xref: /freebsd/sys/fs/fuse/fuse_vnops.c (revision bd001d86d679e10d179ef00b9866f0e65b6fa7fd)
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/filio.h>
73 #include <sys/uio.h>
74 #include <sys/malloc.h>
75 #include <sys/queue.h>
76 #include <sys/limits.h>
77 #include <sys/lock.h>
78 #include <sys/rwlock.h>
79 #include <sys/sx.h>
80 #include <sys/proc.h>
81 #include <sys/mount.h>
82 #include <sys/vnode.h>
83 #include <sys/namei.h>
84 #include <sys/extattr.h>
85 #include <sys/stat.h>
86 #include <sys/unistd.h>
87 #include <sys/filedesc.h>
88 #include <sys/file.h>
89 #include <sys/fcntl.h>
90 #include <sys/dirent.h>
91 #include <sys/bio.h>
92 #include <sys/buf.h>
93 #include <sys/sysctl.h>
94 #include <sys/vmmeter.h>
95 
96 #include <vm/vm.h>
97 #include <vm/vm_extern.h>
98 #include <vm/pmap.h>
99 #include <vm/vm_map.h>
100 #include <vm/vm_page.h>
101 #include <vm/vm_param.h>
102 #include <vm/vm_object.h>
103 #include <vm/vm_pager.h>
104 #include <vm/vnode_pager.h>
105 #include <vm/vm_object.h>
106 
107 #include "fuse.h"
108 #include "fuse_file.h"
109 #include "fuse_internal.h"
110 #include "fuse_ipc.h"
111 #include "fuse_node.h"
112 #include "fuse_io.h"
113 
114 #include <sys/priv.h>
115 
116 /* Maximum number of hardlinks to a single FUSE file */
117 #define FUSE_LINK_MAX                      UINT32_MAX
118 
119 SDT_PROVIDER_DECLARE(fusefs);
120 /*
121  * Fuse trace probe:
122  * arg0: verbosity.  Higher numbers give more verbose messages
123  * arg1: Textual message
124  */
125 SDT_PROBE_DEFINE2(fusefs, , vnops, trace, "int", "char*");
126 
127 /* vnode ops */
128 static vop_access_t fuse_vnop_access;
129 static vop_advlock_t fuse_vnop_advlock;
130 static vop_allocate_t fuse_vnop_allocate;
131 static vop_bmap_t fuse_vnop_bmap;
132 static vop_close_t fuse_fifo_close;
133 static vop_close_t fuse_vnop_close;
134 static vop_copy_file_range_t fuse_vnop_copy_file_range;
135 static vop_create_t fuse_vnop_create;
136 static vop_deallocate_t fuse_vnop_deallocate;
137 static vop_deleteextattr_t fuse_vnop_deleteextattr;
138 static vop_fdatasync_t fuse_vnop_fdatasync;
139 static vop_fsync_t fuse_vnop_fsync;
140 static vop_getattr_t fuse_vnop_getattr;
141 static vop_getextattr_t fuse_vnop_getextattr;
142 static vop_inactive_t fuse_vnop_inactive;
143 static vop_ioctl_t fuse_vnop_ioctl;
144 static vop_link_t fuse_vnop_link;
145 static vop_listextattr_t fuse_vnop_listextattr;
146 static vop_lookup_t fuse_vnop_lookup;
147 static vop_mkdir_t fuse_vnop_mkdir;
148 static vop_mknod_t fuse_vnop_mknod;
149 static vop_open_t fuse_vnop_open;
150 static vop_pathconf_t fuse_vnop_pathconf;
151 static vop_read_t fuse_vnop_read;
152 static vop_readdir_t fuse_vnop_readdir;
153 static vop_readlink_t fuse_vnop_readlink;
154 static vop_reclaim_t fuse_vnop_reclaim;
155 static vop_remove_t fuse_vnop_remove;
156 static vop_rename_t fuse_vnop_rename;
157 static vop_rmdir_t fuse_vnop_rmdir;
158 static vop_setattr_t fuse_vnop_setattr;
159 static vop_setextattr_t fuse_vnop_setextattr;
160 static vop_strategy_t fuse_vnop_strategy;
161 static vop_symlink_t fuse_vnop_symlink;
162 static vop_write_t fuse_vnop_write;
163 static vop_getpages_t fuse_vnop_getpages;
164 static vop_print_t fuse_vnop_print;
165 static vop_vptofh_t fuse_vnop_vptofh;
166 
167 struct vop_vector fuse_fifoops = {
168 	.vop_default =		&fifo_specops,
169 	.vop_access =		fuse_vnop_access,
170 	.vop_close =		fuse_fifo_close,
171 	.vop_fsync =		fuse_vnop_fsync,
172 	.vop_getattr =		fuse_vnop_getattr,
173 	.vop_inactive =		fuse_vnop_inactive,
174 	.vop_pathconf =		fuse_vnop_pathconf,
175 	.vop_print =		fuse_vnop_print,
176 	.vop_read =		VOP_PANIC,
177 	.vop_reclaim =		fuse_vnop_reclaim,
178 	.vop_setattr =		fuse_vnop_setattr,
179 	.vop_write =		VOP_PANIC,
180 	.vop_vptofh =		fuse_vnop_vptofh,
181 };
182 VFS_VOP_VECTOR_REGISTER(fuse_fifoops);
183 
184 struct vop_vector fuse_vnops = {
185 	.vop_allocate =	fuse_vnop_allocate,
186 	.vop_default = &default_vnodeops,
187 	.vop_access = fuse_vnop_access,
188 	.vop_advlock = fuse_vnop_advlock,
189 	.vop_bmap = fuse_vnop_bmap,
190 	.vop_close = fuse_vnop_close,
191 	.vop_copy_file_range = fuse_vnop_copy_file_range,
192 	.vop_create = fuse_vnop_create,
193 	.vop_deallocate = fuse_vnop_deallocate,
194 	.vop_deleteextattr = fuse_vnop_deleteextattr,
195 	.vop_fsync = fuse_vnop_fsync,
196 	.vop_fdatasync = fuse_vnop_fdatasync,
197 	.vop_getattr = fuse_vnop_getattr,
198 	.vop_getextattr = fuse_vnop_getextattr,
199 	.vop_inactive = fuse_vnop_inactive,
200 	.vop_ioctl = fuse_vnop_ioctl,
201 	.vop_link = fuse_vnop_link,
202 	.vop_listextattr = fuse_vnop_listextattr,
203 	.vop_lookup = fuse_vnop_lookup,
204 	.vop_mkdir = fuse_vnop_mkdir,
205 	.vop_mknod = fuse_vnop_mknod,
206 	.vop_open = fuse_vnop_open,
207 	.vop_pathconf = fuse_vnop_pathconf,
208 	/*
209 	 * TODO: implement vop_poll after upgrading to protocol 7.21.
210 	 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
211 	 * 7.21, which adds the ability for the client to choose which poll
212 	 * events it wants, and for a client to deregister a file handle
213 	 */
214 	.vop_read = fuse_vnop_read,
215 	.vop_readdir = fuse_vnop_readdir,
216 	.vop_readlink = fuse_vnop_readlink,
217 	.vop_reclaim = fuse_vnop_reclaim,
218 	.vop_remove = fuse_vnop_remove,
219 	.vop_rename = fuse_vnop_rename,
220 	.vop_rmdir = fuse_vnop_rmdir,
221 	.vop_setattr = fuse_vnop_setattr,
222 	.vop_setextattr = fuse_vnop_setextattr,
223 	.vop_strategy = fuse_vnop_strategy,
224 	.vop_symlink = fuse_vnop_symlink,
225 	.vop_write = fuse_vnop_write,
226 	.vop_getpages = fuse_vnop_getpages,
227 	.vop_print = fuse_vnop_print,
228 	.vop_vptofh = fuse_vnop_vptofh,
229 };
230 VFS_VOP_VECTOR_REGISTER(fuse_vnops);
231 
232 uma_zone_t fuse_pbuf_zone;
233 
234 /* Check permission for extattr operations, much like extattr_check_cred */
235 static int
236 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
237 	struct thread *td, accmode_t accmode)
238 {
239 	struct mount *mp = vnode_mount(vp);
240 	struct fuse_data *data = fuse_get_mpdata(mp);
241 	int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
242 
243 	/*
244 	 * Kernel-invoked always succeeds.
245 	 */
246 	if (cred == NOCRED)
247 		return (0);
248 
249 	/*
250 	 * Do not allow privileged processes in jail to directly manipulate
251 	 * system attributes.
252 	 */
253 	switch (ns) {
254 	case EXTATTR_NAMESPACE_SYSTEM:
255 		if (default_permissions) {
256 			return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
257 		}
258 		return (0);
259 	case EXTATTR_NAMESPACE_USER:
260 		if (default_permissions) {
261 			return (fuse_internal_access(vp, accmode, td, cred));
262 		}
263 		return (0);
264 	default:
265 		return (EPERM);
266 	}
267 }
268 
269 /* Get a filehandle for a directory */
270 static int
271 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
272 	struct ucred *cred, pid_t pid)
273 {
274 	if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
275 		return 0;
276 	return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
277 }
278 
279 /* Send FUSE_FLUSH for this vnode */
280 static int
281 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
282 {
283 	struct fuse_flush_in *ffi;
284 	struct fuse_filehandle *fufh;
285 	struct fuse_dispatcher fdi;
286 	struct thread *td = curthread;
287 	struct mount *mp = vnode_mount(vp);
288 	int err;
289 
290 	if (fsess_not_impl(vnode_mount(vp), FUSE_FLUSH))
291 		return 0;
292 
293 	err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
294 	if (err)
295 		return err;
296 
297 	fdisp_init(&fdi, sizeof(*ffi));
298 	fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
299 	ffi = fdi.indata;
300 	ffi->fh = fufh->fh_id;
301 	/*
302 	 * If the file has a POSIX lock then we're supposed to set lock_owner.
303 	 * If not, then lock_owner is undefined.  So we may as well always set
304 	 * it.
305 	 */
306 	ffi->lock_owner = td->td_proc->p_pid;
307 
308 	err = fdisp_wait_answ(&fdi);
309 	if (err == ENOSYS) {
310 		fsess_set_notimpl(mp, FUSE_FLUSH);
311 		err = 0;
312 	}
313 	fdisp_destroy(&fdi);
314 	return err;
315 }
316 
317 /* Close wrapper for fifos.  */
318 static int
319 fuse_fifo_close(struct vop_close_args *ap)
320 {
321 	return (fifo_specops.vop_close(ap));
322 }
323 
324 /* Invalidate a range of cached data, whether dirty of not */
325 static int
326 fuse_inval_buf_range(struct vnode *vp, off_t filesize, off_t start, off_t end)
327 {
328 	struct buf *bp;
329 	daddr_t left_lbn, end_lbn, right_lbn;
330 	off_t new_filesize;
331 	int iosize, left_on, right_on, right_blksize;
332 
333 	iosize = fuse_iosize(vp);
334 	left_lbn = start / iosize;
335 	end_lbn = howmany(end, iosize);
336 	left_on = start & (iosize - 1);
337 	if (left_on != 0) {
338 		bp = getblk(vp, left_lbn, iosize, PCATCH, 0, 0);
339 		if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyend >= left_on) {
340 			/*
341 			 * Flush the dirty buffer, because we don't have a
342 			 * byte-granular way to record which parts of the
343 			 * buffer are valid.
344 			 */
345 			bwrite(bp);
346 			if (bp->b_error)
347 				return (bp->b_error);
348 		} else {
349 			brelse(bp);
350 		}
351 	}
352 	right_on = end & (iosize - 1);
353 	if (right_on != 0) {
354 		right_lbn = end / iosize;
355 		new_filesize = MAX(filesize, end);
356 		right_blksize = MIN(iosize, new_filesize - iosize * right_lbn);
357 		bp = getblk(vp, right_lbn, right_blksize, PCATCH, 0, 0);
358 		if ((bp->b_flags & B_CACHE) != 0 && bp->b_dirtyoff < right_on) {
359 			/*
360 			 * Flush the dirty buffer, because we don't have a
361 			 * byte-granular way to record which parts of the
362 			 * buffer are valid.
363 			 */
364 			bwrite(bp);
365 			if (bp->b_error)
366 				return (bp->b_error);
367 		} else {
368 			brelse(bp);
369 		}
370 	}
371 
372 	v_inval_buf_range(vp, left_lbn, end_lbn, iosize);
373 	return (0);
374 }
375 
376 
377 /* Send FUSE_LSEEK for this node */
378 static int
379 fuse_vnop_do_lseek(struct vnode *vp, struct thread *td, struct ucred *cred,
380 	pid_t pid, off_t *offp, int whence)
381 {
382 	struct fuse_dispatcher fdi;
383 	struct fuse_filehandle *fufh;
384 	struct fuse_lseek_in *flsi;
385 	struct fuse_lseek_out *flso;
386 	struct mount *mp = vnode_mount(vp);
387 	int err;
388 
389 	ASSERT_VOP_LOCKED(vp, __func__);
390 
391 	err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
392 	if (err)
393 		return (err);
394 	fdisp_init(&fdi, sizeof(*flsi));
395 	fdisp_make_vp(&fdi, FUSE_LSEEK, vp, td, cred);
396 	flsi = fdi.indata;
397 	flsi->fh = fufh->fh_id;
398 	flsi->offset = *offp;
399 	flsi->whence = whence;
400 	err = fdisp_wait_answ(&fdi);
401 	if (err == ENOSYS) {
402 		fsess_set_notimpl(mp, FUSE_LSEEK);
403 	} else if (err == 0) {
404 		fsess_set_impl(mp, FUSE_LSEEK);
405 		flso = fdi.answ;
406 		*offp = flso->offset;
407 	}
408 	fdisp_destroy(&fdi);
409 
410 	return (err);
411 }
412 
413 /*
414     struct vnop_access_args {
415 	struct vnode *a_vp;
416 #if VOP_ACCESS_TAKES_ACCMODE_T
417 	accmode_t a_accmode;
418 #else
419 	int a_mode;
420 #endif
421 	struct ucred *a_cred;
422 	struct thread *a_td;
423     };
424 */
425 static int
426 fuse_vnop_access(struct vop_access_args *ap)
427 {
428 	struct vnode *vp = ap->a_vp;
429 	int accmode = ap->a_accmode;
430 	struct ucred *cred = ap->a_cred;
431 
432 	struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
433 
434 	int err;
435 
436 	if (fuse_isdeadfs(vp)) {
437 		if (vnode_isvroot(vp)) {
438 			return 0;
439 		}
440 		return ENXIO;
441 	}
442 	if (!(data->dataflags & FSESS_INITED)) {
443 		if (vnode_isvroot(vp)) {
444 			if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
445 			    (fuse_match_cred(data->daemoncred, cred) == 0)) {
446 				return 0;
447 			}
448 		}
449 		return EBADF;
450 	}
451 	if (vnode_islnk(vp)) {
452 		return 0;
453 	}
454 
455 	err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
456 	return err;
457 }
458 
459 /*
460  * struct vop_advlock_args {
461  *	struct vop_generic_args a_gen;
462  *	struct vnode *a_vp;
463  *	void *a_id;
464  *	int a_op;
465  *	struct flock *a_fl;
466  *	int a_flags;
467  * }
468  */
469 static int
470 fuse_vnop_advlock(struct vop_advlock_args *ap)
471 {
472 	struct vnode *vp = ap->a_vp;
473 	struct flock *fl = ap->a_fl;
474 	struct thread *td = curthread;
475 	struct ucred *cred = td->td_ucred;
476 	pid_t pid = td->td_proc->p_pid;
477 	struct fuse_filehandle *fufh;
478 	struct fuse_dispatcher fdi;
479 	struct fuse_lk_in *fli;
480 	struct fuse_lk_out *flo;
481 	enum fuse_opcode op;
482 	int dataflags, err;
483 	int flags = ap->a_flags;
484 
485 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
486 
487 	if (fuse_isdeadfs(vp)) {
488 		return ENXIO;
489 	}
490 
491 	switch(ap->a_op) {
492 	case F_GETLK:
493 		op = FUSE_GETLK;
494 		break;
495 	case F_SETLK:
496 		if (flags & F_WAIT)
497 			op = FUSE_SETLKW;
498 		else
499 			op = FUSE_SETLK;
500 		break;
501 	case F_UNLCK:
502 		op = FUSE_SETLK;
503 		break;
504 	default:
505 		return EINVAL;
506 	}
507 
508 	if (!(dataflags & FSESS_POSIX_LOCKS))
509 		return vop_stdadvlock(ap);
510 	/* FUSE doesn't properly support flock until protocol 7.17 */
511 	if (flags & F_FLOCK)
512 		return vop_stdadvlock(ap);
513 
514 	vn_lock(vp, LK_SHARED | LK_RETRY);
515 
516 	err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
517 	if (err)
518 		goto out;
519 
520 	fdisp_init(&fdi, sizeof(*fli));
521 
522 	fdisp_make_vp(&fdi, op, vp, td, cred);
523 	fli = fdi.indata;
524 	fli->fh = fufh->fh_id;
525 	fli->owner = td->td_proc->p_pid;
526 	fli->lk.start = fl->l_start;
527 	if (fl->l_len != 0)
528 		fli->lk.end = fl->l_start + fl->l_len - 1;
529 	else
530 		fli->lk.end = INT64_MAX;
531 	fli->lk.type = fl->l_type;
532 	fli->lk.pid = td->td_proc->p_pid;
533 
534 	err = fdisp_wait_answ(&fdi);
535 	fdisp_destroy(&fdi);
536 
537 	if (err == 0 && op == FUSE_GETLK) {
538 		flo = fdi.answ;
539 		fl->l_type = flo->lk.type;
540 		fl->l_pid = flo->lk.pid;
541 		if (flo->lk.type != F_UNLCK) {
542 			fl->l_start = flo->lk.start;
543 			if (flo->lk.end == INT64_MAX)
544 				fl->l_len = 0;
545 			else
546 				fl->l_len = flo->lk.end - flo->lk.start + 1;
547 			fl->l_start = flo->lk.start;
548 		}
549 	}
550 
551 out:
552 	VOP_UNLOCK(vp);
553 	return err;
554 }
555 
556 static int
557 fuse_vnop_allocate(struct vop_allocate_args *ap)
558 {
559 	struct vnode *vp = ap->a_vp;
560 	off_t *len = ap->a_len;
561 	off_t *offset = ap->a_offset;
562 	struct ucred *cred = ap->a_cred;
563 	struct fuse_filehandle *fufh;
564 	struct mount *mp = vnode_mount(vp);
565 	struct fuse_dispatcher fdi;
566 	struct fuse_fallocate_in *ffi;
567 	struct uio io;
568 	pid_t pid = curthread->td_proc->p_pid;
569 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
570 	off_t filesize;
571 	int err;
572 
573 	if (fuse_isdeadfs(vp))
574 		return (ENXIO);
575 
576 	switch (vp->v_type) {
577 	case VFIFO:
578 		return (ESPIPE);
579 	case VLNK:
580 	case VREG:
581 		if (vfs_isrdonly(mp))
582 			return (EROFS);
583 		break;
584 	default:
585 		return (ENODEV);
586 	}
587 
588 	if (vfs_isrdonly(mp))
589 		return (EROFS);
590 
591 	if (fsess_not_impl(mp, FUSE_FALLOCATE))
592 		return (EINVAL);
593 
594 	io.uio_offset = *offset;
595 	io.uio_resid = *len;
596 	err = vn_rlimit_fsize(vp, &io, curthread);
597 	if (err)
598 		return (err);
599 
600 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
601 	if (err)
602 		return (err);
603 
604 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
605 
606 	err = fuse_vnode_size(vp, &filesize, cred, curthread);
607 	if (err)
608 		return (err);
609 	fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
610 
611 	fdisp_init(&fdi, sizeof(*ffi));
612 	fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
613 	ffi = fdi.indata;
614 	ffi->fh = fufh->fh_id;
615 	ffi->offset = *offset;
616 	ffi->length = *len;
617 	ffi->mode = 0;
618 	err = fdisp_wait_answ(&fdi);
619 
620 	if (err == ENOSYS) {
621 		fsess_set_notimpl(mp, FUSE_FALLOCATE);
622 		err = EINVAL;
623 	} else if (err == EOPNOTSUPP) {
624 		/*
625 		 * The file system server does not support FUSE_FALLOCATE with
626 		 * the supplied mode for this particular file.
627 		 */
628 		err = EINVAL;
629 	} else if (!err) {
630 		*offset += *len;
631 		*len = 0;
632 		fuse_vnode_undirty_cached_timestamps(vp, false);
633 		fuse_internal_clear_suid_on_write(vp, cred, curthread);
634 		if (*offset > fvdat->cached_attrs.va_size) {
635 			fuse_vnode_setsize(vp, *offset, false);
636 			getnanouptime(&fvdat->last_local_modify);
637 		}
638 	}
639 
640 	return (err);
641 }
642 
643 /* {
644 	struct vnode *a_vp;
645 	daddr_t a_bn;
646 	struct bufobj **a_bop;
647 	daddr_t *a_bnp;
648 	int *a_runp;
649 	int *a_runb;
650 } */
651 static int
652 fuse_vnop_bmap(struct vop_bmap_args *ap)
653 {
654 	struct vnode *vp = ap->a_vp;
655 	struct bufobj **bo = ap->a_bop;
656 	struct thread *td = curthread;
657 	struct mount *mp;
658 	struct fuse_dispatcher fdi;
659 	struct fuse_bmap_in *fbi;
660 	struct fuse_bmap_out *fbo;
661 	struct fuse_data *data;
662 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
663 	uint64_t biosize;
664 	off_t fsize;
665 	daddr_t lbn = ap->a_bn;
666 	daddr_t *pbn = ap->a_bnp;
667 	int *runp = ap->a_runp;
668 	int *runb = ap->a_runb;
669 	int error = 0;
670 	int maxrun;
671 
672 	if (fuse_isdeadfs(vp)) {
673 		return ENXIO;
674 	}
675 
676 	mp = vnode_mount(vp);
677 	data = fuse_get_mpdata(mp);
678 	biosize = fuse_iosize(vp);
679 	maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
680 		data->max_readahead_blocks);
681 
682 	if (bo != NULL)
683 		*bo = &vp->v_bufobj;
684 
685 	/*
686 	 * The FUSE_BMAP operation does not include the runp and runb
687 	 * variables, so we must guess.  Report nonzero contiguous runs so
688 	 * cluster_read will combine adjacent reads.  It's worthwhile to reduce
689 	 * upcalls even if we don't know the true physical layout of the file.
690 	 *
691 	 * FUSE file systems may opt out of read clustering in two ways:
692 	 * * mounting with -onoclusterr
693 	 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
694 	 */
695 	if (runb != NULL)
696 		*runb = MIN(lbn, maxrun);
697 	if (runp != NULL && maxrun == 0)
698 		*runp = 0;
699 	else if (runp != NULL) {
700 		/*
701 		 * If the file's size is cached, use that value to calculate
702 		 * runp, even if the cache is expired.  runp is only advisory,
703 		 * and the risk of getting it wrong is not worth the cost of
704 		 * another upcall.
705 		 */
706 		if (fvdat->cached_attrs.va_size != VNOVAL)
707 			fsize = fvdat->cached_attrs.va_size;
708 		else
709 			error = fuse_vnode_size(vp, &fsize, td->td_ucred, td);
710 		if (error == 0)
711 			*runp = MIN(MAX(0, fsize / (off_t)biosize - lbn - 1),
712 				    maxrun);
713 		else
714 			*runp = 0;
715 	}
716 
717 	if (fsess_maybe_impl(mp, FUSE_BMAP)) {
718 		fdisp_init(&fdi, sizeof(*fbi));
719 		fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
720 		fbi = fdi.indata;
721 		fbi->block = lbn;
722 		fbi->blocksize = biosize;
723 		error = fdisp_wait_answ(&fdi);
724 		if (error == ENOSYS) {
725 			fdisp_destroy(&fdi);
726 			fsess_set_notimpl(mp, FUSE_BMAP);
727 			error = 0;
728 		} else {
729 			fbo = fdi.answ;
730 			if (error == 0 && pbn != NULL)
731 				*pbn = fbo->block;
732 			fdisp_destroy(&fdi);
733 			return error;
734 		}
735 	}
736 
737 	/* If the daemon doesn't support BMAP, make up a sensible default */
738 	if (pbn != NULL)
739 		*pbn = lbn * btodb(biosize);
740 	return (error);
741 }
742 
743 /*
744     struct vop_close_args {
745 	struct vnode *a_vp;
746 	int  a_fflag;
747 	struct ucred *a_cred;
748 	struct thread *a_td;
749     };
750 */
751 static int
752 fuse_vnop_close(struct vop_close_args *ap)
753 {
754 	struct vnode *vp = ap->a_vp;
755 	struct ucred *cred = ap->a_cred;
756 	int fflag = ap->a_fflag;
757 	struct thread *td = ap->a_td;
758 	pid_t pid = td->td_proc->p_pid;
759 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
760 	int err = 0;
761 
762 	if (fuse_isdeadfs(vp))
763 		return 0;
764 	if (vnode_isdir(vp))
765 		return 0;
766 	if (fflag & IO_NDELAY)
767 		return 0;
768 
769 	err = fuse_flush(vp, cred, pid, fflag);
770 	if (err == 0 && (fvdat->flag & FN_ATIMECHANGE)) {
771 		struct vattr vap;
772 
773 		VATTR_NULL(&vap);
774 		vap.va_atime = fvdat->cached_attrs.va_atime;
775 		err = fuse_internal_setattr(vp, &vap, td, NULL);
776 	}
777 	/* TODO: close the file handle, if we're sure it's no longer used */
778 	if ((fvdat->flag & FN_SIZECHANGE) != 0) {
779 		fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
780 	}
781 	return err;
782 }
783 
784 /*
785    struct vop_copy_file_range_args {
786 	struct vop_generic_args a_gen;
787 	struct vnode *a_invp;
788 	off_t *a_inoffp;
789 	struct vnode *a_outvp;
790 	off_t *a_outoffp;
791 	size_t *a_lenp;
792 	unsigned int a_flags;
793 	struct ucred *a_incred;
794 	struct ucred *a_outcred;
795 	struct thread *a_fsizetd;
796 }
797  */
798 static int
799 fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap)
800 {
801 	struct vnode *invp = ap->a_invp;
802 	struct vnode *outvp = ap->a_outvp;
803 	struct mount *mp = vnode_mount(invp);
804 	struct fuse_vnode_data *outfvdat = VTOFUD(outvp);
805 	struct fuse_dispatcher fdi;
806 	struct fuse_filehandle *infufh, *outfufh;
807 	struct fuse_copy_file_range_in *fcfri;
808 	struct ucred *incred = ap->a_incred;
809 	struct ucred *outcred = ap->a_outcred;
810 	struct fuse_write_out *fwo;
811 	struct thread *td;
812 	struct uio io;
813 	off_t outfilesize;
814 	pid_t pid;
815 	int err;
816 
817 	if (mp != vnode_mount(outvp))
818 		goto fallback;
819 
820 	if (incred->cr_uid != outcred->cr_uid)
821 		goto fallback;
822 
823 	if (incred->cr_groups[0] != outcred->cr_groups[0])
824 		goto fallback;
825 
826 	if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE))
827 		goto fallback;
828 
829 	if (ap->a_fsizetd == NULL)
830 		td = curthread;
831 	else
832 		td = ap->a_fsizetd;
833 	pid = td->td_proc->p_pid;
834 
835 	/* Lock both vnodes, avoiding risk of deadlock. */
836 	do {
837 		err = vn_lock(outvp, LK_EXCLUSIVE);
838 		if (invp == outvp)
839 			break;
840 		if (err == 0) {
841 			err = vn_lock(invp, LK_SHARED | LK_NOWAIT);
842 			if (err == 0)
843 				break;
844 			VOP_UNLOCK(outvp);
845 			err = vn_lock(invp, LK_SHARED);
846 			if (err == 0)
847 				VOP_UNLOCK(invp);
848 		}
849 	} while (err == 0);
850 	if (err != 0)
851 		return (err);
852 
853 	err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid);
854 	if (err)
855 		goto unlock;
856 
857 	err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid);
858 	if (err)
859 		goto unlock;
860 
861 	if (ap->a_fsizetd) {
862 		io.uio_offset = *ap->a_outoffp;
863 		io.uio_resid = *ap->a_lenp;
864 		err = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd);
865 		if (err)
866 			goto unlock;
867 	}
868 
869 	err = fuse_vnode_size(outvp, &outfilesize, outcred, curthread);
870 	if (err)
871 		goto unlock;
872 
873 	err = fuse_inval_buf_range(outvp, outfilesize, *ap->a_outoffp,
874 		*ap->a_outoffp + *ap->a_lenp);
875 	if (err)
876 		goto unlock;
877 
878 	fdisp_init(&fdi, sizeof(*fcfri));
879 	fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred);
880 	fcfri = fdi.indata;
881 	fcfri->fh_in = infufh->fh_id;
882 	fcfri->off_in = *ap->a_inoffp;
883 	fcfri->nodeid_out = VTOI(outvp);
884 	fcfri->fh_out = outfufh->fh_id;
885 	fcfri->off_out = *ap->a_outoffp;
886 	fcfri->len = *ap->a_lenp;
887 	fcfri->flags = 0;
888 
889 	err = fdisp_wait_answ(&fdi);
890 	if (err == 0) {
891 		fwo = fdi.answ;
892 		*ap->a_lenp = fwo->size;
893 		*ap->a_inoffp += fwo->size;
894 		*ap->a_outoffp += fwo->size;
895 		fuse_internal_clear_suid_on_write(outvp, outcred, td);
896 		if (*ap->a_outoffp > outfvdat->cached_attrs.va_size) {
897                         fuse_vnode_setsize(outvp, *ap->a_outoffp, false);
898 			getnanouptime(&outfvdat->last_local_modify);
899 		}
900 		fuse_vnode_update(invp, FN_ATIMECHANGE);
901 		fuse_vnode_update(outvp, FN_MTIMECHANGE | FN_CTIMECHANGE);
902 	}
903 	fdisp_destroy(&fdi);
904 
905 unlock:
906 	if (invp != outvp)
907 		VOP_UNLOCK(invp);
908 	VOP_UNLOCK(outvp);
909 
910 	if (err == ENOSYS) {
911 		fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE);
912 fallback:
913 		err = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
914 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
915 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
916 	}
917 
918 	return (err);
919 }
920 
921 static void
922 fdisp_make_mknod_for_fallback(
923 	struct fuse_dispatcher *fdip,
924 	struct componentname *cnp,
925 	struct vnode *dvp,
926 	uint64_t parentnid,
927 	struct thread *td,
928 	struct ucred *cred,
929 	mode_t mode,
930 	enum fuse_opcode *op)
931 {
932 	struct fuse_mknod_in *fmni;
933 
934 	fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
935 	*op = FUSE_MKNOD;
936 	fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
937 	fmni = fdip->indata;
938 	fmni->mode = mode;
939 	fmni->rdev = 0;
940 	memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
941 	    cnp->cn_namelen);
942 	((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
943 }
944 /*
945     struct vnop_create_args {
946 	struct vnode *a_dvp;
947 	struct vnode **a_vpp;
948 	struct componentname *a_cnp;
949 	struct vattr *a_vap;
950     };
951 */
952 static int
953 fuse_vnop_create(struct vop_create_args *ap)
954 {
955 	struct vnode *dvp = ap->a_dvp;
956 	struct vnode **vpp = ap->a_vpp;
957 	struct componentname *cnp = ap->a_cnp;
958 	struct vattr *vap = ap->a_vap;
959 	struct thread *td = curthread;
960 	struct ucred *cred = cnp->cn_cred;
961 
962 	struct fuse_data *data;
963 	struct fuse_create_in *fci;
964 	struct fuse_entry_out *feo;
965 	struct fuse_open_out *foo;
966 	struct fuse_dispatcher fdi, fdi2;
967 	struct fuse_dispatcher *fdip = &fdi;
968 	struct fuse_dispatcher *fdip2 = NULL;
969 
970 	int err;
971 
972 	struct mount *mp = vnode_mount(dvp);
973 	data = fuse_get_mpdata(mp);
974 	uint64_t parentnid = VTOFUD(dvp)->nid;
975 	mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
976 	enum fuse_opcode op;
977 	int flags;
978 
979 	if (fuse_isdeadfs(dvp))
980 		return ENXIO;
981 
982 	/* FUSE expects sockets to be created with FUSE_MKNOD */
983 	if (vap->va_type == VSOCK)
984 		return fuse_internal_mknod(dvp, vpp, cnp, vap);
985 
986 	/*
987 	 * VOP_CREATE doesn't tell us the open(2) flags, so we guess.  Only a
988 	 * writable mode makes sense, and we might as well include readability
989 	 * too.
990 	 */
991 	flags = O_RDWR;
992 
993 	bzero(&fdi, sizeof(fdi));
994 
995 	if (vap->va_type != VREG)
996 		return (EINVAL);
997 
998 	if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
999 		/* Fallback to FUSE_MKNOD/FUSE_OPEN */
1000 		fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
1001 			cred, mode, &op);
1002 	} else {
1003 		/* Use FUSE_CREATE */
1004 		size_t insize;
1005 
1006 		op = FUSE_CREATE;
1007 		fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
1008 		fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
1009 		fci = fdip->indata;
1010 		fci->mode = mode;
1011 		fci->flags = O_CREAT | flags;
1012 		if (fuse_libabi_geq(data, 7, 12)) {
1013 			insize = sizeof(*fci);
1014 			fci->umask = td->td_proc->p_pd->pd_cmask;
1015 		} else {
1016 			insize = sizeof(struct fuse_open_in);
1017 		}
1018 
1019 		memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
1020 		    cnp->cn_namelen);
1021 		((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
1022 	}
1023 
1024 	err = fdisp_wait_answ(fdip);
1025 
1026 	if (err) {
1027 		if (err == ENOSYS && op == FUSE_CREATE) {
1028 			fsess_set_notimpl(mp, FUSE_CREATE);
1029 			fdisp_destroy(fdip);
1030 			fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
1031 				parentnid, td, cred, mode, &op);
1032 			err = fdisp_wait_answ(fdip);
1033 		}
1034 		if (err)
1035 			goto out;
1036 	}
1037 
1038 	feo = fdip->answ;
1039 
1040 	if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
1041 		goto out;
1042 	}
1043 
1044 	if (op == FUSE_CREATE) {
1045 		if (fuse_libabi_geq(data, 7, 9))
1046 			foo = (struct fuse_open_out*)(feo + 1);
1047 		else
1048 			foo = (struct fuse_open_out*)((char*)feo +
1049 				FUSE_COMPAT_ENTRY_OUT_SIZE);
1050 	} else {
1051 		/* Issue a separate FUSE_OPEN */
1052 		struct fuse_open_in *foi;
1053 
1054 		fdip2 = &fdi2;
1055 		fdisp_init(fdip2, sizeof(*foi));
1056 		fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
1057 			cred);
1058 		foi = fdip2->indata;
1059 		foi->flags = flags;
1060 		err = fdisp_wait_answ(fdip2);
1061 		if (err)
1062 			goto out;
1063 		foo = fdip2->answ;
1064 	}
1065 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
1066 	if (err) {
1067 		struct fuse_release_in *fri;
1068 		uint64_t nodeid = feo->nodeid;
1069 		uint64_t fh_id = foo->fh;
1070 
1071 		fdisp_init(fdip, sizeof(*fri));
1072 		fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
1073 		fri = fdip->indata;
1074 		fri->fh = fh_id;
1075 		fri->flags = flags;
1076 		fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
1077 		fuse_insert_message(fdip->tick, false);
1078 		goto out;
1079 	}
1080 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
1081 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
1082 		feo->attr_valid_nsec, NULL, true);
1083 
1084 	fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
1085 	fuse_vnode_open(*vpp, foo->open_flags, td);
1086 	/*
1087 	 * Purge the parent's attribute cache because the daemon should've
1088 	 * updated its mtime and ctime
1089 	 */
1090 	fuse_vnode_clear_attr_cache(dvp);
1091 	cache_purge_negative(dvp);
1092 
1093 out:
1094 	if (fdip2)
1095 		fdisp_destroy(fdip2);
1096 	fdisp_destroy(fdip);
1097 	return err;
1098 }
1099 
1100 /*
1101     struct vnop_fdatasync_args {
1102 	struct vop_generic_args a_gen;
1103 	struct vnode * a_vp;
1104 	struct thread * a_td;
1105     };
1106 */
1107 static int
1108 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
1109 {
1110 	struct vnode *vp = ap->a_vp;
1111 	struct thread *td = ap->a_td;
1112 	int waitfor = MNT_WAIT;
1113 
1114 	int err = 0;
1115 
1116 	if (fuse_isdeadfs(vp)) {
1117 		return 0;
1118 	}
1119 	if ((err = vop_stdfdatasync_buf(ap)))
1120 		return err;
1121 
1122 	return fuse_internal_fsync(vp, td, waitfor, true);
1123 }
1124 
1125 /*
1126     struct vnop_fsync_args {
1127 	struct vop_generic_args a_gen;
1128 	struct vnode * a_vp;
1129 	int  a_waitfor;
1130 	struct thread * a_td;
1131     };
1132 */
1133 static int
1134 fuse_vnop_fsync(struct vop_fsync_args *ap)
1135 {
1136 	struct vnode *vp = ap->a_vp;
1137 	struct thread *td = ap->a_td;
1138 	int waitfor = ap->a_waitfor;
1139 	int err = 0;
1140 
1141 	if (fuse_isdeadfs(vp)) {
1142 		return 0;
1143 	}
1144 	if ((err = vop_stdfsync(ap)))
1145 		return err;
1146 
1147 	return fuse_internal_fsync(vp, td, waitfor, false);
1148 }
1149 
1150 /*
1151     struct vnop_getattr_args {
1152 	struct vnode *a_vp;
1153 	struct vattr *a_vap;
1154 	struct ucred *a_cred;
1155 	struct thread *a_td;
1156     };
1157 */
1158 static int
1159 fuse_vnop_getattr(struct vop_getattr_args *ap)
1160 {
1161 	struct vnode *vp = ap->a_vp;
1162 	struct vattr *vap = ap->a_vap;
1163 	struct ucred *cred = ap->a_cred;
1164 	struct thread *td = curthread;
1165 
1166 	int err = 0;
1167 	int dataflags;
1168 
1169 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
1170 
1171 	/* Note that we are not bailing out on a dead file system just yet. */
1172 
1173 	if (!(dataflags & FSESS_INITED)) {
1174 		if (!vnode_isvroot(vp)) {
1175 			fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
1176 			err = ENOTCONN;
1177 			return err;
1178 		} else {
1179 			goto fake;
1180 		}
1181 	}
1182 	err = fuse_internal_getattr(vp, vap, cred, td);
1183 	if (err == ENOTCONN && vnode_isvroot(vp)) {
1184 		/* see comment in fuse_vfsop_statfs() */
1185 		goto fake;
1186 	} else {
1187 		return err;
1188 	}
1189 
1190 fake:
1191 	bzero(vap, sizeof(*vap));
1192 	vap->va_type = vnode_vtype(vp);
1193 
1194 	return 0;
1195 }
1196 
1197 /*
1198     struct vnop_inactive_args {
1199 	struct vnode *a_vp;
1200     };
1201 */
1202 static int
1203 fuse_vnop_inactive(struct vop_inactive_args *ap)
1204 {
1205 	struct vnode *vp = ap->a_vp;
1206 	struct thread *td = curthread;
1207 
1208 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1209 	struct fuse_filehandle *fufh, *fufh_tmp;
1210 
1211 	int need_flush = 1;
1212 
1213 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1214 		if (need_flush && vp->v_type == VREG) {
1215 			if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
1216 				fuse_vnode_savesize(vp, NULL, 0);
1217 			}
1218 			if ((fvdat->flag & FN_REVOKED) != 0)
1219 				fuse_io_invalbuf(vp, td);
1220 			else
1221 				fuse_io_flushbuf(vp, MNT_WAIT, td);
1222 			need_flush = 0;
1223 		}
1224 		fuse_filehandle_close(vp, fufh, td, NULL);
1225 	}
1226 
1227 	if ((fvdat->flag & FN_REVOKED) != 0)
1228 		vrecycle(vp);
1229 
1230 	return 0;
1231 }
1232 
1233 /*
1234     struct vnop_ioctl_args {
1235 	struct vnode *a_vp;
1236 	u_long a_command;
1237 	caddr_t a_data;
1238 	int a_fflag;
1239 	struct ucred *a_cred;
1240 	struct thread *a_td;
1241     };
1242 */
1243 static int
1244 fuse_vnop_ioctl(struct vop_ioctl_args *ap)
1245 {
1246 	struct vnode *vp = ap->a_vp;
1247 	struct mount *mp = vnode_mount(vp);
1248 	struct ucred *cred = ap->a_cred;
1249 	off_t *offp;
1250 	pid_t pid = ap->a_td->td_proc->p_pid;
1251 	int err;
1252 
1253 	switch (ap->a_command) {
1254 	case FIOSEEKDATA:
1255 	case FIOSEEKHOLE:
1256 		/* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */
1257 		if (fsess_maybe_impl(mp, FUSE_LSEEK)) {
1258 			int whence;
1259 
1260 			offp = ap->a_data;
1261 			if (ap->a_command == FIOSEEKDATA)
1262 				whence = SEEK_DATA;
1263 			else
1264 				whence = SEEK_HOLE;
1265 
1266 			vn_lock(vp, LK_SHARED | LK_RETRY);
1267 			err = fuse_vnop_do_lseek(vp, ap->a_td, cred, pid, offp,
1268 			    whence);
1269 			VOP_UNLOCK(vp);
1270 		}
1271 		if (fsess_not_impl(mp, FUSE_LSEEK))
1272 			err = vop_stdioctl(ap);
1273 		break;
1274 	default:
1275 		/* TODO: implement FUSE_IOCTL */
1276 		err = ENOTTY;
1277 		break;
1278 	}
1279 	return (err);
1280 }
1281 
1282 
1283 /*
1284     struct vnop_link_args {
1285 	struct vnode *a_tdvp;
1286 	struct vnode *a_vp;
1287 	struct componentname *a_cnp;
1288     };
1289 */
1290 static int
1291 fuse_vnop_link(struct vop_link_args *ap)
1292 {
1293 	struct vnode *vp = ap->a_vp;
1294 	struct vnode *tdvp = ap->a_tdvp;
1295 	struct componentname *cnp = ap->a_cnp;
1296 
1297 	struct vattr *vap = VTOVA(vp);
1298 
1299 	struct fuse_dispatcher fdi;
1300 	struct fuse_entry_out *feo;
1301 	struct fuse_link_in fli;
1302 
1303 	int err;
1304 
1305 	if (fuse_isdeadfs(vp)) {
1306 		return ENXIO;
1307 	}
1308 	if (vnode_mount(tdvp) != vnode_mount(vp)) {
1309 		return EXDEV;
1310 	}
1311 
1312 	/*
1313 	 * This is a seatbelt check to protect naive userspace filesystems from
1314 	 * themselves and the limitations of the FUSE IPC protocol.  If a
1315 	 * filesystem does not allow attribute caching, assume it is capable of
1316 	 * validating that nlink does not overflow.
1317 	 */
1318 	if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
1319 		return EMLINK;
1320 	fli.oldnodeid = VTOI(vp);
1321 
1322 	fdisp_init(&fdi, 0);
1323 	fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
1324 	    FUSE_LINK, &fli, sizeof(fli), &fdi);
1325 	if ((err = fdisp_wait_answ(&fdi))) {
1326 		goto out;
1327 	}
1328 	feo = fdi.answ;
1329 
1330 	err = fuse_internal_checkentry(feo, vnode_vtype(vp));
1331 	if (!err) {
1332 		/*
1333 		 * Purge the parent's attribute cache because the daemon
1334 		 * should've updated its mtime and ctime
1335 		 */
1336 		fuse_vnode_clear_attr_cache(tdvp);
1337 		fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
1338 			feo->attr_valid_nsec, NULL, true);
1339 	}
1340 out:
1341 	fdisp_destroy(&fdi);
1342 	return err;
1343 }
1344 
1345 struct fuse_lookup_alloc_arg {
1346 	struct fuse_entry_out *feo;
1347 	struct componentname *cnp;
1348 	uint64_t nid;
1349 	enum vtype vtyp;
1350 };
1351 
1352 /* Callback for vn_get_ino */
1353 static int
1354 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
1355 {
1356 	struct fuse_lookup_alloc_arg *flaa = arg;
1357 
1358 	return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
1359 		flaa->vtyp);
1360 }
1361 
1362 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
1363 	"int", "struct timespec*", "struct timespec*");
1364 /*
1365     struct vnop_lookup_args {
1366 	struct vnodeop_desc *a_desc;
1367 	struct vnode *a_dvp;
1368 	struct vnode **a_vpp;
1369 	struct componentname *a_cnp;
1370     };
1371 */
1372 int
1373 fuse_vnop_lookup(struct vop_lookup_args *ap)
1374 {
1375 	struct vnode *dvp = ap->a_dvp;
1376 	struct vnode **vpp = ap->a_vpp;
1377 	struct componentname *cnp = ap->a_cnp;
1378 	struct thread *td = curthread;
1379 	struct ucred *cred = cnp->cn_cred;
1380 	struct timespec now;
1381 
1382 	int nameiop = cnp->cn_nameiop;
1383 	int flags = cnp->cn_flags;
1384 	int wantparent = flags & (LOCKPARENT | WANTPARENT);
1385 	int islastcn = flags & ISLASTCN;
1386 	struct mount *mp = vnode_mount(dvp);
1387 	struct fuse_data *data = fuse_get_mpdata(mp);
1388 	int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
1389 
1390 	int err = 0;
1391 	int lookup_err = 0;
1392 	struct vnode *vp = NULL;
1393 
1394 	struct fuse_dispatcher fdi;
1395 	bool did_lookup = false;
1396 	struct fuse_entry_out *feo = NULL;
1397 	enum vtype vtyp;	/* vnode type of target */
1398 
1399 	uint64_t nid;
1400 
1401 	if (fuse_isdeadfs(dvp)) {
1402 		*vpp = NULL;
1403 		return ENXIO;
1404 	}
1405 	if (!vnode_isdir(dvp))
1406 		return ENOTDIR;
1407 
1408 	if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1409 		return EROFS;
1410 
1411 	if ((cnp->cn_flags & NOEXECCHECK) != 0)
1412 		cnp->cn_flags &= ~NOEXECCHECK;
1413 	else if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1414 		return err;
1415 
1416 	if ((flags & ISDOTDOT) && !(data->dataflags & FSESS_EXPORT_SUPPORT))
1417 	{
1418 		if (!(VTOFUD(dvp)->flag & FN_PARENT_NID)) {
1419 			/*
1420 			 * Since the file system doesn't support ".." lookups,
1421 			 * we have no way to find this entry.
1422 			 */
1423 			return ESTALE;
1424 		}
1425 		nid = VTOFUD(dvp)->parent_nid;
1426 		if (nid == 0)
1427 			return ENOENT;
1428 		/* .. is obviously a directory */
1429 		vtyp = VDIR;
1430 	} else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
1431 		nid = VTOI(dvp);
1432 		/* . is obviously a directory */
1433 		vtyp = VDIR;
1434 	} else {
1435 		struct timespec timeout;
1436 		int ncpticks; /* here to accommodate for API contract */
1437 
1438 		err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks);
1439 		getnanouptime(&now);
1440 		SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1441 		switch (err) {
1442 		case -1:		/* positive match */
1443 			if (timespeccmp(&timeout, &now, >)) {
1444 				counter_u64_add(fuse_lookup_cache_hits, 1);
1445 			} else {
1446 				/* Cache timeout */
1447 				counter_u64_add(fuse_lookup_cache_misses, 1);
1448 				bintime_clear(
1449 					&VTOFUD(*vpp)->entry_cache_timeout);
1450 				cache_purge(*vpp);
1451 				if (dvp != *vpp)
1452 					vput(*vpp);
1453 				else
1454 					vrele(*vpp);
1455 				*vpp = NULL;
1456 				break;
1457 			}
1458 			return 0;
1459 
1460 		case 0:		/* no match in cache */
1461 			counter_u64_add(fuse_lookup_cache_misses, 1);
1462 			break;
1463 
1464 		case ENOENT:		/* negative match */
1465 			if (timespeccmp(&timeout, &now, <=)) {
1466 				/* Cache timeout */
1467 				cache_purge_negative(dvp);
1468 				break;
1469 			}
1470 			/* fall through */
1471 		default:
1472 			return err;
1473 		}
1474 
1475 		fdisp_init(&fdi, cnp->cn_namelen + 1);
1476 		fdisp_make(&fdi, FUSE_LOOKUP, mp, VTOI(dvp), td, cred);
1477 
1478 		memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1479 		((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1480 		lookup_err = fdisp_wait_answ(&fdi);
1481 		did_lookup = true;
1482 
1483 		if (!lookup_err) {
1484 			/* lookup call succeeded */
1485 			feo = (struct fuse_entry_out *)fdi.answ;
1486 			nid = feo->nodeid;
1487 			if (nid == 0) {
1488 				/* zero nodeid means ENOENT and cache it */
1489 				struct timespec timeout;
1490 
1491 				fdi.answ_stat = ENOENT;
1492 				lookup_err = ENOENT;
1493 				if (cnp->cn_flags & MAKEENTRY) {
1494 					fuse_validity_2_timespec(feo, &timeout);
1495 					/* Use the same entry_time for .. as for
1496 					 * the file itself.  That doesn't honor
1497 					 * exactly what the fuse server tells
1498 					 * us, but to do otherwise would require
1499 					 * another cache lookup at this point.
1500 					 */
1501 					struct timespec *dtsp = NULL;
1502 					cache_enter_time(dvp, *vpp, cnp,
1503 						&timeout, dtsp);
1504 				}
1505 			}
1506 			vtyp = IFTOVT(feo->attr.mode);
1507 		}
1508 		if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1509 			fdisp_destroy(&fdi);
1510 			return lookup_err;
1511 		}
1512 	}
1513 	/* lookup_err, if non-zero, must be ENOENT at this point */
1514 
1515 	if (lookup_err) {
1516 		/* Entry not found */
1517 		if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1518 			if (default_permissions)
1519 				err = fuse_internal_access(dvp, VWRITE, td,
1520 				    cred);
1521 			else
1522 				err = 0;
1523 			if (!err) {
1524 				/*
1525 				 * Set the SAVENAME flag to hold onto the
1526 				 * pathname for use later in VOP_CREATE or
1527 				 * VOP_RENAME.
1528 				 */
1529 				cnp->cn_flags |= SAVENAME;
1530 
1531 				err = EJUSTRETURN;
1532 			}
1533 		} else {
1534 			err = ENOENT;
1535 		}
1536 	} else {
1537 		/* Entry was found */
1538 		if (flags & ISDOTDOT) {
1539 			struct fuse_lookup_alloc_arg flaa;
1540 
1541 			flaa.nid = nid;
1542 			flaa.feo = feo;
1543 			flaa.cnp = cnp;
1544 			flaa.vtyp = vtyp;
1545 			err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1546 				&vp);
1547 			*vpp = vp;
1548 		} else if (nid == VTOI(dvp)) {
1549 			vref(dvp);
1550 			*vpp = dvp;
1551 		} else {
1552 			struct fuse_vnode_data *fvdat;
1553 
1554 			err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1555 			    &vp, cnp, vtyp);
1556 			if (err)
1557 				goto out;
1558 			*vpp = vp;
1559 			fvdat = VTOFUD(vp);
1560 
1561 			MPASS(feo != NULL);
1562 			if (timespeccmp(&now, &fvdat->last_local_modify, >)) {
1563 				/*
1564 				 * Attributes from the server are definitely
1565 				 * newer than the last attributes we sent to
1566 				 * the server, so cache them.
1567 				 */
1568 				fuse_internal_cache_attrs(*vpp, &feo->attr,
1569 					feo->attr_valid, feo->attr_valid_nsec,
1570 					NULL, true);
1571 			}
1572 			fuse_validity_2_bintime(feo->entry_valid,
1573 				feo->entry_valid_nsec,
1574 				&fvdat->entry_cache_timeout);
1575 
1576 			if ((nameiop == DELETE || nameiop == RENAME) &&
1577 				islastcn && default_permissions)
1578 			{
1579 				struct vattr dvattr;
1580 
1581 				err = fuse_internal_access(dvp, VWRITE, td,
1582 					cred);
1583 				if (err != 0)
1584 					goto out;
1585 				/*
1586 				 * if the parent's sticky bit is set, check
1587 				 * whether we're allowed to remove the file.
1588 				 * Need to figure out the vnode locking to make
1589 				 * this work.
1590 				 */
1591 				fuse_internal_getattr(dvp, &dvattr, cred, td);
1592 				if ((dvattr.va_mode & S_ISTXT) &&
1593 					fuse_internal_access(dvp, VADMIN, td,
1594 						cred) &&
1595 					fuse_internal_access(*vpp, VADMIN, td,
1596 						cred)) {
1597 					err = EPERM;
1598 					goto out;
1599 				}
1600 			}
1601 
1602 			if (islastcn && (
1603 				(nameiop == DELETE) ||
1604 				(nameiop == RENAME && wantparent))) {
1605 				cnp->cn_flags |= SAVENAME;
1606 			}
1607 		}
1608 	}
1609 out:
1610 	if (err) {
1611 		if (vp != NULL && dvp != vp)
1612 			vput(vp);
1613 		else if (vp != NULL)
1614 			vrele(vp);
1615 		*vpp = NULL;
1616 	}
1617 	if (did_lookup)
1618 		fdisp_destroy(&fdi);
1619 
1620 	return err;
1621 }
1622 
1623 /*
1624     struct vnop_mkdir_args {
1625 	struct vnode *a_dvp;
1626 	struct vnode **a_vpp;
1627 	struct componentname *a_cnp;
1628 	struct vattr *a_vap;
1629     };
1630 */
1631 static int
1632 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1633 {
1634 	struct vnode *dvp = ap->a_dvp;
1635 	struct vnode **vpp = ap->a_vpp;
1636 	struct componentname *cnp = ap->a_cnp;
1637 	struct vattr *vap = ap->a_vap;
1638 
1639 	struct fuse_mkdir_in fmdi;
1640 
1641 	if (fuse_isdeadfs(dvp)) {
1642 		return ENXIO;
1643 	}
1644 	fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1645 	fmdi.umask = curthread->td_proc->p_pd->pd_cmask;
1646 
1647 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1648 	    sizeof(fmdi), VDIR));
1649 }
1650 
1651 /*
1652     struct vnop_mknod_args {
1653 	struct vnode *a_dvp;
1654 	struct vnode **a_vpp;
1655 	struct componentname *a_cnp;
1656 	struct vattr *a_vap;
1657     };
1658 */
1659 static int
1660 fuse_vnop_mknod(struct vop_mknod_args *ap)
1661 {
1662 
1663 	struct vnode *dvp = ap->a_dvp;
1664 	struct vnode **vpp = ap->a_vpp;
1665 	struct componentname *cnp = ap->a_cnp;
1666 	struct vattr *vap = ap->a_vap;
1667 
1668 	if (fuse_isdeadfs(dvp))
1669 		return ENXIO;
1670 
1671 	return fuse_internal_mknod(dvp, vpp, cnp, vap);
1672 }
1673 
1674 /*
1675     struct vop_open_args {
1676 	struct vnode *a_vp;
1677 	int  a_mode;
1678 	struct ucred *a_cred;
1679 	struct thread *a_td;
1680 	int a_fdidx; / struct file *a_fp;
1681     };
1682 */
1683 static int
1684 fuse_vnop_open(struct vop_open_args *ap)
1685 {
1686 	struct vnode *vp = ap->a_vp;
1687 	int a_mode = ap->a_mode;
1688 	struct thread *td = ap->a_td;
1689 	struct ucred *cred = ap->a_cred;
1690 	pid_t pid = td->td_proc->p_pid;
1691 
1692 	if (fuse_isdeadfs(vp))
1693 		return ENXIO;
1694 	if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1695 		return (EOPNOTSUPP);
1696 	if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1697 		return EINVAL;
1698 
1699 	if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1700 		fuse_vnode_open(vp, 0, td);
1701 		return 0;
1702 	}
1703 
1704 	return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1705 }
1706 
1707 static int
1708 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1709 {
1710 	struct vnode *vp = ap->a_vp;
1711 	struct mount *mp;
1712 
1713 	switch (ap->a_name) {
1714 	case _PC_FILESIZEBITS:
1715 		*ap->a_retval = 64;
1716 		return (0);
1717 	case _PC_NAME_MAX:
1718 		*ap->a_retval = NAME_MAX;
1719 		return (0);
1720 	case _PC_LINK_MAX:
1721 		*ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1722 		return (0);
1723 	case _PC_SYMLINK_MAX:
1724 		*ap->a_retval = MAXPATHLEN;
1725 		return (0);
1726 	case _PC_NO_TRUNC:
1727 		*ap->a_retval = 1;
1728 		return (0);
1729 	case _PC_MIN_HOLE_SIZE:
1730 		/*
1731 		 * The FUSE protocol provides no mechanism for a server to
1732 		 * report _PC_MIN_HOLE_SIZE.  It's a protocol bug.  Instead,
1733 		 * return EINVAL if the server does not support FUSE_LSEEK, or
1734 		 * 1 if it does.
1735 		 */
1736 		mp = vnode_mount(vp);
1737 		if (!fsess_is_impl(mp, FUSE_LSEEK) &&
1738 		    !fsess_not_impl(mp, FUSE_LSEEK)) {
1739 			off_t offset = 0;
1740 
1741 			/* Issue a FUSE_LSEEK to find out if it's implemented */
1742 			fuse_vnop_do_lseek(vp, curthread, curthread->td_ucred,
1743 			    curthread->td_proc->p_pid, &offset, SEEK_DATA);
1744 		}
1745 
1746 		if (fsess_is_impl(mp, FUSE_LSEEK)) {
1747 			*ap->a_retval = 1;
1748 			return (0);
1749 		} else {
1750 			/*
1751 			 * Probably FUSE_LSEEK is not implemented.  It might
1752 			 * be, if the FUSE_LSEEK above returned an error like
1753 			 * EACCES, but in that case we can't tell, so it's
1754 			 * safest to report EINVAL anyway.
1755 			 */
1756 			return (EINVAL);
1757 		}
1758 	default:
1759 		return (vop_stdpathconf(ap));
1760 	}
1761 }
1762 
1763 SDT_PROBE_DEFINE3(fusefs, , vnops, filehandles_closed, "struct vnode*",
1764     "struct uio*", "struct ucred*");
1765 /*
1766     struct vnop_read_args {
1767 	struct vnode *a_vp;
1768 	struct uio *a_uio;
1769 	int  a_ioflag;
1770 	struct ucred *a_cred;
1771     };
1772 */
1773 static int
1774 fuse_vnop_read(struct vop_read_args *ap)
1775 {
1776 	struct vnode *vp = ap->a_vp;
1777 	struct uio *uio = ap->a_uio;
1778 	int ioflag = ap->a_ioflag;
1779 	struct ucred *cred = ap->a_cred;
1780 	pid_t pid = curthread->td_proc->p_pid;
1781 	struct fuse_filehandle *fufh;
1782 	int err;
1783 	bool closefufh = false, directio;
1784 
1785 	MPASS(vp->v_type == VREG || vp->v_type == VDIR);
1786 
1787 	if (fuse_isdeadfs(vp)) {
1788 		return ENXIO;
1789 	}
1790 
1791 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1792 		ioflag |= IO_DIRECT;
1793 	}
1794 
1795 	err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
1796 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1797 		/*
1798 		 * nfsd will do I/O without first doing VOP_OPEN.  We
1799 		 * must implicitly open the file here
1800 		 */
1801 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1802 		closefufh = true;
1803 	}
1804 	if (err) {
1805 		SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
1806 		return err;
1807 	}
1808 
1809 	/*
1810          * Ideally, when the daemon asks for direct io at open time, the
1811          * standard file flag should be set according to this, so that would
1812          * just change the default mode, which later on could be changed via
1813          * fcntl(2).
1814          * But this doesn't work, the O_DIRECT flag gets cleared at some point
1815          * (don't know where). So to make any use of the Fuse direct_io option,
1816          * we hardwire it into the file's private data (similarly to Linux,
1817          * btw.).
1818          */
1819 	directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
1820 
1821 	fuse_vnode_update(vp, FN_ATIMECHANGE);
1822 	if (directio) {
1823 		SDT_PROBE2(fusefs, , vnops, trace, 1, "direct read of vnode");
1824 		err = fuse_read_directbackend(vp, uio, cred, fufh);
1825 	} else {
1826 		SDT_PROBE2(fusefs, , vnops, trace, 1, "buffered read of vnode");
1827 		err = fuse_read_biobackend(vp, uio, ioflag, cred, fufh, pid);
1828 	}
1829 
1830 	if (closefufh)
1831 		fuse_filehandle_close(vp, fufh, curthread, cred);
1832 
1833 	return (err);
1834 }
1835 
1836 /*
1837     struct vnop_readdir_args {
1838 	struct vnode *a_vp;
1839 	struct uio *a_uio;
1840 	struct ucred *a_cred;
1841 	int *a_eofflag;
1842 	int *a_ncookies;
1843 	uint64_t **a_cookies;
1844     };
1845 */
1846 static int
1847 fuse_vnop_readdir(struct vop_readdir_args *ap)
1848 {
1849 	struct vnode *vp = ap->a_vp;
1850 	struct uio *uio = ap->a_uio;
1851 	struct ucred *cred = ap->a_cred;
1852 	struct fuse_filehandle *fufh = NULL;
1853 	struct mount *mp = vnode_mount(vp);
1854 	struct fuse_iov cookediov;
1855 	int err = 0;
1856 	uint64_t *cookies;
1857 	ssize_t tresid;
1858 	int ncookies;
1859 	bool closefufh = false;
1860 	pid_t pid = curthread->td_proc->p_pid;
1861 
1862 	if (ap->a_eofflag)
1863 		*ap->a_eofflag = 0;
1864 	if (fuse_isdeadfs(vp)) {
1865 		return ENXIO;
1866 	}
1867 	if (				/* XXXIP ((uio_iovcnt(uio) > 1)) || */
1868 	    (uio_resid(uio) < sizeof(struct dirent))) {
1869 		return EINVAL;
1870 	}
1871 
1872 	tresid = uio->uio_resid;
1873 	err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1874 	if (err == EBADF && mp->mnt_flag & MNT_EXPORTED) {
1875 		KASSERT(fuse_get_mpdata(mp)->dataflags
1876 				& FSESS_NO_OPENDIR_SUPPORT,
1877 			("FUSE file systems that don't set "
1878 			 "FUSE_NO_OPENDIR_SUPPORT should not be exported"));
1879 		/*
1880 		 * nfsd will do VOP_READDIR without first doing VOP_OPEN.  We
1881 		 * must implicitly open the directory here.
1882 		 */
1883 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1884 		closefufh = true;
1885 	}
1886 	if (err)
1887 		return (err);
1888 	if (ap->a_ncookies != NULL) {
1889 		ncookies = uio->uio_resid /
1890 			(offsetof(struct dirent, d_name) + 4) + 1;
1891 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1892 		*ap->a_ncookies = ncookies;
1893 		*ap->a_cookies = cookies;
1894 	} else {
1895 		ncookies = 0;
1896 		cookies = NULL;
1897 	}
1898 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1899 	fiov_init(&cookediov, DIRCOOKEDSIZE);
1900 
1901 	err = fuse_internal_readdir(vp, uio, fufh, &cookediov,
1902 		&ncookies, cookies);
1903 
1904 	fiov_teardown(&cookediov);
1905 	if (closefufh)
1906 		fuse_filehandle_close(vp, fufh, curthread, cred);
1907 
1908 	if (ap->a_ncookies != NULL) {
1909 		if (err == 0) {
1910 			*ap->a_ncookies -= ncookies;
1911 		} else {
1912 			free(*ap->a_cookies, M_TEMP);
1913 			*ap->a_ncookies = 0;
1914 			*ap->a_cookies = NULL;
1915 		}
1916 	}
1917 	if (err == 0 && tresid == uio->uio_resid)
1918 		*ap->a_eofflag = 1;
1919 
1920 	return err;
1921 }
1922 
1923 /*
1924     struct vnop_readlink_args {
1925 	struct vnode *a_vp;
1926 	struct uio *a_uio;
1927 	struct ucred *a_cred;
1928     };
1929 */
1930 static int
1931 fuse_vnop_readlink(struct vop_readlink_args *ap)
1932 {
1933 	struct vnode *vp = ap->a_vp;
1934 	struct uio *uio = ap->a_uio;
1935 	struct ucred *cred = ap->a_cred;
1936 
1937 	struct fuse_dispatcher fdi;
1938 	int err;
1939 
1940 	if (fuse_isdeadfs(vp)) {
1941 		return ENXIO;
1942 	}
1943 	if (!vnode_islnk(vp)) {
1944 		return EINVAL;
1945 	}
1946 	fdisp_init(&fdi, 0);
1947 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1948 	if (err) {
1949 		goto out;
1950 	}
1951 	if (((char *)fdi.answ)[0] == '/' &&
1952 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1953 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1954 
1955 		err = uiomove(mpth, strlen(mpth), uio);
1956 	}
1957 	if (!err) {
1958 		err = uiomove(fdi.answ, fdi.iosize, uio);
1959 	}
1960 out:
1961 	fdisp_destroy(&fdi);
1962 	return err;
1963 }
1964 
1965 /*
1966     struct vnop_reclaim_args {
1967 	struct vnode *a_vp;
1968     };
1969 */
1970 static int
1971 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1972 {
1973 	struct vnode *vp = ap->a_vp;
1974 	struct thread *td = curthread;
1975 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1976 	struct fuse_filehandle *fufh, *fufh_tmp;
1977 
1978 	if (!fvdat) {
1979 		panic("FUSE: no vnode data during recycling");
1980 	}
1981 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1982 		printf("FUSE: vnode being reclaimed with open fufh "
1983 			"(type=%#x)", fufh->fufh_type);
1984 		fuse_filehandle_close(vp, fufh, td, NULL);
1985 	}
1986 
1987 	if (VTOI(vp) == 1) {
1988 		/*
1989 		 * Don't send FUSE_FORGET for the root inode, because
1990 		 * we never send FUSE_LOOKUP for it (see
1991 		 * fuse_vfsop_root) and we don't want the server to see
1992 		 * mismatched lookup counts.
1993 		 */
1994 		struct fuse_data *data;
1995 		struct vnode *vroot;
1996 
1997 		data = fuse_get_mpdata(vnode_mount(vp));
1998 		FUSE_LOCK();
1999 		vroot = data->vroot;
2000 		data->vroot = NULL;
2001 		FUSE_UNLOCK();
2002 		if (vroot)
2003 			vrele(vroot);
2004 	} else if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
2005 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
2006 		    fvdat->nlookup);
2007 	}
2008 	cache_purge(vp);
2009 	vfs_hash_remove(vp);
2010 	fuse_vnode_destroy(vp);
2011 
2012 	return 0;
2013 }
2014 
2015 /*
2016     struct vnop_remove_args {
2017 	struct vnode *a_dvp;
2018 	struct vnode *a_vp;
2019 	struct componentname *a_cnp;
2020     };
2021 */
2022 static int
2023 fuse_vnop_remove(struct vop_remove_args *ap)
2024 {
2025 	struct vnode *dvp = ap->a_dvp;
2026 	struct vnode *vp = ap->a_vp;
2027 	struct componentname *cnp = ap->a_cnp;
2028 
2029 	int err;
2030 
2031 	if (fuse_isdeadfs(vp)) {
2032 		return ENXIO;
2033 	}
2034 	if (vnode_isdir(vp)) {
2035 		return EPERM;
2036 	}
2037 
2038 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
2039 
2040 	return err;
2041 }
2042 
2043 /*
2044     struct vnop_rename_args {
2045 	struct vnode *a_fdvp;
2046 	struct vnode *a_fvp;
2047 	struct componentname *a_fcnp;
2048 	struct vnode *a_tdvp;
2049 	struct vnode *a_tvp;
2050 	struct componentname *a_tcnp;
2051     };
2052 */
2053 static int
2054 fuse_vnop_rename(struct vop_rename_args *ap)
2055 {
2056 	struct vnode *fdvp = ap->a_fdvp;
2057 	struct vnode *fvp = ap->a_fvp;
2058 	struct componentname *fcnp = ap->a_fcnp;
2059 	struct vnode *tdvp = ap->a_tdvp;
2060 	struct vnode *tvp = ap->a_tvp;
2061 	struct componentname *tcnp = ap->a_tcnp;
2062 	struct fuse_data *data;
2063 	bool newparent = fdvp != tdvp;
2064 	bool isdir = fvp->v_type == VDIR;
2065 	int err = 0;
2066 
2067 	if (fuse_isdeadfs(fdvp)) {
2068 		return ENXIO;
2069 	}
2070 	if (fvp->v_mount != tdvp->v_mount ||
2071 	    (tvp && fvp->v_mount != tvp->v_mount)) {
2072 		SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
2073 		err = EXDEV;
2074 		goto out;
2075 	}
2076 	cache_purge(fvp);
2077 
2078 	/*
2079 	 * FUSE library is expected to check if target directory is not
2080 	 * under the source directory in the file system tree.
2081 	 * Linux performs this check at VFS level.
2082 	 */
2083 	/*
2084 	 * If source is a directory, and it will get a new parent, user must
2085 	 * have write permission to it, so ".." can be modified.
2086 	 */
2087 	data = fuse_get_mpdata(vnode_mount(tdvp));
2088 	if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
2089 		err = fuse_internal_access(fvp, VWRITE,
2090 			curthread, tcnp->cn_cred);
2091 		if (err)
2092 			goto out;
2093 	}
2094 	sx_xlock(&data->rename_lock);
2095 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
2096 	if (err == 0) {
2097 		if (tdvp != fdvp)
2098 			fuse_vnode_setparent(fvp, tdvp);
2099 		if (tvp != NULL)
2100 			fuse_vnode_setparent(tvp, NULL);
2101 	}
2102 	sx_unlock(&data->rename_lock);
2103 
2104 	if (tvp != NULL && tvp != fvp) {
2105 		cache_purge(tvp);
2106 	}
2107 	if (vnode_isdir(fvp)) {
2108 		if (((tvp != NULL) && vnode_isdir(tvp)) || vnode_isdir(fvp)) {
2109 			cache_purge(tdvp);
2110 		}
2111 		cache_purge(fdvp);
2112 	}
2113 out:
2114 	if (tdvp == tvp) {
2115 		vrele(tdvp);
2116 	} else {
2117 		vput(tdvp);
2118 	}
2119 	if (tvp != NULL) {
2120 		vput(tvp);
2121 	}
2122 	vrele(fdvp);
2123 	vrele(fvp);
2124 
2125 	return err;
2126 }
2127 
2128 /*
2129     struct vnop_rmdir_args {
2130 	    struct vnode *a_dvp;
2131 	    struct vnode *a_vp;
2132 	    struct componentname *a_cnp;
2133     } *ap;
2134 */
2135 static int
2136 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
2137 {
2138 	struct vnode *dvp = ap->a_dvp;
2139 	struct vnode *vp = ap->a_vp;
2140 
2141 	int err;
2142 
2143 	if (fuse_isdeadfs(vp)) {
2144 		return ENXIO;
2145 	}
2146 	if (VTOFUD(vp) == VTOFUD(dvp)) {
2147 		return EINVAL;
2148 	}
2149 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
2150 
2151 	return err;
2152 }
2153 
2154 /*
2155     struct vnop_setattr_args {
2156 	struct vnode *a_vp;
2157 	struct vattr *a_vap;
2158 	struct ucred *a_cred;
2159 	struct thread *a_td;
2160     };
2161 */
2162 static int
2163 fuse_vnop_setattr(struct vop_setattr_args *ap)
2164 {
2165 	struct vnode *vp = ap->a_vp;
2166 	struct vattr *vap = ap->a_vap;
2167 	struct ucred *cred = ap->a_cred;
2168 	struct thread *td = curthread;
2169 	struct mount *mp;
2170 	struct fuse_data *data;
2171 	struct vattr old_va;
2172 	int dataflags;
2173 	int err = 0, err2;
2174 	accmode_t accmode = 0;
2175 	bool checkperm;
2176 	bool drop_suid = false;
2177 	gid_t cr_gid;
2178 
2179 	mp = vnode_mount(vp);
2180 	data = fuse_get_mpdata(mp);
2181 	dataflags = data->dataflags;
2182 	checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
2183 	if (cred->cr_ngroups > 0)
2184 		cr_gid = cred->cr_groups[0];
2185 	else
2186 		cr_gid = 0;
2187 
2188 	if (fuse_isdeadfs(vp)) {
2189 		return ENXIO;
2190 	}
2191 
2192 	if (vap->va_uid != (uid_t)VNOVAL) {
2193 		if (checkperm) {
2194 			/* Only root may change a file's owner */
2195 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2196 			if (err) {
2197 				/* As a special case, allow the null chown */
2198 				err2 = fuse_internal_getattr(vp, &old_va, cred,
2199 					td);
2200 				if (err2)
2201 					return (err2);
2202 				if (vap->va_uid != old_va.va_uid)
2203 					return err;
2204 				else
2205 					accmode |= VADMIN;
2206 				drop_suid = true;
2207 			} else
2208 				accmode |= VADMIN;
2209 		} else
2210 			accmode |= VADMIN;
2211 	}
2212 	if (vap->va_gid != (gid_t)VNOVAL) {
2213 		if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
2214 			drop_suid = true;
2215 		if (checkperm && !groupmember(vap->va_gid, cred))
2216 		{
2217 			/*
2218 			 * Non-root users may only chgrp to one of their own
2219 			 * groups
2220 			 */
2221 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2222 			if (err) {
2223 				/* As a special case, allow the null chgrp */
2224 				err2 = fuse_internal_getattr(vp, &old_va, cred,
2225 					td);
2226 				if (err2)
2227 					return (err2);
2228 				if (vap->va_gid != old_va.va_gid)
2229 					return err;
2230 				accmode |= VADMIN;
2231 			} else
2232 				accmode |= VADMIN;
2233 		} else
2234 			accmode |= VADMIN;
2235 	}
2236 	if (vap->va_size != VNOVAL) {
2237 		switch (vp->v_type) {
2238 		case VDIR:
2239 			return (EISDIR);
2240 		case VLNK:
2241 		case VREG:
2242 			if (vfs_isrdonly(mp))
2243 				return (EROFS);
2244 			break;
2245 		default:
2246 			/*
2247 			 * According to POSIX, the result is unspecified
2248 			 * for file types other than regular files,
2249 			 * directories and shared memory objects.  We
2250 			 * don't support shared memory objects in the file
2251 			 * system, and have dubious support for truncating
2252 			 * symlinks.  Just ignore the request in other cases.
2253 			 */
2254 			return (0);
2255 		}
2256 		/* Don't set accmode.  Permission to trunc is checked upstack */
2257 	}
2258 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
2259 		if (vap->va_vaflags & VA_UTIMES_NULL)
2260 			accmode |= VWRITE;
2261 		else
2262 			accmode |= VADMIN;
2263 	}
2264 	if (drop_suid) {
2265 		if (vap->va_mode != (mode_t)VNOVAL)
2266 			vap->va_mode &= ~(S_ISUID | S_ISGID);
2267 		else {
2268 			err = fuse_internal_getattr(vp, &old_va, cred, td);
2269 			if (err)
2270 				return (err);
2271 			vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
2272 		}
2273 	}
2274 	if (vap->va_mode != (mode_t)VNOVAL) {
2275 		/* Only root may set the sticky bit on non-directories */
2276 		if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
2277 		    && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
2278 			return EFTYPE;
2279 		if (checkperm && (vap->va_mode & S_ISGID)) {
2280 			err = fuse_internal_getattr(vp, &old_va, cred, td);
2281 			if (err)
2282 				return (err);
2283 			if (!groupmember(old_va.va_gid, cred)) {
2284 				err = priv_check_cred(cred, PRIV_VFS_SETGID);
2285 				if (err)
2286 					return (err);
2287 			}
2288 		}
2289 		accmode |= VADMIN;
2290 	}
2291 
2292 	if (vfs_isrdonly(mp))
2293 		return EROFS;
2294 
2295 	if (checkperm) {
2296 		err = fuse_internal_access(vp, accmode, td, cred);
2297 	} else {
2298 		err = 0;
2299 	}
2300 	if (err)
2301 		return err;
2302 	else
2303 		return fuse_internal_setattr(vp, vap, td, cred);
2304 }
2305 
2306 /*
2307     struct vnop_strategy_args {
2308 	struct vnode *a_vp;
2309 	struct buf *a_bp;
2310     };
2311 */
2312 static int
2313 fuse_vnop_strategy(struct vop_strategy_args *ap)
2314 {
2315 	struct vnode *vp = ap->a_vp;
2316 	struct buf *bp = ap->a_bp;
2317 
2318 	if (!vp || fuse_isdeadfs(vp)) {
2319 		bp->b_ioflags |= BIO_ERROR;
2320 		bp->b_error = ENXIO;
2321 		bufdone(bp);
2322 		return 0;
2323 	}
2324 
2325 	/*
2326 	 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
2327 	 * fuse_io_strategy sets bp's error fields
2328 	 */
2329 	(void)fuse_io_strategy(vp, bp);
2330 
2331 	return 0;
2332 }
2333 
2334 /*
2335     struct vnop_symlink_args {
2336 	struct vnode *a_dvp;
2337 	struct vnode **a_vpp;
2338 	struct componentname *a_cnp;
2339 	struct vattr *a_vap;
2340 	char *a_target;
2341     };
2342 */
2343 static int
2344 fuse_vnop_symlink(struct vop_symlink_args *ap)
2345 {
2346 	struct vnode *dvp = ap->a_dvp;
2347 	struct vnode **vpp = ap->a_vpp;
2348 	struct componentname *cnp = ap->a_cnp;
2349 	const char *target = ap->a_target;
2350 
2351 	struct fuse_dispatcher fdi;
2352 
2353 	int err;
2354 	size_t len;
2355 
2356 	if (fuse_isdeadfs(dvp)) {
2357 		return ENXIO;
2358 	}
2359 	/*
2360 	 * Unlike the other creator type calls, here we have to create a message
2361 	 * where the name of the new entry comes first, and the data describing
2362 	 * the entry comes second.
2363 	 * Hence we can't rely on our handy fuse_internal_newentry() routine,
2364 	 * but put together the message manually and just call the core part.
2365 	 */
2366 
2367 	len = strlen(target) + 1;
2368 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
2369 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
2370 
2371 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
2372 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
2373 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
2374 
2375 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
2376 	fdisp_destroy(&fdi);
2377 	return err;
2378 }
2379 
2380 /*
2381     struct vnop_write_args {
2382 	struct vnode *a_vp;
2383 	struct uio *a_uio;
2384 	int  a_ioflag;
2385 	struct ucred *a_cred;
2386     };
2387 */
2388 static int
2389 fuse_vnop_write(struct vop_write_args *ap)
2390 {
2391 	struct vnode *vp = ap->a_vp;
2392 	struct uio *uio = ap->a_uio;
2393 	int ioflag = ap->a_ioflag;
2394 	struct ucred *cred = ap->a_cred;
2395 	pid_t pid = curthread->td_proc->p_pid;
2396 	struct fuse_filehandle *fufh;
2397 	int err;
2398 	bool closefufh = false, directio;
2399 
2400 	MPASS(vp->v_type == VREG || vp->v_type == VDIR);
2401 
2402 	if (fuse_isdeadfs(vp)) {
2403 		return ENXIO;
2404 	}
2405 
2406 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
2407 		ioflag |= IO_DIRECT;
2408 	}
2409 
2410 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2411 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2412 		/*
2413 		 * nfsd will do I/O without first doing VOP_OPEN.  We
2414 		 * must implicitly open the file here
2415 		 */
2416 		err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2417 		closefufh = true;
2418 	}
2419 	if (err) {
2420 		SDT_PROBE3(fusefs, , vnops, filehandles_closed, vp, uio, cred);
2421 		return err;
2422 	}
2423 
2424 	/*
2425          * Ideally, when the daemon asks for direct io at open time, the
2426          * standard file flag should be set according to this, so that would
2427          * just change the default mode, which later on could be changed via
2428          * fcntl(2).
2429          * But this doesn't work, the O_DIRECT flag gets cleared at some point
2430          * (don't know where). So to make any use of the Fuse direct_io option,
2431          * we hardwire it into the file's private data (similarly to Linux,
2432          * btw.).
2433          */
2434 	directio = (ioflag & IO_DIRECT) || !fsess_opt_datacache(vnode_mount(vp));
2435 
2436 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
2437 	if (directio) {
2438 		off_t start, end, filesize;
2439 		bool pages = (ioflag & IO_VMIO) != 0;
2440 
2441 		SDT_PROBE2(fusefs, , vnops, trace, 1, "direct write of vnode");
2442 
2443 		err = fuse_vnode_size(vp, &filesize, cred, curthread);
2444 		if (err)
2445 			goto out;
2446 
2447 		start = uio->uio_offset;
2448 		end = start + uio->uio_resid;
2449 		if (!pages) {
2450 			err = fuse_inval_buf_range(vp, filesize, start,
2451 			    end);
2452 			if (err)
2453 				goto out;
2454 		}
2455 		err = fuse_write_directbackend(vp, uio, cred, fufh,
2456 			filesize, ioflag, pages);
2457 	} else {
2458 		SDT_PROBE2(fusefs, , vnops, trace, 1,
2459 			"buffered write of vnode");
2460 		if (!fsess_opt_writeback(vnode_mount(vp)))
2461 			ioflag |= IO_SYNC;
2462 		err = fuse_write_biobackend(vp, uio, cred, fufh, ioflag, pid);
2463 	}
2464 	fuse_internal_clear_suid_on_write(vp, cred, uio->uio_td);
2465 
2466 out:
2467 	if (closefufh)
2468 		fuse_filehandle_close(vp, fufh, curthread, cred);
2469 
2470 	return (err);
2471 }
2472 
2473 static daddr_t
2474 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
2475 {
2476 	const int biosize = fuse_iosize(vp);
2477 
2478 	return (off / biosize);
2479 }
2480 
2481 static int
2482 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *blksz)
2483 {
2484 	off_t filesize;
2485 	int err;
2486 	const int biosize = fuse_iosize(vp);
2487 
2488 	err = fuse_vnode_size(vp, &filesize, NULL, NULL);
2489 	if (err) {
2490 		/* This will turn into a SIGBUS */
2491 		return (EIO);
2492 	} else if ((off_t)lbn * biosize >= filesize) {
2493 		*blksz = 0;
2494 	} else if ((off_t)(lbn + 1) * biosize > filesize) {
2495 		*blksz = filesize - (off_t)lbn *biosize;
2496 	} else {
2497 		*blksz = biosize;
2498 	}
2499 	return (0);
2500 }
2501 
2502 /*
2503     struct vnop_getpages_args {
2504 	struct vnode *a_vp;
2505 	vm_page_t *a_m;
2506 	int a_count;
2507 	int a_reqpage;
2508     };
2509 */
2510 static int
2511 fuse_vnop_getpages(struct vop_getpages_args *ap)
2512 {
2513 	struct vnode *vp = ap->a_vp;
2514 
2515 	if (!fsess_opt_mmap(vnode_mount(vp))) {
2516 		SDT_PROBE2(fusefs, , vnops, trace, 1,
2517 			"called on non-cacheable vnode??\n");
2518 		return (VM_PAGER_ERROR);
2519 	}
2520 
2521 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
2522 	    ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
2523 }
2524 
2525 static const char extattr_namespace_separator = '.';
2526 
2527 /*
2528     struct vop_getextattr_args {
2529 	struct vop_generic_args a_gen;
2530 	struct vnode *a_vp;
2531 	int a_attrnamespace;
2532 	const char *a_name;
2533 	struct uio *a_uio;
2534 	size_t *a_size;
2535 	struct ucred *a_cred;
2536 	struct thread *a_td;
2537     };
2538 */
2539 static int
2540 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2541 {
2542 	struct vnode *vp = ap->a_vp;
2543 	struct uio *uio = ap->a_uio;
2544 	struct fuse_dispatcher fdi;
2545 	struct fuse_getxattr_in *get_xattr_in;
2546 	struct fuse_getxattr_out *get_xattr_out;
2547 	struct mount *mp = vnode_mount(vp);
2548 	struct thread *td = ap->a_td;
2549 	struct ucred *cred = ap->a_cred;
2550 	char *prefix;
2551 	char *attr_str;
2552 	size_t len;
2553 	int err;
2554 
2555 	if (fuse_isdeadfs(vp))
2556 		return (ENXIO);
2557 
2558 	if (fsess_not_impl(mp, FUSE_GETXATTR))
2559 		return EOPNOTSUPP;
2560 
2561 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2562 	if (err)
2563 		return err;
2564 
2565 	/* Default to looking for user attributes. */
2566 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2567 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2568 	else
2569 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2570 
2571 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2572 	    strlen(ap->a_name) + 1;
2573 
2574 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2575 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2576 
2577 	get_xattr_in = fdi.indata;
2578 	/*
2579 	 * Check to see whether we're querying the available size or
2580 	 * issuing the actual request.  If we pass in 0, we get back struct
2581 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2582 	 * that much data, without the struct fuse_getxattr_out header.
2583 	 */
2584 	if (uio == NULL)
2585 		get_xattr_in->size = 0;
2586 	else
2587 		get_xattr_in->size = uio->uio_resid;
2588 
2589 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2590 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2591 	    ap->a_name);
2592 
2593 	err = fdisp_wait_answ(&fdi);
2594 	if (err != 0) {
2595 		if (err == ENOSYS) {
2596 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2597 			err = EOPNOTSUPP;
2598 		}
2599 		goto out;
2600 	}
2601 
2602 	get_xattr_out = fdi.answ;
2603 
2604 	if (ap->a_size != NULL)
2605 		*ap->a_size = get_xattr_out->size;
2606 
2607 	if (uio != NULL)
2608 		err = uiomove(fdi.answ, fdi.iosize, uio);
2609 
2610 out:
2611 	fdisp_destroy(&fdi);
2612 	return (err);
2613 }
2614 
2615 /*
2616     struct vop_setextattr_args {
2617 	struct vop_generic_args a_gen;
2618 	struct vnode *a_vp;
2619 	int a_attrnamespace;
2620 	const char *a_name;
2621 	struct uio *a_uio;
2622 	struct ucred *a_cred;
2623 	struct thread *a_td;
2624     };
2625 */
2626 static int
2627 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2628 {
2629 	struct vnode *vp = ap->a_vp;
2630 	struct uio *uio = ap->a_uio;
2631 	struct fuse_dispatcher fdi;
2632 	struct fuse_setxattr_in *set_xattr_in;
2633 	struct mount *mp = vnode_mount(vp);
2634 	struct thread *td = ap->a_td;
2635 	struct ucred *cred = ap->a_cred;
2636 	char *prefix;
2637 	size_t len;
2638 	char *attr_str;
2639 	int err;
2640 
2641 	if (fuse_isdeadfs(vp))
2642 		return (ENXIO);
2643 
2644 	if (fsess_not_impl(mp, FUSE_SETXATTR))
2645 		return EOPNOTSUPP;
2646 
2647 	if (vfs_isrdonly(mp))
2648 		return EROFS;
2649 
2650 	/* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2651 	if (ap->a_uio == NULL) {
2652 		/*
2653 		 * If we got here as fallback from VOP_DELETEEXTATTR, then
2654 		 * return EOPNOTSUPP.
2655 		 */
2656 		if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
2657 			return (EOPNOTSUPP);
2658 		else
2659 			return (EINVAL);
2660 	}
2661 
2662 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2663 		VWRITE);
2664 	if (err)
2665 		return err;
2666 
2667 	/* Default to looking for user attributes. */
2668 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2669 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2670 	else
2671 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2672 
2673 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2674 	    strlen(ap->a_name) + 1;
2675 
2676 	fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2677 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2678 
2679 	set_xattr_in = fdi.indata;
2680 	set_xattr_in->size = uio->uio_resid;
2681 
2682 	attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2683 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2684 	    ap->a_name);
2685 
2686 	err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2687 	    uio->uio_resid, uio);
2688 	if (err != 0) {
2689 		goto out;
2690 	}
2691 
2692 	err = fdisp_wait_answ(&fdi);
2693 
2694 	if (err == ENOSYS) {
2695 		fsess_set_notimpl(mp, FUSE_SETXATTR);
2696 		err = EOPNOTSUPP;
2697 	}
2698 	if (err == ERESTART) {
2699 		/* Can't restart after calling uiomove */
2700 		err = EINTR;
2701 	}
2702 
2703 out:
2704 	fdisp_destroy(&fdi);
2705 	return (err);
2706 }
2707 
2708 /*
2709  * The Linux / FUSE extended attribute list is simply a collection of
2710  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2711  * byte length followed by a non-NUL terminated string.  So, this allows
2712  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2713  * Linux attribute names are reported with the namespace as a prefix (e.g.
2714  * "user.attribute_name"), but in FreeBSD they are reported without the
2715  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2716  *
2717  * user.attr_name1\0user.attr_name2\0
2718  *
2719  * to:
2720  *
2721  * <num>attr_name1<num>attr_name2
2722  *
2723  * Where "<num>" is a single byte number of characters in the attribute name.
2724  *
2725  * Args:
2726  * prefix - exattr namespace prefix string
2727  * list, list_len - input list with namespace prefixes
2728  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2729  */
2730 static int
2731 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2732     char *bsd_list, int *bsd_list_len)
2733 {
2734 	int len, pos, dist_to_next, prefix_len;
2735 
2736 	pos = 0;
2737 	*bsd_list_len = 0;
2738 	prefix_len = strlen(prefix);
2739 
2740 	while (pos < list_len && list[pos] != '\0') {
2741 		dist_to_next = strlen(&list[pos]) + 1;
2742 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2743 		    list[pos + prefix_len] == extattr_namespace_separator) {
2744 			len = dist_to_next -
2745 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2746 			if (len >= EXTATTR_MAXNAMELEN)
2747 				return (ENAMETOOLONG);
2748 
2749 			bsd_list[*bsd_list_len] = len;
2750 			memcpy(&bsd_list[*bsd_list_len + 1],
2751 			    &list[pos + prefix_len +
2752 			    sizeof(extattr_namespace_separator)], len);
2753 
2754 			*bsd_list_len += len + 1;
2755 		}
2756 
2757 		pos += dist_to_next;
2758 	}
2759 
2760 	return (0);
2761 }
2762 
2763 /*
2764  * List extended attributes
2765  *
2766  * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
2767  * has a number of differences compared to its FreeBSD equivalent,
2768  * extattr_list_file:
2769  *
2770  * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
2771  *   whereas listxattr(2) only returns attributes for a single namespace
2772  * - FUSE_LISTXATTR prepends each attribute name with "namespace."
2773  * - If the provided buffer is not large enough to hold the result,
2774  *   FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
2775  *   return as many results as will fit.
2776  */
2777 /*
2778     struct vop_listextattr_args {
2779 	struct vop_generic_args a_gen;
2780 	struct vnode *a_vp;
2781 	int a_attrnamespace;
2782 	struct uio *a_uio;
2783 	size_t *a_size;
2784 	struct ucred *a_cred;
2785 	struct thread *a_td;
2786     };
2787 */
2788 static int
2789 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2790 {
2791 	struct vnode *vp = ap->a_vp;
2792 	struct uio *uio = ap->a_uio;
2793 	struct fuse_dispatcher fdi;
2794 	struct fuse_listxattr_in *list_xattr_in;
2795 	struct fuse_listxattr_out *list_xattr_out;
2796 	struct mount *mp = vnode_mount(vp);
2797 	struct thread *td = ap->a_td;
2798 	struct ucred *cred = ap->a_cred;
2799 	char *prefix;
2800 	char *bsd_list = NULL;
2801 	char *linux_list;
2802 	int bsd_list_len;
2803 	int linux_list_len;
2804 	int err;
2805 
2806 	if (fuse_isdeadfs(vp))
2807 		return (ENXIO);
2808 
2809 	if (fsess_not_impl(mp, FUSE_LISTXATTR))
2810 		return EOPNOTSUPP;
2811 
2812 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2813 	if (err)
2814 		return err;
2815 
2816 	/*
2817 	 * Add space for a NUL and the period separator if enabled.
2818 	 * Default to looking for user attributes.
2819 	 */
2820 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2821 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2822 	else
2823 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2824 
2825 	fdisp_init(&fdi, sizeof(*list_xattr_in));
2826 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2827 
2828 	/*
2829 	 * Retrieve Linux / FUSE compatible list size.
2830 	 */
2831 	list_xattr_in = fdi.indata;
2832 	list_xattr_in->size = 0;
2833 
2834 	err = fdisp_wait_answ(&fdi);
2835 	if (err != 0) {
2836 		if (err == ENOSYS) {
2837 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
2838 			err = EOPNOTSUPP;
2839 		}
2840 		goto out;
2841 	}
2842 
2843 	list_xattr_out = fdi.answ;
2844 	linux_list_len = list_xattr_out->size;
2845 	if (linux_list_len == 0) {
2846 		if (ap->a_size != NULL)
2847 			*ap->a_size = linux_list_len;
2848 		goto out;
2849 	}
2850 
2851 	/*
2852 	 * Retrieve Linux / FUSE compatible list values.
2853 	 */
2854 	fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2855 	list_xattr_in = fdi.indata;
2856 	list_xattr_in->size = linux_list_len;
2857 
2858 	err = fdisp_wait_answ(&fdi);
2859 	if (err == ERANGE) {
2860 		/*
2861 		 * Race detected.  The attribute list must've grown since the
2862 		 * first FUSE_LISTXATTR call.  Start over.  Go all the way back
2863 		 * to userland so we can process signals, if necessary, before
2864 		 * restarting.
2865 		 */
2866 		err = ERESTART;
2867 		goto out;
2868 	} else if (err != 0)
2869 		goto out;
2870 
2871 	linux_list = fdi.answ;
2872 	/* FUSE doesn't allow the server to return more data than requested */
2873 	if (fdi.iosize > linux_list_len) {
2874 		struct fuse_data *data = fuse_get_mpdata(mp);
2875 
2876 		fuse_warn(data, FSESS_WARN_LSEXTATTR_LONG,
2877 			"server returned "
2878 			"more extended attribute data than requested; "
2879 			"should've returned ERANGE instead.");
2880 	} else {
2881 		/* But returning less data is fine */
2882 		linux_list_len = fdi.iosize;
2883 	}
2884 
2885 	/*
2886 	 * Retrieve the BSD compatible list values.
2887 	 * The Linux / FUSE attribute list format isn't the same
2888 	 * as FreeBSD's format. So we need to transform it into
2889 	 * FreeBSD's format before giving it to the user.
2890 	 */
2891 	bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2892 	err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2893 	    bsd_list, &bsd_list_len);
2894 	if (err != 0)
2895 		goto out;
2896 
2897 	if (ap->a_size != NULL)
2898 		*ap->a_size = bsd_list_len;
2899 
2900 	if (uio != NULL)
2901 		err = uiomove(bsd_list, bsd_list_len, uio);
2902 
2903 out:
2904 	free(bsd_list, M_TEMP);
2905 	fdisp_destroy(&fdi);
2906 	return (err);
2907 }
2908 
2909 /*
2910     struct vop_deallocate_args {
2911 	struct vop_generic_args a_gen;
2912 	struct vnode *a_vp;
2913 	off_t *a_offset;
2914 	off_t *a_len;
2915 	int a_flags;
2916 	int a_ioflag;
2917         struct ucred *a_cred;
2918     };
2919 */
2920 static int
2921 fuse_vnop_deallocate(struct vop_deallocate_args *ap)
2922 {
2923 	struct vnode *vp = ap->a_vp;
2924 	struct mount *mp = vnode_mount(vp);
2925 	struct fuse_filehandle *fufh;
2926 	struct fuse_dispatcher fdi;
2927 	struct fuse_fallocate_in *ffi;
2928 	struct ucred *cred = ap->a_cred;
2929 	pid_t pid = curthread->td_proc->p_pid;
2930 	off_t *len = ap->a_len;
2931 	off_t *offset = ap->a_offset;
2932 	int ioflag = ap->a_ioflag;
2933 	off_t filesize;
2934 	int err;
2935 	bool closefufh = false;
2936 
2937 	if (fuse_isdeadfs(vp))
2938 		return (ENXIO);
2939 
2940 	if (vfs_isrdonly(mp))
2941 		return (EROFS);
2942 
2943 	if (fsess_not_impl(mp, FUSE_FALLOCATE))
2944 		goto fallback;
2945 
2946 	err = fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid);
2947 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
2948 		/*
2949 		 * nfsd will do I/O without first doing VOP_OPEN.  We
2950 		 * must implicitly open the file here
2951 		 */
2952 		err = fuse_filehandle_open(vp, FWRITE, &fufh, curthread, cred);
2953 		closefufh = true;
2954 	}
2955 	if (err)
2956 		return (err);
2957 
2958 	fuse_vnode_update(vp, FN_MTIMECHANGE | FN_CTIMECHANGE);
2959 
2960 	err = fuse_vnode_size(vp, &filesize, cred, curthread);
2961 	if (err)
2962 		goto out;
2963 	fuse_inval_buf_range(vp, filesize, *offset, *offset + *len);
2964 
2965 	fdisp_init(&fdi, sizeof(*ffi));
2966 	fdisp_make_vp(&fdi, FUSE_FALLOCATE, vp, curthread, cred);
2967 	ffi = fdi.indata;
2968 	ffi->fh = fufh->fh_id;
2969 	ffi->offset = *offset;
2970 	ffi->length = *len;
2971 	/*
2972 	 * FreeBSD's fspacectl is equivalent to Linux's fallocate with
2973 	 * mode == FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE
2974 	 */
2975 	ffi->mode = FUSE_FALLOC_FL_PUNCH_HOLE | FUSE_FALLOC_FL_KEEP_SIZE;
2976 	err = fdisp_wait_answ(&fdi);
2977 
2978 	if (err == ENOSYS) {
2979 		fsess_set_notimpl(mp, FUSE_FALLOCATE);
2980 		goto fallback;
2981 	} else if (err == EOPNOTSUPP) {
2982 		/*
2983 		 * The file system server does not support FUSE_FALLOCATE with
2984 		 * the supplied mode for this particular file.
2985 		 */
2986 		goto fallback;
2987 	} else if (!err) {
2988 		/*
2989 		 * Clip the returned offset to EoF.  Do it here rather than
2990 		 * before FUSE_FALLOCATE just in case the kernel's cached file
2991 		 * size is out of date.  Unfortunately, FUSE does not return
2992 		 * any information about filesize from that operation.
2993 		 */
2994 		*offset = MIN(*offset + *len, filesize);
2995 		*len = 0;
2996 		fuse_vnode_undirty_cached_timestamps(vp, false);
2997 		fuse_internal_clear_suid_on_write(vp, cred, curthread);
2998 
2999 		if (ioflag & IO_SYNC)
3000 			err = fuse_internal_fsync(vp, curthread, MNT_WAIT,
3001 			    false);
3002 	}
3003 
3004 out:
3005 	if (closefufh)
3006 		fuse_filehandle_close(vp, fufh, curthread, cred);
3007 
3008 	return (err);
3009 
3010 fallback:
3011 	if (closefufh)
3012 		fuse_filehandle_close(vp, fufh, curthread, cred);
3013 
3014 	return (vop_stddeallocate(ap));
3015 }
3016 
3017 /*
3018     struct vop_deleteextattr_args {
3019 	struct vop_generic_args a_gen;
3020 	struct vnode *a_vp;
3021 	int a_attrnamespace;
3022 	const char *a_name;
3023 	struct ucred *a_cred;
3024 	struct thread *a_td;
3025     };
3026 */
3027 static int
3028 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
3029 {
3030 	struct vnode *vp = ap->a_vp;
3031 	struct fuse_dispatcher fdi;
3032 	struct mount *mp = vnode_mount(vp);
3033 	struct thread *td = ap->a_td;
3034 	struct ucred *cred = ap->a_cred;
3035 	char *prefix;
3036 	size_t len;
3037 	char *attr_str;
3038 	int err;
3039 
3040 	if (fuse_isdeadfs(vp))
3041 		return (ENXIO);
3042 
3043 	if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
3044 		return EOPNOTSUPP;
3045 
3046 	if (vfs_isrdonly(mp))
3047 		return EROFS;
3048 
3049 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
3050 		VWRITE);
3051 	if (err)
3052 		return err;
3053 
3054 	/* Default to looking for user attributes. */
3055 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
3056 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
3057 	else
3058 		prefix = EXTATTR_NAMESPACE_USER_STRING;
3059 
3060 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
3061 	    strlen(ap->a_name) + 1;
3062 
3063 	fdisp_init(&fdi, len);
3064 	fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
3065 
3066 	attr_str = fdi.indata;
3067 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
3068 	    ap->a_name);
3069 
3070 	err = fdisp_wait_answ(&fdi);
3071 	if (err == ENOSYS) {
3072 		fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
3073 		err = EOPNOTSUPP;
3074 	}
3075 
3076 	fdisp_destroy(&fdi);
3077 	return (err);
3078 }
3079 
3080 /*
3081     struct vnop_print_args {
3082 	struct vnode *a_vp;
3083     };
3084 */
3085 static int
3086 fuse_vnop_print(struct vop_print_args *ap)
3087 {
3088 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
3089 
3090 	printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
3091 	    (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
3092 	    (uintmax_t)fvdat->nlookup,
3093 	    fvdat->flag);
3094 
3095 	return 0;
3096 }
3097 
3098 /*
3099  * Get an NFS filehandle for a FUSE file.
3100  *
3101  * This will only work for FUSE file systems that guarantee the uniqueness of
3102  * nodeid:generation, which most don't.
3103  */
3104 /*
3105 vop_vptofh {
3106 	IN struct vnode *a_vp;
3107 	IN struct fid *a_fhp;
3108 };
3109 */
3110 static int
3111 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
3112 {
3113 	struct vnode *vp = ap->a_vp;
3114 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
3115 	struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
3116 	_Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
3117 		"FUSE fid type is too big");
3118 	struct mount *mp = vnode_mount(vp);
3119 	struct fuse_data *data = fuse_get_mpdata(mp);
3120 	struct vattr va;
3121 	int err;
3122 
3123 	if (!(data->dataflags & FSESS_EXPORT_SUPPORT)) {
3124 		/* NFS requires lookups for "." and ".." */
3125 		SDT_PROBE2(fusefs, , vnops, trace, 1,
3126 			"VOP_VPTOFH without FUSE_EXPORT_SUPPORT");
3127 		return EOPNOTSUPP;
3128 	}
3129 	if ((mp->mnt_flag & MNT_EXPORTED) &&
3130 		!(data->dataflags & FSESS_NO_OPENDIR_SUPPORT))
3131 	{
3132 		/*
3133 		 * NFS is stateless, so nfsd must reopen a directory on every
3134 		 * call to VOP_READDIR, passing in the d_off field from the
3135 		 * final dirent of the previous invocation.  But without
3136 		 * FUSE_NO_OPENDIR_SUPPORT, the FUSE protocol does not
3137 		 * guarantee that d_off will be valid after a directory is
3138 		 * closed and reopened.  So prohibit exporting FUSE file
3139 		 * systems that don't set that flag.
3140 		 *
3141 		 * But userspace NFS servers don't have this problem.
3142                  */
3143 		SDT_PROBE2(fusefs, , vnops, trace, 1,
3144 			"VOP_VPTOFH without FUSE_NO_OPENDIR_SUPPORT");
3145 		return EOPNOTSUPP;
3146 	}
3147 
3148 	err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
3149 	if (err)
3150 		return err;
3151 
3152 	/*ip = VTOI(ap->a_vp);*/
3153 	/*ufhp = (struct ufid *)ap->a_fhp;*/
3154 	fhp->len = sizeof(struct fuse_fid);
3155 	fhp->nid = fvdat->nid;
3156 	if (fvdat->generation <= UINT32_MAX)
3157 		fhp->gen = fvdat->generation;
3158 	else
3159 		return EOVERFLOW;
3160 	return (0);
3161 }
3162