xref: /freebsd/sys/fs/fuse/fuse_vnops.c (revision 396c556d77189a5c474d35cec6f44a762e310b7d)
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_read_t fuse_vnop_read;
130 static vop_readdir_t fuse_vnop_readdir;
131 static vop_readlink_t fuse_vnop_readlink;
132 static vop_reclaim_t fuse_vnop_reclaim;
133 static vop_remove_t fuse_vnop_remove;
134 static vop_rename_t fuse_vnop_rename;
135 static vop_rmdir_t fuse_vnop_rmdir;
136 static vop_setattr_t fuse_vnop_setattr;
137 static vop_setextattr_t fuse_vnop_setextattr;
138 static vop_strategy_t fuse_vnop_strategy;
139 static vop_symlink_t fuse_vnop_symlink;
140 static vop_write_t fuse_vnop_write;
141 static vop_getpages_t fuse_vnop_getpages;
142 static vop_putpages_t fuse_vnop_putpages;
143 static vop_print_t fuse_vnop_print;
144 
145 struct vop_vector fuse_vnops = {
146 	.vop_default = &default_vnodeops,
147 	.vop_access = fuse_vnop_access,
148 	.vop_close = fuse_vnop_close,
149 	.vop_create = fuse_vnop_create,
150 	.vop_deleteextattr = fuse_vnop_deleteextattr,
151 	.vop_fsync = fuse_vnop_fsync,
152 	.vop_getattr = fuse_vnop_getattr,
153 	.vop_getextattr = fuse_vnop_getextattr,
154 	.vop_inactive = fuse_vnop_inactive,
155 	.vop_link = fuse_vnop_link,
156 	.vop_listextattr = fuse_vnop_listextattr,
157 	.vop_lookup = fuse_vnop_lookup,
158 	.vop_mkdir = fuse_vnop_mkdir,
159 	.vop_mknod = fuse_vnop_mknod,
160 	.vop_open = fuse_vnop_open,
161 	.vop_pathconf = vop_stdpathconf,
162 	.vop_read = fuse_vnop_read,
163 	.vop_readdir = fuse_vnop_readdir,
164 	.vop_readlink = fuse_vnop_readlink,
165 	.vop_reclaim = fuse_vnop_reclaim,
166 	.vop_remove = fuse_vnop_remove,
167 	.vop_rename = fuse_vnop_rename,
168 	.vop_rmdir = fuse_vnop_rmdir,
169 	.vop_setattr = fuse_vnop_setattr,
170 	.vop_setextattr = fuse_vnop_setextattr,
171 	.vop_strategy = fuse_vnop_strategy,
172 	.vop_symlink = fuse_vnop_symlink,
173 	.vop_write = fuse_vnop_write,
174 	.vop_getpages = fuse_vnop_getpages,
175 	.vop_putpages = fuse_vnop_putpages,
176 	.vop_print = fuse_vnop_print,
177 };
178 
179 static u_long fuse_lookup_cache_hits = 0;
180 
181 SYSCTL_ULONG(_vfs_fuse, OID_AUTO, lookup_cache_hits, CTLFLAG_RD,
182     &fuse_lookup_cache_hits, 0, "");
183 
184 static u_long fuse_lookup_cache_misses = 0;
185 
186 SYSCTL_ULONG(_vfs_fuse, OID_AUTO, lookup_cache_misses, CTLFLAG_RD,
187     &fuse_lookup_cache_misses, 0, "");
188 
189 int	fuse_lookup_cache_enable = 1;
190 
191 SYSCTL_INT(_vfs_fuse, OID_AUTO, lookup_cache_enable, CTLFLAG_RW,
192     &fuse_lookup_cache_enable, 0, "");
193 
194 /*
195  * XXX: This feature is highly experimental and can bring to instabilities,
196  * needs revisiting before to be enabled by default.
197  */
198 static int fuse_reclaim_revoked = 0;
199 
200 SYSCTL_INT(_vfs_fuse, OID_AUTO, reclaim_revoked, CTLFLAG_RW,
201     &fuse_reclaim_revoked, 0, "");
202 
203 int	fuse_pbuf_freecnt = -1;
204 
205 #define fuse_vm_page_lock(m)		vm_page_lock((m));
206 #define fuse_vm_page_unlock(m)		vm_page_unlock((m));
207 #define fuse_vm_page_lock_queues()	((void)0)
208 #define fuse_vm_page_unlock_queues()	((void)0)
209 
210 /*
211     struct vnop_access_args {
212         struct vnode *a_vp;
213 #if VOP_ACCESS_TAKES_ACCMODE_T
214         accmode_t a_accmode;
215 #else
216         int a_mode;
217 #endif
218         struct ucred *a_cred;
219         struct thread *a_td;
220     };
221 */
222 static int
223 fuse_vnop_access(struct vop_access_args *ap)
224 {
225 	struct vnode *vp = ap->a_vp;
226 	int accmode = ap->a_accmode;
227 	struct ucred *cred = ap->a_cred;
228 
229 	struct fuse_access_param facp;
230 	struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp));
231 
232 	int err;
233 
234 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
235 
236 	if (fuse_isdeadfs(vp)) {
237 		if (vnode_isvroot(vp)) {
238 			return 0;
239 		}
240 		return ENXIO;
241 	}
242 	if (!(data->dataflags & FSESS_INITED)) {
243 		if (vnode_isvroot(vp)) {
244 			if (priv_check_cred(cred, PRIV_VFS_ADMIN, 0) ||
245 			    (fuse_match_cred(data->daemoncred, cred) == 0)) {
246 				return 0;
247 			}
248 		}
249 		return EBADF;
250 	}
251 	if (vnode_islnk(vp)) {
252 		return 0;
253 	}
254 	bzero(&facp, sizeof(facp));
255 
256 	err = fuse_internal_access(vp, accmode, &facp, ap->a_td, ap->a_cred);
257 	FS_DEBUG2G("err=%d accmode=0x%x\n", err, accmode);
258 	return err;
259 }
260 
261 /*
262     struct vnop_close_args {
263 	struct vnode *a_vp;
264 	int  a_fflag;
265 	struct ucred *a_cred;
266 	struct thread *a_td;
267     };
268 */
269 static int
270 fuse_vnop_close(struct vop_close_args *ap)
271 {
272 	struct vnode *vp = ap->a_vp;
273 	struct ucred *cred = ap->a_cred;
274 	int fflag = ap->a_fflag;
275 	fufh_type_t fufh_type;
276 
277 	fuse_trace_printf_vnop();
278 
279 	if (fuse_isdeadfs(vp)) {
280 		return 0;
281 	}
282 	if (vnode_isdir(vp)) {
283 		if (fuse_filehandle_valid(vp, FUFH_RDONLY)) {
284 			fuse_filehandle_close(vp, FUFH_RDONLY, NULL, cred);
285 		}
286 		return 0;
287 	}
288 	if (fflag & IO_NDELAY) {
289 		return 0;
290 	}
291 	fufh_type = fuse_filehandle_xlate_from_fflags(fflag);
292 
293 	if (!fuse_filehandle_valid(vp, fufh_type)) {
294 		int i;
295 
296 		for (i = 0; i < FUFH_MAXTYPE; i++)
297 			if (fuse_filehandle_valid(vp, i))
298 				break;
299 		if (i == FUFH_MAXTYPE)
300 			panic("FUSE: fufh type %d found to be invalid in close"
301 			      " (fflag=0x%x)\n",
302 			      fufh_type, fflag);
303 	}
304 	if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
305 		fuse_vnode_savesize(vp, cred);
306 	}
307 	return 0;
308 }
309 
310 /*
311     struct vnop_create_args {
312 	struct vnode *a_dvp;
313 	struct vnode **a_vpp;
314 	struct componentname *a_cnp;
315 	struct vattr *a_vap;
316     };
317 */
318 static int
319 fuse_vnop_create(struct vop_create_args *ap)
320 {
321 	struct vnode *dvp = ap->a_dvp;
322 	struct vnode **vpp = ap->a_vpp;
323 	struct componentname *cnp = ap->a_cnp;
324 	struct vattr *vap = ap->a_vap;
325 	struct thread *td = cnp->cn_thread;
326 	struct ucred *cred = cnp->cn_cred;
327 
328 	struct fuse_open_in *foi;
329 	struct fuse_entry_out *feo;
330 	struct fuse_dispatcher fdi;
331 	struct fuse_dispatcher *fdip = &fdi;
332 
333 	int err;
334 
335 	struct mount *mp = vnode_mount(dvp);
336 	uint64_t parentnid = VTOFUD(dvp)->nid;
337 	mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode);
338 	uint64_t x_fh_id;
339 	uint32_t x_open_flags;
340 
341 	fuse_trace_printf_vnop();
342 
343 	if (fuse_isdeadfs(dvp)) {
344 		return ENXIO;
345 	}
346 	bzero(&fdi, sizeof(fdi));
347 
348 	/* XXX:	Will we ever want devices ? */
349 	if ((vap->va_type != VREG)) {
350 		printf("fuse_vnop_create: unsupported va_type %d\n",
351 		    vap->va_type);
352 		return (EINVAL);
353 	}
354 	debug_printf("parent nid = %ju, mode = %x\n", (uintmax_t)parentnid,
355 	    mode);
356 
357 	fdisp_init(fdip, sizeof(*foi) + cnp->cn_namelen + 1);
358 	if (!fsess_isimpl(mp, FUSE_CREATE)) {
359 		debug_printf("eh, daemon doesn't implement create?\n");
360 		return (EINVAL);
361 	}
362 	fdisp_make(fdip, FUSE_CREATE, vnode_mount(dvp), parentnid, td, cred);
363 
364 	foi = fdip->indata;
365 	foi->mode = mode;
366 	foi->flags = O_CREAT | O_RDWR;
367 
368 	memcpy((char *)fdip->indata + sizeof(*foi), cnp->cn_nameptr,
369 	    cnp->cn_namelen);
370 	((char *)fdip->indata)[sizeof(*foi) + cnp->cn_namelen] = '\0';
371 
372 	err = fdisp_wait_answ(fdip);
373 
374 	if (err) {
375 		if (err == ENOSYS)
376 			fsess_set_notimpl(mp, FUSE_CREATE);
377 		debug_printf("create: got err=%d from daemon\n", err);
378 		goto out;
379 	}
380 
381 	feo = fdip->answ;
382 
383 	if ((err = fuse_internal_checkentry(feo, VREG))) {
384 		goto out;
385 	}
386 	err = fuse_vnode_get(mp, feo->nodeid, dvp, vpp, cnp, VREG);
387 	if (err) {
388 		struct fuse_release_in *fri;
389 		uint64_t nodeid = feo->nodeid;
390 		uint64_t fh_id = ((struct fuse_open_out *)(feo + 1))->fh;
391 
392 		fdisp_init(fdip, sizeof(*fri));
393 		fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred);
394 		fri = fdip->indata;
395 		fri->fh = fh_id;
396 		fri->flags = OFLAGS(mode);
397 		fuse_insert_callback(fdip->tick, fuse_internal_forget_callback);
398 		fuse_insert_message(fdip->tick);
399 		return err;
400 	}
401 	ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create");
402 
403 	fdip->answ = feo + 1;
404 
405 	x_fh_id = ((struct fuse_open_out *)(feo + 1))->fh;
406 	x_open_flags = ((struct fuse_open_out *)(feo + 1))->open_flags;
407 	fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, x_fh_id);
408 	fuse_vnode_open(*vpp, x_open_flags, td);
409 	cache_purge_negative(dvp);
410 
411 out:
412 	fdisp_destroy(fdip);
413 	return err;
414 }
415 
416 /*
417  * Our vnop_fsync roughly corresponds to the FUSE_FSYNC method. The Linux
418  * version of FUSE also has a FUSE_FLUSH method.
419  *
420  * On Linux, fsync() synchronizes a file's complete in-core state with that
421  * on disk. The call is not supposed to return until the system has completed
422  * that action or until an error is detected.
423  *
424  * Linux also has an fdatasync() call that is similar to fsync() but is not
425  * required to update the metadata such as access time and modification time.
426  */
427 
428 /*
429     struct vnop_fsync_args {
430 	struct vnodeop_desc *a_desc;
431 	struct vnode * a_vp;
432 	struct ucred * a_cred;
433 	int  a_waitfor;
434 	struct thread * a_td;
435     };
436 */
437 static int
438 fuse_vnop_fsync(struct vop_fsync_args *ap)
439 {
440 	struct vnode *vp = ap->a_vp;
441 	struct thread *td = ap->a_td;
442 
443 	struct fuse_filehandle *fufh;
444 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
445 
446 	int type, err = 0;
447 
448 	fuse_trace_printf_vnop();
449 
450 	if (fuse_isdeadfs(vp)) {
451 		return 0;
452 	}
453 	if ((err = vop_stdfsync(ap)))
454 		return err;
455 
456 	if (!fsess_isimpl(vnode_mount(vp),
457 	    (vnode_vtype(vp) == VDIR ? FUSE_FSYNCDIR : FUSE_FSYNC))) {
458 		goto out;
459 	}
460 	for (type = 0; type < FUFH_MAXTYPE; type++) {
461 		fufh = &(fvdat->fufh[type]);
462 		if (FUFH_IS_VALID(fufh)) {
463 			fuse_internal_fsync(vp, td, NULL, fufh);
464 		}
465 	}
466 
467 out:
468 	return 0;
469 }
470 
471 /*
472     struct vnop_getattr_args {
473 	struct vnode *a_vp;
474 	struct vattr *a_vap;
475 	struct ucred *a_cred;
476 	struct thread *a_td;
477     };
478 */
479 static int
480 fuse_vnop_getattr(struct vop_getattr_args *ap)
481 {
482 	struct vnode *vp = ap->a_vp;
483 	struct vattr *vap = ap->a_vap;
484 	struct ucred *cred = ap->a_cred;
485 	struct thread *td = curthread;
486 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
487 
488 	int err = 0;
489 	int dataflags;
490 	struct fuse_dispatcher fdi;
491 
492 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
493 
494 	dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags;
495 
496 	/* Note that we are not bailing out on a dead file system just yet. */
497 
498 	if (!(dataflags & FSESS_INITED)) {
499 		if (!vnode_isvroot(vp)) {
500 			fdata_set_dead(fuse_get_mpdata(vnode_mount(vp)));
501 			err = ENOTCONN;
502 			debug_printf("fuse_getattr b: returning ENOTCONN\n");
503 			return err;
504 		} else {
505 			goto fake;
506 		}
507 	}
508 	fdisp_init(&fdi, 0);
509 	if ((err = fdisp_simple_putget_vp(&fdi, FUSE_GETATTR, vp, td, cred))) {
510 		if ((err == ENOTCONN) && vnode_isvroot(vp)) {
511 			/* see comment at similar place in fuse_statfs() */
512 			fdisp_destroy(&fdi);
513 			goto fake;
514 		}
515 		if (err == ENOENT) {
516 			fuse_internal_vnode_disappear(vp);
517 		}
518 		goto out;
519 	}
520 	cache_attrs(vp, (struct fuse_attr_out *)fdi.answ);
521 	if (vap != VTOVA(vp)) {
522 		memcpy(vap, VTOVA(vp), sizeof(*vap));
523 	}
524 	if (vap->va_type != vnode_vtype(vp)) {
525 		fuse_internal_vnode_disappear(vp);
526 		err = ENOENT;
527 		goto out;
528 	}
529 	if ((fvdat->flag & FN_SIZECHANGE) != 0)
530 		vap->va_size = fvdat->filesize;
531 
532 	if (vnode_isreg(vp) && (fvdat->flag & FN_SIZECHANGE) == 0) {
533 		/*
534 	         * This is for those cases when the file size changed without us
535 	         * knowing, and we want to catch up.
536 	         */
537 		off_t new_filesize = ((struct fuse_attr_out *)
538 				      fdi.answ)->attr.size;
539 
540 		if (fvdat->filesize != new_filesize) {
541 			fuse_vnode_setsize(vp, cred, new_filesize);
542 		}
543 	}
544 	debug_printf("fuse_getattr e: returning 0\n");
545 
546 out:
547 	fdisp_destroy(&fdi);
548 	return err;
549 
550 fake:
551 	bzero(vap, sizeof(*vap));
552 	vap->va_type = vnode_vtype(vp);
553 
554 	return 0;
555 }
556 
557 /*
558     struct vnop_inactive_args {
559 	struct vnode *a_vp;
560 	struct thread *a_td;
561     };
562 */
563 static int
564 fuse_vnop_inactive(struct vop_inactive_args *ap)
565 {
566 	struct vnode *vp = ap->a_vp;
567 	struct thread *td = ap->a_td;
568 
569 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
570 	struct fuse_filehandle *fufh = NULL;
571 
572 	int type, need_flush = 1;
573 
574 	FS_DEBUG("inode=%ju\n", (uintmax_t)VTOI(vp));
575 
576 	for (type = 0; type < FUFH_MAXTYPE; type++) {
577 		fufh = &(fvdat->fufh[type]);
578 		if (FUFH_IS_VALID(fufh)) {
579 			if (need_flush && vp->v_type == VREG) {
580 				if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) {
581 					fuse_vnode_savesize(vp, NULL);
582 				}
583 				if (fuse_data_cache_invalidate ||
584 				    (fvdat->flag & FN_REVOKED) != 0)
585 					fuse_io_invalbuf(vp, td);
586 				else
587 					fuse_io_flushbuf(vp, MNT_WAIT, td);
588 				need_flush = 0;
589 			}
590 			fuse_filehandle_close(vp, type, td, NULL);
591 		}
592 	}
593 
594 	if ((fvdat->flag & FN_REVOKED) != 0 && fuse_reclaim_revoked) {
595 		vrecycle(vp);
596 	}
597 	return 0;
598 }
599 
600 /*
601     struct vnop_link_args {
602 	struct vnode *a_tdvp;
603 	struct vnode *a_vp;
604 	struct componentname *a_cnp;
605     };
606 */
607 static int
608 fuse_vnop_link(struct vop_link_args *ap)
609 {
610 	struct vnode *vp = ap->a_vp;
611 	struct vnode *tdvp = ap->a_tdvp;
612 	struct componentname *cnp = ap->a_cnp;
613 
614 	struct vattr *vap = VTOVA(vp);
615 
616 	struct fuse_dispatcher fdi;
617 	struct fuse_entry_out *feo;
618 	struct fuse_link_in fli;
619 
620 	int err;
621 
622 	fuse_trace_printf_vnop();
623 
624 	if (fuse_isdeadfs(vp)) {
625 		return ENXIO;
626 	}
627 	if (vnode_mount(tdvp) != vnode_mount(vp)) {
628 		return EXDEV;
629 	}
630 	if (vap->va_nlink >= FUSE_LINK_MAX) {
631 		return EMLINK;
632 	}
633 	fli.oldnodeid = VTOI(vp);
634 
635 	fdisp_init(&fdi, 0);
636 	fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp,
637 	    FUSE_LINK, &fli, sizeof(fli), &fdi);
638 	if ((err = fdisp_wait_answ(&fdi))) {
639 		goto out;
640 	}
641 	feo = fdi.answ;
642 
643 	err = fuse_internal_checkentry(feo, vnode_vtype(vp));
644 out:
645 	fdisp_destroy(&fdi);
646 	return err;
647 }
648 
649 /*
650     struct vnop_lookup_args {
651 	struct vnodeop_desc *a_desc;
652 	struct vnode *a_dvp;
653 	struct vnode **a_vpp;
654 	struct componentname *a_cnp;
655     };
656 */
657 int
658 fuse_vnop_lookup(struct vop_lookup_args *ap)
659 {
660 	struct vnode *dvp = ap->a_dvp;
661 	struct vnode **vpp = ap->a_vpp;
662 	struct componentname *cnp = ap->a_cnp;
663 	struct thread *td = cnp->cn_thread;
664 	struct ucred *cred = cnp->cn_cred;
665 
666 	int nameiop = cnp->cn_nameiop;
667 	int flags = cnp->cn_flags;
668 	int wantparent = flags & (LOCKPARENT | WANTPARENT);
669 	int islastcn = flags & ISLASTCN;
670 	struct mount *mp = vnode_mount(dvp);
671 
672 	int err = 0;
673 	int lookup_err = 0;
674 	struct vnode *vp = NULL;
675 
676 	struct fuse_dispatcher fdi;
677 	enum fuse_opcode op;
678 
679 	uint64_t nid;
680 	struct fuse_access_param facp;
681 
682 	FS_DEBUG2G("parent_inode=%ju - %*s\n",
683 	    (uintmax_t)VTOI(dvp), (int)cnp->cn_namelen, cnp->cn_nameptr);
684 
685 	if (fuse_isdeadfs(dvp)) {
686 		*vpp = NULL;
687 		return ENXIO;
688 	}
689 	if (!vnode_isdir(dvp)) {
690 		return ENOTDIR;
691 	}
692 	if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP)) {
693 		return EROFS;
694 	}
695 	/*
696          * We do access check prior to doing anything else only in the case
697          * when we are at fs root (we'd like to say, "we are at the first
698          * component", but that's not exactly the same... nevermind).
699          * See further comments at further access checks.
700          */
701 
702 	bzero(&facp, sizeof(facp));
703 	if (vnode_isvroot(dvp)) {	/* early permission check hack */
704 		if ((err = fuse_internal_access(dvp, VEXEC, &facp, td, cred))) {
705 			return err;
706 		}
707 	}
708 	if (flags & ISDOTDOT) {
709 		nid = VTOFUD(dvp)->parent_nid;
710 		if (nid == 0) {
711 			return ENOENT;
712 		}
713 		fdisp_init(&fdi, 0);
714 		op = FUSE_GETATTR;
715 		goto calldaemon;
716 	} else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') {
717 		nid = VTOI(dvp);
718 		fdisp_init(&fdi, 0);
719 		op = FUSE_GETATTR;
720 		goto calldaemon;
721 	} else if (fuse_lookup_cache_enable) {
722 		err = cache_lookup(dvp, vpp, cnp, NULL, NULL);
723 		switch (err) {
724 
725 		case -1:		/* positive match */
726 			atomic_add_acq_long(&fuse_lookup_cache_hits, 1);
727 			return 0;
728 
729 		case 0:		/* no match in cache */
730 			atomic_add_acq_long(&fuse_lookup_cache_misses, 1);
731 			break;
732 
733 		case ENOENT:		/* negative match */
734 			/* fall through */
735 		default:
736 			return err;
737 		}
738 	}
739 	nid = VTOI(dvp);
740 	fdisp_init(&fdi, cnp->cn_namelen + 1);
741 	op = FUSE_LOOKUP;
742 
743 calldaemon:
744 	fdisp_make(&fdi, op, mp, nid, td, cred);
745 
746 	if (op == FUSE_LOOKUP) {
747 		memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
748 		((char *)fdi.indata)[cnp->cn_namelen] = '\0';
749 	}
750 	lookup_err = fdisp_wait_answ(&fdi);
751 
752 	if ((op == FUSE_LOOKUP) && !lookup_err) {	/* lookup call succeeded */
753 		nid = ((struct fuse_entry_out *)fdi.answ)->nodeid;
754 		if (!nid) {
755 			/*
756 	                 * zero nodeid is the same as "not found",
757 	                 * but it's also cacheable (which we keep
758 	                 * keep on doing not as of writing this)
759 	                 */
760 			lookup_err = ENOENT;
761 		} else if (nid == FUSE_ROOT_ID) {
762 			lookup_err = EINVAL;
763 		}
764 	}
765 	if (lookup_err &&
766 	    (!fdi.answ_stat || lookup_err != ENOENT || op != FUSE_LOOKUP)) {
767 		fdisp_destroy(&fdi);
768 		return lookup_err;
769 	}
770 	/* lookup_err, if non-zero, must be ENOENT at this point */
771 
772 	if (lookup_err) {
773 
774 		if ((nameiop == CREATE || nameiop == RENAME) && islastcn
775 		     /* && directory dvp has not been removed */ ) {
776 
777 			if (vfs_isrdonly(mp)) {
778 				err = EROFS;
779 				goto out;
780 			}
781 #if 0 /* THINK_ABOUT_THIS */
782 			if ((err = fuse_internal_access(dvp, VWRITE, cred, td, &facp))) {
783 				goto out;
784 			}
785 #endif
786 
787 			/*
788 	                 * Possibly record the position of a slot in the
789 	                 * directory large enough for the new component name.
790 	                 * This can be recorded in the vnode private data for
791 	                 * dvp. Set the SAVENAME flag to hold onto the
792 	                 * pathname for use later in VOP_CREATE or VOP_RENAME.
793 	                 */
794 			cnp->cn_flags |= SAVENAME;
795 
796 			err = EJUSTRETURN;
797 			goto out;
798 		}
799 		/* Consider inserting name into cache. */
800 
801 		/*
802 	         * No we can't use negative caching, as the fs
803 	         * changes are out of our control.
804 	         * False positives' falseness turns out just as things
805 	         * go by, but false negatives' falseness doesn't.
806 	         * (and aiding the caching mechanism with extra control
807 	         * mechanisms comes quite close to beating the whole purpose
808 	         * caching...)
809 	         */
810 #if 0
811 		if ((cnp->cn_flags & MAKEENTRY) != 0) {
812 			FS_DEBUG("inserting NULL into cache\n");
813 			cache_enter(dvp, NULL, cnp);
814 		}
815 #endif
816 		err = ENOENT;
817 		goto out;
818 
819 	} else {
820 
821 		/* !lookup_err */
822 
823 		struct fuse_entry_out *feo = NULL;
824 		struct fuse_attr *fattr = NULL;
825 
826 		if (op == FUSE_GETATTR) {
827 			fattr = &((struct fuse_attr_out *)fdi.answ)->attr;
828 		} else {
829 			feo = (struct fuse_entry_out *)fdi.answ;
830 			fattr = &(feo->attr);
831 		}
832 
833 		/*
834 	         * If deleting, and at end of pathname, return parameters
835 	         * which can be used to remove file.  If the wantparent flag
836 	         * isn't set, we return only the directory, otherwise we go on
837 	         * and lock the inode, being careful with ".".
838 	         */
839 		if (nameiop == DELETE && islastcn) {
840 			/*
841 	                 * Check for write access on directory.
842 	                 */
843 			facp.xuid = fattr->uid;
844 			facp.facc_flags |= FACCESS_STICKY;
845 			err = fuse_internal_access(dvp, VWRITE, &facp, td, cred);
846 			facp.facc_flags &= ~FACCESS_XQUERIES;
847 
848 			if (err) {
849 				goto out;
850 			}
851 			if (nid == VTOI(dvp)) {
852 				vref(dvp);
853 				*vpp = dvp;
854 			} else {
855 				err = fuse_vnode_get(dvp->v_mount, nid, dvp,
856 				    &vp, cnp, IFTOVT(fattr->mode));
857 				if (err)
858 					goto out;
859 				*vpp = vp;
860 			}
861 
862 			/*
863 			 * Save the name for use in VOP_RMDIR and VOP_REMOVE
864 			 * later.
865 			 */
866 			cnp->cn_flags |= SAVENAME;
867 			goto out;
868 
869 		}
870 		/*
871 	         * If rewriting (RENAME), return the inode and the
872 	         * information required to rewrite the present directory
873 	         * Must get inode of directory entry to verify it's a
874 	         * regular file, or empty directory.
875 	         */
876 		if (nameiop == RENAME && wantparent && islastcn) {
877 
878 #if 0 /* THINK_ABOUT_THIS */
879 			if ((err = fuse_internal_access(dvp, VWRITE, cred, td, &facp))) {
880 				goto out;
881 			}
882 #endif
883 
884 			/*
885 	                 * Check for "."
886 	                 */
887 			if (nid == VTOI(dvp)) {
888 				err = EISDIR;
889 				goto out;
890 			}
891 			err = fuse_vnode_get(vnode_mount(dvp),
892 			    nid,
893 			    dvp,
894 			    &vp,
895 			    cnp,
896 			    IFTOVT(fattr->mode));
897 			if (err) {
898 				goto out;
899 			}
900 			*vpp = vp;
901 			/*
902 	                 * Save the name for use in VOP_RENAME later.
903 	                 */
904 			cnp->cn_flags |= SAVENAME;
905 
906 			goto out;
907 		}
908 		if (flags & ISDOTDOT) {
909 			struct mount *mp;
910 			int ltype;
911 
912 			/*
913 			 * Expanded copy of vn_vget_ino() so that
914 			 * fuse_vnode_get() can be used.
915 			 */
916 			mp = dvp->v_mount;
917 			ltype = VOP_ISLOCKED(dvp);
918 			err = vfs_busy(mp, MBF_NOWAIT);
919 			if (err != 0) {
920 				vfs_ref(mp);
921 				VOP_UNLOCK(dvp, 0);
922 				err = vfs_busy(mp, 0);
923 				vn_lock(dvp, ltype | LK_RETRY);
924 				vfs_rel(mp);
925 				if (err)
926 					goto out;
927 				if ((dvp->v_iflag & VI_DOOMED) != 0) {
928 					err = ENOENT;
929 					vfs_unbusy(mp);
930 					goto out;
931 				}
932 			}
933 			VOP_UNLOCK(dvp, 0);
934 			err = fuse_vnode_get(vnode_mount(dvp),
935 			    nid,
936 			    NULL,
937 			    &vp,
938 			    cnp,
939 			    IFTOVT(fattr->mode));
940 			vfs_unbusy(mp);
941 			vn_lock(dvp, ltype | LK_RETRY);
942 			if ((dvp->v_iflag & VI_DOOMED) != 0) {
943 				if (err == 0)
944 					vput(vp);
945 				err = ENOENT;
946 			}
947 			if (err)
948 				goto out;
949 			*vpp = vp;
950 		} else if (nid == VTOI(dvp)) {
951 			vref(dvp);
952 			*vpp = dvp;
953 		} else {
954 			err = fuse_vnode_get(vnode_mount(dvp),
955 			    nid,
956 			    dvp,
957 			    &vp,
958 			    cnp,
959 			    IFTOVT(fattr->mode));
960 			if (err) {
961 				goto out;
962 			}
963 			fuse_vnode_setparent(vp, dvp);
964 			*vpp = vp;
965 		}
966 
967 		if (op == FUSE_GETATTR) {
968 			cache_attrs(*vpp, (struct fuse_attr_out *)fdi.answ);
969 		} else {
970 			cache_attrs(*vpp, (struct fuse_entry_out *)fdi.answ);
971 		}
972 
973 		/* Insert name into cache if appropriate. */
974 
975 		/*
976 	         * Nooo, caching is evil. With caching, we can't avoid stale
977 	         * information taking over the playground (cached info is not
978 	         * just positive/negative, it does have qualitative aspects,
979 	         * too). And a (VOP/FUSE)_GETATTR is always thrown anyway, when
980 	         * walking down along cached path components, and that's not
981 	         * any cheaper than FUSE_LOOKUP. This might change with
982 	         * implementing kernel side attr caching, but... In Linux,
983 	         * lookup results are not cached, and the daemon is bombarded
984 	         * with FUSE_LOOKUPS on and on. This shows that by design, the
985 	         * daemon is expected to handle frequent lookup queries
986 	         * efficiently, do its caching in userspace, and so on.
987 	         *
988 	         * So just leave the name cache alone.
989 	         */
990 
991 		/*
992 	         * Well, now I know, Linux caches lookups, but with a
993 	         * timeout... So it's the same thing as attribute caching:
994 	         * we can deal with it when implement timeouts.
995 	         */
996 #if 0
997 		if (cnp->cn_flags & MAKEENTRY) {
998 			cache_enter(dvp, *vpp, cnp);
999 		}
1000 #endif
1001 	}
1002 out:
1003 	if (!lookup_err) {
1004 
1005 		/* No lookup error; need to clean up. */
1006 
1007 		if (err) {		/* Found inode; exit with no vnode. */
1008 			if (op == FUSE_LOOKUP) {
1009 				fuse_internal_forget_send(vnode_mount(dvp), td, cred,
1010 				    nid, 1);
1011 			}
1012 			fdisp_destroy(&fdi);
1013 			return err;
1014 		} else {
1015 #ifndef NO_EARLY_PERM_CHECK_HACK
1016 			if (!islastcn) {
1017 				/*
1018 				 * We have the attributes of the next item
1019 				 * *now*, and it's a fact, and we do not
1020 				 * have to do extra work for it (ie, beg the
1021 				 * daemon), and it neither depends on such
1022 				 * accidental things like attr caching. So
1023 				 * the big idea: check credentials *now*,
1024 				 * not at the beginning of the next call to
1025 				 * lookup.
1026 				 *
1027 				 * The first item of the lookup chain (fs root)
1028 				 * won't be checked then here, of course, as
1029 				 * its never "the next". But go and see that
1030 				 * the root is taken care about at the very
1031 				 * beginning of this function.
1032 				 *
1033 				 * Now, given we want to do the access check
1034 				 * this way, one might ask: so then why not
1035 				 * do the access check just after fetching
1036 				 * the inode and its attributes from the
1037 				 * daemon? Why bother with producing the
1038 				 * corresponding vnode at all if something
1039 				 * is not OK? We know what's the deal as
1040 				 * soon as we get those attrs... There is
1041 				 * one bit of info though not given us by
1042 				 * the daemon: whether his response is
1043 				 * authoritative or not... His response should
1044 				 * be ignored if something is mounted over
1045 				 * the dir in question. But that can be
1046 				 * known only by having the vnode...
1047 				 */
1048 				int tmpvtype = vnode_vtype(*vpp);
1049 
1050 				bzero(&facp, sizeof(facp));
1051 				/*the early perm check hack */
1052 				    facp.facc_flags |= FACCESS_VA_VALID;
1053 
1054 				if ((tmpvtype != VDIR) && (tmpvtype != VLNK)) {
1055 					err = ENOTDIR;
1056 				}
1057 				if (!err && !vnode_mountedhere(*vpp)) {
1058 					err = fuse_internal_access(*vpp, VEXEC, &facp, td, cred);
1059 				}
1060 				if (err) {
1061 					if (tmpvtype == VLNK)
1062 						FS_DEBUG("weird, permission error with a symlink?\n");
1063 					vput(*vpp);
1064 					*vpp = NULL;
1065 				}
1066 			}
1067 #endif
1068 		}
1069 	}
1070 	fdisp_destroy(&fdi);
1071 
1072 	return err;
1073 }
1074 
1075 /*
1076     struct vnop_mkdir_args {
1077 	struct vnode *a_dvp;
1078 	struct vnode **a_vpp;
1079 	struct componentname *a_cnp;
1080 	struct vattr *a_vap;
1081     };
1082 */
1083 static int
1084 fuse_vnop_mkdir(struct vop_mkdir_args *ap)
1085 {
1086 	struct vnode *dvp = ap->a_dvp;
1087 	struct vnode **vpp = ap->a_vpp;
1088 	struct componentname *cnp = ap->a_cnp;
1089 	struct vattr *vap = ap->a_vap;
1090 
1091 	struct fuse_mkdir_in fmdi;
1092 
1093 	fuse_trace_printf_vnop();
1094 
1095 	if (fuse_isdeadfs(dvp)) {
1096 		return ENXIO;
1097 	}
1098 	fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode);
1099 
1100 	return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi,
1101 	    sizeof(fmdi), VDIR));
1102 }
1103 
1104 /*
1105     struct vnop_mknod_args {
1106 	struct vnode *a_dvp;
1107 	struct vnode **a_vpp;
1108 	struct componentname *a_cnp;
1109 	struct vattr *a_vap;
1110     };
1111 */
1112 static int
1113 fuse_vnop_mknod(struct vop_mknod_args *ap)
1114 {
1115 
1116 	return (EINVAL);
1117 }
1118 
1119 
1120 /*
1121     struct vnop_open_args {
1122 	struct vnode *a_vp;
1123 	int  a_mode;
1124 	struct ucred *a_cred;
1125 	struct thread *a_td;
1126 	int a_fdidx; / struct file *a_fp;
1127     };
1128 */
1129 static int
1130 fuse_vnop_open(struct vop_open_args *ap)
1131 {
1132 	struct vnode *vp = ap->a_vp;
1133 	int mode = ap->a_mode;
1134 	struct thread *td = ap->a_td;
1135 	struct ucred *cred = ap->a_cred;
1136 
1137 	fufh_type_t fufh_type;
1138 	struct fuse_vnode_data *fvdat;
1139 
1140 	int error, isdir = 0;
1141 	int32_t fuse_open_flags;
1142 
1143 	FS_DEBUG2G("inode=%ju mode=0x%x\n", (uintmax_t)VTOI(vp), mode);
1144 
1145 	if (fuse_isdeadfs(vp)) {
1146 		return ENXIO;
1147 	}
1148 	fvdat = VTOFUD(vp);
1149 
1150 	if (vnode_isdir(vp)) {
1151 		isdir = 1;
1152 	}
1153 	fuse_open_flags = 0;
1154 	if (isdir) {
1155 		fufh_type = FUFH_RDONLY;
1156 	} else {
1157 		fufh_type = fuse_filehandle_xlate_from_fflags(mode);
1158 		/*
1159 		 * For WRONLY opens, force DIRECT_IO.  This is necessary
1160 		 * since writing a partial block through the buffer cache
1161 		 * will result in a read of the block and that read won't
1162 		 * be allowed by the WRONLY open.
1163 		 */
1164 		if (fufh_type == FUFH_WRONLY ||
1165 		    (fvdat->flag & FN_DIRECTIO) != 0)
1166 			fuse_open_flags = FOPEN_DIRECT_IO;
1167 	}
1168 
1169 	if (fuse_filehandle_validrw(vp, fufh_type) != FUFH_INVALID) {
1170 		fuse_vnode_open(vp, fuse_open_flags, td);
1171 		return 0;
1172 	}
1173 	error = fuse_filehandle_open(vp, fufh_type, NULL, td, cred);
1174 
1175 	return error;
1176 }
1177 
1178 /*
1179     struct vnop_read_args {
1180 	struct vnode *a_vp;
1181 	struct uio *a_uio;
1182 	int  a_ioflag;
1183 	struct ucred *a_cred;
1184     };
1185 */
1186 static int
1187 fuse_vnop_read(struct vop_read_args *ap)
1188 {
1189 	struct vnode *vp = ap->a_vp;
1190 	struct uio *uio = ap->a_uio;
1191 	int ioflag = ap->a_ioflag;
1192 	struct ucred *cred = ap->a_cred;
1193 
1194 	FS_DEBUG2G("inode=%ju offset=%jd resid=%zd\n",
1195 	    (uintmax_t)VTOI(vp), uio->uio_offset, uio->uio_resid);
1196 
1197 	if (fuse_isdeadfs(vp)) {
1198 		return ENXIO;
1199 	}
1200 
1201 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1202 		ioflag |= IO_DIRECT;
1203 	}
1204 
1205 	return fuse_io_dispatch(vp, uio, ioflag, cred);
1206 }
1207 
1208 /*
1209     struct vnop_readdir_args {
1210 	struct vnode *a_vp;
1211 	struct uio *a_uio;
1212 	struct ucred *a_cred;
1213 	int *a_eofflag;
1214 	int *ncookies;
1215 	u_long **a_cookies;
1216     };
1217 */
1218 static int
1219 fuse_vnop_readdir(struct vop_readdir_args *ap)
1220 {
1221 	struct vnode *vp = ap->a_vp;
1222 	struct uio *uio = ap->a_uio;
1223 	struct ucred *cred = ap->a_cred;
1224 
1225 	struct fuse_filehandle *fufh = NULL;
1226 	struct fuse_vnode_data *fvdat;
1227 	struct fuse_iov cookediov;
1228 
1229 	int err = 0;
1230 	int freefufh = 0;
1231 
1232 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
1233 
1234 	if (fuse_isdeadfs(vp)) {
1235 		return ENXIO;
1236 	}
1237 	if (				/* XXXIP ((uio_iovcnt(uio) > 1)) || */
1238 	    (uio_resid(uio) < sizeof(struct dirent))) {
1239 		return EINVAL;
1240 	}
1241 	fvdat = VTOFUD(vp);
1242 
1243 	if (!fuse_filehandle_valid(vp, FUFH_RDONLY)) {
1244 		FS_DEBUG("calling readdir() before open()");
1245 		err = fuse_filehandle_open(vp, FUFH_RDONLY, &fufh, NULL, cred);
1246 		freefufh = 1;
1247 	} else {
1248 		err = fuse_filehandle_get(vp, FUFH_RDONLY, &fufh);
1249 	}
1250 	if (err) {
1251 		return (err);
1252 	}
1253 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1254 	fiov_init(&cookediov, DIRCOOKEDSIZE);
1255 
1256 	err = fuse_internal_readdir(vp, uio, fufh, &cookediov);
1257 
1258 	fiov_teardown(&cookediov);
1259 	if (freefufh) {
1260 		fuse_filehandle_close(vp, FUFH_RDONLY, NULL, cred);
1261 	}
1262 	return err;
1263 }
1264 
1265 /*
1266     struct vnop_readlink_args {
1267 	struct vnode *a_vp;
1268 	struct uio *a_uio;
1269 	struct ucred *a_cred;
1270     };
1271 */
1272 static int
1273 fuse_vnop_readlink(struct vop_readlink_args *ap)
1274 {
1275 	struct vnode *vp = ap->a_vp;
1276 	struct uio *uio = ap->a_uio;
1277 	struct ucred *cred = ap->a_cred;
1278 
1279 	struct fuse_dispatcher fdi;
1280 	int err;
1281 
1282 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
1283 
1284 	if (fuse_isdeadfs(vp)) {
1285 		return ENXIO;
1286 	}
1287 	if (!vnode_islnk(vp)) {
1288 		return EINVAL;
1289 	}
1290 	fdisp_init(&fdi, 0);
1291 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1292 	if (err) {
1293 		goto out;
1294 	}
1295 	if (((char *)fdi.answ)[0] == '/' &&
1296 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1297 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1298 
1299 		err = uiomove(mpth, strlen(mpth), uio);
1300 	}
1301 	if (!err) {
1302 		err = uiomove(fdi.answ, fdi.iosize, uio);
1303 	}
1304 out:
1305 	fdisp_destroy(&fdi);
1306 	return err;
1307 }
1308 
1309 /*
1310     struct vnop_reclaim_args {
1311 	struct vnode *a_vp;
1312 	struct thread *a_td;
1313     };
1314 */
1315 static int
1316 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1317 {
1318 	struct vnode *vp = ap->a_vp;
1319 	struct thread *td = ap->a_td;
1320 
1321 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1322 	struct fuse_filehandle *fufh = NULL;
1323 
1324 	int type;
1325 
1326 	if (!fvdat) {
1327 		panic("FUSE: no vnode data during recycling");
1328 	}
1329 	FS_DEBUG("inode=%ju\n", (uintmax_t)VTOI(vp));
1330 
1331 	for (type = 0; type < FUFH_MAXTYPE; type++) {
1332 		fufh = &(fvdat->fufh[type]);
1333 		if (FUFH_IS_VALID(fufh)) {
1334 			printf("FUSE: vnode being reclaimed but fufh (type=%d) is valid",
1335 			    type);
1336 			fuse_filehandle_close(vp, type, td, NULL);
1337 		}
1338 	}
1339 
1340 	if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) {
1341 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1342 		    fvdat->nlookup);
1343 	}
1344 	fuse_vnode_setparent(vp, NULL);
1345 	cache_purge(vp);
1346 	vfs_hash_remove(vp);
1347 	vnode_destroy_vobject(vp);
1348 	fuse_vnode_destroy(vp);
1349 
1350 	return 0;
1351 }
1352 
1353 /*
1354     struct vnop_remove_args {
1355 	struct vnode *a_dvp;
1356 	struct vnode *a_vp;
1357 	struct componentname *a_cnp;
1358     };
1359 */
1360 static int
1361 fuse_vnop_remove(struct vop_remove_args *ap)
1362 {
1363 	struct vnode *dvp = ap->a_dvp;
1364 	struct vnode *vp = ap->a_vp;
1365 	struct componentname *cnp = ap->a_cnp;
1366 
1367 	int err;
1368 
1369 	FS_DEBUG2G("inode=%ju name=%*s\n",
1370 	    (uintmax_t)VTOI(vp), (int)cnp->cn_namelen, cnp->cn_nameptr);
1371 
1372 	if (fuse_isdeadfs(vp)) {
1373 		return ENXIO;
1374 	}
1375 	if (vnode_isdir(vp)) {
1376 		return EPERM;
1377 	}
1378 	cache_purge(vp);
1379 
1380 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1381 
1382 	if (err == 0)
1383 		fuse_internal_vnode_disappear(vp);
1384 	return err;
1385 }
1386 
1387 /*
1388     struct vnop_rename_args {
1389 	struct vnode *a_fdvp;
1390 	struct vnode *a_fvp;
1391 	struct componentname *a_fcnp;
1392 	struct vnode *a_tdvp;
1393 	struct vnode *a_tvp;
1394 	struct componentname *a_tcnp;
1395     };
1396 */
1397 static int
1398 fuse_vnop_rename(struct vop_rename_args *ap)
1399 {
1400 	struct vnode *fdvp = ap->a_fdvp;
1401 	struct vnode *fvp = ap->a_fvp;
1402 	struct componentname *fcnp = ap->a_fcnp;
1403 	struct vnode *tdvp = ap->a_tdvp;
1404 	struct vnode *tvp = ap->a_tvp;
1405 	struct componentname *tcnp = ap->a_tcnp;
1406 	struct fuse_data *data;
1407 
1408 	int err = 0;
1409 
1410 	FS_DEBUG2G("from: inode=%ju name=%*s -> to: inode=%ju name=%*s\n",
1411 	    (uintmax_t)VTOI(fvp), (int)fcnp->cn_namelen, fcnp->cn_nameptr,
1412 	    (uintmax_t)(tvp == NULL ? -1 : VTOI(tvp)),
1413 	    (int)tcnp->cn_namelen, tcnp->cn_nameptr);
1414 
1415 	if (fuse_isdeadfs(fdvp)) {
1416 		return ENXIO;
1417 	}
1418 	if (fvp->v_mount != tdvp->v_mount ||
1419 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1420 		FS_DEBUG("cross-device rename: %s -> %s\n",
1421 		    fcnp->cn_nameptr, (tcnp != NULL ? tcnp->cn_nameptr : "(NULL)"));
1422 		err = EXDEV;
1423 		goto out;
1424 	}
1425 	cache_purge(fvp);
1426 
1427 	/*
1428          * FUSE library is expected to check if target directory is not
1429          * under the source directory in the file system tree.
1430          * Linux performs this check at VFS level.
1431          */
1432 	data = fuse_get_mpdata(vnode_mount(tdvp));
1433 	sx_xlock(&data->rename_lock);
1434 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1435 	if (err == 0) {
1436 		if (tdvp != fdvp)
1437 			fuse_vnode_setparent(fvp, tdvp);
1438 		if (tvp != NULL)
1439 			fuse_vnode_setparent(tvp, NULL);
1440 	}
1441 	sx_unlock(&data->rename_lock);
1442 
1443 	if (tvp != NULL && tvp != fvp) {
1444 		cache_purge(tvp);
1445 	}
1446 	if (vnode_isdir(fvp)) {
1447 		if ((tvp != NULL) && vnode_isdir(tvp)) {
1448 			cache_purge(tdvp);
1449 		}
1450 		cache_purge(fdvp);
1451 	}
1452 out:
1453 	if (tdvp == tvp) {
1454 		vrele(tdvp);
1455 	} else {
1456 		vput(tdvp);
1457 	}
1458 	if (tvp != NULL) {
1459 		vput(tvp);
1460 	}
1461 	vrele(fdvp);
1462 	vrele(fvp);
1463 
1464 	return err;
1465 }
1466 
1467 /*
1468     struct vnop_rmdir_args {
1469 	    struct vnode *a_dvp;
1470 	    struct vnode *a_vp;
1471 	    struct componentname *a_cnp;
1472     } *ap;
1473 */
1474 static int
1475 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1476 {
1477 	struct vnode *dvp = ap->a_dvp;
1478 	struct vnode *vp = ap->a_vp;
1479 
1480 	int err;
1481 
1482 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
1483 
1484 	if (fuse_isdeadfs(vp)) {
1485 		return ENXIO;
1486 	}
1487 	if (VTOFUD(vp) == VTOFUD(dvp)) {
1488 		return EINVAL;
1489 	}
1490 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1491 
1492 	if (err == 0)
1493 		fuse_internal_vnode_disappear(vp);
1494 	return err;
1495 }
1496 
1497 /*
1498     struct vnop_setattr_args {
1499 	struct vnode *a_vp;
1500 	struct vattr *a_vap;
1501 	struct ucred *a_cred;
1502 	struct thread *a_td;
1503     };
1504 */
1505 static int
1506 fuse_vnop_setattr(struct vop_setattr_args *ap)
1507 {
1508 	struct vnode *vp = ap->a_vp;
1509 	struct vattr *vap = ap->a_vap;
1510 	struct ucred *cred = ap->a_cred;
1511 	struct thread *td = curthread;
1512 
1513 	struct fuse_dispatcher fdi;
1514 	struct fuse_setattr_in *fsai;
1515 	struct fuse_access_param facp;
1516 
1517 	int err = 0;
1518 	enum vtype vtyp;
1519 	int sizechanged = 0;
1520 	uint64_t newsize = 0;
1521 
1522 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
1523 
1524 	if (fuse_isdeadfs(vp)) {
1525 		return ENXIO;
1526 	}
1527 	fdisp_init(&fdi, sizeof(*fsai));
1528 	fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
1529 	fsai = fdi.indata;
1530 	fsai->valid = 0;
1531 
1532 	bzero(&facp, sizeof(facp));
1533 
1534 	facp.xuid = vap->va_uid;
1535 	facp.xgid = vap->va_gid;
1536 
1537 	if (vap->va_uid != (uid_t)VNOVAL) {
1538 		facp.facc_flags |= FACCESS_CHOWN;
1539 		fsai->uid = vap->va_uid;
1540 		fsai->valid |= FATTR_UID;
1541 	}
1542 	if (vap->va_gid != (gid_t)VNOVAL) {
1543 		facp.facc_flags |= FACCESS_CHOWN;
1544 		fsai->gid = vap->va_gid;
1545 		fsai->valid |= FATTR_GID;
1546 	}
1547 	if (vap->va_size != VNOVAL) {
1548 
1549 		struct fuse_filehandle *fufh = NULL;
1550 
1551 		/*Truncate to a new value. */
1552 		    fsai->size = vap->va_size;
1553 		sizechanged = 1;
1554 		newsize = vap->va_size;
1555 		fsai->valid |= FATTR_SIZE;
1556 
1557 		fuse_filehandle_getrw(vp, FUFH_WRONLY, &fufh);
1558 		if (fufh) {
1559 			fsai->fh = fufh->fh_id;
1560 			fsai->valid |= FATTR_FH;
1561 		}
1562 	}
1563 	if (vap->va_atime.tv_sec != VNOVAL) {
1564 		fsai->atime = vap->va_atime.tv_sec;
1565 		fsai->atimensec = vap->va_atime.tv_nsec;
1566 		fsai->valid |= FATTR_ATIME;
1567 	}
1568 	if (vap->va_mtime.tv_sec != VNOVAL) {
1569 		fsai->mtime = vap->va_mtime.tv_sec;
1570 		fsai->mtimensec = vap->va_mtime.tv_nsec;
1571 		fsai->valid |= FATTR_MTIME;
1572 	}
1573 	if (vap->va_mode != (mode_t)VNOVAL) {
1574 		fsai->mode = vap->va_mode & ALLPERMS;
1575 		fsai->valid |= FATTR_MODE;
1576 	}
1577 	if (!fsai->valid) {
1578 		goto out;
1579 	}
1580 	vtyp = vnode_vtype(vp);
1581 
1582 	if (fsai->valid & FATTR_SIZE && vtyp == VDIR) {
1583 		err = EISDIR;
1584 		goto out;
1585 	}
1586 	if (vfs_isrdonly(vnode_mount(vp)) && (fsai->valid & ~FATTR_SIZE || vtyp == VREG)) {
1587 		err = EROFS;
1588 		goto out;
1589 	}
1590 	if (fsai->valid & ~FATTR_SIZE) {
1591 	  /*err = fuse_internal_access(vp, VADMIN, context, &facp); */
1592 	  /*XXX */
1593 		    err = 0;
1594 	}
1595 	facp.facc_flags &= ~FACCESS_XQUERIES;
1596 
1597 	if (err && !(fsai->valid & ~(FATTR_ATIME | FATTR_MTIME)) &&
1598 	    vap->va_vaflags & VA_UTIMES_NULL) {
1599 		err = fuse_internal_access(vp, VWRITE, &facp, td, cred);
1600 	}
1601 	if (err)
1602 		goto out;
1603 	if ((err = fdisp_wait_answ(&fdi)))
1604 		goto out;
1605 	vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode);
1606 
1607 	if (vnode_vtype(vp) != vtyp) {
1608 		if (vnode_vtype(vp) == VNON && vtyp != VNON) {
1609 			debug_printf("FUSE: Dang! vnode_vtype is VNON and vtype isn't.\n");
1610 		} else {
1611 			/*
1612 	                 * STALE vnode, ditch
1613 	                 *
1614 	                 * The vnode has changed its type "behind our back". There's
1615 	                 * nothing really we can do, so let us just force an internal
1616 	                 * revocation and tell the caller to try again, if interested.
1617 	                 */
1618 			fuse_internal_vnode_disappear(vp);
1619 			err = EAGAIN;
1620 		}
1621 	}
1622 	if (!err && !sizechanged) {
1623 		cache_attrs(vp, (struct fuse_attr_out *)fdi.answ);
1624 	}
1625 out:
1626 	fdisp_destroy(&fdi);
1627 	if (!err && sizechanged) {
1628 		fuse_vnode_setsize(vp, cred, newsize);
1629 		VTOFUD(vp)->flag &= ~FN_SIZECHANGE;
1630 	}
1631 	return err;
1632 }
1633 
1634 /*
1635     struct vnop_strategy_args {
1636 	struct vnode *a_vp;
1637 	struct buf *a_bp;
1638     };
1639 */
1640 static int
1641 fuse_vnop_strategy(struct vop_strategy_args *ap)
1642 {
1643 	struct vnode *vp = ap->a_vp;
1644 	struct buf *bp = ap->a_bp;
1645 
1646 	fuse_trace_printf_vnop();
1647 
1648 	if (!vp || fuse_isdeadfs(vp)) {
1649 		bp->b_ioflags |= BIO_ERROR;
1650 		bp->b_error = ENXIO;
1651 		bufdone(bp);
1652 		return ENXIO;
1653 	}
1654 	if (bp->b_iocmd == BIO_WRITE)
1655 		fuse_vnode_refreshsize(vp, NOCRED);
1656 
1657 	(void)fuse_io_strategy(vp, bp);
1658 
1659 	/*
1660 	 * This is a dangerous function. If returns error, that might mean a
1661 	 * panic. We prefer pretty much anything over being forced to panic
1662 	 * by a malicious daemon (a demon?). So we just return 0 anyway. You
1663 	 * should never mind this: this function has its own error
1664 	 * propagation mechanism via the argument buffer, so
1665 	 * not-that-melodramatic residents of the call chain still will be
1666 	 * able to know what to do.
1667 	 */
1668 	return 0;
1669 }
1670 
1671 
1672 /*
1673     struct vnop_symlink_args {
1674 	struct vnode *a_dvp;
1675 	struct vnode **a_vpp;
1676 	struct componentname *a_cnp;
1677 	struct vattr *a_vap;
1678 	char *a_target;
1679     };
1680 */
1681 static int
1682 fuse_vnop_symlink(struct vop_symlink_args *ap)
1683 {
1684 	struct vnode *dvp = ap->a_dvp;
1685 	struct vnode **vpp = ap->a_vpp;
1686 	struct componentname *cnp = ap->a_cnp;
1687 	char *target = ap->a_target;
1688 
1689 	struct fuse_dispatcher fdi;
1690 
1691 	int err;
1692 	size_t len;
1693 
1694 	FS_DEBUG2G("inode=%ju name=%*s\n",
1695 	    (uintmax_t)VTOI(dvp), (int)cnp->cn_namelen, cnp->cn_nameptr);
1696 
1697 	if (fuse_isdeadfs(dvp)) {
1698 		return ENXIO;
1699 	}
1700 	/*
1701          * Unlike the other creator type calls, here we have to create a message
1702          * where the name of the new entry comes first, and the data describing
1703          * the entry comes second.
1704          * Hence we can't rely on our handy fuse_internal_newentry() routine,
1705          * but put together the message manually and just call the core part.
1706          */
1707 
1708 	len = strlen(target) + 1;
1709 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1710 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1711 
1712 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1713 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1714 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1715 
1716 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1717 	fdisp_destroy(&fdi);
1718 	return err;
1719 }
1720 
1721 /*
1722     struct vnop_write_args {
1723 	struct vnode *a_vp;
1724 	struct uio *a_uio;
1725 	int  a_ioflag;
1726 	struct ucred *a_cred;
1727     };
1728 */
1729 static int
1730 fuse_vnop_write(struct vop_write_args *ap)
1731 {
1732 	struct vnode *vp = ap->a_vp;
1733 	struct uio *uio = ap->a_uio;
1734 	int ioflag = ap->a_ioflag;
1735 	struct ucred *cred = ap->a_cred;
1736 
1737 	fuse_trace_printf_vnop();
1738 
1739 	if (fuse_isdeadfs(vp)) {
1740 		return ENXIO;
1741 	}
1742 	fuse_vnode_refreshsize(vp, cred);
1743 
1744 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1745 		ioflag |= IO_DIRECT;
1746 	}
1747 
1748 	return fuse_io_dispatch(vp, uio, ioflag, cred);
1749 }
1750 
1751 /*
1752     struct vnop_getpages_args {
1753         struct vnode *a_vp;
1754         vm_page_t *a_m;
1755         int a_count;
1756         int a_reqpage;
1757     };
1758 */
1759 static int
1760 fuse_vnop_getpages(struct vop_getpages_args *ap)
1761 {
1762 	int i, error, nextoff, size, toff, count, npages;
1763 	struct uio uio;
1764 	struct iovec iov;
1765 	vm_offset_t kva;
1766 	struct buf *bp;
1767 	struct vnode *vp;
1768 	struct thread *td;
1769 	struct ucred *cred;
1770 	vm_page_t *pages;
1771 
1772 	FS_DEBUG2G("heh\n");
1773 
1774 	vp = ap->a_vp;
1775 	KASSERT(vp->v_object, ("objectless vp passed to getpages"));
1776 	td = curthread;			/* XXX */
1777 	cred = curthread->td_ucred;	/* XXX */
1778 	pages = ap->a_m;
1779 	npages = ap->a_count;
1780 
1781 	if (!fsess_opt_mmap(vnode_mount(vp))) {
1782 		FS_DEBUG("called on non-cacheable vnode??\n");
1783 		return (VM_PAGER_ERROR);
1784 	}
1785 
1786 	/*
1787 	 * If the last page is partially valid, just return it and allow
1788 	 * the pager to zero-out the blanks.  Partially valid pages can
1789 	 * only occur at the file EOF.
1790 	 *
1791 	 * XXXGL: is that true for FUSE, which is a local filesystem,
1792 	 * but still somewhat disconnected from the kernel?
1793 	 */
1794 	VM_OBJECT_WLOCK(vp->v_object);
1795 	if (pages[npages - 1]->valid != 0 && --npages == 0)
1796 		goto out;
1797 	VM_OBJECT_WUNLOCK(vp->v_object);
1798 
1799 	/*
1800 	 * We use only the kva address for the buffer, but this is extremely
1801 	 * convenient and fast.
1802 	 */
1803 	bp = getpbuf(&fuse_pbuf_freecnt);
1804 
1805 	kva = (vm_offset_t)bp->b_data;
1806 	pmap_qenter(kva, pages, npages);
1807 	VM_CNT_INC(v_vnodein);
1808 	VM_CNT_ADD(v_vnodepgsin, npages);
1809 
1810 	count = npages << PAGE_SHIFT;
1811 	iov.iov_base = (caddr_t)kva;
1812 	iov.iov_len = count;
1813 	uio.uio_iov = &iov;
1814 	uio.uio_iovcnt = 1;
1815 	uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
1816 	uio.uio_resid = count;
1817 	uio.uio_segflg = UIO_SYSSPACE;
1818 	uio.uio_rw = UIO_READ;
1819 	uio.uio_td = td;
1820 
1821 	error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred);
1822 	pmap_qremove(kva, npages);
1823 
1824 	relpbuf(bp, &fuse_pbuf_freecnt);
1825 
1826 	if (error && (uio.uio_resid == count)) {
1827 		FS_DEBUG("error %d\n", error);
1828 		return VM_PAGER_ERROR;
1829 	}
1830 	/*
1831 	 * Calculate the number of bytes read and validate only that number
1832 	 * of bytes.  Note that due to pending writes, size may be 0.  This
1833 	 * does not mean that the remaining data is invalid!
1834 	 */
1835 
1836 	size = count - uio.uio_resid;
1837 	VM_OBJECT_WLOCK(vp->v_object);
1838 	fuse_vm_page_lock_queues();
1839 	for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
1840 		vm_page_t m;
1841 
1842 		nextoff = toff + PAGE_SIZE;
1843 		m = pages[i];
1844 
1845 		if (nextoff <= size) {
1846 			/*
1847 			 * Read operation filled an entire page
1848 			 */
1849 			m->valid = VM_PAGE_BITS_ALL;
1850 			KASSERT(m->dirty == 0,
1851 			    ("fuse_getpages: page %p is dirty", m));
1852 		} else if (size > toff) {
1853 			/*
1854 			 * Read operation filled a partial page.
1855 			 */
1856 			m->valid = 0;
1857 			vm_page_set_valid_range(m, 0, size - toff);
1858 			KASSERT(m->dirty == 0,
1859 			    ("fuse_getpages: page %p is dirty", m));
1860 		} else {
1861 			/*
1862 			 * Read operation was short.  If no error occurred
1863 			 * we may have hit a zero-fill section.   We simply
1864 			 * leave valid set to 0.
1865 			 */
1866 			;
1867 		}
1868 	}
1869 	fuse_vm_page_unlock_queues();
1870 out:
1871 	VM_OBJECT_WUNLOCK(vp->v_object);
1872 	if (ap->a_rbehind)
1873 		*ap->a_rbehind = 0;
1874 	if (ap->a_rahead)
1875 		*ap->a_rahead = 0;
1876 	return (VM_PAGER_OK);
1877 }
1878 
1879 /*
1880     struct vnop_putpages_args {
1881         struct vnode *a_vp;
1882         vm_page_t *a_m;
1883         int a_count;
1884         int a_sync;
1885         int *a_rtvals;
1886         vm_ooffset_t a_offset;
1887     };
1888 */
1889 static int
1890 fuse_vnop_putpages(struct vop_putpages_args *ap)
1891 {
1892 	struct uio uio;
1893 	struct iovec iov;
1894 	vm_offset_t kva;
1895 	struct buf *bp;
1896 	int i, error, npages, count;
1897 	off_t offset;
1898 	int *rtvals;
1899 	struct vnode *vp;
1900 	struct thread *td;
1901 	struct ucred *cred;
1902 	vm_page_t *pages;
1903 	vm_ooffset_t fsize;
1904 
1905 	FS_DEBUG2G("heh\n");
1906 
1907 	vp = ap->a_vp;
1908 	KASSERT(vp->v_object, ("objectless vp passed to putpages"));
1909 	fsize = vp->v_object->un_pager.vnp.vnp_size;
1910 	td = curthread;			/* XXX */
1911 	cred = curthread->td_ucred;	/* XXX */
1912 	pages = ap->a_m;
1913 	count = ap->a_count;
1914 	rtvals = ap->a_rtvals;
1915 	npages = btoc(count);
1916 	offset = IDX_TO_OFF(pages[0]->pindex);
1917 
1918 	if (!fsess_opt_mmap(vnode_mount(vp))) {
1919 		FS_DEBUG("called on non-cacheable vnode??\n");
1920 	}
1921 	for (i = 0; i < npages; i++)
1922 		rtvals[i] = VM_PAGER_AGAIN;
1923 
1924 	/*
1925 	 * When putting pages, do not extend file past EOF.
1926 	 */
1927 
1928 	if (offset + count > fsize) {
1929 		count = fsize - offset;
1930 		if (count < 0)
1931 			count = 0;
1932 	}
1933 	/*
1934 	 * We use only the kva address for the buffer, but this is extremely
1935 	 * convenient and fast.
1936 	 */
1937 	bp = getpbuf(&fuse_pbuf_freecnt);
1938 
1939 	kva = (vm_offset_t)bp->b_data;
1940 	pmap_qenter(kva, pages, npages);
1941 	VM_CNT_INC(v_vnodeout);
1942 	VM_CNT_ADD(v_vnodepgsout, count);
1943 
1944 	iov.iov_base = (caddr_t)kva;
1945 	iov.iov_len = count;
1946 	uio.uio_iov = &iov;
1947 	uio.uio_iovcnt = 1;
1948 	uio.uio_offset = offset;
1949 	uio.uio_resid = count;
1950 	uio.uio_segflg = UIO_SYSSPACE;
1951 	uio.uio_rw = UIO_WRITE;
1952 	uio.uio_td = td;
1953 
1954 	error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred);
1955 
1956 	pmap_qremove(kva, npages);
1957 	relpbuf(bp, &fuse_pbuf_freecnt);
1958 
1959 	if (!error) {
1960 		int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE;
1961 
1962 		for (i = 0; i < nwritten; i++) {
1963 			rtvals[i] = VM_PAGER_OK;
1964 			VM_OBJECT_WLOCK(pages[i]->object);
1965 			vm_page_undirty(pages[i]);
1966 			VM_OBJECT_WUNLOCK(pages[i]->object);
1967 		}
1968 	}
1969 	return rtvals[0];
1970 }
1971 
1972 static const char extattr_namespace_separator = '.';
1973 
1974 /*
1975     struct vop_getextattr_args {
1976         struct vop_generic_args a_gen;
1977         struct vnode *a_vp;
1978         int a_attrnamespace;
1979         const char *a_name;
1980         struct uio *a_uio;
1981         size_t *a_size;
1982         struct ucred *a_cred;
1983         struct thread *a_td;
1984     };
1985 */
1986 static int
1987 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
1988 {
1989 	struct vnode *vp = ap->a_vp;
1990 	struct uio *uio = ap->a_uio;
1991 	struct fuse_dispatcher fdi = {0};
1992 	struct fuse_getxattr_in *get_xattr_in;
1993 	struct fuse_getxattr_out *get_xattr_out;
1994 	struct mount *mp = vnode_mount(vp);
1995 	char *prefix;
1996 	size_t len;
1997 	char *attr_str;
1998 	struct thread *td = ap->a_td;
1999 	struct ucred *cred = ap->a_cred;
2000 	int err = 0;
2001 
2002 	fuse_trace_printf_vnop();
2003 
2004 	if (fuse_isdeadfs(vp))
2005 		return ENXIO;
2006 
2007 	/* Default to looking for user attributes. */
2008 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2009 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2010 	else
2011 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2012 
2013 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2014 	    strlen(ap->a_name) + 1;
2015 
2016 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2017 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2018 
2019 	get_xattr_in = fdi.indata;
2020 	/*
2021 	 * Check to see whether we're querying the available size or
2022 	 * issuing the actual request.  If we pass in 0, we get back struct
2023 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2024 	 * that much data, without the struct fuse_getxattr_out header.
2025 	 */
2026 	if (ap->a_size != NULL)
2027 		get_xattr_in->size = 0;
2028 	else
2029 		get_xattr_in->size = uio->uio_resid;
2030 
2031 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2032 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2033 	    ap->a_name);
2034 
2035 	err = fdisp_wait_answ(&fdi);
2036 
2037 	if (err != 0) {
2038 		if (err == ENOSYS)
2039 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2040 		debug_printf("getxattr: got err=%d from daemon\n", err);
2041 		goto out;
2042 	}
2043 
2044 	/*
2045 	 * If we get to this point (i.e. no error), we should have a valid
2046 	 * answer of some sort.  i.e. non-zero iosize and a valid pointer.
2047 	 */
2048 	if ((fdi.answ == NULL) || (fdi.iosize == 0)) {
2049 		debug_printf("getxattr: err = 0, but answ = %p, iosize = %zu\n",
2050 		    fdi.answ, fdi.iosize);
2051 		err = EINVAL;
2052 		goto out;
2053 	}
2054 	get_xattr_out = fdi.answ;
2055 
2056 	if (ap->a_size != NULL) {
2057 		*ap->a_size = get_xattr_out->size;
2058 	} else if (fdi.iosize > 0) {
2059 		err = uiomove(fdi.answ, fdi.iosize, uio);
2060 	} else {
2061 		err = EINVAL;
2062 	}
2063 
2064 out:
2065 	fdisp_destroy(&fdi);
2066 	return (err);
2067 }
2068 
2069 /*
2070     struct vop_setextattr_args {
2071         struct vop_generic_args a_gen;
2072         struct vnode *a_vp;
2073         int a_attrnamespace;
2074         const char *a_name;
2075         struct uio *a_uio;
2076         struct ucred *a_cred;
2077         struct thread *a_td;
2078     };
2079 */
2080 static int
2081 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2082 {
2083 	struct vnode *vp = ap->a_vp;
2084 	struct uio *uio = ap->a_uio;
2085 	struct fuse_dispatcher fdi = {0};
2086 	struct fuse_setxattr_in *set_xattr_in;
2087 	struct mount *mp = vnode_mount(vp);
2088 	char *prefix;
2089 	size_t len;
2090 	char *attr_str;
2091 	struct thread *td = ap->a_td;
2092 	struct ucred *cred = ap->a_cred;
2093 	int err = 0;
2094 
2095 	fuse_trace_printf_vnop();
2096 
2097 	if (fuse_isdeadfs(vp))
2098 		return ENXIO;
2099 
2100 	/* Default to looking for user attributes. */
2101 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2102 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2103 	else
2104 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2105 
2106 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2107 	    strlen(ap->a_name) + 1;
2108 
2109 	fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2110 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2111 
2112 	set_xattr_in = fdi.indata;
2113 	set_xattr_in->size = uio->uio_resid;
2114 
2115 	attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2116 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2117 	    ap->a_name);
2118 
2119 	err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2120 	    uio->uio_resid, uio);
2121 	if (err != 0) {
2122 		debug_printf("setxattr: got error %d from uiomove\n", err);
2123 		goto out;
2124 	}
2125 
2126 	err = fdisp_wait_answ(&fdi);
2127 
2128 	if (err != 0) {
2129 		if (err == ENOSYS)
2130 			fsess_set_notimpl(mp, FUSE_SETXATTR);
2131 		debug_printf("setxattr: got err=%d from daemon\n", err);
2132 		goto out;
2133 	}
2134 
2135 out:
2136 	fdisp_destroy(&fdi);
2137 	return (err);
2138 }
2139 
2140 /*
2141  * The Linux / FUSE extended attribute list is simply a collection of
2142  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2143  * byte length followed by a non-NUL terminated string.  So, this allows
2144  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2145  * Linux attribute names are reported with the namespace as a prefix (e.g.
2146  * "user.attribute_name"), but in FreeBSD they are reported without the
2147  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2148  *
2149  * user.attr_name1\0user.attr_name2\0
2150  *
2151  * to:
2152  *
2153  * <num>attr_name1<num>attr_name2
2154  *
2155  * Where "<num>" is a single byte number of characters in the attribute name.
2156  *
2157  * Args:
2158  * prefix - exattr namespace prefix string
2159  * list, list_len - input list with namespace prefixes
2160  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2161  */
2162 static int
2163 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2164     char *bsd_list, int *bsd_list_len)
2165 {
2166 	int len, pos, dist_to_next, prefix_len;
2167 
2168 	pos = 0;
2169 	*bsd_list_len = 0;
2170 	prefix_len = strlen(prefix);
2171 
2172 	while (pos < list_len && list[pos] != '\0') {
2173 		dist_to_next = strlen(&list[pos]) + 1;
2174 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2175 		    list[pos + prefix_len] == extattr_namespace_separator) {
2176 			len = dist_to_next -
2177 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2178 			if (len >= EXTATTR_MAXNAMELEN)
2179 				return (ENAMETOOLONG);
2180 
2181 			bsd_list[*bsd_list_len] = len;
2182 			memcpy(&bsd_list[*bsd_list_len + 1],
2183 			    &list[pos + prefix_len +
2184 			    sizeof(extattr_namespace_separator)], len);
2185 
2186 			*bsd_list_len += len + 1;
2187 		}
2188 
2189 		pos += dist_to_next;
2190 	}
2191 
2192 	return (0);
2193 }
2194 
2195 /*
2196     struct vop_listextattr_args {
2197         struct vop_generic_args a_gen;
2198         struct vnode *a_vp;
2199         int a_attrnamespace;
2200         struct uio *a_uio;
2201         size_t *a_size;
2202         struct ucred *a_cred;
2203         struct thread *a_td;
2204     };
2205 */
2206 static int
2207 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2208 {
2209 	struct vnode *vp = ap->a_vp;
2210 	struct uio *uio = ap->a_uio;
2211 	struct fuse_dispatcher fdi = {0};
2212 	struct fuse_getxattr_in *get_xattr_in;
2213 	struct fuse_getxattr_out *get_xattr_out;
2214 	struct mount *mp = vnode_mount(vp);
2215 	size_t len;
2216 	char *prefix;
2217 	char *attr_str;
2218 	char *bsd_list = NULL;
2219 	int bsd_list_len;
2220 	struct thread *td = ap->a_td;
2221 	struct ucred *cred = ap->a_cred;
2222 	int err = 0;
2223 
2224 	fuse_trace_printf_vnop();
2225 
2226 	if (fuse_isdeadfs(vp))
2227 		return ENXIO;
2228 
2229 	/*
2230 	 * Add space for a NUL and the period separator if enabled.
2231 	 * Default to looking for user attributes.
2232 	 */
2233 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2234 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2235 	else
2236 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2237 
2238 	len = strlen(prefix) + sizeof(extattr_namespace_separator) + 1;
2239 
2240 	fdisp_init(&fdi, sizeof(*get_xattr_in) + len);
2241 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2242 
2243 	get_xattr_in = fdi.indata;
2244 	if (ap->a_size != NULL)
2245 		get_xattr_in->size = 0;
2246 	else
2247 		get_xattr_in->size = uio->uio_resid + sizeof(*get_xattr_out);
2248 
2249 
2250 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2251 	snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2252 
2253 	err = fdisp_wait_answ(&fdi);
2254 	if (err != 0) {
2255 		if (err == ENOSYS)
2256 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
2257 		debug_printf("listextattr: got err=%d from daemon\n", err);
2258 		goto out;
2259 	}
2260 
2261 	if ((fdi.answ == NULL) || (fdi.iosize == 0)) {
2262 		err = EINVAL;
2263 		goto out;
2264 	}
2265 	get_xattr_out = fdi.answ;
2266 
2267 	if (ap->a_size != NULL) {
2268 		*ap->a_size = get_xattr_out->size;
2269 	} else if (fdi.iosize > 0) {
2270 		/*
2271 		 * The Linux / FUSE attribute list format isn't the same
2272 		 * as FreeBSD's format.  So we need to transform it into
2273 		 * FreeBSD's format before giving it to the user.
2274 		 */
2275 		bsd_list = malloc(fdi.iosize, M_TEMP, M_WAITOK);
2276 		err = fuse_xattrlist_convert(prefix, fdi.answ, fdi.iosize,
2277 		    bsd_list, &bsd_list_len);
2278 		if (err != 0)
2279 			goto out;
2280 
2281 		err = uiomove(bsd_list, bsd_list_len, uio);
2282 	} else {
2283 		debug_printf("listextattr: returned iosize %zu for %s attribute list is "
2284 		    "too small\n", fdi.iosize, prefix);
2285 		err = EINVAL;
2286 	}
2287 
2288 out:
2289 	free(bsd_list, M_TEMP);
2290 	fdisp_destroy(&fdi);
2291 	return (err);
2292 }
2293 
2294 /*
2295     struct vop_deleteextattr_args {
2296         struct vop_generic_args a_gen;
2297         struct vnode *a_vp;
2298         int a_attrnamespace;
2299         const char *a_name;
2300         struct ucred *a_cred;
2301         struct thread *a_td;
2302     };
2303 */
2304 static int
2305 fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap)
2306 {
2307 	struct vnode *vp = ap->a_vp;
2308 	struct fuse_dispatcher fdi = {0};
2309 	struct mount *mp = vnode_mount(vp);
2310 	char *prefix;
2311 	size_t len;
2312 	char *attr_str;
2313 	struct thread *td = ap->a_td;
2314 	struct ucred *cred = ap->a_cred;
2315 	int err;
2316 
2317 	fuse_trace_printf_vnop();
2318 
2319 	if (fuse_isdeadfs(vp))
2320 		return ENXIO;
2321 
2322 	/* Default to looking for user attributes. */
2323 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2324 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2325 	else
2326 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2327 
2328 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2329 	    strlen(ap->a_name) + 1;
2330 
2331 	fdisp_init(&fdi, len);
2332 	fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred);
2333 
2334 	attr_str = fdi.indata;
2335 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2336 	    ap->a_name);
2337 
2338 	err = fdisp_wait_answ(&fdi);
2339 	if (err != 0) {
2340 		if (err == ENOSYS)
2341 			fsess_set_notimpl(mp, FUSE_REMOVEXATTR);
2342 		debug_printf("removexattr: got err=%d from daemon\n", err);
2343 	}
2344 
2345 	fdisp_destroy(&fdi);
2346 	return (err);
2347 }
2348 
2349 /*
2350     struct vnop_print_args {
2351         struct vnode *a_vp;
2352     };
2353 */
2354 static int
2355 fuse_vnop_print(struct vop_print_args *ap)
2356 {
2357 	struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp);
2358 
2359 	printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n",
2360 	    (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid,
2361 	    (uintmax_t)fvdat->nlookup,
2362 	    fvdat->flag);
2363 
2364 	return 0;
2365 }
2366