xref: /netbsd-src/lib/libperfuse/ops.c (revision 9aa0541bdf64142d9a27c2cf274394d60182818f)
1 /*  $NetBSD: ops.c,v 1.42 2011/09/10 10:06:10 tron 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 <machine/vmparam.h>
41 
42 #include "perfuse_priv.h"
43 #include "fuse.h"
44 
45 extern int perfuse_diagflags;
46 
47 static int xchg_msg(struct puffs_usermount *, puffs_cookie_t,
48     perfuse_msg_t *, size_t, enum perfuse_xchg_pb_reply);
49 static int mode_access(puffs_cookie_t, const struct puffs_cred *, mode_t);
50 static int sticky_access(struct puffs_node *, const struct puffs_cred *);
51 static void fuse_attr_to_vap(struct perfuse_state *,
52     struct vattr *, struct fuse_attr *);
53 static int node_lookup_dir_nodot(struct puffs_usermount *,
54     puffs_cookie_t, char *, size_t, struct puffs_node **);
55 static int node_lookup_common(struct puffs_usermount *, puffs_cookie_t,
56     const char *, const struct puffs_cred *, struct puffs_node **);
57 static int node_mk_common(struct puffs_usermount *, puffs_cookie_t,
58     struct puffs_newinfo *, const struct puffs_cn *pcn, perfuse_msg_t *);
59 static int node_mk_common_final(struct puffs_usermount *, puffs_cookie_t,
60     struct puffs_node *, const struct puffs_cn *pcn);
61 static uint64_t readdir_last_cookie(struct fuse_dirent *, size_t);
62 static ssize_t fuse_to_dirent(struct puffs_usermount *, puffs_cookie_t,
63     struct fuse_dirent *, size_t);
64 static int readdir_buffered(puffs_cookie_t, struct dirent *, off_t *,
65     size_t *);
66 static void requeue_request(struct puffs_usermount *,
67     puffs_cookie_t opc, enum perfuse_qtype);
68 static int dequeue_requests(struct perfuse_state *,
69     puffs_cookie_t opc, enum perfuse_qtype, int);
70 #define DEQUEUE_ALL 0
71 
72 /*
73  *  From <sys/vnode>, inside #ifdef _KERNEL section
74  */
75 #define IO_SYNC		(0x40|IO_DSYNC)
76 #define IO_DSYNC	0x00200
77 #define IO_DIRECT	0x02000
78 
79 /*
80  *  From <fcntl>, inside #ifdef _KERNEL section
81  */
82 #define F_WAIT		0x010
83 #define F_FLOCK		0x020
84 #define OFLAGS(fflags)  ((fflags) - 1)
85 
86 /*
87  * Borrowed from src/sys/kern/vfs_subr.c and src/sys/sys/vnode.h
88  */
89 const enum vtype iftovt_tab[16] = {
90 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
91         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
92 };
93 const int vttoif_tab[9] = {
94 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
95         S_IFSOCK, S_IFIFO, S_IFMT,
96 };
97 
98 #define IFTOVT(mode) (iftovt_tab[((mode) & S_IFMT) >> 12])
99 #define VTTOIF(indx) (vttoif_tab[(int)(indx)])
100 
101 int
102 perfuse_node_close_common(pu, opc, mode)
103 	struct puffs_usermount *pu;
104 	puffs_cookie_t opc;
105 	int mode;
106 {
107 	struct perfuse_state *ps;
108 	perfuse_msg_t *pm;
109 	int op;
110 	uint64_t fh;
111 	struct fuse_release_in *fri;
112 	struct perfuse_node_data *pnd;
113 	struct puffs_node *pn;
114 	int error;
115 
116 	ps = puffs_getspecific(pu);
117 	pn = (struct puffs_node *)opc;
118 	pnd = PERFUSE_NODE_DATA(pn);
119 
120 	if (puffs_pn_getvap(pn)->va_type == VDIR) {
121 		op = FUSE_RELEASEDIR;
122 		mode = FREAD;
123 	} else {
124 		op = FUSE_RELEASE;
125 	}
126 
127 	/*
128 	 * Destroy the filehandle before sending the
129 	 * request to the FUSE filesystem, otherwise
130 	 * we may get a second close() while we wait
131 	 * for the reply, and we would end up closing
132 	 * the same fh twice instead of closng both.
133 	 */
134 	fh = perfuse_get_fh(opc, mode);
135 	perfuse_destroy_fh(pn, fh);
136 
137 	/*
138 	 * release_flags may be set to FUSE_RELEASE_FLUSH
139 	 * to flush locks. lock_owner must be set in that case
140 	 *
141 	 * ps_new_msg() is called with NULL creds, which will
142 	 * be interpreted as FUSE superuser. We come here from the
143 	 * inactive method, which provides no creds, but obviously
144 	 * runs with kernel privilege.
145 	 */
146 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*fri), NULL);
147 	fri = GET_INPAYLOAD(ps, pm, fuse_release_in);
148 	fri->fh = fh;
149 	fri->flags = 0;
150 	fri->release_flags = 0;
151 	fri->lock_owner = pnd->pnd_lock_owner;
152 	fri->flags = (fri->lock_owner != 0) ? FUSE_RELEASE_FLUSH : 0;
153 
154 #ifdef PERFUSE_DEBUG
155 	if (perfuse_diagflags & PDF_FH)
156 		DPRINTF("%s: opc = %p, ino = %"PRId64", fh = 0x%"PRIx64"\n",
157 			 __func__, (void *)opc, pnd->pnd_ino, fri->fh);
158 #endif
159 
160 	if ((error = xchg_msg(pu, opc, pm,
161 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
162 		goto out;
163 
164 	ps->ps_destroy_msg(pm);
165 
166 	error = 0;
167 
168 out:
169 	if (error != 0)
170 		DERRX(EX_SOFTWARE, "%s: freed fh = 0x%"PRIx64" but filesystem "
171 		      "returned error = %d", __func__, fh, error);
172 
173 	return error;
174 }
175 
176 /* ARGSUSED1 */
177 static int
178 xchg_msg(pu, opc, pm, len, wait)
179 	struct puffs_usermount *pu;
180 	puffs_cookie_t opc;
181 	perfuse_msg_t *pm;
182 	size_t len;
183      	enum perfuse_xchg_pb_reply wait;
184 {
185 	struct perfuse_state *ps;
186 	struct perfuse_node_data *pnd;
187 	int error;
188 
189 	ps = puffs_getspecific(pu);
190 	pnd = NULL;
191 	if ((struct puffs_node *)opc != NULL)
192 		pnd = PERFUSE_NODE_DATA(opc);
193 
194 #ifdef PERFUSE_DEBUG
195 	if ((perfuse_diagflags & PDF_FILENAME) && (opc != 0))
196 		DPRINTF("file = \"%s\" flags = 0x%x\n",
197 			perfuse_node_path(opc),
198 			PERFUSE_NODE_DATA(opc)->pnd_flags);
199 #endif
200 	if (pnd)
201 		pnd->pnd_flags |= PND_INXCHG;
202 
203 	error = ps->ps_xchg_msg(pu, pm, len, wait);
204 
205 	if (pnd) {
206 		pnd->pnd_flags &= ~PND_INXCHG;
207 		(void)dequeue_requests(ps, opc, PCQ_AFTERXCHG, DEQUEUE_ALL);
208 	}
209 
210 	return error;
211 }
212 
213 static int
214 mode_access(opc, pcr, mode)
215 	puffs_cookie_t opc;
216 	const struct puffs_cred *pcr;
217 	mode_t mode;
218 {
219 	struct puffs_node *pn;
220 	struct vattr *va;
221 
222 	/*
223 	 * pcr is NULL for self open through fsync or readdir.
224 	 * In both case, access control is useless, as it was
225 	 * done before, at open time.
226 	 */
227 	if (pcr == NULL)
228 		return 0;
229 
230 	pn = (struct puffs_node *)opc;
231 	va = puffs_pn_getvap(pn);
232 	return puffs_access(va->va_type, va->va_mode,
233 			    va->va_uid, va->va_gid,
234 			    mode, pcr);
235 }
236 
237 static int
238 sticky_access(targ, pcr)
239 	struct puffs_node *targ;
240 	const struct puffs_cred *pcr;
241 {
242 	uid_t uid;
243 	struct puffs_node *tdir;
244 	int sticky, owner;
245 
246 	tdir = PERFUSE_NODE_DATA(targ)->pnd_parent;
247 
248 	/*
249 	 * This covers the case where the kernel requests a DELETE
250 	 * or RENAME on its own, and where puffs_cred_getuid would
251 	 * return -1. While such a situation should not happen,
252 	 * we allow it here.
253 	 *
254 	 * This also allows root to tamper with other users' files
255 	 * that have the sticky bit.
256 	 */
257 	if (puffs_cred_isjuggernaut(pcr))
258 		return 0;
259 
260 	if (puffs_cred_getuid(pcr, &uid) != 0)
261 		DERRX(EX_SOFTWARE, "puffs_cred_getuid fails in %s", __func__);
262 
263 	sticky = puffs_pn_getvap(tdir)->va_mode & S_ISTXT;
264 	owner = puffs_pn_getvap(targ)->va_uid == uid;
265 
266 	if (sticky && !owner)
267 		return EACCES;
268 
269 	return 0;
270 }
271 
272 
273 static void
274 fuse_attr_to_vap(ps, vap, fa)
275 	struct perfuse_state *ps;
276 	struct vattr *vap;
277 	struct fuse_attr *fa;
278 {
279 	vap->va_type = IFTOVT(fa->mode);
280 	vap->va_mode = fa->mode & ALLPERMS;
281 	vap->va_nlink = fa->nlink;
282 	vap->va_uid = fa->uid;
283 	vap->va_gid = fa->gid;
284 	vap->va_fsid = (long)ps->ps_fsid;
285 	vap->va_fileid = fa->ino;
286 	vap->va_size = fa->size;
287 	vap->va_blocksize = fa->blksize;
288 	vap->va_atime.tv_sec = (time_t)fa->atime;
289 	vap->va_atime.tv_nsec = (long) fa->atimensec;
290 	vap->va_mtime.tv_sec = (time_t)fa->mtime;
291 	vap->va_mtime.tv_nsec = (long)fa->mtimensec;
292 	vap->va_ctime.tv_sec = (time_t)fa->ctime;
293 	vap->va_ctime.tv_nsec = (long)fa->ctimensec;
294 	vap->va_birthtime.tv_sec = 0;
295 	vap->va_birthtime.tv_nsec = 0;
296 	vap->va_gen = 0;
297 	vap->va_flags = 0;
298 	vap->va_rdev = fa->rdev;
299 	vap->va_bytes = fa->size;
300 	vap->va_filerev = (u_quad_t)PUFFS_VNOVAL;
301 	vap->va_vaflags = 0;
302 
303 	if (vap->va_blocksize == 0)
304 		vap->va_blocksize = DEV_BSIZE;
305 
306 	if (vap->va_size == (size_t)PUFFS_VNOVAL) /* XXX */
307 		vap->va_size = 0;
308 
309 	return;
310 }
311 
312 
313 /*
314  * Lookup name in directory opc
315  * We take special care of name being . or ..
316  * These are returned by readdir and deserve tweaks.
317  */
318 static int
319 node_lookup_dir_nodot(pu, opc, name, namelen, pnp)
320 	struct puffs_usermount *pu;
321 	puffs_cookie_t opc;
322 	char *name;
323 	size_t namelen;
324 	struct puffs_node **pnp;
325 {
326 	char *path;
327 	struct puffs_node *dpn = (struct puffs_node *)opc;
328 	int error;
329 
330 	/*
331 	 *  is easy as we already know it
332 	 */
333 	if (strncmp(name, ".", namelen) == 0) {
334 		*pnp = (struct puffs_node *)opc;
335 		return 0;
336 	}
337 
338 	/*
339 	 * For .. we just forget the name part
340 	 */
341 	if (strncmp(name, "..", namelen) == 0)
342 		namelen = 0;
343 
344 	namelen = PNPLEN(dpn) + 1 + namelen + 1;
345 	if ((path = malloc(namelen)) == NULL)
346 		DERR(EX_OSERR, "%s: malloc failed", __func__);
347 	(void)snprintf(path, namelen, "%s/%s",
348 		       perfuse_node_path((puffs_cookie_t)dpn), name);
349 
350 	error = node_lookup_common(pu, opc, path, NULL, pnp);
351 
352 	free(path);
353 
354 	return error;
355 }
356 
357 static int
358 node_lookup_common(pu, opc, path, pcr, pnp)
359 	struct puffs_usermount *pu;
360 	puffs_cookie_t opc;
361 	const char *path;
362 	const struct puffs_cred *pcr;
363 	struct puffs_node **pnp;
364 {
365 	struct perfuse_state *ps;
366 	struct perfuse_node_data *pnd;
367 	perfuse_msg_t *pm;
368 	struct fuse_entry_out *feo;
369 	struct puffs_node *pn;
370 	size_t len;
371 	int error;
372 
373 	ps = puffs_getspecific(pu);
374 
375 #ifdef PERFUSE_DEBUG
376 	if (perfuse_diagflags & PDF_FILENAME)
377 		DPRINTF("%s: opc = %p, file = \"%s\" looking up \"%s\"\n",
378 			__func__, (void *)opc, perfuse_node_path(opc), path);
379 #endif
380 	/*
381 	 * Is the node already known?
382 	 */
383 	TAILQ_FOREACH(pnd, &PERFUSE_NODE_DATA(opc)->pnd_children, pnd_next) {
384 		if ((pnd->pnd_flags & PND_REMOVED) ||
385 		    (strcmp(pnd->pnd_name, path) != 0))
386 			continue;
387 
388 		/*
389 		 * We have a match
390 		 */
391 		if (pnp != NULL)
392 			*pnp = (struct puffs_node *)(pnd->pnd_pn);
393 
394 #ifdef PERFUSE_DEBUG
395 		if (perfuse_diagflags & PDF_FILENAME)
396 			DPRINTF("%s: opc = %p, file = \"%s\" found "
397 				"cookie = %p, ino = %"PRId64" for \"%s\"\n",
398 				__func__, (void *)opc, perfuse_node_path(opc),
399 				(void *)pnd->pnd_pn, pnd->pnd_ino, path);
400 #endif
401 		return 0;
402 	}
403 
404 	len = strlen(path) + 1;
405 
406 	pm = ps->ps_new_msg(pu, opc, FUSE_LOOKUP, len, pcr);
407 	(void)strlcpy(_GET_INPAYLOAD(ps, pm, char *), path, len);
408 
409 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
410 		goto out;
411 
412 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
413 
414 	pn = perfuse_new_pn(pu, path, opc);
415 	PERFUSE_NODE_DATA(pn)->pnd_ino = feo->nodeid;
416 
417 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
418 	pn->pn_va.va_gen = (u_long)(feo->generation);
419 
420 	if (pnp != NULL)
421 		*pnp = pn;
422 
423 #ifdef PERFUSE_DEBUG
424 	if (perfuse_diagflags & PDF_FILENAME)
425 		DPRINTF("%s: opc = %p, looked up opc = %p, ino = %"PRId64" "
426 			"file = \"%s\"\n", __func__, (void *)opc, pn,
427 			feo->nodeid, path);
428 #endif
429 out:
430 	ps->ps_destroy_msg(pm);
431 
432 	return error;
433 }
434 
435 
436 /*
437  * Common code for methods that create objects:
438  * perfuse_node_mkdir
439  * perfuse_node_mknod
440  * perfuse_node_symlink
441  */
442 static int
443 node_mk_common(pu, opc, pni, pcn, pm)
444 	struct puffs_usermount *pu;
445 	puffs_cookie_t opc;
446 	struct puffs_newinfo *pni;
447 	const struct puffs_cn *pcn;
448 	perfuse_msg_t *pm;
449 {
450 	struct perfuse_state *ps;
451 	struct puffs_node *pn;
452 	struct fuse_entry_out *feo;
453 	int error;
454 
455 	ps =  puffs_getspecific(pu);
456 
457 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
458 		goto out;
459 
460 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
461 	if (feo->nodeid == PERFUSE_UNKNOWN_INO)
462 		DERRX(EX_SOFTWARE, "%s: no ino", __func__);
463 
464 	pn = perfuse_new_pn(pu, pcn->pcn_name, opc);
465 	PERFUSE_NODE_DATA(pn)->pnd_ino = feo->nodeid;
466 
467 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
468 	pn->pn_va.va_gen = (u_long)(feo->generation);
469 
470 	puffs_newinfo_setcookie(pni, pn);
471 
472 #ifdef PERFUSE_DEBUG
473 	if (perfuse_diagflags & PDF_FILENAME)
474 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
475 			"ino = %"PRId64"\n",
476 			__func__, (void *)pn, pcn->pcn_name,
477 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid);
478 #endif
479 	ps->ps_destroy_msg(pm);
480 
481 	return node_mk_common_final(pu, opc, pn, pcn);
482 
483 out:
484 	ps->ps_destroy_msg(pm);
485 
486 	return error;
487 }
488 
489 /*
490  * Common final code for methods that create objects:
491  * perfuse_node_mkdir via node_mk_common
492  * perfuse_node_mknod via node_mk_common
493  * perfuse_node_symlink via node_mk_common
494  * perfuse_node_create
495  */
496 static int
497 node_mk_common_final(pu, opc, pn, pcn)
498 	struct puffs_usermount *pu;
499 	puffs_cookie_t opc;
500 	struct puffs_node *pn;
501 	const struct puffs_cn *pcn;
502 {
503 	struct perfuse_state *ps;
504 	perfuse_msg_t *pm;
505 	struct fuse_setattr_in *fsi;
506 	struct fuse_attr_out *fao;
507 	int error;
508 
509 	ps =  puffs_getspecific(pu);
510 
511 	/*
512 	 * Set owner and group. The kernel cannot create a file
513 	 * on its own (puffs_cred_getuid would return -1), right?
514 	 */
515 	if (puffs_cred_getuid(pcn->pcn_cred, &pn->pn_va.va_uid) != 0)
516 		DERRX(EX_SOFTWARE, "puffs_cred_getuid fails in %s", __func__);
517 	if (puffs_cred_getgid(pcn->pcn_cred, &pn->pn_va.va_gid) != 0)
518 		DERRX(EX_SOFTWARE, "puffs_cred_getgid fails in %s", __func__);
519 
520 	pm = ps->ps_new_msg(pu, (puffs_cookie_t)pn,
521 			    FUSE_SETATTR, sizeof(*fsi), pcn->pcn_cred);
522 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
523 	fsi->uid = pn->pn_va.va_uid;
524 	fsi->gid = pn->pn_va.va_gid;
525 	fsi->valid = FUSE_FATTR_UID|FUSE_FATTR_GID;
526 
527 	if ((error = xchg_msg(pu, (puffs_cookie_t)pn, pm,
528 			      sizeof(*fao), wait_reply)) != 0)
529 		goto out;
530 
531 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
532 	fuse_attr_to_vap(ps, &pn->pn_va, &fao->attr);
533 
534 	/*
535 	 * The parent directory needs a sync
536 	 */
537 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
538 
539 out:
540 	if (pm != NULL)
541 		ps->ps_destroy_msg(pm);
542 
543 	return error;
544 }
545 
546 static uint64_t
547 readdir_last_cookie(fd, fd_len)
548 	struct fuse_dirent *fd;
549 	size_t fd_len;
550 {
551 	size_t len;
552 	size_t seen = 0;
553 	char *ndp;
554 
555 	do {
556 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
557 		seen += len;
558 
559 		if (seen >= fd_len)
560 			break;
561 
562 		ndp = (char *)(void *)fd + (size_t)len;
563 		fd = (struct fuse_dirent *)(void *)ndp;
564 	} while (1 /* CONSTCOND */);
565 
566 	return fd->off;
567 }
568 
569 static ssize_t
570 fuse_to_dirent(pu, opc, fd, fd_len)
571 	struct puffs_usermount *pu;
572 	puffs_cookie_t opc;
573 	struct fuse_dirent *fd;
574 	size_t fd_len;
575 {
576 	struct dirent *dents;
577 	size_t dents_len;
578 	ssize_t written;
579 	uint64_t fd_offset;
580 	struct fuse_dirent *fd_base;
581 	size_t len;
582 
583 	fd_base = fd;
584 	fd_offset = 0;
585 	written = 0;
586 	dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
587 	dents_len = (size_t)PERFUSE_NODE_DATA(opc)->pnd_dirent_len;
588 
589 	do {
590 		char *ndp;
591 		size_t reclen;
592 
593 		reclen = _DIRENT_RECLEN(dents, fd->namelen);
594 
595 		/*
596 		 * Check we do not overflow the output buffer
597 		 * struct fuse_dirent is bigger than struct dirent,
598 		 * so we should always use fd_len and never reallocate
599 		 * later.
600 		 * If we have to reallocate,try to double the buffer
601 		 * each time so that we do not have to do it too often.
602 		 */
603 		if (written + reclen > dents_len) {
604 			if (dents_len == 0)
605 				dents_len = fd_len;
606 			else
607 				dents_len =
608 				   MAX(2 * dents_len, written + reclen);
609 
610 			dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
611 			if ((dents = realloc(dents, dents_len)) == NULL)
612 				DERR(EX_OSERR, "%s: malloc failed", __func__);
613 
614 			PERFUSE_NODE_DATA(opc)->pnd_dirent = dents;
615 			PERFUSE_NODE_DATA(opc)->pnd_dirent_len = dents_len;
616 
617 			/*
618 			 * (void *) for delint
619 			 */
620 			ndp = (char *)(void *)dents + written;
621 			dents = (struct dirent *)(void *)ndp;
622 		}
623 
624 		/*
625 		 * Filesystem was mounted without -o use_ino
626 		 * Perform a lookup to find it.
627 		 * XXX still broken
628 		 */
629 		if (fd->ino == PERFUSE_UNKNOWN_INO) {
630 			struct puffs_node *pn;
631 
632 			if (node_lookup_dir_nodot(pu, opc, fd->name,
633 						  fd->namelen, &pn) != 0)
634 				DERRX(EX_SOFTWARE,
635 				     "node_lookup_dir_nodot failed");
636 
637 			fd->ino = PERFUSE_NODE_DATA(pn)->pnd_ino;
638 		}
639 
640 		dents->d_fileno = fd->ino;
641 		dents->d_reclen = (unsigned short)reclen;
642 		dents->d_namlen = fd->namelen;
643 		dents->d_type = fd->type;
644 		strlcpy(dents->d_name, fd->name, fd->namelen + 1);
645 
646 #ifdef PERFUSE_DEBUG
647 		if (perfuse_diagflags & PDF_READDIR)
648 			DPRINTF("%s: translated \"%s\" ino = %"PRId64"\n",
649 				__func__, dents->d_name, dents->d_fileno);
650 #endif
651 
652 		dents = _DIRENT_NEXT(dents);
653 		written += reclen;
654 
655 		/*
656 		 * Move to the next record.
657 		 * fd->off is not the offset, it is an opaque cookie
658 		 * given by the filesystem to keep state across multiple
659 		 * readdir() operation.
660 		 * Use record alignement instead.
661 		 */
662 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
663 #ifdef PERFUSE_DEBUG
664 		if (perfuse_diagflags & PDF_READDIR)
665 			DPRINTF("%s: record at %"PRId64"/0x%"PRIx64" "
666 				"length = %zd/0x%zx. "
667 				"next record at %"PRId64"/0x%"PRIx64" "
668 				"max %zd/0x%zx\n",
669 				__func__, fd_offset, fd_offset, len, len,
670 				fd_offset + len, fd_offset + len,
671 				fd_len, fd_len);
672 #endif
673 		fd_offset += len;
674 
675 		/*
676 		 * Check if next record is still within the packet
677 		 * If it is not, we reached the end of the buffer.
678 		 */
679 		if (fd_offset >= fd_len)
680 			break;
681 
682 		/*
683 		 * (void *) for delint
684 		 */
685 		ndp = (char *)(void *)fd_base + (size_t)fd_offset;
686 		fd = (struct fuse_dirent *)(void *)ndp;
687 
688 	} while (1 /* CONSTCOND */);
689 
690 	/*
691 	 * Adjust the dirent output length
692 	 */
693 	if (written != -1)
694 		PERFUSE_NODE_DATA(opc)->pnd_dirent_len = written;
695 
696 	return written;
697 }
698 
699 static int
700 readdir_buffered(opc, dent, readoff, reslen)
701 	puffs_cookie_t opc;
702 	struct dirent *dent;
703 	off_t *readoff;
704 	size_t *reslen;
705 {
706 	struct dirent *fromdent;
707 	struct perfuse_node_data *pnd;
708 	char *ndp;
709 
710 	pnd = PERFUSE_NODE_DATA(opc);
711 
712 	while (*readoff < pnd->pnd_dirent_len) {
713 		/*
714 		 * (void *) for delint
715 		 */
716 		ndp = (char *)(void *)pnd->pnd_dirent + (size_t)*readoff;
717 		fromdent = (struct dirent *)(void *)ndp;
718 
719 		if (*reslen < _DIRENT_SIZE(fromdent))
720 			break;
721 
722 		memcpy(dent, fromdent, _DIRENT_SIZE(fromdent));
723 		*readoff += _DIRENT_SIZE(fromdent);
724 		*reslen -= _DIRENT_SIZE(fromdent);
725 
726 		dent = _DIRENT_NEXT(dent);
727 	}
728 
729 #ifdef PERFUSE_DEBUG
730 	if (perfuse_diagflags & PDF_READDIR)
731 		DPRINTF("%s: readoff = %"PRId64",  "
732 			"pnd->pnd_dirent_len = %"PRId64"\n",
733 			__func__, *readoff, pnd->pnd_dirent_len);
734 #endif
735 	if (*readoff >=  pnd->pnd_dirent_len) {
736 		free(pnd->pnd_dirent);
737 		pnd->pnd_dirent = NULL;
738 		pnd->pnd_dirent_len = 0;
739 	}
740 
741 	return 0;
742 }
743 
744 static void
745 requeue_request(pu, opc, type)
746 	struct puffs_usermount *pu;
747 	puffs_cookie_t opc;
748 	enum perfuse_qtype type;
749 {
750 	struct perfuse_cc_queue pcq;
751 	struct perfuse_node_data *pnd;
752 #ifdef PERFUSE_DEBUG
753 	struct perfuse_state *ps;
754 
755 	ps = perfuse_getspecific(pu);
756 #endif
757 
758 	pnd = PERFUSE_NODE_DATA(opc);
759 	pcq.pcq_type = type;
760 	pcq.pcq_cc = puffs_cc_getcc(pu);
761 	TAILQ_INSERT_TAIL(&pnd->pnd_pcq, &pcq, pcq_next);
762 
763 #ifdef PERFUSE_DEBUG
764 	if (perfuse_diagflags & PDF_REQUEUE)
765 		DPRINTF("%s: REQUEUE opc = %p, pcc = %p (%s)\n",
766 		        __func__, (void *)opc, pcq.pcq_cc,
767 			perfuse_qtypestr[type]);
768 #endif
769 
770 	puffs_cc_yield(pcq.pcq_cc);
771 	TAILQ_REMOVE(&pnd->pnd_pcq, &pcq, pcq_next);
772 
773 #ifdef PERFUSE_DEBUG
774 	if (perfuse_diagflags & PDF_REQUEUE)
775 		DPRINTF("%s: RESUME opc = %p, pcc = %p (%s)\n",
776 		        __func__, (void *)opc, pcq.pcq_cc,
777 			perfuse_qtypestr[type]);
778 #endif
779 
780 	return;
781 }
782 
783 /* ARGSUSED0 */
784 static int
785 dequeue_requests(ps, opc, type, max)
786 	struct perfuse_state *ps;
787 	puffs_cookie_t opc;
788 	enum perfuse_qtype type;
789 	int max;
790 {
791 	struct perfuse_cc_queue *pcq;
792 	struct perfuse_node_data *pnd;
793 	int dequeued;
794 
795 	pnd = PERFUSE_NODE_DATA(opc);
796 	dequeued = 0;
797 	TAILQ_FOREACH(pcq, &pnd->pnd_pcq, pcq_next) {
798 		if (pcq->pcq_type != type)
799 			continue;
800 
801 #ifdef PERFUSE_DEBUG
802 		if (perfuse_diagflags & PDF_REQUEUE)
803 			DPRINTF("%s: SCHEDULE opc = %p, pcc = %p (%s)\n",
804 				__func__, (void *)opc, pcq->pcq_cc,
805 				 perfuse_qtypestr[type]);
806 #endif
807 		puffs_cc_schedule(pcq->pcq_cc);
808 
809 		if (++dequeued == max)
810 			break;
811 	}
812 
813 #ifdef PERFUSE_DEBUG
814 	if (perfuse_diagflags & PDF_REQUEUE)
815 		DPRINTF("%s: DONE  opc = %p\n", __func__, (void *)opc);
816 #endif
817 
818 	return dequeued;
819 }
820 
821 void
822 perfuse_fs_init(pu)
823 	struct puffs_usermount *pu;
824 {
825 	struct perfuse_state *ps;
826 	perfuse_msg_t *pm;
827 	struct fuse_init_in *fii;
828 	struct fuse_init_out *fio;
829 	int error;
830 
831 	ps = puffs_getspecific(pu);
832 
833         if (puffs_mount(pu, ps->ps_target, ps->ps_mountflags, ps->ps_root) != 0)
834                 DERR(EX_OSERR, "%s: puffs_mount failed", __func__);
835 
836 	/*
837 	 * Linux 2.6.34.1 sends theses flags:
838 	 * FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC
839 	 * FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK
840 	 *
841 	 * Linux also sets max_readahead at 32 pages (128 kB)
842 	 *
843 	 * ps_new_msg() is called with NULL creds, which will
844 	 * be interpreted as FUSE superuser.
845 	 */
846 	pm = ps->ps_new_msg(pu, 0, FUSE_INIT, sizeof(*fii), NULL);
847 	fii = GET_INPAYLOAD(ps, pm, fuse_init_in);
848 	fii->major = FUSE_KERNEL_VERSION;
849 	fii->minor = FUSE_KERNEL_MINOR_VERSION;
850 	fii->max_readahead = (unsigned int)(32 * sysconf(_SC_PAGESIZE));
851 	fii->flags = (FUSE_ASYNC_READ|FUSE_POSIX_LOCKS|FUSE_ATOMIC_O_TRUNC);
852 
853 	if ((error = xchg_msg(pu, 0, pm, sizeof(*fio), wait_reply)) != 0)
854 		DERRX(EX_SOFTWARE, "init message exchange failed (%d)", error);
855 
856 	fio = GET_OUTPAYLOAD(ps, pm, fuse_init_out);
857 	ps->ps_max_readahead = fio->max_readahead;
858 	ps->ps_max_write = fio->max_write;
859 
860 	ps->ps_destroy_msg(pm);
861 
862 	return;
863 }
864 
865 int
866 perfuse_fs_unmount(pu, flags)
867 	struct puffs_usermount *pu;
868 	int flags;
869 {
870 	perfuse_msg_t *pm;
871 	struct perfuse_state *ps;
872 	puffs_cookie_t opc;
873 	int error;
874 
875 	ps = puffs_getspecific(pu);
876 	opc = (puffs_cookie_t)puffs_getroot(pu);
877 
878 	/*
879 	 * ps_new_msg() is called with NULL creds, which will
880 	 * be interpreted as FUSE superuser.
881 	 */
882 	pm = ps->ps_new_msg(pu, opc, FUSE_DESTROY, 0, NULL);
883 
884 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0){
885 		DWARN("unmount %s", ps->ps_target);
886 		if (!(flags & MNT_FORCE))
887 			goto out;
888 	}
889 
890 	ps->ps_umount(pu);
891 
892 	if (perfuse_diagflags & PDF_MISC)
893 		DPRINTF("%s unmounted, exit\n", ps->ps_target);
894 
895 	return 0;
896 out:
897 	ps->ps_destroy_msg(pm);
898 
899 	return error;
900 }
901 
902 int
903 perfuse_fs_statvfs(pu, svfsb)
904 	struct puffs_usermount *pu;
905 	struct statvfs *svfsb;
906 {
907 	struct perfuse_state *ps;
908 	perfuse_msg_t *pm;
909 	puffs_cookie_t opc;
910 	struct fuse_statfs_out *fso;
911 	int error;
912 
913 	ps = puffs_getspecific(pu);
914 	opc = (puffs_cookie_t)puffs_getroot(pu);
915 
916 	/*
917 	 * ps_new_msg() is called with NULL creds, which will
918 	 * be interpreted as FUSE superuser.
919 	 */
920 	pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
921 
922 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply)) != 0)
923 		goto out;
924 
925 	fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
926 	svfsb->f_flag = ps->ps_mountflags;
927 	svfsb->f_bsize = fso->st.bsize;
928 	svfsb->f_frsize = fso->st.frsize;
929 	svfsb->f_iosize = ((struct puffs_node *)opc)->pn_va.va_blocksize;
930 	svfsb->f_blocks = fso->st.blocks;
931 	svfsb->f_bfree = fso->st.bfree;
932 	svfsb->f_bavail = fso->st.bavail;
933 	svfsb->f_bresvd = fso->st.bfree - fso->st.bavail;
934 	svfsb->f_files = fso->st.files;
935 	svfsb->f_ffree = fso->st.ffree;
936 	svfsb->f_favail = fso->st.ffree;/* files not reserved for root */
937 	svfsb->f_fresvd = 0;		/* files reserved for root */
938 
939 	svfsb->f_syncreads = ps->ps_syncreads;
940 	svfsb->f_syncwrites = ps->ps_syncwrites;
941 
942 	svfsb->f_asyncreads = ps->ps_asyncreads;
943 	svfsb->f_asyncwrites = ps->ps_asyncwrites;
944 
945 	(void)memcpy(&svfsb->f_fsidx, &ps->ps_fsid, sizeof(ps->ps_fsid));
946 	svfsb->f_fsid = (unsigned long)ps->ps_fsid;
947 	svfsb->f_namemax = MAXPATHLEN;	/* XXX */
948 	svfsb->f_owner = ps->ps_owner_uid;
949 
950 	(void)strlcpy(svfsb->f_mntonname, ps->ps_target, _VFS_NAMELEN);
951 
952 	if (ps->ps_filesystemtype != NULL)
953 		(void)strlcpy(svfsb->f_fstypename,
954 			      ps->ps_filesystemtype, _VFS_NAMELEN);
955 	else
956 		(void)strlcpy(svfsb->f_fstypename, "fuse", _VFS_NAMELEN);
957 
958 	if (ps->ps_source != NULL)
959 		strlcpy(svfsb->f_mntfromname, ps->ps_source, _VFS_NAMELEN);
960 	else
961 		strlcpy(svfsb->f_mntfromname, _PATH_FUSE, _VFS_NAMELEN);
962 out:
963 	ps->ps_destroy_msg(pm);
964 
965 	return error;
966 }
967 
968 int
969 perfuse_fs_sync(pu, waitfor, pcr)
970 	struct puffs_usermount *pu;
971 	int waitfor;
972 	const struct puffs_cred *pcr;
973 {
974 	/*
975 	 * FUSE does not seem to have a FS sync callback.
976 	 * Maybe do not even register this callback
977 	 */
978 	return puffs_fsnop_sync(pu, waitfor, pcr);
979 }
980 
981 /* ARGSUSED0 */
982 int
983 perfuse_fs_fhtonode(pu, fid, fidsize, pni)
984 	struct puffs_usermount *pu;
985 	void *fid;
986 	size_t fidsize;
987 	struct puffs_newinfo *pni;
988 {
989 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
990 	return 0;
991 }
992 
993 /* ARGSUSED0 */
994 int
995 perfuse_fs_nodetofh(pu, cookie, fid, fidsize)
996 	struct puffs_usermount *pu;
997 	puffs_cookie_t cookie;
998 	void *fid;
999 	size_t *fidsize;
1000 {
1001 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1002 	return 0;
1003 }
1004 
1005 #if 0
1006 /* ARGSUSED0 */
1007 void
1008 perfuse_fs_extattrctl(pu, cmd, cookie, flags, namespace, attrname)
1009 	struct puffs_usermount *pu;
1010 	int cmd,
1011 	puffs_cookie_t *cookie;
1012 	int flags;
1013 	int namespace;
1014 	const char *attrname;
1015 {
1016 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1017 	return 0;
1018 }
1019 #endif /* 0 */
1020 
1021 /* ARGSUSED0 */
1022 void
1023 perfuse_fs_suspend(pu, status)
1024 	struct puffs_usermount *pu;
1025 	int status;
1026 {
1027 	return;
1028 }
1029 
1030 
1031 
1032 int
1033 perfuse_node_lookup(pu, opc, pni, pcn)
1034 	struct puffs_usermount *pu;
1035 	puffs_cookie_t opc;
1036 	struct puffs_newinfo *pni;
1037 	const struct puffs_cn *pcn;
1038 {
1039 	struct puffs_node *pn;
1040 	mode_t mode;
1041 	int error;
1042 
1043 	/*
1044 	 * Check permissions
1045 	 */
1046 	switch(pcn->pcn_nameiop) {
1047 	case NAMEI_DELETE: /* FALLTHROUGH */
1048 	case NAMEI_RENAME: /* FALLTHROUGH */
1049 	case NAMEI_CREATE:
1050 		if (pcn->pcn_flags & NAMEI_ISLASTCN)
1051 			mode = PUFFS_VEXEC|PUFFS_VWRITE;
1052 		else
1053 			mode = PUFFS_VEXEC;
1054 		break;
1055 	case NAMEI_LOOKUP: /* FALLTHROUGH */
1056 	default:
1057 		mode = PUFFS_VEXEC;
1058 		break;
1059 	}
1060 
1061 	if ((error = mode_access(opc, pcn->pcn_cred, mode)) != 0)
1062 		return error;
1063 
1064 	/*
1065 	 * Special case for ..
1066 	 */
1067 	if (strcmp(pcn->pcn_name, "..") == 0)
1068 		pn = PERFUSE_NODE_DATA(opc)->pnd_parent;
1069 	else
1070 		error = node_lookup_common(pu, (puffs_cookie_t)opc,
1071 					   pcn->pcn_name, pcn->pcn_cred, &pn);
1072 	if (error != 0)
1073 		return error;
1074 
1075 	/*
1076 	 * Removed node
1077 	 */
1078 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_REMOVED)
1079 		return ENOENT;
1080 
1081 	/*
1082 	 * Check for sticky bit. Unfortunately there is no way to
1083 	 * do this before creating the puffs_node, since we require
1084 	 * this operation to get the node owner.
1085 	 */
1086 	switch (pcn->pcn_nameiop) {
1087 	case NAMEI_DELETE: /* FALLTHROUGH */
1088 	case NAMEI_RENAME:
1089 		error = sticky_access(pn, pcn->pcn_cred);
1090 		if (error != 0) {
1091 			/*
1092 			 * kernel will never know about it and will
1093 			 * not reclaim it. The filesystem needs to
1094 			 * clean it up anyway, therefore mimick a forget.
1095 			 */
1096 			PERFUSE_NODE_DATA(pn)->pnd_flags |= PND_RECLAIMED;
1097 			(void)perfuse_node_reclaim(pu, (puffs_cookie_t)pn);
1098 			return error;
1099 		}
1100 		break;
1101 	default:
1102 		break;
1103 	}
1104 
1105 	/*
1106 	 * If that node had a pending reclaim, wipe it out.
1107 	 */
1108 	PERFUSE_NODE_DATA(pn)->pnd_flags &= ~PND_RECLAIMED;
1109 
1110 	puffs_newinfo_setcookie(pni, pn);
1111 	puffs_newinfo_setvtype(pni, pn->pn_va.va_type);
1112 	puffs_newinfo_setsize(pni, (voff_t)pn->pn_va.va_size);
1113 	puffs_newinfo_setrdev(pni, pn->pn_va.va_rdev);
1114 
1115 	return error;
1116 }
1117 
1118 int
1119 perfuse_node_create(pu, opc, pni, pcn, vap)
1120 	struct puffs_usermount *pu;
1121 	puffs_cookie_t opc;
1122 	struct puffs_newinfo *pni;
1123 	const struct puffs_cn *pcn;
1124 	const struct vattr *vap;
1125 {
1126 	perfuse_msg_t *pm;
1127 	struct perfuse_state *ps;
1128 	struct fuse_create_in *fci;
1129 	struct fuse_entry_out *feo;
1130 	struct fuse_open_out *foo;
1131 	struct puffs_node *pn;
1132 	const char *name;
1133 	size_t namelen;
1134 	size_t len;
1135 	int error;
1136 
1137 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1138 		return ENOENT;
1139 
1140 	/*
1141 	 * If create is unimplemented: Check that it does not
1142 	 * already exists, and if not, do mknod and open
1143 	 */
1144 	ps = puffs_getspecific(pu);
1145 	if (ps->ps_flags & PS_NO_CREAT) {
1146 		error = node_lookup_common(pu, opc, pcn->pcn_name,
1147 					   pcn->pcn_cred, &pn);
1148 		if (error == 0)
1149 			return EEXIST;
1150 
1151 		error = perfuse_node_mknod(pu, opc, pni, pcn, vap);
1152 		if (error != 0)
1153 			return error;
1154 
1155 		error = node_lookup_common(pu, opc, pcn->pcn_name,
1156 					   pcn->pcn_cred, &pn);
1157 		if (error != 0)
1158 			return error;
1159 
1160 		/*
1161 		 * FUSE does the open at create time, while
1162 		 * NetBSD will open in a subsequent operation.
1163 		 * We need to open now, in order to retain FUSE
1164 		 * semantics. The calling process will not get
1165 		 * a file descriptor before the kernel sends
1166 		 * the open operation.
1167 		 */
1168 		opc = (puffs_cookie_t)pn;
1169 		error = perfuse_node_open(pu, opc, FWRITE, pcn->pcn_cred);
1170 		if (error != 0)
1171 			return error;
1172 
1173 		return 0;
1174 	}
1175 
1176 	name = pcn->pcn_name;
1177 	namelen = pcn->pcn_namelen + 1;
1178 	len = sizeof(*fci) + namelen;
1179 
1180 	/*
1181 	 * flags should use O_WRONLY instead of O_RDWR, but it
1182 	 * breaks when the caller tries to read from file.
1183 	 *
1184 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1185 	 */
1186 	pm = ps->ps_new_msg(pu, opc, FUSE_CREATE, len, pcn->pcn_cred);
1187 	fci = GET_INPAYLOAD(ps, pm, fuse_create_in);
1188 	fci->flags = O_CREAT | O_TRUNC | O_RDWR;
1189 	fci->mode = vap->va_mode | VTTOIF(vap->va_type);
1190 	fci->umask = 0; 	/* Seems unused by libfuse */
1191 	(void)strlcpy((char*)(void *)(fci + 1), name, namelen);
1192 
1193 	len = sizeof(*feo) + sizeof(*foo);
1194 	if ((error = xchg_msg(pu, opc, pm, len, wait_reply)) != 0)
1195 		goto out;
1196 
1197 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
1198 	foo = (struct fuse_open_out *)(void *)(feo + 1);
1199 	if (feo->nodeid == PERFUSE_UNKNOWN_INO)
1200 		DERRX(EX_SOFTWARE, "%s: no ino", __func__);
1201 
1202 	/*
1203 	 * Save the file handle and inode in node private data
1204 	 * so that we can reuse it later
1205 	 */
1206 	pn = perfuse_new_pn(pu, name, opc);
1207 	perfuse_new_fh((puffs_cookie_t)pn, foo->fh, FWRITE);
1208 	PERFUSE_NODE_DATA(pn)->pnd_ino = feo->nodeid;
1209 
1210 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
1211 	pn->pn_va.va_gen = (u_long)(feo->generation);
1212 
1213 	puffs_newinfo_setcookie(pni, pn);
1214 
1215 #ifdef PERFUSE_DEBUG
1216 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1217 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
1218 			"ino = %"PRId64", wfh = 0x%"PRIx64"\n",
1219 			__func__, (void *)pn, pcn->pcn_name,
1220 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid, foo->fh);
1221 #endif
1222 
1223 	ps->ps_destroy_msg(pm);
1224 
1225 	return node_mk_common_final(pu, opc, pn, pcn);
1226 
1227 out:
1228 	ps->ps_destroy_msg(pm);
1229 
1230 	/*
1231 	 * create is unimplmented, remember it for later,
1232 	 * and start over using mknod and open instead.
1233 	 */
1234 	if (error == ENOSYS) {
1235 		ps->ps_flags |= PS_NO_CREAT;
1236 		return perfuse_node_create(pu, opc, pni, pcn, vap);
1237 	}
1238 
1239 	return error;
1240 }
1241 
1242 
1243 int
1244 perfuse_node_mknod(pu, opc, pni, pcn, vap)
1245 	struct puffs_usermount *pu;
1246 	puffs_cookie_t opc;
1247 	struct puffs_newinfo *pni;
1248 	const struct puffs_cn *pcn;
1249 	const struct vattr *vap;
1250 {
1251 	struct perfuse_state *ps;
1252 	perfuse_msg_t *pm;
1253 	struct fuse_mknod_in *fmi;
1254 	const char* path;
1255 	size_t len;
1256 
1257 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1258 		return ENOENT;
1259 
1260 	/*
1261 	 * Only superuser can mknod objects other than
1262 	 * directories, files, socks, fifo and links.
1263 	 *
1264 	 * Create an object require -WX permission in the parent directory
1265 	 */
1266 	switch (vap->va_type) {
1267 	case VDIR:	/* FALLTHROUGH */
1268 	case VREG:	/* FALLTHROUGH */
1269 	case VFIFO:	/* FALLTHROUGH */
1270 	case VSOCK:
1271 		break;
1272 	default:	/* VNON, VBLK, VCHR, VBAD */
1273 		if (!puffs_cred_isjuggernaut(pcn->pcn_cred))
1274 			return EACCES;
1275 		break;
1276 	}
1277 
1278 
1279 	ps = puffs_getspecific(pu);
1280 	path = pcn->pcn_name;
1281 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
1282 
1283 	/*
1284 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1285 	 */
1286 	pm = ps->ps_new_msg(pu, opc, FUSE_MKNOD, len, pcn->pcn_cred);
1287 	fmi = GET_INPAYLOAD(ps, pm, fuse_mknod_in);
1288 	fmi->mode = vap->va_mode | VTTOIF(vap->va_type);
1289 	fmi->rdev = (uint32_t)vap->va_rdev;
1290 	fmi->umask = 0; 	/* Seems unused bu libfuse */
1291 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
1292 
1293 	return node_mk_common(pu, opc, pni, pcn, pm);
1294 }
1295 
1296 
1297 int
1298 perfuse_node_open(pu, opc, mode, pcr)
1299 	struct puffs_usermount *pu;
1300 	puffs_cookie_t opc;
1301 	int mode;
1302 	const struct puffs_cred *pcr;
1303 {
1304 	struct perfuse_state *ps;
1305 	struct perfuse_node_data *pnd;
1306 	perfuse_msg_t *pm;
1307 	mode_t fmode;
1308 	int op;
1309 	struct fuse_open_in *foi;
1310 	struct fuse_open_out *foo;
1311 	struct puffs_node *pn;
1312 	int error;
1313 
1314 	ps = puffs_getspecific(pu);
1315 	pn = (struct puffs_node *)opc;
1316 	pnd = PERFUSE_NODE_DATA(opc);
1317 	pm = NULL;
1318 	error = 0;
1319 
1320 	if (pnd->pnd_flags & PND_REMOVED)
1321 		return ENOENT;
1322 
1323 	if (puffs_pn_getvap(pn)->va_type == VDIR)
1324 		op = FUSE_OPENDIR;
1325 	else
1326 		op = FUSE_OPEN;
1327 
1328 	/*
1329 	 * libfuse docs says
1330 	 * - O_CREAT and O_EXCL should never be set.
1331 	 * - O_TRUNC may be used if mount option atomic_o_trunc is used XXX
1332 	 *
1333 	 * O_APPEND makes no sense since FUSE always sends
1334 	 * the file offset for write operations. If the
1335 	 * filesystem uses pwrite(), O_APPEND would cause
1336 	 * the offset to be ignored and cause file corruption.
1337 	 */
1338 	mode &= ~(O_CREAT|O_EXCL|O_APPEND);
1339 
1340 	/*
1341 	 * Do not open twice, and do not reopen for reading
1342 	 * if we already have write handle.
1343 	 */
1344 	if (((mode & FREAD) && (pnd->pnd_flags & PND_RFH)) ||
1345 	    ((mode & FREAD) && (pnd->pnd_flags & PND_WFH)) ||
1346 	    ((mode & FWRITE) && (pnd->pnd_flags & PND_WFH))) {
1347 		error = 0;
1348 		goto out;
1349 	}
1350 
1351 	/*
1352 	 * Queue open on a node so that we do not open
1353 	 * twice. This would be better with read and
1354 	 * write distinguished.
1355 	 */
1356 	while (pnd->pnd_flags & PND_INOPEN)
1357 		requeue_request(pu, opc, PCQ_OPEN);
1358 	pnd->pnd_flags |= PND_INOPEN;
1359 
1360 	/*
1361 	 * Convert PUFFS mode to FUSE mode: convert FREAD/FWRITE
1362 	 * to O_RDONLY/O_WRONLY while perserving the other options.
1363 	 */
1364 	fmode = mode & ~(FREAD|FWRITE);
1365 	fmode |= (mode & FWRITE) ? O_RDWR : O_RDONLY;
1366 
1367 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*foi), pcr);
1368 	foi = GET_INPAYLOAD(ps, pm, fuse_open_in);
1369 	foi->flags = fmode;
1370 	foi->unused = 0;
1371 
1372 	if ((error = xchg_msg(pu, opc, pm, sizeof(*foo), wait_reply)) != 0)
1373 		goto out;
1374 
1375 	foo = GET_OUTPAYLOAD(ps, pm, fuse_open_out);
1376 
1377 	/*
1378 	 * Save the file handle in node private data
1379 	 * so that we can reuse it later
1380 	 */
1381 	perfuse_new_fh(opc, foo->fh, mode);
1382 
1383 #ifdef PERFUSE_DEBUG
1384 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1385 		DPRINTF("%s: opc = %p, file = \"%s\", "
1386 			"ino = %"PRId64", %s%sfh = 0x%"PRIx64"\n",
1387 			__func__, (void *)opc, perfuse_node_path(opc),
1388 			pnd->pnd_ino, mode & FREAD ? "r" : "",
1389 			mode & FWRITE ? "w" : "", foo->fh);
1390 #endif
1391 
1392 out:
1393 	if (pm != NULL)
1394 		ps->ps_destroy_msg(pm);
1395 
1396 	pnd->pnd_flags &= ~PND_INOPEN;
1397 	(void)dequeue_requests(ps, opc, PCQ_OPEN, DEQUEUE_ALL);
1398 
1399 	return error;
1400 }
1401 
1402 /* ARGSUSED0 */
1403 int
1404 perfuse_node_close(pu, opc, flags, pcr)
1405 	struct puffs_usermount *pu;
1406 	puffs_cookie_t opc;
1407 	int flags;
1408 	const struct puffs_cred *pcr;
1409 {
1410 	struct perfuse_node_data *pnd;
1411 
1412 	pnd = PERFUSE_NODE_DATA(opc);
1413 
1414 	if (!(pnd->pnd_flags & PND_OPEN))
1415 		return EBADF;
1416 
1417 	/*
1418 	 * Actual close is postponed at inactive time.
1419 	 */
1420 	return 0;
1421 }
1422 
1423 int
1424 perfuse_node_access(pu, opc, mode, pcr)
1425 	struct puffs_usermount *pu;
1426 	puffs_cookie_t opc;
1427 	int mode;
1428 	const struct puffs_cred *pcr;
1429 {
1430 	perfuse_msg_t *pm;
1431 	struct perfuse_state *ps;
1432 	struct fuse_access_in *fai;
1433 	int error;
1434 
1435 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1436 		return ENOENT;
1437 
1438 	/*
1439 	 * If we previously detected the filesystem does not
1440 	 * implement access(), short-circuit the call and skip
1441 	 * to libpuffs access() emulation.
1442 	 */
1443 	ps = puffs_getspecific(pu);
1444 	if (ps->ps_flags & PS_NO_ACCESS) {
1445 		const struct vattr *vap;
1446 
1447 		vap = puffs_pn_getvap((struct puffs_node *)opc);
1448 
1449 		error = puffs_access(IFTOVT(vap->va_mode),
1450 				     vap->va_mode & ACCESSPERMS,
1451 				     vap->va_uid, vap->va_gid,
1452 				     (mode_t)mode, pcr);
1453 		return error;
1454 	}
1455 
1456 	/*
1457 	 * Plain access call
1458 	 */
1459 	pm = ps->ps_new_msg(pu, opc, FUSE_ACCESS, sizeof(*fai), pcr);
1460 	fai = GET_INPAYLOAD(ps, pm, fuse_access_in);
1461 	fai->mask = 0;
1462 	fai->mask |= (mode & PUFFS_VREAD) ? R_OK : 0;
1463 	fai->mask |= (mode & PUFFS_VWRITE) ? W_OK : 0;
1464 	fai->mask |= (mode & PUFFS_VEXEC) ? X_OK : 0;
1465 
1466 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
1467 
1468 	ps->ps_destroy_msg(pm);
1469 
1470 	/*
1471 	 * If unimplemented, start over with emulation
1472 	 */
1473 	if (error == ENOSYS) {
1474 		ps->ps_flags |= PS_NO_ACCESS;
1475 		return perfuse_node_access(pu, opc, mode, pcr);
1476 	}
1477 
1478 	return error;
1479 }
1480 
1481 int
1482 perfuse_node_getattr(pu, opc, vap, pcr)
1483 	struct puffs_usermount *pu;
1484 	puffs_cookie_t opc;
1485 	struct vattr *vap;
1486 	const struct puffs_cred *pcr;
1487 {
1488 	perfuse_msg_t *pm;
1489 	struct perfuse_state *ps;
1490 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
1491 	struct fuse_getattr_in *fgi;
1492 	struct fuse_attr_out *fao;
1493 	int error;
1494 
1495 	if (pnd->pnd_flags & PND_REMOVED)
1496 		return ENOENT;
1497 
1498 	/*
1499 	 * Serialize size access, see comment in perfuse_node_setattr().
1500 	 */
1501 	while (pnd->pnd_flags & PND_INRESIZE)
1502 		requeue_request(pu, opc, PCQ_RESIZE);
1503 	pnd->pnd_flags |= PND_INRESIZE;
1504 
1505 	ps = puffs_getspecific(pu);
1506 
1507 	/*
1508 	 * FUSE_GETATTR_FH must be set in fgi->flags
1509 	 * if we use for fgi->fh
1510 	 */
1511 	pm = ps->ps_new_msg(pu, opc, FUSE_GETATTR, sizeof(*fgi), pcr);
1512 	fgi = GET_INPAYLOAD(ps, pm, fuse_getattr_in);
1513 	fgi->getattr_flags = 0;
1514 	fgi->dummy = 0;
1515 	fgi->fh = 0;
1516 
1517 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_OPEN) {
1518 		fgi->fh = perfuse_get_fh(opc, FREAD);
1519 		fgi->getattr_flags |= FUSE_GETATTR_FH;
1520 	}
1521 
1522 #ifdef PERFUSE_DEBUG
1523 	if (perfuse_diagflags & PDF_RESIZE)
1524 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__, (void *)opc,
1525 		    vap->va_size);
1526 #endif
1527 
1528 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
1529 		goto out;
1530 
1531 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1532 
1533 #ifdef PERFUSE_DEBUG
1534 	if (perfuse_diagflags & PDF_RESIZE)
1535 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1536 		    (void *)opc, vap->va_size, fao->attr.size);
1537 #endif
1538 
1539 	/*
1540 	 * The message from filesystem has a cache timeout
1541 	 * XXX this is ignored yet, is that right?
1542 	 *
1543 	 * We also set birthtime, flags, filerev,vaflags to 0.
1544 	 * This seems the best bet, since the information is
1545 	 * not available from filesystem.
1546 	 */
1547 	fuse_attr_to_vap(ps, vap, &fao->attr);
1548 
1549 out:
1550 	ps->ps_destroy_msg(pm);
1551 
1552 	pnd->pnd_flags &= ~PND_INRESIZE;
1553 	(void)dequeue_requests(ps, opc, PCQ_RESIZE, DEQUEUE_ALL);
1554 
1555 	return error;
1556 }
1557 
1558 int
1559 perfuse_node_setattr(pu, opc, vap, pcr)
1560 	struct puffs_usermount *pu;
1561 	puffs_cookie_t opc;
1562 	const struct vattr *vap;
1563 	const struct puffs_cred *pcr;
1564 {
1565 	perfuse_msg_t *pm;
1566 	uint64_t fh;
1567 	struct perfuse_state *ps;
1568 	struct perfuse_node_data *pnd;
1569 	struct fuse_setattr_in *fsi;
1570 	struct fuse_attr_out *fao;
1571 	struct vattr *old_va;
1572 	int error;
1573 #ifdef PERFUSE_DEBUG
1574 	struct vattr *old_vap;
1575 	int resize_debug = 0;
1576 #endif
1577 
1578 	ps = puffs_getspecific(pu);
1579 	pnd = PERFUSE_NODE_DATA(opc);
1580 	pm = NULL;
1581 
1582 	/*
1583 	 * The only operation we can do once the file is removed
1584 	 * is to resize it, and we can do it only if it is open.
1585 	 * Do not even send the operation to the filesystem: the
1586 	 * file is not there anymore.
1587 	 */
1588 	if (pnd->pnd_flags & PND_REMOVED) {
1589 		if (!(pnd->pnd_flags & PND_OPEN))
1590 			return ENOENT;
1591 
1592 		error = 0;
1593 		goto out;
1594 	}
1595 
1596 	old_va = puffs_pn_getvap((struct puffs_node *)opc);
1597 
1598 	/*
1599 	 * Check for permission to change size
1600 	 */
1601 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1602 	    (error = mode_access(opc, pcr, PUFFS_VWRITE)) != 0)
1603 		return error;
1604 
1605 	/*
1606 	 * Check for permission to change dates
1607 	 */
1608 	if (((vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1609 	     (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL)) &&
1610 	    (puffs_access_times(old_va->va_uid, old_va->va_gid,
1611 				old_va->va_mode, 0, pcr) != 0))
1612 		return EACCES;
1613 
1614 	/*
1615 	 * Check for permission to change owner and group
1616 	 */
1617 	if (((vap->va_uid != (uid_t)PUFFS_VNOVAL) ||
1618 	     (vap->va_gid != (gid_t)PUFFS_VNOVAL)) &&
1619 	    (puffs_access_chown(old_va->va_uid, old_va->va_gid,
1620 				vap->va_uid, vap->va_gid, pcr)) != 0)
1621 		return EACCES;
1622 
1623 	/*
1624 	 * Check for permission to change permissions
1625 	 */
1626 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1627 	    (puffs_access_chmod(old_va->va_uid, old_va->va_gid,
1628 				old_va->va_type, vap->va_mode, pcr)) != 0)
1629 		return EACCES;
1630 
1631 	pm = ps->ps_new_msg(pu, opc, FUSE_SETATTR, sizeof(*fsi), pcr);
1632 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
1633 	fsi->valid = 0;
1634 
1635 	/*
1636 	 * Get a fh if the node is open for writing
1637 	 */
1638 	if (pnd->pnd_flags & PND_WFH) {
1639 		fh = perfuse_get_fh(opc, FWRITE);
1640 		fsi->fh = fh;
1641 		fsi->valid |= FUSE_FATTR_FH;
1642 	}
1643 
1644 	if (vap->va_size != (u_quad_t)PUFFS_VNOVAL) {
1645 		fsi->size = vap->va_size;
1646 		fsi->valid |= FUSE_FATTR_SIZE;
1647 
1648 		/*
1649 		 * Serialize anything that can touch file size
1650 		 * to avoid reordered GETATTR and SETATTR.
1651 		 * Out of order SETATTR can report stale size,
1652 		 * which will cause the kernel to truncate the file.
1653 		 */
1654 		while (pnd->pnd_flags & PND_INRESIZE)
1655 			requeue_request(pu, opc, PCQ_RESIZE);
1656 		pnd->pnd_flags |= PND_INRESIZE;
1657 	}
1658 
1659 	/*
1660  	 * Setting mtime without atime or vice versa leads to
1661 	 * dates being reset to Epoch on glusterfs. If one
1662 	 * is missing, use the old value.
1663  	 */
1664 	if ((vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1665 	    (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL)) {
1666 
1667 		if (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) {
1668 			fsi->atime = vap->va_atime.tv_sec;
1669 			fsi->atimensec = (uint32_t)vap->va_atime.tv_nsec;
1670 		} else {
1671 			fsi->atime = old_va->va_atime.tv_sec;
1672 			fsi->atimensec = (uint32_t)old_va->va_atime.tv_nsec;
1673 		}
1674 
1675 		if (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) {
1676 			fsi->mtime = vap->va_mtime.tv_sec;
1677 			fsi->mtimensec = (uint32_t)vap->va_mtime.tv_nsec;
1678 		} else {
1679 			fsi->mtime = old_va->va_mtime.tv_sec;
1680 			fsi->mtimensec = (uint32_t)old_va->va_mtime.tv_nsec;
1681 		}
1682 
1683 		fsi->valid |= (FUSE_FATTR_MTIME|FUSE_FATTR_ATIME);
1684 	}
1685 
1686 	if (vap->va_mode != (mode_t)PUFFS_VNOVAL) {
1687 		fsi->mode = vap->va_mode;
1688 		fsi->valid |= FUSE_FATTR_MODE;
1689 	}
1690 
1691 	if (vap->va_uid != (uid_t)PUFFS_VNOVAL) {
1692 		fsi->uid = vap->va_uid;
1693 		fsi->valid |= FUSE_FATTR_UID;
1694 	}
1695 
1696 	if (vap->va_gid != (gid_t)PUFFS_VNOVAL) {
1697 		fsi->gid = vap->va_gid;
1698 		fsi->valid |= FUSE_FATTR_GID;
1699 	}
1700 
1701 	if (pnd->pnd_lock_owner != 0) {
1702 		fsi->lock_owner = pnd->pnd_lock_owner;
1703 		fsi->valid |= FUSE_FATTR_LOCKOWNER;
1704 	}
1705 
1706 	/*
1707 	 * ftruncate() sends only va_size, and metadata cache
1708 	 * flush adds va_atime and va_mtime. Some FUSE
1709 	 * filesystems will attempt to detect ftruncate by
1710 	 * checking for FATTR_SIZE being set without
1711 	 * FATTR_UID|FATTR_GID|FATTR_ATIME|FATTR_MTIME|FATTR_MODE
1712 	 *
1713 	 * Try to adapt and remove FATTR_ATIME|FATTR_MTIME
1714 	 * if we suspect a ftruncate().
1715 	 */
1716 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1717 	    ((vap->va_mode == (mode_t)PUFFS_VNOVAL) &&
1718 	     (vap->va_uid == (uid_t)PUFFS_VNOVAL) &&
1719 	     (vap->va_gid == (gid_t)PUFFS_VNOVAL))) {
1720 		fsi->atime = 0;
1721 		fsi->atimensec = 0;
1722 		fsi->mtime = 0;
1723 		fsi->mtimensec = 0;
1724 		fsi->valid &= ~(FUSE_FATTR_ATIME|FUSE_FATTR_MTIME);
1725 	}
1726 
1727 	/*
1728 	 * If nothing remain, discard the operation.
1729 	 */
1730 	if (!(fsi->valid & (FUSE_FATTR_SIZE|FUSE_FATTR_ATIME|FUSE_FATTR_MTIME|
1731 			    FUSE_FATTR_MODE|FUSE_FATTR_UID|FUSE_FATTR_GID))) {
1732 		error = 0;
1733 		goto out;
1734 	}
1735 
1736 #ifdef PERFUSE_DEBUG
1737 	old_vap = puffs_pn_getvap((struct puffs_node *)opc);
1738 
1739 	if ((perfuse_diagflags & PDF_RESIZE) &&
1740 	    (old_vap->va_size != (u_quad_t)PUFFS_VNOVAL)) {
1741 		resize_debug = 1;
1742 
1743 		DPRINTF(">> %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1744 		    (void *)opc,
1745 		    puffs_pn_getvap((struct puffs_node *)opc)->va_size,
1746 		    fsi->size);
1747 	}
1748 #endif
1749 
1750 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
1751 		goto out;
1752 
1753 	/*
1754 	 * Copy back the new values
1755 	 */
1756 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1757 
1758 #ifdef PERFUSE_DEBUG
1759 	if (resize_debug)
1760 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1761 		    (void *)opc, old_vap->va_size, fao->attr.size);
1762 #endif
1763 
1764 	fuse_attr_to_vap(ps, old_va, &fao->attr);
1765 out:
1766 	if (pm != NULL)
1767 		ps->ps_destroy_msg(pm);
1768 
1769 	if (pnd->pnd_flags & PND_INRESIZE) {
1770 		pnd->pnd_flags &= ~PND_INRESIZE;
1771 		(void)dequeue_requests(ps, opc, PCQ_RESIZE, DEQUEUE_ALL);
1772 	}
1773 
1774 	return error;
1775 }
1776 
1777 int
1778 perfuse_node_poll(pu, opc, events)
1779 	struct puffs_usermount *pu;
1780 	puffs_cookie_t opc;
1781 	int *events;
1782 {
1783 	struct perfuse_state *ps;
1784 	perfuse_msg_t *pm;
1785 	struct fuse_poll_in *fpi;
1786 	struct fuse_poll_out *fpo;
1787 	int error;
1788 
1789 	ps = puffs_getspecific(pu);
1790 	/*
1791 	 * kh is set if FUSE_POLL_SCHEDULE_NOTIFY is set.
1792 	 *
1793 	 * XXX ps_new_msg() is called with NULL creds, which will
1794 	 * be interpreted as FUSE superuser. We have no way to
1795 	 * know the requesting process' credential, but since poll
1796 	 * is supposed to operate on a file that has been open,
1797 	 * permission should have already been checked at open time.
1798 	 * That still may breaks on filesystems that provides odd
1799 	 * semantics.
1800  	 */
1801 	pm = ps->ps_new_msg(pu, opc, FUSE_POLL, sizeof(*fpi), NULL);
1802 	fpi = GET_INPAYLOAD(ps, pm, fuse_poll_in);
1803 	fpi->fh = perfuse_get_fh(opc, FREAD);
1804 	fpi->kh = 0;
1805 	fpi->flags = 0;
1806 
1807 #ifdef PERFUSE_DEBUG
1808 	if (perfuse_diagflags & PDF_FH)
1809 		DPRINTF("%s: opc = %p, ino = %"PRId64", fh = 0x%"PRIx64"\n",
1810 			__func__, (void *)opc,
1811 			PERFUSE_NODE_DATA(opc)->pnd_ino, fpi->fh);
1812 #endif
1813 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fpo), wait_reply)) != 0)
1814 		goto out;
1815 
1816 	fpo = GET_OUTPAYLOAD(ps, pm, fuse_poll_out);
1817 	*events = fpo->revents;
1818 out:
1819 	ps->ps_destroy_msg(pm);
1820 
1821 	return error;
1822 }
1823 
1824 /* ARGSUSED0 */
1825 int
1826 perfuse_node_mmap(pu, opc, flags, pcr)
1827 	struct puffs_usermount *pu;
1828 	puffs_cookie_t opc;
1829 	int flags;
1830 	const struct puffs_cred *pcr;
1831 {
1832 	/*
1833 	 * Not implemented anymore in libfuse
1834 	 */
1835 	return ENOSYS;
1836 }
1837 
1838 /* ARGSUSED2 */
1839 int
1840 perfuse_node_fsync(pu, opc, pcr, flags, offlo, offhi)
1841 	struct puffs_usermount *pu;
1842 	puffs_cookie_t opc;
1843 	const struct puffs_cred *pcr;
1844 	int flags;
1845 	off_t offlo;
1846 	off_t offhi;
1847 {
1848 	int op;
1849 	perfuse_msg_t *pm;
1850 	struct perfuse_state *ps;
1851 	struct perfuse_node_data *pnd;
1852 	struct fuse_fsync_in *ffi;
1853 	uint64_t fh;
1854 	int error;
1855 
1856 	pm = NULL;
1857 	ps = puffs_getspecific(pu);
1858 	pnd = PERFUSE_NODE_DATA(opc);
1859 
1860 	/*
1861 	 * No need to sync a removed node
1862 	 */
1863 	if (pnd->pnd_flags & PND_REMOVED)
1864 		return 0;
1865 
1866 	/*
1867 	 * We do not sync closed files. They have been
1868 	 * sync at inactive time already.
1869 	 */
1870 	if (!(pnd->pnd_flags & PND_OPEN))
1871 		return 0;
1872 
1873 	if (puffs_pn_getvap((struct puffs_node *)opc)->va_type == VDIR)
1874 		op = FUSE_FSYNCDIR;
1875 	else 		/* VREG but also other types such as VLNK */
1876 		op = FUSE_FSYNC;
1877 
1878 	/*
1879 	 * Do not sync if there are no change to sync
1880 	 * XXX remove that test on files if we implement mmap
1881 	 */
1882 #ifdef PERFUSE_DEBUG
1883 	if (perfuse_diagflags & PDF_SYNC)
1884 		DPRINTF("%s: TEST opc = %p, file = \"%s\" is %sdirty\n",
1885 			__func__, (void*)opc, perfuse_node_path(opc),
1886 			pnd->pnd_flags & PND_DIRTY ? "" : "not ");
1887 #endif
1888 	if (!(pnd->pnd_flags & PND_DIRTY))
1889 		return 0;
1890 
1891 	/*
1892 	 * It seems NetBSD can call fsync without open first
1893 	 * glusterfs complain in such a situation:
1894 	 * "FSYNC() ERR => -1 (Invalid argument)"
1895 	 * The file will be closed at inactive time.
1896 	 *
1897 	 * We open the directory for reading in order to sync.
1898 	 * This sounds rather counterintuitive, but it works.
1899 	 */
1900 	if (!(pnd->pnd_flags & PND_WFH)) {
1901 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
1902 			goto out;
1903 	}
1904 
1905 	if (op == FUSE_FSYNCDIR)
1906 		fh = perfuse_get_fh(opc, FREAD);
1907 	else
1908 		fh = perfuse_get_fh(opc, FWRITE);
1909 
1910 	/*
1911 	 * If fsync_flags  is set, meta data should not be flushed.
1912 	 */
1913 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*ffi), pcr);
1914 	ffi = GET_INPAYLOAD(ps, pm, fuse_fsync_in);
1915 	ffi->fh = fh;
1916 	ffi->fsync_flags = (flags & FFILESYNC) ? 0 : 1;
1917 
1918 #ifdef PERFUSE_DEBUG
1919 	if (perfuse_diagflags & PDF_FH)
1920 		DPRINTF("%s: opc = %p, ino = %"PRId64", fh = 0x%"PRIx64"\n",
1921 			__func__, (void *)opc,
1922 			PERFUSE_NODE_DATA(opc)->pnd_ino, ffi->fh);
1923 #endif
1924 
1925 	if ((error = xchg_msg(pu, opc, pm,
1926 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
1927 		goto out;
1928 
1929 	/*
1930 	 * No reply beyond fuse_out_header: nothing to do on success
1931 	 * just clear the dirty flag
1932 	 */
1933 	pnd->pnd_flags &= ~PND_DIRTY;
1934 
1935 #ifdef PERFUSE_DEBUG
1936 	if (perfuse_diagflags & PDF_SYNC)
1937 		DPRINTF("%s: CLEAR opc = %p, file = \"%s\"\n",
1938 			__func__, (void*)opc, perfuse_node_path(opc));
1939 #endif
1940 
1941 out:
1942 	/*
1943 	 * ENOSYS is not returned to kernel,
1944 	 */
1945 	if (error == ENOSYS)
1946 		error = 0;
1947 
1948 	if (pm != NULL)
1949 		ps->ps_destroy_msg(pm);
1950 
1951 	return error;
1952 }
1953 
1954 /* ARGSUSED0 */
1955 int
1956 perfuse_node_seek(pu, opc, oldoff, newoff,  pcr)
1957 	struct puffs_usermount *pu;
1958 	puffs_cookie_t opc;
1959 	off_t oldoff;
1960 	off_t newoff;
1961 	const struct puffs_cred *pcr;
1962 {
1963 	return 0;
1964 }
1965 
1966 int
1967 perfuse_node_remove(pu, opc, targ, pcn)
1968 	struct puffs_usermount *pu;
1969 	puffs_cookie_t opc;
1970 	puffs_cookie_t targ;
1971 	const struct puffs_cn *pcn;
1972 {
1973 	struct perfuse_state *ps;
1974 	struct perfuse_node_data *pnd;
1975 	perfuse_msg_t *pm;
1976 	char *path;
1977 	const char *name;
1978 	size_t len;
1979 	int error;
1980 
1981 	pnd = PERFUSE_NODE_DATA(opc);
1982 
1983 	if ((pnd->pnd_flags & PND_REMOVED) ||
1984 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
1985 		return ENOENT;
1986 
1987 #ifdef PERFUSE_DEBUG
1988 	if (targ == NULL)
1989 		DERRX(EX_SOFTWARE, "%s: targ is NULL", __func__);
1990 
1991 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1992 		DPRINTF("%s: opc = %p, remove opc = %p, file = \"%s\"\n",
1993 			__func__, (void *)opc, (void *)targ, pcn->pcn_name);
1994 #endif
1995 	/*
1996 	 * Await for all operations on the deleted node to drain,
1997 	 * as the filesystem may be confused to have it deleted
1998 	 * during a getattr
1999 	 */
2000 	while (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_INXCHG)
2001 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2002 
2003 	ps = puffs_getspecific(pu);
2004 	pnd = PERFUSE_NODE_DATA(opc);
2005 	name = pcn->pcn_name;
2006 	len = pcn->pcn_namelen + 1;
2007 
2008 	pm = ps->ps_new_msg(pu, opc, FUSE_UNLINK, len, pcn->pcn_cred);
2009 	path = _GET_INPAYLOAD(ps, pm, char *);
2010 	(void)strlcpy(path, name, len);
2011 
2012 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2013 		goto out;
2014 
2015 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2016 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2017 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2018 
2019 	/*
2020 	 * The parent directory needs a sync
2021 	 */
2022 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2023 
2024 #ifdef PERFUSE_DEBUG
2025 	if (perfuse_diagflags & PDF_FILENAME)
2026 		DPRINTF("%s: remove nodeid = %"PRId64" file = \"%s\"\n",
2027 			__func__, PERFUSE_NODE_DATA(targ)->pnd_ino,
2028 			pcn->pcn_name);
2029 #endif
2030 out:
2031 	ps->ps_destroy_msg(pm);
2032 
2033 	return error;
2034 }
2035 
2036 int
2037 perfuse_node_link(pu, opc, targ, pcn)
2038 	struct puffs_usermount *pu;
2039 	puffs_cookie_t opc;
2040 	puffs_cookie_t targ;
2041 	const struct puffs_cn *pcn;
2042 {
2043 	struct perfuse_state *ps;
2044 	perfuse_msg_t *pm;
2045 	const char *name;
2046 	size_t len;
2047 	struct puffs_node *pn;
2048 	struct fuse_link_in *fli;
2049 	int error;
2050 
2051 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2052 		return ENOENT;
2053 
2054 	ps = puffs_getspecific(pu);
2055 	pn = (struct puffs_node *)targ;
2056 	name = pcn->pcn_name;
2057 	len =  sizeof(*fli) + pcn->pcn_namelen + 1;
2058 
2059 	pm = ps->ps_new_msg(pu, opc, FUSE_LINK, len, pcn->pcn_cred);
2060 	fli = GET_INPAYLOAD(ps, pm, fuse_link_in);
2061 	fli->oldnodeid = PERFUSE_NODE_DATA(pn)->pnd_ino;
2062 	(void)strlcpy((char *)(void *)(fli + 1), name, len - sizeof(*fli));
2063 
2064 	error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
2065 
2066 	ps->ps_destroy_msg(pm);
2067 
2068 	return error;
2069 }
2070 
2071 int
2072 perfuse_node_rename(pu, opc, src, pcn_src, targ_dir, targ, pcn_targ)
2073 	struct puffs_usermount *pu;
2074 	puffs_cookie_t opc;
2075 	puffs_cookie_t src;
2076 	const struct puffs_cn *pcn_src;
2077 	puffs_cookie_t targ_dir;
2078 	puffs_cookie_t targ;
2079 	const struct puffs_cn *pcn_targ;
2080 {
2081 	struct perfuse_state *ps;
2082 	perfuse_msg_t *pm;
2083 	struct fuse_rename_in *fri;
2084 	const char *newname;
2085 	const char *oldname;
2086 	char *np;
2087 	int error;
2088 	size_t len;
2089 	size_t newname_len;
2090 	size_t oldname_len;
2091 
2092 	if ((PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED) ||
2093 	    (PERFUSE_NODE_DATA(src)->pnd_flags & PND_REMOVED) ||
2094 	    (PERFUSE_NODE_DATA(targ_dir)->pnd_flags & PND_REMOVED))
2095 		return ENOENT;
2096 
2097 	/*
2098 	 * Await for all operations on the deleted node to drain,
2099 	 * as the filesystem may be confused to have it deleted
2100 	 * during a getattr
2101 	 */
2102 	if ((struct puffs_node *)targ != NULL) {
2103 		while (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_INXCHG)
2104 			requeue_request(pu, targ, PCQ_AFTERXCHG);
2105 	} else {
2106 		while (PERFUSE_NODE_DATA(src)->pnd_flags & PND_INXCHG)
2107 			requeue_request(pu, src, PCQ_AFTERXCHG);
2108 	}
2109 
2110 	ps = puffs_getspecific(pu);
2111 	newname =  pcn_targ->pcn_name;
2112 	newname_len = pcn_targ->pcn_namelen + 1;
2113 	oldname =  pcn_src->pcn_name;
2114 	oldname_len = pcn_src->pcn_namelen + 1;
2115 
2116 	len = sizeof(*fri) + oldname_len + newname_len;
2117 	pm = ps->ps_new_msg(pu, opc, FUSE_RENAME, len, pcn_targ->pcn_cred);
2118 	fri = GET_INPAYLOAD(ps, pm, fuse_rename_in);
2119 	fri->newdir = PERFUSE_NODE_DATA(targ_dir)->pnd_ino;
2120 	np = (char *)(void *)(fri + 1);
2121 	(void)strlcpy(np, oldname, oldname_len);
2122 	np += oldname_len;
2123 	(void)strlcpy(np, newname, newname_len);
2124 
2125 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2126 		goto out;
2127 
2128 	if (opc != targ_dir) {
2129 		struct perfuse_node_data *srcdir_pnd;
2130 		struct perfuse_node_data *dstdir_pnd;
2131 		struct perfuse_node_data *src_pnd;
2132 
2133 		srcdir_pnd = PERFUSE_NODE_DATA(opc);
2134 		dstdir_pnd = PERFUSE_NODE_DATA(targ_dir);
2135 		src_pnd = PERFUSE_NODE_DATA(src);
2136 
2137 		TAILQ_REMOVE(&srcdir_pnd->pnd_children, src_pnd, pnd_next);
2138 		TAILQ_INSERT_TAIL(&dstdir_pnd->pnd_children, src_pnd, pnd_next);
2139 
2140 		srcdir_pnd->pnd_childcount--;
2141 		dstdir_pnd->pnd_childcount++;
2142 
2143 		src_pnd->pnd_parent = targ_dir;
2144 
2145 		PERFUSE_NODE_DATA(targ_dir)->pnd_flags |= PND_DIRTY;
2146 	}
2147 
2148 	(void)strlcpy(PERFUSE_NODE_DATA(src)->pnd_name, newname, MAXPATHLEN);
2149 
2150 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2151 
2152 	if ((struct puffs_node *)targ != NULL)
2153 		PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2154 
2155 #ifdef PERFUSE_DEBUG
2156 	if (perfuse_diagflags & PDF_FILENAME)
2157 		DPRINTF("%s: nodeid = %"PRId64" file = \"%s\" renamed \"%s\" "
2158 			"nodeid = %"PRId64" -> nodeid = %"PRId64" \"%s\"\n",
2159 	 		__func__, PERFUSE_NODE_DATA(src)->pnd_ino,
2160 			pcn_src->pcn_name, pcn_targ->pcn_name,
2161 			PERFUSE_NODE_DATA(opc)->pnd_ino,
2162 			PERFUSE_NODE_DATA(targ_dir)->pnd_ino,
2163 			perfuse_node_path(targ_dir));
2164 #endif
2165 
2166 out:
2167 	if (pm != NULL)
2168 		ps->ps_destroy_msg(pm);
2169 
2170 	return error;
2171 }
2172 
2173 int
2174 perfuse_node_mkdir(pu, opc, pni, pcn, vap)
2175 	struct puffs_usermount *pu;
2176 	puffs_cookie_t opc;
2177 	struct puffs_newinfo *pni;
2178 	const struct puffs_cn *pcn;
2179 	const struct vattr *vap;
2180 {
2181 	struct perfuse_state *ps;
2182 	perfuse_msg_t *pm;
2183 	struct fuse_mkdir_in *fmi;
2184 	const char *path;
2185 	size_t len;
2186 
2187 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2188 		return ENOENT;
2189 
2190 	ps = puffs_getspecific(pu);
2191 	path = pcn->pcn_name;
2192 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
2193 
2194 	pm = ps->ps_new_msg(pu, opc, FUSE_MKDIR, len, pcn->pcn_cred);
2195 	fmi = GET_INPAYLOAD(ps, pm, fuse_mkdir_in);
2196 	fmi->mode = vap->va_mode;
2197 	fmi->umask = 0; 	/* Seems unused by libfuse? */
2198 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
2199 
2200 	return node_mk_common(pu, opc, pni, pcn, pm);
2201 }
2202 
2203 
2204 int
2205 perfuse_node_rmdir(pu, opc, targ, pcn)
2206 	struct puffs_usermount *pu;
2207 	puffs_cookie_t opc;
2208 	puffs_cookie_t targ;
2209 	const struct puffs_cn *pcn;
2210 {
2211 	struct perfuse_state *ps;
2212 	struct perfuse_node_data *pnd;
2213 	perfuse_msg_t *pm;
2214 	char *path;
2215 	const char *name;
2216 	size_t len;
2217 	int error;
2218 
2219 	pnd = PERFUSE_NODE_DATA(opc);
2220 	if (pnd->pnd_flags & PND_REMOVED)
2221 		return ENOENT;
2222 
2223 	/*
2224 	 * Await for all operations on the deleted node to drain,
2225 	 * as the filesystem may be confused to have it deleted
2226 	 * during a getattr
2227 	 */
2228 	while (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_INXCHG)
2229 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2230 
2231 	ps = puffs_getspecific(pu);
2232 	name = pcn->pcn_name;
2233 	len = pcn->pcn_namelen + 1;
2234 
2235 	pm = ps->ps_new_msg(pu, opc, FUSE_RMDIR, len, pcn->pcn_cred);
2236 	path = _GET_INPAYLOAD(ps, pm, char *);
2237 	(void)strlcpy(path, name, len);
2238 
2239 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2240 		goto out;
2241 
2242 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2243 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2244 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2245 
2246 	/*
2247 	 * The parent directory needs a sync
2248 	 */
2249 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2250 
2251 #ifdef PERFUSE_DEBUG
2252 	if (perfuse_diagflags & PDF_FILENAME)
2253 		DPRINTF("%s: remove nodeid = %"PRId64" file = \"%s\"\n",
2254 			__func__, PERFUSE_NODE_DATA(targ)->pnd_ino,
2255 			perfuse_node_path(targ));
2256 #endif
2257 out:
2258 	ps->ps_destroy_msg(pm);
2259 
2260 	return error;
2261 }
2262 
2263 /* vap is unused */
2264 /* ARGSUSED4 */
2265 int
2266 perfuse_node_symlink(pu, opc, pni, pcn_src, vap, link_target)
2267 	struct puffs_usermount *pu;
2268 	puffs_cookie_t opc;
2269 	struct puffs_newinfo *pni;
2270 	const struct puffs_cn *pcn_src;
2271 	const struct vattr *vap;
2272 	const char *link_target;
2273 {
2274 	struct perfuse_state *ps;
2275 	perfuse_msg_t *pm;
2276 	char *np;
2277 	const char *path;
2278 	size_t path_len;
2279 	size_t linkname_len;
2280 	size_t len;
2281 
2282 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2283 		return ENOENT;
2284 
2285 	ps = puffs_getspecific(pu);
2286 	path = pcn_src->pcn_name;
2287 	path_len = pcn_src->pcn_namelen + 1;
2288 	linkname_len = strlen(link_target) + 1;
2289 	len = path_len + linkname_len;
2290 
2291 	pm = ps->ps_new_msg(pu, opc, FUSE_SYMLINK, len, pcn_src->pcn_cred);
2292 	np = _GET_INPAYLOAD(ps, pm, char *);
2293 	(void)strlcpy(np, path, path_len);
2294 	np += path_len;
2295 	(void)strlcpy(np, link_target, linkname_len);
2296 
2297 	return node_mk_common(pu, opc, pni, pcn_src, pm);
2298 }
2299 
2300 /* ARGSUSED4 */
2301 int
2302 perfuse_node_readdir(pu, opc, dent, readoff,
2303 		     reslen, pcr, eofflag, cookies, ncookies)
2304 	struct puffs_usermount *pu;
2305 	puffs_cookie_t opc;
2306 	struct dirent *dent;
2307 	off_t *readoff;
2308 	size_t *reslen;
2309 	const struct puffs_cred *pcr;
2310 	int *eofflag;
2311 	off_t *cookies;
2312 	size_t *ncookies;
2313 {
2314 	perfuse_msg_t *pm;
2315 	uint64_t fh;
2316 	struct perfuse_state *ps;
2317 	struct perfuse_node_data *pnd;
2318 	struct fuse_read_in *fri;
2319 	struct fuse_out_header *foh;
2320 	struct fuse_dirent *fd;
2321 	size_t foh_len;
2322 	int error;
2323 	size_t fd_maxlen;
2324 
2325 	pm = NULL;
2326 	error = 0;
2327 	ps = puffs_getspecific(pu);
2328 
2329 	/*
2330 	 * readdir state is kept at node level, and several readdir
2331 	 * requests can be issued at the same time on the same node.
2332 	 * We need to queue requests so that only one is in readdir
2333 	 * code at the same time.
2334 	 */
2335 	pnd = PERFUSE_NODE_DATA(opc);
2336 	while (pnd->pnd_flags & PND_INREADDIR)
2337 		requeue_request(pu, opc, PCQ_READDIR);
2338 	pnd->pnd_flags |= PND_INREADDIR;
2339 
2340 #ifdef PERFUSE_DEBUG
2341 	if (perfuse_diagflags & PDF_READDIR)
2342 		DPRINTF("%s: READDIR opc = %p enter critical section\n",
2343 			__func__, (void *)opc);
2344 #endif
2345 	/*
2346 	 * Re-initialize pnd->pnd_fd_cookie on the first readdir for a node
2347 	 */
2348 	if (*readoff == 0)
2349 		pnd->pnd_fd_cookie = 0;
2350 
2351 	/*
2352 	 * Do we already have the data bufered?
2353 	 */
2354 	if (pnd->pnd_dirent != NULL)
2355 		goto out;
2356 	pnd->pnd_dirent_len = 0;
2357 
2358 	/*
2359 	 * It seems NetBSD can call readdir without open first
2360 	 * libfuse will crash if it is done that way, hence open first.
2361 	 */
2362 	if (!(pnd->pnd_flags & PND_OPEN)) {
2363 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2364 			goto out;
2365 	}
2366 
2367 	fh = perfuse_get_fh(opc, FREAD);
2368 
2369 #ifdef PERFUSE_DEBUG
2370 	if (perfuse_diagflags & PDF_FH)
2371 		DPRINTF("%s: opc = %p, ino = %"PRId64", rfh = 0x%"PRIx64"\n",
2372 			__func__, (void *)opc,
2373 			PERFUSE_NODE_DATA(opc)->pnd_ino, fh);
2374 #endif
2375 
2376 	pnd->pnd_all_fd = NULL;
2377 	pnd->pnd_all_fd_len = 0;
2378 	fd_maxlen = ps->ps_max_readahead - sizeof(*foh);
2379 
2380 	do {
2381 		size_t fd_len;
2382 		char *afdp;
2383 
2384 		pm = ps->ps_new_msg(pu, opc, FUSE_READDIR, sizeof(*fri), pcr);
2385 
2386 		/*
2387 		 * read_flags, lock_owner and flags are unused in libfuse
2388 		 */
2389 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
2390 		fri->fh = fh;
2391 		fri->offset = pnd->pnd_fd_cookie;
2392 		fri->size = (uint32_t)fd_maxlen;
2393 		fri->read_flags = 0;
2394 		fri->lock_owner = 0;
2395 		fri->flags = 0;
2396 
2397 		if ((error = xchg_msg(pu, opc, pm,
2398 				      UNSPEC_REPLY_LEN, wait_reply)) != 0)
2399 			goto out;
2400 
2401 		/*
2402 		 * There are many puffs_framebufs calls later,
2403 		 * therefore foh will not be valid for a long time.
2404 		 * Just get the length and forget it.
2405 		 */
2406 		foh = GET_OUTHDR(ps, pm);
2407 		foh_len = foh->len;
2408 
2409 		/*
2410 		 * Empty read: we reached the end of the buffer.
2411 		 */
2412 		if (foh_len == sizeof(*foh)) {
2413 			*eofflag = 1;
2414 			break;
2415 		}
2416 
2417 		/*
2418 		 * Check for corrupted message.
2419 		 */
2420 		if (foh_len < sizeof(*foh) + sizeof(*fd)) {
2421 			DWARNX("readdir reply too short");
2422 			error = EIO;
2423 			goto out;
2424 		}
2425 
2426 
2427 		fd = GET_OUTPAYLOAD(ps, pm, fuse_dirent);
2428 		fd_len = foh_len - sizeof(*foh);
2429 
2430 		pnd->pnd_all_fd = realloc(pnd->pnd_all_fd,
2431 					  pnd->pnd_all_fd_len + fd_len);
2432 		if (pnd->pnd_all_fd  == NULL)
2433 			DERR(EX_OSERR, "%s: malloc failed", __func__);
2434 
2435 		afdp = (char *)(void *)pnd->pnd_all_fd + pnd->pnd_all_fd_len;
2436 		(void)memcpy(afdp, fd, fd_len);
2437 
2438 		pnd->pnd_all_fd_len += fd_len;
2439 
2440 		/*
2441 		 * The fd->off field is used as a cookie for
2442 		 * resuming the next readdir() where this one was left.
2443 	 	 */
2444 		pnd->pnd_fd_cookie = readdir_last_cookie(fd, fd_len);
2445 
2446 		ps->ps_destroy_msg(pm);
2447 		pm = NULL;
2448 	} while (1 /* CONSTCOND */);
2449 
2450 	if (pnd->pnd_all_fd != NULL) {
2451 		if (fuse_to_dirent(pu, opc, pnd->pnd_all_fd,
2452 				   pnd->pnd_all_fd_len) == -1)
2453 			error = EIO;
2454 	}
2455 
2456 out:
2457 	if (pnd->pnd_all_fd != NULL) {
2458 		free(pnd->pnd_all_fd);
2459 		pnd->pnd_all_fd = NULL;
2460 		pnd->pnd_all_fd_len = 0;
2461 	}
2462 
2463 	if (pm != NULL)
2464 		ps->ps_destroy_msg(pm);
2465 
2466 	if (error == 0)
2467 		error = readdir_buffered(opc, dent, readoff, reslen);
2468 
2469 	/*
2470 	 * Schedule queued readdir requests
2471 	 */
2472 	pnd->pnd_flags &= ~PND_INREADDIR;
2473 	(void)dequeue_requests(ps, opc, PCQ_READDIR, DEQUEUE_ALL);
2474 
2475 #ifdef PERFUSE_DEBUG
2476 	if (perfuse_diagflags & PDF_READDIR)
2477 		DPRINTF("%s: READDIR opc = %p exit critical section\n",
2478 			__func__, (void *)opc);
2479 #endif
2480 
2481 	return error;
2482 }
2483 
2484 int
2485 perfuse_node_readlink(pu, opc, pcr, linkname, linklen)
2486 	struct puffs_usermount *pu;
2487 	puffs_cookie_t opc;
2488 	const struct puffs_cred *pcr;
2489 	char *linkname;
2490 	size_t *linklen;
2491 {
2492 	struct perfuse_state *ps;
2493 	perfuse_msg_t *pm;
2494 	int error;
2495 	size_t len;
2496 	struct fuse_out_header *foh;
2497 
2498 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2499 		return ENOENT;
2500 
2501 	ps = puffs_getspecific(pu);
2502 
2503 	pm = ps->ps_new_msg(pu, opc, FUSE_READLINK, 0, pcr);
2504 
2505 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2506 		goto out;
2507 
2508 	foh = GET_OUTHDR(ps, pm);
2509 	len = foh->len - sizeof(*foh);
2510 	if (len > *linklen)
2511 		DERRX(EX_PROTOCOL, "path len = %zd too long", len);
2512 	if (len == 0)
2513 		DERRX(EX_PROTOCOL, "path len = %zd too short", len);
2514 
2515 	/*
2516 	 * FUSE filesystems return a NUL terminated string, we
2517 	 * do not want to trailing \0
2518 	 */
2519 	*linklen = len - 1;
2520 	(void)memcpy(linkname, _GET_OUTPAYLOAD(ps, pm, char *), len);
2521 out:
2522 	ps->ps_destroy_msg(pm);
2523 
2524 	return error;
2525 }
2526 
2527 int
2528 perfuse_node_reclaim(pu, opc)
2529 	struct puffs_usermount *pu;
2530 	puffs_cookie_t opc;
2531 {
2532 	struct perfuse_state *ps;
2533 	perfuse_msg_t *pm;
2534 	struct perfuse_node_data *pnd;
2535 	struct fuse_forget_in *ffi;
2536 	struct puffs_node *pn;
2537 	struct puffs_node *pn_root;
2538 
2539 	ps = puffs_getspecific(pu);
2540 	pnd = PERFUSE_NODE_DATA(opc);
2541 
2542 	/*
2543 	 * Never forget the root.
2544 	 */
2545 	if (pnd->pnd_ino == FUSE_ROOT_ID)
2546 		return 0;
2547 
2548 	pnd->pnd_flags |= PND_RECLAIMED;
2549 
2550 #ifdef PERFUSE_DEBUG
2551 	if (perfuse_diagflags & PDF_RECLAIM)
2552 		DPRINTF("%s (nodeid %"PRId64") reclaimed\n",
2553 			perfuse_node_path(opc), pnd->pnd_ino);
2554 #endif
2555 
2556 	pn_root = puffs_getroot(pu);
2557 	pn = (struct puffs_node *)opc;
2558 	while (pn != pn_root) {
2559 		struct puffs_node *parent_pn;
2560 
2561 		pnd = PERFUSE_NODE_DATA(pn);
2562 
2563 #ifdef PERFUSE_DEBUG
2564 	if (perfuse_diagflags & PDF_RECLAIM)
2565 		DPRINTF("%s (nodeid %"PRId64") is %sreclaimed, "
2566 			"has childcount %d %s%s%s%s, pending ops:%s%s%s\n",
2567 		        perfuse_node_path((puffs_cookie_t)pn), pnd->pnd_ino,
2568 		        pnd->pnd_flags & PND_RECLAIMED ? "" : "not ",
2569 		        pnd->pnd_childcount,
2570 			pnd->pnd_flags & PND_OPEN ? "open " : "not open",
2571 			pnd->pnd_flags & PND_RFH ? "r" : "",
2572 			pnd->pnd_flags & PND_WFH ? "w" : "",
2573 			pnd->pnd_flags & PND_BUSY ? "" : " none",
2574 			pnd->pnd_flags & PND_INREADDIR ? " readdir" : "",
2575 			pnd->pnd_flags & PND_INWRITE ? " write" : "",
2576 			pnd->pnd_flags & PND_INOPEN ? " open" : "");
2577 #endif
2578 
2579 		if (!(pnd->pnd_flags & PND_RECLAIMED) ||
2580 		    (pnd->pnd_childcount != 0))
2581 			return 0;
2582 
2583 #ifdef PERFUSE_DEBUG
2584 		if ((pnd->pnd_flags & PND_OPEN) ||
2585 		       !TAILQ_EMPTY(&pnd->pnd_pcq))
2586 			DERRX(EX_SOFTWARE, "%s: opc = %p: still open",
2587 			      __func__, (void *)opc);
2588 
2589 		if ((pnd->pnd_flags & PND_BUSY) ||
2590 		       !TAILQ_EMPTY(&pnd->pnd_pcq))
2591 			DERRX(EX_SOFTWARE, "%s: opc = %p: ongoing operations",
2592 			      __func__, (void *)opc);
2593 #endif
2594 
2595 		/*
2596 		 * Send the FORGET message
2597 		 *
2598 		 * ps_new_msg() is called with NULL creds, which will
2599 		 * be interpreted as FUSE superuser. This is obviously
2600 		 * fine since we operate with kernel creds here.
2601 		 */
2602 		pm = ps->ps_new_msg(pu, (puffs_cookie_t)pn, FUSE_FORGET,
2603 			      sizeof(*ffi), NULL);
2604 		ffi = GET_INPAYLOAD(ps, pm, fuse_forget_in);
2605 		ffi->nlookup = pnd->pnd_nlookup;
2606 
2607 		/*
2608 		 * No reply is expected, pm is freed in xchg_msg
2609 		 */
2610 		(void)xchg_msg(pu, (puffs_cookie_t)pn,
2611 			       pm, UNSPEC_REPLY_LEN, no_reply);
2612 
2613 		parent_pn = pnd->pnd_parent;
2614 
2615 		perfuse_destroy_pn(pn);
2616 
2617 		pn = parent_pn;
2618 	}
2619 
2620 	return 0;
2621 }
2622 
2623 int
2624 perfuse_node_inactive(pu, opc)
2625 	struct puffs_usermount *pu;
2626 	puffs_cookie_t opc;
2627 {
2628 	struct perfuse_state *ps;
2629 	struct perfuse_node_data *pnd;
2630 
2631 	ps = puffs_getspecific(pu);
2632 	pnd = PERFUSE_NODE_DATA(opc);
2633 
2634 	if (!(pnd->pnd_flags & (PND_OPEN|PND_REMOVED)))
2635 		return 0;
2636 
2637 	/*
2638 	 * Make sure all operation are finished
2639 	 * There can be an ongoing write. Other
2640 	 * operation wait for all data before
2641 	 * the close/inactive.
2642 	 */
2643 	while (pnd->pnd_flags & PND_INWRITE)
2644 		requeue_request(pu, opc, PCQ_AFTERWRITE);
2645 
2646 	/*
2647 	 * The inactive operation may be cancelled.
2648 	 * If no open is in progress, set PND_INOPEN
2649 	 * so that a new open will be queued.
2650 	 */
2651 	if (pnd->pnd_flags & PND_INOPEN)
2652 		return 0;
2653 
2654 	pnd->pnd_flags |= PND_INOPEN;
2655 
2656 	/*
2657 	 * Sync data
2658 	 */
2659 	if (pnd->pnd_flags & PND_DIRTY)
2660 		(void)perfuse_node_fsync(pu, opc, NULL, 0, 0, 0);
2661 
2662 	/*
2663 	 * Close handles
2664 	 */
2665 	if (pnd->pnd_flags & PND_WFH)
2666 		(void)perfuse_node_close_common(pu, opc, FWRITE);
2667 
2668 	if (pnd->pnd_flags & PND_RFH)
2669 		(void)perfuse_node_close_common(pu, opc, FREAD);
2670 
2671 	/*
2672 	 * This will cause a reclaim to be sent
2673 	 */
2674 	if (pnd->pnd_flags & PND_REMOVED)
2675 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N1);
2676 
2677 	/*
2678 	 * Schedule awaiting operations
2679 	 */
2680 	pnd->pnd_flags &= ~PND_INOPEN;
2681 	(void)dequeue_requests(ps, opc, PCQ_OPEN, DEQUEUE_ALL);
2682 
2683 	return 0;
2684 }
2685 
2686 
2687 /* ARGSUSED0 */
2688 int
2689 perfuse_node_print(pu, opc)
2690 	struct puffs_usermount *pu;
2691 	puffs_cookie_t opc;
2692 {
2693 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2694 	return 0;
2695 }
2696 
2697 /* ARGSUSED0 */
2698 int
2699 perfuse_node_pathconf(pu, opc, name, retval)
2700 	struct puffs_usermount *pu;
2701 	puffs_cookie_t opc;
2702 	int name;
2703 	int *retval;
2704 {
2705 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2706 	return 0;
2707 }
2708 
2709 /* id is unused */
2710 /* ARGSUSED2 */
2711 int
2712 perfuse_node_advlock(pu, opc, id, op, fl, flags)
2713 	struct puffs_usermount *pu;
2714 	puffs_cookie_t opc;
2715 	void *id;
2716 	int op;
2717 	struct flock *fl;
2718 	int flags;
2719 {
2720 	struct perfuse_state *ps;
2721 	int fop;
2722 	perfuse_msg_t *pm;
2723 	struct fuse_lk_in *fli;
2724 	struct fuse_out_header *foh;
2725 	struct fuse_lk_out *flo;
2726 	uint32_t owner;
2727 	size_t len;
2728 	int error;
2729 
2730 	ps = puffs_getspecific(pu);
2731 
2732 	if (op == F_GETLK)
2733 		fop = FUSE_GETLK;
2734 	else
2735 		fop = (flags & F_WAIT) ? FUSE_SETLKW : FUSE_SETLK;
2736 
2737 	/*
2738 	 * XXX ps_new_msg() is called with NULL creds, which will
2739 	 * be interpreted as FUSE superuser. We have no way to
2740 	 * know the requesting process' credential, but since advlock()
2741 	 * is supposed to operate on a file that has been open(),
2742 	 * permission should have already been checked at open() time.
2743 	 */
2744 	pm = ps->ps_new_msg(pu, opc, fop, sizeof(*fli), NULL);
2745 	fli = GET_INPAYLOAD(ps, pm, fuse_lk_in);
2746 	fli->fh = perfuse_get_fh(opc, FWRITE);
2747 	fli->owner = fl->l_pid;
2748 	fli->lk.start = fl->l_start;
2749 	fli->lk.end = fl->l_start + fl->l_len;
2750 	fli->lk.type = fl->l_type;
2751 	fli->lk.pid = fl->l_pid;
2752 	fli->lk_flags = (flags & F_FLOCK) ? FUSE_LK_FLOCK : 0;
2753 
2754 	owner = fl->l_pid;
2755 
2756 #ifdef PERFUSE_DEBUG
2757 	if (perfuse_diagflags & PDF_FH)
2758 		DPRINTF("%s: opc = %p, ino = %"PRId64", fh = 0x%"PRIx64"\n",
2759 			__func__, (void *)opc,
2760 			PERFUSE_NODE_DATA(opc)->pnd_ino, fli->fh);
2761 #endif
2762 
2763 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2764 		goto out;
2765 
2766 	foh = GET_OUTHDR(ps, pm);
2767 	len = foh->len - sizeof(*foh);
2768 
2769 	/*
2770 	 * Save or clear the lock
2771 	 */
2772 	switch (op) {
2773 	case F_GETLK:
2774 		if (len != sizeof(*flo))
2775 			DERRX(EX_SOFTWARE,
2776 			      "%s: Unexpected lock reply len %zd",
2777 			      __func__, len);
2778 
2779 		flo = GET_OUTPAYLOAD(ps, pm, fuse_lk_out);
2780 		fl->l_start = flo->lk.start;
2781 		fl->l_len = flo->lk.end - flo->lk.start;
2782 		fl->l_pid = flo->lk.pid;
2783 		fl->l_type = flo->lk.type;
2784 		fl->l_whence = SEEK_SET;	/* libfuse hardcodes it */
2785 
2786 		PERFUSE_NODE_DATA(opc)->pnd_lock_owner = flo->lk.pid;
2787 		break;
2788 	case F_UNLCK:
2789 		owner = 0;
2790 		/* FALLTHROUGH */
2791 	case F_SETLK:
2792 		/* FALLTHROUGH */
2793 	case F_SETLKW:
2794 		if (error != 0)
2795 			PERFUSE_NODE_DATA(opc)->pnd_lock_owner = owner;
2796 
2797 		if (len != 0)
2798 			DERRX(EX_SOFTWARE,
2799 			      "%s: Unexpected unlock reply len %zd",
2800 			      __func__, len);
2801 
2802 		break;
2803 	default:
2804 		DERRX(EX_SOFTWARE, "%s: Unexpected op %d", __func__, op);
2805 		break;
2806 	}
2807 
2808 out:
2809 	ps->ps_destroy_msg(pm);
2810 
2811 	return error;
2812 }
2813 
2814 int
2815 perfuse_node_read(pu, opc, buf, offset, resid, pcr, ioflag)
2816 	struct puffs_usermount *pu;
2817 	puffs_cookie_t opc;
2818 	uint8_t *buf;
2819 	off_t offset;
2820 	size_t *resid;
2821 	const struct puffs_cred *pcr;
2822 	int ioflag;
2823 {
2824 	struct perfuse_state *ps;
2825 	struct perfuse_node_data *pnd;
2826 	const struct vattr *vap;
2827 	perfuse_msg_t *pm;
2828 	struct fuse_read_in *fri;
2829 	struct fuse_out_header *foh;
2830 	size_t readen;
2831 	int error;
2832 
2833 	ps = puffs_getspecific(pu);
2834 	pnd = PERFUSE_NODE_DATA(opc);
2835 	vap = puffs_pn_getvap((struct puffs_node *)opc);
2836 	pm = NULL;
2837 
2838 	if ((u_quad_t)offset + *resid > vap->va_size)
2839 		DWARNX("%s %p read %lld@%zu beyond EOF %" PRIu64 "\n",
2840 		       __func__, (void *)opc, (long long)offset,
2841 		       *resid, vap->va_size);
2842 
2843 	do {
2844 		size_t max_read;
2845 
2846 		max_read = ps->ps_max_readahead - sizeof(*foh);
2847 		/*
2848 		 * flags may be set to FUSE_READ_LOCKOWNER
2849 		 * if lock_owner is provided.
2850 		 */
2851 		pm = ps->ps_new_msg(pu, opc, FUSE_READ, sizeof(*fri), pcr);
2852 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
2853 		fri->fh = perfuse_get_fh(opc, FREAD);
2854 		fri->offset = offset;
2855 		fri->size = (uint32_t)MIN(*resid, max_read);
2856 		fri->read_flags = 0; /* XXX Unused by libfuse? */
2857 		fri->lock_owner = pnd->pnd_lock_owner;
2858 		fri->flags = 0;
2859 		fri->flags |= (fri->lock_owner != 0) ? FUSE_READ_LOCKOWNER : 0;
2860 
2861 #ifdef PERFUSE_DEBUG
2862 	if (perfuse_diagflags & PDF_FH)
2863 		DPRINTF("%s: opc = %p, ino = %"PRId64", fh = 0x%"PRIx64"\n",
2864 			__func__, (void *)opc, pnd->pnd_ino, fri->fh);
2865 #endif
2866 		error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
2867 
2868 		if (error  != 0)
2869 			goto out;
2870 
2871 		foh = GET_OUTHDR(ps, pm);
2872 		readen = foh->len - sizeof(*foh);
2873 
2874 #ifdef PERFUSE_DEBUG
2875 		if (readen > *resid)
2876 			DERRX(EX_SOFTWARE, "%s: Unexpected big read %zd",
2877 			      __func__, readen);
2878 #endif
2879 
2880 		(void)memcpy(buf,  _GET_OUTPAYLOAD(ps, pm, char *), readen);
2881 
2882 		buf += readen;
2883 		offset += readen;
2884 		*resid -= readen;
2885 
2886 		ps->ps_destroy_msg(pm);
2887 		pm = NULL;
2888 	} while ((*resid != 0) && (readen != 0));
2889 
2890 	if (ioflag & (IO_SYNC|IO_DSYNC))
2891 		ps->ps_syncreads++;
2892 	else
2893 		ps->ps_asyncreads++;
2894 
2895 out:
2896 	if (pm != NULL)
2897 		ps->ps_destroy_msg(pm);
2898 
2899 	return error;
2900 }
2901 
2902 int
2903 perfuse_node_write(pu, opc, buf, offset, resid, pcr, ioflag)
2904 	struct puffs_usermount *pu;
2905 	puffs_cookie_t opc;
2906 	uint8_t *buf;
2907 	off_t offset;
2908 	size_t *resid;
2909 	const struct puffs_cred *pcr;
2910 	int ioflag;
2911 {
2912 	struct perfuse_state *ps;
2913 	struct perfuse_node_data *pnd;
2914 	struct vattr *vap;
2915 	perfuse_msg_t *pm;
2916 	struct fuse_write_in *fwi;
2917 	struct fuse_write_out *fwo;
2918 	size_t data_len;
2919 	size_t payload_len;
2920 	size_t written;
2921 	int inresize;
2922 	int error;
2923 
2924 	ps = puffs_getspecific(pu);
2925 	pnd = PERFUSE_NODE_DATA(opc);
2926 	vap = puffs_pn_getvap((struct puffs_node *)opc);
2927 	written = 0;
2928 	inresize = 0;
2929 	pm = NULL;
2930 
2931 	if (vap->va_type == VDIR)
2932 		return EBADF;
2933 
2934 	/*
2935 	 * We need to queue write requests in order to avoid
2936 	 * dequeueing PCQ_AFTERWRITE when there are pending writes.
2937 	 */
2938 	while (pnd->pnd_flags & PND_INWRITE)
2939 		requeue_request(pu, opc, PCQ_WRITE);
2940 	pnd->pnd_flags |= PND_INWRITE;
2941 
2942 	/*
2943 	 * Serialize size access, see comment in perfuse_node_setattr().
2944 	 */
2945 	if ((u_quad_t)offset + *resid > vap->va_size) {
2946 		while (pnd->pnd_flags & PND_INRESIZE)
2947 			requeue_request(pu, opc, PCQ_RESIZE);
2948 		pnd->pnd_flags |= PND_INRESIZE;
2949 		inresize = 1;
2950 	}
2951 
2952 	/*
2953 	 * append flag: re-read the file size so that
2954 	 * we get the latest value.
2955 	 */
2956 	if (ioflag & PUFFS_IO_APPEND) {
2957 		DWARNX("%s: PUFFS_IO_APPEND set, untested code", __func__);
2958 
2959 		if ((error = perfuse_node_getattr(pu, opc, vap, pcr)) != 0)
2960 			goto out;
2961 
2962 		offset = vap->va_size;
2963 	}
2964 
2965 	pm = NULL;
2966 
2967 #ifdef PERFUSE_DEBUG
2968 	if (perfuse_diagflags & PDF_RESIZE)
2969 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__,
2970 			(void *)opc, vap->va_size);
2971 #endif
2972 
2973 	do {
2974 		size_t max_write;
2975 		/*
2976 		 * There is a writepage flag when data
2977 		 * is aligned to page size. Use it for
2978 		 * everything but the data after the last
2979 		 * page boundary.
2980 		 */
2981 		max_write = ps->ps_max_write - sizeof(*fwi);
2982 
2983 		data_len = MIN(*resid, max_write);
2984 		if (data_len > (size_t)sysconf(_SC_PAGESIZE))
2985 			data_len = data_len & ~(sysconf(_SC_PAGESIZE) - 1);
2986 
2987 		payload_len = data_len + sizeof(*fwi);
2988 
2989 		/*
2990 		 * flags may be set to FUSE_WRITE_CACHE (XXX usage?)
2991 		 * or FUSE_WRITE_LOCKOWNER, if lock_owner is provided.
2992 		 * write_flags is set to 1 for writepage.
2993 		 */
2994 		pm = ps->ps_new_msg(pu, opc, FUSE_WRITE, payload_len, pcr);
2995 		fwi = GET_INPAYLOAD(ps, pm, fuse_write_in);
2996 		fwi->fh = perfuse_get_fh(opc, FWRITE);
2997 		fwi->offset = offset;
2998 		fwi->size = (uint32_t)data_len;
2999 		fwi->write_flags = (fwi->size % sysconf(_SC_PAGESIZE)) ? 0 : 1;
3000 		fwi->lock_owner = pnd->pnd_lock_owner;
3001 		fwi->flags = 0;
3002 		fwi->flags |= (fwi->lock_owner != 0) ? FUSE_WRITE_LOCKOWNER : 0;
3003 		fwi->flags |= (ioflag & IO_DIRECT) ? 0 : FUSE_WRITE_CACHE;
3004 		(void)memcpy((fwi + 1), buf, data_len);
3005 
3006 
3007 #ifdef PERFUSE_DEBUG
3008 		if (perfuse_diagflags & PDF_FH)
3009 			DPRINTF("%s: opc = %p, ino = %"PRId64", "
3010 				"fh = 0x%"PRIx64"\n", __func__,
3011 				(void *)opc, pnd->pnd_ino, fwi->fh);
3012 #endif
3013 		if ((error = xchg_msg(pu, opc, pm,
3014 				      sizeof(*fwo), wait_reply)) != 0)
3015 			goto out;
3016 
3017 		fwo = GET_OUTPAYLOAD(ps, pm, fuse_write_out);
3018 		written = fwo->size;
3019 #ifdef PERFUSE_DEBUG
3020 		if (written > *resid)
3021 			DERRX(EX_SOFTWARE, "%s: Unexpected big write %zd",
3022 			      __func__, written);
3023 #endif
3024 		*resid -= written;
3025 		offset += written;
3026 		buf += written;
3027 
3028 		ps->ps_destroy_msg(pm);
3029 		pm = NULL;
3030 	} while (*resid != 0);
3031 
3032 	/*
3033 	 * puffs_ops(3) says
3034 	 *  "everything must be written or an error will be generated"
3035 	 */
3036 	if (*resid != 0)
3037 		error = EFBIG;
3038 
3039 #ifdef PERFUSE_DEBUG
3040 	if (perfuse_diagflags & PDF_RESIZE) {
3041 		if (offset > (off_t)vap->va_size)
3042 			DPRINTF("<< %s %p %" PRIu64 " -> %lld\n", __func__,
3043 				(void *)opc, vap->va_size, (long long)offset);
3044 		else
3045 			DPRINTF("<< %s %p \n", __func__, (void *)opc);
3046 	}
3047 #endif
3048 
3049 	/*
3050 	 * Update file size if we wrote beyond the end
3051 	 */
3052 	if (offset > (off_t)vap->va_size)
3053 		vap->va_size = offset;
3054 
3055 	if (inresize) {
3056 #ifdef PERFUSE_DEBUG
3057 		if (!(pnd->pnd_flags & PND_INRESIZE))
3058 			DERRX(EX_SOFTWARE, "file write grow without resize");
3059 #endif
3060 		pnd->pnd_flags &= ~PND_INRESIZE;
3061 		(void)dequeue_requests(ps, opc, PCQ_RESIZE, DEQUEUE_ALL);
3062 	}
3063 
3064 
3065 	/*
3066 	 * Statistics
3067 	 */
3068 	if (ioflag & (IO_SYNC|IO_DSYNC))
3069 		ps->ps_syncwrites++;
3070 	else
3071 		ps->ps_asyncwrites++;
3072 
3073 	/*
3074 	 * Remember to sync the file
3075 	 */
3076 	pnd->pnd_flags |= PND_DIRTY;
3077 
3078 #ifdef PERFUSE_DEBUG
3079 	if (perfuse_diagflags & PDF_SYNC)
3080 		DPRINTF("%s: DIRTY opc = %p, file = \"%s\"\n",
3081 			__func__, (void*)opc, perfuse_node_path(opc));
3082 #endif
3083 out:
3084 	if (pm != NULL)
3085 		ps->ps_destroy_msg(pm);
3086 
3087 	/*
3088 	 * If there are no more queued write, we can resume
3089 	 * an operation awaiting write completion.
3090 	 */
3091 	pnd->pnd_flags &= ~PND_INWRITE;
3092 	if (dequeue_requests(ps, opc, PCQ_WRITE, 1) == 0)
3093 		(void)dequeue_requests(ps, opc, PCQ_AFTERWRITE, DEQUEUE_ALL);
3094 
3095 	return error;
3096 }
3097 
3098 /* ARGSUSED0 */
3099 void
3100 perfuse_cache_write(pu, opc, size, runs)
3101 	struct puffs_usermount *pu;
3102 	puffs_cookie_t opc;
3103 	size_t size;
3104 	struct puffs_cacherun *runs;
3105 {
3106 	return;
3107 }
3108 
3109 /* ARGSUSED4 */
3110 int
3111 perfuse_node_getextattr(pu, opc, attrns, attrname, attrsize, attr, resid, pcr)
3112 	struct puffs_usermount *pu;
3113 	puffs_cookie_t opc;
3114 	int attrns;
3115 	const char *attrname;
3116 	size_t *attrsize;
3117 	uint8_t *attr;
3118 	size_t *resid;
3119 	const struct puffs_cred *pcr;
3120 {
3121 	struct perfuse_state *ps;
3122 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3123 	perfuse_msg_t *pm;
3124 	struct fuse_getxattr_in *fgi;
3125 	struct fuse_getxattr_out *fgo;
3126 	struct fuse_out_header *foh;
3127 	size_t attrnamelen;
3128 	size_t len;
3129 	char *np;
3130 	int error;
3131 
3132 	ps = puffs_getspecific(pu);
3133 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3134 	attrnamelen = strlen(attrname) + 1;
3135 	len = sizeof(*fgi) + attrnamelen;
3136 
3137 	pm = ps->ps_new_msg(pu, opc, FUSE_GETXATTR, len, pcr);
3138 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3139 	fgi->size = (unsigned int)((resid != NULL) ? *resid : 0);
3140 	np = (char *)(void *)(fgi + 1);
3141 	(void)strlcpy(np, attrname, attrnamelen);
3142 
3143 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3144 		goto out;
3145 
3146 	/*
3147 	 * We just get fuse_getattr_out with list size if we requested
3148 	 * a null size.
3149 	 */
3150 	if (resid == NULL) {
3151 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3152 
3153 		if (attrsize != NULL)
3154 			*attrsize = fgo->size;
3155 
3156 		goto out;
3157 	}
3158 
3159 	/*
3160 	 * And with a non null requested size, we get the list just
3161 	 * after the header
3162 	 */
3163 	foh = GET_OUTHDR(ps, pm);
3164 	np = (char *)(void *)(foh + 1);
3165 
3166 	if (resid != NULL) {
3167 		len = MAX(foh->len - sizeof(*foh), *resid);
3168 		(void)memcpy(attr, np, len);
3169 		*resid -= len;
3170 	}
3171 
3172 out:
3173 	ps->ps_destroy_msg(pm);
3174 
3175 	return error;
3176 }
3177 
3178 int
3179 perfuse_node_setextattr(pu, opc, attrns, attrname, attr, resid, pcr)
3180 	struct puffs_usermount *pu;
3181 	puffs_cookie_t opc;
3182 	int attrns;
3183 	const char *attrname;
3184 	uint8_t *attr;
3185 	size_t *resid;
3186 	const struct puffs_cred *pcr;
3187 {
3188 	struct perfuse_state *ps;
3189 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3190 	perfuse_msg_t *pm;
3191 	struct fuse_setxattr_in *fsi;
3192 	size_t attrnamelen;
3193 	size_t len;
3194 	char *np;
3195 	int error;
3196 
3197 	ps = puffs_getspecific(pu);
3198 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3199 	attrnamelen = strlen(attrname) + 1;
3200 	len = sizeof(*fsi) + attrnamelen + *resid;
3201 
3202 	pm = ps->ps_new_msg(pu, opc, FUSE_SETXATTR, len, pcr);
3203 	fsi = GET_INPAYLOAD(ps, pm, fuse_setxattr_in);
3204 	fsi->size = (unsigned int)*resid;
3205 	fsi->flags = 0;
3206 	np = (char *)(void *)(fsi + 1);
3207 	(void)strlcpy(np, attrname, attrnamelen);
3208 	np += attrnamelen;
3209 	(void)memcpy(np, (char *)attr, *resid);
3210 
3211 	if ((error = xchg_msg(pu, opc, pm,
3212 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
3213 		goto out;
3214 
3215 	*resid = 0;
3216 out:
3217 	ps->ps_destroy_msg(pm);
3218 
3219 	return error;
3220 }
3221 
3222 /* ARGSUSED2 */
3223 int
3224 perfuse_node_listextattr(pu, opc, attrns, attrsize, attrs, resid, flag, pcr)
3225 	struct puffs_usermount *pu;
3226 	puffs_cookie_t opc;
3227 	int attrns;
3228 	size_t *attrsize;
3229 	uint8_t *attrs;
3230 	size_t *resid;
3231 	int flag;
3232 	const struct puffs_cred *pcr;
3233 {
3234 	struct perfuse_state *ps;
3235 	perfuse_msg_t *pm;
3236 	struct fuse_getxattr_in *fgi;
3237 	struct fuse_getxattr_out *fgo;
3238 	struct fuse_out_header *foh;
3239 	char *np;
3240 	size_t len, puffs_len;
3241 	int error;
3242 
3243 	ps = puffs_getspecific(pu);
3244 	len = sizeof(*fgi);
3245 
3246 	pm = ps->ps_new_msg(pu, opc, FUSE_LISTXATTR, len, pcr);
3247 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3248 	if (resid != NULL)
3249 		fgi->size = (unsigned int)*resid;
3250 	else
3251 		fgi->size = 0;
3252 
3253 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3254 		goto out;
3255 
3256 	/*
3257 	 * We just get fuse_getattr_out with list size if we requested
3258 	 * a null size.
3259 	 */
3260 	if (resid == NULL) {
3261 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3262 
3263 		if (attrsize != NULL)
3264 			*attrsize = fgo->size;
3265 
3266 		goto out;
3267 	}
3268 
3269 	/*
3270 	 * And with a non null requested size, we get the list just
3271 	 * after the header
3272 	 */
3273 	foh = GET_OUTHDR(ps, pm);
3274 	np = (char *)(void *)(foh + 1);
3275 	puffs_len = foh->len - sizeof(*foh);
3276 
3277 	if (attrs != NULL) {
3278 #ifdef PUFFS_EXTATTR_LIST_LENPREFIX
3279 		/*
3280 		 * Convert the FUSE reply to length prefixed strings
3281 		 * if this is what the kernel wants.
3282 		 */
3283 		if (flag & PUFFS_EXTATTR_LIST_LENPREFIX) {
3284 			size_t i, attrlen;
3285 
3286 			for (i = 0; i < puffs_len; i += attrlen + 1) {
3287 				attrlen = strlen(np + i);
3288 				(void)memmove(np + i + 1, np + i, attrlen);
3289 				*(np + i) = (uint8_t)attrlen;
3290 			}
3291 		}
3292 #endif /* PUFFS_EXTATTR_LIST_LENPREFIX */
3293 		(void)memcpy(attrs, np, puffs_len);
3294 		*resid -= puffs_len;
3295 	}
3296 
3297 	if (attrsize != NULL)
3298 		*attrsize = puffs_len;
3299 
3300 out:
3301 	ps->ps_destroy_msg(pm);
3302 
3303 	return error;
3304 }
3305 
3306 int
3307 perfuse_node_deleteextattr(pu, opc, attrns, attrname, pcr)
3308 	struct puffs_usermount *pu;
3309 	puffs_cookie_t opc;
3310 	int attrns;
3311 	const char *attrname;
3312 	const struct puffs_cred *pcr;
3313 {
3314 	struct perfuse_state *ps;
3315 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3316 	perfuse_msg_t *pm;
3317 	size_t attrnamelen;
3318 	char *np;
3319 	int error;
3320 
3321 	ps = puffs_getspecific(pu);
3322 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3323 	attrnamelen = strlen(attrname) + 1;
3324 
3325 	pm = ps->ps_new_msg(pu, opc, FUSE_REMOVEXATTR, attrnamelen, pcr);
3326 	np = _GET_INPAYLOAD(ps, pm, char *);
3327 	(void)strlcpy(np, attrname, attrnamelen);
3328 
3329 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
3330 
3331 	ps->ps_destroy_msg(pm);
3332 
3333 	return error;
3334 }
3335