xref: /netbsd-src/lib/libperfuse/ops.c (revision deb6f0161a9109e7de9b519dc8dfb9478668dcdd)
1 /*  $NetBSD: ops.c,v 1.85 2018/11/16 02:39:02 manu Exp $ */
2 
3 /*-
4  *  Copyright (c) 2010-2011 Emmanuel Dreyfus. All rights reserved.
5  *
6  *  Redistribution and use in source and binary forms, with or without
7  *  modification, are permitted provided that the following conditions
8  *  are met:
9  *  1. Redistributions of source code must retain the above copyright
10  *     notice, this list of conditions and the following disclaimer.
11  *  2. Redistributions in binary form must reproduce the above copyright
12  *     notice, this list of conditions and the following disclaimer in the
13  *     documentation and/or other materials provided with the distribution.
14  *
15  *  THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
16  *  ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
17  *  TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
18  *  PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
19  *  BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
20  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
21  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
22  *  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
23  *  CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
24  *  ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
25  *  POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 #include <stdio.h>
29 #include <unistd.h>
30 #include <stdlib.h>
31 #include <libgen.h>
32 #include <errno.h>
33 #include <err.h>
34 #include <sysexits.h>
35 #include <syslog.h>
36 #include <puffs.h>
37 #include <sys/socket.h>
38 #include <sys/socket.h>
39 #include <sys/extattr.h>
40 #include <sys/time.h>
41 #include <machine/vmparam.h>
42 
43 #include "perfuse_priv.h"
44 #include "fuse.h"
45 
46 extern int perfuse_diagflags;
47 
48 #if 0
49 static void print_node(const char *, puffs_cookie_t);
50 #endif
51 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
52 static void perfuse_newinfo_setttl(struct puffs_newinfo *,
53     struct puffs_node *, struct fuse_entry_out *, struct fuse_attr_out *);
54 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
55 static int xchg_msg(struct puffs_usermount *, puffs_cookie_t,
56     perfuse_msg_t *, size_t, enum perfuse_xchg_pb_reply);
57 static int mode_access(puffs_cookie_t, const struct puffs_cred *, mode_t);
58 static int sticky_access(puffs_cookie_t, struct puffs_node *,
59     const struct puffs_cred *);
60 static void fuse_attr_to_vap(struct perfuse_state *,
61     struct vattr *, struct fuse_attr *);
62 static int node_lookup_common(struct puffs_usermount *, puffs_cookie_t,
63     struct puffs_newinfo *, const char *, const struct puffs_cred *,
64     struct puffs_node **);
65 static int node_mk_common(struct puffs_usermount *, puffs_cookie_t,
66     struct puffs_newinfo *, const struct puffs_cn *pcn, perfuse_msg_t *);
67 static uint64_t readdir_last_cookie(struct fuse_dirent *, size_t);
68 static ssize_t fuse_to_dirent(struct puffs_usermount *, puffs_cookie_t,
69     struct fuse_dirent *, size_t);
70 static void readdir_buffered(puffs_cookie_t, struct dirent *, off_t *,
71     size_t *);
72 static void node_ref(puffs_cookie_t);
73 static void node_rele(puffs_cookie_t);
74 static void requeue_request(struct puffs_usermount *,
75     puffs_cookie_t opc, enum perfuse_qtype);
76 static int dequeue_requests(puffs_cookie_t opc, enum perfuse_qtype, int);
77 #define DEQUEUE_ALL 0
78 
79 /*
80  *  From <sys/vnode>, inside #ifdef _KERNEL section
81  */
82 #define IO_SYNC		(0x40|IO_DSYNC)
83 #define IO_DSYNC	0x00200
84 #define IO_DIRECT	0x02000
85 
86 /*
87  *  From <fcntl>, inside #ifdef _KERNEL section
88  */
89 #define F_WAIT		0x010
90 #define F_FLOCK		0x020
91 #define OFLAGS(fflags)  ((fflags) - 1)
92 
93 /*
94  * Borrowed from src/sys/kern/vfs_subr.c and src/sys/sys/vnode.h
95  */
96 const enum vtype iftovt_tab[16] = {
97 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
98         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
99 };
100 const int vttoif_tab[9] = {
101 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
102         S_IFSOCK, S_IFIFO, S_IFMT,
103 };
104 
105 #define IFTOVT(mode) (iftovt_tab[((mode) & S_IFMT) >> 12])
106 #define VTTOIF(indx) (vttoif_tab[(int)(indx)])
107 
108 #if 0
109 static void
110 print_node(const char *func, puffs_cookie_t opc)
111 {
112 	struct puffs_node *pn;
113 	struct perfuse_node_data *pnd;
114 	struct vattr *vap;
115 
116 	pn = (struct puffs_node *)opc;
117 	pnd = PERFUSE_NODE_DATA(opc);
118 	vap = &pn->pn_va;
119 
120 	printf("%s: \"%s\", opc = %p, nodeid = 0x%"PRIx64" ino = %"PRIu64"\n",
121 	       func, pnd->pnd_name, opc, pnd->pnd_nodeid, vap->va_fileid);
122 
123 	return;
124 }
125 #endif /* PERFUSE_DEBUG */
126 
127 int
128 perfuse_node_close_common(struct puffs_usermount *pu, puffs_cookie_t opc,
129 	int mode)
130 {
131 	struct perfuse_state *ps;
132 	perfuse_msg_t *pm;
133 	int op;
134 	uint64_t fh;
135 	struct fuse_release_in *fri;
136 	struct perfuse_node_data *pnd;
137 	struct puffs_node *pn;
138 	int error;
139 
140 	ps = puffs_getspecific(pu);
141 	pn = (struct puffs_node *)opc;
142 	pnd = PERFUSE_NODE_DATA(pn);
143 
144 	if (puffs_pn_getvap(pn)->va_type == VDIR) {
145 		op = FUSE_RELEASEDIR;
146 		mode = FREAD;
147 	} else {
148 		op = FUSE_RELEASE;
149 	}
150 
151 	/*
152 	 * Destroy the filehandle before sending the
153 	 * request to the FUSE filesystem, otherwise
154 	 * we may get a second close() while we wait
155 	 * for the reply, and we would end up closing
156 	 * the same fh twice instead of closng both.
157 	 */
158 	fh = perfuse_get_fh(opc, mode);
159 	perfuse_destroy_fh(pn, fh);
160 
161 	/*
162 	 * release_flags may be set to FUSE_RELEASE_FLUSH
163 	 * to flush locks. lock_owner must be set in that case
164 	 *
165 	 * ps_new_msg() is called with NULL creds, which will
166 	 * be interpreted as FUSE superuser. We come here from the
167 	 * inactive method, which provides no creds, but obviously
168 	 * runs with kernel privilege.
169 	 */
170 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*fri), NULL);
171 	fri = GET_INPAYLOAD(ps, pm, fuse_release_in);
172 	fri->fh = fh;
173 	fri->flags = 0;
174 	fri->release_flags = 0;
175 	fri->lock_owner = pnd->pnd_lock_owner;
176 	fri->flags = (fri->lock_owner != 0) ? FUSE_RELEASE_FLUSH : 0;
177 
178 #ifdef PERFUSE_DEBUG
179 	if (perfuse_diagflags & PDF_FH)
180 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
181 			 __func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
182 #endif
183 
184 	if ((error = xchg_msg(pu, opc, pm,
185 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
186 		DERRX(EX_SOFTWARE, "%s: freed fh = 0x%"PRIx64" but filesystem "
187 		      "returned error = %d", __func__, fh, error);
188 
189 	ps->ps_destroy_msg(pm);
190 
191 	return 0;
192 }
193 
194 static int
195 xchg_msg(struct puffs_usermount *pu, puffs_cookie_t opc, perfuse_msg_t *pm,
196 	size_t len, enum perfuse_xchg_pb_reply wait)
197 {
198 	struct perfuse_state *ps;
199 	struct perfuse_node_data *pnd;
200 	struct perfuse_trace *pt = NULL;
201 	int error;
202 
203 	ps = puffs_getspecific(pu);
204 	pnd = NULL;
205 	if ((struct puffs_node *)opc != NULL)
206 		pnd = PERFUSE_NODE_DATA(opc);
207 
208 #ifdef PERFUSE_DEBUG
209 	if ((perfuse_diagflags & PDF_FILENAME) && (opc != 0))
210 		DPRINTF("file = \"%s\", ino = %"PRIu64" flags = 0x%x\n",
211 			perfuse_node_path(ps, opc),
212 			((struct puffs_node *)opc)->pn_va.va_fileid,
213 			PERFUSE_NODE_DATA(opc)->pnd_flags);
214 #endif
215 	ps->ps_xchgcount++;
216 	if (pnd)
217 		pnd->pnd_inxchg++;
218 
219 	/*
220 	 * Record FUSE call start if requested
221 	 */
222 	if (perfuse_diagflags & PDF_TRACE)
223 		pt = perfuse_trace_begin(ps, opc, pm);
224 
225 	/*
226 	 * Do actual FUSE exchange
227 	 */
228 	if ((error = ps->ps_xchg_msg(pu, pm, len, wait)) != 0)
229 		ps->ps_destroy_msg(pm);
230 
231 	/*
232 	 * Record FUSE call end if requested
233 	 */
234 	if (pt != NULL)
235 		perfuse_trace_end(ps, pt, error);
236 
237 	ps->ps_xchgcount--;
238 	if (pnd) {
239 		pnd->pnd_inxchg--;
240 		(void)dequeue_requests(opc, PCQ_AFTERXCHG, DEQUEUE_ALL);
241 	}
242 
243 	return error;
244 }
245 
246 static int
247 mode_access(puffs_cookie_t opc, const struct puffs_cred *pcr, mode_t mode)
248 {
249 	struct puffs_node *pn;
250 	struct vattr *va;
251 
252 	/*
253 	 * pcr is NULL for self open through fsync or readdir.
254 	 * In both case, access control is useless, as it was
255 	 * done before, at open time.
256 	 */
257 	if (pcr == NULL)
258 		return 0;
259 
260 	pn = (struct puffs_node *)opc;
261 	va = puffs_pn_getvap(pn);
262 	return puffs_access(va->va_type, va->va_mode,
263 			    va->va_uid, va->va_gid,
264 			    mode, pcr);
265 }
266 
267 static int
268 sticky_access(puffs_cookie_t opc, struct puffs_node *targ,
269 	      const struct puffs_cred *pcr)
270 {
271 	uid_t uid;
272 	int sticky, owner, parent_owner;
273 
274 	/*
275 	 * This covers the case where the kernel requests a DELETE
276 	 * or RENAME on its own, and where puffs_cred_getuid would
277 	 * return -1. While such a situation should not happen,
278 	 * we allow it here.
279 	 *
280 	 * This also allows root to tamper with other users' files
281 	 * that have the sticky bit.
282 	 */
283 	if (puffs_cred_isjuggernaut(pcr))
284 		return 0;
285 
286 	if (puffs_cred_getuid(pcr, &uid) != 0)
287 		DERRX(EX_SOFTWARE, "puffs_cred_getuid fails in %s", __func__);
288 
289 	sticky = puffs_pn_getvap(opc)->va_mode & S_ISTXT;
290 	owner = puffs_pn_getvap(targ)->va_uid == uid;
291 	parent_owner = puffs_pn_getvap(opc)->va_uid == uid;
292 
293 	if (sticky && !owner && !parent_owner)
294 		return EPERM;
295 
296 	return 0;
297 }
298 
299 
300 static void
301 fuse_attr_to_vap(struct perfuse_state *ps, struct vattr *vap,
302 	struct fuse_attr *fa)
303 {
304 	vap->va_type = IFTOVT(fa->mode);
305 	vap->va_mode = fa->mode & ALLPERMS;
306 	vap->va_nlink = fa->nlink;
307 	vap->va_uid = fa->uid;
308 	vap->va_gid = fa->gid;
309 	vap->va_fsid = (long)ps->ps_fsid;
310 	vap->va_fileid = fa->ino;
311 	vap->va_size = fa->size;
312 	vap->va_blocksize = fa->blksize;
313 	vap->va_atime.tv_sec = (time_t)fa->atime;
314 	vap->va_atime.tv_nsec = (long) fa->atimensec;
315 	vap->va_mtime.tv_sec = (time_t)fa->mtime;
316 	vap->va_mtime.tv_nsec = (long)fa->mtimensec;
317 	vap->va_ctime.tv_sec = (time_t)fa->ctime;
318 	vap->va_ctime.tv_nsec = (long)fa->ctimensec;
319 	vap->va_birthtime.tv_sec = 0;
320 	vap->va_birthtime.tv_nsec = 0;
321 	vap->va_gen = 0;
322 	vap->va_flags = 0;
323 	vap->va_rdev = fa->rdev;
324 	vap->va_bytes = fa->blocks * S_BLKSIZE;
325 	vap->va_filerev = (u_quad_t)PUFFS_VNOVAL;
326 	vap->va_vaflags = 0;
327 
328 	if (vap->va_blocksize == 0)
329 		vap->va_blocksize = DEV_BSIZE;
330 
331 	if (vap->va_size == (size_t)PUFFS_VNOVAL) /* XXX */
332 		vap->va_size = 0;
333 
334 	return;
335 }
336 
337 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
338 static void
339 perfuse_newinfo_setttl(struct puffs_newinfo *pni,
340     struct puffs_node *pn, struct fuse_entry_out *feo,
341     struct fuse_attr_out *fao)
342 {
343 #ifdef PERFUSE_DEBUG
344 	if ((feo == NULL) && (fao == NULL))
345 		DERRX(EX_SOFTWARE, "%s: feo and fao NULL", __func__);
346 
347 	if ((feo != NULL) && (fao != NULL))
348 		DERRX(EX_SOFTWARE, "%s: feo and fao != NULL", __func__);
349 #endif /* PERFUSE_DEBUG */
350 
351 	if (fao != NULL) {
352 		struct timespec va_ttl;
353 
354 		va_ttl.tv_sec = fao->attr_valid;
355 		va_ttl.tv_nsec = fao->attr_valid_nsec;
356 
357 		puffs_newinfo_setvattl(pni, &va_ttl);
358 	}
359 
360 	if (feo != NULL) {
361 		struct timespec va_ttl;
362 		struct timespec cn_ttl;
363 		struct timespec now;
364 		struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(pn);
365 
366 		va_ttl.tv_sec = feo->attr_valid;
367 		va_ttl.tv_nsec = feo->attr_valid_nsec;
368 		cn_ttl.tv_sec = feo->entry_valid;
369 		cn_ttl.tv_nsec = feo->entry_valid_nsec;
370 
371 		puffs_newinfo_setvattl(pni, &va_ttl);
372 		puffs_newinfo_setcnttl(pni, &cn_ttl);
373 
374 		if (clock_gettime(CLOCK_REALTIME, &now) != 0)
375 			DERR(EX_OSERR, "clock_gettime failed");
376 
377                 timespecadd(&now, &cn_ttl, &pnd->pnd_cn_expire);
378 	}
379 
380 	return;
381 }
382 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
383 
384 static int
385 node_lookup_common(struct puffs_usermount *pu, puffs_cookie_t opc,
386 	struct puffs_newinfo *pni, const char *path,
387 	const struct puffs_cred *pcr, struct puffs_node **pnp)
388 {
389 	struct perfuse_state *ps;
390 	struct perfuse_node_data *oldpnd;
391 	perfuse_msg_t *pm;
392 	struct fuse_entry_out *feo;
393 	struct puffs_node *pn;
394 	size_t len;
395 	int error;
396 
397 	/*
398 	 * Prevent further lookups if the parent was removed
399 	 */
400 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
401 		return ESTALE;
402 
403 	if (pnp == NULL)
404 		DERRX(EX_SOFTWARE, "pnp must be != NULL");
405 
406 	ps = puffs_getspecific(pu);
407 
408 #ifdef PERFUSE_DEBUG
409 	if (perfuse_diagflags & PDF_FILENAME)
410 		DPRINTF("%s: opc = %p, file = \"%s\" looking up \"%s\"\n",
411 			__func__, (void *)opc,
412 			perfuse_node_path(ps, opc), path);
413 
414 	if (strcmp(path, ".") == 0)
415 		DERRX(EX_SOFTWARE, "unexpected dot-lookup");
416 
417 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_RECLAIMED)
418 		DERRX(EX_SOFTWARE,
419 		      "looking up reclaimed node opc = %p, name = \"%s\"",
420 		      opc, path);
421 
422 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_INVALID)
423 		DERRX(EX_SOFTWARE,
424 		      "looking up freed node opc = %p, name = \"%s\"",
425 		      opc, path);
426 #endif /* PERFUSE_DEBUG */
427 
428 	len = strlen(path) + 1;
429 	pm = ps->ps_new_msg(pu, opc, FUSE_LOOKUP, len, pcr);
430 	(void)strlcpy(_GET_INPAYLOAD(ps, pm, char *), path, len);
431 
432 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
433 		return error;
434 
435 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
436 
437 	/*
438 	 * Starting with ABI 7.4, inode number 0 means ENOENT,
439 	 * with entry_valid / entry_valid_nsec giving negative
440 	 * cache timeout (which we do not implement yet).
441 	 */
442 	if (feo->attr.ino == 0) {
443 		ps->ps_destroy_msg(pm);
444 		return ENOENT;
445 	}
446 
447 	/*
448 	 * Check for a known node, not reclaimed, with another name.
449 	 * It may have been moved, or we can lookup ../
450 	 */
451 	if (((oldpnd = perfuse_node_bynodeid(ps, feo->nodeid)) != NULL) &&
452 	    !(oldpnd->pnd_flags & PND_RECLAIMED)) {
453 		/*
454 		 * Save the new node name if not ..
455 		 */
456 		if (strncmp(path, "..", len) != 0)
457 			(void)strlcpy(oldpnd->pnd_name,
458 				      path, MAXPATHLEN);
459 		pn = oldpnd->pnd_pn;
460 
461 	} else {
462 		pn = perfuse_new_pn(pu, path, opc);
463 		PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
464 		perfuse_node_cache(ps, pn);
465 	}
466 
467 #ifdef PERFUSE_DEBUG
468 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_RECLAIMED)
469 		DERRX(EX_SOFTWARE,
470 		      "reclaimed in lookup opc = %p, name = \"%s\", ck = %p",
471 		      opc, path, pn);
472 
473 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_INVALID)
474 		DERRX(EX_SOFTWARE,
475 		      "freed in lookup opc = %p, name = \"%s\", ck = %p",
476 		      opc, path, pn);
477 #endif /* PERFUSE_DEBUG */
478 
479 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
480 	pn->pn_va.va_gen = (u_long)(feo->generation);
481 	PERFUSE_NODE_DATA(pn)->pnd_fuse_nlookup++;
482 
483 	*pnp = pn;
484 
485 #ifdef PERFUSE_DEBUG
486 	if (perfuse_diagflags & PDF_FILENAME)
487 		DPRINTF("%s: opc = %p, looked up opc = %p, "
488 			"nodeid = 0x%"PRIx64" file = \"%s\"\n", __func__,
489 			(void *)opc, pn, feo->nodeid, path);
490 #endif
491 
492 	if (pni != NULL) {
493 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
494 		puffs_newinfo_setva(pni, &pn->pn_va);
495 		perfuse_newinfo_setttl(pni, pn, feo, NULL);
496 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
497 		puffs_newinfo_setcookie(pni, pn);
498 		puffs_newinfo_setvtype(pni, pn->pn_va.va_type);
499 		puffs_newinfo_setsize(pni, (voff_t)pn->pn_va.va_size);
500 		puffs_newinfo_setrdev(pni, pn->pn_va.va_rdev);
501 	}
502 
503 	ps->ps_destroy_msg(pm);
504 
505 	return 0;
506 }
507 
508 
509 /*
510  * Common code for methods that create objects:
511  * perfuse_node_mkdir
512  * perfuse_node_mknod
513  * perfuse_node_symlink
514  */
515 static int
516 node_mk_common(struct puffs_usermount *pu, puffs_cookie_t opc,
517 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
518 	perfuse_msg_t *pm)
519 {
520 	struct perfuse_state *ps;
521 	struct puffs_node *pn;
522 	struct fuse_entry_out *feo;
523 	int error;
524 
525 	ps =  puffs_getspecific(pu);
526 
527 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
528 		return error;
529 
530 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
531 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
532 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
533 
534 	pn = perfuse_new_pn(pu, pcn->pcn_name, opc);
535 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
536 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
537 	perfuse_node_cache(ps, pn);
538 
539 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
540 	pn->pn_va.va_gen = (u_long)(feo->generation);
541 
542 	puffs_newinfo_setcookie(pni, pn);
543 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
544 	puffs_newinfo_setva(pni, &pn->pn_va);
545 	perfuse_newinfo_setttl(pni, pn, feo, NULL);
546 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
547 
548 
549 #ifdef PERFUSE_DEBUG
550 	if (perfuse_diagflags & PDF_FILENAME)
551 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
552 			"nodeid = 0x%"PRIx64"\n",
553 			__func__, (void *)pn, pcn->pcn_name,
554 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid);
555 #endif
556 	ps->ps_destroy_msg(pm);
557 
558 	/* Parents is now dirty */
559 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
560 
561 	return 0;
562 }
563 
564 static uint64_t
565 readdir_last_cookie(struct fuse_dirent *fd, size_t fd_len)
566 {
567 	size_t len;
568 	size_t seen = 0;
569 	char *ndp;
570 
571 	do {
572 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
573 		seen += len;
574 
575 		if (seen >= fd_len)
576 			break;
577 
578 		ndp = (char *)(void *)fd + (size_t)len;
579 		fd = (struct fuse_dirent *)(void *)ndp;
580 	} while (1 /* CONSTCOND */);
581 
582 	return fd->off;
583 }
584 
585 static ssize_t
586 fuse_to_dirent(struct puffs_usermount *pu, puffs_cookie_t opc,
587 	struct fuse_dirent *fd, size_t fd_len)
588 {
589 	struct dirent *dents;
590 	size_t dents_len;
591 	ssize_t written;
592 	uint64_t fd_offset;
593 	struct fuse_dirent *fd_base;
594 	size_t len;
595 
596 	fd_base = fd;
597 	fd_offset = 0;
598 	written = 0;
599 	dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
600 	dents_len = (size_t)PERFUSE_NODE_DATA(opc)->pnd_dirent_len;
601 
602 	do {
603 		char *ndp;
604 		size_t reclen;
605 		char name[MAXPATHLEN];
606 
607 		reclen = _DIRENT_RECLEN(dents, fd->namelen);
608 
609 		/*
610 		 * Check we do not overflow the output buffer
611 		 * struct fuse_dirent is bigger than struct dirent,
612 		 * so we should always use fd_len and never reallocate
613 		 * later.
614 		 * If we have to reallocate,try to double the buffer
615 		 * each time so that we do not have to do it too often.
616 		 */
617 		if (written + reclen > dents_len) {
618 			if (dents_len == 0)
619 				dents_len = fd_len;
620 			else
621 				dents_len =
622 				   MAX(2 * dents_len, written + reclen);
623 
624 			dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
625 			if ((dents = realloc(dents, dents_len)) == NULL)
626 				DERR(EX_OSERR, "%s: malloc failed", __func__);
627 
628 			PERFUSE_NODE_DATA(opc)->pnd_dirent = dents;
629 			PERFUSE_NODE_DATA(opc)->pnd_dirent_len = dents_len;
630 
631 			/*
632 			 * (void *) for delint
633 			 */
634 			ndp = (char *)(void *)dents + written;
635 			dents = (struct dirent *)(void *)ndp;
636 		}
637 
638 		strncpy(name, fd->name, fd->namelen);
639 		name[fd->namelen] = '\0';
640 
641 		/*
642 		 * Filesystem was mounted without -o use_ino
643 		 * Perform a lookup to find it.
644 		 */
645 		if (fd->ino == PERFUSE_UNKNOWN_INO) {
646 			struct puffs_node *pn;
647 			struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
648 
649 			if (strcmp(name, "..") == 0) {
650 				/*
651 				 * Avoid breaking out of fs
652 				 * by lookup to .. on root
653 				 */
654 				if (pnd->pnd_nodeid == FUSE_ROOT_ID)
655 					fd->ino = FUSE_ROOT_ID;
656 				else
657 					fd->ino = pnd->pnd_parent_nodeid;
658 			} else if (strcmp(name, ".") == 0 ) {
659 				fd->ino = pnd->pnd_nodeid;
660 			} else {
661 				int error;
662 
663 				error = node_lookup_common(pu, opc, NULL,
664 							   name, NULL, &pn);
665 				if (error != 0) {
666 					DWARNX("node_lookup_common %s "
667 					       "failed: %d", name, error);
668 				} else {
669 					fd->ino = pn->pn_va.va_fileid;
670 					(void)perfuse_node_reclaim2(pu, pn, 1);
671 				}
672 			}
673 		}
674 
675 		dents->d_fileno = fd->ino;
676 		dents->d_reclen = (unsigned short)reclen;
677 		dents->d_namlen = fd->namelen;
678 		dents->d_type = fd->type;
679 		strlcpy(dents->d_name, name, fd->namelen + 1);
680 
681 #ifdef PERFUSE_DEBUG
682 		if (perfuse_diagflags & PDF_READDIR)
683 			DPRINTF("%s: translated \"%s\" ino = %"PRIu64"\n",
684 				__func__, dents->d_name, dents->d_fileno);
685 #endif
686 
687 		dents = _DIRENT_NEXT(dents);
688 		written += reclen;
689 
690 		/*
691 		 * Move to the next record.
692 		 * fd->off is not the offset, it is an opaque cookie
693 		 * given by the filesystem to keep state across multiple
694 		 * readdir() operation.
695 		 * Use record alignement instead.
696 		 */
697 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
698 #ifdef PERFUSE_DEBUG
699 		if (perfuse_diagflags & PDF_READDIR)
700 			DPRINTF("%s: record at %"PRId64"/0x%"PRIx64" "
701 				"length = %zd/0x%zx. "
702 				"next record at %"PRId64"/0x%"PRIx64" "
703 				"max %zd/0x%zx\n",
704 				__func__, fd_offset, fd_offset, len, len,
705 				fd_offset + len, fd_offset + len,
706 				fd_len, fd_len);
707 #endif
708 		fd_offset += len;
709 
710 		/*
711 		 * Check if next record is still within the packet
712 		 * If it is not, we reached the end of the buffer.
713 		 */
714 		if (fd_offset >= fd_len)
715 			break;
716 
717 		/*
718 		 * (void *) for delint
719 		 */
720 		ndp = (char *)(void *)fd_base + (size_t)fd_offset;
721 		fd = (struct fuse_dirent *)(void *)ndp;
722 
723 	} while (1 /* CONSTCOND */);
724 
725 	/*
726 	 * Adjust the dirent output length
727 	 */
728 	if (written != -1)
729 		PERFUSE_NODE_DATA(opc)->pnd_dirent_len = written;
730 
731 	return written;
732 }
733 
734 static void
735 readdir_buffered(puffs_cookie_t opc, struct dirent *dent, off_t *readoff,
736 	size_t *reslen)
737 {
738 	struct dirent *fromdent;
739 	struct perfuse_node_data *pnd;
740 	char *ndp;
741 
742 	pnd = PERFUSE_NODE_DATA(opc);
743 
744 	while (*readoff < pnd->pnd_dirent_len) {
745 		/*
746 		 * (void *) for delint
747 		 */
748 		ndp = (char *)(void *)pnd->pnd_dirent + (size_t)*readoff;
749 		fromdent = (struct dirent *)(void *)ndp;
750 
751 		if (*reslen < _DIRENT_SIZE(fromdent))
752 			break;
753 
754 		memcpy(dent, fromdent, _DIRENT_SIZE(fromdent));
755 		*readoff += _DIRENT_SIZE(fromdent);
756 		*reslen -= _DIRENT_SIZE(fromdent);
757 
758 		dent = _DIRENT_NEXT(dent);
759 	}
760 
761 #ifdef PERFUSE_DEBUG
762 	if (perfuse_diagflags & PDF_READDIR)
763 		DPRINTF("%s: readoff = %"PRId64",  "
764 			"pnd->pnd_dirent_len = %"PRId64"\n",
765 			__func__, *readoff, pnd->pnd_dirent_len);
766 #endif
767 	if (*readoff >=  pnd->pnd_dirent_len) {
768 		free(pnd->pnd_dirent);
769 		pnd->pnd_dirent = NULL;
770 		pnd->pnd_dirent_len = 0;
771 	}
772 
773 	return;
774 }
775 
776 
777 static void
778 node_ref(puffs_cookie_t opc)
779 {
780 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
781 
782 #ifdef PERFUSE_DEBUG
783 	if (pnd->pnd_flags & PND_INVALID)
784 		DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
785 #endif /* PERFUSE_DEBUG */
786 
787 	pnd->pnd_ref++;
788 	return;
789 }
790 
791 static void
792 node_rele(puffs_cookie_t opc)
793 {
794 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
795 
796 #ifdef PERFUSE_DEBUG
797 	if (pnd->pnd_flags & PND_INVALID)
798 		DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
799 #endif /* PERFUSE_DEBUG */
800 
801 	pnd->pnd_ref--;
802 
803 	if (pnd->pnd_ref == 0)
804 		(void)dequeue_requests(opc, PCQ_REF, DEQUEUE_ALL);
805 
806 	return;
807 }
808 
809 static void
810 requeue_request(struct puffs_usermount *pu, puffs_cookie_t opc,
811 	enum perfuse_qtype type)
812 {
813 	struct perfuse_cc_queue pcq;
814 	struct perfuse_node_data *pnd;
815 
816 	pnd = PERFUSE_NODE_DATA(opc);
817 	pcq.pcq_type = type;
818 	pcq.pcq_cc = puffs_cc_getcc(pu);
819 	TAILQ_INSERT_TAIL(&pnd->pnd_pcq, &pcq, pcq_next);
820 
821 #ifdef PERFUSE_DEBUG
822 	if (perfuse_diagflags & PDF_REQUEUE)
823 		DPRINTF("%s: REQUEUE opc = %p, pcc = %p (%s)\n",
824 		        __func__, (void *)opc, pcq.pcq_cc,
825 			perfuse_qtypestr[type]);
826 #endif
827 
828 	puffs_cc_yield(pcq.pcq_cc);
829 	TAILQ_REMOVE(&pnd->pnd_pcq, &pcq, pcq_next);
830 
831 #ifdef PERFUSE_DEBUG
832 	if (perfuse_diagflags & PDF_REQUEUE)
833 		DPRINTF("%s: RESUME opc = %p, pcc = %p (%s)\n",
834 		        __func__, (void *)opc, pcq.pcq_cc,
835 			perfuse_qtypestr[type]);
836 #endif
837 
838 	return;
839 }
840 
841 static int
842 dequeue_requests(puffs_cookie_t opc, enum perfuse_qtype type, int max)
843 {
844 	struct perfuse_cc_queue *pcq;
845 	struct perfuse_node_data *pnd;
846 	int dequeued;
847 
848 	pnd = PERFUSE_NODE_DATA(opc);
849 	dequeued = 0;
850 	TAILQ_FOREACH(pcq, &pnd->pnd_pcq, pcq_next) {
851 		if (pcq->pcq_type != type)
852 			continue;
853 
854 #ifdef PERFUSE_DEBUG
855 		if (perfuse_diagflags & PDF_REQUEUE)
856 			DPRINTF("%s: SCHEDULE opc = %p, pcc = %p (%s)\n",
857 				__func__, (void *)opc, pcq->pcq_cc,
858 				 perfuse_qtypestr[type]);
859 #endif
860 		puffs_cc_schedule(pcq->pcq_cc);
861 
862 		if (++dequeued == max)
863 			break;
864 	}
865 
866 #ifdef PERFUSE_DEBUG
867 	if (perfuse_diagflags & PDF_REQUEUE)
868 		DPRINTF("%s: DONE  opc = %p\n", __func__, (void *)opc);
869 #endif
870 
871 	return dequeued;
872 }
873 
874 void
875 perfuse_fs_init(struct puffs_usermount *pu)
876 {
877 	struct perfuse_state *ps;
878 	perfuse_msg_t *pm;
879 	struct fuse_init_in *fii;
880 	struct fuse_init_out *fio;
881 	int error;
882 
883 	ps = puffs_getspecific(pu);
884 
885         if (puffs_mount(pu, ps->ps_target, ps->ps_mountflags, ps->ps_root) != 0)
886                 DERR(EX_OSERR, "%s: puffs_mount failed", __func__);
887 
888 	/*
889 	 * Linux 2.6.34.1 sends theses flags:
890 	 * FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC
891 	 * FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK
892 	 *
893 	 * Linux also sets max_readahead at 32 pages (128 kB)
894 	 *
895 	 * ps_new_msg() is called with NULL creds, which will
896 	 * be interpreted as FUSE superuser.
897 	 */
898 	pm = ps->ps_new_msg(pu, 0, FUSE_INIT, sizeof(*fii), NULL);
899 	fii = GET_INPAYLOAD(ps, pm, fuse_init_in);
900 	fii->major = FUSE_KERNEL_VERSION;
901 	fii->minor = FUSE_KERNEL_MINOR_VERSION;
902 	fii->max_readahead = (unsigned int)(32 * sysconf(_SC_PAGESIZE));
903 	fii->flags = (FUSE_ASYNC_READ|FUSE_POSIX_LOCKS|FUSE_ATOMIC_O_TRUNC);
904 
905 	if ((error = xchg_msg(pu, 0, pm, sizeof(*fio), wait_reply)) != 0)
906 		DERRX(EX_SOFTWARE, "init message exchange failed (%d)", error);
907 
908 	fio = GET_OUTPAYLOAD(ps, pm, fuse_init_out);
909 	ps->ps_max_readahead = fio->max_readahead;
910 	ps->ps_max_write = fio->max_write;
911 
912 	ps->ps_destroy_msg(pm);
913 
914 	return;
915 }
916 
917 int
918 perfuse_fs_unmount(struct puffs_usermount *pu, int flags)
919 {
920 	perfuse_msg_t *pm;
921 	struct perfuse_state *ps;
922 	puffs_cookie_t opc;
923 	int error;
924 
925 	ps = puffs_getspecific(pu);
926 	opc = (puffs_cookie_t)puffs_getroot(pu);
927 
928 	/*
929 	 * ps_new_msg() is called with NULL creds, which will
930 	 * be interpreted as FUSE superuser.
931 	 */
932 	pm = ps->ps_new_msg(pu, opc, FUSE_DESTROY, 0, NULL);
933 
934 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0){
935 		DWARN("unmount %s", ps->ps_target);
936 		if (!(flags & MNT_FORCE))
937 			return error;
938 		else
939 			error = 0;
940 	} else {
941 		ps->ps_destroy_msg(pm);
942 	}
943 
944 	ps->ps_umount(pu);
945 
946 	if (perfuse_diagflags & PDF_MISC)
947 		DPRINTF("%s unmounted, exit\n", ps->ps_target);
948 
949 	return 0;
950 }
951 
952 int
953 perfuse_fs_statvfs(struct puffs_usermount *pu, struct statvfs *svfsb)
954 {
955 	struct perfuse_state *ps;
956 	perfuse_msg_t *pm;
957 	puffs_cookie_t opc;
958 	struct fuse_statfs_out *fso;
959 	int error;
960 
961 	ps = puffs_getspecific(pu);
962 	opc = (puffs_cookie_t)puffs_getroot(pu);
963 
964 	/*
965 	 * ps_new_msg() is called with NULL creds, which will
966 	 * be interpreted as FUSE superuser.
967 	 */
968 	pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
969 
970 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply)) != 0)
971 		return error;
972 
973 	fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
974 	svfsb->f_flag = ps->ps_mountflags;
975 	svfsb->f_bsize = fso->st.bsize;
976 	svfsb->f_frsize = fso->st.frsize;
977 	svfsb->f_iosize = ((struct puffs_node *)opc)->pn_va.va_blocksize;
978 	svfsb->f_blocks = fso->st.blocks;
979 	svfsb->f_bfree = fso->st.bfree;
980 	svfsb->f_bavail = fso->st.bavail;
981 	svfsb->f_bresvd = fso->st.bfree - fso->st.bavail;
982 	svfsb->f_files = fso->st.files;
983 	svfsb->f_ffree = fso->st.ffree;
984 	svfsb->f_favail = fso->st.ffree;/* files not reserved for root */
985 	svfsb->f_fresvd = 0;		/* files reserved for root */
986 
987 	svfsb->f_syncreads = ps->ps_syncreads;
988 	svfsb->f_syncwrites = ps->ps_syncwrites;
989 
990 	svfsb->f_asyncreads = ps->ps_asyncreads;
991 	svfsb->f_asyncwrites = ps->ps_asyncwrites;
992 
993 	(void)memcpy(&svfsb->f_fsidx, &ps->ps_fsid, sizeof(ps->ps_fsid));
994 	svfsb->f_fsid = (unsigned long)ps->ps_fsid;
995 	svfsb->f_namemax = MAXPATHLEN;	/* XXX */
996 	svfsb->f_owner = ps->ps_owner_uid;
997 
998 	(void)strlcpy(svfsb->f_mntonname, ps->ps_target, _VFS_NAMELEN);
999 
1000 	if (ps->ps_filesystemtype != NULL)
1001 		(void)strlcpy(svfsb->f_fstypename,
1002 			      ps->ps_filesystemtype, _VFS_NAMELEN);
1003 	else
1004 		(void)strlcpy(svfsb->f_fstypename, "fuse", _VFS_NAMELEN);
1005 
1006 	if (ps->ps_source != NULL)
1007 		strlcpy(svfsb->f_mntfromname, ps->ps_source, _VFS_NAMELEN);
1008 	else
1009 		strlcpy(svfsb->f_mntfromname, _PATH_FUSE, _VFS_NAMELEN);
1010 
1011 	ps->ps_destroy_msg(pm);
1012 
1013 	return 0;
1014 }
1015 
1016 int
1017 perfuse_fs_sync(struct puffs_usermount *pu, int waitfor,
1018 	const struct puffs_cred *pcr)
1019 {
1020 	/*
1021 	 * FUSE does not seem to have a FS sync callback.
1022 	 * Maybe do not even register this callback
1023 	 */
1024 	return puffs_fsnop_sync(pu, waitfor, pcr);
1025 }
1026 
1027 /* ARGSUSED0 */
1028 int
1029 perfuse_fs_fhtonode(struct puffs_usermount *pu, void *fid, size_t fidsize,
1030 	struct puffs_newinfo *pni)
1031 {
1032 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1033 	return 0;
1034 }
1035 
1036 /* ARGSUSED0 */
1037 int
1038 perfuse_fs_nodetofh(struct puffs_usermount *pu, puffs_cookie_t cookie,
1039 	void *fid, size_t *fidsize)
1040 {
1041 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1042 	return 0;
1043 }
1044 
1045 #if 0
1046 /* ARGSUSED0 */
1047 void
1048 perfuse_fs_extattrctl(struct puffs_usermount *pu, int cmd,
1049 	puffs_cookie_t *cookie, int flags, int namespace, const char *attrname)
1050 {
1051 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1052 	return 0;
1053 }
1054 #endif /* 0 */
1055 
1056 /* ARGSUSED0 */
1057 void
1058 perfuse_fs_suspend(struct puffs_usermount *pu, int status)
1059 {
1060 	return;
1061 }
1062 
1063 
1064 int
1065 perfuse_node_lookup(struct puffs_usermount *pu, puffs_cookie_t opc,
1066 	struct puffs_newinfo *pni, const struct puffs_cn *pcn)
1067 {
1068 	struct perfuse_state *ps;
1069 	struct puffs_node *pn;
1070 	mode_t mode;
1071 	int error;
1072 
1073 	ps = puffs_getspecific(pu);
1074 	node_ref(opc);
1075 
1076 	/*
1077 	 * Check permissions
1078 	 */
1079 	switch(pcn->pcn_nameiop) {
1080 	case NAMEI_DELETE: /* FALLTHROUGH */
1081 	case NAMEI_RENAME: /* FALLTHROUGH */
1082 	case NAMEI_CREATE:
1083 		if (pcn->pcn_flags & NAMEI_ISLASTCN)
1084 			mode = PUFFS_VEXEC|PUFFS_VWRITE;
1085 		else
1086 			mode = PUFFS_VEXEC;
1087 		break;
1088 	case NAMEI_LOOKUP: /* FALLTHROUGH */
1089 	default:
1090 		mode = PUFFS_VEXEC;
1091 		break;
1092 	}
1093 
1094 	if ((error = mode_access(opc, pcn->pcn_cred, mode)) != 0)
1095 		goto out;
1096 
1097 	error = node_lookup_common(pu, (puffs_cookie_t)opc, pni,
1098 				   pcn->pcn_name, pcn->pcn_cred, &pn);
1099 
1100 	if (error != 0)
1101 		goto out;
1102 
1103 	/*
1104 	 * Kernel would kill us if the filesystem returned the parent
1105 	 * itself. If we want to live, hide that!
1106 	 */
1107 	if ((opc == (puffs_cookie_t)pn) && (strcmp(pcn->pcn_name, ".") != 0)) {
1108 		DERRX(EX_SOFTWARE, "lookup \"%s\" in \"%s\" returned parent",
1109 		      pcn->pcn_name, perfuse_node_path(ps, opc));
1110 		/* NOTREACHED */
1111 		error = ESTALE;
1112 		goto out;
1113 	}
1114 
1115 	/*
1116 	 * Removed node
1117 	 */
1118 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_REMOVED) {
1119 		error = ENOENT;
1120 		goto out;
1121 	}
1122 
1123 	/*
1124 	 * Check for sticky bit. Unfortunately there is no way to
1125 	 * do this before creating the puffs_node, since we require
1126 	 * this operation to get the node owner.
1127 	 */
1128 	switch (pcn->pcn_nameiop) {
1129 	case NAMEI_DELETE: /* FALLTHROUGH */
1130 	case NAMEI_RENAME:
1131 		error = sticky_access(opc, pn, pcn->pcn_cred);
1132 		if (error != 0) {
1133 			(void)perfuse_node_reclaim2(pu, pn, 1);
1134 			goto out;
1135 		}
1136 		break;
1137 	default:
1138 		break;
1139 	}
1140 
1141 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
1142 
1143 	error = 0;
1144 
1145 out:
1146 	node_rele(opc);
1147 	return error;
1148 }
1149 
1150 int
1151 perfuse_node_create(struct puffs_usermount *pu, puffs_cookie_t opc,
1152 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
1153 	const struct vattr *vap)
1154 {
1155 	perfuse_msg_t *pm;
1156 	struct perfuse_state *ps;
1157 	struct fuse_create_in *fci;
1158 	struct fuse_entry_out *feo;
1159 	struct fuse_open_out *foo;
1160 	struct puffs_node *pn;
1161 	const char *name;
1162 	size_t namelen;
1163 	size_t len;
1164 	int error;
1165 
1166 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1167 		return ENOENT;
1168 
1169 	node_ref(opc);
1170 
1171 	/*
1172 	 * If create is unimplemented: Check that it does not
1173 	 * already exists, and if not, do mknod and open
1174 	 */
1175 	ps = puffs_getspecific(pu);
1176 	if (ps->ps_flags & PS_NO_CREAT) {
1177 		error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
1178 					   pcn->pcn_cred, &pn);
1179 		if (error == 0)	{
1180 			(void)perfuse_node_reclaim2(pu, pn, 1);
1181 			error = EEXIST;
1182 			goto out;
1183 		}
1184 
1185 		error = perfuse_node_mknod(pu, opc, pni, pcn, vap);
1186 		if (error != 0)
1187 			goto out;
1188 
1189 		error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
1190 					   pcn->pcn_cred, &pn);
1191 		if (error != 0)
1192 			goto out;
1193 
1194 		/*
1195 		 * FUSE does the open at create time, while
1196 		 * NetBSD will open in a subsequent operation.
1197 		 * We need to open now, in order to retain FUSE
1198 		 * semantics. The calling process will not get
1199 		 * a file descriptor before the kernel sends
1200 		 * the open operation.
1201 		 */
1202 		error = perfuse_node_open(pu, (puffs_cookie_t)pn,
1203 					  FWRITE, pcn->pcn_cred);
1204 		goto out;
1205 	}
1206 
1207 	name = pcn->pcn_name;
1208 	namelen = pcn->pcn_namelen + 1;
1209 	len = sizeof(*fci) + namelen;
1210 
1211 	/*
1212 	 * flags should use O_WRONLY instead of O_RDWR, but it
1213 	 * breaks when the caller tries to read from file.
1214 	 *
1215 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1216 	 */
1217 	pm = ps->ps_new_msg(pu, opc, FUSE_CREATE, len, pcn->pcn_cred);
1218 	fci = GET_INPAYLOAD(ps, pm, fuse_create_in);
1219 	fci->flags = O_CREAT | O_TRUNC | O_RDWR;
1220 	fci->mode = vap->va_mode | VTTOIF(vap->va_type);
1221 	fci->umask = 0; 	/* Seems unused by libfuse */
1222 	(void)strlcpy((char*)(void *)(fci + 1), name, namelen);
1223 
1224 	len = sizeof(*feo) + sizeof(*foo);
1225 	if ((error = xchg_msg(pu, opc, pm, len, wait_reply)) != 0) {
1226 		/*
1227 		 * create is unimplmented, remember it for later,
1228 		 * and start over using mknod and open instead.
1229 		 */
1230 		if (error == ENOSYS) {
1231 			ps->ps_flags |= PS_NO_CREAT;
1232 			error = perfuse_node_create(pu, opc, pni, pcn, vap);
1233 		}
1234 
1235 		goto out;
1236 	}
1237 
1238 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
1239 	foo = (struct fuse_open_out *)(void *)(feo + 1);
1240 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
1241 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
1242 
1243 	/*
1244 	 * Save the file handle and inode in node private data
1245 	 * so that we can reuse it later
1246 	 */
1247 	pn = perfuse_new_pn(pu, name, opc);
1248 	perfuse_new_fh((puffs_cookie_t)pn, foo->fh, FWRITE);
1249 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
1250 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
1251 	perfuse_node_cache(ps, pn);
1252 
1253 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
1254 	pn->pn_va.va_gen = (u_long)(feo->generation);
1255 
1256 	puffs_newinfo_setcookie(pni, pn);
1257 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
1258 	puffs_newinfo_setva(pni, &pn->pn_va);
1259 	perfuse_newinfo_setttl(pni, pn, feo, NULL);
1260 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
1261 
1262 #ifdef PERFUSE_DEBUG
1263 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1264 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
1265 			"nodeid = 0x%"PRIx64", wfh = 0x%"PRIx64"\n",
1266 			__func__, (void *)pn, pcn->pcn_name,
1267 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid,
1268 			foo->fh);
1269 #endif
1270 
1271 	ps->ps_destroy_msg(pm);
1272 	error = 0;
1273 
1274 out:
1275 	node_rele(opc);
1276 	return error;
1277 }
1278 
1279 
1280 int
1281 perfuse_node_mknod(struct puffs_usermount *pu, puffs_cookie_t opc,
1282 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
1283 	const struct vattr *vap)
1284 {
1285 	struct perfuse_state *ps;
1286 	perfuse_msg_t *pm;
1287 	struct fuse_mknod_in *fmi;
1288 	const char* path;
1289 	size_t len;
1290 	int error;
1291 
1292 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1293 		return ENOENT;
1294 
1295 	node_ref(opc);
1296 
1297 	/*
1298 	 * Only superuser can mknod objects other than
1299 	 * directories, files, socks, fifo and links.
1300 	 *
1301 	 * Create an object require -WX permission in the parent directory
1302 	 */
1303 	switch (vap->va_type) {
1304 	case VDIR:	/* FALLTHROUGH */
1305 	case VREG:	/* FALLTHROUGH */
1306 	case VFIFO:	/* FALLTHROUGH */
1307 	case VSOCK:
1308 		break;
1309 	default:	/* VNON, VBLK, VCHR, VBAD */
1310 		if (!puffs_cred_isjuggernaut(pcn->pcn_cred)) {
1311 			error = EPERM;
1312 			goto out;
1313 		}
1314 		break;
1315 	}
1316 
1317 
1318 	ps = puffs_getspecific(pu);
1319 	path = pcn->pcn_name;
1320 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
1321 
1322 	/*
1323 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1324 	 */
1325 	pm = ps->ps_new_msg(pu, opc, FUSE_MKNOD, len, pcn->pcn_cred);
1326 	fmi = GET_INPAYLOAD(ps, pm, fuse_mknod_in);
1327 	fmi->mode = vap->va_mode | VTTOIF(vap->va_type);
1328 	fmi->rdev = (uint32_t)vap->va_rdev;
1329 	fmi->umask = 0; 	/* Seems unused bu libfuse */
1330 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
1331 
1332 	error = node_mk_common(pu, opc, pni, pcn, pm);
1333 
1334 out:
1335 	node_rele(opc);
1336 	return error;
1337 }
1338 
1339 
1340 int
1341 perfuse_node_open(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
1342 	const struct puffs_cred *pcr)
1343 {
1344 	return perfuse_node_open2(pu, opc, mode, pcr, NULL);
1345 }
1346 
1347 int
1348 perfuse_node_open2(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
1349 	const struct puffs_cred *pcr, int *oflags)
1350 {
1351 	struct perfuse_state *ps;
1352 	struct perfuse_node_data *pnd;
1353 	perfuse_msg_t *pm;
1354 	mode_t fmode;
1355 	int op;
1356 	struct fuse_open_in *foi;
1357 	struct fuse_open_out *foo;
1358 	struct puffs_node *pn;
1359 	int error;
1360 
1361 	ps = puffs_getspecific(pu);
1362 	pn = (struct puffs_node *)opc;
1363 	pnd = PERFUSE_NODE_DATA(opc);
1364 	error = 0;
1365 
1366 	if (pnd->pnd_flags & PND_REMOVED)
1367 		return ENOENT;
1368 
1369 	node_ref(opc);
1370 
1371 	if (puffs_pn_getvap(pn)->va_type == VDIR)
1372 		op = FUSE_OPENDIR;
1373 	else
1374 		op = FUSE_OPEN;
1375 
1376 	/*
1377 	 * libfuse docs says
1378 	 * - O_CREAT and O_EXCL should never be set.
1379 	 * - O_TRUNC may be used if mount option atomic_o_trunc is used XXX
1380 	 *
1381 	 * O_APPEND makes no sense since FUSE always sends
1382 	 * the file offset for write operations. If the
1383 	 * filesystem uses pwrite(), O_APPEND would cause
1384 	 * the offset to be ignored and cause file corruption.
1385 	 */
1386 	mode &= ~(O_CREAT|O_EXCL|O_APPEND);
1387 
1388 	/*
1389 	 * Do not open twice, and do not reopen for reading
1390 	 * if we already have write handle.
1391 	 */
1392 	switch (mode & (FREAD|FWRITE)) {
1393 	case FREAD:
1394 		if (pnd->pnd_flags & (PND_RFH|PND_WFH))
1395 			goto out;
1396 		break;
1397 	case FWRITE:
1398 		if (pnd->pnd_flags & PND_WFH)
1399 			goto out;
1400 		break;
1401 	case FREAD|FWRITE:
1402 		if (pnd->pnd_flags & PND_WFH)
1403 			goto out;
1404 
1405 		/*
1406 		 * Corner case: if already open for reading (PND_RFH)
1407 		 * and re-opening FREAD|FWRITE, we need to reopen,
1408 		 * but only for writing. Note the change on mode
1409 		 * will only affect perfuse_new_fh()
1410 		 */
1411 		if (pnd->pnd_flags & PND_RFH)
1412 			mode &= ~FREAD;
1413 		break;
1414 	default:
1415 		DWARNX("open without either FREAD nor FWRITE");
1416 		error = EPERM;
1417 		goto out;
1418 	}
1419 
1420 	/*
1421 	 * Queue open on a node so that we do not open
1422 	 * twice. This would be better with read and
1423 	 * write distinguished.
1424 	 */
1425 	while (pnd->pnd_flags & PND_INOPEN)
1426 		requeue_request(pu, opc, PCQ_OPEN);
1427 	pnd->pnd_flags |= PND_INOPEN;
1428 
1429 	/*
1430 	 * Convert PUFFS mode to FUSE mode: convert FREAD/FWRITE
1431 	 * to O_RDONLY/O_WRONLY while perserving the other options.
1432 	 */
1433 	fmode = mode & ~(FREAD|FWRITE);
1434 	fmode |= (mode & FWRITE) ? O_RDWR : O_RDONLY;
1435 
1436 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*foi), pcr);
1437 	foi = GET_INPAYLOAD(ps, pm, fuse_open_in);
1438 	foi->flags = fmode;
1439 	foi->unused = 0;
1440 
1441 	if ((error = xchg_msg(pu, opc, pm, sizeof(*foo), wait_reply)) != 0)
1442 		goto out;
1443 
1444 	foo = GET_OUTPAYLOAD(ps, pm, fuse_open_out);
1445 
1446 	/*
1447 	 * Save the file handle in node private data
1448 	 * so that we can reuse it later
1449 	 */
1450 	perfuse_new_fh(opc, foo->fh, mode);
1451 
1452 	/*
1453 	 * Set direct I/O if the filesystems forces it
1454 	 */
1455 	if ((foo->open_flags & FUSE_FOPEN_DIRECT_IO) && (oflags != NULL))
1456 		*oflags |= PUFFS_OPEN_IO_DIRECT;
1457 
1458 #ifdef PERFUSE_DEBUG
1459 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1460 		DPRINTF("%s: opc = %p, file = \"%s\", "
1461 			"nodeid = 0x%"PRIx64", %s%sfh = 0x%"PRIx64"\n",
1462 			__func__, (void *)opc, perfuse_node_path(ps, opc),
1463 			pnd->pnd_nodeid, mode & FREAD ? "r" : "",
1464 			mode & FWRITE ? "w" : "", foo->fh);
1465 #endif
1466 
1467 	ps->ps_destroy_msg(pm);
1468 out:
1469 
1470 	pnd->pnd_flags &= ~PND_INOPEN;
1471 	(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
1472 
1473 	node_rele(opc);
1474 	return error;
1475 }
1476 
1477 /* ARGSUSED0 */
1478 int
1479 perfuse_node_close(struct puffs_usermount *pu, puffs_cookie_t opc, int flags,
1480 	const struct puffs_cred *pcr)
1481 {
1482 	struct perfuse_node_data *pnd;
1483 
1484 	pnd = PERFUSE_NODE_DATA(opc);
1485 
1486 	if (!(pnd->pnd_flags & PND_OPEN))
1487 		return EBADF;
1488 
1489 	/*
1490 	 * Actual close is postponed at inactive time.
1491 	 */
1492 	return 0;
1493 }
1494 
1495 int
1496 perfuse_node_access(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
1497 	const struct puffs_cred *pcr)
1498 {
1499 	perfuse_msg_t *pm;
1500 	struct perfuse_state *ps;
1501 	struct fuse_access_in *fai;
1502 	int error;
1503 
1504 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1505 		return ENOENT;
1506 
1507 	node_ref(opc);
1508 
1509 	/*
1510 	 * If we previously detected the filesystem does not
1511 	 * implement access(), short-circuit the call and skip
1512 	 * to libpuffs access() emulation.
1513 	 */
1514 	ps = puffs_getspecific(pu);
1515 	if (ps->ps_flags & PS_NO_ACCESS) {
1516 		const struct vattr *vap;
1517 
1518 		vap = puffs_pn_getvap((struct puffs_node *)opc);
1519 
1520 		error = puffs_access(IFTOVT(vap->va_mode),
1521 				     vap->va_mode & ACCESSPERMS,
1522 				     vap->va_uid, vap->va_gid,
1523 				     (mode_t)mode, pcr);
1524 		goto out;
1525 	}
1526 
1527 	/*
1528 	 * Plain access call
1529 	 */
1530 	pm = ps->ps_new_msg(pu, opc, FUSE_ACCESS, sizeof(*fai), pcr);
1531 	fai = GET_INPAYLOAD(ps, pm, fuse_access_in);
1532 	fai->mask = 0;
1533 	fai->mask |= (mode & PUFFS_VREAD) ? R_OK : 0;
1534 	fai->mask |= (mode & PUFFS_VWRITE) ? W_OK : 0;
1535 	fai->mask |= (mode & PUFFS_VEXEC) ? X_OK : 0;
1536 
1537 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
1538 
1539 	ps->ps_destroy_msg(pm);
1540 
1541 	/*
1542 	 * If unimplemented, start over with emulation
1543 	 */
1544 	if (error == ENOSYS) {
1545 		ps->ps_flags |= PS_NO_ACCESS;
1546 		error = perfuse_node_access(pu, opc, mode, pcr);
1547 	}
1548 
1549 out:
1550 	node_rele(opc);
1551 	return error;
1552 }
1553 
1554 int
1555 perfuse_node_getattr(struct puffs_usermount *pu, puffs_cookie_t opc,
1556 	struct vattr *vap, const struct puffs_cred *pcr)
1557 {
1558 	return perfuse_node_getattr_ttl(pu, opc, vap, pcr, NULL);
1559 }
1560 
1561 int
1562 perfuse_node_getattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
1563 	struct vattr *vap, const struct puffs_cred *pcr,
1564 	struct timespec *va_ttl)
1565 {
1566 	perfuse_msg_t *pm = NULL;
1567 	struct perfuse_state *ps;
1568 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
1569 	struct fuse_getattr_in *fgi;
1570 	struct fuse_attr_out *fao;
1571 	int error = 0;
1572 
1573 	if ((pnd->pnd_flags & PND_REMOVED) && !(pnd->pnd_flags & PND_OPEN))
1574 		return ENOENT;
1575 
1576 	node_ref(opc);
1577 
1578 	/*
1579 	 * Serialize size access, see comment in perfuse_node_setattr().
1580 	 */
1581 	while (pnd->pnd_flags & PND_INRESIZE)
1582 		requeue_request(pu, opc, PCQ_RESIZE);
1583 	pnd->pnd_flags |= PND_INRESIZE;
1584 
1585 	ps = puffs_getspecific(pu);
1586 
1587 	/*
1588 	 * FUSE_GETATTR_FH must be set in fgi->flags
1589 	 * if we use for fgi->fh
1590 	 */
1591 	pm = ps->ps_new_msg(pu, opc, FUSE_GETATTR, sizeof(*fgi), pcr);
1592 	fgi = GET_INPAYLOAD(ps, pm, fuse_getattr_in);
1593 	fgi->getattr_flags = 0;
1594 	fgi->dummy = 0;
1595 	fgi->fh = 0;
1596 
1597 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_OPEN) {
1598 		fgi->fh = perfuse_get_fh(opc, FREAD);
1599 		fgi->getattr_flags |= FUSE_GETATTR_FH;
1600 	}
1601 
1602 #ifdef PERFUSE_DEBUG
1603 	if (perfuse_diagflags & PDF_RESIZE)
1604 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__, (void *)opc,
1605 		    vap->va_size);
1606 #endif
1607 
1608 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
1609 		goto out;
1610 
1611 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1612 
1613 #ifdef PERFUSE_DEBUG
1614 	if (perfuse_diagflags & PDF_RESIZE)
1615 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1616 		    (void *)opc, vap->va_size, fao->attr.size);
1617 #endif
1618 
1619 	/*
1620 	 * We set birthtime, flags, filerev,vaflags to 0.
1621 	 * This seems the best bet, since the information is
1622 	 * not available from filesystem.
1623 	 */
1624 	fuse_attr_to_vap(ps, vap, &fao->attr);
1625 
1626 	if (va_ttl != NULL) {
1627 		va_ttl->tv_sec = fao->attr_valid;
1628 		va_ttl->tv_nsec = fao->attr_valid_nsec;
1629 	}
1630 
1631 	ps->ps_destroy_msg(pm);
1632 	error = 0;
1633 out:
1634 
1635 	pnd->pnd_flags &= ~PND_INRESIZE;
1636 	(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
1637 
1638 	node_rele(opc);
1639 	return error;
1640 }
1641 
1642 int
1643 perfuse_node_setattr(struct puffs_usermount *pu, puffs_cookie_t opc,
1644 	const struct vattr *vap, const struct puffs_cred *pcr)
1645 {
1646 	return perfuse_node_setattr_ttl(pu, opc,
1647 					__UNCONST(vap), pcr, NULL, 0);
1648 }
1649 
1650 int
1651 perfuse_node_setattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
1652 	struct vattr *vap, const struct puffs_cred *pcr,
1653 	struct timespec *va_ttl, int xflag)
1654 {
1655 	perfuse_msg_t *pm;
1656 	uint64_t fh;
1657 	struct perfuse_state *ps;
1658 	struct perfuse_node_data *pnd;
1659 	struct fuse_setattr_in *fsi;
1660 	struct fuse_attr_out *fao;
1661 	struct vattr *old_va;
1662 	enum perfuse_xchg_pb_reply reply;
1663 	int error;
1664 #ifdef PERFUSE_DEBUG
1665 	struct vattr *old_vap;
1666 	int resize_debug = 0;
1667 #endif
1668 	ps = puffs_getspecific(pu);
1669 	pnd = PERFUSE_NODE_DATA(opc);
1670 
1671 	/*
1672 	 * The only operation we can do once the file is removed
1673 	 * is to resize it, and we can do it only if it is open.
1674 	 * Do not even send the operation to the filesystem: the
1675 	 * file is not there anymore.
1676 	 */
1677 	if (pnd->pnd_flags & PND_REMOVED) {
1678 		if (!(pnd->pnd_flags & PND_OPEN))
1679 			return ENOENT;
1680 
1681 		return 0;
1682 	}
1683 
1684 	old_va = puffs_pn_getvap((struct puffs_node *)opc);
1685 
1686 	/*
1687 	 * Check for permission to change size
1688 	 * It is always allowed if we already have a write file handle
1689 	 */
1690 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1691 	    !(pnd->pnd_flags & PND_WFH) &&
1692 	    (error = mode_access(opc, pcr, PUFFS_VWRITE)) != 0)
1693 		return error;
1694 
1695 	/*
1696 	 * Check for permission to change dates
1697 	 */
1698 	if (((vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1699 	     (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL)) &&
1700 	    (puffs_access_times(old_va->va_uid, old_va->va_gid,
1701 				old_va->va_mode, 0, pcr) != 0))
1702 		return EPERM;
1703 
1704 	/*
1705 	 * Check for permission to change owner and group
1706 	 */
1707 	if (((vap->va_uid != (uid_t)PUFFS_VNOVAL) ||
1708 	     (vap->va_gid != (gid_t)PUFFS_VNOVAL)) &&
1709 	    (puffs_access_chown(old_va->va_uid, old_va->va_gid,
1710 				vap->va_uid, vap->va_gid, pcr)) != 0)
1711 		return EPERM;
1712 
1713 	/*
1714 	 * Check for sticky bit on non-directory by non root user
1715 	 */
1716 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1717 	    (vap->va_mode & S_ISTXT) && (old_va->va_type != VDIR) &&
1718 	    !puffs_cred_isjuggernaut(pcr))
1719 		return EFTYPE;
1720 
1721 	/*
1722 	 * Check for permission to change permissions
1723 	 */
1724 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1725 	    (puffs_access_chmod(old_va->va_uid, old_va->va_gid,
1726 				old_va->va_type, vap->va_mode, pcr)) != 0)
1727 		return EPERM;
1728 
1729 	node_ref(opc);
1730 
1731 	if (pnd->pnd_flags & PND_WFH)
1732 		fh = perfuse_get_fh(opc, FWRITE);
1733 	else
1734 		fh = FUSE_UNKNOWN_FH;
1735 
1736 	/*
1737 	 * fchmod() sets mode and fh, and it may carry
1738 	 * a resize as well. That may break if the
1739 	 * filesystem does chmod then resize, and fails
1740 	 * because it does not have permission anymore.
1741 	 * We work this around by splitting into two setattr.
1742 	 */
1743 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1744 	    (vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1745 	    (fh != FUSE_UNKNOWN_FH)) {
1746 		struct vattr resize_va;
1747 
1748 		(void)memcpy(&resize_va, vap, sizeof(resize_va));
1749 		resize_va.va_mode = (mode_t)PUFFS_VNOVAL;
1750 		if ((error = perfuse_node_setattr_ttl(pu, opc, &resize_va,
1751 						      pcr, va_ttl, xflag)) != 0)
1752 			goto out2;
1753 
1754 		vap->va_size = (u_quad_t)PUFFS_VNOVAL;
1755 	}
1756 
1757 	pm = ps->ps_new_msg(pu, opc, FUSE_SETATTR, sizeof(*fsi), pcr);
1758 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
1759 	fsi->valid = 0;
1760 
1761 	/*
1762 	 * Get a fh if the node is open for writing
1763 	 */
1764 	if (fh != FUSE_UNKNOWN_FH) {
1765 		fsi->fh = fh;
1766 		fsi->valid |= FUSE_FATTR_FH;
1767 	}
1768 
1769 
1770 	if (vap->va_size != (u_quad_t)PUFFS_VNOVAL) {
1771 		fsi->size = vap->va_size;
1772 		fsi->valid |= FUSE_FATTR_SIZE;
1773 
1774 		/*
1775 		 * Serialize anything that can touch file size
1776 		 * to avoid reordered GETATTR and SETATTR.
1777 		 * Out of order SETATTR can report stale size,
1778 		 * which will cause the kernel to truncate the file.
1779 		 * XXX Probably useless now we have a lock on GETATTR
1780 		 */
1781 		while (pnd->pnd_flags & PND_INRESIZE)
1782 			requeue_request(pu, opc, PCQ_RESIZE);
1783 		pnd->pnd_flags |= PND_INRESIZE;
1784 	}
1785 
1786 	/*
1787  	 * When not sending a time field, still fill with
1788 	 * current value, as the filesystem may just reset
1789 	 * the field to Epoch even if fsi->valid bit is
1790 	 * not set (GlusterFS does that).
1791  	 */
1792 	if (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) {
1793 		fsi->atime = vap->va_atime.tv_sec;
1794 		fsi->atimensec = (uint32_t)vap->va_atime.tv_nsec;
1795 		fsi->valid |= FUSE_FATTR_ATIME;
1796 	} else {
1797 		fsi->atime = old_va->va_atime.tv_sec;
1798 		fsi->atimensec = (uint32_t)old_va->va_atime.tv_nsec;
1799 	}
1800 
1801 	if (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) {
1802 		fsi->mtime = vap->va_mtime.tv_sec;
1803 		fsi->mtimensec = (uint32_t)vap->va_mtime.tv_nsec;
1804 		fsi->valid |= FUSE_FATTR_MTIME;
1805 	} else {
1806 		fsi->mtime = old_va->va_mtime.tv_sec;
1807 		fsi->mtimensec = (uint32_t)old_va->va_mtime.tv_nsec;
1808 	}
1809 
1810 	if (vap->va_mode != (mode_t)PUFFS_VNOVAL) {
1811 		fsi->mode = vap->va_mode;
1812 		fsi->valid |= FUSE_FATTR_MODE;
1813 	}
1814 
1815 	if (vap->va_uid != (uid_t)PUFFS_VNOVAL) {
1816 		fsi->uid = vap->va_uid;
1817 		fsi->valid |= FUSE_FATTR_UID;
1818 	}
1819 
1820 	if (vap->va_gid != (gid_t)PUFFS_VNOVAL) {
1821 		fsi->gid = vap->va_gid;
1822 		fsi->valid |= FUSE_FATTR_GID;
1823 	}
1824 
1825 	if (pnd->pnd_lock_owner != 0) {
1826 		fsi->lock_owner = pnd->pnd_lock_owner;
1827 		fsi->valid |= FUSE_FATTR_LOCKOWNER;
1828 	}
1829 
1830 #ifndef PUFFS_KFLAG_NOFLUSH_META
1831 	/*
1832 	 * ftruncate() sends only va_size, and metadata cache
1833 	 * flush adds va_atime and va_mtime. Some FUSE
1834 	 * filesystems will attempt to detect ftruncate by
1835 	 * checking for FATTR_SIZE being set without
1836 	 * FATTR_UID|FATTR_GID|FATTR_ATIME|FATTR_MTIME|FATTR_MODE
1837 	 *
1838 	 * Try to adapt and remove FATTR_ATIME|FATTR_MTIME
1839 	 * if we suspect a ftruncate().
1840 	 */
1841 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1842 	    ((vap->va_mode == (mode_t)PUFFS_VNOVAL) &&
1843 	     (vap->va_uid == (uid_t)PUFFS_VNOVAL) &&
1844 	     (vap->va_gid == (gid_t)PUFFS_VNOVAL))) {
1845 		fsi->atime = 0;
1846 		fsi->atimensec = 0;
1847 		fsi->mtime = 0;
1848 		fsi->mtimensec = 0;
1849 		fsi->valid &= ~(FUSE_FATTR_ATIME|FUSE_FATTR_MTIME);
1850 	}
1851 
1852 	/*
1853 	 * If only atime is changed, discard the operation: it
1854 	 * happens after read, and in that case the filesystem
1855 	 * already updaed atime. NB: utimes() also change mtime.
1856 	 */
1857 	if (fsi->valid == FUSE_FATTR_ATIME)
1858 		fsi->valid &= ~FUSE_FATTR_ATIME;
1859 #endif /* PUFFS_KFLAG_NOFLUSH_META */
1860 
1861 	/*
1862 	 * If nothing remain, discard the operation.
1863 	 */
1864 	if (!(fsi->valid & (FUSE_FATTR_SIZE|FUSE_FATTR_ATIME|FUSE_FATTR_MTIME|
1865 			    FUSE_FATTR_MODE|FUSE_FATTR_UID|FUSE_FATTR_GID))) {
1866 		error = 0;
1867 		ps->ps_destroy_msg(pm);
1868 		goto out;
1869 	}
1870 
1871 #ifdef PERFUSE_DEBUG
1872 	old_vap = puffs_pn_getvap((struct puffs_node *)opc);
1873 
1874 	if ((perfuse_diagflags & PDF_RESIZE) &&
1875 	    (old_vap->va_size != (u_quad_t)PUFFS_VNOVAL)) {
1876 		resize_debug = 1;
1877 
1878 		DPRINTF(">> %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1879 		    (void *)opc,
1880 		    puffs_pn_getvap((struct puffs_node *)opc)->va_size,
1881 		    fsi->size);
1882 	}
1883 #endif
1884 
1885 	/*
1886 	 * Do not honour FAF when changing size. How do
1887 	 * you want such a thing to work?
1888 	 */
1889 	reply = wait_reply;
1890 #ifdef PUFFS_SETATTR_FAF
1891 	if ((xflag & PUFFS_SETATTR_FAF) && !(fsi->valid & FUSE_FATTR_SIZE))
1892 		reply = no_reply;
1893 #endif
1894 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), reply)) != 0)
1895 		goto out;
1896 
1897 	if (reply == no_reply)
1898 		goto out;
1899 
1900 	/*
1901 	 * Copy back the new values
1902 	 */
1903 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1904 
1905 #ifdef PERFUSE_DEBUG
1906 	if (resize_debug)
1907 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1908 		    (void *)opc, old_vap->va_size, fao->attr.size);
1909 #endif
1910 
1911 	fuse_attr_to_vap(ps, old_va, &fao->attr);
1912 
1913 	if (va_ttl != NULL) {
1914 		va_ttl->tv_sec = fao->attr_valid;
1915 		va_ttl->tv_nsec = fao->attr_valid_nsec;
1916 		(void)memcpy(vap, old_va, sizeof(*vap));
1917 	}
1918 
1919 	ps->ps_destroy_msg(pm);
1920 	error = 0;
1921 
1922 out:
1923 	if (pnd->pnd_flags & PND_INRESIZE) {
1924 		pnd->pnd_flags &= ~PND_INRESIZE;
1925 		(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
1926 	}
1927 
1928 out2:
1929 	node_rele(opc);
1930 	return error;
1931 }
1932 
1933 int
1934 perfuse_node_poll(struct puffs_usermount *pu, puffs_cookie_t opc, int *events)
1935 {
1936 	struct perfuse_state *ps;
1937 	perfuse_msg_t *pm;
1938 	struct fuse_poll_in *fpi;
1939 	struct fuse_poll_out *fpo;
1940 	int error;
1941 
1942 	node_ref(opc);
1943 	ps = puffs_getspecific(pu);
1944 	/*
1945 	 * kh is set if FUSE_POLL_SCHEDULE_NOTIFY is set.
1946 	 *
1947 	 * XXX ps_new_msg() is called with NULL creds, which will
1948 	 * be interpreted as FUSE superuser. We have no way to
1949 	 * know the requesting process' credential, but since poll
1950 	 * is supposed to operate on a file that has been open,
1951 	 * permission should have already been checked at open time.
1952 	 * That still may breaks on filesystems that provides odd
1953 	 * semantics.
1954  	 */
1955 	pm = ps->ps_new_msg(pu, opc, FUSE_POLL, sizeof(*fpi), NULL);
1956 	fpi = GET_INPAYLOAD(ps, pm, fuse_poll_in);
1957 	fpi->fh = perfuse_get_fh(opc, FREAD);
1958 	fpi->kh = 0;
1959 	fpi->flags = 0;
1960 
1961 #ifdef PERFUSE_DEBUG
1962 	if (perfuse_diagflags & PDF_FH)
1963 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
1964 			"fh = 0x%"PRIx64"\n", __func__, (void *)opc,
1965 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fpi->fh);
1966 #endif
1967 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fpo), wait_reply)) != 0)
1968 		goto out;
1969 
1970 	fpo = GET_OUTPAYLOAD(ps, pm, fuse_poll_out);
1971 	*events = fpo->revents;
1972 
1973 	ps->ps_destroy_msg(pm);
1974 	error = 0;
1975 
1976 out:
1977 	node_rele(opc);
1978 	return error;
1979 }
1980 
1981 /* ARGSUSED2 */
1982 int
1983 perfuse_node_fsync(struct puffs_usermount *pu, puffs_cookie_t opc,
1984 	const struct puffs_cred *pcr, int flags, off_t offlo, off_t offhi)
1985 {
1986 	int op;
1987 	perfuse_msg_t *pm;
1988 	struct perfuse_state *ps;
1989 	struct perfuse_node_data *pnd;
1990 	struct fuse_fsync_in *ffi;
1991 	uint64_t fh;
1992 	int error = 0;
1993 
1994 	pm = NULL;
1995 	ps = puffs_getspecific(pu);
1996 	pnd = PERFUSE_NODE_DATA(opc);
1997 
1998 	/*
1999 	 * No need to sync a removed node
2000 	 */
2001 	if (pnd->pnd_flags & PND_REMOVED)
2002 		return 0;
2003 
2004 	/*
2005 	 * We do not sync closed files. They have been
2006 	 * sync at inactive time already.
2007 	 */
2008 	if (!(pnd->pnd_flags & PND_OPEN))
2009 		return 0;
2010 
2011 	node_ref(opc);
2012 
2013 	if (puffs_pn_getvap((struct puffs_node *)opc)->va_type == VDIR)
2014 		op = FUSE_FSYNCDIR;
2015 	else 		/* VREG but also other types such as VLNK */
2016 		op = FUSE_FSYNC;
2017 
2018 	/*
2019 	 * Do not sync if there are no change to sync
2020 	 * XXX remove that test on files if we implement mmap
2021 	 */
2022 #ifdef PERFUSE_DEBUG
2023 	if (perfuse_diagflags & PDF_SYNC)
2024 		DPRINTF("%s: TEST opc = %p, file = \"%s\" is %sdirty\n",
2025 			__func__, (void*)opc, perfuse_node_path(ps, opc),
2026 			pnd->pnd_flags & PND_DIRTY ? "" : "not ");
2027 #endif
2028 	if (!(pnd->pnd_flags & PND_DIRTY))
2029 		goto out;
2030 
2031 	/*
2032 	 * It seems NetBSD can call fsync without open first
2033 	 * glusterfs complain in such a situation:
2034 	 * "FSYNC() ERR => -1 (Invalid argument)"
2035 	 * The file will be closed at inactive time.
2036 	 *
2037 	 * We open the directory for reading in order to sync.
2038 	 * This sounds rather counterintuitive, but it works.
2039 	 */
2040 	if (!(pnd->pnd_flags & PND_WFH)) {
2041 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2042 			goto out;
2043 	}
2044 
2045 	if (op == FUSE_FSYNCDIR)
2046 		fh = perfuse_get_fh(opc, FREAD);
2047 	else
2048 		fh = perfuse_get_fh(opc, FWRITE);
2049 
2050 	/*
2051 	 * If fsync_flags  is set, meta data should not be flushed.
2052 	 */
2053 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*ffi), pcr);
2054 	ffi = GET_INPAYLOAD(ps, pm, fuse_fsync_in);
2055 	ffi->fh = fh;
2056 	ffi->fsync_flags = (flags & FFILESYNC) ? 0 : 1;
2057 
2058 #ifdef PERFUSE_DEBUG
2059 	if (perfuse_diagflags & PDF_FH)
2060 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
2061 			__func__, (void *)opc,
2062 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, ffi->fh);
2063 #endif
2064 
2065 	if ((error = xchg_msg(pu, opc, pm,
2066 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
2067 		goto out;
2068 
2069 	/*
2070 	 * No reply beyond fuse_out_header: nothing to do on success
2071 	 * just clear the dirty flag
2072 	 */
2073 	pnd->pnd_flags &= ~PND_DIRTY;
2074 
2075 #ifdef PERFUSE_DEBUG
2076 	if (perfuse_diagflags & PDF_SYNC)
2077 		DPRINTF("%s: CLEAR opc = %p, file = \"%s\"\n",
2078 			__func__, (void*)opc, perfuse_node_path(ps, opc));
2079 #endif
2080 
2081 	ps->ps_destroy_msg(pm);
2082 	error = 0;
2083 
2084 out:
2085 	/*
2086 	 * ENOSYS is not returned to kernel,
2087 	 */
2088 	if (error == ENOSYS)
2089 		error = 0;
2090 
2091 	node_rele(opc);
2092 	return error;
2093 }
2094 
2095 int
2096 perfuse_node_remove(struct puffs_usermount *pu, puffs_cookie_t opc,
2097 	puffs_cookie_t targ, const struct puffs_cn *pcn)
2098 {
2099 	struct perfuse_state *ps;
2100 	struct perfuse_node_data *pnd;
2101 	perfuse_msg_t *pm;
2102 	char *path;
2103 	const char *name;
2104 	size_t len;
2105 	int error;
2106 
2107 	pnd = PERFUSE_NODE_DATA(opc);
2108 
2109 	if ((pnd->pnd_flags & PND_REMOVED) ||
2110 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
2111 		return ENOENT;
2112 
2113 #ifdef PERFUSE_DEBUG
2114 	if (targ == NULL)
2115 		DERRX(EX_SOFTWARE, "%s: targ is NULL", __func__);
2116 
2117 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
2118 		DPRINTF("%s: opc = %p, remove opc = %p, file = \"%s\"\n",
2119 			__func__, (void *)opc, (void *)targ, pcn->pcn_name);
2120 #endif
2121 	node_ref(opc);
2122 	node_ref(targ);
2123 
2124 	/*
2125 	 * Await for all operations on the deleted node to drain,
2126 	 * as the filesystem may be confused to have it deleted
2127 	 * during a getattr
2128 	 */
2129 	while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
2130 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2131 
2132 	ps = puffs_getspecific(pu);
2133 	pnd = PERFUSE_NODE_DATA(opc);
2134 	name = pcn->pcn_name;
2135 	len = pcn->pcn_namelen + 1;
2136 
2137 	pm = ps->ps_new_msg(pu, opc, FUSE_UNLINK, len, pcn->pcn_cred);
2138 	path = _GET_INPAYLOAD(ps, pm, char *);
2139 	(void)strlcpy(path, name, len);
2140 
2141 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2142 		goto out;
2143 
2144 	perfuse_cache_flush(targ);
2145 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2146 
2147 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2148 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2149 
2150 	/*
2151 	 * The parent directory needs a sync
2152 	 */
2153 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2154 
2155 #ifdef PERFUSE_DEBUG
2156 	if (perfuse_diagflags & PDF_FILENAME)
2157 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
2158 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
2159 			pcn->pcn_name);
2160 #endif
2161 	ps->ps_destroy_msg(pm);
2162 	error = 0;
2163 
2164 out:
2165 	node_rele(opc);
2166 	node_rele(targ);
2167 	return error;
2168 }
2169 
2170 int
2171 perfuse_node_link(struct puffs_usermount *pu, puffs_cookie_t opc,
2172 	puffs_cookie_t targ, const struct puffs_cn *pcn)
2173 {
2174 	struct perfuse_state *ps;
2175 	perfuse_msg_t *pm;
2176 	const char *name;
2177 	size_t len;
2178 	struct puffs_node *pn;
2179 	struct fuse_link_in *fli;
2180 	int error;
2181 
2182 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2183 		return ENOENT;
2184 
2185 	node_ref(opc);
2186 	node_ref(targ);
2187 	ps = puffs_getspecific(pu);
2188 	pn = (struct puffs_node *)targ;
2189 	name = pcn->pcn_name;
2190 	len =  sizeof(*fli) + pcn->pcn_namelen + 1;
2191 
2192 	pm = ps->ps_new_msg(pu, opc, FUSE_LINK, len, pcn->pcn_cred);
2193 	fli = GET_INPAYLOAD(ps, pm, fuse_link_in);
2194 	fli->oldnodeid = PERFUSE_NODE_DATA(pn)->pnd_nodeid;
2195 	(void)strlcpy((char *)(void *)(fli + 1), name, len - sizeof(*fli));
2196 
2197 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2198 		goto out;
2199 
2200 	ps->ps_destroy_msg(pm);
2201 	error = 0;
2202 
2203 out:
2204 	node_rele(opc);
2205 	node_rele(targ);
2206 	return error;
2207 }
2208 
2209 int
2210 perfuse_node_rename(struct puffs_usermount *pu, puffs_cookie_t opc,
2211 	puffs_cookie_t src, const struct puffs_cn *pcn_src,
2212 	puffs_cookie_t targ_dir, puffs_cookie_t targ,
2213 	const struct puffs_cn *pcn_targ)
2214 {
2215 	struct perfuse_state *ps;
2216 	struct perfuse_node_data *dstdir_pnd;
2217 	perfuse_msg_t *pm;
2218 	struct fuse_rename_in *fri;
2219 	const char *newname;
2220 	const char *oldname;
2221 	char *np;
2222 	int error;
2223 	size_t len;
2224 	size_t newname_len;
2225 	size_t oldname_len;
2226 
2227 	if ((PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED) ||
2228 	    (PERFUSE_NODE_DATA(src)->pnd_flags & PND_REMOVED) ||
2229 	    (PERFUSE_NODE_DATA(targ_dir)->pnd_flags & PND_REMOVED))
2230 		return ENOENT;
2231 
2232 	node_ref(opc);
2233 	node_ref(src);
2234 
2235 	/*
2236 	 * Await for all operations on the deleted node to drain,
2237 	 * as the filesystem may be confused to have it deleted
2238 	 * during a getattr
2239 	 */
2240 	if ((struct puffs_node *)targ != NULL) {
2241 		node_ref(targ);
2242 		while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
2243 			requeue_request(pu, targ, PCQ_AFTERXCHG);
2244 	} else {
2245 		while (PERFUSE_NODE_DATA(src)->pnd_inxchg)
2246 			requeue_request(pu, src, PCQ_AFTERXCHG);
2247 	}
2248 
2249 	ps = puffs_getspecific(pu);
2250 	newname =  pcn_targ->pcn_name;
2251 	newname_len = pcn_targ->pcn_namelen + 1;
2252 	oldname =  pcn_src->pcn_name;
2253 	oldname_len = pcn_src->pcn_namelen + 1;
2254 
2255 	len = sizeof(*fri) + oldname_len + newname_len;
2256 	pm = ps->ps_new_msg(pu, opc, FUSE_RENAME, len, pcn_targ->pcn_cred);
2257 	fri = GET_INPAYLOAD(ps, pm, fuse_rename_in);
2258 	fri->newdir = PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid;
2259 	np = (char *)(void *)(fri + 1);
2260 	(void)strlcpy(np, oldname, oldname_len);
2261 	np += oldname_len;
2262 	(void)strlcpy(np, newname, newname_len);
2263 
2264 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2265 		goto out;
2266 
2267 
2268 	/*
2269 	 * Record new parent nodeid
2270 	 */
2271 	dstdir_pnd = PERFUSE_NODE_DATA(targ_dir);
2272 	PERFUSE_NODE_DATA(src)->pnd_parent_nodeid = dstdir_pnd->pnd_nodeid;
2273 
2274 	if (opc != targ_dir)
2275 		dstdir_pnd->pnd_flags |= PND_DIRTY;
2276 
2277 	if (strcmp(newname, "..") != 0)
2278 		(void)strlcpy(PERFUSE_NODE_DATA(src)->pnd_name,
2279 		    newname, MAXPATHLEN);
2280 	else
2281 		PERFUSE_NODE_DATA(src)->pnd_name[0] = 0; /* forget name */
2282 
2283 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2284 
2285 	if ((struct puffs_node *)targ != NULL) {
2286 		perfuse_cache_flush(targ);
2287 		PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2288 	}
2289 
2290 #ifdef PERFUSE_DEBUG
2291 	if (perfuse_diagflags & PDF_FILENAME)
2292 		DPRINTF("%s: nodeid = 0x%"PRIx64" file = \"%s\" renamed \"%s\" "
2293 			"nodeid = 0x%"PRIx64" -> nodeid = 0x%"PRIx64" \"%s\"\n",
2294 	 		__func__, PERFUSE_NODE_DATA(src)->pnd_nodeid,
2295 			pcn_src->pcn_name, pcn_targ->pcn_name,
2296 			PERFUSE_NODE_DATA(opc)->pnd_nodeid,
2297 			PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid,
2298 			perfuse_node_path(ps, targ_dir));
2299 #endif
2300 
2301 	ps->ps_destroy_msg(pm);
2302 	error = 0;
2303 
2304 out:
2305 	node_rele(opc);
2306 	node_rele(src);
2307 	if ((struct puffs_node *)targ != NULL)
2308 		node_rele(targ);
2309 
2310 	return error;
2311 }
2312 
2313 int
2314 perfuse_node_mkdir(struct puffs_usermount *pu, puffs_cookie_t opc,
2315 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
2316 	const struct vattr *vap)
2317 {
2318 	struct perfuse_state *ps;
2319 	perfuse_msg_t *pm;
2320 	struct fuse_mkdir_in *fmi;
2321 	const char *path;
2322 	size_t len;
2323 	int error;
2324 
2325 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2326 		return ENOENT;
2327 
2328 	node_ref(opc);
2329 	ps = puffs_getspecific(pu);
2330 	path = pcn->pcn_name;
2331 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
2332 
2333 	pm = ps->ps_new_msg(pu, opc, FUSE_MKDIR, len, pcn->pcn_cred);
2334 	fmi = GET_INPAYLOAD(ps, pm, fuse_mkdir_in);
2335 	fmi->mode = vap->va_mode;
2336 	fmi->umask = 0; 	/* Seems unused by libfuse? */
2337 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
2338 
2339 	error = node_mk_common(pu, opc, pni, pcn, pm);
2340 
2341 	node_rele(opc);
2342 	return error;
2343 }
2344 
2345 
2346 int
2347 perfuse_node_rmdir(struct puffs_usermount *pu, puffs_cookie_t opc,
2348 	puffs_cookie_t targ, const struct puffs_cn *pcn)
2349 {
2350 	struct perfuse_state *ps;
2351 	struct perfuse_node_data *pnd;
2352 	perfuse_msg_t *pm;
2353 	char *path;
2354 	const char *name;
2355 	size_t len;
2356 	int error;
2357 
2358 	pnd = PERFUSE_NODE_DATA(opc);
2359 
2360 	if ((pnd->pnd_flags & PND_REMOVED) ||
2361 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
2362 		return ENOENT;
2363 
2364 	/*
2365 	 * Attempt to rmdir dir/.. shoud raise ENOTEMPTY
2366 	 */
2367 	if (PERFUSE_NODE_DATA(targ)->pnd_nodeid == pnd->pnd_parent_nodeid)
2368 		return ENOTEMPTY;
2369 
2370 	node_ref(opc);
2371 	node_ref(targ);
2372 
2373 	/*
2374 	 * Await for all operations on the deleted node to drain,
2375 	 * as the filesystem may be confused to have it deleted
2376 	 * during a getattr
2377 	 */
2378 	while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
2379 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2380 
2381 	ps = puffs_getspecific(pu);
2382 	name = pcn->pcn_name;
2383 	len = pcn->pcn_namelen + 1;
2384 
2385 	pm = ps->ps_new_msg(pu, opc, FUSE_RMDIR, len, pcn->pcn_cred);
2386 	path = _GET_INPAYLOAD(ps, pm, char *);
2387 	(void)strlcpy(path, name, len);
2388 
2389 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2390 		goto out;
2391 
2392 	perfuse_cache_flush(targ);
2393 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2394 
2395 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2396 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2397 
2398 	/*
2399 	 * The parent directory needs a sync
2400 	 */
2401 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2402 
2403 #ifdef PERFUSE_DEBUG
2404 	if (perfuse_diagflags & PDF_FILENAME)
2405 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
2406 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
2407 			perfuse_node_path(ps, targ));
2408 #endif
2409 	ps->ps_destroy_msg(pm);
2410 	error = 0;
2411 
2412 out:
2413 	node_rele(opc);
2414 	node_rele(targ);
2415 	return error;
2416 }
2417 
2418 /* vap is unused */
2419 /* ARGSUSED4 */
2420 int
2421 perfuse_node_symlink(struct puffs_usermount *pu, puffs_cookie_t opc,
2422 	struct puffs_newinfo *pni, const struct puffs_cn *pcn_src,
2423 	const struct vattr *vap, const char *link_target)
2424 {
2425 	struct perfuse_state *ps;
2426 	perfuse_msg_t *pm;
2427 	char *np;
2428 	const char *path;
2429 	size_t path_len;
2430 	size_t linkname_len;
2431 	size_t len;
2432 	int error;
2433 
2434 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2435 		return ENOENT;
2436 
2437 	node_ref(opc);
2438 	ps = puffs_getspecific(pu);
2439 	path = pcn_src->pcn_name;
2440 	path_len = pcn_src->pcn_namelen + 1;
2441 	linkname_len = strlen(link_target) + 1;
2442 	len = path_len + linkname_len;
2443 
2444 	pm = ps->ps_new_msg(pu, opc, FUSE_SYMLINK, len, pcn_src->pcn_cred);
2445 	np = _GET_INPAYLOAD(ps, pm, char *);
2446 	(void)strlcpy(np, path, path_len);
2447 	np += path_len;
2448 	(void)strlcpy(np, link_target, linkname_len);
2449 
2450 	error = node_mk_common(pu, opc, pni, pcn_src, pm);
2451 
2452 	node_rele(opc);
2453 	return error;
2454 }
2455 
2456 /* ARGSUSED4 */
2457 int
2458 perfuse_node_readdir(struct puffs_usermount *pu, puffs_cookie_t opc,
2459 	struct dirent *dent, off_t *readoff, size_t *reslen,
2460 	const struct puffs_cred *pcr, int *eofflag, off_t *cookies,
2461 	size_t *ncookies)
2462 {
2463 	perfuse_msg_t *pm;
2464 	uint64_t fh;
2465 	struct perfuse_state *ps;
2466 	struct perfuse_node_data *pnd;
2467 	struct fuse_read_in *fri;
2468 	struct fuse_out_header *foh;
2469 	struct fuse_dirent *fd;
2470 	size_t foh_len;
2471 	int error;
2472 	size_t fd_maxlen;
2473 
2474 	error = 0;
2475 	node_ref(opc);
2476 	ps = puffs_getspecific(pu);
2477 
2478 	/*
2479 	 * readdir state is kept at node level, and several readdir
2480 	 * requests can be issued at the same time on the same node.
2481 	 * We need to queue requests so that only one is in readdir
2482 	 * code at the same time.
2483 	 */
2484 	pnd = PERFUSE_NODE_DATA(opc);
2485 	while (pnd->pnd_flags & PND_INREADDIR)
2486 		requeue_request(pu, opc, PCQ_READDIR);
2487 	pnd->pnd_flags |= PND_INREADDIR;
2488 
2489 #ifdef PERFUSE_DEBUG
2490 	if (perfuse_diagflags & PDF_READDIR)
2491 		DPRINTF("%s: READDIR opc = %p enter critical section\n",
2492 			__func__, (void *)opc);
2493 #endif
2494 	/*
2495 	 * Re-initialize pnd->pnd_fd_cookie on the first readdir for a node
2496 	 */
2497 	if (*readoff == 0)
2498 		pnd->pnd_fd_cookie = 0;
2499 
2500 	/*
2501 	 * Do we already have the data bufered?
2502 	 */
2503 	if (pnd->pnd_dirent != NULL)
2504 		goto out;
2505 	pnd->pnd_dirent_len = 0;
2506 
2507 	/*
2508 	 * It seems NetBSD can call readdir without open first
2509 	 * libfuse will crash if it is done that way, hence open first.
2510 	 */
2511 	if (!(pnd->pnd_flags & PND_OPEN)) {
2512 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2513 			goto out;
2514 	}
2515 
2516 	fh = perfuse_get_fh(opc, FREAD);
2517 
2518 #ifdef PERFUSE_DEBUG
2519 	if (perfuse_diagflags & PDF_FH)
2520 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
2521 			"rfh = 0x%"PRIx64"\n", __func__, (void *)opc,
2522 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fh);
2523 #endif
2524 
2525 	pnd->pnd_all_fd = NULL;
2526 	pnd->pnd_all_fd_len = 0;
2527 	fd_maxlen = ps->ps_max_readahead - sizeof(*foh);
2528 
2529 	do {
2530 		size_t fd_len;
2531 		char *afdp;
2532 
2533 		pm = ps->ps_new_msg(pu, opc, FUSE_READDIR, sizeof(*fri), pcr);
2534 
2535 		/*
2536 		 * read_flags, lock_owner and flags are unused in libfuse
2537 		 */
2538 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
2539 		fri->fh = fh;
2540 		fri->offset = pnd->pnd_fd_cookie;
2541 		fri->size = (uint32_t)fd_maxlen;
2542 		fri->read_flags = 0;
2543 		fri->lock_owner = 0;
2544 		fri->flags = 0;
2545 
2546 		if ((error = xchg_msg(pu, opc, pm,
2547 				      UNSPEC_REPLY_LEN, wait_reply)) != 0)
2548 			goto out;
2549 
2550 		/*
2551 		 * There are many puffs_framebufs calls later,
2552 		 * therefore foh will not be valid for a long time.
2553 		 * Just get the length and forget it.
2554 		 */
2555 		foh = GET_OUTHDR(ps, pm);
2556 		foh_len = foh->len;
2557 
2558 		/*
2559 		 * Empty read: we reached the end of the buffer.
2560 		 */
2561 		if (foh_len == sizeof(*foh)) {
2562 			ps->ps_destroy_msg(pm);
2563 			*eofflag = 1;
2564 			break;
2565 		}
2566 
2567 		/*
2568 		 * Check for corrupted message.
2569 		 */
2570 		if (foh_len < sizeof(*foh) + sizeof(*fd)) {
2571 			ps->ps_destroy_msg(pm);
2572 			DWARNX("readdir reply too short");
2573 			error = EIO;
2574 			goto out;
2575 		}
2576 
2577 
2578 		fd = GET_OUTPAYLOAD(ps, pm, fuse_dirent);
2579 		fd_len = foh_len - sizeof(*foh);
2580 
2581 		pnd->pnd_all_fd = realloc(pnd->pnd_all_fd,
2582 					  pnd->pnd_all_fd_len + fd_len);
2583 		if (pnd->pnd_all_fd  == NULL)
2584 			DERR(EX_OSERR, "%s: malloc failed", __func__);
2585 
2586 		afdp = (char *)(void *)pnd->pnd_all_fd + pnd->pnd_all_fd_len;
2587 		(void)memcpy(afdp, fd, fd_len);
2588 
2589 		pnd->pnd_all_fd_len += fd_len;
2590 
2591 		/*
2592 		 * The fd->off field is used as a cookie for
2593 		 * resuming the next readdir() where this one was left.
2594 	 	 */
2595 		pnd->pnd_fd_cookie = readdir_last_cookie(fd, fd_len);
2596 
2597 		ps->ps_destroy_msg(pm);
2598 	} while (1 /* CONSTCOND */);
2599 
2600 	if (pnd->pnd_all_fd != NULL) {
2601 		if (fuse_to_dirent(pu, opc, pnd->pnd_all_fd,
2602 				   pnd->pnd_all_fd_len) == -1)
2603 			error = EIO;
2604 	}
2605 
2606 out:
2607 	if (pnd->pnd_all_fd != NULL) {
2608 		free(pnd->pnd_all_fd);
2609 		pnd->pnd_all_fd = NULL;
2610 		pnd->pnd_all_fd_len = 0;
2611 	}
2612 
2613 	if (error == 0)
2614 		readdir_buffered(opc, dent, readoff, reslen);
2615 
2616 	/*
2617 	 * Schedule queued readdir requests
2618 	 */
2619 	pnd->pnd_flags &= ~PND_INREADDIR;
2620 	(void)dequeue_requests(opc, PCQ_READDIR, DEQUEUE_ALL);
2621 
2622 #ifdef PERFUSE_DEBUG
2623 	if (perfuse_diagflags & PDF_READDIR)
2624 		DPRINTF("%s: READDIR opc = %p exit critical section\n",
2625 			__func__, (void *)opc);
2626 #endif
2627 
2628 	node_rele(opc);
2629 	return error;
2630 }
2631 
2632 int
2633 perfuse_node_readlink(struct puffs_usermount *pu, puffs_cookie_t opc,
2634 	const struct puffs_cred *pcr, char *linkname, size_t *linklen)
2635 {
2636 	struct perfuse_state *ps;
2637 	perfuse_msg_t *pm;
2638 	int error;
2639 	size_t len;
2640 	struct fuse_out_header *foh;
2641 
2642 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2643 		return ENOENT;
2644 
2645 	node_ref(opc);
2646 	ps = puffs_getspecific(pu);
2647 
2648 	pm = ps->ps_new_msg(pu, opc, FUSE_READLINK, 0, pcr);
2649 
2650 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2651 		goto out;
2652 
2653 	foh = GET_OUTHDR(ps, pm);
2654 	len = foh->len - sizeof(*foh);
2655 	if (len > *linklen)
2656 		DERRX(EX_PROTOCOL, "path len = %zd too long", len);
2657 	if (len == 0)
2658 		DERRX(EX_PROTOCOL, "path len = %zd too short", len);
2659 
2660 	(void)memcpy(linkname, _GET_OUTPAYLOAD(ps, pm, char *), len);
2661 
2662 	/*
2663 	 * FUSE filesystems return a NUL terminated string, we
2664 	 * do not want the trailing \0
2665 	 */
2666 	while (len > 0 && linkname[len - 1] == '\0')
2667 		len--;
2668 
2669 	*linklen = len;
2670 
2671 	ps->ps_destroy_msg(pm);
2672 	error = 0;
2673 
2674 out:
2675 	node_rele(opc);
2676 	return error;
2677 }
2678 
2679 int
2680 perfuse_node_reclaim2(struct puffs_usermount *pu,
2681 		      puffs_cookie_t opc, int nlookup)
2682 {
2683 	struct perfuse_state *ps;
2684 	perfuse_msg_t *pm;
2685 	struct perfuse_node_data *pnd;
2686 	struct fuse_forget_in *ffi;
2687 
2688 #ifdef PERFUSE_DEBUG
2689 		if (perfuse_diagflags & PDF_RECLAIM)
2690 			DPRINTF("%s called with opc = %p, nlookup = %d\n",
2691 				__func__, (void *)opc, nlookup);
2692 #endif
2693 	if (opc == 0 || nlookup == 0) {
2694 		return 0;
2695 	}
2696 
2697 	ps = puffs_getspecific(pu);
2698 	pnd = PERFUSE_NODE_DATA(opc);
2699 
2700 	/*
2701 	 * Never forget the root.
2702 	 */
2703 	if (pnd->pnd_nodeid == FUSE_ROOT_ID)
2704 		return 0;
2705 
2706 #ifdef PERFUSE_DEBUG
2707 	if (perfuse_diagflags & PDF_RECLAIM)
2708 		DPRINTF("%s (nodeid %"PRId64") reclaimed, nlookup = %d/%d\n",
2709 			perfuse_node_path(ps, opc), pnd->pnd_nodeid,
2710 			nlookup, pnd->pnd_puffs_nlookup);
2711 #endif
2712 	/*
2713 	 * The kernel tells us how many lookups it made, which allows
2714 	 * us to detect that we have an uncompleted lookup and that the
2715 	 * node should not dispear.
2716 	 */
2717 	pnd->pnd_puffs_nlookup -= nlookup;
2718 	if (pnd->pnd_puffs_nlookup > 0)
2719 		return 0;
2720 
2721 	node_ref(opc);
2722 	pnd->pnd_flags |= PND_RECLAIMED;
2723 
2724 #ifdef PERFUSE_DEBUG
2725 	if (perfuse_diagflags & PDF_RECLAIM)
2726 		DPRINTF("%s (nodeid %"PRId64") is %sreclaimed, nlookup = %d "
2727 			"%s%s%s%s, pending ops:%s%s%s\n",
2728 		        perfuse_node_path(ps, opc), pnd->pnd_nodeid,
2729 		        pnd->pnd_flags & PND_RECLAIMED ? "" : "not ",
2730 			pnd->pnd_puffs_nlookup,
2731 			pnd->pnd_flags & PND_OPEN ? "open " : "not open",
2732 			pnd->pnd_flags & PND_RFH ? "r" : "",
2733 			pnd->pnd_flags & PND_WFH ? "w" : "",
2734 			pnd->pnd_flags & PND_BUSY ? " busy" : "",
2735 			pnd->pnd_flags & PND_INREADDIR ? " readdir" : "",
2736 			pnd->pnd_flags & PND_INWRITE ? " write" : "",
2737 			pnd->pnd_flags & PND_INOPEN ? " open" : "");
2738 #endif
2739 	/*
2740 	 * Make sure it is not looked up again
2741 	 */
2742 	if (!(pnd->pnd_flags & PND_REMOVED))
2743 		perfuse_cache_flush(opc);
2744 
2745 	/*
2746 	 * Purge any activity on the node, while checking
2747 	 * that it remains eligible for a reclaim.
2748 	 */
2749 	while (pnd->pnd_ref > 1)
2750 		requeue_request(pu, opc, PCQ_REF);
2751 
2752 #ifdef PERFUSE_DEBUG
2753 	if ((pnd->pnd_flags & PND_OPEN) ||
2754 	       !TAILQ_EMPTY(&pnd->pnd_pcq))
2755 		DERRX(EX_SOFTWARE, "%s: opc = %p \"%s\": still open",
2756 		      __func__, opc, pnd->pnd_name);
2757 
2758 	if ((pnd->pnd_flags & PND_BUSY) ||
2759 	       !TAILQ_EMPTY(&pnd->pnd_pcq))
2760 		DERRX(EX_SOFTWARE, "%s: opc = %p: queued operations",
2761 		      __func__, opc);
2762 
2763 	if (pnd->pnd_inxchg != 0)
2764 		DERRX(EX_SOFTWARE, "%s: opc = %p: ongoing operations",
2765 		      __func__, opc);
2766 #endif
2767 
2768 	/*
2769 	 * Send the FORGET message
2770 	 *
2771 	 * ps_new_msg() is called with NULL creds, which will
2772 	 * be interpreted as FUSE superuser. This is obviously
2773 	 * fine since we operate with kernel creds here.
2774 	 */
2775 	pm = ps->ps_new_msg(pu, opc, FUSE_FORGET,
2776 		      sizeof(*ffi), NULL);
2777 	ffi = GET_INPAYLOAD(ps, pm, fuse_forget_in);
2778 	ffi->nlookup = pnd->pnd_fuse_nlookup;
2779 
2780 	/*
2781 	 * No reply is expected, pm is freed in xchg_msg
2782 	 */
2783 	(void)xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, no_reply);
2784 
2785 	perfuse_destroy_pn(pu, opc);
2786 
2787 	return 0;
2788 }
2789 
2790 int
2791 perfuse_node_reclaim(struct puffs_usermount *pu, puffs_cookie_t opc)
2792 {
2793 #ifdef PERFUSE_DEBUG
2794 	if (perfuse_diagflags & PDF_RECLAIM)
2795 		DPRINTF("perfuse_node_reclaim called\n");
2796 #endif
2797 	return perfuse_node_reclaim2(pu, opc, 1);
2798 }
2799 
2800 int
2801 perfuse_node_inactive(struct puffs_usermount *pu, puffs_cookie_t opc)
2802 {
2803 	struct perfuse_node_data *pnd;
2804 	int error;
2805 
2806 	if (opc == 0)
2807 		return 0;
2808 
2809 	pnd = PERFUSE_NODE_DATA(opc);
2810 	if (!(pnd->pnd_flags & (PND_OPEN|PND_REMOVED)))
2811 		return 0;
2812 
2813 	node_ref(opc);
2814 
2815 	/*
2816 	 * Make sure all operation are finished
2817 	 * There can be an ongoing write. Other
2818 	 * operation wait for all data before
2819 	 * the close/inactive.
2820 	 */
2821 	while (pnd->pnd_flags & PND_INWRITE)
2822 		requeue_request(pu, opc, PCQ_AFTERWRITE);
2823 
2824 	/*
2825 	 * The inactive operation may be cancelled,
2826 	 * If no open is in progress, set PND_INOPEN
2827 	 * so that a new open will be queued.
2828 	 */
2829 	if (pnd->pnd_flags & PND_INOPEN)
2830 		goto out;
2831 
2832 	pnd->pnd_flags |= PND_INOPEN;
2833 
2834 	/*
2835 	 * Sync data
2836 	 */
2837 	if (pnd->pnd_flags & PND_DIRTY) {
2838 		if ((error = perfuse_node_fsync(pu, opc, NULL, 0, 0, 0)) != 0)
2839 			DWARN("%s: perfuse_node_fsync failed error = %d",
2840 			      __func__, error);
2841 	}
2842 
2843 
2844 	/*
2845 	 * Close handles
2846 	 */
2847 	if (pnd->pnd_flags & PND_WFH) {
2848 		if ((error = perfuse_node_close_common(pu, opc, FWRITE)) != 0)
2849 			DWARN("%s: close write FH failed error = %d",
2850 			      __func__, error);
2851 	}
2852 
2853 	if (pnd->pnd_flags & PND_RFH) {
2854 		if ((error = perfuse_node_close_common(pu, opc, FREAD)) != 0)
2855 			DWARN("%s: close read FH failed error = %d",
2856 			      __func__, error);
2857 	}
2858 
2859 	/*
2860 	 * This will cause a reclaim to be sent
2861 	 */
2862 	if (pnd->pnd_flags & PND_REMOVED)
2863 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N1);
2864 
2865 	/*
2866 	 * Schedule awaiting operations
2867 	 */
2868 	pnd->pnd_flags &= ~PND_INOPEN;
2869 	(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
2870 
2871 	/*
2872 	 * errors are ignored, since the kernel ignores the return code.
2873 	 */
2874 out:
2875 	node_rele(opc);
2876 	return 0;
2877 }
2878 
2879 
2880 /* ARGSUSED0 */
2881 int
2882 perfuse_node_print(struct puffs_usermount *pu, puffs_cookie_t opc)
2883 {
2884 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2885 	return 0;
2886 }
2887 
2888 int
2889 perfuse_node_pathconf(struct puffs_usermount *pu, puffs_cookie_t opc,
2890 	int name, register_t *retval)
2891 {
2892 	perfuse_msg_t *pm;
2893 	struct perfuse_state *ps;
2894 	struct fuse_statfs_out *fso;
2895 	int error = 0;
2896 
2897 	/*
2898 	 * Static values copied from UFS
2899 	 * in src/sys/ufs/ufs/ufs_vnops.c
2900 	 */
2901 	switch (name) {
2902 	case _PC_LINK_MAX:
2903 		*retval = LINK_MAX;
2904 		break;
2905 	case _PC_PATH_MAX:
2906 		*retval = PATH_MAX;
2907 		break;
2908 	case _PC_PIPE_BUF:
2909 		*retval = PIPE_BUF;
2910 		break;
2911 	case _PC_CHOWN_RESTRICTED:
2912 		*retval = 1;
2913 		break;
2914 	case _PC_NO_TRUNC:
2915 		*retval = 1;
2916 		break;
2917 	case _PC_SYNC_IO:
2918 		*retval = 1;
2919 		break;
2920 	case _PC_FILESIZEBITS:
2921 		*retval = 42;
2922 		break;
2923 	case _PC_SYMLINK_MAX:
2924 		*retval = MAXPATHLEN;
2925 		break;
2926 	case _PC_2_SYMLINKS:
2927 		*retval = 1;
2928 		break;
2929 	case _PC_NAME_MAX:
2930 		ps = puffs_getspecific(pu);
2931 		pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
2932 
2933 		error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply);
2934 		if (error != 0)
2935 			return error;
2936 
2937 		fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
2938 		*retval = fso->st.namelen;
2939 
2940 		ps->ps_destroy_msg(pm);
2941 
2942 		break;
2943 	default:
2944 		DWARN("Unimplemented pathconf for name = %d", name);
2945 		error = ENOSYS;
2946 		break;
2947 	}
2948 
2949 	return error;
2950 }
2951 
2952 int
2953 perfuse_node_advlock(struct puffs_usermount *pu, puffs_cookie_t opc,
2954 	void *id, int op, struct flock *fl, int flags)
2955 {
2956 	struct perfuse_state *ps;
2957 	int fop;
2958 	perfuse_msg_t *pm;
2959 	uint64_t fh;
2960 	struct fuse_lk_in *fli;
2961 	struct fuse_out_header *foh;
2962 	struct fuse_lk_out *flo;
2963 	uint32_t owner;
2964 	size_t len;
2965 	int error;
2966 
2967 	node_ref(opc);
2968 
2969 	/*
2970 	 * Make sure we do have a filehandle, as the FUSE filesystem
2971 	 * expect one. E.g.: if we provide none, GlusterFS logs an error
2972 	 * "0-glusterfs-fuse: xl is NULL"
2973 	 *
2974 	 * We need the read file handle if the file is open read only,
2975 	 * in order to support shared locks on read-only files.
2976 	 * NB: The kernel always sends advlock for read-only
2977 	 * files at exit time when the process used lock, see
2978 	 * sys_exit -> exit1 -> fd_free -> fd_close -> VOP_ADVLOCK
2979 	 */
2980 	if ((fh = perfuse_get_fh(opc, FREAD)) == FUSE_UNKNOWN_FH) {
2981 		error = EBADF;
2982 		goto out;
2983 	}
2984 
2985 	ps = puffs_getspecific(pu);
2986 
2987 	if (op == F_GETLK)
2988 		fop = FUSE_GETLK;
2989 	else
2990 		fop = (flags & F_WAIT) ? FUSE_SETLKW : FUSE_SETLK;
2991 
2992 	/*
2993 	 * XXX ps_new_msg() is called with NULL creds, which will
2994 	 * be interpreted as FUSE superuser. We have no way to
2995 	 * know the requesting process' credential, but since advlock()
2996 	 * is supposed to operate on a file that has been open(),
2997 	 * permission should have already been checked at open() time.
2998 	 */
2999 	pm = ps->ps_new_msg(pu, opc, fop, sizeof(*fli), NULL);
3000 	fli = GET_INPAYLOAD(ps, pm, fuse_lk_in);
3001 	fli->fh = fh;
3002 	fli->owner = (uint64_t)(vaddr_t)id;
3003 	fli->lk.start = fl->l_start;
3004 	fli->lk.end = fl->l_start + fl->l_len;
3005 	fli->lk.type = fl->l_type;
3006 	fli->lk.pid = fl->l_pid;
3007 	fli->lk_flags = (flags & F_FLOCK) ? FUSE_LK_FLOCK : 0;
3008 
3009 	owner = (uint32_t)(vaddr_t)id;
3010 
3011 #ifdef PERFUSE_DEBUG
3012 	if (perfuse_diagflags & PDF_FH)
3013 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
3014 			__func__, (void *)opc,
3015 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fli->fh);
3016 #endif
3017 
3018 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3019 		goto out;
3020 
3021 	foh = GET_OUTHDR(ps, pm);
3022 	len = foh->len - sizeof(*foh);
3023 
3024 	/*
3025 	 * Save or clear the lock
3026 	 */
3027 	switch (op) {
3028 	case F_GETLK:
3029 		if (len != sizeof(*flo))
3030 			DERRX(EX_SOFTWARE,
3031 			      "%s: Unexpected lock reply len %zd",
3032 			      __func__, len);
3033 
3034 		flo = GET_OUTPAYLOAD(ps, pm, fuse_lk_out);
3035 		fl->l_start = flo->lk.start;
3036 		fl->l_len = flo->lk.end - flo->lk.start;
3037 		fl->l_pid = flo->lk.pid;
3038 		fl->l_type = flo->lk.type;
3039 		fl->l_whence = SEEK_SET;	/* libfuse hardcodes it */
3040 
3041 		PERFUSE_NODE_DATA(opc)->pnd_lock_owner = flo->lk.pid;
3042 		break;
3043 	case F_UNLCK:
3044 		owner = 0;
3045 		/* FALLTHROUGH */
3046 	case F_SETLK:
3047 		/* FALLTHROUGH */
3048 	case F_SETLKW:
3049 		if (error != 0)
3050 			PERFUSE_NODE_DATA(opc)->pnd_lock_owner = owner;
3051 
3052 		if (len != 0)
3053 			DERRX(EX_SOFTWARE,
3054 			      "%s: Unexpected unlock reply len %zd",
3055 			      __func__, len);
3056 
3057 		break;
3058 	default:
3059 		DERRX(EX_SOFTWARE, "%s: Unexpected op %d", __func__, op);
3060 		break;
3061 	}
3062 
3063 	ps->ps_destroy_msg(pm);
3064 	error = 0;
3065 
3066 out:
3067 	node_rele(opc);
3068 	return error;
3069 }
3070 
3071 int
3072 perfuse_node_read(struct puffs_usermount *pu, puffs_cookie_t opc, uint8_t *buf,
3073 	off_t offset, size_t *resid, const struct puffs_cred *pcr, int ioflag)
3074 {
3075 	struct perfuse_state *ps;
3076 	struct perfuse_node_data *pnd;
3077 	const struct vattr *vap;
3078 	perfuse_msg_t *pm;
3079 	struct fuse_read_in *fri;
3080 	struct fuse_out_header *foh;
3081 	size_t readen;
3082 	int error;
3083 
3084 	ps = puffs_getspecific(pu);
3085 	pnd = PERFUSE_NODE_DATA(opc);
3086 	vap = puffs_pn_getvap((struct puffs_node *)opc);
3087 
3088 	/*
3089 	 * NetBSD turns that into a getdents(2) output
3090 	 * We just do a EISDIR as this feature is of little use.
3091 	 */
3092 	if (vap->va_type == VDIR)
3093 		return EISDIR;
3094 
3095 	do {
3096 		size_t max_read;
3097 
3098 		max_read = ps->ps_max_readahead - sizeof(*foh);
3099 		/*
3100 		 * flags may be set to FUSE_READ_LOCKOWNER
3101 		 * if lock_owner is provided.
3102 		 */
3103 		pm = ps->ps_new_msg(pu, opc, FUSE_READ, sizeof(*fri), pcr);
3104 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
3105 		fri->fh = perfuse_get_fh(opc, FREAD);
3106 		fri->offset = offset;
3107 		fri->size = (uint32_t)MIN(*resid, max_read);
3108 		fri->read_flags = 0; /* XXX Unused by libfuse? */
3109 		fri->lock_owner = pnd->pnd_lock_owner;
3110 		fri->flags = 0;
3111 		fri->flags |= (fri->lock_owner != 0) ? FUSE_READ_LOCKOWNER : 0;
3112 
3113 #ifdef PERFUSE_DEBUG
3114 	if (perfuse_diagflags & PDF_FH)
3115 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
3116 			__func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
3117 #endif
3118 		error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
3119 		if (error  != 0)
3120 			return error;
3121 
3122 		foh = GET_OUTHDR(ps, pm);
3123 		readen = foh->len - sizeof(*foh);
3124 
3125 #ifdef PERFUSE_DEBUG
3126 		if (readen > *resid)
3127 			DERRX(EX_SOFTWARE, "%s: Unexpected big read %zd",
3128 			      __func__, readen);
3129 #endif
3130 
3131 		(void)memcpy(buf,  _GET_OUTPAYLOAD(ps, pm, char *), readen);
3132 
3133 		buf += readen;
3134 		offset += readen;
3135 		*resid -= readen;
3136 
3137 		ps->ps_destroy_msg(pm);
3138 	} while ((*resid != 0) && (readen != 0));
3139 
3140 	if (ioflag & (IO_SYNC|IO_DSYNC))
3141 		ps->ps_syncreads++;
3142 	else
3143 		ps->ps_asyncreads++;
3144 
3145 	return 0;
3146 }
3147 
3148 int
3149 perfuse_node_write(struct puffs_usermount *pu, puffs_cookie_t opc,
3150 	uint8_t *buf, off_t offset, size_t *resid,
3151 	const struct puffs_cred *pcr, int ioflag)
3152 {
3153 	return perfuse_node_write2(pu, opc, buf, offset, resid, pcr, ioflag, 0);
3154 }
3155 
3156 /* ARGSUSED7 */
3157 int
3158 perfuse_node_write2(struct puffs_usermount *pu, puffs_cookie_t opc,
3159 	uint8_t *buf, off_t offset, size_t *resid,
3160 	const struct puffs_cred *pcr, int ioflag, int xflag)
3161 {
3162 	struct perfuse_state *ps;
3163 	struct perfuse_node_data *pnd;
3164 	struct vattr *vap;
3165 	perfuse_msg_t *pm;
3166 	struct fuse_write_in *fwi;
3167 	struct fuse_write_out *fwo;
3168 	size_t data_len;
3169 	size_t payload_len;
3170 	size_t written;
3171 	int inresize;
3172 	int error;
3173 
3174 	ps = puffs_getspecific(pu);
3175 	pnd = PERFUSE_NODE_DATA(opc);
3176 	vap = puffs_pn_getvap((struct puffs_node *)opc);
3177 	written = 0;
3178 	inresize = 0;
3179 	error = 0;
3180 
3181 	if (vap->va_type == VDIR)
3182 		return EISDIR;
3183 
3184 	node_ref(opc);
3185 
3186 	/*
3187 	 * We need to queue write requests in order to avoid
3188 	 * dequeueing PCQ_AFTERWRITE when there are pending writes.
3189 	 */
3190 	while (pnd->pnd_flags & PND_INWRITE)
3191 		requeue_request(pu, opc, PCQ_WRITE);
3192 	pnd->pnd_flags |= PND_INWRITE;
3193 
3194 	/*
3195 	 * append flag: re-read the file size so that
3196 	 * we get the latest value.
3197 	 */
3198 	if (ioflag & PUFFS_IO_APPEND) {
3199 		if ((error = perfuse_node_getattr(pu, opc, vap, pcr)) != 0)
3200 			goto out;
3201 
3202 		offset = vap->va_size;
3203 	}
3204 
3205 	/*
3206 	 * Serialize size access, see comment in perfuse_node_setattr().
3207 	 */
3208 	if ((u_quad_t)offset + *resid > vap->va_size) {
3209 		while (pnd->pnd_flags & PND_INRESIZE)
3210 			requeue_request(pu, opc, PCQ_RESIZE);
3211 		pnd->pnd_flags |= PND_INRESIZE;
3212 		inresize = 1;
3213 	}
3214 
3215 #ifdef PERFUSE_DEBUG
3216 	if (perfuse_diagflags & PDF_RESIZE)
3217 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__,
3218 			(void *)opc, vap->va_size);
3219 #endif
3220 
3221 	do {
3222 		size_t max_write;
3223 		/*
3224 		 * There is a writepage flag when data
3225 		 * is aligned to page size. Use it for
3226 		 * everything but the data after the last
3227 		 * page boundary.
3228 		 */
3229 		max_write = ps->ps_max_write - sizeof(*fwi);
3230 
3231 		data_len = MIN(*resid, max_write);
3232 		if (data_len > (size_t)sysconf(_SC_PAGESIZE))
3233 			data_len = data_len & ~(sysconf(_SC_PAGESIZE) - 1);
3234 
3235 		payload_len = data_len + sizeof(*fwi);
3236 
3237 		/*
3238 		 * flags may be set to FUSE_WRITE_CACHE (XXX usage?)
3239 		 * or FUSE_WRITE_LOCKOWNER, if lock_owner is provided.
3240 		 * write_flags is set to 1 for writepage.
3241 		 */
3242 		pm = ps->ps_new_msg(pu, opc, FUSE_WRITE, payload_len, pcr);
3243 		fwi = GET_INPAYLOAD(ps, pm, fuse_write_in);
3244 		fwi->fh = perfuse_get_fh(opc, FWRITE);
3245 		fwi->offset = offset;
3246 		fwi->size = (uint32_t)data_len;
3247 		fwi->write_flags = (fwi->size % sysconf(_SC_PAGESIZE)) ? 0 : 1;
3248 		fwi->lock_owner = pnd->pnd_lock_owner;
3249 		fwi->flags = 0;
3250 		fwi->flags |= (fwi->lock_owner != 0) ? FUSE_WRITE_LOCKOWNER : 0;
3251 		fwi->flags |= (ioflag & IO_DIRECT) ? 0 : FUSE_WRITE_CACHE;
3252 		(void)memcpy((fwi + 1), buf, data_len);
3253 
3254 
3255 #ifdef PERFUSE_DEBUG
3256 		if (perfuse_diagflags & PDF_FH)
3257 			DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
3258 				"fh = 0x%"PRIx64"\n", __func__,
3259 				(void *)opc, pnd->pnd_nodeid, fwi->fh);
3260 #endif
3261 		if ((error = xchg_msg(pu, opc, pm,
3262 				      sizeof(*fwo), wait_reply)) != 0)
3263 			goto out;
3264 
3265 		fwo = GET_OUTPAYLOAD(ps, pm, fuse_write_out);
3266 		written = fwo->size;
3267 		ps->ps_destroy_msg(pm);
3268 
3269 #ifdef PERFUSE_DEBUG
3270 		if (written > *resid)
3271 			DERRX(EX_SOFTWARE, "%s: Unexpected big write %zd",
3272 			      __func__, written);
3273 #endif
3274 		*resid -= written;
3275 		offset += written;
3276 		buf += written;
3277 
3278 	} while (*resid != 0);
3279 
3280 	/*
3281 	 * puffs_ops(3) says
3282 	 *  "everything must be written or an error will be generated"
3283 	 */
3284 	if (*resid != 0)
3285 		error = EFBIG;
3286 
3287 out:
3288 #ifdef PERFUSE_DEBUG
3289 	if (perfuse_diagflags & PDF_RESIZE) {
3290 		if (offset > (off_t)vap->va_size)
3291 			DPRINTF("<< %s %p %" PRIu64 " -> %lld\n", __func__,
3292 				(void *)opc, vap->va_size, (long long)offset);
3293 		else
3294 			DPRINTF("<< %s %p \n", __func__, (void *)opc);
3295 	}
3296 #endif
3297 
3298 	/*
3299 	 * Update file size if we wrote beyond the end
3300 	 */
3301 	if (offset > (off_t)vap->va_size)
3302 		vap->va_size = offset;
3303 
3304 	/*
3305 	 * Statistics
3306 	 */
3307 	if (ioflag & (IO_SYNC|IO_DSYNC))
3308 		ps->ps_syncwrites++;
3309 	else
3310 		ps->ps_asyncwrites++;
3311 
3312 	/*
3313 	 * Remember to sync the file
3314 	 */
3315 	pnd->pnd_flags |= PND_DIRTY;
3316 
3317 #ifdef PERFUSE_DEBUG
3318 	if (perfuse_diagflags & PDF_SYNC)
3319 		DPRINTF("%s: DIRTY opc = %p, file = \"%s\"\n",
3320 			__func__, (void*)opc, perfuse_node_path(ps, opc));
3321 #endif
3322 
3323 	if (inresize) {
3324 #ifdef PERFUSE_DEBUG
3325 		if (!(pnd->pnd_flags & PND_INRESIZE))
3326 			DERRX(EX_SOFTWARE, "file write grow without resize");
3327 #endif
3328 		pnd->pnd_flags &= ~PND_INRESIZE;
3329 		(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
3330 	}
3331 
3332 	/*
3333 	 * VOP_PUTPAGE causes FAF write where kernel does not
3334 	 * check operation result. At least warn if it failed.
3335 	 */
3336 #ifdef PUFFS_WRITE_FAF
3337 	if (error && (xflag & PUFFS_WRITE_FAF))
3338 		DWARN("Data loss caused by FAF write failed on \"%s\"",
3339 		      pnd->pnd_name);
3340 #endif /* PUFFS_WRITE_FAF */
3341 
3342 	/*
3343 	 * If there are no more queued write, we can resume
3344 	 * an operation awaiting write completion.
3345 	 */
3346 	pnd->pnd_flags &= ~PND_INWRITE;
3347 	if (dequeue_requests(opc, PCQ_WRITE, 1) == 0)
3348 		(void)dequeue_requests(opc, PCQ_AFTERWRITE, DEQUEUE_ALL);
3349 
3350 	node_rele(opc);
3351 	return error;
3352 }
3353 
3354 /* ARGSUSED0 */
3355 void
3356 perfuse_cache_write(struct puffs_usermount *pu, puffs_cookie_t opc, size_t size,
3357 	struct puffs_cacherun *runs)
3358 {
3359 	return;
3360 }
3361 
3362 /* ARGSUSED4 */
3363 int
3364 perfuse_node_getextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3365 	int attrns, const char *attrname, size_t *attrsize, uint8_t *attr,
3366 	size_t *resid, const struct puffs_cred *pcr)
3367 {
3368 	struct perfuse_state *ps;
3369 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3370 	perfuse_msg_t *pm;
3371 	struct fuse_getxattr_in *fgi;
3372 	struct fuse_getxattr_out *fgo;
3373 	struct fuse_out_header *foh;
3374 	size_t attrnamelen;
3375 	size_t len;
3376 	char *np;
3377 	int error;
3378 
3379 	/* system namespace attrs are not accessible to non root users */
3380 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
3381 		return EPERM;
3382 
3383 	node_ref(opc);
3384 	ps = puffs_getspecific(pu);
3385 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3386 	attrnamelen = strlen(attrname) + 1;
3387 	len = sizeof(*fgi) + attrnamelen;
3388 
3389 	pm = ps->ps_new_msg(pu, opc, FUSE_GETXATTR, len, pcr);
3390 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3391 	fgi->size = (unsigned int)((resid != NULL) ? *resid : 0);
3392 	np = (char *)(void *)(fgi + 1);
3393 	(void)strlcpy(np, attrname, attrnamelen);
3394 
3395 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3396 		goto out;
3397 
3398 	/*
3399 	 * We just get fuse_getattr_out with list size if we requested
3400 	 * a null size.
3401 	 */
3402 	if (resid == NULL) {
3403 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3404 
3405 		if (attrsize != NULL)
3406 			*attrsize = fgo->size;
3407 
3408 		ps->ps_destroy_msg(pm);
3409 		error = 0;
3410 		goto out;
3411 	}
3412 
3413 	/*
3414 	 * And with a non null requested size, we get the list just
3415 	 * after the header
3416 	 */
3417 	foh = GET_OUTHDR(ps, pm);
3418 	np = (char *)(void *)(foh + 1);
3419 	len = foh->len - sizeof(*foh);
3420 
3421 	if (attrsize != NULL)
3422 		*attrsize = len;
3423 
3424 	if (resid != NULL) {
3425 		if (*resid < len) {
3426 			error = ERANGE;
3427 			ps->ps_destroy_msg(pm);
3428 			goto out;
3429 		}
3430 
3431 		(void)memcpy(attr, np, len);
3432 		*resid -= len;
3433 	}
3434 
3435 	ps->ps_destroy_msg(pm);
3436 	error = 0;
3437 
3438 out:
3439 	node_rele(opc);
3440 	return error;
3441 }
3442 
3443 int
3444 perfuse_node_setextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3445 	int attrns, const char *attrname, uint8_t *attr, size_t *resid,
3446 	const struct puffs_cred *pcr)
3447 {
3448 	struct perfuse_state *ps;
3449 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3450 	perfuse_msg_t *pm;
3451 	struct fuse_setxattr_in *fsi;
3452 	size_t attrnamelen;
3453 	size_t datalen;
3454 	size_t len;
3455 	char *np;
3456 	int error;
3457 
3458 	/* system namespace attrs are not accessible to non root users */
3459 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
3460 		return EPERM;
3461 
3462 	node_ref(opc);
3463 	ps = puffs_getspecific(pu);
3464 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3465 	attrnamelen = strlen(attrname) + 1;
3466 
3467 	datalen = (resid != NULL) ? *resid : 0;
3468 	len = sizeof(*fsi) + attrnamelen + datalen;
3469 
3470 	pm = ps->ps_new_msg(pu, opc, FUSE_SETXATTR, len, pcr);
3471 	fsi = GET_INPAYLOAD(ps, pm, fuse_setxattr_in);
3472 	fsi->size = (unsigned int)datalen;
3473 	fsi->flags = 0;
3474 	np = (char *)(void *)(fsi + 1);
3475 	(void)strlcpy(np, attrname, attrnamelen);
3476 	np += attrnamelen;
3477 	if (datalen)
3478 		(void)memcpy(np, (char *)attr, datalen);
3479 
3480 	if ((error = xchg_msg(pu, opc, pm,
3481 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
3482 		goto out;
3483 
3484 	ps->ps_destroy_msg(pm);
3485 	if (resid)
3486 		*resid = 0;
3487 	error = 0;
3488 
3489 out:
3490 	node_rele(opc);
3491 	return error;
3492 }
3493 
3494 /* ARGSUSED2 */
3495 int
3496 perfuse_node_listextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3497 	int attrns, size_t *attrsize, uint8_t *attrs, size_t *resid, int flag,
3498 	const struct puffs_cred *pcr)
3499 {
3500 	struct perfuse_state *ps;
3501 	perfuse_msg_t *pm;
3502 	struct fuse_getxattr_in *fgi;
3503 	struct fuse_getxattr_out *fgo;
3504 	struct fuse_out_header *foh;
3505 	char *np;
3506 	size_t len, puffs_len, i, attrlen, outlen;
3507 	int error;
3508 
3509 	/* system namespace attrs are not accessible to non root users */
3510 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
3511 		return EPERM;
3512 
3513 	node_ref(opc);
3514 
3515 	ps = puffs_getspecific(pu);
3516 	len = sizeof(*fgi);
3517 
3518 	pm = ps->ps_new_msg(pu, opc, FUSE_LISTXATTR, len, pcr);
3519 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3520 	if (resid != NULL)
3521 		fgi->size = (unsigned int)*resid;
3522 	else
3523 		fgi->size = 0;
3524 
3525 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3526 		goto out;
3527 
3528 	/*
3529 	 * We just get fuse_getattr_out with list size if we requested
3530 	 * a null size.
3531 	 */
3532 	if (resid == NULL) {
3533 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3534 
3535 		if (attrsize != NULL)
3536 			*attrsize = fgo->size;
3537 
3538 		ps->ps_destroy_msg(pm);
3539 
3540 		error = 0;
3541 		goto out;
3542 	}
3543 
3544 	/*
3545 	 * And with a non null requested size, we get the list just
3546 	 * after the header
3547 	 */
3548 	foh = GET_OUTHDR(ps, pm);
3549 	np = (char *)(void *)(foh + 1);
3550 	puffs_len = foh->len - sizeof(*foh);
3551 
3552 	if (attrsize != NULL)
3553 		*attrsize = puffs_len;
3554 
3555 	if (attrs != NULL) {
3556 		if (*resid < puffs_len) {
3557 			error = ERANGE;
3558 			ps->ps_destroy_msg(pm);
3559 			goto out;
3560 		}
3561 
3562 		outlen = 0;
3563 
3564 		for (i = 0; i < puffs_len; i += attrlen + 1) {
3565 			attrlen = strlen(np + i);
3566 
3567 			/*
3568 			 * Filter attributes per namespace
3569 			 */
3570 			if (!perfuse_ns_match(attrns, np + i))
3571 				continue;
3572 
3573 #ifdef PUFFS_EXTATTR_LIST_LENPREFIX
3574 			/*
3575 			 * Convert the FUSE reply to length prefixed strings
3576 			 * if this is what the kernel wants.
3577 			 */
3578 			if (flag & PUFFS_EXTATTR_LIST_LENPREFIX) {
3579 				(void)memcpy(attrs + outlen + 1,
3580 					     np + i, attrlen);
3581 				*(attrs + outlen) = (uint8_t)attrlen;
3582 			} else
3583 #endif /* PUFFS_EXTATTR_LIST_LENPREFIX */
3584 			(void)memcpy(attrs + outlen, np + i, attrlen + 1);
3585 			outlen += attrlen + 1;
3586 		}
3587 
3588 		*resid -= outlen;
3589 	}
3590 
3591 	ps->ps_destroy_msg(pm);
3592 	error = 0;
3593 
3594 out:
3595 	node_rele(opc);
3596 	return error;
3597 }
3598 
3599 int
3600 perfuse_node_deleteextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3601 	int attrns, const char *attrname, const struct puffs_cred *pcr)
3602 {
3603 	struct perfuse_state *ps;
3604 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3605 	perfuse_msg_t *pm;
3606 	size_t attrnamelen;
3607 	char *np;
3608 	int error;
3609 
3610 	/* system namespace attrs are not accessible to non root users */
3611 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
3612 		return EPERM;
3613 
3614 	node_ref(opc);
3615 
3616 	ps = puffs_getspecific(pu);
3617 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3618 	attrnamelen = strlen(attrname) + 1;
3619 
3620 	pm = ps->ps_new_msg(pu, opc, FUSE_REMOVEXATTR, attrnamelen, pcr);
3621 	np = _GET_INPAYLOAD(ps, pm, char *);
3622 	(void)strlcpy(np, attrname, attrnamelen);
3623 
3624 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
3625 	if (error != 0)
3626 		goto out;
3627 
3628 	ps->ps_destroy_msg(pm);
3629 
3630 out:
3631 	node_rele(opc);
3632 	return error;
3633 }
3634 
3635 int
3636 perfuse_node_fallocate(struct puffs_usermount *pu, puffs_cookie_t opc,
3637 	off_t off, off_t len)
3638 {
3639 	struct perfuse_state *ps;
3640 	perfuse_msg_t *pm;
3641 	struct fuse_fallocate_in *fai;
3642 	int error;
3643 
3644 	ps = puffs_getspecific(pu);
3645 	if (ps->ps_flags & PS_NO_FALLOCATE)
3646 		return EOPNOTSUPP;
3647 
3648 	node_ref(opc);
3649 
3650 	pm = ps->ps_new_msg(pu, opc, FUSE_FALLOCATE, sizeof(*fai), NULL);
3651 
3652 	fai = GET_INPAYLOAD(ps, pm, fuse_fallocate_in);
3653 	fai->fh = perfuse_get_fh(opc, FWRITE);
3654 	fai->offset = off;
3655 	fai->length = len;
3656 	fai->mode = 0;
3657 
3658 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
3659 	if (error == EOPNOTSUPP || error == ENOSYS) {
3660 		ps->ps_flags |= PS_NO_FALLOCATE;
3661 		error = EOPNOTSUPP;
3662 	}
3663 	if (error != 0)
3664 		goto out;
3665 
3666 	ps->ps_destroy_msg(pm);
3667 
3668 out:
3669 	node_rele(opc);
3670 	return error;
3671 }
3672