xref: /netbsd-src/lib/libperfuse/ops.c (revision 9ddb6ab554e70fb9bbd90c3d96b812bc57755a14)
1 /*  $NetBSD: ops.c,v 1.51 2012/03/08 14:58:57 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 error;
211 
212 	ps = puffs_getspecific(pu);
213 	pnd = NULL;
214 	if ((struct puffs_node *)opc != NULL)
215 		pnd = PERFUSE_NODE_DATA(opc);
216 
217 #ifdef PERFUSE_DEBUG
218 	if ((perfuse_diagflags & PDF_FILENAME) && (opc != 0))
219 		DPRINTF("file = \"%s\", ino = %"PRIu64" flags = 0x%x\n",
220 			perfuse_node_path(opc),
221 			((struct puffs_node *)opc)->pn_va.va_fileid,
222 			PERFUSE_NODE_DATA(opc)->pnd_flags);
223 #endif
224 	if (pnd)
225 		pnd->pnd_flags |= PND_INXCHG;
226 
227 	/*
228 	 * Record FUSE call start if requested
229 	 */
230 	if (perfuse_diagflags & PDF_TRACE)
231 		pt = perfuse_trace_begin(ps, opc, pm);
232 
233 	/*
234 	 * Do actual FUSE exchange
235 	 */
236 	if ((error = ps->ps_xchg_msg(pu, pm, len, wait)) != 0)
237 		ps->ps_destroy_msg(pm);
238 
239 	/*
240 	 * Record FUSE call end if requested
241 	 */
242 	if (pt != NULL)
243 		perfuse_trace_end(ps, pt, error);
244 
245 	if (pnd) {
246 		pnd->pnd_flags &= ~PND_INXCHG;
247 		(void)dequeue_requests(ps, opc, PCQ_AFTERXCHG, DEQUEUE_ALL);
248 	}
249 
250 	return error;
251 }
252 
253 static int
254 mode_access(opc, pcr, mode)
255 	puffs_cookie_t opc;
256 	const struct puffs_cred *pcr;
257 	mode_t mode;
258 {
259 	struct puffs_node *pn;
260 	struct vattr *va;
261 
262 	/*
263 	 * pcr is NULL for self open through fsync or readdir.
264 	 * In both case, access control is useless, as it was
265 	 * done before, at open time.
266 	 */
267 	if (pcr == NULL)
268 		return 0;
269 
270 	pn = (struct puffs_node *)opc;
271 	va = puffs_pn_getvap(pn);
272 	return puffs_access(va->va_type, va->va_mode,
273 			    va->va_uid, va->va_gid,
274 			    mode, pcr);
275 }
276 
277 static int
278 sticky_access(targ, pcr)
279 	struct puffs_node *targ;
280 	const struct puffs_cred *pcr;
281 {
282 	uid_t uid;
283 	struct puffs_node *tdir;
284 	int sticky, owner;
285 
286 	tdir = PERFUSE_NODE_DATA(targ)->pnd_parent;
287 
288 	/*
289 	 * This covers the case where the kernel requests a DELETE
290 	 * or RENAME on its own, and where puffs_cred_getuid would
291 	 * return -1. While such a situation should not happen,
292 	 * we allow it here.
293 	 *
294 	 * This also allows root to tamper with other users' files
295 	 * that have the sticky bit.
296 	 */
297 	if (puffs_cred_isjuggernaut(pcr))
298 		return 0;
299 
300 	if (puffs_cred_getuid(pcr, &uid) != 0)
301 		DERRX(EX_SOFTWARE, "puffs_cred_getuid fails in %s", __func__);
302 
303 	sticky = puffs_pn_getvap(tdir)->va_mode & S_ISTXT;
304 	owner = puffs_pn_getvap(targ)->va_uid == uid;
305 
306 	if (sticky && !owner)
307 		return EACCES;
308 
309 	return 0;
310 }
311 
312 
313 static void
314 fuse_attr_to_vap(ps, vap, fa)
315 	struct perfuse_state *ps;
316 	struct vattr *vap;
317 	struct fuse_attr *fa;
318 {
319 	vap->va_type = IFTOVT(fa->mode);
320 	vap->va_mode = fa->mode & ALLPERMS;
321 	vap->va_nlink = fa->nlink;
322 	vap->va_uid = fa->uid;
323 	vap->va_gid = fa->gid;
324 	vap->va_fsid = (long)ps->ps_fsid;
325 	vap->va_fileid = fa->ino;
326 	vap->va_size = fa->size;
327 	vap->va_blocksize = fa->blksize;
328 	vap->va_atime.tv_sec = (time_t)fa->atime;
329 	vap->va_atime.tv_nsec = (long) fa->atimensec;
330 	vap->va_mtime.tv_sec = (time_t)fa->mtime;
331 	vap->va_mtime.tv_nsec = (long)fa->mtimensec;
332 	vap->va_ctime.tv_sec = (time_t)fa->ctime;
333 	vap->va_ctime.tv_nsec = (long)fa->ctimensec;
334 	vap->va_birthtime.tv_sec = 0;
335 	vap->va_birthtime.tv_nsec = 0;
336 	vap->va_gen = 0;
337 	vap->va_flags = 0;
338 	vap->va_rdev = fa->rdev;
339 	vap->va_bytes = fa->size;
340 	vap->va_filerev = (u_quad_t)PUFFS_VNOVAL;
341 	vap->va_vaflags = 0;
342 
343 	if (vap->va_blocksize == 0)
344 		vap->va_blocksize = DEV_BSIZE;
345 
346 	if (vap->va_size == (size_t)PUFFS_VNOVAL) /* XXX */
347 		vap->va_size = 0;
348 
349 	return;
350 }
351 
352 static void
353 set_expire(opc, feo, fao)
354 	puffs_cookie_t opc;
355 	struct fuse_entry_out *feo;
356 	struct fuse_attr_out *fao;
357 {
358 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
359 	struct timespec entry_ts;
360 	struct timespec attr_ts;
361 	struct timespec now;
362 
363 	if ((feo == NULL) && (fao == NULL))
364 		DERRX(EX_SOFTWARE, "%s: feo and fao NULL", __func__);
365 
366 	if ((feo != NULL) && (fao != NULL))
367 		DERRX(EX_SOFTWARE, "%s: feo and fao != NULL", __func__);
368 
369 	if (clock_gettime(CLOCK_REALTIME, &now) != 0)
370 		DERR(EX_OSERR, "clock_gettime failed");
371 
372 	if (feo != NULL) {
373 		entry_ts.tv_sec = (time_t)feo->entry_valid;
374 		entry_ts.tv_nsec = (long)feo->entry_valid_nsec;
375 
376 		timespecadd(&now, &entry_ts, &pnd->pnd_entry_expire);
377 
378 		attr_ts.tv_sec = (time_t)feo->attr_valid;
379 		attr_ts.tv_nsec = (long)feo->attr_valid_nsec;
380 
381 		timespecadd(&now, &attr_ts, &pnd->pnd_attr_expire);
382 	}
383 
384 	if (fao != NULL) {
385 		attr_ts.tv_sec = (time_t)fao->attr_valid;
386 		attr_ts.tv_nsec = (long)fao->attr_valid_nsec;
387 
388 		timespecadd(&now, &attr_ts, &pnd->pnd_attr_expire);
389 	}
390 
391 	return;
392 }
393 
394 static int
395 attr_expired(opc)
396 	puffs_cookie_t opc;
397 {
398 	struct perfuse_node_data *pnd;
399 	struct timespec expire;
400 	struct timespec now;
401 
402 	pnd = PERFUSE_NODE_DATA(opc);
403 	expire = pnd->pnd_attr_expire;
404 
405 	if (clock_gettime(CLOCK_REALTIME, &now) != 0)
406 		DERR(EX_OSERR, "clock_gettime failed");
407 
408 	return timespeccmp(&expire, &now, <);
409 }
410 
411 static int
412 entry_expired(opc)
413 	puffs_cookie_t opc;
414 {
415 	struct perfuse_node_data *pnd;
416 	struct timespec expire;
417 	struct timespec now;
418 
419 	pnd = PERFUSE_NODE_DATA(opc);
420 	expire = pnd->pnd_entry_expire;
421 
422 	if (clock_gettime(CLOCK_REALTIME, &now) != 0)
423 		DERR(EX_OSERR, "clock_gettime failed");
424 
425 	return timespeccmp(&expire, &now, <);
426 }
427 
428 
429 /*
430  * Lookup name in directory opc
431  * We take special care of name being . or ..
432  * These are returned by readdir and deserve tweaks.
433  */
434 static int
435 node_lookup_dir_nodot(pu, opc, name, namelen, pnp)
436 	struct puffs_usermount *pu;
437 	puffs_cookie_t opc;
438 	char *name;
439 	size_t namelen;
440 	struct puffs_node **pnp;
441 {
442 	/*
443 	 * "dot" is easy as we already know it
444 	 */
445 	if (strncmp(name, ".", namelen) == 0) {
446 		*pnp = (struct puffs_node *)opc;
447 		return 0;
448 	}
449 
450 	/*
451 	 * "dotdot" is also known
452 	 */
453 	if (strncmp(name, "..", namelen) == 0) {
454 		*pnp = PERFUSE_NODE_DATA(opc)->pnd_parent;
455 		return 0;
456 	}
457 
458 	return node_lookup_common(pu, opc, name, NULL, pnp);
459 }
460 
461 static int
462 node_lookup_common(pu, opc, path, pcr, pnp)
463 	struct puffs_usermount *pu;
464 	puffs_cookie_t opc;
465 	const char *path;
466 	const struct puffs_cred *pcr;
467 	struct puffs_node **pnp;
468 {
469 	struct perfuse_state *ps;
470 	struct perfuse_node_data *oldpnd;
471 	perfuse_msg_t *pm;
472 	struct fuse_entry_out *feo;
473 	struct puffs_node *pn;
474 	size_t len;
475 	int error;
476 
477 	/*
478 	 * Prevent further lookups if the parent was removed
479 	 */
480 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
481 		return ESTALE;
482 
483 	if (pnp == NULL)
484 		DERRX(EX_SOFTWARE, "pnp must be != NULL");
485 
486 	ps = puffs_getspecific(pu);
487 
488 #ifdef PERFUSE_DEBUG
489 	if (perfuse_diagflags & PDF_FILENAME)
490 		DPRINTF("%s: opc = %p, file = \"%s\" looking up \"%s\"\n",
491 			__func__, (void *)opc, perfuse_node_path(opc), path);
492 #endif
493 	/*
494 	 * Is the node already known?
495 	 */
496 	TAILQ_FOREACH(oldpnd, &PERFUSE_NODE_DATA(opc)->pnd_children, pnd_next) {
497 		if ((oldpnd->pnd_flags & PND_REMOVED) ||
498 		    (strcmp(oldpnd->pnd_name, path) != 0))
499 			continue;
500 
501 #ifdef PERFUSE_DEBUG
502 		if (perfuse_diagflags & PDF_FILENAME)
503 			DPRINTF("%s: opc = %p, file = \"%s\" found "
504 				"cookie = %p, nodeid = 0x%"PRIx64" "
505 				"for \"%s\"\n", __func__,
506 				(void *)opc, perfuse_node_path(opc),
507 				(void *)oldpnd->pnd_pn, oldpnd->pnd_nodeid,
508 				path);
509 #endif
510 		break;
511 	}
512 
513 	/*
514 	 * Check for cached name
515 	 */
516 	if ((oldpnd != NULL) && !entry_expired(oldpnd->pnd_pn)) {
517 		oldpnd->pnd_puffs_nlookup++;
518 		*pnp = oldpnd->pnd_pn;
519 		return 0;
520 	}
521 
522 	len = strlen(path) + 1;
523 
524 	pm = ps->ps_new_msg(pu, opc, FUSE_LOOKUP, len, pcr);
525 	(void)strlcpy(_GET_INPAYLOAD(ps, pm, char *), path, len);
526 
527 	error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply);
528 
529 	switch (error) {
530 	case 0:
531 		break;
532 	case ENOENT:
533 		if (oldpnd != NULL) {
534 			oldpnd->pnd_flags |= PND_REMOVED;
535 #ifdef PERFUSE_DEBUG
536 			if (perfuse_diagflags & PDF_FILENAME)
537 				DPRINTF("%s: opc = %p nodeid = 0x%"PRIx64" "
538 					"file = \"%s\" removed\n", __func__,
539 					oldpnd->pnd_pn, oldpnd->pnd_nodeid,
540 					oldpnd->pnd_name);
541 #endif
542 		}
543 		/* FALLTHROUGH */
544 	default:
545 		return error;
546 		/* NOTREACHED */
547 		break;
548 	}
549 
550 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
551 
552 	if (oldpnd != NULL) {
553 		if (oldpnd->pnd_nodeid == feo->nodeid) {
554 			oldpnd->pnd_fuse_nlookup++;
555 			oldpnd->pnd_puffs_nlookup++;
556 			*pnp = oldpnd->pnd_pn;
557 
558 			ps->ps_destroy_msg(pm);
559 			return 0;
560 		} else {
561 			oldpnd->pnd_flags |= PND_REMOVED;
562 #ifdef PERFUSE_DEBUG
563 			if (perfuse_diagflags & PDF_FILENAME)
564 				DPRINTF("%s: opc = %p nodeid = 0x%"PRIx64" "
565 					"file = \"%s\" replaced\n", __func__,
566 					oldpnd->pnd_pn, oldpnd->pnd_nodeid,
567 					oldpnd->pnd_name);
568 #endif
569 		}
570 	}
571 
572 	pn = perfuse_new_pn(pu, path, opc);
573 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
574 
575 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
576 	pn->pn_va.va_gen = (u_long)(feo->generation);
577 	set_expire((puffs_cookie_t)pn, feo, NULL);
578 
579 	*pnp = pn;
580 
581 #ifdef PERFUSE_DEBUG
582 	if (perfuse_diagflags & PDF_FILENAME)
583 		DPRINTF("%s: opc = %p, looked up opc = %p, "
584 			"nodeid = 0x%"PRIx64" file = \"%s\"\n", __func__,
585 			(void *)opc, pn, feo->nodeid, path);
586 #endif
587 
588 	ps->ps_destroy_msg(pm);
589 
590 	return 0;
591 }
592 
593 
594 /*
595  * Common code for methods that create objects:
596  * perfuse_node_mkdir
597  * perfuse_node_mknod
598  * perfuse_node_symlink
599  */
600 static int
601 node_mk_common(pu, opc, pni, pcn, pm)
602 	struct puffs_usermount *pu;
603 	puffs_cookie_t opc;
604 	struct puffs_newinfo *pni;
605 	const struct puffs_cn *pcn;
606 	perfuse_msg_t *pm;
607 {
608 	struct perfuse_state *ps;
609 	struct puffs_node *pn;
610 	struct fuse_entry_out *feo;
611 	int error;
612 
613 	ps =  puffs_getspecific(pu);
614 
615 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
616 		return error;
617 
618 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
619 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
620 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
621 
622 	pn = perfuse_new_pn(pu, pcn->pcn_name, opc);
623 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
624 
625 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
626 	pn->pn_va.va_gen = (u_long)(feo->generation);
627 	set_expire((puffs_cookie_t)pn, feo, NULL);
628 
629 	puffs_newinfo_setcookie(pni, pn);
630 
631 #ifdef PERFUSE_DEBUG
632 	if (perfuse_diagflags & PDF_FILENAME)
633 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
634 			"nodeid = 0x%"PRIx64"\n",
635 			__func__, (void *)pn, pcn->pcn_name,
636 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid);
637 #endif
638 	ps->ps_destroy_msg(pm);
639 
640 	return node_mk_common_final(pu, opc, pn, pcn);
641 }
642 
643 /*
644  * Common final code for methods that create objects:
645  * perfuse_node_mkdir via node_mk_common
646  * perfuse_node_mknod via node_mk_common
647  * perfuse_node_symlink via node_mk_common
648  * perfuse_node_create
649  */
650 static int
651 node_mk_common_final(pu, opc, pn, pcn)
652 	struct puffs_usermount *pu;
653 	puffs_cookie_t opc;
654 	struct puffs_node *pn;
655 	const struct puffs_cn *pcn;
656 {
657 	struct perfuse_state *ps;
658 	perfuse_msg_t *pm;
659 	struct fuse_setattr_in *fsi;
660 	struct fuse_attr_out *fao;
661 	int error;
662 
663 	ps =  puffs_getspecific(pu);
664 
665 	/*
666 	 * Set owner and group. The kernel cannot create a file
667 	 * on its own (puffs_cred_getuid would return -1), right?
668 	 */
669 	if (puffs_cred_getuid(pcn->pcn_cred, &pn->pn_va.va_uid) != 0)
670 		DERRX(EX_SOFTWARE, "puffs_cred_getuid fails in %s", __func__);
671 	if (puffs_cred_getgid(pcn->pcn_cred, &pn->pn_va.va_gid) != 0)
672 		DERRX(EX_SOFTWARE, "puffs_cred_getgid fails in %s", __func__);
673 
674 	pm = ps->ps_new_msg(pu, (puffs_cookie_t)pn,
675 			    FUSE_SETATTR, sizeof(*fsi), pcn->pcn_cred);
676 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
677 	fsi->uid = pn->pn_va.va_uid;
678 	fsi->gid = pn->pn_va.va_gid;
679 	fsi->valid = FUSE_FATTR_UID|FUSE_FATTR_GID;
680 
681 	if ((error = xchg_msg(pu, (puffs_cookie_t)pn, pm,
682 			      sizeof(*fao), wait_reply)) != 0)
683 		return error;
684 
685 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
686 	fuse_attr_to_vap(ps, &pn->pn_va, &fao->attr);
687 	set_expire((puffs_cookie_t)pn, NULL, fao);
688 
689 	/*
690 	 * The parent directory needs a sync
691 	 */
692 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
693 
694 	ps->ps_destroy_msg(pm);
695 
696 	return 0;
697 }
698 
699 static uint64_t
700 readdir_last_cookie(fd, fd_len)
701 	struct fuse_dirent *fd;
702 	size_t fd_len;
703 {
704 	size_t len;
705 	size_t seen = 0;
706 	char *ndp;
707 
708 	do {
709 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
710 		seen += len;
711 
712 		if (seen >= fd_len)
713 			break;
714 
715 		ndp = (char *)(void *)fd + (size_t)len;
716 		fd = (struct fuse_dirent *)(void *)ndp;
717 	} while (1 /* CONSTCOND */);
718 
719 	return fd->off;
720 }
721 
722 static ssize_t
723 fuse_to_dirent(pu, opc, fd, fd_len)
724 	struct puffs_usermount *pu;
725 	puffs_cookie_t opc;
726 	struct fuse_dirent *fd;
727 	size_t fd_len;
728 {
729 	struct dirent *dents;
730 	size_t dents_len;
731 	ssize_t written;
732 	uint64_t fd_offset;
733 	struct fuse_dirent *fd_base;
734 	size_t len;
735 
736 	fd_base = fd;
737 	fd_offset = 0;
738 	written = 0;
739 	dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
740 	dents_len = (size_t)PERFUSE_NODE_DATA(opc)->pnd_dirent_len;
741 
742 	do {
743 		char *ndp;
744 		size_t reclen;
745 
746 		reclen = _DIRENT_RECLEN(dents, fd->namelen);
747 
748 		/*
749 		 * Check we do not overflow the output buffer
750 		 * struct fuse_dirent is bigger than struct dirent,
751 		 * so we should always use fd_len and never reallocate
752 		 * later.
753 		 * If we have to reallocate,try to double the buffer
754 		 * each time so that we do not have to do it too often.
755 		 */
756 		if (written + reclen > dents_len) {
757 			if (dents_len == 0)
758 				dents_len = fd_len;
759 			else
760 				dents_len =
761 				   MAX(2 * dents_len, written + reclen);
762 
763 			dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
764 			if ((dents = realloc(dents, dents_len)) == NULL)
765 				DERR(EX_OSERR, "%s: malloc failed", __func__);
766 
767 			PERFUSE_NODE_DATA(opc)->pnd_dirent = dents;
768 			PERFUSE_NODE_DATA(opc)->pnd_dirent_len = dents_len;
769 
770 			/*
771 			 * (void *) for delint
772 			 */
773 			ndp = (char *)(void *)dents + written;
774 			dents = (struct dirent *)(void *)ndp;
775 		}
776 
777 		/*
778 		 * Filesystem was mounted without -o use_ino
779 		 * Perform a lookup to find it.
780 		 */
781 		if (fd->ino == PERFUSE_UNKNOWN_INO) {
782 			struct puffs_node *pn;
783 
784 			if (node_lookup_dir_nodot(pu, opc, fd->name,
785 						  fd->namelen, &pn) != 0) {
786 				DWARNX("node_lookup_dir_nodot failed");
787 			} else {
788 				fd->ino = pn->pn_va.va_fileid;
789 			}
790 		}
791 
792 		dents->d_fileno = fd->ino;
793 		dents->d_reclen = (unsigned short)reclen;
794 		dents->d_namlen = fd->namelen;
795 		dents->d_type = fd->type;
796 		strlcpy(dents->d_name, fd->name, fd->namelen + 1);
797 
798 #ifdef PERFUSE_DEBUG
799 		if (perfuse_diagflags & PDF_READDIR)
800 			DPRINTF("%s: translated \"%s\" ino = %"PRIu64"\n",
801 				__func__, dents->d_name, dents->d_fileno);
802 #endif
803 
804 		dents = _DIRENT_NEXT(dents);
805 		written += reclen;
806 
807 		/*
808 		 * Move to the next record.
809 		 * fd->off is not the offset, it is an opaque cookie
810 		 * given by the filesystem to keep state across multiple
811 		 * readdir() operation.
812 		 * Use record alignement instead.
813 		 */
814 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
815 #ifdef PERFUSE_DEBUG
816 		if (perfuse_diagflags & PDF_READDIR)
817 			DPRINTF("%s: record at %"PRId64"/0x%"PRIx64" "
818 				"length = %zd/0x%zx. "
819 				"next record at %"PRId64"/0x%"PRIx64" "
820 				"max %zd/0x%zx\n",
821 				__func__, fd_offset, fd_offset, len, len,
822 				fd_offset + len, fd_offset + len,
823 				fd_len, fd_len);
824 #endif
825 		fd_offset += len;
826 
827 		/*
828 		 * Check if next record is still within the packet
829 		 * If it is not, we reached the end of the buffer.
830 		 */
831 		if (fd_offset >= fd_len)
832 			break;
833 
834 		/*
835 		 * (void *) for delint
836 		 */
837 		ndp = (char *)(void *)fd_base + (size_t)fd_offset;
838 		fd = (struct fuse_dirent *)(void *)ndp;
839 
840 	} while (1 /* CONSTCOND */);
841 
842 	/*
843 	 * Adjust the dirent output length
844 	 */
845 	if (written != -1)
846 		PERFUSE_NODE_DATA(opc)->pnd_dirent_len = written;
847 
848 	return written;
849 }
850 
851 static int
852 readdir_buffered(opc, dent, readoff, reslen)
853 	puffs_cookie_t opc;
854 	struct dirent *dent;
855 	off_t *readoff;
856 	size_t *reslen;
857 {
858 	struct dirent *fromdent;
859 	struct perfuse_node_data *pnd;
860 	char *ndp;
861 
862 	pnd = PERFUSE_NODE_DATA(opc);
863 
864 	while (*readoff < pnd->pnd_dirent_len) {
865 		/*
866 		 * (void *) for delint
867 		 */
868 		ndp = (char *)(void *)pnd->pnd_dirent + (size_t)*readoff;
869 		fromdent = (struct dirent *)(void *)ndp;
870 
871 		if (*reslen < _DIRENT_SIZE(fromdent))
872 			break;
873 
874 		memcpy(dent, fromdent, _DIRENT_SIZE(fromdent));
875 		*readoff += _DIRENT_SIZE(fromdent);
876 		*reslen -= _DIRENT_SIZE(fromdent);
877 
878 		dent = _DIRENT_NEXT(dent);
879 	}
880 
881 #ifdef PERFUSE_DEBUG
882 	if (perfuse_diagflags & PDF_READDIR)
883 		DPRINTF("%s: readoff = %"PRId64",  "
884 			"pnd->pnd_dirent_len = %"PRId64"\n",
885 			__func__, *readoff, pnd->pnd_dirent_len);
886 #endif
887 	if (*readoff >=  pnd->pnd_dirent_len) {
888 		free(pnd->pnd_dirent);
889 		pnd->pnd_dirent = NULL;
890 		pnd->pnd_dirent_len = 0;
891 	}
892 
893 	return 0;
894 }
895 
896 static void
897 requeue_request(pu, opc, type)
898 	struct puffs_usermount *pu;
899 	puffs_cookie_t opc;
900 	enum perfuse_qtype type;
901 {
902 	struct perfuse_cc_queue pcq;
903 	struct perfuse_node_data *pnd;
904 #ifdef PERFUSE_DEBUG
905 	struct perfuse_state *ps;
906 
907 	ps = perfuse_getspecific(pu);
908 #endif
909 
910 	pnd = PERFUSE_NODE_DATA(opc);
911 	pcq.pcq_type = type;
912 	pcq.pcq_cc = puffs_cc_getcc(pu);
913 	TAILQ_INSERT_TAIL(&pnd->pnd_pcq, &pcq, pcq_next);
914 
915 #ifdef PERFUSE_DEBUG
916 	if (perfuse_diagflags & PDF_REQUEUE)
917 		DPRINTF("%s: REQUEUE opc = %p, pcc = %p (%s)\n",
918 		        __func__, (void *)opc, pcq.pcq_cc,
919 			perfuse_qtypestr[type]);
920 #endif
921 
922 	puffs_cc_yield(pcq.pcq_cc);
923 	TAILQ_REMOVE(&pnd->pnd_pcq, &pcq, pcq_next);
924 
925 #ifdef PERFUSE_DEBUG
926 	if (perfuse_diagflags & PDF_REQUEUE)
927 		DPRINTF("%s: RESUME opc = %p, pcc = %p (%s)\n",
928 		        __func__, (void *)opc, pcq.pcq_cc,
929 			perfuse_qtypestr[type]);
930 #endif
931 
932 	return;
933 }
934 
935 /* ARGSUSED0 */
936 static int
937 dequeue_requests(ps, opc, type, max)
938 	struct perfuse_state *ps;
939 	puffs_cookie_t opc;
940 	enum perfuse_qtype type;
941 	int max;
942 {
943 	struct perfuse_cc_queue *pcq;
944 	struct perfuse_node_data *pnd;
945 	int dequeued;
946 
947 	pnd = PERFUSE_NODE_DATA(opc);
948 	dequeued = 0;
949 	TAILQ_FOREACH(pcq, &pnd->pnd_pcq, pcq_next) {
950 		if (pcq->pcq_type != type)
951 			continue;
952 
953 #ifdef PERFUSE_DEBUG
954 		if (perfuse_diagflags & PDF_REQUEUE)
955 			DPRINTF("%s: SCHEDULE opc = %p, pcc = %p (%s)\n",
956 				__func__, (void *)opc, pcq->pcq_cc,
957 				 perfuse_qtypestr[type]);
958 #endif
959 		puffs_cc_schedule(pcq->pcq_cc);
960 
961 		if (++dequeued == max)
962 			break;
963 	}
964 
965 #ifdef PERFUSE_DEBUG
966 	if (perfuse_diagflags & PDF_REQUEUE)
967 		DPRINTF("%s: DONE  opc = %p\n", __func__, (void *)opc);
968 #endif
969 
970 	return dequeued;
971 }
972 
973 void
974 perfuse_fs_init(pu)
975 	struct puffs_usermount *pu;
976 {
977 	struct perfuse_state *ps;
978 	perfuse_msg_t *pm;
979 	struct fuse_init_in *fii;
980 	struct fuse_init_out *fio;
981 	int error;
982 
983 	ps = puffs_getspecific(pu);
984 
985         if (puffs_mount(pu, ps->ps_target, ps->ps_mountflags, ps->ps_root) != 0)
986                 DERR(EX_OSERR, "%s: puffs_mount failed", __func__);
987 
988 	/*
989 	 * Linux 2.6.34.1 sends theses flags:
990 	 * FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC
991 	 * FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK
992 	 *
993 	 * Linux also sets max_readahead at 32 pages (128 kB)
994 	 *
995 	 * ps_new_msg() is called with NULL creds, which will
996 	 * be interpreted as FUSE superuser.
997 	 */
998 	pm = ps->ps_new_msg(pu, 0, FUSE_INIT, sizeof(*fii), NULL);
999 	fii = GET_INPAYLOAD(ps, pm, fuse_init_in);
1000 	fii->major = FUSE_KERNEL_VERSION;
1001 	fii->minor = FUSE_KERNEL_MINOR_VERSION;
1002 	fii->max_readahead = (unsigned int)(32 * sysconf(_SC_PAGESIZE));
1003 	fii->flags = (FUSE_ASYNC_READ|FUSE_POSIX_LOCKS|FUSE_ATOMIC_O_TRUNC);
1004 
1005 	if ((error = xchg_msg(pu, 0, pm, sizeof(*fio), wait_reply)) != 0)
1006 		DERRX(EX_SOFTWARE, "init message exchange failed (%d)", error);
1007 
1008 	fio = GET_OUTPAYLOAD(ps, pm, fuse_init_out);
1009 	ps->ps_max_readahead = fio->max_readahead;
1010 	ps->ps_max_write = fio->max_write;
1011 
1012 	ps->ps_destroy_msg(pm);
1013 
1014 	return;
1015 }
1016 
1017 int
1018 perfuse_fs_unmount(pu, flags)
1019 	struct puffs_usermount *pu;
1020 	int flags;
1021 {
1022 	perfuse_msg_t *pm;
1023 	struct perfuse_state *ps;
1024 	puffs_cookie_t opc;
1025 	int error;
1026 
1027 	ps = puffs_getspecific(pu);
1028 	opc = (puffs_cookie_t)puffs_getroot(pu);
1029 
1030 	/*
1031 	 * ps_new_msg() is called with NULL creds, which will
1032 	 * be interpreted as FUSE superuser.
1033 	 */
1034 	pm = ps->ps_new_msg(pu, opc, FUSE_DESTROY, 0, NULL);
1035 
1036 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0){
1037 		DWARN("unmount %s", ps->ps_target);
1038 		if (!(flags & MNT_FORCE))
1039 			return error;
1040 		else
1041 			error = 0;
1042 	} else {
1043 		ps->ps_destroy_msg(pm);
1044 	}
1045 
1046 	ps->ps_umount(pu);
1047 
1048 	if (perfuse_diagflags & PDF_MISC)
1049 		DPRINTF("%s unmounted, exit\n", ps->ps_target);
1050 
1051 	return 0;
1052 }
1053 
1054 int
1055 perfuse_fs_statvfs(pu, svfsb)
1056 	struct puffs_usermount *pu;
1057 	struct statvfs *svfsb;
1058 {
1059 	struct perfuse_state *ps;
1060 	perfuse_msg_t *pm;
1061 	puffs_cookie_t opc;
1062 	struct fuse_statfs_out *fso;
1063 	int error;
1064 
1065 	ps = puffs_getspecific(pu);
1066 	opc = (puffs_cookie_t)puffs_getroot(pu);
1067 
1068 	/*
1069 	 * ps_new_msg() is called with NULL creds, which will
1070 	 * be interpreted as FUSE superuser.
1071 	 */
1072 	pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
1073 
1074 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply)) != 0)
1075 		return error;
1076 
1077 	fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
1078 	svfsb->f_flag = ps->ps_mountflags;
1079 	svfsb->f_bsize = fso->st.bsize;
1080 	svfsb->f_frsize = fso->st.frsize;
1081 	svfsb->f_iosize = ((struct puffs_node *)opc)->pn_va.va_blocksize;
1082 	svfsb->f_blocks = fso->st.blocks;
1083 	svfsb->f_bfree = fso->st.bfree;
1084 	svfsb->f_bavail = fso->st.bavail;
1085 	svfsb->f_bresvd = fso->st.bfree - fso->st.bavail;
1086 	svfsb->f_files = fso->st.files;
1087 	svfsb->f_ffree = fso->st.ffree;
1088 	svfsb->f_favail = fso->st.ffree;/* files not reserved for root */
1089 	svfsb->f_fresvd = 0;		/* files reserved for root */
1090 
1091 	svfsb->f_syncreads = ps->ps_syncreads;
1092 	svfsb->f_syncwrites = ps->ps_syncwrites;
1093 
1094 	svfsb->f_asyncreads = ps->ps_asyncreads;
1095 	svfsb->f_asyncwrites = ps->ps_asyncwrites;
1096 
1097 	(void)memcpy(&svfsb->f_fsidx, &ps->ps_fsid, sizeof(ps->ps_fsid));
1098 	svfsb->f_fsid = (unsigned long)ps->ps_fsid;
1099 	svfsb->f_namemax = MAXPATHLEN;	/* XXX */
1100 	svfsb->f_owner = ps->ps_owner_uid;
1101 
1102 	(void)strlcpy(svfsb->f_mntonname, ps->ps_target, _VFS_NAMELEN);
1103 
1104 	if (ps->ps_filesystemtype != NULL)
1105 		(void)strlcpy(svfsb->f_fstypename,
1106 			      ps->ps_filesystemtype, _VFS_NAMELEN);
1107 	else
1108 		(void)strlcpy(svfsb->f_fstypename, "fuse", _VFS_NAMELEN);
1109 
1110 	if (ps->ps_source != NULL)
1111 		strlcpy(svfsb->f_mntfromname, ps->ps_source, _VFS_NAMELEN);
1112 	else
1113 		strlcpy(svfsb->f_mntfromname, _PATH_FUSE, _VFS_NAMELEN);
1114 
1115 	ps->ps_destroy_msg(pm);
1116 
1117 	return 0;
1118 }
1119 
1120 int
1121 perfuse_fs_sync(pu, waitfor, pcr)
1122 	struct puffs_usermount *pu;
1123 	int waitfor;
1124 	const struct puffs_cred *pcr;
1125 {
1126 	/*
1127 	 * FUSE does not seem to have a FS sync callback.
1128 	 * Maybe do not even register this callback
1129 	 */
1130 	return puffs_fsnop_sync(pu, waitfor, pcr);
1131 }
1132 
1133 /* ARGSUSED0 */
1134 int
1135 perfuse_fs_fhtonode(pu, fid, fidsize, pni)
1136 	struct puffs_usermount *pu;
1137 	void *fid;
1138 	size_t fidsize;
1139 	struct puffs_newinfo *pni;
1140 {
1141 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1142 	return 0;
1143 }
1144 
1145 /* ARGSUSED0 */
1146 int
1147 perfuse_fs_nodetofh(pu, cookie, fid, fidsize)
1148 	struct puffs_usermount *pu;
1149 	puffs_cookie_t cookie;
1150 	void *fid;
1151 	size_t *fidsize;
1152 {
1153 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1154 	return 0;
1155 }
1156 
1157 #if 0
1158 /* ARGSUSED0 */
1159 void
1160 perfuse_fs_extattrctl(pu, cmd, cookie, flags, namespace, attrname)
1161 	struct puffs_usermount *pu;
1162 	int cmd,
1163 	puffs_cookie_t *cookie;
1164 	int flags;
1165 	int namespace;
1166 	const char *attrname;
1167 {
1168 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1169 	return 0;
1170 }
1171 #endif /* 0 */
1172 
1173 /* ARGSUSED0 */
1174 void
1175 perfuse_fs_suspend(pu, status)
1176 	struct puffs_usermount *pu;
1177 	int status;
1178 {
1179 	return;
1180 }
1181 
1182 
1183 
1184 int
1185 perfuse_node_lookup(pu, opc, pni, pcn)
1186 	struct puffs_usermount *pu;
1187 	puffs_cookie_t opc;
1188 	struct puffs_newinfo *pni;
1189 	const struct puffs_cn *pcn;
1190 {
1191 	struct puffs_node *pn;
1192 	mode_t mode;
1193 	int error;
1194 
1195 	/*
1196 	 * Check permissions
1197 	 */
1198 	switch(pcn->pcn_nameiop) {
1199 	case NAMEI_DELETE: /* FALLTHROUGH */
1200 	case NAMEI_RENAME: /* FALLTHROUGH */
1201 	case NAMEI_CREATE:
1202 		if (pcn->pcn_flags & NAMEI_ISLASTCN)
1203 			mode = PUFFS_VEXEC|PUFFS_VWRITE;
1204 		else
1205 			mode = PUFFS_VEXEC;
1206 		break;
1207 	case NAMEI_LOOKUP: /* FALLTHROUGH */
1208 	default:
1209 		mode = PUFFS_VEXEC;
1210 		break;
1211 	}
1212 
1213 	if ((error = mode_access(opc, pcn->pcn_cred, mode)) != 0)
1214 		return error;
1215 
1216 	/*
1217 	 * Special case for ..
1218 	 */
1219 	if (strcmp(pcn->pcn_name, "..") == 0)
1220 		pn = PERFUSE_NODE_DATA(opc)->pnd_parent;
1221 	else
1222 		error = node_lookup_common(pu, (puffs_cookie_t)opc,
1223 					   pcn->pcn_name, pcn->pcn_cred, &pn);
1224 	if (error != 0)
1225 		return error;
1226 
1227 	/*
1228 	 * Kernel would kill us if the filesystem returned the parent
1229 	 * itself. If we want to live, hide that!
1230 	 */
1231 	if ((opc == (puffs_cookie_t)pn) && (strcmp(pcn->pcn_name, ".") != 0)) {
1232 		DWARNX("lookup returned parent");
1233 		return ESTALE;
1234 	}
1235 
1236 	/*
1237 	 * Removed node
1238 	 */
1239 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_REMOVED)
1240 		return ENOENT;
1241 
1242 	/*
1243 	 * Check for sticky bit. Unfortunately there is no way to
1244 	 * do this before creating the puffs_node, since we require
1245 	 * this operation to get the node owner.
1246 	 */
1247 	switch (pcn->pcn_nameiop) {
1248 	case NAMEI_DELETE: /* FALLTHROUGH */
1249 	case NAMEI_RENAME:
1250 		error = sticky_access(pn, pcn->pcn_cred);
1251 		if (error != 0) {
1252 			/*
1253 			 * kernel will never know about it and will
1254 			 * not reclaim it. The filesystem needs to
1255 			 * clean it up anyway, therefore mimick a forget.
1256 			 */
1257 			PERFUSE_NODE_DATA(pn)->pnd_flags |= PND_RECLAIMED;
1258 			(void)perfuse_node_reclaim(pu, (puffs_cookie_t)pn);
1259 			return error;
1260 		}
1261 		break;
1262 	default:
1263 		break;
1264 	}
1265 
1266 	/*
1267 	 * If that node had a pending reclaim, wipe it out.
1268 	 */
1269 	PERFUSE_NODE_DATA(pn)->pnd_flags &= ~PND_RECLAIMED;
1270 
1271 	puffs_newinfo_setcookie(pni, pn);
1272 	puffs_newinfo_setvtype(pni, pn->pn_va.va_type);
1273 	puffs_newinfo_setsize(pni, (voff_t)pn->pn_va.va_size);
1274 	puffs_newinfo_setrdev(pni, pn->pn_va.va_rdev);
1275 
1276 	return error;
1277 }
1278 
1279 int
1280 perfuse_node_create(pu, opc, pni, pcn, vap)
1281 	struct puffs_usermount *pu;
1282 	puffs_cookie_t opc;
1283 	struct puffs_newinfo *pni;
1284 	const struct puffs_cn *pcn;
1285 	const struct vattr *vap;
1286 {
1287 	perfuse_msg_t *pm;
1288 	struct perfuse_state *ps;
1289 	struct fuse_create_in *fci;
1290 	struct fuse_entry_out *feo;
1291 	struct fuse_open_out *foo;
1292 	struct puffs_node *pn;
1293 	const char *name;
1294 	size_t namelen;
1295 	size_t len;
1296 	int error;
1297 
1298 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1299 		return ENOENT;
1300 
1301 	/*
1302 	 * If create is unimplemented: Check that it does not
1303 	 * already exists, and if not, do mknod and open
1304 	 */
1305 	ps = puffs_getspecific(pu);
1306 	if (ps->ps_flags & PS_NO_CREAT) {
1307 		error = node_lookup_common(pu, opc, pcn->pcn_name,
1308 					   pcn->pcn_cred, &pn);
1309 		if (error == 0)
1310 			return EEXIST;
1311 
1312 		error = perfuse_node_mknod(pu, opc, pni, pcn, vap);
1313 		if (error != 0)
1314 			return error;
1315 
1316 		error = node_lookup_common(pu, opc, pcn->pcn_name,
1317 					   pcn->pcn_cred, &pn);
1318 		if (error != 0)
1319 			return error;
1320 
1321 		/*
1322 		 * FUSE does the open at create time, while
1323 		 * NetBSD will open in a subsequent operation.
1324 		 * We need to open now, in order to retain FUSE
1325 		 * semantics. The calling process will not get
1326 		 * a file descriptor before the kernel sends
1327 		 * the open operation.
1328 		 */
1329 		opc = (puffs_cookie_t)pn;
1330 		error = perfuse_node_open(pu, opc, FWRITE, pcn->pcn_cred);
1331 		if (error != 0)
1332 			return error;
1333 
1334 		return 0;
1335 	}
1336 
1337 	name = pcn->pcn_name;
1338 	namelen = pcn->pcn_namelen + 1;
1339 	len = sizeof(*fci) + namelen;
1340 
1341 	/*
1342 	 * flags should use O_WRONLY instead of O_RDWR, but it
1343 	 * breaks when the caller tries to read from file.
1344 	 *
1345 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1346 	 */
1347 	pm = ps->ps_new_msg(pu, opc, FUSE_CREATE, len, pcn->pcn_cred);
1348 	fci = GET_INPAYLOAD(ps, pm, fuse_create_in);
1349 	fci->flags = O_CREAT | O_TRUNC | O_RDWR;
1350 	fci->mode = vap->va_mode | VTTOIF(vap->va_type);
1351 	fci->umask = 0; 	/* Seems unused by libfuse */
1352 	(void)strlcpy((char*)(void *)(fci + 1), name, namelen);
1353 
1354 	len = sizeof(*feo) + sizeof(*foo);
1355 	if ((error = xchg_msg(pu, opc, pm, len, wait_reply)) != 0) {
1356 		/*
1357 		 * create is unimplmented, remember it for later,
1358 		 * and start over using mknod and open instead.
1359 		 */
1360 		if (error == ENOSYS) {
1361 			ps->ps_flags |= PS_NO_CREAT;
1362 			return perfuse_node_create(pu, opc, pni, pcn, vap);
1363 		}
1364 
1365 		return error;
1366 	}
1367 
1368 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
1369 	foo = (struct fuse_open_out *)(void *)(feo + 1);
1370 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
1371 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
1372 
1373 	/*
1374 	 * Save the file handle and inode in node private data
1375 	 * so that we can reuse it later
1376 	 */
1377 	pn = perfuse_new_pn(pu, name, opc);
1378 	perfuse_new_fh((puffs_cookie_t)pn, foo->fh, FWRITE);
1379 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
1380 
1381 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
1382 	pn->pn_va.va_gen = (u_long)(feo->generation);
1383 	set_expire((puffs_cookie_t)pn, feo, NULL);
1384 
1385 	puffs_newinfo_setcookie(pni, pn);
1386 
1387 #ifdef PERFUSE_DEBUG
1388 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1389 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
1390 			"nodeid = 0x%"PRIx64", wfh = 0x%"PRIx64"\n",
1391 			__func__, (void *)pn, pcn->pcn_name,
1392 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid,
1393 			foo->fh);
1394 #endif
1395 
1396 	ps->ps_destroy_msg(pm);
1397 
1398 	return node_mk_common_final(pu, opc, pn, pcn);
1399 }
1400 
1401 
1402 int
1403 perfuse_node_mknod(pu, opc, pni, pcn, vap)
1404 	struct puffs_usermount *pu;
1405 	puffs_cookie_t opc;
1406 	struct puffs_newinfo *pni;
1407 	const struct puffs_cn *pcn;
1408 	const struct vattr *vap;
1409 {
1410 	struct perfuse_state *ps;
1411 	perfuse_msg_t *pm;
1412 	struct fuse_mknod_in *fmi;
1413 	const char* path;
1414 	size_t len;
1415 
1416 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1417 		return ENOENT;
1418 
1419 	/*
1420 	 * Only superuser can mknod objects other than
1421 	 * directories, files, socks, fifo and links.
1422 	 *
1423 	 * Create an object require -WX permission in the parent directory
1424 	 */
1425 	switch (vap->va_type) {
1426 	case VDIR:	/* FALLTHROUGH */
1427 	case VREG:	/* FALLTHROUGH */
1428 	case VFIFO:	/* FALLTHROUGH */
1429 	case VSOCK:
1430 		break;
1431 	default:	/* VNON, VBLK, VCHR, VBAD */
1432 		if (!puffs_cred_isjuggernaut(pcn->pcn_cred))
1433 			return EACCES;
1434 		break;
1435 	}
1436 
1437 
1438 	ps = puffs_getspecific(pu);
1439 	path = pcn->pcn_name;
1440 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
1441 
1442 	/*
1443 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1444 	 */
1445 	pm = ps->ps_new_msg(pu, opc, FUSE_MKNOD, len, pcn->pcn_cred);
1446 	fmi = GET_INPAYLOAD(ps, pm, fuse_mknod_in);
1447 	fmi->mode = vap->va_mode | VTTOIF(vap->va_type);
1448 	fmi->rdev = (uint32_t)vap->va_rdev;
1449 	fmi->umask = 0; 	/* Seems unused bu libfuse */
1450 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
1451 
1452 	return node_mk_common(pu, opc, pni, pcn, pm);
1453 }
1454 
1455 
1456 int
1457 perfuse_node_open(pu, opc, mode, pcr)
1458 	struct puffs_usermount *pu;
1459 	puffs_cookie_t opc;
1460 	int mode;
1461 	const struct puffs_cred *pcr;
1462 {
1463 	struct perfuse_state *ps;
1464 	struct perfuse_node_data *pnd;
1465 	perfuse_msg_t *pm;
1466 	mode_t fmode;
1467 	int op;
1468 	struct fuse_open_in *foi;
1469 	struct fuse_open_out *foo;
1470 	struct puffs_node *pn;
1471 	int error;
1472 
1473 	ps = puffs_getspecific(pu);
1474 	pn = (struct puffs_node *)opc;
1475 	pnd = PERFUSE_NODE_DATA(opc);
1476 	error = 0;
1477 
1478 	if (pnd->pnd_flags & PND_REMOVED)
1479 		return ENOENT;
1480 
1481 	if (puffs_pn_getvap(pn)->va_type == VDIR)
1482 		op = FUSE_OPENDIR;
1483 	else
1484 		op = FUSE_OPEN;
1485 
1486 	/*
1487 	 * libfuse docs says
1488 	 * - O_CREAT and O_EXCL should never be set.
1489 	 * - O_TRUNC may be used if mount option atomic_o_trunc is used XXX
1490 	 *
1491 	 * O_APPEND makes no sense since FUSE always sends
1492 	 * the file offset for write operations. If the
1493 	 * filesystem uses pwrite(), O_APPEND would cause
1494 	 * the offset to be ignored and cause file corruption.
1495 	 */
1496 	mode &= ~(O_CREAT|O_EXCL|O_APPEND);
1497 
1498 	/*
1499 	 * Do not open twice, and do not reopen for reading
1500 	 * if we already have write handle.
1501 	 */
1502 	if (((mode & FREAD) && (pnd->pnd_flags & PND_RFH)) ||
1503 	    ((mode & FREAD) && (pnd->pnd_flags & PND_WFH)) ||
1504 	    ((mode & FWRITE) && (pnd->pnd_flags & PND_WFH)))
1505 		goto out;
1506 
1507 	/*
1508 	 * Queue open on a node so that we do not open
1509 	 * twice. This would be better with read and
1510 	 * write distinguished.
1511 	 */
1512 	while (pnd->pnd_flags & PND_INOPEN)
1513 		requeue_request(pu, opc, PCQ_OPEN);
1514 	pnd->pnd_flags |= PND_INOPEN;
1515 
1516 	/*
1517 	 * Convert PUFFS mode to FUSE mode: convert FREAD/FWRITE
1518 	 * to O_RDONLY/O_WRONLY while perserving the other options.
1519 	 */
1520 	fmode = mode & ~(FREAD|FWRITE);
1521 	fmode |= (mode & FWRITE) ? O_RDWR : O_RDONLY;
1522 
1523 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*foi), pcr);
1524 	foi = GET_INPAYLOAD(ps, pm, fuse_open_in);
1525 	foi->flags = fmode;
1526 	foi->unused = 0;
1527 
1528 	if ((error = xchg_msg(pu, opc, pm, sizeof(*foo), wait_reply)) != 0)
1529 		goto out;
1530 
1531 	foo = GET_OUTPAYLOAD(ps, pm, fuse_open_out);
1532 
1533 	/*
1534 	 * Save the file handle in node private data
1535 	 * so that we can reuse it later
1536 	 */
1537 	perfuse_new_fh(opc, foo->fh, mode);
1538 
1539 #ifdef PERFUSE_DEBUG
1540 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1541 		DPRINTF("%s: opc = %p, file = \"%s\", "
1542 			"nodeid = 0x%"PRIx64", %s%sfh = 0x%"PRIx64"\n",
1543 			__func__, (void *)opc, perfuse_node_path(opc),
1544 			pnd->pnd_nodeid, mode & FREAD ? "r" : "",
1545 			mode & FWRITE ? "w" : "", foo->fh);
1546 #endif
1547 
1548 	ps->ps_destroy_msg(pm);
1549 out:
1550 
1551 	pnd->pnd_flags &= ~PND_INOPEN;
1552 	(void)dequeue_requests(ps, opc, PCQ_OPEN, DEQUEUE_ALL);
1553 
1554 	return error;
1555 }
1556 
1557 /* ARGSUSED0 */
1558 int
1559 perfuse_node_close(pu, opc, flags, pcr)
1560 	struct puffs_usermount *pu;
1561 	puffs_cookie_t opc;
1562 	int flags;
1563 	const struct puffs_cred *pcr;
1564 {
1565 	struct perfuse_node_data *pnd;
1566 
1567 	pnd = PERFUSE_NODE_DATA(opc);
1568 
1569 	if (!(pnd->pnd_flags & PND_OPEN))
1570 		return EBADF;
1571 
1572 	/*
1573 	 * Actual close is postponed at inactive time.
1574 	 */
1575 	return 0;
1576 }
1577 
1578 int
1579 perfuse_node_access(pu, opc, mode, pcr)
1580 	struct puffs_usermount *pu;
1581 	puffs_cookie_t opc;
1582 	int mode;
1583 	const struct puffs_cred *pcr;
1584 {
1585 	perfuse_msg_t *pm;
1586 	struct perfuse_state *ps;
1587 	struct fuse_access_in *fai;
1588 	int error;
1589 
1590 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1591 		return ENOENT;
1592 
1593 	/*
1594 	 * If we previously detected the filesystem does not
1595 	 * implement access(), short-circuit the call and skip
1596 	 * to libpuffs access() emulation.
1597 	 */
1598 	ps = puffs_getspecific(pu);
1599 	if (ps->ps_flags & PS_NO_ACCESS) {
1600 		const struct vattr *vap;
1601 
1602 		vap = puffs_pn_getvap((struct puffs_node *)opc);
1603 
1604 		error = puffs_access(IFTOVT(vap->va_mode),
1605 				     vap->va_mode & ACCESSPERMS,
1606 				     vap->va_uid, vap->va_gid,
1607 				     (mode_t)mode, pcr);
1608 		return error;
1609 	}
1610 
1611 	/*
1612 	 * Plain access call
1613 	 */
1614 	pm = ps->ps_new_msg(pu, opc, FUSE_ACCESS, sizeof(*fai), pcr);
1615 	fai = GET_INPAYLOAD(ps, pm, fuse_access_in);
1616 	fai->mask = 0;
1617 	fai->mask |= (mode & PUFFS_VREAD) ? R_OK : 0;
1618 	fai->mask |= (mode & PUFFS_VWRITE) ? W_OK : 0;
1619 	fai->mask |= (mode & PUFFS_VEXEC) ? X_OK : 0;
1620 
1621 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
1622 
1623 	ps->ps_destroy_msg(pm);
1624 
1625 	/*
1626 	 * If unimplemented, start over with emulation
1627 	 */
1628 	if (error == ENOSYS) {
1629 		ps->ps_flags |= PS_NO_ACCESS;
1630 		return perfuse_node_access(pu, opc, mode, pcr);
1631 	}
1632 
1633 	return error;
1634 }
1635 
1636 int
1637 perfuse_node_getattr(pu, opc, vap, pcr)
1638 	struct puffs_usermount *pu;
1639 	puffs_cookie_t opc;
1640 	struct vattr *vap;
1641 	const struct puffs_cred *pcr;
1642 {
1643 	perfuse_msg_t *pm = NULL;
1644 	struct perfuse_state *ps;
1645 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
1646 	struct fuse_getattr_in *fgi;
1647 	struct fuse_attr_out *fao;
1648 	int error = 0;
1649 
1650 	if (pnd->pnd_flags & PND_REMOVED)
1651 		return ENOENT;
1652 
1653 	/*
1654 	 * Serialize size access, see comment in perfuse_node_setattr().
1655 	 */
1656 	while (pnd->pnd_flags & PND_INRESIZE)
1657 		requeue_request(pu, opc, PCQ_RESIZE);
1658 	pnd->pnd_flags |= PND_INRESIZE;
1659 
1660 	ps = puffs_getspecific(pu);
1661 
1662 	/*
1663 	 * Check for cached attributes
1664 	 * This still require serialized access to size.
1665 	 */
1666 	if (!attr_expired(opc)) {
1667 		(void)memcpy(vap, puffs_pn_getvap((struct puffs_node *)opc),
1668 			     sizeof(*vap));
1669 		goto out;
1670 	}
1671 
1672 	/*
1673 	 * FUSE_GETATTR_FH must be set in fgi->flags
1674 	 * if we use for fgi->fh
1675 	 */
1676 	pm = ps->ps_new_msg(pu, opc, FUSE_GETATTR, sizeof(*fgi), pcr);
1677 	fgi = GET_INPAYLOAD(ps, pm, fuse_getattr_in);
1678 	fgi->getattr_flags = 0;
1679 	fgi->dummy = 0;
1680 	fgi->fh = 0;
1681 
1682 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_OPEN) {
1683 		fgi->fh = perfuse_get_fh(opc, FREAD);
1684 		fgi->getattr_flags |= FUSE_GETATTR_FH;
1685 	}
1686 
1687 #ifdef PERFUSE_DEBUG
1688 	if (perfuse_diagflags & PDF_RESIZE)
1689 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__, (void *)opc,
1690 		    vap->va_size);
1691 #endif
1692 
1693 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
1694 		goto out;
1695 
1696 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1697 
1698 #ifdef PERFUSE_DEBUG
1699 	if (perfuse_diagflags & PDF_RESIZE)
1700 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1701 		    (void *)opc, vap->va_size, fao->attr.size);
1702 #endif
1703 
1704 	/*
1705 	 * We set birthtime, flags, filerev,vaflags to 0.
1706 	 * This seems the best bet, since the information is
1707 	 * not available from filesystem.
1708 	 */
1709 	fuse_attr_to_vap(ps, vap, &fao->attr);
1710 	set_expire(opc, NULL, fao);
1711 
1712 	ps->ps_destroy_msg(pm);
1713 out:
1714 
1715 	pnd->pnd_flags &= ~PND_INRESIZE;
1716 	(void)dequeue_requests(ps, opc, PCQ_RESIZE, DEQUEUE_ALL);
1717 
1718 	return error;
1719 }
1720 
1721 int
1722 perfuse_node_setattr(pu, opc, vap, pcr)
1723 	struct puffs_usermount *pu;
1724 	puffs_cookie_t opc;
1725 	const struct vattr *vap;
1726 	const struct puffs_cred *pcr;
1727 {
1728 	perfuse_msg_t *pm;
1729 	uint64_t fh;
1730 	struct perfuse_state *ps;
1731 	struct perfuse_node_data *pnd;
1732 	struct fuse_setattr_in *fsi;
1733 	struct fuse_attr_out *fao;
1734 	struct vattr *old_va;
1735 	int error;
1736 #ifdef PERFUSE_DEBUG
1737 	struct vattr *old_vap;
1738 	int resize_debug = 0;
1739 #endif
1740 
1741 	ps = puffs_getspecific(pu);
1742 	pnd = PERFUSE_NODE_DATA(opc);
1743 
1744 	/*
1745 	 * The only operation we can do once the file is removed
1746 	 * is to resize it, and we can do it only if it is open.
1747 	 * Do not even send the operation to the filesystem: the
1748 	 * file is not there anymore.
1749 	 */
1750 	if (pnd->pnd_flags & PND_REMOVED) {
1751 		if (!(pnd->pnd_flags & PND_OPEN))
1752 			return ENOENT;
1753 
1754 		error = 0;
1755 		goto out;
1756 	}
1757 
1758 	old_va = puffs_pn_getvap((struct puffs_node *)opc);
1759 
1760 	/*
1761 	 * Check for permission to change size
1762 	 */
1763 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1764 	    (error = mode_access(opc, pcr, PUFFS_VWRITE)) != 0)
1765 		return error;
1766 
1767 	/*
1768 	 * Check for permission to change dates
1769 	 */
1770 	if (((vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1771 	     (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL)) &&
1772 	    (puffs_access_times(old_va->va_uid, old_va->va_gid,
1773 				old_va->va_mode, 0, pcr) != 0))
1774 		return EACCES;
1775 
1776 	/*
1777 	 * Check for permission to change owner and group
1778 	 */
1779 	if (((vap->va_uid != (uid_t)PUFFS_VNOVAL) ||
1780 	     (vap->va_gid != (gid_t)PUFFS_VNOVAL)) &&
1781 	    (puffs_access_chown(old_va->va_uid, old_va->va_gid,
1782 				vap->va_uid, vap->va_gid, pcr)) != 0)
1783 		return EACCES;
1784 
1785 	/*
1786 	 * Check for permission to change permissions
1787 	 */
1788 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1789 	    (puffs_access_chmod(old_va->va_uid, old_va->va_gid,
1790 				old_va->va_type, vap->va_mode, pcr)) != 0)
1791 		return EACCES;
1792 
1793 	pm = ps->ps_new_msg(pu, opc, FUSE_SETATTR, sizeof(*fsi), pcr);
1794 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
1795 	fsi->valid = 0;
1796 
1797 	/*
1798 	 * Get a fh if the node is open for writing
1799 	 */
1800 	if (pnd->pnd_flags & PND_WFH) {
1801 		fh = perfuse_get_fh(opc, FWRITE);
1802 		fsi->fh = fh;
1803 		fsi->valid |= FUSE_FATTR_FH;
1804 	}
1805 
1806 	if (vap->va_size != (u_quad_t)PUFFS_VNOVAL) {
1807 		fsi->size = vap->va_size;
1808 		fsi->valid |= FUSE_FATTR_SIZE;
1809 
1810 		/*
1811 		 * Serialize anything that can touch file size
1812 		 * to avoid reordered GETATTR and SETATTR.
1813 		 * Out of order SETATTR can report stale size,
1814 		 * which will cause the kernel to truncate the file.
1815 		 */
1816 		while (pnd->pnd_flags & PND_INRESIZE)
1817 			requeue_request(pu, opc, PCQ_RESIZE);
1818 		pnd->pnd_flags |= PND_INRESIZE;
1819 	}
1820 
1821 	/*
1822  	 * Setting mtime without atime or vice versa leads to
1823 	 * dates being reset to Epoch on glusterfs. If one
1824 	 * is missing, use the old value.
1825  	 */
1826 	if ((vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1827 	    (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL)) {
1828 
1829 		if (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) {
1830 			fsi->atime = vap->va_atime.tv_sec;
1831 			fsi->atimensec = (uint32_t)vap->va_atime.tv_nsec;
1832 		} else {
1833 			fsi->atime = old_va->va_atime.tv_sec;
1834 			fsi->atimensec = (uint32_t)old_va->va_atime.tv_nsec;
1835 		}
1836 
1837 		if (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) {
1838 			fsi->mtime = vap->va_mtime.tv_sec;
1839 			fsi->mtimensec = (uint32_t)vap->va_mtime.tv_nsec;
1840 		} else {
1841 			fsi->mtime = old_va->va_mtime.tv_sec;
1842 			fsi->mtimensec = (uint32_t)old_va->va_mtime.tv_nsec;
1843 		}
1844 
1845 		fsi->valid |= (FUSE_FATTR_MTIME|FUSE_FATTR_ATIME);
1846 	}
1847 
1848 	if (vap->va_mode != (mode_t)PUFFS_VNOVAL) {
1849 		fsi->mode = vap->va_mode;
1850 		fsi->valid |= FUSE_FATTR_MODE;
1851 	}
1852 
1853 	if (vap->va_uid != (uid_t)PUFFS_VNOVAL) {
1854 		fsi->uid = vap->va_uid;
1855 		fsi->valid |= FUSE_FATTR_UID;
1856 	}
1857 
1858 	if (vap->va_gid != (gid_t)PUFFS_VNOVAL) {
1859 		fsi->gid = vap->va_gid;
1860 		fsi->valid |= FUSE_FATTR_GID;
1861 	}
1862 
1863 	if (pnd->pnd_lock_owner != 0) {
1864 		fsi->lock_owner = pnd->pnd_lock_owner;
1865 		fsi->valid |= FUSE_FATTR_LOCKOWNER;
1866 	}
1867 
1868 	/*
1869 	 * ftruncate() sends only va_size, and metadata cache
1870 	 * flush adds va_atime and va_mtime. Some FUSE
1871 	 * filesystems will attempt to detect ftruncate by
1872 	 * checking for FATTR_SIZE being set without
1873 	 * FATTR_UID|FATTR_GID|FATTR_ATIME|FATTR_MTIME|FATTR_MODE
1874 	 *
1875 	 * Try to adapt and remove FATTR_ATIME|FATTR_MTIME
1876 	 * if we suspect a ftruncate().
1877 	 */
1878 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1879 	    ((vap->va_mode == (mode_t)PUFFS_VNOVAL) &&
1880 	     (vap->va_uid == (uid_t)PUFFS_VNOVAL) &&
1881 	     (vap->va_gid == (gid_t)PUFFS_VNOVAL))) {
1882 		fsi->atime = 0;
1883 		fsi->atimensec = 0;
1884 		fsi->mtime = 0;
1885 		fsi->mtimensec = 0;
1886 		fsi->valid &= ~(FUSE_FATTR_ATIME|FUSE_FATTR_MTIME);
1887 	}
1888 
1889 	/*
1890 	 * If nothing remain, discard the operation.
1891 	 */
1892 	if (!(fsi->valid & (FUSE_FATTR_SIZE|FUSE_FATTR_ATIME|FUSE_FATTR_MTIME|
1893 			    FUSE_FATTR_MODE|FUSE_FATTR_UID|FUSE_FATTR_GID))) {
1894 		error = 0;
1895 		goto out;
1896 	}
1897 
1898 #ifdef PERFUSE_DEBUG
1899 	old_vap = puffs_pn_getvap((struct puffs_node *)opc);
1900 
1901 	if ((perfuse_diagflags & PDF_RESIZE) &&
1902 	    (old_vap->va_size != (u_quad_t)PUFFS_VNOVAL)) {
1903 		resize_debug = 1;
1904 
1905 		DPRINTF(">> %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1906 		    (void *)opc,
1907 		    puffs_pn_getvap((struct puffs_node *)opc)->va_size,
1908 		    fsi->size);
1909 	}
1910 #endif
1911 
1912 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
1913 		goto out;
1914 
1915 	/*
1916 	 * Copy back the new values
1917 	 */
1918 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1919 
1920 #ifdef PERFUSE_DEBUG
1921 	if (resize_debug)
1922 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1923 		    (void *)opc, old_vap->va_size, fao->attr.size);
1924 #endif
1925 
1926 	fuse_attr_to_vap(ps, old_va, &fao->attr);
1927 	set_expire(opc, NULL, fao);
1928 
1929 	ps->ps_destroy_msg(pm);
1930 
1931 out:
1932 	if (pnd->pnd_flags & PND_INRESIZE) {
1933 		pnd->pnd_flags &= ~PND_INRESIZE;
1934 		(void)dequeue_requests(ps, opc, PCQ_RESIZE, DEQUEUE_ALL);
1935 	}
1936 
1937 	return error;
1938 }
1939 
1940 int
1941 perfuse_node_poll(pu, opc, events)
1942 	struct puffs_usermount *pu;
1943 	puffs_cookie_t opc;
1944 	int *events;
1945 {
1946 	struct perfuse_state *ps;
1947 	perfuse_msg_t *pm;
1948 	struct fuse_poll_in *fpi;
1949 	struct fuse_poll_out *fpo;
1950 	int error;
1951 
1952 	ps = puffs_getspecific(pu);
1953 	/*
1954 	 * kh is set if FUSE_POLL_SCHEDULE_NOTIFY is set.
1955 	 *
1956 	 * XXX ps_new_msg() is called with NULL creds, which will
1957 	 * be interpreted as FUSE superuser. We have no way to
1958 	 * know the requesting process' credential, but since poll
1959 	 * is supposed to operate on a file that has been open,
1960 	 * permission should have already been checked at open time.
1961 	 * That still may breaks on filesystems that provides odd
1962 	 * semantics.
1963  	 */
1964 	pm = ps->ps_new_msg(pu, opc, FUSE_POLL, sizeof(*fpi), NULL);
1965 	fpi = GET_INPAYLOAD(ps, pm, fuse_poll_in);
1966 	fpi->fh = perfuse_get_fh(opc, FREAD);
1967 	fpi->kh = 0;
1968 	fpi->flags = 0;
1969 
1970 #ifdef PERFUSE_DEBUG
1971 	if (perfuse_diagflags & PDF_FH)
1972 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
1973 			"fh = 0x%"PRIx64"\n", __func__, (void *)opc,
1974 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fpi->fh);
1975 #endif
1976 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fpo), wait_reply)) != 0)
1977 		return error;
1978 
1979 	fpo = GET_OUTPAYLOAD(ps, pm, fuse_poll_out);
1980 	*events = fpo->revents;
1981 
1982 	ps->ps_destroy_msg(pm);
1983 
1984 	return 0;
1985 }
1986 
1987 /* ARGSUSED0 */
1988 int
1989 perfuse_node_mmap(pu, opc, flags, pcr)
1990 	struct puffs_usermount *pu;
1991 	puffs_cookie_t opc;
1992 	int flags;
1993 	const struct puffs_cred *pcr;
1994 {
1995 	/*
1996 	 * Not implemented anymore in libfuse
1997 	 */
1998 	return ENOSYS;
1999 }
2000 
2001 /* ARGSUSED2 */
2002 int
2003 perfuse_node_fsync(pu, opc, pcr, flags, offlo, offhi)
2004 	struct puffs_usermount *pu;
2005 	puffs_cookie_t opc;
2006 	const struct puffs_cred *pcr;
2007 	int flags;
2008 	off_t offlo;
2009 	off_t offhi;
2010 {
2011 	int op;
2012 	perfuse_msg_t *pm;
2013 	struct perfuse_state *ps;
2014 	struct perfuse_node_data *pnd;
2015 	struct fuse_fsync_in *ffi;
2016 	uint64_t fh;
2017 	int error;
2018 
2019 	pm = NULL;
2020 	ps = puffs_getspecific(pu);
2021 	pnd = PERFUSE_NODE_DATA(opc);
2022 
2023 	/*
2024 	 * No need to sync a removed node
2025 	 */
2026 	if (pnd->pnd_flags & PND_REMOVED)
2027 		return 0;
2028 
2029 	/*
2030 	 * We do not sync closed files. They have been
2031 	 * sync at inactive time already.
2032 	 */
2033 	if (!(pnd->pnd_flags & PND_OPEN))
2034 		return 0;
2035 
2036 	if (puffs_pn_getvap((struct puffs_node *)opc)->va_type == VDIR)
2037 		op = FUSE_FSYNCDIR;
2038 	else 		/* VREG but also other types such as VLNK */
2039 		op = FUSE_FSYNC;
2040 
2041 	/*
2042 	 * Do not sync if there are no change to sync
2043 	 * XXX remove that test on files if we implement mmap
2044 	 */
2045 #ifdef PERFUSE_DEBUG
2046 	if (perfuse_diagflags & PDF_SYNC)
2047 		DPRINTF("%s: TEST opc = %p, file = \"%s\" is %sdirty\n",
2048 			__func__, (void*)opc, perfuse_node_path(opc),
2049 			pnd->pnd_flags & PND_DIRTY ? "" : "not ");
2050 #endif
2051 	if (!(pnd->pnd_flags & PND_DIRTY))
2052 		return 0;
2053 
2054 	/*
2055 	 * It seems NetBSD can call fsync without open first
2056 	 * glusterfs complain in such a situation:
2057 	 * "FSYNC() ERR => -1 (Invalid argument)"
2058 	 * The file will be closed at inactive time.
2059 	 *
2060 	 * We open the directory for reading in order to sync.
2061 	 * This sounds rather counterintuitive, but it works.
2062 	 */
2063 	if (!(pnd->pnd_flags & PND_WFH)) {
2064 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2065 			goto out;
2066 	}
2067 
2068 	if (op == FUSE_FSYNCDIR)
2069 		fh = perfuse_get_fh(opc, FREAD);
2070 	else
2071 		fh = perfuse_get_fh(opc, FWRITE);
2072 
2073 	/*
2074 	 * If fsync_flags  is set, meta data should not be flushed.
2075 	 */
2076 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*ffi), pcr);
2077 	ffi = GET_INPAYLOAD(ps, pm, fuse_fsync_in);
2078 	ffi->fh = fh;
2079 	ffi->fsync_flags = (flags & FFILESYNC) ? 0 : 1;
2080 
2081 #ifdef PERFUSE_DEBUG
2082 	if (perfuse_diagflags & PDF_FH)
2083 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
2084 			__func__, (void *)opc,
2085 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, ffi->fh);
2086 #endif
2087 
2088 	if ((error = xchg_msg(pu, opc, pm,
2089 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
2090 		goto out;
2091 
2092 	/*
2093 	 * No reply beyond fuse_out_header: nothing to do on success
2094 	 * just clear the dirty flag
2095 	 */
2096 	pnd->pnd_flags &= ~PND_DIRTY;
2097 
2098 #ifdef PERFUSE_DEBUG
2099 	if (perfuse_diagflags & PDF_SYNC)
2100 		DPRINTF("%s: CLEAR opc = %p, file = \"%s\"\n",
2101 			__func__, (void*)opc, perfuse_node_path(opc));
2102 #endif
2103 
2104 	ps->ps_destroy_msg(pm);
2105 
2106 out:
2107 	/*
2108 	 * ENOSYS is not returned to kernel,
2109 	 */
2110 	if (error == ENOSYS)
2111 		error = 0;
2112 
2113 	return error;
2114 }
2115 
2116 /* ARGSUSED0 */
2117 int
2118 perfuse_node_seek(pu, opc, oldoff, newoff,  pcr)
2119 	struct puffs_usermount *pu;
2120 	puffs_cookie_t opc;
2121 	off_t oldoff;
2122 	off_t newoff;
2123 	const struct puffs_cred *pcr;
2124 {
2125 	return 0;
2126 }
2127 
2128 int
2129 perfuse_node_remove(pu, opc, targ, pcn)
2130 	struct puffs_usermount *pu;
2131 	puffs_cookie_t opc;
2132 	puffs_cookie_t targ;
2133 	const struct puffs_cn *pcn;
2134 {
2135 	struct perfuse_state *ps;
2136 	struct perfuse_node_data *pnd;
2137 	perfuse_msg_t *pm;
2138 	char *path;
2139 	const char *name;
2140 	size_t len;
2141 	int error;
2142 
2143 	pnd = PERFUSE_NODE_DATA(opc);
2144 
2145 	if ((pnd->pnd_flags & PND_REMOVED) ||
2146 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
2147 		return ENOENT;
2148 
2149 #ifdef PERFUSE_DEBUG
2150 	if (targ == NULL)
2151 		DERRX(EX_SOFTWARE, "%s: targ is NULL", __func__);
2152 
2153 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
2154 		DPRINTF("%s: opc = %p, remove opc = %p, file = \"%s\"\n",
2155 			__func__, (void *)opc, (void *)targ, pcn->pcn_name);
2156 #endif
2157 	/*
2158 	 * Await for all operations on the deleted node to drain,
2159 	 * as the filesystem may be confused to have it deleted
2160 	 * during a getattr
2161 	 */
2162 	while (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_INXCHG)
2163 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2164 
2165 	ps = puffs_getspecific(pu);
2166 	pnd = PERFUSE_NODE_DATA(opc);
2167 	name = pcn->pcn_name;
2168 	len = pcn->pcn_namelen + 1;
2169 
2170 	pm = ps->ps_new_msg(pu, opc, FUSE_UNLINK, len, pcn->pcn_cred);
2171 	path = _GET_INPAYLOAD(ps, pm, char *);
2172 	(void)strlcpy(path, name, len);
2173 
2174 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2175 		return error;
2176 
2177 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2178 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2179 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2180 
2181 	/*
2182 	 * The parent directory needs a sync
2183 	 */
2184 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2185 
2186 #ifdef PERFUSE_DEBUG
2187 	if (perfuse_diagflags & PDF_FILENAME)
2188 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
2189 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
2190 			pcn->pcn_name);
2191 #endif
2192 	ps->ps_destroy_msg(pm);
2193 
2194 	return 0;
2195 }
2196 
2197 int
2198 perfuse_node_link(pu, opc, targ, pcn)
2199 	struct puffs_usermount *pu;
2200 	puffs_cookie_t opc;
2201 	puffs_cookie_t targ;
2202 	const struct puffs_cn *pcn;
2203 {
2204 	struct perfuse_state *ps;
2205 	perfuse_msg_t *pm;
2206 	const char *name;
2207 	size_t len;
2208 	struct puffs_node *pn;
2209 	struct fuse_link_in *fli;
2210 	int error;
2211 
2212 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2213 		return ENOENT;
2214 
2215 	ps = puffs_getspecific(pu);
2216 	pn = (struct puffs_node *)targ;
2217 	name = pcn->pcn_name;
2218 	len =  sizeof(*fli) + pcn->pcn_namelen + 1;
2219 
2220 	pm = ps->ps_new_msg(pu, opc, FUSE_LINK, len, pcn->pcn_cred);
2221 	fli = GET_INPAYLOAD(ps, pm, fuse_link_in);
2222 	fli->oldnodeid = PERFUSE_NODE_DATA(pn)->pnd_nodeid;
2223 	(void)strlcpy((char *)(void *)(fli + 1), name, len - sizeof(*fli));
2224 
2225 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2226 		return error;
2227 
2228 	ps->ps_destroy_msg(pm);
2229 
2230 	return 0;
2231 }
2232 
2233 int
2234 perfuse_node_rename(pu, opc, src, pcn_src, targ_dir, targ, pcn_targ)
2235 	struct puffs_usermount *pu;
2236 	puffs_cookie_t opc;
2237 	puffs_cookie_t src;
2238 	const struct puffs_cn *pcn_src;
2239 	puffs_cookie_t targ_dir;
2240 	puffs_cookie_t targ;
2241 	const struct puffs_cn *pcn_targ;
2242 {
2243 	struct perfuse_state *ps;
2244 	perfuse_msg_t *pm;
2245 	struct fuse_rename_in *fri;
2246 	const char *newname;
2247 	const char *oldname;
2248 	char *np;
2249 	int error;
2250 	size_t len;
2251 	size_t newname_len;
2252 	size_t oldname_len;
2253 
2254 	if ((PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED) ||
2255 	    (PERFUSE_NODE_DATA(src)->pnd_flags & PND_REMOVED) ||
2256 	    (PERFUSE_NODE_DATA(targ_dir)->pnd_flags & PND_REMOVED))
2257 		return ENOENT;
2258 
2259 	/*
2260 	 * Await for all operations on the deleted node to drain,
2261 	 * as the filesystem may be confused to have it deleted
2262 	 * during a getattr
2263 	 */
2264 	if ((struct puffs_node *)targ != NULL) {
2265 		while (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_INXCHG)
2266 			requeue_request(pu, targ, PCQ_AFTERXCHG);
2267 	} else {
2268 		while (PERFUSE_NODE_DATA(src)->pnd_flags & PND_INXCHG)
2269 			requeue_request(pu, src, PCQ_AFTERXCHG);
2270 	}
2271 
2272 	ps = puffs_getspecific(pu);
2273 	newname =  pcn_targ->pcn_name;
2274 	newname_len = pcn_targ->pcn_namelen + 1;
2275 	oldname =  pcn_src->pcn_name;
2276 	oldname_len = pcn_src->pcn_namelen + 1;
2277 
2278 	len = sizeof(*fri) + oldname_len + newname_len;
2279 	pm = ps->ps_new_msg(pu, opc, FUSE_RENAME, len, pcn_targ->pcn_cred);
2280 	fri = GET_INPAYLOAD(ps, pm, fuse_rename_in);
2281 	fri->newdir = PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid;
2282 	np = (char *)(void *)(fri + 1);
2283 	(void)strlcpy(np, oldname, oldname_len);
2284 	np += oldname_len;
2285 	(void)strlcpy(np, newname, newname_len);
2286 
2287 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2288 		return error;
2289 
2290 	if (opc != targ_dir) {
2291 		struct perfuse_node_data *srcdir_pnd;
2292 		struct perfuse_node_data *dstdir_pnd;
2293 		struct perfuse_node_data *src_pnd;
2294 
2295 		srcdir_pnd = PERFUSE_NODE_DATA(opc);
2296 		dstdir_pnd = PERFUSE_NODE_DATA(targ_dir);
2297 		src_pnd = PERFUSE_NODE_DATA(src);
2298 
2299 		TAILQ_REMOVE(&srcdir_pnd->pnd_children, src_pnd, pnd_next);
2300 		TAILQ_INSERT_TAIL(&dstdir_pnd->pnd_children, src_pnd, pnd_next);
2301 
2302 		srcdir_pnd->pnd_childcount--;
2303 		dstdir_pnd->pnd_childcount++;
2304 
2305 		src_pnd->pnd_parent = targ_dir;
2306 
2307 		PERFUSE_NODE_DATA(targ_dir)->pnd_flags |= PND_DIRTY;
2308 	}
2309 
2310 	(void)strlcpy(PERFUSE_NODE_DATA(src)->pnd_name, newname, MAXPATHLEN);
2311 
2312 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2313 
2314 	if ((struct puffs_node *)targ != NULL)
2315 		PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2316 
2317 #ifdef PERFUSE_DEBUG
2318 	if (perfuse_diagflags & PDF_FILENAME)
2319 		DPRINTF("%s: nodeid = 0x%"PRIx64" file = \"%s\" renamed \"%s\" "
2320 			"nodeid = 0x%"PRIx64" -> nodeid = 0x%"PRIx64" \"%s\"\n",
2321 	 		__func__, PERFUSE_NODE_DATA(src)->pnd_nodeid,
2322 			pcn_src->pcn_name, pcn_targ->pcn_name,
2323 			PERFUSE_NODE_DATA(opc)->pnd_nodeid,
2324 			PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid,
2325 			perfuse_node_path(targ_dir));
2326 #endif
2327 
2328 	ps->ps_destroy_msg(pm);
2329 
2330 	return 0;
2331 }
2332 
2333 int
2334 perfuse_node_mkdir(pu, opc, pni, pcn, vap)
2335 	struct puffs_usermount *pu;
2336 	puffs_cookie_t opc;
2337 	struct puffs_newinfo *pni;
2338 	const struct puffs_cn *pcn;
2339 	const struct vattr *vap;
2340 {
2341 	struct perfuse_state *ps;
2342 	perfuse_msg_t *pm;
2343 	struct fuse_mkdir_in *fmi;
2344 	const char *path;
2345 	size_t len;
2346 
2347 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2348 		return ENOENT;
2349 
2350 	ps = puffs_getspecific(pu);
2351 	path = pcn->pcn_name;
2352 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
2353 
2354 	pm = ps->ps_new_msg(pu, opc, FUSE_MKDIR, len, pcn->pcn_cred);
2355 	fmi = GET_INPAYLOAD(ps, pm, fuse_mkdir_in);
2356 	fmi->mode = vap->va_mode;
2357 	fmi->umask = 0; 	/* Seems unused by libfuse? */
2358 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
2359 
2360 	return node_mk_common(pu, opc, pni, pcn, pm);
2361 }
2362 
2363 
2364 int
2365 perfuse_node_rmdir(pu, opc, targ, pcn)
2366 	struct puffs_usermount *pu;
2367 	puffs_cookie_t opc;
2368 	puffs_cookie_t targ;
2369 	const struct puffs_cn *pcn;
2370 {
2371 	struct perfuse_state *ps;
2372 	struct perfuse_node_data *pnd;
2373 	perfuse_msg_t *pm;
2374 	char *path;
2375 	const char *name;
2376 	size_t len;
2377 	int error;
2378 
2379 	pnd = PERFUSE_NODE_DATA(opc);
2380 
2381 	if ((pnd->pnd_flags & PND_REMOVED) ||
2382 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
2383 		return ENOENT;
2384 
2385 	/*
2386 	 * Await for all operations on the deleted node to drain,
2387 	 * as the filesystem may be confused to have it deleted
2388 	 * during a getattr
2389 	 */
2390 	while (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_INXCHG)
2391 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2392 
2393 	ps = puffs_getspecific(pu);
2394 	name = pcn->pcn_name;
2395 	len = pcn->pcn_namelen + 1;
2396 
2397 	pm = ps->ps_new_msg(pu, opc, FUSE_RMDIR, len, pcn->pcn_cred);
2398 	path = _GET_INPAYLOAD(ps, pm, char *);
2399 	(void)strlcpy(path, name, len);
2400 
2401 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2402 		return error;
2403 
2404 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2405 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2406 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2407 
2408 	/*
2409 	 * The parent directory needs a sync
2410 	 */
2411 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2412 
2413 #ifdef PERFUSE_DEBUG
2414 	if (perfuse_diagflags & PDF_FILENAME)
2415 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
2416 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
2417 			perfuse_node_path(targ));
2418 #endif
2419 	ps->ps_destroy_msg(pm);
2420 
2421 	return 0;
2422 }
2423 
2424 /* vap is unused */
2425 /* ARGSUSED4 */
2426 int
2427 perfuse_node_symlink(pu, opc, pni, pcn_src, vap, link_target)
2428 	struct puffs_usermount *pu;
2429 	puffs_cookie_t opc;
2430 	struct puffs_newinfo *pni;
2431 	const struct puffs_cn *pcn_src;
2432 	const struct vattr *vap;
2433 	const char *link_target;
2434 {
2435 	struct perfuse_state *ps;
2436 	perfuse_msg_t *pm;
2437 	char *np;
2438 	const char *path;
2439 	size_t path_len;
2440 	size_t linkname_len;
2441 	size_t len;
2442 
2443 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2444 		return ENOENT;
2445 
2446 	ps = puffs_getspecific(pu);
2447 	path = pcn_src->pcn_name;
2448 	path_len = pcn_src->pcn_namelen + 1;
2449 	linkname_len = strlen(link_target) + 1;
2450 	len = path_len + linkname_len;
2451 
2452 	pm = ps->ps_new_msg(pu, opc, FUSE_SYMLINK, len, pcn_src->pcn_cred);
2453 	np = _GET_INPAYLOAD(ps, pm, char *);
2454 	(void)strlcpy(np, path, path_len);
2455 	np += path_len;
2456 	(void)strlcpy(np, link_target, linkname_len);
2457 
2458 	return node_mk_common(pu, opc, pni, pcn_src, pm);
2459 }
2460 
2461 /* ARGSUSED4 */
2462 int
2463 perfuse_node_readdir(pu, opc, dent, readoff,
2464 		     reslen, pcr, eofflag, cookies, ncookies)
2465 	struct puffs_usermount *pu;
2466 	puffs_cookie_t opc;
2467 	struct dirent *dent;
2468 	off_t *readoff;
2469 	size_t *reslen;
2470 	const struct puffs_cred *pcr;
2471 	int *eofflag;
2472 	off_t *cookies;
2473 	size_t *ncookies;
2474 {
2475 	perfuse_msg_t *pm;
2476 	uint64_t fh;
2477 	struct perfuse_state *ps;
2478 	struct perfuse_node_data *pnd;
2479 	struct fuse_read_in *fri;
2480 	struct fuse_out_header *foh;
2481 	struct fuse_dirent *fd;
2482 	size_t foh_len;
2483 	int error;
2484 	size_t fd_maxlen;
2485 
2486 	error = 0;
2487 	ps = puffs_getspecific(pu);
2488 
2489 	/*
2490 	 * readdir state is kept at node level, and several readdir
2491 	 * requests can be issued at the same time on the same node.
2492 	 * We need to queue requests so that only one is in readdir
2493 	 * code at the same time.
2494 	 */
2495 	pnd = PERFUSE_NODE_DATA(opc);
2496 	while (pnd->pnd_flags & PND_INREADDIR)
2497 		requeue_request(pu, opc, PCQ_READDIR);
2498 	pnd->pnd_flags |= PND_INREADDIR;
2499 
2500 #ifdef PERFUSE_DEBUG
2501 	if (perfuse_diagflags & PDF_READDIR)
2502 		DPRINTF("%s: READDIR opc = %p enter critical section\n",
2503 			__func__, (void *)opc);
2504 #endif
2505 	/*
2506 	 * Re-initialize pnd->pnd_fd_cookie on the first readdir for a node
2507 	 */
2508 	if (*readoff == 0)
2509 		pnd->pnd_fd_cookie = 0;
2510 
2511 	/*
2512 	 * Do we already have the data bufered?
2513 	 */
2514 	if (pnd->pnd_dirent != NULL)
2515 		goto out;
2516 	pnd->pnd_dirent_len = 0;
2517 
2518 	/*
2519 	 * It seems NetBSD can call readdir without open first
2520 	 * libfuse will crash if it is done that way, hence open first.
2521 	 */
2522 	if (!(pnd->pnd_flags & PND_OPEN)) {
2523 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2524 			goto out;
2525 	}
2526 
2527 	fh = perfuse_get_fh(opc, FREAD);
2528 
2529 #ifdef PERFUSE_DEBUG
2530 	if (perfuse_diagflags & PDF_FH)
2531 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
2532 			"rfh = 0x%"PRIx64"\n", __func__, (void *)opc,
2533 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fh);
2534 #endif
2535 
2536 	pnd->pnd_all_fd = NULL;
2537 	pnd->pnd_all_fd_len = 0;
2538 	fd_maxlen = ps->ps_max_readahead - sizeof(*foh);
2539 
2540 	do {
2541 		size_t fd_len;
2542 		char *afdp;
2543 
2544 		pm = ps->ps_new_msg(pu, opc, FUSE_READDIR, sizeof(*fri), pcr);
2545 
2546 		/*
2547 		 * read_flags, lock_owner and flags are unused in libfuse
2548 		 */
2549 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
2550 		fri->fh = fh;
2551 		fri->offset = pnd->pnd_fd_cookie;
2552 		fri->size = (uint32_t)fd_maxlen;
2553 		fri->read_flags = 0;
2554 		fri->lock_owner = 0;
2555 		fri->flags = 0;
2556 
2557 		if ((error = xchg_msg(pu, opc, pm,
2558 				      UNSPEC_REPLY_LEN, wait_reply)) != 0)
2559 			goto out;
2560 
2561 		/*
2562 		 * There are many puffs_framebufs calls later,
2563 		 * therefore foh will not be valid for a long time.
2564 		 * Just get the length and forget it.
2565 		 */
2566 		foh = GET_OUTHDR(ps, pm);
2567 		foh_len = foh->len;
2568 
2569 		/*
2570 		 * Empty read: we reached the end of the buffer.
2571 		 */
2572 		if (foh_len == sizeof(*foh)) {
2573 			ps->ps_destroy_msg(pm);
2574 			*eofflag = 1;
2575 			break;
2576 		}
2577 
2578 		/*
2579 		 * Check for corrupted message.
2580 		 */
2581 		if (foh_len < sizeof(*foh) + sizeof(*fd)) {
2582 			ps->ps_destroy_msg(pm);
2583 			DWARNX("readdir reply too short");
2584 			error = EIO;
2585 			goto out;
2586 		}
2587 
2588 
2589 		fd = GET_OUTPAYLOAD(ps, pm, fuse_dirent);
2590 		fd_len = foh_len - sizeof(*foh);
2591 
2592 		pnd->pnd_all_fd = realloc(pnd->pnd_all_fd,
2593 					  pnd->pnd_all_fd_len + fd_len);
2594 		if (pnd->pnd_all_fd  == NULL)
2595 			DERR(EX_OSERR, "%s: malloc failed", __func__);
2596 
2597 		afdp = (char *)(void *)pnd->pnd_all_fd + pnd->pnd_all_fd_len;
2598 		(void)memcpy(afdp, fd, fd_len);
2599 
2600 		pnd->pnd_all_fd_len += fd_len;
2601 
2602 		/*
2603 		 * The fd->off field is used as a cookie for
2604 		 * resuming the next readdir() where this one was left.
2605 	 	 */
2606 		pnd->pnd_fd_cookie = readdir_last_cookie(fd, fd_len);
2607 
2608 		ps->ps_destroy_msg(pm);
2609 	} while (1 /* CONSTCOND */);
2610 
2611 	if (pnd->pnd_all_fd != NULL) {
2612 		if (fuse_to_dirent(pu, opc, pnd->pnd_all_fd,
2613 				   pnd->pnd_all_fd_len) == -1)
2614 			error = EIO;
2615 	}
2616 
2617 out:
2618 	if (pnd->pnd_all_fd != NULL) {
2619 		free(pnd->pnd_all_fd);
2620 		pnd->pnd_all_fd = NULL;
2621 		pnd->pnd_all_fd_len = 0;
2622 	}
2623 
2624 	if (error == 0)
2625 		error = readdir_buffered(opc, dent, readoff, reslen);
2626 
2627 	/*
2628 	 * Schedule queued readdir requests
2629 	 */
2630 	pnd->pnd_flags &= ~PND_INREADDIR;
2631 	(void)dequeue_requests(ps, opc, PCQ_READDIR, DEQUEUE_ALL);
2632 
2633 #ifdef PERFUSE_DEBUG
2634 	if (perfuse_diagflags & PDF_READDIR)
2635 		DPRINTF("%s: READDIR opc = %p exit critical section\n",
2636 			__func__, (void *)opc);
2637 #endif
2638 
2639 	return error;
2640 }
2641 
2642 int
2643 perfuse_node_readlink(pu, opc, pcr, linkname, linklen)
2644 	struct puffs_usermount *pu;
2645 	puffs_cookie_t opc;
2646 	const struct puffs_cred *pcr;
2647 	char *linkname;
2648 	size_t *linklen;
2649 {
2650 	struct perfuse_state *ps;
2651 	perfuse_msg_t *pm;
2652 	int error;
2653 	size_t len;
2654 	struct fuse_out_header *foh;
2655 
2656 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2657 		return ENOENT;
2658 
2659 	ps = puffs_getspecific(pu);
2660 
2661 	pm = ps->ps_new_msg(pu, opc, FUSE_READLINK, 0, pcr);
2662 
2663 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2664 		return error;
2665 
2666 	foh = GET_OUTHDR(ps, pm);
2667 	len = foh->len - sizeof(*foh);
2668 	if (len > *linklen)
2669 		DERRX(EX_PROTOCOL, "path len = %zd too long", len);
2670 	if (len == 0)
2671 		DERRX(EX_PROTOCOL, "path len = %zd too short", len);
2672 
2673 	/*
2674 	 * FUSE filesystems return a NUL terminated string, we
2675 	 * do not want to trailing \0
2676 	 */
2677 	*linklen = len - 1;
2678 	(void)memcpy(linkname, _GET_OUTPAYLOAD(ps, pm, char *), len);
2679 
2680 	ps->ps_destroy_msg(pm);
2681 
2682 	return 0;
2683 }
2684 
2685 int
2686 perfuse_node_reclaim(pu, opc)
2687 	struct puffs_usermount *pu;
2688 	puffs_cookie_t opc;
2689 {
2690 	struct perfuse_state *ps;
2691 	perfuse_msg_t *pm;
2692 	struct perfuse_node_data *pnd;
2693 	struct fuse_forget_in *ffi;
2694 	struct puffs_node *pn;
2695 	struct puffs_node *pn_root;
2696 
2697 	ps = puffs_getspecific(pu);
2698 	pnd = PERFUSE_NODE_DATA(opc);
2699 
2700 	/*
2701 	 * Never forget the root.
2702 	 */
2703 	if (pnd->pnd_nodeid == FUSE_ROOT_ID)
2704 		return 0;
2705 
2706 	pnd->pnd_flags |= PND_RECLAIMED;
2707 	pnd->pnd_puffs_nlookup--;
2708 
2709 #ifdef PERFUSE_DEBUG
2710 	if (perfuse_diagflags & PDF_RECLAIM)
2711 		DPRINTF("%s (nodeid %"PRId64") reclaimed\n",
2712 			perfuse_node_path(opc), pnd->pnd_nodeid);
2713 #endif
2714 
2715 	pn_root = puffs_getroot(pu);
2716 	pn = (struct puffs_node *)opc;
2717 	while (pn != pn_root) {
2718 		struct puffs_node *parent_pn;
2719 
2720 		pnd = PERFUSE_NODE_DATA(pn);
2721 
2722 #ifdef PERFUSE_DEBUG
2723 	if (perfuse_diagflags & PDF_RECLAIM)
2724 		DPRINTF("%s (nodeid %"PRId64") is %sreclaimed, nlookup = %d "
2725 			"has childcount %d %s%s%s%s, pending ops:%s%s%s\n",
2726 		        perfuse_node_path((puffs_cookie_t)pn), pnd->pnd_nodeid,
2727 		        pnd->pnd_flags & PND_RECLAIMED ? "" : "not ",
2728 			pnd->pnd_puffs_nlookup, pnd->pnd_childcount,
2729 			pnd->pnd_flags & PND_OPEN ? "open " : "not open",
2730 			pnd->pnd_flags & PND_RFH ? "r" : "",
2731 			pnd->pnd_flags & PND_WFH ? "w" : "",
2732 			pnd->pnd_flags & PND_BUSY ? "" : " none",
2733 			pnd->pnd_flags & PND_INREADDIR ? " readdir" : "",
2734 			pnd->pnd_flags & PND_INWRITE ? " write" : "",
2735 			pnd->pnd_flags & PND_INOPEN ? " open" : "");
2736 #endif
2737 		if (!(pnd->pnd_flags & PND_RECLAIMED) ||
2738 		    (pnd->pnd_puffs_nlookup != 0) ||
2739 		    (pnd->pnd_childcount != 0))
2740 			return 0;
2741 
2742 #ifdef PERFUSE_DEBUG
2743 		if ((pnd->pnd_flags & PND_OPEN) ||
2744 		       !TAILQ_EMPTY(&pnd->pnd_pcq))
2745 			DERRX(EX_SOFTWARE, "%s: opc = %p: still open",
2746 			      __func__, (void *)opc);
2747 
2748 		if ((pnd->pnd_flags & PND_BUSY) ||
2749 		       !TAILQ_EMPTY(&pnd->pnd_pcq))
2750 			DERRX(EX_SOFTWARE, "%s: opc = %p: ongoing operations",
2751 			      __func__, (void *)opc);
2752 #endif
2753 
2754 		/*
2755 		 * Send the FORGET message
2756 		 *
2757 		 * ps_new_msg() is called with NULL creds, which will
2758 		 * be interpreted as FUSE superuser. This is obviously
2759 		 * fine since we operate with kernel creds here.
2760 		 */
2761 		pm = ps->ps_new_msg(pu, (puffs_cookie_t)pn, FUSE_FORGET,
2762 			      sizeof(*ffi), NULL);
2763 		ffi = GET_INPAYLOAD(ps, pm, fuse_forget_in);
2764 		ffi->nlookup = pnd->pnd_fuse_nlookup;
2765 
2766 		/*
2767 		 * No reply is expected, pm is freed in xchg_msg
2768 		 */
2769 		(void)xchg_msg(pu, (puffs_cookie_t)pn,
2770 			       pm, UNSPEC_REPLY_LEN, no_reply);
2771 
2772 		parent_pn = pnd->pnd_parent;
2773 
2774 		perfuse_destroy_pn(pn);
2775 
2776 		pn = parent_pn;
2777 	}
2778 
2779 	return 0;
2780 }
2781 
2782 int
2783 perfuse_node_inactive(pu, opc)
2784 	struct puffs_usermount *pu;
2785 	puffs_cookie_t opc;
2786 {
2787 	struct perfuse_state *ps;
2788 	struct perfuse_node_data *pnd;
2789 
2790 	ps = puffs_getspecific(pu);
2791 	pnd = PERFUSE_NODE_DATA(opc);
2792 
2793 	if (!(pnd->pnd_flags & (PND_OPEN|PND_REMOVED)))
2794 		return 0;
2795 
2796 	/*
2797 	 * Make sure all operation are finished
2798 	 * There can be an ongoing write. Other
2799 	 * operation wait for all data before
2800 	 * the close/inactive.
2801 	 */
2802 	while (pnd->pnd_flags & PND_INWRITE)
2803 		requeue_request(pu, opc, PCQ_AFTERWRITE);
2804 
2805 	/*
2806 	 * The inactive operation may be cancelled.
2807 	 * If no open is in progress, set PND_INOPEN
2808 	 * so that a new open will be queued.
2809 	 */
2810 	if (pnd->pnd_flags & PND_INOPEN)
2811 		return 0;
2812 
2813 	pnd->pnd_flags |= PND_INOPEN;
2814 
2815 	/*
2816 	 * Sync data
2817 	 */
2818 	if (pnd->pnd_flags & PND_DIRTY)
2819 		(void)perfuse_node_fsync(pu, opc, NULL, 0, 0, 0);
2820 
2821 	/*
2822 	 * Close handles
2823 	 */
2824 	if (pnd->pnd_flags & PND_WFH)
2825 		(void)perfuse_node_close_common(pu, opc, FWRITE);
2826 
2827 	if (pnd->pnd_flags & PND_RFH)
2828 		(void)perfuse_node_close_common(pu, opc, FREAD);
2829 
2830 	/*
2831 	 * This will cause a reclaim to be sent
2832 	 */
2833 	if (pnd->pnd_flags & PND_REMOVED)
2834 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N1);
2835 
2836 	/*
2837 	 * Schedule awaiting operations
2838 	 */
2839 	pnd->pnd_flags &= ~PND_INOPEN;
2840 	(void)dequeue_requests(ps, opc, PCQ_OPEN, DEQUEUE_ALL);
2841 
2842 	return 0;
2843 }
2844 
2845 
2846 /* ARGSUSED0 */
2847 int
2848 perfuse_node_print(pu, opc)
2849 	struct puffs_usermount *pu;
2850 	puffs_cookie_t opc;
2851 {
2852 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2853 	return 0;
2854 }
2855 
2856 /* ARGSUSED0 */
2857 int
2858 perfuse_node_pathconf(pu, opc, name, retval)
2859 	struct puffs_usermount *pu;
2860 	puffs_cookie_t opc;
2861 	int name;
2862 	int *retval;
2863 {
2864 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2865 	return 0;
2866 }
2867 
2868 int
2869 perfuse_node_advlock(pu, opc, id, op, fl, flags)
2870 	struct puffs_usermount *pu;
2871 	puffs_cookie_t opc;
2872 	void *id;
2873 	int op;
2874 	struct flock *fl;
2875 	int flags;
2876 {
2877 	struct perfuse_state *ps;
2878 	int fop;
2879 	perfuse_msg_t *pm;
2880 	uint64_t fh;
2881 	struct fuse_lk_in *fli;
2882 	struct fuse_out_header *foh;
2883 	struct fuse_lk_out *flo;
2884 	uint32_t owner;
2885 	size_t len;
2886 	int error;
2887 
2888 	/*
2889 	 * Make sure we do have a filehandle, as the FUSE filesystem
2890 	 * expect one. E.g.: if we provide none, GlusterFS logs an error
2891 	 * "0-glusterfs-fuse: xl is NULL"
2892 	 *
2893 	 * We need the read file handle if the file is open read only,
2894 	 * in order to support shared locks on read-only files.
2895 	 * NB: The kernel always sends advlock for read-only
2896 	 * files at exit time when the process used lock, see
2897 	 * sys_exit -> exit1 -> fd_free -> fd_close -> VOP_ADVLOCK
2898 	 */
2899 	if ((fh = perfuse_get_fh(opc, FREAD)) == FUSE_UNKNOWN_FH)
2900 		return EBADF;
2901 
2902 	ps = puffs_getspecific(pu);
2903 
2904 	if (op == F_GETLK)
2905 		fop = FUSE_GETLK;
2906 	else
2907 		fop = (flags & F_WAIT) ? FUSE_SETLKW : FUSE_SETLK;
2908 
2909 	/*
2910 	 * XXX ps_new_msg() is called with NULL creds, which will
2911 	 * be interpreted as FUSE superuser. We have no way to
2912 	 * know the requesting process' credential, but since advlock()
2913 	 * is supposed to operate on a file that has been open(),
2914 	 * permission should have already been checked at open() time.
2915 	 */
2916 	pm = ps->ps_new_msg(pu, opc, fop, sizeof(*fli), NULL);
2917 	fli = GET_INPAYLOAD(ps, pm, fuse_lk_in);
2918 	fli->fh = fh;
2919 	fli->owner = (uint64_t)(vaddr_t)id;
2920 	fli->lk.start = fl->l_start;
2921 	fli->lk.end = fl->l_start + fl->l_len;
2922 	fli->lk.type = fl->l_type;
2923 	fli->lk.pid = fl->l_pid;
2924 	fli->lk_flags = (flags & F_FLOCK) ? FUSE_LK_FLOCK : 0;
2925 
2926 	owner = (uint32_t)(vaddr_t)id;
2927 
2928 #ifdef PERFUSE_DEBUG
2929 	if (perfuse_diagflags & PDF_FH)
2930 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
2931 			__func__, (void *)opc,
2932 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fli->fh);
2933 #endif
2934 
2935 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2936 		return error;
2937 
2938 	foh = GET_OUTHDR(ps, pm);
2939 	len = foh->len - sizeof(*foh);
2940 
2941 	/*
2942 	 * Save or clear the lock
2943 	 */
2944 	switch (op) {
2945 	case F_GETLK:
2946 		if (len != sizeof(*flo))
2947 			DERRX(EX_SOFTWARE,
2948 			      "%s: Unexpected lock reply len %zd",
2949 			      __func__, len);
2950 
2951 		flo = GET_OUTPAYLOAD(ps, pm, fuse_lk_out);
2952 		fl->l_start = flo->lk.start;
2953 		fl->l_len = flo->lk.end - flo->lk.start;
2954 		fl->l_pid = flo->lk.pid;
2955 		fl->l_type = flo->lk.type;
2956 		fl->l_whence = SEEK_SET;	/* libfuse hardcodes it */
2957 
2958 		PERFUSE_NODE_DATA(opc)->pnd_lock_owner = flo->lk.pid;
2959 		break;
2960 	case F_UNLCK:
2961 		owner = 0;
2962 		/* FALLTHROUGH */
2963 	case F_SETLK:
2964 		/* FALLTHROUGH */
2965 	case F_SETLKW:
2966 		if (error != 0)
2967 			PERFUSE_NODE_DATA(opc)->pnd_lock_owner = owner;
2968 
2969 		if (len != 0)
2970 			DERRX(EX_SOFTWARE,
2971 			      "%s: Unexpected unlock reply len %zd",
2972 			      __func__, len);
2973 
2974 		break;
2975 	default:
2976 		DERRX(EX_SOFTWARE, "%s: Unexpected op %d", __func__, op);
2977 		break;
2978 	}
2979 
2980 	ps->ps_destroy_msg(pm);
2981 
2982 	return 0;
2983 }
2984 
2985 int
2986 perfuse_node_read(pu, opc, buf, offset, resid, pcr, ioflag)
2987 	struct puffs_usermount *pu;
2988 	puffs_cookie_t opc;
2989 	uint8_t *buf;
2990 	off_t offset;
2991 	size_t *resid;
2992 	const struct puffs_cred *pcr;
2993 	int ioflag;
2994 {
2995 	struct perfuse_state *ps;
2996 	struct perfuse_node_data *pnd;
2997 	const struct vattr *vap;
2998 	perfuse_msg_t *pm;
2999 	struct fuse_read_in *fri;
3000 	struct fuse_out_header *foh;
3001 	size_t readen;
3002 	int error;
3003 
3004 	ps = puffs_getspecific(pu);
3005 	pnd = PERFUSE_NODE_DATA(opc);
3006 	vap = puffs_pn_getvap((struct puffs_node *)opc);
3007 
3008 	/*
3009 	 * NetBSD turns that into a getdents(2) output
3010 	 * We just do a EISDIR as this feature is of little use.
3011 	 */
3012 	if (vap->va_type == VDIR)
3013 		return EISDIR;
3014 
3015 	if ((u_quad_t)offset + *resid > vap->va_size)
3016 		DWARNX("%s %p read %lld@%zu beyond EOF %" PRIu64 "\n",
3017 		       __func__, (void *)opc, (long long)offset,
3018 		       *resid, vap->va_size);
3019 
3020 	do {
3021 		size_t max_read;
3022 
3023 		max_read = ps->ps_max_readahead - sizeof(*foh);
3024 		/*
3025 		 * flags may be set to FUSE_READ_LOCKOWNER
3026 		 * if lock_owner is provided.
3027 		 */
3028 		pm = ps->ps_new_msg(pu, opc, FUSE_READ, sizeof(*fri), pcr);
3029 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
3030 		fri->fh = perfuse_get_fh(opc, FREAD);
3031 		fri->offset = offset;
3032 		fri->size = (uint32_t)MIN(*resid, max_read);
3033 		fri->read_flags = 0; /* XXX Unused by libfuse? */
3034 		fri->lock_owner = pnd->pnd_lock_owner;
3035 		fri->flags = 0;
3036 		fri->flags |= (fri->lock_owner != 0) ? FUSE_READ_LOCKOWNER : 0;
3037 
3038 #ifdef PERFUSE_DEBUG
3039 	if (perfuse_diagflags & PDF_FH)
3040 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
3041 			__func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
3042 #endif
3043 		error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
3044 		if (error  != 0)
3045 			return error;
3046 
3047 		foh = GET_OUTHDR(ps, pm);
3048 		readen = foh->len - sizeof(*foh);
3049 
3050 #ifdef PERFUSE_DEBUG
3051 		if (readen > *resid)
3052 			DERRX(EX_SOFTWARE, "%s: Unexpected big read %zd",
3053 			      __func__, readen);
3054 #endif
3055 
3056 		(void)memcpy(buf,  _GET_OUTPAYLOAD(ps, pm, char *), readen);
3057 
3058 		buf += readen;
3059 		offset += readen;
3060 		*resid -= readen;
3061 
3062 		ps->ps_destroy_msg(pm);
3063 	} while ((*resid != 0) && (readen != 0));
3064 
3065 	if (ioflag & (IO_SYNC|IO_DSYNC))
3066 		ps->ps_syncreads++;
3067 	else
3068 		ps->ps_asyncreads++;
3069 
3070 	return 0;
3071 }
3072 
3073 int
3074 perfuse_node_write(pu, opc, buf, offset, resid, pcr, ioflag)
3075 	struct puffs_usermount *pu;
3076 	puffs_cookie_t opc;
3077 	uint8_t *buf;
3078 	off_t offset;
3079 	size_t *resid;
3080 	const struct puffs_cred *pcr;
3081 	int ioflag;
3082 {
3083 	struct perfuse_state *ps;
3084 	struct perfuse_node_data *pnd;
3085 	struct vattr *vap;
3086 	perfuse_msg_t *pm;
3087 	struct fuse_write_in *fwi;
3088 	struct fuse_write_out *fwo;
3089 	size_t data_len;
3090 	size_t payload_len;
3091 	size_t written;
3092 	int inresize;
3093 	int error;
3094 
3095 	ps = puffs_getspecific(pu);
3096 	pnd = PERFUSE_NODE_DATA(opc);
3097 	vap = puffs_pn_getvap((struct puffs_node *)opc);
3098 	written = 0;
3099 	inresize = 0;
3100 	error = 0;
3101 
3102 	if (vap->va_type == VDIR)
3103 		return EISDIR;
3104 
3105 	/*
3106 	 * We need to queue write requests in order to avoid
3107 	 * dequeueing PCQ_AFTERWRITE when there are pending writes.
3108 	 */
3109 	while (pnd->pnd_flags & PND_INWRITE)
3110 		requeue_request(pu, opc, PCQ_WRITE);
3111 	pnd->pnd_flags |= PND_INWRITE;
3112 
3113 	/*
3114 	 * Serialize size access, see comment in perfuse_node_setattr().
3115 	 */
3116 	if ((u_quad_t)offset + *resid > vap->va_size) {
3117 		while (pnd->pnd_flags & PND_INRESIZE)
3118 			requeue_request(pu, opc, PCQ_RESIZE);
3119 		pnd->pnd_flags |= PND_INRESIZE;
3120 		inresize = 1;
3121 	}
3122 
3123 	/*
3124 	 * append flag: re-read the file size so that
3125 	 * we get the latest value.
3126 	 */
3127 	if (ioflag & PUFFS_IO_APPEND) {
3128 		DWARNX("%s: PUFFS_IO_APPEND set, untested code", __func__);
3129 
3130 		if ((error = perfuse_node_getattr(pu, opc, vap, pcr)) != 0)
3131 			goto out;
3132 
3133 		offset = vap->va_size;
3134 	}
3135 
3136 #ifdef PERFUSE_DEBUG
3137 	if (perfuse_diagflags & PDF_RESIZE)
3138 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__,
3139 			(void *)opc, vap->va_size);
3140 #endif
3141 
3142 	do {
3143 		size_t max_write;
3144 		/*
3145 		 * There is a writepage flag when data
3146 		 * is aligned to page size. Use it for
3147 		 * everything but the data after the last
3148 		 * page boundary.
3149 		 */
3150 		max_write = ps->ps_max_write - sizeof(*fwi);
3151 
3152 		data_len = MIN(*resid, max_write);
3153 		if (data_len > (size_t)sysconf(_SC_PAGESIZE))
3154 			data_len = data_len & ~(sysconf(_SC_PAGESIZE) - 1);
3155 
3156 		payload_len = data_len + sizeof(*fwi);
3157 
3158 		/*
3159 		 * flags may be set to FUSE_WRITE_CACHE (XXX usage?)
3160 		 * or FUSE_WRITE_LOCKOWNER, if lock_owner is provided.
3161 		 * write_flags is set to 1 for writepage.
3162 		 */
3163 		pm = ps->ps_new_msg(pu, opc, FUSE_WRITE, payload_len, pcr);
3164 		fwi = GET_INPAYLOAD(ps, pm, fuse_write_in);
3165 		fwi->fh = perfuse_get_fh(opc, FWRITE);
3166 		fwi->offset = offset;
3167 		fwi->size = (uint32_t)data_len;
3168 		fwi->write_flags = (fwi->size % sysconf(_SC_PAGESIZE)) ? 0 : 1;
3169 		fwi->lock_owner = pnd->pnd_lock_owner;
3170 		fwi->flags = 0;
3171 		fwi->flags |= (fwi->lock_owner != 0) ? FUSE_WRITE_LOCKOWNER : 0;
3172 		fwi->flags |= (ioflag & IO_DIRECT) ? 0 : FUSE_WRITE_CACHE;
3173 		(void)memcpy((fwi + 1), buf, data_len);
3174 
3175 
3176 #ifdef PERFUSE_DEBUG
3177 		if (perfuse_diagflags & PDF_FH)
3178 			DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
3179 				"fh = 0x%"PRIx64"\n", __func__,
3180 				(void *)opc, pnd->pnd_nodeid, fwi->fh);
3181 #endif
3182 		if ((error = xchg_msg(pu, opc, pm,
3183 				      sizeof(*fwo), wait_reply)) != 0)
3184 			goto out;
3185 
3186 		fwo = GET_OUTPAYLOAD(ps, pm, fuse_write_out);
3187 		written = fwo->size;
3188 #ifdef PERFUSE_DEBUG
3189 		if (written > *resid)
3190 			DERRX(EX_SOFTWARE, "%s: Unexpected big write %zd",
3191 			      __func__, written);
3192 #endif
3193 		*resid -= written;
3194 		offset += written;
3195 		buf += written;
3196 
3197 		ps->ps_destroy_msg(pm);
3198 	} while (*resid != 0);
3199 
3200 	/*
3201 	 * puffs_ops(3) says
3202 	 *  "everything must be written or an error will be generated"
3203 	 */
3204 	if (*resid != 0)
3205 		error = EFBIG;
3206 
3207 #ifdef PERFUSE_DEBUG
3208 	if (perfuse_diagflags & PDF_RESIZE) {
3209 		if (offset > (off_t)vap->va_size)
3210 			DPRINTF("<< %s %p %" PRIu64 " -> %lld\n", __func__,
3211 				(void *)opc, vap->va_size, (long long)offset);
3212 		else
3213 			DPRINTF("<< %s %p \n", __func__, (void *)opc);
3214 	}
3215 #endif
3216 
3217 	/*
3218 	 * Update file size if we wrote beyond the end
3219 	 */
3220 	if (offset > (off_t)vap->va_size)
3221 		vap->va_size = offset;
3222 
3223 	if (inresize) {
3224 #ifdef PERFUSE_DEBUG
3225 		if (!(pnd->pnd_flags & PND_INRESIZE))
3226 			DERRX(EX_SOFTWARE, "file write grow without resize");
3227 #endif
3228 		pnd->pnd_flags &= ~PND_INRESIZE;
3229 		(void)dequeue_requests(ps, opc, PCQ_RESIZE, DEQUEUE_ALL);
3230 	}
3231 
3232 
3233 	/*
3234 	 * Statistics
3235 	 */
3236 	if (ioflag & (IO_SYNC|IO_DSYNC))
3237 		ps->ps_syncwrites++;
3238 	else
3239 		ps->ps_asyncwrites++;
3240 
3241 	/*
3242 	 * Remember to sync the file
3243 	 */
3244 	pnd->pnd_flags |= PND_DIRTY;
3245 
3246 #ifdef PERFUSE_DEBUG
3247 	if (perfuse_diagflags & PDF_SYNC)
3248 		DPRINTF("%s: DIRTY opc = %p, file = \"%s\"\n",
3249 			__func__, (void*)opc, perfuse_node_path(opc));
3250 #endif
3251 
3252 out:
3253 	/*
3254 	 * If there are no more queued write, we can resume
3255 	 * an operation awaiting write completion.
3256 	 */
3257 	pnd->pnd_flags &= ~PND_INWRITE;
3258 	if (dequeue_requests(ps, opc, PCQ_WRITE, 1) == 0)
3259 		(void)dequeue_requests(ps, opc, PCQ_AFTERWRITE, DEQUEUE_ALL);
3260 
3261 	return error;
3262 }
3263 
3264 /* ARGSUSED0 */
3265 void
3266 perfuse_cache_write(pu, opc, size, runs)
3267 	struct puffs_usermount *pu;
3268 	puffs_cookie_t opc;
3269 	size_t size;
3270 	struct puffs_cacherun *runs;
3271 {
3272 	return;
3273 }
3274 
3275 /* ARGSUSED4 */
3276 int
3277 perfuse_node_getextattr(pu, opc, attrns, attrname, attrsize, attr, resid, pcr)
3278 	struct puffs_usermount *pu;
3279 	puffs_cookie_t opc;
3280 	int attrns;
3281 	const char *attrname;
3282 	size_t *attrsize;
3283 	uint8_t *attr;
3284 	size_t *resid;
3285 	const struct puffs_cred *pcr;
3286 {
3287 	struct perfuse_state *ps;
3288 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3289 	perfuse_msg_t *pm;
3290 	struct fuse_getxattr_in *fgi;
3291 	struct fuse_getxattr_out *fgo;
3292 	struct fuse_out_header *foh;
3293 	size_t attrnamelen;
3294 	size_t len;
3295 	char *np;
3296 	int error;
3297 
3298 	ps = puffs_getspecific(pu);
3299 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3300 	attrnamelen = strlen(attrname) + 1;
3301 	len = sizeof(*fgi) + attrnamelen;
3302 
3303 	pm = ps->ps_new_msg(pu, opc, FUSE_GETXATTR, len, pcr);
3304 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3305 	fgi->size = (unsigned int)((resid != NULL) ? *resid : 0);
3306 	np = (char *)(void *)(fgi + 1);
3307 	(void)strlcpy(np, attrname, attrnamelen);
3308 
3309 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3310 		return error;
3311 
3312 	/*
3313 	 * We just get fuse_getattr_out with list size if we requested
3314 	 * a null size.
3315 	 */
3316 	if (resid == NULL) {
3317 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3318 
3319 		if (attrsize != NULL)
3320 			*attrsize = fgo->size;
3321 
3322 		ps->ps_destroy_msg(pm);
3323 		return 0;
3324 	}
3325 
3326 	/*
3327 	 * And with a non null requested size, we get the list just
3328 	 * after the header
3329 	 */
3330 	foh = GET_OUTHDR(ps, pm);
3331 	np = (char *)(void *)(foh + 1);
3332 
3333 	if (resid != NULL) {
3334 		len = MAX(foh->len - sizeof(*foh), *resid);
3335 		(void)memcpy(attr, np, len);
3336 		*resid -= len;
3337 	}
3338 
3339 	ps->ps_destroy_msg(pm);
3340 
3341 	return 0;
3342 }
3343 
3344 int
3345 perfuse_node_setextattr(pu, opc, attrns, attrname, attr, resid, pcr)
3346 	struct puffs_usermount *pu;
3347 	puffs_cookie_t opc;
3348 	int attrns;
3349 	const char *attrname;
3350 	uint8_t *attr;
3351 	size_t *resid;
3352 	const struct puffs_cred *pcr;
3353 {
3354 	struct perfuse_state *ps;
3355 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3356 	perfuse_msg_t *pm;
3357 	struct fuse_setxattr_in *fsi;
3358 	size_t attrnamelen;
3359 	size_t len;
3360 	char *np;
3361 	int error;
3362 
3363 	ps = puffs_getspecific(pu);
3364 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3365 	attrnamelen = strlen(attrname) + 1;
3366 	len = sizeof(*fsi) + attrnamelen + *resid;
3367 
3368 	pm = ps->ps_new_msg(pu, opc, FUSE_SETXATTR, len, pcr);
3369 	fsi = GET_INPAYLOAD(ps, pm, fuse_setxattr_in);
3370 	fsi->size = (unsigned int)*resid;
3371 	fsi->flags = 0;
3372 	np = (char *)(void *)(fsi + 1);
3373 	(void)strlcpy(np, attrname, attrnamelen);
3374 	np += attrnamelen;
3375 	(void)memcpy(np, (char *)attr, *resid);
3376 
3377 	if ((error = xchg_msg(pu, opc, pm,
3378 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
3379 		return error;
3380 
3381 	*resid = 0;
3382 	ps->ps_destroy_msg(pm);
3383 
3384 	return 0;
3385 }
3386 
3387 /* ARGSUSED2 */
3388 int
3389 perfuse_node_listextattr(pu, opc, attrns, attrsize, attrs, resid, flag, pcr)
3390 	struct puffs_usermount *pu;
3391 	puffs_cookie_t opc;
3392 	int attrns;
3393 	size_t *attrsize;
3394 	uint8_t *attrs;
3395 	size_t *resid;
3396 	int flag;
3397 	const struct puffs_cred *pcr;
3398 {
3399 	struct perfuse_state *ps;
3400 	perfuse_msg_t *pm;
3401 	struct fuse_getxattr_in *fgi;
3402 	struct fuse_getxattr_out *fgo;
3403 	struct fuse_out_header *foh;
3404 	char *np;
3405 	size_t len, puffs_len;
3406 	int error;
3407 
3408 	ps = puffs_getspecific(pu);
3409 	len = sizeof(*fgi);
3410 
3411 	pm = ps->ps_new_msg(pu, opc, FUSE_LISTXATTR, len, pcr);
3412 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3413 	if (resid != NULL)
3414 		fgi->size = (unsigned int)*resid;
3415 	else
3416 		fgi->size = 0;
3417 
3418 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3419 		return error;
3420 
3421 	/*
3422 	 * We just get fuse_getattr_out with list size if we requested
3423 	 * a null size.
3424 	 */
3425 	if (resid == NULL) {
3426 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3427 
3428 		if (attrsize != NULL)
3429 			*attrsize = fgo->size;
3430 
3431 		ps->ps_destroy_msg(pm);
3432 
3433 		return 0;
3434 	}
3435 
3436 	/*
3437 	 * And with a non null requested size, we get the list just
3438 	 * after the header
3439 	 */
3440 	foh = GET_OUTHDR(ps, pm);
3441 	np = (char *)(void *)(foh + 1);
3442 	puffs_len = foh->len - sizeof(*foh);
3443 
3444 	if (attrs != NULL) {
3445 #ifdef PUFFS_EXTATTR_LIST_LENPREFIX
3446 		/*
3447 		 * Convert the FUSE reply to length prefixed strings
3448 		 * if this is what the kernel wants.
3449 		 */
3450 		if (flag & PUFFS_EXTATTR_LIST_LENPREFIX) {
3451 			size_t i, attrlen;
3452 
3453 			for (i = 0; i < puffs_len; i += attrlen + 1) {
3454 				attrlen = strlen(np + i);
3455 				(void)memmove(np + i + 1, np + i, attrlen);
3456 				*(np + i) = (uint8_t)attrlen;
3457 			}
3458 		}
3459 #endif /* PUFFS_EXTATTR_LIST_LENPREFIX */
3460 		(void)memcpy(attrs, np, puffs_len);
3461 		*resid -= puffs_len;
3462 	}
3463 
3464 	if (attrsize != NULL)
3465 		*attrsize = puffs_len;
3466 
3467 	ps->ps_destroy_msg(pm);
3468 
3469 	return 0;
3470 }
3471 
3472 int
3473 perfuse_node_deleteextattr(pu, opc, attrns, attrname, pcr)
3474 	struct puffs_usermount *pu;
3475 	puffs_cookie_t opc;
3476 	int attrns;
3477 	const char *attrname;
3478 	const struct puffs_cred *pcr;
3479 {
3480 	struct perfuse_state *ps;
3481 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3482 	perfuse_msg_t *pm;
3483 	size_t attrnamelen;
3484 	char *np;
3485 	int error;
3486 
3487 	ps = puffs_getspecific(pu);
3488 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3489 	attrnamelen = strlen(attrname) + 1;
3490 
3491 	pm = ps->ps_new_msg(pu, opc, FUSE_REMOVEXATTR, attrnamelen, pcr);
3492 	np = _GET_INPAYLOAD(ps, pm, char *);
3493 	(void)strlcpy(np, attrname, attrnamelen);
3494 
3495 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
3496 
3497 	ps->ps_destroy_msg(pm);
3498 
3499 	return error;
3500 }
3501