xref: /linux/fs/fuse/dir.c (revision d5acba26bfa097a618be425522b1ec4269d3edaf)
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/file.h>
13 #include <linux/sched.h>
14 #include <linux/namei.h>
15 #include <linux/slab.h>
16 #include <linux/xattr.h>
17 #include <linux/posix_acl.h>
18 
19 static bool fuse_use_readdirplus(struct inode *dir, struct dir_context *ctx)
20 {
21 	struct fuse_conn *fc = get_fuse_conn(dir);
22 	struct fuse_inode *fi = get_fuse_inode(dir);
23 
24 	if (!fc->do_readdirplus)
25 		return false;
26 	if (!fc->readdirplus_auto)
27 		return true;
28 	if (test_and_clear_bit(FUSE_I_ADVISE_RDPLUS, &fi->state))
29 		return true;
30 	if (ctx->pos == 0)
31 		return true;
32 	return false;
33 }
34 
35 static void fuse_advise_use_readdirplus(struct inode *dir)
36 {
37 	struct fuse_inode *fi = get_fuse_inode(dir);
38 
39 	set_bit(FUSE_I_ADVISE_RDPLUS, &fi->state);
40 }
41 
42 union fuse_dentry {
43 	u64 time;
44 	struct rcu_head rcu;
45 };
46 
47 static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
48 {
49 	((union fuse_dentry *) entry->d_fsdata)->time = time;
50 }
51 
52 static inline u64 fuse_dentry_time(struct dentry *entry)
53 {
54 	return ((union fuse_dentry *) entry->d_fsdata)->time;
55 }
56 
57 /*
58  * FUSE caches dentries and attributes with separate timeout.  The
59  * time in jiffies until the dentry/attributes are valid is stored in
60  * dentry->d_fsdata and fuse_inode->i_time respectively.
61  */
62 
63 /*
64  * Calculate the time in jiffies until a dentry/attributes are valid
65  */
66 static u64 time_to_jiffies(u64 sec, u32 nsec)
67 {
68 	if (sec || nsec) {
69 		struct timespec64 ts = {
70 			sec,
71 			min_t(u32, nsec, NSEC_PER_SEC - 1)
72 		};
73 
74 		return get_jiffies_64() + timespec64_to_jiffies(&ts);
75 	} else
76 		return 0;
77 }
78 
79 /*
80  * Set dentry and possibly attribute timeouts from the lookup/mk*
81  * replies
82  */
83 static void fuse_change_entry_timeout(struct dentry *entry,
84 				      struct fuse_entry_out *o)
85 {
86 	fuse_dentry_settime(entry,
87 		time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
88 }
89 
90 static u64 attr_timeout(struct fuse_attr_out *o)
91 {
92 	return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
93 }
94 
95 static u64 entry_attr_timeout(struct fuse_entry_out *o)
96 {
97 	return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
98 }
99 
100 /*
101  * Mark the attributes as stale, so that at the next call to
102  * ->getattr() they will be fetched from userspace
103  */
104 void fuse_invalidate_attr(struct inode *inode)
105 {
106 	get_fuse_inode(inode)->i_time = 0;
107 }
108 
109 /**
110  * Mark the attributes as stale due to an atime change.  Avoid the invalidate if
111  * atime is not used.
112  */
113 void fuse_invalidate_atime(struct inode *inode)
114 {
115 	if (!IS_RDONLY(inode))
116 		fuse_invalidate_attr(inode);
117 }
118 
119 /*
120  * Just mark the entry as stale, so that a next attempt to look it up
121  * will result in a new lookup call to userspace
122  *
123  * This is called when a dentry is about to become negative and the
124  * timeout is unknown (unlink, rmdir, rename and in some cases
125  * lookup)
126  */
127 void fuse_invalidate_entry_cache(struct dentry *entry)
128 {
129 	fuse_dentry_settime(entry, 0);
130 }
131 
132 /*
133  * Same as fuse_invalidate_entry_cache(), but also try to remove the
134  * dentry from the hash
135  */
136 static void fuse_invalidate_entry(struct dentry *entry)
137 {
138 	d_invalidate(entry);
139 	fuse_invalidate_entry_cache(entry);
140 }
141 
142 static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_args *args,
143 			     u64 nodeid, const struct qstr *name,
144 			     struct fuse_entry_out *outarg)
145 {
146 	memset(outarg, 0, sizeof(struct fuse_entry_out));
147 	args->in.h.opcode = FUSE_LOOKUP;
148 	args->in.h.nodeid = nodeid;
149 	args->in.numargs = 1;
150 	args->in.args[0].size = name->len + 1;
151 	args->in.args[0].value = name->name;
152 	args->out.numargs = 1;
153 	args->out.args[0].size = sizeof(struct fuse_entry_out);
154 	args->out.args[0].value = outarg;
155 }
156 
157 u64 fuse_get_attr_version(struct fuse_conn *fc)
158 {
159 	u64 curr_version;
160 
161 	/*
162 	 * The spin lock isn't actually needed on 64bit archs, but we
163 	 * don't yet care too much about such optimizations.
164 	 */
165 	spin_lock(&fc->lock);
166 	curr_version = fc->attr_version;
167 	spin_unlock(&fc->lock);
168 
169 	return curr_version;
170 }
171 
172 /*
173  * Check whether the dentry is still valid
174  *
175  * If the entry validity timeout has expired and the dentry is
176  * positive, try to redo the lookup.  If the lookup results in a
177  * different inode, then let the VFS invalidate the dentry and redo
178  * the lookup once more.  If the lookup results in the same inode,
179  * then refresh the attributes, timeouts and mark the dentry valid.
180  */
181 static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
182 {
183 	struct inode *inode;
184 	struct dentry *parent;
185 	struct fuse_conn *fc;
186 	struct fuse_inode *fi;
187 	int ret;
188 
189 	inode = d_inode_rcu(entry);
190 	if (inode && is_bad_inode(inode))
191 		goto invalid;
192 	else if (time_before64(fuse_dentry_time(entry), get_jiffies_64()) ||
193 		 (flags & LOOKUP_REVAL)) {
194 		struct fuse_entry_out outarg;
195 		FUSE_ARGS(args);
196 		struct fuse_forget_link *forget;
197 		u64 attr_version;
198 
199 		/* For negative dentries, always do a fresh lookup */
200 		if (!inode)
201 			goto invalid;
202 
203 		ret = -ECHILD;
204 		if (flags & LOOKUP_RCU)
205 			goto out;
206 
207 		fc = get_fuse_conn(inode);
208 
209 		forget = fuse_alloc_forget();
210 		ret = -ENOMEM;
211 		if (!forget)
212 			goto out;
213 
214 		attr_version = fuse_get_attr_version(fc);
215 
216 		parent = dget_parent(entry);
217 		fuse_lookup_init(fc, &args, get_node_id(d_inode(parent)),
218 				 &entry->d_name, &outarg);
219 		ret = fuse_simple_request(fc, &args);
220 		dput(parent);
221 		/* Zero nodeid is same as -ENOENT */
222 		if (!ret && !outarg.nodeid)
223 			ret = -ENOENT;
224 		if (!ret) {
225 			fi = get_fuse_inode(inode);
226 			if (outarg.nodeid != get_node_id(inode)) {
227 				fuse_queue_forget(fc, forget, outarg.nodeid, 1);
228 				goto invalid;
229 			}
230 			spin_lock(&fc->lock);
231 			fi->nlookup++;
232 			spin_unlock(&fc->lock);
233 		}
234 		kfree(forget);
235 		if (ret == -ENOMEM)
236 			goto out;
237 		if (ret || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
238 			goto invalid;
239 
240 		forget_all_cached_acls(inode);
241 		fuse_change_attributes(inode, &outarg.attr,
242 				       entry_attr_timeout(&outarg),
243 				       attr_version);
244 		fuse_change_entry_timeout(entry, &outarg);
245 	} else if (inode) {
246 		fi = get_fuse_inode(inode);
247 		if (flags & LOOKUP_RCU) {
248 			if (test_bit(FUSE_I_INIT_RDPLUS, &fi->state))
249 				return -ECHILD;
250 		} else if (test_and_clear_bit(FUSE_I_INIT_RDPLUS, &fi->state)) {
251 			parent = dget_parent(entry);
252 			fuse_advise_use_readdirplus(d_inode(parent));
253 			dput(parent);
254 		}
255 	}
256 	ret = 1;
257 out:
258 	return ret;
259 
260 invalid:
261 	ret = 0;
262 	goto out;
263 }
264 
265 static int invalid_nodeid(u64 nodeid)
266 {
267 	return !nodeid || nodeid == FUSE_ROOT_ID;
268 }
269 
270 static int fuse_dentry_init(struct dentry *dentry)
271 {
272 	dentry->d_fsdata = kzalloc(sizeof(union fuse_dentry), GFP_KERNEL);
273 
274 	return dentry->d_fsdata ? 0 : -ENOMEM;
275 }
276 static void fuse_dentry_release(struct dentry *dentry)
277 {
278 	union fuse_dentry *fd = dentry->d_fsdata;
279 
280 	kfree_rcu(fd, rcu);
281 }
282 
283 const struct dentry_operations fuse_dentry_operations = {
284 	.d_revalidate	= fuse_dentry_revalidate,
285 	.d_init		= fuse_dentry_init,
286 	.d_release	= fuse_dentry_release,
287 };
288 
289 const struct dentry_operations fuse_root_dentry_operations = {
290 	.d_init		= fuse_dentry_init,
291 	.d_release	= fuse_dentry_release,
292 };
293 
294 int fuse_valid_type(int m)
295 {
296 	return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
297 		S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
298 }
299 
300 int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
301 		     struct fuse_entry_out *outarg, struct inode **inode)
302 {
303 	struct fuse_conn *fc = get_fuse_conn_super(sb);
304 	FUSE_ARGS(args);
305 	struct fuse_forget_link *forget;
306 	u64 attr_version;
307 	int err;
308 
309 	*inode = NULL;
310 	err = -ENAMETOOLONG;
311 	if (name->len > FUSE_NAME_MAX)
312 		goto out;
313 
314 
315 	forget = fuse_alloc_forget();
316 	err = -ENOMEM;
317 	if (!forget)
318 		goto out;
319 
320 	attr_version = fuse_get_attr_version(fc);
321 
322 	fuse_lookup_init(fc, &args, nodeid, name, outarg);
323 	err = fuse_simple_request(fc, &args);
324 	/* Zero nodeid is same as -ENOENT, but with valid timeout */
325 	if (err || !outarg->nodeid)
326 		goto out_put_forget;
327 
328 	err = -EIO;
329 	if (!outarg->nodeid)
330 		goto out_put_forget;
331 	if (!fuse_valid_type(outarg->attr.mode))
332 		goto out_put_forget;
333 
334 	*inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
335 			   &outarg->attr, entry_attr_timeout(outarg),
336 			   attr_version);
337 	err = -ENOMEM;
338 	if (!*inode) {
339 		fuse_queue_forget(fc, forget, outarg->nodeid, 1);
340 		goto out;
341 	}
342 	err = 0;
343 
344  out_put_forget:
345 	kfree(forget);
346  out:
347 	return err;
348 }
349 
350 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
351 				  unsigned int flags)
352 {
353 	int err;
354 	struct fuse_entry_out outarg;
355 	struct inode *inode;
356 	struct dentry *newent;
357 	bool outarg_valid = true;
358 
359 	fuse_lock_inode(dir);
360 	err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
361 			       &outarg, &inode);
362 	fuse_unlock_inode(dir);
363 	if (err == -ENOENT) {
364 		outarg_valid = false;
365 		err = 0;
366 	}
367 	if (err)
368 		goto out_err;
369 
370 	err = -EIO;
371 	if (inode && get_node_id(inode) == FUSE_ROOT_ID)
372 		goto out_iput;
373 
374 	newent = d_splice_alias(inode, entry);
375 	err = PTR_ERR(newent);
376 	if (IS_ERR(newent))
377 		goto out_err;
378 
379 	entry = newent ? newent : entry;
380 	if (outarg_valid)
381 		fuse_change_entry_timeout(entry, &outarg);
382 	else
383 		fuse_invalidate_entry_cache(entry);
384 
385 	fuse_advise_use_readdirplus(dir);
386 	return newent;
387 
388  out_iput:
389 	iput(inode);
390  out_err:
391 	return ERR_PTR(err);
392 }
393 
394 /*
395  * Atomic create+open operation
396  *
397  * If the filesystem doesn't support this, then fall back to separate
398  * 'mknod' + 'open' requests.
399  */
400 static int fuse_create_open(struct inode *dir, struct dentry *entry,
401 			    struct file *file, unsigned flags,
402 			    umode_t mode)
403 {
404 	int err;
405 	struct inode *inode;
406 	struct fuse_conn *fc = get_fuse_conn(dir);
407 	FUSE_ARGS(args);
408 	struct fuse_forget_link *forget;
409 	struct fuse_create_in inarg;
410 	struct fuse_open_out outopen;
411 	struct fuse_entry_out outentry;
412 	struct fuse_file *ff;
413 
414 	/* Userspace expects S_IFREG in create mode */
415 	BUG_ON((mode & S_IFMT) != S_IFREG);
416 
417 	forget = fuse_alloc_forget();
418 	err = -ENOMEM;
419 	if (!forget)
420 		goto out_err;
421 
422 	err = -ENOMEM;
423 	ff = fuse_file_alloc(fc);
424 	if (!ff)
425 		goto out_put_forget_req;
426 
427 	if (!fc->dont_mask)
428 		mode &= ~current_umask();
429 
430 	flags &= ~O_NOCTTY;
431 	memset(&inarg, 0, sizeof(inarg));
432 	memset(&outentry, 0, sizeof(outentry));
433 	inarg.flags = flags;
434 	inarg.mode = mode;
435 	inarg.umask = current_umask();
436 	args.in.h.opcode = FUSE_CREATE;
437 	args.in.h.nodeid = get_node_id(dir);
438 	args.in.numargs = 2;
439 	args.in.args[0].size = sizeof(inarg);
440 	args.in.args[0].value = &inarg;
441 	args.in.args[1].size = entry->d_name.len + 1;
442 	args.in.args[1].value = entry->d_name.name;
443 	args.out.numargs = 2;
444 	args.out.args[0].size = sizeof(outentry);
445 	args.out.args[0].value = &outentry;
446 	args.out.args[1].size = sizeof(outopen);
447 	args.out.args[1].value = &outopen;
448 	err = fuse_simple_request(fc, &args);
449 	if (err)
450 		goto out_free_ff;
451 
452 	err = -EIO;
453 	if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
454 		goto out_free_ff;
455 
456 	ff->fh = outopen.fh;
457 	ff->nodeid = outentry.nodeid;
458 	ff->open_flags = outopen.open_flags;
459 	inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
460 			  &outentry.attr, entry_attr_timeout(&outentry), 0);
461 	if (!inode) {
462 		flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
463 		fuse_sync_release(ff, flags);
464 		fuse_queue_forget(fc, forget, outentry.nodeid, 1);
465 		err = -ENOMEM;
466 		goto out_err;
467 	}
468 	kfree(forget);
469 	d_instantiate(entry, inode);
470 	fuse_change_entry_timeout(entry, &outentry);
471 	fuse_invalidate_attr(dir);
472 	err = finish_open(file, entry, generic_file_open);
473 	if (err) {
474 		fuse_sync_release(ff, flags);
475 	} else {
476 		file->private_data = ff;
477 		fuse_finish_open(inode, file);
478 	}
479 	return err;
480 
481 out_free_ff:
482 	fuse_file_free(ff);
483 out_put_forget_req:
484 	kfree(forget);
485 out_err:
486 	return err;
487 }
488 
489 static int fuse_mknod(struct inode *, struct dentry *, umode_t, dev_t);
490 static int fuse_atomic_open(struct inode *dir, struct dentry *entry,
491 			    struct file *file, unsigned flags,
492 			    umode_t mode)
493 {
494 	int err;
495 	struct fuse_conn *fc = get_fuse_conn(dir);
496 	struct dentry *res = NULL;
497 
498 	if (d_in_lookup(entry)) {
499 		res = fuse_lookup(dir, entry, 0);
500 		if (IS_ERR(res))
501 			return PTR_ERR(res);
502 
503 		if (res)
504 			entry = res;
505 	}
506 
507 	if (!(flags & O_CREAT) || d_really_is_positive(entry))
508 		goto no_open;
509 
510 	/* Only creates */
511 	file->f_mode |= FMODE_CREATED;
512 
513 	if (fc->no_create)
514 		goto mknod;
515 
516 	err = fuse_create_open(dir, entry, file, flags, mode);
517 	if (err == -ENOSYS) {
518 		fc->no_create = 1;
519 		goto mknod;
520 	}
521 out_dput:
522 	dput(res);
523 	return err;
524 
525 mknod:
526 	err = fuse_mknod(dir, entry, mode, 0);
527 	if (err)
528 		goto out_dput;
529 no_open:
530 	return finish_no_open(file, res);
531 }
532 
533 /*
534  * Code shared between mknod, mkdir, symlink and link
535  */
536 static int create_new_entry(struct fuse_conn *fc, struct fuse_args *args,
537 			    struct inode *dir, struct dentry *entry,
538 			    umode_t mode)
539 {
540 	struct fuse_entry_out outarg;
541 	struct inode *inode;
542 	struct dentry *d;
543 	int err;
544 	struct fuse_forget_link *forget;
545 
546 	forget = fuse_alloc_forget();
547 	if (!forget)
548 		return -ENOMEM;
549 
550 	memset(&outarg, 0, sizeof(outarg));
551 	args->in.h.nodeid = get_node_id(dir);
552 	args->out.numargs = 1;
553 	args->out.args[0].size = sizeof(outarg);
554 	args->out.args[0].value = &outarg;
555 	err = fuse_simple_request(fc, args);
556 	if (err)
557 		goto out_put_forget_req;
558 
559 	err = -EIO;
560 	if (invalid_nodeid(outarg.nodeid))
561 		goto out_put_forget_req;
562 
563 	if ((outarg.attr.mode ^ mode) & S_IFMT)
564 		goto out_put_forget_req;
565 
566 	inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
567 			  &outarg.attr, entry_attr_timeout(&outarg), 0);
568 	if (!inode) {
569 		fuse_queue_forget(fc, forget, outarg.nodeid, 1);
570 		return -ENOMEM;
571 	}
572 	kfree(forget);
573 
574 	d_drop(entry);
575 	d = d_splice_alias(inode, entry);
576 	if (IS_ERR(d))
577 		return PTR_ERR(d);
578 
579 	if (d) {
580 		fuse_change_entry_timeout(d, &outarg);
581 		dput(d);
582 	} else {
583 		fuse_change_entry_timeout(entry, &outarg);
584 	}
585 	fuse_invalidate_attr(dir);
586 	return 0;
587 
588  out_put_forget_req:
589 	kfree(forget);
590 	return err;
591 }
592 
593 static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
594 		      dev_t rdev)
595 {
596 	struct fuse_mknod_in inarg;
597 	struct fuse_conn *fc = get_fuse_conn(dir);
598 	FUSE_ARGS(args);
599 
600 	if (!fc->dont_mask)
601 		mode &= ~current_umask();
602 
603 	memset(&inarg, 0, sizeof(inarg));
604 	inarg.mode = mode;
605 	inarg.rdev = new_encode_dev(rdev);
606 	inarg.umask = current_umask();
607 	args.in.h.opcode = FUSE_MKNOD;
608 	args.in.numargs = 2;
609 	args.in.args[0].size = sizeof(inarg);
610 	args.in.args[0].value = &inarg;
611 	args.in.args[1].size = entry->d_name.len + 1;
612 	args.in.args[1].value = entry->d_name.name;
613 	return create_new_entry(fc, &args, dir, entry, mode);
614 }
615 
616 static int fuse_create(struct inode *dir, struct dentry *entry, umode_t mode,
617 		       bool excl)
618 {
619 	return fuse_mknod(dir, entry, mode, 0);
620 }
621 
622 static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
623 {
624 	struct fuse_mkdir_in inarg;
625 	struct fuse_conn *fc = get_fuse_conn(dir);
626 	FUSE_ARGS(args);
627 
628 	if (!fc->dont_mask)
629 		mode &= ~current_umask();
630 
631 	memset(&inarg, 0, sizeof(inarg));
632 	inarg.mode = mode;
633 	inarg.umask = current_umask();
634 	args.in.h.opcode = FUSE_MKDIR;
635 	args.in.numargs = 2;
636 	args.in.args[0].size = sizeof(inarg);
637 	args.in.args[0].value = &inarg;
638 	args.in.args[1].size = entry->d_name.len + 1;
639 	args.in.args[1].value = entry->d_name.name;
640 	return create_new_entry(fc, &args, dir, entry, S_IFDIR);
641 }
642 
643 static int fuse_symlink(struct inode *dir, struct dentry *entry,
644 			const char *link)
645 {
646 	struct fuse_conn *fc = get_fuse_conn(dir);
647 	unsigned len = strlen(link) + 1;
648 	FUSE_ARGS(args);
649 
650 	args.in.h.opcode = FUSE_SYMLINK;
651 	args.in.numargs = 2;
652 	args.in.args[0].size = entry->d_name.len + 1;
653 	args.in.args[0].value = entry->d_name.name;
654 	args.in.args[1].size = len;
655 	args.in.args[1].value = link;
656 	return create_new_entry(fc, &args, dir, entry, S_IFLNK);
657 }
658 
659 void fuse_update_ctime(struct inode *inode)
660 {
661 	if (!IS_NOCMTIME(inode)) {
662 		inode->i_ctime = current_time(inode);
663 		mark_inode_dirty_sync(inode);
664 	}
665 }
666 
667 static int fuse_unlink(struct inode *dir, struct dentry *entry)
668 {
669 	int err;
670 	struct fuse_conn *fc = get_fuse_conn(dir);
671 	FUSE_ARGS(args);
672 
673 	args.in.h.opcode = FUSE_UNLINK;
674 	args.in.h.nodeid = get_node_id(dir);
675 	args.in.numargs = 1;
676 	args.in.args[0].size = entry->d_name.len + 1;
677 	args.in.args[0].value = entry->d_name.name;
678 	err = fuse_simple_request(fc, &args);
679 	if (!err) {
680 		struct inode *inode = d_inode(entry);
681 		struct fuse_inode *fi = get_fuse_inode(inode);
682 
683 		spin_lock(&fc->lock);
684 		fi->attr_version = ++fc->attr_version;
685 		/*
686 		 * If i_nlink == 0 then unlink doesn't make sense, yet this can
687 		 * happen if userspace filesystem is careless.  It would be
688 		 * difficult to enforce correct nlink usage so just ignore this
689 		 * condition here
690 		 */
691 		if (inode->i_nlink > 0)
692 			drop_nlink(inode);
693 		spin_unlock(&fc->lock);
694 		fuse_invalidate_attr(inode);
695 		fuse_invalidate_attr(dir);
696 		fuse_invalidate_entry_cache(entry);
697 		fuse_update_ctime(inode);
698 	} else if (err == -EINTR)
699 		fuse_invalidate_entry(entry);
700 	return err;
701 }
702 
703 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
704 {
705 	int err;
706 	struct fuse_conn *fc = get_fuse_conn(dir);
707 	FUSE_ARGS(args);
708 
709 	args.in.h.opcode = FUSE_RMDIR;
710 	args.in.h.nodeid = get_node_id(dir);
711 	args.in.numargs = 1;
712 	args.in.args[0].size = entry->d_name.len + 1;
713 	args.in.args[0].value = entry->d_name.name;
714 	err = fuse_simple_request(fc, &args);
715 	if (!err) {
716 		clear_nlink(d_inode(entry));
717 		fuse_invalidate_attr(dir);
718 		fuse_invalidate_entry_cache(entry);
719 	} else if (err == -EINTR)
720 		fuse_invalidate_entry(entry);
721 	return err;
722 }
723 
724 static int fuse_rename_common(struct inode *olddir, struct dentry *oldent,
725 			      struct inode *newdir, struct dentry *newent,
726 			      unsigned int flags, int opcode, size_t argsize)
727 {
728 	int err;
729 	struct fuse_rename2_in inarg;
730 	struct fuse_conn *fc = get_fuse_conn(olddir);
731 	FUSE_ARGS(args);
732 
733 	memset(&inarg, 0, argsize);
734 	inarg.newdir = get_node_id(newdir);
735 	inarg.flags = flags;
736 	args.in.h.opcode = opcode;
737 	args.in.h.nodeid = get_node_id(olddir);
738 	args.in.numargs = 3;
739 	args.in.args[0].size = argsize;
740 	args.in.args[0].value = &inarg;
741 	args.in.args[1].size = oldent->d_name.len + 1;
742 	args.in.args[1].value = oldent->d_name.name;
743 	args.in.args[2].size = newent->d_name.len + 1;
744 	args.in.args[2].value = newent->d_name.name;
745 	err = fuse_simple_request(fc, &args);
746 	if (!err) {
747 		/* ctime changes */
748 		fuse_invalidate_attr(d_inode(oldent));
749 		fuse_update_ctime(d_inode(oldent));
750 
751 		if (flags & RENAME_EXCHANGE) {
752 			fuse_invalidate_attr(d_inode(newent));
753 			fuse_update_ctime(d_inode(newent));
754 		}
755 
756 		fuse_invalidate_attr(olddir);
757 		if (olddir != newdir)
758 			fuse_invalidate_attr(newdir);
759 
760 		/* newent will end up negative */
761 		if (!(flags & RENAME_EXCHANGE) && d_really_is_positive(newent)) {
762 			fuse_invalidate_attr(d_inode(newent));
763 			fuse_invalidate_entry_cache(newent);
764 			fuse_update_ctime(d_inode(newent));
765 		}
766 	} else if (err == -EINTR) {
767 		/* If request was interrupted, DEITY only knows if the
768 		   rename actually took place.  If the invalidation
769 		   fails (e.g. some process has CWD under the renamed
770 		   directory), then there can be inconsistency between
771 		   the dcache and the real filesystem.  Tough luck. */
772 		fuse_invalidate_entry(oldent);
773 		if (d_really_is_positive(newent))
774 			fuse_invalidate_entry(newent);
775 	}
776 
777 	return err;
778 }
779 
780 static int fuse_rename2(struct inode *olddir, struct dentry *oldent,
781 			struct inode *newdir, struct dentry *newent,
782 			unsigned int flags)
783 {
784 	struct fuse_conn *fc = get_fuse_conn(olddir);
785 	int err;
786 
787 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
788 		return -EINVAL;
789 
790 	if (flags) {
791 		if (fc->no_rename2 || fc->minor < 23)
792 			return -EINVAL;
793 
794 		err = fuse_rename_common(olddir, oldent, newdir, newent, flags,
795 					 FUSE_RENAME2,
796 					 sizeof(struct fuse_rename2_in));
797 		if (err == -ENOSYS) {
798 			fc->no_rename2 = 1;
799 			err = -EINVAL;
800 		}
801 	} else {
802 		err = fuse_rename_common(olddir, oldent, newdir, newent, 0,
803 					 FUSE_RENAME,
804 					 sizeof(struct fuse_rename_in));
805 	}
806 
807 	return err;
808 }
809 
810 static int fuse_link(struct dentry *entry, struct inode *newdir,
811 		     struct dentry *newent)
812 {
813 	int err;
814 	struct fuse_link_in inarg;
815 	struct inode *inode = d_inode(entry);
816 	struct fuse_conn *fc = get_fuse_conn(inode);
817 	FUSE_ARGS(args);
818 
819 	memset(&inarg, 0, sizeof(inarg));
820 	inarg.oldnodeid = get_node_id(inode);
821 	args.in.h.opcode = FUSE_LINK;
822 	args.in.numargs = 2;
823 	args.in.args[0].size = sizeof(inarg);
824 	args.in.args[0].value = &inarg;
825 	args.in.args[1].size = newent->d_name.len + 1;
826 	args.in.args[1].value = newent->d_name.name;
827 	err = create_new_entry(fc, &args, newdir, newent, inode->i_mode);
828 	/* Contrary to "normal" filesystems it can happen that link
829 	   makes two "logical" inodes point to the same "physical"
830 	   inode.  We invalidate the attributes of the old one, so it
831 	   will reflect changes in the backing inode (link count,
832 	   etc.)
833 	*/
834 	if (!err) {
835 		struct fuse_inode *fi = get_fuse_inode(inode);
836 
837 		spin_lock(&fc->lock);
838 		fi->attr_version = ++fc->attr_version;
839 		inc_nlink(inode);
840 		spin_unlock(&fc->lock);
841 		fuse_invalidate_attr(inode);
842 		fuse_update_ctime(inode);
843 	} else if (err == -EINTR) {
844 		fuse_invalidate_attr(inode);
845 	}
846 	return err;
847 }
848 
849 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
850 			  struct kstat *stat)
851 {
852 	unsigned int blkbits;
853 	struct fuse_conn *fc = get_fuse_conn(inode);
854 
855 	/* see the comment in fuse_change_attributes() */
856 	if (fc->writeback_cache && S_ISREG(inode->i_mode)) {
857 		attr->size = i_size_read(inode);
858 		attr->mtime = inode->i_mtime.tv_sec;
859 		attr->mtimensec = inode->i_mtime.tv_nsec;
860 		attr->ctime = inode->i_ctime.tv_sec;
861 		attr->ctimensec = inode->i_ctime.tv_nsec;
862 	}
863 
864 	stat->dev = inode->i_sb->s_dev;
865 	stat->ino = attr->ino;
866 	stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
867 	stat->nlink = attr->nlink;
868 	stat->uid = make_kuid(fc->user_ns, attr->uid);
869 	stat->gid = make_kgid(fc->user_ns, attr->gid);
870 	stat->rdev = inode->i_rdev;
871 	stat->atime.tv_sec = attr->atime;
872 	stat->atime.tv_nsec = attr->atimensec;
873 	stat->mtime.tv_sec = attr->mtime;
874 	stat->mtime.tv_nsec = attr->mtimensec;
875 	stat->ctime.tv_sec = attr->ctime;
876 	stat->ctime.tv_nsec = attr->ctimensec;
877 	stat->size = attr->size;
878 	stat->blocks = attr->blocks;
879 
880 	if (attr->blksize != 0)
881 		blkbits = ilog2(attr->blksize);
882 	else
883 		blkbits = inode->i_sb->s_blocksize_bits;
884 
885 	stat->blksize = 1 << blkbits;
886 }
887 
888 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
889 			   struct file *file)
890 {
891 	int err;
892 	struct fuse_getattr_in inarg;
893 	struct fuse_attr_out outarg;
894 	struct fuse_conn *fc = get_fuse_conn(inode);
895 	FUSE_ARGS(args);
896 	u64 attr_version;
897 
898 	attr_version = fuse_get_attr_version(fc);
899 
900 	memset(&inarg, 0, sizeof(inarg));
901 	memset(&outarg, 0, sizeof(outarg));
902 	/* Directories have separate file-handle space */
903 	if (file && S_ISREG(inode->i_mode)) {
904 		struct fuse_file *ff = file->private_data;
905 
906 		inarg.getattr_flags |= FUSE_GETATTR_FH;
907 		inarg.fh = ff->fh;
908 	}
909 	args.in.h.opcode = FUSE_GETATTR;
910 	args.in.h.nodeid = get_node_id(inode);
911 	args.in.numargs = 1;
912 	args.in.args[0].size = sizeof(inarg);
913 	args.in.args[0].value = &inarg;
914 	args.out.numargs = 1;
915 	args.out.args[0].size = sizeof(outarg);
916 	args.out.args[0].value = &outarg;
917 	err = fuse_simple_request(fc, &args);
918 	if (!err) {
919 		if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
920 			make_bad_inode(inode);
921 			err = -EIO;
922 		} else {
923 			fuse_change_attributes(inode, &outarg.attr,
924 					       attr_timeout(&outarg),
925 					       attr_version);
926 			if (stat)
927 				fuse_fillattr(inode, &outarg.attr, stat);
928 		}
929 	}
930 	return err;
931 }
932 
933 static int fuse_update_get_attr(struct inode *inode, struct file *file,
934 				struct kstat *stat, unsigned int flags)
935 {
936 	struct fuse_inode *fi = get_fuse_inode(inode);
937 	int err = 0;
938 	bool sync;
939 
940 	if (flags & AT_STATX_FORCE_SYNC)
941 		sync = true;
942 	else if (flags & AT_STATX_DONT_SYNC)
943 		sync = false;
944 	else
945 		sync = time_before64(fi->i_time, get_jiffies_64());
946 
947 	if (sync) {
948 		forget_all_cached_acls(inode);
949 		err = fuse_do_getattr(inode, stat, file);
950 	} else if (stat) {
951 		generic_fillattr(inode, stat);
952 		stat->mode = fi->orig_i_mode;
953 		stat->ino = fi->orig_ino;
954 	}
955 
956 	return err;
957 }
958 
959 int fuse_update_attributes(struct inode *inode, struct file *file)
960 {
961 	return fuse_update_get_attr(inode, file, NULL, 0);
962 }
963 
964 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
965 			     u64 child_nodeid, struct qstr *name)
966 {
967 	int err = -ENOTDIR;
968 	struct inode *parent;
969 	struct dentry *dir;
970 	struct dentry *entry;
971 
972 	parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
973 	if (!parent)
974 		return -ENOENT;
975 
976 	inode_lock(parent);
977 	if (!S_ISDIR(parent->i_mode))
978 		goto unlock;
979 
980 	err = -ENOENT;
981 	dir = d_find_alias(parent);
982 	if (!dir)
983 		goto unlock;
984 
985 	name->hash = full_name_hash(dir, name->name, name->len);
986 	entry = d_lookup(dir, name);
987 	dput(dir);
988 	if (!entry)
989 		goto unlock;
990 
991 	fuse_invalidate_attr(parent);
992 	fuse_invalidate_entry(entry);
993 
994 	if (child_nodeid != 0 && d_really_is_positive(entry)) {
995 		inode_lock(d_inode(entry));
996 		if (get_node_id(d_inode(entry)) != child_nodeid) {
997 			err = -ENOENT;
998 			goto badentry;
999 		}
1000 		if (d_mountpoint(entry)) {
1001 			err = -EBUSY;
1002 			goto badentry;
1003 		}
1004 		if (d_is_dir(entry)) {
1005 			shrink_dcache_parent(entry);
1006 			if (!simple_empty(entry)) {
1007 				err = -ENOTEMPTY;
1008 				goto badentry;
1009 			}
1010 			d_inode(entry)->i_flags |= S_DEAD;
1011 		}
1012 		dont_mount(entry);
1013 		clear_nlink(d_inode(entry));
1014 		err = 0;
1015  badentry:
1016 		inode_unlock(d_inode(entry));
1017 		if (!err)
1018 			d_delete(entry);
1019 	} else {
1020 		err = 0;
1021 	}
1022 	dput(entry);
1023 
1024  unlock:
1025 	inode_unlock(parent);
1026 	iput(parent);
1027 	return err;
1028 }
1029 
1030 /*
1031  * Calling into a user-controlled filesystem gives the filesystem
1032  * daemon ptrace-like capabilities over the current process.  This
1033  * means, that the filesystem daemon is able to record the exact
1034  * filesystem operations performed, and can also control the behavior
1035  * of the requester process in otherwise impossible ways.  For example
1036  * it can delay the operation for arbitrary length of time allowing
1037  * DoS against the requester.
1038  *
1039  * For this reason only those processes can call into the filesystem,
1040  * for which the owner of the mount has ptrace privilege.  This
1041  * excludes processes started by other users, suid or sgid processes.
1042  */
1043 int fuse_allow_current_process(struct fuse_conn *fc)
1044 {
1045 	const struct cred *cred;
1046 
1047 	if (fc->allow_other)
1048 		return current_in_userns(fc->user_ns);
1049 
1050 	cred = current_cred();
1051 	if (uid_eq(cred->euid, fc->user_id) &&
1052 	    uid_eq(cred->suid, fc->user_id) &&
1053 	    uid_eq(cred->uid,  fc->user_id) &&
1054 	    gid_eq(cred->egid, fc->group_id) &&
1055 	    gid_eq(cred->sgid, fc->group_id) &&
1056 	    gid_eq(cred->gid,  fc->group_id))
1057 		return 1;
1058 
1059 	return 0;
1060 }
1061 
1062 static int fuse_access(struct inode *inode, int mask)
1063 {
1064 	struct fuse_conn *fc = get_fuse_conn(inode);
1065 	FUSE_ARGS(args);
1066 	struct fuse_access_in inarg;
1067 	int err;
1068 
1069 	BUG_ON(mask & MAY_NOT_BLOCK);
1070 
1071 	if (fc->no_access)
1072 		return 0;
1073 
1074 	memset(&inarg, 0, sizeof(inarg));
1075 	inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1076 	args.in.h.opcode = FUSE_ACCESS;
1077 	args.in.h.nodeid = get_node_id(inode);
1078 	args.in.numargs = 1;
1079 	args.in.args[0].size = sizeof(inarg);
1080 	args.in.args[0].value = &inarg;
1081 	err = fuse_simple_request(fc, &args);
1082 	if (err == -ENOSYS) {
1083 		fc->no_access = 1;
1084 		err = 0;
1085 	}
1086 	return err;
1087 }
1088 
1089 static int fuse_perm_getattr(struct inode *inode, int mask)
1090 {
1091 	if (mask & MAY_NOT_BLOCK)
1092 		return -ECHILD;
1093 
1094 	forget_all_cached_acls(inode);
1095 	return fuse_do_getattr(inode, NULL, NULL);
1096 }
1097 
1098 /*
1099  * Check permission.  The two basic access models of FUSE are:
1100  *
1101  * 1) Local access checking ('default_permissions' mount option) based
1102  * on file mode.  This is the plain old disk filesystem permission
1103  * modell.
1104  *
1105  * 2) "Remote" access checking, where server is responsible for
1106  * checking permission in each inode operation.  An exception to this
1107  * is if ->permission() was invoked from sys_access() in which case an
1108  * access request is sent.  Execute permission is still checked
1109  * locally based on file mode.
1110  */
1111 static int fuse_permission(struct inode *inode, int mask)
1112 {
1113 	struct fuse_conn *fc = get_fuse_conn(inode);
1114 	bool refreshed = false;
1115 	int err = 0;
1116 
1117 	if (!fuse_allow_current_process(fc))
1118 		return -EACCES;
1119 
1120 	/*
1121 	 * If attributes are needed, refresh them before proceeding
1122 	 */
1123 	if (fc->default_permissions ||
1124 	    ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1125 		struct fuse_inode *fi = get_fuse_inode(inode);
1126 
1127 		if (time_before64(fi->i_time, get_jiffies_64())) {
1128 			refreshed = true;
1129 
1130 			err = fuse_perm_getattr(inode, mask);
1131 			if (err)
1132 				return err;
1133 		}
1134 	}
1135 
1136 	if (fc->default_permissions) {
1137 		err = generic_permission(inode, mask);
1138 
1139 		/* If permission is denied, try to refresh file
1140 		   attributes.  This is also needed, because the root
1141 		   node will at first have no permissions */
1142 		if (err == -EACCES && !refreshed) {
1143 			err = fuse_perm_getattr(inode, mask);
1144 			if (!err)
1145 				err = generic_permission(inode, mask);
1146 		}
1147 
1148 		/* Note: the opposite of the above test does not
1149 		   exist.  So if permissions are revoked this won't be
1150 		   noticed immediately, only after the attribute
1151 		   timeout has expired */
1152 	} else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1153 		err = fuse_access(inode, mask);
1154 	} else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
1155 		if (!(inode->i_mode & S_IXUGO)) {
1156 			if (refreshed)
1157 				return -EACCES;
1158 
1159 			err = fuse_perm_getattr(inode, mask);
1160 			if (!err && !(inode->i_mode & S_IXUGO))
1161 				return -EACCES;
1162 		}
1163 	}
1164 	return err;
1165 }
1166 
1167 static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
1168 			 struct dir_context *ctx)
1169 {
1170 	while (nbytes >= FUSE_NAME_OFFSET) {
1171 		struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
1172 		size_t reclen = FUSE_DIRENT_SIZE(dirent);
1173 		if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1174 			return -EIO;
1175 		if (reclen > nbytes)
1176 			break;
1177 		if (memchr(dirent->name, '/', dirent->namelen) != NULL)
1178 			return -EIO;
1179 
1180 		if (!dir_emit(ctx, dirent->name, dirent->namelen,
1181 			       dirent->ino, dirent->type))
1182 			break;
1183 
1184 		buf += reclen;
1185 		nbytes -= reclen;
1186 		ctx->pos = dirent->off;
1187 	}
1188 
1189 	return 0;
1190 }
1191 
1192 static int fuse_direntplus_link(struct file *file,
1193 				struct fuse_direntplus *direntplus,
1194 				u64 attr_version)
1195 {
1196 	struct fuse_entry_out *o = &direntplus->entry_out;
1197 	struct fuse_dirent *dirent = &direntplus->dirent;
1198 	struct dentry *parent = file->f_path.dentry;
1199 	struct qstr name = QSTR_INIT(dirent->name, dirent->namelen);
1200 	struct dentry *dentry;
1201 	struct dentry *alias;
1202 	struct inode *dir = d_inode(parent);
1203 	struct fuse_conn *fc;
1204 	struct inode *inode;
1205 	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1206 
1207 	if (!o->nodeid) {
1208 		/*
1209 		 * Unlike in the case of fuse_lookup, zero nodeid does not mean
1210 		 * ENOENT. Instead, it only means the userspace filesystem did
1211 		 * not want to return attributes/handle for this entry.
1212 		 *
1213 		 * So do nothing.
1214 		 */
1215 		return 0;
1216 	}
1217 
1218 	if (name.name[0] == '.') {
1219 		/*
1220 		 * We could potentially refresh the attributes of the directory
1221 		 * and its parent?
1222 		 */
1223 		if (name.len == 1)
1224 			return 0;
1225 		if (name.name[1] == '.' && name.len == 2)
1226 			return 0;
1227 	}
1228 
1229 	if (invalid_nodeid(o->nodeid))
1230 		return -EIO;
1231 	if (!fuse_valid_type(o->attr.mode))
1232 		return -EIO;
1233 
1234 	fc = get_fuse_conn(dir);
1235 
1236 	name.hash = full_name_hash(parent, name.name, name.len);
1237 	dentry = d_lookup(parent, &name);
1238 	if (!dentry) {
1239 retry:
1240 		dentry = d_alloc_parallel(parent, &name, &wq);
1241 		if (IS_ERR(dentry))
1242 			return PTR_ERR(dentry);
1243 	}
1244 	if (!d_in_lookup(dentry)) {
1245 		struct fuse_inode *fi;
1246 		inode = d_inode(dentry);
1247 		if (!inode ||
1248 		    get_node_id(inode) != o->nodeid ||
1249 		    ((o->attr.mode ^ inode->i_mode) & S_IFMT)) {
1250 			d_invalidate(dentry);
1251 			dput(dentry);
1252 			goto retry;
1253 		}
1254 		if (is_bad_inode(inode)) {
1255 			dput(dentry);
1256 			return -EIO;
1257 		}
1258 
1259 		fi = get_fuse_inode(inode);
1260 		spin_lock(&fc->lock);
1261 		fi->nlookup++;
1262 		spin_unlock(&fc->lock);
1263 
1264 		forget_all_cached_acls(inode);
1265 		fuse_change_attributes(inode, &o->attr,
1266 				       entry_attr_timeout(o),
1267 				       attr_version);
1268 		/*
1269 		 * The other branch comes via fuse_iget()
1270 		 * which bumps nlookup inside
1271 		 */
1272 	} else {
1273 		inode = fuse_iget(dir->i_sb, o->nodeid, o->generation,
1274 				  &o->attr, entry_attr_timeout(o),
1275 				  attr_version);
1276 		if (!inode)
1277 			inode = ERR_PTR(-ENOMEM);
1278 
1279 		alias = d_splice_alias(inode, dentry);
1280 		d_lookup_done(dentry);
1281 		if (alias) {
1282 			dput(dentry);
1283 			dentry = alias;
1284 		}
1285 		if (IS_ERR(dentry))
1286 			return PTR_ERR(dentry);
1287 	}
1288 	if (fc->readdirplus_auto)
1289 		set_bit(FUSE_I_INIT_RDPLUS, &get_fuse_inode(inode)->state);
1290 	fuse_change_entry_timeout(dentry, o);
1291 
1292 	dput(dentry);
1293 	return 0;
1294 }
1295 
1296 static int parse_dirplusfile(char *buf, size_t nbytes, struct file *file,
1297 			     struct dir_context *ctx, u64 attr_version)
1298 {
1299 	struct fuse_direntplus *direntplus;
1300 	struct fuse_dirent *dirent;
1301 	size_t reclen;
1302 	int over = 0;
1303 	int ret;
1304 
1305 	while (nbytes >= FUSE_NAME_OFFSET_DIRENTPLUS) {
1306 		direntplus = (struct fuse_direntplus *) buf;
1307 		dirent = &direntplus->dirent;
1308 		reclen = FUSE_DIRENTPLUS_SIZE(direntplus);
1309 
1310 		if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1311 			return -EIO;
1312 		if (reclen > nbytes)
1313 			break;
1314 		if (memchr(dirent->name, '/', dirent->namelen) != NULL)
1315 			return -EIO;
1316 
1317 		if (!over) {
1318 			/* We fill entries into dstbuf only as much as
1319 			   it can hold. But we still continue iterating
1320 			   over remaining entries to link them. If not,
1321 			   we need to send a FORGET for each of those
1322 			   which we did not link.
1323 			*/
1324 			over = !dir_emit(ctx, dirent->name, dirent->namelen,
1325 				       dirent->ino, dirent->type);
1326 			if (!over)
1327 				ctx->pos = dirent->off;
1328 		}
1329 
1330 		buf += reclen;
1331 		nbytes -= reclen;
1332 
1333 		ret = fuse_direntplus_link(file, direntplus, attr_version);
1334 		if (ret)
1335 			fuse_force_forget(file, direntplus->entry_out.nodeid);
1336 	}
1337 
1338 	return 0;
1339 }
1340 
1341 static int fuse_readdir(struct file *file, struct dir_context *ctx)
1342 {
1343 	int plus, err;
1344 	size_t nbytes;
1345 	struct page *page;
1346 	struct inode *inode = file_inode(file);
1347 	struct fuse_conn *fc = get_fuse_conn(inode);
1348 	struct fuse_req *req;
1349 	u64 attr_version = 0;
1350 
1351 	if (is_bad_inode(inode))
1352 		return -EIO;
1353 
1354 	req = fuse_get_req(fc, 1);
1355 	if (IS_ERR(req))
1356 		return PTR_ERR(req);
1357 
1358 	page = alloc_page(GFP_KERNEL);
1359 	if (!page) {
1360 		fuse_put_request(fc, req);
1361 		return -ENOMEM;
1362 	}
1363 
1364 	plus = fuse_use_readdirplus(inode, ctx);
1365 	req->out.argpages = 1;
1366 	req->num_pages = 1;
1367 	req->pages[0] = page;
1368 	req->page_descs[0].length = PAGE_SIZE;
1369 	if (plus) {
1370 		attr_version = fuse_get_attr_version(fc);
1371 		fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1372 			       FUSE_READDIRPLUS);
1373 	} else {
1374 		fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1375 			       FUSE_READDIR);
1376 	}
1377 	fuse_lock_inode(inode);
1378 	fuse_request_send(fc, req);
1379 	fuse_unlock_inode(inode);
1380 	nbytes = req->out.args[0].size;
1381 	err = req->out.h.error;
1382 	fuse_put_request(fc, req);
1383 	if (!err) {
1384 		if (plus) {
1385 			err = parse_dirplusfile(page_address(page), nbytes,
1386 						file, ctx,
1387 						attr_version);
1388 		} else {
1389 			err = parse_dirfile(page_address(page), nbytes, file,
1390 					    ctx);
1391 		}
1392 	}
1393 
1394 	__free_page(page);
1395 	fuse_invalidate_atime(inode);
1396 	return err;
1397 }
1398 
1399 static const char *fuse_get_link(struct dentry *dentry,
1400 				 struct inode *inode,
1401 				 struct delayed_call *done)
1402 {
1403 	struct fuse_conn *fc = get_fuse_conn(inode);
1404 	FUSE_ARGS(args);
1405 	char *link;
1406 	ssize_t ret;
1407 
1408 	if (!dentry)
1409 		return ERR_PTR(-ECHILD);
1410 
1411 	link = kmalloc(PAGE_SIZE, GFP_KERNEL);
1412 	if (!link)
1413 		return ERR_PTR(-ENOMEM);
1414 
1415 	args.in.h.opcode = FUSE_READLINK;
1416 	args.in.h.nodeid = get_node_id(inode);
1417 	args.out.argvar = 1;
1418 	args.out.numargs = 1;
1419 	args.out.args[0].size = PAGE_SIZE - 1;
1420 	args.out.args[0].value = link;
1421 	ret = fuse_simple_request(fc, &args);
1422 	if (ret < 0) {
1423 		kfree(link);
1424 		link = ERR_PTR(ret);
1425 	} else {
1426 		link[ret] = '\0';
1427 		set_delayed_call(done, kfree_link, link);
1428 	}
1429 	fuse_invalidate_atime(inode);
1430 	return link;
1431 }
1432 
1433 static int fuse_dir_open(struct inode *inode, struct file *file)
1434 {
1435 	return fuse_open_common(inode, file, true);
1436 }
1437 
1438 static int fuse_dir_release(struct inode *inode, struct file *file)
1439 {
1440 	fuse_release_common(file, FUSE_RELEASEDIR);
1441 
1442 	return 0;
1443 }
1444 
1445 static int fuse_dir_fsync(struct file *file, loff_t start, loff_t end,
1446 			  int datasync)
1447 {
1448 	return fuse_fsync_common(file, start, end, datasync, 1);
1449 }
1450 
1451 static long fuse_dir_ioctl(struct file *file, unsigned int cmd,
1452 			    unsigned long arg)
1453 {
1454 	struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1455 
1456 	/* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1457 	if (fc->minor < 18)
1458 		return -ENOTTY;
1459 
1460 	return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_DIR);
1461 }
1462 
1463 static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
1464 				   unsigned long arg)
1465 {
1466 	struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1467 
1468 	if (fc->minor < 18)
1469 		return -ENOTTY;
1470 
1471 	return fuse_ioctl_common(file, cmd, arg,
1472 				 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
1473 }
1474 
1475 static bool update_mtime(unsigned ivalid, bool trust_local_mtime)
1476 {
1477 	/* Always update if mtime is explicitly set  */
1478 	if (ivalid & ATTR_MTIME_SET)
1479 		return true;
1480 
1481 	/* Or if kernel i_mtime is the official one */
1482 	if (trust_local_mtime)
1483 		return true;
1484 
1485 	/* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1486 	if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1487 		return false;
1488 
1489 	/* In all other cases update */
1490 	return true;
1491 }
1492 
1493 static void iattr_to_fattr(struct fuse_conn *fc, struct iattr *iattr,
1494 			   struct fuse_setattr_in *arg, bool trust_local_cmtime)
1495 {
1496 	unsigned ivalid = iattr->ia_valid;
1497 
1498 	if (ivalid & ATTR_MODE)
1499 		arg->valid |= FATTR_MODE,   arg->mode = iattr->ia_mode;
1500 	if (ivalid & ATTR_UID)
1501 		arg->valid |= FATTR_UID,    arg->uid = from_kuid(fc->user_ns, iattr->ia_uid);
1502 	if (ivalid & ATTR_GID)
1503 		arg->valid |= FATTR_GID,    arg->gid = from_kgid(fc->user_ns, iattr->ia_gid);
1504 	if (ivalid & ATTR_SIZE)
1505 		arg->valid |= FATTR_SIZE,   arg->size = iattr->ia_size;
1506 	if (ivalid & ATTR_ATIME) {
1507 		arg->valid |= FATTR_ATIME;
1508 		arg->atime = iattr->ia_atime.tv_sec;
1509 		arg->atimensec = iattr->ia_atime.tv_nsec;
1510 		if (!(ivalid & ATTR_ATIME_SET))
1511 			arg->valid |= FATTR_ATIME_NOW;
1512 	}
1513 	if ((ivalid & ATTR_MTIME) && update_mtime(ivalid, trust_local_cmtime)) {
1514 		arg->valid |= FATTR_MTIME;
1515 		arg->mtime = iattr->ia_mtime.tv_sec;
1516 		arg->mtimensec = iattr->ia_mtime.tv_nsec;
1517 		if (!(ivalid & ATTR_MTIME_SET) && !trust_local_cmtime)
1518 			arg->valid |= FATTR_MTIME_NOW;
1519 	}
1520 	if ((ivalid & ATTR_CTIME) && trust_local_cmtime) {
1521 		arg->valid |= FATTR_CTIME;
1522 		arg->ctime = iattr->ia_ctime.tv_sec;
1523 		arg->ctimensec = iattr->ia_ctime.tv_nsec;
1524 	}
1525 }
1526 
1527 /*
1528  * Prevent concurrent writepages on inode
1529  *
1530  * This is done by adding a negative bias to the inode write counter
1531  * and waiting for all pending writes to finish.
1532  */
1533 void fuse_set_nowrite(struct inode *inode)
1534 {
1535 	struct fuse_conn *fc = get_fuse_conn(inode);
1536 	struct fuse_inode *fi = get_fuse_inode(inode);
1537 
1538 	BUG_ON(!inode_is_locked(inode));
1539 
1540 	spin_lock(&fc->lock);
1541 	BUG_ON(fi->writectr < 0);
1542 	fi->writectr += FUSE_NOWRITE;
1543 	spin_unlock(&fc->lock);
1544 	wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1545 }
1546 
1547 /*
1548  * Allow writepages on inode
1549  *
1550  * Remove the bias from the writecounter and send any queued
1551  * writepages.
1552  */
1553 static void __fuse_release_nowrite(struct inode *inode)
1554 {
1555 	struct fuse_inode *fi = get_fuse_inode(inode);
1556 
1557 	BUG_ON(fi->writectr != FUSE_NOWRITE);
1558 	fi->writectr = 0;
1559 	fuse_flush_writepages(inode);
1560 }
1561 
1562 void fuse_release_nowrite(struct inode *inode)
1563 {
1564 	struct fuse_conn *fc = get_fuse_conn(inode);
1565 
1566 	spin_lock(&fc->lock);
1567 	__fuse_release_nowrite(inode);
1568 	spin_unlock(&fc->lock);
1569 }
1570 
1571 static void fuse_setattr_fill(struct fuse_conn *fc, struct fuse_args *args,
1572 			      struct inode *inode,
1573 			      struct fuse_setattr_in *inarg_p,
1574 			      struct fuse_attr_out *outarg_p)
1575 {
1576 	args->in.h.opcode = FUSE_SETATTR;
1577 	args->in.h.nodeid = get_node_id(inode);
1578 	args->in.numargs = 1;
1579 	args->in.args[0].size = sizeof(*inarg_p);
1580 	args->in.args[0].value = inarg_p;
1581 	args->out.numargs = 1;
1582 	args->out.args[0].size = sizeof(*outarg_p);
1583 	args->out.args[0].value = outarg_p;
1584 }
1585 
1586 /*
1587  * Flush inode->i_mtime to the server
1588  */
1589 int fuse_flush_times(struct inode *inode, struct fuse_file *ff)
1590 {
1591 	struct fuse_conn *fc = get_fuse_conn(inode);
1592 	FUSE_ARGS(args);
1593 	struct fuse_setattr_in inarg;
1594 	struct fuse_attr_out outarg;
1595 
1596 	memset(&inarg, 0, sizeof(inarg));
1597 	memset(&outarg, 0, sizeof(outarg));
1598 
1599 	inarg.valid = FATTR_MTIME;
1600 	inarg.mtime = inode->i_mtime.tv_sec;
1601 	inarg.mtimensec = inode->i_mtime.tv_nsec;
1602 	if (fc->minor >= 23) {
1603 		inarg.valid |= FATTR_CTIME;
1604 		inarg.ctime = inode->i_ctime.tv_sec;
1605 		inarg.ctimensec = inode->i_ctime.tv_nsec;
1606 	}
1607 	if (ff) {
1608 		inarg.valid |= FATTR_FH;
1609 		inarg.fh = ff->fh;
1610 	}
1611 	fuse_setattr_fill(fc, &args, inode, &inarg, &outarg);
1612 
1613 	return fuse_simple_request(fc, &args);
1614 }
1615 
1616 /*
1617  * Set attributes, and at the same time refresh them.
1618  *
1619  * Truncation is slightly complicated, because the 'truncate' request
1620  * may fail, in which case we don't want to touch the mapping.
1621  * vmtruncate() doesn't allow for this case, so do the rlimit checking
1622  * and the actual truncation by hand.
1623  */
1624 int fuse_do_setattr(struct dentry *dentry, struct iattr *attr,
1625 		    struct file *file)
1626 {
1627 	struct inode *inode = d_inode(dentry);
1628 	struct fuse_conn *fc = get_fuse_conn(inode);
1629 	struct fuse_inode *fi = get_fuse_inode(inode);
1630 	FUSE_ARGS(args);
1631 	struct fuse_setattr_in inarg;
1632 	struct fuse_attr_out outarg;
1633 	bool is_truncate = false;
1634 	bool is_wb = fc->writeback_cache;
1635 	loff_t oldsize;
1636 	int err;
1637 	bool trust_local_cmtime = is_wb && S_ISREG(inode->i_mode);
1638 
1639 	if (!fc->default_permissions)
1640 		attr->ia_valid |= ATTR_FORCE;
1641 
1642 	err = setattr_prepare(dentry, attr);
1643 	if (err)
1644 		return err;
1645 
1646 	if (attr->ia_valid & ATTR_OPEN) {
1647 		/* This is coming from open(..., ... | O_TRUNC); */
1648 		WARN_ON(!(attr->ia_valid & ATTR_SIZE));
1649 		WARN_ON(attr->ia_size != 0);
1650 		if (fc->atomic_o_trunc) {
1651 			/*
1652 			 * No need to send request to userspace, since actual
1653 			 * truncation has already been done by OPEN.  But still
1654 			 * need to truncate page cache.
1655 			 */
1656 			i_size_write(inode, 0);
1657 			truncate_pagecache(inode, 0);
1658 			return 0;
1659 		}
1660 		file = NULL;
1661 	}
1662 
1663 	if (attr->ia_valid & ATTR_SIZE)
1664 		is_truncate = true;
1665 
1666 	if (is_truncate) {
1667 		fuse_set_nowrite(inode);
1668 		set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1669 		if (trust_local_cmtime && attr->ia_size != inode->i_size)
1670 			attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1671 	}
1672 
1673 	memset(&inarg, 0, sizeof(inarg));
1674 	memset(&outarg, 0, sizeof(outarg));
1675 	iattr_to_fattr(fc, attr, &inarg, trust_local_cmtime);
1676 	if (file) {
1677 		struct fuse_file *ff = file->private_data;
1678 		inarg.valid |= FATTR_FH;
1679 		inarg.fh = ff->fh;
1680 	}
1681 	if (attr->ia_valid & ATTR_SIZE) {
1682 		/* For mandatory locking in truncate */
1683 		inarg.valid |= FATTR_LOCKOWNER;
1684 		inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1685 	}
1686 	fuse_setattr_fill(fc, &args, inode, &inarg, &outarg);
1687 	err = fuse_simple_request(fc, &args);
1688 	if (err) {
1689 		if (err == -EINTR)
1690 			fuse_invalidate_attr(inode);
1691 		goto error;
1692 	}
1693 
1694 	if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1695 		make_bad_inode(inode);
1696 		err = -EIO;
1697 		goto error;
1698 	}
1699 
1700 	spin_lock(&fc->lock);
1701 	/* the kernel maintains i_mtime locally */
1702 	if (trust_local_cmtime) {
1703 		if (attr->ia_valid & ATTR_MTIME)
1704 			inode->i_mtime = attr->ia_mtime;
1705 		if (attr->ia_valid & ATTR_CTIME)
1706 			inode->i_ctime = attr->ia_ctime;
1707 		/* FIXME: clear I_DIRTY_SYNC? */
1708 	}
1709 
1710 	fuse_change_attributes_common(inode, &outarg.attr,
1711 				      attr_timeout(&outarg));
1712 	oldsize = inode->i_size;
1713 	/* see the comment in fuse_change_attributes() */
1714 	if (!is_wb || is_truncate || !S_ISREG(inode->i_mode))
1715 		i_size_write(inode, outarg.attr.size);
1716 
1717 	if (is_truncate) {
1718 		/* NOTE: this may release/reacquire fc->lock */
1719 		__fuse_release_nowrite(inode);
1720 	}
1721 	spin_unlock(&fc->lock);
1722 
1723 	/*
1724 	 * Only call invalidate_inode_pages2() after removing
1725 	 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1726 	 */
1727 	if ((is_truncate || !is_wb) &&
1728 	    S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1729 		truncate_pagecache(inode, outarg.attr.size);
1730 		invalidate_inode_pages2(inode->i_mapping);
1731 	}
1732 
1733 	clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1734 	return 0;
1735 
1736 error:
1737 	if (is_truncate)
1738 		fuse_release_nowrite(inode);
1739 
1740 	clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1741 	return err;
1742 }
1743 
1744 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1745 {
1746 	struct inode *inode = d_inode(entry);
1747 	struct fuse_conn *fc = get_fuse_conn(inode);
1748 	struct file *file = (attr->ia_valid & ATTR_FILE) ? attr->ia_file : NULL;
1749 	int ret;
1750 
1751 	if (!fuse_allow_current_process(get_fuse_conn(inode)))
1752 		return -EACCES;
1753 
1754 	if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) {
1755 		attr->ia_valid &= ~(ATTR_KILL_SUID | ATTR_KILL_SGID |
1756 				    ATTR_MODE);
1757 
1758 		/*
1759 		 * The only sane way to reliably kill suid/sgid is to do it in
1760 		 * the userspace filesystem
1761 		 *
1762 		 * This should be done on write(), truncate() and chown().
1763 		 */
1764 		if (!fc->handle_killpriv) {
1765 			/*
1766 			 * ia_mode calculation may have used stale i_mode.
1767 			 * Refresh and recalculate.
1768 			 */
1769 			ret = fuse_do_getattr(inode, NULL, file);
1770 			if (ret)
1771 				return ret;
1772 
1773 			attr->ia_mode = inode->i_mode;
1774 			if (inode->i_mode & S_ISUID) {
1775 				attr->ia_valid |= ATTR_MODE;
1776 				attr->ia_mode &= ~S_ISUID;
1777 			}
1778 			if ((inode->i_mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1779 				attr->ia_valid |= ATTR_MODE;
1780 				attr->ia_mode &= ~S_ISGID;
1781 			}
1782 		}
1783 	}
1784 	if (!attr->ia_valid)
1785 		return 0;
1786 
1787 	ret = fuse_do_setattr(entry, attr, file);
1788 	if (!ret) {
1789 		/*
1790 		 * If filesystem supports acls it may have updated acl xattrs in
1791 		 * the filesystem, so forget cached acls for the inode.
1792 		 */
1793 		if (fc->posix_acl)
1794 			forget_all_cached_acls(inode);
1795 
1796 		/* Directory mode changed, may need to revalidate access */
1797 		if (d_is_dir(entry) && (attr->ia_valid & ATTR_MODE))
1798 			fuse_invalidate_entry_cache(entry);
1799 	}
1800 	return ret;
1801 }
1802 
1803 static int fuse_getattr(const struct path *path, struct kstat *stat,
1804 			u32 request_mask, unsigned int flags)
1805 {
1806 	struct inode *inode = d_inode(path->dentry);
1807 	struct fuse_conn *fc = get_fuse_conn(inode);
1808 
1809 	if (!fuse_allow_current_process(fc))
1810 		return -EACCES;
1811 
1812 	return fuse_update_get_attr(inode, NULL, stat, flags);
1813 }
1814 
1815 static const struct inode_operations fuse_dir_inode_operations = {
1816 	.lookup		= fuse_lookup,
1817 	.mkdir		= fuse_mkdir,
1818 	.symlink	= fuse_symlink,
1819 	.unlink		= fuse_unlink,
1820 	.rmdir		= fuse_rmdir,
1821 	.rename		= fuse_rename2,
1822 	.link		= fuse_link,
1823 	.setattr	= fuse_setattr,
1824 	.create		= fuse_create,
1825 	.atomic_open	= fuse_atomic_open,
1826 	.mknod		= fuse_mknod,
1827 	.permission	= fuse_permission,
1828 	.getattr	= fuse_getattr,
1829 	.listxattr	= fuse_listxattr,
1830 	.get_acl	= fuse_get_acl,
1831 	.set_acl	= fuse_set_acl,
1832 };
1833 
1834 static const struct file_operations fuse_dir_operations = {
1835 	.llseek		= generic_file_llseek,
1836 	.read		= generic_read_dir,
1837 	.iterate_shared	= fuse_readdir,
1838 	.open		= fuse_dir_open,
1839 	.release	= fuse_dir_release,
1840 	.fsync		= fuse_dir_fsync,
1841 	.unlocked_ioctl	= fuse_dir_ioctl,
1842 	.compat_ioctl	= fuse_dir_compat_ioctl,
1843 };
1844 
1845 static const struct inode_operations fuse_common_inode_operations = {
1846 	.setattr	= fuse_setattr,
1847 	.permission	= fuse_permission,
1848 	.getattr	= fuse_getattr,
1849 	.listxattr	= fuse_listxattr,
1850 	.get_acl	= fuse_get_acl,
1851 	.set_acl	= fuse_set_acl,
1852 };
1853 
1854 static const struct inode_operations fuse_symlink_inode_operations = {
1855 	.setattr	= fuse_setattr,
1856 	.get_link	= fuse_get_link,
1857 	.getattr	= fuse_getattr,
1858 	.listxattr	= fuse_listxattr,
1859 };
1860 
1861 void fuse_init_common(struct inode *inode)
1862 {
1863 	inode->i_op = &fuse_common_inode_operations;
1864 }
1865 
1866 void fuse_init_dir(struct inode *inode)
1867 {
1868 	inode->i_op = &fuse_dir_inode_operations;
1869 	inode->i_fop = &fuse_dir_operations;
1870 }
1871 
1872 void fuse_init_symlink(struct inode *inode)
1873 {
1874 	inode->i_op = &fuse_symlink_inode_operations;
1875 }
1876