xref: /freebsd/sys/fs/fuse/fuse_vnops.c (revision 0bf48626aaa33768078f5872b922b1487b3a9296)
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 
178 struct vop_vector fuse_vnops = {
179 	.vop_allocate =	VOP_EINVAL,
180 	.vop_default = &default_vnodeops,
181 	.vop_access = fuse_vnop_access,
182 	.vop_advlock = fuse_vnop_advlock,
183 	.vop_bmap = fuse_vnop_bmap,
184 	.vop_close = fuse_vnop_close,
185 	.vop_create = fuse_vnop_create,
186 	.vop_deleteextattr = fuse_vnop_deleteextattr,
187 	.vop_fsync = fuse_vnop_fsync,
188 	.vop_fdatasync = fuse_vnop_fdatasync,
189 	.vop_getattr = fuse_vnop_getattr,
190 	.vop_getextattr = fuse_vnop_getextattr,
191 	.vop_inactive = fuse_vnop_inactive,
192 	/*
193 	 * TODO: implement vop_ioctl after upgrading to protocol 7.16.
194 	 * FUSE_IOCTL was added in 7.11, but 32-bit compat is broken until
195 	 * 7.16.
196 	 */
197 	.vop_link = fuse_vnop_link,
198 	.vop_listextattr = fuse_vnop_listextattr,
199 	.vop_lookup = fuse_vnop_lookup,
200 	.vop_mkdir = fuse_vnop_mkdir,
201 	.vop_mknod = fuse_vnop_mknod,
202 	.vop_open = fuse_vnop_open,
203 	.vop_pathconf = fuse_vnop_pathconf,
204 	/*
205 	 * TODO: implement vop_poll after upgrading to protocol 7.21.
206 	 * FUSE_POLL was added in protocol 7.11, but it's kind of broken until
207 	 * 7.21, which adds the ability for the client to choose which poll
208 	 * events it wants, and for a client to deregister a file handle
209 	 */
210 	.vop_read = fuse_vnop_read,
211 	.vop_readdir = fuse_vnop_readdir,
212 	.vop_readlink = fuse_vnop_readlink,
213 	.vop_reclaim = fuse_vnop_reclaim,
214 	.vop_remove = fuse_vnop_remove,
215 	.vop_rename = fuse_vnop_rename,
216 	.vop_rmdir = fuse_vnop_rmdir,
217 	.vop_setattr = fuse_vnop_setattr,
218 	.vop_setextattr = fuse_vnop_setextattr,
219 	.vop_strategy = fuse_vnop_strategy,
220 	.vop_symlink = fuse_vnop_symlink,
221 	.vop_write = fuse_vnop_write,
222 	.vop_getpages = fuse_vnop_getpages,
223 	.vop_print = fuse_vnop_print,
224 	.vop_vptofh = fuse_vnop_vptofh,
225 };
226 
227 uma_zone_t fuse_pbuf_zone;
228 
229 #define fuse_vm_page_lock(m)		vm_page_lock((m));
230 #define fuse_vm_page_unlock(m)		vm_page_unlock((m));
231 #define fuse_vm_page_lock_queues()	((void)0)
232 #define fuse_vm_page_unlock_queues()	((void)0)
233 
234 /* Check permission for extattr operations, much like extattr_check_cred */
235 static int
236 fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred,
237 	struct thread *td, accmode_t accmode)
238 {
239 	struct mount *mp = vnode_mount(vp);
240 	struct fuse_data *data = fuse_get_mpdata(mp);
241 
242 	/*
243 	 * Kernel-invoked always succeeds.
244 	 */
245 	if (cred == NOCRED)
246 		return (0);
247 
248 	/*
249 	 * Do not allow privileged processes in jail to directly manipulate
250 	 * system attributes.
251 	 */
252 	switch (ns) {
253 	case EXTATTR_NAMESPACE_SYSTEM:
254 		if (data->dataflags & FSESS_DEFAULT_PERMISSIONS) {
255 			return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM));
256 		}
257 		/* FALLTHROUGH */
258 	case EXTATTR_NAMESPACE_USER:
259 		return (fuse_internal_access(vp, accmode, td, cred));
260 	default:
261 		return (EPERM);
262 	}
263 }
264 
265 /* Get a filehandle for a directory */
266 static int
267 fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp,
268 	struct ucred *cred, pid_t pid)
269 {
270 	if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0)
271 		return 0;
272 	return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid);
273 }
274 
275 /* Send FUSE_FLUSH for this vnode */
276 static int
277 fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag)
278 {
279 	struct fuse_flush_in *ffi;
280 	struct fuse_filehandle *fufh;
281 	struct fuse_dispatcher fdi;
282 	struct thread *td = curthread;
283 	struct mount *mp = vnode_mount(vp);
284 	int err;
285 
286 	if (!fsess_isimpl(vnode_mount(vp), FUSE_FLUSH))
287 		return 0;
288 
289 	err = fuse_filehandle_getrw(vp, fflag, &fufh, cred, pid);
290 	if (err)
291 		return err;
292 
293 	fdisp_init(&fdi, sizeof(*ffi));
294 	fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred);
295 	ffi = fdi.indata;
296 	ffi->fh = fufh->fh_id;
297 	/*
298 	 * If the file has a POSIX lock then we're supposed to set lock_owner.
299 	 * If not, then lock_owner is undefined.  So we may as well always set
300 	 * it.
301 	 */
302 	ffi->lock_owner = td->td_proc->p_pid;
303 
304 	err = fdisp_wait_answ(&fdi);
305 	if (err == ENOSYS) {
306 		fsess_set_notimpl(mp, FUSE_FLUSH);
307 		err = 0;
308 	}
309 	fdisp_destroy(&fdi);
310 	return err;
311 }
312 
313 /* Close wrapper for fifos.  */
314 static int
315 fuse_fifo_close(struct vop_close_args *ap)
316 {
317 	return (fifo_specops.vop_close(ap));
318 }
319 
320 /*
321     struct vnop_access_args {
322 	struct vnode *a_vp;
323 #if VOP_ACCESS_TAKES_ACCMODE_T
324 	accmode_t a_accmode;
325 #else
326 	int a_mode;
327 #endif
328 	struct ucred *a_cred;
329 	struct thread *a_td;
330     };
331 */
332 static int
333 fuse_vnop_access(struct vop_access_args *ap)
334 {
335 	struct vnode *vp = ap->a_vp;
336 	int accmode = ap->a_accmode;
337 	struct ucred *cred = ap->a_cred;
338 
339 	struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
340 
341 	int err;
342 
343 	if (fuse_isdeadfs(vp)) {
344 		if (vnode_isvroot(vp)) {
345 			return 0;
346 		}
347 		return ENXIO;
348 	}
349 	if (!(data->dataflags & FSESS_INITED)) {
350 		if (vnode_isvroot(vp)) {
351 			if (priv_check_cred(cred, PRIV_VFS_ADMIN) ||
352 			    (fuse_match_cred(data->daemoncred, cred) == 0)) {
353 				return 0;
354 			}
355 		}
356 		return EBADF;
357 	}
358 	if (vnode_islnk(vp)) {
359 		return 0;
360 	}
361 
362 	err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred);
363 	return err;
364 }
365 
366 /*
367  * struct vop_advlock_args {
368  *	struct vop_generic_args a_gen;
369  *	struct vnode *a_vp;
370  *	void *a_id;
371  *	int a_op;
372  *	struct flock *a_fl;
373  *	int a_flags;
374  * }
375  */
376 static int
377 fuse_vnop_advlock(struct vop_advlock_args *ap)
378 {
379 	struct vnode *vp = ap->a_vp;
380 	struct flock *fl = ap->a_fl;
381 	struct thread *td = curthread;
382 	struct ucred *cred = td->td_ucred;
383 	pid_t pid = td->td_proc->p_pid;
384 	struct fuse_filehandle *fufh;
385 	struct fuse_dispatcher fdi;
386 	struct fuse_lk_in *fli;
387 	struct fuse_lk_out *flo;
388 	enum fuse_opcode op;
389 	int dataflags, err;
390 	int flags = ap->a_flags;
391 
392 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
393 
394 	if (fuse_isdeadfs(vp)) {
395 		return ENXIO;
396 	}
397 
398 	if (!(dataflags & FSESS_POSIX_LOCKS))
399 		return vop_stdadvlock(ap);
400 	/* FUSE doesn't properly support flock until protocol 7.17 */
401 	if (flags & F_FLOCK)
402 		return vop_stdadvlock(ap);
403 
404 	err = fuse_filehandle_get_anyflags(vp, &fufh, cred, pid);
405 	if (err)
406 		return err;
407 
408 	fdisp_init(&fdi, sizeof(*fli));
409 
410 	switch(ap->a_op) {
411 	case F_GETLK:
412 		op = FUSE_GETLK;
413 		break;
414 	case F_SETLK:
415 		op = FUSE_SETLK;
416 		break;
417 	case F_SETLKW:
418 		op = FUSE_SETLKW;
419 		break;
420 	default:
421 		return EINVAL;
422 	}
423 
424 	fdisp_make_vp(&fdi, op, vp, td, cred);
425 	fli = fdi.indata;
426 	fli->fh = fufh->fh_id;
427 	fli->owner = fl->l_pid;
428 	fli->lk.start = fl->l_start;
429 	if (fl->l_len != 0)
430 		fli->lk.end = fl->l_start + fl->l_len - 1;
431 	else
432 		fli->lk.end = INT64_MAX;
433 	fli->lk.type = fl->l_type;
434 	fli->lk.pid = fl->l_pid;
435 
436 	err = fdisp_wait_answ(&fdi);
437 	fdisp_destroy(&fdi);
438 
439 	if (err == 0 && op == FUSE_GETLK) {
440 		flo = fdi.answ;
441 		fl->l_type = flo->lk.type;
442 		fl->l_pid = flo->lk.pid;
443 		if (flo->lk.type != F_UNLCK) {
444 			fl->l_start = flo->lk.start;
445 			if (flo->lk.end == INT64_MAX)
446 				fl->l_len = 0;
447 			else
448 				fl->l_len = flo->lk.end - flo->lk.start + 1;
449 			fl->l_start = flo->lk.start;
450 		}
451 	}
452 
453 	return err;
454 }
455 
456 /* {
457 	struct vnode *a_vp;
458 	daddr_t a_bn;
459 	struct bufobj **a_bop;
460 	daddr_t *a_bnp;
461 	int *a_runp;
462 	int *a_runb;
463 } */
464 static int
465 fuse_vnop_bmap(struct vop_bmap_args *ap)
466 {
467 	struct vnode *vp = ap->a_vp;
468 	struct bufobj **bo = ap->a_bop;
469 	struct thread *td = curthread;
470 	struct mount *mp;
471 	struct fuse_dispatcher fdi;
472 	struct fuse_bmap_in *fbi;
473 	struct fuse_bmap_out *fbo;
474 	struct fuse_data *data;
475 	uint64_t biosize;
476 	off_t filesize;
477 	daddr_t lbn = ap->a_bn;
478 	daddr_t *pbn = ap->a_bnp;
479 	int *runp = ap->a_runp;
480 	int *runb = ap->a_runb;
481 	int error = 0;
482 	int maxrun;
483 
484 	if (fuse_isdeadfs(vp)) {
485 		return ENXIO;
486 	}
487 
488 	mp = vnode_mount(vp);
489 	data = fuse_get_mpdata(mp);
490 	biosize = fuse_iosize(vp);
491 	maxrun = MIN(vp->v_mount->mnt_iosize_max / biosize - 1,
492 		data->max_readahead_blocks);
493 
494 	if (bo != NULL)
495 		*bo = &vp->v_bufobj;
496 
497 	/*
498 	 * The FUSE_BMAP operation does not include the runp and runb
499 	 * variables, so we must guess.  Report nonzero contiguous runs so
500 	 * cluster_read will combine adjacent reads.  It's worthwhile to reduce
501 	 * upcalls even if we don't know the true physical layout of the file.
502 	 *
503 	 * FUSE file systems may opt out of read clustering in two ways:
504 	 * * mounting with -onoclusterr
505 	 * * Setting max_readahead <= maxbcachebuf during FUSE_INIT
506 	 */
507 	if (runb != NULL)
508 		*runb = MIN(lbn, maxrun);
509 	if (runp != NULL) {
510 		error = fuse_vnode_size(vp, &filesize, td->td_ucred, td);
511 		if (error == 0)
512 			*runp = MIN(MAX(0, filesize / biosize - lbn - 1),
513 				    maxrun);
514 		else
515 			*runp = 0;
516 	}
517 
518 	if (fsess_isimpl(mp, FUSE_BMAP)) {
519 		fdisp_init(&fdi, sizeof(*fbi));
520 		fdisp_make_vp(&fdi, FUSE_BMAP, vp, td, td->td_ucred);
521 		fbi = fdi.indata;
522 		fbi->block = lbn;
523 		fbi->blocksize = biosize;
524 		error = fdisp_wait_answ(&fdi);
525 		if (error == ENOSYS) {
526 			fdisp_destroy(&fdi);
527 			fsess_set_notimpl(mp, FUSE_BMAP);
528 			error = 0;
529 		} else {
530 			fbo = fdi.answ;
531 			if (error == 0 && pbn != NULL)
532 				*pbn = fbo->block;
533 			fdisp_destroy(&fdi);
534 			return error;
535 		}
536 	}
537 
538 	/* If the daemon doesn't support BMAP, make up a sensible default */
539 	if (pbn != NULL)
540 		*pbn = lbn * btodb(biosize);
541 	return (error);
542 }
543 
544 /*
545     struct vop_close_args {
546 	struct vnode *a_vp;
547 	int  a_fflag;
548 	struct ucred *a_cred;
549 	struct thread *a_td;
550     };
551 */
552 static int
553 fuse_vnop_close(struct vop_close_args *ap)
554 {
555 	struct vnode *vp = ap->a_vp;
556 	struct ucred *cred = ap->a_cred;
557 	int fflag = ap->a_fflag;
558 	struct thread *td = ap->a_td;
559 	pid_t pid = td->td_proc->p_pid;
560 	int err = 0;
561 
562 	if (fuse_isdeadfs(vp))
563 		return 0;
564 	if (vnode_isdir(vp))
565 		return 0;
566 	if (fflag & IO_NDELAY)
567 		return 0;
568 
569 	err = fuse_flush(vp, cred, pid, fflag);
570 	/* TODO: close the file handle, if we're sure it's no longer used */
571 	if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
572 		fuse_vnode_savesize(vp, cred, td->td_proc->p_pid);
573 	}
574 	return err;
575 }
576 
577 static void
578 fdisp_make_mknod_for_fallback(
579 	struct fuse_dispatcher *fdip,
580 	struct componentname *cnp,
581 	struct vnode *dvp,
582 	uint64_t parentnid,
583 	struct thread *td,
584 	struct ucred *cred,
585 	mode_t mode,
586 	enum fuse_opcode *op)
587 {
588 	struct fuse_mknod_in *fmni;
589 
590 	fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1);
591 	*op = FUSE_MKNOD;
592 	fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred);
593 	fmni = fdip->indata;
594 	fmni->mode = mode;
595 	fmni->rdev = 0;
596 	memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr,
597 	    cnp->cn_namelen);
598 	((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0';
599 }
600 /*
601     struct vnop_create_args {
602 	struct vnode *a_dvp;
603 	struct vnode **a_vpp;
604 	struct componentname *a_cnp;
605 	struct vattr *a_vap;
606     };
607 */
608 static int
609 fuse_vnop_create(struct vop_create_args *ap)
610 {
611 	struct vnode *dvp = ap->a_dvp;
612 	struct vnode **vpp = ap->a_vpp;
613 	struct componentname *cnp = ap->a_cnp;
614 	struct vattr *vap = ap->a_vap;
615 	struct thread *td = cnp->cn_thread;
616 	struct ucred *cred = cnp->cn_cred;
617 
618 	struct fuse_data *data;
619 	struct fuse_create_in *fci;
620 	struct fuse_entry_out *feo;
621 	struct fuse_open_out *foo;
622 	struct fuse_dispatcher fdi, fdi2;
623 	struct fuse_dispatcher *fdip = &fdi;
624 	struct fuse_dispatcher *fdip2 = NULL;
625 
626 	int err;
627 
628 	struct mount *mp = vnode_mount(dvp);
629 	data = fuse_get_mpdata(mp);
630 	uint64_t parentnid = VTOFUD(dvp)->nid;
631 	mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
632 	enum fuse_opcode op;
633 	int flags;
634 
635 	if (fuse_isdeadfs(dvp))
636 		return ENXIO;
637 
638 	/* FUSE expects sockets to be created with FUSE_MKNOD */
639 	if (vap->va_type == VSOCK)
640 		return fuse_internal_mknod(dvp, vpp, cnp, vap);
641 
642 	/*
643 	 * VOP_CREATE doesn't tell us the open(2) flags, so we guess.  Only a
644 	 * writable mode makes sense, and we might as well include readability
645 	 * too.
646 	 */
647 	flags = O_RDWR;
648 
649 	bzero(&fdi, sizeof(fdi));
650 
651 	if (vap->va_type != VREG)
652 		return (EINVAL);
653 
654 	if (!fsess_isimpl(mp, FUSE_CREATE) || vap->va_type == VSOCK) {
655 		/* Fallback to FUSE_MKNOD/FUSE_OPEN */
656 		fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td,
657 			cred, mode, &op);
658 	} else {
659 		/* Use FUSE_CREATE */
660 		size_t insize;
661 
662 		op = FUSE_CREATE;
663 		fdisp_init(fdip, sizeof(*fci) + cnp->cn_namelen + 1);
664 		fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred);
665 		fci = fdip->indata;
666 		fci->mode = mode;
667 		fci->flags = O_CREAT | flags;
668 		if (fuse_libabi_geq(data, 7, 12)) {
669 			insize = sizeof(*fci);
670 			fci->umask = td->td_proc->p_fd->fd_cmask;
671 		} else {
672 			insize = sizeof(struct fuse_open_in);
673 		}
674 
675 		memcpy((char *)fdip->indata + insize, cnp->cn_nameptr,
676 		    cnp->cn_namelen);
677 		((char *)fdip->indata)[insize + cnp->cn_namelen] = '\0';
678 	}
679 
680 	err = fdisp_wait_answ(fdip);
681 
682 	if (err) {
683 		if (err == ENOSYS && op == FUSE_CREATE) {
684 			fsess_set_notimpl(mp, FUSE_CREATE);
685 			fdisp_destroy(fdip);
686 			fdisp_make_mknod_for_fallback(fdip, cnp, dvp,
687 				parentnid, td, cred, mode, &op);
688 			err = fdisp_wait_answ(fdip);
689 		}
690 		if (err)
691 			goto out;
692 	}
693 
694 	feo = fdip->answ;
695 
696 	if ((err = fuse_internal_checkentry(feo, vap->va_type))) {
697 		goto out;
698 	}
699 
700 	if (op == FUSE_CREATE) {
701 		foo = (struct fuse_open_out*)(feo + 1);
702 	} else {
703 		/* Issue a separate FUSE_OPEN */
704 		struct fuse_open_in *foi;
705 
706 		fdip2 = &fdi2;
707 		fdisp_init(fdip2, sizeof(*foi));
708 		fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td,
709 			cred);
710 		foi = fdip2->indata;
711 		foi->flags = flags;
712 		err = fdisp_wait_answ(fdip2);
713 		if (err)
714 			goto out;
715 		foo = fdip2->answ;
716 	}
717 	err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vap->va_type);
718 	if (err) {
719 		struct fuse_release_in *fri;
720 		uint64_t nodeid = feo->nodeid;
721 		uint64_t fh_id = foo->fh;
722 
723 		fdisp_init(fdip, sizeof(*fri));
724 		fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
725 		fri = fdip->indata;
726 		fri->fh = fh_id;
727 		fri->flags = flags;
728 		fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
729 		fuse_insert_message(fdip->tick, false);
730 		goto out;
731 	}
732 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
733 	fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid,
734 		feo->attr_valid_nsec, NULL);
735 
736 	fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo);
737 	fuse_vnode_open(*vpp, foo->open_flags, td);
738 	/*
739 	 * Purge the parent's attribute cache because the daemon should've
740 	 * updated its mtime and ctime
741 	 */
742 	fuse_vnode_clear_attr_cache(dvp);
743 	cache_purge_negative(dvp);
744 
745 out:
746 	if (fdip2)
747 		fdisp_destroy(fdip2);
748 	fdisp_destroy(fdip);
749 	return err;
750 }
751 
752 /*
753     struct vnop_fdatasync_args {
754 	struct vop_generic_args a_gen;
755 	struct vnode * a_vp;
756 	struct thread * a_td;
757     };
758 */
759 static int
760 fuse_vnop_fdatasync(struct vop_fdatasync_args *ap)
761 {
762 	struct vnode *vp = ap->a_vp;
763 	struct thread *td = ap->a_td;
764 	int waitfor = MNT_WAIT;
765 
766 	int err = 0;
767 
768 	if (fuse_isdeadfs(vp)) {
769 		return 0;
770 	}
771 	if ((err = vop_stdfdatasync_buf(ap)))
772 		return err;
773 
774 	return fuse_internal_fsync(vp, td, waitfor, true);
775 }
776 
777 /*
778     struct vnop_fsync_args {
779 	struct vop_generic_args a_gen;
780 	struct vnode * a_vp;
781 	int  a_waitfor;
782 	struct thread * a_td;
783     };
784 */
785 static int
786 fuse_vnop_fsync(struct vop_fsync_args *ap)
787 {
788 	struct vnode *vp = ap->a_vp;
789 	struct thread *td = ap->a_td;
790 	int waitfor = ap->a_waitfor;
791 	int err = 0;
792 
793 	if (fuse_isdeadfs(vp)) {
794 		return 0;
795 	}
796 	if ((err = vop_stdfsync(ap)))
797 		return err;
798 
799 	return fuse_internal_fsync(vp, td, waitfor, false);
800 }
801 
802 /*
803     struct vnop_getattr_args {
804 	struct vnode *a_vp;
805 	struct vattr *a_vap;
806 	struct ucred *a_cred;
807 	struct thread *a_td;
808     };
809 */
810 static int
811 fuse_vnop_getattr(struct vop_getattr_args *ap)
812 {
813 	struct vnode *vp = ap->a_vp;
814 	struct vattr *vap = ap->a_vap;
815 	struct ucred *cred = ap->a_cred;
816 	struct thread *td = curthread;
817 
818 	int err = 0;
819 	int dataflags;
820 
821 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
822 
823 	/* Note that we are not bailing out on a dead file system just yet. */
824 
825 	if (!(dataflags & FSESS_INITED)) {
826 		if (!vnode_isvroot(vp)) {
827 			fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
828 			err = ENOTCONN;
829 			return err;
830 		} else {
831 			goto fake;
832 		}
833 	}
834 	err = fuse_internal_getattr(vp, vap, cred, td);
835 	if (err == ENOTCONN && vnode_isvroot(vp)) {
836 		/* see comment in fuse_vfsop_statfs() */
837 		goto fake;
838 	} else {
839 		return err;
840 	}
841 
842 fake:
843 	bzero(vap, sizeof(*vap));
844 	vap->va_type = vnode_vtype(vp);
845 
846 	return 0;
847 }
848 
849 /*
850     struct vnop_inactive_args {
851 	struct vnode *a_vp;
852 	struct thread *a_td;
853     };
854 */
855 static int
856 fuse_vnop_inactive(struct vop_inactive_args *ap)
857 {
858 	struct vnode *vp = ap->a_vp;
859 	struct thread *td = ap->a_td;
860 
861 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
862 	struct fuse_filehandle *fufh, *fufh_tmp;
863 
864 	int need_flush = 1;
865 
866 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
867 		if (need_flush && vp->v_type == VREG) {
868 			if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
869 				fuse_vnode_savesize(vp, NULL, 0);
870 			}
871 			if ((fvdat->flag & FN_REVOKED) != 0)
872 				fuse_io_invalbuf(vp, td);
873 			else
874 				fuse_io_flushbuf(vp, MNT_WAIT, td);
875 			need_flush = 0;
876 		}
877 		fuse_filehandle_close(vp, fufh, td, NULL);
878 	}
879 
880 	if ((fvdat->flag & FN_REVOKED) != 0)
881 		vrecycle(vp);
882 
883 	return 0;
884 }
885 
886 /*
887     struct vnop_link_args {
888 	struct vnode *a_tdvp;
889 	struct vnode *a_vp;
890 	struct componentname *a_cnp;
891     };
892 */
893 static int
894 fuse_vnop_link(struct vop_link_args *ap)
895 {
896 	struct vnode *vp = ap->a_vp;
897 	struct vnode *tdvp = ap->a_tdvp;
898 	struct componentname *cnp = ap->a_cnp;
899 
900 	struct vattr *vap = VTOVA(vp);
901 
902 	struct fuse_dispatcher fdi;
903 	struct fuse_entry_out *feo;
904 	struct fuse_link_in fli;
905 
906 	int err;
907 
908 	if (fuse_isdeadfs(vp)) {
909 		return ENXIO;
910 	}
911 	if (vnode_mount(tdvp) != vnode_mount(vp)) {
912 		return EXDEV;
913 	}
914 
915 	/*
916 	 * This is a seatbelt check to protect naive userspace filesystems from
917 	 * themselves and the limitations of the FUSE IPC protocol.  If a
918 	 * filesystem does not allow attribute caching, assume it is capable of
919 	 * validating that nlink does not overflow.
920 	 */
921 	if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX)
922 		return EMLINK;
923 	fli.oldnodeid = VTOI(vp);
924 
925 	fdisp_init(&fdi, 0);
926 	fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
927 	    FUSE_LINK, &fli, sizeof(fli), &fdi);
928 	if ((err = fdisp_wait_answ(&fdi))) {
929 		goto out;
930 	}
931 	feo = fdi.answ;
932 
933 	err = fuse_internal_checkentry(feo, vnode_vtype(vp));
934 	if (!err) {
935 		/*
936 		 * Purge the parent's attribute cache because the daemon
937 		 * should've updated its mtime and ctime
938 		 */
939 		fuse_vnode_clear_attr_cache(tdvp);
940 		fuse_internal_cache_attrs(vp, &feo->attr, feo->attr_valid,
941 			feo->attr_valid_nsec, NULL);
942 	}
943 out:
944 	fdisp_destroy(&fdi);
945 	return err;
946 }
947 
948 struct fuse_lookup_alloc_arg {
949 	struct fuse_entry_out *feo;
950 	struct componentname *cnp;
951 	uint64_t nid;
952 	enum vtype vtyp;
953 };
954 
955 /* Callback for vn_get_ino */
956 static int
957 fuse_lookup_alloc(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
958 {
959 	struct fuse_lookup_alloc_arg *flaa = arg;
960 
961 	return fuse_vnode_get(mp, flaa->feo, flaa->nid, NULL, vpp, flaa->cnp,
962 		flaa->vtyp);
963 }
964 
965 SDT_PROBE_DEFINE3(fusefs, , vnops, cache_lookup,
966 	"int", "struct timespec*", "struct timespec*");
967 /*
968     struct vnop_lookup_args {
969 	struct vnodeop_desc *a_desc;
970 	struct vnode *a_dvp;
971 	struct vnode **a_vpp;
972 	struct componentname *a_cnp;
973     };
974 */
975 int
976 fuse_vnop_lookup(struct vop_lookup_args *ap)
977 {
978 	struct vnode *dvp = ap->a_dvp;
979 	struct vnode **vpp = ap->a_vpp;
980 	struct componentname *cnp = ap->a_cnp;
981 	struct thread *td = cnp->cn_thread;
982 	struct ucred *cred = cnp->cn_cred;
983 
984 	int nameiop = cnp->cn_nameiop;
985 	int flags = cnp->cn_flags;
986 	int wantparent = flags & (LOCKPARENT | WANTPARENT);
987 	int islastcn = flags & ISLASTCN;
988 	struct mount *mp = vnode_mount(dvp);
989 
990 	int err = 0;
991 	int lookup_err = 0;
992 	struct vnode *vp = NULL;
993 
994 	struct fuse_dispatcher fdi;
995 	bool did_lookup = false;
996 	struct fuse_entry_out *feo = NULL;
997 	enum vtype vtyp;	/* vnode type of target */
998 	off_t filesize;		/* filesize of target */
999 
1000 	uint64_t nid;
1001 
1002 	if (fuse_isdeadfs(dvp)) {
1003 		*vpp = NULL;
1004 		return ENXIO;
1005 	}
1006 	if (!vnode_isdir(dvp))
1007 		return ENOTDIR;
1008 
1009 	if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1010 		return EROFS;
1011 
1012 	if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1013 		return err;
1014 
1015 	if (flags & ISDOTDOT) {
1016 		KASSERT(VTOFUD(dvp)->flag & FN_PARENT_NID,
1017 			("Looking up .. is TODO"));
1018 		nid = VTOFUD(dvp)->parent_nid;
1019 		if (nid == 0)
1020 			return ENOENT;
1021 		/* .. is obviously a directory */
1022 		vtyp = VDIR;
1023 		filesize = 0;
1024 	} else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
1025 		nid = VTOI(dvp);
1026 		/* . is obviously a directory */
1027 		vtyp = VDIR;
1028 		filesize = 0;
1029 	} else {
1030 		struct timespec now, timeout;
1031 
1032 		err = cache_lookup(dvp, vpp, cnp, &timeout, NULL);
1033 		getnanouptime(&now);
1034 		SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1035 		switch (err) {
1036 		case -1:		/* positive match */
1037 			if (timespeccmp(&timeout, &now, >)) {
1038 				counter_u64_add(fuse_lookup_cache_hits, 1);
1039 			} else {
1040 				/* Cache timeout */
1041 				counter_u64_add(fuse_lookup_cache_misses, 1);
1042 				bintime_clear(
1043 					&VTOFUD(*vpp)->entry_cache_timeout);
1044 				cache_purge(*vpp);
1045 				if (dvp != *vpp)
1046 					vput(*vpp);
1047 				else
1048 					vrele(*vpp);
1049 				*vpp = NULL;
1050 				break;
1051 			}
1052 			return 0;
1053 
1054 		case 0:		/* no match in cache */
1055 			counter_u64_add(fuse_lookup_cache_misses, 1);
1056 			break;
1057 
1058 		case ENOENT:		/* negative match */
1059 			getnanouptime(&now);
1060 			if (timespeccmp(&timeout, &now, <=)) {
1061 				/* Cache timeout */
1062 				cache_purge_negative(dvp);
1063 				break;
1064 			}
1065 			/* fall through */
1066 		default:
1067 			return err;
1068 		}
1069 
1070 		nid = VTOI(dvp);
1071 		fdisp_init(&fdi, cnp->cn_namelen + 1);
1072 		fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred);
1073 
1074 		memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1075 		((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1076 		lookup_err = fdisp_wait_answ(&fdi);
1077 		did_lookup = true;
1078 
1079 		if (!lookup_err) {
1080 			/* lookup call succeeded */
1081 			feo = (struct fuse_entry_out *)fdi.answ;
1082 			nid = feo->nodeid;
1083 			if (nid == 0) {
1084 				/* zero nodeid means ENOENT and cache it */
1085 				struct timespec timeout;
1086 
1087 				fdi.answ_stat = ENOENT;
1088 				lookup_err = ENOENT;
1089 				if (cnp->cn_flags & MAKEENTRY) {
1090 					fuse_validity_2_timespec(feo, &timeout);
1091 					cache_enter_time(dvp, *vpp, cnp,
1092 						&timeout, NULL);
1093 				}
1094 			} else if (nid == FUSE_ROOT_ID) {
1095 				lookup_err = EINVAL;
1096 			}
1097 			vtyp = IFTOVT(feo->attr.mode);
1098 			filesize = feo->attr.size;
1099 		}
1100 		if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1101 			fdisp_destroy(&fdi);
1102 			return lookup_err;
1103 		}
1104 	}
1105 	/* lookup_err, if non-zero, must be ENOENT at this point */
1106 
1107 	if (lookup_err) {
1108 		/* Entry not found */
1109 		if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1110 			err = fuse_internal_access(dvp, VWRITE, td, cred);
1111 			if (!err) {
1112 				/*
1113 				 * Set the SAVENAME flag to hold onto the
1114 				 * pathname for use later in VOP_CREATE or
1115 				 * VOP_RENAME.
1116 				 */
1117 				cnp->cn_flags |= SAVENAME;
1118 
1119 				err = EJUSTRETURN;
1120 			}
1121 		} else {
1122 			err = ENOENT;
1123 		}
1124 	} else {
1125 		/* Entry was found */
1126 		if (flags & ISDOTDOT) {
1127 			struct fuse_lookup_alloc_arg flaa;
1128 
1129 			flaa.nid = nid;
1130 			flaa.feo = feo;
1131 			flaa.cnp = cnp;
1132 			flaa.vtyp = vtyp;
1133 			err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1134 				&vp);
1135 			*vpp = vp;
1136 		} else if (nid == VTOI(dvp)) {
1137 			vref(dvp);
1138 			*vpp = dvp;
1139 		} else {
1140 			struct fuse_vnode_data *fvdat;
1141 
1142 			err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1143 			    &vp, cnp, vtyp);
1144 			if (err)
1145 				goto out;
1146 			*vpp = vp;
1147 
1148 			/*
1149 			 * In the case where we are looking up a FUSE node
1150 			 * represented by an existing cached vnode, and the
1151 			 * true size reported by FUSE_LOOKUP doesn't match
1152 			 * the vnode's cached size, then any cached writes
1153 			 * beyond the file's current size are lost.
1154 			 *
1155 			 * We can get here:
1156 			 * * following attribute cache expiration, or
1157 			 * * due a bug in the daemon, or
1158 			 */
1159 			fvdat = VTOFUD(vp);
1160 			if (vnode_isreg(vp) &&
1161 			    filesize != fvdat->cached_attrs.va_size &&
1162 			    fvdat->flag & FN_SIZECHANGE) {
1163 				/*
1164 				 * The FN_SIZECHANGE flag reflects a dirty
1165 				 * append.  If userspace lets us know our cache
1166 				 * is invalid, that write was lost.  (Dirty
1167 				 * writes that do not cause append are also
1168 				 * lost, but we don't detect them here.)
1169 				 *
1170 				 * XXX: Maybe disable WB caching on this mount.
1171 				 */
1172 				printf("%s: WB cache incoherent on %s!\n",
1173 				    __func__,
1174 				    vnode_mount(vp)->mnt_stat.f_mntonname);
1175 
1176 				fvdat->flag &= ~FN_SIZECHANGE;
1177 			}
1178 
1179 			MPASS(feo != NULL);
1180 			fuse_internal_cache_attrs(*vpp, &feo->attr,
1181 				feo->attr_valid, feo->attr_valid_nsec, NULL);
1182 			fuse_validity_2_bintime(feo->entry_valid,
1183 				feo->entry_valid_nsec,
1184 				&fvdat->entry_cache_timeout);
1185 
1186 			if ((nameiop == DELETE || nameiop == RENAME) &&
1187 				islastcn)
1188 			{
1189 				struct vattr dvattr;
1190 
1191 				err = fuse_internal_access(dvp, VWRITE, td,
1192 					cred);
1193 				if (err != 0)
1194 					goto out;
1195 				/*
1196 				 * if the parent's sticky bit is set, check
1197 				 * whether we're allowed to remove the file.
1198 				 * Need to figure out the vnode locking to make
1199 				 * this work.
1200 				 */
1201 				fuse_internal_getattr(dvp, &dvattr, cred, td);
1202 				if ((dvattr.va_mode & S_ISTXT) &&
1203 					fuse_internal_access(dvp, VADMIN, td,
1204 						cred) &&
1205 					fuse_internal_access(*vpp, VADMIN, td,
1206 						cred)) {
1207 					err = EPERM;
1208 					goto out;
1209 				}
1210 			}
1211 
1212 			if (islastcn && (
1213 				(nameiop == DELETE) ||
1214 				(nameiop == RENAME && wantparent))) {
1215 				cnp->cn_flags |= SAVENAME;
1216 			}
1217 
1218 		}
1219 	}
1220 out:
1221 	if (err) {
1222 		if (vp != NULL && dvp != vp)
1223 			vput(vp);
1224 		else if (vp != NULL)
1225 			vrele(vp);
1226 		*vpp = NULL;
1227 	}
1228 	if (did_lookup)
1229 		fdisp_destroy(&fdi);
1230 
1231 	return err;
1232 }
1233 
1234 /*
1235     struct vnop_mkdir_args {
1236 	struct vnode *a_dvp;
1237 	struct vnode **a_vpp;
1238 	struct componentname *a_cnp;
1239 	struct vattr *a_vap;
1240     };
1241 */
1242 static int
1243 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1244 {
1245 	struct vnode *dvp = ap->a_dvp;
1246 	struct vnode **vpp = ap->a_vpp;
1247 	struct componentname *cnp = ap->a_cnp;
1248 	struct vattr *vap = ap->a_vap;
1249 
1250 	struct fuse_mkdir_in fmdi;
1251 
1252 	if (fuse_isdeadfs(dvp)) {
1253 		return ENXIO;
1254 	}
1255 	fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1256 	fmdi.umask = curthread->td_proc->p_fd->fd_cmask;
1257 
1258 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1259 	    sizeof(fmdi), VDIR));
1260 }
1261 
1262 /*
1263     struct vnop_mknod_args {
1264 	struct vnode *a_dvp;
1265 	struct vnode **a_vpp;
1266 	struct componentname *a_cnp;
1267 	struct vattr *a_vap;
1268     };
1269 */
1270 static int
1271 fuse_vnop_mknod(struct vop_mknod_args *ap)
1272 {
1273 
1274 	struct vnode *dvp = ap->a_dvp;
1275 	struct vnode **vpp = ap->a_vpp;
1276 	struct componentname *cnp = ap->a_cnp;
1277 	struct vattr *vap = ap->a_vap;
1278 
1279 	if (fuse_isdeadfs(dvp))
1280 		return ENXIO;
1281 
1282 	return fuse_internal_mknod(dvp, vpp, cnp, vap);
1283 }
1284 
1285 /*
1286     struct vop_open_args {
1287 	struct vnode *a_vp;
1288 	int  a_mode;
1289 	struct ucred *a_cred;
1290 	struct thread *a_td;
1291 	int a_fdidx; / struct file *a_fp;
1292     };
1293 */
1294 static int
1295 fuse_vnop_open(struct vop_open_args *ap)
1296 {
1297 	struct vnode *vp = ap->a_vp;
1298 	int a_mode = ap->a_mode;
1299 	struct thread *td = ap->a_td;
1300 	struct ucred *cred = ap->a_cred;
1301 	pid_t pid = td->td_proc->p_pid;
1302 	struct fuse_vnode_data *fvdat;
1303 
1304 	if (fuse_isdeadfs(vp))
1305 		return ENXIO;
1306 	if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1307 		return (EOPNOTSUPP);
1308 	if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1309 		return EINVAL;
1310 
1311 	fvdat = VTOFUD(vp);
1312 
1313 	if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1314 		fuse_vnode_open(vp, 0, td);
1315 		return 0;
1316 	}
1317 
1318 	return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1319 }
1320 
1321 static int
1322 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1323 {
1324 
1325 	switch (ap->a_name) {
1326 	case _PC_FILESIZEBITS:
1327 		*ap->a_retval = 64;
1328 		return (0);
1329 	case _PC_NAME_MAX:
1330 		*ap->a_retval = NAME_MAX;
1331 		return (0);
1332 	case _PC_LINK_MAX:
1333 		*ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1334 		return (0);
1335 	case _PC_SYMLINK_MAX:
1336 		*ap->a_retval = MAXPATHLEN;
1337 		return (0);
1338 	case _PC_NO_TRUNC:
1339 		*ap->a_retval = 1;
1340 		return (0);
1341 	default:
1342 		return (vop_stdpathconf(ap));
1343 	}
1344 }
1345 
1346 /*
1347     struct vnop_read_args {
1348 	struct vnode *a_vp;
1349 	struct uio *a_uio;
1350 	int  a_ioflag;
1351 	struct ucred *a_cred;
1352     };
1353 */
1354 static int
1355 fuse_vnop_read(struct vop_read_args *ap)
1356 {
1357 	struct vnode *vp = ap->a_vp;
1358 	struct uio *uio = ap->a_uio;
1359 	int ioflag = ap->a_ioflag;
1360 	struct ucred *cred = ap->a_cred;
1361 	pid_t pid = curthread->td_proc->p_pid;
1362 
1363 	if (fuse_isdeadfs(vp)) {
1364 		return ENXIO;
1365 	}
1366 
1367 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1368 		ioflag |= IO_DIRECT;
1369 	}
1370 
1371 	return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1372 }
1373 
1374 /*
1375     struct vnop_readdir_args {
1376 	struct vnode *a_vp;
1377 	struct uio *a_uio;
1378 	struct ucred *a_cred;
1379 	int *a_eofflag;
1380 	int *a_ncookies;
1381 	u_long **a_cookies;
1382     };
1383 */
1384 static int
1385 fuse_vnop_readdir(struct vop_readdir_args *ap)
1386 {
1387 	struct vnode *vp = ap->a_vp;
1388 	struct uio *uio = ap->a_uio;
1389 	struct ucred *cred = ap->a_cred;
1390 	struct fuse_filehandle *fufh = NULL;
1391 	struct fuse_iov cookediov;
1392 	int err = 0;
1393 	u_long *cookies;
1394 	off_t startoff;
1395 	ssize_t tresid;
1396 	int ncookies;
1397 	bool closefufh = false;
1398 	pid_t pid = curthread->td_proc->p_pid;
1399 
1400 	if (ap->a_eofflag)
1401 		*ap->a_eofflag = 0;
1402 	if (fuse_isdeadfs(vp)) {
1403 		return ENXIO;
1404 	}
1405 	if (				/* XXXIP ((uio_iovcnt(uio) > 1)) || */
1406 	    (uio_resid(uio) < sizeof(struct dirent))) {
1407 		return EINVAL;
1408 	}
1409 
1410 	tresid = uio->uio_resid;
1411 	startoff = uio->uio_offset;
1412 	err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1413 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1414 		/*
1415 		 * nfsd will do VOP_READDIR without first doing VOP_OPEN.  We
1416 		 * must implicitly open the directory here
1417 		 */
1418 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1419 		if (err == 0) {
1420 			/*
1421 			 * When a directory is opened, it must be read from
1422 			 * the beginning.  Hopefully, the "startoff" still
1423 			 * exists as an offset cookie for the directory.
1424 			 * If not, it will read the entire directory without
1425 			 * returning any entries and just return eof.
1426 			 */
1427 			uio->uio_offset = 0;
1428 		}
1429 		closefufh = true;
1430 	}
1431 	if (err)
1432 		return (err);
1433 	if (ap->a_ncookies != NULL) {
1434 		ncookies = uio->uio_resid /
1435 			(offsetof(struct dirent, d_name) + 4) + 1;
1436 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1437 		*ap->a_ncookies = ncookies;
1438 		*ap->a_cookies = cookies;
1439 	} else {
1440 		ncookies = 0;
1441 		cookies = NULL;
1442 	}
1443 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1444 	fiov_init(&cookediov, DIRCOOKEDSIZE);
1445 
1446 	err = fuse_internal_readdir(vp, uio, startoff, fufh, &cookediov,
1447 		&ncookies, cookies);
1448 
1449 	fiov_teardown(&cookediov);
1450 	if (closefufh)
1451 		fuse_filehandle_close(vp, fufh, curthread, cred);
1452 
1453 	if (ap->a_ncookies != NULL) {
1454 		if (err == 0) {
1455 			*ap->a_ncookies -= ncookies;
1456 		} else {
1457 			free(*ap->a_cookies, M_TEMP);
1458 			*ap->a_ncookies = 0;
1459 			*ap->a_cookies = NULL;
1460 		}
1461 	}
1462 	if (err == 0 && tresid == uio->uio_resid)
1463 		*ap->a_eofflag = 1;
1464 
1465 	return err;
1466 }
1467 
1468 /*
1469     struct vnop_readlink_args {
1470 	struct vnode *a_vp;
1471 	struct uio *a_uio;
1472 	struct ucred *a_cred;
1473     };
1474 */
1475 static int
1476 fuse_vnop_readlink(struct vop_readlink_args *ap)
1477 {
1478 	struct vnode *vp = ap->a_vp;
1479 	struct uio *uio = ap->a_uio;
1480 	struct ucred *cred = ap->a_cred;
1481 
1482 	struct fuse_dispatcher fdi;
1483 	int err;
1484 
1485 	if (fuse_isdeadfs(vp)) {
1486 		return ENXIO;
1487 	}
1488 	if (!vnode_islnk(vp)) {
1489 		return EINVAL;
1490 	}
1491 	fdisp_init(&fdi, 0);
1492 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1493 	if (err) {
1494 		goto out;
1495 	}
1496 	if (((char *)fdi.answ)[0] == '/' &&
1497 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1498 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1499 
1500 		err = uiomove(mpth, strlen(mpth), uio);
1501 	}
1502 	if (!err) {
1503 		err = uiomove(fdi.answ, fdi.iosize, uio);
1504 	}
1505 out:
1506 	fdisp_destroy(&fdi);
1507 	return err;
1508 }
1509 
1510 /*
1511     struct vnop_reclaim_args {
1512 	struct vnode *a_vp;
1513 	struct thread *a_td;
1514     };
1515 */
1516 static int
1517 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1518 {
1519 	struct vnode *vp = ap->a_vp;
1520 	struct thread *td = ap->a_td;
1521 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1522 	struct fuse_filehandle *fufh, *fufh_tmp;
1523 
1524 	if (!fvdat) {
1525 		panic("FUSE: no vnode data during recycling");
1526 	}
1527 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1528 		printf("FUSE: vnode being reclaimed with open fufh "
1529 			"(type=%#x)", fufh->fufh_type);
1530 		fuse_filehandle_close(vp, fufh, td, NULL);
1531 	}
1532 
1533 	if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) {
1534 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1535 		    fvdat->nlookup);
1536 	}
1537 	fuse_vnode_setparent(vp, NULL);
1538 	cache_purge(vp);
1539 	vfs_hash_remove(vp);
1540 	vnode_destroy_vobject(vp);
1541 	fuse_vnode_destroy(vp);
1542 
1543 	return 0;
1544 }
1545 
1546 /*
1547     struct vnop_remove_args {
1548 	struct vnode *a_dvp;
1549 	struct vnode *a_vp;
1550 	struct componentname *a_cnp;
1551     };
1552 */
1553 static int
1554 fuse_vnop_remove(struct vop_remove_args *ap)
1555 {
1556 	struct vnode *dvp = ap->a_dvp;
1557 	struct vnode *vp = ap->a_vp;
1558 	struct componentname *cnp = ap->a_cnp;
1559 
1560 	int err;
1561 
1562 	if (fuse_isdeadfs(vp)) {
1563 		return ENXIO;
1564 	}
1565 	if (vnode_isdir(vp)) {
1566 		return EPERM;
1567 	}
1568 
1569 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1570 
1571 	return err;
1572 }
1573 
1574 /*
1575     struct vnop_rename_args {
1576 	struct vnode *a_fdvp;
1577 	struct vnode *a_fvp;
1578 	struct componentname *a_fcnp;
1579 	struct vnode *a_tdvp;
1580 	struct vnode *a_tvp;
1581 	struct componentname *a_tcnp;
1582     };
1583 */
1584 static int
1585 fuse_vnop_rename(struct vop_rename_args *ap)
1586 {
1587 	struct vnode *fdvp = ap->a_fdvp;
1588 	struct vnode *fvp = ap->a_fvp;
1589 	struct componentname *fcnp = ap->a_fcnp;
1590 	struct vnode *tdvp = ap->a_tdvp;
1591 	struct vnode *tvp = ap->a_tvp;
1592 	struct componentname *tcnp = ap->a_tcnp;
1593 	struct fuse_data *data;
1594 	bool newparent = fdvp != tdvp;
1595 	bool isdir = fvp->v_type == VDIR;
1596 	int err = 0;
1597 
1598 	if (fuse_isdeadfs(fdvp)) {
1599 		return ENXIO;
1600 	}
1601 	if (fvp->v_mount != tdvp->v_mount ||
1602 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1603 		SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
1604 		err = EXDEV;
1605 		goto out;
1606 	}
1607 	cache_purge(fvp);
1608 
1609 	/*
1610 	 * FUSE library is expected to check if target directory is not
1611 	 * under the source directory in the file system tree.
1612 	 * Linux performs this check at VFS level.
1613 	 */
1614 	/*
1615 	 * If source is a directory, and it will get a new parent, user must
1616 	 * have write permission to it, so ".." can be modified.
1617 	 */
1618 	data = fuse_get_mpdata(vnode_mount(tdvp));
1619 	if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
1620 		err = fuse_internal_access(fvp, VWRITE,
1621 			tcnp->cn_thread, tcnp->cn_cred);
1622 		if (err)
1623 			goto out;
1624 	}
1625 	sx_xlock(&data->rename_lock);
1626 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1627 	if (err == 0) {
1628 		if (tdvp != fdvp)
1629 			fuse_vnode_setparent(fvp, tdvp);
1630 		if (tvp != NULL)
1631 			fuse_vnode_setparent(tvp, NULL);
1632 	}
1633 	sx_unlock(&data->rename_lock);
1634 
1635 	if (tvp != NULL && tvp != fvp) {
1636 		cache_purge(tvp);
1637 	}
1638 	if (vnode_isdir(fvp)) {
1639 		if ((tvp != NULL) && vnode_isdir(tvp)) {
1640 			cache_purge(tdvp);
1641 		}
1642 		cache_purge(fdvp);
1643 	}
1644 out:
1645 	if (tdvp == tvp) {
1646 		vrele(tdvp);
1647 	} else {
1648 		vput(tdvp);
1649 	}
1650 	if (tvp != NULL) {
1651 		vput(tvp);
1652 	}
1653 	vrele(fdvp);
1654 	vrele(fvp);
1655 
1656 	return err;
1657 }
1658 
1659 /*
1660     struct vnop_rmdir_args {
1661 	    struct vnode *a_dvp;
1662 	    struct vnode *a_vp;
1663 	    struct componentname *a_cnp;
1664     } *ap;
1665 */
1666 static int
1667 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1668 {
1669 	struct vnode *dvp = ap->a_dvp;
1670 	struct vnode *vp = ap->a_vp;
1671 
1672 	int err;
1673 
1674 	if (fuse_isdeadfs(vp)) {
1675 		return ENXIO;
1676 	}
1677 	if (VTOFUD(vp) == VTOFUD(dvp)) {
1678 		return EINVAL;
1679 	}
1680 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1681 
1682 	return err;
1683 }
1684 
1685 /*
1686     struct vnop_setattr_args {
1687 	struct vnode *a_vp;
1688 	struct vattr *a_vap;
1689 	struct ucred *a_cred;
1690 	struct thread *a_td;
1691     };
1692 */
1693 static int
1694 fuse_vnop_setattr(struct vop_setattr_args *ap)
1695 {
1696 	struct vnode *vp = ap->a_vp;
1697 	struct vattr *vap = ap->a_vap;
1698 	struct ucred *cred = ap->a_cred;
1699 	struct thread *td = curthread;
1700 	struct mount *mp;
1701 	struct fuse_data *data;
1702 	struct vattr old_va;
1703 	int dataflags;
1704 	int err = 0, err2;
1705 	accmode_t accmode = 0;
1706 	bool checkperm;
1707 	bool drop_suid = false;
1708 	gid_t cr_gid;
1709 
1710 	mp = vnode_mount(vp);
1711 	data = fuse_get_mpdata(mp);
1712 	dataflags = data->dataflags;
1713 	checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
1714 	if (cred->cr_ngroups > 0)
1715 		cr_gid = cred->cr_groups[0];
1716 	else
1717 		cr_gid = 0;
1718 
1719 	if (fuse_isdeadfs(vp)) {
1720 		return ENXIO;
1721 	}
1722 
1723 	if (vap->va_uid != (uid_t)VNOVAL) {
1724 		if (checkperm) {
1725 			/* Only root may change a file's owner */
1726 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1727 			if (err) {
1728 				/* As a special case, allow the null chown */
1729 				err2 = fuse_internal_getattr(vp, &old_va, cred,
1730 					td);
1731 				if (err2)
1732 					return (err2);
1733 				if (vap->va_uid != old_va.va_uid)
1734 					return err;
1735 				else
1736 					accmode |= VADMIN;
1737 				drop_suid = true;
1738 			} else
1739 				accmode |= VADMIN;
1740 		} else
1741 			accmode |= VADMIN;
1742 	}
1743 	if (vap->va_gid != (gid_t)VNOVAL) {
1744 		if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
1745 			drop_suid = true;
1746 		if (checkperm && !groupmember(vap->va_gid, cred))
1747 		{
1748 			/*
1749 			 * Non-root users may only chgrp to one of their own
1750 			 * groups
1751 			 */
1752 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1753 			if (err) {
1754 				/* As a special case, allow the null chgrp */
1755 				err2 = fuse_internal_getattr(vp, &old_va, cred,
1756 					td);
1757 				if (err2)
1758 					return (err2);
1759 				if (vap->va_gid != old_va.va_gid)
1760 					return err;
1761 				accmode |= VADMIN;
1762 			} else
1763 				accmode |= VADMIN;
1764 		} else
1765 			accmode |= VADMIN;
1766 	}
1767 	if (vap->va_size != VNOVAL) {
1768 		switch (vp->v_type) {
1769 		case VDIR:
1770 			return (EISDIR);
1771 		case VLNK:
1772 		case VREG:
1773 			if (vfs_isrdonly(mp))
1774 				return (EROFS);
1775 			break;
1776 		default:
1777 			/*
1778 			 * According to POSIX, the result is unspecified
1779 			 * for file types other than regular files,
1780 			 * directories and shared memory objects.  We
1781 			 * don't support shared memory objects in the file
1782 			 * system, and have dubious support for truncating
1783 			 * symlinks.  Just ignore the request in other cases.
1784 			 */
1785 			return (0);
1786 		}
1787 		/* Don't set accmode.  Permission to trunc is checked upstack */
1788 	}
1789 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1790 		if (vap->va_vaflags & VA_UTIMES_NULL)
1791 			accmode |= VWRITE;
1792 		else
1793 			accmode |= VADMIN;
1794 	}
1795 	if (drop_suid) {
1796 		if (vap->va_mode != (mode_t)VNOVAL)
1797 			vap->va_mode &= ~(S_ISUID | S_ISGID);
1798 		else {
1799 			err = fuse_internal_getattr(vp, &old_va, cred, td);
1800 			if (err)
1801 				return (err);
1802 			vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
1803 		}
1804 	}
1805 	if (vap->va_mode != (mode_t)VNOVAL) {
1806 		/* Only root may set the sticky bit on non-directories */
1807 		if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
1808 		    && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1809 			return EFTYPE;
1810 		if (checkperm && (vap->va_mode & S_ISGID)) {
1811 			err = fuse_internal_getattr(vp, &old_va, cred, td);
1812 			if (err)
1813 				return (err);
1814 			if (!groupmember(old_va.va_gid, cred)) {
1815 				err = priv_check_cred(cred, PRIV_VFS_SETGID);
1816 				if (err)
1817 					return (err);
1818 			}
1819 		}
1820 		accmode |= VADMIN;
1821 	}
1822 
1823 	if (vfs_isrdonly(mp))
1824 		return EROFS;
1825 
1826 	err = fuse_internal_access(vp, accmode, td, cred);
1827 	if (err)
1828 		return err;
1829 	else
1830 		return fuse_internal_setattr(vp, vap, td, cred);
1831 }
1832 
1833 /*
1834     struct vnop_strategy_args {
1835 	struct vnode *a_vp;
1836 	struct buf *a_bp;
1837     };
1838 */
1839 static int
1840 fuse_vnop_strategy(struct vop_strategy_args *ap)
1841 {
1842 	struct vnode *vp = ap->a_vp;
1843 	struct buf *bp = ap->a_bp;
1844 
1845 	if (!vp || fuse_isdeadfs(vp)) {
1846 		bp->b_ioflags |= BIO_ERROR;
1847 		bp->b_error = ENXIO;
1848 		bufdone(bp);
1849 		return 0;
1850 	}
1851 
1852 	/*
1853 	 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
1854 	 * fuse_io_strategy sets bp's error fields
1855 	 */
1856 	(void)fuse_io_strategy(vp, bp);
1857 
1858 	return 0;
1859 }
1860 
1861 
1862 /*
1863     struct vnop_symlink_args {
1864 	struct vnode *a_dvp;
1865 	struct vnode **a_vpp;
1866 	struct componentname *a_cnp;
1867 	struct vattr *a_vap;
1868 	char *a_target;
1869     };
1870 */
1871 static int
1872 fuse_vnop_symlink(struct vop_symlink_args *ap)
1873 {
1874 	struct vnode *dvp = ap->a_dvp;
1875 	struct vnode **vpp = ap->a_vpp;
1876 	struct componentname *cnp = ap->a_cnp;
1877 	const char *target = ap->a_target;
1878 
1879 	struct fuse_dispatcher fdi;
1880 
1881 	int err;
1882 	size_t len;
1883 
1884 	if (fuse_isdeadfs(dvp)) {
1885 		return ENXIO;
1886 	}
1887 	/*
1888 	 * Unlike the other creator type calls, here we have to create a message
1889 	 * where the name of the new entry comes first, and the data describing
1890 	 * the entry comes second.
1891 	 * Hence we can't rely on our handy fuse_internal_newentry() routine,
1892 	 * but put together the message manually and just call the core part.
1893 	 */
1894 
1895 	len = strlen(target) + 1;
1896 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1897 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1898 
1899 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1900 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1901 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1902 
1903 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1904 	fdisp_destroy(&fdi);
1905 	return err;
1906 }
1907 
1908 /*
1909     struct vnop_write_args {
1910 	struct vnode *a_vp;
1911 	struct uio *a_uio;
1912 	int  a_ioflag;
1913 	struct ucred *a_cred;
1914     };
1915 */
1916 static int
1917 fuse_vnop_write(struct vop_write_args *ap)
1918 {
1919 	struct vnode *vp = ap->a_vp;
1920 	struct uio *uio = ap->a_uio;
1921 	int ioflag = ap->a_ioflag;
1922 	struct ucred *cred = ap->a_cred;
1923 	pid_t pid = curthread->td_proc->p_pid;
1924 
1925 	if (fuse_isdeadfs(vp)) {
1926 		return ENXIO;
1927 	}
1928 
1929 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1930 		ioflag |= IO_DIRECT;
1931 	}
1932 
1933 	return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1934 }
1935 
1936 static daddr_t
1937 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1938 {
1939 	const int biosize = fuse_iosize(vp);
1940 
1941 	return (off / biosize);
1942 }
1943 
1944 static int
1945 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1946 {
1947 	off_t filesize;
1948 	int blksz, err;
1949 	const int biosize = fuse_iosize(vp);
1950 
1951 	err = fuse_vnode_size(vp, &filesize, NULL, NULL);
1952 	KASSERT(err == 0, ("vfs_bio_getpages can't handle errors here"));
1953 	if (err)
1954 		return biosize;
1955 
1956 	if ((off_t)lbn * biosize >= filesize) {
1957 		blksz = 0;
1958 	} else if ((off_t)(lbn + 1) * biosize > filesize) {
1959 		blksz = filesize - (off_t)lbn *biosize;
1960 	} else {
1961 		blksz = biosize;
1962 	}
1963 	return (blksz);
1964 }
1965 
1966 /*
1967     struct vnop_getpages_args {
1968 	struct vnode *a_vp;
1969 	vm_page_t *a_m;
1970 	int a_count;
1971 	int a_reqpage;
1972     };
1973 */
1974 static int
1975 fuse_vnop_getpages(struct vop_getpages_args *ap)
1976 {
1977 	struct vnode *vp = ap->a_vp;
1978 
1979 	if (!fsess_opt_mmap(vnode_mount(vp))) {
1980 		SDT_PROBE2(fusefs, , vnops, trace, 1,
1981 			"called on non-cacheable vnode??\n");
1982 		return (VM_PAGER_ERROR);
1983 	}
1984 
1985 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1986 	    ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
1987 }
1988 
1989 static const char extattr_namespace_separator = '.';
1990 
1991 /*
1992     struct vop_getextattr_args {
1993 	struct vop_generic_args a_gen;
1994 	struct vnode *a_vp;
1995 	int a_attrnamespace;
1996 	const char *a_name;
1997 	struct uio *a_uio;
1998 	size_t *a_size;
1999 	struct ucred *a_cred;
2000 	struct thread *a_td;
2001     };
2002 */
2003 static int
2004 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2005 {
2006 	struct vnode *vp = ap->a_vp;
2007 	struct uio *uio = ap->a_uio;
2008 	struct fuse_dispatcher fdi;
2009 	struct fuse_getxattr_in *get_xattr_in;
2010 	struct fuse_getxattr_out *get_xattr_out;
2011 	struct mount *mp = vnode_mount(vp);
2012 	struct thread *td = ap->a_td;
2013 	struct ucred *cred = ap->a_cred;
2014 	char *prefix;
2015 	char *attr_str;
2016 	size_t len;
2017 	int err;
2018 
2019 	if (fuse_isdeadfs(vp))
2020 		return (ENXIO);
2021 
2022 	if (!fsess_isimpl(mp, FUSE_GETXATTR))
2023 		return EOPNOTSUPP;
2024 
2025 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2026 	if (err)
2027 		return err;
2028 
2029 	/* Default to looking for user attributes. */
2030 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2031 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2032 	else
2033 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2034 
2035 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2036 	    strlen(ap->a_name) + 1;
2037 
2038 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2039 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2040 
2041 	get_xattr_in = fdi.indata;
2042 	/*
2043 	 * Check to see whether we're querying the available size or
2044 	 * issuing the actual request.  If we pass in 0, we get back struct
2045 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2046 	 * that much data, without the struct fuse_getxattr_out header.
2047 	 */
2048 	if (uio == NULL)
2049 		get_xattr_in->size = 0;
2050 	else
2051 		get_xattr_in->size = uio->uio_resid;
2052 
2053 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2054 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2055 	    ap->a_name);
2056 
2057 	err = fdisp_wait_answ(&fdi);
2058 	if (err != 0) {
2059 		if (err == ENOSYS) {
2060 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2061 			err = EOPNOTSUPP;
2062 		}
2063 		goto out;
2064 	}
2065 
2066 	get_xattr_out = fdi.answ;
2067 
2068 	if (ap->a_size != NULL)
2069 		*ap->a_size = get_xattr_out->size;
2070 
2071 	if (uio != NULL)
2072 		err = uiomove(fdi.answ, fdi.iosize, uio);
2073 
2074 out:
2075 	fdisp_destroy(&fdi);
2076 	return (err);
2077 }
2078 
2079 /*
2080     struct vop_setextattr_args {
2081 	struct vop_generic_args a_gen;
2082 	struct vnode *a_vp;
2083 	int a_attrnamespace;
2084 	const char *a_name;
2085 	struct uio *a_uio;
2086 	struct ucred *a_cred;
2087 	struct thread *a_td;
2088     };
2089 */
2090 static int
2091 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2092 {
2093 	struct vnode *vp = ap->a_vp;
2094 	struct uio *uio = ap->a_uio;
2095 	struct fuse_dispatcher fdi;
2096 	struct fuse_setxattr_in *set_xattr_in;
2097 	struct mount *mp = vnode_mount(vp);
2098 	struct thread *td = ap->a_td;
2099 	struct ucred *cred = ap->a_cred;
2100 	char *prefix;
2101 	size_t len;
2102 	char *attr_str;
2103 	int err;
2104 
2105 	if (fuse_isdeadfs(vp))
2106 		return (ENXIO);
2107 
2108 	if (!fsess_isimpl(mp, FUSE_SETXATTR))
2109 		return EOPNOTSUPP;
2110 
2111 	if (vfs_isrdonly(mp))
2112 		return EROFS;
2113 
2114 	/* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2115 	if (ap->a_uio == NULL) {
2116 		/*
2117 		 * If we got here as fallback from VOP_DELETEEXTATTR, then
2118 		 * return EOPNOTSUPP.
2119 		 */
2120 		if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2121 			return (EOPNOTSUPP);
2122 		else
2123 			return (EINVAL);
2124 	}
2125 
2126 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2127 		VWRITE);
2128 	if (err)
2129 		return err;
2130 
2131 	/* Default to looking for user attributes. */
2132 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2133 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2134 	else
2135 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2136 
2137 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2138 	    strlen(ap->a_name) + 1;
2139 
2140 	fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2141 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2142 
2143 	set_xattr_in = fdi.indata;
2144 	set_xattr_in->size = uio->uio_resid;
2145 
2146 	attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2147 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2148 	    ap->a_name);
2149 
2150 	err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2151 	    uio->uio_resid, uio);
2152 	if (err != 0) {
2153 		goto out;
2154 	}
2155 
2156 	err = fdisp_wait_answ(&fdi);
2157 
2158 	if (err == ENOSYS) {
2159 		fsess_set_notimpl(mp, FUSE_SETXATTR);
2160 		err = EOPNOTSUPP;
2161 	}
2162 	if (err == ERESTART) {
2163 		/* Can't restart after calling uiomove */
2164 		err = EINTR;
2165 	}
2166 
2167 out:
2168 	fdisp_destroy(&fdi);
2169 	return (err);
2170 }
2171 
2172 /*
2173  * The Linux / FUSE extended attribute list is simply a collection of
2174  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2175  * byte length followed by a non-NUL terminated string.  So, this allows
2176  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2177  * Linux attribute names are reported with the namespace as a prefix (e.g.
2178  * "user.attribute_name"), but in FreeBSD they are reported without the
2179  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2180  *
2181  * user.attr_name1\0user.attr_name2\0
2182  *
2183  * to:
2184  *
2185  * <num>attr_name1<num>attr_name2
2186  *
2187  * Where "<num>" is a single byte number of characters in the attribute name.
2188  *
2189  * Args:
2190  * prefix - exattr namespace prefix string
2191  * list, list_len - input list with namespace prefixes
2192  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2193  */
2194 static int
2195 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2196     char *bsd_list, int *bsd_list_len)
2197 {
2198 	int len, pos, dist_to_next, prefix_len;
2199 
2200 	pos = 0;
2201 	*bsd_list_len = 0;
2202 	prefix_len = strlen(prefix);
2203 
2204 	while (pos < list_len && list[pos] != '\0') {
2205 		dist_to_next = strlen(&list[pos]) + 1;
2206 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2207 		    list[pos + prefix_len] == extattr_namespace_separator) {
2208 			len = dist_to_next -
2209 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2210 			if (len >= EXTATTR_MAXNAMELEN)
2211 				return (ENAMETOOLONG);
2212 
2213 			bsd_list[*bsd_list_len] = len;
2214 			memcpy(&bsd_list[*bsd_list_len + 1],
2215 			    &list[pos + prefix_len +
2216 			    sizeof(extattr_namespace_separator)], len);
2217 
2218 			*bsd_list_len += len + 1;
2219 		}
2220 
2221 		pos += dist_to_next;
2222 	}
2223 
2224 	return (0);
2225 }
2226 
2227 /*
2228     struct vop_listextattr_args {
2229 	struct vop_generic_args a_gen;
2230 	struct vnode *a_vp;
2231 	int a_attrnamespace;
2232 	struct uio *a_uio;
2233 	size_t *a_size;
2234 	struct ucred *a_cred;
2235 	struct thread *a_td;
2236     };
2237 */
2238 static int
2239 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2240 {
2241 	struct vnode *vp = ap->a_vp;
2242 	struct uio *uio = ap->a_uio;
2243 	struct fuse_dispatcher fdi;
2244 	struct fuse_listxattr_in *list_xattr_in;
2245 	struct fuse_listxattr_out *list_xattr_out;
2246 	struct mount *mp = vnode_mount(vp);
2247 	struct thread *td = ap->a_td;
2248 	struct ucred *cred = ap->a_cred;
2249 	size_t len;
2250 	char *prefix;
2251 	char *attr_str;
2252 	char *bsd_list = NULL;
2253 	char *linux_list;
2254 	int bsd_list_len;
2255 	int linux_list_len;
2256 	int err;
2257 
2258 	if (fuse_isdeadfs(vp))
2259 		return (ENXIO);
2260 
2261 	if (!fsess_isimpl(mp, FUSE_LISTXATTR))
2262 		return EOPNOTSUPP;
2263 
2264 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2265 	if (err)
2266 		return err;
2267 
2268 	/*
2269 	 * Add space for a NUL and the period separator if enabled.
2270 	 * Default to looking for user attributes.
2271 	 */
2272 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2273 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2274 	else
2275 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2276 
2277 	len = strlen(prefix) + sizeof(extattr_namespace_separator) + 1;
2278 
2279 	fdisp_init(&fdi, sizeof(*list_xattr_in) + len);
2280 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2281 
2282 	/*
2283 	 * Retrieve Linux / FUSE compatible list size.
2284 	 */
2285 	list_xattr_in = fdi.indata;
2286 	list_xattr_in->size = 0;
2287 	attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2288 	snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2289 
2290 	err = fdisp_wait_answ(&fdi);
2291 	if (err != 0) {
2292 		if (err == ENOSYS) {
2293 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
2294 			err = EOPNOTSUPP;
2295 		}
2296 		goto out;
2297 	}
2298 
2299 	list_xattr_out = fdi.answ;
2300 	linux_list_len = list_xattr_out->size;
2301 	if (linux_list_len == 0) {
2302 		if (ap->a_size != NULL)
2303 			*ap->a_size = linux_list_len;
2304 		goto out;
2305 	}
2306 
2307 	/*
2308 	 * Retrieve Linux / FUSE compatible list values.
2309 	 */
2310 	fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2311 	list_xattr_in = fdi.indata;
2312 	list_xattr_in->size = linux_list_len + sizeof(*list_xattr_out);
2313 	attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2314 	snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2315 
2316 	err = fdisp_wait_answ(&fdi);
2317 	if (err != 0)
2318 		goto out;
2319 
2320 	linux_list = fdi.answ;
2321 	linux_list_len = fdi.iosize;
2322 
2323 	/*
2324 	 * Retrieve the BSD compatible list values.
2325 	 * The Linux / FUSE attribute list format isn't the same
2326 	 * as FreeBSD's format. So we need to transform it into
2327 	 * FreeBSD's format before giving it to the user.
2328 	 */
2329 	bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2330 	err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2331 	    bsd_list, &bsd_list_len);
2332 	if (err != 0)
2333 		goto out;
2334 
2335 	if (ap->a_size != NULL)
2336 		*ap->a_size = bsd_list_len;
2337 
2338 	if (uio != NULL)
2339 		err = uiomove(bsd_list, bsd_list_len, uio);
2340 
2341 out:
2342 	free(bsd_list, M_TEMP);
2343 	fdisp_destroy(&fdi);
2344 	return (err);
2345 }
2346 
2347 /*
2348     struct vop_deleteextattr_args {
2349 	struct vop_generic_args a_gen;
2350 	struct vnode *a_vp;
2351 	int a_attrnamespace;
2352 	const char *a_name;
2353 	struct ucred *a_cred;
2354 	struct thread *a_td;
2355     };
2356 */
2357 static int
2358 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2359 {
2360 	struct vnode *vp = ap->a_vp;
2361 	struct fuse_dispatcher fdi;
2362 	struct mount *mp = vnode_mount(vp);
2363 	struct thread *td = ap->a_td;
2364 	struct ucred *cred = ap->a_cred;
2365 	char *prefix;
2366 	size_t len;
2367 	char *attr_str;
2368 	int err;
2369 
2370 	if (fuse_isdeadfs(vp))
2371 		return (ENXIO);
2372 
2373 	if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2374 		return EOPNOTSUPP;
2375 
2376 	if (vfs_isrdonly(mp))
2377 		return EROFS;
2378 
2379 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2380 		VWRITE);
2381 	if (err)
2382 		return err;
2383 
2384 	/* Default to looking for user attributes. */
2385 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2386 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2387 	else
2388 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2389 
2390 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2391 	    strlen(ap->a_name) + 1;
2392 
2393 	fdisp_init(&fdi, len);
2394 	fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2395 
2396 	attr_str = fdi.indata;
2397 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2398 	    ap->a_name);
2399 
2400 	err = fdisp_wait_answ(&fdi);
2401 	if (err == ENOSYS) {
2402 		fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2403 		err = EOPNOTSUPP;
2404 	}
2405 
2406 	fdisp_destroy(&fdi);
2407 	return (err);
2408 }
2409 
2410 /*
2411     struct vnop_print_args {
2412 	struct vnode *a_vp;
2413     };
2414 */
2415 static int
2416 fuse_vnop_print(struct vop_print_args *ap)
2417 {
2418 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2419 
2420 	printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2421 	    (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2422 	    (uintmax_t)fvdat->nlookup,
2423 	    fvdat->flag);
2424 
2425 	return 0;
2426 }
2427 
2428 /*
2429  * Get an NFS filehandle for a FUSE file.
2430  *
2431  * This will only work for FUSE file systems that guarantee the uniqueness of
2432  * nodeid:generation, which most don't.
2433  */
2434 /*
2435 vop_vptofh {
2436 	IN struct vnode *a_vp;
2437 	IN struct fid *a_fhp;
2438 };
2439 */
2440 static int
2441 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
2442 {
2443 	struct vnode *vp = ap->a_vp;
2444 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
2445 	struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
2446 	_Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
2447 		"FUSE fid type is too big");
2448 	struct mount *mp = vnode_mount(vp);
2449 	struct fuse_data *data = fuse_get_mpdata(mp);
2450 	struct vattr va;
2451 	int err;
2452 
2453 	if (!(data->dataflags & FSESS_EXPORT_SUPPORT))
2454 		return EOPNOTSUPP;
2455 
2456 	err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
2457 	if (err)
2458 		return err;
2459 
2460 	/*ip = VTOI(ap->a_vp);*/
2461 	/*ufhp = (struct ufid *)ap->a_fhp;*/
2462 	fhp->len = sizeof(struct fuse_fid);
2463 	fhp->nid = fvdat->nid;
2464 	if (fvdat->generation <= UINT32_MAX)
2465 		fhp->gen = fvdat->generation;
2466 	else
2467 		return EOVERFLOW;
2468 	return (0);
2469 }
2470 
2471 
2472