xref: /freebsd/sys/fs/fuse/fuse_vnops.c (revision e796cc77c586c2955b2f3940dbf4991b31e8d289)
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_vnode_data *fvdat;
1253 	struct fuse_iov cookediov;
1254 
1255 	int err = 0;
1256 	int freefufh = 0;
1257 
1258 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
1259 
1260 	if (fuse_isdeadfs(vp)) {
1261 		return ENXIO;
1262 	}
1263 	if (				/* XXXIP ((uio_iovcnt(uio) > 1)) || */
1264 	    (uio_resid(uio) < sizeof(struct dirent))) {
1265 		return EINVAL;
1266 	}
1267 	fvdat = VTOFUD(vp);
1268 
1269 	if (!fuse_filehandle_valid(vp, FUFH_RDONLY)) {
1270 		FS_DEBUG("calling readdir() before open()");
1271 		err = fuse_filehandle_open(vp, FUFH_RDONLY, &fufh, NULL, cred);
1272 		freefufh = 1;
1273 	} else {
1274 		err = fuse_filehandle_get(vp, FUFH_RDONLY, &fufh);
1275 	}
1276 	if (err) {
1277 		return (err);
1278 	}
1279 #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1)
1280 	fiov_init(&cookediov, DIRCOOKEDSIZE);
1281 
1282 	err = fuse_internal_readdir(vp, uio, fufh, &cookediov);
1283 
1284 	fiov_teardown(&cookediov);
1285 	if (freefufh) {
1286 		fuse_filehandle_close(vp, FUFH_RDONLY, NULL, cred);
1287 	}
1288 	return err;
1289 }
1290 
1291 /*
1292     struct vnop_readlink_args {
1293 	struct vnode *a_vp;
1294 	struct uio *a_uio;
1295 	struct ucred *a_cred;
1296     };
1297 */
1298 static int
1299 fuse_vnop_readlink(struct vop_readlink_args *ap)
1300 {
1301 	struct vnode *vp = ap->a_vp;
1302 	struct uio *uio = ap->a_uio;
1303 	struct ucred *cred = ap->a_cred;
1304 
1305 	struct fuse_dispatcher fdi;
1306 	int err;
1307 
1308 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
1309 
1310 	if (fuse_isdeadfs(vp)) {
1311 		return ENXIO;
1312 	}
1313 	if (!vnode_islnk(vp)) {
1314 		return EINVAL;
1315 	}
1316 	fdisp_init(&fdi, 0);
1317 	err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred);
1318 	if (err) {
1319 		goto out;
1320 	}
1321 	if (((char *)fdi.answ)[0] == '/' &&
1322 	    fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) {
1323 		char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname;
1324 
1325 		err = uiomove(mpth, strlen(mpth), uio);
1326 	}
1327 	if (!err) {
1328 		err = uiomove(fdi.answ, fdi.iosize, uio);
1329 	}
1330 out:
1331 	fdisp_destroy(&fdi);
1332 	return err;
1333 }
1334 
1335 /*
1336     struct vnop_reclaim_args {
1337 	struct vnode *a_vp;
1338 	struct thread *a_td;
1339     };
1340 */
1341 static int
1342 fuse_vnop_reclaim(struct vop_reclaim_args *ap)
1343 {
1344 	struct vnode *vp = ap->a_vp;
1345 	struct thread *td = ap->a_td;
1346 
1347 	struct fuse_vnode_data *fvdat = VTOFUD(vp);
1348 	struct fuse_filehandle *fufh = NULL;
1349 
1350 	int type;
1351 
1352 	if (!fvdat) {
1353 		panic("FUSE: no vnode data during recycling");
1354 	}
1355 	FS_DEBUG("inode=%ju\n", (uintmax_t)VTOI(vp));
1356 
1357 	for (type = 0; type < FUFH_MAXTYPE; type++) {
1358 		fufh = &(fvdat->fufh[type]);
1359 		if (FUFH_IS_VALID(fufh)) {
1360 			printf("FUSE: vnode being reclaimed but fufh (type=%d) is valid",
1361 			    type);
1362 			fuse_filehandle_close(vp, type, td, NULL);
1363 		}
1364 	}
1365 
1366 	if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) {
1367 		fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp),
1368 		    fvdat->nlookup);
1369 	}
1370 	fuse_vnode_setparent(vp, NULL);
1371 	cache_purge(vp);
1372 	vfs_hash_remove(vp);
1373 	vnode_destroy_vobject(vp);
1374 	fuse_vnode_destroy(vp);
1375 
1376 	return 0;
1377 }
1378 
1379 /*
1380     struct vnop_remove_args {
1381 	struct vnode *a_dvp;
1382 	struct vnode *a_vp;
1383 	struct componentname *a_cnp;
1384     };
1385 */
1386 static int
1387 fuse_vnop_remove(struct vop_remove_args *ap)
1388 {
1389 	struct vnode *dvp = ap->a_dvp;
1390 	struct vnode *vp = ap->a_vp;
1391 	struct componentname *cnp = ap->a_cnp;
1392 
1393 	int err;
1394 
1395 	FS_DEBUG2G("inode=%ju name=%*s\n",
1396 	    (uintmax_t)VTOI(vp), (int)cnp->cn_namelen, cnp->cn_nameptr);
1397 
1398 	if (fuse_isdeadfs(vp)) {
1399 		return ENXIO;
1400 	}
1401 	if (vnode_isdir(vp)) {
1402 		return EPERM;
1403 	}
1404 	cache_purge(vp);
1405 
1406 	err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK);
1407 
1408 	if (err == 0)
1409 		fuse_internal_vnode_disappear(vp);
1410 	return err;
1411 }
1412 
1413 /*
1414     struct vnop_rename_args {
1415 	struct vnode *a_fdvp;
1416 	struct vnode *a_fvp;
1417 	struct componentname *a_fcnp;
1418 	struct vnode *a_tdvp;
1419 	struct vnode *a_tvp;
1420 	struct componentname *a_tcnp;
1421     };
1422 */
1423 static int
1424 fuse_vnop_rename(struct vop_rename_args *ap)
1425 {
1426 	struct vnode *fdvp = ap->a_fdvp;
1427 	struct vnode *fvp = ap->a_fvp;
1428 	struct componentname *fcnp = ap->a_fcnp;
1429 	struct vnode *tdvp = ap->a_tdvp;
1430 	struct vnode *tvp = ap->a_tvp;
1431 	struct componentname *tcnp = ap->a_tcnp;
1432 	struct fuse_data *data;
1433 
1434 	int err = 0;
1435 
1436 	FS_DEBUG2G("from: inode=%ju name=%*s -> to: inode=%ju name=%*s\n",
1437 	    (uintmax_t)VTOI(fvp), (int)fcnp->cn_namelen, fcnp->cn_nameptr,
1438 	    (uintmax_t)(tvp == NULL ? -1 : VTOI(tvp)),
1439 	    (int)tcnp->cn_namelen, tcnp->cn_nameptr);
1440 
1441 	if (fuse_isdeadfs(fdvp)) {
1442 		return ENXIO;
1443 	}
1444 	if (fvp->v_mount != tdvp->v_mount ||
1445 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1446 		FS_DEBUG("cross-device rename: %s -> %s\n",
1447 		    fcnp->cn_nameptr, (tcnp != NULL ? tcnp->cn_nameptr : "(NULL)"));
1448 		err = EXDEV;
1449 		goto out;
1450 	}
1451 	cache_purge(fvp);
1452 
1453 	/*
1454          * FUSE library is expected to check if target directory is not
1455          * under the source directory in the file system tree.
1456          * Linux performs this check at VFS level.
1457          */
1458 	data = fuse_get_mpdata(vnode_mount(tdvp));
1459 	sx_xlock(&data->rename_lock);
1460 	err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp);
1461 	if (err == 0) {
1462 		if (tdvp != fdvp)
1463 			fuse_vnode_setparent(fvp, tdvp);
1464 		if (tvp != NULL)
1465 			fuse_vnode_setparent(tvp, NULL);
1466 	}
1467 	sx_unlock(&data->rename_lock);
1468 
1469 	if (tvp != NULL && tvp != fvp) {
1470 		cache_purge(tvp);
1471 	}
1472 	if (vnode_isdir(fvp)) {
1473 		if ((tvp != NULL) && vnode_isdir(tvp)) {
1474 			cache_purge(tdvp);
1475 		}
1476 		cache_purge(fdvp);
1477 	}
1478 out:
1479 	if (tdvp == tvp) {
1480 		vrele(tdvp);
1481 	} else {
1482 		vput(tdvp);
1483 	}
1484 	if (tvp != NULL) {
1485 		vput(tvp);
1486 	}
1487 	vrele(fdvp);
1488 	vrele(fvp);
1489 
1490 	return err;
1491 }
1492 
1493 /*
1494     struct vnop_rmdir_args {
1495 	    struct vnode *a_dvp;
1496 	    struct vnode *a_vp;
1497 	    struct componentname *a_cnp;
1498     } *ap;
1499 */
1500 static int
1501 fuse_vnop_rmdir(struct vop_rmdir_args *ap)
1502 {
1503 	struct vnode *dvp = ap->a_dvp;
1504 	struct vnode *vp = ap->a_vp;
1505 
1506 	int err;
1507 
1508 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
1509 
1510 	if (fuse_isdeadfs(vp)) {
1511 		return ENXIO;
1512 	}
1513 	if (VTOFUD(vp) == VTOFUD(dvp)) {
1514 		return EINVAL;
1515 	}
1516 	err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR);
1517 
1518 	if (err == 0)
1519 		fuse_internal_vnode_disappear(vp);
1520 	return err;
1521 }
1522 
1523 /*
1524     struct vnop_setattr_args {
1525 	struct vnode *a_vp;
1526 	struct vattr *a_vap;
1527 	struct ucred *a_cred;
1528 	struct thread *a_td;
1529     };
1530 */
1531 static int
1532 fuse_vnop_setattr(struct vop_setattr_args *ap)
1533 {
1534 	struct vnode *vp = ap->a_vp;
1535 	struct vattr *vap = ap->a_vap;
1536 	struct ucred *cred = ap->a_cred;
1537 	struct thread *td = curthread;
1538 
1539 	struct fuse_dispatcher fdi;
1540 	struct fuse_setattr_in *fsai;
1541 	struct fuse_access_param facp;
1542 
1543 	int err = 0;
1544 	enum vtype vtyp;
1545 	int sizechanged = 0;
1546 	uint64_t newsize = 0;
1547 
1548 	FS_DEBUG2G("inode=%ju\n", (uintmax_t)VTOI(vp));
1549 
1550 	if (fuse_isdeadfs(vp)) {
1551 		return ENXIO;
1552 	}
1553 	fdisp_init(&fdi, sizeof(*fsai));
1554 	fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred);
1555 	fsai = fdi.indata;
1556 	fsai->valid = 0;
1557 
1558 	bzero(&facp, sizeof(facp));
1559 
1560 	facp.xuid = vap->va_uid;
1561 	facp.xgid = vap->va_gid;
1562 
1563 	if (vap->va_uid != (uid_t)VNOVAL) {
1564 		facp.facc_flags |= FACCESS_CHOWN;
1565 		fsai->uid = vap->va_uid;
1566 		fsai->valid |= FATTR_UID;
1567 	}
1568 	if (vap->va_gid != (gid_t)VNOVAL) {
1569 		facp.facc_flags |= FACCESS_CHOWN;
1570 		fsai->gid = vap->va_gid;
1571 		fsai->valid |= FATTR_GID;
1572 	}
1573 	if (vap->va_size != VNOVAL) {
1574 
1575 		struct fuse_filehandle *fufh = NULL;
1576 
1577 		/*Truncate to a new value. */
1578 		    fsai->size = vap->va_size;
1579 		sizechanged = 1;
1580 		newsize = vap->va_size;
1581 		fsai->valid |= FATTR_SIZE;
1582 
1583 		fuse_filehandle_getrw(vp, FUFH_WRONLY, &fufh);
1584 		if (fufh) {
1585 			fsai->fh = fufh->fh_id;
1586 			fsai->valid |= FATTR_FH;
1587 		}
1588 	}
1589 	if (vap->va_atime.tv_sec != VNOVAL) {
1590 		fsai->atime = vap->va_atime.tv_sec;
1591 		fsai->atimensec = vap->va_atime.tv_nsec;
1592 		fsai->valid |= FATTR_ATIME;
1593 	}
1594 	if (vap->va_mtime.tv_sec != VNOVAL) {
1595 		fsai->mtime = vap->va_mtime.tv_sec;
1596 		fsai->mtimensec = vap->va_mtime.tv_nsec;
1597 		fsai->valid |= FATTR_MTIME;
1598 	}
1599 	if (vap->va_mode != (mode_t)VNOVAL) {
1600 		fsai->mode = vap->va_mode & ALLPERMS;
1601 		fsai->valid |= FATTR_MODE;
1602 	}
1603 	if (!fsai->valid) {
1604 		goto out;
1605 	}
1606 	vtyp = vnode_vtype(vp);
1607 
1608 	if (fsai->valid & FATTR_SIZE && vtyp == VDIR) {
1609 		err = EISDIR;
1610 		goto out;
1611 	}
1612 	if (vfs_isrdonly(vnode_mount(vp)) && (fsai->valid & ~FATTR_SIZE || vtyp == VREG)) {
1613 		err = EROFS;
1614 		goto out;
1615 	}
1616 	if (fsai->valid & ~FATTR_SIZE) {
1617 	  /*err = fuse_internal_access(vp, VADMIN, context, &facp); */
1618 	  /*XXX */
1619 		    err = 0;
1620 	}
1621 	facp.facc_flags &= ~FACCESS_XQUERIES;
1622 
1623 	if (err && !(fsai->valid & ~(FATTR_ATIME | FATTR_MTIME)) &&
1624 	    vap->va_vaflags & VA_UTIMES_NULL) {
1625 		err = fuse_internal_access(vp, VWRITE, &facp, td, cred);
1626 	}
1627 	if (err)
1628 		goto out;
1629 	if ((err = fdisp_wait_answ(&fdi)))
1630 		goto out;
1631 	vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode);
1632 
1633 	if (vnode_vtype(vp) != vtyp) {
1634 		if (vnode_vtype(vp) == VNON && vtyp != VNON) {
1635 			debug_printf("FUSE: Dang! vnode_vtype is VNON and vtype isn't.\n");
1636 		} else {
1637 			/*
1638 	                 * STALE vnode, ditch
1639 	                 *
1640 	                 * The vnode has changed its type "behind our back". There's
1641 	                 * nothing really we can do, so let us just force an internal
1642 	                 * revocation and tell the caller to try again, if interested.
1643 	                 */
1644 			fuse_internal_vnode_disappear(vp);
1645 			err = EAGAIN;
1646 		}
1647 	}
1648 	if (!err && !sizechanged) {
1649 		cache_attrs(vp, (struct fuse_attr_out *)fdi.answ);
1650 	}
1651 out:
1652 	fdisp_destroy(&fdi);
1653 	if (!err && sizechanged) {
1654 		fuse_vnode_setsize(vp, cred, newsize);
1655 		VTOFUD(vp)->flag &= ~FN_SIZECHANGE;
1656 	}
1657 	return err;
1658 }
1659 
1660 /*
1661     struct vnop_strategy_args {
1662 	struct vnode *a_vp;
1663 	struct buf *a_bp;
1664     };
1665 */
1666 static int
1667 fuse_vnop_strategy(struct vop_strategy_args *ap)
1668 {
1669 	struct vnode *vp = ap->a_vp;
1670 	struct buf *bp = ap->a_bp;
1671 
1672 	fuse_trace_printf_vnop();
1673 
1674 	if (!vp || fuse_isdeadfs(vp)) {
1675 		bp->b_ioflags |= BIO_ERROR;
1676 		bp->b_error = ENXIO;
1677 		bufdone(bp);
1678 		return ENXIO;
1679 	}
1680 	if (bp->b_iocmd == BIO_WRITE)
1681 		fuse_vnode_refreshsize(vp, NOCRED);
1682 
1683 	(void)fuse_io_strategy(vp, bp);
1684 
1685 	/*
1686 	 * This is a dangerous function. If returns error, that might mean a
1687 	 * panic. We prefer pretty much anything over being forced to panic
1688 	 * by a malicious daemon (a demon?). So we just return 0 anyway. You
1689 	 * should never mind this: this function has its own error
1690 	 * propagation mechanism via the argument buffer, so
1691 	 * not-that-melodramatic residents of the call chain still will be
1692 	 * able to know what to do.
1693 	 */
1694 	return 0;
1695 }
1696 
1697 
1698 /*
1699     struct vnop_symlink_args {
1700 	struct vnode *a_dvp;
1701 	struct vnode **a_vpp;
1702 	struct componentname *a_cnp;
1703 	struct vattr *a_vap;
1704 	char *a_target;
1705     };
1706 */
1707 static int
1708 fuse_vnop_symlink(struct vop_symlink_args *ap)
1709 {
1710 	struct vnode *dvp = ap->a_dvp;
1711 	struct vnode **vpp = ap->a_vpp;
1712 	struct componentname *cnp = ap->a_cnp;
1713 	char *target = ap->a_target;
1714 
1715 	struct fuse_dispatcher fdi;
1716 
1717 	int err;
1718 	size_t len;
1719 
1720 	FS_DEBUG2G("inode=%ju name=%*s\n",
1721 	    (uintmax_t)VTOI(dvp), (int)cnp->cn_namelen, cnp->cn_nameptr);
1722 
1723 	if (fuse_isdeadfs(dvp)) {
1724 		return ENXIO;
1725 	}
1726 	/*
1727          * Unlike the other creator type calls, here we have to create a message
1728          * where the name of the new entry comes first, and the data describing
1729          * the entry comes second.
1730          * Hence we can't rely on our handy fuse_internal_newentry() routine,
1731          * but put together the message manually and just call the core part.
1732          */
1733 
1734 	len = strlen(target) + 1;
1735 	fdisp_init(&fdi, len + cnp->cn_namelen + 1);
1736 	fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL);
1737 
1738 	memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen);
1739 	((char *)fdi.indata)[cnp->cn_namelen] = '\0';
1740 	memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len);
1741 
1742 	err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi);
1743 	fdisp_destroy(&fdi);
1744 	return err;
1745 }
1746 
1747 /*
1748     struct vnop_write_args {
1749 	struct vnode *a_vp;
1750 	struct uio *a_uio;
1751 	int  a_ioflag;
1752 	struct ucred *a_cred;
1753     };
1754 */
1755 static int
1756 fuse_vnop_write(struct vop_write_args *ap)
1757 {
1758 	struct vnode *vp = ap->a_vp;
1759 	struct uio *uio = ap->a_uio;
1760 	int ioflag = ap->a_ioflag;
1761 	struct ucred *cred = ap->a_cred;
1762 
1763 	fuse_trace_printf_vnop();
1764 
1765 	if (fuse_isdeadfs(vp)) {
1766 		return ENXIO;
1767 	}
1768 	fuse_vnode_refreshsize(vp, cred);
1769 
1770 	if (VTOFUD(vp)->flag & FN_DIRECTIO) {
1771 		ioflag |= IO_DIRECT;
1772 	}
1773 
1774 	return fuse_io_dispatch(vp, uio, ioflag, cred);
1775 }
1776 
1777 /*
1778     struct vnop_getpages_args {
1779         struct vnode *a_vp;
1780         vm_page_t *a_m;
1781         int a_count;
1782         int a_reqpage;
1783     };
1784 */
1785 static int
1786 fuse_vnop_getpages(struct vop_getpages_args *ap)
1787 {
1788 	int i, error, nextoff, size, toff, count, npages;
1789 	struct uio uio;
1790 	struct iovec iov;
1791 	vm_offset_t kva;
1792 	struct buf *bp;
1793 	struct vnode *vp;
1794 	struct thread *td;
1795 	struct ucred *cred;
1796 	vm_page_t *pages;
1797 
1798 	FS_DEBUG2G("heh\n");
1799 
1800 	vp = ap->a_vp;
1801 	KASSERT(vp->v_object, ("objectless vp passed to getpages"));
1802 	td = curthread;			/* XXX */
1803 	cred = curthread->td_ucred;	/* XXX */
1804 	pages = ap->a_m;
1805 	npages = ap->a_count;
1806 
1807 	if (!fsess_opt_mmap(vnode_mount(vp))) {
1808 		FS_DEBUG("called on non-cacheable vnode??\n");
1809 		return (VM_PAGER_ERROR);
1810 	}
1811 
1812 	/*
1813 	 * If the last page is partially valid, just return it and allow
1814 	 * the pager to zero-out the blanks.  Partially valid pages can
1815 	 * only occur at the file EOF.
1816 	 *
1817 	 * XXXGL: is that true for FUSE, which is a local filesystem,
1818 	 * but still somewhat disconnected from the kernel?
1819 	 */
1820 	VM_OBJECT_WLOCK(vp->v_object);
1821 	if (pages[npages - 1]->valid != 0 && --npages == 0)
1822 		goto out;
1823 	VM_OBJECT_WUNLOCK(vp->v_object);
1824 
1825 	/*
1826 	 * We use only the kva address for the buffer, but this is extremely
1827 	 * convenient and fast.
1828 	 */
1829 	bp = getpbuf(&fuse_pbuf_freecnt);
1830 
1831 	kva = (vm_offset_t)bp->b_data;
1832 	pmap_qenter(kva, pages, npages);
1833 	VM_CNT_INC(v_vnodein);
1834 	VM_CNT_ADD(v_vnodepgsin, npages);
1835 
1836 	count = npages << PAGE_SHIFT;
1837 	iov.iov_base = (caddr_t)kva;
1838 	iov.iov_len = count;
1839 	uio.uio_iov = &iov;
1840 	uio.uio_iovcnt = 1;
1841 	uio.uio_offset = IDX_TO_OFF(pages[0]->pindex);
1842 	uio.uio_resid = count;
1843 	uio.uio_segflg = UIO_SYSSPACE;
1844 	uio.uio_rw = UIO_READ;
1845 	uio.uio_td = td;
1846 
1847 	error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred);
1848 	pmap_qremove(kva, npages);
1849 
1850 	relpbuf(bp, &fuse_pbuf_freecnt);
1851 
1852 	if (error && (uio.uio_resid == count)) {
1853 		FS_DEBUG("error %d\n", error);
1854 		return VM_PAGER_ERROR;
1855 	}
1856 	/*
1857 	 * Calculate the number of bytes read and validate only that number
1858 	 * of bytes.  Note that due to pending writes, size may be 0.  This
1859 	 * does not mean that the remaining data is invalid!
1860 	 */
1861 
1862 	size = count - uio.uio_resid;
1863 	VM_OBJECT_WLOCK(vp->v_object);
1864 	fuse_vm_page_lock_queues();
1865 	for (i = 0, toff = 0; i < npages; i++, toff = nextoff) {
1866 		vm_page_t m;
1867 
1868 		nextoff = toff + PAGE_SIZE;
1869 		m = pages[i];
1870 
1871 		if (nextoff <= size) {
1872 			/*
1873 			 * Read operation filled an entire page
1874 			 */
1875 			m->valid = VM_PAGE_BITS_ALL;
1876 			KASSERT(m->dirty == 0,
1877 			    ("fuse_getpages: page %p is dirty", m));
1878 		} else if (size > toff) {
1879 			/*
1880 			 * Read operation filled a partial page.
1881 			 */
1882 			m->valid = 0;
1883 			vm_page_set_valid_range(m, 0, size - toff);
1884 			KASSERT(m->dirty == 0,
1885 			    ("fuse_getpages: page %p is dirty", m));
1886 		} else {
1887 			/*
1888 			 * Read operation was short.  If no error occurred
1889 			 * we may have hit a zero-fill section.   We simply
1890 			 * leave valid set to 0.
1891 			 */
1892 			;
1893 		}
1894 	}
1895 	fuse_vm_page_unlock_queues();
1896 out:
1897 	VM_OBJECT_WUNLOCK(vp->v_object);
1898 	if (ap->a_rbehind)
1899 		*ap->a_rbehind = 0;
1900 	if (ap->a_rahead)
1901 		*ap->a_rahead = 0;
1902 	return (VM_PAGER_OK);
1903 }
1904 
1905 /*
1906     struct vnop_putpages_args {
1907         struct vnode *a_vp;
1908         vm_page_t *a_m;
1909         int a_count;
1910         int a_sync;
1911         int *a_rtvals;
1912         vm_ooffset_t a_offset;
1913     };
1914 */
1915 static int
1916 fuse_vnop_putpages(struct vop_putpages_args *ap)
1917 {
1918 	struct uio uio;
1919 	struct iovec iov;
1920 	vm_offset_t kva;
1921 	struct buf *bp;
1922 	int i, error, npages, count;
1923 	off_t offset;
1924 	int *rtvals;
1925 	struct vnode *vp;
1926 	struct thread *td;
1927 	struct ucred *cred;
1928 	vm_page_t *pages;
1929 	vm_ooffset_t fsize;
1930 
1931 	FS_DEBUG2G("heh\n");
1932 
1933 	vp = ap->a_vp;
1934 	KASSERT(vp->v_object, ("objectless vp passed to putpages"));
1935 	fsize = vp->v_object->un_pager.vnp.vnp_size;
1936 	td = curthread;			/* XXX */
1937 	cred = curthread->td_ucred;	/* XXX */
1938 	pages = ap->a_m;
1939 	count = ap->a_count;
1940 	rtvals = ap->a_rtvals;
1941 	npages = btoc(count);
1942 	offset = IDX_TO_OFF(pages[0]->pindex);
1943 
1944 	if (!fsess_opt_mmap(vnode_mount(vp))) {
1945 		FS_DEBUG("called on non-cacheable vnode??\n");
1946 	}
1947 	for (i = 0; i < npages; i++)
1948 		rtvals[i] = VM_PAGER_AGAIN;
1949 
1950 	/*
1951 	 * When putting pages, do not extend file past EOF.
1952 	 */
1953 
1954 	if (offset + count > fsize) {
1955 		count = fsize - offset;
1956 		if (count < 0)
1957 			count = 0;
1958 	}
1959 	/*
1960 	 * We use only the kva address for the buffer, but this is extremely
1961 	 * convenient and fast.
1962 	 */
1963 	bp = getpbuf(&fuse_pbuf_freecnt);
1964 
1965 	kva = (vm_offset_t)bp->b_data;
1966 	pmap_qenter(kva, pages, npages);
1967 	VM_CNT_INC(v_vnodeout);
1968 	VM_CNT_ADD(v_vnodepgsout, count);
1969 
1970 	iov.iov_base = (caddr_t)kva;
1971 	iov.iov_len = count;
1972 	uio.uio_iov = &iov;
1973 	uio.uio_iovcnt = 1;
1974 	uio.uio_offset = offset;
1975 	uio.uio_resid = count;
1976 	uio.uio_segflg = UIO_SYSSPACE;
1977 	uio.uio_rw = UIO_WRITE;
1978 	uio.uio_td = td;
1979 
1980 	error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred);
1981 
1982 	pmap_qremove(kva, npages);
1983 	relpbuf(bp, &fuse_pbuf_freecnt);
1984 
1985 	if (!error) {
1986 		int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE;
1987 
1988 		for (i = 0; i < nwritten; i++) {
1989 			rtvals[i] = VM_PAGER_OK;
1990 			VM_OBJECT_WLOCK(pages[i]->object);
1991 			vm_page_undirty(pages[i]);
1992 			VM_OBJECT_WUNLOCK(pages[i]->object);
1993 		}
1994 	}
1995 	return rtvals[0];
1996 }
1997 
1998 static const char extattr_namespace_separator = '.';
1999 
2000 /*
2001     struct vop_getextattr_args {
2002         struct vop_generic_args a_gen;
2003         struct vnode *a_vp;
2004         int a_attrnamespace;
2005         const char *a_name;
2006         struct uio *a_uio;
2007         size_t *a_size;
2008         struct ucred *a_cred;
2009         struct thread *a_td;
2010     };
2011 */
2012 static int
2013 fuse_vnop_getextattr(struct vop_getextattr_args *ap)
2014 {
2015 	struct vnode *vp = ap->a_vp;
2016 	struct uio *uio = ap->a_uio;
2017 	struct fuse_dispatcher fdi = {0};
2018 	struct fuse_getxattr_in *get_xattr_in;
2019 	struct fuse_getxattr_out *get_xattr_out;
2020 	struct mount *mp = vnode_mount(vp);
2021 	char *prefix;
2022 	size_t len;
2023 	char *attr_str;
2024 	struct thread *td = ap->a_td;
2025 	struct ucred *cred = ap->a_cred;
2026 	int err = 0;
2027 
2028 	fuse_trace_printf_vnop();
2029 
2030 	if (fuse_isdeadfs(vp))
2031 		return ENXIO;
2032 
2033 	/* Default to looking for user attributes. */
2034 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2035 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2036 	else
2037 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2038 
2039 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2040 	    strlen(ap->a_name) + 1;
2041 
2042 	fdisp_init(&fdi, len + sizeof(*get_xattr_in));
2043 	fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred);
2044 
2045 	get_xattr_in = fdi.indata;
2046 	/*
2047 	 * Check to see whether we're querying the available size or
2048 	 * issuing the actual request.  If we pass in 0, we get back struct
2049 	 * fuse_getxattr_out.  If we pass in a non-zero size, we get back
2050 	 * that much data, without the struct fuse_getxattr_out header.
2051 	 */
2052 	if (ap->a_size != NULL)
2053 		get_xattr_in->size = 0;
2054 	else
2055 		get_xattr_in->size = uio->uio_resid;
2056 
2057 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2058 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2059 	    ap->a_name);
2060 
2061 	err = fdisp_wait_answ(&fdi);
2062 
2063 	if (err != 0) {
2064 		if (err == ENOSYS)
2065 			fsess_set_notimpl(mp, FUSE_GETXATTR);
2066 		debug_printf("getxattr: got err=%d from daemon\n", err);
2067 		goto out;
2068 	}
2069 
2070 	/*
2071 	 * If we get to this point (i.e. no error), we should have a valid
2072 	 * answer of some sort.  i.e. non-zero iosize and a valid pointer.
2073 	 */
2074 	if ((fdi.answ == NULL) || (fdi.iosize == 0)) {
2075 		debug_printf("getxattr: err = 0, but answ = %p, iosize = %zu\n",
2076 		    fdi.answ, fdi.iosize);
2077 		err = EINVAL;
2078 		goto out;
2079 	}
2080 	get_xattr_out = fdi.answ;
2081 
2082 	if (ap->a_size != NULL) {
2083 		*ap->a_size = get_xattr_out->size;
2084 	} else if (fdi.iosize > 0) {
2085 		err = uiomove(fdi.answ, fdi.iosize, uio);
2086 	} else {
2087 		err = EINVAL;
2088 	}
2089 
2090 out:
2091 	fdisp_destroy(&fdi);
2092 	return (err);
2093 }
2094 
2095 /*
2096     struct vop_setextattr_args {
2097         struct vop_generic_args a_gen;
2098         struct vnode *a_vp;
2099         int a_attrnamespace;
2100         const char *a_name;
2101         struct uio *a_uio;
2102         struct ucred *a_cred;
2103         struct thread *a_td;
2104     };
2105 */
2106 static int
2107 fuse_vnop_setextattr(struct vop_setextattr_args *ap)
2108 {
2109 	struct vnode *vp = ap->a_vp;
2110 	struct uio *uio = ap->a_uio;
2111 	struct fuse_dispatcher fdi = {0};
2112 	struct fuse_setxattr_in *set_xattr_in;
2113 	struct mount *mp = vnode_mount(vp);
2114 	char *prefix;
2115 	size_t len;
2116 	char *attr_str;
2117 	struct thread *td = ap->a_td;
2118 	struct ucred *cred = ap->a_cred;
2119 	int err = 0;
2120 
2121 	fuse_trace_printf_vnop();
2122 
2123 	if (fuse_isdeadfs(vp))
2124 		return ENXIO;
2125 
2126 	/* Default to looking for user attributes. */
2127 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2128 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2129 	else
2130 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2131 
2132 	len = strlen(prefix) + sizeof(extattr_namespace_separator) +
2133 	    strlen(ap->a_name) + 1;
2134 
2135 	fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid);
2136 	fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred);
2137 
2138 	set_xattr_in = fdi.indata;
2139 	set_xattr_in->size = uio->uio_resid;
2140 
2141 	attr_str = (char *)fdi.indata + sizeof(*set_xattr_in);
2142 	snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator,
2143 	    ap->a_name);
2144 
2145 	err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len,
2146 	    uio->uio_resid, uio);
2147 	if (err != 0) {
2148 		debug_printf("setxattr: got error %d from uiomove\n", err);
2149 		goto out;
2150 	}
2151 
2152 	err = fdisp_wait_answ(&fdi);
2153 
2154 	if (err != 0) {
2155 		if (err == ENOSYS)
2156 			fsess_set_notimpl(mp, FUSE_SETXATTR);
2157 		debug_printf("setxattr: got err=%d from daemon\n", err);
2158 		goto out;
2159 	}
2160 
2161 out:
2162 	fdisp_destroy(&fdi);
2163 	return (err);
2164 }
2165 
2166 /*
2167  * The Linux / FUSE extended attribute list is simply a collection of
2168  * NUL-terminated strings.  The FreeBSD extended attribute list is a single
2169  * byte length followed by a non-NUL terminated string.  So, this allows
2170  * conversion of the Linux / FUSE format to the FreeBSD format in place.
2171  * Linux attribute names are reported with the namespace as a prefix (e.g.
2172  * "user.attribute_name"), but in FreeBSD they are reported without the
2173  * namespace prefix (e.g. "attribute_name").  So, we're going from:
2174  *
2175  * user.attr_name1\0user.attr_name2\0
2176  *
2177  * to:
2178  *
2179  * <num>attr_name1<num>attr_name2
2180  *
2181  * Where "<num>" is a single byte number of characters in the attribute name.
2182  *
2183  * Args:
2184  * prefix - exattr namespace prefix string
2185  * list, list_len - input list with namespace prefixes
2186  * bsd_list, bsd_list_len - output list compatible with bsd vfs
2187  */
2188 static int
2189 fuse_xattrlist_convert(char *prefix, const char *list, int list_len,
2190     char *bsd_list, int *bsd_list_len)
2191 {
2192 	int len, pos, dist_to_next, prefix_len;
2193 
2194 	pos = 0;
2195 	*bsd_list_len = 0;
2196 	prefix_len = strlen(prefix);
2197 
2198 	while (pos < list_len && list[pos] != '\0') {
2199 		dist_to_next = strlen(&list[pos]) + 1;
2200 		if (bcmp(&list[pos], prefix, prefix_len) == 0 &&
2201 		    list[pos + prefix_len] == extattr_namespace_separator) {
2202 			len = dist_to_next -
2203 			    (prefix_len + sizeof(extattr_namespace_separator)) - 1;
2204 			if (len >= EXTATTR_MAXNAMELEN)
2205 				return (ENAMETOOLONG);
2206 
2207 			bsd_list[*bsd_list_len] = len;
2208 			memcpy(&bsd_list[*bsd_list_len + 1],
2209 			    &list[pos + prefix_len +
2210 			    sizeof(extattr_namespace_separator)], len);
2211 
2212 			*bsd_list_len += len + 1;
2213 		}
2214 
2215 		pos += dist_to_next;
2216 	}
2217 
2218 	return (0);
2219 }
2220 
2221 /*
2222     struct vop_listextattr_args {
2223         struct vop_generic_args a_gen;
2224         struct vnode *a_vp;
2225         int a_attrnamespace;
2226         struct uio *a_uio;
2227         size_t *a_size;
2228         struct ucred *a_cred;
2229         struct thread *a_td;
2230     };
2231 */
2232 static int
2233 fuse_vnop_listextattr(struct vop_listextattr_args *ap)
2234 {
2235 	struct vnode *vp = ap->a_vp;
2236 	struct uio *uio = ap->a_uio;
2237 	struct fuse_dispatcher fdi = {0};
2238 	struct fuse_getxattr_in *get_xattr_in;
2239 	struct fuse_getxattr_out *get_xattr_out;
2240 	struct mount *mp = vnode_mount(vp);
2241 	size_t len;
2242 	char *prefix;
2243 	char *attr_str;
2244 	char *bsd_list = NULL;
2245 	int bsd_list_len;
2246 	struct thread *td = ap->a_td;
2247 	struct ucred *cred = ap->a_cred;
2248 	int err = 0;
2249 
2250 	fuse_trace_printf_vnop();
2251 
2252 	if (fuse_isdeadfs(vp))
2253 		return ENXIO;
2254 
2255 	/*
2256 	 * Add space for a NUL and the period separator if enabled.
2257 	 * Default to looking for user attributes.
2258 	 */
2259 	if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM)
2260 		prefix = EXTATTR_NAMESPACE_SYSTEM_STRING;
2261 	else
2262 		prefix = EXTATTR_NAMESPACE_USER_STRING;
2263 
2264 	len = strlen(prefix) + sizeof(extattr_namespace_separator) + 1;
2265 
2266 	fdisp_init(&fdi, sizeof(*get_xattr_in) + len);
2267 	fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred);
2268 
2269 	get_xattr_in = fdi.indata;
2270 	if (ap->a_size != NULL)
2271 		get_xattr_in->size = 0;
2272 	else
2273 		get_xattr_in->size = uio->uio_resid + sizeof(*get_xattr_out);
2274 
2275 
2276 	attr_str = (char *)fdi.indata + sizeof(*get_xattr_in);
2277 	snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator);
2278 
2279 	err = fdisp_wait_answ(&fdi);
2280 	if (err != 0) {
2281 		if (err == ENOSYS)
2282 			fsess_set_notimpl(mp, FUSE_LISTXATTR);
2283 		debug_printf("listextattr: got err=%d from daemon\n", err);
2284 		goto out;
2285 	}
2286 
2287 	if ((fdi.answ == NULL) || (fdi.iosize == 0)) {
2288 		err = EINVAL;
2289 		goto out;
2290 	}
2291 	get_xattr_out = fdi.answ;
2292 
2293 	if (ap->a_size != NULL) {
2294 		*ap->a_size = get_xattr_out->size;
2295 	} else if (fdi.iosize > 0) {
2296 		/*
2297 		 * The Linux / FUSE attribute list format isn't the same
2298 		 * as FreeBSD's format.  So we need to transform it into
2299 		 * FreeBSD's format before giving it to the user.
2300 		 */
2301 		bsd_list = malloc(fdi.iosize, M_TEMP, M_WAITOK);
2302 		err = fuse_xattrlist_convert(prefix, fdi.answ, fdi.iosize,
2303 		    bsd_list, &bsd_list_len);
2304 		if (err != 0)
2305 			goto out;
2306 
2307 		err = uiomove(bsd_list, bsd_list_len, uio);
2308 	} else {
2309 		debug_printf("listextattr: returned iosize %zu for %s attribute list is "
2310 		    "too small\n", fdi.iosize, prefix);
2311 		err = EINVAL;
2312 	}
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 = {0};
2335 	struct mount *mp = vnode_mount(vp);
2336 	char *prefix;
2337 	size_t len;
2338 	char *attr_str;
2339 	struct thread *td = ap->a_td;
2340 	struct ucred *cred = ap->a_cred;
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