xref: /netbsd-src/lib/libperfuse/ops.c (revision 946379e7b37692fc43f68eb0d1c10daa0a7f3b6c)
1 /*  $NetBSD: ops.c,v 1.84 2015/06/03 14:07:05 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 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_NODELEAK) {
504 		PERFUSE_NODE_DATA(pn)->pnd_flags &= ~PND_NODELEAK;
505 		ps->ps_nodeleakcount--;
506 	}
507 
508 	ps->ps_destroy_msg(pm);
509 
510 	return 0;
511 }
512 
513 
514 /*
515  * Common code for methods that create objects:
516  * perfuse_node_mkdir
517  * perfuse_node_mknod
518  * perfuse_node_symlink
519  */
520 static int
521 node_mk_common(struct puffs_usermount *pu, puffs_cookie_t opc,
522 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
523 	perfuse_msg_t *pm)
524 {
525 	struct perfuse_state *ps;
526 	struct puffs_node *pn;
527 	struct fuse_entry_out *feo;
528 	int error;
529 
530 	ps =  puffs_getspecific(pu);
531 
532 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
533 		return error;
534 
535 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
536 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
537 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
538 
539 	pn = perfuse_new_pn(pu, pcn->pcn_name, opc);
540 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
541 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
542 	perfuse_node_cache(ps, pn);
543 
544 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
545 	pn->pn_va.va_gen = (u_long)(feo->generation);
546 
547 	puffs_newinfo_setcookie(pni, pn);
548 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
549 	puffs_newinfo_setva(pni, &pn->pn_va);
550 	perfuse_newinfo_setttl(pni, pn, feo, NULL);
551 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
552 
553 
554 #ifdef PERFUSE_DEBUG
555 	if (perfuse_diagflags & PDF_FILENAME)
556 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
557 			"nodeid = 0x%"PRIx64"\n",
558 			__func__, (void *)pn, pcn->pcn_name,
559 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid);
560 #endif
561 	ps->ps_destroy_msg(pm);
562 
563 	/* Parents is now dirty */
564 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
565 
566 	return 0;
567 }
568 
569 static uint64_t
570 readdir_last_cookie(struct fuse_dirent *fd, size_t fd_len)
571 {
572 	size_t len;
573 	size_t seen = 0;
574 	char *ndp;
575 
576 	do {
577 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
578 		seen += len;
579 
580 		if (seen >= fd_len)
581 			break;
582 
583 		ndp = (char *)(void *)fd + (size_t)len;
584 		fd = (struct fuse_dirent *)(void *)ndp;
585 	} while (1 /* CONSTCOND */);
586 
587 	return fd->off;
588 }
589 
590 static ssize_t
591 fuse_to_dirent(struct puffs_usermount *pu, puffs_cookie_t opc,
592 	struct fuse_dirent *fd, size_t fd_len)
593 {
594 	struct dirent *dents;
595 	size_t dents_len;
596 	ssize_t written;
597 	uint64_t fd_offset;
598 	struct fuse_dirent *fd_base;
599 	size_t len;
600 
601 	fd_base = fd;
602 	fd_offset = 0;
603 	written = 0;
604 	dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
605 	dents_len = (size_t)PERFUSE_NODE_DATA(opc)->pnd_dirent_len;
606 
607 	do {
608 		char *ndp;
609 		size_t reclen;
610 		char name[MAXPATHLEN];
611 
612 		reclen = _DIRENT_RECLEN(dents, fd->namelen);
613 
614 		/*
615 		 * Check we do not overflow the output buffer
616 		 * struct fuse_dirent is bigger than struct dirent,
617 		 * so we should always use fd_len and never reallocate
618 		 * later.
619 		 * If we have to reallocate,try to double the buffer
620 		 * each time so that we do not have to do it too often.
621 		 */
622 		if (written + reclen > dents_len) {
623 			if (dents_len == 0)
624 				dents_len = fd_len;
625 			else
626 				dents_len =
627 				   MAX(2 * dents_len, written + reclen);
628 
629 			dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
630 			if ((dents = realloc(dents, dents_len)) == NULL)
631 				DERR(EX_OSERR, "%s: malloc failed", __func__);
632 
633 			PERFUSE_NODE_DATA(opc)->pnd_dirent = dents;
634 			PERFUSE_NODE_DATA(opc)->pnd_dirent_len = dents_len;
635 
636 			/*
637 			 * (void *) for delint
638 			 */
639 			ndp = (char *)(void *)dents + written;
640 			dents = (struct dirent *)(void *)ndp;
641 		}
642 
643 		strncpy(name, fd->name, fd->namelen);
644 		name[fd->namelen] = '\0';
645 
646 		/*
647 		 * Filesystem was mounted without -o use_ino
648 		 * Perform a lookup to find it.
649 		 */
650 		if (fd->ino == PERFUSE_UNKNOWN_INO) {
651 			struct puffs_node *pn;
652 			struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
653 
654 			if (strcmp(name, "..") == 0) {
655 				/*
656 				 * Avoid breaking out of fs
657 				 * by lookup to .. on root
658 				 */
659 				if (pnd->pnd_nodeid == FUSE_ROOT_ID)
660 					fd->ino = FUSE_ROOT_ID;
661 				else
662 					fd->ino = pnd->pnd_parent_nodeid;
663 			} else if (strcmp(name, ".") == 0 ) {
664 				fd->ino = pnd->pnd_nodeid;
665 			} else {
666 				int error;
667 
668 				error = node_lookup_common(pu, opc, NULL,
669 							   name, NULL, &pn);
670 				if (error != 0) {
671 					DWARNX("node_lookup_common %s "
672 					       "failed: %d", name, error);
673 				} else {
674 					fd->ino = pn->pn_va.va_fileid;
675 					(void)perfuse_node_reclaim(pu, pn);
676 				}
677 			}
678 		}
679 
680 		dents->d_fileno = fd->ino;
681 		dents->d_reclen = (unsigned short)reclen;
682 		dents->d_namlen = fd->namelen;
683 		dents->d_type = fd->type;
684 		strlcpy(dents->d_name, name, fd->namelen + 1);
685 
686 #ifdef PERFUSE_DEBUG
687 		if (perfuse_diagflags & PDF_READDIR)
688 			DPRINTF("%s: translated \"%s\" ino = %"PRIu64"\n",
689 				__func__, dents->d_name, dents->d_fileno);
690 #endif
691 
692 		dents = _DIRENT_NEXT(dents);
693 		written += reclen;
694 
695 		/*
696 		 * Move to the next record.
697 		 * fd->off is not the offset, it is an opaque cookie
698 		 * given by the filesystem to keep state across multiple
699 		 * readdir() operation.
700 		 * Use record alignement instead.
701 		 */
702 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
703 #ifdef PERFUSE_DEBUG
704 		if (perfuse_diagflags & PDF_READDIR)
705 			DPRINTF("%s: record at %"PRId64"/0x%"PRIx64" "
706 				"length = %zd/0x%zx. "
707 				"next record at %"PRId64"/0x%"PRIx64" "
708 				"max %zd/0x%zx\n",
709 				__func__, fd_offset, fd_offset, len, len,
710 				fd_offset + len, fd_offset + len,
711 				fd_len, fd_len);
712 #endif
713 		fd_offset += len;
714 
715 		/*
716 		 * Check if next record is still within the packet
717 		 * If it is not, we reached the end of the buffer.
718 		 */
719 		if (fd_offset >= fd_len)
720 			break;
721 
722 		/*
723 		 * (void *) for delint
724 		 */
725 		ndp = (char *)(void *)fd_base + (size_t)fd_offset;
726 		fd = (struct fuse_dirent *)(void *)ndp;
727 
728 	} while (1 /* CONSTCOND */);
729 
730 	/*
731 	 * Adjust the dirent output length
732 	 */
733 	if (written != -1)
734 		PERFUSE_NODE_DATA(opc)->pnd_dirent_len = written;
735 
736 	return written;
737 }
738 
739 static void
740 readdir_buffered(puffs_cookie_t opc, struct dirent *dent, off_t *readoff,
741 	size_t *reslen)
742 {
743 	struct dirent *fromdent;
744 	struct perfuse_node_data *pnd;
745 	char *ndp;
746 
747 	pnd = PERFUSE_NODE_DATA(opc);
748 
749 	while (*readoff < pnd->pnd_dirent_len) {
750 		/*
751 		 * (void *) for delint
752 		 */
753 		ndp = (char *)(void *)pnd->pnd_dirent + (size_t)*readoff;
754 		fromdent = (struct dirent *)(void *)ndp;
755 
756 		if (*reslen < _DIRENT_SIZE(fromdent))
757 			break;
758 
759 		memcpy(dent, fromdent, _DIRENT_SIZE(fromdent));
760 		*readoff += _DIRENT_SIZE(fromdent);
761 		*reslen -= _DIRENT_SIZE(fromdent);
762 
763 		dent = _DIRENT_NEXT(dent);
764 	}
765 
766 #ifdef PERFUSE_DEBUG
767 	if (perfuse_diagflags & PDF_READDIR)
768 		DPRINTF("%s: readoff = %"PRId64",  "
769 			"pnd->pnd_dirent_len = %"PRId64"\n",
770 			__func__, *readoff, pnd->pnd_dirent_len);
771 #endif
772 	if (*readoff >=  pnd->pnd_dirent_len) {
773 		free(pnd->pnd_dirent);
774 		pnd->pnd_dirent = NULL;
775 		pnd->pnd_dirent_len = 0;
776 	}
777 
778 	return;
779 }
780 
781 
782 static void
783 node_ref(puffs_cookie_t opc)
784 {
785 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
786 
787 #ifdef PERFUSE_DEBUG
788 	if (pnd->pnd_flags & PND_INVALID)
789 		DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
790 #endif /* PERFUSE_DEBUG */
791 
792 	pnd->pnd_ref++;
793 	return;
794 }
795 
796 static void
797 node_rele(puffs_cookie_t opc)
798 {
799 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
800 
801 #ifdef PERFUSE_DEBUG
802 	if (pnd->pnd_flags & PND_INVALID)
803 		DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
804 #endif /* PERFUSE_DEBUG */
805 
806 	pnd->pnd_ref--;
807 
808 	if (pnd->pnd_ref == 0)
809 		(void)dequeue_requests(opc, PCQ_REF, DEQUEUE_ALL);
810 
811 	return;
812 }
813 
814 static void
815 requeue_request(struct puffs_usermount *pu, puffs_cookie_t opc,
816 	enum perfuse_qtype type)
817 {
818 	struct perfuse_cc_queue pcq;
819 	struct perfuse_node_data *pnd;
820 
821 	pnd = PERFUSE_NODE_DATA(opc);
822 	pcq.pcq_type = type;
823 	pcq.pcq_cc = puffs_cc_getcc(pu);
824 	TAILQ_INSERT_TAIL(&pnd->pnd_pcq, &pcq, pcq_next);
825 
826 #ifdef PERFUSE_DEBUG
827 	if (perfuse_diagflags & PDF_REQUEUE)
828 		DPRINTF("%s: REQUEUE opc = %p, pcc = %p (%s)\n",
829 		        __func__, (void *)opc, pcq.pcq_cc,
830 			perfuse_qtypestr[type]);
831 #endif
832 
833 	puffs_cc_yield(pcq.pcq_cc);
834 	TAILQ_REMOVE(&pnd->pnd_pcq, &pcq, pcq_next);
835 
836 #ifdef PERFUSE_DEBUG
837 	if (perfuse_diagflags & PDF_REQUEUE)
838 		DPRINTF("%s: RESUME opc = %p, pcc = %p (%s)\n",
839 		        __func__, (void *)opc, pcq.pcq_cc,
840 			perfuse_qtypestr[type]);
841 #endif
842 
843 	return;
844 }
845 
846 static int
847 dequeue_requests(puffs_cookie_t opc, enum perfuse_qtype type, int max)
848 {
849 	struct perfuse_cc_queue *pcq;
850 	struct perfuse_node_data *pnd;
851 	int dequeued;
852 
853 	pnd = PERFUSE_NODE_DATA(opc);
854 	dequeued = 0;
855 	TAILQ_FOREACH(pcq, &pnd->pnd_pcq, pcq_next) {
856 		if (pcq->pcq_type != type)
857 			continue;
858 
859 #ifdef PERFUSE_DEBUG
860 		if (perfuse_diagflags & PDF_REQUEUE)
861 			DPRINTF("%s: SCHEDULE opc = %p, pcc = %p (%s)\n",
862 				__func__, (void *)opc, pcq->pcq_cc,
863 				 perfuse_qtypestr[type]);
864 #endif
865 		puffs_cc_schedule(pcq->pcq_cc);
866 
867 		if (++dequeued == max)
868 			break;
869 	}
870 
871 #ifdef PERFUSE_DEBUG
872 	if (perfuse_diagflags & PDF_REQUEUE)
873 		DPRINTF("%s: DONE  opc = %p\n", __func__, (void *)opc);
874 #endif
875 
876 	return dequeued;
877 }
878 
879 void
880 perfuse_fs_init(struct puffs_usermount *pu)
881 {
882 	struct perfuse_state *ps;
883 	perfuse_msg_t *pm;
884 	struct fuse_init_in *fii;
885 	struct fuse_init_out *fio;
886 	int error;
887 
888 	ps = puffs_getspecific(pu);
889 
890         if (puffs_mount(pu, ps->ps_target, ps->ps_mountflags, ps->ps_root) != 0)
891                 DERR(EX_OSERR, "%s: puffs_mount failed", __func__);
892 
893 	/*
894 	 * Linux 2.6.34.1 sends theses flags:
895 	 * FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC
896 	 * FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK
897 	 *
898 	 * Linux also sets max_readahead at 32 pages (128 kB)
899 	 *
900 	 * ps_new_msg() is called with NULL creds, which will
901 	 * be interpreted as FUSE superuser.
902 	 */
903 	pm = ps->ps_new_msg(pu, 0, FUSE_INIT, sizeof(*fii), NULL);
904 	fii = GET_INPAYLOAD(ps, pm, fuse_init_in);
905 	fii->major = FUSE_KERNEL_VERSION;
906 	fii->minor = FUSE_KERNEL_MINOR_VERSION;
907 	fii->max_readahead = (unsigned int)(32 * sysconf(_SC_PAGESIZE));
908 	fii->flags = (FUSE_ASYNC_READ|FUSE_POSIX_LOCKS|FUSE_ATOMIC_O_TRUNC);
909 
910 	if ((error = xchg_msg(pu, 0, pm, sizeof(*fio), wait_reply)) != 0)
911 		DERRX(EX_SOFTWARE, "init message exchange failed (%d)", error);
912 
913 	fio = GET_OUTPAYLOAD(ps, pm, fuse_init_out);
914 	ps->ps_max_readahead = fio->max_readahead;
915 	ps->ps_max_write = fio->max_write;
916 
917 	ps->ps_destroy_msg(pm);
918 
919 	return;
920 }
921 
922 int
923 perfuse_fs_unmount(struct puffs_usermount *pu, int flags)
924 {
925 	perfuse_msg_t *pm;
926 	struct perfuse_state *ps;
927 	puffs_cookie_t opc;
928 	int error;
929 
930 	ps = puffs_getspecific(pu);
931 	opc = (puffs_cookie_t)puffs_getroot(pu);
932 
933 	/*
934 	 * ps_new_msg() is called with NULL creds, which will
935 	 * be interpreted as FUSE superuser.
936 	 */
937 	pm = ps->ps_new_msg(pu, opc, FUSE_DESTROY, 0, NULL);
938 
939 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0){
940 		DWARN("unmount %s", ps->ps_target);
941 		if (!(flags & MNT_FORCE))
942 			return error;
943 		else
944 			error = 0;
945 	} else {
946 		ps->ps_destroy_msg(pm);
947 	}
948 
949 	ps->ps_umount(pu);
950 
951 	if (perfuse_diagflags & PDF_MISC)
952 		DPRINTF("%s unmounted, exit\n", ps->ps_target);
953 
954 	return 0;
955 }
956 
957 int
958 perfuse_fs_statvfs(struct puffs_usermount *pu, struct statvfs *svfsb)
959 {
960 	struct perfuse_state *ps;
961 	perfuse_msg_t *pm;
962 	puffs_cookie_t opc;
963 	struct fuse_statfs_out *fso;
964 	int error;
965 
966 	ps = puffs_getspecific(pu);
967 	opc = (puffs_cookie_t)puffs_getroot(pu);
968 
969 	/*
970 	 * ps_new_msg() is called with NULL creds, which will
971 	 * be interpreted as FUSE superuser.
972 	 */
973 	pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
974 
975 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply)) != 0)
976 		return error;
977 
978 	fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
979 	svfsb->f_flag = ps->ps_mountflags;
980 	svfsb->f_bsize = fso->st.bsize;
981 	svfsb->f_frsize = fso->st.frsize;
982 	svfsb->f_iosize = ((struct puffs_node *)opc)->pn_va.va_blocksize;
983 	svfsb->f_blocks = fso->st.blocks;
984 	svfsb->f_bfree = fso->st.bfree;
985 	svfsb->f_bavail = fso->st.bavail;
986 	svfsb->f_bresvd = fso->st.bfree - fso->st.bavail;
987 	svfsb->f_files = fso->st.files;
988 	svfsb->f_ffree = fso->st.ffree;
989 	svfsb->f_favail = fso->st.ffree;/* files not reserved for root */
990 	svfsb->f_fresvd = 0;		/* files reserved for root */
991 
992 	svfsb->f_syncreads = ps->ps_syncreads;
993 	svfsb->f_syncwrites = ps->ps_syncwrites;
994 
995 	svfsb->f_asyncreads = ps->ps_asyncreads;
996 	svfsb->f_asyncwrites = ps->ps_asyncwrites;
997 
998 	(void)memcpy(&svfsb->f_fsidx, &ps->ps_fsid, sizeof(ps->ps_fsid));
999 	svfsb->f_fsid = (unsigned long)ps->ps_fsid;
1000 	svfsb->f_namemax = MAXPATHLEN;	/* XXX */
1001 	svfsb->f_owner = ps->ps_owner_uid;
1002 
1003 	(void)strlcpy(svfsb->f_mntonname, ps->ps_target, _VFS_NAMELEN);
1004 
1005 	if (ps->ps_filesystemtype != NULL)
1006 		(void)strlcpy(svfsb->f_fstypename,
1007 			      ps->ps_filesystemtype, _VFS_NAMELEN);
1008 	else
1009 		(void)strlcpy(svfsb->f_fstypename, "fuse", _VFS_NAMELEN);
1010 
1011 	if (ps->ps_source != NULL)
1012 		strlcpy(svfsb->f_mntfromname, ps->ps_source, _VFS_NAMELEN);
1013 	else
1014 		strlcpy(svfsb->f_mntfromname, _PATH_FUSE, _VFS_NAMELEN);
1015 
1016 	ps->ps_destroy_msg(pm);
1017 
1018 	return 0;
1019 }
1020 
1021 int
1022 perfuse_fs_sync(struct puffs_usermount *pu, int waitfor,
1023 	const struct puffs_cred *pcr)
1024 {
1025 	/*
1026 	 * FUSE does not seem to have a FS sync callback.
1027 	 * Maybe do not even register this callback
1028 	 */
1029 	return puffs_fsnop_sync(pu, waitfor, pcr);
1030 }
1031 
1032 /* ARGSUSED0 */
1033 int
1034 perfuse_fs_fhtonode(struct puffs_usermount *pu, void *fid, size_t fidsize,
1035 	struct puffs_newinfo *pni)
1036 {
1037 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1038 	return 0;
1039 }
1040 
1041 /* ARGSUSED0 */
1042 int
1043 perfuse_fs_nodetofh(struct puffs_usermount *pu, puffs_cookie_t cookie,
1044 	void *fid, size_t *fidsize)
1045 {
1046 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1047 	return 0;
1048 }
1049 
1050 #if 0
1051 /* ARGSUSED0 */
1052 void
1053 perfuse_fs_extattrctl(struct puffs_usermount *pu, int cmd,
1054 	puffs_cookie_t *cookie, int flags, int namespace, const char *attrname)
1055 {
1056 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1057 	return 0;
1058 }
1059 #endif /* 0 */
1060 
1061 /* ARGSUSED0 */
1062 void
1063 perfuse_fs_suspend(struct puffs_usermount *pu, int status)
1064 {
1065 	return;
1066 }
1067 
1068 
1069 int
1070 perfuse_node_lookup(struct puffs_usermount *pu, puffs_cookie_t opc,
1071 	struct puffs_newinfo *pni, const struct puffs_cn *pcn)
1072 {
1073 	struct perfuse_state *ps;
1074 	struct puffs_node *pn;
1075 	mode_t mode;
1076 	int error;
1077 
1078 	ps = puffs_getspecific(pu);
1079 	node_ref(opc);
1080 
1081 	/*
1082 	 * Check permissions
1083 	 */
1084 	switch(pcn->pcn_nameiop) {
1085 	case NAMEI_DELETE: /* FALLTHROUGH */
1086 	case NAMEI_RENAME: /* FALLTHROUGH */
1087 	case NAMEI_CREATE:
1088 		if (pcn->pcn_flags & NAMEI_ISLASTCN)
1089 			mode = PUFFS_VEXEC|PUFFS_VWRITE;
1090 		else
1091 			mode = PUFFS_VEXEC;
1092 		break;
1093 	case NAMEI_LOOKUP: /* FALLTHROUGH */
1094 	default:
1095 		mode = PUFFS_VEXEC;
1096 		break;
1097 	}
1098 
1099 	if ((error = mode_access(opc, pcn->pcn_cred, mode)) != 0)
1100 		goto out;
1101 
1102 	error = node_lookup_common(pu, (puffs_cookie_t)opc, pni,
1103 				   pcn->pcn_name, pcn->pcn_cred, &pn);
1104 
1105 	if (error != 0)
1106 		goto out;
1107 
1108 	/*
1109 	 * Kernel would kill us if the filesystem returned the parent
1110 	 * itself. If we want to live, hide that!
1111 	 */
1112 	if ((opc == (puffs_cookie_t)pn) && (strcmp(pcn->pcn_name, ".") != 0)) {
1113 		DERRX(EX_SOFTWARE, "lookup \"%s\" in \"%s\" returned parent",
1114 		      pcn->pcn_name, perfuse_node_path(ps, opc));
1115 		/* NOTREACHED */
1116 		error = ESTALE;
1117 		goto out;
1118 	}
1119 
1120 	/*
1121 	 * Removed node
1122 	 */
1123 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_REMOVED) {
1124 		error = ENOENT;
1125 		goto out;
1126 	}
1127 
1128 	/*
1129 	 * Check for sticky bit. Unfortunately there is no way to
1130 	 * do this before creating the puffs_node, since we require
1131 	 * this operation to get the node owner.
1132 	 */
1133 	switch (pcn->pcn_nameiop) {
1134 	case NAMEI_DELETE: /* FALLTHROUGH */
1135 	case NAMEI_RENAME:
1136 		error = sticky_access(opc, pn, pcn->pcn_cred);
1137 		if (error != 0) {
1138 			(void)perfuse_node_reclaim(pu, pn);
1139 			goto out;
1140 		}
1141 		break;
1142 	default:
1143 		break;
1144 	}
1145 
1146 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
1147 
1148 	error = 0;
1149 
1150 out:
1151 	node_rele(opc);
1152 	return error;
1153 }
1154 
1155 int
1156 perfuse_node_create(struct puffs_usermount *pu, puffs_cookie_t opc,
1157 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
1158 	const struct vattr *vap)
1159 {
1160 	perfuse_msg_t *pm;
1161 	struct perfuse_state *ps;
1162 	struct fuse_create_in *fci;
1163 	struct fuse_entry_out *feo;
1164 	struct fuse_open_out *foo;
1165 	struct puffs_node *pn;
1166 	const char *name;
1167 	size_t namelen;
1168 	size_t len;
1169 	int error;
1170 
1171 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1172 		return ENOENT;
1173 
1174 	node_ref(opc);
1175 
1176 	/*
1177 	 * If create is unimplemented: Check that it does not
1178 	 * already exists, and if not, do mknod and open
1179 	 */
1180 	ps = puffs_getspecific(pu);
1181 	if (ps->ps_flags & PS_NO_CREAT) {
1182 		error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
1183 					   pcn->pcn_cred, &pn);
1184 		if (error == 0)	{
1185 			(void)perfuse_node_reclaim(pu, pn);
1186 			error = EEXIST;
1187 			goto out;
1188 		}
1189 
1190 		error = perfuse_node_mknod(pu, opc, pni, pcn, vap);
1191 		if (error != 0)
1192 			goto out;
1193 
1194 		error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
1195 					   pcn->pcn_cred, &pn);
1196 		if (error != 0)
1197 			goto out;
1198 
1199 		/*
1200 		 * FUSE does the open at create time, while
1201 		 * NetBSD will open in a subsequent operation.
1202 		 * We need to open now, in order to retain FUSE
1203 		 * semantics. The calling process will not get
1204 		 * a file descriptor before the kernel sends
1205 		 * the open operation.
1206 		 */
1207 		error = perfuse_node_open(pu, (puffs_cookie_t)pn,
1208 					  FWRITE, pcn->pcn_cred);
1209 		goto out;
1210 	}
1211 
1212 	name = pcn->pcn_name;
1213 	namelen = pcn->pcn_namelen + 1;
1214 	len = sizeof(*fci) + namelen;
1215 
1216 	/*
1217 	 * flags should use O_WRONLY instead of O_RDWR, but it
1218 	 * breaks when the caller tries to read from file.
1219 	 *
1220 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1221 	 */
1222 	pm = ps->ps_new_msg(pu, opc, FUSE_CREATE, len, pcn->pcn_cred);
1223 	fci = GET_INPAYLOAD(ps, pm, fuse_create_in);
1224 	fci->flags = O_CREAT | O_TRUNC | O_RDWR;
1225 	fci->mode = vap->va_mode | VTTOIF(vap->va_type);
1226 	fci->umask = 0; 	/* Seems unused by libfuse */
1227 	(void)strlcpy((char*)(void *)(fci + 1), name, namelen);
1228 
1229 	len = sizeof(*feo) + sizeof(*foo);
1230 	if ((error = xchg_msg(pu, opc, pm, len, wait_reply)) != 0) {
1231 		/*
1232 		 * create is unimplmented, remember it for later,
1233 		 * and start over using mknod and open instead.
1234 		 */
1235 		if (error == ENOSYS) {
1236 			ps->ps_flags |= PS_NO_CREAT;
1237 			error = perfuse_node_create(pu, opc, pni, pcn, vap);
1238 		}
1239 
1240 		goto out;
1241 	}
1242 
1243 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
1244 	foo = (struct fuse_open_out *)(void *)(feo + 1);
1245 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
1246 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
1247 
1248 	/*
1249 	 * Save the file handle and inode in node private data
1250 	 * so that we can reuse it later
1251 	 */
1252 	pn = perfuse_new_pn(pu, name, opc);
1253 	perfuse_new_fh((puffs_cookie_t)pn, foo->fh, FWRITE);
1254 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
1255 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
1256 	perfuse_node_cache(ps, pn);
1257 
1258 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
1259 	pn->pn_va.va_gen = (u_long)(feo->generation);
1260 
1261 	puffs_newinfo_setcookie(pni, pn);
1262 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
1263 	puffs_newinfo_setva(pni, &pn->pn_va);
1264 	perfuse_newinfo_setttl(pni, pn, feo, NULL);
1265 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
1266 
1267 #ifdef PERFUSE_DEBUG
1268 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1269 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
1270 			"nodeid = 0x%"PRIx64", wfh = 0x%"PRIx64"\n",
1271 			__func__, (void *)pn, pcn->pcn_name,
1272 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid,
1273 			foo->fh);
1274 #endif
1275 
1276 	ps->ps_destroy_msg(pm);
1277 	error = 0;
1278 
1279 out:
1280 	node_rele(opc);
1281 	return error;
1282 }
1283 
1284 
1285 int
1286 perfuse_node_mknod(struct puffs_usermount *pu, puffs_cookie_t opc,
1287 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
1288 	const struct vattr *vap)
1289 {
1290 	struct perfuse_state *ps;
1291 	perfuse_msg_t *pm;
1292 	struct fuse_mknod_in *fmi;
1293 	const char* path;
1294 	size_t len;
1295 	int error;
1296 
1297 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1298 		return ENOENT;
1299 
1300 	node_ref(opc);
1301 
1302 	/*
1303 	 * Only superuser can mknod objects other than
1304 	 * directories, files, socks, fifo and links.
1305 	 *
1306 	 * Create an object require -WX permission in the parent directory
1307 	 */
1308 	switch (vap->va_type) {
1309 	case VDIR:	/* FALLTHROUGH */
1310 	case VREG:	/* FALLTHROUGH */
1311 	case VFIFO:	/* FALLTHROUGH */
1312 	case VSOCK:
1313 		break;
1314 	default:	/* VNON, VBLK, VCHR, VBAD */
1315 		if (!puffs_cred_isjuggernaut(pcn->pcn_cred)) {
1316 			error = EPERM;
1317 			goto out;
1318 		}
1319 		break;
1320 	}
1321 
1322 
1323 	ps = puffs_getspecific(pu);
1324 	path = pcn->pcn_name;
1325 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
1326 
1327 	/*
1328 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1329 	 */
1330 	pm = ps->ps_new_msg(pu, opc, FUSE_MKNOD, len, pcn->pcn_cred);
1331 	fmi = GET_INPAYLOAD(ps, pm, fuse_mknod_in);
1332 	fmi->mode = vap->va_mode | VTTOIF(vap->va_type);
1333 	fmi->rdev = (uint32_t)vap->va_rdev;
1334 	fmi->umask = 0; 	/* Seems unused bu libfuse */
1335 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
1336 
1337 	error = node_mk_common(pu, opc, pni, pcn, pm);
1338 
1339 out:
1340 	node_rele(opc);
1341 	return error;
1342 }
1343 
1344 
1345 int
1346 perfuse_node_open(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
1347 	const struct puffs_cred *pcr)
1348 {
1349 	return perfuse_node_open2(pu, opc, mode, pcr, NULL);
1350 }
1351 
1352 int
1353 perfuse_node_open2(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
1354 	const struct puffs_cred *pcr, int *oflags)
1355 {
1356 	struct perfuse_state *ps;
1357 	struct perfuse_node_data *pnd;
1358 	perfuse_msg_t *pm;
1359 	mode_t fmode;
1360 	int op;
1361 	struct fuse_open_in *foi;
1362 	struct fuse_open_out *foo;
1363 	struct puffs_node *pn;
1364 	int error;
1365 
1366 	ps = puffs_getspecific(pu);
1367 	pn = (struct puffs_node *)opc;
1368 	pnd = PERFUSE_NODE_DATA(opc);
1369 	error = 0;
1370 
1371 	if (pnd->pnd_flags & PND_REMOVED)
1372 		return ENOENT;
1373 
1374 	node_ref(opc);
1375 
1376 	if (puffs_pn_getvap(pn)->va_type == VDIR)
1377 		op = FUSE_OPENDIR;
1378 	else
1379 		op = FUSE_OPEN;
1380 
1381 	/*
1382 	 * libfuse docs says
1383 	 * - O_CREAT and O_EXCL should never be set.
1384 	 * - O_TRUNC may be used if mount option atomic_o_trunc is used XXX
1385 	 *
1386 	 * O_APPEND makes no sense since FUSE always sends
1387 	 * the file offset for write operations. If the
1388 	 * filesystem uses pwrite(), O_APPEND would cause
1389 	 * the offset to be ignored and cause file corruption.
1390 	 */
1391 	mode &= ~(O_CREAT|O_EXCL|O_APPEND);
1392 
1393 	/*
1394 	 * Do not open twice, and do not reopen for reading
1395 	 * if we already have write handle.
1396 	 */
1397 	switch (mode & (FREAD|FWRITE)) {
1398 	case FREAD:
1399 		if (pnd->pnd_flags & (PND_RFH|PND_WFH))
1400 			goto out;
1401 		break;
1402 	case FWRITE:
1403 		if (pnd->pnd_flags & PND_WFH)
1404 			goto out;
1405 		break;
1406 	case FREAD|FWRITE:
1407 		if (pnd->pnd_flags & PND_WFH)
1408 			goto out;
1409 
1410 		/*
1411 		 * Corner case: if already open for reading (PND_RFH)
1412 		 * and re-opening FREAD|FWRITE, we need to reopen,
1413 		 * but only for writing. Note the change on mode
1414 		 * will only affect perfuse_new_fh()
1415 		 */
1416 		if (pnd->pnd_flags & PND_RFH)
1417 			mode &= ~FREAD;
1418 		break;
1419 	default:
1420 		DWARNX("open without either FREAD nor FWRITE");
1421 		error = EPERM;
1422 		goto out;
1423 	}
1424 
1425 	/*
1426 	 * Queue open on a node so that we do not open
1427 	 * twice. This would be better with read and
1428 	 * write distinguished.
1429 	 */
1430 	while (pnd->pnd_flags & PND_INOPEN)
1431 		requeue_request(pu, opc, PCQ_OPEN);
1432 	pnd->pnd_flags |= PND_INOPEN;
1433 
1434 	/*
1435 	 * Convert PUFFS mode to FUSE mode: convert FREAD/FWRITE
1436 	 * to O_RDONLY/O_WRONLY while perserving the other options.
1437 	 */
1438 	fmode = mode & ~(FREAD|FWRITE);
1439 	fmode |= (mode & FWRITE) ? O_RDWR : O_RDONLY;
1440 
1441 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*foi), pcr);
1442 	foi = GET_INPAYLOAD(ps, pm, fuse_open_in);
1443 	foi->flags = fmode;
1444 	foi->unused = 0;
1445 
1446 	if ((error = xchg_msg(pu, opc, pm, sizeof(*foo), wait_reply)) != 0)
1447 		goto out;
1448 
1449 	foo = GET_OUTPAYLOAD(ps, pm, fuse_open_out);
1450 
1451 	/*
1452 	 * Save the file handle in node private data
1453 	 * so that we can reuse it later
1454 	 */
1455 	perfuse_new_fh(opc, foo->fh, mode);
1456 
1457 	/*
1458 	 * Set direct I/O if the filesystems forces it
1459 	 */
1460 	if ((foo->open_flags & FUSE_FOPEN_DIRECT_IO) && (oflags != NULL))
1461 		*oflags |= PUFFS_OPEN_IO_DIRECT;
1462 
1463 #ifdef PERFUSE_DEBUG
1464 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1465 		DPRINTF("%s: opc = %p, file = \"%s\", "
1466 			"nodeid = 0x%"PRIx64", %s%sfh = 0x%"PRIx64"\n",
1467 			__func__, (void *)opc, perfuse_node_path(ps, opc),
1468 			pnd->pnd_nodeid, mode & FREAD ? "r" : "",
1469 			mode & FWRITE ? "w" : "", foo->fh);
1470 #endif
1471 
1472 	ps->ps_destroy_msg(pm);
1473 out:
1474 
1475 	pnd->pnd_flags &= ~PND_INOPEN;
1476 	(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
1477 
1478 	node_rele(opc);
1479 	return error;
1480 }
1481 
1482 /* ARGSUSED0 */
1483 int
1484 perfuse_node_close(struct puffs_usermount *pu, puffs_cookie_t opc, int flags,
1485 	const struct puffs_cred *pcr)
1486 {
1487 	struct perfuse_node_data *pnd;
1488 
1489 	pnd = PERFUSE_NODE_DATA(opc);
1490 
1491 	if (!(pnd->pnd_flags & PND_OPEN))
1492 		return EBADF;
1493 
1494 	/*
1495 	 * Actual close is postponed at inactive time.
1496 	 */
1497 	return 0;
1498 }
1499 
1500 int
1501 perfuse_node_access(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
1502 	const struct puffs_cred *pcr)
1503 {
1504 	perfuse_msg_t *pm;
1505 	struct perfuse_state *ps;
1506 	struct fuse_access_in *fai;
1507 	int error;
1508 
1509 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1510 		return ENOENT;
1511 
1512 	node_ref(opc);
1513 
1514 	/*
1515 	 * If we previously detected the filesystem does not
1516 	 * implement access(), short-circuit the call and skip
1517 	 * to libpuffs access() emulation.
1518 	 */
1519 	ps = puffs_getspecific(pu);
1520 	if (ps->ps_flags & PS_NO_ACCESS) {
1521 		const struct vattr *vap;
1522 
1523 		vap = puffs_pn_getvap((struct puffs_node *)opc);
1524 
1525 		error = puffs_access(IFTOVT(vap->va_mode),
1526 				     vap->va_mode & ACCESSPERMS,
1527 				     vap->va_uid, vap->va_gid,
1528 				     (mode_t)mode, pcr);
1529 		goto out;
1530 	}
1531 
1532 	/*
1533 	 * Plain access call
1534 	 */
1535 	pm = ps->ps_new_msg(pu, opc, FUSE_ACCESS, sizeof(*fai), pcr);
1536 	fai = GET_INPAYLOAD(ps, pm, fuse_access_in);
1537 	fai->mask = 0;
1538 	fai->mask |= (mode & PUFFS_VREAD) ? R_OK : 0;
1539 	fai->mask |= (mode & PUFFS_VWRITE) ? W_OK : 0;
1540 	fai->mask |= (mode & PUFFS_VEXEC) ? X_OK : 0;
1541 
1542 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
1543 
1544 	ps->ps_destroy_msg(pm);
1545 
1546 	/*
1547 	 * If unimplemented, start over with emulation
1548 	 */
1549 	if (error == ENOSYS) {
1550 		ps->ps_flags |= PS_NO_ACCESS;
1551 		error = perfuse_node_access(pu, opc, mode, pcr);
1552 	}
1553 
1554 out:
1555 	node_rele(opc);
1556 	return error;
1557 }
1558 
1559 int
1560 perfuse_node_getattr(struct puffs_usermount *pu, puffs_cookie_t opc,
1561 	struct vattr *vap, const struct puffs_cred *pcr)
1562 {
1563 	return perfuse_node_getattr_ttl(pu, opc, vap, pcr, NULL);
1564 }
1565 
1566 int
1567 perfuse_node_getattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
1568 	struct vattr *vap, const struct puffs_cred *pcr,
1569 	struct timespec *va_ttl)
1570 {
1571 	perfuse_msg_t *pm = NULL;
1572 	struct perfuse_state *ps;
1573 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
1574 	struct fuse_getattr_in *fgi;
1575 	struct fuse_attr_out *fao;
1576 	int error = 0;
1577 
1578 	if ((pnd->pnd_flags & PND_REMOVED) && !(pnd->pnd_flags & PND_OPEN))
1579 		return ENOENT;
1580 
1581 	node_ref(opc);
1582 
1583 	/*
1584 	 * Serialize size access, see comment in perfuse_node_setattr().
1585 	 */
1586 	while (pnd->pnd_flags & PND_INRESIZE)
1587 		requeue_request(pu, opc, PCQ_RESIZE);
1588 	pnd->pnd_flags |= PND_INRESIZE;
1589 
1590 	ps = puffs_getspecific(pu);
1591 
1592 	/*
1593 	 * FUSE_GETATTR_FH must be set in fgi->flags
1594 	 * if we use for fgi->fh
1595 	 */
1596 	pm = ps->ps_new_msg(pu, opc, FUSE_GETATTR, sizeof(*fgi), pcr);
1597 	fgi = GET_INPAYLOAD(ps, pm, fuse_getattr_in);
1598 	fgi->getattr_flags = 0;
1599 	fgi->dummy = 0;
1600 	fgi->fh = 0;
1601 
1602 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_OPEN) {
1603 		fgi->fh = perfuse_get_fh(opc, FREAD);
1604 		fgi->getattr_flags |= FUSE_GETATTR_FH;
1605 	}
1606 
1607 #ifdef PERFUSE_DEBUG
1608 	if (perfuse_diagflags & PDF_RESIZE)
1609 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__, (void *)opc,
1610 		    vap->va_size);
1611 #endif
1612 
1613 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
1614 		goto out;
1615 
1616 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1617 
1618 #ifdef PERFUSE_DEBUG
1619 	if (perfuse_diagflags & PDF_RESIZE)
1620 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1621 		    (void *)opc, vap->va_size, fao->attr.size);
1622 #endif
1623 
1624 	/*
1625 	 * We set birthtime, flags, filerev,vaflags to 0.
1626 	 * This seems the best bet, since the information is
1627 	 * not available from filesystem.
1628 	 */
1629 	fuse_attr_to_vap(ps, vap, &fao->attr);
1630 
1631 	if (va_ttl != NULL) {
1632 		va_ttl->tv_sec = fao->attr_valid;
1633 		va_ttl->tv_nsec = fao->attr_valid_nsec;
1634 	}
1635 
1636 	ps->ps_destroy_msg(pm);
1637 	error = 0;
1638 out:
1639 
1640 	pnd->pnd_flags &= ~PND_INRESIZE;
1641 	(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
1642 
1643 	node_rele(opc);
1644 	return error;
1645 }
1646 
1647 int
1648 perfuse_node_setattr(struct puffs_usermount *pu, puffs_cookie_t opc,
1649 	const struct vattr *vap, const struct puffs_cred *pcr)
1650 {
1651 	return perfuse_node_setattr_ttl(pu, opc,
1652 					__UNCONST(vap), pcr, NULL, 0);
1653 }
1654 
1655 int
1656 perfuse_node_setattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
1657 	struct vattr *vap, const struct puffs_cred *pcr,
1658 	struct timespec *va_ttl, int xflag)
1659 {
1660 	perfuse_msg_t *pm;
1661 	uint64_t fh;
1662 	struct perfuse_state *ps;
1663 	struct perfuse_node_data *pnd;
1664 	struct fuse_setattr_in *fsi;
1665 	struct fuse_attr_out *fao;
1666 	struct vattr *old_va;
1667 	enum perfuse_xchg_pb_reply reply;
1668 	int error;
1669 #ifdef PERFUSE_DEBUG
1670 	struct vattr *old_vap;
1671 	int resize_debug = 0;
1672 #endif
1673 	ps = puffs_getspecific(pu);
1674 	pnd = PERFUSE_NODE_DATA(opc);
1675 
1676 	/*
1677 	 * The only operation we can do once the file is removed
1678 	 * is to resize it, and we can do it only if it is open.
1679 	 * Do not even send the operation to the filesystem: the
1680 	 * file is not there anymore.
1681 	 */
1682 	if (pnd->pnd_flags & PND_REMOVED) {
1683 		if (!(pnd->pnd_flags & PND_OPEN))
1684 			return ENOENT;
1685 
1686 		return 0;
1687 	}
1688 
1689 	old_va = puffs_pn_getvap((struct puffs_node *)opc);
1690 
1691 	/*
1692 	 * Check for permission to change size
1693 	 * It is always allowed if we already have a write file handle
1694 	 */
1695 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1696 	    !(pnd->pnd_flags & PND_WFH) &&
1697 	    (error = mode_access(opc, pcr, PUFFS_VWRITE)) != 0)
1698 		return error;
1699 
1700 	/*
1701 	 * Check for permission to change dates
1702 	 */
1703 	if (((vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1704 	     (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL)) &&
1705 	    (puffs_access_times(old_va->va_uid, old_va->va_gid,
1706 				old_va->va_mode, 0, pcr) != 0))
1707 		return EPERM;
1708 
1709 	/*
1710 	 * Check for permission to change owner and group
1711 	 */
1712 	if (((vap->va_uid != (uid_t)PUFFS_VNOVAL) ||
1713 	     (vap->va_gid != (gid_t)PUFFS_VNOVAL)) &&
1714 	    (puffs_access_chown(old_va->va_uid, old_va->va_gid,
1715 				vap->va_uid, vap->va_gid, pcr)) != 0)
1716 		return EPERM;
1717 
1718 	/*
1719 	 * Check for sticky bit on non-directory by non root user
1720 	 */
1721 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1722 	    (vap->va_mode & S_ISTXT) && (old_va->va_type != VDIR) &&
1723 	    !puffs_cred_isjuggernaut(pcr))
1724 		return EFTYPE;
1725 
1726 	/*
1727 	 * Check for permission to change permissions
1728 	 */
1729 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1730 	    (puffs_access_chmod(old_va->va_uid, old_va->va_gid,
1731 				old_va->va_type, vap->va_mode, pcr)) != 0)
1732 		return EPERM;
1733 
1734 	node_ref(opc);
1735 
1736 	if (pnd->pnd_flags & PND_WFH)
1737 		fh = perfuse_get_fh(opc, FWRITE);
1738 	else
1739 		fh = FUSE_UNKNOWN_FH;
1740 
1741 	/*
1742 	 * fchmod() sets mode and fh, and it may carry
1743 	 * a resize as well. That may break if the
1744 	 * filesystem does chmod then resize, and fails
1745 	 * because it does not have permission anymore.
1746 	 * We work this around by splitting into two setattr.
1747 	 */
1748 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1749 	    (vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1750 	    (fh != FUSE_UNKNOWN_FH)) {
1751 		struct vattr resize_va;
1752 
1753 		(void)memcpy(&resize_va, vap, sizeof(resize_va));
1754 		resize_va.va_mode = (mode_t)PUFFS_VNOVAL;
1755 		if ((error = perfuse_node_setattr_ttl(pu, opc, &resize_va,
1756 						      pcr, va_ttl, xflag)) != 0)
1757 			goto out2;
1758 
1759 		vap->va_size = (u_quad_t)PUFFS_VNOVAL;
1760 	}
1761 
1762 	pm = ps->ps_new_msg(pu, opc, FUSE_SETATTR, sizeof(*fsi), pcr);
1763 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
1764 	fsi->valid = 0;
1765 
1766 	/*
1767 	 * Get a fh if the node is open for writing
1768 	 */
1769 	if (fh != FUSE_UNKNOWN_FH) {
1770 		fsi->fh = fh;
1771 		fsi->valid |= FUSE_FATTR_FH;
1772 	}
1773 
1774 
1775 	if (vap->va_size != (u_quad_t)PUFFS_VNOVAL) {
1776 		fsi->size = vap->va_size;
1777 		fsi->valid |= FUSE_FATTR_SIZE;
1778 
1779 		/*
1780 		 * Serialize anything that can touch file size
1781 		 * to avoid reordered GETATTR and SETATTR.
1782 		 * Out of order SETATTR can report stale size,
1783 		 * which will cause the kernel to truncate the file.
1784 		 * XXX Probably useless now we have a lock on GETATTR
1785 		 */
1786 		while (pnd->pnd_flags & PND_INRESIZE)
1787 			requeue_request(pu, opc, PCQ_RESIZE);
1788 		pnd->pnd_flags |= PND_INRESIZE;
1789 	}
1790 
1791 	/*
1792  	 * When not sending a time field, still fill with
1793 	 * current value, as the filesystem may just reset
1794 	 * the field to Epoch even if fsi->valid bit is
1795 	 * not set (GlusterFS does that).
1796  	 */
1797 	if (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) {
1798 		fsi->atime = vap->va_atime.tv_sec;
1799 		fsi->atimensec = (uint32_t)vap->va_atime.tv_nsec;
1800 		fsi->valid |= FUSE_FATTR_ATIME;
1801 	} else {
1802 		fsi->atime = old_va->va_atime.tv_sec;
1803 		fsi->atimensec = (uint32_t)old_va->va_atime.tv_nsec;
1804 	}
1805 
1806 	if (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) {
1807 		fsi->mtime = vap->va_mtime.tv_sec;
1808 		fsi->mtimensec = (uint32_t)vap->va_mtime.tv_nsec;
1809 		fsi->valid |= FUSE_FATTR_MTIME;
1810 	} else {
1811 		fsi->mtime = old_va->va_mtime.tv_sec;
1812 		fsi->mtimensec = (uint32_t)old_va->va_mtime.tv_nsec;
1813 	}
1814 
1815 	if (vap->va_mode != (mode_t)PUFFS_VNOVAL) {
1816 		fsi->mode = vap->va_mode;
1817 		fsi->valid |= FUSE_FATTR_MODE;
1818 	}
1819 
1820 	if (vap->va_uid != (uid_t)PUFFS_VNOVAL) {
1821 		fsi->uid = vap->va_uid;
1822 		fsi->valid |= FUSE_FATTR_UID;
1823 	}
1824 
1825 	if (vap->va_gid != (gid_t)PUFFS_VNOVAL) {
1826 		fsi->gid = vap->va_gid;
1827 		fsi->valid |= FUSE_FATTR_GID;
1828 	}
1829 
1830 	if (pnd->pnd_lock_owner != 0) {
1831 		fsi->lock_owner = pnd->pnd_lock_owner;
1832 		fsi->valid |= FUSE_FATTR_LOCKOWNER;
1833 	}
1834 
1835 #ifndef PUFFS_KFLAG_NOFLUSH_META
1836 	/*
1837 	 * ftruncate() sends only va_size, and metadata cache
1838 	 * flush adds va_atime and va_mtime. Some FUSE
1839 	 * filesystems will attempt to detect ftruncate by
1840 	 * checking for FATTR_SIZE being set without
1841 	 * FATTR_UID|FATTR_GID|FATTR_ATIME|FATTR_MTIME|FATTR_MODE
1842 	 *
1843 	 * Try to adapt and remove FATTR_ATIME|FATTR_MTIME
1844 	 * if we suspect a ftruncate().
1845 	 */
1846 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1847 	    ((vap->va_mode == (mode_t)PUFFS_VNOVAL) &&
1848 	     (vap->va_uid == (uid_t)PUFFS_VNOVAL) &&
1849 	     (vap->va_gid == (gid_t)PUFFS_VNOVAL))) {
1850 		fsi->atime = 0;
1851 		fsi->atimensec = 0;
1852 		fsi->mtime = 0;
1853 		fsi->mtimensec = 0;
1854 		fsi->valid &= ~(FUSE_FATTR_ATIME|FUSE_FATTR_MTIME);
1855 	}
1856 
1857 	/*
1858 	 * If only atime is changed, discard the operation: it
1859 	 * happens after read, and in that case the filesystem
1860 	 * already updaed atime. NB: utimes() also change mtime.
1861 	 */
1862 	if (fsi->valid == FUSE_FATTR_ATIME)
1863 		fsi->valid &= ~FUSE_FATTR_ATIME;
1864 #endif /* PUFFS_KFLAG_NOFLUSH_META */
1865 
1866 	/*
1867 	 * If nothing remain, discard the operation.
1868 	 */
1869 	if (!(fsi->valid & (FUSE_FATTR_SIZE|FUSE_FATTR_ATIME|FUSE_FATTR_MTIME|
1870 			    FUSE_FATTR_MODE|FUSE_FATTR_UID|FUSE_FATTR_GID))) {
1871 		error = 0;
1872 		ps->ps_destroy_msg(pm);
1873 		goto out;
1874 	}
1875 
1876 #ifdef PERFUSE_DEBUG
1877 	old_vap = puffs_pn_getvap((struct puffs_node *)opc);
1878 
1879 	if ((perfuse_diagflags & PDF_RESIZE) &&
1880 	    (old_vap->va_size != (u_quad_t)PUFFS_VNOVAL)) {
1881 		resize_debug = 1;
1882 
1883 		DPRINTF(">> %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1884 		    (void *)opc,
1885 		    puffs_pn_getvap((struct puffs_node *)opc)->va_size,
1886 		    fsi->size);
1887 	}
1888 #endif
1889 
1890 	/*
1891 	 * Do not honour FAF when changing size. How do
1892 	 * you want such a thing to work?
1893 	 */
1894 	reply = wait_reply;
1895 #ifdef PUFFS_SETATTR_FAF
1896 	if ((xflag & PUFFS_SETATTR_FAF) && !(fsi->valid & FUSE_FATTR_SIZE))
1897 		reply = no_reply;
1898 #endif
1899 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), reply)) != 0)
1900 		goto out;
1901 
1902 	if (reply == no_reply)
1903 		goto out;
1904 
1905 	/*
1906 	 * Copy back the new values
1907 	 */
1908 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1909 
1910 #ifdef PERFUSE_DEBUG
1911 	if (resize_debug)
1912 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1913 		    (void *)opc, old_vap->va_size, fao->attr.size);
1914 #endif
1915 
1916 	fuse_attr_to_vap(ps, old_va, &fao->attr);
1917 
1918 	if (va_ttl != NULL) {
1919 		va_ttl->tv_sec = fao->attr_valid;
1920 		va_ttl->tv_nsec = fao->attr_valid_nsec;
1921 		(void)memcpy(vap, old_va, sizeof(*vap));
1922 	}
1923 
1924 	ps->ps_destroy_msg(pm);
1925 	error = 0;
1926 
1927 out:
1928 	if (pnd->pnd_flags & PND_INRESIZE) {
1929 		pnd->pnd_flags &= ~PND_INRESIZE;
1930 		(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
1931 	}
1932 
1933 out2:
1934 	node_rele(opc);
1935 	return error;
1936 }
1937 
1938 int
1939 perfuse_node_poll(struct puffs_usermount *pu, puffs_cookie_t opc, int *events)
1940 {
1941 	struct perfuse_state *ps;
1942 	perfuse_msg_t *pm;
1943 	struct fuse_poll_in *fpi;
1944 	struct fuse_poll_out *fpo;
1945 	int error;
1946 
1947 	node_ref(opc);
1948 	ps = puffs_getspecific(pu);
1949 	/*
1950 	 * kh is set if FUSE_POLL_SCHEDULE_NOTIFY is set.
1951 	 *
1952 	 * XXX ps_new_msg() is called with NULL creds, which will
1953 	 * be interpreted as FUSE superuser. We have no way to
1954 	 * know the requesting process' credential, but since poll
1955 	 * is supposed to operate on a file that has been open,
1956 	 * permission should have already been checked at open time.
1957 	 * That still may breaks on filesystems that provides odd
1958 	 * semantics.
1959  	 */
1960 	pm = ps->ps_new_msg(pu, opc, FUSE_POLL, sizeof(*fpi), NULL);
1961 	fpi = GET_INPAYLOAD(ps, pm, fuse_poll_in);
1962 	fpi->fh = perfuse_get_fh(opc, FREAD);
1963 	fpi->kh = 0;
1964 	fpi->flags = 0;
1965 
1966 #ifdef PERFUSE_DEBUG
1967 	if (perfuse_diagflags & PDF_FH)
1968 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
1969 			"fh = 0x%"PRIx64"\n", __func__, (void *)opc,
1970 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fpi->fh);
1971 #endif
1972 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fpo), wait_reply)) != 0)
1973 		goto out;
1974 
1975 	fpo = GET_OUTPAYLOAD(ps, pm, fuse_poll_out);
1976 	*events = fpo->revents;
1977 
1978 	ps->ps_destroy_msg(pm);
1979 	error = 0;
1980 
1981 out:
1982 	node_rele(opc);
1983 	return error;
1984 }
1985 
1986 /* ARGSUSED2 */
1987 int
1988 perfuse_node_fsync(struct puffs_usermount *pu, puffs_cookie_t opc,
1989 	const struct puffs_cred *pcr, int flags, off_t offlo, off_t offhi)
1990 {
1991 	int op;
1992 	perfuse_msg_t *pm;
1993 	struct perfuse_state *ps;
1994 	struct perfuse_node_data *pnd;
1995 	struct fuse_fsync_in *ffi;
1996 	uint64_t fh;
1997 	int error = 0;
1998 
1999 	pm = NULL;
2000 	ps = puffs_getspecific(pu);
2001 	pnd = PERFUSE_NODE_DATA(opc);
2002 
2003 	/*
2004 	 * No need to sync a removed node
2005 	 */
2006 	if (pnd->pnd_flags & PND_REMOVED)
2007 		return 0;
2008 
2009 	/*
2010 	 * We do not sync closed files. They have been
2011 	 * sync at inactive time already.
2012 	 */
2013 	if (!(pnd->pnd_flags & PND_OPEN))
2014 		return 0;
2015 
2016 	node_ref(opc);
2017 
2018 	if (puffs_pn_getvap((struct puffs_node *)opc)->va_type == VDIR)
2019 		op = FUSE_FSYNCDIR;
2020 	else 		/* VREG but also other types such as VLNK */
2021 		op = FUSE_FSYNC;
2022 
2023 	/*
2024 	 * Do not sync if there are no change to sync
2025 	 * XXX remove that test on files if we implement mmap
2026 	 */
2027 #ifdef PERFUSE_DEBUG
2028 	if (perfuse_diagflags & PDF_SYNC)
2029 		DPRINTF("%s: TEST opc = %p, file = \"%s\" is %sdirty\n",
2030 			__func__, (void*)opc, perfuse_node_path(ps, opc),
2031 			pnd->pnd_flags & PND_DIRTY ? "" : "not ");
2032 #endif
2033 	if (!(pnd->pnd_flags & PND_DIRTY))
2034 		goto out;
2035 
2036 	/*
2037 	 * It seems NetBSD can call fsync without open first
2038 	 * glusterfs complain in such a situation:
2039 	 * "FSYNC() ERR => -1 (Invalid argument)"
2040 	 * The file will be closed at inactive time.
2041 	 *
2042 	 * We open the directory for reading in order to sync.
2043 	 * This sounds rather counterintuitive, but it works.
2044 	 */
2045 	if (!(pnd->pnd_flags & PND_WFH)) {
2046 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2047 			goto out;
2048 	}
2049 
2050 	if (op == FUSE_FSYNCDIR)
2051 		fh = perfuse_get_fh(opc, FREAD);
2052 	else
2053 		fh = perfuse_get_fh(opc, FWRITE);
2054 
2055 	/*
2056 	 * If fsync_flags  is set, meta data should not be flushed.
2057 	 */
2058 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*ffi), pcr);
2059 	ffi = GET_INPAYLOAD(ps, pm, fuse_fsync_in);
2060 	ffi->fh = fh;
2061 	ffi->fsync_flags = (flags & FFILESYNC) ? 0 : 1;
2062 
2063 #ifdef PERFUSE_DEBUG
2064 	if (perfuse_diagflags & PDF_FH)
2065 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
2066 			__func__, (void *)opc,
2067 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, ffi->fh);
2068 #endif
2069 
2070 	if ((error = xchg_msg(pu, opc, pm,
2071 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
2072 		goto out;
2073 
2074 	/*
2075 	 * No reply beyond fuse_out_header: nothing to do on success
2076 	 * just clear the dirty flag
2077 	 */
2078 	pnd->pnd_flags &= ~PND_DIRTY;
2079 
2080 #ifdef PERFUSE_DEBUG
2081 	if (perfuse_diagflags & PDF_SYNC)
2082 		DPRINTF("%s: CLEAR opc = %p, file = \"%s\"\n",
2083 			__func__, (void*)opc, perfuse_node_path(ps, opc));
2084 #endif
2085 
2086 	ps->ps_destroy_msg(pm);
2087 	error = 0;
2088 
2089 out:
2090 	/*
2091 	 * ENOSYS is not returned to kernel,
2092 	 */
2093 	if (error == ENOSYS)
2094 		error = 0;
2095 
2096 	node_rele(opc);
2097 	return error;
2098 }
2099 
2100 int
2101 perfuse_node_remove(struct puffs_usermount *pu, puffs_cookie_t opc,
2102 	puffs_cookie_t targ, const struct puffs_cn *pcn)
2103 {
2104 	struct perfuse_state *ps;
2105 	struct perfuse_node_data *pnd;
2106 	perfuse_msg_t *pm;
2107 	char *path;
2108 	const char *name;
2109 	size_t len;
2110 	int error;
2111 
2112 	pnd = PERFUSE_NODE_DATA(opc);
2113 
2114 	if ((pnd->pnd_flags & PND_REMOVED) ||
2115 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
2116 		return ENOENT;
2117 
2118 #ifdef PERFUSE_DEBUG
2119 	if (targ == NULL)
2120 		DERRX(EX_SOFTWARE, "%s: targ is NULL", __func__);
2121 
2122 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
2123 		DPRINTF("%s: opc = %p, remove opc = %p, file = \"%s\"\n",
2124 			__func__, (void *)opc, (void *)targ, pcn->pcn_name);
2125 #endif
2126 	node_ref(opc);
2127 	node_ref(targ);
2128 
2129 	/*
2130 	 * Await for all operations on the deleted node to drain,
2131 	 * as the filesystem may be confused to have it deleted
2132 	 * during a getattr
2133 	 */
2134 	while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
2135 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2136 
2137 	ps = puffs_getspecific(pu);
2138 	pnd = PERFUSE_NODE_DATA(opc);
2139 	name = pcn->pcn_name;
2140 	len = pcn->pcn_namelen + 1;
2141 
2142 	pm = ps->ps_new_msg(pu, opc, FUSE_UNLINK, len, pcn->pcn_cred);
2143 	path = _GET_INPAYLOAD(ps, pm, char *);
2144 	(void)strlcpy(path, name, len);
2145 
2146 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2147 		goto out;
2148 
2149 	perfuse_cache_flush(targ);
2150 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2151 
2152 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2153 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2154 
2155 	/*
2156 	 * The parent directory needs a sync
2157 	 */
2158 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2159 
2160 #ifdef PERFUSE_DEBUG
2161 	if (perfuse_diagflags & PDF_FILENAME)
2162 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
2163 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
2164 			pcn->pcn_name);
2165 #endif
2166 	ps->ps_destroy_msg(pm);
2167 	error = 0;
2168 
2169 out:
2170 	node_rele(opc);
2171 	node_rele(targ);
2172 	return error;
2173 }
2174 
2175 int
2176 perfuse_node_link(struct puffs_usermount *pu, puffs_cookie_t opc,
2177 	puffs_cookie_t targ, const struct puffs_cn *pcn)
2178 {
2179 	struct perfuse_state *ps;
2180 	perfuse_msg_t *pm;
2181 	const char *name;
2182 	size_t len;
2183 	struct puffs_node *pn;
2184 	struct fuse_link_in *fli;
2185 	int error;
2186 
2187 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2188 		return ENOENT;
2189 
2190 	node_ref(opc);
2191 	node_ref(targ);
2192 	ps = puffs_getspecific(pu);
2193 	pn = (struct puffs_node *)targ;
2194 	name = pcn->pcn_name;
2195 	len =  sizeof(*fli) + pcn->pcn_namelen + 1;
2196 
2197 	pm = ps->ps_new_msg(pu, opc, FUSE_LINK, len, pcn->pcn_cred);
2198 	fli = GET_INPAYLOAD(ps, pm, fuse_link_in);
2199 	fli->oldnodeid = PERFUSE_NODE_DATA(pn)->pnd_nodeid;
2200 	(void)strlcpy((char *)(void *)(fli + 1), name, len - sizeof(*fli));
2201 
2202 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2203 		goto out;
2204 
2205 	ps->ps_destroy_msg(pm);
2206 	error = 0;
2207 
2208 out:
2209 	node_rele(opc);
2210 	node_rele(targ);
2211 	return error;
2212 }
2213 
2214 int
2215 perfuse_node_rename(struct puffs_usermount *pu, puffs_cookie_t opc,
2216 	puffs_cookie_t src, const struct puffs_cn *pcn_src,
2217 	puffs_cookie_t targ_dir, puffs_cookie_t targ,
2218 	const struct puffs_cn *pcn_targ)
2219 {
2220 	struct perfuse_state *ps;
2221 	struct perfuse_node_data *dstdir_pnd;
2222 	perfuse_msg_t *pm;
2223 	struct fuse_rename_in *fri;
2224 	const char *newname;
2225 	const char *oldname;
2226 	char *np;
2227 	int error;
2228 	size_t len;
2229 	size_t newname_len;
2230 	size_t oldname_len;
2231 
2232 	if ((PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED) ||
2233 	    (PERFUSE_NODE_DATA(src)->pnd_flags & PND_REMOVED) ||
2234 	    (PERFUSE_NODE_DATA(targ_dir)->pnd_flags & PND_REMOVED))
2235 		return ENOENT;
2236 
2237 	node_ref(opc);
2238 	node_ref(src);
2239 
2240 	/*
2241 	 * Await for all operations on the deleted node to drain,
2242 	 * as the filesystem may be confused to have it deleted
2243 	 * during a getattr
2244 	 */
2245 	if ((struct puffs_node *)targ != NULL) {
2246 		node_ref(targ);
2247 		while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
2248 			requeue_request(pu, targ, PCQ_AFTERXCHG);
2249 	} else {
2250 		while (PERFUSE_NODE_DATA(src)->pnd_inxchg)
2251 			requeue_request(pu, src, PCQ_AFTERXCHG);
2252 	}
2253 
2254 	ps = puffs_getspecific(pu);
2255 	newname =  pcn_targ->pcn_name;
2256 	newname_len = pcn_targ->pcn_namelen + 1;
2257 	oldname =  pcn_src->pcn_name;
2258 	oldname_len = pcn_src->pcn_namelen + 1;
2259 
2260 	len = sizeof(*fri) + oldname_len + newname_len;
2261 	pm = ps->ps_new_msg(pu, opc, FUSE_RENAME, len, pcn_targ->pcn_cred);
2262 	fri = GET_INPAYLOAD(ps, pm, fuse_rename_in);
2263 	fri->newdir = PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid;
2264 	np = (char *)(void *)(fri + 1);
2265 	(void)strlcpy(np, oldname, oldname_len);
2266 	np += oldname_len;
2267 	(void)strlcpy(np, newname, newname_len);
2268 
2269 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2270 		goto out;
2271 
2272 
2273 	/*
2274 	 * Record new parent nodeid
2275 	 */
2276 	dstdir_pnd = PERFUSE_NODE_DATA(targ_dir);
2277 	PERFUSE_NODE_DATA(src)->pnd_parent_nodeid = dstdir_pnd->pnd_nodeid;
2278 
2279 	if (opc != targ_dir)
2280 		dstdir_pnd->pnd_flags |= PND_DIRTY;
2281 
2282 	if (strcmp(newname, "..") != 0)
2283 		(void)strlcpy(PERFUSE_NODE_DATA(src)->pnd_name,
2284 		    newname, MAXPATHLEN);
2285 	else
2286 		PERFUSE_NODE_DATA(src)->pnd_name[0] = 0; /* forget name */
2287 
2288 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2289 
2290 	if ((struct puffs_node *)targ != NULL) {
2291 		perfuse_cache_flush(targ);
2292 		PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2293 	}
2294 
2295 #ifdef PERFUSE_DEBUG
2296 	if (perfuse_diagflags & PDF_FILENAME)
2297 		DPRINTF("%s: nodeid = 0x%"PRIx64" file = \"%s\" renamed \"%s\" "
2298 			"nodeid = 0x%"PRIx64" -> nodeid = 0x%"PRIx64" \"%s\"\n",
2299 	 		__func__, PERFUSE_NODE_DATA(src)->pnd_nodeid,
2300 			pcn_src->pcn_name, pcn_targ->pcn_name,
2301 			PERFUSE_NODE_DATA(opc)->pnd_nodeid,
2302 			PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid,
2303 			perfuse_node_path(ps, targ_dir));
2304 #endif
2305 
2306 	ps->ps_destroy_msg(pm);
2307 	error = 0;
2308 
2309 out:
2310 	node_rele(opc);
2311 	node_rele(src);
2312 	if ((struct puffs_node *)targ != NULL)
2313 		node_rele(targ);
2314 
2315 	return error;
2316 }
2317 
2318 int
2319 perfuse_node_mkdir(struct puffs_usermount *pu, puffs_cookie_t opc,
2320 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
2321 	const struct vattr *vap)
2322 {
2323 	struct perfuse_state *ps;
2324 	perfuse_msg_t *pm;
2325 	struct fuse_mkdir_in *fmi;
2326 	const char *path;
2327 	size_t len;
2328 	int error;
2329 
2330 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2331 		return ENOENT;
2332 
2333 	node_ref(opc);
2334 	ps = puffs_getspecific(pu);
2335 	path = pcn->pcn_name;
2336 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
2337 
2338 	pm = ps->ps_new_msg(pu, opc, FUSE_MKDIR, len, pcn->pcn_cred);
2339 	fmi = GET_INPAYLOAD(ps, pm, fuse_mkdir_in);
2340 	fmi->mode = vap->va_mode;
2341 	fmi->umask = 0; 	/* Seems unused by libfuse? */
2342 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
2343 
2344 	error = node_mk_common(pu, opc, pni, pcn, pm);
2345 
2346 	node_rele(opc);
2347 	return error;
2348 }
2349 
2350 
2351 int
2352 perfuse_node_rmdir(struct puffs_usermount *pu, puffs_cookie_t opc,
2353 	puffs_cookie_t targ, const struct puffs_cn *pcn)
2354 {
2355 	struct perfuse_state *ps;
2356 	struct perfuse_node_data *pnd;
2357 	perfuse_msg_t *pm;
2358 	char *path;
2359 	const char *name;
2360 	size_t len;
2361 	int error;
2362 
2363 	pnd = PERFUSE_NODE_DATA(opc);
2364 
2365 	if ((pnd->pnd_flags & PND_REMOVED) ||
2366 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
2367 		return ENOENT;
2368 
2369 	/*
2370 	 * Attempt to rmdir dir/.. shoud raise ENOTEMPTY
2371 	 */
2372 	if (PERFUSE_NODE_DATA(targ)->pnd_nodeid == pnd->pnd_parent_nodeid)
2373 		return ENOTEMPTY;
2374 
2375 	node_ref(opc);
2376 	node_ref(targ);
2377 
2378 	/*
2379 	 * Await for all operations on the deleted node to drain,
2380 	 * as the filesystem may be confused to have it deleted
2381 	 * during a getattr
2382 	 */
2383 	while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
2384 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2385 
2386 	ps = puffs_getspecific(pu);
2387 	name = pcn->pcn_name;
2388 	len = pcn->pcn_namelen + 1;
2389 
2390 	pm = ps->ps_new_msg(pu, opc, FUSE_RMDIR, len, pcn->pcn_cred);
2391 	path = _GET_INPAYLOAD(ps, pm, char *);
2392 	(void)strlcpy(path, name, len);
2393 
2394 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2395 		goto out;
2396 
2397 	perfuse_cache_flush(targ);
2398 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2399 
2400 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2401 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2402 
2403 	/*
2404 	 * The parent directory needs a sync
2405 	 */
2406 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2407 
2408 #ifdef PERFUSE_DEBUG
2409 	if (perfuse_diagflags & PDF_FILENAME)
2410 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
2411 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
2412 			perfuse_node_path(ps, targ));
2413 #endif
2414 	ps->ps_destroy_msg(pm);
2415 	error = 0;
2416 
2417 out:
2418 	node_rele(opc);
2419 	node_rele(targ);
2420 	return error;
2421 }
2422 
2423 /* vap is unused */
2424 /* ARGSUSED4 */
2425 int
2426 perfuse_node_symlink(struct puffs_usermount *pu, puffs_cookie_t opc,
2427 	struct puffs_newinfo *pni, const struct puffs_cn *pcn_src,
2428 	const struct vattr *vap, const char *link_target)
2429 {
2430 	struct perfuse_state *ps;
2431 	perfuse_msg_t *pm;
2432 	char *np;
2433 	const char *path;
2434 	size_t path_len;
2435 	size_t linkname_len;
2436 	size_t len;
2437 	int error;
2438 
2439 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2440 		return ENOENT;
2441 
2442 	node_ref(opc);
2443 	ps = puffs_getspecific(pu);
2444 	path = pcn_src->pcn_name;
2445 	path_len = pcn_src->pcn_namelen + 1;
2446 	linkname_len = strlen(link_target) + 1;
2447 	len = path_len + linkname_len;
2448 
2449 	pm = ps->ps_new_msg(pu, opc, FUSE_SYMLINK, len, pcn_src->pcn_cred);
2450 	np = _GET_INPAYLOAD(ps, pm, char *);
2451 	(void)strlcpy(np, path, path_len);
2452 	np += path_len;
2453 	(void)strlcpy(np, link_target, linkname_len);
2454 
2455 	error = node_mk_common(pu, opc, pni, pcn_src, pm);
2456 
2457 	node_rele(opc);
2458 	return error;
2459 }
2460 
2461 /* ARGSUSED4 */
2462 int
2463 perfuse_node_readdir(struct puffs_usermount *pu, puffs_cookie_t opc,
2464 	struct dirent *dent, off_t *readoff, size_t *reslen,
2465 	const struct puffs_cred *pcr, int *eofflag, off_t *cookies,
2466 	size_t *ncookies)
2467 {
2468 	perfuse_msg_t *pm;
2469 	uint64_t fh;
2470 	struct perfuse_state *ps;
2471 	struct perfuse_node_data *pnd;
2472 	struct fuse_read_in *fri;
2473 	struct fuse_out_header *foh;
2474 	struct fuse_dirent *fd;
2475 	size_t foh_len;
2476 	int error;
2477 	size_t fd_maxlen;
2478 
2479 	error = 0;
2480 	node_ref(opc);
2481 	ps = puffs_getspecific(pu);
2482 
2483 	/*
2484 	 * readdir state is kept at node level, and several readdir
2485 	 * requests can be issued at the same time on the same node.
2486 	 * We need to queue requests so that only one is in readdir
2487 	 * code at the same time.
2488 	 */
2489 	pnd = PERFUSE_NODE_DATA(opc);
2490 	while (pnd->pnd_flags & PND_INREADDIR)
2491 		requeue_request(pu, opc, PCQ_READDIR);
2492 	pnd->pnd_flags |= PND_INREADDIR;
2493 
2494 #ifdef PERFUSE_DEBUG
2495 	if (perfuse_diagflags & PDF_READDIR)
2496 		DPRINTF("%s: READDIR opc = %p enter critical section\n",
2497 			__func__, (void *)opc);
2498 #endif
2499 	/*
2500 	 * Re-initialize pnd->pnd_fd_cookie on the first readdir for a node
2501 	 */
2502 	if (*readoff == 0)
2503 		pnd->pnd_fd_cookie = 0;
2504 
2505 	/*
2506 	 * Do we already have the data bufered?
2507 	 */
2508 	if (pnd->pnd_dirent != NULL)
2509 		goto out;
2510 	pnd->pnd_dirent_len = 0;
2511 
2512 	/*
2513 	 * It seems NetBSD can call readdir without open first
2514 	 * libfuse will crash if it is done that way, hence open first.
2515 	 */
2516 	if (!(pnd->pnd_flags & PND_OPEN)) {
2517 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2518 			goto out;
2519 	}
2520 
2521 	fh = perfuse_get_fh(opc, FREAD);
2522 
2523 #ifdef PERFUSE_DEBUG
2524 	if (perfuse_diagflags & PDF_FH)
2525 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
2526 			"rfh = 0x%"PRIx64"\n", __func__, (void *)opc,
2527 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fh);
2528 #endif
2529 
2530 	pnd->pnd_all_fd = NULL;
2531 	pnd->pnd_all_fd_len = 0;
2532 	fd_maxlen = ps->ps_max_readahead - sizeof(*foh);
2533 
2534 	do {
2535 		size_t fd_len;
2536 		char *afdp;
2537 
2538 		pm = ps->ps_new_msg(pu, opc, FUSE_READDIR, sizeof(*fri), pcr);
2539 
2540 		/*
2541 		 * read_flags, lock_owner and flags are unused in libfuse
2542 		 */
2543 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
2544 		fri->fh = fh;
2545 		fri->offset = pnd->pnd_fd_cookie;
2546 		fri->size = (uint32_t)fd_maxlen;
2547 		fri->read_flags = 0;
2548 		fri->lock_owner = 0;
2549 		fri->flags = 0;
2550 
2551 		if ((error = xchg_msg(pu, opc, pm,
2552 				      UNSPEC_REPLY_LEN, wait_reply)) != 0)
2553 			goto out;
2554 
2555 		/*
2556 		 * There are many puffs_framebufs calls later,
2557 		 * therefore foh will not be valid for a long time.
2558 		 * Just get the length and forget it.
2559 		 */
2560 		foh = GET_OUTHDR(ps, pm);
2561 		foh_len = foh->len;
2562 
2563 		/*
2564 		 * Empty read: we reached the end of the buffer.
2565 		 */
2566 		if (foh_len == sizeof(*foh)) {
2567 			ps->ps_destroy_msg(pm);
2568 			*eofflag = 1;
2569 			break;
2570 		}
2571 
2572 		/*
2573 		 * Check for corrupted message.
2574 		 */
2575 		if (foh_len < sizeof(*foh) + sizeof(*fd)) {
2576 			ps->ps_destroy_msg(pm);
2577 			DWARNX("readdir reply too short");
2578 			error = EIO;
2579 			goto out;
2580 		}
2581 
2582 
2583 		fd = GET_OUTPAYLOAD(ps, pm, fuse_dirent);
2584 		fd_len = foh_len - sizeof(*foh);
2585 
2586 		pnd->pnd_all_fd = realloc(pnd->pnd_all_fd,
2587 					  pnd->pnd_all_fd_len + fd_len);
2588 		if (pnd->pnd_all_fd  == NULL)
2589 			DERR(EX_OSERR, "%s: malloc failed", __func__);
2590 
2591 		afdp = (char *)(void *)pnd->pnd_all_fd + pnd->pnd_all_fd_len;
2592 		(void)memcpy(afdp, fd, fd_len);
2593 
2594 		pnd->pnd_all_fd_len += fd_len;
2595 
2596 		/*
2597 		 * The fd->off field is used as a cookie for
2598 		 * resuming the next readdir() where this one was left.
2599 	 	 */
2600 		pnd->pnd_fd_cookie = readdir_last_cookie(fd, fd_len);
2601 
2602 		ps->ps_destroy_msg(pm);
2603 	} while (1 /* CONSTCOND */);
2604 
2605 	if (pnd->pnd_all_fd != NULL) {
2606 		if (fuse_to_dirent(pu, opc, pnd->pnd_all_fd,
2607 				   pnd->pnd_all_fd_len) == -1)
2608 			error = EIO;
2609 	}
2610 
2611 out:
2612 	if (pnd->pnd_all_fd != NULL) {
2613 		free(pnd->pnd_all_fd);
2614 		pnd->pnd_all_fd = NULL;
2615 		pnd->pnd_all_fd_len = 0;
2616 	}
2617 
2618 	if (error == 0)
2619 		readdir_buffered(opc, dent, readoff, reslen);
2620 
2621 	/*
2622 	 * Schedule queued readdir requests
2623 	 */
2624 	pnd->pnd_flags &= ~PND_INREADDIR;
2625 	(void)dequeue_requests(opc, PCQ_READDIR, DEQUEUE_ALL);
2626 
2627 #ifdef PERFUSE_DEBUG
2628 	if (perfuse_diagflags & PDF_READDIR)
2629 		DPRINTF("%s: READDIR opc = %p exit critical section\n",
2630 			__func__, (void *)opc);
2631 #endif
2632 
2633 	node_rele(opc);
2634 	return error;
2635 }
2636 
2637 int
2638 perfuse_node_readlink(struct puffs_usermount *pu, puffs_cookie_t opc,
2639 	const struct puffs_cred *pcr, char *linkname, size_t *linklen)
2640 {
2641 	struct perfuse_state *ps;
2642 	perfuse_msg_t *pm;
2643 	int error;
2644 	size_t len;
2645 	struct fuse_out_header *foh;
2646 
2647 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2648 		return ENOENT;
2649 
2650 	node_ref(opc);
2651 	ps = puffs_getspecific(pu);
2652 
2653 	pm = ps->ps_new_msg(pu, opc, FUSE_READLINK, 0, pcr);
2654 
2655 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2656 		goto out;
2657 
2658 	foh = GET_OUTHDR(ps, pm);
2659 	len = foh->len - sizeof(*foh);
2660 	if (len > *linklen)
2661 		DERRX(EX_PROTOCOL, "path len = %zd too long", len);
2662 	if (len == 0)
2663 		DERRX(EX_PROTOCOL, "path len = %zd too short", len);
2664 
2665 	(void)memcpy(linkname, _GET_OUTPAYLOAD(ps, pm, char *), len);
2666 
2667 	/*
2668 	 * FUSE filesystems return a NUL terminated string, we
2669 	 * do not want the trailing \0
2670 	 */
2671 	while (len > 0 && linkname[len - 1] == '\0')
2672 		len--;
2673 
2674 	*linklen = len;
2675 
2676 	ps->ps_destroy_msg(pm);
2677 	error = 0;
2678 
2679 out:
2680 	node_rele(opc);
2681 	return error;
2682 }
2683 
2684 int
2685 perfuse_node_reclaim(struct puffs_usermount *pu, puffs_cookie_t opc)
2686 {
2687 	struct perfuse_state *ps;
2688 	perfuse_msg_t *pm;
2689 	struct perfuse_node_data *pnd;
2690 	struct fuse_forget_in *ffi;
2691 	int nlookup;
2692 	struct timespec now;
2693 
2694 	if (opc == 0)
2695 		return 0;
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 	/*
2707 	 * There is a race condition between reclaim and lookup.
2708 	 * When looking up an already known node, the kernel cannot
2709 	 * hold a reference on the result until it gets the PUFFS
2710 	 * reply. It mayy therefore reclaim the node after the
2711 	 * userland looked it up, and before it gets the reply.
2712 	 * On rely, the kernel re-creates the node, but at that
2713 	 * time the node has been reclaimed in userland.
2714 	 *
2715 	 * In order to avoid this, we refuse reclaiming nodes that
2716 	 * are too young since the last lookup - and that we do
2717 	 * not have removed on our own, of course.
2718 	 */
2719 	if (clock_gettime(CLOCK_REALTIME, &now) != 0)
2720 		DERR(EX_OSERR, "clock_gettime failed");
2721 
2722 	if (timespeccmp(&pnd->pnd_cn_expire, &now, >) &&
2723 	    !(pnd->pnd_flags & PND_REMOVED)) {
2724 		if (!(pnd->pnd_flags & PND_NODELEAK)) {
2725 			ps->ps_nodeleakcount++;
2726 			pnd->pnd_flags |= PND_NODELEAK;
2727 		}
2728 		DWARNX("possible leaked node:: opc = %p \"%s\"",
2729 		       opc, pnd->pnd_name);
2730 		return 0;
2731 	}
2732 
2733 	node_ref(opc);
2734 	pnd->pnd_flags |= PND_RECLAIMED;
2735 	pnd->pnd_puffs_nlookup--;
2736 	nlookup = pnd->pnd_puffs_nlookup;
2737 
2738 #ifdef PERFUSE_DEBUG
2739 	if (perfuse_diagflags & PDF_RECLAIM)
2740 		DPRINTF("%s (nodeid %"PRId64") reclaimed\n",
2741 			perfuse_node_path(ps, opc), pnd->pnd_nodeid);
2742 #endif
2743 
2744 #ifdef PERFUSE_DEBUG
2745 	if (perfuse_diagflags & PDF_RECLAIM)
2746 		DPRINTF("%s (nodeid %"PRId64") is %sreclaimed, nlookup = %d "
2747 			"%s%s%s%s, pending ops:%s%s%s\n",
2748 		        perfuse_node_path(ps, opc), pnd->pnd_nodeid,
2749 		        pnd->pnd_flags & PND_RECLAIMED ? "" : "not ",
2750 			pnd->pnd_puffs_nlookup,
2751 			pnd->pnd_flags & PND_OPEN ? "open " : "not open",
2752 			pnd->pnd_flags & PND_RFH ? "r" : "",
2753 			pnd->pnd_flags & PND_WFH ? "w" : "",
2754 			pnd->pnd_flags & PND_BUSY ? "" : " none",
2755 			pnd->pnd_flags & PND_INREADDIR ? " readdir" : "",
2756 			pnd->pnd_flags & PND_INWRITE ? " write" : "",
2757 			pnd->pnd_flags & PND_INOPEN ? " open" : "");
2758 #endif
2759 	/*
2760 	 * Make sure it is not looked up again
2761 	 */
2762 	if (!(pnd->pnd_flags & PND_REMOVED))
2763 		perfuse_cache_flush(opc);
2764 
2765 	/*
2766 	 * Purge any activity on the node, while checking
2767 	 * that it remains eligible for a reclaim.
2768 	 */
2769 	while (pnd->pnd_ref > 1)
2770 		requeue_request(pu, opc, PCQ_REF);
2771 
2772 	/*
2773 	 * reclaim cancel?
2774 	 */
2775 	if (pnd->pnd_puffs_nlookup > nlookup) {
2776 		pnd->pnd_flags &= ~PND_RECLAIMED;
2777 		perfuse_node_cache(ps, opc);
2778 		node_rele(opc);
2779 		return 0;
2780 	}
2781 
2782 
2783 #ifdef PERFUSE_DEBUG
2784 	if ((pnd->pnd_flags & PND_OPEN) ||
2785 	       !TAILQ_EMPTY(&pnd->pnd_pcq))
2786 		DERRX(EX_SOFTWARE, "%s: opc = %p \"%s\": still open",
2787 		      __func__, opc, pnd->pnd_name);
2788 
2789 	if ((pnd->pnd_flags & PND_BUSY) ||
2790 	       !TAILQ_EMPTY(&pnd->pnd_pcq))
2791 		DERRX(EX_SOFTWARE, "%s: opc = %p: queued operations",
2792 		      __func__, opc);
2793 
2794 	if (pnd->pnd_inxchg != 0)
2795 		DERRX(EX_SOFTWARE, "%s: opc = %p: ongoing operations",
2796 		      __func__, opc);
2797 #endif
2798 
2799 	/*
2800 	 * Send the FORGET message
2801 	 *
2802 	 * ps_new_msg() is called with NULL creds, which will
2803 	 * be interpreted as FUSE superuser. This is obviously
2804 	 * fine since we operate with kernel creds here.
2805 	 */
2806 	pm = ps->ps_new_msg(pu, opc, FUSE_FORGET,
2807 		      sizeof(*ffi), NULL);
2808 	ffi = GET_INPAYLOAD(ps, pm, fuse_forget_in);
2809 	ffi->nlookup = pnd->pnd_fuse_nlookup;
2810 
2811 	/*
2812 	 * No reply is expected, pm is freed in xchg_msg
2813 	 */
2814 	(void)xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, no_reply);
2815 
2816 	perfuse_destroy_pn(pu, opc);
2817 
2818 	return 0;
2819 }
2820 
2821 int
2822 perfuse_node_inactive(struct puffs_usermount *pu, puffs_cookie_t opc)
2823 {
2824 	struct perfuse_node_data *pnd;
2825 	int error;
2826 
2827 	if (opc == 0)
2828 		return 0;
2829 
2830 	pnd = PERFUSE_NODE_DATA(opc);
2831 	if (!(pnd->pnd_flags & (PND_OPEN|PND_REMOVED)))
2832 		return 0;
2833 
2834 	node_ref(opc);
2835 
2836 	/*
2837 	 * Make sure all operation are finished
2838 	 * There can be an ongoing write. Other
2839 	 * operation wait for all data before
2840 	 * the close/inactive.
2841 	 */
2842 	while (pnd->pnd_flags & PND_INWRITE)
2843 		requeue_request(pu, opc, PCQ_AFTERWRITE);
2844 
2845 	/*
2846 	 * The inactive operation may be cancelled,
2847 	 * If no open is in progress, set PND_INOPEN
2848 	 * so that a new open will be queued.
2849 	 */
2850 	if (pnd->pnd_flags & PND_INOPEN)
2851 		goto out;
2852 
2853 	pnd->pnd_flags |= PND_INOPEN;
2854 
2855 	/*
2856 	 * Sync data
2857 	 */
2858 	if (pnd->pnd_flags & PND_DIRTY) {
2859 		if ((error = perfuse_node_fsync(pu, opc, NULL, 0, 0, 0)) != 0)
2860 			DWARN("%s: perfuse_node_fsync failed error = %d",
2861 			      __func__, error);
2862 	}
2863 
2864 
2865 	/*
2866 	 * Close handles
2867 	 */
2868 	if (pnd->pnd_flags & PND_WFH) {
2869 		if ((error = perfuse_node_close_common(pu, opc, FWRITE)) != 0)
2870 			DWARN("%s: close write FH failed error = %d",
2871 			      __func__, error);
2872 	}
2873 
2874 	if (pnd->pnd_flags & PND_RFH) {
2875 		if ((error = perfuse_node_close_common(pu, opc, FREAD)) != 0)
2876 			DWARN("%s: close read FH failed error = %d",
2877 			      __func__, error);
2878 	}
2879 
2880 	/*
2881 	 * This will cause a reclaim to be sent
2882 	 */
2883 	if (pnd->pnd_flags & PND_REMOVED)
2884 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N1);
2885 
2886 	/*
2887 	 * Schedule awaiting operations
2888 	 */
2889 	pnd->pnd_flags &= ~PND_INOPEN;
2890 	(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
2891 
2892 	/*
2893 	 * errors are ignored, since the kernel ignores the return code.
2894 	 */
2895 out:
2896 	node_rele(opc);
2897 	return 0;
2898 }
2899 
2900 
2901 /* ARGSUSED0 */
2902 int
2903 perfuse_node_print(struct puffs_usermount *pu, puffs_cookie_t opc)
2904 {
2905 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2906 	return 0;
2907 }
2908 
2909 int
2910 perfuse_node_pathconf(struct puffs_usermount *pu, puffs_cookie_t opc,
2911 	int name, register_t *retval)
2912 {
2913 	perfuse_msg_t *pm;
2914 	struct perfuse_state *ps;
2915 	struct fuse_statfs_out *fso;
2916 	int error = 0;
2917 
2918 	/*
2919 	 * Static values copied from UFS
2920 	 * in src/sys/ufs/ufs/ufs_vnops.c
2921 	 */
2922 	switch (name) {
2923 	case _PC_LINK_MAX:
2924 		*retval = LINK_MAX;
2925 		break;
2926 	case _PC_PATH_MAX:
2927 		*retval = PATH_MAX;
2928 		break;
2929 	case _PC_PIPE_BUF:
2930 		*retval = PIPE_BUF;
2931 		break;
2932 	case _PC_CHOWN_RESTRICTED:
2933 		*retval = 1;
2934 		break;
2935 	case _PC_NO_TRUNC:
2936 		*retval = 1;
2937 		break;
2938 	case _PC_SYNC_IO:
2939 		*retval = 1;
2940 		break;
2941 	case _PC_FILESIZEBITS:
2942 		*retval = 42;
2943 		break;
2944 	case _PC_SYMLINK_MAX:
2945 		*retval = MAXPATHLEN;
2946 		break;
2947 	case _PC_2_SYMLINKS:
2948 		*retval = 1;
2949 		break;
2950 	case _PC_NAME_MAX:
2951 		ps = puffs_getspecific(pu);
2952 		pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
2953 
2954 		error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply);
2955 		if (error != 0)
2956 			return error;
2957 
2958 		fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
2959 		*retval = fso->st.namelen;
2960 
2961 		ps->ps_destroy_msg(pm);
2962 
2963 		break;
2964 	default:
2965 		DWARN("Unimplemented pathconf for name = %d", name);
2966 		error = ENOSYS;
2967 		break;
2968 	}
2969 
2970 	return error;
2971 }
2972 
2973 int
2974 perfuse_node_advlock(struct puffs_usermount *pu, puffs_cookie_t opc,
2975 	void *id, int op, struct flock *fl, int flags)
2976 {
2977 	struct perfuse_state *ps;
2978 	int fop;
2979 	perfuse_msg_t *pm;
2980 	uint64_t fh;
2981 	struct fuse_lk_in *fli;
2982 	struct fuse_out_header *foh;
2983 	struct fuse_lk_out *flo;
2984 	uint32_t owner;
2985 	size_t len;
2986 	int error;
2987 
2988 	node_ref(opc);
2989 
2990 	/*
2991 	 * Make sure we do have a filehandle, as the FUSE filesystem
2992 	 * expect one. E.g.: if we provide none, GlusterFS logs an error
2993 	 * "0-glusterfs-fuse: xl is NULL"
2994 	 *
2995 	 * We need the read file handle if the file is open read only,
2996 	 * in order to support shared locks on read-only files.
2997 	 * NB: The kernel always sends advlock for read-only
2998 	 * files at exit time when the process used lock, see
2999 	 * sys_exit -> exit1 -> fd_free -> fd_close -> VOP_ADVLOCK
3000 	 */
3001 	if ((fh = perfuse_get_fh(opc, FREAD)) == FUSE_UNKNOWN_FH) {
3002 		error = EBADF;
3003 		goto out;
3004 	}
3005 
3006 	ps = puffs_getspecific(pu);
3007 
3008 	if (op == F_GETLK)
3009 		fop = FUSE_GETLK;
3010 	else
3011 		fop = (flags & F_WAIT) ? FUSE_SETLKW : FUSE_SETLK;
3012 
3013 	/*
3014 	 * XXX ps_new_msg() is called with NULL creds, which will
3015 	 * be interpreted as FUSE superuser. We have no way to
3016 	 * know the requesting process' credential, but since advlock()
3017 	 * is supposed to operate on a file that has been open(),
3018 	 * permission should have already been checked at open() time.
3019 	 */
3020 	pm = ps->ps_new_msg(pu, opc, fop, sizeof(*fli), NULL);
3021 	fli = GET_INPAYLOAD(ps, pm, fuse_lk_in);
3022 	fli->fh = fh;
3023 	fli->owner = (uint64_t)(vaddr_t)id;
3024 	fli->lk.start = fl->l_start;
3025 	fli->lk.end = fl->l_start + fl->l_len;
3026 	fli->lk.type = fl->l_type;
3027 	fli->lk.pid = fl->l_pid;
3028 	fli->lk_flags = (flags & F_FLOCK) ? FUSE_LK_FLOCK : 0;
3029 
3030 	owner = (uint32_t)(vaddr_t)id;
3031 
3032 #ifdef PERFUSE_DEBUG
3033 	if (perfuse_diagflags & PDF_FH)
3034 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
3035 			__func__, (void *)opc,
3036 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fli->fh);
3037 #endif
3038 
3039 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3040 		goto out;
3041 
3042 	foh = GET_OUTHDR(ps, pm);
3043 	len = foh->len - sizeof(*foh);
3044 
3045 	/*
3046 	 * Save or clear the lock
3047 	 */
3048 	switch (op) {
3049 	case F_GETLK:
3050 		if (len != sizeof(*flo))
3051 			DERRX(EX_SOFTWARE,
3052 			      "%s: Unexpected lock reply len %zd",
3053 			      __func__, len);
3054 
3055 		flo = GET_OUTPAYLOAD(ps, pm, fuse_lk_out);
3056 		fl->l_start = flo->lk.start;
3057 		fl->l_len = flo->lk.end - flo->lk.start;
3058 		fl->l_pid = flo->lk.pid;
3059 		fl->l_type = flo->lk.type;
3060 		fl->l_whence = SEEK_SET;	/* libfuse hardcodes it */
3061 
3062 		PERFUSE_NODE_DATA(opc)->pnd_lock_owner = flo->lk.pid;
3063 		break;
3064 	case F_UNLCK:
3065 		owner = 0;
3066 		/* FALLTHROUGH */
3067 	case F_SETLK:
3068 		/* FALLTHROUGH */
3069 	case F_SETLKW:
3070 		if (error != 0)
3071 			PERFUSE_NODE_DATA(opc)->pnd_lock_owner = owner;
3072 
3073 		if (len != 0)
3074 			DERRX(EX_SOFTWARE,
3075 			      "%s: Unexpected unlock reply len %zd",
3076 			      __func__, len);
3077 
3078 		break;
3079 	default:
3080 		DERRX(EX_SOFTWARE, "%s: Unexpected op %d", __func__, op);
3081 		break;
3082 	}
3083 
3084 	ps->ps_destroy_msg(pm);
3085 	error = 0;
3086 
3087 out:
3088 	node_rele(opc);
3089 	return error;
3090 }
3091 
3092 int
3093 perfuse_node_read(struct puffs_usermount *pu, puffs_cookie_t opc, uint8_t *buf,
3094 	off_t offset, size_t *resid, const struct puffs_cred *pcr, int ioflag)
3095 {
3096 	struct perfuse_state *ps;
3097 	struct perfuse_node_data *pnd;
3098 	const struct vattr *vap;
3099 	perfuse_msg_t *pm;
3100 	struct fuse_read_in *fri;
3101 	struct fuse_out_header *foh;
3102 	size_t readen;
3103 	int error;
3104 
3105 	ps = puffs_getspecific(pu);
3106 	pnd = PERFUSE_NODE_DATA(opc);
3107 	vap = puffs_pn_getvap((struct puffs_node *)opc);
3108 
3109 	/*
3110 	 * NetBSD turns that into a getdents(2) output
3111 	 * We just do a EISDIR as this feature is of little use.
3112 	 */
3113 	if (vap->va_type == VDIR)
3114 		return EISDIR;
3115 
3116 	do {
3117 		size_t max_read;
3118 
3119 		max_read = ps->ps_max_readahead - sizeof(*foh);
3120 		/*
3121 		 * flags may be set to FUSE_READ_LOCKOWNER
3122 		 * if lock_owner is provided.
3123 		 */
3124 		pm = ps->ps_new_msg(pu, opc, FUSE_READ, sizeof(*fri), pcr);
3125 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
3126 		fri->fh = perfuse_get_fh(opc, FREAD);
3127 		fri->offset = offset;
3128 		fri->size = (uint32_t)MIN(*resid, max_read);
3129 		fri->read_flags = 0; /* XXX Unused by libfuse? */
3130 		fri->lock_owner = pnd->pnd_lock_owner;
3131 		fri->flags = 0;
3132 		fri->flags |= (fri->lock_owner != 0) ? FUSE_READ_LOCKOWNER : 0;
3133 
3134 #ifdef PERFUSE_DEBUG
3135 	if (perfuse_diagflags & PDF_FH)
3136 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
3137 			__func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
3138 #endif
3139 		error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
3140 		if (error  != 0)
3141 			return error;
3142 
3143 		foh = GET_OUTHDR(ps, pm);
3144 		readen = foh->len - sizeof(*foh);
3145 
3146 #ifdef PERFUSE_DEBUG
3147 		if (readen > *resid)
3148 			DERRX(EX_SOFTWARE, "%s: Unexpected big read %zd",
3149 			      __func__, readen);
3150 #endif
3151 
3152 		(void)memcpy(buf,  _GET_OUTPAYLOAD(ps, pm, char *), readen);
3153 
3154 		buf += readen;
3155 		offset += readen;
3156 		*resid -= readen;
3157 
3158 		ps->ps_destroy_msg(pm);
3159 	} while ((*resid != 0) && (readen != 0));
3160 
3161 	if (ioflag & (IO_SYNC|IO_DSYNC))
3162 		ps->ps_syncreads++;
3163 	else
3164 		ps->ps_asyncreads++;
3165 
3166 	return 0;
3167 }
3168 
3169 int
3170 perfuse_node_write(struct puffs_usermount *pu, puffs_cookie_t opc,
3171 	uint8_t *buf, off_t offset, size_t *resid,
3172 	const struct puffs_cred *pcr, int ioflag)
3173 {
3174 	return perfuse_node_write2(pu, opc, buf, offset, resid, pcr, ioflag, 0);
3175 }
3176 
3177 /* ARGSUSED7 */
3178 int
3179 perfuse_node_write2(struct puffs_usermount *pu, puffs_cookie_t opc,
3180 	uint8_t *buf, off_t offset, size_t *resid,
3181 	const struct puffs_cred *pcr, int ioflag, int xflag)
3182 {
3183 	struct perfuse_state *ps;
3184 	struct perfuse_node_data *pnd;
3185 	struct vattr *vap;
3186 	perfuse_msg_t *pm;
3187 	struct fuse_write_in *fwi;
3188 	struct fuse_write_out *fwo;
3189 	size_t data_len;
3190 	size_t payload_len;
3191 	size_t written;
3192 	int inresize;
3193 	int error;
3194 
3195 	ps = puffs_getspecific(pu);
3196 	pnd = PERFUSE_NODE_DATA(opc);
3197 	vap = puffs_pn_getvap((struct puffs_node *)opc);
3198 	written = 0;
3199 	inresize = 0;
3200 	error = 0;
3201 
3202 	if (vap->va_type == VDIR)
3203 		return EISDIR;
3204 
3205 	node_ref(opc);
3206 
3207 	/*
3208 	 * We need to queue write requests in order to avoid
3209 	 * dequeueing PCQ_AFTERWRITE when there are pending writes.
3210 	 */
3211 	while (pnd->pnd_flags & PND_INWRITE)
3212 		requeue_request(pu, opc, PCQ_WRITE);
3213 	pnd->pnd_flags |= PND_INWRITE;
3214 
3215 	/*
3216 	 * append flag: re-read the file size so that
3217 	 * we get the latest value.
3218 	 */
3219 	if (ioflag & PUFFS_IO_APPEND) {
3220 		if ((error = perfuse_node_getattr(pu, opc, vap, pcr)) != 0)
3221 			goto out;
3222 
3223 		offset = vap->va_size;
3224 	}
3225 
3226 	/*
3227 	 * Serialize size access, see comment in perfuse_node_setattr().
3228 	 */
3229 	if ((u_quad_t)offset + *resid > vap->va_size) {
3230 		while (pnd->pnd_flags & PND_INRESIZE)
3231 			requeue_request(pu, opc, PCQ_RESIZE);
3232 		pnd->pnd_flags |= PND_INRESIZE;
3233 		inresize = 1;
3234 	}
3235 
3236 #ifdef PERFUSE_DEBUG
3237 	if (perfuse_diagflags & PDF_RESIZE)
3238 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__,
3239 			(void *)opc, vap->va_size);
3240 #endif
3241 
3242 	do {
3243 		size_t max_write;
3244 		/*
3245 		 * There is a writepage flag when data
3246 		 * is aligned to page size. Use it for
3247 		 * everything but the data after the last
3248 		 * page boundary.
3249 		 */
3250 		max_write = ps->ps_max_write - sizeof(*fwi);
3251 
3252 		data_len = MIN(*resid, max_write);
3253 		if (data_len > (size_t)sysconf(_SC_PAGESIZE))
3254 			data_len = data_len & ~(sysconf(_SC_PAGESIZE) - 1);
3255 
3256 		payload_len = data_len + sizeof(*fwi);
3257 
3258 		/*
3259 		 * flags may be set to FUSE_WRITE_CACHE (XXX usage?)
3260 		 * or FUSE_WRITE_LOCKOWNER, if lock_owner is provided.
3261 		 * write_flags is set to 1 for writepage.
3262 		 */
3263 		pm = ps->ps_new_msg(pu, opc, FUSE_WRITE, payload_len, pcr);
3264 		fwi = GET_INPAYLOAD(ps, pm, fuse_write_in);
3265 		fwi->fh = perfuse_get_fh(opc, FWRITE);
3266 		fwi->offset = offset;
3267 		fwi->size = (uint32_t)data_len;
3268 		fwi->write_flags = (fwi->size % sysconf(_SC_PAGESIZE)) ? 0 : 1;
3269 		fwi->lock_owner = pnd->pnd_lock_owner;
3270 		fwi->flags = 0;
3271 		fwi->flags |= (fwi->lock_owner != 0) ? FUSE_WRITE_LOCKOWNER : 0;
3272 		fwi->flags |= (ioflag & IO_DIRECT) ? 0 : FUSE_WRITE_CACHE;
3273 		(void)memcpy((fwi + 1), buf, data_len);
3274 
3275 
3276 #ifdef PERFUSE_DEBUG
3277 		if (perfuse_diagflags & PDF_FH)
3278 			DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
3279 				"fh = 0x%"PRIx64"\n", __func__,
3280 				(void *)opc, pnd->pnd_nodeid, fwi->fh);
3281 #endif
3282 		if ((error = xchg_msg(pu, opc, pm,
3283 				      sizeof(*fwo), wait_reply)) != 0)
3284 			goto out;
3285 
3286 		fwo = GET_OUTPAYLOAD(ps, pm, fuse_write_out);
3287 		written = fwo->size;
3288 		ps->ps_destroy_msg(pm);
3289 
3290 #ifdef PERFUSE_DEBUG
3291 		if (written > *resid)
3292 			DERRX(EX_SOFTWARE, "%s: Unexpected big write %zd",
3293 			      __func__, written);
3294 #endif
3295 		*resid -= written;
3296 		offset += written;
3297 		buf += written;
3298 
3299 	} while (*resid != 0);
3300 
3301 	/*
3302 	 * puffs_ops(3) says
3303 	 *  "everything must be written or an error will be generated"
3304 	 */
3305 	if (*resid != 0)
3306 		error = EFBIG;
3307 
3308 out:
3309 #ifdef PERFUSE_DEBUG
3310 	if (perfuse_diagflags & PDF_RESIZE) {
3311 		if (offset > (off_t)vap->va_size)
3312 			DPRINTF("<< %s %p %" PRIu64 " -> %lld\n", __func__,
3313 				(void *)opc, vap->va_size, (long long)offset);
3314 		else
3315 			DPRINTF("<< %s %p \n", __func__, (void *)opc);
3316 	}
3317 #endif
3318 
3319 	/*
3320 	 * Update file size if we wrote beyond the end
3321 	 */
3322 	if (offset > (off_t)vap->va_size)
3323 		vap->va_size = offset;
3324 
3325 	/*
3326 	 * Statistics
3327 	 */
3328 	if (ioflag & (IO_SYNC|IO_DSYNC))
3329 		ps->ps_syncwrites++;
3330 	else
3331 		ps->ps_asyncwrites++;
3332 
3333 	/*
3334 	 * Remember to sync the file
3335 	 */
3336 	pnd->pnd_flags |= PND_DIRTY;
3337 
3338 #ifdef PERFUSE_DEBUG
3339 	if (perfuse_diagflags & PDF_SYNC)
3340 		DPRINTF("%s: DIRTY opc = %p, file = \"%s\"\n",
3341 			__func__, (void*)opc, perfuse_node_path(ps, opc));
3342 #endif
3343 
3344 	if (inresize) {
3345 #ifdef PERFUSE_DEBUG
3346 		if (!(pnd->pnd_flags & PND_INRESIZE))
3347 			DERRX(EX_SOFTWARE, "file write grow without resize");
3348 #endif
3349 		pnd->pnd_flags &= ~PND_INRESIZE;
3350 		(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
3351 	}
3352 
3353 	/*
3354 	 * VOP_PUTPAGE causes FAF write where kernel does not
3355 	 * check operation result. At least warn if it failed.
3356 	 */
3357 #ifdef PUFFS_WRITE_FAF
3358 	if (error && (xflag & PUFFS_WRITE_FAF))
3359 		DWARN("Data loss caused by FAF write failed on \"%s\"",
3360 		      pnd->pnd_name);
3361 #endif /* PUFFS_WRITE_FAF */
3362 
3363 	/*
3364 	 * If there are no more queued write, we can resume
3365 	 * an operation awaiting write completion.
3366 	 */
3367 	pnd->pnd_flags &= ~PND_INWRITE;
3368 	if (dequeue_requests(opc, PCQ_WRITE, 1) == 0)
3369 		(void)dequeue_requests(opc, PCQ_AFTERWRITE, DEQUEUE_ALL);
3370 
3371 	node_rele(opc);
3372 	return error;
3373 }
3374 
3375 /* ARGSUSED0 */
3376 void
3377 perfuse_cache_write(struct puffs_usermount *pu, puffs_cookie_t opc, size_t size,
3378 	struct puffs_cacherun *runs)
3379 {
3380 	return;
3381 }
3382 
3383 /* ARGSUSED4 */
3384 int
3385 perfuse_node_getextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3386 	int attrns, const char *attrname, size_t *attrsize, uint8_t *attr,
3387 	size_t *resid, const struct puffs_cred *pcr)
3388 {
3389 	struct perfuse_state *ps;
3390 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3391 	perfuse_msg_t *pm;
3392 	struct fuse_getxattr_in *fgi;
3393 	struct fuse_getxattr_out *fgo;
3394 	struct fuse_out_header *foh;
3395 	size_t attrnamelen;
3396 	size_t len;
3397 	char *np;
3398 	int error;
3399 
3400 	/* system namespace attrs are not accessible to non root users */
3401 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
3402 		return EPERM;
3403 
3404 	node_ref(opc);
3405 	ps = puffs_getspecific(pu);
3406 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3407 	attrnamelen = strlen(attrname) + 1;
3408 	len = sizeof(*fgi) + attrnamelen;
3409 
3410 	pm = ps->ps_new_msg(pu, opc, FUSE_GETXATTR, len, pcr);
3411 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3412 	fgi->size = (unsigned int)((resid != NULL) ? *resid : 0);
3413 	np = (char *)(void *)(fgi + 1);
3414 	(void)strlcpy(np, attrname, attrnamelen);
3415 
3416 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3417 		goto out;
3418 
3419 	/*
3420 	 * We just get fuse_getattr_out with list size if we requested
3421 	 * a null size.
3422 	 */
3423 	if (resid == NULL) {
3424 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3425 
3426 		if (attrsize != NULL)
3427 			*attrsize = fgo->size;
3428 
3429 		ps->ps_destroy_msg(pm);
3430 		error = 0;
3431 		goto out;
3432 	}
3433 
3434 	/*
3435 	 * And with a non null requested size, we get the list just
3436 	 * after the header
3437 	 */
3438 	foh = GET_OUTHDR(ps, pm);
3439 	np = (char *)(void *)(foh + 1);
3440 	len = foh->len - sizeof(*foh);
3441 
3442 	if (attrsize != NULL)
3443 		*attrsize = len;
3444 
3445 	if (resid != NULL) {
3446 		if (*resid < len) {
3447 			error = ERANGE;
3448 			ps->ps_destroy_msg(pm);
3449 			goto out;
3450 		}
3451 
3452 		(void)memcpy(attr, np, len);
3453 		*resid -= len;
3454 	}
3455 
3456 	ps->ps_destroy_msg(pm);
3457 	error = 0;
3458 
3459 out:
3460 	node_rele(opc);
3461 	return error;
3462 }
3463 
3464 int
3465 perfuse_node_setextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3466 	int attrns, const char *attrname, uint8_t *attr, size_t *resid,
3467 	const struct puffs_cred *pcr)
3468 {
3469 	struct perfuse_state *ps;
3470 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3471 	perfuse_msg_t *pm;
3472 	struct fuse_setxattr_in *fsi;
3473 	size_t attrnamelen;
3474 	size_t datalen;
3475 	size_t len;
3476 	char *np;
3477 	int error;
3478 
3479 	/* system namespace attrs are not accessible to non root users */
3480 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
3481 		return EPERM;
3482 
3483 	node_ref(opc);
3484 	ps = puffs_getspecific(pu);
3485 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3486 	attrnamelen = strlen(attrname) + 1;
3487 
3488 	datalen = (resid != NULL) ? *resid : 0;
3489 	len = sizeof(*fsi) + attrnamelen + datalen;
3490 
3491 	pm = ps->ps_new_msg(pu, opc, FUSE_SETXATTR, len, pcr);
3492 	fsi = GET_INPAYLOAD(ps, pm, fuse_setxattr_in);
3493 	fsi->size = (unsigned int)datalen;
3494 	fsi->flags = 0;
3495 	np = (char *)(void *)(fsi + 1);
3496 	(void)strlcpy(np, attrname, attrnamelen);
3497 	np += attrnamelen;
3498 	if (datalen)
3499 		(void)memcpy(np, (char *)attr, datalen);
3500 
3501 	if ((error = xchg_msg(pu, opc, pm,
3502 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
3503 		goto out;
3504 
3505 	ps->ps_destroy_msg(pm);
3506 	if (resid)
3507 		*resid = 0;
3508 	error = 0;
3509 
3510 out:
3511 	node_rele(opc);
3512 	return error;
3513 }
3514 
3515 /* ARGSUSED2 */
3516 int
3517 perfuse_node_listextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3518 	int attrns, size_t *attrsize, uint8_t *attrs, size_t *resid, int flag,
3519 	const struct puffs_cred *pcr)
3520 {
3521 	struct perfuse_state *ps;
3522 	perfuse_msg_t *pm;
3523 	struct fuse_getxattr_in *fgi;
3524 	struct fuse_getxattr_out *fgo;
3525 	struct fuse_out_header *foh;
3526 	char *np;
3527 	size_t len, puffs_len, i, attrlen, outlen;
3528 	int error;
3529 
3530 	/* system namespace attrs are not accessible to non root users */
3531 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
3532 		return EPERM;
3533 
3534 	node_ref(opc);
3535 
3536 	ps = puffs_getspecific(pu);
3537 	len = sizeof(*fgi);
3538 
3539 	pm = ps->ps_new_msg(pu, opc, FUSE_LISTXATTR, len, pcr);
3540 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3541 	if (resid != NULL)
3542 		fgi->size = (unsigned int)*resid;
3543 	else
3544 		fgi->size = 0;
3545 
3546 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3547 		goto out;
3548 
3549 	/*
3550 	 * We just get fuse_getattr_out with list size if we requested
3551 	 * a null size.
3552 	 */
3553 	if (resid == NULL) {
3554 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3555 
3556 		if (attrsize != NULL)
3557 			*attrsize = fgo->size;
3558 
3559 		ps->ps_destroy_msg(pm);
3560 
3561 		error = 0;
3562 		goto out;
3563 	}
3564 
3565 	/*
3566 	 * And with a non null requested size, we get the list just
3567 	 * after the header
3568 	 */
3569 	foh = GET_OUTHDR(ps, pm);
3570 	np = (char *)(void *)(foh + 1);
3571 	puffs_len = foh->len - sizeof(*foh);
3572 
3573 	if (attrsize != NULL)
3574 		*attrsize = puffs_len;
3575 
3576 	if (attrs != NULL) {
3577 		if (*resid < puffs_len) {
3578 			error = ERANGE;
3579 			ps->ps_destroy_msg(pm);
3580 			goto out;
3581 		}
3582 
3583 		outlen = 0;
3584 
3585 		for (i = 0; i < puffs_len; i += attrlen + 1) {
3586 			attrlen = strlen(np + i);
3587 
3588 			/*
3589 			 * Filter attributes per namespace
3590 			 */
3591 			if (!perfuse_ns_match(attrns, np + i))
3592 				continue;
3593 
3594 #ifdef PUFFS_EXTATTR_LIST_LENPREFIX
3595 			/*
3596 			 * Convert the FUSE reply to length prefixed strings
3597 			 * if this is what the kernel wants.
3598 			 */
3599 			if (flag & PUFFS_EXTATTR_LIST_LENPREFIX) {
3600 				(void)memcpy(attrs + outlen + 1,
3601 					     np + i, attrlen);
3602 				*(attrs + outlen) = (uint8_t)attrlen;
3603 			} else
3604 #endif /* PUFFS_EXTATTR_LIST_LENPREFIX */
3605 			(void)memcpy(attrs + outlen, np + i, attrlen + 1);
3606 			outlen += attrlen + 1;
3607 		}
3608 
3609 		*resid -= outlen;
3610 	}
3611 
3612 	ps->ps_destroy_msg(pm);
3613 	error = 0;
3614 
3615 out:
3616 	node_rele(opc);
3617 	return error;
3618 }
3619 
3620 int
3621 perfuse_node_deleteextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3622 	int attrns, const char *attrname, const struct puffs_cred *pcr)
3623 {
3624 	struct perfuse_state *ps;
3625 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3626 	perfuse_msg_t *pm;
3627 	size_t attrnamelen;
3628 	char *np;
3629 	int error;
3630 
3631 	/* system namespace attrs are not accessible to non root users */
3632 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
3633 		return EPERM;
3634 
3635 	node_ref(opc);
3636 
3637 	ps = puffs_getspecific(pu);
3638 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3639 	attrnamelen = strlen(attrname) + 1;
3640 
3641 	pm = ps->ps_new_msg(pu, opc, FUSE_REMOVEXATTR, attrnamelen, pcr);
3642 	np = _GET_INPAYLOAD(ps, pm, char *);
3643 	(void)strlcpy(np, attrname, attrnamelen);
3644 
3645 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
3646 	if (error != 0)
3647 		goto out;
3648 
3649 	ps->ps_destroy_msg(pm);
3650 
3651 out:
3652 	node_rele(opc);
3653 	return error;
3654 }
3655 
3656 int
3657 perfuse_node_fallocate(struct puffs_usermount *pu, puffs_cookie_t opc,
3658 	off_t off, off_t len)
3659 {
3660 	struct perfuse_state *ps;
3661 	perfuse_msg_t *pm;
3662 	struct fuse_fallocate_in *fai;
3663 	int error;
3664 
3665 	ps = puffs_getspecific(pu);
3666 	if (ps->ps_flags & PS_NO_FALLOCATE)
3667 		return EOPNOTSUPP;
3668 
3669 	node_ref(opc);
3670 
3671 	pm = ps->ps_new_msg(pu, opc, FUSE_FALLOCATE, sizeof(*fai), NULL);
3672 
3673 	fai = GET_INPAYLOAD(ps, pm, fuse_fallocate_in);
3674 	fai->fh = perfuse_get_fh(opc, FWRITE);
3675 	fai->offset = off;
3676 	fai->length = len;
3677 	fai->mode = 0;
3678 
3679 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
3680 	if (error == EOPNOTSUPP || error == ENOSYS) {
3681 		ps->ps_flags |= PS_NO_FALLOCATE;
3682 		error = EOPNOTSUPP;
3683 	}
3684 	if (error != 0)
3685 		goto out;
3686 
3687 	ps->ps_destroy_msg(pm);
3688 
3689 out:
3690 	node_rele(opc);
3691 	return error;
3692 }
3693