xref: /netbsd-src/lib/libperfuse/ops.c (revision 62f324d0121177eaf2e0384f92fd9ca2a751c795)
1 /*  $NetBSD: ops.c,v 1.60 2012/11/03 15:43:20 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;
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 
292 	if (sticky && !owner)
293 		return EACCES;
294 
295 	return 0;
296 }
297 
298 
299 static void
300 fuse_attr_to_vap(struct perfuse_state *ps, struct vattr *vap,
301 	struct fuse_attr *fa)
302 {
303 	vap->va_type = IFTOVT(fa->mode);
304 	vap->va_mode = fa->mode & ALLPERMS;
305 	vap->va_nlink = fa->nlink;
306 	vap->va_uid = fa->uid;
307 	vap->va_gid = fa->gid;
308 	vap->va_fsid = (long)ps->ps_fsid;
309 	vap->va_fileid = fa->ino;
310 	vap->va_size = fa->size;
311 	vap->va_blocksize = fa->blksize;
312 	vap->va_atime.tv_sec = (time_t)fa->atime;
313 	vap->va_atime.tv_nsec = (long) fa->atimensec;
314 	vap->va_mtime.tv_sec = (time_t)fa->mtime;
315 	vap->va_mtime.tv_nsec = (long)fa->mtimensec;
316 	vap->va_ctime.tv_sec = (time_t)fa->ctime;
317 	vap->va_ctime.tv_nsec = (long)fa->ctimensec;
318 	vap->va_birthtime.tv_sec = 0;
319 	vap->va_birthtime.tv_nsec = 0;
320 	vap->va_gen = 0;
321 	vap->va_flags = 0;
322 	vap->va_rdev = fa->rdev;
323 	vap->va_bytes = fa->size;
324 	vap->va_filerev = (u_quad_t)PUFFS_VNOVAL;
325 	vap->va_vaflags = 0;
326 
327 	if (vap->va_blocksize == 0)
328 		vap->va_blocksize = DEV_BSIZE;
329 
330 	if (vap->va_size == (size_t)PUFFS_VNOVAL) /* XXX */
331 		vap->va_size = 0;
332 
333 	return;
334 }
335 
336 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
337 static void
338 perfuse_newinfo_setttl(struct puffs_newinfo *pni,
339     struct puffs_node *pn, struct fuse_entry_out *feo,
340     struct fuse_attr_out *fao)
341 {
342 #ifdef PERFUSE_DEBUG
343 	if ((feo == NULL) && (fao == NULL))
344 		DERRX(EX_SOFTWARE, "%s: feo and fao NULL", __func__);
345 
346 	if ((feo != NULL) && (fao != NULL))
347 		DERRX(EX_SOFTWARE, "%s: feo and fao != NULL", __func__);
348 #endif /* PERFUSE_DEBUG */
349 
350 	if (fao != NULL) {
351 		struct timespec va_ttl;
352 
353 		va_ttl.tv_sec = fao->attr_valid;
354 		va_ttl.tv_nsec = fao->attr_valid_nsec;
355 
356 		puffs_newinfo_setvattl(pni, &va_ttl);
357 	}
358 
359 	if (feo != NULL) {
360 		struct timespec va_ttl;
361 		struct timespec cn_ttl;
362 		struct timespec now;
363 		struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(pn);
364 
365 		va_ttl.tv_sec = feo->attr_valid;
366 		va_ttl.tv_nsec = feo->attr_valid_nsec;
367 		cn_ttl.tv_sec = feo->entry_valid;
368 		cn_ttl.tv_nsec = feo->entry_valid_nsec;
369 
370 		puffs_newinfo_setvattl(pni, &va_ttl);
371 		puffs_newinfo_setcnttl(pni, &cn_ttl);
372 
373 		if (clock_gettime(CLOCK_REALTIME, &now) != 0)
374 			DERR(EX_OSERR, "clock_gettime failed");
375 
376                 timespecadd(&now, &cn_ttl, &pnd->pnd_cn_expire);
377 	}
378 
379 	return;
380 }
381 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
382 
383 static int
384 node_lookup_common(struct puffs_usermount *pu, puffs_cookie_t opc,
385 	struct puffs_newinfo *pni, const char *path,
386 	const struct puffs_cred *pcr, struct puffs_node **pnp)
387 {
388 	struct perfuse_state *ps;
389 	struct perfuse_node_data *oldpnd;
390 	perfuse_msg_t *pm;
391 	struct fuse_entry_out *feo;
392 	struct puffs_node *pn;
393 	size_t len;
394 	int error;
395 
396 	/*
397 	 * Prevent further lookups if the parent was removed
398 	 */
399 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
400 		return ESTALE;
401 
402 	if (pnp == NULL)
403 		DERRX(EX_SOFTWARE, "pnp must be != NULL");
404 
405 	ps = puffs_getspecific(pu);
406 
407 #ifdef PERFUSE_DEBUG
408 	if (perfuse_diagflags & PDF_FILENAME)
409 		DPRINTF("%s: opc = %p, file = \"%s\" looking up \"%s\"\n",
410 			__func__, (void *)opc,
411 			perfuse_node_path(ps, opc), path);
412 
413 	if (strcmp(path, ".") == 0)
414 		DERRX(EX_SOFTWARE, "unexpected dot-lookup");
415 
416 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_RECLAIMED)
417 		DERRX(EX_SOFTWARE,
418 		      "looking up reclaimed node opc = %p, name = \"%s\"",
419 		      opc, path);
420 
421 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_INVALID)
422 		DERRX(EX_SOFTWARE,
423 		      "looking up freed node opc = %p, name = \"%s\"",
424 		      opc, path);
425 #endif /* PERFUSE_DEBUG */
426 
427 	len = strlen(path) + 1;
428 	pm = ps->ps_new_msg(pu, opc, FUSE_LOOKUP, len, pcr);
429 	(void)strlcpy(_GET_INPAYLOAD(ps, pm, char *), path, len);
430 
431 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
432 		return error;
433 
434 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
435 
436 	/*
437 	 * Starting with ABI 7.4, inode number 0 means ENOENT,
438 	 * with entry_valid / entry_valid_nsec giving negative
439 	 * cache timeout (which we do not implement yet).
440 	 */
441 	if (feo->attr.ino == 0) {
442 		ps->ps_destroy_msg(pm);
443 		return ENOENT;
444 	}
445 
446 	/*
447 	 * Check for a known node, not reclaimed, with another name.
448 	 * It may have been moved, or we can lookup ../
449 	 */
450 	if (((oldpnd = perfuse_node_bynodeid(ps, feo->nodeid)) != NULL) &&
451 	    !(oldpnd->pnd_flags & PND_RECLAIMED)) {
452 		/*
453 		 * Save the new node name if not ..
454 		 */
455 		if (strncmp(path, "..", len) != 0)
456 			(void)strlcpy(oldpnd->pnd_name,
457 				      path, MAXPATHLEN);
458 		pn = oldpnd->pnd_pn;
459 
460 	} else {
461 		pn = perfuse_new_pn(pu, path, opc);
462 		PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
463 		perfuse_node_cache(ps, pn);
464 	}
465 
466 #ifdef PERFUSE_DEBUG
467 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_RECLAIMED)
468 		DERRX(EX_SOFTWARE,
469 		      "reclaimed in lookup opc = %p, name = \"%s\", ck = %p",
470 		      opc, path, pn);
471 
472 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_INVALID)
473 		DERRX(EX_SOFTWARE,
474 		      "freed in lookup opc = %p, name = \"%s\", ck = %p",
475 		      opc, path, pn);
476 #endif /* PERFUSE_DEBUG */
477 
478 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
479 	pn->pn_va.va_gen = (u_long)(feo->generation);
480 	PERFUSE_NODE_DATA(pn)->pnd_fuse_nlookup++;
481 
482 	*pnp = pn;
483 
484 #ifdef PERFUSE_DEBUG
485 	if (perfuse_diagflags & PDF_FILENAME)
486 		DPRINTF("%s: opc = %p, looked up opc = %p, "
487 			"nodeid = 0x%"PRIx64" file = \"%s\"\n", __func__,
488 			(void *)opc, pn, feo->nodeid, path);
489 #endif
490 
491 	if (pni != NULL) {
492 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
493 		puffs_newinfo_setva(pni, &pn->pn_va);
494 		perfuse_newinfo_setttl(pni, pn, feo, NULL);
495 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
496 		puffs_newinfo_setcookie(pni, pn);
497 		puffs_newinfo_setvtype(pni, pn->pn_va.va_type);
498 		puffs_newinfo_setsize(pni, (voff_t)pn->pn_va.va_size);
499 		puffs_newinfo_setrdev(pni, pn->pn_va.va_rdev);
500 	}
501 
502 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_NODELEAK) {
503 		PERFUSE_NODE_DATA(pn)->pnd_flags &= ~PND_NODELEAK;
504 		ps->ps_nodeleakcount--;
505 	}
506 
507 	ps->ps_destroy_msg(pm);
508 
509 	return 0;
510 }
511 
512 
513 /*
514  * Common code for methods that create objects:
515  * perfuse_node_mkdir
516  * perfuse_node_mknod
517  * perfuse_node_symlink
518  */
519 static int
520 node_mk_common(struct puffs_usermount *pu, puffs_cookie_t opc,
521 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
522 	perfuse_msg_t *pm)
523 {
524 	struct perfuse_state *ps;
525 	struct puffs_node *pn;
526 	struct fuse_entry_out *feo;
527 	int error;
528 
529 	ps =  puffs_getspecific(pu);
530 
531 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
532 		return error;
533 
534 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
535 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
536 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
537 
538 	pn = perfuse_new_pn(pu, pcn->pcn_name, opc);
539 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
540 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
541 	perfuse_node_cache(ps, pn);
542 
543 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
544 	pn->pn_va.va_gen = (u_long)(feo->generation);
545 
546 	puffs_newinfo_setcookie(pni, pn);
547 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
548 	puffs_newinfo_setva(pni, &pn->pn_va);
549 	perfuse_newinfo_setttl(pni, pn, feo, NULL);
550 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
551 
552 
553 #ifdef PERFUSE_DEBUG
554 	if (perfuse_diagflags & PDF_FILENAME)
555 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
556 			"nodeid = 0x%"PRIx64"\n",
557 			__func__, (void *)pn, pcn->pcn_name,
558 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid);
559 #endif
560 	ps->ps_destroy_msg(pm);
561 
562 	/* Parents is now dirty */
563 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
564 
565 	return 0;
566 }
567 
568 static uint64_t
569 readdir_last_cookie(struct fuse_dirent *fd, size_t fd_len)
570 {
571 	size_t len;
572 	size_t seen = 0;
573 	char *ndp;
574 
575 	do {
576 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
577 		seen += len;
578 
579 		if (seen >= fd_len)
580 			break;
581 
582 		ndp = (char *)(void *)fd + (size_t)len;
583 		fd = (struct fuse_dirent *)(void *)ndp;
584 	} while (1 /* CONSTCOND */);
585 
586 	return fd->off;
587 }
588 
589 static ssize_t
590 fuse_to_dirent(struct puffs_usermount *pu, puffs_cookie_t opc,
591 	struct fuse_dirent *fd, size_t fd_len)
592 {
593 	struct dirent *dents;
594 	size_t dents_len;
595 	ssize_t written;
596 	uint64_t fd_offset;
597 	struct fuse_dirent *fd_base;
598 	size_t len;
599 
600 	fd_base = fd;
601 	fd_offset = 0;
602 	written = 0;
603 	dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
604 	dents_len = (size_t)PERFUSE_NODE_DATA(opc)->pnd_dirent_len;
605 
606 	do {
607 		char *ndp;
608 		size_t reclen;
609 
610 		reclen = _DIRENT_RECLEN(dents, fd->namelen);
611 
612 		/*
613 		 * Check we do not overflow the output buffer
614 		 * struct fuse_dirent is bigger than struct dirent,
615 		 * so we should always use fd_len and never reallocate
616 		 * later.
617 		 * If we have to reallocate,try to double the buffer
618 		 * each time so that we do not have to do it too often.
619 		 */
620 		if (written + reclen > dents_len) {
621 			if (dents_len == 0)
622 				dents_len = fd_len;
623 			else
624 				dents_len =
625 				   MAX(2 * dents_len, written + reclen);
626 
627 			dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
628 			if ((dents = realloc(dents, dents_len)) == NULL)
629 				DERR(EX_OSERR, "%s: malloc failed", __func__);
630 
631 			PERFUSE_NODE_DATA(opc)->pnd_dirent = dents;
632 			PERFUSE_NODE_DATA(opc)->pnd_dirent_len = dents_len;
633 
634 			/*
635 			 * (void *) for delint
636 			 */
637 			ndp = (char *)(void *)dents + written;
638 			dents = (struct dirent *)(void *)ndp;
639 		}
640 
641 		/*
642 		 * Filesystem was mounted without -o use_ino
643 		 * Perform a lookup to find it.
644 		 */
645 		if (fd->ino == PERFUSE_UNKNOWN_INO) {
646 			struct puffs_node *pn;
647 
648 			if (node_lookup_common(pu, opc, NULL, fd->name,
649 					       NULL, &pn) != 0) {
650 				DWARNX("node_lookup_common failed");
651 			} else {
652 				fd->ino = pn->pn_va.va_fileid;
653 				(void)perfuse_node_reclaim(pu, pn);
654 			}
655 		}
656 
657 		dents->d_fileno = fd->ino;
658 		dents->d_reclen = (unsigned short)reclen;
659 		dents->d_namlen = fd->namelen;
660 		dents->d_type = fd->type;
661 		strlcpy(dents->d_name, fd->name, fd->namelen + 1);
662 
663 #ifdef PERFUSE_DEBUG
664 		if (perfuse_diagflags & PDF_READDIR)
665 			DPRINTF("%s: translated \"%s\" ino = %"PRIu64"\n",
666 				__func__, dents->d_name, dents->d_fileno);
667 #endif
668 
669 		dents = _DIRENT_NEXT(dents);
670 		written += reclen;
671 
672 		/*
673 		 * Move to the next record.
674 		 * fd->off is not the offset, it is an opaque cookie
675 		 * given by the filesystem to keep state across multiple
676 		 * readdir() operation.
677 		 * Use record alignement instead.
678 		 */
679 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
680 #ifdef PERFUSE_DEBUG
681 		if (perfuse_diagflags & PDF_READDIR)
682 			DPRINTF("%s: record at %"PRId64"/0x%"PRIx64" "
683 				"length = %zd/0x%zx. "
684 				"next record at %"PRId64"/0x%"PRIx64" "
685 				"max %zd/0x%zx\n",
686 				__func__, fd_offset, fd_offset, len, len,
687 				fd_offset + len, fd_offset + len,
688 				fd_len, fd_len);
689 #endif
690 		fd_offset += len;
691 
692 		/*
693 		 * Check if next record is still within the packet
694 		 * If it is not, we reached the end of the buffer.
695 		 */
696 		if (fd_offset >= fd_len)
697 			break;
698 
699 		/*
700 		 * (void *) for delint
701 		 */
702 		ndp = (char *)(void *)fd_base + (size_t)fd_offset;
703 		fd = (struct fuse_dirent *)(void *)ndp;
704 
705 	} while (1 /* CONSTCOND */);
706 
707 	/*
708 	 * Adjust the dirent output length
709 	 */
710 	if (written != -1)
711 		PERFUSE_NODE_DATA(opc)->pnd_dirent_len = written;
712 
713 	return written;
714 }
715 
716 static void
717 readdir_buffered(puffs_cookie_t opc, struct dirent *dent, off_t *readoff,
718 	size_t *reslen)
719 {
720 	struct dirent *fromdent;
721 	struct perfuse_node_data *pnd;
722 	char *ndp;
723 
724 	pnd = PERFUSE_NODE_DATA(opc);
725 
726 	while (*readoff < pnd->pnd_dirent_len) {
727 		/*
728 		 * (void *) for delint
729 		 */
730 		ndp = (char *)(void *)pnd->pnd_dirent + (size_t)*readoff;
731 		fromdent = (struct dirent *)(void *)ndp;
732 
733 		if (*reslen < _DIRENT_SIZE(fromdent))
734 			break;
735 
736 		memcpy(dent, fromdent, _DIRENT_SIZE(fromdent));
737 		*readoff += _DIRENT_SIZE(fromdent);
738 		*reslen -= _DIRENT_SIZE(fromdent);
739 
740 		dent = _DIRENT_NEXT(dent);
741 	}
742 
743 #ifdef PERFUSE_DEBUG
744 	if (perfuse_diagflags & PDF_READDIR)
745 		DPRINTF("%s: readoff = %"PRId64",  "
746 			"pnd->pnd_dirent_len = %"PRId64"\n",
747 			__func__, *readoff, pnd->pnd_dirent_len);
748 #endif
749 	if (*readoff >=  pnd->pnd_dirent_len) {
750 		free(pnd->pnd_dirent);
751 		pnd->pnd_dirent = NULL;
752 		pnd->pnd_dirent_len = 0;
753 	}
754 
755 	return;
756 }
757 
758 
759 static void
760 node_ref(puffs_cookie_t opc)
761 {
762 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
763 
764 #ifdef PERFUSE_DEBUG
765 	if (pnd->pnd_flags & PND_INVALID)
766 		DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
767 #endif /* PERFUSE_DEBUG */
768 
769 	pnd->pnd_ref++;
770 	return;
771 }
772 
773 static void
774 node_rele(puffs_cookie_t opc)
775 {
776 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
777 
778 #ifdef PERFUSE_DEBUG
779 	if (pnd->pnd_flags & PND_INVALID)
780 		DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
781 #endif /* PERFUSE_DEBUG */
782 
783 	pnd->pnd_ref--;
784 
785 	if (pnd->pnd_ref == 0)
786 		(void)dequeue_requests(opc, PCQ_REF, DEQUEUE_ALL);
787 
788 	return;
789 }
790 
791 static void
792 requeue_request(struct puffs_usermount *pu, puffs_cookie_t opc,
793 	enum perfuse_qtype type)
794 {
795 	struct perfuse_cc_queue pcq;
796 	struct perfuse_node_data *pnd;
797 #ifdef PERFUSE_DEBUG
798 	struct perfuse_state *ps;
799 
800 	ps = perfuse_getspecific(pu);
801 #endif
802 
803 	pnd = PERFUSE_NODE_DATA(opc);
804 	pcq.pcq_type = type;
805 	pcq.pcq_cc = puffs_cc_getcc(pu);
806 	TAILQ_INSERT_TAIL(&pnd->pnd_pcq, &pcq, pcq_next);
807 
808 #ifdef PERFUSE_DEBUG
809 	if (perfuse_diagflags & PDF_REQUEUE)
810 		DPRINTF("%s: REQUEUE opc = %p, pcc = %p (%s)\n",
811 		        __func__, (void *)opc, pcq.pcq_cc,
812 			perfuse_qtypestr[type]);
813 #endif
814 
815 	puffs_cc_yield(pcq.pcq_cc);
816 	TAILQ_REMOVE(&pnd->pnd_pcq, &pcq, pcq_next);
817 
818 #ifdef PERFUSE_DEBUG
819 	if (perfuse_diagflags & PDF_REQUEUE)
820 		DPRINTF("%s: RESUME opc = %p, pcc = %p (%s)\n",
821 		        __func__, (void *)opc, pcq.pcq_cc,
822 			perfuse_qtypestr[type]);
823 #endif
824 
825 	return;
826 }
827 
828 static int
829 dequeue_requests(puffs_cookie_t opc, enum perfuse_qtype type, int max)
830 {
831 	struct perfuse_cc_queue *pcq;
832 	struct perfuse_node_data *pnd;
833 	int dequeued;
834 
835 	pnd = PERFUSE_NODE_DATA(opc);
836 	dequeued = 0;
837 	TAILQ_FOREACH(pcq, &pnd->pnd_pcq, pcq_next) {
838 		if (pcq->pcq_type != type)
839 			continue;
840 
841 #ifdef PERFUSE_DEBUG
842 		if (perfuse_diagflags & PDF_REQUEUE)
843 			DPRINTF("%s: SCHEDULE opc = %p, pcc = %p (%s)\n",
844 				__func__, (void *)opc, pcq->pcq_cc,
845 				 perfuse_qtypestr[type]);
846 #endif
847 		puffs_cc_schedule(pcq->pcq_cc);
848 
849 		if (++dequeued == max)
850 			break;
851 	}
852 
853 #ifdef PERFUSE_DEBUG
854 	if (perfuse_diagflags & PDF_REQUEUE)
855 		DPRINTF("%s: DONE  opc = %p\n", __func__, (void *)opc);
856 #endif
857 
858 	return dequeued;
859 }
860 
861 void
862 perfuse_fs_init(struct puffs_usermount *pu)
863 {
864 	struct perfuse_state *ps;
865 	perfuse_msg_t *pm;
866 	struct fuse_init_in *fii;
867 	struct fuse_init_out *fio;
868 	int error;
869 
870 	ps = puffs_getspecific(pu);
871 
872         if (puffs_mount(pu, ps->ps_target, ps->ps_mountflags, ps->ps_root) != 0)
873                 DERR(EX_OSERR, "%s: puffs_mount failed", __func__);
874 
875 	/*
876 	 * Linux 2.6.34.1 sends theses flags:
877 	 * FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC
878 	 * FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK
879 	 *
880 	 * Linux also sets max_readahead at 32 pages (128 kB)
881 	 *
882 	 * ps_new_msg() is called with NULL creds, which will
883 	 * be interpreted as FUSE superuser.
884 	 */
885 	pm = ps->ps_new_msg(pu, 0, FUSE_INIT, sizeof(*fii), NULL);
886 	fii = GET_INPAYLOAD(ps, pm, fuse_init_in);
887 	fii->major = FUSE_KERNEL_VERSION;
888 	fii->minor = FUSE_KERNEL_MINOR_VERSION;
889 	fii->max_readahead = (unsigned int)(32 * sysconf(_SC_PAGESIZE));
890 	fii->flags = (FUSE_ASYNC_READ|FUSE_POSIX_LOCKS|FUSE_ATOMIC_O_TRUNC);
891 
892 	if ((error = xchg_msg(pu, 0, pm, sizeof(*fio), wait_reply)) != 0)
893 		DERRX(EX_SOFTWARE, "init message exchange failed (%d)", error);
894 
895 	fio = GET_OUTPAYLOAD(ps, pm, fuse_init_out);
896 	ps->ps_max_readahead = fio->max_readahead;
897 	ps->ps_max_write = fio->max_write;
898 
899 	ps->ps_destroy_msg(pm);
900 
901 	return;
902 }
903 
904 int
905 perfuse_fs_unmount(struct puffs_usermount *pu, int flags)
906 {
907 	perfuse_msg_t *pm;
908 	struct perfuse_state *ps;
909 	puffs_cookie_t opc;
910 	int error;
911 
912 	ps = puffs_getspecific(pu);
913 	opc = (puffs_cookie_t)puffs_getroot(pu);
914 
915 	/*
916 	 * ps_new_msg() is called with NULL creds, which will
917 	 * be interpreted as FUSE superuser.
918 	 */
919 	pm = ps->ps_new_msg(pu, opc, FUSE_DESTROY, 0, NULL);
920 
921 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0){
922 		DWARN("unmount %s", ps->ps_target);
923 		if (!(flags & MNT_FORCE))
924 			return error;
925 		else
926 			error = 0;
927 	} else {
928 		ps->ps_destroy_msg(pm);
929 	}
930 
931 	ps->ps_umount(pu);
932 
933 	if (perfuse_diagflags & PDF_MISC)
934 		DPRINTF("%s unmounted, exit\n", ps->ps_target);
935 
936 	return 0;
937 }
938 
939 int
940 perfuse_fs_statvfs(struct puffs_usermount *pu, struct statvfs *svfsb)
941 {
942 	struct perfuse_state *ps;
943 	perfuse_msg_t *pm;
944 	puffs_cookie_t opc;
945 	struct fuse_statfs_out *fso;
946 	int error;
947 
948 	ps = puffs_getspecific(pu);
949 	opc = (puffs_cookie_t)puffs_getroot(pu);
950 
951 	/*
952 	 * ps_new_msg() is called with NULL creds, which will
953 	 * be interpreted as FUSE superuser.
954 	 */
955 	pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
956 
957 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply)) != 0)
958 		return error;
959 
960 	fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
961 	svfsb->f_flag = ps->ps_mountflags;
962 	svfsb->f_bsize = fso->st.bsize;
963 	svfsb->f_frsize = fso->st.frsize;
964 	svfsb->f_iosize = ((struct puffs_node *)opc)->pn_va.va_blocksize;
965 	svfsb->f_blocks = fso->st.blocks;
966 	svfsb->f_bfree = fso->st.bfree;
967 	svfsb->f_bavail = fso->st.bavail;
968 	svfsb->f_bresvd = fso->st.bfree - fso->st.bavail;
969 	svfsb->f_files = fso->st.files;
970 	svfsb->f_ffree = fso->st.ffree;
971 	svfsb->f_favail = fso->st.ffree;/* files not reserved for root */
972 	svfsb->f_fresvd = 0;		/* files reserved for root */
973 
974 	svfsb->f_syncreads = ps->ps_syncreads;
975 	svfsb->f_syncwrites = ps->ps_syncwrites;
976 
977 	svfsb->f_asyncreads = ps->ps_asyncreads;
978 	svfsb->f_asyncwrites = ps->ps_asyncwrites;
979 
980 	(void)memcpy(&svfsb->f_fsidx, &ps->ps_fsid, sizeof(ps->ps_fsid));
981 	svfsb->f_fsid = (unsigned long)ps->ps_fsid;
982 	svfsb->f_namemax = MAXPATHLEN;	/* XXX */
983 	svfsb->f_owner = ps->ps_owner_uid;
984 
985 	(void)strlcpy(svfsb->f_mntonname, ps->ps_target, _VFS_NAMELEN);
986 
987 	if (ps->ps_filesystemtype != NULL)
988 		(void)strlcpy(svfsb->f_fstypename,
989 			      ps->ps_filesystemtype, _VFS_NAMELEN);
990 	else
991 		(void)strlcpy(svfsb->f_fstypename, "fuse", _VFS_NAMELEN);
992 
993 	if (ps->ps_source != NULL)
994 		strlcpy(svfsb->f_mntfromname, ps->ps_source, _VFS_NAMELEN);
995 	else
996 		strlcpy(svfsb->f_mntfromname, _PATH_FUSE, _VFS_NAMELEN);
997 
998 	ps->ps_destroy_msg(pm);
999 
1000 	return 0;
1001 }
1002 
1003 int
1004 perfuse_fs_sync(struct puffs_usermount *pu, int waitfor,
1005 	const struct puffs_cred *pcr)
1006 {
1007 	/*
1008 	 * FUSE does not seem to have a FS sync callback.
1009 	 * Maybe do not even register this callback
1010 	 */
1011 	return puffs_fsnop_sync(pu, waitfor, pcr);
1012 }
1013 
1014 /* ARGSUSED0 */
1015 int
1016 perfuse_fs_fhtonode(struct puffs_usermount *pu, void *fid, size_t fidsize,
1017 	struct puffs_newinfo *pni)
1018 {
1019 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1020 	return 0;
1021 }
1022 
1023 /* ARGSUSED0 */
1024 int
1025 perfuse_fs_nodetofh(struct puffs_usermount *pu, puffs_cookie_t cookie,
1026 	void *fid, size_t *fidsize)
1027 {
1028 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1029 	return 0;
1030 }
1031 
1032 #if 0
1033 /* ARGSUSED0 */
1034 void
1035 perfuse_fs_extattrctl(struct puffs_usermount *pu, int cmd,
1036 	puffs_cookie_t *cookie, int flags, int namespace, const char *attrname)
1037 {
1038 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
1039 	return 0;
1040 }
1041 #endif /* 0 */
1042 
1043 /* ARGSUSED0 */
1044 void
1045 perfuse_fs_suspend(struct puffs_usermount *pu, int status)
1046 {
1047 	return;
1048 }
1049 
1050 
1051 int
1052 perfuse_node_lookup(struct puffs_usermount *pu, puffs_cookie_t opc,
1053 	struct puffs_newinfo *pni, const struct puffs_cn *pcn)
1054 {
1055 	struct perfuse_state *ps;
1056 	struct puffs_node *pn;
1057 	mode_t mode;
1058 	int error;
1059 
1060 	ps = puffs_getspecific(pu);
1061 	node_ref(opc);
1062 
1063 	/*
1064 	 * Check permissions
1065 	 */
1066 	switch(pcn->pcn_nameiop) {
1067 	case NAMEI_DELETE: /* FALLTHROUGH */
1068 	case NAMEI_RENAME: /* FALLTHROUGH */
1069 	case NAMEI_CREATE:
1070 		if (pcn->pcn_flags & NAMEI_ISLASTCN)
1071 			mode = PUFFS_VEXEC|PUFFS_VWRITE;
1072 		else
1073 			mode = PUFFS_VEXEC;
1074 		break;
1075 	case NAMEI_LOOKUP: /* FALLTHROUGH */
1076 	default:
1077 		mode = PUFFS_VEXEC;
1078 		break;
1079 	}
1080 
1081 	if ((error = mode_access(opc, pcn->pcn_cred, mode)) != 0)
1082 		goto out;
1083 
1084 	error = node_lookup_common(pu, (puffs_cookie_t)opc, pni,
1085 				   pcn->pcn_name, pcn->pcn_cred, &pn);
1086 
1087 	if (error != 0)
1088 		goto out;
1089 
1090 	/*
1091 	 * Kernel would kill us if the filesystem returned the parent
1092 	 * itself. If we want to live, hide that!
1093 	 */
1094 	if ((opc == (puffs_cookie_t)pn) && (strcmp(pcn->pcn_name, ".") != 0)) {
1095 		DERRX(EX_SOFTWARE, "lookup \"%s\" in \"%s\" returned parent",
1096 		      pcn->pcn_name, perfuse_node_path(ps, opc));
1097 		/* NOTREACHED */
1098 		error = ESTALE;
1099 		goto out;
1100 	}
1101 
1102 	/*
1103 	 * Removed node
1104 	 */
1105 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_REMOVED) {
1106 		error = ENOENT;
1107 		goto out;
1108 	}
1109 
1110 	/*
1111 	 * Check for sticky bit. Unfortunately there is no way to
1112 	 * do this before creating the puffs_node, since we require
1113 	 * this operation to get the node owner.
1114 	 */
1115 	switch (pcn->pcn_nameiop) {
1116 	case NAMEI_DELETE: /* FALLTHROUGH */
1117 	case NAMEI_RENAME:
1118 		error = sticky_access(opc, pn, pcn->pcn_cred);
1119 		if (error != 0) {
1120 			(void)perfuse_node_reclaim(pu, pn);
1121 			goto out;
1122 		}
1123 		break;
1124 	default:
1125 		break;
1126 	}
1127 
1128 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
1129 
1130 	error = 0;
1131 
1132 out:
1133 	node_rele(opc);
1134 	return error;
1135 }
1136 
1137 int
1138 perfuse_node_create(struct puffs_usermount *pu, puffs_cookie_t opc,
1139 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
1140 	const struct vattr *vap)
1141 {
1142 	perfuse_msg_t *pm;
1143 	struct perfuse_state *ps;
1144 	struct fuse_create_in *fci;
1145 	struct fuse_entry_out *feo;
1146 	struct fuse_open_out *foo;
1147 	struct puffs_node *pn;
1148 	const char *name;
1149 	size_t namelen;
1150 	size_t len;
1151 	int error;
1152 
1153 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1154 		return ENOENT;
1155 
1156 	node_ref(opc);
1157 
1158 	/*
1159 	 * If create is unimplemented: Check that it does not
1160 	 * already exists, and if not, do mknod and open
1161 	 */
1162 	ps = puffs_getspecific(pu);
1163 	if (ps->ps_flags & PS_NO_CREAT) {
1164 		error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
1165 					   pcn->pcn_cred, &pn);
1166 		if (error == 0)	{
1167 			(void)perfuse_node_reclaim(pu, pn);
1168 			error = EEXIST;
1169 			goto out;
1170 		}
1171 
1172 		error = perfuse_node_mknod(pu, opc, pni, pcn, vap);
1173 		if (error != 0)
1174 			goto out;
1175 
1176 		error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
1177 					   pcn->pcn_cred, &pn);
1178 		if (error != 0)
1179 			goto out;
1180 
1181 		/*
1182 		 * FUSE does the open at create time, while
1183 		 * NetBSD will open in a subsequent operation.
1184 		 * We need to open now, in order to retain FUSE
1185 		 * semantics. The calling process will not get
1186 		 * a file descriptor before the kernel sends
1187 		 * the open operation.
1188 		 */
1189 		error = perfuse_node_open(pu, (puffs_cookie_t)pn,
1190 					  FWRITE, pcn->pcn_cred);
1191 		goto out;
1192 	}
1193 
1194 	name = pcn->pcn_name;
1195 	namelen = pcn->pcn_namelen + 1;
1196 	len = sizeof(*fci) + namelen;
1197 
1198 	/*
1199 	 * flags should use O_WRONLY instead of O_RDWR, but it
1200 	 * breaks when the caller tries to read from file.
1201 	 *
1202 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1203 	 */
1204 	pm = ps->ps_new_msg(pu, opc, FUSE_CREATE, len, pcn->pcn_cred);
1205 	fci = GET_INPAYLOAD(ps, pm, fuse_create_in);
1206 	fci->flags = O_CREAT | O_TRUNC | O_RDWR;
1207 	fci->mode = vap->va_mode | VTTOIF(vap->va_type);
1208 	fci->umask = 0; 	/* Seems unused by libfuse */
1209 	(void)strlcpy((char*)(void *)(fci + 1), name, namelen);
1210 
1211 	len = sizeof(*feo) + sizeof(*foo);
1212 	if ((error = xchg_msg(pu, opc, pm, len, wait_reply)) != 0) {
1213 		/*
1214 		 * create is unimplmented, remember it for later,
1215 		 * and start over using mknod and open instead.
1216 		 */
1217 		if (error == ENOSYS) {
1218 			ps->ps_flags |= PS_NO_CREAT;
1219 			error = perfuse_node_create(pu, opc, pni, pcn, vap);
1220 		}
1221 
1222 		goto out;
1223 	}
1224 
1225 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
1226 	foo = (struct fuse_open_out *)(void *)(feo + 1);
1227 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
1228 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
1229 
1230 	/*
1231 	 * Save the file handle and inode in node private data
1232 	 * so that we can reuse it later
1233 	 */
1234 	pn = perfuse_new_pn(pu, name, opc);
1235 	perfuse_new_fh((puffs_cookie_t)pn, foo->fh, FWRITE);
1236 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
1237 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
1238 	perfuse_node_cache(ps, pn);
1239 
1240 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
1241 	pn->pn_va.va_gen = (u_long)(feo->generation);
1242 
1243 	puffs_newinfo_setcookie(pni, pn);
1244 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
1245 	puffs_newinfo_setva(pni, &pn->pn_va);
1246 	perfuse_newinfo_setttl(pni, pn, feo, NULL);
1247 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
1248 
1249 #ifdef PERFUSE_DEBUG
1250 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1251 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
1252 			"nodeid = 0x%"PRIx64", wfh = 0x%"PRIx64"\n",
1253 			__func__, (void *)pn, pcn->pcn_name,
1254 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid,
1255 			foo->fh);
1256 #endif
1257 
1258 	ps->ps_destroy_msg(pm);
1259 	error = 0;
1260 
1261 out:
1262 	node_rele(opc);
1263 	return error;
1264 }
1265 
1266 
1267 int
1268 perfuse_node_mknod(struct puffs_usermount *pu, puffs_cookie_t opc,
1269 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
1270 	const struct vattr *vap)
1271 {
1272 	struct perfuse_state *ps;
1273 	perfuse_msg_t *pm;
1274 	struct fuse_mknod_in *fmi;
1275 	const char* path;
1276 	size_t len;
1277 	int error;
1278 
1279 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1280 		return ENOENT;
1281 
1282 	node_ref(opc);
1283 
1284 	/*
1285 	 * Only superuser can mknod objects other than
1286 	 * directories, files, socks, fifo and links.
1287 	 *
1288 	 * Create an object require -WX permission in the parent directory
1289 	 */
1290 	switch (vap->va_type) {
1291 	case VDIR:	/* FALLTHROUGH */
1292 	case VREG:	/* FALLTHROUGH */
1293 	case VFIFO:	/* FALLTHROUGH */
1294 	case VSOCK:
1295 		break;
1296 	default:	/* VNON, VBLK, VCHR, VBAD */
1297 		if (!puffs_cred_isjuggernaut(pcn->pcn_cred)) {
1298 			error = EACCES;
1299 			goto out;
1300 		}
1301 		break;
1302 	}
1303 
1304 
1305 	ps = puffs_getspecific(pu);
1306 	path = pcn->pcn_name;
1307 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
1308 
1309 	/*
1310 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
1311 	 */
1312 	pm = ps->ps_new_msg(pu, opc, FUSE_MKNOD, len, pcn->pcn_cred);
1313 	fmi = GET_INPAYLOAD(ps, pm, fuse_mknod_in);
1314 	fmi->mode = vap->va_mode | VTTOIF(vap->va_type);
1315 	fmi->rdev = (uint32_t)vap->va_rdev;
1316 	fmi->umask = 0; 	/* Seems unused bu libfuse */
1317 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
1318 
1319 	error = node_mk_common(pu, opc, pni, pcn, pm);
1320 
1321 out:
1322 	node_rele(opc);
1323 	return error;
1324 }
1325 
1326 
1327 int
1328 perfuse_node_open(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
1329 	const struct puffs_cred *pcr)
1330 {
1331 	struct perfuse_state *ps;
1332 	struct perfuse_node_data *pnd;
1333 	perfuse_msg_t *pm;
1334 	mode_t fmode;
1335 	int op;
1336 	struct fuse_open_in *foi;
1337 	struct fuse_open_out *foo;
1338 	struct puffs_node *pn;
1339 	int error;
1340 
1341 	ps = puffs_getspecific(pu);
1342 	pn = (struct puffs_node *)opc;
1343 	pnd = PERFUSE_NODE_DATA(opc);
1344 	error = 0;
1345 
1346 	if (pnd->pnd_flags & PND_REMOVED)
1347 		return ENOENT;
1348 
1349 	node_ref(opc);
1350 
1351 	if (puffs_pn_getvap(pn)->va_type == VDIR)
1352 		op = FUSE_OPENDIR;
1353 	else
1354 		op = FUSE_OPEN;
1355 
1356 	/*
1357 	 * libfuse docs says
1358 	 * - O_CREAT and O_EXCL should never be set.
1359 	 * - O_TRUNC may be used if mount option atomic_o_trunc is used XXX
1360 	 *
1361 	 * O_APPEND makes no sense since FUSE always sends
1362 	 * the file offset for write operations. If the
1363 	 * filesystem uses pwrite(), O_APPEND would cause
1364 	 * the offset to be ignored and cause file corruption.
1365 	 */
1366 	mode &= ~(O_CREAT|O_EXCL|O_APPEND);
1367 
1368 	/*
1369 	 * Do not open twice, and do not reopen for reading
1370 	 * if we already have write handle.
1371 	 */
1372 	if (((mode & FREAD) && (pnd->pnd_flags & PND_RFH)) ||
1373 	    ((mode & FREAD) && (pnd->pnd_flags & PND_WFH)) ||
1374 	    ((mode & FWRITE) && (pnd->pnd_flags & PND_WFH)))
1375 		goto out;
1376 
1377 	/*
1378 	 * Queue open on a node so that we do not open
1379 	 * twice. This would be better with read and
1380 	 * write distinguished.
1381 	 */
1382 	while (pnd->pnd_flags & PND_INOPEN)
1383 		requeue_request(pu, opc, PCQ_OPEN);
1384 	pnd->pnd_flags |= PND_INOPEN;
1385 
1386 	/*
1387 	 * Convert PUFFS mode to FUSE mode: convert FREAD/FWRITE
1388 	 * to O_RDONLY/O_WRONLY while perserving the other options.
1389 	 */
1390 	fmode = mode & ~(FREAD|FWRITE);
1391 	fmode |= (mode & FWRITE) ? O_RDWR : O_RDONLY;
1392 
1393 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*foi), pcr);
1394 	foi = GET_INPAYLOAD(ps, pm, fuse_open_in);
1395 	foi->flags = fmode;
1396 	foi->unused = 0;
1397 
1398 	if ((error = xchg_msg(pu, opc, pm, sizeof(*foo), wait_reply)) != 0)
1399 		goto out;
1400 
1401 	foo = GET_OUTPAYLOAD(ps, pm, fuse_open_out);
1402 
1403 	/*
1404 	 * Save the file handle in node private data
1405 	 * so that we can reuse it later
1406 	 */
1407 	perfuse_new_fh(opc, foo->fh, mode);
1408 
1409 #ifdef PERFUSE_DEBUG
1410 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
1411 		DPRINTF("%s: opc = %p, file = \"%s\", "
1412 			"nodeid = 0x%"PRIx64", %s%sfh = 0x%"PRIx64"\n",
1413 			__func__, (void *)opc, perfuse_node_path(ps, opc),
1414 			pnd->pnd_nodeid, mode & FREAD ? "r" : "",
1415 			mode & FWRITE ? "w" : "", foo->fh);
1416 #endif
1417 
1418 	ps->ps_destroy_msg(pm);
1419 out:
1420 
1421 	pnd->pnd_flags &= ~PND_INOPEN;
1422 	(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
1423 
1424 	node_rele(opc);
1425 	return error;
1426 }
1427 
1428 /* ARGSUSED0 */
1429 int
1430 perfuse_node_close(struct puffs_usermount *pu, puffs_cookie_t opc, int flags,
1431 	const struct puffs_cred *pcr)
1432 {
1433 	struct perfuse_node_data *pnd;
1434 
1435 	pnd = PERFUSE_NODE_DATA(opc);
1436 
1437 	if (!(pnd->pnd_flags & PND_OPEN))
1438 		return EBADF;
1439 
1440 	/*
1441 	 * Actual close is postponed at inactive time.
1442 	 */
1443 	return 0;
1444 }
1445 
1446 int
1447 perfuse_node_access(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
1448 	const struct puffs_cred *pcr)
1449 {
1450 	perfuse_msg_t *pm;
1451 	struct perfuse_state *ps;
1452 	struct fuse_access_in *fai;
1453 	int error;
1454 
1455 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
1456 		return ENOENT;
1457 
1458 	node_ref(opc);
1459 
1460 	/*
1461 	 * If we previously detected the filesystem does not
1462 	 * implement access(), short-circuit the call and skip
1463 	 * to libpuffs access() emulation.
1464 	 */
1465 	ps = puffs_getspecific(pu);
1466 	if (ps->ps_flags & PS_NO_ACCESS) {
1467 		const struct vattr *vap;
1468 
1469 		vap = puffs_pn_getvap((struct puffs_node *)opc);
1470 
1471 		error = puffs_access(IFTOVT(vap->va_mode),
1472 				     vap->va_mode & ACCESSPERMS,
1473 				     vap->va_uid, vap->va_gid,
1474 				     (mode_t)mode, pcr);
1475 		goto out;
1476 	}
1477 
1478 	/*
1479 	 * Plain access call
1480 	 */
1481 	pm = ps->ps_new_msg(pu, opc, FUSE_ACCESS, sizeof(*fai), pcr);
1482 	fai = GET_INPAYLOAD(ps, pm, fuse_access_in);
1483 	fai->mask = 0;
1484 	fai->mask |= (mode & PUFFS_VREAD) ? R_OK : 0;
1485 	fai->mask |= (mode & PUFFS_VWRITE) ? W_OK : 0;
1486 	fai->mask |= (mode & PUFFS_VEXEC) ? X_OK : 0;
1487 
1488 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
1489 
1490 	ps->ps_destroy_msg(pm);
1491 
1492 	/*
1493 	 * If unimplemented, start over with emulation
1494 	 */
1495 	if (error == ENOSYS) {
1496 		ps->ps_flags |= PS_NO_ACCESS;
1497 		error = perfuse_node_access(pu, opc, mode, pcr);
1498 	}
1499 
1500 out:
1501 	node_rele(opc);
1502 	return error;
1503 }
1504 
1505 int
1506 perfuse_node_getattr(struct puffs_usermount *pu, puffs_cookie_t opc,
1507 	struct vattr *vap, const struct puffs_cred *pcr)
1508 {
1509 	return perfuse_node_getattr_ttl(pu, opc, vap, pcr, NULL);
1510 }
1511 
1512 int
1513 perfuse_node_getattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
1514 	struct vattr *vap, const struct puffs_cred *pcr,
1515 	struct timespec *va_ttl)
1516 {
1517 	perfuse_msg_t *pm = NULL;
1518 	struct perfuse_state *ps;
1519 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
1520 	struct fuse_getattr_in *fgi;
1521 	struct fuse_attr_out *fao;
1522 	int error = 0;
1523 
1524 	if (pnd->pnd_flags & PND_REMOVED)
1525 		return ENOENT;
1526 
1527 	node_ref(opc);
1528 
1529 	/*
1530 	 * Serialize size access, see comment in perfuse_node_setattr().
1531 	 */
1532 	while (pnd->pnd_flags & PND_INRESIZE)
1533 		requeue_request(pu, opc, PCQ_RESIZE);
1534 	pnd->pnd_flags |= PND_INRESIZE;
1535 
1536 	ps = puffs_getspecific(pu);
1537 
1538 	/*
1539 	 * FUSE_GETATTR_FH must be set in fgi->flags
1540 	 * if we use for fgi->fh
1541 	 */
1542 	pm = ps->ps_new_msg(pu, opc, FUSE_GETATTR, sizeof(*fgi), pcr);
1543 	fgi = GET_INPAYLOAD(ps, pm, fuse_getattr_in);
1544 	fgi->getattr_flags = 0;
1545 	fgi->dummy = 0;
1546 	fgi->fh = 0;
1547 
1548 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_OPEN) {
1549 		fgi->fh = perfuse_get_fh(opc, FREAD);
1550 		fgi->getattr_flags |= FUSE_GETATTR_FH;
1551 	}
1552 
1553 #ifdef PERFUSE_DEBUG
1554 	if (perfuse_diagflags & PDF_RESIZE)
1555 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__, (void *)opc,
1556 		    vap->va_size);
1557 #endif
1558 
1559 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
1560 		goto out;
1561 
1562 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1563 
1564 #ifdef PERFUSE_DEBUG
1565 	if (perfuse_diagflags & PDF_RESIZE)
1566 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1567 		    (void *)opc, vap->va_size, fao->attr.size);
1568 #endif
1569 
1570 	/*
1571 	 * We set birthtime, flags, filerev,vaflags to 0.
1572 	 * This seems the best bet, since the information is
1573 	 * not available from filesystem.
1574 	 */
1575 	fuse_attr_to_vap(ps, vap, &fao->attr);
1576 
1577 	if (va_ttl != NULL) {
1578 		va_ttl->tv_sec = fao->attr_valid;
1579 		va_ttl->tv_nsec = fao->attr_valid_nsec;
1580 	}
1581 
1582 	ps->ps_destroy_msg(pm);
1583 	error = 0;
1584 out:
1585 
1586 	pnd->pnd_flags &= ~PND_INRESIZE;
1587 	(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
1588 
1589 	node_rele(opc);
1590 	return error;
1591 }
1592 
1593 int
1594 perfuse_node_setattr(struct puffs_usermount *pu, puffs_cookie_t opc,
1595 	const struct vattr *vap, const struct puffs_cred *pcr)
1596 {
1597 	return perfuse_node_setattr_ttl(pu, opc,
1598 					__UNCONST(vap), pcr, NULL, 0);
1599 }
1600 
1601 int
1602 perfuse_node_setattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
1603 	struct vattr *vap, const struct puffs_cred *pcr,
1604 	struct timespec *va_ttl, int xflag)
1605 {
1606 	perfuse_msg_t *pm;
1607 	uint64_t fh;
1608 	struct perfuse_state *ps;
1609 	struct perfuse_node_data *pnd;
1610 	struct fuse_setattr_in *fsi;
1611 	struct fuse_attr_out *fao;
1612 	struct vattr *old_va;
1613 	enum perfuse_xchg_pb_reply reply;
1614 	int error;
1615 #ifdef PERFUSE_DEBUG
1616 	struct vattr *old_vap;
1617 	int resize_debug = 0;
1618 #endif
1619 	ps = puffs_getspecific(pu);
1620 	pnd = PERFUSE_NODE_DATA(opc);
1621 
1622 	/*
1623 	 * The only operation we can do once the file is removed
1624 	 * is to resize it, and we can do it only if it is open.
1625 	 * Do not even send the operation to the filesystem: the
1626 	 * file is not there anymore.
1627 	 */
1628 	if (pnd->pnd_flags & PND_REMOVED) {
1629 		if (!(pnd->pnd_flags & PND_OPEN))
1630 			return ENOENT;
1631 
1632 		return 0;
1633 	}
1634 
1635 	old_va = puffs_pn_getvap((struct puffs_node *)opc);
1636 
1637 	/*
1638 	 * Check for permission to change size
1639 	 * It is always allowed if we already have a write file handle
1640 	 */
1641 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1642 	    !(pnd->pnd_flags & PND_WFH) &&
1643 	    (error = mode_access(opc, pcr, PUFFS_VWRITE)) != 0)
1644 		return error;
1645 
1646 	/*
1647 	 * Check for permission to change dates
1648 	 */
1649 	if (((vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1650 	     (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL)) &&
1651 	    (puffs_access_times(old_va->va_uid, old_va->va_gid,
1652 				old_va->va_mode, 0, pcr) != 0))
1653 		return EACCES;
1654 
1655 	/*
1656 	 * Check for permission to change owner and group
1657 	 */
1658 	if (((vap->va_uid != (uid_t)PUFFS_VNOVAL) ||
1659 	     (vap->va_gid != (gid_t)PUFFS_VNOVAL)) &&
1660 	    (puffs_access_chown(old_va->va_uid, old_va->va_gid,
1661 				vap->va_uid, vap->va_gid, pcr)) != 0)
1662 		return EACCES;
1663 
1664 	/*
1665 	 * Check for permission to change permissions
1666 	 */
1667 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1668 	    (puffs_access_chmod(old_va->va_uid, old_va->va_gid,
1669 				old_va->va_type, vap->va_mode, pcr)) != 0)
1670 		return EACCES;
1671 
1672 	node_ref(opc);
1673 
1674 	if (pnd->pnd_flags & PND_WFH)
1675 		fh = perfuse_get_fh(opc, FWRITE);
1676 	else
1677 		fh = FUSE_UNKNOWN_FH;
1678 
1679 	/*
1680 	 * fchmod() sets mode and fh, and it may carry
1681 	 * a resize as well. That may break if the
1682 	 * filesystem does chmod then resize, and fails
1683 	 * because it does not have permission anymore.
1684 	 * We work this around by splitting into two setattr.
1685 	 */
1686 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1687 	    (vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
1688 	    (fh != FUSE_UNKNOWN_FH)) {
1689 		struct vattr resize_va;
1690 
1691 		(void)memcpy(&resize_va, vap, sizeof(resize_va));
1692 		resize_va.va_mode = (mode_t)PUFFS_VNOVAL;
1693 		if ((error = perfuse_node_setattr_ttl(pu, opc, &resize_va,
1694 						      pcr, va_ttl, xflag)) != 0)
1695 			goto out2;
1696 
1697 		vap->va_size = (u_quad_t)PUFFS_VNOVAL;
1698 	}
1699 
1700 	pm = ps->ps_new_msg(pu, opc, FUSE_SETATTR, sizeof(*fsi), pcr);
1701 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
1702 	fsi->valid = 0;
1703 
1704 	/*
1705 	 * Get a fh if the node is open for writing
1706 	 */
1707 	if (fh != FUSE_UNKNOWN_FH) {
1708 		fsi->fh = fh;
1709 		fsi->valid |= FUSE_FATTR_FH;
1710 	}
1711 
1712 
1713 	if (vap->va_size != (u_quad_t)PUFFS_VNOVAL) {
1714 		fsi->size = vap->va_size;
1715 		fsi->valid |= FUSE_FATTR_SIZE;
1716 
1717 		/*
1718 		 * Serialize anything that can touch file size
1719 		 * to avoid reordered GETATTR and SETATTR.
1720 		 * Out of order SETATTR can report stale size,
1721 		 * which will cause the kernel to truncate the file.
1722 		 * XXX Probably useless now we have a lock on GETATTR
1723 		 */
1724 		while (pnd->pnd_flags & PND_INRESIZE)
1725 			requeue_request(pu, opc, PCQ_RESIZE);
1726 		pnd->pnd_flags |= PND_INRESIZE;
1727 	}
1728 
1729 	/*
1730  	 * Setting mtime without atime or vice versa leads to
1731 	 * dates being reset to Epoch on glusterfs. If one
1732 	 * is missing, use the old value.
1733  	 */
1734 	if ((vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) ||
1735 	    (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL)) {
1736 
1737 		if (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) {
1738 			fsi->atime = vap->va_atime.tv_sec;
1739 			fsi->atimensec = (uint32_t)vap->va_atime.tv_nsec;
1740 		} else {
1741 			fsi->atime = old_va->va_atime.tv_sec;
1742 			fsi->atimensec = (uint32_t)old_va->va_atime.tv_nsec;
1743 		}
1744 
1745 		if (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) {
1746 			fsi->mtime = vap->va_mtime.tv_sec;
1747 			fsi->mtimensec = (uint32_t)vap->va_mtime.tv_nsec;
1748 		} else {
1749 			fsi->mtime = old_va->va_mtime.tv_sec;
1750 			fsi->mtimensec = (uint32_t)old_va->va_mtime.tv_nsec;
1751 		}
1752 
1753 		fsi->valid |= (FUSE_FATTR_MTIME|FUSE_FATTR_ATIME);
1754 	}
1755 
1756 	if (vap->va_mode != (mode_t)PUFFS_VNOVAL) {
1757 		fsi->mode = vap->va_mode;
1758 		fsi->valid |= FUSE_FATTR_MODE;
1759 	}
1760 
1761 	if (vap->va_uid != (uid_t)PUFFS_VNOVAL) {
1762 		fsi->uid = vap->va_uid;
1763 		fsi->valid |= FUSE_FATTR_UID;
1764 	}
1765 
1766 	if (vap->va_gid != (gid_t)PUFFS_VNOVAL) {
1767 		fsi->gid = vap->va_gid;
1768 		fsi->valid |= FUSE_FATTR_GID;
1769 	}
1770 
1771 	if (pnd->pnd_lock_owner != 0) {
1772 		fsi->lock_owner = pnd->pnd_lock_owner;
1773 		fsi->valid |= FUSE_FATTR_LOCKOWNER;
1774 	}
1775 
1776 	/*
1777 	 * ftruncate() sends only va_size, and metadata cache
1778 	 * flush adds va_atime and va_mtime. Some FUSE
1779 	 * filesystems will attempt to detect ftruncate by
1780 	 * checking for FATTR_SIZE being set without
1781 	 * FATTR_UID|FATTR_GID|FATTR_ATIME|FATTR_MTIME|FATTR_MODE
1782 	 *
1783 	 * Try to adapt and remove FATTR_ATIME|FATTR_MTIME
1784 	 * if we suspect a ftruncate().
1785 	 */
1786 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
1787 	    ((vap->va_mode == (mode_t)PUFFS_VNOVAL) &&
1788 	     (vap->va_uid == (uid_t)PUFFS_VNOVAL) &&
1789 	     (vap->va_gid == (gid_t)PUFFS_VNOVAL))) {
1790 		fsi->atime = 0;
1791 		fsi->atimensec = 0;
1792 		fsi->mtime = 0;
1793 		fsi->mtimensec = 0;
1794 		fsi->valid &= ~(FUSE_FATTR_ATIME|FUSE_FATTR_MTIME);
1795 	}
1796 
1797 	/*
1798 	 * If nothing remain, discard the operation.
1799 	 */
1800 	if (!(fsi->valid & (FUSE_FATTR_SIZE|FUSE_FATTR_ATIME|FUSE_FATTR_MTIME|
1801 			    FUSE_FATTR_MODE|FUSE_FATTR_UID|FUSE_FATTR_GID))) {
1802 		error = 0;
1803 		ps->ps_destroy_msg(pm);
1804 		goto out;
1805 	}
1806 
1807 #ifdef PERFUSE_DEBUG
1808 	old_vap = puffs_pn_getvap((struct puffs_node *)opc);
1809 
1810 	if ((perfuse_diagflags & PDF_RESIZE) &&
1811 	    (old_vap->va_size != (u_quad_t)PUFFS_VNOVAL)) {
1812 		resize_debug = 1;
1813 
1814 		DPRINTF(">> %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1815 		    (void *)opc,
1816 		    puffs_pn_getvap((struct puffs_node *)opc)->va_size,
1817 		    fsi->size);
1818 	}
1819 #endif
1820 
1821 	/*
1822 	 * Do not honour FAF when changing size. How do
1823 	 * you want such a thing to work?
1824 	 */
1825 	reply = wait_reply;
1826 #ifdef PUFFS_SETATTR_FAF
1827 	if ((xflag & PUFFS_SETATTR_FAF) && !(fsi->valid & FUSE_FATTR_SIZE))
1828 		reply = no_reply;
1829 #endif
1830 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), reply)) != 0)
1831 		goto out;
1832 
1833 	if (reply == no_reply)
1834 		goto out;
1835 
1836 	/*
1837 	 * Copy back the new values
1838 	 */
1839 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
1840 
1841 #ifdef PERFUSE_DEBUG
1842 	if (resize_debug)
1843 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
1844 		    (void *)opc, old_vap->va_size, fao->attr.size);
1845 #endif
1846 
1847 	fuse_attr_to_vap(ps, old_va, &fao->attr);
1848 
1849 	if (va_ttl != NULL) {
1850 		va_ttl->tv_sec = fao->attr_valid;
1851 		va_ttl->tv_nsec = fao->attr_valid_nsec;
1852 		(void)memcpy(vap, old_va, sizeof(*vap));
1853 	}
1854 
1855 	ps->ps_destroy_msg(pm);
1856 	error = 0;
1857 
1858 out:
1859 	if (pnd->pnd_flags & PND_INRESIZE) {
1860 		pnd->pnd_flags &= ~PND_INRESIZE;
1861 		(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
1862 	}
1863 
1864 out2:
1865 	node_rele(opc);
1866 	return error;
1867 }
1868 
1869 int
1870 perfuse_node_poll(struct puffs_usermount *pu, puffs_cookie_t opc, int *events)
1871 {
1872 	struct perfuse_state *ps;
1873 	perfuse_msg_t *pm;
1874 	struct fuse_poll_in *fpi;
1875 	struct fuse_poll_out *fpo;
1876 	int error;
1877 
1878 	node_ref(opc);
1879 	ps = puffs_getspecific(pu);
1880 	/*
1881 	 * kh is set if FUSE_POLL_SCHEDULE_NOTIFY is set.
1882 	 *
1883 	 * XXX ps_new_msg() is called with NULL creds, which will
1884 	 * be interpreted as FUSE superuser. We have no way to
1885 	 * know the requesting process' credential, but since poll
1886 	 * is supposed to operate on a file that has been open,
1887 	 * permission should have already been checked at open time.
1888 	 * That still may breaks on filesystems that provides odd
1889 	 * semantics.
1890  	 */
1891 	pm = ps->ps_new_msg(pu, opc, FUSE_POLL, sizeof(*fpi), NULL);
1892 	fpi = GET_INPAYLOAD(ps, pm, fuse_poll_in);
1893 	fpi->fh = perfuse_get_fh(opc, FREAD);
1894 	fpi->kh = 0;
1895 	fpi->flags = 0;
1896 
1897 #ifdef PERFUSE_DEBUG
1898 	if (perfuse_diagflags & PDF_FH)
1899 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
1900 			"fh = 0x%"PRIx64"\n", __func__, (void *)opc,
1901 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fpi->fh);
1902 #endif
1903 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fpo), wait_reply)) != 0)
1904 		goto out;
1905 
1906 	fpo = GET_OUTPAYLOAD(ps, pm, fuse_poll_out);
1907 	*events = fpo->revents;
1908 
1909 	ps->ps_destroy_msg(pm);
1910 	error = 0;
1911 
1912 out:
1913 	node_rele(opc);
1914 	return error;
1915 }
1916 
1917 /* ARGSUSED0 */
1918 int
1919 perfuse_node_mmap(struct puffs_usermount *pu, puffs_cookie_t opc, int flags,
1920 	const struct puffs_cred *pcr)
1921 {
1922 	/*
1923 	 * Not implemented anymore in libfuse
1924 	 */
1925 	return ENOSYS;
1926 }
1927 
1928 /* ARGSUSED2 */
1929 int
1930 perfuse_node_fsync(struct puffs_usermount *pu, puffs_cookie_t opc,
1931 	const struct puffs_cred *pcr, int flags, off_t offlo, off_t offhi)
1932 {
1933 	int op;
1934 	perfuse_msg_t *pm;
1935 	struct perfuse_state *ps;
1936 	struct perfuse_node_data *pnd;
1937 	struct fuse_fsync_in *ffi;
1938 	uint64_t fh;
1939 	int error = 0;
1940 
1941 	pm = NULL;
1942 	ps = puffs_getspecific(pu);
1943 	pnd = PERFUSE_NODE_DATA(opc);
1944 
1945 	/*
1946 	 * No need to sync a removed node
1947 	 */
1948 	if (pnd->pnd_flags & PND_REMOVED)
1949 		return 0;
1950 
1951 	/*
1952 	 * We do not sync closed files. They have been
1953 	 * sync at inactive time already.
1954 	 */
1955 	if (!(pnd->pnd_flags & PND_OPEN))
1956 		return 0;
1957 
1958 	node_ref(opc);
1959 
1960 	if (puffs_pn_getvap((struct puffs_node *)opc)->va_type == VDIR)
1961 		op = FUSE_FSYNCDIR;
1962 	else 		/* VREG but also other types such as VLNK */
1963 		op = FUSE_FSYNC;
1964 
1965 	/*
1966 	 * Do not sync if there are no change to sync
1967 	 * XXX remove that test on files if we implement mmap
1968 	 */
1969 #ifdef PERFUSE_DEBUG
1970 	if (perfuse_diagflags & PDF_SYNC)
1971 		DPRINTF("%s: TEST opc = %p, file = \"%s\" is %sdirty\n",
1972 			__func__, (void*)opc, perfuse_node_path(ps, opc),
1973 			pnd->pnd_flags & PND_DIRTY ? "" : "not ");
1974 #endif
1975 	if (!(pnd->pnd_flags & PND_DIRTY))
1976 		goto out;
1977 
1978 	/*
1979 	 * It seems NetBSD can call fsync without open first
1980 	 * glusterfs complain in such a situation:
1981 	 * "FSYNC() ERR => -1 (Invalid argument)"
1982 	 * The file will be closed at inactive time.
1983 	 *
1984 	 * We open the directory for reading in order to sync.
1985 	 * This sounds rather counterintuitive, but it works.
1986 	 */
1987 	if (!(pnd->pnd_flags & PND_WFH)) {
1988 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
1989 			goto out;
1990 	}
1991 
1992 	if (op == FUSE_FSYNCDIR)
1993 		fh = perfuse_get_fh(opc, FREAD);
1994 	else
1995 		fh = perfuse_get_fh(opc, FWRITE);
1996 
1997 	/*
1998 	 * If fsync_flags  is set, meta data should not be flushed.
1999 	 */
2000 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*ffi), pcr);
2001 	ffi = GET_INPAYLOAD(ps, pm, fuse_fsync_in);
2002 	ffi->fh = fh;
2003 	ffi->fsync_flags = (flags & FFILESYNC) ? 0 : 1;
2004 
2005 #ifdef PERFUSE_DEBUG
2006 	if (perfuse_diagflags & PDF_FH)
2007 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
2008 			__func__, (void *)opc,
2009 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, ffi->fh);
2010 #endif
2011 
2012 	if ((error = xchg_msg(pu, opc, pm,
2013 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
2014 		goto out;
2015 
2016 	/*
2017 	 * No reply beyond fuse_out_header: nothing to do on success
2018 	 * just clear the dirty flag
2019 	 */
2020 	pnd->pnd_flags &= ~PND_DIRTY;
2021 
2022 #ifdef PERFUSE_DEBUG
2023 	if (perfuse_diagflags & PDF_SYNC)
2024 		DPRINTF("%s: CLEAR opc = %p, file = \"%s\"\n",
2025 			__func__, (void*)opc, perfuse_node_path(ps, opc));
2026 #endif
2027 
2028 	ps->ps_destroy_msg(pm);
2029 	error = 0;
2030 
2031 out:
2032 	/*
2033 	 * ENOSYS is not returned to kernel,
2034 	 */
2035 	if (error == ENOSYS)
2036 		error = 0;
2037 
2038 	node_rele(opc);
2039 	return error;
2040 }
2041 
2042 /* ARGSUSED0 */
2043 int
2044 perfuse_node_seek(struct puffs_usermount *pu, puffs_cookie_t opc,
2045 	off_t oldoff, off_t newoff, const struct puffs_cred *pcr)
2046 {
2047 	return 0;
2048 }
2049 
2050 int
2051 perfuse_node_remove(struct puffs_usermount *pu, puffs_cookie_t opc,
2052 	puffs_cookie_t targ, const struct puffs_cn *pcn)
2053 {
2054 	struct perfuse_state *ps;
2055 	struct perfuse_node_data *pnd;
2056 	perfuse_msg_t *pm;
2057 	char *path;
2058 	const char *name;
2059 	size_t len;
2060 	int error;
2061 
2062 	pnd = PERFUSE_NODE_DATA(opc);
2063 
2064 	if ((pnd->pnd_flags & PND_REMOVED) ||
2065 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
2066 		return ENOENT;
2067 
2068 #ifdef PERFUSE_DEBUG
2069 	if (targ == NULL)
2070 		DERRX(EX_SOFTWARE, "%s: targ is NULL", __func__);
2071 
2072 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
2073 		DPRINTF("%s: opc = %p, remove opc = %p, file = \"%s\"\n",
2074 			__func__, (void *)opc, (void *)targ, pcn->pcn_name);
2075 #endif
2076 	node_ref(opc);
2077 	node_ref(targ);
2078 
2079 	/*
2080 	 * Await for all operations on the deleted node to drain,
2081 	 * as the filesystem may be confused to have it deleted
2082 	 * during a getattr
2083 	 */
2084 	while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
2085 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2086 
2087 	ps = puffs_getspecific(pu);
2088 	pnd = PERFUSE_NODE_DATA(opc);
2089 	name = pcn->pcn_name;
2090 	len = pcn->pcn_namelen + 1;
2091 
2092 	pm = ps->ps_new_msg(pu, opc, FUSE_UNLINK, len, pcn->pcn_cred);
2093 	path = _GET_INPAYLOAD(ps, pm, char *);
2094 	(void)strlcpy(path, name, len);
2095 
2096 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2097 		goto out;
2098 
2099 	perfuse_cache_flush(targ);
2100 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2101 
2102 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2103 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2104 
2105 	/*
2106 	 * The parent directory needs a sync
2107 	 */
2108 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2109 
2110 #ifdef PERFUSE_DEBUG
2111 	if (perfuse_diagflags & PDF_FILENAME)
2112 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
2113 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
2114 			pcn->pcn_name);
2115 #endif
2116 	ps->ps_destroy_msg(pm);
2117 	error = 0;
2118 
2119 out:
2120 	node_rele(opc);
2121 	node_rele(targ);
2122 	return error;
2123 }
2124 
2125 int
2126 perfuse_node_link(struct puffs_usermount *pu, puffs_cookie_t opc,
2127 	puffs_cookie_t targ, const struct puffs_cn *pcn)
2128 {
2129 	struct perfuse_state *ps;
2130 	perfuse_msg_t *pm;
2131 	const char *name;
2132 	size_t len;
2133 	struct puffs_node *pn;
2134 	struct fuse_link_in *fli;
2135 	int error;
2136 
2137 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2138 		return ENOENT;
2139 
2140 	node_ref(opc);
2141 	node_ref(targ);
2142 	ps = puffs_getspecific(pu);
2143 	pn = (struct puffs_node *)targ;
2144 	name = pcn->pcn_name;
2145 	len =  sizeof(*fli) + pcn->pcn_namelen + 1;
2146 
2147 	pm = ps->ps_new_msg(pu, opc, FUSE_LINK, len, pcn->pcn_cred);
2148 	fli = GET_INPAYLOAD(ps, pm, fuse_link_in);
2149 	fli->oldnodeid = PERFUSE_NODE_DATA(pn)->pnd_nodeid;
2150 	(void)strlcpy((char *)(void *)(fli + 1), name, len - sizeof(*fli));
2151 
2152 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2153 		goto out;
2154 
2155 	ps->ps_destroy_msg(pm);
2156 	error = 0;
2157 
2158 out:
2159 	node_rele(opc);
2160 	node_rele(targ);
2161 	return error;
2162 }
2163 
2164 int
2165 perfuse_node_rename(struct puffs_usermount *pu, puffs_cookie_t opc,
2166 	puffs_cookie_t src, const struct puffs_cn *pcn_src,
2167 	puffs_cookie_t targ_dir, puffs_cookie_t targ,
2168 	const struct puffs_cn *pcn_targ)
2169 {
2170 	struct perfuse_state *ps;
2171 	struct perfuse_node_data *dstdir_pnd;
2172 	perfuse_msg_t *pm;
2173 	struct fuse_rename_in *fri;
2174 	const char *newname;
2175 	const char *oldname;
2176 	char *np;
2177 	int error;
2178 	size_t len;
2179 	size_t newname_len;
2180 	size_t oldname_len;
2181 
2182 	if ((PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED) ||
2183 	    (PERFUSE_NODE_DATA(src)->pnd_flags & PND_REMOVED) ||
2184 	    (PERFUSE_NODE_DATA(targ_dir)->pnd_flags & PND_REMOVED))
2185 		return ENOENT;
2186 
2187 	node_ref(opc);
2188 	node_ref(src);
2189 
2190 	/*
2191 	 * Await for all operations on the deleted node to drain,
2192 	 * as the filesystem may be confused to have it deleted
2193 	 * during a getattr
2194 	 */
2195 	if ((struct puffs_node *)targ != NULL) {
2196 		node_ref(targ);
2197 		while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
2198 			requeue_request(pu, targ, PCQ_AFTERXCHG);
2199 	} else {
2200 		while (PERFUSE_NODE_DATA(src)->pnd_inxchg)
2201 			requeue_request(pu, src, PCQ_AFTERXCHG);
2202 	}
2203 
2204 	ps = puffs_getspecific(pu);
2205 	newname =  pcn_targ->pcn_name;
2206 	newname_len = pcn_targ->pcn_namelen + 1;
2207 	oldname =  pcn_src->pcn_name;
2208 	oldname_len = pcn_src->pcn_namelen + 1;
2209 
2210 	len = sizeof(*fri) + oldname_len + newname_len;
2211 	pm = ps->ps_new_msg(pu, opc, FUSE_RENAME, len, pcn_targ->pcn_cred);
2212 	fri = GET_INPAYLOAD(ps, pm, fuse_rename_in);
2213 	fri->newdir = PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid;
2214 	np = (char *)(void *)(fri + 1);
2215 	(void)strlcpy(np, oldname, oldname_len);
2216 	np += oldname_len;
2217 	(void)strlcpy(np, newname, newname_len);
2218 
2219 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2220 		goto out;
2221 
2222 
2223 	/*
2224 	 * Record new parent nodeid
2225 	 */
2226 	dstdir_pnd = PERFUSE_NODE_DATA(targ_dir);
2227 	PERFUSE_NODE_DATA(src)->pnd_parent_nodeid = dstdir_pnd->pnd_nodeid;
2228 
2229 	if (opc != targ_dir)
2230 		dstdir_pnd->pnd_flags |= PND_DIRTY;
2231 
2232 	if (strcmp(newname, "..") != 0)
2233 		(void)strlcpy(PERFUSE_NODE_DATA(src)->pnd_name,
2234 		    newname, MAXPATHLEN);
2235 	else
2236 		PERFUSE_NODE_DATA(src)->pnd_name[0] = 0; /* forget name */
2237 
2238 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2239 
2240 	if ((struct puffs_node *)targ != NULL) {
2241 		perfuse_cache_flush(targ);
2242 		PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2243 	}
2244 
2245 #ifdef PERFUSE_DEBUG
2246 	if (perfuse_diagflags & PDF_FILENAME)
2247 		DPRINTF("%s: nodeid = 0x%"PRIx64" file = \"%s\" renamed \"%s\" "
2248 			"nodeid = 0x%"PRIx64" -> nodeid = 0x%"PRIx64" \"%s\"\n",
2249 	 		__func__, PERFUSE_NODE_DATA(src)->pnd_nodeid,
2250 			pcn_src->pcn_name, pcn_targ->pcn_name,
2251 			PERFUSE_NODE_DATA(opc)->pnd_nodeid,
2252 			PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid,
2253 			perfuse_node_path(ps, targ_dir));
2254 #endif
2255 
2256 	ps->ps_destroy_msg(pm);
2257 	error = 0;
2258 
2259 out:
2260 	node_rele(opc);
2261 	node_rele(src);
2262 	if ((struct puffs_node *)targ != NULL)
2263 		node_rele(targ);
2264 
2265 	return error;
2266 }
2267 
2268 int
2269 perfuse_node_mkdir(struct puffs_usermount *pu, puffs_cookie_t opc,
2270 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
2271 	const struct vattr *vap)
2272 {
2273 	struct perfuse_state *ps;
2274 	perfuse_msg_t *pm;
2275 	struct fuse_mkdir_in *fmi;
2276 	const char *path;
2277 	size_t len;
2278 	int error;
2279 
2280 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2281 		return ENOENT;
2282 
2283 	node_ref(opc);
2284 	ps = puffs_getspecific(pu);
2285 	path = pcn->pcn_name;
2286 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
2287 
2288 	pm = ps->ps_new_msg(pu, opc, FUSE_MKDIR, len, pcn->pcn_cred);
2289 	fmi = GET_INPAYLOAD(ps, pm, fuse_mkdir_in);
2290 	fmi->mode = vap->va_mode;
2291 	fmi->umask = 0; 	/* Seems unused by libfuse? */
2292 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
2293 
2294 	error = node_mk_common(pu, opc, pni, pcn, pm);
2295 
2296 	node_rele(opc);
2297 	return error;
2298 }
2299 
2300 
2301 int
2302 perfuse_node_rmdir(struct puffs_usermount *pu, puffs_cookie_t opc,
2303 	puffs_cookie_t targ, const struct puffs_cn *pcn)
2304 {
2305 	struct perfuse_state *ps;
2306 	struct perfuse_node_data *pnd;
2307 	perfuse_msg_t *pm;
2308 	char *path;
2309 	const char *name;
2310 	size_t len;
2311 	int error;
2312 
2313 	pnd = PERFUSE_NODE_DATA(opc);
2314 
2315 	if ((pnd->pnd_flags & PND_REMOVED) ||
2316 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
2317 		return ENOENT;
2318 
2319 	node_ref(opc);
2320 	node_ref(targ);
2321 
2322 	/*
2323 	 * Await for all operations on the deleted node to drain,
2324 	 * as the filesystem may be confused to have it deleted
2325 	 * during a getattr
2326 	 */
2327 	while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
2328 		requeue_request(pu, targ, PCQ_AFTERXCHG);
2329 
2330 	ps = puffs_getspecific(pu);
2331 	name = pcn->pcn_name;
2332 	len = pcn->pcn_namelen + 1;
2333 
2334 	pm = ps->ps_new_msg(pu, opc, FUSE_RMDIR, len, pcn->pcn_cred);
2335 	path = _GET_INPAYLOAD(ps, pm, char *);
2336 	(void)strlcpy(path, name, len);
2337 
2338 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2339 		goto out;
2340 
2341 	perfuse_cache_flush(targ);
2342 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
2343 
2344 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
2345 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
2346 
2347 	/*
2348 	 * The parent directory needs a sync
2349 	 */
2350 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
2351 
2352 #ifdef PERFUSE_DEBUG
2353 	if (perfuse_diagflags & PDF_FILENAME)
2354 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
2355 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
2356 			perfuse_node_path(ps, targ));
2357 #endif
2358 	ps->ps_destroy_msg(pm);
2359 	error = 0;
2360 
2361 out:
2362 	node_rele(opc);
2363 	node_rele(targ);
2364 	return error;
2365 }
2366 
2367 /* vap is unused */
2368 /* ARGSUSED4 */
2369 int
2370 perfuse_node_symlink(struct puffs_usermount *pu, puffs_cookie_t opc,
2371 	struct puffs_newinfo *pni, const struct puffs_cn *pcn_src,
2372 	const struct vattr *vap, const char *link_target)
2373 {
2374 	struct perfuse_state *ps;
2375 	perfuse_msg_t *pm;
2376 	char *np;
2377 	const char *path;
2378 	size_t path_len;
2379 	size_t linkname_len;
2380 	size_t len;
2381 	int error;
2382 
2383 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2384 		return ENOENT;
2385 
2386 	node_ref(opc);
2387 	ps = puffs_getspecific(pu);
2388 	path = pcn_src->pcn_name;
2389 	path_len = pcn_src->pcn_namelen + 1;
2390 	linkname_len = strlen(link_target) + 1;
2391 	len = path_len + linkname_len;
2392 
2393 	pm = ps->ps_new_msg(pu, opc, FUSE_SYMLINK, len, pcn_src->pcn_cred);
2394 	np = _GET_INPAYLOAD(ps, pm, char *);
2395 	(void)strlcpy(np, path, path_len);
2396 	np += path_len;
2397 	(void)strlcpy(np, link_target, linkname_len);
2398 
2399 	error = node_mk_common(pu, opc, pni, pcn_src, pm);
2400 
2401 	node_rele(opc);
2402 	return error;
2403 }
2404 
2405 /* ARGSUSED4 */
2406 int
2407 perfuse_node_readdir(struct puffs_usermount *pu, puffs_cookie_t opc,
2408 	struct dirent *dent, off_t *readoff, size_t *reslen,
2409 	const struct puffs_cred *pcr, int *eofflag, off_t *cookies,
2410 	size_t *ncookies)
2411 {
2412 	perfuse_msg_t *pm;
2413 	uint64_t fh;
2414 	struct perfuse_state *ps;
2415 	struct perfuse_node_data *pnd;
2416 	struct fuse_read_in *fri;
2417 	struct fuse_out_header *foh;
2418 	struct fuse_dirent *fd;
2419 	size_t foh_len;
2420 	int error;
2421 	size_t fd_maxlen;
2422 
2423 	error = 0;
2424 	node_ref(opc);
2425 	ps = puffs_getspecific(pu);
2426 
2427 	/*
2428 	 * readdir state is kept at node level, and several readdir
2429 	 * requests can be issued at the same time on the same node.
2430 	 * We need to queue requests so that only one is in readdir
2431 	 * code at the same time.
2432 	 */
2433 	pnd = PERFUSE_NODE_DATA(opc);
2434 	while (pnd->pnd_flags & PND_INREADDIR)
2435 		requeue_request(pu, opc, PCQ_READDIR);
2436 	pnd->pnd_flags |= PND_INREADDIR;
2437 
2438 #ifdef PERFUSE_DEBUG
2439 	if (perfuse_diagflags & PDF_READDIR)
2440 		DPRINTF("%s: READDIR opc = %p enter critical section\n",
2441 			__func__, (void *)opc);
2442 #endif
2443 	/*
2444 	 * Re-initialize pnd->pnd_fd_cookie on the first readdir for a node
2445 	 */
2446 	if (*readoff == 0)
2447 		pnd->pnd_fd_cookie = 0;
2448 
2449 	/*
2450 	 * Do we already have the data bufered?
2451 	 */
2452 	if (pnd->pnd_dirent != NULL)
2453 		goto out;
2454 	pnd->pnd_dirent_len = 0;
2455 
2456 	/*
2457 	 * It seems NetBSD can call readdir without open first
2458 	 * libfuse will crash if it is done that way, hence open first.
2459 	 */
2460 	if (!(pnd->pnd_flags & PND_OPEN)) {
2461 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
2462 			goto out;
2463 	}
2464 
2465 	fh = perfuse_get_fh(opc, FREAD);
2466 
2467 #ifdef PERFUSE_DEBUG
2468 	if (perfuse_diagflags & PDF_FH)
2469 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
2470 			"rfh = 0x%"PRIx64"\n", __func__, (void *)opc,
2471 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fh);
2472 #endif
2473 
2474 	pnd->pnd_all_fd = NULL;
2475 	pnd->pnd_all_fd_len = 0;
2476 	fd_maxlen = ps->ps_max_readahead - sizeof(*foh);
2477 
2478 	do {
2479 		size_t fd_len;
2480 		char *afdp;
2481 
2482 		pm = ps->ps_new_msg(pu, opc, FUSE_READDIR, sizeof(*fri), pcr);
2483 
2484 		/*
2485 		 * read_flags, lock_owner and flags are unused in libfuse
2486 		 */
2487 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
2488 		fri->fh = fh;
2489 		fri->offset = pnd->pnd_fd_cookie;
2490 		fri->size = (uint32_t)fd_maxlen;
2491 		fri->read_flags = 0;
2492 		fri->lock_owner = 0;
2493 		fri->flags = 0;
2494 
2495 		if ((error = xchg_msg(pu, opc, pm,
2496 				      UNSPEC_REPLY_LEN, wait_reply)) != 0)
2497 			goto out;
2498 
2499 		/*
2500 		 * There are many puffs_framebufs calls later,
2501 		 * therefore foh will not be valid for a long time.
2502 		 * Just get the length and forget it.
2503 		 */
2504 		foh = GET_OUTHDR(ps, pm);
2505 		foh_len = foh->len;
2506 
2507 		/*
2508 		 * Empty read: we reached the end of the buffer.
2509 		 */
2510 		if (foh_len == sizeof(*foh)) {
2511 			ps->ps_destroy_msg(pm);
2512 			*eofflag = 1;
2513 			break;
2514 		}
2515 
2516 		/*
2517 		 * Check for corrupted message.
2518 		 */
2519 		if (foh_len < sizeof(*foh) + sizeof(*fd)) {
2520 			ps->ps_destroy_msg(pm);
2521 			DWARNX("readdir reply too short");
2522 			error = EIO;
2523 			goto out;
2524 		}
2525 
2526 
2527 		fd = GET_OUTPAYLOAD(ps, pm, fuse_dirent);
2528 		fd_len = foh_len - sizeof(*foh);
2529 
2530 		pnd->pnd_all_fd = realloc(pnd->pnd_all_fd,
2531 					  pnd->pnd_all_fd_len + fd_len);
2532 		if (pnd->pnd_all_fd  == NULL)
2533 			DERR(EX_OSERR, "%s: malloc failed", __func__);
2534 
2535 		afdp = (char *)(void *)pnd->pnd_all_fd + pnd->pnd_all_fd_len;
2536 		(void)memcpy(afdp, fd, fd_len);
2537 
2538 		pnd->pnd_all_fd_len += fd_len;
2539 
2540 		/*
2541 		 * The fd->off field is used as a cookie for
2542 		 * resuming the next readdir() where this one was left.
2543 	 	 */
2544 		pnd->pnd_fd_cookie = readdir_last_cookie(fd, fd_len);
2545 
2546 		ps->ps_destroy_msg(pm);
2547 	} while (1 /* CONSTCOND */);
2548 
2549 	if (pnd->pnd_all_fd != NULL) {
2550 		if (fuse_to_dirent(pu, opc, pnd->pnd_all_fd,
2551 				   pnd->pnd_all_fd_len) == -1)
2552 			error = EIO;
2553 	}
2554 
2555 out:
2556 	if (pnd->pnd_all_fd != NULL) {
2557 		free(pnd->pnd_all_fd);
2558 		pnd->pnd_all_fd = NULL;
2559 		pnd->pnd_all_fd_len = 0;
2560 	}
2561 
2562 	if (error == 0)
2563 		readdir_buffered(opc, dent, readoff, reslen);
2564 
2565 	/*
2566 	 * Schedule queued readdir requests
2567 	 */
2568 	pnd->pnd_flags &= ~PND_INREADDIR;
2569 	(void)dequeue_requests(opc, PCQ_READDIR, DEQUEUE_ALL);
2570 
2571 #ifdef PERFUSE_DEBUG
2572 	if (perfuse_diagflags & PDF_READDIR)
2573 		DPRINTF("%s: READDIR opc = %p exit critical section\n",
2574 			__func__, (void *)opc);
2575 #endif
2576 
2577 	node_rele(opc);
2578 	return error;
2579 }
2580 
2581 int
2582 perfuse_node_readlink(struct puffs_usermount *pu, puffs_cookie_t opc,
2583 	const struct puffs_cred *pcr, char *linkname, size_t *linklen)
2584 {
2585 	struct perfuse_state *ps;
2586 	perfuse_msg_t *pm;
2587 	int error;
2588 	size_t len;
2589 	struct fuse_out_header *foh;
2590 
2591 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
2592 		return ENOENT;
2593 
2594 	node_ref(opc);
2595 	ps = puffs_getspecific(pu);
2596 
2597 	pm = ps->ps_new_msg(pu, opc, FUSE_READLINK, 0, pcr);
2598 
2599 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2600 		goto out;
2601 
2602 	foh = GET_OUTHDR(ps, pm);
2603 	len = foh->len - sizeof(*foh);
2604 	if (len > *linklen)
2605 		DERRX(EX_PROTOCOL, "path len = %zd too long", len);
2606 	if (len == 0)
2607 		DERRX(EX_PROTOCOL, "path len = %zd too short", len);
2608 
2609 	/*
2610 	 * FUSE filesystems return a NUL terminated string, we
2611 	 * do not want to trailing \0
2612 	 */
2613 	*linklen = len - 1;
2614 	(void)memcpy(linkname, _GET_OUTPAYLOAD(ps, pm, char *), len);
2615 
2616 	ps->ps_destroy_msg(pm);
2617 	error = 0;
2618 
2619 out:
2620 	node_rele(opc);
2621 	return error;
2622 }
2623 
2624 int
2625 perfuse_node_reclaim(struct puffs_usermount *pu, puffs_cookie_t opc)
2626 {
2627 	struct perfuse_state *ps;
2628 	perfuse_msg_t *pm;
2629 	struct perfuse_node_data *pnd;
2630 	struct fuse_forget_in *ffi;
2631 	int nlookup;
2632 	struct timespec now;
2633 
2634 	if (opc == 0)
2635 		return 0;
2636 
2637 	ps = puffs_getspecific(pu);
2638 	pnd = PERFUSE_NODE_DATA(opc);
2639 
2640 	/*
2641 	 * Never forget the root.
2642 	 */
2643 	if (pnd->pnd_nodeid == FUSE_ROOT_ID)
2644 		return 0;
2645 
2646 	/*
2647 	 * There is a race condition between reclaim and lookup.
2648 	 * When looking up an already known node, the kernel cannot
2649 	 * hold a reference on the result until it gets the PUFFS
2650 	 * reply. It mayy therefore reclaim the node after the
2651 	 * userland looked it up, and before it gets the reply.
2652 	 * On rely, the kernel re-creates the node, but at that
2653 	 * time the node has been reclaimed in userland.
2654 	 *
2655 	 * In order to avoid this, we refuse reclaiming nodes that
2656 	 * are too young since the last lookup - and that we do
2657 	 * not have removed on our own, of course.
2658 	 */
2659 	if (clock_gettime(CLOCK_REALTIME, &now) != 0)
2660 		DERR(EX_OSERR, "clock_gettime failed");
2661 
2662 	if (timespeccmp(&pnd->pnd_cn_expire, &now, >) &&
2663 	    !(pnd->pnd_flags & PND_REMOVED)) {
2664 		if (!(pnd->pnd_flags & PND_NODELEAK)) {
2665 			ps->ps_nodeleakcount++;
2666 			pnd->pnd_flags |= PND_NODELEAK;
2667 		}
2668 		DWARNX("possible leaked node:: opc = %p \"%s\"",
2669 		       opc, pnd->pnd_name);
2670 		return 0;
2671 	}
2672 
2673 	node_ref(opc);
2674 	pnd->pnd_flags |= PND_RECLAIMED;
2675 	pnd->pnd_puffs_nlookup--;
2676 	nlookup = pnd->pnd_puffs_nlookup;
2677 
2678 #ifdef PERFUSE_DEBUG
2679 	if (perfuse_diagflags & PDF_RECLAIM)
2680 		DPRINTF("%s (nodeid %"PRId64") reclaimed\n",
2681 			perfuse_node_path(ps, opc), pnd->pnd_nodeid);
2682 #endif
2683 
2684 #ifdef PERFUSE_DEBUG
2685 	if (perfuse_diagflags & PDF_RECLAIM)
2686 		DPRINTF("%s (nodeid %"PRId64") is %sreclaimed, nlookup = %d "
2687 			"%s%s%s%s, pending ops:%s%s%s\n",
2688 		        perfuse_node_path(ps, opc), pnd->pnd_nodeid,
2689 		        pnd->pnd_flags & PND_RECLAIMED ? "" : "not ",
2690 			pnd->pnd_puffs_nlookup,
2691 			pnd->pnd_flags & PND_OPEN ? "open " : "not open",
2692 			pnd->pnd_flags & PND_RFH ? "r" : "",
2693 			pnd->pnd_flags & PND_WFH ? "w" : "",
2694 			pnd->pnd_flags & PND_BUSY ? "" : " none",
2695 			pnd->pnd_flags & PND_INREADDIR ? " readdir" : "",
2696 			pnd->pnd_flags & PND_INWRITE ? " write" : "",
2697 			pnd->pnd_flags & PND_INOPEN ? " open" : "");
2698 #endif
2699 	/*
2700 	 * Make sure it is not looked up again
2701 	 */
2702 	if (!(pnd->pnd_flags & PND_REMOVED))
2703 		perfuse_cache_flush(opc);
2704 
2705 	/*
2706 	 * Purge any activity on the node, while checking
2707 	 * that it remains eligible for a reclaim.
2708 	 */
2709 	while (pnd->pnd_ref > 1)
2710 		requeue_request(pu, opc, PCQ_REF);
2711 
2712 	/*
2713 	 * reclaim cancel?
2714 	 */
2715 	if (pnd->pnd_puffs_nlookup > nlookup) {
2716 		pnd->pnd_flags &= ~PND_RECLAIMED;
2717 		perfuse_node_cache(ps, opc);
2718 		node_rele(opc);
2719 		return 0;
2720 	}
2721 
2722 
2723 #ifdef PERFUSE_DEBUG
2724 	if ((pnd->pnd_flags & PND_OPEN) ||
2725 	       !TAILQ_EMPTY(&pnd->pnd_pcq))
2726 		DERRX(EX_SOFTWARE, "%s: opc = %p: still open",
2727 		      __func__, opc);
2728 
2729 	if ((pnd->pnd_flags & PND_BUSY) ||
2730 	       !TAILQ_EMPTY(&pnd->pnd_pcq))
2731 		DERRX(EX_SOFTWARE, "%s: opc = %p: queued operations",
2732 		      __func__, opc);
2733 
2734 	if (pnd->pnd_inxchg != 0)
2735 		DERRX(EX_SOFTWARE, "%s: opc = %p: ongoing operations",
2736 		      __func__, opc);
2737 #endif
2738 
2739 	/*
2740 	 * Send the FORGET message
2741 	 *
2742 	 * ps_new_msg() is called with NULL creds, which will
2743 	 * be interpreted as FUSE superuser. This is obviously
2744 	 * fine since we operate with kernel creds here.
2745 	 */
2746 	pm = ps->ps_new_msg(pu, opc, FUSE_FORGET,
2747 		      sizeof(*ffi), NULL);
2748 	ffi = GET_INPAYLOAD(ps, pm, fuse_forget_in);
2749 	ffi->nlookup = pnd->pnd_fuse_nlookup;
2750 
2751 	/*
2752 	 * No reply is expected, pm is freed in xchg_msg
2753 	 */
2754 	(void)xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, no_reply);
2755 
2756 	perfuse_destroy_pn(pu, opc);
2757 
2758 	return 0;
2759 }
2760 
2761 int
2762 perfuse_node_inactive(struct puffs_usermount *pu, puffs_cookie_t opc)
2763 {
2764 	struct perfuse_node_data *pnd;
2765 	int error;
2766 
2767 	if (opc == 0)
2768 		return 0;
2769 
2770 	node_ref(opc);
2771 	pnd = PERFUSE_NODE_DATA(opc);
2772 
2773 	if (!(pnd->pnd_flags & (PND_OPEN|PND_REMOVED)))
2774 		goto out;
2775 
2776 	/*
2777 	 * Make sure all operation are finished
2778 	 * There can be an ongoing write. Other
2779 	 * operation wait for all data before
2780 	 * the close/inactive.
2781 	 */
2782 	while (pnd->pnd_flags & PND_INWRITE)
2783 		requeue_request(pu, opc, PCQ_AFTERWRITE);
2784 
2785 	/*
2786 	 * The inactive operation may be cancelled,
2787 	 * If no open is in progress, set PND_INOPEN
2788 	 * so that a new open will be queued.
2789 	 */
2790 	if (pnd->pnd_flags & PND_INOPEN)
2791 		goto out;
2792 
2793 	pnd->pnd_flags |= PND_INOPEN;
2794 
2795 	/*
2796 	 * Sync data
2797 	 */
2798 	if (pnd->pnd_flags & PND_DIRTY) {
2799 		if ((error = perfuse_node_fsync(pu, opc, NULL, 0, 0, 0)) != 0)
2800 			DWARN("%s: perfuse_node_fsync failed error = %d",
2801 			      __func__, error);
2802 	}
2803 
2804 
2805 	/*
2806 	 * Close handles
2807 	 */
2808 	if (pnd->pnd_flags & PND_WFH) {
2809 		if ((error = perfuse_node_close_common(pu, opc, FWRITE)) != 0)
2810 			DWARN("%s: close write FH failed error = %d",
2811 			      __func__, error);
2812 	}
2813 
2814 	if (pnd->pnd_flags & PND_RFH) {
2815 		if ((error = perfuse_node_close_common(pu, opc, FREAD)) != 0)
2816 			DWARN("%s: close read FH failed error = %d",
2817 			      __func__, error);
2818 	}
2819 
2820 	/*
2821 	 * This will cause a reclaim to be sent
2822 	 */
2823 	if (pnd->pnd_flags & PND_REMOVED)
2824 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N1);
2825 
2826 	/*
2827 	 * Schedule awaiting operations
2828 	 */
2829 	pnd->pnd_flags &= ~PND_INOPEN;
2830 	(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
2831 
2832 	/*
2833 	 * errors are ignored, since the kernel ignores the return code.
2834 	 */
2835 out:
2836 	node_rele(opc);
2837 	return 0;
2838 }
2839 
2840 
2841 /* ARGSUSED0 */
2842 int
2843 perfuse_node_print(struct puffs_usermount *pu, puffs_cookie_t opc)
2844 {
2845 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2846 	return 0;
2847 }
2848 
2849 /* ARGSUSED0 */
2850 int
2851 perfuse_node_pathconf(struct puffs_usermount *pu, puffs_cookie_t opc,
2852 	int name, int *retval)
2853 {
2854 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
2855 	return 0;
2856 }
2857 
2858 int
2859 perfuse_node_advlock(struct puffs_usermount *pu, puffs_cookie_t opc,
2860 	void *id, int op, struct flock *fl, int flags)
2861 {
2862 	struct perfuse_state *ps;
2863 	int fop;
2864 	perfuse_msg_t *pm;
2865 	uint64_t fh;
2866 	struct fuse_lk_in *fli;
2867 	struct fuse_out_header *foh;
2868 	struct fuse_lk_out *flo;
2869 	uint32_t owner;
2870 	size_t len;
2871 	int error;
2872 
2873 	node_ref(opc);
2874 
2875 	/*
2876 	 * Make sure we do have a filehandle, as the FUSE filesystem
2877 	 * expect one. E.g.: if we provide none, GlusterFS logs an error
2878 	 * "0-glusterfs-fuse: xl is NULL"
2879 	 *
2880 	 * We need the read file handle if the file is open read only,
2881 	 * in order to support shared locks on read-only files.
2882 	 * NB: The kernel always sends advlock for read-only
2883 	 * files at exit time when the process used lock, see
2884 	 * sys_exit -> exit1 -> fd_free -> fd_close -> VOP_ADVLOCK
2885 	 */
2886 	if ((fh = perfuse_get_fh(opc, FREAD)) == FUSE_UNKNOWN_FH) {
2887 		error = EBADF;
2888 		goto out;
2889 	}
2890 
2891 	ps = puffs_getspecific(pu);
2892 
2893 	if (op == F_GETLK)
2894 		fop = FUSE_GETLK;
2895 	else
2896 		fop = (flags & F_WAIT) ? FUSE_SETLKW : FUSE_SETLK;
2897 
2898 	/*
2899 	 * XXX ps_new_msg() is called with NULL creds, which will
2900 	 * be interpreted as FUSE superuser. We have no way to
2901 	 * know the requesting process' credential, but since advlock()
2902 	 * is supposed to operate on a file that has been open(),
2903 	 * permission should have already been checked at open() time.
2904 	 */
2905 	pm = ps->ps_new_msg(pu, opc, fop, sizeof(*fli), NULL);
2906 	fli = GET_INPAYLOAD(ps, pm, fuse_lk_in);
2907 	fli->fh = fh;
2908 	fli->owner = (uint64_t)(vaddr_t)id;
2909 	fli->lk.start = fl->l_start;
2910 	fli->lk.end = fl->l_start + fl->l_len;
2911 	fli->lk.type = fl->l_type;
2912 	fli->lk.pid = fl->l_pid;
2913 	fli->lk_flags = (flags & F_FLOCK) ? FUSE_LK_FLOCK : 0;
2914 
2915 	owner = (uint32_t)(vaddr_t)id;
2916 
2917 #ifdef PERFUSE_DEBUG
2918 	if (perfuse_diagflags & PDF_FH)
2919 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
2920 			__func__, (void *)opc,
2921 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fli->fh);
2922 #endif
2923 
2924 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
2925 		goto out;
2926 
2927 	foh = GET_OUTHDR(ps, pm);
2928 	len = foh->len - sizeof(*foh);
2929 
2930 	/*
2931 	 * Save or clear the lock
2932 	 */
2933 	switch (op) {
2934 	case F_GETLK:
2935 		if (len != sizeof(*flo))
2936 			DERRX(EX_SOFTWARE,
2937 			      "%s: Unexpected lock reply len %zd",
2938 			      __func__, len);
2939 
2940 		flo = GET_OUTPAYLOAD(ps, pm, fuse_lk_out);
2941 		fl->l_start = flo->lk.start;
2942 		fl->l_len = flo->lk.end - flo->lk.start;
2943 		fl->l_pid = flo->lk.pid;
2944 		fl->l_type = flo->lk.type;
2945 		fl->l_whence = SEEK_SET;	/* libfuse hardcodes it */
2946 
2947 		PERFUSE_NODE_DATA(opc)->pnd_lock_owner = flo->lk.pid;
2948 		break;
2949 	case F_UNLCK:
2950 		owner = 0;
2951 		/* FALLTHROUGH */
2952 	case F_SETLK:
2953 		/* FALLTHROUGH */
2954 	case F_SETLKW:
2955 		if (error != 0)
2956 			PERFUSE_NODE_DATA(opc)->pnd_lock_owner = owner;
2957 
2958 		if (len != 0)
2959 			DERRX(EX_SOFTWARE,
2960 			      "%s: Unexpected unlock reply len %zd",
2961 			      __func__, len);
2962 
2963 		break;
2964 	default:
2965 		DERRX(EX_SOFTWARE, "%s: Unexpected op %d", __func__, op);
2966 		break;
2967 	}
2968 
2969 	ps->ps_destroy_msg(pm);
2970 	error = 0;
2971 
2972 out:
2973 	node_rele(opc);
2974 	return error;
2975 }
2976 
2977 int
2978 perfuse_node_read(struct puffs_usermount *pu, puffs_cookie_t opc, uint8_t *buf,
2979 	off_t offset, size_t *resid, const struct puffs_cred *pcr, int ioflag)
2980 {
2981 	struct perfuse_state *ps;
2982 	struct perfuse_node_data *pnd;
2983 	const struct vattr *vap;
2984 	perfuse_msg_t *pm;
2985 	struct fuse_read_in *fri;
2986 	struct fuse_out_header *foh;
2987 	size_t readen;
2988 	int error;
2989 
2990 	ps = puffs_getspecific(pu);
2991 	pnd = PERFUSE_NODE_DATA(opc);
2992 	vap = puffs_pn_getvap((struct puffs_node *)opc);
2993 
2994 	/*
2995 	 * NetBSD turns that into a getdents(2) output
2996 	 * We just do a EISDIR as this feature is of little use.
2997 	 */
2998 	if (vap->va_type == VDIR)
2999 		return EISDIR;
3000 
3001 	if ((u_quad_t)offset + *resid > vap->va_size)
3002 		DWARNX("%s %p read %lld@%zu beyond EOF %" PRIu64 "\n",
3003 		       __func__, (void *)opc, (long long)offset,
3004 		       *resid, vap->va_size);
3005 
3006 	do {
3007 		size_t max_read;
3008 
3009 		max_read = ps->ps_max_readahead - sizeof(*foh);
3010 		/*
3011 		 * flags may be set to FUSE_READ_LOCKOWNER
3012 		 * if lock_owner is provided.
3013 		 */
3014 		pm = ps->ps_new_msg(pu, opc, FUSE_READ, sizeof(*fri), pcr);
3015 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
3016 		fri->fh = perfuse_get_fh(opc, FREAD);
3017 		fri->offset = offset;
3018 		fri->size = (uint32_t)MIN(*resid, max_read);
3019 		fri->read_flags = 0; /* XXX Unused by libfuse? */
3020 		fri->lock_owner = pnd->pnd_lock_owner;
3021 		fri->flags = 0;
3022 		fri->flags |= (fri->lock_owner != 0) ? FUSE_READ_LOCKOWNER : 0;
3023 
3024 #ifdef PERFUSE_DEBUG
3025 	if (perfuse_diagflags & PDF_FH)
3026 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
3027 			__func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
3028 #endif
3029 		error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
3030 		if (error  != 0)
3031 			return error;
3032 
3033 		foh = GET_OUTHDR(ps, pm);
3034 		readen = foh->len - sizeof(*foh);
3035 
3036 #ifdef PERFUSE_DEBUG
3037 		if (readen > *resid)
3038 			DERRX(EX_SOFTWARE, "%s: Unexpected big read %zd",
3039 			      __func__, readen);
3040 #endif
3041 
3042 		(void)memcpy(buf,  _GET_OUTPAYLOAD(ps, pm, char *), readen);
3043 
3044 		buf += readen;
3045 		offset += readen;
3046 		*resid -= readen;
3047 
3048 		ps->ps_destroy_msg(pm);
3049 	} while ((*resid != 0) && (readen != 0));
3050 
3051 	if (ioflag & (IO_SYNC|IO_DSYNC))
3052 		ps->ps_syncreads++;
3053 	else
3054 		ps->ps_asyncreads++;
3055 
3056 	return 0;
3057 }
3058 
3059 int
3060 perfuse_node_write(struct puffs_usermount *pu, puffs_cookie_t opc,
3061 	uint8_t *buf, off_t offset, size_t *resid,
3062 	const struct puffs_cred *pcr, int ioflag)
3063 {
3064 	return perfuse_node_write2(pu, opc, buf, offset, resid, pcr, ioflag, 0);
3065 }
3066 
3067 /* ARGSUSED7 */
3068 int
3069 perfuse_node_write2(struct puffs_usermount *pu, puffs_cookie_t opc,
3070 	uint8_t *buf, off_t offset, size_t *resid,
3071 	const struct puffs_cred *pcr, int ioflag, int xflag)
3072 {
3073 	struct perfuse_state *ps;
3074 	struct perfuse_node_data *pnd;
3075 	struct vattr *vap;
3076 	perfuse_msg_t *pm;
3077 	struct fuse_write_in *fwi;
3078 	struct fuse_write_out *fwo;
3079 	size_t data_len;
3080 	size_t payload_len;
3081 	size_t written;
3082 	int inresize;
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 	written = 0;
3089 	inresize = 0;
3090 	error = 0;
3091 
3092 	if (vap->va_type == VDIR)
3093 		return EISDIR;
3094 
3095 	node_ref(opc);
3096 
3097 	/*
3098 	 * We need to queue write requests in order to avoid
3099 	 * dequeueing PCQ_AFTERWRITE when there are pending writes.
3100 	 */
3101 	while (pnd->pnd_flags & PND_INWRITE)
3102 		requeue_request(pu, opc, PCQ_WRITE);
3103 	pnd->pnd_flags |= PND_INWRITE;
3104 
3105 	/*
3106 	 * Serialize size access, see comment in perfuse_node_setattr().
3107 	 */
3108 	if ((u_quad_t)offset + *resid > vap->va_size) {
3109 		while (pnd->pnd_flags & PND_INRESIZE)
3110 			requeue_request(pu, opc, PCQ_RESIZE);
3111 		pnd->pnd_flags |= PND_INRESIZE;
3112 		inresize = 1;
3113 	}
3114 
3115 	/*
3116 	 * append flag: re-read the file size so that
3117 	 * we get the latest value.
3118 	 */
3119 	if (ioflag & PUFFS_IO_APPEND) {
3120 		DWARNX("%s: PUFFS_IO_APPEND set, untested code", __func__);
3121 
3122 		if ((error = perfuse_node_getattr(pu, opc, vap, pcr)) != 0)
3123 			goto out;
3124 
3125 		offset = vap->va_size;
3126 	}
3127 
3128 #ifdef PERFUSE_DEBUG
3129 	if (perfuse_diagflags & PDF_RESIZE)
3130 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__,
3131 			(void *)opc, vap->va_size);
3132 #endif
3133 
3134 	do {
3135 		size_t max_write;
3136 		/*
3137 		 * There is a writepage flag when data
3138 		 * is aligned to page size. Use it for
3139 		 * everything but the data after the last
3140 		 * page boundary.
3141 		 */
3142 		max_write = ps->ps_max_write - sizeof(*fwi);
3143 
3144 		data_len = MIN(*resid, max_write);
3145 		if (data_len > (size_t)sysconf(_SC_PAGESIZE))
3146 			data_len = data_len & ~(sysconf(_SC_PAGESIZE) - 1);
3147 
3148 		payload_len = data_len + sizeof(*fwi);
3149 
3150 		/*
3151 		 * flags may be set to FUSE_WRITE_CACHE (XXX usage?)
3152 		 * or FUSE_WRITE_LOCKOWNER, if lock_owner is provided.
3153 		 * write_flags is set to 1 for writepage.
3154 		 */
3155 		pm = ps->ps_new_msg(pu, opc, FUSE_WRITE, payload_len, pcr);
3156 		fwi = GET_INPAYLOAD(ps, pm, fuse_write_in);
3157 		fwi->fh = perfuse_get_fh(opc, FWRITE);
3158 		fwi->offset = offset;
3159 		fwi->size = (uint32_t)data_len;
3160 		fwi->write_flags = (fwi->size % sysconf(_SC_PAGESIZE)) ? 0 : 1;
3161 		fwi->lock_owner = pnd->pnd_lock_owner;
3162 		fwi->flags = 0;
3163 		fwi->flags |= (fwi->lock_owner != 0) ? FUSE_WRITE_LOCKOWNER : 0;
3164 		fwi->flags |= (ioflag & IO_DIRECT) ? 0 : FUSE_WRITE_CACHE;
3165 		(void)memcpy((fwi + 1), buf, data_len);
3166 
3167 
3168 #ifdef PERFUSE_DEBUG
3169 		if (perfuse_diagflags & PDF_FH)
3170 			DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
3171 				"fh = 0x%"PRIx64"\n", __func__,
3172 				(void *)opc, pnd->pnd_nodeid, fwi->fh);
3173 #endif
3174 		if ((error = xchg_msg(pu, opc, pm,
3175 				      sizeof(*fwo), wait_reply)) != 0)
3176 			goto out;
3177 
3178 		fwo = GET_OUTPAYLOAD(ps, pm, fuse_write_out);
3179 		written = fwo->size;
3180 		ps->ps_destroy_msg(pm);
3181 
3182 #ifdef PERFUSE_DEBUG
3183 		if (written > *resid)
3184 			DERRX(EX_SOFTWARE, "%s: Unexpected big write %zd",
3185 			      __func__, written);
3186 #endif
3187 		*resid -= written;
3188 		offset += written;
3189 		buf += written;
3190 
3191 	} while (*resid != 0);
3192 
3193 	/*
3194 	 * puffs_ops(3) says
3195 	 *  "everything must be written or an error will be generated"
3196 	 */
3197 	if (*resid != 0)
3198 		error = EFBIG;
3199 
3200 #ifdef PERFUSE_DEBUG
3201 	if (perfuse_diagflags & PDF_RESIZE) {
3202 		if (offset > (off_t)vap->va_size)
3203 			DPRINTF("<< %s %p %" PRIu64 " -> %lld\n", __func__,
3204 				(void *)opc, vap->va_size, (long long)offset);
3205 		else
3206 			DPRINTF("<< %s %p \n", __func__, (void *)opc);
3207 	}
3208 #endif
3209 
3210 	/*
3211 	 * Update file size if we wrote beyond the end
3212 	 */
3213 	if (offset > (off_t)vap->va_size)
3214 		vap->va_size = offset;
3215 
3216 	if (inresize) {
3217 #ifdef PERFUSE_DEBUG
3218 		if (!(pnd->pnd_flags & PND_INRESIZE))
3219 			DERRX(EX_SOFTWARE, "file write grow without resize");
3220 #endif
3221 		pnd->pnd_flags &= ~PND_INRESIZE;
3222 		(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
3223 	}
3224 
3225 
3226 	/*
3227 	 * Statistics
3228 	 */
3229 	if (ioflag & (IO_SYNC|IO_DSYNC))
3230 		ps->ps_syncwrites++;
3231 	else
3232 		ps->ps_asyncwrites++;
3233 
3234 	/*
3235 	 * Remember to sync the file
3236 	 */
3237 	pnd->pnd_flags |= PND_DIRTY;
3238 
3239 #ifdef PERFUSE_DEBUG
3240 	if (perfuse_diagflags & PDF_SYNC)
3241 		DPRINTF("%s: DIRTY opc = %p, file = \"%s\"\n",
3242 			__func__, (void*)opc, perfuse_node_path(ps, opc));
3243 #endif
3244 
3245 out:
3246 	/*
3247 	 * VOP_PUTPAGE causes FAF write where kernel does not
3248 	 * check operation result. At least warn if it failed.
3249 	 */
3250 #ifdef PUFFS_WRITE_FAF
3251 	if (error && (xflag & PUFFS_WRITE_FAF))
3252 		DWARN("Data loss caused by FAF write failed on \"%s\"",
3253 		      pnd->pnd_name);
3254 #endif /* PUFFS_WRITE_FAF */
3255 
3256 	/*
3257 	 * If there are no more queued write, we can resume
3258 	 * an operation awaiting write completion.
3259 	 */
3260 	pnd->pnd_flags &= ~PND_INWRITE;
3261 	if (dequeue_requests(opc, PCQ_WRITE, 1) == 0)
3262 		(void)dequeue_requests(opc, PCQ_AFTERWRITE, DEQUEUE_ALL);
3263 
3264 	node_rele(opc);
3265 	return error;
3266 }
3267 
3268 /* ARGSUSED0 */
3269 void
3270 perfuse_cache_write(struct puffs_usermount *pu, puffs_cookie_t opc, size_t size,
3271 	struct puffs_cacherun *runs)
3272 {
3273 	return;
3274 }
3275 
3276 /* ARGSUSED4 */
3277 int
3278 perfuse_node_getextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3279 	int attrns, const char *attrname, size_t *attrsize, uint8_t *attr,
3280 	size_t *resid, const struct puffs_cred *pcr)
3281 {
3282 	struct perfuse_state *ps;
3283 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3284 	perfuse_msg_t *pm;
3285 	struct fuse_getxattr_in *fgi;
3286 	struct fuse_getxattr_out *fgo;
3287 	struct fuse_out_header *foh;
3288 	size_t attrnamelen;
3289 	size_t len;
3290 	char *np;
3291 	int error;
3292 
3293 	node_ref(opc);
3294 	ps = puffs_getspecific(pu);
3295 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3296 	attrnamelen = strlen(attrname) + 1;
3297 	len = sizeof(*fgi) + attrnamelen;
3298 
3299 	pm = ps->ps_new_msg(pu, opc, FUSE_GETXATTR, len, pcr);
3300 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3301 	fgi->size = (unsigned int)((resid != NULL) ? *resid : 0);
3302 	np = (char *)(void *)(fgi + 1);
3303 	(void)strlcpy(np, attrname, attrnamelen);
3304 
3305 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3306 		goto out;
3307 
3308 	/*
3309 	 * We just get fuse_getattr_out with list size if we requested
3310 	 * a null size.
3311 	 */
3312 	if (resid == NULL) {
3313 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3314 
3315 		if (attrsize != NULL)
3316 			*attrsize = fgo->size;
3317 
3318 		ps->ps_destroy_msg(pm);
3319 		error = 0;
3320 		goto out;
3321 	}
3322 
3323 	/*
3324 	 * And with a non null requested size, we get the list just
3325 	 * after the header
3326 	 */
3327 	foh = GET_OUTHDR(ps, pm);
3328 	np = (char *)(void *)(foh + 1);
3329 
3330 	if (resid != NULL) {
3331 		len = MAX(foh->len - sizeof(*foh), *resid);
3332 		(void)memcpy(attr, np, len);
3333 		*resid -= len;
3334 	}
3335 
3336 	ps->ps_destroy_msg(pm);
3337 	error = 0;
3338 
3339 out:
3340 	node_rele(opc);
3341 	return error;
3342 }
3343 
3344 int
3345 perfuse_node_setextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3346 	int attrns, const char *attrname, uint8_t *attr, size_t *resid,
3347 	const struct puffs_cred *pcr)
3348 {
3349 	struct perfuse_state *ps;
3350 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3351 	perfuse_msg_t *pm;
3352 	struct fuse_setxattr_in *fsi;
3353 	size_t attrnamelen;
3354 	size_t len;
3355 	char *np;
3356 	int error;
3357 
3358 	node_ref(opc);
3359 	ps = puffs_getspecific(pu);
3360 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3361 	attrnamelen = strlen(attrname) + 1;
3362 	len = sizeof(*fsi) + attrnamelen + *resid;
3363 
3364 	pm = ps->ps_new_msg(pu, opc, FUSE_SETXATTR, len, pcr);
3365 	fsi = GET_INPAYLOAD(ps, pm, fuse_setxattr_in);
3366 	fsi->size = (unsigned int)*resid;
3367 	fsi->flags = 0;
3368 	np = (char *)(void *)(fsi + 1);
3369 	(void)strlcpy(np, attrname, attrnamelen);
3370 	np += attrnamelen;
3371 	(void)memcpy(np, (char *)attr, *resid);
3372 
3373 	if ((error = xchg_msg(pu, opc, pm,
3374 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
3375 		goto out;
3376 
3377 	ps->ps_destroy_msg(pm);
3378 	*resid = 0;
3379 	error = 0;
3380 
3381 out:
3382 	node_rele(opc);
3383 	return error;
3384 }
3385 
3386 /* ARGSUSED2 */
3387 int
3388 perfuse_node_listextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3389 	int attrns, size_t *attrsize, uint8_t *attrs, size_t *resid, int flag,
3390 	const struct puffs_cred *pcr)
3391 {
3392 	struct perfuse_state *ps;
3393 	perfuse_msg_t *pm;
3394 	struct fuse_getxattr_in *fgi;
3395 	struct fuse_getxattr_out *fgo;
3396 	struct fuse_out_header *foh;
3397 	char *np;
3398 	size_t len, puffs_len;
3399 	int error;
3400 
3401 	node_ref(opc);
3402 
3403 	ps = puffs_getspecific(pu);
3404 	len = sizeof(*fgi);
3405 
3406 	pm = ps->ps_new_msg(pu, opc, FUSE_LISTXATTR, len, pcr);
3407 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
3408 	if (resid != NULL)
3409 		fgi->size = (unsigned int)*resid;
3410 	else
3411 		fgi->size = 0;
3412 
3413 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
3414 		goto out;
3415 
3416 	/*
3417 	 * We just get fuse_getattr_out with list size if we requested
3418 	 * a null size.
3419 	 */
3420 	if (resid == NULL) {
3421 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
3422 
3423 		if (attrsize != NULL)
3424 			*attrsize = fgo->size;
3425 
3426 		ps->ps_destroy_msg(pm);
3427 
3428 		error = 0;
3429 		goto out;
3430 	}
3431 
3432 	/*
3433 	 * And with a non null requested size, we get the list just
3434 	 * after the header
3435 	 */
3436 	foh = GET_OUTHDR(ps, pm);
3437 	np = (char *)(void *)(foh + 1);
3438 	puffs_len = foh->len - sizeof(*foh);
3439 
3440 	if (attrs != NULL) {
3441 #ifdef PUFFS_EXTATTR_LIST_LENPREFIX
3442 		/*
3443 		 * Convert the FUSE reply to length prefixed strings
3444 		 * if this is what the kernel wants.
3445 		 */
3446 		if (flag & PUFFS_EXTATTR_LIST_LENPREFIX) {
3447 			size_t i, attrlen;
3448 
3449 			for (i = 0; i < puffs_len; i += attrlen + 1) {
3450 				attrlen = strlen(np + i);
3451 				(void)memmove(np + i + 1, np + i, attrlen);
3452 				*(np + i) = (uint8_t)attrlen;
3453 			}
3454 		}
3455 #endif /* PUFFS_EXTATTR_LIST_LENPREFIX */
3456 		(void)memcpy(attrs, np, puffs_len);
3457 		*resid -= puffs_len;
3458 	}
3459 
3460 	if (attrsize != NULL)
3461 		*attrsize = puffs_len;
3462 
3463 	ps->ps_destroy_msg(pm);
3464 	error = 0;
3465 
3466 out:
3467 	node_rele(opc);
3468 	return error;
3469 }
3470 
3471 int
3472 perfuse_node_deleteextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
3473 	int attrns, const char *attrname, const struct puffs_cred *pcr)
3474 {
3475 	struct perfuse_state *ps;
3476 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
3477 	perfuse_msg_t *pm;
3478 	size_t attrnamelen;
3479 	char *np;
3480 	int error;
3481 
3482 	node_ref(opc);
3483 
3484 	ps = puffs_getspecific(pu);
3485 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
3486 	attrnamelen = strlen(attrname) + 1;
3487 
3488 	pm = ps->ps_new_msg(pu, opc, FUSE_REMOVEXATTR, attrnamelen, pcr);
3489 	np = _GET_INPAYLOAD(ps, pm, char *);
3490 	(void)strlcpy(np, attrname, attrnamelen);
3491 
3492 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
3493 
3494 	ps->ps_destroy_msg(pm);
3495 
3496 	node_rele(opc);
3497 	return error;
3498 }
3499