[3] << 32;
880	parent.nodeid |= (u64) fid->raw[4];
881	parent.generation = fid->raw[5];
882	return fuse_get_dentry(sb, &parent);
883}
884
885static struct dentry *fuse_get_parent(struct dentry *child)
886{
887	struct inode *child_inode = d_inode(child);
888	struct fuse_conn *fc = get_fuse_conn(child_inode);
889	struct inode *inode;
890	struct dentry *parent;
891	struct fuse_entry_out outarg;
892	const struct qstr name = QSTR_INIT("..", 2);
893	int err;
894
895	if (!fc->export_support)
896		return ERR_PTR(-ESTALE);
897
898	err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
899			       &name, &outarg, &inode);
900	if (err) {
901		if (err == -ENOENT)
902			return ERR_PTR(-ESTALE);
903		return ERR_PTR(err);
904	}
905
906	parent = d_obtain_alias(inode);
907	if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
908		fuse_invalidate_entry_cache(parent);
909
910	return parent;
911}
912
913static const struct export_operations fuse_export_operations = {
914	.fh_to_dentry	= fuse_fh_to_dentry,
915	.fh_to_parent	= fuse_fh_to_parent,
916	.encode_fh	= fuse_encode_fh,
917	.get_parent	= fuse_get_parent,
918};
919
920static const struct super_operations fuse_super_operations = {
921	.alloc_inode    = fuse_alloc_inode,
922	.free_inode     = fuse_free_inode,
923	.evict_inode	= fuse_evict_inode,
924	.write_inode	= fuse_write_inode,
925	.drop_inode	= generic_delete_inode,
926	.put_super	= fuse_put_super,
927	.umount_begin	= fuse_umount_begin,
928	.statfs		= fuse_statfs,
929	.show_options	= fuse_show_options,
930};
931
932static void sanitize_global_limit(unsigned *limit)
933{
934	/*
935	 * The default maximum number of async requests is calculated to consume
936	 * 1/2^13 of the total memory, assuming 392 bytes per request.
937	 */
938	if (*limit == 0)
939		*limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
940
941	if (*limit >= 1 << 16)
942		*limit = (1 << 16) - 1;
943}
944
945static int set_global_limit(const char *val, const struct kernel_param *kp)
946{
947	int rv;
948
949	rv = param_set_uint(val, kp);
950	if (rv)
951		return rv;
952
953	sanitize_global_limit((unsigned *)kp->arg);
954
955	return 0;
956}
957
958static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
959{
960	int cap_sys_admin = capable(CAP_SYS_ADMIN);
961
962	if (arg->minor < 13)
963		return;
964
965	sanitize_global_limit(&max_user_bgreq);
966	sanitize_global_limit(&max_user_congthresh);
967
968	spin_lock(&fc->bg_lock);
969	if (arg->max_background) {
970		fc->max_background = arg->max_background;
971
972		if (!cap_sys_admin && fc->max_background > max_user_bgreq)
973			fc->max_background = max_user_bgreq;
974	}
975	if (arg->congestion_threshold) {
976		fc->congestion_threshold = arg->congestion_threshold;
977
978		if (!cap_sys_admin &&
979		    fc->congestion_threshold > max_user_congthresh)
980			fc->congestion_threshold = max_user_congthresh;
981	}
982	spin_unlock(&fc->bg_lock);
983}
984
985struct fuse_init_args {
986	struct fuse_args args;
987	struct fuse_init_in in;
988	struct fuse_init_out out;
989};
990
991static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
992			       int error)
993{
994	struct fuse_conn *fc = fm->fc;
995	struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
996	struct fuse_init_out *arg = &ia->out;
997	bool ok = true;
998
999	if (error || arg->major != FUSE_KERNEL_VERSION)
1000		ok = false;
1001	else {
1002		unsigned long ra_pages;
1003
1004		process_init_limits(fc, arg);
1005
1006		if (arg->minor >= 6) {
1007			ra_pages = arg->max_readahead / PAGE_SIZE;
1008			if (arg->flags & FUSE_ASYNC_READ)
1009				fc->async_read = 1;
1010			if (!(arg->flags & FUSE_POSIX_LOCKS))
1011				fc->no_lock = 1;
1012			if (arg->minor >= 17) {
1013				if (!(arg->flags & FUSE_FLOCK_LOCKS))
1014					fc->no_flock = 1;
1015			} else {
1016				if (!(arg->flags & FUSE_POSIX_LOCKS))
1017					fc->no_flock = 1;
1018			}
1019			if (arg->flags & FUSE_ATOMIC_O_TRUNC)
1020				fc->atomic_o_trunc = 1;
1021			if (arg->minor >= 9) {
1022				/* LOOKUP has dependency on proto version */
1023				if (arg->flags & FUSE_EXPORT_SUPPORT)
1024					fc->export_support = 1;
1025			}
1026			if (arg->flags & FUSE_BIG_WRITES)
1027				fc->big_writes = 1;
1028			if (arg->flags & FUSE_DONT_MASK)
1029				fc->dont_mask = 1;
1030			if (arg->flags & FUSE_AUTO_INVAL_DATA)
1031				fc->auto_inval_data = 1;
1032			else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
1033				fc->explicit_inval_data = 1;
1034			if (arg->flags & FUSE_DO_READDIRPLUS) {
1035				fc->do_readdirplus = 1;
1036				if (arg->flags & FUSE_READDIRPLUS_AUTO)
1037					fc->readdirplus_auto = 1;
1038			}
1039			if (arg->flags & FUSE_ASYNC_DIO)
1040				fc->async_dio = 1;
1041			if (arg->flags & FUSE_WRITEBACK_CACHE)
1042				fc->writeback_cache = 1;
1043			if (arg->flags & FUSE_PARALLEL_DIROPS)
1044				fc->parallel_dirops = 1;
1045			if (arg->flags & FUSE_HANDLE_KILLPRIV)
1046				fc->handle_killpriv = 1;
1047			if (arg->time_gran && arg->time_gran <= 1000000000)
1048				fm->sb->s_time_gran = arg->time_gran;
1049			if ((arg->flags & FUSE_POSIX_ACL)) {
1050				fc->default_permissions = 1;
1051				fc->posix_acl = 1;
1052				fm->sb->s_xattr = fuse_acl_xattr_handlers;
1053			}
1054			if (arg->flags & FUSE_CACHE_SYMLINKS)
1055				fc->cache_symlinks = 1;
1056			if (arg->flags & FUSE_ABORT_ERROR)
1057				fc->abort_err = 1;
1058			if (arg->flags & FUSE_MAX_PAGES) {
1059				fc->max_pages =
1060					min_t(unsigned int, fc->max_pages_limit,
1061					max_t(unsigned int, arg->max_pages, 1));
1062			}
1063			if (IS_ENABLED(CONFIG_FUSE_DAX) &&
1064			    arg->flags & FUSE_MAP_ALIGNMENT &&
1065			    !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1066				ok = false;
1067			}
1068		} else {
1069			ra_pages = fc->max_read / PAGE_SIZE;
1070			fc->no_lock = 1;
1071			fc->no_flock = 1;
1072		}
1073
1074		fm->sb->s_bdi->ra_pages =
1075				min(fm->sb->s_bdi->ra_pages, ra_pages);
1076		fc->minor = arg->minor;
1077		fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1078		fc->max_write = max_t(unsigned, 4096, fc->max_write);
1079		fc->conn_init = 1;
1080	}
1081	kfree(ia);
1082
1083	if (!ok) {
1084		fc->conn_init = 0;
1085		fc->conn_error = 1;
1086	}
1087
1088	fuse_set_initialized(fc);
1089	wake_up_all(&fc->blocked_waitq);
1090}
1091
1092void fuse_send_init(struct fuse_mount *fm)
1093{
1094	struct fuse_init_args *ia;
1095
1096	ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1097
1098	ia->in.major = FUSE_KERNEL_VERSION;
1099	ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1100	ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1101	ia->in.flags |=
1102		FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1103		FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1104		FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1105		FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1106		FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1107		FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1108		FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1109		FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1110		FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
1111#ifdef CONFIG_FUSE_DAX
1112	if (fm->fc->dax)
1113		ia->in.flags |= FUSE_MAP_ALIGNMENT;
1114#endif
1115	if (fm->fc->auto_submounts)
1116		ia->in.flags |= FUSE_SUBMOUNTS;
1117
1118	ia->args.opcode = FUSE_INIT;
1119	ia->args.in_numargs = 1;
1120	ia->args.in_args[0].size = sizeof(ia->in);
1121	ia->args.in_args[0].value = &ia->in;
1122	ia->args.out_numargs = 1;
1123	/* Variable length argument used for backward compatibility
1124	   with interface version < 7.5.  Rest of init_out is zeroed
1125	   by do_get_request(), so a short reply is not a problem */
1126	ia->args.out_argvar = true;
1127	ia->args.out_args[0].size = sizeof(ia->out);
1128	ia->args.out_args[0].value = &ia->out;
1129	ia->args.force = true;
1130	ia->args.nocreds = true;
1131	ia->args.end = process_init_reply;
1132
1133	if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1134		process_init_reply(fm, &ia->args, -ENOTCONN);
1135}
1136EXPORT_SYMBOL_GPL(fuse_send_init);
1137
1138void fuse_free_conn(struct fuse_conn *fc)
1139{
1140	WARN_ON(!list_empty(&fc->devices));
1141	kfree_rcu(fc, rcu);
1142}
1143EXPORT_SYMBOL_GPL(fuse_free_conn);
1144
1145static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1146{
1147	int err;
1148	char *suffix = "";
1149
1150	if (sb->s_bdev) {
1151		suffix = "-fuseblk";
1152		/*
1153		 * sb->s_bdi points to blkdev's bdi however we want to redirect
1154		 * it to our private bdi...
1155		 */
1156		bdi_put(sb->s_bdi);
1157		sb->s_bdi = &noop_backing_dev_info;
1158	}
1159	err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1160				   MINOR(fc->dev), suffix);
1161	if (err)
1162		return err;
1163
1164	/* fuse does it's own writeback accounting */
1165	sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1166	sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1167
1168	/*
1169	 * For a single fuse filesystem use max 1% of dirty +
1170	 * writeback threshold.
1171	 *
1172	 * This gives about 1M of write buffer for memory maps on a
1173	 * machine with 1G and 10% dirty_ratio, which should be more
1174	 * than enough.
1175	 *
1176	 * Privileged users can raise it by writing to
1177	 *
1178	 *    /sys/class/bdi/<bdi>/max_ratio
1179	 */
1180	bdi_set_max_ratio(sb->s_bdi, 1);
1181
1182	return 0;
1183}
1184
1185struct fuse_dev *fuse_dev_alloc(void)
1186{
1187	struct fuse_dev *fud;
1188	struct list_head *pq;
1189
1190	fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1191	if (!fud)
1192		return NULL;
1193
1194	pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1195	if (!pq) {
1196		kfree(fud);
1197		return NULL;
1198	}
1199
1200	fud->pq.processing = pq;
1201	fuse_pqueue_init(&fud->pq);
1202
1203	return fud;
1204}
1205EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1206
1207void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1208{
1209	fud->fc = fuse_conn_get(fc);
1210	spin_lock(&fc->lock);
1211	list_add_tail(&fud->entry, &fc->devices);
1212	spin_unlock(&fc->lock);
1213}
1214EXPORT_SYMBOL_GPL(fuse_dev_install);
1215
1216struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1217{
1218	struct fuse_dev *fud;
1219
1220	fud = fuse_dev_alloc();
1221	if (!fud)
1222		return NULL;
1223
1224	fuse_dev_install(fud, fc);
1225	return fud;
1226}
1227EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1228
1229void fuse_dev_free(struct fuse_dev *fud)
1230{
1231	struct fuse_conn *fc = fud->fc;
1232
1233	if (fc) {
1234		spin_lock(&fc->lock);
1235		list_del(&fud->entry);
1236		spin_unlock(&fc->lock);
1237
1238		fuse_conn_put(fc);
1239	}
1240	kfree(fud->pq.processing);
1241	kfree(fud);
1242}
1243EXPORT_SYMBOL_GPL(fuse_dev_free);
1244
1245static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1246				      const struct fuse_inode *fi)
1247{
1248	*attr = (struct fuse_attr){
1249		.ino		= fi->inode.i_ino,
1250		.size		= fi->inode.i_size,
1251		.blocks		= fi->inode.i_blocks,
1252		.atime		= fi->inode.i_atime.tv_sec,
1253		.mtime		= fi->inode.i_mtime.tv_sec,
1254		.ctime		= fi->inode.i_ctime.tv_sec,
1255		.atimensec	= fi->inode.i_atime.tv_nsec,
1256		.mtimensec	= fi->inode.i_mtime.tv_nsec,
1257		.ctimensec	= fi->inode.i_ctime.tv_nsec,
1258		.mode		= fi->inode.i_mode,
1259		.nlink		= fi->inode.i_nlink,
1260		.uid		= fi->inode.i_uid.val,
1261		.gid		= fi->inode.i_gid.val,
1262		.rdev		= fi->inode.i_rdev,
1263		.blksize	= 1u << fi->inode.i_blkbits,
1264	};
1265}
1266
1267static void fuse_sb_defaults(struct super_block *sb)
1268{
1269	sb->s_magic = FUSE_SUPER_MAGIC;
1270	sb->s_op = &fuse_super_operations;
1271	sb->s_xattr = fuse_xattr_handlers;
1272	sb->s_maxbytes = MAX_LFS_FILESIZE;
1273	sb->s_time_gran = 1;
1274	sb->s_export_op = &fuse_export_operations;
1275	sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1276	if (sb->s_user_ns != &init_user_ns)
1277		sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1278	sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1279
1280	/*
1281	 * If we are not in the initial user namespace posix
1282	 * acls must be translated.
1283	 */
1284	if (sb->s_user_ns != &init_user_ns)
1285		sb->s_xattr = fuse_no_acl_xattr_handlers;
1286}
1287
1288int fuse_fill_super_submount(struct super_block *sb,
1289			     struct fuse_inode *parent_fi)
1290{
1291	struct fuse_mount *fm = get_fuse_mount_super(sb);
1292	struct super_block *parent_sb = parent_fi->inode.i_sb;
1293	struct fuse_attr root_attr;
1294	struct inode *root;
1295
1296	fuse_sb_defaults(sb);
1297	fm->sb = sb;
1298
1299	WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1300	sb->s_bdi = bdi_get(parent_sb->s_bdi);
1301
1302	sb->s_xattr = parent_sb->s_xattr;
1303	sb->s_time_gran = parent_sb->s_time_gran;
1304	sb->s_blocksize = parent_sb->s_blocksize;
1305	sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1306	sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1307	if (parent_sb->s_subtype && !sb->s_subtype)
1308		return -ENOMEM;
1309
1310	fuse_fill_attr_from_inode(&root_attr, parent_fi);
1311	root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1312	/*
1313	 * This inode is just a duplicate, so it is not looked up and
1314	 * its nlookup should not be incremented.  fuse_iget() does
1315	 * that, though, so undo it here.
1316	 */
1317	get_fuse_inode(root)->nlookup--;
1318	sb->s_d_op = &fuse_dentry_operations;
1319	sb->s_root = d_make_root(root);
1320	if (!sb->s_root)
1321		return -ENOMEM;
1322
1323	return 0;
1324}
1325
1326int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1327{
1328	struct fuse_dev *fud = NULL;
1329	struct fuse_mount *fm = get_fuse_mount_super(sb);
1330	struct fuse_conn *fc = fm->fc;
1331	struct inode *root;
1332	struct dentry *root_dentry;
1333	int err;
1334
1335	err = -EINVAL;
1336	if (sb->s_flags & SB_MANDLOCK)
1337		goto err;
1338
1339	fuse_sb_defaults(sb);
1340
1341	if (ctx->is_bdev) {
1342#ifdef CONFIG_BLOCK
1343		err = -EINVAL;
1344		if (!sb_set_blocksize(sb, ctx->blksize))
1345			goto err;
1346#endif
1347	} else {
1348		sb->s_blocksize = PAGE_SIZE;
1349		sb->s_blocksize_bits = PAGE_SHIFT;
1350	}
1351
1352	sb->s_subtype = ctx->subtype;
1353	ctx->subtype = NULL;
1354	if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1355		err = fuse_dax_conn_alloc(fc, ctx->dax_dev);
1356		if (err)
1357			goto err;
1358	}
1359
1360	if (ctx->fudptr) {
1361		err = -ENOMEM;
1362		fud = fuse_dev_alloc_install(fc);
1363		if (!fud)
1364			goto err_free_dax;
1365	}
1366
1367	fc->dev = sb->s_dev;
1368	fm->sb = sb;
1369	err = fuse_bdi_init(fc, sb);
1370	if (err)
1371		goto err_dev_free;
1372
1373	/* Handle umasking inside the fuse code */
1374	if (sb->s_flags & SB_POSIXACL)
1375		fc->dont_mask = 1;
1376	sb->s_flags |= SB_POSIXACL;
1377
1378	fc->default_permissions = ctx->default_permissions;
1379	fc->allow_other = ctx->allow_other;
1380	fc->user_id = ctx->user_id;
1381	fc->group_id = ctx->group_id;
1382	fc->legacy_opts_show = ctx->legacy_opts_show;
1383	fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1384	fc->destroy = ctx->destroy;
1385	fc->no_control = ctx->no_control;
1386	fc->no_force_umount = ctx->no_force_umount;
1387
1388	err = -ENOMEM;
1389	root = fuse_get_root_inode(sb, ctx->rootmode);
1390	sb->s_d_op = &fuse_root_dentry_operations;
1391	root_dentry = d_make_root(root);
1392	if (!root_dentry)
1393		goto err_dev_free;
1394	/* Root dentry doesn't have .d_revalidate */
1395	sb->s_d_op = &fuse_dentry_operations;
1396
1397	mutex_lock(&fuse_mutex);
1398	err = -EINVAL;
1399	if (ctx->fudptr && *ctx->fudptr)
1400		goto err_unlock;
1401
1402	err = fuse_ctl_add_conn(fc);
1403	if (err)
1404		goto err_unlock;
1405
1406	list_add_tail(&fc->entry, &fuse_conn_list);
1407	sb->s_root = root_dentry;
1408	if (ctx->fudptr)
1409		*ctx->fudptr = fud;
1410	mutex_unlock(&fuse_mutex);
1411	return 0;
1412
1413 err_unlock:
1414	mutex_unlock(&fuse_mutex);
1415	dput(root_dentry);
1416 err_dev_free:
1417	if (fud)
1418		fuse_dev_free(fud);
1419 err_free_dax:
1420	if (IS_ENABLED(CONFIG_FUSE_DAX))
1421		fuse_dax_conn_free(fc);
1422 err:
1423	return err;
1424}
1425EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1426
1427static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1428{
1429	struct fuse_fs_context *ctx = fsc->fs_private;
1430	struct file *file;
1431	int err;
1432	struct fuse_conn *fc;
1433	struct fuse_mount *fm;
1434
1435	err = -EINVAL;
1436	file = fget(ctx->fd);
1437	if (!file)
1438		goto err;
1439
1440	/*
1441	 * Require mount to happen from the same user namespace which
1442	 * opened /dev/fuse to prevent potential attacks.
1443	 */
1444	if ((file->f_op != &fuse_dev_operations) ||
1445	    (file->f_cred->user_ns != sb->s_user_ns))
1446		goto err_fput;
1447	ctx->fudptr = &file->private_data;
1448
1449	fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1450	err = -ENOMEM;
1451	if (!fc)
1452		goto err_fput;
1453
1454	fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1455	if (!fm) {
1456		kfree(fc);
1457		goto err_fput;
1458	}
1459
1460	fuse_conn_init(fc, fm, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1461	fc->release = fuse_free_conn;
1462
1463	sb->s_fs_info = fm;
1464
1465	err = fuse_fill_super_common(sb, ctx);
1466	if (err)
1467		goto err_put_conn;
1468	/*
1469	 * atomic_dec_and_test() in fput() provides the necessary
1470	 * memory barrier for file->private_data to be visible on all
1471	 * CPUs after this
1472	 */
1473	fput(file);
1474	fuse_send_init(get_fuse_mount_super(sb));
1475	return 0;
1476
1477 err_put_conn:
1478	fuse_mount_put(fm);
1479	sb->s_fs_info = NULL;
1480 err_fput:
1481	fput(file);
1482 err:
1483	return err;
1484}
1485
1486static int fuse_get_tree(struct fs_context *fc)
1487{
1488	struct fuse_fs_context *ctx = fc->fs_private;
1489
1490	if (!ctx->fd_present || !ctx->rootmode_present ||
1491	    !ctx->user_id_present || !ctx->group_id_present)
1492		return -EINVAL;
1493
1494#ifdef CONFIG_BLOCK
1495	if (ctx->is_bdev)
1496		return get_tree_bdev(fc, fuse_fill_super);
1497#endif
1498
1499	return get_tree_nodev(fc, fuse_fill_super);
1500}
1501
1502static const struct fs_context_operations fuse_context_ops = {
1503	.free		= fuse_free_fc,
1504	.parse_param	= fuse_parse_param,
1505	.reconfigure	= fuse_reconfigure,
1506	.get_tree	= fuse_get_tree,
1507};
1508
1509/*
1510 * Set up the filesystem mount context.
1511 */
1512static int fuse_init_fs_context(struct fs_context *fc)
1513{
1514	struct fuse_fs_context *ctx;
1515
1516	ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1517	if (!ctx)
1518		return -ENOMEM;
1519
1520	ctx->max_read = ~0;
1521	ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1522	ctx->legacy_opts_show = true;
1523
1524#ifdef CONFIG_BLOCK
1525	if (fc->fs_type == &fuseblk_fs_type) {
1526		ctx->is_bdev = true;
1527		ctx->destroy = true;
1528	}
1529#endif
1530
1531	fc->fs_private = ctx;
1532	fc->ops = &fuse_context_ops;
1533	return 0;
1534}
1535
1536bool fuse_mount_remove(struct fuse_mount *fm)
1537{
1538	struct fuse_conn *fc = fm->fc;
1539	bool last = false;
1540
1541	down_write(&fc->killsb);
1542	list_del_init(&fm->fc_entry);
1543	if (list_empty(&fc->mounts))
1544		last = true;
1545	up_write(&fc->killsb);
1546
1547	return last;
1548}
1549EXPORT_SYMBOL_GPL(fuse_mount_remove);
1550
1551void fuse_conn_destroy(struct fuse_mount *fm)
1552{
1553	struct fuse_conn *fc = fm->fc;
1554
1555	if (fc->destroy)
1556		fuse_send_destroy(fm);
1557
1558	fuse_abort_conn(fc);
1559	fuse_wait_aborted(fc);
1560
1561	if (!list_empty(&fc->entry)) {
1562		mutex_lock(&fuse_mutex);
1563		list_del(&fc->entry);
1564		fuse_ctl_remove_conn(fc);
1565		mutex_unlock(&fuse_mutex);
1566	}
1567}
1568EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1569
1570static void fuse_kill_sb_anon(struct super_block *sb)
1571{
1572	struct fuse_mount *fm = get_fuse_mount_super(sb);
1573	bool last;
1574
1575	if (fm) {
1576		last = fuse_mount_remove(fm);
1577		if (last)
1578			fuse_conn_destroy(fm);
1579	}
1580	kill_anon_super(sb);
1581}
1582
1583static struct file_system_type fuse_fs_type = {
1584	.owner		= THIS_MODULE,
1585	.name		= "fuse",
1586	.fs_flags	= FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1587	.init_fs_context = fuse_init_fs_context,
1588	.parameters	= fuse_fs_parameters,
1589	.kill_sb	= fuse_kill_sb_anon,
1590};
1591MODULE_ALIAS_FS("fuse");
1592
1593#ifdef CONFIG_BLOCK
1594static void fuse_kill_sb_blk(struct super_block *sb)
1595{
1596	struct fuse_mount *fm = get_fuse_mount_super(sb);
1597	bool last;
1598
1599	if (sb->s_root) {
1600		last = fuse_mount_remove(fm);
1601		if (last)
1602			fuse_conn_destroy(fm);
1603	}
1604	kill_block_super(sb);
1605}
1606
1607static struct file_system_type fuseblk_fs_type = {
1608	.owner		= THIS_MODULE,
1609	.name		= "fuseblk",
1610	.init_fs_context = fuse_init_fs_context,
1611	.parameters	= fuse_fs_parameters,
1612	.kill_sb	= fuse_kill_sb_blk,
1613	.fs_flags	= FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1614};
1615MODULE_ALIAS_FS("fuseblk");
1616
1617static inline int register_fuseblk(void)
1618{
1619	return register_filesystem(&fuseblk_fs_type);
1620}
1621
1622static inline void unregister_fuseblk(void)
1623{
1624	unregister_filesystem(&fuseblk_fs_type);
1625}
1626#else
1627static inline int register_fuseblk(void)
1628{
1629	return 0;
1630}
1631
1632static inline void unregister_fuseblk(void)
1633{
1634}
1635#endif
1636
1637static void fuse_inode_init_once(void *foo)
1638{
1639	struct inode *inode = foo;
1640
1641	inode_init_once(inode);
1642}
1643
1644static int __init fuse_fs_init(void)
1645{
1646	int err;
1647
1648	fuse_inode_cachep = kmem_cache_create("fuse_inode",
1649			sizeof(struct fuse_inode), 0,
1650			SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1651			fuse_inode_init_once);
1652	err = -ENOMEM;
1653	if (!fuse_inode_cachep)
1654		goto out;
1655
1656	err = register_fuseblk();
1657	if (err)
1658		goto out2;
1659
1660	err = register_filesystem(&fuse_fs_type);
1661	if (err)
1662		goto out3;
1663
1664	return 0;
1665
1666 out3:
1667	unregister_fuseblk();
1668 out2:
1669	kmem_cache_destroy(fuse_inode_cachep);
1670 out:
1671	return err;
1672}
1673
1674static void fuse_fs_cleanup(void)
1675{
1676	unregister_filesystem(&fuse_fs_type);
1677	unregister_fuseblk();
1678
1679	/*
1680	 * Make sure all delayed rcu free inodes are flushed before we
1681	 * destroy cache.
1682	 */
1683	rcu_barrier();
1684	kmem_cache_destroy(fuse_inode_cachep);
1685}
1686
1687static struct kobject *fuse_kobj;
1688
1689static int fuse_sysfs_init(void)
1690{
1691	int err;
1692
1693	fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1694	if (!fuse_kobj) {
1695		err = -ENOMEM;
1696		goto out_err;
1697	}
1698
1699	err = sysfs_create_mount_point(fuse_kobj, "connections");
1700	if (err)
1701		goto out_fuse_unregister;
1702
1703	return 0;
1704
1705 out_fuse_unregister:
1706	kobject_put(fuse_kobj);
1707 out_err:
1708	return err;
1709}
1710
1711static void fuse_sysfs_cleanup(void)
1712{
1713	sysfs_remove_mount_point(fuse_kobj, "connections");
1714	kobject_put(fuse_kobj);
1715}
1716
1717static int __init fuse_init(void)
1718{
1719	int res;
1720
1721	pr_info("init (API version %i.%i)\n",
1722		FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1723
1724	INIT_LIST_HEAD(&fuse_conn_list);
1725	res = fuse_fs_init();
1726	if (res)
1727		goto err;
1728
1729	res = fuse_dev_init();
1730	if (res)
1731		goto err_fs_cleanup;
1732
1733	res = fuse_sysfs_init();
1734	if (res)
1735		goto err_dev_cleanup;
1736
1737	res = fuse_ctl_init();
1738	if (res)
1739		goto err_sysfs_cleanup;
1740
1741	sanitize_global_limit(&max_user_bgreq);
1742	sanitize_global_limit(&max_user_congthresh);
1743
1744	return 0;
1745
1746 err_sysfs_cleanup:
1747	fuse_sysfs_cleanup();
1748 err_dev_cleanup:
1749	fuse_dev_cleanup();
1750 err_fs_cleanup:
1751	fuse_fs_cleanup();
1752 err:
1753	return res;
1754}
1755
1756static void __exit fuse_exit(void)
1757{
1758	pr_debug("exit\n");
1759
1760	fuse_ctl_cleanup();
1761	fuse_sysfs_cleanup();
1762	fuse_fs_cleanup();
1763	fuse_dev_cleanup();
1764}
1765
1766module_init(fuse_init);
1767module_exit(fuse_exit);
1768