xref: /netbsd-src/lib/libperfuse/ops.c (revision 230b95665bbd3a9d1a53658a36b1053f8382a519)
1 /*  $NetBSD: ops.c,v 1.71 2014/09/03 16:01:45 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->size;
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 
611 		reclen = _DIRENT_RECLEN(dents, fd->namelen);
612 
613 		/*
614 		 * Check we do not overflow the output buffer
615 		 * struct fuse_dirent is bigger than struct dirent,
616 		 * so we should always use fd_len and never reallocate
617 		 * later.
618 		 * If we have to reallocate,try to double the buffer
619 		 * each time so that we do not have to do it too often.
620 		 */
621 		if (written + reclen > dents_len) {
622 			if (dents_len == 0)
623 				dents_len = fd_len;
624 			else
625 				dents_len =
626 				   MAX(2 * dents_len, written + reclen);
627 
628 			dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
629 			if ((dents = realloc(dents, dents_len)) == NULL)
630 				DERR(EX_OSERR, "%s: malloc failed", __func__);
631 
632 			PERFUSE_NODE_DATA(opc)->pnd_dirent = dents;
633 			PERFUSE_NODE_DATA(opc)->pnd_dirent_len = dents_len;
634 
635 			/*
636 			 * (void *) for delint
637 			 */
638 			ndp = (char *)(void *)dents + written;
639 			dents = (struct dirent *)(void *)ndp;
640 		}
641 
642 		/*
643 		 * Filesystem was mounted without -o use_ino
644 		 * Perform a lookup to find it.
645 		 */
646 		if (fd->ino == PERFUSE_UNKNOWN_INO) {
647 			struct puffs_node *pn;
648 			struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
649 
650 			/*
651 			 * Avoid breaking out of fs
652 			 * by lookup to .. on root
653 			 */
654 			if ((strcmp(fd->name, "..") == 0) &&
655 			    (pnd->pnd_nodeid == FUSE_ROOT_ID)) {
656 				fd->ino = FUSE_ROOT_ID;
657 			} else {
658 				if (node_lookup_common(pu, opc, NULL, fd->name,
659 						       NULL, &pn) != 0) {
660 					DWARNX("node_lookup_common failed");
661 				} else {
662 					fd->ino = pn->pn_va.va_fileid;
663 					(void)perfuse_node_reclaim(pu, pn);
664 				}
665 			}
666 		}
667 
668 		dents->d_fileno = fd->ino;
669 		dents->d_reclen = (unsigned short)reclen;
670 		dents->d_namlen = fd->namelen;
671 		dents->d_type = fd->type;
672 		strlcpy(dents->d_name, fd->name, fd->namelen + 1);
673 
674 #ifdef PERFUSE_DEBUG
675 		if (perfuse_diagflags & PDF_READDIR)
676 			DPRINTF("%s: translated \"%s\" ino = %"PRIu64"\n",
677 				__func__, dents->d_name, dents->d_fileno);
678 #endif
679 
680 		dents = _DIRENT_NEXT(dents);
681 		written += reclen;
682 
683 		/*
684 		 * Move to the next record.
685 		 * fd->off is not the offset, it is an opaque cookie
686 		 * given by the filesystem to keep state across multiple
687 		 * readdir() operation.
688 		 * Use record alignement instead.
689 		 */
690 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
691 #ifdef PERFUSE_DEBUG
692 		if (perfuse_diagflags & PDF_READDIR)
693 			DPRINTF("%s: record at %"PRId64"/0x%"PRIx64" "
694 				"length = %zd/0x%zx. "
695 				"next record at %"PRId64"/0x%"PRIx64" "
696 				"max %zd/0x%zx\n",
697 				__func__, fd_offset, fd_offset, len, len,
698 				fd_offset + len, fd_offset + len,
699 				fd_len, fd_len);
700 #endif
701 		fd_offset += len;
702 
703 		/*
704 		 * Check if next record is still within the packet
705 		 * If it is not, we reached the end of the buffer.
706 		 */
707 		if (fd_offset >= fd_len)
708 			break;
709 
710 		/*
711 		 * (void *) for delint
712 		 */
713 		ndp = (char *)(void *)fd_base + (size_t)fd_offset;
714 		fd = (struct fuse_dirent *)(void *)ndp;
715 
716 	} while (1 /* CONSTCOND */);
717 
718 	/*
719 	 * Adjust the dirent output length
720 	 */
721 	if (written != -1)
722 		PERFUSE_NODE_DATA(opc)->pnd_dirent_len = written;
723 
724 	return written;
725 }
726 
727 static void
728 readdir_buffered(puffs_cookie_t opc, struct dirent *dent, off_t *readoff,
729 	size_t *reslen)
730 {
731 	struct dirent *fromdent;
732 	struct perfuse_node_data *pnd;
733 	char *ndp;
734 
735 	pnd = PERFUSE_NODE_DATA(opc);
736 
737 	while (*readoff < pnd->pnd_dirent_len) {
738 		/*
739 		 * (void *) for delint
740 		 */
741 		ndp = (char *)(void *)pnd->pnd_dirent + (size_t)*readoff;
742 		fromdent = (struct dirent *)(void *)ndp;
743 
744 		if (*reslen < _DIRENT_SIZE(fromdent))
745 			break;
746 
747 		memcpy(dent, fromdent, _DIRENT_SIZE(fromdent));
748 		*readoff += _DIRENT_SIZE(fromdent);
749 		*reslen -= _DIRENT_SIZE(fromdent);
750 
751 		dent = _DIRENT_NEXT(dent);
752 	}
753 
754 #ifdef PERFUSE_DEBUG
755 	if (perfuse_diagflags & PDF_READDIR)
756 		DPRINTF("%s: readoff = %"PRId64",  "
757 			"pnd->pnd_dirent_len = %"PRId64"\n",
758 			__func__, *readoff, pnd->pnd_dirent_len);
759 #endif
760 	if (*readoff >=  pnd->pnd_dirent_len) {
761 		free(pnd->pnd_dirent);
762 		pnd->pnd_dirent = NULL;
763 		pnd->pnd_dirent_len = 0;
764 	}
765 
766 	return;
767 }
768 
769 
770 static void
771 node_ref(puffs_cookie_t opc)
772 {
773 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
774 
775 #ifdef PERFUSE_DEBUG
776 	if (pnd->pnd_flags & PND_INVALID)
777 		DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
778 #endif /* PERFUSE_DEBUG */
779 
780 	pnd->pnd_ref++;
781 	return;
782 }
783 
784 static void
785 node_rele(puffs_cookie_t opc)
786 {
787 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
788 
789 #ifdef PERFUSE_DEBUG
790 	if (pnd->pnd_flags & PND_INVALID)
791 		DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
792 #endif /* PERFUSE_DEBUG */
793 
794 	pnd->pnd_ref--;
795 
796 	if (pnd->pnd_ref == 0)
797 		(void)dequeue_requests(opc, PCQ_REF, DEQUEUE_ALL);
798 
799 	return;
800 }
801 
802 static void
803 requeue_request(struct puffs_usermount *pu, puffs_cookie_t opc,
804 	enum perfuse_qtype type)
805 {
806 	struct perfuse_cc_queue pcq;
807 	struct perfuse_node_data *pnd;
808 #ifdef PERFUSE_DEBUG
809 	struct perfuse_state *ps;
810 
811 	ps = perfuse_getspecific(pu);
812 #endif
813 
814 	pnd = PERFUSE_NODE_DATA(opc);
815 	pcq.pcq_type = type;
816 	pcq.pcq_cc = puffs_cc_getcc(pu);
817 	TAILQ_INSERT_TAIL(&pnd->pnd_pcq, &pcq, pcq_next);
818 
819 #ifdef PERFUSE_DEBUG
820 	if (perfuse_diagflags & PDF_REQUEUE)
821 		DPRINTF("%s: REQUEUE opc = %p, pcc = %p (%s)\n",
822 		        __func__, (void *)opc, pcq.pcq_cc,
823 			perfuse_qtypestr[type]);
824 #endif
825 
826 	puffs_cc_yield(pcq.pcq_cc);
827 	TAILQ_REMOVE(&pnd->pnd_pcq, &pcq, pcq_next);
828 
829 #ifdef PERFUSE_DEBUG
830 	if (perfuse_diagflags & PDF_REQUEUE)
831 		DPRINTF("%s: RESUME opc = %p, pcc = %p (%s)\n",
832 		        __func__, (void *)opc, pcq.pcq_cc,
833 			perfuse_qtypestr[type]);
834 #endif
835 
836 	return;
837 }
838 
839 static int
840 dequeue_requests(puffs_cookie_t opc, enum perfuse_qtype type, int max)
841 {
842 	struct perfuse_cc_queue *pcq;
843 	struct perfuse_node_data *pnd;
844 	int dequeued;
845 
846 	pnd = PERFUSE_NODE_DATA(opc);
847 	dequeued = 0;
848 	TAILQ_FOREACH(pcq, &pnd->pnd_pcq, pcq_next) {
849 		if (pcq->pcq_type != type)
850 			continue;
851 
852 #ifdef PERFUSE_DEBUG
853 		if (perfuse_diagflags & PDF_REQUEUE)
854 			DPRINTF("%s: SCHEDULE opc = %p, pcc = %p (%s)\n",
855 				__func__, (void *)opc, pcq->pcq_cc,
856 				 perfuse_qtypestr[type]);
857 #endif
858 		puffs_cc_schedule(pcq->pcq_cc);
859 
860 		if (++dequeued == max)
861 			break;
862 	}
863 
864 #ifdef PERFUSE_DEBUG
865 	if (perfuse_diagflags & PDF_REQUEUE)
866 		DPRINTF("%s: DONE  opc = %p\n", __func__, (void *)opc);
867 #endif
868 
869 	return dequeued;
870 }
871 
872 void
873 perfuse_fs_init(struct puffs_usermount *pu)
874 {
875 	struct perfuse_state *ps;
876 	perfuse_msg_t *pm;
877 	struct fuse_init_in *fii;
878 	struct fuse_init_out *fio;
879 	int error;
880 
881 	ps = puffs_getspecific(pu);
882 
883         if (puffs_mount(pu, ps->ps_target, ps->ps_mountflags, ps->ps_root) != 0)
884                 DERR(EX_OSERR, "%s: puffs_mount failed", __func__);
885 
886 	/*
887 	 * Linux 2.6.34.1 sends theses flags:
888 	 * FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC
889 	 * FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK
890 	 *
891 	 * Linux also sets max_readahead at 32 pages (128 kB)
892 	 *
893 	 * ps_new_msg() is called with NULL creds, which will
894 	 * be interpreted as FUSE superuser.
895 	 */
896 	pm = ps->ps_new_msg(pu, 0, FUSE_INIT, sizeof(*fii), NULL);
897 	fii = GET_INPAYLOAD(ps, pm, fuse_init_in);
898 	fii->major = FUSE_KERNEL_VERSION;
899 	fii->minor = FUSE_KERNEL_MINOR_VERSION;
900 	fii->max_readahead = (unsigned int)(32 * sysconf(_SC_PAGESIZE));
901 	fii->flags = (FUSE_ASYNC_READ|FUSE_POSIX_LOCKS|FUSE_ATOMIC_O_TRUNC);
902 
903 	if ((error = xchg_msg(pu, 0, pm, sizeof(*fio), wait_reply)) != 0)
904 		DERRX(EX_SOFTWARE, "init message exchange failed (%d)", error);
905 
906 	fio = GET_OUTPAYLOAD(ps, pm, fuse_init_out);
907 	ps->ps_max_readahead = fio->max_readahead;
908 	ps->ps_max_write = fio->max_write;
909 
910 	ps->ps_destroy_msg(pm);
911 
912 	return;
913 }
914 
915 int
916 perfuse_fs_unmount(struct puffs_usermount *pu, int flags)
917 {
918 	perfuse_msg_t *pm;
919 	struct perfuse_state *ps;
920 	puffs_cookie_t opc;
921 	int error;
922 
923 	ps = puffs_getspecific(pu);
924 	opc = (puffs_cookie_t)puffs_getroot(pu);
925 
926 	/*
927 	 * ps_new_msg() is called with NULL creds, which will
928 	 * be interpreted as FUSE superuser.
929 	 */
930 	pm = ps->ps_new_msg(pu, opc, FUSE_DESTROY, 0, NULL);
931 
932 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0){
933 		DWARN("unmount %s", ps->ps_target);
934 		if (!(flags & MNT_FORCE))
935 			return error;
936 		else
937 			error = 0;
938 	} else {
939 		ps->ps_destroy_msg(pm);
940 	}
941 
942 	ps->ps_umount(pu);
943 
944 	if (perfuse_diagflags & PDF_MISC)
945 		DPRINTF("%s unmounted, exit\n", ps->ps_target);
946 
947 	return 0;
948 }
949 
950 int
951 perfuse_fs_statvfs(struct puffs_usermount *pu, struct statvfs *svfsb)
952 {
953 	struct perfuse_state *ps;
954 	perfuse_msg_t *pm;
955 	puffs_cookie_t opc;
956 	struct fuse_statfs_out *fso;
957 	int error;
958 
959 	ps = puffs_getspecific(pu);
960 	opc = (puffs_cookie_t)puffs_getroot(pu);
961 
962 	/*
963 	 * ps_new_msg() is called with NULL creds, which will
964 	 * be interpreted as FUSE superuser.
965 	 */
966 	pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
967 
968 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply)) != 0)
969 		return error;
970 
971 	fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
972 	svfsb->f_flag = ps->ps_mountflags;
973 	svfsb->f_bsize = fso->st.bsize;
974 	svfsb->f_frsize = fso->st.frsize;
975 	svfsb->f_iosize = ((struct puffs_node *)opc)->pn_va.va_blocksize;
976 	svfsb->f_blocks = fso->st.blocks;
977 	svfsb->f_bfree = fso->st.bfree;
978 	svfsb->f_bavail = fso->st.bavail;
979 	svfsb->f_bresvd = fso->st.bfree - fso->st.bavail;
980 	svfsb->f_files = fso->st.files;
981 	svfsb->f_ffree = fso->st.ffree;
982 	svfsb->f_favail = fso->st.ffree;/* files not reserved for root */
983 	svfsb->f_fresvd = 0;		/* files reserved for root */
984 
985 	svfsb->f_syncreads = ps->ps_syncreads;
986 	svfsb->f_syncwrites = ps->ps_syncwrites;
987 
988 	svfsb->f_asyncreads = ps->ps_asyncreads;
989 	svfsb->f_asyncwrites = ps->ps_asyncwrites;
990 
991 	(void)memcpy(&svfsb->f_fsidx, &ps->ps_fsid, sizeof(ps->ps_fsid));
992 	svfsb->f_fsid = (unsigned long)ps->ps_fsid;
993 	svfsb->f_namemax = MAXPATHLEN;	/* XXX */
994 	svfsb->f_owner = ps->ps_owner_uid;
995 
996 	(void)strlcpy(svfsb->f_mntonname, ps->ps_target, _VFS_NAMELEN);
997 
998 	if (ps->ps_filesystemtype != NULL)
999 		(void)strlcpy(svfsb->f_fstypename,
1000 			      ps->ps_filesystemtype, _VFS_NAMELEN);
1001 	else
1002 		(void)strlcpy(svfsb->f_fstypename, "fuse", _VFS_NAMELEN);
1003 
1004 	if (ps->ps_source != NULL)
1005 		strlcpy(svfsb->f_mntfromname, ps->ps_source, _VFS_NAMELEN);
1006 	else
1007 		strlcpy(svfsb->f_mntfromname, _PATH_FUSE, _VFS_NAMELEN);
1008 
1009 	ps->ps_destroy_msg(pm);
1010 
1011 	return 0;
1012 }
1013 
1014 int
1015 perfuse_fs_sync(struct puffs_usermount *pu, int waitfor,
1016 	const struct puffs_cred *pcr)
1017 {
1018 	/*
1019 	 * FUSE does not seem to have a FS sync callback.
1020 	 * Maybe do not even register this callback
1021 	 */
1022 	return puffs_fsnop_sync(pu, waitfor, pcr);
1023 }
1024 
1025 /* ARGSUSED0 */
1026 int
1027 perfuse_fs_fhtonode(struct puffs_usermount *pu, void *fid, size_t fidsize,
1028 	struct puffs_newinfo *pni)
1029 {
1030 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1031 	return 0;
1032 }
1033 
1034 /* ARGSUSED0 */
1035 int
1036 perfuse_fs_nodetofh(struct puffs_usermount *pu, puffs_cookie_t cookie,
1037 	void *fid, size_t *fidsize)
1038 {
1039 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1040 	return 0;
1041 }
1042 
1043 #if 0
1044 /* ARGSUSED0 */
1045 void
1046 perfuse_fs_extattrctl(struct puffs_usermount *pu, int cmd,
1047 	puffs_cookie_t *cookie, int flags, int namespace, const char *attrname)
1048 {
1049 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1050 	return 0;
1051 }
1052 #endif /* 0 */
1053 
1054 /* ARGSUSED0 */
1055 void
1056 perfuse_fs_suspend(struct puffs_usermount *pu, int status)
1057 {
1058 	return;
1059 }
1060 
1061 
1062 int
1063 perfuse_node_lookup(struct puffs_usermount *pu, puffs_cookie_t opc,
1064 	struct puffs_newinfo *pni, const struct puffs_cn *pcn)
1065 {
1066 	struct perfuse_state *ps;
1067 	struct puffs_node *pn;
1068 	mode_t mode;
1069 	int error;
1070 
1071 	ps = puffs_getspecific(pu);
1072 	node_ref(opc);
1073 
1074 	/*
1075 	 * Check permissions
1076 	 */
1077 	switch(pcn->pcn_nameiop) {
1078 	case NAMEI_DELETE: /* FALLTHROUGH */
1079 	case NAMEI_RENAME: /* FALLTHROUGH */
1080 	case NAMEI_CREATE:
1081 		if (pcn->pcn_flags & NAMEI_ISLASTCN)
1082 			mode = PUFFS_VEXEC|PUFFS_VWRITE;
1083 		else
1084 			mode = PUFFS_VEXEC;
1085 		break;
1086 	case NAMEI_LOOKUP: /* FALLTHROUGH */
1087 	default:
1088 		mode = PUFFS_VEXEC;
1089 		break;
1090 	}
1091 
1092 	if ((error = mode_access(opc, pcn->pcn_cred, mode)) != 0)
1093 		goto out;
1094 
1095 	error = node_lookup_common(pu, (puffs_cookie_t)opc, pni,
1096 				   pcn->pcn_name, pcn->pcn_cred, &pn);
1097 
1098 	if (error != 0)
1099 		goto out;
1100 
1101 	/*
1102 	 * Kernel would kill us if the filesystem returned the parent
1103 	 * itself. If we want to live, hide that!
1104 	 */
1105 	if ((opc == (puffs_cookie_t)pn) && (strcmp(pcn->pcn_name, ".") != 0)) {
1106 		DERRX(EX_SOFTWARE, "lookup \"%s\" in \"%s\" returned parent",
1107 		      pcn->pcn_name, perfuse_node_path(ps, opc));
1108 		/* NOTREACHED */
1109 		error = ESTALE;
1110 		goto out;
1111 	}
1112 
1113 	/*
1114 	 * Removed node
1115 	 */
1116 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_REMOVED) {
1117 		error = ENOENT;
1118 		goto out;
1119 	}
1120 
1121 	/*
1122 	 * Check for sticky bit. Unfortunately there is no way to
1123 	 * do this before creating the puffs_node, since we require
1124 	 * this operation to get the node owner.
1125 	 */
1126 	switch (pcn->pcn_nameiop) {
1127 	case NAMEI_DELETE: /* FALLTHROUGH */
1128 	case NAMEI_RENAME:
1129 		error = sticky_access(opc, pn, pcn->pcn_cred);
1130 		if (error != 0) {
1131 			(void)perfuse_node_reclaim(pu, pn);
1132 			goto out;
1133 		}
1134 		break;
1135 	default:
1136 		break;
1137 	}
1138 
1139 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
1140 
1141 	error = 0;
1142 
1143 out:
1144 	node_rele(opc);
1145 	return error;
1146 }
1147 
1148 int
1149 perfuse_node_create(struct puffs_usermount *pu, puffs_cookie_t opc,
1150 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
1151 	const struct vattr *vap)
1152 {
1153 	perfuse_msg_t *pm;
1154 	struct perfuse_state *ps;
1155 	struct fuse_create_in *fci;
1156 	struct fuse_entry_out *feo;
1157 	struct fuse_open_out *foo;
1158 	struct puffs_node *pn;
1159 	const char *name;
1160 	size_t namelen;
1161 	size_t len;
1162 	int error;
1163 
1164 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1165 		return ENOENT;
1166 
1167 	node_ref(opc);
1168 
1169 	/*
1170 	 * If create is unimplemented: Check that it does not
1171 	 * already exists, and if not, do mknod and open
1172 	 */
1173 	ps = puffs_getspecific(pu);
1174 	if (ps->ps_flags & PS_NO_CREAT) {
1175 		error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
1176 					   pcn->pcn_cred, &pn);
1177 		if (error == 0)	{
1178 			(void)perfuse_node_reclaim(pu, pn);
1179 			error = EEXIST;
1180 			goto out;
1181 		}
1182 
1183 		error = perfuse_node_mknod(pu, opc, pni, pcn, vap);
1184 		if (error != 0)
1185 			goto out;
1186 
1187 		error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
1188 					   pcn->pcn_cred, &pn);
1189 		if (error != 0)
1190 			goto out;
1191 
1192 		/*
1193 		 * FUSE does the open at create time, while
1194 		 * NetBSD will open in a subsequent operation.
1195 		 * We need to open now, in order to retain FUSE
1196 		 * semantics. The calling process will not get
1197 		 * a file descriptor before the kernel sends
1198 		 * the open operation.
1199 		 */
1200 		error = perfuse_node_open(pu, (puffs_cookie_t)pn,
1201 					  FWRITE, pcn->pcn_cred);
1202 		goto out;
1203 	}
1204 
1205 	name = pcn->pcn_name;
1206 	namelen = pcn->pcn_namelen + 1;
1207 	len = sizeof(*fci) + namelen;
1208 
1209 	/*
1210 	 * flags should use O_WRONLY instead of O_RDWR, but it
1211 	 * breaks when the caller tries to read from file.
1212 	 *
1213 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1214 	 */
1215 	pm = ps->ps_new_msg(pu, opc, FUSE_CREATE, len, pcn->pcn_cred);
1216 	fci = GET_INPAYLOAD(ps, pm, fuse_create_in);
1217 	fci->flags = O_CREAT | O_TRUNC | O_RDWR;
1218 	fci->mode = vap->va_mode | VTTOIF(vap->va_type);
1219 	fci->umask = 0; 	/* Seems unused by libfuse */
1220 	(void)strlcpy((char*)(void *)(fci + 1), name, namelen);
1221 
1222 	len = sizeof(*feo) + sizeof(*foo);
1223 	if ((error = xchg_msg(pu, opc, pm, len, wait_reply)) != 0) {
1224 		/*
1225 		 * create is unimplmented, remember it for later,
1226 		 * and start over using mknod and open instead.
1227 		 */
1228 		if (error == ENOSYS) {
1229 			ps->ps_flags |= PS_NO_CREAT;
1230 			error = perfuse_node_create(pu, opc, pni, pcn, vap);
1231 		}
1232 
1233 		goto out;
1234 	}
1235 
1236 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
1237 	foo = (struct fuse_open_out *)(void *)(feo + 1);
1238 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
1239 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
1240 
1241 	/*
1242 	 * Save the file handle and inode in node private data
1243 	 * so that we can reuse it later
1244 	 */
1245 	pn = perfuse_new_pn(pu, name, opc);
1246 	perfuse_new_fh((puffs_cookie_t)pn, foo->fh, FWRITE);
1247 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
1248 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
1249 	perfuse_node_cache(ps, pn);
1250 
1251 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
1252 	pn->pn_va.va_gen = (u_long)(feo->generation);
1253 
1254 	puffs_newinfo_setcookie(pni, pn);
1255 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
1256 	puffs_newinfo_setva(pni, &pn->pn_va);
1257 	perfuse_newinfo_setttl(pni, pn, feo, NULL);
1258 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
1259 
1260 #ifdef PERFUSE_DEBUG
1261 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1262 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
1263 			"nodeid = 0x%"PRIx64", wfh = 0x%"PRIx64"\n",
1264 			__func__, (void *)pn, pcn->pcn_name,
1265 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid,
1266 			foo->fh);
1267 #endif
1268 
1269 	ps->ps_destroy_msg(pm);
1270 	error = 0;
1271 
1272 out:
1273 	node_rele(opc);
1274 	return error;
1275 }
1276 
1277 
1278 int
1279 perfuse_node_mknod(struct puffs_usermount *pu, puffs_cookie_t opc,
1280 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
1281 	const struct vattr *vap)
1282 {
1283 	struct perfuse_state *ps;
1284 	perfuse_msg_t *pm;
1285 	struct fuse_mknod_in *fmi;
1286 	const char* path;
1287 	size_t len;
1288 	int error;
1289 
1290 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1291 		return ENOENT;
1292 
1293 	node_ref(opc);
1294 
1295 	/*
1296 	 * Only superuser can mknod objects other than
1297 	 * directories, files, socks, fifo and links.
1298 	 *
1299 	 * Create an object require -WX permission in the parent directory
1300 	 */
1301 	switch (vap->va_type) {
1302 	case VDIR:	/* FALLTHROUGH */
1303 	case VREG:	/* FALLTHROUGH */
1304 	case VFIFO:	/* FALLTHROUGH */
1305 	case VSOCK:
1306 		break;
1307 	default:	/* VNON, VBLK, VCHR, VBAD */
1308 		if (!puffs_cred_isjuggernaut(pcn->pcn_cred)) {
1309 			error = EPERM;
1310 			goto out;
1311 		}
1312 		break;
1313 	}
1314 
1315 
1316 	ps = puffs_getspecific(pu);
1317 	path = pcn->pcn_name;
1318 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
1319 
1320 	/*
1321 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1322 	 */
1323 	pm = ps->ps_new_msg(pu, opc, FUSE_MKNOD, len, pcn->pcn_cred);
1324 	fmi = GET_INPAYLOAD(ps, pm, fuse_mknod_in);
1325 	fmi->mode = vap->va_mode | VTTOIF(vap->va_type);
1326 	fmi->rdev = (uint32_t)vap->va_rdev;
1327 	fmi->umask = 0; 	/* Seems unused bu libfuse */
1328 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
1329 
1330 	error = node_mk_common(pu, opc, pni, pcn, pm);
1331 
1332 out:
1333 	node_rele(opc);
1334 	return error;
1335 }
1336 
1337 
1338 int
1339 perfuse_node_open(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
1340 	const struct puffs_cred *pcr)
1341 {
1342 	return perfuse_node_open2(pu, opc, mode, pcr, NULL);
1343 }
1344 
1345 int
1346 perfuse_node_open2(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
1347 	const struct puffs_cred *pcr, int *oflags)
1348 {
1349 	struct perfuse_state *ps;
1350 	struct perfuse_node_data *pnd;
1351 	perfuse_msg_t *pm;
1352 	mode_t fmode;
1353 	int op;
1354 	struct fuse_open_in *foi;
1355 	struct fuse_open_out *foo;
1356 	struct puffs_node *pn;
1357 	int error;
1358 
1359 	ps = puffs_getspecific(pu);
1360 	pn = (struct puffs_node *)opc;
1361 	pnd = PERFUSE_NODE_DATA(opc);
1362 	error = 0;
1363 
1364 	if (pnd->pnd_flags & PND_REMOVED)
1365 		return ENOENT;
1366 
1367 	node_ref(opc);
1368 
1369 	if (puffs_pn_getvap(pn)->va_type == VDIR)
1370 		op = FUSE_OPENDIR;
1371 	else
1372 		op = FUSE_OPEN;
1373 
1374 	/*
1375 	 * libfuse docs says
1376 	 * - O_CREAT and O_EXCL should never be set.
1377 	 * - O_TRUNC may be used if mount option atomic_o_trunc is used XXX
1378 	 *
1379 	 * O_APPEND makes no sense since FUSE always sends
1380 	 * the file offset for write operations. If the
1381 	 * filesystem uses pwrite(), O_APPEND would cause
1382 	 * the offset to be ignored and cause file corruption.
1383 	 */
1384 	mode &= ~(O_CREAT|O_EXCL|O_APPEND);
1385 
1386 	/*
1387 	 * Do not open twice, and do not reopen for reading
1388 	 * if we already have write handle.
1389 	 */
1390 	switch (mode & (FREAD|FWRITE)) {
1391 	case FREAD:
1392 		if (pnd->pnd_flags & (PND_RFH|PND_WFH))
1393 			goto out;
1394 		break;
1395 	case FWRITE:
1396 		if (pnd->pnd_flags & PND_WFH)
1397 			goto out;
1398 		break;
1399 	case FREAD|FWRITE:
1400 		if (pnd->pnd_flags & PND_WFH)
1401 			goto out;
1402 
1403 		/*
1404 		 * Corner case: if already open for reading (PND_RFH)
1405 		 * and re-opening FREAD|FWRITE, we need to reopen,
1406 		 * but only for writing. Note the change on mode
1407 		 * will only affect perfuse_new_fh()
1408 		 */
1409 		if (pnd->pnd_flags & PND_RFH)
1410 			mode &= ~FREAD;
1411 		break;
1412 	default:
1413 		DWARNX("open without either FREAD nor FWRITE");
1414 		error = EPERM;
1415 		goto out;
1416 	}
1417 
1418 	/*
1419 	 * Queue open on a node so that we do not open
1420 	 * twice. This would be better with read and
1421 	 * write distinguished.
1422 	 */
1423 	while (pnd->pnd_flags & PND_INOPEN)
1424 		requeue_request(pu, opc, PCQ_OPEN);
1425 	pnd->pnd_flags |= PND_INOPEN;
1426 
1427 	/*
1428 	 * Convert PUFFS mode to FUSE mode: convert FREAD/FWRITE
1429 	 * to O_RDONLY/O_WRONLY while perserving the other options.
1430 	 */
1431 	fmode = mode & ~(FREAD|FWRITE);
1432 	fmode |= (mode & FWRITE) ? O_RDWR : O_RDONLY;
1433 
1434 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*foi), pcr);
1435 	foi = GET_INPAYLOAD(ps, pm, fuse_open_in);
1436 	foi->flags = fmode;
1437 	foi->unused = 0;
1438 
1439 	if ((error = xchg_msg(pu, opc, pm, sizeof(*foo), wait_reply)) != 0)
1440 		goto out;
1441 
1442 	foo = GET_OUTPAYLOAD(ps, pm, fuse_open_out);
1443 
1444 	/*
1445 	 * Save the file handle in node private data
1446 	 * so that we can reuse it later
1447 	 */
1448 	perfuse_new_fh(opc, foo->fh, mode);
1449 
1450 	/*
1451 	 * Set direct I/O if the filesystems forces it
1452 	 */
1453 	if ((foo->open_flags & FUSE_FOPEN_DIRECT_IO) && (oflags != NULL))
1454 		*oflags |= PUFFS_OPEN_IO_DIRECT;
1455 
1456 #ifdef PERFUSE_DEBUG
1457 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1458 		DPRINTF("%s: opc = %p, file = \"%s\", "
1459 			"nodeid = 0x%"PRIx64", %s%sfh = 0x%"PRIx64"\n",
1460 			__func__, (void *)opc, perfuse_node_path(ps, opc),
1461 			pnd->pnd_nodeid, mode & FREAD ? "r" : "",
1462 			mode & FWRITE ? "w" : "", foo->fh);
1463 #endif
1464 
1465 	ps->ps_destroy_msg(pm);
1466 out:
1467 
1468 	pnd->pnd_flags &= ~PND_INOPEN;
1469 	(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
1470 
1471 	node_rele(opc);
1472 	return error;
1473 }
1474 
1475 /* ARGSUSED0 */
1476 int
1477 perfuse_node_close(struct puffs_usermount *pu, puffs_cookie_t opc, int flags,
1478 	const struct puffs_cred *pcr)
1479 {
1480 	struct perfuse_node_data *pnd;
1481 
1482 	pnd = PERFUSE_NODE_DATA(opc);
1483 
1484 	if (!(pnd->pnd_flags & PND_OPEN))
1485 		return EBADF;
1486 
1487 	/*
1488 	 * Actual close is postponed at inactive time.
1489 	 */
1490 	return 0;
1491 }
1492 
1493 int
1494 perfuse_node_access(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
1495 	const struct puffs_cred *pcr)
1496 {
1497 	perfuse_msg_t *pm;
1498 	struct perfuse_state *ps;
1499 	struct fuse_access_in *fai;
1500 	int error;
1501 
1502 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1503 		return ENOENT;
1504 
1505 	node_ref(opc);
1506 
1507 	/*
1508 	 * If we previously detected the filesystem does not
1509 	 * implement access(), short-circuit the call and skip
1510 	 * to libpuffs access() emulation.
1511 	 */
1512 	ps = puffs_getspecific(pu);
1513 	if (ps->ps_flags & PS_NO_ACCESS) {
1514 		const struct vattr *vap;
1515 
1516 		vap = puffs_pn_getvap((struct puffs_node *)opc);
1517 
1518 		error = puffs_access(IFTOVT(vap->va_mode),
1519 				     vap->va_mode & ACCESSPERMS,
1520 				     vap->va_uid, vap->va_gid,
1521 				     (mode_t)mode, pcr);
1522 		goto out;
1523 	}
1524 
1525 	/*
1526 	 * Plain access call
1527 	 */
1528 	pm = ps->ps_new_msg(pu, opc, FUSE_ACCESS, sizeof(*fai), pcr);
1529 	fai = GET_INPAYLOAD(ps, pm, fuse_access_in);
1530 	fai->mask = 0;
1531 	fai->mask |= (mode & PUFFS_VREAD) ? R_OK : 0;
1532 	fai->mask |= (mode & PUFFS_VWRITE) ? W_OK : 0;
1533 	fai->mask |= (mode & PUFFS_VEXEC) ? X_OK : 0;
1534 
1535 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
1536 
1537 	ps->ps_destroy_msg(pm);
1538 
1539 	/*
1540 	 * If unimplemented, start over with emulation
1541 	 */
1542 	if (error == ENOSYS) {
1543 		ps->ps_flags |= PS_NO_ACCESS;
1544 		error = perfuse_node_access(pu, opc, mode, pcr);
1545 	}
1546 
1547 out:
1548 	node_rele(opc);
1549 	return error;
1550 }
1551 
1552 int
1553 perfuse_node_getattr(struct puffs_usermount *pu, puffs_cookie_t opc,
1554 	struct vattr *vap, const struct puffs_cred *pcr)
1555 {
1556 	return perfuse_node_getattr_ttl(pu, opc, vap, pcr, NULL);
1557 }
1558 
1559 int
1560 perfuse_node_getattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
1561 	struct vattr *vap, const struct puffs_cred *pcr,
1562 	struct timespec *va_ttl)
1563 {
1564 	perfuse_msg_t *pm = NULL;
1565 	struct perfuse_state *ps;
1566 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
1567 	struct fuse_getattr_in *fgi;
1568 	struct fuse_attr_out *fao;
1569 	int error = 0;
1570 
1571 	if ((pnd->pnd_flags & PND_REMOVED) && !(pnd->pnd_flags & PND_OPEN))
1572 		return ENOENT;
1573 
1574 	node_ref(opc);
1575 
1576 	/*
1577 	 * Serialize size access, see comment in perfuse_node_setattr().
1578 	 */
1579 	while (pnd->pnd_flags & PND_INRESIZE)
1580 		requeue_request(pu, opc, PCQ_RESIZE);
1581 	pnd->pnd_flags |= PND_INRESIZE;
1582 
1583 	ps = puffs_getspecific(pu);
1584 
1585 	/*
1586 	 * FUSE_GETATTR_FH must be set in fgi->flags
1587 	 * if we use for fgi->fh
1588 	 */
1589 	pm = ps->ps_new_msg(pu, opc, FUSE_GETATTR, sizeof(*fgi), pcr);
1590 	fgi = GET_INPAYLOAD(ps, pm, fuse_getattr_in);
1591 	fgi->getattr_flags = 0;
1592 	fgi->dummy = 0;
1593 	fgi->fh = 0;
1594 
1595 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_OPEN) {
1596 		fgi->fh = perfuse_get_fh(opc, FREAD);
1597 		fgi->getattr_flags |= FUSE_GETATTR_FH;
1598 	}
1599 
1600 #ifdef PERFUSE_DEBUG
1601 	if (perfuse_diagflags & PDF_RESIZE)
1602 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__, (void *)opc,
1603 		    vap->va_size);
1604 #endif
1605 
1606 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
1607 		goto out;
1608 
1609 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1610 
1611 #ifdef PERFUSE_DEBUG
1612 	if (perfuse_diagflags & PDF_RESIZE)
1613 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1614 		    (void *)opc, vap->va_size, fao->attr.size);
1615 #endif
1616 
1617 	/*
1618 	 * We set birthtime, flags, filerev,vaflags to 0.
1619 	 * This seems the best bet, since the information is
1620 	 * not available from filesystem.
1621 	 */
1622 	fuse_attr_to_vap(ps, vap, &fao->attr);
1623 
1624 	if (va_ttl != NULL) {
1625 		va_ttl->tv_sec = fao->attr_valid;
1626 		va_ttl->tv_nsec = fao->attr_valid_nsec;
1627 	}
1628 
1629 	ps->ps_destroy_msg(pm);
1630 	error = 0;
1631 out:
1632 
1633 	pnd->pnd_flags &= ~PND_INRESIZE;
1634 	(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
1635 
1636 	node_rele(opc);
1637 	return error;
1638 }
1639 
1640 int
1641 perfuse_node_setattr(struct puffs_usermount *pu, puffs_cookie_t opc,
1642 	const struct vattr *vap, const struct puffs_cred *pcr)
1643 {
1644 	return perfuse_node_setattr_ttl(pu, opc,
1645 					__UNCONST(vap), pcr, NULL, 0);
1646 }
1647 
1648 int
1649 perfuse_node_setattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
1650 	struct vattr *vap, const struct puffs_cred *pcr,
1651 	struct timespec *va_ttl, int xflag)
1652 {
1653 	perfuse_msg_t *pm;
1654 	uint64_t fh;
1655 	struct perfuse_state *ps;
1656 	struct perfuse_node_data *pnd;
1657 	struct fuse_setattr_in *fsi;
1658 	struct fuse_attr_out *fao;
1659 	struct vattr *old_va;
1660 	enum perfuse_xchg_pb_reply reply;
1661 	int error;
1662 #ifdef PERFUSE_DEBUG
1663 	struct vattr *old_vap;
1664 	int resize_debug = 0;
1665 #endif
1666 	ps = puffs_getspecific(pu);
1667 	pnd = PERFUSE_NODE_DATA(opc);
1668 
1669 	/*
1670 	 * The only operation we can do once the file is removed
1671 	 * is to resize it, and we can do it only if it is open.
1672 	 * Do not even send the operation to the filesystem: the
1673 	 * file is not there anymore.
1674 	 */
1675 	if (pnd->pnd_flags & PND_REMOVED) {
1676 		if (!(pnd->pnd_flags & PND_OPEN))
1677 			return ENOENT;
1678 
1679 		return 0;
1680 	}
1681 
1682 	old_va = puffs_pn_getvap((struct puffs_node *)opc);
1683 
1684 	/*
1685 	 * Check for permission to change size
1686 	 * It is always allowed if we already have a write file handle
1687 	 */
1688 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1689 	    !(pnd->pnd_flags & PND_WFH) &&
1690 	    (error = mode_access(opc, pcr, PUFFS_VWRITE)) != 0)
1691 		return error;
1692 
1693 	/*
1694 	 * Check for permission to change dates
1695 	 */
1696 	if (((vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1697 	     (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL)) &&
1698 	    (puffs_access_times(old_va->va_uid, old_va->va_gid,
1699 				old_va->va_mode, 0, pcr) != 0))
1700 		return EPERM;
1701 
1702 	/*
1703 	 * Check for permission to change owner and group
1704 	 */
1705 	if (((vap->va_uid != (uid_t)PUFFS_VNOVAL) ||
1706 	     (vap->va_gid != (gid_t)PUFFS_VNOVAL)) &&
1707 	    (puffs_access_chown(old_va->va_uid, old_va->va_gid,
1708 				vap->va_uid, vap->va_gid, pcr)) != 0)
1709 		return EPERM;
1710 
1711 	/*
1712 	 * Check for sticky bit on non-directory by non root user
1713 	 */
1714 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1715 	    (vap->va_mode & S_ISTXT) && (old_va->va_type != VDIR) &&
1716 	    !puffs_cred_isjuggernaut(pcr))
1717 		return EFTYPE;
1718 
1719 	/*
1720 	 * Check for permission to change permissions
1721 	 */
1722 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1723 	    (puffs_access_chmod(old_va->va_uid, old_va->va_gid,
1724 				old_va->va_type, vap->va_mode, pcr)) != 0)
1725 		return EPERM;
1726 
1727 	node_ref(opc);
1728 
1729 	if (pnd->pnd_flags & PND_WFH)
1730 		fh = perfuse_get_fh(opc, FWRITE);
1731 	else
1732 		fh = FUSE_UNKNOWN_FH;
1733 
1734 	/*
1735 	 * fchmod() sets mode and fh, and it may carry
1736 	 * a resize as well. That may break if the
1737 	 * filesystem does chmod then resize, and fails
1738 	 * because it does not have permission anymore.
1739 	 * We work this around by splitting into two setattr.
1740 	 */
1741 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1742 	    (vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1743 	    (fh != FUSE_UNKNOWN_FH)) {
1744 		struct vattr resize_va;
1745 
1746 		(void)memcpy(&resize_va, vap, sizeof(resize_va));
1747 		resize_va.va_mode = (mode_t)PUFFS_VNOVAL;
1748 		if ((error = perfuse_node_setattr_ttl(pu, opc, &resize_va,
1749 						      pcr, va_ttl, xflag)) != 0)
1750 			goto out2;
1751 
1752 		vap->va_size = (u_quad_t)PUFFS_VNOVAL;
1753 	}
1754 
1755 	pm = ps->ps_new_msg(pu, opc, FUSE_SETATTR, sizeof(*fsi), pcr);
1756 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
1757 	fsi->valid = 0;
1758 
1759 	/*
1760 	 * Get a fh if the node is open for writing
1761 	 */
1762 	if (fh != FUSE_UNKNOWN_FH) {
1763 		fsi->fh = fh;
1764 		fsi->valid |= FUSE_FATTR_FH;
1765 	}
1766 
1767 
1768 	if (vap->va_size != (u_quad_t)PUFFS_VNOVAL) {
1769 		fsi->size = vap->va_size;
1770 		fsi->valid |= FUSE_FATTR_SIZE;
1771 
1772 		/*
1773 		 * Serialize anything that can touch file size
1774 		 * to avoid reordered GETATTR and SETATTR.
1775 		 * Out of order SETATTR can report stale size,
1776 		 * which will cause the kernel to truncate the file.
1777 		 * XXX Probably useless now we have a lock on GETATTR
1778 		 */
1779 		while (pnd->pnd_flags & PND_INRESIZE)
1780 			requeue_request(pu, opc, PCQ_RESIZE);
1781 		pnd->pnd_flags |= PND_INRESIZE;
1782 	}
1783 
1784 	/*
1785  	 * Setting mtime without atime or vice versa leads to
1786 	 * dates being reset to Epoch on glusterfs. If one
1787 	 * is missing, use the old value.
1788  	 */
1789 	if ((vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1790 	    (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL)) {
1791 
1792 		if (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) {
1793 			fsi->atime = vap->va_atime.tv_sec;
1794 			fsi->atimensec = (uint32_t)vap->va_atime.tv_nsec;
1795 		} else {
1796 			fsi->atime = old_va->va_atime.tv_sec;
1797 			fsi->atimensec = (uint32_t)old_va->va_atime.tv_nsec;
1798 		}
1799 
1800 		if (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) {
1801 			fsi->mtime = vap->va_mtime.tv_sec;
1802 			fsi->mtimensec = (uint32_t)vap->va_mtime.tv_nsec;
1803 		} else {
1804 			fsi->mtime = old_va->va_mtime.tv_sec;
1805 			fsi->mtimensec = (uint32_t)old_va->va_mtime.tv_nsec;
1806 		}
1807 
1808 		fsi->valid |= (FUSE_FATTR_MTIME|FUSE_FATTR_ATIME);
1809 	}
1810 
1811 	if (vap->va_mode != (mode_t)PUFFS_VNOVAL) {
1812 		fsi->mode = vap->va_mode;
1813 		fsi->valid |= FUSE_FATTR_MODE;
1814 	}
1815 
1816 	if (vap->va_uid != (uid_t)PUFFS_VNOVAL) {
1817 		fsi->uid = vap->va_uid;
1818 		fsi->valid |= FUSE_FATTR_UID;
1819 	}
1820 
1821 	if (vap->va_gid != (gid_t)PUFFS_VNOVAL) {
1822 		fsi->gid = vap->va_gid;
1823 		fsi->valid |= FUSE_FATTR_GID;
1824 	}
1825 
1826 	if (pnd->pnd_lock_owner != 0) {
1827 		fsi->lock_owner = pnd->pnd_lock_owner;
1828 		fsi->valid |= FUSE_FATTR_LOCKOWNER;
1829 	}
1830 
1831 	/*
1832 	 * ftruncate() sends only va_size, and metadata cache
1833 	 * flush adds va_atime and va_mtime. Some FUSE
1834 	 * filesystems will attempt to detect ftruncate by
1835 	 * checking for FATTR_SIZE being set without
1836 	 * FATTR_UID|FATTR_GID|FATTR_ATIME|FATTR_MTIME|FATTR_MODE
1837 	 *
1838 	 * Try to adapt and remove FATTR_ATIME|FATTR_MTIME
1839 	 * if we suspect a ftruncate().
1840 	 */
1841 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1842 	    ((vap->va_mode == (mode_t)PUFFS_VNOVAL) &&
1843 	     (vap->va_uid == (uid_t)PUFFS_VNOVAL) &&
1844 	     (vap->va_gid == (gid_t)PUFFS_VNOVAL))) {
1845 		fsi->atime = 0;
1846 		fsi->atimensec = 0;
1847 		fsi->mtime = 0;
1848 		fsi->mtimensec = 0;
1849 		fsi->valid &= ~(FUSE_FATTR_ATIME|FUSE_FATTR_MTIME);
1850 	}
1851 
1852 	/*
1853 	 * If nothing remain, discard the operation.
1854 	 */
1855 	if (!(fsi->valid & (FUSE_FATTR_SIZE|FUSE_FATTR_ATIME|FUSE_FATTR_MTIME|
1856 			    FUSE_FATTR_MODE|FUSE_FATTR_UID|FUSE_FATTR_GID))) {
1857 		error = 0;
1858 		ps->ps_destroy_msg(pm);
1859 		goto out;
1860 	}
1861 
1862 #ifdef PERFUSE_DEBUG
1863 	old_vap = puffs_pn_getvap((struct puffs_node *)opc);
1864 
1865 	if ((perfuse_diagflags & PDF_RESIZE) &&
1866 	    (old_vap->va_size != (u_quad_t)PUFFS_VNOVAL)) {
1867 		resize_debug = 1;
1868 
1869 		DPRINTF(">> %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1870 		    (void *)opc,
1871 		    puffs_pn_getvap((struct puffs_node *)opc)->va_size,
1872 		    fsi->size);
1873 	}
1874 #endif
1875 
1876 	/*
1877 	 * Do not honour FAF when changing size. How do
1878 	 * you want such a thing to work?
1879 	 */
1880 	reply = wait_reply;
1881 #ifdef PUFFS_SETATTR_FAF
1882 	if ((xflag & PUFFS_SETATTR_FAF) && !(fsi->valid & FUSE_FATTR_SIZE))
1883 		reply = no_reply;
1884 #endif
1885 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), reply)) != 0)
1886 		goto out;
1887 
1888 	if (reply == no_reply)
1889 		goto out;
1890 
1891 	/*
1892 	 * Copy back the new values
1893 	 */
1894 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1895 
1896 #ifdef PERFUSE_DEBUG
1897 	if (resize_debug)
1898 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1899 		    (void *)opc, old_vap->va_size, fao->attr.size);
1900 #endif
1901 
1902 	fuse_attr_to_vap(ps, old_va, &fao->attr);
1903 
1904 	if (va_ttl != NULL) {
1905 		va_ttl->tv_sec = fao->attr_valid;
1906 		va_ttl->tv_nsec = fao->attr_valid_nsec;
1907 		(void)memcpy(vap, old_va, sizeof(*vap));
1908 	}
1909 
1910 	ps->ps_destroy_msg(pm);
1911 	error = 0;
1912 
1913 out:
1914 	if (pnd->pnd_flags & PND_INRESIZE) {
1915 		pnd->pnd_flags &= ~PND_INRESIZE;
1916 		(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
1917 	}
1918 
1919 out2:
1920 	node_rele(opc);
1921 	return error;
1922 }
1923 
1924 int
1925 perfuse_node_poll(struct puffs_usermount *pu, puffs_cookie_t opc, int *events)
1926 {
1927 	struct perfuse_state *ps;
1928 	perfuse_msg_t *pm;
1929 	struct fuse_poll_in *fpi;
1930 	struct fuse_poll_out *fpo;
1931 	int error;
1932 
1933 	node_ref(opc);
1934 	ps = puffs_getspecific(pu);
1935 	/*
1936 	 * kh is set if FUSE_POLL_SCHEDULE_NOTIFY is set.
1937 	 *
1938 	 * XXX ps_new_msg() is called with NULL creds, which will
1939 	 * be interpreted as FUSE superuser. We have no way to
1940 	 * know the requesting process' credential, but since poll
1941 	 * is supposed to operate on a file that has been open,
1942 	 * permission should have already been checked at open time.
1943 	 * That still may breaks on filesystems that provides odd
1944 	 * semantics.
1945  	 */
1946 	pm = ps->ps_new_msg(pu, opc, FUSE_POLL, sizeof(*fpi), NULL);
1947 	fpi = GET_INPAYLOAD(ps, pm, fuse_poll_in);
1948 	fpi->fh = perfuse_get_fh(opc, FREAD);
1949 	fpi->kh = 0;
1950 	fpi->flags = 0;
1951 
1952 #ifdef PERFUSE_DEBUG
1953 	if (perfuse_diagflags & PDF_FH)
1954 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
1955 			"fh = 0x%"PRIx64"\n", __func__, (void *)opc,
1956 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fpi->fh);
1957 #endif
1958 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fpo), wait_reply)) != 0)
1959 		goto out;
1960 
1961 	fpo = GET_OUTPAYLOAD(ps, pm, fuse_poll_out);
1962 	*events = fpo->revents;
1963 
1964 	ps->ps_destroy_msg(pm);
1965 	error = 0;
1966 
1967 out:
1968 	node_rele(opc);
1969 	return error;
1970 }
1971 
1972 /* ARGSUSED2 */
1973 int
1974 perfuse_node_fsync(struct puffs_usermount *pu, puffs_cookie_t opc,
1975 	const struct puffs_cred *pcr, int flags, off_t offlo, off_t offhi)
1976 {
1977 	int op;
1978 	perfuse_msg_t *pm;
1979 	struct perfuse_state *ps;
1980 	struct perfuse_node_data *pnd;
1981 	struct fuse_fsync_in *ffi;
1982 	uint64_t fh;
1983 	int error = 0;
1984 
1985 	pm = NULL;
1986 	ps = puffs_getspecific(pu);
1987 	pnd = PERFUSE_NODE_DATA(opc);
1988 
1989 	/*
1990 	 * No need to sync a removed node
1991 	 */
1992 	if (pnd->pnd_flags & PND_REMOVED)
1993 		return 0;
1994 
1995 	/*
1996 	 * We do not sync closed files. They have been
1997 	 * sync at inactive time already.
1998 	 */
1999 	if (!(pnd->pnd_flags & PND_OPEN))
2000 		return 0;
2001 
2002 	node_ref(opc);
2003 
2004 	if (puffs_pn_getvap((struct puffs_node *)opc)->va_type == VDIR)
2005 		op = FUSE_FSYNCDIR;
2006 	else 		/* VREG but also other types such as VLNK */
2007 		op = FUSE_FSYNC;
2008 
2009 	/*
2010 	 * Do not sync if there are no change to sync
2011 	 * XXX remove that test on files if we implement mmap
2012 	 */
2013 #ifdef PERFUSE_DEBUG
2014 	if (perfuse_diagflags & PDF_SYNC)
2015 		DPRINTF("%s: TEST opc = %p, file = \"%s\" is %sdirty\n",
2016 			__func__, (void*)opc, perfuse_node_path(ps, opc),
2017 			pnd->pnd_flags & PND_DIRTY ? "" : "not ");
2018 #endif
2019 	if (!(pnd->pnd_flags & PND_DIRTY))
2020 		goto out;
2021 
2022 	/*
2023 	 * It seems NetBSD can call fsync without open first
2024 	 * glusterfs complain in such a situation:
2025 	 * "FSYNC() ERR => -1 (Invalid argument)"
2026 	 * The file will be closed at inactive time.
2027 	 *
2028 	 * We open the directory for reading in order to sync.
2029 	 * This sounds rather counterintuitive, but it works.
2030 	 */
2031 	if (!(pnd->pnd_flags & PND_WFH)) {
2032 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2033 			goto out;
2034 	}
2035 
2036 	if (op == FUSE_FSYNCDIR)
2037 		fh = perfuse_get_fh(opc, FREAD);
2038 	else
2039 		fh = perfuse_get_fh(opc, FWRITE);
2040 
2041 	/*
2042 	 * If fsync_flags  is set, meta data should not be flushed.
2043 	 */
2044 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*ffi), pcr);
2045 	ffi = GET_INPAYLOAD(ps, pm, fuse_fsync_in);
2046 	ffi->fh = fh;
2047 	ffi->fsync_flags = (flags & FFILESYNC) ? 0 : 1;
2048 
2049 #ifdef PERFUSE_DEBUG
2050 	if (perfuse_diagflags & PDF_FH)
2051 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
2052 			__func__, (void *)opc,
2053 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, ffi->fh);
2054 #endif
2055 
2056 	if ((error = xchg_msg(pu, opc, pm,
2057 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
2058 		goto out;
2059 
2060 	/*
2061 	 * No reply beyond fuse_out_header: nothing to do on success
2062 	 * just clear the dirty flag
2063 	 */
2064 	pnd->pnd_flags &= ~PND_DIRTY;
2065 
2066 #ifdef PERFUSE_DEBUG
2067 	if (perfuse_diagflags & PDF_SYNC)
2068 		DPRINTF("%s: CLEAR opc = %p, file = \"%s\"\n",
2069 			__func__, (void*)opc, perfuse_node_path(ps, opc));
2070 #endif
2071 
2072 	ps->ps_destroy_msg(pm);
2073 	error = 0;
2074 
2075 out:
2076 	/*
2077 	 * ENOSYS is not returned to kernel,
2078 	 */
2079 	if (error == ENOSYS)
2080 		error = 0;
2081 
2082 	node_rele(opc);
2083 	return error;
2084 }
2085 
2086 int
2087 perfuse_node_remove(struct puffs_usermount *pu, puffs_cookie_t opc,
2088 	puffs_cookie_t targ, const struct puffs_cn *pcn)
2089 {
2090 	struct perfuse_state *ps;
2091 	struct perfuse_node_data *pnd;
2092 	perfuse_msg_t *pm;
2093 	char *path;
2094 	const char *name;
2095 	size_t len;
2096 	int error;
2097 
2098 	pnd = PERFUSE_NODE_DATA(opc);
2099 
2100 	if ((pnd->pnd_flags & PND_REMOVED) ||
2101 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
2102 		return ENOENT;
2103 
2104 #ifdef PERFUSE_DEBUG
2105 	if (targ == NULL)
2106 		DERRX(EX_SOFTWARE, "%s: targ is NULL", __func__);
2107 
2108 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
2109 		DPRINTF("%s: opc = %p, remove opc = %p, file = \"%s\"\n",
2110 			__func__, (void *)opc, (void *)targ, pcn->pcn_name);
2111 #endif
2112 	node_ref(opc);
2113 	node_ref(targ);
2114 
2115 	/*
2116 	 * Await for all operations on the deleted node to drain,
2117 	 * as the filesystem may be confused to have it deleted
2118 	 * during a getattr
2119 	 */
2120 	while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
2121 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2122 
2123 	ps = puffs_getspecific(pu);
2124 	pnd = PERFUSE_NODE_DATA(opc);
2125 	name = pcn->pcn_name;
2126 	len = pcn->pcn_namelen + 1;
2127 
2128 	pm = ps->ps_new_msg(pu, opc, FUSE_UNLINK, len, pcn->pcn_cred);
2129 	path = _GET_INPAYLOAD(ps, pm, char *);
2130 	(void)strlcpy(path, name, len);
2131 
2132 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2133 		goto out;
2134 
2135 	perfuse_cache_flush(targ);
2136 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2137 
2138 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2139 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2140 
2141 	/*
2142 	 * The parent directory needs a sync
2143 	 */
2144 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2145 
2146 #ifdef PERFUSE_DEBUG
2147 	if (perfuse_diagflags & PDF_FILENAME)
2148 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
2149 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
2150 			pcn->pcn_name);
2151 #endif
2152 	ps->ps_destroy_msg(pm);
2153 	error = 0;
2154 
2155 out:
2156 	node_rele(opc);
2157 	node_rele(targ);
2158 	return error;
2159 }
2160 
2161 int
2162 perfuse_node_link(struct puffs_usermount *pu, puffs_cookie_t opc,
2163 	puffs_cookie_t targ, const struct puffs_cn *pcn)
2164 {
2165 	struct perfuse_state *ps;
2166 	perfuse_msg_t *pm;
2167 	const char *name;
2168 	size_t len;
2169 	struct puffs_node *pn;
2170 	struct fuse_link_in *fli;
2171 	int error;
2172 
2173 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2174 		return ENOENT;
2175 
2176 	node_ref(opc);
2177 	node_ref(targ);
2178 	ps = puffs_getspecific(pu);
2179 	pn = (struct puffs_node *)targ;
2180 	name = pcn->pcn_name;
2181 	len =  sizeof(*fli) + pcn->pcn_namelen + 1;
2182 
2183 	pm = ps->ps_new_msg(pu, opc, FUSE_LINK, len, pcn->pcn_cred);
2184 	fli = GET_INPAYLOAD(ps, pm, fuse_link_in);
2185 	fli->oldnodeid = PERFUSE_NODE_DATA(pn)->pnd_nodeid;
2186 	(void)strlcpy((char *)(void *)(fli + 1), name, len - sizeof(*fli));
2187 
2188 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2189 		goto out;
2190 
2191 	ps->ps_destroy_msg(pm);
2192 	error = 0;
2193 
2194 out:
2195 	node_rele(opc);
2196 	node_rele(targ);
2197 	return error;
2198 }
2199 
2200 int
2201 perfuse_node_rename(struct puffs_usermount *pu, puffs_cookie_t opc,
2202 	puffs_cookie_t src, const struct puffs_cn *pcn_src,
2203 	puffs_cookie_t targ_dir, puffs_cookie_t targ,
2204 	const struct puffs_cn *pcn_targ)
2205 {
2206 	struct perfuse_state *ps;
2207 	struct perfuse_node_data *dstdir_pnd;
2208 	perfuse_msg_t *pm;
2209 	struct fuse_rename_in *fri;
2210 	const char *newname;
2211 	const char *oldname;
2212 	char *np;
2213 	int error;
2214 	size_t len;
2215 	size_t newname_len;
2216 	size_t oldname_len;
2217 
2218 	if ((PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED) ||
2219 	    (PERFUSE_NODE_DATA(src)->pnd_flags & PND_REMOVED) ||
2220 	    (PERFUSE_NODE_DATA(targ_dir)->pnd_flags & PND_REMOVED))
2221 		return ENOENT;
2222 
2223 	node_ref(opc);
2224 	node_ref(src);
2225 
2226 	/*
2227 	 * Await for all operations on the deleted node to drain,
2228 	 * as the filesystem may be confused to have it deleted
2229 	 * during a getattr
2230 	 */
2231 	if ((struct puffs_node *)targ != NULL) {
2232 		node_ref(targ);
2233 		while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
2234 			requeue_request(pu, targ, PCQ_AFTERXCHG);
2235 	} else {
2236 		while (PERFUSE_NODE_DATA(src)->pnd_inxchg)
2237 			requeue_request(pu, src, PCQ_AFTERXCHG);
2238 	}
2239 
2240 	ps = puffs_getspecific(pu);
2241 	newname =  pcn_targ->pcn_name;
2242 	newname_len = pcn_targ->pcn_namelen + 1;
2243 	oldname =  pcn_src->pcn_name;
2244 	oldname_len = pcn_src->pcn_namelen + 1;
2245 
2246 	len = sizeof(*fri) + oldname_len + newname_len;
2247 	pm = ps->ps_new_msg(pu, opc, FUSE_RENAME, len, pcn_targ->pcn_cred);
2248 	fri = GET_INPAYLOAD(ps, pm, fuse_rename_in);
2249 	fri->newdir = PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid;
2250 	np = (char *)(void *)(fri + 1);
2251 	(void)strlcpy(np, oldname, oldname_len);
2252 	np += oldname_len;
2253 	(void)strlcpy(np, newname, newname_len);
2254 
2255 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2256 		goto out;
2257 
2258 
2259 	/*
2260 	 * Record new parent nodeid
2261 	 */
2262 	dstdir_pnd = PERFUSE_NODE_DATA(targ_dir);
2263 	PERFUSE_NODE_DATA(src)->pnd_parent_nodeid = dstdir_pnd->pnd_nodeid;
2264 
2265 	if (opc != targ_dir)
2266 		dstdir_pnd->pnd_flags |= PND_DIRTY;
2267 
2268 	if (strcmp(newname, "..") != 0)
2269 		(void)strlcpy(PERFUSE_NODE_DATA(src)->pnd_name,
2270 		    newname, MAXPATHLEN);
2271 	else
2272 		PERFUSE_NODE_DATA(src)->pnd_name[0] = 0; /* forget name */
2273 
2274 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2275 
2276 	if ((struct puffs_node *)targ != NULL) {
2277 		perfuse_cache_flush(targ);
2278 		PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2279 	}
2280 
2281 #ifdef PERFUSE_DEBUG
2282 	if (perfuse_diagflags & PDF_FILENAME)
2283 		DPRINTF("%s: nodeid = 0x%"PRIx64" file = \"%s\" renamed \"%s\" "
2284 			"nodeid = 0x%"PRIx64" -> nodeid = 0x%"PRIx64" \"%s\"\n",
2285 	 		__func__, PERFUSE_NODE_DATA(src)->pnd_nodeid,
2286 			pcn_src->pcn_name, pcn_targ->pcn_name,
2287 			PERFUSE_NODE_DATA(opc)->pnd_nodeid,
2288 			PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid,
2289 			perfuse_node_path(ps, targ_dir));
2290 #endif
2291 
2292 	ps->ps_destroy_msg(pm);
2293 	error = 0;
2294 
2295 out:
2296 	node_rele(opc);
2297 	node_rele(src);
2298 	if ((struct puffs_node *)targ != NULL)
2299 		node_rele(targ);
2300 
2301 	return error;
2302 }
2303 
2304 int
2305 perfuse_node_mkdir(struct puffs_usermount *pu, puffs_cookie_t opc,
2306 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
2307 	const struct vattr *vap)
2308 {
2309 	struct perfuse_state *ps;
2310 	perfuse_msg_t *pm;
2311 	struct fuse_mkdir_in *fmi;
2312 	const char *path;
2313 	size_t len;
2314 	int error;
2315 
2316 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2317 		return ENOENT;
2318 
2319 	node_ref(opc);
2320 	ps = puffs_getspecific(pu);
2321 	path = pcn->pcn_name;
2322 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
2323 
2324 	pm = ps->ps_new_msg(pu, opc, FUSE_MKDIR, len, pcn->pcn_cred);
2325 	fmi = GET_INPAYLOAD(ps, pm, fuse_mkdir_in);
2326 	fmi->mode = vap->va_mode;
2327 	fmi->umask = 0; 	/* Seems unused by libfuse? */
2328 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
2329 
2330 	error = node_mk_common(pu, opc, pni, pcn, pm);
2331 
2332 	node_rele(opc);
2333 	return error;
2334 }
2335 
2336 
2337 int
2338 perfuse_node_rmdir(struct puffs_usermount *pu, puffs_cookie_t opc,
2339 	puffs_cookie_t targ, const struct puffs_cn *pcn)
2340 {
2341 	struct perfuse_state *ps;
2342 	struct perfuse_node_data *pnd;
2343 	perfuse_msg_t *pm;
2344 	char *path;
2345 	const char *name;
2346 	size_t len;
2347 	int error;
2348 
2349 	pnd = PERFUSE_NODE_DATA(opc);
2350 
2351 	if ((pnd->pnd_flags & PND_REMOVED) ||
2352 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
2353 		return ENOENT;
2354 
2355 	node_ref(opc);
2356 	node_ref(targ);
2357 
2358 	/*
2359 	 * Await for all operations on the deleted node to drain,
2360 	 * as the filesystem may be confused to have it deleted
2361 	 * during a getattr
2362 	 */
2363 	while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
2364 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2365 
2366 	ps = puffs_getspecific(pu);
2367 	name = pcn->pcn_name;
2368 	len = pcn->pcn_namelen + 1;
2369 
2370 	pm = ps->ps_new_msg(pu, opc, FUSE_RMDIR, len, pcn->pcn_cred);
2371 	path = _GET_INPAYLOAD(ps, pm, char *);
2372 	(void)strlcpy(path, name, len);
2373 
2374 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2375 		goto out;
2376 
2377 	perfuse_cache_flush(targ);
2378 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2379 
2380 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2381 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2382 
2383 	/*
2384 	 * The parent directory needs a sync
2385 	 */
2386 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2387 
2388 #ifdef PERFUSE_DEBUG
2389 	if (perfuse_diagflags & PDF_FILENAME)
2390 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
2391 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
2392 			perfuse_node_path(ps, targ));
2393 #endif
2394 	ps->ps_destroy_msg(pm);
2395 	error = 0;
2396 
2397 out:
2398 	node_rele(opc);
2399 	node_rele(targ);
2400 	return error;
2401 }
2402 
2403 /* vap is unused */
2404 /* ARGSUSED4 */
2405 int
2406 perfuse_node_symlink(struct puffs_usermount *pu, puffs_cookie_t opc,
2407 	struct puffs_newinfo *pni, const struct puffs_cn *pcn_src,
2408 	const struct vattr *vap, const char *link_target)
2409 {
2410 	struct perfuse_state *ps;
2411 	perfuse_msg_t *pm;
2412 	char *np;
2413 	const char *path;
2414 	size_t path_len;
2415 	size_t linkname_len;
2416 	size_t len;
2417 	int error;
2418 
2419 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2420 		return ENOENT;
2421 
2422 	node_ref(opc);
2423 	ps = puffs_getspecific(pu);
2424 	path = pcn_src->pcn_name;
2425 	path_len = pcn_src->pcn_namelen + 1;
2426 	linkname_len = strlen(link_target) + 1;
2427 	len = path_len + linkname_len;
2428 
2429 	pm = ps->ps_new_msg(pu, opc, FUSE_SYMLINK, len, pcn_src->pcn_cred);
2430 	np = _GET_INPAYLOAD(ps, pm, char *);
2431 	(void)strlcpy(np, path, path_len);
2432 	np += path_len;
2433 	(void)strlcpy(np, link_target, linkname_len);
2434 
2435 	error = node_mk_common(pu, opc, pni, pcn_src, pm);
2436 
2437 	node_rele(opc);
2438 	return error;
2439 }
2440 
2441 /* ARGSUSED4 */
2442 int
2443 perfuse_node_readdir(struct puffs_usermount *pu, puffs_cookie_t opc,
2444 	struct dirent *dent, off_t *readoff, size_t *reslen,
2445 	const struct puffs_cred *pcr, int *eofflag, off_t *cookies,
2446 	size_t *ncookies)
2447 {
2448 	perfuse_msg_t *pm;
2449 	uint64_t fh;
2450 	struct perfuse_state *ps;
2451 	struct perfuse_node_data *pnd;
2452 	struct fuse_read_in *fri;
2453 	struct fuse_out_header *foh;
2454 	struct fuse_dirent *fd;
2455 	size_t foh_len;
2456 	int error;
2457 	size_t fd_maxlen;
2458 
2459 	error = 0;
2460 	node_ref(opc);
2461 	ps = puffs_getspecific(pu);
2462 
2463 	/*
2464 	 * readdir state is kept at node level, and several readdir
2465 	 * requests can be issued at the same time on the same node.
2466 	 * We need to queue requests so that only one is in readdir
2467 	 * code at the same time.
2468 	 */
2469 	pnd = PERFUSE_NODE_DATA(opc);
2470 	while (pnd->pnd_flags & PND_INREADDIR)
2471 		requeue_request(pu, opc, PCQ_READDIR);
2472 	pnd->pnd_flags |= PND_INREADDIR;
2473 
2474 #ifdef PERFUSE_DEBUG
2475 	if (perfuse_diagflags & PDF_READDIR)
2476 		DPRINTF("%s: READDIR opc = %p enter critical section\n",
2477 			__func__, (void *)opc);
2478 #endif
2479 	/*
2480 	 * Re-initialize pnd->pnd_fd_cookie on the first readdir for a node
2481 	 */
2482 	if (*readoff == 0)
2483 		pnd->pnd_fd_cookie = 0;
2484 
2485 	/*
2486 	 * Do we already have the data bufered?
2487 	 */
2488 	if (pnd->pnd_dirent != NULL)
2489 		goto out;
2490 	pnd->pnd_dirent_len = 0;
2491 
2492 	/*
2493 	 * It seems NetBSD can call readdir without open first
2494 	 * libfuse will crash if it is done that way, hence open first.
2495 	 */
2496 	if (!(pnd->pnd_flags & PND_OPEN)) {
2497 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2498 			goto out;
2499 	}
2500 
2501 	fh = perfuse_get_fh(opc, FREAD);
2502 
2503 #ifdef PERFUSE_DEBUG
2504 	if (perfuse_diagflags & PDF_FH)
2505 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
2506 			"rfh = 0x%"PRIx64"\n", __func__, (void *)opc,
2507 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fh);
2508 #endif
2509 
2510 	pnd->pnd_all_fd = NULL;
2511 	pnd->pnd_all_fd_len = 0;
2512 	fd_maxlen = ps->ps_max_readahead - sizeof(*foh);
2513 
2514 	do {
2515 		size_t fd_len;
2516 		char *afdp;
2517 
2518 		pm = ps->ps_new_msg(pu, opc, FUSE_READDIR, sizeof(*fri), pcr);
2519 
2520 		/*
2521 		 * read_flags, lock_owner and flags are unused in libfuse
2522 		 */
2523 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
2524 		fri->fh = fh;
2525 		fri->offset = pnd->pnd_fd_cookie;
2526 		fri->size = (uint32_t)fd_maxlen;
2527 		fri->read_flags = 0;
2528 		fri->lock_owner = 0;
2529 		fri->flags = 0;
2530 
2531 		if ((error = xchg_msg(pu, opc, pm,
2532 				      UNSPEC_REPLY_LEN, wait_reply)) != 0)
2533 			goto out;
2534 
2535 		/*
2536 		 * There are many puffs_framebufs calls later,
2537 		 * therefore foh will not be valid for a long time.
2538 		 * Just get the length and forget it.
2539 		 */
2540 		foh = GET_OUTHDR(ps, pm);
2541 		foh_len = foh->len;
2542 
2543 		/*
2544 		 * Empty read: we reached the end of the buffer.
2545 		 */
2546 		if (foh_len == sizeof(*foh)) {
2547 			ps->ps_destroy_msg(pm);
2548 			*eofflag = 1;
2549 			break;
2550 		}
2551 
2552 		/*
2553 		 * Check for corrupted message.
2554 		 */
2555 		if (foh_len < sizeof(*foh) + sizeof(*fd)) {
2556 			ps->ps_destroy_msg(pm);
2557 			DWARNX("readdir reply too short");
2558 			error = EIO;
2559 			goto out;
2560 		}
2561 
2562 
2563 		fd = GET_OUTPAYLOAD(ps, pm, fuse_dirent);
2564 		fd_len = foh_len - sizeof(*foh);
2565 
2566 		pnd->pnd_all_fd = realloc(pnd->pnd_all_fd,
2567 					  pnd->pnd_all_fd_len + fd_len);
2568 		if (pnd->pnd_all_fd  == NULL)
2569 			DERR(EX_OSERR, "%s: malloc failed", __func__);
2570 
2571 		afdp = (char *)(void *)pnd->pnd_all_fd + pnd->pnd_all_fd_len;
2572 		(void)memcpy(afdp, fd, fd_len);
2573 
2574 		pnd->pnd_all_fd_len += fd_len;
2575 
2576 		/*
2577 		 * The fd->off field is used as a cookie for
2578 		 * resuming the next readdir() where this one was left.
2579 	 	 */
2580 		pnd->pnd_fd_cookie = readdir_last_cookie(fd, fd_len);
2581 
2582 		ps->ps_destroy_msg(pm);
2583 	} while (1 /* CONSTCOND */);
2584 
2585 	if (pnd->pnd_all_fd != NULL) {
2586 		if (fuse_to_dirent(pu, opc, pnd->pnd_all_fd,
2587 				   pnd->pnd_all_fd_len) == -1)
2588 			error = EIO;
2589 	}
2590 
2591 out:
2592 	if (pnd->pnd_all_fd != NULL) {
2593 		free(pnd->pnd_all_fd);
2594 		pnd->pnd_all_fd = NULL;
2595 		pnd->pnd_all_fd_len = 0;
2596 	}
2597 
2598 	if (error == 0)
2599 		readdir_buffered(opc, dent, readoff, reslen);
2600 
2601 	/*
2602 	 * Schedule queued readdir requests
2603 	 */
2604 	pnd->pnd_flags &= ~PND_INREADDIR;
2605 	(void)dequeue_requests(opc, PCQ_READDIR, DEQUEUE_ALL);
2606 
2607 #ifdef PERFUSE_DEBUG
2608 	if (perfuse_diagflags & PDF_READDIR)
2609 		DPRINTF("%s: READDIR opc = %p exit critical section\n",
2610 			__func__, (void *)opc);
2611 #endif
2612 
2613 	node_rele(opc);
2614 	return error;
2615 }
2616 
2617 int
2618 perfuse_node_readlink(struct puffs_usermount *pu, puffs_cookie_t opc,
2619 	const struct puffs_cred *pcr, char *linkname, size_t *linklen)
2620 {
2621 	struct perfuse_state *ps;
2622 	perfuse_msg_t *pm;
2623 	int error;
2624 	size_t len;
2625 	struct fuse_out_header *foh;
2626 
2627 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2628 		return ENOENT;
2629 
2630 	node_ref(opc);
2631 	ps = puffs_getspecific(pu);
2632 
2633 	pm = ps->ps_new_msg(pu, opc, FUSE_READLINK, 0, pcr);
2634 
2635 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2636 		goto out;
2637 
2638 	foh = GET_OUTHDR(ps, pm);
2639 	len = foh->len - sizeof(*foh);
2640 	if (len > *linklen)
2641 		DERRX(EX_PROTOCOL, "path len = %zd too long", len);
2642 	if (len == 0)
2643 		DERRX(EX_PROTOCOL, "path len = %zd too short", len);
2644 
2645 	(void)memcpy(linkname, _GET_OUTPAYLOAD(ps, pm, char *), len);
2646 
2647 	/*
2648 	 * FUSE filesystems return a NUL terminated string, we
2649 	 * do not want the trailing \0
2650 	 */
2651 	while (len > 0 && linkname[len - 1] == '\0')
2652 		len--;
2653 
2654 	*linklen = len;
2655 
2656 	ps->ps_destroy_msg(pm);
2657 	error = 0;
2658 
2659 out:
2660 	node_rele(opc);
2661 	return error;
2662 }
2663 
2664 int
2665 perfuse_node_reclaim(struct puffs_usermount *pu, puffs_cookie_t opc)
2666 {
2667 	struct perfuse_state *ps;
2668 	perfuse_msg_t *pm;
2669 	struct perfuse_node_data *pnd;
2670 	struct fuse_forget_in *ffi;
2671 	int nlookup;
2672 	struct timespec now;
2673 
2674 	if (opc == 0)
2675 		return 0;
2676 
2677 	ps = puffs_getspecific(pu);
2678 	pnd = PERFUSE_NODE_DATA(opc);
2679 
2680 	/*
2681 	 * Never forget the root.
2682 	 */
2683 	if (pnd->pnd_nodeid == FUSE_ROOT_ID)
2684 		return 0;
2685 
2686 	/*
2687 	 * There is a race condition between reclaim and lookup.
2688 	 * When looking up an already known node, the kernel cannot
2689 	 * hold a reference on the result until it gets the PUFFS
2690 	 * reply. It mayy therefore reclaim the node after the
2691 	 * userland looked it up, and before it gets the reply.
2692 	 * On rely, the kernel re-creates the node, but at that
2693 	 * time the node has been reclaimed in userland.
2694 	 *
2695 	 * In order to avoid this, we refuse reclaiming nodes that
2696 	 * are too young since the last lookup - and that we do
2697 	 * not have removed on our own, of course.
2698 	 */
2699 	if (clock_gettime(CLOCK_REALTIME, &now) != 0)
2700 		DERR(EX_OSERR, "clock_gettime failed");
2701 
2702 	if (timespeccmp(&pnd->pnd_cn_expire, &now, >) &&
2703 	    !(pnd->pnd_flags & PND_REMOVED)) {
2704 		if (!(pnd->pnd_flags & PND_NODELEAK)) {
2705 			ps->ps_nodeleakcount++;
2706 			pnd->pnd_flags |= PND_NODELEAK;
2707 		}
2708 		DWARNX("possible leaked node:: opc = %p \"%s\"",
2709 		       opc, pnd->pnd_name);
2710 		return 0;
2711 	}
2712 
2713 	node_ref(opc);
2714 	pnd->pnd_flags |= PND_RECLAIMED;
2715 	pnd->pnd_puffs_nlookup--;
2716 	nlookup = pnd->pnd_puffs_nlookup;
2717 
2718 #ifdef PERFUSE_DEBUG
2719 	if (perfuse_diagflags & PDF_RECLAIM)
2720 		DPRINTF("%s (nodeid %"PRId64") reclaimed\n",
2721 			perfuse_node_path(ps, opc), pnd->pnd_nodeid);
2722 #endif
2723 
2724 #ifdef PERFUSE_DEBUG
2725 	if (perfuse_diagflags & PDF_RECLAIM)
2726 		DPRINTF("%s (nodeid %"PRId64") is %sreclaimed, nlookup = %d "
2727 			"%s%s%s%s, pending ops:%s%s%s\n",
2728 		        perfuse_node_path(ps, opc), pnd->pnd_nodeid,
2729 		        pnd->pnd_flags & PND_RECLAIMED ? "" : "not ",
2730 			pnd->pnd_puffs_nlookup,
2731 			pnd->pnd_flags & PND_OPEN ? "open " : "not open",
2732 			pnd->pnd_flags & PND_RFH ? "r" : "",
2733 			pnd->pnd_flags & PND_WFH ? "w" : "",
2734 			pnd->pnd_flags & PND_BUSY ? "" : " none",
2735 			pnd->pnd_flags & PND_INREADDIR ? " readdir" : "",
2736 			pnd->pnd_flags & PND_INWRITE ? " write" : "",
2737 			pnd->pnd_flags & PND_INOPEN ? " open" : "");
2738 #endif
2739 	/*
2740 	 * Make sure it is not looked up again
2741 	 */
2742 	if (!(pnd->pnd_flags & PND_REMOVED))
2743 		perfuse_cache_flush(opc);
2744 
2745 	/*
2746 	 * Purge any activity on the node, while checking
2747 	 * that it remains eligible for a reclaim.
2748 	 */
2749 	while (pnd->pnd_ref > 1)
2750 		requeue_request(pu, opc, PCQ_REF);
2751 
2752 	/*
2753 	 * reclaim cancel?
2754 	 */
2755 	if (pnd->pnd_puffs_nlookup > nlookup) {
2756 		pnd->pnd_flags &= ~PND_RECLAIMED;
2757 		perfuse_node_cache(ps, opc);
2758 		node_rele(opc);
2759 		return 0;
2760 	}
2761 
2762 
2763 #ifdef PERFUSE_DEBUG
2764 	if ((pnd->pnd_flags & PND_OPEN) ||
2765 	       !TAILQ_EMPTY(&pnd->pnd_pcq))
2766 		DERRX(EX_SOFTWARE, "%s: opc = %p \"%s\": still open",
2767 		      __func__, opc, pnd->pnd_name);
2768 
2769 	if ((pnd->pnd_flags & PND_BUSY) ||
2770 	       !TAILQ_EMPTY(&pnd->pnd_pcq))
2771 		DERRX(EX_SOFTWARE, "%s: opc = %p: queued operations",
2772 		      __func__, opc);
2773 
2774 	if (pnd->pnd_inxchg != 0)
2775 		DERRX(EX_SOFTWARE, "%s: opc = %p: ongoing operations",
2776 		      __func__, opc);
2777 #endif
2778 
2779 	/*
2780 	 * Send the FORGET message
2781 	 *
2782 	 * ps_new_msg() is called with NULL creds, which will
2783 	 * be interpreted as FUSE superuser. This is obviously
2784 	 * fine since we operate with kernel creds here.
2785 	 */
2786 	pm = ps->ps_new_msg(pu, opc, FUSE_FORGET,
2787 		      sizeof(*ffi), NULL);
2788 	ffi = GET_INPAYLOAD(ps, pm, fuse_forget_in);
2789 	ffi->nlookup = pnd->pnd_fuse_nlookup;
2790 
2791 	/*
2792 	 * No reply is expected, pm is freed in xchg_msg
2793 	 */
2794 	(void)xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, no_reply);
2795 
2796 	perfuse_destroy_pn(pu, opc);
2797 
2798 	return 0;
2799 }
2800 
2801 int
2802 perfuse_node_inactive(struct puffs_usermount *pu, puffs_cookie_t opc)
2803 {
2804 	struct perfuse_node_data *pnd;
2805 	int error;
2806 
2807 	if (opc == 0)
2808 		return 0;
2809 
2810 	node_ref(opc);
2811 	pnd = PERFUSE_NODE_DATA(opc);
2812 
2813 	if (!(pnd->pnd_flags & (PND_OPEN|PND_REMOVED)))
2814 		goto out;
2815 
2816 	/*
2817 	 * Make sure all operation are finished
2818 	 * There can be an ongoing write. Other
2819 	 * operation wait for all data before
2820 	 * the close/inactive.
2821 	 */
2822 	while (pnd->pnd_flags & PND_INWRITE)
2823 		requeue_request(pu, opc, PCQ_AFTERWRITE);
2824 
2825 	/*
2826 	 * The inactive operation may be cancelled,
2827 	 * If no open is in progress, set PND_INOPEN
2828 	 * so that a new open will be queued.
2829 	 */
2830 	if (pnd->pnd_flags & PND_INOPEN)
2831 		goto out;
2832 
2833 	pnd->pnd_flags |= PND_INOPEN;
2834 
2835 	/*
2836 	 * Sync data
2837 	 */
2838 	if (pnd->pnd_flags & PND_DIRTY) {
2839 		if ((error = perfuse_node_fsync(pu, opc, NULL, 0, 0, 0)) != 0)
2840 			DWARN("%s: perfuse_node_fsync failed error = %d",
2841 			      __func__, error);
2842 	}
2843 
2844 
2845 	/*
2846 	 * Close handles
2847 	 */
2848 	if (pnd->pnd_flags & PND_WFH) {
2849 		if ((error = perfuse_node_close_common(pu, opc, FWRITE)) != 0)
2850 			DWARN("%s: close write FH failed error = %d",
2851 			      __func__, error);
2852 	}
2853 
2854 	if (pnd->pnd_flags & PND_RFH) {
2855 		if ((error = perfuse_node_close_common(pu, opc, FREAD)) != 0)
2856 			DWARN("%s: close read FH failed error = %d",
2857 			      __func__, error);
2858 	}
2859 
2860 	/*
2861 	 * This will cause a reclaim to be sent
2862 	 */
2863 	if (pnd->pnd_flags & PND_REMOVED)
2864 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N1);
2865 
2866 	/*
2867 	 * Schedule awaiting operations
2868 	 */
2869 	pnd->pnd_flags &= ~PND_INOPEN;
2870 	(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
2871 
2872 	/*
2873 	 * errors are ignored, since the kernel ignores the return code.
2874 	 */
2875 out:
2876 	node_rele(opc);
2877 	return 0;
2878 }
2879 
2880 
2881 /* ARGSUSED0 */
2882 int
2883 perfuse_node_print(struct puffs_usermount *pu, puffs_cookie_t opc)
2884 {
2885 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2886 	return 0;
2887 }
2888 
2889 int
2890 perfuse_node_pathconf(struct puffs_usermount *pu, puffs_cookie_t opc,
2891 	int name, int *retval)
2892 {
2893 	perfuse_msg_t *pm;
2894 	struct perfuse_state *ps;
2895 	struct fuse_statfs_out *fso;
2896 	int error = 0;
2897 
2898 	/*
2899 	 * Static values copied from UFS
2900 	 * in src/sys/ufs/ufs/ufs_vnops.c
2901 	 */
2902 	switch (name) {
2903 	case _PC_LINK_MAX:
2904 		*retval = LINK_MAX;
2905 		break;
2906 	case _PC_PATH_MAX:
2907 		*retval = PATH_MAX;
2908 		break;
2909 	case _PC_PIPE_BUF:
2910 		*retval = PIPE_BUF;
2911 		break;
2912 	case _PC_CHOWN_RESTRICTED:
2913 		*retval = 1;
2914 		break;
2915 	case _PC_NO_TRUNC:
2916 		*retval = 1;
2917 		break;
2918 	case _PC_SYNC_IO:
2919 		*retval = 1;
2920 		break;
2921 	case _PC_FILESIZEBITS:
2922 		*retval = 42;
2923 		break;
2924 	case _PC_SYMLINK_MAX:
2925 		*retval = MAXPATHLEN;
2926 		break;
2927 	case _PC_2_SYMLINKS:
2928 		*retval = 1;
2929 		break;
2930 	case _PC_NAME_MAX:
2931 		ps = puffs_getspecific(pu);
2932 		pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
2933 
2934 		error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply);
2935 		if (error != 0)
2936 			return error;
2937 
2938 		fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
2939 		*retval = fso->st.namelen;
2940 
2941 		ps->ps_destroy_msg(pm);
2942 
2943 		break;
2944 	default:
2945 		DWARN("Unimplemented pathconf for name = %d", name);
2946 		error = ENOSYS;
2947 		break;
2948 	}
2949 
2950 	return error;
2951 }
2952 
2953 int
2954 perfuse_node_advlock(struct puffs_usermount *pu, puffs_cookie_t opc,
2955 	void *id, int op, struct flock *fl, int flags)
2956 {
2957 	struct perfuse_state *ps;
2958 	int fop;
2959 	perfuse_msg_t *pm;
2960 	uint64_t fh;
2961 	struct fuse_lk_in *fli;
2962 	struct fuse_out_header *foh;
2963 	struct fuse_lk_out *flo;
2964 	uint32_t owner;
2965 	size_t len;
2966 	int error;
2967 
2968 	node_ref(opc);
2969 
2970 	/*
2971 	 * Make sure we do have a filehandle, as the FUSE filesystem
2972 	 * expect one. E.g.: if we provide none, GlusterFS logs an error
2973 	 * "0-glusterfs-fuse: xl is NULL"
2974 	 *
2975 	 * We need the read file handle if the file is open read only,
2976 	 * in order to support shared locks on read-only files.
2977 	 * NB: The kernel always sends advlock for read-only
2978 	 * files at exit time when the process used lock, see
2979 	 * sys_exit -> exit1 -> fd_free -> fd_close -> VOP_ADVLOCK
2980 	 */
2981 	if ((fh = perfuse_get_fh(opc, FREAD)) == FUSE_UNKNOWN_FH) {
2982 		error = EBADF;
2983 		goto out;
2984 	}
2985 
2986 	ps = puffs_getspecific(pu);
2987 
2988 	if (op == F_GETLK)
2989 		fop = FUSE_GETLK;
2990 	else
2991 		fop = (flags & F_WAIT) ? FUSE_SETLKW : FUSE_SETLK;
2992 
2993 	/*
2994 	 * XXX ps_new_msg() is called with NULL creds, which will
2995 	 * be interpreted as FUSE superuser. We have no way to
2996 	 * know the requesting process' credential, but since advlock()
2997 	 * is supposed to operate on a file that has been open(),
2998 	 * permission should have already been checked at open() time.
2999 	 */
3000 	pm = ps->ps_new_msg(pu, opc, fop, sizeof(*fli), NULL);
3001 	fli = GET_INPAYLOAD(ps, pm, fuse_lk_in);
3002 	fli->fh = fh;
3003 	fli->owner = (uint64_t)(vaddr_t)id;
3004 	fli->lk.start = fl->l_start;
3005 	fli->lk.end = fl->l_start + fl->l_len;
3006 	fli->lk.type = fl->l_type;
3007 	fli->lk.pid = fl->l_pid;
3008 	fli->lk_flags = (flags & F_FLOCK) ? FUSE_LK_FLOCK : 0;
3009 
3010 	owner = (uint32_t)(vaddr_t)id;
3011 
3012 #ifdef PERFUSE_DEBUG
3013 	if (perfuse_diagflags & PDF_FH)
3014 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
3015 			__func__, (void *)opc,
3016 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fli->fh);
3017 #endif
3018 
3019 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3020 		goto out;
3021 
3022 	foh = GET_OUTHDR(ps, pm);
3023 	len = foh->len - sizeof(*foh);
3024 
3025 	/*
3026 	 * Save or clear the lock
3027 	 */
3028 	switch (op) {
3029 	case F_GETLK:
3030 		if (len != sizeof(*flo))
3031 			DERRX(EX_SOFTWARE,
3032 			      "%s: Unexpected lock reply len %zd",
3033 			      __func__, len);
3034 
3035 		flo = GET_OUTPAYLOAD(ps, pm, fuse_lk_out);
3036 		fl->l_start = flo->lk.start;
3037 		fl->l_len = flo->lk.end - flo->lk.start;
3038 		fl->l_pid = flo->lk.pid;
3039 		fl->l_type = flo->lk.type;
3040 		fl->l_whence = SEEK_SET;	/* libfuse hardcodes it */
3041 
3042 		PERFUSE_NODE_DATA(opc)->pnd_lock_owner = flo->lk.pid;
3043 		break;
3044 	case F_UNLCK:
3045 		owner = 0;
3046 		/* FALLTHROUGH */
3047 	case F_SETLK:
3048 		/* FALLTHROUGH */
3049 	case F_SETLKW:
3050 		if (error != 0)
3051 			PERFUSE_NODE_DATA(opc)->pnd_lock_owner = owner;
3052 
3053 		if (len != 0)
3054 			DERRX(EX_SOFTWARE,
3055 			      "%s: Unexpected unlock reply len %zd",
3056 			      __func__, len);
3057 
3058 		break;
3059 	default:
3060 		DERRX(EX_SOFTWARE, "%s: Unexpected op %d", __func__, op);
3061 		break;
3062 	}
3063 
3064 	ps->ps_destroy_msg(pm);
3065 	error = 0;
3066 
3067 out:
3068 	node_rele(opc);
3069 	return error;
3070 }
3071 
3072 int
3073 perfuse_node_read(struct puffs_usermount *pu, puffs_cookie_t opc, uint8_t *buf,
3074 	off_t offset, size_t *resid, const struct puffs_cred *pcr, int ioflag)
3075 {
3076 	struct perfuse_state *ps;
3077 	struct perfuse_node_data *pnd;
3078 	const struct vattr *vap;
3079 	perfuse_msg_t *pm;
3080 	struct fuse_read_in *fri;
3081 	struct fuse_out_header *foh;
3082 	size_t readen;
3083 	int error;
3084 
3085 	ps = puffs_getspecific(pu);
3086 	pnd = PERFUSE_NODE_DATA(opc);
3087 	vap = puffs_pn_getvap((struct puffs_node *)opc);
3088 
3089 	/*
3090 	 * NetBSD turns that into a getdents(2) output
3091 	 * We just do a EISDIR as this feature is of little use.
3092 	 */
3093 	if (vap->va_type == VDIR)
3094 		return EISDIR;
3095 
3096 	do {
3097 		size_t max_read;
3098 
3099 		max_read = ps->ps_max_readahead - sizeof(*foh);
3100 		/*
3101 		 * flags may be set to FUSE_READ_LOCKOWNER
3102 		 * if lock_owner is provided.
3103 		 */
3104 		pm = ps->ps_new_msg(pu, opc, FUSE_READ, sizeof(*fri), pcr);
3105 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
3106 		fri->fh = perfuse_get_fh(opc, FREAD);
3107 		fri->offset = offset;
3108 		fri->size = (uint32_t)MIN(*resid, max_read);
3109 		fri->read_flags = 0; /* XXX Unused by libfuse? */
3110 		fri->lock_owner = pnd->pnd_lock_owner;
3111 		fri->flags = 0;
3112 		fri->flags |= (fri->lock_owner != 0) ? FUSE_READ_LOCKOWNER : 0;
3113 
3114 #ifdef PERFUSE_DEBUG
3115 	if (perfuse_diagflags & PDF_FH)
3116 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
3117 			__func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
3118 #endif
3119 		error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
3120 		if (error  != 0)
3121 			return error;
3122 
3123 		foh = GET_OUTHDR(ps, pm);
3124 		readen = foh->len - sizeof(*foh);
3125 
3126 #ifdef PERFUSE_DEBUG
3127 		if (readen > *resid)
3128 			DERRX(EX_SOFTWARE, "%s: Unexpected big read %zd",
3129 			      __func__, readen);
3130 #endif
3131 
3132 		(void)memcpy(buf,  _GET_OUTPAYLOAD(ps, pm, char *), readen);
3133 
3134 		buf += readen;
3135 		offset += readen;
3136 		*resid -= readen;
3137 
3138 		ps->ps_destroy_msg(pm);
3139 	} while ((*resid != 0) && (readen != 0));
3140 
3141 	if (ioflag & (IO_SYNC|IO_DSYNC))
3142 		ps->ps_syncreads++;
3143 	else
3144 		ps->ps_asyncreads++;
3145 
3146 	return 0;
3147 }
3148 
3149 int
3150 perfuse_node_write(struct puffs_usermount *pu, puffs_cookie_t opc,
3151 	uint8_t *buf, off_t offset, size_t *resid,
3152 	const struct puffs_cred *pcr, int ioflag)
3153 {
3154 	return perfuse_node_write2(pu, opc, buf, offset, resid, pcr, ioflag, 0);
3155 }
3156 
3157 /* ARGSUSED7 */
3158 int
3159 perfuse_node_write2(struct puffs_usermount *pu, puffs_cookie_t opc,
3160 	uint8_t *buf, off_t offset, size_t *resid,
3161 	const struct puffs_cred *pcr, int ioflag, int xflag)
3162 {
3163 	struct perfuse_state *ps;
3164 	struct perfuse_node_data *pnd;
3165 	struct vattr *vap;
3166 	perfuse_msg_t *pm;
3167 	struct fuse_write_in *fwi;
3168 	struct fuse_write_out *fwo;
3169 	size_t data_len;
3170 	size_t payload_len;
3171 	size_t written;
3172 	int inresize;
3173 	int error;
3174 
3175 	ps = puffs_getspecific(pu);
3176 	pnd = PERFUSE_NODE_DATA(opc);
3177 	vap = puffs_pn_getvap((struct puffs_node *)opc);
3178 	written = 0;
3179 	inresize = 0;
3180 	error = 0;
3181 
3182 	if (vap->va_type == VDIR)
3183 		return EISDIR;
3184 
3185 	node_ref(opc);
3186 
3187 	/*
3188 	 * We need to queue write requests in order to avoid
3189 	 * dequeueing PCQ_AFTERWRITE when there are pending writes.
3190 	 */
3191 	while (pnd->pnd_flags & PND_INWRITE)
3192 		requeue_request(pu, opc, PCQ_WRITE);
3193 	pnd->pnd_flags |= PND_INWRITE;
3194 
3195 	/*
3196 	 * Serialize size access, see comment in perfuse_node_setattr().
3197 	 */
3198 	if ((u_quad_t)offset + *resid > vap->va_size) {
3199 		while (pnd->pnd_flags & PND_INRESIZE)
3200 			requeue_request(pu, opc, PCQ_RESIZE);
3201 		pnd->pnd_flags |= PND_INRESIZE;
3202 		inresize = 1;
3203 	}
3204 
3205 	/*
3206 	 * append flag: re-read the file size so that
3207 	 * we get the latest value.
3208 	 */
3209 	if (ioflag & PUFFS_IO_APPEND) {
3210 		if ((error = perfuse_node_getattr(pu, opc, vap, pcr)) != 0)
3211 			goto out;
3212 
3213 		offset = vap->va_size;
3214 	}
3215 
3216 #ifdef PERFUSE_DEBUG
3217 	if (perfuse_diagflags & PDF_RESIZE)
3218 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__,
3219 			(void *)opc, vap->va_size);
3220 #endif
3221 
3222 	do {
3223 		size_t max_write;
3224 		/*
3225 		 * There is a writepage flag when data
3226 		 * is aligned to page size. Use it for
3227 		 * everything but the data after the last
3228 		 * page boundary.
3229 		 */
3230 		max_write = ps->ps_max_write - sizeof(*fwi);
3231 
3232 		data_len = MIN(*resid, max_write);
3233 		if (data_len > (size_t)sysconf(_SC_PAGESIZE))
3234 			data_len = data_len & ~(sysconf(_SC_PAGESIZE) - 1);
3235 
3236 		payload_len = data_len + sizeof(*fwi);
3237 
3238 		/*
3239 		 * flags may be set to FUSE_WRITE_CACHE (XXX usage?)
3240 		 * or FUSE_WRITE_LOCKOWNER, if lock_owner is provided.
3241 		 * write_flags is set to 1 for writepage.
3242 		 */
3243 		pm = ps->ps_new_msg(pu, opc, FUSE_WRITE, payload_len, pcr);
3244 		fwi = GET_INPAYLOAD(ps, pm, fuse_write_in);
3245 		fwi->fh = perfuse_get_fh(opc, FWRITE);
3246 		fwi->offset = offset;
3247 		fwi->size = (uint32_t)data_len;
3248 		fwi->write_flags = (fwi->size % sysconf(_SC_PAGESIZE)) ? 0 : 1;
3249 		fwi->lock_owner = pnd->pnd_lock_owner;
3250 		fwi->flags = 0;
3251 		fwi->flags |= (fwi->lock_owner != 0) ? FUSE_WRITE_LOCKOWNER : 0;
3252 		fwi->flags |= (ioflag & IO_DIRECT) ? 0 : FUSE_WRITE_CACHE;
3253 		(void)memcpy((fwi + 1), buf, data_len);
3254 
3255 
3256 #ifdef PERFUSE_DEBUG
3257 		if (perfuse_diagflags & PDF_FH)
3258 			DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
3259 				"fh = 0x%"PRIx64"\n", __func__,
3260 				(void *)opc, pnd->pnd_nodeid, fwi->fh);
3261 #endif
3262 		if ((error = xchg_msg(pu, opc, pm,
3263 				      sizeof(*fwo), wait_reply)) != 0)
3264 			goto out;
3265 
3266 		fwo = GET_OUTPAYLOAD(ps, pm, fuse_write_out);
3267 		written = fwo->size;
3268 		ps->ps_destroy_msg(pm);
3269 
3270 #ifdef PERFUSE_DEBUG
3271 		if (written > *resid)
3272 			DERRX(EX_SOFTWARE, "%s: Unexpected big write %zd",
3273 			      __func__, written);
3274 #endif
3275 		*resid -= written;
3276 		offset += written;
3277 		buf += written;
3278 
3279 	} while (*resid != 0);
3280 
3281 	/*
3282 	 * puffs_ops(3) says
3283 	 *  "everything must be written or an error will be generated"
3284 	 */
3285 	if (*resid != 0)
3286 		error = EFBIG;
3287 
3288 #ifdef PERFUSE_DEBUG
3289 	if (perfuse_diagflags & PDF_RESIZE) {
3290 		if (offset > (off_t)vap->va_size)
3291 			DPRINTF("<< %s %p %" PRIu64 " -> %lld\n", __func__,
3292 				(void *)opc, vap->va_size, (long long)offset);
3293 		else
3294 			DPRINTF("<< %s %p \n", __func__, (void *)opc);
3295 	}
3296 #endif
3297 
3298 	/*
3299 	 * Update file size if we wrote beyond the end
3300 	 */
3301 	if (offset > (off_t)vap->va_size)
3302 		vap->va_size = offset;
3303 
3304 	if (inresize) {
3305 #ifdef PERFUSE_DEBUG
3306 		if (!(pnd->pnd_flags & PND_INRESIZE))
3307 			DERRX(EX_SOFTWARE, "file write grow without resize");
3308 #endif
3309 		pnd->pnd_flags &= ~PND_INRESIZE;
3310 		(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
3311 	}
3312 
3313 
3314 	/*
3315 	 * Statistics
3316 	 */
3317 	if (ioflag & (IO_SYNC|IO_DSYNC))
3318 		ps->ps_syncwrites++;
3319 	else
3320 		ps->ps_asyncwrites++;
3321 
3322 	/*
3323 	 * Remember to sync the file
3324 	 */
3325 	pnd->pnd_flags |= PND_DIRTY;
3326 
3327 #ifdef PERFUSE_DEBUG
3328 	if (perfuse_diagflags & PDF_SYNC)
3329 		DPRINTF("%s: DIRTY opc = %p, file = \"%s\"\n",
3330 			__func__, (void*)opc, perfuse_node_path(ps, opc));
3331 #endif
3332 
3333 out:
3334 	/*
3335 	 * VOP_PUTPAGE causes FAF write where kernel does not
3336 	 * check operation result. At least warn if it failed.
3337 	 */
3338 #ifdef PUFFS_WRITE_FAF
3339 	if (error && (xflag & PUFFS_WRITE_FAF))
3340 		DWARN("Data loss caused by FAF write failed on \"%s\"",
3341 		      pnd->pnd_name);
3342 #endif /* PUFFS_WRITE_FAF */
3343 
3344 	/*
3345 	 * If there are no more queued write, we can resume
3346 	 * an operation awaiting write completion.
3347 	 */
3348 	pnd->pnd_flags &= ~PND_INWRITE;
3349 	if (dequeue_requests(opc, PCQ_WRITE, 1) == 0)
3350 		(void)dequeue_requests(opc, PCQ_AFTERWRITE, DEQUEUE_ALL);
3351 
3352 	node_rele(opc);
3353 	return error;
3354 }
3355 
3356 /* ARGSUSED0 */
3357 void
3358 perfuse_cache_write(struct puffs_usermount *pu, puffs_cookie_t opc, size_t size,
3359 	struct puffs_cacherun *runs)
3360 {
3361 	return;
3362 }
3363 
3364 /* ARGSUSED4 */
3365 int
3366 perfuse_node_getextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3367 	int attrns, const char *attrname, size_t *attrsize, uint8_t *attr,
3368 	size_t *resid, const struct puffs_cred *pcr)
3369 {
3370 	struct perfuse_state *ps;
3371 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3372 	perfuse_msg_t *pm;
3373 	struct fuse_getxattr_in *fgi;
3374 	struct fuse_getxattr_out *fgo;
3375 	struct fuse_out_header *foh;
3376 	size_t attrnamelen;
3377 	size_t len;
3378 	char *np;
3379 	int error;
3380 
3381 	/* system namespace attrs are not accessible to non root users */
3382 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
3383 		return EPERM;
3384 
3385 	node_ref(opc);
3386 	ps = puffs_getspecific(pu);
3387 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3388 	attrnamelen = strlen(attrname) + 1;
3389 	len = sizeof(*fgi) + attrnamelen;
3390 
3391 	pm = ps->ps_new_msg(pu, opc, FUSE_GETXATTR, len, pcr);
3392 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3393 	fgi->size = (unsigned int)((resid != NULL) ? *resid : 0);
3394 	np = (char *)(void *)(fgi + 1);
3395 	(void)strlcpy(np, attrname, attrnamelen);
3396 
3397 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3398 		goto out;
3399 
3400 	/*
3401 	 * We just get fuse_getattr_out with list size if we requested
3402 	 * a null size.
3403 	 */
3404 	if (resid == NULL) {
3405 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3406 
3407 		if (attrsize != NULL)
3408 			*attrsize = fgo->size;
3409 
3410 		ps->ps_destroy_msg(pm);
3411 		error = 0;
3412 		goto out;
3413 	}
3414 
3415 	/*
3416 	 * And with a non null requested size, we get the list just
3417 	 * after the header
3418 	 */
3419 	foh = GET_OUTHDR(ps, pm);
3420 	np = (char *)(void *)(foh + 1);
3421 	len = foh->len - sizeof(*foh);
3422 
3423 	if (attrsize != NULL)
3424 		*attrsize = len;
3425 
3426 	if (resid != NULL) {
3427 		if (*resid < len) {
3428 			error = ERANGE;
3429 			ps->ps_destroy_msg(pm);
3430 			goto out;
3431 		}
3432 
3433 		(void)memcpy(attr, np, len);
3434 		*resid -= len;
3435 	}
3436 
3437 	ps->ps_destroy_msg(pm);
3438 	error = 0;
3439 
3440 out:
3441 	node_rele(opc);
3442 	return error;
3443 }
3444 
3445 int
3446 perfuse_node_setextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3447 	int attrns, const char *attrname, uint8_t *attr, size_t *resid,
3448 	const struct puffs_cred *pcr)
3449 {
3450 	struct perfuse_state *ps;
3451 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3452 	perfuse_msg_t *pm;
3453 	struct fuse_setxattr_in *fsi;
3454 	size_t attrnamelen;
3455 	size_t len;
3456 	char *np;
3457 	int error;
3458 
3459 	/* system namespace attrs are not accessible to non root users */
3460 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
3461 		return EPERM;
3462 
3463 	node_ref(opc);
3464 	ps = puffs_getspecific(pu);
3465 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3466 	attrnamelen = strlen(attrname) + 1;
3467 	len = sizeof(*fsi) + attrnamelen + *resid;
3468 
3469 	pm = ps->ps_new_msg(pu, opc, FUSE_SETXATTR, len, pcr);
3470 	fsi = GET_INPAYLOAD(ps, pm, fuse_setxattr_in);
3471 	fsi->size = (unsigned int)*resid;
3472 	fsi->flags = 0;
3473 	np = (char *)(void *)(fsi + 1);
3474 	(void)strlcpy(np, attrname, attrnamelen);
3475 	np += attrnamelen;
3476 	(void)memcpy(np, (char *)attr, *resid);
3477 
3478 	if ((error = xchg_msg(pu, opc, pm,
3479 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
3480 		goto out;
3481 
3482 	ps->ps_destroy_msg(pm);
3483 	*resid = 0;
3484 	error = 0;
3485 
3486 out:
3487 	node_rele(opc);
3488 	return error;
3489 }
3490 
3491 /* ARGSUSED2 */
3492 int
3493 perfuse_node_listextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3494 	int attrns, size_t *attrsize, uint8_t *attrs, size_t *resid, int flag,
3495 	const struct puffs_cred *pcr)
3496 {
3497 	struct perfuse_state *ps;
3498 	perfuse_msg_t *pm;
3499 	struct fuse_getxattr_in *fgi;
3500 	struct fuse_getxattr_out *fgo;
3501 	struct fuse_out_header *foh;
3502 	char *np;
3503 	size_t len, puffs_len, i, attrlen, outlen;
3504 	int error;
3505 
3506 	/* system namespace attrs are not accessible to non root users */
3507 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
3508 		return EPERM;
3509 
3510 	node_ref(opc);
3511 
3512 	ps = puffs_getspecific(pu);
3513 	len = sizeof(*fgi);
3514 
3515 	pm = ps->ps_new_msg(pu, opc, FUSE_LISTXATTR, len, pcr);
3516 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3517 	if (resid != NULL)
3518 		fgi->size = (unsigned int)*resid;
3519 	else
3520 		fgi->size = 0;
3521 
3522 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3523 		goto out;
3524 
3525 	/*
3526 	 * We just get fuse_getattr_out with list size if we requested
3527 	 * a null size.
3528 	 */
3529 	if (resid == NULL) {
3530 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3531 
3532 		if (attrsize != NULL)
3533 			*attrsize = fgo->size;
3534 
3535 		ps->ps_destroy_msg(pm);
3536 
3537 		error = 0;
3538 		goto out;
3539 	}
3540 
3541 	/*
3542 	 * And with a non null requested size, we get the list just
3543 	 * after the header
3544 	 */
3545 	foh = GET_OUTHDR(ps, pm);
3546 	np = (char *)(void *)(foh + 1);
3547 	puffs_len = foh->len - sizeof(*foh);
3548 
3549 	if (attrsize != NULL)
3550 		*attrsize = puffs_len;
3551 
3552 	if (attrs != NULL) {
3553 		if (*resid < puffs_len) {
3554 			error = ERANGE;
3555 			ps->ps_destroy_msg(pm);
3556 			goto out;
3557 		}
3558 
3559 		outlen = 0;
3560 
3561 		for (i = 0; i < puffs_len; i += attrlen + 1) {
3562 			attrlen = strlen(np + i);
3563 
3564 			/*
3565 			 * Filter attributes per namespace
3566 			 */
3567 			if (!perfuse_ns_match(attrns, np + i))
3568 				continue;
3569 
3570 #ifdef PUFFS_EXTATTR_LIST_LENPREFIX
3571 			/*
3572 			 * Convert the FUSE reply to length prefixed strings
3573 			 * if this is what the kernel wants.
3574 			 */
3575 			if (flag & PUFFS_EXTATTR_LIST_LENPREFIX) {
3576 				(void)memcpy(attrs + outlen + 1,
3577 					     np + i, attrlen);
3578 				*(attrs + outlen) = (uint8_t)attrlen;
3579 			} else
3580 #endif /* PUFFS_EXTATTR_LIST_LENPREFIX */
3581 			(void)memcpy(attrs + outlen, np + i, attrlen + 1);
3582 			outlen += attrlen + 1;
3583 		}
3584 
3585 		*resid -= outlen;
3586 	}
3587 
3588 	ps->ps_destroy_msg(pm);
3589 	error = 0;
3590 
3591 out:
3592 	node_rele(opc);
3593 	return error;
3594 }
3595 
3596 int
3597 perfuse_node_deleteextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3598 	int attrns, const char *attrname, const struct puffs_cred *pcr)
3599 {
3600 	struct perfuse_state *ps;
3601 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3602 	perfuse_msg_t *pm;
3603 	size_t attrnamelen;
3604 	char *np;
3605 	int error;
3606 
3607 	/* system namespace attrs are not accessible to non root users */
3608 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
3609 		return EPERM;
3610 
3611 	node_ref(opc);
3612 
3613 	ps = puffs_getspecific(pu);
3614 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3615 	attrnamelen = strlen(attrname) + 1;
3616 
3617 	pm = ps->ps_new_msg(pu, opc, FUSE_REMOVEXATTR, attrnamelen, pcr);
3618 	np = _GET_INPAYLOAD(ps, pm, char *);
3619 	(void)strlcpy(np, attrname, attrnamelen);
3620 
3621 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
3622 
3623 	ps->ps_destroy_msg(pm);
3624 
3625 	node_rele(opc);
3626 	return error;
3627 }
3628