xref: /netbsd-src/lib/libperfuse/ops.c (revision 4fee23f98c45552038ad6b5bd05124a41302fb01)
1 /*  $NetBSD: ops.c,v 1.30 2011/06/01 15:54:10 manu Exp $ */
2 
3 /*-
4  *  Copyright (c) 2010-2011 Emmanuel Dreyfus. All rights reserved.
5  *
6  *  Redistribution and use in source and binary forms, with or without
7  *  modification, are permitted provided that the following conditions
8  *  are met:
9  *  1. Redistributions of source code must retain the above copyright
10  *     notice, this list of conditions and the following disclaimer.
11  *  2. Redistributions in binary form must reproduce the above copyright
12  *     notice, this list of conditions and the following disclaimer in the
13  *     documentation and/or other materials provided with the distribution.
14  *
15  *  THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
16  *  ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
17  *  TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
18  *  PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
19  *  BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
20  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
21  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
22  *  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
23  *  CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
24  *  ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
25  *  POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 #include <stdio.h>
29 #include <unistd.h>
30 #include <stdlib.h>
31 #include <libgen.h>
32 #include <errno.h>
33 #include <err.h>
34 #include <sysexits.h>
35 #include <syslog.h>
36 #include <puffs.h>
37 #include <sys/socket.h>
38 #include <sys/socket.h>
39 #include <machine/vmparam.h>
40 
41 #include "perfuse_priv.h"
42 #include "fuse.h"
43 
44 extern int perfuse_diagflags;
45 
46 static int xchg_msg(struct puffs_usermount *, puffs_cookie_t,
47     perfuse_msg_t *, size_t, enum perfuse_xchg_pb_reply);
48 static int mode_access(puffs_cookie_t, const struct puffs_cred *, mode_t);
49 static int sticky_access(struct puffs_node *, const struct puffs_cred *);
50 static void fuse_attr_to_vap(struct perfuse_state *,
51     struct vattr *, struct fuse_attr *);
52 static int node_lookup_dir_nodot(struct puffs_usermount *,
53     puffs_cookie_t, char *, size_t, struct puffs_node **);
54 static int node_lookup_common(struct puffs_usermount *, puffs_cookie_t,
55     const char*, struct puffs_node **);
56 static int node_mk_common(struct puffs_usermount *, puffs_cookie_t,
57     struct puffs_newinfo *, const struct puffs_cn *pcn, perfuse_msg_t *);
58 static int node_mk_common_final(struct puffs_usermount *, puffs_cookie_t,
59     struct puffs_node *, const struct puffs_cn *pcn);
60 static ssize_t fuse_to_dirent(struct puffs_usermount *, puffs_cookie_t,
61     struct fuse_dirent *, size_t);
62 static int readdir_buffered(puffs_cookie_t, struct dirent *, off_t *,
63     size_t *, const struct puffs_cred *, int *, off_t *, size_t *);
64 static void requeue_request(struct puffs_usermount *,
65     puffs_cookie_t opc, enum perfuse_qtype);
66 static int dequeue_requests(struct perfuse_state *,
67     puffs_cookie_t opc, enum perfuse_qtype, int);
68 #define DEQUEUE_ALL 0
69 
70 /*
71  *  From <sys/vnode>, inside #ifdef _KERNEL section
72  */
73 #define IO_SYNC		(0x40|IO_DSYNC)
74 #define IO_DSYNC	0x00200
75 #define IO_DIRECT	0x02000
76 
77 /*
78  *  From <fcntl>, inside #ifdef _KERNEL section
79  */
80 #define F_WAIT		0x010
81 #define F_FLOCK		0x020
82 #define OFLAGS(fflags)  ((fflags) - 1)
83 
84 /*
85  * Borrowed from src/sys/kern/vfs_subr.c and src/sys/sys/vnode.h
86  */
87 const enum vtype iftovt_tab[16] = {
88 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
89         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
90 };
91 const int vttoif_tab[9] = {
92 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
93         S_IFSOCK, S_IFIFO, S_IFMT,
94 };
95 
96 #define IFTOVT(mode) (iftovt_tab[((mode) & S_IFMT) >> 12])
97 #define VTTOIF(indx) (vttoif_tab[(int)(indx)])
98 
99 int
100 perfuse_node_close_common(pu, opc, mode)
101 	struct puffs_usermount *pu;
102 	puffs_cookie_t opc;
103 	int mode;
104 {
105 	struct perfuse_state *ps;
106 	perfuse_msg_t *pm;
107 	int op;
108 	uint64_t fh;
109 	struct fuse_release_in *fri;
110 	struct perfuse_node_data *pnd;
111 	struct puffs_node *pn;
112 	int error;
113 
114 	ps = puffs_getspecific(pu);
115 	pn = (struct puffs_node *)opc;
116 	pnd = PERFUSE_NODE_DATA(pn);
117 
118 	if (puffs_pn_getvap(pn)->va_type == VDIR) {
119 		op = FUSE_RELEASEDIR;
120 		mode = FREAD;
121 	} else {
122 		op = FUSE_RELEASE;
123 	}
124 
125 	/*
126 	 * Destroy the filehandle before sending the
127 	 * request to the FUSE filesystem, otherwise
128 	 * we may get a second close() while we wait
129 	 * for the reply, and we would end up closing
130 	 * the same fh twice instead of closng both.
131 	 */
132 	fh = perfuse_get_fh(opc, mode);
133 	perfuse_destroy_fh(pn, fh);
134 
135 	/*
136 	 * release_flags may be set to FUSE_RELEASE_FLUSH
137 	 * to flush locks. lock_owner must be set in that case
138 	 */
139 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*fri), NULL);
140 	fri = GET_INPAYLOAD(ps, pm, fuse_release_in);
141 	fri->fh = fh;
142 	fri->flags = 0;
143 	fri->release_flags = 0;
144 	fri->lock_owner = pnd->pnd_lock_owner;
145 	fri->flags = (fri->lock_owner != 0) ? FUSE_RELEASE_FLUSH : 0;
146 
147 #ifdef PERFUSE_DEBUG
148 	if (perfuse_diagflags & PDF_FH)
149 		DPRINTF("%s: opc = %p, ino = %"PRId64", fh = 0x%"PRIx64"\n",
150 			 __func__, (void *)opc, pnd->pnd_ino, fri->fh);
151 #endif
152 
153 	if ((error = xchg_msg(pu, opc, pm,
154 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
155 		goto out;
156 
157 	ps->ps_destroy_msg(pm);
158 
159 	error = 0;
160 
161 out:
162 	if (error != 0)
163 		DERRX(EX_SOFTWARE, "%s: freed fh = 0x%"PRIx64" but filesystem "
164 		      "returned error = %d", __func__, fh, error);
165 
166 	return error;
167 }
168 
169 /* ARGSUSED1 */
170 static int
171 xchg_msg(pu, opc, pm, len, wait)
172 	struct puffs_usermount *pu;
173 	puffs_cookie_t opc;
174 	perfuse_msg_t *pm;
175 	size_t len;
176      	enum perfuse_xchg_pb_reply wait;
177 {
178 	struct perfuse_state *ps;
179 	struct perfuse_node_data *pnd;
180 	int error;
181 
182 	ps = puffs_getspecific(pu);
183 	pnd = NULL;
184 	if ((struct puffs_node *)opc != NULL)
185 		pnd = PERFUSE_NODE_DATA(opc);
186 
187 #ifdef PERFUSE_DEBUG
188 	if ((perfuse_diagflags & PDF_FILENAME) && (opc != 0))
189 		DPRINTF("file = \"%s\" flags = 0x%x\n",
190 			perfuse_node_path(opc),
191 			PERFUSE_NODE_DATA(opc)->pnd_flags);
192 #endif
193 	if (pnd)
194 		pnd->pnd_flags |= PND_INXCHG;
195 
196 	error = ps->ps_xchg_msg(pu, pm, len, wait);
197 
198 	if (pnd) {
199 		pnd->pnd_flags &= ~PND_INXCHG;
200 		(void)dequeue_requests(ps, opc, PCQ_AFTERXCHG, DEQUEUE_ALL);
201 	}
202 
203 	return error;
204 }
205 
206 static int
207 mode_access(opc, pcr, mode)
208 	puffs_cookie_t opc;
209 	const struct puffs_cred *pcr;
210 	mode_t mode;
211 {
212 	struct puffs_node *pn;
213 	struct vattr *va;
214 
215 	/*
216 	 * pcr is NULL for self open through fsync or readdir.
217 	 * In both case, access control is useless, as it was
218 	 * done before, at open time.
219 	 */
220 	if (pcr == NULL)
221 		return 0;
222 
223 	pn = (struct puffs_node *)opc;
224 	va = puffs_pn_getvap(pn);
225 	return puffs_access(va->va_type, va->va_mode,
226 			    va->va_uid, va->va_gid,
227 			    mode, pcr);
228 }
229 
230 static int
231 sticky_access(targ, pcr)
232 	struct puffs_node *targ;
233 	const struct puffs_cred *pcr;
234 {
235 	uid_t uid;
236 	struct puffs_node *tdir;
237 	int sticky, owner, error;
238 
239 	tdir = PERFUSE_NODE_DATA(targ)->pnd_parent;
240 
241 	if ((error = puffs_cred_getuid(pcr, &uid)) != 0)
242 		return error;
243 
244 	sticky = puffs_pn_getvap(tdir)->va_mode & S_ISTXT;
245 	owner = puffs_pn_getvap(targ)->va_uid == uid;
246 
247 	if (sticky && !owner)
248 		error = EACCES;
249 
250 	return error;
251 }
252 
253 
254 static void
255 fuse_attr_to_vap(ps, vap, fa)
256 	struct perfuse_state *ps;
257 	struct vattr *vap;
258 	struct fuse_attr *fa;
259 {
260 	vap->va_type = IFTOVT(fa->mode);
261 	vap->va_mode = fa->mode & ALLPERMS;
262 	vap->va_nlink = fa->nlink;
263 	vap->va_uid = fa->uid;
264 	vap->va_gid = fa->gid;
265 	vap->va_fsid = (long)ps->ps_fsid;
266 	vap->va_fileid = fa->ino;
267 	vap->va_size = fa->size;
268 	vap->va_blocksize = fa->blksize;
269 	vap->va_atime.tv_sec = (time_t)fa->atime;
270 	vap->va_atime.tv_nsec = (long) fa->atimensec;
271 	vap->va_mtime.tv_sec = (time_t)fa->mtime;
272 	vap->va_mtime.tv_nsec = (long)fa->mtimensec;
273 	vap->va_ctime.tv_sec = (time_t)fa->ctime;
274 	vap->va_ctime.tv_nsec = (long)fa->ctimensec;
275 	vap->va_birthtime.tv_sec = 0;
276 	vap->va_birthtime.tv_nsec = 0;
277 	vap->va_gen = 0;
278 	vap->va_flags = 0;
279 	vap->va_rdev = fa->rdev;
280 	vap->va_bytes = fa->size;
281 	vap->va_filerev = (u_quad_t)PUFFS_VNOVAL;
282 	vap->va_vaflags = 0;
283 
284 	if (vap->va_blocksize == 0)
285 		vap->va_blocksize = DEV_BSIZE;
286 
287 	if (vap->va_size == (size_t)PUFFS_VNOVAL) /* XXX */
288 		vap->va_size = 0;
289 
290 	return;
291 }
292 
293 
294 /*
295  * Lookup name in directory opc
296  * We take special care of name being . or ..
297  * These are returned by readdir and deserve tweaks.
298  */
299 static int
300 node_lookup_dir_nodot(pu, opc, name, namelen, pnp)
301 	struct puffs_usermount *pu;
302 	puffs_cookie_t opc;
303 	char *name;
304 	size_t namelen;
305 	struct puffs_node **pnp;
306 {
307 	char *path;
308 	struct puffs_node *dpn = (struct puffs_node *)opc;
309 	int error;
310 
311 	/*
312 	 *  is easy as we already know it
313 	 */
314 	if (strncmp(name, ".", namelen) == 0) {
315 		*pnp = (struct puffs_node *)opc;
316 		return 0;
317 	}
318 
319 	/*
320 	 * For .. we just forget the name part
321 	 */
322 	if (strncmp(name, "..", namelen) == 0)
323 		namelen = 0;
324 
325 	namelen = PNPLEN(dpn) + 1 + namelen + 1;
326 	if ((path = malloc(namelen)) == NULL)
327 		DERR(EX_OSERR, "malloc failed");
328 	(void)snprintf(path, namelen, "%s/%s",
329 		       perfuse_node_path((puffs_cookie_t)dpn), name);
330 
331 	error = node_lookup_common(pu, opc, path, pnp);
332 
333 	free(path);
334 
335 	return error;
336 }
337 
338 static int
339 node_lookup_common(pu, opc, path, pnp)
340 	struct puffs_usermount *pu;
341 	puffs_cookie_t opc;
342 	const char *path;
343 	struct puffs_node **pnp;
344 {
345 	struct perfuse_state *ps;
346 	struct perfuse_node_data *pnd;
347 	perfuse_msg_t *pm;
348 	struct fuse_entry_out *feo;
349 	struct puffs_node *pn;
350 	size_t len;
351 	int error;
352 
353 	ps = puffs_getspecific(pu);
354 
355 #ifdef PERFUSE_DEBUG
356 	if (perfuse_diagflags & PDF_FILENAME)
357 		DPRINTF("%s: opc = %p, file = \"%s\" looking up \"%s\"\n",
358 			__func__, (void *)opc, perfuse_node_path(opc), path);
359 #endif
360 	/*
361 	 * Is the node already known?
362 	 */
363 	TAILQ_FOREACH(pnd, &PERFUSE_NODE_DATA(opc)->pnd_children, pnd_next) {
364 		if ((pnd->pnd_flags & PND_REMOVED) ||
365 		    (strcmp(pnd->pnd_name, path) != 0))
366 			continue;
367 
368 		/*
369 		 * We have a match
370 		 */
371 		if (pnp != NULL)
372 			*pnp = (struct puffs_node *)(pnd->pnd_pn);
373 
374 #ifdef PERFUSE_DEBUG
375 		if (perfuse_diagflags & PDF_FILENAME)
376 			DPRINTF("%s: opc = %p, file = \"%s\" found "
377 				"cookie = %p, ino = %"PRId64" for \"%s\"\n",
378 				__func__, (void *)opc, perfuse_node_path(opc),
379 				(void *)pnd->pnd_pn, pnd->pnd_ino, path);
380 #endif
381 		return 0;
382 	}
383 
384 	len = strlen(path) + 1;
385 
386 	pm = ps->ps_new_msg(pu, opc, FUSE_LOOKUP, len, NULL);
387 	(void)strlcpy(_GET_INPAYLOAD(ps, pm, char *), path, len);
388 
389 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
390 		goto out;
391 
392 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
393 
394 	pn = perfuse_new_pn(pu, path, opc);
395 	PERFUSE_NODE_DATA(pn)->pnd_ino = feo->nodeid;
396 
397 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
398 	pn->pn_va.va_gen = (u_long)(feo->generation);
399 
400 	if (pnp != NULL)
401 		*pnp = pn;
402 
403 #ifdef PERFUSE_DEBUG
404 	if (perfuse_diagflags & PDF_FILENAME)
405 		DPRINTF("%s: opc = %p, looked up opc = %p, ino = %"PRId64" "
406 			"file = \"%s\"\n", __func__, (void *)opc, pn,
407 			feo->nodeid, path);
408 #endif
409 out:
410 	ps->ps_destroy_msg(pm);
411 
412 	return error;
413 }
414 
415 
416 /*
417  * Common code for methods that create objects:
418  * perfuse_node_mkdir
419  * perfuse_node_mknod
420  * perfuse_node_symlink
421  */
422 static int
423 node_mk_common(pu, opc, pni, pcn, pm)
424 	struct puffs_usermount *pu;
425 	puffs_cookie_t opc;
426 	struct puffs_newinfo *pni;
427 	const struct puffs_cn *pcn;
428 	perfuse_msg_t *pm;
429 {
430 	struct perfuse_state *ps;
431 	struct puffs_node *pn;
432 	struct fuse_entry_out *feo;
433 	int error;
434 
435 	ps =  puffs_getspecific(pu);
436 
437 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
438 		goto out;
439 
440 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
441 	if (feo->nodeid == PERFUSE_UNKNOWN_INO)
442 		DERRX(EX_SOFTWARE, "%s: no ino", __func__);
443 
444 	pn = perfuse_new_pn(pu, pcn->pcn_name, opc);
445 	PERFUSE_NODE_DATA(pn)->pnd_ino = feo->nodeid;
446 
447 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
448 	pn->pn_va.va_gen = (u_long)(feo->generation);
449 
450 	puffs_newinfo_setcookie(pni, pn);
451 
452 #ifdef PERFUSE_DEBUG
453 	if (perfuse_diagflags & PDF_FILENAME)
454 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
455 			"ino = %"PRId64"\n",
456 			__func__, (void *)pn, pcn->pcn_name,
457 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid);
458 #endif
459 	ps->ps_destroy_msg(pm);
460 
461 	return node_mk_common_final(pu, opc, pn, pcn);
462 
463 out:
464 	ps->ps_destroy_msg(pm);
465 
466 	return error;
467 }
468 
469 /*
470  * Common final code for methods that create objects:
471  * perfuse_node_mkdir via node_mk_common
472  * perfuse_node_mknod via node_mk_common
473  * perfuse_node_symlink via node_mk_common
474  * perfuse_node_create
475  */
476 static int
477 node_mk_common_final(pu, opc, pn, pcn)
478 	struct puffs_usermount *pu;
479 	puffs_cookie_t opc;
480 	struct puffs_node *pn;
481 	const struct puffs_cn *pcn;
482 {
483 	struct perfuse_state *ps;
484 	perfuse_msg_t *pm;
485 	struct fuse_setattr_in *fsi;
486 	struct fuse_attr_out *fao;
487 	int error;
488 
489 	ps =  puffs_getspecific(pu);
490 
491 	/*
492 	 * Set owner and group
493 	 */
494 	(void)puffs_cred_getuid(pcn->pcn_cred, &pn->pn_va.va_uid);
495 	(void)puffs_cred_getgid(pcn->pcn_cred, &pn->pn_va.va_gid);
496 
497 	pm = ps->ps_new_msg(pu, (puffs_cookie_t)pn,
498 			    FUSE_SETATTR, sizeof(*fsi), NULL);
499 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
500 	fsi->uid = pn->pn_va.va_uid;
501 	fsi->gid = pn->pn_va.va_gid;
502 	fsi->valid = FUSE_FATTR_UID|FUSE_FATTR_GID;
503 
504 	if ((error = xchg_msg(pu, (puffs_cookie_t)pn, pm,
505 			      sizeof(*fao), wait_reply)) != 0)
506 		goto out;
507 
508 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
509 	fuse_attr_to_vap(ps, &pn->pn_va, &fao->attr);
510 
511 	/*
512 	 * The parent directory needs a sync
513 	 */
514 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
515 
516 out:
517 	if (pm != NULL)
518 		ps->ps_destroy_msg(pm);
519 
520 	return error;
521 }
522 
523 
524 static ssize_t
525 fuse_to_dirent(pu, opc, fd, fd_len)
526 	struct puffs_usermount *pu;
527 	puffs_cookie_t opc;
528 	struct fuse_dirent *fd;
529 	size_t fd_len;
530 {
531 	struct dirent *dents;
532 	size_t dents_len;
533 	ssize_t written;
534 	uint64_t fd_offset;
535 	struct fuse_dirent *fd_base;
536 	size_t len;
537 
538 	fd_base = fd;
539 	fd_offset = 0;
540 	written = 0;
541 	dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
542 	dents_len = (size_t)PERFUSE_NODE_DATA(opc)->pnd_dirent_len;
543 
544 	do {
545 		char *ndp;
546 		size_t reclen;
547 
548 		reclen = _DIRENT_RECLEN(dents, fd->namelen);
549 
550 		/*
551 		 * Check we do not overflow the output buffer
552 		 * struct fuse_dirent is bigger than struct dirent,
553 		 * so we should always use fd_len and never reallocate
554 		 * later.
555 		 * If we have to reallocate,try to double the buffer
556 		 * each time so that we do not have to do it too often.
557 		 */
558 		if (written + reclen > dents_len) {
559 			if (dents_len == 0)
560 				dents_len = fd_len;
561 			else
562 				dents_len =
563 				   MAX(2 * dents_len, written + reclen);
564 
565 			dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
566 			if ((dents = realloc(dents, dents_len)) == NULL)
567 				DERR(EX_OSERR, "malloc failed");
568 
569 			PERFUSE_NODE_DATA(opc)->pnd_dirent = dents;
570 			PERFUSE_NODE_DATA(opc)->pnd_dirent_len = dents_len;
571 
572 			/*
573 			 * (void *) for delint
574 			 */
575 			ndp = (char *)(void *)dents + written;
576 			dents = (struct dirent *)(void *)ndp;
577 		}
578 
579 
580 
581 		/*
582 		 * Filesystem was mounted without -o use_ino
583 		 * Perform a lookup to find it.
584 		 * XXX still broken
585 		 */
586 		if (fd->ino == PERFUSE_UNKNOWN_INO) {
587 			struct puffs_node *pn;
588 
589 			if (node_lookup_dir_nodot(pu, opc, fd->name,
590 						  fd->namelen, &pn) != 0)
591 				DERRX(EX_SOFTWARE,
592 				     "node_lookup_dir_nodot failed");
593 
594 			fd->ino = PERFUSE_NODE_DATA(pn)->pnd_ino;
595 		}
596 
597 		dents->d_fileno = fd->ino;
598 		dents->d_reclen = (unsigned short)reclen;
599 		dents->d_namlen = fd->namelen;
600 		dents->d_type = fd->type;
601 		strlcpy(dents->d_name, fd->name, fd->namelen + 1);
602 
603 #ifdef PERFUSE_DEBUG
604 		if (perfuse_diagflags & PDF_READDIR)
605 			DPRINTF("%s: translated \"%s\" ino = %"PRId64"\n",
606 				__func__, dents->d_name, dents->d_fileno);
607 #endif
608 
609 		dents = _DIRENT_NEXT(dents);
610 		written += reclen;
611 
612 		/*
613 		 * Move to the next record.
614 		 * fd->off seems unreliable, for instance, flusterfs
615 		 * does not clear the unused bits, and we get
616 		 * 0xffffffffb9b95040 instead of just 0x40. Use
617 		 * record alignement instead.
618 		 */
619 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
620 #ifdef PERFUSE_DEBUG
621 		if (perfuse_diagflags & PDF_READDIR)
622 			DPRINTF("%s: record at %"PRId64"/0x%"PRIx64" "
623 				"length = %zd/0x%zx. "
624 				"next record at %"PRId64"/0x%"PRIx64" "
625 				"max %zd/0x%zx\n",
626 				__func__, fd_offset, fd_offset, len, len,
627 				fd_offset + len, fd_offset + len,
628 				fd_len, fd_len);
629 #endif
630 		fd_offset += len;
631 
632 		/*
633 		 * Check if next record is still within the packet
634 		 * If it is not, we reached the end of the buffer.
635 		 */
636 		if (fd_offset >= fd_len)
637 			break;
638 
639 		/*
640 		 * (void *) for delint
641 		 */
642 		ndp = (char *)(void *)fd_base + (size_t)fd_offset;
643 		fd = (struct fuse_dirent *)(void *)ndp;
644 
645 	} while (1 /* CONSTCOND */);
646 
647 	/*
648 	 * Adjust the dirent output length
649 	 */
650 	if (written != -1)
651 		PERFUSE_NODE_DATA(opc)->pnd_dirent_len = written;
652 
653 	return written;
654 }
655 
656 /* ARGSUSED4 */
657 static int
658 readdir_buffered(opc, dent, readoff,
659 		 reslen, pcr, eofflag, cookies, ncookies)
660 	puffs_cookie_t opc;
661 	struct dirent *dent;
662 	off_t *readoff;
663 	size_t *reslen;
664 	const struct puffs_cred *pcr;
665 	int *eofflag;
666 	off_t *cookies;
667 	size_t *ncookies;
668 {
669 	struct dirent *fromdent;
670 	struct perfuse_node_data *pnd;
671 	char *ndp;
672 
673 	pnd = PERFUSE_NODE_DATA(opc);
674 
675 	while (*readoff < pnd->pnd_dirent_len) {
676 		/*
677 		 * (void *) for delint
678 		 */
679 		ndp = (char *)(void *)pnd->pnd_dirent + (size_t)*readoff;
680 		fromdent = (struct dirent *)(void *)ndp;
681 
682 		if (*reslen < _DIRENT_SIZE(fromdent))
683 			break;
684 
685 		memcpy(dent, fromdent, _DIRENT_SIZE(fromdent));
686 		*readoff += _DIRENT_SIZE(fromdent);
687 		*reslen -= _DIRENT_SIZE(fromdent);
688 
689 		dent = _DIRENT_NEXT(dent);
690 	}
691 
692 #ifdef PERFUSE_DEBUG
693 	if (perfuse_diagflags & PDF_READDIR)
694 		DPRINTF("%s: readoff = %"PRId64",  "
695 			"pnd->pnd_dirent_len = %"PRId64"\n",
696 			__func__, *readoff, pnd->pnd_dirent_len);
697 #endif
698 	if (*readoff >=  pnd->pnd_dirent_len) {
699 		free(pnd->pnd_dirent);
700 		pnd->pnd_dirent = NULL;
701 		pnd->pnd_dirent_len = 0;
702 		*eofflag = 1;
703 	}
704 
705 	return 0;
706 }
707 
708 static void
709 requeue_request(pu, opc, type)
710 	struct puffs_usermount *pu;
711 	puffs_cookie_t opc;
712 	enum perfuse_qtype type;
713 {
714 	struct perfuse_cc_queue pcq;
715 	struct perfuse_node_data *pnd;
716 #ifdef PERFUSE_DEBUG
717 	struct perfuse_state *ps;
718 
719 	ps = perfuse_getspecific(pu);
720 #endif
721 
722 	pnd = PERFUSE_NODE_DATA(opc);
723 	pcq.pcq_type = type;
724 	pcq.pcq_cc = puffs_cc_getcc(pu);
725 	TAILQ_INSERT_TAIL(&pnd->pnd_pcq, &pcq, pcq_next);
726 
727 #ifdef PERFUSE_DEBUG
728 	if (perfuse_diagflags & PDF_REQUEUE)
729 		DPRINTF("%s: REQUEUE opc = %p, pcc = %p (%s)\n",
730 		        __func__, (void *)opc, pcq.pcq_cc,
731 			perfuse_qtypestr[type]);
732 #endif
733 
734 	puffs_cc_yield(pcq.pcq_cc);
735 	TAILQ_REMOVE(&pnd->pnd_pcq, &pcq, pcq_next);
736 
737 #ifdef PERFUSE_DEBUG
738 	if (perfuse_diagflags & PDF_REQUEUE)
739 		DPRINTF("%s: RESUME opc = %p, pcc = %p (%s)\n",
740 		        __func__, (void *)opc, pcq.pcq_cc,
741 			perfuse_qtypestr[type]);
742 #endif
743 
744 	return;
745 }
746 
747 /* ARGSUSED0 */
748 static int
749 dequeue_requests(ps, opc, type, max)
750 	struct perfuse_state *ps;
751 	puffs_cookie_t opc;
752 	enum perfuse_qtype type;
753 	int max;
754 {
755 	struct perfuse_cc_queue *pcq;
756 	struct perfuse_node_data *pnd;
757 	int dequeued;
758 
759 	pnd = PERFUSE_NODE_DATA(opc);
760 	dequeued = 0;
761 	TAILQ_FOREACH(pcq, &pnd->pnd_pcq, pcq_next) {
762 		if (pcq->pcq_type != type)
763 			continue;
764 
765 #ifdef PERFUSE_DEBUG
766 		if (perfuse_diagflags & PDF_REQUEUE)
767 			DPRINTF("%s: SCHEDULE opc = %p, pcc = %p (%s)\n",
768 				__func__, (void *)opc, pcq->pcq_cc,
769 				 perfuse_qtypestr[type]);
770 #endif
771 		puffs_cc_schedule(pcq->pcq_cc);
772 
773 		if (++dequeued == max)
774 			break;
775 	}
776 
777 #ifdef PERFUSE_DEBUG
778 	if (perfuse_diagflags & PDF_REQUEUE)
779 		DPRINTF("%s: DONE  opc = %p\n", __func__, (void *)opc);
780 #endif
781 
782 	return dequeued;
783 }
784 
785 void
786 perfuse_fs_init(pu)
787 	struct puffs_usermount *pu;
788 {
789 	struct perfuse_state *ps;
790 	perfuse_msg_t *pm;
791 	struct fuse_init_in *fii;
792 	struct fuse_init_out *fio;
793 	int error;
794 
795 	ps = puffs_getspecific(pu);
796 
797         if (puffs_mount(pu, ps->ps_target, ps->ps_mountflags, ps->ps_root) != 0)
798                 DERR(EX_OSERR, "puffs_mount failed");
799 
800 	/*
801 	 * Linux 2.6.34.1 sends theses flags:
802 	 * FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC
803 	 * FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK
804 	 *
805 	 * Linux also sets max_readahead at 32 pages (128 kB)
806 	 */
807 	pm = ps->ps_new_msg(pu, 0, FUSE_INIT, sizeof(*fii), NULL);
808 	fii = GET_INPAYLOAD(ps, pm, fuse_init_in);
809 	fii->major = FUSE_KERNEL_VERSION;
810 	fii->minor = FUSE_KERNEL_MINOR_VERSION;
811 	fii->max_readahead = 32 * sysconf(_SC_PAGESIZE);
812 	fii->flags = (FUSE_ASYNC_READ|FUSE_POSIX_LOCKS|FUSE_ATOMIC_O_TRUNC);
813 
814 	if ((error = xchg_msg(pu, 0, pm, sizeof(*fio), wait_reply)) != 0)
815 		DERRX(EX_SOFTWARE, "init message exchange failed (%d)", error);
816 
817 	fio = GET_OUTPAYLOAD(ps, pm, fuse_init_out);
818 	ps->ps_max_readahead = fio->max_readahead;
819 	ps->ps_max_write = fio->max_write;
820 
821 	ps->ps_destroy_msg(pm);
822 
823 	return;
824 }
825 
826 int
827 perfuse_fs_unmount(pu, flags)
828 	struct puffs_usermount *pu;
829 	int flags;
830 {
831 	perfuse_msg_t *pm;
832 	struct perfuse_state *ps;
833 	puffs_cookie_t opc;
834 	int error;
835 
836 	ps = puffs_getspecific(pu);
837 
838 	opc = (puffs_cookie_t)puffs_getroot(pu);
839 	pm = ps->ps_new_msg(pu, opc, FUSE_DESTROY, 0, NULL);
840 
841 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0){
842 		DWARN("unmount %s", ps->ps_target);
843 		if (!(flags & MNT_FORCE))
844 			goto out;
845 	}
846 
847 	ps->ps_umount(pu);
848 
849 	if (perfuse_diagflags & PDF_MISC)
850 		DPRINTF("%s unmounted, exit\n", ps->ps_target);
851 
852 	return 0;
853 out:
854 	ps->ps_destroy_msg(pm);
855 
856 	return error;
857 }
858 
859 int
860 perfuse_fs_statvfs(pu, svfsb)
861 	struct puffs_usermount *pu;
862 	struct statvfs *svfsb;
863 {
864 	struct perfuse_state *ps;
865 	perfuse_msg_t *pm;
866 	puffs_cookie_t opc;
867 	struct fuse_statfs_out *fso;
868 	int error;
869 
870 	ps = puffs_getspecific(pu);
871 	opc = (puffs_cookie_t)puffs_getroot(pu);
872 	pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
873 
874 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply)) != 0)
875 		goto out;
876 
877 	fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
878 	svfsb->f_flag = ps->ps_mountflags;
879 	svfsb->f_bsize = fso->st.bsize;
880 	svfsb->f_frsize = fso->st.frsize;
881 	svfsb->f_iosize = ((struct puffs_node *)opc)->pn_va.va_blocksize;
882 	svfsb->f_blocks = fso->st.blocks;
883 	svfsb->f_bfree = fso->st.bfree;
884 	svfsb->f_bavail = fso->st.bavail;
885 	svfsb->f_bresvd = fso->st.bfree - fso->st.bavail;
886 	svfsb->f_files = fso->st.files;
887 	svfsb->f_ffree = fso->st.ffree;
888 	svfsb->f_favail = fso->st.ffree;/* files not reserved for root */
889 	svfsb->f_fresvd = 0;		/* files reserved for root */
890 
891 	svfsb->f_syncreads = ps->ps_syncreads;
892 	svfsb->f_syncwrites = ps->ps_syncwrites;
893 
894 	svfsb->f_asyncreads = ps->ps_asyncreads;
895 	svfsb->f_asyncwrites = ps->ps_asyncwrites;
896 
897 	(void)memcpy(&svfsb->f_fsidx, &ps->ps_fsid, sizeof(ps->ps_fsid));
898 	svfsb->f_fsid = (unsigned long)ps->ps_fsid;
899 	svfsb->f_namemax = MAXPATHLEN;	/* XXX */
900 	svfsb->f_owner = ps->ps_owner_uid;
901 
902 	(void)strlcpy(svfsb->f_mntonname, ps->ps_target, _VFS_NAMELEN);
903 
904 	if (ps->ps_filesystemtype != NULL)
905 		(void)strlcpy(svfsb->f_fstypename,
906 			      ps->ps_filesystemtype, _VFS_NAMELEN);
907 	else
908 		(void)strlcpy(svfsb->f_fstypename, "fuse", _VFS_NAMELEN);
909 
910 	if (ps->ps_source != NULL)
911 		strlcpy(svfsb->f_mntfromname, ps->ps_source, _VFS_NAMELEN);
912 	else
913 		strlcpy(svfsb->f_mntfromname, _PATH_FUSE, _VFS_NAMELEN);
914 out:
915 	ps->ps_destroy_msg(pm);
916 
917 	return error;
918 }
919 
920 int
921 perfuse_fs_sync(pu, waitfor, pcr)
922 	struct puffs_usermount *pu;
923 	int waitfor;
924 	const struct puffs_cred *pcr;
925 {
926 	/*
927 	 * FUSE does not seem to have a FS sync callback.
928 	 * Maybe do not even register this callback
929 	 */
930 	return puffs_fsnop_sync(pu, waitfor, pcr);
931 }
932 
933 /* ARGSUSED0 */
934 int
935 perfuse_fs_fhtonode(pu, fid, fidsize, pni)
936 	struct puffs_usermount *pu;
937 	void *fid;
938 	size_t fidsize;
939 	struct puffs_newinfo *pni;
940 {
941 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
942 	return 0;
943 }
944 
945 /* ARGSUSED0 */
946 int
947 perfuse_fs_nodetofh(pu, cookie, fid, fidsize)
948 	struct puffs_usermount *pu;
949 	puffs_cookie_t cookie;
950 	void *fid;
951 	size_t *fidsize;
952 {
953 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
954 	return 0;
955 }
956 
957 #if 0
958 /* ARGSUSED0 */
959 void
960 perfuse_fs_extattrctl(pu, cmd, cookie, flags, namespace, attrname)
961 	struct puffs_usermount *pu;
962 	int cmd,
963 	puffs_cookie_t *cookie;
964 	int flags;
965 	int namespace;
966 	const char *attrname;
967 {
968 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
969 	return 0;
970 }
971 #endif /* 0 */
972 
973 /* ARGSUSED0 */
974 void
975 perfuse_fs_suspend(pu, status)
976 	struct puffs_usermount *pu;
977 	int status;
978 {
979 	return;
980 }
981 
982 
983 
984 int
985 perfuse_node_lookup(pu, opc, pni, pcn)
986 	struct puffs_usermount *pu;
987 	puffs_cookie_t opc;
988 	struct puffs_newinfo *pni;
989 	const struct puffs_cn *pcn;
990 {
991 	struct puffs_node *pn;
992 	mode_t mode;
993 	int error;
994 
995 	/*
996 	 * Check permissions
997 	 */
998 	switch(pcn->pcn_nameiop) {
999 	case NAMEI_DELETE: /* FALLTHROUGH */
1000 	case NAMEI_RENAME: /* FALLTHROUGH */
1001 	case NAMEI_CREATE:
1002 		if (pcn->pcn_flags & NAMEI_ISLASTCN)
1003 			mode = PUFFS_VEXEC|PUFFS_VWRITE;
1004 		else
1005 			mode = PUFFS_VEXEC;
1006 		break;
1007 	case NAMEI_LOOKUP: /* FALLTHROUGH */
1008 	default:
1009 		mode = PUFFS_VEXEC;
1010 		break;
1011 	}
1012 
1013 	if ((error = mode_access(opc, pcn->pcn_cred, mode)) != 0)
1014 		return error;
1015 
1016 	/*
1017 	 * Special case for ..
1018 	 */
1019 	if (strcmp(pcn->pcn_name, "..") == 0)
1020 		pn = PERFUSE_NODE_DATA(opc)->pnd_parent;
1021 	else
1022 		error = node_lookup_common(pu, (puffs_cookie_t)opc,
1023 					   pcn->pcn_name, &pn);
1024 	if (error != 0)
1025 		return error;
1026 
1027 	/*
1028 	 * Removed node
1029 	 */
1030 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_REMOVED)
1031 		return ENOENT;
1032 
1033 	/*
1034 	 * Check for sticky bit. Unfortunately there is no way to
1035 	 * do this before creating the puffs_node, since we require
1036 	 * this operation to get the node owner.
1037 	 */
1038 	switch (pcn->pcn_nameiop) {
1039 	case NAMEI_DELETE: /* FALLTHROUGH */
1040 	case NAMEI_RENAME:
1041 		error = sticky_access(pn, pcn->pcn_cred);
1042 		if (error != 0) {
1043 			/*
1044 			 * kernel will never know about it and will
1045 			 * not reclaim it. The filesystem needs to
1046 			 * clean it up anyway, therefore mimick a forget.
1047 			 */
1048 			PERFUSE_NODE_DATA(pn)->pnd_flags |= PND_RECLAIMED;
1049 			(void)perfuse_node_reclaim(pu, (puffs_cookie_t)pn);
1050 			return error;
1051 		}
1052 		break;
1053 	default:
1054 		break;
1055 	}
1056 
1057 	/*
1058 	 * If that node had a pending reclaim, wipe it out.
1059 	 */
1060 	PERFUSE_NODE_DATA(pn)->pnd_flags &= ~PND_RECLAIMED;
1061 
1062 	puffs_newinfo_setcookie(pni, pn);
1063 	puffs_newinfo_setvtype(pni, pn->pn_va.va_type);
1064 	puffs_newinfo_setsize(pni, (voff_t)pn->pn_va.va_size);
1065 	puffs_newinfo_setrdev(pni, pn->pn_va.va_rdev);
1066 
1067 	return error;
1068 }
1069 
1070 int
1071 perfuse_node_create(pu, opc, pni, pcn, vap)
1072 	struct puffs_usermount *pu;
1073 	puffs_cookie_t opc;
1074 	struct puffs_newinfo *pni;
1075 	const struct puffs_cn *pcn;
1076 	const struct vattr *vap;
1077 {
1078 	perfuse_msg_t *pm;
1079 	struct perfuse_state *ps;
1080 	struct fuse_create_in *fci;
1081 	struct fuse_entry_out *feo;
1082 	struct fuse_open_out *foo;
1083 	struct puffs_node *pn;
1084 	const char *name;
1085 	size_t namelen;
1086 	size_t len;
1087 	int error;
1088 
1089 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1090 		return ENOENT;
1091 
1092 	/*
1093 	 * If create is unimplemented: Check that it does not
1094 	 * already exists, and if not, do mknod and open
1095 	 */
1096 	ps = puffs_getspecific(pu);
1097 	if (ps->ps_flags & PS_NO_CREAT) {
1098 		error = node_lookup_common(pu, opc, pcn->pcn_name, &pn);
1099 		if (error == 0)
1100 			return EEXIST;
1101 
1102 		error = perfuse_node_mknod(pu, opc, pni, pcn, vap);
1103 		if (error != 0)
1104 			return error;
1105 
1106 		error = node_lookup_common(pu, opc, pcn->pcn_name, &pn);
1107 		if (error != 0)
1108 			return error;
1109 
1110 		/*
1111 		 * FUSE does the open at create time, while
1112 		 * NetBSD will open in a subsequent operation.
1113 		 * We need to open now, in order to retain FUSE
1114 		 * semantics. The calling process will not get
1115 		 * a file descriptor before the kernel sends
1116 		 * the open operation.
1117 		 */
1118 		opc = (puffs_cookie_t)pn;
1119 		error = perfuse_node_open(pu, opc, FWRITE, pcn->pcn_cred);
1120 		if (error != 0)
1121 			return error;
1122 
1123 		return 0;
1124 	}
1125 
1126 	name = pcn->pcn_name;
1127 	namelen = pcn->pcn_namelen + 1;
1128 	len = sizeof(*fci) + namelen;
1129 
1130 	/*
1131 	 * flags should use O_WRONLY instead of O_RDWR, but it
1132 	 * breaks when the caller tries to read from file.
1133 	 *
1134 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1135 	 */
1136 	pm = ps->ps_new_msg(pu, opc, FUSE_CREATE, len, NULL);
1137 	fci = GET_INPAYLOAD(ps, pm, fuse_create_in);
1138 	fci->flags = O_CREAT | O_TRUNC | O_RDWR;
1139 	fci->mode = vap->va_mode | VTTOIF(vap->va_type);
1140 	fci->umask = 0; 	/* Seems unused by libfuse */
1141 	(void)strlcpy((char*)(void *)(fci + 1), name, namelen);
1142 
1143 	len = sizeof(*feo) + sizeof(*foo);
1144 	if ((error = xchg_msg(pu, opc, pm, len, wait_reply)) != 0)
1145 		goto out;
1146 
1147 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
1148 	foo = (struct fuse_open_out *)(void *)(feo + 1);
1149 	if (feo->nodeid == PERFUSE_UNKNOWN_INO)
1150 		DERRX(EX_SOFTWARE, "%s: no ino", __func__);
1151 
1152 	/*
1153 	 * Save the file handle and inode in node private data
1154 	 * so that we can reuse it later
1155 	 */
1156 	pn = perfuse_new_pn(pu, name, opc);
1157 	perfuse_new_fh((puffs_cookie_t)pn, foo->fh, FWRITE);
1158 	PERFUSE_NODE_DATA(pn)->pnd_ino = feo->nodeid;
1159 
1160 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
1161 	pn->pn_va.va_gen = (u_long)(feo->generation);
1162 
1163 	puffs_newinfo_setcookie(pni, pn);
1164 
1165 #ifdef PERFUSE_DEBUG
1166 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1167 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
1168 			"ino = %"PRId64", wfh = 0x%"PRIx64"\n",
1169 			__func__, (void *)pn, pcn->pcn_name,
1170 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid, foo->fh);
1171 #endif
1172 
1173 	ps->ps_destroy_msg(pm);
1174 
1175 	return node_mk_common_final(pu, opc, pn, pcn);
1176 
1177 out:
1178 	ps->ps_destroy_msg(pm);
1179 
1180 	/*
1181 	 * create is unimplmented, remember it for later,
1182 	 * and start over using mknod and open instead.
1183 	 */
1184 	if (error == ENOSYS) {
1185 		ps->ps_flags |= PS_NO_CREAT;
1186 		return perfuse_node_create(pu, opc, pni, pcn, vap);
1187 	}
1188 
1189 	return error;
1190 }
1191 
1192 
1193 int
1194 perfuse_node_mknod(pu, opc, pni, pcn, vap)
1195 	struct puffs_usermount *pu;
1196 	puffs_cookie_t opc;
1197 	struct puffs_newinfo *pni;
1198 	const struct puffs_cn *pcn;
1199 	const struct vattr *vap;
1200 {
1201 	struct perfuse_state *ps;
1202 	perfuse_msg_t *pm;
1203 	struct fuse_mknod_in *fmi;
1204 	const char* path;
1205 	size_t len;
1206 
1207 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1208 		return ENOENT;
1209 
1210 	/*
1211 	 * Only superuser can mknod objects other than
1212 	 * directories, files, socks, fifo and links.
1213 	 *
1214 	 * Create an object require -WX permission in the parent directory
1215 	 */
1216 	switch (vap->va_type) {
1217 	case VDIR:	/* FALLTHROUGH */
1218 	case VREG:	/* FALLTHROUGH */
1219 	case VFIFO:	/* FALLTHROUGH */
1220 	case VSOCK:
1221 		break;
1222 	default:	/* VNON, VBLK, VCHR, VBAD */
1223 		if (!puffs_cred_isjuggernaut(pcn->pcn_cred))
1224 			return EACCES;
1225 		break;
1226 	}
1227 
1228 
1229 	ps = puffs_getspecific(pu);
1230 	path = pcn->pcn_name;
1231 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
1232 
1233 	/*
1234 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1235 	 */
1236 	pm = ps->ps_new_msg(pu, opc, FUSE_MKNOD, len, NULL);
1237 	fmi = GET_INPAYLOAD(ps, pm, fuse_mknod_in);
1238 	fmi->mode = vap->va_mode | VTTOIF(vap->va_type);
1239 	fmi->rdev = (uint32_t)vap->va_rdev;
1240 	fmi->umask = 0; 	/* Seems unused bu libfuse */
1241 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
1242 
1243 	return node_mk_common(pu, opc, pni, pcn, pm);
1244 }
1245 
1246 
1247 int
1248 perfuse_node_open(pu, opc, mode, pcr)
1249 	struct puffs_usermount *pu;
1250 	puffs_cookie_t opc;
1251 	int mode;
1252 	const struct puffs_cred *pcr;
1253 {
1254 	struct perfuse_state *ps;
1255 	struct perfuse_node_data *pnd;
1256 	perfuse_msg_t *pm;
1257 	mode_t fmode;
1258 	int op;
1259 	struct fuse_open_in *foi;
1260 	struct fuse_open_out *foo;
1261 	struct puffs_node *pn;
1262 	int error;
1263 
1264 	ps = puffs_getspecific(pu);
1265 	pn = (struct puffs_node *)opc;
1266 	pnd = PERFUSE_NODE_DATA(opc);
1267 	pm = NULL;
1268 	error = 0;
1269 
1270 	if (pnd->pnd_flags & PND_REMOVED)
1271 		return ENOENT;
1272 
1273 	if (puffs_pn_getvap(pn)->va_type == VDIR)
1274 		op = FUSE_OPENDIR;
1275 	else
1276 		op = FUSE_OPEN;
1277 
1278 	/*
1279 	 * libfuse docs says
1280 	 * - O_CREAT and O_EXCL should never be set.
1281 	 * - O_TRUNC may be used if mount option atomic_o_trunc is used XXX
1282 	 *
1283 	 * O_APPEND makes no sense since FUSE always sends
1284 	 * the file offset for write operations. If the
1285 	 * filesystem uses pwrite(), O_APPEND would cause
1286 	 * the offset to be ignored and cause file corruption.
1287 	 */
1288 	mode &= ~(O_CREAT|O_EXCL|O_APPEND);
1289 
1290 	/*
1291 	 * Do not open twice, and do not reopen for reading
1292 	 * if we already have write handle.
1293 	 */
1294 	if (((mode & FREAD) && (pnd->pnd_flags & PND_RFH)) ||
1295 	    ((mode & FREAD) && (pnd->pnd_flags & PND_WFH)) ||
1296 	    ((mode & FWRITE) && (pnd->pnd_flags & PND_WFH))) {
1297 		error = 0;
1298 		goto out;
1299 	}
1300 
1301 	/*
1302 	 * Queue open on a node so that we do not open
1303 	 * twice. This would be better with read and
1304 	 * write distinguished.
1305 	 */
1306 	while (pnd->pnd_flags & PND_INOPEN)
1307 		requeue_request(pu, opc, PCQ_OPEN);
1308 	pnd->pnd_flags |= PND_INOPEN;
1309 
1310 	/*
1311 	 * Convert PUFFS mode to FUSE mode: convert FREAD/FWRITE
1312 	 * to O_RDONLY/O_WRONLY while perserving the other options.
1313 	 */
1314 	fmode = mode & ~(FREAD|FWRITE);
1315 	fmode |= (mode & FWRITE) ? O_RDWR : O_RDONLY;
1316 
1317 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*foi), pcr);
1318 	foi = GET_INPAYLOAD(ps, pm, fuse_open_in);
1319 	foi->flags = fmode;
1320 	foi->unused = 0;
1321 
1322 	if ((error = xchg_msg(pu, opc, pm, sizeof(*foo), wait_reply)) != 0)
1323 		goto out;
1324 
1325 	foo = GET_OUTPAYLOAD(ps, pm, fuse_open_out);
1326 
1327 	/*
1328 	 * Save the file handle in node private data
1329 	 * so that we can reuse it later
1330 	 */
1331 	perfuse_new_fh(opc, foo->fh, mode);
1332 
1333 #ifdef PERFUSE_DEBUG
1334 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1335 		DPRINTF("%s: opc = %p, file = \"%s\", "
1336 			"ino = %"PRId64", %s%sfh = 0x%"PRIx64"\n",
1337 			__func__, (void *)opc, perfuse_node_path(opc),
1338 			pnd->pnd_ino, mode & FREAD ? "r" : "",
1339 			mode & FWRITE ? "w" : "", foo->fh);
1340 #endif
1341 
1342 out:
1343 	if (pm != NULL)
1344 		ps->ps_destroy_msg(pm);
1345 
1346 	pnd->pnd_flags &= ~PND_INOPEN;
1347 	(void)dequeue_requests(ps, opc, PCQ_OPEN, DEQUEUE_ALL);
1348 
1349 	return error;
1350 }
1351 
1352 /* ARGSUSED0 */
1353 int
1354 perfuse_node_close(pu, opc, flags, pcr)
1355 	struct puffs_usermount *pu;
1356 	puffs_cookie_t opc;
1357 	int flags;
1358 	const struct puffs_cred *pcr;
1359 {
1360 	struct perfuse_node_data *pnd;
1361 
1362 	pnd = PERFUSE_NODE_DATA(opc);
1363 
1364 	if (!(pnd->pnd_flags & PND_OPEN))
1365 		return EBADF;
1366 
1367 	/*
1368 	 * Actual close is postponed at inactive time.
1369 	 */
1370 	return 0;
1371 }
1372 
1373 int
1374 perfuse_node_access(pu, opc, mode, pcr)
1375 	struct puffs_usermount *pu;
1376 	puffs_cookie_t opc;
1377 	int mode;
1378 	const struct puffs_cred *pcr;
1379 {
1380 	perfuse_msg_t *pm;
1381 	struct perfuse_state *ps;
1382 	struct fuse_access_in *fai;
1383 	int error;
1384 
1385 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1386 		return ENOENT;
1387 
1388 	/*
1389 	 * If we previously detected the filesystem does not
1390 	 * implement access(), short-circuit the call and skip
1391 	 * to libpuffs access() emulation.
1392 	 */
1393 	ps = puffs_getspecific(pu);
1394 	if (ps->ps_flags & PS_NO_ACCESS) {
1395 		const struct vattr *vap;
1396 
1397 		vap = puffs_pn_getvap((struct puffs_node *)opc);
1398 
1399 		error = puffs_access(IFTOVT(vap->va_mode),
1400 				     vap->va_mode & ACCESSPERMS,
1401 				     vap->va_uid, vap->va_gid,
1402 				     (mode_t)mode, pcr);
1403 		return error;
1404 	}
1405 
1406 	/*
1407 	 * Plain access call
1408 	 */
1409 	pm = ps->ps_new_msg(pu, opc, FUSE_ACCESS, sizeof(*fai), pcr);
1410 	fai = GET_INPAYLOAD(ps, pm, fuse_access_in);
1411 	fai->mask = 0;
1412 	fai->mask |= (mode & PUFFS_VREAD) ? R_OK : 0;
1413 	fai->mask |= (mode & PUFFS_VWRITE) ? W_OK : 0;
1414 	fai->mask |= (mode & PUFFS_VEXEC) ? X_OK : 0;
1415 
1416 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
1417 
1418 	ps->ps_destroy_msg(pm);
1419 
1420 	/*
1421 	 * If unimplemented, start over with emulation
1422 	 */
1423 	if (error == ENOSYS) {
1424 		ps->ps_flags |= PS_NO_ACCESS;
1425 		return perfuse_node_access(pu, opc, mode, pcr);
1426 	}
1427 
1428 	return error;
1429 }
1430 
1431 int
1432 perfuse_node_getattr(pu, opc, vap, pcr)
1433 	struct puffs_usermount *pu;
1434 	puffs_cookie_t opc;
1435 	struct vattr *vap;
1436 	const struct puffs_cred *pcr;
1437 {
1438 	perfuse_msg_t *pm;
1439 	struct perfuse_state *ps;
1440 	struct fuse_getattr_in *fgi;
1441 	struct fuse_attr_out *fao;
1442 	u_quad_t va_size;
1443 	int error;
1444 
1445 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1446 		return ENOENT;
1447 
1448 	ps = puffs_getspecific(pu);
1449 	va_size = vap->va_size;
1450 
1451 	/*
1452 	 * FUSE_GETATTR_FH must be set in fgi->flags
1453 	 * if we use for fgi->fh
1454 	 */
1455 	pm = ps->ps_new_msg(pu, opc, FUSE_GETATTR, sizeof(*fgi), pcr);
1456 	fgi = GET_INPAYLOAD(ps, pm, fuse_getattr_in);
1457 	fgi->getattr_flags = 0;
1458 	fgi->dummy = 0;
1459 	fgi->fh = 0;
1460 
1461 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_OPEN) {
1462 		fgi->fh = perfuse_get_fh(opc, FREAD);
1463 		fgi->getattr_flags |= FUSE_GETATTR_FH;
1464 	}
1465 
1466 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
1467 		goto out;
1468 
1469 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1470 
1471 	/*
1472 	 * The message from filesystem has a cache timeout
1473 	 * XXX this is ignored yet, is that right?
1474 	 *
1475 	 * We also set birthtime, flags, filerev,vaflags to 0.
1476 	 * This seems the best bet, since the information is
1477 	 * not available from filesystem.
1478 	 */
1479 	fuse_attr_to_vap(ps, vap, &fao->attr);
1480 
1481 	/*
1482 	 * If a write is in progress, do not trust filesystem opinion
1483 	 * of file size, use the one from kernel.
1484 	 */
1485 	if ((PERFUSE_NODE_DATA(opc)->pnd_flags & PND_INWRITE) &&
1486 	    (va_size != (u_quad_t)PUFFS_VNOVAL))
1487 		vap->va_size = MAX(va_size, vap->va_size);;
1488 out:
1489 	ps->ps_destroy_msg(pm);
1490 
1491 	return error;
1492 }
1493 
1494 int
1495 perfuse_node_setattr(pu, opc, vap, pcr)
1496 	struct puffs_usermount *pu;
1497 	puffs_cookie_t opc;
1498 	const struct vattr *vap;
1499 	const struct puffs_cred *pcr;
1500 {
1501 	perfuse_msg_t *pm;
1502 	uint64_t fh;
1503 	struct perfuse_state *ps;
1504 	struct perfuse_node_data *pnd;
1505 	struct fuse_setattr_in *fsi;
1506 	struct fuse_attr_out *fao;
1507 	struct vattr *old_va;
1508 	u_quad_t va_size;
1509 	int error;
1510 
1511 	ps = puffs_getspecific(pu);
1512 	pnd = PERFUSE_NODE_DATA(opc);
1513 	pm = NULL;
1514 
1515 	/*
1516 	 * The only operation we can do once the file is removed
1517 	 * is to resize it, and we can do it only if it is open.
1518 	 * Do not even send the operation to the filesystem: the
1519 	 * file is not there anymore.
1520 	 */
1521 	if (pnd->pnd_flags & PND_REMOVED) {
1522 		if (!(pnd->pnd_flags & PND_OPEN))
1523 			return ENOENT;
1524 
1525 		error = 0;
1526 		goto out;
1527 	}
1528 
1529 	old_va = puffs_pn_getvap((struct puffs_node *)opc);
1530 
1531 	/*
1532 	 * Check for permission to change size
1533 	 */
1534 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1535 	    (error = mode_access(opc, pcr, PUFFS_VWRITE)) != 0)
1536 		return error;
1537 
1538 	/*
1539 	 * Check for permission to change dates
1540 	 */
1541 	if (((vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1542 	     (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL)) &&
1543 	    (puffs_access_times(old_va->va_uid, old_va->va_gid,
1544 				old_va->va_mode, 0, pcr) != 0))
1545 		return EACCES;
1546 
1547 	/*
1548 	 * Check for permission to change owner and group
1549 	 */
1550 	if (((vap->va_uid != (uid_t)PUFFS_VNOVAL) ||
1551 	     (vap->va_gid != (gid_t)PUFFS_VNOVAL)) &&
1552 	    (puffs_access_chown(old_va->va_uid, old_va->va_gid,
1553 				vap->va_uid, vap->va_gid, pcr)) != 0)
1554 		return EACCES;
1555 
1556 	/*
1557 	 * Check for permission to change permissions
1558 	 */
1559 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1560 	    (puffs_access_chmod(old_va->va_uid, old_va->va_gid,
1561 				old_va->va_type, vap->va_mode, pcr)) != 0)
1562 		return EACCES;
1563 
1564 	/*
1565 	 * If a write is in progress, set the highest
1566 	 * value in the filesystem, otherwise we break
1567 	 * IO_APPEND.
1568 	 */
1569 	va_size = vap->va_size;
1570 	if ((pnd->pnd_flags & PND_INWRITE) &&
1571 	    (va_size != (u_quad_t)PUFFS_VNOVAL))
1572 		va_size = MAX(va_size, old_va->va_size);
1573 
1574 	pm = ps->ps_new_msg(pu, opc, FUSE_SETATTR, sizeof(*fsi), pcr);
1575 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
1576 	fsi->valid = 0;
1577 
1578 	/*
1579 	 * Get a fh if the node is open for writing
1580 	 */
1581 	if (pnd->pnd_flags & PND_WFH) {
1582 		fh = perfuse_get_fh(opc, FWRITE);
1583 		fsi->fh = fh;
1584 		fsi->valid |= FUSE_FATTR_FH;
1585 	}
1586 
1587 	if (va_size != (u_quad_t)PUFFS_VNOVAL) {
1588 		fsi->size = va_size;
1589 		fsi->valid |= FUSE_FATTR_SIZE;
1590 	}
1591 
1592 	/*
1593  	 * Setting mtime without atime or vice versa leads to
1594 	 * dates being reset to Epoch on glusterfs. If one
1595 	 * is missing, use the old value.
1596  	 */
1597 	if ((vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1598 	    (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL)) {
1599 
1600 		if (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) {
1601 			fsi->atime = vap->va_atime.tv_sec;
1602 			fsi->atimensec = (uint32_t)vap->va_atime.tv_nsec;
1603 		} else {
1604 			fsi->atime = old_va->va_atime.tv_sec;
1605 			fsi->atimensec = (uint32_t)old_va->va_atime.tv_nsec;
1606 		}
1607 
1608 		if (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) {
1609 			fsi->mtime = vap->va_mtime.tv_sec;
1610 			fsi->mtimensec = (uint32_t)vap->va_mtime.tv_nsec;
1611 		} else {
1612 			fsi->mtime = old_va->va_mtime.tv_sec;
1613 			fsi->mtimensec = (uint32_t)old_va->va_mtime.tv_nsec;
1614 		}
1615 
1616 		fsi->valid |= (FUSE_FATTR_MTIME|FUSE_FATTR_ATIME);
1617 	}
1618 
1619 	if (vap->va_mode != (mode_t)PUFFS_VNOVAL) {
1620 		fsi->mode = vap->va_mode;
1621 		fsi->valid |= FUSE_FATTR_MODE;
1622 	}
1623 
1624 	if (vap->va_uid != (uid_t)PUFFS_VNOVAL) {
1625 		fsi->uid = vap->va_uid;
1626 		fsi->valid |= FUSE_FATTR_UID;
1627 	}
1628 
1629 	if (vap->va_gid != (gid_t)PUFFS_VNOVAL) {
1630 		fsi->gid = vap->va_gid;
1631 		fsi->valid |= FUSE_FATTR_GID;
1632 	}
1633 
1634 	if (pnd->pnd_lock_owner != 0) {
1635 		fsi->lock_owner = pnd->pnd_lock_owner;
1636 		fsi->valid |= FUSE_FATTR_LOCKOWNER;
1637 	}
1638 
1639 	/*
1640 	 * If nothing remain, discard the operation.
1641 	 */
1642 	if (!(fsi->valid & (FUSE_FATTR_SIZE|FUSE_FATTR_ATIME|FUSE_FATTR_MTIME|
1643 			    FUSE_FATTR_MODE|FUSE_FATTR_UID|FUSE_FATTR_GID))) {
1644 		error = 0;
1645 		goto out;
1646 	}
1647 
1648 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
1649 		goto out;
1650 
1651 	/*
1652 	 * Copy back the new values
1653 	 */
1654 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1655 	fuse_attr_to_vap(ps, old_va, &fao->attr);
1656 out:
1657 
1658 	if (pm != NULL)
1659 		ps->ps_destroy_msg(pm);
1660 
1661 	return error;
1662 }
1663 
1664 int
1665 perfuse_node_poll(pu, opc, events)
1666 	struct puffs_usermount *pu;
1667 	puffs_cookie_t opc;
1668 	int *events;
1669 {
1670 	struct perfuse_state *ps;
1671 	perfuse_msg_t *pm;
1672 	struct fuse_poll_in *fpi;
1673 	struct fuse_poll_out *fpo;
1674 	int error;
1675 
1676 	ps = puffs_getspecific(pu);
1677 	/*
1678 	 * kh is set if FUSE_POLL_SCHEDULE_NOTIFY is set.
1679  	 */
1680 	pm = ps->ps_new_msg(pu, opc, FUSE_POLL, sizeof(*fpi), NULL);
1681 	fpi = GET_INPAYLOAD(ps, pm, fuse_poll_in);
1682 	fpi->fh = perfuse_get_fh(opc, FREAD);
1683 	fpi->kh = 0;
1684 	fpi->flags = 0;
1685 
1686 #ifdef PERFUSE_DEBUG
1687 	if (perfuse_diagflags & PDF_FH)
1688 		DPRINTF("%s: opc = %p, ino = %"PRId64", fh = 0x%"PRIx64"\n",
1689 			__func__, (void *)opc,
1690 			PERFUSE_NODE_DATA(opc)->pnd_ino, fpi->fh);
1691 #endif
1692 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fpo), wait_reply)) != 0)
1693 		goto out;
1694 
1695 	fpo = GET_OUTPAYLOAD(ps, pm, fuse_poll_out);
1696 	*events = fpo->revents;
1697 out:
1698 	ps->ps_destroy_msg(pm);
1699 
1700 	return error;
1701 }
1702 
1703 /* ARGSUSED0 */
1704 int
1705 perfuse_node_mmap(pu, opc, flags, pcr)
1706 	struct puffs_usermount *pu;
1707 	puffs_cookie_t opc;
1708 	int flags;
1709 	const struct puffs_cred *pcr;
1710 {
1711 	/*
1712 	 * Not implemented anymore in libfuse
1713 	 */
1714 	return ENOSYS;
1715 }
1716 
1717 /* ARGSUSED2 */
1718 int
1719 perfuse_node_fsync(pu, opc, pcr, flags, offlo, offhi)
1720 	struct puffs_usermount *pu;
1721 	puffs_cookie_t opc;
1722 	const struct puffs_cred *pcr;
1723 	int flags;
1724 	off_t offlo;
1725 	off_t offhi;
1726 {
1727 	int op;
1728 	perfuse_msg_t *pm;
1729 	struct perfuse_state *ps;
1730 	struct perfuse_node_data *pnd;
1731 	struct fuse_fsync_in *ffi;
1732 	uint64_t fh;
1733 	int error;
1734 
1735 	pm = NULL;
1736 	ps = puffs_getspecific(pu);
1737 	pnd = PERFUSE_NODE_DATA(opc);
1738 
1739 	/*
1740 	 * No need to sync a removed node
1741 	 */
1742 	if (pnd->pnd_flags & PND_REMOVED)
1743 		return 0;
1744 
1745 	/*
1746 	 * We do not sync closed files. They have been
1747 	 * sync at inactive time already.
1748 	 */
1749 	if (!(pnd->pnd_flags & PND_OPEN))
1750 		return 0;
1751 
1752 	if (puffs_pn_getvap((struct puffs_node *)opc)->va_type == VDIR)
1753 		op = FUSE_FSYNCDIR;
1754 	else 		/* VREG but also other types such as VLNK */
1755 		op = FUSE_FSYNC;
1756 
1757 	/*
1758 	 * Do not sync if there are no change to sync
1759 	 * XXX remove that test on files if we implement mmap
1760 	 */
1761 #ifdef PERFUSE_DEBUG
1762 	if (perfuse_diagflags & PDF_SYNC)
1763 		DPRINTF("%s: TEST opc = %p, file = \"%s\" is %sdirty\n",
1764 			__func__, (void*)opc, perfuse_node_path(opc),
1765 			pnd->pnd_flags & PND_DIRTY ? "" : "not ");
1766 #endif
1767 	if (!(pnd->pnd_flags & PND_DIRTY))
1768 		return 0;
1769 
1770 	/*
1771 	 * It seems NetBSD can call fsync without open first
1772 	 * glusterfs complain in such a situation:
1773 	 * "FSYNC() ERR => -1 (Invalid argument)"
1774 	 * The file will be closed at inactive time.
1775 	 *
1776 	 * We open the directory for reading in order to sync.
1777 	 * This sounds rather counterintuitive, but it works.
1778 	 */
1779 	if (!(pnd->pnd_flags & PND_WFH)) {
1780 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
1781 			goto out;
1782 	}
1783 
1784 	if (op == FUSE_FSYNCDIR)
1785 		fh = perfuse_get_fh(opc, FREAD);
1786 	else
1787 		fh = perfuse_get_fh(opc, FWRITE);
1788 
1789 	/*
1790 	 * If fsync_flags  is set, meta data should not be flushed.
1791 	 */
1792 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*ffi), NULL);
1793 	ffi = GET_INPAYLOAD(ps, pm, fuse_fsync_in);
1794 	ffi->fh = fh;
1795 	ffi->fsync_flags = (flags & FFILESYNC) ? 0 : 1;
1796 
1797 #ifdef PERFUSE_DEBUG
1798 	if (perfuse_diagflags & PDF_FH)
1799 		DPRINTF("%s: opc = %p, ino = %"PRId64", fh = 0x%"PRIx64"\n",
1800 			__func__, (void *)opc,
1801 			PERFUSE_NODE_DATA(opc)->pnd_ino, ffi->fh);
1802 #endif
1803 
1804 	if ((error = xchg_msg(pu, opc, pm,
1805 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
1806 		goto out;
1807 
1808 	/*
1809 	 * No reply beyond fuse_out_header: nothing to do on success
1810 	 * just clear the dirty flag
1811 	 */
1812 	pnd->pnd_flags &= ~PND_DIRTY;
1813 
1814 #ifdef PERFUSE_DEBUG
1815 	if (perfuse_diagflags & PDF_SYNC)
1816 		DPRINTF("%s: CLEAR opc = %p, file = \"%s\"\n",
1817 			__func__, (void*)opc, perfuse_node_path(opc));
1818 #endif
1819 
1820 out:
1821 	/*
1822 	 * ENOSYS is not returned to kernel,
1823 	 */
1824 	if (error == ENOSYS)
1825 		error = 0;
1826 
1827 	if (pm != NULL)
1828 		ps->ps_destroy_msg(pm);
1829 
1830 	return error;
1831 }
1832 
1833 /* ARGSUSED0 */
1834 int
1835 perfuse_node_seek(pu, opc, oldoff, newoff,  pcr)
1836 	struct puffs_usermount *pu;
1837 	puffs_cookie_t opc;
1838 	off_t oldoff;
1839 	off_t newoff;
1840 	const struct puffs_cred *pcr;
1841 {
1842 	return 0;
1843 }
1844 
1845 int
1846 perfuse_node_remove(pu, opc, targ, pcn)
1847 	struct puffs_usermount *pu;
1848 	puffs_cookie_t opc;
1849 	puffs_cookie_t targ;
1850 	const struct puffs_cn *pcn;
1851 {
1852 	struct perfuse_state *ps;
1853 	struct perfuse_node_data *pnd;
1854 	perfuse_msg_t *pm;
1855 	char *path;
1856 	const char *name;
1857 	size_t len;
1858 	int error;
1859 
1860 	pnd = PERFUSE_NODE_DATA(opc);
1861 
1862 	if ((pnd->pnd_flags & PND_REMOVED) ||
1863 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
1864 		return ENOENT;
1865 
1866 #ifdef PERFUSE_DEBUG
1867 	if (targ == NULL)
1868 		DERRX(EX_SOFTWARE, "%s: targ is NULL", __func__);
1869 
1870 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1871 		DPRINTF("%s: opc = %p, remove opc = %p, file = \"%s\"\n",
1872 			__func__, (void *)opc, (void *)targ, pcn->pcn_name);
1873 #endif
1874 	/*
1875 	 * Await for all operations on the deleted node to drain,
1876 	 * as the filesystem may be confused to have it deleted
1877 	 * during a getattr
1878 	 */
1879 	while (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_INXCHG)
1880 		requeue_request(pu, targ, PCQ_AFTERXCHG);
1881 
1882 	ps = puffs_getspecific(pu);
1883 	pnd = PERFUSE_NODE_DATA(opc);
1884 	name = pcn->pcn_name;
1885 	len = pcn->pcn_namelen + 1;
1886 
1887 	pm = ps->ps_new_msg(pu, opc, FUSE_UNLINK, len, NULL);
1888 	path = _GET_INPAYLOAD(ps, pm, char *);
1889 	(void)strlcpy(path, name, len);
1890 
1891 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
1892 		goto out;
1893 
1894 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
1895 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
1896 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
1897 
1898 	/*
1899 	 * The parent directory needs a sync
1900 	 */
1901 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
1902 
1903 #ifdef PERFUSE_DEBUG
1904 	if (perfuse_diagflags & PDF_FILENAME)
1905 		DPRINTF("%s: remove nodeid = %"PRId64" file = \"%s\"\n",
1906 			__func__, PERFUSE_NODE_DATA(targ)->pnd_ino,
1907 			pcn->pcn_name);
1908 #endif
1909 out:
1910 	ps->ps_destroy_msg(pm);
1911 
1912 	return error;
1913 }
1914 
1915 int
1916 perfuse_node_link(pu, opc, targ, pcn)
1917 	struct puffs_usermount *pu;
1918 	puffs_cookie_t opc;
1919 	puffs_cookie_t targ;
1920 	const struct puffs_cn *pcn;
1921 {
1922 	struct perfuse_state *ps;
1923 	perfuse_msg_t *pm;
1924 	const char *name;
1925 	size_t len;
1926 	struct puffs_node *pn;
1927 	struct fuse_link_in *fli;
1928 	int error;
1929 
1930 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1931 		return ENOENT;
1932 
1933 	ps = puffs_getspecific(pu);
1934 	pn = (struct puffs_node *)targ;
1935 	name = pcn->pcn_name;
1936 	len =  sizeof(*fli) + pcn->pcn_namelen + 1;
1937 
1938 	pm = ps->ps_new_msg(pu, opc, FUSE_LINK, len, NULL);
1939 	fli = GET_INPAYLOAD(ps, pm, fuse_link_in);
1940 	fli->oldnodeid = PERFUSE_NODE_DATA(pn)->pnd_ino;
1941 	(void)strlcpy((char *)(void *)(fli + 1), name, len - sizeof(*fli));
1942 
1943 	error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
1944 
1945 	ps->ps_destroy_msg(pm);
1946 
1947 	return error;
1948 }
1949 
1950 int
1951 perfuse_node_rename(pu, opc, src, pcn_src, targ_dir, targ, pcn_targ)
1952 	struct puffs_usermount *pu;
1953 	puffs_cookie_t opc;
1954 	puffs_cookie_t src;
1955 	const struct puffs_cn *pcn_src;
1956 	puffs_cookie_t targ_dir;
1957 	puffs_cookie_t targ;
1958 	const struct puffs_cn *pcn_targ;
1959 {
1960 	struct perfuse_state *ps;
1961 	perfuse_msg_t *pm;
1962 	struct fuse_rename_in *fri;
1963 	const char *newname;
1964 	const char *oldname;
1965 	char *np;
1966 	int error;
1967 	size_t len;
1968 	size_t newname_len;
1969 	size_t oldname_len;
1970 
1971 	if ((PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED) ||
1972 	    (PERFUSE_NODE_DATA(src)->pnd_flags & PND_REMOVED) ||
1973 	    (PERFUSE_NODE_DATA(targ_dir)->pnd_flags & PND_REMOVED))
1974 		return ENOENT;
1975 
1976 	/*
1977 	 * Await for all operations on the deleted node to drain,
1978 	 * as the filesystem may be confused to have it deleted
1979 	 * during a getattr
1980 	 */
1981 	if ((struct puffs_node *)targ != NULL) {
1982 		while (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_INXCHG)
1983 			requeue_request(pu, targ, PCQ_AFTERXCHG);
1984 	} else {
1985 		while (PERFUSE_NODE_DATA(src)->pnd_flags & PND_INXCHG)
1986 			requeue_request(pu, src, PCQ_AFTERXCHG);
1987 	}
1988 
1989 	ps = puffs_getspecific(pu);
1990 	newname =  pcn_targ->pcn_name;
1991 	newname_len = pcn_targ->pcn_namelen + 1;
1992 	oldname =  pcn_src->pcn_name;
1993 	oldname_len = pcn_src->pcn_namelen + 1;
1994 
1995 	len = sizeof(*fri) + oldname_len + newname_len;
1996 	pm = ps->ps_new_msg(pu, opc, FUSE_RENAME, len, NULL);
1997 	fri = GET_INPAYLOAD(ps, pm, fuse_rename_in);
1998 	fri->newdir = PERFUSE_NODE_DATA(targ_dir)->pnd_ino;
1999 	np = (char *)(void *)(fri + 1);
2000 	(void)strlcpy(np, oldname, oldname_len);
2001 	np += oldname_len;
2002 	(void)strlcpy(np, newname, newname_len);
2003 
2004 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2005 		goto out;
2006 
2007 	if (opc != targ_dir) {
2008 		struct perfuse_node_data *srcdir_pnd;
2009 		struct perfuse_node_data *dstdir_pnd;
2010 		struct perfuse_node_data *src_pnd;
2011 
2012 		srcdir_pnd = PERFUSE_NODE_DATA(opc);
2013 		dstdir_pnd = PERFUSE_NODE_DATA(targ_dir);
2014 		src_pnd = PERFUSE_NODE_DATA(src);
2015 
2016 		TAILQ_REMOVE(&srcdir_pnd->pnd_children, src_pnd, pnd_next);
2017 		TAILQ_INSERT_TAIL(&dstdir_pnd->pnd_children, src_pnd, pnd_next);
2018 
2019 		srcdir_pnd->pnd_childcount--;
2020 		dstdir_pnd->pnd_childcount++;
2021 
2022 		src_pnd->pnd_parent = targ_dir;
2023 
2024 		PERFUSE_NODE_DATA(targ_dir)->pnd_flags |= PND_DIRTY;
2025 	}
2026 
2027 	(void)strlcpy(PERFUSE_NODE_DATA(src)->pnd_name, newname, MAXPATHLEN);
2028 
2029 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2030 
2031 	if ((struct puffs_node *)targ != NULL)
2032 		PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2033 
2034 #ifdef PERFUSE_DEBUG
2035 	if (perfuse_diagflags & PDF_FILENAME)
2036 		DPRINTF("%s: nodeid = %"PRId64" file = \"%s\" renamed \"%s\" "
2037 			"nodeid = %"PRId64" -> nodeid = %"PRId64" \"%s\"\n",
2038 	 		__func__, PERFUSE_NODE_DATA(src)->pnd_ino,
2039 			pcn_src->pcn_name, pcn_targ->pcn_name,
2040 			PERFUSE_NODE_DATA(opc)->pnd_ino,
2041 			PERFUSE_NODE_DATA(targ_dir)->pnd_ino,
2042 			perfuse_node_path(targ_dir));
2043 #endif
2044 
2045 out:
2046 	if (pm != NULL)
2047 		ps->ps_destroy_msg(pm);
2048 
2049 	return error;
2050 }
2051 
2052 int
2053 perfuse_node_mkdir(pu, opc, pni, pcn, vap)
2054 	struct puffs_usermount *pu;
2055 	puffs_cookie_t opc;
2056 	struct puffs_newinfo *pni;
2057 	const struct puffs_cn *pcn;
2058 	const struct vattr *vap;
2059 {
2060 	struct perfuse_state *ps;
2061 	perfuse_msg_t *pm;
2062 	struct fuse_mkdir_in *fmi;
2063 	const char *path;
2064 	size_t len;
2065 
2066 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2067 		return ENOENT;
2068 
2069 	ps = puffs_getspecific(pu);
2070 	path = pcn->pcn_name;
2071 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
2072 
2073 	pm = ps->ps_new_msg(pu, opc, FUSE_MKDIR, len, NULL);
2074 	fmi = GET_INPAYLOAD(ps, pm, fuse_mkdir_in);
2075 	fmi->mode = vap->va_mode;
2076 	fmi->umask = 0; 	/* Seems unused by libfuse? */
2077 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
2078 
2079 	return node_mk_common(pu, opc, pni, pcn, pm);
2080 }
2081 
2082 
2083 int
2084 perfuse_node_rmdir(pu, opc, targ, pcn)
2085 	struct puffs_usermount *pu;
2086 	puffs_cookie_t opc;
2087 	puffs_cookie_t targ;
2088 	const struct puffs_cn *pcn;
2089 {
2090 	struct perfuse_state *ps;
2091 	struct perfuse_node_data *pnd;
2092 	perfuse_msg_t *pm;
2093 	char *path;
2094 	const char *name;
2095 	size_t len;
2096 	int error;
2097 
2098 	pnd = PERFUSE_NODE_DATA(opc);
2099 	if (pnd->pnd_flags & PND_REMOVED)
2100 		return ENOENT;
2101 
2102 	/*
2103 	 * Await for all operations on the deleted node to drain,
2104 	 * as the filesystem may be confused to have it deleted
2105 	 * during a getattr
2106 	 */
2107 	while (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_INXCHG)
2108 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2109 
2110 	ps = puffs_getspecific(pu);
2111 	name = pcn->pcn_name;
2112 	len = pcn->pcn_namelen + 1;
2113 
2114 	pm = ps->ps_new_msg(pu, opc, FUSE_RMDIR, len, NULL);
2115 	path = _GET_INPAYLOAD(ps, pm, char *);
2116 	(void)strlcpy(path, name, len);
2117 
2118 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2119 		goto out;
2120 
2121 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2122 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2123 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2124 
2125 	/*
2126 	 * The parent directory needs a sync
2127 	 */
2128 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2129 
2130 #ifdef PERFUSE_DEBUG
2131 	if (perfuse_diagflags & PDF_FILENAME)
2132 		DPRINTF("%s: remove nodeid = %"PRId64" file = \"%s\"\n",
2133 			__func__, PERFUSE_NODE_DATA(targ)->pnd_ino,
2134 			perfuse_node_path(targ));
2135 #endif
2136 out:
2137 	ps->ps_destroy_msg(pm);
2138 
2139 	return error;
2140 }
2141 
2142 /* vap is unused */
2143 /* ARGSUSED4 */
2144 int
2145 perfuse_node_symlink(pu, opc, pni, pcn_src, vap, link_target)
2146 	struct puffs_usermount *pu;
2147 	puffs_cookie_t opc;
2148 	struct puffs_newinfo *pni;
2149 	const struct puffs_cn *pcn_src;
2150 	const struct vattr *vap;
2151 	const char *link_target;
2152 {
2153 	struct perfuse_state *ps;
2154 	perfuse_msg_t *pm;
2155 	char *np;
2156 	const char *path;
2157 	size_t path_len;
2158 	size_t linkname_len;
2159 	size_t len;
2160 
2161 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2162 		return ENOENT;
2163 
2164 	ps = puffs_getspecific(pu);
2165 	path = pcn_src->pcn_name;
2166 	path_len = pcn_src->pcn_namelen + 1;
2167 	linkname_len = strlen(link_target) + 1;
2168 	len = path_len + linkname_len;
2169 
2170 	pm = ps->ps_new_msg(pu, opc, FUSE_SYMLINK, len, NULL);
2171 	np = _GET_INPAYLOAD(ps, pm, char *);
2172 	(void)strlcpy(np, path, path_len);
2173 	np += path_len;
2174 	(void)strlcpy(np, link_target, linkname_len);
2175 
2176 	return node_mk_common(pu, opc, pni, pcn_src, pm);
2177 }
2178 
2179 int
2180 perfuse_node_readdir(pu, opc, dent, readoff,
2181 		     reslen, pcr, eofflag, cookies, ncookies)
2182 	struct puffs_usermount *pu;
2183 	puffs_cookie_t opc;
2184 	struct dirent *dent;
2185 	off_t *readoff;
2186 	size_t *reslen;
2187 	const struct puffs_cred *pcr;
2188 	int *eofflag;
2189 	off_t *cookies;
2190 	size_t *ncookies;
2191 {
2192 	perfuse_msg_t *pm;
2193 	uint64_t fh;
2194 	struct perfuse_state *ps;
2195 	struct perfuse_node_data *pnd;
2196 	struct fuse_read_in *fri;
2197 	struct fuse_out_header *foh;
2198 	struct fuse_dirent *fd;
2199 	size_t foh_len;
2200 	int error;
2201 	uint64_t fd_offset;
2202 	size_t fd_maxlen;
2203 
2204 	pm = NULL;
2205 	error = 0;
2206 	ps = puffs_getspecific(pu);
2207 
2208 	/*
2209 	 * readdir state is kept at node level, and several readdir
2210 	 * requests can be issued at the same time on the same node.
2211 	 * We need to queue requests so that only one is in readdir
2212 	 * code at the same time.
2213 	 */
2214 	pnd = PERFUSE_NODE_DATA(opc);
2215 	while (pnd->pnd_flags & PND_INREADDIR)
2216 		requeue_request(pu, opc, PCQ_READDIR);
2217 	pnd->pnd_flags |= PND_INREADDIR;
2218 
2219 #ifdef PERFUSE_DEBUG
2220 	if (perfuse_diagflags & PDF_READDIR)
2221 		DPRINTF("%s: READDIR opc = %p enter critical section\n",
2222 			__func__, (void *)opc);
2223 #endif
2224 	/*
2225 	 * Do we already have the data bufered?
2226 	 */
2227 	if (pnd->pnd_dirent != NULL)
2228 		goto out;
2229 	pnd->pnd_dirent_len = 0;
2230 
2231 	/*
2232 	 * It seems NetBSD can call readdir without open first
2233 	 * libfuse will crash if it is done that way, hence open first.
2234 	 */
2235 	if (!(pnd->pnd_flags & PND_OPEN)) {
2236 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2237 			goto out;
2238 	}
2239 
2240 	fh = perfuse_get_fh(opc, FREAD);
2241 
2242 #ifdef PERFUSE_DEBUG
2243 	if (perfuse_diagflags & PDF_FH)
2244 		DPRINTF("%s: opc = %p, ino = %"PRId64", rfh = 0x%"PRIx64"\n",
2245 			__func__, (void *)opc,
2246 			PERFUSE_NODE_DATA(opc)->pnd_ino, fh);
2247 #endif
2248 
2249 	pnd->pnd_all_fd = NULL;
2250 	pnd->pnd_all_fd_len = 0;
2251 	fd_offset = 0;
2252 	fd_maxlen = ps->ps_max_readahead - sizeof(*foh);
2253 
2254 	do {
2255 		size_t fd_len;
2256 		char *afdp;
2257 
2258 		pm = ps->ps_new_msg(pu, opc, FUSE_READDIR, sizeof(*fri), pcr);
2259 
2260 		/*
2261 		 * read_flags, lock_owner and flags are unused in libfuse
2262 		 */
2263 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
2264 		fri->fh = fh;
2265 		fri->offset = fd_offset;
2266 		fri->size = (uint32_t)fd_maxlen;
2267 		fri->read_flags = 0;
2268 		fri->lock_owner = 0;
2269 		fri->flags = 0;
2270 
2271 		if ((error = xchg_msg(pu, opc, pm,
2272 				      UNSPEC_REPLY_LEN, wait_reply)) != 0)
2273 			goto out;
2274 
2275 		/*
2276 		 * There are many puffs_framebufs calls later,
2277 		 * therefore foh will not be valid for a long time.
2278 		 * Just get the length and forget it.
2279 		 */
2280 		foh = GET_OUTHDR(ps, pm);
2281 		foh_len = foh->len;
2282 
2283 		/*
2284 		 * Empty read: we reached the end of the buffer.
2285 		 */
2286 		if (foh_len == sizeof(*foh))
2287 			break;
2288 
2289 		/*
2290 		 * Corrupted message.
2291 		 */
2292 		if (foh_len < sizeof(*foh) + sizeof(*fd)) {
2293 			DWARNX("readdir reply too short");
2294 			error = EIO;
2295 			goto out;
2296 		}
2297 
2298 
2299 		fd = GET_OUTPAYLOAD(ps, pm, fuse_dirent);
2300 		fd_len = foh_len - sizeof(*foh);
2301 
2302 		pnd->pnd_all_fd = realloc(pnd->pnd_all_fd,
2303 					  pnd->pnd_all_fd_len + fd_len);
2304 		if (pnd->pnd_all_fd  == NULL)
2305 			DERR(EX_OSERR, "malloc failed");
2306 
2307 		afdp = (char *)(void *)pnd->pnd_all_fd + pnd->pnd_all_fd_len;
2308 		(void)memcpy(afdp, fd, fd_len);
2309 
2310 		pnd->pnd_all_fd_len += fd_len;
2311 		fd_offset += fd_len;
2312 
2313 		ps->ps_destroy_msg(pm);
2314 		pm = NULL;
2315 
2316 		/*
2317 		 * If the buffer was not completely filled,
2318 		 * that is, if there is room for the biggest
2319 		 * struct dirent possible, then we are done:
2320 		 * no need to issue another READDIR to see
2321 		 * an empty reply.
2322 		 */
2323 	} while (foh_len >= fd_maxlen - (sizeof(*fd) + MAXPATHLEN));
2324 
2325 	if (fuse_to_dirent(pu, opc, pnd->pnd_all_fd, pnd->pnd_all_fd_len) == -1)
2326 		error = EIO;
2327 
2328 out:
2329 	if (pnd->pnd_all_fd != NULL) {
2330 		free(pnd->pnd_all_fd);
2331 		pnd->pnd_all_fd = NULL;
2332 		pnd->pnd_all_fd_len = 0;
2333 	}
2334 
2335 	if (pm != NULL)
2336 		ps->ps_destroy_msg(pm);
2337 
2338 	if (error == 0)
2339 		error = readdir_buffered(opc, dent, readoff,
2340 			reslen, pcr, eofflag, cookies, ncookies);
2341 
2342 	/*
2343 	 * Schedule queued readdir requests
2344 	 */
2345 	pnd->pnd_flags &= ~PND_INREADDIR;
2346 	(void)dequeue_requests(ps, opc, PCQ_READDIR, DEQUEUE_ALL);
2347 
2348 #ifdef PERFUSE_DEBUG
2349 	if (perfuse_diagflags & PDF_READDIR)
2350 		DPRINTF("%s: READDIR opc = %p exit critical section\n",
2351 			__func__, (void *)opc);
2352 #endif
2353 
2354 	return error;
2355 }
2356 
2357 int
2358 perfuse_node_readlink(pu, opc, pcr, linkname, linklen)
2359 	struct puffs_usermount *pu;
2360 	puffs_cookie_t opc;
2361 	const struct puffs_cred *pcr;
2362 	char *linkname;
2363 	size_t *linklen;
2364 {
2365 	struct perfuse_state *ps;
2366 	perfuse_msg_t *pm;
2367 	int error;
2368 	size_t len;
2369 	struct fuse_out_header *foh;
2370 
2371 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2372 		return ENOENT;
2373 
2374 	ps = puffs_getspecific(pu);
2375 
2376 	pm = ps->ps_new_msg(pu, opc, FUSE_READLINK, 0, pcr);
2377 
2378 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2379 		goto out;
2380 
2381 	foh = GET_OUTHDR(ps, pm);
2382 	len = foh->len - sizeof(*foh);
2383 	if (len > *linklen)
2384 		DERRX(EX_PROTOCOL, "path len = %zd too long", len);
2385 	if (len == 0)
2386 		DERRX(EX_PROTOCOL, "path len = %zd too short", len);
2387 
2388 	/*
2389 	 * FUSE filesystems return a NUL terminated string, we
2390 	 * do not want to trailing \0
2391 	 */
2392 	*linklen = len - 1;
2393 	(void)memcpy(linkname, _GET_OUTPAYLOAD(ps, pm, char *), len);
2394 out:
2395 	ps->ps_destroy_msg(pm);
2396 
2397 	return error;
2398 }
2399 
2400 int
2401 perfuse_node_reclaim(pu, opc)
2402 	struct puffs_usermount *pu;
2403 	puffs_cookie_t opc;
2404 {
2405 	struct perfuse_state *ps;
2406 	perfuse_msg_t *pm;
2407 	struct perfuse_node_data *pnd;
2408 	struct fuse_forget_in *ffi;
2409 	struct puffs_node *pn;
2410 	struct puffs_node *pn_root;
2411 
2412 	ps = puffs_getspecific(pu);
2413 	pnd = PERFUSE_NODE_DATA(opc);
2414 
2415 	/*
2416 	 * Never forget the root.
2417 	 */
2418 	if (pnd->pnd_ino == FUSE_ROOT_ID)
2419 		return 0;
2420 
2421 	pnd->pnd_flags |= PND_RECLAIMED;
2422 
2423 #ifdef PERFUSE_DEBUG
2424 	if (perfuse_diagflags & PDF_RECLAIM)
2425 		DPRINTF("%s (nodeid %"PRId64") reclaimed\n",
2426 			perfuse_node_path(opc), pnd->pnd_ino);
2427 #endif
2428 
2429 	pn_root = puffs_getroot(pu);
2430 	pn = (struct puffs_node *)opc;
2431 	while (pn != pn_root) {
2432 		struct puffs_node *parent_pn;
2433 
2434 		pnd = PERFUSE_NODE_DATA(pn);
2435 
2436 #ifdef PERFUSE_DEBUG
2437 	if (perfuse_diagflags & PDF_RECLAIM)
2438 		DPRINTF("%s (nodeid %"PRId64") is %sreclaimed, "
2439 			"has childcount %d %s%s%s%s, pending ops:%s%s%s\n",
2440 		        perfuse_node_path((puffs_cookie_t)pn), pnd->pnd_ino,
2441 		        pnd->pnd_flags & PND_RECLAIMED ? "" : "not ",
2442 		        pnd->pnd_childcount,
2443 			pnd->pnd_flags & PND_OPEN ? "open " : "not open",
2444 			pnd->pnd_flags & PND_RFH ? "r" : "",
2445 			pnd->pnd_flags & PND_WFH ? "w" : "",
2446 			pnd->pnd_flags & PND_BUSY ? "" : " none",
2447 			pnd->pnd_flags & PND_INREADDIR ? " readdir" : "",
2448 			pnd->pnd_flags & PND_INWRITE ? " write" : "",
2449 			pnd->pnd_flags & PND_INOPEN ? " open" : "");
2450 #endif
2451 
2452 		if (!(pnd->pnd_flags & PND_RECLAIMED) ||
2453 		    (pnd->pnd_childcount != 0))
2454 			return 0;
2455 
2456 #ifdef PERFUSE_DEBUG
2457 		if ((pnd->pnd_flags & PND_OPEN) ||
2458 		       !TAILQ_EMPTY(&pnd->pnd_pcq))
2459 			DERRX(EX_SOFTWARE, "%s: opc = %p: still open",
2460 			      __func__, (void *)opc);
2461 
2462 		if ((pnd->pnd_flags & PND_BUSY) ||
2463 		       !TAILQ_EMPTY(&pnd->pnd_pcq))
2464 			DERRX(EX_SOFTWARE, "%s: opc = %p: ongoing operations",
2465 			      __func__, (void *)opc);
2466 #endif
2467 
2468 		/*
2469 		 * Send the FORGET message
2470 		 */
2471 		pm = ps->ps_new_msg(pu, (puffs_cookie_t)pn, FUSE_FORGET,
2472 			      sizeof(*ffi), NULL);
2473 		ffi = GET_INPAYLOAD(ps, pm, fuse_forget_in);
2474 		ffi->nlookup = pnd->pnd_nlookup;
2475 
2476 		/*
2477 		 * No reply is expected, pm is freed in xchg_msg
2478 		 */
2479 		(void)xchg_msg(pu, (puffs_cookie_t)pn,
2480 			       pm, UNSPEC_REPLY_LEN, no_reply);
2481 
2482 		parent_pn = pnd->pnd_parent;
2483 
2484 		perfuse_destroy_pn(pn);
2485 
2486 		pn = parent_pn;
2487 	}
2488 
2489 	return 0;
2490 }
2491 
2492 int
2493 perfuse_node_inactive(pu, opc)
2494 	struct puffs_usermount *pu;
2495 	puffs_cookie_t opc;
2496 {
2497 	struct perfuse_state *ps;
2498 	struct perfuse_node_data *pnd;
2499 
2500 	ps = puffs_getspecific(pu);
2501 	pnd = PERFUSE_NODE_DATA(opc);
2502 
2503 	if (!(pnd->pnd_flags & (PND_OPEN|PND_REMOVED)))
2504 		return 0;
2505 
2506 	/*
2507 	 * Make sure all operation are finished
2508 	 * There can be an ongoing write. Other
2509 	 * operation wait for all data before
2510 	 * the close/inactive.
2511 	 */
2512 	while (pnd->pnd_flags & PND_INWRITE)
2513 		requeue_request(pu, opc, PCQ_AFTERWRITE);
2514 
2515 	/*
2516 	 * The inactive operation may be cancelled.
2517 	 * If no open is in progress, set PND_INOPEN
2518 	 * so that a new open will be queued.
2519 	 */
2520 	if (pnd->pnd_flags & PND_INOPEN)
2521 		return 0;
2522 
2523 	pnd->pnd_flags |= PND_INOPEN;
2524 
2525 	/*
2526 	 * Sync data
2527 	 */
2528 	if (pnd->pnd_flags & PND_DIRTY)
2529 		(void)perfuse_node_fsync(pu, opc, NULL, 0, 0, 0);
2530 
2531 	/*
2532 	 * Close handles
2533 	 */
2534 	if (pnd->pnd_flags & PND_WFH)
2535 		(void)perfuse_node_close_common(pu, opc, FWRITE);
2536 
2537 	if (pnd->pnd_flags & PND_RFH)
2538 		(void)perfuse_node_close_common(pu, opc, FREAD);
2539 
2540 	/*
2541 	 * This will cause a reclaim to be sent
2542 	 */
2543 	if (pnd->pnd_flags & PND_REMOVED)
2544 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N1);
2545 
2546 	/*
2547 	 * Schedule awaiting operations
2548 	 */
2549 	pnd->pnd_flags &= ~PND_INOPEN;
2550 	(void)dequeue_requests(ps, opc, PCQ_OPEN, DEQUEUE_ALL);
2551 
2552 	return 0;
2553 }
2554 
2555 
2556 /* ARGSUSED0 */
2557 int
2558 perfuse_node_print(pu, opc)
2559 	struct puffs_usermount *pu;
2560 	puffs_cookie_t opc;
2561 {
2562 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2563 	return 0;
2564 }
2565 
2566 /* ARGSUSED0 */
2567 int
2568 perfuse_node_pathconf(pu, opc, name, retval)
2569 	struct puffs_usermount *pu;
2570 	puffs_cookie_t opc;
2571 	int name;
2572 	int *retval;
2573 {
2574 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2575 	return 0;
2576 }
2577 
2578 /* id is unused */
2579 /* ARGSUSED2 */
2580 int
2581 perfuse_node_advlock(pu, opc, id, op, fl, flags)
2582 	struct puffs_usermount *pu;
2583 	puffs_cookie_t opc;
2584 	void *id;
2585 	int op;
2586 	struct flock *fl;
2587 	int flags;
2588 {
2589 	struct perfuse_state *ps;
2590 	int fop;
2591 	perfuse_msg_t *pm;
2592 	struct fuse_lk_in *fli;
2593 	struct fuse_out_header *foh;
2594 	struct fuse_lk_out *flo;
2595 	uint32_t owner;
2596 	size_t len;
2597 	int error;
2598 
2599 	ps = puffs_getspecific(pu);
2600 
2601 	if (op == F_GETLK)
2602 		fop = FUSE_GETLK;
2603 	else
2604 		fop = (flags & F_WAIT) ? FUSE_SETLKW : FUSE_SETLK;
2605 
2606 	pm = ps->ps_new_msg(pu, opc, fop, sizeof(*fli), NULL);
2607 	fli = GET_INPAYLOAD(ps, pm, fuse_lk_in);
2608 	fli->fh = perfuse_get_fh(opc, FWRITE);
2609 	fli->owner = fl->l_pid;
2610 	fli->lk.start = fl->l_start;
2611 	fli->lk.end = fl->l_start + fl->l_len;
2612 	fli->lk.type = fl->l_type;
2613 	fli->lk.pid = fl->l_pid;
2614 	fli->lk_flags = (flags & F_FLOCK) ? FUSE_LK_FLOCK : 0;
2615 
2616 	owner = fl->l_pid;
2617 
2618 #ifdef PERFUSE_DEBUG
2619 	if (perfuse_diagflags & PDF_FH)
2620 		DPRINTF("%s: opc = %p, ino = %"PRId64", fh = 0x%"PRIx64"\n",
2621 			__func__, (void *)opc,
2622 			PERFUSE_NODE_DATA(opc)->pnd_ino, fli->fh);
2623 #endif
2624 
2625 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2626 		goto out;
2627 
2628 	foh = GET_OUTHDR(ps, pm);
2629 	len = foh->len - sizeof(*foh);
2630 
2631 	/*
2632 	 * Save or clear the lock
2633 	 */
2634 	switch (op) {
2635 	case F_GETLK:
2636 		if (len != sizeof(*flo))
2637 			DERRX(EX_SOFTWARE,
2638 			      "%s: Unexpected lock reply len %zd",
2639 			      __func__, len);
2640 
2641 		flo = GET_OUTPAYLOAD(ps, pm, fuse_lk_out);
2642 		fl->l_start = flo->lk.start;
2643 		fl->l_len = flo->lk.end - flo->lk.start;
2644 		fl->l_pid = flo->lk.pid;
2645 		fl->l_type = flo->lk.type;
2646 		fl->l_whence = SEEK_SET;	/* libfuse hardcodes it */
2647 
2648 		PERFUSE_NODE_DATA(opc)->pnd_lock_owner = flo->lk.pid;
2649 		break;
2650 	case F_UNLCK:
2651 		owner = 0;
2652 		/* FALLTHROUGH */
2653 	case F_SETLK:
2654 		/* FALLTHROUGH */
2655 	case F_SETLKW:
2656 		if (error != 0)
2657 			PERFUSE_NODE_DATA(opc)->pnd_lock_owner = owner;
2658 
2659 		if (len != 0)
2660 			DERRX(EX_SOFTWARE,
2661 			      "%s: Unexpected unlock reply len %zd",
2662 			      __func__, len);
2663 
2664 		break;
2665 	default:
2666 		DERRX(EX_SOFTWARE, "%s: Unexpected op %d", __func__, op);
2667 		break;
2668 	}
2669 
2670 out:
2671 	ps->ps_destroy_msg(pm);
2672 
2673 	return error;
2674 }
2675 
2676 int
2677 perfuse_node_read(pu, opc, buf, offset, resid, pcr, ioflag)
2678 	struct puffs_usermount *pu;
2679 	puffs_cookie_t opc;
2680 	uint8_t *buf;
2681 	off_t offset;
2682 	size_t *resid;
2683 	const struct puffs_cred *pcr;
2684 	int ioflag;
2685 {
2686 	struct perfuse_state *ps;
2687 	struct perfuse_node_data *pnd;
2688 	perfuse_msg_t *pm;
2689 	struct fuse_read_in *fri;
2690 	struct fuse_out_header *foh;
2691 	size_t readen;
2692 	int error;
2693 
2694 	ps = puffs_getspecific(pu);
2695 	pnd = PERFUSE_NODE_DATA(opc);
2696 	pm = NULL;
2697 
2698 	if (puffs_pn_getvap((struct puffs_node *)opc)->va_type == VDIR)
2699 		return EBADF;
2700 
2701 	do {
2702 		size_t max_read;
2703 
2704 		max_read = ps->ps_max_readahead - sizeof(*foh);
2705 		/*
2706 		 * flags may be set to FUSE_READ_LOCKOWNER
2707 		 * if lock_owner is provided.
2708 		 */
2709 		pm = ps->ps_new_msg(pu, opc, FUSE_READ, sizeof(*fri), pcr);
2710 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
2711 		fri->fh = perfuse_get_fh(opc, FREAD);
2712 		fri->offset = offset;
2713 		fri->size = (uint32_t)MIN(*resid, max_read);
2714 		fri->read_flags = 0; /* XXX Unused by libfuse? */
2715 		fri->lock_owner = pnd->pnd_lock_owner;
2716 		fri->flags = 0;
2717 		fri->flags |= (fri->lock_owner != 0) ? FUSE_READ_LOCKOWNER : 0;
2718 
2719 #ifdef PERFUSE_DEBUG
2720 	if (perfuse_diagflags & PDF_FH)
2721 		DPRINTF("%s: opc = %p, ino = %"PRId64", fh = 0x%"PRIx64"\n",
2722 			__func__, (void *)opc, pnd->pnd_ino, fri->fh);
2723 #endif
2724 		error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
2725 
2726 		if (error  != 0)
2727 			goto out;
2728 
2729 		foh = GET_OUTHDR(ps, pm);
2730 		readen = foh->len - sizeof(*foh);
2731 
2732 #ifdef PERFUSE_DEBUG
2733 		if (readen > *resid)
2734 			DERRX(EX_SOFTWARE, "%s: Unexpected big read %zd",
2735 			      __func__, readen);
2736 #endif
2737 
2738 		(void)memcpy(buf,  _GET_OUTPAYLOAD(ps, pm, char *), readen);
2739 
2740 		buf += readen;
2741 		offset += readen;
2742 		*resid -= readen;
2743 
2744 		ps->ps_destroy_msg(pm);
2745 		pm = NULL;
2746 	} while ((*resid != 0) && (readen != 0));
2747 
2748 	if (ioflag & (IO_SYNC|IO_DSYNC))
2749 		ps->ps_syncreads++;
2750 	else
2751 		ps->ps_asyncreads++;
2752 
2753 out:
2754 	if (pm != NULL)
2755 		ps->ps_destroy_msg(pm);
2756 
2757 	return error;
2758 }
2759 
2760 int
2761 perfuse_node_write(pu, opc, buf, offset, resid, pcr, ioflag)
2762 	struct puffs_usermount *pu;
2763 	puffs_cookie_t opc;
2764 	uint8_t *buf;
2765 	off_t offset;
2766 	size_t *resid;
2767 	const struct puffs_cred *pcr;
2768 	int ioflag;
2769 {
2770 	struct perfuse_state *ps;
2771 	struct perfuse_node_data *pnd;
2772 	struct vattr *vap;
2773 	perfuse_msg_t *pm;
2774 	struct fuse_write_in *fwi;
2775 	struct fuse_write_out *fwo;
2776 	size_t data_len;
2777 	size_t payload_len;
2778 	size_t written;
2779 	int error;
2780 
2781 	ps = puffs_getspecific(pu);
2782 	pnd = PERFUSE_NODE_DATA(opc);
2783 	vap = puffs_pn_getvap((struct puffs_node *)opc);
2784 	written = 0;
2785 	pm = NULL;
2786 
2787 	if (vap->va_type == VDIR)
2788 		return EBADF;
2789 
2790 	/*
2791 	 * We need to queue write requests in order to avoid
2792 	 * dequeueing PCQ_AFTERWRITE when there are pending writes.
2793 	 */
2794 	while (pnd->pnd_flags & PND_INWRITE)
2795 		requeue_request(pu, opc, PCQ_WRITE);
2796 	pnd->pnd_flags |= PND_INWRITE;
2797 
2798 	/*
2799 	 * append flag: re-read the file size so that
2800 	 * we get the latest value.
2801 	 */
2802 	if (ioflag & PUFFS_IO_APPEND) {
2803 		if ((error = perfuse_node_getattr(pu, opc, vap, pcr)) != 0)
2804 			goto out;
2805 
2806 		offset = vap->va_size;
2807 	}
2808 
2809 	pm = NULL;
2810 
2811 	do {
2812 		size_t max_write;
2813 		/*
2814 		 * There is a writepage flag when data
2815 		 * is aligned to page size. Use it for
2816 		 * everything but the data after the last
2817 		 * page boundary.
2818 		 */
2819 		max_write = ps->ps_max_write - sizeof(*fwi);
2820 
2821 		data_len = MIN(*resid, max_write);
2822 		if (data_len > (size_t)sysconf(_SC_PAGESIZE))
2823 			data_len = data_len & ~(sysconf(_SC_PAGESIZE) - 1);
2824 
2825 		payload_len = data_len + sizeof(*fwi);
2826 
2827 		/*
2828 		 * flags may be set to FUSE_WRITE_CACHE (XXX usage?)
2829 		 * or FUSE_WRITE_LOCKOWNER, if lock_owner is provided.
2830 		 * write_flags is set to 1 for writepage.
2831 		 */
2832 		pm = ps->ps_new_msg(pu, opc, FUSE_WRITE, payload_len, pcr);
2833 		fwi = GET_INPAYLOAD(ps, pm, fuse_write_in);
2834 		fwi->fh = perfuse_get_fh(opc, FWRITE);
2835 		fwi->offset = offset;
2836 		fwi->size = (uint32_t)data_len;
2837 		fwi->write_flags = (fwi->size % sysconf(_SC_PAGESIZE)) ? 0 : 1;
2838 		fwi->lock_owner = pnd->pnd_lock_owner;
2839 		fwi->flags = 0;
2840 		fwi->flags |= (fwi->lock_owner != 0) ? FUSE_WRITE_LOCKOWNER : 0;
2841 		fwi->flags |= (ioflag & IO_DIRECT) ? 0 : FUSE_WRITE_CACHE;
2842 		(void)memcpy((fwi + 1), buf, data_len);
2843 
2844 
2845 #ifdef PERFUSE_DEBUG
2846 		if (perfuse_diagflags & PDF_FH)
2847 			DPRINTF("%s: opc = %p, ino = %"PRId64", "
2848 				"fh = 0x%"PRIx64"\n", __func__,
2849 				(void *)opc, pnd->pnd_ino, fwi->fh);
2850 #endif
2851 		if ((error = xchg_msg(pu, opc, pm,
2852 				      sizeof(*fwo), wait_reply)) != 0)
2853 			goto out;
2854 
2855 		fwo = GET_OUTPAYLOAD(ps, pm, fuse_write_out);
2856 		written = fwo->size;
2857 #ifdef PERFUSE_DEBUG
2858 		if (written > *resid)
2859 			DERRX(EX_SOFTWARE, "%s: Unexpected big write %zd",
2860 			      __func__, written);
2861 #endif
2862 		*resid -= written;
2863 		offset += written;
2864 		buf += written;
2865 
2866 		ps->ps_destroy_msg(pm);
2867 		pm = NULL;
2868 	} while (*resid != 0);
2869 
2870 	/*
2871 	 * puffs_ops(3) says
2872 	 *  "everything must be written or an error will be generated"
2873 	 */
2874 	if (*resid != 0)
2875 		error = EFBIG;
2876 
2877 	/*
2878 	 * Update file size if we wrote beyond the end
2879 	 */
2880 	if (offset > (off_t)vap->va_size)
2881 		vap->va_size = offset;
2882 
2883 	/*
2884 	 * Statistics
2885 	 */
2886 	if (ioflag & (IO_SYNC|IO_DSYNC))
2887 		ps->ps_syncwrites++;
2888 	else
2889 		ps->ps_asyncwrites++;
2890 
2891 	/*
2892 	 * Remember to sync the file
2893 	 */
2894 	pnd->pnd_flags |= PND_DIRTY;
2895 
2896 #ifdef PERFUSE_DEBUG
2897 	if (perfuse_diagflags & PDF_SYNC)
2898 		DPRINTF("%s: DIRTY opc = %p, file = \"%s\"\n",
2899 			__func__, (void*)opc, perfuse_node_path(opc));
2900 #endif
2901 out:
2902 	if (pm != NULL)
2903 		ps->ps_destroy_msg(pm);
2904 
2905 	/*
2906 	 * If there are no more queued write, we can resume
2907 	 * an operation awaiting write completion.
2908 	 */
2909 	pnd->pnd_flags &= ~PND_INWRITE;
2910 	if (dequeue_requests(ps, opc, PCQ_WRITE, 1) == 0)
2911 		(void)dequeue_requests(ps, opc, PCQ_AFTERWRITE, DEQUEUE_ALL);
2912 
2913 	return error;
2914 }
2915 
2916 /* ARGSUSED0 */
2917 void
2918 perfuse_cache_write(pu, opc, size, runs)
2919 	struct puffs_usermount *pu;
2920 	puffs_cookie_t opc;
2921 	size_t size;
2922 	struct puffs_cacherun *runs;
2923 {
2924 	return;
2925 }
2926 
2927