xref: /freebsd/sys/fs/fuse/fuse_vnops.c (revision 4b50c451720d8b427757a6da1dd2bb4c52cd9e35)
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 /*
965     struct vnop_lookup_args {
966 	struct vnodeop_desc *a_desc;
967 	struct vnode *a_dvp;
968 	struct vnode **a_vpp;
969 	struct componentname *a_cnp;
970     };
971 */
972 int
973 fuse_vnop_lookup(struct vop_lookup_args *ap)
974 {
975 	struct vnode *dvp = ap->a_dvp;
976 	struct vnode **vpp = ap->a_vpp;
977 	struct componentname *cnp = ap->a_cnp;
978 	struct thread *td = cnp->cn_thread;
979 	struct ucred *cred = cnp->cn_cred;
980 
981 	int nameiop = cnp->cn_nameiop;
982 	int flags = cnp->cn_flags;
983 	int wantparent = flags & (LOCKPARENT | WANTPARENT);
984 	int islastcn = flags & ISLASTCN;
985 	struct mount *mp = vnode_mount(dvp);
986 
987 	int err = 0;
988 	int lookup_err = 0;
989 	struct vnode *vp = NULL;
990 
991 	struct fuse_dispatcher fdi;
992 	bool did_lookup = false;
993 	struct fuse_entry_out *feo = NULL;
994 	enum vtype vtyp;	/* vnode type of target */
995 	off_t filesize;		/* filesize of target */
996 
997 	uint64_t nid;
998 
999 	if (fuse_isdeadfs(dvp)) {
1000 		*vpp = NULL;
1001 		return ENXIO;
1002 	}
1003 	if (!vnode_isdir(dvp))
1004 		return ENOTDIR;
1005 
1006 	if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP))
1007 		return EROFS;
1008 
1009 	if ((err = fuse_internal_access(dvp, VEXEC, td, cred)))
1010 		return err;
1011 
1012 	if (flags & ISDOTDOT) {
1013 		KASSERT(VTOFUD(dvp)->flag & FN_PARENT_NID,
1014 			("Looking up .. is TODO"));
1015 		nid = VTOFUD(dvp)->parent_nid;
1016 		if (nid == 0)
1017 			return ENOENT;
1018 		/* .. is obviously a directory */
1019 		vtyp = VDIR;
1020 		filesize = 0;
1021 	} else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
1022 		nid = VTOI(dvp);
1023 		/* . is obviously a directory */
1024 		vtyp = VDIR;
1025 		filesize = 0;
1026 	} else {
1027 		struct timespec now, timeout;
1028 
1029 		err = cache_lookup(dvp, vpp, cnp, &timeout, NULL);
1030 		getnanouptime(&now);
1031 		SDT_PROBE3(fusefs, , vnops, cache_lookup, err, &timeout, &now);
1032 		switch (err) {
1033 		case -1:		/* positive match */
1034 			if (timespeccmp(&timeout, &now, >)) {
1035 				counter_u64_add(fuse_lookup_cache_hits, 1);
1036 			} else {
1037 				/* Cache timeout */
1038 				counter_u64_add(fuse_lookup_cache_misses, 1);
1039 				bintime_clear(
1040 					&VTOFUD(*vpp)->entry_cache_timeout);
1041 				cache_purge(*vpp);
1042 				if (dvp != *vpp)
1043 					vput(*vpp);
1044 				else
1045 					vrele(*vpp);
1046 				*vpp = NULL;
1047 				break;
1048 			}
1049 			return 0;
1050 
1051 		case 0:		/* no match in cache */
1052 			counter_u64_add(fuse_lookup_cache_misses, 1);
1053 			break;
1054 
1055 		case ENOENT:		/* negative match */
1056 			getnanouptime(&now);
1057 			if (timespeccmp(&timeout, &now, <=)) {
1058 				/* Cache timeout */
1059 				cache_purge_negative(dvp);
1060 				break;
1061 			}
1062 			/* fall through */
1063 		default:
1064 			return err;
1065 		}
1066 
1067 		nid = VTOI(dvp);
1068 		fdisp_init(&fdi, cnp->cn_namelen + 1);
1069 		fdisp_make(&fdi, FUSE_LOOKUP, mp, nid, td, cred);
1070 
1071 		memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1072 		((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1073 		lookup_err = fdisp_wait_answ(&fdi);
1074 		did_lookup = true;
1075 
1076 		if (!lookup_err) {
1077 			/* lookup call succeeded */
1078 			feo = (struct fuse_entry_out *)fdi.answ;
1079 			nid = feo->nodeid;
1080 			if (nid == 0) {
1081 				/* zero nodeid means ENOENT and cache it */
1082 				struct timespec timeout;
1083 
1084 				fdi.answ_stat = ENOENT;
1085 				lookup_err = ENOENT;
1086 				if (cnp->cn_flags & MAKEENTRY) {
1087 					fuse_validity_2_timespec(feo, &timeout);
1088 					cache_enter_time(dvp, *vpp, cnp,
1089 						&timeout, NULL);
1090 				}
1091 			} else if (nid == FUSE_ROOT_ID) {
1092 				lookup_err = EINVAL;
1093 			}
1094 			vtyp = IFTOVT(feo->attr.mode);
1095 			filesize = feo->attr.size;
1096 		}
1097 		if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT)) {
1098 			fdisp_destroy(&fdi);
1099 			return lookup_err;
1100 		}
1101 	}
1102 	/* lookup_err, if non-zero, must be ENOENT at this point */
1103 
1104 	if (lookup_err) {
1105 		/* Entry not found */
1106 		if ((nameiop == CREATE || nameiop == RENAME) && islastcn) {
1107 			err = fuse_internal_access(dvp, VWRITE, td, cred);
1108 			if (!err) {
1109 				/*
1110 				 * Set the SAVENAME flag to hold onto the
1111 				 * pathname for use later in VOP_CREATE or
1112 				 * VOP_RENAME.
1113 				 */
1114 				cnp->cn_flags |= SAVENAME;
1115 
1116 				err = EJUSTRETURN;
1117 			}
1118 		} else {
1119 			err = ENOENT;
1120 		}
1121 	} else {
1122 		/* Entry was found */
1123 		if (flags & ISDOTDOT) {
1124 			struct fuse_lookup_alloc_arg flaa;
1125 
1126 			flaa.nid = nid;
1127 			flaa.feo = feo;
1128 			flaa.cnp = cnp;
1129 			flaa.vtyp = vtyp;
1130 			err = vn_vget_ino_gen(dvp, fuse_lookup_alloc, &flaa, 0,
1131 				&vp);
1132 			*vpp = vp;
1133 		} else if (nid == VTOI(dvp)) {
1134 			vref(dvp);
1135 			*vpp = dvp;
1136 		} else {
1137 			struct fuse_vnode_data *fvdat;
1138 
1139 			err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp,
1140 			    &vp, cnp, vtyp);
1141 			if (err)
1142 				goto out;
1143 			*vpp = vp;
1144 
1145 			/*
1146 			 * In the case where we are looking up a FUSE node
1147 			 * represented by an existing cached vnode, and the
1148 			 * true size reported by FUSE_LOOKUP doesn't match
1149 			 * the vnode's cached size, then any cached writes
1150 			 * beyond the file's current size are lost.
1151 			 *
1152 			 * We can get here:
1153 			 * * following attribute cache expiration, or
1154 			 * * due a bug in the daemon, or
1155 			 */
1156 			fvdat = VTOFUD(vp);
1157 			if (vnode_isreg(vp) &&
1158 			    filesize != fvdat->cached_attrs.va_size &&
1159 			    fvdat->flag & FN_SIZECHANGE) {
1160 				/*
1161 				 * The FN_SIZECHANGE flag reflects a dirty
1162 				 * append.  If userspace lets us know our cache
1163 				 * is invalid, that write was lost.  (Dirty
1164 				 * writes that do not cause append are also
1165 				 * lost, but we don't detect them here.)
1166 				 *
1167 				 * XXX: Maybe disable WB caching on this mount.
1168 				 */
1169 				printf("%s: WB cache incoherent on %s!\n",
1170 				    __func__,
1171 				    vnode_mount(vp)->mnt_stat.f_mntonname);
1172 
1173 				fvdat->flag &= ~FN_SIZECHANGE;
1174 			}
1175 
1176 			MPASS(feo != NULL);
1177 			fuse_internal_cache_attrs(*vpp, &feo->attr,
1178 				feo->attr_valid, feo->attr_valid_nsec, NULL);
1179 			fuse_validity_2_bintime(feo->entry_valid,
1180 				feo->entry_valid_nsec,
1181 				&fvdat->entry_cache_timeout);
1182 
1183 			if ((nameiop == DELETE || nameiop == RENAME) &&
1184 				islastcn)
1185 			{
1186 				struct vattr dvattr;
1187 
1188 				err = fuse_internal_access(dvp, VWRITE, td,
1189 					cred);
1190 				if (err != 0)
1191 					goto out;
1192 				/*
1193 				 * if the parent's sticky bit is set, check
1194 				 * whether we're allowed to remove the file.
1195 				 * Need to figure out the vnode locking to make
1196 				 * this work.
1197 				 */
1198 				fuse_internal_getattr(dvp, &dvattr, cred, td);
1199 				if ((dvattr.va_mode & S_ISTXT) &&
1200 					fuse_internal_access(dvp, VADMIN, td,
1201 						cred) &&
1202 					fuse_internal_access(*vpp, VADMIN, td,
1203 						cred)) {
1204 					err = EPERM;
1205 					goto out;
1206 				}
1207 			}
1208 
1209 			if (islastcn && (
1210 				(nameiop == DELETE) ||
1211 				(nameiop == RENAME && wantparent))) {
1212 				cnp->cn_flags |= SAVENAME;
1213 			}
1214 
1215 		}
1216 	}
1217 out:
1218 	if (err) {
1219 		if (vp != NULL && dvp != vp)
1220 			vput(vp);
1221 		else if (vp != NULL)
1222 			vrele(vp);
1223 		*vpp = NULL;
1224 	}
1225 	if (did_lookup)
1226 		fdisp_destroy(&fdi);
1227 
1228 	return err;
1229 }
1230 
1231 /*
1232     struct vnop_mkdir_args {
1233 	struct vnode *a_dvp;
1234 	struct vnode **a_vpp;
1235 	struct componentname *a_cnp;
1236 	struct vattr *a_vap;
1237     };
1238 */
1239 static int
1240 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1241 {
1242 	struct vnode *dvp = ap->a_dvp;
1243 	struct vnode **vpp = ap->a_vpp;
1244 	struct componentname *cnp = ap->a_cnp;
1245 	struct vattr *vap = ap->a_vap;
1246 
1247 	struct fuse_mkdir_in fmdi;
1248 
1249 	if (fuse_isdeadfs(dvp)) {
1250 		return ENXIO;
1251 	}
1252 	fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1253 	fmdi.umask = curthread->td_proc->p_fd->fd_cmask;
1254 
1255 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1256 	    sizeof(fmdi), VDIR));
1257 }
1258 
1259 /*
1260     struct vnop_mknod_args {
1261 	struct vnode *a_dvp;
1262 	struct vnode **a_vpp;
1263 	struct componentname *a_cnp;
1264 	struct vattr *a_vap;
1265     };
1266 */
1267 static int
1268 fuse_vnop_mknod(struct vop_mknod_args *ap)
1269 {
1270 
1271 	struct vnode *dvp = ap->a_dvp;
1272 	struct vnode **vpp = ap->a_vpp;
1273 	struct componentname *cnp = ap->a_cnp;
1274 	struct vattr *vap = ap->a_vap;
1275 
1276 	if (fuse_isdeadfs(dvp))
1277 		return ENXIO;
1278 
1279 	return fuse_internal_mknod(dvp, vpp, cnp, vap);
1280 }
1281 
1282 /*
1283     struct vop_open_args {
1284 	struct vnode *a_vp;
1285 	int  a_mode;
1286 	struct ucred *a_cred;
1287 	struct thread *a_td;
1288 	int a_fdidx; / struct file *a_fp;
1289     };
1290 */
1291 static int
1292 fuse_vnop_open(struct vop_open_args *ap)
1293 {
1294 	struct vnode *vp = ap->a_vp;
1295 	int a_mode = ap->a_mode;
1296 	struct thread *td = ap->a_td;
1297 	struct ucred *cred = ap->a_cred;
1298 	pid_t pid = td->td_proc->p_pid;
1299 	struct fuse_vnode_data *fvdat;
1300 
1301 	if (fuse_isdeadfs(vp))
1302 		return ENXIO;
1303 	if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO)
1304 		return (EOPNOTSUPP);
1305 	if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0)
1306 		return EINVAL;
1307 
1308 	fvdat = VTOFUD(vp);
1309 
1310 	if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) {
1311 		fuse_vnode_open(vp, 0, td);
1312 		return 0;
1313 	}
1314 
1315 	return fuse_filehandle_open(vp, a_mode, NULL, td, cred);
1316 }
1317 
1318 static int
1319 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1320 {
1321 
1322 	switch (ap->a_name) {
1323 	case _PC_FILESIZEBITS:
1324 		*ap->a_retval = 64;
1325 		return (0);
1326 	case _PC_NAME_MAX:
1327 		*ap->a_retval = NAME_MAX;
1328 		return (0);
1329 	case _PC_LINK_MAX:
1330 		*ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1331 		return (0);
1332 	case _PC_SYMLINK_MAX:
1333 		*ap->a_retval = MAXPATHLEN;
1334 		return (0);
1335 	case _PC_NO_TRUNC:
1336 		*ap->a_retval = 1;
1337 		return (0);
1338 	default:
1339 		return (vop_stdpathconf(ap));
1340 	}
1341 }
1342 
1343 /*
1344     struct vnop_read_args {
1345 	struct vnode *a_vp;
1346 	struct uio *a_uio;
1347 	int  a_ioflag;
1348 	struct ucred *a_cred;
1349     };
1350 */
1351 static int
1352 fuse_vnop_read(struct vop_read_args *ap)
1353 {
1354 	struct vnode *vp = ap->a_vp;
1355 	struct uio *uio = ap->a_uio;
1356 	int ioflag = ap->a_ioflag;
1357 	struct ucred *cred = ap->a_cred;
1358 	pid_t pid = curthread->td_proc->p_pid;
1359 
1360 	if (fuse_isdeadfs(vp)) {
1361 		return ENXIO;
1362 	}
1363 
1364 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1365 		ioflag |= IO_DIRECT;
1366 	}
1367 
1368 	return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1369 }
1370 
1371 /*
1372     struct vnop_readdir_args {
1373 	struct vnode *a_vp;
1374 	struct uio *a_uio;
1375 	struct ucred *a_cred;
1376 	int *a_eofflag;
1377 	int *a_ncookies;
1378 	u_long **a_cookies;
1379     };
1380 */
1381 static int
1382 fuse_vnop_readdir(struct vop_readdir_args *ap)
1383 {
1384 	struct vnode *vp = ap->a_vp;
1385 	struct uio *uio = ap->a_uio;
1386 	struct ucred *cred = ap->a_cred;
1387 	struct fuse_filehandle *fufh = NULL;
1388 	struct fuse_iov cookediov;
1389 	int err = 0;
1390 	u_long *cookies;
1391 	off_t startoff;
1392 	ssize_t tresid;
1393 	int ncookies;
1394 	bool closefufh = false;
1395 	pid_t pid = curthread->td_proc->p_pid;
1396 
1397 	if (ap->a_eofflag)
1398 		*ap->a_eofflag = 0;
1399 	if (fuse_isdeadfs(vp)) {
1400 		return ENXIO;
1401 	}
1402 	if (				/* XXXIP ((uio_iovcnt(uio) > 1)) || */
1403 	    (uio_resid(uio) < sizeof(struct dirent))) {
1404 		return EINVAL;
1405 	}
1406 
1407 	tresid = uio->uio_resid;
1408 	startoff = uio->uio_offset;
1409 	err = fuse_filehandle_get_dir(vp, &fufh, cred, pid);
1410 	if (err == EBADF && vnode_mount(vp)->mnt_flag & MNT_EXPORTED) {
1411 		/*
1412 		 * nfsd will do VOP_READDIR without first doing VOP_OPEN.  We
1413 		 * must implicitly open the directory here
1414 		 */
1415 		err = fuse_filehandle_open(vp, FREAD, &fufh, curthread, cred);
1416 		if (err == 0) {
1417 			/*
1418 			 * When a directory is opened, it must be read from
1419 			 * the beginning.  Hopefully, the "startoff" still
1420 			 * exists as an offset cookie for the directory.
1421 			 * If not, it will read the entire directory without
1422 			 * returning any entries and just return eof.
1423 			 */
1424 			uio->uio_offset = 0;
1425 		}
1426 		closefufh = true;
1427 	}
1428 	if (err)
1429 		return (err);
1430 	if (ap->a_ncookies != NULL) {
1431 		ncookies = uio->uio_resid /
1432 			(offsetof(struct dirent, d_name) + 4) + 1;
1433 		cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
1434 		*ap->a_ncookies = ncookies;
1435 		*ap->a_cookies = cookies;
1436 	} else {
1437 		ncookies = 0;
1438 		cookies = NULL;
1439 	}
1440 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1441 	fiov_init(&cookediov, DIRCOOKEDSIZE);
1442 
1443 	err = fuse_internal_readdir(vp, uio, startoff, fufh, &cookediov,
1444 		&ncookies, cookies);
1445 
1446 	fiov_teardown(&cookediov);
1447 	if (closefufh)
1448 		fuse_filehandle_close(vp, fufh, curthread, cred);
1449 
1450 	if (ap->a_ncookies != NULL) {
1451 		if (err == 0) {
1452 			*ap->a_ncookies -= ncookies;
1453 		} else {
1454 			free(*ap->a_cookies, M_TEMP);
1455 			*ap->a_ncookies = 0;
1456 			*ap->a_cookies = NULL;
1457 		}
1458 	}
1459 	if (err == 0 && tresid == uio->uio_resid)
1460 		*ap->a_eofflag = 1;
1461 
1462 	return err;
1463 }
1464 
1465 /*
1466     struct vnop_readlink_args {
1467 	struct vnode *a_vp;
1468 	struct uio *a_uio;
1469 	struct ucred *a_cred;
1470     };
1471 */
1472 static int
1473 fuse_vnop_readlink(struct vop_readlink_args *ap)
1474 {
1475 	struct vnode *vp = ap->a_vp;
1476 	struct uio *uio = ap->a_uio;
1477 	struct ucred *cred = ap->a_cred;
1478 
1479 	struct fuse_dispatcher fdi;
1480 	int err;
1481 
1482 	if (fuse_isdeadfs(vp)) {
1483 		return ENXIO;
1484 	}
1485 	if (!vnode_islnk(vp)) {
1486 		return EINVAL;
1487 	}
1488 	fdisp_init(&fdi, 0);
1489 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1490 	if (err) {
1491 		goto out;
1492 	}
1493 	if (((char *)fdi.answ)[0] == '/' &&
1494 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1495 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1496 
1497 		err = uiomove(mpth, strlen(mpth), uio);
1498 	}
1499 	if (!err) {
1500 		err = uiomove(fdi.answ, fdi.iosize, uio);
1501 	}
1502 out:
1503 	fdisp_destroy(&fdi);
1504 	return err;
1505 }
1506 
1507 /*
1508     struct vnop_reclaim_args {
1509 	struct vnode *a_vp;
1510 	struct thread *a_td;
1511     };
1512 */
1513 static int
1514 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1515 {
1516 	struct vnode *vp = ap->a_vp;
1517 	struct thread *td = ap->a_td;
1518 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1519 	struct fuse_filehandle *fufh, *fufh_tmp;
1520 
1521 	if (!fvdat) {
1522 		panic("FUSE: no vnode data during recycling");
1523 	}
1524 	LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) {
1525 		printf("FUSE: vnode being reclaimed with open fufh "
1526 			"(type=%#x)", fufh->fufh_type);
1527 		fuse_filehandle_close(vp, fufh, td, NULL);
1528 	}
1529 
1530 	if (!fuse_isdeadfs(vp) && fvdat->nlookup > 0) {
1531 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1532 		    fvdat->nlookup);
1533 	}
1534 	cache_purge(vp);
1535 	vfs_hash_remove(vp);
1536 	fuse_vnode_destroy(vp);
1537 
1538 	return 0;
1539 }
1540 
1541 /*
1542     struct vnop_remove_args {
1543 	struct vnode *a_dvp;
1544 	struct vnode *a_vp;
1545 	struct componentname *a_cnp;
1546     };
1547 */
1548 static int
1549 fuse_vnop_remove(struct vop_remove_args *ap)
1550 {
1551 	struct vnode *dvp = ap->a_dvp;
1552 	struct vnode *vp = ap->a_vp;
1553 	struct componentname *cnp = ap->a_cnp;
1554 
1555 	int err;
1556 
1557 	if (fuse_isdeadfs(vp)) {
1558 		return ENXIO;
1559 	}
1560 	if (vnode_isdir(vp)) {
1561 		return EPERM;
1562 	}
1563 
1564 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1565 
1566 	return err;
1567 }
1568 
1569 /*
1570     struct vnop_rename_args {
1571 	struct vnode *a_fdvp;
1572 	struct vnode *a_fvp;
1573 	struct componentname *a_fcnp;
1574 	struct vnode *a_tdvp;
1575 	struct vnode *a_tvp;
1576 	struct componentname *a_tcnp;
1577     };
1578 */
1579 static int
1580 fuse_vnop_rename(struct vop_rename_args *ap)
1581 {
1582 	struct vnode *fdvp = ap->a_fdvp;
1583 	struct vnode *fvp = ap->a_fvp;
1584 	struct componentname *fcnp = ap->a_fcnp;
1585 	struct vnode *tdvp = ap->a_tdvp;
1586 	struct vnode *tvp = ap->a_tvp;
1587 	struct componentname *tcnp = ap->a_tcnp;
1588 	struct fuse_data *data;
1589 	bool newparent = fdvp != tdvp;
1590 	bool isdir = fvp->v_type == VDIR;
1591 	int err = 0;
1592 
1593 	if (fuse_isdeadfs(fdvp)) {
1594 		return ENXIO;
1595 	}
1596 	if (fvp->v_mount != tdvp->v_mount ||
1597 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1598 		SDT_PROBE2(fusefs, , vnops, trace, 1, "cross-device rename");
1599 		err = EXDEV;
1600 		goto out;
1601 	}
1602 	cache_purge(fvp);
1603 
1604 	/*
1605 	 * FUSE library is expected to check if target directory is not
1606 	 * under the source directory in the file system tree.
1607 	 * Linux performs this check at VFS level.
1608 	 */
1609 	/*
1610 	 * If source is a directory, and it will get a new parent, user must
1611 	 * have write permission to it, so ".." can be modified.
1612 	 */
1613 	data = fuse_get_mpdata(vnode_mount(tdvp));
1614 	if (data->dataflags & FSESS_DEFAULT_PERMISSIONS && isdir && newparent) {
1615 		err = fuse_internal_access(fvp, VWRITE,
1616 			tcnp->cn_thread, tcnp->cn_cred);
1617 		if (err)
1618 			goto out;
1619 	}
1620 	sx_xlock(&data->rename_lock);
1621 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1622 	if (err == 0) {
1623 		if (tdvp != fdvp)
1624 			fuse_vnode_setparent(fvp, tdvp);
1625 		if (tvp != NULL)
1626 			fuse_vnode_setparent(tvp, NULL);
1627 	}
1628 	sx_unlock(&data->rename_lock);
1629 
1630 	if (tvp != NULL && tvp != fvp) {
1631 		cache_purge(tvp);
1632 	}
1633 	if (vnode_isdir(fvp)) {
1634 		if ((tvp != NULL) && vnode_isdir(tvp)) {
1635 			cache_purge(tdvp);
1636 		}
1637 		cache_purge(fdvp);
1638 	}
1639 out:
1640 	if (tdvp == tvp) {
1641 		vrele(tdvp);
1642 	} else {
1643 		vput(tdvp);
1644 	}
1645 	if (tvp != NULL) {
1646 		vput(tvp);
1647 	}
1648 	vrele(fdvp);
1649 	vrele(fvp);
1650 
1651 	return err;
1652 }
1653 
1654 /*
1655     struct vnop_rmdir_args {
1656 	    struct vnode *a_dvp;
1657 	    struct vnode *a_vp;
1658 	    struct componentname *a_cnp;
1659     } *ap;
1660 */
1661 static int
1662 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1663 {
1664 	struct vnode *dvp = ap->a_dvp;
1665 	struct vnode *vp = ap->a_vp;
1666 
1667 	int err;
1668 
1669 	if (fuse_isdeadfs(vp)) {
1670 		return ENXIO;
1671 	}
1672 	if (VTOFUD(vp) == VTOFUD(dvp)) {
1673 		return EINVAL;
1674 	}
1675 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1676 
1677 	return err;
1678 }
1679 
1680 /*
1681     struct vnop_setattr_args {
1682 	struct vnode *a_vp;
1683 	struct vattr *a_vap;
1684 	struct ucred *a_cred;
1685 	struct thread *a_td;
1686     };
1687 */
1688 static int
1689 fuse_vnop_setattr(struct vop_setattr_args *ap)
1690 {
1691 	struct vnode *vp = ap->a_vp;
1692 	struct vattr *vap = ap->a_vap;
1693 	struct ucred *cred = ap->a_cred;
1694 	struct thread *td = curthread;
1695 	struct mount *mp;
1696 	struct fuse_data *data;
1697 	struct vattr old_va;
1698 	int dataflags;
1699 	int err = 0, err2;
1700 	accmode_t accmode = 0;
1701 	bool checkperm;
1702 	bool drop_suid = false;
1703 	gid_t cr_gid;
1704 
1705 	mp = vnode_mount(vp);
1706 	data = fuse_get_mpdata(mp);
1707 	dataflags = data->dataflags;
1708 	checkperm = dataflags & FSESS_DEFAULT_PERMISSIONS;
1709 	if (cred->cr_ngroups > 0)
1710 		cr_gid = cred->cr_groups[0];
1711 	else
1712 		cr_gid = 0;
1713 
1714 	if (fuse_isdeadfs(vp)) {
1715 		return ENXIO;
1716 	}
1717 
1718 	if (vap->va_uid != (uid_t)VNOVAL) {
1719 		if (checkperm) {
1720 			/* Only root may change a file's owner */
1721 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1722 			if (err) {
1723 				/* As a special case, allow the null chown */
1724 				err2 = fuse_internal_getattr(vp, &old_va, cred,
1725 					td);
1726 				if (err2)
1727 					return (err2);
1728 				if (vap->va_uid != old_va.va_uid)
1729 					return err;
1730 				else
1731 					accmode |= VADMIN;
1732 				drop_suid = true;
1733 			} else
1734 				accmode |= VADMIN;
1735 		} else
1736 			accmode |= VADMIN;
1737 	}
1738 	if (vap->va_gid != (gid_t)VNOVAL) {
1739 		if (checkperm && priv_check_cred(cred, PRIV_VFS_CHOWN))
1740 			drop_suid = true;
1741 		if (checkperm && !groupmember(vap->va_gid, cred))
1742 		{
1743 			/*
1744 			 * Non-root users may only chgrp to one of their own
1745 			 * groups
1746 			 */
1747 			err = priv_check_cred(cred, PRIV_VFS_CHOWN);
1748 			if (err) {
1749 				/* As a special case, allow the null chgrp */
1750 				err2 = fuse_internal_getattr(vp, &old_va, cred,
1751 					td);
1752 				if (err2)
1753 					return (err2);
1754 				if (vap->va_gid != old_va.va_gid)
1755 					return err;
1756 				accmode |= VADMIN;
1757 			} else
1758 				accmode |= VADMIN;
1759 		} else
1760 			accmode |= VADMIN;
1761 	}
1762 	if (vap->va_size != VNOVAL) {
1763 		switch (vp->v_type) {
1764 		case VDIR:
1765 			return (EISDIR);
1766 		case VLNK:
1767 		case VREG:
1768 			if (vfs_isrdonly(mp))
1769 				return (EROFS);
1770 			break;
1771 		default:
1772 			/*
1773 			 * According to POSIX, the result is unspecified
1774 			 * for file types other than regular files,
1775 			 * directories and shared memory objects.  We
1776 			 * don't support shared memory objects in the file
1777 			 * system, and have dubious support for truncating
1778 			 * symlinks.  Just ignore the request in other cases.
1779 			 */
1780 			return (0);
1781 		}
1782 		/* Don't set accmode.  Permission to trunc is checked upstack */
1783 	}
1784 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1785 		if (vap->va_vaflags & VA_UTIMES_NULL)
1786 			accmode |= VWRITE;
1787 		else
1788 			accmode |= VADMIN;
1789 	}
1790 	if (drop_suid) {
1791 		if (vap->va_mode != (mode_t)VNOVAL)
1792 			vap->va_mode &= ~(S_ISUID | S_ISGID);
1793 		else {
1794 			err = fuse_internal_getattr(vp, &old_va, cred, td);
1795 			if (err)
1796 				return (err);
1797 			vap->va_mode = old_va.va_mode & ~(S_ISUID | S_ISGID);
1798 		}
1799 	}
1800 	if (vap->va_mode != (mode_t)VNOVAL) {
1801 		/* Only root may set the sticky bit on non-directories */
1802 		if (checkperm && vp->v_type != VDIR && (vap->va_mode & S_ISTXT)
1803 		    && priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1804 			return EFTYPE;
1805 		if (checkperm && (vap->va_mode & S_ISGID)) {
1806 			err = fuse_internal_getattr(vp, &old_va, cred, td);
1807 			if (err)
1808 				return (err);
1809 			if (!groupmember(old_va.va_gid, cred)) {
1810 				err = priv_check_cred(cred, PRIV_VFS_SETGID);
1811 				if (err)
1812 					return (err);
1813 			}
1814 		}
1815 		accmode |= VADMIN;
1816 	}
1817 
1818 	if (vfs_isrdonly(mp))
1819 		return EROFS;
1820 
1821 	err = fuse_internal_access(vp, accmode, td, cred);
1822 	if (err)
1823 		return err;
1824 	else
1825 		return fuse_internal_setattr(vp, vap, td, cred);
1826 }
1827 
1828 /*
1829     struct vnop_strategy_args {
1830 	struct vnode *a_vp;
1831 	struct buf *a_bp;
1832     };
1833 */
1834 static int
1835 fuse_vnop_strategy(struct vop_strategy_args *ap)
1836 {
1837 	struct vnode *vp = ap->a_vp;
1838 	struct buf *bp = ap->a_bp;
1839 
1840 	if (!vp || fuse_isdeadfs(vp)) {
1841 		bp->b_ioflags |= BIO_ERROR;
1842 		bp->b_error = ENXIO;
1843 		bufdone(bp);
1844 		return 0;
1845 	}
1846 
1847 	/*
1848 	 * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags.
1849 	 * fuse_io_strategy sets bp's error fields
1850 	 */
1851 	(void)fuse_io_strategy(vp, bp);
1852 
1853 	return 0;
1854 }
1855 
1856 
1857 /*
1858     struct vnop_symlink_args {
1859 	struct vnode *a_dvp;
1860 	struct vnode **a_vpp;
1861 	struct componentname *a_cnp;
1862 	struct vattr *a_vap;
1863 	char *a_target;
1864     };
1865 */
1866 static int
1867 fuse_vnop_symlink(struct vop_symlink_args *ap)
1868 {
1869 	struct vnode *dvp = ap->a_dvp;
1870 	struct vnode **vpp = ap->a_vpp;
1871 	struct componentname *cnp = ap->a_cnp;
1872 	const char *target = ap->a_target;
1873 
1874 	struct fuse_dispatcher fdi;
1875 
1876 	int err;
1877 	size_t len;
1878 
1879 	if (fuse_isdeadfs(dvp)) {
1880 		return ENXIO;
1881 	}
1882 	/*
1883 	 * Unlike the other creator type calls, here we have to create a message
1884 	 * where the name of the new entry comes first, and the data describing
1885 	 * the entry comes second.
1886 	 * Hence we can't rely on our handy fuse_internal_newentry() routine,
1887 	 * but put together the message manually and just call the core part.
1888 	 */
1889 
1890 	len = strlen(target) + 1;
1891 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1892 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1893 
1894 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1895 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1896 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1897 
1898 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1899 	fdisp_destroy(&fdi);
1900 	return err;
1901 }
1902 
1903 /*
1904     struct vnop_write_args {
1905 	struct vnode *a_vp;
1906 	struct uio *a_uio;
1907 	int  a_ioflag;
1908 	struct ucred *a_cred;
1909     };
1910 */
1911 static int
1912 fuse_vnop_write(struct vop_write_args *ap)
1913 {
1914 	struct vnode *vp = ap->a_vp;
1915 	struct uio *uio = ap->a_uio;
1916 	int ioflag = ap->a_ioflag;
1917 	struct ucred *cred = ap->a_cred;
1918 	pid_t pid = curthread->td_proc->p_pid;
1919 
1920 	if (fuse_isdeadfs(vp)) {
1921 		return ENXIO;
1922 	}
1923 
1924 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1925 		ioflag |= IO_DIRECT;
1926 	}
1927 
1928 	return fuse_io_dispatch(vp, uio, ioflag, cred, pid);
1929 }
1930 
1931 static daddr_t
1932 fuse_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1933 {
1934 	const int biosize = fuse_iosize(vp);
1935 
1936 	return (off / biosize);
1937 }
1938 
1939 static int
1940 fuse_gbp_getblksz(struct vnode *vp, daddr_t lbn)
1941 {
1942 	off_t filesize;
1943 	int blksz, err;
1944 	const int biosize = fuse_iosize(vp);
1945 
1946 	err = fuse_vnode_size(vp, &filesize, NULL, NULL);
1947 	KASSERT(err == 0, ("vfs_bio_getpages can't handle errors here"));
1948 	if (err)
1949 		return biosize;
1950 
1951 	if ((off_t)lbn * biosize >= filesize) {
1952 		blksz = 0;
1953 	} else if ((off_t)(lbn + 1) * biosize > filesize) {
1954 		blksz = filesize - (off_t)lbn *biosize;
1955 	} else {
1956 		blksz = biosize;
1957 	}
1958 	return (blksz);
1959 }
1960 
1961 /*
1962     struct vnop_getpages_args {
1963 	struct vnode *a_vp;
1964 	vm_page_t *a_m;
1965 	int a_count;
1966 	int a_reqpage;
1967     };
1968 */
1969 static int
1970 fuse_vnop_getpages(struct vop_getpages_args *ap)
1971 {
1972 	struct vnode *vp = ap->a_vp;
1973 
1974 	if (!fsess_opt_mmap(vnode_mount(vp))) {
1975 		SDT_PROBE2(fusefs, , vnops, trace, 1,
1976 			"called on non-cacheable vnode??\n");
1977 		return (VM_PAGER_ERROR);
1978 	}
1979 
1980 	return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind,
1981 	    ap->a_rahead, fuse_gbp_getblkno, fuse_gbp_getblksz));
1982 }
1983 
1984 static const char extattr_namespace_separator = '.';
1985 
1986 /*
1987     struct vop_getextattr_args {
1988 	struct vop_generic_args a_gen;
1989 	struct vnode *a_vp;
1990 	int a_attrnamespace;
1991 	const char *a_name;
1992 	struct uio *a_uio;
1993 	size_t *a_size;
1994 	struct ucred *a_cred;
1995 	struct thread *a_td;
1996     };
1997 */
1998 static int
1999 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2000 {
2001 	struct vnode *vp = ap->a_vp;
2002 	struct uio *uio = ap->a_uio;
2003 	struct fuse_dispatcher fdi;
2004 	struct fuse_getxattr_in *get_xattr_in;
2005 	struct fuse_getxattr_out *get_xattr_out;
2006 	struct mount *mp = vnode_mount(vp);
2007 	struct thread *td = ap->a_td;
2008 	struct ucred *cred = ap->a_cred;
2009 	char *prefix;
2010 	char *attr_str;
2011 	size_t len;
2012 	int err;
2013 
2014 	if (fuse_isdeadfs(vp))
2015 		return (ENXIO);
2016 
2017 	if (!fsess_isimpl(mp, FUSE_GETXATTR))
2018 		return EOPNOTSUPP;
2019 
2020 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2021 	if (err)
2022 		return err;
2023 
2024 	/* Default to looking for user attributes. */
2025 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2026 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2027 	else
2028 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2029 
2030 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2031 	    strlen(ap->a_name) + 1;
2032 
2033 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2034 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2035 
2036 	get_xattr_in = fdi.indata;
2037 	/*
2038 	 * Check to see whether we're querying the available size or
2039 	 * issuing the actual request.  If we pass in 0, we get back struct
2040 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2041 	 * that much data, without the struct fuse_getxattr_out header.
2042 	 */
2043 	if (uio == NULL)
2044 		get_xattr_in->size = 0;
2045 	else
2046 		get_xattr_in->size = uio->uio_resid;
2047 
2048 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2049 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2050 	    ap->a_name);
2051 
2052 	err = fdisp_wait_answ(&fdi);
2053 	if (err != 0) {
2054 		if (err == ENOSYS) {
2055 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2056 			err = EOPNOTSUPP;
2057 		}
2058 		goto out;
2059 	}
2060 
2061 	get_xattr_out = fdi.answ;
2062 
2063 	if (ap->a_size != NULL)
2064 		*ap->a_size = get_xattr_out->size;
2065 
2066 	if (uio != NULL)
2067 		err = uiomove(fdi.answ, fdi.iosize, uio);
2068 
2069 out:
2070 	fdisp_destroy(&fdi);
2071 	return (err);
2072 }
2073 
2074 /*
2075     struct vop_setextattr_args {
2076 	struct vop_generic_args a_gen;
2077 	struct vnode *a_vp;
2078 	int a_attrnamespace;
2079 	const char *a_name;
2080 	struct uio *a_uio;
2081 	struct ucred *a_cred;
2082 	struct thread *a_td;
2083     };
2084 */
2085 static int
2086 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2087 {
2088 	struct vnode *vp = ap->a_vp;
2089 	struct uio *uio = ap->a_uio;
2090 	struct fuse_dispatcher fdi;
2091 	struct fuse_setxattr_in *set_xattr_in;
2092 	struct mount *mp = vnode_mount(vp);
2093 	struct thread *td = ap->a_td;
2094 	struct ucred *cred = ap->a_cred;
2095 	char *prefix;
2096 	size_t len;
2097 	char *attr_str;
2098 	int err;
2099 
2100 	if (fuse_isdeadfs(vp))
2101 		return (ENXIO);
2102 
2103 	if (!fsess_isimpl(mp, FUSE_SETXATTR))
2104 		return EOPNOTSUPP;
2105 
2106 	if (vfs_isrdonly(mp))
2107 		return EROFS;
2108 
2109 	/* Deleting xattrs must use VOP_DELETEEXTATTR instead */
2110 	if (ap->a_uio == NULL) {
2111 		/*
2112 		 * If we got here as fallback from VOP_DELETEEXTATTR, then
2113 		 * return EOPNOTSUPP.
2114 		 */
2115 		if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2116 			return (EOPNOTSUPP);
2117 		else
2118 			return (EINVAL);
2119 	}
2120 
2121 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2122 		VWRITE);
2123 	if (err)
2124 		return err;
2125 
2126 	/* Default to looking for user attributes. */
2127 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2128 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2129 	else
2130 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2131 
2132 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2133 	    strlen(ap->a_name) + 1;
2134 
2135 	fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2136 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2137 
2138 	set_xattr_in = fdi.indata;
2139 	set_xattr_in->size = uio->uio_resid;
2140 
2141 	attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2142 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2143 	    ap->a_name);
2144 
2145 	err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2146 	    uio->uio_resid, uio);
2147 	if (err != 0) {
2148 		goto out;
2149 	}
2150 
2151 	err = fdisp_wait_answ(&fdi);
2152 
2153 	if (err == ENOSYS) {
2154 		fsess_set_notimpl(mp, FUSE_SETXATTR);
2155 		err = EOPNOTSUPP;
2156 	}
2157 	if (err == ERESTART) {
2158 		/* Can't restart after calling uiomove */
2159 		err = EINTR;
2160 	}
2161 
2162 out:
2163 	fdisp_destroy(&fdi);
2164 	return (err);
2165 }
2166 
2167 /*
2168  * The Linux / FUSE extended attribute list is simply a collection of
2169  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2170  * byte length followed by a non-NUL terminated string.  So, this allows
2171  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2172  * Linux attribute names are reported with the namespace as a prefix (e.g.
2173  * "user.attribute_name"), but in FreeBSD they are reported without the
2174  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2175  *
2176  * user.attr_name1\0user.attr_name2\0
2177  *
2178  * to:
2179  *
2180  * <num>attr_name1<num>attr_name2
2181  *
2182  * Where "<num>" is a single byte number of characters in the attribute name.
2183  *
2184  * Args:
2185  * prefix - exattr namespace prefix string
2186  * list, list_len - input list with namespace prefixes
2187  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2188  */
2189 static int
2190 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2191     char *bsd_list, int *bsd_list_len)
2192 {
2193 	int len, pos, dist_to_next, prefix_len;
2194 
2195 	pos = 0;
2196 	*bsd_list_len = 0;
2197 	prefix_len = strlen(prefix);
2198 
2199 	while (pos < list_len && list[pos] != '\0') {
2200 		dist_to_next = strlen(&list[pos]) + 1;
2201 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2202 		    list[pos + prefix_len] == extattr_namespace_separator) {
2203 			len = dist_to_next -
2204 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2205 			if (len >= EXTATTR_MAXNAMELEN)
2206 				return (ENAMETOOLONG);
2207 
2208 			bsd_list[*bsd_list_len] = len;
2209 			memcpy(&bsd_list[*bsd_list_len + 1],
2210 			    &list[pos + prefix_len +
2211 			    sizeof(extattr_namespace_separator)], len);
2212 
2213 			*bsd_list_len += len + 1;
2214 		}
2215 
2216 		pos += dist_to_next;
2217 	}
2218 
2219 	return (0);
2220 }
2221 
2222 /*
2223  * List extended attributes
2224  *
2225  * The FUSE_LISTXATTR operation is based on Linux's listxattr(2) syscall, which
2226  * has a number of differences compared to its FreeBSD equivalent,
2227  * extattr_list_file:
2228  *
2229  * - FUSE_LISTXATTR returns all extended attributes across all namespaces,
2230  *   whereas listxattr(2) only returns attributes for a single namespace
2231  * - FUSE_LISTXATTR prepends each attribute name with "namespace."
2232  * - If the provided buffer is not large enough to hold the result,
2233  *   FUSE_LISTXATTR should return ERANGE, whereas listxattr is expected to
2234  *   return as many results as will fit.
2235  */
2236 /*
2237     struct vop_listextattr_args {
2238 	struct vop_generic_args a_gen;
2239 	struct vnode *a_vp;
2240 	int a_attrnamespace;
2241 	struct uio *a_uio;
2242 	size_t *a_size;
2243 	struct ucred *a_cred;
2244 	struct thread *a_td;
2245     };
2246 */
2247 static int
2248 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2249 {
2250 	struct vnode *vp = ap->a_vp;
2251 	struct uio *uio = ap->a_uio;
2252 	struct fuse_dispatcher fdi;
2253 	struct fuse_listxattr_in *list_xattr_in;
2254 	struct fuse_listxattr_out *list_xattr_out;
2255 	struct mount *mp = vnode_mount(vp);
2256 	struct thread *td = ap->a_td;
2257 	struct ucred *cred = ap->a_cred;
2258 	char *prefix;
2259 	char *bsd_list = NULL;
2260 	char *linux_list;
2261 	int bsd_list_len;
2262 	int linux_list_len;
2263 	int err;
2264 
2265 	if (fuse_isdeadfs(vp))
2266 		return (ENXIO);
2267 
2268 	if (!fsess_isimpl(mp, FUSE_LISTXATTR))
2269 		return EOPNOTSUPP;
2270 
2271 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD);
2272 	if (err)
2273 		return err;
2274 
2275 	/*
2276 	 * Add space for a NUL and the period separator if enabled.
2277 	 * Default to looking for user attributes.
2278 	 */
2279 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2280 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2281 	else
2282 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2283 
2284 	fdisp_init(&fdi, sizeof(*list_xattr_in));
2285 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2286 
2287 	/*
2288 	 * Retrieve Linux / FUSE compatible list size.
2289 	 */
2290 	list_xattr_in = fdi.indata;
2291 	list_xattr_in->size = 0;
2292 
2293 	err = fdisp_wait_answ(&fdi);
2294 	if (err != 0) {
2295 		if (err == ENOSYS) {
2296 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
2297 			err = EOPNOTSUPP;
2298 		}
2299 		goto out;
2300 	}
2301 
2302 	list_xattr_out = fdi.answ;
2303 	linux_list_len = list_xattr_out->size;
2304 	if (linux_list_len == 0) {
2305 		if (ap->a_size != NULL)
2306 			*ap->a_size = linux_list_len;
2307 		goto out;
2308 	}
2309 
2310 	/*
2311 	 * Retrieve Linux / FUSE compatible list values.
2312 	 */
2313 	fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2314 	list_xattr_in = fdi.indata;
2315 	list_xattr_in->size = linux_list_len;
2316 
2317 	err = fdisp_wait_answ(&fdi);
2318 	if (err == ERANGE) {
2319 		/*
2320 		 * Race detected.  The attribute list must've grown since the
2321 		 * first FUSE_LISTXATTR call.  Start over.  Go all the way back
2322 		 * to userland so we can process signals, if necessary, before
2323 		 * restarting.
2324 		 */
2325 		err = ERESTART;
2326 		goto out;
2327 	} else if (err != 0)
2328 		goto out;
2329 
2330 	linux_list = fdi.answ;
2331 	/* FUSE doesn't allow the server to return more data than requested */
2332 	if (fdi.iosize > linux_list_len) {
2333 		printf("WARNING: FUSE protocol violation.  Server returned "
2334 			"more extended attribute data than requested; "
2335 			"should've returned ERANGE instead");
2336 	} else {
2337 		/* But returning less data is fine */
2338 		linux_list_len = fdi.iosize;
2339 	}
2340 
2341 	/*
2342 	 * Retrieve the BSD compatible list values.
2343 	 * The Linux / FUSE attribute list format isn't the same
2344 	 * as FreeBSD's format. So we need to transform it into
2345 	 * FreeBSD's format before giving it to the user.
2346 	 */
2347 	bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2348 	err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2349 	    bsd_list, &bsd_list_len);
2350 	if (err != 0)
2351 		goto out;
2352 
2353 	if (ap->a_size != NULL)
2354 		*ap->a_size = bsd_list_len;
2355 
2356 	if (uio != NULL)
2357 		err = uiomove(bsd_list, bsd_list_len, uio);
2358 
2359 out:
2360 	free(bsd_list, M_TEMP);
2361 	fdisp_destroy(&fdi);
2362 	return (err);
2363 }
2364 
2365 /*
2366     struct vop_deleteextattr_args {
2367 	struct vop_generic_args a_gen;
2368 	struct vnode *a_vp;
2369 	int a_attrnamespace;
2370 	const char *a_name;
2371 	struct ucred *a_cred;
2372 	struct thread *a_td;
2373     };
2374 */
2375 static int
2376 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2377 {
2378 	struct vnode *vp = ap->a_vp;
2379 	struct fuse_dispatcher fdi;
2380 	struct mount *mp = vnode_mount(vp);
2381 	struct thread *td = ap->a_td;
2382 	struct ucred *cred = ap->a_cred;
2383 	char *prefix;
2384 	size_t len;
2385 	char *attr_str;
2386 	int err;
2387 
2388 	if (fuse_isdeadfs(vp))
2389 		return (ENXIO);
2390 
2391 	if (!fsess_isimpl(mp, FUSE_REMOVEXATTR))
2392 		return EOPNOTSUPP;
2393 
2394 	if (vfs_isrdonly(mp))
2395 		return EROFS;
2396 
2397 	err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td,
2398 		VWRITE);
2399 	if (err)
2400 		return err;
2401 
2402 	/* Default to looking for user attributes. */
2403 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2404 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2405 	else
2406 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2407 
2408 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2409 	    strlen(ap->a_name) + 1;
2410 
2411 	fdisp_init(&fdi, len);
2412 	fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2413 
2414 	attr_str = fdi.indata;
2415 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2416 	    ap->a_name);
2417 
2418 	err = fdisp_wait_answ(&fdi);
2419 	if (err == ENOSYS) {
2420 		fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2421 		err = EOPNOTSUPP;
2422 	}
2423 
2424 	fdisp_destroy(&fdi);
2425 	return (err);
2426 }
2427 
2428 /*
2429     struct vnop_print_args {
2430 	struct vnode *a_vp;
2431     };
2432 */
2433 static int
2434 fuse_vnop_print(struct vop_print_args *ap)
2435 {
2436 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2437 
2438 	printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2439 	    (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2440 	    (uintmax_t)fvdat->nlookup,
2441 	    fvdat->flag);
2442 
2443 	return 0;
2444 }
2445 
2446 /*
2447  * Get an NFS filehandle for a FUSE file.
2448  *
2449  * This will only work for FUSE file systems that guarantee the uniqueness of
2450  * nodeid:generation, which most don't.
2451  */
2452 /*
2453 vop_vptofh {
2454 	IN struct vnode *a_vp;
2455 	IN struct fid *a_fhp;
2456 };
2457 */
2458 static int
2459 fuse_vnop_vptofh(struct vop_vptofh_args *ap)
2460 {
2461 	struct vnode *vp = ap->a_vp;
2462 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
2463 	struct fuse_fid *fhp = (struct fuse_fid *)(ap->a_fhp);
2464 	_Static_assert(sizeof(struct fuse_fid) <= sizeof(struct fid),
2465 		"FUSE fid type is too big");
2466 	struct mount *mp = vnode_mount(vp);
2467 	struct fuse_data *data = fuse_get_mpdata(mp);
2468 	struct vattr va;
2469 	int err;
2470 
2471 	if (!(data->dataflags & FSESS_EXPORT_SUPPORT))
2472 		return EOPNOTSUPP;
2473 
2474 	err = fuse_internal_getattr(vp, &va, curthread->td_ucred, curthread);
2475 	if (err)
2476 		return err;
2477 
2478 	/*ip = VTOI(ap->a_vp);*/
2479 	/*ufhp = (struct ufid *)ap->a_fhp;*/
2480 	fhp->len = sizeof(struct fuse_fid);
2481 	fhp->nid = fvdat->nid;
2482 	if (fvdat->generation <= UINT32_MAX)
2483 		fhp->gen = fvdat->generation;
2484 	else
2485 		return EOVERFLOW;
2486 	return (0);
2487 }
2488 
2489 
2490