xref: /freebsd/sys/fs/fuse/fuse_vnops.c (revision 190cef3d52236565eb22e18b33e9e865ec634aa3)
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  * Redistribution and use in source and binary forms, with or without
37  * modification, are permitted provided that the following conditions
38  * are met:
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  *
45  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND
46  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48  * ARE DISCLAIMED.  IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
49  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55  * SUCH DAMAGE.
56  */
57 
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD$");
60 
61 #include <sys/types.h>
62 #include <sys/module.h>
63 #include <sys/systm.h>
64 #include <sys/errno.h>
65 #include <sys/param.h>
66 #include <sys/kernel.h>
67 #include <sys/conf.h>
68 #include <sys/uio.h>
69 #include <sys/malloc.h>
70 #include <sys/queue.h>
71 #include <sys/lock.h>
72 #include <sys/rwlock.h>
73 #include <sys/sx.h>
74 #include <sys/proc.h>
75 #include <sys/mount.h>
76 #include <sys/vnode.h>
77 #include <sys/namei.h>
78 #include <sys/extattr.h>
79 #include <sys/stat.h>
80 #include <sys/unistd.h>
81 #include <sys/filedesc.h>
82 #include <sys/file.h>
83 #include <sys/fcntl.h>
84 #include <sys/dirent.h>
85 #include <sys/bio.h>
86 #include <sys/buf.h>
87 #include <sys/sysctl.h>
88 #include <sys/vmmeter.h>
89 
90 #include <vm/vm.h>
91 #include <vm/vm_extern.h>
92 #include <vm/pmap.h>
93 #include <vm/vm_map.h>
94 #include <vm/vm_page.h>
95 #include <vm/vm_param.h>
96 #include <vm/vm_object.h>
97 #include <vm/vm_pager.h>
98 #include <vm/vnode_pager.h>
99 #include <vm/vm_object.h>
100 
101 #include "fuse.h"
102 #include "fuse_file.h"
103 #include "fuse_internal.h"
104 #include "fuse_ipc.h"
105 #include "fuse_node.h"
106 #include "fuse_param.h"
107 #include "fuse_io.h"
108 
109 #include <sys/priv.h>
110 
111 #define FUSE_DEBUG_MODULE VNOPS
112 #include "fuse_debug.h"
113 
114 /* vnode ops */
115 static vop_access_t fuse_vnop_access;
116 static vop_close_t fuse_vnop_close;
117 static vop_create_t fuse_vnop_create;
118 static vop_deleteextattr_t fuse_vnop_deleteextattr;
119 static vop_fsync_t fuse_vnop_fsync;
120 static vop_getattr_t fuse_vnop_getattr;
121 static vop_getextattr_t fuse_vnop_getextattr;
122 static vop_inactive_t fuse_vnop_inactive;
123 static vop_link_t fuse_vnop_link;
124 static vop_listextattr_t fuse_vnop_listextattr;
125 static vop_lookup_t fuse_vnop_lookup;
126 static vop_mkdir_t fuse_vnop_mkdir;
127 static vop_mknod_t fuse_vnop_mknod;
128 static vop_open_t fuse_vnop_open;
129 static vop_pathconf_t fuse_vnop_pathconf;
130 static vop_read_t fuse_vnop_read;
131 static vop_readdir_t fuse_vnop_readdir;
132 static vop_readlink_t fuse_vnop_readlink;
133 static vop_reclaim_t fuse_vnop_reclaim;
134 static vop_remove_t fuse_vnop_remove;
135 static vop_rename_t fuse_vnop_rename;
136 static vop_rmdir_t fuse_vnop_rmdir;
137 static vop_setattr_t fuse_vnop_setattr;
138 static vop_setextattr_t fuse_vnop_setextattr;
139 static vop_strategy_t fuse_vnop_strategy;
140 static vop_symlink_t fuse_vnop_symlink;
141 static vop_write_t fuse_vnop_write;
142 static vop_getpages_t fuse_vnop_getpages;
143 static vop_putpages_t fuse_vnop_putpages;
144 static vop_print_t fuse_vnop_print;
145 
146 struct vop_vector fuse_vnops = {
147 	.vop_default = &default_vnodeops,
148 	.vop_access = fuse_vnop_access,
149 	.vop_close = fuse_vnop_close,
150 	.vop_create = fuse_vnop_create,
151 	.vop_deleteextattr = fuse_vnop_deleteextattr,
152 	.vop_fsync = fuse_vnop_fsync,
153 	.vop_getattr = fuse_vnop_getattr,
154 	.vop_getextattr = fuse_vnop_getextattr,
155 	.vop_inactive = fuse_vnop_inactive,
156 	.vop_link = fuse_vnop_link,
157 	.vop_listextattr = fuse_vnop_listextattr,
158 	.vop_lookup = fuse_vnop_lookup,
159 	.vop_mkdir = fuse_vnop_mkdir,
160 	.vop_mknod = fuse_vnop_mknod,
161 	.vop_open = fuse_vnop_open,
162 	.vop_pathconf = fuse_vnop_pathconf,
163 	.vop_read = fuse_vnop_read,
164 	.vop_readdir = fuse_vnop_readdir,
165 	.vop_readlink = fuse_vnop_readlink,
166 	.vop_reclaim = fuse_vnop_reclaim,
167 	.vop_remove = fuse_vnop_remove,
168 	.vop_rename = fuse_vnop_rename,
169 	.vop_rmdir = fuse_vnop_rmdir,
170 	.vop_setattr = fuse_vnop_setattr,
171 	.vop_setextattr = fuse_vnop_setextattr,
172 	.vop_strategy = fuse_vnop_strategy,
173 	.vop_symlink = fuse_vnop_symlink,
174 	.vop_write = fuse_vnop_write,
175 	.vop_getpages = fuse_vnop_getpages,
176 	.vop_putpages = fuse_vnop_putpages,
177 	.vop_print = fuse_vnop_print,
178 };
179 
180 static u_long fuse_lookup_cache_hits = 0;
181 
182 SYSCTL_ULONG(_vfs_fuse, OID_AUTO, lookup_cache_hits, CTLFLAG_RD,
183     &fuse_lookup_cache_hits, 0, "");
184 
185 static u_long fuse_lookup_cache_misses = 0;
186 
187 SYSCTL_ULONG(_vfs_fuse, OID_AUTO, lookup_cache_misses, CTLFLAG_RD,
188     &fuse_lookup_cache_misses, 0, "");
189 
190 int	fuse_lookup_cache_enable = 1;
191 
192 SYSCTL_INT(_vfs_fuse, OID_AUTO, lookup_cache_enable, CTLFLAG_RW,
193     &fuse_lookup_cache_enable, 0, "");
194 
195 /*
196  * XXX: This feature is highly experimental and can bring to instabilities,
197  * needs revisiting before to be enabled by default.
198  */
199 static int fuse_reclaim_revoked = 0;
200 
201 SYSCTL_INT(_vfs_fuse, OID_AUTO, reclaim_revoked, CTLFLAG_RW,
202     &fuse_reclaim_revoked, 0, "");
203 
204 int	fuse_pbuf_freecnt = -1;
205 
206 #define fuse_vm_page_lock(m)		vm_page_lock((m));
207 #define fuse_vm_page_unlock(m)		vm_page_unlock((m));
208 #define fuse_vm_page_lock_queues()	((void)0)
209 #define fuse_vm_page_unlock_queues()	((void)0)
210 
211 /*
212     struct vnop_access_args {
213         struct vnode *a_vp;
214 #if VOP_ACCESS_TAKES_ACCMODE_T
215         accmode_t a_accmode;
216 #else
217         int a_mode;
218 #endif
219         struct ucred *a_cred;
220         struct thread *a_td;
221     };
222 */
223 static int
224 fuse_vnop_access(struct vop_access_args *ap)
225 {
226 	struct vnode *vp = ap->a_vp;
227 	int accmode = ap->a_accmode;
228 	struct ucred *cred = ap->a_cred;
229 
230 	struct fuse_access_param facp;
231 	struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
232 
233 	int err;
234 
235 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
236 
237 	if (fuse_isdeadfs(vp)) {
238 		if (vnode_isvroot(vp)) {
239 			return 0;
240 		}
241 		return ENXIO;
242 	}
243 	if (!(data->dataflags & FSESS_INITED)) {
244 		if (vnode_isvroot(vp)) {
245 			if (priv_check_cred(cred, PRIV_VFS_ADMIN, 0) ||
246 			    (fuse_match_cred(data->daemoncred, cred) == 0)) {
247 				return 0;
248 			}
249 		}
250 		return EBADF;
251 	}
252 	if (vnode_islnk(vp)) {
253 		return 0;
254 	}
255 	bzero(&facp, sizeof(facp));
256 
257 	err = fuse_internal_access(vp, accmode, &facp, ap->a_td, ap->a_cred);
258 	FS_DEBUG2G("err=%d accmode=0x%x\n", err, accmode);
259 	return err;
260 }
261 
262 /*
263     struct vnop_close_args {
264 	struct vnode *a_vp;
265 	int  a_fflag;
266 	struct ucred *a_cred;
267 	struct thread *a_td;
268     };
269 */
270 static int
271 fuse_vnop_close(struct vop_close_args *ap)
272 {
273 	struct vnode *vp = ap->a_vp;
274 	struct ucred *cred = ap->a_cred;
275 	int fflag = ap->a_fflag;
276 	fufh_type_t fufh_type;
277 
278 	fuse_trace_printf_vnop();
279 
280 	if (fuse_isdeadfs(vp)) {
281 		return 0;
282 	}
283 	if (vnode_isdir(vp)) {
284 		if (fuse_filehandle_valid(vp, FUFH_RDONLY)) {
285 			fuse_filehandle_close(vp, FUFH_RDONLY, NULL, cred);
286 		}
287 		return 0;
288 	}
289 	if (fflag & IO_NDELAY) {
290 		return 0;
291 	}
292 	fufh_type = fuse_filehandle_xlate_from_fflags(fflag);
293 
294 	if (!fuse_filehandle_valid(vp, fufh_type)) {
295 		int i;
296 
297 		for (i = 0; i < FUFH_MAXTYPE; i++)
298 			if (fuse_filehandle_valid(vp, i))
299 				break;
300 		if (i == FUFH_MAXTYPE)
301 			panic("FUSE: fufh type %d found to be invalid in close"
302 			      " (fflag=0x%x)\n",
303 			      fufh_type, fflag);
304 	}
305 	if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
306 		fuse_vnode_savesize(vp, cred);
307 	}
308 	return 0;
309 }
310 
311 /*
312     struct vnop_create_args {
313 	struct vnode *a_dvp;
314 	struct vnode **a_vpp;
315 	struct componentname *a_cnp;
316 	struct vattr *a_vap;
317     };
318 */
319 static int
320 fuse_vnop_create(struct vop_create_args *ap)
321 {
322 	struct vnode *dvp = ap->a_dvp;
323 	struct vnode **vpp = ap->a_vpp;
324 	struct componentname *cnp = ap->a_cnp;
325 	struct vattr *vap = ap->a_vap;
326 	struct thread *td = cnp->cn_thread;
327 	struct ucred *cred = cnp->cn_cred;
328 
329 	struct fuse_open_in *foi;
330 	struct fuse_entry_out *feo;
331 	struct fuse_dispatcher fdi;
332 	struct fuse_dispatcher *fdip = &fdi;
333 
334 	int err;
335 
336 	struct mount *mp = vnode_mount(dvp);
337 	uint64_t parentnid = VTOFUD(dvp)->nid;
338 	mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
339 	uint64_t x_fh_id;
340 	uint32_t x_open_flags;
341 
342 	fuse_trace_printf_vnop();
343 
344 	if (fuse_isdeadfs(dvp)) {
345 		return ENXIO;
346 	}
347 	bzero(&fdi, sizeof(fdi));
348 
349 	/* XXX:	Will we ever want devices ? */
350 	if ((vap->va_type != VREG)) {
351 		printf("fuse_vnop_create: unsupported va_type %d\n",
352 		    vap->va_type);
353 		return (EINVAL);
354 	}
355 	debug_printf("parent nid = %ju, mode = %x\n", (uintmax_t)parentnid,
356 	    mode);
357 
358 	fdisp_init(fdip, sizeof(*foi) + cnp->cn_namelen + 1);
359 	if (!fsess_isimpl(mp, FUSE_CREATE)) {
360 		debug_printf("eh, daemon doesn't implement create?\n");
361 		return (EINVAL);
362 	}
363 	fdisp_make(fdip, FUSE_CREATE, vnode_mount(dvp), parentnid, td, cred);
364 
365 	foi = fdip->indata;
366 	foi->mode = mode;
367 	foi->flags = O_CREAT | O_RDWR;
368 
369 	memcpy((char *)fdip->indata + sizeof(*foi), cnp->cn_nameptr,
370 	    cnp->cn_namelen);
371 	((char *)fdip->indata)[sizeof(*foi) + cnp->cn_namelen] = '\0';
372 
373 	err = fdisp_wait_answ(fdip);
374 
375 	if (err) {
376 		if (err == ENOSYS)
377 			fsess_set_notimpl(mp, FUSE_CREATE);
378 		debug_printf("create: got err=%d from daemon\n", err);
379 		goto out;
380 	}
381 
382 	feo = fdip->answ;
383 
384 	if ((err = fuse_internal_checkentry(feo, VREG))) {
385 		goto out;
386 	}
387 	err = fuse_vnode_get(mp, feo->nodeid, dvp, vpp, cnp, VREG);
388 	if (err) {
389 		struct fuse_release_in *fri;
390 		uint64_t nodeid = feo->nodeid;
391 		uint64_t fh_id = ((struct fuse_open_out *)(feo + 1))->fh;
392 
393 		fdisp_init(fdip, sizeof(*fri));
394 		fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
395 		fri = fdip->indata;
396 		fri->fh = fh_id;
397 		fri->flags = OFLAGS(mode);
398 		fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
399 		fuse_insert_message(fdip->tick);
400 		return err;
401 	}
402 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
403 
404 	fdip->answ = feo + 1;
405 
406 	x_fh_id = ((struct fuse_open_out *)(feo + 1))->fh;
407 	x_open_flags = ((struct fuse_open_out *)(feo + 1))->open_flags;
408 	fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, x_fh_id);
409 	fuse_vnode_open(*vpp, x_open_flags, td);
410 	cache_purge_negative(dvp);
411 
412 out:
413 	fdisp_destroy(fdip);
414 	return err;
415 }
416 
417 /*
418  * Our vnop_fsync roughly corresponds to the FUSE_FSYNC method. The Linux
419  * version of FUSE also has a FUSE_FLUSH method.
420  *
421  * On Linux, fsync() synchronizes a file's complete in-core state with that
422  * on disk. The call is not supposed to return until the system has completed
423  * that action or until an error is detected.
424  *
425  * Linux also has an fdatasync() call that is similar to fsync() but is not
426  * required to update the metadata such as access time and modification time.
427  */
428 
429 /*
430     struct vnop_fsync_args {
431 	struct vnodeop_desc *a_desc;
432 	struct vnode * a_vp;
433 	struct ucred * a_cred;
434 	int  a_waitfor;
435 	struct thread * a_td;
436     };
437 */
438 static int
439 fuse_vnop_fsync(struct vop_fsync_args *ap)
440 {
441 	struct vnode *vp = ap->a_vp;
442 	struct thread *td = ap->a_td;
443 
444 	struct fuse_filehandle *fufh;
445 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
446 
447 	int type, err = 0;
448 
449 	fuse_trace_printf_vnop();
450 
451 	if (fuse_isdeadfs(vp)) {
452 		return 0;
453 	}
454 	if ((err = vop_stdfsync(ap)))
455 		return err;
456 
457 	if (!fsess_isimpl(vnode_mount(vp),
458 	    (vnode_vtype(vp) == VDIR ? FUSE_FSYNCDIR : FUSE_FSYNC))) {
459 		goto out;
460 	}
461 	for (type = 0; type < FUFH_MAXTYPE; type++) {
462 		fufh = &(fvdat->fufh[type]);
463 		if (FUFH_IS_VALID(fufh)) {
464 			fuse_internal_fsync(vp, td, NULL, fufh);
465 		}
466 	}
467 
468 out:
469 	return 0;
470 }
471 
472 /*
473     struct vnop_getattr_args {
474 	struct vnode *a_vp;
475 	struct vattr *a_vap;
476 	struct ucred *a_cred;
477 	struct thread *a_td;
478     };
479 */
480 static int
481 fuse_vnop_getattr(struct vop_getattr_args *ap)
482 {
483 	struct vnode *vp = ap->a_vp;
484 	struct vattr *vap = ap->a_vap;
485 	struct ucred *cred = ap->a_cred;
486 	struct thread *td = curthread;
487 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
488 
489 	int err = 0;
490 	int dataflags;
491 	struct fuse_dispatcher fdi;
492 
493 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
494 
495 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
496 
497 	/* Note that we are not bailing out on a dead file system just yet. */
498 
499 	if (!(dataflags & FSESS_INITED)) {
500 		if (!vnode_isvroot(vp)) {
501 			fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
502 			err = ENOTCONN;
503 			debug_printf("fuse_getattr b: returning ENOTCONN\n");
504 			return err;
505 		} else {
506 			goto fake;
507 		}
508 	}
509 	fdisp_init(&fdi, 0);
510 	if ((err = fdisp_simple_putget_vp(&fdi, FUSE_GETATTR, vp, td, cred))) {
511 		if ((err == ENOTCONN) && vnode_isvroot(vp)) {
512 			/* see comment at similar place in fuse_statfs() */
513 			fdisp_destroy(&fdi);
514 			goto fake;
515 		}
516 		if (err == ENOENT) {
517 			fuse_internal_vnode_disappear(vp);
518 		}
519 		goto out;
520 	}
521 	cache_attrs(vp, (struct fuse_attr_out *)fdi.answ);
522 	if (vap != VTOVA(vp)) {
523 		memcpy(vap, VTOVA(vp), sizeof(*vap));
524 	}
525 	if (vap->va_type != vnode_vtype(vp)) {
526 		fuse_internal_vnode_disappear(vp);
527 		err = ENOENT;
528 		goto out;
529 	}
530 	if ((fvdat->flag & FN_SIZECHANGE) != 0)
531 		vap->va_size = fvdat->filesize;
532 
533 	if (vnode_isreg(vp) && (fvdat->flag & FN_SIZECHANGE) == 0) {
534 		/*
535 	         * This is for those cases when the file size changed without us
536 	         * knowing, and we want to catch up.
537 	         */
538 		off_t new_filesize = ((struct fuse_attr_out *)
539 				      fdi.answ)->attr.size;
540 
541 		if (fvdat->filesize != new_filesize) {
542 			fuse_vnode_setsize(vp, cred, new_filesize);
543 		}
544 	}
545 	debug_printf("fuse_getattr e: returning 0\n");
546 
547 out:
548 	fdisp_destroy(&fdi);
549 	return err;
550 
551 fake:
552 	bzero(vap, sizeof(*vap));
553 	vap->va_type = vnode_vtype(vp);
554 
555 	return 0;
556 }
557 
558 /*
559     struct vnop_inactive_args {
560 	struct vnode *a_vp;
561 	struct thread *a_td;
562     };
563 */
564 static int
565 fuse_vnop_inactive(struct vop_inactive_args *ap)
566 {
567 	struct vnode *vp = ap->a_vp;
568 	struct thread *td = ap->a_td;
569 
570 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
571 	struct fuse_filehandle *fufh = NULL;
572 
573 	int type, need_flush = 1;
574 
575 	FS_DEBUG("inode=%ju\n", (uintmax_t)VTOI(vp));
576 
577 	for (type = 0; type < FUFH_MAXTYPE; type++) {
578 		fufh = &(fvdat->fufh[type]);
579 		if (FUFH_IS_VALID(fufh)) {
580 			if (need_flush && vp->v_type == VREG) {
581 				if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
582 					fuse_vnode_savesize(vp, NULL);
583 				}
584 				if (fuse_data_cache_invalidate ||
585 				    (fvdat->flag & FN_REVOKED) != 0)
586 					fuse_io_invalbuf(vp, td);
587 				else
588 					fuse_io_flushbuf(vp, MNT_WAIT, td);
589 				need_flush = 0;
590 			}
591 			fuse_filehandle_close(vp, type, td, NULL);
592 		}
593 	}
594 
595 	if ((fvdat->flag & FN_REVOKED) != 0 && fuse_reclaim_revoked) {
596 		vrecycle(vp);
597 	}
598 	return 0;
599 }
600 
601 /*
602     struct vnop_link_args {
603 	struct vnode *a_tdvp;
604 	struct vnode *a_vp;
605 	struct componentname *a_cnp;
606     };
607 */
608 static int
609 fuse_vnop_link(struct vop_link_args *ap)
610 {
611 	struct vnode *vp = ap->a_vp;
612 	struct vnode *tdvp = ap->a_tdvp;
613 	struct componentname *cnp = ap->a_cnp;
614 
615 	struct vattr *vap = VTOVA(vp);
616 
617 	struct fuse_dispatcher fdi;
618 	struct fuse_entry_out *feo;
619 	struct fuse_link_in fli;
620 
621 	int err;
622 
623 	fuse_trace_printf_vnop();
624 
625 	if (fuse_isdeadfs(vp)) {
626 		return ENXIO;
627 	}
628 	if (vnode_mount(tdvp) != vnode_mount(vp)) {
629 		return EXDEV;
630 	}
631 	if (vap->va_nlink >= FUSE_LINK_MAX) {
632 		return EMLINK;
633 	}
634 	fli.oldnodeid = VTOI(vp);
635 
636 	fdisp_init(&fdi, 0);
637 	fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
638 	    FUSE_LINK, &fli, sizeof(fli), &fdi);
639 	if ((err = fdisp_wait_answ(&fdi))) {
640 		goto out;
641 	}
642 	feo = fdi.answ;
643 
644 	err = fuse_internal_checkentry(feo, vnode_vtype(vp));
645 out:
646 	fdisp_destroy(&fdi);
647 	return err;
648 }
649 
650 /*
651     struct vnop_lookup_args {
652 	struct vnodeop_desc *a_desc;
653 	struct vnode *a_dvp;
654 	struct vnode **a_vpp;
655 	struct componentname *a_cnp;
656     };
657 */
658 int
659 fuse_vnop_lookup(struct vop_lookup_args *ap)
660 {
661 	struct vnode *dvp = ap->a_dvp;
662 	struct vnode **vpp = ap->a_vpp;
663 	struct componentname *cnp = ap->a_cnp;
664 	struct thread *td = cnp->cn_thread;
665 	struct ucred *cred = cnp->cn_cred;
666 
667 	int nameiop = cnp->cn_nameiop;
668 	int flags = cnp->cn_flags;
669 	int wantparent = flags & (LOCKPARENT | WANTPARENT);
670 	int islastcn = flags & ISLASTCN;
671 	struct mount *mp = vnode_mount(dvp);
672 
673 	int err = 0;
674 	int lookup_err = 0;
675 	struct vnode *vp = NULL;
676 
677 	struct fuse_dispatcher fdi;
678 	enum fuse_opcode op;
679 
680 	uint64_t nid;
681 	struct fuse_access_param facp;
682 
683 	FS_DEBUG2G("parent_inode=%ju - %*s\n",
684 	    (uintmax_t)VTOI(dvp), (int)cnp->cn_namelen, cnp->cn_nameptr);
685 
686 	if (fuse_isdeadfs(dvp)) {
687 		*vpp = NULL;
688 		return ENXIO;
689 	}
690 	if (!vnode_isdir(dvp)) {
691 		return ENOTDIR;
692 	}
693 	if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP)) {
694 		return EROFS;
695 	}
696 	/*
697          * We do access check prior to doing anything else only in the case
698          * when we are at fs root (we'd like to say, "we are at the first
699          * component", but that's not exactly the same... nevermind).
700          * See further comments at further access checks.
701          */
702 
703 	bzero(&facp, sizeof(facp));
704 	if (vnode_isvroot(dvp)) {	/* early permission check hack */
705 		if ((err = fuse_internal_access(dvp, VEXEC, &facp, td, cred))) {
706 			return err;
707 		}
708 	}
709 	if (flags & ISDOTDOT) {
710 		nid = VTOFUD(dvp)->parent_nid;
711 		if (nid == 0) {
712 			return ENOENT;
713 		}
714 		fdisp_init(&fdi, 0);
715 		op = FUSE_GETATTR;
716 		goto calldaemon;
717 	} else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
718 		nid = VTOI(dvp);
719 		fdisp_init(&fdi, 0);
720 		op = FUSE_GETATTR;
721 		goto calldaemon;
722 	} else if (fuse_lookup_cache_enable) {
723 		err = cache_lookup(dvp, vpp, cnp, NULL, NULL);
724 		switch (err) {
725 
726 		case -1:		/* positive match */
727 			atomic_add_acq_long(&fuse_lookup_cache_hits, 1);
728 			return 0;
729 
730 		case 0:		/* no match in cache */
731 			atomic_add_acq_long(&fuse_lookup_cache_misses, 1);
732 			break;
733 
734 		case ENOENT:		/* negative match */
735 			/* fall through */
736 		default:
737 			return err;
738 		}
739 	}
740 	nid = VTOI(dvp);
741 	fdisp_init(&fdi, cnp->cn_namelen + 1);
742 	op = FUSE_LOOKUP;
743 
744 calldaemon:
745 	fdisp_make(&fdi, op, mp, nid, td, cred);
746 
747 	if (op == FUSE_LOOKUP) {
748 		memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
749 		((char *)fdi.indata)[cnp->cn_namelen] = '\0';
750 	}
751 	lookup_err = fdisp_wait_answ(&fdi);
752 
753 	if ((op == FUSE_LOOKUP) && !lookup_err) {	/* lookup call succeeded */
754 		nid = ((struct fuse_entry_out *)fdi.answ)->nodeid;
755 		if (!nid) {
756 			/*
757 	                 * zero nodeid is the same as "not found",
758 	                 * but it's also cacheable (which we keep
759 	                 * keep on doing not as of writing this)
760 	                 */
761 			lookup_err = ENOENT;
762 		} else if (nid == FUSE_ROOT_ID) {
763 			lookup_err = EINVAL;
764 		}
765 	}
766 	if (lookup_err &&
767 	    (!fdi.answ_stat || lookup_err != ENOENT || op != FUSE_LOOKUP)) {
768 		fdisp_destroy(&fdi);
769 		return lookup_err;
770 	}
771 	/* lookup_err, if non-zero, must be ENOENT at this point */
772 
773 	if (lookup_err) {
774 
775 		if ((nameiop == CREATE || nameiop == RENAME) && islastcn
776 		     /* && directory dvp has not been removed */ ) {
777 
778 			if (vfs_isrdonly(mp)) {
779 				err = EROFS;
780 				goto out;
781 			}
782 #if 0 /* THINK_ABOUT_THIS */
783 			if ((err = fuse_internal_access(dvp, VWRITE, cred, td, &facp))) {
784 				goto out;
785 			}
786 #endif
787 
788 			/*
789 	                 * Possibly record the position of a slot in the
790 	                 * directory large enough for the new component name.
791 	                 * This can be recorded in the vnode private data for
792 	                 * dvp. Set the SAVENAME flag to hold onto the
793 	                 * pathname for use later in VOP_CREATE or VOP_RENAME.
794 	                 */
795 			cnp->cn_flags |= SAVENAME;
796 
797 			err = EJUSTRETURN;
798 			goto out;
799 		}
800 		/* Consider inserting name into cache. */
801 
802 		/*
803 	         * No we can't use negative caching, as the fs
804 	         * changes are out of our control.
805 	         * False positives' falseness turns out just as things
806 	         * go by, but false negatives' falseness doesn't.
807 	         * (and aiding the caching mechanism with extra control
808 	         * mechanisms comes quite close to beating the whole purpose
809 	         * caching...)
810 	         */
811 #if 0
812 		if ((cnp->cn_flags & MAKEENTRY) != 0) {
813 			FS_DEBUG("inserting NULL into cache\n");
814 			cache_enter(dvp, NULL, cnp);
815 		}
816 #endif
817 		err = ENOENT;
818 		goto out;
819 
820 	} else {
821 
822 		/* !lookup_err */
823 
824 		struct fuse_entry_out *feo = NULL;
825 		struct fuse_attr *fattr = NULL;
826 
827 		if (op == FUSE_GETATTR) {
828 			fattr = &((struct fuse_attr_out *)fdi.answ)->attr;
829 		} else {
830 			feo = (struct fuse_entry_out *)fdi.answ;
831 			fattr = &(feo->attr);
832 		}
833 
834 		/*
835 	         * If deleting, and at end of pathname, return parameters
836 	         * which can be used to remove file.  If the wantparent flag
837 	         * isn't set, we return only the directory, otherwise we go on
838 	         * and lock the inode, being careful with ".".
839 	         */
840 		if (nameiop == DELETE && islastcn) {
841 			/*
842 	                 * Check for write access on directory.
843 	                 */
844 			facp.xuid = fattr->uid;
845 			facp.facc_flags |= FACCESS_STICKY;
846 			err = fuse_internal_access(dvp, VWRITE, &facp, td, cred);
847 			facp.facc_flags &= ~FACCESS_XQUERIES;
848 
849 			if (err) {
850 				goto out;
851 			}
852 			if (nid == VTOI(dvp)) {
853 				vref(dvp);
854 				*vpp = dvp;
855 			} else {
856 				err = fuse_vnode_get(dvp->v_mount, nid, dvp,
857 				    &vp, cnp, IFTOVT(fattr->mode));
858 				if (err)
859 					goto out;
860 				*vpp = vp;
861 			}
862 
863 			/*
864 			 * Save the name for use in VOP_RMDIR and VOP_REMOVE
865 			 * later.
866 			 */
867 			cnp->cn_flags |= SAVENAME;
868 			goto out;
869 
870 		}
871 		/*
872 	         * If rewriting (RENAME), return the inode and the
873 	         * information required to rewrite the present directory
874 	         * Must get inode of directory entry to verify it's a
875 	         * regular file, or empty directory.
876 	         */
877 		if (nameiop == RENAME && wantparent && islastcn) {
878 
879 #if 0 /* THINK_ABOUT_THIS */
880 			if ((err = fuse_internal_access(dvp, VWRITE, cred, td, &facp))) {
881 				goto out;
882 			}
883 #endif
884 
885 			/*
886 	                 * Check for "."
887 	                 */
888 			if (nid == VTOI(dvp)) {
889 				err = EISDIR;
890 				goto out;
891 			}
892 			err = fuse_vnode_get(vnode_mount(dvp),
893 			    nid,
894 			    dvp,
895 			    &vp,
896 			    cnp,
897 			    IFTOVT(fattr->mode));
898 			if (err) {
899 				goto out;
900 			}
901 			*vpp = vp;
902 			/*
903 	                 * Save the name for use in VOP_RENAME later.
904 	                 */
905 			cnp->cn_flags |= SAVENAME;
906 
907 			goto out;
908 		}
909 		if (flags & ISDOTDOT) {
910 			struct mount *mp;
911 			int ltype;
912 
913 			/*
914 			 * Expanded copy of vn_vget_ino() so that
915 			 * fuse_vnode_get() can be used.
916 			 */
917 			mp = dvp->v_mount;
918 			ltype = VOP_ISLOCKED(dvp);
919 			err = vfs_busy(mp, MBF_NOWAIT);
920 			if (err != 0) {
921 				vfs_ref(mp);
922 				VOP_UNLOCK(dvp, 0);
923 				err = vfs_busy(mp, 0);
924 				vn_lock(dvp, ltype | LK_RETRY);
925 				vfs_rel(mp);
926 				if (err)
927 					goto out;
928 				if ((dvp->v_iflag & VI_DOOMED) != 0) {
929 					err = ENOENT;
930 					vfs_unbusy(mp);
931 					goto out;
932 				}
933 			}
934 			VOP_UNLOCK(dvp, 0);
935 			err = fuse_vnode_get(vnode_mount(dvp),
936 			    nid,
937 			    NULL,
938 			    &vp,
939 			    cnp,
940 			    IFTOVT(fattr->mode));
941 			vfs_unbusy(mp);
942 			vn_lock(dvp, ltype | LK_RETRY);
943 			if ((dvp->v_iflag & VI_DOOMED) != 0) {
944 				if (err == 0)
945 					vput(vp);
946 				err = ENOENT;
947 			}
948 			if (err)
949 				goto out;
950 			*vpp = vp;
951 		} else if (nid == VTOI(dvp)) {
952 			vref(dvp);
953 			*vpp = dvp;
954 		} else {
955 			err = fuse_vnode_get(vnode_mount(dvp),
956 			    nid,
957 			    dvp,
958 			    &vp,
959 			    cnp,
960 			    IFTOVT(fattr->mode));
961 			if (err) {
962 				goto out;
963 			}
964 			fuse_vnode_setparent(vp, dvp);
965 			*vpp = vp;
966 		}
967 
968 		if (op == FUSE_GETATTR) {
969 			cache_attrs(*vpp, (struct fuse_attr_out *)fdi.answ);
970 		} else {
971 			cache_attrs(*vpp, (struct fuse_entry_out *)fdi.answ);
972 		}
973 
974 		/* Insert name into cache if appropriate. */
975 
976 		/*
977 	         * Nooo, caching is evil. With caching, we can't avoid stale
978 	         * information taking over the playground (cached info is not
979 	         * just positive/negative, it does have qualitative aspects,
980 	         * too). And a (VOP/FUSE)_GETATTR is always thrown anyway, when
981 	         * walking down along cached path components, and that's not
982 	         * any cheaper than FUSE_LOOKUP. This might change with
983 	         * implementing kernel side attr caching, but... In Linux,
984 	         * lookup results are not cached, and the daemon is bombarded
985 	         * with FUSE_LOOKUPS on and on. This shows that by design, the
986 	         * daemon is expected to handle frequent lookup queries
987 	         * efficiently, do its caching in userspace, and so on.
988 	         *
989 	         * So just leave the name cache alone.
990 	         */
991 
992 		/*
993 	         * Well, now I know, Linux caches lookups, but with a
994 	         * timeout... So it's the same thing as attribute caching:
995 	         * we can deal with it when implement timeouts.
996 	         */
997 #if 0
998 		if (cnp->cn_flags & MAKEENTRY) {
999 			cache_enter(dvp, *vpp, cnp);
1000 		}
1001 #endif
1002 	}
1003 out:
1004 	if (!lookup_err) {
1005 
1006 		/* No lookup error; need to clean up. */
1007 
1008 		if (err) {		/* Found inode; exit with no vnode. */
1009 			if (op == FUSE_LOOKUP) {
1010 				fuse_internal_forget_send(vnode_mount(dvp), td, cred,
1011 				    nid, 1);
1012 			}
1013 			fdisp_destroy(&fdi);
1014 			return err;
1015 		} else {
1016 #ifndef NO_EARLY_PERM_CHECK_HACK
1017 			if (!islastcn) {
1018 				/*
1019 				 * We have the attributes of the next item
1020 				 * *now*, and it's a fact, and we do not
1021 				 * have to do extra work for it (ie, beg the
1022 				 * daemon), and it neither depends on such
1023 				 * accidental things like attr caching. So
1024 				 * the big idea: check credentials *now*,
1025 				 * not at the beginning of the next call to
1026 				 * lookup.
1027 				 *
1028 				 * The first item of the lookup chain (fs root)
1029 				 * won't be checked then here, of course, as
1030 				 * its never "the next". But go and see that
1031 				 * the root is taken care about at the very
1032 				 * beginning of this function.
1033 				 *
1034 				 * Now, given we want to do the access check
1035 				 * this way, one might ask: so then why not
1036 				 * do the access check just after fetching
1037 				 * the inode and its attributes from the
1038 				 * daemon? Why bother with producing the
1039 				 * corresponding vnode at all if something
1040 				 * is not OK? We know what's the deal as
1041 				 * soon as we get those attrs... There is
1042 				 * one bit of info though not given us by
1043 				 * the daemon: whether his response is
1044 				 * authoritative or not... His response should
1045 				 * be ignored if something is mounted over
1046 				 * the dir in question. But that can be
1047 				 * known only by having the vnode...
1048 				 */
1049 				int tmpvtype = vnode_vtype(*vpp);
1050 
1051 				bzero(&facp, sizeof(facp));
1052 				/*the early perm check hack */
1053 				    facp.facc_flags |= FACCESS_VA_VALID;
1054 
1055 				if ((tmpvtype != VDIR) && (tmpvtype != VLNK)) {
1056 					err = ENOTDIR;
1057 				}
1058 				if (!err && !vnode_mountedhere(*vpp)) {
1059 					err = fuse_internal_access(*vpp, VEXEC, &facp, td, cred);
1060 				}
1061 				if (err) {
1062 					if (tmpvtype == VLNK)
1063 						FS_DEBUG("weird, permission error with a symlink?\n");
1064 					vput(*vpp);
1065 					*vpp = NULL;
1066 				}
1067 			}
1068 #endif
1069 		}
1070 	}
1071 	fdisp_destroy(&fdi);
1072 
1073 	return err;
1074 }
1075 
1076 /*
1077     struct vnop_mkdir_args {
1078 	struct vnode *a_dvp;
1079 	struct vnode **a_vpp;
1080 	struct componentname *a_cnp;
1081 	struct vattr *a_vap;
1082     };
1083 */
1084 static int
1085 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1086 {
1087 	struct vnode *dvp = ap->a_dvp;
1088 	struct vnode **vpp = ap->a_vpp;
1089 	struct componentname *cnp = ap->a_cnp;
1090 	struct vattr *vap = ap->a_vap;
1091 
1092 	struct fuse_mkdir_in fmdi;
1093 
1094 	fuse_trace_printf_vnop();
1095 
1096 	if (fuse_isdeadfs(dvp)) {
1097 		return ENXIO;
1098 	}
1099 	fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1100 
1101 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1102 	    sizeof(fmdi), VDIR));
1103 }
1104 
1105 /*
1106     struct vnop_mknod_args {
1107 	struct vnode *a_dvp;
1108 	struct vnode **a_vpp;
1109 	struct componentname *a_cnp;
1110 	struct vattr *a_vap;
1111     };
1112 */
1113 static int
1114 fuse_vnop_mknod(struct vop_mknod_args *ap)
1115 {
1116 
1117 	return (EINVAL);
1118 }
1119 
1120 
1121 /*
1122     struct vnop_open_args {
1123 	struct vnode *a_vp;
1124 	int  a_mode;
1125 	struct ucred *a_cred;
1126 	struct thread *a_td;
1127 	int a_fdidx; / struct file *a_fp;
1128     };
1129 */
1130 static int
1131 fuse_vnop_open(struct vop_open_args *ap)
1132 {
1133 	struct vnode *vp = ap->a_vp;
1134 	int mode = ap->a_mode;
1135 	struct thread *td = ap->a_td;
1136 	struct ucred *cred = ap->a_cred;
1137 
1138 	fufh_type_t fufh_type;
1139 	struct fuse_vnode_data *fvdat;
1140 
1141 	int error, isdir = 0;
1142 	int32_t fuse_open_flags;
1143 
1144 	FS_DEBUG2G("inode=%ju mode=0x%x\n", (uintmax_t)VTOI(vp), mode);
1145 
1146 	if (fuse_isdeadfs(vp)) {
1147 		return ENXIO;
1148 	}
1149 	fvdat = VTOFUD(vp);
1150 
1151 	if (vnode_isdir(vp)) {
1152 		isdir = 1;
1153 	}
1154 	fuse_open_flags = 0;
1155 	if (isdir) {
1156 		fufh_type = FUFH_RDONLY;
1157 	} else {
1158 		fufh_type = fuse_filehandle_xlate_from_fflags(mode);
1159 		/*
1160 		 * For WRONLY opens, force DIRECT_IO.  This is necessary
1161 		 * since writing a partial block through the buffer cache
1162 		 * will result in a read of the block and that read won't
1163 		 * be allowed by the WRONLY open.
1164 		 */
1165 		if (fufh_type == FUFH_WRONLY ||
1166 		    (fvdat->flag & FN_DIRECTIO) != 0)
1167 			fuse_open_flags = FOPEN_DIRECT_IO;
1168 	}
1169 
1170 	if (fuse_filehandle_validrw(vp, fufh_type) != FUFH_INVALID) {
1171 		fuse_vnode_open(vp, fuse_open_flags, td);
1172 		return 0;
1173 	}
1174 	error = fuse_filehandle_open(vp, fufh_type, NULL, td, cred);
1175 
1176 	return error;
1177 }
1178 
1179 static int
1180 fuse_vnop_pathconf(struct vop_pathconf_args *ap)
1181 {
1182 
1183 	switch (ap->a_name) {
1184 	case _PC_FILESIZEBITS:
1185 		*ap->a_retval = 64;
1186 		return (0);
1187 	case _PC_NAME_MAX:
1188 		*ap->a_retval = NAME_MAX;
1189 		return (0);
1190 	case _PC_LINK_MAX:
1191 		*ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX);
1192 		return (0);
1193 	case _PC_SYMLINK_MAX:
1194 		*ap->a_retval = MAXPATHLEN;
1195 		return (0);
1196 	case _PC_NO_TRUNC:
1197 		*ap->a_retval = 1;
1198 		return (0);
1199 	default:
1200 		return (vop_stdpathconf(ap));
1201 	}
1202 }
1203 
1204 /*
1205     struct vnop_read_args {
1206 	struct vnode *a_vp;
1207 	struct uio *a_uio;
1208 	int  a_ioflag;
1209 	struct ucred *a_cred;
1210     };
1211 */
1212 static int
1213 fuse_vnop_read(struct vop_read_args *ap)
1214 {
1215 	struct vnode *vp = ap->a_vp;
1216 	struct uio *uio = ap->a_uio;
1217 	int ioflag = ap->a_ioflag;
1218 	struct ucred *cred = ap->a_cred;
1219 
1220 	FS_DEBUG2G("inode=%ju offset=%jd resid=%zd\n",
1221 	    (uintmax_t)VTOI(vp), uio->uio_offset, uio->uio_resid);
1222 
1223 	if (fuse_isdeadfs(vp)) {
1224 		return ENXIO;
1225 	}
1226 
1227 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1228 		ioflag |= IO_DIRECT;
1229 	}
1230 
1231 	return fuse_io_dispatch(vp, uio, ioflag, cred);
1232 }
1233 
1234 /*
1235     struct vnop_readdir_args {
1236 	struct vnode *a_vp;
1237 	struct uio *a_uio;
1238 	struct ucred *a_cred;
1239 	int *a_eofflag;
1240 	int *ncookies;
1241 	u_long **a_cookies;
1242     };
1243 */
1244 static int
1245 fuse_vnop_readdir(struct vop_readdir_args *ap)
1246 {
1247 	struct vnode *vp = ap->a_vp;
1248 	struct uio *uio = ap->a_uio;
1249 	struct ucred *cred = ap->a_cred;
1250 
1251 	struct fuse_filehandle *fufh = NULL;
1252 	struct fuse_iov cookediov;
1253 
1254 	int err = 0;
1255 	int freefufh = 0;
1256 
1257 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
1258 
1259 	if (fuse_isdeadfs(vp)) {
1260 		return ENXIO;
1261 	}
1262 	if (				/* XXXIP ((uio_iovcnt(uio) > 1)) || */
1263 	    (uio_resid(uio) < sizeof(struct dirent))) {
1264 		return EINVAL;
1265 	}
1266 
1267 	if (!fuse_filehandle_valid(vp, FUFH_RDONLY)) {
1268 		FS_DEBUG("calling readdir() before open()");
1269 		err = fuse_filehandle_open(vp, FUFH_RDONLY, &fufh, NULL, cred);
1270 		freefufh = 1;
1271 	} else {
1272 		err = fuse_filehandle_get(vp, FUFH_RDONLY, &fufh);
1273 	}
1274 	if (err) {
1275 		return (err);
1276 	}
1277 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1278 	fiov_init(&cookediov, DIRCOOKEDSIZE);
1279 
1280 	err = fuse_internal_readdir(vp, uio, fufh, &cookediov);
1281 
1282 	fiov_teardown(&cookediov);
1283 	if (freefufh) {
1284 		fuse_filehandle_close(vp, FUFH_RDONLY, NULL, cred);
1285 	}
1286 	return err;
1287 }
1288 
1289 /*
1290     struct vnop_readlink_args {
1291 	struct vnode *a_vp;
1292 	struct uio *a_uio;
1293 	struct ucred *a_cred;
1294     };
1295 */
1296 static int
1297 fuse_vnop_readlink(struct vop_readlink_args *ap)
1298 {
1299 	struct vnode *vp = ap->a_vp;
1300 	struct uio *uio = ap->a_uio;
1301 	struct ucred *cred = ap->a_cred;
1302 
1303 	struct fuse_dispatcher fdi;
1304 	int err;
1305 
1306 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
1307 
1308 	if (fuse_isdeadfs(vp)) {
1309 		return ENXIO;
1310 	}
1311 	if (!vnode_islnk(vp)) {
1312 		return EINVAL;
1313 	}
1314 	fdisp_init(&fdi, 0);
1315 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1316 	if (err) {
1317 		goto out;
1318 	}
1319 	if (((char *)fdi.answ)[0] == '/' &&
1320 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1321 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1322 
1323 		err = uiomove(mpth, strlen(mpth), uio);
1324 	}
1325 	if (!err) {
1326 		err = uiomove(fdi.answ, fdi.iosize, uio);
1327 	}
1328 out:
1329 	fdisp_destroy(&fdi);
1330 	return err;
1331 }
1332 
1333 /*
1334     struct vnop_reclaim_args {
1335 	struct vnode *a_vp;
1336 	struct thread *a_td;
1337     };
1338 */
1339 static int
1340 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1341 {
1342 	struct vnode *vp = ap->a_vp;
1343 	struct thread *td = ap->a_td;
1344 
1345 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1346 	struct fuse_filehandle *fufh = NULL;
1347 
1348 	int type;
1349 
1350 	if (!fvdat) {
1351 		panic("FUSE: no vnode data during recycling");
1352 	}
1353 	FS_DEBUG("inode=%ju\n", (uintmax_t)VTOI(vp));
1354 
1355 	for (type = 0; type < FUFH_MAXTYPE; type++) {
1356 		fufh = &(fvdat->fufh[type]);
1357 		if (FUFH_IS_VALID(fufh)) {
1358 			printf("FUSE: vnode being reclaimed but fufh (type=%d) is valid",
1359 			    type);
1360 			fuse_filehandle_close(vp, type, td, NULL);
1361 		}
1362 	}
1363 
1364 	if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) {
1365 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1366 		    fvdat->nlookup);
1367 	}
1368 	fuse_vnode_setparent(vp, NULL);
1369 	cache_purge(vp);
1370 	vfs_hash_remove(vp);
1371 	vnode_destroy_vobject(vp);
1372 	fuse_vnode_destroy(vp);
1373 
1374 	return 0;
1375 }
1376 
1377 /*
1378     struct vnop_remove_args {
1379 	struct vnode *a_dvp;
1380 	struct vnode *a_vp;
1381 	struct componentname *a_cnp;
1382     };
1383 */
1384 static int
1385 fuse_vnop_remove(struct vop_remove_args *ap)
1386 {
1387 	struct vnode *dvp = ap->a_dvp;
1388 	struct vnode *vp = ap->a_vp;
1389 	struct componentname *cnp = ap->a_cnp;
1390 
1391 	int err;
1392 
1393 	FS_DEBUG2G("inode=%ju name=%*s\n",
1394 	    (uintmax_t)VTOI(vp), (int)cnp->cn_namelen, cnp->cn_nameptr);
1395 
1396 	if (fuse_isdeadfs(vp)) {
1397 		return ENXIO;
1398 	}
1399 	if (vnode_isdir(vp)) {
1400 		return EPERM;
1401 	}
1402 	cache_purge(vp);
1403 
1404 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1405 
1406 	if (err == 0)
1407 		fuse_internal_vnode_disappear(vp);
1408 	return err;
1409 }
1410 
1411 /*
1412     struct vnop_rename_args {
1413 	struct vnode *a_fdvp;
1414 	struct vnode *a_fvp;
1415 	struct componentname *a_fcnp;
1416 	struct vnode *a_tdvp;
1417 	struct vnode *a_tvp;
1418 	struct componentname *a_tcnp;
1419     };
1420 */
1421 static int
1422 fuse_vnop_rename(struct vop_rename_args *ap)
1423 {
1424 	struct vnode *fdvp = ap->a_fdvp;
1425 	struct vnode *fvp = ap->a_fvp;
1426 	struct componentname *fcnp = ap->a_fcnp;
1427 	struct vnode *tdvp = ap->a_tdvp;
1428 	struct vnode *tvp = ap->a_tvp;
1429 	struct componentname *tcnp = ap->a_tcnp;
1430 	struct fuse_data *data;
1431 
1432 	int err = 0;
1433 
1434 	FS_DEBUG2G("from: inode=%ju name=%*s -> to: inode=%ju name=%*s\n",
1435 	    (uintmax_t)VTOI(fvp), (int)fcnp->cn_namelen, fcnp->cn_nameptr,
1436 	    (uintmax_t)(tvp == NULL ? -1 : VTOI(tvp)),
1437 	    (int)tcnp->cn_namelen, tcnp->cn_nameptr);
1438 
1439 	if (fuse_isdeadfs(fdvp)) {
1440 		return ENXIO;
1441 	}
1442 	if (fvp->v_mount != tdvp->v_mount ||
1443 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1444 		FS_DEBUG("cross-device rename: %s -> %s\n",
1445 		    fcnp->cn_nameptr, (tcnp != NULL ? tcnp->cn_nameptr : "(NULL)"));
1446 		err = EXDEV;
1447 		goto out;
1448 	}
1449 	cache_purge(fvp);
1450 
1451 	/*
1452          * FUSE library is expected to check if target directory is not
1453          * under the source directory in the file system tree.
1454          * Linux performs this check at VFS level.
1455          */
1456 	data = fuse_get_mpdata(vnode_mount(tdvp));
1457 	sx_xlock(&data->rename_lock);
1458 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1459 	if (err == 0) {
1460 		if (tdvp != fdvp)
1461 			fuse_vnode_setparent(fvp, tdvp);
1462 		if (tvp != NULL)
1463 			fuse_vnode_setparent(tvp, NULL);
1464 	}
1465 	sx_unlock(&data->rename_lock);
1466 
1467 	if (tvp != NULL && tvp != fvp) {
1468 		cache_purge(tvp);
1469 	}
1470 	if (vnode_isdir(fvp)) {
1471 		if ((tvp != NULL) && vnode_isdir(tvp)) {
1472 			cache_purge(tdvp);
1473 		}
1474 		cache_purge(fdvp);
1475 	}
1476 out:
1477 	if (tdvp == tvp) {
1478 		vrele(tdvp);
1479 	} else {
1480 		vput(tdvp);
1481 	}
1482 	if (tvp != NULL) {
1483 		vput(tvp);
1484 	}
1485 	vrele(fdvp);
1486 	vrele(fvp);
1487 
1488 	return err;
1489 }
1490 
1491 /*
1492     struct vnop_rmdir_args {
1493 	    struct vnode *a_dvp;
1494 	    struct vnode *a_vp;
1495 	    struct componentname *a_cnp;
1496     } *ap;
1497 */
1498 static int
1499 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1500 {
1501 	struct vnode *dvp = ap->a_dvp;
1502 	struct vnode *vp = ap->a_vp;
1503 
1504 	int err;
1505 
1506 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
1507 
1508 	if (fuse_isdeadfs(vp)) {
1509 		return ENXIO;
1510 	}
1511 	if (VTOFUD(vp) == VTOFUD(dvp)) {
1512 		return EINVAL;
1513 	}
1514 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1515 
1516 	if (err == 0)
1517 		fuse_internal_vnode_disappear(vp);
1518 	return err;
1519 }
1520 
1521 /*
1522     struct vnop_setattr_args {
1523 	struct vnode *a_vp;
1524 	struct vattr *a_vap;
1525 	struct ucred *a_cred;
1526 	struct thread *a_td;
1527     };
1528 */
1529 static int
1530 fuse_vnop_setattr(struct vop_setattr_args *ap)
1531 {
1532 	struct vnode *vp = ap->a_vp;
1533 	struct vattr *vap = ap->a_vap;
1534 	struct ucred *cred = ap->a_cred;
1535 	struct thread *td = curthread;
1536 
1537 	struct fuse_dispatcher fdi;
1538 	struct fuse_setattr_in *fsai;
1539 	struct fuse_access_param facp;
1540 
1541 	int err = 0;
1542 	enum vtype vtyp;
1543 	int sizechanged = 0;
1544 	uint64_t newsize = 0;
1545 
1546 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
1547 
1548 	if (fuse_isdeadfs(vp)) {
1549 		return ENXIO;
1550 	}
1551 	fdisp_init(&fdi, sizeof(*fsai));
1552 	fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
1553 	fsai = fdi.indata;
1554 	fsai->valid = 0;
1555 
1556 	bzero(&facp, sizeof(facp));
1557 
1558 	facp.xuid = vap->va_uid;
1559 	facp.xgid = vap->va_gid;
1560 
1561 	if (vap->va_uid != (uid_t)VNOVAL) {
1562 		facp.facc_flags |= FACCESS_CHOWN;
1563 		fsai->uid = vap->va_uid;
1564 		fsai->valid |= FATTR_UID;
1565 	}
1566 	if (vap->va_gid != (gid_t)VNOVAL) {
1567 		facp.facc_flags |= FACCESS_CHOWN;
1568 		fsai->gid = vap->va_gid;
1569 		fsai->valid |= FATTR_GID;
1570 	}
1571 	if (vap->va_size != VNOVAL) {
1572 
1573 		struct fuse_filehandle *fufh = NULL;
1574 
1575 		/*Truncate to a new value. */
1576 		    fsai->size = vap->va_size;
1577 		sizechanged = 1;
1578 		newsize = vap->va_size;
1579 		fsai->valid |= FATTR_SIZE;
1580 
1581 		fuse_filehandle_getrw(vp, FUFH_WRONLY, &fufh);
1582 		if (fufh) {
1583 			fsai->fh = fufh->fh_id;
1584 			fsai->valid |= FATTR_FH;
1585 		}
1586 	}
1587 	if (vap->va_atime.tv_sec != VNOVAL) {
1588 		fsai->atime = vap->va_atime.tv_sec;
1589 		fsai->atimensec = vap->va_atime.tv_nsec;
1590 		fsai->valid |= FATTR_ATIME;
1591 	}
1592 	if (vap->va_mtime.tv_sec != VNOVAL) {
1593 		fsai->mtime = vap->va_mtime.tv_sec;
1594 		fsai->mtimensec = vap->va_mtime.tv_nsec;
1595 		fsai->valid |= FATTR_MTIME;
1596 	}
1597 	if (vap->va_mode != (mode_t)VNOVAL) {
1598 		fsai->mode = vap->va_mode & ALLPERMS;
1599 		fsai->valid |= FATTR_MODE;
1600 	}
1601 	if (!fsai->valid) {
1602 		goto out;
1603 	}
1604 	vtyp = vnode_vtype(vp);
1605 
1606 	if (fsai->valid & FATTR_SIZE && vtyp == VDIR) {
1607 		err = EISDIR;
1608 		goto out;
1609 	}
1610 	if (vfs_isrdonly(vnode_mount(vp)) && (fsai->valid & ~FATTR_SIZE || vtyp == VREG)) {
1611 		err = EROFS;
1612 		goto out;
1613 	}
1614 	if (fsai->valid & ~FATTR_SIZE) {
1615 	  /*err = fuse_internal_access(vp, VADMIN, context, &facp); */
1616 	  /*XXX */
1617 		    err = 0;
1618 	}
1619 	facp.facc_flags &= ~FACCESS_XQUERIES;
1620 
1621 	if (err && !(fsai->valid & ~(FATTR_ATIME | FATTR_MTIME)) &&
1622 	    vap->va_vaflags & VA_UTIMES_NULL) {
1623 		err = fuse_internal_access(vp, VWRITE, &facp, td, cred);
1624 	}
1625 	if (err)
1626 		goto out;
1627 	if ((err = fdisp_wait_answ(&fdi)))
1628 		goto out;
1629 	vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode);
1630 
1631 	if (vnode_vtype(vp) != vtyp) {
1632 		if (vnode_vtype(vp) == VNON && vtyp != VNON) {
1633 			debug_printf("FUSE: Dang! vnode_vtype is VNON and vtype isn't.\n");
1634 		} else {
1635 			/*
1636 	                 * STALE vnode, ditch
1637 	                 *
1638 	                 * The vnode has changed its type "behind our back". There's
1639 	                 * nothing really we can do, so let us just force an internal
1640 	                 * revocation and tell the caller to try again, if interested.
1641 	                 */
1642 			fuse_internal_vnode_disappear(vp);
1643 			err = EAGAIN;
1644 		}
1645 	}
1646 	if (!err && !sizechanged) {
1647 		cache_attrs(vp, (struct fuse_attr_out *)fdi.answ);
1648 	}
1649 out:
1650 	fdisp_destroy(&fdi);
1651 	if (!err && sizechanged) {
1652 		fuse_vnode_setsize(vp, cred, newsize);
1653 		VTOFUD(vp)->flag &= ~FN_SIZECHANGE;
1654 	}
1655 	return err;
1656 }
1657 
1658 /*
1659     struct vnop_strategy_args {
1660 	struct vnode *a_vp;
1661 	struct buf *a_bp;
1662     };
1663 */
1664 static int
1665 fuse_vnop_strategy(struct vop_strategy_args *ap)
1666 {
1667 	struct vnode *vp = ap->a_vp;
1668 	struct buf *bp = ap->a_bp;
1669 
1670 	fuse_trace_printf_vnop();
1671 
1672 	if (!vp || fuse_isdeadfs(vp)) {
1673 		bp->b_ioflags |= BIO_ERROR;
1674 		bp->b_error = ENXIO;
1675 		bufdone(bp);
1676 		return ENXIO;
1677 	}
1678 	if (bp->b_iocmd == BIO_WRITE)
1679 		fuse_vnode_refreshsize(vp, NOCRED);
1680 
1681 	(void)fuse_io_strategy(vp, bp);
1682 
1683 	/*
1684 	 * This is a dangerous function. If returns error, that might mean a
1685 	 * panic. We prefer pretty much anything over being forced to panic
1686 	 * by a malicious daemon (a demon?). So we just return 0 anyway. You
1687 	 * should never mind this: this function has its own error
1688 	 * propagation mechanism via the argument buffer, so
1689 	 * not-that-melodramatic residents of the call chain still will be
1690 	 * able to know what to do.
1691 	 */
1692 	return 0;
1693 }
1694 
1695 
1696 /*
1697     struct vnop_symlink_args {
1698 	struct vnode *a_dvp;
1699 	struct vnode **a_vpp;
1700 	struct componentname *a_cnp;
1701 	struct vattr *a_vap;
1702 	char *a_target;
1703     };
1704 */
1705 static int
1706 fuse_vnop_symlink(struct vop_symlink_args *ap)
1707 {
1708 	struct vnode *dvp = ap->a_dvp;
1709 	struct vnode **vpp = ap->a_vpp;
1710 	struct componentname *cnp = ap->a_cnp;
1711 	char *target = ap->a_target;
1712 
1713 	struct fuse_dispatcher fdi;
1714 
1715 	int err;
1716 	size_t len;
1717 
1718 	FS_DEBUG2G("inode=%ju name=%*s\n",
1719 	    (uintmax_t)VTOI(dvp), (int)cnp->cn_namelen, cnp->cn_nameptr);
1720 
1721 	if (fuse_isdeadfs(dvp)) {
1722 		return ENXIO;
1723 	}
1724 	/*
1725          * Unlike the other creator type calls, here we have to create a message
1726          * where the name of the new entry comes first, and the data describing
1727          * the entry comes second.
1728          * Hence we can't rely on our handy fuse_internal_newentry() routine,
1729          * but put together the message manually and just call the core part.
1730          */
1731 
1732 	len = strlen(target) + 1;
1733 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1734 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1735 
1736 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1737 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1738 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1739 
1740 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1741 	fdisp_destroy(&fdi);
1742 	return err;
1743 }
1744 
1745 /*
1746     struct vnop_write_args {
1747 	struct vnode *a_vp;
1748 	struct uio *a_uio;
1749 	int  a_ioflag;
1750 	struct ucred *a_cred;
1751     };
1752 */
1753 static int
1754 fuse_vnop_write(struct vop_write_args *ap)
1755 {
1756 	struct vnode *vp = ap->a_vp;
1757 	struct uio *uio = ap->a_uio;
1758 	int ioflag = ap->a_ioflag;
1759 	struct ucred *cred = ap->a_cred;
1760 
1761 	fuse_trace_printf_vnop();
1762 
1763 	if (fuse_isdeadfs(vp)) {
1764 		return ENXIO;
1765 	}
1766 	fuse_vnode_refreshsize(vp, cred);
1767 
1768 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1769 		ioflag |= IO_DIRECT;
1770 	}
1771 
1772 	return fuse_io_dispatch(vp, uio, ioflag, cred);
1773 }
1774 
1775 /*
1776     struct vnop_getpages_args {
1777         struct vnode *a_vp;
1778         vm_page_t *a_m;
1779         int a_count;
1780         int a_reqpage;
1781     };
1782 */
1783 static int
1784 fuse_vnop_getpages(struct vop_getpages_args *ap)
1785 {
1786 	int i, error, nextoff, size, toff, count, npages;
1787 	struct uio uio;
1788 	struct iovec iov;
1789 	vm_offset_t kva;
1790 	struct buf *bp;
1791 	struct vnode *vp;
1792 	struct thread *td;
1793 	struct ucred *cred;
1794 	vm_page_t *pages;
1795 
1796 	FS_DEBUG2G("heh\n");
1797 
1798 	vp = ap->a_vp;
1799 	KASSERT(vp->v_object, ("objectless vp passed to getpages"));
1800 	td = curthread;			/* XXX */
1801 	cred = curthread->td_ucred;	/* XXX */
1802 	pages = ap->a_m;
1803 	npages = ap->a_count;
1804 
1805 	if (!fsess_opt_mmap(vnode_mount(vp))) {
1806 		FS_DEBUG("called on non-cacheable vnode??\n");
1807 		return (VM_PAGER_ERROR);
1808 	}
1809 
1810 	/*
1811 	 * If the last page is partially valid, just return it and allow
1812 	 * the pager to zero-out the blanks.  Partially valid pages can
1813 	 * only occur at the file EOF.
1814 	 *
1815 	 * XXXGL: is that true for FUSE, which is a local filesystem,
1816 	 * but still somewhat disconnected from the kernel?
1817 	 */
1818 	VM_OBJECT_WLOCK(vp->v_object);
1819 	if (pages[npages - 1]->valid != 0 && --npages == 0)
1820 		goto out;
1821 	VM_OBJECT_WUNLOCK(vp->v_object);
1822 
1823 	/*
1824 	 * We use only the kva address for the buffer, but this is extremely
1825 	 * convenient and fast.
1826 	 */
1827 	bp = getpbuf(&fuse_pbuf_freecnt);
1828 
1829 	kva = (vm_offset_t)bp->b_data;
1830 	pmap_qenter(kva, pages, npages);
1831 	VM_CNT_INC(v_vnodein);
1832 	VM_CNT_ADD(v_vnodepgsin, npages);
1833 
1834 	count = npages << PAGE_SHIFT;
1835 	iov.iov_base = (caddr_t)kva;
1836 	iov.iov_len = count;
1837 	uio.uio_iov = &iov;
1838 	uio.uio_iovcnt = 1;
1839 	uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
1840 	uio.uio_resid = count;
1841 	uio.uio_segflg = UIO_SYSSPACE;
1842 	uio.uio_rw = UIO_READ;
1843 	uio.uio_td = td;
1844 
1845 	error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred);
1846 	pmap_qremove(kva, npages);
1847 
1848 	relpbuf(bp, &fuse_pbuf_freecnt);
1849 
1850 	if (error && (uio.uio_resid == count)) {
1851 		FS_DEBUG("error %d\n", error);
1852 		return VM_PAGER_ERROR;
1853 	}
1854 	/*
1855 	 * Calculate the number of bytes read and validate only that number
1856 	 * of bytes.  Note that due to pending writes, size may be 0.  This
1857 	 * does not mean that the remaining data is invalid!
1858 	 */
1859 
1860 	size = count - uio.uio_resid;
1861 	VM_OBJECT_WLOCK(vp->v_object);
1862 	fuse_vm_page_lock_queues();
1863 	for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
1864 		vm_page_t m;
1865 
1866 		nextoff = toff + PAGE_SIZE;
1867 		m = pages[i];
1868 
1869 		if (nextoff <= size) {
1870 			/*
1871 			 * Read operation filled an entire page
1872 			 */
1873 			m->valid = VM_PAGE_BITS_ALL;
1874 			KASSERT(m->dirty == 0,
1875 			    ("fuse_getpages: page %p is dirty", m));
1876 		} else if (size > toff) {
1877 			/*
1878 			 * Read operation filled a partial page.
1879 			 */
1880 			m->valid = 0;
1881 			vm_page_set_valid_range(m, 0, size - toff);
1882 			KASSERT(m->dirty == 0,
1883 			    ("fuse_getpages: page %p is dirty", m));
1884 		} else {
1885 			/*
1886 			 * Read operation was short.  If no error occurred
1887 			 * we may have hit a zero-fill section.   We simply
1888 			 * leave valid set to 0.
1889 			 */
1890 			;
1891 		}
1892 	}
1893 	fuse_vm_page_unlock_queues();
1894 out:
1895 	VM_OBJECT_WUNLOCK(vp->v_object);
1896 	if (ap->a_rbehind)
1897 		*ap->a_rbehind = 0;
1898 	if (ap->a_rahead)
1899 		*ap->a_rahead = 0;
1900 	return (VM_PAGER_OK);
1901 }
1902 
1903 /*
1904     struct vnop_putpages_args {
1905         struct vnode *a_vp;
1906         vm_page_t *a_m;
1907         int a_count;
1908         int a_sync;
1909         int *a_rtvals;
1910         vm_ooffset_t a_offset;
1911     };
1912 */
1913 static int
1914 fuse_vnop_putpages(struct vop_putpages_args *ap)
1915 {
1916 	struct uio uio;
1917 	struct iovec iov;
1918 	vm_offset_t kva;
1919 	struct buf *bp;
1920 	int i, error, npages, count;
1921 	off_t offset;
1922 	int *rtvals;
1923 	struct vnode *vp;
1924 	struct thread *td;
1925 	struct ucred *cred;
1926 	vm_page_t *pages;
1927 	vm_ooffset_t fsize;
1928 
1929 	FS_DEBUG2G("heh\n");
1930 
1931 	vp = ap->a_vp;
1932 	KASSERT(vp->v_object, ("objectless vp passed to putpages"));
1933 	fsize = vp->v_object->un_pager.vnp.vnp_size;
1934 	td = curthread;			/* XXX */
1935 	cred = curthread->td_ucred;	/* XXX */
1936 	pages = ap->a_m;
1937 	count = ap->a_count;
1938 	rtvals = ap->a_rtvals;
1939 	npages = btoc(count);
1940 	offset = IDX_TO_OFF(pages[0]->pindex);
1941 
1942 	if (!fsess_opt_mmap(vnode_mount(vp))) {
1943 		FS_DEBUG("called on non-cacheable vnode??\n");
1944 	}
1945 	for (i = 0; i < npages; i++)
1946 		rtvals[i] = VM_PAGER_AGAIN;
1947 
1948 	/*
1949 	 * When putting pages, do not extend file past EOF.
1950 	 */
1951 
1952 	if (offset + count > fsize) {
1953 		count = fsize - offset;
1954 		if (count < 0)
1955 			count = 0;
1956 	}
1957 	/*
1958 	 * We use only the kva address for the buffer, but this is extremely
1959 	 * convenient and fast.
1960 	 */
1961 	bp = getpbuf(&fuse_pbuf_freecnt);
1962 
1963 	kva = (vm_offset_t)bp->b_data;
1964 	pmap_qenter(kva, pages, npages);
1965 	VM_CNT_INC(v_vnodeout);
1966 	VM_CNT_ADD(v_vnodepgsout, count);
1967 
1968 	iov.iov_base = (caddr_t)kva;
1969 	iov.iov_len = count;
1970 	uio.uio_iov = &iov;
1971 	uio.uio_iovcnt = 1;
1972 	uio.uio_offset = offset;
1973 	uio.uio_resid = count;
1974 	uio.uio_segflg = UIO_SYSSPACE;
1975 	uio.uio_rw = UIO_WRITE;
1976 	uio.uio_td = td;
1977 
1978 	error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred);
1979 
1980 	pmap_qremove(kva, npages);
1981 	relpbuf(bp, &fuse_pbuf_freecnt);
1982 
1983 	if (!error) {
1984 		int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE;
1985 
1986 		for (i = 0; i < nwritten; i++) {
1987 			rtvals[i] = VM_PAGER_OK;
1988 			VM_OBJECT_WLOCK(pages[i]->object);
1989 			vm_page_undirty(pages[i]);
1990 			VM_OBJECT_WUNLOCK(pages[i]->object);
1991 		}
1992 	}
1993 	return rtvals[0];
1994 }
1995 
1996 static const char extattr_namespace_separator = '.';
1997 
1998 /*
1999     struct vop_getextattr_args {
2000         struct vop_generic_args a_gen;
2001         struct vnode *a_vp;
2002         int a_attrnamespace;
2003         const char *a_name;
2004         struct uio *a_uio;
2005         size_t *a_size;
2006         struct ucred *a_cred;
2007         struct thread *a_td;
2008     };
2009 */
2010 static int
2011 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2012 {
2013 	struct vnode *vp = ap->a_vp;
2014 	struct uio *uio = ap->a_uio;
2015 	struct fuse_dispatcher fdi;
2016 	struct fuse_getxattr_in *get_xattr_in;
2017 	struct fuse_getxattr_out *get_xattr_out;
2018 	struct mount *mp = vnode_mount(vp);
2019 	struct thread *td = ap->a_td;
2020 	struct ucred *cred = ap->a_cred;
2021 	char *prefix;
2022 	char *attr_str;
2023 	size_t len;
2024 	int err;
2025 
2026 	fuse_trace_printf_vnop();
2027 
2028 	if (fuse_isdeadfs(vp))
2029 		return (ENXIO);
2030 
2031 	/* Default to looking for user attributes. */
2032 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2033 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2034 	else
2035 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2036 
2037 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2038 	    strlen(ap->a_name) + 1;
2039 
2040 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2041 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2042 
2043 	get_xattr_in = fdi.indata;
2044 	/*
2045 	 * Check to see whether we're querying the available size or
2046 	 * issuing the actual request.  If we pass in 0, we get back struct
2047 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2048 	 * that much data, without the struct fuse_getxattr_out header.
2049 	 */
2050 	if (uio == NULL)
2051 		get_xattr_in->size = 0;
2052 	else
2053 		get_xattr_in->size = uio->uio_resid;
2054 
2055 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2056 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2057 	    ap->a_name);
2058 
2059 	err = fdisp_wait_answ(&fdi);
2060 	if (err != 0) {
2061 		if (err == ENOSYS)
2062 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2063 		debug_printf("getxattr: got err=%d from daemon\n", err);
2064 		goto out;
2065 	}
2066 
2067 	get_xattr_out = fdi.answ;
2068 
2069 	if (ap->a_size != NULL)
2070 		*ap->a_size = get_xattr_out->size;
2071 
2072 	if (uio != NULL)
2073 		err = uiomove(fdi.answ, fdi.iosize, uio);
2074 
2075 out:
2076 	fdisp_destroy(&fdi);
2077 	return (err);
2078 }
2079 
2080 /*
2081     struct vop_setextattr_args {
2082         struct vop_generic_args a_gen;
2083         struct vnode *a_vp;
2084         int a_attrnamespace;
2085         const char *a_name;
2086         struct uio *a_uio;
2087         struct ucred *a_cred;
2088         struct thread *a_td;
2089     };
2090 */
2091 static int
2092 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2093 {
2094 	struct vnode *vp = ap->a_vp;
2095 	struct uio *uio = ap->a_uio;
2096 	struct fuse_dispatcher fdi;
2097 	struct fuse_setxattr_in *set_xattr_in;
2098 	struct mount *mp = vnode_mount(vp);
2099 	struct thread *td = ap->a_td;
2100 	struct ucred *cred = ap->a_cred;
2101 	char *prefix;
2102 	size_t len;
2103 	char *attr_str;
2104 	int err;
2105 
2106 	fuse_trace_printf_vnop();
2107 
2108 	if (fuse_isdeadfs(vp))
2109 		return (ENXIO);
2110 
2111 	/* Default to looking for user attributes. */
2112 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2113 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2114 	else
2115 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2116 
2117 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2118 	    strlen(ap->a_name) + 1;
2119 
2120 	fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2121 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2122 
2123 	set_xattr_in = fdi.indata;
2124 	set_xattr_in->size = uio->uio_resid;
2125 
2126 	attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2127 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2128 	    ap->a_name);
2129 
2130 	err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2131 	    uio->uio_resid, uio);
2132 	if (err != 0) {
2133 		debug_printf("setxattr: got error %d from uiomove\n", err);
2134 		goto out;
2135 	}
2136 
2137 	err = fdisp_wait_answ(&fdi);
2138 
2139 	if (err != 0) {
2140 		if (err == ENOSYS)
2141 			fsess_set_notimpl(mp, FUSE_SETXATTR);
2142 		debug_printf("setxattr: got err=%d from daemon\n", err);
2143 		goto out;
2144 	}
2145 
2146 out:
2147 	fdisp_destroy(&fdi);
2148 	return (err);
2149 }
2150 
2151 /*
2152  * The Linux / FUSE extended attribute list is simply a collection of
2153  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2154  * byte length followed by a non-NUL terminated string.  So, this allows
2155  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2156  * Linux attribute names are reported with the namespace as a prefix (e.g.
2157  * "user.attribute_name"), but in FreeBSD they are reported without the
2158  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2159  *
2160  * user.attr_name1\0user.attr_name2\0
2161  *
2162  * to:
2163  *
2164  * <num>attr_name1<num>attr_name2
2165  *
2166  * Where "<num>" is a single byte number of characters in the attribute name.
2167  *
2168  * Args:
2169  * prefix - exattr namespace prefix string
2170  * list, list_len - input list with namespace prefixes
2171  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2172  */
2173 static int
2174 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2175     char *bsd_list, int *bsd_list_len)
2176 {
2177 	int len, pos, dist_to_next, prefix_len;
2178 
2179 	pos = 0;
2180 	*bsd_list_len = 0;
2181 	prefix_len = strlen(prefix);
2182 
2183 	while (pos < list_len && list[pos] != '\0') {
2184 		dist_to_next = strlen(&list[pos]) + 1;
2185 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2186 		    list[pos + prefix_len] == extattr_namespace_separator) {
2187 			len = dist_to_next -
2188 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2189 			if (len >= EXTATTR_MAXNAMELEN)
2190 				return (ENAMETOOLONG);
2191 
2192 			bsd_list[*bsd_list_len] = len;
2193 			memcpy(&bsd_list[*bsd_list_len + 1],
2194 			    &list[pos + prefix_len +
2195 			    sizeof(extattr_namespace_separator)], len);
2196 
2197 			*bsd_list_len += len + 1;
2198 		}
2199 
2200 		pos += dist_to_next;
2201 	}
2202 
2203 	return (0);
2204 }
2205 
2206 /*
2207     struct vop_listextattr_args {
2208         struct vop_generic_args a_gen;
2209         struct vnode *a_vp;
2210         int a_attrnamespace;
2211         struct uio *a_uio;
2212         size_t *a_size;
2213         struct ucred *a_cred;
2214         struct thread *a_td;
2215     };
2216 */
2217 static int
2218 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2219 {
2220 	struct vnode *vp = ap->a_vp;
2221 	struct uio *uio = ap->a_uio;
2222 	struct fuse_dispatcher fdi;
2223 	struct fuse_listxattr_in *list_xattr_in;
2224 	struct fuse_listxattr_out *list_xattr_out;
2225 	struct mount *mp = vnode_mount(vp);
2226 	struct thread *td = ap->a_td;
2227 	struct ucred *cred = ap->a_cred;
2228 	size_t len;
2229 	char *prefix;
2230 	char *attr_str;
2231 	char *bsd_list = NULL;
2232 	char *linux_list;
2233 	int bsd_list_len;
2234 	int linux_list_len;
2235 	int err;
2236 
2237 	fuse_trace_printf_vnop();
2238 
2239 	if (fuse_isdeadfs(vp))
2240 		return (ENXIO);
2241 
2242 	/*
2243 	 * Add space for a NUL and the period separator if enabled.
2244 	 * Default to looking for user attributes.
2245 	 */
2246 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2247 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2248 	else
2249 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2250 
2251 	len = strlen(prefix) + sizeof(extattr_namespace_separator) + 1;
2252 
2253 	fdisp_init(&fdi, sizeof(*list_xattr_in) + len);
2254 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2255 
2256 	/*
2257 	 * Retrieve Linux / FUSE compatible list size.
2258 	 */
2259 	list_xattr_in = fdi.indata;
2260 	list_xattr_in->size = 0;
2261 	attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2262 	snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2263 
2264 	err = fdisp_wait_answ(&fdi);
2265 	if (err != 0) {
2266 		if (err == ENOSYS)
2267 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
2268 		debug_printf("listextattr: got err=%d from daemon\n", err);
2269 		goto out;
2270 	}
2271 
2272 	list_xattr_out = fdi.answ;
2273 	linux_list_len = list_xattr_out->size;
2274 	if (linux_list_len == 0) {
2275 		if (ap->a_size != NULL)
2276 			*ap->a_size = linux_list_len;
2277 		goto out;
2278 	}
2279 
2280 	/*
2281 	 * Retrieve Linux / FUSE compatible list values.
2282 	 */
2283 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2284 	list_xattr_in = fdi.indata;
2285 	list_xattr_in->size = linux_list_len + sizeof(*list_xattr_out);
2286 	attr_str = (char *)fdi.indata + sizeof(*list_xattr_in);
2287 	snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2288 
2289 	err = fdisp_wait_answ(&fdi);
2290 	if (err != 0)
2291 		goto out;
2292 
2293 	linux_list = fdi.answ;
2294 	linux_list_len = fdi.iosize;
2295 
2296 	/*
2297 	 * Retrieve the BSD compatible list values.
2298 	 * The Linux / FUSE attribute list format isn't the same
2299 	 * as FreeBSD's format. So we need to transform it into
2300 	 * FreeBSD's format before giving it to the user.
2301 	 */
2302 	bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK);
2303 	err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len,
2304 	    bsd_list, &bsd_list_len);
2305 	if (err != 0)
2306 		goto out;
2307 
2308 	if (ap->a_size != NULL)
2309 		*ap->a_size = bsd_list_len;
2310 
2311 	if (uio != NULL)
2312 		err = uiomove(bsd_list, bsd_list_len, uio);
2313 
2314 out:
2315 	free(bsd_list, M_TEMP);
2316 	fdisp_destroy(&fdi);
2317 	return (err);
2318 }
2319 
2320 /*
2321     struct vop_deleteextattr_args {
2322         struct vop_generic_args a_gen;
2323         struct vnode *a_vp;
2324         int a_attrnamespace;
2325         const char *a_name;
2326         struct ucred *a_cred;
2327         struct thread *a_td;
2328     };
2329 */
2330 static int
2331 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2332 {
2333 	struct vnode *vp = ap->a_vp;
2334 	struct fuse_dispatcher fdi;
2335 	struct mount *mp = vnode_mount(vp);
2336 	struct thread *td = ap->a_td;
2337 	struct ucred *cred = ap->a_cred;
2338 	char *prefix;
2339 	size_t len;
2340 	char *attr_str;
2341 	int err;
2342 
2343 	fuse_trace_printf_vnop();
2344 
2345 	if (fuse_isdeadfs(vp))
2346 		return (ENXIO);
2347 
2348 	/* Default to looking for user attributes. */
2349 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2350 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2351 	else
2352 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2353 
2354 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2355 	    strlen(ap->a_name) + 1;
2356 
2357 	fdisp_init(&fdi, len);
2358 	fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2359 
2360 	attr_str = fdi.indata;
2361 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2362 	    ap->a_name);
2363 
2364 	err = fdisp_wait_answ(&fdi);
2365 	if (err != 0) {
2366 		if (err == ENOSYS)
2367 			fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2368 		debug_printf("removexattr: got err=%d from daemon\n", err);
2369 	}
2370 
2371 	fdisp_destroy(&fdi);
2372 	return (err);
2373 }
2374 
2375 /*
2376     struct vnop_print_args {
2377         struct vnode *a_vp;
2378     };
2379 */
2380 static int
2381 fuse_vnop_print(struct vop_print_args *ap)
2382 {
2383 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2384 
2385 	printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2386 	    (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2387 	    (uintmax_t)fvdat->nlookup,
2388 	    fvdat->flag);
2389 
2390 	return 0;
2391 }
2392