xref: /freebsd/sys/fs/fuse/fuse_vnops.c (revision c66ec88fed842fbaad62c30d510644ceb7bd2d71)
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_bmap_t fuse_vnop_bmap;
131 static vop_close_t fuse_fifo_close;
132 static vop_close_t fuse_vnop_close;
133 static vop_copy_file_range_t fuse_vnop_copy_file_range;
134 static vop_create_t fuse_vnop_create;
135 static vop_deleteextattr_t fuse_vnop_deleteextattr;
136 static vop_fdatasync_t fuse_vnop_fdatasync;
137 static vop_fsync_t fuse_vnop_fsync;
138 static vop_getattr_t fuse_vnop_getattr;
139 static vop_getextattr_t fuse_vnop_getextattr;
140 static vop_inactive_t fuse_vnop_inactive;
141 static vop_ioctl_t fuse_vnop_ioctl;
142 static vop_link_t fuse_vnop_link;
143 static vop_listextattr_t fuse_vnop_listextattr;
144 static vop_lookup_t fuse_vnop_lookup;
145 static vop_mkdir_t fuse_vnop_mkdir;
146 static vop_mknod_t fuse_vnop_mknod;
147 static vop_open_t fuse_vnop_open;
148 static vop_pathconf_t fuse_vnop_pathconf;
149 static vop_read_t fuse_vnop_read;
150 static vop_readdir_t fuse_vnop_readdir;
151 static vop_readlink_t fuse_vnop_readlink;
152 static vop_reclaim_t fuse_vnop_reclaim;
153 static vop_remove_t fuse_vnop_remove;
154 static vop_rename_t fuse_vnop_rename;
155 static vop_rmdir_t fuse_vnop_rmdir;
156 static vop_setattr_t fuse_vnop_setattr;
157 static vop_setextattr_t fuse_vnop_setextattr;
158 static vop_strategy_t fuse_vnop_strategy;
159 static vop_symlink_t fuse_vnop_symlink;
160 static vop_write_t fuse_vnop_write;
161 static vop_getpages_t fuse_vnop_getpages;
162 static vop_print_t fuse_vnop_print;
163 static vop_vptofh_t fuse_vnop_vptofh;
164 
165 struct vop_vector fuse_fifoops = {
166 	.vop_default =		&fifo_specops,
167 	.vop_access =		fuse_vnop_access,
168 	.vop_close =		fuse_fifo_close,
169 	.vop_fsync =		fuse_vnop_fsync,
170 	.vop_getattr =		fuse_vnop_getattr,
171 	.vop_inactive =		fuse_vnop_inactive,
172 	.vop_pathconf =		fuse_vnop_pathconf,
173 	.vop_print =		fuse_vnop_print,
174 	.vop_read =		VOP_PANIC,
175 	.vop_reclaim =		fuse_vnop_reclaim,
176 	.vop_setattr =		fuse_vnop_setattr,
177 	.vop_write =		VOP_PANIC,
178 	.vop_vptofh =		fuse_vnop_vptofh,
179 };
180 VFS_VOP_VECTOR_REGISTER(fuse_fifoops);
181 
182 struct vop_vector fuse_vnops = {
183 	.vop_allocate =	VOP_EINVAL,
184 	.vop_default = &default_vnodeops,
185 	.vop_access = fuse_vnop_access,
186 	.vop_advlock = fuse_vnop_advlock,
187 	.vop_bmap = fuse_vnop_bmap,
188 	.vop_close = fuse_vnop_close,
189 	.vop_copy_file_range = fuse_vnop_copy_file_range,
190 	.vop_create = fuse_vnop_create,
191 	.vop_deleteextattr = fuse_vnop_deleteextattr,
192 	.vop_fsync = fuse_vnop_fsync,
193 	.vop_fdatasync = fuse_vnop_fdatasync,
194 	.vop_getattr = fuse_vnop_getattr,
195 	.vop_getextattr = fuse_vnop_getextattr,
196 	.vop_inactive = fuse_vnop_inactive,
197 	.vop_ioctl = fuse_vnop_ioctl,
198 	.vop_link = fuse_vnop_link,
199 	.vop_listextattr = fuse_vnop_listextattr,
200 	.vop_lookup = fuse_vnop_lookup,
201 	.vop_mkdir = fuse_vnop_mkdir,
202 	.vop_mknod = fuse_vnop_mknod,
203 	.vop_open = fuse_vnop_open,
204 	.vop_pathconf = fuse_vnop_pathconf,
205 	/*
206 	 * TODO: implement vop_poll after upgrading to protocol 7.21.
207 	 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
208 	 * 7.21, which adds the ability for the client to choose which poll
209 	 * events it wants, and for a client to deregister a file handle
210 	 */
211 	.vop_read = fuse_vnop_read,
212 	.vop_readdir = fuse_vnop_readdir,
213 	.vop_readlink = fuse_vnop_readlink,
214 	.vop_reclaim = fuse_vnop_reclaim,
215 	.vop_remove = fuse_vnop_remove,
216 	.vop_rename = fuse_vnop_rename,
217 	.vop_rmdir = fuse_vnop_rmdir,
218 	.vop_setattr = fuse_vnop_setattr,
219 	.vop_setextattr = fuse_vnop_setextattr,
220 	.vop_strategy = fuse_vnop_strategy,
221 	.vop_symlink = fuse_vnop_symlink,
222 	.vop_write = fuse_vnop_write,
223 	.vop_getpages = fuse_vnop_getpages,
224 	.vop_print = fuse_vnop_print,
225 	.vop_vptofh = fuse_vnop_vptofh,
226 };
227 VFS_VOP_VECTOR_REGISTER(fuse_vnops);
228 
229 uma_zone_t fuse_pbuf_zone;
230 
231 /* Check permission for extattr operations, much like extattr_check_cred */
232 static int
233 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
234 	struct thread *td, accmode_t accmode)
235 {
236 	struct mount *mp = vnode_mount(vp);
237 	struct fuse_data *data = fuse_get_mpdata(mp);
238 	int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
239 
240 	/*
241 	 * Kernel-invoked always succeeds.
242 	 */
243 	if (cred == NOCRED)
244 		return (0);
245 
246 	/*
247 	 * Do not allow privileged processes in jail to directly manipulate
248 	 * system attributes.
249 	 */
250 	switch (ns) {
251 	case EXTATTR_NAMESPACE_SYSTEM:
252 		if (default_permissions) {
253 			return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
254 		}
255 		return (0);
256 	case EXTATTR_NAMESPACE_USER:
257 		if (default_permissions) {
258 			return (fuse_internal_access(vp, accmode, td, cred));
259 		}
260 		return (0);
261 	default:
262 		return (EPERM);
263 	}
264 }
265 
266 /* Get a filehandle for a directory */
267 static int
268 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
269 	struct ucred *cred, pid_t pid)
270 {
271 	if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
272 		return 0;
273 	return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
274 }
275 
276 /* Send FUSE_FLUSH for this vnode */
277 static int
278 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
279 {
280 	struct fuse_flush_in *ffi;
281 	struct fuse_filehandle *fufh;
282 	struct fuse_dispatcher fdi;
283 	struct thread *td = curthread;
284 	struct mount *mp = vnode_mount(vp);
285 	int err;
286 
287 	if (fsess_not_impl(vnode_mount(vp), FUSE_FLUSH))
288 		return 0;
289 
290 	err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
291 	if (err)
292 		return err;
293 
294 	fdisp_init(&fdi, sizeof(*ffi));
295 	fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
296 	ffi = fdi.indata;
297 	ffi->fh = fufh->fh_id;
298 	/*
299 	 * If the file has a POSIX lock then we're supposed to set lock_owner.
300 	 * If not, then lock_owner is undefined.  So we may as well always set
301 	 * it.
302 	 */
303 	ffi->lock_owner = td->td_proc->p_pid;
304 
305 	err = fdisp_wait_answ(&fdi);
306 	if (err == ENOSYS) {
307 		fsess_set_notimpl(mp, FUSE_FLUSH);
308 		err = 0;
309 	}
310 	fdisp_destroy(&fdi);
311 	return err;
312 }
313 
314 /* Close wrapper for fifos.  */
315 static int
316 fuse_fifo_close(struct vop_close_args *ap)
317 {
318 	return (fifo_specops.vop_close(ap));
319 }
320 
321 /* Send FUSE_LSEEK for this node */
322 static int
323 fuse_vnop_do_lseek(struct vnode *vp, struct thread *td, struct ucred *cred,
324 	pid_t pid, off_t *offp, int whence)
325 {
326 	struct fuse_dispatcher fdi;
327 	struct fuse_filehandle *fufh;
328 	struct fuse_lseek_in *flsi;
329 	struct fuse_lseek_out *flso;
330 	struct mount *mp = vnode_mount(vp);
331 	int err;
332 
333 	ASSERT_VOP_LOCKED(vp, __func__);
334 
335 	err = fuse_filehandle_getrw(vp, FREAD, &fufh, cred, pid);
336 	if (err)
337 		return (err);
338 	fdisp_init(&fdi, sizeof(*flsi));
339 	fdisp_make_vp(&fdi, FUSE_LSEEK, vp, td, cred);
340 	flsi = fdi.indata;
341 	flsi->fh = fufh->fh_id;
342 	flsi->offset = *offp;
343 	flsi->whence = whence;
344 	err = fdisp_wait_answ(&fdi);
345 	if (err == ENOSYS) {
346 		fsess_set_notimpl(mp, FUSE_LSEEK);
347 	} else if (err == 0) {
348 		fsess_set_impl(mp, FUSE_LSEEK);
349 		flso = fdi.answ;
350 		*offp = flso->offset;
351 	}
352 	fdisp_destroy(&fdi);
353 
354 	return (err);
355 }
356 
357 /*
358     struct vnop_access_args {
359 	struct vnode *a_vp;
360 #if VOP_ACCESS_TAKES_ACCMODE_T
361 	accmode_t a_accmode;
362 #else
363 	int a_mode;
364 #endif
365 	struct ucred *a_cred;
366 	struct thread *a_td;
367     };
368 */
369 static int
370 fuse_vnop_access(struct vop_access_args *ap)
371 {
372 	struct vnode *vp = ap->a_vp;
373 	int accmode = ap->a_accmode;
374 	struct ucred *cred = ap->a_cred;
375 
376 	struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
377 
378 	int err;
379 
380 	if (fuse_isdeadfs(vp)) {
381 		if (vnode_isvroot(vp)) {
382 			return 0;
383 		}
384 		return ENXIO;
385 	}
386 	if (!(data->dataflags & FSESS_INITED)) {
387 		if (vnode_isvroot(vp)) {
388 			if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
389 			    (fuse_match_cred(data->daemoncred, cred) == 0)) {
390 				return 0;
391 			}
392 		}
393 		return EBADF;
394 	}
395 	if (vnode_islnk(vp)) {
396 		return 0;
397 	}
398 
399 	err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
400 	return err;
401 }
402 
403 /*
404  * struct vop_advlock_args {
405  *	struct vop_generic_args a_gen;
406  *	struct vnode *a_vp;
407  *	void *a_id;
408  *	int a_op;
409  *	struct flock *a_fl;
410  *	int a_flags;
411  * }
412  */
413 static int
414 fuse_vnop_advlock(struct vop_advlock_args *ap)
415 {
416 	struct vnode *vp = ap->a_vp;
417 	struct flock *fl = ap->a_fl;
418 	struct thread *td = curthread;
419 	struct ucred *cred = td->td_ucred;
420 	pid_t pid = td->td_proc->p_pid;
421 	struct fuse_filehandle *fufh;
422 	struct fuse_dispatcher fdi;
423 	struct fuse_lk_in *fli;
424 	struct fuse_lk_out *flo;
425 	enum fuse_opcode op;
426 	int dataflags, err;
427 	int flags = ap->a_flags;
428 
429 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
430 
431 	if (fuse_isdeadfs(vp)) {
432 		return ENXIO;
433 	}
434 
435 	switch(ap->a_op) {
436 	case F_GETLK:
437 		op = FUSE_GETLK;
438 		break;
439 	case F_SETLK:
440 		op = FUSE_SETLK;
441 		break;
442 	case F_SETLKW:
443 		op = FUSE_SETLKW;
444 		break;
445 	default:
446 		return EINVAL;
447 	}
448 
449 	if (!(dataflags & FSESS_POSIX_LOCKS))
450 		return vop_stdadvlock(ap);
451 	/* FUSE doesn't properly support flock until protocol 7.17 */
452 	if (flags & F_FLOCK)
453 		return vop_stdadvlock(ap);
454 
455 	vn_lock(vp, LK_SHARED | LK_RETRY);
456 
457 	err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
458 	if (err)
459 		goto out;
460 
461 	fdisp_init(&fdi, sizeof(*fli));
462 
463 	fdisp_make_vp(&fdi, op, vp, td, cred);
464 	fli = fdi.indata;
465 	fli->fh = fufh->fh_id;
466 	fli->owner = fl->l_pid;
467 	fli->lk.start = fl->l_start;
468 	if (fl->l_len != 0)
469 		fli->lk.end = fl->l_start + fl->l_len - 1;
470 	else
471 		fli->lk.end = INT64_MAX;
472 	fli->lk.type = fl->l_type;
473 	fli->lk.pid = fl->l_pid;
474 
475 	err = fdisp_wait_answ(&fdi);
476 	fdisp_destroy(&fdi);
477 
478 	if (err == 0 && op == FUSE_GETLK) {
479 		flo = fdi.answ;
480 		fl->l_type = flo->lk.type;
481 		fl->l_pid = flo->lk.pid;
482 		if (flo->lk.type != F_UNLCK) {
483 			fl->l_start = flo->lk.start;
484 			if (flo->lk.end == INT64_MAX)
485 				fl->l_len = 0;
486 			else
487 				fl->l_len = flo->lk.end - flo->lk.start + 1;
488 			fl->l_start = flo->lk.start;
489 		}
490 	}
491 
492 out:
493 	VOP_UNLOCK(vp);
494 	return err;
495 }
496 
497 /* {
498 	struct vnode *a_vp;
499 	daddr_t a_bn;
500 	struct bufobj **a_bop;
501 	daddr_t *a_bnp;
502 	int *a_runp;
503 	int *a_runb;
504 } */
505 static int
506 fuse_vnop_bmap(struct vop_bmap_args *ap)
507 {
508 	struct vnode *vp = ap->a_vp;
509 	struct bufobj **bo = ap->a_bop;
510 	struct thread *td = curthread;
511 	struct mount *mp;
512 	struct fuse_dispatcher fdi;
513 	struct fuse_bmap_in *fbi;
514 	struct fuse_bmap_out *fbo;
515 	struct fuse_data *data;
516 	uint64_t biosize;
517 	off_t filesize;
518 	daddr_t lbn = ap->a_bn;
519 	daddr_t *pbn = ap->a_bnp;
520 	int *runp = ap->a_runp;
521 	int *runb = ap->a_runb;
522 	int error = 0;
523 	int maxrun;
524 
525 	if (fuse_isdeadfs(vp)) {
526 		return ENXIO;
527 	}
528 
529 	mp = vnode_mount(vp);
530 	data = fuse_get_mpdata(mp);
531 	biosize = fuse_iosize(vp);
532 	maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
533 		data->max_readahead_blocks);
534 
535 	if (bo != NULL)
536 		*bo = &vp->v_bufobj;
537 
538 	/*
539 	 * The FUSE_BMAP operation does not include the runp and runb
540 	 * variables, so we must guess.  Report nonzero contiguous runs so
541 	 * cluster_read will combine adjacent reads.  It's worthwhile to reduce
542 	 * upcalls even if we don't know the true physical layout of the file.
543 	 *
544 	 * FUSE file systems may opt out of read clustering in two ways:
545 	 * * mounting with -onoclusterr
546 	 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
547 	 */
548 	if (runb != NULL)
549 		*runb = MIN(lbn, maxrun);
550 	if (runp != NULL) {
551 		error = fuse_vnode_size(vp, &filesize, td->td_ucred, td);
552 		if (error == 0)
553 			*runp = MIN(MAX(0, filesize / (off_t)biosize - lbn - 1),
554 				    maxrun);
555 		else
556 			*runp = 0;
557 	}
558 
559 	if (fsess_maybe_impl(mp, FUSE_BMAP)) {
560 		fdisp_init(&fdi, sizeof(*fbi));
561 		fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
562 		fbi = fdi.indata;
563 		fbi->block = lbn;
564 		fbi->blocksize = biosize;
565 		error = fdisp_wait_answ(&fdi);
566 		if (error == ENOSYS) {
567 			fdisp_destroy(&fdi);
568 			fsess_set_notimpl(mp, FUSE_BMAP);
569 			error = 0;
570 		} else {
571 			fbo = fdi.answ;
572 			if (error == 0 && pbn != NULL)
573 				*pbn = fbo->block;
574 			fdisp_destroy(&fdi);
575 			return error;
576 		}
577 	}
578 
579 	/* If the daemon doesn't support BMAP, make up a sensible default */
580 	if (pbn != NULL)
581 		*pbn = lbn * btodb(biosize);
582 	return (error);
583 }
584 
585 /*
586     struct vop_close_args {
587 	struct vnode *a_vp;
588 	int  a_fflag;
589 	struct ucred *a_cred;
590 	struct thread *a_td;
591     };
592 */
593 static int
594 fuse_vnop_close(struct vop_close_args *ap)
595 {
596 	struct vnode *vp = ap->a_vp;
597 	struct ucred *cred = ap->a_cred;
598 	int fflag = ap->a_fflag;
599 	struct thread *td = ap->a_td;
600 	pid_t pid = td->td_proc->p_pid;
601 	int err = 0;
602 
603 	if (fuse_isdeadfs(vp))
604 		return 0;
605 	if (vnode_isdir(vp))
606 		return 0;
607 	if (fflag & IO_NDELAY)
608 		return 0;
609 
610 	err = fuse_flush(vp, cred, pid, fflag);
611 	/* TODO: close the file handle, if we're sure it's no longer used */
612 	if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
613 		fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
614 	}
615 	return err;
616 }
617 
618 /*
619    struct vop_copy_file_range_args {
620 	struct vop_generic_args a_gen;
621 	struct vnode *a_invp;
622 	off_t *a_inoffp;
623 	struct vnode *a_outvp;
624 	off_t *a_outoffp;
625 	size_t *a_lenp;
626 	unsigned int a_flags;
627 	struct ucred *a_incred;
628 	struct ucred *a_outcred;
629 	struct thread *a_fsizetd;
630 }
631  */
632 static int
633 fuse_vnop_copy_file_range(struct vop_copy_file_range_args *ap)
634 {
635 	struct vnode *invp = ap->a_invp;
636 	struct vnode *outvp = ap->a_outvp;
637 	struct mount *mp = vnode_mount(invp);
638 	struct fuse_dispatcher fdi;
639 	struct fuse_filehandle *infufh, *outfufh;
640 	struct fuse_copy_file_range_in *fcfri;
641 	struct ucred *incred = ap->a_incred;
642 	struct ucred *outcred = ap->a_outcred;
643 	struct fuse_write_out *fwo;
644 	struct thread *td;
645 	struct uio io;
646 	pid_t pid;
647 	int err;
648 
649 	if (mp != vnode_mount(outvp))
650 		goto fallback;
651 
652 	if (incred->cr_uid != outcred->cr_uid)
653 		goto fallback;
654 
655 	if (incred->cr_groups[0] != outcred->cr_groups[0])
656 		goto fallback;
657 
658 	if (fsess_not_impl(mp, FUSE_COPY_FILE_RANGE))
659 		goto fallback;
660 
661 	if (ap->a_fsizetd == NULL)
662 		td = curthread;
663 	else
664 		td = ap->a_fsizetd;
665 	pid = td->td_proc->p_pid;
666 
667 	/* Lock both vnodes, avoiding risk of deadlock. */
668 	do {
669 		err = vn_lock(outvp, LK_EXCLUSIVE);
670 		if (invp == outvp)
671 			break;
672 		if (err == 0) {
673 			err = vn_lock(invp, LK_SHARED | LK_NOWAIT);
674 			if (err == 0)
675 				break;
676 			VOP_UNLOCK(outvp);
677 			err = vn_lock(invp, LK_SHARED);
678 			if (err == 0)
679 				VOP_UNLOCK(invp);
680 		}
681 	} while (err == 0);
682 	if (err != 0)
683 		return (err);
684 
685 	err = fuse_filehandle_getrw(invp, FREAD, &infufh, incred, pid);
686 	if (err)
687 		goto unlock;
688 
689 	err = fuse_filehandle_getrw(outvp, FWRITE, &outfufh, outcred, pid);
690 	if (err)
691 		goto unlock;
692 
693 	if (ap->a_fsizetd) {
694 		io.uio_offset = *ap->a_outoffp;
695 		io.uio_resid = *ap->a_lenp;
696 		err = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd);
697 		if (err)
698 			goto unlock;
699 	}
700 
701 	fdisp_init(&fdi, sizeof(*fcfri));
702 	fdisp_make_vp(&fdi, FUSE_COPY_FILE_RANGE, invp, td, incred);
703 	fcfri = fdi.indata;
704 	fcfri->fh_in = infufh->fh_id;
705 	fcfri->off_in = *ap->a_inoffp;
706 	fcfri->nodeid_out = VTOI(outvp);
707 	fcfri->fh_out = outfufh->fh_id;
708 	fcfri->off_out = *ap->a_outoffp;
709 	fcfri->len = *ap->a_lenp;
710 	fcfri->flags = 0;
711 
712 	err = fdisp_wait_answ(&fdi);
713 	if (err == 0) {
714 		fwo = fdi.answ;
715 		*ap->a_lenp = fwo->size;
716 		*ap->a_inoffp += fwo->size;
717 		*ap->a_outoffp += fwo->size;
718 		fuse_internal_clear_suid_on_write(outvp, outcred, td);
719 	}
720 	fdisp_destroy(&fdi);
721 
722 unlock:
723 	if (invp != outvp)
724 		VOP_UNLOCK(invp);
725 	VOP_UNLOCK(outvp);
726 
727 	if (err == ENOSYS) {
728 		fsess_set_notimpl(mp, FUSE_COPY_FILE_RANGE);
729 fallback:
730 		err = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
731 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
732 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
733 	}
734 
735 	return (err);
736 }
737 
738 static void
739 fdisp_make_mknod_for_fallback(
740 	struct fuse_dispatcher *fdip,
741 	struct componentname *cnp,
742 	struct vnode *dvp,
743 	uint64_t parentnid,
744 	struct thread *td,
745 	struct ucred *cred,
746 	mode_t mode,
747 	enum fuse_opcode *op)
748 {
749 	struct fuse_mknod_in *fmni;
750 
751 	fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
752 	*op = FUSE_MKNOD;
753 	fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
754 	fmni = fdip->indata;
755 	fmni->mode = mode;
756 	fmni->rdev = 0;
757 	memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
758 	    cnp->cn_namelen);
759 	((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
760 }
761 /*
762     struct vnop_create_args {
763 	struct vnode *a_dvp;
764 	struct vnode **a_vpp;
765 	struct componentname *a_cnp;
766 	struct vattr *a_vap;
767     };
768 */
769 static int
770 fuse_vnop_create(struct vop_create_args *ap)
771 {
772 	struct vnode *dvp = ap->a_dvp;
773 	struct vnode **vpp = ap->a_vpp;
774 	struct componentname *cnp = ap->a_cnp;
775 	struct vattr *vap = ap->a_vap;
776 	struct thread *td = cnp->cn_thread;
777 	struct ucred *cred = cnp->cn_cred;
778 
779 	struct fuse_data *data;
780 	struct fuse_create_in *fci;
781 	struct fuse_entry_out *feo;
782 	struct fuse_open_out *foo;
783 	struct fuse_dispatcher fdi, fdi2;
784 	struct fuse_dispatcher *fdip = &fdi;
785 	struct fuse_dispatcher *fdip2 = NULL;
786 
787 	int err;
788 
789 	struct mount *mp = vnode_mount(dvp);
790 	data = fuse_get_mpdata(mp);
791 	uint64_t parentnid = VTOFUD(dvp)->nid;
792 	mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
793 	enum fuse_opcode op;
794 	int flags;
795 
796 	if (fuse_isdeadfs(dvp))
797 		return ENXIO;
798 
799 	/* FUSE expects sockets to be created with FUSE_MKNOD */
800 	if (vap->va_type == VSOCK)
801 		return fuse_internal_mknod(dvp, vpp, cnp, vap);
802 
803 	/*
804 	 * VOP_CREATE doesn't tell us the open(2) flags, so we guess.  Only a
805 	 * writable mode makes sense, and we might as well include readability
806 	 * too.
807 	 */
808 	flags = O_RDWR;
809 
810 	bzero(&fdi, sizeof(fdi));
811 
812 	if (vap->va_type != VREG)
813 		return (EINVAL);
814 
815 	if (fsess_not_impl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
816 		/* Fallback to FUSE_MKNOD/FUSE_OPEN */
817 		fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
818 			cred, mode, &op);
819 	} else {
820 		/* Use FUSE_CREATE */
821 		size_t insize;
822 
823 		op = FUSE_CREATE;
824 		fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
825 		fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
826 		fci = fdip->indata;
827 		fci->mode = mode;
828 		fci->flags = O_CREAT | flags;
829 		if (fuse_libabi_geq(data, 7, 12)) {
830 			insize = sizeof(*fci);
831 			fci->umask = td->td_proc->p_pd->pd_cmask;
832 		} else {
833 			insize = sizeof(struct fuse_open_in);
834 		}
835 
836 		memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
837 		    cnp->cn_namelen);
838 		((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
839 	}
840 
841 	err = fdisp_wait_answ(fdip);
842 
843 	if (err) {
844 		if (err == ENOSYS && op == FUSE_CREATE) {
845 			fsess_set_notimpl(mp, FUSE_CREATE);
846 			fdisp_destroy(fdip);
847 			fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
848 				parentnid, td, cred, mode, &op);
849 			err = fdisp_wait_answ(fdip);
850 		}
851 		if (err)
852 			goto out;
853 	}
854 
855 	feo = fdip->answ;
856 
857 	if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
858 		goto out;
859 	}
860 
861 	if (op == FUSE_CREATE) {
862 		foo = (struct fuse_open_out*)(feo + 1);
863 	} else {
864 		/* Issue a separate FUSE_OPEN */
865 		struct fuse_open_in *foi;
866 
867 		fdip2 = &fdi2;
868 		fdisp_init(fdip2, sizeof(*foi));
869 		fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
870 			cred);
871 		foi = fdip2->indata;
872 		foi->flags = flags;
873 		err = fdisp_wait_answ(fdip2);
874 		if (err)
875 			goto out;
876 		foo = fdip2->answ;
877 	}
878 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
879 	if (err) {
880 		struct fuse_release_in *fri;
881 		uint64_t nodeid = feo->nodeid;
882 		uint64_t fh_id = foo->fh;
883 
884 		fdisp_init(fdip, sizeof(*fri));
885 		fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
886 		fri = fdip->indata;
887 		fri->fh = fh_id;
888 		fri->flags = flags;
889 		fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
890 		fuse_insert_message(fdip->tick, false);
891 		goto out;
892 	}
893 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
894 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
895 		feo->attr_valid_nsec, NULL);
896 
897 	fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
898 	fuse_vnode_open(*vpp, foo->open_flags, td);
899 	/*
900 	 * Purge the parent's attribute cache because the daemon should've
901 	 * updated its mtime and ctime
902 	 */
903 	fuse_vnode_clear_attr_cache(dvp);
904 	cache_purge_negative(dvp);
905 
906 out:
907 	if (fdip2)
908 		fdisp_destroy(fdip2);
909 	fdisp_destroy(fdip);
910 	return err;
911 }
912 
913 /*
914     struct vnop_fdatasync_args {
915 	struct vop_generic_args a_gen;
916 	struct vnode * a_vp;
917 	struct thread * a_td;
918     };
919 */
920 static int
921 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
922 {
923 	struct vnode *vp = ap->a_vp;
924 	struct thread *td = ap->a_td;
925 	int waitfor = MNT_WAIT;
926 
927 	int err = 0;
928 
929 	if (fuse_isdeadfs(vp)) {
930 		return 0;
931 	}
932 	if ((err = vop_stdfdatasync_buf(ap)))
933 		return err;
934 
935 	return fuse_internal_fsync(vp, td, waitfor, true);
936 }
937 
938 /*
939     struct vnop_fsync_args {
940 	struct vop_generic_args a_gen;
941 	struct vnode * a_vp;
942 	int  a_waitfor;
943 	struct thread * a_td;
944     };
945 */
946 static int
947 fuse_vnop_fsync(struct vop_fsync_args *ap)
948 {
949 	struct vnode *vp = ap->a_vp;
950 	struct thread *td = ap->a_td;
951 	int waitfor = ap->a_waitfor;
952 	int err = 0;
953 
954 	if (fuse_isdeadfs(vp)) {
955 		return 0;
956 	}
957 	if ((err = vop_stdfsync(ap)))
958 		return err;
959 
960 	return fuse_internal_fsync(vp, td, waitfor, false);
961 }
962 
963 /*
964     struct vnop_getattr_args {
965 	struct vnode *a_vp;
966 	struct vattr *a_vap;
967 	struct ucred *a_cred;
968 	struct thread *a_td;
969     };
970 */
971 static int
972 fuse_vnop_getattr(struct vop_getattr_args *ap)
973 {
974 	struct vnode *vp = ap->a_vp;
975 	struct vattr *vap = ap->a_vap;
976 	struct ucred *cred = ap->a_cred;
977 	struct thread *td = curthread;
978 
979 	int err = 0;
980 	int dataflags;
981 
982 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
983 
984 	/* Note that we are not bailing out on a dead file system just yet. */
985 
986 	if (!(dataflags & FSESS_INITED)) {
987 		if (!vnode_isvroot(vp)) {
988 			fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
989 			err = ENOTCONN;
990 			return err;
991 		} else {
992 			goto fake;
993 		}
994 	}
995 	err = fuse_internal_getattr(vp, vap, cred, td);
996 	if (err == ENOTCONN && vnode_isvroot(vp)) {
997 		/* see comment in fuse_vfsop_statfs() */
998 		goto fake;
999 	} else {
1000 		return err;
1001 	}
1002 
1003 fake:
1004 	bzero(vap, sizeof(*vap));
1005 	vap->va_type = vnode_vtype(vp);
1006 
1007 	return 0;
1008 }
1009 
1010 /*
1011     struct vnop_inactive_args {
1012 	struct vnode *a_vp;
1013     };
1014 */
1015 static int
1016 fuse_vnop_inactive(struct vop_inactive_args *ap)
1017 {
1018 	struct vnode *vp = ap->a_vp;
1019 	struct thread *td = curthread;
1020 
1021 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1022 	struct fuse_filehandle *fufh, *fufh_tmp;
1023 
1024 	int need_flush = 1;
1025 
1026 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1027 		if (need_flush && vp->v_type == VREG) {
1028 			if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
1029 				fuse_vnode_savesize(vp, NULL, 0);
1030 			}
1031 			if ((fvdat->flag & FN_REVOKED) != 0)
1032 				fuse_io_invalbuf(vp, td);
1033 			else
1034 				fuse_io_flushbuf(vp, MNT_WAIT, td);
1035 			need_flush = 0;
1036 		}
1037 		fuse_filehandle_close(vp, fufh, td, NULL);
1038 	}
1039 
1040 	if ((fvdat->flag & FN_REVOKED) != 0)
1041 		vrecycle(vp);
1042 
1043 	return 0;
1044 }
1045 
1046 /*
1047     struct vnop_ioctl_args {
1048 	struct vnode *a_vp;
1049 	u_long a_command;
1050 	caddr_t a_data;
1051 	int a_fflag;
1052 	struct ucred *a_cred;
1053 	struct thread *a_td;
1054     };
1055 */
1056 static int
1057 fuse_vnop_ioctl(struct vop_ioctl_args *ap)
1058 {
1059 	struct vnode *vp = ap->a_vp;
1060 	struct mount *mp = vnode_mount(vp);
1061 	struct ucred *cred = ap->a_cred;
1062 	off_t *offp;
1063 	pid_t pid = ap->a_td->td_proc->p_pid;
1064 	int err;
1065 
1066 	switch (ap->a_command) {
1067 	case FIOSEEKDATA:
1068 	case FIOSEEKHOLE:
1069 		/* Call FUSE_LSEEK, if we can, or fall back to vop_stdioctl */
1070 		if (fsess_maybe_impl(mp, FUSE_LSEEK)) {
1071 			int whence;
1072 
1073 			offp = ap->a_data;
1074 			if (ap->a_command == FIOSEEKDATA)
1075 				whence = SEEK_DATA;
1076 			else
1077 				whence = SEEK_HOLE;
1078 
1079 			vn_lock(vp, LK_SHARED | LK_RETRY);
1080 			err = fuse_vnop_do_lseek(vp, ap->a_td, cred, pid, offp,
1081 			    whence);
1082 			VOP_UNLOCK(vp);
1083 		}
1084 		if (fsess_not_impl(mp, FUSE_LSEEK))
1085 			err = vop_stdioctl(ap);
1086 		break;
1087 	default:
1088 		/* TODO: implement FUSE_IOCTL */
1089 		err = ENOTTY;
1090 		break;
1091 	}
1092 	return (err);
1093 }
1094 
1095 
1096 /*
1097     struct vnop_link_args {
1098 	struct vnode *a_tdvp;
1099 	struct vnode *a_vp;
1100 	struct componentname *a_cnp;
1101     };
1102 */
1103 static int
1104 fuse_vnop_link(struct vop_link_args *ap)
1105 {
1106 	struct vnode *vp = ap->a_vp;
1107 	struct vnode *tdvp = ap->a_tdvp;
1108 	struct componentname *cnp = ap->a_cnp;
1109 
1110 	struct vattr *vap = VTOVA(vp);
1111 
1112 	struct fuse_dispatcher fdi;
1113 	struct fuse_entry_out *feo;
1114 	struct fuse_link_in fli;
1115 
1116 	int err;
1117 
1118 	if (fuse_isdeadfs(vp)) {
1119 		return ENXIO;
1120 	}
1121 	if (vnode_mount(tdvp) != vnode_mount(vp)) {
1122 		return EXDEV;
1123 	}
1124 
1125 	/*
1126 	 * This is a seatbelt check to protect naive userspace filesystems from
1127 	 * themselves and the limitations of the FUSE IPC protocol.  If a
1128 	 * filesystem does not allow attribute caching, assume it is capable of
1129 	 * validating that nlink does not overflow.
1130 	 */
1131 	if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
1132 		return EMLINK;
1133 	fli.oldnodeid = VTOI(vp);
1134 
1135 	fdisp_init(&fdi, 0);
1136 	fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
1137 	    FUSE_LINK, &fli, sizeof(fli), &fdi);
1138 	if ((err = fdisp_wait_answ(&fdi))) {
1139 		goto out;
1140 	}
1141 	feo = fdi.answ;
1142 
1143 	err = fuse_internal_checkentry(feo, vnode_vtype(vp));
1144 	if (!err) {
1145 		/*
1146 		 * Purge the parent's attribute cache because the daemon
1147 		 * should've updated its mtime and ctime
1148 		 */
1149 		fuse_vnode_clear_attr_cache(tdvp);
1150 		fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
1151 			feo->attr_valid_nsec, NULL);
1152 	}
1153 out:
1154 	fdisp_destroy(&fdi);
1155 	return err;
1156 }
1157 
1158 struct fuse_lookup_alloc_arg {
1159 	struct fuse_entry_out *feo;
1160 	struct componentname *cnp;
1161 	uint64_t nid;
1162 	enum vtype vtyp;
1163 };
1164 
1165 /* Callback for vn_get_ino */
1166 static int
1167 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
1168 {
1169 	struct fuse_lookup_alloc_arg *flaa = arg;
1170 
1171 	return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
1172 		flaa->vtyp);
1173 }
1174 
1175 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
1176 	"int", "struct timespec*", "struct timespec*");
1177 SDT_PROBE_DEFINE2(fusefs, , vnops, lookup_cache_incoherent,
1178 	"struct vnode*", "struct fuse_entry_out*");
1179 /*
1180     struct vnop_lookup_args {
1181 	struct vnodeop_desc *a_desc;
1182 	struct vnode *a_dvp;
1183 	struct vnode **a_vpp;
1184 	struct componentname *a_cnp;
1185     };
1186 */
1187 int
1188 fuse_vnop_lookup(struct vop_lookup_args *ap)
1189 {
1190 	struct vnode *dvp = ap->a_dvp;
1191 	struct vnode **vpp = ap->a_vpp;
1192 	struct componentname *cnp = ap->a_cnp;
1193 	struct thread *td = cnp->cn_thread;
1194 	struct ucred *cred = cnp->cn_cred;
1195 
1196 	int nameiop = cnp->cn_nameiop;
1197 	int flags = cnp->cn_flags;
1198 	int wantparent = flags & (LOCKPARENT | WANTPARENT);
1199 	int islastcn = flags & ISLASTCN;
1200 	struct mount *mp = vnode_mount(dvp);
1201 	struct fuse_data *data = fuse_get_mpdata(mp);
1202 	int default_permissions = data->dataflags & FSESS_DEFAULT_PERMISSIONS;
1203 
1204 	int err = 0;
1205 	int lookup_err = 0;
1206 	struct vnode *vp = NULL;
1207 
1208 	struct fuse_dispatcher fdi;
1209 	bool did_lookup = false;
1210 	struct fuse_entry_out *feo = NULL;
1211 	enum vtype vtyp;	/* vnode type of target */
1212 	off_t filesize;		/* filesize of target */
1213 
1214 	uint64_t nid;
1215 
1216 	if (fuse_isdeadfs(dvp)) {
1217 		*vpp = NULL;
1218 		return ENXIO;
1219 	}
1220 	if (!vnode_isdir(dvp))
1221 		return ENOTDIR;
1222 
1223 	if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1224 		return EROFS;
1225 
1226 	if ((cnp->cn_flags & NOEXECCHECK) != 0)
1227 		cnp->cn_flags &= ~NOEXECCHECK;
1228 	else if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1229 		return err;
1230 
1231 	if (flags & ISDOTDOT) {
1232 		KASSERT(VTOFUD(dvp)->flag & FN_PARENT_NID,
1233 			("Looking up .. is TODO"));
1234 		nid = VTOFUD(dvp)->parent_nid;
1235 		if (nid == 0)
1236 			return ENOENT;
1237 		/* .. is obviously a directory */
1238 		vtyp = VDIR;
1239 		filesize = 0;
1240 	} else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
1241 		nid = VTOI(dvp);
1242 		/* . is obviously a directory */
1243 		vtyp = VDIR;
1244 		filesize = 0;
1245 	} else {
1246 		struct timespec now, timeout;
1247 		int ncpticks; /* here to accomodate for API contract */
1248 
1249 		err = cache_lookup(dvp, vpp, cnp, &timeout, &ncpticks);
1250 		getnanouptime(&now);
1251 		SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1252 		switch (err) {
1253 		case -1:		/* positive match */
1254 			if (timespeccmp(&timeout, &now, >)) {
1255 				counter_u64_add(fuse_lookup_cache_hits, 1);
1256 			} else {
1257 				/* Cache timeout */
1258 				counter_u64_add(fuse_lookup_cache_misses, 1);
1259 				bintime_clear(
1260 					&VTOFUD(*vpp)->entry_cache_timeout);
1261 				cache_purge(*vpp);
1262 				if (dvp != *vpp)
1263 					vput(*vpp);
1264 				else
1265 					vrele(*vpp);
1266 				*vpp = NULL;
1267 				break;
1268 			}
1269 			return 0;
1270 
1271 		case 0:		/* no match in cache */
1272 			counter_u64_add(fuse_lookup_cache_misses, 1);
1273 			break;
1274 
1275 		case ENOENT:		/* negative match */
1276 			getnanouptime(&now);
1277 			if (timespeccmp(&timeout, &now, <=)) {
1278 				/* Cache timeout */
1279 				cache_purge_negative(dvp);
1280 				break;
1281 			}
1282 			/* fall through */
1283 		default:
1284 			return err;
1285 		}
1286 
1287 		nid = VTOI(dvp);
1288 		fdisp_init(&fdi, cnp->cn_namelen + 1);
1289 		fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred);
1290 
1291 		memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1292 		((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1293 		lookup_err = fdisp_wait_answ(&fdi);
1294 		did_lookup = true;
1295 
1296 		if (!lookup_err) {
1297 			/* lookup call succeeded */
1298 			feo = (struct fuse_entry_out *)fdi.answ;
1299 			nid = feo->nodeid;
1300 			if (nid == 0) {
1301 				/* zero nodeid means ENOENT and cache it */
1302 				struct timespec timeout;
1303 
1304 				fdi.answ_stat = ENOENT;
1305 				lookup_err = ENOENT;
1306 				if (cnp->cn_flags & MAKEENTRY) {
1307 					fuse_validity_2_timespec(feo, &timeout);
1308 					cache_enter_time(dvp, *vpp, cnp,
1309 						&timeout, NULL);
1310 				}
1311 			} else if (nid == FUSE_ROOT_ID) {
1312 				lookup_err = EINVAL;
1313 			}
1314 			vtyp = IFTOVT(feo->attr.mode);
1315 			filesize = feo->attr.size;
1316 		}
1317 		if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1318 			fdisp_destroy(&fdi);
1319 			return lookup_err;
1320 		}
1321 	}
1322 	/* lookup_err, if non-zero, must be ENOENT at this point */
1323 
1324 	if (lookup_err) {
1325 		/* Entry not found */
1326 		if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1327 			if (default_permissions)
1328 				err = fuse_internal_access(dvp, VWRITE, td,
1329 				    cred);
1330 			else
1331 				err = 0;
1332 			if (!err) {
1333 				/*
1334 				 * Set the SAVENAME flag to hold onto the
1335 				 * pathname for use later in VOP_CREATE or
1336 				 * VOP_RENAME.
1337 				 */
1338 				cnp->cn_flags |= SAVENAME;
1339 
1340 				err = EJUSTRETURN;
1341 			}
1342 		} else {
1343 			err = ENOENT;
1344 		}
1345 	} else {
1346 		/* Entry was found */
1347 		if (flags & ISDOTDOT) {
1348 			struct fuse_lookup_alloc_arg flaa;
1349 
1350 			flaa.nid = nid;
1351 			flaa.feo = feo;
1352 			flaa.cnp = cnp;
1353 			flaa.vtyp = vtyp;
1354 			err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1355 				&vp);
1356 			*vpp = vp;
1357 		} else if (nid == VTOI(dvp)) {
1358 			vref(dvp);
1359 			*vpp = dvp;
1360 		} else {
1361 			struct fuse_vnode_data *fvdat;
1362 			struct vattr *vap;
1363 
1364 			err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1365 			    &vp, cnp, vtyp);
1366 			if (err)
1367 				goto out;
1368 			*vpp = vp;
1369 
1370 			/*
1371 			 * In the case where we are looking up a FUSE node
1372 			 * represented by an existing cached vnode, and the
1373 			 * true size reported by FUSE_LOOKUP doesn't match
1374 			 * the vnode's cached size, then any cached writes
1375 			 * beyond the file's current size are lost.
1376 			 *
1377 			 * We can get here:
1378 			 * * following attribute cache expiration, or
1379 			 * * due a bug in the daemon, or
1380 			 */
1381 			fvdat = VTOFUD(vp);
1382 			if (vnode_isreg(vp) &&
1383 			    ((filesize != fvdat->cached_attrs.va_size &&
1384 			      fvdat->flag & FN_SIZECHANGE) ||
1385 			     ((vap = VTOVA(vp)) &&
1386 			      filesize != vap->va_size)))
1387 			{
1388 				SDT_PROBE2(fusefs, , vnops, lookup_cache_incoherent, vp, feo);
1389 				fvdat->flag &= ~FN_SIZECHANGE;
1390 				/*
1391 				 * The server changed the file's size even
1392 				 * though we had it cached, or had dirty writes
1393 				 * in the WB cache!
1394 				 */
1395 				printf("%s: cache incoherent on %s!  "
1396 		    		    "Buggy FUSE server detected.  To prevent "
1397 				    "data corruption, disable the data cache "
1398 				    "by mounting with -o direct_io, or as "
1399 				    "directed otherwise by your FUSE server's "
1400 		    		    "documentation\n", __func__,
1401 				    vnode_mount(vp)->mnt_stat.f_mntonname);
1402 				int iosize = fuse_iosize(vp);
1403 				v_inval_buf_range(vp, 0, INT64_MAX, iosize);
1404 			}
1405 
1406 			MPASS(feo != NULL);
1407 			fuse_internal_cache_attrs(*vpp, &feo->attr,
1408 				feo->attr_valid, feo->attr_valid_nsec, NULL);
1409 			fuse_validity_2_bintime(feo->entry_valid,
1410 				feo->entry_valid_nsec,
1411 				&fvdat->entry_cache_timeout);
1412 
1413 			if ((nameiop == DELETE || nameiop == RENAME) &&
1414 				islastcn && default_permissions)
1415 			{
1416 				struct vattr dvattr;
1417 
1418 				err = fuse_internal_access(dvp, VWRITE, td,
1419 					cred);
1420 				if (err != 0)
1421 					goto out;
1422 				/*
1423 				 * if the parent's sticky bit is set, check
1424 				 * whether we're allowed to remove the file.
1425 				 * Need to figure out the vnode locking to make
1426 				 * this work.
1427 				 */
1428 				fuse_internal_getattr(dvp, &dvattr, cred, td);
1429 				if ((dvattr.va_mode & S_ISTXT) &&
1430 					fuse_internal_access(dvp, VADMIN, td,
1431 						cred) &&
1432 					fuse_internal_access(*vpp, VADMIN, td,
1433 						cred)) {
1434 					err = EPERM;
1435 					goto out;
1436 				}
1437 			}
1438 
1439 			if (islastcn && (
1440 				(nameiop == DELETE) ||
1441 				(nameiop == RENAME && wantparent))) {
1442 				cnp->cn_flags |= SAVENAME;
1443 			}
1444 		}
1445 	}
1446 out:
1447 	if (err) {
1448 		if (vp != NULL && dvp != vp)
1449 			vput(vp);
1450 		else if (vp != NULL)
1451 			vrele(vp);
1452 		*vpp = NULL;
1453 	}
1454 	if (did_lookup)
1455 		fdisp_destroy(&fdi);
1456 
1457 	return err;
1458 }
1459 
1460 /*
1461     struct vnop_mkdir_args {
1462 	struct vnode *a_dvp;
1463 	struct vnode **a_vpp;
1464 	struct componentname *a_cnp;
1465 	struct vattr *a_vap;
1466     };
1467 */
1468 static int
1469 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1470 {
1471 	struct vnode *dvp = ap->a_dvp;
1472 	struct vnode **vpp = ap->a_vpp;
1473 	struct componentname *cnp = ap->a_cnp;
1474 	struct vattr *vap = ap->a_vap;
1475 
1476 	struct fuse_mkdir_in fmdi;
1477 
1478 	if (fuse_isdeadfs(dvp)) {
1479 		return ENXIO;
1480 	}
1481 	fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1482 	fmdi.umask = curthread->td_proc->p_pd->pd_cmask;
1483 
1484 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1485 	    sizeof(fmdi), VDIR));
1486 }
1487 
1488 /*
1489     struct vnop_mknod_args {
1490 	struct vnode *a_dvp;
1491 	struct vnode **a_vpp;
1492 	struct componentname *a_cnp;
1493 	struct vattr *a_vap;
1494     };
1495 */
1496 static int
1497 fuse_vnop_mknod(struct vop_mknod_args *ap)
1498 {
1499 
1500 	struct vnode *dvp = ap->a_dvp;
1501 	struct vnode **vpp = ap->a_vpp;
1502 	struct componentname *cnp = ap->a_cnp;
1503 	struct vattr *vap = ap->a_vap;
1504 
1505 	if (fuse_isdeadfs(dvp))
1506 		return ENXIO;
1507 
1508 	return fuse_internal_mknod(dvp, vpp, cnp, vap);
1509 }
1510 
1511 /*
1512     struct vop_open_args {
1513 	struct vnode *a_vp;
1514 	int  a_mode;
1515 	struct ucred *a_cred;
1516 	struct thread *a_td;
1517 	int a_fdidx; / struct file *a_fp;
1518     };
1519 */
1520 static int
1521 fuse_vnop_open(struct vop_open_args *ap)
1522 {
1523 	struct vnode *vp = ap->a_vp;
1524 	int a_mode = ap->a_mode;
1525 	struct thread *td = ap->a_td;
1526 	struct ucred *cred = ap->a_cred;
1527 	pid_t pid = td->td_proc->p_pid;
1528 	struct fuse_vnode_data *fvdat;
1529 
1530 	if (fuse_isdeadfs(vp))
1531 		return ENXIO;
1532 	if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1533 		return (EOPNOTSUPP);
1534 	if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1535 		return EINVAL;
1536 
1537 	fvdat = VTOFUD(vp);
1538 
1539 	if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1540 		fuse_vnode_open(vp, 0, td);
1541 		return 0;
1542 	}
1543 
1544 	return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1545 }
1546 
1547 static int
1548 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1549 {
1550 	struct vnode *vp = ap->a_vp;
1551 	struct mount *mp;
1552 
1553 	switch (ap->a_name) {
1554 	case _PC_FILESIZEBITS:
1555 		*ap->a_retval = 64;
1556 		return (0);
1557 	case _PC_NAME_MAX:
1558 		*ap->a_retval = NAME_MAX;
1559 		return (0);
1560 	case _PC_LINK_MAX:
1561 		*ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1562 		return (0);
1563 	case _PC_SYMLINK_MAX:
1564 		*ap->a_retval = MAXPATHLEN;
1565 		return (0);
1566 	case _PC_NO_TRUNC:
1567 		*ap->a_retval = 1;
1568 		return (0);
1569 	case _PC_MIN_HOLE_SIZE:
1570 		/*
1571 		 * The FUSE protocol provides no mechanism for a server to
1572 		 * report _PC_MIN_HOLE_SIZE.  It's a protocol bug.  Instead,
1573 		 * return EINVAL if the server does not support FUSE_LSEEK, or
1574 		 * 1 if it does.
1575 		 */
1576 		mp = vnode_mount(vp);
1577 		if (!fsess_is_impl(mp, FUSE_LSEEK) &&
1578 		    !fsess_not_impl(mp, FUSE_LSEEK)) {
1579 			off_t offset = 0;
1580 
1581 			/* Issue a FUSE_LSEEK to find out if it's implemented */
1582 			fuse_vnop_do_lseek(vp, curthread, curthread->td_ucred,
1583 			    curthread->td_proc->p_pid, &offset, SEEK_DATA);
1584 		}
1585 
1586 		if (fsess_is_impl(mp, FUSE_LSEEK)) {
1587 			*ap->a_retval = 1;
1588 			return (0);
1589 		} else {
1590 			/*
1591 			 * Probably FUSE_LSEEK is not implemented.  It might
1592 			 * be, if the FUSE_LSEEK above returned an error like
1593 			 * EACCES, but in that case we can't tell, so it's
1594 			 * safest to report EINVAL anyway.
1595 			 */
1596 			return (EINVAL);
1597 		}
1598 	default:
1599 		return (vop_stdpathconf(ap));
1600 	}
1601 }
1602 
1603 /*
1604     struct vnop_read_args {
1605 	struct vnode *a_vp;
1606 	struct uio *a_uio;
1607 	int  a_ioflag;
1608 	struct ucred *a_cred;
1609     };
1610 */
1611 static int
1612 fuse_vnop_read(struct vop_read_args *ap)
1613 {
1614 	struct vnode *vp = ap->a_vp;
1615 	struct uio *uio = ap->a_uio;
1616 	int ioflag = ap->a_ioflag;
1617 	struct ucred *cred = ap->a_cred;
1618 	pid_t pid = curthread->td_proc->p_pid;
1619 
1620 	if (fuse_isdeadfs(vp)) {
1621 		return ENXIO;
1622 	}
1623 
1624 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1625 		ioflag |= IO_DIRECT;
1626 	}
1627 
1628 	return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1629 }
1630 
1631 /*
1632     struct vnop_readdir_args {
1633 	struct vnode *a_vp;
1634 	struct uio *a_uio;
1635 	struct ucred *a_cred;
1636 	int *a_eofflag;
1637 	int *a_ncookies;
1638 	u_long **a_cookies;
1639     };
1640 */
1641 static int
1642 fuse_vnop_readdir(struct vop_readdir_args *ap)
1643 {
1644 	struct vnode *vp = ap->a_vp;
1645 	struct uio *uio = ap->a_uio;
1646 	struct ucred *cred = ap->a_cred;
1647 	struct fuse_filehandle *fufh = NULL;
1648 	struct fuse_iov cookediov;
1649 	int err = 0;
1650 	u_long *cookies;
1651 	off_t startoff;
1652 	ssize_t tresid;
1653 	int ncookies;
1654 	bool closefufh = false;
1655 	pid_t pid = curthread->td_proc->p_pid;
1656 
1657 	if (ap->a_eofflag)
1658 		*ap->a_eofflag = 0;
1659 	if (fuse_isdeadfs(vp)) {
1660 		return ENXIO;
1661 	}
1662 	if (				/* XXXIP ((uio_iovcnt(uio) > 1)) || */
1663 	    (uio_resid(uio) < sizeof(struct dirent))) {
1664 		return EINVAL;
1665 	}
1666 
1667 	tresid = uio->uio_resid;
1668 	startoff = uio->uio_offset;
1669 	err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1670 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1671 		/*
1672 		 * nfsd will do VOP_READDIR without first doing VOP_OPEN.  We
1673 		 * must implicitly open the directory here
1674 		 */
1675 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1676 		if (err == 0) {
1677 			/*
1678 			 * When a directory is opened, it must be read from
1679 			 * the beginning.  Hopefully, the "startoff" still
1680 			 * exists as an offset cookie for the directory.
1681 			 * If not, it will read the entire directory without
1682 			 * returning any entries and just return eof.
1683 			 */
1684 			uio->uio_offset = 0;
1685 		}
1686 		closefufh = true;
1687 	}
1688 	if (err)
1689 		return (err);
1690 	if (ap->a_ncookies != NULL) {
1691 		ncookies = uio->uio_resid /
1692 			(offsetof(struct dirent, d_name) + 4) + 1;
1693 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1694 		*ap->a_ncookies = ncookies;
1695 		*ap->a_cookies = cookies;
1696 	} else {
1697 		ncookies = 0;
1698 		cookies = NULL;
1699 	}
1700 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1701 	fiov_init(&cookediov, DIRCOOKEDSIZE);
1702 
1703 	err = fuse_internal_readdir(vp, uio, startoff, fufh, &cookediov,
1704 		&ncookies, cookies);
1705 
1706 	fiov_teardown(&cookediov);
1707 	if (closefufh)
1708 		fuse_filehandle_close(vp, fufh, curthread, cred);
1709 
1710 	if (ap->a_ncookies != NULL) {
1711 		if (err == 0) {
1712 			*ap->a_ncookies -= ncookies;
1713 		} else {
1714 			free(*ap->a_cookies, M_TEMP);
1715 			*ap->a_ncookies = 0;
1716 			*ap->a_cookies = NULL;
1717 		}
1718 	}
1719 	if (err == 0 && tresid == uio->uio_resid)
1720 		*ap->a_eofflag = 1;
1721 
1722 	return err;
1723 }
1724 
1725 /*
1726     struct vnop_readlink_args {
1727 	struct vnode *a_vp;
1728 	struct uio *a_uio;
1729 	struct ucred *a_cred;
1730     };
1731 */
1732 static int
1733 fuse_vnop_readlink(struct vop_readlink_args *ap)
1734 {
1735 	struct vnode *vp = ap->a_vp;
1736 	struct uio *uio = ap->a_uio;
1737 	struct ucred *cred = ap->a_cred;
1738 
1739 	struct fuse_dispatcher fdi;
1740 	int err;
1741 
1742 	if (fuse_isdeadfs(vp)) {
1743 		return ENXIO;
1744 	}
1745 	if (!vnode_islnk(vp)) {
1746 		return EINVAL;
1747 	}
1748 	fdisp_init(&fdi, 0);
1749 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1750 	if (err) {
1751 		goto out;
1752 	}
1753 	if (((char *)fdi.answ)[0] == '/' &&
1754 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1755 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1756 
1757 		err = uiomove(mpth, strlen(mpth), uio);
1758 	}
1759 	if (!err) {
1760 		err = uiomove(fdi.answ, fdi.iosize, uio);
1761 	}
1762 out:
1763 	fdisp_destroy(&fdi);
1764 	return err;
1765 }
1766 
1767 /*
1768     struct vnop_reclaim_args {
1769 	struct vnode *a_vp;
1770     };
1771 */
1772 static int
1773 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1774 {
1775 	struct vnode *vp = ap->a_vp;
1776 	struct thread *td = curthread;
1777 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1778 	struct fuse_filehandle *fufh, *fufh_tmp;
1779 
1780 	if (!fvdat) {
1781 		panic("FUSE: no vnode data during recycling");
1782 	}
1783 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1784 		printf("FUSE: vnode being reclaimed with open fufh "
1785 			"(type=%#x)", fufh->fufh_type);
1786 		fuse_filehandle_close(vp, fufh, td, NULL);
1787 	}
1788 
1789 	if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
1790 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1791 		    fvdat->nlookup);
1792 	}
1793 	cache_purge(vp);
1794 	vfs_hash_remove(vp);
1795 	fuse_vnode_destroy(vp);
1796 
1797 	return 0;
1798 }
1799 
1800 /*
1801     struct vnop_remove_args {
1802 	struct vnode *a_dvp;
1803 	struct vnode *a_vp;
1804 	struct componentname *a_cnp;
1805     };
1806 */
1807 static int
1808 fuse_vnop_remove(struct vop_remove_args *ap)
1809 {
1810 	struct vnode *dvp = ap->a_dvp;
1811 	struct vnode *vp = ap->a_vp;
1812 	struct componentname *cnp = ap->a_cnp;
1813 
1814 	int err;
1815 
1816 	if (fuse_isdeadfs(vp)) {
1817 		return ENXIO;
1818 	}
1819 	if (vnode_isdir(vp)) {
1820 		return EPERM;
1821 	}
1822 
1823 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1824 
1825 	return err;
1826 }
1827 
1828 /*
1829     struct vnop_rename_args {
1830 	struct vnode *a_fdvp;
1831 	struct vnode *a_fvp;
1832 	struct componentname *a_fcnp;
1833 	struct vnode *a_tdvp;
1834 	struct vnode *a_tvp;
1835 	struct componentname *a_tcnp;
1836     };
1837 */
1838 static int
1839 fuse_vnop_rename(struct vop_rename_args *ap)
1840 {
1841 	struct vnode *fdvp = ap->a_fdvp;
1842 	struct vnode *fvp = ap->a_fvp;
1843 	struct componentname *fcnp = ap->a_fcnp;
1844 	struct vnode *tdvp = ap->a_tdvp;
1845 	struct vnode *tvp = ap->a_tvp;
1846 	struct componentname *tcnp = ap->a_tcnp;
1847 	struct fuse_data *data;
1848 	bool newparent = fdvp != tdvp;
1849 	bool isdir = fvp->v_type == VDIR;
1850 	int err = 0;
1851 
1852 	if (fuse_isdeadfs(fdvp)) {
1853 		return ENXIO;
1854 	}
1855 	if (fvp->v_mount != tdvp->v_mount ||
1856 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1857 		SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
1858 		err = EXDEV;
1859 		goto out;
1860 	}
1861 	cache_purge(fvp);
1862 
1863 	/*
1864 	 * FUSE library is expected to check if target directory is not
1865 	 * under the source directory in the file system tree.
1866 	 * Linux performs this check at VFS level.
1867 	 */
1868 	/*
1869 	 * If source is a directory, and it will get a new parent, user must
1870 	 * have write permission to it, so ".." can be modified.
1871 	 */
1872 	data = fuse_get_mpdata(vnode_mount(tdvp));
1873 	if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
1874 		err = fuse_internal_access(fvp, VWRITE,
1875 			tcnp->cn_thread, tcnp->cn_cred);
1876 		if (err)
1877 			goto out;
1878 	}
1879 	sx_xlock(&data->rename_lock);
1880 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1881 	if (err == 0) {
1882 		if (tdvp != fdvp)
1883 			fuse_vnode_setparent(fvp, tdvp);
1884 		if (tvp != NULL)
1885 			fuse_vnode_setparent(tvp, NULL);
1886 	}
1887 	sx_unlock(&data->rename_lock);
1888 
1889 	if (tvp != NULL && tvp != fvp) {
1890 		cache_purge(tvp);
1891 	}
1892 	if (vnode_isdir(fvp)) {
1893 		if ((tvp != NULL) && vnode_isdir(tvp)) {
1894 			cache_purge(tdvp);
1895 		}
1896 		cache_purge(fdvp);
1897 	}
1898 out:
1899 	if (tdvp == tvp) {
1900 		vrele(tdvp);
1901 	} else {
1902 		vput(tdvp);
1903 	}
1904 	if (tvp != NULL) {
1905 		vput(tvp);
1906 	}
1907 	vrele(fdvp);
1908 	vrele(fvp);
1909 
1910 	return err;
1911 }
1912 
1913 /*
1914     struct vnop_rmdir_args {
1915 	    struct vnode *a_dvp;
1916 	    struct vnode *a_vp;
1917 	    struct componentname *a_cnp;
1918     } *ap;
1919 */
1920 static int
1921 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1922 {
1923 	struct vnode *dvp = ap->a_dvp;
1924 	struct vnode *vp = ap->a_vp;
1925 
1926 	int err;
1927 
1928 	if (fuse_isdeadfs(vp)) {
1929 		return ENXIO;
1930 	}
1931 	if (VTOFUD(vp) == VTOFUD(dvp)) {
1932 		return EINVAL;
1933 	}
1934 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1935 
1936 	return err;
1937 }
1938 
1939 /*
1940     struct vnop_setattr_args {
1941 	struct vnode *a_vp;
1942 	struct vattr *a_vap;
1943 	struct ucred *a_cred;
1944 	struct thread *a_td;
1945     };
1946 */
1947 static int
1948 fuse_vnop_setattr(struct vop_setattr_args *ap)
1949 {
1950 	struct vnode *vp = ap->a_vp;
1951 	struct vattr *vap = ap->a_vap;
1952 	struct ucred *cred = ap->a_cred;
1953 	struct thread *td = curthread;
1954 	struct mount *mp;
1955 	struct fuse_data *data;
1956 	struct vattr old_va;
1957 	int dataflags;
1958 	int err = 0, err2;
1959 	accmode_t accmode = 0;
1960 	bool checkperm;
1961 	bool drop_suid = false;
1962 	gid_t cr_gid;
1963 
1964 	mp = vnode_mount(vp);
1965 	data = fuse_get_mpdata(mp);
1966 	dataflags = data->dataflags;
1967 	checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
1968 	if (cred->cr_ngroups > 0)
1969 		cr_gid = cred->cr_groups[0];
1970 	else
1971 		cr_gid = 0;
1972 
1973 	if (fuse_isdeadfs(vp)) {
1974 		return ENXIO;
1975 	}
1976 
1977 	if (vap->va_uid != (uid_t)VNOVAL) {
1978 		if (checkperm) {
1979 			/* Only root may change a file's owner */
1980 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1981 			if (err) {
1982 				/* As a special case, allow the null chown */
1983 				err2 = fuse_internal_getattr(vp, &old_va, cred,
1984 					td);
1985 				if (err2)
1986 					return (err2);
1987 				if (vap->va_uid != old_va.va_uid)
1988 					return err;
1989 				else
1990 					accmode |= VADMIN;
1991 				drop_suid = true;
1992 			} else
1993 				accmode |= VADMIN;
1994 		} else
1995 			accmode |= VADMIN;
1996 	}
1997 	if (vap->va_gid != (gid_t)VNOVAL) {
1998 		if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
1999 			drop_suid = true;
2000 		if (checkperm && !groupmember(vap->va_gid, cred))
2001 		{
2002 			/*
2003 			 * Non-root users may only chgrp to one of their own
2004 			 * groups
2005 			 */
2006 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
2007 			if (err) {
2008 				/* As a special case, allow the null chgrp */
2009 				err2 = fuse_internal_getattr(vp, &old_va, cred,
2010 					td);
2011 				if (err2)
2012 					return (err2);
2013 				if (vap->va_gid != old_va.va_gid)
2014 					return err;
2015 				accmode |= VADMIN;
2016 			} else
2017 				accmode |= VADMIN;
2018 		} else
2019 			accmode |= VADMIN;
2020 	}
2021 	if (vap->va_size != VNOVAL) {
2022 		switch (vp->v_type) {
2023 		case VDIR:
2024 			return (EISDIR);
2025 		case VLNK:
2026 		case VREG:
2027 			if (vfs_isrdonly(mp))
2028 				return (EROFS);
2029 			break;
2030 		default:
2031 			/*
2032 			 * According to POSIX, the result is unspecified
2033 			 * for file types other than regular files,
2034 			 * directories and shared memory objects.  We
2035 			 * don't support shared memory objects in the file
2036 			 * system, and have dubious support for truncating
2037 			 * symlinks.  Just ignore the request in other cases.
2038 			 */
2039 			return (0);
2040 		}
2041 		/* Don't set accmode.  Permission to trunc is checked upstack */
2042 	}
2043 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
2044 		if (vap->va_vaflags & VA_UTIMES_NULL)
2045 			accmode |= VWRITE;
2046 		else
2047 			accmode |= VADMIN;
2048 	}
2049 	if (drop_suid) {
2050 		if (vap->va_mode != (mode_t)VNOVAL)
2051 			vap->va_mode &= ~(S_ISUID | S_ISGID);
2052 		else {
2053 			err = fuse_internal_getattr(vp, &old_va, cred, td);
2054 			if (err)
2055 				return (err);
2056 			vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
2057 		}
2058 	}
2059 	if (vap->va_mode != (mode_t)VNOVAL) {
2060 		/* Only root may set the sticky bit on non-directories */
2061 		if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
2062 		    && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
2063 			return EFTYPE;
2064 		if (checkperm && (vap->va_mode & S_ISGID)) {
2065 			err = fuse_internal_getattr(vp, &old_va, cred, td);
2066 			if (err)
2067 				return (err);
2068 			if (!groupmember(old_va.va_gid, cred)) {
2069 				err = priv_check_cred(cred, PRIV_VFS_SETGID);
2070 				if (err)
2071 					return (err);
2072 			}
2073 		}
2074 		accmode |= VADMIN;
2075 	}
2076 
2077 	if (vfs_isrdonly(mp))
2078 		return EROFS;
2079 
2080 	if (checkperm) {
2081 		err = fuse_internal_access(vp, accmode, td, cred);
2082 	} else {
2083 		err = 0;
2084 	}
2085 	if (err)
2086 		return err;
2087 	else
2088 		return fuse_internal_setattr(vp, vap, td, cred);
2089 }
2090 
2091 /*
2092     struct vnop_strategy_args {
2093 	struct vnode *a_vp;
2094 	struct buf *a_bp;
2095     };
2096 */
2097 static int
2098 fuse_vnop_strategy(struct vop_strategy_args *ap)
2099 {
2100 	struct vnode *vp = ap->a_vp;
2101 	struct buf *bp = ap->a_bp;
2102 
2103 	if (!vp || fuse_isdeadfs(vp)) {
2104 		bp->b_ioflags |= BIO_ERROR;
2105 		bp->b_error = ENXIO;
2106 		bufdone(bp);
2107 		return 0;
2108 	}
2109 
2110 	/*
2111 	 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
2112 	 * fuse_io_strategy sets bp's error fields
2113 	 */
2114 	(void)fuse_io_strategy(vp, bp);
2115 
2116 	return 0;
2117 }
2118 
2119 /*
2120     struct vnop_symlink_args {
2121 	struct vnode *a_dvp;
2122 	struct vnode **a_vpp;
2123 	struct componentname *a_cnp;
2124 	struct vattr *a_vap;
2125 	char *a_target;
2126     };
2127 */
2128 static int
2129 fuse_vnop_symlink(struct vop_symlink_args *ap)
2130 {
2131 	struct vnode *dvp = ap->a_dvp;
2132 	struct vnode **vpp = ap->a_vpp;
2133 	struct componentname *cnp = ap->a_cnp;
2134 	const char *target = ap->a_target;
2135 
2136 	struct fuse_dispatcher fdi;
2137 
2138 	int err;
2139 	size_t len;
2140 
2141 	if (fuse_isdeadfs(dvp)) {
2142 		return ENXIO;
2143 	}
2144 	/*
2145 	 * Unlike the other creator type calls, here we have to create a message
2146 	 * where the name of the new entry comes first, and the data describing
2147 	 * the entry comes second.
2148 	 * Hence we can't rely on our handy fuse_internal_newentry() routine,
2149 	 * but put together the message manually and just call the core part.
2150 	 */
2151 
2152 	len = strlen(target) + 1;
2153 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
2154 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
2155 
2156 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
2157 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
2158 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
2159 
2160 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
2161 	fdisp_destroy(&fdi);
2162 	return err;
2163 }
2164 
2165 /*
2166     struct vnop_write_args {
2167 	struct vnode *a_vp;
2168 	struct uio *a_uio;
2169 	int  a_ioflag;
2170 	struct ucred *a_cred;
2171     };
2172 */
2173 static int
2174 fuse_vnop_write(struct vop_write_args *ap)
2175 {
2176 	struct vnode *vp = ap->a_vp;
2177 	struct uio *uio = ap->a_uio;
2178 	int ioflag = ap->a_ioflag;
2179 	struct ucred *cred = ap->a_cred;
2180 	pid_t pid = curthread->td_proc->p_pid;
2181 
2182 	if (fuse_isdeadfs(vp)) {
2183 		return ENXIO;
2184 	}
2185 
2186 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
2187 		ioflag |= IO_DIRECT;
2188 	}
2189 
2190 	return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
2191 }
2192 
2193 static daddr_t
2194 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
2195 {
2196 	const int biosize = fuse_iosize(vp);
2197 
2198 	return (off / biosize);
2199 }
2200 
2201 static int
2202 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn)
2203 {
2204 	off_t filesize;
2205 	int blksz, err;
2206 	const int biosize = fuse_iosize(vp);
2207 
2208 	err = fuse_vnode_size(vp, &filesize, NULL, NULL);
2209 	KASSERT(err == 0, ("vfs_bio_getpages can't handle errors here"));
2210 	if (err)
2211 		return biosize;
2212 
2213 	if ((off_t)lbn * biosize >= filesize) {
2214 		blksz = 0;
2215 	} else if ((off_t)(lbn + 1) * biosize > filesize) {
2216 		blksz = filesize - (off_t)lbn *biosize;
2217 	} else {
2218 		blksz = biosize;
2219 	}
2220 	return (blksz);
2221 }
2222 
2223 /*
2224     struct vnop_getpages_args {
2225 	struct vnode *a_vp;
2226 	vm_page_t *a_m;
2227 	int a_count;
2228 	int a_reqpage;
2229     };
2230 */
2231 static int
2232 fuse_vnop_getpages(struct vop_getpages_args *ap)
2233 {
2234 	struct vnode *vp = ap->a_vp;
2235 
2236 	if (!fsess_opt_mmap(vnode_mount(vp))) {
2237 		SDT_PROBE2(fusefs, , vnops, trace, 1,
2238 			"called on non-cacheable vnode??\n");
2239 		return (VM_PAGER_ERROR);
2240 	}
2241 
2242 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
2243 	    ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
2244 }
2245 
2246 static const char extattr_namespace_separator = '.';
2247 
2248 /*
2249     struct vop_getextattr_args {
2250 	struct vop_generic_args a_gen;
2251 	struct vnode *a_vp;
2252 	int a_attrnamespace;
2253 	const char *a_name;
2254 	struct uio *a_uio;
2255 	size_t *a_size;
2256 	struct ucred *a_cred;
2257 	struct thread *a_td;
2258     };
2259 */
2260 static int
2261 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2262 {
2263 	struct vnode *vp = ap->a_vp;
2264 	struct uio *uio = ap->a_uio;
2265 	struct fuse_dispatcher fdi;
2266 	struct fuse_getxattr_in *get_xattr_in;
2267 	struct fuse_getxattr_out *get_xattr_out;
2268 	struct mount *mp = vnode_mount(vp);
2269 	struct thread *td = ap->a_td;
2270 	struct ucred *cred = ap->a_cred;
2271 	char *prefix;
2272 	char *attr_str;
2273 	size_t len;
2274 	int err;
2275 
2276 	if (fuse_isdeadfs(vp))
2277 		return (ENXIO);
2278 
2279 	if (fsess_not_impl(mp, FUSE_GETXATTR))
2280 		return EOPNOTSUPP;
2281 
2282 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2283 	if (err)
2284 		return err;
2285 
2286 	/* Default to looking for user attributes. */
2287 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2288 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2289 	else
2290 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2291 
2292 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2293 	    strlen(ap->a_name) + 1;
2294 
2295 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2296 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2297 
2298 	get_xattr_in = fdi.indata;
2299 	/*
2300 	 * Check to see whether we're querying the available size or
2301 	 * issuing the actual request.  If we pass in 0, we get back struct
2302 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2303 	 * that much data, without the struct fuse_getxattr_out header.
2304 	 */
2305 	if (uio == NULL)
2306 		get_xattr_in->size = 0;
2307 	else
2308 		get_xattr_in->size = uio->uio_resid;
2309 
2310 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2311 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2312 	    ap->a_name);
2313 
2314 	err = fdisp_wait_answ(&fdi);
2315 	if (err != 0) {
2316 		if (err == ENOSYS) {
2317 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2318 			err = EOPNOTSUPP;
2319 		}
2320 		goto out;
2321 	}
2322 
2323 	get_xattr_out = fdi.answ;
2324 
2325 	if (ap->a_size != NULL)
2326 		*ap->a_size = get_xattr_out->size;
2327 
2328 	if (uio != NULL)
2329 		err = uiomove(fdi.answ, fdi.iosize, uio);
2330 
2331 out:
2332 	fdisp_destroy(&fdi);
2333 	return (err);
2334 }
2335 
2336 /*
2337     struct vop_setextattr_args {
2338 	struct vop_generic_args a_gen;
2339 	struct vnode *a_vp;
2340 	int a_attrnamespace;
2341 	const char *a_name;
2342 	struct uio *a_uio;
2343 	struct ucred *a_cred;
2344 	struct thread *a_td;
2345     };
2346 */
2347 static int
2348 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2349 {
2350 	struct vnode *vp = ap->a_vp;
2351 	struct uio *uio = ap->a_uio;
2352 	struct fuse_dispatcher fdi;
2353 	struct fuse_setxattr_in *set_xattr_in;
2354 	struct mount *mp = vnode_mount(vp);
2355 	struct thread *td = ap->a_td;
2356 	struct ucred *cred = ap->a_cred;
2357 	char *prefix;
2358 	size_t len;
2359 	char *attr_str;
2360 	int err;
2361 
2362 	if (fuse_isdeadfs(vp))
2363 		return (ENXIO);
2364 
2365 	if (fsess_not_impl(mp, FUSE_SETXATTR))
2366 		return EOPNOTSUPP;
2367 
2368 	if (vfs_isrdonly(mp))
2369 		return EROFS;
2370 
2371 	/* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2372 	if (ap->a_uio == NULL) {
2373 		/*
2374 		 * If we got here as fallback from VOP_DELETEEXTATTR, then
2375 		 * return EOPNOTSUPP.
2376 		 */
2377 		if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
2378 			return (EOPNOTSUPP);
2379 		else
2380 			return (EINVAL);
2381 	}
2382 
2383 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2384 		VWRITE);
2385 	if (err)
2386 		return err;
2387 
2388 	/* Default to looking for user attributes. */
2389 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2390 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2391 	else
2392 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2393 
2394 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2395 	    strlen(ap->a_name) + 1;
2396 
2397 	fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2398 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2399 
2400 	set_xattr_in = fdi.indata;
2401 	set_xattr_in->size = uio->uio_resid;
2402 
2403 	attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2404 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2405 	    ap->a_name);
2406 
2407 	err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2408 	    uio->uio_resid, uio);
2409 	if (err != 0) {
2410 		goto out;
2411 	}
2412 
2413 	err = fdisp_wait_answ(&fdi);
2414 
2415 	if (err == ENOSYS) {
2416 		fsess_set_notimpl(mp, FUSE_SETXATTR);
2417 		err = EOPNOTSUPP;
2418 	}
2419 	if (err == ERESTART) {
2420 		/* Can't restart after calling uiomove */
2421 		err = EINTR;
2422 	}
2423 
2424 out:
2425 	fdisp_destroy(&fdi);
2426 	return (err);
2427 }
2428 
2429 /*
2430  * The Linux / FUSE extended attribute list is simply a collection of
2431  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2432  * byte length followed by a non-NUL terminated string.  So, this allows
2433  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2434  * Linux attribute names are reported with the namespace as a prefix (e.g.
2435  * "user.attribute_name"), but in FreeBSD they are reported without the
2436  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2437  *
2438  * user.attr_name1\0user.attr_name2\0
2439  *
2440  * to:
2441  *
2442  * <num>attr_name1<num>attr_name2
2443  *
2444  * Where "<num>" is a single byte number of characters in the attribute name.
2445  *
2446  * Args:
2447  * prefix - exattr namespace prefix string
2448  * list, list_len - input list with namespace prefixes
2449  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2450  */
2451 static int
2452 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2453     char *bsd_list, int *bsd_list_len)
2454 {
2455 	int len, pos, dist_to_next, prefix_len;
2456 
2457 	pos = 0;
2458 	*bsd_list_len = 0;
2459 	prefix_len = strlen(prefix);
2460 
2461 	while (pos < list_len && list[pos] != '\0') {
2462 		dist_to_next = strlen(&list[pos]) + 1;
2463 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2464 		    list[pos + prefix_len] == extattr_namespace_separator) {
2465 			len = dist_to_next -
2466 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2467 			if (len >= EXTATTR_MAXNAMELEN)
2468 				return (ENAMETOOLONG);
2469 
2470 			bsd_list[*bsd_list_len] = len;
2471 			memcpy(&bsd_list[*bsd_list_len + 1],
2472 			    &list[pos + prefix_len +
2473 			    sizeof(extattr_namespace_separator)], len);
2474 
2475 			*bsd_list_len += len + 1;
2476 		}
2477 
2478 		pos += dist_to_next;
2479 	}
2480 
2481 	return (0);
2482 }
2483 
2484 /*
2485  * List extended attributes
2486  *
2487  * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
2488  * has a number of differences compared to its FreeBSD equivalent,
2489  * extattr_list_file:
2490  *
2491  * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
2492  *   whereas listxattr(2) only returns attributes for a single namespace
2493  * - FUSE_LISTXATTR prepends each attribute name with "namespace."
2494  * - If the provided buffer is not large enough to hold the result,
2495  *   FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
2496  *   return as many results as will fit.
2497  */
2498 /*
2499     struct vop_listextattr_args {
2500 	struct vop_generic_args a_gen;
2501 	struct vnode *a_vp;
2502 	int a_attrnamespace;
2503 	struct uio *a_uio;
2504 	size_t *a_size;
2505 	struct ucred *a_cred;
2506 	struct thread *a_td;
2507     };
2508 */
2509 static int
2510 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2511 {
2512 	struct vnode *vp = ap->a_vp;
2513 	struct uio *uio = ap->a_uio;
2514 	struct fuse_dispatcher fdi;
2515 	struct fuse_listxattr_in *list_xattr_in;
2516 	struct fuse_listxattr_out *list_xattr_out;
2517 	struct mount *mp = vnode_mount(vp);
2518 	struct thread *td = ap->a_td;
2519 	struct ucred *cred = ap->a_cred;
2520 	char *prefix;
2521 	char *bsd_list = NULL;
2522 	char *linux_list;
2523 	int bsd_list_len;
2524 	int linux_list_len;
2525 	int err;
2526 
2527 	if (fuse_isdeadfs(vp))
2528 		return (ENXIO);
2529 
2530 	if (fsess_not_impl(mp, FUSE_LISTXATTR))
2531 		return EOPNOTSUPP;
2532 
2533 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2534 	if (err)
2535 		return err;
2536 
2537 	/*
2538 	 * Add space for a NUL and the period separator if enabled.
2539 	 * Default to looking for user attributes.
2540 	 */
2541 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2542 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2543 	else
2544 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2545 
2546 	fdisp_init(&fdi, sizeof(*list_xattr_in));
2547 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2548 
2549 	/*
2550 	 * Retrieve Linux / FUSE compatible list size.
2551 	 */
2552 	list_xattr_in = fdi.indata;
2553 	list_xattr_in->size = 0;
2554 
2555 	err = fdisp_wait_answ(&fdi);
2556 	if (err != 0) {
2557 		if (err == ENOSYS) {
2558 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
2559 			err = EOPNOTSUPP;
2560 		}
2561 		goto out;
2562 	}
2563 
2564 	list_xattr_out = fdi.answ;
2565 	linux_list_len = list_xattr_out->size;
2566 	if (linux_list_len == 0) {
2567 		if (ap->a_size != NULL)
2568 			*ap->a_size = linux_list_len;
2569 		goto out;
2570 	}
2571 
2572 	/*
2573 	 * Retrieve Linux / FUSE compatible list values.
2574 	 */
2575 	fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2576 	list_xattr_in = fdi.indata;
2577 	list_xattr_in->size = linux_list_len;
2578 
2579 	err = fdisp_wait_answ(&fdi);
2580 	if (err == ERANGE) {
2581 		/*
2582 		 * Race detected.  The attribute list must've grown since the
2583 		 * first FUSE_LISTXATTR call.  Start over.  Go all the way back
2584 		 * to userland so we can process signals, if necessary, before
2585 		 * restarting.
2586 		 */
2587 		err = ERESTART;
2588 		goto out;
2589 	} else if (err != 0)
2590 		goto out;
2591 
2592 	linux_list = fdi.answ;
2593 	/* FUSE doesn't allow the server to return more data than requested */
2594 	if (fdi.iosize > linux_list_len) {
2595 		printf("WARNING: FUSE protocol violation.  Server returned "
2596 			"more extended attribute data than requested; "
2597 			"should've returned ERANGE instead");
2598 	} else {
2599 		/* But returning less data is fine */
2600 		linux_list_len = fdi.iosize;
2601 	}
2602 
2603 	/*
2604 	 * Retrieve the BSD compatible list values.
2605 	 * The Linux / FUSE attribute list format isn't the same
2606 	 * as FreeBSD's format. So we need to transform it into
2607 	 * FreeBSD's format before giving it to the user.
2608 	 */
2609 	bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2610 	err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2611 	    bsd_list, &bsd_list_len);
2612 	if (err != 0)
2613 		goto out;
2614 
2615 	if (ap->a_size != NULL)
2616 		*ap->a_size = bsd_list_len;
2617 
2618 	if (uio != NULL)
2619 		err = uiomove(bsd_list, bsd_list_len, uio);
2620 
2621 out:
2622 	free(bsd_list, M_TEMP);
2623 	fdisp_destroy(&fdi);
2624 	return (err);
2625 }
2626 
2627 /*
2628     struct vop_deleteextattr_args {
2629 	struct vop_generic_args a_gen;
2630 	struct vnode *a_vp;
2631 	int a_attrnamespace;
2632 	const char *a_name;
2633 	struct ucred *a_cred;
2634 	struct thread *a_td;
2635     };
2636 */
2637 static int
2638 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2639 {
2640 	struct vnode *vp = ap->a_vp;
2641 	struct fuse_dispatcher fdi;
2642 	struct mount *mp = vnode_mount(vp);
2643 	struct thread *td = ap->a_td;
2644 	struct ucred *cred = ap->a_cred;
2645 	char *prefix;
2646 	size_t len;
2647 	char *attr_str;
2648 	int err;
2649 
2650 	if (fuse_isdeadfs(vp))
2651 		return (ENXIO);
2652 
2653 	if (fsess_not_impl(mp, FUSE_REMOVEXATTR))
2654 		return EOPNOTSUPP;
2655 
2656 	if (vfs_isrdonly(mp))
2657 		return EROFS;
2658 
2659 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2660 		VWRITE);
2661 	if (err)
2662 		return err;
2663 
2664 	/* Default to looking for user attributes. */
2665 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2666 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2667 	else
2668 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2669 
2670 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2671 	    strlen(ap->a_name) + 1;
2672 
2673 	fdisp_init(&fdi, len);
2674 	fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2675 
2676 	attr_str = fdi.indata;
2677 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2678 	    ap->a_name);
2679 
2680 	err = fdisp_wait_answ(&fdi);
2681 	if (err == ENOSYS) {
2682 		fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2683 		err = EOPNOTSUPP;
2684 	}
2685 
2686 	fdisp_destroy(&fdi);
2687 	return (err);
2688 }
2689 
2690 /*
2691     struct vnop_print_args {
2692 	struct vnode *a_vp;
2693     };
2694 */
2695 static int
2696 fuse_vnop_print(struct vop_print_args *ap)
2697 {
2698 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2699 
2700 	printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2701 	    (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2702 	    (uintmax_t)fvdat->nlookup,
2703 	    fvdat->flag);
2704 
2705 	return 0;
2706 }
2707 
2708 /*
2709  * Get an NFS filehandle for a FUSE file.
2710  *
2711  * This will only work for FUSE file systems that guarantee the uniqueness of
2712  * nodeid:generation, which most don't.
2713  */
2714 /*
2715 vop_vptofh {
2716 	IN struct vnode *a_vp;
2717 	IN struct fid *a_fhp;
2718 };
2719 */
2720 static int
2721 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
2722 {
2723 	struct vnode *vp = ap->a_vp;
2724 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
2725 	struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
2726 	_Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
2727 		"FUSE fid type is too big");
2728 	struct mount *mp = vnode_mount(vp);
2729 	struct fuse_data *data = fuse_get_mpdata(mp);
2730 	struct vattr va;
2731 	int err;
2732 
2733 	if (!(data->dataflags & FSESS_EXPORT_SUPPORT))
2734 		return EOPNOTSUPP;
2735 
2736 	err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
2737 	if (err)
2738 		return err;
2739 
2740 	/*ip = VTOI(ap->a_vp);*/
2741 	/*ufhp = (struct ufid *)ap->a_fhp;*/
2742 	fhp->len = sizeof(struct fuse_fid);
2743 	fhp->nid = fvdat->nid;
2744 	if (fvdat->generation <= UINT32_MAX)
2745 		fhp->gen = fvdat->generation;
2746 	else
2747 		return EOVERFLOW;
2748 	return (0);
2749 }
2750