xref: /freebsd/sys/fs/fuse/fuse_vnops.c (revision 7cc42f6d25ef2e19059d088fa7d4853fe9afefb5)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2007-2009 Google Inc. and Amit Singh
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions are
9  * met:
10  *
11  * * Redistributions of source code must retain the above copyright
12  *   notice, this list of conditions and the following disclaimer.
13  * * Redistributions in binary form must reproduce the above
14  *   copyright notice, this list of conditions and the following disclaimer
15  *   in the documentation and/or other materials provided with the
16  *   distribution.
17  * * Neither the name of Google Inc. nor the names of its
18  *   contributors may be used to endorse or promote products derived from
19  *   this software without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  *
33  * Copyright (C) 2005 Csaba Henk.
34  * All rights reserved.
35  *
36  * Copyright (c) 2019 The FreeBSD Foundation
37  *
38  * Portions of this software were developed by BFF Storage Systems, LLC under
39  * sponsorship from the FreeBSD Foundation.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  *
50  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  */
62 
63 #include <sys/cdefs.h>
64 __FBSDID("$FreeBSD$");
65 
66 #include <sys/param.h>
67 #include <sys/module.h>
68 #include <sys/systm.h>
69 #include <sys/errno.h>
70 #include <sys/kernel.h>
71 #include <sys/conf.h>
72 #include <sys/uio.h>
73 #include <sys/malloc.h>
74 #include <sys/queue.h>
75 #include <sys/limits.h>
76 #include <sys/lock.h>
77 #include <sys/rwlock.h>
78 #include <sys/sx.h>
79 #include <sys/proc.h>
80 #include <sys/mount.h>
81 #include <sys/vnode.h>
82 #include <sys/namei.h>
83 #include <sys/extattr.h>
84 #include <sys/stat.h>
85 #include <sys/unistd.h>
86 #include <sys/filedesc.h>
87 #include <sys/file.h>
88 #include <sys/fcntl.h>
89 #include <sys/dirent.h>
90 #include <sys/bio.h>
91 #include <sys/buf.h>
92 #include <sys/sysctl.h>
93 #include <sys/vmmeter.h>
94 
95 #include <vm/vm.h>
96 #include <vm/vm_extern.h>
97 #include <vm/pmap.h>
98 #include <vm/vm_map.h>
99 #include <vm/vm_page.h>
100 #include <vm/vm_param.h>
101 #include <vm/vm_object.h>
102 #include <vm/vm_pager.h>
103 #include <vm/vnode_pager.h>
104 #include <vm/vm_object.h>
105 
106 #include "fuse.h"
107 #include "fuse_file.h"
108 #include "fuse_internal.h"
109 #include "fuse_ipc.h"
110 #include "fuse_node.h"
111 #include "fuse_io.h"
112 
113 #include <sys/priv.h>
114 
115 /* Maximum number of hardlinks to a single FUSE file */
116 #define FUSE_LINK_MAX                      UINT32_MAX
117 
118 SDT_PROVIDER_DECLARE(fusefs);
119 /*
120  * Fuse trace probe:
121  * arg0: verbosity.  Higher numbers give more verbose messages
122  * arg1: Textual message
123  */
124 SDT_PROBE_DEFINE2(fusefs, , vnops, trace, "int", "char*");
125 
126 /* vnode ops */
127 static vop_access_t fuse_vnop_access;
128 static vop_advlock_t fuse_vnop_advlock;
129 static vop_bmap_t fuse_vnop_bmap;
130 static vop_close_t fuse_fifo_close;
131 static vop_close_t fuse_vnop_close;
132 static vop_create_t fuse_vnop_create;
133 static vop_deleteextattr_t fuse_vnop_deleteextattr;
134 static vop_fdatasync_t fuse_vnop_fdatasync;
135 static vop_fsync_t fuse_vnop_fsync;
136 static vop_getattr_t fuse_vnop_getattr;
137 static vop_getextattr_t fuse_vnop_getextattr;
138 static vop_inactive_t fuse_vnop_inactive;
139 static vop_link_t fuse_vnop_link;
140 static vop_listextattr_t fuse_vnop_listextattr;
141 static vop_lookup_t fuse_vnop_lookup;
142 static vop_mkdir_t fuse_vnop_mkdir;
143 static vop_mknod_t fuse_vnop_mknod;
144 static vop_open_t fuse_vnop_open;
145 static vop_pathconf_t fuse_vnop_pathconf;
146 static vop_read_t fuse_vnop_read;
147 static vop_readdir_t fuse_vnop_readdir;
148 static vop_readlink_t fuse_vnop_readlink;
149 static vop_reclaim_t fuse_vnop_reclaim;
150 static vop_remove_t fuse_vnop_remove;
151 static vop_rename_t fuse_vnop_rename;
152 static vop_rmdir_t fuse_vnop_rmdir;
153 static vop_setattr_t fuse_vnop_setattr;
154 static vop_setextattr_t fuse_vnop_setextattr;
155 static vop_strategy_t fuse_vnop_strategy;
156 static vop_symlink_t fuse_vnop_symlink;
157 static vop_write_t fuse_vnop_write;
158 static vop_getpages_t fuse_vnop_getpages;
159 static vop_print_t fuse_vnop_print;
160 static vop_vptofh_t fuse_vnop_vptofh;
161 
162 struct vop_vector fuse_fifoops = {
163 	.vop_default =		&fifo_specops,
164 	.vop_access =		fuse_vnop_access,
165 	.vop_close =		fuse_fifo_close,
166 	.vop_fsync =		fuse_vnop_fsync,
167 	.vop_getattr =		fuse_vnop_getattr,
168 	.vop_inactive =		fuse_vnop_inactive,
169 	.vop_pathconf =		fuse_vnop_pathconf,
170 	.vop_print =		fuse_vnop_print,
171 	.vop_read =		VOP_PANIC,
172 	.vop_reclaim =		fuse_vnop_reclaim,
173 	.vop_setattr =		fuse_vnop_setattr,
174 	.vop_write =		VOP_PANIC,
175 	.vop_vptofh =		fuse_vnop_vptofh,
176 };
177 VFS_VOP_VECTOR_REGISTER(fuse_fifoops);
178 
179 struct vop_vector fuse_vnops = {
180 	.vop_allocate =	VOP_EINVAL,
181 	.vop_default = &default_vnodeops,
182 	.vop_access = fuse_vnop_access,
183 	.vop_advlock = fuse_vnop_advlock,
184 	.vop_bmap = fuse_vnop_bmap,
185 	.vop_close = fuse_vnop_close,
186 	.vop_create = fuse_vnop_create,
187 	.vop_deleteextattr = fuse_vnop_deleteextattr,
188 	.vop_fsync = fuse_vnop_fsync,
189 	.vop_fdatasync = fuse_vnop_fdatasync,
190 	.vop_getattr = fuse_vnop_getattr,
191 	.vop_getextattr = fuse_vnop_getextattr,
192 	.vop_inactive = fuse_vnop_inactive,
193 	/*
194 	 * TODO: implement vop_ioctl after upgrading to protocol 7.16.
195 	 * FUSE_IOCTL was added in 7.11, but 32-bit compat is broken until
196 	 * 7.16.
197 	 */
198 	.vop_link = fuse_vnop_link,
199 	.vop_listextattr = fuse_vnop_listextattr,
200 	.vop_lookup = fuse_vnop_lookup,
201 	.vop_mkdir = fuse_vnop_mkdir,
202 	.vop_mknod = fuse_vnop_mknod,
203 	.vop_open = fuse_vnop_open,
204 	.vop_pathconf = fuse_vnop_pathconf,
205 	/*
206 	 * TODO: implement vop_poll after upgrading to protocol 7.21.
207 	 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
208 	 * 7.21, which adds the ability for the client to choose which poll
209 	 * events it wants, and for a client to deregister a file handle
210 	 */
211 	.vop_read = fuse_vnop_read,
212 	.vop_readdir = fuse_vnop_readdir,
213 	.vop_readlink = fuse_vnop_readlink,
214 	.vop_reclaim = fuse_vnop_reclaim,
215 	.vop_remove = fuse_vnop_remove,
216 	.vop_rename = fuse_vnop_rename,
217 	.vop_rmdir = fuse_vnop_rmdir,
218 	.vop_setattr = fuse_vnop_setattr,
219 	.vop_setextattr = fuse_vnop_setextattr,
220 	.vop_strategy = fuse_vnop_strategy,
221 	.vop_symlink = fuse_vnop_symlink,
222 	.vop_write = fuse_vnop_write,
223 	.vop_getpages = fuse_vnop_getpages,
224 	.vop_print = fuse_vnop_print,
225 	.vop_vptofh = fuse_vnop_vptofh,
226 };
227 VFS_VOP_VECTOR_REGISTER(fuse_vnops);
228 
229 uma_zone_t fuse_pbuf_zone;
230 
231 /* Check permission for extattr operations, much like extattr_check_cred */
232 static int
233 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
234 	struct thread *td, accmode_t accmode)
235 {
236 	struct mount *mp = vnode_mount(vp);
237 	struct fuse_data *data = fuse_get_mpdata(mp);
238 	int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
239 
240 	/*
241 	 * Kernel-invoked always succeeds.
242 	 */
243 	if (cred == NOCRED)
244 		return (0);
245 
246 	/*
247 	 * Do not allow privileged processes in jail to directly manipulate
248 	 * system attributes.
249 	 */
250 	switch (ns) {
251 	case EXTATTR_NAMESPACE_SYSTEM:
252 		if (default_permissions) {
253 			return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
254 		}
255 		return (0);
256 	case EXTATTR_NAMESPACE_USER:
257 		if (default_permissions) {
258 			return (fuse_internal_access(vp, accmode, td, cred));
259 		}
260 		return (0);
261 	default:
262 		return (EPERM);
263 	}
264 }
265 
266 /* Get a filehandle for a directory */
267 static int
268 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
269 	struct ucred *cred, pid_t pid)
270 {
271 	if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
272 		return 0;
273 	return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
274 }
275 
276 /* Send FUSE_FLUSH for this vnode */
277 static int
278 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
279 {
280 	struct fuse_flush_in *ffi;
281 	struct fuse_filehandle *fufh;
282 	struct fuse_dispatcher fdi;
283 	struct thread *td = curthread;
284 	struct mount *mp = vnode_mount(vp);
285 	int err;
286 
287 	if (!fsess_isimpl(vnode_mount(vp), FUSE_FLUSH))
288 		return 0;
289 
290 	err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
291 	if (err)
292 		return err;
293 
294 	fdisp_init(&fdi, sizeof(*ffi));
295 	fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
296 	ffi = fdi.indata;
297 	ffi->fh = fufh->fh_id;
298 	/*
299 	 * If the file has a POSIX lock then we're supposed to set lock_owner.
300 	 * If not, then lock_owner is undefined.  So we may as well always set
301 	 * it.
302 	 */
303 	ffi->lock_owner = td->td_proc->p_pid;
304 
305 	err = fdisp_wait_answ(&fdi);
306 	if (err == ENOSYS) {
307 		fsess_set_notimpl(mp, FUSE_FLUSH);
308 		err = 0;
309 	}
310 	fdisp_destroy(&fdi);
311 	return err;
312 }
313 
314 /* Close wrapper for fifos.  */
315 static int
316 fuse_fifo_close(struct vop_close_args *ap)
317 {
318 	return (fifo_specops.vop_close(ap));
319 }
320 
321 /*
322     struct vnop_access_args {
323 	struct vnode *a_vp;
324 #if VOP_ACCESS_TAKES_ACCMODE_T
325 	accmode_t a_accmode;
326 #else
327 	int a_mode;
328 #endif
329 	struct ucred *a_cred;
330 	struct thread *a_td;
331     };
332 */
333 static int
334 fuse_vnop_access(struct vop_access_args *ap)
335 {
336 	struct vnode *vp = ap->a_vp;
337 	int accmode = ap->a_accmode;
338 	struct ucred *cred = ap->a_cred;
339 
340 	struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
341 
342 	int err;
343 
344 	if (fuse_isdeadfs(vp)) {
345 		if (vnode_isvroot(vp)) {
346 			return 0;
347 		}
348 		return ENXIO;
349 	}
350 	if (!(data->dataflags & FSESS_INITED)) {
351 		if (vnode_isvroot(vp)) {
352 			if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
353 			    (fuse_match_cred(data->daemoncred, cred) == 0)) {
354 				return 0;
355 			}
356 		}
357 		return EBADF;
358 	}
359 	if (vnode_islnk(vp)) {
360 		return 0;
361 	}
362 
363 	err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
364 	return err;
365 }
366 
367 /*
368  * struct vop_advlock_args {
369  *	struct vop_generic_args a_gen;
370  *	struct vnode *a_vp;
371  *	void *a_id;
372  *	int a_op;
373  *	struct flock *a_fl;
374  *	int a_flags;
375  * }
376  */
377 static int
378 fuse_vnop_advlock(struct vop_advlock_args *ap)
379 {
380 	struct vnode *vp = ap->a_vp;
381 	struct flock *fl = ap->a_fl;
382 	struct thread *td = curthread;
383 	struct ucred *cred = td->td_ucred;
384 	pid_t pid = td->td_proc->p_pid;
385 	struct fuse_filehandle *fufh;
386 	struct fuse_dispatcher fdi;
387 	struct fuse_lk_in *fli;
388 	struct fuse_lk_out *flo;
389 	enum fuse_opcode op;
390 	int dataflags, err;
391 	int flags = ap->a_flags;
392 
393 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
394 
395 	if (fuse_isdeadfs(vp)) {
396 		return ENXIO;
397 	}
398 
399 	if (!(dataflags & FSESS_POSIX_LOCKS))
400 		return vop_stdadvlock(ap);
401 	/* FUSE doesn't properly support flock until protocol 7.17 */
402 	if (flags & F_FLOCK)
403 		return vop_stdadvlock(ap);
404 
405 	err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
406 	if (err)
407 		return err;
408 
409 	fdisp_init(&fdi, sizeof(*fli));
410 
411 	switch(ap->a_op) {
412 	case F_GETLK:
413 		op = FUSE_GETLK;
414 		break;
415 	case F_SETLK:
416 		op = FUSE_SETLK;
417 		break;
418 	case F_SETLKW:
419 		op = FUSE_SETLKW;
420 		break;
421 	default:
422 		return EINVAL;
423 	}
424 
425 	fdisp_make_vp(&fdi, op, vp, td, cred);
426 	fli = fdi.indata;
427 	fli->fh = fufh->fh_id;
428 	fli->owner = fl->l_pid;
429 	fli->lk.start = fl->l_start;
430 	if (fl->l_len != 0)
431 		fli->lk.end = fl->l_start + fl->l_len - 1;
432 	else
433 		fli->lk.end = INT64_MAX;
434 	fli->lk.type = fl->l_type;
435 	fli->lk.pid = fl->l_pid;
436 
437 	err = fdisp_wait_answ(&fdi);
438 	fdisp_destroy(&fdi);
439 
440 	if (err == 0 && op == FUSE_GETLK) {
441 		flo = fdi.answ;
442 		fl->l_type = flo->lk.type;
443 		fl->l_pid = flo->lk.pid;
444 		if (flo->lk.type != F_UNLCK) {
445 			fl->l_start = flo->lk.start;
446 			if (flo->lk.end == INT64_MAX)
447 				fl->l_len = 0;
448 			else
449 				fl->l_len = flo->lk.end - flo->lk.start + 1;
450 			fl->l_start = flo->lk.start;
451 		}
452 	}
453 
454 	return err;
455 }
456 
457 /* {
458 	struct vnode *a_vp;
459 	daddr_t a_bn;
460 	struct bufobj **a_bop;
461 	daddr_t *a_bnp;
462 	int *a_runp;
463 	int *a_runb;
464 } */
465 static int
466 fuse_vnop_bmap(struct vop_bmap_args *ap)
467 {
468 	struct vnode *vp = ap->a_vp;
469 	struct bufobj **bo = ap->a_bop;
470 	struct thread *td = curthread;
471 	struct mount *mp;
472 	struct fuse_dispatcher fdi;
473 	struct fuse_bmap_in *fbi;
474 	struct fuse_bmap_out *fbo;
475 	struct fuse_data *data;
476 	uint64_t biosize;
477 	off_t filesize;
478 	daddr_t lbn = ap->a_bn;
479 	daddr_t *pbn = ap->a_bnp;
480 	int *runp = ap->a_runp;
481 	int *runb = ap->a_runb;
482 	int error = 0;
483 	int maxrun;
484 
485 	if (fuse_isdeadfs(vp)) {
486 		return ENXIO;
487 	}
488 
489 	mp = vnode_mount(vp);
490 	data = fuse_get_mpdata(mp);
491 	biosize = fuse_iosize(vp);
492 	maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
493 		data->max_readahead_blocks);
494 
495 	if (bo != NULL)
496 		*bo = &vp->v_bufobj;
497 
498 	/*
499 	 * The FUSE_BMAP operation does not include the runp and runb
500 	 * variables, so we must guess.  Report nonzero contiguous runs so
501 	 * cluster_read will combine adjacent reads.  It's worthwhile to reduce
502 	 * upcalls even if we don't know the true physical layout of the file.
503 	 *
504 	 * FUSE file systems may opt out of read clustering in two ways:
505 	 * * mounting with -onoclusterr
506 	 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
507 	 */
508 	if (runb != NULL)
509 		*runb = MIN(lbn, maxrun);
510 	if (runp != NULL) {
511 		error = fuse_vnode_size(vp, &filesize, td->td_ucred, td);
512 		if (error == 0)
513 			*runp = MIN(MAX(0, filesize / (off_t)biosize - lbn - 1),
514 				    maxrun);
515 		else
516 			*runp = 0;
517 	}
518 
519 	if (fsess_isimpl(mp, FUSE_BMAP)) {
520 		fdisp_init(&fdi, sizeof(*fbi));
521 		fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
522 		fbi = fdi.indata;
523 		fbi->block = lbn;
524 		fbi->blocksize = biosize;
525 		error = fdisp_wait_answ(&fdi);
526 		if (error == ENOSYS) {
527 			fdisp_destroy(&fdi);
528 			fsess_set_notimpl(mp, FUSE_BMAP);
529 			error = 0;
530 		} else {
531 			fbo = fdi.answ;
532 			if (error == 0 && pbn != NULL)
533 				*pbn = fbo->block;
534 			fdisp_destroy(&fdi);
535 			return error;
536 		}
537 	}
538 
539 	/* If the daemon doesn't support BMAP, make up a sensible default */
540 	if (pbn != NULL)
541 		*pbn = lbn * btodb(biosize);
542 	return (error);
543 }
544 
545 /*
546     struct vop_close_args {
547 	struct vnode *a_vp;
548 	int  a_fflag;
549 	struct ucred *a_cred;
550 	struct thread *a_td;
551     };
552 */
553 static int
554 fuse_vnop_close(struct vop_close_args *ap)
555 {
556 	struct vnode *vp = ap->a_vp;
557 	struct ucred *cred = ap->a_cred;
558 	int fflag = ap->a_fflag;
559 	struct thread *td = ap->a_td;
560 	pid_t pid = td->td_proc->p_pid;
561 	int err = 0;
562 
563 	if (fuse_isdeadfs(vp))
564 		return 0;
565 	if (vnode_isdir(vp))
566 		return 0;
567 	if (fflag & IO_NDELAY)
568 		return 0;
569 
570 	err = fuse_flush(vp, cred, pid, fflag);
571 	/* TODO: close the file handle, if we're sure it's no longer used */
572 	if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
573 		fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
574 	}
575 	return err;
576 }
577 
578 static void
579 fdisp_make_mknod_for_fallback(
580 	struct fuse_dispatcher *fdip,
581 	struct componentname *cnp,
582 	struct vnode *dvp,
583 	uint64_t parentnid,
584 	struct thread *td,
585 	struct ucred *cred,
586 	mode_t mode,
587 	enum fuse_opcode *op)
588 {
589 	struct fuse_mknod_in *fmni;
590 
591 	fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
592 	*op = FUSE_MKNOD;
593 	fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
594 	fmni = fdip->indata;
595 	fmni->mode = mode;
596 	fmni->rdev = 0;
597 	memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
598 	    cnp->cn_namelen);
599 	((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
600 }
601 /*
602     struct vnop_create_args {
603 	struct vnode *a_dvp;
604 	struct vnode **a_vpp;
605 	struct componentname *a_cnp;
606 	struct vattr *a_vap;
607     };
608 */
609 static int
610 fuse_vnop_create(struct vop_create_args *ap)
611 {
612 	struct vnode *dvp = ap->a_dvp;
613 	struct vnode **vpp = ap->a_vpp;
614 	struct componentname *cnp = ap->a_cnp;
615 	struct vattr *vap = ap->a_vap;
616 	struct thread *td = cnp->cn_thread;
617 	struct ucred *cred = cnp->cn_cred;
618 
619 	struct fuse_data *data;
620 	struct fuse_create_in *fci;
621 	struct fuse_entry_out *feo;
622 	struct fuse_open_out *foo;
623 	struct fuse_dispatcher fdi, fdi2;
624 	struct fuse_dispatcher *fdip = &fdi;
625 	struct fuse_dispatcher *fdip2 = NULL;
626 
627 	int err;
628 
629 	struct mount *mp = vnode_mount(dvp);
630 	data = fuse_get_mpdata(mp);
631 	uint64_t parentnid = VTOFUD(dvp)->nid;
632 	mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
633 	enum fuse_opcode op;
634 	int flags;
635 
636 	if (fuse_isdeadfs(dvp))
637 		return ENXIO;
638 
639 	/* FUSE expects sockets to be created with FUSE_MKNOD */
640 	if (vap->va_type == VSOCK)
641 		return fuse_internal_mknod(dvp, vpp, cnp, vap);
642 
643 	/*
644 	 * VOP_CREATE doesn't tell us the open(2) flags, so we guess.  Only a
645 	 * writable mode makes sense, and we might as well include readability
646 	 * too.
647 	 */
648 	flags = O_RDWR;
649 
650 	bzero(&fdi, sizeof(fdi));
651 
652 	if (vap->va_type != VREG)
653 		return (EINVAL);
654 
655 	if (!fsess_isimpl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
656 		/* Fallback to FUSE_MKNOD/FUSE_OPEN */
657 		fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
658 			cred, mode, &op);
659 	} else {
660 		/* Use FUSE_CREATE */
661 		size_t insize;
662 
663 		op = FUSE_CREATE;
664 		fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
665 		fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
666 		fci = fdip->indata;
667 		fci->mode = mode;
668 		fci->flags = O_CREAT | flags;
669 		if (fuse_libabi_geq(data, 7, 12)) {
670 			insize = sizeof(*fci);
671 			fci->umask = td->td_proc->p_fd->fd_cmask;
672 		} else {
673 			insize = sizeof(struct fuse_open_in);
674 		}
675 
676 		memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
677 		    cnp->cn_namelen);
678 		((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
679 	}
680 
681 	err = fdisp_wait_answ(fdip);
682 
683 	if (err) {
684 		if (err == ENOSYS && op == FUSE_CREATE) {
685 			fsess_set_notimpl(mp, FUSE_CREATE);
686 			fdisp_destroy(fdip);
687 			fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
688 				parentnid, td, cred, mode, &op);
689 			err = fdisp_wait_answ(fdip);
690 		}
691 		if (err)
692 			goto out;
693 	}
694 
695 	feo = fdip->answ;
696 
697 	if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
698 		goto out;
699 	}
700 
701 	if (op == FUSE_CREATE) {
702 		foo = (struct fuse_open_out*)(feo + 1);
703 	} else {
704 		/* Issue a separate FUSE_OPEN */
705 		struct fuse_open_in *foi;
706 
707 		fdip2 = &fdi2;
708 		fdisp_init(fdip2, sizeof(*foi));
709 		fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
710 			cred);
711 		foi = fdip2->indata;
712 		foi->flags = flags;
713 		err = fdisp_wait_answ(fdip2);
714 		if (err)
715 			goto out;
716 		foo = fdip2->answ;
717 	}
718 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
719 	if (err) {
720 		struct fuse_release_in *fri;
721 		uint64_t nodeid = feo->nodeid;
722 		uint64_t fh_id = foo->fh;
723 
724 		fdisp_init(fdip, sizeof(*fri));
725 		fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
726 		fri = fdip->indata;
727 		fri->fh = fh_id;
728 		fri->flags = flags;
729 		fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
730 		fuse_insert_message(fdip->tick, false);
731 		goto out;
732 	}
733 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
734 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
735 		feo->attr_valid_nsec, NULL);
736 
737 	fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
738 	fuse_vnode_open(*vpp, foo->open_flags, td);
739 	/*
740 	 * Purge the parent's attribute cache because the daemon should've
741 	 * updated its mtime and ctime
742 	 */
743 	fuse_vnode_clear_attr_cache(dvp);
744 	cache_purge_negative(dvp);
745 
746 out:
747 	if (fdip2)
748 		fdisp_destroy(fdip2);
749 	fdisp_destroy(fdip);
750 	return err;
751 }
752 
753 /*
754     struct vnop_fdatasync_args {
755 	struct vop_generic_args a_gen;
756 	struct vnode * a_vp;
757 	struct thread * a_td;
758     };
759 */
760 static int
761 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
762 {
763 	struct vnode *vp = ap->a_vp;
764 	struct thread *td = ap->a_td;
765 	int waitfor = MNT_WAIT;
766 
767 	int err = 0;
768 
769 	if (fuse_isdeadfs(vp)) {
770 		return 0;
771 	}
772 	if ((err = vop_stdfdatasync_buf(ap)))
773 		return err;
774 
775 	return fuse_internal_fsync(vp, td, waitfor, true);
776 }
777 
778 /*
779     struct vnop_fsync_args {
780 	struct vop_generic_args a_gen;
781 	struct vnode * a_vp;
782 	int  a_waitfor;
783 	struct thread * a_td;
784     };
785 */
786 static int
787 fuse_vnop_fsync(struct vop_fsync_args *ap)
788 {
789 	struct vnode *vp = ap->a_vp;
790 	struct thread *td = ap->a_td;
791 	int waitfor = ap->a_waitfor;
792 	int err = 0;
793 
794 	if (fuse_isdeadfs(vp)) {
795 		return 0;
796 	}
797 	if ((err = vop_stdfsync(ap)))
798 		return err;
799 
800 	return fuse_internal_fsync(vp, td, waitfor, false);
801 }
802 
803 /*
804     struct vnop_getattr_args {
805 	struct vnode *a_vp;
806 	struct vattr *a_vap;
807 	struct ucred *a_cred;
808 	struct thread *a_td;
809     };
810 */
811 static int
812 fuse_vnop_getattr(struct vop_getattr_args *ap)
813 {
814 	struct vnode *vp = ap->a_vp;
815 	struct vattr *vap = ap->a_vap;
816 	struct ucred *cred = ap->a_cred;
817 	struct thread *td = curthread;
818 
819 	int err = 0;
820 	int dataflags;
821 
822 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
823 
824 	/* Note that we are not bailing out on a dead file system just yet. */
825 
826 	if (!(dataflags & FSESS_INITED)) {
827 		if (!vnode_isvroot(vp)) {
828 			fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
829 			err = ENOTCONN;
830 			return err;
831 		} else {
832 			goto fake;
833 		}
834 	}
835 	err = fuse_internal_getattr(vp, vap, cred, td);
836 	if (err == ENOTCONN && vnode_isvroot(vp)) {
837 		/* see comment in fuse_vfsop_statfs() */
838 		goto fake;
839 	} else {
840 		return err;
841 	}
842 
843 fake:
844 	bzero(vap, sizeof(*vap));
845 	vap->va_type = vnode_vtype(vp);
846 
847 	return 0;
848 }
849 
850 /*
851     struct vnop_inactive_args {
852 	struct vnode *a_vp;
853 	struct thread *a_td;
854     };
855 */
856 static int
857 fuse_vnop_inactive(struct vop_inactive_args *ap)
858 {
859 	struct vnode *vp = ap->a_vp;
860 	struct thread *td = ap->a_td;
861 
862 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
863 	struct fuse_filehandle *fufh, *fufh_tmp;
864 
865 	int need_flush = 1;
866 
867 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
868 		if (need_flush && vp->v_type == VREG) {
869 			if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
870 				fuse_vnode_savesize(vp, NULL, 0);
871 			}
872 			if ((fvdat->flag & FN_REVOKED) != 0)
873 				fuse_io_invalbuf(vp, td);
874 			else
875 				fuse_io_flushbuf(vp, MNT_WAIT, td);
876 			need_flush = 0;
877 		}
878 		fuse_filehandle_close(vp, fufh, td, NULL);
879 	}
880 
881 	if ((fvdat->flag & FN_REVOKED) != 0)
882 		vrecycle(vp);
883 
884 	return 0;
885 }
886 
887 /*
888     struct vnop_link_args {
889 	struct vnode *a_tdvp;
890 	struct vnode *a_vp;
891 	struct componentname *a_cnp;
892     };
893 */
894 static int
895 fuse_vnop_link(struct vop_link_args *ap)
896 {
897 	struct vnode *vp = ap->a_vp;
898 	struct vnode *tdvp = ap->a_tdvp;
899 	struct componentname *cnp = ap->a_cnp;
900 
901 	struct vattr *vap = VTOVA(vp);
902 
903 	struct fuse_dispatcher fdi;
904 	struct fuse_entry_out *feo;
905 	struct fuse_link_in fli;
906 
907 	int err;
908 
909 	if (fuse_isdeadfs(vp)) {
910 		return ENXIO;
911 	}
912 	if (vnode_mount(tdvp) != vnode_mount(vp)) {
913 		return EXDEV;
914 	}
915 
916 	/*
917 	 * This is a seatbelt check to protect naive userspace filesystems from
918 	 * themselves and the limitations of the FUSE IPC protocol.  If a
919 	 * filesystem does not allow attribute caching, assume it is capable of
920 	 * validating that nlink does not overflow.
921 	 */
922 	if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
923 		return EMLINK;
924 	fli.oldnodeid = VTOI(vp);
925 
926 	fdisp_init(&fdi, 0);
927 	fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
928 	    FUSE_LINK, &fli, sizeof(fli), &fdi);
929 	if ((err = fdisp_wait_answ(&fdi))) {
930 		goto out;
931 	}
932 	feo = fdi.answ;
933 
934 	err = fuse_internal_checkentry(feo, vnode_vtype(vp));
935 	if (!err) {
936 		/*
937 		 * Purge the parent's attribute cache because the daemon
938 		 * should've updated its mtime and ctime
939 		 */
940 		fuse_vnode_clear_attr_cache(tdvp);
941 		fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
942 			feo->attr_valid_nsec, NULL);
943 	}
944 out:
945 	fdisp_destroy(&fdi);
946 	return err;
947 }
948 
949 struct fuse_lookup_alloc_arg {
950 	struct fuse_entry_out *feo;
951 	struct componentname *cnp;
952 	uint64_t nid;
953 	enum vtype vtyp;
954 };
955 
956 /* Callback for vn_get_ino */
957 static int
958 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
959 {
960 	struct fuse_lookup_alloc_arg *flaa = arg;
961 
962 	return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
963 		flaa->vtyp);
964 }
965 
966 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
967 	"int", "struct timespec*", "struct timespec*");
968 SDT_PROBE_DEFINE2(fusefs, , vnops, lookup_cache_incoherent,
969 	"struct vnode*", "struct fuse_entry_out*");
970 /*
971     struct vnop_lookup_args {
972 	struct vnodeop_desc *a_desc;
973 	struct vnode *a_dvp;
974 	struct vnode **a_vpp;
975 	struct componentname *a_cnp;
976     };
977 */
978 int
979 fuse_vnop_lookup(struct vop_lookup_args *ap)
980 {
981 	struct vnode *dvp = ap->a_dvp;
982 	struct vnode **vpp = ap->a_vpp;
983 	struct componentname *cnp = ap->a_cnp;
984 	struct thread *td = cnp->cn_thread;
985 	struct ucred *cred = cnp->cn_cred;
986 
987 	int nameiop = cnp->cn_nameiop;
988 	int flags = cnp->cn_flags;
989 	int wantparent = flags & (LOCKPARENT | WANTPARENT);
990 	int islastcn = flags & ISLASTCN;
991 	struct mount *mp = vnode_mount(dvp);
992 	struct fuse_data *data = fuse_get_mpdata(mp);
993 	int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
994 
995 	int err = 0;
996 	int lookup_err = 0;
997 	struct vnode *vp = NULL;
998 
999 	struct fuse_dispatcher fdi;
1000 	bool did_lookup = false;
1001 	struct fuse_entry_out *feo = NULL;
1002 	enum vtype vtyp;	/* vnode type of target */
1003 	off_t filesize;		/* filesize of target */
1004 
1005 	uint64_t nid;
1006 
1007 	if (fuse_isdeadfs(dvp)) {
1008 		*vpp = NULL;
1009 		return ENXIO;
1010 	}
1011 	if (!vnode_isdir(dvp))
1012 		return ENOTDIR;
1013 
1014 	if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1015 		return EROFS;
1016 
1017 	if ((cnp->cn_flags & NOEXECCHECK) != 0)
1018 		cnp->cn_flags &= ~NOEXECCHECK;
1019 	else if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1020 		return err;
1021 
1022 	if (flags & ISDOTDOT) {
1023 		KASSERT(VTOFUD(dvp)->flag & FN_PARENT_NID,
1024 			("Looking up .. is TODO"));
1025 		nid = VTOFUD(dvp)->parent_nid;
1026 		if (nid == 0)
1027 			return ENOENT;
1028 		/* .. is obviously a directory */
1029 		vtyp = VDIR;
1030 		filesize = 0;
1031 	} else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
1032 		nid = VTOI(dvp);
1033 		/* . is obviously a directory */
1034 		vtyp = VDIR;
1035 		filesize = 0;
1036 	} else {
1037 		struct timespec now, timeout;
1038 		int ncpticks; /* here to accomodate for API contract */
1039 
1040 		err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks);
1041 		getnanouptime(&now);
1042 		SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1043 		switch (err) {
1044 		case -1:		/* positive match */
1045 			if (timespeccmp(&timeout, &now, >)) {
1046 				counter_u64_add(fuse_lookup_cache_hits, 1);
1047 			} else {
1048 				/* Cache timeout */
1049 				counter_u64_add(fuse_lookup_cache_misses, 1);
1050 				bintime_clear(
1051 					&VTOFUD(*vpp)->entry_cache_timeout);
1052 				cache_purge(*vpp);
1053 				if (dvp != *vpp)
1054 					vput(*vpp);
1055 				else
1056 					vrele(*vpp);
1057 				*vpp = NULL;
1058 				break;
1059 			}
1060 			return 0;
1061 
1062 		case 0:		/* no match in cache */
1063 			counter_u64_add(fuse_lookup_cache_misses, 1);
1064 			break;
1065 
1066 		case ENOENT:		/* negative match */
1067 			getnanouptime(&now);
1068 			if (timespeccmp(&timeout, &now, <=)) {
1069 				/* Cache timeout */
1070 				cache_purge_negative(dvp);
1071 				break;
1072 			}
1073 			/* fall through */
1074 		default:
1075 			return err;
1076 		}
1077 
1078 		nid = VTOI(dvp);
1079 		fdisp_init(&fdi, cnp->cn_namelen + 1);
1080 		fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred);
1081 
1082 		memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1083 		((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1084 		lookup_err = fdisp_wait_answ(&fdi);
1085 		did_lookup = true;
1086 
1087 		if (!lookup_err) {
1088 			/* lookup call succeeded */
1089 			feo = (struct fuse_entry_out *)fdi.answ;
1090 			nid = feo->nodeid;
1091 			if (nid == 0) {
1092 				/* zero nodeid means ENOENT and cache it */
1093 				struct timespec timeout;
1094 
1095 				fdi.answ_stat = ENOENT;
1096 				lookup_err = ENOENT;
1097 				if (cnp->cn_flags & MAKEENTRY) {
1098 					fuse_validity_2_timespec(feo, &timeout);
1099 					cache_enter_time(dvp, *vpp, cnp,
1100 						&timeout, NULL);
1101 				}
1102 			} else if (nid == FUSE_ROOT_ID) {
1103 				lookup_err = EINVAL;
1104 			}
1105 			vtyp = IFTOVT(feo->attr.mode);
1106 			filesize = feo->attr.size;
1107 		}
1108 		if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1109 			fdisp_destroy(&fdi);
1110 			return lookup_err;
1111 		}
1112 	}
1113 	/* lookup_err, if non-zero, must be ENOENT at this point */
1114 
1115 	if (lookup_err) {
1116 		/* Entry not found */
1117 		if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1118 			if (default_permissions)
1119 				err = fuse_internal_access(dvp, VWRITE, td,
1120 				    cred);
1121 			else
1122 				err = 0;
1123 			if (!err) {
1124 				/*
1125 				 * Set the SAVENAME flag to hold onto the
1126 				 * pathname for use later in VOP_CREATE or
1127 				 * VOP_RENAME.
1128 				 */
1129 				cnp->cn_flags |= SAVENAME;
1130 
1131 				err = EJUSTRETURN;
1132 			}
1133 		} else {
1134 			err = ENOENT;
1135 		}
1136 	} else {
1137 		/* Entry was found */
1138 		if (flags & ISDOTDOT) {
1139 			struct fuse_lookup_alloc_arg flaa;
1140 
1141 			flaa.nid = nid;
1142 			flaa.feo = feo;
1143 			flaa.cnp = cnp;
1144 			flaa.vtyp = vtyp;
1145 			err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1146 				&vp);
1147 			*vpp = vp;
1148 		} else if (nid == VTOI(dvp)) {
1149 			vref(dvp);
1150 			*vpp = dvp;
1151 		} else {
1152 			struct fuse_vnode_data *fvdat;
1153 			struct vattr *vap;
1154 
1155 			err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1156 			    &vp, cnp, vtyp);
1157 			if (err)
1158 				goto out;
1159 			*vpp = vp;
1160 
1161 			/*
1162 			 * In the case where we are looking up a FUSE node
1163 			 * represented by an existing cached vnode, and the
1164 			 * true size reported by FUSE_LOOKUP doesn't match
1165 			 * the vnode's cached size, then any cached writes
1166 			 * beyond the file's current size are lost.
1167 			 *
1168 			 * We can get here:
1169 			 * * following attribute cache expiration, or
1170 			 * * due a bug in the daemon, or
1171 			 */
1172 			fvdat = VTOFUD(vp);
1173 			if (vnode_isreg(vp) &&
1174 			    ((filesize != fvdat->cached_attrs.va_size &&
1175 			      fvdat->flag & FN_SIZECHANGE) ||
1176 			     ((vap = VTOVA(vp)) &&
1177 			      filesize != vap->va_size)))
1178 			{
1179 				SDT_PROBE2(fusefs, , vnops, lookup_cache_incoherent, vp, feo);
1180 				fvdat->flag &= ~FN_SIZECHANGE;
1181 				/*
1182 				 * The server changed the file's size even
1183 				 * though we had it cached, or had dirty writes
1184 				 * in the WB cache!
1185 				 */
1186 				printf("%s: cache incoherent on %s!  "
1187 		    		    "Buggy FUSE server detected.  To prevent "
1188 				    "data corruption, disable the data cache "
1189 				    "by mounting with -o direct_io, or as "
1190 				    "directed otherwise by your FUSE server's "
1191 		    		    "documentation\n", __func__,
1192 				    vnode_mount(vp)->mnt_stat.f_mntonname);
1193 				int iosize = fuse_iosize(vp);
1194 				v_inval_buf_range(vp, 0, INT64_MAX, iosize);
1195 			}
1196 
1197 			MPASS(feo != NULL);
1198 			fuse_internal_cache_attrs(*vpp, &feo->attr,
1199 				feo->attr_valid, feo->attr_valid_nsec, NULL);
1200 			fuse_validity_2_bintime(feo->entry_valid,
1201 				feo->entry_valid_nsec,
1202 				&fvdat->entry_cache_timeout);
1203 
1204 			if ((nameiop == DELETE || nameiop == RENAME) &&
1205 				islastcn && default_permissions)
1206 			{
1207 				struct vattr dvattr;
1208 
1209 				err = fuse_internal_access(dvp, VWRITE, td,
1210 					cred);
1211 				if (err != 0)
1212 					goto out;
1213 				/*
1214 				 * if the parent's sticky bit is set, check
1215 				 * whether we're allowed to remove the file.
1216 				 * Need to figure out the vnode locking to make
1217 				 * this work.
1218 				 */
1219 				fuse_internal_getattr(dvp, &dvattr, cred, td);
1220 				if ((dvattr.va_mode & S_ISTXT) &&
1221 					fuse_internal_access(dvp, VADMIN, td,
1222 						cred) &&
1223 					fuse_internal_access(*vpp, VADMIN, td,
1224 						cred)) {
1225 					err = EPERM;
1226 					goto out;
1227 				}
1228 			}
1229 
1230 			if (islastcn && (
1231 				(nameiop == DELETE) ||
1232 				(nameiop == RENAME && wantparent))) {
1233 				cnp->cn_flags |= SAVENAME;
1234 			}
1235 		}
1236 	}
1237 out:
1238 	if (err) {
1239 		if (vp != NULL && dvp != vp)
1240 			vput(vp);
1241 		else if (vp != NULL)
1242 			vrele(vp);
1243 		*vpp = NULL;
1244 	}
1245 	if (did_lookup)
1246 		fdisp_destroy(&fdi);
1247 
1248 	return err;
1249 }
1250 
1251 /*
1252     struct vnop_mkdir_args {
1253 	struct vnode *a_dvp;
1254 	struct vnode **a_vpp;
1255 	struct componentname *a_cnp;
1256 	struct vattr *a_vap;
1257     };
1258 */
1259 static int
1260 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1261 {
1262 	struct vnode *dvp = ap->a_dvp;
1263 	struct vnode **vpp = ap->a_vpp;
1264 	struct componentname *cnp = ap->a_cnp;
1265 	struct vattr *vap = ap->a_vap;
1266 
1267 	struct fuse_mkdir_in fmdi;
1268 
1269 	if (fuse_isdeadfs(dvp)) {
1270 		return ENXIO;
1271 	}
1272 	fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1273 	fmdi.umask = curthread->td_proc->p_fd->fd_cmask;
1274 
1275 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1276 	    sizeof(fmdi), VDIR));
1277 }
1278 
1279 /*
1280     struct vnop_mknod_args {
1281 	struct vnode *a_dvp;
1282 	struct vnode **a_vpp;
1283 	struct componentname *a_cnp;
1284 	struct vattr *a_vap;
1285     };
1286 */
1287 static int
1288 fuse_vnop_mknod(struct vop_mknod_args *ap)
1289 {
1290 
1291 	struct vnode *dvp = ap->a_dvp;
1292 	struct vnode **vpp = ap->a_vpp;
1293 	struct componentname *cnp = ap->a_cnp;
1294 	struct vattr *vap = ap->a_vap;
1295 
1296 	if (fuse_isdeadfs(dvp))
1297 		return ENXIO;
1298 
1299 	return fuse_internal_mknod(dvp, vpp, cnp, vap);
1300 }
1301 
1302 /*
1303     struct vop_open_args {
1304 	struct vnode *a_vp;
1305 	int  a_mode;
1306 	struct ucred *a_cred;
1307 	struct thread *a_td;
1308 	int a_fdidx; / struct file *a_fp;
1309     };
1310 */
1311 static int
1312 fuse_vnop_open(struct vop_open_args *ap)
1313 {
1314 	struct vnode *vp = ap->a_vp;
1315 	int a_mode = ap->a_mode;
1316 	struct thread *td = ap->a_td;
1317 	struct ucred *cred = ap->a_cred;
1318 	pid_t pid = td->td_proc->p_pid;
1319 	struct fuse_vnode_data *fvdat;
1320 
1321 	if (fuse_isdeadfs(vp))
1322 		return ENXIO;
1323 	if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1324 		return (EOPNOTSUPP);
1325 	if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1326 		return EINVAL;
1327 
1328 	fvdat = VTOFUD(vp);
1329 
1330 	if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1331 		fuse_vnode_open(vp, 0, td);
1332 		return 0;
1333 	}
1334 
1335 	return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1336 }
1337 
1338 static int
1339 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1340 {
1341 
1342 	switch (ap->a_name) {
1343 	case _PC_FILESIZEBITS:
1344 		*ap->a_retval = 64;
1345 		return (0);
1346 	case _PC_NAME_MAX:
1347 		*ap->a_retval = NAME_MAX;
1348 		return (0);
1349 	case _PC_LINK_MAX:
1350 		*ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1351 		return (0);
1352 	case _PC_SYMLINK_MAX:
1353 		*ap->a_retval = MAXPATHLEN;
1354 		return (0);
1355 	case _PC_NO_TRUNC:
1356 		*ap->a_retval = 1;
1357 		return (0);
1358 	default:
1359 		return (vop_stdpathconf(ap));
1360 	}
1361 }
1362 
1363 /*
1364     struct vnop_read_args {
1365 	struct vnode *a_vp;
1366 	struct uio *a_uio;
1367 	int  a_ioflag;
1368 	struct ucred *a_cred;
1369     };
1370 */
1371 static int
1372 fuse_vnop_read(struct vop_read_args *ap)
1373 {
1374 	struct vnode *vp = ap->a_vp;
1375 	struct uio *uio = ap->a_uio;
1376 	int ioflag = ap->a_ioflag;
1377 	struct ucred *cred = ap->a_cred;
1378 	pid_t pid = curthread->td_proc->p_pid;
1379 
1380 	if (fuse_isdeadfs(vp)) {
1381 		return ENXIO;
1382 	}
1383 
1384 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1385 		ioflag |= IO_DIRECT;
1386 	}
1387 
1388 	return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1389 }
1390 
1391 /*
1392     struct vnop_readdir_args {
1393 	struct vnode *a_vp;
1394 	struct uio *a_uio;
1395 	struct ucred *a_cred;
1396 	int *a_eofflag;
1397 	int *a_ncookies;
1398 	u_long **a_cookies;
1399     };
1400 */
1401 static int
1402 fuse_vnop_readdir(struct vop_readdir_args *ap)
1403 {
1404 	struct vnode *vp = ap->a_vp;
1405 	struct uio *uio = ap->a_uio;
1406 	struct ucred *cred = ap->a_cred;
1407 	struct fuse_filehandle *fufh = NULL;
1408 	struct fuse_iov cookediov;
1409 	int err = 0;
1410 	u_long *cookies;
1411 	off_t startoff;
1412 	ssize_t tresid;
1413 	int ncookies;
1414 	bool closefufh = false;
1415 	pid_t pid = curthread->td_proc->p_pid;
1416 
1417 	if (ap->a_eofflag)
1418 		*ap->a_eofflag = 0;
1419 	if (fuse_isdeadfs(vp)) {
1420 		return ENXIO;
1421 	}
1422 	if (				/* XXXIP ((uio_iovcnt(uio) > 1)) || */
1423 	    (uio_resid(uio) < sizeof(struct dirent))) {
1424 		return EINVAL;
1425 	}
1426 
1427 	tresid = uio->uio_resid;
1428 	startoff = uio->uio_offset;
1429 	err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1430 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1431 		/*
1432 		 * nfsd will do VOP_READDIR without first doing VOP_OPEN.  We
1433 		 * must implicitly open the directory here
1434 		 */
1435 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1436 		if (err == 0) {
1437 			/*
1438 			 * When a directory is opened, it must be read from
1439 			 * the beginning.  Hopefully, the "startoff" still
1440 			 * exists as an offset cookie for the directory.
1441 			 * If not, it will read the entire directory without
1442 			 * returning any entries and just return eof.
1443 			 */
1444 			uio->uio_offset = 0;
1445 		}
1446 		closefufh = true;
1447 	}
1448 	if (err)
1449 		return (err);
1450 	if (ap->a_ncookies != NULL) {
1451 		ncookies = uio->uio_resid /
1452 			(offsetof(struct dirent, d_name) + 4) + 1;
1453 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1454 		*ap->a_ncookies = ncookies;
1455 		*ap->a_cookies = cookies;
1456 	} else {
1457 		ncookies = 0;
1458 		cookies = NULL;
1459 	}
1460 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1461 	fiov_init(&cookediov, DIRCOOKEDSIZE);
1462 
1463 	err = fuse_internal_readdir(vp, uio, startoff, fufh, &cookediov,
1464 		&ncookies, cookies);
1465 
1466 	fiov_teardown(&cookediov);
1467 	if (closefufh)
1468 		fuse_filehandle_close(vp, fufh, curthread, cred);
1469 
1470 	if (ap->a_ncookies != NULL) {
1471 		if (err == 0) {
1472 			*ap->a_ncookies -= ncookies;
1473 		} else {
1474 			free(*ap->a_cookies, M_TEMP);
1475 			*ap->a_ncookies = 0;
1476 			*ap->a_cookies = NULL;
1477 		}
1478 	}
1479 	if (err == 0 && tresid == uio->uio_resid)
1480 		*ap->a_eofflag = 1;
1481 
1482 	return err;
1483 }
1484 
1485 /*
1486     struct vnop_readlink_args {
1487 	struct vnode *a_vp;
1488 	struct uio *a_uio;
1489 	struct ucred *a_cred;
1490     };
1491 */
1492 static int
1493 fuse_vnop_readlink(struct vop_readlink_args *ap)
1494 {
1495 	struct vnode *vp = ap->a_vp;
1496 	struct uio *uio = ap->a_uio;
1497 	struct ucred *cred = ap->a_cred;
1498 
1499 	struct fuse_dispatcher fdi;
1500 	int err;
1501 
1502 	if (fuse_isdeadfs(vp)) {
1503 		return ENXIO;
1504 	}
1505 	if (!vnode_islnk(vp)) {
1506 		return EINVAL;
1507 	}
1508 	fdisp_init(&fdi, 0);
1509 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1510 	if (err) {
1511 		goto out;
1512 	}
1513 	if (((char *)fdi.answ)[0] == '/' &&
1514 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1515 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1516 
1517 		err = uiomove(mpth, strlen(mpth), uio);
1518 	}
1519 	if (!err) {
1520 		err = uiomove(fdi.answ, fdi.iosize, uio);
1521 	}
1522 out:
1523 	fdisp_destroy(&fdi);
1524 	return err;
1525 }
1526 
1527 /*
1528     struct vnop_reclaim_args {
1529 	struct vnode *a_vp;
1530     };
1531 */
1532 static int
1533 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1534 {
1535 	struct vnode *vp = ap->a_vp;
1536 	struct thread *td = curthread;
1537 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1538 	struct fuse_filehandle *fufh, *fufh_tmp;
1539 
1540 	if (!fvdat) {
1541 		panic("FUSE: no vnode data during recycling");
1542 	}
1543 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1544 		printf("FUSE: vnode being reclaimed with open fufh "
1545 			"(type=%#x)", fufh->fufh_type);
1546 		fuse_filehandle_close(vp, fufh, td, NULL);
1547 	}
1548 
1549 	if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
1550 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1551 		    fvdat->nlookup);
1552 	}
1553 	cache_purge(vp);
1554 	vfs_hash_remove(vp);
1555 	fuse_vnode_destroy(vp);
1556 
1557 	return 0;
1558 }
1559 
1560 /*
1561     struct vnop_remove_args {
1562 	struct vnode *a_dvp;
1563 	struct vnode *a_vp;
1564 	struct componentname *a_cnp;
1565     };
1566 */
1567 static int
1568 fuse_vnop_remove(struct vop_remove_args *ap)
1569 {
1570 	struct vnode *dvp = ap->a_dvp;
1571 	struct vnode *vp = ap->a_vp;
1572 	struct componentname *cnp = ap->a_cnp;
1573 
1574 	int err;
1575 
1576 	if (fuse_isdeadfs(vp)) {
1577 		return ENXIO;
1578 	}
1579 	if (vnode_isdir(vp)) {
1580 		return EPERM;
1581 	}
1582 
1583 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1584 
1585 	return err;
1586 }
1587 
1588 /*
1589     struct vnop_rename_args {
1590 	struct vnode *a_fdvp;
1591 	struct vnode *a_fvp;
1592 	struct componentname *a_fcnp;
1593 	struct vnode *a_tdvp;
1594 	struct vnode *a_tvp;
1595 	struct componentname *a_tcnp;
1596     };
1597 */
1598 static int
1599 fuse_vnop_rename(struct vop_rename_args *ap)
1600 {
1601 	struct vnode *fdvp = ap->a_fdvp;
1602 	struct vnode *fvp = ap->a_fvp;
1603 	struct componentname *fcnp = ap->a_fcnp;
1604 	struct vnode *tdvp = ap->a_tdvp;
1605 	struct vnode *tvp = ap->a_tvp;
1606 	struct componentname *tcnp = ap->a_tcnp;
1607 	struct fuse_data *data;
1608 	bool newparent = fdvp != tdvp;
1609 	bool isdir = fvp->v_type == VDIR;
1610 	int err = 0;
1611 
1612 	if (fuse_isdeadfs(fdvp)) {
1613 		return ENXIO;
1614 	}
1615 	if (fvp->v_mount != tdvp->v_mount ||
1616 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1617 		SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
1618 		err = EXDEV;
1619 		goto out;
1620 	}
1621 	cache_purge(fvp);
1622 
1623 	/*
1624 	 * FUSE library is expected to check if target directory is not
1625 	 * under the source directory in the file system tree.
1626 	 * Linux performs this check at VFS level.
1627 	 */
1628 	/*
1629 	 * If source is a directory, and it will get a new parent, user must
1630 	 * have write permission to it, so ".." can be modified.
1631 	 */
1632 	data = fuse_get_mpdata(vnode_mount(tdvp));
1633 	if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
1634 		err = fuse_internal_access(fvp, VWRITE,
1635 			tcnp->cn_thread, tcnp->cn_cred);
1636 		if (err)
1637 			goto out;
1638 	}
1639 	sx_xlock(&data->rename_lock);
1640 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1641 	if (err == 0) {
1642 		if (tdvp != fdvp)
1643 			fuse_vnode_setparent(fvp, tdvp);
1644 		if (tvp != NULL)
1645 			fuse_vnode_setparent(tvp, NULL);
1646 	}
1647 	sx_unlock(&data->rename_lock);
1648 
1649 	if (tvp != NULL && tvp != fvp) {
1650 		cache_purge(tvp);
1651 	}
1652 	if (vnode_isdir(fvp)) {
1653 		if ((tvp != NULL) && vnode_isdir(tvp)) {
1654 			cache_purge(tdvp);
1655 		}
1656 		cache_purge(fdvp);
1657 	}
1658 out:
1659 	if (tdvp == tvp) {
1660 		vrele(tdvp);
1661 	} else {
1662 		vput(tdvp);
1663 	}
1664 	if (tvp != NULL) {
1665 		vput(tvp);
1666 	}
1667 	vrele(fdvp);
1668 	vrele(fvp);
1669 
1670 	return err;
1671 }
1672 
1673 /*
1674     struct vnop_rmdir_args {
1675 	    struct vnode *a_dvp;
1676 	    struct vnode *a_vp;
1677 	    struct componentname *a_cnp;
1678     } *ap;
1679 */
1680 static int
1681 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1682 {
1683 	struct vnode *dvp = ap->a_dvp;
1684 	struct vnode *vp = ap->a_vp;
1685 
1686 	int err;
1687 
1688 	if (fuse_isdeadfs(vp)) {
1689 		return ENXIO;
1690 	}
1691 	if (VTOFUD(vp) == VTOFUD(dvp)) {
1692 		return EINVAL;
1693 	}
1694 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1695 
1696 	return err;
1697 }
1698 
1699 /*
1700     struct vnop_setattr_args {
1701 	struct vnode *a_vp;
1702 	struct vattr *a_vap;
1703 	struct ucred *a_cred;
1704 	struct thread *a_td;
1705     };
1706 */
1707 static int
1708 fuse_vnop_setattr(struct vop_setattr_args *ap)
1709 {
1710 	struct vnode *vp = ap->a_vp;
1711 	struct vattr *vap = ap->a_vap;
1712 	struct ucred *cred = ap->a_cred;
1713 	struct thread *td = curthread;
1714 	struct mount *mp;
1715 	struct fuse_data *data;
1716 	struct vattr old_va;
1717 	int dataflags;
1718 	int err = 0, err2;
1719 	accmode_t accmode = 0;
1720 	bool checkperm;
1721 	bool drop_suid = false;
1722 	gid_t cr_gid;
1723 
1724 	mp = vnode_mount(vp);
1725 	data = fuse_get_mpdata(mp);
1726 	dataflags = data->dataflags;
1727 	checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
1728 	if (cred->cr_ngroups > 0)
1729 		cr_gid = cred->cr_groups[0];
1730 	else
1731 		cr_gid = 0;
1732 
1733 	if (fuse_isdeadfs(vp)) {
1734 		return ENXIO;
1735 	}
1736 
1737 	if (vap->va_uid != (uid_t)VNOVAL) {
1738 		if (checkperm) {
1739 			/* Only root may change a file's owner */
1740 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1741 			if (err) {
1742 				/* As a special case, allow the null chown */
1743 				err2 = fuse_internal_getattr(vp, &old_va, cred,
1744 					td);
1745 				if (err2)
1746 					return (err2);
1747 				if (vap->va_uid != old_va.va_uid)
1748 					return err;
1749 				else
1750 					accmode |= VADMIN;
1751 				drop_suid = true;
1752 			} else
1753 				accmode |= VADMIN;
1754 		} else
1755 			accmode |= VADMIN;
1756 	}
1757 	if (vap->va_gid != (gid_t)VNOVAL) {
1758 		if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
1759 			drop_suid = true;
1760 		if (checkperm && !groupmember(vap->va_gid, cred))
1761 		{
1762 			/*
1763 			 * Non-root users may only chgrp to one of their own
1764 			 * groups
1765 			 */
1766 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1767 			if (err) {
1768 				/* As a special case, allow the null chgrp */
1769 				err2 = fuse_internal_getattr(vp, &old_va, cred,
1770 					td);
1771 				if (err2)
1772 					return (err2);
1773 				if (vap->va_gid != old_va.va_gid)
1774 					return err;
1775 				accmode |= VADMIN;
1776 			} else
1777 				accmode |= VADMIN;
1778 		} else
1779 			accmode |= VADMIN;
1780 	}
1781 	if (vap->va_size != VNOVAL) {
1782 		switch (vp->v_type) {
1783 		case VDIR:
1784 			return (EISDIR);
1785 		case VLNK:
1786 		case VREG:
1787 			if (vfs_isrdonly(mp))
1788 				return (EROFS);
1789 			break;
1790 		default:
1791 			/*
1792 			 * According to POSIX, the result is unspecified
1793 			 * for file types other than regular files,
1794 			 * directories and shared memory objects.  We
1795 			 * don't support shared memory objects in the file
1796 			 * system, and have dubious support for truncating
1797 			 * symlinks.  Just ignore the request in other cases.
1798 			 */
1799 			return (0);
1800 		}
1801 		/* Don't set accmode.  Permission to trunc is checked upstack */
1802 	}
1803 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1804 		if (vap->va_vaflags & VA_UTIMES_NULL)
1805 			accmode |= VWRITE;
1806 		else
1807 			accmode |= VADMIN;
1808 	}
1809 	if (drop_suid) {
1810 		if (vap->va_mode != (mode_t)VNOVAL)
1811 			vap->va_mode &= ~(S_ISUID | S_ISGID);
1812 		else {
1813 			err = fuse_internal_getattr(vp, &old_va, cred, td);
1814 			if (err)
1815 				return (err);
1816 			vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
1817 		}
1818 	}
1819 	if (vap->va_mode != (mode_t)VNOVAL) {
1820 		/* Only root may set the sticky bit on non-directories */
1821 		if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
1822 		    && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1823 			return EFTYPE;
1824 		if (checkperm && (vap->va_mode & S_ISGID)) {
1825 			err = fuse_internal_getattr(vp, &old_va, cred, td);
1826 			if (err)
1827 				return (err);
1828 			if (!groupmember(old_va.va_gid, cred)) {
1829 				err = priv_check_cred(cred, PRIV_VFS_SETGID);
1830 				if (err)
1831 					return (err);
1832 			}
1833 		}
1834 		accmode |= VADMIN;
1835 	}
1836 
1837 	if (vfs_isrdonly(mp))
1838 		return EROFS;
1839 
1840 	if (checkperm) {
1841 		err = fuse_internal_access(vp, accmode, td, cred);
1842 	} else {
1843 		err = 0;
1844 	}
1845 	if (err)
1846 		return err;
1847 	else
1848 		return fuse_internal_setattr(vp, vap, td, cred);
1849 }
1850 
1851 /*
1852     struct vnop_strategy_args {
1853 	struct vnode *a_vp;
1854 	struct buf *a_bp;
1855     };
1856 */
1857 static int
1858 fuse_vnop_strategy(struct vop_strategy_args *ap)
1859 {
1860 	struct vnode *vp = ap->a_vp;
1861 	struct buf *bp = ap->a_bp;
1862 
1863 	if (!vp || fuse_isdeadfs(vp)) {
1864 		bp->b_ioflags |= BIO_ERROR;
1865 		bp->b_error = ENXIO;
1866 		bufdone(bp);
1867 		return 0;
1868 	}
1869 
1870 	/*
1871 	 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
1872 	 * fuse_io_strategy sets bp's error fields
1873 	 */
1874 	(void)fuse_io_strategy(vp, bp);
1875 
1876 	return 0;
1877 }
1878 
1879 /*
1880     struct vnop_symlink_args {
1881 	struct vnode *a_dvp;
1882 	struct vnode **a_vpp;
1883 	struct componentname *a_cnp;
1884 	struct vattr *a_vap;
1885 	char *a_target;
1886     };
1887 */
1888 static int
1889 fuse_vnop_symlink(struct vop_symlink_args *ap)
1890 {
1891 	struct vnode *dvp = ap->a_dvp;
1892 	struct vnode **vpp = ap->a_vpp;
1893 	struct componentname *cnp = ap->a_cnp;
1894 	const char *target = ap->a_target;
1895 
1896 	struct fuse_dispatcher fdi;
1897 
1898 	int err;
1899 	size_t len;
1900 
1901 	if (fuse_isdeadfs(dvp)) {
1902 		return ENXIO;
1903 	}
1904 	/*
1905 	 * Unlike the other creator type calls, here we have to create a message
1906 	 * where the name of the new entry comes first, and the data describing
1907 	 * the entry comes second.
1908 	 * Hence we can't rely on our handy fuse_internal_newentry() routine,
1909 	 * but put together the message manually and just call the core part.
1910 	 */
1911 
1912 	len = strlen(target) + 1;
1913 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1914 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1915 
1916 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1917 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1918 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1919 
1920 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1921 	fdisp_destroy(&fdi);
1922 	return err;
1923 }
1924 
1925 /*
1926     struct vnop_write_args {
1927 	struct vnode *a_vp;
1928 	struct uio *a_uio;
1929 	int  a_ioflag;
1930 	struct ucred *a_cred;
1931     };
1932 */
1933 static int
1934 fuse_vnop_write(struct vop_write_args *ap)
1935 {
1936 	struct vnode *vp = ap->a_vp;
1937 	struct uio *uio = ap->a_uio;
1938 	int ioflag = ap->a_ioflag;
1939 	struct ucred *cred = ap->a_cred;
1940 	pid_t pid = curthread->td_proc->p_pid;
1941 
1942 	if (fuse_isdeadfs(vp)) {
1943 		return ENXIO;
1944 	}
1945 
1946 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1947 		ioflag |= IO_DIRECT;
1948 	}
1949 
1950 	return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1951 }
1952 
1953 static daddr_t
1954 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1955 {
1956 	const int biosize = fuse_iosize(vp);
1957 
1958 	return (off / biosize);
1959 }
1960 
1961 static int
1962 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1963 {
1964 	off_t filesize;
1965 	int blksz, err;
1966 	const int biosize = fuse_iosize(vp);
1967 
1968 	err = fuse_vnode_size(vp, &filesize, NULL, NULL);
1969 	KASSERT(err == 0, ("vfs_bio_getpages can't handle errors here"));
1970 	if (err)
1971 		return biosize;
1972 
1973 	if ((off_t)lbn * biosize >= filesize) {
1974 		blksz = 0;
1975 	} else if ((off_t)(lbn + 1) * biosize > filesize) {
1976 		blksz = filesize - (off_t)lbn *biosize;
1977 	} else {
1978 		blksz = biosize;
1979 	}
1980 	return (blksz);
1981 }
1982 
1983 /*
1984     struct vnop_getpages_args {
1985 	struct vnode *a_vp;
1986 	vm_page_t *a_m;
1987 	int a_count;
1988 	int a_reqpage;
1989     };
1990 */
1991 static int
1992 fuse_vnop_getpages(struct vop_getpages_args *ap)
1993 {
1994 	struct vnode *vp = ap->a_vp;
1995 
1996 	if (!fsess_opt_mmap(vnode_mount(vp))) {
1997 		SDT_PROBE2(fusefs, , vnops, trace, 1,
1998 			"called on non-cacheable vnode??\n");
1999 		return (VM_PAGER_ERROR);
2000 	}
2001 
2002 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
2003 	    ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
2004 }
2005 
2006 static const char extattr_namespace_separator = '.';
2007 
2008 /*
2009     struct vop_getextattr_args {
2010 	struct vop_generic_args a_gen;
2011 	struct vnode *a_vp;
2012 	int a_attrnamespace;
2013 	const char *a_name;
2014 	struct uio *a_uio;
2015 	size_t *a_size;
2016 	struct ucred *a_cred;
2017 	struct thread *a_td;
2018     };
2019 */
2020 static int
2021 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2022 {
2023 	struct vnode *vp = ap->a_vp;
2024 	struct uio *uio = ap->a_uio;
2025 	struct fuse_dispatcher fdi;
2026 	struct fuse_getxattr_in *get_xattr_in;
2027 	struct fuse_getxattr_out *get_xattr_out;
2028 	struct mount *mp = vnode_mount(vp);
2029 	struct thread *td = ap->a_td;
2030 	struct ucred *cred = ap->a_cred;
2031 	char *prefix;
2032 	char *attr_str;
2033 	size_t len;
2034 	int err;
2035 
2036 	if (fuse_isdeadfs(vp))
2037 		return (ENXIO);
2038 
2039 	if (!fsess_isimpl(mp, FUSE_GETXATTR))
2040 		return EOPNOTSUPP;
2041 
2042 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2043 	if (err)
2044 		return err;
2045 
2046 	/* Default to looking for user attributes. */
2047 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2048 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2049 	else
2050 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2051 
2052 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2053 	    strlen(ap->a_name) + 1;
2054 
2055 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2056 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2057 
2058 	get_xattr_in = fdi.indata;
2059 	/*
2060 	 * Check to see whether we're querying the available size or
2061 	 * issuing the actual request.  If we pass in 0, we get back struct
2062 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2063 	 * that much data, without the struct fuse_getxattr_out header.
2064 	 */
2065 	if (uio == NULL)
2066 		get_xattr_in->size = 0;
2067 	else
2068 		get_xattr_in->size = uio->uio_resid;
2069 
2070 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2071 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2072 	    ap->a_name);
2073 
2074 	err = fdisp_wait_answ(&fdi);
2075 	if (err != 0) {
2076 		if (err == ENOSYS) {
2077 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2078 			err = EOPNOTSUPP;
2079 		}
2080 		goto out;
2081 	}
2082 
2083 	get_xattr_out = fdi.answ;
2084 
2085 	if (ap->a_size != NULL)
2086 		*ap->a_size = get_xattr_out->size;
2087 
2088 	if (uio != NULL)
2089 		err = uiomove(fdi.answ, fdi.iosize, uio);
2090 
2091 out:
2092 	fdisp_destroy(&fdi);
2093 	return (err);
2094 }
2095 
2096 /*
2097     struct vop_setextattr_args {
2098 	struct vop_generic_args a_gen;
2099 	struct vnode *a_vp;
2100 	int a_attrnamespace;
2101 	const char *a_name;
2102 	struct uio *a_uio;
2103 	struct ucred *a_cred;
2104 	struct thread *a_td;
2105     };
2106 */
2107 static int
2108 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2109 {
2110 	struct vnode *vp = ap->a_vp;
2111 	struct uio *uio = ap->a_uio;
2112 	struct fuse_dispatcher fdi;
2113 	struct fuse_setxattr_in *set_xattr_in;
2114 	struct mount *mp = vnode_mount(vp);
2115 	struct thread *td = ap->a_td;
2116 	struct ucred *cred = ap->a_cred;
2117 	char *prefix;
2118 	size_t len;
2119 	char *attr_str;
2120 	int err;
2121 
2122 	if (fuse_isdeadfs(vp))
2123 		return (ENXIO);
2124 
2125 	if (!fsess_isimpl(mp, FUSE_SETXATTR))
2126 		return EOPNOTSUPP;
2127 
2128 	if (vfs_isrdonly(mp))
2129 		return EROFS;
2130 
2131 	/* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2132 	if (ap->a_uio == NULL) {
2133 		/*
2134 		 * If we got here as fallback from VOP_DELETEEXTATTR, then
2135 		 * return EOPNOTSUPP.
2136 		 */
2137 		if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2138 			return (EOPNOTSUPP);
2139 		else
2140 			return (EINVAL);
2141 	}
2142 
2143 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2144 		VWRITE);
2145 	if (err)
2146 		return err;
2147 
2148 	/* Default to looking for user attributes. */
2149 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2150 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2151 	else
2152 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2153 
2154 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2155 	    strlen(ap->a_name) + 1;
2156 
2157 	fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2158 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2159 
2160 	set_xattr_in = fdi.indata;
2161 	set_xattr_in->size = uio->uio_resid;
2162 
2163 	attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2164 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2165 	    ap->a_name);
2166 
2167 	err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2168 	    uio->uio_resid, uio);
2169 	if (err != 0) {
2170 		goto out;
2171 	}
2172 
2173 	err = fdisp_wait_answ(&fdi);
2174 
2175 	if (err == ENOSYS) {
2176 		fsess_set_notimpl(mp, FUSE_SETXATTR);
2177 		err = EOPNOTSUPP;
2178 	}
2179 	if (err == ERESTART) {
2180 		/* Can't restart after calling uiomove */
2181 		err = EINTR;
2182 	}
2183 
2184 out:
2185 	fdisp_destroy(&fdi);
2186 	return (err);
2187 }
2188 
2189 /*
2190  * The Linux / FUSE extended attribute list is simply a collection of
2191  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2192  * byte length followed by a non-NUL terminated string.  So, this allows
2193  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2194  * Linux attribute names are reported with the namespace as a prefix (e.g.
2195  * "user.attribute_name"), but in FreeBSD they are reported without the
2196  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2197  *
2198  * user.attr_name1\0user.attr_name2\0
2199  *
2200  * to:
2201  *
2202  * <num>attr_name1<num>attr_name2
2203  *
2204  * Where "<num>" is a single byte number of characters in the attribute name.
2205  *
2206  * Args:
2207  * prefix - exattr namespace prefix string
2208  * list, list_len - input list with namespace prefixes
2209  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2210  */
2211 static int
2212 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2213     char *bsd_list, int *bsd_list_len)
2214 {
2215 	int len, pos, dist_to_next, prefix_len;
2216 
2217 	pos = 0;
2218 	*bsd_list_len = 0;
2219 	prefix_len = strlen(prefix);
2220 
2221 	while (pos < list_len && list[pos] != '\0') {
2222 		dist_to_next = strlen(&list[pos]) + 1;
2223 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2224 		    list[pos + prefix_len] == extattr_namespace_separator) {
2225 			len = dist_to_next -
2226 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2227 			if (len >= EXTATTR_MAXNAMELEN)
2228 				return (ENAMETOOLONG);
2229 
2230 			bsd_list[*bsd_list_len] = len;
2231 			memcpy(&bsd_list[*bsd_list_len + 1],
2232 			    &list[pos + prefix_len +
2233 			    sizeof(extattr_namespace_separator)], len);
2234 
2235 			*bsd_list_len += len + 1;
2236 		}
2237 
2238 		pos += dist_to_next;
2239 	}
2240 
2241 	return (0);
2242 }
2243 
2244 /*
2245  * List extended attributes
2246  *
2247  * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
2248  * has a number of differences compared to its FreeBSD equivalent,
2249  * extattr_list_file:
2250  *
2251  * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
2252  *   whereas listxattr(2) only returns attributes for a single namespace
2253  * - FUSE_LISTXATTR prepends each attribute name with "namespace."
2254  * - If the provided buffer is not large enough to hold the result,
2255  *   FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
2256  *   return as many results as will fit.
2257  */
2258 /*
2259     struct vop_listextattr_args {
2260 	struct vop_generic_args a_gen;
2261 	struct vnode *a_vp;
2262 	int a_attrnamespace;
2263 	struct uio *a_uio;
2264 	size_t *a_size;
2265 	struct ucred *a_cred;
2266 	struct thread *a_td;
2267     };
2268 */
2269 static int
2270 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2271 {
2272 	struct vnode *vp = ap->a_vp;
2273 	struct uio *uio = ap->a_uio;
2274 	struct fuse_dispatcher fdi;
2275 	struct fuse_listxattr_in *list_xattr_in;
2276 	struct fuse_listxattr_out *list_xattr_out;
2277 	struct mount *mp = vnode_mount(vp);
2278 	struct thread *td = ap->a_td;
2279 	struct ucred *cred = ap->a_cred;
2280 	char *prefix;
2281 	char *bsd_list = NULL;
2282 	char *linux_list;
2283 	int bsd_list_len;
2284 	int linux_list_len;
2285 	int err;
2286 
2287 	if (fuse_isdeadfs(vp))
2288 		return (ENXIO);
2289 
2290 	if (!fsess_isimpl(mp, FUSE_LISTXATTR))
2291 		return EOPNOTSUPP;
2292 
2293 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2294 	if (err)
2295 		return err;
2296 
2297 	/*
2298 	 * Add space for a NUL and the period separator if enabled.
2299 	 * Default to looking for user attributes.
2300 	 */
2301 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2302 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2303 	else
2304 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2305 
2306 	fdisp_init(&fdi, sizeof(*list_xattr_in));
2307 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2308 
2309 	/*
2310 	 * Retrieve Linux / FUSE compatible list size.
2311 	 */
2312 	list_xattr_in = fdi.indata;
2313 	list_xattr_in->size = 0;
2314 
2315 	err = fdisp_wait_answ(&fdi);
2316 	if (err != 0) {
2317 		if (err == ENOSYS) {
2318 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
2319 			err = EOPNOTSUPP;
2320 		}
2321 		goto out;
2322 	}
2323 
2324 	list_xattr_out = fdi.answ;
2325 	linux_list_len = list_xattr_out->size;
2326 	if (linux_list_len == 0) {
2327 		if (ap->a_size != NULL)
2328 			*ap->a_size = linux_list_len;
2329 		goto out;
2330 	}
2331 
2332 	/*
2333 	 * Retrieve Linux / FUSE compatible list values.
2334 	 */
2335 	fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2336 	list_xattr_in = fdi.indata;
2337 	list_xattr_in->size = linux_list_len;
2338 
2339 	err = fdisp_wait_answ(&fdi);
2340 	if (err == ERANGE) {
2341 		/*
2342 		 * Race detected.  The attribute list must've grown since the
2343 		 * first FUSE_LISTXATTR call.  Start over.  Go all the way back
2344 		 * to userland so we can process signals, if necessary, before
2345 		 * restarting.
2346 		 */
2347 		err = ERESTART;
2348 		goto out;
2349 	} else if (err != 0)
2350 		goto out;
2351 
2352 	linux_list = fdi.answ;
2353 	/* FUSE doesn't allow the server to return more data than requested */
2354 	if (fdi.iosize > linux_list_len) {
2355 		printf("WARNING: FUSE protocol violation.  Server returned "
2356 			"more extended attribute data than requested; "
2357 			"should've returned ERANGE instead");
2358 	} else {
2359 		/* But returning less data is fine */
2360 		linux_list_len = fdi.iosize;
2361 	}
2362 
2363 	/*
2364 	 * Retrieve the BSD compatible list values.
2365 	 * The Linux / FUSE attribute list format isn't the same
2366 	 * as FreeBSD's format. So we need to transform it into
2367 	 * FreeBSD's format before giving it to the user.
2368 	 */
2369 	bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2370 	err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2371 	    bsd_list, &bsd_list_len);
2372 	if (err != 0)
2373 		goto out;
2374 
2375 	if (ap->a_size != NULL)
2376 		*ap->a_size = bsd_list_len;
2377 
2378 	if (uio != NULL)
2379 		err = uiomove(bsd_list, bsd_list_len, uio);
2380 
2381 out:
2382 	free(bsd_list, M_TEMP);
2383 	fdisp_destroy(&fdi);
2384 	return (err);
2385 }
2386 
2387 /*
2388     struct vop_deleteextattr_args {
2389 	struct vop_generic_args a_gen;
2390 	struct vnode *a_vp;
2391 	int a_attrnamespace;
2392 	const char *a_name;
2393 	struct ucred *a_cred;
2394 	struct thread *a_td;
2395     };
2396 */
2397 static int
2398 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2399 {
2400 	struct vnode *vp = ap->a_vp;
2401 	struct fuse_dispatcher fdi;
2402 	struct mount *mp = vnode_mount(vp);
2403 	struct thread *td = ap->a_td;
2404 	struct ucred *cred = ap->a_cred;
2405 	char *prefix;
2406 	size_t len;
2407 	char *attr_str;
2408 	int err;
2409 
2410 	if (fuse_isdeadfs(vp))
2411 		return (ENXIO);
2412 
2413 	if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2414 		return EOPNOTSUPP;
2415 
2416 	if (vfs_isrdonly(mp))
2417 		return EROFS;
2418 
2419 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2420 		VWRITE);
2421 	if (err)
2422 		return err;
2423 
2424 	/* Default to looking for user attributes. */
2425 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2426 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2427 	else
2428 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2429 
2430 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2431 	    strlen(ap->a_name) + 1;
2432 
2433 	fdisp_init(&fdi, len);
2434 	fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2435 
2436 	attr_str = fdi.indata;
2437 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2438 	    ap->a_name);
2439 
2440 	err = fdisp_wait_answ(&fdi);
2441 	if (err == ENOSYS) {
2442 		fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2443 		err = EOPNOTSUPP;
2444 	}
2445 
2446 	fdisp_destroy(&fdi);
2447 	return (err);
2448 }
2449 
2450 /*
2451     struct vnop_print_args {
2452 	struct vnode *a_vp;
2453     };
2454 */
2455 static int
2456 fuse_vnop_print(struct vop_print_args *ap)
2457 {
2458 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2459 
2460 	printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2461 	    (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2462 	    (uintmax_t)fvdat->nlookup,
2463 	    fvdat->flag);
2464 
2465 	return 0;
2466 }
2467 
2468 /*
2469  * Get an NFS filehandle for a FUSE file.
2470  *
2471  * This will only work for FUSE file systems that guarantee the uniqueness of
2472  * nodeid:generation, which most don't.
2473  */
2474 /*
2475 vop_vptofh {
2476 	IN struct vnode *a_vp;
2477 	IN struct fid *a_fhp;
2478 };
2479 */
2480 static int
2481 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
2482 {
2483 	struct vnode *vp = ap->a_vp;
2484 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
2485 	struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
2486 	_Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
2487 		"FUSE fid type is too big");
2488 	struct mount *mp = vnode_mount(vp);
2489 	struct fuse_data *data = fuse_get_mpdata(mp);
2490 	struct vattr va;
2491 	int err;
2492 
2493 	if (!(data->dataflags & FSESS_EXPORT_SUPPORT))
2494 		return EOPNOTSUPP;
2495 
2496 	err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
2497 	if (err)
2498 		return err;
2499 
2500 	/*ip = VTOI(ap->a_vp);*/
2501 	/*ufhp = (struct ufid *)ap->a_fhp;*/
2502 	fhp->len = sizeof(struct fuse_fid);
2503 	fhp->nid = fvdat->nid;
2504 	if (fvdat->generation <= UINT32_MAX)
2505 		fhp->gen = fvdat->generation;
2506 	else
2507 		return EOVERFLOW;
2508 	return (0);
2509 }
2510