xref: /openbmc/linux/fs/fuse/dir.c (revision 7b6d864b)
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 
17 static bool fuse_use_readdirplus(struct inode *dir, struct dir_context *ctx)
18 {
19 	struct fuse_conn *fc = get_fuse_conn(dir);
20 	struct fuse_inode *fi = get_fuse_inode(dir);
21 
22 	if (!fc->do_readdirplus)
23 		return false;
24 	if (!fc->readdirplus_auto)
25 		return true;
26 	if (test_and_clear_bit(FUSE_I_ADVISE_RDPLUS, &fi->state))
27 		return true;
28 	if (ctx->pos == 0)
29 		return true;
30 	return false;
31 }
32 
33 static void fuse_advise_use_readdirplus(struct inode *dir)
34 {
35 	struct fuse_inode *fi = get_fuse_inode(dir);
36 
37 	set_bit(FUSE_I_ADVISE_RDPLUS, &fi->state);
38 }
39 
40 #if BITS_PER_LONG >= 64
41 static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
42 {
43 	entry->d_time = time;
44 }
45 
46 static inline u64 fuse_dentry_time(struct dentry *entry)
47 {
48 	return entry->d_time;
49 }
50 #else
51 /*
52  * On 32 bit archs store the high 32 bits of time in d_fsdata
53  */
54 static void fuse_dentry_settime(struct dentry *entry, u64 time)
55 {
56 	entry->d_time = time;
57 	entry->d_fsdata = (void *) (unsigned long) (time >> 32);
58 }
59 
60 static u64 fuse_dentry_time(struct dentry *entry)
61 {
62 	return (u64) entry->d_time +
63 		((u64) (unsigned long) entry->d_fsdata << 32);
64 }
65 #endif
66 
67 /*
68  * FUSE caches dentries and attributes with separate timeout.  The
69  * time in jiffies until the dentry/attributes are valid is stored in
70  * dentry->d_time and fuse_inode->i_time respectively.
71  */
72 
73 /*
74  * Calculate the time in jiffies until a dentry/attributes are valid
75  */
76 static u64 time_to_jiffies(unsigned long sec, unsigned long nsec)
77 {
78 	if (sec || nsec) {
79 		struct timespec ts = {sec, nsec};
80 		return get_jiffies_64() + timespec_to_jiffies(&ts);
81 	} else
82 		return 0;
83 }
84 
85 /*
86  * Set dentry and possibly attribute timeouts from the lookup/mk*
87  * replies
88  */
89 static void fuse_change_entry_timeout(struct dentry *entry,
90 				      struct fuse_entry_out *o)
91 {
92 	fuse_dentry_settime(entry,
93 		time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
94 }
95 
96 static u64 attr_timeout(struct fuse_attr_out *o)
97 {
98 	return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
99 }
100 
101 static u64 entry_attr_timeout(struct fuse_entry_out *o)
102 {
103 	return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
104 }
105 
106 /*
107  * Mark the attributes as stale, so that at the next call to
108  * ->getattr() they will be fetched from userspace
109  */
110 void fuse_invalidate_attr(struct inode *inode)
111 {
112 	get_fuse_inode(inode)->i_time = 0;
113 }
114 
115 /*
116  * Just mark the entry as stale, so that a next attempt to look it up
117  * will result in a new lookup call to userspace
118  *
119  * This is called when a dentry is about to become negative and the
120  * timeout is unknown (unlink, rmdir, rename and in some cases
121  * lookup)
122  */
123 void fuse_invalidate_entry_cache(struct dentry *entry)
124 {
125 	fuse_dentry_settime(entry, 0);
126 }
127 
128 /*
129  * Same as fuse_invalidate_entry_cache(), but also try to remove the
130  * dentry from the hash
131  */
132 static void fuse_invalidate_entry(struct dentry *entry)
133 {
134 	d_invalidate(entry);
135 	fuse_invalidate_entry_cache(entry);
136 }
137 
138 static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_req *req,
139 			     u64 nodeid, struct qstr *name,
140 			     struct fuse_entry_out *outarg)
141 {
142 	memset(outarg, 0, sizeof(struct fuse_entry_out));
143 	req->in.h.opcode = FUSE_LOOKUP;
144 	req->in.h.nodeid = nodeid;
145 	req->in.numargs = 1;
146 	req->in.args[0].size = name->len + 1;
147 	req->in.args[0].value = name->name;
148 	req->out.numargs = 1;
149 	if (fc->minor < 9)
150 		req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
151 	else
152 		req->out.args[0].size = sizeof(struct fuse_entry_out);
153 	req->out.args[0].value = outarg;
154 }
155 
156 u64 fuse_get_attr_version(struct fuse_conn *fc)
157 {
158 	u64 curr_version;
159 
160 	/*
161 	 * The spin lock isn't actually needed on 64bit archs, but we
162 	 * don't yet care too much about such optimizations.
163 	 */
164 	spin_lock(&fc->lock);
165 	curr_version = fc->attr_version;
166 	spin_unlock(&fc->lock);
167 
168 	return curr_version;
169 }
170 
171 /*
172  * Check whether the dentry is still valid
173  *
174  * If the entry validity timeout has expired and the dentry is
175  * positive, try to redo the lookup.  If the lookup results in a
176  * different inode, then let the VFS invalidate the dentry and redo
177  * the lookup once more.  If the lookup results in the same inode,
178  * then refresh the attributes, timeouts and mark the dentry valid.
179  */
180 static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
181 {
182 	struct inode *inode;
183 	struct dentry *parent;
184 	struct fuse_conn *fc;
185 
186 	inode = ACCESS_ONCE(entry->d_inode);
187 	if (inode && is_bad_inode(inode))
188 		return 0;
189 	else if (fuse_dentry_time(entry) < get_jiffies_64()) {
190 		int err;
191 		struct fuse_entry_out outarg;
192 		struct fuse_req *req;
193 		struct fuse_forget_link *forget;
194 		u64 attr_version;
195 
196 		/* For negative dentries, always do a fresh lookup */
197 		if (!inode)
198 			return 0;
199 
200 		if (flags & LOOKUP_RCU)
201 			return -ECHILD;
202 
203 		fc = get_fuse_conn(inode);
204 		req = fuse_get_req_nopages(fc);
205 		if (IS_ERR(req))
206 			return 0;
207 
208 		forget = fuse_alloc_forget();
209 		if (!forget) {
210 			fuse_put_request(fc, req);
211 			return 0;
212 		}
213 
214 		attr_version = fuse_get_attr_version(fc);
215 
216 		parent = dget_parent(entry);
217 		fuse_lookup_init(fc, req, get_node_id(parent->d_inode),
218 				 &entry->d_name, &outarg);
219 		fuse_request_send(fc, req);
220 		dput(parent);
221 		err = req->out.h.error;
222 		fuse_put_request(fc, req);
223 		/* Zero nodeid is same as -ENOENT */
224 		if (!err && !outarg.nodeid)
225 			err = -ENOENT;
226 		if (!err) {
227 			struct fuse_inode *fi = get_fuse_inode(inode);
228 			if (outarg.nodeid != get_node_id(inode)) {
229 				fuse_queue_forget(fc, forget, outarg.nodeid, 1);
230 				return 0;
231 			}
232 			spin_lock(&fc->lock);
233 			fi->nlookup++;
234 			spin_unlock(&fc->lock);
235 		}
236 		kfree(forget);
237 		if (err || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
238 			return 0;
239 
240 		fuse_change_attributes(inode, &outarg.attr,
241 				       entry_attr_timeout(&outarg),
242 				       attr_version);
243 		fuse_change_entry_timeout(entry, &outarg);
244 	} else if (inode) {
245 		fc = get_fuse_conn(inode);
246 		if (fc->readdirplus_auto) {
247 			parent = dget_parent(entry);
248 			fuse_advise_use_readdirplus(parent->d_inode);
249 			dput(parent);
250 		}
251 	}
252 	return 1;
253 }
254 
255 static int invalid_nodeid(u64 nodeid)
256 {
257 	return !nodeid || nodeid == FUSE_ROOT_ID;
258 }
259 
260 const struct dentry_operations fuse_dentry_operations = {
261 	.d_revalidate	= fuse_dentry_revalidate,
262 };
263 
264 int fuse_valid_type(int m)
265 {
266 	return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
267 		S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
268 }
269 
270 /*
271  * Add a directory inode to a dentry, ensuring that no other dentry
272  * refers to this inode.  Called with fc->inst_mutex.
273  */
274 static struct dentry *fuse_d_add_directory(struct dentry *entry,
275 					   struct inode *inode)
276 {
277 	struct dentry *alias = d_find_alias(inode);
278 	if (alias && !(alias->d_flags & DCACHE_DISCONNECTED)) {
279 		/* This tries to shrink the subtree below alias */
280 		fuse_invalidate_entry(alias);
281 		dput(alias);
282 		if (!hlist_empty(&inode->i_dentry))
283 			return ERR_PTR(-EBUSY);
284 	} else {
285 		dput(alias);
286 	}
287 	return d_splice_alias(inode, entry);
288 }
289 
290 int fuse_lookup_name(struct super_block *sb, u64 nodeid, struct qstr *name,
291 		     struct fuse_entry_out *outarg, struct inode **inode)
292 {
293 	struct fuse_conn *fc = get_fuse_conn_super(sb);
294 	struct fuse_req *req;
295 	struct fuse_forget_link *forget;
296 	u64 attr_version;
297 	int err;
298 
299 	*inode = NULL;
300 	err = -ENAMETOOLONG;
301 	if (name->len > FUSE_NAME_MAX)
302 		goto out;
303 
304 	req = fuse_get_req_nopages(fc);
305 	err = PTR_ERR(req);
306 	if (IS_ERR(req))
307 		goto out;
308 
309 	forget = fuse_alloc_forget();
310 	err = -ENOMEM;
311 	if (!forget) {
312 		fuse_put_request(fc, req);
313 		goto out;
314 	}
315 
316 	attr_version = fuse_get_attr_version(fc);
317 
318 	fuse_lookup_init(fc, req, nodeid, name, outarg);
319 	fuse_request_send(fc, req);
320 	err = req->out.h.error;
321 	fuse_put_request(fc, req);
322 	/* Zero nodeid is same as -ENOENT, but with valid timeout */
323 	if (err || !outarg->nodeid)
324 		goto out_put_forget;
325 
326 	err = -EIO;
327 	if (!outarg->nodeid)
328 		goto out_put_forget;
329 	if (!fuse_valid_type(outarg->attr.mode))
330 		goto out_put_forget;
331 
332 	*inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
333 			   &outarg->attr, entry_attr_timeout(outarg),
334 			   attr_version);
335 	err = -ENOMEM;
336 	if (!*inode) {
337 		fuse_queue_forget(fc, forget, outarg->nodeid, 1);
338 		goto out;
339 	}
340 	err = 0;
341 
342  out_put_forget:
343 	kfree(forget);
344  out:
345 	return err;
346 }
347 
348 static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
349 				  unsigned int flags)
350 {
351 	int err;
352 	struct fuse_entry_out outarg;
353 	struct inode *inode;
354 	struct dentry *newent;
355 	struct fuse_conn *fc = get_fuse_conn(dir);
356 	bool outarg_valid = true;
357 
358 	err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
359 			       &outarg, &inode);
360 	if (err == -ENOENT) {
361 		outarg_valid = false;
362 		err = 0;
363 	}
364 	if (err)
365 		goto out_err;
366 
367 	err = -EIO;
368 	if (inode && get_node_id(inode) == FUSE_ROOT_ID)
369 		goto out_iput;
370 
371 	if (inode && S_ISDIR(inode->i_mode)) {
372 		mutex_lock(&fc->inst_mutex);
373 		newent = fuse_d_add_directory(entry, inode);
374 		mutex_unlock(&fc->inst_mutex);
375 		err = PTR_ERR(newent);
376 		if (IS_ERR(newent))
377 			goto out_iput;
378 	} else {
379 		newent = d_splice_alias(inode, entry);
380 	}
381 
382 	entry = newent ? newent : entry;
383 	if (outarg_valid)
384 		fuse_change_entry_timeout(entry, &outarg);
385 	else
386 		fuse_invalidate_entry_cache(entry);
387 
388 	fuse_advise_use_readdirplus(dir);
389 	return newent;
390 
391  out_iput:
392 	iput(inode);
393  out_err:
394 	return ERR_PTR(err);
395 }
396 
397 /*
398  * Atomic create+open operation
399  *
400  * If the filesystem doesn't support this, then fall back to separate
401  * 'mknod' + 'open' requests.
402  */
403 static int fuse_create_open(struct inode *dir, struct dentry *entry,
404 			    struct file *file, unsigned flags,
405 			    umode_t mode, int *opened)
406 {
407 	int err;
408 	struct inode *inode;
409 	struct fuse_conn *fc = get_fuse_conn(dir);
410 	struct fuse_req *req;
411 	struct fuse_forget_link *forget;
412 	struct fuse_create_in inarg;
413 	struct fuse_open_out outopen;
414 	struct fuse_entry_out outentry;
415 	struct fuse_file *ff;
416 
417 	/* Userspace expects S_IFREG in create mode */
418 	BUG_ON((mode & S_IFMT) != S_IFREG);
419 
420 	forget = fuse_alloc_forget();
421 	err = -ENOMEM;
422 	if (!forget)
423 		goto out_err;
424 
425 	req = fuse_get_req_nopages(fc);
426 	err = PTR_ERR(req);
427 	if (IS_ERR(req))
428 		goto out_put_forget_req;
429 
430 	err = -ENOMEM;
431 	ff = fuse_file_alloc(fc);
432 	if (!ff)
433 		goto out_put_request;
434 
435 	if (!fc->dont_mask)
436 		mode &= ~current_umask();
437 
438 	flags &= ~O_NOCTTY;
439 	memset(&inarg, 0, sizeof(inarg));
440 	memset(&outentry, 0, sizeof(outentry));
441 	inarg.flags = flags;
442 	inarg.mode = mode;
443 	inarg.umask = current_umask();
444 	req->in.h.opcode = FUSE_CREATE;
445 	req->in.h.nodeid = get_node_id(dir);
446 	req->in.numargs = 2;
447 	req->in.args[0].size = fc->minor < 12 ? sizeof(struct fuse_open_in) :
448 						sizeof(inarg);
449 	req->in.args[0].value = &inarg;
450 	req->in.args[1].size = entry->d_name.len + 1;
451 	req->in.args[1].value = entry->d_name.name;
452 	req->out.numargs = 2;
453 	if (fc->minor < 9)
454 		req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
455 	else
456 		req->out.args[0].size = sizeof(outentry);
457 	req->out.args[0].value = &outentry;
458 	req->out.args[1].size = sizeof(outopen);
459 	req->out.args[1].value = &outopen;
460 	fuse_request_send(fc, req);
461 	err = req->out.h.error;
462 	if (err)
463 		goto out_free_ff;
464 
465 	err = -EIO;
466 	if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
467 		goto out_free_ff;
468 
469 	fuse_put_request(fc, req);
470 	ff->fh = outopen.fh;
471 	ff->nodeid = outentry.nodeid;
472 	ff->open_flags = outopen.open_flags;
473 	inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
474 			  &outentry.attr, entry_attr_timeout(&outentry), 0);
475 	if (!inode) {
476 		flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
477 		fuse_sync_release(ff, flags);
478 		fuse_queue_forget(fc, forget, outentry.nodeid, 1);
479 		err = -ENOMEM;
480 		goto out_err;
481 	}
482 	kfree(forget);
483 	d_instantiate(entry, inode);
484 	fuse_change_entry_timeout(entry, &outentry);
485 	fuse_invalidate_attr(dir);
486 	err = finish_open(file, entry, generic_file_open, opened);
487 	if (err) {
488 		fuse_sync_release(ff, flags);
489 	} else {
490 		file->private_data = fuse_file_get(ff);
491 		fuse_finish_open(inode, file);
492 	}
493 	return err;
494 
495 out_free_ff:
496 	fuse_file_free(ff);
497 out_put_request:
498 	fuse_put_request(fc, req);
499 out_put_forget_req:
500 	kfree(forget);
501 out_err:
502 	return err;
503 }
504 
505 static int fuse_mknod(struct inode *, struct dentry *, umode_t, dev_t);
506 static int fuse_atomic_open(struct inode *dir, struct dentry *entry,
507 			    struct file *file, unsigned flags,
508 			    umode_t mode, int *opened)
509 {
510 	int err;
511 	struct fuse_conn *fc = get_fuse_conn(dir);
512 	struct dentry *res = NULL;
513 
514 	if (d_unhashed(entry)) {
515 		res = fuse_lookup(dir, entry, 0);
516 		if (IS_ERR(res))
517 			return PTR_ERR(res);
518 
519 		if (res)
520 			entry = res;
521 	}
522 
523 	if (!(flags & O_CREAT) || entry->d_inode)
524 		goto no_open;
525 
526 	/* Only creates */
527 	*opened |= FILE_CREATED;
528 
529 	if (fc->no_create)
530 		goto mknod;
531 
532 	err = fuse_create_open(dir, entry, file, flags, mode, opened);
533 	if (err == -ENOSYS) {
534 		fc->no_create = 1;
535 		goto mknod;
536 	}
537 out_dput:
538 	dput(res);
539 	return err;
540 
541 mknod:
542 	err = fuse_mknod(dir, entry, mode, 0);
543 	if (err)
544 		goto out_dput;
545 no_open:
546 	return finish_no_open(file, res);
547 }
548 
549 /*
550  * Code shared between mknod, mkdir, symlink and link
551  */
552 static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
553 			    struct inode *dir, struct dentry *entry,
554 			    umode_t mode)
555 {
556 	struct fuse_entry_out outarg;
557 	struct inode *inode;
558 	int err;
559 	struct fuse_forget_link *forget;
560 
561 	forget = fuse_alloc_forget();
562 	if (!forget) {
563 		fuse_put_request(fc, req);
564 		return -ENOMEM;
565 	}
566 
567 	memset(&outarg, 0, sizeof(outarg));
568 	req->in.h.nodeid = get_node_id(dir);
569 	req->out.numargs = 1;
570 	if (fc->minor < 9)
571 		req->out.args[0].size = FUSE_COMPAT_ENTRY_OUT_SIZE;
572 	else
573 		req->out.args[0].size = sizeof(outarg);
574 	req->out.args[0].value = &outarg;
575 	fuse_request_send(fc, req);
576 	err = req->out.h.error;
577 	fuse_put_request(fc, req);
578 	if (err)
579 		goto out_put_forget_req;
580 
581 	err = -EIO;
582 	if (invalid_nodeid(outarg.nodeid))
583 		goto out_put_forget_req;
584 
585 	if ((outarg.attr.mode ^ mode) & S_IFMT)
586 		goto out_put_forget_req;
587 
588 	inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
589 			  &outarg.attr, entry_attr_timeout(&outarg), 0);
590 	if (!inode) {
591 		fuse_queue_forget(fc, forget, outarg.nodeid, 1);
592 		return -ENOMEM;
593 	}
594 	kfree(forget);
595 
596 	if (S_ISDIR(inode->i_mode)) {
597 		struct dentry *alias;
598 		mutex_lock(&fc->inst_mutex);
599 		alias = d_find_alias(inode);
600 		if (alias) {
601 			/* New directory must have moved since mkdir */
602 			mutex_unlock(&fc->inst_mutex);
603 			dput(alias);
604 			iput(inode);
605 			return -EBUSY;
606 		}
607 		d_instantiate(entry, inode);
608 		mutex_unlock(&fc->inst_mutex);
609 	} else
610 		d_instantiate(entry, inode);
611 
612 	fuse_change_entry_timeout(entry, &outarg);
613 	fuse_invalidate_attr(dir);
614 	return 0;
615 
616  out_put_forget_req:
617 	kfree(forget);
618 	return err;
619 }
620 
621 static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
622 		      dev_t rdev)
623 {
624 	struct fuse_mknod_in inarg;
625 	struct fuse_conn *fc = get_fuse_conn(dir);
626 	struct fuse_req *req = fuse_get_req_nopages(fc);
627 	if (IS_ERR(req))
628 		return PTR_ERR(req);
629 
630 	if (!fc->dont_mask)
631 		mode &= ~current_umask();
632 
633 	memset(&inarg, 0, sizeof(inarg));
634 	inarg.mode = mode;
635 	inarg.rdev = new_encode_dev(rdev);
636 	inarg.umask = current_umask();
637 	req->in.h.opcode = FUSE_MKNOD;
638 	req->in.numargs = 2;
639 	req->in.args[0].size = fc->minor < 12 ? FUSE_COMPAT_MKNOD_IN_SIZE :
640 						sizeof(inarg);
641 	req->in.args[0].value = &inarg;
642 	req->in.args[1].size = entry->d_name.len + 1;
643 	req->in.args[1].value = entry->d_name.name;
644 	return create_new_entry(fc, req, dir, entry, mode);
645 }
646 
647 static int fuse_create(struct inode *dir, struct dentry *entry, umode_t mode,
648 		       bool excl)
649 {
650 	return fuse_mknod(dir, entry, mode, 0);
651 }
652 
653 static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
654 {
655 	struct fuse_mkdir_in inarg;
656 	struct fuse_conn *fc = get_fuse_conn(dir);
657 	struct fuse_req *req = fuse_get_req_nopages(fc);
658 	if (IS_ERR(req))
659 		return PTR_ERR(req);
660 
661 	if (!fc->dont_mask)
662 		mode &= ~current_umask();
663 
664 	memset(&inarg, 0, sizeof(inarg));
665 	inarg.mode = mode;
666 	inarg.umask = current_umask();
667 	req->in.h.opcode = FUSE_MKDIR;
668 	req->in.numargs = 2;
669 	req->in.args[0].size = sizeof(inarg);
670 	req->in.args[0].value = &inarg;
671 	req->in.args[1].size = entry->d_name.len + 1;
672 	req->in.args[1].value = entry->d_name.name;
673 	return create_new_entry(fc, req, dir, entry, S_IFDIR);
674 }
675 
676 static int fuse_symlink(struct inode *dir, struct dentry *entry,
677 			const char *link)
678 {
679 	struct fuse_conn *fc = get_fuse_conn(dir);
680 	unsigned len = strlen(link) + 1;
681 	struct fuse_req *req = fuse_get_req_nopages(fc);
682 	if (IS_ERR(req))
683 		return PTR_ERR(req);
684 
685 	req->in.h.opcode = FUSE_SYMLINK;
686 	req->in.numargs = 2;
687 	req->in.args[0].size = entry->d_name.len + 1;
688 	req->in.args[0].value = entry->d_name.name;
689 	req->in.args[1].size = len;
690 	req->in.args[1].value = link;
691 	return create_new_entry(fc, req, dir, entry, S_IFLNK);
692 }
693 
694 static int fuse_unlink(struct inode *dir, struct dentry *entry)
695 {
696 	int err;
697 	struct fuse_conn *fc = get_fuse_conn(dir);
698 	struct fuse_req *req = fuse_get_req_nopages(fc);
699 	if (IS_ERR(req))
700 		return PTR_ERR(req);
701 
702 	req->in.h.opcode = FUSE_UNLINK;
703 	req->in.h.nodeid = get_node_id(dir);
704 	req->in.numargs = 1;
705 	req->in.args[0].size = entry->d_name.len + 1;
706 	req->in.args[0].value = entry->d_name.name;
707 	fuse_request_send(fc, req);
708 	err = req->out.h.error;
709 	fuse_put_request(fc, req);
710 	if (!err) {
711 		struct inode *inode = entry->d_inode;
712 		struct fuse_inode *fi = get_fuse_inode(inode);
713 
714 		spin_lock(&fc->lock);
715 		fi->attr_version = ++fc->attr_version;
716 		/*
717 		 * If i_nlink == 0 then unlink doesn't make sense, yet this can
718 		 * happen if userspace filesystem is careless.  It would be
719 		 * difficult to enforce correct nlink usage so just ignore this
720 		 * condition here
721 		 */
722 		if (inode->i_nlink > 0)
723 			drop_nlink(inode);
724 		spin_unlock(&fc->lock);
725 		fuse_invalidate_attr(inode);
726 		fuse_invalidate_attr(dir);
727 		fuse_invalidate_entry_cache(entry);
728 	} else if (err == -EINTR)
729 		fuse_invalidate_entry(entry);
730 	return err;
731 }
732 
733 static int fuse_rmdir(struct inode *dir, struct dentry *entry)
734 {
735 	int err;
736 	struct fuse_conn *fc = get_fuse_conn(dir);
737 	struct fuse_req *req = fuse_get_req_nopages(fc);
738 	if (IS_ERR(req))
739 		return PTR_ERR(req);
740 
741 	req->in.h.opcode = FUSE_RMDIR;
742 	req->in.h.nodeid = get_node_id(dir);
743 	req->in.numargs = 1;
744 	req->in.args[0].size = entry->d_name.len + 1;
745 	req->in.args[0].value = entry->d_name.name;
746 	fuse_request_send(fc, req);
747 	err = req->out.h.error;
748 	fuse_put_request(fc, req);
749 	if (!err) {
750 		clear_nlink(entry->d_inode);
751 		fuse_invalidate_attr(dir);
752 		fuse_invalidate_entry_cache(entry);
753 	} else if (err == -EINTR)
754 		fuse_invalidate_entry(entry);
755 	return err;
756 }
757 
758 static int fuse_rename(struct inode *olddir, struct dentry *oldent,
759 		       struct inode *newdir, struct dentry *newent)
760 {
761 	int err;
762 	struct fuse_rename_in inarg;
763 	struct fuse_conn *fc = get_fuse_conn(olddir);
764 	struct fuse_req *req = fuse_get_req_nopages(fc);
765 
766 	if (IS_ERR(req))
767 		return PTR_ERR(req);
768 
769 	memset(&inarg, 0, sizeof(inarg));
770 	inarg.newdir = get_node_id(newdir);
771 	req->in.h.opcode = FUSE_RENAME;
772 	req->in.h.nodeid = get_node_id(olddir);
773 	req->in.numargs = 3;
774 	req->in.args[0].size = sizeof(inarg);
775 	req->in.args[0].value = &inarg;
776 	req->in.args[1].size = oldent->d_name.len + 1;
777 	req->in.args[1].value = oldent->d_name.name;
778 	req->in.args[2].size = newent->d_name.len + 1;
779 	req->in.args[2].value = newent->d_name.name;
780 	fuse_request_send(fc, req);
781 	err = req->out.h.error;
782 	fuse_put_request(fc, req);
783 	if (!err) {
784 		/* ctime changes */
785 		fuse_invalidate_attr(oldent->d_inode);
786 
787 		fuse_invalidate_attr(olddir);
788 		if (olddir != newdir)
789 			fuse_invalidate_attr(newdir);
790 
791 		/* newent will end up negative */
792 		if (newent->d_inode) {
793 			fuse_invalidate_attr(newent->d_inode);
794 			fuse_invalidate_entry_cache(newent);
795 		}
796 	} else if (err == -EINTR) {
797 		/* If request was interrupted, DEITY only knows if the
798 		   rename actually took place.  If the invalidation
799 		   fails (e.g. some process has CWD under the renamed
800 		   directory), then there can be inconsistency between
801 		   the dcache and the real filesystem.  Tough luck. */
802 		fuse_invalidate_entry(oldent);
803 		if (newent->d_inode)
804 			fuse_invalidate_entry(newent);
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 = entry->d_inode;
816 	struct fuse_conn *fc = get_fuse_conn(inode);
817 	struct fuse_req *req = fuse_get_req_nopages(fc);
818 	if (IS_ERR(req))
819 		return PTR_ERR(req);
820 
821 	memset(&inarg, 0, sizeof(inarg));
822 	inarg.oldnodeid = get_node_id(inode);
823 	req->in.h.opcode = FUSE_LINK;
824 	req->in.numargs = 2;
825 	req->in.args[0].size = sizeof(inarg);
826 	req->in.args[0].value = &inarg;
827 	req->in.args[1].size = newent->d_name.len + 1;
828 	req->in.args[1].value = newent->d_name.name;
829 	err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
830 	/* Contrary to "normal" filesystems it can happen that link
831 	   makes two "logical" inodes point to the same "physical"
832 	   inode.  We invalidate the attributes of the old one, so it
833 	   will reflect changes in the backing inode (link count,
834 	   etc.)
835 	*/
836 	if (!err) {
837 		struct fuse_inode *fi = get_fuse_inode(inode);
838 
839 		spin_lock(&fc->lock);
840 		fi->attr_version = ++fc->attr_version;
841 		inc_nlink(inode);
842 		spin_unlock(&fc->lock);
843 		fuse_invalidate_attr(inode);
844 	} else if (err == -EINTR) {
845 		fuse_invalidate_attr(inode);
846 	}
847 	return err;
848 }
849 
850 static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
851 			  struct kstat *stat)
852 {
853 	unsigned int blkbits;
854 
855 	stat->dev = inode->i_sb->s_dev;
856 	stat->ino = attr->ino;
857 	stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
858 	stat->nlink = attr->nlink;
859 	stat->uid = make_kuid(&init_user_ns, attr->uid);
860 	stat->gid = make_kgid(&init_user_ns, attr->gid);
861 	stat->rdev = inode->i_rdev;
862 	stat->atime.tv_sec = attr->atime;
863 	stat->atime.tv_nsec = attr->atimensec;
864 	stat->mtime.tv_sec = attr->mtime;
865 	stat->mtime.tv_nsec = attr->mtimensec;
866 	stat->ctime.tv_sec = attr->ctime;
867 	stat->ctime.tv_nsec = attr->ctimensec;
868 	stat->size = attr->size;
869 	stat->blocks = attr->blocks;
870 
871 	if (attr->blksize != 0)
872 		blkbits = ilog2(attr->blksize);
873 	else
874 		blkbits = inode->i_sb->s_blocksize_bits;
875 
876 	stat->blksize = 1 << blkbits;
877 }
878 
879 static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
880 			   struct file *file)
881 {
882 	int err;
883 	struct fuse_getattr_in inarg;
884 	struct fuse_attr_out outarg;
885 	struct fuse_conn *fc = get_fuse_conn(inode);
886 	struct fuse_req *req;
887 	u64 attr_version;
888 
889 	req = fuse_get_req_nopages(fc);
890 	if (IS_ERR(req))
891 		return PTR_ERR(req);
892 
893 	attr_version = fuse_get_attr_version(fc);
894 
895 	memset(&inarg, 0, sizeof(inarg));
896 	memset(&outarg, 0, sizeof(outarg));
897 	/* Directories have separate file-handle space */
898 	if (file && S_ISREG(inode->i_mode)) {
899 		struct fuse_file *ff = file->private_data;
900 
901 		inarg.getattr_flags |= FUSE_GETATTR_FH;
902 		inarg.fh = ff->fh;
903 	}
904 	req->in.h.opcode = FUSE_GETATTR;
905 	req->in.h.nodeid = get_node_id(inode);
906 	req->in.numargs = 1;
907 	req->in.args[0].size = sizeof(inarg);
908 	req->in.args[0].value = &inarg;
909 	req->out.numargs = 1;
910 	if (fc->minor < 9)
911 		req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
912 	else
913 		req->out.args[0].size = sizeof(outarg);
914 	req->out.args[0].value = &outarg;
915 	fuse_request_send(fc, req);
916 	err = req->out.h.error;
917 	fuse_put_request(fc, req);
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 int fuse_update_attributes(struct inode *inode, struct kstat *stat,
934 			   struct file *file, bool *refreshed)
935 {
936 	struct fuse_inode *fi = get_fuse_inode(inode);
937 	int err;
938 	bool r;
939 
940 	if (fi->i_time < get_jiffies_64()) {
941 		r = true;
942 		err = fuse_do_getattr(inode, stat, file);
943 	} else {
944 		r = false;
945 		err = 0;
946 		if (stat) {
947 			generic_fillattr(inode, stat);
948 			stat->mode = fi->orig_i_mode;
949 			stat->ino = fi->orig_ino;
950 		}
951 	}
952 
953 	if (refreshed != NULL)
954 		*refreshed = r;
955 
956 	return err;
957 }
958 
959 int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
960 			     u64 child_nodeid, struct qstr *name)
961 {
962 	int err = -ENOTDIR;
963 	struct inode *parent;
964 	struct dentry *dir;
965 	struct dentry *entry;
966 
967 	parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
968 	if (!parent)
969 		return -ENOENT;
970 
971 	mutex_lock(&parent->i_mutex);
972 	if (!S_ISDIR(parent->i_mode))
973 		goto unlock;
974 
975 	err = -ENOENT;
976 	dir = d_find_alias(parent);
977 	if (!dir)
978 		goto unlock;
979 
980 	entry = d_lookup(dir, name);
981 	dput(dir);
982 	if (!entry)
983 		goto unlock;
984 
985 	fuse_invalidate_attr(parent);
986 	fuse_invalidate_entry(entry);
987 
988 	if (child_nodeid != 0 && entry->d_inode) {
989 		mutex_lock(&entry->d_inode->i_mutex);
990 		if (get_node_id(entry->d_inode) != child_nodeid) {
991 			err = -ENOENT;
992 			goto badentry;
993 		}
994 		if (d_mountpoint(entry)) {
995 			err = -EBUSY;
996 			goto badentry;
997 		}
998 		if (S_ISDIR(entry->d_inode->i_mode)) {
999 			shrink_dcache_parent(entry);
1000 			if (!simple_empty(entry)) {
1001 				err = -ENOTEMPTY;
1002 				goto badentry;
1003 			}
1004 			entry->d_inode->i_flags |= S_DEAD;
1005 		}
1006 		dont_mount(entry);
1007 		clear_nlink(entry->d_inode);
1008 		err = 0;
1009  badentry:
1010 		mutex_unlock(&entry->d_inode->i_mutex);
1011 		if (!err)
1012 			d_delete(entry);
1013 	} else {
1014 		err = 0;
1015 	}
1016 	dput(entry);
1017 
1018  unlock:
1019 	mutex_unlock(&parent->i_mutex);
1020 	iput(parent);
1021 	return err;
1022 }
1023 
1024 /*
1025  * Calling into a user-controlled filesystem gives the filesystem
1026  * daemon ptrace-like capabilities over the current process.  This
1027  * means, that the filesystem daemon is able to record the exact
1028  * filesystem operations performed, and can also control the behavior
1029  * of the requester process in otherwise impossible ways.  For example
1030  * it can delay the operation for arbitrary length of time allowing
1031  * DoS against the requester.
1032  *
1033  * For this reason only those processes can call into the filesystem,
1034  * for which the owner of the mount has ptrace privilege.  This
1035  * excludes processes started by other users, suid or sgid processes.
1036  */
1037 int fuse_allow_current_process(struct fuse_conn *fc)
1038 {
1039 	const struct cred *cred;
1040 
1041 	if (fc->flags & FUSE_ALLOW_OTHER)
1042 		return 1;
1043 
1044 	cred = current_cred();
1045 	if (uid_eq(cred->euid, fc->user_id) &&
1046 	    uid_eq(cred->suid, fc->user_id) &&
1047 	    uid_eq(cred->uid,  fc->user_id) &&
1048 	    gid_eq(cred->egid, fc->group_id) &&
1049 	    gid_eq(cred->sgid, fc->group_id) &&
1050 	    gid_eq(cred->gid,  fc->group_id))
1051 		return 1;
1052 
1053 	return 0;
1054 }
1055 
1056 static int fuse_access(struct inode *inode, int mask)
1057 {
1058 	struct fuse_conn *fc = get_fuse_conn(inode);
1059 	struct fuse_req *req;
1060 	struct fuse_access_in inarg;
1061 	int err;
1062 
1063 	if (fc->no_access)
1064 		return 0;
1065 
1066 	req = fuse_get_req_nopages(fc);
1067 	if (IS_ERR(req))
1068 		return PTR_ERR(req);
1069 
1070 	memset(&inarg, 0, sizeof(inarg));
1071 	inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1072 	req->in.h.opcode = FUSE_ACCESS;
1073 	req->in.h.nodeid = get_node_id(inode);
1074 	req->in.numargs = 1;
1075 	req->in.args[0].size = sizeof(inarg);
1076 	req->in.args[0].value = &inarg;
1077 	fuse_request_send(fc, req);
1078 	err = req->out.h.error;
1079 	fuse_put_request(fc, req);
1080 	if (err == -ENOSYS) {
1081 		fc->no_access = 1;
1082 		err = 0;
1083 	}
1084 	return err;
1085 }
1086 
1087 static int fuse_perm_getattr(struct inode *inode, int mask)
1088 {
1089 	if (mask & MAY_NOT_BLOCK)
1090 		return -ECHILD;
1091 
1092 	return fuse_do_getattr(inode, NULL, NULL);
1093 }
1094 
1095 /*
1096  * Check permission.  The two basic access models of FUSE are:
1097  *
1098  * 1) Local access checking ('default_permissions' mount option) based
1099  * on file mode.  This is the plain old disk filesystem permission
1100  * modell.
1101  *
1102  * 2) "Remote" access checking, where server is responsible for
1103  * checking permission in each inode operation.  An exception to this
1104  * is if ->permission() was invoked from sys_access() in which case an
1105  * access request is sent.  Execute permission is still checked
1106  * locally based on file mode.
1107  */
1108 static int fuse_permission(struct inode *inode, int mask)
1109 {
1110 	struct fuse_conn *fc = get_fuse_conn(inode);
1111 	bool refreshed = false;
1112 	int err = 0;
1113 
1114 	if (!fuse_allow_current_process(fc))
1115 		return -EACCES;
1116 
1117 	/*
1118 	 * If attributes are needed, refresh them before proceeding
1119 	 */
1120 	if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
1121 	    ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1122 		struct fuse_inode *fi = get_fuse_inode(inode);
1123 
1124 		if (fi->i_time < get_jiffies_64()) {
1125 			refreshed = true;
1126 
1127 			err = fuse_perm_getattr(inode, mask);
1128 			if (err)
1129 				return err;
1130 		}
1131 	}
1132 
1133 	if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1134 		err = generic_permission(inode, mask);
1135 
1136 		/* If permission is denied, try to refresh file
1137 		   attributes.  This is also needed, because the root
1138 		   node will at first have no permissions */
1139 		if (err == -EACCES && !refreshed) {
1140 			err = fuse_perm_getattr(inode, mask);
1141 			if (!err)
1142 				err = generic_permission(inode, mask);
1143 		}
1144 
1145 		/* Note: the opposite of the above test does not
1146 		   exist.  So if permissions are revoked this won't be
1147 		   noticed immediately, only after the attribute
1148 		   timeout has expired */
1149 	} else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1150 		if (mask & MAY_NOT_BLOCK)
1151 			return -ECHILD;
1152 
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 
1178 		if (!dir_emit(ctx, dirent->name, dirent->namelen,
1179 			       dirent->ino, dirent->type))
1180 			break;
1181 
1182 		buf += reclen;
1183 		nbytes -= reclen;
1184 		ctx->pos = dirent->off;
1185 	}
1186 
1187 	return 0;
1188 }
1189 
1190 static int fuse_direntplus_link(struct file *file,
1191 				struct fuse_direntplus *direntplus,
1192 				u64 attr_version)
1193 {
1194 	int err;
1195 	struct fuse_entry_out *o = &direntplus->entry_out;
1196 	struct fuse_dirent *dirent = &direntplus->dirent;
1197 	struct dentry *parent = file->f_path.dentry;
1198 	struct qstr name = QSTR_INIT(dirent->name, dirent->namelen);
1199 	struct dentry *dentry;
1200 	struct dentry *alias;
1201 	struct inode *dir = parent->d_inode;
1202 	struct fuse_conn *fc;
1203 	struct inode *inode;
1204 
1205 	if (!o->nodeid) {
1206 		/*
1207 		 * Unlike in the case of fuse_lookup, zero nodeid does not mean
1208 		 * ENOENT. Instead, it only means the userspace filesystem did
1209 		 * not want to return attributes/handle for this entry.
1210 		 *
1211 		 * So do nothing.
1212 		 */
1213 		return 0;
1214 	}
1215 
1216 	if (name.name[0] == '.') {
1217 		/*
1218 		 * We could potentially refresh the attributes of the directory
1219 		 * and its parent?
1220 		 */
1221 		if (name.len == 1)
1222 			return 0;
1223 		if (name.name[1] == '.' && name.len == 2)
1224 			return 0;
1225 	}
1226 	fc = get_fuse_conn(dir);
1227 
1228 	name.hash = full_name_hash(name.name, name.len);
1229 	dentry = d_lookup(parent, &name);
1230 	if (dentry && dentry->d_inode) {
1231 		inode = dentry->d_inode;
1232 		if (get_node_id(inode) == o->nodeid) {
1233 			struct fuse_inode *fi;
1234 			fi = get_fuse_inode(inode);
1235 			spin_lock(&fc->lock);
1236 			fi->nlookup++;
1237 			spin_unlock(&fc->lock);
1238 
1239 			/*
1240 			 * The other branch to 'found' comes via fuse_iget()
1241 			 * which bumps nlookup inside
1242 			 */
1243 			goto found;
1244 		}
1245 		err = d_invalidate(dentry);
1246 		if (err)
1247 			goto out;
1248 		dput(dentry);
1249 		dentry = NULL;
1250 	}
1251 
1252 	dentry = d_alloc(parent, &name);
1253 	err = -ENOMEM;
1254 	if (!dentry)
1255 		goto out;
1256 
1257 	inode = fuse_iget(dir->i_sb, o->nodeid, o->generation,
1258 			  &o->attr, entry_attr_timeout(o), attr_version);
1259 	if (!inode)
1260 		goto out;
1261 
1262 	alias = d_materialise_unique(dentry, inode);
1263 	err = PTR_ERR(alias);
1264 	if (IS_ERR(alias))
1265 		goto out;
1266 	if (alias) {
1267 		dput(dentry);
1268 		dentry = alias;
1269 	}
1270 
1271 found:
1272 	fuse_change_attributes(inode, &o->attr, entry_attr_timeout(o),
1273 			       attr_version);
1274 
1275 	fuse_change_entry_timeout(dentry, o);
1276 
1277 	err = 0;
1278 out:
1279 	if (dentry)
1280 		dput(dentry);
1281 	return err;
1282 }
1283 
1284 static int parse_dirplusfile(char *buf, size_t nbytes, struct file *file,
1285 			     struct dir_context *ctx, u64 attr_version)
1286 {
1287 	struct fuse_direntplus *direntplus;
1288 	struct fuse_dirent *dirent;
1289 	size_t reclen;
1290 	int over = 0;
1291 	int ret;
1292 
1293 	while (nbytes >= FUSE_NAME_OFFSET_DIRENTPLUS) {
1294 		direntplus = (struct fuse_direntplus *) buf;
1295 		dirent = &direntplus->dirent;
1296 		reclen = FUSE_DIRENTPLUS_SIZE(direntplus);
1297 
1298 		if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
1299 			return -EIO;
1300 		if (reclen > nbytes)
1301 			break;
1302 
1303 		if (!over) {
1304 			/* We fill entries into dstbuf only as much as
1305 			   it can hold. But we still continue iterating
1306 			   over remaining entries to link them. If not,
1307 			   we need to send a FORGET for each of those
1308 			   which we did not link.
1309 			*/
1310 			over = !dir_emit(ctx, dirent->name, dirent->namelen,
1311 				       dirent->ino, dirent->type);
1312 			ctx->pos = dirent->off;
1313 		}
1314 
1315 		buf += reclen;
1316 		nbytes -= reclen;
1317 
1318 		ret = fuse_direntplus_link(file, direntplus, attr_version);
1319 		if (ret)
1320 			fuse_force_forget(file, direntplus->entry_out.nodeid);
1321 	}
1322 
1323 	return 0;
1324 }
1325 
1326 static int fuse_readdir(struct file *file, struct dir_context *ctx)
1327 {
1328 	int plus, err;
1329 	size_t nbytes;
1330 	struct page *page;
1331 	struct inode *inode = file_inode(file);
1332 	struct fuse_conn *fc = get_fuse_conn(inode);
1333 	struct fuse_req *req;
1334 	u64 attr_version = 0;
1335 
1336 	if (is_bad_inode(inode))
1337 		return -EIO;
1338 
1339 	req = fuse_get_req(fc, 1);
1340 	if (IS_ERR(req))
1341 		return PTR_ERR(req);
1342 
1343 	page = alloc_page(GFP_KERNEL);
1344 	if (!page) {
1345 		fuse_put_request(fc, req);
1346 		return -ENOMEM;
1347 	}
1348 
1349 	plus = fuse_use_readdirplus(inode, ctx);
1350 	req->out.argpages = 1;
1351 	req->num_pages = 1;
1352 	req->pages[0] = page;
1353 	req->page_descs[0].length = PAGE_SIZE;
1354 	if (plus) {
1355 		attr_version = fuse_get_attr_version(fc);
1356 		fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1357 			       FUSE_READDIRPLUS);
1358 	} else {
1359 		fuse_read_fill(req, file, ctx->pos, PAGE_SIZE,
1360 			       FUSE_READDIR);
1361 	}
1362 	fuse_request_send(fc, req);
1363 	nbytes = req->out.args[0].size;
1364 	err = req->out.h.error;
1365 	fuse_put_request(fc, req);
1366 	if (!err) {
1367 		if (plus) {
1368 			err = parse_dirplusfile(page_address(page), nbytes,
1369 						file, ctx,
1370 						attr_version);
1371 		} else {
1372 			err = parse_dirfile(page_address(page), nbytes, file,
1373 					    ctx);
1374 		}
1375 	}
1376 
1377 	__free_page(page);
1378 	fuse_invalidate_attr(inode); /* atime changed */
1379 	return err;
1380 }
1381 
1382 static char *read_link(struct dentry *dentry)
1383 {
1384 	struct inode *inode = dentry->d_inode;
1385 	struct fuse_conn *fc = get_fuse_conn(inode);
1386 	struct fuse_req *req = fuse_get_req_nopages(fc);
1387 	char *link;
1388 
1389 	if (IS_ERR(req))
1390 		return ERR_CAST(req);
1391 
1392 	link = (char *) __get_free_page(GFP_KERNEL);
1393 	if (!link) {
1394 		link = ERR_PTR(-ENOMEM);
1395 		goto out;
1396 	}
1397 	req->in.h.opcode = FUSE_READLINK;
1398 	req->in.h.nodeid = get_node_id(inode);
1399 	req->out.argvar = 1;
1400 	req->out.numargs = 1;
1401 	req->out.args[0].size = PAGE_SIZE - 1;
1402 	req->out.args[0].value = link;
1403 	fuse_request_send(fc, req);
1404 	if (req->out.h.error) {
1405 		free_page((unsigned long) link);
1406 		link = ERR_PTR(req->out.h.error);
1407 	} else
1408 		link[req->out.args[0].size] = '\0';
1409  out:
1410 	fuse_put_request(fc, req);
1411 	fuse_invalidate_attr(inode); /* atime changed */
1412 	return link;
1413 }
1414 
1415 static void free_link(char *link)
1416 {
1417 	if (!IS_ERR(link))
1418 		free_page((unsigned long) link);
1419 }
1420 
1421 static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
1422 {
1423 	nd_set_link(nd, read_link(dentry));
1424 	return NULL;
1425 }
1426 
1427 static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1428 {
1429 	free_link(nd_get_link(nd));
1430 }
1431 
1432 static int fuse_dir_open(struct inode *inode, struct file *file)
1433 {
1434 	return fuse_open_common(inode, file, true);
1435 }
1436 
1437 static int fuse_dir_release(struct inode *inode, struct file *file)
1438 {
1439 	fuse_release_common(file, FUSE_RELEASEDIR);
1440 
1441 	return 0;
1442 }
1443 
1444 static int fuse_dir_fsync(struct file *file, loff_t start, loff_t end,
1445 			  int datasync)
1446 {
1447 	return fuse_fsync_common(file, start, end, datasync, 1);
1448 }
1449 
1450 static long fuse_dir_ioctl(struct file *file, unsigned int cmd,
1451 			    unsigned long arg)
1452 {
1453 	struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1454 
1455 	/* FUSE_IOCTL_DIR only supported for API version >= 7.18 */
1456 	if (fc->minor < 18)
1457 		return -ENOTTY;
1458 
1459 	return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_DIR);
1460 }
1461 
1462 static long fuse_dir_compat_ioctl(struct file *file, unsigned int cmd,
1463 				   unsigned long arg)
1464 {
1465 	struct fuse_conn *fc = get_fuse_conn(file->f_mapping->host);
1466 
1467 	if (fc->minor < 18)
1468 		return -ENOTTY;
1469 
1470 	return fuse_ioctl_common(file, cmd, arg,
1471 				 FUSE_IOCTL_COMPAT | FUSE_IOCTL_DIR);
1472 }
1473 
1474 static bool update_mtime(unsigned ivalid)
1475 {
1476 	/* Always update if mtime is explicitly set  */
1477 	if (ivalid & ATTR_MTIME_SET)
1478 		return true;
1479 
1480 	/* If it's an open(O_TRUNC) or an ftruncate(), don't update */
1481 	if ((ivalid & ATTR_SIZE) && (ivalid & (ATTR_OPEN | ATTR_FILE)))
1482 		return false;
1483 
1484 	/* In all other cases update */
1485 	return true;
1486 }
1487 
1488 static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
1489 {
1490 	unsigned ivalid = iattr->ia_valid;
1491 
1492 	if (ivalid & ATTR_MODE)
1493 		arg->valid |= FATTR_MODE,   arg->mode = iattr->ia_mode;
1494 	if (ivalid & ATTR_UID)
1495 		arg->valid |= FATTR_UID,    arg->uid = from_kuid(&init_user_ns, iattr->ia_uid);
1496 	if (ivalid & ATTR_GID)
1497 		arg->valid |= FATTR_GID,    arg->gid = from_kgid(&init_user_ns, iattr->ia_gid);
1498 	if (ivalid & ATTR_SIZE)
1499 		arg->valid |= FATTR_SIZE,   arg->size = iattr->ia_size;
1500 	if (ivalid & ATTR_ATIME) {
1501 		arg->valid |= FATTR_ATIME;
1502 		arg->atime = iattr->ia_atime.tv_sec;
1503 		arg->atimensec = iattr->ia_atime.tv_nsec;
1504 		if (!(ivalid & ATTR_ATIME_SET))
1505 			arg->valid |= FATTR_ATIME_NOW;
1506 	}
1507 	if ((ivalid & ATTR_MTIME) && update_mtime(ivalid)) {
1508 		arg->valid |= FATTR_MTIME;
1509 		arg->mtime = iattr->ia_mtime.tv_sec;
1510 		arg->mtimensec = iattr->ia_mtime.tv_nsec;
1511 		if (!(ivalid & ATTR_MTIME_SET))
1512 			arg->valid |= FATTR_MTIME_NOW;
1513 	}
1514 }
1515 
1516 /*
1517  * Prevent concurrent writepages on inode
1518  *
1519  * This is done by adding a negative bias to the inode write counter
1520  * and waiting for all pending writes to finish.
1521  */
1522 void fuse_set_nowrite(struct inode *inode)
1523 {
1524 	struct fuse_conn *fc = get_fuse_conn(inode);
1525 	struct fuse_inode *fi = get_fuse_inode(inode);
1526 
1527 	BUG_ON(!mutex_is_locked(&inode->i_mutex));
1528 
1529 	spin_lock(&fc->lock);
1530 	BUG_ON(fi->writectr < 0);
1531 	fi->writectr += FUSE_NOWRITE;
1532 	spin_unlock(&fc->lock);
1533 	wait_event(fi->page_waitq, fi->writectr == FUSE_NOWRITE);
1534 }
1535 
1536 /*
1537  * Allow writepages on inode
1538  *
1539  * Remove the bias from the writecounter and send any queued
1540  * writepages.
1541  */
1542 static void __fuse_release_nowrite(struct inode *inode)
1543 {
1544 	struct fuse_inode *fi = get_fuse_inode(inode);
1545 
1546 	BUG_ON(fi->writectr != FUSE_NOWRITE);
1547 	fi->writectr = 0;
1548 	fuse_flush_writepages(inode);
1549 }
1550 
1551 void fuse_release_nowrite(struct inode *inode)
1552 {
1553 	struct fuse_conn *fc = get_fuse_conn(inode);
1554 
1555 	spin_lock(&fc->lock);
1556 	__fuse_release_nowrite(inode);
1557 	spin_unlock(&fc->lock);
1558 }
1559 
1560 /*
1561  * Set attributes, and at the same time refresh them.
1562  *
1563  * Truncation is slightly complicated, because the 'truncate' request
1564  * may fail, in which case we don't want to touch the mapping.
1565  * vmtruncate() doesn't allow for this case, so do the rlimit checking
1566  * and the actual truncation by hand.
1567  */
1568 int fuse_do_setattr(struct inode *inode, struct iattr *attr,
1569 		    struct file *file)
1570 {
1571 	struct fuse_conn *fc = get_fuse_conn(inode);
1572 	struct fuse_req *req;
1573 	struct fuse_setattr_in inarg;
1574 	struct fuse_attr_out outarg;
1575 	bool is_truncate = false;
1576 	loff_t oldsize;
1577 	int err;
1578 
1579 	if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
1580 		attr->ia_valid |= ATTR_FORCE;
1581 
1582 	err = inode_change_ok(inode, attr);
1583 	if (err)
1584 		return err;
1585 
1586 	if (attr->ia_valid & ATTR_OPEN) {
1587 		if (fc->atomic_o_trunc)
1588 			return 0;
1589 		file = NULL;
1590 	}
1591 
1592 	if (attr->ia_valid & ATTR_SIZE)
1593 		is_truncate = true;
1594 
1595 	req = fuse_get_req_nopages(fc);
1596 	if (IS_ERR(req))
1597 		return PTR_ERR(req);
1598 
1599 	if (is_truncate)
1600 		fuse_set_nowrite(inode);
1601 
1602 	memset(&inarg, 0, sizeof(inarg));
1603 	memset(&outarg, 0, sizeof(outarg));
1604 	iattr_to_fattr(attr, &inarg);
1605 	if (file) {
1606 		struct fuse_file *ff = file->private_data;
1607 		inarg.valid |= FATTR_FH;
1608 		inarg.fh = ff->fh;
1609 	}
1610 	if (attr->ia_valid & ATTR_SIZE) {
1611 		/* For mandatory locking in truncate */
1612 		inarg.valid |= FATTR_LOCKOWNER;
1613 		inarg.lock_owner = fuse_lock_owner_id(fc, current->files);
1614 	}
1615 	req->in.h.opcode = FUSE_SETATTR;
1616 	req->in.h.nodeid = get_node_id(inode);
1617 	req->in.numargs = 1;
1618 	req->in.args[0].size = sizeof(inarg);
1619 	req->in.args[0].value = &inarg;
1620 	req->out.numargs = 1;
1621 	if (fc->minor < 9)
1622 		req->out.args[0].size = FUSE_COMPAT_ATTR_OUT_SIZE;
1623 	else
1624 		req->out.args[0].size = sizeof(outarg);
1625 	req->out.args[0].value = &outarg;
1626 	fuse_request_send(fc, req);
1627 	err = req->out.h.error;
1628 	fuse_put_request(fc, req);
1629 	if (err) {
1630 		if (err == -EINTR)
1631 			fuse_invalidate_attr(inode);
1632 		goto error;
1633 	}
1634 
1635 	if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
1636 		make_bad_inode(inode);
1637 		err = -EIO;
1638 		goto error;
1639 	}
1640 
1641 	spin_lock(&fc->lock);
1642 	fuse_change_attributes_common(inode, &outarg.attr,
1643 				      attr_timeout(&outarg));
1644 	oldsize = inode->i_size;
1645 	i_size_write(inode, outarg.attr.size);
1646 
1647 	if (is_truncate) {
1648 		/* NOTE: this may release/reacquire fc->lock */
1649 		__fuse_release_nowrite(inode);
1650 	}
1651 	spin_unlock(&fc->lock);
1652 
1653 	/*
1654 	 * Only call invalidate_inode_pages2() after removing
1655 	 * FUSE_NOWRITE, otherwise fuse_launder_page() would deadlock.
1656 	 */
1657 	if (S_ISREG(inode->i_mode) && oldsize != outarg.attr.size) {
1658 		truncate_pagecache(inode, oldsize, outarg.attr.size);
1659 		invalidate_inode_pages2(inode->i_mapping);
1660 	}
1661 
1662 	return 0;
1663 
1664 error:
1665 	if (is_truncate)
1666 		fuse_release_nowrite(inode);
1667 
1668 	return err;
1669 }
1670 
1671 static int fuse_setattr(struct dentry *entry, struct iattr *attr)
1672 {
1673 	struct inode *inode = entry->d_inode;
1674 
1675 	if (!fuse_allow_current_process(get_fuse_conn(inode)))
1676 		return -EACCES;
1677 
1678 	if (attr->ia_valid & ATTR_FILE)
1679 		return fuse_do_setattr(inode, attr, attr->ia_file);
1680 	else
1681 		return fuse_do_setattr(inode, attr, NULL);
1682 }
1683 
1684 static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
1685 			struct kstat *stat)
1686 {
1687 	struct inode *inode = entry->d_inode;
1688 	struct fuse_conn *fc = get_fuse_conn(inode);
1689 
1690 	if (!fuse_allow_current_process(fc))
1691 		return -EACCES;
1692 
1693 	return fuse_update_attributes(inode, stat, NULL, NULL);
1694 }
1695 
1696 static int fuse_setxattr(struct dentry *entry, const char *name,
1697 			 const void *value, size_t size, int flags)
1698 {
1699 	struct inode *inode = entry->d_inode;
1700 	struct fuse_conn *fc = get_fuse_conn(inode);
1701 	struct fuse_req *req;
1702 	struct fuse_setxattr_in inarg;
1703 	int err;
1704 
1705 	if (fc->no_setxattr)
1706 		return -EOPNOTSUPP;
1707 
1708 	req = fuse_get_req_nopages(fc);
1709 	if (IS_ERR(req))
1710 		return PTR_ERR(req);
1711 
1712 	memset(&inarg, 0, sizeof(inarg));
1713 	inarg.size = size;
1714 	inarg.flags = flags;
1715 	req->in.h.opcode = FUSE_SETXATTR;
1716 	req->in.h.nodeid = get_node_id(inode);
1717 	req->in.numargs = 3;
1718 	req->in.args[0].size = sizeof(inarg);
1719 	req->in.args[0].value = &inarg;
1720 	req->in.args[1].size = strlen(name) + 1;
1721 	req->in.args[1].value = name;
1722 	req->in.args[2].size = size;
1723 	req->in.args[2].value = value;
1724 	fuse_request_send(fc, req);
1725 	err = req->out.h.error;
1726 	fuse_put_request(fc, req);
1727 	if (err == -ENOSYS) {
1728 		fc->no_setxattr = 1;
1729 		err = -EOPNOTSUPP;
1730 	}
1731 	return err;
1732 }
1733 
1734 static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1735 			     void *value, size_t size)
1736 {
1737 	struct inode *inode = entry->d_inode;
1738 	struct fuse_conn *fc = get_fuse_conn(inode);
1739 	struct fuse_req *req;
1740 	struct fuse_getxattr_in inarg;
1741 	struct fuse_getxattr_out outarg;
1742 	ssize_t ret;
1743 
1744 	if (fc->no_getxattr)
1745 		return -EOPNOTSUPP;
1746 
1747 	req = fuse_get_req_nopages(fc);
1748 	if (IS_ERR(req))
1749 		return PTR_ERR(req);
1750 
1751 	memset(&inarg, 0, sizeof(inarg));
1752 	inarg.size = size;
1753 	req->in.h.opcode = FUSE_GETXATTR;
1754 	req->in.h.nodeid = get_node_id(inode);
1755 	req->in.numargs = 2;
1756 	req->in.args[0].size = sizeof(inarg);
1757 	req->in.args[0].value = &inarg;
1758 	req->in.args[1].size = strlen(name) + 1;
1759 	req->in.args[1].value = name;
1760 	/* This is really two different operations rolled into one */
1761 	req->out.numargs = 1;
1762 	if (size) {
1763 		req->out.argvar = 1;
1764 		req->out.args[0].size = size;
1765 		req->out.args[0].value = value;
1766 	} else {
1767 		req->out.args[0].size = sizeof(outarg);
1768 		req->out.args[0].value = &outarg;
1769 	}
1770 	fuse_request_send(fc, req);
1771 	ret = req->out.h.error;
1772 	if (!ret)
1773 		ret = size ? req->out.args[0].size : outarg.size;
1774 	else {
1775 		if (ret == -ENOSYS) {
1776 			fc->no_getxattr = 1;
1777 			ret = -EOPNOTSUPP;
1778 		}
1779 	}
1780 	fuse_put_request(fc, req);
1781 	return ret;
1782 }
1783 
1784 static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1785 {
1786 	struct inode *inode = entry->d_inode;
1787 	struct fuse_conn *fc = get_fuse_conn(inode);
1788 	struct fuse_req *req;
1789 	struct fuse_getxattr_in inarg;
1790 	struct fuse_getxattr_out outarg;
1791 	ssize_t ret;
1792 
1793 	if (!fuse_allow_current_process(fc))
1794 		return -EACCES;
1795 
1796 	if (fc->no_listxattr)
1797 		return -EOPNOTSUPP;
1798 
1799 	req = fuse_get_req_nopages(fc);
1800 	if (IS_ERR(req))
1801 		return PTR_ERR(req);
1802 
1803 	memset(&inarg, 0, sizeof(inarg));
1804 	inarg.size = size;
1805 	req->in.h.opcode = FUSE_LISTXATTR;
1806 	req->in.h.nodeid = get_node_id(inode);
1807 	req->in.numargs = 1;
1808 	req->in.args[0].size = sizeof(inarg);
1809 	req->in.args[0].value = &inarg;
1810 	/* This is really two different operations rolled into one */
1811 	req->out.numargs = 1;
1812 	if (size) {
1813 		req->out.argvar = 1;
1814 		req->out.args[0].size = size;
1815 		req->out.args[0].value = list;
1816 	} else {
1817 		req->out.args[0].size = sizeof(outarg);
1818 		req->out.args[0].value = &outarg;
1819 	}
1820 	fuse_request_send(fc, req);
1821 	ret = req->out.h.error;
1822 	if (!ret)
1823 		ret = size ? req->out.args[0].size : outarg.size;
1824 	else {
1825 		if (ret == -ENOSYS) {
1826 			fc->no_listxattr = 1;
1827 			ret = -EOPNOTSUPP;
1828 		}
1829 	}
1830 	fuse_put_request(fc, req);
1831 	return ret;
1832 }
1833 
1834 static int fuse_removexattr(struct dentry *entry, const char *name)
1835 {
1836 	struct inode *inode = entry->d_inode;
1837 	struct fuse_conn *fc = get_fuse_conn(inode);
1838 	struct fuse_req *req;
1839 	int err;
1840 
1841 	if (fc->no_removexattr)
1842 		return -EOPNOTSUPP;
1843 
1844 	req = fuse_get_req_nopages(fc);
1845 	if (IS_ERR(req))
1846 		return PTR_ERR(req);
1847 
1848 	req->in.h.opcode = FUSE_REMOVEXATTR;
1849 	req->in.h.nodeid = get_node_id(inode);
1850 	req->in.numargs = 1;
1851 	req->in.args[0].size = strlen(name) + 1;
1852 	req->in.args[0].value = name;
1853 	fuse_request_send(fc, req);
1854 	err = req->out.h.error;
1855 	fuse_put_request(fc, req);
1856 	if (err == -ENOSYS) {
1857 		fc->no_removexattr = 1;
1858 		err = -EOPNOTSUPP;
1859 	}
1860 	return err;
1861 }
1862 
1863 static const struct inode_operations fuse_dir_inode_operations = {
1864 	.lookup		= fuse_lookup,
1865 	.mkdir		= fuse_mkdir,
1866 	.symlink	= fuse_symlink,
1867 	.unlink		= fuse_unlink,
1868 	.rmdir		= fuse_rmdir,
1869 	.rename		= fuse_rename,
1870 	.link		= fuse_link,
1871 	.setattr	= fuse_setattr,
1872 	.create		= fuse_create,
1873 	.atomic_open	= fuse_atomic_open,
1874 	.mknod		= fuse_mknod,
1875 	.permission	= fuse_permission,
1876 	.getattr	= fuse_getattr,
1877 	.setxattr	= fuse_setxattr,
1878 	.getxattr	= fuse_getxattr,
1879 	.listxattr	= fuse_listxattr,
1880 	.removexattr	= fuse_removexattr,
1881 };
1882 
1883 static const struct file_operations fuse_dir_operations = {
1884 	.llseek		= generic_file_llseek,
1885 	.read		= generic_read_dir,
1886 	.iterate	= fuse_readdir,
1887 	.open		= fuse_dir_open,
1888 	.release	= fuse_dir_release,
1889 	.fsync		= fuse_dir_fsync,
1890 	.unlocked_ioctl	= fuse_dir_ioctl,
1891 	.compat_ioctl	= fuse_dir_compat_ioctl,
1892 };
1893 
1894 static const struct inode_operations fuse_common_inode_operations = {
1895 	.setattr	= fuse_setattr,
1896 	.permission	= fuse_permission,
1897 	.getattr	= fuse_getattr,
1898 	.setxattr	= fuse_setxattr,
1899 	.getxattr	= fuse_getxattr,
1900 	.listxattr	= fuse_listxattr,
1901 	.removexattr	= fuse_removexattr,
1902 };
1903 
1904 static const struct inode_operations fuse_symlink_inode_operations = {
1905 	.setattr	= fuse_setattr,
1906 	.follow_link	= fuse_follow_link,
1907 	.put_link	= fuse_put_link,
1908 	.readlink	= generic_readlink,
1909 	.getattr	= fuse_getattr,
1910 	.setxattr	= fuse_setxattr,
1911 	.getxattr	= fuse_getxattr,
1912 	.listxattr	= fuse_listxattr,
1913 	.removexattr	= fuse_removexattr,
1914 };
1915 
1916 void fuse_init_common(struct inode *inode)
1917 {
1918 	inode->i_op = &fuse_common_inode_operations;
1919 }
1920 
1921 void fuse_init_dir(struct inode *inode)
1922 {
1923 	inode->i_op = &fuse_dir_inode_operations;
1924 	inode->i_fop = &fuse_dir_operations;
1925 }
1926 
1927 void fuse_init_symlink(struct inode *inode)
1928 {
1929 	inode->i_op = &fuse_symlink_inode_operations;
1930 }
1931