xref: /freebsd/sys/fs/fuse/fuse_vnops.c (revision 6f63e88c0166ed3e5f2805a9e667c7d24d304cf1)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2007-2009 Google Inc. and Amit Singh
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions are
9  * met:
10  *
11  * * Redistributions of source code must retain the above copyright
12  *   notice, this list of conditions and the following disclaimer.
13  * * Redistributions in binary form must reproduce the above
14  *   copyright notice, this list of conditions and the following disclaimer
15  *   in the documentation and/or other materials provided with the
16  *   distribution.
17  * * Neither the name of Google Inc. nor the names of its
18  *   contributors may be used to endorse or promote products derived from
19  *   this software without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  *
33  * Copyright (C) 2005 Csaba Henk.
34  * All rights reserved.
35  *
36  * Copyright (c) 2019 The FreeBSD Foundation
37  *
38  * Portions of this software were developed by BFF Storage Systems, LLC under
39  * sponsorship from the FreeBSD Foundation.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  *
50  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  */
62 
63 #include <sys/cdefs.h>
64 __FBSDID("$FreeBSD$");
65 
66 #include <sys/param.h>
67 #include <sys/module.h>
68 #include <sys/systm.h>
69 #include <sys/errno.h>
70 #include <sys/kernel.h>
71 #include <sys/conf.h>
72 #include <sys/uio.h>
73 #include <sys/malloc.h>
74 #include <sys/queue.h>
75 #include <sys/limits.h>
76 #include <sys/lock.h>
77 #include <sys/rwlock.h>
78 #include <sys/sx.h>
79 #include <sys/proc.h>
80 #include <sys/mount.h>
81 #include <sys/vnode.h>
82 #include <sys/namei.h>
83 #include <sys/extattr.h>
84 #include <sys/stat.h>
85 #include <sys/unistd.h>
86 #include <sys/filedesc.h>
87 #include <sys/file.h>
88 #include <sys/fcntl.h>
89 #include <sys/dirent.h>
90 #include <sys/bio.h>
91 #include <sys/buf.h>
92 #include <sys/sysctl.h>
93 #include <sys/vmmeter.h>
94 
95 #include <vm/vm.h>
96 #include <vm/vm_extern.h>
97 #include <vm/pmap.h>
98 #include <vm/vm_map.h>
99 #include <vm/vm_page.h>
100 #include <vm/vm_param.h>
101 #include <vm/vm_object.h>
102 #include <vm/vm_pager.h>
103 #include <vm/vnode_pager.h>
104 #include <vm/vm_object.h>
105 
106 #include "fuse.h"
107 #include "fuse_file.h"
108 #include "fuse_internal.h"
109 #include "fuse_ipc.h"
110 #include "fuse_node.h"
111 #include "fuse_io.h"
112 
113 #include <sys/priv.h>
114 
115 /* Maximum number of hardlinks to a single FUSE file */
116 #define FUSE_LINK_MAX                      UINT32_MAX
117 
118 SDT_PROVIDER_DECLARE(fusefs);
119 /*
120  * Fuse trace probe:
121  * arg0: verbosity.  Higher numbers give more verbose messages
122  * arg1: Textual message
123  */
124 SDT_PROBE_DEFINE2(fusefs, , vnops, trace, "int", "char*");
125 
126 /* vnode ops */
127 static vop_access_t fuse_vnop_access;
128 static vop_advlock_t fuse_vnop_advlock;
129 static vop_bmap_t fuse_vnop_bmap;
130 static vop_close_t fuse_fifo_close;
131 static vop_close_t fuse_vnop_close;
132 static vop_create_t fuse_vnop_create;
133 static vop_deleteextattr_t fuse_vnop_deleteextattr;
134 static vop_fdatasync_t fuse_vnop_fdatasync;
135 static vop_fsync_t fuse_vnop_fsync;
136 static vop_getattr_t fuse_vnop_getattr;
137 static vop_getextattr_t fuse_vnop_getextattr;
138 static vop_inactive_t fuse_vnop_inactive;
139 static vop_link_t fuse_vnop_link;
140 static vop_listextattr_t fuse_vnop_listextattr;
141 static vop_lookup_t fuse_vnop_lookup;
142 static vop_mkdir_t fuse_vnop_mkdir;
143 static vop_mknod_t fuse_vnop_mknod;
144 static vop_open_t fuse_vnop_open;
145 static vop_pathconf_t fuse_vnop_pathconf;
146 static vop_read_t fuse_vnop_read;
147 static vop_readdir_t fuse_vnop_readdir;
148 static vop_readlink_t fuse_vnop_readlink;
149 static vop_reclaim_t fuse_vnop_reclaim;
150 static vop_remove_t fuse_vnop_remove;
151 static vop_rename_t fuse_vnop_rename;
152 static vop_rmdir_t fuse_vnop_rmdir;
153 static vop_setattr_t fuse_vnop_setattr;
154 static vop_setextattr_t fuse_vnop_setextattr;
155 static vop_strategy_t fuse_vnop_strategy;
156 static vop_symlink_t fuse_vnop_symlink;
157 static vop_write_t fuse_vnop_write;
158 static vop_getpages_t fuse_vnop_getpages;
159 static vop_print_t fuse_vnop_print;
160 static vop_vptofh_t fuse_vnop_vptofh;
161 
162 struct vop_vector fuse_fifoops = {
163 	.vop_default =		&fifo_specops,
164 	.vop_access =		fuse_vnop_access,
165 	.vop_close =		fuse_fifo_close,
166 	.vop_fsync =		fuse_vnop_fsync,
167 	.vop_getattr =		fuse_vnop_getattr,
168 	.vop_inactive =		fuse_vnop_inactive,
169 	.vop_pathconf =		fuse_vnop_pathconf,
170 	.vop_print =		fuse_vnop_print,
171 	.vop_read =		VOP_PANIC,
172 	.vop_reclaim =		fuse_vnop_reclaim,
173 	.vop_setattr =		fuse_vnop_setattr,
174 	.vop_write =		VOP_PANIC,
175 	.vop_vptofh =		fuse_vnop_vptofh,
176 };
177 VFS_VOP_VECTOR_REGISTER(fuse_fifoops);
178 
179 struct vop_vector fuse_vnops = {
180 	.vop_allocate =	VOP_EINVAL,
181 	.vop_default = &default_vnodeops,
182 	.vop_access = fuse_vnop_access,
183 	.vop_advlock = fuse_vnop_advlock,
184 	.vop_bmap = fuse_vnop_bmap,
185 	.vop_close = fuse_vnop_close,
186 	.vop_create = fuse_vnop_create,
187 	.vop_deleteextattr = fuse_vnop_deleteextattr,
188 	.vop_fsync = fuse_vnop_fsync,
189 	.vop_fdatasync = fuse_vnop_fdatasync,
190 	.vop_getattr = fuse_vnop_getattr,
191 	.vop_getextattr = fuse_vnop_getextattr,
192 	.vop_inactive = fuse_vnop_inactive,
193 	/*
194 	 * TODO: implement vop_ioctl after upgrading to protocol 7.16.
195 	 * FUSE_IOCTL was added in 7.11, but 32-bit compat is broken until
196 	 * 7.16.
197 	 */
198 	.vop_link = fuse_vnop_link,
199 	.vop_listextattr = fuse_vnop_listextattr,
200 	.vop_lookup = fuse_vnop_lookup,
201 	.vop_mkdir = fuse_vnop_mkdir,
202 	.vop_mknod = fuse_vnop_mknod,
203 	.vop_open = fuse_vnop_open,
204 	.vop_pathconf = fuse_vnop_pathconf,
205 	/*
206 	 * TODO: implement vop_poll after upgrading to protocol 7.21.
207 	 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
208 	 * 7.21, which adds the ability for the client to choose which poll
209 	 * events it wants, and for a client to deregister a file handle
210 	 */
211 	.vop_read = fuse_vnop_read,
212 	.vop_readdir = fuse_vnop_readdir,
213 	.vop_readlink = fuse_vnop_readlink,
214 	.vop_reclaim = fuse_vnop_reclaim,
215 	.vop_remove = fuse_vnop_remove,
216 	.vop_rename = fuse_vnop_rename,
217 	.vop_rmdir = fuse_vnop_rmdir,
218 	.vop_setattr = fuse_vnop_setattr,
219 	.vop_setextattr = fuse_vnop_setextattr,
220 	.vop_strategy = fuse_vnop_strategy,
221 	.vop_symlink = fuse_vnop_symlink,
222 	.vop_write = fuse_vnop_write,
223 	.vop_getpages = fuse_vnop_getpages,
224 	.vop_print = fuse_vnop_print,
225 	.vop_vptofh = fuse_vnop_vptofh,
226 };
227 VFS_VOP_VECTOR_REGISTER(fuse_vnops);
228 
229 uma_zone_t fuse_pbuf_zone;
230 
231 /* Check permission for extattr operations, much like extattr_check_cred */
232 static int
233 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
234 	struct thread *td, accmode_t accmode)
235 {
236 	struct mount *mp = vnode_mount(vp);
237 	struct fuse_data *data = fuse_get_mpdata(mp);
238 
239 	/*
240 	 * Kernel-invoked always succeeds.
241 	 */
242 	if (cred == NOCRED)
243 		return (0);
244 
245 	/*
246 	 * Do not allow privileged processes in jail to directly manipulate
247 	 * system attributes.
248 	 */
249 	switch (ns) {
250 	case EXTATTR_NAMESPACE_SYSTEM:
251 		if (data->dataflags & FSESS_DEFAULT_PERMISSIONS) {
252 			return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
253 		}
254 		/* FALLTHROUGH */
255 	case EXTATTR_NAMESPACE_USER:
256 		return (fuse_internal_access(vp, accmode, td, cred));
257 	default:
258 		return (EPERM);
259 	}
260 }
261 
262 /* Get a filehandle for a directory */
263 static int
264 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
265 	struct ucred *cred, pid_t pid)
266 {
267 	if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
268 		return 0;
269 	return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
270 }
271 
272 /* Send FUSE_FLUSH for this vnode */
273 static int
274 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
275 {
276 	struct fuse_flush_in *ffi;
277 	struct fuse_filehandle *fufh;
278 	struct fuse_dispatcher fdi;
279 	struct thread *td = curthread;
280 	struct mount *mp = vnode_mount(vp);
281 	int err;
282 
283 	if (!fsess_isimpl(vnode_mount(vp), FUSE_FLUSH))
284 		return 0;
285 
286 	err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
287 	if (err)
288 		return err;
289 
290 	fdisp_init(&fdi, sizeof(*ffi));
291 	fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
292 	ffi = fdi.indata;
293 	ffi->fh = fufh->fh_id;
294 	/*
295 	 * If the file has a POSIX lock then we're supposed to set lock_owner.
296 	 * If not, then lock_owner is undefined.  So we may as well always set
297 	 * it.
298 	 */
299 	ffi->lock_owner = td->td_proc->p_pid;
300 
301 	err = fdisp_wait_answ(&fdi);
302 	if (err == ENOSYS) {
303 		fsess_set_notimpl(mp, FUSE_FLUSH);
304 		err = 0;
305 	}
306 	fdisp_destroy(&fdi);
307 	return err;
308 }
309 
310 /* Close wrapper for fifos.  */
311 static int
312 fuse_fifo_close(struct vop_close_args *ap)
313 {
314 	return (fifo_specops.vop_close(ap));
315 }
316 
317 /*
318     struct vnop_access_args {
319 	struct vnode *a_vp;
320 #if VOP_ACCESS_TAKES_ACCMODE_T
321 	accmode_t a_accmode;
322 #else
323 	int a_mode;
324 #endif
325 	struct ucred *a_cred;
326 	struct thread *a_td;
327     };
328 */
329 static int
330 fuse_vnop_access(struct vop_access_args *ap)
331 {
332 	struct vnode *vp = ap->a_vp;
333 	int accmode = ap->a_accmode;
334 	struct ucred *cred = ap->a_cred;
335 
336 	struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
337 
338 	int err;
339 
340 	if (fuse_isdeadfs(vp)) {
341 		if (vnode_isvroot(vp)) {
342 			return 0;
343 		}
344 		return ENXIO;
345 	}
346 	if (!(data->dataflags & FSESS_INITED)) {
347 		if (vnode_isvroot(vp)) {
348 			if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
349 			    (fuse_match_cred(data->daemoncred, cred) == 0)) {
350 				return 0;
351 			}
352 		}
353 		return EBADF;
354 	}
355 	if (vnode_islnk(vp)) {
356 		return 0;
357 	}
358 
359 	err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
360 	return err;
361 }
362 
363 /*
364  * struct vop_advlock_args {
365  *	struct vop_generic_args a_gen;
366  *	struct vnode *a_vp;
367  *	void *a_id;
368  *	int a_op;
369  *	struct flock *a_fl;
370  *	int a_flags;
371  * }
372  */
373 static int
374 fuse_vnop_advlock(struct vop_advlock_args *ap)
375 {
376 	struct vnode *vp = ap->a_vp;
377 	struct flock *fl = ap->a_fl;
378 	struct thread *td = curthread;
379 	struct ucred *cred = td->td_ucred;
380 	pid_t pid = td->td_proc->p_pid;
381 	struct fuse_filehandle *fufh;
382 	struct fuse_dispatcher fdi;
383 	struct fuse_lk_in *fli;
384 	struct fuse_lk_out *flo;
385 	enum fuse_opcode op;
386 	int dataflags, err;
387 	int flags = ap->a_flags;
388 
389 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
390 
391 	if (fuse_isdeadfs(vp)) {
392 		return ENXIO;
393 	}
394 
395 	if (!(dataflags & FSESS_POSIX_LOCKS))
396 		return vop_stdadvlock(ap);
397 	/* FUSE doesn't properly support flock until protocol 7.17 */
398 	if (flags & F_FLOCK)
399 		return vop_stdadvlock(ap);
400 
401 	err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
402 	if (err)
403 		return err;
404 
405 	fdisp_init(&fdi, sizeof(*fli));
406 
407 	switch(ap->a_op) {
408 	case F_GETLK:
409 		op = FUSE_GETLK;
410 		break;
411 	case F_SETLK:
412 		op = FUSE_SETLK;
413 		break;
414 	case F_SETLKW:
415 		op = FUSE_SETLKW;
416 		break;
417 	default:
418 		return EINVAL;
419 	}
420 
421 	fdisp_make_vp(&fdi, op, vp, td, cred);
422 	fli = fdi.indata;
423 	fli->fh = fufh->fh_id;
424 	fli->owner = fl->l_pid;
425 	fli->lk.start = fl->l_start;
426 	if (fl->l_len != 0)
427 		fli->lk.end = fl->l_start + fl->l_len - 1;
428 	else
429 		fli->lk.end = INT64_MAX;
430 	fli->lk.type = fl->l_type;
431 	fli->lk.pid = fl->l_pid;
432 
433 	err = fdisp_wait_answ(&fdi);
434 	fdisp_destroy(&fdi);
435 
436 	if (err == 0 && op == FUSE_GETLK) {
437 		flo = fdi.answ;
438 		fl->l_type = flo->lk.type;
439 		fl->l_pid = flo->lk.pid;
440 		if (flo->lk.type != F_UNLCK) {
441 			fl->l_start = flo->lk.start;
442 			if (flo->lk.end == INT64_MAX)
443 				fl->l_len = 0;
444 			else
445 				fl->l_len = flo->lk.end - flo->lk.start + 1;
446 			fl->l_start = flo->lk.start;
447 		}
448 	}
449 
450 	return err;
451 }
452 
453 /* {
454 	struct vnode *a_vp;
455 	daddr_t a_bn;
456 	struct bufobj **a_bop;
457 	daddr_t *a_bnp;
458 	int *a_runp;
459 	int *a_runb;
460 } */
461 static int
462 fuse_vnop_bmap(struct vop_bmap_args *ap)
463 {
464 	struct vnode *vp = ap->a_vp;
465 	struct bufobj **bo = ap->a_bop;
466 	struct thread *td = curthread;
467 	struct mount *mp;
468 	struct fuse_dispatcher fdi;
469 	struct fuse_bmap_in *fbi;
470 	struct fuse_bmap_out *fbo;
471 	struct fuse_data *data;
472 	uint64_t biosize;
473 	off_t filesize;
474 	daddr_t lbn = ap->a_bn;
475 	daddr_t *pbn = ap->a_bnp;
476 	int *runp = ap->a_runp;
477 	int *runb = ap->a_runb;
478 	int error = 0;
479 	int maxrun;
480 
481 	if (fuse_isdeadfs(vp)) {
482 		return ENXIO;
483 	}
484 
485 	mp = vnode_mount(vp);
486 	data = fuse_get_mpdata(mp);
487 	biosize = fuse_iosize(vp);
488 	maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
489 		data->max_readahead_blocks);
490 
491 	if (bo != NULL)
492 		*bo = &vp->v_bufobj;
493 
494 	/*
495 	 * The FUSE_BMAP operation does not include the runp and runb
496 	 * variables, so we must guess.  Report nonzero contiguous runs so
497 	 * cluster_read will combine adjacent reads.  It's worthwhile to reduce
498 	 * upcalls even if we don't know the true physical layout of the file.
499 	 *
500 	 * FUSE file systems may opt out of read clustering in two ways:
501 	 * * mounting with -onoclusterr
502 	 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
503 	 */
504 	if (runb != NULL)
505 		*runb = MIN(lbn, maxrun);
506 	if (runp != NULL) {
507 		error = fuse_vnode_size(vp, &filesize, td->td_ucred, td);
508 		if (error == 0)
509 			*runp = MIN(MAX(0, filesize / (off_t)biosize - lbn - 1),
510 				    maxrun);
511 		else
512 			*runp = 0;
513 	}
514 
515 	if (fsess_isimpl(mp, FUSE_BMAP)) {
516 		fdisp_init(&fdi, sizeof(*fbi));
517 		fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
518 		fbi = fdi.indata;
519 		fbi->block = lbn;
520 		fbi->blocksize = biosize;
521 		error = fdisp_wait_answ(&fdi);
522 		if (error == ENOSYS) {
523 			fdisp_destroy(&fdi);
524 			fsess_set_notimpl(mp, FUSE_BMAP);
525 			error = 0;
526 		} else {
527 			fbo = fdi.answ;
528 			if (error == 0 && pbn != NULL)
529 				*pbn = fbo->block;
530 			fdisp_destroy(&fdi);
531 			return error;
532 		}
533 	}
534 
535 	/* If the daemon doesn't support BMAP, make up a sensible default */
536 	if (pbn != NULL)
537 		*pbn = lbn * btodb(biosize);
538 	return (error);
539 }
540 
541 /*
542     struct vop_close_args {
543 	struct vnode *a_vp;
544 	int  a_fflag;
545 	struct ucred *a_cred;
546 	struct thread *a_td;
547     };
548 */
549 static int
550 fuse_vnop_close(struct vop_close_args *ap)
551 {
552 	struct vnode *vp = ap->a_vp;
553 	struct ucred *cred = ap->a_cred;
554 	int fflag = ap->a_fflag;
555 	struct thread *td = ap->a_td;
556 	pid_t pid = td->td_proc->p_pid;
557 	int err = 0;
558 
559 	if (fuse_isdeadfs(vp))
560 		return 0;
561 	if (vnode_isdir(vp))
562 		return 0;
563 	if (fflag & IO_NDELAY)
564 		return 0;
565 
566 	err = fuse_flush(vp, cred, pid, fflag);
567 	/* TODO: close the file handle, if we're sure it's no longer used */
568 	if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
569 		fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
570 	}
571 	return err;
572 }
573 
574 static void
575 fdisp_make_mknod_for_fallback(
576 	struct fuse_dispatcher *fdip,
577 	struct componentname *cnp,
578 	struct vnode *dvp,
579 	uint64_t parentnid,
580 	struct thread *td,
581 	struct ucred *cred,
582 	mode_t mode,
583 	enum fuse_opcode *op)
584 {
585 	struct fuse_mknod_in *fmni;
586 
587 	fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
588 	*op = FUSE_MKNOD;
589 	fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
590 	fmni = fdip->indata;
591 	fmni->mode = mode;
592 	fmni->rdev = 0;
593 	memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
594 	    cnp->cn_namelen);
595 	((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
596 }
597 /*
598     struct vnop_create_args {
599 	struct vnode *a_dvp;
600 	struct vnode **a_vpp;
601 	struct componentname *a_cnp;
602 	struct vattr *a_vap;
603     };
604 */
605 static int
606 fuse_vnop_create(struct vop_create_args *ap)
607 {
608 	struct vnode *dvp = ap->a_dvp;
609 	struct vnode **vpp = ap->a_vpp;
610 	struct componentname *cnp = ap->a_cnp;
611 	struct vattr *vap = ap->a_vap;
612 	struct thread *td = cnp->cn_thread;
613 	struct ucred *cred = cnp->cn_cred;
614 
615 	struct fuse_data *data;
616 	struct fuse_create_in *fci;
617 	struct fuse_entry_out *feo;
618 	struct fuse_open_out *foo;
619 	struct fuse_dispatcher fdi, fdi2;
620 	struct fuse_dispatcher *fdip = &fdi;
621 	struct fuse_dispatcher *fdip2 = NULL;
622 
623 	int err;
624 
625 	struct mount *mp = vnode_mount(dvp);
626 	data = fuse_get_mpdata(mp);
627 	uint64_t parentnid = VTOFUD(dvp)->nid;
628 	mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
629 	enum fuse_opcode op;
630 	int flags;
631 
632 	if (fuse_isdeadfs(dvp))
633 		return ENXIO;
634 
635 	/* FUSE expects sockets to be created with FUSE_MKNOD */
636 	if (vap->va_type == VSOCK)
637 		return fuse_internal_mknod(dvp, vpp, cnp, vap);
638 
639 	/*
640 	 * VOP_CREATE doesn't tell us the open(2) flags, so we guess.  Only a
641 	 * writable mode makes sense, and we might as well include readability
642 	 * too.
643 	 */
644 	flags = O_RDWR;
645 
646 	bzero(&fdi, sizeof(fdi));
647 
648 	if (vap->va_type != VREG)
649 		return (EINVAL);
650 
651 	if (!fsess_isimpl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
652 		/* Fallback to FUSE_MKNOD/FUSE_OPEN */
653 		fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
654 			cred, mode, &op);
655 	} else {
656 		/* Use FUSE_CREATE */
657 		size_t insize;
658 
659 		op = FUSE_CREATE;
660 		fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
661 		fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
662 		fci = fdip->indata;
663 		fci->mode = mode;
664 		fci->flags = O_CREAT | flags;
665 		if (fuse_libabi_geq(data, 7, 12)) {
666 			insize = sizeof(*fci);
667 			fci->umask = td->td_proc->p_fd->fd_cmask;
668 		} else {
669 			insize = sizeof(struct fuse_open_in);
670 		}
671 
672 		memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
673 		    cnp->cn_namelen);
674 		((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
675 	}
676 
677 	err = fdisp_wait_answ(fdip);
678 
679 	if (err) {
680 		if (err == ENOSYS && op == FUSE_CREATE) {
681 			fsess_set_notimpl(mp, FUSE_CREATE);
682 			fdisp_destroy(fdip);
683 			fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
684 				parentnid, td, cred, mode, &op);
685 			err = fdisp_wait_answ(fdip);
686 		}
687 		if (err)
688 			goto out;
689 	}
690 
691 	feo = fdip->answ;
692 
693 	if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
694 		goto out;
695 	}
696 
697 	if (op == FUSE_CREATE) {
698 		foo = (struct fuse_open_out*)(feo + 1);
699 	} else {
700 		/* Issue a separate FUSE_OPEN */
701 		struct fuse_open_in *foi;
702 
703 		fdip2 = &fdi2;
704 		fdisp_init(fdip2, sizeof(*foi));
705 		fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
706 			cred);
707 		foi = fdip2->indata;
708 		foi->flags = flags;
709 		err = fdisp_wait_answ(fdip2);
710 		if (err)
711 			goto out;
712 		foo = fdip2->answ;
713 	}
714 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
715 	if (err) {
716 		struct fuse_release_in *fri;
717 		uint64_t nodeid = feo->nodeid;
718 		uint64_t fh_id = foo->fh;
719 
720 		fdisp_init(fdip, sizeof(*fri));
721 		fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
722 		fri = fdip->indata;
723 		fri->fh = fh_id;
724 		fri->flags = flags;
725 		fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
726 		fuse_insert_message(fdip->tick, false);
727 		goto out;
728 	}
729 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
730 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
731 		feo->attr_valid_nsec, NULL);
732 
733 	fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
734 	fuse_vnode_open(*vpp, foo->open_flags, td);
735 	/*
736 	 * Purge the parent's attribute cache because the daemon should've
737 	 * updated its mtime and ctime
738 	 */
739 	fuse_vnode_clear_attr_cache(dvp);
740 	cache_purge_negative(dvp);
741 
742 out:
743 	if (fdip2)
744 		fdisp_destroy(fdip2);
745 	fdisp_destroy(fdip);
746 	return err;
747 }
748 
749 /*
750     struct vnop_fdatasync_args {
751 	struct vop_generic_args a_gen;
752 	struct vnode * a_vp;
753 	struct thread * a_td;
754     };
755 */
756 static int
757 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
758 {
759 	struct vnode *vp = ap->a_vp;
760 	struct thread *td = ap->a_td;
761 	int waitfor = MNT_WAIT;
762 
763 	int err = 0;
764 
765 	if (fuse_isdeadfs(vp)) {
766 		return 0;
767 	}
768 	if ((err = vop_stdfdatasync_buf(ap)))
769 		return err;
770 
771 	return fuse_internal_fsync(vp, td, waitfor, true);
772 }
773 
774 /*
775     struct vnop_fsync_args {
776 	struct vop_generic_args a_gen;
777 	struct vnode * a_vp;
778 	int  a_waitfor;
779 	struct thread * a_td;
780     };
781 */
782 static int
783 fuse_vnop_fsync(struct vop_fsync_args *ap)
784 {
785 	struct vnode *vp = ap->a_vp;
786 	struct thread *td = ap->a_td;
787 	int waitfor = ap->a_waitfor;
788 	int err = 0;
789 
790 	if (fuse_isdeadfs(vp)) {
791 		return 0;
792 	}
793 	if ((err = vop_stdfsync(ap)))
794 		return err;
795 
796 	return fuse_internal_fsync(vp, td, waitfor, false);
797 }
798 
799 /*
800     struct vnop_getattr_args {
801 	struct vnode *a_vp;
802 	struct vattr *a_vap;
803 	struct ucred *a_cred;
804 	struct thread *a_td;
805     };
806 */
807 static int
808 fuse_vnop_getattr(struct vop_getattr_args *ap)
809 {
810 	struct vnode *vp = ap->a_vp;
811 	struct vattr *vap = ap->a_vap;
812 	struct ucred *cred = ap->a_cred;
813 	struct thread *td = curthread;
814 
815 	int err = 0;
816 	int dataflags;
817 
818 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
819 
820 	/* Note that we are not bailing out on a dead file system just yet. */
821 
822 	if (!(dataflags & FSESS_INITED)) {
823 		if (!vnode_isvroot(vp)) {
824 			fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
825 			err = ENOTCONN;
826 			return err;
827 		} else {
828 			goto fake;
829 		}
830 	}
831 	err = fuse_internal_getattr(vp, vap, cred, td);
832 	if (err == ENOTCONN && vnode_isvroot(vp)) {
833 		/* see comment in fuse_vfsop_statfs() */
834 		goto fake;
835 	} else {
836 		return err;
837 	}
838 
839 fake:
840 	bzero(vap, sizeof(*vap));
841 	vap->va_type = vnode_vtype(vp);
842 
843 	return 0;
844 }
845 
846 /*
847     struct vnop_inactive_args {
848 	struct vnode *a_vp;
849 	struct thread *a_td;
850     };
851 */
852 static int
853 fuse_vnop_inactive(struct vop_inactive_args *ap)
854 {
855 	struct vnode *vp = ap->a_vp;
856 	struct thread *td = ap->a_td;
857 
858 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
859 	struct fuse_filehandle *fufh, *fufh_tmp;
860 
861 	int need_flush = 1;
862 
863 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
864 		if (need_flush && vp->v_type == VREG) {
865 			if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
866 				fuse_vnode_savesize(vp, NULL, 0);
867 			}
868 			if ((fvdat->flag & FN_REVOKED) != 0)
869 				fuse_io_invalbuf(vp, td);
870 			else
871 				fuse_io_flushbuf(vp, MNT_WAIT, td);
872 			need_flush = 0;
873 		}
874 		fuse_filehandle_close(vp, fufh, td, NULL);
875 	}
876 
877 	if ((fvdat->flag & FN_REVOKED) != 0)
878 		vrecycle(vp);
879 
880 	return 0;
881 }
882 
883 /*
884     struct vnop_link_args {
885 	struct vnode *a_tdvp;
886 	struct vnode *a_vp;
887 	struct componentname *a_cnp;
888     };
889 */
890 static int
891 fuse_vnop_link(struct vop_link_args *ap)
892 {
893 	struct vnode *vp = ap->a_vp;
894 	struct vnode *tdvp = ap->a_tdvp;
895 	struct componentname *cnp = ap->a_cnp;
896 
897 	struct vattr *vap = VTOVA(vp);
898 
899 	struct fuse_dispatcher fdi;
900 	struct fuse_entry_out *feo;
901 	struct fuse_link_in fli;
902 
903 	int err;
904 
905 	if (fuse_isdeadfs(vp)) {
906 		return ENXIO;
907 	}
908 	if (vnode_mount(tdvp) != vnode_mount(vp)) {
909 		return EXDEV;
910 	}
911 
912 	/*
913 	 * This is a seatbelt check to protect naive userspace filesystems from
914 	 * themselves and the limitations of the FUSE IPC protocol.  If a
915 	 * filesystem does not allow attribute caching, assume it is capable of
916 	 * validating that nlink does not overflow.
917 	 */
918 	if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
919 		return EMLINK;
920 	fli.oldnodeid = VTOI(vp);
921 
922 	fdisp_init(&fdi, 0);
923 	fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
924 	    FUSE_LINK, &fli, sizeof(fli), &fdi);
925 	if ((err = fdisp_wait_answ(&fdi))) {
926 		goto out;
927 	}
928 	feo = fdi.answ;
929 
930 	err = fuse_internal_checkentry(feo, vnode_vtype(vp));
931 	if (!err) {
932 		/*
933 		 * Purge the parent's attribute cache because the daemon
934 		 * should've updated its mtime and ctime
935 		 */
936 		fuse_vnode_clear_attr_cache(tdvp);
937 		fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
938 			feo->attr_valid_nsec, NULL);
939 	}
940 out:
941 	fdisp_destroy(&fdi);
942 	return err;
943 }
944 
945 struct fuse_lookup_alloc_arg {
946 	struct fuse_entry_out *feo;
947 	struct componentname *cnp;
948 	uint64_t nid;
949 	enum vtype vtyp;
950 };
951 
952 /* Callback for vn_get_ino */
953 static int
954 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
955 {
956 	struct fuse_lookup_alloc_arg *flaa = arg;
957 
958 	return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
959 		flaa->vtyp);
960 }
961 
962 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
963 	"int", "struct timespec*", "struct timespec*");
964 SDT_PROBE_DEFINE2(fusefs, , vnops, lookup_cache_incoherent,
965 	"struct vnode*", "struct fuse_entry_out*");
966 /*
967     struct vnop_lookup_args {
968 	struct vnodeop_desc *a_desc;
969 	struct vnode *a_dvp;
970 	struct vnode **a_vpp;
971 	struct componentname *a_cnp;
972     };
973 */
974 int
975 fuse_vnop_lookup(struct vop_lookup_args *ap)
976 {
977 	struct vnode *dvp = ap->a_dvp;
978 	struct vnode **vpp = ap->a_vpp;
979 	struct componentname *cnp = ap->a_cnp;
980 	struct thread *td = cnp->cn_thread;
981 	struct ucred *cred = cnp->cn_cred;
982 
983 	int nameiop = cnp->cn_nameiop;
984 	int flags = cnp->cn_flags;
985 	int wantparent = flags & (LOCKPARENT | WANTPARENT);
986 	int islastcn = flags & ISLASTCN;
987 	struct mount *mp = vnode_mount(dvp);
988 
989 	int err = 0;
990 	int lookup_err = 0;
991 	struct vnode *vp = NULL;
992 
993 	struct fuse_dispatcher fdi;
994 	bool did_lookup = false;
995 	struct fuse_entry_out *feo = NULL;
996 	enum vtype vtyp;	/* vnode type of target */
997 	off_t filesize;		/* filesize of target */
998 
999 	uint64_t nid;
1000 
1001 	if (fuse_isdeadfs(dvp)) {
1002 		*vpp = NULL;
1003 		return ENXIO;
1004 	}
1005 	if (!vnode_isdir(dvp))
1006 		return ENOTDIR;
1007 
1008 	if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1009 		return EROFS;
1010 
1011 	if ((cnp->cn_flags & NOEXECCHECK) != 0)
1012 		cnp->cn_flags &= ~NOEXECCHECK;
1013 	else if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1014 		return err;
1015 
1016 	if (flags & ISDOTDOT) {
1017 		KASSERT(VTOFUD(dvp)->flag & FN_PARENT_NID,
1018 			("Looking up .. is TODO"));
1019 		nid = VTOFUD(dvp)->parent_nid;
1020 		if (nid == 0)
1021 			return ENOENT;
1022 		/* .. is obviously a directory */
1023 		vtyp = VDIR;
1024 		filesize = 0;
1025 	} else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
1026 		nid = VTOI(dvp);
1027 		/* . is obviously a directory */
1028 		vtyp = VDIR;
1029 		filesize = 0;
1030 	} else {
1031 		struct timespec now, timeout;
1032 
1033 		err = cache_lookup(dvp, vpp, cnp, &timeout, NULL);
1034 		getnanouptime(&now);
1035 		SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1036 		switch (err) {
1037 		case -1:		/* positive match */
1038 			if (timespeccmp(&timeout, &now, >)) {
1039 				counter_u64_add(fuse_lookup_cache_hits, 1);
1040 			} else {
1041 				/* Cache timeout */
1042 				counter_u64_add(fuse_lookup_cache_misses, 1);
1043 				bintime_clear(
1044 					&VTOFUD(*vpp)->entry_cache_timeout);
1045 				cache_purge(*vpp);
1046 				if (dvp != *vpp)
1047 					vput(*vpp);
1048 				else
1049 					vrele(*vpp);
1050 				*vpp = NULL;
1051 				break;
1052 			}
1053 			return 0;
1054 
1055 		case 0:		/* no match in cache */
1056 			counter_u64_add(fuse_lookup_cache_misses, 1);
1057 			break;
1058 
1059 		case ENOENT:		/* negative match */
1060 			getnanouptime(&now);
1061 			if (timespeccmp(&timeout, &now, <=)) {
1062 				/* Cache timeout */
1063 				cache_purge_negative(dvp);
1064 				break;
1065 			}
1066 			/* fall through */
1067 		default:
1068 			return err;
1069 		}
1070 
1071 		nid = VTOI(dvp);
1072 		fdisp_init(&fdi, cnp->cn_namelen + 1);
1073 		fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred);
1074 
1075 		memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1076 		((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1077 		lookup_err = fdisp_wait_answ(&fdi);
1078 		did_lookup = true;
1079 
1080 		if (!lookup_err) {
1081 			/* lookup call succeeded */
1082 			feo = (struct fuse_entry_out *)fdi.answ;
1083 			nid = feo->nodeid;
1084 			if (nid == 0) {
1085 				/* zero nodeid means ENOENT and cache it */
1086 				struct timespec timeout;
1087 
1088 				fdi.answ_stat = ENOENT;
1089 				lookup_err = ENOENT;
1090 				if (cnp->cn_flags & MAKEENTRY) {
1091 					fuse_validity_2_timespec(feo, &timeout);
1092 					cache_enter_time(dvp, *vpp, cnp,
1093 						&timeout, NULL);
1094 				}
1095 			} else if (nid == FUSE_ROOT_ID) {
1096 				lookup_err = EINVAL;
1097 			}
1098 			vtyp = IFTOVT(feo->attr.mode);
1099 			filesize = feo->attr.size;
1100 		}
1101 		if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1102 			fdisp_destroy(&fdi);
1103 			return lookup_err;
1104 		}
1105 	}
1106 	/* lookup_err, if non-zero, must be ENOENT at this point */
1107 
1108 	if (lookup_err) {
1109 		/* Entry not found */
1110 		if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1111 			err = fuse_internal_access(dvp, VWRITE, td, cred);
1112 			if (!err) {
1113 				/*
1114 				 * Set the SAVENAME flag to hold onto the
1115 				 * pathname for use later in VOP_CREATE or
1116 				 * VOP_RENAME.
1117 				 */
1118 				cnp->cn_flags |= SAVENAME;
1119 
1120 				err = EJUSTRETURN;
1121 			}
1122 		} else {
1123 			err = ENOENT;
1124 		}
1125 	} else {
1126 		/* Entry was found */
1127 		if (flags & ISDOTDOT) {
1128 			struct fuse_lookup_alloc_arg flaa;
1129 
1130 			flaa.nid = nid;
1131 			flaa.feo = feo;
1132 			flaa.cnp = cnp;
1133 			flaa.vtyp = vtyp;
1134 			err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1135 				&vp);
1136 			*vpp = vp;
1137 		} else if (nid == VTOI(dvp)) {
1138 			vref(dvp);
1139 			*vpp = dvp;
1140 		} else {
1141 			struct fuse_vnode_data *fvdat;
1142 			struct vattr *vap;
1143 
1144 			err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1145 			    &vp, cnp, vtyp);
1146 			if (err)
1147 				goto out;
1148 			*vpp = vp;
1149 
1150 			/*
1151 			 * In the case where we are looking up a FUSE node
1152 			 * represented by an existing cached vnode, and the
1153 			 * true size reported by FUSE_LOOKUP doesn't match
1154 			 * the vnode's cached size, then any cached writes
1155 			 * beyond the file's current size are lost.
1156 			 *
1157 			 * We can get here:
1158 			 * * following attribute cache expiration, or
1159 			 * * due a bug in the daemon, or
1160 			 */
1161 			fvdat = VTOFUD(vp);
1162 			if (vnode_isreg(vp) &&
1163 			    ((filesize != fvdat->cached_attrs.va_size &&
1164 			      fvdat->flag & FN_SIZECHANGE) ||
1165 			     ((vap = VTOVA(vp)) &&
1166 			      filesize != vap->va_size)))
1167 			{
1168 				SDT_PROBE2(fusefs, , vnops, lookup_cache_incoherent, vp, feo);
1169 				fvdat->flag &= ~FN_SIZECHANGE;
1170 				/*
1171 				 * The server changed the file's size even
1172 				 * though we had it cached, or had dirty writes
1173 				 * in the WB cache!
1174 				 */
1175 				printf("%s: cache incoherent on %s!  "
1176 		    		    "Buggy FUSE server detected.  To prevent "
1177 				    "data corruption, disable the data cache "
1178 				    "by mounting with -o direct_io, or as "
1179 				    "directed otherwise by your FUSE server's "
1180 		    		    "documentation\n", __func__,
1181 				    vnode_mount(vp)->mnt_stat.f_mntonname);
1182 				int iosize = fuse_iosize(vp);
1183 				v_inval_buf_range(vp, 0, INT64_MAX, iosize);
1184 			}
1185 
1186 			MPASS(feo != NULL);
1187 			fuse_internal_cache_attrs(*vpp, &feo->attr,
1188 				feo->attr_valid, feo->attr_valid_nsec, NULL);
1189 			fuse_validity_2_bintime(feo->entry_valid,
1190 				feo->entry_valid_nsec,
1191 				&fvdat->entry_cache_timeout);
1192 
1193 			if ((nameiop == DELETE || nameiop == RENAME) &&
1194 				islastcn)
1195 			{
1196 				struct vattr dvattr;
1197 
1198 				err = fuse_internal_access(dvp, VWRITE, td,
1199 					cred);
1200 				if (err != 0)
1201 					goto out;
1202 				/*
1203 				 * if the parent's sticky bit is set, check
1204 				 * whether we're allowed to remove the file.
1205 				 * Need to figure out the vnode locking to make
1206 				 * this work.
1207 				 */
1208 				fuse_internal_getattr(dvp, &dvattr, cred, td);
1209 				if ((dvattr.va_mode & S_ISTXT) &&
1210 					fuse_internal_access(dvp, VADMIN, td,
1211 						cred) &&
1212 					fuse_internal_access(*vpp, VADMIN, td,
1213 						cred)) {
1214 					err = EPERM;
1215 					goto out;
1216 				}
1217 			}
1218 
1219 			if (islastcn && (
1220 				(nameiop == DELETE) ||
1221 				(nameiop == RENAME && wantparent))) {
1222 				cnp->cn_flags |= SAVENAME;
1223 			}
1224 
1225 		}
1226 	}
1227 out:
1228 	if (err) {
1229 		if (vp != NULL && dvp != vp)
1230 			vput(vp);
1231 		else if (vp != NULL)
1232 			vrele(vp);
1233 		*vpp = NULL;
1234 	}
1235 	if (did_lookup)
1236 		fdisp_destroy(&fdi);
1237 
1238 	return err;
1239 }
1240 
1241 /*
1242     struct vnop_mkdir_args {
1243 	struct vnode *a_dvp;
1244 	struct vnode **a_vpp;
1245 	struct componentname *a_cnp;
1246 	struct vattr *a_vap;
1247     };
1248 */
1249 static int
1250 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1251 {
1252 	struct vnode *dvp = ap->a_dvp;
1253 	struct vnode **vpp = ap->a_vpp;
1254 	struct componentname *cnp = ap->a_cnp;
1255 	struct vattr *vap = ap->a_vap;
1256 
1257 	struct fuse_mkdir_in fmdi;
1258 
1259 	if (fuse_isdeadfs(dvp)) {
1260 		return ENXIO;
1261 	}
1262 	fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1263 	fmdi.umask = curthread->td_proc->p_fd->fd_cmask;
1264 
1265 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1266 	    sizeof(fmdi), VDIR));
1267 }
1268 
1269 /*
1270     struct vnop_mknod_args {
1271 	struct vnode *a_dvp;
1272 	struct vnode **a_vpp;
1273 	struct componentname *a_cnp;
1274 	struct vattr *a_vap;
1275     };
1276 */
1277 static int
1278 fuse_vnop_mknod(struct vop_mknod_args *ap)
1279 {
1280 
1281 	struct vnode *dvp = ap->a_dvp;
1282 	struct vnode **vpp = ap->a_vpp;
1283 	struct componentname *cnp = ap->a_cnp;
1284 	struct vattr *vap = ap->a_vap;
1285 
1286 	if (fuse_isdeadfs(dvp))
1287 		return ENXIO;
1288 
1289 	return fuse_internal_mknod(dvp, vpp, cnp, vap);
1290 }
1291 
1292 /*
1293     struct vop_open_args {
1294 	struct vnode *a_vp;
1295 	int  a_mode;
1296 	struct ucred *a_cred;
1297 	struct thread *a_td;
1298 	int a_fdidx; / struct file *a_fp;
1299     };
1300 */
1301 static int
1302 fuse_vnop_open(struct vop_open_args *ap)
1303 {
1304 	struct vnode *vp = ap->a_vp;
1305 	int a_mode = ap->a_mode;
1306 	struct thread *td = ap->a_td;
1307 	struct ucred *cred = ap->a_cred;
1308 	pid_t pid = td->td_proc->p_pid;
1309 	struct fuse_vnode_data *fvdat;
1310 
1311 	if (fuse_isdeadfs(vp))
1312 		return ENXIO;
1313 	if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1314 		return (EOPNOTSUPP);
1315 	if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1316 		return EINVAL;
1317 
1318 	fvdat = VTOFUD(vp);
1319 
1320 	if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1321 		fuse_vnode_open(vp, 0, td);
1322 		return 0;
1323 	}
1324 
1325 	return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1326 }
1327 
1328 static int
1329 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1330 {
1331 
1332 	switch (ap->a_name) {
1333 	case _PC_FILESIZEBITS:
1334 		*ap->a_retval = 64;
1335 		return (0);
1336 	case _PC_NAME_MAX:
1337 		*ap->a_retval = NAME_MAX;
1338 		return (0);
1339 	case _PC_LINK_MAX:
1340 		*ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1341 		return (0);
1342 	case _PC_SYMLINK_MAX:
1343 		*ap->a_retval = MAXPATHLEN;
1344 		return (0);
1345 	case _PC_NO_TRUNC:
1346 		*ap->a_retval = 1;
1347 		return (0);
1348 	default:
1349 		return (vop_stdpathconf(ap));
1350 	}
1351 }
1352 
1353 /*
1354     struct vnop_read_args {
1355 	struct vnode *a_vp;
1356 	struct uio *a_uio;
1357 	int  a_ioflag;
1358 	struct ucred *a_cred;
1359     };
1360 */
1361 static int
1362 fuse_vnop_read(struct vop_read_args *ap)
1363 {
1364 	struct vnode *vp = ap->a_vp;
1365 	struct uio *uio = ap->a_uio;
1366 	int ioflag = ap->a_ioflag;
1367 	struct ucred *cred = ap->a_cred;
1368 	pid_t pid = curthread->td_proc->p_pid;
1369 
1370 	if (fuse_isdeadfs(vp)) {
1371 		return ENXIO;
1372 	}
1373 
1374 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1375 		ioflag |= IO_DIRECT;
1376 	}
1377 
1378 	return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1379 }
1380 
1381 /*
1382     struct vnop_readdir_args {
1383 	struct vnode *a_vp;
1384 	struct uio *a_uio;
1385 	struct ucred *a_cred;
1386 	int *a_eofflag;
1387 	int *a_ncookies;
1388 	u_long **a_cookies;
1389     };
1390 */
1391 static int
1392 fuse_vnop_readdir(struct vop_readdir_args *ap)
1393 {
1394 	struct vnode *vp = ap->a_vp;
1395 	struct uio *uio = ap->a_uio;
1396 	struct ucred *cred = ap->a_cred;
1397 	struct fuse_filehandle *fufh = NULL;
1398 	struct fuse_iov cookediov;
1399 	int err = 0;
1400 	u_long *cookies;
1401 	off_t startoff;
1402 	ssize_t tresid;
1403 	int ncookies;
1404 	bool closefufh = false;
1405 	pid_t pid = curthread->td_proc->p_pid;
1406 
1407 	if (ap->a_eofflag)
1408 		*ap->a_eofflag = 0;
1409 	if (fuse_isdeadfs(vp)) {
1410 		return ENXIO;
1411 	}
1412 	if (				/* XXXIP ((uio_iovcnt(uio) > 1)) || */
1413 	    (uio_resid(uio) < sizeof(struct dirent))) {
1414 		return EINVAL;
1415 	}
1416 
1417 	tresid = uio->uio_resid;
1418 	startoff = uio->uio_offset;
1419 	err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1420 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1421 		/*
1422 		 * nfsd will do VOP_READDIR without first doing VOP_OPEN.  We
1423 		 * must implicitly open the directory here
1424 		 */
1425 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1426 		if (err == 0) {
1427 			/*
1428 			 * When a directory is opened, it must be read from
1429 			 * the beginning.  Hopefully, the "startoff" still
1430 			 * exists as an offset cookie for the directory.
1431 			 * If not, it will read the entire directory without
1432 			 * returning any entries and just return eof.
1433 			 */
1434 			uio->uio_offset = 0;
1435 		}
1436 		closefufh = true;
1437 	}
1438 	if (err)
1439 		return (err);
1440 	if (ap->a_ncookies != NULL) {
1441 		ncookies = uio->uio_resid /
1442 			(offsetof(struct dirent, d_name) + 4) + 1;
1443 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1444 		*ap->a_ncookies = ncookies;
1445 		*ap->a_cookies = cookies;
1446 	} else {
1447 		ncookies = 0;
1448 		cookies = NULL;
1449 	}
1450 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1451 	fiov_init(&cookediov, DIRCOOKEDSIZE);
1452 
1453 	err = fuse_internal_readdir(vp, uio, startoff, fufh, &cookediov,
1454 		&ncookies, cookies);
1455 
1456 	fiov_teardown(&cookediov);
1457 	if (closefufh)
1458 		fuse_filehandle_close(vp, fufh, curthread, cred);
1459 
1460 	if (ap->a_ncookies != NULL) {
1461 		if (err == 0) {
1462 			*ap->a_ncookies -= ncookies;
1463 		} else {
1464 			free(*ap->a_cookies, M_TEMP);
1465 			*ap->a_ncookies = 0;
1466 			*ap->a_cookies = NULL;
1467 		}
1468 	}
1469 	if (err == 0 && tresid == uio->uio_resid)
1470 		*ap->a_eofflag = 1;
1471 
1472 	return err;
1473 }
1474 
1475 /*
1476     struct vnop_readlink_args {
1477 	struct vnode *a_vp;
1478 	struct uio *a_uio;
1479 	struct ucred *a_cred;
1480     };
1481 */
1482 static int
1483 fuse_vnop_readlink(struct vop_readlink_args *ap)
1484 {
1485 	struct vnode *vp = ap->a_vp;
1486 	struct uio *uio = ap->a_uio;
1487 	struct ucred *cred = ap->a_cred;
1488 
1489 	struct fuse_dispatcher fdi;
1490 	int err;
1491 
1492 	if (fuse_isdeadfs(vp)) {
1493 		return ENXIO;
1494 	}
1495 	if (!vnode_islnk(vp)) {
1496 		return EINVAL;
1497 	}
1498 	fdisp_init(&fdi, 0);
1499 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1500 	if (err) {
1501 		goto out;
1502 	}
1503 	if (((char *)fdi.answ)[0] == '/' &&
1504 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1505 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1506 
1507 		err = uiomove(mpth, strlen(mpth), uio);
1508 	}
1509 	if (!err) {
1510 		err = uiomove(fdi.answ, fdi.iosize, uio);
1511 	}
1512 out:
1513 	fdisp_destroy(&fdi);
1514 	return err;
1515 }
1516 
1517 /*
1518     struct vnop_reclaim_args {
1519 	struct vnode *a_vp;
1520 	struct thread *a_td;
1521     };
1522 */
1523 static int
1524 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1525 {
1526 	struct vnode *vp = ap->a_vp;
1527 	struct thread *td = ap->a_td;
1528 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1529 	struct fuse_filehandle *fufh, *fufh_tmp;
1530 
1531 	if (!fvdat) {
1532 		panic("FUSE: no vnode data during recycling");
1533 	}
1534 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1535 		printf("FUSE: vnode being reclaimed with open fufh "
1536 			"(type=%#x)", fufh->fufh_type);
1537 		fuse_filehandle_close(vp, fufh, td, NULL);
1538 	}
1539 
1540 	if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
1541 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1542 		    fvdat->nlookup);
1543 	}
1544 	cache_purge(vp);
1545 	vfs_hash_remove(vp);
1546 	fuse_vnode_destroy(vp);
1547 
1548 	return 0;
1549 }
1550 
1551 /*
1552     struct vnop_remove_args {
1553 	struct vnode *a_dvp;
1554 	struct vnode *a_vp;
1555 	struct componentname *a_cnp;
1556     };
1557 */
1558 static int
1559 fuse_vnop_remove(struct vop_remove_args *ap)
1560 {
1561 	struct vnode *dvp = ap->a_dvp;
1562 	struct vnode *vp = ap->a_vp;
1563 	struct componentname *cnp = ap->a_cnp;
1564 
1565 	int err;
1566 
1567 	if (fuse_isdeadfs(vp)) {
1568 		return ENXIO;
1569 	}
1570 	if (vnode_isdir(vp)) {
1571 		return EPERM;
1572 	}
1573 
1574 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1575 
1576 	return err;
1577 }
1578 
1579 /*
1580     struct vnop_rename_args {
1581 	struct vnode *a_fdvp;
1582 	struct vnode *a_fvp;
1583 	struct componentname *a_fcnp;
1584 	struct vnode *a_tdvp;
1585 	struct vnode *a_tvp;
1586 	struct componentname *a_tcnp;
1587     };
1588 */
1589 static int
1590 fuse_vnop_rename(struct vop_rename_args *ap)
1591 {
1592 	struct vnode *fdvp = ap->a_fdvp;
1593 	struct vnode *fvp = ap->a_fvp;
1594 	struct componentname *fcnp = ap->a_fcnp;
1595 	struct vnode *tdvp = ap->a_tdvp;
1596 	struct vnode *tvp = ap->a_tvp;
1597 	struct componentname *tcnp = ap->a_tcnp;
1598 	struct fuse_data *data;
1599 	bool newparent = fdvp != tdvp;
1600 	bool isdir = fvp->v_type == VDIR;
1601 	int err = 0;
1602 
1603 	if (fuse_isdeadfs(fdvp)) {
1604 		return ENXIO;
1605 	}
1606 	if (fvp->v_mount != tdvp->v_mount ||
1607 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1608 		SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
1609 		err = EXDEV;
1610 		goto out;
1611 	}
1612 	cache_purge(fvp);
1613 
1614 	/*
1615 	 * FUSE library is expected to check if target directory is not
1616 	 * under the source directory in the file system tree.
1617 	 * Linux performs this check at VFS level.
1618 	 */
1619 	/*
1620 	 * If source is a directory, and it will get a new parent, user must
1621 	 * have write permission to it, so ".." can be modified.
1622 	 */
1623 	data = fuse_get_mpdata(vnode_mount(tdvp));
1624 	if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
1625 		err = fuse_internal_access(fvp, VWRITE,
1626 			tcnp->cn_thread, tcnp->cn_cred);
1627 		if (err)
1628 			goto out;
1629 	}
1630 	sx_xlock(&data->rename_lock);
1631 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1632 	if (err == 0) {
1633 		if (tdvp != fdvp)
1634 			fuse_vnode_setparent(fvp, tdvp);
1635 		if (tvp != NULL)
1636 			fuse_vnode_setparent(tvp, NULL);
1637 	}
1638 	sx_unlock(&data->rename_lock);
1639 
1640 	if (tvp != NULL && tvp != fvp) {
1641 		cache_purge(tvp);
1642 	}
1643 	if (vnode_isdir(fvp)) {
1644 		if ((tvp != NULL) && vnode_isdir(tvp)) {
1645 			cache_purge(tdvp);
1646 		}
1647 		cache_purge(fdvp);
1648 	}
1649 out:
1650 	if (tdvp == tvp) {
1651 		vrele(tdvp);
1652 	} else {
1653 		vput(tdvp);
1654 	}
1655 	if (tvp != NULL) {
1656 		vput(tvp);
1657 	}
1658 	vrele(fdvp);
1659 	vrele(fvp);
1660 
1661 	return err;
1662 }
1663 
1664 /*
1665     struct vnop_rmdir_args {
1666 	    struct vnode *a_dvp;
1667 	    struct vnode *a_vp;
1668 	    struct componentname *a_cnp;
1669     } *ap;
1670 */
1671 static int
1672 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1673 {
1674 	struct vnode *dvp = ap->a_dvp;
1675 	struct vnode *vp = ap->a_vp;
1676 
1677 	int err;
1678 
1679 	if (fuse_isdeadfs(vp)) {
1680 		return ENXIO;
1681 	}
1682 	if (VTOFUD(vp) == VTOFUD(dvp)) {
1683 		return EINVAL;
1684 	}
1685 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1686 
1687 	return err;
1688 }
1689 
1690 /*
1691     struct vnop_setattr_args {
1692 	struct vnode *a_vp;
1693 	struct vattr *a_vap;
1694 	struct ucred *a_cred;
1695 	struct thread *a_td;
1696     };
1697 */
1698 static int
1699 fuse_vnop_setattr(struct vop_setattr_args *ap)
1700 {
1701 	struct vnode *vp = ap->a_vp;
1702 	struct vattr *vap = ap->a_vap;
1703 	struct ucred *cred = ap->a_cred;
1704 	struct thread *td = curthread;
1705 	struct mount *mp;
1706 	struct fuse_data *data;
1707 	struct vattr old_va;
1708 	int dataflags;
1709 	int err = 0, err2;
1710 	accmode_t accmode = 0;
1711 	bool checkperm;
1712 	bool drop_suid = false;
1713 	gid_t cr_gid;
1714 
1715 	mp = vnode_mount(vp);
1716 	data = fuse_get_mpdata(mp);
1717 	dataflags = data->dataflags;
1718 	checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
1719 	if (cred->cr_ngroups > 0)
1720 		cr_gid = cred->cr_groups[0];
1721 	else
1722 		cr_gid = 0;
1723 
1724 	if (fuse_isdeadfs(vp)) {
1725 		return ENXIO;
1726 	}
1727 
1728 	if (vap->va_uid != (uid_t)VNOVAL) {
1729 		if (checkperm) {
1730 			/* Only root may change a file's owner */
1731 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1732 			if (err) {
1733 				/* As a special case, allow the null chown */
1734 				err2 = fuse_internal_getattr(vp, &old_va, cred,
1735 					td);
1736 				if (err2)
1737 					return (err2);
1738 				if (vap->va_uid != old_va.va_uid)
1739 					return err;
1740 				else
1741 					accmode |= VADMIN;
1742 				drop_suid = true;
1743 			} else
1744 				accmode |= VADMIN;
1745 		} else
1746 			accmode |= VADMIN;
1747 	}
1748 	if (vap->va_gid != (gid_t)VNOVAL) {
1749 		if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
1750 			drop_suid = true;
1751 		if (checkperm && !groupmember(vap->va_gid, cred))
1752 		{
1753 			/*
1754 			 * Non-root users may only chgrp to one of their own
1755 			 * groups
1756 			 */
1757 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1758 			if (err) {
1759 				/* As a special case, allow the null chgrp */
1760 				err2 = fuse_internal_getattr(vp, &old_va, cred,
1761 					td);
1762 				if (err2)
1763 					return (err2);
1764 				if (vap->va_gid != old_va.va_gid)
1765 					return err;
1766 				accmode |= VADMIN;
1767 			} else
1768 				accmode |= VADMIN;
1769 		} else
1770 			accmode |= VADMIN;
1771 	}
1772 	if (vap->va_size != VNOVAL) {
1773 		switch (vp->v_type) {
1774 		case VDIR:
1775 			return (EISDIR);
1776 		case VLNK:
1777 		case VREG:
1778 			if (vfs_isrdonly(mp))
1779 				return (EROFS);
1780 			break;
1781 		default:
1782 			/*
1783 			 * According to POSIX, the result is unspecified
1784 			 * for file types other than regular files,
1785 			 * directories and shared memory objects.  We
1786 			 * don't support shared memory objects in the file
1787 			 * system, and have dubious support for truncating
1788 			 * symlinks.  Just ignore the request in other cases.
1789 			 */
1790 			return (0);
1791 		}
1792 		/* Don't set accmode.  Permission to trunc is checked upstack */
1793 	}
1794 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1795 		if (vap->va_vaflags & VA_UTIMES_NULL)
1796 			accmode |= VWRITE;
1797 		else
1798 			accmode |= VADMIN;
1799 	}
1800 	if (drop_suid) {
1801 		if (vap->va_mode != (mode_t)VNOVAL)
1802 			vap->va_mode &= ~(S_ISUID | S_ISGID);
1803 		else {
1804 			err = fuse_internal_getattr(vp, &old_va, cred, td);
1805 			if (err)
1806 				return (err);
1807 			vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
1808 		}
1809 	}
1810 	if (vap->va_mode != (mode_t)VNOVAL) {
1811 		/* Only root may set the sticky bit on non-directories */
1812 		if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
1813 		    && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1814 			return EFTYPE;
1815 		if (checkperm && (vap->va_mode & S_ISGID)) {
1816 			err = fuse_internal_getattr(vp, &old_va, cred, td);
1817 			if (err)
1818 				return (err);
1819 			if (!groupmember(old_va.va_gid, cred)) {
1820 				err = priv_check_cred(cred, PRIV_VFS_SETGID);
1821 				if (err)
1822 					return (err);
1823 			}
1824 		}
1825 		accmode |= VADMIN;
1826 	}
1827 
1828 	if (vfs_isrdonly(mp))
1829 		return EROFS;
1830 
1831 	err = fuse_internal_access(vp, accmode, td, cred);
1832 	if (err)
1833 		return err;
1834 	else
1835 		return fuse_internal_setattr(vp, vap, td, cred);
1836 }
1837 
1838 /*
1839     struct vnop_strategy_args {
1840 	struct vnode *a_vp;
1841 	struct buf *a_bp;
1842     };
1843 */
1844 static int
1845 fuse_vnop_strategy(struct vop_strategy_args *ap)
1846 {
1847 	struct vnode *vp = ap->a_vp;
1848 	struct buf *bp = ap->a_bp;
1849 
1850 	if (!vp || fuse_isdeadfs(vp)) {
1851 		bp->b_ioflags |= BIO_ERROR;
1852 		bp->b_error = ENXIO;
1853 		bufdone(bp);
1854 		return 0;
1855 	}
1856 
1857 	/*
1858 	 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
1859 	 * fuse_io_strategy sets bp's error fields
1860 	 */
1861 	(void)fuse_io_strategy(vp, bp);
1862 
1863 	return 0;
1864 }
1865 
1866 
1867 /*
1868     struct vnop_symlink_args {
1869 	struct vnode *a_dvp;
1870 	struct vnode **a_vpp;
1871 	struct componentname *a_cnp;
1872 	struct vattr *a_vap;
1873 	char *a_target;
1874     };
1875 */
1876 static int
1877 fuse_vnop_symlink(struct vop_symlink_args *ap)
1878 {
1879 	struct vnode *dvp = ap->a_dvp;
1880 	struct vnode **vpp = ap->a_vpp;
1881 	struct componentname *cnp = ap->a_cnp;
1882 	const char *target = ap->a_target;
1883 
1884 	struct fuse_dispatcher fdi;
1885 
1886 	int err;
1887 	size_t len;
1888 
1889 	if (fuse_isdeadfs(dvp)) {
1890 		return ENXIO;
1891 	}
1892 	/*
1893 	 * Unlike the other creator type calls, here we have to create a message
1894 	 * where the name of the new entry comes first, and the data describing
1895 	 * the entry comes second.
1896 	 * Hence we can't rely on our handy fuse_internal_newentry() routine,
1897 	 * but put together the message manually and just call the core part.
1898 	 */
1899 
1900 	len = strlen(target) + 1;
1901 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1902 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1903 
1904 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1905 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1906 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1907 
1908 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1909 	fdisp_destroy(&fdi);
1910 	return err;
1911 }
1912 
1913 /*
1914     struct vnop_write_args {
1915 	struct vnode *a_vp;
1916 	struct uio *a_uio;
1917 	int  a_ioflag;
1918 	struct ucred *a_cred;
1919     };
1920 */
1921 static int
1922 fuse_vnop_write(struct vop_write_args *ap)
1923 {
1924 	struct vnode *vp = ap->a_vp;
1925 	struct uio *uio = ap->a_uio;
1926 	int ioflag = ap->a_ioflag;
1927 	struct ucred *cred = ap->a_cred;
1928 	pid_t pid = curthread->td_proc->p_pid;
1929 
1930 	if (fuse_isdeadfs(vp)) {
1931 		return ENXIO;
1932 	}
1933 
1934 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1935 		ioflag |= IO_DIRECT;
1936 	}
1937 
1938 	return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1939 }
1940 
1941 static daddr_t
1942 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1943 {
1944 	const int biosize = fuse_iosize(vp);
1945 
1946 	return (off / biosize);
1947 }
1948 
1949 static int
1950 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1951 {
1952 	off_t filesize;
1953 	int blksz, err;
1954 	const int biosize = fuse_iosize(vp);
1955 
1956 	err = fuse_vnode_size(vp, &filesize, NULL, NULL);
1957 	KASSERT(err == 0, ("vfs_bio_getpages can't handle errors here"));
1958 	if (err)
1959 		return biosize;
1960 
1961 	if ((off_t)lbn * biosize >= filesize) {
1962 		blksz = 0;
1963 	} else if ((off_t)(lbn + 1) * biosize > filesize) {
1964 		blksz = filesize - (off_t)lbn *biosize;
1965 	} else {
1966 		blksz = biosize;
1967 	}
1968 	return (blksz);
1969 }
1970 
1971 /*
1972     struct vnop_getpages_args {
1973 	struct vnode *a_vp;
1974 	vm_page_t *a_m;
1975 	int a_count;
1976 	int a_reqpage;
1977     };
1978 */
1979 static int
1980 fuse_vnop_getpages(struct vop_getpages_args *ap)
1981 {
1982 	struct vnode *vp = ap->a_vp;
1983 
1984 	if (!fsess_opt_mmap(vnode_mount(vp))) {
1985 		SDT_PROBE2(fusefs, , vnops, trace, 1,
1986 			"called on non-cacheable vnode??\n");
1987 		return (VM_PAGER_ERROR);
1988 	}
1989 
1990 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1991 	    ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
1992 }
1993 
1994 static const char extattr_namespace_separator = '.';
1995 
1996 /*
1997     struct vop_getextattr_args {
1998 	struct vop_generic_args a_gen;
1999 	struct vnode *a_vp;
2000 	int a_attrnamespace;
2001 	const char *a_name;
2002 	struct uio *a_uio;
2003 	size_t *a_size;
2004 	struct ucred *a_cred;
2005 	struct thread *a_td;
2006     };
2007 */
2008 static int
2009 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2010 {
2011 	struct vnode *vp = ap->a_vp;
2012 	struct uio *uio = ap->a_uio;
2013 	struct fuse_dispatcher fdi;
2014 	struct fuse_getxattr_in *get_xattr_in;
2015 	struct fuse_getxattr_out *get_xattr_out;
2016 	struct mount *mp = vnode_mount(vp);
2017 	struct thread *td = ap->a_td;
2018 	struct ucred *cred = ap->a_cred;
2019 	char *prefix;
2020 	char *attr_str;
2021 	size_t len;
2022 	int err;
2023 
2024 	if (fuse_isdeadfs(vp))
2025 		return (ENXIO);
2026 
2027 	if (!fsess_isimpl(mp, FUSE_GETXATTR))
2028 		return EOPNOTSUPP;
2029 
2030 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2031 	if (err)
2032 		return err;
2033 
2034 	/* Default to looking for user attributes. */
2035 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2036 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2037 	else
2038 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2039 
2040 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2041 	    strlen(ap->a_name) + 1;
2042 
2043 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2044 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2045 
2046 	get_xattr_in = fdi.indata;
2047 	/*
2048 	 * Check to see whether we're querying the available size or
2049 	 * issuing the actual request.  If we pass in 0, we get back struct
2050 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2051 	 * that much data, without the struct fuse_getxattr_out header.
2052 	 */
2053 	if (uio == NULL)
2054 		get_xattr_in->size = 0;
2055 	else
2056 		get_xattr_in->size = uio->uio_resid;
2057 
2058 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2059 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2060 	    ap->a_name);
2061 
2062 	err = fdisp_wait_answ(&fdi);
2063 	if (err != 0) {
2064 		if (err == ENOSYS) {
2065 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2066 			err = EOPNOTSUPP;
2067 		}
2068 		goto out;
2069 	}
2070 
2071 	get_xattr_out = fdi.answ;
2072 
2073 	if (ap->a_size != NULL)
2074 		*ap->a_size = get_xattr_out->size;
2075 
2076 	if (uio != NULL)
2077 		err = uiomove(fdi.answ, fdi.iosize, uio);
2078 
2079 out:
2080 	fdisp_destroy(&fdi);
2081 	return (err);
2082 }
2083 
2084 /*
2085     struct vop_setextattr_args {
2086 	struct vop_generic_args a_gen;
2087 	struct vnode *a_vp;
2088 	int a_attrnamespace;
2089 	const char *a_name;
2090 	struct uio *a_uio;
2091 	struct ucred *a_cred;
2092 	struct thread *a_td;
2093     };
2094 */
2095 static int
2096 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2097 {
2098 	struct vnode *vp = ap->a_vp;
2099 	struct uio *uio = ap->a_uio;
2100 	struct fuse_dispatcher fdi;
2101 	struct fuse_setxattr_in *set_xattr_in;
2102 	struct mount *mp = vnode_mount(vp);
2103 	struct thread *td = ap->a_td;
2104 	struct ucred *cred = ap->a_cred;
2105 	char *prefix;
2106 	size_t len;
2107 	char *attr_str;
2108 	int err;
2109 
2110 	if (fuse_isdeadfs(vp))
2111 		return (ENXIO);
2112 
2113 	if (!fsess_isimpl(mp, FUSE_SETXATTR))
2114 		return EOPNOTSUPP;
2115 
2116 	if (vfs_isrdonly(mp))
2117 		return EROFS;
2118 
2119 	/* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2120 	if (ap->a_uio == NULL) {
2121 		/*
2122 		 * If we got here as fallback from VOP_DELETEEXTATTR, then
2123 		 * return EOPNOTSUPP.
2124 		 */
2125 		if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2126 			return (EOPNOTSUPP);
2127 		else
2128 			return (EINVAL);
2129 	}
2130 
2131 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2132 		VWRITE);
2133 	if (err)
2134 		return err;
2135 
2136 	/* Default to looking for user attributes. */
2137 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2138 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2139 	else
2140 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2141 
2142 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2143 	    strlen(ap->a_name) + 1;
2144 
2145 	fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2146 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2147 
2148 	set_xattr_in = fdi.indata;
2149 	set_xattr_in->size = uio->uio_resid;
2150 
2151 	attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2152 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2153 	    ap->a_name);
2154 
2155 	err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2156 	    uio->uio_resid, uio);
2157 	if (err != 0) {
2158 		goto out;
2159 	}
2160 
2161 	err = fdisp_wait_answ(&fdi);
2162 
2163 	if (err == ENOSYS) {
2164 		fsess_set_notimpl(mp, FUSE_SETXATTR);
2165 		err = EOPNOTSUPP;
2166 	}
2167 	if (err == ERESTART) {
2168 		/* Can't restart after calling uiomove */
2169 		err = EINTR;
2170 	}
2171 
2172 out:
2173 	fdisp_destroy(&fdi);
2174 	return (err);
2175 }
2176 
2177 /*
2178  * The Linux / FUSE extended attribute list is simply a collection of
2179  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2180  * byte length followed by a non-NUL terminated string.  So, this allows
2181  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2182  * Linux attribute names are reported with the namespace as a prefix (e.g.
2183  * "user.attribute_name"), but in FreeBSD they are reported without the
2184  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2185  *
2186  * user.attr_name1\0user.attr_name2\0
2187  *
2188  * to:
2189  *
2190  * <num>attr_name1<num>attr_name2
2191  *
2192  * Where "<num>" is a single byte number of characters in the attribute name.
2193  *
2194  * Args:
2195  * prefix - exattr namespace prefix string
2196  * list, list_len - input list with namespace prefixes
2197  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2198  */
2199 static int
2200 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2201     char *bsd_list, int *bsd_list_len)
2202 {
2203 	int len, pos, dist_to_next, prefix_len;
2204 
2205 	pos = 0;
2206 	*bsd_list_len = 0;
2207 	prefix_len = strlen(prefix);
2208 
2209 	while (pos < list_len && list[pos] != '\0') {
2210 		dist_to_next = strlen(&list[pos]) + 1;
2211 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2212 		    list[pos + prefix_len] == extattr_namespace_separator) {
2213 			len = dist_to_next -
2214 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2215 			if (len >= EXTATTR_MAXNAMELEN)
2216 				return (ENAMETOOLONG);
2217 
2218 			bsd_list[*bsd_list_len] = len;
2219 			memcpy(&bsd_list[*bsd_list_len + 1],
2220 			    &list[pos + prefix_len +
2221 			    sizeof(extattr_namespace_separator)], len);
2222 
2223 			*bsd_list_len += len + 1;
2224 		}
2225 
2226 		pos += dist_to_next;
2227 	}
2228 
2229 	return (0);
2230 }
2231 
2232 /*
2233  * List extended attributes
2234  *
2235  * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
2236  * has a number of differences compared to its FreeBSD equivalent,
2237  * extattr_list_file:
2238  *
2239  * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
2240  *   whereas listxattr(2) only returns attributes for a single namespace
2241  * - FUSE_LISTXATTR prepends each attribute name with "namespace."
2242  * - If the provided buffer is not large enough to hold the result,
2243  *   FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
2244  *   return as many results as will fit.
2245  */
2246 /*
2247     struct vop_listextattr_args {
2248 	struct vop_generic_args a_gen;
2249 	struct vnode *a_vp;
2250 	int a_attrnamespace;
2251 	struct uio *a_uio;
2252 	size_t *a_size;
2253 	struct ucred *a_cred;
2254 	struct thread *a_td;
2255     };
2256 */
2257 static int
2258 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2259 {
2260 	struct vnode *vp = ap->a_vp;
2261 	struct uio *uio = ap->a_uio;
2262 	struct fuse_dispatcher fdi;
2263 	struct fuse_listxattr_in *list_xattr_in;
2264 	struct fuse_listxattr_out *list_xattr_out;
2265 	struct mount *mp = vnode_mount(vp);
2266 	struct thread *td = ap->a_td;
2267 	struct ucred *cred = ap->a_cred;
2268 	char *prefix;
2269 	char *bsd_list = NULL;
2270 	char *linux_list;
2271 	int bsd_list_len;
2272 	int linux_list_len;
2273 	int err;
2274 
2275 	if (fuse_isdeadfs(vp))
2276 		return (ENXIO);
2277 
2278 	if (!fsess_isimpl(mp, FUSE_LISTXATTR))
2279 		return EOPNOTSUPP;
2280 
2281 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2282 	if (err)
2283 		return err;
2284 
2285 	/*
2286 	 * Add space for a NUL and the period separator if enabled.
2287 	 * Default to looking for user attributes.
2288 	 */
2289 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2290 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2291 	else
2292 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2293 
2294 	fdisp_init(&fdi, sizeof(*list_xattr_in));
2295 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2296 
2297 	/*
2298 	 * Retrieve Linux / FUSE compatible list size.
2299 	 */
2300 	list_xattr_in = fdi.indata;
2301 	list_xattr_in->size = 0;
2302 
2303 	err = fdisp_wait_answ(&fdi);
2304 	if (err != 0) {
2305 		if (err == ENOSYS) {
2306 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
2307 			err = EOPNOTSUPP;
2308 		}
2309 		goto out;
2310 	}
2311 
2312 	list_xattr_out = fdi.answ;
2313 	linux_list_len = list_xattr_out->size;
2314 	if (linux_list_len == 0) {
2315 		if (ap->a_size != NULL)
2316 			*ap->a_size = linux_list_len;
2317 		goto out;
2318 	}
2319 
2320 	/*
2321 	 * Retrieve Linux / FUSE compatible list values.
2322 	 */
2323 	fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2324 	list_xattr_in = fdi.indata;
2325 	list_xattr_in->size = linux_list_len;
2326 
2327 	err = fdisp_wait_answ(&fdi);
2328 	if (err == ERANGE) {
2329 		/*
2330 		 * Race detected.  The attribute list must've grown since the
2331 		 * first FUSE_LISTXATTR call.  Start over.  Go all the way back
2332 		 * to userland so we can process signals, if necessary, before
2333 		 * restarting.
2334 		 */
2335 		err = ERESTART;
2336 		goto out;
2337 	} else if (err != 0)
2338 		goto out;
2339 
2340 	linux_list = fdi.answ;
2341 	/* FUSE doesn't allow the server to return more data than requested */
2342 	if (fdi.iosize > linux_list_len) {
2343 		printf("WARNING: FUSE protocol violation.  Server returned "
2344 			"more extended attribute data than requested; "
2345 			"should've returned ERANGE instead");
2346 	} else {
2347 		/* But returning less data is fine */
2348 		linux_list_len = fdi.iosize;
2349 	}
2350 
2351 	/*
2352 	 * Retrieve the BSD compatible list values.
2353 	 * The Linux / FUSE attribute list format isn't the same
2354 	 * as FreeBSD's format. So we need to transform it into
2355 	 * FreeBSD's format before giving it to the user.
2356 	 */
2357 	bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2358 	err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2359 	    bsd_list, &bsd_list_len);
2360 	if (err != 0)
2361 		goto out;
2362 
2363 	if (ap->a_size != NULL)
2364 		*ap->a_size = bsd_list_len;
2365 
2366 	if (uio != NULL)
2367 		err = uiomove(bsd_list, bsd_list_len, uio);
2368 
2369 out:
2370 	free(bsd_list, M_TEMP);
2371 	fdisp_destroy(&fdi);
2372 	return (err);
2373 }
2374 
2375 /*
2376     struct vop_deleteextattr_args {
2377 	struct vop_generic_args a_gen;
2378 	struct vnode *a_vp;
2379 	int a_attrnamespace;
2380 	const char *a_name;
2381 	struct ucred *a_cred;
2382 	struct thread *a_td;
2383     };
2384 */
2385 static int
2386 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2387 {
2388 	struct vnode *vp = ap->a_vp;
2389 	struct fuse_dispatcher fdi;
2390 	struct mount *mp = vnode_mount(vp);
2391 	struct thread *td = ap->a_td;
2392 	struct ucred *cred = ap->a_cred;
2393 	char *prefix;
2394 	size_t len;
2395 	char *attr_str;
2396 	int err;
2397 
2398 	if (fuse_isdeadfs(vp))
2399 		return (ENXIO);
2400 
2401 	if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2402 		return EOPNOTSUPP;
2403 
2404 	if (vfs_isrdonly(mp))
2405 		return EROFS;
2406 
2407 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2408 		VWRITE);
2409 	if (err)
2410 		return err;
2411 
2412 	/* Default to looking for user attributes. */
2413 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2414 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2415 	else
2416 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2417 
2418 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2419 	    strlen(ap->a_name) + 1;
2420 
2421 	fdisp_init(&fdi, len);
2422 	fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2423 
2424 	attr_str = fdi.indata;
2425 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2426 	    ap->a_name);
2427 
2428 	err = fdisp_wait_answ(&fdi);
2429 	if (err == ENOSYS) {
2430 		fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2431 		err = EOPNOTSUPP;
2432 	}
2433 
2434 	fdisp_destroy(&fdi);
2435 	return (err);
2436 }
2437 
2438 /*
2439     struct vnop_print_args {
2440 	struct vnode *a_vp;
2441     };
2442 */
2443 static int
2444 fuse_vnop_print(struct vop_print_args *ap)
2445 {
2446 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2447 
2448 	printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2449 	    (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2450 	    (uintmax_t)fvdat->nlookup,
2451 	    fvdat->flag);
2452 
2453 	return 0;
2454 }
2455 
2456 /*
2457  * Get an NFS filehandle for a FUSE file.
2458  *
2459  * This will only work for FUSE file systems that guarantee the uniqueness of
2460  * nodeid:generation, which most don't.
2461  */
2462 /*
2463 vop_vptofh {
2464 	IN struct vnode *a_vp;
2465 	IN struct fid *a_fhp;
2466 };
2467 */
2468 static int
2469 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
2470 {
2471 	struct vnode *vp = ap->a_vp;
2472 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
2473 	struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
2474 	_Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
2475 		"FUSE fid type is too big");
2476 	struct mount *mp = vnode_mount(vp);
2477 	struct fuse_data *data = fuse_get_mpdata(mp);
2478 	struct vattr va;
2479 	int err;
2480 
2481 	if (!(data->dataflags & FSESS_EXPORT_SUPPORT))
2482 		return EOPNOTSUPP;
2483 
2484 	err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
2485 	if (err)
2486 		return err;
2487 
2488 	/*ip = VTOI(ap->a_vp);*/
2489 	/*ufhp = (struct ufid *)ap->a_fhp;*/
2490 	fhp->len = sizeof(struct fuse_fid);
2491 	fhp->nid = fvdat->nid;
2492 	if (fvdat->generation <= UINT32_MAX)
2493 		fhp->gen = fvdat->generation;
2494 	else
2495 		return EOVERFLOW;
2496 	return (0);
2497 }
2498 
2499 
2500