xref: /netbsd-src/lib/libperfuse/ops.c (revision e39ef1d61eee3ccba837ee281f1e098c864487aa)
1 /*  $NetBSD: ops.c,v 1.49 2011/12/28 17:33:53 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 <sys/extattr.h>
40 #include <sys/time.h>
41 #include <machine/vmparam.h>
42 
43 #include "perfuse_priv.h"
44 #include "fuse.h"
45 
46 extern int perfuse_diagflags;
47 
48 #if 0
49 static void print_node(const char *, puffs_cookie_t);
50 #endif
51 static void set_expire(puffs_cookie_t, struct fuse_entry_out *,
52    struct fuse_attr_out *);
53 static int attr_expired(puffs_cookie_t);
54 static int entry_expired(puffs_cookie_t);
55 static int xchg_msg(struct puffs_usermount *, puffs_cookie_t,
56     perfuse_msg_t *, size_t, enum perfuse_xchg_pb_reply);
57 static int mode_access(puffs_cookie_t, const struct puffs_cred *, mode_t);
58 static int sticky_access(struct puffs_node *, const struct puffs_cred *);
59 static void fuse_attr_to_vap(struct perfuse_state *,
60     struct vattr *, struct fuse_attr *);
61 static int node_lookup_dir_nodot(struct puffs_usermount *,
62     puffs_cookie_t, char *, size_t, struct puffs_node **);
63 static int node_lookup_common(struct puffs_usermount *, puffs_cookie_t,
64     const char *, const struct puffs_cred *, struct puffs_node **);
65 static int node_mk_common(struct puffs_usermount *, puffs_cookie_t,
66     struct puffs_newinfo *, const struct puffs_cn *pcn, perfuse_msg_t *);
67 static int node_mk_common_final(struct puffs_usermount *, puffs_cookie_t,
68     struct puffs_node *, const struct puffs_cn *pcn);
69 static uint64_t readdir_last_cookie(struct fuse_dirent *, size_t);
70 static ssize_t fuse_to_dirent(struct puffs_usermount *, puffs_cookie_t,
71     struct fuse_dirent *, size_t);
72 static int readdir_buffered(puffs_cookie_t, struct dirent *, off_t *,
73     size_t *);
74 static void requeue_request(struct puffs_usermount *,
75     puffs_cookie_t opc, enum perfuse_qtype);
76 static int dequeue_requests(struct perfuse_state *,
77     puffs_cookie_t opc, enum perfuse_qtype, int);
78 #define DEQUEUE_ALL 0
79 
80 /*
81  *  From <sys/vnode>, inside #ifdef _KERNEL section
82  */
83 #define IO_SYNC		(0x40|IO_DSYNC)
84 #define IO_DSYNC	0x00200
85 #define IO_DIRECT	0x02000
86 
87 /*
88  *  From <fcntl>, inside #ifdef _KERNEL section
89  */
90 #define F_WAIT		0x010
91 #define F_FLOCK		0x020
92 #define OFLAGS(fflags)  ((fflags) - 1)
93 
94 /*
95  * Borrowed from src/sys/kern/vfs_subr.c and src/sys/sys/vnode.h
96  */
97 const enum vtype iftovt_tab[16] = {
98 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
99         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
100 };
101 const int vttoif_tab[9] = {
102 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
103         S_IFSOCK, S_IFIFO, S_IFMT,
104 };
105 
106 #define IFTOVT(mode) (iftovt_tab[((mode) & S_IFMT) >> 12])
107 #define VTTOIF(indx) (vttoif_tab[(int)(indx)])
108 
109 #if 0
110 static void
111 print_node(func, opc)
112 	const char *func;
113 	puffs_cookie_t opc;
114 {
115 	struct puffs_node *pn;
116 	struct perfuse_node_data *pnd;
117 	struct vattr *vap;
118 
119 	pn = (struct puffs_node *)opc;
120 	pnd = PERFUSE_NODE_DATA(opc);
121 	vap = &pn->pn_va;
122 
123 	printf("%s: \"%s\", opc = %p, nodeid = 0x%"PRIx64" ino = %"PRIu64"\n",
124 	       func, pnd->pnd_name, opc, pnd->pnd_nodeid, vap->va_fileid);
125 
126 	return;
127 }
128 #endif /* PERFUSE_DEBUG */
129 
130 int
131 perfuse_node_close_common(pu, opc, mode)
132 	struct puffs_usermount *pu;
133 	puffs_cookie_t opc;
134 	int mode;
135 {
136 	struct perfuse_state *ps;
137 	perfuse_msg_t *pm;
138 	int op;
139 	uint64_t fh;
140 	struct fuse_release_in *fri;
141 	struct perfuse_node_data *pnd;
142 	struct puffs_node *pn;
143 	int error;
144 
145 	ps = puffs_getspecific(pu);
146 	pn = (struct puffs_node *)opc;
147 	pnd = PERFUSE_NODE_DATA(pn);
148 
149 	if (puffs_pn_getvap(pn)->va_type == VDIR) {
150 		op = FUSE_RELEASEDIR;
151 		mode = FREAD;
152 	} else {
153 		op = FUSE_RELEASE;
154 	}
155 
156 	/*
157 	 * Destroy the filehandle before sending the
158 	 * request to the FUSE filesystem, otherwise
159 	 * we may get a second close() while we wait
160 	 * for the reply, and we would end up closing
161 	 * the same fh twice instead of closng both.
162 	 */
163 	fh = perfuse_get_fh(opc, mode);
164 	perfuse_destroy_fh(pn, fh);
165 
166 	/*
167 	 * release_flags may be set to FUSE_RELEASE_FLUSH
168 	 * to flush locks. lock_owner must be set in that case
169 	 *
170 	 * ps_new_msg() is called with NULL creds, which will
171 	 * be interpreted as FUSE superuser. We come here from the
172 	 * inactive method, which provides no creds, but obviously
173 	 * runs with kernel privilege.
174 	 */
175 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*fri), NULL);
176 	fri = GET_INPAYLOAD(ps, pm, fuse_release_in);
177 	fri->fh = fh;
178 	fri->flags = 0;
179 	fri->release_flags = 0;
180 	fri->lock_owner = pnd->pnd_lock_owner;
181 	fri->flags = (fri->lock_owner != 0) ? FUSE_RELEASE_FLUSH : 0;
182 
183 #ifdef PERFUSE_DEBUG
184 	if (perfuse_diagflags & PDF_FH)
185 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
186 			 __func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
187 #endif
188 
189 	if ((error = xchg_msg(pu, opc, pm,
190 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
191 		DERRX(EX_SOFTWARE, "%s: freed fh = 0x%"PRIx64" but filesystem "
192 		      "returned error = %d", __func__, fh, error);
193 
194 	ps->ps_destroy_msg(pm);
195 
196 	return 0;
197 }
198 
199 static int
200 xchg_msg(pu, opc, pm, len, wait)
201 	struct puffs_usermount *pu;
202 	puffs_cookie_t opc;
203 	perfuse_msg_t *pm;
204 	size_t len;
205      	enum perfuse_xchg_pb_reply wait;
206 {
207 	struct perfuse_state *ps;
208 	struct perfuse_node_data *pnd;
209 	struct perfuse_trace *pt = NULL;
210 	int opcode;
211 	int error;
212 
213 	ps = puffs_getspecific(pu);
214 	pnd = NULL;
215 	if ((struct puffs_node *)opc != NULL)
216 		pnd = PERFUSE_NODE_DATA(opc);
217 
218 #ifdef PERFUSE_DEBUG
219 	if ((perfuse_diagflags & PDF_FILENAME) && (opc != 0))
220 		DPRINTF("file = \"%s\", ino = %"PRIu64" flags = 0x%x\n",
221 			perfuse_node_path(opc),
222 			((struct puffs_node *)opc)->pn_va.va_fileid,
223 			PERFUSE_NODE_DATA(opc)->pnd_flags);
224 #endif
225 	if (pnd)
226 		pnd->pnd_flags |= PND_INXCHG;
227 
228 	/*
229 	 * Record FUSE call start if requested
230 	 */
231 	opcode = ps->ps_get_inhdr(pm)->opcode;
232 	if (perfuse_diagflags & PDF_TRACE) {
233 		if ((pt = malloc(sizeof(*pt))) == NULL)
234 			DERR(EX_OSERR, "malloc failed");
235 
236 		pt->pt_opcode = opcode;
237 		pt->pt_status = inxchg;
238 
239 		if (opc == 0)
240 			(void)strcpy(pt->pt_path, "");
241 		else
242 			(void)strlcpy(pt->pt_path,
243 				      perfuse_node_path(opc),
244 				      sizeof(pt->pt_path));
245 
246 		(void)strlcpy(pt->pt_extra,
247 			      perfuse_opdump_in(ps, pm),
248 			      sizeof(pt->pt_extra));
249 
250 		if (clock_gettime(CLOCK_REALTIME, &pt->pt_start) != 0)
251 			DERR(EX_OSERR, "clock_gettime failed");
252 
253 		TAILQ_INSERT_TAIL(&ps->ps_trace, pt, pt_list);
254 		ps->ps_tracecount++;
255 	}
256 
257 	/*
258 	 * Do actual FUSE exchange
259 	 */
260 	if ((error = ps->ps_xchg_msg(pu, pm, len, wait)) != 0)
261 		ps->ps_destroy_msg(pm);
262 
263 	/*
264 	 * Record FUSE call end if requested
265 	 */
266 	if (perfuse_diagflags & PDF_TRACE) {
267 		if (clock_gettime(CLOCK_REALTIME, &pt->pt_end) != 0)
268 			DERR(EX_OSERR, "clock_gettime failed");
269 
270 		pt->pt_status = done;
271 		pt->pt_error = error;
272 		while (ps->ps_tracecount > PERFUSE_TRACECOUNT_MAX) {
273 			struct perfuse_trace *fpt = TAILQ_FIRST(&ps->ps_trace);
274 
275 			if (fpt->pt_status != done)
276 				break;
277 
278 			TAILQ_REMOVE(&ps->ps_trace, fpt, pt_list);
279 			ps->ps_tracecount--;
280 			free(fpt);
281 		}
282 	}
283 
284 	if (pnd) {
285 		pnd->pnd_flags &= ~PND_INXCHG;
286 		(void)dequeue_requests(ps, opc, PCQ_AFTERXCHG, DEQUEUE_ALL);
287 	}
288 
289 	return error;
290 }
291 
292 static int
293 mode_access(opc, pcr, mode)
294 	puffs_cookie_t opc;
295 	const struct puffs_cred *pcr;
296 	mode_t mode;
297 {
298 	struct puffs_node *pn;
299 	struct vattr *va;
300 
301 	/*
302 	 * pcr is NULL for self open through fsync or readdir.
303 	 * In both case, access control is useless, as it was
304 	 * done before, at open time.
305 	 */
306 	if (pcr == NULL)
307 		return 0;
308 
309 	pn = (struct puffs_node *)opc;
310 	va = puffs_pn_getvap(pn);
311 	return puffs_access(va->va_type, va->va_mode,
312 			    va->va_uid, va->va_gid,
313 			    mode, pcr);
314 }
315 
316 static int
317 sticky_access(targ, pcr)
318 	struct puffs_node *targ;
319 	const struct puffs_cred *pcr;
320 {
321 	uid_t uid;
322 	struct puffs_node *tdir;
323 	int sticky, owner;
324 
325 	tdir = PERFUSE_NODE_DATA(targ)->pnd_parent;
326 
327 	/*
328 	 * This covers the case where the kernel requests a DELETE
329 	 * or RENAME on its own, and where puffs_cred_getuid would
330 	 * return -1. While such a situation should not happen,
331 	 * we allow it here.
332 	 *
333 	 * This also allows root to tamper with other users' files
334 	 * that have the sticky bit.
335 	 */
336 	if (puffs_cred_isjuggernaut(pcr))
337 		return 0;
338 
339 	if (puffs_cred_getuid(pcr, &uid) != 0)
340 		DERRX(EX_SOFTWARE, "puffs_cred_getuid fails in %s", __func__);
341 
342 	sticky = puffs_pn_getvap(tdir)->va_mode & S_ISTXT;
343 	owner = puffs_pn_getvap(targ)->va_uid == uid;
344 
345 	if (sticky && !owner)
346 		return EACCES;
347 
348 	return 0;
349 }
350 
351 
352 static void
353 fuse_attr_to_vap(ps, vap, fa)
354 	struct perfuse_state *ps;
355 	struct vattr *vap;
356 	struct fuse_attr *fa;
357 {
358 	vap->va_type = IFTOVT(fa->mode);
359 	vap->va_mode = fa->mode & ALLPERMS;
360 	vap->va_nlink = fa->nlink;
361 	vap->va_uid = fa->uid;
362 	vap->va_gid = fa->gid;
363 	vap->va_fsid = (long)ps->ps_fsid;
364 	vap->va_fileid = fa->ino;
365 	vap->va_size = fa->size;
366 	vap->va_blocksize = fa->blksize;
367 	vap->va_atime.tv_sec = (time_t)fa->atime;
368 	vap->va_atime.tv_nsec = (long) fa->atimensec;
369 	vap->va_mtime.tv_sec = (time_t)fa->mtime;
370 	vap->va_mtime.tv_nsec = (long)fa->mtimensec;
371 	vap->va_ctime.tv_sec = (time_t)fa->ctime;
372 	vap->va_ctime.tv_nsec = (long)fa->ctimensec;
373 	vap->va_birthtime.tv_sec = 0;
374 	vap->va_birthtime.tv_nsec = 0;
375 	vap->va_gen = 0;
376 	vap->va_flags = 0;
377 	vap->va_rdev = fa->rdev;
378 	vap->va_bytes = fa->size;
379 	vap->va_filerev = (u_quad_t)PUFFS_VNOVAL;
380 	vap->va_vaflags = 0;
381 
382 	if (vap->va_blocksize == 0)
383 		vap->va_blocksize = DEV_BSIZE;
384 
385 	if (vap->va_size == (size_t)PUFFS_VNOVAL) /* XXX */
386 		vap->va_size = 0;
387 
388 	return;
389 }
390 
391 static void
392 set_expire(opc, feo, fao)
393 	puffs_cookie_t opc;
394 	struct fuse_entry_out *feo;
395 	struct fuse_attr_out *fao;
396 {
397 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
398 	struct timespec entry_ts;
399 	struct timespec attr_ts;
400 	struct timespec now;
401 
402 	if ((feo == NULL) && (fao == NULL))
403 		DERRX(EX_SOFTWARE, "%s: feo and fao NULL", __func__);
404 
405 	if ((feo != NULL) && (fao != NULL))
406 		DERRX(EX_SOFTWARE, "%s: feo and fao != NULL", __func__);
407 
408 	if (clock_gettime(CLOCK_REALTIME, &now) != 0)
409 		DERR(EX_OSERR, "clock_gettime failed");
410 
411 	if (feo != NULL) {
412 		entry_ts.tv_sec = (time_t)feo->entry_valid;
413 		entry_ts.tv_nsec = (long)feo->entry_valid_nsec;
414 
415 		timespecadd(&now, &entry_ts, &pnd->pnd_entry_expire);
416 
417 		attr_ts.tv_sec = (time_t)feo->attr_valid;
418 		attr_ts.tv_nsec = (long)feo->attr_valid_nsec;
419 
420 		timespecadd(&now, &attr_ts, &pnd->pnd_attr_expire);
421 	}
422 
423 	if (fao != NULL) {
424 		attr_ts.tv_sec = (time_t)fao->attr_valid;
425 		attr_ts.tv_nsec = (long)fao->attr_valid_nsec;
426 
427 		timespecadd(&now, &attr_ts, &pnd->pnd_attr_expire);
428 	}
429 
430 	return;
431 }
432 
433 static int
434 attr_expired(opc)
435 	puffs_cookie_t opc;
436 {
437 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
438 	struct timespec now;
439 
440 	if (clock_gettime(CLOCK_REALTIME, &now) != 0)
441 		DERR(EX_OSERR, "clock_gettime failed");
442 
443 	return timespeccmp(&pnd->pnd_attr_expire, &now, <);
444 }
445 
446 static int
447 entry_expired(opc)
448 	puffs_cookie_t opc;
449 {
450 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
451 	struct timespec expire = pnd->pnd_entry_expire;
452 	struct timespec now;
453 
454 	if (clock_gettime(CLOCK_REALTIME, &now) != 0)
455 		DERR(EX_OSERR, "clock_gettime failed");
456 
457 	return timespeccmp(&expire, &now, <);
458 }
459 
460 
461 /*
462  * Lookup name in directory opc
463  * We take special care of name being . or ..
464  * These are returned by readdir and deserve tweaks.
465  */
466 static int
467 node_lookup_dir_nodot(pu, opc, name, namelen, pnp)
468 	struct puffs_usermount *pu;
469 	puffs_cookie_t opc;
470 	char *name;
471 	size_t namelen;
472 	struct puffs_node **pnp;
473 {
474 	/*
475 	 * "dot" is easy as we already know it
476 	 */
477 	if (strncmp(name, ".", namelen) == 0) {
478 		*pnp = (struct puffs_node *)opc;
479 		return 0;
480 	}
481 
482 	/*
483 	 * "dotdot" is also known
484 	 */
485 	if (strncmp(name, "..", namelen) == 0) {
486 		*pnp = PERFUSE_NODE_DATA(opc)->pnd_parent;
487 		return 0;
488 	}
489 
490 	return node_lookup_common(pu, opc, name, NULL, pnp);
491 }
492 
493 static int
494 node_lookup_common(pu, opc, path, pcr, pnp)
495 	struct puffs_usermount *pu;
496 	puffs_cookie_t opc;
497 	const char *path;
498 	const struct puffs_cred *pcr;
499 	struct puffs_node **pnp;
500 {
501 	struct perfuse_state *ps;
502 	struct perfuse_node_data *oldpnd;
503 	perfuse_msg_t *pm;
504 	struct fuse_entry_out *feo;
505 	struct puffs_node *pn;
506 	size_t len;
507 	int error;
508 
509 	/*
510 	 * Prevent further lookups if the parent was removed
511 	 */
512 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
513 		return ESTALE;
514 
515 	if (pnp == NULL)
516 		DERRX(EX_SOFTWARE, "pnp must be != NULL");
517 
518 	ps = puffs_getspecific(pu);
519 
520 #ifdef PERFUSE_DEBUG
521 	if (perfuse_diagflags & PDF_FILENAME)
522 		DPRINTF("%s: opc = %p, file = \"%s\" looking up \"%s\"\n",
523 			__func__, (void *)opc, perfuse_node_path(opc), path);
524 #endif
525 	/*
526 	 * Is the node already known?
527 	 */
528 	TAILQ_FOREACH(oldpnd, &PERFUSE_NODE_DATA(opc)->pnd_children, pnd_next) {
529 		if ((oldpnd->pnd_flags & PND_REMOVED) ||
530 		    (strcmp(oldpnd->pnd_name, path) != 0))
531 			continue;
532 
533 #ifdef PERFUSE_DEBUG
534 		if (perfuse_diagflags & PDF_FILENAME)
535 			DPRINTF("%s: opc = %p, file = \"%s\" found "
536 				"cookie = %p, nodeid = 0x%"PRIx64" "
537 				"for \"%s\"\n", __func__,
538 				(void *)opc, perfuse_node_path(opc),
539 				(void *)oldpnd->pnd_pn, oldpnd->pnd_nodeid,
540 				path);
541 #endif
542 		break;
543 	}
544 
545 	/*
546 	 * Check for cached name
547 	 */
548 	if ((oldpnd != NULL) && !entry_expired(oldpnd->pnd_pn)) {
549 		*pnp = oldpnd->pnd_pn;
550 		return 0;
551 	}
552 
553 	len = strlen(path) + 1;
554 
555 	pm = ps->ps_new_msg(pu, opc, FUSE_LOOKUP, len, pcr);
556 	(void)strlcpy(_GET_INPAYLOAD(ps, pm, char *), path, len);
557 
558 	error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply);
559 
560 	switch (error) {
561 	case 0:
562 		break;
563 	case ENOENT:
564 		if (oldpnd != NULL) {
565 			oldpnd->pnd_flags |= PND_REMOVED;
566 #ifdef PERFUSE_DEBUG
567 			if (perfuse_diagflags & PDF_FILENAME)
568 				DPRINTF("%s: opc = %p nodeid = 0x%"PRIx64" "
569 					"file = \"%s\" removed\n", __func__,
570 					oldpnd->pnd_pn, oldpnd->pnd_nodeid,
571 					oldpnd->pnd_name);
572 #endif
573 		}
574 		/* FALLTHROUGH */
575 	default:
576 		return error;
577 		/* NOTREACHED */
578 		break;
579 	}
580 
581 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
582 
583 	if (oldpnd != NULL) {
584 		if (oldpnd->pnd_nodeid == feo->nodeid) {
585 			oldpnd->pnd_nlookup++;
586 			*pnp = oldpnd->pnd_pn;
587 
588 			ps->ps_destroy_msg(pm);
589 			return 0;
590 		} else {
591 			oldpnd->pnd_flags |= PND_REMOVED;
592 #ifdef PERFUSE_DEBUG
593 			if (perfuse_diagflags & PDF_FILENAME)
594 				DPRINTF("%s: opc = %p nodeid = 0x%"PRIx64" "
595 					"file = \"%s\" replaced\n", __func__,
596 					oldpnd->pnd_pn, oldpnd->pnd_nodeid,
597 					oldpnd->pnd_name);
598 #endif
599 		}
600 	}
601 
602 	pn = perfuse_new_pn(pu, path, opc);
603 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
604 
605 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
606 	pn->pn_va.va_gen = (u_long)(feo->generation);
607 	set_expire((puffs_cookie_t)pn, feo, NULL);
608 
609 	*pnp = pn;
610 
611 #ifdef PERFUSE_DEBUG
612 	if (perfuse_diagflags & PDF_FILENAME)
613 		DPRINTF("%s: opc = %p, looked up opc = %p, "
614 			"nodeid = 0x%"PRIx64" file = \"%s\"\n", __func__,
615 			(void *)opc, pn, feo->nodeid, path);
616 #endif
617 
618 	ps->ps_destroy_msg(pm);
619 
620 	return 0;
621 }
622 
623 
624 /*
625  * Common code for methods that create objects:
626  * perfuse_node_mkdir
627  * perfuse_node_mknod
628  * perfuse_node_symlink
629  */
630 static int
631 node_mk_common(pu, opc, pni, pcn, pm)
632 	struct puffs_usermount *pu;
633 	puffs_cookie_t opc;
634 	struct puffs_newinfo *pni;
635 	const struct puffs_cn *pcn;
636 	perfuse_msg_t *pm;
637 {
638 	struct perfuse_state *ps;
639 	struct puffs_node *pn;
640 	struct fuse_entry_out *feo;
641 	int error;
642 
643 	ps =  puffs_getspecific(pu);
644 
645 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
646 		return error;
647 
648 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
649 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
650 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
651 
652 	pn = perfuse_new_pn(pu, pcn->pcn_name, opc);
653 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
654 
655 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
656 	pn->pn_va.va_gen = (u_long)(feo->generation);
657 	set_expire((puffs_cookie_t)pn, feo, NULL);
658 
659 	puffs_newinfo_setcookie(pni, pn);
660 
661 #ifdef PERFUSE_DEBUG
662 	if (perfuse_diagflags & PDF_FILENAME)
663 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
664 			"nodeid = 0x%"PRIx64"\n",
665 			__func__, (void *)pn, pcn->pcn_name,
666 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid);
667 #endif
668 	ps->ps_destroy_msg(pm);
669 
670 	return node_mk_common_final(pu, opc, pn, pcn);
671 }
672 
673 /*
674  * Common final code for methods that create objects:
675  * perfuse_node_mkdir via node_mk_common
676  * perfuse_node_mknod via node_mk_common
677  * perfuse_node_symlink via node_mk_common
678  * perfuse_node_create
679  */
680 static int
681 node_mk_common_final(pu, opc, pn, pcn)
682 	struct puffs_usermount *pu;
683 	puffs_cookie_t opc;
684 	struct puffs_node *pn;
685 	const struct puffs_cn *pcn;
686 {
687 	struct perfuse_state *ps;
688 	perfuse_msg_t *pm;
689 	struct fuse_setattr_in *fsi;
690 	struct fuse_attr_out *fao;
691 	int error;
692 
693 	ps =  puffs_getspecific(pu);
694 
695 	/*
696 	 * Set owner and group. The kernel cannot create a file
697 	 * on its own (puffs_cred_getuid would return -1), right?
698 	 */
699 	if (puffs_cred_getuid(pcn->pcn_cred, &pn->pn_va.va_uid) != 0)
700 		DERRX(EX_SOFTWARE, "puffs_cred_getuid fails in %s", __func__);
701 	if (puffs_cred_getgid(pcn->pcn_cred, &pn->pn_va.va_gid) != 0)
702 		DERRX(EX_SOFTWARE, "puffs_cred_getgid fails in %s", __func__);
703 
704 	pm = ps->ps_new_msg(pu, (puffs_cookie_t)pn,
705 			    FUSE_SETATTR, sizeof(*fsi), pcn->pcn_cred);
706 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
707 	fsi->uid = pn->pn_va.va_uid;
708 	fsi->gid = pn->pn_va.va_gid;
709 	fsi->valid = FUSE_FATTR_UID|FUSE_FATTR_GID;
710 
711 	if ((error = xchg_msg(pu, (puffs_cookie_t)pn, pm,
712 			      sizeof(*fao), wait_reply)) != 0)
713 		return error;
714 
715 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
716 	fuse_attr_to_vap(ps, &pn->pn_va, &fao->attr);
717 	set_expire((puffs_cookie_t)pn, NULL, fao);
718 
719 	/*
720 	 * The parent directory needs a sync
721 	 */
722 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
723 
724 	ps->ps_destroy_msg(pm);
725 
726 	return 0;
727 }
728 
729 static uint64_t
730 readdir_last_cookie(fd, fd_len)
731 	struct fuse_dirent *fd;
732 	size_t fd_len;
733 {
734 	size_t len;
735 	size_t seen = 0;
736 	char *ndp;
737 
738 	do {
739 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
740 		seen += len;
741 
742 		if (seen >= fd_len)
743 			break;
744 
745 		ndp = (char *)(void *)fd + (size_t)len;
746 		fd = (struct fuse_dirent *)(void *)ndp;
747 	} while (1 /* CONSTCOND */);
748 
749 	return fd->off;
750 }
751 
752 static ssize_t
753 fuse_to_dirent(pu, opc, fd, fd_len)
754 	struct puffs_usermount *pu;
755 	puffs_cookie_t opc;
756 	struct fuse_dirent *fd;
757 	size_t fd_len;
758 {
759 	struct dirent *dents;
760 	size_t dents_len;
761 	ssize_t written;
762 	uint64_t fd_offset;
763 	struct fuse_dirent *fd_base;
764 	size_t len;
765 
766 	fd_base = fd;
767 	fd_offset = 0;
768 	written = 0;
769 	dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
770 	dents_len = (size_t)PERFUSE_NODE_DATA(opc)->pnd_dirent_len;
771 
772 	do {
773 		char *ndp;
774 		size_t reclen;
775 
776 		reclen = _DIRENT_RECLEN(dents, fd->namelen);
777 
778 		/*
779 		 * Check we do not overflow the output buffer
780 		 * struct fuse_dirent is bigger than struct dirent,
781 		 * so we should always use fd_len and never reallocate
782 		 * later.
783 		 * If we have to reallocate,try to double the buffer
784 		 * each time so that we do not have to do it too often.
785 		 */
786 		if (written + reclen > dents_len) {
787 			if (dents_len == 0)
788 				dents_len = fd_len;
789 			else
790 				dents_len =
791 				   MAX(2 * dents_len, written + reclen);
792 
793 			dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
794 			if ((dents = realloc(dents, dents_len)) == NULL)
795 				DERR(EX_OSERR, "%s: malloc failed", __func__);
796 
797 			PERFUSE_NODE_DATA(opc)->pnd_dirent = dents;
798 			PERFUSE_NODE_DATA(opc)->pnd_dirent_len = dents_len;
799 
800 			/*
801 			 * (void *) for delint
802 			 */
803 			ndp = (char *)(void *)dents + written;
804 			dents = (struct dirent *)(void *)ndp;
805 		}
806 
807 		/*
808 		 * Filesystem was mounted without -o use_ino
809 		 * Perform a lookup to find it.
810 		 */
811 		if (fd->ino == PERFUSE_UNKNOWN_INO) {
812 			struct puffs_node *pn;
813 
814 			if (node_lookup_dir_nodot(pu, opc, fd->name,
815 						  fd->namelen, &pn) != 0) {
816 				DWARNX("node_lookup_dir_nodot failed");
817 			} else {
818 				fd->ino = pn->pn_va.va_fileid;
819 			}
820 		}
821 
822 		dents->d_fileno = fd->ino;
823 		dents->d_reclen = (unsigned short)reclen;
824 		dents->d_namlen = fd->namelen;
825 		dents->d_type = fd->type;
826 		strlcpy(dents->d_name, fd->name, fd->namelen + 1);
827 
828 #ifdef PERFUSE_DEBUG
829 		if (perfuse_diagflags & PDF_READDIR)
830 			DPRINTF("%s: translated \"%s\" ino = %"PRIu64"\n",
831 				__func__, dents->d_name, dents->d_fileno);
832 #endif
833 
834 		dents = _DIRENT_NEXT(dents);
835 		written += reclen;
836 
837 		/*
838 		 * Move to the next record.
839 		 * fd->off is not the offset, it is an opaque cookie
840 		 * given by the filesystem to keep state across multiple
841 		 * readdir() operation.
842 		 * Use record alignement instead.
843 		 */
844 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
845 #ifdef PERFUSE_DEBUG
846 		if (perfuse_diagflags & PDF_READDIR)
847 			DPRINTF("%s: record at %"PRId64"/0x%"PRIx64" "
848 				"length = %zd/0x%zx. "
849 				"next record at %"PRId64"/0x%"PRIx64" "
850 				"max %zd/0x%zx\n",
851 				__func__, fd_offset, fd_offset, len, len,
852 				fd_offset + len, fd_offset + len,
853 				fd_len, fd_len);
854 #endif
855 		fd_offset += len;
856 
857 		/*
858 		 * Check if next record is still within the packet
859 		 * If it is not, we reached the end of the buffer.
860 		 */
861 		if (fd_offset >= fd_len)
862 			break;
863 
864 		/*
865 		 * (void *) for delint
866 		 */
867 		ndp = (char *)(void *)fd_base + (size_t)fd_offset;
868 		fd = (struct fuse_dirent *)(void *)ndp;
869 
870 	} while (1 /* CONSTCOND */);
871 
872 	/*
873 	 * Adjust the dirent output length
874 	 */
875 	if (written != -1)
876 		PERFUSE_NODE_DATA(opc)->pnd_dirent_len = written;
877 
878 	return written;
879 }
880 
881 static int
882 readdir_buffered(opc, dent, readoff, reslen)
883 	puffs_cookie_t opc;
884 	struct dirent *dent;
885 	off_t *readoff;
886 	size_t *reslen;
887 {
888 	struct dirent *fromdent;
889 	struct perfuse_node_data *pnd;
890 	char *ndp;
891 
892 	pnd = PERFUSE_NODE_DATA(opc);
893 
894 	while (*readoff < pnd->pnd_dirent_len) {
895 		/*
896 		 * (void *) for delint
897 		 */
898 		ndp = (char *)(void *)pnd->pnd_dirent + (size_t)*readoff;
899 		fromdent = (struct dirent *)(void *)ndp;
900 
901 		if (*reslen < _DIRENT_SIZE(fromdent))
902 			break;
903 
904 		memcpy(dent, fromdent, _DIRENT_SIZE(fromdent));
905 		*readoff += _DIRENT_SIZE(fromdent);
906 		*reslen -= _DIRENT_SIZE(fromdent);
907 
908 		dent = _DIRENT_NEXT(dent);
909 	}
910 
911 #ifdef PERFUSE_DEBUG
912 	if (perfuse_diagflags & PDF_READDIR)
913 		DPRINTF("%s: readoff = %"PRId64",  "
914 			"pnd->pnd_dirent_len = %"PRId64"\n",
915 			__func__, *readoff, pnd->pnd_dirent_len);
916 #endif
917 	if (*readoff >=  pnd->pnd_dirent_len) {
918 		free(pnd->pnd_dirent);
919 		pnd->pnd_dirent = NULL;
920 		pnd->pnd_dirent_len = 0;
921 	}
922 
923 	return 0;
924 }
925 
926 static void
927 requeue_request(pu, opc, type)
928 	struct puffs_usermount *pu;
929 	puffs_cookie_t opc;
930 	enum perfuse_qtype type;
931 {
932 	struct perfuse_cc_queue pcq;
933 	struct perfuse_node_data *pnd;
934 #ifdef PERFUSE_DEBUG
935 	struct perfuse_state *ps;
936 
937 	ps = perfuse_getspecific(pu);
938 #endif
939 
940 	pnd = PERFUSE_NODE_DATA(opc);
941 	pcq.pcq_type = type;
942 	pcq.pcq_cc = puffs_cc_getcc(pu);
943 	TAILQ_INSERT_TAIL(&pnd->pnd_pcq, &pcq, pcq_next);
944 
945 #ifdef PERFUSE_DEBUG
946 	if (perfuse_diagflags & PDF_REQUEUE)
947 		DPRINTF("%s: REQUEUE opc = %p, pcc = %p (%s)\n",
948 		        __func__, (void *)opc, pcq.pcq_cc,
949 			perfuse_qtypestr[type]);
950 #endif
951 
952 	puffs_cc_yield(pcq.pcq_cc);
953 	TAILQ_REMOVE(&pnd->pnd_pcq, &pcq, pcq_next);
954 
955 #ifdef PERFUSE_DEBUG
956 	if (perfuse_diagflags & PDF_REQUEUE)
957 		DPRINTF("%s: RESUME opc = %p, pcc = %p (%s)\n",
958 		        __func__, (void *)opc, pcq.pcq_cc,
959 			perfuse_qtypestr[type]);
960 #endif
961 
962 	return;
963 }
964 
965 /* ARGSUSED0 */
966 static int
967 dequeue_requests(ps, opc, type, max)
968 	struct perfuse_state *ps;
969 	puffs_cookie_t opc;
970 	enum perfuse_qtype type;
971 	int max;
972 {
973 	struct perfuse_cc_queue *pcq;
974 	struct perfuse_node_data *pnd;
975 	int dequeued;
976 
977 	pnd = PERFUSE_NODE_DATA(opc);
978 	dequeued = 0;
979 	TAILQ_FOREACH(pcq, &pnd->pnd_pcq, pcq_next) {
980 		if (pcq->pcq_type != type)
981 			continue;
982 
983 #ifdef PERFUSE_DEBUG
984 		if (perfuse_diagflags & PDF_REQUEUE)
985 			DPRINTF("%s: SCHEDULE opc = %p, pcc = %p (%s)\n",
986 				__func__, (void *)opc, pcq->pcq_cc,
987 				 perfuse_qtypestr[type]);
988 #endif
989 		puffs_cc_schedule(pcq->pcq_cc);
990 
991 		if (++dequeued == max)
992 			break;
993 	}
994 
995 #ifdef PERFUSE_DEBUG
996 	if (perfuse_diagflags & PDF_REQUEUE)
997 		DPRINTF("%s: DONE  opc = %p\n", __func__, (void *)opc);
998 #endif
999 
1000 	return dequeued;
1001 }
1002 
1003 void
1004 perfuse_fs_init(pu)
1005 	struct puffs_usermount *pu;
1006 {
1007 	struct perfuse_state *ps;
1008 	perfuse_msg_t *pm;
1009 	struct fuse_init_in *fii;
1010 	struct fuse_init_out *fio;
1011 	int error;
1012 
1013 	ps = puffs_getspecific(pu);
1014 
1015         if (puffs_mount(pu, ps->ps_target, ps->ps_mountflags, ps->ps_root) != 0)
1016                 DERR(EX_OSERR, "%s: puffs_mount failed", __func__);
1017 
1018 	/*
1019 	 * Linux 2.6.34.1 sends theses flags:
1020 	 * FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC
1021 	 * FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK
1022 	 *
1023 	 * Linux also sets max_readahead at 32 pages (128 kB)
1024 	 *
1025 	 * ps_new_msg() is called with NULL creds, which will
1026 	 * be interpreted as FUSE superuser.
1027 	 */
1028 	pm = ps->ps_new_msg(pu, 0, FUSE_INIT, sizeof(*fii), NULL);
1029 	fii = GET_INPAYLOAD(ps, pm, fuse_init_in);
1030 	fii->major = FUSE_KERNEL_VERSION;
1031 	fii->minor = FUSE_KERNEL_MINOR_VERSION;
1032 	fii->max_readahead = (unsigned int)(32 * sysconf(_SC_PAGESIZE));
1033 	fii->flags = (FUSE_ASYNC_READ|FUSE_POSIX_LOCKS|FUSE_ATOMIC_O_TRUNC);
1034 
1035 	if ((error = xchg_msg(pu, 0, pm, sizeof(*fio), wait_reply)) != 0)
1036 		DERRX(EX_SOFTWARE, "init message exchange failed (%d)", error);
1037 
1038 	fio = GET_OUTPAYLOAD(ps, pm, fuse_init_out);
1039 	ps->ps_max_readahead = fio->max_readahead;
1040 	ps->ps_max_write = fio->max_write;
1041 
1042 	ps->ps_destroy_msg(pm);
1043 
1044 	return;
1045 }
1046 
1047 int
1048 perfuse_fs_unmount(pu, flags)
1049 	struct puffs_usermount *pu;
1050 	int flags;
1051 {
1052 	perfuse_msg_t *pm;
1053 	struct perfuse_state *ps;
1054 	puffs_cookie_t opc;
1055 	int error;
1056 
1057 	ps = puffs_getspecific(pu);
1058 	opc = (puffs_cookie_t)puffs_getroot(pu);
1059 
1060 	/*
1061 	 * ps_new_msg() is called with NULL creds, which will
1062 	 * be interpreted as FUSE superuser.
1063 	 */
1064 	pm = ps->ps_new_msg(pu, opc, FUSE_DESTROY, 0, NULL);
1065 
1066 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0){
1067 		DWARN("unmount %s", ps->ps_target);
1068 		if (!(flags & MNT_FORCE))
1069 			return error;
1070 		else
1071 			error = 0;
1072 	} else {
1073 		ps->ps_destroy_msg(pm);
1074 	}
1075 
1076 	ps->ps_umount(pu);
1077 
1078 	if (perfuse_diagflags & PDF_MISC)
1079 		DPRINTF("%s unmounted, exit\n", ps->ps_target);
1080 
1081 	return 0;
1082 }
1083 
1084 int
1085 perfuse_fs_statvfs(pu, svfsb)
1086 	struct puffs_usermount *pu;
1087 	struct statvfs *svfsb;
1088 {
1089 	struct perfuse_state *ps;
1090 	perfuse_msg_t *pm;
1091 	puffs_cookie_t opc;
1092 	struct fuse_statfs_out *fso;
1093 	int error;
1094 
1095 	ps = puffs_getspecific(pu);
1096 	opc = (puffs_cookie_t)puffs_getroot(pu);
1097 
1098 	/*
1099 	 * ps_new_msg() is called with NULL creds, which will
1100 	 * be interpreted as FUSE superuser.
1101 	 */
1102 	pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
1103 
1104 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply)) != 0)
1105 		return error;
1106 
1107 	fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
1108 	svfsb->f_flag = ps->ps_mountflags;
1109 	svfsb->f_bsize = fso->st.bsize;
1110 	svfsb->f_frsize = fso->st.frsize;
1111 	svfsb->f_iosize = ((struct puffs_node *)opc)->pn_va.va_blocksize;
1112 	svfsb->f_blocks = fso->st.blocks;
1113 	svfsb->f_bfree = fso->st.bfree;
1114 	svfsb->f_bavail = fso->st.bavail;
1115 	svfsb->f_bresvd = fso->st.bfree - fso->st.bavail;
1116 	svfsb->f_files = fso->st.files;
1117 	svfsb->f_ffree = fso->st.ffree;
1118 	svfsb->f_favail = fso->st.ffree;/* files not reserved for root */
1119 	svfsb->f_fresvd = 0;		/* files reserved for root */
1120 
1121 	svfsb->f_syncreads = ps->ps_syncreads;
1122 	svfsb->f_syncwrites = ps->ps_syncwrites;
1123 
1124 	svfsb->f_asyncreads = ps->ps_asyncreads;
1125 	svfsb->f_asyncwrites = ps->ps_asyncwrites;
1126 
1127 	(void)memcpy(&svfsb->f_fsidx, &ps->ps_fsid, sizeof(ps->ps_fsid));
1128 	svfsb->f_fsid = (unsigned long)ps->ps_fsid;
1129 	svfsb->f_namemax = MAXPATHLEN;	/* XXX */
1130 	svfsb->f_owner = ps->ps_owner_uid;
1131 
1132 	(void)strlcpy(svfsb->f_mntonname, ps->ps_target, _VFS_NAMELEN);
1133 
1134 	if (ps->ps_filesystemtype != NULL)
1135 		(void)strlcpy(svfsb->f_fstypename,
1136 			      ps->ps_filesystemtype, _VFS_NAMELEN);
1137 	else
1138 		(void)strlcpy(svfsb->f_fstypename, "fuse", _VFS_NAMELEN);
1139 
1140 	if (ps->ps_source != NULL)
1141 		strlcpy(svfsb->f_mntfromname, ps->ps_source, _VFS_NAMELEN);
1142 	else
1143 		strlcpy(svfsb->f_mntfromname, _PATH_FUSE, _VFS_NAMELEN);
1144 
1145 	ps->ps_destroy_msg(pm);
1146 
1147 	return 0;
1148 }
1149 
1150 int
1151 perfuse_fs_sync(pu, waitfor, pcr)
1152 	struct puffs_usermount *pu;
1153 	int waitfor;
1154 	const struct puffs_cred *pcr;
1155 {
1156 	/*
1157 	 * FUSE does not seem to have a FS sync callback.
1158 	 * Maybe do not even register this callback
1159 	 */
1160 	return puffs_fsnop_sync(pu, waitfor, pcr);
1161 }
1162 
1163 /* ARGSUSED0 */
1164 int
1165 perfuse_fs_fhtonode(pu, fid, fidsize, pni)
1166 	struct puffs_usermount *pu;
1167 	void *fid;
1168 	size_t fidsize;
1169 	struct puffs_newinfo *pni;
1170 {
1171 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1172 	return 0;
1173 }
1174 
1175 /* ARGSUSED0 */
1176 int
1177 perfuse_fs_nodetofh(pu, cookie, fid, fidsize)
1178 	struct puffs_usermount *pu;
1179 	puffs_cookie_t cookie;
1180 	void *fid;
1181 	size_t *fidsize;
1182 {
1183 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1184 	return 0;
1185 }
1186 
1187 #if 0
1188 /* ARGSUSED0 */
1189 void
1190 perfuse_fs_extattrctl(pu, cmd, cookie, flags, namespace, attrname)
1191 	struct puffs_usermount *pu;
1192 	int cmd,
1193 	puffs_cookie_t *cookie;
1194 	int flags;
1195 	int namespace;
1196 	const char *attrname;
1197 {
1198 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1199 	return 0;
1200 }
1201 #endif /* 0 */
1202 
1203 /* ARGSUSED0 */
1204 void
1205 perfuse_fs_suspend(pu, status)
1206 	struct puffs_usermount *pu;
1207 	int status;
1208 {
1209 	return;
1210 }
1211 
1212 
1213 
1214 int
1215 perfuse_node_lookup(pu, opc, pni, pcn)
1216 	struct puffs_usermount *pu;
1217 	puffs_cookie_t opc;
1218 	struct puffs_newinfo *pni;
1219 	const struct puffs_cn *pcn;
1220 {
1221 	struct puffs_node *pn;
1222 	mode_t mode;
1223 	int error;
1224 
1225 	/*
1226 	 * Check permissions
1227 	 */
1228 	switch(pcn->pcn_nameiop) {
1229 	case NAMEI_DELETE: /* FALLTHROUGH */
1230 	case NAMEI_RENAME: /* FALLTHROUGH */
1231 	case NAMEI_CREATE:
1232 		if (pcn->pcn_flags & NAMEI_ISLASTCN)
1233 			mode = PUFFS_VEXEC|PUFFS_VWRITE;
1234 		else
1235 			mode = PUFFS_VEXEC;
1236 		break;
1237 	case NAMEI_LOOKUP: /* FALLTHROUGH */
1238 	default:
1239 		mode = PUFFS_VEXEC;
1240 		break;
1241 	}
1242 
1243 	if ((error = mode_access(opc, pcn->pcn_cred, mode)) != 0)
1244 		return error;
1245 
1246 	/*
1247 	 * Special case for ..
1248 	 */
1249 	if (strcmp(pcn->pcn_name, "..") == 0)
1250 		pn = PERFUSE_NODE_DATA(opc)->pnd_parent;
1251 	else
1252 		error = node_lookup_common(pu, (puffs_cookie_t)opc,
1253 					   pcn->pcn_name, pcn->pcn_cred, &pn);
1254 	if (error != 0)
1255 		return error;
1256 
1257 	/*
1258 	 * Kernel would kill us if the filesystem returned the parent
1259 	 * itself. If we want to live, hide that!
1260 	 */
1261 	if ((opc == (puffs_cookie_t)pn) && (strcmp(pcn->pcn_name, ".") != 0)) {
1262 		DWARNX("lookup returned parent");
1263 		return ESTALE;
1264 	}
1265 
1266 	/*
1267 	 * Removed node
1268 	 */
1269 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_REMOVED)
1270 		return ENOENT;
1271 
1272 	/*
1273 	 * Check for sticky bit. Unfortunately there is no way to
1274 	 * do this before creating the puffs_node, since we require
1275 	 * this operation to get the node owner.
1276 	 */
1277 	switch (pcn->pcn_nameiop) {
1278 	case NAMEI_DELETE: /* FALLTHROUGH */
1279 	case NAMEI_RENAME:
1280 		error = sticky_access(pn, pcn->pcn_cred);
1281 		if (error != 0) {
1282 			/*
1283 			 * kernel will never know about it and will
1284 			 * not reclaim it. The filesystem needs to
1285 			 * clean it up anyway, therefore mimick a forget.
1286 			 */
1287 			PERFUSE_NODE_DATA(pn)->pnd_flags |= PND_RECLAIMED;
1288 			(void)perfuse_node_reclaim(pu, (puffs_cookie_t)pn);
1289 			return error;
1290 		}
1291 		break;
1292 	default:
1293 		break;
1294 	}
1295 
1296 	/*
1297 	 * If that node had a pending reclaim, wipe it out.
1298 	 */
1299 	PERFUSE_NODE_DATA(pn)->pnd_flags &= ~PND_RECLAIMED;
1300 
1301 	puffs_newinfo_setcookie(pni, pn);
1302 	puffs_newinfo_setvtype(pni, pn->pn_va.va_type);
1303 	puffs_newinfo_setsize(pni, (voff_t)pn->pn_va.va_size);
1304 	puffs_newinfo_setrdev(pni, pn->pn_va.va_rdev);
1305 
1306 	return error;
1307 }
1308 
1309 int
1310 perfuse_node_create(pu, opc, pni, pcn, vap)
1311 	struct puffs_usermount *pu;
1312 	puffs_cookie_t opc;
1313 	struct puffs_newinfo *pni;
1314 	const struct puffs_cn *pcn;
1315 	const struct vattr *vap;
1316 {
1317 	perfuse_msg_t *pm;
1318 	struct perfuse_state *ps;
1319 	struct fuse_create_in *fci;
1320 	struct fuse_entry_out *feo;
1321 	struct fuse_open_out *foo;
1322 	struct puffs_node *pn;
1323 	const char *name;
1324 	size_t namelen;
1325 	size_t len;
1326 	int error;
1327 
1328 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1329 		return ENOENT;
1330 
1331 	/*
1332 	 * If create is unimplemented: Check that it does not
1333 	 * already exists, and if not, do mknod and open
1334 	 */
1335 	ps = puffs_getspecific(pu);
1336 	if (ps->ps_flags & PS_NO_CREAT) {
1337 		error = node_lookup_common(pu, opc, pcn->pcn_name,
1338 					   pcn->pcn_cred, &pn);
1339 		if (error == 0)
1340 			return EEXIST;
1341 
1342 		error = perfuse_node_mknod(pu, opc, pni, pcn, vap);
1343 		if (error != 0)
1344 			return error;
1345 
1346 		error = node_lookup_common(pu, opc, pcn->pcn_name,
1347 					   pcn->pcn_cred, &pn);
1348 		if (error != 0)
1349 			return error;
1350 
1351 		/*
1352 		 * FUSE does the open at create time, while
1353 		 * NetBSD will open in a subsequent operation.
1354 		 * We need to open now, in order to retain FUSE
1355 		 * semantics. The calling process will not get
1356 		 * a file descriptor before the kernel sends
1357 		 * the open operation.
1358 		 */
1359 		opc = (puffs_cookie_t)pn;
1360 		error = perfuse_node_open(pu, opc, FWRITE, pcn->pcn_cred);
1361 		if (error != 0)
1362 			return error;
1363 
1364 		return 0;
1365 	}
1366 
1367 	name = pcn->pcn_name;
1368 	namelen = pcn->pcn_namelen + 1;
1369 	len = sizeof(*fci) + namelen;
1370 
1371 	/*
1372 	 * flags should use O_WRONLY instead of O_RDWR, but it
1373 	 * breaks when the caller tries to read from file.
1374 	 *
1375 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1376 	 */
1377 	pm = ps->ps_new_msg(pu, opc, FUSE_CREATE, len, pcn->pcn_cred);
1378 	fci = GET_INPAYLOAD(ps, pm, fuse_create_in);
1379 	fci->flags = O_CREAT | O_TRUNC | O_RDWR;
1380 	fci->mode = vap->va_mode | VTTOIF(vap->va_type);
1381 	fci->umask = 0; 	/* Seems unused by libfuse */
1382 	(void)strlcpy((char*)(void *)(fci + 1), name, namelen);
1383 
1384 	len = sizeof(*feo) + sizeof(*foo);
1385 	if ((error = xchg_msg(pu, opc, pm, len, wait_reply)) != 0) {
1386 		/*
1387 		 * create is unimplmented, remember it for later,
1388 		 * and start over using mknod and open instead.
1389 		 */
1390 		if (error == ENOSYS) {
1391 			ps->ps_flags |= PS_NO_CREAT;
1392 			return perfuse_node_create(pu, opc, pni, pcn, vap);
1393 		}
1394 
1395 		return error;
1396 	}
1397 
1398 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
1399 	foo = (struct fuse_open_out *)(void *)(feo + 1);
1400 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
1401 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
1402 
1403 	/*
1404 	 * Save the file handle and inode in node private data
1405 	 * so that we can reuse it later
1406 	 */
1407 	pn = perfuse_new_pn(pu, name, opc);
1408 	perfuse_new_fh((puffs_cookie_t)pn, foo->fh, FWRITE);
1409 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
1410 
1411 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
1412 	pn->pn_va.va_gen = (u_long)(feo->generation);
1413 	set_expire((puffs_cookie_t)pn, feo, NULL);
1414 
1415 	puffs_newinfo_setcookie(pni, pn);
1416 
1417 #ifdef PERFUSE_DEBUG
1418 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1419 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
1420 			"nodeid = 0x%"PRIx64", wfh = 0x%"PRIx64"\n",
1421 			__func__, (void *)pn, pcn->pcn_name,
1422 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid,
1423 			foo->fh);
1424 #endif
1425 
1426 	ps->ps_destroy_msg(pm);
1427 
1428 	return node_mk_common_final(pu, opc, pn, pcn);
1429 }
1430 
1431 
1432 int
1433 perfuse_node_mknod(pu, opc, pni, pcn, vap)
1434 	struct puffs_usermount *pu;
1435 	puffs_cookie_t opc;
1436 	struct puffs_newinfo *pni;
1437 	const struct puffs_cn *pcn;
1438 	const struct vattr *vap;
1439 {
1440 	struct perfuse_state *ps;
1441 	perfuse_msg_t *pm;
1442 	struct fuse_mknod_in *fmi;
1443 	const char* path;
1444 	size_t len;
1445 
1446 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1447 		return ENOENT;
1448 
1449 	/*
1450 	 * Only superuser can mknod objects other than
1451 	 * directories, files, socks, fifo and links.
1452 	 *
1453 	 * Create an object require -WX permission in the parent directory
1454 	 */
1455 	switch (vap->va_type) {
1456 	case VDIR:	/* FALLTHROUGH */
1457 	case VREG:	/* FALLTHROUGH */
1458 	case VFIFO:	/* FALLTHROUGH */
1459 	case VSOCK:
1460 		break;
1461 	default:	/* VNON, VBLK, VCHR, VBAD */
1462 		if (!puffs_cred_isjuggernaut(pcn->pcn_cred))
1463 			return EACCES;
1464 		break;
1465 	}
1466 
1467 
1468 	ps = puffs_getspecific(pu);
1469 	path = pcn->pcn_name;
1470 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
1471 
1472 	/*
1473 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1474 	 */
1475 	pm = ps->ps_new_msg(pu, opc, FUSE_MKNOD, len, pcn->pcn_cred);
1476 	fmi = GET_INPAYLOAD(ps, pm, fuse_mknod_in);
1477 	fmi->mode = vap->va_mode | VTTOIF(vap->va_type);
1478 	fmi->rdev = (uint32_t)vap->va_rdev;
1479 	fmi->umask = 0; 	/* Seems unused bu libfuse */
1480 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
1481 
1482 	return node_mk_common(pu, opc, pni, pcn, pm);
1483 }
1484 
1485 
1486 int
1487 perfuse_node_open(pu, opc, mode, pcr)
1488 	struct puffs_usermount *pu;
1489 	puffs_cookie_t opc;
1490 	int mode;
1491 	const struct puffs_cred *pcr;
1492 {
1493 	struct perfuse_state *ps;
1494 	struct perfuse_node_data *pnd;
1495 	perfuse_msg_t *pm;
1496 	mode_t fmode;
1497 	int op;
1498 	struct fuse_open_in *foi;
1499 	struct fuse_open_out *foo;
1500 	struct puffs_node *pn;
1501 	int error;
1502 
1503 	ps = puffs_getspecific(pu);
1504 	pn = (struct puffs_node *)opc;
1505 	pnd = PERFUSE_NODE_DATA(opc);
1506 	error = 0;
1507 
1508 	if (pnd->pnd_flags & PND_REMOVED)
1509 		return ENOENT;
1510 
1511 	if (puffs_pn_getvap(pn)->va_type == VDIR)
1512 		op = FUSE_OPENDIR;
1513 	else
1514 		op = FUSE_OPEN;
1515 
1516 	/*
1517 	 * libfuse docs says
1518 	 * - O_CREAT and O_EXCL should never be set.
1519 	 * - O_TRUNC may be used if mount option atomic_o_trunc is used XXX
1520 	 *
1521 	 * O_APPEND makes no sense since FUSE always sends
1522 	 * the file offset for write operations. If the
1523 	 * filesystem uses pwrite(), O_APPEND would cause
1524 	 * the offset to be ignored and cause file corruption.
1525 	 */
1526 	mode &= ~(O_CREAT|O_EXCL|O_APPEND);
1527 
1528 	/*
1529 	 * Do not open twice, and do not reopen for reading
1530 	 * if we already have write handle.
1531 	 */
1532 	if (((mode & FREAD) && (pnd->pnd_flags & PND_RFH)) ||
1533 	    ((mode & FREAD) && (pnd->pnd_flags & PND_WFH)) ||
1534 	    ((mode & FWRITE) && (pnd->pnd_flags & PND_WFH)))
1535 		goto out;
1536 
1537 	/*
1538 	 * Queue open on a node so that we do not open
1539 	 * twice. This would be better with read and
1540 	 * write distinguished.
1541 	 */
1542 	while (pnd->pnd_flags & PND_INOPEN)
1543 		requeue_request(pu, opc, PCQ_OPEN);
1544 	pnd->pnd_flags |= PND_INOPEN;
1545 
1546 	/*
1547 	 * Convert PUFFS mode to FUSE mode: convert FREAD/FWRITE
1548 	 * to O_RDONLY/O_WRONLY while perserving the other options.
1549 	 */
1550 	fmode = mode & ~(FREAD|FWRITE);
1551 	fmode |= (mode & FWRITE) ? O_RDWR : O_RDONLY;
1552 
1553 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*foi), pcr);
1554 	foi = GET_INPAYLOAD(ps, pm, fuse_open_in);
1555 	foi->flags = fmode;
1556 	foi->unused = 0;
1557 
1558 	if ((error = xchg_msg(pu, opc, pm, sizeof(*foo), wait_reply)) != 0)
1559 		goto out;
1560 
1561 	foo = GET_OUTPAYLOAD(ps, pm, fuse_open_out);
1562 
1563 	/*
1564 	 * Save the file handle in node private data
1565 	 * so that we can reuse it later
1566 	 */
1567 	perfuse_new_fh(opc, foo->fh, mode);
1568 
1569 #ifdef PERFUSE_DEBUG
1570 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1571 		DPRINTF("%s: opc = %p, file = \"%s\", "
1572 			"nodeid = 0x%"PRIx64", %s%sfh = 0x%"PRIx64"\n",
1573 			__func__, (void *)opc, perfuse_node_path(opc),
1574 			pnd->pnd_nodeid, mode & FREAD ? "r" : "",
1575 			mode & FWRITE ? "w" : "", foo->fh);
1576 #endif
1577 
1578 	ps->ps_destroy_msg(pm);
1579 out:
1580 
1581 	pnd->pnd_flags &= ~PND_INOPEN;
1582 	(void)dequeue_requests(ps, opc, PCQ_OPEN, DEQUEUE_ALL);
1583 
1584 	return error;
1585 }
1586 
1587 /* ARGSUSED0 */
1588 int
1589 perfuse_node_close(pu, opc, flags, pcr)
1590 	struct puffs_usermount *pu;
1591 	puffs_cookie_t opc;
1592 	int flags;
1593 	const struct puffs_cred *pcr;
1594 {
1595 	struct perfuse_node_data *pnd;
1596 
1597 	pnd = PERFUSE_NODE_DATA(opc);
1598 
1599 	if (!(pnd->pnd_flags & PND_OPEN))
1600 		return EBADF;
1601 
1602 	/*
1603 	 * Actual close is postponed at inactive time.
1604 	 */
1605 	return 0;
1606 }
1607 
1608 int
1609 perfuse_node_access(pu, opc, mode, pcr)
1610 	struct puffs_usermount *pu;
1611 	puffs_cookie_t opc;
1612 	int mode;
1613 	const struct puffs_cred *pcr;
1614 {
1615 	perfuse_msg_t *pm;
1616 	struct perfuse_state *ps;
1617 	struct fuse_access_in *fai;
1618 	int error;
1619 
1620 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1621 		return ENOENT;
1622 
1623 	/*
1624 	 * If we previously detected the filesystem does not
1625 	 * implement access(), short-circuit the call and skip
1626 	 * to libpuffs access() emulation.
1627 	 */
1628 	ps = puffs_getspecific(pu);
1629 	if (ps->ps_flags & PS_NO_ACCESS) {
1630 		const struct vattr *vap;
1631 
1632 		vap = puffs_pn_getvap((struct puffs_node *)opc);
1633 
1634 		error = puffs_access(IFTOVT(vap->va_mode),
1635 				     vap->va_mode & ACCESSPERMS,
1636 				     vap->va_uid, vap->va_gid,
1637 				     (mode_t)mode, pcr);
1638 		return error;
1639 	}
1640 
1641 	/*
1642 	 * Plain access call
1643 	 */
1644 	pm = ps->ps_new_msg(pu, opc, FUSE_ACCESS, sizeof(*fai), pcr);
1645 	fai = GET_INPAYLOAD(ps, pm, fuse_access_in);
1646 	fai->mask = 0;
1647 	fai->mask |= (mode & PUFFS_VREAD) ? R_OK : 0;
1648 	fai->mask |= (mode & PUFFS_VWRITE) ? W_OK : 0;
1649 	fai->mask |= (mode & PUFFS_VEXEC) ? X_OK : 0;
1650 
1651 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
1652 
1653 	ps->ps_destroy_msg(pm);
1654 
1655 	/*
1656 	 * If unimplemented, start over with emulation
1657 	 */
1658 	if (error == ENOSYS) {
1659 		ps->ps_flags |= PS_NO_ACCESS;
1660 		return perfuse_node_access(pu, opc, mode, pcr);
1661 	}
1662 
1663 	return error;
1664 }
1665 
1666 int
1667 perfuse_node_getattr(pu, opc, vap, pcr)
1668 	struct puffs_usermount *pu;
1669 	puffs_cookie_t opc;
1670 	struct vattr *vap;
1671 	const struct puffs_cred *pcr;
1672 {
1673 	perfuse_msg_t *pm = NULL;
1674 	struct perfuse_state *ps;
1675 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
1676 	struct fuse_getattr_in *fgi;
1677 	struct fuse_attr_out *fao;
1678 	int error = 0;
1679 
1680 	if (pnd->pnd_flags & PND_REMOVED)
1681 		return ENOENT;
1682 
1683 	/*
1684 	 * Serialize size access, see comment in perfuse_node_setattr().
1685 	 */
1686 	while (pnd->pnd_flags & PND_INRESIZE)
1687 		requeue_request(pu, opc, PCQ_RESIZE);
1688 	pnd->pnd_flags |= PND_INRESIZE;
1689 
1690 	ps = puffs_getspecific(pu);
1691 
1692 	/*
1693 	 * Check for cached attributes
1694 	 * This still require serialized access to size.
1695 	 */
1696 	if (!attr_expired(opc)) {
1697 		(void)memcpy(vap, puffs_pn_getvap((struct puffs_node *)opc),
1698 			     sizeof(*vap));
1699 		goto out;
1700 	}
1701 
1702 	/*
1703 	 * FUSE_GETATTR_FH must be set in fgi->flags
1704 	 * if we use for fgi->fh
1705 	 */
1706 	pm = ps->ps_new_msg(pu, opc, FUSE_GETATTR, sizeof(*fgi), pcr);
1707 	fgi = GET_INPAYLOAD(ps, pm, fuse_getattr_in);
1708 	fgi->getattr_flags = 0;
1709 	fgi->dummy = 0;
1710 	fgi->fh = 0;
1711 
1712 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_OPEN) {
1713 		fgi->fh = perfuse_get_fh(opc, FREAD);
1714 		fgi->getattr_flags |= FUSE_GETATTR_FH;
1715 	}
1716 
1717 #ifdef PERFUSE_DEBUG
1718 	if (perfuse_diagflags & PDF_RESIZE)
1719 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__, (void *)opc,
1720 		    vap->va_size);
1721 #endif
1722 
1723 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
1724 		goto out;
1725 
1726 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1727 
1728 #ifdef PERFUSE_DEBUG
1729 	if (perfuse_diagflags & PDF_RESIZE)
1730 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1731 		    (void *)opc, vap->va_size, fao->attr.size);
1732 #endif
1733 
1734 	/*
1735 	 * We set birthtime, flags, filerev,vaflags to 0.
1736 	 * This seems the best bet, since the information is
1737 	 * not available from filesystem.
1738 	 */
1739 	fuse_attr_to_vap(ps, vap, &fao->attr);
1740 	set_expire(opc, NULL, fao);
1741 
1742 	ps->ps_destroy_msg(pm);
1743 out:
1744 
1745 	pnd->pnd_flags &= ~PND_INRESIZE;
1746 	(void)dequeue_requests(ps, opc, PCQ_RESIZE, DEQUEUE_ALL);
1747 
1748 	return error;
1749 }
1750 
1751 int
1752 perfuse_node_setattr(pu, opc, vap, pcr)
1753 	struct puffs_usermount *pu;
1754 	puffs_cookie_t opc;
1755 	const struct vattr *vap;
1756 	const struct puffs_cred *pcr;
1757 {
1758 	perfuse_msg_t *pm;
1759 	uint64_t fh;
1760 	struct perfuse_state *ps;
1761 	struct perfuse_node_data *pnd;
1762 	struct fuse_setattr_in *fsi;
1763 	struct fuse_attr_out *fao;
1764 	struct vattr *old_va;
1765 	int error;
1766 #ifdef PERFUSE_DEBUG
1767 	struct vattr *old_vap;
1768 	int resize_debug = 0;
1769 #endif
1770 
1771 	ps = puffs_getspecific(pu);
1772 	pnd = PERFUSE_NODE_DATA(opc);
1773 
1774 	/*
1775 	 * The only operation we can do once the file is removed
1776 	 * is to resize it, and we can do it only if it is open.
1777 	 * Do not even send the operation to the filesystem: the
1778 	 * file is not there anymore.
1779 	 */
1780 	if (pnd->pnd_flags & PND_REMOVED) {
1781 		if (!(pnd->pnd_flags & PND_OPEN))
1782 			return ENOENT;
1783 
1784 		error = 0;
1785 		goto out;
1786 	}
1787 
1788 	old_va = puffs_pn_getvap((struct puffs_node *)opc);
1789 
1790 	/*
1791 	 * Check for permission to change size
1792 	 */
1793 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1794 	    (error = mode_access(opc, pcr, PUFFS_VWRITE)) != 0)
1795 		return error;
1796 
1797 	/*
1798 	 * Check for permission to change dates
1799 	 */
1800 	if (((vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1801 	     (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL)) &&
1802 	    (puffs_access_times(old_va->va_uid, old_va->va_gid,
1803 				old_va->va_mode, 0, pcr) != 0))
1804 		return EACCES;
1805 
1806 	/*
1807 	 * Check for permission to change owner and group
1808 	 */
1809 	if (((vap->va_uid != (uid_t)PUFFS_VNOVAL) ||
1810 	     (vap->va_gid != (gid_t)PUFFS_VNOVAL)) &&
1811 	    (puffs_access_chown(old_va->va_uid, old_va->va_gid,
1812 				vap->va_uid, vap->va_gid, pcr)) != 0)
1813 		return EACCES;
1814 
1815 	/*
1816 	 * Check for permission to change permissions
1817 	 */
1818 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1819 	    (puffs_access_chmod(old_va->va_uid, old_va->va_gid,
1820 				old_va->va_type, vap->va_mode, pcr)) != 0)
1821 		return EACCES;
1822 
1823 	pm = ps->ps_new_msg(pu, opc, FUSE_SETATTR, sizeof(*fsi), pcr);
1824 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
1825 	fsi->valid = 0;
1826 
1827 	/*
1828 	 * Get a fh if the node is open for writing
1829 	 */
1830 	if (pnd->pnd_flags & PND_WFH) {
1831 		fh = perfuse_get_fh(opc, FWRITE);
1832 		fsi->fh = fh;
1833 		fsi->valid |= FUSE_FATTR_FH;
1834 	}
1835 
1836 	if (vap->va_size != (u_quad_t)PUFFS_VNOVAL) {
1837 		fsi->size = vap->va_size;
1838 		fsi->valid |= FUSE_FATTR_SIZE;
1839 
1840 		/*
1841 		 * Serialize anything that can touch file size
1842 		 * to avoid reordered GETATTR and SETATTR.
1843 		 * Out of order SETATTR can report stale size,
1844 		 * which will cause the kernel to truncate the file.
1845 		 */
1846 		while (pnd->pnd_flags & PND_INRESIZE)
1847 			requeue_request(pu, opc, PCQ_RESIZE);
1848 		pnd->pnd_flags |= PND_INRESIZE;
1849 	}
1850 
1851 	/*
1852  	 * Setting mtime without atime or vice versa leads to
1853 	 * dates being reset to Epoch on glusterfs. If one
1854 	 * is missing, use the old value.
1855  	 */
1856 	if ((vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1857 	    (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL)) {
1858 
1859 		if (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) {
1860 			fsi->atime = vap->va_atime.tv_sec;
1861 			fsi->atimensec = (uint32_t)vap->va_atime.tv_nsec;
1862 		} else {
1863 			fsi->atime = old_va->va_atime.tv_sec;
1864 			fsi->atimensec = (uint32_t)old_va->va_atime.tv_nsec;
1865 		}
1866 
1867 		if (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) {
1868 			fsi->mtime = vap->va_mtime.tv_sec;
1869 			fsi->mtimensec = (uint32_t)vap->va_mtime.tv_nsec;
1870 		} else {
1871 			fsi->mtime = old_va->va_mtime.tv_sec;
1872 			fsi->mtimensec = (uint32_t)old_va->va_mtime.tv_nsec;
1873 		}
1874 
1875 		fsi->valid |= (FUSE_FATTR_MTIME|FUSE_FATTR_ATIME);
1876 	}
1877 
1878 	if (vap->va_mode != (mode_t)PUFFS_VNOVAL) {
1879 		fsi->mode = vap->va_mode;
1880 		fsi->valid |= FUSE_FATTR_MODE;
1881 	}
1882 
1883 	if (vap->va_uid != (uid_t)PUFFS_VNOVAL) {
1884 		fsi->uid = vap->va_uid;
1885 		fsi->valid |= FUSE_FATTR_UID;
1886 	}
1887 
1888 	if (vap->va_gid != (gid_t)PUFFS_VNOVAL) {
1889 		fsi->gid = vap->va_gid;
1890 		fsi->valid |= FUSE_FATTR_GID;
1891 	}
1892 
1893 	if (pnd->pnd_lock_owner != 0) {
1894 		fsi->lock_owner = pnd->pnd_lock_owner;
1895 		fsi->valid |= FUSE_FATTR_LOCKOWNER;
1896 	}
1897 
1898 	/*
1899 	 * ftruncate() sends only va_size, and metadata cache
1900 	 * flush adds va_atime and va_mtime. Some FUSE
1901 	 * filesystems will attempt to detect ftruncate by
1902 	 * checking for FATTR_SIZE being set without
1903 	 * FATTR_UID|FATTR_GID|FATTR_ATIME|FATTR_MTIME|FATTR_MODE
1904 	 *
1905 	 * Try to adapt and remove FATTR_ATIME|FATTR_MTIME
1906 	 * if we suspect a ftruncate().
1907 	 */
1908 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1909 	    ((vap->va_mode == (mode_t)PUFFS_VNOVAL) &&
1910 	     (vap->va_uid == (uid_t)PUFFS_VNOVAL) &&
1911 	     (vap->va_gid == (gid_t)PUFFS_VNOVAL))) {
1912 		fsi->atime = 0;
1913 		fsi->atimensec = 0;
1914 		fsi->mtime = 0;
1915 		fsi->mtimensec = 0;
1916 		fsi->valid &= ~(FUSE_FATTR_ATIME|FUSE_FATTR_MTIME);
1917 	}
1918 
1919 	/*
1920 	 * If nothing remain, discard the operation.
1921 	 */
1922 	if (!(fsi->valid & (FUSE_FATTR_SIZE|FUSE_FATTR_ATIME|FUSE_FATTR_MTIME|
1923 			    FUSE_FATTR_MODE|FUSE_FATTR_UID|FUSE_FATTR_GID))) {
1924 		error = 0;
1925 		goto out;
1926 	}
1927 
1928 #ifdef PERFUSE_DEBUG
1929 	old_vap = puffs_pn_getvap((struct puffs_node *)opc);
1930 
1931 	if ((perfuse_diagflags & PDF_RESIZE) &&
1932 	    (old_vap->va_size != (u_quad_t)PUFFS_VNOVAL)) {
1933 		resize_debug = 1;
1934 
1935 		DPRINTF(">> %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1936 		    (void *)opc,
1937 		    puffs_pn_getvap((struct puffs_node *)opc)->va_size,
1938 		    fsi->size);
1939 	}
1940 #endif
1941 
1942 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
1943 		goto out;
1944 
1945 	/*
1946 	 * Copy back the new values
1947 	 */
1948 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1949 
1950 #ifdef PERFUSE_DEBUG
1951 	if (resize_debug)
1952 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1953 		    (void *)opc, old_vap->va_size, fao->attr.size);
1954 #endif
1955 
1956 	fuse_attr_to_vap(ps, old_va, &fao->attr);
1957 	set_expire(opc, NULL, fao);
1958 
1959 	ps->ps_destroy_msg(pm);
1960 
1961 out:
1962 	if (pnd->pnd_flags & PND_INRESIZE) {
1963 		pnd->pnd_flags &= ~PND_INRESIZE;
1964 		(void)dequeue_requests(ps, opc, PCQ_RESIZE, DEQUEUE_ALL);
1965 	}
1966 
1967 	return error;
1968 }
1969 
1970 int
1971 perfuse_node_poll(pu, opc, events)
1972 	struct puffs_usermount *pu;
1973 	puffs_cookie_t opc;
1974 	int *events;
1975 {
1976 	struct perfuse_state *ps;
1977 	perfuse_msg_t *pm;
1978 	struct fuse_poll_in *fpi;
1979 	struct fuse_poll_out *fpo;
1980 	int error;
1981 
1982 	ps = puffs_getspecific(pu);
1983 	/*
1984 	 * kh is set if FUSE_POLL_SCHEDULE_NOTIFY is set.
1985 	 *
1986 	 * XXX ps_new_msg() is called with NULL creds, which will
1987 	 * be interpreted as FUSE superuser. We have no way to
1988 	 * know the requesting process' credential, but since poll
1989 	 * is supposed to operate on a file that has been open,
1990 	 * permission should have already been checked at open time.
1991 	 * That still may breaks on filesystems that provides odd
1992 	 * semantics.
1993  	 */
1994 	pm = ps->ps_new_msg(pu, opc, FUSE_POLL, sizeof(*fpi), NULL);
1995 	fpi = GET_INPAYLOAD(ps, pm, fuse_poll_in);
1996 	fpi->fh = perfuse_get_fh(opc, FREAD);
1997 	fpi->kh = 0;
1998 	fpi->flags = 0;
1999 
2000 #ifdef PERFUSE_DEBUG
2001 	if (perfuse_diagflags & PDF_FH)
2002 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
2003 			"fh = 0x%"PRIx64"\n", __func__, (void *)opc,
2004 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fpi->fh);
2005 #endif
2006 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fpo), wait_reply)) != 0)
2007 		return error;
2008 
2009 	fpo = GET_OUTPAYLOAD(ps, pm, fuse_poll_out);
2010 	*events = fpo->revents;
2011 
2012 	ps->ps_destroy_msg(pm);
2013 
2014 	return 0;
2015 }
2016 
2017 /* ARGSUSED0 */
2018 int
2019 perfuse_node_mmap(pu, opc, flags, pcr)
2020 	struct puffs_usermount *pu;
2021 	puffs_cookie_t opc;
2022 	int flags;
2023 	const struct puffs_cred *pcr;
2024 {
2025 	/*
2026 	 * Not implemented anymore in libfuse
2027 	 */
2028 	return ENOSYS;
2029 }
2030 
2031 /* ARGSUSED2 */
2032 int
2033 perfuse_node_fsync(pu, opc, pcr, flags, offlo, offhi)
2034 	struct puffs_usermount *pu;
2035 	puffs_cookie_t opc;
2036 	const struct puffs_cred *pcr;
2037 	int flags;
2038 	off_t offlo;
2039 	off_t offhi;
2040 {
2041 	int op;
2042 	perfuse_msg_t *pm;
2043 	struct perfuse_state *ps;
2044 	struct perfuse_node_data *pnd;
2045 	struct fuse_fsync_in *ffi;
2046 	uint64_t fh;
2047 	int error;
2048 
2049 	pm = NULL;
2050 	ps = puffs_getspecific(pu);
2051 	pnd = PERFUSE_NODE_DATA(opc);
2052 
2053 	/*
2054 	 * No need to sync a removed node
2055 	 */
2056 	if (pnd->pnd_flags & PND_REMOVED)
2057 		return 0;
2058 
2059 	/*
2060 	 * We do not sync closed files. They have been
2061 	 * sync at inactive time already.
2062 	 */
2063 	if (!(pnd->pnd_flags & PND_OPEN))
2064 		return 0;
2065 
2066 	if (puffs_pn_getvap((struct puffs_node *)opc)->va_type == VDIR)
2067 		op = FUSE_FSYNCDIR;
2068 	else 		/* VREG but also other types such as VLNK */
2069 		op = FUSE_FSYNC;
2070 
2071 	/*
2072 	 * Do not sync if there are no change to sync
2073 	 * XXX remove that test on files if we implement mmap
2074 	 */
2075 #ifdef PERFUSE_DEBUG
2076 	if (perfuse_diagflags & PDF_SYNC)
2077 		DPRINTF("%s: TEST opc = %p, file = \"%s\" is %sdirty\n",
2078 			__func__, (void*)opc, perfuse_node_path(opc),
2079 			pnd->pnd_flags & PND_DIRTY ? "" : "not ");
2080 #endif
2081 	if (!(pnd->pnd_flags & PND_DIRTY))
2082 		return 0;
2083 
2084 	/*
2085 	 * It seems NetBSD can call fsync without open first
2086 	 * glusterfs complain in such a situation:
2087 	 * "FSYNC() ERR => -1 (Invalid argument)"
2088 	 * The file will be closed at inactive time.
2089 	 *
2090 	 * We open the directory for reading in order to sync.
2091 	 * This sounds rather counterintuitive, but it works.
2092 	 */
2093 	if (!(pnd->pnd_flags & PND_WFH)) {
2094 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2095 			goto out;
2096 	}
2097 
2098 	if (op == FUSE_FSYNCDIR)
2099 		fh = perfuse_get_fh(opc, FREAD);
2100 	else
2101 		fh = perfuse_get_fh(opc, FWRITE);
2102 
2103 	/*
2104 	 * If fsync_flags  is set, meta data should not be flushed.
2105 	 */
2106 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*ffi), pcr);
2107 	ffi = GET_INPAYLOAD(ps, pm, fuse_fsync_in);
2108 	ffi->fh = fh;
2109 	ffi->fsync_flags = (flags & FFILESYNC) ? 0 : 1;
2110 
2111 #ifdef PERFUSE_DEBUG
2112 	if (perfuse_diagflags & PDF_FH)
2113 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
2114 			__func__, (void *)opc,
2115 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, ffi->fh);
2116 #endif
2117 
2118 	if ((error = xchg_msg(pu, opc, pm,
2119 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
2120 		goto out;
2121 
2122 	/*
2123 	 * No reply beyond fuse_out_header: nothing to do on success
2124 	 * just clear the dirty flag
2125 	 */
2126 	pnd->pnd_flags &= ~PND_DIRTY;
2127 
2128 #ifdef PERFUSE_DEBUG
2129 	if (perfuse_diagflags & PDF_SYNC)
2130 		DPRINTF("%s: CLEAR opc = %p, file = \"%s\"\n",
2131 			__func__, (void*)opc, perfuse_node_path(opc));
2132 #endif
2133 
2134 	ps->ps_destroy_msg(pm);
2135 
2136 out:
2137 	/*
2138 	 * ENOSYS is not returned to kernel,
2139 	 */
2140 	if (error == ENOSYS)
2141 		error = 0;
2142 
2143 	return error;
2144 }
2145 
2146 /* ARGSUSED0 */
2147 int
2148 perfuse_node_seek(pu, opc, oldoff, newoff,  pcr)
2149 	struct puffs_usermount *pu;
2150 	puffs_cookie_t opc;
2151 	off_t oldoff;
2152 	off_t newoff;
2153 	const struct puffs_cred *pcr;
2154 {
2155 	return 0;
2156 }
2157 
2158 int
2159 perfuse_node_remove(pu, opc, targ, pcn)
2160 	struct puffs_usermount *pu;
2161 	puffs_cookie_t opc;
2162 	puffs_cookie_t targ;
2163 	const struct puffs_cn *pcn;
2164 {
2165 	struct perfuse_state *ps;
2166 	struct perfuse_node_data *pnd;
2167 	perfuse_msg_t *pm;
2168 	char *path;
2169 	const char *name;
2170 	size_t len;
2171 	int error;
2172 
2173 	pnd = PERFUSE_NODE_DATA(opc);
2174 
2175 	if ((pnd->pnd_flags & PND_REMOVED) ||
2176 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
2177 		return ENOENT;
2178 
2179 #ifdef PERFUSE_DEBUG
2180 	if (targ == NULL)
2181 		DERRX(EX_SOFTWARE, "%s: targ is NULL", __func__);
2182 
2183 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
2184 		DPRINTF("%s: opc = %p, remove opc = %p, file = \"%s\"\n",
2185 			__func__, (void *)opc, (void *)targ, pcn->pcn_name);
2186 #endif
2187 	/*
2188 	 * Await for all operations on the deleted node to drain,
2189 	 * as the filesystem may be confused to have it deleted
2190 	 * during a getattr
2191 	 */
2192 	while (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_INXCHG)
2193 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2194 
2195 	ps = puffs_getspecific(pu);
2196 	pnd = PERFUSE_NODE_DATA(opc);
2197 	name = pcn->pcn_name;
2198 	len = pcn->pcn_namelen + 1;
2199 
2200 	pm = ps->ps_new_msg(pu, opc, FUSE_UNLINK, len, pcn->pcn_cred);
2201 	path = _GET_INPAYLOAD(ps, pm, char *);
2202 	(void)strlcpy(path, name, len);
2203 
2204 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2205 		return error;
2206 
2207 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2208 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2209 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2210 
2211 	/*
2212 	 * The parent directory needs a sync
2213 	 */
2214 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2215 
2216 #ifdef PERFUSE_DEBUG
2217 	if (perfuse_diagflags & PDF_FILENAME)
2218 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
2219 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
2220 			pcn->pcn_name);
2221 #endif
2222 	ps->ps_destroy_msg(pm);
2223 
2224 	return 0;
2225 }
2226 
2227 int
2228 perfuse_node_link(pu, opc, targ, pcn)
2229 	struct puffs_usermount *pu;
2230 	puffs_cookie_t opc;
2231 	puffs_cookie_t targ;
2232 	const struct puffs_cn *pcn;
2233 {
2234 	struct perfuse_state *ps;
2235 	perfuse_msg_t *pm;
2236 	const char *name;
2237 	size_t len;
2238 	struct puffs_node *pn;
2239 	struct fuse_link_in *fli;
2240 	int error;
2241 
2242 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2243 		return ENOENT;
2244 
2245 	ps = puffs_getspecific(pu);
2246 	pn = (struct puffs_node *)targ;
2247 	name = pcn->pcn_name;
2248 	len =  sizeof(*fli) + pcn->pcn_namelen + 1;
2249 
2250 	pm = ps->ps_new_msg(pu, opc, FUSE_LINK, len, pcn->pcn_cred);
2251 	fli = GET_INPAYLOAD(ps, pm, fuse_link_in);
2252 	fli->oldnodeid = PERFUSE_NODE_DATA(pn)->pnd_nodeid;
2253 	(void)strlcpy((char *)(void *)(fli + 1), name, len - sizeof(*fli));
2254 
2255 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2256 		return error;
2257 
2258 	ps->ps_destroy_msg(pm);
2259 
2260 	return 0;
2261 }
2262 
2263 int
2264 perfuse_node_rename(pu, opc, src, pcn_src, targ_dir, targ, pcn_targ)
2265 	struct puffs_usermount *pu;
2266 	puffs_cookie_t opc;
2267 	puffs_cookie_t src;
2268 	const struct puffs_cn *pcn_src;
2269 	puffs_cookie_t targ_dir;
2270 	puffs_cookie_t targ;
2271 	const struct puffs_cn *pcn_targ;
2272 {
2273 	struct perfuse_state *ps;
2274 	perfuse_msg_t *pm;
2275 	struct fuse_rename_in *fri;
2276 	const char *newname;
2277 	const char *oldname;
2278 	char *np;
2279 	int error;
2280 	size_t len;
2281 	size_t newname_len;
2282 	size_t oldname_len;
2283 
2284 	if ((PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED) ||
2285 	    (PERFUSE_NODE_DATA(src)->pnd_flags & PND_REMOVED) ||
2286 	    (PERFUSE_NODE_DATA(targ_dir)->pnd_flags & PND_REMOVED))
2287 		return ENOENT;
2288 
2289 	/*
2290 	 * Await for all operations on the deleted node to drain,
2291 	 * as the filesystem may be confused to have it deleted
2292 	 * during a getattr
2293 	 */
2294 	if ((struct puffs_node *)targ != NULL) {
2295 		while (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_INXCHG)
2296 			requeue_request(pu, targ, PCQ_AFTERXCHG);
2297 	} else {
2298 		while (PERFUSE_NODE_DATA(src)->pnd_flags & PND_INXCHG)
2299 			requeue_request(pu, src, PCQ_AFTERXCHG);
2300 	}
2301 
2302 	ps = puffs_getspecific(pu);
2303 	newname =  pcn_targ->pcn_name;
2304 	newname_len = pcn_targ->pcn_namelen + 1;
2305 	oldname =  pcn_src->pcn_name;
2306 	oldname_len = pcn_src->pcn_namelen + 1;
2307 
2308 	len = sizeof(*fri) + oldname_len + newname_len;
2309 	pm = ps->ps_new_msg(pu, opc, FUSE_RENAME, len, pcn_targ->pcn_cred);
2310 	fri = GET_INPAYLOAD(ps, pm, fuse_rename_in);
2311 	fri->newdir = PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid;
2312 	np = (char *)(void *)(fri + 1);
2313 	(void)strlcpy(np, oldname, oldname_len);
2314 	np += oldname_len;
2315 	(void)strlcpy(np, newname, newname_len);
2316 
2317 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2318 		return error;
2319 
2320 	if (opc != targ_dir) {
2321 		struct perfuse_node_data *srcdir_pnd;
2322 		struct perfuse_node_data *dstdir_pnd;
2323 		struct perfuse_node_data *src_pnd;
2324 
2325 		srcdir_pnd = PERFUSE_NODE_DATA(opc);
2326 		dstdir_pnd = PERFUSE_NODE_DATA(targ_dir);
2327 		src_pnd = PERFUSE_NODE_DATA(src);
2328 
2329 		TAILQ_REMOVE(&srcdir_pnd->pnd_children, src_pnd, pnd_next);
2330 		TAILQ_INSERT_TAIL(&dstdir_pnd->pnd_children, src_pnd, pnd_next);
2331 
2332 		srcdir_pnd->pnd_childcount--;
2333 		dstdir_pnd->pnd_childcount++;
2334 
2335 		src_pnd->pnd_parent = targ_dir;
2336 
2337 		PERFUSE_NODE_DATA(targ_dir)->pnd_flags |= PND_DIRTY;
2338 	}
2339 
2340 	(void)strlcpy(PERFUSE_NODE_DATA(src)->pnd_name, newname, MAXPATHLEN);
2341 
2342 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2343 
2344 	if ((struct puffs_node *)targ != NULL)
2345 		PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2346 
2347 #ifdef PERFUSE_DEBUG
2348 	if (perfuse_diagflags & PDF_FILENAME)
2349 		DPRINTF("%s: nodeid = 0x%"PRIx64" file = \"%s\" renamed \"%s\" "
2350 			"nodeid = 0x%"PRIx64" -> nodeid = 0x%"PRIx64" \"%s\"\n",
2351 	 		__func__, PERFUSE_NODE_DATA(src)->pnd_nodeid,
2352 			pcn_src->pcn_name, pcn_targ->pcn_name,
2353 			PERFUSE_NODE_DATA(opc)->pnd_nodeid,
2354 			PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid,
2355 			perfuse_node_path(targ_dir));
2356 #endif
2357 
2358 	ps->ps_destroy_msg(pm);
2359 
2360 	return 0;
2361 }
2362 
2363 int
2364 perfuse_node_mkdir(pu, opc, pni, pcn, vap)
2365 	struct puffs_usermount *pu;
2366 	puffs_cookie_t opc;
2367 	struct puffs_newinfo *pni;
2368 	const struct puffs_cn *pcn;
2369 	const struct vattr *vap;
2370 {
2371 	struct perfuse_state *ps;
2372 	perfuse_msg_t *pm;
2373 	struct fuse_mkdir_in *fmi;
2374 	const char *path;
2375 	size_t len;
2376 
2377 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2378 		return ENOENT;
2379 
2380 	ps = puffs_getspecific(pu);
2381 	path = pcn->pcn_name;
2382 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
2383 
2384 	pm = ps->ps_new_msg(pu, opc, FUSE_MKDIR, len, pcn->pcn_cred);
2385 	fmi = GET_INPAYLOAD(ps, pm, fuse_mkdir_in);
2386 	fmi->mode = vap->va_mode;
2387 	fmi->umask = 0; 	/* Seems unused by libfuse? */
2388 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
2389 
2390 	return node_mk_common(pu, opc, pni, pcn, pm);
2391 }
2392 
2393 
2394 int
2395 perfuse_node_rmdir(pu, opc, targ, pcn)
2396 	struct puffs_usermount *pu;
2397 	puffs_cookie_t opc;
2398 	puffs_cookie_t targ;
2399 	const struct puffs_cn *pcn;
2400 {
2401 	struct perfuse_state *ps;
2402 	struct perfuse_node_data *pnd;
2403 	perfuse_msg_t *pm;
2404 	char *path;
2405 	const char *name;
2406 	size_t len;
2407 	int error;
2408 
2409 	pnd = PERFUSE_NODE_DATA(opc);
2410 
2411 	if ((pnd->pnd_flags & PND_REMOVED) ||
2412 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
2413 		return ENOENT;
2414 
2415 	/*
2416 	 * Await for all operations on the deleted node to drain,
2417 	 * as the filesystem may be confused to have it deleted
2418 	 * during a getattr
2419 	 */
2420 	while (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_INXCHG)
2421 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2422 
2423 	ps = puffs_getspecific(pu);
2424 	name = pcn->pcn_name;
2425 	len = pcn->pcn_namelen + 1;
2426 
2427 	pm = ps->ps_new_msg(pu, opc, FUSE_RMDIR, len, pcn->pcn_cred);
2428 	path = _GET_INPAYLOAD(ps, pm, char *);
2429 	(void)strlcpy(path, name, len);
2430 
2431 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2432 		return error;
2433 
2434 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2435 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2436 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2437 
2438 	/*
2439 	 * The parent directory needs a sync
2440 	 */
2441 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2442 
2443 #ifdef PERFUSE_DEBUG
2444 	if (perfuse_diagflags & PDF_FILENAME)
2445 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
2446 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
2447 			perfuse_node_path(targ));
2448 #endif
2449 	ps->ps_destroy_msg(pm);
2450 
2451 	return 0;
2452 }
2453 
2454 /* vap is unused */
2455 /* ARGSUSED4 */
2456 int
2457 perfuse_node_symlink(pu, opc, pni, pcn_src, vap, link_target)
2458 	struct puffs_usermount *pu;
2459 	puffs_cookie_t opc;
2460 	struct puffs_newinfo *pni;
2461 	const struct puffs_cn *pcn_src;
2462 	const struct vattr *vap;
2463 	const char *link_target;
2464 {
2465 	struct perfuse_state *ps;
2466 	perfuse_msg_t *pm;
2467 	char *np;
2468 	const char *path;
2469 	size_t path_len;
2470 	size_t linkname_len;
2471 	size_t len;
2472 
2473 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2474 		return ENOENT;
2475 
2476 	ps = puffs_getspecific(pu);
2477 	path = pcn_src->pcn_name;
2478 	path_len = pcn_src->pcn_namelen + 1;
2479 	linkname_len = strlen(link_target) + 1;
2480 	len = path_len + linkname_len;
2481 
2482 	pm = ps->ps_new_msg(pu, opc, FUSE_SYMLINK, len, pcn_src->pcn_cred);
2483 	np = _GET_INPAYLOAD(ps, pm, char *);
2484 	(void)strlcpy(np, path, path_len);
2485 	np += path_len;
2486 	(void)strlcpy(np, link_target, linkname_len);
2487 
2488 	return node_mk_common(pu, opc, pni, pcn_src, pm);
2489 }
2490 
2491 /* ARGSUSED4 */
2492 int
2493 perfuse_node_readdir(pu, opc, dent, readoff,
2494 		     reslen, pcr, eofflag, cookies, ncookies)
2495 	struct puffs_usermount *pu;
2496 	puffs_cookie_t opc;
2497 	struct dirent *dent;
2498 	off_t *readoff;
2499 	size_t *reslen;
2500 	const struct puffs_cred *pcr;
2501 	int *eofflag;
2502 	off_t *cookies;
2503 	size_t *ncookies;
2504 {
2505 	perfuse_msg_t *pm;
2506 	uint64_t fh;
2507 	struct perfuse_state *ps;
2508 	struct perfuse_node_data *pnd;
2509 	struct fuse_read_in *fri;
2510 	struct fuse_out_header *foh;
2511 	struct fuse_dirent *fd;
2512 	size_t foh_len;
2513 	int error;
2514 	size_t fd_maxlen;
2515 
2516 	error = 0;
2517 	ps = puffs_getspecific(pu);
2518 
2519 	/*
2520 	 * readdir state is kept at node level, and several readdir
2521 	 * requests can be issued at the same time on the same node.
2522 	 * We need to queue requests so that only one is in readdir
2523 	 * code at the same time.
2524 	 */
2525 	pnd = PERFUSE_NODE_DATA(opc);
2526 	while (pnd->pnd_flags & PND_INREADDIR)
2527 		requeue_request(pu, opc, PCQ_READDIR);
2528 	pnd->pnd_flags |= PND_INREADDIR;
2529 
2530 #ifdef PERFUSE_DEBUG
2531 	if (perfuse_diagflags & PDF_READDIR)
2532 		DPRINTF("%s: READDIR opc = %p enter critical section\n",
2533 			__func__, (void *)opc);
2534 #endif
2535 	/*
2536 	 * Re-initialize pnd->pnd_fd_cookie on the first readdir for a node
2537 	 */
2538 	if (*readoff == 0)
2539 		pnd->pnd_fd_cookie = 0;
2540 
2541 	/*
2542 	 * Do we already have the data bufered?
2543 	 */
2544 	if (pnd->pnd_dirent != NULL)
2545 		goto out;
2546 	pnd->pnd_dirent_len = 0;
2547 
2548 	/*
2549 	 * It seems NetBSD can call readdir without open first
2550 	 * libfuse will crash if it is done that way, hence open first.
2551 	 */
2552 	if (!(pnd->pnd_flags & PND_OPEN)) {
2553 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2554 			goto out;
2555 	}
2556 
2557 	fh = perfuse_get_fh(opc, FREAD);
2558 
2559 #ifdef PERFUSE_DEBUG
2560 	if (perfuse_diagflags & PDF_FH)
2561 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
2562 			"rfh = 0x%"PRIx64"\n", __func__, (void *)opc,
2563 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fh);
2564 #endif
2565 
2566 	pnd->pnd_all_fd = NULL;
2567 	pnd->pnd_all_fd_len = 0;
2568 	fd_maxlen = ps->ps_max_readahead - sizeof(*foh);
2569 
2570 	do {
2571 		size_t fd_len;
2572 		char *afdp;
2573 
2574 		pm = ps->ps_new_msg(pu, opc, FUSE_READDIR, sizeof(*fri), pcr);
2575 
2576 		/*
2577 		 * read_flags, lock_owner and flags are unused in libfuse
2578 		 */
2579 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
2580 		fri->fh = fh;
2581 		fri->offset = pnd->pnd_fd_cookie;
2582 		fri->size = (uint32_t)fd_maxlen;
2583 		fri->read_flags = 0;
2584 		fri->lock_owner = 0;
2585 		fri->flags = 0;
2586 
2587 		if ((error = xchg_msg(pu, opc, pm,
2588 				      UNSPEC_REPLY_LEN, wait_reply)) != 0)
2589 			goto out;
2590 
2591 		/*
2592 		 * There are many puffs_framebufs calls later,
2593 		 * therefore foh will not be valid for a long time.
2594 		 * Just get the length and forget it.
2595 		 */
2596 		foh = GET_OUTHDR(ps, pm);
2597 		foh_len = foh->len;
2598 
2599 		/*
2600 		 * Empty read: we reached the end of the buffer.
2601 		 */
2602 		if (foh_len == sizeof(*foh)) {
2603 			ps->ps_destroy_msg(pm);
2604 			*eofflag = 1;
2605 			break;
2606 		}
2607 
2608 		/*
2609 		 * Check for corrupted message.
2610 		 */
2611 		if (foh_len < sizeof(*foh) + sizeof(*fd)) {
2612 			ps->ps_destroy_msg(pm);
2613 			DWARNX("readdir reply too short");
2614 			error = EIO;
2615 			goto out;
2616 		}
2617 
2618 
2619 		fd = GET_OUTPAYLOAD(ps, pm, fuse_dirent);
2620 		fd_len = foh_len - sizeof(*foh);
2621 
2622 		pnd->pnd_all_fd = realloc(pnd->pnd_all_fd,
2623 					  pnd->pnd_all_fd_len + fd_len);
2624 		if (pnd->pnd_all_fd  == NULL)
2625 			DERR(EX_OSERR, "%s: malloc failed", __func__);
2626 
2627 		afdp = (char *)(void *)pnd->pnd_all_fd + pnd->pnd_all_fd_len;
2628 		(void)memcpy(afdp, fd, fd_len);
2629 
2630 		pnd->pnd_all_fd_len += fd_len;
2631 
2632 		/*
2633 		 * The fd->off field is used as a cookie for
2634 		 * resuming the next readdir() where this one was left.
2635 	 	 */
2636 		pnd->pnd_fd_cookie = readdir_last_cookie(fd, fd_len);
2637 
2638 		ps->ps_destroy_msg(pm);
2639 	} while (1 /* CONSTCOND */);
2640 
2641 	if (pnd->pnd_all_fd != NULL) {
2642 		if (fuse_to_dirent(pu, opc, pnd->pnd_all_fd,
2643 				   pnd->pnd_all_fd_len) == -1)
2644 			error = EIO;
2645 	}
2646 
2647 out:
2648 	if (pnd->pnd_all_fd != NULL) {
2649 		free(pnd->pnd_all_fd);
2650 		pnd->pnd_all_fd = NULL;
2651 		pnd->pnd_all_fd_len = 0;
2652 	}
2653 
2654 	if (error == 0)
2655 		error = readdir_buffered(opc, dent, readoff, reslen);
2656 
2657 	/*
2658 	 * Schedule queued readdir requests
2659 	 */
2660 	pnd->pnd_flags &= ~PND_INREADDIR;
2661 	(void)dequeue_requests(ps, opc, PCQ_READDIR, DEQUEUE_ALL);
2662 
2663 #ifdef PERFUSE_DEBUG
2664 	if (perfuse_diagflags & PDF_READDIR)
2665 		DPRINTF("%s: READDIR opc = %p exit critical section\n",
2666 			__func__, (void *)opc);
2667 #endif
2668 
2669 	return error;
2670 }
2671 
2672 int
2673 perfuse_node_readlink(pu, opc, pcr, linkname, linklen)
2674 	struct puffs_usermount *pu;
2675 	puffs_cookie_t opc;
2676 	const struct puffs_cred *pcr;
2677 	char *linkname;
2678 	size_t *linklen;
2679 {
2680 	struct perfuse_state *ps;
2681 	perfuse_msg_t *pm;
2682 	int error;
2683 	size_t len;
2684 	struct fuse_out_header *foh;
2685 
2686 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2687 		return ENOENT;
2688 
2689 	ps = puffs_getspecific(pu);
2690 
2691 	pm = ps->ps_new_msg(pu, opc, FUSE_READLINK, 0, pcr);
2692 
2693 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2694 		return error;
2695 
2696 	foh = GET_OUTHDR(ps, pm);
2697 	len = foh->len - sizeof(*foh);
2698 	if (len > *linklen)
2699 		DERRX(EX_PROTOCOL, "path len = %zd too long", len);
2700 	if (len == 0)
2701 		DERRX(EX_PROTOCOL, "path len = %zd too short", len);
2702 
2703 	/*
2704 	 * FUSE filesystems return a NUL terminated string, we
2705 	 * do not want to trailing \0
2706 	 */
2707 	*linklen = len - 1;
2708 	(void)memcpy(linkname, _GET_OUTPAYLOAD(ps, pm, char *), len);
2709 
2710 	ps->ps_destroy_msg(pm);
2711 
2712 	return 0;
2713 }
2714 
2715 int
2716 perfuse_node_reclaim(pu, opc)
2717 	struct puffs_usermount *pu;
2718 	puffs_cookie_t opc;
2719 {
2720 	struct perfuse_state *ps;
2721 	perfuse_msg_t *pm;
2722 	struct perfuse_node_data *pnd;
2723 	struct fuse_forget_in *ffi;
2724 	struct puffs_node *pn;
2725 	struct puffs_node *pn_root;
2726 
2727 	ps = puffs_getspecific(pu);
2728 	pnd = PERFUSE_NODE_DATA(opc);
2729 
2730 	/*
2731 	 * Never forget the root.
2732 	 */
2733 	if (pnd->pnd_nodeid == FUSE_ROOT_ID)
2734 		return 0;
2735 
2736 	pnd->pnd_flags |= PND_RECLAIMED;
2737 
2738 #ifdef PERFUSE_DEBUG
2739 	if (perfuse_diagflags & PDF_RECLAIM)
2740 		DPRINTF("%s (nodeid %"PRId64") reclaimed\n",
2741 			perfuse_node_path(opc), pnd->pnd_nodeid);
2742 #endif
2743 
2744 	pn_root = puffs_getroot(pu);
2745 	pn = (struct puffs_node *)opc;
2746 	while (pn != pn_root) {
2747 		struct puffs_node *parent_pn;
2748 
2749 		pnd = PERFUSE_NODE_DATA(pn);
2750 
2751 #ifdef PERFUSE_DEBUG
2752 	if (perfuse_diagflags & PDF_RECLAIM)
2753 		DPRINTF("%s (nodeid %"PRId64") is %sreclaimed, "
2754 			"has childcount %d %s%s%s%s, pending ops:%s%s%s\n",
2755 		        perfuse_node_path((puffs_cookie_t)pn), pnd->pnd_nodeid,
2756 		        pnd->pnd_flags & PND_RECLAIMED ? "" : "not ",
2757 		        pnd->pnd_childcount,
2758 			pnd->pnd_flags & PND_OPEN ? "open " : "not open",
2759 			pnd->pnd_flags & PND_RFH ? "r" : "",
2760 			pnd->pnd_flags & PND_WFH ? "w" : "",
2761 			pnd->pnd_flags & PND_BUSY ? "" : " none",
2762 			pnd->pnd_flags & PND_INREADDIR ? " readdir" : "",
2763 			pnd->pnd_flags & PND_INWRITE ? " write" : "",
2764 			pnd->pnd_flags & PND_INOPEN ? " open" : "");
2765 #endif
2766 
2767 		if (!(pnd->pnd_flags & PND_RECLAIMED) ||
2768 		    (pnd->pnd_childcount != 0))
2769 			return 0;
2770 
2771 #ifdef PERFUSE_DEBUG
2772 		if ((pnd->pnd_flags & PND_OPEN) ||
2773 		       !TAILQ_EMPTY(&pnd->pnd_pcq))
2774 			DERRX(EX_SOFTWARE, "%s: opc = %p: still open",
2775 			      __func__, (void *)opc);
2776 
2777 		if ((pnd->pnd_flags & PND_BUSY) ||
2778 		       !TAILQ_EMPTY(&pnd->pnd_pcq))
2779 			DERRX(EX_SOFTWARE, "%s: opc = %p: ongoing operations",
2780 			      __func__, (void *)opc);
2781 #endif
2782 
2783 		/*
2784 		 * Send the FORGET message
2785 		 *
2786 		 * ps_new_msg() is called with NULL creds, which will
2787 		 * be interpreted as FUSE superuser. This is obviously
2788 		 * fine since we operate with kernel creds here.
2789 		 */
2790 		pm = ps->ps_new_msg(pu, (puffs_cookie_t)pn, FUSE_FORGET,
2791 			      sizeof(*ffi), NULL);
2792 		ffi = GET_INPAYLOAD(ps, pm, fuse_forget_in);
2793 		ffi->nlookup = pnd->pnd_nlookup;
2794 
2795 		/*
2796 		 * No reply is expected, pm is freed in xchg_msg
2797 		 */
2798 		(void)xchg_msg(pu, (puffs_cookie_t)pn,
2799 			       pm, UNSPEC_REPLY_LEN, no_reply);
2800 
2801 		parent_pn = pnd->pnd_parent;
2802 
2803 		perfuse_destroy_pn(pn);
2804 
2805 		pn = parent_pn;
2806 	}
2807 
2808 	return 0;
2809 }
2810 
2811 int
2812 perfuse_node_inactive(pu, opc)
2813 	struct puffs_usermount *pu;
2814 	puffs_cookie_t opc;
2815 {
2816 	struct perfuse_state *ps;
2817 	struct perfuse_node_data *pnd;
2818 
2819 	ps = puffs_getspecific(pu);
2820 	pnd = PERFUSE_NODE_DATA(opc);
2821 
2822 	if (!(pnd->pnd_flags & (PND_OPEN|PND_REMOVED)))
2823 		return 0;
2824 
2825 	/*
2826 	 * Make sure all operation are finished
2827 	 * There can be an ongoing write. Other
2828 	 * operation wait for all data before
2829 	 * the close/inactive.
2830 	 */
2831 	while (pnd->pnd_flags & PND_INWRITE)
2832 		requeue_request(pu, opc, PCQ_AFTERWRITE);
2833 
2834 	/*
2835 	 * The inactive operation may be cancelled.
2836 	 * If no open is in progress, set PND_INOPEN
2837 	 * so that a new open will be queued.
2838 	 */
2839 	if (pnd->pnd_flags & PND_INOPEN)
2840 		return 0;
2841 
2842 	pnd->pnd_flags |= PND_INOPEN;
2843 
2844 	/*
2845 	 * Sync data
2846 	 */
2847 	if (pnd->pnd_flags & PND_DIRTY)
2848 		(void)perfuse_node_fsync(pu, opc, NULL, 0, 0, 0);
2849 
2850 	/*
2851 	 * Close handles
2852 	 */
2853 	if (pnd->pnd_flags & PND_WFH)
2854 		(void)perfuse_node_close_common(pu, opc, FWRITE);
2855 
2856 	if (pnd->pnd_flags & PND_RFH)
2857 		(void)perfuse_node_close_common(pu, opc, FREAD);
2858 
2859 	/*
2860 	 * This will cause a reclaim to be sent
2861 	 */
2862 	if (pnd->pnd_flags & PND_REMOVED)
2863 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N1);
2864 
2865 	/*
2866 	 * Schedule awaiting operations
2867 	 */
2868 	pnd->pnd_flags &= ~PND_INOPEN;
2869 	(void)dequeue_requests(ps, opc, PCQ_OPEN, DEQUEUE_ALL);
2870 
2871 	return 0;
2872 }
2873 
2874 
2875 /* ARGSUSED0 */
2876 int
2877 perfuse_node_print(pu, opc)
2878 	struct puffs_usermount *pu;
2879 	puffs_cookie_t opc;
2880 {
2881 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2882 	return 0;
2883 }
2884 
2885 /* ARGSUSED0 */
2886 int
2887 perfuse_node_pathconf(pu, opc, name, retval)
2888 	struct puffs_usermount *pu;
2889 	puffs_cookie_t opc;
2890 	int name;
2891 	int *retval;
2892 {
2893 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2894 	return 0;
2895 }
2896 
2897 int
2898 perfuse_node_advlock(pu, opc, id, op, fl, flags)
2899 	struct puffs_usermount *pu;
2900 	puffs_cookie_t opc;
2901 	void *id;
2902 	int op;
2903 	struct flock *fl;
2904 	int flags;
2905 {
2906 	struct perfuse_state *ps;
2907 	int fop;
2908 	perfuse_msg_t *pm;
2909 	uint64_t fh;
2910 	struct fuse_lk_in *fli;
2911 	struct fuse_out_header *foh;
2912 	struct fuse_lk_out *flo;
2913 	uint32_t owner;
2914 	size_t len;
2915 	int error;
2916 
2917 	/*
2918 	 * Make sure we do have a filehandle, as the FUSE filesystem
2919 	 * expect one. E.g.: if we provide none, GlusterFS logs an error
2920 	 * "0-glusterfs-fuse: xl is NULL"
2921 	 *
2922 	 * We need the read file handle if the file is open read only,
2923 	 * in order to support shared locks on read-only files.
2924 	 * NB: The kernel always sends advlock for read-only
2925 	 * files at exit time when the process used lock, see
2926 	 * sys_exit -> exit1 -> fd_free -> fd_close -> VOP_ADVLOCK
2927 	 */
2928 	if ((fh = perfuse_get_fh(opc, FREAD)) == FUSE_UNKNOWN_FH)
2929 		return EBADF;
2930 
2931 	ps = puffs_getspecific(pu);
2932 
2933 	if (op == F_GETLK)
2934 		fop = FUSE_GETLK;
2935 	else
2936 		fop = (flags & F_WAIT) ? FUSE_SETLKW : FUSE_SETLK;
2937 
2938 	/*
2939 	 * XXX ps_new_msg() is called with NULL creds, which will
2940 	 * be interpreted as FUSE superuser. We have no way to
2941 	 * know the requesting process' credential, but since advlock()
2942 	 * is supposed to operate on a file that has been open(),
2943 	 * permission should have already been checked at open() time.
2944 	 */
2945 	pm = ps->ps_new_msg(pu, opc, fop, sizeof(*fli), NULL);
2946 	fli = GET_INPAYLOAD(ps, pm, fuse_lk_in);
2947 	fli->fh = fh;
2948 	fli->owner = (uint64_t)(vaddr_t)id;
2949 	fli->lk.start = fl->l_start;
2950 	fli->lk.end = fl->l_start + fl->l_len;
2951 	fli->lk.type = fl->l_type;
2952 	fli->lk.pid = fl->l_pid;
2953 	fli->lk_flags = (flags & F_FLOCK) ? FUSE_LK_FLOCK : 0;
2954 
2955 	owner = (uint32_t)(vaddr_t)id;
2956 
2957 #ifdef PERFUSE_DEBUG
2958 	if (perfuse_diagflags & PDF_FH)
2959 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
2960 			__func__, (void *)opc,
2961 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fli->fh);
2962 #endif
2963 
2964 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2965 		return error;
2966 
2967 	foh = GET_OUTHDR(ps, pm);
2968 	len = foh->len - sizeof(*foh);
2969 
2970 	/*
2971 	 * Save or clear the lock
2972 	 */
2973 	switch (op) {
2974 	case F_GETLK:
2975 		if (len != sizeof(*flo))
2976 			DERRX(EX_SOFTWARE,
2977 			      "%s: Unexpected lock reply len %zd",
2978 			      __func__, len);
2979 
2980 		flo = GET_OUTPAYLOAD(ps, pm, fuse_lk_out);
2981 		fl->l_start = flo->lk.start;
2982 		fl->l_len = flo->lk.end - flo->lk.start;
2983 		fl->l_pid = flo->lk.pid;
2984 		fl->l_type = flo->lk.type;
2985 		fl->l_whence = SEEK_SET;	/* libfuse hardcodes it */
2986 
2987 		PERFUSE_NODE_DATA(opc)->pnd_lock_owner = flo->lk.pid;
2988 		break;
2989 	case F_UNLCK:
2990 		owner = 0;
2991 		/* FALLTHROUGH */
2992 	case F_SETLK:
2993 		/* FALLTHROUGH */
2994 	case F_SETLKW:
2995 		if (error != 0)
2996 			PERFUSE_NODE_DATA(opc)->pnd_lock_owner = owner;
2997 
2998 		if (len != 0)
2999 			DERRX(EX_SOFTWARE,
3000 			      "%s: Unexpected unlock reply len %zd",
3001 			      __func__, len);
3002 
3003 		break;
3004 	default:
3005 		DERRX(EX_SOFTWARE, "%s: Unexpected op %d", __func__, op);
3006 		break;
3007 	}
3008 
3009 	ps->ps_destroy_msg(pm);
3010 
3011 	return 0;
3012 }
3013 
3014 int
3015 perfuse_node_read(pu, opc, buf, offset, resid, pcr, ioflag)
3016 	struct puffs_usermount *pu;
3017 	puffs_cookie_t opc;
3018 	uint8_t *buf;
3019 	off_t offset;
3020 	size_t *resid;
3021 	const struct puffs_cred *pcr;
3022 	int ioflag;
3023 {
3024 	struct perfuse_state *ps;
3025 	struct perfuse_node_data *pnd;
3026 	const struct vattr *vap;
3027 	perfuse_msg_t *pm;
3028 	struct fuse_read_in *fri;
3029 	struct fuse_out_header *foh;
3030 	size_t readen;
3031 	int error;
3032 
3033 	ps = puffs_getspecific(pu);
3034 	pnd = PERFUSE_NODE_DATA(opc);
3035 	vap = puffs_pn_getvap((struct puffs_node *)opc);
3036 
3037 	/*
3038 	 * NetBSD turns that into a getdents(2) output
3039 	 * We just do a EISDIR as this feature is of little use.
3040 	 */
3041 	if (vap->va_type == VDIR)
3042 		return EISDIR;
3043 
3044 	if ((u_quad_t)offset + *resid > vap->va_size)
3045 		DWARNX("%s %p read %lld@%zu beyond EOF %" PRIu64 "\n",
3046 		       __func__, (void *)opc, (long long)offset,
3047 		       *resid, vap->va_size);
3048 
3049 	do {
3050 		size_t max_read;
3051 
3052 		max_read = ps->ps_max_readahead - sizeof(*foh);
3053 		/*
3054 		 * flags may be set to FUSE_READ_LOCKOWNER
3055 		 * if lock_owner is provided.
3056 		 */
3057 		pm = ps->ps_new_msg(pu, opc, FUSE_READ, sizeof(*fri), pcr);
3058 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
3059 		fri->fh = perfuse_get_fh(opc, FREAD);
3060 		fri->offset = offset;
3061 		fri->size = (uint32_t)MIN(*resid, max_read);
3062 		fri->read_flags = 0; /* XXX Unused by libfuse? */
3063 		fri->lock_owner = pnd->pnd_lock_owner;
3064 		fri->flags = 0;
3065 		fri->flags |= (fri->lock_owner != 0) ? FUSE_READ_LOCKOWNER : 0;
3066 
3067 #ifdef PERFUSE_DEBUG
3068 	if (perfuse_diagflags & PDF_FH)
3069 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
3070 			__func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
3071 #endif
3072 		error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
3073 		if (error  != 0)
3074 			return error;
3075 
3076 		foh = GET_OUTHDR(ps, pm);
3077 		readen = foh->len - sizeof(*foh);
3078 
3079 #ifdef PERFUSE_DEBUG
3080 		if (readen > *resid)
3081 			DERRX(EX_SOFTWARE, "%s: Unexpected big read %zd",
3082 			      __func__, readen);
3083 #endif
3084 
3085 		(void)memcpy(buf,  _GET_OUTPAYLOAD(ps, pm, char *), readen);
3086 
3087 		buf += readen;
3088 		offset += readen;
3089 		*resid -= readen;
3090 
3091 		ps->ps_destroy_msg(pm);
3092 	} while ((*resid != 0) && (readen != 0));
3093 
3094 	if (ioflag & (IO_SYNC|IO_DSYNC))
3095 		ps->ps_syncreads++;
3096 	else
3097 		ps->ps_asyncreads++;
3098 
3099 	return 0;
3100 }
3101 
3102 int
3103 perfuse_node_write(pu, opc, buf, offset, resid, pcr, ioflag)
3104 	struct puffs_usermount *pu;
3105 	puffs_cookie_t opc;
3106 	uint8_t *buf;
3107 	off_t offset;
3108 	size_t *resid;
3109 	const struct puffs_cred *pcr;
3110 	int ioflag;
3111 {
3112 	struct perfuse_state *ps;
3113 	struct perfuse_node_data *pnd;
3114 	struct vattr *vap;
3115 	perfuse_msg_t *pm;
3116 	struct fuse_write_in *fwi;
3117 	struct fuse_write_out *fwo;
3118 	size_t data_len;
3119 	size_t payload_len;
3120 	size_t written;
3121 	int inresize;
3122 	int error;
3123 
3124 	ps = puffs_getspecific(pu);
3125 	pnd = PERFUSE_NODE_DATA(opc);
3126 	vap = puffs_pn_getvap((struct puffs_node *)opc);
3127 	written = 0;
3128 	inresize = 0;
3129 	error = 0;
3130 
3131 	if (vap->va_type == VDIR)
3132 		return EISDIR;
3133 
3134 	/*
3135 	 * We need to queue write requests in order to avoid
3136 	 * dequeueing PCQ_AFTERWRITE when there are pending writes.
3137 	 */
3138 	while (pnd->pnd_flags & PND_INWRITE)
3139 		requeue_request(pu, opc, PCQ_WRITE);
3140 	pnd->pnd_flags |= PND_INWRITE;
3141 
3142 	/*
3143 	 * Serialize size access, see comment in perfuse_node_setattr().
3144 	 */
3145 	if ((u_quad_t)offset + *resid > vap->va_size) {
3146 		while (pnd->pnd_flags & PND_INRESIZE)
3147 			requeue_request(pu, opc, PCQ_RESIZE);
3148 		pnd->pnd_flags |= PND_INRESIZE;
3149 		inresize = 1;
3150 	}
3151 
3152 	/*
3153 	 * append flag: re-read the file size so that
3154 	 * we get the latest value.
3155 	 */
3156 	if (ioflag & PUFFS_IO_APPEND) {
3157 		DWARNX("%s: PUFFS_IO_APPEND set, untested code", __func__);
3158 
3159 		if ((error = perfuse_node_getattr(pu, opc, vap, pcr)) != 0)
3160 			goto out;
3161 
3162 		offset = vap->va_size;
3163 	}
3164 
3165 #ifdef PERFUSE_DEBUG
3166 	if (perfuse_diagflags & PDF_RESIZE)
3167 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__,
3168 			(void *)opc, vap->va_size);
3169 #endif
3170 
3171 	do {
3172 		size_t max_write;
3173 		/*
3174 		 * There is a writepage flag when data
3175 		 * is aligned to page size. Use it for
3176 		 * everything but the data after the last
3177 		 * page boundary.
3178 		 */
3179 		max_write = ps->ps_max_write - sizeof(*fwi);
3180 
3181 		data_len = MIN(*resid, max_write);
3182 		if (data_len > (size_t)sysconf(_SC_PAGESIZE))
3183 			data_len = data_len & ~(sysconf(_SC_PAGESIZE) - 1);
3184 
3185 		payload_len = data_len + sizeof(*fwi);
3186 
3187 		/*
3188 		 * flags may be set to FUSE_WRITE_CACHE (XXX usage?)
3189 		 * or FUSE_WRITE_LOCKOWNER, if lock_owner is provided.
3190 		 * write_flags is set to 1 for writepage.
3191 		 */
3192 		pm = ps->ps_new_msg(pu, opc, FUSE_WRITE, payload_len, pcr);
3193 		fwi = GET_INPAYLOAD(ps, pm, fuse_write_in);
3194 		fwi->fh = perfuse_get_fh(opc, FWRITE);
3195 		fwi->offset = offset;
3196 		fwi->size = (uint32_t)data_len;
3197 		fwi->write_flags = (fwi->size % sysconf(_SC_PAGESIZE)) ? 0 : 1;
3198 		fwi->lock_owner = pnd->pnd_lock_owner;
3199 		fwi->flags = 0;
3200 		fwi->flags |= (fwi->lock_owner != 0) ? FUSE_WRITE_LOCKOWNER : 0;
3201 		fwi->flags |= (ioflag & IO_DIRECT) ? 0 : FUSE_WRITE_CACHE;
3202 		(void)memcpy((fwi + 1), buf, data_len);
3203 
3204 
3205 #ifdef PERFUSE_DEBUG
3206 		if (perfuse_diagflags & PDF_FH)
3207 			DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
3208 				"fh = 0x%"PRIx64"\n", __func__,
3209 				(void *)opc, pnd->pnd_nodeid, fwi->fh);
3210 #endif
3211 		if ((error = xchg_msg(pu, opc, pm,
3212 				      sizeof(*fwo), wait_reply)) != 0)
3213 			goto out;
3214 
3215 		fwo = GET_OUTPAYLOAD(ps, pm, fuse_write_out);
3216 		written = fwo->size;
3217 #ifdef PERFUSE_DEBUG
3218 		if (written > *resid)
3219 			DERRX(EX_SOFTWARE, "%s: Unexpected big write %zd",
3220 			      __func__, written);
3221 #endif
3222 		*resid -= written;
3223 		offset += written;
3224 		buf += written;
3225 
3226 		ps->ps_destroy_msg(pm);
3227 	} while (*resid != 0);
3228 
3229 	/*
3230 	 * puffs_ops(3) says
3231 	 *  "everything must be written or an error will be generated"
3232 	 */
3233 	if (*resid != 0)
3234 		error = EFBIG;
3235 
3236 #ifdef PERFUSE_DEBUG
3237 	if (perfuse_diagflags & PDF_RESIZE) {
3238 		if (offset > (off_t)vap->va_size)
3239 			DPRINTF("<< %s %p %" PRIu64 " -> %lld\n", __func__,
3240 				(void *)opc, vap->va_size, (long long)offset);
3241 		else
3242 			DPRINTF("<< %s %p \n", __func__, (void *)opc);
3243 	}
3244 #endif
3245 
3246 	/*
3247 	 * Update file size if we wrote beyond the end
3248 	 */
3249 	if (offset > (off_t)vap->va_size)
3250 		vap->va_size = offset;
3251 
3252 	if (inresize) {
3253 #ifdef PERFUSE_DEBUG
3254 		if (!(pnd->pnd_flags & PND_INRESIZE))
3255 			DERRX(EX_SOFTWARE, "file write grow without resize");
3256 #endif
3257 		pnd->pnd_flags &= ~PND_INRESIZE;
3258 		(void)dequeue_requests(ps, opc, PCQ_RESIZE, DEQUEUE_ALL);
3259 	}
3260 
3261 
3262 	/*
3263 	 * Statistics
3264 	 */
3265 	if (ioflag & (IO_SYNC|IO_DSYNC))
3266 		ps->ps_syncwrites++;
3267 	else
3268 		ps->ps_asyncwrites++;
3269 
3270 	/*
3271 	 * Remember to sync the file
3272 	 */
3273 	pnd->pnd_flags |= PND_DIRTY;
3274 
3275 #ifdef PERFUSE_DEBUG
3276 	if (perfuse_diagflags & PDF_SYNC)
3277 		DPRINTF("%s: DIRTY opc = %p, file = \"%s\"\n",
3278 			__func__, (void*)opc, perfuse_node_path(opc));
3279 #endif
3280 
3281 out:
3282 	/*
3283 	 * If there are no more queued write, we can resume
3284 	 * an operation awaiting write completion.
3285 	 */
3286 	pnd->pnd_flags &= ~PND_INWRITE;
3287 	if (dequeue_requests(ps, opc, PCQ_WRITE, 1) == 0)
3288 		(void)dequeue_requests(ps, opc, PCQ_AFTERWRITE, DEQUEUE_ALL);
3289 
3290 	return error;
3291 }
3292 
3293 /* ARGSUSED0 */
3294 void
3295 perfuse_cache_write(pu, opc, size, runs)
3296 	struct puffs_usermount *pu;
3297 	puffs_cookie_t opc;
3298 	size_t size;
3299 	struct puffs_cacherun *runs;
3300 {
3301 	return;
3302 }
3303 
3304 /* ARGSUSED4 */
3305 int
3306 perfuse_node_getextattr(pu, opc, attrns, attrname, attrsize, attr, resid, pcr)
3307 	struct puffs_usermount *pu;
3308 	puffs_cookie_t opc;
3309 	int attrns;
3310 	const char *attrname;
3311 	size_t *attrsize;
3312 	uint8_t *attr;
3313 	size_t *resid;
3314 	const struct puffs_cred *pcr;
3315 {
3316 	struct perfuse_state *ps;
3317 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3318 	perfuse_msg_t *pm;
3319 	struct fuse_getxattr_in *fgi;
3320 	struct fuse_getxattr_out *fgo;
3321 	struct fuse_out_header *foh;
3322 	size_t attrnamelen;
3323 	size_t len;
3324 	char *np;
3325 	int error;
3326 
3327 	ps = puffs_getspecific(pu);
3328 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3329 	attrnamelen = strlen(attrname) + 1;
3330 	len = sizeof(*fgi) + attrnamelen;
3331 
3332 	pm = ps->ps_new_msg(pu, opc, FUSE_GETXATTR, len, pcr);
3333 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3334 	fgi->size = (unsigned int)((resid != NULL) ? *resid : 0);
3335 	np = (char *)(void *)(fgi + 1);
3336 	(void)strlcpy(np, attrname, attrnamelen);
3337 
3338 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3339 		return error;
3340 
3341 	/*
3342 	 * We just get fuse_getattr_out with list size if we requested
3343 	 * a null size.
3344 	 */
3345 	if (resid == NULL) {
3346 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3347 
3348 		if (attrsize != NULL)
3349 			*attrsize = fgo->size;
3350 
3351 		ps->ps_destroy_msg(pm);
3352 		return 0;
3353 	}
3354 
3355 	/*
3356 	 * And with a non null requested size, we get the list just
3357 	 * after the header
3358 	 */
3359 	foh = GET_OUTHDR(ps, pm);
3360 	np = (char *)(void *)(foh + 1);
3361 
3362 	if (resid != NULL) {
3363 		len = MAX(foh->len - sizeof(*foh), *resid);
3364 		(void)memcpy(attr, np, len);
3365 		*resid -= len;
3366 	}
3367 
3368 	ps->ps_destroy_msg(pm);
3369 
3370 	return 0;
3371 }
3372 
3373 int
3374 perfuse_node_setextattr(pu, opc, attrns, attrname, attr, resid, pcr)
3375 	struct puffs_usermount *pu;
3376 	puffs_cookie_t opc;
3377 	int attrns;
3378 	const char *attrname;
3379 	uint8_t *attr;
3380 	size_t *resid;
3381 	const struct puffs_cred *pcr;
3382 {
3383 	struct perfuse_state *ps;
3384 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3385 	perfuse_msg_t *pm;
3386 	struct fuse_setxattr_in *fsi;
3387 	size_t attrnamelen;
3388 	size_t len;
3389 	char *np;
3390 	int error;
3391 
3392 	ps = puffs_getspecific(pu);
3393 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3394 	attrnamelen = strlen(attrname) + 1;
3395 	len = sizeof(*fsi) + attrnamelen + *resid;
3396 
3397 	pm = ps->ps_new_msg(pu, opc, FUSE_SETXATTR, len, pcr);
3398 	fsi = GET_INPAYLOAD(ps, pm, fuse_setxattr_in);
3399 	fsi->size = (unsigned int)*resid;
3400 	fsi->flags = 0;
3401 	np = (char *)(void *)(fsi + 1);
3402 	(void)strlcpy(np, attrname, attrnamelen);
3403 	np += attrnamelen;
3404 	(void)memcpy(np, (char *)attr, *resid);
3405 
3406 	if ((error = xchg_msg(pu, opc, pm,
3407 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
3408 		return error;
3409 
3410 	*resid = 0;
3411 	ps->ps_destroy_msg(pm);
3412 
3413 	return 0;
3414 }
3415 
3416 /* ARGSUSED2 */
3417 int
3418 perfuse_node_listextattr(pu, opc, attrns, attrsize, attrs, resid, flag, pcr)
3419 	struct puffs_usermount *pu;
3420 	puffs_cookie_t opc;
3421 	int attrns;
3422 	size_t *attrsize;
3423 	uint8_t *attrs;
3424 	size_t *resid;
3425 	int flag;
3426 	const struct puffs_cred *pcr;
3427 {
3428 	struct perfuse_state *ps;
3429 	perfuse_msg_t *pm;
3430 	struct fuse_getxattr_in *fgi;
3431 	struct fuse_getxattr_out *fgo;
3432 	struct fuse_out_header *foh;
3433 	char *np;
3434 	size_t len, puffs_len;
3435 	int error;
3436 
3437 	ps = puffs_getspecific(pu);
3438 	len = sizeof(*fgi);
3439 
3440 	pm = ps->ps_new_msg(pu, opc, FUSE_LISTXATTR, len, pcr);
3441 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3442 	if (resid != NULL)
3443 		fgi->size = (unsigned int)*resid;
3444 	else
3445 		fgi->size = 0;
3446 
3447 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3448 		return error;
3449 
3450 	/*
3451 	 * We just get fuse_getattr_out with list size if we requested
3452 	 * a null size.
3453 	 */
3454 	if (resid == NULL) {
3455 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3456 
3457 		if (attrsize != NULL)
3458 			*attrsize = fgo->size;
3459 
3460 		ps->ps_destroy_msg(pm);
3461 
3462 		return 0;
3463 	}
3464 
3465 	/*
3466 	 * And with a non null requested size, we get the list just
3467 	 * after the header
3468 	 */
3469 	foh = GET_OUTHDR(ps, pm);
3470 	np = (char *)(void *)(foh + 1);
3471 	puffs_len = foh->len - sizeof(*foh);
3472 
3473 	if (attrs != NULL) {
3474 #ifdef PUFFS_EXTATTR_LIST_LENPREFIX
3475 		/*
3476 		 * Convert the FUSE reply to length prefixed strings
3477 		 * if this is what the kernel wants.
3478 		 */
3479 		if (flag & PUFFS_EXTATTR_LIST_LENPREFIX) {
3480 			size_t i, attrlen;
3481 
3482 			for (i = 0; i < puffs_len; i += attrlen + 1) {
3483 				attrlen = strlen(np + i);
3484 				(void)memmove(np + i + 1, np + i, attrlen);
3485 				*(np + i) = (uint8_t)attrlen;
3486 			}
3487 		}
3488 #endif /* PUFFS_EXTATTR_LIST_LENPREFIX */
3489 		(void)memcpy(attrs, np, puffs_len);
3490 		*resid -= puffs_len;
3491 	}
3492 
3493 	if (attrsize != NULL)
3494 		*attrsize = puffs_len;
3495 
3496 	ps->ps_destroy_msg(pm);
3497 
3498 	return 0;
3499 }
3500 
3501 int
3502 perfuse_node_deleteextattr(pu, opc, attrns, attrname, pcr)
3503 	struct puffs_usermount *pu;
3504 	puffs_cookie_t opc;
3505 	int attrns;
3506 	const char *attrname;
3507 	const struct puffs_cred *pcr;
3508 {
3509 	struct perfuse_state *ps;
3510 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3511 	perfuse_msg_t *pm;
3512 	size_t attrnamelen;
3513 	char *np;
3514 	int error;
3515 
3516 	ps = puffs_getspecific(pu);
3517 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3518 	attrnamelen = strlen(attrname) + 1;
3519 
3520 	pm = ps->ps_new_msg(pu, opc, FUSE_REMOVEXATTR, attrnamelen, pcr);
3521 	np = _GET_INPAYLOAD(ps, pm, char *);
3522 	(void)strlcpy(np, attrname, attrnamelen);
3523 
3524 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
3525 
3526 	ps->ps_destroy_msg(pm);
3527 
3528 	return error;
3529 }
3530